Mercurial > hg > openjdk > hsx19
changeset 1346:09ac706c2623
Merge
| author | asaha |
|---|---|
| date | Wed, 24 Mar 2010 17:16:33 -0700 |
| parents | f5dd08ad65df (current diff) bf823ef06b4f (diff) |
| children | 895d9ade6111 5b29c2368d93 |
| files | src/share/vm/ci/ciEnv.cpp src/share/vm/classfile/loaderConstraints.cpp src/share/vm/classfile/loaderConstraints.hpp src/share/vm/classfile/systemDictionary.cpp src/share/vm/gc_implementation/g1/ptrQueue.inline.hpp src/share/vm/opto/type.cpp |
| diffstat | 432 files changed, 16265 insertions(+), 4287 deletions(-) [+] |
line wrap: on
line diff
--- a/.hgignore Mon Mar 15 15:51:36 2010 -0400 +++ b/.hgignore Wed Mar 24 17:16:33 2010 -0700 @@ -1,6 +1,6 @@ ^build/ ^dist/ -^nbproject/private/ +/nbproject/private/ ^src/share/tools/hsdis/build/ ^src/share/tools/IdealGraphVisualizer/[a-zA-Z0-9]*/build/ ^src/share/tools/IdealGraphVisualizer/build/
--- a/.hgtags Mon Mar 15 15:51:36 2010 -0400 +++ b/.hgtags Wed Mar 24 17:16:33 2010 -0700 @@ -51,3 +51,35 @@ f4b900403d6e4b0af51447bd13bbe23fe3a1dac7 jdk7-b74 d8dd291a362acb656026a9c0a9da48501505a1e7 jdk7-b75 9174bb32e934965288121f75394874eeb1fcb649 jdk7-b76 +455105fc81d941482f8f8056afaa7aa0949c9300 jdk7-b77 +e703499b4b51e3af756ae77c3d5e8b3058a14e4e jdk7-b78 +a5a6adfca6ecefb5894a848debabfe442ff50e25 jdk7-b79 +3003ddd1d4330b06cb4691ae74d600d3685899eb jdk7-b80 +1f9b07674480c224828852ffe137beea36b3cab5 jdk7-b81 +1999f5b12482d66c8b0daf6709daea4f51893a04 jdk7-b82 +a94714c550658fd6741793ef036cb9625dc2ab1a hs17-b01 +faf94d94786b621f8e13cbcc941ca69c6d967c3f hs17-b02 +f4b900403d6e4b0af51447bd13bbe23fe3a1dac7 hs17-b03 +d8dd291a362acb656026a9c0a9da48501505a1e7 hs17-b04 +9174bb32e934965288121f75394874eeb1fcb649 hs17-b05 +a5a6adfca6ecefb5894a848debabfe442ff50e25 hs17-b06 +3003ddd1d4330b06cb4691ae74d600d3685899eb hs17-b07 +1f9b07674480c224828852ffe137beea36b3cab5 hs17-b08 +ff3232b68fbb35185b338d7ff4695b52460243f3 hs17-b09 +981375ca07b7f0605f92f57aad95122e8c385a4d hs16-b01 +f4cbf78110c726919f46b59a3b054c54c7e889b4 hs16-b02 +07c1c01e031513bfe6a7d17c6cf30d2752824ae9 hs16-b03 +08f86fa55a31113df626a75c8a626e66a543a1bd hs16-b04 +32c83fb84370a35344676991a48440378e6b6c8a hs16-b05 +ba313800759b678979434d6da8ed3bf49eb8bea4 hs16-b06 +3c0f729815607e1678bd0c41ae68494c700dcc71 hs16-b07 +ac59d4e6dae51ac5fc31a9a4940d1857f91161b1 hs16-b08 +3f844a28c5f4912bd04043b44f21b25b0805ffc2 hs15-b01 +1605bb4eb5a7a1703b13d5b077a22cc665fe45f7 hs15-b02 +2581d90c6c9b2012da930eb4742add94a03069a0 hs15-b03 +9ab385cb0c42997e16a7761ebcd25c90560a2714 hs15-b04 +fafab5d5349c7c066d677538db67a1ee0fb33bd2 hs15-b05 +3f370a32906eb5ba993fabd7b4279be7f31052b9 jdk7-b83 +ffc8d176b84bcfb5ac21302b4feb3b0c0d69b97c jdk7-b84 +6c9796468b91dcbb39e09dfa1baf9779ac45eb66 jdk7-b85 +418bc80ce13995149eadc9eecbba21d7a9fa02ae hs17-b10
--- a/agent/src/share/classes/sun/jvm/hotspot/memory/SystemDictionary.java Mon Mar 15 15:51:36 2010 -0400 +++ b/agent/src/share/classes/sun/jvm/hotspot/memory/SystemDictionary.java Wed Mar 24 17:16:33 2010 -0700 @@ -63,12 +63,12 @@ javaSystemLoaderField = type.getOopField("_java_system_loader"); nofBuckets = db.lookupIntConstant("SystemDictionary::_nof_buckets").intValue(); - objectKlassField = type.getOopField(WK_KLASS("object_klass")); - classLoaderKlassField = type.getOopField(WK_KLASS("classloader_klass")); - stringKlassField = type.getOopField(WK_KLASS("string_klass")); - systemKlassField = type.getOopField(WK_KLASS("system_klass")); - threadKlassField = type.getOopField(WK_KLASS("thread_klass")); - threadGroupKlassField = type.getOopField(WK_KLASS("threadGroup_klass")); + objectKlassField = type.getOopField(WK_KLASS("Object_klass")); + classLoaderKlassField = type.getOopField(WK_KLASS("ClassLoader_klass")); + stringKlassField = type.getOopField(WK_KLASS("String_klass")); + systemKlassField = type.getOopField(WK_KLASS("System_klass")); + threadKlassField = type.getOopField(WK_KLASS("Thread_klass")); + threadGroupKlassField = type.getOopField(WK_KLASS("ThreadGroup_klass")); } // This WK functions must follow the definitions in systemDictionary.hpp:
--- a/make/Makefile Mon Mar 15 15:51:36 2010 -0400 +++ b/make/Makefile Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ # -# Copyright 2005-2008 Sun Microsystems, Inc. All Rights Reserved. +# Copyright 2005-2010 Sun Microsystems, Inc. All Rights Reserved. # DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. # # This code is free software; you can redistribute it and/or modify it @@ -312,10 +312,13 @@ $(EXPORT_LIB_DIR)/%.jar: $(GEN_DIR)/%.jar $(install-file) -# Include files (jvmti.h, jni.h, $(JDK_INCLUDE_SUBDIR)/jni_md.h, jmm.h) +# Include files (jvmti.h, jvmticmlr.h, jni.h, $(JDK_INCLUDE_SUBDIR)/jni_md.h, jmm.h) $(EXPORT_INCLUDE_DIR)/%: $(GEN_DIR)/jvmtifiles/% $(install-file) +$(EXPORT_INCLUDE_DIR)/%: $(HS_SRC_DIR)/share/vm/code/% + $(install-file) + $(EXPORT_INCLUDE_DIR)/%: $(HS_SRC_DIR)/share/vm/prims/% $(install-file)
--- a/make/defs.make Mon Mar 15 15:51:36 2010 -0400 +++ b/make/defs.make Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ # -# Copyright 2006-2008 Sun Microsystems, Inc. All Rights Reserved. +# Copyright 2006-2010 Sun Microsystems, Inc. All Rights Reserved. # DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. # # This code is free software; you can redistribute it and/or modify it @@ -261,6 +261,7 @@ # Common export list of files EXPORT_LIST += $(EXPORT_INCLUDE_DIR)/jvmti.h +EXPORT_LIST += $(EXPORT_INCLUDE_DIR)/jvmticmlr.h EXPORT_LIST += $(EXPORT_INCLUDE_DIR)/jni.h EXPORT_LIST += $(EXPORT_INCLUDE_DIR)/$(JDK_INCLUDE_SUBDIR)/jni_md.h EXPORT_LIST += $(EXPORT_INCLUDE_DIR)/jmm.h
--- a/make/hotspot_version Mon Mar 15 15:51:36 2010 -0400 +++ b/make/hotspot_version Wed Mar 24 17:16:33 2010 -0700 @@ -35,7 +35,7 @@ HS_MAJOR_VER=17 HS_MINOR_VER=0 -HS_BUILD_NUMBER=05 +HS_BUILD_NUMBER=10 JDK_MAJOR_VER=1 JDK_MINOR_VER=7
--- a/make/linux/makefiles/debug.make Mon Mar 15 15:51:36 2010 -0400 +++ b/make/linux/makefiles/debug.make Wed Mar 24 17:16:33 2010 -0700 @@ -38,7 +38,7 @@ "Please use 'make jvmg' to build debug JVM. \n" \ "----------------------------------------------------------------------\n") -G_SUFFIX = +G_SUFFIX = _g VERSION = debug SYSDEFS += -DASSERT -DDEBUG PICFLAGS = DEFAULT
--- a/make/linux/makefiles/fastdebug.make Mon Mar 15 15:51:36 2010 -0400 +++ b/make/linux/makefiles/fastdebug.make Wed Mar 24 17:16:33 2010 -0700 @@ -58,7 +58,7 @@ # Linker mapfile MAPFILE = $(GAMMADIR)/make/linux/makefiles/mapfile-vers-debug -G_SUFFIX = +G_SUFFIX = _g VERSION = optimized SYSDEFS += -DASSERT -DFASTDEBUG PICFLAGS = DEFAULT
--- a/make/linux/makefiles/jsig.make Mon Mar 15 15:51:36 2010 -0400 +++ b/make/linux/makefiles/jsig.make Wed Mar 24 17:16:33 2010 -0700 @@ -25,9 +25,12 @@ # Rules to build signal interposition library, used by vm.make # libjsig[_g].so: signal interposition library -JSIG = jsig$(G_SUFFIX) +JSIG = jsig LIBJSIG = lib$(JSIG).so +JSIG_G = $(JSIG)$(G_SUFFIX) +LIBJSIG_G = lib$(JSIG_G).so + JSIGSRCDIR = $(GAMMADIR)/src/os/$(Platform_os_family)/vm DEST_JSIG = $(JDK_LIBDIR)/$(LIBJSIG) @@ -50,6 +53,7 @@ @echo Making signal interposition lib... $(QUIETLY) $(CC) $(SYMFLAG) $(ARCHFLAG) $(SHARED_FLAG) $(PICFLAG) \ $(LFLAGS_JSIG) $(JSIG_DEBUG_CFLAGS) -o $@ $< -ldl + $(QUIETLY) [ -f $(LIBJSIG_G) ] || { ln -s $@ $(LIBJSIG_G); } install_jsig: $(LIBJSIG) @echo "Copying $(LIBJSIG) to $(DEST_JSIG)"
--- a/make/linux/makefiles/jvmg.make Mon Mar 15 15:51:36 2010 -0400 +++ b/make/linux/makefiles/jvmg.make Wed Mar 24 17:16:33 2010 -0700 @@ -35,7 +35,7 @@ # Linker mapfile MAPFILE = $(GAMMADIR)/make/linux/makefiles/mapfile-vers-debug -G_SUFFIX = +G_SUFFIX = _g VERSION = debug SYSDEFS += -DASSERT -DDEBUG PICFLAGS = DEFAULT
--- a/make/linux/makefiles/launcher.make Mon Mar 15 15:51:36 2010 -0400 +++ b/make/linux/makefiles/launcher.make Wed Mar 24 17:16:33 2010 -0700 @@ -25,7 +25,9 @@ # Rules to build gamma launcher, used by vm.make # gamma[_g]: launcher -LAUNCHER = gamma$(G_SUFFIX) + +LAUNCHER = gamma +LAUNCHER_G = $(LAUNCHER)$(G_SUFFIX) LAUNCHERDIR = $(GAMMADIR)/src/os/$(Platform_os_family)/launcher LAUNCHERFLAGS = $(ARCHFLAG) \ @@ -70,4 +72,5 @@ $(LINK_LAUNCHER/PRE_HOOK) \ $(LINK_LAUNCHER) $(LFLAGS_LAUNCHER) -o $@ $(LAUNCHER.o) $(LIBS_LAUNCHER); \ $(LINK_LAUNCHER/POST_HOOK) \ + [ -f $(LAUNCHER_G) ] || { ln -s $@ $(LAUNCHER_G); }; \ }
--- a/make/linux/makefiles/saproc.make Mon Mar 15 15:51:36 2010 -0400 +++ b/make/linux/makefiles/saproc.make Wed Mar 24 17:16:33 2010 -0700 @@ -25,9 +25,13 @@ # Rules to build serviceability agent library, used by vm.make # libsaproc[_g].so: serviceability agent -SAPROC = saproc$(G_SUFFIX) + +SAPROC = saproc LIBSAPROC = lib$(SAPROC).so +SAPROC_G = $(SAPROC)$(G_SUFFIX) +LIBSAPROC_G = lib$(SAPROC_G).so + AGENT_DIR = $(GAMMADIR)/agent SASRCDIR = $(AGENT_DIR)/src/os/$(Platform_os_family) @@ -75,6 +79,7 @@ $(SA_DEBUG_CFLAGS) \ -o $@ \ -lthread_db + $(QUIETLY) [ -f $(LIBSAPROC_G) ] || { ln -s $@ $(LIBSAPROC_G); } install_saproc: checkAndBuildSA $(QUIETLY) if [ -e $(LIBSAPROC) ] ; then \
--- a/make/linux/makefiles/vm.make Mon Mar 15 15:51:36 2010 -0400 +++ b/make/linux/makefiles/vm.make Wed Mar 24 17:16:33 2010 -0700 @@ -113,8 +113,9 @@ #---------------------------------------------------------------------- # JVM -JVM = jvm$(G_SUFFIX) -LIBJVM = lib$(JVM).so +JVM = jvm +LIBJVM = lib$(JVM).so +LIBJVM_G = lib$(JVM)$(G_SUFFIX).so JVM_OBJ_FILES = $(Obj_Files) @@ -201,6 +202,7 @@ $(LFLAGS_VM) -o $@ $(LIBJVM.o) $(LIBS_VM); \ $(LINK_LIB.CC/POST_HOOK) \ rm -f $@.1; ln -s $@ $@.1; \ + [ -f $(LIBJVM_G) ] || { ln -s $@ $(LIBJVM_G); ln -s $@.1 $(LIBJVM_G).1; }; \ if [ -x /usr/sbin/selinuxenabled ] ; then \ /usr/sbin/selinuxenabled; \ if [ $$? = 0 ] ; then \
--- a/make/solaris/makefiles/debug.make Mon Mar 15 15:51:36 2010 -0400 +++ b/make/solaris/makefiles/debug.make Wed Mar 24 17:16:33 2010 -0700 @@ -54,7 +54,7 @@ "Please use 'gnumake jvmg' to build debug JVM. \n" \ "-------------------------------------------------------------------------\n") -G_SUFFIX = +G_SUFFIX = _g VERSION = debug SYSDEFS += -DASSERT -DDEBUG PICFLAGS = DEFAULT
--- a/make/solaris/makefiles/dtrace.make Mon Mar 15 15:51:36 2010 -0400 +++ b/make/solaris/makefiles/dtrace.make Wed Mar 24 17:16:33 2010 -0700 @@ -24,8 +24,8 @@ # Rules to build jvm_db/dtrace, used by vm.make -# we build libjvm_dtrace/libjvm_db/dtrace for COMPILER1 and COMPILER2 -# but not for CORE configuration +# We build libjvm_dtrace/libjvm_db/dtrace for COMPILER1 and COMPILER2 +# but not for CORE or KERNEL configurations. ifneq ("${TYPE}", "CORE") ifneq ("${TYPE}", "KERNEL") @@ -37,12 +37,13 @@ else - JVM_DB = libjvm_db -LIBJVM_DB = libjvm$(G_SUFFIX)_db.so +LIBJVM_DB = libjvm_db.so +LIBJVM_DB_G = libjvm$(G_SUFFIX)_db.so JVM_DTRACE = jvm_dtrace -LIBJVM_DTRACE = libjvm$(G_SUFFIX)_dtrace.so +LIBJVM_DTRACE = libjvm_dtrace.so +LIBJVM_DTRACE_G = libjvm$(G_SUFFIX)_dtrace.so JVMOFFS = JvmOffsets JVMOFFS.o = $(JVMOFFS).o @@ -77,7 +78,7 @@ LFLAGS_JVM_DTRACE += -D_REENTRANT $(PICFLAG) else LFLAGS_JVM_DB += -mt $(PICFLAG) -xnolib -LFLAGS_JVM_DTRACE += -mt $(PICFLAG) -xnolib +LFLAGS_JVM_DTRACE += -mt $(PICFLAG) -xnolib -ldl endif ISA = $(subst i386,i486,$(shell isainfo -n)) @@ -86,18 +87,24 @@ ifneq ("${ISA}","${BUILDARCH}") XLIBJVM_DB = 64/$(LIBJVM_DB) +XLIBJVM_DB_G = 64/$(LIBJVM_DB_G) XLIBJVM_DTRACE = 64/$(LIBJVM_DTRACE) +XLIBJVM_DTRACE_G = 64/$(LIBJVM_DTRACE_G) $(XLIBJVM_DB): $(DTRACE_SRCDIR)/$(JVM_DB).c $(JVMOFFS).h $(LIBJVM_DB_MAPFILE) @echo Making $@ $(QUIETLY) mkdir -p 64/ ; \ $(CC) $(SYMFLAG) $(ARCHFLAG/$(ISA)) -D$(TYPE) -I. -I$(GENERATED) \ $(SHARED_FLAG) $(LFLAGS_JVM_DB) -o $@ $(DTRACE_SRCDIR)/$(JVM_DB).c -lc + [ -f $(XLIBJVM_DB_G) ] || { ln -s $(LIBJVM_DB) $(XLIBJVM_DB_G); } + $(XLIBJVM_DTRACE): $(DTRACE_SRCDIR)/$(JVM_DTRACE).c $(DTRACE_SRCDIR)/$(JVM_DTRACE).h $(LIBJVM_DTRACE_MAPFILE) @echo Making $@ $(QUIETLY) mkdir -p 64/ ; \ $(CC) $(SYMFLAG) $(ARCHFLAG/$(ISA)) -D$(TYPE) -I. \ $(SHARED_FLAG) $(LFLAGS_JVM_DTRACE) -o $@ $(DTRACE_SRCDIR)/$(JVM_DTRACE).c -lc -lthread -ldoor + [ -f $(XLIBJVM_DTRACE_G) ] || { ln -s $(LIBJVM_DTRACE) $(XLIBJVM_DTRACE_G); } + endif # ifneq ("${ISA}","${BUILDARCH}") ifdef USE_GCC @@ -142,11 +149,13 @@ @echo Making $@ $(QUIETLY) $(CC) $(SYMFLAG) $(ARCHFLAG) -D$(TYPE) -I. -I$(GENERATED) \ $(SHARED_FLAG) $(LFLAGS_JVM_DB) -o $@ $(DTRACE_SRCDIR)/$(JVM_DB).c -lc + [ -f $(LIBJVM_DB_G) ] || { ln -s $@ $(LIBJVM_DB_G); } $(LIBJVM_DTRACE): $(DTRACE_SRCDIR)/$(JVM_DTRACE).c $(XLIBJVM_DTRACE) $(DTRACE_SRCDIR)/$(JVM_DTRACE).h $(LIBJVM_DTRACE_MAPFILE) @echo Making $@ $(QUIETLY) $(CC) $(SYMFLAG) $(ARCHFLAG) -D$(TYPE) -I. \ $(SHARED_FLAG) $(LFLAGS_JVM_DTRACE) -o $@ $(DTRACE_SRCDIR)/$(JVM_DTRACE).c -lc -lthread -ldoor + [ -f $(LIBJVM_DTRACE_G) ] || { ln -s $@ $(LIBJVM_DTRACE_G); } $(DTRACE).d: $(DTRACE_SRCDIR)/hotspot.d $(DTRACE_SRCDIR)/hotspot_jni.d \ $(DTRACE_SRCDIR)/hs_private.d $(DTRACE_SRCDIR)/jhelper.d
--- a/make/solaris/makefiles/fastdebug.make Mon Mar 15 15:51:36 2010 -0400 +++ b/make/solaris/makefiles/fastdebug.make Wed Mar 24 17:16:33 2010 -0700 @@ -90,7 +90,6 @@ # for this method for now. (fix this when dtrace bug 6258412 is fixed) OPT_CFLAGS/ciEnv.o = $(OPT_CFLAGS) -xinline=no%__1cFciEnvbFpost_compiled_method_load_event6MpnHnmethod__v_ - # (OPT_CFLAGS/SLOWER is also available, to alter compilation of buggy files) # If you set HOTSPARC_GENERIC=yes, you disable all OPT_CFLAGS settings @@ -115,8 +114,7 @@ # and mustn't be otherwise. MAPFILE_DTRACE = $(GAMMADIR)/make/solaris/makefiles/mapfile-vers-$(TYPE) - -G_SUFFIX = +G_SUFFIX = _g VERSION = optimized SYSDEFS += -DASSERT -DFASTDEBUG -DCHECK_UNHANDLED_OOPS PICFLAGS = DEFAULT
--- a/make/solaris/makefiles/jsig.make Mon Mar 15 15:51:36 2010 -0400 +++ b/make/solaris/makefiles/jsig.make Wed Mar 24 17:16:33 2010 -0700 @@ -25,8 +25,11 @@ # Rules to build signal interposition library, used by vm.make # libjsig[_g].so: signal interposition library -JSIG = jsig$(G_SUFFIX) -LIBJSIG = lib$(JSIG).so +JSIG = jsig +LIBJSIG = lib$(JSIG).so + +JSIG_G = $(JSIG)$(G_SUFFIX) +LIBJSIG_G = lib$(JSIG_G).so JSIGSRCDIR = $(GAMMADIR)/src/os/$(Platform_os_family)/vm @@ -46,6 +49,7 @@ @echo Making signal interposition lib... $(QUIETLY) $(CC) $(SYMFLAG) $(ARCHFLAG) $(SHARED_FLAG) $(PICFLAG) \ $(LFLAGS_JSIG) -o $@ $< -ldl + [ -f $(LIBJSIG_G) ] || { ln -s $@ $(LIBJSIG_G); } install_jsig: $(LIBJSIG) @echo "Copying $(LIBJSIG) to $(DEST_JSIG)"
--- a/make/solaris/makefiles/jvmg.make Mon Mar 15 15:51:36 2010 -0400 +++ b/make/solaris/makefiles/jvmg.make Wed Mar 24 17:16:33 2010 -0700 @@ -51,7 +51,7 @@ # and mustn't be otherwise. MAPFILE_DTRACE = $(GAMMADIR)/make/solaris/makefiles/mapfile-vers-$(TYPE) -G_SUFFIX = +G_SUFFIX = _g VERSION = debug SYSDEFS += -DASSERT -DDEBUG PICFLAGS = DEFAULT
--- a/make/solaris/makefiles/launcher.make Mon Mar 15 15:51:36 2010 -0400 +++ b/make/solaris/makefiles/launcher.make Wed Mar 24 17:16:33 2010 -0700 @@ -25,7 +25,8 @@ # Rules to build gamma launcher, used by vm.make # gamma[_g]: launcher -LAUNCHER = gamma$(G_SUFFIX) +LAUNCHER = gamma +LAUNCHER_G = $(LAUNCHER)$(G_SUFFIX) LAUNCHERDIR = $(GAMMADIR)/src/os/$(Platform_os_family)/launcher LAUNCHERFLAGS = $(ARCHFLAG) \ @@ -88,5 +89,6 @@ $(LINK_LAUNCHER/PRE_HOOK) \ $(LINK_LAUNCHER) $(LFLAGS_LAUNCHER) -o $@ $(LAUNCHER.o) $(LIBS_LAUNCHER); \ $(LINK_LAUNCHER/POST_HOOK) \ + [ -f $(LAUNCHER_G) ] || { ln -s $@ $(LAUNCHER_G); }; \ ;; \ esac
--- a/make/solaris/makefiles/saproc.make Mon Mar 15 15:51:36 2010 -0400 +++ b/make/solaris/makefiles/saproc.make Wed Mar 24 17:16:33 2010 -0700 @@ -25,9 +25,13 @@ # Rules to build serviceability agent library, used by vm.make # libsaproc[_g].so: serviceability agent -SAPROC = saproc$(G_SUFFIX) + +SAPROC = saproc LIBSAPROC = lib$(SAPROC).so +SAPROC_G = $(SAPROC)$(G_SUFFIX) +LIBSAPROC_G = lib$(SAPROC_G).so + AGENT_DIR = $(GAMMADIR)/agent SASRCDIR = $(AGENT_DIR)/src/os/$(Platform_os_family)/proc @@ -69,6 +73,7 @@ $(SA_LFLAGS) \ -o $@ \ -ldl -ldemangle -lthread -lc + [ -f $(LIBSAPROC_G) ] || { ln -s $@ $(LIBSAPROC_G); } install_saproc: checkAndBuildSA $(QUIETLY) if [ -f $(LIBSAPROC) ] ; then \
--- a/make/solaris/makefiles/sparcWorks.make Mon Mar 15 15:51:36 2010 -0400 +++ b/make/solaris/makefiles/sparcWorks.make Wed Mar 24 17:16:33 2010 -0700 @@ -281,8 +281,6 @@ OPT_CFLAGS=-xO4 $(EXTRA_OPT_CFLAGS) endif -CFLAGS += $(GAMMADIR)/src/os_cpu/solaris_sparc/vm/solaris_sparc.il - endif # sparc ifeq ("${Platform_arch_model}", "x86_32") @@ -293,13 +291,14 @@ # [phh] Is this still true for 6.1? OPT_CFLAGS+=-xO3 -CFLAGS += $(GAMMADIR)/src/os_cpu/solaris_x86/vm/solaris_x86_32.il - endif # 32bit x86 # no more exceptions CFLAGS/NOEX=-noex +# Inline functions +CFLAGS += $(GAMMADIR)/src/os_cpu/solaris_${Platform_arch}/vm/solaris_${Platform_arch_model}.il + # Reduce code bloat by reverting back to 5.0 behavior for static initializers CFLAGS += -Qoption ccfe -one_static_init @@ -312,6 +311,15 @@ PICFLAG/BETTER = $(PICFLAG/DEFAULT) PICFLAG/BYFILE = $(PICFLAG/$@)$(PICFLAG/DEFAULT$(PICFLAG/$@)) +# Use $(MAPFLAG:FILENAME=real_file_name) to specify a map file. +MAPFLAG = -M FILENAME + +# Use $(SONAMEFLAG:SONAME=soname) to specify the intrinsic name of a shared obj +SONAMEFLAG = -h SONAME + +# Build shared library +SHARED_FLAG = -G + # Would be better if these weren't needed, since we link with CC, but # at present removing them causes run-time errors LFLAGS += -library=Crun
--- a/make/solaris/makefiles/vm.make Mon Mar 15 15:51:36 2010 -0400 +++ b/make/solaris/makefiles/vm.make Wed Mar 24 17:16:33 2010 -0700 @@ -108,11 +108,16 @@ # older libm before libCrun, just to make sure it's found and used first. LIBS += -lsocket -lsched -ldl $(LIBM) -lCrun -lthread -ldoor -lc else +ifeq ($(COMPILER_REV_NUMERIC), 502) +# SC6.1 has it's own libm.so: specifying anything else provokes a name conflict. +LIBS += -ldl -lthread -lsocket -lm -lsched -ldoor +else LIBS += -ldl -lthread -lsocket $(LIBM) -lsched -ldoor -endif +endif # 502 +endif # 505 else LIBS += -lsocket -lsched -ldl $(LIBM) -lthread -lc -endif +endif # sparcWorks # By default, link the *.o into the library, not the executable. LINK_INTO$(LINK_INTO) = LIBJVM @@ -126,8 +131,9 @@ #---------------------------------------------------------------------- # JVM -JVM = jvm$(G_SUFFIX) -LIBJVM = lib$(JVM).so +JVM = jvm +LIBJVM = lib$(JVM).so +LIBJVM_G = lib$(JVM)$(G_SUFFIX).so JVM_OBJ_FILES = $(Obj_Files) $(DTRACE_OBJS) @@ -173,11 +179,12 @@ -sbfast|-xsbfast) \ ;; \ *) \ - echo Linking vm...; \ - $(LINK_LIB.CC/PRE_HOOK) \ - $(LINK_VM) $(LFLAGS_VM) -o $@ $(LIBJVM.o) $(LIBS_VM); \ - $(LINK_LIB.CC/POST_HOOK) \ - rm -f $@.1; ln -s $@ $@.1; \ + echo Linking vm...; \ + $(LINK_LIB.CC/PRE_HOOK) \ + $(LINK_VM) $(LFLAGS_VM) -o $@ $(LIBJVM.o) $(LIBS_VM); \ + $(LINK_LIB.CC/POST_HOOK) \ + rm -f $@.1; ln -s $@ $@.1; \ + [ -f $(LIBJVM_G) ] || { ln -s $@ $(LIBJVM_G); ln -s $@.1 $(LIBJVM_G).1; }; \ ;; \ esac
--- a/src/cpu/sparc/vm/c1_LIRAssembler_sparc.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/sparc/vm/c1_LIRAssembler_sparc.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2000-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2000-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -189,14 +189,17 @@ Register OSR_buf = osrBufferPointer()->as_register(); { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below"); int monitor_offset = BytesPerWord * method()->max_locals() + - (BasicObjectLock::size() * BytesPerWord) * (number_of_locks - 1); + (2 * BytesPerWord) * (number_of_locks - 1); + // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in + // the OSR buffer using 2 word entries: first the lock and then + // the oop. for (int i = 0; i < number_of_locks; i++) { - int slot_offset = monitor_offset - ((i * BasicObjectLock::size()) * BytesPerWord); + int slot_offset = monitor_offset - ((i * 2) * BytesPerWord); #ifdef ASSERT // verify the interpreter's monitor has a non-null object { Label L; - __ ld_ptr(OSR_buf, slot_offset + BasicObjectLock::obj_offset_in_bytes(), O7); + __ ld_ptr(OSR_buf, slot_offset + 1*BytesPerWord, O7); __ cmp(G0, O7); __ br(Assembler::notEqual, false, Assembler::pt, L); __ delayed()->nop(); @@ -205,9 +208,9 @@ } #endif // ASSERT // Copy the lock field into the compiled activation. - __ ld_ptr(OSR_buf, slot_offset + BasicObjectLock::lock_offset_in_bytes(), O7); + __ ld_ptr(OSR_buf, slot_offset + 0, O7); __ st_ptr(O7, frame_map()->address_for_monitor_lock(i)); - __ ld_ptr(OSR_buf, slot_offset + BasicObjectLock::obj_offset_in_bytes(), O7); + __ ld_ptr(OSR_buf, slot_offset + 1*BytesPerWord, O7); __ st_ptr(O7, frame_map()->address_for_monitor_object(i)); } } @@ -354,7 +357,7 @@ } -void LIR_Assembler::emit_exception_handler() { +int LIR_Assembler::emit_exception_handler() { // if the last instruction is a call (typically to do a throw which // is coming at the end after block reordering) the return address // must still point into the code area in order to avoid assertion @@ -370,15 +373,12 @@ if (handler_base == NULL) { // not enough space left for the handler bailout("exception handler overflow"); - return; + return -1; } -#ifdef ASSERT + int offset = code_offset(); -#endif // ASSERT - compilation()->offsets()->set_value(CodeOffsets::Exceptions, code_offset()); - - - if (compilation()->has_exception_handlers() || compilation()->env()->jvmti_can_post_exceptions()) { + + if (compilation()->has_exception_handlers() || compilation()->env()->jvmti_can_post_on_exceptions()) { __ call(Runtime1::entry_for(Runtime1::handle_exception_id), relocInfo::runtime_call_type); __ delayed()->nop(); } @@ -387,11 +387,13 @@ __ delayed()->nop(); debug_only(__ stop("should have gone to the caller");) assert(code_offset() - offset <= exception_handler_size, "overflow"); - __ end_a_stub(); + + return offset; } -void LIR_Assembler::emit_deopt_handler() { + +int LIR_Assembler::emit_deopt_handler() { // if the last instruction is a call (typically to do a throw which // is coming at the end after block reordering) the return address // must still point into the code area in order to avoid assertion @@ -405,23 +407,18 @@ if (handler_base == NULL) { // not enough space left for the handler bailout("deopt handler overflow"); - return; + return -1; } -#ifdef ASSERT + int offset = code_offset(); -#endif // ASSERT - compilation()->offsets()->set_value(CodeOffsets::Deopt, code_offset()); - AddressLiteral deopt_blob(SharedRuntime::deopt_blob()->unpack()); - __ JUMP(deopt_blob, G3_scratch, 0); // sethi;jmp __ delayed()->nop(); - assert(code_offset() - offset <= deopt_handler_size, "overflow"); - debug_only(__ stop("should have gone to the caller");) - __ end_a_stub(); + + return offset; } @@ -953,9 +950,11 @@ } else { #ifdef _LP64 assert(base != to_reg->as_register_lo(), "can't handle this"); + assert(O7 != to_reg->as_register_lo(), "can't handle this"); __ ld(base, offset + hi_word_offset_in_bytes, to_reg->as_register_lo()); + __ lduw(base, offset + lo_word_offset_in_bytes, O7); // in case O7 is base or offset, use it last __ sllx(to_reg->as_register_lo(), 32, to_reg->as_register_lo()); - __ ld(base, offset + lo_word_offset_in_bytes, to_reg->as_register_lo()); + __ or3(to_reg->as_register_lo(), O7, to_reg->as_register_lo()); #else if (base == to_reg->as_register_lo()) { __ ld(base, offset + hi_word_offset_in_bytes, to_reg->as_register_hi()); @@ -976,8 +975,8 @@ FloatRegister reg = to_reg->as_double_reg(); // split unaligned loads if (unaligned || PatchALot) { - __ ldf(FloatRegisterImpl::S, base, offset + BytesPerWord, reg->successor()); - __ ldf(FloatRegisterImpl::S, base, offset, reg); + __ ldf(FloatRegisterImpl::S, base, offset + 4, reg->successor()); + __ ldf(FloatRegisterImpl::S, base, offset, reg); } else { __ ldf(FloatRegisterImpl::D, base, offset, to_reg->as_double_reg()); } @@ -2200,6 +2199,7 @@ Register len = O2; __ add(src, arrayOopDesc::base_offset_in_bytes(basic_type), src_ptr); + LP64_ONLY(__ sra(src_pos, 0, src_pos);) //higher 32bits must be null if (shift == 0) { __ add(src_ptr, src_pos, src_ptr); } else { @@ -2208,6 +2208,7 @@ } __ add(dst, arrayOopDesc::base_offset_in_bytes(basic_type), dst_ptr); + LP64_ONLY(__ sra(dst_pos, 0, dst_pos);) //higher 32bits must be null if (shift == 0) { __ add(dst_ptr, dst_pos, dst_ptr); } else { @@ -2729,9 +2730,6 @@ } Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias); - __ lduw(counter_addr, tmp1); - __ add(tmp1, DataLayout::counter_increment, tmp1); - __ stw(tmp1, counter_addr); Bytecodes::Code bc = method->java_code_at_bci(bci); // Perform additional virtual call profiling for invokevirtual and // invokeinterface bytecodes @@ -2821,15 +2819,23 @@ __ set(DataLayout::counter_increment, tmp1); __ st_ptr(tmp1, mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)) - mdo_offset_bias); - if (i < (VirtualCallData::row_limit() - 1)) { - __ br(Assembler::always, false, Assembler::pt, update_done); - __ delayed()->nop(); - } + __ br(Assembler::always, false, Assembler::pt, update_done); + __ delayed()->nop(); __ bind(next_test); } + // Receiver did not match any saved receiver and there is no empty row for it. + // Increment total counter to indicate polymorphic case. + __ lduw(counter_addr, tmp1); + __ add(tmp1, DataLayout::counter_increment, tmp1); + __ stw(tmp1, counter_addr); __ bind(update_done); } + } else { + // Static call + __ lduw(counter_addr, tmp1); + __ add(tmp1, DataLayout::counter_increment, tmp1); + __ stw(tmp1, counter_addr); } }
--- a/src/cpu/sparc/vm/c1_LIRGenerator_sparc.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/sparc/vm/c1_LIRGenerator_sparc.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -144,17 +144,17 @@ if (index->is_register()) { // apply the shift and accumulate the displacement if (shift > 0) { - LIR_Opr tmp = new_register(T_INT); + LIR_Opr tmp = new_pointer_register(); __ shift_left(index, shift, tmp); index = tmp; } if (disp != 0) { - LIR_Opr tmp = new_register(T_INT); + LIR_Opr tmp = new_pointer_register(); if (Assembler::is_simm13(disp)) { - __ add(tmp, LIR_OprFact::intConst(disp), tmp); + __ add(tmp, LIR_OprFact::intptrConst(disp), tmp); index = tmp; } else { - __ move(LIR_OprFact::intConst(disp), tmp); + __ move(LIR_OprFact::intptrConst(disp), tmp); __ add(tmp, index, tmp); index = tmp; } @@ -162,8 +162,8 @@ } } else if (disp != 0 && !Assembler::is_simm13(disp)) { // index is illegal so replace it with the displacement loaded into a register - index = new_register(T_INT); - __ move(LIR_OprFact::intConst(disp), index); + index = new_pointer_register(); + __ move(LIR_OprFact::intptrConst(disp), index); disp = 0; }
--- a/src/cpu/sparc/vm/c1_globals_sparc.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/sparc/vm/c1_globals_sparc.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -22,10 +22,9 @@ * */ -// // Sets the default values for platform dependent flags used by the client compiler. // (see c1_globals.hpp) -// + #ifndef TIERED define_pd_global(bool, BackgroundCompilation, true ); define_pd_global(bool, CICompileOSR, true ); @@ -48,27 +47,24 @@ define_pd_global(bool, UseTLAB, true ); define_pd_global(bool, ProfileInterpreter, false); define_pd_global(intx, FreqInlineSize, 325 ); -define_pd_global(intx, NewRatio, 8 ); // Design center runs on 1.3.1 define_pd_global(bool, ResizeTLAB, true ); define_pd_global(intx, ReservedCodeCacheSize, 32*M ); define_pd_global(intx, CodeCacheExpansionSize, 32*K ); define_pd_global(uintx,CodeCacheMinBlockLength, 1); -define_pd_global(uintx, PermSize, 12*M ); -define_pd_global(uintx, MaxPermSize, 64*M ); -define_pd_global(bool, NeverActAsServerClassMachine, true); +define_pd_global(uintx,PermSize, 12*M ); +define_pd_global(uintx,MaxPermSize, 64*M ); +define_pd_global(bool, NeverActAsServerClassMachine, true ); define_pd_global(intx, NewSizeThreadIncrease, 16*K ); -define_pd_global(uintx, DefaultMaxRAM, 1*G); +define_pd_global(uint64_t,MaxRAM, 1ULL*G); define_pd_global(intx, InitialCodeCacheSize, 160*K); -#endif // TIERED +#endif // !TIERED define_pd_global(bool, UseTypeProfile, false); define_pd_global(bool, RoundFPResults, false); - -define_pd_global(bool, LIRFillDelaySlots, true); +define_pd_global(bool, LIRFillDelaySlots, true ); define_pd_global(bool, OptimizeSinglePrecision, false); -define_pd_global(bool, CSEArrayLength, true); +define_pd_global(bool, CSEArrayLength, true ); define_pd_global(bool, TwoOperandLIRForm, false); - -define_pd_global(intx, SafepointPollOffset, 0); +define_pd_global(intx, SafepointPollOffset, 0 );
--- a/src/cpu/sparc/vm/c2_globals_sparc.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/sparc/vm/c2_globals_sparc.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -59,7 +59,6 @@ define_pd_global(intx, FreqInlineSize, 175); define_pd_global(intx, INTPRESSURE, 48); // large register set define_pd_global(intx, InteriorEntryAlignment, 16); // = CodeEntryAlignment -define_pd_global(intx, NewRatio, 2); define_pd_global(intx, NewSizeThreadIncrease, ScaleForWordSize(4*K)); // The default setting 16/16 seems to work best. // (For _228_jack 16/16 is 2% better than 4/4, 16/4, 32/32, 32/16, or 16/32.) @@ -83,25 +82,25 @@ // sequence of instructions to load a 64 bit pointer. // // InitialCodeCacheSize derived from specjbb2000 run. -define_pd_global(intx, InitialCodeCacheSize, 2048*K); // Integral multiple of CodeCacheExpansionSize -define_pd_global(intx, ReservedCodeCacheSize, 48*M); -define_pd_global(intx, CodeCacheExpansionSize, 64*K); +define_pd_global(intx, InitialCodeCacheSize, 2048*K); // Integral multiple of CodeCacheExpansionSize +define_pd_global(intx, ReservedCodeCacheSize, 48*M); +define_pd_global(intx, CodeCacheExpansionSize, 64*K); // Ergonomics related flags -define_pd_global(uintx, DefaultMaxRAM, 32*G); +define_pd_global(uint64_t,MaxRAM, 128ULL*G); #else // InitialCodeCacheSize derived from specjbb2000 run. -define_pd_global(intx, InitialCodeCacheSize, 1536*K); // Integral multiple of CodeCacheExpansionSize -define_pd_global(intx, ReservedCodeCacheSize, 32*M); -define_pd_global(intx, CodeCacheExpansionSize, 32*K); +define_pd_global(intx, InitialCodeCacheSize, 1536*K); // Integral multiple of CodeCacheExpansionSize +define_pd_global(intx, ReservedCodeCacheSize, 32*M); +define_pd_global(intx, CodeCacheExpansionSize, 32*K); // Ergonomics related flags -define_pd_global(uintx, DefaultMaxRAM, 1*G); +define_pd_global(uint64_t,MaxRAM, 4ULL*G); #endif -define_pd_global(uintx,CodeCacheMinBlockLength, 4); +define_pd_global(uintx,CodeCacheMinBlockLength, 4); // Heap related flags -define_pd_global(uintx, PermSize, ScaleForWordSize(16*M)); -define_pd_global(uintx, MaxPermSize, ScaleForWordSize(64*M)); +define_pd_global(uintx,PermSize, ScaleForWordSize(16*M)); +define_pd_global(uintx,MaxPermSize, ScaleForWordSize(64*M)); // Ergonomics related flags define_pd_global(bool, NeverActAsServerClassMachine, false);
--- a/src/cpu/sparc/vm/frame_sparc.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/sparc/vm/frame_sparc.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -366,8 +366,9 @@ // as get_original_pc() needs correct value for unextended_sp() if (_pc != NULL) { _cb = CodeCache::find_blob(_pc); - if (_cb != NULL && _cb->is_nmethod() && ((nmethod*)_cb)->is_deopt_pc(_pc)) { - _pc = ((nmethod*)_cb)->get_original_pc(this); + address original_pc = nmethod::get_deopt_original_pc(this); + if (original_pc != NULL) { + _pc = original_pc; _deopt_state = is_deoptimized; } else { _deopt_state = not_deoptimized; @@ -519,9 +520,9 @@ _cb = CodeCache::find_blob(pc); *O7_addr() = pc - pc_return_offset; _cb = CodeCache::find_blob(_pc); - if (_cb != NULL && _cb->is_nmethod() && ((nmethod*)_cb)->is_deopt_pc(_pc)) { - address orig = ((nmethod*)_cb)->get_original_pc(this); - assert(orig == _pc, "expected original to be stored before patching"); + address original_pc = nmethod::get_deopt_original_pc(this); + if (original_pc != NULL) { + assert(original_pc == _pc, "expected original to be stored before patching"); _deopt_state = is_deoptimized; } else { _deopt_state = not_deoptimized;
--- a/src/cpu/sparc/vm/globals_sparc.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/sparc/vm/globals_sparc.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -22,10 +22,8 @@ * */ -// // Sets the default values for platform dependent flags used by the runtime system. // (see globals.hpp) -// // For sparc we do not do call backs when a thread is in the interpreter, because the // interpreter dispatch needs at least two instructions - first to load the dispatch address @@ -41,26 +39,23 @@ define_pd_global(bool, ImplicitNullChecks, true); // Generate code for implicit null checks define_pd_global(bool, UncommonNullCast, true); // Uncommon-trap NULLs past to check cast -define_pd_global(intx, CodeEntryAlignment, 32); -define_pd_global(uintx, TLABSize, 0); -define_pd_global(uintx, NewSize, ScaleForWordSize((2048 * K) + (2 * (64 * K)))); -define_pd_global(intx, SurvivorRatio, 8); -define_pd_global(intx, InlineFrequencyCount, 50); // we can use more inlining on the SPARC -define_pd_global(intx, InlineSmallCode, 1500); +define_pd_global(intx, CodeEntryAlignment, 32); +define_pd_global(intx, InlineFrequencyCount, 50); // we can use more inlining on the SPARC +define_pd_global(intx, InlineSmallCode, 1500); #ifdef _LP64 // Stack slots are 2X larger in LP64 than in the 32 bit VM. -define_pd_global(intx, ThreadStackSize, 1024); -define_pd_global(intx, VMThreadStackSize, 1024); +define_pd_global(intx, ThreadStackSize, 1024); +define_pd_global(intx, VMThreadStackSize, 1024); #else -define_pd_global(intx, ThreadStackSize, 512); -define_pd_global(intx, VMThreadStackSize, 512); +define_pd_global(intx, ThreadStackSize, 512); +define_pd_global(intx, VMThreadStackSize, 512); #endif define_pd_global(intx, StackYellowPages, 2); define_pd_global(intx, StackRedPages, 1); define_pd_global(intx, StackShadowPages, 3 DEBUG_ONLY(+1)); -define_pd_global(intx, PreInflateSpin, 40); // Determined by running design center +define_pd_global(intx, PreInflateSpin, 40); // Determined by running design center define_pd_global(bool, RewriteBytecodes, true); define_pd_global(bool, RewriteFrequentPairs, true);
--- a/src/cpu/sparc/vm/interp_masm_sparc.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/sparc/vm/interp_masm_sparc.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -1681,11 +1681,8 @@ // If no method data exists, go to profile_continue. test_method_data_pointer(profile_continue); - // We are making a call. Increment the count. - increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); - // Record the receiver type. - record_klass_in_profile(receiver, scratch); + record_klass_in_profile(receiver, scratch, true); // The method data pointer needs to be updated to reflect the new target. update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); @@ -1695,9 +1692,13 @@ void InterpreterMacroAssembler::record_klass_in_profile_helper( Register receiver, Register scratch, - int start_row, Label& done) { - if (TypeProfileWidth == 0) + int start_row, Label& done, bool is_virtual_call) { + if (TypeProfileWidth == 0) { + if (is_virtual_call) { + increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); + } return; + } int last_row = VirtualCallData::row_limit() - 1; assert(start_row <= last_row, "must be work left to do"); @@ -1714,6 +1715,7 @@ // See if the receiver is receiver[n]. int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); test_mdp_data_at(recvr_offset, receiver, next_test, scratch); + // delayed()->tst(scratch); // The receiver is receiver[n]. Increment count[n]. int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); @@ -1723,20 +1725,31 @@ bind(next_test); if (test_for_null_also) { + Label found_null; // Failed the equality check on receiver[n]... Test for null. if (start_row == last_row) { // The only thing left to do is handle the null case. - brx(Assembler::notZero, false, Assembler::pt, done); - delayed()->nop(); + if (is_virtual_call) { + brx(Assembler::zero, false, Assembler::pn, found_null); + delayed()->nop(); + // Receiver did not match any saved receiver and there is no empty row for it. + // Increment total counter to indicate polymorphic case. + increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); + ba(false, done); + delayed()->nop(); + bind(found_null); + } else { + brx(Assembler::notZero, false, Assembler::pt, done); + delayed()->nop(); + } break; } // Since null is rare, make it be the branch-taken case. - Label found_null; brx(Assembler::zero, false, Assembler::pn, found_null); delayed()->nop(); // Put all the "Case 3" tests here. - record_klass_in_profile_helper(receiver, scratch, start_row + 1, done); + record_klass_in_profile_helper(receiver, scratch, start_row + 1, done, is_virtual_call); // Found a null. Keep searching for a matching receiver, // but remember that this is an empty (unused) slot. @@ -1753,16 +1766,18 @@ int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); mov(DataLayout::counter_increment, scratch); set_mdp_data_at(count_offset, scratch); - ba(false, done); - delayed()->nop(); + if (start_row > 0) { + ba(false, done); + delayed()->nop(); + } } void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, - Register scratch) { + Register scratch, bool is_virtual_call) { assert(ProfileInterpreter, "must be profiling"); Label done; - record_klass_in_profile_helper(receiver, scratch, 0, done); + record_klass_in_profile_helper(receiver, scratch, 0, done, is_virtual_call); bind (done); } @@ -1840,7 +1855,7 @@ mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); // Record the object type. - record_klass_in_profile(klass, scratch); + record_klass_in_profile(klass, scratch, false); } // The method data pointer needs to be updated.
--- a/src/cpu/sparc/vm/interp_masm_sparc.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/sparc/vm/interp_masm_sparc.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -290,9 +290,9 @@ void test_mdp_data_at(int offset, Register value, Label& not_equal_continue, Register scratch); - void record_klass_in_profile(Register receiver, Register scratch); + void record_klass_in_profile(Register receiver, Register scratch, bool is_virtual_call); void record_klass_in_profile_helper(Register receiver, Register scratch, - int start_row, Label& done); + int start_row, Label& done, bool is_virtual_call); void update_mdp_by_offset(int offset_of_disp, Register scratch); void update_mdp_by_offset(Register reg, int offset_of_disp,
--- a/src/cpu/sparc/vm/interpreter_sparc.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/sparc/vm/interpreter_sparc.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -394,6 +394,11 @@ } +bool AbstractInterpreter::can_be_compiled(methodHandle m) { + // No special entry points that preclude compilation + return true; +} + // This method tells the deoptimizer how big an interpreted frame must be: int AbstractInterpreter::size_activation(methodOop method, int tempcount,
--- a/src/cpu/sparc/vm/sharedRuntime_sparc.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/sparc/vm/sharedRuntime_sparc.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2003-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2003-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -851,10 +851,10 @@ __ set(reg2offset(r_1) + extraspace + bias, ld_off); #else int ld_off = reg2offset(r_1) + extraspace + bias; +#endif // _LP64 #ifdef ASSERT G1_forced = true; #endif // ASSERT -#endif // _LP64 r_1 = G1_scratch->as_VMReg();// as part of the load/store shuffle if (!r_2->is_valid()) __ ld (base, ld_off, G1_scratch); else __ ldx(base, ld_off, G1_scratch); @@ -865,9 +865,11 @@ if (sig_bt[i] == T_OBJECT || sig_bt[i] == T_ARRAY) { store_c2i_object(r, base, st_off); } else if (sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) { +#ifndef _LP64 if (TieredCompilation) { assert(G1_forced || sig_bt[i] != T_LONG, "should not use register args for longs"); } +#endif // _LP64 store_c2i_long(r, base, st_off, r_2->is_stack()); } else { store_c2i_int(r, base, st_off); @@ -1189,7 +1191,8 @@ // VMReg max_arg, int comp_args_on_stack, // VMRegStackSlots const BasicType *sig_bt, - const VMRegPair *regs) { + const VMRegPair *regs, + AdapterFingerPrint* fingerprint) { address i2c_entry = __ pc(); AdapterGenerator agen(masm); @@ -1258,7 +1261,7 @@ agen.gen_c2i_adapter(total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup); __ flush(); - return new AdapterHandlerEntry(i2c_entry, c2i_entry, c2i_unverified_entry); + return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry); }
--- a/src/cpu/sparc/vm/sparc.ad Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/sparc/vm/sparc.ad Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ // -// Copyright 1998-2009 Sun Microsystems, Inc. All Rights Reserved. +// Copyright 1998-2010 Sun Microsystems, Inc. All Rights Reserved. // DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. // // This code is free software; you can redistribute it and/or modify it @@ -1885,6 +1885,10 @@ return RegMask(); } +const RegMask Matcher::method_handle_invoke_SP_save_mask() { + return RegMask(); +} + %} @@ -6664,7 +6668,7 @@ ins_pipe(ialu_imm); %} -instruct cmovII_U_reg(cmpOpU cmp, flagsRegU icc, iRegI dst, iRegI src) %{ +instruct cmovIIu_reg(cmpOpU cmp, flagsRegU icc, iRegI dst, iRegI src) %{ match(Set dst (CMoveI (Binary cmp icc) (Binary dst src))); ins_cost(150); size(4); @@ -6673,7 +6677,7 @@ ins_pipe(ialu_reg); %} -instruct cmovII_U_imm(cmpOpU cmp, flagsRegU icc, iRegI dst, immI11 src) %{ +instruct cmovIIu_imm(cmpOpU cmp, flagsRegU icc, iRegI dst, immI11 src) %{ match(Set dst (CMoveI (Binary cmp icc) (Binary dst src))); ins_cost(140); size(4); @@ -6719,6 +6723,16 @@ ins_pipe(ialu_reg); %} +// This instruction also works with CmpN so we don't need cmovNN_reg. +instruct cmovNIu_reg(cmpOpU cmp, flagsRegU icc, iRegN dst, iRegN src) %{ + match(Set dst (CMoveN (Binary cmp icc) (Binary dst src))); + ins_cost(150); + size(4); + format %{ "MOV$cmp $icc,$src,$dst" %} + ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) ); + ins_pipe(ialu_reg); +%} + instruct cmovNF_reg(cmpOpF cmp, flagsRegF fcc, iRegN dst, iRegN src) %{ match(Set dst (CMoveN (Binary cmp fcc) (Binary dst src))); ins_cost(150); @@ -6756,6 +6770,16 @@ ins_pipe(ialu_reg); %} +instruct cmovPIu_reg(cmpOpU cmp, flagsRegU icc, iRegP dst, iRegP src) %{ + match(Set dst (CMoveP (Binary cmp icc) (Binary dst src))); + ins_cost(150); + + size(4); + format %{ "MOV$cmp $icc,$src,$dst\t! ptr" %} + ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) ); + ins_pipe(ialu_reg); +%} + instruct cmovPI_imm(cmpOp cmp, flagsReg icc, iRegP dst, immP0 src) %{ match(Set dst (CMoveP (Binary cmp icc) (Binary dst src))); ins_cost(140); @@ -6766,6 +6790,16 @@ ins_pipe(ialu_imm); %} +instruct cmovPIu_imm(cmpOpU cmp, flagsRegU icc, iRegP dst, immP0 src) %{ + match(Set dst (CMoveP (Binary cmp icc) (Binary dst src))); + ins_cost(140); + + size(4); + format %{ "MOV$cmp $icc,$src,$dst\t! ptr" %} + ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) ); + ins_pipe(ialu_imm); +%} + instruct cmovPF_reg(cmpOpF cmp, flagsRegF fcc, iRegP dst, iRegP src) %{ match(Set dst (CMoveP (Binary cmp fcc) (Binary dst src))); ins_cost(150); @@ -6805,6 +6839,17 @@ ins_pipe(int_conditional_float_move); %} +instruct cmovFIu_reg(cmpOpU cmp, flagsRegU icc, regF dst, regF src) %{ + match(Set dst (CMoveF (Binary cmp icc) (Binary dst src))); + ins_cost(150); + + size(4); + format %{ "FMOVS$cmp $icc,$src,$dst" %} + opcode(0x101); + ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) ); + ins_pipe(int_conditional_float_move); +%} + // Conditional move, instruct cmovFF_reg(cmpOpF cmp, flagsRegF fcc, regF dst, regF src) %{ match(Set dst (CMoveF (Binary cmp fcc) (Binary dst src))); @@ -6838,6 +6883,17 @@ ins_pipe(int_conditional_double_move); %} +instruct cmovDIu_reg(cmpOpU cmp, flagsRegU icc, regD dst, regD src) %{ + match(Set dst (CMoveD (Binary cmp icc) (Binary dst src))); + ins_cost(150); + + size(4); + format %{ "FMOVD$cmp $icc,$src,$dst" %} + opcode(0x102); + ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) ); + ins_pipe(int_conditional_double_move); +%} + // Conditional move, instruct cmovDF_reg(cmpOpF cmp, flagsRegF fcc, regD dst, regD src) %{ match(Set dst (CMoveD (Binary cmp fcc) (Binary dst src))); @@ -6877,6 +6933,17 @@ %} +instruct cmovLIu_reg(cmpOpU cmp, flagsRegU icc, iRegL dst, iRegL src) %{ + match(Set dst (CMoveL (Binary cmp icc) (Binary dst src))); + ins_cost(150); + + size(4); + format %{ "MOV$cmp $icc,$src,$dst\t! long" %} + ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) ); + ins_pipe(ialu_reg); +%} + + instruct cmovLF_reg(cmpOpF cmp, flagsRegF fcc, iRegL dst, iRegL src) %{ match(Set dst (CMoveL (Binary cmp fcc) (Binary dst src))); ins_cost(150);
--- a/src/cpu/sparc/vm/stubGenerator_sparc.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/sparc/vm/stubGenerator_sparc.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -2862,6 +2862,9 @@ // arraycopy stubs used by compilers generate_arraycopy_stubs(); + + // Don't initialize the platform math functions since sparc + // doesn't have intrinsics for these operations. }
--- a/src/cpu/sparc/vm/templateInterpreter_sparc.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/sparc/vm/templateInterpreter_sparc.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -150,8 +150,7 @@ } -address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, bool unbox) { - assert(!unbox, "NYI");//6815692// +address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) { address compiled_entry = __ pc(); Label cont;
--- a/src/cpu/x86/vm/assembler_x86.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/assembler_x86.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -2251,6 +2251,7 @@ emit_byte(0x9D); } +#ifndef _LP64 // no 32bit push/pop on amd64 void Assembler::popl(Address dst) { // NOTE: this will adjust stack by 8byte on 64bits InstructionMark im(this); @@ -2258,6 +2259,7 @@ emit_byte(0x8F); emit_operand(rax, dst); } +#endif void Assembler::prefetch_prefix(Address src) { prefix(src); @@ -2428,6 +2430,7 @@ emit_byte(0x9C); } +#ifndef _LP64 // no 32bit push/pop on amd64 void Assembler::pushl(Address src) { // Note this will push 64bit on 64bit InstructionMark im(this); @@ -2435,6 +2438,7 @@ emit_byte(0xFF); emit_operand(rsi, src); } +#endif void Assembler::pxor(XMMRegister dst, Address src) { NOT_LP64(assert(VM_Version::supports_sse2(), "")); @@ -5591,7 +5595,12 @@ } void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src) { - andpd(dst, as_Address(src)); + if (reachable(src)) { + andpd(dst, as_Address(src)); + } else { + lea(rscratch1, src); + andpd(dst, Address(rscratch1, 0)); + } } void MacroAssembler::andptr(Register dst, int32_t imm32) { @@ -6078,11 +6087,21 @@ } void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src) { - comisd(dst, as_Address(src)); + if (reachable(src)) { + comisd(dst, as_Address(src)); + } else { + lea(rscratch1, src); + comisd(dst, Address(rscratch1, 0)); + } } void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src) { - comiss(dst, as_Address(src)); + if (reachable(src)) { + comiss(dst, as_Address(src)); + } else { + lea(rscratch1, src); + comiss(dst, Address(rscratch1, 0)); + } } @@ -7647,7 +7666,7 @@ #ifdef ASSERT Label L; - testl(tmp, tmp); + testptr(tmp, tmp); jccb(Assembler::notZero, L); hlt(); bind(L);
--- a/src/cpu/x86/vm/assembler_x86.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/assembler_x86.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1244,7 +1244,9 @@ void pcmpestri(XMMRegister xmm1, XMMRegister xmm2, int imm8); void pcmpestri(XMMRegister xmm1, Address src, int imm8); +#ifndef _LP64 // no 32bit push/pop on amd64 void popl(Address dst); +#endif #ifdef _LP64 void popq(Address dst); @@ -1285,7 +1287,9 @@ // Interleave Low Bytes void punpcklbw(XMMRegister dst, XMMRegister src); +#ifndef _LP64 // no 32bit push/pop on amd64 void pushl(Address src); +#endif void pushq(Address src);
--- a/src/cpu/x86/vm/c1_LIRAssembler_x86.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/c1_LIRAssembler_x86.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2000-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2000-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -301,22 +301,25 @@ Register OSR_buf = osrBufferPointer()->as_pointer_register(); { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below"); int monitor_offset = BytesPerWord * method()->max_locals() + - (BasicObjectLock::size() * BytesPerWord) * (number_of_locks - 1); + (2 * BytesPerWord) * (number_of_locks - 1); + // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in + // the OSR buffer using 2 word entries: first the lock and then + // the oop. for (int i = 0; i < number_of_locks; i++) { - int slot_offset = monitor_offset - ((i * BasicObjectLock::size()) * BytesPerWord); + int slot_offset = monitor_offset - ((i * 2) * BytesPerWord); #ifdef ASSERT // verify the interpreter's monitor has a non-null object { Label L; - __ cmpptr(Address(OSR_buf, slot_offset + BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD); + __ cmpptr(Address(OSR_buf, slot_offset + 1*BytesPerWord), (int32_t)NULL_WORD); __ jcc(Assembler::notZero, L); __ stop("locked object is NULL"); __ bind(L); } #endif - __ movptr(rbx, Address(OSR_buf, slot_offset + BasicObjectLock::lock_offset_in_bytes())); + __ movptr(rbx, Address(OSR_buf, slot_offset + 0)); __ movptr(frame_map()->address_for_monitor_lock(i), rbx); - __ movptr(rbx, Address(OSR_buf, slot_offset + BasicObjectLock::obj_offset_in_bytes())); + __ movptr(rbx, Address(OSR_buf, slot_offset + 1*BytesPerWord)); __ movptr(frame_map()->address_for_monitor_object(i), rbx); } } @@ -415,13 +418,12 @@ } -void LIR_Assembler::emit_exception_handler() { +int LIR_Assembler::emit_exception_handler() { // if the last instruction is a call (typically to do a throw which // is coming at the end after block reordering) the return address // must still point into the code area in order to avoid assertion // failures when searching for the corresponding bci => add a nop // (was bug 5/14/1999 - gri) - __ nop(); // generate code for exception handler @@ -429,17 +431,14 @@ if (handler_base == NULL) { // not enough space left for the handler bailout("exception handler overflow"); - return; + return -1; } -#ifdef ASSERT + int offset = code_offset(); -#endif // ASSERT - - compilation()->offsets()->set_value(CodeOffsets::Exceptions, code_offset()); // if the method does not have an exception handler, then there is // no reason to search for one - if (compilation()->has_exception_handlers() || compilation()->env()->jvmti_can_post_exceptions()) { + if (compilation()->has_exception_handlers() || compilation()->env()->jvmti_can_post_on_exceptions()) { // the exception oop and pc are in rax, and rdx // no other registers need to be preserved, so invalidate them __ invalidate_registers(false, true, true, false, true, true); @@ -471,19 +470,19 @@ // unwind activation and forward exception to caller // rax,: exception __ jump(RuntimeAddress(Runtime1::entry_for(Runtime1::unwind_exception_id))); - assert(code_offset() - offset <= exception_handler_size, "overflow"); - __ end_a_stub(); + + return offset; } -void LIR_Assembler::emit_deopt_handler() { + +int LIR_Assembler::emit_deopt_handler() { // if the last instruction is a call (typically to do a throw which // is coming at the end after block reordering) the return address // must still point into the code area in order to avoid assertion // failures when searching for the corresponding bci => add a nop // (was bug 5/14/1999 - gri) - __ nop(); // generate code for exception handler @@ -491,23 +490,17 @@ if (handler_base == NULL) { // not enough space left for the handler bailout("deopt handler overflow"); - return; + return -1; } -#ifdef ASSERT + int offset = code_offset(); -#endif // ASSERT - - compilation()->offsets()->set_value(CodeOffsets::Deopt, code_offset()); - InternalAddress here(__ pc()); __ pushptr(here.addr()); - __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack())); - assert(code_offset() - offset <= deopt_handler_size, "overflow"); - __ end_a_stub(); + return offset; } @@ -785,7 +778,13 @@ ShouldNotReachHere(); __ movoop(as_Address(addr, noreg), c->as_jobject()); } else { +#ifdef _LP64 + __ movoop(rscratch1, c->as_jobject()); + null_check_here = code_offset(); + __ movptr(as_Address_lo(addr), rscratch1); +#else __ movoop(as_Address(addr), c->as_jobject()); +#endif } } break; @@ -1118,8 +1117,14 @@ __ pushptr(frame_map()->address_for_slot(src ->single_stack_ix())); __ popptr (frame_map()->address_for_slot(dest->single_stack_ix())); } else { +#ifndef _LP64 __ pushl(frame_map()->address_for_slot(src ->single_stack_ix())); __ popl (frame_map()->address_for_slot(dest->single_stack_ix())); +#else + //no pushl on 64bits + __ movl(rscratch1, frame_map()->address_for_slot(src ->single_stack_ix())); + __ movl(frame_map()->address_for_slot(dest->single_stack_ix()), rscratch1); +#endif } } else if (src->is_double_stack()) { @@ -3136,8 +3141,10 @@ #ifdef _LP64 assert_different_registers(c_rarg0, dst, dst_pos, length); + __ movl2ptr(src_pos, src_pos); //higher 32bits must be null __ lea(c_rarg0, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type))); assert_different_registers(c_rarg1, length); + __ movl2ptr(dst_pos, dst_pos); //higher 32bits must be null __ lea(c_rarg1, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type))); __ mov(c_rarg2, length); @@ -3202,7 +3209,6 @@ Register mdo = op->mdo()->as_register(); __ movoop(mdo, md->constant_encoding()); Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset())); - __ addl(counter_addr, DataLayout::counter_increment); Bytecodes::Code bc = method->java_code_at_bci(bci); // Perform additional virtual call profiling for invokevirtual and // invokeinterface bytecodes @@ -3269,14 +3275,18 @@ __ jcc(Assembler::notEqual, next_test); __ movptr(recv_addr, recv); __ movl(Address(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i))), DataLayout::counter_increment); - if (i < (VirtualCallData::row_limit() - 1)) { - __ jmp(update_done); - } + __ jmp(update_done); __ bind(next_test); } + // Receiver did not match any saved receiver and there is no empty row for it. + // Increment total counter to indicate polymorphic case. + __ addl(counter_addr, DataLayout::counter_increment); __ bind(update_done); } + } else { + // Static call + __ addl(counter_addr, DataLayout::counter_increment); } }
--- a/src/cpu/x86/vm/c1_LIRGenerator_x86.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/c1_LIRGenerator_x86.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -755,8 +755,19 @@ } LIR_Opr addr = new_pointer_register(); - __ move(obj.result(), addr); - __ add(addr, offset.result(), addr); + LIR_Address* a; + if(offset.result()->is_constant()) { + a = new LIR_Address(obj.result(), + NOT_LP64(offset.result()->as_constant_ptr()->as_jint()) LP64_ONLY((int)offset.result()->as_constant_ptr()->as_jlong()), + as_BasicType(type)); + } else { + a = new LIR_Address(obj.result(), + offset.result(), + LIR_Address::times_1, + 0, + as_BasicType(type)); + } + __ leal(LIR_OprFact::address(a), addr); if (type == objectType) { // Write-barrier needed for Object fields. // Do the pre-write barrier, if any.
--- a/src/cpu/x86/vm/c1_globals_x86.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/c1_globals_x86.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -22,10 +22,8 @@ * */ -// // Sets the default values for platform dependent flags used by the client compiler. // (see c1_globals.hpp) -// #ifndef TIERED define_pd_global(bool, BackgroundCompilation, true ); @@ -48,27 +46,24 @@ define_pd_global(intx, OnStackReplacePercentage, 933 ); define_pd_global(intx, FreqInlineSize, 325 ); -define_pd_global(intx, NewRatio, 12 ); define_pd_global(intx, NewSizeThreadIncrease, 4*K ); define_pd_global(intx, InitialCodeCacheSize, 160*K); define_pd_global(intx, ReservedCodeCacheSize, 32*M ); define_pd_global(bool, ProfileInterpreter, false); define_pd_global(intx, CodeCacheExpansionSize, 32*K ); define_pd_global(uintx,CodeCacheMinBlockLength, 1); -define_pd_global(uintx, PermSize, 12*M ); -define_pd_global(uintx, MaxPermSize, 64*M ); -define_pd_global(bool, NeverActAsServerClassMachine, true); -define_pd_global(uintx, DefaultMaxRAM, 1*G); +define_pd_global(uintx,PermSize, 12*M ); +define_pd_global(uintx,MaxPermSize, 64*M ); +define_pd_global(bool, NeverActAsServerClassMachine, true ); +define_pd_global(uint64_t,MaxRAM, 1ULL*G); define_pd_global(bool, CICompileOSR, true ); -#endif // TIERED +#endif // !TIERED define_pd_global(bool, UseTypeProfile, false); define_pd_global(bool, RoundFPResults, true ); - define_pd_global(bool, LIRFillDelaySlots, false); -define_pd_global(bool, OptimizeSinglePrecision, true); +define_pd_global(bool, OptimizeSinglePrecision, true ); define_pd_global(bool, CSEArrayLength, false); -define_pd_global(bool, TwoOperandLIRForm, true); +define_pd_global(bool, TwoOperandLIRForm, true ); - -define_pd_global(intx, SafepointPollOffset, 256); +define_pd_global(intx, SafepointPollOffset, 256 );
--- a/src/cpu/x86/vm/c2_globals_x86.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/c2_globals_x86.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -22,7 +22,6 @@ * */ -// // Sets the default values for platform dependent flags used by the server compiler. // (see c2_globals.hpp). Alpha-sorted. @@ -46,8 +45,8 @@ define_pd_global(intx, CompileThreshold, 10000); #endif // TIERED define_pd_global(intx, Tier2CompileThreshold, 10000); -define_pd_global(intx, Tier3CompileThreshold, 20000 ); -define_pd_global(intx, Tier4CompileThreshold, 40000 ); +define_pd_global(intx, Tier3CompileThreshold, 20000); +define_pd_global(intx, Tier4CompileThreshold, 40000); define_pd_global(intx, BackEdgeThreshold, 100000); define_pd_global(intx, Tier2BackEdgeThreshold, 100000); @@ -61,7 +60,6 @@ #ifdef AMD64 define_pd_global(intx, INTPRESSURE, 13); define_pd_global(intx, InteriorEntryAlignment, 16); -define_pd_global(intx, NewRatio, 2); define_pd_global(intx, NewSizeThreadIncrease, ScaleForWordSize(4*K)); define_pd_global(intx, LoopUnrollLimit, 60); // InitialCodeCacheSize derived from specjbb2000 run. @@ -69,19 +67,18 @@ define_pd_global(intx, CodeCacheExpansionSize, 64*K); // Ergonomics related flags -define_pd_global(uintx, DefaultMaxRAM, 32*G); +define_pd_global(uint64_t,MaxRAM, 128ULL*G); #else define_pd_global(intx, INTPRESSURE, 6); define_pd_global(intx, InteriorEntryAlignment, 4); -define_pd_global(intx, NewRatio, 8); // Design center runs on 1.3.1 define_pd_global(intx, NewSizeThreadIncrease, 4*K); -define_pd_global(intx, LoopUnrollLimit, 50); // Design center runs on 1.3.1 +define_pd_global(intx, LoopUnrollLimit, 50); // Design center runs on 1.3.1 // InitialCodeCacheSize derived from specjbb2000 run. define_pd_global(intx, InitialCodeCacheSize, 2304*K); // Integral multiple of CodeCacheExpansionSize define_pd_global(intx, CodeCacheExpansionSize, 32*K); // Ergonomics related flags -define_pd_global(uintx, DefaultMaxRAM, 1*G); +define_pd_global(uint64_t,MaxRAM, 4ULL*G); #endif // AMD64 define_pd_global(intx, OptoLoopAlignment, 16); define_pd_global(intx, RegisterCostAreaRatio, 16000); @@ -97,8 +94,8 @@ define_pd_global(uintx,CodeCacheMinBlockLength, 4); // Heap related flags -define_pd_global(uintx, PermSize, ScaleForWordSize(16*M)); -define_pd_global(uintx, MaxPermSize, ScaleForWordSize(64*M)); +define_pd_global(uintx,PermSize, ScaleForWordSize(16*M)); +define_pd_global(uintx,MaxPermSize, ScaleForWordSize(64*M)); // Ergonomics related flags define_pd_global(bool, NeverActAsServerClassMachine, false);
--- a/src/cpu/x86/vm/frame_x86.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/frame_x86.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -222,9 +222,9 @@ } ((address *)sp())[-1] = pc; _cb = CodeCache::find_blob(pc); - if (_cb != NULL && _cb->is_nmethod() && ((nmethod*)_cb)->is_deopt_pc(_pc)) { - address orig = (((nmethod*)_cb)->get_original_pc(this)); - assert(orig == _pc, "expected original to be stored before patching"); + address original_pc = nmethod::get_deopt_original_pc(this); + if (original_pc != NULL) { + assert(original_pc == _pc, "expected original PC to be stored before patching"); _deopt_state = is_deoptimized; // leave _pc as is } else { @@ -323,13 +323,63 @@ return fr; } + +//------------------------------------------------------------------------------ +// frame::verify_deopt_original_pc +// +// Verifies the calculated original PC of a deoptimization PC for the +// given unextended SP. The unextended SP might also be the saved SP +// for MethodHandle call sites. +#if ASSERT +void frame::verify_deopt_original_pc(nmethod* nm, intptr_t* unextended_sp, bool is_method_handle_return) { + frame fr; + + // This is ugly but it's better than to change {get,set}_original_pc + // to take an SP value as argument. And it's only a debugging + // method anyway. + fr._unextended_sp = unextended_sp; + + address original_pc = nm->get_original_pc(&fr); + assert(nm->code_contains(original_pc), "original PC must be in nmethod"); + assert(nm->is_method_handle_return(original_pc) == is_method_handle_return, "must be"); +} +#endif + + +//------------------------------------------------------------------------------ +// frame::sender_for_interpreter_frame frame frame::sender_for_interpreter_frame(RegisterMap* map) const { - // sp is the raw sp from the sender after adapter or interpreter extension - intptr_t* sp = (intptr_t*) addr_at(sender_sp_offset); + // SP is the raw SP from the sender after adapter or interpreter + // extension. + intptr_t* sender_sp = this->sender_sp(); // This is the sp before any possible extension (adapter/locals). intptr_t* unextended_sp = interpreter_frame_sender_sp(); + // Stored FP. + intptr_t* saved_fp = link(); + + address sender_pc = this->sender_pc(); + CodeBlob* sender_cb = CodeCache::find_blob_unsafe(sender_pc); + assert(sender_cb, "sanity"); + nmethod* sender_nm = sender_cb->as_nmethod_or_null(); + + if (sender_nm != NULL) { + // If the sender PC is a deoptimization point, get the original + // PC. For MethodHandle call site the unextended_sp is stored in + // saved_fp. + if (sender_nm->is_deopt_mh_entry(sender_pc)) { + DEBUG_ONLY(verify_deopt_mh_original_pc(sender_nm, saved_fp)); + unextended_sp = saved_fp; + } + else if (sender_nm->is_deopt_entry(sender_pc)) { + DEBUG_ONLY(verify_deopt_original_pc(sender_nm, unextended_sp)); + } + else if (sender_nm->is_method_handle_return(sender_pc)) { + unextended_sp = saved_fp; + } + } + // The interpreter and compiler(s) always save EBP/RBP in a known // location on entry. We must record where that location is // so this if EBP/RBP was live on callout from c2 we can find @@ -351,29 +401,52 @@ } #endif // AMD64 } -#endif /* COMPILER2 */ - return frame(sp, unextended_sp, link(), sender_pc()); +#endif // COMPILER2 + + return frame(sender_sp, unextended_sp, saved_fp, sender_pc); } -//------------------------------sender_for_compiled_frame----------------------- +//------------------------------------------------------------------------------ +// frame::sender_for_compiled_frame frame frame::sender_for_compiled_frame(RegisterMap* map) const { assert(map != NULL, "map must be set"); - const bool c1_compiled = _cb->is_compiled_by_c1(); // frame owned by optimizing compiler - intptr_t* sender_sp = NULL; - assert(_cb->frame_size() >= 0, "must have non-zero frame size"); - sender_sp = unextended_sp() + _cb->frame_size(); + intptr_t* sender_sp = unextended_sp() + _cb->frame_size(); + intptr_t* unextended_sp = sender_sp; // On Intel the return_address is always the word on the stack address sender_pc = (address) *(sender_sp-1); - // This is the saved value of ebp which may or may not really be an fp. - // it is only an fp if the sender is an interpreter frame (or c1?) + // This is the saved value of EBP which may or may not really be an FP. + // It is only an FP if the sender is an interpreter frame (or C1?). + intptr_t* saved_fp = (intptr_t*) *(sender_sp - frame::sender_sp_offset); + + // If we are returning to a compiled MethodHandle call site, the + // saved_fp will in fact be a saved value of the unextended SP. The + // simplest way to tell whether we are returning to such a call site + // is as follows: + CodeBlob* sender_cb = CodeCache::find_blob_unsafe(sender_pc); + assert(sender_cb, "sanity"); + nmethod* sender_nm = sender_cb->as_nmethod_or_null(); - intptr_t *saved_fp = (intptr_t*)*(sender_sp - frame::sender_sp_offset); + if (sender_nm != NULL) { + // If the sender PC is a deoptimization point, get the original + // PC. For MethodHandle call site the unextended_sp is stored in + // saved_fp. + if (sender_nm->is_deopt_mh_entry(sender_pc)) { + DEBUG_ONLY(verify_deopt_mh_original_pc(sender_nm, saved_fp)); + unextended_sp = saved_fp; + } + else if (sender_nm->is_deopt_entry(sender_pc)) { + DEBUG_ONLY(verify_deopt_original_pc(sender_nm, unextended_sp)); + } + else if (sender_nm->is_method_handle_return(sender_pc)) { + unextended_sp = saved_fp; + } + } if (map->update_map()) { // Tell GC to use argument oopmaps for some runtime stubs that need it. @@ -383,7 +456,7 @@ if (_cb->oop_maps() != NULL) { OopMapSet::update_register_map(this, map); } - // Since the prolog does the save and restore of epb there is no oopmap + // Since the prolog does the save and restore of EBP there is no oopmap // for it so we must fill in its location as if there was an oopmap entry // since if our caller was compiled code there could be live jvm state in it. map->set_location(rbp->as_VMReg(), (address) (sender_sp - frame::sender_sp_offset)); @@ -399,9 +472,12 @@ } assert(sender_sp != sp(), "must have changed"); - return frame(sender_sp, saved_fp, sender_pc); + return frame(sender_sp, unextended_sp, saved_fp, sender_pc); } + +//------------------------------------------------------------------------------ +// frame::sender frame frame::sender(RegisterMap* map) const { // Default is we done have to follow them. The sender_for_xxx will // update it accordingly
--- a/src/cpu/x86/vm/frame_x86.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/frame_x86.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -163,6 +163,14 @@ return (intptr_t*) addr_at(offset); } +#if ASSERT + // Used in frame::sender_for_{interpreter,compiled}_frame + static void verify_deopt_original_pc( nmethod* nm, intptr_t* unextended_sp, bool is_method_handle_return = false); + static void verify_deopt_mh_original_pc(nmethod* nm, intptr_t* unextended_sp) { + verify_deopt_original_pc(nm, unextended_sp, true); + } +#endif + public: // Constructors
--- a/src/cpu/x86/vm/frame_x86.inline.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/frame_x86.inline.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -35,32 +35,35 @@ _deopt_state = unknown; } -inline frame:: frame(intptr_t* sp, intptr_t* fp, address pc) { +inline frame::frame(intptr_t* sp, intptr_t* fp, address pc) { _sp = sp; _unextended_sp = sp; _fp = fp; _pc = pc; assert(pc != NULL, "no pc?"); _cb = CodeCache::find_blob(pc); - _deopt_state = not_deoptimized; - if (_cb != NULL && _cb->is_nmethod() && ((nmethod*)_cb)->is_deopt_pc(_pc)) { - _pc = (((nmethod*)_cb)->get_original_pc(this)); + + address original_pc = nmethod::get_deopt_original_pc(this); + if (original_pc != NULL) { + _pc = original_pc; _deopt_state = is_deoptimized; } else { _deopt_state = not_deoptimized; } } -inline frame:: frame(intptr_t* sp, intptr_t* unextended_sp, intptr_t* fp, address pc) { +inline frame::frame(intptr_t* sp, intptr_t* unextended_sp, intptr_t* fp, address pc) { _sp = sp; _unextended_sp = unextended_sp; _fp = fp; _pc = pc; assert(pc != NULL, "no pc?"); _cb = CodeCache::find_blob(pc); - _deopt_state = not_deoptimized; - if (_cb != NULL && _cb->is_nmethod() && ((nmethod*)_cb)->is_deopt_pc(_pc)) { - _pc = (((nmethod*)_cb)->get_original_pc(this)); + + address original_pc = nmethod::get_deopt_original_pc(this); + if (original_pc != NULL) { + _pc = original_pc; + assert(((nmethod*)_cb)->code_contains(_pc), "original PC must be in nmethod"); _deopt_state = is_deoptimized; } else { _deopt_state = not_deoptimized; @@ -86,9 +89,9 @@ _cb = CodeCache::find_blob(_pc); - _deopt_state = not_deoptimized; - if (_cb != NULL && _cb->is_nmethod() && ((nmethod*)_cb)->is_deopt_pc(_pc)) { - _pc = (((nmethod*)_cb)->get_original_pc(this)); + address original_pc = nmethod::get_deopt_original_pc(this); + if (original_pc != NULL) { + _pc = original_pc; _deopt_state = is_deoptimized; } else { _deopt_state = not_deoptimized; @@ -225,11 +228,13 @@ // top of expression stack inline intptr_t* frame::interpreter_frame_tos_address() const { intptr_t* last_sp = interpreter_frame_last_sp(); - if (last_sp == NULL ) { + if (last_sp == NULL) { return sp(); } else { - // sp() may have been extended by an adapter - assert(last_sp < fp() && last_sp >= sp(), "bad tos"); + // sp() may have been extended or shrunk by an adapter. At least + // check that we don't fall behind the legal region. + // For top deoptimized frame last_sp == interpreter_frame_monitor_end. + assert(last_sp <= (intptr_t*) interpreter_frame_monitor_end(), "bad tos"); return last_sp; } }
--- a/src/cpu/x86/vm/globals_x86.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/globals_x86.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -22,17 +22,16 @@ * */ -// // Sets the default values for platform dependent flags used by the runtime system. // (see globals.hpp) -// -define_pd_global(bool, ConvertSleepToYield, true); -define_pd_global(bool, ShareVtableStubs, true); -define_pd_global(bool, CountInterpCalls, true); +define_pd_global(bool, ConvertSleepToYield, true); +define_pd_global(bool, ShareVtableStubs, true); +define_pd_global(bool, CountInterpCalls, true); +define_pd_global(bool, NeedsDeoptSuspend, false); // only register window machines need this -define_pd_global(bool, ImplicitNullChecks, true); // Generate code for implicit null checks -define_pd_global(bool, UncommonNullCast, true); // Uncommon-trap NULLs past to check cast +define_pd_global(bool, ImplicitNullChecks, true); // Generate code for implicit null checks +define_pd_global(bool, UncommonNullCast, true); // Uncommon-trap NULLs past to check cast // See 4827828 for this change. There is no globals_core_i486.hpp. I can't // assign a different value for C2 without touching a number of files. Use @@ -42,29 +41,24 @@ // the uep and the vep doesn't get real alignment but just slops on by // only assured that the entry instruction meets the 5 byte size requirement. #ifdef COMPILER2 -define_pd_global(intx, CodeEntryAlignment, 32); +define_pd_global(intx, CodeEntryAlignment, 32); #else -define_pd_global(intx, CodeEntryAlignment, 16); +define_pd_global(intx, CodeEntryAlignment, 16); #endif // COMPILER2 +define_pd_global(intx, InlineFrequencyCount, 100); +define_pd_global(intx, InlineSmallCode, 1000); -define_pd_global(bool, NeedsDeoptSuspend, false); // only register window machines need this - -define_pd_global(uintx, TLABSize, 0); +define_pd_global(intx, StackYellowPages, 2); +define_pd_global(intx, StackRedPages, 1); #ifdef AMD64 -define_pd_global(uintx, NewSize, ScaleForWordSize(2048 * K)); // Very large C++ stack frames using solaris-amd64 optimized builds // due to lack of optimization caused by C++ compiler bugs define_pd_global(intx, StackShadowPages, SOLARIS_ONLY(20) NOT_SOLARIS(6) DEBUG_ONLY(+2)); #else -define_pd_global(uintx, NewSize, 1024 * K); define_pd_global(intx, StackShadowPages, 3 DEBUG_ONLY(+1)); #endif // AMD64 -define_pd_global(intx, InlineFrequencyCount, 100); -define_pd_global(intx, InlineSmallCode, 1000); -define_pd_global(intx, PreInflateSpin, 10); -define_pd_global(intx, StackYellowPages, 2); -define_pd_global(intx, StackRedPages, 1); +define_pd_global(intx, PreInflateSpin, 10); define_pd_global(bool, RewriteBytecodes, true); define_pd_global(bool, RewriteFrequentPairs, true);
--- a/src/cpu/x86/vm/interp_masm_x86_32.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/interp_masm_x86_32.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -196,6 +196,9 @@ } else { assert(EnableInvokeDynamic, "giant index used only for EnableInvokeDynamic"); movl(reg, Address(rsi, bcp_offset)); + // Check if the secondary index definition is still ~x, otherwise + // we have to change the following assembler code to calculate the + // plain index. assert(constantPoolCacheOopDesc::decode_secondary_index(~123) == 123, "else change next line"); notl(reg); // convert to plain index } @@ -1236,17 +1239,19 @@ // If no method data exists, go to profile_continue. test_method_data_pointer(mdp, profile_continue); - // We are making a call. Increment the count. - increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); - Label skip_receiver_profile; if (receiver_can_be_null) { + Label not_null; testptr(receiver, receiver); - jcc(Assembler::zero, skip_receiver_profile); + jccb(Assembler::notZero, not_null); + // We are making a call. Increment the count for null receiver. + increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); + jmp(skip_receiver_profile); + bind(not_null); } // Record the receiver type. - record_klass_in_profile(receiver, mdp, reg2); + record_klass_in_profile(receiver, mdp, reg2, true); bind(skip_receiver_profile); // The method data pointer needs to be updated to reflect the new target. @@ -1260,10 +1265,14 @@ void InterpreterMacroAssembler::record_klass_in_profile_helper( Register receiver, Register mdp, - Register reg2, - int start_row, Label& done) { - if (TypeProfileWidth == 0) + Register reg2, int start_row, + Label& done, bool is_virtual_call) { + if (TypeProfileWidth == 0) { + if (is_virtual_call) { + increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); + } return; + } int last_row = VirtualCallData::row_limit() - 1; assert(start_row <= last_row, "must be work left to do"); @@ -1291,19 +1300,28 @@ bind(next_test); if (row == start_row) { + Label found_null; // Failed the equality check on receiver[n]... Test for null. testptr(reg2, reg2); if (start_row == last_row) { // The only thing left to do is handle the null case. - jcc(Assembler::notZero, done); + if (is_virtual_call) { + jccb(Assembler::zero, found_null); + // Receiver did not match any saved receiver and there is no empty row for it. + // Increment total counter to indicate polymorphic case. + increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); + jmp(done); + bind(found_null); + } else { + jcc(Assembler::notZero, done); + } break; } // Since null is rare, make it be the branch-taken case. - Label found_null; jcc(Assembler::zero, found_null); // Put all the "Case 3" tests here. - record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done); + record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call); // Found a null. Keep searching for a matching receiver, // but remember that this is an empty (unused) slot. @@ -1320,16 +1338,18 @@ int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); movptr(reg2, (int32_t)DataLayout::counter_increment); set_mdp_data_at(mdp, count_offset, reg2); - jmp(done); + if (start_row > 0) { + jmp(done); + } } void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, - Register mdp, - Register reg2) { + Register mdp, Register reg2, + bool is_virtual_call) { assert(ProfileInterpreter, "must be profiling"); Label done; - record_klass_in_profile_helper(receiver, mdp, reg2, 0, done); + record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call); bind (done); } @@ -1422,7 +1442,7 @@ mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); // Record the object type. - record_klass_in_profile(klass, mdp, reg2); + record_klass_in_profile(klass, mdp, reg2, false); assert(reg2 == rdi, "we know how to fix this blown reg"); restore_locals(); // Restore EDI }
--- a/src/cpu/x86/vm/interp_masm_x86_32.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/interp_masm_x86_32.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -213,10 +213,10 @@ Label& not_equal_continue); void record_klass_in_profile(Register receiver, Register mdp, - Register reg2); + Register reg2, bool is_virtual_call); void record_klass_in_profile_helper(Register receiver, Register mdp, - Register reg2, - int start_row, Label& done); + Register reg2, int start_row, + Label& done, bool is_virtual_call); void update_mdp_by_offset(Register mdp_in, int offset_of_offset); void update_mdp_by_offset(Register mdp_in, Register reg, int offset_of_disp);
--- a/src/cpu/x86/vm/interp_masm_x86_64.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/interp_masm_x86_64.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2003-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2003-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -185,12 +185,30 @@ } +void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index, + int bcp_offset, + bool giant_index) { + assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); + if (!giant_index) { + load_unsigned_short(index, Address(r13, bcp_offset)); + } else { + assert(EnableInvokeDynamic, "giant index used only for EnableInvokeDynamic"); + movl(index, Address(r13, bcp_offset)); + // Check if the secondary index definition is still ~x, otherwise + // we have to change the following assembler code to calculate the + // plain index. + assert(constantPoolCacheOopDesc::decode_secondary_index(~123) == 123, "else change next line"); + notl(index); // convert to plain index + } +} + + void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register index, - int bcp_offset) { - assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); + int bcp_offset, + bool giant_index) { assert(cache != index, "must use different registers"); - load_unsigned_short(index, Address(r13, bcp_offset)); + get_cache_index_at_bcp(index, bcp_offset, giant_index); movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below"); // convert from field index to ConstantPoolCacheEntry index @@ -200,10 +218,10 @@ void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp, - int bcp_offset) { - assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); + int bcp_offset, + bool giant_index) { assert(cache != tmp, "must use different register"); - load_unsigned_short(tmp, Address(r13, bcp_offset)); + get_cache_index_at_bcp(tmp, bcp_offset, giant_index); assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below"); // convert from field index to ConstantPoolCacheEntry index // and from word offset to byte offset @@ -1236,18 +1254,28 @@ void InterpreterMacroAssembler::profile_virtual_call(Register receiver, Register mdp, - Register reg2) { + Register reg2, + bool receiver_can_be_null) { if (ProfileInterpreter) { Label profile_continue; // If no method data exists, go to profile_continue. test_method_data_pointer(mdp, profile_continue); - // We are making a call. Increment the count. - increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); + Label skip_receiver_profile; + if (receiver_can_be_null) { + Label not_null; + testptr(receiver, receiver); + jccb(Assembler::notZero, not_null); + // We are making a call. Increment the count for null receiver. + increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); + jmp(skip_receiver_profile); + bind(not_null); + } // Record the receiver type. - record_klass_in_profile(receiver, mdp, reg2); + record_klass_in_profile(receiver, mdp, reg2, true); + bind(skip_receiver_profile); // The method data pointer needs to be updated to reflect the new target. update_mdp_by_constant(mdp, @@ -1270,10 +1298,14 @@ // See below for example code. void InterpreterMacroAssembler::record_klass_in_profile_helper( Register receiver, Register mdp, - Register reg2, - int start_row, Label& done) { - if (TypeProfileWidth == 0) + Register reg2, int start_row, + Label& done, bool is_virtual_call) { + if (TypeProfileWidth == 0) { + if (is_virtual_call) { + increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); + } return; + } int last_row = VirtualCallData::row_limit() - 1; assert(start_row <= last_row, "must be work left to do"); @@ -1301,19 +1333,28 @@ bind(next_test); if (test_for_null_also) { + Label found_null; // Failed the equality check on receiver[n]... Test for null. testptr(reg2, reg2); if (start_row == last_row) { // The only thing left to do is handle the null case. - jcc(Assembler::notZero, done); + if (is_virtual_call) { + jccb(Assembler::zero, found_null); + // Receiver did not match any saved receiver and there is no empty row for it. + // Increment total counter to indicate polymorphic case. + increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); + jmp(done); + bind(found_null); + } else { + jcc(Assembler::notZero, done); + } break; } // Since null is rare, make it be the branch-taken case. - Label found_null; jcc(Assembler::zero, found_null); // Put all the "Case 3" tests here. - record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done); + record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call); // Found a null. Keep searching for a matching receiver, // but remember that this is an empty (unused) slot. @@ -1330,7 +1371,9 @@ int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); movl(reg2, DataLayout::counter_increment); set_mdp_data_at(mdp, count_offset, reg2); - jmp(done); + if (start_row > 0) { + jmp(done); + } } // Example state machine code for three profile rows: @@ -1342,7 +1385,7 @@ // if (row[1].rec != NULL) { // // degenerate decision tree, rooted at row[2] // if (row[2].rec == rec) { row[2].incr(); goto done; } -// if (row[2].rec != NULL) { goto done; } // overflow +// if (row[2].rec != NULL) { count.incr(); goto done; } // overflow // row[2].init(rec); goto done; // } else { // // remember row[1] is empty @@ -1355,14 +1398,15 @@ // if (row[2].rec == rec) { row[2].incr(); goto done; } // row[0].init(rec); goto done; // } +// done: void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, - Register mdp, - Register reg2) { + Register mdp, Register reg2, + bool is_virtual_call) { assert(ProfileInterpreter, "must be profiling"); Label done; - record_klass_in_profile_helper(receiver, mdp, reg2, 0, done); + record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call); bind (done); } @@ -1458,7 +1502,7 @@ mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); // Record the object type. - record_klass_in_profile(klass, mdp, reg2); + record_klass_in_profile(klass, mdp, reg2, false); } update_mdp_by_constant(mdp, mdp_delta);
--- a/src/cpu/x86/vm/interp_masm_x86_64.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/interp_masm_x86_64.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -95,9 +95,10 @@ void get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset); void get_cache_and_index_at_bcp(Register cache, Register index, - int bcp_offset); + int bcp_offset, bool giant_index = false); void get_cache_entry_pointer_at_bcp(Register cache, Register tmp, - int bcp_offset); + int bcp_offset, bool giant_index = false); + void get_cache_index_at_bcp(Register index, int bcp_offset, bool giant_index = false); void pop_ptr(Register r = rax); @@ -221,10 +222,10 @@ Label& not_equal_continue); void record_klass_in_profile(Register receiver, Register mdp, - Register reg2); + Register reg2, bool is_virtual_call); void record_klass_in_profile_helper(Register receiver, Register mdp, - Register reg2, - int start_row, Label& done); + Register reg2, int start_row, + Label& done, bool is_virtual_call); void update_mdp_by_offset(Register mdp_in, int offset_of_offset); void update_mdp_by_offset(Register mdp_in, Register reg, int offset_of_disp); @@ -236,7 +237,8 @@ void profile_call(Register mdp); void profile_final_call(Register mdp); void profile_virtual_call(Register receiver, Register mdp, - Register scratch2); + Register scratch2, + bool receiver_can_be_null = false); void profile_ret(Register return_bci, Register mdp); void profile_null_seen(Register mdp); void profile_typecheck(Register mdp, Register klass, Register scratch);
--- a/src/cpu/x86/vm/interpreter_x86_64.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/interpreter_x86_64.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -277,12 +277,11 @@ address entry_point = __ pc(); // abstract method entry - // remove return address. Not really needed, since exception - // handling throws away expression stack - __ pop(rbx); - // adjust stack to what a normal return would do - __ mov(rsp, r13); + // pop return address, reset last_sp to NULL + __ empty_expression_stack(); + __ restore_bcp(); // rsi must be correct for exception handler (was destroyed) + __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) // throw exception __ call_VM(noreg, CAST_FROM_FN_PTR(address, @@ -300,7 +299,10 @@ if (!EnableMethodHandles) { return generate_abstract_entry(); } - return generate_abstract_entry(); //6815692// + + address entry_point = MethodHandles::generate_method_handle_interpreter_entry(_masm); + + return entry_point; }
--- a/src/cpu/x86/vm/methodHandles_x86.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/methodHandles_x86.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -65,9 +65,9 @@ // Verify that argslot lies within (rsp, rbp]. Label L_ok, L_bad; __ cmpptr(rax_argslot, rbp); - __ jcc(Assembler::above, L_bad); + __ jccb(Assembler::above, L_bad); __ cmpptr(rsp, rax_argslot); - __ jcc(Assembler::below, L_ok); + __ jccb(Assembler::below, L_ok); __ bind(L_bad); __ stop(error_message); __ bind(L_ok); @@ -136,9 +136,9 @@ if (arg_slots.is_register()) { Label L_ok, L_bad; __ cmpptr(arg_slots.as_register(), (int32_t) NULL_WORD); - __ jcc(Assembler::greater, L_bad); + __ jccb(Assembler::greater, L_bad); __ testl(arg_slots.as_register(), -stack_move_unit() - 1); - __ jcc(Assembler::zero, L_ok); + __ jccb(Assembler::zero, L_ok); __ bind(L_bad); __ stop("assert arg_slots <= 0 and clear low bits"); __ bind(L_ok); @@ -173,7 +173,7 @@ __ movptr(Address(rdx_temp, arg_slots, Address::times_ptr), rbx_temp); __ addptr(rdx_temp, wordSize); __ cmpptr(rdx_temp, rax_argslot); - __ jcc(Assembler::less, loop); + __ jccb(Assembler::less, loop); } // Now move the argslot down, to point to the opened-up space. @@ -211,9 +211,9 @@ Label L_ok, L_bad; __ lea(rbx_temp, Address(rax_argslot, arg_slots, Address::times_ptr)); __ cmpptr(rbx_temp, rbp); - __ jcc(Assembler::above, L_bad); + __ jccb(Assembler::above, L_bad); __ cmpptr(rsp, rax_argslot); - __ jcc(Assembler::below, L_ok); + __ jccb(Assembler::below, L_ok); __ bind(L_bad); __ stop("deleted argument(s) must fall within current frame"); __ bind(L_ok); @@ -221,9 +221,9 @@ if (arg_slots.is_register()) { Label L_ok, L_bad; __ cmpptr(arg_slots.as_register(), (int32_t) NULL_WORD); - __ jcc(Assembler::less, L_bad); + __ jccb(Assembler::less, L_bad); __ testl(arg_slots.as_register(), -stack_move_unit() - 1); - __ jcc(Assembler::zero, L_ok); + __ jccb(Assembler::zero, L_ok); __ bind(L_bad); __ stop("assert arg_slots >= 0 and clear low bits"); __ bind(L_ok); @@ -258,7 +258,7 @@ __ movptr(Address(rdx_temp, arg_slots, Address::times_ptr), rbx_temp); __ addptr(rdx_temp, -wordSize); __ cmpptr(rdx_temp, rsp); - __ jcc(Assembler::greaterEqual, loop); + __ jccb(Assembler::greaterEqual, loop); } // Now move the argslot up, to point to the just-copied block. @@ -268,8 +268,9 @@ } #ifndef PRODUCT +extern "C" void print_method_handle(oop mh); void trace_method_handle_stub(const char* adaptername, - oopDesc* mh, + oop mh, intptr_t* entry_sp, intptr_t* saved_sp, intptr_t* saved_bp) { @@ -280,6 +281,7 @@ adaptername, (intptr_t)mh, (intptr_t)entry_sp, (intptr_t)(saved_sp - entry_sp), (intptr_t)(base_sp - last_sp), (intptr_t)saved_bp); if (last_sp != saved_sp) printf("*** last_sp="INTPTR_FORMAT"\n", (intptr_t)last_sp); + if (Verbose) print_method_handle(mh); } #endif //PRODUCT @@ -382,11 +384,11 @@ // FIXME: fill in _raise_exception_method with a suitable sun.dyn method __ movptr(rbx_method, ExternalAddress((address) &_raise_exception_method)); __ testptr(rbx_method, rbx_method); - __ jcc(Assembler::zero, no_method); + __ jccb(Assembler::zero, no_method); int jobject_oop_offset = 0; __ movptr(rbx_method, Address(rbx_method, jobject_oop_offset)); // dereference the jobject __ testptr(rbx_method, rbx_method); - __ jcc(Assembler::zero, no_method); + __ jccb(Assembler::zero, no_method); __ verify_oop(rbx_method); __ push(rdi_pc); // and restore caller PC __ jmp(rbx_method_fie); @@ -448,7 +450,7 @@ rbx_index, Address::times_ptr, base + vtableEntry::method_offset_in_bytes()); Register rbx_method = rbx_temp; - __ movl(rbx_method, vtable_entry_addr); + __ movptr(rbx_method, vtable_entry_addr); __ verify_oop(rbx_method); __ jmp(rbx_method_fie); @@ -533,16 +535,15 @@ if (arg_type == T_OBJECT) { __ movptr(Address(rax_argslot, 0), rbx_temp); } else { - __ load_sized_value(rbx_temp, prim_value_addr, + __ load_sized_value(rdx_temp, prim_value_addr, type2aelembytes(arg_type), is_signed_subword_type(arg_type)); - __ movptr(Address(rax_argslot, 0), rbx_temp); + __ movptr(Address(rax_argslot, 0), rdx_temp); #ifndef _LP64 if (arg_slots == 2) { - __ movl(rbx_temp, prim_value_addr.plus_disp(wordSize)); - __ movl(Address(rax_argslot, Interpreter::stackElementSize()), rbx_temp); + __ movl(rdx_temp, prim_value_addr.plus_disp(wordSize)); + __ movl(Address(rax_argslot, Interpreter::stackElementSize()), rdx_temp); } #endif //_LP64 - break; } if (direct_to_method) { @@ -584,7 +585,7 @@ Label done; __ movptr(rdx_temp, vmarg); __ testl(rdx_temp, rdx_temp); - __ jcc(Assembler::zero, done); // no cast if null + __ jccb(Assembler::zero, done); // no cast if null __ load_klass(rdx_temp, rdx_temp); // live at this point: @@ -675,24 +676,24 @@ // (now we are done with the old MH) // original 32-bit vmdata word must be of this form: - // | MBZ:16 | signBitCount:8 | srcDstTypes:8 | conversionOp:8 | - __ xchgl(rcx, rbx_vminfo); // free rcx for shifts + // | MBZ:6 | signBitCount:8 | srcDstTypes:8 | conversionOp:8 | + __ xchgptr(rcx, rbx_vminfo); // free rcx for shifts __ shll(rdx_temp /*, rcx*/); Label zero_extend, done; __ testl(rcx, CONV_VMINFO_SIGN_FLAG); - __ jcc(Assembler::zero, zero_extend); + __ jccb(Assembler::zero, zero_extend); // this path is taken for int->byte, int->short __ sarl(rdx_temp /*, rcx*/); - __ jmp(done); + __ jmpb(done); __ bind(zero_extend); // this is taken for int->char __ shrl(rdx_temp /*, rcx*/); __ bind(done); - __ movptr(vmarg, rdx_temp); - __ xchgl(rcx, rbx_vminfo); // restore rcx_recv + __ movl(vmarg, rdx_temp); + __ xchgptr(rcx, rbx_vminfo); // restore rcx_recv __ jump_to_method_handle_entry(rcx_recv, rdx_temp); } @@ -861,7 +862,7 @@ // Verify that argslot > destslot, by at least swap_bytes. Label L_ok; __ cmpptr(rax_argslot, rbx_destslot); - __ jcc(Assembler::aboveEqual, L_ok); + __ jccb(Assembler::aboveEqual, L_ok); __ stop("source must be above destination (upward rotation)"); __ bind(L_ok); } @@ -877,7 +878,7 @@ __ movptr(Address(rax_argslot, swap_bytes), rdx_temp); __ addptr(rax_argslot, -wordSize); __ cmpptr(rax_argslot, rbx_destslot); - __ jcc(Assembler::aboveEqual, loop); + __ jccb(Assembler::aboveEqual, loop); } else { __ addptr(rax_argslot, swap_bytes); #ifdef ASSERT @@ -885,7 +886,7 @@ // Verify that argslot < destslot, by at least swap_bytes. Label L_ok; __ cmpptr(rax_argslot, rbx_destslot); - __ jcc(Assembler::belowEqual, L_ok); + __ jccb(Assembler::belowEqual, L_ok); __ stop("source must be below destination (downward rotation)"); __ bind(L_ok); } @@ -901,7 +902,7 @@ __ movptr(Address(rax_argslot, -swap_bytes), rdx_temp); __ addptr(rax_argslot, wordSize); __ cmpptr(rax_argslot, rbx_destslot); - __ jcc(Assembler::belowEqual, loop); + __ jccb(Assembler::belowEqual, loop); } // pop the original first chunk into the destination slot, now free @@ -967,7 +968,7 @@ __ addptr(rax_argslot, wordSize); __ addptr(rdx_newarg, wordSize); __ cmpptr(rdx_newarg, rbx_oldarg); - __ jcc(Assembler::less, loop); + __ jccb(Assembler::less, loop); __ pop(rdi); // restore temp @@ -1119,7 +1120,7 @@ } __ addptr(rax_argslot, Interpreter::stackElementSize()); __ cmpptr(rax_argslot, rdx_argslot_limit); - __ jcc(Assembler::less, loop); + __ jccb(Assembler::less, loop); } else if (length_constant == 0) { __ bind(skip_array_check); // nothing to copy
--- a/src/cpu/x86/vm/runtime_x86_32.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/runtime_x86_32.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -43,11 +43,11 @@ // This code is entered with a jmp. // // Arguments: -// rax,: exception oop +// rax: exception oop // rdx: exception pc // // Results: -// rax,: exception oop +// rax: exception oop // rdx: exception pc in caller or ??? // destination: exception handler of caller // @@ -113,17 +113,17 @@ __ addptr(rsp, return_off * wordSize); // Epilog! __ pop(rdx); // Exception pc + // rax: exception handler for given <exception oop/exception pc> - // rax,: exception handler for given <exception oop/exception pc> + // Restore SP from BP if the exception PC is a MethodHandle call. + __ cmpl(Address(rcx, JavaThread::is_method_handle_exception_offset()), 0); + __ cmovptr(Assembler::notEqual, rsp, rbp); // We have a handler in rax, (could be deopt blob) // rdx - throwing pc, deopt blob will need it. __ push(rax); - // rcx contains handler address - - __ get_thread(rcx); // TLS // Get the exception __ movptr(rax, Address(rcx, JavaThread::exception_oop_offset())); // Get the exception pc in case we are deoptimized @@ -137,7 +137,7 @@ __ pop(rcx); - // rax,: exception oop + // rax: exception oop // rcx: exception handler // rdx: exception pc __ jmp (rcx);
--- a/src/cpu/x86/vm/sharedRuntime_x86_32.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/sharedRuntime_x86_32.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2003-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2003-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -907,7 +907,8 @@ int total_args_passed, int comp_args_on_stack, const BasicType *sig_bt, - const VMRegPair *regs) { + const VMRegPair *regs, + AdapterFingerPrint* fingerprint) { address i2c_entry = __ pc(); gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs); @@ -954,7 +955,7 @@ gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup); __ flush(); - return new AdapterHandlerEntry(i2c_entry, c2i_entry, c2i_unverified_entry); + return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry); } int SharedRuntime::c_calling_convention(const BasicType *sig_bt,
--- a/src/cpu/x86/vm/sharedRuntime_x86_64.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/sharedRuntime_x86_64.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2003-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2003-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -638,6 +638,10 @@ __ movptr(rax, Address(rsp, 0)); + // Must preserve original SP for loading incoming arguments because + // we need to align the outgoing SP for compiled code. + __ movptr(r11, rsp); + // Cut-out for having no stack args. Since up to 2 int/oop args are passed // in registers, we will occasionally have no stack args. int comp_words_on_stack = 0; @@ -661,6 +665,10 @@ // as far as the placement of the call instruction __ push(rax); + // Put saved SP in another register + const Register saved_sp = rax; + __ movptr(saved_sp, r11); + // Will jump to the compiled code just as if compiled code was doing it. // Pre-load the register-jump target early, to schedule it better. __ movptr(r11, Address(rbx, in_bytes(methodOopDesc::from_compiled_offset()))); @@ -680,11 +688,7 @@ assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(), "scrambled load targets?"); // Load in argument order going down. - // int ld_off = (total_args_passed + comp_words_on_stack -i)*wordSize; - // base ld_off on r13 (sender_sp) as the stack alignment makes offsets from rsp - // unpredictable - int ld_off = ((total_args_passed - 1) - i)*Interpreter::stackElementSize(); - + int ld_off = (total_args_passed - i)*Interpreter::stackElementSize() + Interpreter::value_offset_in_bytes(); // Point to interpreter value (vs. tag) int next_off = ld_off - Interpreter::stackElementSize(); // @@ -699,10 +703,14 @@ if (r_1->is_stack()) { // Convert stack slot to an SP offset (+ wordSize to account for return address ) int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size + wordSize; + + // We can use r13 as a temp here because compiled code doesn't need r13 as an input + // and if we end up going thru a c2i because of a miss a reasonable value of r13 + // will be generated. if (!r_2->is_valid()) { // sign extend??? - __ movl(rax, Address(r13, ld_off)); - __ movptr(Address(rsp, st_off), rax); + __ movl(r13, Address(saved_sp, ld_off)); + __ movptr(Address(rsp, st_off), r13); } else { // // We are using two optoregs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE @@ -715,9 +723,9 @@ // ld_off is MSW so get LSW const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)? next_off : ld_off; - __ movq(rax, Address(r13, offset)); + __ movq(r13, Address(saved_sp, offset)); // st_off is LSW (i.e. reg.first()) - __ movq(Address(rsp, st_off), rax); + __ movq(Address(rsp, st_off), r13); } } else if (r_1->is_Register()) { // Register argument Register r = r_1->as_Register(); @@ -732,16 +740,16 @@ next_off : ld_off; // this can be a misaligned move - __ movq(r, Address(r13, offset)); + __ movq(r, Address(saved_sp, offset)); } else { // sign extend and use a full word? - __ movl(r, Address(r13, ld_off)); + __ movl(r, Address(saved_sp, ld_off)); } } else { if (!r_2->is_valid()) { - __ movflt(r_1->as_XMMRegister(), Address(r13, ld_off)); + __ movflt(r_1->as_XMMRegister(), Address(saved_sp, ld_off)); } else { - __ movdbl(r_1->as_XMMRegister(), Address(r13, next_off)); + __ movdbl(r_1->as_XMMRegister(), Address(saved_sp, next_off)); } } } @@ -770,7 +778,8 @@ int total_args_passed, int comp_args_on_stack, const BasicType *sig_bt, - const VMRegPair *regs) { + const VMRegPair *regs, + AdapterFingerPrint* fingerprint) { address i2c_entry = __ pc(); gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs); @@ -816,7 +825,7 @@ gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup); __ flush(); - return new AdapterHandlerEntry(i2c_entry, c2i_entry, c2i_unverified_entry); + return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry); } int SharedRuntime::c_calling_convention(const BasicType *sig_bt, @@ -3319,6 +3328,10 @@ // rax: exception handler + // Restore SP from BP if the exception PC is a MethodHandle call. + __ cmpl(Address(r15_thread, JavaThread::is_method_handle_exception_offset()), 0); + __ cmovptr(Assembler::notEqual, rsp, rbp); + // We have a handler in rax (could be deopt blob). __ mov(r8, rax);
--- a/src/cpu/x86/vm/stubGenerator_x86_32.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/stubGenerator_x86_32.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -718,10 +718,8 @@ case BarrierSet::G1SATBCTLogging: { __ pusha(); // push registers - __ push(count); - __ push(start); - __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre))); - __ addptr(rsp, 2*wordSize); + __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), + start, count); __ popa(); } break; @@ -752,10 +750,8 @@ case BarrierSet::G1SATBCTLogging: { __ pusha(); // push registers - __ push(count); - __ push(start); - __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post))); - __ addptr(rsp, 2*wordSize); + __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), + start, count); __ popa(); } break; @@ -2030,6 +2026,54 @@ entry_checkcast_arraycopy); } + void generate_math_stubs() { + { + StubCodeMark mark(this, "StubRoutines", "log"); + StubRoutines::_intrinsic_log = (double (*)(double)) __ pc(); + + __ fld_d(Address(rsp, 4)); + __ flog(); + __ ret(0); + } + { + StubCodeMark mark(this, "StubRoutines", "log10"); + StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc(); + + __ fld_d(Address(rsp, 4)); + __ flog10(); + __ ret(0); + } + { + StubCodeMark mark(this, "StubRoutines", "sin"); + StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc(); + + __ fld_d(Address(rsp, 4)); + __ trigfunc('s'); + __ ret(0); + } + { + StubCodeMark mark(this, "StubRoutines", "cos"); + StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc(); + + __ fld_d(Address(rsp, 4)); + __ trigfunc('c'); + __ ret(0); + } + { + StubCodeMark mark(this, "StubRoutines", "tan"); + StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc(); + + __ fld_d(Address(rsp, 4)); + __ trigfunc('t'); + __ ret(0); + } + + // The intrinsic version of these seem to return the same value as + // the strict version. + StubRoutines::_intrinsic_exp = SharedRuntime::dexp; + StubRoutines::_intrinsic_pow = SharedRuntime::dpow; + } + public: // Information about frame layout at time of blocking runtime call. // Note that we only have to preserve callee-saved registers since @@ -2228,6 +2272,8 @@ MethodHandles::generate_method_handle_stub(_masm, ek); } } + + generate_math_stubs(); }
--- a/src/cpu/x86/vm/stubGenerator_x86_64.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/stubGenerator_x86_64.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2003-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2003-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -1172,7 +1172,7 @@ __ movptr(c_rarg0, addr); __ movptr(c_rarg1, count); } - __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre))); + __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 2); __ popa(); } break; @@ -1212,7 +1212,7 @@ __ shrptr(scratch, LogBytesPerHeapOop); // convert to element count __ mov(c_rarg0, start); __ mov(c_rarg1, scratch); - __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post))); + __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 2); __ popa(); } break; @@ -2731,6 +2731,79 @@ StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy; } + void generate_math_stubs() { + { + StubCodeMark mark(this, "StubRoutines", "log"); + StubRoutines::_intrinsic_log = (double (*)(double)) __ pc(); + + __ subq(rsp, 8); + __ movdbl(Address(rsp, 0), xmm0); + __ fld_d(Address(rsp, 0)); + __ flog(); + __ fstp_d(Address(rsp, 0)); + __ movdbl(xmm0, Address(rsp, 0)); + __ addq(rsp, 8); + __ ret(0); + } + { + StubCodeMark mark(this, "StubRoutines", "log10"); + StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc(); + + __ subq(rsp, 8); + __ movdbl(Address(rsp, 0), xmm0); + __ fld_d(Address(rsp, 0)); + __ flog10(); + __ fstp_d(Address(rsp, 0)); + __ movdbl(xmm0, Address(rsp, 0)); + __ addq(rsp, 8); + __ ret(0); + } + { + StubCodeMark mark(this, "StubRoutines", "sin"); + StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc(); + + __ subq(rsp, 8); + __ movdbl(Address(rsp, 0), xmm0); + __ fld_d(Address(rsp, 0)); + __ trigfunc('s'); + __ fstp_d(Address(rsp, 0)); + __ movdbl(xmm0, Address(rsp, 0)); + __ addq(rsp, 8); + __ ret(0); + } + { + StubCodeMark mark(this, "StubRoutines", "cos"); + StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc(); + + __ subq(rsp, 8); + __ movdbl(Address(rsp, 0), xmm0); + __ fld_d(Address(rsp, 0)); + __ trigfunc('c'); + __ fstp_d(Address(rsp, 0)); + __ movdbl(xmm0, Address(rsp, 0)); + __ addq(rsp, 8); + __ ret(0); + } + { + StubCodeMark mark(this, "StubRoutines", "tan"); + StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc(); + + __ subq(rsp, 8); + __ movdbl(Address(rsp, 0), xmm0); + __ fld_d(Address(rsp, 0)); + __ trigfunc('t'); + __ fstp_d(Address(rsp, 0)); + __ movdbl(xmm0, Address(rsp, 0)); + __ addq(rsp, 8); + __ ret(0); + } + + // The intrinsic version of these seem to return the same value as + // the strict version. + StubRoutines::_intrinsic_exp = SharedRuntime::dexp; + StubRoutines::_intrinsic_pow = SharedRuntime::dpow; + } + #undef __ #define __ masm-> @@ -2935,6 +3008,18 @@ // arraycopy stubs used by compilers generate_arraycopy_stubs(); + + // generic method handle stubs + if (EnableMethodHandles && SystemDictionary::MethodHandle_klass() != NULL) { + for (MethodHandles::EntryKind ek = MethodHandles::_EK_FIRST; + ek < MethodHandles::_EK_LIMIT; + ek = MethodHandles::EntryKind(1 + (int)ek)) { + StubCodeMark mark(this, "MethodHandle", MethodHandles::entry_name(ek)); + MethodHandles::generate_method_handle_stub(_masm, ek); + } + } + + generate_math_stubs(); } public:
--- a/src/cpu/x86/vm/templateInterpreter_x86_32.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/templateInterpreter_x86_32.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -155,15 +155,8 @@ } -address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, bool unbox) { +address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) { TosState incoming_state = state; - if (EnableInvokeDynamic) { - if (unbox) { - incoming_state = atos; - } - } else { - assert(!unbox, "old behavior"); - } Label interpreter_entry; address compiled_entry = __ pc(); @@ -216,46 +209,6 @@ __ restore_bcp(); __ restore_locals(); - Label L_fail; - - if (unbox && state != atos) { - // cast and unbox - BasicType type = as_BasicType(state); - if (type == T_BYTE) type = T_BOOLEAN; // FIXME - KlassHandle boxk = SystemDictionaryHandles::box_klass(type); - __ mov32(rbx, ExternalAddress((address) boxk.raw_value())); - __ testl(rax, rax); - Label L_got_value, L_get_value; - // convert nulls to zeroes (avoid NPEs here) - if (!(type == T_FLOAT || type == T_DOUBLE)) { - // if rax already contains zero bits, forge ahead - __ jcc(Assembler::zero, L_got_value); - } else { - __ jcc(Assembler::notZero, L_get_value); - __ fldz(); - __ jmp(L_got_value); - } - __ bind(L_get_value); - __ cmp32(rbx, Address(rax, oopDesc::klass_offset_in_bytes())); - __ jcc(Assembler::notEqual, L_fail); - int offset = java_lang_boxing_object::value_offset_in_bytes(type); - // Cf. TemplateTable::getfield_or_static - switch (type) { - case T_BYTE: // fall through: - case T_BOOLEAN: __ load_signed_byte(rax, Address(rax, offset)); break; - case T_CHAR: __ load_unsigned_short(rax, Address(rax, offset)); break; - case T_SHORT: __ load_signed_short(rax, Address(rax, offset)); break; - case T_INT: __ movl(rax, Address(rax, offset)); break; - case T_FLOAT: __ fld_s(Address(rax, offset)); break; - case T_DOUBLE: __ fld_d(Address(rax, offset)); break; - // Access to java.lang.Double.value does not need to be atomic: - case T_LONG: { __ movl(rdx, Address(rax, offset + 4)); - __ movl(rax, Address(rax, offset + 0)); } break; - default: ShouldNotReachHere(); - } - __ bind(L_got_value); - } - Label L_got_cache, L_giant_index; if (EnableInvokeDynamic) { __ cmpb(Address(rsi, 0), Bytecodes::_invokedynamic); @@ -263,32 +216,6 @@ } __ get_cache_and_index_at_bcp(rbx, rcx, 1, false); __ bind(L_got_cache); - if (unbox && state == atos) { - // insert a casting conversion, to keep verifier sane - Label L_ok, L_ok_pops; - __ testl(rax, rax); - __ jcc(Assembler::zero, L_ok); - __ push(rax); // save the object to check - __ push(rbx); // save CP cache reference - __ movl(rdx, Address(rax, oopDesc::klass_offset_in_bytes())); - __ movl(rbx, Address(rbx, rcx, - Address::times_4, constantPoolCacheOopDesc::base_offset() + - ConstantPoolCacheEntry::f1_offset())); - __ movl(rbx, Address(rbx, __ delayed_value(sun_dyn_CallSiteImpl::type_offset_in_bytes, rcx))); - __ movl(rbx, Address(rbx, __ delayed_value(java_dyn_MethodType::rtype_offset_in_bytes, rcx))); - __ movl(rax, Address(rbx, __ delayed_value(java_lang_Class::klass_offset_in_bytes, rcx))); - __ check_klass_subtype(rdx, rax, rbx, L_ok_pops); - __ pop(rcx); // pop and discard CP cache - __ mov(rbx, rax); // target supertype into rbx for L_fail - __ pop(rax); // failed object into rax for L_fail - __ jmp(L_fail); - - __ bind(L_ok_pops); - // restore pushed temp regs: - __ pop(rbx); - __ pop(rax); - __ bind(L_ok); - } __ movl(rbx, Address(rbx, rcx, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset())); @@ -301,14 +228,6 @@ __ bind(L_giant_index); __ get_cache_and_index_at_bcp(rbx, rcx, 1, true); __ jmp(L_got_cache); - - if (unbox) { - __ bind(L_fail); - __ push(rbx); // missed klass (required) - __ push(rax); // bad object (actual) - __ movptr(rdx, ExternalAddress((address) &Interpreter::_throw_WrongMethodType_entry)); - __ call(rdx); - } } return entry; @@ -1512,6 +1431,23 @@ } +// These should never be compiled since the interpreter will prefer +// the compiled version to the intrinsic version. +bool AbstractInterpreter::can_be_compiled(methodHandle m) { + switch (method_kind(m)) { + case Interpreter::java_lang_math_sin : // fall thru + case Interpreter::java_lang_math_cos : // fall thru + case Interpreter::java_lang_math_tan : // fall thru + case Interpreter::java_lang_math_abs : // fall thru + case Interpreter::java_lang_math_log : // fall thru + case Interpreter::java_lang_math_log10 : // fall thru + case Interpreter::java_lang_math_sqrt : + return false; + default: + return true; + } +} + // How much stack a method activation needs in words. int AbstractInterpreter::size_top_interpreter_activation(methodOop method) { @@ -1569,7 +1505,10 @@ if (interpreter_frame != NULL) { #ifdef ASSERT - assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable"); + if (!EnableMethodHandles) + // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences? + // Probably, since deoptimization doesn't work yet. + assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable"); assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)"); #endif
--- a/src/cpu/x86/vm/templateInterpreter_x86_64.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/templateInterpreter_x86_64.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2003-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2003-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -100,21 +100,26 @@ return entry; } -// Arguments are: required type in rarg1, failing object (or NULL) in rarg2 +// Arguments are: required type at TOS+8, failing object (or NULL) at TOS+4. address TemplateInterpreterGenerator::generate_WrongMethodType_handler() { address entry = __ pc(); __ pop(c_rarg2); // failing object is at TOS __ pop(c_rarg1); // required type is at TOS+8 - // expression stack must be empty before entering the VM if an - // exception happened + __ verify_oop(c_rarg1); + __ verify_oop(c_rarg2); + + // Various method handle types use interpreter registers as temps. + __ restore_bcp(); + __ restore_locals(); + + // Expression stack must be empty before entering the VM for an exception. __ empty_expression_stack(); __ call_VM(noreg, CAST_FROM_FN_PTR(address, - InterpreterRuntime:: - throw_WrongMethodTypeException), + InterpreterRuntime::throw_WrongMethodTypeException), // pass required type, failing object (or NULL) c_rarg1, c_rarg2); return entry; @@ -166,8 +171,7 @@ address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, - int step, bool unbox) { - assert(!unbox, "NYI");//6815692// + int step) { // amd64 doesn't need to do anything special about compiled returns // to the interpreter so the code that exists on x86 to place a sentinel @@ -183,15 +187,29 @@ __ restore_bcp(); __ restore_locals(); - __ get_cache_and_index_at_bcp(rbx, rcx, 1); + Label L_got_cache, L_giant_index; + if (EnableInvokeDynamic) { + __ cmpb(Address(r13, 0), Bytecodes::_invokedynamic); + __ jcc(Assembler::equal, L_giant_index); + } + __ get_cache_and_index_at_bcp(rbx, rcx, 1, false); + __ bind(L_got_cache); __ movl(rbx, Address(rbx, rcx, - Address::times_8, + Address::times_ptr, in_bytes(constantPoolCacheOopDesc::base_offset()) + 3 * wordSize)); __ andl(rbx, 0xFF); if (TaggedStackInterpreter) __ shll(rbx, 1); // 2 slots per parameter. __ lea(rsp, Address(rsp, rbx, Address::times_8)); __ dispatch_next(state, step); + + // out of the main line of code... + if (EnableInvokeDynamic) { + __ bind(L_giant_index); + __ get_cache_and_index_at_bcp(rbx, rcx, 1, true); + __ jmp(L_got_cache); + } + return entry; } @@ -431,8 +449,12 @@ __ addptr(rax, stack_base); __ subptr(rax, stack_size); + // Use the maximum number of pages we might bang. + const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages : + (StackRedPages+StackYellowPages); + // add in the red and yellow zone sizes - __ addptr(rax, (StackRedPages + StackYellowPages) * page_size); + __ addptr(rax, max_pages * page_size); // check against the current stack bottom __ cmpptr(rsp, rax); @@ -1434,6 +1456,23 @@ generate_normal_entry(synchronized); } +// These should never be compiled since the interpreter will prefer +// the compiled version to the intrinsic version. +bool AbstractInterpreter::can_be_compiled(methodHandle m) { + switch (method_kind(m)) { + case Interpreter::java_lang_math_sin : // fall thru + case Interpreter::java_lang_math_cos : // fall thru + case Interpreter::java_lang_math_tan : // fall thru + case Interpreter::java_lang_math_abs : // fall thru + case Interpreter::java_lang_math_log : // fall thru + case Interpreter::java_lang_math_log10 : // fall thru + case Interpreter::java_lang_math_sqrt : + return false; + default: + return true; + } +} + // How much stack a method activation needs in words. int AbstractInterpreter::size_top_interpreter_activation(methodOop method) { const int entry_size = frame::interpreter_frame_monitor_size(); @@ -1484,8 +1523,10 @@ tempcount* Interpreter::stackElementWords() + popframe_extra_args; if (interpreter_frame != NULL) { #ifdef ASSERT - assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), - "Frame not properly walkable"); + if (!EnableMethodHandles) + // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences? + // Probably, since deoptimization doesn't work yet. + assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable"); assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)"); #endif
--- a/src/cpu/x86/vm/templateTable_x86_32.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/templateTable_x86_32.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -2890,9 +2890,6 @@ void TemplateTable::prepare_invoke(Register method, Register index, int byte_no) { - bool is_invdyn_bootstrap = (byte_no < 0); - if (is_invdyn_bootstrap) byte_no = -byte_no; - // determine flags Bytecodes::Code code = bytecode(); const bool is_invokeinterface = code == Bytecodes::_invokeinterface; @@ -2907,8 +2904,6 @@ const Register flags = rdx; assert_different_registers(method, index, recv, flags); - assert(!is_invdyn_bootstrap || is_invokedynamic, "byte_no<0 hack only for invdyn"); - // save 'interpreter return address' __ save_bcp(); @@ -2944,9 +2939,7 @@ // load return address { address table_addr; - if (is_invdyn_bootstrap) - table_addr = (address)Interpreter::return_5_unbox_addrs_by_index_table(); - else if (is_invokeinterface || is_invokedynamic) + if (is_invokeinterface || is_invokedynamic) table_addr = (address)Interpreter::return_5_addrs_by_index_table(); else table_addr = (address)Interpreter::return_3_addrs_by_index_table(); @@ -3153,54 +3146,10 @@ __ profile_call(rsi); } - Label handle_unlinked_site; - __ movptr(rcx, Address(rax, __ delayed_value(sun_dyn_CallSiteImpl::target_offset_in_bytes, rcx))); - __ testptr(rcx, rcx); - __ jcc(Assembler::zero, handle_unlinked_site); - + __ movptr(rcx, Address(rax, __ delayed_value(java_dyn_CallSite::target_offset_in_bytes, rcx))); + __ null_check(rcx); __ prepare_to_jump_from_interpreted(); __ jump_to_method_handle_entry(rcx, rdx); - - // Initial calls come here... - __ bind(handle_unlinked_site); - __ pop(rcx); // remove return address pushed by prepare_invoke - - // box stacked arguments into an array for the bootstrap method - address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::bootstrap_invokedynamic); - __ restore_bcp(); // rsi must be correct for call_VM - __ call_VM(rax, entry, rax); - __ movl(rdi, rax); // protect bootstrap MH from prepare_invoke - - // recompute return address - __ restore_bcp(); // rsi must be correct for prepare_invoke - prepare_invoke(rax, rbx, -byte_no); // smashes rcx, rdx - // rax: CallSite object (f1) - // rbx: unused (f2) - // rdi: bootstrap MH - // rdx: flags - - // now load up the arglist, which has been neatly boxed - __ get_thread(rcx); - __ movptr(rdx, Address(rcx, JavaThread::vm_result_2_offset())); - __ movptr(Address(rcx, JavaThread::vm_result_2_offset()), NULL_WORD); - __ verify_oop(rdx); - // rdx = arglist - - // save SP now, before we add the bootstrap call to the stack - // We must preserve a fiction that the original arguments are outgoing, - // because the return sequence will reset the stack to this point - // and then pop all those arguments. It seems error-prone to use - // a different argument list size just for bootstrapping. - __ prepare_to_jump_from_interpreted(); - - // Now let's play adapter, pushing the real arguments on the stack. - __ pop(rbx); // return PC - __ push(rdi); // boot MH - __ push(rax); // call site - __ push(rdx); // arglist - __ push(rbx); // return PC, again - __ mov(rcx, rdi); - __ jump_to_method_handle_entry(rcx, rdx); } //----------------------------------------------------------------------------------------------------
--- a/src/cpu/x86/vm/templateTable_x86_64.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/templateTable_x86_64.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -203,18 +203,15 @@ __ jcc(Assembler::notEqual, fast_patch); __ get_method(scratch); // Let breakpoint table handling rewrite to quicker bytecode - __ call_VM(noreg, - CAST_FROM_FN_PTR(address, - InterpreterRuntime::set_original_bytecode_at), - scratch, r13, bc); + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), scratch, r13, bc); #ifndef ASSERT __ jmpb(patch_done); +#else + __ jmp(patch_done); +#endif __ bind(fast_patch); } -#else - __ jmp(patch_done); - __ bind(fast_patch); - } +#ifdef ASSERT Label okay; __ load_unsigned_byte(scratch, at_bcp(0)); __ cmpl(scratch, (int) Bytecodes::java_code(bytecode)); @@ -2054,26 +2051,28 @@ } } -void TemplateTable::resolve_cache_and_index(int byte_no, - Register Rcache, - Register index) { +void TemplateTable::resolve_cache_and_index(int byte_no, Register Rcache, Register index) { assert(byte_no == 1 || byte_no == 2, "byte_no out of range"); + bool is_invokedynamic = (bytecode() == Bytecodes::_invokedynamic); const Register temp = rbx; assert_different_registers(Rcache, index, temp); const int shift_count = (1 + byte_no) * BitsPerByte; Label resolved; - __ get_cache_and_index_at_bcp(Rcache, index, 1); - __ movl(temp, Address(Rcache, - index, Address::times_8, - constantPoolCacheOopDesc::base_offset() + - ConstantPoolCacheEntry::indices_offset())); - __ shrl(temp, shift_count); - // have we resolved this bytecode? - __ andl(temp, 0xFF); - __ cmpl(temp, (int) bytecode()); - __ jcc(Assembler::equal, resolved); + __ get_cache_and_index_at_bcp(Rcache, index, 1, is_invokedynamic); + if (is_invokedynamic) { + // we are resolved if the f1 field contains a non-null CallSite object + __ cmpptr(Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f1_offset()), (int32_t) NULL_WORD); + __ jcc(Assembler::notEqual, resolved); + } else { + __ movl(temp, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset())); + __ shrl(temp, shift_count); + // have we resolved this bytecode? + __ andl(temp, 0xFF); + __ cmpl(temp, (int) bytecode()); + __ jcc(Assembler::equal, resolved); + } // resolve first time through address entry; @@ -2090,6 +2089,9 @@ case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break; + case Bytecodes::_invokedynamic: + entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic); + break; default: ShouldNotReachHere(); break; @@ -2098,7 +2100,7 @@ __ call_VM(noreg, entry, temp); // Update registers with resolved info - __ get_cache_and_index_at_bcp(Rcache, index, 1); + __ get_cache_and_index_at_bcp(Rcache, index, 1, is_invokedynamic); __ bind(resolved); } @@ -2832,15 +2834,14 @@ ShouldNotReachHere(); } -void TemplateTable::prepare_invoke(Register method, - Register index, - int byte_no, - Bytecodes::Code code) { +void TemplateTable::prepare_invoke(Register method, Register index, int byte_no) { // determine flags + Bytecodes::Code code = bytecode(); const bool is_invokeinterface = code == Bytecodes::_invokeinterface; + const bool is_invokedynamic = code == Bytecodes::_invokedynamic; const bool is_invokevirtual = code == Bytecodes::_invokevirtual; const bool is_invokespecial = code == Bytecodes::_invokespecial; - const bool load_receiver = code != Bytecodes::_invokestatic; + const bool load_receiver = (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic); const bool receiver_null_check = is_invokespecial; const bool save_flags = is_invokeinterface || is_invokevirtual; // setup registers & access constant pool cache @@ -2858,9 +2859,13 @@ __ movl(recv, flags); __ andl(recv, 0xFF); if (TaggedStackInterpreter) __ shll(recv, 1); // index*2 - __ movptr(recv, Address(rsp, recv, Address::times_8, - -Interpreter::expr_offset_in_bytes(1))); - __ verify_oop(recv); + Address recv_addr(rsp, recv, Address::times_8, -Interpreter::expr_offset_in_bytes(1)); + if (is_invokedynamic) { + __ lea(recv, recv_addr); + } else { + __ movptr(recv, recv_addr); + __ verify_oop(recv); + } } // do null check if needed @@ -2878,10 +2883,14 @@ ConstantPoolCacheEntry::verify_tosBits(); // load return address { - ExternalAddress return_5((address)Interpreter::return_5_addrs_by_index_table()); - ExternalAddress return_3((address)Interpreter::return_3_addrs_by_index_table()); - __ lea(rscratch1, (is_invokeinterface ? return_5 : return_3)); - __ movptr(flags, Address(rscratch1, flags, Address::times_8)); + address table_addr; + if (is_invokeinterface || is_invokedynamic) + table_addr = (address)Interpreter::return_5_addrs_by_index_table(); + else + table_addr = (address)Interpreter::return_3_addrs_by_index_table(); + ExternalAddress table(table_addr); + __ lea(rscratch1, table); + __ movptr(flags, Address(rscratch1, flags, Address::times_ptr)); } // push return address @@ -2947,7 +2956,7 @@ void TemplateTable::invokevirtual(int byte_no) { transition(vtos, vtos); - prepare_invoke(rbx, noreg, byte_no, bytecode()); + prepare_invoke(rbx, noreg, byte_no); // rbx: index // rcx: receiver @@ -2959,7 +2968,7 @@ void TemplateTable::invokespecial(int byte_no) { transition(vtos, vtos); - prepare_invoke(rbx, noreg, byte_no, bytecode()); + prepare_invoke(rbx, noreg, byte_no); // do the call __ verify_oop(rbx); __ profile_call(rax); @@ -2969,7 +2978,7 @@ void TemplateTable::invokestatic(int byte_no) { transition(vtos, vtos); - prepare_invoke(rbx, noreg, byte_no, bytecode()); + prepare_invoke(rbx, noreg, byte_no); // do the call __ verify_oop(rbx); __ profile_call(rax); @@ -2983,7 +2992,7 @@ void TemplateTable::invokeinterface(int byte_no) { transition(vtos, vtos); - prepare_invoke(rax, rbx, byte_no, bytecode()); + prepare_invoke(rax, rbx, byte_no); // rax: Interface // rbx: index @@ -3072,7 +3081,24 @@ return; } - __ stop("invokedynamic NYI");//6815692// + prepare_invoke(rax, rbx, byte_no); + + // rax: CallSite object (f1) + // rbx: unused (f2) + // rcx: receiver address + // rdx: flags (unused) + + if (ProfileInterpreter) { + Label L; + // %%% should make a type profile for any invokedynamic that takes a ref argument + // profile this call + __ profile_call(r13); + } + + __ movptr(rcx, Address(rax, __ delayed_value(java_dyn_CallSite::target_offset_in_bytes, rcx))); + __ null_check(rcx); + __ prepare_to_jump_from_interpreted(); + __ jump_to_method_handle_entry(rcx, rdx); } @@ -3212,17 +3238,19 @@ __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code) __ store_klass_gap(rax, rcx); // zero klass gap for compressed oops __ store_klass(rax, rsi); // store klass last + + { + SkipIfEqual skip(_masm, &DTraceAllocProbes, false); + // Trigger dtrace event for fastpath + __ push(atos); // save the return value + __ call_VM_leaf( + CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax); + __ pop(atos); // restore the return value + + } __ jmp(done); } - { - SkipIfEqual skip(_masm, &DTraceAllocProbes, false); - // Trigger dtrace event for fastpath - __ push(atos); // save the return value - __ call_VM_leaf( - CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax); - __ pop(atos); // restore the return value - } // slow case __ bind(slow_case);
--- a/src/cpu/x86/vm/templateTable_x86_64.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/templateTable_x86_64.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -22,8 +22,7 @@ * */ - static void prepare_invoke(Register method, Register index, int byte_no, - Bytecodes::Code code); + static void prepare_invoke(Register method, Register index, int byte_no); static void invokevirtual_helper(Register index, Register recv, Register flags); static void volatile_barrier(Assembler::Membar_mask_bits order_constraint);
--- a/src/cpu/x86/vm/vm_version_x86.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/vm_version_x86.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -255,6 +255,8 @@ if (!VM_Version::supports_sse2()) { vm_exit_during_initialization("Unknown x64 processor: SSE2 not supported"); } + // in 64 bit the use of SSE2 is the minimum + if (UseSSE < 2) UseSSE = 2; #endif // If the OS doesn't support SSE, we can't use this feature even if the HW does
--- a/src/cpu/x86/vm/x86_32.ad Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/x86_32.ad Wed Mar 24 17:16:33 2010 -0700 @@ -235,6 +235,11 @@ //----------SOURCE BLOCK------------------------------------------------------- // This is a block of C++ code which provides values, functions, and // definitions necessary in the rest of the architecture description +source_hpp %{ +// Must be visible to the DFA in dfa_x86_32.cpp +extern bool is_operand_hi32_zero(Node* n); +%} + source %{ #define RELOC_IMM32 Assembler::imm_operand #define RELOC_DISP32 Assembler::disp32_operand @@ -268,22 +273,36 @@ static jlong *float_signflip_pool = double_quadword(&fp_signmask_pool[3*2], CONST64(0x8000000080000000), CONST64(0x8000000080000000)); static jlong *double_signflip_pool = double_quadword(&fp_signmask_pool[4*2], CONST64(0x8000000000000000), CONST64(0x8000000000000000)); +// Offset hacking within calls. +static int pre_call_FPU_size() { + if (Compile::current()->in_24_bit_fp_mode()) + return 6; // fldcw + return 0; +} + +static int preserve_SP_size() { + return LP64_ONLY(1 +) 2; // [rex,] op, rm(reg/reg) +} + // !!!!! Special hack to get all type of calls to specify the byte offset // from the start of the call to the point where the return address // will point. int MachCallStaticJavaNode::ret_addr_offset() { - return 5 + (Compile::current()->in_24_bit_fp_mode() ? 6 : 0); // 5 bytes from start of call to where return address points + int offset = 5 + pre_call_FPU_size(); // 5 bytes from start of call to where return address points + if (_method_handle_invoke) + offset += preserve_SP_size(); + return offset; } int MachCallDynamicJavaNode::ret_addr_offset() { - return 10 + (Compile::current()->in_24_bit_fp_mode() ? 6 : 0); // 10 bytes from start of call to where return address points + return 10 + pre_call_FPU_size(); // 10 bytes from start of call to where return address points } static int sizeof_FFree_Float_Stack_All = -1; int MachCallRuntimeNode::ret_addr_offset() { assert(sizeof_FFree_Float_Stack_All != -1, "must have been emitted already"); - return sizeof_FFree_Float_Stack_All + 5 + (Compile::current()->in_24_bit_fp_mode() ? 6 : 0); + return sizeof_FFree_Float_Stack_All + 5 + pre_call_FPU_size(); } // Indicate if the safepoint node needs the polling page as an input. @@ -299,8 +318,16 @@ // The address of the call instruction needs to be 4-byte aligned to // ensure that it does not span a cache line so that it can be patched. int CallStaticJavaDirectNode::compute_padding(int current_offset) const { - if (Compile::current()->in_24_bit_fp_mode()) - current_offset += 6; // skip fldcw in pre_call_FPU, if any + current_offset += pre_call_FPU_size(); // skip fldcw, if any + current_offset += 1; // skip call opcode byte + return round_to(current_offset, alignment_required()) - current_offset; +} + +// The address of the call instruction needs to be 4-byte aligned to +// ensure that it does not span a cache line so that it can be patched. +int CallStaticJavaHandleNode::compute_padding(int current_offset) const { + current_offset += pre_call_FPU_size(); // skip fldcw, if any + current_offset += preserve_SP_size(); // skip mov rbp, rsp current_offset += 1; // skip call opcode byte return round_to(current_offset, alignment_required()) - current_offset; } @@ -308,8 +335,7 @@ // The address of the call instruction needs to be 4-byte aligned to // ensure that it does not span a cache line so that it can be patched. int CallDynamicJavaDirectNode::compute_padding(int current_offset) const { - if (Compile::current()->in_24_bit_fp_mode()) - current_offset += 6; // skip fldcw in pre_call_FPU, if any + current_offset += pre_call_FPU_size(); // skip fldcw, if any current_offset += 5; // skip MOV instruction current_offset += 1; // skip call opcode byte return round_to(current_offset, alignment_required()) - current_offset; @@ -1460,6 +1486,25 @@ return RegMask(); } +const RegMask Matcher::method_handle_invoke_SP_save_mask() { + return EBP_REG_mask; +} + +// Returns true if the high 32 bits of the value is known to be zero. +bool is_operand_hi32_zero(Node* n) { + int opc = n->Opcode(); + if (opc == Op_LoadUI2L) { + return true; + } + if (opc == Op_AndL) { + Node* o2 = n->in(2); + if (o2->is_Con() && (o2->get_long() & 0xFFFFFFFF00000000LL) == 0LL) { + return true; + } + } + return false; +} + %} //----------ENCODING BLOCK----------------------------------------------------- @@ -1772,10 +1817,13 @@ enc_class pre_call_FPU %{ // If method sets FPU control word restore it here + debug_only(int off0 = cbuf.code_size()); if( Compile::current()->in_24_bit_fp_mode() ) { MacroAssembler masm(&cbuf); masm.fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); } + debug_only(int off1 = cbuf.code_size()); + assert(off1 - off0 == pre_call_FPU_size(), "correct size prediction"); %} enc_class post_call_FPU %{ @@ -1786,6 +1834,21 @@ } %} + enc_class preserve_SP %{ + debug_only(int off0 = cbuf.code_size()); + MacroAssembler _masm(&cbuf); + // RBP is preserved across all calls, even compiled calls. + // Use it to preserve RSP in places where the callee might change the SP. + __ movptr(rbp, rsp); + debug_only(int off1 = cbuf.code_size()); + assert(off1 - off0 == preserve_SP_size(), "correct size prediction"); + %} + + enc_class restore_SP %{ + MacroAssembler _masm(&cbuf); + __ movptr(rsp, rbp); + %} + enc_class Java_Static_Call (method meth) %{ // JAVA STATIC CALL // CALL to fixup routine. Fixup routine uses ScopeDesc info to determine // who we intended to call. @@ -8556,6 +8619,63 @@ ins_pipe( pipe_slow ); %} +// Multiply Register Long where the left operand's high 32 bits are zero +instruct mulL_eReg_lhi0(eADXRegL dst, eRegL src, eRegI tmp, eFlagsReg cr) %{ + predicate(is_operand_hi32_zero(n->in(1))); + match(Set dst (MulL dst src)); + effect(KILL cr, TEMP tmp); + ins_cost(2*100+2*400); +// Basic idea: lo(result) = lo(x_lo * y_lo) +// hi(result) = hi(x_lo * y_lo) + lo(x_lo * y_hi) where lo(x_hi * y_lo) = 0 because x_hi = 0 + format %{ "MOV $tmp,$src.hi\n\t" + "IMUL $tmp,EAX\n\t" + "MUL EDX:EAX,$src.lo\n\t" + "ADD EDX,$tmp" %} + ins_encode %{ + __ movl($tmp$$Register, HIGH_FROM_LOW($src$$Register)); + __ imull($tmp$$Register, rax); + __ mull($src$$Register); + __ addl(rdx, $tmp$$Register); + %} + ins_pipe( pipe_slow ); +%} + +// Multiply Register Long where the right operand's high 32 bits are zero +instruct mulL_eReg_rhi0(eADXRegL dst, eRegL src, eRegI tmp, eFlagsReg cr) %{ + predicate(is_operand_hi32_zero(n->in(2))); + match(Set dst (MulL dst src)); + effect(KILL cr, TEMP tmp); + ins_cost(2*100+2*400); +// Basic idea: lo(result) = lo(x_lo * y_lo) +// hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) where lo(x_lo * y_hi) = 0 because y_hi = 0 + format %{ "MOV $tmp,$src.lo\n\t" + "IMUL $tmp,EDX\n\t" + "MUL EDX:EAX,$src.lo\n\t" + "ADD EDX,$tmp" %} + ins_encode %{ + __ movl($tmp$$Register, $src$$Register); + __ imull($tmp$$Register, rdx); + __ mull($src$$Register); + __ addl(rdx, $tmp$$Register); + %} + ins_pipe( pipe_slow ); +%} + +// Multiply Register Long where the left and the right operands' high 32 bits are zero +instruct mulL_eReg_hi0(eADXRegL dst, eRegL src, eFlagsReg cr) %{ + predicate(is_operand_hi32_zero(n->in(1)) && is_operand_hi32_zero(n->in(2))); + match(Set dst (MulL dst src)); + effect(KILL cr); + ins_cost(1*400); +// Basic idea: lo(result) = lo(x_lo * y_lo) +// hi(result) = hi(x_lo * y_lo) where lo(x_hi * y_lo) = 0 and lo(x_lo * y_hi) = 0 because x_hi = 0 and y_hi = 0 + format %{ "MUL EDX:EAX,$src.lo\n\t" %} + ins_encode %{ + __ mull($src$$Register); + %} + ins_pipe( pipe_slow ); +%} + // Multiply Register Long by small constant instruct mulL_eReg_con(eADXRegL dst, immL_127 src, eRegI tmp, eFlagsReg cr) %{ match(Set dst (MulL dst src)); @@ -13406,6 +13526,7 @@ // compute_padding() functions will have to be adjusted. instruct CallStaticJavaDirect(method meth) %{ match(CallStaticJava); + predicate(! ((CallStaticJavaNode*)n)->is_method_handle_invoke()); effect(USE meth); ins_cost(300); @@ -13420,6 +13541,30 @@ ins_alignment(4); %} +// Call Java Static Instruction (method handle version) +// Note: If this code changes, the corresponding ret_addr_offset() and +// compute_padding() functions will have to be adjusted. +instruct CallStaticJavaHandle(method meth, eBPRegP ebp) %{ + match(CallStaticJava); + predicate(((CallStaticJavaNode*)n)->is_method_handle_invoke()); + effect(USE meth); + // EBP is saved by all callees (for interpreter stack correction). + // We use it here for a similar purpose, in {preserve,restore}_SP. + + ins_cost(300); + format %{ "CALL,static/MethodHandle " %} + opcode(0xE8); /* E8 cd */ + ins_encode( pre_call_FPU, + preserve_SP, + Java_Static_Call( meth ), + restore_SP, + call_epilog, + post_call_FPU ); + ins_pipe( pipe_slow ); + ins_pc_relative(1); + ins_alignment(4); +%} + // Call Java Dynamic Instruction // Note: If this code changes, the corresponding ret_addr_offset() and // compute_padding() functions will have to be adjusted.
--- a/src/cpu/x86/vm/x86_64.ad Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/x86/vm/x86_64.ad Wed Mar 24 17:16:33 2010 -0700 @@ -551,12 +551,19 @@ #define __ _masm. +static int preserve_SP_size() { + return LP64_ONLY(1 +) 2; // [rex,] op, rm(reg/reg) +} + // !!!!! Special hack to get all types of calls to specify the byte offset // from the start of the call to the point where the return address // will point. int MachCallStaticJavaNode::ret_addr_offset() { - return 5; // 5 bytes from start of call to where return address points + int offset = 5; // 5 bytes from start of call to where return address points + if (_method_handle_invoke) + offset += preserve_SP_size(); + return offset; } int MachCallDynamicJavaNode::ret_addr_offset() @@ -589,6 +596,15 @@ // The address of the call instruction needs to be 4-byte aligned to // ensure that it does not span a cache line so that it can be patched. +int CallStaticJavaHandleNode::compute_padding(int current_offset) const +{ + current_offset += preserve_SP_size(); // skip mov rbp, rsp + current_offset += 1; // skip call opcode byte + return round_to(current_offset, alignment_required()) - current_offset; +} + +// The address of the call instruction needs to be 4-byte aligned to +// ensure that it does not span a cache line so that it can be patched. int CallDynamicJavaDirectNode::compute_padding(int current_offset) const { current_offset += 11; // skip movq instruction + call opcode byte @@ -2113,6 +2129,10 @@ return LONG_RDX_REG_mask; } +const RegMask Matcher::method_handle_invoke_SP_save_mask() { + return PTR_RBP_REG_mask; +} + static Address build_address(int b, int i, int s, int d) { Register index = as_Register(i); Address::ScaleFactor scale = (Address::ScaleFactor)s; @@ -2608,6 +2628,21 @@ RELOC_DISP32); %} + enc_class preserve_SP %{ + debug_only(int off0 = cbuf.code_size()); + MacroAssembler _masm(&cbuf); + // RBP is preserved across all calls, even compiled calls. + // Use it to preserve RSP in places where the callee might change the SP. + __ movptr(rbp, rsp); + debug_only(int off1 = cbuf.code_size()); + assert(off1 - off0 == preserve_SP_size(), "correct size prediction"); + %} + + enc_class restore_SP %{ + MacroAssembler _masm(&cbuf); + __ movptr(rsp, rbp); + %} + enc_class Java_Static_Call(method meth) %{ // JAVA STATIC CALL @@ -12526,9 +12561,9 @@ // Call Java Static Instruction // Note: If this code changes, the corresponding ret_addr_offset() and // compute_padding() functions will have to be adjusted. -instruct CallStaticJavaDirect(method meth) -%{ +instruct CallStaticJavaDirect(method meth) %{ match(CallStaticJava); + predicate(!((CallStaticJavaNode*) n)->is_method_handle_invoke()); effect(USE meth); ins_cost(300); @@ -12540,6 +12575,28 @@ ins_alignment(4); %} +// Call Java Static Instruction (method handle version) +// Note: If this code changes, the corresponding ret_addr_offset() and +// compute_padding() functions will have to be adjusted. +instruct CallStaticJavaHandle(method meth, rbp_RegP rbp) %{ + match(CallStaticJava); + predicate(((CallStaticJavaNode*) n)->is_method_handle_invoke()); + effect(USE meth); + // RBP is saved by all callees (for interpreter stack correction). + // We use it here for a similar purpose, in {preserve,restore}_SP. + + ins_cost(300); + format %{ "call,static/MethodHandle " %} + opcode(0xE8); /* E8 cd */ + ins_encode(preserve_SP, + Java_Static_Call(meth), + restore_SP, + call_epilog); + ins_pipe(pipe_slow); + ins_pc_relative(1); + ins_alignment(4); +%} + // Call Java Dynamic Instruction // Note: If this code changes, the corresponding ret_addr_offset() and // compute_padding() functions will have to be adjusted.
--- a/src/cpu/zero/vm/cppInterpreter_zero.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/zero/vm/cppInterpreter_zero.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,6 +1,6 @@ /* * Copyright 2003-2007 Sun Microsystems, Inc. All Rights Reserved. - * Copyright 2007, 2008, 2009 Red Hat, Inc. + * Copyright 2007, 2008, 2009, 2010 Red Hat, Inc. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -145,7 +145,7 @@ } else if (istate->msg() == BytecodeInterpreter::return_from_method) { // Copy the result into the caller's frame - result_slots = type2size[method->result_type()]; + result_slots = type2size[result_type_of(method)]; assert(result_slots >= 0 && result_slots <= 2, "what?"); result = istate->stack() + result_slots; break; @@ -204,6 +204,20 @@ goto unwind_and_return; } + // Update the invocation counter + if ((UseCompiler || CountCompiledCalls) && !method->is_synchronized()) { + thread->set_do_not_unlock(); + InvocationCounter *counter = method->invocation_counter(); + counter->increment(); + if (counter->reached_InvocationLimit()) { + CALL_VM_NOCHECK( + InterpreterRuntime::frequency_counter_overflow(thread, NULL)); + if (HAS_PENDING_EXCEPTION) + goto unwind_and_return; + } + thread->clr_do_not_unlock(); + } + // Lock if necessary BasicObjectLock *monitor; monitor = NULL; @@ -231,7 +245,7 @@ if (handlerAddr == NULL) { CALL_VM_NOCHECK(InterpreterRuntime::prepare_native_call(thread, method)); if (HAS_PENDING_EXCEPTION) - goto unwind_and_return; + goto unlock_unwind_and_return; handlerAddr = method->signature_handler(); assert(handlerAddr != NULL, "eh?"); @@ -240,7 +254,7 @@ CALL_VM_NOCHECK(handlerAddr = InterpreterRuntime::slow_signature_handler(thread, method, NULL,NULL)); if (HAS_PENDING_EXCEPTION) - goto unwind_and_return; + goto unlock_unwind_and_return; } handler = \ InterpreterRuntime::SignatureHandler::from_handlerAddr(handlerAddr); @@ -351,10 +365,10 @@ // Reset handle block thread->active_handles()->clear(); - // Unlock if necessary. It seems totally wrong that this - // is skipped in the event of an exception but apparently - // the template interpreter does this so we do too. - if (monitor && !HAS_PENDING_EXCEPTION) { + unlock_unwind_and_return: + + // Unlock if necessary + if (monitor) { BasicLock *lock = monitor->lock(); markOop header = lock->displaced_header(); oop rcvr = monitor->obj(); @@ -380,9 +394,10 @@ // Push our result if (!HAS_PENDING_EXCEPTION) { - stack->set_sp(stack->sp() - type2size[method->result_type()]); + BasicType type = result_type_of(method); + stack->set_sp(stack->sp() - type2size[type]); - switch (method->result_type()) { + switch (type) { case T_VOID: break; @@ -693,6 +708,26 @@ return i; } +BasicType CppInterpreter::result_type_of(methodOop method) { + BasicType t; + switch (method->result_index()) { + case 0 : t = T_BOOLEAN; break; + case 1 : t = T_CHAR; break; + case 2 : t = T_BYTE; break; + case 3 : t = T_SHORT; break; + case 4 : t = T_INT; break; + case 5 : t = T_LONG; break; + case 6 : t = T_VOID; break; + case 7 : t = T_FLOAT; break; + case 8 : t = T_DOUBLE; break; + case 9 : t = T_OBJECT; break; + default: ShouldNotReachHere(); + } + assert(AbstractInterpreter::BasicType_as_index(t) == method->result_index(), + "out of step with AbstractInterpreter::BasicType_as_index"); + return t; +} + address InterpreterGenerator::generate_empty_entry() { if (!UseFastEmptyMethods) return NULL;
--- a/src/cpu/zero/vm/cppInterpreter_zero.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/zero/vm/cppInterpreter_zero.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,6 +1,6 @@ /* * Copyright 1997-2007 Sun Microsystems, Inc. All Rights Reserved. - * Copyright 2007, 2008 Red Hat, Inc. + * Copyright 2007, 2008, 2010 Red Hat, Inc. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -41,3 +41,7 @@ private: // Stack overflow checks static bool stack_overflow_imminent(JavaThread *thread); + + private: + // Fast result type determination + static BasicType result_type_of(methodOop method);
--- a/src/cpu/zero/vm/frame_zero.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/zero/vm/frame_zero.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -36,11 +36,8 @@ return zeroframe()->is_interpreter_frame(); } -bool frame::is_fake_stub_frame() const { - return zeroframe()->is_fake_stub_frame(); -} - frame frame::sender_for_entry_frame(RegisterMap *map) const { + assert(zeroframe()->is_entry_frame(), "wrong type of frame"); assert(map != NULL, "map must be set"); assert(!entry_frame_is_first(), "next Java fp must be non zero"); assert(entry_frame_call_wrapper()->anchor()->last_Java_sp() == sender_sp(), @@ -50,15 +47,10 @@ return frame(sender_sp(), sp() + 1); } -frame frame::sender_for_interpreter_frame(RegisterMap *map) const { - return frame(sender_sp(), sp() + 1); -} - -frame frame::sender_for_compiled_frame(RegisterMap *map) const { - return frame(sender_sp(), sp() + 1); -} - -frame frame::sender_for_fake_stub_frame(RegisterMap *map) const { +frame frame::sender_for_nonentry_frame(RegisterMap *map) const { + assert(zeroframe()->is_interpreter_frame() || + zeroframe()->is_shark_frame() || + zeroframe()->is_fake_stub_frame(), "wrong type of frame"); return frame(sender_sp(), sp() + 1); } @@ -69,17 +61,8 @@ if (is_entry_frame()) return sender_for_entry_frame(map); - - if (is_interpreted_frame()) - return sender_for_interpreter_frame(map); - - if (is_compiled_frame()) - return sender_for_compiled_frame(map); - - if (is_fake_stub_frame()) - return sender_for_fake_stub_frame(map); - - ShouldNotReachHere(); + else + return sender_for_nonentry_frame(map); } #ifdef CC_INTERP
--- a/src/cpu/zero/vm/frame_zero.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/zero/vm/frame_zero.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -65,10 +65,7 @@ } public: - bool is_fake_stub_frame() const; - - public: - frame sender_for_fake_stub_frame(RegisterMap* map) const; + frame sender_for_nonentry_frame(RegisterMap* map) const; public: void zero_print_on_error(int index,
--- a/src/cpu/zero/vm/globals_zero.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/zero/vm/globals_zero.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,6 +1,6 @@ /* * Copyright 2000-2006 Sun Microsystems, Inc. All Rights Reserved. - * Copyright 2007, 2008, 2009 Red Hat, Inc. + * Copyright 2007, 2008, 2009, 2010 Red Hat, Inc. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -23,10 +23,8 @@ * */ -// // Set the default values for platform dependent flags used by the // runtime system. See globals.hpp for details of what they do. -// define_pd_global(bool, ConvertSleepToYield, true); define_pd_global(bool, ShareVtableStubs, true); @@ -37,19 +35,12 @@ define_pd_global(bool, UncommonNullCast, true); define_pd_global(intx, CodeEntryAlignment, 32); -define_pd_global(uintx, TLABSize, 0); -#ifdef _LP64 -define_pd_global(uintx, NewSize, ScaleForWordSize(2048 * K)); -#else -define_pd_global(uintx, NewSize, ScaleForWordSize(1024 * K)); -#endif // _LP64 define_pd_global(intx, InlineFrequencyCount, 100); -define_pd_global(intx, InlineSmallCode, 1000); define_pd_global(intx, PreInflateSpin, 10); define_pd_global(intx, StackYellowPages, 2); define_pd_global(intx, StackRedPages, 1); -define_pd_global(intx, StackShadowPages, 3 LP64_ONLY(+3) DEBUG_ONLY(+3)); +define_pd_global(intx, StackShadowPages, 5 LP64_ONLY(+1) DEBUG_ONLY(+3)); define_pd_global(bool, RewriteBytecodes, true); define_pd_global(bool, RewriteFrequentPairs, true);
--- a/src/cpu/zero/vm/interpreter_zero.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/zero/vm/interpreter_zero.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,6 +1,6 @@ /* * Copyright 2003-2007 Sun Microsystems, Inc. All Rights Reserved. - * Copyright 2007, 2008 Red Hat, Inc. + * Copyright 2007, 2008, 2009, 2010 Red Hat, Inc. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -47,6 +47,10 @@ return ShouldNotCallThisEntry(); } +bool AbstractInterpreter::can_be_compiled(methodHandle m) { + return true; +} + int AbstractInterpreter::size_activation(methodOop method, int tempcount, int popframe_extra_args,
--- a/src/cpu/zero/vm/sharedRuntime_zero.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/zero/vm/sharedRuntime_zero.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,6 +1,6 @@ /* * Copyright 2003-2007 Sun Microsystems, Inc. All Rights Reserved. - * Copyright 2007, 2008 Red Hat, Inc. + * Copyright 2007, 2008, 2009, 2010 Red Hat, Inc. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -47,8 +47,10 @@ int total_args_passed, int comp_args_on_stack, const BasicType *sig_bt, - const VMRegPair *regs) { - return new AdapterHandlerEntry( + const VMRegPair *regs, + AdapterFingerPrint *fingerprint) { + return AdapterHandlerLibrary::new_entry( + fingerprint, ShouldNotCallThisStub(), ShouldNotCallThisStub(), ShouldNotCallThisStub()); @@ -61,7 +63,14 @@ BasicType *in_sig_bt, VMRegPair *in_regs, BasicType ret_type) { +#ifdef SHARK + return SharkCompiler::compiler()->generate_native_wrapper(masm, + method, + in_sig_bt, + ret_type); +#else ShouldNotCallThis(); +#endif // SHARK } int Deoptimization::last_frame_adjust(int callee_parameters,
--- a/src/cpu/zero/vm/sharkFrame_zero.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/cpu/zero/vm/sharkFrame_zero.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,6 +1,6 @@ /* * Copyright 2003-2007 Sun Microsystems, Inc. All Rights Reserved. - * Copyright 2008 Red Hat, Inc. + * Copyright 2008, 2009 Red Hat, Inc. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -41,7 +41,7 @@ // | ... | class SharkFrame : public ZeroFrame { - friend class SharkFunction; + friend class SharkStack; private: SharkFrame() : ZeroFrame() {
--- a/src/os/linux/vm/os_linux.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/os/linux/vm/os_linux.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -223,8 +223,8 @@ "environment on Linux when /proc filesystem is not mounted."; void os::Linux::initialize_system_info() { - _processor_count = sysconf(_SC_NPROCESSORS_CONF); - if (_processor_count == 1) { + set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); + if (processor_count() == 1) { pid_t pid = os::Linux::gettid(); char fname[32]; jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); @@ -236,7 +236,7 @@ } } _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); - assert(_processor_count > 0, "linux error"); + assert(processor_count() > 0, "linux error"); } void os::init_system_properties_values() { @@ -4683,6 +4683,7 @@ // Return immediately if a permit is available. if (_counter > 0) { _counter = 0 ; + OrderAccess::fence(); return ; } @@ -4725,6 +4726,7 @@ _counter = 0; status = pthread_mutex_unlock(_mutex); assert (status == 0, "invariant") ; + OrderAccess::fence(); return; } @@ -4765,6 +4767,7 @@ jt->java_suspend_self(); } + OrderAccess::fence(); } void Parker::unpark() {
--- a/src/os/solaris/dtrace/libjvm_db.c Mon Mar 15 15:51:36 2010 -0400 +++ b/src/os/solaris/dtrace/libjvm_db.c Wed Mar 24 17:16:33 2010 -0700 @@ -937,54 +937,56 @@ return err; } -static int -scopeDesc_chain(Nmethod_t *N) -{ +static int scopeDesc_chain(Nmethod_t *N) { int32_t decode_offset = 0; int32_t err; - if (debug > 2) - fprintf(stderr, "\t scopeDesc_chain: BEGIN\n"); + if (debug > 2) { + fprintf(stderr, "\t scopeDesc_chain: BEGIN\n"); + } err = ps_pread(N->J->P, N->pc_desc + OFFSET_PcDesc_scope_decode_offset, &decode_offset, SZ32); CHECK_FAIL(err); while (decode_offset > 0) { - if (debug > 2) - fprintf(stderr, "\t scopeDesc_chain: decode_offset: %#x\n", decode_offset); + Vframe_t *vf = &N->vframes[N->vf_cnt]; - Vframe_t *vf = &N->vframes[N->vf_cnt]; + if (debug > 2) { + fprintf(stderr, "\t scopeDesc_chain: decode_offset: %#x\n", decode_offset); + } + + err = scope_desc_at(N, decode_offset, vf); + CHECK_FAIL(err); - err = scope_desc_at(N, decode_offset, vf); - CHECK_FAIL(err); + if (vf->methodIdx > N->oops_len) { + fprintf(stderr, "\t scopeDesc_chain: (methodIdx > oops_len) !\n"); + return -1; + } + err = read_pointer(N->J, N->nm + N->oops_beg + (vf->methodIdx-1)*POINTER_SIZE, + &vf->methodOop); + CHECK_FAIL(err); - if (vf->methodIdx > N->oops_len) { - fprintf(stderr, "\t scopeDesc_chain: (methodIdx > oops_len) !\n"); - return -1; - } - err = read_pointer(N->J, N->nm + N->oops_beg + (vf->methodIdx-1)*POINTER_SIZE, - &vf->methodOop); + if (vf->methodOop) { + N->vf_cnt++; + err = line_number_from_bci(N->J, vf); CHECK_FAIL(err); - - if (vf->methodOop) { - N->vf_cnt++; - err = line_number_from_bci(N->J, vf); - CHECK_FAIL(err); - if (debug > 2) { - fprintf(stderr, "\t scopeDesc_chain: methodOop: %#8llx, line: %ld\n", - vf->methodOop, vf->line); - } + if (debug > 2) { + fprintf(stderr, "\t scopeDesc_chain: methodOop: %#8llx, line: %ld\n", + vf->methodOop, vf->line); } - decode_offset = vf->sender_decode_offset; + } + decode_offset = vf->sender_decode_offset; } - if (debug > 2) - fprintf(stderr, "\t scopeDesc_chain: END \n\n"); + if (debug > 2) { + fprintf(stderr, "\t scopeDesc_chain: END \n\n"); + } return PS_OK; fail: - if (debug) - fprintf(stderr, "\t scopeDesc_chain: FAIL \n\n"); + if (debug) { + fprintf(stderr, "\t scopeDesc_chain: FAIL \n\n"); + } return err; }
--- a/src/os/solaris/vm/os_solaris.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/os/solaris/vm/os_solaris.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -457,7 +457,7 @@ void os::Solaris::initialize_system_info() { - _processor_count = sysconf(_SC_NPROCESSORS_CONF); + set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); _processors_online = sysconf (_SC_NPROCESSORS_ONLN); _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); } @@ -5803,6 +5803,7 @@ // Return immediately if a permit is available. if (_counter > 0) { _counter = 0 ; + OrderAccess::fence(); return ; } @@ -5846,6 +5847,7 @@ _counter = 0; status = os::Solaris::mutex_unlock(_mutex); assert (status == 0, "invariant") ; + OrderAccess::fence(); return; } @@ -5892,6 +5894,7 @@ jt->java_suspend_self(); } + OrderAccess::fence(); } void Parker::unpark() {
--- a/src/os/windows/vm/os_windows.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/os/windows/vm/os_windows.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -142,6 +142,9 @@ } #ifndef _WIN64 +// previous UnhandledExceptionFilter, if there is one +static LPTOP_LEVEL_EXCEPTION_FILTER prev_uef_handler = NULL; + LONG WINAPI Handle_FLT_Exception(struct _EXCEPTION_POINTERS* exceptionInfo); #endif void os::init_system_properties_values() { @@ -260,7 +263,8 @@ } #ifndef _WIN64 - SetUnhandledExceptionFilter(Handle_FLT_Exception); + // set our UnhandledExceptionFilter and save any previous one + prev_uef_handler = SetUnhandledExceptionFilter(Handle_FLT_Exception); #endif // Done @@ -1969,7 +1973,7 @@ #ifndef _WIN64 //----------------------------------------------------------------------------- LONG WINAPI Handle_FLT_Exception(struct _EXCEPTION_POINTERS* exceptionInfo) { - // handle exception caused by native mothod modifying control word + // handle exception caused by native method modifying control word PCONTEXT ctx = exceptionInfo->ContextRecord; DWORD exception_code = exceptionInfo->ExceptionRecord->ExceptionCode; @@ -1990,6 +1994,13 @@ return EXCEPTION_CONTINUE_EXECUTION; } } + + if (prev_uef_handler != NULL) { + // We didn't handle this exception so pass it to the previous + // UnhandledExceptionFilter. + return (prev_uef_handler)(exceptionInfo); + } + return EXCEPTION_CONTINUE_SEARCH; } #else //_WIN64 @@ -3150,7 +3161,7 @@ _vm_allocation_granularity = si.dwAllocationGranularity; _processor_type = si.dwProcessorType; _processor_level = si.wProcessorLevel; - _processor_count = si.dwNumberOfProcessors; + set_processor_count(si.dwNumberOfProcessors); MEMORYSTATUSEX ms; ms.dwLength = sizeof(ms);
--- a/src/os_cpu/linux_x86/vm/globals_linux_x86.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/os_cpu/linux_x86/vm/globals_linux_x86.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -22,10 +22,9 @@ * */ -// // Sets the default values for platform dependent flags used by the runtime system. // (see globals.hpp) -// + define_pd_global(bool, DontYieldALot, false); #ifdef AMD64 define_pd_global(intx, ThreadStackSize, 1024); // 0 => use system default @@ -39,11 +38,10 @@ #endif // AMD64 define_pd_global(intx, CompilerThreadStackSize, 0); -define_pd_global(intx, SurvivorRatio, 8); -define_pd_global(uintx, JVMInvokeMethodSlack, 8192); +define_pd_global(uintx,JVMInvokeMethodSlack, 8192); // Only used on 64 bit platforms -define_pd_global(uintx, HeapBaseMinAddress, 2*G); +define_pd_global(uintx,HeapBaseMinAddress, 2*G); // Only used on 64 bit Windows platforms define_pd_global(bool, UseVectoredExceptions, false);
--- a/src/os_cpu/linux_zero/vm/os_linux_zero.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/os_cpu/linux_zero/vm/os_linux_zero.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,6 +1,6 @@ /* * Copyright 2003-2007 Sun Microsystems, Inc. All Rights Reserved. - * Copyright 2007, 2008 Red Hat, Inc. + * Copyright 2007, 2008, 2009, 2010 Red Hat, Inc. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -239,7 +239,21 @@ } bool os::is_allocatable(size_t bytes) { - ShouldNotCallThis(); +#ifdef _LP64 + return true; +#else + if (bytes < 2 * G) { + return true; + } + + char* addr = reserve_memory(bytes, NULL); + + if (addr != NULL) { + release_memory(addr, bytes); + } + + return addr != NULL; +#endif // _LP64 } ///////////////////////////////////////////////////////////////////////////////
--- a/src/os_cpu/linux_zero/vm/os_linux_zero.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/os_cpu/linux_zero/vm/os_linux_zero.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,6 +1,6 @@ /* * Copyright 2003-2004 Sun Microsystems, Inc. All Rights Reserved. - * Copyright 2007, 2008 Red Hat, Inc. + * Copyright 2007, 2008, 2010 Red Hat, Inc. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -39,7 +39,13 @@ "stfd %0, 0(%2)\n" : "=f"(tmp) : "b"(src), "b"(dst)); +#elif defined(S390) && !defined(_LP64) + double tmp; + asm volatile ("ld %0, 0(%1)\n" + "std %0, 0(%2)\n" + : "=r"(tmp) + : "a"(src), "a"(dst)); #else *(jlong *) dst = *(jlong *) src; -#endif // PPC && !_LP64 +#endif }
--- a/src/os_cpu/solaris_x86/vm/globals_solaris_x86.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/os_cpu/solaris_x86/vm/globals_solaris_x86.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -22,31 +22,25 @@ * */ -// // Sets the default values for platform dependent flags used by the runtime system. // (see globals.hpp) -// + define_pd_global(bool, DontYieldALot, true); // Determined in the design center #ifdef AMD64 define_pd_global(intx, ThreadStackSize, 1024); // 0 => use system default define_pd_global(intx, VMThreadStackSize, 1024); -define_pd_global(intx, SurvivorRatio, 6); -define_pd_global(uintx, JVMInvokeMethodSlack, 8*K); +define_pd_global(uintx,JVMInvokeMethodSlack, 8*K); #else -// UseStackBanging is not pd -// define_pd_global(bool, UseStackBanging, true); - // ThreadStackSize 320 allows TaggedStackInterpreter and a couple of test cases // to run while keeping the number of threads that can be created high. define_pd_global(intx, ThreadStackSize, 320); define_pd_global(intx, VMThreadStackSize, 512); -define_pd_global(intx, SurvivorRatio, 8); -define_pd_global(uintx, JVMInvokeMethodSlack, 10*K); +define_pd_global(uintx,JVMInvokeMethodSlack, 10*K); #endif // AMD64 define_pd_global(intx, CompilerThreadStackSize, 0); // Only used on 64 bit platforms -define_pd_global(uintx, HeapBaseMinAddress, 256*M); +define_pd_global(uintx,HeapBaseMinAddress, 256*M); // Only used on 64 bit Windows platforms define_pd_global(bool, UseVectoredExceptions, false);
--- a/src/os_cpu/solaris_x86/vm/os_solaris_x86.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/os_cpu/solaris_x86/vm/os_solaris_x86.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -730,11 +730,12 @@ st->print(", RSI=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RSI]); st->print(", RDI=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RDI]); st->cr(); - st->print(", R8=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R8]); + st->print( "R8=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R8]); st->print(", R9=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R9]); st->print(", R10=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R10]); st->print(", R11=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R11]); - st->print(", R12=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R12]); + st->cr(); + st->print( "R12=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R12]); st->print(", R13=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R13]); st->print(", R14=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R14]); st->print(", R15=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R15]);
--- a/src/os_cpu/windows_x86/vm/globals_windows_x86.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/os_cpu/windows_x86/vm/globals_windows_x86.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -22,10 +22,9 @@ * */ -// // Sets the default values for platform dependent flags used by the runtime system. // (see globals.hpp) -// + define_pd_global(bool, DontYieldALot, false); // Default stack size on Windows is determined by the executable (java.exe @@ -35,8 +34,6 @@ define_pd_global(intx, ThreadStackSize, 0); // 0 => use system default define_pd_global(intx, VMThreadStackSize, 0); // 0 => use system default -define_pd_global(intx, SurvivorRatio, 8); - #ifdef ASSERT define_pd_global(intx, CompilerThreadStackSize, 1024); #else
--- a/src/share/vm/adlc/output_c.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/adlc/output_c.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1998-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1998-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -1496,7 +1496,7 @@ unsigned i; // Generate Expand function header - fprintf(fp,"MachNode *%sNode::Expand(State *state, Node_List &proj_list) {\n", node->_ident); + fprintf(fp,"MachNode *%sNode::Expand(State *state, Node_List &proj_list, Node* mem) {\n", node->_ident); fprintf(fp,"Compile* C = Compile::current();\n"); // Generate expand code if( node->expands() ) { @@ -1546,15 +1546,16 @@ // Build a mapping from operand index to input edges fprintf(fp," unsigned idx0 = oper_input_base();\n"); - // The order in which inputs are added to a node is very + // The order in which the memory input is added to a node is very // strange. Store nodes get a memory input before Expand is - // called and all other nodes get it afterwards so - // oper_input_base is wrong during expansion. This code adjusts - // is so that expansion will work correctly. - bool missing_memory_edge = node->_matrule->needs_ideal_memory_edge(_globalNames) && - node->is_ideal_store() == Form::none; - if (missing_memory_edge) { - fprintf(fp," idx0--; // Adjust base because memory edge hasn't been inserted yet\n"); + // called and other nodes get it afterwards or before depending on + // match order so oper_input_base is wrong during expansion. This + // code adjusts it so that expansion will work correctly. + int has_memory_edge = node->_matrule->needs_ideal_memory_edge(_globalNames); + if (has_memory_edge) { + fprintf(fp," if (mem == (Node*)1) {\n"); + fprintf(fp," idx0--; // Adjust base because memory edge hasn't been inserted yet\n"); + fprintf(fp," }\n"); } for( i = 0; i < node->num_opnds(); i++ ) { @@ -1611,9 +1612,11 @@ int node_mem_op = node->memory_operand(_globalNames); assert( node_mem_op != InstructForm::NO_MEMORY_OPERAND, "expand rule member needs memory but top-level inst doesn't have any" ); - if (!missing_memory_edge) { + if (has_memory_edge) { // Copy memory edge - fprintf(fp," n%d->add_req(_in[1]);\t// Add memory edge\n", cnt); + fprintf(fp," if (mem != (Node*)1) {\n"); + fprintf(fp," n%d->add_req(_in[1]);\t// Add memory edge\n", cnt); + fprintf(fp," }\n"); } } @@ -1689,7 +1692,7 @@ } // done iterating over a new instruction's operands // Invoke Expand() for the newly created instruction. - fprintf(fp," result = n%d->Expand( state, proj_list );\n", cnt); + fprintf(fp," result = n%d->Expand( state, proj_list, mem );\n", cnt); assert( !new_inst->expands(), "Do not have complete support for recursive expansion"); } // done iterating over new instructions fprintf(fp,"\n");
--- a/src/share/vm/adlc/output_h.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/adlc/output_h.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1998-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1998-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -1754,7 +1754,7 @@ instr->has_temps() || instr->_matrule != NULL && instr->num_opnds() != instr->num_unique_opnds() ) { - fprintf(fp," virtual MachNode *Expand(State *state, Node_List &proj_list);\n"); + fprintf(fp," virtual MachNode *Expand(State *state, Node_List &proj_list, Node* mem);\n"); } if (instr->is_pinned(_globalNames)) {
--- a/src/share/vm/asm/codeBuffer.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/asm/codeBuffer.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -39,6 +39,7 @@ Dtrace_trap = OSR_Entry, // dtrace probes can never have an OSR entry so reuse it Exceptions, // Offset where exception handler lives Deopt, // Offset where deopt handler lives + DeoptMH, // Offset where MethodHandle deopt handler lives max_Entries }; // special value to note codeBlobs where profile (forte) stack walking is @@ -51,12 +52,13 @@ public: CodeOffsets() { - _values[Entry] = 0; + _values[Entry ] = 0; _values[Verified_Entry] = 0; _values[Frame_Complete] = frame_never_safe; - _values[OSR_Entry] = 0; - _values[Exceptions] = -1; - _values[Deopt] = -1; + _values[OSR_Entry ] = 0; + _values[Exceptions ] = -1; + _values[Deopt ] = -1; + _values[DeoptMH ] = -1; } int value(Entries e) { return _values[e]; }
--- a/src/share/vm/c1/c1_Compilation.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/c1/c1_Compilation.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -205,6 +205,8 @@ void Compilation::emit_code_epilog(LIR_Assembler* assembler) { CHECK_BAILOUT(); + CodeOffsets* code_offsets = assembler->offsets(); + // generate code or slow cases assembler->emit_slow_case_stubs(); CHECK_BAILOUT(); @@ -213,10 +215,18 @@ assembler->emit_exception_entries(exception_info_list()); CHECK_BAILOUT(); - // generate code for exception handler - assembler->emit_exception_handler(); + // Generate code for exception handler. + code_offsets->set_value(CodeOffsets::Exceptions, assembler->emit_exception_handler()); CHECK_BAILOUT(); - assembler->emit_deopt_handler(); + + // Generate code for deopt handler. + code_offsets->set_value(CodeOffsets::Deopt, assembler->emit_deopt_handler()); + CHECK_BAILOUT(); + + // Generate code for MethodHandle deopt handler. We can use the + // same code as for the normal deopt handler, we just need a + // different entry point address. + code_offsets->set_value(CodeOffsets::DeoptMH, assembler->emit_deopt_handler()); CHECK_BAILOUT(); // done
--- a/src/share/vm/c1/c1_GraphBuilder.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/c1/c1_GraphBuilder.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -365,7 +365,7 @@ if (_next_loop_index < 31) _next_loop_index++; } else { // block already marked as loop header - assert(is_power_of_2(_loop_map.at(block->block_id())), "exactly one bit must be set"); + assert(is_power_of_2((unsigned int)_loop_map.at(block->block_id())), "exactly one bit must be set"); } }
--- a/src/share/vm/c1/c1_IR.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/c1/c1_IR.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -251,8 +251,10 @@ DebugToken* expvals = recorder->create_scope_values(expressions()); DebugToken* monvals = recorder->create_monitor_values(monitors()); // reexecute allowed only for the topmost frame - bool reexecute = topmost ? should_reexecute() : false; - recorder->describe_scope(pc_offset, scope()->method(), bci(), reexecute, locvals, expvals, monvals); + bool reexecute = topmost ? should_reexecute() : false; + bool is_method_handle_invoke = false; + bool return_oop = false; // This flag will be ignored since it used only for C2 with escape analysis. + recorder->describe_scope(pc_offset, scope()->method(), bci(), reexecute, is_method_handle_invoke, return_oop, locvals, expvals, monvals); } };
--- a/src/share/vm/c1/c1_LIR.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/c1/c1_LIR.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2000-2008 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2000-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -2000,7 +2000,7 @@ typedef enum { inputMode, firstMode = inputMode, tempMode, outputMode, numModes, invalidMode = -1 } OprMode; enum { - maxNumberOfOperands = 14, + maxNumberOfOperands = 16, maxNumberOfInfos = 4 };
--- a/src/share/vm/c1/c1_LIRAssembler.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/c1/c1_LIRAssembler.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2000-2008 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2000-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -133,9 +133,9 @@ void add_call_info_here(CodeEmitInfo* info) { add_call_info(code_offset(), info); } // code patterns - void emit_exception_handler(); + int emit_exception_handler(); void emit_exception_entries(ExceptionInfoList* info_list); - void emit_deopt_handler(); + int emit_deopt_handler(); void emit_code(BlockList* hir); void emit_block(BlockBegin* block);
--- a/src/share/vm/c1/c1_LIRGenerator.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/c1/c1_LIRGenerator.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2005-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2005-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -1765,7 +1765,7 @@ __ null_check(exception_opr, new CodeEmitInfo(info, true)); } - if (compilation()->env()->jvmti_can_post_exceptions() && + if (compilation()->env()->jvmti_can_post_on_exceptions() && !block()->is_set(BlockBegin::default_exception_handler_flag)) { // we need to go through the exception lookup path to get JVMTI // notification done @@ -1855,12 +1855,26 @@ addr = new LIR_Address(base_op, index_op->as_jint(), dst_type); } else { #ifdef X86 +#ifdef _LP64 + if (!index_op->is_illegal() && index_op->type() == T_INT) { + LIR_Opr tmp = new_pointer_register(); + __ convert(Bytecodes::_i2l, index_op, tmp); + index_op = tmp; + } +#endif addr = new LIR_Address(base_op, index_op, LIR_Address::Scale(log2_scale), 0, dst_type); #else if (index_op->is_illegal() || log2_scale == 0) { +#ifdef _LP64 + if (!index_op->is_illegal() && index_op->type() == T_INT) { + LIR_Opr tmp = new_pointer_register(); + __ convert(Bytecodes::_i2l, index_op, tmp); + index_op = tmp; + } +#endif addr = new LIR_Address(base_op, index_op, dst_type); } else { - LIR_Opr tmp = new_register(T_INT); + LIR_Opr tmp = new_pointer_register(); __ shift_left(index_op, log2_scale, tmp); addr = new LIR_Address(base_op, tmp, dst_type); } @@ -1915,10 +1929,25 @@ LIR_Opr index_op = idx.result(); if (log2_scale != 0) { // temporary fix (platform dependent code without shift on Intel would be better) - index_op = new_register(T_INT); - __ move(idx.result(), index_op); + index_op = new_pointer_register(); +#ifdef _LP64 + if(idx.result()->type() == T_INT) { + __ convert(Bytecodes::_i2l, idx.result(), index_op); + } else { +#endif + __ move(idx.result(), index_op); +#ifdef _LP64 + } +#endif __ shift_left(index_op, log2_scale, index_op); } +#ifdef _LP64 + else if(!index_op->is_illegal() && index_op->type() == T_INT) { + LIR_Opr tmp = new_pointer_register(); + __ convert(Bytecodes::_i2l, index_op, tmp); + index_op = tmp; + } +#endif LIR_Address* addr = new LIR_Address(base_op, index_op, x->basic_type()); __ move(value.result(), addr);
--- a/src/share/vm/c1/c1_LinearScan.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/c1/c1_LinearScan.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -2464,6 +2464,10 @@ case T_LONG: // fall through case T_DOUBLE: { +#ifdef _LP64 + scope_values->append(&_int_0_scope_value); + scope_values->append(new ConstantLongValue(c->as_jlong_bits())); +#else if (hi_word_offset_in_bytes > lo_word_offset_in_bytes) { scope_values->append(new ConstantIntValue(c->as_jint_hi_bits())); scope_values->append(new ConstantIntValue(c->as_jint_lo_bits())); @@ -2471,7 +2475,7 @@ scope_values->append(new ConstantIntValue(c->as_jint_lo_bits())); scope_values->append(new ConstantIntValue(c->as_jint_hi_bits())); } - +#endif return 2; } @@ -2503,17 +2507,18 @@ } else if (opr->is_single_cpu()) { bool is_oop = opr->is_oop_register(); int cache_idx = opr->cpu_regnr() * 2 + (is_oop ? 1 : 0); + Location::Type int_loc_type = NOT_LP64(Location::normal) LP64_ONLY(Location::int_in_long); ScopeValue* sv = _scope_value_cache.at(cache_idx); if (sv == NULL) { - Location::Type loc_type = is_oop ? Location::oop : Location::normal; + Location::Type loc_type = is_oop ? Location::oop : int_loc_type; VMReg rname = frame_map()->regname(opr); sv = new LocationValue(Location::new_reg_loc(loc_type, rname)); _scope_value_cache.at_put(cache_idx, sv); } // check if cached value is correct - DEBUG_ONLY(assert_equal(sv, new LocationValue(Location::new_reg_loc(is_oop ? Location::oop : Location::normal, frame_map()->regname(opr))))); + DEBUG_ONLY(assert_equal(sv, new LocationValue(Location::new_reg_loc(is_oop ? Location::oop : int_loc_type, frame_map()->regname(opr))))); scope_values->append(sv); return 1;
--- a/src/share/vm/c1/c1_Runtime1.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/c1/c1_Runtime1.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -110,8 +110,8 @@ RegisterMap reg_map(thread, false); frame runtime_frame = thread->last_frame(); frame caller_frame = runtime_frame.sender(®_map); - VM_DeoptimizeFrame deopt(thread, caller_frame.id()); - VMThread::execute(&deopt); + // bypass VM_DeoptimizeFrame and deoptimize the frame directly + Deoptimization::deoptimize_frame(thread, caller_frame.id()); assert(caller_is_deopted(), "Must be deoptimized"); } } @@ -354,7 +354,7 @@ JRT_ENTRY(void, Runtime1::post_jvmti_exception_throw(JavaThread* thread)) - if (JvmtiExport::can_post_exceptions()) { + if (JvmtiExport::can_post_on_exceptions()) { vframeStream vfst(thread, true); address bcp = vfst.method()->bcp_from(vfst.bci()); JvmtiExport::post_exception_throw(thread, vfst.method(), bcp, thread->exception_oop()); @@ -425,7 +425,7 @@ assert(exception.not_null(), "NULL exceptions should be handled by throw_exception"); assert(exception->is_oop(), "just checking"); // Check that exception is a subclass of Throwable, otherwise we have a VerifyError - if (!(exception->is_a(SystemDictionary::throwable_klass()))) { + if (!(exception->is_a(SystemDictionary::Throwable_klass()))) { if (ExitVMOnVerifyError) vm_exit(-1); ShouldNotReachHere(); } @@ -437,7 +437,7 @@ bool guard_pages_enabled = thread->stack_yellow_zone_enabled(); if (!guard_pages_enabled) guard_pages_enabled = thread->reguard_stack(); - if (JvmtiExport::can_post_exceptions()) { + if (JvmtiExport::can_post_on_exceptions()) { // To ensure correct notification of exception catches and throws // we have to deoptimize here. If we attempted to notify the // catches and throws during this exception lookup it's possible @@ -1075,6 +1075,7 @@ }; +// Below length is the # elements copied. template <class T> int obj_arraycopy_work(oopDesc* src, T* src_addr, oopDesc* dst, T* dst_addr, int length) { @@ -1083,22 +1084,22 @@ // barrier. The assert will fail if this is not the case. // Note that we use the non-virtual inlineable variant of write_ref_array. BarrierSet* bs = Universe::heap()->barrier_set(); - assert(bs->has_write_ref_array_opt(), - "Barrier set must have ref array opt"); + assert(bs->has_write_ref_array_opt(), "Barrier set must have ref array opt"); + assert(bs->has_write_ref_array_pre_opt(), "For pre-barrier as well."); if (src == dst) { // same object, no check + bs->write_ref_array_pre(dst_addr, length); Copy::conjoint_oops_atomic(src_addr, dst_addr, length); - bs->write_ref_array(MemRegion((HeapWord*)dst_addr, - (HeapWord*)(dst_addr + length))); + bs->write_ref_array((HeapWord*)dst_addr, length); return ac_ok; } else { klassOop bound = objArrayKlass::cast(dst->klass())->element_klass(); klassOop stype = objArrayKlass::cast(src->klass())->element_klass(); if (stype == bound || Klass::cast(stype)->is_subtype_of(bound)) { // Elements are guaranteed to be subtypes, so no check necessary + bs->write_ref_array_pre(dst_addr, length); Copy::conjoint_oops_atomic(src_addr, dst_addr, length); - bs->write_ref_array(MemRegion((HeapWord*)dst_addr, - (HeapWord*)(dst_addr + length))); + bs->write_ref_array((HeapWord*)dst_addr, length); return ac_ok; } } @@ -1162,9 +1163,16 @@ #endif if (num == 0) return; + BarrierSet* bs = Universe::heap()->barrier_set(); + assert(bs->has_write_ref_array_opt(), "Barrier set must have ref array opt"); + assert(bs->has_write_ref_array_pre_opt(), "For pre-barrier as well."); + if (UseCompressedOops) { + bs->write_ref_array_pre((narrowOop*)dst, num); + } else { + bs->write_ref_array_pre((oop*)dst, num); + } Copy::conjoint_oops_atomic((oop*) src, (oop*) dst, num); - BarrierSet* bs = Universe::heap()->barrier_set(); - bs->write_ref_array(MemRegion(dst, dst + num)); + bs->write_ref_array(dst, num); JRT_END
--- a/src/share/vm/ci/bcEscapeAnalyzer.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/bcEscapeAnalyzer.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -61,9 +61,11 @@ BCEscapeAnalyzer* _parent; int _level; + public: class ArgumentMap; class StateInfo; + private: // helper functions bool is_argument(int i) { return i >= 0 && i < _arg_size; }
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/share/vm/ci/ciCPCache.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,49 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +#include "incls/_precompiled.incl" +#include "incls/_ciCPCache.cpp.incl" + +// ciCPCache + +// ------------------------------------------------------------------ +// ciCPCache::get_f1_offset +size_t ciCPCache::get_f1_offset(int index) { + // Calculate the offset from the constantPoolCacheOop to the f1 + // field. + ByteSize f1_offset = + constantPoolCacheOopDesc::entry_offset(index) + + ConstantPoolCacheEntry::f1_offset(); + + return in_bytes(f1_offset); +} + + +// ------------------------------------------------------------------ +// ciCPCache::print +// +// Print debugging information about the cache. +void ciCPCache::print() { + Unimplemented(); +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/share/vm/ci/ciCPCache.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,43 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +// ciCPCache +// +// This class represents a constant pool cache. +// +// Note: This class is called ciCPCache as ciConstantPoolCache is used +// for something different. +class ciCPCache : public ciObject { +public: + ciCPCache(constantPoolCacheHandle cpcache) : ciObject(cpcache) {} + + // What kind of ciObject is this? + bool is_cpcache() const { return true; } + + // Get the offset in bytes from the oop to the f1 field of the + // requested entry. + size_t get_f1_offset(int index); + + void print(); +};
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/share/vm/ci/ciCallSite.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,46 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +#include "incls/_precompiled.incl" +#include "incls/_ciCallSite.cpp.incl" + +// ciCallSite + +// ------------------------------------------------------------------ +// ciCallSite::get_target +// +// Return the target MethodHandle of this CallSite. +ciMethodHandle* ciCallSite::get_target() const { + VM_ENTRY_MARK; + oop method_handle_oop = java_dyn_CallSite::target(get_oop()); + return CURRENT_ENV->get_object(method_handle_oop)->as_method_handle(); +} + +// ------------------------------------------------------------------ +// ciCallSite::print +// +// Print debugging information about the CallSite. +void ciCallSite::print() { + Unimplemented(); +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/share/vm/ci/ciCallSite.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,39 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +// ciCallSite +// +// The class represents a java.dyn.CallSite object. +class ciCallSite : public ciInstance { +public: + ciCallSite(instanceHandle h_i) : ciInstance(h_i) {} + + // What kind of ciObject is this? + bool is_call_site() const { return true; } + + // Return the target MethodHandle of this CallSite. + ciMethodHandle* get_target() const; + + void print(); +};
--- a/src/share/vm/ci/ciClassList.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciClassList.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -25,6 +25,7 @@ class ciEnv; class ciObjectFactory; class ciConstantPoolCache; +class ciCPCache; class ciField; class ciConstant; @@ -42,6 +43,8 @@ class ciObject; class ciNullObject; class ciInstance; +class ciCallSite; +class ciMethodHandle; class ciMethod; class ciMethodData; class ciReceiverTypeData; // part of ciMethodData @@ -78,6 +81,7 @@ // Any more access must be given explicitly. #define CI_PACKAGE_ACCESS_TO \ friend class ciObjectFactory; \ +friend class ciCallSite; \ friend class ciConstantPoolCache; \ friend class ciField; \ friend class ciConstant; \ @@ -93,6 +97,7 @@ friend class ciInstance; \ friend class ciMethod; \ friend class ciMethodData; \ +friend class ciMethodHandle; \ friend class ciReceiverTypeData; \ friend class ciSymbol; \ friend class ciArray; \
--- a/src/share/vm/ci/ciEnv.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciEnv.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -38,14 +38,9 @@ ciTypeArrayKlassKlass* ciEnv::_type_array_klass_klass_instance; ciObjArrayKlassKlass* ciEnv::_obj_array_klass_klass_instance; -ciInstanceKlass* ciEnv::_ArrayStoreException; -ciInstanceKlass* ciEnv::_Class; -ciInstanceKlass* ciEnv::_ClassCastException; -ciInstanceKlass* ciEnv::_Object; -ciInstanceKlass* ciEnv::_Throwable; -ciInstanceKlass* ciEnv::_Thread; -ciInstanceKlass* ciEnv::_OutOfMemoryError; -ciInstanceKlass* ciEnv::_String; +#define WK_KLASS_DEFN(name, ignore_s, ignore_o) ciInstanceKlass* ciEnv::_##name = NULL; +WK_KLASSES_DO(WK_KLASS_DEFN) +#undef WK_KLASS_DEFN ciSymbol* ciEnv::_unloaded_cisymbol = NULL; ciInstanceKlass* ciEnv::_unloaded_ciinstance_klass = NULL; @@ -110,6 +105,8 @@ _ArrayIndexOutOfBoundsException_instance = NULL; _ArrayStoreException_instance = NULL; _ClassCastException_instance = NULL; + _the_null_string = NULL; + _the_min_jint_string = NULL; } ciEnv::ciEnv(Arena* arena) { @@ -163,6 +160,8 @@ _ArrayIndexOutOfBoundsException_instance = NULL; _ArrayStoreException_instance = NULL; _ClassCastException_instance = NULL; + _the_null_string = NULL; + _the_min_jint_string = NULL; } ciEnv::~ciEnv() { @@ -179,7 +178,7 @@ _jvmti_can_hotswap_or_post_breakpoint = JvmtiExport::can_hotswap_or_post_breakpoint(); _jvmti_can_examine_or_deopt_anywhere = JvmtiExport::can_examine_or_deopt_anywhere(); _jvmti_can_access_local_variables = JvmtiExport::can_access_local_variables(); - _jvmti_can_post_exceptions = JvmtiExport::can_post_exceptions(); + _jvmti_can_post_on_exceptions = JvmtiExport::can_post_on_exceptions(); } // ------------------------------------------------------------------ @@ -248,6 +247,22 @@ return _ClassCastException_instance; } +ciInstance* ciEnv::the_null_string() { + if (_the_null_string == NULL) { + VM_ENTRY_MARK; + _the_null_string = get_object(Universe::the_null_string())->as_instance(); + } + return _the_null_string; +} + +ciInstance* ciEnv::the_min_jint_string() { + if (_the_min_jint_string == NULL) { + VM_ENTRY_MARK; + _the_min_jint_string = get_object(Universe::the_min_jint_string())->as_instance(); + } + return _the_min_jint_string; +} + // ------------------------------------------------------------------ // ciEnv::get_method_from_handle ciMethod* ciEnv::get_method_from_handle(jobject method) { @@ -419,12 +434,11 @@ // ciEnv::get_klass_by_index_impl // // Implementation of get_klass_by_index. -ciKlass* ciEnv::get_klass_by_index_impl(ciInstanceKlass* accessor, +ciKlass* ciEnv::get_klass_by_index_impl(constantPoolHandle cpool, int index, - bool& is_accessible) { - assert(accessor->get_instanceKlass()->is_linked(), "must be linked before accessing constant pool"); + bool& is_accessible, + ciInstanceKlass* accessor) { EXCEPTION_CONTEXT; - constantPoolHandle cpool(THREAD, accessor->get_instanceKlass()->constants()); KlassHandle klass (THREAD, constantPoolOopDesc::klass_at_if_loaded(cpool, index)); symbolHandle klass_name; if (klass.is_null()) { @@ -486,22 +500,21 @@ // ciEnv::get_klass_by_index // // Get a klass from the constant pool. -ciKlass* ciEnv::get_klass_by_index(ciInstanceKlass* accessor, +ciKlass* ciEnv::get_klass_by_index(constantPoolHandle cpool, int index, - bool& is_accessible) { - GUARDED_VM_ENTRY(return get_klass_by_index_impl(accessor, index, is_accessible);) + bool& is_accessible, + ciInstanceKlass* accessor) { + GUARDED_VM_ENTRY(return get_klass_by_index_impl(cpool, index, is_accessible, accessor);) } // ------------------------------------------------------------------ // ciEnv::get_constant_by_index_impl // // Implementation of get_constant_by_index(). -ciConstant ciEnv::get_constant_by_index_impl(ciInstanceKlass* accessor, - int index) { +ciConstant ciEnv::get_constant_by_index_impl(constantPoolHandle cpool, + int index, + ciInstanceKlass* accessor) { EXCEPTION_CONTEXT; - instanceKlass* ik_accessor = accessor->get_instanceKlass(); - assert(ik_accessor->is_linked(), "must be linked before accessing constant pool"); - constantPoolOop cpool = ik_accessor->constants(); constantTag tag = cpool->tag_at(index); if (tag.is_int()) { return ciConstant(T_INT, (jint)cpool->int_at(index)); @@ -529,7 +542,7 @@ } else if (tag.is_klass() || tag.is_unresolved_klass()) { // 4881222: allow ldc to take a class type bool ignore; - ciKlass* klass = get_klass_by_index_impl(accessor, index, ignore); + ciKlass* klass = get_klass_by_index_impl(cpool, index, ignore, accessor); if (HAS_PENDING_EXCEPTION) { CLEAR_PENDING_EXCEPTION; record_out_of_memory_failure(); @@ -538,6 +551,11 @@ assert (klass->is_instance_klass() || klass->is_array_klass(), "must be an instance or array klass "); return ciConstant(T_OBJECT, klass); + } else if (tag.is_object()) { + oop obj = cpool->object_at(index); + assert(obj->is_instance(), "must be an instance"); + ciObject* ciobj = get_object(obj); + return ciConstant(T_OBJECT, ciobj); } else { ShouldNotReachHere(); return ciConstant(); @@ -574,9 +592,10 @@ // Pull a constant out of the constant pool. How appropriate. // // Implementation note: this query is currently in no way cached. -ciConstant ciEnv::get_constant_by_index(ciInstanceKlass* accessor, - int index) { - GUARDED_VM_ENTRY(return get_constant_by_index_impl(accessor, index); ) +ciConstant ciEnv::get_constant_by_index(constantPoolHandle cpool, + int index, + ciInstanceKlass* accessor) { + GUARDED_VM_ENTRY(return get_constant_by_index_impl(cpool, index, accessor);) } // ------------------------------------------------------------------ @@ -586,7 +605,7 @@ // // Implementation note: this query is currently in no way cached. bool ciEnv::is_unresolved_string(ciInstanceKlass* accessor, - int index) const { + int index) const { GUARDED_VM_ENTRY(return is_unresolved_string_impl(accessor->get_instanceKlass(), index); ) } @@ -597,7 +616,7 @@ // // Implementation note: this query is currently in no way cached. bool ciEnv::is_unresolved_klass(ciInstanceKlass* accessor, - int index) const { + int index) const { GUARDED_VM_ENTRY(return is_unresolved_klass_impl(accessor->get_instanceKlass(), index); ) } @@ -678,22 +697,17 @@ // ------------------------------------------------------------------ // ciEnv::get_method_by_index_impl -ciMethod* ciEnv::get_method_by_index_impl(ciInstanceKlass* accessor, - int index, Bytecodes::Code bc) { - // Get the method's declared holder. - - assert(accessor->get_instanceKlass()->is_linked(), "must be linked before accessing constant pool"); - constantPoolHandle cpool = accessor->get_instanceKlass()->constants(); +ciMethod* ciEnv::get_method_by_index_impl(constantPoolHandle cpool, + int index, Bytecodes::Code bc, + ciInstanceKlass* accessor) { int holder_index = cpool->klass_ref_index_at(index); bool holder_is_accessible; - ciKlass* holder = get_klass_by_index_impl(accessor, holder_index, holder_is_accessible); + ciKlass* holder = get_klass_by_index_impl(cpool, holder_index, holder_is_accessible, accessor); ciInstanceKlass* declared_holder = get_instance_klass_for_declared_method_holder(holder); // Get the method's name and signature. - int nt_index = cpool->name_and_type_ref_index_at(index); - int sig_index = cpool->signature_ref_index_at(nt_index); symbolOop name_sym = cpool->name_ref_at(index); - symbolOop sig_sym = cpool->symbol_at(sig_index); + symbolOop sig_sym = cpool->signature_ref_at(index); if (holder_is_accessible) { // Our declared holder is loaded. instanceKlass* lookup = declared_holder->get_instanceKlass(); @@ -715,6 +729,33 @@ // ------------------------------------------------------------------ +// ciEnv::get_fake_invokedynamic_method_impl +ciMethod* ciEnv::get_fake_invokedynamic_method_impl(constantPoolHandle cpool, + int index, Bytecodes::Code bc) { + assert(bc == Bytecodes::_invokedynamic, "must be invokedynamic"); + + // Get the CallSite from the constant pool cache. + ConstantPoolCacheEntry* cpc_entry = cpool->cache()->secondary_entry_at(index); + assert(cpc_entry != NULL && cpc_entry->is_secondary_entry(), "sanity"); + Handle call_site = cpc_entry->f1(); + + // Call site might not be linked yet. + if (call_site.is_null()) { + ciInstanceKlass* mh_klass = get_object(SystemDictionary::MethodHandle_klass())->as_instance_klass(); + ciSymbol* sig_sym = get_object(cpool->signature_ref_at(index))->as_symbol(); + return get_unloaded_method(mh_klass, ciSymbol::invoke_name(), sig_sym); + } + + // Get the methodOop from the CallSite. + methodOop method_oop = (methodOop) java_dyn_CallSite::vmmethod(call_site()); + assert(method_oop != NULL, "sanity"); + assert(method_oop->is_method_handle_invoke(), "consistent"); + + return get_object(method_oop)->as_method(); +} + + +// ------------------------------------------------------------------ // ciEnv::get_instance_klass_for_declared_method_holder ciInstanceKlass* ciEnv::get_instance_klass_for_declared_method_holder(ciKlass* method_holder) { // For the case of <array>.clone(), the method holder can be a ciArrayKlass @@ -736,15 +777,19 @@ } - - // ------------------------------------------------------------------ // ciEnv::get_method_by_index -ciMethod* ciEnv::get_method_by_index(ciInstanceKlass* accessor, - int index, Bytecodes::Code bc) { - GUARDED_VM_ENTRY(return get_method_by_index_impl(accessor, index, bc);) +ciMethod* ciEnv::get_method_by_index(constantPoolHandle cpool, + int index, Bytecodes::Code bc, + ciInstanceKlass* accessor) { + if (bc == Bytecodes::_invokedynamic) { + GUARDED_VM_ENTRY(return get_fake_invokedynamic_method_impl(cpool, index, bc);) + } else { + GUARDED_VM_ENTRY(return get_method_by_index_impl(cpool, index, bc, accessor);) + } } + // ------------------------------------------------------------------ // ciEnv::name_buffer char *ciEnv::name_buffer(int req_len) { @@ -846,8 +891,8 @@ JvmtiExport::can_examine_or_deopt_anywhere()) || (!jvmti_can_access_local_variables() && JvmtiExport::can_access_local_variables()) || - (!jvmti_can_post_exceptions() && - JvmtiExport::can_post_exceptions()) )) { + (!jvmti_can_post_on_exceptions() && + JvmtiExport::can_post_on_exceptions()) )) { record_failure("Jvmti state change invalidated dependencies"); } @@ -917,18 +962,10 @@ if (nm == NULL) { // The CodeCache is full. Print out warning and disable compilation. record_failure("code cache is full"); - UseInterpreter = true; - if (UseCompiler || AlwaysCompileLoopMethods ) { -#ifndef PRODUCT - warning("CodeCache is full. Compiler has been disabled"); - if (CompileTheWorld || ExitOnFullCodeCache) { - before_exit(JavaThread::current()); - exit_globals(); // will delete tty - vm_direct_exit(CompileTheWorld ? 0 : 1); - } -#endif - UseCompiler = false; - AlwaysCompileLoopMethods = false; + { + MutexUnlocker ml(Compile_lock); + MutexUnlocker locker(MethodCompileQueue_lock); + CompileBroker::handle_full_code_cache(); } } else { NOT_PRODUCT(nm->set_has_debug_info(has_debug_info); )
--- a/src/share/vm/ci/ciEnv.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciEnv.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -57,7 +57,7 @@ bool _jvmti_can_hotswap_or_post_breakpoint; bool _jvmti_can_examine_or_deopt_anywhere; bool _jvmti_can_access_local_variables; - bool _jvmti_can_post_exceptions; + bool _jvmti_can_post_on_exceptions; // Cache DTrace flags bool _dtrace_extended_probes; @@ -74,14 +74,9 @@ static ciTypeArrayKlassKlass* _type_array_klass_klass_instance; static ciObjArrayKlassKlass* _obj_array_klass_klass_instance; - static ciInstanceKlass* _ArrayStoreException; - static ciInstanceKlass* _Class; - static ciInstanceKlass* _ClassCastException; - static ciInstanceKlass* _Object; - static ciInstanceKlass* _Throwable; - static ciInstanceKlass* _Thread; - static ciInstanceKlass* _OutOfMemoryError; - static ciInstanceKlass* _String; +#define WK_KLASS_DECL(name, ignore_s, ignore_o) static ciInstanceKlass* _##name; + WK_KLASSES_DO(WK_KLASS_DECL) +#undef WK_KLASS_DECL static ciSymbol* _unloaded_cisymbol; static ciInstanceKlass* _unloaded_ciinstance_klass; @@ -97,6 +92,9 @@ ciInstance* _ArrayStoreException_instance; ciInstance* _ClassCastException_instance; + ciInstance* _the_null_string; // The Java string "null" + ciInstance* _the_min_jint_string; // The Java string "-2147483648" + // Look up a klass by name from a particular class loader (the accessor's). // If require_local, result must be defined in that class loader, or NULL. // If !require_local, a result from remote class loader may be reported, @@ -114,37 +112,45 @@ bool require_local); // Constant pool access. - ciKlass* get_klass_by_index(ciInstanceKlass* loading_klass, + ciKlass* get_klass_by_index(constantPoolHandle cpool, int klass_index, - bool& is_accessible); - ciConstant get_constant_by_index(ciInstanceKlass* loading_klass, - int constant_index); + bool& is_accessible, + ciInstanceKlass* loading_klass); + ciConstant get_constant_by_index(constantPoolHandle cpool, + int constant_index, + ciInstanceKlass* accessor); bool is_unresolved_string(ciInstanceKlass* loading_klass, int constant_index) const; bool is_unresolved_klass(ciInstanceKlass* loading_klass, int constant_index) const; ciField* get_field_by_index(ciInstanceKlass* loading_klass, int field_index); - ciMethod* get_method_by_index(ciInstanceKlass* loading_klass, - int method_index, Bytecodes::Code bc); + ciMethod* get_method_by_index(constantPoolHandle cpool, + int method_index, Bytecodes::Code bc, + ciInstanceKlass* loading_klass); // Implementation methods for loading and constant pool access. ciKlass* get_klass_by_name_impl(ciKlass* accessing_klass, ciSymbol* klass_name, bool require_local); - ciKlass* get_klass_by_index_impl(ciInstanceKlass* loading_klass, + ciKlass* get_klass_by_index_impl(constantPoolHandle cpool, int klass_index, - bool& is_accessible); - ciConstant get_constant_by_index_impl(ciInstanceKlass* loading_klass, - int constant_index); + bool& is_accessible, + ciInstanceKlass* loading_klass); + ciConstant get_constant_by_index_impl(constantPoolHandle cpool, + int constant_index, + ciInstanceKlass* loading_klass); bool is_unresolved_string_impl (instanceKlass* loading_klass, int constant_index) const; bool is_unresolved_klass_impl (instanceKlass* loading_klass, int constant_index) const; ciField* get_field_by_index_impl(ciInstanceKlass* loading_klass, int field_index); - ciMethod* get_method_by_index_impl(ciInstanceKlass* loading_klass, - int method_index, Bytecodes::Code bc); + ciMethod* get_method_by_index_impl(constantPoolHandle cpool, + int method_index, Bytecodes::Code bc, + ciInstanceKlass* loading_klass); + ciMethod* get_fake_invokedynamic_method_impl(constantPoolHandle cpool, + int index, Bytecodes::Code bc); // Helper methods bool check_klass_accessibility(ciKlass* accessing_klass, @@ -253,7 +259,7 @@ bool jvmti_can_hotswap_or_post_breakpoint() const { return _jvmti_can_hotswap_or_post_breakpoint; } bool jvmti_can_examine_or_deopt_anywhere() const { return _jvmti_can_examine_or_deopt_anywhere; } bool jvmti_can_access_local_variables() const { return _jvmti_can_access_local_variables; } - bool jvmti_can_post_exceptions() const { return _jvmti_can_post_exceptions; } + bool jvmti_can_post_on_exceptions() const { return _jvmti_can_post_on_exceptions; } // Cache DTrace flags void cache_dtrace_flags(); @@ -286,30 +292,13 @@ // Access to certain well known ciObjects. - ciInstanceKlass* ArrayStoreException_klass() { - return _ArrayStoreException; - } - ciInstanceKlass* Class_klass() { - return _Class; - } - ciInstanceKlass* ClassCastException_klass() { - return _ClassCastException; - } - ciInstanceKlass* Object_klass() { - return _Object; +#define WK_KLASS_FUNC(name, ignore_s, ignore_o) \ + ciInstanceKlass* name() { \ + return _##name;\ } - ciInstanceKlass* Throwable_klass() { - return _Throwable; - } - ciInstanceKlass* Thread_klass() { - return _Thread; - } - ciInstanceKlass* OutOfMemoryError_klass() { - return _OutOfMemoryError; - } - ciInstanceKlass* String_klass() { - return _String; - } + WK_KLASSES_DO(WK_KLASS_FUNC) +#undef WK_KLASS_FUNC + ciInstance* NullPointerException_instance() { assert(_NullPointerException_instance != NULL, "initialization problem"); return _NullPointerException_instance; @@ -324,6 +313,9 @@ ciInstance* ArrayStoreException_instance(); ciInstance* ClassCastException_instance(); + ciInstance* the_null_string(); + ciInstance* the_min_jint_string(); + static ciSymbol* unloaded_cisymbol() { return _unloaded_cisymbol; }
--- a/src/share/vm/ci/ciExceptionHandler.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciExceptionHandler.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2003 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -34,12 +34,16 @@ // // Get the exception klass that this handler catches. ciInstanceKlass* ciExceptionHandler::catch_klass() { + VM_ENTRY_MARK; assert(!is_catch_all(), "bad index"); if (_catch_klass == NULL) { bool will_link; - ciKlass* k = CURRENT_ENV->get_klass_by_index(_loading_klass, + assert(_loading_klass->get_instanceKlass()->is_linked(), "must be linked before accessing constant pool"); + constantPoolHandle cpool(_loading_klass->get_instanceKlass()->constants()); + ciKlass* k = CURRENT_ENV->get_klass_by_index(cpool, _catch_klass_index, - will_link); + will_link, + _loading_klass); if (!will_link && k->is_loaded()) { GUARDED_VM_ENTRY( k = CURRENT_ENV->get_unloaded_klass(_loading_klass, k->name());
--- a/src/share/vm/ci/ciField.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciField.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2007 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -86,7 +86,7 @@ bool ignore; // This is not really a class reference; the index always refers to the // field's type signature, as a symbol. Linkage checks do not apply. - _type = ciEnv::current(thread)->get_klass_by_index(klass, sig_index, ignore); + _type = ciEnv::current(thread)->get_klass_by_index(cpool, sig_index, ignore, klass); } else { _type = ciType::make(field_type); } @@ -100,9 +100,9 @@ int holder_index = cpool->klass_ref_index_at(index); bool holder_is_accessible; ciInstanceKlass* declared_holder = - ciEnv::current(thread)->get_klass_by_index(klass, holder_index, - holder_is_accessible) - ->as_instance_klass(); + ciEnv::current(thread)->get_klass_by_index(cpool, holder_index, + holder_is_accessible, + klass)->as_instance_klass(); // The declared holder of this field may not have been loaded. // Bail out with partial field information. @@ -161,6 +161,18 @@ "bootstrap classes must not create & cache unshared fields"); } +static bool trust_final_non_static_fields(ciInstanceKlass* holder) { + if (holder == NULL) + return false; + if (holder->name() == ciSymbol::java_lang_System()) + // Never trust strangely unstable finals: System.out, etc. + return false; + // Even if general trusting is disabled, trust system-built closures in these packages. + if (holder->is_in_package("java/dyn") || holder->is_in_package("sun/dyn")) + return true; + return TrustFinalNonStaticFields; +} + void ciField::initialize_from(fieldDescriptor* fd) { // Get the flags, offset, and canonical holder of the field. _flags = ciFlags(fd->access_flags()); @@ -168,8 +180,18 @@ _holder = CURRENT_ENV->get_object(fd->field_holder())->as_instance_klass(); // Check to see if the field is constant. - if (_holder->is_initialized() && - this->is_final() && this->is_static()) { + if (_holder->is_initialized() && this->is_final()) { + if (!this->is_static()) { + // A field can be constant if it's a final static field or if it's + // a final non-static field of a trusted class ({java,sun}.dyn). + if (trust_final_non_static_fields(_holder)) { + _is_constant = true; + return; + } + _is_constant = false; + return; + } + // This field just may be constant. The only cases where it will // not be constant are: // @@ -182,8 +204,8 @@ // java.lang.System.out, and java.lang.System.err. klassOop k = _holder->get_klassOop(); - assert( SystemDictionary::system_klass() != NULL, "Check once per vm"); - if( k == SystemDictionary::system_klass() ) { + assert( SystemDictionary::System_klass() != NULL, "Check once per vm"); + if( k == SystemDictionary::System_klass() ) { // Check offsets for case 2: System.in, System.out, or System.err if( _offset == java_lang_System::in_offset_in_bytes() || _offset == java_lang_System::out_offset_in_bytes() ||
--- a/src/share/vm/ci/ciField.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciField.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -138,10 +138,18 @@ // Get the constant value of this field. ciConstant constant_value() { - assert(is_constant(), "illegal call to constant_value()"); + assert(is_static() && is_constant(), "illegal call to constant_value()"); return _constant_value; } + // Get the constant value of non-static final field in the given + // object. + ciConstant constant_value_of(ciObject* object) { + assert(!is_static() && is_constant(), "only if field is non-static constant"); + assert(object->is_instance(), "must be instance"); + return object->as_instance()->field_value(this); + } + // Check for link time errors. Accessing a field from a // certain class via a certain bytecode may or may not be legal. // This call checks to see if an exception may be raised by
--- a/src/share/vm/ci/ciInstance.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciInstance.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -36,7 +36,7 @@ VM_ENTRY_MARK; oop m = get_oop(); // Return NULL if it is not java.lang.Class. - if (m == NULL || m->klass() != SystemDictionary::class_klass()) { + if (m == NULL || m->klass() != SystemDictionary::Class_klass()) { return NULL; } // Return either a primitive type or a klass.
--- a/src/share/vm/ci/ciInstanceKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciInstanceKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -75,7 +75,7 @@ _java_mirror = NULL; if (is_shared()) { - if (h_k() != SystemDictionary::object_klass()) { + if (h_k() != SystemDictionary::Object_klass()) { super(); } java_mirror(); @@ -232,8 +232,48 @@ // ------------------------------------------------------------------ // ciInstanceKlass::uses_default_loader bool ciInstanceKlass::uses_default_loader() { - VM_ENTRY_MARK; - return loader() == NULL; + // Note: We do not need to resolve the handle or enter the VM + // in order to test null-ness. + return _loader == NULL; +} + +// ------------------------------------------------------------------ +// ciInstanceKlass::is_in_package +// +// Is this klass in the given package? +bool ciInstanceKlass::is_in_package(const char* packagename, int len) { + // To avoid class loader mischief, this test always rejects application classes. + if (!uses_default_loader()) + return false; + GUARDED_VM_ENTRY( + return is_in_package_impl(packagename, len); + ) +} + +bool ciInstanceKlass::is_in_package_impl(const char* packagename, int len) { + ASSERT_IN_VM; + + // If packagename contains trailing '/' exclude it from the + // prefix-test since we test for it explicitly. + if (packagename[len - 1] == '/') + len--; + + if (!name()->starts_with(packagename, len)) + return false; + + // Test if the class name is something like "java/lang". + if ((len + 1) > name()->utf8_length()) + return false; + + // Test for trailing '/' + if ((char) name()->byte_at(len) != '/') + return false; + + // Make sure it's not actually in a subpackage: + if (name()->index_of_at(len+1, "/", 1) >= 0) + return false; + + return true; } // ------------------------------------------------------------------ @@ -341,6 +381,20 @@ } // ------------------------------------------------------------------ +// ciInstanceKlass::get_field_by_name +ciField* ciInstanceKlass::get_field_by_name(ciSymbol* name, ciSymbol* signature, bool is_static) { + VM_ENTRY_MARK; + instanceKlass* k = get_instanceKlass(); + fieldDescriptor fd; + klassOop def = k->find_field(name->get_symbolOop(), signature->get_symbolOop(), is_static, &fd); + if (def == NULL) { + return NULL; + } + ciField* field = new (CURRENT_THREAD_ENV->arena()) ciField(&fd); + return field; +} + +// ------------------------------------------------------------------ // ciInstanceKlass::non_static_fields. class NonStaticFieldFiller: public FieldClosure {
--- a/src/share/vm/ci/ciInstanceKlass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciInstanceKlass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -29,10 +29,11 @@ // be loaded. class ciInstanceKlass : public ciKlass { CI_PACKAGE_ACCESS + friend class ciBytecodeStream; friend class ciEnv; + friend class ciExceptionHandler; friend class ciMethod; friend class ciField; - friend class ciBytecodeStream; private: jobject _loader; @@ -78,6 +79,8 @@ const char* type_string() { return "ciInstanceKlass"; } + bool is_in_package_impl(const char* packagename, int len); + void print_impl(outputStream* st); ciConstantPoolCache* field_cache(); @@ -148,6 +151,7 @@ ciInstanceKlass* get_canonical_holder(int offset); ciField* get_field_by_offset(int field_offset, bool is_static); + ciField* get_field_by_name(ciSymbol* name, ciSymbol* signature, bool is_static); GrowableArray<ciField*>* non_static_fields(); @@ -195,6 +199,12 @@ bool is_java_lang_Object(); + // Is this klass in the given package? + bool is_in_package(const char* packagename) { + return is_in_package(packagename, (int) strlen(packagename)); + } + bool is_in_package(const char* packagename, int len); + // What kind of ciObject is this? bool is_instance_klass() { return true; } bool is_java_klass() { return true; }
--- a/src/share/vm/ci/ciKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2007 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it
--- a/src/share/vm/ci/ciKlass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciKlass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2007 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -69,7 +69,7 @@ ciKlass(KlassHandle k_h); // What is the name of this klass? - ciSymbol* name() { return _name; } + ciSymbol* name() const { return _name; } // What is its layout helper value? jint layout_helper() { return _layout_helper; }
--- a/src/share/vm/ci/ciMethod.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciMethod.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -436,15 +436,21 @@ // we will set result._method also. } // Determine call site's morphism. - // The call site count could be == (receivers_count_total + 1) - // not only in the case of a polymorphic call but also in the case - // when a method data snapshot is taken after the site count was updated - // but before receivers counters were updated. - if (morphism == result._limit) { - // There were no array klasses and morphism <= MorphismLimit. - if (morphism < ciCallProfile::MorphismLimit || - morphism == ciCallProfile::MorphismLimit && - (receivers_count_total+1) >= count) { + // The call site count is 0 with known morphism (onlt 1 or 2 receivers) + // or < 0 in the case of a type check failured for checkcast, aastore, instanceof. + // The call site count is > 0 in the case of a polymorphic virtual call. + if (morphism > 0 && morphism == result._limit) { + // The morphism <= MorphismLimit. + if ((morphism < ciCallProfile::MorphismLimit) || + (morphism == ciCallProfile::MorphismLimit && count == 0)) { +#ifdef ASSERT + if (count > 0) { + this->print_short_name(tty); + tty->print_cr(" @ bci:%d", bci); + this->print_codes(); + assert(false, "this call site should not be polymorphic"); + } +#endif result._morphism = morphism; } } @@ -452,10 +458,8 @@ // zero or less, presume that this is a typecheck profile and // do nothing. Otherwise, increase count to be the sum of all // receiver's counts. - if (count > 0) { - if (count < receivers_count_total) { - count = receivers_count_total; - } + if (count >= 0) { + count += receivers_count_total; } } result._count = count; @@ -687,7 +691,7 @@ // ------------------------------------------------------------------ // invokedynamic support // -bool ciMethod::is_method_handle_invoke() { +bool ciMethod::is_method_handle_invoke() const { check_is_loaded(); bool flag = ((flags().as_int() & JVM_MH_INVOKE_BITS) == JVM_MH_INVOKE_BITS); #ifdef ASSERT @@ -700,6 +704,12 @@ return flag; } +bool ciMethod::is_method_handle_adapter() const { + check_is_loaded(); + VM_ENTRY_MARK; + return get_methodOop()->is_method_handle_adapter(); +} + ciInstance* ciMethod::method_handle_type() { check_is_loaded(); VM_ENTRY_MARK;
--- a/src/share/vm/ci/ciMethod.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciMethod.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -38,6 +38,8 @@ CI_PACKAGE_ACCESS friend class ciEnv; friend class ciExceptionHandlerStream; + friend class ciBytecodeStream; + friend class ciMethodHandle; private: // General method information. @@ -213,7 +215,10 @@ bool check_call(int refinfo_index, bool is_static) const; void build_method_data(); // make sure it exists in the VM also int scale_count(int count, float prof_factor = 1.); // make MDO count commensurate with IIC - bool is_method_handle_invoke(); + + // JSR 292 support + bool is_method_handle_invoke() const; + bool is_method_handle_adapter() const; ciInstance* method_handle_type(); // What kind of ciObject is this? @@ -251,4 +256,10 @@ // Print the name of this method in various incarnations. void print_name(outputStream* st = tty); void print_short_name(outputStream* st = tty); + + methodOop get_method_handle_target() { + klassOop receiver_limit_oop = NULL; + int flags = 0; + return MethodHandles::decode_method(get_oop(), receiver_limit_oop, flags); + } };
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/share/vm/ci/ciMethodHandle.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,52 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +#include "incls/_precompiled.incl" +#include "incls/_ciMethodHandle.cpp.incl" + +// ciMethodHandle + +// ------------------------------------------------------------------ +// ciMethodHandle::get_adapter +// +// Return an adapter for this MethodHandle. +ciMethod* ciMethodHandle::get_adapter(bool is_invokedynamic) const { + VM_ENTRY_MARK; + + Handle h(get_oop()); + methodHandle callee(_callee->get_methodOop()); + MethodHandleCompiler mhc(h, callee, is_invokedynamic, THREAD); + methodHandle m = mhc.compile(CHECK_NULL); + return CURRENT_ENV->get_object(m())->as_method(); +} + + +// ------------------------------------------------------------------ +// ciMethodHandle::print_impl +// +// Implementation of the print method. +void ciMethodHandle::print_impl(outputStream* st) { + st->print(" type="); + get_oop()->print(); +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/share/vm/ci/ciMethodHandle.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,56 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +// ciMethodHandle +// +// The class represents a java.dyn.MethodHandle object. +class ciMethodHandle : public ciInstance { +private: + ciMethod* _callee; + + // Return an adapter for this MethodHandle. + ciMethod* get_adapter(bool is_invokedynamic) const; + +protected: + void print_impl(outputStream* st); + +public: + ciMethodHandle(instanceHandle h_i) : ciInstance(h_i) {}; + + // What kind of ciObject is this? + bool is_method_handle() const { return true; } + + ciMethod* callee() const { return _callee; } + void set_callee(ciMethod* m) { _callee = m; } + + // Return an adapter for a MethodHandle call. + ciMethod* get_method_handle_adapter() const { + return get_adapter(false); + } + + // Return an adapter for an invokedynamic call. + ciMethod* get_invokedynamic_adapter() const { + return get_adapter(true); + } +};
--- a/src/share/vm/ci/ciObject.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciObject.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -131,9 +131,12 @@ // What kind of ciObject is this? virtual bool is_null_object() const { return false; } + virtual bool is_call_site() const { return false; } + virtual bool is_cpcache() const { return false; } virtual bool is_instance() { return false; } virtual bool is_method() { return false; } virtual bool is_method_data() { return false; } + virtual bool is_method_handle() const { return false; } virtual bool is_array() { return false; } virtual bool is_obj_array() { return false; } virtual bool is_type_array() { return false; } @@ -185,6 +188,14 @@ assert(is_null_object(), "bad cast"); return (ciNullObject*)this; } + ciCallSite* as_call_site() { + assert(is_call_site(), "bad cast"); + return (ciCallSite*) this; + } + ciCPCache* as_cpcache() { + assert(is_cpcache(), "bad cast"); + return (ciCPCache*) this; + } ciInstance* as_instance() { assert(is_instance(), "bad cast"); return (ciInstance*)this; @@ -197,6 +208,10 @@ assert(is_method_data(), "bad cast"); return (ciMethodData*)this; } + ciMethodHandle* as_method_handle() { + assert(is_method_handle(), "bad cast"); + return (ciMethodHandle*) this; + } ciArray* as_array() { assert(is_array(), "bad cast"); return (ciArray*)this;
--- a/src/share/vm/ci/ciObjectFactory.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciObjectFactory.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -144,30 +144,13 @@ ciEnv::_obj_array_klass_klass_instance = get(Universe::objArrayKlassKlassObj()) ->as_obj_array_klass_klass(); - ciEnv::_ArrayStoreException = - get(SystemDictionary::ArrayStoreException_klass()) - ->as_instance_klass(); - ciEnv::_Class = - get(SystemDictionary::class_klass()) - ->as_instance_klass(); - ciEnv::_ClassCastException = - get(SystemDictionary::ClassCastException_klass()) - ->as_instance_klass(); - ciEnv::_Object = - get(SystemDictionary::object_klass()) - ->as_instance_klass(); - ciEnv::_Throwable = - get(SystemDictionary::throwable_klass()) - ->as_instance_klass(); - ciEnv::_Thread = - get(SystemDictionary::thread_klass()) - ->as_instance_klass(); - ciEnv::_OutOfMemoryError = - get(SystemDictionary::OutOfMemoryError_klass()) - ->as_instance_klass(); - ciEnv::_String = - get(SystemDictionary::string_klass()) - ->as_instance_klass(); + +#define WK_KLASS_DEFN(name, ignore_s, opt) \ + if (SystemDictionary::name() != NULL) \ + ciEnv::_##name = get(SystemDictionary::name())->as_instance_klass(); + + WK_KLASSES_DO(WK_KLASS_DEFN) +#undef WK_KLASS_DEFN for (int len = -1; len != _ci_objects->length(); ) { len = _ci_objects->length(); @@ -324,13 +307,21 @@ return new (arena()) ciMethodData(h_md); } else if (o->is_instance()) { instanceHandle h_i(THREAD, (instanceOop)o); - return new (arena()) ciInstance(h_i); + if (java_dyn_CallSite::is_instance(o)) + return new (arena()) ciCallSite(h_i); + else if (java_dyn_MethodHandle::is_instance(o)) + return new (arena()) ciMethodHandle(h_i); + else + return new (arena()) ciInstance(h_i); } else if (o->is_objArray()) { objArrayHandle h_oa(THREAD, (objArrayOop)o); return new (arena()) ciObjArray(h_oa); } else if (o->is_typeArray()) { typeArrayHandle h_ta(THREAD, (typeArrayOop)o); return new (arena()) ciTypeArray(h_ta); + } else if (o->is_constantPoolCache()) { + constantPoolCacheHandle h_cpc(THREAD, (constantPoolCacheOop) o); + return new (arena()) ciCPCache(h_cpc); } // The oop is of some type not supported by the compiler interface. @@ -567,7 +558,7 @@ if (key->is_perm() && _non_perm_count == 0) { return emptyBucket; } else if (key->is_instance()) { - if (key->klass() == SystemDictionary::class_klass()) { + if (key->klass() == SystemDictionary::Class_klass()) { // class mirror instances are always perm return emptyBucket; }
--- a/src/share/vm/ci/ciStreams.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciStreams.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -186,8 +186,9 @@ // If this bytecode is a new, newarray, multianewarray, instanceof, // or checkcast, get the referenced klass. ciKlass* ciBytecodeStream::get_klass(bool& will_link) { - return CURRENT_ENV->get_klass_by_index(_holder, get_klass_index(), - will_link); + VM_ENTRY_MARK; + constantPoolHandle cpool(_method->get_methodOop()->constants()); + return CURRENT_ENV->get_klass_by_index(cpool, get_klass_index(), will_link, _holder); } // ------------------------------------------------------------------ @@ -213,7 +214,9 @@ // If this bytecode is one of the ldc variants, get the referenced // constant. ciConstant ciBytecodeStream::get_constant() { - return CURRENT_ENV->get_constant_by_index(_holder, get_constant_index()); + VM_ENTRY_MARK; + constantPoolHandle cpool(_method->get_methodOop()->constants()); + return CURRENT_ENV->get_constant_by_index(cpool, get_constant_index(), _holder); } // ------------------------------------------------------------------ @@ -264,9 +267,11 @@ // There is no "will_link" result passed back. The user is responsible // for checking linkability when retrieving the associated field. ciInstanceKlass* ciBytecodeStream::get_declared_field_holder() { + VM_ENTRY_MARK; + constantPoolHandle cpool(_method->get_methodOop()->constants()); int holder_index = get_field_holder_index(); bool ignore; - return CURRENT_ENV->get_klass_by_index(_holder, holder_index, ignore) + return CURRENT_ENV->get_klass_by_index(cpool, holder_index, ignore, _holder) ->as_instance_klass(); } @@ -277,9 +282,10 @@ // referenced by the current bytecode. Used for generating // deoptimization information. int ciBytecodeStream::get_field_holder_index() { - VM_ENTRY_MARK; - constantPoolOop cpool = _holder->get_instanceKlass()->constants(); - return cpool->klass_ref_index_at(get_field_index()); + GUARDED_VM_ENTRY( + constantPoolOop cpool = _holder->get_instanceKlass()->constants(); + return cpool->klass_ref_index_at(get_field_index()); + ) } // ------------------------------------------------------------------ @@ -321,7 +327,9 @@ // // If this is a method invocation bytecode, get the invoked method. ciMethod* ciBytecodeStream::get_method(bool& will_link) { - ciMethod* m = CURRENT_ENV->get_method_by_index(_holder, get_method_index(),cur_bc()); + VM_ENTRY_MARK; + constantPoolHandle cpool(_method->get_methodOop()->constants()); + ciMethod* m = CURRENT_ENV->get_method_by_index(cpool, get_method_index(), cur_bc(), _holder); will_link = m->is_loaded(); return m; } @@ -338,11 +346,13 @@ // There is no "will_link" result passed back. The user is responsible // for checking linkability when retrieving the associated method. ciKlass* ciBytecodeStream::get_declared_method_holder() { + VM_ENTRY_MARK; + constantPoolHandle cpool(_method->get_methodOop()->constants()); bool ignore; - // report as Dynamic for invokedynamic, which is syntactically classless + // report as InvokeDynamic for invokedynamic, which is syntactically classless if (cur_bc() == Bytecodes::_invokedynamic) - return CURRENT_ENV->get_klass_by_name(_holder, ciSymbol::java_dyn_Dynamic(), false); - return CURRENT_ENV->get_klass_by_index(_holder, get_method_holder_index(), ignore); + return CURRENT_ENV->get_klass_by_name(_holder, ciSymbol::java_dyn_InvokeDynamic(), false); + return CURRENT_ENV->get_klass_by_index(cpool, get_method_holder_index(), ignore, _holder); } // ------------------------------------------------------------------ @@ -352,8 +362,7 @@ // referenced by the current bytecode. Used for generating // deoptimization information. int ciBytecodeStream::get_method_holder_index() { - VM_ENTRY_MARK; - constantPoolOop cpool = _holder->get_instanceKlass()->constants(); + constantPoolOop cpool = _method->get_methodOop()->constants(); return cpool->klass_ref_index_at(get_method_index()); } @@ -370,3 +379,31 @@ int name_and_type_index = cpool->name_and_type_ref_index_at(method_index); return cpool->signature_ref_index_at(name_and_type_index); } + +// ------------------------------------------------------------------ +// ciBytecodeStream::get_cpcache +ciCPCache* ciBytecodeStream::get_cpcache() { + VM_ENTRY_MARK; + // Get the constant pool. + constantPoolOop cpool = _holder->get_instanceKlass()->constants(); + constantPoolCacheOop cpcache = cpool->cache(); + + return CURRENT_ENV->get_object(cpcache)->as_cpcache(); +} + +// ------------------------------------------------------------------ +// ciBytecodeStream::get_call_site +ciCallSite* ciBytecodeStream::get_call_site() { + VM_ENTRY_MARK; + // Get the constant pool. + constantPoolOop cpool = _holder->get_instanceKlass()->constants(); + constantPoolCacheOop cpcache = cpool->cache(); + + // Get the CallSite from the constant pool cache. + int method_index = get_method_index(); + ConstantPoolCacheEntry* cpcache_entry = cpcache->secondary_entry_at(method_index); + oop call_site_oop = cpcache_entry->f1(); + + // Create a CallSite object and return it. + return CURRENT_ENV->get_object(call_site_oop)->as_call_site(); +}
--- a/src/share/vm/ci/ciStreams.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciStreams.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -232,6 +232,9 @@ int get_method_holder_index(); int get_method_signature_index(); + ciCPCache* get_cpcache(); + ciCallSite* get_call_site(); + private: void assert_index_size(int required_size) const { #ifdef ASSERT
--- a/src/share/vm/ci/ciSymbol.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciSymbol.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2007 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -60,6 +60,22 @@ } // ------------------------------------------------------------------ +// ciSymbol::starts_with +// +// Tests if the symbol starts with the given prefix. +bool ciSymbol::starts_with(const char* prefix, int len) const { + GUARDED_VM_ENTRY(return get_symbolOop()->starts_with(prefix, len);) +} + +// ------------------------------------------------------------------ +// ciSymbol::index_of +// +// Determines where the symbol contains the given substring. +int ciSymbol::index_of_at(int i, const char* str, int len) const { + GUARDED_VM_ENTRY(return get_symbolOop()->index_of_at(i, str, len);) +} + +// ------------------------------------------------------------------ // ciSymbol::utf8_length int ciSymbol::utf8_length() { GUARDED_VM_ENTRY(return get_symbolOop()->utf8_length();)
--- a/src/share/vm/ci/ciSymbol.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciSymbol.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2001 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -28,6 +28,7 @@ // machine. class ciSymbol : public ciObject { CI_PACKAGE_ACCESS + // These friends all make direct use of get_symbolOop: friend class ciEnv; friend class ciInstanceKlass; friend class ciSignature; @@ -38,13 +39,13 @@ ciSymbol(symbolOop s) : ciObject(s) {} ciSymbol(symbolHandle s); // for use with vmSymbolHandles - symbolOop get_symbolOop() { return (symbolOop)get_oop(); } + symbolOop get_symbolOop() const { return (symbolOop)get_oop(); } const char* type_string() { return "ciSymbol"; } void print_impl(outputStream* st); - int byte_at(int i); + // This is public in symbolOop but private here, because the base can move: jbyte* base(); // Make a ciSymbol from a C string (implementation). @@ -55,6 +56,15 @@ const char* as_utf8(); int utf8_length(); + // Return the i-th utf8 byte, where i < utf8_length + int byte_at(int i); + + // Tests if the symbol starts with the given prefix. + bool starts_with(const char* prefix, int len) const; + + // Determines where the symbol contains the given substring. + int index_of_at(int i, const char* str, int len) const; + // What kind of ciObject is this? bool is_symbol() { return true; }
--- a/src/share/vm/ci/ciType.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciType.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -111,7 +111,7 @@ // short, etc. // Note: Bare T_ADDRESS means a raw pointer type, not a return_address. assert((uint)t < T_CONFLICT+1, "range check"); - if (t == T_OBJECT) return ciEnv::_Object; // java/lang/Object + if (t == T_OBJECT) return ciEnv::_Object_klass; // java/lang/Object assert(_basic_types[t] != NULL, "domain check"); return _basic_types[t]; }
--- a/src/share/vm/ci/ciTypeFlow.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciTypeFlow.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -635,8 +635,15 @@ ciMethod* method = str->get_method(will_link); if (!will_link) { // We weren't able to find the method. - ciKlass* unloaded_holder = method->holder(); - trap(str, unloaded_holder, str->get_method_holder_index()); + if (str->cur_bc() == Bytecodes::_invokedynamic) { + trap(str, NULL, + Deoptimization::make_trap_request + (Deoptimization::Reason_uninitialized, + Deoptimization::Action_reinterpret)); + } else { + ciKlass* unloaded_holder = method->holder(); + trap(str, unloaded_holder, str->get_method_holder_index()); + } } else { ciSignature* signature = method->signature(); ciSignatureStream sigstr(signature); @@ -1292,8 +1299,8 @@ case Bytecodes::_invokeinterface: do_invoke(str, true); break; case Bytecodes::_invokespecial: do_invoke(str, true); break; case Bytecodes::_invokestatic: do_invoke(str, false); break; - case Bytecodes::_invokevirtual: do_invoke(str, true); break; + case Bytecodes::_invokedynamic: do_invoke(str, false); break; case Bytecodes::_istore: store_local_int(str->get_index()); break; case Bytecodes::_istore_0: store_local_int(0); break;
--- a/src/share/vm/ci/ciUtilities.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/ci/ciUtilities.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -79,7 +79,7 @@ THREAD); \ if (HAS_PENDING_EXCEPTION) { \ if (PENDING_EXCEPTION->klass() == \ - SystemDictionary::threaddeath_klass()) { \ + SystemDictionary::ThreadDeath_klass()) { \ /* Kill the compilation. */ \ fatal("unhandled ci exception"); \ return (result); \
--- a/src/share/vm/classfile/classFileParser.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/classfile/classFileParser.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -430,7 +430,7 @@ case JVM_CONSTANT_UnresolvedClass : // Patching a class means pre-resolving it. // The name in the constant pool is ignored. - if (patch->klass() == SystemDictionary::class_klass()) { // %%% java_lang_Class::is_instance + if (patch->klass() == SystemDictionary::Class_klass()) { // %%% java_lang_Class::is_instance guarantee_property(!java_lang_Class::is_primitive(patch()), "Illegal class patch at %d in class file %s", index, CHECK); @@ -643,7 +643,7 @@ guarantee_property(value_type.is_int(), "Inconsistent constant value type in class file %s", CHECK); break; case T_OBJECT: - guarantee_property((cp->symbol_at(signature_index)->equals("Ljava/lang/String;", 18) + guarantee_property((cp->symbol_at(signature_index)->equals("Ljava/lang/String;") && (value_type.is_string() || value_type.is_unresolved_string())), "Bad string initial value in class file %s", CHECK); break; @@ -1718,9 +1718,7 @@ m->set_exception_table(exception_handlers()); // Copy byte codes - if (code_length > 0) { - memcpy(m->code_base(), code_start, code_length); - } + m->set_code(code_start); // Copy line number table if (linenumber_table != NULL) { @@ -2511,23 +2509,12 @@ fac_ptr->nonstatic_byte_count -= 1; (*fields_ptr)->ushort_at_put(i + instanceKlass::signature_index_offset, word_sig_index); - if (wordSize == jintSize) { - fac_ptr->nonstatic_word_count += 1; - } else { - fac_ptr->nonstatic_double_count += 1; - } - - FieldAllocationType atype = (FieldAllocationType) (*fields_ptr)->ushort_at(i+4); + fac_ptr->nonstatic_word_count += 1; + + FieldAllocationType atype = (FieldAllocationType) (*fields_ptr)->ushort_at(i + instanceKlass::low_offset); assert(atype == NONSTATIC_BYTE, ""); FieldAllocationType new_atype = NONSTATIC_WORD; - if (wordSize > jintSize) { - if (Universe::field_type_should_be_aligned(T_LONG)) { - atype = NONSTATIC_ALIGNED_DOUBLE; - } else { - atype = NONSTATIC_DOUBLE; - } - } - (*fields_ptr)->ushort_at_put(i+4, new_atype); + (*fields_ptr)->ushort_at_put(i + instanceKlass::low_offset, new_atype); found_vmentry = true; break; @@ -3085,7 +3072,7 @@ int len = fields->length(); for (int i = 0; i < len; i += instanceKlass::next_offset) { int real_offset; - FieldAllocationType atype = (FieldAllocationType) fields->ushort_at(i+4); + FieldAllocationType atype = (FieldAllocationType) fields->ushort_at(i + instanceKlass::low_offset); switch (atype) { case STATIC_OOP: real_offset = next_static_oop_offset; @@ -3173,8 +3160,8 @@ default: ShouldNotReachHere(); } - fields->short_at_put(i+4, extract_low_short_from_int(real_offset) ); - fields->short_at_put(i+5, extract_high_short_from_int(real_offset) ); + fields->short_at_put(i + instanceKlass::low_offset, extract_low_short_from_int(real_offset)); + fields->short_at_put(i + instanceKlass::high_offset, extract_high_short_from_int(real_offset)); } // Size of instances @@ -3482,8 +3469,8 @@ #endif // Check if this klass supports the java.lang.Cloneable interface - if (SystemDictionary::cloneable_klass_loaded()) { - if (k->is_subtype_of(SystemDictionary::cloneable_klass())) { + if (SystemDictionary::Cloneable_klass_loaded()) { + if (k->is_subtype_of(SystemDictionary::Cloneable_klass())) { k->set_is_cloneable(); } } @@ -3766,8 +3753,9 @@ } bool ClassFileParser::is_supported_version(u2 major, u2 minor) { - u2 max_version = JDK_Version::is_gte_jdk17x_version() ? - JAVA_MAX_SUPPORTED_VERSION : JAVA_6_VERSION; + u2 max_version = + JDK_Version::is_gte_jdk17x_version() ? JAVA_MAX_SUPPORTED_VERSION : + (JDK_Version::is_gte_jdk16x_version() ? JAVA_6_VERSION : JAVA_1_5_VERSION); return (major >= JAVA_MIN_SUPPORTED_VERSION) && (major <= max_version) && ((major != max_version) || @@ -4188,7 +4176,7 @@ // Check if ch is Java identifier start or is Java identifier part // 4672820: call java.lang.Character methods directly without generating separate tables. EXCEPTION_MARK; - instanceKlassHandle klass (THREAD, SystemDictionary::char_klass()); + instanceKlassHandle klass (THREAD, SystemDictionary::Character_klass()); // return value JavaValue result(T_BOOLEAN);
--- a/src/share/vm/classfile/classLoader.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/classfile/classLoader.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -819,7 +819,7 @@ _package_hash_table->copy_pkgnames(packages); } // Allocate objArray and fill with java.lang.String - objArrayOop r = oopFactory::new_objArray(SystemDictionary::string_klass(), + objArrayOop r = oopFactory::new_objArray(SystemDictionary::String_klass(), nof_entries, CHECK_0); objArrayHandle result(THREAD, r); for (int i = 0; i < nof_entries; i++) { @@ -1249,6 +1249,7 @@ } int ClassLoader::_compile_the_world_counter = 0; +static int _codecache_sweep_counter = 0; void ClassLoader::compile_the_world_in(char* name, Handle loader, TRAPS) { int len = (int)strlen(name); @@ -1293,6 +1294,13 @@ for (int n = 0; n < k->methods()->length(); n++) { methodHandle m (THREAD, methodOop(k->methods()->obj_at(n))); if (CompilationPolicy::canBeCompiled(m)) { + + if (++_codecache_sweep_counter == CompileTheWorldSafepointInterval) { + // Give sweeper a chance to keep up with CTW + VM_ForceSafepoint op; + VMThread::execute(&op); + _codecache_sweep_counter = 0; + } // Force compilation CompileBroker::compile_method(m, InvocationEntryBci, methodHandle(), 0, "CTW", THREAD);
--- a/src/share/vm/classfile/javaAssertions.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/classfile/javaAssertions.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -95,14 +95,14 @@ int len; typeArrayOop t; len = OptionList::count(_packages); - objArrayOop pn = oopFactory::new_objArray(SystemDictionary::string_klass(), len, CHECK_NULL); + objArrayOop pn = oopFactory::new_objArray(SystemDictionary::String_klass(), len, CHECK_NULL); objArrayHandle pkgNames (THREAD, pn); t = oopFactory::new_typeArray(T_BOOLEAN, len, CHECK_NULL); typeArrayHandle pkgEnabled(THREAD, t); fillJavaArrays(_packages, len, pkgNames, pkgEnabled, CHECK_NULL); len = OptionList::count(_classes); - objArrayOop cn = oopFactory::new_objArray(SystemDictionary::string_klass(), len, CHECK_NULL); + objArrayOop cn = oopFactory::new_objArray(SystemDictionary::String_klass(), len, CHECK_NULL); objArrayHandle classNames (THREAD, cn); t = oopFactory::new_typeArray(T_BOOLEAN, len, CHECK_NULL); typeArrayHandle classEnabled(THREAD, t);
--- a/src/share/vm/classfile/javaClasses.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/classfile/javaClasses.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -68,9 +68,9 @@ // and the char array it points to end up in the same cache line. oop obj; if (tenured) { - obj = instanceKlass::cast(SystemDictionary::string_klass())->allocate_permanent_instance(CHECK_NH); + obj = instanceKlass::cast(SystemDictionary::String_klass())->allocate_permanent_instance(CHECK_NH); } else { - obj = instanceKlass::cast(SystemDictionary::string_klass())->allocate_instance(CHECK_NH); + obj = instanceKlass::cast(SystemDictionary::String_klass())->allocate_instance(CHECK_NH); } // Create the char array. The String object must be handlized here @@ -293,7 +293,7 @@ bool java_lang_String::equals(oop java_string, jchar* chars, int len) { assert(SharedSkipVerify || - java_string->klass() == SystemDictionary::string_klass(), + java_string->klass() == SystemDictionary::String_klass(), "must be java_string"); typeArrayOop value = java_lang_String::value(java_string); int offset = java_lang_String::offset(java_string); @@ -311,7 +311,7 @@ void java_lang_String::print(Handle java_string, outputStream* st) { oop obj = java_string(); - assert(obj->klass() == SystemDictionary::string_klass(), "must be java_string"); + assert(obj->klass() == SystemDictionary::String_klass(), "must be java_string"); typeArrayOop value = java_lang_String::value(obj); int offset = java_lang_String::offset(obj); int length = java_lang_String::length(obj); @@ -339,9 +339,9 @@ // class is put into the system dictionary. int computed_modifiers = k->compute_modifier_flags(CHECK_0); k->set_modifier_flags(computed_modifiers); - if (SystemDictionary::class_klass_loaded()) { + if (SystemDictionary::Class_klass_loaded()) { // Allocate mirror (java.lang.Class instance) - Handle mirror = instanceKlass::cast(SystemDictionary::class_klass())->allocate_permanent_instance(CHECK_0); + Handle mirror = instanceKlass::cast(SystemDictionary::Class_klass())->allocate_permanent_instance(CHECK_0); // Setup indirections mirror->obj_field_put(klass_offset, k()); k->set_java_mirror(mirror()); @@ -378,7 +378,7 @@ oop java_lang_Class::create_basic_type_mirror(const char* basic_type_name, BasicType type, TRAPS) { // This should be improved by adding a field at the Java level or by // introducing a new VM klass (see comment in ClassFileParser) - oop java_class = instanceKlass::cast(SystemDictionary::class_klass())->allocate_permanent_instance(CHECK_0); + oop java_class = instanceKlass::cast(SystemDictionary::Class_klass())->allocate_permanent_instance(CHECK_0); if (type != T_VOID) { klassOop aklass = Universe::typeArrayKlassObj(type); assert(aklass != NULL, "correct bootstrap"); @@ -502,7 +502,7 @@ oop java_lang_Class::primitive_mirror(BasicType t) { oop mirror = Universe::java_mirror(t); - assert(mirror != NULL && mirror->is_a(SystemDictionary::class_klass()), "must be a Class"); + assert(mirror != NULL && mirror->is_a(SystemDictionary::Class_klass()), "must be a Class"); assert(java_lang_Class::is_primitive(mirror), "must be primitive"); return mirror; } @@ -515,14 +515,14 @@ assert(!offsets_computed, "offsets should be initialized only once"); offsets_computed = true; - klassOop k = SystemDictionary::class_klass(); + klassOop k = SystemDictionary::Class_klass(); // The classRedefinedCount field is only present starting in 1.5, // so don't go fatal. compute_optional_offset(classRedefinedCount_offset, k, vmSymbols::classRedefinedCount_name(), vmSymbols::int_signature()); // The field indicating parallelCapable (parallelLockMap) is only present starting in 7, - klassOop k1 = SystemDictionary::classloader_klass(); + klassOop k1 = SystemDictionary::ClassLoader_klass(); compute_optional_offset(parallelCapable_offset, k1, vmSymbols::parallelCapable_name(), vmSymbols::concurrenthashmap_signature()); } @@ -588,7 +588,7 @@ void java_lang_Thread::compute_offsets() { assert(_group_offset == 0, "offsets should be initialized only once"); - klassOop k = SystemDictionary::thread_klass(); + klassOop k = SystemDictionary::Thread_klass(); compute_offset(_name_offset, k, vmSymbols::name_name(), vmSymbols::char_array_signature()); compute_offset(_group_offset, k, vmSymbols::group_name(), vmSymbols::threadgroup_signature()); compute_offset(_contextClassLoader_offset, k, vmSymbols::contextClassLoader_name(), vmSymbols::classloader_signature()); @@ -847,7 +847,7 @@ void java_lang_ThreadGroup::compute_offsets() { assert(_parent_offset == 0, "offsets should be initialized only once"); - klassOop k = SystemDictionary::threadGroup_klass(); + klassOop k = SystemDictionary::ThreadGroup_klass(); compute_offset(_parent_offset, k, vmSymbols::parent_name(), vmSymbols::threadgroup_signature()); compute_offset(_name_offset, k, vmSymbols::name_name(), vmSymbols::string_signature()); @@ -1121,11 +1121,23 @@ } void flush() { + // The following appears to have been an optimization to save from + // doing a barrier for each individual store into the _methods array, + // but rather to do it for the entire array after the series of writes. + // That optimization seems to have been lost when compressed oops was + // implemented. However, the extra card-marks below was left in place, + // but is now redundant because the individual stores into the + // _methods array already execute the barrier code. CR 6918185 has + // been filed so the original code may be restored by deferring the + // barriers until after the entire sequence of stores, thus re-enabling + // the intent of the original optimization. In the meantime the redundant + // card mark below is now disabled. if (_dirty && _methods != NULL) { +#if 0 BarrierSet* bs = Universe::heap()->barrier_set(); assert(bs->has_write_ref_array_opt(), "Barrier set must have ref array opt"); - bs->write_ref_array(MemRegion((HeapWord*)_methods->base(), - _methods->array_size())); + bs->write_ref_array((HeapWord*)_methods->base(), _methods->length()); +#endif _dirty = false; } } @@ -1169,9 +1181,7 @@ method = mhandle(); } - _methods->obj_at_put(_index, method); - // bad for UseCompressedOops - // *_methods->obj_at_addr(_index) = method; + _methods->obj_at_put(_index, method); _bcis->ushort_at_put(_index, bci); _index++; _dirty = true; @@ -1345,7 +1355,7 @@ // No-op if stack trace is disabled if (!StackTraceInThrowable) return; - assert(throwable->is_a(SystemDictionary::throwable_klass()), "sanity check"); + assert(throwable->is_a(SystemDictionary::Throwable_klass()), "sanity check"); oop backtrace = java_lang_Throwable::backtrace(throwable()); assert(backtrace != NULL, "backtrace not preallocated"); @@ -1450,7 +1460,7 @@ assert(JDK_Version::is_gte_jdk14x_version(), "should only be called in >= 1.4"); // Allocate java.lang.StackTraceElement instance - klassOop k = SystemDictionary::stackTraceElement_klass(); + klassOop k = SystemDictionary::StackTraceElement_klass(); assert(k != NULL, "must be loaded in 1.4+"); instanceKlassHandle ik (THREAD, k); if (ik->should_be_initialized()) { @@ -1488,7 +1498,7 @@ void java_lang_reflect_AccessibleObject::compute_offsets() { - klassOop k = SystemDictionary::reflect_accessible_object_klass(); + klassOop k = SystemDictionary::reflect_AccessibleObject_klass(); compute_offset(override_offset, k, vmSymbols::override_name(), vmSymbols::bool_signature()); } @@ -1503,7 +1513,7 @@ } void java_lang_reflect_Method::compute_offsets() { - klassOop k = SystemDictionary::reflect_method_klass(); + klassOop k = SystemDictionary::reflect_Method_klass(); compute_offset(clazz_offset, k, vmSymbols::clazz_name(), vmSymbols::class_signature()); compute_offset(name_offset, k, vmSymbols::name_name(), vmSymbols::string_signature()); compute_offset(returnType_offset, k, vmSymbols::returnType_name(), vmSymbols::class_signature()); @@ -1524,7 +1534,7 @@ Handle java_lang_reflect_Method::create(TRAPS) { assert(Universe::is_fully_initialized(), "Need to find another solution to the reflection problem"); - klassOop klass = SystemDictionary::reflect_method_klass(); + klassOop klass = SystemDictionary::reflect_Method_klass(); // This class is eagerly initialized during VM initialization, since we keep a refence // to one of the methods assert(instanceKlass::cast(klass)->is_initialized(), "must be initialized"); @@ -1666,7 +1676,7 @@ } void java_lang_reflect_Constructor::compute_offsets() { - klassOop k = SystemDictionary::reflect_constructor_klass(); + klassOop k = SystemDictionary::reflect_Constructor_klass(); compute_offset(clazz_offset, k, vmSymbols::clazz_name(), vmSymbols::class_signature()); compute_offset(parameterTypes_offset, k, vmSymbols::parameterTypes_name(), vmSymbols::class_array_signature()); compute_offset(exceptionTypes_offset, k, vmSymbols::exceptionTypes_name(), vmSymbols::class_array_signature()); @@ -1790,7 +1800,7 @@ } void java_lang_reflect_Field::compute_offsets() { - klassOop k = SystemDictionary::reflect_field_klass(); + klassOop k = SystemDictionary::reflect_Field_klass(); compute_offset(clazz_offset, k, vmSymbols::clazz_name(), vmSymbols::class_signature()); compute_offset(name_offset, k, vmSymbols::name_name(), vmSymbols::string_signature()); compute_offset(type_offset, k, vmSymbols::type_name(), vmSymbols::class_signature()); @@ -1897,7 +1907,7 @@ void sun_reflect_ConstantPool::compute_offsets() { - klassOop k = SystemDictionary::reflect_constant_pool_klass(); + klassOop k = SystemDictionary::reflect_ConstantPool_klass(); // This null test can be removed post beta if (k != NULL) { compute_offset(_cp_oop_offset, k, vmSymbols::constantPoolOop_name(), vmSymbols::object_signature()); @@ -1907,7 +1917,7 @@ Handle sun_reflect_ConstantPool::create(TRAPS) { assert(Universe::is_fully_initialized(), "Need to find another solution to the reflection problem"); - klassOop k = SystemDictionary::reflect_constant_pool_klass(); + klassOop k = SystemDictionary::reflect_ConstantPool_klass(); instanceKlassHandle klass (THREAD, k); // Ensure it is initialized klass->initialize(CHECK_NH); @@ -1927,7 +1937,7 @@ } void sun_reflect_UnsafeStaticFieldAccessorImpl::compute_offsets() { - klassOop k = SystemDictionary::reflect_unsafe_static_field_accessor_impl_klass(); + klassOop k = SystemDictionary::reflect_UnsafeStaticFieldAccessorImpl_klass(); // This null test can be removed post beta if (k != NULL) { compute_offset(_base_offset, k, @@ -2073,7 +2083,7 @@ // Support for java_lang_ref_Reference oop java_lang_ref_Reference::pending_list_lock() { - instanceKlass* ik = instanceKlass::cast(SystemDictionary::reference_klass()); + instanceKlass* ik = instanceKlass::cast(SystemDictionary::Reference_klass()); char *addr = (((char *)ik->start_of_static_fields()) + static_lock_offset); if (UseCompressedOops) { return oopDesc::load_decode_heap_oop((narrowOop *)addr); @@ -2083,7 +2093,7 @@ } HeapWord *java_lang_ref_Reference::pending_list_addr() { - instanceKlass* ik = instanceKlass::cast(SystemDictionary::reference_klass()); + instanceKlass* ik = instanceKlass::cast(SystemDictionary::Reference_klass()); char *addr = (((char *)ik->start_of_static_fields()) + static_pending_offset); // XXX This might not be HeapWord aligned, almost rather be char *. return (HeapWord*)addr; @@ -2106,17 +2116,17 @@ } jlong java_lang_ref_SoftReference::clock() { - instanceKlass* ik = instanceKlass::cast(SystemDictionary::soft_reference_klass()); + instanceKlass* ik = instanceKlass::cast(SystemDictionary::SoftReference_klass()); int offset = ik->offset_of_static_fields() + static_clock_offset; - return SystemDictionary::soft_reference_klass()->long_field(offset); + return SystemDictionary::SoftReference_klass()->long_field(offset); } void java_lang_ref_SoftReference::set_clock(jlong value) { - instanceKlass* ik = instanceKlass::cast(SystemDictionary::soft_reference_klass()); + instanceKlass* ik = instanceKlass::cast(SystemDictionary::SoftReference_klass()); int offset = ik->offset_of_static_fields() + static_clock_offset; - SystemDictionary::soft_reference_klass()->long_field_put(offset, value); + SystemDictionary::SoftReference_klass()->long_field_put(offset, value); } @@ -2404,6 +2414,10 @@ return ptypes(mt)->obj_at(idx); } +int java_dyn_MethodType::ptype_count(oop mt) { + return ptypes(mt)->length(); +} + // Support for java_dyn_MethodTypeForm @@ -2430,15 +2444,15 @@ } -// Support for sun_dyn_CallSiteImpl - -int sun_dyn_CallSiteImpl::_type_offset; -int sun_dyn_CallSiteImpl::_target_offset; -int sun_dyn_CallSiteImpl::_vmmethod_offset; - -void sun_dyn_CallSiteImpl::compute_offsets() { +// Support for java_dyn_CallSite + +int java_dyn_CallSite::_type_offset; +int java_dyn_CallSite::_target_offset; +int java_dyn_CallSite::_vmmethod_offset; + +void java_dyn_CallSite::compute_offsets() { if (!EnableInvokeDynamic) return; - klassOop k = SystemDictionary::CallSiteImpl_klass(); + klassOop k = SystemDictionary::CallSite_klass(); if (k != NULL) { compute_offset(_type_offset, k, vmSymbols::type_name(), vmSymbols::java_dyn_MethodType_signature(), true); compute_offset(_target_offset, k, vmSymbols::target_name(), vmSymbols::java_dyn_MethodHandle_signature(), true); @@ -2446,23 +2460,23 @@ } } -oop sun_dyn_CallSiteImpl::type(oop site) { +oop java_dyn_CallSite::type(oop site) { return site->obj_field(_type_offset); } -oop sun_dyn_CallSiteImpl::target(oop site) { +oop java_dyn_CallSite::target(oop site) { return site->obj_field(_target_offset); } -void sun_dyn_CallSiteImpl::set_target(oop site, oop target) { +void java_dyn_CallSite::set_target(oop site, oop target) { site->obj_field_put(_target_offset, target); } -oop sun_dyn_CallSiteImpl::vmmethod(oop site) { +oop java_dyn_CallSite::vmmethod(oop site) { return site->obj_field(_vmmethod_offset); } -void sun_dyn_CallSiteImpl::set_vmmethod(oop site, oop ref) { +void java_dyn_CallSite::set_vmmethod(oop site, oop ref) { site->obj_field_put(_vmmethod_offset, ref); } @@ -2535,7 +2549,7 @@ // the generated bytecodes for reflection, and if so, "magically" // delegate to its parent to prevent class loading from occurring // in places where applications using reflection didn't expect it. - klassOop delegating_cl_class = SystemDictionary::reflect_delegating_classloader_klass(); + klassOop delegating_cl_class = SystemDictionary::reflect_DelegatingClassLoader_klass(); // This might be null in non-1.4 JDKs if (delegating_cl_class != NULL && loader->is_a(delegating_cl_class)) { return parent(loader); @@ -2550,7 +2564,7 @@ void java_lang_System::compute_offsets() { assert(offset_of_static_fields == 0, "offsets should be initialized only once"); - instanceKlass* ik = instanceKlass::cast(SystemDictionary::system_klass()); + instanceKlass* ik = instanceKlass::cast(SystemDictionary::System_klass()); offset_of_static_fields = ik->offset_of_static_fields(); } @@ -2811,7 +2825,7 @@ java_dyn_MethodTypeForm::compute_offsets(); } if (EnableInvokeDynamic) { - sun_dyn_CallSiteImpl::compute_offsets(); + java_dyn_CallSite::compute_offsets(); } java_security_AccessControlContext::compute_offsets(); // Initialize reflection classes. The layouts of these classes
--- a/src/share/vm/classfile/javaClasses.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/classfile/javaClasses.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -111,7 +111,7 @@ // Testers static bool is_instance(oop obj) { - return obj != NULL && obj->klass() == SystemDictionary::string_klass(); + return obj != NULL && obj->klass() == SystemDictionary::String_klass(); } // Debugging @@ -161,7 +161,7 @@ static void print_signature(oop java_class, outputStream *st); // Testing static bool is_instance(oop obj) { - return obj != NULL && obj->klass() == SystemDictionary::class_klass(); + return obj != NULL && obj->klass() == SystemDictionary::Class_klass(); } static bool is_primitive(oop java_class); static BasicType primitive_type(oop java_class); @@ -1027,6 +1027,7 @@ static oop form(oop mt); static oop ptype(oop mt, int index); + static int ptype_count(oop mt); static symbolOop as_signature(oop mt, bool intern_if_not_found, TRAPS); static void print_signature(oop mt, outputStream* st); @@ -1061,9 +1062,9 @@ }; -// Interface to sun.dyn.CallSiteImpl objects +// Interface to java.dyn.CallSite objects -class sun_dyn_CallSiteImpl: AllStatic { +class java_dyn_CallSite: AllStatic { friend class JavaClasses; private: @@ -1083,6 +1084,14 @@ static oop vmmethod(oop site); static void set_vmmethod(oop site, oop ref); + // Testers + static bool is_subclass(klassOop klass) { + return Klass::cast(klass)->is_subclass_of(SystemDictionary::CallSite_klass()); + } + static bool is_instance(oop obj) { + return obj != NULL && is_subclass(obj->klass()); + } + // Accessors for code generation: static int target_offset_in_bytes() { return _target_offset; } static int type_offset_in_bytes() { return _type_offset; }
--- a/src/share/vm/classfile/loaderConstraints.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/classfile/loaderConstraints.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -430,7 +430,8 @@ } -void LoaderConstraintTable::verify(Dictionary* dictionary) { +void LoaderConstraintTable::verify(Dictionary* dictionary, + PlaceholderTable* placeholders) { Thread *thread = Thread::current(); for (int cindex = 0; cindex < _loader_constraint_size; cindex++) { for (LoaderConstraintEntry* probe = bucket(cindex); @@ -445,7 +446,23 @@ unsigned int d_hash = dictionary->compute_hash(name, loader); int d_index = dictionary->hash_to_index(d_hash); klassOop k = dictionary->find_class(d_index, d_hash, name, loader); - guarantee(k == probe->klass(), "klass should be in dictionary"); + if (k != NULL) { + // We found the class in the system dictionary, so we should + // make sure that the klassOop matches what we already have. + guarantee(k == probe->klass(), "klass should be in dictionary"); + } else { + // If we don't find the class in the system dictionary, it + // has to be in the placeholders table. + unsigned int p_hash = placeholders->compute_hash(name, loader); + int p_index = placeholders->hash_to_index(p_hash); + PlaceholderEntry* entry = placeholders->get_entry(p_index, p_hash, + name, loader); + + // The instanceKlass might not be on the entry, so the only + // thing we can check here is whether we were successful in + // finding the class in the placeholders table. + guarantee(entry != NULL, "klass should be in the placeholders"); + } } for (int n = 0; n< probe->num_loaders(); n++) { guarantee(probe->loader(n)->is_oop_or_null(), "should be oop");
--- a/src/share/vm/classfile/loaderConstraints.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/classfile/loaderConstraints.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -81,7 +81,7 @@ void purge_loader_constraints(BoolObjectClosure* is_alive); - void verify(Dictionary* dictionary); + void verify(Dictionary* dictionary, PlaceholderTable* placeholders); #ifndef PRODUCT void print(); #endif
--- a/src/share/vm/classfile/systemDictionary.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/classfile/systemDictionary.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -60,10 +60,10 @@ } void SystemDictionary::compute_java_system_loader(TRAPS) { - KlassHandle system_klass(THREAD, WK_KLASS(classloader_klass)); + KlassHandle system_klass(THREAD, WK_KLASS(ClassLoader_klass)); JavaValue result(T_OBJECT); JavaCalls::call_static(&result, - KlassHandle(THREAD, WK_KLASS(classloader_klass)), + KlassHandle(THREAD, WK_KLASS(ClassLoader_klass)), vmSymbolHandles::getSystemClassLoader_name(), vmSymbolHandles::void_classloader_signature(), CHECK); @@ -99,6 +99,15 @@ return java_lang_Class::parallelCapable(class_loader()); } // ---------------------------------------------------------------------------- +// ParallelDefineClass flag does not apply to bootclass loader +bool SystemDictionary::is_parallelDefine(Handle class_loader) { + if (class_loader.is_null()) return false; + if (AllowParallelDefineClass && java_lang_Class::parallelCapable(class_loader())) { + return true; + } + return false; +} +// ---------------------------------------------------------------------------- // Resolving of classes // Forwards to resolve_or_null @@ -119,7 +128,7 @@ // in which case we have to check whether the pending exception is a ClassNotFoundException, // and if so convert it to a NoClassDefFoundError // And chain the original ClassNotFoundException - if (throw_error && PENDING_EXCEPTION->is_a(SystemDictionary::classNotFoundException_klass())) { + if (throw_error && PENDING_EXCEPTION->is_a(SystemDictionary::ClassNotFoundException_klass())) { ResourceMark rm(THREAD); assert(klass_h() == NULL, "Should not have result with exception pending"); Handle e(THREAD, PENDING_EXCEPTION); @@ -350,7 +359,7 @@ assert(class_loader() != NULL, "should not have non-null protection domain for null classloader"); - KlassHandle system_loader(THREAD, SystemDictionary::classloader_klass()); + KlassHandle system_loader(THREAD, SystemDictionary::ClassLoader_klass()); JavaCalls::call_special(&result, class_loader, system_loader, @@ -724,17 +733,17 @@ // Do actual loading k = load_instance_class(name, class_loader, THREAD); - // For UnsyncloadClass and AllowParallelDefineClass only: + // For UnsyncloadClass only // If they got a linkageError, check if a parallel class load succeeded. // If it did, then for bytecode resolution the specification requires // that we return the same result we did for the other thread, i.e. the // successfully loaded instanceKlass // Should not get here for classloaders that support parallelism - // with the new cleaner mechanism + // with the new cleaner mechanism, even with AllowParallelDefineClass // Bootstrap goes through here to allow for an extra guarantee check if (UnsyncloadClass || (class_loader.is_null())) { if (k.is_null() && HAS_PENDING_EXCEPTION - && PENDING_EXCEPTION->is_a(SystemDictionary::linkageError_klass())) { + && PENDING_EXCEPTION->is_a(SystemDictionary::LinkageError_klass())) { MutexLocker mu(SystemDictionary_lock, THREAD); klassOop check = find_class(d_index, d_hash, name, class_loader); if (check != NULL) { @@ -1358,7 +1367,7 @@ JavaValue result(T_OBJECT); - KlassHandle spec_klass (THREAD, SystemDictionary::classloader_klass()); + KlassHandle spec_klass (THREAD, SystemDictionary::ClassLoader_klass()); // Call public unsynchronized loadClass(String) directly for all class loaders // for parallelCapable class loaders. JDK >=7, loadClass(String, boolean) will @@ -1483,14 +1492,17 @@ } // Support parallel classloading -// Initial implementation for bootstrap classloader -// For custom class loaders that support parallel classloading, +// All parallel class loaders, including bootstrap classloader +// lock a placeholder entry for this class/class_loader pair +// to allow parallel defines of different classes for this class loader // With AllowParallelDefine flag==true, in case they do not synchronize around // FindLoadedClass/DefineClass, calls, we check for parallel // loading for them, wait if a defineClass is in progress // and return the initial requestor's results +// This flag does not apply to the bootstrap classloader. // With AllowParallelDefine flag==false, call through to define_instance_class // which will throw LinkageError: duplicate class definition. +// False is the requested default. // For better performance, the class loaders should synchronize // findClass(), i.e. FindLoadedClass/DefineClassIfAbsent or they // potentially waste time reading and parsing the bytestream. @@ -1511,9 +1523,11 @@ { MutexLocker mu(SystemDictionary_lock, THREAD); // First check if class already defined - klassOop check = find_class(d_index, d_hash, name_h, class_loader); - if (check != NULL) { - return(instanceKlassHandle(THREAD, check)); + if (UnsyncloadClass || (is_parallelDefine(class_loader))) { + klassOop check = find_class(d_index, d_hash, name_h, class_loader); + if (check != NULL) { + return(instanceKlassHandle(THREAD, check)); + } } // Acquire define token for this class/classloader @@ -1529,7 +1543,7 @@ // Only special cases allow parallel defines and can use other thread's results // Other cases fall through, and may run into duplicate defines // caught by finding an entry in the SystemDictionary - if ((UnsyncloadClass || AllowParallelDefineClass) && (probe->instanceKlass() != NULL)) { + if ((UnsyncloadClass || is_parallelDefine(class_loader)) && (probe->instanceKlass() != NULL)) { probe->remove_seen_thread(THREAD, PlaceholderTable::DEFINE_CLASS); placeholders()->find_and_remove(p_index, p_hash, name_h, class_loader, THREAD); SystemDictionary_lock->notify_all(); @@ -1930,13 +1944,13 @@ void SystemDictionary::initialize_preloaded_classes(TRAPS) { - assert(WK_KLASS(object_klass) == NULL, "preloaded classes should only be initialized once"); + assert(WK_KLASS(Object_klass) == NULL, "preloaded classes should only be initialized once"); // Preload commonly used klasses WKID scan = FIRST_WKID; // first do Object, String, Class - initialize_wk_klasses_through(WK_KLASS_ENUM_NAME(class_klass), scan, CHECK); + initialize_wk_klasses_through(WK_KLASS_ENUM_NAME(Class_klass), scan, CHECK); - debug_only(instanceKlass::verify_class_klass_nonstatic_oop_maps(WK_KLASS(class_klass))); + debug_only(instanceKlass::verify_class_klass_nonstatic_oop_maps(WK_KLASS(Class_klass))); // Fixup mirrors for classes loaded before java.lang.Class. // These calls iterate over the objects currently in the perm gen @@ -1947,17 +1961,17 @@ Universe::fixup_mirrors(CHECK); // do a bunch more: - initialize_wk_klasses_through(WK_KLASS_ENUM_NAME(reference_klass), scan, CHECK); + initialize_wk_klasses_through(WK_KLASS_ENUM_NAME(Reference_klass), scan, CHECK); // Preload ref klasses and set reference types - instanceKlass::cast(WK_KLASS(reference_klass))->set_reference_type(REF_OTHER); - instanceRefKlass::update_nonstatic_oop_maps(WK_KLASS(reference_klass)); + instanceKlass::cast(WK_KLASS(Reference_klass))->set_reference_type(REF_OTHER); + instanceRefKlass::update_nonstatic_oop_maps(WK_KLASS(Reference_klass)); - initialize_wk_klasses_through(WK_KLASS_ENUM_NAME(phantom_reference_klass), scan, CHECK); - instanceKlass::cast(WK_KLASS(soft_reference_klass))->set_reference_type(REF_SOFT); - instanceKlass::cast(WK_KLASS(weak_reference_klass))->set_reference_type(REF_WEAK); - instanceKlass::cast(WK_KLASS(final_reference_klass))->set_reference_type(REF_FINAL); - instanceKlass::cast(WK_KLASS(phantom_reference_klass))->set_reference_type(REF_PHANTOM); + initialize_wk_klasses_through(WK_KLASS_ENUM_NAME(PhantomReference_klass), scan, CHECK); + instanceKlass::cast(WK_KLASS(SoftReference_klass))->set_reference_type(REF_SOFT); + instanceKlass::cast(WK_KLASS(WeakReference_klass))->set_reference_type(REF_WEAK); + instanceKlass::cast(WK_KLASS(FinalReference_klass))->set_reference_type(REF_FINAL); + instanceKlass::cast(WK_KLASS(PhantomReference_klass))->set_reference_type(REF_PHANTOM); WKID meth_group_start = WK_KLASS_ENUM_NAME(MethodHandle_klass); WKID meth_group_end = WK_KLASS_ENUM_NAME(WrongMethodTypeException_klass); @@ -1970,10 +1984,10 @@ scan = WKID(meth_group_end+1); } WKID indy_group_start = WK_KLASS_ENUM_NAME(Linkage_klass); - WKID indy_group_end = WK_KLASS_ENUM_NAME(Dynamic_klass); + WKID indy_group_end = WK_KLASS_ENUM_NAME(InvokeDynamic_klass); initialize_wk_klasses_until(indy_group_start, scan, CHECK); if (EnableInvokeDynamic) { - initialize_wk_klasses_through(indy_group_start, scan, CHECK); + initialize_wk_klasses_through(indy_group_end, scan, CHECK); } if (_well_known_klasses[indy_group_start] == NULL) { // Skip the rest of the dynamic typing classes, if Linkage is not loaded. @@ -1982,14 +1996,14 @@ initialize_wk_klasses_until(WKID_LIMIT, scan, CHECK); - _box_klasses[T_BOOLEAN] = WK_KLASS(boolean_klass); - _box_klasses[T_CHAR] = WK_KLASS(char_klass); - _box_klasses[T_FLOAT] = WK_KLASS(float_klass); - _box_klasses[T_DOUBLE] = WK_KLASS(double_klass); - _box_klasses[T_BYTE] = WK_KLASS(byte_klass); - _box_klasses[T_SHORT] = WK_KLASS(short_klass); - _box_klasses[T_INT] = WK_KLASS(int_klass); - _box_klasses[T_LONG] = WK_KLASS(long_klass); + _box_klasses[T_BOOLEAN] = WK_KLASS(Boolean_klass); + _box_klasses[T_CHAR] = WK_KLASS(Character_klass); + _box_klasses[T_FLOAT] = WK_KLASS(Float_klass); + _box_klasses[T_DOUBLE] = WK_KLASS(Double_klass); + _box_klasses[T_BYTE] = WK_KLASS(Byte_klass); + _box_klasses[T_SHORT] = WK_KLASS(Short_klass); + _box_klasses[T_INT] = WK_KLASS(Integer_klass); + _box_klasses[T_LONG] = WK_KLASS(Long_klass); //_box_klasses[T_OBJECT] = WK_KLASS(object_klass); //_box_klasses[T_ARRAY] = WK_KLASS(object_klass); @@ -2000,11 +2014,11 @@ #endif // KERNEL { // Compute whether we should use loadClass or loadClassInternal when loading classes. - methodOop method = instanceKlass::cast(classloader_klass())->find_method(vmSymbols::loadClassInternal_name(), vmSymbols::string_class_signature()); + methodOop method = instanceKlass::cast(ClassLoader_klass())->find_method(vmSymbols::loadClassInternal_name(), vmSymbols::string_class_signature()); _has_loadClassInternal = (method != NULL); } { // Compute whether we should use checkPackageAccess or NOT - methodOop method = instanceKlass::cast(classloader_klass())->find_method(vmSymbols::checkPackageAccess_name(), vmSymbols::class_protectiondomain_signature()); + methodOop method = instanceKlass::cast(ClassLoader_klass())->find_method(vmSymbols::checkPackageAccess_name(), vmSymbols::class_protectiondomain_signature()); _has_checkPackageAccess = (method != NULL); } } @@ -2343,6 +2357,8 @@ SymbolPropertyEntry* spe = invoke_method_table()->find_entry(index, hash, signature); if (spe == NULL || spe->property_oop() == NULL) { // Must create lots of stuff here, but outside of the SystemDictionary lock. + if (THREAD->is_Compiler_thread()) + return NULL; // do not attempt from within compiler Handle mt = compute_method_handle_type(signature(), class_loader, protection_domain, CHECK_NULL); @@ -2375,7 +2391,7 @@ TRAPS) { Handle empty; int npts = ArgumentCount(signature()).size(); - objArrayHandle pts = oopFactory::new_objArray(SystemDictionary::class_klass(), npts, CHECK_(empty)); + objArrayHandle pts = oopFactory::new_objArray(SystemDictionary::Class_klass(), npts, CHECK_(empty)); int arg = 0; Handle rt; // the return type from the signature for (SignatureStream ss(signature()); !ss.is_done(); ss.next()) { @@ -2421,7 +2437,7 @@ methodHandle mh_invdyn, TRAPS) { Handle empty; - // call sun.dyn.CallSiteImpl::makeSite(caller, name, mtype, cmid, cbci) + // call java.dyn.CallSite::makeSite(caller, name, mtype, cmid, cbci) oop name_str_oop = StringTable::intern(name(), CHECK_(empty)); // not a handle! JavaCallArguments args(Handle(THREAD, caller->java_mirror())); args.push_oop(name_str_oop); @@ -2430,17 +2446,19 @@ args.push_int(caller_bci); JavaValue result(T_OBJECT); JavaCalls::call_static(&result, - SystemDictionary::CallSiteImpl_klass(), + SystemDictionary::CallSite_klass(), vmSymbols::makeSite_name(), vmSymbols::makeSite_signature(), &args, CHECK_(empty)); oop call_site_oop = (oop) result.get_jobject(); assert(call_site_oop->is_oop() - /*&& sun_dyn_CallSiteImpl::is_instance(call_site_oop)*/, "must be sane"); - sun_dyn_CallSiteImpl::set_vmmethod(call_site_oop, mh_invdyn()); + /*&& java_dyn_CallSite::is_instance(call_site_oop)*/, "must be sane"); + java_dyn_CallSite::set_vmmethod(call_site_oop, mh_invdyn()); if (TraceMethodHandles) { +#ifndef PRODUCT tty->print_cr("Linked invokedynamic bci=%d site="INTPTR_FORMAT":", caller_bci, call_site_oop); call_site_oop->print(); tty->cr(); +#endif //PRODUCT } return call_site_oop; } @@ -2453,9 +2471,17 @@ instanceKlassHandle ik(THREAD, caller()); - if (ik->bootstrap_method() != NULL) { - return Handle(THREAD, ik->bootstrap_method()); + oop boot_method_oop = ik->bootstrap_method(); + if (boot_method_oop != NULL) { + if (TraceMethodHandles) { + tty->print_cr("bootstrap method for "PTR_FORMAT" cached as "PTR_FORMAT":", ik(), boot_method_oop); + } + NOT_PRODUCT(if (!boot_method_oop->is_oop()) { tty->print_cr("*** boot MH of "PTR_FORMAT" = "PTR_FORMAT, ik(), boot_method_oop); ik()->print(); }); + assert(boot_method_oop->is_oop() + && java_dyn_MethodHandle::is_instance(boot_method_oop), "must be sane"); + return Handle(THREAD, boot_method_oop); } + boot_method_oop = NULL; // GC safety // call java.dyn.Linkage::findBootstrapMethod(caller, sbk) JavaCallArguments args(Handle(THREAD, ik->java_mirror())); @@ -2469,9 +2495,18 @@ vmSymbols::findBootstrapMethod_name(), vmSymbols::findBootstrapMethod_signature(), &args, CHECK_(empty)); - oop boot_method_oop = (oop) result.get_jobject(); + boot_method_oop = (oop) result.get_jobject(); if (boot_method_oop != NULL) { + if (TraceMethodHandles) { +#ifndef PRODUCT + tty->print_cr("--------"); + tty->print_cr("bootstrap method for "PTR_FORMAT" computed as "PTR_FORMAT":", ik(), boot_method_oop); + ik()->print(); + boot_method_oop->print(); + tty->print_cr("========"); +#endif //PRODUCT + } assert(boot_method_oop->is_oop() && java_dyn_MethodHandle::is_instance(boot_method_oop), "must be sane"); // probably no race conditions, but let's be careful: @@ -2480,6 +2515,14 @@ else boot_method_oop = ik->bootstrap_method(); } else { + if (TraceMethodHandles) { +#ifndef PRODUCT + tty->print_cr("--------"); + tty->print_cr("bootstrap method for "PTR_FORMAT" computed as NULL:", ik()); + ik()->print(); + tty->print_cr("========"); +#endif //PRODUCT + } boot_method_oop = ik->bootstrap_method(); } @@ -2547,7 +2590,7 @@ // Verify constraint table guarantee(constraints() != NULL, "Verify of loader constraints failed"); - constraints()->verify(dictionary()); + constraints()->verify(dictionary(), placeholders()); }
--- a/src/share/vm/classfile/systemDictionary.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/classfile/systemDictionary.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -82,55 +82,55 @@ #define WK_KLASSES_DO(template) \ /* well-known classes */ \ - template(object_klass, java_lang_Object, Pre) \ - template(string_klass, java_lang_String, Pre) \ - template(class_klass, java_lang_Class, Pre) \ - template(cloneable_klass, java_lang_Cloneable, Pre) \ - template(classloader_klass, java_lang_ClassLoader, Pre) \ - template(serializable_klass, java_io_Serializable, Pre) \ - template(system_klass, java_lang_System, Pre) \ - template(throwable_klass, java_lang_Throwable, Pre) \ - template(error_klass, java_lang_Error, Pre) \ - template(threaddeath_klass, java_lang_ThreadDeath, Pre) \ - template(exception_klass, java_lang_Exception, Pre) \ - template(runtime_exception_klass, java_lang_RuntimeException, Pre) \ - template(protectionDomain_klass, java_security_ProtectionDomain, Pre) \ + template(Object_klass, java_lang_Object, Pre) \ + template(String_klass, java_lang_String, Pre) \ + template(Class_klass, java_lang_Class, Pre) \ + template(Cloneable_klass, java_lang_Cloneable, Pre) \ + template(ClassLoader_klass, java_lang_ClassLoader, Pre) \ + template(Serializable_klass, java_io_Serializable, Pre) \ + template(System_klass, java_lang_System, Pre) \ + template(Throwable_klass, java_lang_Throwable, Pre) \ + template(Error_klass, java_lang_Error, Pre) \ + template(ThreadDeath_klass, java_lang_ThreadDeath, Pre) \ + template(Exception_klass, java_lang_Exception, Pre) \ + template(RuntimeException_klass, java_lang_RuntimeException, Pre) \ + template(ProtectionDomain_klass, java_security_ProtectionDomain, Pre) \ template(AccessControlContext_klass, java_security_AccessControlContext, Pre) \ - template(classNotFoundException_klass, java_lang_ClassNotFoundException, Pre) \ - template(noClassDefFoundError_klass, java_lang_NoClassDefFoundError, Pre) \ - template(linkageError_klass, java_lang_LinkageError, Pre) \ + template(ClassNotFoundException_klass, java_lang_ClassNotFoundException, Pre) \ + template(NoClassDefFoundError_klass, java_lang_NoClassDefFoundError, Pre) \ + template(LinkageError_klass, java_lang_LinkageError, Pre) \ template(ClassCastException_klass, java_lang_ClassCastException, Pre) \ template(ArrayStoreException_klass, java_lang_ArrayStoreException, Pre) \ - template(virtualMachineError_klass, java_lang_VirtualMachineError, Pre) \ + template(VirtualMachineError_klass, java_lang_VirtualMachineError, Pre) \ template(OutOfMemoryError_klass, java_lang_OutOfMemoryError, Pre) \ template(StackOverflowError_klass, java_lang_StackOverflowError, Pre) \ template(IllegalMonitorStateException_klass, java_lang_IllegalMonitorStateException, Pre) \ - template(reference_klass, java_lang_ref_Reference, Pre) \ + template(Reference_klass, java_lang_ref_Reference, Pre) \ \ /* Preload ref klasses and set reference types */ \ - template(soft_reference_klass, java_lang_ref_SoftReference, Pre) \ - template(weak_reference_klass, java_lang_ref_WeakReference, Pre) \ - template(final_reference_klass, java_lang_ref_FinalReference, Pre) \ - template(phantom_reference_klass, java_lang_ref_PhantomReference, Pre) \ - template(finalizer_klass, java_lang_ref_Finalizer, Pre) \ + template(SoftReference_klass, java_lang_ref_SoftReference, Pre) \ + template(WeakReference_klass, java_lang_ref_WeakReference, Pre) \ + template(FinalReference_klass, java_lang_ref_FinalReference, Pre) \ + template(PhantomReference_klass, java_lang_ref_PhantomReference, Pre) \ + template(Finalizer_klass, java_lang_ref_Finalizer, Pre) \ \ - template(thread_klass, java_lang_Thread, Pre) \ - template(threadGroup_klass, java_lang_ThreadGroup, Pre) \ - template(properties_klass, java_util_Properties, Pre) \ - template(reflect_accessible_object_klass, java_lang_reflect_AccessibleObject, Pre) \ - template(reflect_field_klass, java_lang_reflect_Field, Pre) \ - template(reflect_method_klass, java_lang_reflect_Method, Pre) \ - template(reflect_constructor_klass, java_lang_reflect_Constructor, Pre) \ + template(Thread_klass, java_lang_Thread, Pre) \ + template(ThreadGroup_klass, java_lang_ThreadGroup, Pre) \ + template(Properties_klass, java_util_Properties, Pre) \ + template(reflect_AccessibleObject_klass, java_lang_reflect_AccessibleObject, Pre) \ + template(reflect_Field_klass, java_lang_reflect_Field, Pre) \ + template(reflect_Method_klass, java_lang_reflect_Method, Pre) \ + template(reflect_Constructor_klass, java_lang_reflect_Constructor, Pre) \ \ /* NOTE: needed too early in bootstrapping process to have checks based on JDK version */ \ /* Universe::is_gte_jdk14x_version() is not set up by this point. */ \ /* It's okay if this turns out to be NULL in non-1.4 JDKs. */ \ - template(reflect_magic_klass, sun_reflect_MagicAccessorImpl, Opt) \ - template(reflect_method_accessor_klass, sun_reflect_MethodAccessorImpl, Opt_Only_JDK14NewRef) \ - template(reflect_constructor_accessor_klass, sun_reflect_ConstructorAccessorImpl, Opt_Only_JDK14NewRef) \ - template(reflect_delegating_classloader_klass, sun_reflect_DelegatingClassLoader, Opt) \ - template(reflect_constant_pool_klass, sun_reflect_ConstantPool, Opt_Only_JDK15) \ - template(reflect_unsafe_static_field_accessor_impl_klass, sun_reflect_UnsafeStaticFieldAccessorImpl, Opt_Only_JDK15) \ + template(reflect_MagicAccessorImpl_klass, sun_reflect_MagicAccessorImpl, Opt) \ + template(reflect_MethodAccessorImpl_klass, sun_reflect_MethodAccessorImpl, Opt_Only_JDK14NewRef) \ + template(reflect_ConstructorAccessorImpl_klass, sun_reflect_ConstructorAccessorImpl, Opt_Only_JDK14NewRef) \ + template(reflect_DelegatingClassLoader_klass, sun_reflect_DelegatingClassLoader, Opt) \ + template(reflect_ConstantPool_klass, sun_reflect_ConstantPool, Opt_Only_JDK15) \ + template(reflect_UnsafeStaticFieldAccessorImpl_klass, sun_reflect_UnsafeStaticFieldAccessorImpl, Opt_Only_JDK15) \ \ /* support for dynamic typing; it's OK if these are NULL in earlier JDKs */ \ template(MethodHandle_klass, java_dyn_MethodHandle, Opt) \ @@ -144,16 +144,14 @@ template(WrongMethodTypeException_klass, java_dyn_WrongMethodTypeException, Opt) \ template(Linkage_klass, java_dyn_Linkage, Opt) \ template(CallSite_klass, java_dyn_CallSite, Opt) \ - template(CallSiteImpl_klass, sun_dyn_CallSiteImpl, Opt) \ - template(Dynamic_klass, java_dyn_Dynamic, Opt) \ - /* Note: MethodHandle must be first, and Dynamic last in group */ \ + template(InvokeDynamic_klass, java_dyn_InvokeDynamic, Opt) \ + /* Note: MethodHandle must be first, and InvokeDynamic last in group */ \ \ - template(vector_klass, java_util_Vector, Pre) \ - template(hashtable_klass, java_util_Hashtable, Pre) \ - template(stringBuffer_klass, java_lang_StringBuffer, Pre) \ + template(StringBuffer_klass, java_lang_StringBuffer, Pre) \ + template(StringBuilder_klass, java_lang_StringBuilder, Pre) \ \ /* It's NULL in non-1.4 JDKs. */ \ - template(stackTraceElement_klass, java_lang_StackTraceElement, Opt) \ + template(StackTraceElement_klass, java_lang_StackTraceElement, Opt) \ /* Universe::is_gte_jdk14x_version() is not set up by this point. */ \ /* It's okay if this turns out to be NULL in non-1.4 JDKs. */ \ template(java_nio_Buffer_klass, java_nio_Buffer, Opt) \ @@ -164,14 +162,14 @@ template(sun_jkernel_DownloadManager_klass, sun_jkernel_DownloadManager, Opt_Kernel) \ \ /* Preload boxing klasses */ \ - template(boolean_klass, java_lang_Boolean, Pre) \ - template(char_klass, java_lang_Character, Pre) \ - template(float_klass, java_lang_Float, Pre) \ - template(double_klass, java_lang_Double, Pre) \ - template(byte_klass, java_lang_Byte, Pre) \ - template(short_klass, java_lang_Short, Pre) \ - template(int_klass, java_lang_Integer, Pre) \ - template(long_klass, java_lang_Long, Pre) \ + template(Boolean_klass, java_lang_Boolean, Pre) \ + template(Character_klass, java_lang_Character, Pre) \ + template(Float_klass, java_lang_Float, Pre) \ + template(Double_klass, java_lang_Double, Pre) \ + template(Byte_klass, java_lang_Byte, Pre) \ + template(Short_klass, java_lang_Short, Pre) \ + template(Integer_klass, java_lang_Integer, Pre) \ + template(Long_klass, java_lang_Long, Pre) \ /*end*/ @@ -438,8 +436,8 @@ // Tells whether ClassLoader.checkPackageAccess is present static bool has_checkPackageAccess() { return _has_checkPackageAccess; } - static bool class_klass_loaded() { return WK_KLASS(class_klass) != NULL; } - static bool cloneable_klass_loaded() { return WK_KLASS(cloneable_klass) != NULL; } + static bool Class_klass_loaded() { return WK_KLASS(Class_klass) != NULL; } + static bool Cloneable_klass_loaded() { return WK_KLASS(Cloneable_klass) != NULL; } // Returns default system loader static oop java_system_loader(); @@ -578,6 +576,7 @@ static Handle compute_loader_lock_object(Handle class_loader, TRAPS); static void check_loader_lock_contention(Handle loader_lock, TRAPS); static bool is_parallelCapable(Handle class_loader); + static bool is_parallelDefine(Handle class_loader); static klassOop find_shared_class(symbolHandle class_name);
--- a/src/share/vm/classfile/verifier.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/classfile/verifier.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -143,7 +143,7 @@ bool Verifier::is_eligible_for_verification(instanceKlassHandle klass, bool should_verify_class) { symbolOop name = klass->name(); - klassOop refl_magic_klass = SystemDictionary::reflect_magic_klass(); + klassOop refl_magic_klass = SystemDictionary::reflect_MagicAccessorImpl_klass(); return (should_verify_for(klass->class_loader(), should_verify_class) && // return if the class is a bootstrapping class @@ -1903,17 +1903,8 @@ verify_cp_type(index, cp, types, CHECK_VERIFY(this)); // Get method name and signature - symbolHandle method_name; - symbolHandle method_sig; - if (opcode == Bytecodes::_invokedynamic) { - int name_index = cp->name_ref_index_at(index); - int sig_index = cp->signature_ref_index_at(index); - method_name = symbolHandle(THREAD, cp->symbol_at(name_index)); - method_sig = symbolHandle(THREAD, cp->symbol_at(sig_index)); - } else { - method_name = symbolHandle(THREAD, cp->name_ref_at(index)); - method_sig = symbolHandle(THREAD, cp->signature_ref_at(index)); - } + symbolHandle method_name(THREAD, cp->name_ref_at(index)); + symbolHandle method_sig(THREAD, cp->signature_ref_at(index)); if (!SignatureVerifier::is_valid_method_signature(method_sig)) { class_format_error(
--- a/src/share/vm/classfile/vmSymbols.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/classfile/vmSymbols.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -70,6 +70,7 @@ void vmSymbols::initialize(TRAPS) { assert((int)SID_LIMIT <= (1<<log2_SID_LIMIT), "must fit in this bitfield"); assert((int)SID_LIMIT*5 > (1<<log2_SID_LIMIT), "make the bitfield smaller, please"); + assert(vmIntrinsics::FLAG_LIMIT <= (1 << vmIntrinsics::log2_FLAG_LIMIT), "must fit in this bitfield"); if (!UseSharedSpaces) { const char* string = &vm_symbol_bodies[0]; @@ -271,6 +272,58 @@ return sid; } +static vmIntrinsics::ID wrapper_intrinsic(BasicType type, bool unboxing) { +#define TYPE2(type, unboxing) ((int)(type)*2 + ((unboxing) ? 1 : 0)) + switch (TYPE2(type, unboxing)) { +#define BASIC_TYPE_CASE(type, box, unbox) \ + case TYPE2(type, false): return vmIntrinsics::box; \ + case TYPE2(type, true): return vmIntrinsics::unbox + BASIC_TYPE_CASE(T_BOOLEAN, _Boolean_valueOf, _booleanValue); + BASIC_TYPE_CASE(T_BYTE, _Byte_valueOf, _byteValue); + BASIC_TYPE_CASE(T_CHAR, _Character_valueOf, _charValue); + BASIC_TYPE_CASE(T_SHORT, _Short_valueOf, _shortValue); + BASIC_TYPE_CASE(T_INT, _Integer_valueOf, _intValue); + BASIC_TYPE_CASE(T_LONG, _Long_valueOf, _longValue); + BASIC_TYPE_CASE(T_FLOAT, _Float_valueOf, _floatValue); + BASIC_TYPE_CASE(T_DOUBLE, _Double_valueOf, _doubleValue); +#undef BASIC_TYPE_CASE + } +#undef TYPE2 + return vmIntrinsics::_none; +} + +vmIntrinsics::ID vmIntrinsics::for_boxing(BasicType type) { + return wrapper_intrinsic(type, false); +} +vmIntrinsics::ID vmIntrinsics::for_unboxing(BasicType type) { + return wrapper_intrinsic(type, true); +} + +vmIntrinsics::ID vmIntrinsics::for_raw_conversion(BasicType src, BasicType dest) { +#define SRC_DEST(s,d) (((int)(s) << 4) + (int)(d)) + switch (SRC_DEST(src, dest)) { + case SRC_DEST(T_INT, T_FLOAT): return vmIntrinsics::_intBitsToFloat; + case SRC_DEST(T_FLOAT, T_INT): return vmIntrinsics::_floatToRawIntBits; + + case SRC_DEST(T_LONG, T_DOUBLE): return vmIntrinsics::_longBitsToDouble; + case SRC_DEST(T_DOUBLE, T_LONG): return vmIntrinsics::_doubleToRawLongBits; + } +#undef SRC_DEST + + return vmIntrinsics::_none; +} + +methodOop vmIntrinsics::method_for(vmIntrinsics::ID id) { + if (id == _none) return NULL; + symbolOop cname = vmSymbols::symbol_at(class_for(id)); + symbolOop mname = vmSymbols::symbol_at(name_for(id)); + symbolOop msig = vmSymbols::symbol_at(signature_for(id)); + if (cname == NULL || mname == NULL || msig == NULL) return NULL; + klassOop k = SystemDictionary::find_well_known_klass(cname); + if (k == NULL) return NULL; + return instanceKlass::cast(k)->find_method(mname, msig); +} + #define VM_INTRINSIC_INITIALIZE(id, klass, name, sig, flags) #id "\0" static const char* vm_intrinsic_name_bodies = @@ -303,6 +356,11 @@ const int neg = JVM_ACC_STATIC | JVM_ACC_SYNCHRONIZED; return (flags & (req | neg)) == req; } +inline bool match_F_Y(jshort flags) { + const int req = JVM_ACC_SYNCHRONIZED; + const int neg = JVM_ACC_STATIC; + return (flags & (req | neg)) == req; +} inline bool match_F_RN(jshort flags) { const int req = JVM_ACC_NATIVE; const int neg = JVM_ACC_STATIC | JVM_ACC_SYNCHRONIZED; @@ -325,15 +383,15 @@ } // These are for forming case labels: -#define ID3(x, y, z) (( jint)(z) + \ - ((jint)(y) << vmSymbols::log2_SID_LIMIT) + \ - ((jint)(x) << (2*vmSymbols::log2_SID_LIMIT)) ) +#define ID3(x, y, z) (( jlong)(z) + \ + ((jlong)(y) << vmSymbols::log2_SID_LIMIT) + \ + ((jlong)(x) << (2*vmSymbols::log2_SID_LIMIT)) ) #define SID_ENUM(n) vmSymbols::VM_SYMBOL_ENUM_NAME(n) -vmIntrinsics::ID vmIntrinsics::find_id(vmSymbols::SID holder, - vmSymbols::SID name, - vmSymbols::SID sig, - jshort flags) { +vmIntrinsics::ID vmIntrinsics::find_id_impl(vmSymbols::SID holder, + vmSymbols::SID name, + vmSymbols::SID sig, + jshort flags) { assert((int)vmSymbols::SID_LIMIT <= (1<<vmSymbols::log2_SID_LIMIT), "must fit"); // Let the C compiler build the decision tree. @@ -361,6 +419,7 @@ const char* sname = vmSymbols::name_for(signature_for(id)); const char* fname = ""; switch (flags_for(id)) { + case F_Y: fname = "synchronized "; break; case F_RN: fname = "native "; break; case F_SN: fname = "native static "; break; case F_S: fname = "static "; break; @@ -377,62 +436,50 @@ } -// These are for friendly printouts of intrinsics: +// These are to get information about intrinsics. + +#define ID4(x, y, z, f) ((ID3(x, y, z) << vmIntrinsics::log2_FLAG_LIMIT) | (jlong) (f)) + +static const jlong intrinsic_info_array[vmIntrinsics::ID_LIMIT+1] = { +#define VM_INTRINSIC_INFO(ignore_id, klass, name, sig, fcode) \ + ID4(SID_ENUM(klass), SID_ENUM(name), SID_ENUM(sig), vmIntrinsics::fcode), + + 0, VM_INTRINSICS_DO(VM_INTRINSIC_INFO, + VM_SYMBOL_IGNORE, VM_SYMBOL_IGNORE, VM_SYMBOL_IGNORE, VM_ALIAS_IGNORE) + 0 +#undef VM_INTRINSIC_INFO +}; + +inline jlong intrinsic_info(vmIntrinsics::ID id) { + return intrinsic_info_array[vmIntrinsics::ID_from((int)id)]; +} vmSymbols::SID vmIntrinsics::class_for(vmIntrinsics::ID id) { -#ifndef PRODUCT -#define VM_INTRINSIC_CASE(id, klass, name, sig, fcode) \ - case id: return SID_ENUM(klass); - - switch (id) { - VM_INTRINSICS_DO(VM_INTRINSIC_CASE, - VM_SYMBOL_IGNORE, VM_SYMBOL_IGNORE, VM_SYMBOL_IGNORE, VM_ALIAS_IGNORE); - } -#undef VM_INTRINSIC_CASE -#endif //PRODUCT - return vmSymbols::NO_SID; + jlong info = intrinsic_info(id); + int shift = 2*vmSymbols::log2_SID_LIMIT + log2_FLAG_LIMIT, mask = right_n_bits(vmSymbols::log2_SID_LIMIT); + assert(((ID4(1021,1022,1023,15) >> shift) & mask) == 1021, ""); + return vmSymbols::SID( (info >> shift) & mask ); } vmSymbols::SID vmIntrinsics::name_for(vmIntrinsics::ID id) { -#ifndef PRODUCT -#define VM_INTRINSIC_CASE(id, klass, name, sig, fcode) \ - case id: return SID_ENUM(name); - - switch (id) { - VM_INTRINSICS_DO(VM_INTRINSIC_CASE, - VM_SYMBOL_IGNORE, VM_SYMBOL_IGNORE, VM_SYMBOL_IGNORE, VM_ALIAS_IGNORE); - } -#undef VM_INTRINSIC_CASE -#endif //PRODUCT - return vmSymbols::NO_SID; + jlong info = intrinsic_info(id); + int shift = vmSymbols::log2_SID_LIMIT + log2_FLAG_LIMIT, mask = right_n_bits(vmSymbols::log2_SID_LIMIT); + assert(((ID4(1021,1022,1023,15) >> shift) & mask) == 1022, ""); + return vmSymbols::SID( (info >> shift) & mask ); } vmSymbols::SID vmIntrinsics::signature_for(vmIntrinsics::ID id) { -#ifndef PRODUCT -#define VM_INTRINSIC_CASE(id, klass, name, sig, fcode) \ - case id: return SID_ENUM(sig); - - switch (id) { - VM_INTRINSICS_DO(VM_INTRINSIC_CASE, - VM_SYMBOL_IGNORE, VM_SYMBOL_IGNORE, VM_SYMBOL_IGNORE, VM_ALIAS_IGNORE); - } -#undef VM_INTRINSIC_CASE -#endif //PRODUCT - return vmSymbols::NO_SID; + jlong info = intrinsic_info(id); + int shift = log2_FLAG_LIMIT, mask = right_n_bits(vmSymbols::log2_SID_LIMIT); + assert(((ID4(1021,1022,1023,15) >> shift) & mask) == 1023, ""); + return vmSymbols::SID( (info >> shift) & mask ); } vmIntrinsics::Flags vmIntrinsics::flags_for(vmIntrinsics::ID id) { -#ifndef PRODUCT -#define VM_INTRINSIC_CASE(id, klass, name, sig, fcode) \ - case id: return fcode; - - switch (id) { - VM_INTRINSICS_DO(VM_INTRINSIC_CASE, - VM_SYMBOL_IGNORE, VM_SYMBOL_IGNORE, VM_SYMBOL_IGNORE, VM_ALIAS_IGNORE); - } -#undef VM_INTRINSIC_CASE -#endif //PRODUCT - return F_none; + jlong info = intrinsic_info(id); + int shift = 0, mask = right_n_bits(log2_FLAG_LIMIT); + assert(((ID4(1021,1022,1023,15) >> shift) & mask) == 15, ""); + return Flags( (info >> shift) & mask ); }
--- a/src/share/vm/classfile/vmSymbols.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/classfile/vmSymbols.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -84,6 +84,7 @@ template(java_lang_reflect_Field, "java/lang/reflect/Field") \ template(java_lang_reflect_Array, "java/lang/reflect/Array") \ template(java_lang_StringBuffer, "java/lang/StringBuffer") \ + template(java_lang_StringBuilder, "java/lang/StringBuilder") \ template(java_lang_CharSequence, "java/lang/CharSequence") \ template(java_security_AccessControlContext, "java/security/AccessControlContext") \ template(java_security_ProtectionDomain, "java/security/ProtectionDomain") \ @@ -104,6 +105,7 @@ template(java_lang_AssertionStatusDirectives, "java/lang/AssertionStatusDirectives") \ template(sun_jkernel_DownloadManager, "sun/jkernel/DownloadManager") \ template(getBootClassPathEntryForClass_name, "getBootClassPathEntryForClass") \ + template(setBootClassLoaderHook_name, "setBootClassLoaderHook") \ \ /* class file format tags */ \ template(tag_source_file, "SourceFile") \ @@ -217,7 +219,7 @@ template(base_name, "base") \ \ /* Support for JSR 292 & invokedynamic (JDK 1.7 and above) */ \ - template(java_dyn_Dynamic, "java/dyn/Dynamic") \ + template(java_dyn_InvokeDynamic, "java/dyn/InvokeDynamic") \ template(java_dyn_Linkage, "java/dyn/Linkage") \ template(java_dyn_CallSite, "java/dyn/CallSite") \ template(java_dyn_MethodHandle, "java/dyn/MethodHandle") \ @@ -233,10 +235,9 @@ template(sun_dyn_AdapterMethodHandle, "sun/dyn/AdapterMethodHandle") \ template(sun_dyn_BoundMethodHandle, "sun/dyn/BoundMethodHandle") \ template(sun_dyn_DirectMethodHandle, "sun/dyn/DirectMethodHandle") \ - template(sun_dyn_CallSiteImpl, "sun/dyn/CallSiteImpl") \ template(makeImpl_name, "makeImpl") /*MethodType::makeImpl*/ \ template(makeImpl_signature, "(Ljava/lang/Class;[Ljava/lang/Class;ZZ)Ljava/dyn/MethodType;") \ - template(makeSite_name, "makeSite") /*CallSiteImpl::makeImpl*/ \ + template(makeSite_name, "makeSite") /*CallSite::makeSite*/ \ template(makeSite_signature, "(Ljava/lang/Class;Ljava/lang/String;Ljava/dyn/MethodType;II)Ljava/dyn/CallSite;") \ template(findBootstrapMethod_name, "findBootstrapMethod") \ template(findBootstrapMethod_signature, "(Ljava/lang/Class;Ljava/lang/Class;)Ljava/dyn/MethodHandle;") \ @@ -335,6 +336,7 @@ template(ptypes_name, "ptypes") \ template(form_name, "form") \ template(erasedType_name, "erasedType") \ + template(append_name, "append") \ \ /* non-intrinsic name/signature pairs: */ \ template(register_method_name, "register") \ @@ -345,9 +347,14 @@ \ /* common signatures names */ \ template(void_method_signature, "()V") \ + template(void_boolean_signature, "()Z") \ + template(void_byte_signature, "()B") \ + template(void_char_signature, "()C") \ + template(void_short_signature, "()S") \ template(void_int_signature, "()I") \ template(void_long_signature, "()J") \ - template(void_boolean_signature, "()Z") \ + template(void_float_signature, "()F") \ + template(void_double_signature, "()D") \ template(int_void_signature, "(I)V") \ template(int_int_signature, "(I)I") \ template(int_bool_signature, "(I)Z") \ @@ -416,6 +423,13 @@ template(string_signature, "Ljava/lang/String;") \ template(reference_signature, "Ljava/lang/ref/Reference;") \ template(concurrenthashmap_signature, "Ljava/util/concurrent/ConcurrentHashMap;") \ + template(String_StringBuilder_signature, "(Ljava/lang/String;)Ljava/lang/StringBuilder;") \ + template(int_StringBuilder_signature, "(I)Ljava/lang/StringBuilder;") \ + template(char_StringBuilder_signature, "(C)Ljava/lang/StringBuilder;") \ + template(String_StringBuffer_signature, "(Ljava/lang/String;)Ljava/lang/StringBuffer;") \ + template(int_StringBuffer_signature, "(I)Ljava/lang/StringBuffer;") \ + template(char_StringBuffer_signature, "(C)Ljava/lang/StringBuffer;") \ + template(int_String_signature, "(I)Ljava/lang/String;") \ /* signature symbols needed by intrinsics */ \ VM_INTRINSICS_DO(VM_INTRINSIC_IGNORE, VM_SYMBOL_IGNORE, VM_SYMBOL_IGNORE, template, VM_ALIAS_IGNORE) \ \ @@ -815,12 +829,76 @@ /*the compiler does have special inlining code for these; bytecode inline is just fine */ \ \ do_intrinsic(_fillInStackTrace, java_lang_Throwable, fillInStackTrace_name, void_throwable_signature, F_RNY) \ - \ - do_intrinsic(_Object_init, java_lang_Object, object_initializer_name, void_method_signature, F_R) \ - /* (symbol object_initializer_name defined above) */ \ - \ + \ + do_intrinsic(_StringBuilder_void, java_lang_StringBuilder, object_initializer_name, void_method_signature, F_R) \ + do_intrinsic(_StringBuilder_int, java_lang_StringBuilder, object_initializer_name, int_void_signature, F_R) \ + do_intrinsic(_StringBuilder_String, java_lang_StringBuilder, object_initializer_name, string_void_signature, F_R) \ + \ + do_intrinsic(_StringBuilder_append_char, java_lang_StringBuilder, append_name, char_StringBuilder_signature, F_R) \ + do_intrinsic(_StringBuilder_append_int, java_lang_StringBuilder, append_name, int_StringBuilder_signature, F_R) \ + do_intrinsic(_StringBuilder_append_String, java_lang_StringBuilder, append_name, String_StringBuilder_signature, F_R) \ + \ + do_intrinsic(_StringBuilder_toString, java_lang_StringBuilder, toString_name, void_string_signature, F_R) \ + \ + do_intrinsic(_StringBuffer_void, java_lang_StringBuffer, object_initializer_name, void_method_signature, F_R) \ + do_intrinsic(_StringBuffer_int, java_lang_StringBuffer, object_initializer_name, int_void_signature, F_R) \ + do_intrinsic(_StringBuffer_String, java_lang_StringBuffer, object_initializer_name, string_void_signature, F_R) \ + \ + do_intrinsic(_StringBuffer_append_char, java_lang_StringBuffer, append_name, char_StringBuffer_signature, F_Y) \ + do_intrinsic(_StringBuffer_append_int, java_lang_StringBuffer, append_name, int_StringBuffer_signature, F_Y) \ + do_intrinsic(_StringBuffer_append_String, java_lang_StringBuffer, append_name, String_StringBuffer_signature, F_Y) \ + \ + do_intrinsic(_StringBuffer_toString, java_lang_StringBuffer, toString_name, void_string_signature, F_Y) \ + \ + do_intrinsic(_Integer_toString, java_lang_Integer, toString_name, int_String_signature, F_S) \ + \ + do_intrinsic(_String_String, java_lang_String, object_initializer_name, string_void_signature, F_R) \ + \ + do_intrinsic(_Object_init, java_lang_Object, object_initializer_name, void_method_signature, F_R) \ + /* (symbol object_initializer_name defined above) */ \ + \ do_intrinsic(_invoke, java_lang_reflect_Method, invoke_name, object_array_object_object_signature, F_R) \ /* (symbols invoke_name and invoke_signature defined above) */ \ + do_intrinsic(_checkSpreadArgument, sun_dyn_MethodHandleImpl, checkSpreadArgument_name, checkSpreadArgument_signature, F_S) \ + do_name( checkSpreadArgument_name, "checkSpreadArgument") \ + do_name( checkSpreadArgument_signature, "(Ljava/lang/Object;I)V") \ + \ + /* unboxing methods: */ \ + do_intrinsic(_booleanValue, java_lang_Boolean, booleanValue_name, void_boolean_signature, F_R) \ + do_name( booleanValue_name, "booleanValue") \ + do_intrinsic(_byteValue, java_lang_Byte, byteValue_name, void_byte_signature, F_R) \ + do_name( byteValue_name, "byteValue") \ + do_intrinsic(_charValue, java_lang_Character, charValue_name, void_char_signature, F_R) \ + do_name( charValue_name, "charValue") \ + do_intrinsic(_shortValue, java_lang_Short, shortValue_name, void_short_signature, F_R) \ + do_name( shortValue_name, "shortValue") \ + do_intrinsic(_intValue, java_lang_Integer, intValue_name, void_int_signature, F_R) \ + do_name( intValue_name, "intValue") \ + do_intrinsic(_longValue, java_lang_Long, longValue_name, void_long_signature, F_R) \ + do_name( longValue_name, "longValue") \ + do_intrinsic(_floatValue, java_lang_Float, floatValue_name, void_float_signature, F_R) \ + do_name( floatValue_name, "floatValue") \ + do_intrinsic(_doubleValue, java_lang_Double, doubleValue_name, void_double_signature, F_R) \ + do_name( doubleValue_name, "doubleValue") \ + \ + /* boxing methods: */ \ + do_name( valueOf_name, "valueOf") \ + do_intrinsic(_Boolean_valueOf, java_lang_Boolean, valueOf_name, Boolean_valueOf_signature, F_S) \ + do_name( Boolean_valueOf_signature, "(Z)Ljava/lang/Boolean;") \ + do_intrinsic(_Byte_valueOf, java_lang_Byte, valueOf_name, Byte_valueOf_signature, F_S) \ + do_name( Byte_valueOf_signature, "(B)Ljava/lang/Byte;") \ + do_intrinsic(_Character_valueOf, java_lang_Character, valueOf_name, Character_valueOf_signature, F_S) \ + do_name( Character_valueOf_signature, "(C)Ljava/lang/Character;") \ + do_intrinsic(_Short_valueOf, java_lang_Short, valueOf_name, Short_valueOf_signature, F_S) \ + do_name( Short_valueOf_signature, "(S)Ljava/lang/Short;") \ + do_intrinsic(_Integer_valueOf, java_lang_Integer, valueOf_name, Integer_valueOf_signature, F_S) \ + do_name( Integer_valueOf_signature, "(I)Ljava/lang/Integer;") \ + do_intrinsic(_Long_valueOf, java_lang_Long, valueOf_name, Long_valueOf_signature, F_S) \ + do_name( Long_valueOf_signature, "(J)Ljava/lang/Long;") \ + do_intrinsic(_Float_valueOf, java_lang_Float, valueOf_name, Float_valueOf_signature, F_S) \ + do_name( Float_valueOf_signature, "(F)Ljava/lang/Float;") \ + do_intrinsic(_Double_valueOf, java_lang_Double, valueOf_name, Double_valueOf_signature, F_S) \ + do_name( Double_valueOf_signature, "(D)Ljava/lang/Double;") \ \ /*end*/ @@ -946,11 +1024,17 @@ enum Flags { // AccessFlags syndromes relevant to intrinsics. F_none = 0, - F_R, // !static !synchronized (R="regular") - F_S, // static !synchronized - F_RN, // !static native !synchronized - F_SN, // static native !synchronized - F_RNY // !static native synchronized + F_R, // !static ?native !synchronized (R="regular") + F_S, // static ?native !synchronized + F_Y, // !static ?native synchronized + F_RN, // !static native !synchronized + F_SN, // static native !synchronized + F_RNY, // !static native synchronized + + FLAG_LIMIT + }; + enum { + log2_FLAG_LIMIT = 4 // checked by an assert at start-up }; public: @@ -962,15 +1046,32 @@ static const char* name_at(ID id); +private: + static ID find_id_impl(vmSymbols::SID holder, + vmSymbols::SID name, + vmSymbols::SID sig, + jshort flags); + +public: // Given a method's class, name, signature, and access flags, report its ID. static ID find_id(vmSymbols::SID holder, vmSymbols::SID name, vmSymbols::SID sig, - jshort flags); + jshort flags) { + ID id = find_id_impl(holder, name, sig, flags); +#ifdef ASSERT + // ID _none does not hold the following asserts. + if (id == _none) return id; +#endif + assert( class_for(id) == holder, "correct id"); + assert( name_for(id) == name, "correct id"); + assert(signature_for(id) == sig, "correct id"); + return id; + } static void verify_method(ID actual_id, methodOop m) PRODUCT_RETURN; - // No need for these in the product: + // Find out the symbols behind an intrinsic: static vmSymbols::SID class_for(ID id); static vmSymbols::SID name_for(ID id); static vmSymbols::SID signature_for(ID id); @@ -980,4 +1081,11 @@ // Access to intrinsic methods: static methodOop method_for(ID id); + + // Wrapper object methods: + static ID for_boxing(BasicType type); + static ID for_unboxing(BasicType type); + + // Raw conversion: + static ID for_raw_conversion(BasicType src, BasicType dest); };
--- a/src/share/vm/code/codeBlob.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/code/codeBlob.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1998-2007 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1998-2009 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -102,6 +102,9 @@ virtual bool is_compiled_by_c2() const { return false; } virtual bool is_compiled_by_c1() const { return false; } + // Casting + nmethod* as_nmethod_or_null() { return is_nmethod() ? (nmethod*) this : NULL; } + // Boundaries address header_begin() const { return (address) this; } address header_end() const { return ((address) this) + _header_size; }; @@ -201,7 +204,8 @@ virtual void print_value_on(outputStream* st) const PRODUCT_RETURN; // Print the comment associated with offset on stream, if there is one - void print_block_comment(outputStream* stream, intptr_t offset) { + virtual void print_block_comment(outputStream* stream, address block_begin) { + intptr_t offset = (intptr_t)(block_begin - instructions_begin()); _comments.print_block_comment(stream, offset); }
--- a/src/share/vm/code/codeCache.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/code/codeCache.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -96,6 +96,7 @@ int CodeCache::_number_of_nmethods_with_dependencies = 0; bool CodeCache::_needs_cache_clean = false; nmethod* CodeCache::_scavenge_root_nmethods = NULL; +nmethod* CodeCache::_saved_nmethods = NULL; CodeBlob* CodeCache::first() { @@ -395,6 +396,85 @@ } #endif //PRODUCT + +nmethod* CodeCache::find_and_remove_saved_code(methodOop m) { + MutexLockerEx mu(CodeCache_lock, Mutex::_no_safepoint_check_flag); + nmethod* saved = _saved_nmethods; + nmethod* prev = NULL; + while (saved != NULL) { + if (saved->is_in_use() && saved->method() == m) { + if (prev != NULL) { + prev->set_saved_nmethod_link(saved->saved_nmethod_link()); + } else { + _saved_nmethods = saved->saved_nmethod_link(); + } + assert(saved->is_speculatively_disconnected(), "shouldn't call for other nmethods"); + saved->set_speculatively_disconnected(false); + saved->set_saved_nmethod_link(NULL); + if (PrintMethodFlushing) { + saved->print_on(tty, " ### nmethod is reconnected"); + } + if (LogCompilation && (xtty != NULL)) { + ttyLocker ttyl; + xtty->begin_elem("nmethod_reconnected compile_id='%3d'", saved->compile_id()); + xtty->method(methodOop(m)); + xtty->stamp(); + xtty->end_elem(); + } + return saved; + } + prev = saved; + saved = saved->saved_nmethod_link(); + } + return NULL; +} + +void CodeCache::remove_saved_code(nmethod* nm) { + MutexLockerEx mu(CodeCache_lock, Mutex::_no_safepoint_check_flag); + assert(nm->is_speculatively_disconnected(), "shouldn't call for other nmethods"); + nmethod* saved = _saved_nmethods; + nmethod* prev = NULL; + while (saved != NULL) { + if (saved == nm) { + if (prev != NULL) { + prev->set_saved_nmethod_link(saved->saved_nmethod_link()); + } else { + _saved_nmethods = saved->saved_nmethod_link(); + } + if (LogCompilation && (xtty != NULL)) { + ttyLocker ttyl; + xtty->begin_elem("nmethod_removed compile_id='%3d'", nm->compile_id()); + xtty->stamp(); + xtty->end_elem(); + } + return; + } + prev = saved; + saved = saved->saved_nmethod_link(); + } + ShouldNotReachHere(); +} + +void CodeCache::speculatively_disconnect(nmethod* nm) { + assert_locked_or_safepoint(CodeCache_lock); + assert(nm->is_in_use() && !nm->is_speculatively_disconnected(), "should only disconnect live nmethods"); + nm->set_saved_nmethod_link(_saved_nmethods); + _saved_nmethods = nm; + if (PrintMethodFlushing) { + nm->print_on(tty, " ### nmethod is speculatively disconnected"); + } + if (LogCompilation && (xtty != NULL)) { + ttyLocker ttyl; + xtty->begin_elem("nmethod_disconnected compile_id='%3d'", nm->compile_id()); + xtty->method(methodOop(nm->method())); + xtty->stamp(); + xtty->end_elem(); + } + nm->method()->clear_code(); + nm->set_speculatively_disconnected(true); +} + + void CodeCache::gc_prologue() { assert(!nmethod::oops_do_marking_is_active(), "oops_do_marking_epilogue must be called"); }
--- a/src/share/vm/code/codeCache.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/code/codeCache.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -46,6 +46,7 @@ static int _number_of_nmethods_with_dependencies; static bool _needs_cache_clean; static nmethod* _scavenge_root_nmethods; // linked via nm->scavenge_root_link() + static nmethod* _saved_nmethods; // linked via nm->saved_nmethod_look() static void verify_if_often() PRODUCT_RETURN; @@ -141,11 +142,16 @@ static size_t capacity() { return _heap->capacity(); } static size_t max_capacity() { return _heap->max_capacity(); } static size_t unallocated_capacity() { return _heap->unallocated_capacity(); } + static bool needs_flushing() { return unallocated_capacity() < CodeCacheFlushingMinimumFreeSpace; } static bool needs_cache_clean() { return _needs_cache_clean; } static void set_needs_cache_clean(bool v) { _needs_cache_clean = v; } static void clear_inline_caches(); // clear all inline caches + static nmethod* find_and_remove_saved_code(methodOop m); + static void remove_saved_code(nmethod* nm); + static void speculatively_disconnect(nmethod* nm); + // Deoptimization static int mark_for_deoptimization(DepChange& changes); #ifdef HOTSWAP
--- a/src/share/vm/code/debugInfoRec.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/code/debugInfoRec.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -281,6 +281,8 @@ ciMethod* method, int bci, bool reexecute, + bool is_method_handle_invoke, + bool return_oop, DebugToken* locals, DebugToken* expressions, DebugToken* monitors) { @@ -292,8 +294,10 @@ int stream_offset = stream()->position(); last_pd->set_scope_decode_offset(stream_offset); - // Record reexecute bit into pcDesc + // Record flags into pcDesc. last_pd->set_should_reexecute(reexecute); + last_pd->set_is_method_handle_invoke(is_method_handle_invoke); + last_pd->set_return_oop(return_oop); // serialize sender stream offest stream()->write_int(sender_stream_offset);
--- a/src/share/vm/code/debugInfoRec.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/code/debugInfoRec.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -88,6 +88,8 @@ ciMethod* method, int bci, bool reexecute, + bool is_method_handle_invoke = false, + bool return_oop = false, DebugToken* locals = NULL, DebugToken* expressions = NULL, DebugToken* monitors = NULL);
--- a/src/share/vm/code/dependencies.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/code/dependencies.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -843,13 +843,15 @@ if (occasional_print || final_stats) { // Every now and then dump a little info about dependency searching. if (xtty != NULL) { - xtty->elem("deps_find_witness calls='%d' steps='%d' recursions='%d' singles='%d'", + ttyLocker ttyl; + xtty->elem("deps_find_witness calls='%d' steps='%d' recursions='%d' singles='%d'", deps_find_witness_calls, deps_find_witness_steps, deps_find_witness_recursions, deps_find_witness_singles); } if (final_stats || (TraceDependencies && WizardMode)) { + ttyLocker ttyl; tty->print_cr("Dependency check (find_witness) " "calls=%d, steps=%d (avg=%.1f), recursions=%d, singles=%d", deps_find_witness_calls, @@ -1528,19 +1530,23 @@ int nsup = 0, nint = 0; for (ContextStream str(*this); str.next(); ) { klassOop k = str.klass(); - switch (str._change_type) { + switch (str.change_type()) { case Change_new_type: tty->print_cr(" dependee = %s", instanceKlass::cast(k)->external_name()); break; case Change_new_sub: - if (!WizardMode) - ++nsup; - else tty->print_cr(" context super = %s", instanceKlass::cast(k)->external_name()); + if (!WizardMode) { + ++nsup; + } else { + tty->print_cr(" context super = %s", instanceKlass::cast(k)->external_name()); + } break; case Change_new_impl: - if (!WizardMode) - ++nint; - else tty->print_cr(" context interface = %s", instanceKlass::cast(k)->external_name()); + if (!WizardMode) { + ++nint; + } else { + tty->print_cr(" context interface = %s", instanceKlass::cast(k)->external_name()); + } break; } }
--- a/src/share/vm/code/dependencies.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/code/dependencies.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -470,7 +470,7 @@ // super types can be context types for a relevant dependency, which the // new type could invalidate. class DepChange : public StackObj { - private: + public: enum ChangeType { NO_CHANGE = 0, // an uninvolved klass Change_new_type, // a newly loaded type @@ -480,6 +480,7 @@ Start_Klass = CHANGE_LIMIT // internal indicator for ContextStream }; + private: // each change set is rooted in exactly one new type (at present): KlassHandle _new_type; @@ -510,15 +511,15 @@ // } class ContextStream : public StackObj { private: - DepChange& _changes; + DepChange& _changes; friend class DepChange; // iteration variables: - ChangeType _change_type; - klassOop _klass; - objArrayOop _ti_base; // i.e., transitive_interfaces - int _ti_index; - int _ti_limit; + ChangeType _change_type; + klassOop _klass; + objArrayOop _ti_base; // i.e., transitive_interfaces + int _ti_index; + int _ti_limit; // start at the beginning: void start() { @@ -530,11 +531,11 @@ _ti_limit = 0; } + public: ContextStream(DepChange& changes) : _changes(changes) { start(); } - public: ContextStream(DepChange& changes, No_Safepoint_Verifier& nsv) : _changes(changes) // the nsv argument makes it safe to hold oops like _klass @@ -542,6 +543,7 @@ bool next(); + ChangeType change_type() { return _change_type; } klassOop klass() { return _klass; } }; friend class DepChange::ContextStream;
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/share/vm/code/jvmticmlr.h Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,115 @@ +/* + * Copyright 2010 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. Sun designates this + * particular file as subject to the "Classpath" exception as provided + * by Sun in the LICENSE file that accompanied this code. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + */ + +/* + * This header file defines the data structures sent by the VM + * through the JVMTI CompiledMethodLoad callback function via the + * "void * compile_info" parameter. The memory pointed to by the + * compile_info parameter may not be referenced after returning from + * the CompiledMethodLoad callback. These are VM implementation + * specific data structures that may evolve in future releases. A + * JVMTI agent should interpret a non-NULL compile_info as a pointer + * to a region of memory containing a list of records. In a typical + * usage scenario, a JVMTI agent would cast each record to a + * jvmtiCompiledMethodLoadRecordHeader, a struct that represents + * arbitrary information. This struct contains a kind field to indicate + * the kind of information being passed, and a pointer to the next + * record. If the kind field indicates inlining information, then the + * agent would cast the record to a jvmtiCompiledMethodLoadInlineRecord. + * This record contains an array of PCStackInfo structs, which indicate + * for every pc address what are the methods on the invocation stack. + * The "methods" and "bcis" fields in each PCStackInfo struct specify a + * 1-1 mapping between these inlined methods and their bytecode indices. + * This can be used to derive the proper source lines of the inlined + * methods. + */ + +#ifndef _JVMTI_CMLR_H_ +#define _JVMTI_CMLR_H_ + +enum { + JVMTI_CMLR_MAJOR_VERSION_1 = 0x00000001, + JVMTI_CMLR_MINOR_VERSION_0 = 0x00000000, + + JVMTI_CMLR_MAJOR_VERSION = 0x00000001, + JVMTI_CMLR_MINOR_VERSION = 0x00000000 + + /* + * This comment is for the "JDK import from HotSpot" sanity check: + * version: 1.0.0 + */ +}; + +typedef enum { + JVMTI_CMLR_DUMMY = 1, + JVMTI_CMLR_INLINE_INFO = 2 +} jvmtiCMLRKind; + +/* + * Record that represents arbitrary information passed through JVMTI + * CompiledMethodLoadEvent void pointer. + */ +typedef struct _jvmtiCompiledMethodLoadRecordHeader { + jvmtiCMLRKind kind; /* id for the kind of info passed in the record */ + jint majorinfoversion; /* major and minor info version values. Init'ed */ + jint minorinfoversion; /* to current version value in jvmtiExport.cpp. */ + + struct _jvmtiCompiledMethodLoadRecordHeader* next; +} jvmtiCompiledMethodLoadRecordHeader; + +/* + * Record that gives information about the methods on the compile-time + * stack at a specific pc address of a compiled method. Each element in + * the methods array maps to same element in the bcis array. + */ +typedef struct _PCStackInfo { + void* pc; /* the pc address for this compiled method */ + jint numstackframes; /* number of methods on the stack */ + jmethodID* methods; /* array of numstackframes method ids */ + jint* bcis; /* array of numstackframes bytecode indices */ +} PCStackInfo; + +/* + * Record that contains inlining information for each pc address of + * an nmethod. + */ +typedef struct _jvmtiCompiledMethodLoadInlineRecord { + jvmtiCompiledMethodLoadRecordHeader header; /* common header for casting */ + jint numpcs; /* number of pc descriptors in this nmethod */ + PCStackInfo* pcinfo; /* array of numpcs pc descriptors */ +} jvmtiCompiledMethodLoadInlineRecord; + +/* + * Dummy record used to test that we can pass records with different + * information through the void pointer provided that they can be cast + * to a jvmtiCompiledMethodLoadRecordHeader. + */ + +typedef struct _jvmtiCompiledMethodLoadDummyRecord { + jvmtiCompiledMethodLoadRecordHeader header; /* common header for casting */ + char message[50]; +} jvmtiCompiledMethodLoadDummyRecord; + +#endif
--- a/src/share/vm/code/nmethod.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/code/nmethod.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -56,13 +56,13 @@ #endif bool nmethod::is_compiled_by_c1() const { + if (compiler() == NULL || method() == NULL) return false; // can happen during debug printing if (is_native_method()) return false; - assert(compiler() != NULL, "must be"); return compiler()->is_c1(); } bool nmethod::is_compiled_by_c2() const { + if (compiler() == NULL || method() == NULL) return false; // can happen during debug printing if (is_native_method()) return false; - assert(compiler() != NULL, "must be"); return compiler()->is_c2(); } @@ -414,9 +414,8 @@ } const char* nmethod::compile_kind() const { - if (method() == NULL) return "unloaded"; - if (is_native_method()) return "c2n"; if (is_osr_method()) return "osr"; + if (method() != NULL && is_native_method()) return "c2n"; return NULL; } @@ -588,11 +587,13 @@ _osr_link = NULL; _scavenge_root_link = NULL; _scavenge_root_state = 0; + _saved_nmethod_link = NULL; _compiler = NULL; // We have no exception handler or deopt handler make the // values something that will never match a pc like the nmethod vtable entry _exception_offset = 0; _deoptimize_offset = 0; + _deoptimize_mh_offset = 0; _orig_pc_offset = 0; #ifdef HAVE_DTRACE_H _trap_offset = 0; @@ -683,6 +684,7 @@ // values something that will never match a pc like the nmethod vtable entry _exception_offset = 0; _deoptimize_offset = 0; + _deoptimize_mh_offset = 0; _trap_offset = offsets->value(CodeOffsets::Dtrace_trap); _orig_pc_offset = 0; _stub_offset = data_offset(); @@ -795,6 +797,7 @@ // Exception handler and deopt handler are in the stub section _exception_offset = _stub_offset + offsets->value(CodeOffsets::Exceptions); _deoptimize_offset = _stub_offset + offsets->value(CodeOffsets::Deopt); + _deoptimize_mh_offset = _stub_offset + offsets->value(CodeOffsets::DeoptMH); _consts_offset = instructions_offset() + code_buffer->total_offset_of(code_buffer->consts()->start()); _scopes_data_offset = data_offset(); _scopes_pcs_offset = _scopes_data_offset + round_to(debug_info->data_size (), oopSize); @@ -985,7 +988,8 @@ PcDesc* pd = pc_desc_at(pc); guarantee(pd != NULL, "scope must be present"); return new ScopeDesc(this, pd->scope_decode_offset(), - pd->obj_decode_offset(), pd->should_reexecute()); + pd->obj_decode_offset(), pd->should_reexecute(), + pd->return_oop()); } @@ -1034,7 +1038,7 @@ if( cb != NULL && cb->is_nmethod() ) { nmethod* nm = (nmethod*)cb; // Clean inline caches pointing to both zombie and not_entrant methods - if (!nm->is_in_use()) ic->set_to_clean(); + if (!nm->is_in_use() || (nm->method()->code() != nm)) ic->set_to_clean(); } break; } @@ -1044,7 +1048,7 @@ if( cb != NULL && cb->is_nmethod() ) { nmethod* nm = (nmethod*)cb; // Clean inline caches pointing to both zombie and not_entrant methods - if (!nm->is_in_use()) csc->set_to_clean(); + if (!nm->is_in_use() || (nm->method()->code() != nm)) csc->set_to_clean(); } break; } @@ -1114,7 +1118,6 @@ if (_method->code() == this) { _method->clear_code(); // Break a cycle } - inc_decompile_count(); // Last chance to make a mark on the MDO _method = NULL; // Clear the method of this dead nmethod } // Make the class unloaded - i.e., change state and notify sweeper @@ -1127,6 +1130,9 @@ } flags.state = unloaded; + // Log the unloading. + log_state_change(); + // The methodOop is gone at this point assert(_method == NULL, "Tautology"); @@ -1137,8 +1143,6 @@ void nmethod::invalidate_osr_method() { assert(_entry_bci != InvocationEntryBci, "wrong kind of nmethod"); - if (_entry_bci != InvalidOSREntryBci) - inc_decompile_count(); // Remove from list of active nmethods if (method() != NULL) instanceKlass::cast(method()->method_holder())->remove_osr_nmethod(this); @@ -1146,59 +1150,65 @@ _entry_bci = InvalidOSREntryBci; } -void nmethod::log_state_change(int state) const { +void nmethod::log_state_change() const { if (LogCompilation) { if (xtty != NULL) { ttyLocker ttyl; // keep the following output all in one block - xtty->begin_elem("make_not_entrant %sthread='" UINTX_FORMAT "'", - (state == zombie ? "zombie='1' " : ""), - os::current_thread_id()); + if (flags.state == unloaded) { + xtty->begin_elem("make_unloaded thread='" UINTX_FORMAT "'", + os::current_thread_id()); + } else { + xtty->begin_elem("make_not_entrant thread='" UINTX_FORMAT "'%s", + os::current_thread_id(), + (flags.state == zombie ? " zombie='1'" : "")); + } log_identity(xtty); xtty->stamp(); xtty->end_elem(); } } - if (PrintCompilation) { - print_on(tty, state == zombie ? "made zombie " : "made not entrant "); + if (PrintCompilation && flags.state != unloaded) { + print_on(tty, flags.state == zombie ? "made zombie " : "made not entrant "); tty->cr(); } } // Common functionality for both make_not_entrant and make_zombie -void nmethod::make_not_entrant_or_zombie(int state) { +bool nmethod::make_not_entrant_or_zombie(unsigned int state) { assert(state == zombie || state == not_entrant, "must be zombie or not_entrant"); - // Code for an on-stack-replacement nmethod is removed when a class gets unloaded. - // They never become zombie/non-entrant, so the nmethod sweeper will never remove - // them. Instead the entry_bci is set to InvalidOSREntryBci, so the osr nmethod - // will never be used anymore. That the nmethods only gets removed when class unloading - // happens, make life much simpler, since the nmethods are not just going to disappear - // out of the blue. - if (is_osr_method()) { - if (osr_entry_bci() != InvalidOSREntryBci) { - // only log this once - log_state_change(state); - } - invalidate_osr_method(); - return; - } - - // If the method is already zombie or set to the state we want, nothing to do - if (is_zombie() || (state == not_entrant && is_not_entrant())) { - return; - } - - log_state_change(state); + bool was_alive = false; // Make sure the nmethod is not flushed in case of a safepoint in code below. nmethodLocker nml(this); { + // If the method is already zombie there is nothing to do + if (is_zombie()) { + return false; + } + + // invalidate osr nmethod before acquiring the patching lock since + // they both acquire leaf locks and we don't want a deadlock. + // This logic is equivalent to the logic below for patching the + // verified entry point of regular methods. + if (is_osr_method()) { + // this effectively makes the osr nmethod not entrant + invalidate_osr_method(); + } + // Enter critical section. Does not block for safepoint. MutexLockerEx pl(Patching_lock, Mutex::_no_safepoint_check_flag); + + if (flags.state == state) { + // another thread already performed this transition so nothing + // to do, but return false to indicate this. + return false; + } + // The caller can be calling the method statically or through an inline // cache call. - if (!is_not_entrant()) { + if (!is_osr_method() && !is_not_entrant()) { NativeJump::patch_verified_entry(entry_point(), verified_entry_point(), SharedRuntime::get_handle_wrong_method_stub()); assert (NativeJump::instruction_size == nmethod::_zombie_instruction_size, ""); @@ -1215,8 +1225,14 @@ assert(state == not_entrant, "other cases may need to be handled differently"); } + was_alive = is_in_use(); // Read state under lock + // Change state flags.state = state; + + // Log the transition once + log_state_change(); + } // leave critical region under Patching_lock if (state == not_entrant) { @@ -1237,9 +1253,11 @@ mark_as_seen_on_stack(); } - // It's a true state change, so mark the method as decompiled. - inc_decompile_count(); - + if (was_alive) { + // It's a true state change, so mark the method as decompiled. + // Do it only for transition from alive. + inc_decompile_count(); + } // zombie only - if a JVMTI agent has enabled the CompiledMethodUnload event // and it hasn't already been reported for this nmethod then report it now. @@ -1268,7 +1286,7 @@ // Check whether method got unloaded at a safepoint before this, // if so we can skip the flushing steps below - if (method() == NULL) return; + if (method() == NULL) return true; // Remove nmethod from method. // We need to check if both the _code and _from_compiled_code_entry_point @@ -1282,6 +1300,8 @@ HandleMark hm; method()->clear_code(); } + + return true; } @@ -1303,7 +1323,8 @@ // completely deallocate this method EventMark m("flushing nmethod " INTPTR_FORMAT " %s", this, ""); if (PrintMethodFlushing) { - tty->print_cr("*flushing nmethod " INTPTR_FORMAT ". Live blobs: %d", this, CodeCache::nof_blobs()); + tty->print_cr("*flushing nmethod %3d/" INTPTR_FORMAT ". Live blobs:" UINT32_FORMAT "/Free CodeCache:" SIZE_FORMAT "Kb", + _compile_id, this, CodeCache::nof_blobs(), CodeCache::unallocated_capacity()/1024); } // We need to deallocate any ExceptionCache data. @@ -1321,6 +1342,10 @@ CodeCache::drop_scavenge_root_nmethod(this); } + if (is_speculatively_disconnected()) { + CodeCache::remove_saved_code(this); + } + ((CodeBlob*)(this))->flush(); CodeCache::free(this); @@ -1715,9 +1740,9 @@ if (!method()->is_native()) { SimpleScopeDesc ssd(this, fr.pc()); Bytecode_invoke* call = Bytecode_invoke_at(ssd.method(), ssd.bci()); - bool is_static = call->is_invokestatic(); + bool has_receiver = call->has_receiver(); symbolOop signature = call->signature(); - fr.oops_compiled_arguments_do(signature, is_static, reg_map, f); + fr.oops_compiled_arguments_do(signature, has_receiver, reg_map, f); } } @@ -1754,6 +1779,14 @@ "must end with a sentinel"); #endif //ASSERT + // Search for MethodHandle invokes and tag the nmethod. + for (int i = 0; i < count; i++) { + if (pcs[i].is_method_handle_invoke()) { + set_has_method_handle_invokes(true); + break; + } + } + int size = count * sizeof(PcDesc); assert(scopes_pcs_size() >= size, "oob"); memcpy(scopes_pcs_begin(), pcs, size); @@ -1978,7 +2011,10 @@ print_pcs(); } #endif - guarantee(cont_offset != 0, "unhandled implicit exception in compiled code"); + if (cont_offset == 0) { + // Let the normal error handling report the exception + return NULL; + } return instructions_begin() + cont_offset; } @@ -2014,9 +2050,33 @@ guarantee(nm->_lock_count >= 0, "unmatched nmethod lock/unlock"); } -bool nmethod::is_deopt_pc(address pc) { - bool ret = pc == deopt_handler_begin(); - return ret; + +// ----------------------------------------------------------------------------- +// nmethod::get_deopt_original_pc +// +// Return the original PC for the given PC if: +// (a) the given PC belongs to a nmethod and +// (b) it is a deopt PC +address nmethod::get_deopt_original_pc(const frame* fr) { + if (fr->cb() == NULL) return NULL; + + nmethod* nm = fr->cb()->as_nmethod_or_null(); + if (nm != NULL && nm->is_deopt_pc(fr->pc())) + return nm->get_original_pc(fr); + + return NULL; +} + + +// ----------------------------------------------------------------------------- +// MethodHandle + +bool nmethod::is_method_handle_return(address return_pc) { + if (!has_method_handle_invokes()) return false; + PcDesc* pd = pc_desc_at(return_pc); + if (pd == NULL) + return false; + return pd->is_method_handle_invoke(); } @@ -2100,7 +2160,8 @@ PcDesc* pd = pc_desc_at(ic->end_of_call()); assert(pd != NULL, "PcDesc must exist"); for (ScopeDesc* sd = new ScopeDesc(this, pd->scope_decode_offset(), - pd->obj_decode_offset(), pd->should_reexecute()); + pd->obj_decode_offset(), pd->should_reexecute(), + pd->return_oop()); !sd->is_top(); sd = sd->sender()) { sd->verify(); } @@ -2365,11 +2426,115 @@ PcDesc* p = pc_desc_near(begin+1); if (p != NULL && p->real_pc(this) <= end) { return new ScopeDesc(this, p->scope_decode_offset(), - p->obj_decode_offset(), p->should_reexecute()); + p->obj_decode_offset(), p->should_reexecute(), + p->return_oop()); } return NULL; } +void nmethod::print_nmethod_labels(outputStream* stream, address block_begin) { + if (block_begin == entry_point()) stream->print_cr("[Entry Point]"); + if (block_begin == verified_entry_point()) stream->print_cr("[Verified Entry Point]"); + if (block_begin == exception_begin()) stream->print_cr("[Exception Handler]"); + if (block_begin == stub_begin()) stream->print_cr("[Stub Code]"); + if (block_begin == deopt_handler_begin()) stream->print_cr("[Deopt Handler Code]"); + if (block_begin == deopt_mh_handler_begin()) stream->print_cr("[Deopt MH Handler Code]"); + if (block_begin == consts_begin()) stream->print_cr("[Constants]"); + if (block_begin == entry_point()) { + methodHandle m = method(); + if (m.not_null()) { + stream->print(" # "); + m->print_value_on(stream); + stream->cr(); + } + if (m.not_null() && !is_osr_method()) { + ResourceMark rm; + int sizeargs = m->size_of_parameters(); + BasicType* sig_bt = NEW_RESOURCE_ARRAY(BasicType, sizeargs); + VMRegPair* regs = NEW_RESOURCE_ARRAY(VMRegPair, sizeargs); + { + int sig_index = 0; + if (!m->is_static()) + sig_bt[sig_index++] = T_OBJECT; // 'this' + for (SignatureStream ss(m->signature()); !ss.at_return_type(); ss.next()) { + BasicType t = ss.type(); + sig_bt[sig_index++] = t; + if (type2size[t] == 2) { + sig_bt[sig_index++] = T_VOID; + } else { + assert(type2size[t] == 1, "size is 1 or 2"); + } + } + assert(sig_index == sizeargs, ""); + } + const char* spname = "sp"; // make arch-specific? + intptr_t out_preserve = SharedRuntime::java_calling_convention(sig_bt, regs, sizeargs, false); + int stack_slot_offset = this->frame_size() * wordSize; + int tab1 = 14, tab2 = 24; + int sig_index = 0; + int arg_index = (m->is_static() ? 0 : -1); + bool did_old_sp = false; + for (SignatureStream ss(m->signature()); !ss.at_return_type(); ) { + bool at_this = (arg_index == -1); + bool at_old_sp = false; + BasicType t = (at_this ? T_OBJECT : ss.type()); + assert(t == sig_bt[sig_index], "sigs in sync"); + if (at_this) + stream->print(" # this: "); + else + stream->print(" # parm%d: ", arg_index); + stream->move_to(tab1); + VMReg fst = regs[sig_index].first(); + VMReg snd = regs[sig_index].second(); + if (fst->is_reg()) { + stream->print("%s", fst->name()); + if (snd->is_valid()) { + stream->print(":%s", snd->name()); + } + } else if (fst->is_stack()) { + int offset = fst->reg2stack() * VMRegImpl::stack_slot_size + stack_slot_offset; + if (offset == stack_slot_offset) at_old_sp = true; + stream->print("[%s+0x%x]", spname, offset); + } else { + stream->print("reg%d:%d??", (int)(intptr_t)fst, (int)(intptr_t)snd); + } + stream->print(" "); + stream->move_to(tab2); + stream->print("= "); + if (at_this) { + m->method_holder()->print_value_on(stream); + } else { + bool did_name = false; + if (!at_this && ss.is_object()) { + symbolOop name = ss.as_symbol_or_null(); + if (name != NULL) { + name->print_value_on(stream); + did_name = true; + } + } + if (!did_name) + stream->print("%s", type2name(t)); + } + if (at_old_sp) { + stream->print(" (%s of caller)", spname); + did_old_sp = true; + } + stream->cr(); + sig_index += type2size[t]; + arg_index += 1; + if (!at_this) ss.next(); + } + if (!did_old_sp) { + stream->print(" # "); + stream->move_to(tab1); + stream->print("[%s+0x%x]", spname, stack_slot_offset); + stream->print(" (%s of caller)", spname); + stream->cr(); + } + } + } +} + void nmethod::print_code_comment_on(outputStream* st, int column, u_char* begin, u_char* end) { // First, find an oopmap in (begin, end]. // We use the odd half-closed interval so that oop maps and scope descs
--- a/src/share/vm/code/nmethod.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/code/nmethod.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -81,18 +81,21 @@ struct nmFlags { friend class VMStructs; - unsigned int version:8; // version number (0 = first version) - unsigned int level:4; // optimization level - unsigned int age:4; // age (in # of sweep steps) + unsigned int version:8; // version number (0 = first version) + unsigned int level:4; // optimization level + unsigned int age:4; // age (in # of sweep steps) - unsigned int state:2; // {alive, zombie, unloaded) + unsigned int state:2; // {alive, zombie, unloaded) - unsigned int isUncommonRecompiled:1; // recompiled because of uncommon trap? - unsigned int isToBeRecompiled:1; // to be recompiled as soon as it matures - unsigned int hasFlushedDependencies:1; // Used for maintenance of dependencies - unsigned int markedForReclamation:1; // Used by NMethodSweeper + unsigned int isUncommonRecompiled:1; // recompiled because of uncommon trap? + unsigned int isToBeRecompiled:1; // to be recompiled as soon as it matures + unsigned int hasFlushedDependencies:1; // Used for maintenance of dependencies + unsigned int markedForReclamation:1; // Used by NMethodSweeper - unsigned int has_unsafe_access:1; // May fault due to unsafe access. + unsigned int has_unsafe_access:1; // May fault due to unsafe access. + unsigned int has_method_handle_invokes:1; // Has this method MethodHandle invokes? + + unsigned int speculatively_disconnected:1; // Marked for potential unload void clear(); }; @@ -136,6 +139,7 @@ // To support simple linked-list chaining of nmethods: nmethod* _osr_link; // from instanceKlass::osr_nmethods_head nmethod* _scavenge_root_link; // from CodeCache::scavenge_root_nmethods + nmethod* _saved_nmethod_link; // from CodeCache::speculatively_disconnect static nmethod* volatile _oops_do_mark_nmethods; nmethod* volatile _oops_do_mark_link; @@ -144,8 +148,12 @@ // Offsets for different nmethod parts int _exception_offset; - // All deoptee's will resume execution at this location described by this offset + // All deoptee's will resume execution at this location described by + // this offset. int _deoptimize_offset; + // All deoptee's at a MethodHandle call site will resume execution + // at this location described by this offset. + int _deoptimize_mh_offset; #ifdef HAVE_DTRACE_H int _trap_offset; #endif // def HAVE_DTRACE_H @@ -252,7 +260,9 @@ void* operator new(size_t size, int nmethod_size); const char* reloc_string_for(u_char* begin, u_char* end); - void make_not_entrant_or_zombie(int state); + // Returns true if this thread changed the state of the nmethod or + // false if another thread performed the transition. + bool make_not_entrant_or_zombie(unsigned int state); void inc_decompile_count(); // used to check that writes to nmFlags are done consistently. @@ -326,24 +336,25 @@ bool is_compiled_by_c2() const; // boundaries for different parts - address code_begin () const { return _entry_point; } - address code_end () const { return header_begin() + _stub_offset ; } - address exception_begin () const { return header_begin() + _exception_offset ; } - address deopt_handler_begin() const { return header_begin() + _deoptimize_offset ; } - address stub_begin () const { return header_begin() + _stub_offset ; } - address stub_end () const { return header_begin() + _consts_offset ; } - address consts_begin () const { return header_begin() + _consts_offset ; } - address consts_end () const { return header_begin() + _scopes_data_offset ; } - address scopes_data_begin () const { return header_begin() + _scopes_data_offset ; } - address scopes_data_end () const { return header_begin() + _scopes_pcs_offset ; } - PcDesc* scopes_pcs_begin () const { return (PcDesc*)(header_begin() + _scopes_pcs_offset ); } - PcDesc* scopes_pcs_end () const { return (PcDesc*)(header_begin() + _dependencies_offset); } - address dependencies_begin () const { return header_begin() + _dependencies_offset ; } - address dependencies_end () const { return header_begin() + _handler_table_offset ; } - address handler_table_begin() const { return header_begin() + _handler_table_offset ; } - address handler_table_end () const { return header_begin() + _nul_chk_table_offset ; } - address nul_chk_table_begin() const { return header_begin() + _nul_chk_table_offset ; } - address nul_chk_table_end () const { return header_begin() + _nmethod_end_offset ; } + address code_begin () const { return _entry_point; } + address code_end () const { return header_begin() + _stub_offset ; } + address exception_begin () const { return header_begin() + _exception_offset ; } + address deopt_handler_begin () const { return header_begin() + _deoptimize_offset ; } + address deopt_mh_handler_begin() const { return header_begin() + _deoptimize_mh_offset ; } + address stub_begin () const { return header_begin() + _stub_offset ; } + address stub_end () const { return header_begin() + _consts_offset ; } + address consts_begin () const { return header_begin() + _consts_offset ; } + address consts_end () const { return header_begin() + _scopes_data_offset ; } + address scopes_data_begin () const { return header_begin() + _scopes_data_offset ; } + address scopes_data_end () const { return header_begin() + _scopes_pcs_offset ; } + PcDesc* scopes_pcs_begin () const { return (PcDesc*)(header_begin() + _scopes_pcs_offset ); } + PcDesc* scopes_pcs_end () const { return (PcDesc*)(header_begin() + _dependencies_offset) ; } + address dependencies_begin () const { return header_begin() + _dependencies_offset ; } + address dependencies_end () const { return header_begin() + _handler_table_offset ; } + address handler_table_begin () const { return header_begin() + _handler_table_offset ; } + address handler_table_end () const { return header_begin() + _nul_chk_table_offset ; } + address nul_chk_table_begin () const { return header_begin() + _nul_chk_table_offset ; } + address nul_chk_table_end () const { return header_begin() + _nmethod_end_offset ; } int code_size () const { return code_end () - code_begin (); } int stub_size () const { return stub_end () - stub_begin (); } @@ -375,10 +386,12 @@ bool is_zombie() const { return flags.state == zombie; } bool is_unloaded() const { return flags.state == unloaded; } - // Make the nmethod non entrant. The nmethod will continue to be alive. - // It is used when an uncommon trap happens. - void make_not_entrant() { make_not_entrant_or_zombie(not_entrant); } - void make_zombie() { make_not_entrant_or_zombie(zombie); } + // Make the nmethod non entrant. The nmethod will continue to be + // alive. It is used when an uncommon trap happens. Returns true + // if this thread changed the state of the nmethod or false if + // another thread performed the transition. + bool make_not_entrant() { return make_not_entrant_or_zombie(not_entrant); } + bool make_zombie() { return make_not_entrant_or_zombie(zombie); } // used by jvmti to track if the unload event has been reported bool unload_reported() { return _unload_reported; } @@ -405,6 +418,12 @@ bool has_unsafe_access() const { return flags.has_unsafe_access; } void set_has_unsafe_access(bool z) { flags.has_unsafe_access = z; } + bool has_method_handle_invokes() const { return flags.has_method_handle_invokes; } + void set_has_method_handle_invokes(bool z) { flags.has_method_handle_invokes = z; } + + bool is_speculatively_disconnected() const { return flags.speculatively_disconnected; } + void set_speculatively_disconnected(bool z) { flags.speculatively_disconnected = z; } + int level() const { return flags.level; } void set_level(int newLevel) { check_safepoint(); flags.level = newLevel; } @@ -429,6 +448,9 @@ nmethod* scavenge_root_link() const { return _scavenge_root_link; } void set_scavenge_root_link(nmethod *n) { _scavenge_root_link = n; } + nmethod* saved_nmethod_link() const { return _saved_nmethod_link; } + void set_saved_nmethod_link(nmethod *n) { _saved_nmethod_link = n; } + public: // Sweeper support @@ -507,7 +529,7 @@ private: ScopeDesc* scope_desc_in(address begin, address end); - address* orig_pc_addr(const frame* fr ) { return (address*) ((address)fr->unextended_sp() + _orig_pc_offset); } + address* orig_pc_addr(const frame* fr) { return (address*) ((address)fr->unextended_sp() + _orig_pc_offset); } PcDesc* find_pc_desc_internal(address pc, bool approximate); @@ -530,13 +552,20 @@ void copy_scopes_pcs(PcDesc* pcs, int count); void copy_scopes_data(address buffer, int size); - // deopt - // return true is the pc is one would expect if the frame is being deopted. - bool is_deopt_pc(address pc); + // Deopt + // Return true is the PC is one would expect if the frame is being deopted. + bool is_deopt_pc (address pc) { return is_deopt_entry(pc) || is_deopt_mh_entry(pc); } + bool is_deopt_entry (address pc) { return pc == deopt_handler_begin(); } + bool is_deopt_mh_entry(address pc) { return pc == deopt_mh_handler_begin(); } // Accessor/mutator for the original pc of a frame before a frame was deopted. address get_original_pc(const frame* fr) { return *orig_pc_addr(fr); } void set_original_pc(const frame* fr, address pc) { *orig_pc_addr(fr) = pc; } + static address get_deopt_original_pc(const frame* fr); + + // MethodHandle + bool is_method_handle_return(address return_pc); + // jvmti support: void post_compiled_method_load_event(); @@ -563,7 +592,14 @@ // Logging void log_identity(xmlStream* log) const; void log_new_nmethod() const; - void log_state_change(int state) const; + void log_state_change() const; + + // Prints block-level comments, including nmethod specific block labels: + virtual void print_block_comment(outputStream* stream, address block_begin) { + print_nmethod_labels(stream, block_begin); + CodeBlob::print_block_comment(stream, block_begin); + } + void print_nmethod_labels(outputStream* stream, address block_begin); // Prints a comment for one native instruction (reloc info, pc desc) void print_code_comment_on(outputStream* st, int column, address begin, address end);
--- a/src/share/vm/code/pcDesc.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/code/pcDesc.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -52,7 +52,8 @@ tty->print(" "); sd->method()->print_short_name(tty); tty->print(" @%d", sd->bci()); - tty->print(" reexecute=%s", sd->should_reexecute()?"true":"false"); + if (sd->should_reexecute()) + tty->print(" reexecute=true"); tty->cr(); } #endif
--- a/src/share/vm/code/pcDesc.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/code/pcDesc.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -38,6 +38,8 @@ int word; struct { unsigned int reexecute: 1; + unsigned int is_method_handle_invoke: 1; + unsigned int return_oop: 1; } bits; bool operator ==(const PcDescFlags& other) { return word == other.word; } } _flags; @@ -72,6 +74,12 @@ _flags == pd->_flags; } + bool is_method_handle_invoke() const { return _flags.bits.is_method_handle_invoke; } + void set_is_method_handle_invoke(bool z) { _flags.bits.is_method_handle_invoke = z; } + + bool return_oop() const { return _flags.bits.return_oop; } + void set_return_oop(bool z) { _flags.bits.return_oop = z; } + // Returns the real pc address real_pc(const nmethod* code) const;
--- a/src/share/vm/code/scopeDesc.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/code/scopeDesc.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -26,19 +26,21 @@ # include "incls/_scopeDesc.cpp.incl" -ScopeDesc::ScopeDesc(const nmethod* code, int decode_offset, int obj_decode_offset, bool reexecute) { +ScopeDesc::ScopeDesc(const nmethod* code, int decode_offset, int obj_decode_offset, bool reexecute, bool return_oop) { _code = code; _decode_offset = decode_offset; _objects = decode_object_values(obj_decode_offset); _reexecute = reexecute; + _return_oop = return_oop; decode_body(); } -ScopeDesc::ScopeDesc(const nmethod* code, int decode_offset, bool reexecute) { +ScopeDesc::ScopeDesc(const nmethod* code, int decode_offset, bool reexecute, bool return_oop) { _code = code; _decode_offset = decode_offset; _objects = decode_object_values(DebugInformationRecorder::serialized_null); _reexecute = reexecute; + _return_oop = return_oop; decode_body(); } @@ -48,6 +50,7 @@ _decode_offset = parent->_sender_decode_offset; _objects = parent->_objects; _reexecute = false; //reexecute only applies to the first scope + _return_oop = false; decode_body(); }
--- a/src/share/vm/code/scopeDesc.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/code/scopeDesc.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -52,17 +52,18 @@ class ScopeDesc : public ResourceObj { public: // Constructor - ScopeDesc(const nmethod* code, int decode_offset, int obj_decode_offset, bool reexecute); + ScopeDesc(const nmethod* code, int decode_offset, int obj_decode_offset, bool reexecute, bool return_oop); // Calls above, giving default value of "serialized_null" to the // "obj_decode_offset" argument. (We don't use a default argument to // avoid a .hpp-.hpp dependency.) - ScopeDesc(const nmethod* code, int decode_offset, bool reexecute); + ScopeDesc(const nmethod* code, int decode_offset, bool reexecute, bool return_oop); // JVM state methodHandle method() const { return _method; } int bci() const { return _bci; } bool should_reexecute() const { return _reexecute; } + bool return_oop() const { return _return_oop; } GrowableArray<ScopeValue*>* locals(); GrowableArray<ScopeValue*>* expressions(); @@ -88,6 +89,7 @@ methodHandle _method; int _bci; bool _reexecute; + bool _return_oop; // Decoding offsets int _decode_offset;
--- a/src/share/vm/compiler/compileBroker.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/compiler/compileBroker.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -69,6 +69,7 @@ bool CompileBroker::_initialized = false; volatile bool CompileBroker::_should_block = false; +volatile jint CompileBroker::_should_compile_new_jobs = run_compilation; // The installed compiler(s) AbstractCompiler* CompileBroker::_compilers[2]; @@ -986,6 +987,13 @@ return method_code; } if (method->is_not_compilable(comp_level)) return NULL; + + nmethod* saved = CodeCache::find_and_remove_saved_code(method()); + if (saved != NULL) { + method->set_code(method, saved); + return saved; + } + } else { // osr compilation #ifndef TIERED @@ -1037,6 +1045,14 @@ method->jmethod_id(); } + // If the compiler is shut off due to code cache flushing or otherwise, + // fail out now so blocking compiles dont hang the java thread + if (!should_compile_new_jobs() || (UseCodeCacheFlushing && CodeCache::needs_flushing())) { + method->invocation_counter()->decay(); + method->backedge_counter()->decay(); + return NULL; + } + // do the compilation if (method->is_native()) { if (!PreferInterpreterNativeStubs) { @@ -1116,7 +1132,7 @@ // the specified level if (is_native && (!CICompileNatives || !compiler(comp_level)->supports_native())) { - method->set_not_compilable(); + method->set_not_compilable_quietly(); return true; } @@ -1140,7 +1156,7 @@ method->print_short_name(tty); tty->cr(); } - method->set_not_compilable(); + method->set_not_compilable_quietly(); } return false; @@ -1173,7 +1189,7 @@ } // Method was not in the appropriate compilation range. - method->set_not_compilable(); + method->set_not_compilable_quietly(); return 0; } @@ -1325,26 +1341,13 @@ { // We need this HandleMark to avoid leaking VM handles. HandleMark hm(thread); + if (CodeCache::unallocated_capacity() < CodeCacheMinimumFreeSpace) { - // The CodeCache is full. Print out warning and disable compilation. - UseInterpreter = true; - if (UseCompiler || AlwaysCompileLoopMethods ) { - if (log != NULL) { - log->begin_elem("code_cache_full"); - log->stamp(); - log->end_elem(); - } -#ifndef PRODUCT - warning("CodeCache is full. Compiler has been disabled"); - if (CompileTheWorld || ExitOnFullCodeCache) { - before_exit(thread); - exit_globals(); // will delete tty - vm_direct_exit(CompileTheWorld ? 0 : 1); - } -#endif - UseCompiler = false; - AlwaysCompileLoopMethods = false; - } + // the code cache is really full + handle_full_code_cache(); + } else if (UseCodeCacheFlushing && CodeCache::needs_flushing()) { + // Attempt to start cleaning the code cache while there is still a little headroom + NMethodSweeper::handle_full_code_cache(false); } CompileTask* task = queue->get(); @@ -1369,7 +1372,7 @@ // Never compile a method if breakpoints are present in it if (method()->number_of_breakpoints() == 0) { // Compile the method. - if (UseCompiler || AlwaysCompileLoopMethods) { + if ((UseCompiler || AlwaysCompileLoopMethods) && CompileBroker::should_compile_new_jobs()) { #ifdef COMPILER1 // Allow repeating compilations for the purpose of benchmarking // compile speed. This is not useful for customers. @@ -1587,10 +1590,10 @@ if (is_osr) { method->set_not_osr_compilable(); } else { - method->set_not_compilable(); + method->set_not_compilable_quietly(); } } else if (compilable == ciEnv::MethodCompilable_not_at_tier) { - method->set_not_compilable(task->comp_level()); + method->set_not_compilable_quietly(task->comp_level()); } // Note that the queued_for_compilation bits are cleared without @@ -1614,6 +1617,38 @@ // ------------------------------------------------------------------ +// CompileBroker::handle_full_code_cache +// +// The CodeCache is full. Print out warning and disable compilation or +// try code cache cleaning so compilation can continue later. +void CompileBroker::handle_full_code_cache() { + UseInterpreter = true; + if (UseCompiler || AlwaysCompileLoopMethods ) { + CompilerThread* thread = CompilerThread::current(); + CompileLog* log = thread->log(); + if (log != NULL) { + log->begin_elem("code_cache_full"); + log->stamp(); + log->end_elem(); + } + #ifndef PRODUCT + warning("CodeCache is full. Compiler has been disabled"); + if (CompileTheWorld || ExitOnFullCodeCache) { + before_exit(JavaThread::current()); + exit_globals(); // will delete tty + vm_direct_exit(CompileTheWorld ? 0 : 1); + } + #endif + if (UseCodeCacheFlushing) { + NMethodSweeper::handle_full_code_cache(true); + } else { + UseCompiler = false; + AlwaysCompileLoopMethods = false; + } + } +} + +// ------------------------------------------------------------------ // CompileBroker::set_last_compile // // Record this compilation for debugging purposes. @@ -1820,9 +1855,11 @@ CompileBroker::_t_standard_compilation.seconds(), CompileBroker::_t_standard_compilation.seconds() / CompileBroker::_total_standard_compile_count); tty->print_cr(" On stack replacement : %6.3f s, Average : %2.3f", CompileBroker::_t_osr_compilation.seconds(), CompileBroker::_t_osr_compilation.seconds() / CompileBroker::_total_osr_compile_count); - compiler(CompLevel_fast_compile)->print_timers(); - if (compiler(CompLevel_fast_compile) != compiler(CompLevel_highest_tier)) { - compiler(CompLevel_highest_tier)->print_timers(); + + if (compiler(CompLevel_fast_compile)) { + compiler(CompLevel_fast_compile)->print_timers(); + if (compiler(CompLevel_fast_compile) != compiler(CompLevel_highest_tier)) + compiler(CompLevel_highest_tier)->print_timers(); } tty->cr();
--- a/src/share/vm/compiler/compileBroker.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/compiler/compileBroker.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -193,6 +193,9 @@ static bool _initialized; static volatile bool _should_block; + // This flag can be used to stop compilation or turn it back on + static volatile jint _should_compile_new_jobs; + // The installed compiler(s) static AbstractCompiler* _compilers[2]; @@ -319,6 +322,7 @@ static void compiler_thread_loop(); + static uint get_compilation_id() { return _compilation_id; } static bool is_idle(); // Set _should_block. @@ -328,6 +332,20 @@ // Call this from the compiler at convenient points, to poll for _should_block. static void maybe_block(); + enum { + // Flags for toggling compiler activity + stop_compilation = 0, + run_compilation = 1 + }; + + static bool should_compile_new_jobs() { return UseCompiler && (_should_compile_new_jobs == run_compilation); } + static bool set_should_compile_new_jobs(jint new_state) { + // Return success if the current caller set it + jint old = Atomic::cmpxchg(new_state, &_should_compile_new_jobs, 1-new_state); + return (old == (1-new_state)); + } + static void handle_full_code_cache(); + // Return total compilation ticks static jlong total_compilation_ticks() { return _perf_total_compilation != NULL ? _perf_total_compilation->get_value() : 0;
--- a/src/share/vm/compiler/compilerOracle.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/compiler/compilerOracle.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -392,18 +392,18 @@ }; static MethodMatcher::Mode check_mode(char name[], const char*& error_msg) { - if (strcmp(name, "*") == 0) return MethodMatcher::Any; - int match = MethodMatcher::Exact; - if (name[0] == '*') { + while (name[0] == '*') { match |= MethodMatcher::Suffix; strcpy(name, name + 1); } + if (strcmp(name, "*") == 0) return MethodMatcher::Any; + size_t len = strlen(name); - if (len > 0 && name[len - 1] == '*') { + while (len > 0 && name[len - 1] == '*') { match |= MethodMatcher::Prefix; - name[len - 1] = '\0'; + name[--len] = '\0'; } if (strstr(name, "*") != NULL) { @@ -610,6 +610,14 @@ CompilerOracle::parse_from_string(CompileCommand, CompilerOracle::parse_from_line); CompilerOracle::parse_from_string(CompileOnly, CompilerOracle::parse_compile_only); CompilerOracle::parse_from_file(); + if (lists[PrintCommand] != NULL) { + if (PrintAssembly) { + warning("CompileCommand and/or .hotspot_compiler file contains 'print' commands, but PrintAssembly is also enabled"); + } else if (FLAG_IS_DEFAULT(DebugNonSafepoints)) { + warning("printing of assembly code is enabled; turning on DebugNonSafepoints to gain additional output"); + DebugNonSafepoints = true; + } + } }
--- a/src/share/vm/compiler/disassembler.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/compiler/disassembler.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -151,8 +151,10 @@ outputStream* st = output(); if (_print_bytes && pc > pc0) print_insn_bytes(pc0, pc); - if (_nm != NULL) + if (_nm != NULL) { _nm->print_code_comment_on(st, COMMENT_COLUMN, pc0, pc); + // this calls reloc_string_for which calls oop::print_value_on + } // Output pc bucket ticks if we have any if (total_ticks() != 0) { @@ -273,8 +275,15 @@ oop obj; if (_nm != NULL && (obj = _nm->embeddedOop_at(cur_insn())) != NULL - && (address) obj == adr) { + && (address) obj == adr + && Universe::heap()->is_in(obj) + && Universe::heap()->is_in(obj->klass())) { + julong c = st->count(); obj->print_value_on(st); + if (st->count() == c) { + // No output. (Can happen in product builds.) + st->print("(a %s)", Klass::cast(obj->klass())->external_name()); + } return; } } @@ -286,17 +295,9 @@ void decode_env::print_insn_labels() { address p = cur_insn(); outputStream* st = output(); - nmethod* nm = _nm; - if (nm != NULL) { - if (p == nm->entry_point()) st->print_cr("[Entry Point]"); - if (p == nm->verified_entry_point()) st->print_cr("[Verified Entry Point]"); - if (p == nm->exception_begin()) st->print_cr("[Exception Handler]"); - if (p == nm->stub_begin()) st->print_cr("[Stub Code]"); - if (p == nm->consts_begin()) st->print_cr("[Constants]"); - } CodeBlob* cb = _code; if (cb != NULL) { - cb->print_block_comment(st, (intptr_t)(p - cb->instructions_begin())); + cb->print_block_comment(st, p); } if (_print_pc) { st->print(" " INTPTR_FORMAT ": ", (intptr_t) p);
--- a/src/share/vm/compiler/methodLiveness.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/compiler/methodLiveness.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -782,6 +782,7 @@ case Bytecodes::_invokespecial: case Bytecodes::_invokestatic: case Bytecodes::_invokeinterface: + case Bytecodes::_invokedynamic: case Bytecodes::_newarray: case Bytecodes::_anewarray: case Bytecodes::_checkcast:
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/binaryTreeDictionary.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/binaryTreeDictionary.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -62,12 +62,13 @@ tl->link_head(tc); tl->link_tail(tc); tl->set_count(1); - tl->init_statistics(); + tl->init_statistics(true /* split_birth */); tl->setParent(NULL); tl->setLeft(NULL); tl->setRight(NULL); return tl; } + TreeList* TreeList::as_TreeList(HeapWord* addr, size_t size) { TreeChunk* tc = (TreeChunk*) addr; assert(size >= sizeof(TreeChunk), "Chunk is too small for a TreeChunk"); @@ -267,6 +268,31 @@ return retTC; } +// Returns the block with the largest heap address amongst +// those in the list for this size; potentially slow and expensive, +// use with caution! +TreeChunk* TreeList::largest_address() { + guarantee(head() != NULL, "The head of the list cannot be NULL"); + FreeChunk* fc = head()->next(); + TreeChunk* retTC; + if (fc == NULL) { + retTC = head_as_TreeChunk(); + } else { + // walk down the list and return the one with the highest + // heap address among chunks of this size. + FreeChunk* last = fc; + while (fc->next() != NULL) { + if ((HeapWord*)last < (HeapWord*)fc) { + last = fc; + } + fc = fc->next(); + } + retTC = TreeChunk::as_TreeChunk(last); + } + assert(retTC->list() == this, "Wrong type of chunk."); + return retTC; +} + BinaryTreeDictionary::BinaryTreeDictionary(MemRegion mr, bool splay): _splay(splay) { @@ -379,7 +405,7 @@ break; } // The evm code reset the hint of the candidate as - // at an interrim point. Why? Seems like this leaves + // at an interim point. Why? Seems like this leaves // the hint pointing to a list that didn't work. // curTL->set_hint(hintTL->size()); } @@ -436,7 +462,7 @@ TreeList *curTL = root(); if (curTL != NULL) { while(curTL->right() != NULL) curTL = curTL->right(); - return curTL->first_available(); + return curTL->largest_address(); } else { return NULL; } @@ -664,7 +690,7 @@ } } TreeChunk* tc = TreeChunk::as_TreeChunk(fc); - // This chunk is being returned to the binary try. It's embedded + // This chunk is being returned to the binary tree. Its embedded // TreeList should be unused at this point. tc->initialize(); if (curTL != NULL) { // exact match @@ -807,6 +833,8 @@ } bool BinaryTreeDictionary::coalDictOverPopulated(size_t size) { + if (FLSAlwaysCoalesceLarge) return true; + TreeList* list_of_size = findList(size); // None of requested size implies overpopulated. return list_of_size == NULL || list_of_size->coalDesired() <= 0 || @@ -854,17 +882,20 @@ double _percentage; float _inter_sweep_current; float _inter_sweep_estimate; + float _intra_sweep_estimate; public: BeginSweepClosure(double p, float inter_sweep_current, - float inter_sweep_estimate) : + float inter_sweep_estimate, + float intra_sweep_estimate) : _percentage(p), _inter_sweep_current(inter_sweep_current), - _inter_sweep_estimate(inter_sweep_estimate) { } + _inter_sweep_estimate(inter_sweep_estimate), + _intra_sweep_estimate(intra_sweep_estimate) { } void do_list(FreeList* fl) { double coalSurplusPercent = _percentage; - fl->compute_desired(_inter_sweep_current, _inter_sweep_estimate); + fl->compute_desired(_inter_sweep_current, _inter_sweep_estimate, _intra_sweep_estimate); fl->set_coalDesired((ssize_t)((double)fl->desired() * coalSurplusPercent)); fl->set_beforeSweep(fl->count()); fl->set_bfrSurp(fl->surplus()); @@ -939,9 +970,10 @@ } void BinaryTreeDictionary::beginSweepDictCensus(double coalSurplusPercent, - float inter_sweep_current, float inter_sweep_estimate) { + float inter_sweep_current, float inter_sweep_estimate, float intra_sweep_estimate) { BeginSweepClosure bsc(coalSurplusPercent, inter_sweep_current, - inter_sweep_estimate); + inter_sweep_estimate, + intra_sweep_estimate); bsc.do_tree(root()); } @@ -1077,13 +1109,13 @@ // Print census information - counts, births, deaths, etc. // for each list in the tree. Also print some summary // information. -class printTreeCensusClosure : public AscendTreeCensusClosure { +class PrintTreeCensusClosure : public AscendTreeCensusClosure { int _print_line; size_t _totalFree; FreeList _total; public: - printTreeCensusClosure() { + PrintTreeCensusClosure() { _print_line = 0; _totalFree = 0; } @@ -1113,7 +1145,7 @@ gclog_or_tty->print("\nBinaryTree\n"); FreeList::print_labels_on(gclog_or_tty, "size"); - printTreeCensusClosure ptc; + PrintTreeCensusClosure ptc; ptc.do_tree(root()); FreeList* total = ptc.total(); @@ -1130,6 +1162,38 @@ /(total->desired() != 0 ? (double)total->desired() : 1.0)); } +class PrintFreeListsClosure : public AscendTreeCensusClosure { + outputStream* _st; + int _print_line; + + public: + PrintFreeListsClosure(outputStream* st) { + _st = st; + _print_line = 0; + } + void do_list(FreeList* fl) { + if (++_print_line >= 40) { + FreeList::print_labels_on(_st, "size"); + _print_line = 0; + } + fl->print_on(gclog_or_tty); + size_t sz = fl->size(); + for (FreeChunk* fc = fl->head(); fc != NULL; + fc = fc->next()) { + _st->print_cr("\t[" PTR_FORMAT "," PTR_FORMAT ") %s", + fc, (HeapWord*)fc + sz, + fc->cantCoalesce() ? "\t CC" : ""); + } + } +}; + +void BinaryTreeDictionary::print_free_lists(outputStream* st) const { + + FreeList::print_labels_on(st, "size"); + PrintFreeListsClosure pflc(st); + pflc.do_tree(root()); +} + // Verify the following tree invariants: // . _root has no parent // . parent and child point to each other
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/binaryTreeDictionary.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/binaryTreeDictionary.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -42,9 +42,6 @@ friend class AscendTreeCensusClosure; friend class DescendTreeCensusClosure; friend class DescendTreeSearchClosure; - TreeList* _parent; - TreeList* _left; - TreeList* _right; protected: TreeList* parent() const { return _parent; } @@ -82,6 +79,11 @@ // to a TreeChunk. TreeChunk* first_available(); + // Returns the block with the largest heap address amongst + // those in the list for this size; potentially slow and expensive, + // use with caution! + TreeChunk* largest_address(); + // removeChunkReplaceIfNeeded() removes the given "tc" from the TreeList. // If "tc" is the first chunk in the list, it is also the // TreeList that is the node in the tree. removeChunkReplaceIfNeeded() @@ -254,8 +256,9 @@ // Methods called at the beginning of a sweep to prepare the // statistics for the sweep. void beginSweepDictCensus(double coalSurplusPercent, - float sweep_current, - float sweep_estimate); + float inter_sweep_current, + float inter_sweep_estimate, + float intra_sweep_estimate); // Methods called after the end of a sweep to modify the // statistics for the sweep. void endSweepDictCensus(double splitSurplusPercent); @@ -269,6 +272,7 @@ // Print the statistcis for all the lists in the tree. Also may // print out summaries. void printDictCensus(void) const; + void print_free_lists(outputStream* st) const; // For debugging. Returns the sum of the _returnedBytes for // all lists in the tree.
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/cmsLockVerifier.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/cmsLockVerifier.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -32,7 +32,9 @@ // threads. The second argument is in support of an extra locking // check for CFL spaces' free list locks. #ifndef PRODUCT -void CMSLockVerifier::assert_locked(const Mutex* lock, const Mutex* p_lock) { +void CMSLockVerifier::assert_locked(const Mutex* lock, + const Mutex* p_lock1, + const Mutex* p_lock2) { if (!Universe::is_fully_initialized()) { return; } @@ -40,7 +42,7 @@ Thread* myThread = Thread::current(); if (lock == NULL) { // a "lock-free" structure, e.g. MUT, protected by CMS token - assert(p_lock == NULL, "Unexpected state"); + assert(p_lock1 == NULL && p_lock2 == NULL, "Unexpected caller error"); if (myThread->is_ConcurrentGC_thread()) { // This test might have to change in the future, if there can be // multiple peer CMS threads. But for now, if we're testing the CMS @@ -60,36 +62,39 @@ return; } - if (ParallelGCThreads == 0) { + if (myThread->is_VM_thread() + || myThread->is_ConcurrentGC_thread() + || myThread->is_Java_thread()) { + // Make sure that we are holding the associated lock. assert_lock_strong(lock); + // The checking of p_lock is a spl case for CFLS' free list + // locks: we make sure that none of the parallel GC work gang + // threads are holding "sub-locks" of freeListLock(). We check only + // the parDictionaryAllocLock because the others are too numerous. + // This spl case code is somewhat ugly and any improvements + // are welcome. + assert(p_lock1 == NULL || !p_lock1->is_locked() || p_lock1->owned_by_self(), + "Possible race between this and parallel GC threads"); + assert(p_lock2 == NULL || !p_lock2->is_locked() || p_lock2->owned_by_self(), + "Possible race between this and parallel GC threads"); + } else if (myThread->is_GC_task_thread()) { + // Make sure that the VM or CMS thread holds lock on our behalf + // XXX If there were a concept of a gang_master for a (set of) + // gang_workers, we could have used the identity of that thread + // for checking ownership here; for now we just disjunct. + assert(lock->owner() == VMThread::vm_thread() || + lock->owner() == ConcurrentMarkSweepThread::cmst(), + "Should be locked by VM thread or CMS thread on my behalf"); + if (p_lock1 != NULL) { + assert_lock_strong(p_lock1); + } + if (p_lock2 != NULL) { + assert_lock_strong(p_lock2); + } } else { - if (myThread->is_VM_thread() - || myThread->is_ConcurrentGC_thread() - || myThread->is_Java_thread()) { - // Make sure that we are holding the associated lock. - assert_lock_strong(lock); - // The checking of p_lock is a spl case for CFLS' free list - // locks: we make sure that none of the parallel GC work gang - // threads are holding "sub-locks" of freeListLock(). We check only - // the parDictionaryAllocLock because the others are too numerous. - // This spl case code is somewhat ugly and any improvements - // are welcome XXX FIX ME!! - if (p_lock != NULL) { - assert(!p_lock->is_locked() || p_lock->owned_by_self(), - "Possible race between this and parallel GC threads"); - } - } else if (myThread->is_GC_task_thread()) { - // Make sure that the VM or CMS thread holds lock on our behalf - // XXX If there were a concept of a gang_master for a (set of) - // gang_workers, we could have used the identity of that thread - // for checking ownership here; for now we just disjunct. - assert(lock->owner() == VMThread::vm_thread() || - lock->owner() == ConcurrentMarkSweepThread::cmst(), - "Should be locked by VM thread or CMS thread on my behalf"); - } else { - // Make sure we didn't miss some obscure corner case - ShouldNotReachHere(); - } + // Make sure we didn't miss some other thread type calling into here; + // perhaps as a result of future VM evolution. + ShouldNotReachHere(); } } #endif
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/cmsLockVerifier.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/cmsLockVerifier.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -29,8 +29,11 @@ // the parallel threads. class CMSLockVerifier: AllStatic { public: - static void assert_locked(const Mutex* lock, const Mutex* p_lock) + static void assert_locked(const Mutex* lock, const Mutex* p_lock1, const Mutex* p_lock2) PRODUCT_RETURN; + static void assert_locked(const Mutex* lock, const Mutex* p_lock) { + assert_locked(lock, p_lock, NULL); + } static void assert_locked(const Mutex* lock) { assert_locked(lock, NULL); }
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -62,18 +62,15 @@ // implementation, namely, the simple binary tree (splaying // temporarily disabled). switch (dictionaryChoice) { - case FreeBlockDictionary::dictionaryBinaryTree: - _dictionary = new BinaryTreeDictionary(mr); - break; case FreeBlockDictionary::dictionarySplayTree: case FreeBlockDictionary::dictionarySkipList: default: warning("dictionaryChoice: selected option not understood; using" " default BinaryTreeDictionary implementation instead."); + case FreeBlockDictionary::dictionaryBinaryTree: _dictionary = new BinaryTreeDictionary(mr); break; } - splitBirth(mr.word_size()); assert(_dictionary != NULL, "CMS dictionary initialization"); // The indexed free lists are initially all empty and are lazily // filled in on demand. Initialize the array elements to NULL. @@ -388,6 +385,105 @@ return res; } +void CompactibleFreeListSpace::print_indexed_free_lists(outputStream* st) +const { + reportIndexedFreeListStatistics(); + gclog_or_tty->print_cr("Layout of Indexed Freelists"); + gclog_or_tty->print_cr("---------------------------"); + FreeList::print_labels_on(st, "size"); + for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) { + _indexedFreeList[i].print_on(gclog_or_tty); + for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL; + fc = fc->next()) { + gclog_or_tty->print_cr("\t[" PTR_FORMAT "," PTR_FORMAT ") %s", + fc, (HeapWord*)fc + i, + fc->cantCoalesce() ? "\t CC" : ""); + } + } +} + +void CompactibleFreeListSpace::print_promo_info_blocks(outputStream* st) +const { + _promoInfo.print_on(st); +} + +void CompactibleFreeListSpace::print_dictionary_free_lists(outputStream* st) +const { + _dictionary->reportStatistics(); + st->print_cr("Layout of Freelists in Tree"); + st->print_cr("---------------------------"); + _dictionary->print_free_lists(st); +} + +class BlkPrintingClosure: public BlkClosure { + const CMSCollector* _collector; + const CompactibleFreeListSpace* _sp; + const CMSBitMap* _live_bit_map; + const bool _post_remark; + outputStream* _st; +public: + BlkPrintingClosure(const CMSCollector* collector, + const CompactibleFreeListSpace* sp, + const CMSBitMap* live_bit_map, + outputStream* st): + _collector(collector), + _sp(sp), + _live_bit_map(live_bit_map), + _post_remark(collector->abstract_state() > CMSCollector::FinalMarking), + _st(st) { } + size_t do_blk(HeapWord* addr); +}; + +size_t BlkPrintingClosure::do_blk(HeapWord* addr) { + size_t sz = _sp->block_size_no_stall(addr, _collector); + assert(sz != 0, "Should always be able to compute a size"); + if (_sp->block_is_obj(addr)) { + const bool dead = _post_remark && !_live_bit_map->isMarked(addr); + _st->print_cr(PTR_FORMAT ": %s object of size " SIZE_FORMAT "%s", + addr, + dead ? "dead" : "live", + sz, + (!dead && CMSPrintObjectsInDump) ? ":" : "."); + if (CMSPrintObjectsInDump && !dead) { + oop(addr)->print_on(_st); + _st->print_cr("--------------------------------------"); + } + } else { // free block + _st->print_cr(PTR_FORMAT ": free block of size " SIZE_FORMAT "%s", + addr, sz, CMSPrintChunksInDump ? ":" : "."); + if (CMSPrintChunksInDump) { + ((FreeChunk*)addr)->print_on(_st); + _st->print_cr("--------------------------------------"); + } + } + return sz; +} + +void CompactibleFreeListSpace::dump_at_safepoint_with_locks(CMSCollector* c, + outputStream* st) { + st->print_cr("\n========================="); + st->print_cr("Block layout in CMS Heap:"); + st->print_cr("========================="); + BlkPrintingClosure bpcl(c, this, c->markBitMap(), st); + blk_iterate(&bpcl); + + st->print_cr("\n======================================="); + st->print_cr("Order & Layout of Promotion Info Blocks"); + st->print_cr("======================================="); + print_promo_info_blocks(st); + + st->print_cr("\n==========================="); + st->print_cr("Order of Indexed Free Lists"); + st->print_cr("========================="); + print_indexed_free_lists(st); + + st->print_cr("\n================================="); + st->print_cr("Order of Free Lists in Dictionary"); + st->print_cr("================================="); + print_dictionary_free_lists(st); +} + + void CompactibleFreeListSpace::reportFreeListStatistics() const { assert_lock_strong(&_freelistLock); assert(PrintFLSStatistics != 0, "Reporting error"); @@ -449,37 +545,37 @@ if (prevEnd != NULL) { // Resize the underlying block offset table. _bt.resize(pointer_delta(value, bottom())); - if (value <= prevEnd) { - assert(value >= unallocated_block(), "New end is below unallocated block"); - } else { - // Now, take this new chunk and add it to the free blocks. - // Note that the BOT has not yet been updated for this block. - size_t newFcSize = pointer_delta(value, prevEnd); - // XXX This is REALLY UGLY and should be fixed up. XXX - if (!_adaptive_freelists && _smallLinearAllocBlock._ptr == NULL) { - // Mark the boundary of the new block in BOT - _bt.mark_block(prevEnd, value); - // put it all in the linAB - if (ParallelGCThreads == 0) { - _smallLinearAllocBlock._ptr = prevEnd; - _smallLinearAllocBlock._word_size = newFcSize; - repairLinearAllocBlock(&_smallLinearAllocBlock); - } else { // ParallelGCThreads > 0 - MutexLockerEx x(parDictionaryAllocLock(), - Mutex::_no_safepoint_check_flag); - _smallLinearAllocBlock._ptr = prevEnd; - _smallLinearAllocBlock._word_size = newFcSize; - repairLinearAllocBlock(&_smallLinearAllocBlock); + if (value <= prevEnd) { + assert(value >= unallocated_block(), "New end is below unallocated block"); + } else { + // Now, take this new chunk and add it to the free blocks. + // Note that the BOT has not yet been updated for this block. + size_t newFcSize = pointer_delta(value, prevEnd); + // XXX This is REALLY UGLY and should be fixed up. XXX + if (!_adaptive_freelists && _smallLinearAllocBlock._ptr == NULL) { + // Mark the boundary of the new block in BOT + _bt.mark_block(prevEnd, value); + // put it all in the linAB + if (ParallelGCThreads == 0) { + _smallLinearAllocBlock._ptr = prevEnd; + _smallLinearAllocBlock._word_size = newFcSize; + repairLinearAllocBlock(&_smallLinearAllocBlock); + } else { // ParallelGCThreads > 0 + MutexLockerEx x(parDictionaryAllocLock(), + Mutex::_no_safepoint_check_flag); + _smallLinearAllocBlock._ptr = prevEnd; + _smallLinearAllocBlock._word_size = newFcSize; + repairLinearAllocBlock(&_smallLinearAllocBlock); + } + // Births of chunks put into a LinAB are not recorded. Births + // of chunks as they are allocated out of a LinAB are. + } else { + // Add the block to the free lists, if possible coalescing it + // with the last free block, and update the BOT and census data. + addChunkToFreeListsAtEndRecordingStats(prevEnd, newFcSize); } - // Births of chunks put into a LinAB are not recorded. Births - // of chunks as they are allocated out of a LinAB are. - } else { - // Add the block to the free lists, if possible coalescing it - // with the last free block, and update the BOT and census data. - addChunkToFreeListsAtEndRecordingStats(prevEnd, newFcSize); } } - } } class FreeListSpace_DCTOC : public Filtering_DCTOC { @@ -732,7 +828,7 @@ void CompactibleFreeListSpace::object_iterate_mem(MemRegion mr, UpwardsObjectClosure* cl) { - assert_locked(); + assert_locked(freelistLock()); NOT_PRODUCT(verify_objects_initialized()); Space::object_iterate_mem(mr, cl); } @@ -1212,12 +1308,15 @@ void CompactibleFreeListSpace::assert_locked() const { CMSLockVerifier::assert_locked(freelistLock(), parDictionaryAllocLock()); } + +void CompactibleFreeListSpace::assert_locked(const Mutex* lock) const { + CMSLockVerifier::assert_locked(lock); +} #endif FreeChunk* CompactibleFreeListSpace::allocateScratch(size_t size) { // In the parallel case, the main thread holds the free list lock // on behalf the parallel threads. - assert_locked(); FreeChunk* fc; { // If GC is parallel, this might be called by several threads. @@ -1298,17 +1397,18 @@ res = blk->_ptr; _bt.allocated(res, blk->_word_size); } else if (size + MinChunkSize <= blk->_refillSize) { + size_t sz = blk->_word_size; // Update _unallocated_block if the size is such that chunk would be // returned to the indexed free list. All other chunks in the indexed // free lists are allocated from the dictionary so that _unallocated_block // has already been adjusted for them. Do it here so that the cost // for all chunks added back to the indexed free lists. - if (blk->_word_size < SmallForDictionary) { - _bt.allocated(blk->_ptr, blk->_word_size); + if (sz < SmallForDictionary) { + _bt.allocated(blk->_ptr, sz); } // Return the chunk that isn't big enough, and then refill below. - addChunkToFreeLists(blk->_ptr, blk->_word_size); - _bt.verify_single_block(blk->_ptr, (blk->_ptr + blk->_word_size)); + addChunkToFreeLists(blk->_ptr, sz); + splitBirth(sz); // Don't keep statistics on adding back chunk from a LinAB. } else { // A refilled block would not satisfy the request. @@ -1376,11 +1476,13 @@ res = getChunkFromIndexedFreeListHelper(size); } _bt.verify_not_unallocated((HeapWord*) res, size); + assert(res == NULL || res->size() == size, "Incorrect block size"); return res; } FreeChunk* -CompactibleFreeListSpace::getChunkFromIndexedFreeListHelper(size_t size) { +CompactibleFreeListSpace::getChunkFromIndexedFreeListHelper(size_t size, + bool replenish) { assert_locked(); FreeChunk* fc = NULL; if (size < SmallForDictionary) { @@ -1398,54 +1500,66 @@ // and replenishing indexed lists from the small linAB. // FreeChunk* newFc = NULL; - size_t replenish_size = CMSIndexedFreeListReplenish * size; + const size_t replenish_size = CMSIndexedFreeListReplenish * size; if (replenish_size < SmallForDictionary) { // Do not replenish from an underpopulated size. if (_indexedFreeList[replenish_size].surplus() > 0 && _indexedFreeList[replenish_size].head() != NULL) { - newFc = - _indexedFreeList[replenish_size].getChunkAtHead(); - } else { + newFc = _indexedFreeList[replenish_size].getChunkAtHead(); + } else if (bestFitFirst()) { newFc = bestFitSmall(replenish_size); } } - if (newFc != NULL) { - splitDeath(replenish_size); - } else if (replenish_size > size) { + if (newFc == NULL && replenish_size > size) { assert(CMSIndexedFreeListReplenish > 1, "ctl pt invariant"); - newFc = - getChunkFromIndexedFreeListHelper(replenish_size); + newFc = getChunkFromIndexedFreeListHelper(replenish_size, false); } + // Note: The stats update re split-death of block obtained above + // will be recorded below precisely when we know we are going to + // be actually splitting it into more than one pieces below. if (newFc != NULL) { - assert(newFc->size() == replenish_size, "Got wrong size"); - size_t i; - FreeChunk *curFc, *nextFc; - // carve up and link blocks 0, ..., CMSIndexedFreeListReplenish - 2 - // The last chunk is not added to the lists but is returned as the - // free chunk. - for (curFc = newFc, nextFc = (FreeChunk*)((HeapWord*)curFc + size), - i = 0; - i < (CMSIndexedFreeListReplenish - 1); - curFc = nextFc, nextFc = (FreeChunk*)((HeapWord*)nextFc + size), - i++) { + if (replenish || CMSReplenishIntermediate) { + // Replenish this list and return one block to caller. + size_t i; + FreeChunk *curFc, *nextFc; + size_t num_blk = newFc->size() / size; + assert(num_blk >= 1, "Smaller than requested?"); + assert(newFc->size() % size == 0, "Should be integral multiple of request"); + if (num_blk > 1) { + // we are sure we will be splitting the block just obtained + // into multiple pieces; record the split-death of the original + splitDeath(replenish_size); + } + // carve up and link blocks 0, ..., num_blk - 2 + // The last chunk is not added to the lists but is returned as the + // free chunk. + for (curFc = newFc, nextFc = (FreeChunk*)((HeapWord*)curFc + size), + i = 0; + i < (num_blk - 1); + curFc = nextFc, nextFc = (FreeChunk*)((HeapWord*)nextFc + size), + i++) { + curFc->setSize(size); + // Don't record this as a return in order to try and + // determine the "returns" from a GC. + _bt.verify_not_unallocated((HeapWord*) fc, size); + _indexedFreeList[size].returnChunkAtTail(curFc, false); + _bt.mark_block((HeapWord*)curFc, size); + splitBirth(size); + // Don't record the initial population of the indexed list + // as a split birth. + } + + // check that the arithmetic was OK above + assert((HeapWord*)nextFc == (HeapWord*)newFc + num_blk*size, + "inconsistency in carving newFc"); curFc->setSize(size); - // Don't record this as a return in order to try and - // determine the "returns" from a GC. - _bt.verify_not_unallocated((HeapWord*) fc, size); - _indexedFreeList[size].returnChunkAtTail(curFc, false); _bt.mark_block((HeapWord*)curFc, size); splitBirth(size); - // Don't record the initial population of the indexed list - // as a split birth. + fc = curFc; + } else { + // Return entire block to caller + fc = newFc; } - - // check that the arithmetic was OK above - assert((HeapWord*)nextFc == (HeapWord*)newFc + replenish_size, - "inconsistency in carving newFc"); - curFc->setSize(size); - _bt.mark_block((HeapWord*)curFc, size); - splitBirth(size); - return curFc; } } } else { @@ -1453,7 +1567,7 @@ // replenish the indexed free list. fc = getChunkFromDictionaryExact(size); } - assert(fc == NULL || fc->isFree(), "Should be returning a free chunk"); + // assert(fc == NULL || fc->isFree(), "Should be returning a free chunk"); return fc; } @@ -1512,6 +1626,11 @@ // adjust _unallocated_block downward, as necessary _bt.freed((HeapWord*)chunk, size); _dictionary->returnChunk(chunk); +#ifndef PRODUCT + if (CMSCollector::abstract_state() != CMSCollector::Sweeping) { + TreeChunk::as_TreeChunk(chunk)->list()->verify_stats(); + } +#endif // PRODUCT } void @@ -1525,6 +1644,11 @@ } else { _indexedFreeList[size].returnChunkAtHead(fc); } +#ifndef PRODUCT + if (CMSCollector::abstract_state() != CMSCollector::Sweeping) { + _indexedFreeList[size].verify_stats(); + } +#endif // PRODUCT } // Add chunk to end of last block -- if it's the largest @@ -1537,7 +1661,6 @@ HeapWord* chunk, size_t size) { // check that the chunk does lie in this space! assert(chunk != NULL && is_in_reserved(chunk), "Not in this space!"); - assert_locked(); // One of the parallel gc task threads may be here // whilst others are allocating. Mutex* lock = NULL; @@ -1991,24 +2114,26 @@ return frag; } -#define CoalSurplusPercent 1.05 -#define SplitSurplusPercent 1.10 - void CompactibleFreeListSpace::beginSweepFLCensus( float inter_sweep_current, - float inter_sweep_estimate) { + float inter_sweep_estimate, + float intra_sweep_estimate) { assert_locked(); size_t i; for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) { FreeList* fl = &_indexedFreeList[i]; - fl->compute_desired(inter_sweep_current, inter_sweep_estimate); - fl->set_coalDesired((ssize_t)((double)fl->desired() * CoalSurplusPercent)); + if (PrintFLSStatistics > 1) { + gclog_or_tty->print("size[%d] : ", i); + } + fl->compute_desired(inter_sweep_current, inter_sweep_estimate, intra_sweep_estimate); + fl->set_coalDesired((ssize_t)((double)fl->desired() * CMSSmallCoalSurplusPercent)); fl->set_beforeSweep(fl->count()); fl->set_bfrSurp(fl->surplus()); } - _dictionary->beginSweepDictCensus(CoalSurplusPercent, + _dictionary->beginSweepDictCensus(CMSLargeCoalSurplusPercent, inter_sweep_current, - inter_sweep_estimate); + inter_sweep_estimate, + intra_sweep_estimate); } void CompactibleFreeListSpace::setFLSurplus() { @@ -2017,7 +2142,7 @@ for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) { FreeList *fl = &_indexedFreeList[i]; fl->set_surplus(fl->count() - - (ssize_t)((double)fl->desired() * SplitSurplusPercent)); + (ssize_t)((double)fl->desired() * CMSSmallSplitSurplusPercent)); } } @@ -2048,6 +2173,11 @@ } void CompactibleFreeListSpace::endSweepFLCensus(size_t sweep_count) { + if (PrintFLSStatistics > 0) { + HeapWord* largestAddr = (HeapWord*) dictionary()->findLargestDict(); + gclog_or_tty->print_cr("CMS: Large block " PTR_FORMAT, + largestAddr); + } setFLSurplus(); setFLHints(); if (PrintGC && PrintFLSCensus > 0) { @@ -2055,7 +2185,7 @@ } clearFLCensus(); assert_locked(); - _dictionary->endSweepDictCensus(SplitSurplusPercent); + _dictionary->endSweepDictCensus(CMSLargeSplitSurplusPercent); } bool CompactibleFreeListSpace::coalOverPopulated(size_t size) { @@ -2312,13 +2442,18 @@ } void CompactibleFreeListSpace::verifyIndexedFreeList(size_t size) const { - FreeChunk* fc = _indexedFreeList[size].head(); + FreeChunk* fc = _indexedFreeList[size].head(); + FreeChunk* tail = _indexedFreeList[size].tail(); + size_t num = _indexedFreeList[size].count(); + size_t n = 0; guarantee((size % 2 == 0) || fc == NULL, "Odd slots should be empty"); - for (; fc != NULL; fc = fc->next()) { + for (; fc != NULL; fc = fc->next(), n++) { guarantee(fc->size() == size, "Size inconsistency"); guarantee(fc->isFree(), "!free?"); guarantee(fc->next() == NULL || fc->next()->prev() == fc, "Broken list"); + guarantee((fc->next() == NULL) == (fc == tail), "Incorrect tail"); } + guarantee(n == num, "Incorrect count"); } #ifndef PRODUCT @@ -2516,11 +2651,41 @@ _tracking = true; } -void PromotionInfo::stopTrackingPromotions() { +#define CMSPrintPromoBlockInfo 1 + +void PromotionInfo::stopTrackingPromotions(uint worker_id) { assert(_spoolHead == _spoolTail && _firstIndex == _nextIndex, "spooling inconsistency?"); _firstIndex = _nextIndex = 1; _tracking = false; + if (CMSPrintPromoBlockInfo > 1) { + print_statistics(worker_id); + } +} + +void PromotionInfo::print_statistics(uint worker_id) const { + assert(_spoolHead == _spoolTail && _firstIndex == _nextIndex, + "Else will undercount"); + assert(CMSPrintPromoBlockInfo > 0, "Else unnecessary call"); + // Count the number of blocks and slots in the free pool + size_t slots = 0; + size_t blocks = 0; + for (SpoolBlock* cur_spool = _spareSpool; + cur_spool != NULL; + cur_spool = cur_spool->nextSpoolBlock) { + // the first entry is just a self-pointer; indices 1 through + // bufferSize - 1 are occupied (thus, bufferSize - 1 slots). + guarantee((void*)cur_spool->displacedHdr == (void*)&cur_spool->displacedHdr, + "first entry of displacedHdr should be self-referential"); + slots += cur_spool->bufferSize - 1; + blocks++; + } + if (_spoolHead != NULL) { + slots += _spoolHead->bufferSize - 1; + blocks++; + } + gclog_or_tty->print_cr(" [worker %d] promo_blocks = %d, promo_slots = %d ", + worker_id, blocks, slots); } // When _spoolTail is not NULL, then the slot <_spoolTail, _nextIndex> @@ -2584,15 +2749,84 @@ guarantee(numDisplacedHdrs == numObjsWithDisplacedHdrs, "Displaced hdr count"); } +void PromotionInfo::print_on(outputStream* st) const { + SpoolBlock* curSpool = NULL; + size_t i = 0; + st->print_cr("start & end indices: [" SIZE_FORMAT ", " SIZE_FORMAT ")", + _firstIndex, _nextIndex); + for (curSpool = _spoolHead; curSpool != _spoolTail && curSpool != NULL; + curSpool = curSpool->nextSpoolBlock) { + curSpool->print_on(st); + st->print_cr(" active "); + i++; + } + for (curSpool = _spoolTail; curSpool != NULL; + curSpool = curSpool->nextSpoolBlock) { + curSpool->print_on(st); + st->print_cr(" inactive "); + i++; + } + for (curSpool = _spareSpool; curSpool != NULL; + curSpool = curSpool->nextSpoolBlock) { + curSpool->print_on(st); + st->print_cr(" free "); + i++; + } + st->print_cr(SIZE_FORMAT " header spooling blocks", i); +} + +void SpoolBlock::print_on(outputStream* st) const { + st->print("[" PTR_FORMAT "," PTR_FORMAT "), " SIZE_FORMAT " HeapWords -> " PTR_FORMAT, + this, (HeapWord*)displacedHdr + bufferSize, + bufferSize, nextSpoolBlock); +} + +/////////////////////////////////////////////////////////////////////////// +// CFLS_LAB +/////////////////////////////////////////////////////////////////////////// + +#define VECTOR_257(x) \ + /* 1 2 3 4 5 6 7 8 9 1x 11 12 13 14 15 16 17 18 19 2x 21 22 23 24 25 26 27 28 29 3x 31 32 */ \ + { x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \ + x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \ + x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \ + x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \ + x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \ + x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \ + x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \ + x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \ + x } + +// Initialize with default setting of CMSParPromoteBlocksToClaim, _not_ +// OldPLABSize, whose static default is different; if overridden at the +// command-line, this will get reinitialized via a call to +// modify_initialization() below. +AdaptiveWeightedAverage CFLS_LAB::_blocks_to_claim[] = + VECTOR_257(AdaptiveWeightedAverage(OldPLABWeight, (float)CMSParPromoteBlocksToClaim)); +size_t CFLS_LAB::_global_num_blocks[] = VECTOR_257(0); +int CFLS_LAB::_global_num_workers[] = VECTOR_257(0); CFLS_LAB::CFLS_LAB(CompactibleFreeListSpace* cfls) : _cfls(cfls) { - _blocks_to_claim = CMSParPromoteBlocksToClaim; + assert(CompactibleFreeListSpace::IndexSetSize == 257, "Modify VECTOR_257() macro above"); for (size_t i = CompactibleFreeListSpace::IndexSetStart; i < CompactibleFreeListSpace::IndexSetSize; i += CompactibleFreeListSpace::IndexSetStride) { _indexedFreeList[i].set_size(i); + _num_blocks[i] = 0; + } +} + +static bool _CFLS_LAB_modified = false; + +void CFLS_LAB::modify_initialization(size_t n, unsigned wt) { + assert(!_CFLS_LAB_modified, "Call only once"); + _CFLS_LAB_modified = true; + for (size_t i = CompactibleFreeListSpace::IndexSetStart; + i < CompactibleFreeListSpace::IndexSetSize; + i += CompactibleFreeListSpace::IndexSetStride) { + _blocks_to_claim[i].modify(n, wt, true /* force */); } } @@ -2607,11 +2841,9 @@ if (res == NULL) return NULL; } else { FreeList* fl = &_indexedFreeList[word_sz]; - bool filled = false; //TRAP if (fl->count() == 0) { - bool filled = true; //TRAP // Attempt to refill this local free list. - _cfls->par_get_chunk_of_blocks(word_sz, _blocks_to_claim, fl); + get_from_global_pool(word_sz, fl); // If it didn't work, give up. if (fl->count() == 0) return NULL; } @@ -2626,80 +2858,190 @@ return (HeapWord*)res; } -void CFLS_LAB::retire() { - for (size_t i = CompactibleFreeListSpace::IndexSetStart; +// Get a chunk of blocks of the right size and update related +// book-keeping stats +void CFLS_LAB::get_from_global_pool(size_t word_sz, FreeList* fl) { + // Get the #blocks we want to claim + size_t n_blks = (size_t)_blocks_to_claim[word_sz].average(); + assert(n_blks > 0, "Error"); + assert(ResizePLAB || n_blks == OldPLABSize, "Error"); + // In some cases, when the application has a phase change, + // there may be a sudden and sharp shift in the object survival + // profile, and updating the counts at the end of a scavenge + // may not be quick enough, giving rise to large scavenge pauses + // during these phase changes. It is beneficial to detect such + // changes on-the-fly during a scavenge and avoid such a phase-change + // pothole. The following code is a heuristic attempt to do that. + // It is protected by a product flag until we have gained + // enough experience with this heuristic and fine-tuned its behaviour. + // WARNING: This might increase fragmentation if we overreact to + // small spikes, so some kind of historical smoothing based on + // previous experience with the greater reactivity might be useful. + // Lacking sufficient experience, CMSOldPLABResizeQuicker is disabled by + // default. + if (ResizeOldPLAB && CMSOldPLABResizeQuicker) { + size_t multiple = _num_blocks[word_sz]/(CMSOldPLABToleranceFactor*CMSOldPLABNumRefills*n_blks); + n_blks += CMSOldPLABReactivityFactor*multiple*n_blks; + n_blks = MIN2(n_blks, CMSOldPLABMax); + } + assert(n_blks > 0, "Error"); + _cfls->par_get_chunk_of_blocks(word_sz, n_blks, fl); + // Update stats table entry for this block size + _num_blocks[word_sz] += fl->count(); +} + +void CFLS_LAB::compute_desired_plab_size() { + for (size_t i = CompactibleFreeListSpace::IndexSetStart; i < CompactibleFreeListSpace::IndexSetSize; i += CompactibleFreeListSpace::IndexSetStride) { - if (_indexedFreeList[i].count() > 0) { - MutexLockerEx x(_cfls->_indexedFreeListParLocks[i], - Mutex::_no_safepoint_check_flag); - _cfls->_indexedFreeList[i].prepend(&_indexedFreeList[i]); - // Reset this list. - _indexedFreeList[i] = FreeList(); - _indexedFreeList[i].set_size(i); + assert((_global_num_workers[i] == 0) == (_global_num_blocks[i] == 0), + "Counter inconsistency"); + if (_global_num_workers[i] > 0) { + // Need to smooth wrt historical average + if (ResizeOldPLAB) { + _blocks_to_claim[i].sample( + MAX2((size_t)CMSOldPLABMin, + MIN2((size_t)CMSOldPLABMax, + _global_num_blocks[i]/(_global_num_workers[i]*CMSOldPLABNumRefills)))); + } + // Reset counters for next round + _global_num_workers[i] = 0; + _global_num_blocks[i] = 0; + if (PrintOldPLAB) { + gclog_or_tty->print_cr("[%d]: %d", i, (size_t)_blocks_to_claim[i].average()); + } } } } -void -CompactibleFreeListSpace:: -par_get_chunk_of_blocks(size_t word_sz, size_t n, FreeList* fl) { +void CFLS_LAB::retire(int tid) { + // We run this single threaded with the world stopped; + // so no need for locks and such. +#define CFLS_LAB_PARALLEL_ACCESS 0 + NOT_PRODUCT(Thread* t = Thread::current();) + assert(Thread::current()->is_VM_thread(), "Error"); + assert(CompactibleFreeListSpace::IndexSetStart == CompactibleFreeListSpace::IndexSetStride, + "Will access to uninitialized slot below"); +#if CFLS_LAB_PARALLEL_ACCESS + for (size_t i = CompactibleFreeListSpace::IndexSetSize - 1; + i > 0; + i -= CompactibleFreeListSpace::IndexSetStride) { +#else // CFLS_LAB_PARALLEL_ACCESS + for (size_t i = CompactibleFreeListSpace::IndexSetStart; + i < CompactibleFreeListSpace::IndexSetSize; + i += CompactibleFreeListSpace::IndexSetStride) { +#endif // !CFLS_LAB_PARALLEL_ACCESS + assert(_num_blocks[i] >= (size_t)_indexedFreeList[i].count(), + "Can't retire more than what we obtained"); + if (_num_blocks[i] > 0) { + size_t num_retire = _indexedFreeList[i].count(); + assert(_num_blocks[i] > num_retire, "Should have used at least one"); + { +#if CFLS_LAB_PARALLEL_ACCESS + MutexLockerEx x(_cfls->_indexedFreeListParLocks[i], + Mutex::_no_safepoint_check_flag); +#endif // CFLS_LAB_PARALLEL_ACCESS + // Update globals stats for num_blocks used + _global_num_blocks[i] += (_num_blocks[i] - num_retire); + _global_num_workers[i]++; + assert(_global_num_workers[i] <= (ssize_t)ParallelGCThreads, "Too big"); + if (num_retire > 0) { + _cfls->_indexedFreeList[i].prepend(&_indexedFreeList[i]); + // Reset this list. + _indexedFreeList[i] = FreeList(); + _indexedFreeList[i].set_size(i); + } + } + if (PrintOldPLAB) { + gclog_or_tty->print_cr("%d[%d]: %d/%d/%d", + tid, i, num_retire, _num_blocks[i], (size_t)_blocks_to_claim[i].average()); + } + // Reset stats for next round + _num_blocks[i] = 0; + } + } +} + +void CompactibleFreeListSpace:: par_get_chunk_of_blocks(size_t word_sz, size_t n, FreeList* fl) { assert(fl->count() == 0, "Precondition."); assert(word_sz < CompactibleFreeListSpace::IndexSetSize, "Precondition"); - // We'll try all multiples of word_sz in the indexed set (starting with - // word_sz itself), then try getting a big chunk and splitting it. - int k = 1; - size_t cur_sz = k * word_sz; - bool found = false; - while (cur_sz < CompactibleFreeListSpace::IndexSetSize && k == 1) { - FreeList* gfl = &_indexedFreeList[cur_sz]; - FreeList fl_for_cur_sz; // Empty. - fl_for_cur_sz.set_size(cur_sz); - { - MutexLockerEx x(_indexedFreeListParLocks[cur_sz], - Mutex::_no_safepoint_check_flag); - if (gfl->count() != 0) { - size_t nn = MAX2(n/k, (size_t)1); - gfl->getFirstNChunksFromList(nn, &fl_for_cur_sz); - found = true; + // We'll try all multiples of word_sz in the indexed set, starting with + // word_sz itself and, if CMSSplitIndexedFreeListBlocks, try larger multiples, + // then try getting a big chunk and splitting it. + { + bool found; + int k; + size_t cur_sz; + for (k = 1, cur_sz = k * word_sz, found = false; + (cur_sz < CompactibleFreeListSpace::IndexSetSize) && + (CMSSplitIndexedFreeListBlocks || k <= 1); + k++, cur_sz = k * word_sz) { + FreeList* gfl = &_indexedFreeList[cur_sz]; + FreeList fl_for_cur_sz; // Empty. + fl_for_cur_sz.set_size(cur_sz); + { + MutexLockerEx x(_indexedFreeListParLocks[cur_sz], + Mutex::_no_safepoint_check_flag); + if (gfl->count() != 0) { + // nn is the number of chunks of size cur_sz that + // we'd need to split k-ways each, in order to create + // "n" chunks of size word_sz each. + const size_t nn = MAX2(n/k, (size_t)1); + gfl->getFirstNChunksFromList(nn, &fl_for_cur_sz); + found = true; + if (k > 1) { + // Update split death stats for the cur_sz-size blocks list: + // we increment the split death count by the number of blocks + // we just took from the cur_sz-size blocks list and which + // we will be splitting below. + ssize_t deaths = _indexedFreeList[cur_sz].splitDeaths() + + fl_for_cur_sz.count(); + _indexedFreeList[cur_sz].set_splitDeaths(deaths); + } + } + } + // Now transfer fl_for_cur_sz to fl. Common case, we hope, is k = 1. + if (found) { + if (k == 1) { + fl->prepend(&fl_for_cur_sz); + } else { + // Divide each block on fl_for_cur_sz up k ways. + FreeChunk* fc; + while ((fc = fl_for_cur_sz.getChunkAtHead()) != NULL) { + // Must do this in reverse order, so that anybody attempting to + // access the main chunk sees it as a single free block until we + // change it. + size_t fc_size = fc->size(); + for (int i = k-1; i >= 0; i--) { + FreeChunk* ffc = (FreeChunk*)((HeapWord*)fc + i * word_sz); + ffc->setSize(word_sz); + ffc->linkNext(NULL); + ffc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads. + // Above must occur before BOT is updated below. + // splitting from the right, fc_size == (k - i + 1) * wordsize + _bt.mark_block((HeapWord*)ffc, word_sz); + fc_size -= word_sz; + _bt.verify_not_unallocated((HeapWord*)ffc, ffc->size()); + _bt.verify_single_block((HeapWord*)fc, fc_size); + _bt.verify_single_block((HeapWord*)ffc, ffc->size()); + // Push this on "fl". + fl->returnChunkAtHead(ffc); + } + // TRAP + assert(fl->tail()->next() == NULL, "List invariant."); + } + } + // Update birth stats for this block size. + size_t num = fl->count(); + MutexLockerEx x(_indexedFreeListParLocks[word_sz], + Mutex::_no_safepoint_check_flag); + ssize_t births = _indexedFreeList[word_sz].splitBirths() + num; + _indexedFreeList[word_sz].set_splitBirths(births); + return; } } - // Now transfer fl_for_cur_sz to fl. Common case, we hope, is k = 1. - if (found) { - if (k == 1) { - fl->prepend(&fl_for_cur_sz); - } else { - // Divide each block on fl_for_cur_sz up k ways. - FreeChunk* fc; - while ((fc = fl_for_cur_sz.getChunkAtHead()) != NULL) { - // Must do this in reverse order, so that anybody attempting to - // access the main chunk sees it as a single free block until we - // change it. - size_t fc_size = fc->size(); - for (int i = k-1; i >= 0; i--) { - FreeChunk* ffc = (FreeChunk*)((HeapWord*)fc + i * word_sz); - ffc->setSize(word_sz); - ffc->linkNext(NULL); - ffc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads. - // Above must occur before BOT is updated below. - // splitting from the right, fc_size == (k - i + 1) * wordsize - _bt.mark_block((HeapWord*)ffc, word_sz); - fc_size -= word_sz; - _bt.verify_not_unallocated((HeapWord*)ffc, ffc->size()); - _bt.verify_single_block((HeapWord*)fc, fc_size); - _bt.verify_single_block((HeapWord*)ffc, ffc->size()); - // Push this on "fl". - fl->returnChunkAtHead(ffc); - } - // TRAP - assert(fl->tail()->next() == NULL, "List invariant."); - } - } - return; - } - k++; cur_sz = k * word_sz; } // Otherwise, we'll split a block from the dictionary. FreeChunk* fc = NULL; @@ -2723,17 +3065,31 @@ } } if (fc == NULL) return; + assert((ssize_t)n >= 1, "Control point invariant"); // Otherwise, split up that block. - size_t nn = fc->size() / word_sz; + const size_t nn = fc->size() / word_sz; n = MIN2(nn, n); + assert((ssize_t)n >= 1, "Control point invariant"); rem = fc->size() - n * word_sz; // If there is a remainder, and it's too small, allocate one fewer. if (rem > 0 && rem < MinChunkSize) { n--; rem += word_sz; } + // Note that at this point we may have n == 0. + assert((ssize_t)n >= 0, "Control point invariant"); + + // If n is 0, the chunk fc that was found is not large + // enough to leave a viable remainder. We are unable to + // allocate even one block. Return fc to the + // dictionary and return, leaving "fl" empty. + if (n == 0) { + returnChunkToDictionary(fc); + return; + } + // First return the remainder, if any. // Note that we hold the lock until we decide if we're going to give - // back the remainder to the dictionary, since a contending allocator + // back the remainder to the dictionary, since a concurrent allocation // may otherwise see the heap as empty. (We're willing to take that // hit if the block is a small block.) if (rem > 0) { @@ -2743,18 +3099,16 @@ rem_fc->linkNext(NULL); rem_fc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads. // Above must occur before BOT is updated below. + assert((ssize_t)n > 0 && prefix_size > 0 && rem_fc > fc, "Error"); _bt.split_block((HeapWord*)fc, fc->size(), prefix_size); if (rem >= IndexSetSize) { returnChunkToDictionary(rem_fc); - dictionary()->dictCensusUpdate(fc->size(), - true /*split*/, - true /*birth*/); + dictionary()->dictCensusUpdate(rem, true /*split*/, true /*birth*/); rem_fc = NULL; } // Otherwise, return it to the small list below. } } - // if (rem_fc != NULL) { MutexLockerEx x(_indexedFreeListParLocks[rem], Mutex::_no_safepoint_check_flag); @@ -2762,7 +3116,7 @@ _indexedFreeList[rem].returnChunkAtHead(rem_fc); smallSplitBirth(rem); } - + assert((ssize_t)n > 0 && fc != NULL, "Consistency"); // Now do the splitting up. // Must do this in reverse order, so that anybody attempting to // access the main chunk sees it as a single free block until we @@ -2792,13 +3146,15 @@ _bt.verify_single_block((HeapWord*)fc, fc->size()); fl->returnChunkAtHead(fc); + assert((ssize_t)n > 0 && (ssize_t)n == fl->count(), "Incorrect number of blocks"); { + // Update the stats for this block size. MutexLockerEx x(_indexedFreeListParLocks[word_sz], Mutex::_no_safepoint_check_flag); - ssize_t new_births = _indexedFreeList[word_sz].splitBirths() + n; - _indexedFreeList[word_sz].set_splitBirths(new_births); - ssize_t new_surplus = _indexedFreeList[word_sz].surplus() + n; - _indexedFreeList[word_sz].set_surplus(new_surplus); + const ssize_t births = _indexedFreeList[word_sz].splitBirths() + n; + _indexedFreeList[word_sz].set_splitBirths(births); + // ssize_t new_surplus = _indexedFreeList[word_sz].surplus() + n; + // _indexedFreeList[word_sz].set_surplus(new_surplus); } // TRAP
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -25,8 +25,6 @@ // Classes in support of keeping track of promotions into a non-Contiguous // space, in this case a CompactibleFreeListSpace. -#define CFLS_LAB_REFILL_STATS 0 - // Forward declarations class CompactibleFreeListSpace; class BlkClosure; @@ -89,6 +87,9 @@ displacedHdr = (markOop*)&displacedHdr; nextSpoolBlock = NULL; } + + void print_on(outputStream* st) const; + void print() const { print_on(gclog_or_tty); } }; class PromotionInfo VALUE_OBJ_CLASS_SPEC { @@ -121,7 +122,7 @@ return _promoHead == NULL; } void startTrackingPromotions(); - void stopTrackingPromotions(); + void stopTrackingPromotions(uint worker_id = 0); bool tracking() const { return _tracking; } void track(PromotedObject* trackOop); // keep track of a promoted oop // The following variant must be used when trackOop is not fully @@ -161,6 +162,9 @@ _nextIndex = 0; } + + void print_on(outputStream* st) const; + void print_statistics(uint worker_id) const; }; class LinearAllocBlock VALUE_OBJ_CLASS_SPEC { @@ -243,6 +247,7 @@ mutable Mutex _freelistLock; // locking verifier convenience function void assert_locked() const PRODUCT_RETURN; + void assert_locked(const Mutex* lock) const PRODUCT_RETURN; // Linear allocation blocks LinearAllocBlock _smallLinearAllocBlock; @@ -281,13 +286,6 @@ // Locks protecting the exact lists during par promotion allocation. Mutex* _indexedFreeListParLocks[IndexSetSize]; -#if CFLS_LAB_REFILL_STATS - // Some statistics. - jint _par_get_chunk_from_small; - jint _par_get_chunk_from_large; -#endif - - // Attempt to obtain up to "n" blocks of the size "word_sz" (which is // required to be smaller than "IndexSetSize".) If successful, // adds them to "fl", which is required to be an empty free list. @@ -320,7 +318,7 @@ // Helper function for getChunkFromIndexedFreeList. // Replenish the indexed free list for this "size". Do not take from an // underpopulated size. - FreeChunk* getChunkFromIndexedFreeListHelper(size_t size); + FreeChunk* getChunkFromIndexedFreeListHelper(size_t size, bool replenish = true); // Get a chunk from the indexed free list. If the indexed free list // does not have a free chunk, try to replenish the indexed free list @@ -430,10 +428,6 @@ void initialize_sequential_subtasks_for_marking(int n_threads, HeapWord* low = NULL); -#if CFLS_LAB_REFILL_STATS - void print_par_alloc_stats(); -#endif - // Space enquiries size_t used() const; size_t free() const; @@ -617,6 +611,12 @@ // Do some basic checks on the the free lists. void checkFreeListConsistency() const PRODUCT_RETURN; + // Printing support + void dump_at_safepoint_with_locks(CMSCollector* c, outputStream* st); + void print_indexed_free_lists(outputStream* st) const; + void print_dictionary_free_lists(outputStream* st) const; + void print_promo_info_blocks(outputStream* st) const; + NOT_PRODUCT ( void initializeIndexedFreeListArrayReturnedBytes(); size_t sumIndexedFreeListArrayReturnedBytes(); @@ -638,8 +638,9 @@ // Statistics functions // Initialize census for lists before the sweep. - void beginSweepFLCensus(float sweep_current, - float sweep_estimate); + void beginSweepFLCensus(float inter_sweep_current, + float inter_sweep_estimate, + float intra_sweep_estimate); // Set the surplus for each of the free lists. void setFLSurplus(); // Set the hint for each of the free lists. @@ -730,16 +731,17 @@ FreeList _indexedFreeList[CompactibleFreeListSpace::IndexSetSize]; // Initialized from a command-line arg. - size_t _blocks_to_claim; -#if CFLS_LAB_REFILL_STATS - // Some statistics. - int _refills; - int _blocksTaken; - static int _tot_refills; - static int _tot_blocksTaken; - static int _next_threshold; -#endif + // Allocation statistics in support of dynamic adjustment of + // #blocks to claim per get_from_global_pool() call below. + static AdaptiveWeightedAverage + _blocks_to_claim [CompactibleFreeListSpace::IndexSetSize]; + static size_t _global_num_blocks [CompactibleFreeListSpace::IndexSetSize]; + static int _global_num_workers[CompactibleFreeListSpace::IndexSetSize]; + size_t _num_blocks [CompactibleFreeListSpace::IndexSetSize]; + + // Internal work method + void get_from_global_pool(size_t word_sz, FreeList* fl); public: CFLS_LAB(CompactibleFreeListSpace* cfls); @@ -748,7 +750,12 @@ HeapWord* alloc(size_t word_sz); // Return any unused portions of the buffer to the global pool. - void retire(); + void retire(int tid); + + // Dynamic OldPLABSize sizing + static void compute_desired_plab_size(); + // When the settings are modified from default static initialization + static void modify_initialization(size_t n, unsigned wt); }; size_t PromotionInfo::refillSize() const {
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -253,7 +253,6 @@ } } - void ConcurrentMarkSweepGeneration::ref_processor_init() { assert(collector() != NULL, "no collector"); collector()->ref_processor_init(); @@ -341,6 +340,14 @@ _icms_duty_cycle = CMSIncrementalDutyCycle; } +double CMSStats::cms_free_adjustment_factor(size_t free) const { + // TBD: CR 6909490 + return 1.0; +} + +void CMSStats::adjust_cms_free_adjustment_factor(bool fail, size_t free) { +} + // If promotion failure handling is on use // the padded average size of the promotion for each // young generation collection. @@ -361,7 +368,11 @@ // Adjust by the safety factor. double cms_free_dbl = (double)cms_free; - cms_free_dbl = cms_free_dbl * (100.0 - CMSIncrementalSafetyFactor) / 100.0; + double cms_adjustment = (100.0 - CMSIncrementalSafetyFactor)/100.0; + // Apply a further correction factor which tries to adjust + // for recent occurance of concurrent mode failures. + cms_adjustment = cms_adjustment * cms_free_adjustment_factor(cms_free); + cms_free_dbl = cms_free_dbl * cms_adjustment; if (PrintGCDetails && Verbose) { gclog_or_tty->print_cr("CMSStats::time_until_cms_gen_full: cms_free " @@ -395,6 +406,8 @@ // late. double work = cms_duration() + gc0_period(); double deadline = time_until_cms_gen_full(); + // If a concurrent mode failure occurred recently, we want to be + // more conservative and halve our expected time_until_cms_gen_full() if (work > deadline) { if (Verbose && PrintGCDetails) { gclog_or_tty->print( @@ -556,7 +569,8 @@ _should_unload_classes(false), _concurrent_cycles_since_last_unload(0), _roots_scanning_options(0), - _sweep_estimate(CMS_SweepWeight, CMS_SweepPadding) + _inter_sweep_estimate(CMS_SweepWeight, CMS_SweepPadding), + _intra_sweep_estimate(CMS_SweepWeight, CMS_SweepPadding) { if (ExplicitGCInvokesConcurrentAndUnloadsClasses) { ExplicitGCInvokesConcurrent = true; @@ -709,7 +723,8 @@ // Support for parallelizing survivor space rescan if (CMSParallelRemarkEnabled && CMSParallelSurvivorRemarkEnabled) { - size_t max_plab_samples = MaxNewSize/((SurvivorRatio+2)*MinTLABSize); + size_t max_plab_samples = cp->max_gen0_size()/ + ((SurvivorRatio+2)*MinTLABSize); _survivor_plab_array = NEW_C_HEAP_ARRAY(ChunkArray, ParallelGCThreads); _survivor_chunk_array = NEW_C_HEAP_ARRAY(HeapWord*, 2*max_plab_samples); _cursor = NEW_C_HEAP_ARRAY(size_t, ParallelGCThreads); @@ -772,7 +787,7 @@ NOT_PRODUCT(_overflow_counter = CMSMarkStackOverflowInterval;) _gc_counters = new CollectorCounters("CMS", 1); _completed_initialization = true; - _sweep_timer.start(); // start of time + _inter_sweep_timer.start(); // start of time } const char* ConcurrentMarkSweepGeneration::name() const { @@ -899,6 +914,14 @@ return result; } +// At a promotion failure dump information on block layout in heap +// (cms old generation). +void ConcurrentMarkSweepGeneration::promotion_failure_occurred() { + if (CMSDumpAtPromotionFailure) { + cmsSpace()->dump_at_safepoint_with_locks(collector(), gclog_or_tty); + } +} + CompactibleSpace* ConcurrentMarkSweepGeneration::first_compaction_space() const { return _cmsSpace; @@ -1367,12 +1390,7 @@ ConcurrentMarkSweepGeneration:: par_promote_alloc_done(int thread_num) { CMSParGCThreadState* ps = _par_gc_thread_states[thread_num]; - ps->lab.retire(); -#if CFLS_LAB_REFILL_STATS - if (thread_num == 0) { - _cmsSpace->print_par_alloc_stats(); - } -#endif + ps->lab.retire(thread_num); } void @@ -1973,11 +1991,14 @@ // We must adjust the allocation statistics being maintained // in the free list space. We do so by reading and clearing // the sweep timer and updating the block flux rate estimates below. - assert(_sweep_timer.is_active(), "We should never see the timer inactive"); - _sweep_timer.stop(); - // Note that we do not use this sample to update the _sweep_estimate. - _cmsGen->cmsSpace()->beginSweepFLCensus((float)(_sweep_timer.seconds()), - _sweep_estimate.padded_average()); + assert(!_intra_sweep_timer.is_active(), "_intra_sweep_timer should be inactive"); + if (_inter_sweep_timer.is_active()) { + _inter_sweep_timer.stop(); + // Note that we do not use this sample to update the _inter_sweep_estimate. + _cmsGen->cmsSpace()->beginSweepFLCensus((float)(_inter_sweep_timer.seconds()), + _inter_sweep_estimate.padded_average(), + _intra_sweep_estimate.padded_average()); + } GenMarkSweep::invoke_at_safepoint(_cmsGen->level(), ref_processor(), clear_all_soft_refs); @@ -2014,10 +2035,10 @@ } // Adjust the per-size allocation stats for the next epoch. - _cmsGen->cmsSpace()->endSweepFLCensus(sweepCount() /* fake */); - // Restart the "sweep timer" for next epoch. - _sweep_timer.reset(); - _sweep_timer.start(); + _cmsGen->cmsSpace()->endSweepFLCensus(sweep_count() /* fake */); + // Restart the "inter sweep timer" for the next epoch. + _inter_sweep_timer.reset(); + _inter_sweep_timer.start(); // Sample collection pause time and reset for collection interval. if (UseAdaptiveSizePolicy) { @@ -2675,7 +2696,7 @@ // Also reset promotion tracking in par gc thread states. if (ParallelGCThreads > 0) { for (uint i = 0; i < ParallelGCThreads; i++) { - _par_gc_thread_states[i]->promo.stopTrackingPromotions(); + _par_gc_thread_states[i]->promo.stopTrackingPromotions(i); } } } @@ -2770,7 +2791,7 @@ bool do_bit(size_t offset) { HeapWord* addr = _marks->offsetToHeapWord(offset); if (!_marks->isMarked(addr)) { - oop(addr)->print(); + oop(addr)->print_on(gclog_or_tty); gclog_or_tty->print_cr(" ("INTPTR_FORMAT" should have been marked)", addr); _failed = true; } @@ -2819,7 +2840,7 @@ // Clear any marks from a previous round verification_mark_bm()->clear_all(); assert(verification_mark_stack()->isEmpty(), "markStack should be empty"); - assert(overflow_list_is_empty(), "overflow list should be empty"); + verify_work_stacks_empty(); GenCollectedHeap* gch = GenCollectedHeap::heap(); gch->ensure_parsability(false); // fill TLABs, but no need to retire them @@ -2892,8 +2913,8 @@ verification_mark_bm()->iterate(&vcl); if (vcl.failed()) { gclog_or_tty->print("Verification failed"); - Universe::heap()->print(); - fatal(" ... aborting"); + Universe::heap()->print_on(gclog_or_tty); + fatal("CMS: failed marking verification after remark"); } } @@ -3313,7 +3334,7 @@ Universe::heap()->barrier_set()->resize_covered_region(mr); // Hmmmm... why doesn't CFLS::set_end verify locking? // This is quite ugly; FIX ME XXX - _cmsSpace->assert_locked(); + _cmsSpace->assert_locked(freelistLock()); _cmsSpace->set_end((HeapWord*)_virtual_space.high()); // update the space and generation capacity counters @@ -3634,9 +3655,7 @@ verify_work_stacks_empty(); verify_overflow_empty(); assert(_revisitStack.isEmpty(), "tabula rasa"); - - DEBUG_ONLY(RememberKlassesChecker cmx(CMSClassUnloadingEnabled);) - + DEBUG_ONLY(RememberKlassesChecker cmx(should_unload_classes());) bool result = false; if (CMSConcurrentMTEnabled && ParallelCMSThreads > 0) { result = do_marking_mt(asynch); @@ -4103,7 +4122,6 @@ void CMSConcMarkingTask::coordinator_yield() { assert(ConcurrentMarkSweepThread::cms_thread_has_cms_token(), "CMS thread should hold CMS token"); - DEBUG_ONLY(RememberKlassesChecker mux(false);) // First give up the locks, then yield, then re-lock // We should probably use a constructor/destructor idiom to @@ -4180,9 +4198,7 @@ // Mutate the Refs discovery so it is MT during the // multi-threaded marking phase. ReferenceProcessorMTMutator mt(ref_processor(), num_workers > 1); - - DEBUG_ONLY(RememberKlassesChecker cmx(CMSClassUnloadingEnabled);) - + DEBUG_ONLY(RememberKlassesChecker cmx(should_unload_classes());) conc_workers()->start_task(&tsk); while (tsk.yielded()) { tsk.coordinator_yield(); @@ -4451,7 +4467,7 @@ // for cleaner interfaces. rp->preclean_discovered_references( rp->is_alive_non_header(), &keep_alive, &complete_trace, - &yield_cl); + &yield_cl, should_unload_classes()); } if (clean_survivor) { // preclean the active survivor space(s) @@ -4473,7 +4489,7 @@ SurvivorSpacePrecleanClosure sss_cl(this, _span, &_markBitMap, &_markStack, &pam_cl, before_count, CMSYield); - DEBUG_ONLY(RememberKlassesChecker mx(CMSClassUnloadingEnabled);) + DEBUG_ONLY(RememberKlassesChecker mx(should_unload_classes());) dng->from()->object_iterate_careful(&sss_cl); dng->to()->object_iterate_careful(&sss_cl); } @@ -4644,7 +4660,7 @@ verify_work_stacks_empty(); verify_overflow_empty(); sample_eden(); - DEBUG_ONLY(RememberKlassesChecker mx(CMSClassUnloadingEnabled);) + DEBUG_ONLY(RememberKlassesChecker mx(should_unload_classes());) stop_point = gen->cmsSpace()->object_iterate_careful_m(dirtyRegion, cl); } @@ -4732,7 +4748,7 @@ sample_eden(); verify_work_stacks_empty(); verify_overflow_empty(); - DEBUG_ONLY(RememberKlassesChecker mx(CMSClassUnloadingEnabled);) + DEBUG_ONLY(RememberKlassesChecker mx(should_unload_classes());) HeapWord* stop_point = gen->cmsSpace()->object_iterate_careful_m(dirtyRegion, cl); if (stop_point != NULL) { @@ -4832,7 +4848,7 @@ assert(haveFreelistLocks(), "must have free list locks"); assert_lock_strong(bitMapLock()); - DEBUG_ONLY(RememberKlassesChecker fmx(CMSClassUnloadingEnabled);) + DEBUG_ONLY(RememberKlassesChecker fmx(should_unload_classes());) if (!init_mark_was_synchronous) { // We might assume that we need not fill TLAB's when // CMSScavengeBeforeRemark is set, because we may have just done @@ -5867,9 +5883,9 @@ check_correct_thread_executing(); verify_work_stacks_empty(); verify_overflow_empty(); - incrementSweepCount(); - _sweep_timer.stop(); - _sweep_estimate.sample(_sweep_timer.seconds()); + increment_sweep_count(); + _inter_sweep_timer.stop(); + _inter_sweep_estimate.sample(_inter_sweep_timer.seconds()); size_policy()->avg_cms_free_at_sweep()->sample(_cmsGen->free()); // PermGen verification support: If perm gen sweeping is disabled in @@ -5892,6 +5908,9 @@ } } + assert(!_intra_sweep_timer.is_active(), "Should not be active"); + _intra_sweep_timer.reset(); + _intra_sweep_timer.start(); if (asynch) { TraceCPUTime tcpu(PrintGCDetails, true, gclog_or_tty); CMSPhaseAccounting pa(this, "sweep", !PrintGCDetails); @@ -5936,8 +5955,11 @@ verify_work_stacks_empty(); verify_overflow_empty(); - _sweep_timer.reset(); - _sweep_timer.start(); + _intra_sweep_timer.stop(); + _intra_sweep_estimate.sample(_intra_sweep_timer.seconds()); + + _inter_sweep_timer.reset(); + _inter_sweep_timer.start(); update_time_of_last_gc(os::javaTimeMillis()); @@ -5980,11 +6002,11 @@ // FIX ME!!! Looks like this belongs in CFLSpace, with // CMSGen merely delegating to it. void ConcurrentMarkSweepGeneration::setNearLargestChunk() { - double nearLargestPercent = 0.999; + double nearLargestPercent = FLSLargestBlockCoalesceProximity; HeapWord* minAddr = _cmsSpace->bottom(); HeapWord* largestAddr = (HeapWord*) _cmsSpace->dictionary()->findLargestDict(); - if (largestAddr == 0) { + if (largestAddr == NULL) { // The dictionary appears to be empty. In this case // try to coalesce at the end of the heap. largestAddr = _cmsSpace->end(); @@ -5992,6 +6014,13 @@ size_t largestOffset = pointer_delta(largestAddr, minAddr); size_t nearLargestOffset = (size_t)((double)largestOffset * nearLargestPercent) - MinChunkSize; + if (PrintFLSStatistics != 0) { + gclog_or_tty->print_cr( + "CMS: Large Block: " PTR_FORMAT ";" + " Proximity: " PTR_FORMAT " -> " PTR_FORMAT, + largestAddr, + _cmsSpace->nearLargestChunk(), minAddr + nearLargestOffset); + } _cmsSpace->set_nearLargestChunk(minAddr + nearLargestOffset); } @@ -6071,9 +6100,11 @@ assert_lock_strong(gen->freelistLock()); assert_lock_strong(bitMapLock()); - assert(!_sweep_timer.is_active(), "Was switched off in an outer context"); - gen->cmsSpace()->beginSweepFLCensus((float)(_sweep_timer.seconds()), - _sweep_estimate.padded_average()); + assert(!_inter_sweep_timer.is_active(), "Was switched off in an outer context"); + assert(_intra_sweep_timer.is_active(), "Was switched on in an outer context"); + gen->cmsSpace()->beginSweepFLCensus((float)(_inter_sweep_timer.seconds()), + _inter_sweep_estimate.padded_average(), + _intra_sweep_estimate.padded_average()); gen->setNearLargestChunk(); { @@ -6086,7 +6117,7 @@ // end-of-sweep-census below will be off by a little bit. } gen->cmsSpace()->sweep_completed(); - gen->cmsSpace()->endSweepFLCensus(sweepCount()); + gen->cmsSpace()->endSweepFLCensus(sweep_count()); if (should_unload_classes()) { // unloaded classes this cycle, _concurrent_cycles_since_last_unload = 0; // ... reset count } else { // did not unload classes,
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -355,6 +355,11 @@ unsigned int new_duty_cycle); unsigned int icms_update_duty_cycle_impl(); + // In support of adjusting of cms trigger ratios based on history + // of concurrent mode failure. + double cms_free_adjustment_factor(size_t free) const; + void adjust_cms_free_adjustment_factor(bool fail, size_t free); + public: CMSStats(ConcurrentMarkSweepGeneration* cms_gen, unsigned int alpha = CMSExpAvgFactor); @@ -570,8 +575,11 @@ // appropriately. void check_gc_time_limit(); // XXX Move these to CMSStats ??? FIX ME !!! - elapsedTimer _sweep_timer; - AdaptivePaddedAverage _sweep_estimate; + elapsedTimer _inter_sweep_timer; // time between sweeps + elapsedTimer _intra_sweep_timer; // time _in_ sweeps + // padded decaying average estimates of the above + AdaptivePaddedAverage _inter_sweep_estimate; + AdaptivePaddedAverage _intra_sweep_estimate; protected: ConcurrentMarkSweepGeneration* _cmsGen; // old gen (CMS) @@ -625,6 +633,7 @@ // . _collectorState <= Idling == post-sweep && pre-mark // . _collectorState in (Idling, Sweeping) == {initial,final}marking || // precleaning || abortablePrecleanb + public: enum CollectorState { Resizing = 0, Resetting = 1, @@ -636,6 +645,7 @@ FinalMarking = 7, Sweeping = 8 }; + protected: static CollectorState _collectorState; // State related to prologue/epilogue invocation for my generations @@ -655,7 +665,7 @@ int _numYields; size_t _numDirtyCards; - uint _sweepCount; + size_t _sweep_count; // number of full gc's since the last concurrent gc. uint _full_gcs_since_conc_gc; @@ -905,7 +915,7 @@ // Check that the currently executing thread is the expected // one (foreground collector or background collector). - void check_correct_thread_executing() PRODUCT_RETURN; + static void check_correct_thread_executing() PRODUCT_RETURN; // XXXPERM void print_statistics() PRODUCT_RETURN; bool is_cms_reachable(HeapWord* addr); @@ -930,8 +940,8 @@ static void set_foregroundGCShouldWait(bool v) { _foregroundGCShouldWait = v; } static bool foregroundGCIsActive() { return _foregroundGCIsActive; } static void set_foregroundGCIsActive(bool v) { _foregroundGCIsActive = v; } - uint sweepCount() const { return _sweepCount; } - void incrementSweepCount() { _sweepCount++; } + size_t sweep_count() const { return _sweep_count; } + void increment_sweep_count() { _sweep_count++; } // Timers/stats for gc scheduling and incremental mode pacing. CMSStats& stats() { return _stats; } @@ -1165,6 +1175,11 @@ virtual bool promotion_attempt_is_safe(size_t promotion_in_bytes, bool younger_handles_promotion_failure) const; + // Inform this (non-young) generation that a promotion failure was + // encountered during a collection of a younger generation that + // promotes into this generation. + virtual void promotion_failure_occurred(); + bool should_collect(bool full, size_t size, bool tlab); virtual bool should_concurrent_collect() const; virtual bool is_too_full() const;
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/freeBlockDictionary.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/freeBlockDictionary.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -55,7 +55,8 @@ virtual void dictCensusUpdate(size_t size, bool split, bool birth) = 0; virtual bool coalDictOverPopulated(size_t size) = 0; virtual void beginSweepDictCensus(double coalSurplusPercent, - float sweep_current, float sweep_ewstimate) = 0; + float inter_sweep_current, float inter_sweep_estimate, + float intra__sweep_current) = 0; virtual void endSweepDictCensus(double splitSurplusPercent) = 0; virtual FreeChunk* findLargestDict() const = 0; // verify that the given chunk is in the dictionary. @@ -79,6 +80,7 @@ } virtual void printDictCensus() const = 0; + virtual void print_free_lists(outputStream* st) const = 0; virtual void verify() const = 0;
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/freeChunk.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/freeChunk.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -67,3 +67,8 @@ } } #endif + +void FreeChunk::print_on(outputStream* st) { + st->print_cr("Next: " PTR_FORMAT " Prev: " PTR_FORMAT " %s", + next(), prev(), cantCoalesce() ? "[can't coalesce]" : ""); +}
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/freeChunk.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/freeChunk.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -129,6 +129,8 @@ void verifyList() const PRODUCT_RETURN; void mangleAllocated(size_t size) PRODUCT_RETURN; void mangleFreed(size_t size) PRODUCT_RETURN; + + void print_on(outputStream* st); }; // Alignment helpers etc.
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/freeList.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/freeList.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -81,8 +81,8 @@ set_hint(hint); } -void FreeList::init_statistics() { - _allocation_stats.initialize(); +void FreeList::init_statistics(bool split_birth) { + _allocation_stats.initialize(split_birth); } FreeChunk* FreeList::getChunkAtHead() { @@ -292,14 +292,31 @@ } #ifndef PRODUCT +void FreeList::verify_stats() const { + // The +1 of the LH comparand is to allow some "looseness" in + // checking: we usually call this interface when adding a block + // and we'll subsequently update the stats; we cannot update the + // stats beforehand because in the case of the large-block BT + // dictionary for example, this might be the first block and + // in that case there would be no place that we could record + // the stats (which are kept in the block itself). + assert(_allocation_stats.prevSweep() + _allocation_stats.splitBirths() + 1 // Total Stock + 1 + >= _allocation_stats.splitDeaths() + (ssize_t)count(), "Conservation Principle"); +} + void FreeList::assert_proper_lock_protection_work() const { -#ifdef ASSERT - if (_protecting_lock != NULL && - SharedHeap::heap()->n_par_threads() > 0) { - // Should become an assert. - guarantee(_protecting_lock->owned_by_self(), "FreeList RACE DETECTED"); + assert(_protecting_lock != NULL, "Don't call this directly"); + assert(ParallelGCThreads > 0, "Don't call this directly"); + Thread* thr = Thread::current(); + if (thr->is_VM_thread() || thr->is_ConcurrentGC_thread()) { + // assert that we are holding the freelist lock + } else if (thr->is_GC_task_thread()) { + assert(_protecting_lock->owned_by_self(), "FreeList RACE DETECTED"); + } else if (thr->is_Java_thread()) { + assert(!SafepointSynchronize::is_at_safepoint(), "Should not be executing"); + } else { + ShouldNotReachHere(); // unaccounted thread type? } -#endif } #endif
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/freeList.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/concurrentMarkSweep/freeList.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -35,18 +35,26 @@ // for that implementation. class Mutex; +class TreeList; class FreeList VALUE_OBJ_CLASS_SPEC { friend class CompactibleFreeListSpace; friend class VMStructs; - friend class printTreeCensusClosure; - FreeChunk* _head; // List of free chunks + friend class PrintTreeCensusClosure; + + protected: + TreeList* _parent; + TreeList* _left; + TreeList* _right; + + private: + FreeChunk* _head; // Head of list of free chunks FreeChunk* _tail; // Tail of list of free chunks - size_t _size; // Size in Heap words of each chunks + size_t _size; // Size in Heap words of each chunk ssize_t _count; // Number of entries in list size_t _hint; // next larger size list with a positive surplus - AllocationStats _allocation_stats; // statistics for smart allocation + AllocationStats _allocation_stats; // allocation-related statistics #ifdef ASSERT Mutex* _protecting_lock; @@ -63,9 +71,12 @@ // Initialize the allocation statistics. protected: - void init_statistics(); + void init_statistics(bool split_birth = false); void set_count(ssize_t v) { _count = v;} - void increment_count() { _count++; } + void increment_count() { + _count++; + } + void decrement_count() { _count--; assert(_count >= 0, "Count should not be negative"); @@ -167,11 +178,13 @@ _allocation_stats.set_desired(v); } void compute_desired(float inter_sweep_current, - float inter_sweep_estimate) { + float inter_sweep_estimate, + float intra_sweep_estimate) { assert_proper_lock_protection(); _allocation_stats.compute_desired(_count, inter_sweep_current, - inter_sweep_estimate); + inter_sweep_estimate, + intra_sweep_estimate); } ssize_t coalDesired() const { return _allocation_stats.coalDesired(); @@ -306,6 +319,9 @@ // found. Return NULL if "fc" is not found. bool verifyChunkInFreeLists(FreeChunk* fc) const; + // Stats verification + void verify_stats() const PRODUCT_RETURN; + // Printing support static void print_labels_on(outputStream* st, const char* c); void print_on(outputStream* st, const char* c = NULL) const;
--- a/src/share/vm/gc_implementation/g1/collectionSetChooser.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/collectionSetChooser.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -351,9 +351,16 @@ CollectionSetChooser::printSortedHeapRegions() { gclog_or_tty->print_cr("Printing %d Heap Regions sorted by amount of known garbage", _numMarkedRegions); + + DEBUG_ONLY(int marked_count = 0;) for (int i = 0; i < _markedRegions.length(); i++) { - printHeapRegion(_markedRegions.at(i)); + HeapRegion* r = _markedRegions.at(i); + if (r != NULL) { + printHeapRegion(r); + DEBUG_ONLY(marked_count++;) + } } + assert(marked_count == _numMarkedRegions, "must be"); gclog_or_tty->print_cr("Done sorted heap region print"); }
--- a/src/share/vm/gc_implementation/g1/concurrentG1Refine.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/concurrentG1Refine.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -42,28 +42,49 @@ _n_periods(0), _threads(NULL), _n_threads(0) { - if (G1ConcRefine) { - _n_threads = (int)thread_num(); - if (_n_threads > 0) { - _threads = NEW_C_HEAP_ARRAY(ConcurrentG1RefineThread*, _n_threads); - int worker_id_offset = (int)DirtyCardQueueSet::num_par_ids(); - ConcurrentG1RefineThread *next = NULL; - for (int i = _n_threads - 1; i >= 0; i--) { - ConcurrentG1RefineThread* t = new ConcurrentG1RefineThread(this, next, worker_id_offset, i); - assert(t != NULL, "Conc refine should have been created"); - assert(t->cg1r() == this, "Conc refine thread should refer to this"); - _threads[i] = t; - next = t; - } - } + + // Ergomonically select initial concurrent refinement parameters + if (FLAG_IS_DEFAULT(G1ConcRefineGreenZone)) { + FLAG_SET_DEFAULT(G1ConcRefineGreenZone, MAX2<int>(ParallelGCThreads, 1)); + } + set_green_zone(G1ConcRefineGreenZone); + + if (FLAG_IS_DEFAULT(G1ConcRefineYellowZone)) { + FLAG_SET_DEFAULT(G1ConcRefineYellowZone, green_zone() * 3); + } + set_yellow_zone(MAX2<int>(G1ConcRefineYellowZone, green_zone())); + + if (FLAG_IS_DEFAULT(G1ConcRefineRedZone)) { + FLAG_SET_DEFAULT(G1ConcRefineRedZone, yellow_zone() * 2); + } + set_red_zone(MAX2<int>(G1ConcRefineRedZone, yellow_zone())); + _n_worker_threads = thread_num(); + // We need one extra thread to do the young gen rset size sampling. + _n_threads = _n_worker_threads + 1; + reset_threshold_step(); + + _threads = NEW_C_HEAP_ARRAY(ConcurrentG1RefineThread*, _n_threads); + int worker_id_offset = (int)DirtyCardQueueSet::num_par_ids(); + ConcurrentG1RefineThread *next = NULL; + for (int i = _n_threads - 1; i >= 0; i--) { + ConcurrentG1RefineThread* t = new ConcurrentG1RefineThread(this, next, worker_id_offset, i); + assert(t != NULL, "Conc refine should have been created"); + assert(t->cg1r() == this, "Conc refine thread should refer to this"); + _threads[i] = t; + next = t; } } -size_t ConcurrentG1Refine::thread_num() { - if (G1ConcRefine) { - return (G1ParallelRSetThreads > 0) ? G1ParallelRSetThreads : ParallelGCThreads; +void ConcurrentG1Refine::reset_threshold_step() { + if (FLAG_IS_DEFAULT(G1ConcRefineThresholdStep)) { + _thread_threshold_step = (yellow_zone() - green_zone()) / (worker_thread_num() + 1); + } else { + _thread_threshold_step = G1ConcRefineThresholdStep; } - return 0; +} + +int ConcurrentG1Refine::thread_num() { + return MAX2<int>((G1ParallelRSetThreads > 0) ? G1ParallelRSetThreads : ParallelGCThreads, 1); } void ConcurrentG1Refine::init() { @@ -123,6 +144,15 @@ } } +void ConcurrentG1Refine::reinitialize_threads() { + reset_threshold_step(); + if (_threads != NULL) { + for (int i = 0; i < _n_threads; i++) { + _threads[i]->initialize(); + } + } +} + ConcurrentG1Refine::~ConcurrentG1Refine() { if (G1ConcRSLogCacheSize > 0) { assert(_card_counts != NULL, "Logic"); @@ -270,7 +300,23 @@ int count; jbyte* cached_ptr = add_card_count(card_ptr, &count, defer); assert(cached_ptr != NULL, "bad cached card ptr"); - assert(!is_young_card(cached_ptr), "shouldn't get a card in young region"); + + if (is_young_card(cached_ptr)) { + // The region containing cached_ptr has been freed during a clean up + // pause, reallocated, and tagged as young. + assert(cached_ptr != card_ptr, "shouldn't be"); + + // We've just inserted a new old-gen card pointer into the card count + // cache and evicted the previous contents of that count slot. + // The evicted card pointer has been determined to be in a young region + // and so cannot be the newly inserted card pointer (that will be + // in an old region). + // The count for newly inserted card will be set to zero during the + // insertion, so we don't want to defer the cleaning of the newly + // inserted card pointer. + assert(*defer == false, "deferring non-hot card"); + return NULL; + } // The card pointer we obtained from card count cache is not hot // so do not store it in the cache; return it for immediate @@ -384,4 +430,3 @@ st->cr(); } } -
--- a/src/share/vm/gc_implementation/g1/concurrentG1Refine.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/concurrentG1Refine.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -29,6 +29,31 @@ class ConcurrentG1Refine: public CHeapObj { ConcurrentG1RefineThread** _threads; int _n_threads; + int _n_worker_threads; + /* + * The value of the update buffer queue length falls into one of 3 zones: + * green, yellow, red. If the value is in [0, green) nothing is + * done, the buffers are left unprocessed to enable the caching effect of the + * dirtied cards. In the yellow zone [green, yellow) the concurrent refinement + * threads are gradually activated. In [yellow, red) all threads are + * running. If the length becomes red (max queue length) the mutators start + * processing the buffers. + * + * There are some interesting cases (with G1AdaptiveConcRefine turned off): + * 1) green = yellow = red = 0. In this case the mutator will process all + * buffers. Except for those that are created by the deferred updates + * machinery during a collection. + * 2) green = 0. Means no caching. Can be a good way to minimize the + * amount of time spent updating rsets during a collection. + */ + int _green_zone; + int _yellow_zone; + int _red_zone; + + int _thread_threshold_step; + + // Reset the threshold step value based of the current zone boundaries. + void reset_threshold_step(); // The cache for card refinement. bool _use_cache; @@ -147,6 +172,8 @@ void init(); // Accomplish some initialization that has to wait. void stop(); + void reinitialize_threads(); + // Iterate over the conc refine threads void threads_do(ThreadClosure *tc); @@ -178,7 +205,20 @@ void clear_and_record_card_counts(); - static size_t thread_num(); + static int thread_num(); void print_worker_threads_on(outputStream* st) const; + + void set_green_zone(int x) { _green_zone = x; } + void set_yellow_zone(int x) { _yellow_zone = x; } + void set_red_zone(int x) { _red_zone = x; } + + int green_zone() const { return _green_zone; } + int yellow_zone() const { return _yellow_zone; } + int red_zone() const { return _red_zone; } + + int total_thread_num() const { return _n_threads; } + int worker_thread_num() const { return _n_worker_threads; } + + int thread_threshold_step() const { return _thread_threshold_step; } };
--- a/src/share/vm/gc_implementation/g1/concurrentG1RefineThread.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/concurrentG1RefineThread.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -25,10 +25,6 @@ #include "incls/_precompiled.incl" #include "incls/_concurrentG1RefineThread.cpp.incl" -// ======= Concurrent Mark Thread ======== - -// The CM thread is created when the G1 garbage collector is used - ConcurrentG1RefineThread:: ConcurrentG1RefineThread(ConcurrentG1Refine* cg1r, ConcurrentG1RefineThread *next, int worker_id_offset, int worker_id) : @@ -37,19 +33,42 @@ _worker_id(worker_id), _active(false), _next(next), + _monitor(NULL), _cg1r(cg1r), - _vtime_accum(0.0), - _interval_ms(5.0) + _vtime_accum(0.0) { + + // Each thread has its own monitor. The i-th thread is responsible for signalling + // to thread i+1 if the number of buffers in the queue exceeds a threashold for this + // thread. Monitors are also used to wake up the threads during termination. + // The 0th worker in notified by mutator threads and has a special monitor. + // The last worker is used for young gen rset size sampling. + if (worker_id > 0) { + _monitor = new Monitor(Mutex::nonleaf, "Refinement monitor", true); + } else { + _monitor = DirtyCardQ_CBL_mon; + } + initialize(); create_and_start(); } +void ConcurrentG1RefineThread::initialize() { + if (_worker_id < cg1r()->worker_thread_num()) { + // Current thread activation threshold + _threshold = MIN2<int>(cg1r()->thread_threshold_step() * (_worker_id + 1) + cg1r()->green_zone(), + cg1r()->yellow_zone()); + // A thread deactivates once the number of buffer reached a deactivation threshold + _deactivation_threshold = MAX2<int>(_threshold - cg1r()->thread_threshold_step(), cg1r()->green_zone()); + } else { + set_active(true); + } +} + void ConcurrentG1RefineThread::sample_young_list_rs_lengths() { G1CollectedHeap* g1h = G1CollectedHeap::heap(); G1CollectorPolicy* g1p = g1h->g1_policy(); if (g1p->adaptive_young_list_length()) { int regions_visited = 0; - g1h->young_list_rs_length_sampling_init(); while (g1h->young_list_rs_length_sampling_more()) { g1h->young_list_rs_length_sampling_next(); @@ -70,99 +89,121 @@ } } -void ConcurrentG1RefineThread::run() { - initialize_in_thread(); +void ConcurrentG1RefineThread::run_young_rs_sampling() { + DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); _vtime_start = os::elapsedVTime(); - wait_for_universe_init(); + while(!_should_terminate) { + _sts.join(); + sample_young_list_rs_lengths(); + _sts.leave(); - while (!_should_terminate) { - DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); - // Wait for completed log buffers to exist. - { - MutexLockerEx x(DirtyCardQ_CBL_mon, Mutex::_no_safepoint_check_flag); - while (((_worker_id == 0 && !dcqs.process_completed_buffers()) || - (_worker_id > 0 && !is_active())) && - !_should_terminate) { - DirtyCardQ_CBL_mon->wait(Mutex::_no_safepoint_check_flag); - } - } - - if (_should_terminate) { - return; + if (os::supports_vtime()) { + _vtime_accum = (os::elapsedVTime() - _vtime_start); + } else { + _vtime_accum = 0.0; } - // Now we take them off (this doesn't hold locks while it applies - // closures.) (If we did a full collection, then we'll do a full - // traversal. - _sts.join(); - int n_logs = 0; - int lower_limit = 0; - double start_vtime_sec; // only used when G1SmoothConcRefine is on - int prev_buffer_num; // only used when G1SmoothConcRefine is on - // This thread activation threshold - int threshold = G1UpdateBufferQueueProcessingThreshold * _worker_id; - // Next thread activation threshold - int next_threshold = threshold + G1UpdateBufferQueueProcessingThreshold; - int deactivation_threshold = MAX2<int>(threshold - G1UpdateBufferQueueProcessingThreshold / 2, 0); + MutexLockerEx x(_monitor, Mutex::_no_safepoint_check_flag); + if (_should_terminate) { + break; + } + _monitor->wait(Mutex::_no_safepoint_check_flag, G1ConcRefineServiceInterval); + } +} - if (G1SmoothConcRefine) { - lower_limit = 0; - start_vtime_sec = os::elapsedVTime(); - prev_buffer_num = (int) dcqs.completed_buffers_num(); - } else { - lower_limit = G1UpdateBufferQueueProcessingThreshold / 4; // For now. +void ConcurrentG1RefineThread::wait_for_completed_buffers() { + DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); + MutexLockerEx x(_monitor, Mutex::_no_safepoint_check_flag); + while (!_should_terminate && !is_active()) { + _monitor->wait(Mutex::_no_safepoint_check_flag); + } +} + +bool ConcurrentG1RefineThread::is_active() { + DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); + return _worker_id > 0 ? _active : dcqs.process_completed_buffers(); +} + +void ConcurrentG1RefineThread::activate() { + MutexLockerEx x(_monitor, Mutex::_no_safepoint_check_flag); + if (_worker_id > 0) { + if (G1TraceConcurrentRefinement) { + DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); + gclog_or_tty->print_cr("G1-Refine-activated worker %d, on threshold %d, current %d", + _worker_id, _threshold, (int)dcqs.completed_buffers_num()); } - while (dcqs.apply_closure_to_completed_buffer(_worker_id + _worker_id_offset, lower_limit)) { - double end_vtime_sec; - double elapsed_vtime_sec; - int elapsed_vtime_ms; - int curr_buffer_num = (int) dcqs.completed_buffers_num(); + set_active(true); + } else { + DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); + dcqs.set_process_completed(true); + } + _monitor->notify(); +} - if (G1SmoothConcRefine) { - end_vtime_sec = os::elapsedVTime(); - elapsed_vtime_sec = end_vtime_sec - start_vtime_sec; - elapsed_vtime_ms = (int) (elapsed_vtime_sec * 1000.0); +void ConcurrentG1RefineThread::deactivate() { + MutexLockerEx x(_monitor, Mutex::_no_safepoint_check_flag); + if (_worker_id > 0) { + if (G1TraceConcurrentRefinement) { + DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); + gclog_or_tty->print_cr("G1-Refine-deactivated worker %d, off threshold %d, current %d", + _worker_id, _deactivation_threshold, (int)dcqs.completed_buffers_num()); + } + set_active(false); + } else { + DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); + dcqs.set_process_completed(false); + } +} + +void ConcurrentG1RefineThread::run() { + initialize_in_thread(); + wait_for_universe_init(); - if (curr_buffer_num > prev_buffer_num || - curr_buffer_num > next_threshold) { - decreaseInterval(elapsed_vtime_ms); - } else if (curr_buffer_num < prev_buffer_num) { - increaseInterval(elapsed_vtime_ms); - } + if (_worker_id >= cg1r()->worker_thread_num()) { + run_young_rs_sampling(); + terminate(); + } + + _vtime_start = os::elapsedVTime(); + while (!_should_terminate) { + DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); + + // Wait for work + wait_for_completed_buffers(); + + if (_should_terminate) { + break; + } + + _sts.join(); + + do { + int curr_buffer_num = (int)dcqs.completed_buffers_num(); + // If the number of the buffers falls down into the yellow zone, + // that means that the transition period after the evacuation pause has ended. + if (dcqs.completed_queue_padding() > 0 && curr_buffer_num <= cg1r()->yellow_zone()) { + dcqs.set_completed_queue_padding(0); } - if (_worker_id == 0) { - sample_young_list_rs_lengths(); - } else if (curr_buffer_num < deactivation_threshold) { + + if (_worker_id > 0 && curr_buffer_num <= _deactivation_threshold) { // If the number of the buffer has fallen below our threshold // we should deactivate. The predecessor will reactivate this // thread should the number of the buffers cross the threshold again. - MutexLockerEx x(DirtyCardQ_CBL_mon, Mutex::_no_safepoint_check_flag); deactivate(); - if (G1TraceConcurrentRefinement) { - gclog_or_tty->print_cr("G1-Refine-deactivated worker %d", _worker_id); - } break; } // Check if we need to activate the next thread. - if (curr_buffer_num > next_threshold && _next != NULL && !_next->is_active()) { - MutexLockerEx x(DirtyCardQ_CBL_mon, Mutex::_no_safepoint_check_flag); + if (_next != NULL && !_next->is_active() && curr_buffer_num > _next->_threshold) { _next->activate(); - DirtyCardQ_CBL_mon->notify_all(); - if (G1TraceConcurrentRefinement) { - gclog_or_tty->print_cr("G1-Refine-activated worker %d", _next->_worker_id); - } } + } while (dcqs.apply_closure_to_completed_buffer(_worker_id + _worker_id_offset, cg1r()->green_zone())); - if (G1SmoothConcRefine) { - prev_buffer_num = curr_buffer_num; - _sts.leave(); - os::sleep(Thread::current(), (jlong) _interval_ms, false); - _sts.join(); - start_vtime_sec = os::elapsedVTime(); - } - n_logs++; + // We can exit the loop above while being active if there was a yield request. + if (is_active()) { + deactivate(); } + _sts.leave(); if (os::supports_vtime()) { @@ -172,7 +213,6 @@ } } assert(_should_terminate, "just checking"); - terminate(); } @@ -191,8 +231,8 @@ } { - MutexLockerEx x(DirtyCardQ_CBL_mon, Mutex::_no_safepoint_check_flag); - DirtyCardQ_CBL_mon->notify_all(); + MutexLockerEx x(_monitor, Mutex::_no_safepoint_check_flag); + _monitor->notify(); } {
--- a/src/share/vm/gc_implementation/g1/concurrentG1RefineThread.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/concurrentG1RefineThread.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -40,42 +40,36 @@ // when the number of the rset update buffer crosses a certain threshold. A successor // would self-deactivate when the number of the buffers falls below the threshold. bool _active; - ConcurrentG1RefineThread * _next; - public: - virtual void run(); - - bool is_active() { return _active; } - void activate() { _active = true; } - void deactivate() { _active = false; } - - private: - ConcurrentG1Refine* _cg1r; - - double _interval_ms; + ConcurrentG1RefineThread* _next; + Monitor* _monitor; + ConcurrentG1Refine* _cg1r; - void decreaseInterval(int processing_time_ms) { - double min_interval_ms = (double) processing_time_ms; - _interval_ms = 0.8 * _interval_ms; - if (_interval_ms < min_interval_ms) - _interval_ms = min_interval_ms; - } - void increaseInterval(int processing_time_ms) { - double max_interval_ms = 9.0 * (double) processing_time_ms; - _interval_ms = 1.1 * _interval_ms; - if (max_interval_ms > 0 && _interval_ms > max_interval_ms) - _interval_ms = max_interval_ms; - } + int _thread_threshold_step; + // This thread activation threshold + int _threshold; + // This thread deactivation threshold + int _deactivation_threshold; - void sleepBeforeNextCycle(); + void sample_young_list_rs_lengths(); + void run_young_rs_sampling(); + void wait_for_completed_buffers(); + + void set_active(bool x) { _active = x; } + bool is_active(); + void activate(); + void deactivate(); // For use by G1CollectedHeap, which is a friend. static SuspendibleThreadSet* sts() { return &_sts; } - public: +public: + virtual void run(); // Constructor ConcurrentG1RefineThread(ConcurrentG1Refine* cg1r, ConcurrentG1RefineThread* next, int worker_id_offset, int worker_id); + void initialize(); + // Printing void print() const; void print_on(outputStream* st) const; @@ -83,13 +77,10 @@ // Total virtual time so far. double vtime_accum() { return _vtime_accum; } - ConcurrentG1Refine* cg1r() { return _cg1r; } - - void sample_young_list_rs_lengths(); + ConcurrentG1Refine* cg1r() { return _cg1r; } // Yield for GC - void yield(); - + void yield(); // shutdown void stop(); };
--- a/src/share/vm/gc_implementation/g1/concurrentMark.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/concurrentMark.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -760,7 +760,6 @@ rp->setup_policy(false); // snapshot the soft ref policy to be used in this cycle SATBMarkQueueSet& satb_mq_set = JavaThread::satb_mark_queue_set(); - satb_mq_set.set_process_completed_threshold(G1SATBProcessCompletedThreshold); satb_mq_set.set_active_all_threads(true); // update_g1_committed() will be called at the end of an evac pause
--- a/src/share/vm/gc_implementation/g1/dirtyCardQueue.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/dirtyCardQueue.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -61,8 +61,8 @@ #pragma warning( disable:4355 ) // 'this' : used in base member initializer list #endif // _MSC_VER -DirtyCardQueueSet::DirtyCardQueueSet() : - PtrQueueSet(true /*notify_when_complete*/), +DirtyCardQueueSet::DirtyCardQueueSet(bool notify_when_complete) : + PtrQueueSet(notify_when_complete), _closure(NULL), _shared_dirty_card_queue(this, true /*perm*/), _free_ids(NULL), @@ -77,12 +77,12 @@ } void DirtyCardQueueSet::initialize(Monitor* cbl_mon, Mutex* fl_lock, + int process_completed_threshold, int max_completed_queue, Mutex* lock, PtrQueueSet* fl_owner) { - PtrQueueSet::initialize(cbl_mon, fl_lock, max_completed_queue, fl_owner); + PtrQueueSet::initialize(cbl_mon, fl_lock, process_completed_threshold, + max_completed_queue, fl_owner); set_buffer_size(G1UpdateBufferSize); - set_process_completed_threshold(G1UpdateBufferQueueProcessingThreshold); - _shared_dirty_card_queue.set_lock(lock); _free_ids = new FreeIdSet((int) num_par_ids(), _cbl_mon); } @@ -154,9 +154,10 @@ return b; } -DirtyCardQueueSet::CompletedBufferNode* -DirtyCardQueueSet::get_completed_buffer_lock(int stop_at) { - CompletedBufferNode* nd = NULL; + +BufferNode* +DirtyCardQueueSet::get_completed_buffer(int stop_at) { + BufferNode* nd = NULL; MutexLockerEx x(_cbl_mon, Mutex::_no_safepoint_check_flag); if ((int)_n_completed_buffers <= stop_at) { @@ -166,53 +167,31 @@ if (_completed_buffers_head != NULL) { nd = _completed_buffers_head; - _completed_buffers_head = nd->next; + _completed_buffers_head = nd->next(); if (_completed_buffers_head == NULL) _completed_buffers_tail = NULL; _n_completed_buffers--; + assert(_n_completed_buffers >= 0, "Invariant"); } debug_only(assert_completed_buffer_list_len_correct_locked()); return nd; } -// We only do this in contexts where there is no concurrent enqueueing. -DirtyCardQueueSet::CompletedBufferNode* -DirtyCardQueueSet::get_completed_buffer_CAS() { - CompletedBufferNode* nd = _completed_buffers_head; - - while (nd != NULL) { - CompletedBufferNode* next = nd->next; - CompletedBufferNode* result = - (CompletedBufferNode*)Atomic::cmpxchg_ptr(next, - &_completed_buffers_head, - nd); - if (result == nd) { - return result; - } else { - nd = _completed_buffers_head; - } - } - assert(_completed_buffers_head == NULL, "Loop post"); - _completed_buffers_tail = NULL; - return NULL; -} - bool DirtyCardQueueSet:: apply_closure_to_completed_buffer_helper(int worker_i, - CompletedBufferNode* nd) { + BufferNode* nd) { if (nd != NULL) { + void **buf = BufferNode::make_buffer_from_node(nd); + size_t index = nd->index(); bool b = - DirtyCardQueue::apply_closure_to_buffer(_closure, nd->buf, - nd->index, _sz, + DirtyCardQueue::apply_closure_to_buffer(_closure, buf, + index, _sz, true, worker_i); - void** buf = nd->buf; - size_t index = nd->index; - delete nd; if (b) { deallocate_buffer(buf); return true; // In normal case, go on to next buffer. } else { - enqueue_complete_buffer(buf, index, true); + enqueue_complete_buffer(buf, index); return false; } } else { @@ -222,40 +201,36 @@ bool DirtyCardQueueSet::apply_closure_to_completed_buffer(int worker_i, int stop_at, - bool with_CAS) + bool during_pause) { - CompletedBufferNode* nd = NULL; - if (with_CAS) { - guarantee(stop_at == 0, "Precondition"); - nd = get_completed_buffer_CAS(); - } else { - nd = get_completed_buffer_lock(stop_at); - } + assert(!during_pause || stop_at == 0, "Should not leave any completed buffers during a pause"); + BufferNode* nd = get_completed_buffer(stop_at); bool res = apply_closure_to_completed_buffer_helper(worker_i, nd); if (res) Atomic::inc(&_processed_buffers_rs_thread); return res; } void DirtyCardQueueSet::apply_closure_to_all_completed_buffers() { - CompletedBufferNode* nd = _completed_buffers_head; + BufferNode* nd = _completed_buffers_head; while (nd != NULL) { bool b = - DirtyCardQueue::apply_closure_to_buffer(_closure, nd->buf, 0, _sz, - false); + DirtyCardQueue::apply_closure_to_buffer(_closure, + BufferNode::make_buffer_from_node(nd), + 0, _sz, false); guarantee(b, "Should not stop early."); - nd = nd->next; + nd = nd->next(); } } void DirtyCardQueueSet::abandon_logs() { assert(SafepointSynchronize::is_at_safepoint(), "Must be at safepoint."); - CompletedBufferNode* buffers_to_delete = NULL; + BufferNode* buffers_to_delete = NULL; { MutexLockerEx x(_cbl_mon, Mutex::_no_safepoint_check_flag); while (_completed_buffers_head != NULL) { - CompletedBufferNode* nd = _completed_buffers_head; - _completed_buffers_head = nd->next; - nd->next = buffers_to_delete; + BufferNode* nd = _completed_buffers_head; + _completed_buffers_head = nd->next(); + nd->set_next(buffers_to_delete); buffers_to_delete = nd; } _n_completed_buffers = 0; @@ -263,10 +238,9 @@ debug_only(assert_completed_buffer_list_len_correct_locked()); } while (buffers_to_delete != NULL) { - CompletedBufferNode* nd = buffers_to_delete; - buffers_to_delete = nd->next; - deallocate_buffer(nd->buf); - delete nd; + BufferNode* nd = buffers_to_delete; + buffers_to_delete = nd->next(); + deallocate_buffer(BufferNode::make_buffer_from_node(nd)); } // Since abandon is done only at safepoints, we can safely manipulate // these queues.
--- a/src/share/vm/gc_implementation/g1/dirtyCardQueue.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/dirtyCardQueue.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2001-2007 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2001-2009 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -84,11 +84,12 @@ jint _processed_buffers_rs_thread; public: - DirtyCardQueueSet(); + DirtyCardQueueSet(bool notify_when_complete = true); void initialize(Monitor* cbl_mon, Mutex* fl_lock, - int max_completed_queue = 0, - Mutex* lock = NULL, PtrQueueSet* fl_owner = NULL); + int process_completed_threshold, + int max_completed_queue, + Mutex* lock, PtrQueueSet* fl_owner = NULL); // The number of parallel ids that can be claimed to allow collector or // mutator threads to do card-processing work. @@ -120,12 +121,13 @@ // is returned to the completed buffer set, and this call returns false. bool apply_closure_to_completed_buffer(int worker_i = 0, int stop_at = 0, - bool with_CAS = false); + bool during_pause = false); + bool apply_closure_to_completed_buffer_helper(int worker_i, - CompletedBufferNode* nd); + BufferNode* nd); - CompletedBufferNode* get_completed_buffer_CAS(); - CompletedBufferNode* get_completed_buffer_lock(int stop_at); + BufferNode* get_completed_buffer(int stop_at); + // Applies the current closure to all completed buffers, // non-consumptively. void apply_closure_to_all_completed_buffers();
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -928,6 +928,8 @@ TraceCPUTime tcpu(PrintGCDetails, true, gclog_or_tty); TraceTime t(full ? "Full GC (System.gc())" : "Full GC", PrintGC, true, gclog_or_tty); + TraceMemoryManagerStats tms(true /* fullGC */); + double start = os::elapsedTime(); g1_policy()->record_full_collection_start(); @@ -1001,6 +1003,8 @@ COMPILER2_PRESENT(DerivedPointerTable::update_pointers()); + MemoryService::track_memory_usage(); + if (VerifyAfterGC && total_collections() >= VerifyGCStartAt) { HandleMark hm; // Discard invalid handles created during verification gclog_or_tty->print(" VerifyAfterGC:"); @@ -1371,6 +1375,7 @@ G1CollectedHeap::G1CollectedHeap(G1CollectorPolicy* policy_) : SharedHeap(policy_), _g1_policy(policy_), + _dirty_card_queue_set(false), _ref_processor(NULL), _process_strong_tasks(new SubTasksDone(G1H_PS_NumElements)), _bot_shared(NULL), @@ -1436,6 +1441,7 @@ } jint G1CollectedHeap::initialize() { + CollectedHeap::pre_initialize(); os::enable_vtime(); // Necessary to satisfy locking discipline assertions. @@ -1456,8 +1462,6 @@ Universe::check_alignment(init_byte_size, HeapRegion::GrainBytes, "g1 heap"); Universe::check_alignment(max_byte_size, HeapRegion::GrainBytes, "g1 heap"); - // We allocate this in any case, but only do no work if the command line - // param is off. _cg1r = new ConcurrentG1Refine(); // Reserve the maximum. @@ -1590,18 +1594,20 @@ JavaThread::satb_mark_queue_set().initialize(SATB_Q_CBL_mon, SATB_Q_FL_lock, - 0, + G1SATBProcessCompletedThreshold, Shared_SATB_Q_lock); JavaThread::dirty_card_queue_set().initialize(DirtyCardQ_CBL_mon, DirtyCardQ_FL_lock, - G1UpdateBufferQueueMaxLength, + concurrent_g1_refine()->yellow_zone(), + concurrent_g1_refine()->red_zone(), Shared_DirtyCardQ_lock); if (G1DeferredRSUpdate) { dirty_card_queue_set().initialize(DirtyCardQ_CBL_mon, DirtyCardQ_FL_lock, - 0, + -1, // never trigger processing + -1, // no limit on length Shared_DirtyCardQ_lock, &JavaThread::dirty_card_queue_set()); } @@ -1732,13 +1738,6 @@ return car->free(); } -void G1CollectedHeap::collect(GCCause::Cause cause) { - // The caller doesn't have the Heap_lock - assert(!Heap_lock->owned_by_self(), "this thread should not own the Heap_lock"); - MutexLocker ml(Heap_lock); - collect_locked(cause); -} - void G1CollectedHeap::collect_as_vm_thread(GCCause::Cause cause) { assert(Thread::current()->is_VM_thread(), "Precondition#1"); assert(Heap_lock->is_locked(), "Precondition#2"); @@ -1755,17 +1754,31 @@ } } - -void G1CollectedHeap::collect_locked(GCCause::Cause cause) { - // Don't want to do a GC until cleanup is completed. - wait_for_cleanup_complete(); - - // Read the GC count while holding the Heap_lock - int gc_count_before = SharedHeap::heap()->total_collections(); +void G1CollectedHeap::collect(GCCause::Cause cause) { + // The caller doesn't have the Heap_lock + assert(!Heap_lock->owned_by_self(), "this thread should not own the Heap_lock"); + + int gc_count_before; { - MutexUnlocker mu(Heap_lock); // give up heap lock, execute gets it back - VM_G1CollectFull op(gc_count_before, cause); - VMThread::execute(&op); + MutexLocker ml(Heap_lock); + // Read the GC count while holding the Heap_lock + gc_count_before = SharedHeap::heap()->total_collections(); + + // Don't want to do a GC until cleanup is completed. + wait_for_cleanup_complete(); + } // We give up heap lock; VMThread::execute gets it back below + switch (cause) { + case GCCause::_scavenge_alot: { + // Do an incremental pause, which might sometimes be abandoned. + VM_G1IncCollectionPause op(gc_count_before, cause); + VMThread::execute(&op); + break; + } + default: { + // In all other cases, we currently do a full gc. + VM_G1CollectFull op(gc_count_before, cause); + VMThread::execute(&op); + } } } @@ -2119,7 +2132,7 @@ } size_t G1CollectedHeap::max_capacity() const { - return _g1_committed.byte_size(); + return g1_reserved_obj_bytes(); } jlong G1CollectedHeap::millis_since_last_gc() { @@ -2492,6 +2505,7 @@ } void G1CollectedHeap::gc_prologue(bool full /* Ignored */) { + // always_do_update_barrier = false; assert(InlineCacheBuffer::is_empty(), "should have cleaned up ICBuffer"); // Call allocation profiler AllocationProfiler::iterate_since_last_gc(); @@ -2505,6 +2519,7 @@ // is set. COMPILER2_PRESENT(assert(DerivedPointerTable::is_empty(), "derived pointer present")); + // always_do_update_barrier = true; } void G1CollectedHeap::do_collection_pause() { @@ -2631,6 +2646,13 @@ // </NEW PREDICTION> +struct PrepareForRSScanningClosure : public HeapRegionClosure { + bool doHeapRegion(HeapRegion *r) { + r->rem_set()->set_iter_claimed(0); + return false; + } +}; + void G1CollectedHeap::do_collection_pause_at_safepoint() { if (PrintHeapAtGC) { @@ -2638,6 +2660,8 @@ } { + ResourceMark rm; + char verbose_str[128]; sprintf(verbose_str, "GC pause "); if (g1_policy()->in_young_gc_mode()) { @@ -2649,8 +2673,6 @@ if (g1_policy()->should_initiate_conc_mark()) strcat(verbose_str, " (initial-mark)"); - GCCauseSetter x(this, GCCause::_g1_inc_collection_pause); - // if PrintGCDetails is on, we'll print long statistics information // in the collector policy code, so let's not print this as the output // is messy if we do. @@ -2658,7 +2680,8 @@ TraceCPUTime tcpu(PrintGCDetails, true, gclog_or_tty); TraceTime t(verbose_str, PrintGC && !PrintGCDetails, true, gclog_or_tty); - ResourceMark rm; + TraceMemoryManagerStats tms(false /* fullGC */); + assert(SafepointSynchronize::is_at_safepoint(), "should be at safepoint"); assert(Thread::current() == VMThread::vm_thread(), "should be in vm thread"); guarantee(!is_gc_active(), "collection is not reentrant"); @@ -2768,6 +2791,8 @@ gclog_or_tty->print_cr("\nAfter pause, heap:"); print(); #endif + PrepareForRSScanningClosure prepare_for_rs_scan; + collection_set_iterate(&prepare_for_rs_scan); setup_surviving_young_words(); @@ -2802,6 +2827,22 @@ _young_list->reset_auxilary_lists(); } } else { + if (_in_cset_fast_test != NULL) { + assert(_in_cset_fast_test_base != NULL, "Since _in_cset_fast_test isn't"); + FREE_C_HEAP_ARRAY(bool, _in_cset_fast_test_base); + // this is more for peace of mind; we're nulling them here and + // we're expecting them to be null at the beginning of the next GC + _in_cset_fast_test = NULL; + _in_cset_fast_test_base = NULL; + } + // This looks confusing, because the DPT should really be empty + // at this point -- since we have not done any collection work, + // there should not be any derived pointers in the table to update; + // however, there is some additional state in the DPT which is + // reset at the end of the (null) "gc" here via the following call. + // A better approach might be to split off that state resetting work + // into a separate method that asserts that the DPT is empty and call + // that here. That is deferred for now. COMPILER2_PRESENT(DerivedPointerTable::update_pointers()); } @@ -2838,6 +2879,8 @@ assert(regions_accounted_for(), "Region leakage."); + MemoryService::track_memory_usage(); + if (VerifyAfterGC && total_collections() >= VerifyGCStartAt) { HandleMark hm; // Discard invalid handles created during verification gclog_or_tty->print(" VerifyAfterGC:"); @@ -3747,22 +3790,16 @@ return obj; } -template <bool do_gen_barrier, G1Barrier barrier, bool do_mark_forwardee, bool skip_cset_test> +template <bool do_gen_barrier, G1Barrier barrier, bool do_mark_forwardee> template <class T> -void G1ParCopyClosure <do_gen_barrier, barrier, do_mark_forwardee, skip_cset_test> +void G1ParCopyClosure <do_gen_barrier, barrier, do_mark_forwardee> ::do_oop_work(T* p) { oop obj = oopDesc::load_decode_heap_oop(p); assert(barrier != G1BarrierRS || obj != NULL, "Precondition: G1BarrierRS implies obj is nonNull"); - // The only time we skip the cset test is when we're scanning - // references popped from the queue. And we only push on the queue - // references that we know point into the cset, so no point in - // checking again. But we'll leave an assert here for peace of mind. - assert(!skip_cset_test || _g1->obj_in_cs(obj), "invariant"); - // here the null check is implicit in the cset_fast_test() test - if (skip_cset_test || _g1->in_cset_fast_test(obj)) { + if (_g1->in_cset_fast_test(obj)) { #if G1_REM_SET_LOGGING gclog_or_tty->print_cr("Loc "PTR_FORMAT" contains pointer "PTR_FORMAT" " "into CS.", p, (void*) obj); @@ -3779,7 +3816,6 @@ } } - // When scanning moved objs, must look at all oops. if (barrier == G1BarrierEvac && obj != NULL) { _par_scan_state->update_rs(_from, p, _par_scan_state->queue_num()); } @@ -3789,8 +3825,8 @@ } } -template void G1ParCopyClosure<false, G1BarrierEvac, false, true>::do_oop_work(oop* p); -template void G1ParCopyClosure<false, G1BarrierEvac, false, true>::do_oop_work(narrowOop* p); +template void G1ParCopyClosure<false, G1BarrierEvac, false>::do_oop_work(oop* p); +template void G1ParCopyClosure<false, G1BarrierEvac, false>::do_oop_work(narrowOop* p); template <class T> void G1ParScanPartialArrayClosure::do_oop_nv(T* p) { assert(has_partial_array_mask(p), "invariant"); @@ -3862,11 +3898,11 @@ assert(UseCompressedOops, "Error"); narrowOop* p = (narrowOop*) stolen_task; assert(has_partial_array_mask(p) || - _g1h->obj_in_cs(oopDesc::load_decode_heap_oop(p)), "Error"); + _g1h->is_in_g1_reserved(oopDesc::load_decode_heap_oop(p)), "Error"); pss->push_on_queue(p); } else { oop* p = (oop*) stolen_task; - assert(has_partial_array_mask(p) || _g1h->obj_in_cs(*p), "Error"); + assert(has_partial_array_mask(p) || _g1h->is_in_g1_reserved(*p), "Error"); pss->push_on_queue(p); } continue; @@ -3928,6 +3964,7 @@ G1ParScanExtRootClosure only_scan_root_cl(_g1h, &pss); G1ParScanPermClosure only_scan_perm_cl(_g1h, &pss); G1ParScanHeapRSClosure only_scan_heap_rs_cl(_g1h, &pss); + G1ParPushHeapRSClosure push_heap_rs_cl(_g1h, &pss); G1ParScanAndMarkExtRootClosure scan_mark_root_cl(_g1h, &pss); G1ParScanAndMarkPermClosure scan_mark_perm_cl(_g1h, &pss); @@ -3951,7 +3988,7 @@ _g1h->g1_process_strong_roots(/* not collecting perm */ false, SharedHeap::SO_AllClasses, scan_root_cl, - &only_scan_heap_rs_cl, + &push_heap_rs_cl, scan_so_cl, scan_perm_cl, i); @@ -4209,10 +4246,11 @@ RedirtyLoggedCardTableEntryFastClosure redirty; dirty_card_queue_set().set_closure(&redirty); dirty_card_queue_set().apply_closure_to_all_completed_buffers(); - JavaThread::dirty_card_queue_set().merge_bufferlists(&dirty_card_queue_set()); + + DirtyCardQueueSet& dcq = JavaThread::dirty_card_queue_set(); + dcq.merge_bufferlists(&dirty_card_queue_set()); assert(dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed"); } - COMPILER2_PRESENT(DerivedPointerTable::update_pointers()); }
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -692,7 +692,7 @@ // Reserved (g1 only; super method includes perm), capacity and the used // portion in bytes. - size_t g1_reserved_obj_bytes() { return _g1_reserved.byte_size(); } + size_t g1_reserved_obj_bytes() const { return _g1_reserved.byte_size(); } virtual size_t capacity() const; virtual size_t used() const; // This should be called when we're not holding the heap lock. The @@ -1004,6 +1004,15 @@ // storage in the heap comes from a young region or not. // See ReduceInitialCardMarks. virtual bool can_elide_tlab_store_barriers() const { + // 6920090: Temporarily disabled, because of lingering + // instabilities related to RICM with G1. In the + // interim, the option ReduceInitialCardMarksForG1 + // below is left solely as a debugging device at least + // until 6920109 fixes the instabilities. + return ReduceInitialCardMarksForG1; + } + + virtual bool card_mark_must_follow_store() const { return true; } @@ -1022,6 +1031,8 @@ // However, non-generational G1 (-XX:-G1Gen) appears to have // bit-rotted so was not tested below. virtual bool can_elide_initializing_store_barrier(oop new_obj) { + // Re 6920090, 6920109 above. + assert(ReduceInitialCardMarksForG1, "Else cannot be here"); assert(G1Gen || !is_in_young(new_obj), "Non-generational G1 should never return true below"); return is_in_young(new_obj); @@ -1612,7 +1623,7 @@ template <class T> void push_on_queue(T* ref) { assert(ref != NULL, "invariant"); assert(has_partial_array_mask(ref) || - _g1h->obj_in_cs(oopDesc::load_decode_heap_oop(ref)), "invariant"); + _g1h->is_in_g1_reserved(oopDesc::load_decode_heap_oop(ref)), "invariant"); #ifdef ASSERT if (has_partial_array_mask(ref)) { oop p = clear_partial_array_mask(ref); @@ -1633,9 +1644,9 @@ assert((oop*)ref != NULL, "pop_local() returned true"); assert(UseCompressedOops || !ref.is_narrow(), "Error"); assert(has_partial_array_mask((oop*)ref) || - _g1h->obj_in_cs(ref.is_narrow() ? oopDesc::load_decode_heap_oop((narrowOop*)ref) - : oopDesc::load_decode_heap_oop((oop*)ref)), - "invariant"); + _g1h->is_in_g1_reserved(ref.is_narrow() ? oopDesc::load_decode_heap_oop((narrowOop*)ref) + : oopDesc::load_decode_heap_oop((oop*)ref)), + "invariant"); IF_G1_DETAILED_STATS(note_pop()); } else { StarTask null_task; @@ -1648,9 +1659,9 @@ assert((oop*)new_ref != NULL, "pop() from a local non-empty stack"); assert(UseCompressedOops || !new_ref.is_narrow(), "Error"); assert(has_partial_array_mask((oop*)new_ref) || - _g1h->obj_in_cs(new_ref.is_narrow() ? oopDesc::load_decode_heap_oop((narrowOop*)new_ref) - : oopDesc::load_decode_heap_oop((oop*)new_ref)), - "invariant"); + _g1h->is_in_g1_reserved(new_ref.is_narrow() ? oopDesc::load_decode_heap_oop((narrowOop*)new_ref) + : oopDesc::load_decode_heap_oop((oop*)new_ref)), + "invariant"); ref = new_ref; } @@ -1814,12 +1825,12 @@ assert(UseCompressedOops, "Error"); narrowOop* p = (narrowOop*)ref_to_scan; assert(!has_partial_array_mask(p) && - _g1h->obj_in_cs(oopDesc::load_decode_heap_oop(p)), "sanity"); + _g1h->is_in_g1_reserved(oopDesc::load_decode_heap_oop(p)), "sanity"); deal_with_reference(p); } else { oop* p = (oop*)ref_to_scan; - assert((has_partial_array_mask(p) && _g1h->obj_in_cs(clear_partial_array_mask(p))) || - _g1h->obj_in_cs(oopDesc::load_decode_heap_oop(p)), "sanity"); + assert((has_partial_array_mask(p) && _g1h->is_in_g1_reserved(clear_partial_array_mask(p))) || + _g1h->is_in_g1_reserved(oopDesc::load_decode_heap_oop(p)), "sanity"); deal_with_reference(p); } } @@ -1833,12 +1844,12 @@ assert(UseCompressedOops, "Error"); narrowOop* p = (narrowOop*)ref_to_scan; assert(!has_partial_array_mask(p) && - _g1h->obj_in_cs(oopDesc::load_decode_heap_oop(p)), "sanity"); + _g1h->is_in_g1_reserved(oopDesc::load_decode_heap_oop(p)), "sanity"); deal_with_reference(p); } else { oop* p = (oop*)ref_to_scan; assert((has_partial_array_mask(p) && _g1h->obj_in_cs(clear_partial_array_mask(p))) || - _g1h->obj_in_cs(oopDesc::load_decode_heap_oop(p)), "sanity"); + _g1h->is_in_g1_reserved(oopDesc::load_decode_heap_oop(p)), "sanity"); deal_with_reference(p); } }
--- a/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -205,6 +205,7 @@ // policy is created before the heap, we have to set this up here, // so it's done as soon as possible. HeapRegion::setup_heap_region_size(Arguments::min_heap_size()); + HeapRegionRemSet::setup_remset_size(); _recent_prev_end_times_for_all_gcs_sec->add(os::elapsedTime()); _prev_collection_pause_end_ms = os::elapsedTime() * 1000.0; @@ -1516,8 +1517,30 @@ (end_time_sec - _recent_prev_end_times_for_all_gcs_sec->oldest()) * 1000.0; update_recent_gc_times(end_time_sec, elapsed_ms); _recent_avg_pause_time_ratio = _recent_gc_times_ms->sum()/interval_ms; - // using 1.01 to account for floating point inaccuracies - assert(recent_avg_pause_time_ratio() < 1.01, "All GC?"); + if (recent_avg_pause_time_ratio() < 0.0 || + (recent_avg_pause_time_ratio() - 1.0 > 0.0)) { +#ifndef PRODUCT + // Dump info to allow post-facto debugging + gclog_or_tty->print_cr("recent_avg_pause_time_ratio() out of bounds"); + gclog_or_tty->print_cr("-------------------------------------------"); + gclog_or_tty->print_cr("Recent GC Times (ms):"); + _recent_gc_times_ms->dump(); + gclog_or_tty->print_cr("(End Time=%3.3f) Recent GC End Times (s):", end_time_sec); + _recent_prev_end_times_for_all_gcs_sec->dump(); + gclog_or_tty->print_cr("GC = %3.3f, Interval = %3.3f, Ratio = %3.3f", + _recent_gc_times_ms->sum(), interval_ms, recent_avg_pause_time_ratio()); + // In debug mode, terminate the JVM if the user wants to debug at this point. + assert(!G1FailOnFPError, "Debugging data for CR 6898948 has been dumped above"); +#endif // !PRODUCT + // Clip ratio between 0.0 and 1.0, and continue. This will be fixed in + // CR 6902692 by redoing the manner in which the ratio is incrementally computed. + if (_recent_avg_pause_time_ratio < 0.0) { + _recent_avg_pause_time_ratio = 0.0; + } else { + assert(_recent_avg_pause_time_ratio - 1.0 > 0.0, "Ctl-point invariant"); + _recent_avg_pause_time_ratio = 1.0; + } + } } if (G1PolicyVerbose > 1) { @@ -1892,6 +1915,10 @@ calculate_young_list_min_length(); calculate_young_list_target_config(); + // Note that _mmu_tracker->max_gc_time() returns the time in seconds. + double update_rs_time_goal_ms = _mmu_tracker->max_gc_time() * MILLIUNITS * G1RSUpdatePauseFractionPercent / 100.0; + adjust_concurrent_refinement(update_rs_time, update_rs_processed_buffers, update_rs_time_goal_ms); + // </NEW PREDICTION> _target_pause_time_ms = -1.0; @@ -1899,6 +1926,47 @@ // <NEW PREDICTION> +void G1CollectorPolicy::adjust_concurrent_refinement(double update_rs_time, + double update_rs_processed_buffers, + double goal_ms) { + DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); + ConcurrentG1Refine *cg1r = G1CollectedHeap::heap()->concurrent_g1_refine(); + + if (G1AdaptiveConcRefine) { + const int k_gy = 3, k_gr = 6; + const double inc_k = 1.1, dec_k = 0.9; + + int g = cg1r->green_zone(); + if (update_rs_time > goal_ms) { + g = (int)(g * dec_k); // Can become 0, that's OK. That would mean a mutator-only processing. + } else { + if (update_rs_time < goal_ms && update_rs_processed_buffers > g) { + g = (int)MAX2(g * inc_k, g + 1.0); + } + } + // Change the refinement threads params + cg1r->set_green_zone(g); + cg1r->set_yellow_zone(g * k_gy); + cg1r->set_red_zone(g * k_gr); + cg1r->reinitialize_threads(); + + int processing_threshold_delta = MAX2((int)(cg1r->green_zone() * sigma()), 1); + int processing_threshold = MIN2(cg1r->green_zone() + processing_threshold_delta, + cg1r->yellow_zone()); + // Change the barrier params + dcqs.set_process_completed_threshold(processing_threshold); + dcqs.set_max_completed_queue(cg1r->red_zone()); + } + + int curr_queue_size = dcqs.completed_buffers_num(); + if (curr_queue_size >= cg1r->yellow_zone()) { + dcqs.set_completed_queue_padding(curr_queue_size); + } else { + dcqs.set_completed_queue_padding(0); + } + dcqs.notify_if_necessary(); +} + double G1CollectorPolicy:: predict_young_collection_elapsed_time_ms(size_t adjustment) { @@ -2825,8 +2893,15 @@ double non_young_start_time_sec; start_recording_regions(); - guarantee(_target_pause_time_ms > -1.0, + guarantee(_target_pause_time_ms > -1.0 + NOT_PRODUCT(|| Universe::heap()->gc_cause() == GCCause::_scavenge_alot), "_target_pause_time_ms should have been set!"); +#ifndef PRODUCT + if (_target_pause_time_ms <= -1.0) { + assert(ScavengeALot && Universe::heap()->gc_cause() == GCCause::_scavenge_alot, "Error"); + _target_pause_time_ms = _mmu_tracker->max_gc_time() * 1000.0; + } +#endif assert(_collection_set == NULL, "Precondition"); double base_time_ms = predict_base_elapsed_time_ms(_pending_cards); @@ -2972,7 +3047,3 @@ G1CollectorPolicy::record_collection_pause_end(abandoned); assert(assertMarkedBytesDataOK(), "Marked regions not OK at pause end."); } - -// Local Variables: *** -// c-indentation-style: gnu *** -// End: ***
--- a/src/share/vm/gc_implementation/g1/g1CollectorPolicy.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1CollectorPolicy.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -316,6 +316,10 @@ bool verify_young_ages(HeapRegion* head, SurvRateGroup *surv_rate_group); #endif // PRODUCT + void adjust_concurrent_refinement(double update_rs_time, + double update_rs_processed_buffers, + double goal_ms); + protected: double _pause_time_target_ms; double _recorded_young_cset_choice_time_ms;
--- a/src/share/vm/gc_implementation/g1/g1MMUTracker.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1MMUTracker.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -86,12 +86,22 @@ // increase the array size (:-) // remove the oldest entry (this might allow more GC time for // the time slice than what's allowed) - // concolidate the two entries with the minimum gap between them - // (this mighte allow less GC time than what's allowed) - guarantee(0, "array full, currently we can't recover"); + // consolidate the two entries with the minimum gap between them + // (this might allow less GC time than what's allowed) + guarantee(NOT_PRODUCT(ScavengeALot ||) G1ForgetfulMMUTracker, + "array full, currently we can't recover unless +G1ForgetfulMMUTracker"); + // In the case where ScavengeALot is true, such overflow is not + // uncommon; in such cases, we can, without much loss of precision + // or performance (we are GC'ing most of the time anyway!), + // simply overwrite the oldest entry in the tracker: this + // is also the behaviour when G1ForgetfulMMUTracker is enabled. + _head_index = trim_index(_head_index + 1); + assert(_head_index == _tail_index, "Because we have a full circular buffer"); + _tail_index = trim_index(_tail_index + 1); + } else { + _head_index = trim_index(_head_index + 1); + ++_no_entries; } - _head_index = trim_index(_head_index + 1); - ++_no_entries; _array[_head_index] = G1MMUTrackerQueueElem(start, end); }
--- a/src/share/vm/gc_implementation/g1/g1MMUTracker.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1MMUTracker.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -99,7 +99,10 @@ // The array is of fixed size and I don't think we'll need more than // two or three entries with the current behaviour of G1 pauses. // If the array is full, an easy fix is to look for the pauses with - // the shortest gap between them and concolidate them. + // the shortest gap between them and consolidate them. + // For now, we have taken the expedient alternative of forgetting + // the oldest entry in the event that +G1ForgetfulMMUTracker, thus + // potentially violating MMU specs for some time thereafter. G1MMUTrackerQueueElem _array[QueueLength]; int _head_index;
--- a/src/share/vm/gc_implementation/g1/g1OopClosures.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1OopClosures.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -53,6 +53,15 @@ bool apply_to_weak_ref_discovered_field() { return true; } }; +class G1ParPushHeapRSClosure : public G1ParClosureSuper { +public: + G1ParPushHeapRSClosure(G1CollectedHeap* g1, G1ParScanThreadState* par_scan_state) : + G1ParClosureSuper(g1, par_scan_state) { } + template <class T> void do_oop_nv(T* p); + virtual void do_oop(oop* p) { do_oop_nv(p); } + virtual void do_oop(narrowOop* p) { do_oop_nv(p); } +}; + class G1ParScanClosure : public G1ParClosureSuper { public: G1ParScanClosure(G1CollectedHeap* g1, G1ParScanThreadState* par_scan_state) : @@ -100,7 +109,7 @@ }; template<bool do_gen_barrier, G1Barrier barrier, - bool do_mark_forwardee, bool skip_cset_test> + bool do_mark_forwardee> class G1ParCopyClosure : public G1ParCopyHelper { G1ParScanClosure _scanner; template <class T> void do_oop_work(T* p); @@ -116,12 +125,13 @@ virtual void do_oop(narrowOop* p) { do_oop_nv(p); } }; -typedef G1ParCopyClosure<false, G1BarrierNone, false, false> G1ParScanExtRootClosure; -typedef G1ParCopyClosure<true, G1BarrierNone, false, false> G1ParScanPermClosure; -typedef G1ParCopyClosure<false, G1BarrierRS, false, false> G1ParScanHeapRSClosure; -typedef G1ParCopyClosure<false, G1BarrierNone, true, false> G1ParScanAndMarkExtRootClosure; -typedef G1ParCopyClosure<true, G1BarrierNone, true, false> G1ParScanAndMarkPermClosure; -typedef G1ParCopyClosure<false, G1BarrierRS, true, false> G1ParScanAndMarkHeapRSClosure; +typedef G1ParCopyClosure<false, G1BarrierNone, false> G1ParScanExtRootClosure; +typedef G1ParCopyClosure<true, G1BarrierNone, false> G1ParScanPermClosure; +typedef G1ParCopyClosure<false, G1BarrierRS, false> G1ParScanHeapRSClosure; +typedef G1ParCopyClosure<false, G1BarrierNone, true> G1ParScanAndMarkExtRootClosure; +typedef G1ParCopyClosure<true, G1BarrierNone, true> G1ParScanAndMarkPermClosure; +typedef G1ParCopyClosure<false, G1BarrierRS, true> G1ParScanAndMarkHeapRSClosure; + // This is the only case when we set skip_cset_test. Basically, this // closure is (should?) only be called directly while we're draining // the overflow and task queues. In that case we know that the @@ -132,7 +142,7 @@ // We need a separate closure to handle references during evacuation // failure processing, as we cannot asume that the reference already // points into the collection set (like G1ParScanHeapEvacClosure does). -typedef G1ParCopyClosure<false, G1BarrierEvac, false, false> G1ParScanHeapEvacFailureClosure; +typedef G1ParCopyClosure<false, G1BarrierEvac, false> G1ParScanHeapEvacFailureClosure; class FilterIntoCSClosure: public OopClosure { G1CollectedHeap* _g1;
--- a/src/share/vm/gc_implementation/g1/g1OopClosures.inline.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1OopClosures.inline.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -104,3 +104,16 @@ } } } + +template <class T> inline void G1ParPushHeapRSClosure::do_oop_nv(T* p) { + T heap_oop = oopDesc::load_heap_oop(p); + + if (!oopDesc::is_null(heap_oop)) { + oop obj = oopDesc::decode_heap_oop_not_null(heap_oop); + if (_g1->in_cset_fast_test(obj)) { + Prefetch::write(obj->mark_addr(), 0); + Prefetch::read(obj->mark_addr(), (HeapWordSize*2)); + _par_scan_state->push_on_queue(p); + } + } +}
--- a/src/share/vm/gc_implementation/g1/g1RemSet.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1RemSet.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -155,8 +155,8 @@ G1BlockOffsetSharedArray* _bot_shared; CardTableModRefBS *_ct_bs; int _worker_i; + int _block_size; bool _try_claimed; - size_t _min_skip_distance, _max_skip_distance; public: ScanRSClosure(OopsInHeapRegionClosure* oc, int worker_i) : _oc(oc), @@ -168,8 +168,7 @@ _g1h = G1CollectedHeap::heap(); _bot_shared = _g1h->bot_shared(); _ct_bs = (CardTableModRefBS*) (_g1h->barrier_set()); - _min_skip_distance = 16; - _max_skip_distance = 2 * _g1h->n_par_threads() * _min_skip_distance; + _block_size = MAX2<int>(G1RSetScanBlockSize, 1); } void set_try_claimed() { _try_claimed = true; } @@ -225,12 +224,15 @@ HeapRegionRemSetIterator* iter = _g1h->rem_set_iterator(_worker_i); hrrs->init_iterator(iter); size_t card_index; - size_t skip_distance = 0, current_card = 0, jump_to_card = 0; - while (iter->has_next(card_index)) { - if (current_card < jump_to_card) { - ++current_card; - continue; + + // We claim cards in block so as to recude the contention. The block size is determined by + // the G1RSetScanBlockSize parameter. + size_t jump_to_card = hrrs->iter_claimed_next(_block_size); + for (size_t current_card = 0; iter->has_next(card_index); current_card++) { + if (current_card >= jump_to_card + _block_size) { + jump_to_card = hrrs->iter_claimed_next(_block_size); } + if (current_card < jump_to_card) continue; HeapWord* card_start = _g1h->bot_shared()->address_for_index(card_index); #if 0 gclog_or_tty->print("Rem set iteration yielded card [" PTR_FORMAT ", " PTR_FORMAT ").\n", @@ -247,22 +249,14 @@ // If the card is dirty, then we will scan it during updateRS. if (!card_region->in_collection_set() && !_ct_bs->is_card_dirty(card_index)) { - if (!_ct_bs->is_card_claimed(card_index) && _ct_bs->claim_card(card_index)) { - scanCard(card_index, card_region); - } else if (_try_claimed) { - if (jump_to_card == 0 || jump_to_card != current_card) { - // We did some useful work in the previous iteration. - // Decrease the distance. - skip_distance = MAX2(skip_distance >> 1, _min_skip_distance); - } else { - // Previous iteration resulted in a claim failure. - // Increase the distance. - skip_distance = MIN2(skip_distance << 1, _max_skip_distance); - } - jump_to_card = current_card + skip_distance; - } + // We make the card as "claimed" lazily (so races are possible but they're benign), + // which reduces the number of duplicate scans (the rsets of the regions in the cset + // can intersect). + if (!_ct_bs->is_card_claimed(card_index)) { + _ct_bs->set_card_claimed(card_index); + scanCard(card_index, card_region); + } } - ++current_card; } if (!_try_claimed) { hrrs->set_iter_complete(); @@ -299,30 +293,18 @@ double rs_time_start = os::elapsedTime(); HeapRegion *startRegion = calculateStartRegion(worker_i); - BufferingOopsInHeapRegionClosure boc(oc); - ScanRSClosure scanRScl(&boc, worker_i); + ScanRSClosure scanRScl(oc, worker_i); _g1->collection_set_iterate_from(startRegion, &scanRScl); scanRScl.set_try_claimed(); _g1->collection_set_iterate_from(startRegion, &scanRScl); - boc.done(); - double closure_app_time_sec = boc.closure_app_seconds(); - double scan_rs_time_sec = (os::elapsedTime() - rs_time_start) - - closure_app_time_sec; - double closure_app_time_ms = closure_app_time_sec * 1000.0; + double scan_rs_time_sec = os::elapsedTime() - rs_time_start; assert( _cards_scanned != NULL, "invariant" ); _cards_scanned[worker_i] = scanRScl.cards_done(); _g1p->record_scan_rs_start_time(worker_i, rs_time_start * 1000.0); _g1p->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0); - - double scan_new_refs_time_ms = _g1p->get_scan_new_refs_time(worker_i); - if (scan_new_refs_time_ms > 0.0) { - closure_app_time_ms += scan_new_refs_time_ms; - } - - _g1p->record_obj_copy_time(worker_i, closure_app_time_ms); } void HRInto_G1RemSet::updateRS(int worker_i) { @@ -449,9 +431,8 @@ oc->do_oop(p); } } - _g1p->record_scan_new_refs_time(worker_i, - (os::elapsedTime() - scan_new_refs_start_sec) - * 1000.0); + double scan_new_refs_time_ms = (os::elapsedTime() - scan_new_refs_start_sec) * 1000.0; + _g1p->record_scan_new_refs_time(worker_i, scan_new_refs_time_ms); } void HRInto_G1RemSet::cleanupHRRS() {
--- a/src/share/vm/gc_implementation/g1/g1_globals.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1_globals.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -85,7 +85,7 @@ diagnostic(bool, G1SummarizeZFStats, false, \ "Summarize zero-filling info") \ \ - develop(bool, G1TraceConcurrentRefinement, false, \ + diagnostic(bool, G1TraceConcurrentRefinement, false, \ "Trace G1 concurrent refinement") \ \ product(intx, G1MarkStackSize, 2 * 1024 * 1024, \ @@ -94,19 +94,6 @@ product(intx, G1MarkRegionStackSize, 1024 * 1024, \ "Size of the region stack for concurrent marking.") \ \ - develop(bool, G1ConcRefine, true, \ - "If true, run concurrent rem set refinement for G1") \ - \ - develop(intx, G1ConcRefineTargTraversals, 4, \ - "Number of concurrent refinement we try to achieve") \ - \ - develop(intx, G1ConcRefineInitialDelta, 4, \ - "Number of heap regions of alloc ahead of starting collection " \ - "pause to start concurrent refinement (initially)") \ - \ - develop(bool, G1SmoothConcRefine, true, \ - "Attempts to smooth out the overhead of concurrent refinement") \ - \ develop(bool, G1ConcZeroFill, true, \ "If true, run concurrent zero-filling thread") \ \ @@ -178,13 +165,38 @@ product(intx, G1UpdateBufferSize, 256, \ "Size of an update buffer") \ \ - product(intx, G1UpdateBufferQueueProcessingThreshold, 5, \ + product(intx, G1ConcRefineYellowZone, 0, \ "Number of enqueued update buffers that will " \ - "trigger concurrent processing") \ + "trigger concurrent processing. Will be selected ergonomically " \ + "by default.") \ + \ + product(intx, G1ConcRefineRedZone, 0, \ + "Maximum number of enqueued update buffers before mutator " \ + "threads start processing new ones instead of enqueueing them. " \ + "Will be selected ergonomically by default. Zero will disable " \ + "concurrent processing.") \ + \ + product(intx, G1ConcRefineGreenZone, 0, \ + "The number of update buffers that are left in the queue by the " \ + "concurrent processing threads. Will be selected ergonomically " \ + "by default.") \ \ - product(intx, G1UpdateBufferQueueMaxLength, 30, \ - "Maximum number of enqueued update buffers before mutator " \ - "threads start processing new ones instead of enqueueing them") \ + product(intx, G1ConcRefineServiceInterval, 300, \ + "The last concurrent refinement thread wakes up every " \ + "specified number of milliseconds to do miscellaneous work.") \ + \ + product(intx, G1ConcRefineThresholdStep, 0, \ + "Each time the rset update queue increases by this amount " \ + "activate the next refinement thread if available. " \ + "Will be selected ergonomically by default.") \ + \ + product(intx, G1RSUpdatePauseFractionPercent, 10, \ + "A target percentage of time that is allowed to be spend on " \ + "process RS update buffers during the collection pause.") \ + \ + product(bool, G1AdaptiveConcRefine, true, \ + "Select green, yellow and red zones adaptively to meet the " \ + "the pause requirements.") \ \ develop(intx, G1ConcRSLogCacheSize, 10, \ "Log base 2 of the length of conc RS hot-card cache.") \ @@ -195,8 +207,20 @@ develop(bool, G1PrintOopAppls, false, \ "When true, print applications of closures to external locs.") \ \ - develop(intx, G1LogRSRegionEntries, 7, \ - "Log_2 of max number of regions for which we keep bitmaps.") \ + develop(intx, G1RSetRegionEntriesBase, 256, \ + "Max number of regions in a fine-grain table per MB.") \ + \ + product(intx, G1RSetRegionEntries, 0, \ + "Max number of regions for which we keep bitmaps." \ + "Will be set ergonomically by default") \ + \ + develop(intx, G1RSetSparseRegionEntriesBase, 4, \ + "Max number of entries per region in a sparse table " \ + "per MB.") \ + \ + product(intx, G1RSetSparseRegionEntries, 0, \ + "Max number of entries per region in a sparse table." \ + "Will be set ergonomically by default.") \ \ develop(bool, G1RecordHRRSOops, false, \ "When true, record recent calls to rem set operations.") \ @@ -242,6 +266,10 @@ product(bool, G1UseSurvivorSpaces, true, \ "When true, use survivor space.") \ \ + develop(bool, G1FailOnFPError, false, \ + "When set, G1 will fail when it encounters an FP 'error', " \ + "so as to allow debugging") \ + \ develop(bool, G1FixedTenuringThreshold, false, \ "When set, G1 will not adjust the tenuring threshold") \ \ @@ -252,6 +280,9 @@ "If non-0 is the size of the G1 survivor space, " \ "otherwise SurvivorRatio is used to determine the size") \ \ + product(bool, G1ForgetfulMMUTracker, false, \ + "If the MMU tracker's memory is full, forget the oldest entry") \ + \ product(uintx, G1HeapRegionSize, 0, \ "Size of the G1 regions.") \ \ @@ -272,6 +303,14 @@ "a particular entry exceeds this value.") \ \ develop(bool, G1VerifyCTCleanup, false, \ - "Verify card table cleanup.") + "Verify card table cleanup.") \ + \ + product(uintx, G1RSetScanBlockSize, 64, \ + "Size of a work unit of cards claimed by a worker thread" \ + "during RSet scanning.") \ + \ + develop(bool, ReduceInitialCardMarksForG1, false, \ + "When ReduceInitialCardMarks is true, this flag setting " \ + " controls whether G1 allows the RICM optimization") G1_FLAGS(DECLARE_DEVELOPER_FLAG, DECLARE_PD_DEVELOPER_FLAG, DECLARE_PRODUCT_FLAG, DECLARE_PD_PRODUCT_FLAG, DECLARE_DIAGNOSTIC_FLAG, DECLARE_EXPERIMENTAL_FLAG, DECLARE_NOTPRODUCT_FLAG, DECLARE_MANAGEABLE_FLAG, DECLARE_PRODUCT_RW_FLAG)
--- a/src/share/vm/gc_implementation/g1/g1_specialized_oop_closures.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1_specialized_oop_closures.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -33,11 +33,12 @@ }; template<bool do_gen_barrier, G1Barrier barrier, - bool do_mark_forwardee, bool skip_cset_test> + bool do_mark_forwardee> class G1ParCopyClosure; class G1ParScanClosure; +class G1ParPushHeapRSClosure; -typedef G1ParCopyClosure<false, G1BarrierEvac, false, true> G1ParScanHeapEvacClosure; +typedef G1ParCopyClosure<false, G1BarrierEvac, false> G1ParScanHeapEvacClosure; class FilterIntoCSClosure; class FilterOutOfRegionClosure; @@ -51,6 +52,7 @@ #define FURTHER_SPECIALIZED_OOP_OOP_ITERATE_CLOSURES(f) \ f(G1ParScanHeapEvacClosure,_nv) \ f(G1ParScanClosure,_nv) \ + f(G1ParPushHeapRSClosure,_nv) \ f(FilterIntoCSClosure,_nv) \ f(FilterOutOfRegionClosure,_nv) \ f(FilterInHeapRegionAndIntoCSClosure,_nv) \
--- a/src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -258,42 +258,6 @@ ReserveParTableExpansion = 1 }; - void par_expand() { - int n = HeapRegionRemSet::num_par_rem_sets()-1; - if (n <= 0) return; - if (_par_tables == NULL) { - PerRegionTable* res = - (PerRegionTable*) - Atomic::cmpxchg_ptr((PerRegionTable*)ReserveParTableExpansion, - &_par_tables, NULL); - if (res != NULL) return; - // Otherwise, we reserved the right to do the expansion. - - PerRegionTable** ptables = NEW_C_HEAP_ARRAY(PerRegionTable*, n); - for (int i = 0; i < n; i++) { - PerRegionTable* ptable = PerRegionTable::alloc(hr()); - ptables[i] = ptable; - } - // Here we do not need an atomic. - _par_tables = ptables; -#if COUNT_PAR_EXPANDS - print_par_expand(); -#endif - // We must put this table on the expanded list. - PosParPRT* exp_head = _par_expanded_list; - while (true) { - set_next_par_expanded(exp_head); - PosParPRT* res = - (PosParPRT*) - Atomic::cmpxchg_ptr(this, &_par_expanded_list, exp_head); - if (res == exp_head) return; - // Otherwise. - exp_head = res; - } - ShouldNotReachHere(); - } - } - void par_contract() { assert(_par_tables != NULL, "Precondition."); int n = HeapRegionRemSet::num_par_rem_sets()-1; @@ -391,13 +355,49 @@ void set_next(PosParPRT* nxt) { _next = nxt; } PosParPRT** next_addr() { return &_next; } + bool should_expand(int tid) { + return par_tables() == NULL && tid > 0 && hr()->is_gc_alloc_region(); + } + + void par_expand() { + int n = HeapRegionRemSet::num_par_rem_sets()-1; + if (n <= 0) return; + if (_par_tables == NULL) { + PerRegionTable* res = + (PerRegionTable*) + Atomic::cmpxchg_ptr((PerRegionTable*)ReserveParTableExpansion, + &_par_tables, NULL); + if (res != NULL) return; + // Otherwise, we reserved the right to do the expansion. + + PerRegionTable** ptables = NEW_C_HEAP_ARRAY(PerRegionTable*, n); + for (int i = 0; i < n; i++) { + PerRegionTable* ptable = PerRegionTable::alloc(hr()); + ptables[i] = ptable; + } + // Here we do not need an atomic. + _par_tables = ptables; +#if COUNT_PAR_EXPANDS + print_par_expand(); +#endif + // We must put this table on the expanded list. + PosParPRT* exp_head = _par_expanded_list; + while (true) { + set_next_par_expanded(exp_head); + PosParPRT* res = + (PosParPRT*) + Atomic::cmpxchg_ptr(this, &_par_expanded_list, exp_head); + if (res == exp_head) return; + // Otherwise. + exp_head = res; + } + ShouldNotReachHere(); + } + } + void add_reference(OopOrNarrowOopStar from, int tid) { // Expand if necessary. PerRegionTable** pt = par_tables(); - if (par_tables() == NULL && tid > 0 && hr()->is_gc_alloc_region()) { - par_expand(); - pt = par_tables(); - } if (pt != NULL) { // We always have to assume that mods to table 0 are in parallel, // because of the claiming scheme in parallel expansion. A thread @@ -505,12 +505,13 @@ typedef PosParPRT* PosParPRTPtr; if (_max_fine_entries == 0) { assert(_mod_max_fine_entries_mask == 0, "Both or none."); - _max_fine_entries = (size_t)(1 << G1LogRSRegionEntries); + size_t max_entries_log = (size_t)log2_long((jlong)G1RSetRegionEntries); + _max_fine_entries = (size_t)(1 << max_entries_log); _mod_max_fine_entries_mask = _max_fine_entries - 1; #if SAMPLE_FOR_EVICTION assert(_fine_eviction_sample_size == 0 && _fine_eviction_stride == 0, "All init at same time."); - _fine_eviction_sample_size = MAX2((size_t)4, (size_t)G1LogRSRegionEntries); + _fine_eviction_sample_size = MAX2((size_t)4, max_entries_log); _fine_eviction_stride = _max_fine_entries / _fine_eviction_sample_size; #endif } @@ -655,13 +656,6 @@ #endif } - // Otherwise, transfer from sparse to fine-grain. - CardIdx_t cards[SparsePRTEntry::CardsPerEntry]; - if (G1HRRSUseSparseTable) { - bool res = _sparse_table.get_cards(from_hrs_ind, &cards[0]); - assert(res, "There should have been an entry"); - } - if (_n_fine_entries == _max_fine_entries) { prt = delete_region_table(); } else { @@ -676,10 +670,12 @@ _fine_grain_regions[ind] = prt; _n_fine_entries++; - // Add in the cards from the sparse table. if (G1HRRSUseSparseTable) { - for (int i = 0; i < SparsePRTEntry::CardsPerEntry; i++) { - CardIdx_t c = cards[i]; + // Transfer from sparse to fine-grain. + SparsePRTEntry *sprt_entry = _sparse_table.get_entry(from_hrs_ind); + assert(sprt_entry != NULL, "There should have been an entry"); + for (int i = 0; i < SparsePRTEntry::cards_num(); i++) { + CardIdx_t c = sprt_entry->card(i); if (c != SparsePRTEntry::NullEntry) { prt->add_card(c); } @@ -696,7 +692,21 @@ // OtherRegionsTable for why this is OK. assert(prt != NULL, "Inv"); - prt->add_reference(from, tid); + if (prt->should_expand(tid)) { + MutexLockerEx x(&_m, Mutex::_no_safepoint_check_flag); + HeapRegion* prt_hr = prt->hr(); + if (prt_hr == from_hr) { + // Make sure the table still corresponds to the same region + prt->par_expand(); + prt->add_reference(from, tid); + } + // else: The table has been concurrently coarsened, evicted, and + // the table data structure re-used for another table. So, we + // don't need to add the reference any more given that the table + // has been coarsened and the whole region will be scanned anyway. + } else { + prt->add_reference(from, tid); + } if (G1RecordHRRSOops) { HeapRegionRemSet::record(hr(), from); #if HRRS_VERBOSE @@ -1070,6 +1080,19 @@ {} +void HeapRegionRemSet::setup_remset_size() { + // Setup sparse and fine-grain tables sizes. + // table_size = base * (log(region_size / 1M) + 1) + int region_size_log_mb = MAX2((int)HeapRegion::LogOfHRGrainBytes - (int)LOG_M, 0); + if (FLAG_IS_DEFAULT(G1RSetSparseRegionEntries)) { + G1RSetSparseRegionEntries = G1RSetSparseRegionEntriesBase * (region_size_log_mb + 1); + } + if (FLAG_IS_DEFAULT(G1RSetRegionEntries)) { + G1RSetRegionEntries = G1RSetRegionEntriesBase * (region_size_log_mb + 1); + } + guarantee(G1RSetSparseRegionEntries > 0 && G1RSetRegionEntries > 0 , "Sanity"); +} + void HeapRegionRemSet::init_for_par_iteration() { _iter_state = Unclaimed; } @@ -1385,7 +1408,7 @@ os::sleep(Thread::current(), (jlong)5000, false); G1CollectedHeap* g1h = G1CollectedHeap::heap(); - // Run with "-XX:G1LogRSRegionEntries=2", so that 1 and 5 end up in same + // Run with "-XX:G1LogRSetRegionEntries=2", so that 1 and 5 end up in same // hash bucket. HeapRegion* hr0 = g1h->region_at(0); HeapRegion* hr1 = g1h->region_at(1);
--- a/src/share/vm/gc_implementation/g1/heapRegionRemSet.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/heapRegionRemSet.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -187,7 +187,8 @@ void clear_outgoing_entries(); enum ParIterState { Unclaimed, Claimed, Complete }; - ParIterState _iter_state; + volatile ParIterState _iter_state; + volatile jlong _iter_claimed; // Unused unless G1RecordHRRSOops is true. @@ -209,6 +210,7 @@ HeapRegion* hr); static int num_par_rem_sets(); + static void setup_remset_size(); HeapRegion* hr() const { return _other_regions.hr(); @@ -272,6 +274,19 @@ // Returns "true" iff the region's iteration is complete. bool iter_is_complete(); + // Support for claiming blocks of cards during iteration + void set_iter_claimed(size_t x) { _iter_claimed = (jlong)x; } + size_t iter_claimed() const { return (size_t)_iter_claimed; } + // Claim the next block of cards + size_t iter_claimed_next(size_t step) { + size_t current, next; + do { + current = iter_claimed(); + next = current + step; + } while (Atomic::cmpxchg((jlong)next, &_iter_claimed, (jlong)current) != (jlong)current); + return current; + } + // Initialize the given iterator to iterate over this rem set. void init_iterator(HeapRegionRemSetIterator* iter) const;
--- a/src/share/vm/gc_implementation/g1/ptrQueue.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/ptrQueue.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -64,8 +64,8 @@ while (_index == 0) { handle_zero_index(); } + assert(_index > 0, "postcondition"); - _index -= oopSize; _buf[byte_index_to_index((int)_index)] = ptr; assert(0 <= _index && _index <= _sz, "Invariant."); @@ -73,7 +73,12 @@ void PtrQueue::locking_enqueue_completed_buffer(void** buf) { assert(_lock->owned_by_self(), "Required."); + + // We have to unlock _lock (which may be Shared_DirtyCardQ_lock) before + // we acquire DirtyCardQ_CBL_mon inside enqeue_complete_buffer as they + // have the same rank and we may get the "possible deadlock" message _lock->unlock(); + qset()->enqueue_complete_buffer(buf); // We must relock only because the caller will unlock, for the normal // case. @@ -99,94 +104,139 @@ assert(_sz > 0, "Didn't set a buffer size."); MutexLockerEx x(_fl_owner->_fl_lock, Mutex::_no_safepoint_check_flag); if (_fl_owner->_buf_free_list != NULL) { - void** res = _fl_owner->_buf_free_list; - _fl_owner->_buf_free_list = (void**)_fl_owner->_buf_free_list[0]; + void** res = BufferNode::make_buffer_from_node(_fl_owner->_buf_free_list); + _fl_owner->_buf_free_list = _fl_owner->_buf_free_list->next(); _fl_owner->_buf_free_list_sz--; - // Just override the next pointer with NULL, just in case we scan this part - // of the buffer. - res[0] = NULL; return res; } else { - return NEW_C_HEAP_ARRAY(void*, _sz); + // Allocate space for the BufferNode in front of the buffer. + char *b = NEW_C_HEAP_ARRAY(char, _sz + BufferNode::aligned_size()); + return BufferNode::make_buffer_from_block(b); } } void PtrQueueSet::deallocate_buffer(void** buf) { assert(_sz > 0, "Didn't set a buffer size."); MutexLockerEx x(_fl_owner->_fl_lock, Mutex::_no_safepoint_check_flag); - buf[0] = (void*)_fl_owner->_buf_free_list; - _fl_owner->_buf_free_list = buf; + BufferNode *node = BufferNode::make_node_from_buffer(buf); + node->set_next(_fl_owner->_buf_free_list); + _fl_owner->_buf_free_list = node; _fl_owner->_buf_free_list_sz++; } void PtrQueueSet::reduce_free_list() { + assert(_fl_owner == this, "Free list reduction is allowed only for the owner"); // For now we'll adopt the strategy of deleting half. MutexLockerEx x(_fl_lock, Mutex::_no_safepoint_check_flag); size_t n = _buf_free_list_sz / 2; while (n > 0) { assert(_buf_free_list != NULL, "_buf_free_list_sz must be wrong."); - void** head = _buf_free_list; - _buf_free_list = (void**)_buf_free_list[0]; - FREE_C_HEAP_ARRAY(void*,head); + void* b = BufferNode::make_block_from_node(_buf_free_list); + _buf_free_list = _buf_free_list->next(); + FREE_C_HEAP_ARRAY(char, b); + _buf_free_list_sz --; n--; } } -void PtrQueueSet::enqueue_complete_buffer(void** buf, size_t index, bool ignore_max_completed) { - // I use explicit locking here because there's a bailout in the middle. - _cbl_mon->lock_without_safepoint_check(); +void PtrQueue::handle_zero_index() { + assert(0 == _index, "Precondition."); + // This thread records the full buffer and allocates a new one (while + // holding the lock if there is one). + if (_buf != NULL) { + if (_lock) { + assert(_lock->owned_by_self(), "Required."); + + // The current PtrQ may be the shared dirty card queue and + // may be being manipulated by more than one worker thread + // during a pause. Since the enqueuing of the completed + // buffer unlocks the Shared_DirtyCardQ_lock more than one + // worker thread can 'race' on reading the shared queue attributes + // (_buf and _index) and multiple threads can call into this + // routine for the same buffer. This will cause the completed + // buffer to be added to the CBL multiple times. - Thread* thread = Thread::current(); - assert( ignore_max_completed || - thread->is_Java_thread() || - SafepointSynchronize::is_at_safepoint(), - "invariant" ); - ignore_max_completed = ignore_max_completed || !thread->is_Java_thread(); + // We "claim" the current buffer by caching value of _buf in + // a local and clearing the field while holding _lock. When + // _lock is released (while enqueueing the completed buffer) + // the thread that acquires _lock will skip this code, + // preventing the subsequent the multiple enqueue, and + // install a newly allocated buffer below. + + void** buf = _buf; // local pointer to completed buffer + _buf = NULL; // clear shared _buf field + + locking_enqueue_completed_buffer(buf); // enqueue completed buffer + + // While the current thread was enqueuing the buffer another thread + // may have a allocated a new buffer and inserted it into this pointer + // queue. If that happens then we just return so that the current + // thread doesn't overwrite the buffer allocated by the other thread + // and potentially losing some dirtied cards. - if (!ignore_max_completed && _max_completed_queue > 0 && - _n_completed_buffers >= (size_t) _max_completed_queue) { - _cbl_mon->unlock(); - bool b = mut_process_buffer(buf); - if (b) { - deallocate_buffer(buf); - return; + if (_buf != NULL) return; + } else { + if (qset()->process_or_enqueue_complete_buffer(_buf)) { + // Recycle the buffer. No allocation. + _sz = qset()->buffer_size(); + _index = _sz; + return; + } } + } + // Reallocate the buffer + _buf = qset()->allocate_buffer(); + _sz = qset()->buffer_size(); + _index = _sz; + assert(0 <= _index && _index <= _sz, "Invariant."); +} - // Otherwise, go ahead and enqueue the buffer. Must reaquire the lock. - _cbl_mon->lock_without_safepoint_check(); +bool PtrQueueSet::process_or_enqueue_complete_buffer(void** buf) { + if (Thread::current()->is_Java_thread()) { + // We don't lock. It is fine to be epsilon-precise here. + if (_max_completed_queue == 0 || _max_completed_queue > 0 && + _n_completed_buffers >= _max_completed_queue + _completed_queue_padding) { + bool b = mut_process_buffer(buf); + if (b) { + // True here means that the buffer hasn't been deallocated and the caller may reuse it. + return true; + } + } } + // The buffer will be enqueued. The caller will have to get a new one. + enqueue_complete_buffer(buf); + return false; +} - // Here we still hold the _cbl_mon. - CompletedBufferNode* cbn = new CompletedBufferNode; - cbn->buf = buf; - cbn->next = NULL; - cbn->index = index; +void PtrQueueSet::enqueue_complete_buffer(void** buf, size_t index) { + MutexLockerEx x(_cbl_mon, Mutex::_no_safepoint_check_flag); + BufferNode* cbn = BufferNode::new_from_buffer(buf); + cbn->set_index(index); if (_completed_buffers_tail == NULL) { assert(_completed_buffers_head == NULL, "Well-formedness"); _completed_buffers_head = cbn; _completed_buffers_tail = cbn; } else { - _completed_buffers_tail->next = cbn; + _completed_buffers_tail->set_next(cbn); _completed_buffers_tail = cbn; } _n_completed_buffers++; - if (!_process_completed && + if (!_process_completed && _process_completed_threshold >= 0 && _n_completed_buffers >= _process_completed_threshold) { _process_completed = true; if (_notify_when_complete) - _cbl_mon->notify_all(); + _cbl_mon->notify(); } debug_only(assert_completed_buffer_list_len_correct_locked()); - _cbl_mon->unlock(); } int PtrQueueSet::completed_buffers_list_length() { int n = 0; - CompletedBufferNode* cbn = _completed_buffers_head; + BufferNode* cbn = _completed_buffers_head; while (cbn != NULL) { n++; - cbn = cbn->next; + cbn = cbn->next(); } return n; } @@ -197,7 +247,7 @@ } void PtrQueueSet::assert_completed_buffer_list_len_correct_locked() { - guarantee((size_t)completed_buffers_list_length() == _n_completed_buffers, + guarantee(completed_buffers_list_length() == _n_completed_buffers, "Completed buffer length is wrong."); } @@ -206,12 +256,8 @@ _sz = sz * oopSize; } -void PtrQueueSet::set_process_completed_threshold(size_t sz) { - _process_completed_threshold = sz; -} - -// Merge lists of buffers. Notify waiting threads if the length of the list -// exceeds threshold. The source queue is emptied as a result. The queues +// Merge lists of buffers. Notify the processing threads. +// The source queue is emptied as a result. The queues // must share the monitor. void PtrQueueSet::merge_bufferlists(PtrQueueSet *src) { assert(_cbl_mon == src->_cbl_mon, "Should share the same lock"); @@ -223,7 +269,7 @@ } else { assert(_completed_buffers_head != NULL, "Well formedness"); if (src->_completed_buffers_head != NULL) { - _completed_buffers_tail->next = src->_completed_buffers_head; + _completed_buffers_tail->set_next(src->_completed_buffers_head); _completed_buffers_tail = src->_completed_buffers_tail; } } @@ -236,31 +282,13 @@ assert(_completed_buffers_head == NULL && _completed_buffers_tail == NULL || _completed_buffers_head != NULL && _completed_buffers_tail != NULL, "Sanity"); - - if (!_process_completed && - _n_completed_buffers >= _process_completed_threshold) { - _process_completed = true; - if (_notify_when_complete) - _cbl_mon->notify_all(); - } } -// Merge free lists of the two queues. The free list of the source -// queue is emptied as a result. The queues must share the same -// mutex that guards free lists. -void PtrQueueSet::merge_freelists(PtrQueueSet* src) { - assert(_fl_lock == src->_fl_lock, "Should share the same lock"); - MutexLockerEx x(_fl_lock, Mutex::_no_safepoint_check_flag); - if (_buf_free_list != NULL) { - void **p = _buf_free_list; - while (*p != NULL) { - p = (void**)*p; - } - *p = src->_buf_free_list; - } else { - _buf_free_list = src->_buf_free_list; +void PtrQueueSet::notify_if_necessary() { + MutexLockerEx x(_cbl_mon, Mutex::_no_safepoint_check_flag); + if (_n_completed_buffers >= _process_completed_threshold || _max_completed_queue == 0) { + _process_completed = true; + if (_notify_when_complete) + _cbl_mon->notify(); } - _buf_free_list_sz += src->_buf_free_list_sz; - src->_buf_free_list = NULL; - src->_buf_free_list_sz = 0; }
--- a/src/share/vm/gc_implementation/g1/ptrQueue.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/ptrQueue.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -27,8 +27,10 @@ // the addresses of modified old-generation objects. This type supports // this operation. +// The definition of placement operator new(size_t, void*) in the <new>. +#include <new> + class PtrQueueSet; - class PtrQueue VALUE_OBJ_CLASS_SPEC { protected: @@ -77,7 +79,7 @@ else enqueue_known_active(ptr); } - inline void handle_zero_index(); + void handle_zero_index(); void locking_enqueue_completed_buffer(void** buf); void enqueue_known_active(void* ptr); @@ -126,34 +128,65 @@ }; +class BufferNode { + size_t _index; + BufferNode* _next; +public: + BufferNode() : _index(0), _next(NULL) { } + BufferNode* next() const { return _next; } + void set_next(BufferNode* n) { _next = n; } + size_t index() const { return _index; } + void set_index(size_t i) { _index = i; } + + // Align the size of the structure to the size of the pointer + static size_t aligned_size() { + static const size_t alignment = round_to(sizeof(BufferNode), sizeof(void*)); + return alignment; + } + + // BufferNode is allocated before the buffer. + // The chunk of memory that holds both of them is a block. + + // Produce a new BufferNode given a buffer. + static BufferNode* new_from_buffer(void** buf) { + return new (make_block_from_buffer(buf)) BufferNode; + } + + // The following are the required conversion routines: + static BufferNode* make_node_from_buffer(void** buf) { + return (BufferNode*)make_block_from_buffer(buf); + } + static void** make_buffer_from_node(BufferNode *node) { + return make_buffer_from_block(node); + } + static void* make_block_from_node(BufferNode *node) { + return (void*)node; + } + static void** make_buffer_from_block(void* p) { + return (void**)((char*)p + aligned_size()); + } + static void* make_block_from_buffer(void** p) { + return (void*)((char*)p - aligned_size()); + } +}; + // A PtrQueueSet represents resources common to a set of pointer queues. // In particular, the individual queues allocate buffers from this shared // set, and return completed buffers to the set. // All these variables are are protected by the TLOQ_CBL_mon. XXX ??? class PtrQueueSet VALUE_OBJ_CLASS_SPEC { - protected: - - class CompletedBufferNode: public CHeapObj { - public: - void** buf; - size_t index; - CompletedBufferNode* next; - CompletedBufferNode() : buf(NULL), - index(0), next(NULL){ } - }; - Monitor* _cbl_mon; // Protects the fields below. - CompletedBufferNode* _completed_buffers_head; - CompletedBufferNode* _completed_buffers_tail; - size_t _n_completed_buffers; - size_t _process_completed_threshold; + BufferNode* _completed_buffers_head; + BufferNode* _completed_buffers_tail; + int _n_completed_buffers; + int _process_completed_threshold; volatile bool _process_completed; // This (and the interpretation of the first element as a "next" // pointer) are protected by the TLOQ_FL_lock. Mutex* _fl_lock; - void** _buf_free_list; + BufferNode* _buf_free_list; size_t _buf_free_list_sz; // Queue set can share a freelist. The _fl_owner variable // specifies the owner. It is set to "this" by default. @@ -170,6 +203,7 @@ // Maximum number of elements allowed on completed queue: after that, // enqueuer does the work itself. Zero indicates no maximum. int _max_completed_queue; + int _completed_queue_padding; int completed_buffers_list_length(); void assert_completed_buffer_list_len_correct_locked(); @@ -191,9 +225,12 @@ // Because of init-order concerns, we can't pass these as constructor // arguments. void initialize(Monitor* cbl_mon, Mutex* fl_lock, - int max_completed_queue = 0, + int process_completed_threshold, + int max_completed_queue, PtrQueueSet *fl_owner = NULL) { _max_completed_queue = max_completed_queue; + _process_completed_threshold = process_completed_threshold; + _completed_queue_padding = 0; assert(cbl_mon != NULL && fl_lock != NULL, "Init order issue?"); _cbl_mon = cbl_mon; _fl_lock = fl_lock; @@ -208,14 +245,17 @@ void deallocate_buffer(void** buf); // Declares that "buf" is a complete buffer. - void enqueue_complete_buffer(void** buf, size_t index = 0, - bool ignore_max_completed = false); + void enqueue_complete_buffer(void** buf, size_t index = 0); + + // To be invoked by the mutator. + bool process_or_enqueue_complete_buffer(void** buf); bool completed_buffers_exist_dirty() { return _n_completed_buffers > 0; } bool process_completed_buffers() { return _process_completed; } + void set_process_completed(bool x) { _process_completed = x; } bool active() { return _all_active; } @@ -226,15 +266,24 @@ // Get the buffer size. size_t buffer_size() { return _sz; } - // Set the number of completed buffers that triggers log processing. - void set_process_completed_threshold(size_t sz); + // Get/Set the number of completed buffers that triggers log processing. + void set_process_completed_threshold(int sz) { _process_completed_threshold = sz; } + int process_completed_threshold() const { return _process_completed_threshold; } // Must only be called at a safe point. Indicates that the buffer free // list size may be reduced, if that is deemed desirable. void reduce_free_list(); - size_t completed_buffers_num() { return _n_completed_buffers; } + int completed_buffers_num() { return _n_completed_buffers; } void merge_bufferlists(PtrQueueSet* src); - void merge_freelists(PtrQueueSet* src); + + void set_max_completed_queue(int m) { _max_completed_queue = m; } + int max_completed_queue() { return _max_completed_queue; } + + void set_completed_queue_padding(int padding) { _completed_queue_padding = padding; } + int completed_queue_padding() { return _completed_queue_padding; } + + // Notify the consumer if the number of buffers crossed the threshold + void notify_if_necessary(); };
--- a/src/share/vm/gc_implementation/g1/ptrQueue.inline.hpp Mon Mar 15 15:51:36 2010 -0400 +++ /dev/null Thu Jan 01 00:00:00 1970 +0000 @@ -1,41 +0,0 @@ -/* - * Copyright 2001-2007 Sun Microsystems, Inc. All Rights Reserved. - * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. - * - * This code is free software; you can redistribute it and/or modify it - * under the terms of the GNU General Public License version 2 only, as - * published by the Free Software Foundation. - * - * This code is distributed in the hope that it will be useful, but WITHOUT - * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or - * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License - * version 2 for more details (a copy is included in the LICENSE file that - * accompanied this code). - * - * You should have received a copy of the GNU General Public License version - * 2 along with this work; if not, write to the Free Software Foundation, - * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. - * - * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, - * CA 95054 USA or visit www.sun.com if you need additional information or - * have any questions. - * - */ - -void PtrQueue::handle_zero_index() { - assert(0 == _index, "Precondition."); - // This thread records the full buffer and allocates a new one (while - // holding the lock if there is one). - void** buf = _buf; - _buf = qset()->allocate_buffer(); - _sz = qset()->buffer_size(); - _index = _sz; - assert(0 <= _index && _index <= _sz, "Invariant."); - if (buf != NULL) { - if (_lock) { - locking_enqueue_completed_buffer(buf); - } else { - qset()->enqueue_complete_buffer(buf); - } - } -}
--- a/src/share/vm/gc_implementation/g1/satbQueue.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/satbQueue.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -67,9 +67,9 @@ {} void SATBMarkQueueSet::initialize(Monitor* cbl_mon, Mutex* fl_lock, - int max_completed_queue, + int process_completed_threshold, Mutex* lock) { - PtrQueueSet::initialize(cbl_mon, fl_lock, max_completed_queue); + PtrQueueSet::initialize(cbl_mon, fl_lock, process_completed_threshold, -1); _shared_satb_queue.set_lock(lock); if (ParallelGCThreads > 0) { _par_closures = NEW_C_HEAP_ARRAY(ObjectClosure*, ParallelGCThreads); @@ -122,12 +122,12 @@ bool SATBMarkQueueSet::apply_closure_to_completed_buffer_work(bool par, int worker) { - CompletedBufferNode* nd = NULL; + BufferNode* nd = NULL; { MutexLockerEx x(_cbl_mon, Mutex::_no_safepoint_check_flag); if (_completed_buffers_head != NULL) { nd = _completed_buffers_head; - _completed_buffers_head = nd->next; + _completed_buffers_head = nd->next(); if (_completed_buffers_head == NULL) _completed_buffers_tail = NULL; _n_completed_buffers--; if (_n_completed_buffers == 0) _process_completed = false; @@ -135,9 +135,9 @@ } ObjectClosure* cl = (par ? _par_closures[worker] : _closure); if (nd != NULL) { - ObjPtrQueue::apply_closure_to_buffer(cl, nd->buf, 0, _sz); - deallocate_buffer(nd->buf); - delete nd; + void **buf = BufferNode::make_buffer_from_node(nd); + ObjPtrQueue::apply_closure_to_buffer(cl, buf, 0, _sz); + deallocate_buffer(buf); return true; } else { return false; @@ -145,13 +145,13 @@ } void SATBMarkQueueSet::abandon_partial_marking() { - CompletedBufferNode* buffers_to_delete = NULL; + BufferNode* buffers_to_delete = NULL; { MutexLockerEx x(_cbl_mon, Mutex::_no_safepoint_check_flag); while (_completed_buffers_head != NULL) { - CompletedBufferNode* nd = _completed_buffers_head; - _completed_buffers_head = nd->next; - nd->next = buffers_to_delete; + BufferNode* nd = _completed_buffers_head; + _completed_buffers_head = nd->next(); + nd->set_next(buffers_to_delete); buffers_to_delete = nd; } _completed_buffers_tail = NULL; @@ -159,10 +159,9 @@ DEBUG_ONLY(assert_completed_buffer_list_len_correct_locked()); } while (buffers_to_delete != NULL) { - CompletedBufferNode* nd = buffers_to_delete; - buffers_to_delete = nd->next; - deallocate_buffer(nd->buf); - delete nd; + BufferNode* nd = buffers_to_delete; + buffers_to_delete = nd->next(); + deallocate_buffer(BufferNode::make_buffer_from_node(nd)); } assert(SafepointSynchronize::is_at_safepoint(), "Must be at safepoint."); // So we can safely manipulate these queues.
--- a/src/share/vm/gc_implementation/g1/satbQueue.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/satbQueue.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -60,8 +60,8 @@ SATBMarkQueueSet(); void initialize(Monitor* cbl_mon, Mutex* fl_lock, - int max_completed_queue = 0, - Mutex* lock = NULL); + int process_completed_threshold, + Mutex* lock); static void handle_zero_index_for_thread(JavaThread* t);
--- a/src/share/vm/gc_implementation/g1/sparsePRT.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/sparsePRT.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -27,7 +27,7 @@ #define SPARSE_PRT_VERBOSE 0 -#define UNROLL_CARD_LOOPS 1 +#define UNROLL_CARD_LOOPS 1 void SparsePRT::init_iterator(SparsePRTIter* sprt_iter) { sprt_iter->init(this); @@ -36,27 +36,32 @@ void SparsePRTEntry::init(RegionIdx_t region_ind) { _region_ind = region_ind; _next_index = NullEntry; + #if UNROLL_CARD_LOOPS - assert(CardsPerEntry == 4, "Assumption. If changes, un-unroll."); - _cards[0] = NullEntry; - _cards[1] = NullEntry; - _cards[2] = NullEntry; - _cards[3] = NullEntry; + assert((cards_num() & (UnrollFactor - 1)) == 0, "Invalid number of cards in the entry"); + for (int i = 0; i < cards_num(); i += UnrollFactor) { + _cards[i] = NullEntry; + _cards[i + 1] = NullEntry; + _cards[i + 2] = NullEntry; + _cards[i + 3] = NullEntry; + } #else - for (int i = 0; i < CardsPerEntry; i++) + for (int i = 0; i < cards_num(); i++) _cards[i] = NullEntry; #endif } bool SparsePRTEntry::contains_card(CardIdx_t card_index) const { #if UNROLL_CARD_LOOPS - assert(CardsPerEntry == 4, "Assumption. If changes, un-unroll."); - if (_cards[0] == card_index) return true; - if (_cards[1] == card_index) return true; - if (_cards[2] == card_index) return true; - if (_cards[3] == card_index) return true; + assert((cards_num() & (UnrollFactor - 1)) == 0, "Invalid number of cards in the entry"); + for (int i = 0; i < cards_num(); i += UnrollFactor) { + if (_cards[i] == card_index || + _cards[i + 1] == card_index || + _cards[i + 2] == card_index || + _cards[i + 3] == card_index) return true; + } #else - for (int i = 0; i < CardsPerEntry; i++) { + for (int i = 0; i < cards_num(); i++) { if (_cards[i] == card_index) return true; } #endif @@ -67,14 +72,16 @@ int SparsePRTEntry::num_valid_cards() const { int sum = 0; #if UNROLL_CARD_LOOPS - assert(CardsPerEntry == 4, "Assumption. If changes, un-unroll."); - if (_cards[0] != NullEntry) sum++; - if (_cards[1] != NullEntry) sum++; - if (_cards[2] != NullEntry) sum++; - if (_cards[3] != NullEntry) sum++; + assert((cards_num() & (UnrollFactor - 1)) == 0, "Invalid number of cards in the entry"); + for (int i = 0; i < cards_num(); i += UnrollFactor) { + sum += (_cards[i] != NullEntry); + sum += (_cards[i + 1] != NullEntry); + sum += (_cards[i + 2] != NullEntry); + sum += (_cards[i + 3] != NullEntry); + } #else - for (int i = 0; i < CardsPerEntry; i++) { - if (_cards[i] != NulLEntry) sum++; + for (int i = 0; i < cards_num(); i++) { + sum += (_cards[i] != NullEntry); } #endif // Otherwise, we're full. @@ -83,27 +90,27 @@ SparsePRTEntry::AddCardResult SparsePRTEntry::add_card(CardIdx_t card_index) { #if UNROLL_CARD_LOOPS - assert(CardsPerEntry == 4, "Assumption. If changes, un-unroll."); - CardIdx_t c = _cards[0]; - if (c == card_index) return found; - if (c == NullEntry) { _cards[0] = card_index; return added; } - c = _cards[1]; - if (c == card_index) return found; - if (c == NullEntry) { _cards[1] = card_index; return added; } - c = _cards[2]; - if (c == card_index) return found; - if (c == NullEntry) { _cards[2] = card_index; return added; } - c = _cards[3]; - if (c == card_index) return found; - if (c == NullEntry) { _cards[3] = card_index; return added; } + assert((cards_num() & (UnrollFactor - 1)) == 0, "Invalid number of cards in the entry"); + CardIdx_t c; + for (int i = 0; i < cards_num(); i += UnrollFactor) { + c = _cards[i]; + if (c == card_index) return found; + if (c == NullEntry) { _cards[i] = card_index; return added; } + c = _cards[i + 1]; + if (c == card_index) return found; + if (c == NullEntry) { _cards[i + 1] = card_index; return added; } + c = _cards[i + 2]; + if (c == card_index) return found; + if (c == NullEntry) { _cards[i + 2] = card_index; return added; } + c = _cards[i + 3]; + if (c == card_index) return found; + if (c == NullEntry) { _cards[i + 3] = card_index; return added; } + } #else - for (int i = 0; i < CardsPerEntry; i++) { + for (int i = 0; i < cards_num(); i++) { CardIdx_t c = _cards[i]; if (c == card_index) return found; - if (c == NullEntry) { - _cards[i] = card_index; - return added; - } + if (c == NullEntry) { _cards[i] = card_index; return added; } } #endif // Otherwise, we're full. @@ -112,13 +119,15 @@ void SparsePRTEntry::copy_cards(CardIdx_t* cards) const { #if UNROLL_CARD_LOOPS - assert(CardsPerEntry == 4, "Assumption. If changes, un-unroll."); - cards[0] = _cards[0]; - cards[1] = _cards[1]; - cards[2] = _cards[2]; - cards[3] = _cards[3]; + assert((cards_num() & (UnrollFactor - 1)) == 0, "Invalid number of cards in the entry"); + for (int i = 0; i < cards_num(); i += UnrollFactor) { + cards[i] = _cards[i]; + cards[i + 1] = _cards[i + 1]; + cards[i + 2] = _cards[i + 2]; + cards[i + 3] = _cards[i + 3]; + } #else - for (int i = 0; i < CardsPerEntry; i++) { + for (int i = 0; i < cards_num(); i++) { cards[i] = _cards[i]; } #endif @@ -133,7 +142,7 @@ RSHashTable::RSHashTable(size_t capacity) : _capacity(capacity), _capacity_mask(capacity-1), _occupied_entries(0), _occupied_cards(0), - _entries(NEW_C_HEAP_ARRAY(SparsePRTEntry, capacity)), + _entries((SparsePRTEntry*)NEW_C_HEAP_ARRAY(char, SparsePRTEntry::size() * capacity)), _buckets(NEW_C_HEAP_ARRAY(int, capacity)), _free_list(NullEntry), _free_region(0) { @@ -161,8 +170,8 @@ "_capacity too large"); // This will put -1 == NullEntry in the key field of all entries. - memset(_entries, -1, _capacity * sizeof(SparsePRTEntry)); - memset(_buckets, -1, _capacity * sizeof(int)); + memset(_entries, NullEntry, _capacity * SparsePRTEntry::size()); + memset(_buckets, NullEntry, _capacity * sizeof(int)); _free_list = NullEntry; _free_region = 0; } @@ -175,8 +184,8 @@ if (res == SparsePRTEntry::added) _occupied_cards++; #if SPARSE_PRT_VERBOSE gclog_or_tty->print_cr(" after add_card[%d]: valid-cards = %d.", - pointer_delta(e, _entries, sizeof(SparsePRTEntry)), - e->num_valid_cards()); + pointer_delta(e, _entries, SparsePRTEntry::size()), + e->num_valid_cards()); #endif assert(e->num_valid_cards() > 0, "Postcondition"); return res != SparsePRTEntry::overflow; @@ -199,6 +208,22 @@ return true; } +SparsePRTEntry* RSHashTable::get_entry(RegionIdx_t region_ind) { + int ind = (int) (region_ind & capacity_mask()); + int cur_ind = _buckets[ind]; + SparsePRTEntry* cur; + while (cur_ind != NullEntry && + (cur = entry(cur_ind))->r_ind() != region_ind) { + cur_ind = cur->next_index(); + } + + if (cur_ind == NullEntry) return NULL; + // Otherwise... + assert(cur->r_ind() == region_ind, "Postcondition of loop + test above."); + assert(cur->num_valid_cards() > 0, "Inv"); + return cur; +} + bool RSHashTable::delete_entry(RegionIdx_t region_ind) { int ind = (int) (region_ind & capacity_mask()); int* prev_loc = &_buckets[ind]; @@ -225,20 +250,8 @@ int ind = (int) (region_ind & capacity_mask()); int cur_ind = _buckets[ind]; SparsePRTEntry* cur; - // XXX - // int k = 0; while (cur_ind != NullEntry && (cur = entry(cur_ind))->r_ind() != region_ind) { - /* - k++; - if (k > 10) { - gclog_or_tty->print_cr("RSHashTable::entry_for_region_ind(%d): " - "k = %d, cur_ind = %d.", region_ind, k, cur_ind); - if (k >= 1000) { - while (1) ; - } - } - */ cur_ind = cur->next_index(); } @@ -319,7 +332,7 @@ bool /* RSHashTable:: */ RSHashTableIter::has_next(size_t& card_index) { _card_ind++; CardIdx_t ci; - if (_card_ind < SparsePRTEntry::CardsPerEntry && + if (_card_ind < SparsePRTEntry::cards_num() && ((ci = _rsht->entry(_bl_ind)->card(_card_ind)) != SparsePRTEntry::NullEntry)) { card_index = compute_card_ind(ci); @@ -359,7 +372,7 @@ size_t RSHashTable::mem_size() const { return sizeof(this) + - capacity() * (sizeof(SparsePRTEntry) + sizeof(int)); + capacity() * (SparsePRTEntry::size() + sizeof(int)); } // ---------------------------------------------------------------------- @@ -446,6 +459,10 @@ return _next->get_cards(region_id, cards); } +SparsePRTEntry* SparsePRT::get_entry(RegionIdx_t region_id) { + return _next->get_entry(region_id); +} + bool SparsePRT::delete_entry(RegionIdx_t region_id) { return _next->delete_entry(region_id); }
--- a/src/share/vm/gc_implementation/g1/sparsePRT.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/sparsePRT.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -32,21 +32,28 @@ // insertions only enqueue old versions for deletions, but do not delete // old versions synchronously. - class SparsePRTEntry: public CHeapObj { public: - enum SomePublicConstants { - CardsPerEntry = 4, - NullEntry = -1 + NullEntry = -1, + UnrollFactor = 4 }; - private: RegionIdx_t _region_ind; int _next_index; - CardIdx_t _cards[CardsPerEntry]; - + CardIdx_t _cards[1]; + // WARNING: Don't put any data members beyond this line. Card array has, in fact, variable length. + // It should always be the last data member. public: + // Returns the size of the entry, used for entry allocation. + static size_t size() { return sizeof(SparsePRTEntry) + sizeof(CardIdx_t) * (cards_num() - 1); } + // Returns the size of the card array. + static int cards_num() { + // The number of cards should be a multiple of 4, because that's our current + // unrolling factor. + static const int s = MAX2<int>(G1RSetSparseRegionEntries & ~(UnrollFactor - 1), UnrollFactor); + return s; + } // Set the region_ind to the given value, and delete all cards. inline void init(RegionIdx_t region_ind); @@ -134,12 +141,15 @@ bool add_card(RegionIdx_t region_id, CardIdx_t card_index); bool get_cards(RegionIdx_t region_id, CardIdx_t* cards); + bool delete_entry(RegionIdx_t region_id); bool contains_card(RegionIdx_t region_id, CardIdx_t card_index) const; void add_entry(SparsePRTEntry* e); + SparsePRTEntry* get_entry(RegionIdx_t region_id); + void clear(); size_t capacity() const { return _capacity; } @@ -148,7 +158,7 @@ size_t occupied_cards() const { return _occupied_cards; } size_t mem_size() const; - SparsePRTEntry* entry(int i) const { return &_entries[i]; } + SparsePRTEntry* entry(int i) const { return (SparsePRTEntry*)((char*)_entries + SparsePRTEntry::size() * i); } void print(); }; @@ -157,7 +167,7 @@ class RSHashTableIter VALUE_OBJ_CLASS_SPEC { int _tbl_ind; // [-1, 0.._rsht->_capacity) int _bl_ind; // [-1, 0.._rsht->_capacity) - short _card_ind; // [0..CardsPerEntry) + short _card_ind; // [0..SparsePRTEntry::cards_num()) RSHashTable* _rsht; size_t _heap_bot_card_ind; @@ -176,7 +186,7 @@ RSHashTableIter(size_t heap_bot_card_ind) : _tbl_ind(RSHashTable::NullEntry), _bl_ind(RSHashTable::NullEntry), - _card_ind((SparsePRTEntry::CardsPerEntry-1)), + _card_ind((SparsePRTEntry::cards_num() - 1)), _rsht(NULL), _heap_bot_card_ind(heap_bot_card_ind) {} @@ -185,7 +195,7 @@ _rsht = rsht; _tbl_ind = -1; // So that first increment gets to 0. _bl_ind = RSHashTable::NullEntry; - _card_ind = (SparsePRTEntry::CardsPerEntry-1); + _card_ind = (SparsePRTEntry::cards_num() - 1); } bool has_next(size_t& card_index); @@ -241,9 +251,13 @@ // If the table hold an entry for "region_ind", Copies its // cards into "cards", which must be an array of length at least - // "CardsPerEntry", and returns "true"; otherwise, returns "false". + // "SparePRTEntry::cards_num()", and returns "true"; otherwise, + // returns "false". bool get_cards(RegionIdx_t region_ind, CardIdx_t* cards); + // Return the pointer to the entry associated with the given region. + SparsePRTEntry* get_entry(RegionIdx_t region_ind); + // If there is an entry for "region_ind", removes it and return "true"; // otherwise returns "false." bool delete_entry(RegionIdx_t region_ind);
--- a/src/share/vm/gc_implementation/g1/vm_operations_g1.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/vm_operations_g1.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -42,7 +42,7 @@ void VM_G1IncCollectionPause::doit() { JvmtiGCForAllocationMarker jgcm; G1CollectedHeap* g1h = G1CollectedHeap::heap(); - GCCauseSetter x(g1h, GCCause::_g1_inc_collection_pause); + GCCauseSetter x(g1h, _gc_cause); g1h->do_collection_pause_at_safepoint(); }
--- a/src/share/vm/gc_implementation/g1/vm_operations_g1.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/vm_operations_g1.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -68,8 +68,9 @@ class VM_G1IncCollectionPause: public VM_GC_Operation { public: - VM_G1IncCollectionPause(int gc_count_before) : - VM_GC_Operation(gc_count_before) {} + VM_G1IncCollectionPause(int gc_count_before, + GCCause::Cause gc_cause = GCCause::_g1_inc_collection_pause) : + VM_GC_Operation(gc_count_before) { _gc_cause = gc_cause; } virtual VMOp_Type type() const { return VMOp_G1IncCollectionPause; } virtual void doit(); virtual const char* name() const {
--- a/src/share/vm/gc_implementation/includeDB_gc_concurrentMarkSweep Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/includeDB_gc_concurrentMarkSweep Wed Mar 24 17:16:33 2010 -0700 @@ -221,6 +221,7 @@ freeList.cpp globals.hpp freeList.cpp mutex.hpp freeList.cpp sharedHeap.hpp +freeList.cpp vmThread.hpp freeList.hpp allocationStats.hpp
--- a/src/share/vm/gc_implementation/includeDB_gc_g1 Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/includeDB_gc_g1 Wed Mar 24 17:16:33 2010 -0700 @@ -109,7 +109,6 @@ dirtyCardQueue.cpp dirtyCardQueue.hpp dirtyCardQueue.cpp heapRegionRemSet.hpp dirtyCardQueue.cpp mutexLocker.hpp -dirtyCardQueue.cpp ptrQueue.inline.hpp dirtyCardQueue.cpp safepoint.hpp dirtyCardQueue.cpp thread.hpp dirtyCardQueue.cpp thread_<os_family>.inline.hpp @@ -222,6 +221,15 @@ g1MarkSweep.hpp timer.hpp g1MarkSweep.hpp universe.hpp +g1MemoryPool.cpp heapRegion.hpp +g1MemoryPool.cpp g1CollectedHeap.inline.hpp +g1MemoryPool.cpp g1CollectedHeap.hpp +g1MemoryPool.cpp g1CollectorPolicy.hpp +g1MemoryPool.cpp g1MemoryPool.hpp + +g1MemoryPool.hpp memoryUsage.hpp +g1MemoryPool.hpp memoryPool.hpp + g1OopClosures.inline.hpp concurrentMark.hpp g1OopClosures.inline.hpp g1OopClosures.hpp g1OopClosures.inline.hpp g1CollectedHeap.hpp @@ -303,12 +311,13 @@ klass.hpp g1OopClosures.hpp +memoryService.cpp g1MemoryPool.hpp + ptrQueue.cpp allocation.hpp ptrQueue.cpp allocation.inline.hpp ptrQueue.cpp mutex.hpp ptrQueue.cpp mutexLocker.hpp ptrQueue.cpp ptrQueue.hpp -ptrQueue.cpp ptrQueue.inline.hpp ptrQueue.cpp thread_<os_family>.inline.hpp ptrQueue.hpp allocation.hpp @@ -318,7 +327,6 @@ satbQueue.cpp allocation.inline.hpp satbQueue.cpp mutexLocker.hpp -satbQueue.cpp ptrQueue.inline.hpp satbQueue.cpp satbQueue.hpp satbQueue.cpp sharedHeap.hpp satbQueue.cpp thread.hpp
--- a/src/share/vm/gc_implementation/includeDB_gc_serial Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/includeDB_gc_serial Wed Mar 24 17:16:33 2010 -0700 @@ -71,6 +71,7 @@ gcUtil.hpp allocation.hpp gcUtil.hpp debug.hpp gcUtil.hpp globalDefinitions.hpp +gcUtil.hpp ostream.hpp gcUtil.hpp timer.hpp generationCounters.cpp generationCounters.hpp
--- a/src/share/vm/gc_implementation/parNew/parNewGeneration.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/parNew/parNewGeneration.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -50,6 +50,7 @@ work_queue_set_, &term_), _is_alive_closure(gen_), _scan_weak_ref_closure(gen_, this), _keep_alive_closure(&_scan_weak_ref_closure), + _promotion_failure_size(0), _pushes(0), _pops(0), _steals(0), _steal_attempts(0), _term_attempts(0), _strong_roots_time(0.0), _term_time(0.0) { @@ -249,6 +250,16 @@ } } +void ParScanThreadState::print_and_clear_promotion_failure_size() { + if (_promotion_failure_size != 0) { + if (PrintPromotionFailure) { + gclog_or_tty->print(" (%d: promotion failure size = " SIZE_FORMAT ") ", + _thread_num, _promotion_failure_size); + } + _promotion_failure_size = 0; + } +} + class ParScanThreadStateSet: private ResourceArray { public: // Initializes states for the specified number of threads; @@ -260,11 +271,11 @@ GrowableArray<oop>** overflow_stacks_, size_t desired_plab_sz, ParallelTaskTerminator& term); - inline ParScanThreadState& thread_sate(int i); + inline ParScanThreadState& thread_state(int i); int pushes() { return _pushes; } int pops() { return _pops; } int steals() { return _steals; } - void reset(); + void reset(bool promotion_failed); void flush(); private: ParallelTaskTerminator& _term; @@ -295,22 +306,31 @@ } } -inline ParScanThreadState& ParScanThreadStateSet::thread_sate(int i) +inline ParScanThreadState& ParScanThreadStateSet::thread_state(int i) { assert(i >= 0 && i < length(), "sanity check!"); return ((ParScanThreadState*)_data)[i]; } -void ParScanThreadStateSet::reset() +void ParScanThreadStateSet::reset(bool promotion_failed) { _term.reset_for_reuse(); + if (promotion_failed) { + for (int i = 0; i < length(); ++i) { + thread_state(i).print_and_clear_promotion_failure_size(); + } + } } void ParScanThreadStateSet::flush() { + // Work in this loop should be kept as lightweight as + // possible since this might otherwise become a bottleneck + // to scaling. Should we add heavy-weight work into this + // loop, consider parallelizing the loop into the worker threads. for (int i = 0; i < length(); ++i) { - ParScanThreadState& par_scan_state = thread_sate(i); + ParScanThreadState& par_scan_state = thread_state(i); // Flush stats related to To-space PLAB activity and // retire the last buffer. @@ -362,6 +382,14 @@ } } } + if (UseConcMarkSweepGC && ParallelGCThreads > 0) { + // We need to call this even when ResizeOldPLAB is disabled + // so as to avoid breaking some asserts. While we may be able + // to avoid this by reorganizing the code a bit, I am loathe + // to do that unless we find cases where ergo leads to bad + // performance. + CFLS_LAB::compute_desired_plab_size(); + } } ParScanClosure::ParScanClosure(ParNewGeneration* g, @@ -475,7 +503,7 @@ Generation* old_gen = gch->next_gen(_gen); - ParScanThreadState& par_scan_state = _state_set->thread_sate(i); + ParScanThreadState& par_scan_state = _state_set->thread_state(i); par_scan_state.set_young_old_boundary(_young_old_boundary); par_scan_state.start_strong_roots(); @@ -659,7 +687,7 @@ { ResourceMark rm; HandleMark hm; - ParScanThreadState& par_scan_state = _state_set.thread_sate(i); + ParScanThreadState& par_scan_state = _state_set.thread_state(i); par_scan_state.set_young_old_boundary(_young_old_boundary); _task.work(i, par_scan_state.is_alive_closure(), par_scan_state.keep_alive_closure(), @@ -693,7 +721,7 @@ ParNewRefProcTaskProxy rp_task(task, _generation, *_generation.next_gen(), _generation.reserved().end(), _state_set); workers->run_task(&rp_task); - _state_set.reset(); + _state_set.reset(_generation.promotion_failed()); } void ParNewRefProcTaskExecutor::execute(EnqueueTask& task) @@ -813,7 +841,7 @@ GenCollectedHeap::StrongRootsScope srs(gch); tsk.work(0); } - thread_state_set.reset(); + thread_state_set.reset(promotion_failed()); if (PAR_STATS_ENABLED && ParallelGCVerbose) { gclog_or_tty->print("Thread totals:\n" @@ -882,6 +910,8 @@ swap_spaces(); // Make life simpler for CMS || rescan; see 6483690. from()->set_next_compaction_space(to()); gch->set_incremental_collection_will_fail(); + // Inform the next generation that a promotion failure occurred. + _next_gen->promotion_failure_occurred(); // Reset the PromotionFailureALot counters. NOT_PRODUCT(Universe::heap()->reset_promotion_should_fail();) @@ -1029,6 +1059,8 @@ new_obj = old; preserve_mark_if_necessary(old, m); + // Log the size of the maiden promotion failure + par_scan_state->log_promotion_failure(sz); } old->forward_to(new_obj); @@ -1150,6 +1182,8 @@ failed_to_promote = true; preserve_mark_if_necessary(old, m); + // Log the size of the maiden promotion failure + par_scan_state->log_promotion_failure(sz); } } else { // Is in to-space; do copying ourselves.
--- a/src/share/vm/gc_implementation/parNew/parNewGeneration.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/parNew/parNewGeneration.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -97,6 +97,9 @@ int _pushes, _pops, _steals, _steal_attempts, _term_attempts; int _overflow_pushes, _overflow_refills, _overflow_refill_objs; + // Stats for promotion failure + size_t _promotion_failure_size; + // Timing numbers. double _start; double _start_strong_roots; @@ -169,6 +172,15 @@ // Undo the most recent allocation ("obj", of "word_sz"). void undo_alloc_in_to_space(HeapWord* obj, size_t word_sz); + // Promotion failure stats + size_t promotion_failure_size() { return promotion_failure_size(); } + void log_promotion_failure(size_t sz) { + if (_promotion_failure_size == 0) { + _promotion_failure_size = sz; + } + } + void print_and_clear_promotion_failure_size(); + int pushes() { return _pushes; } int pops() { return _pops; } int steals() { return _steals; }
--- a/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -51,6 +51,8 @@ } jint ParallelScavengeHeap::initialize() { + CollectedHeap::pre_initialize(); + // Cannot be initialized until after the flags are parsed GenerationSizer flag_parser; @@ -717,10 +719,6 @@ return young_gen()->allocate(size, true); } -void ParallelScavengeHeap::fill_all_tlabs(bool retire) { - CollectedHeap::fill_all_tlabs(retire); -} - void ParallelScavengeHeap::accumulate_statistics_all_tlabs() { CollectedHeap::accumulate_statistics_all_tlabs(); }
--- a/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -54,7 +54,6 @@ protected: static inline size_t total_invocations(); HeapWord* allocate_new_tlab(size_t size); - void fill_all_tlabs(bool retire); public: ParallelScavengeHeap() : CollectedHeap() { @@ -191,6 +190,10 @@ return true; } + virtual bool card_mark_must_follow_store() const { + return false; + } + // Return true if we don't we need a store barrier for // initializing stores to an object at this address. virtual bool can_elide_initializing_store_barrier(oop new_obj);
--- a/src/share/vm/gc_implementation/parallelScavenge/psGCAdaptivePolicyCounters.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/parallelScavenge/psGCAdaptivePolicyCounters.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -51,7 +51,7 @@ cname = PerfDataManager::counter_name(name_space(), "oldCapacity"); _old_capacity = PerfDataManager::create_variable(SUN_GC, cname, - PerfData::U_Bytes, (jlong) Arguments::initial_heap_size(), CHECK); + PerfData::U_Bytes, (jlong) InitialHeapSize, CHECK); cname = PerfDataManager::counter_name(name_space(), "boundaryMoved"); _boundary_moved = PerfDataManager::create_variable(SUN_GC, cname,
--- a/src/share/vm/gc_implementation/shared/allocationStats.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/shared/allocationStats.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -31,7 +31,7 @@ // beginning of this sweep: // Count(end_last_sweep) - Count(start_this_sweep) // + splitBirths(between) - splitDeaths(between) - // The above number divided by the time since the start [END???] of the + // The above number divided by the time since the end of the // previous sweep gives us a time rate of demand for blocks // of this size. We compute a padded average of this rate as // our current estimate for the time rate of demand for blocks @@ -41,7 +41,7 @@ // estimates. AdaptivePaddedAverage _demand_rate_estimate; - ssize_t _desired; // Estimate computed as described above + ssize_t _desired; // Demand stimate computed as described above ssize_t _coalDesired; // desired +/- small-percent for tuning coalescing ssize_t _surplus; // count - (desired +/- small-percent), @@ -53,9 +53,9 @@ ssize_t _coalDeaths; // loss from coalescing ssize_t _splitBirths; // additional chunks from splitting ssize_t _splitDeaths; // loss from splitting - size_t _returnedBytes; // number of bytes returned to list. + size_t _returnedBytes; // number of bytes returned to list. public: - void initialize() { + void initialize(bool split_birth = false) { AdaptivePaddedAverage* dummy = new (&_demand_rate_estimate) AdaptivePaddedAverage(CMS_FLSWeight, CMS_FLSPadding); @@ -67,7 +67,7 @@ _beforeSweep = 0; _coalBirths = 0; _coalDeaths = 0; - _splitBirths = 0; + _splitBirths = split_birth? 1 : 0; _splitDeaths = 0; _returnedBytes = 0; } @@ -75,10 +75,12 @@ AllocationStats() { initialize(); } + // The rate estimate is in blocks per second. void compute_desired(size_t count, float inter_sweep_current, - float inter_sweep_estimate) { + float inter_sweep_estimate, + float intra_sweep_estimate) { // If the latest inter-sweep time is below our granularity // of measurement, we may call in here with // inter_sweep_current == 0. However, even for suitably small @@ -88,12 +90,31 @@ // vulnerable to noisy glitches. In such cases, we // ignore the current sample and use currently available // historical estimates. + // XXX NEEDS TO BE FIXED + // assert(prevSweep() + splitBirths() >= splitDeaths() + (ssize_t)count, "Conservation Principle"); + // ^^^^^^^^^^^^^^^^^^^^^^^^^^^ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + // "Total Stock" "Not used at this block size" if (inter_sweep_current > _threshold) { - ssize_t demand = prevSweep() - count + splitBirths() - splitDeaths(); + ssize_t demand = prevSweep() - (ssize_t)count + splitBirths() - splitDeaths(); + // XXX NEEDS TO BE FIXED + // assert(demand >= 0, "Demand should be non-negative"); + // Defensive: adjust for imprecision in event counting + if (demand < 0) { + demand = 0; + } + float old_rate = _demand_rate_estimate.padded_average(); float rate = ((float)demand)/inter_sweep_current; _demand_rate_estimate.sample(rate); - _desired = (ssize_t)(_demand_rate_estimate.padded_average() - *inter_sweep_estimate); + float new_rate = _demand_rate_estimate.padded_average(); + ssize_t old_desired = _desired; + _desired = (ssize_t)(new_rate * (inter_sweep_estimate + + CMSExtrapolateSweep + ? intra_sweep_estimate + : 0.0)); + if (PrintFLSStatistics > 1) { + gclog_or_tty->print_cr("demand: %d, old_rate: %f, current_rate: %f, new_rate: %f, old_desired: %d, new_desired: %d", + demand, old_rate, rate, new_rate, old_desired, _desired); + } } }
--- a/src/share/vm/gc_implementation/shared/gcUtil.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/shared/gcUtil.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -52,11 +52,35 @@ _last_sample = new_sample; } +void AdaptiveWeightedAverage::print() const { + print_on(tty); +} + +void AdaptiveWeightedAverage::print_on(outputStream* st) const { + guarantee(false, "NYI"); +} + +void AdaptivePaddedAverage::print() const { + print_on(tty); +} + +void AdaptivePaddedAverage::print_on(outputStream* st) const { + guarantee(false, "NYI"); +} + +void AdaptivePaddedNoZeroDevAverage::print() const { + print_on(tty); +} + +void AdaptivePaddedNoZeroDevAverage::print_on(outputStream* st) const { + guarantee(false, "NYI"); +} + void AdaptivePaddedAverage::sample(float new_sample) { - // Compute our parent classes sample information + // Compute new adaptive weighted average based on new sample. AdaptiveWeightedAverage::sample(new_sample); - // Now compute the deviation and the new padded sample + // Now update the deviation and the padded average. float new_avg = average(); float new_dev = compute_adaptive_average(fabsd(new_sample - new_avg), deviation());
--- a/src/share/vm/gc_implementation/shared/gcUtil.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_implementation/shared/gcUtil.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -54,8 +54,8 @@ public: // Input weight must be between 0 and 100 - AdaptiveWeightedAverage(unsigned weight) : - _average(0.0), _sample_count(0), _weight(weight), _last_sample(0.0) { + AdaptiveWeightedAverage(unsigned weight, float avg = 0.0) : + _average(avg), _sample_count(0), _weight(weight), _last_sample(0.0) { } void clear() { @@ -64,6 +64,13 @@ _last_sample = 0; } + // Useful for modifying static structures after startup. + void modify(size_t avg, unsigned wt, bool force = false) { + assert(force, "Are you sure you want to call this?"); + _average = (float)avg; + _weight = wt; + } + // Accessors float average() const { return _average; } unsigned weight() const { return _weight; } @@ -83,6 +90,10 @@ // Convert to float and back to avoid integer overflow. return (size_t)exp_avg((float)avg, (float)sample, weight); } + + // Printing + void print_on(outputStream* st) const; + void print() const; }; @@ -129,6 +140,10 @@ // Override void sample(float new_sample); + + // Printing + void print_on(outputStream* st) const; + void print() const; }; // A weighted average that includes a deviation from the average, @@ -146,7 +161,12 @@ AdaptivePaddedAverage(weight, padding) {} // Override void sample(float new_sample); + + // Printing + void print_on(outputStream* st) const; + void print() const; }; + // Use a least squares fit to a set of data to generate a linear // equation. // y = intercept + slope * x
--- a/src/share/vm/gc_interface/collectedHeap.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_interface/collectedHeap.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -59,8 +59,19 @@ PerfDataManager::create_string_variable(SUN_GC, "lastCause", 80, GCCause::to_string(_gc_lastcause), CHECK); } + _defer_initial_card_mark = false; // strengthened by subclass in pre_initialize() below. } +void CollectedHeap::pre_initialize() { + // Used for ReduceInitialCardMarks (when COMPILER2 is used); + // otherwise remains unused. +#ifdef COMPLER2 + _defer_initial_card_mark = ReduceInitialCardMarks && can_elide_tlab_store_barriers() + && (DeferInitialCardMark || card_mark_must_follow_store()); +#else + assert(_defer_initial_card_mark == false, "Who would set it?"); +#endif +} #ifndef PRODUCT void CollectedHeap::check_for_bad_heap_word_value(HeapWord* addr, size_t size) { @@ -140,12 +151,13 @@ void CollectedHeap::flush_deferred_store_barrier(JavaThread* thread) { MemRegion deferred = thread->deferred_card_mark(); if (!deferred.is_empty()) { + assert(_defer_initial_card_mark, "Otherwise should be empty"); { // Verify that the storage points to a parsable object in heap DEBUG_ONLY(oop old_obj = oop(deferred.start());) assert(is_in(old_obj), "Not in allocated heap"); assert(!can_elide_initializing_store_barrier(old_obj), - "Else should have been filtered in defer_store_barrier()"); + "Else should have been filtered in new_store_pre_barrier()"); assert(!is_in_permanent(old_obj), "Sanity: not expected"); assert(old_obj->is_oop(true), "Not an oop"); assert(old_obj->is_parsable(), "Will not be concurrently parsable"); @@ -174,9 +186,7 @@ // so long as the card-mark is completed before the next // scavenge. For all these cases, we can do a card mark // at the point at which we do a slow path allocation -// in the old gen. For uniformity, however, we end -// up using the same scheme (see below) for all three -// cases (deferring the card-mark appropriately). +// in the old gen, i.e. in this call. // (b) GenCollectedHeap(ConcurrentMarkSweepGeneration) requires // in addition that the card-mark for an old gen allocated // object strictly follow any associated initializing stores. @@ -199,12 +209,13 @@ // but, like in CMS, because of the presence of concurrent refinement // (much like CMS' precleaning), must strictly follow the oop-store. // Thus, using the same protocol for maintaining the intended -// invariants turns out, serendepitously, to be the same for all -// three collectors/heap types above. +// invariants turns out, serendepitously, to be the same for both +// G1 and CMS. // -// For each future collector, this should be reexamined with -// that specific collector in mind. -oop CollectedHeap::defer_store_barrier(JavaThread* thread, oop new_obj) { +// For any future collector, this code should be reexamined with +// that specific collector in mind, and the documentation above suitably +// extended and updated. +oop CollectedHeap::new_store_pre_barrier(JavaThread* thread, oop new_obj) { // If a previous card-mark was deferred, flush it now. flush_deferred_store_barrier(thread); if (can_elide_initializing_store_barrier(new_obj)) { @@ -212,10 +223,17 @@ // following the flush above. assert(thread->deferred_card_mark().is_empty(), "Error"); } else { - // Remember info for the newly deferred store barrier - MemRegion deferred = MemRegion((HeapWord*)new_obj, new_obj->size()); - assert(!deferred.is_empty(), "Error"); - thread->set_deferred_card_mark(deferred); + MemRegion mr((HeapWord*)new_obj, new_obj->size()); + assert(!mr.is_empty(), "Error"); + if (_defer_initial_card_mark) { + // Defer the card mark + thread->set_deferred_card_mark(mr); + } else { + // Do the card mark + BarrierSet* bs = barrier_set(); + assert(bs->has_write_region_opt(), "No write_region() on BarrierSet"); + bs->write_region(mr); + } } return new_obj; } @@ -241,9 +259,9 @@ assert(Universe::heap()->is_in_reserved(start + words - 1), "not in heap"); } -void CollectedHeap::zap_filler_array(HeapWord* start, size_t words) +void CollectedHeap::zap_filler_array(HeapWord* start, size_t words, bool zap) { - if (ZapFillerObjects) { + if (ZapFillerObjects && zap) { Copy::fill_to_words(start + filler_array_hdr_size(), words - filler_array_hdr_size(), 0XDEAFBABE); } @@ -251,7 +269,7 @@ #endif // ASSERT void -CollectedHeap::fill_with_array(HeapWord* start, size_t words) +CollectedHeap::fill_with_array(HeapWord* start, size_t words, bool zap) { assert(words >= filler_array_min_size(), "too small for an array"); assert(words <= filler_array_max_size(), "too big for a single object"); @@ -262,31 +280,31 @@ // Set the length first for concurrent GC. ((arrayOop)start)->set_length((int)len); post_allocation_setup_common(Universe::intArrayKlassObj(), start, words); - DEBUG_ONLY(zap_filler_array(start, words);) + DEBUG_ONLY(zap_filler_array(start, words, zap);) } void -CollectedHeap::fill_with_object_impl(HeapWord* start, size_t words) +CollectedHeap::fill_with_object_impl(HeapWord* start, size_t words, bool zap) { assert(words <= filler_array_max_size(), "too big for a single object"); if (words >= filler_array_min_size()) { - fill_with_array(start, words); + fill_with_array(start, words, zap); } else if (words > 0) { assert(words == min_fill_size(), "unaligned size"); - post_allocation_setup_common(SystemDictionary::object_klass(), start, + post_allocation_setup_common(SystemDictionary::Object_klass(), start, words); } } -void CollectedHeap::fill_with_object(HeapWord* start, size_t words) +void CollectedHeap::fill_with_object(HeapWord* start, size_t words, bool zap) { DEBUG_ONLY(fill_args_check(start, words);) HandleMark hm; // Free handles before leaving. - fill_with_object_impl(start, words); + fill_with_object_impl(start, words, zap); } -void CollectedHeap::fill_with_objects(HeapWord* start, size_t words) +void CollectedHeap::fill_with_objects(HeapWord* start, size_t words, bool zap) { DEBUG_ONLY(fill_args_check(start, words);) HandleMark hm; // Free handles before leaving. @@ -299,13 +317,13 @@ const size_t max = filler_array_max_size(); while (words > max) { const size_t cur = words - max >= min ? max : max - min; - fill_with_array(start, cur); + fill_with_array(start, cur, zap); start += cur; words -= cur; } #endif - fill_with_object_impl(start, words); + fill_with_object_impl(start, words, zap); } HeapWord* CollectedHeap::allocate_new_tlab(size_t size) { @@ -313,22 +331,6 @@ return NULL; } -void CollectedHeap::fill_all_tlabs(bool retire) { - assert(UseTLAB, "should not reach here"); - // See note in ensure_parsability() below. - assert(SafepointSynchronize::is_at_safepoint() || - !is_init_completed(), - "should only fill tlabs at safepoint"); - // The main thread starts allocating via a TLAB even before it - // has added itself to the threads list at vm boot-up. - assert(Threads::first() != NULL, - "Attempt to fill tlabs before main thread has been added" - " to threads list is doomed to failure!"); - for(JavaThread *thread = Threads::first(); thread; thread = thread->next()) { - thread->tlab().make_parsable(retire); - } -} - void CollectedHeap::ensure_parsability(bool retire_tlabs) { // The second disjunct in the assertion below makes a concession // for the start-up verification done while the VM is being @@ -343,8 +345,24 @@ "Should only be called at a safepoint or at start-up" " otherwise concurrent mutator activity may make heap " " unparsable again"); - if (UseTLAB) { - fill_all_tlabs(retire_tlabs); + const bool use_tlab = UseTLAB; + const bool deferred = _defer_initial_card_mark; + // The main thread starts allocating via a TLAB even before it + // has added itself to the threads list at vm boot-up. + assert(!use_tlab || Threads::first() != NULL, + "Attempt to fill tlabs before main thread has been added" + " to threads list is doomed to failure!"); + for (JavaThread *thread = Threads::first(); thread; thread = thread->next()) { + if (use_tlab) thread->tlab().make_parsable(retire_tlabs); +#ifdef COMPILER2 + // The deferred store barriers must all have been flushed to the + // card-table (or other remembered set structure) before GC starts + // processing the card-table (or other remembered set). + if (deferred) flush_deferred_store_barrier(thread); +#else + assert(!deferred, "Should be false"); + assert(thread->deferred_card_mark().is_empty(), "Should be empty"); +#endif } }
--- a/src/share/vm/gc_interface/collectedHeap.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/gc_interface/collectedHeap.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -51,6 +51,9 @@ // Used for filler objects (static, but initialized in ctor). static size_t _filler_array_max_size; + // Used in support of ReduceInitialCardMarks; only consulted if COMPILER2 is being used + bool _defer_initial_card_mark; + protected: MemRegion _reserved; BarrierSet* _barrier_set; @@ -70,13 +73,16 @@ // Constructor CollectedHeap(); + // Do common initializations that must follow instance construction, + // for example, those needing virtual calls. + // This code could perhaps be moved into initialize() but would + // be slightly more awkward because we want the latter to be a + // pure virtual. + void pre_initialize(); + // Create a new tlab virtual HeapWord* allocate_new_tlab(size_t size); - // Fix up tlabs to make the heap well-formed again, - // optionally retiring the tlabs. - virtual void fill_all_tlabs(bool retire); - // Accumulate statistics on all tlabs. virtual void accumulate_statistics_all_tlabs(); @@ -127,14 +133,14 @@ static inline size_t filler_array_max_size(); DEBUG_ONLY(static void fill_args_check(HeapWord* start, size_t words);) - DEBUG_ONLY(static void zap_filler_array(HeapWord* start, size_t words);) + DEBUG_ONLY(static void zap_filler_array(HeapWord* start, size_t words, bool zap = true);) // Fill with a single array; caller must ensure filler_array_min_size() <= // words <= filler_array_max_size(). - static inline void fill_with_array(HeapWord* start, size_t words); + static inline void fill_with_array(HeapWord* start, size_t words, bool zap = true); // Fill with a single object (either an int array or a java.lang.Object). - static inline void fill_with_object_impl(HeapWord* start, size_t words); + static inline void fill_with_object_impl(HeapWord* start, size_t words, bool zap = true); // Verification functions virtual void check_for_bad_heap_word_value(HeapWord* addr, size_t size) @@ -338,14 +344,14 @@ return size_t(align_object_size(oopDesc::header_size())); } - static void fill_with_objects(HeapWord* start, size_t words); + static void fill_with_objects(HeapWord* start, size_t words, bool zap = true); - static void fill_with_object(HeapWord* start, size_t words); - static void fill_with_object(MemRegion region) { - fill_with_object(region.start(), region.word_size()); + static void fill_with_object(HeapWord* start, size_t words, bool zap = true); + static void fill_with_object(MemRegion region, bool zap = true) { + fill_with_object(region.start(), region.word_size(), zap); } - static void fill_with_object(HeapWord* start, HeapWord* end) { - fill_with_object(start, pointer_delta(end, start)); + static void fill_with_object(HeapWord* start, HeapWord* end, bool zap = true) { + fill_with_object(start, pointer_delta(end, start), zap); } // Some heaps may offer a contiguous region for shared non-blocking @@ -431,14 +437,25 @@ // promises to call this function on such a slow-path-allocated // object before performing initializations that have elided // store barriers. Returns new_obj, or maybe a safer copy thereof. - virtual oop defer_store_barrier(JavaThread* thread, oop new_obj); + virtual oop new_store_pre_barrier(JavaThread* thread, oop new_obj); // Answers whether an initializing store to a new object currently - // allocated at the given address doesn't need a (deferred) store + // allocated at the given address doesn't need a store // barrier. Returns "true" if it doesn't need an initializing // store barrier; answers "false" if it does. virtual bool can_elide_initializing_store_barrier(oop new_obj) = 0; + // If a compiler is eliding store barriers for TLAB-allocated objects, + // we will be informed of a slow-path allocation by a call + // to new_store_pre_barrier() above. Such a call precedes the + // initialization of the object itself, and no post-store-barriers will + // be issued. Some heap types require that the barrier strictly follows + // the initializing stores. (This is currently implemented by deferring the + // barrier until the next slow-path allocation or gc-related safepoint.) + // This interface answers whether a particular heap type needs the card + // mark to be thus strictly sequenced after the stores. + virtual bool card_mark_must_follow_store() const = 0; + // If the CollectedHeap was asked to defer a store barrier above, // this informs it to flush such a deferred store barrier to the // remembered set.
--- a/src/share/vm/includeDB_compiler2 Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/includeDB_compiler2 Wed Mar 24 17:16:33 2010 -0700 @@ -149,11 +149,15 @@ c2compiler.hpp abstractCompiler.hpp callGenerator.cpp addnode.hpp +callGenerator.cpp bcEscapeAnalyzer.hpp callGenerator.cpp callGenerator.hpp callGenerator.cpp callnode.hpp callGenerator.cpp cfgnode.hpp callGenerator.cpp compileLog.hpp callGenerator.cpp connode.hpp +callGenerator.cpp ciCPCache.hpp +callGenerator.cpp ciMethodHandle.hpp +callGenerator.cpp javaClasses.hpp callGenerator.cpp parse.hpp callGenerator.cpp rootnode.hpp callGenerator.cpp runtime.hpp @@ -321,6 +325,7 @@ compile.cpp rootnode.hpp compile.cpp runtime.hpp compile.cpp signature.hpp +compile.cpp stringopts.hpp compile.cpp stubRoutines.hpp compile.cpp systemDictionary.hpp compile.cpp timer.hpp @@ -389,6 +394,9 @@ doCall.cpp addnode.hpp doCall.cpp callGenerator.hpp +doCall.cpp ciCallSite.hpp +doCall.cpp ciCPCache.hpp +doCall.cpp ciMethodHandle.hpp doCall.cpp cfgnode.hpp doCall.cpp compileLog.hpp doCall.cpp linkResolver.hpp @@ -476,12 +484,16 @@ graphKit.cpp runtime.hpp graphKit.cpp sharedRuntime.hpp +graphKit.hpp addnode.hpp graphKit.hpp callnode.hpp graphKit.hpp cfgnode.hpp graphKit.hpp ciEnv.hpp +graphKit.hpp divnode.hpp graphKit.hpp compile.hpp graphKit.hpp deoptimization.hpp graphKit.hpp phaseX.hpp +graphKit.hpp mulnode.hpp +graphKit.hpp subnode.hpp graphKit.hpp type.hpp idealKit.cpp addnode.hpp @@ -490,7 +502,10 @@ idealKit.cpp idealKit.hpp idealKit.cpp runtime.hpp +idealKit.hpp addnode.hpp +idealKit.hpp cfgnode.hpp idealKit.hpp connode.hpp +idealKit.hpp divnode.hpp idealKit.hpp mulnode.hpp idealKit.hpp phaseX.hpp idealKit.hpp subnode.hpp @@ -586,6 +601,7 @@ loopTransform.cpp addnode.hpp loopTransform.cpp allocation.inline.hpp +loopTransform.cpp callnode.hpp loopTransform.cpp connode.hpp loopTransform.cpp compileLog.hpp loopTransform.cpp divnode.hpp @@ -641,6 +657,7 @@ macro.cpp callnode.hpp macro.cpp cfgnode.hpp macro.cpp compile.hpp +macro.cpp compileLog.hpp macro.cpp connode.hpp macro.cpp locknode.hpp macro.cpp loopnode.hpp @@ -758,6 +775,7 @@ output.cpp assembler.inline.hpp output.cpp callnode.hpp output.cpp cfgnode.hpp +output.cpp compileBroker.hpp output.cpp debugInfo.hpp output.cpp debugInfoRec.hpp output.cpp handles.inline.hpp @@ -993,6 +1011,21 @@ split_if.cpp connode.hpp split_if.cpp loopnode.hpp +stringopts.hpp phaseX.hpp +stringopts.hpp node.hpp + +stringopts.cpp addnode.hpp +stringopts.cpp callnode.hpp +stringopts.cpp callGenerator.hpp +stringopts.cpp compileLog.hpp +stringopts.cpp divnode.hpp +stringopts.cpp idealKit.hpp +stringopts.cpp graphKit.hpp +stringopts.cpp rootnode.hpp +stringopts.cpp runtime.hpp +stringopts.cpp subnode.hpp +stringopts.cpp stringopts.hpp + stubGenerator_<arch_model>.cpp runtime.hpp stubRoutines.cpp runtime.hpp
--- a/src/share/vm/includeDB_core Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/includeDB_core Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ // -// Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. +// Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. // DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. // // This code is free software; you can redistribute it and/or modify it @@ -175,6 +175,7 @@ arguments.cpp management.hpp arguments.cpp oop.inline.hpp arguments.cpp os_<os_family>.inline.hpp +arguments.cpp referenceProcessor.hpp arguments.cpp universe.inline.hpp arguments.cpp vm_version_<arch>.hpp @@ -289,7 +290,7 @@ attachListener.hpp debug.hpp attachListener.hpp ostream.hpp -barrierSet.cpp barrierSet.hpp +barrierSet.cpp barrierSet.inline.hpp barrierSet.cpp collectedHeap.hpp barrierSet.cpp universe.hpp @@ -516,6 +517,11 @@ ciCallProfile.hpp ciClassList.hpp +ciCallSite.cpp ciCallSite.hpp +ciCallSite.cpp ciUtilities.hpp + +ciCallSite.hpp ciInstance.hpp + ciConstant.cpp allocation.hpp ciConstant.cpp allocation.inline.hpp ciConstant.cpp ciConstant.hpp @@ -532,6 +538,12 @@ ciConstantPoolCache.hpp growableArray.hpp ciConstantPoolCache.hpp resourceArea.hpp +ciCPCache.cpp cpCacheOop.hpp +ciCPCache.cpp ciCPCache.hpp + +ciCPCache.hpp ciClassList.hpp +ciCPCache.hpp ciObject.hpp + ciEnv.cpp allocation.inline.hpp ciEnv.cpp ciConstant.hpp ciEnv.cpp ciEnv.hpp @@ -570,6 +582,7 @@ ciEnv.hpp dependencies.hpp ciEnv.hpp exceptionHandlerTable.hpp ciEnv.hpp oopMap.hpp +ciEnv.hpp systemDictionary.hpp ciEnv.hpp thread.hpp ciExceptionHandler.cpp ciExceptionHandler.hpp @@ -592,6 +605,7 @@ ciField.hpp ciClassList.hpp ciField.hpp ciConstant.hpp ciField.hpp ciFlags.hpp +ciField.hpp ciInstance.hpp ciFlags.cpp ciFlags.hpp @@ -678,6 +692,7 @@ ciMethod.hpp ciInstanceKlass.hpp ciMethod.hpp ciObject.hpp ciMethod.hpp ciSignature.hpp +ciMethod.hpp methodHandles.hpp ciMethod.hpp methodLiveness.hpp ciMethodBlocks.cpp bytecode.hpp @@ -709,6 +724,15 @@ ciMethodKlass.hpp ciKlass.hpp ciMethodKlass.hpp ciSymbol.hpp +ciMethodHandle.cpp ciClassList.hpp +ciMethodHandle.cpp ciInstance.hpp +ciMethodHandle.cpp ciMethodHandle.hpp +ciMethodHandle.cpp ciUtilities.hpp +ciMethodHandle.cpp methodHandles.hpp +ciMethodHandle.cpp methodHandleWalk.hpp + +ciMethodHandle.hpp methodHandles.hpp + ciNullObject.cpp ciNullObject.hpp ciNullObject.hpp ciClassList.hpp @@ -754,11 +778,14 @@ ciObject.hpp jniHandles.hpp ciObjectFactory.cpp allocation.inline.hpp +ciObjectFactory.cpp ciCallSite.hpp +ciObjectFactory.cpp ciCPCache.hpp ciObjectFactory.cpp ciInstance.hpp ciObjectFactory.cpp ciInstanceKlass.hpp ciObjectFactory.cpp ciInstanceKlassKlass.hpp ciObjectFactory.cpp ciMethod.hpp ciObjectFactory.cpp ciMethodData.hpp +ciObjectFactory.cpp ciMethodHandle.hpp ciObjectFactory.cpp ciMethodKlass.hpp ciObjectFactory.cpp ciNullObject.hpp ciObjectFactory.cpp ciObjArray.hpp @@ -792,6 +819,7 @@ ciSignature.hpp globalDefinitions.hpp ciSignature.hpp growableArray.hpp +ciStreams.cpp ciCallSite.hpp ciStreams.cpp ciConstant.hpp ciStreams.cpp ciField.hpp ciStreams.cpp ciStreams.hpp @@ -894,6 +922,7 @@ classLoader.cpp allocation.inline.hpp classLoader.cpp arguments.hpp +classLoader.cpp bytecodeStream.hpp classLoader.cpp classFileParser.hpp classLoader.cpp classFileStream.hpp classLoader.cpp classLoader.hpp @@ -921,6 +950,7 @@ classLoader.cpp management.hpp classLoader.cpp oop.inline.hpp classLoader.cpp oopFactory.hpp +classLoader.cpp oopMapCache.hpp classLoader.cpp os_<os_family>.inline.hpp classLoader.cpp symbolOop.hpp classLoader.cpp systemDictionary.hpp @@ -1003,6 +1033,7 @@ codeCache.cpp oop.inline.hpp codeCache.cpp pcDesc.hpp codeCache.cpp resourceArea.hpp +codeCache.cpp xmlstream.hpp codeCache.hpp allocation.hpp codeCache.hpp codeBlob.hpp @@ -1091,6 +1122,7 @@ compileBroker.cpp oop.inline.hpp compileBroker.cpp os.hpp compileBroker.cpp sharedRuntime.hpp +compileBroker.cpp sweeper.hpp compileBroker.cpp systemDictionary.hpp compileBroker.cpp vmSymbols.hpp @@ -1291,6 +1323,7 @@ cpCacheOop.cpp markSweep.inline.hpp cpCacheOop.cpp objArrayOop.hpp cpCacheOop.cpp oop.inline.hpp +cpCacheOop.cpp rewriter.hpp cpCacheOop.cpp universe.inline.hpp cpCacheOop.hpp allocation.hpp @@ -1451,6 +1484,7 @@ deoptimization.cpp vframe.hpp deoptimization.cpp vframeArray.hpp deoptimization.cpp vframe_hp.hpp +deoptimization.cpp vmreg_<arch>.inline.hpp deoptimization.cpp xmlstream.hpp deoptimization.hpp allocation.hpp @@ -1497,6 +1531,7 @@ disassembler.cpp fprofiler.hpp disassembler.cpp handles.inline.hpp disassembler.cpp hpi.hpp +disassembler.cpp javaClasses.hpp disassembler.cpp stubCodeGenerator.hpp disassembler.cpp stubRoutines.hpp @@ -2501,6 +2536,7 @@ jvmtiExport.hpp handles.hpp jvmtiExport.hpp iterator.hpp jvmtiExport.hpp jvmti.h +jvmtiExport.hpp jvmticmlr.h jvmtiExport.hpp oop.hpp jvmtiExport.hpp oopsHierarchy.hpp @@ -2619,6 +2655,7 @@ loaderConstraints.cpp safepoint.hpp loaderConstraints.hpp dictionary.hpp +loaderConstraints.hpp placeholders.hpp loaderConstraints.hpp hashtable.hpp location.cpp debugInfo.hpp @@ -2812,6 +2849,12 @@ methodDataOop.hpp orderAccess.hpp methodDataOop.hpp universe.hpp +methodHandleWalk.hpp methodHandles.hpp + +methodHandleWalk.cpp methodHandleWalk.hpp +methodHandleWalk.cpp oopFactory.hpp +methodHandleWalk.cpp rewriter.hpp + methodHandles.hpp frame.inline.hpp methodHandles.hpp globals.hpp methodHandles.hpp interfaceSupport.hpp @@ -3469,6 +3512,7 @@ reflection.cpp javaClasses.hpp reflection.cpp jvm.h reflection.cpp linkResolver.hpp +reflection.cpp methodHandleWalk.hpp reflection.cpp objArrayKlass.hpp reflection.cpp objArrayOop.hpp reflection.cpp oopFactory.hpp @@ -3680,6 +3724,7 @@ sharedRuntime.cpp abstractCompiler.hpp sharedRuntime.cpp arguments.hpp sharedRuntime.cpp biasedLocking.hpp +sharedRuntime.cpp compileBroker.hpp sharedRuntime.cpp compiledIC.hpp sharedRuntime.cpp compilerOracle.hpp sharedRuntime.cpp copy.hpp @@ -3688,6 +3733,7 @@ sharedRuntime.cpp forte.hpp sharedRuntime.cpp gcLocker.inline.hpp sharedRuntime.cpp handles.inline.hpp +sharedRuntime.cpp hashtable.inline.hpp sharedRuntime.cpp init.hpp sharedRuntime.cpp interfaceSupport.hpp sharedRuntime.cpp interpreterRuntime.hpp @@ -3715,6 +3761,7 @@ sharedRuntime.hpp allocation.hpp sharedRuntime.hpp bytecodeHistogram.hpp sharedRuntime.hpp bytecodeTracer.hpp +sharedRuntime.hpp hashtable.hpp sharedRuntime.hpp linkResolver.hpp sharedRuntime.hpp resourceArea.hpp sharedRuntime.hpp threadLocalStorage.hpp @@ -3932,6 +3979,7 @@ sweeper.cpp atomic.hpp sweeper.cpp codeCache.hpp +sweeper.cpp compileBroker.hpp sweeper.cpp events.hpp sweeper.cpp methodOop.hpp sweeper.cpp mutexLocker.hpp @@ -3939,6 +3987,8 @@ sweeper.cpp os.hpp sweeper.cpp resourceArea.hpp sweeper.cpp sweeper.hpp +sweeper.cpp vm_operations.hpp +sweeper.cpp xmlstream.hpp symbolKlass.cpp gcLocker.hpp symbolKlass.cpp handles.inline.hpp @@ -4592,6 +4642,7 @@ vm_operations.cpp interfaceSupport.hpp vm_operations.cpp isGCActiveMark.hpp vm_operations.cpp resourceArea.hpp +vm_operations.cpp sweeper.hpp vm_operations.cpp threadService.hpp vm_operations.cpp thread_<os_family>.inline.hpp vm_operations.cpp vmSymbols.hpp
--- a/src/share/vm/includeDB_gc_parallel Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/includeDB_gc_parallel Wed Mar 24 17:16:33 2010 -0700 @@ -21,6 +21,8 @@ // have any questions. // +arguments.cpp compactibleFreeListSpace.hpp + assembler_<arch>.cpp g1SATBCardTableModRefBS.hpp assembler_<arch>.cpp g1CollectedHeap.inline.hpp assembler_<arch>.cpp heapRegion.hpp
--- a/src/share/vm/interpreter/abstractInterpreter.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/interpreter/abstractInterpreter.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -109,6 +109,8 @@ static void print_method_kind(MethodKind kind) PRODUCT_RETURN; + static bool can_be_compiled(methodHandle m); + // Runtime support // length = invoke bytecode length (to advance to next bytecode)
--- a/src/share/vm/interpreter/bytecode.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/interpreter/bytecode.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -102,7 +102,9 @@ KlassHandle resolved_klass; constantPoolHandle constants(THREAD, _method->constants()); - if (adjusted_invoke_code() != Bytecodes::_invokeinterface) { + if (adjusted_invoke_code() == Bytecodes::_invokedynamic) { + LinkResolver::resolve_dynamic_method(m, resolved_klass, constants, index(), CHECK_(methodHandle())); + } else if (adjusted_invoke_code() != Bytecodes::_invokeinterface) { LinkResolver::resolve_method(m, resolved_klass, constants, index(), CHECK_(methodHandle())); } else { LinkResolver::resolve_interface_method(m, resolved_klass, constants, index(), CHECK_(methodHandle()));
--- a/src/share/vm/interpreter/bytecode.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/interpreter/bytecode.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -205,12 +205,14 @@ bool is_invokespecial() const { return adjusted_invoke_code() == Bytecodes::_invokespecial; } bool is_invokedynamic() const { return adjusted_invoke_code() == Bytecodes::_invokedynamic; } + bool has_receiver() const { return !is_invokestatic() && !is_invokedynamic(); } bool has_giant_index() const { return is_invokedynamic(); } bool is_valid() const { return is_invokeinterface() || is_invokevirtual() || is_invokestatic() || - is_invokespecial(); } + is_invokespecial() || + is_invokedynamic(); } // Creation inline friend Bytecode_invoke* Bytecode_invoke_at(methodHandle method, int bci);
--- a/src/share/vm/interpreter/bytecodeInterpreter.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/interpreter/bytecodeInterpreter.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -281,7 +281,7 @@ #define DO_BACKEDGE_CHECKS(skip, branch_pc) \ if ((skip) <= 0) { \ - if (UseCompiler && UseLoopCounter) { \ + if (UseLoopCounter) { \ bool do_OSR = UseOnStackReplacement; \ BACKEDGE_COUNT->increment(); \ if (do_OSR) do_OSR = BACKEDGE_COUNT->reached_InvocationLimit(); \ @@ -289,16 +289,12 @@ nmethod* osr_nmethod; \ OSR_REQUEST(osr_nmethod, branch_pc); \ if (osr_nmethod != NULL && osr_nmethod->osr_entry_bci() != InvalidOSREntryBci) { \ - intptr_t* buf; \ - CALL_VM(buf=SharedRuntime::OSR_migration_begin(THREAD), handle_exception); \ + intptr_t* buf = SharedRuntime::OSR_migration_begin(THREAD); \ istate->set_msg(do_osr); \ istate->set_osr_buf((address)buf); \ istate->set_osr_entry(osr_nmethod->osr_entry()); \ return; \ } \ - } else { \ - INCR_INVOCATION_COUNT; \ - SAFEPOINT; \ } \ } /* UseCompiler ... */ \ INCR_INVOCATION_COUNT; \ @@ -1281,12 +1277,7 @@ jfloat f; jdouble r; f = STACK_FLOAT(-1); -#ifdef IA64 - // IA64 gcc bug - r = ( f == 0.0f ) ? (jdouble) f : (jdouble) f + ia64_double_zero; -#else r = (jdouble) f; -#endif MORE_STACK(-1); // POP SET_STACK_DOUBLE(r, 1); UPDATE_PC_AND_TOS_AND_CONTINUE(1, 2);
--- a/src/share/vm/interpreter/bytecodeTracer.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/interpreter/bytecodeTracer.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -270,6 +270,8 @@ st->print_cr(" %s", constants->resolved_klass_at(i)->klass_part()->external_name()); } else if (tag.is_unresolved_klass()) { st->print_cr(" <unresolved klass at %d>", i); + } else if (tag.is_object()) { + st->print_cr(" " PTR_FORMAT, constants->object_at(i)); } else { st->print_cr(" bad tag=%d at %d", tag.value(), i); } @@ -282,18 +284,21 @@ constantPoolOop constants = method()->constants(); constantTag tag = constants->tag_at(i); + int nt_index = -1; + switch (tag.value()) { case JVM_CONSTANT_InterfaceMethodref: case JVM_CONSTANT_Methodref: case JVM_CONSTANT_Fieldref: + case JVM_CONSTANT_NameAndType: break; default: st->print_cr(" bad tag=%d at %d", tag.value(), i); return; } - symbolOop name = constants->name_ref_at(orig_i); - symbolOop signature = constants->signature_ref_at(orig_i); + symbolOop name = constants->uncached_name_ref_at(i); + symbolOop signature = constants->uncached_signature_ref_at(i); st->print_cr(" %d <%s> <%s> ", i, name->as_C_string(), signature->as_C_string()); }
--- a/src/share/vm/interpreter/bytecodes.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/interpreter/bytecodes.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -357,7 +357,7 @@ def(_invokespecial , "invokespecial" , "bjj" , NULL , T_ILLEGAL, -1, true); def(_invokestatic , "invokestatic" , "bjj" , NULL , T_ILLEGAL, 0, true); def(_invokeinterface , "invokeinterface" , "bjj__", NULL , T_ILLEGAL, -1, true); - def(_invokedynamic , "invokedynamic" , "bjjjj", NULL , T_ILLEGAL, -1, true ); + def(_invokedynamic , "invokedynamic" , "bjjjj", NULL , T_ILLEGAL, 0, true ); def(_new , "new" , "bii" , NULL , T_OBJECT , 1, true ); def(_newarray , "newarray" , "bc" , NULL , T_OBJECT , 0, true ); def(_anewarray , "anewarray" , "bii" , NULL , T_OBJECT , 0, true );
--- a/src/share/vm/interpreter/interpreter.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/interpreter/interpreter.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -314,6 +314,20 @@ break; } + case Bytecodes::_invokedynamic: { + Thread *thread = Thread::current(); + ResourceMark rm(thread); + methodHandle mh(thread, method); + type = Bytecode_invoke_at(mh, bci)->result_type(thread); + // since the cache entry might not be initialized: + // (NOT needed for the old calling convension) + if (!is_top_frame) { + int index = Bytes::get_native_u4(bcp+1); + method->constants()->cache()->secondary_entry_at(index)->set_parameter_size(callee_parameters); + } + break; + } + case Bytecodes::_ldc : type = constant_pool_type( method, *(bcp+1) ); break;
--- a/src/share/vm/interpreter/interpreterRuntime.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/interpreter/interpreterRuntime.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -353,7 +353,7 @@ assert(h_exception.not_null(), "NULL exceptions should be handled by athrow"); assert(h_exception->is_oop(), "just checking"); // Check that exception is a subclass of Throwable, otherwise we have a VerifyError - if (!(h_exception->is_a(SystemDictionary::throwable_klass()))) { + if (!(h_exception->is_a(SystemDictionary::Throwable_klass()))) { if (ExitVMOnVerifyError) vm_exit(-1); ShouldNotReachHere(); } @@ -397,7 +397,7 @@ // notify JVMTI of an exception throw; JVMTI will detect if this is a first // time throw or a stack unwinding throw and accordingly notify the debugger - if (JvmtiExport::can_post_exceptions()) { + if (JvmtiExport::can_post_on_exceptions()) { JvmtiExport::post_exception_throw(thread, h_method(), bcp(thread), h_exception()); } @@ -426,7 +426,7 @@ } // notify debugger of an exception catch // (this is good for exceptions caught in native methods as well) - if (JvmtiExport::can_post_exceptions()) { + if (JvmtiExport::can_post_on_exceptions()) { JvmtiExport::notice_unwind_due_to_exception(thread, h_method(), handler_pc, h_exception(), (handler_pc != NULL)); } @@ -585,7 +585,7 @@ Handle exception(thread, thread->vm_result()); assert(exception() != NULL, "vm result should be set"); thread->set_vm_result(NULL); // clear vm result before continuing (may cause memory leaks and assert failures) - if (!exception->is_a(SystemDictionary::threaddeath_klass())) { + if (!exception->is_a(SystemDictionary::ThreadDeath_klass())) { exception = get_preinitialized_exception( SystemDictionary::IllegalMonitorStateException_klass(), CATCH); @@ -660,7 +660,7 @@ tty->print_cr("Resolving: klass: %s to method: %s", info.resolved_klass()->name()->as_C_string(), info.resolved_method()->name()->as_C_string()); } if (info.resolved_method()->method_holder() == - SystemDictionary::object_klass()) { + SystemDictionary::Object_klass()) { // NOTE: THIS IS A FIX FOR A CORNER CASE in the JVM spec // (see also cpCacheOop.cpp for details) methodHandle rm = info.resolved_method(); @@ -681,7 +681,7 @@ IRT_END -// First time execution: Resolve symbols, create a permanent CallSiteImpl object. +// First time execution: Resolve symbols, create a permanent CallSite object. IRT_ENTRY(void, InterpreterRuntime::resolve_invokedynamic(JavaThread* thread)) { ResourceMark rm(thread); @@ -708,21 +708,16 @@ constantPoolHandle pool(thread, caller_method->constants()); pool->set_invokedynamic(); // mark header to flag active call sites - int raw_index = four_byte_index(thread); - assert(constantPoolCacheOopDesc::is_secondary_index(raw_index), "invokedynamic indexes marked specially"); - - // there are two CPC entries that are of interest: - int site_index = constantPoolCacheOopDesc::decode_secondary_index(raw_index); - int main_index = pool->cache()->entry_at(site_index)->main_entry_index(); - // and there is one CP entry, a NameAndType: - int nt_index = pool->map_instruction_operand_to_index(raw_index); + int site_index = four_byte_index(thread); + // there is a second CPC entries that is of interest; it caches signature info: + int main_index = pool->cache()->secondary_entry_at(site_index)->main_entry_index(); // first resolve the signature to a MH.invoke methodOop if (!pool->cache()->entry_at(main_index)->is_resolved(bytecode)) { JvmtiHideSingleStepping jhss(thread); CallInfo info; LinkResolver::resolve_invoke(info, Handle(), pool, - raw_index, bytecode, CHECK); + site_index, bytecode, CHECK); // The main entry corresponds to a JVM_CONSTANT_NameAndType, and serves // as a common reference point for all invokedynamic call sites with // that exact call descriptor. We will link it in the CP cache exactly @@ -741,7 +736,7 @@ assert(mh_invdyn.not_null() && mh_invdyn->is_method() && mh_invdyn->is_method_handle_invoke(), "correct result from LinkResolver::resolve_invokedynamic"); - symbolHandle call_site_name(THREAD, pool->nt_name_ref_at(nt_index)); + symbolHandle call_site_name(THREAD, pool->name_ref_at(site_index)); Handle call_site = SystemDictionary::make_dynamic_call_site(caller_method->method_holder(), caller_method->method_idnum(), @@ -753,61 +748,11 @@ // In the secondary entry, the f1 field is the call site, and the f2 (index) // field is some data about the invoke site. int extra_data = 0; - pool->cache()->entry_at(site_index)->set_dynamic_call(call_site(), extra_data); + pool->cache()->secondary_entry_at(site_index)->set_dynamic_call(call_site(), extra_data); } IRT_END -// Called on first time execution, and also whenever the CallSite.target is null. -// FIXME: Do more of this in Java code. -IRT_ENTRY(void, InterpreterRuntime::bootstrap_invokedynamic(JavaThread* thread, oopDesc* call_site)) { - methodHandle mh_invdyn(thread, (methodOop) sun_dyn_CallSiteImpl::vmmethod(call_site)); - Handle mh_type(thread, mh_invdyn->method_handle_type()); - objArrayHandle mh_ptypes(thread, java_dyn_MethodType::ptypes(mh_type())); - - // squish the arguments down to a single array - int nargs = mh_ptypes->length(); - objArrayHandle arg_array; - { - objArrayOop aaoop = oopFactory::new_objArray(SystemDictionary::object_klass(), nargs, CHECK); - arg_array = objArrayHandle(thread, aaoop); - } - frame fr = thread->last_frame(); - assert(fr.interpreter_frame_bcp() != NULL, "sanity"); - int tos_offset = 0; - for (int i = nargs; --i >= 0; ) { - intptr_t* slot_addr = fr.interpreter_frame_tos_at(tos_offset++); - oop ptype = mh_ptypes->obj_at(i); - oop arg = NULL; - if (!java_lang_Class::is_primitive(ptype)) { - arg = *(oop*) slot_addr; - } else { - BasicType bt = java_lang_Class::primitive_type(ptype); - assert(frame::interpreter_frame_expression_stack_direction() < 0, "else reconsider this code"); - jvalue value; - Interpreter::get_jvalue_in_slot(slot_addr, bt, &value); - tos_offset += type2size[bt]-1; - arg = java_lang_boxing_object::create(bt, &value, CHECK); - // FIXME: These boxing objects are not canonicalized under - // the Java autoboxing rules. They should be... - // The best approach would be to push the arglist creation into Java. - // The JVM should use a lower-level interface to communicate argument lists. - } - arg_array->obj_at_put(i, arg); - } - - // now find the bootstrap method - oop bootstrap_mh_oop = instanceKlass::cast(fr.interpreter_frame_method()->method_holder())->bootstrap_method(); - assert(bootstrap_mh_oop != NULL, "resolve_invokedynamic ensures a BSM"); - - // return the bootstrap method and argument array via vm_result/_2 - thread->set_vm_result(bootstrap_mh_oop); - thread->set_vm_result_2(arg_array()); -} -IRT_END - - - //------------------------------------------------------------------------------------------------------------------------ // Miscellaneous @@ -1305,7 +1250,7 @@ methodHandle mh(thread, fr.interpreter_frame_method()); Bytecode_invoke* invoke = Bytecode_invoke_at(mh, bci); ArgumentSizeComputer asc(invoke->signature()); - int size_of_arguments = (asc.size() + (invoke->is_invokestatic() ? 0 : 1)); // receiver + int size_of_arguments = (asc.size() + (invoke->has_receiver() ? 1 : 0)); // receiver Copy::conjoint_bytes(src_address, dest_address, size_of_arguments * Interpreter::stackElementSize()); IRT_END
--- a/src/share/vm/interpreter/interpreterRuntime.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/interpreter/interpreterRuntime.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -91,7 +91,6 @@ // Calls static void resolve_invoke (JavaThread* thread, Bytecodes::Code bytecode); static void resolve_invokedynamic(JavaThread* thread); - static void bootstrap_invokedynamic(JavaThread* thread, oopDesc* call_site); // Breakpoints static void _breakpoint(JavaThread* thread, methodOopDesc* method, address bcp);
--- a/src/share/vm/interpreter/linkResolver.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/interpreter/linkResolver.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -55,7 +55,7 @@ // we should pick the vtable index from the resolved method. // Other than that case, there is no valid vtable index to specify. int vtable_index = methodOopDesc::invalid_vtable_index; - if (resolved_method->method_holder() == SystemDictionary::object_klass()) { + if (resolved_method->method_holder() == SystemDictionary::Object_klass()) { assert(resolved_method->vtable_index() == selected_method->vtable_index(), "sanity check"); vtable_index = resolved_method->vtable_index(); } @@ -75,6 +75,8 @@ _selected_method = selected_method; _vtable_index = vtable_index; if (CompilationPolicy::mustBeCompiled(selected_method)) { + // This path is unusual, mostly used by the '-Xcomp' stress test mode. + // Note: with several active threads, the mustBeCompiled may be true // while canBeCompiled is false; remove assert // assert(CompilationPolicy::canBeCompiled(selected_method), "cannot compile"); @@ -82,6 +84,16 @@ // don't force compilation, resolve was on behalf of compiler return; } + if (instanceKlass::cast(selected_method->method_holder())->is_not_initialized()) { + // 'is_not_initialized' means not only '!is_initialized', but also that + // initialization has not been started yet ('!being_initialized') + // Do not force compilation of methods in uninitialized classes. + // Note that doing this would throw an assert later, + // in CompileBroker::compile_method. + // We sometimes use the link resolver to do reflective lookups + // even before classes are initialized. + return; + } CompileBroker::compile_method(selected_method, InvocationEntryBci, methodHandle(), 0, "mustBeCompiled", CHECK); } @@ -181,7 +193,7 @@ // We'll check for the method name first, as that's most likely // to be false (so we'll short-circuit out of these tests). if (sel_method->name() == vmSymbols::clone_name() && - sel_klass() == SystemDictionary::object_klass() && + sel_klass() == SystemDictionary::Object_klass() && resolved_klass->oop_is_array()) { // We need to change "protected" to "public". assert(flags.is_protected(), "clone not protected?"); @@ -223,6 +235,18 @@ resolve_method(resolved_method, resolved_klass, method_name, method_signature, current_klass, true, CHECK); } +void LinkResolver::resolve_dynamic_method(methodHandle& resolved_method, KlassHandle& resolved_klass, constantPoolHandle pool, int index, TRAPS) { + // The class is java.dyn.MethodHandle + resolved_klass = SystemDictionaryHandles::MethodHandle_klass(); + + symbolHandle method_name = vmSymbolHandles::invoke_name(); + + symbolHandle method_signature(THREAD, pool->signature_ref_at(index)); + KlassHandle current_klass (THREAD, pool->pool_holder()); + + resolve_method(resolved_method, resolved_klass, method_name, method_signature, current_klass, true, CHECK); +} + void LinkResolver::resolve_interface_method(methodHandle& resolved_method, KlassHandle& resolved_klass, constantPoolHandle pool, int index, TRAPS) { // resolve klass @@ -1015,11 +1039,8 @@ // This guy is reached from InterpreterRuntime::resolve_invokedynamic. - assert(constantPoolCacheOopDesc::is_secondary_index(raw_index), "must be secondary index"); - int nt_index = pool->map_instruction_operand_to_index(raw_index); - // At this point, we only need the signature, and can ignore the name. - symbolHandle method_signature(THREAD, pool->nt_signature_ref_at(nt_index)); + symbolHandle method_signature(THREAD, pool->signature_ref_at(raw_index)); // raw_index works directly symbolHandle method_name = vmSymbolHandles::invoke_name(); KlassHandle resolved_klass = SystemDictionaryHandles::MethodHandle_klass();
--- a/src/share/vm/interpreter/linkResolver.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/interpreter/linkResolver.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -133,6 +133,7 @@ // static resolving for all calls except interface calls static void resolve_method (methodHandle& method_result, KlassHandle& klass_result, constantPoolHandle pool, int index, TRAPS); + static void resolve_dynamic_method (methodHandle& resolved_method, KlassHandle& resolved_klass, constantPoolHandle pool, int index, TRAPS); static void resolve_interface_method(methodHandle& method_result, KlassHandle& klass_result, constantPoolHandle pool, int index, TRAPS); // runtime/static resolving for fields
--- a/src/share/vm/interpreter/rewriter.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/interpreter/rewriter.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -48,16 +48,6 @@ } -int Rewriter::add_extra_cp_cache_entry(int main_entry) { - // Hack: We put it on the map as an encoded value. - // The only place that consumes this is ConstantPoolCacheEntry::set_initial_state - int encoded = constantPoolCacheOopDesc::encode_secondary_index(main_entry); - int plain_secondary_index = _cp_cache_map.append(encoded); - return constantPoolCacheOopDesc::encode_secondary_index(plain_secondary_index); -} - - - // Creates a constant pool cache given a CPC map // This creates the constant pool cache initially in a state // that is unsafe for concurrent GC processing but sets it to @@ -127,7 +117,7 @@ assert(p[-1] == Bytecodes::_invokedynamic, ""); int cp_index = Bytes::get_Java_u2(p); int cpc = maybe_add_cp_cache_entry(cp_index); // add lazily - int cpc2 = add_extra_cp_cache_entry(cpc); + int cpc2 = add_secondary_cp_cache_entry(cpc); // Replace the trailing four bytes with a CPC index for the dynamic // call site. Unlike other CPC entries, there is one per bytecode, @@ -137,7 +127,7 @@ // all these entries. That is the main reason invokedynamic // must have a five-byte instruction format. (Of course, other JVM // implementations can use the bytes for other purposes.) - Bytes::put_native_u4(p, cpc2); + Bytes::put_native_u4(p, constantPoolCacheOopDesc::encode_secondary_index(cpc2)); // Note: We use native_u4 format exclusively for 4-byte indexes. } @@ -257,15 +247,22 @@ void Rewriter::rewrite(instanceKlassHandle klass, TRAPS) { ResourceMark rm(THREAD); - Rewriter rw(klass, CHECK); + Rewriter rw(klass, klass->constants(), klass->methods(), CHECK); // (That's all, folks.) } -Rewriter::Rewriter(instanceKlassHandle klass, TRAPS) + +void Rewriter::rewrite(instanceKlassHandle klass, constantPoolHandle cpool, objArrayHandle methods, TRAPS) { + ResourceMark rm(THREAD); + Rewriter rw(klass, cpool, methods, CHECK); + // (That's all, folks.) +} + + +Rewriter::Rewriter(instanceKlassHandle klass, constantPoolHandle cpool, objArrayHandle methods, TRAPS) : _klass(klass), - // gather starting points - _pool( THREAD, klass->constants()), - _methods(THREAD, klass->methods()) + _pool(cpool), + _methods(methods) { assert(_pool->cache() == NULL, "constant pool cache must not be set yet");
--- a/src/share/vm/interpreter/rewriter.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/interpreter/rewriter.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -43,16 +43,21 @@ bool has_cp_cache(int i) { return (uint)i < (uint)_cp_map.length() && _cp_map[i] >= 0; } int maybe_add_cp_cache_entry(int i) { return has_cp_cache(i) ? _cp_map[i] : add_cp_cache_entry(i); } int add_cp_cache_entry(int cp_index) { + assert((cp_index & _secondary_entry_tag) == 0, "bad tag"); assert(_cp_map[cp_index] == -1, "not twice on same cp_index"); int cache_index = _cp_cache_map.append(cp_index); _cp_map.at_put(cp_index, cache_index); assert(cp_entry_to_cp_cache(cp_index) == cache_index, ""); return cache_index; } - int add_extra_cp_cache_entry(int main_entry); + int add_secondary_cp_cache_entry(int main_cpc_entry) { + assert(main_cpc_entry < _cp_cache_map.length(), "must be earlier CP cache entry"); + int cache_index = _cp_cache_map.append(main_cpc_entry | _secondary_entry_tag); + return cache_index; + } // All the work goes in here: - Rewriter(instanceKlassHandle klass, TRAPS); + Rewriter(instanceKlassHandle klass, constantPoolHandle cpool, objArrayHandle methods, TRAPS); void compute_index_maps(); void make_constant_pool_cache(TRAPS); @@ -65,4 +70,9 @@ public: // Driver routine: static void rewrite(instanceKlassHandle klass, TRAPS); + static void rewrite(instanceKlassHandle klass, constantPoolHandle cpool, objArrayHandle methods, TRAPS); + + enum { + _secondary_entry_tag = nth_bit(30) + }; };
--- a/src/share/vm/interpreter/templateInterpreter.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/interpreter/templateInterpreter.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -178,14 +178,12 @@ #endif // !PRODUCT EntryPoint TemplateInterpreter::_return_entry[TemplateInterpreter::number_of_return_entries]; EntryPoint TemplateInterpreter::_earlyret_entry; -EntryPoint TemplateInterpreter::_return_unbox_entry; EntryPoint TemplateInterpreter::_deopt_entry [TemplateInterpreter::number_of_deopt_entries ]; EntryPoint TemplateInterpreter::_continuation_entry; EntryPoint TemplateInterpreter::_safept_entry; address TemplateInterpreter::_return_3_addrs_by_index[TemplateInterpreter::number_of_return_addrs]; address TemplateInterpreter::_return_5_addrs_by_index[TemplateInterpreter::number_of_return_addrs]; -address TemplateInterpreter::_return_5_unbox_addrs_by_index[TemplateInterpreter::number_of_return_addrs]; DispatchTable TemplateInterpreter::_active_table; DispatchTable TemplateInterpreter::_normal_table; @@ -253,22 +251,6 @@ } } - if (EnableInvokeDynamic) { - CodeletMark cm(_masm, "unboxing return entry points"); - Interpreter::_return_unbox_entry = - EntryPoint( - generate_return_unbox_entry_for(btos, 5), - generate_return_unbox_entry_for(ctos, 5), - generate_return_unbox_entry_for(stos, 5), - generate_return_unbox_entry_for(atos, 5), // cast conversion - generate_return_unbox_entry_for(itos, 5), - generate_return_unbox_entry_for(ltos, 5), - generate_return_unbox_entry_for(ftos, 5), - generate_return_unbox_entry_for(dtos, 5), - Interpreter::_return_entry[5].entry(vtos) // no unboxing for void - ); - } - { CodeletMark cm(_masm, "earlyret entry points"); Interpreter::_earlyret_entry = EntryPoint( @@ -319,8 +301,6 @@ int index = Interpreter::TosState_as_index(states[j]); Interpreter::_return_3_addrs_by_index[index] = Interpreter::return_entry(states[j], 3); Interpreter::_return_5_addrs_by_index[index] = Interpreter::return_entry(states[j], 5); - if (EnableInvokeDynamic) - Interpreter::_return_5_unbox_addrs_by_index[index] = Interpreter::return_unbox_entry(states[j], 5); } { CodeletMark cm(_masm, "continuation entry points"); @@ -485,9 +465,11 @@ void TemplateInterpreterGenerator::set_short_entry_points(Template* t, address& bep, address& cep, address& sep, address& aep, address& iep, address& lep, address& fep, address& dep, address& vep) { assert(t->is_valid(), "template must exist"); switch (t->tos_in()) { - case btos: vep = __ pc(); __ pop(btos); bep = __ pc(); generate_and_dispatch(t); break; - case ctos: vep = __ pc(); __ pop(ctos); sep = __ pc(); generate_and_dispatch(t); break; - case stos: vep = __ pc(); __ pop(stos); sep = __ pc(); generate_and_dispatch(t); break; + case btos: + case ctos: + case stos: + ShouldNotReachHere(); // btos/ctos/stos should use itos. + break; case atos: vep = __ pc(); __ pop(atos); aep = __ pc(); generate_and_dispatch(t); break; case itos: vep = __ pc(); __ pop(itos); iep = __ pc(); generate_and_dispatch(t); break; case ltos: vep = __ pc(); __ pop(ltos); lep = __ pc(); generate_and_dispatch(t); break; @@ -547,18 +529,6 @@ } -address TemplateInterpreter::return_unbox_entry(TosState state, int length) { - assert(EnableInvokeDynamic, ""); - if (state == vtos) { - // no unboxing to do, actually - return return_entry(state, length); - } else { - assert(length == 5, "unboxing entries generated for invokedynamic only"); - return _return_unbox_entry.entry(state); - } -} - - address TemplateInterpreter::deopt_entry(TosState state, int length) { guarantee(0 <= length && length < Interpreter::number_of_deopt_entries, "illegal length"); return _deopt_entry[length].entry(state);
--- a/src/share/vm/interpreter/templateInterpreter.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/interpreter/templateInterpreter.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -110,14 +110,12 @@ #endif // !PRODUCT static EntryPoint _return_entry[number_of_return_entries]; // entry points to return to from a call static EntryPoint _earlyret_entry; // entry point to return early from a call - static EntryPoint _return_unbox_entry; // entry point to unbox a return value from a call static EntryPoint _deopt_entry[number_of_deopt_entries]; // entry points to return to from a deoptimization static EntryPoint _continuation_entry; static EntryPoint _safept_entry; static address _return_3_addrs_by_index[number_of_return_addrs]; // for invokevirtual return entries static address _return_5_addrs_by_index[number_of_return_addrs]; // for invokeinterface return entries - static address _return_5_unbox_addrs_by_index[number_of_return_addrs]; // for invokedynamic bootstrap methods static DispatchTable _active_table; // the active dispatch table (used by the interpreter for dispatch) static DispatchTable _normal_table; // the normal dispatch table (used to set the active table in normal mode) @@ -159,12 +157,10 @@ // Support for invokes static address* return_3_addrs_by_index_table() { return _return_3_addrs_by_index; } static address* return_5_addrs_by_index_table() { return _return_5_addrs_by_index; } - static address* return_5_unbox_addrs_by_index_table() { return _return_5_unbox_addrs_by_index; } static int TosState_as_index(TosState state); // computes index into return_3_entry_by_index table static address return_entry (TosState state, int length); static address deopt_entry (TosState state, int length); - static address return_unbox_entry(TosState state, int length); // Safepoint support static void notice_safepoints(); // stops the thread when reaching a safepoint
--- a/src/share/vm/interpreter/templateInterpreterGenerator.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/interpreter/templateInterpreterGenerator.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -51,10 +51,7 @@ address generate_WrongMethodType_handler(); address generate_ArrayIndexOutOfBounds_handler(const char* name); address generate_continuation_for(TosState state); - address generate_return_entry_for(TosState state, int step, bool unbox = false); - address generate_return_unbox_entry_for(TosState state, int step) { - return generate_return_entry_for(state, step, true); - } + address generate_return_entry_for(TosState state, int step); address generate_earlyret_entry_for(TosState state); address generate_deopt_entry_for(TosState state, int step); address generate_safept_entry_for(TosState state, address runtime_entry);
--- a/src/share/vm/memory/barrierSet.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/barrierSet.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -41,11 +41,6 @@ // count is number of array elements being written void BarrierSet::static_write_ref_array_post(HeapWord* start, size_t count) { - assert(count <= (size_t)max_intx, "count too large"); - HeapWord* end = start + objArrayOopDesc::array_size((int)count); -#if 0 - warning("Post:\t" INTPTR_FORMAT "[" SIZE_FORMAT "] : [" INTPTR_FORMAT","INTPTR_FORMAT")\t", - start, count, start, end); -#endif - Universe::heap()->barrier_set()->write_ref_array_work(MemRegion(start, end)); + // simply delegate to instance method + Universe::heap()->barrier_set()->write_ref_array(start, count); }
--- a/src/share/vm/memory/barrierSet.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/barrierSet.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -121,17 +121,18 @@ virtual void read_ref_array(MemRegion mr) = 0; virtual void read_prim_array(MemRegion mr) = 0; + // Below length is the # array elements being written virtual void write_ref_array_pre( oop* dst, int length) {} virtual void write_ref_array_pre(narrowOop* dst, int length) {} - inline void write_ref_array(MemRegion mr); + // Below count is the # array elements being written, starting + // at the address "start", which may not necessarily be HeapWord-aligned + inline void write_ref_array(HeapWord* start, size_t count); - // Static versions, suitable for calling from generated code. + // Static versions, suitable for calling from generated code; + // count is # array elements being written, starting with "start", + // which may not necessarily be HeapWord-aligned. static void static_write_ref_array_pre(HeapWord* start, size_t count); static void static_write_ref_array_post(HeapWord* start, size_t count); - // Narrow oop versions of the above; count is # of array elements being written, - // starting with "start", which is HeapWord-aligned. - static void static_write_ref_array_pre_narrow(HeapWord* start, size_t count); - static void static_write_ref_array_post_narrow(HeapWord* start, size_t count); protected: virtual void write_ref_array_work(MemRegion mr) = 0;
--- a/src/share/vm/memory/barrierSet.inline.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/barrierSet.inline.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -42,14 +42,34 @@ } } -void BarrierSet::write_ref_array(MemRegion mr) { - if (kind() == CardTableModRef) { - ((CardTableModRefBS*)this)->inline_write_ref_array(mr); - } else { - write_ref_array_work(mr); - } +// count is number of array elements being written +void BarrierSet::write_ref_array(HeapWord* start, size_t count) { + assert(count <= (size_t)max_intx, "count too large"); + HeapWord* end = (HeapWord*)((char*)start + (count*heapOopSize)); + // In the case of compressed oops, start and end may potentially be misaligned; + // so we need to conservatively align the first downward (this is not + // strictly necessary for current uses, but a case of good hygiene and, + // if you will, aesthetics) and the second upward (this is essential for + // current uses) to a HeapWord boundary, so we mark all cards overlapping + // this write. If this evolves in the future to calling a + // logging barrier of narrow oop granularity, like the pre-barrier for G1 + // (mentioned here merely by way of example), we will need to change this + // interface, so it is "exactly precise" (if i may be allowed the adverbial + // redundancy for emphasis) and does not include narrow oop slots not + // included in the original write interval. + HeapWord* aligned_start = (HeapWord*)align_size_down((uintptr_t)start, HeapWordSize); + HeapWord* aligned_end = (HeapWord*)align_size_up ((uintptr_t)end, HeapWordSize); + // If compressed oops were not being used, these should already be aligned + assert(UseCompressedOops || (aligned_start == start && aligned_end == end), + "Expected heap word alignment of start and end"); +#if 0 + warning("Post:\t" INTPTR_FORMAT "[" SIZE_FORMAT "] : [" INTPTR_FORMAT","INTPTR_FORMAT")\t", + start, count, aligned_start, aligned_end); +#endif + write_ref_array_work(MemRegion(aligned_start, aligned_end)); } + void BarrierSet::write_region(MemRegion mr) { if (kind() == CardTableModRef) { ((CardTableModRefBS*)this)->inline_write_region(mr);
--- a/src/share/vm/memory/cardTableModRefBS.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/cardTableModRefBS.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -511,6 +511,8 @@ } void CardTableModRefBS::dirty_MemRegion(MemRegion mr) { + assert((HeapWord*)align_size_down((uintptr_t)mr.start(), HeapWordSize) == mr.start(), "Unaligned start"); + assert((HeapWord*)align_size_up ((uintptr_t)mr.end(), HeapWordSize) == mr.end(), "Unaligned end" ); jbyte* cur = byte_for(mr.start()); jbyte* last = byte_after(mr.last()); while (cur < last) { @@ -520,6 +522,8 @@ } void CardTableModRefBS::invalidate(MemRegion mr, bool whole_heap) { + assert((HeapWord*)align_size_down((uintptr_t)mr.start(), HeapWordSize) == mr.start(), "Unaligned start"); + assert((HeapWord*)align_size_up ((uintptr_t)mr.end(), HeapWordSize) == mr.end(), "Unaligned end" ); for (int i = 0; i < _cur_covered_regions; i++) { MemRegion mri = mr.intersection(_covered[i]); if (!mri.is_empty()) dirty_MemRegion(mri);
--- a/src/share/vm/memory/cardTableModRefBS.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/cardTableModRefBS.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -339,6 +339,16 @@ return (val & (clean_card_mask_val() | claimed_card_val())) == claimed_card_val(); } + void set_card_claimed(size_t card_index) { + jbyte val = _byte_map[card_index]; + if (val == clean_card_val()) { + val = (jbyte)claimed_card_val(); + } else { + val |= (jbyte)claimed_card_val(); + } + _byte_map[card_index] = val; + } + bool claim_card(size_t card_index); bool is_card_clean(size_t card_index) {
--- a/src/share/vm/memory/classify.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/classify.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -49,7 +49,7 @@ Klass* k = obj->blueprint(); - if (k->as_klassOop() == SystemDictionary::object_klass()) { + if (k->as_klassOop() == SystemDictionary::Object_klass()) { tty->print_cr("Found the class!"); }
--- a/src/share/vm/memory/collectorPolicy.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/collectorPolicy.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -55,7 +55,7 @@ void CollectorPolicy::initialize_size_info() { // User inputs from -mx and ms are aligned - set_initial_heap_byte_size(Arguments::initial_heap_size()); + set_initial_heap_byte_size(InitialHeapSize); if (initial_heap_byte_size() == 0) { set_initial_heap_byte_size(NewSize + OldSize); }
--- a/src/share/vm/memory/defNewGeneration.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/defNewGeneration.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -609,7 +609,7 @@ remove_forwarding_pointers(); if (PrintGCDetails) { - gclog_or_tty->print(" (promotion failed)"); + gclog_or_tty->print(" (promotion failed) "); } // Add to-space to the list of space to compact // when a promotion failure has occurred. In that @@ -620,6 +620,9 @@ from()->set_next_compaction_space(to()); gch->set_incremental_collection_will_fail(); + // Inform the next generation that a promotion failure occurred. + _next_gen->promotion_failure_occurred(); + // Reset the PromotionFailureALot counters. NOT_PRODUCT(Universe::heap()->reset_promotion_should_fail();) } @@ -679,6 +682,11 @@ void DefNewGeneration::handle_promotion_failure(oop old) { preserve_mark_if_necessary(old, old->mark()); + if (!_promotion_failed && PrintPromotionFailure) { + gclog_or_tty->print(" (promotion failure size = " SIZE_FORMAT ") ", + old->size()); + } + // forward to self old->forward_to(old); _promotion_failed = true;
--- a/src/share/vm/memory/dump.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/dump.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -63,7 +63,7 @@ void do_oop(oop* p) { if (p != NULL) { oop obj = *p; - if (obj->klass() == SystemDictionary::string_klass()) { + if (obj->klass() == SystemDictionary::String_klass()) { int hash; typeArrayOop value = java_lang_String::value(obj); @@ -625,11 +625,11 @@ if (obj->is_klass() || obj->is_instance()) { if (obj->is_klass() || - obj->is_a(SystemDictionary::class_klass()) || - obj->is_a(SystemDictionary::throwable_klass())) { + obj->is_a(SystemDictionary::Class_klass()) || + obj->is_a(SystemDictionary::Throwable_klass())) { // Do nothing } - else if (obj->is_a(SystemDictionary::string_klass())) { + else if (obj->is_a(SystemDictionary::String_klass())) { // immutable objects. } else { // someone added an object we hadn't accounted for.
--- a/src/share/vm/memory/genCollectedHeap.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/genCollectedHeap.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -51,6 +51,8 @@ } jint GenCollectedHeap::initialize() { + CollectedHeap::pre_initialize(); + int i; _n_gens = gen_policy()->number_of_generations(); @@ -129,6 +131,7 @@ _rem_set = collector_policy()->create_rem_set(_reserved, n_covered_regions); set_barrier_set(rem_set()->bs()); + _gch = this; for (i = 0; i < _n_gens; i++) { @@ -925,6 +928,8 @@ guarantee(VerifyBeforeGC || VerifyDuringGC || VerifyBeforeExit || + PrintAssembly || + tty->count() != 0 || // already printing VerifyAfterGC, "too expensive"); #endif // This might be sped up with a cache of the last generation that
--- a/src/share/vm/memory/genCollectedHeap.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/genCollectedHeap.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -260,6 +260,10 @@ return true; } + virtual bool card_mark_must_follow_store() const { + return UseConcMarkSweepGC; + } + // We don't need barriers for stores to objects in the // young gen and, a fortiori, for initializing stores to // objects therein. This applies to {DefNew,ParNew}+{Tenured,CMS}
--- a/src/share/vm/memory/generation.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/generation.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -181,6 +181,12 @@ virtual bool promotion_attempt_is_safe(size_t promotion_in_bytes, bool younger_handles_promotion_failure) const; + // For a non-young generation, this interface can be used to inform a + // generation that a promotion attempt into that generation failed. + // Typically used to enable diagnostic output for post-mortem analysis, + // but other uses of the interface are not ruled out. + virtual void promotion_failure_occurred() { /* does nothing */ } + // Return an estimate of the maximum allocation that could be performed // in the generation without triggering any collection or expansion // activity. It is "unsafe" because no locks are taken; the result
--- a/src/share/vm/memory/heap.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/heap.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -464,7 +464,7 @@ } // Verify that freelist contains the right amount of free space - guarantee(len == _free_segments, "wrong freelist"); + // guarantee(len == _free_segments, "wrong freelist"); // Verify that the number of free blocks is not out of hand. static int free_block_threshold = 10000; @@ -479,5 +479,5 @@ for(HeapBlock *h = first_block(); h != NULL; h = next_block(h)) { if (h->free()) count--; } - guarantee(count == 0, "missing free blocks"); + // guarantee(count == 0, "missing free blocks"); }
--- a/src/share/vm/memory/iterator.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/iterator.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -296,23 +296,32 @@ // RememberKlassesChecker can be passed "false" to turn off checking. // It is used by CMS when CMS yields to a different collector. class RememberKlassesChecker: StackObj { - bool _state; - bool _skip; + bool _saved_state; + bool _do_check; public: - RememberKlassesChecker(bool checking_on) : _state(false), _skip(false) { - _skip = !(ClassUnloading && !UseConcMarkSweepGC || - CMSClassUnloadingEnabled && UseConcMarkSweepGC); - if (_skip) { - return; + RememberKlassesChecker(bool checking_on) : _saved_state(false), + _do_check(true) { + // The ClassUnloading unloading flag affects the collectors except + // for CMS. + // CMS unloads classes if CMSClassUnloadingEnabled is true or + // if ExplicitGCInvokesConcurrentAndUnloadsClasses is true and + // the current collection is an explicit collection. Turning + // on the checking in general for + // ExplicitGCInvokesConcurrentAndUnloadsClasses and + // UseConcMarkSweepGC should not lead to false positives. + _do_check = + ClassUnloading && !UseConcMarkSweepGC || + CMSClassUnloadingEnabled && UseConcMarkSweepGC || + ExplicitGCInvokesConcurrentAndUnloadsClasses && UseConcMarkSweepGC; + if (_do_check) { + _saved_state = OopClosure::must_remember_klasses(); + OopClosure::set_must_remember_klasses(checking_on); } - _state = OopClosure::must_remember_klasses(); - OopClosure::set_must_remember_klasses(checking_on); } ~RememberKlassesChecker() { - if (_skip) { - return; + if (_do_check) { + OopClosure::set_must_remember_klasses(_saved_state); } - OopClosure::set_must_remember_klasses(_state); } }; #endif // ASSERT
--- a/src/share/vm/memory/referenceProcessor.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/referenceProcessor.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -71,7 +71,7 @@ assert(_sentinelRef == NULL, "should be initialized precisely once"); EXCEPTION_MARK; _sentinelRef = instanceKlass::cast( - SystemDictionary::reference_klass())-> + SystemDictionary::Reference_klass())-> allocate_permanent_instance(THREAD); // Initialize the master soft ref clock. @@ -299,8 +299,8 @@ template <class T> -static bool enqueue_discovered_ref_helper(ReferenceProcessor* ref, - AbstractRefProcTaskExecutor* task_executor) { +bool enqueue_discovered_ref_helper(ReferenceProcessor* ref, + AbstractRefProcTaskExecutor* task_executor) { // Remember old value of pending references list T* pending_list_addr = (T*)java_lang_ref_Reference::pending_list_addr(); @@ -1227,13 +1227,16 @@ BoolObjectClosure* is_alive, OopClosure* keep_alive, VoidClosure* complete_gc, - YieldClosure* yield) { + YieldClosure* yield, + bool should_unload_classes) { NOT_PRODUCT(verify_ok_to_handle_reflists()); #ifdef ASSERT bool must_remember_klasses = ClassUnloading && !UseConcMarkSweepGC || - CMSClassUnloadingEnabled && UseConcMarkSweepGC; + CMSClassUnloadingEnabled && UseConcMarkSweepGC || + ExplicitGCInvokesConcurrentAndUnloadsClasses && + UseConcMarkSweepGC && should_unload_classes; RememberKlassesChecker mx(must_remember_klasses); #endif // Soft references
--- a/src/share/vm/memory/referenceProcessor.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/referenceProcessor.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -170,11 +170,13 @@ // The caller is responsible for taking care of potential // interference with concurrent operations on these lists // (or predicates involved) by other threads. Currently - // only used by the CMS collector. + // only used by the CMS collector. should_unload_classes is + // used to aid assertion checking when classes are collected. void preclean_discovered_references(BoolObjectClosure* is_alive, OopClosure* keep_alive, VoidClosure* complete_gc, - YieldClosure* yield); + YieldClosure* yield, + bool should_unload_classes); // Delete entries in the discovered lists that have // either a null referent or are not active. Such @@ -261,10 +263,13 @@ int parallel_gc_threads = 1, bool mt_processing = false, bool discovered_list_needs_barrier = false); + // RefDiscoveryPolicy values - enum { + enum DiscoveryPolicy { ReferenceBasedDiscovery = 0, - ReferentBasedDiscovery = 1 + ReferentBasedDiscovery = 1, + DiscoveryPolicyMin = ReferenceBasedDiscovery, + DiscoveryPolicyMax = ReferentBasedDiscovery }; static void init_statics();
--- a/src/share/vm/memory/sharedHeap.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/sharedHeap.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -224,10 +224,6 @@ CodeBlobClosure* code_roots, OopClosure* non_root_closure); - - // Like CollectedHeap::collect, but assume that the caller holds the Heap_lock. - virtual void collect_locked(GCCause::Cause cause) = 0; - // The functions below are helper functions that a subclass of // "SharedHeap" can use in the implementation of its virtual // functions.
--- a/src/share/vm/memory/space.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/space.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -876,7 +876,7 @@ instanceOop obj = (instanceOop) allocate(size); obj->set_mark(markOopDesc::prototype()); obj->set_klass_gap(0); - obj->set_klass(SystemDictionary::object_klass()); + obj->set_klass(SystemDictionary::Object_klass()); } }
--- a/src/share/vm/memory/threadLocalAllocBuffer.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/threadLocalAllocBuffer.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -100,7 +100,7 @@ void ThreadLocalAllocBuffer::make_parsable(bool retire) { if (end() != NULL) { invariants(); - CollectedHeap::fill_with_object(top(), hard_end()); + CollectedHeap::fill_with_object(top(), hard_end(), retire); if (retire || ZeroTLAB) { // "Reset" the TLAB set_start(NULL);
--- a/src/share/vm/memory/threadLocalAllocBuffer.inline.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/threadLocalAllocBuffer.inline.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2007 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -27,8 +27,13 @@ HeapWord* obj = top(); if (pointer_delta(end(), obj) >= size) { // successful thread-local allocation - - DEBUG_ONLY(Copy::fill_to_words(obj, size, badHeapWordVal)); +#ifdef ASSERT + // Skip mangling the space corresponding to the object header to + // ensure that the returned space is not considered parsable by + // any concurrent GC thread. + size_t hdr_size = CollectedHeap::min_fill_size(); + Copy::fill_to_words(obj + hdr_size, size - hdr_size, badHeapWordVal); +#endif // ASSERT // This addition is safe because we know that top is // at least size below end, so the add can't wrap. set_top(obj + size);
--- a/src/share/vm/memory/universe.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/universe.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -67,6 +67,8 @@ objArrayOop Universe::_the_empty_system_obj_array = NULL; objArrayOop Universe::_the_empty_class_klass_array = NULL; objArrayOop Universe::_the_array_interfaces_array = NULL; +oop Universe::_the_null_string = NULL; +oop Universe::_the_min_jint_string = NULL; LatestMethodOopCache* Universe::_finalizer_register_cache = NULL; LatestMethodOopCache* Universe::_loader_addClass_cache = NULL; ActiveMethodOopsCache* Universe::_reflect_invoke_cache = NULL; @@ -187,6 +189,8 @@ f->do_oop((oop*)&_the_empty_system_obj_array); f->do_oop((oop*)&_the_empty_class_klass_array); f->do_oop((oop*)&_the_array_interfaces_array); + f->do_oop((oop*)&_the_null_string); + f->do_oop((oop*)&_the_min_jint_string); _finalizer_register_cache->oops_do(f); _loader_addClass_cache->oops_do(f); _reflect_invoke_cache->oops_do(f); @@ -287,14 +291,17 @@ SystemDictionary::initialize(CHECK); - klassOop ok = SystemDictionary::object_klass(); + klassOop ok = SystemDictionary::Object_klass(); + + _the_null_string = StringTable::intern("null", CHECK); + _the_min_jint_string = StringTable::intern("-2147483648", CHECK); if (UseSharedSpaces) { // Verify shared interfaces array. assert(_the_array_interfaces_array->obj_at(0) == - SystemDictionary::cloneable_klass(), "u3"); + SystemDictionary::Cloneable_klass(), "u3"); assert(_the_array_interfaces_array->obj_at(1) == - SystemDictionary::serializable_klass(), "u3"); + SystemDictionary::Serializable_klass(), "u3"); // Verify element klass for system obj array klass assert(objArrayKlass::cast(_systemObjArrayKlassObj)->element_klass() == ok, "u1"); @@ -313,8 +320,8 @@ assert(Klass::cast(systemObjArrayKlassObj())->super() == ok, "u3"); } else { // Set up shared interfaces array. (Do this before supers are set up.) - _the_array_interfaces_array->obj_at_put(0, SystemDictionary::cloneable_klass()); - _the_array_interfaces_array->obj_at_put(1, SystemDictionary::serializable_klass()); + _the_array_interfaces_array->obj_at_put(0, SystemDictionary::Cloneable_klass()); + _the_array_interfaces_array->obj_at_put(1, SystemDictionary::Serializable_klass()); // Set element klass for system obj array klass objArrayKlass::cast(_systemObjArrayKlassObj)->set_element_klass(ok); @@ -358,7 +365,7 @@ // Initialize _objectArrayKlass after core bootstraping to make // sure the super class is set up properly for _objectArrayKlass. _objectArrayKlassObj = instanceKlass:: - cast(SystemDictionary::object_klass())->array_klass(1, CHECK); + cast(SystemDictionary::Object_klass())->array_klass(1, CHECK); // Add the class to the class hierarchy manually to make sure that // its vtable is initialized after core bootstrapping is completed. Klass::cast(_objectArrayKlassObj)->append_to_sibling_list(); @@ -419,11 +426,11 @@ while (i < size) { if (!UseConcMarkSweepGC) { // Allocate dummy in old generation - oop dummy = instanceKlass::cast(SystemDictionary::object_klass())->allocate_instance(CHECK); + oop dummy = instanceKlass::cast(SystemDictionary::Object_klass())->allocate_instance(CHECK); dummy_array->obj_at_put(i++, dummy); } // Allocate dummy in permanent generation - oop dummy = instanceKlass::cast(SystemDictionary::object_klass())->allocate_permanent_instance(CHECK); + oop dummy = instanceKlass::cast(SystemDictionary::Object_klass())->allocate_permanent_instance(CHECK); dummy_array->obj_at_put(i++, dummy); } { @@ -533,7 +540,7 @@ // but we cannot do that for classes created before java.lang.Class is loaded. Here we simply // walk over permanent objects created so far (mostly classes) and fixup their mirrors. Note // that the number of objects allocated at this point is very small. - assert(SystemDictionary::class_klass_loaded(), "java.lang.Class should be loaded"); + assert(SystemDictionary::Class_klass_loaded(), "java.lang.Class should be loaded"); FixupMirrorClosure blk; Universe::heap()->permanent_object_iterate(&blk); } @@ -549,7 +556,7 @@ if (TraceReferenceGC) tty->print_cr("Callback to run finalizers on exit"); { PRESERVE_EXCEPTION_MARK; - KlassHandle finalizer_klass(THREAD, SystemDictionary::finalizer_klass()); + KlassHandle finalizer_klass(THREAD, SystemDictionary::Finalizer_klass()); JavaValue result(T_VOID); JavaCalls::call_static( &result, @@ -744,22 +751,22 @@ static const uint64_t OopEncodingHeapMax = NarrowOopHeapMax << LogMinObjAlignmentInBytes; char* Universe::preferred_heap_base(size_t heap_size, NARROW_OOP_MODE mode) { + size_t base = 0; #ifdef _LP64 if (UseCompressedOops) { assert(mode == UnscaledNarrowOop || mode == ZeroBasedNarrowOop || mode == HeapBasedNarrowOop, "mode is invalid"); + const size_t total_size = heap_size + HeapBaseMinAddress; // Return specified base for the first request. if (!FLAG_IS_DEFAULT(HeapBaseMinAddress) && (mode == UnscaledNarrowOop)) { - return (char*)HeapBaseMinAddress; - } - const size_t total_size = heap_size + HeapBaseMinAddress; - if (total_size <= OopEncodingHeapMax && (mode != HeapBasedNarrowOop)) { + base = HeapBaseMinAddress; + } else if (total_size <= OopEncodingHeapMax && (mode != HeapBasedNarrowOop)) { if (total_size <= NarrowOopHeapMax && (mode == UnscaledNarrowOop) && (Universe::narrow_oop_shift() == 0)) { // Use 32-bits oops without encoding and // place heap's top on the 4Gb boundary - return (char*)(NarrowOopHeapMax - heap_size); + base = (NarrowOopHeapMax - heap_size); } else { // Can't reserve with NarrowOopShift == 0 Universe::set_narrow_oop_shift(LogMinObjAlignmentInBytes); @@ -768,16 +775,38 @@ // Use zero based compressed oops with encoding and // place heap's top on the 32Gb boundary in case // total_size > 4Gb or failed to reserve below 4Gb. - return (char*)(OopEncodingHeapMax - heap_size); + base = (OopEncodingHeapMax - heap_size); } } } else { // Can't reserve below 32Gb. Universe::set_narrow_oop_shift(LogMinObjAlignmentInBytes); } + // Set narrow_oop_base and narrow_oop_use_implicit_null_checks + // used in ReservedHeapSpace() constructors. + // The final values will be set in initialize_heap() below. + if (base != 0 && (base + heap_size) <= OopEncodingHeapMax) { + // Use zero based compressed oops + Universe::set_narrow_oop_base(NULL); + // Don't need guard page for implicit checks in indexed + // addressing mode with zero based Compressed Oops. + Universe::set_narrow_oop_use_implicit_null_checks(true); + } else { + // Set to a non-NULL value so the ReservedSpace ctor computes + // the correct no-access prefix. + // The final value will be set in initialize_heap() below. + Universe::set_narrow_oop_base((address)NarrowOopHeapMax); +#ifdef _WIN64 + if (UseLargePages) { + // Cannot allocate guard pages for implicit checks in indexed + // addressing mode when large pages are specified on windows. + Universe::set_narrow_oop_use_implicit_null_checks(false); + } +#endif // _WIN64 + } } #endif - return NULL; // also return NULL (don't care) for 32-bit VM + return (char*)base; // also return NULL (don't care) for 32-bit VM } jint Universe::initialize_heap() { @@ -921,7 +950,7 @@ { ResourceMark rm; Interpreter::initialize(); // needed for interpreter entry points if (!UseSharedSpaces) { - KlassHandle ok_h(THREAD, SystemDictionary::object_klass()); + KlassHandle ok_h(THREAD, SystemDictionary::Object_klass()); Universe::reinitialize_vtable_of(ok_h, CHECK_false); Universe::reinitialize_itables(CHECK_false); } @@ -931,7 +960,7 @@ instanceKlassHandle k_h; if (!UseSharedSpaces) { // Setup preallocated empty java.lang.Class array - Universe::_the_empty_class_klass_array = oopFactory::new_objArray(SystemDictionary::class_klass(), 0, CHECK_false); + Universe::_the_empty_class_klass_array = oopFactory::new_objArray(SystemDictionary::Class_klass(), 0, CHECK_false); // Setup preallocated OutOfMemoryError errors k = SystemDictionary::resolve_or_fail(vmSymbolHandles::java_lang_OutOfMemoryError(), true, CHECK_false); k_h = instanceKlassHandle(THREAD, k); @@ -998,8 +1027,8 @@ // Setup static method for registering finalizers // The finalizer klass must be linked before looking up the method, in // case it needs to get rewritten. - instanceKlass::cast(SystemDictionary::finalizer_klass())->link_class(CHECK_false); - methodOop m = instanceKlass::cast(SystemDictionary::finalizer_klass())->find_method( + instanceKlass::cast(SystemDictionary::Finalizer_klass())->link_class(CHECK_false); + methodOop m = instanceKlass::cast(SystemDictionary::Finalizer_klass())->find_method( vmSymbols::register_method_name(), vmSymbols::register_method_signature()); if (m == NULL || !m->is_static()) { @@ -1007,7 +1036,7 @@ "java.lang.ref.Finalizer.register", false); } Universe::_finalizer_register_cache->init( - SystemDictionary::finalizer_klass(), m, CHECK_false); + SystemDictionary::Finalizer_klass(), m, CHECK_false); // Resolve on first use and initialize class. // Note: No race-condition here, since a resolve will always return the same result @@ -1024,14 +1053,14 @@ Universe::_reflect_invoke_cache->init(k_h(), m, CHECK_false); // Setup method for registering loaded classes in class loader vector - instanceKlass::cast(SystemDictionary::classloader_klass())->link_class(CHECK_false); - m = instanceKlass::cast(SystemDictionary::classloader_klass())->find_method(vmSymbols::addClass_name(), vmSymbols::class_void_signature()); + instanceKlass::cast(SystemDictionary::ClassLoader_klass())->link_class(CHECK_false); + m = instanceKlass::cast(SystemDictionary::ClassLoader_klass())->find_method(vmSymbols::addClass_name(), vmSymbols::class_void_signature()); if (m == NULL || m->is_static()) { THROW_MSG_(vmSymbols::java_lang_NoSuchMethodException(), "java.lang.ClassLoader.addClass", false); } Universe::_loader_addClass_cache->init( - SystemDictionary::classloader_klass(), m, CHECK_false); + SystemDictionary::ClassLoader_klass(), m, CHECK_false); // The folowing is initializing converter functions for serialization in // JVM.cpp. If we clean up the StrictMath code above we may want to find
--- a/src/share/vm/memory/universe.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/memory/universe.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -169,6 +169,8 @@ static objArrayOop _the_empty_system_obj_array; // Canonicalized system obj array static objArrayOop _the_empty_class_klass_array; // Canonicalized obj array of type java.lang.Class static objArrayOop _the_array_interfaces_array; // Canonicalized 2-array of cloneable & serializable klasses + static oop _the_null_string; // A cache of "null" as a Java string + static oop _the_min_jint_string; // A cache of "-2147483648" as a Java string static LatestMethodOopCache* _finalizer_register_cache; // static method for registering finalizable objects static LatestMethodOopCache* _loader_addClass_cache; // method for registering loaded classes in class loader vector static ActiveMethodOopsCache* _reflect_invoke_cache; // method for security checks @@ -310,6 +312,8 @@ static objArrayOop the_empty_system_obj_array () { return _the_empty_system_obj_array; } static objArrayOop the_empty_class_klass_array () { return _the_empty_class_klass_array; } static objArrayOop the_array_interfaces_array() { return _the_array_interfaces_array; } + static oop the_null_string() { return _the_null_string; } + static oop the_min_jint_string() { return _the_min_jint_string; } static methodOop finalizer_register_method() { return _finalizer_register_cache->get_methodOop(); } static methodOop loader_addClass_method() { return _loader_addClass_cache->get_methodOop(); } static ActiveMethodOopsCache* reflect_invoke_cache() { return _reflect_invoke_cache; }
--- a/src/share/vm/oops/arrayKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/arrayKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -43,7 +43,7 @@ if (super() == NULL) return NULL; // bootstrap case // Array klasses have primary supertypes which are not reported to Java. // Example super chain: String[][] -> Object[][] -> Object[] -> Object - return SystemDictionary::object_klass(); + return SystemDictionary::Object_klass(); } @@ -82,7 +82,7 @@ k = arrayKlassHandle(THREAD, base_klass()); assert(!k()->is_parsable(), "not expecting parsability yet."); - k->set_super(Universe::is_bootstrapping() ? (klassOop)NULL : SystemDictionary::object_klass()); + k->set_super(Universe::is_bootstrapping() ? (klassOop)NULL : SystemDictionary::Object_klass()); k->set_layout_helper(Klass::_lh_neutral_value); k->set_dimension(1); k->set_higher_dimension(NULL); @@ -117,9 +117,9 @@ bool arrayKlass::compute_is_subtype_of(klassOop k) { // An array is a subtype of Serializable, Clonable, and Object - return k == SystemDictionary::object_klass() - || k == SystemDictionary::cloneable_klass() - || k == SystemDictionary::serializable_klass(); + return k == SystemDictionary::Object_klass() + || k == SystemDictionary::Cloneable_klass() + || k == SystemDictionary::Serializable_klass(); }
--- a/src/share/vm/oops/arrayKlass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/arrayKlass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -67,7 +67,7 @@ // Compiler/Interpreter offset static ByteSize component_mirror_offset() { return byte_offset_of(arrayKlass, _component_mirror); } - virtual klassOop java_super() const;//{ return SystemDictionary::object_klass(); } + virtual klassOop java_super() const;//{ return SystemDictionary::Object_klass(); } // Allocation // Sizes points to the first dimension of the array, subsequent dimensions
--- a/src/share/vm/oops/arrayKlassKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/arrayKlassKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -159,7 +159,7 @@ assert(obj->is_klass(), "must be klass"); klassKlass::oop_print_on(obj, st); } - +#endif //PRODUCT void arrayKlassKlass::oop_print_value_on(oop obj, outputStream* st) { assert(obj->is_klass(), "must be klass"); @@ -168,7 +168,6 @@ st->print("[]"); } } -#endif const char* arrayKlassKlass::internal_name() const {
--- a/src/share/vm/oops/arrayKlassKlass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/arrayKlassKlass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -55,14 +55,13 @@ int oop_oop_iterate(oop obj, OopClosure* blk); int oop_oop_iterate_m(oop obj, OopClosure* blk, MemRegion mr); -#ifndef PRODUCT public: // Printing + void oop_print_value_on(oop obj, outputStream* st); +#ifndef PRODUCT void oop_print_on(oop obj, outputStream* st); - void oop_print_value_on(oop obj, outputStream* st); -#endif +#endif //PRODUCT - public: // Verification const char* internal_name() const; void oop_verify_on(oop obj, outputStream* st);
--- a/src/share/vm/oops/compiledICHolderKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/compiledICHolderKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -166,12 +166,12 @@ st->print(" - klass: "); c->holder_klass()->print_value_on(st); st->cr(); } +#endif //PRODUCT void compiledICHolderKlass::oop_print_value_on(oop obj, outputStream* st) { assert(obj->is_compiledICHolder(), "must be compiledICHolder"); Klass::oop_print_value_on(obj, st); } -#endif const char* compiledICHolderKlass::internal_name() const { return "{compiledICHolder}";
--- a/src/share/vm/oops/compiledICHolderKlass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/compiledICHolderKlass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -68,14 +68,13 @@ int oop_oop_iterate(oop obj, OopClosure* blk); int oop_oop_iterate_m(oop obj, OopClosure* blk, MemRegion mr); -#ifndef PRODUCT public: // Printing + void oop_print_value_on(oop obj, outputStream* st); +#ifndef PRODUCT void oop_print_on (oop obj, outputStream* st); - void oop_print_value_on(oop obj, outputStream* st); -#endif +#endif //PRODUCT - public: // Verification const char* internal_name() const; void oop_verify_on(oop obj, outputStream* st);
--- a/src/share/vm/oops/constMethodKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/constMethodKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -216,6 +216,7 @@ } } +#endif //PRODUCT // Short version of printing constMethodOop - just print the name of the // method it belongs to. @@ -226,8 +227,6 @@ m->method()->print_value_on(st); } -#endif // PRODUCT - const char* constMethodKlass::internal_name() const { return "{constMethod}"; }
--- a/src/share/vm/oops/constMethodKlass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/constMethodKlass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -77,15 +77,13 @@ int oop_oop_iterate(oop obj, OopClosure* blk); int oop_oop_iterate_m(oop obj, OopClosure* blk, MemRegion mr); -#ifndef PRODUCT public: // Printing + void oop_print_value_on(oop obj, outputStream* st); +#ifndef PRODUCT void oop_print_on (oop obj, outputStream* st); - void oop_print_value_on(oop obj, outputStream* st); +#endif //PRODUCT -#endif - - public: // Verify operations const char* internal_name() const; void oop_verify_on(oop obj, outputStream* st);
--- a/src/share/vm/oops/constMethodOop.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/constMethodOop.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -258,6 +258,11 @@ LocalVariableTableElement* localvariable_table_start() const; // byte codes + void set_code(address code) { + if (code_size() > 0) { + memcpy(code_base(), code, code_size()); + } + } address code_base() const { return (address) (this+1); } address code_end() const { return code_base() + code_size(); } bool contains(address bcp) const { return code_base() <= bcp
--- a/src/share/vm/oops/constantPoolKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/constantPoolKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -387,8 +387,18 @@ cp->set_cache(cache()); } +#endif -#endif +void constantPoolKlass::oop_print_value_on(oop obj, outputStream* st) { + assert(obj->is_constantPool(), "must be constantPool"); + constantPoolOop cp = constantPoolOop(obj); + st->print("constant pool [%d]", cp->length()); + if (cp->has_pseudo_string()) st->print("/pseudo_string"); + if (cp->has_invokedynamic()) st->print("/invokedynamic"); + cp->print_address_on(st); + st->print(" for "); + cp->pool_holder()->print_value_on(st); +} const char* constantPoolKlass::internal_name() const { return "{constant pool}";
--- a/src/share/vm/oops/constantPoolKlass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/constantPoolKlass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -65,9 +65,10 @@ juint alloc_size() const { return _alloc_size; } void set_alloc_size(juint n) { _alloc_size = n; } -#ifndef PRODUCT public: // Printing + void oop_print_value_on(oop obj, outputStream* st); +#ifndef PRODUCT void oop_print_on(oop obj, outputStream* st); #endif
--- a/src/share/vm/oops/constantPoolOop.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/constantPoolOop.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -110,7 +110,7 @@ } if (!PENDING_EXCEPTION-> - is_a(SystemDictionary::linkageError_klass())) { + is_a(SystemDictionary::LinkageError_klass())) { // Just throw the exception and don't prevent these classes from // being loaded due to virtual machine errors like StackOverflow // and OutOfMemoryError, etc, or if the thread was hit by stop() @@ -262,25 +262,48 @@ int constantPoolOopDesc::impl_name_and_type_ref_index_at(int which, bool uncached) { - jint ref_index = field_or_method_at(which, uncached); + int i = which; + if (!uncached && cache() != NULL) { + if (constantPoolCacheOopDesc::is_secondary_index(which)) + // Invokedynamic indexes are always processed in native order + // so there is no question of reading a native u2 in Java order here. + return cache()->main_entry_at(which)->constant_pool_index(); + // change byte-ordering and go via cache + i = remap_instruction_operand_from_cache(which); + } else { + if (tag_at(which).is_name_and_type()) + // invokedynamic index is a simple name-and-type + return which; + } + assert(tag_at(i).is_field_or_method(), "Corrupted constant pool"); + jint ref_index = *int_at_addr(i); return extract_high_short_from_int(ref_index); } int constantPoolOopDesc::impl_klass_ref_index_at(int which, bool uncached) { - jint ref_index = field_or_method_at(which, uncached); + guarantee(!constantPoolCacheOopDesc::is_secondary_index(which), + "an invokedynamic instruction does not have a klass"); + int i = which; + if (!uncached && cache() != NULL) { + // change byte-ordering and go via cache + i = remap_instruction_operand_from_cache(which); + } + assert(tag_at(i).is_field_or_method(), "Corrupted constant pool"); + jint ref_index = *int_at_addr(i); return extract_low_short_from_int(ref_index); } -int constantPoolOopDesc::map_instruction_operand_to_index(int operand) { - if (constantPoolCacheOopDesc::is_secondary_index(operand)) { - return cache()->main_entry_at(operand)->constant_pool_index(); - } +int constantPoolOopDesc::remap_instruction_operand_from_cache(int operand) { + // Operand was fetched by a stream using get_Java_u2, yet was stored + // by Rewriter::rewrite_member_reference in native order. + // So now we have to fix the damage by swapping back to native order. assert((int)(u2)operand == operand, "clean u2"); - int index = Bytes::swap_u2(operand); - return cache()->entry_at(index)->constant_pool_index(); + int cpc_index = Bytes::swap_u2(operand); + int member_index = cache()->entry_at(cpc_index)->constant_pool_index(); + return member_index; }
--- a/src/share/vm/oops/constantPoolOop.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/constantPoolOop.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -191,6 +191,16 @@ } } + void object_at_put(int which, oop str) { + oop_store((volatile oop*) obj_at_addr(which), str); + release_tag_at_put(which, JVM_CONSTANT_Object); + if (UseConcMarkSweepGC) { + // In case the earlier card-mark was consumed by a concurrent + // marking thread before the tag was updated, redirty the card. + oop_store_without_check((volatile oop*) obj_at_addr(which), str); + } + } + // For temporary use while constructing constant pool void string_index_at_put(int which, int string_index) { tag_at_put(which, JVM_CONSTANT_StringIndex); @@ -228,7 +238,8 @@ tag.is_unresolved_klass() || tag.is_symbol() || tag.is_unresolved_string() || - tag.is_string(); + tag.is_string() || + tag.is_object(); } // Fetching constants @@ -291,6 +302,11 @@ return string_at_impl(h_this, which, CHECK_NULL); } + oop object_at(int which) { + assert(tag_at(which).is_object(), "Corrupted constant pool"); + return *obj_at_addr(which); + } + // A "pseudo-string" is an non-string oop that has found is way into // a String entry. // Under AnonymousClasses this can happen if the user patches a live @@ -342,12 +358,14 @@ } // The following methods (name/signature/klass_ref_at, klass_ref_at_noresolve, - // name_and_type_ref_index_at) all expect constant pool indices - // from the bytecodes to be passed in, which are actually potentially byte-swapped - // or rewritten constant pool cache indices. They all call map_instruction_operand_to_index. - int map_instruction_operand_to_index(int operand); + // name_and_type_ref_index_at) all expect to be passed indices obtained + // directly from the bytecode, and extracted according to java byte order. + // If the indices are meant to refer to fields or methods, they are + // actually potentially byte-swapped, rewritten constant pool cache indices. + // The routine remap_instruction_operand_from_cache manages the adjustment + // of these values back to constant pool indices. - // There are also "uncached" versions which do not map the operand index; see below. + // There are also "uncached" versions which do not adjust the operand index; see below. // Lookup for entries consisting of (klass_index, name_and_type index) klassOop klass_ref_at(int which, TRAPS); @@ -361,8 +379,6 @@ // Lookup for entries consisting of (name_index, signature_index) int name_ref_index_at(int which_nt); // == low-order jshort of name_and_type_at(which_nt) int signature_ref_index_at(int which_nt); // == high-order jshort of name_and_type_at(which_nt) - symbolOop nt_name_ref_at(int which_nt) { return symbol_at(name_ref_index_at(which_nt)); } - symbolOop nt_signature_ref_at(int which_nt) { return symbol_at(signature_ref_index_at(which_nt)); } BasicType basic_type_for_signature_at(int which); @@ -425,18 +441,7 @@ int impl_klass_ref_index_at(int which, bool uncached); int impl_name_and_type_ref_index_at(int which, bool uncached); - // Takes either a constant pool cache index in possibly byte-swapped - // byte order (which comes from the bytecodes after rewriting) or, - // if "uncached" is true, a vanilla constant pool index - jint field_or_method_at(int which, bool uncached) { - int i = which; - if (!uncached && cache() != NULL) { - // change byte-ordering and go via cache - i = map_instruction_operand_to_index(which); - } - assert(tag_at(i).is_field_or_method(), "Corrupted constant pool"); - return *int_at_addr(i); - } + int remap_instruction_operand_from_cache(int operand); // Used while constructing constant pool (only by ClassFileParser) jint klass_index_at(int which) {
--- a/src/share/vm/oops/cpCacheKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/cpCacheKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -261,6 +261,15 @@ #endif +void constantPoolCacheKlass::oop_print_value_on(oop obj, outputStream* st) { + assert(obj->is_constantPoolCache(), "obj must be constant pool cache"); + constantPoolCacheOop cache = (constantPoolCacheOop)obj; + st->print("cache [%d]", cache->length()); + cache->print_address_on(st); + st->print(" for "); + cache->constant_pool()->print_value_on(st); +} + void constantPoolCacheKlass::oop_verify_on(oop obj, outputStream* st) { guarantee(obj->is_constantPoolCache(), "obj must be constant pool cache"); constantPoolCacheOop cache = (constantPoolCacheOop)obj;
--- a/src/share/vm/oops/cpCacheKlass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/cpCacheKlass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -61,9 +61,10 @@ juint alloc_size() const { return _alloc_size; } void set_alloc_size(juint n) { _alloc_size = n; } -#ifndef PRODUCT public: // Printing + void oop_print_value_on(oop obj, outputStream* st); +#ifndef PRODUCT void oop_print_on(oop obj, outputStream* st); #endif
--- a/src/share/vm/oops/cpCacheOop.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/cpCacheOop.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -28,21 +28,17 @@ // Implememtation of ConstantPoolCacheEntry -void ConstantPoolCacheEntry::set_initial_state(int index) { - if (constantPoolCacheOopDesc::is_secondary_index(index)) { - // Hack: The rewriter is trying to say that this entry itself - // will be a secondary entry. - int main_index = constantPoolCacheOopDesc::decode_secondary_index(index); - assert(0 <= main_index && main_index < 0x10000, "sanity check"); - _indices = (main_index << 16); - assert(main_entry_index() == main_index, ""); - return; - } +void ConstantPoolCacheEntry::initialize_entry(int index) { assert(0 < index && index < 0x10000, "sanity check"); _indices = index; assert(constant_pool_index() == index, ""); } +void ConstantPoolCacheEntry::initialize_secondary_entry(int main_index) { + assert(0 <= main_index && main_index < 0x10000, "sanity check"); + _indices = (main_index << 16); + assert(main_entry_index() == main_index, ""); +} int ConstantPoolCacheEntry::as_flags(TosState state, bool is_final, bool is_vfinal, bool is_volatile, @@ -223,10 +219,10 @@ void ConstantPoolCacheEntry::set_dynamic_call(Handle call_site, int extra_data) { - methodOop method = (methodOop) sun_dyn_CallSiteImpl::vmmethod(call_site()); + methodOop method = (methodOop) java_dyn_CallSite::vmmethod(call_site()); assert(method->is_method(), "must be initialized properly"); int param_size = method->size_of_parameters(); - assert(param_size > 1, "method argument size must include MH.this & initial dynamic receiver"); + assert(param_size >= 1, "method argument size must include MH.this"); param_size -= 1; // do not count MH.this; it is not stacked for invokedynamic if (Atomic::cmpxchg_ptr(call_site(), &_f1, NULL) == NULL) { // racing threads might be trying to install their own favorites @@ -439,7 +435,18 @@ void constantPoolCacheOopDesc::initialize(intArray& inverse_index_map) { assert(inverse_index_map.length() == length(), "inverse index map must have same length as cache"); - for (int i = 0; i < length(); i++) entry_at(i)->set_initial_state(inverse_index_map[i]); + for (int i = 0; i < length(); i++) { + ConstantPoolCacheEntry* e = entry_at(i); + int original_index = inverse_index_map[i]; + if ((original_index & Rewriter::_secondary_entry_tag) != 0) { + int main_index = (original_index - Rewriter::_secondary_entry_tag); + assert(!entry_at(main_index)->is_secondary_entry(), "valid main index"); + e->initialize_secondary_entry(main_index); + } else { + e->initialize_entry(original_index); + } + assert(entry_at(i) == e, "sanity"); + } } // RedefineClasses() API support:
--- a/src/share/vm/oops/cpCacheOop.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/cpCacheOop.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -154,7 +154,8 @@ }; // Initialization - void set_initial_state(int index); // sets entry to initial state + void initialize_entry(int original_index); // initialize primary entry + void initialize_secondary_entry(int main_index); // initialize secondary entry void set_field( // sets entry to resolved field state Bytecodes::Code get_code, // the bytecode used for reading the field @@ -251,6 +252,7 @@ // Code generation support static WordSize size() { return in_WordSize(sizeof(ConstantPoolCacheEntry) / HeapWordSize); } + static ByteSize size_in_bytes() { return in_ByteSize(sizeof(ConstantPoolCacheEntry)); } static ByteSize indices_offset() { return byte_offset_of(ConstantPoolCacheEntry, _indices); } static ByteSize f1_offset() { return byte_offset_of(ConstantPoolCacheEntry, _f1); } static ByteSize f2_offset() { return byte_offset_of(ConstantPoolCacheEntry, _f2); } @@ -321,6 +323,7 @@ ConstantPoolCacheEntry* base() const { return (ConstantPoolCacheEntry*)((address)this + in_bytes(base_offset())); } friend class constantPoolCacheKlass; + friend class ConstantPoolCacheEntry; public: // Initialization @@ -329,7 +332,8 @@ // Secondary indexes. // They must look completely different from normal indexes. // The main reason is that byte swapping is sometimes done on normal indexes. - // Also, it is helpful for debugging to tell the two apart. + // Also, some of the CP accessors do different things for secondary indexes. + // Finally, it is helpful for debugging to tell the two apart. static bool is_secondary_index(int i) { return (i < 0); } static int decode_secondary_index(int i) { assert(is_secondary_index(i), ""); return ~i; } static int encode_secondary_index(int i) { assert(!is_secondary_index(i), ""); return ~i; } @@ -337,18 +341,35 @@ // Accessors void set_constant_pool(constantPoolOop pool) { oop_store_without_check((oop*)&_constant_pool, (oop)pool); } constantPoolOop constant_pool() const { return _constant_pool; } - ConstantPoolCacheEntry* entry_at(int i) const { assert(0 <= i && i < length(), "index out of bounds"); return base() + i; } + // Fetches the entry at the given index. + // The entry may be either primary or secondary. + // In either case the index must not be encoded or byte-swapped in any way. + ConstantPoolCacheEntry* entry_at(int i) const { + assert(0 <= i && i < length(), "index out of bounds"); + return base() + i; + } + // Fetches the secondary entry referred to by index. + // The index may be a secondary index, and must not be byte-swapped. + ConstantPoolCacheEntry* secondary_entry_at(int i) const { + int raw_index = i; + if (is_secondary_index(i)) { // correct these on the fly + raw_index = decode_secondary_index(i); + } + assert(entry_at(raw_index)->is_secondary_entry(), "not a secondary entry"); + return entry_at(raw_index); + } + // Given a primary or secondary index, fetch the corresponding primary entry. + // Indirect through the secondary entry, if the index is encoded as a secondary index. + // The index must not be byte-swapped. ConstantPoolCacheEntry* main_entry_at(int i) const { - ConstantPoolCacheEntry* e; + int primary_index = i; if (is_secondary_index(i)) { // run through an extra level of indirection: - i = decode_secondary_index(i); - e = entry_at(i); - i = e->main_entry_index(); + int raw_index = decode_secondary_index(i); + primary_index = entry_at(raw_index)->main_entry_index(); } - e = entry_at(i); - assert(!e->is_secondary_entry(), "only one level of indirection"); - return e; + assert(!entry_at(primary_index)->is_secondary_entry(), "only one level of indirection"); + return entry_at(primary_index); } // GC support @@ -359,6 +380,12 @@ // Code generation static ByteSize base_offset() { return in_ByteSize(sizeof(constantPoolCacheOopDesc)); } + static ByteSize entry_offset(int raw_index) { + int index = raw_index; + if (is_secondary_index(raw_index)) + index = decode_secondary_index(raw_index); + return (base_offset() + ConstantPoolCacheEntry::size_in_bytes() * index); + } // RedefineClasses() API support: // If any entry of this constantPoolCache points to any of
--- a/src/share/vm/oops/generateOopMap.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/generateOopMap.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1559,7 +1559,7 @@ case Bytecodes::_invokevirtual: case Bytecodes::_invokespecial: do_method(false, false, itr->get_index_big(), itr->bci()); break; case Bytecodes::_invokestatic: do_method(true, false, itr->get_index_big(), itr->bci()); break; - case Bytecodes::_invokedynamic: do_method(false, true, itr->get_index_int(), itr->bci()); break; + case Bytecodes::_invokedynamic: do_method(true, false, itr->get_index_int(), itr->bci()); break; case Bytecodes::_invokeinterface: do_method(false, true, itr->get_index_big(), itr->bci()); break; case Bytecodes::_newarray: case Bytecodes::_anewarray: pp_new_ref(vCTS, itr->bci()); break; @@ -1830,12 +1830,8 @@ void GenerateOopMap::do_ldc(int idx, int bci) { - constantPoolOop cp = method()->constants(); - constantTag tag = cp->tag_at(idx); - - CellTypeState cts = (tag.is_string() || tag.is_unresolved_string() || - tag.is_klass() || tag.is_unresolved_klass()) - ? CellTypeState::make_line_ref(bci) : valCTS; + constantPoolOop cp = method()->constants(); + CellTypeState cts = cp->is_pointer_entry(idx) ? CellTypeState::make_line_ref(bci) : valCTS; ppush1(cts); } @@ -1900,11 +1896,9 @@ } void GenerateOopMap::do_method(int is_static, int is_interface, int idx, int bci) { - // Dig up signature for field in constant pool - constantPoolOop cp = _method->constants(); - int nameAndTypeIdx = cp->name_and_type_ref_index_at(idx); - int signatureIdx = cp->signature_ref_index_at(nameAndTypeIdx); // @@@@@ - symbolOop signature = cp->symbol_at(signatureIdx); + // Dig up signature for field in constant pool + constantPoolOop cp = _method->constants(); + symbolOop signature = cp->signature_ref_at(idx); // Parse method signature CellTypeState out[4];
--- a/src/share/vm/oops/instanceKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/instanceKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -383,7 +383,7 @@ this_oop->set_initialization_state_and_notify(initialization_error, THREAD); CLEAR_PENDING_EXCEPTION; // ignore any exception thrown, class initialization error is thrown below } - if (e->is_a(SystemDictionary::error_klass())) { + if (e->is_a(SystemDictionary::Error_klass())) { THROW_OOP(e()); } else { JavaCallArguments args(e); @@ -568,7 +568,7 @@ THROW_MSG(throwError ? vmSymbols::java_lang_InstantiationError() : vmSymbols::java_lang_InstantiationException(), external_name()); } - if (as_klassOop() == SystemDictionary::class_klass()) { + if (as_klassOop() == SystemDictionary::Class_klass()) { ResourceMark rm(THREAD); THROW_MSG(throwError ? vmSymbols::java_lang_IllegalAccessError() : vmSymbols::java_lang_IllegalAccessException(), external_name()); @@ -2045,8 +2045,9 @@ // As we walk along, look for equalities between outer1 and class2. // Eventually, the walks will terminate as outer1 stops // at the top-level class around the original class. - symbolOop ignore_name; - klassOop next = outer1->compute_enclosing_class(ignore_name, CHECK_false); + bool ignore_inner_is_member; + klassOop next = outer1->compute_enclosing_class(&ignore_inner_is_member, + CHECK_false); if (next == NULL) break; if (next == class2()) return true; outer1 = instanceKlassHandle(THREAD, next); @@ -2055,8 +2056,9 @@ // Now do the same for class2. instanceKlassHandle outer2 = class2; for (;;) { - symbolOop ignore_name; - klassOop next = outer2->compute_enclosing_class(ignore_name, CHECK_false); + bool ignore_inner_is_member; + klassOop next = outer2->compute_enclosing_class(&ignore_inner_is_member, + CHECK_false); if (next == NULL) break; // Might as well check the new outer against all available values. if (next == class1()) return true; @@ -2223,7 +2225,7 @@ void instanceKlass::oop_print_on(oop obj, outputStream* st) { Klass::oop_print_on(obj, st); - if (as_klassOop() == SystemDictionary::string_klass()) { + if (as_klassOop() == SystemDictionary::String_klass()) { typeArrayOop value = java_lang_String::value(obj); juint offset = java_lang_String::offset(obj); juint length = java_lang_String::length(obj); @@ -2243,7 +2245,7 @@ FieldPrinter print_nonstatic_field(st, obj); do_nonstatic_fields(&print_nonstatic_field); - if (as_klassOop() == SystemDictionary::class_klass()) { + if (as_klassOop() == SystemDictionary::Class_klass()) { st->print(BULLET"signature: "); java_lang_Class::print_signature(obj, st); st->cr(); @@ -2266,11 +2268,13 @@ } } +#endif //PRODUCT + void instanceKlass::oop_print_value_on(oop obj, outputStream* st) { st->print("a "); name()->print_value_on(st); obj->print_address_on(st); - if (as_klassOop() == SystemDictionary::string_klass() + if (as_klassOop() == SystemDictionary::String_klass() && java_lang_String::value(obj) != NULL) { ResourceMark rm; int len = java_lang_String::length(obj); @@ -2279,7 +2283,7 @@ st->print(" = \"%s\"", str); if (len > plen) st->print("...[%d]", len); - } else if (as_klassOop() == SystemDictionary::class_klass()) { + } else if (as_klassOop() == SystemDictionary::Class_klass()) { klassOop k = java_lang_Class::as_klassOop(obj); st->print(" = "); if (k != NULL) { @@ -2297,8 +2301,6 @@ } } -#endif // ndef PRODUCT - const char* instanceKlass::internal_name() const { return external_name(); } @@ -2346,7 +2348,7 @@ // Check that we have the right class static bool first_time = true; - guarantee(k == SystemDictionary::class_klass() && first_time, "Invalid verify of maps"); + guarantee(k == SystemDictionary::Class_klass() && first_time, "Invalid verify of maps"); first_time = false; const int extra = java_lang_Class::number_of_fake_oop_fields; guarantee(ik->nonstatic_field_size() == extra, "just checking");
--- a/src/share/vm/oops/instanceKlass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/instanceKlass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -337,12 +337,12 @@ static bool is_same_class_package(oop class_loader1, symbolOop class_name1, oop class_loader2, symbolOop class_name2); // find an enclosing class (defined where original code was, in jvm.cpp!) - klassOop compute_enclosing_class(symbolOop& simple_name_result, TRAPS) { + klassOop compute_enclosing_class(bool* inner_is_member, TRAPS) { instanceKlassHandle self(THREAD, this->as_klassOop()); - return compute_enclosing_class_impl(self, simple_name_result, THREAD); + return compute_enclosing_class_impl(self, inner_is_member, THREAD); } static klassOop compute_enclosing_class_impl(instanceKlassHandle self, - symbolOop& simple_name_result, TRAPS); + bool* inner_is_member, TRAPS); // tell if two classes have the same enclosing class (at package level) bool is_same_package_member(klassOop class2, TRAPS) { @@ -839,17 +839,16 @@ // JVMTI support jint jvmti_class_status() const; -#ifndef PRODUCT public: // Printing + void oop_print_value_on(oop obj, outputStream* st); +#ifndef PRODUCT void oop_print_on (oop obj, outputStream* st); - void oop_print_value_on(oop obj, outputStream* st); void print_dependent_nmethods(bool verbose = false); bool is_dependent_nmethod(nmethod* nm); #endif - public: // Verification const char* internal_name() const; void oop_verify_on(oop obj, outputStream* st);
--- a/src/share/vm/oops/instanceKlassKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/instanceKlassKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -317,6 +317,11 @@ pm->claim_or_forward_breadth(sg_addr); } + oop* bsm_addr = ik->adr_bootstrap_method(); + if (PSScavenge::should_scavenge(bsm_addr)) { + pm->claim_or_forward_breadth(bsm_addr); + } + klassKlass::oop_copy_contents(pm, obj); } @@ -345,6 +350,11 @@ pm->claim_or_forward_depth(sg_addr); } + oop* bsm_addr = ik->adr_bootstrap_method(); + if (PSScavenge::should_scavenge(bsm_addr)) { + pm->claim_or_forward_depth(bsm_addr); + } + klassKlass::oop_copy_contents(pm, obj); } @@ -628,6 +638,7 @@ st->cr(); } +#endif //PRODUCT void instanceKlassKlass::oop_print_value_on(oop obj, outputStream* st) { assert(obj->is_klass(), "must be klass"); @@ -635,8 +646,6 @@ ik->name()->print_value_on(st); } -#endif // PRODUCT - const char* instanceKlassKlass::internal_name() const { return "{instance class}"; }
--- a/src/share/vm/oops/instanceKlassKlass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/instanceKlassKlass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -69,14 +69,13 @@ // Apply closure to the InstanceKlass oops that are outside the java heap. inline void iterate_c_heap_oops(instanceKlass* ik, OopClosure* closure); -#ifndef PRODUCT public: // Printing + void oop_print_value_on(oop obj, outputStream* st); +#ifndef PRODUCT void oop_print_on(oop obj, outputStream* st); - void oop_print_value_on(oop obj, outputStream* st); #endif - public: // Verification const char* internal_name() const; void oop_verify_on(oop obj, outputStream* st);
--- a/src/share/vm/oops/instanceRefKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/instanceRefKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -78,9 +78,9 @@ #ifndef SERIALGC template <class T> -static void specialized_oop_follow_contents(instanceRefKlass* ref, - ParCompactionManager* cm, - oop obj) { +void specialized_oop_follow_contents(instanceRefKlass* ref, + ParCompactionManager* cm, + oop obj) { T* referent_addr = (T*)java_lang_ref_Reference::referent_addr(obj); T heap_oop = oopDesc::load_heap_oop(referent_addr); debug_only( @@ -397,7 +397,7 @@ // Check that we have the right class debug_only(static bool first_time = true); - assert(k == SystemDictionary::reference_klass() && first_time, + assert(k == SystemDictionary::Reference_klass() && first_time, "Invalid update of maps"); debug_only(first_time = false); assert(ik->nonstatic_oop_map_count() == 1, "just checking");
--- a/src/share/vm/oops/klass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/klass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -217,8 +217,8 @@ set_super(NULL); oop_store_without_check((oop*) &_primary_supers[0], (oop) this->as_klassOop()); assert(super_depth() == 0, "Object must already be initialized properly"); - } else if (k != super() || k == SystemDictionary::object_klass()) { - assert(super() == NULL || super() == SystemDictionary::object_klass(), + } else if (k != super() || k == SystemDictionary::Object_klass()) { + assert(super() == NULL || super() == SystemDictionary::Object_klass(), "initialize this only once to a non-trivial value"); set_super(k); Klass* sup = k->klass_part(); @@ -370,7 +370,7 @@ void Klass::remove_from_sibling_list() { // remove receiver from sibling list instanceKlass* super = superklass(); - assert(super != NULL || as_klassOop() == SystemDictionary::object_klass(), "should have super"); + assert(super != NULL || as_klassOop() == SystemDictionary::Object_klass(), "should have super"); if (super == NULL) return; // special case: class Object if (super->subklass() == this) { // first subklass @@ -541,6 +541,7 @@ st->cr(); } +#endif //PRODUCT void Klass::oop_print_value_on(oop obj, outputStream* st) { // print title @@ -549,8 +550,6 @@ obj->print_address_on(st); } -#endif - // Verification void Klass::oop_verify_on(oop obj, outputStream* st) {
--- a/src/share/vm/oops/klass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/klass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -776,14 +776,13 @@ // JVMTI support virtual jint jvmti_class_status() const; -#ifndef PRODUCT public: // Printing + virtual void oop_print_value_on(oop obj, outputStream* st); +#ifndef PRODUCT virtual void oop_print_on (oop obj, outputStream* st); - virtual void oop_print_value_on(oop obj, outputStream* st); -#endif +#endif //PRODUCT - public: // Verification virtual const char* internal_name() const = 0; virtual void oop_verify_on(oop obj, outputStream* st);
--- a/src/share/vm/oops/klassKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/klassKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -202,13 +202,12 @@ Klass::oop_print_on(obj, st); } +#endif //PRODUCT void klassKlass::oop_print_value_on(oop obj, outputStream* st) { Klass::oop_print_value_on(obj, st); } -#endif - const char* klassKlass::internal_name() const { return "{other class}"; }
--- a/src/share/vm/oops/klassKlass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/klassKlass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -67,14 +67,13 @@ juint alloc_size() const { return _alloc_size; } void set_alloc_size(juint n) { _alloc_size = n; } -#ifndef PRODUCT public: // Printing + void oop_print_value_on(oop obj, outputStream* st); +#ifndef PRODUCT void oop_print_on (oop obj, outputStream* st); - void oop_print_value_on(oop obj, outputStream* st); -#endif +#endif //PRODUCT - public: // Verification const char* internal_name() const; void oop_verify_on(oop obj, outputStream* st);
--- a/src/share/vm/oops/methodDataKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/methodDataKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -214,6 +214,8 @@ m->print_data_on(st); } +#endif //PRODUCT + void methodDataKlass::oop_print_value_on(oop obj, outputStream* st) { assert(obj->is_methodData(), "should be method data"); methodDataOop m = methodDataOop(obj); @@ -221,8 +223,6 @@ m->method()->print_value_on(st); } -#endif // !PRODUCT - const char* methodDataKlass::internal_name() const { return "{method data}"; }
--- a/src/share/vm/oops/methodDataKlass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/methodDataKlass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -71,14 +71,13 @@ int oop_oop_iterate(oop obj, OopClosure* blk); int oop_oop_iterate_m(oop obj, OopClosure* blk, MemRegion mr); -#ifndef PRODUCT public: // Printing + void oop_print_value_on(oop obj, outputStream* st); +#ifndef PRODUCT void oop_print_on (oop obj, outputStream* st); - void oop_print_value_on(oop obj, outputStream* st); -#endif // !PRODUCT +#endif //PRODUCT - public: // Verify operations const char* internal_name() const; void oop_verify_on(oop obj, outputStream* st);
--- a/src/share/vm/oops/methodDataOop.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/methodDataOop.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2000-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2000-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -545,6 +545,10 @@ return cell_offset(counter_cell_count); } + void set_count(uint count) { + set_uint_at(count_off, count); + } + #ifndef PRODUCT void print_data_on(outputStream* st); #endif @@ -692,6 +696,23 @@ void clear_row(uint row) { assert(row < row_limit(), "oob"); + // Clear total count - indicator of polymorphic call site. + // The site may look like as monomorphic after that but + // it allow to have more accurate profiling information because + // there was execution phase change since klasses were unloaded. + // If the site is still polymorphic then MDO will be updated + // to reflect it. But it could be the case that the site becomes + // only bimorphic. Then keeping total count not 0 will be wrong. + // Even if we use monomorphic (when it is not) for compilation + // we will only have trap, deoptimization and recompile again + // with updated MDO after executing method in Interpreter. + // An additional receiver will be recorded in the cleaned row + // during next call execution. + // + // Note: our profiling logic works with empty rows in any slot. + // We do sorting a profiling info (ciCallProfile) for compilation. + // + set_count(0); set_receiver(row, NULL); set_receiver_count(row, 0); } @@ -1391,6 +1412,9 @@ } void inc_decompile_count() { _nof_decompiles += 1; + if (decompile_count() > (uint)PerMethodRecompilationCutoff) { + method()->set_not_compilable(); + } } // Support for code generation
--- a/src/share/vm/oops/methodKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/methodKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -308,6 +308,7 @@ } } +#endif //PRODUCT void methodKlass::oop_print_value_on(oop obj, outputStream* st) { assert(obj->is_method(), "must be method"); @@ -323,8 +324,6 @@ if (WizardMode && m->code() != NULL) st->print(" ((nmethod*)%p)", m->code()); } -#endif // PRODUCT - const char* methodKlass::internal_name() const { return "{method}"; }
--- a/src/share/vm/oops/methodKlass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/methodKlass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -68,14 +68,13 @@ int oop_oop_iterate(oop obj, OopClosure* blk); int oop_oop_iterate_m(oop obj, OopClosure* blk, MemRegion mr); -#ifndef PRODUCT public: // Printing + void oop_print_value_on(oop obj, outputStream* st); +#ifndef PRODUCT void oop_print_on (oop obj, outputStream* st); - void oop_print_value_on(oop obj, outputStream* st); -#endif +#endif //PRODUCT - public: // Verify operations const char* internal_name() const; void oop_verify_on(oop obj, outputStream* st);
--- a/src/share/vm/oops/methodOop.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/methodOop.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -456,12 +456,12 @@ return objArrayHandle(THREAD, Universe::the_empty_class_klass_array()); } else { methodHandle h_this(THREAD, this_oop); - objArrayOop m_oop = oopFactory::new_objArray(SystemDictionary::class_klass(), length, CHECK_(objArrayHandle())); + objArrayOop m_oop = oopFactory::new_objArray(SystemDictionary::Class_klass(), length, CHECK_(objArrayHandle())); objArrayHandle mirrors (THREAD, m_oop); for (int i = 0; i < length; i++) { CheckedExceptionElement* table = h_this->checked_exceptions_start(); // recompute on each iteration, not gc safe klassOop k = h_this->constants()->klass_at(table[i].class_cp_index, CHECK_(objArrayHandle())); - assert(Klass::cast(k)->is_subclass_of(SystemDictionary::throwable_klass()), "invalid exception class"); + assert(Klass::cast(k)->is_subclass_of(SystemDictionary::Throwable_klass()), "invalid exception class"); mirrors->obj_at_put(i, Klass::cast(k)->java_mirror()); } return mirrors; @@ -575,12 +575,6 @@ return true; } - methodDataOop mdo = method_data(); - if (mdo != NULL - && (uint)mdo->decompile_count() > (uint)PerMethodRecompilationCutoff) { - // Since (uint)-1 is large, -1 really means 'no cutoff'. - return true; - } #ifdef COMPILER2 if (is_tier1_compile(comp_level)) { if (is_not_tier1_compilable()) { @@ -593,7 +587,16 @@ } // call this when compiler finds that this method is not compilable -void methodOopDesc::set_not_compilable(int comp_level) { +void methodOopDesc::set_not_compilable(int comp_level, bool report) { + if (PrintCompilation && report) { + ttyLocker ttyl; + tty->print("made not compilable "); + this->print_short_name(tty); + int size = this->code_size(); + if (size > 0) + tty->print(" (%d bytes)", size); + tty->cr(); + } if ((TraceDeoptimization || LogCompilation) && (xtty != NULL)) { ttyLocker ttyl; xtty->begin_elem("make_not_compilable thread='%d'", (int) os::current_thread_id()); @@ -688,7 +691,7 @@ // so making them eagerly shouldn't be too expensive. AdapterHandlerEntry* adapter = AdapterHandlerLibrary::get_adapter(mh); if (adapter == NULL ) { - THROW_0(vmSymbols::java_lang_OutOfMemoryError()); + THROW_MSG_NULL(vmSymbols::java_lang_VirtualMachineError(), "out of space in CodeCache for adapters"); } mh->set_adapter_entry(adapter); @@ -705,6 +708,16 @@ // This function must not hit a safepoint! address methodOopDesc::verified_code_entry() { debug_only(No_Safepoint_Verifier nsv;) + nmethod *code = (nmethod *)OrderAccess::load_ptr_acquire(&_code); + if (code == NULL && UseCodeCacheFlushing) { + nmethod *saved_code = CodeCache::find_and_remove_saved_code(this); + if (saved_code != NULL) { + methodHandle method(this); + assert( ! saved_code->is_osr_method(), "should not get here for osr" ); + set_code( method, saved_code ); + } + } + assert(_from_compiled_entry != NULL, "must be set"); return _from_compiled_entry; } @@ -733,8 +746,8 @@ int comp_level = code->comp_level(); // In theory there could be a race here. In practice it is unlikely // and not worth worrying about. - if (comp_level > highest_tier_compile()) { - set_highest_tier_compile(comp_level); + if (comp_level > mh->highest_tier_compile()) { + mh->set_highest_tier_compile(comp_level); } OrderAccess::storestore(); @@ -821,6 +834,18 @@ return pchase; } +//------------------------------------------------------------------------------ +// methodOopDesc::is_method_handle_adapter +// +// Tests if this method is an internal adapter frame from the +// MethodHandleCompiler. +bool methodOopDesc::is_method_handle_adapter() const { + return ((name() == vmSymbols::invoke_name() && + method_holder() == SystemDictionary::MethodHandle_klass()) + || + method_holder() == SystemDictionary::InvokeDynamic_klass()); +} + methodHandle methodOopDesc::make_invoke_method(KlassHandle holder, symbolHandle signature, Handle method_type, TRAPS) { @@ -1032,8 +1057,8 @@ // We are loading classes eagerly. If a ClassNotFoundException or // a LinkageError was generated, be sure to ignore it. if (HAS_PENDING_EXCEPTION) { - if (PENDING_EXCEPTION->is_a(SystemDictionary::classNotFoundException_klass()) || - PENDING_EXCEPTION->is_a(SystemDictionary::linkageError_klass())) { + if (PENDING_EXCEPTION->is_a(SystemDictionary::ClassNotFoundException_klass()) || + PENDING_EXCEPTION->is_a(SystemDictionary::LinkageError_klass())) { CLEAR_PENDING_EXCEPTION; } else { return false;
--- a/src/share/vm/oops/methodOop.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/methodOop.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -303,7 +303,7 @@ bool check_code() const; // Not inline to avoid circular ref nmethod* volatile code() const { assert( check_code(), "" ); return (nmethod *)OrderAccess::load_ptr_acquire(&_code); } void clear_code(); // Clear out any compiled code - void set_code(methodHandle mh, nmethod* code); + static void set_code(methodHandle mh, nmethod* code); void set_adapter_entry(AdapterHandlerEntry* adapter) { _adapter = adapter; } address get_i2c_entry(); address get_c2i_entry(); @@ -365,6 +365,7 @@ #endif // byte codes + void set_code(address code) { return constMethod()->set_code(code); } address code_base() const { return constMethod()->code_base(); } bool contains(address bcp) const { return constMethod()->contains(bcp); } @@ -524,6 +525,9 @@ // JSR 292 support bool is_method_handle_invoke() const { return access_flags().is_method_handle_invoke(); } + // Tests if this method is an internal adapter frame from the + // MethodHandleCompiler. + bool is_method_handle_adapter() const; static methodHandle make_invoke_method(KlassHandle holder, symbolHandle signature, Handle method_type, @@ -537,6 +541,7 @@ // all without checking for a stack overflow static int extra_stack_entries() { return (EnableMethodHandles ? (int)MethodHandlePushLimit : 0) + (EnableInvokeDynamic ? 3 : 0); } static int extra_stack_words(); // = extra_stack_entries() * Interpreter::stackElementSize() + // RedefineClasses() support: bool is_old() const { return access_flags().is_old(); } void set_is_old() { _access_flags.set_is_old(); } @@ -591,7 +596,10 @@ // whether it is not compilable for another reason like having a // breakpoint set in it. bool is_not_compilable(int comp_level = CompLevel_highest_tier) const; - void set_not_compilable(int comp_level = CompLevel_highest_tier); + void set_not_compilable(int comp_level = CompLevel_highest_tier, bool report = true); + void set_not_compilable_quietly(int comp_level = CompLevel_highest_tier) { + set_not_compilable(comp_level, false); + } bool is_not_osr_compilable() const { return is_not_compilable() || access_flags().is_not_osr_compilable(); } void set_not_osr_compilable() { _access_flags.set_not_osr_compilable(); }
--- a/src/share/vm/oops/objArrayKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/objArrayKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -127,16 +127,14 @@ // pointer delta is scaled to number of elements (length field in // objArrayOop) which we assume is 32 bit. assert(pd == (size_t)(int)pd, "length field overflow"); - const size_t done_word_len = objArrayOopDesc::array_size((int)pd); - bs->write_ref_array(MemRegion((HeapWord*)dst, done_word_len)); + bs->write_ref_array((HeapWord*)dst, pd); THROW(vmSymbols::java_lang_ArrayStoreException()); return; } } } } - const size_t word_len = objArrayOopDesc::array_size(length); - bs->write_ref_array(MemRegion((HeapWord*)dst, word_len)); + bs->write_ref_array((HeapWord*)dst, length); } void objArrayKlass::copy_array(arrayOop s, int src_pos, arrayOop d, @@ -248,8 +246,8 @@ } else { objArrayOop sec_oop = oopFactory::new_system_objArray(num_secondaries, CHECK_NULL); objArrayHandle secondaries(THREAD, sec_oop); - secondaries->obj_at_put(num_extra_slots+0, SystemDictionary::cloneable_klass()); - secondaries->obj_at_put(num_extra_slots+1, SystemDictionary::serializable_klass()); + secondaries->obj_at_put(num_extra_slots+0, SystemDictionary::Cloneable_klass()); + secondaries->obj_at_put(num_extra_slots+1, SystemDictionary::Serializable_klass()); for (int i = 0; i < num_elem_supers; i++) { klassOop elem_super = (klassOop) elem_supers->obj_at(i); klassOop array_super = elem_super->klass_part()->array_klass_or_null(); @@ -501,6 +499,8 @@ } } +#endif //PRODUCT + static int max_objArray_print_length = 4; void objArrayKlass::oop_print_value_on(oop obj, outputStream* st) { @@ -510,7 +510,7 @@ int len = objArrayOop(obj)->length(); st->print("[%d] ", len); obj->print_address_on(st); - if (PrintOopAddress || PrintMiscellaneous && (WizardMode || Verbose)) { + if (NOT_PRODUCT(PrintOopAddress ||) PrintMiscellaneous && (WizardMode || Verbose)) { st->print("{"); for (int i = 0; i < len; i++) { if (i > max_objArray_print_length) { @@ -522,8 +522,6 @@ } } -#endif // PRODUCT - const char* objArrayKlass::internal_name() const { return external_name(); }
--- a/src/share/vm/oops/objArrayKlass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/objArrayKlass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -119,14 +119,13 @@ private: static klassOop array_klass_impl (objArrayKlassHandle this_oop, bool or_null, int n, TRAPS); -#ifndef PRODUCT public: // Printing + void oop_print_value_on(oop obj, outputStream* st); +#ifndef PRODUCT void oop_print_on (oop obj, outputStream* st); - void oop_print_value_on(oop obj, outputStream* st); -#endif +#endif //PRODUCT - public: // Verification const char* internal_name() const; void oop_verify_on(oop obj, outputStream* st);
--- a/src/share/vm/oops/objArrayKlassKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/objArrayKlassKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -99,7 +99,7 @@ } } else { // The element type is already Object. Object[] has direct super of Object. - super_klass = KlassHandle(THREAD, SystemDictionary::object_klass()); + super_klass = KlassHandle(THREAD, SystemDictionary::Object_klass()); } } @@ -278,6 +278,7 @@ st->cr(); } +#endif //PRODUCT void objArrayKlassKlass::oop_print_value_on(oop obj, outputStream* st) { assert(obj->is_klass(), "must be klass"); @@ -287,8 +288,6 @@ st->print("[]"); } -#endif - const char* objArrayKlassKlass::internal_name() const { return "{object array class}"; }
--- a/src/share/vm/oops/objArrayKlassKlass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/objArrayKlassKlass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -64,14 +64,13 @@ // helpers static klassOop allocate_objArray_klass_impl(objArrayKlassKlassHandle this_oop, int n, KlassHandle element_klass, TRAPS); -#ifndef PRODUCT public: // Printing + void oop_print_value_on(oop obj, outputStream* st); +#ifndef PRODUCT void oop_print_on(oop obj, outputStream* st); - void oop_print_value_on(oop obj, outputStream* st); -#endif +#endif //PRODUCT - public: // Verification const char* internal_name() const; void oop_verify_on(oop obj, outputStream* st);
--- a/src/share/vm/oops/objArrayOop.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/objArrayOop.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -37,6 +37,32 @@ return &((T*)base())[index]; } +private: + // Give size of objArrayOop in HeapWords minus the header + static int array_size(int length) { + const int OopsPerHeapWord = HeapWordSize/heapOopSize; + assert(OopsPerHeapWord >= 1 && (HeapWordSize % heapOopSize == 0), + "Else the following (new) computation would be in error"); +#ifdef ASSERT + // The old code is left in for sanity-checking; it'll + // go away pretty soon. XXX + // Without UseCompressedOops, this is simply: + // oop->length() * HeapWordsPerOop; + // With narrowOops, HeapWordsPerOop is 1/2 or equal 0 as an integer. + // The oop elements are aligned up to wordSize + const int HeapWordsPerOop = heapOopSize/HeapWordSize; + int old_res; + if (HeapWordsPerOop > 0) { + old_res = length * HeapWordsPerOop; + } else { + old_res = align_size_up(length, OopsPerHeapWord)/OopsPerHeapWord; + } +#endif // ASSERT + int res = ((uint)length + OopsPerHeapWord - 1)/OopsPerHeapWord; + assert(res == old_res, "Inconsistency between old and new."); + return res; + } + public: // Returns the offset of the first element. static int base_offset_in_bytes() { @@ -67,27 +93,14 @@ // Sizing static int header_size() { return arrayOopDesc::header_size(T_OBJECT); } int object_size() { return object_size(length()); } - int array_size() { return array_size(length()); } static int object_size(int length) { // This returns the object size in HeapWords. - return align_object_size(header_size() + array_size(length)); - } - - // Give size of objArrayOop in HeapWords minus the header - static int array_size(int length) { - // Without UseCompressedOops, this is simply: - // oop->length() * HeapWordsPerOop; - // With narrowOops, HeapWordsPerOop is 1/2 or equal 0 as an integer. - // The oop elements are aligned up to wordSize - const int HeapWordsPerOop = heapOopSize/HeapWordSize; - if (HeapWordsPerOop > 0) { - return length * HeapWordsPerOop; - } else { - const int OopsPerHeapWord = HeapWordSize/heapOopSize; - int word_len = align_size_up(length, OopsPerHeapWord)/OopsPerHeapWord; - return word_len; - } + uint asz = array_size(length); + uint osz = align_object_size(header_size() + asz); + assert(osz >= asz, "no overflow"); + assert((int)osz > 0, "no overflow"); + return (int)osz; } // special iterators for index ranges, returns size of object
--- a/src/share/vm/oops/oop.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/oop.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -31,14 +31,13 @@ #ifdef PRODUCT void oopDesc::print_on(outputStream* st) const {} -void oopDesc::print_value_on(outputStream* st) const {} void oopDesc::print_address_on(outputStream* st) const {} -char* oopDesc::print_value_string() { return NULL; } char* oopDesc::print_string() { return NULL; } void oopDesc::print() {} -void oopDesc::print_value() {} void oopDesc::print_address() {} -#else + +#else //PRODUCT + void oopDesc::print_on(outputStream* st) const { if (this == NULL) { st->print_cr("NULL"); @@ -47,22 +46,6 @@ } } -void oopDesc::print_value_on(outputStream* st) const { - oop obj = oop(this); - if (this == NULL) { - st->print("NULL"); - } else if (java_lang_String::is_instance(obj)) { - java_lang_String::print(obj, st); - if (PrintOopAddress) print_address_on(st); -#ifdef ASSERT - } else if (!Universe::heap()->is_in(obj) || !Universe::heap()->is_in(klass())) { - st->print("### BAD OOP %p ###", (address)obj); -#endif - } else { - blueprint()->oop_print_value_on(obj, st); - } -} - void oopDesc::print_address_on(outputStream* st) const { if (PrintOopAddress) { st->print("{"INTPTR_FORMAT"}", this); @@ -71,23 +54,47 @@ void oopDesc::print() { print_on(tty); } -void oopDesc::print_value() { print_value_on(tty); } - void oopDesc::print_address() { print_address_on(tty); } char* oopDesc::print_string() { - stringStream* st = new stringStream(); - print_on(st); - return st->as_string(); + stringStream st; + print_on(&st); + return st.as_string(); +} + +#endif // PRODUCT + +// The print_value functions are present in all builds, to support the disassembler. + +void oopDesc::print_value() { + print_value_on(tty); } char* oopDesc::print_value_string() { - stringStream* st = new stringStream(); - print_value_on(st); - return st->as_string(); + char buf[100]; + stringStream st(buf, sizeof(buf)); + print_value_on(&st); + return st.as_string(); } -#endif // PRODUCT +void oopDesc::print_value_on(outputStream* st) const { + oop obj = oop(this); + if (this == NULL) { + st->print("NULL"); + } else if (java_lang_String::is_instance(obj)) { + java_lang_String::print(obj, st); +#ifndef PRODUCT + if (PrintOopAddress) print_address_on(st); +#endif //PRODUCT +#ifdef ASSERT + } else if (!Universe::heap()->is_in(obj) || !Universe::heap()->is_in(klass())) { + st->print("### BAD OOP %p ###", (address)obj); +#endif //ASSERT + } else { + blueprint()->oop_print_value_on(obj, st); + } +} + void oopDesc::verify_on(outputStream* st) { if (this != NULL) {
--- a/src/share/vm/oops/oop.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/oop.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -30,13 +30,12 @@ // no virtual functions allowed // store into oop with store check -template <class T> void oop_store(T* p, oop v); -template <class T> void oop_store(volatile T* p, oop v); +template <class T> inline void oop_store(T* p, oop v); +template <class T> inline void oop_store(volatile T* p, oop v); // store into oop without store check -template <class T> void oop_store_without_check(T* p, oop v); -template <class T> void oop_store_without_check(volatile T* p, oop v); - +template <class T> inline void oop_store_without_check(T* p, oop v); +template <class T> inline void oop_store_without_check(volatile T* p, oop v); extern bool always_do_update_barrier;
--- a/src/share/vm/oops/symbolKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/symbolKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -213,6 +213,8 @@ st->print("'"); } +#endif //PRODUCT + void symbolKlass::oop_print_value_on(oop obj, outputStream* st) { symbolOop sym = symbolOop(obj); st->print("'"); @@ -222,8 +224,6 @@ st->print("'"); } -#endif //PRODUCT - const char* symbolKlass::internal_name() const { return "{symbol}"; }
--- a/src/share/vm/oops/symbolKlass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/symbolKlass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -65,10 +65,10 @@ int oop_oop_iterate(oop obj, OopClosure* blk); int oop_oop_iterate_m(oop obj, OopClosure* blk, MemRegion mr); -#ifndef PRODUCT // Printing void oop_print_value_on(oop obj, outputStream* st); +#ifndef PRODUCT void oop_print_on(oop obj, outputStream* st); -#endif +#endif //PRODUCT const char* internal_name() const; };
--- a/src/share/vm/oops/symbolOop.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/symbolOop.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2005 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -25,6 +25,11 @@ # include "incls/_precompiled.incl" # include "incls/_symbolOop.cpp.incl" + +// ------------------------------------------------------------------ +// symbolOopDesc::equals +// +// Compares the symbol with a string of the given length. bool symbolOopDesc::equals(const char* str, int len) const { int l = utf8_length(); if (l != len) return false; @@ -36,6 +41,48 @@ return true; } + +// ------------------------------------------------------------------ +// symbolOopDesc::starts_with +// +// Tests if the symbol starts with the specified prefix of the given +// length. +bool symbolOopDesc::starts_with(const char* prefix, int len) const { + if (len > utf8_length()) return false; + while (len-- > 0) { + if (prefix[len] != (char) byte_at(len)) + return false; + } + assert(len == -1, "we should be at the beginning"); + return true; +} + + +// ------------------------------------------------------------------ +// symbolOopDesc::index_of +// +// Finds if the given string is a substring of this symbol's utf8 bytes. +// Return -1 on failure. Otherwise return the first index where str occurs. +int symbolOopDesc::index_of_at(int i, const char* str, int len) const { + assert(i >= 0 && i <= utf8_length(), "oob"); + if (len <= 0) return 0; + char first_char = str[0]; + address bytes = (address) ((symbolOopDesc*)this)->base(); + address limit = bytes + utf8_length() - len; // inclusive limit + address scan = bytes + i; + if (scan > limit) + return -1; + for (;;) { + scan = (address) memchr(scan, first_char, (limit + 1 - scan)); + if (scan == NULL) + return -1; // not found + assert(scan >= bytes+i && scan <= limit, "scan oob"); + if (memcmp(scan, str, len) == 0) + return (int)(scan - bytes); + } +} + + char* symbolOopDesc::as_C_string(char* buf, int size) const { if (size > 0) { int len = MIN2(size - 1, utf8_length());
--- a/src/share/vm/oops/symbolOop.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/symbolOop.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2005 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -70,8 +70,21 @@ void set_utf8_length(int len) { _length = len; } - // Compares the symbol with a string + // Compares the symbol with a string. bool equals(const char* str, int len) const; + bool equals(const char* str) const { return equals(str, (int) strlen(str)); } + + // Tests if the symbol starts with the given prefix. + bool starts_with(const char* prefix, int len) const; + bool starts_with(const char* prefix) const { + return starts_with(prefix, (int) strlen(prefix)); + } + + // Tests if the symbol starts with the given prefix. + int index_of_at(int i, const char* str, int len) const; + int index_of_at(int i, const char* str) const { + return index_of_at(i, str, (int) strlen(str)); + } // Three-way compare for sorting; returns -1/0/1 if receiver is </==/> than arg // note that the ordering is not alfabetical
--- a/src/share/vm/oops/typeArrayKlassKlass.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/typeArrayKlassKlass.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -45,6 +45,7 @@ Klass:: oop_print_on(obj, st); } +#endif //PRODUCT void typeArrayKlassKlass::oop_print_value_on(oop obj, outputStream* st) { assert(obj->is_klass(), "must be klass"); @@ -63,8 +64,6 @@ st->print("}"); } -#endif - const char* typeArrayKlassKlass::internal_name() const { return "{type array class}"; }
--- a/src/share/vm/oops/typeArrayKlassKlass.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/oops/typeArrayKlassKlass.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -47,12 +47,12 @@ static int header_size() { return oopDesc::header_size() + sizeof(typeArrayKlassKlass)/HeapWordSize; } int object_size() const { return align_object_size(header_size()); } -#ifndef PRODUCT public: // Printing + void oop_print_value_on(oop obj, outputStream* st); +#ifndef PRODUCT void oop_print_on(oop obj, outputStream* st); - void oop_print_value_on(oop obj, outputStream* st); -#endif - public: +#endif //PRODUCT + const char* internal_name() const; };
--- a/src/share/vm/opto/bytecodeInfo.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/bytecodeInfo.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -27,11 +27,16 @@ //============================================================================= //------------------------------InlineTree------------------------------------- -InlineTree::InlineTree( Compile* c, const InlineTree *caller_tree, ciMethod* callee, JVMState* caller_jvms, int caller_bci, float site_invoke_ratio ) +InlineTree::InlineTree( Compile* c, + const InlineTree *caller_tree, ciMethod* callee, + JVMState* caller_jvms, int caller_bci, + float site_invoke_ratio, int site_depth_adjust) : C(c), _caller_jvms(caller_jvms), _caller_tree((InlineTree*)caller_tree), _method(callee), _site_invoke_ratio(site_invoke_ratio), - _count_inline_bcs(method()->code_size()) { + _site_depth_adjust(site_depth_adjust), + _count_inline_bcs(method()->code_size()) +{ NOT_PRODUCT(_count_inlines = 0;) if (_caller_jvms != NULL) { // Keep a private copy of the caller_jvms: @@ -40,7 +45,7 @@ assert(!caller_jvms->should_reexecute(), "there should be no reexecute bytecode with inlining"); } assert(_caller_jvms->same_calls_as(caller_jvms), "consistent JVMS"); - assert((caller_tree == NULL ? 0 : caller_tree->inline_depth() + 1) == inline_depth(), "correct (redundant) depth parameter"); + assert((caller_tree == NULL ? 0 : caller_tree->stack_depth() + 1) == stack_depth(), "correct (redundant) depth parameter"); assert(caller_bci == this->caller_bci(), "correct (redundant) bci parameter"); if (UseOldInlining) { // Update hierarchical counts, count_inline_bcs() and count_inlines() @@ -52,10 +57,13 @@ } } -InlineTree::InlineTree(Compile* c, ciMethod* callee_method, JVMState* caller_jvms, float site_invoke_ratio) +InlineTree::InlineTree(Compile* c, ciMethod* callee_method, JVMState* caller_jvms, + float site_invoke_ratio, int site_depth_adjust) : C(c), _caller_jvms(caller_jvms), _caller_tree(NULL), _method(callee_method), _site_invoke_ratio(site_invoke_ratio), - _count_inline_bcs(method()->code_size()) { + _site_depth_adjust(site_depth_adjust), + _count_inline_bcs(method()->code_size()) +{ NOT_PRODUCT(_count_inlines = 0;) assert(!UseOldInlining, "do not use for old stuff"); } @@ -180,6 +188,10 @@ return NULL; } + // Always inline MethodHandle methods. + if (callee_method->is_method_handle_invoke()) + return NULL; + // First check all inlining restrictions which are required for correctness if (callee_method->is_abstract()) return "abstract method"; // note: we allow ik->is_abstract() @@ -265,10 +277,13 @@ return msg; } - bool is_accessor = InlineAccessors && callee_method->is_accessor(); + if (InlineAccessors && callee_method->is_accessor()) { + // accessor methods are not subject to any of the following limits. + return NULL; + } // suppress a few checks for accessors and trivial methods - if (!is_accessor && callee_method->code_size() > MaxTrivialSize) { + if (callee_method->code_size() > MaxTrivialSize) { // don't inline into giant methods if (C->unique() > (uint)NodeCountInliningCutoff) { @@ -287,7 +302,7 @@ } } - if (!C->do_inlining() && InlineAccessors && !is_accessor) { + if (!C->do_inlining() && InlineAccessors) { return "not an accessor"; } if( inline_depth() > MaxInlineLevel ) { @@ -322,14 +337,17 @@ // stricter than callee_holder->is_initialized() ciBytecodeStream iter(caller_method); iter.force_bci(caller_bci); - int index = iter.get_index_int(); - if( !caller_method->is_klass_loaded(index, true) ) { - return false; - } - // Try to do constant pool resolution if running Xcomp Bytecodes::Code call_bc = iter.cur_bc(); - if( !caller_method->check_call(index, call_bc == Bytecodes::_invokestatic) ) { - return false; + // An invokedynamic instruction does not have a klass. + if (call_bc != Bytecodes::_invokedynamic) { + int index = iter.get_index_int(); + if (!caller_method->is_klass_loaded(index, true)) { + return false; + } + // Try to do constant pool resolution if running Xcomp + if( !caller_method->check_call(index, call_bc == Bytecodes::_invokestatic) ) { + return false; + } } } // We will attempt to see if a class/field/etc got properly loaded. If it @@ -457,7 +475,30 @@ if (old_ilt != NULL) { return old_ilt; } - InlineTree *ilt = new InlineTree( C, this, callee_method, caller_jvms, caller_bci, recur_frequency ); + int new_depth_adjust = 0; + if (caller_jvms->method() != NULL) { + if ((caller_jvms->method()->name() == ciSymbol::invoke_name() && + caller_jvms->method()->holder()->name() == ciSymbol::java_dyn_MethodHandle()) + || caller_jvms->method()->holder()->name() == ciSymbol::java_dyn_InvokeDynamic()) + /* @@@ FIXME: + if (caller_jvms->method()->is_method_handle_adapter()) + */ + new_depth_adjust -= 1; // don't count actions in MH or indy adapter frames + else if (callee_method->is_method_handle_invoke()) { + new_depth_adjust -= 1; // don't count method handle calls from java.dyn implem + } + if (new_depth_adjust != 0 && PrintInlining) { + stringStream nm1; caller_jvms->method()->print_name(&nm1); + stringStream nm2; callee_method->print_name(&nm2); + tty->print_cr("discounting inlining depth from %s to %s", nm1.base(), nm2.base()); + } + if (new_depth_adjust != 0 && C->log()) { + int id1 = C->log()->identify(caller_jvms->method()); + int id2 = C->log()->identify(callee_method); + C->log()->elem("inline_depth_discount caller='%d' callee='%d'", id1, id2); + } + } + InlineTree *ilt = new InlineTree(C, this, callee_method, caller_jvms, caller_bci, recur_frequency, _site_depth_adjust + new_depth_adjust); _subtrees.append( ilt ); NOT_PRODUCT( _count_inlines += 1; ) @@ -483,7 +524,7 @@ Compile* C = Compile::current(); // Root of inline tree - InlineTree *ilt = new InlineTree(C, NULL, C->method(), NULL, -1, 1.0F); + InlineTree *ilt = new InlineTree(C, NULL, C->method(), NULL, -1, 1.0F, 0); return ilt; }
--- a/src/share/vm/opto/c2_globals.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/c2_globals.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -25,4 +25,4 @@ # include "incls/_precompiled.incl" # include "incls/_c2_globals.cpp.incl" -C2_FLAGS(MATERIALIZE_DEVELOPER_FLAG, MATERIALIZE_PD_DEVELOPER_FLAG, MATERIALIZE_PRODUCT_FLAG, MATERIALIZE_PD_PRODUCT_FLAG, MATERIALIZE_DIAGNOSTIC_FLAG, MATERIALIZE_NOTPRODUCT_FLAG) +C2_FLAGS(MATERIALIZE_DEVELOPER_FLAG, MATERIALIZE_PD_DEVELOPER_FLAG, MATERIALIZE_PRODUCT_FLAG, MATERIALIZE_PD_PRODUCT_FLAG, MATERIALIZE_DIAGNOSTIC_FLAG, MATERIALIZE_EXPERIMENTAL_FLAG, MATERIALIZE_NOTPRODUCT_FLAG)
--- a/src/share/vm/opto/c2_globals.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/c2_globals.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -26,7 +26,7 @@ // Defines all globals flags used by the server compiler. // -#define C2_FLAGS(develop, develop_pd, product, product_pd, diagnostic, notproduct) \ +#define C2_FLAGS(develop, develop_pd, product, product_pd, diagnostic, experimental, notproduct) \ \ notproduct(intx, CompileZapFirst, 0, \ "If +ZapDeadCompiledLocals, " \ @@ -154,6 +154,12 @@ notproduct(bool, TraceProfileTripCount, false, \ "Trace profile loop trip count information") \ \ + product(bool, UseLoopPredicate, true, \ + "Generate a predicate to select fast/slow loop versions") \ + \ + develop(bool, TraceLoopPredicate, false, \ + "Trace generation of loop predicates") \ + \ develop(bool, OptoCoalesce, true, \ "Use Conservative Copy Coalescing in the Register Allocator") \ \ @@ -394,6 +400,12 @@ product(bool, UseOptoBiasInlining, true, \ "Generate biased locking code in C2 ideal graph") \ \ + product(bool, OptimizeStringConcat, false, \ + "Optimize the construction of Strings by StringBuilder") \ + \ + notproduct(bool, PrintOptimizeStringConcat, false, \ + "Print information about transformations performed on Strings") \ + \ product(intx, ValueSearchLimit, 1000, \ "Recursion limit in PhaseMacroExpand::value_from_mem_phi") \ \ @@ -413,4 +425,4 @@ product(bool, BlockLayoutRotateLoops, true, \ "Allow back branches to be fall throughs in the block layour") \ -C2_FLAGS(DECLARE_DEVELOPER_FLAG, DECLARE_PD_DEVELOPER_FLAG, DECLARE_PRODUCT_FLAG, DECLARE_PD_PRODUCT_FLAG, DECLARE_DIAGNOSTIC_FLAG, DECLARE_NOTPRODUCT_FLAG) +C2_FLAGS(DECLARE_DEVELOPER_FLAG, DECLARE_PD_DEVELOPER_FLAG, DECLARE_PRODUCT_FLAG, DECLARE_PD_PRODUCT_FLAG, DECLARE_DIAGNOSTIC_FLAG, DECLARE_EXPERIMENTAL_FLAG, DECLARE_NOTPRODUCT_FLAG)
--- a/src/share/vm/opto/callGenerator.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/callGenerator.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2000-2008 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2000-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -98,12 +98,21 @@ //---------------------------DirectCallGenerator------------------------------ // Internal class which handles all out-of-line calls w/o receiver type checks. class DirectCallGenerator : public CallGenerator { -public: - DirectCallGenerator(ciMethod* method) - : CallGenerator(method) + private: + CallStaticJavaNode* _call_node; + // Force separate memory and I/O projections for the exceptional + // paths to facilitate late inlinig. + bool _separate_io_proj; + + public: + DirectCallGenerator(ciMethod* method, bool separate_io_proj) + : CallGenerator(method), + _separate_io_proj(separate_io_proj) { } virtual JVMState* generate(JVMState* jvms); + + CallStaticJavaNode* call_node() const { return _call_node; } }; JVMState* DirectCallGenerator::generate(JVMState* jvms) { @@ -127,14 +136,85 @@ } // Mark the call node as virtual, sort of: call->set_optimized_virtual(true); + if (method()->is_method_handle_invoke()) { + call->set_method_handle_invoke(true); + kit.C->set_has_method_handle_invokes(true); + } } kit.set_arguments_for_java_call(call); + kit.set_edges_for_java_call(call, false, _separate_io_proj); + Node* ret = kit.set_results_for_java_call(call, _separate_io_proj); + kit.push_node(method()->return_type()->basic_type(), ret); + _call_node = call; // Save the call node in case we need it later + return kit.transfer_exceptions_into_jvms(); +} + +//---------------------------DynamicCallGenerator----------------------------- +// Internal class which handles all out-of-line invokedynamic calls. +class DynamicCallGenerator : public CallGenerator { +public: + DynamicCallGenerator(ciMethod* method) + : CallGenerator(method) + { + } + virtual JVMState* generate(JVMState* jvms); +}; + +JVMState* DynamicCallGenerator::generate(JVMState* jvms) { + GraphKit kit(jvms); + + if (kit.C->log() != NULL) { + kit.C->log()->elem("dynamic_call bci='%d'", jvms->bci()); + } + + // Get the constant pool cache from the caller class. + ciMethod* caller_method = jvms->method(); + ciBytecodeStream str(caller_method); + str.force_bci(jvms->bci()); // Set the stream to the invokedynamic bci. + assert(str.cur_bc() == Bytecodes::_invokedynamic, "wrong place to issue a dynamic call!"); + ciCPCache* cpcache = str.get_cpcache(); + + // Get the offset of the CallSite from the constant pool cache + // pointer. + int index = str.get_method_index(); + size_t call_site_offset = cpcache->get_f1_offset(index); + + // Load the CallSite object from the constant pool cache. + const TypeOopPtr* cpcache_ptr = TypeOopPtr::make_from_constant(cpcache); + Node* cpcache_adr = kit.makecon(cpcache_ptr); + Node* call_site_adr = kit.basic_plus_adr(cpcache_adr, cpcache_adr, call_site_offset); + Node* call_site = kit.make_load(kit.control(), call_site_adr, TypeInstPtr::BOTTOM, T_OBJECT, Compile::AliasIdxRaw); + + // Load the target MethodHandle from the CallSite object. + Node* target_mh_adr = kit.basic_plus_adr(call_site, call_site, java_dyn_CallSite::target_offset_in_bytes()); + Node* target_mh = kit.make_load(kit.control(), target_mh_adr, TypeInstPtr::BOTTOM, T_OBJECT); + + address resolve_stub = SharedRuntime::get_resolve_opt_virtual_call_stub(); + + CallStaticJavaNode *call = new (kit.C, tf()->domain()->cnt()) CallStaticJavaNode(tf(), resolve_stub, method(), kit.bci()); + // invokedynamic is treated as an optimized invokevirtual. + call->set_optimized_virtual(true); + // Take extra care (in the presence of argument motion) not to trash the SP: + call->set_method_handle_invoke(true); + kit.C->set_has_method_handle_invokes(true); + + // Pass the target MethodHandle as first argument and shift the + // other arguments. + call->init_req(0 + TypeFunc::Parms, target_mh); + uint nargs = call->method()->arg_size(); + for (uint i = 1; i < nargs; i++) { + Node* arg = kit.argument(i - 1); + call->init_req(i + TypeFunc::Parms, arg); + } + kit.set_edges_for_java_call(call); Node* ret = kit.set_results_for_java_call(call); kit.push_node(method()->return_type()->basic_type(), ret); return kit.transfer_exceptions_into_jvms(); } +//--------------------------VirtualCallGenerator------------------------------ +// Internal class which handles all out-of-line calls checking receiver type. class VirtualCallGenerator : public CallGenerator { private: int _vtable_index; @@ -149,8 +229,6 @@ virtual JVMState* generate(JVMState* jvms); }; -//--------------------------VirtualCallGenerator------------------------------ -// Internal class which handles all out-of-line calls checking receiver type. JVMState* VirtualCallGenerator::generate(JVMState* jvms) { GraphKit kit(jvms); Node* receiver = kit.argument(0); @@ -238,16 +316,124 @@ return new ParseGenerator(m, expected_uses, true); } -CallGenerator* CallGenerator::for_direct_call(ciMethod* m) { +CallGenerator* CallGenerator::for_direct_call(ciMethod* m, bool separate_io_proj) { assert(!m->is_abstract(), "for_direct_call mismatch"); - return new DirectCallGenerator(m); + return new DirectCallGenerator(m, separate_io_proj); +} + +CallGenerator* CallGenerator::for_dynamic_call(ciMethod* m) { + assert(m->is_method_handle_invoke(), "for_dynamic_call mismatch"); + return new DynamicCallGenerator(m); } CallGenerator* CallGenerator::for_virtual_call(ciMethod* m, int vtable_index) { assert(!m->is_static(), "for_virtual_call mismatch"); + assert(!m->is_method_handle_invoke(), "should be a direct call"); return new VirtualCallGenerator(m, vtable_index); } +// Allow inlining decisions to be delayed +class LateInlineCallGenerator : public DirectCallGenerator { + CallGenerator* _inline_cg; + + public: + LateInlineCallGenerator(ciMethod* method, CallGenerator* inline_cg) : + DirectCallGenerator(method, true), _inline_cg(inline_cg) {} + + virtual bool is_late_inline() const { return true; } + + // Convert the CallStaticJava into an inline + virtual void do_late_inline(); + + JVMState* generate(JVMState* jvms) { + // Record that this call site should be revisited once the main + // parse is finished. + Compile::current()->add_late_inline(this); + + // Emit the CallStaticJava and request separate projections so + // that the late inlining logic can distinguish between fall + // through and exceptional uses of the memory and io projections + // as is done for allocations and macro expansion. + return DirectCallGenerator::generate(jvms); + } + +}; + + +void LateInlineCallGenerator::do_late_inline() { + // Can't inline it + if (call_node() == NULL || call_node()->outcnt() == 0 || + call_node()->in(0) == NULL || call_node()->in(0)->is_top()) + return; + + CallStaticJavaNode* call = call_node(); + + // Make a clone of the JVMState that appropriate to use for driving a parse + Compile* C = Compile::current(); + JVMState* jvms = call->jvms()->clone_shallow(C); + uint size = call->req(); + SafePointNode* map = new (C, size) SafePointNode(size, jvms); + for (uint i1 = 0; i1 < size; i1++) { + map->init_req(i1, call->in(i1)); + } + + // Make sure the state is a MergeMem for parsing. + if (!map->in(TypeFunc::Memory)->is_MergeMem()) { + map->set_req(TypeFunc::Memory, MergeMemNode::make(C, map->in(TypeFunc::Memory))); + } + + // Make enough space for the expression stack and transfer the incoming arguments + int nargs = method()->arg_size(); + jvms->set_map(map); + map->ensure_stack(jvms, jvms->method()->max_stack()); + if (nargs > 0) { + for (int i1 = 0; i1 < nargs; i1++) { + map->set_req(i1 + jvms->argoff(), call->in(TypeFunc::Parms + i1)); + } + } + + CompileLog* log = C->log(); + if (log != NULL) { + log->head("late_inline method='%d'", log->identify(method())); + JVMState* p = jvms; + while (p != NULL) { + log->elem("jvms bci='%d' method='%d'", p->bci(), log->identify(p->method())); + p = p->caller(); + } + log->tail("late_inline"); + } + + // Setup default node notes to be picked up by the inlining + Node_Notes* old_nn = C->default_node_notes(); + if (old_nn != NULL) { + Node_Notes* entry_nn = old_nn->clone(C); + entry_nn->set_jvms(jvms); + C->set_default_node_notes(entry_nn); + } + + // Now perform the inling using the synthesized JVMState + JVMState* new_jvms = _inline_cg->generate(jvms); + if (new_jvms == NULL) return; // no change + if (C->failing()) return; + + // Capture any exceptional control flow + GraphKit kit(new_jvms); + + // Find the result object + Node* result = C->top(); + int result_size = method()->return_type()->size(); + if (result_size != 0 && !kit.stopped()) { + result = (result_size == 1) ? kit.pop() : kit.pop_pair(); + } + + kit.replace_call(call, result); +} + + +CallGenerator* CallGenerator::for_late_inline(ciMethod* method, CallGenerator* inline_cg) { + return new LateInlineCallGenerator(method, inline_cg); +} + //---------------------------WarmCallGenerator-------------------------------- // Internal class which handles initial deferral of inlining decisions. @@ -315,70 +501,7 @@ } void WarmCallInfo::make_hot() { - Compile* C = Compile::current(); - // Replace the callnode with something better. - CallJavaNode* call = this->call()->as_CallJava(); - ciMethod* method = call->method(); - int nargs = method->arg_size(); - JVMState* jvms = call->jvms()->clone_shallow(C); - uint size = TypeFunc::Parms + MAX2(2, nargs); - SafePointNode* map = new (C, size) SafePointNode(size, jvms); - for (uint i1 = 0; i1 < (uint)(TypeFunc::Parms + nargs); i1++) { - map->init_req(i1, call->in(i1)); - } - jvms->set_map(map); - jvms->set_offsets(map->req()); - jvms->set_locoff(TypeFunc::Parms); - jvms->set_stkoff(TypeFunc::Parms); - GraphKit kit(jvms); - - JVMState* new_jvms = _hot_cg->generate(kit.jvms()); - if (new_jvms == NULL) return; // no change - if (C->failing()) return; - - kit.set_jvms(new_jvms); - Node* res = C->top(); - int res_size = method->return_type()->size(); - if (res_size != 0) { - kit.inc_sp(-res_size); - res = kit.argument(0); - } - GraphKit ekit(kit.combine_and_pop_all_exception_states()->jvms()); - - // Replace the call: - for (DUIterator i = call->outs(); call->has_out(i); i++) { - Node* n = call->out(i); - Node* nn = NULL; // replacement - if (n->is_Proj()) { - ProjNode* nproj = n->as_Proj(); - assert(nproj->_con < (uint)(TypeFunc::Parms + (res_size ? 1 : 0)), "sane proj"); - if (nproj->_con == TypeFunc::Parms) { - nn = res; - } else { - nn = kit.map()->in(nproj->_con); - } - if (nproj->_con == TypeFunc::I_O) { - for (DUIterator j = nproj->outs(); nproj->has_out(j); j++) { - Node* e = nproj->out(j); - if (e->Opcode() == Op_CreateEx) { - e->replace_by(ekit.argument(0)); - } else if (e->Opcode() == Op_Catch) { - for (DUIterator k = e->outs(); e->has_out(k); k++) { - CatchProjNode* p = e->out(j)->as_CatchProj(); - if (p->is_handler_proj()) { - p->replace_by(ekit.control()); - } else { - p->replace_by(kit.control()); - } - } - } - } - } - } - NOT_PRODUCT(if (!nn) n->dump(2)); - assert(nn != NULL, "don't know what to do with this user"); - n->replace_by(nn); - } + Unimplemented(); } void WarmCallInfo::make_cold() { @@ -527,6 +650,155 @@ } +//------------------------PredictedDynamicCallGenerator----------------------- +// Internal class which handles all out-of-line calls checking receiver type. +class PredictedDynamicCallGenerator : public CallGenerator { + ciMethodHandle* _predicted_method_handle; + CallGenerator* _if_missed; + CallGenerator* _if_hit; + float _hit_prob; + +public: + PredictedDynamicCallGenerator(ciMethodHandle* predicted_method_handle, + CallGenerator* if_missed, + CallGenerator* if_hit, + float hit_prob) + : CallGenerator(if_missed->method()), + _predicted_method_handle(predicted_method_handle), + _if_missed(if_missed), + _if_hit(if_hit), + _hit_prob(hit_prob) + {} + + virtual bool is_inline() const { return _if_hit->is_inline(); } + virtual bool is_deferred() const { return _if_hit->is_deferred(); } + + virtual JVMState* generate(JVMState* jvms); +}; + + +CallGenerator* CallGenerator::for_predicted_dynamic_call(ciMethodHandle* predicted_method_handle, + CallGenerator* if_missed, + CallGenerator* if_hit, + float hit_prob) { + return new PredictedDynamicCallGenerator(predicted_method_handle, if_missed, if_hit, hit_prob); +} + + +JVMState* PredictedDynamicCallGenerator::generate(JVMState* jvms) { + GraphKit kit(jvms); + PhaseGVN& gvn = kit.gvn(); + + CompileLog* log = kit.C->log(); + if (log != NULL) { + log->elem("predicted_dynamic_call bci='%d'", jvms->bci()); + } + + // Get the constant pool cache from the caller class. + ciMethod* caller_method = jvms->method(); + ciBytecodeStream str(caller_method); + str.force_bci(jvms->bci()); // Set the stream to the invokedynamic bci. + ciCPCache* cpcache = str.get_cpcache(); + + // Get the offset of the CallSite from the constant pool cache + // pointer. + int index = str.get_method_index(); + size_t call_site_offset = cpcache->get_f1_offset(index); + + // Load the CallSite object from the constant pool cache. + const TypeOopPtr* cpcache_ptr = TypeOopPtr::make_from_constant(cpcache); + Node* cpcache_adr = kit.makecon(cpcache_ptr); + Node* call_site_adr = kit.basic_plus_adr(cpcache_adr, cpcache_adr, call_site_offset); + Node* call_site = kit.make_load(kit.control(), call_site_adr, TypeInstPtr::BOTTOM, T_OBJECT, Compile::AliasIdxRaw); + + // Load the target MethodHandle from the CallSite object. + Node* target_adr = kit.basic_plus_adr(call_site, call_site, java_dyn_CallSite::target_offset_in_bytes()); + Node* target_mh = kit.make_load(kit.control(), target_adr, TypeInstPtr::BOTTOM, T_OBJECT); + + // Check if the MethodHandle is still the same. + const TypeOopPtr* predicted_mh_ptr = TypeOopPtr::make_from_constant(_predicted_method_handle, true); + Node* predicted_mh = kit.makecon(predicted_mh_ptr); + + Node* cmp = gvn.transform(new(kit.C, 3) CmpPNode(target_mh, predicted_mh)); + Node* bol = gvn.transform(new(kit.C, 2) BoolNode(cmp, BoolTest::eq) ); + IfNode* iff = kit.create_and_xform_if(kit.control(), bol, _hit_prob, COUNT_UNKNOWN); + kit.set_control( gvn.transform(new(kit.C, 1) IfTrueNode (iff))); + Node* slow_ctl = gvn.transform(new(kit.C, 1) IfFalseNode(iff)); + + SafePointNode* slow_map = NULL; + JVMState* slow_jvms; + { PreserveJVMState pjvms(&kit); + kit.set_control(slow_ctl); + if (!kit.stopped()) { + slow_jvms = _if_missed->generate(kit.sync_jvms()); + assert(slow_jvms != NULL, "miss path must not fail to generate"); + kit.add_exception_states_from(slow_jvms); + kit.set_map(slow_jvms->map()); + if (!kit.stopped()) + slow_map = kit.stop(); + } + } + + if (kit.stopped()) { + // Instance exactly does not matches the desired type. + kit.set_jvms(slow_jvms); + return kit.transfer_exceptions_into_jvms(); + } + + // Make the hot call: + JVMState* new_jvms = _if_hit->generate(kit.sync_jvms()); + if (new_jvms == NULL) { + // Inline failed, so make a direct call. + assert(_if_hit->is_inline(), "must have been a failed inline"); + CallGenerator* cg = CallGenerator::for_direct_call(_if_hit->method()); + new_jvms = cg->generate(kit.sync_jvms()); + } + kit.add_exception_states_from(new_jvms); + kit.set_jvms(new_jvms); + + // Need to merge slow and fast? + if (slow_map == NULL) { + // The fast path is the only path remaining. + return kit.transfer_exceptions_into_jvms(); + } + + if (kit.stopped()) { + // Inlined method threw an exception, so it's just the slow path after all. + kit.set_jvms(slow_jvms); + return kit.transfer_exceptions_into_jvms(); + } + + // Finish the diamond. + kit.C->set_has_split_ifs(true); // Has chance for split-if optimization + RegionNode* region = new (kit.C, 3) RegionNode(3); + region->init_req(1, kit.control()); + region->init_req(2, slow_map->control()); + kit.set_control(gvn.transform(region)); + Node* iophi = PhiNode::make(region, kit.i_o(), Type::ABIO); + iophi->set_req(2, slow_map->i_o()); + kit.set_i_o(gvn.transform(iophi)); + kit.merge_memory(slow_map->merged_memory(), region, 2); + uint tos = kit.jvms()->stkoff() + kit.sp(); + uint limit = slow_map->req(); + for (uint i = TypeFunc::Parms; i < limit; i++) { + // Skip unused stack slots; fast forward to monoff(); + if (i == tos) { + i = kit.jvms()->monoff(); + if( i >= limit ) break; + } + Node* m = kit.map()->in(i); + Node* n = slow_map->in(i); + if (m != n) { + const Type* t = gvn.type(m)->meet(gvn.type(n)); + Node* phi = PhiNode::make(region, m, t); + phi->set_req(2, n); + kit.map()->set_req(i, gvn.transform(phi)); + } + } + return kit.transfer_exceptions_into_jvms(); +} + + //-------------------------UncommonTrapCallGenerator----------------------------- // Internal class which handles all out-of-line calls checking receiver type. class UncommonTrapCallGenerator : public CallGenerator {
--- a/src/share/vm/opto/callGenerator.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/callGenerator.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -57,6 +57,13 @@ // is_trap: Does not return to the caller. (E.g., uncommon trap.) virtual bool is_trap() const { return false; } + // is_late_inline: supports conversion of call into an inline + virtual bool is_late_inline() const { return false; } + // Replace the call with an inline version of the code + virtual void do_late_inline() { ShouldNotReachHere(); } + + virtual CallStaticJavaNode* call_node() const { ShouldNotReachHere(); return NULL; } + // Note: It is possible for a CG to be both inline and virtual. // (The hashCode intrinsic does a vtable check and an inlined fast path.) @@ -92,9 +99,13 @@ static CallGenerator* for_osr(ciMethod* m, int osr_bci); // How to generate vanilla out-of-line call sites: - static CallGenerator* for_direct_call(ciMethod* m); // static, special + static CallGenerator* for_direct_call(ciMethod* m, bool separate_io_projs = false); // static, special + static CallGenerator* for_dynamic_call(ciMethod* m); // invokedynamic static CallGenerator* for_virtual_call(ciMethod* m, int vtable_index); // virtual, interface + // How to generate a replace a direct call with an inline version + static CallGenerator* for_late_inline(ciMethod* m, CallGenerator* inline_cg); + // How to make a call but defer the decision whether to inline or not. static CallGenerator* for_warm_call(WarmCallInfo* ci, CallGenerator* if_cold, @@ -106,6 +117,12 @@ CallGenerator* if_hit, float hit_prob); + // How to make a call that optimistically assumes a MethodHandle target: + static CallGenerator* for_predicted_dynamic_call(ciMethodHandle* predicted_method_handle, + CallGenerator* if_missed, + CallGenerator* if_hit, + float hit_prob); + // How to make a call that gives up and goes back to the interpreter: static CallGenerator* for_uncommon_trap(ciMethod* m, Deoptimization::DeoptReason reason,
--- a/src/share/vm/opto/callnode.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/callnode.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -693,6 +693,84 @@ } +void CallNode::extract_projections(CallProjections* projs, bool separate_io_proj) { + projs->fallthrough_proj = NULL; + projs->fallthrough_catchproj = NULL; + projs->fallthrough_ioproj = NULL; + projs->catchall_ioproj = NULL; + projs->catchall_catchproj = NULL; + projs->fallthrough_memproj = NULL; + projs->catchall_memproj = NULL; + projs->resproj = NULL; + projs->exobj = NULL; + + for (DUIterator_Fast imax, i = fast_outs(imax); i < imax; i++) { + ProjNode *pn = fast_out(i)->as_Proj(); + if (pn->outcnt() == 0) continue; + switch (pn->_con) { + case TypeFunc::Control: + { + // For Control (fallthrough) and I_O (catch_all_index) we have CatchProj -> Catch -> Proj + projs->fallthrough_proj = pn; + DUIterator_Fast jmax, j = pn->fast_outs(jmax); + const Node *cn = pn->fast_out(j); + if (cn->is_Catch()) { + ProjNode *cpn = NULL; + for (DUIterator_Fast kmax, k = cn->fast_outs(kmax); k < kmax; k++) { + cpn = cn->fast_out(k)->as_Proj(); + assert(cpn->is_CatchProj(), "must be a CatchProjNode"); + if (cpn->_con == CatchProjNode::fall_through_index) + projs->fallthrough_catchproj = cpn; + else { + assert(cpn->_con == CatchProjNode::catch_all_index, "must be correct index."); + projs->catchall_catchproj = cpn; + } + } + } + break; + } + case TypeFunc::I_O: + if (pn->_is_io_use) + projs->catchall_ioproj = pn; + else + projs->fallthrough_ioproj = pn; + for (DUIterator j = pn->outs(); pn->has_out(j); j++) { + Node* e = pn->out(j); + if (e->Opcode() == Op_CreateEx && e->in(0)->is_CatchProj()) { + assert(projs->exobj == NULL, "only one"); + projs->exobj = e; + } + } + break; + case TypeFunc::Memory: + if (pn->_is_io_use) + projs->catchall_memproj = pn; + else + projs->fallthrough_memproj = pn; + break; + case TypeFunc::Parms: + projs->resproj = pn; + break; + default: + assert(false, "unexpected projection from allocation node."); + } + } + + // The resproj may not exist because the result couuld be ignored + // and the exception object may not exist if an exception handler + // swallows the exception but all the other must exist and be found. + assert(projs->fallthrough_proj != NULL, "must be found"); + assert(projs->fallthrough_catchproj != NULL, "must be found"); + assert(projs->fallthrough_memproj != NULL, "must be found"); + assert(projs->fallthrough_ioproj != NULL, "must be found"); + assert(projs->catchall_catchproj != NULL, "must be found"); + if (separate_io_proj) { + assert(projs->catchall_memproj != NULL, "must be found"); + assert(projs->catchall_ioproj != NULL, "must be found"); + } +} + + //============================================================================= uint CallJavaNode::size_of() const { return sizeof(*this); } uint CallJavaNode::cmp( const Node &n ) const {
--- a/src/share/vm/opto/callnode.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/callnode.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -470,6 +470,23 @@ #endif }; + +// Simple container for the outgoing projections of a call. Useful +// for serious surgery on calls. +class CallProjections : public StackObj { +public: + Node* fallthrough_proj; + Node* fallthrough_catchproj; + Node* fallthrough_memproj; + Node* fallthrough_ioproj; + Node* catchall_catchproj; + Node* catchall_memproj; + Node* catchall_ioproj; + Node* resproj; + Node* exobj; +}; + + //------------------------------CallNode--------------------------------------- // Call nodes now subsume the function of debug nodes at callsites, so they // contain the functionality of a full scope chain of debug nodes. @@ -521,6 +538,11 @@ // or returns NULL if there is no one. Node *result_cast(); + // Collect all the interesting edges from a call for use in + // replacing the call by something else. Used by macro expansion + // and the late inlining support. + void extract_projections(CallProjections* projs, bool separate_io_proj); + virtual uint match_edge(uint idx) const; #ifndef PRODUCT @@ -529,6 +551,7 @@ #endif }; + //------------------------------CallJavaNode----------------------------------- // Make a static or dynamic subroutine call node using Java calling // convention. (The "Java" calling convention is the compiler's calling @@ -539,12 +562,15 @@ virtual uint size_of() const; // Size is bigger bool _optimized_virtual; + bool _method_handle_invoke; ciMethod* _method; // Method being direct called public: const int _bci; // Byte Code Index of call byte code CallJavaNode(const TypeFunc* tf , address addr, ciMethod* method, int bci) : CallNode(tf, addr, TypePtr::BOTTOM), - _method(method), _bci(bci), _optimized_virtual(false) + _method(method), _bci(bci), + _optimized_virtual(false), + _method_handle_invoke(false) { init_class_id(Class_CallJava); } @@ -554,6 +580,8 @@ void set_method(ciMethod *m) { _method = m; } void set_optimized_virtual(bool f) { _optimized_virtual = f; } bool is_optimized_virtual() const { return _optimized_virtual; } + void set_method_handle_invoke(bool f) { _method_handle_invoke = f; } + bool is_method_handle_invoke() const { return _method_handle_invoke; } #ifndef PRODUCT virtual void dump_spec(outputStream *st) const;
--- a/src/share/vm/opto/compile.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/compile.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -224,6 +224,32 @@ } +void Compile::gvn_replace_by(Node* n, Node* nn) { + for (DUIterator_Last imin, i = n->last_outs(imin); i >= imin; ) { + Node* use = n->last_out(i); + bool is_in_table = initial_gvn()->hash_delete(use); + uint uses_found = 0; + for (uint j = 0; j < use->len(); j++) { + if (use->in(j) == n) { + if (j < use->req()) + use->set_req(j, nn); + else + use->set_prec(j, nn); + uses_found++; + } + } + if (is_in_table) { + // reinsert into table + initial_gvn()->hash_find_insert(use); + } + record_for_igvn(use); + i -= uses_found; // we deleted 1 or more copies of this edge + } +} + + + + // Identify all nodes that are reachable from below, useful. // Use breadth-first pass that records state in a Unique_Node_List, // recursive traversal is slower. @@ -439,6 +465,7 @@ _code_buffer("Compile::Fill_buffer"), _orig_pc_slot(0), _orig_pc_slot_offset_in_bytes(0), + _has_method_handle_invokes(false), _node_bundling_limit(0), _node_bundling_base(NULL), _java_calls(0), @@ -554,6 +581,28 @@ rethrow_exceptions(kit.transfer_exceptions_into_jvms()); } + if (!failing() && has_stringbuilder()) { + { + // remove useless nodes to make the usage analysis simpler + ResourceMark rm; + PhaseRemoveUseless pru(initial_gvn(), &for_igvn); + } + + { + ResourceMark rm; + print_method("Before StringOpts", 3); + PhaseStringOpts pso(initial_gvn(), &for_igvn); + print_method("After StringOpts", 3); + } + + // now inline anything that we skipped the first time around + while (_late_inlines.length() > 0) { + CallGenerator* cg = _late_inlines.pop(); + cg->do_late_inline(); + } + } + assert(_late_inlines.length() == 0, "should have been processed"); + print_method("Before RemoveUseless", 3); // Remove clutter produced by parsing. @@ -711,6 +760,7 @@ _do_escape_analysis(false), _failure_reason(NULL), _code_buffer("Compile::Fill_buffer"), + _has_method_handle_invokes(false), _node_bundling_limit(0), _node_bundling_base(NULL), _java_calls(0), @@ -820,6 +870,7 @@ _fixed_slots = 0; set_has_split_ifs(false); set_has_loops(has_method() && method()->has_loops()); // first approximation + set_has_stringbuilder(false); _deopt_happens = true; // start out assuming the worst _trap_can_recompile = false; // no traps emitted yet _major_progress = true; // start out assuming good things will happen @@ -883,6 +934,7 @@ _intrinsics = NULL; _macro_nodes = new GrowableArray<Node*>(comp_arena(), 8, 0, NULL); + _predicate_opaqs = new GrowableArray<Node*>(comp_arena(), 8, 0, NULL); register_library_intrinsics(); } @@ -1504,6 +1556,19 @@ } } +//---------------------cleanup_loop_predicates----------------------- +// Remove the opaque nodes that protect the predicates so that all unused +// checks and uncommon_traps will be eliminated from the ideal graph +void Compile::cleanup_loop_predicates(PhaseIterGVN &igvn) { + if (predicate_count()==0) return; + for (int i = predicate_count(); i > 0; i--) { + Node * n = predicate_opaque1_node(i-1); + assert(n->Opcode() == Op_Opaque1, "must be"); + igvn.replace_node(n, n->in(1)); + } + assert(predicate_count()==0, "should be clean!"); + igvn.optimize(); +} //------------------------------Optimize--------------------------------------- // Given a graph, optimize it. @@ -1545,7 +1610,7 @@ if((loop_opts_cnt > 0) && (has_loops() || has_split_ifs())) { { TracePhase t2("idealLoop", &_t_idealLoop, true); - PhaseIdealLoop ideal_loop( igvn, true ); + PhaseIdealLoop ideal_loop( igvn, true, UseLoopPredicate); loop_opts_cnt--; if (major_progress()) print_method("PhaseIdealLoop 1", 2); if (failing()) return; @@ -1553,7 +1618,7 @@ // Loop opts pass if partial peeling occurred in previous pass if(PartialPeelLoop && major_progress() && (loop_opts_cnt > 0)) { TracePhase t3("idealLoop", &_t_idealLoop, true); - PhaseIdealLoop ideal_loop( igvn, false ); + PhaseIdealLoop ideal_loop( igvn, false, UseLoopPredicate); loop_opts_cnt--; if (major_progress()) print_method("PhaseIdealLoop 2", 2); if (failing()) return; @@ -1561,7 +1626,7 @@ // Loop opts pass for loop-unrolling before CCP if(major_progress() && (loop_opts_cnt > 0)) { TracePhase t4("idealLoop", &_t_idealLoop, true); - PhaseIdealLoop ideal_loop( igvn, false ); + PhaseIdealLoop ideal_loop( igvn, false, UseLoopPredicate); loop_opts_cnt--; if (major_progress()) print_method("PhaseIdealLoop 3", 2); } @@ -1599,13 +1664,21 @@ // peeling, unrolling, etc. if(loop_opts_cnt > 0) { debug_only( int cnt = 0; ); + bool loop_predication = UseLoopPredicate; while(major_progress() && (loop_opts_cnt > 0)) { TracePhase t2("idealLoop", &_t_idealLoop, true); assert( cnt++ < 40, "infinite cycle in loop optimization" ); - PhaseIdealLoop ideal_loop( igvn, true ); + PhaseIdealLoop ideal_loop( igvn, true, loop_predication); loop_opts_cnt--; if (major_progress()) print_method("PhaseIdealLoop iterations", 2); if (failing()) return; + // Perform loop predication optimization during first iteration after CCP. + // After that switch it off and cleanup unused loop predicates. + if (loop_predication) { + loop_predication = false; + cleanup_loop_predicates(igvn); + if (failing()) return; + } } } @@ -1803,6 +1876,7 @@ !n->is_Phi() && // a few noisely useless nodes !n->is_Proj() && !n->is_MachTemp() && + !n->is_SafePointScalarObject() && !n->is_Catch() && // Would be nice to print exception table targets !n->is_MergeMem() && // Not very interesting !n->is_top() && // Debug info table constants @@ -2240,6 +2314,30 @@ break; } + case Op_Proj: { + if (OptimizeStringConcat) { + ProjNode* p = n->as_Proj(); + if (p->_is_io_use) { + // Separate projections were used for the exception path which + // are normally removed by a late inline. If it wasn't inlined + // then they will hang around and should just be replaced with + // the original one. + Node* proj = NULL; + // Replace with just one + for (SimpleDUIterator i(p->in(0)); i.has_next(); i.next()) { + Node *use = i.get(); + if (use->is_Proj() && p != use && use->as_Proj()->_con == p->_con) { + proj = use; + break; + } + } + assert(p != NULL, "must be found"); + p->subsume_by(proj); + } + } + break; + } + case Op_Phi: if (n->as_Phi()->bottom_type()->isa_narrowoop()) { // The EncodeP optimization may create Phi with the same edges
--- a/src/share/vm/opto/compile.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/compile.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -38,6 +38,7 @@ class OptoReg; class PhaseCFG; class PhaseGVN; +class PhaseIterGVN; class PhaseRegAlloc; class PhaseCCP; class PhaseCCP_DCE; @@ -149,6 +150,7 @@ bool _has_loops; // True if the method _may_ have some loops bool _has_split_ifs; // True if the method _may_ have some split-if bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores. + bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated uint _trap_hist[trapHistLength]; // Cumulative traps bool _trap_can_recompile; // Have we emitted a recompiling trap? uint _decompile_count; // Cumulative decompilation counts. @@ -164,6 +166,9 @@ bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing #endif + // JSR 292 + bool _has_method_handle_invokes; // True if this method has MethodHandle invokes. + // Compilation environment. Arena _comp_arena; // Arena with lifetime equivalent to Compile ciEnv* _env; // CI interface @@ -171,6 +176,7 @@ const char* _failure_reason; // for record_failure/failing pattern GrowableArray<CallGenerator*>* _intrinsics; // List of intrinsics. GrowableArray<Node*>* _macro_nodes; // List of nodes which need to be expanded before matching. + GrowableArray<Node*>* _predicate_opaqs; // List of Opaque1 nodes for the loop predicates. ConnectionGraph* _congraph; #ifndef PRODUCT IdealGraphPrinter* _printer; @@ -219,6 +225,9 @@ Unique_Node_List* _for_igvn; // Initial work-list for next round of Iterative GVN WarmCallInfo* _warm_calls; // Sorted work-list for heat-based inlining. + GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after + // main parsing has finished. + // Matching, CFG layout, allocation, code generation PhaseCFG* _cfg; // Results of CFG finding bool _select_24_bit_instr; // We selected an instruction with a 24-bit result @@ -298,6 +307,8 @@ void set_has_split_ifs(bool z) { _has_split_ifs = z; } bool has_unsafe_access() const { return _has_unsafe_access; } void set_has_unsafe_access(bool z) { _has_unsafe_access = z; } + bool has_stringbuilder() const { return _has_stringbuilder; } + void set_has_stringbuilder(bool z) { _has_stringbuilder = z; } void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; } uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; } bool trap_can_recompile() const { return _trap_can_recompile; } @@ -328,6 +339,10 @@ void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; } #endif + // JSR 292 + bool has_method_handle_invokes() const { return _has_method_handle_invokes; } + void set_has_method_handle_invokes(bool z) { _has_method_handle_invokes = z; } + void begin_method() { #ifndef PRODUCT if (_printer) _printer->begin_method(this); @@ -345,7 +360,9 @@ } int macro_count() { return _macro_nodes->length(); } + int predicate_count() { return _predicate_opaqs->length();} Node* macro_node(int idx) { return _macro_nodes->at(idx); } + Node* predicate_opaque1_node(int idx) { return _predicate_opaqs->at(idx);} ConnectionGraph* congraph() { return _congraph;} void add_macro_node(Node * n) { //assert(n->is_macro(), "must be a macro node"); @@ -357,7 +374,19 @@ // that the node is in the array before attempting to remove it if (_macro_nodes->contains(n)) _macro_nodes->remove(n); + // remove from _predicate_opaqs list also if it is there + if (predicate_count() > 0 && _predicate_opaqs->contains(n)){ + _predicate_opaqs->remove(n); + } } + void add_predicate_opaq(Node * n) { + assert(!_predicate_opaqs->contains(n), " duplicate entry in predicate opaque1"); + assert(_macro_nodes->contains(n), "should have already been in macro list"); + _predicate_opaqs->append(n); + } + // remove the opaque nodes that protect the predicates so that the unused checks and + // uncommon traps will be eliminated from the graph. + void cleanup_loop_predicates(PhaseIterGVN &igvn); // Compilation environment. Arena* comp_arena() { return &_comp_arena; } @@ -475,6 +504,7 @@ // Decide how to build a call. // The profile factor is a discount to apply to this site's interp. profile. CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_is_virtual, JVMState* jvms, bool allow_inline, float profile_factor); + bool should_delay_inlining(ciMethod* call_method, JVMState* jvms); // Report if there were too many traps at a current method and bci. // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded. @@ -495,6 +525,11 @@ void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; } void set_for_igvn(Unique_Node_List *for_igvn) { _for_igvn = for_igvn; } + // Replace n by nn using initial_gvn, calling hash_delete and + // record_for_igvn as needed. + void gvn_replace_by(Node* n, Node* nn); + + void identify_useful_nodes(Unique_Node_List &useful); void remove_useless_nodes (Unique_Node_List &useful); @@ -502,6 +537,9 @@ void set_warm_calls(WarmCallInfo* l) { _warm_calls = l; } WarmCallInfo* pop_warm_call(); + // Record this CallGenerator for inlining at the end of parsing. + void add_late_inline(CallGenerator* cg) { _late_inlines.push(cg); } + // Matching, CFG layout, allocation, code generation PhaseCFG* cfg() { return _cfg; } bool select_24_bit_instr() const { return _select_24_bit_instr; }
--- a/src/share/vm/opto/divnode.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/divnode.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -114,7 +114,8 @@ if( andconi_t && andconi_t->is_con() ) { jint andconi = andconi_t->get_con(); if( andconi < 0 && is_power_of_2(-andconi) && (-andconi) >= d ) { - dividend = dividend->in(1); + if( (-andconi) == d ) // Remove AND if it clears bits which will be shifted + dividend = dividend->in(1); needs_rounding = false; } } @@ -356,7 +357,8 @@ if( andconl_t && andconl_t->is_con() ) { jlong andconl = andconl_t->get_con(); if( andconl < 0 && is_power_of_2_long(-andconl) && (-andconl) >= d ) { - dividend = dividend->in(1); + if( (-andconl) == d ) // Remove AND if it clears bits which will be shifted + dividend = dividend->in(1); needs_rounding = false; } }
--- a/src/share/vm/opto/doCall.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/doCall.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1998-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1998-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -43,7 +43,9 @@ } #endif -CallGenerator* Compile::call_generator(ciMethod* call_method, int vtable_index, bool call_is_virtual, JVMState* jvms, bool allow_inline, float prof_factor) { +CallGenerator* Compile::call_generator(ciMethod* call_method, int vtable_index, bool call_is_virtual, + JVMState* jvms, bool allow_inline, + float prof_factor) { CallGenerator* cg; // Dtrace currently doesn't work unless all calls are vanilla @@ -68,7 +70,7 @@ CompileLog* log = this->log(); if (log != NULL) { int rid = (receiver_count >= 0)? log->identify(profile.receiver(0)): -1; - int r2id = (profile.morphism() == 2)? log->identify(profile.receiver(1)):-1; + int r2id = (rid != -1 && profile.has_receiver(1))? log->identify(profile.receiver(1)):-1; log->begin_elem("call method='%d' count='%d' prof_factor='%g'", log->identify(call_method), site_count, prof_factor); if (call_is_virtual) log->print(" virtual='1'"); @@ -116,7 +118,7 @@ // TO DO: When UseOldInlining is removed, copy the ILT code elsewhere. float site_invoke_ratio = prof_factor; // Note: ilt is for the root of this parse, not the present call site. - ilt = new InlineTree(this, jvms->method(), jvms->caller(), site_invoke_ratio); + ilt = new InlineTree(this, jvms->method(), jvms->caller(), site_invoke_ratio, 0); } WarmCallInfo scratch_ci; if (!UseOldInlining) @@ -128,6 +130,12 @@ if (allow_inline) { CallGenerator* cg = CallGenerator::for_inline(call_method, expected_uses); + if (require_inline && cg != NULL && should_delay_inlining(call_method, jvms)) { + // Delay the inlining of this method to give us the + // opportunity to perform some high level optimizations + // first. + return CallGenerator::for_late_inline(call_method, cg); + } if (cg == NULL) { // Fall through. } else if (require_inline || !InlineWarmCalls) { @@ -174,26 +182,16 @@ } } CallGenerator* miss_cg; + Deoptimization::DeoptReason reason = (profile.morphism() == 2) ? + Deoptimization::Reason_bimorphic : + Deoptimization::Reason_class_check; if (( profile.morphism() == 1 || (profile.morphism() == 2 && next_hit_cg != NULL) ) && - - !too_many_traps(Deoptimization::Reason_class_check) - - // Check only total number of traps per method to allow - // the transition from monomorphic to bimorphic case between - // compilations without falling into virtual call. - // A monomorphic case may have the class_check trap flag is set - // due to the time gap between the uncommon trap processing - // when flags are set in MDO and the call site bytecode execution - // in Interpreter when MDO counters are updated. - // There was also class_check trap in monomorphic case due to - // the bug 6225440. - + !too_many_traps(jvms->method(), jvms->bci(), reason) ) { // Generate uncommon trap for class check failure path // in case of monomorphic or bimorphic virtual call site. - miss_cg = CallGenerator::for_uncommon_trap(call_method, - Deoptimization::Reason_class_check, + miss_cg = CallGenerator::for_uncommon_trap(call_method, reason, Deoptimization::Action_maybe_recompile); } else { // Generate virtual call for class check failure path @@ -218,6 +216,57 @@ } } + // Do MethodHandle calls. + if (call_method->is_method_handle_invoke()) { + if (jvms->method()->java_code_at_bci(jvms->bci()) != Bytecodes::_invokedynamic) { + GraphKit kit(jvms); + Node* n = kit.argument(0); + + if (n->Opcode() == Op_ConP) { + const TypeOopPtr* oop_ptr = n->bottom_type()->is_oopptr(); + ciObject* const_oop = oop_ptr->const_oop(); + ciMethodHandle* method_handle = const_oop->as_method_handle(); + + // Set the actually called method to have access to the class + // and signature in the MethodHandleCompiler. + method_handle->set_callee(call_method); + + // Get an adapter for the MethodHandle. + ciMethod* target_method = method_handle->get_method_handle_adapter(); + + CallGenerator* hit_cg = this->call_generator(target_method, vtable_index, false, jvms, true, prof_factor); + if (hit_cg != NULL && hit_cg->is_inline()) + return hit_cg; + } + + return CallGenerator::for_direct_call(call_method); + } + else { + // Get the MethodHandle from the CallSite. + ciMethod* caller_method = jvms->method(); + ciBytecodeStream str(caller_method); + str.force_bci(jvms->bci()); // Set the stream to the invokedynamic bci. + ciCallSite* call_site = str.get_call_site(); + ciMethodHandle* method_handle = call_site->get_target(); + + // Set the actually called method to have access to the class + // and signature in the MethodHandleCompiler. + method_handle->set_callee(call_method); + + // Get an adapter for the MethodHandle. + ciMethod* target_method = method_handle->get_invokedynamic_adapter(); + + CallGenerator* hit_cg = this->call_generator(target_method, vtable_index, false, jvms, true, prof_factor); + if (hit_cg != NULL && hit_cg->is_inline()) { + CallGenerator* miss_cg = CallGenerator::for_dynamic_call(call_method); + return CallGenerator::for_predicted_dynamic_call(method_handle, miss_cg, hit_cg, prof_factor); + } + + // If something failed, generate a normal dynamic call. + return CallGenerator::for_dynamic_call(call_method); + } + } + // There was no special inlining tactic, or it bailed out. // Use a more generic tactic, like a simple call. if (call_is_virtual) { @@ -225,10 +274,63 @@ } else { // Class Hierarchy Analysis or Type Profile reveals a unique target, // or it is a static or special call. - return CallGenerator::for_direct_call(call_method); + return CallGenerator::for_direct_call(call_method, should_delay_inlining(call_method, jvms)); } } +// Return true for methods that shouldn't be inlined early so that +// they are easier to analyze and optimize as intrinsics. +bool Compile::should_delay_inlining(ciMethod* call_method, JVMState* jvms) { + if (has_stringbuilder()) { + + if ((call_method->holder() == C->env()->StringBuilder_klass() || + call_method->holder() == C->env()->StringBuffer_klass()) && + (jvms->method()->holder() == C->env()->StringBuilder_klass() || + jvms->method()->holder() == C->env()->StringBuffer_klass())) { + // Delay SB calls only when called from non-SB code + return false; + } + + switch (call_method->intrinsic_id()) { + case vmIntrinsics::_StringBuilder_void: + case vmIntrinsics::_StringBuilder_int: + case vmIntrinsics::_StringBuilder_String: + case vmIntrinsics::_StringBuilder_append_char: + case vmIntrinsics::_StringBuilder_append_int: + case vmIntrinsics::_StringBuilder_append_String: + case vmIntrinsics::_StringBuilder_toString: + case vmIntrinsics::_StringBuffer_void: + case vmIntrinsics::_StringBuffer_int: + case vmIntrinsics::_StringBuffer_String: + case vmIntrinsics::_StringBuffer_append_char: + case vmIntrinsics::_StringBuffer_append_int: + case vmIntrinsics::_StringBuffer_append_String: + case vmIntrinsics::_StringBuffer_toString: + case vmIntrinsics::_Integer_toString: + return true; + + case vmIntrinsics::_String_String: + { + Node* receiver = jvms->map()->in(jvms->argoff() + 1); + if (receiver->is_Proj() && receiver->in(0)->is_CallStaticJava()) { + CallStaticJavaNode* csj = receiver->in(0)->as_CallStaticJava(); + ciMethod* m = csj->method(); + if (m != NULL && + (m->intrinsic_id() == vmIntrinsics::_StringBuffer_toString || + m->intrinsic_id() == vmIntrinsics::_StringBuilder_toString)) + // Delay String.<init>(new SB()) + return true; + } + return false; + } + + default: + return false; + } + } + return false; +} + // uncommon-trap call-sites where callee is unloaded, uninitialized or will not link bool Parse::can_not_compile_call_site(ciMethod *dest_method, ciInstanceKlass* klass) { @@ -240,7 +342,7 @@ // Interface classes can be loaded & linked and never get around to // being initialized. Uncommon-trap for not-initialized static or // v-calls. Let interface calls happen. - ciInstanceKlass* holder_klass = dest_method->holder(); + ciInstanceKlass* holder_klass = dest_method->holder(); if (!holder_klass->is_initialized() && !holder_klass->is_interface()) { uncommon_trap(Deoptimization::Reason_uninitialized, @@ -248,14 +350,6 @@ holder_klass); return true; } - if (dest_method->is_method_handle_invoke() - && holder_klass->name() == ciSymbol::java_dyn_Dynamic()) { - // FIXME: NYI - uncommon_trap(Deoptimization::Reason_unhandled, - Deoptimization::Action_none, - holder_klass); - return true; - } assert(dest_method->will_link(method()->holder(), klass, bc()), "dest_method: typeflow responsibility"); return false; @@ -274,6 +368,7 @@ bool is_virtual = bc() == Bytecodes::_invokevirtual; bool is_virtual_or_interface = is_virtual || bc() == Bytecodes::_invokeinterface; bool has_receiver = is_virtual_or_interface || bc() == Bytecodes::_invokespecial; + bool is_invokedynamic = bc() == Bytecodes::_invokedynamic; // Find target being called bool will_link; @@ -282,7 +377,8 @@ ciKlass* holder = iter().get_declared_method_holder(); ciInstanceKlass* klass = ciEnv::get_instance_klass_for_declared_method_holder(holder); - int nargs = dest_method->arg_size(); + int nargs = dest_method->arg_size(); + if (is_invokedynamic) nargs -= 1; // uncommon-trap when callee is unloaded, uninitialized or will not link // bailout when too many arguments for register representation @@ -296,7 +392,7 @@ return; } assert(holder_klass->is_loaded(), ""); - assert(dest_method->is_static() == !has_receiver, "must match bc"); + assert((dest_method->is_static() || is_invokedynamic) == !has_receiver , "must match bc"); // Note: this takes into account invokeinterface of methods declared in java/lang/Object, // which should be invokevirtuals but according to the VM spec may be invokeinterfaces assert(holder_klass->is_interface() || holder_klass->super() == NULL || (bc() != Bytecodes::_invokeinterface), "must match bc");
--- a/src/share/vm/opto/escape.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/escape.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -537,11 +537,13 @@ } const TypeOopPtr *tinst = base_t->add_offset(t->offset())->is_oopptr(); - // Do NOT remove the next call: ensure an new alias index is allocated - // for the instance type + // Do NOT remove the next line: ensure a new alias index is allocated + // for the instance type. Note: C++ will not remove it since the call + // has side effect. int alias_idx = _compile->get_alias_index(tinst); igvn->set_type(addp, tinst); // record the allocation in the node map + assert(ptnode_adr(addp->_idx)->_node != NULL, "should be registered"); set_map(addp->_idx, get_map(base->_idx)); // Set addp's Base and Address to 'base'. @@ -617,9 +619,14 @@ const TypePtr *atype = C->get_adr_type(alias_idx); result = PhiNode::make(orig_phi->in(0), NULL, Type::MEMORY, atype); C->copy_node_notes_to(result, orig_phi); - set_map_phi(orig_phi->_idx, result); igvn->set_type(result, result->bottom_type()); record_for_optimizer(result); + + debug_only(Node* pn = ptnode_adr(orig_phi->_idx)->_node;) + assert(pn == NULL || pn == orig_phi, "wrong node"); + set_map(orig_phi->_idx, result); + ptnode_adr(orig_phi->_idx)->_node = orig_phi; + new_created = true; return result; } @@ -710,6 +717,81 @@ } // +// Move memory users to their memory slices. +// +void ConnectionGraph::move_inst_mem(Node* n, GrowableArray<PhiNode *> &orig_phis, PhaseGVN *igvn) { + Compile* C = _compile; + + const TypePtr* tp = igvn->type(n->in(MemNode::Address))->isa_ptr(); + assert(tp != NULL, "ptr type"); + int alias_idx = C->get_alias_index(tp); + int general_idx = C->get_general_index(alias_idx); + + // Move users first + for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { + Node* use = n->fast_out(i); + if (use->is_MergeMem()) { + MergeMemNode* mmem = use->as_MergeMem(); + assert(n == mmem->memory_at(alias_idx), "should be on instance memory slice"); + if (n != mmem->memory_at(general_idx) || alias_idx == general_idx) { + continue; // Nothing to do + } + // Replace previous general reference to mem node. + uint orig_uniq = C->unique(); + Node* m = find_inst_mem(n, general_idx, orig_phis, igvn); + assert(orig_uniq == C->unique(), "no new nodes"); + mmem->set_memory_at(general_idx, m); + --imax; + --i; + } else if (use->is_MemBar()) { + assert(!use->is_Initialize(), "initializing stores should not be moved"); + if (use->req() > MemBarNode::Precedent && + use->in(MemBarNode::Precedent) == n) { + // Don't move related membars. + record_for_optimizer(use); + continue; + } + tp = use->as_MemBar()->adr_type()->isa_ptr(); + if (tp != NULL && C->get_alias_index(tp) == alias_idx || + alias_idx == general_idx) { + continue; // Nothing to do + } + // Move to general memory slice. + uint orig_uniq = C->unique(); + Node* m = find_inst_mem(n, general_idx, orig_phis, igvn); + assert(orig_uniq == C->unique(), "no new nodes"); + igvn->hash_delete(use); + imax -= use->replace_edge(n, m); + igvn->hash_insert(use); + record_for_optimizer(use); + --i; +#ifdef ASSERT + } else if (use->is_Mem()) { + if (use->Opcode() == Op_StoreCM && use->in(MemNode::OopStore) == n) { + // Don't move related cardmark. + continue; + } + // Memory nodes should have new memory input. + tp = igvn->type(use->in(MemNode::Address))->isa_ptr(); + assert(tp != NULL, "ptr type"); + int idx = C->get_alias_index(tp); + assert(get_map(use->_idx) != NULL || idx == alias_idx, + "Following memory nodes should have new memory input or be on the same memory slice"); + } else if (use->is_Phi()) { + // Phi nodes should be split and moved already. + tp = use->as_Phi()->adr_type()->isa_ptr(); + assert(tp != NULL, "ptr type"); + int idx = C->get_alias_index(tp); + assert(idx == alias_idx, "Following Phi nodes should be on the same memory slice"); + } else { + use->dump(); + assert(false, "should not be here"); +#endif + } + } +} + +// // Search memory chain of "mem" to find a MemNode whose address // is the specified alias index. // @@ -774,10 +856,18 @@ C->get_alias_index(result->as_Phi()->adr_type()) != alias_idx) { Node *un = result->as_Phi()->unique_input(phase); if (un != NULL) { + orig_phis.append_if_missing(result->as_Phi()); result = un; } else { break; } + } else if (result->is_ClearArray()) { + if (!ClearArrayNode::step_through(&result, (uint)tinst->instance_id(), phase)) { + // Can not bypass initialization of the instance + // we are looking for. + break; + } + // Otherwise skip it (the call updated 'result' value). } else if (result->Opcode() == Op_SCMemProj) { assert(result->in(0)->is_LoadStore(), "sanity"); const Type *at = phase->type(result->in(0)->in(MemNode::Address)); @@ -807,7 +897,6 @@ return result; } - // // Convert the types of unescaped object to instance types where possible, // propagate the new type information through the graph, and update memory @@ -899,12 +988,13 @@ // void ConnectionGraph::split_unique_types(GrowableArray<Node *> &alloc_worklist) { GrowableArray<Node *> memnode_worklist; - GrowableArray<Node *> mergemem_worklist; GrowableArray<PhiNode *> orig_phis; + PhaseGVN *igvn = _compile->initial_gvn(); uint new_index_start = (uint) _compile->num_alias_types(); - VectorSet visited(Thread::current()->resource_area()); - VectorSet ptset(Thread::current()->resource_area()); + Arena* arena = Thread::current()->resource_area(); + VectorSet visited(arena); + VectorSet ptset(arena); // Phase 1: Process possible allocations from alloc_worklist. @@ -980,6 +1070,8 @@ // - non-escaping // - eligible to be a unique type // - not determined to be ineligible by escape analysis + assert(ptnode_adr(alloc->_idx)->_node != NULL && + ptnode_adr(n->_idx)->_node != NULL, "should be registered"); set_map(alloc->_idx, n); set_map(n->_idx, alloc); const TypeOopPtr *t = igvn->type(n)->isa_oopptr(); @@ -1024,7 +1116,7 @@ alloc_worklist.append_if_missing(addp2); } alloc_worklist.append_if_missing(use); - } else if (use->is_Initialize()) { + } else if (use->is_MemBar()) { memnode_worklist.append_if_missing(use); } } @@ -1034,10 +1126,12 @@ PointsTo(ptset, get_addp_base(n), igvn); assert(ptset.Size() == 1, "AddP address is unique"); uint elem = ptset.getelem(); // Allocation node's index - if (elem == _phantom_object) + if (elem == _phantom_object) { + assert(false, "escaped allocation"); continue; // Assume the value was set outside this method. + } Node *base = get_map(elem); // CheckCastPP node - if (!split_AddP(n, base, igvn)) continue; // wrong type + if (!split_AddP(n, base, igvn)) continue; // wrong type from dead path tinst = igvn->type(base)->isa_oopptr(); } else if (n->is_Phi() || n->is_CheckCastPP() || @@ -1052,8 +1146,10 @@ PointsTo(ptset, n, igvn); if (ptset.Size() == 1) { uint elem = ptset.getelem(); // Allocation node's index - if (elem == _phantom_object) + if (elem == _phantom_object) { + assert(false, "escaped allocation"); continue; // Assume the value was set outside this method. + } Node *val = get_map(elem); // CheckCastPP node TypeNode *tn = n->as_Type(); tinst = igvn->type(val)->isa_oopptr(); @@ -1068,8 +1164,7 @@ tn_t = tn_type->isa_oopptr(); } - if (tn_t != NULL && - tinst->cast_to_instance_id(TypeOopPtr::InstanceBot)->higher_equal(tn_t)) { + if (tn_t != NULL && tinst->klass()->is_subtype_of(tn_t->klass())) { if (tn_type->isa_narrowoop()) { tn_type = tinst->make_narrowoop(); } else { @@ -1081,33 +1176,25 @@ igvn->hash_insert(tn); record_for_optimizer(n); } else { - continue; // wrong type + assert(tn_type == TypePtr::NULL_PTR || + tn_t != NULL && !tinst->klass()->is_subtype_of(tn_t->klass()), + "unexpected type"); + continue; // Skip dead path with different type } } } else { + debug_only(n->dump();) + assert(false, "EA: unexpected node"); continue; } - // push users on appropriate worklist + // push allocation's users on appropriate worklist for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { Node *use = n->fast_out(i); if(use->is_Mem() && use->in(MemNode::Address) == n) { - memnode_worklist.append_if_missing(use); - } else if (use->is_Initialize()) { + // Load/store to instance's field memnode_worklist.append_if_missing(use); - } else if (use->is_MergeMem()) { - mergemem_worklist.append_if_missing(use); - } else if (use->is_SafePoint() && tinst != NULL) { - // Look for MergeMem nodes for calls which reference unique allocation - // (through CheckCastPP nodes) even for debug info. - Node* m = use->in(TypeFunc::Memory); - uint iid = tinst->instance_id(); - while (m->is_Proj() && m->in(0)->is_SafePoint() && - m->in(0) != use && !m->in(0)->_idx != iid) { - m = m->in(0)->in(TypeFunc::Memory); - } - if (m->is_MergeMem()) { - mergemem_worklist.append_if_missing(m); - } + } else if (use->is_MemBar()) { + memnode_worklist.append_if_missing(use); } else if (use->is_AddP() && use->outcnt() > 0) { // No dead nodes Node* addp2 = find_second_addp(use, n); if (addp2 != NULL) { @@ -1120,6 +1207,29 @@ use->is_DecodeN() || (use->is_ConstraintCast() && use->Opcode() == Op_CastPP)) { alloc_worklist.append_if_missing(use); +#ifdef ASSERT + } else if (use->is_Mem()) { + assert(use->in(MemNode::Address) != n, "EA: missing allocation reference path"); + } else if (use->is_MergeMem()) { + assert(_mergemem_worklist.contains(use->as_MergeMem()), "EA: missing MergeMem node in the worklist"); + } else if (use->is_SafePoint()) { + // Look for MergeMem nodes for calls which reference unique allocation + // (through CheckCastPP nodes) even for debug info. + Node* m = use->in(TypeFunc::Memory); + if (m->is_MergeMem()) { + assert(_mergemem_worklist.contains(m->as_MergeMem()), "EA: missing MergeMem node in the worklist"); + } + } else { + uint op = use->Opcode(); + if (!(op == Op_CmpP || op == Op_Conv2B || + op == Op_CastP2X || op == Op_StoreCM || + op == Op_FastLock || op == Op_AryEq || op == Op_StrComp || + op == Op_StrEquals || op == Op_StrIndexOf)) { + n->dump(); + use->dump(); + assert(false, "EA: missing allocation reference path"); + } +#endif } } @@ -1137,19 +1247,16 @@ Node *n = memnode_worklist.pop(); if (visited.test_set(n->_idx)) continue; - if (n->is_Phi()) { - assert(n->as_Phi()->adr_type() != TypePtr::BOTTOM, "narrow memory slice required"); - // we don't need to do anything, but the users must be pushed if we haven't processed - // this Phi before - } else if (n->is_Initialize()) { - // we don't need to do anything, but the users of the memory projection must be pushed - n = n->as_Initialize()->proj_out(TypeFunc::Memory); + if (n->is_Phi() || n->is_ClearArray()) { + // we don't need to do anything, but the users must be pushed + } else if (n->is_MemBar()) { // Initialize, MemBar nodes + // we don't need to do anything, but the users must be pushed + n = n->as_MemBar()->proj_out(TypeFunc::Memory); if (n == NULL) continue; } else { assert(n->is_Mem(), "memory node required."); Node *addr = n->in(MemNode::Address); - assert(addr->is_AddP(), "AddP required"); const Type *addr_t = igvn->type(addr); if (addr_t == Type::TOP) continue; @@ -1161,6 +1268,10 @@ return; } if (mem != n->in(MemNode::Memory)) { + // We delay the memory edge update since we need old one in + // MergeMem code below when instances memory slices are separated. + debug_only(Node* pn = ptnode_adr(n->_idx)->_node;) + assert(pn == NULL || pn == n, "wrong node"); set_map(n->_idx, mem); ptnode_adr(n->_idx)->_node = n; } @@ -1181,36 +1292,55 @@ // push user on appropriate worklist for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { Node *use = n->fast_out(i); - if (use->is_Phi()) { + if (use->is_Phi() || use->is_ClearArray()) { memnode_worklist.append_if_missing(use); } else if(use->is_Mem() && use->in(MemNode::Memory) == n) { + if (use->Opcode() == Op_StoreCM) // Ignore cardmark stores + continue; memnode_worklist.append_if_missing(use); - } else if (use->is_Initialize()) { + } else if (use->is_MemBar()) { memnode_worklist.append_if_missing(use); +#ifdef ASSERT + } else if(use->is_Mem()) { + assert(use->in(MemNode::Memory) != n, "EA: missing memory path"); } else if (use->is_MergeMem()) { - mergemem_worklist.append_if_missing(use); + assert(_mergemem_worklist.contains(use->as_MergeMem()), "EA: missing MergeMem node in the worklist"); + } else { + uint op = use->Opcode(); + if (!(op == Op_StoreCM || + (op == Op_CallLeaf && use->as_CallLeaf()->_name != NULL && + strcmp(use->as_CallLeaf()->_name, "g1_wb_pre") == 0) || + op == Op_AryEq || op == Op_StrComp || + op == Op_StrEquals || op == Op_StrIndexOf)) { + n->dump(); + use->dump(); + assert(false, "EA: missing memory path"); + } +#endif } } } // Phase 3: Process MergeMem nodes from mergemem_worklist. - // Walk each memory moving the first node encountered of each + // Walk each memory slice moving the first node encountered of each // instance type to the the input corresponding to its alias index. - while (mergemem_worklist.length() != 0) { - Node *n = mergemem_worklist.pop(); - assert(n->is_MergeMem(), "MergeMem node required."); - if (visited.test_set(n->_idx)) - continue; - MergeMemNode *nmm = n->as_MergeMem(); + uint length = _mergemem_worklist.length(); + for( uint next = 0; next < length; ++next ) { + MergeMemNode* nmm = _mergemem_worklist.at(next); + assert(!visited.test_set(nmm->_idx), "should not be visited before"); // Note: we don't want to use MergeMemStream here because we only want to - // scan inputs which exist at the start, not ones we add during processing. + // scan inputs which exist at the start, not ones we add during processing. + // Note 2: MergeMem may already contains instance memory slices added + // during find_inst_mem() call when memory nodes were processed above. + igvn->hash_delete(nmm); uint nslices = nmm->req(); - igvn->hash_delete(nmm); for (uint i = Compile::AliasIdxRaw+1; i < nslices; i++) { Node* mem = nmm->in(i); Node* cur = NULL; if (mem == NULL || mem->is_top()) continue; + // First, update mergemem by moving memory nodes to corresponding slices + // if their type became more precise since this mergemem was created. while (mem->is_Mem()) { const Type *at = igvn->type(mem->in(MemNode::Address)); if (at != Type::TOP) { @@ -1229,7 +1359,7 @@ } nmm->set_memory_at(i, (cur != NULL) ? cur : mem); // Find any instance of the current type if we haven't encountered - // a value of the instance along the chain. + // already a memory slice of the instance along the memory chain. for (uint ni = new_index_start; ni < new_index_end; ni++) { if((uint)_compile->get_general_index(ni) == i) { Node *m = (ni >= nmm->req()) ? nmm->empty_memory() : nmm->in(ni); @@ -1245,11 +1375,11 @@ } // Find the rest of instances values for (uint ni = new_index_start; ni < new_index_end; ni++) { - const TypeOopPtr *tinst = igvn->C->get_adr_type(ni)->isa_oopptr(); + const TypeOopPtr *tinst = _compile->get_adr_type(ni)->isa_oopptr(); Node* result = step_through_mergemem(nmm, ni, tinst); if (result == nmm->base_memory()) { // Didn't find instance memory, search through general slice recursively. - result = nmm->memory_at(igvn->C->get_general_index(ni)); + result = nmm->memory_at(_compile->get_general_index(ni)); result = find_inst_mem(result, ni, orig_phis, igvn); if (_compile->failing()) { return; @@ -1259,41 +1389,6 @@ } igvn->hash_insert(nmm); record_for_optimizer(nmm); - - // Propagate new memory slices to following MergeMem nodes. - for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { - Node *use = n->fast_out(i); - if (use->is_Call()) { - CallNode* in = use->as_Call(); - if (in->proj_out(TypeFunc::Memory) != NULL) { - Node* m = in->proj_out(TypeFunc::Memory); - for (DUIterator_Fast jmax, j = m->fast_outs(jmax); j < jmax; j++) { - Node* mm = m->fast_out(j); - if (mm->is_MergeMem()) { - mergemem_worklist.append_if_missing(mm); - } - } - } - if (use->is_Allocate()) { - use = use->as_Allocate()->initialization(); - if (use == NULL) { - continue; - } - } - } - if (use->is_Initialize()) { - InitializeNode* in = use->as_Initialize(); - if (in->proj_out(TypeFunc::Memory) != NULL) { - Node* m = in->proj_out(TypeFunc::Memory); - for (DUIterator_Fast jmax, j = m->fast_outs(jmax); j < jmax; j++) { - Node* mm = m->fast_out(j); - if (mm->is_MergeMem()) { - mergemem_worklist.append_if_missing(mm); - } - } - } - } - } } // Phase 4: Update the inputs of non-instance memory Phis and @@ -1322,19 +1417,48 @@ } // Update the memory inputs of MemNodes with the value we computed - // in Phase 2. + // in Phase 2 and move stores memory users to corresponding memory slices. +#ifdef ASSERT + visited.Clear(); + Node_Stack old_mems(arena, _compile->unique() >> 2); +#endif for (uint i = 0; i < nodes_size(); i++) { Node *nmem = get_map(i); if (nmem != NULL) { Node *n = ptnode_adr(i)->_node; - if (n != NULL && n->is_Mem()) { + assert(n != NULL, "sanity"); + if (n->is_Mem()) { +#ifdef ASSERT + Node* old_mem = n->in(MemNode::Memory); + if (!visited.test_set(old_mem->_idx)) { + old_mems.push(old_mem, old_mem->outcnt()); + } +#endif + assert(n->in(MemNode::Memory) != nmem, "sanity"); + if (!n->is_Load()) { + // Move memory users of a store first. + move_inst_mem(n, orig_phis, igvn); + } + // Now update memory input igvn->hash_delete(n); n->set_req(MemNode::Memory, nmem); igvn->hash_insert(n); record_for_optimizer(n); + } else { + assert(n->is_Allocate() || n->is_CheckCastPP() || + n->is_AddP() || n->is_Phi(), "unknown node used for set_map()"); } } } +#ifdef ASSERT + // Verify that memory was split correctly + while (old_mems.is_nonempty()) { + Node* old_mem = old_mems.node(); + uint old_cnt = old_mems.index(); + old_mems.pop(); + assert(old_cnt = old_mem->outcnt(), "old mem could be lost"); + } +#endif } bool ConnectionGraph::has_candidates(Compile *C) { @@ -1381,8 +1505,20 @@ ptnode_adr(n->_idx)->node_type() == PointsToNode::JavaObject) { has_allocations = true; } - if(n->is_AddP()) - cg_worklist.append(n->_idx); + if(n->is_AddP()) { + // Collect address nodes which directly reference an allocation. + // Use them during stage 3 below to build initial connection graph + // field edges. Other field edges could be added after StoreP/LoadP + // nodes are processed during stage 4 below. + Node* base = get_addp_base(n); + if(base->is_Proj() && base->in(0)->is_Allocate()) { + cg_worklist.append(n->_idx); + } + } else if (n->is_MergeMem()) { + // Collect all MergeMem nodes to add memory slices for + // scalar replaceable objects in split_unique_types(). + _mergemem_worklist.append(n->as_MergeMem()); + } for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { Node* m = n->fast_out(i); // Get user worklist_init.push(m); @@ -1423,12 +1559,13 @@ } } - VectorSet ptset(Thread::current()->resource_area()); + Arena* arena = Thread::current()->resource_area(); + VectorSet ptset(arena); GrowableArray<uint> deferred_edges; - VectorSet visited(Thread::current()->resource_area()); + VectorSet visited(arena); - // 5. Remove deferred edges from the graph and collect - // information needed for type splitting. + // 5. Remove deferred edges from the graph and adjust + // escape state of nonescaping objects. cg_length = cg_worklist.length(); for( uint next = 0; next < cg_length; ++next ) { int ni = cg_worklist.at(next); @@ -1438,98 +1575,9 @@ remove_deferred(ni, &deferred_edges, &visited); Node *n = ptn->_node; if (n->is_AddP()) { - // Search for objects which are not scalar replaceable. - // Mark their escape state as ArgEscape to propagate the state - // to referenced objects. - // Note: currently there are no difference in compiler optimizations - // for ArgEscape objects and NoEscape objects which are not - // scalar replaceable. - - int offset = ptn->offset(); - Node *base = get_addp_base(n); - ptset.Clear(); - PointsTo(ptset, base, igvn); - int ptset_size = ptset.Size(); - - // Check if a field's initializing value is recorded and add - // a corresponding NULL field's value if it is not recorded. - // Connection Graph does not record a default initialization by NULL - // captured by Initialize node. - // - // Note: it will disable scalar replacement in some cases: - // - // Point p[] = new Point[1]; - // p[0] = new Point(); // Will be not scalar replaced - // - // but it will save us from incorrect optimizations in next cases: - // - // Point p[] = new Point[1]; - // if ( x ) p[0] = new Point(); // Will be not scalar replaced - // - // Without a control flow analysis we can't distinguish above cases. - // - if (offset != Type::OffsetBot && ptset_size == 1) { - uint elem = ptset.getelem(); // Allocation node's index - // It does not matter if it is not Allocation node since - // only non-escaping allocations are scalar replaced. - if (ptnode_adr(elem)->_node->is_Allocate() && - ptnode_adr(elem)->escape_state() == PointsToNode::NoEscape) { - AllocateNode* alloc = ptnode_adr(elem)->_node->as_Allocate(); - InitializeNode* ini = alloc->initialization(); - Node* value = NULL; - if (ini != NULL) { - BasicType ft = UseCompressedOops ? T_NARROWOOP : T_OBJECT; - Node* store = ini->find_captured_store(offset, type2aelembytes(ft), igvn); - if (store != NULL && store->is_Store()) - value = store->in(MemNode::ValueIn); - } - if (value == NULL || value != ptnode_adr(value->_idx)->_node) { - // A field's initializing value was not recorded. Add NULL. - uint null_idx = UseCompressedOops ? _noop_null : _oop_null; - add_pointsto_edge(ni, null_idx); - } - } - } - - // An object is not scalar replaceable if the field which may point - // to it has unknown offset (unknown element of an array of objects). - // - if (offset == Type::OffsetBot) { - uint e_cnt = ptn->edge_count(); - for (uint ei = 0; ei < e_cnt; ei++) { - uint npi = ptn->edge_target(ei); - set_escape_state(npi, PointsToNode::ArgEscape); - ptnode_adr(npi)->_scalar_replaceable = false; - } - } - - // Currently an object is not scalar replaceable if a LoadStore node - // access its field since the field value is unknown after it. - // - bool has_LoadStore = false; - for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { - Node *use = n->fast_out(i); - if (use->is_LoadStore()) { - has_LoadStore = true; - break; - } - } - // An object is not scalar replaceable if the address points - // to unknown field (unknown element for arrays, offset is OffsetBot). - // - // Or the address may point to more then one object. This may produce - // the false positive result (set scalar_replaceable to false) - // since the flow-insensitive escape analysis can't separate - // the case when stores overwrite the field's value from the case - // when stores happened on different control branches. - // - if (ptset_size > 1 || ptset_size != 0 && - (has_LoadStore || offset == Type::OffsetBot)) { - for( VectorSetI j(&ptset); j.test(); ++j ) { - set_escape_state(j.elem, PointsToNode::ArgEscape); - ptnode_adr(j.elem)->_scalar_replaceable = false; - } - } + // Search for objects which are not scalar replaceable + // and adjust their escape state. + verify_escape_state(ni, ptset, igvn); } } } @@ -1646,6 +1694,150 @@ return has_non_escaping_obj; } +// Search for objects which are not scalar replaceable. +void ConnectionGraph::verify_escape_state(int nidx, VectorSet& ptset, PhaseTransform* phase) { + PointsToNode* ptn = ptnode_adr(nidx); + Node* n = ptn->_node; + assert(n->is_AddP(), "Should be called for AddP nodes only"); + // Search for objects which are not scalar replaceable. + // Mark their escape state as ArgEscape to propagate the state + // to referenced objects. + // Note: currently there are no difference in compiler optimizations + // for ArgEscape objects and NoEscape objects which are not + // scalar replaceable. + + Compile* C = _compile; + + int offset = ptn->offset(); + Node* base = get_addp_base(n); + ptset.Clear(); + PointsTo(ptset, base, phase); + int ptset_size = ptset.Size(); + + // Check if a oop field's initializing value is recorded and add + // a corresponding NULL field's value if it is not recorded. + // Connection Graph does not record a default initialization by NULL + // captured by Initialize node. + // + // Note: it will disable scalar replacement in some cases: + // + // Point p[] = new Point[1]; + // p[0] = new Point(); // Will be not scalar replaced + // + // but it will save us from incorrect optimizations in next cases: + // + // Point p[] = new Point[1]; + // if ( x ) p[0] = new Point(); // Will be not scalar replaced + // + // Do a simple control flow analysis to distinguish above cases. + // + if (offset != Type::OffsetBot && ptset_size == 1) { + uint elem = ptset.getelem(); // Allocation node's index + // It does not matter if it is not Allocation node since + // only non-escaping allocations are scalar replaced. + if (ptnode_adr(elem)->_node->is_Allocate() && + ptnode_adr(elem)->escape_state() == PointsToNode::NoEscape) { + AllocateNode* alloc = ptnode_adr(elem)->_node->as_Allocate(); + InitializeNode* ini = alloc->initialization(); + + // Check only oop fields. + const Type* adr_type = n->as_AddP()->bottom_type(); + BasicType basic_field_type = T_INT; + if (adr_type->isa_instptr()) { + ciField* field = C->alias_type(adr_type->isa_instptr())->field(); + if (field != NULL) { + basic_field_type = field->layout_type(); + } else { + // Ignore non field load (for example, klass load) + } + } else if (adr_type->isa_aryptr()) { + const Type* elemtype = adr_type->isa_aryptr()->elem(); + basic_field_type = elemtype->array_element_basic_type(); + } else { + // Raw pointers are used for initializing stores so skip it. + assert(adr_type->isa_rawptr() && base->is_Proj() && + (base->in(0) == alloc),"unexpected pointer type"); + } + if (basic_field_type == T_OBJECT || + basic_field_type == T_NARROWOOP || + basic_field_type == T_ARRAY) { + Node* value = NULL; + if (ini != NULL) { + BasicType ft = UseCompressedOops ? T_NARROWOOP : T_OBJECT; + Node* store = ini->find_captured_store(offset, type2aelembytes(ft), phase); + if (store != NULL && store->is_Store()) { + value = store->in(MemNode::ValueIn); + } else if (ptn->edge_count() > 0) { // Are there oop stores? + // Check for a store which follows allocation without branches. + // For example, a volatile field store is not collected + // by Initialize node. TODO: it would be nice to use idom() here. + for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { + store = n->fast_out(i); + if (store->is_Store() && store->in(0) != NULL) { + Node* ctrl = store->in(0); + while(!(ctrl == ini || ctrl == alloc || ctrl == NULL || + ctrl == C->root() || ctrl == C->top() || ctrl->is_Region() || + ctrl->is_IfTrue() || ctrl->is_IfFalse())) { + ctrl = ctrl->in(0); + } + if (ctrl == ini || ctrl == alloc) { + value = store->in(MemNode::ValueIn); + break; + } + } + } + } + } + if (value == NULL || value != ptnode_adr(value->_idx)->_node) { + // A field's initializing value was not recorded. Add NULL. + uint null_idx = UseCompressedOops ? _noop_null : _oop_null; + add_pointsto_edge(nidx, null_idx); + } + } + } + } + + // An object is not scalar replaceable if the field which may point + // to it has unknown offset (unknown element of an array of objects). + // + if (offset == Type::OffsetBot) { + uint e_cnt = ptn->edge_count(); + for (uint ei = 0; ei < e_cnt; ei++) { + uint npi = ptn->edge_target(ei); + set_escape_state(npi, PointsToNode::ArgEscape); + ptnode_adr(npi)->_scalar_replaceable = false; + } + } + + // Currently an object is not scalar replaceable if a LoadStore node + // access its field since the field value is unknown after it. + // + bool has_LoadStore = false; + for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { + Node *use = n->fast_out(i); + if (use->is_LoadStore()) { + has_LoadStore = true; + break; + } + } + // An object is not scalar replaceable if the address points + // to unknown field (unknown element for arrays, offset is OffsetBot). + // + // Or the address may point to more then one object. This may produce + // the false positive result (set scalar_replaceable to false) + // since the flow-insensitive escape analysis can't separate + // the case when stores overwrite the field's value from the case + // when stores happened on different control branches. + // + if (ptset_size > 1 || ptset_size != 0 && + (has_LoadStore || offset == Type::OffsetBot)) { + for( VectorSetI j(&ptset); j.test(); ++j ) { + set_escape_state(j.elem, PointsToNode::ArgEscape); + ptnode_adr(j.elem)->_scalar_replaceable = false; + } + } +} + void ConnectionGraph::process_call_arguments(CallNode *call, PhaseTransform *phase) { switch (call->Opcode()) { @@ -1657,6 +1849,7 @@ assert(false, "should be done already"); break; #endif + case Op_CallLeaf: case Op_CallLeafNoFP: { // Stub calls, objects do not escape but they are not scale replaceable. @@ -1667,9 +1860,23 @@ const Type* at = d->field_at(i); Node *arg = call->in(i)->uncast(); const Type *aat = phase->type(arg); - if (!arg->is_top() && at->isa_ptr() && aat->isa_ptr()) { + if (!arg->is_top() && at->isa_ptr() && aat->isa_ptr() && + ptnode_adr(arg->_idx)->escape_state() < PointsToNode::ArgEscape) { + assert(aat == Type::TOP || aat == TypePtr::NULL_PTR || aat->isa_ptr() != NULL, "expecting an Ptr"); +#ifdef ASSERT + if (!(call->Opcode() == Op_CallLeafNoFP && + call->as_CallLeaf()->_name != NULL && + (strstr(call->as_CallLeaf()->_name, "arraycopy") != 0) || + call->as_CallLeaf()->_name != NULL && + (strcmp(call->as_CallLeaf()->_name, "g1_wb_pre") == 0 || + strcmp(call->as_CallLeaf()->_name, "g1_wb_post") == 0 )) + ) { + call->dump(); + assert(false, "EA: unexpected CallLeaf"); + } +#endif set_escape_state(arg->_idx, PointsToNode::ArgEscape); if (arg->is_AddP()) { // @@ -1706,9 +1913,10 @@ for (uint i = TypeFunc::Parms; i < d->cnt(); i++) { const Type* at = d->field_at(i); int k = i - TypeFunc::Parms; + Node *arg = call->in(i)->uncast(); - if (at->isa_oopptr() != NULL) { - Node *arg = call->in(i)->uncast(); + if (at->isa_oopptr() != NULL && + ptnode_adr(arg->_idx)->escape_state() < PointsToNode::GlobalEscape) { bool global_escapes = false; bool fields_escapes = false; @@ -1942,20 +2150,23 @@ record_for_optimizer(n); _processed.set(n->_idx); } else { - // Have to process call's arguments first. + // Don't mark as processed since call's arguments have to be processed. PointsToNode::NodeType nt = PointsToNode::UnknownType; + PointsToNode::EscapeState es = PointsToNode::UnknownEscape; // Check if a call returns an object. const TypeTuple *r = n->as_Call()->tf()->range(); - if (n->is_CallStaticJava() && r->cnt() > TypeFunc::Parms && + if (r->cnt() > TypeFunc::Parms && + r->field_at(TypeFunc::Parms)->isa_ptr() && n->as_Call()->proj_out(TypeFunc::Parms) != NULL) { - // Note: use isa_ptr() instead of isa_oopptr() here because - // the _multianewarray functions return a TypeRawPtr. - if (r->field_at(TypeFunc::Parms)->isa_ptr() != NULL) { - nt = PointsToNode::JavaObject; + nt = PointsToNode::JavaObject; + if (!n->is_CallStaticJava()) { + // Since the called mathod is statically unknown assume + // the worst case that the returned value globally escapes. + es = PointsToNode::GlobalEscape; } } - add_node(n, nt, PointsToNode::UnknownEscape, false); + add_node(n, nt, es, false); } return; } @@ -2088,18 +2299,27 @@ } case Op_Proj: { - // we are only interested in the result projection from a call + // we are only interested in the oop result projection from a call if (n->as_Proj()->_con == TypeFunc::Parms && n->in(0)->is_Call() ) { - add_node(n, PointsToNode::LocalVar, PointsToNode::UnknownEscape, false); - process_call_result(n->as_Proj(), phase); - if (!_processed.test(n->_idx)) { - // The call's result may need to be processed later if the call - // returns it's argument and the argument is not processed yet. - _delayed_worklist.push(n); + const TypeTuple *r = n->in(0)->as_Call()->tf()->range(); + assert(r->cnt() > TypeFunc::Parms, "sanity"); + if (r->field_at(TypeFunc::Parms)->isa_ptr() != NULL) { + add_node(n, PointsToNode::LocalVar, PointsToNode::UnknownEscape, false); + int ti = n->in(0)->_idx; + // The call may not be registered yet (since not all its inputs are registered) + // if this is the projection from backbranch edge of Phi. + if (ptnode_adr(ti)->node_type() != PointsToNode::UnknownType) { + process_call_result(n->as_Proj(), phase); + } + if (!_processed.test(n->_idx)) { + // The call's result may need to be processed later if the call + // returns it's argument and the argument is not processed yet. + _delayed_worklist.push(n); + } + break; } - } else { - _processed.set(n->_idx); } + _processed.set(n->_idx); break; } case Op_Return: @@ -2160,6 +2380,15 @@ } break; } + case Op_AryEq: + case Op_StrComp: + case Op_StrEquals: + case Op_StrIndexOf: + { + // char[] arrays passed to string intrinsics are not scalar replaceable. + add_node(n, PointsToNode::UnknownType, PointsToNode::UnknownEscape, false); + break; + } case Op_ThreadLocal: { add_node(n, PointsToNode::JavaObject, PointsToNode::ArgEscape, true); @@ -2174,6 +2403,7 @@ void ConnectionGraph::build_connection_graph(Node *n, PhaseTransform *phase) { uint n_idx = n->_idx; + assert(ptnode_adr(n_idx)->_node != NULL, "node should be registered"); // Don't set processed bit for AddP, LoadP, StoreP since // they may need more then one pass to process. @@ -2211,6 +2441,7 @@ case Op_DecodeN: { int ti = n->in(1)->_idx; + assert(ptnode_adr(ti)->node_type() != PointsToNode::UnknownType, "all nodes should be registered"); if (ptnode_adr(ti)->node_type() == PointsToNode::JavaObject) { add_pointsto_edge(n_idx, ti); } else { @@ -2250,7 +2481,6 @@ #endif Node* adr = n->in(MemNode::Address)->uncast(); - const Type *adr_type = phase->type(adr); Node* adr_base; if (adr->is_AddP()) { adr_base = get_addp_base(adr); @@ -2302,13 +2532,19 @@ } case Op_Proj: { - // we are only interested in the result projection from a call + // we are only interested in the oop result projection from a call if (n->as_Proj()->_con == TypeFunc::Parms && n->in(0)->is_Call() ) { - process_call_result(n->as_Proj(), phase); - assert(_processed.test(n_idx), "all call results should be processed"); - } else { - assert(false, "Op_Proj"); + assert(ptnode_adr(n->in(0)->_idx)->node_type() != PointsToNode::UnknownType, + "all nodes should be registered"); + const TypeTuple *r = n->in(0)->as_Call()->tf()->range(); + assert(r->cnt() > TypeFunc::Parms, "sanity"); + if (r->field_at(TypeFunc::Parms)->isa_ptr() != NULL) { + process_call_result(n->as_Proj(), phase); + assert(_processed.test(n_idx), "all call results should be processed"); + break; + } } + assert(false, "Op_Proj"); break; } case Op_Return: @@ -2320,6 +2556,7 @@ } #endif int ti = n->in(TypeFunc::Parms)->_idx; + assert(ptnode_adr(ti)->node_type() != PointsToNode::UnknownType, "node should be registered"); if (ptnode_adr(ti)->node_type() == PointsToNode::JavaObject) { add_pointsto_edge(n_idx, ti); } else { @@ -2354,14 +2591,38 @@ } break; } + case Op_AryEq: + case Op_StrComp: + case Op_StrEquals: + case Op_StrIndexOf: + { + // char[] arrays passed to string intrinsic do not escape but + // they are not scalar replaceable. Adjust escape state for them. + // Start from in(2) edge since in(1) is memory edge. + for (uint i = 2; i < n->req(); i++) { + Node* adr = n->in(i)->uncast(); + const Type *at = phase->type(adr); + if (!adr->is_top() && at->isa_ptr()) { + assert(at == Type::TOP || at == TypePtr::NULL_PTR || + at->isa_ptr() != NULL, "expecting an Ptr"); + if (adr->is_AddP()) { + adr = get_addp_base(adr); + } + // Mark as ArgEscape everything "adr" could point to. + set_escape_state(adr->_idx, PointsToNode::ArgEscape); + } + } + _processed.set(n_idx); + break; + } case Op_ThreadLocal: { assert(false, "Op_ThreadLocal"); break; } default: - ; - // nothing to do + // This method should be called only for EA specific nodes. + ShouldNotReachHere(); } }
--- a/src/share/vm/opto/escape.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/escape.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -210,6 +210,8 @@ Unique_Node_List _delayed_worklist; // Nodes to be processed before // the call build_connection_graph(). + GrowableArray<MergeMemNode *> _mergemem_worklist; // List of all MergeMem nodes + VectorSet _processed; // Records which nodes have been // processed. @@ -289,7 +291,7 @@ bool split_AddP(Node *addp, Node *base, PhaseGVN *igvn); PhiNode *create_split_phi(PhiNode *orig_phi, int alias_idx, GrowableArray<PhiNode *> &orig_phi_worklist, PhaseGVN *igvn, bool &new_created); PhiNode *split_memory_phi(PhiNode *orig_phi, int alias_idx, GrowableArray<PhiNode *> &orig_phi_worklist, PhaseGVN *igvn); - Node *find_mem(Node *mem, int alias_idx, PhaseGVN *igvn); + void move_inst_mem(Node* n, GrowableArray<PhiNode *> &orig_phis, PhaseGVN *igvn); Node *find_inst_mem(Node *mem, int alias_idx,GrowableArray<PhiNode *> &orig_phi_worklist, PhaseGVN *igvn); // Propagate unique types created for unescaped allocated objects @@ -298,7 +300,6 @@ // manage entries in _node_map void set_map(int idx, Node *n) { _node_map.map(idx, n); } - void set_map_phi(int idx, PhiNode *p) { _node_map.map(idx, (Node *) p); } Node *get_map(int idx) { return _node_map[idx]; } PhiNode *get_map_phi(int idx) { Node *phi = _node_map[idx]; @@ -315,6 +316,9 @@ // Set the escape state of a node void set_escape_state(uint ni, PointsToNode::EscapeState es); + // Search for objects which are not scalar replaceable. + void verify_escape_state(int nidx, VectorSet& ptset, PhaseTransform* phase); + public: ConnectionGraph(Compile *C);
--- a/src/share/vm/opto/graphKit.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/graphKit.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2001-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2001-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -455,16 +455,44 @@ return Bytecodes::_illegal; } +void GraphKit::uncommon_trap_if_should_post_on_exceptions(Deoptimization::DeoptReason reason, + bool must_throw) { + // if the exception capability is set, then we will generate code + // to check the JavaThread.should_post_on_exceptions flag to see + // if we actually need to report exception events (for this + // thread). If we don't need to report exception events, we will + // take the normal fast path provided by add_exception_events. If + // exception event reporting is enabled for this thread, we will + // take the uncommon_trap in the BuildCutout below. + + // first must access the should_post_on_exceptions_flag in this thread's JavaThread + Node* jthread = _gvn.transform(new (C, 1) ThreadLocalNode()); + Node* adr = basic_plus_adr(top(), jthread, in_bytes(JavaThread::should_post_on_exceptions_flag_offset())); + Node* should_post_flag = make_load(control(), adr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, false); + + // Test the should_post_on_exceptions_flag vs. 0 + Node* chk = _gvn.transform( new (C, 3) CmpINode(should_post_flag, intcon(0)) ); + Node* tst = _gvn.transform( new (C, 2) BoolNode(chk, BoolTest::eq) ); + + // Branch to slow_path if should_post_on_exceptions_flag was true + { BuildCutout unless(this, tst, PROB_MAX); + // Do not try anything fancy if we're notifying the VM on every throw. + // Cf. case Bytecodes::_athrow in parse2.cpp. + uncommon_trap(reason, Deoptimization::Action_none, + (ciKlass*)NULL, (char*)NULL, must_throw); + } + +} + //------------------------------builtin_throw---------------------------------- void GraphKit::builtin_throw(Deoptimization::DeoptReason reason, Node* arg) { bool must_throw = true; - if (env()->jvmti_can_post_exceptions()) { - // Do not try anything fancy if we're notifying the VM on every throw. - // Cf. case Bytecodes::_athrow in parse2.cpp. - uncommon_trap(reason, Deoptimization::Action_none, - (ciKlass*)NULL, (char*)NULL, must_throw); - return; + if (env()->jvmti_can_post_on_exceptions()) { + // check if we must post exception events, take uncommon trap if so + uncommon_trap_if_should_post_on_exceptions(reason, must_throw); + // here if should_post_on_exceptions is false + // continue on with the normal codegen } // If this particular condition has not yet happened at this @@ -752,12 +780,20 @@ // Helper function for enforcing certain bytecodes to reexecute if // deoptimization happens -static bool should_reexecute_implied_by_bytecode(JVMState *jvms) { +static bool should_reexecute_implied_by_bytecode(JVMState *jvms, bool is_anewarray) { ciMethod* cur_method = jvms->method(); int cur_bci = jvms->bci(); if (cur_method != NULL && cur_bci != InvocationEntryBci) { Bytecodes::Code code = cur_method->java_code_at_bci(cur_bci); - return Interpreter::bytecode_should_reexecute(code); + return Interpreter::bytecode_should_reexecute(code) || + is_anewarray && code == Bytecodes::_multianewarray; + // Reexecute _multianewarray bytecode which was replaced with + // sequence of [a]newarray. See Parse::do_multianewarray(). + // + // Note: interpreter should not have it set since this optimization + // is limited by dimensions and guarded by flag so in some cases + // multianewarray() runtime calls will be generated and + // the bytecode should not be reexecutes (stack will not be reset). } else return false; } @@ -808,7 +844,7 @@ // For a known set of bytecodes, the interpreter should reexecute them if // deoptimization happens. We set the reexecute state for them here if (out_jvms->is_reexecute_undefined() && //don't change if already specified - should_reexecute_implied_by_bytecode(out_jvms)) { + should_reexecute_implied_by_bytecode(out_jvms, call->is_AllocateArray())) { out_jvms->set_should_reexecute(true); //NOTE: youngest_jvms not changed } @@ -981,14 +1017,19 @@ case Bytecodes::_invokedynamic: case Bytecodes::_invokeinterface: { - bool is_static = (depth == 0); bool ignore; ciBytecodeStream iter(method()); iter.reset_to_bci(bci()); iter.next(); ciMethod* method = iter.get_method(ignore); inputs = method->arg_size_no_receiver(); - if (!is_static) inputs += 1; + // Add a receiver argument, maybe: + if (code != Bytecodes::_invokestatic && + code != Bytecodes::_invokedynamic) + inputs += 1; + // (Do not use ciMethod::arg_size(), because + // it might be an unloaded method, which doesn't + // know whether it is static or not.) int size = method->return_type()->size(); depth = size - inputs; } @@ -1351,8 +1392,8 @@ } //------------------------------set_all_memory_call---------------------------- -void GraphKit::set_all_memory_call(Node* call) { - Node* newmem = _gvn.transform( new (C, 1) ProjNode(call, TypeFunc::Memory) ); +void GraphKit::set_all_memory_call(Node* call, bool separate_io_proj) { + Node* newmem = _gvn.transform( new (C, 1) ProjNode(call, TypeFunc::Memory, separate_io_proj) ); set_all_memory(newmem); } @@ -1573,7 +1614,7 @@ //---------------------------set_edges_for_java_call--------------------------- // Connect a newly created call into the current JVMS. // A return value node (if any) is returned from set_edges_for_java_call. -void GraphKit::set_edges_for_java_call(CallJavaNode* call, bool must_throw) { +void GraphKit::set_edges_for_java_call(CallJavaNode* call, bool must_throw, bool separate_io_proj) { // Add the predefined inputs: call->init_req( TypeFunc::Control, control() ); @@ -1595,13 +1636,13 @@ // Re-use the current map to produce the result. set_control(_gvn.transform(new (C, 1) ProjNode(call, TypeFunc::Control))); - set_i_o( _gvn.transform(new (C, 1) ProjNode(call, TypeFunc::I_O ))); - set_all_memory_call(xcall); + set_i_o( _gvn.transform(new (C, 1) ProjNode(call, TypeFunc::I_O , separate_io_proj))); + set_all_memory_call(xcall, separate_io_proj); //return xcall; // no need, caller already has it } -Node* GraphKit::set_results_for_java_call(CallJavaNode* call) { +Node* GraphKit::set_results_for_java_call(CallJavaNode* call, bool separate_io_proj) { if (stopped()) return top(); // maybe the call folded up? // Capture the return value, if any. @@ -1614,8 +1655,15 @@ // Note: Since any out-of-line call can produce an exception, // we always insert an I_O projection from the call into the result. - make_slow_call_ex(call, env()->Throwable_klass(), false); - + make_slow_call_ex(call, env()->Throwable_klass(), separate_io_proj); + + if (separate_io_proj) { + // The caller requested separate projections be used by the fall + // through and exceptional paths, so replace the projections for + // the fall through path. + set_i_o(_gvn.transform( new (C, 1) ProjNode(call, TypeFunc::I_O) )); + set_all_memory(_gvn.transform( new (C, 1) ProjNode(call, TypeFunc::Memory) )); + } return ret; } @@ -1678,6 +1726,64 @@ } } + +// Replace the call with the current state of the kit. +void GraphKit::replace_call(CallNode* call, Node* result) { + JVMState* ejvms = NULL; + if (has_exceptions()) { + ejvms = transfer_exceptions_into_jvms(); + } + + SafePointNode* final_state = stop(); + + // Find all the needed outputs of this call + CallProjections callprojs; + call->extract_projections(&callprojs, true); + + // Replace all the old call edges with the edges from the inlining result + C->gvn_replace_by(callprojs.fallthrough_catchproj, final_state->in(TypeFunc::Control)); + C->gvn_replace_by(callprojs.fallthrough_memproj, final_state->in(TypeFunc::Memory)); + C->gvn_replace_by(callprojs.fallthrough_ioproj, final_state->in(TypeFunc::I_O)); + + // Replace the result with the new result if it exists and is used + if (callprojs.resproj != NULL && result != NULL) { + C->gvn_replace_by(callprojs.resproj, result); + } + + if (ejvms == NULL) { + // No exception edges to simply kill off those paths + C->gvn_replace_by(callprojs.catchall_catchproj, C->top()); + C->gvn_replace_by(callprojs.catchall_memproj, C->top()); + C->gvn_replace_by(callprojs.catchall_ioproj, C->top()); + + // Replace the old exception object with top + if (callprojs.exobj != NULL) { + C->gvn_replace_by(callprojs.exobj, C->top()); + } + } else { + GraphKit ekit(ejvms); + + // Load my combined exception state into the kit, with all phis transformed: + SafePointNode* ex_map = ekit.combine_and_pop_all_exception_states(); + + Node* ex_oop = ekit.use_exception_state(ex_map); + + C->gvn_replace_by(callprojs.catchall_catchproj, ekit.control()); + C->gvn_replace_by(callprojs.catchall_memproj, ekit.reset_memory()); + C->gvn_replace_by(callprojs.catchall_ioproj, ekit.i_o()); + + // Replace the old exception object with the newly created one + if (callprojs.exobj != NULL) { + C->gvn_replace_by(callprojs.exobj, ex_oop); + } + } + + // Disconnect the call from the graph + call->disconnect_inputs(NULL); + C->gvn_replace_by(call, C->top()); +} + + //------------------------------increment_counter------------------------------ // for statistics: increment a VM counter by 1 @@ -3189,9 +3295,10 @@ if (use_ReduceInitialCardMarks() && obj == just_allocated_object(control())) { // We can skip marks on a freshly-allocated object in Eden. - // Keep this code in sync with maybe_defer_card_mark() in runtime.cpp. - // That routine informs GC to take appropriate compensating steps - // so as to make this card-mark elision safe. + // Keep this code in sync with new_store_pre_barrier() in runtime.cpp. + // That routine informs GC to take appropriate compensating steps, + // upon a slow-path allocation, so as to make this card-mark + // elision safe. return; } @@ -3459,4 +3566,3 @@ sync_kit(ideal); } #undef __ -
--- a/src/share/vm/opto/graphKit.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/graphKit.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2001-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2001-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -251,6 +251,11 @@ // via an uncommon trap. void builtin_throw(Deoptimization::DeoptReason reason, Node* arg = NULL); + // Helper to check the JavaThread::_should_post_on_exceptions flag + // and branch to an uncommon_trap if it is true (with the specified reason and must_throw) + void uncommon_trap_if_should_post_on_exceptions(Deoptimization::DeoptReason reason, + bool must_throw) ; + // Helper Functions for adding debug information void kill_dead_locals(); #ifdef ASSERT @@ -279,6 +284,34 @@ } Node* basic_plus_adr(Node* base, Node* ptr, Node* offset); + + // Some convenient shortcuts for common nodes + Node* IfTrue(IfNode* iff) { return _gvn.transform(new (C,1) IfTrueNode(iff)); } + Node* IfFalse(IfNode* iff) { return _gvn.transform(new (C,1) IfFalseNode(iff)); } + + Node* AddI(Node* l, Node* r) { return _gvn.transform(new (C,3) AddINode(l, r)); } + Node* SubI(Node* l, Node* r) { return _gvn.transform(new (C,3) SubINode(l, r)); } + Node* MulI(Node* l, Node* r) { return _gvn.transform(new (C,3) MulINode(l, r)); } + Node* DivI(Node* ctl, Node* l, Node* r) { return _gvn.transform(new (C,3) DivINode(ctl, l, r)); } + + Node* AndI(Node* l, Node* r) { return _gvn.transform(new (C,3) AndINode(l, r)); } + Node* OrI(Node* l, Node* r) { return _gvn.transform(new (C,3) OrINode(l, r)); } + Node* XorI(Node* l, Node* r) { return _gvn.transform(new (C,3) XorINode(l, r)); } + + Node* MaxI(Node* l, Node* r) { return _gvn.transform(new (C,3) MaxINode(l, r)); } + Node* MinI(Node* l, Node* r) { return _gvn.transform(new (C,3) MinINode(l, r)); } + + Node* LShiftI(Node* l, Node* r) { return _gvn.transform(new (C,3) LShiftINode(l, r)); } + Node* RShiftI(Node* l, Node* r) { return _gvn.transform(new (C,3) RShiftINode(l, r)); } + Node* URShiftI(Node* l, Node* r) { return _gvn.transform(new (C,3) URShiftINode(l, r)); } + + Node* CmpI(Node* l, Node* r) { return _gvn.transform(new (C,3) CmpINode(l, r)); } + Node* CmpL(Node* l, Node* r) { return _gvn.transform(new (C,3) CmpLNode(l, r)); } + Node* CmpP(Node* l, Node* r) { return _gvn.transform(new (C,3) CmpPNode(l, r)); } + Node* Bool(Node* cmp, BoolTest::mask relop) { return _gvn.transform(new (C,2) BoolNode(cmp, relop)); } + + Node* AddP(Node* b, Node* a, Node* o) { return _gvn.transform(new (C,4) AddPNode(b, a, o)); } + // Convert between int and long, and size_t. // (See macros ConvI2X, etc., in type.hpp for ConvI2X, etc.) Node* ConvI2L(Node* offset); @@ -400,7 +433,7 @@ void set_all_memory(Node* newmem); // Create a memory projection from the call, then set_all_memory. - void set_all_memory_call(Node* call); + void set_all_memory_call(Node* call, bool separate_io_proj = false); // Create a LoadNode, reading from the parser's memory state. // (Note: require_atomic_access is useful only with T_LONG.) @@ -543,12 +576,12 @@ // Transform the call, and update the basics: control, i_o, memory. // (The next step is usually to call set_results_for_java_call.) void set_edges_for_java_call(CallJavaNode* call, - bool must_throw = false); + bool must_throw = false, bool separate_io_proj = false); // Finish up a java call that was started by set_edges_for_java_call. // Call add_exception on any throw arising from the call. // Return the call result (transformed). - Node* set_results_for_java_call(CallJavaNode* call); + Node* set_results_for_java_call(CallJavaNode* call, bool separate_io_proj = false); // Similar to set_edges_for_java_call, but simplified for runtime calls. void set_predefined_output_for_runtime_call(Node* call) { @@ -559,6 +592,11 @@ const TypePtr* hook_mem); Node* set_predefined_input_for_runtime_call(SafePointNode* call); + // Replace the call with the current state of the kit. Requires + // that the call was generated with separate io_projs so that + // exceptional control flow can be handled properly. + void replace_call(CallNode* call, Node* result); + // helper functions for statistics void increment_counter(address counter_addr); // increment a debug counter void increment_counter(Node* counter_addr); // increment a debug counter
--- a/src/share/vm/opto/ifg.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/ifg.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1998-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1998-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it
--- a/src/share/vm/opto/ifnode.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/ifnode.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -531,6 +531,9 @@ if (linear_only) return NULL; + if( dom->is_Root() ) + return NULL; + // Else hit a Region. Check for a loop header if( dom->is_Loop() ) return dom->in(1); // Skip up thru loops
--- a/src/share/vm/opto/lcm.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/lcm.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -120,6 +120,7 @@ case Op_LoadRange: case Op_LoadD_unaligned: case Op_LoadL_unaligned: + assert(mach->in(2) == val, "should be address"); break; case Op_StoreB: case Op_StoreC: @@ -146,6 +147,21 @@ default: // Also check for embedded loads if( !mach->needs_anti_dependence_check() ) continue; // Not an memory op; skip it + { + // Check that value is used in memory address. + Node* base; + Node* index; + const MachOper* oper = mach->memory_inputs(base, index); + if (oper == NULL || oper == (MachOper*)-1) { + continue; // Not an memory op; skip it + } + if (val == base || + val == index && val->bottom_type()->isa_narrowoop()) { + break; // Found it + } else { + continue; // Skip it + } + } break; } // check if the offset is not too high for implicit exception @@ -542,6 +558,16 @@ // pointers as far as the kill mask goes. bool exclude_soe = op == Op_CallRuntime; + // If the call is a MethodHandle invoke, we need to exclude the + // register which is used to save the SP value over MH invokes from + // the mask. Otherwise this register could be used for + // deoptimization information. + if (op == Op_CallStaticJava) { + MachCallStaticJavaNode* mcallstaticjava = (MachCallStaticJavaNode*) mcall; + if (mcallstaticjava->_method_handle_invoke) + proj->_rout.OR(Matcher::method_handle_invoke_SP_save_mask()); + } + // Fill in the kill mask for the call for( OptoReg::Name r = OptoReg::Name(0); r < _last_Mach_Reg; r=OptoReg::add(r,1) ) { if( !regs.Member(r) ) { // Not already defined by the call @@ -616,8 +642,9 @@ assert(cfg->_bbs[oop_store->_idx]->_dom_depth <= this->_dom_depth, "oop_store must dominate card-mark"); } } - if( n->is_Mach() && n->as_Mach()->ideal_Opcode() == Op_MemBarAcquire && - n->req() > TypeFunc::Parms ) { + if( n->is_Mach() && n->req() > TypeFunc::Parms && + (n->as_Mach()->ideal_Opcode() == Op_MemBarAcquire || + n->as_Mach()->ideal_Opcode() == Op_MemBarVolatile) ) { // MemBarAcquire could be created without Precedent edge. // del_req() replaces the specified edge with the last input edge // and then removes the last edge. If the specified edge > number of
--- a/src/share/vm/opto/library_call.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/library_call.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -3697,12 +3697,14 @@ // Helper routine for above bool LibraryCallKit::is_method_invoke_or_aux_frame(JVMState* jvms) { + ciMethod* method = jvms->method(); + // Is this the Method.invoke method itself? - if (jvms->method()->intrinsic_id() == vmIntrinsics::_invoke) + if (method->intrinsic_id() == vmIntrinsics::_invoke) return true; // Is this a helper, defined somewhere underneath MethodAccessorImpl. - ciKlass* k = jvms->method()->holder(); + ciKlass* k = method->holder(); if (k->is_instance_klass()) { ciInstanceKlass* ik = k->as_instance_klass(); for (; ik != NULL; ik = ik->super()) { @@ -3712,6 +3714,10 @@ } } } + else if (method->is_method_handle_adapter()) { + // This is an internal adapter frame from the MethodHandleCompiler -- skip it + return true; + } return false; }
--- a/src/share/vm/opto/loopTransform.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/loopTransform.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -549,6 +549,10 @@ // Comparing trip+off vs limit Node *bol = iff->in(1); if( bol->req() != 2 ) continue; // dead constant test + if (!bol->is_Bool()) { + assert(UseLoopPredicate && bol->Opcode() == Op_Conv2B, "predicate check only"); + continue; + } Node *cmp = bol->in(1); Node *rc_exp = cmp->in(1); @@ -875,7 +879,7 @@ //------------------------------is_invariant----------------------------- // Return true if n is invariant bool IdealLoopTree::is_invariant(Node* n) const { - Node *n_c = _phase->get_ctrl(n); + Node *n_c = _phase->has_ctrl(n) ? _phase->get_ctrl(n) : n; if (n_c->is_top()) return false; return !is_member(_phase->get_loop(n_c)); } @@ -1594,7 +1598,7 @@ bool IdealLoopTree::iteration_split_impl( PhaseIdealLoop *phase, Node_List &old_new ) { // Check and remove empty loops (spam micro-benchmarks) if( policy_do_remove_empty_loop(phase) ) - return true; // Here we removed an empty loop + return true; // Here we removed an empty loop bool should_peel = policy_peeling(phase); // Should we peel? @@ -1688,8 +1692,8 @@ // an even number of trips). If we are peeling, we might enable some RCE // and we'd rather unroll the post-RCE'd loop SO... do not unroll if // peeling. - if( should_unroll && !should_peel ) - phase->do_unroll(this,old_new, true); + if( should_unroll && !should_peel ) + phase->do_unroll(this,old_new, true); // Adjust the pre-loop limits to align the main body // iterations. @@ -1731,9 +1735,9 @@ _allow_optimizations && !tail()->is_top() ) { // Also ignore the occasional dead backedge if (!_has_call) { - if (!iteration_split_impl( phase, old_new )) { - return false; - } + if (!iteration_split_impl( phase, old_new )) { + return false; + } } else if (policy_unswitching(phase)) { phase->do_unswitching(this, old_new); } @@ -1746,3 +1750,576 @@ return false; return true; } + +//-------------------------------is_uncommon_trap_proj---------------------------- +// Return true if proj is the form of "proj->[region->..]call_uct" +bool PhaseIdealLoop::is_uncommon_trap_proj(ProjNode* proj, bool must_reason_predicate) { + int path_limit = 10; + assert(proj, "invalid argument"); + Node* out = proj; + for (int ct = 0; ct < path_limit; ct++) { + out = out->unique_ctrl_out(); + if (out == NULL || out->is_Root() || out->is_Start()) + return false; + if (out->is_CallStaticJava()) { + int req = out->as_CallStaticJava()->uncommon_trap_request(); + if (req != 0) { + Deoptimization::DeoptReason reason = Deoptimization::trap_request_reason(req); + if (!must_reason_predicate || reason == Deoptimization::Reason_predicate){ + return true; + } + } + return false; // don't do further after call + } + } + return false; +} + +//-------------------------------is_uncommon_trap_if_pattern------------------------- +// Return true for "if(test)-> proj -> ... +// | +// V +// other_proj->[region->..]call_uct" +// +// "must_reason_predicate" means the uct reason must be Reason_predicate +bool PhaseIdealLoop::is_uncommon_trap_if_pattern(ProjNode *proj, bool must_reason_predicate) { + Node *in0 = proj->in(0); + if (!in0->is_If()) return false; + IfNode* iff = in0->as_If(); + + // we need "If(Conv2B(Opaque1(...)))" pattern for must_reason_predicate + if (must_reason_predicate) { + if (iff->in(1)->Opcode() != Op_Conv2B || + iff->in(1)->in(1)->Opcode() != Op_Opaque1) { + return false; + } + } + + ProjNode* other_proj = iff->proj_out(1-proj->_con)->as_Proj(); + return is_uncommon_trap_proj(other_proj, must_reason_predicate); +} + +//------------------------------create_new_if_for_predicate------------------------ +// create a new if above the uct_if_pattern for the predicate to be promoted. +// +// before after +// ---------- ---------- +// ctrl ctrl +// | | +// | | +// v v +// iff new_iff +// / \ / \ +// / \ / \ +// v v v v +// uncommon_proj cont_proj if_uct if_cont +// \ | | | | +// \ | | | | +// v v v | v +// rgn loop | iff +// | | / \ +// | | / \ +// v | v v +// uncommon_trap | uncommon_proj cont_proj +// \ \ | | +// \ \ | | +// v v v v +// rgn loop +// | +// | +// v +// uncommon_trap +// +// +// We will create a region to guard the uct call if there is no one there. +// The true projecttion (if_cont) of the new_iff is returned. +ProjNode* PhaseIdealLoop::create_new_if_for_predicate(ProjNode* cont_proj) { + assert(is_uncommon_trap_if_pattern(cont_proj, true), "must be a uct if pattern!"); + IfNode* iff = cont_proj->in(0)->as_If(); + + ProjNode *uncommon_proj = iff->proj_out(1 - cont_proj->_con); + Node *rgn = uncommon_proj->unique_ctrl_out(); + assert(rgn->is_Region() || rgn->is_Call(), "must be a region or call uct"); + + if (!rgn->is_Region()) { // create a region to guard the call + assert(rgn->is_Call(), "must be call uct"); + CallNode* call = rgn->as_Call(); + rgn = new (C, 1) RegionNode(1); + _igvn.set_type(rgn, rgn->bottom_type()); + rgn->add_req(uncommon_proj); + set_idom(rgn, idom(uncommon_proj), dom_depth(uncommon_proj)+1); + _igvn.hash_delete(call); + call->set_req(0, rgn); + } + + // Create new_iff + uint iffdd = dom_depth(iff); + IdealLoopTree* lp = get_loop(iff); + IfNode *new_iff = new (C, 2) IfNode(iff->in(0), NULL, iff->_prob, iff->_fcnt); + register_node(new_iff, lp, idom(iff), iffdd); + Node *if_cont = new (C, 1) IfTrueNode(new_iff); + Node *if_uct = new (C, 1) IfFalseNode(new_iff); + if (cont_proj->is_IfFalse()) { + // Swap + Node* tmp = if_uct; if_uct = if_cont; if_cont = tmp; + } + register_node(if_cont, lp, new_iff, iffdd); + register_node(if_uct, get_loop(rgn), new_iff, iffdd); + + // if_cont to iff + _igvn.hash_delete(iff); + iff->set_req(0, if_cont); + set_idom(iff, if_cont, dom_depth(iff)); + + // if_uct to rgn + _igvn.hash_delete(rgn); + rgn->add_req(if_uct); + Node* ridom = idom(rgn); + Node* nrdom = dom_lca(ridom, new_iff); + set_idom(rgn, nrdom, dom_depth(rgn)); + + // rgn must have no phis + assert(!rgn->as_Region()->has_phi(), "region must have no phis"); + + return if_cont->as_Proj(); +} + +//------------------------------find_predicate_insertion_point-------------------------- +// Find a good location to insert a predicate +ProjNode* PhaseIdealLoop::find_predicate_insertion_point(Node* start_c) { + if (start_c == C->root() || !start_c->is_Proj()) + return NULL; + if (is_uncommon_trap_if_pattern(start_c->as_Proj(), true/*Reason_Predicate*/)) { + return start_c->as_Proj(); + } + return NULL; +} + +//------------------------------Invariance----------------------------------- +// Helper class for loop_predication_impl to compute invariance on the fly and +// clone invariants. +class Invariance : public StackObj { + VectorSet _visited, _invariant; + Node_Stack _stack; + VectorSet _clone_visited; + Node_List _old_new; // map of old to new (clone) + IdealLoopTree* _lpt; + PhaseIdealLoop* _phase; + + // Helper function to set up the invariance for invariance computation + // If n is a known invariant, set up directly. Otherwise, look up the + // the possibility to push n onto the stack for further processing. + void visit(Node* use, Node* n) { + if (_lpt->is_invariant(n)) { // known invariant + _invariant.set(n->_idx); + } else if (!n->is_CFG()) { + Node *n_ctrl = _phase->ctrl_or_self(n); + Node *u_ctrl = _phase->ctrl_or_self(use); // self if use is a CFG + if (_phase->is_dominator(n_ctrl, u_ctrl)) { + _stack.push(n, n->in(0) == NULL ? 1 : 0); + } + } + } + + // Compute invariance for "the_node" and (possibly) all its inputs recursively + // on the fly + void compute_invariance(Node* n) { + assert(_visited.test(n->_idx), "must be"); + visit(n, n); + while (_stack.is_nonempty()) { + Node* n = _stack.node(); + uint idx = _stack.index(); + if (idx == n->req()) { // all inputs are processed + _stack.pop(); + // n is invariant if it's inputs are all invariant + bool all_inputs_invariant = true; + for (uint i = 0; i < n->req(); i++) { + Node* in = n->in(i); + if (in == NULL) continue; + assert(_visited.test(in->_idx), "must have visited input"); + if (!_invariant.test(in->_idx)) { // bad guy + all_inputs_invariant = false; + break; + } + } + if (all_inputs_invariant) { + _invariant.set(n->_idx); // I am a invariant too + } + } else { // process next input + _stack.set_index(idx + 1); + Node* m = n->in(idx); + if (m != NULL && !_visited.test_set(m->_idx)) { + visit(n, m); + } + } + } + } + + // Helper function to set up _old_new map for clone_nodes. + // If n is a known invariant, set up directly ("clone" of n == n). + // Otherwise, push n onto the stack for real cloning. + void clone_visit(Node* n) { + assert(_invariant.test(n->_idx), "must be invariant"); + if (_lpt->is_invariant(n)) { // known invariant + _old_new.map(n->_idx, n); + } else{ // to be cloned + assert (!n->is_CFG(), "should not see CFG here"); + _stack.push(n, n->in(0) == NULL ? 1 : 0); + } + } + + // Clone "n" and (possibly) all its inputs recursively + void clone_nodes(Node* n, Node* ctrl) { + clone_visit(n); + while (_stack.is_nonempty()) { + Node* n = _stack.node(); + uint idx = _stack.index(); + if (idx == n->req()) { // all inputs processed, clone n! + _stack.pop(); + // clone invariant node + Node* n_cl = n->clone(); + _old_new.map(n->_idx, n_cl); + _phase->register_new_node(n_cl, ctrl); + for (uint i = 0; i < n->req(); i++) { + Node* in = n_cl->in(i); + if (in == NULL) continue; + n_cl->set_req(i, _old_new[in->_idx]); + } + } else { // process next input + _stack.set_index(idx + 1); + Node* m = n->in(idx); + if (m != NULL && !_clone_visited.test_set(m->_idx)) { + clone_visit(m); // visit the input + } + } + } + } + + public: + Invariance(Arena* area, IdealLoopTree* lpt) : + _lpt(lpt), _phase(lpt->_phase), + _visited(area), _invariant(area), _stack(area, 10 /* guess */), + _clone_visited(area), _old_new(area) + {} + + // Map old to n for invariance computation and clone + void map_ctrl(Node* old, Node* n) { + assert(old->is_CFG() && n->is_CFG(), "must be"); + _old_new.map(old->_idx, n); // "clone" of old is n + _invariant.set(old->_idx); // old is invariant + _clone_visited.set(old->_idx); + } + + // Driver function to compute invariance + bool is_invariant(Node* n) { + if (!_visited.test_set(n->_idx)) + compute_invariance(n); + return (_invariant.test(n->_idx) != 0); + } + + // Driver function to clone invariant + Node* clone(Node* n, Node* ctrl) { + assert(ctrl->is_CFG(), "must be"); + assert(_invariant.test(n->_idx), "must be an invariant"); + if (!_clone_visited.test(n->_idx)) + clone_nodes(n, ctrl); + return _old_new[n->_idx]; + } +}; + +//------------------------------is_range_check_if ----------------------------------- +// Returns true if the predicate of iff is in "scale*iv + offset u< load_range(ptr)" format +// Note: this function is particularly designed for loop predication. We require load_range +// and offset to be loop invariant computed on the fly by "invar" +bool IdealLoopTree::is_range_check_if(IfNode *iff, PhaseIdealLoop *phase, Invariance& invar) const { + if (!is_loop_exit(iff)) { + return false; + } + if (!iff->in(1)->is_Bool()) { + return false; + } + const BoolNode *bol = iff->in(1)->as_Bool(); + if (bol->_test._test != BoolTest::lt) { + return false; + } + if (!bol->in(1)->is_Cmp()) { + return false; + } + const CmpNode *cmp = bol->in(1)->as_Cmp(); + if (cmp->Opcode() != Op_CmpU ) { + return false; + } + if (cmp->in(2)->Opcode() != Op_LoadRange) { + return false; + } + LoadRangeNode* lr = (LoadRangeNode*)cmp->in(2); + if (!invar.is_invariant(lr)) { // loadRange must be invariant + return false; + } + Node *iv = _head->as_CountedLoop()->phi(); + int scale = 0; + Node *offset = NULL; + if (!phase->is_scaled_iv_plus_offset(cmp->in(1), iv, &scale, &offset)) { + return false; + } + if(offset && !invar.is_invariant(offset)) { // offset must be invariant + return false; + } + return true; +} + +//------------------------------rc_predicate----------------------------------- +// Create a range check predicate +// +// for (i = init; i < limit; i += stride) { +// a[scale*i+offset] +// } +// +// Compute max(scale*i + offset) for init <= i < limit and build the predicate +// as "max(scale*i + offset) u< a.length". +// +// There are two cases for max(scale*i + offset): +// (1) stride*scale > 0 +// max(scale*i + offset) = scale*(limit-stride) + offset +// (2) stride*scale < 0 +// max(scale*i + offset) = scale*init + offset +BoolNode* PhaseIdealLoop::rc_predicate(Node* ctrl, + int scale, Node* offset, + Node* init, Node* limit, Node* stride, + Node* range) { + Node* max_idx_expr = init; + int stride_con = stride->get_int(); + if ((stride_con > 0) == (scale > 0)) { + max_idx_expr = new (C, 3) SubINode(limit, stride); + register_new_node(max_idx_expr, ctrl); + } + + if (scale != 1) { + ConNode* con_scale = _igvn.intcon(scale); + max_idx_expr = new (C, 3) MulINode(max_idx_expr, con_scale); + register_new_node(max_idx_expr, ctrl); + } + + if (offset && (!offset->is_Con() || offset->get_int() != 0)){ + max_idx_expr = new (C, 3) AddINode(max_idx_expr, offset); + register_new_node(max_idx_expr, ctrl); + } + + CmpUNode* cmp = new (C, 3) CmpUNode(max_idx_expr, range); + register_new_node(cmp, ctrl); + BoolNode* bol = new (C, 2) BoolNode(cmp, BoolTest::lt); + register_new_node(bol, ctrl); + return bol; +} + +//------------------------------ loop_predication_impl-------------------------- +// Insert loop predicates for null checks and range checks +bool PhaseIdealLoop::loop_predication_impl(IdealLoopTree *loop) { + if (!UseLoopPredicate) return false; + + // Too many traps seen? + bool tmt = C->too_many_traps(C->method(), 0, Deoptimization::Reason_predicate); + int tc = C->trap_count(Deoptimization::Reason_predicate); + if (tmt || tc > 0) { + if (TraceLoopPredicate) { + tty->print_cr("too many predicate traps: %d", tc); + C->method()->print(); // which method has too many predicate traps + tty->print_cr(""); + } + return false; + } + + CountedLoopNode *cl = NULL; + if (loop->_head->is_CountedLoop()) { + cl = loop->_head->as_CountedLoop(); + // do nothing for iteration-splitted loops + if(!cl->is_normal_loop()) return false; + } + + LoopNode *lpn = loop->_head->as_Loop(); + Node* entry = lpn->in(LoopNode::EntryControl); + + ProjNode *predicate_proj = find_predicate_insertion_point(entry); + if (!predicate_proj){ +#ifndef PRODUCT + if (TraceLoopPredicate) { + tty->print("missing predicate:"); + loop->dump_head(); + } +#endif + return false; + } + + ConNode* zero = _igvn.intcon(0); + set_ctrl(zero, C->root()); + Node *cond_false = new (C, 2) Conv2BNode(zero); + register_new_node(cond_false, C->root()); + ConNode* one = _igvn.intcon(1); + set_ctrl(one, C->root()); + Node *cond_true = new (C, 2) Conv2BNode(one); + register_new_node(cond_true, C->root()); + + ResourceArea *area = Thread::current()->resource_area(); + Invariance invar(area, loop); + + // Create list of if-projs such that a newer proj dominates all older + // projs in the list, and they all dominate loop->tail() + Node_List if_proj_list(area); + LoopNode *head = loop->_head->as_Loop(); + Node *current_proj = loop->tail(); //start from tail + while ( current_proj != head ) { + if (loop == get_loop(current_proj) && // still in the loop ? + current_proj->is_Proj() && // is a projection ? + current_proj->in(0)->Opcode() == Op_If) { // is a if projection ? + if_proj_list.push(current_proj); + } + current_proj = idom(current_proj); + } + + bool hoisted = false; // true if at least one proj is promoted + while (if_proj_list.size() > 0) { + // Following are changed to nonnull when a predicate can be hoisted + ProjNode* new_predicate_proj = NULL; + BoolNode* new_predicate_bol = NULL; + + ProjNode* proj = if_proj_list.pop()->as_Proj(); + IfNode* iff = proj->in(0)->as_If(); + + if (!is_uncommon_trap_if_pattern(proj)) { + if (loop->is_loop_exit(iff)) { + // stop processing the remaining projs in the list because the execution of them + // depends on the condition of "iff" (iff->in(1)). + break; + } else { + // Both arms are inside the loop. There are two cases: + // (1) there is one backward branch. In this case, any remaining proj + // in the if_proj list post-dominates "iff". So, the condition of "iff" + // does not determine the execution the remining projs directly, and we + // can safely continue. + // (2) both arms are forwarded, i.e. a diamond shape. In this case, "proj" + // does not dominate loop->tail(), so it can not be in the if_proj list. + continue; + } + } + + Node* test = iff->in(1); + if (!test->is_Bool()){ //Conv2B, ... + continue; + } + BoolNode* bol = test->as_Bool(); + if (invar.is_invariant(bol)) { + // Invariant test + new_predicate_proj = create_new_if_for_predicate(predicate_proj); + Node* ctrl = new_predicate_proj->in(0)->as_If()->in(0); + new_predicate_bol = invar.clone(bol, ctrl)->as_Bool(); + if (TraceLoopPredicate) tty->print("invariant"); + } else if (cl != NULL && loop->is_range_check_if(iff, this, invar)) { + // Range check (only for counted loops) + new_predicate_proj = create_new_if_for_predicate(predicate_proj); + Node *ctrl = new_predicate_proj->in(0)->as_If()->in(0); + const Node* cmp = bol->in(1)->as_Cmp(); + Node* idx = cmp->in(1); + assert(!invar.is_invariant(idx), "index is variant"); + assert(cmp->in(2)->Opcode() == Op_LoadRange, "must be"); + LoadRangeNode* ld_rng = (LoadRangeNode*)cmp->in(2); // LoadRangeNode + assert(invar.is_invariant(ld_rng), "load range must be invariant"); + ld_rng = (LoadRangeNode*)invar.clone(ld_rng, ctrl); + int scale = 1; + Node* offset = zero; + bool ok = is_scaled_iv_plus_offset(idx, cl->phi(), &scale, &offset); + assert(ok, "must be index expression"); + if (offset && offset != zero) { + assert(invar.is_invariant(offset), "offset must be loop invariant"); + offset = invar.clone(offset, ctrl); + } + Node* init = cl->init_trip(); + Node* limit = cl->limit(); + Node* stride = cl->stride(); + new_predicate_bol = rc_predicate(ctrl, scale, offset, init, limit, stride, ld_rng); + if (TraceLoopPredicate) tty->print("range check"); + } + + if (new_predicate_proj == NULL) { + // The other proj of the "iff" is a uncommon trap projection, and we can assume + // the other proj will not be executed ("executed" means uct raised). + continue; + } else { + // Success - attach condition (new_predicate_bol) to predicate if + invar.map_ctrl(proj, new_predicate_proj); // so that invariance test can be appropriate + IfNode* new_iff = new_predicate_proj->in(0)->as_If(); + + // Negate test if necessary + if (proj->_con != predicate_proj->_con) { + new_predicate_bol = new (C, 2) BoolNode(new_predicate_bol->in(1), new_predicate_bol->_test.negate()); + register_new_node(new_predicate_bol, new_iff->in(0)); + if (TraceLoopPredicate) tty->print_cr(" if negated: %d", iff->_idx); + } else { + if (TraceLoopPredicate) tty->print_cr(" if: %d", iff->_idx); + } + + _igvn.hash_delete(new_iff); + new_iff->set_req(1, new_predicate_bol); + + _igvn.hash_delete(iff); + iff->set_req(1, proj->is_IfFalse() ? cond_false : cond_true); + + Node* ctrl = new_predicate_proj; // new control + ProjNode* dp = proj; // old control + assert(get_loop(dp) == loop, "guarenteed at the time of collecting proj"); + // Find nodes (depends only on the test) off the surviving projection; + // move them outside the loop with the control of proj_clone + for (DUIterator_Fast imax, i = dp->fast_outs(imax); i < imax; i++) { + Node* cd = dp->fast_out(i); // Control-dependent node + if (cd->depends_only_on_test()) { + assert(cd->in(0) == dp, ""); + _igvn.hash_delete(cd); + cd->set_req(0, ctrl); // ctrl, not NULL + set_early_ctrl(cd); + _igvn._worklist.push(cd); + IdealLoopTree *new_loop = get_loop(get_ctrl(cd)); + if (new_loop != loop) { + if (!loop->_child) loop->_body.yank(cd); + if (!new_loop->_child ) new_loop->_body.push(cd); + } + --i; + --imax; + } + } + + hoisted = true; + C->set_major_progress(); + } + } // end while + +#ifndef PRODUCT + // report that the loop predication has been actually performed + // for this loop + if (TraceLoopPredicate && hoisted) { + tty->print("Loop Predication Performed:"); + loop->dump_head(); + } +#endif + + return hoisted; +} + +//------------------------------loop_predication-------------------------------- +// driver routine for loop predication optimization +bool IdealLoopTree::loop_predication( PhaseIdealLoop *phase) { + bool hoisted = false; + // Recursively promote predicates + if ( _child ) { + hoisted = _child->loop_predication( phase); + } + + // self + if (!_irreducible && !tail()->is_top()) { + hoisted |= phase->loop_predication_impl(this); + } + + if ( _next ) { //sibling + hoisted |= _next->loop_predication( phase); + } + + return hoisted; +}
--- a/src/share/vm/opto/loopnode.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/loopnode.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1279,7 +1279,8 @@ // Visit all children, looking for Phis for (DUIterator i = cl->outs(); cl->has_out(i); i++) { Node *out = cl->out(i); - if (!out->is_Phi() || out == phi) continue; // Looking for other phis + // Look for other phis (secondary IVs). Skip dead ones + if (!out->is_Phi() || out == phi || !phase->has_node(out)) continue; PhiNode* phi2 = out->as_Phi(); Node *incr2 = phi2->in( LoopNode::LoopBackControl ); // Look for induction variables of the form: X += constant @@ -1419,11 +1420,57 @@ } } +//---------------------collect_potentially_useful_predicates----------------------- +// Helper function to collect potentially useful predicates to prevent them from +// being eliminated by PhaseIdealLoop::eliminate_useless_predicates +void PhaseIdealLoop::collect_potentially_useful_predicates( + IdealLoopTree * loop, Unique_Node_List &useful_predicates) { + if (loop->_child) { // child + collect_potentially_useful_predicates(loop->_child, useful_predicates); + } + + // self (only loops that we can apply loop predication may use their predicates) + if (loop->_head->is_Loop() && + !loop->_irreducible && + !loop->tail()->is_top()) { + LoopNode *lpn = loop->_head->as_Loop(); + Node* entry = lpn->in(LoopNode::EntryControl); + ProjNode *predicate_proj = find_predicate_insertion_point(entry); + if (predicate_proj != NULL ) { // right pattern that can be used by loop predication + assert(entry->in(0)->in(1)->in(1)->Opcode()==Op_Opaque1, "must be"); + useful_predicates.push(entry->in(0)->in(1)->in(1)); // good one + } + } + + if ( loop->_next ) { // sibling + collect_potentially_useful_predicates(loop->_next, useful_predicates); + } +} + +//------------------------eliminate_useless_predicates----------------------------- +// Eliminate all inserted predicates if they could not be used by loop predication. +void PhaseIdealLoop::eliminate_useless_predicates() { + if (C->predicate_count() == 0) return; // no predicate left + + Unique_Node_List useful_predicates; // to store useful predicates + if (C->has_loops()) { + collect_potentially_useful_predicates(_ltree_root->_child, useful_predicates); + } + + for (int i = C->predicate_count(); i > 0; i--) { + Node * n = C->predicate_opaque1_node(i-1); + assert(n->Opcode() == Op_Opaque1, "must be"); + if (!useful_predicates.member(n)) { // not in the useful list + _igvn.replace_node(n, n->in(1)); + } + } +} + //============================================================================= //----------------------------build_and_optimize------------------------------- // Create a PhaseLoop. Build the ideal Loop tree. Map each Ideal Node to // its corresponding LoopNode. If 'optimize' is true, do some loop cleanups. -void PhaseIdealLoop::build_and_optimize(bool do_split_ifs) { +void PhaseIdealLoop::build_and_optimize(bool do_split_ifs, bool do_loop_pred) { int old_progress = C->major_progress(); // Reset major-progress flag for the driver's heuristics @@ -1576,6 +1623,12 @@ return; } + // some parser-inserted loop predicates could never be used by loop + // predication. Eliminate them before loop optimization + if (UseLoopPredicate) { + eliminate_useless_predicates(); + } + // clear out the dead code while(_deadlist.size()) { _igvn.remove_globally_dead_node(_deadlist.pop()); @@ -1602,7 +1655,7 @@ // Because RCE opportunities can be masked by split_thru_phi, // look for RCE candidates and inhibit split_thru_phi // on just their loop-phi's for this pass of loop opts - if( SplitIfBlocks && do_split_ifs ) { + if (SplitIfBlocks && do_split_ifs) { if (lpt->policy_range_check(this)) { lpt->_rce_candidate = 1; // = true } @@ -1618,12 +1671,17 @@ NOT_PRODUCT( if( VerifyLoopOptimizations ) verify(); ); } + // Perform loop predication before iteration splitting + if (do_loop_pred && C->has_loops() && !C->major_progress()) { + _ltree_root->_child->loop_predication(this); + } + // Perform iteration-splitting on inner loops. Split iterations to avoid // range checks or one-shot null checks. // If split-if's didn't hack the graph too bad (no CFG changes) // then do loop opts. - if( C->has_loops() && !C->major_progress() ) { + if (C->has_loops() && !C->major_progress()) { memset( worklist.adr(), 0, worklist.Size()*sizeof(Node*) ); _ltree_root->_child->iteration_split( this, worklist ); // No verify after peeling! GCM has hoisted code out of the loop. @@ -1635,7 +1693,7 @@ // Do verify graph edges in any case NOT_PRODUCT( C->verify_graph_edges(); ); - if( !do_split_ifs ) { + if (!do_split_ifs) { // We saw major progress in Split-If to get here. We forced a // pass with unrolling and not split-if, however more split-if's // might make progress. If the unrolling didn't make progress @@ -2762,6 +2820,22 @@ Node *legal = LCA; // Walk 'legal' up the IDOM chain Node *least = legal; // Best legal position so far while( early != legal ) { // While not at earliest legal +#ifdef ASSERT + if (legal->is_Start() && !early->is_Root()) { + // Bad graph. Print idom path and fail. + tty->print_cr( "Bad graph detected in build_loop_late"); + tty->print("n: ");n->dump(); tty->cr(); + tty->print("early: ");early->dump(); tty->cr(); + int ct = 0; + Node *dbg_legal = LCA; + while(!dbg_legal->is_Start() && ct < 100) { + tty->print("idom[%d] ",ct); dbg_legal->dump(); tty->cr(); + ct++; + dbg_legal = idom(dbg_legal); + } + assert(false, "Bad graph detected in build_loop_late"); + } +#endif // Find least loop nesting depth legal = idom(legal); // Bump up the IDOM tree // Check for lower nesting depth
--- a/src/share/vm/opto/loopnode.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/loopnode.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -30,6 +30,7 @@ class Node; class PhaseIdealLoop; class VectorSet; +class Invariance; struct small_cache; // @@ -325,6 +326,10 @@ // Returns TRUE if loop tree is structurally changed. bool beautify_loops( PhaseIdealLoop *phase ); + // Perform optimization to use the loop predicates for null checks and range checks. + // Applies to any loop level (not just the innermost one) + bool loop_predication( PhaseIdealLoop *phase); + // Perform iteration-splitting on inner loops. Split iterations to // avoid range checks or one-shot null checks. Returns false if the // current round of loop opts should stop. @@ -395,6 +400,9 @@ // into longer memory ops, we may want to increase alignment. bool policy_align( PhaseIdealLoop *phase ) const; + // Return TRUE if "iff" is a range check. + bool is_range_check_if(IfNode *iff, PhaseIdealLoop *phase, Invariance& invar) const; + // Compute loop trip count from profile data void compute_profile_trip_cnt( PhaseIdealLoop *phase ); @@ -521,9 +529,6 @@ } Node *dom_lca_for_get_late_ctrl_internal( Node *lca, Node *n, Node *tag ); - // true if CFG node d dominates CFG node n - bool is_dominator(Node *d, Node *n); - // Helper function for directing control inputs away from CFG split // points. Node *find_non_split_ctrl( Node *ctrl ) const { @@ -572,6 +577,17 @@ assert(n == find_non_split_ctrl(n), "must return legal ctrl" ); return n; } + // true if CFG node d dominates CFG node n + bool is_dominator(Node *d, Node *n); + // return get_ctrl for a data node and self(n) for a CFG node + Node* ctrl_or_self(Node* n) { + if (has_ctrl(n)) + return get_ctrl(n); + else { + assert (n->is_CFG(), "must be a CFG node"); + return n; + } + } private: Node *get_ctrl_no_update( Node *i ) const { @@ -600,7 +616,7 @@ // Lazy-dazy update of 'get_ctrl' and 'idom_at' mechanisms. Replace // the 'old_node' with 'new_node'. Kill old-node. Add a reference // from old_node to new_node to support the lazy update. Reference - // replaces loop reference, since that is not neede for dead node. + // replaces loop reference, since that is not needed for dead node. public: void lazy_update( Node *old_node, Node *new_node ) { assert( old_node != new_node, "no cycles please" ); @@ -679,11 +695,11 @@ _dom_lca_tags(C->comp_arena()), _verify_me(NULL), _verify_only(true) { - build_and_optimize(false); + build_and_optimize(false, false); } // build the loop tree and perform any requested optimizations - void build_and_optimize(bool do_split_if); + void build_and_optimize(bool do_split_if, bool do_loop_pred); public: // Dominators for the sea of nodes @@ -694,13 +710,13 @@ Node *dom_lca_internal( Node *n1, Node *n2 ) const; // Compute the Ideal Node to Loop mapping - PhaseIdealLoop( PhaseIterGVN &igvn, bool do_split_ifs) : + PhaseIdealLoop( PhaseIterGVN &igvn, bool do_split_ifs, bool do_loop_pred) : PhaseTransform(Ideal_Loop), _igvn(igvn), _dom_lca_tags(C->comp_arena()), _verify_me(NULL), _verify_only(false) { - build_and_optimize(do_split_ifs); + build_and_optimize(do_split_ifs, do_loop_pred); } // Verify that verify_me made the same decisions as a fresh run. @@ -710,7 +726,7 @@ _dom_lca_tags(C->comp_arena()), _verify_me(verify_me), _verify_only(false) { - build_and_optimize(false); + build_and_optimize(false, false); } // Build and verify the loop tree without modifying the graph. This @@ -790,6 +806,30 @@ // Return true if exp is a scaled induction var plus (or minus) constant bool is_scaled_iv_plus_offset(Node* exp, Node* iv, int* p_scale, Node** p_offset, int depth = 0); + // Return true if proj is for "proj->[region->..]call_uct" + bool is_uncommon_trap_proj(ProjNode* proj, bool must_reason_predicate = false); + // Return true for "if(test)-> proj -> ... + // | + // V + // other_proj->[region->..]call_uct" + bool is_uncommon_trap_if_pattern(ProjNode* proj, bool must_reason_predicate = false); + // Create a new if above the uncommon_trap_if_pattern for the predicate to be promoted + ProjNode* create_new_if_for_predicate(ProjNode* cont_proj); + // Find a good location to insert a predicate + ProjNode* find_predicate_insertion_point(Node* start_c); + // Construct a range check for a predicate if + BoolNode* rc_predicate(Node* ctrl, + int scale, Node* offset, + Node* init, Node* limit, Node* stride, + Node* range); + + // Implementation of the loop predication to promote checks outside the loop + bool loop_predication_impl(IdealLoopTree *loop); + + // Helper function to collect predicate for eliminating the useless ones + void collect_potentially_useful_predicates(IdealLoopTree *loop, Unique_Node_List &predicate_opaque1); + void eliminate_useless_predicates(); + // Eliminate range-checks and other trip-counter vs loop-invariant tests. void do_range_check( IdealLoopTree *loop, Node_List &old_new ); @@ -906,7 +946,6 @@ const TypeInt* filtered_type_from_dominators( Node* val, Node *val_ctrl); // Helper functions - void register_new_node( Node *n, Node *blk ); Node *spinup( Node *iff, Node *new_false, Node *new_true, Node *region, Node *phi, small_cache *cache ); Node *find_use_block( Node *use, Node *def, Node *old_false, Node *new_false, Node *old_true, Node *new_true ); void handle_use( Node *use, Node *def, small_cache *cache, Node *region_dom, Node *new_false, Node *new_true, Node *old_false, Node *old_true ); @@ -918,6 +957,7 @@ public: void set_created_loop_node() { _created_loop_node = true; } bool created_loop_node() { return _created_loop_node; } + void register_new_node( Node *n, Node *blk ); #ifndef PRODUCT void dump( ) const;
--- a/src/share/vm/opto/loopopts.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/loopopts.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1999-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -47,7 +47,7 @@ int offset = t_oop->offset(); phi = new (C,region->req()) PhiNode(region, type, NULL, iid, index, offset); } else { - phi = new (C,region->req()) PhiNode(region, type); + phi = PhiNode::make_blank(region, n); } uint old_unique = C->unique(); for( uint i = 1; i < region->req(); i++ ) {
--- a/src/share/vm/opto/machnode.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/machnode.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -636,7 +636,9 @@ } #ifndef PRODUCT void MachCallJavaNode::dump_spec(outputStream *st) const { - if( _method ) { + if (_method_handle_invoke) + st->print("MethodHandle "); + if (_method) { _method->print_short_name(st); st->print(" "); } @@ -644,6 +646,20 @@ } #endif +//------------------------------Registers-------------------------------------- +const RegMask &MachCallJavaNode::in_RegMask(uint idx) const { + // Values in the domain use the users calling convention, embodied in the + // _in_rms array of RegMasks. + if (idx < tf()->domain()->cnt()) return _in_rms[idx]; + // Values outside the domain represent debug info + Matcher* m = Compile::current()->matcher(); + // If this call is a MethodHandle invoke we have to use a different + // debugmask which does not include the register we use to save the + // SP over MH invokes. + RegMask** debugmask = _method_handle_invoke ? m->idealreg2mhdebugmask : m->idealreg2debugmask; + return *debugmask[in(idx)->ideal_reg()]; +} + //============================================================================= uint MachCallStaticJavaNode::size_of() const { return sizeof(*this); } uint MachCallStaticJavaNode::cmp( const Node &n ) const {
--- a/src/share/vm/opto/machnode.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/machnode.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2007 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -232,7 +232,7 @@ // Expand method for MachNode, replaces nodes representing pseudo // instructions with a set of nodes which represent real machine // instructions and compute the same value. - virtual MachNode *Expand( State *, Node_List &proj_list ) { return this; } + virtual MachNode *Expand( State *, Node_List &proj_list, Node* mem ) { return this; } // Bottom_type call; value comes from operand0 virtual const class Type *bottom_type() const { return _opnds[0]->type(); } @@ -662,9 +662,13 @@ ciMethod* _method; // Method being direct called int _bci; // Byte Code index of call byte code bool _optimized_virtual; // Tells if node is a static call or an optimized virtual + bool _method_handle_invoke; // Tells if the call has to preserve SP MachCallJavaNode() : MachCallNode() { init_class_id(Class_MachCallJava); } + + virtual const RegMask &in_RegMask(uint) const; + #ifndef PRODUCT virtual void dump_spec(outputStream *st) const; #endif
--- a/src/share/vm/opto/macro.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/macro.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -316,6 +316,21 @@ assert(adr_idx == Compile::AliasIdxRaw, "address must match or be raw"); } mem = mem->in(MemNode::Memory); + } else if (mem->is_ClearArray()) { + if (!ClearArrayNode::step_through(&mem, alloc->_idx, phase)) { + // Can not bypass initialization of the instance + // we are looking. + debug_only(intptr_t offset;) + assert(alloc == AllocateNode::Ideal_allocation(mem->in(3), phase, offset), "sanity"); + InitializeNode* init = alloc->as_Allocate()->initialization(); + // We are looking for stored value, return Initialize node + // or memory edge from Allocate node. + if (init != NULL) + return init; + else + return alloc->in(TypeFunc::Memory); // It will produce zero value (see callers). + } + // Otherwise skip it (the call updated 'mem' value). } else if (mem->Opcode() == Op_SCMemProj) { assert(mem->in(0)->is_LoadStore(), "sanity"); const TypePtr* atype = mem->in(0)->in(MemNode::Address)->bottom_type()->is_ptr(); @@ -823,6 +838,18 @@ Node *n = use->last_out(k); uint oc2 = use->outcnt(); if (n->is_Store()) { +#ifdef ASSERT + // Verify that there is no dependent MemBarVolatile nodes, + // they should be removed during IGVN, see MemBarNode::Ideal(). + for (DUIterator_Fast pmax, p = n->fast_outs(pmax); + p < pmax; p++) { + Node* mb = n->fast_out(p); + assert(mb->is_Initialize() || !mb->is_MemBar() || + mb->req() <= MemBarNode::Precedent || + mb->in(MemBarNode::Precedent) != n, + "MemBarVolatile should be eliminated for non-escaping object"); + } +#endif _igvn.replace_node(n, n->in(MemNode::Memory)); } else { eliminate_card_mark(n); @@ -912,15 +939,29 @@ return false; } + CompileLog* log = C->log(); + if (log != NULL) { + Node* klass = alloc->in(AllocateNode::KlassNode); + const TypeKlassPtr* tklass = _igvn.type(klass)->is_klassptr(); + log->head("eliminate_allocation type='%d'", + log->identify(tklass->klass())); + JVMState* p = alloc->jvms(); + while (p != NULL) { + log->elem("jvms bci='%d' method='%d'", p->bci(), log->identify(p->method())); + p = p->caller(); + } + log->tail("eliminate_allocation"); + } + process_users_of_allocation(alloc); #ifndef PRODUCT -if (PrintEliminateAllocations) { - if (alloc->is_AllocateArray()) - tty->print_cr("++++ Eliminated: %d AllocateArray", alloc->_idx); - else - tty->print_cr("++++ Eliminated: %d Allocate", alloc->_idx); -} + if (PrintEliminateAllocations) { + if (alloc->is_AllocateArray()) + tty->print_cr("++++ Eliminated: %d AllocateArray", alloc->_idx); + else + tty->print_cr("++++ Eliminated: %d Allocate", alloc->_idx); + } #endif return true; @@ -1639,6 +1680,18 @@ } // if (!oldbox->is_eliminated()) } // if (alock->is_Lock() && !lock->is_coarsened()) + CompileLog* log = C->log(); + if (log != NULL) { + log->head("eliminate_lock lock='%d'", + alock->is_Lock()); + JVMState* p = alock->jvms(); + while (p != NULL) { + log->elem("jvms bci='%d' method='%d'", p->bci(), log->identify(p->method())); + p = p->caller(); + } + log->tail("eliminate_lock"); + } + #ifndef PRODUCT if (PrintEliminateLocks) { if (alock->is_Lock()) {
--- a/src/share/vm/opto/matcher.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/matcher.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -70,19 +70,27 @@ _dontcare(&_states_arena) { C->set_matcher(this); - idealreg2spillmask[Op_RegI] = NULL; - idealreg2spillmask[Op_RegN] = NULL; - idealreg2spillmask[Op_RegL] = NULL; - idealreg2spillmask[Op_RegF] = NULL; - idealreg2spillmask[Op_RegD] = NULL; - idealreg2spillmask[Op_RegP] = NULL; + idealreg2spillmask [Op_RegI] = NULL; + idealreg2spillmask [Op_RegN] = NULL; + idealreg2spillmask [Op_RegL] = NULL; + idealreg2spillmask [Op_RegF] = NULL; + idealreg2spillmask [Op_RegD] = NULL; + idealreg2spillmask [Op_RegP] = NULL; - idealreg2debugmask[Op_RegI] = NULL; - idealreg2debugmask[Op_RegN] = NULL; - idealreg2debugmask[Op_RegL] = NULL; - idealreg2debugmask[Op_RegF] = NULL; - idealreg2debugmask[Op_RegD] = NULL; - idealreg2debugmask[Op_RegP] = NULL; + idealreg2debugmask [Op_RegI] = NULL; + idealreg2debugmask [Op_RegN] = NULL; + idealreg2debugmask [Op_RegL] = NULL; + idealreg2debugmask [Op_RegF] = NULL; + idealreg2debugmask [Op_RegD] = NULL; + idealreg2debugmask [Op_RegP] = NULL; + + idealreg2mhdebugmask[Op_RegI] = NULL; + idealreg2mhdebugmask[Op_RegN] = NULL; + idealreg2mhdebugmask[Op_RegL] = NULL; + idealreg2mhdebugmask[Op_RegF] = NULL; + idealreg2mhdebugmask[Op_RegD] = NULL; + idealreg2mhdebugmask[Op_RegP] = NULL; + debug_only(_mem_node = NULL;) // Ideal memory node consumed by mach node } @@ -389,19 +397,28 @@ void Matcher::init_first_stack_mask() { // Allocate storage for spill masks as masks for the appropriate load type. - RegMask *rms = (RegMask*)C->comp_arena()->Amalloc_D(sizeof(RegMask)*12); - idealreg2spillmask[Op_RegN] = &rms[0]; - idealreg2spillmask[Op_RegI] = &rms[1]; - idealreg2spillmask[Op_RegL] = &rms[2]; - idealreg2spillmask[Op_RegF] = &rms[3]; - idealreg2spillmask[Op_RegD] = &rms[4]; - idealreg2spillmask[Op_RegP] = &rms[5]; - idealreg2debugmask[Op_RegN] = &rms[6]; - idealreg2debugmask[Op_RegI] = &rms[7]; - idealreg2debugmask[Op_RegL] = &rms[8]; - idealreg2debugmask[Op_RegF] = &rms[9]; - idealreg2debugmask[Op_RegD] = &rms[10]; - idealreg2debugmask[Op_RegP] = &rms[11]; + RegMask *rms = (RegMask*)C->comp_arena()->Amalloc_D(sizeof(RegMask) * 3*6); + + idealreg2spillmask [Op_RegN] = &rms[0]; + idealreg2spillmask [Op_RegI] = &rms[1]; + idealreg2spillmask [Op_RegL] = &rms[2]; + idealreg2spillmask [Op_RegF] = &rms[3]; + idealreg2spillmask [Op_RegD] = &rms[4]; + idealreg2spillmask [Op_RegP] = &rms[5]; + + idealreg2debugmask [Op_RegN] = &rms[6]; + idealreg2debugmask [Op_RegI] = &rms[7]; + idealreg2debugmask [Op_RegL] = &rms[8]; + idealreg2debugmask [Op_RegF] = &rms[9]; + idealreg2debugmask [Op_RegD] = &rms[10]; + idealreg2debugmask [Op_RegP] = &rms[11]; + + idealreg2mhdebugmask[Op_RegN] = &rms[12]; + idealreg2mhdebugmask[Op_RegI] = &rms[13]; + idealreg2mhdebugmask[Op_RegL] = &rms[14]; + idealreg2mhdebugmask[Op_RegF] = &rms[15]; + idealreg2mhdebugmask[Op_RegD] = &rms[16]; + idealreg2mhdebugmask[Op_RegP] = &rms[17]; OptoReg::Name i; @@ -442,12 +459,19 @@ // Make up debug masks. Any spill slot plus callee-save registers. // Caller-save registers are assumed to be trashable by the various // inline-cache fixup routines. - *idealreg2debugmask[Op_RegN]= *idealreg2spillmask[Op_RegN]; - *idealreg2debugmask[Op_RegI]= *idealreg2spillmask[Op_RegI]; - *idealreg2debugmask[Op_RegL]= *idealreg2spillmask[Op_RegL]; - *idealreg2debugmask[Op_RegF]= *idealreg2spillmask[Op_RegF]; - *idealreg2debugmask[Op_RegD]= *idealreg2spillmask[Op_RegD]; - *idealreg2debugmask[Op_RegP]= *idealreg2spillmask[Op_RegP]; + *idealreg2debugmask [Op_RegN]= *idealreg2spillmask[Op_RegN]; + *idealreg2debugmask [Op_RegI]= *idealreg2spillmask[Op_RegI]; + *idealreg2debugmask [Op_RegL]= *idealreg2spillmask[Op_RegL]; + *idealreg2debugmask [Op_RegF]= *idealreg2spillmask[Op_RegF]; + *idealreg2debugmask [Op_RegD]= *idealreg2spillmask[Op_RegD]; + *idealreg2debugmask [Op_RegP]= *idealreg2spillmask[Op_RegP]; + + *idealreg2mhdebugmask[Op_RegN]= *idealreg2spillmask[Op_RegN]; + *idealreg2mhdebugmask[Op_RegI]= *idealreg2spillmask[Op_RegI]; + *idealreg2mhdebugmask[Op_RegL]= *idealreg2spillmask[Op_RegL]; + *idealreg2mhdebugmask[Op_RegF]= *idealreg2spillmask[Op_RegF]; + *idealreg2mhdebugmask[Op_RegD]= *idealreg2spillmask[Op_RegD]; + *idealreg2mhdebugmask[Op_RegP]= *idealreg2spillmask[Op_RegP]; // Prevent stub compilations from attempting to reference // callee-saved registers from debug info @@ -458,14 +482,31 @@ if( _register_save_policy[i] == 'C' || _register_save_policy[i] == 'A' || (_register_save_policy[i] == 'E' && exclude_soe) ) { - idealreg2debugmask[Op_RegN]->Remove(i); - idealreg2debugmask[Op_RegI]->Remove(i); // Exclude save-on-call - idealreg2debugmask[Op_RegL]->Remove(i); // registers from debug - idealreg2debugmask[Op_RegF]->Remove(i); // masks - idealreg2debugmask[Op_RegD]->Remove(i); - idealreg2debugmask[Op_RegP]->Remove(i); + idealreg2debugmask [Op_RegN]->Remove(i); + idealreg2debugmask [Op_RegI]->Remove(i); // Exclude save-on-call + idealreg2debugmask [Op_RegL]->Remove(i); // registers from debug + idealreg2debugmask [Op_RegF]->Remove(i); // masks + idealreg2debugmask [Op_RegD]->Remove(i); + idealreg2debugmask [Op_RegP]->Remove(i); + + idealreg2mhdebugmask[Op_RegN]->Remove(i); + idealreg2mhdebugmask[Op_RegI]->Remove(i); + idealreg2mhdebugmask[Op_RegL]->Remove(i); + idealreg2mhdebugmask[Op_RegF]->Remove(i); + idealreg2mhdebugmask[Op_RegD]->Remove(i); + idealreg2mhdebugmask[Op_RegP]->Remove(i); } } + + // Subtract the register we use to save the SP for MethodHandle + // invokes to from the debug mask. + const RegMask save_mask = method_handle_invoke_SP_save_mask(); + idealreg2mhdebugmask[Op_RegN]->SUBTRACT(save_mask); + idealreg2mhdebugmask[Op_RegI]->SUBTRACT(save_mask); + idealreg2mhdebugmask[Op_RegL]->SUBTRACT(save_mask); + idealreg2mhdebugmask[Op_RegF]->SUBTRACT(save_mask); + idealreg2mhdebugmask[Op_RegD]->SUBTRACT(save_mask); + idealreg2mhdebugmask[Op_RegP]->SUBTRACT(save_mask); } //---------------------------is_save_on_entry---------------------------------- @@ -989,6 +1030,7 @@ CallNode *call; const TypeTuple *domain; ciMethod* method = NULL; + bool is_method_handle_invoke = false; // for special kill effects if( sfpt->is_Call() ) { call = sfpt->as_Call(); domain = call->tf()->domain(); @@ -1013,6 +1055,8 @@ mcall_java->_method = method; mcall_java->_bci = call_java->_bci; mcall_java->_optimized_virtual = call_java->is_optimized_virtual(); + is_method_handle_invoke = call_java->is_method_handle_invoke(); + mcall_java->_method_handle_invoke = is_method_handle_invoke; if( mcall_java->is_MachCallStaticJava() ) mcall_java->as_MachCallStaticJava()->_name = call_java->as_CallStaticJava()->_name; @@ -1126,6 +1170,15 @@ mcall->_argsize = out_arg_limit_per_call - begin_out_arg_area; } + if (is_method_handle_invoke) { + // Kill some extra stack space in case method handles want to do + // a little in-place argument insertion. + int regs_per_word = NOT_LP64(1) LP64_ONLY(2); // %%% make a global const! + out_arg_limit_per_call += MethodHandlePushLimit * regs_per_word; + // Do not update mcall->_argsize because (a) the extra space is not + // pushed as arguments and (b) _argsize is dead (not used anywhere). + } + // Compute the max stack slot killed by any call. These will not be // available for debug info, and will be used to adjust FIRST_STACK_mask // after all call sites have been visited. @@ -1527,7 +1580,7 @@ uint num_proj = _proj_list.size(); // Perform any 1-to-many expansions required - MachNode *ex = mach->Expand(s,_proj_list); + MachNode *ex = mach->Expand(s,_proj_list, mem); if( ex != mach ) { assert(ex->ideal_reg() == mach->ideal_reg(), "ideal types should match"); if( ex->in(1)->is_Con() ) @@ -1832,67 +1885,23 @@ case Op_Binary: // These are introduced in the Post_Visit state. ShouldNotReachHere(); break; - case Op_StoreB: // Do match these, despite no ideal reg - case Op_StoreC: - case Op_StoreCM: - case Op_StoreD: - case Op_StoreF: - case Op_StoreI: - case Op_StoreL: - case Op_StoreP: - case Op_StoreN: - case Op_Store16B: - case Op_Store8B: - case Op_Store4B: - case Op_Store8C: - case Op_Store4C: - case Op_Store2C: - case Op_Store4I: - case Op_Store2I: - case Op_Store2L: - case Op_Store4F: - case Op_Store2F: - case Op_Store2D: case Op_ClearArray: case Op_SafePoint: mem_op = true; break; - case Op_LoadB: - case Op_LoadUS: - case Op_LoadD: - case Op_LoadF: - case Op_LoadI: - case Op_LoadKlass: - case Op_LoadNKlass: - case Op_LoadL: - case Op_LoadS: - case Op_LoadP: - case Op_LoadN: - case Op_LoadRange: - case Op_LoadD_unaligned: - case Op_LoadL_unaligned: - case Op_Load16B: - case Op_Load8B: - case Op_Load4B: - case Op_Load4C: - case Op_Load2C: - case Op_Load8C: - case Op_Load8S: - case Op_Load4S: - case Op_Load2S: - case Op_Load4I: - case Op_Load2I: - case Op_Load2L: - case Op_Load4F: - case Op_Load2F: - case Op_Load2D: - mem_op = true; - // Must be root of match tree due to prior load conflict - if( C->subsume_loads() == false ) { - set_shared(n); + default: + if( n->is_Store() ) { + // Do match stores, despite no ideal reg + mem_op = true; + break; + } + if( n->is_Mem() ) { // Loads and LoadStores + mem_op = true; + // Loads must be root of match tree due to prior load conflict + if( C->subsume_loads() == false ) + set_shared(n); } // Fall into default case - default: if( !n->ideal_reg() ) set_dontcare(n); // Unmatchable Nodes } // end_switch @@ -1913,15 +1922,15 @@ continue; // for(int i = ...) } - // Clone addressing expressions as they are "free" in most instructions + if( mop == Op_AddP && m->in(AddPNode::Base)->Opcode() == Op_DecodeN ) { + // Bases used in addresses must be shared but since + // they are shared through a DecodeN they may appear + // to have a single use so force sharing here. + set_shared(m->in(AddPNode::Base)->in(1)); + } + + // Clone addressing expressions as they are "free" in memory access instructions if( mem_op && i == MemNode::Address && mop == Op_AddP ) { - if (m->in(AddPNode::Base)->Opcode() == Op_DecodeN) { - // Bases used in addresses must be shared but since - // they are shared through a DecodeN they may appear - // to have a single use so force sharing here. - set_shared(m->in(AddPNode::Base)->in(1)); - } - // Some inputs for address expression are not put on stack // to avoid marking them as shared and forcing them into register // if they are used only in address expressions.
--- a/src/share/vm/opto/matcher.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/matcher.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -117,8 +117,9 @@ static const int base2reg[]; // Map Types to machine register types // Convert ideal machine register to a register mask for spill-loads static const RegMask *idealreg2regmask[]; - RegMask *idealreg2spillmask[_last_machine_leaf]; - RegMask *idealreg2debugmask[_last_machine_leaf]; + RegMask *idealreg2spillmask [_last_machine_leaf]; + RegMask *idealreg2debugmask [_last_machine_leaf]; + RegMask *idealreg2mhdebugmask[_last_machine_leaf]; void init_spill_mask( Node *ret ); // Convert machine register number to register mask static uint mreg2regmask_max; @@ -297,6 +298,8 @@ // Register for MODL projection of divmodL static RegMask modL_proj_mask(); + static const RegMask method_handle_invoke_SP_save_mask(); + // Java-Interpreter calling convention // (what you use when calling between compiled-Java and Interpreted-Java
--- a/src/share/vm/opto/memnode.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/memnode.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -123,6 +123,13 @@ } else { assert(false, "unexpected projection"); } + } else if (result->is_ClearArray()) { + if (!ClearArrayNode::step_through(&result, instance_id, phase)) { + // Can not bypass initialization of the instance + // we are looking for. + break; + } + // Otherwise skip it (the call updated 'result' value). } else if (result->is_MergeMem()) { result = step_through_mergemem(phase, result->as_MergeMem(), t_adr, NULL, tty); } @@ -255,6 +262,13 @@ return NodeSentinel; // caller will return NULL } + // Do NOT remove or optimize the next lines: ensure a new alias index + // is allocated for an oop pointer type before Escape Analysis. + // Note: C++ will not remove it since the call has side effect. + if ( t_adr->isa_oopptr() ) { + int alias_idx = phase->C->get_alias_index(t_adr->is_ptr()); + } + #ifdef ASSERT Node* base = NULL; if (address->is_AddP()) @@ -530,6 +544,15 @@ } else if (mem->is_Proj() && mem->in(0)->is_MemBar()) { mem = mem->in(0)->in(TypeFunc::Memory); continue; // (a) advance through independent MemBar memory + } else if (mem->is_ClearArray()) { + if (ClearArrayNode::step_through(&mem, (uint)addr_t->instance_id(), phase)) { + // (the call updated 'mem' value) + continue; // (a) advance through independent allocation memory + } else { + // Can not bypass initialization of the instance + // we are looking for. + return mem; + } } else if (mem->is_MergeMem()) { int alias_idx = phase->C->get_alias_index(adr_type()); mem = mem->as_MergeMem()->memory_at(alias_idx); @@ -1496,6 +1519,8 @@ } } } else if (tp->base() == Type::InstPtr) { + const TypeInstPtr* tinst = tp->is_instptr(); + ciKlass* klass = tinst->klass(); assert( off != Type::OffsetBot || // arrays can be cast to Objects tp->is_oopptr()->klass()->is_java_lang_Object() || @@ -1503,6 +1528,25 @@ phase->C->has_unsafe_access(), "Field accesses must be precise" ); // For oop loads, we expect the _type to be precise + if (OptimizeStringConcat && klass == phase->C->env()->String_klass() && + adr->is_AddP() && off != Type::OffsetBot) { + // For constant Strings treat the fields as compile time constants. + Node* base = adr->in(AddPNode::Base); + if (base->Opcode() == Op_ConP) { + const TypeOopPtr* t = phase->type(base)->isa_oopptr(); + ciObject* string = t->const_oop(); + ciConstant constant = string->as_instance()->field_value_by_offset(off); + if (constant.basic_type() == T_INT) { + return TypeInt::make(constant.as_int()); + } else if (constant.basic_type() == T_ARRAY) { + if (adr->bottom_type()->is_ptr_to_narrowoop()) { + return TypeNarrowOop::make_from_constant(constant.as_object()); + } else { + return TypeOopPtr::make_from_constant(constant.as_object()); + } + } + } + } } else if (tp->base() == Type::KlassPtr) { assert( off != Type::OffsetBot || // arrays can be cast to Objects @@ -2426,6 +2470,31 @@ return mem; } +//----------------------------step_through---------------------------------- +// Return allocation input memory edge if it is different instance +// or itself if it is the one we are looking for. +bool ClearArrayNode::step_through(Node** np, uint instance_id, PhaseTransform* phase) { + Node* n = *np; + assert(n->is_ClearArray(), "sanity"); + intptr_t offset; + AllocateNode* alloc = AllocateNode::Ideal_allocation(n->in(3), phase, offset); + // This method is called only before Allocate nodes are expanded during + // macro nodes expansion. Before that ClearArray nodes are only generated + // in LibraryCallKit::generate_arraycopy() which follows allocations. + assert(alloc != NULL, "should have allocation"); + if (alloc->_idx == instance_id) { + // Can not bypass initialization of the instance we are looking for. + return false; + } + // Otherwise skip it. + InitializeNode* init = alloc->initialization(); + if (init != NULL) + *np = init->in(TypeFunc::Memory); + else + *np = alloc->in(TypeFunc::Memory); + return true; +} + //----------------------------clear_memory------------------------------------- // Generate code to initialize object storage to zero. Node* ClearArrayNode::clear_memory(Node* ctl, Node* mem, Node* dest, @@ -2599,7 +2668,30 @@ // Return a node which is more "ideal" than the current node. Strip out // control copies Node *MemBarNode::Ideal(PhaseGVN *phase, bool can_reshape) { - return remove_dead_region(phase, can_reshape) ? this : NULL; + if (remove_dead_region(phase, can_reshape)) return this; + + // Eliminate volatile MemBars for scalar replaced objects. + if (can_reshape && req() == (Precedent+1) && + (Opcode() == Op_MemBarAcquire || Opcode() == Op_MemBarVolatile)) { + // Volatile field loads and stores. + Node* my_mem = in(MemBarNode::Precedent); + if (my_mem != NULL && my_mem->is_Mem()) { + const TypeOopPtr* t_oop = my_mem->in(MemNode::Address)->bottom_type()->isa_oopptr(); + // Check for scalar replaced object reference. + if( t_oop != NULL && t_oop->is_known_instance_field() && + t_oop->offset() != Type::OffsetBot && + t_oop->offset() != Type::OffsetTop) { + // Replace MemBar projections by its inputs. + PhaseIterGVN* igvn = phase->is_IterGVN(); + igvn->replace_node(proj_out(TypeFunc::Memory), in(TypeFunc::Memory)); + igvn->replace_node(proj_out(TypeFunc::Control), in(TypeFunc::Control)); + // Must return either the original node (now dead) or a new node + // (Do not return a top here, since that would break the uniqueness of top.) + return new (phase->C, 1) ConINode(TypeInt::ZERO); + } + } + } + return NULL; } //------------------------------Value------------------------------------------
--- a/src/share/vm/opto/memnode.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/memnode.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -583,9 +583,22 @@ // Preceeding equivalent StoreCMs may be eliminated. class StoreCMNode : public StoreNode { private: + virtual uint hash() const { return StoreNode::hash() + _oop_alias_idx; } + virtual uint cmp( const Node &n ) const { + return _oop_alias_idx == ((StoreCMNode&)n)._oop_alias_idx + && StoreNode::cmp(n); + } + virtual uint size_of() const { return sizeof(*this); } int _oop_alias_idx; // The alias_idx of OopStore + public: - StoreCMNode( Node *c, Node *mem, Node *adr, const TypePtr* at, Node *val, Node *oop_store, int oop_alias_idx ) : StoreNode(c,mem,adr,at,val,oop_store), _oop_alias_idx(oop_alias_idx) {} + StoreCMNode( Node *c, Node *mem, Node *adr, const TypePtr* at, Node *val, Node *oop_store, int oop_alias_idx ) : + StoreNode(c,mem,adr,at,val,oop_store), + _oop_alias_idx(oop_alias_idx) { + assert(_oop_alias_idx >= Compile::AliasIdxRaw || + _oop_alias_idx == Compile::AliasIdxBot && Compile::current()->AliasLevel() == 0, + "bad oop alias idx"); + } virtual int Opcode() const; virtual Node *Identity( PhaseTransform *phase ); virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); @@ -717,7 +730,10 @@ //------------------------------ClearArray------------------------------------- class ClearArrayNode: public Node { public: - ClearArrayNode( Node *ctrl, Node *arymem, Node *word_cnt, Node *base ) : Node(ctrl,arymem,word_cnt,base) {} + ClearArrayNode( Node *ctrl, Node *arymem, Node *word_cnt, Node *base ) + : Node(ctrl,arymem,word_cnt,base) { + init_class_id(Class_ClearArray); + } virtual int Opcode() const; virtual const Type *bottom_type() const { return Type::MEMORY; } // ClearArray modifies array elements, and so affects only the @@ -743,6 +759,9 @@ Node* start_offset, Node* end_offset, PhaseGVN* phase); + // Return allocation input memory edge if it is different instance + // or itself if it is the one we are looking for. + static bool step_through(Node** np, uint instance_id, PhaseTransform* phase); }; //------------------------------StrComp-------------------------------------
--- a/src/share/vm/opto/node.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/node.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -47,6 +47,7 @@ class CatchNode; class CatchProjNode; class CheckCastPPNode; +class ClearArrayNode; class CmpNode; class CodeBuffer; class ConstraintCastNode; @@ -599,8 +600,9 @@ DEFINE_CLASS_ID(BoxLock, Node, 10) DEFINE_CLASS_ID(Add, Node, 11) DEFINE_CLASS_ID(Mul, Node, 12) + DEFINE_CLASS_ID(ClearArray, Node, 13) - _max_classes = ClassMask_Mul + _max_classes = ClassMask_ClearArray }; #undef DEFINE_CLASS_ID @@ -661,18 +663,25 @@ return (_flags & Flag_is_Call) != 0; } + CallNode* isa_Call() const { + return is_Call() ? as_Call() : NULL; + } + CallNode *as_Call() const { // Only for CallNode (not for MachCallNode) assert((_class_id & ClassMask_Call) == Class_Call, "invalid node class"); return (CallNode*)this; } - #define DEFINE_CLASS_QUERY(type) \ - bool is_##type() const { \ + #define DEFINE_CLASS_QUERY(type) \ + bool is_##type() const { \ return ((_class_id & ClassMask_##type) == Class_##type); \ - } \ - type##Node *as_##type() const { \ - assert(is_##type(), "invalid node class"); \ - return (type##Node*)this; \ + } \ + type##Node *as_##type() const { \ + assert(is_##type(), "invalid node class"); \ + return (type##Node*)this; \ + } \ + type##Node* isa_##type() const { \ + return (is_##type()) ? as_##type() : NULL; \ } DEFINE_CLASS_QUERY(AbstractLock) @@ -691,6 +700,7 @@ DEFINE_CLASS_QUERY(CatchProj) DEFINE_CLASS_QUERY(CheckCastPP) DEFINE_CLASS_QUERY(ConstraintCast) + DEFINE_CLASS_QUERY(ClearArray) DEFINE_CLASS_QUERY(CMove) DEFINE_CLASS_QUERY(Cmp) DEFINE_CLASS_QUERY(CountedLoop) @@ -1249,6 +1259,24 @@ #undef I_VDUI_ONLY #undef VDUI_ONLY +// An Iterator that truly follows the iterator pattern. Doesn't +// support deletion but could be made to. +// +// for (SimpleDUIterator i(n); i.has_next(); i.next()) { +// Node* m = i.get(); +// +class SimpleDUIterator : public StackObj { + private: + Node* node; + DUIterator_Fast i; + DUIterator_Fast imax; + public: + SimpleDUIterator(Node* n): node(n), i(n->fast_outs(imax)) {} + bool has_next() { return i < imax; } + void next() { i++; } + Node* get() { return node->fast_out(i); } +}; + //----------------------------------------------------------------------------- // Map dense integer indices to Nodes. Uses classic doubling-array trick. @@ -1290,6 +1318,12 @@ public: Node_List() : Node_Array(Thread::current()->resource_area()), _cnt(0) {} Node_List(Arena *a) : Node_Array(a), _cnt(0) {} + bool contains(Node* n) { + for (uint e = 0; e < size(); e++) { + if (at(e) == n) return true; + } + return false; + } void insert( uint i, Node *n ) { Node_Array::insert(i,n); _cnt++; } void remove( uint i ) { Node_Array::remove(i); _cnt--; } void push( Node *b ) { map(_cnt++,b); }
--- a/src/share/vm/opto/output.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/output.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1998-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1998-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -794,6 +794,8 @@ #endif int safepoint_pc_offset = current_offset; + bool is_method_handle_invoke = false; + bool return_oop = false; // Add the safepoint in the DebugInfoRecorder if( !mach->is_MachCall() ) { @@ -801,6 +803,20 @@ debug_info()->add_safepoint(safepoint_pc_offset, sfn->_oop_map); } else { mcall = mach->as_MachCall(); + + // Is the call a MethodHandle call? + if (mcall->is_MachCallJava()) { + if (mcall->as_MachCallJava()->_method_handle_invoke) { + assert(has_method_handle_invokes(), "must have been set during call generation"); + is_method_handle_invoke = true; + } + } + + // Check if a call returns an object. + if (mcall->return_value_is_used() && + mcall->tf()->range()->field_at(TypeFunc::Parms)->isa_ptr()) { + return_oop = true; + } safepoint_pc_offset += mcall->ret_addr_offset(); debug_info()->add_safepoint(safepoint_pc_offset, mcall->_oop_map); } @@ -911,9 +927,9 @@ ciMethod* scope_method = method ? method : _method; // Describe the scope here assert(jvms->bci() >= InvocationEntryBci && jvms->bci() <= 0x10000, "must be a valid or entry BCI"); - assert(!jvms->should_reexecute() || depth==max_depth, "reexecute allowed only for the youngest"); + assert(!jvms->should_reexecute() || depth == max_depth, "reexecute allowed only for the youngest"); // Now we can describe the scope. - debug_info()->describe_scope(safepoint_pc_offset,scope_method,jvms->bci(),jvms->should_reexecute(),locvals,expvals,monvals); + debug_info()->describe_scope(safepoint_pc_offset, scope_method, jvms->bci(), jvms->should_reexecute(), is_method_handle_invoke, return_oop, locvals, expvals, monvals); } // End jvms loop // Mark the end of the scope set. @@ -1080,14 +1096,26 @@ deopt_handler_req += MAX_stubs_size; // add marginal slop for handler stub_req += MAX_stubs_size; // ensure per-stub margin code_req += MAX_inst_size; // ensure per-instruction margin + if (StressCodeBuffers) code_req = const_req = stub_req = exception_handler_req = deopt_handler_req = 0x10; // force expansion - int total_req = code_req + pad_req + stub_req + exception_handler_req + deopt_handler_req + const_req; + + int total_req = + code_req + + pad_req + + stub_req + + exception_handler_req + + deopt_handler_req + // deopt handler + const_req; + + if (has_method_handle_invokes()) + total_req += deopt_handler_req; // deopt MH handler + CodeBuffer* cb = code_buffer(); cb->initialize(total_req, locs_req); // Have we run out of code space? - if (cb->blob() == NULL) { + if ((cb->blob() == NULL) || (!CompileBroker::should_compile_new_jobs())) { turn_off_compiler(this); return; } @@ -1308,7 +1336,7 @@ // Verify that there is sufficient space remaining cb->insts()->maybe_expand_to_ensure_remaining(MAX_inst_size); - if (cb->blob() == NULL) { + if ((cb->blob() == NULL) || (!CompileBroker::should_compile_new_jobs())) { turn_off_compiler(this); return; } @@ -1424,10 +1452,17 @@ _code_offsets.set_value(CodeOffsets::Exceptions, emit_exception_handler(*cb)); // Emit the deopt handler code. _code_offsets.set_value(CodeOffsets::Deopt, emit_deopt_handler(*cb)); + + // Emit the MethodHandle deopt handler code (if required). + if (has_method_handle_invokes()) { + // We can use the same code as for the normal deopt handler, we + // just need a different entry point address. + _code_offsets.set_value(CodeOffsets::DeoptMH, emit_deopt_handler(*cb)); + } } // One last check for failed CodeBuffer::expand: - if (cb->blob() == NULL) { + if ((cb->blob() == NULL) || (!CompileBroker::should_compile_new_jobs())) { turn_off_compiler(this); return; }
--- a/src/share/vm/opto/parse.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/parse.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -39,6 +39,7 @@ // Always between 0.0 and 1.0. Represents the percentage of the method's // total execution time used at this call site. const float _site_invoke_ratio; + const int _site_depth_adjust; float compute_callee_frequency( int caller_bci ) const; GrowableArray<InlineTree*> _subtrees; @@ -50,7 +51,8 @@ ciMethod* callee_method, JVMState* caller_jvms, int caller_bci, - float site_invoke_ratio); + float site_invoke_ratio, + int site_depth_adjust); InlineTree *build_inline_tree_for_callee(ciMethod* callee_method, JVMState* caller_jvms, int caller_bci); @@ -61,14 +63,15 @@ InlineTree *caller_tree() const { return _caller_tree; } InlineTree* callee_at(int bci, ciMethod* m) const; - int inline_depth() const { return _caller_jvms ? _caller_jvms->depth() : 0; } + int inline_depth() const { return stack_depth() + _site_depth_adjust; } + int stack_depth() const { return _caller_jvms ? _caller_jvms->depth() : 0; } public: static InlineTree* build_inline_tree_root(); static InlineTree* find_subtree_from_root(InlineTree* root, JVMState* jvms, ciMethod* callee, bool create_if_not_found = false); // For temporary (stack-allocated, stateless) ilts: - InlineTree(Compile* c, ciMethod* callee_method, JVMState* caller_jvms, float site_invoke_ratio); + InlineTree(Compile* c, ciMethod* callee_method, JVMState* caller_jvms, float site_invoke_ratio, int site_depth_adjust); // InlineTree enum enum InlineStyle { @@ -427,6 +430,11 @@ } } + // Return true if the parser should add a loop predicate + bool should_add_predicate(int target_bci); + // Insert a loop predicate into the graph + void add_predicate(); + // Note: Intrinsic generation routines may be found in library_call.cpp. // Helper function to setup Ideal Call nodes @@ -488,7 +496,7 @@ void do_ifnull(BoolTest::mask btest, Node* c); void do_if(BoolTest::mask btest, Node* c); - void repush_if_args(); + int repush_if_args(); void adjust_map_after_if(BoolTest::mask btest, Node* c, float prob, Block* path, Block* other_path); IfNode* jump_if_fork_int(Node* a, Node* b, BoolTest::mask mask);
--- a/src/share/vm/opto/parse1.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/parse1.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -231,12 +231,13 @@ // Use the raw liveness computation to make sure that unexpected // values don't propagate into the OSR frame. - MethodLivenessResult live_locals = method()->raw_liveness_at_bci(osr_bci()); + MethodLivenessResult live_locals = method()->liveness_at_bci(osr_bci()); if (!live_locals.is_valid()) { // Degenerate or breakpointed method. C->record_method_not_compilable("OSR in empty or breakpointed method"); return; } + MethodLivenessResult raw_live_locals = method()->raw_liveness_at_bci(osr_bci()); // Extract the needed locals from the interpreter frame. Node *locals_addr = basic_plus_adr(osr_buf, osr_buf, (max_locals-1)*wordSize); @@ -316,6 +317,10 @@ continue; } } + if (type->basic_type() == T_ADDRESS && !raw_live_locals.at(index)) { + // Skip type check for dead address locals + continue; + } set_local(index, check_interpreter_type(l, type, bad_type_exit)); } @@ -819,7 +824,6 @@ case Bytecodes::_ddiv: case Bytecodes::_checkcast: case Bytecodes::_instanceof: - case Bytecodes::_athrow: case Bytecodes::_anewarray: case Bytecodes::_newarray: case Bytecodes::_multianewarray: @@ -829,6 +833,8 @@ return true; break; + // Don't rerun athrow since it's part of the exception path. + case Bytecodes::_athrow: case Bytecodes::_invokestatic: case Bytecodes::_invokedynamic: case Bytecodes::_invokespecial: @@ -1378,6 +1384,10 @@ set_parse_bci(iter().cur_bci()); if (bci() == block()->limit()) { + // insert a predicate if it falls through to a loop head block + if (should_add_predicate(bci())){ + add_predicate(); + } // Do not walk into the next block until directed by do_all_blocks. merge(bci()); break; @@ -2078,6 +2088,37 @@ } } +//------------------------------should_add_predicate-------------------------- +bool Parse::should_add_predicate(int target_bci) { + if (!UseLoopPredicate) return false; + Block* target = successor_for_bci(target_bci); + if (target != NULL && + target->is_loop_head() && + block()->rpo() < target->rpo()) { + return true; + } + return false; +} + +//------------------------------add_predicate--------------------------------- +void Parse::add_predicate() { + assert(UseLoopPredicate,"use only for loop predicate"); + Node *cont = _gvn.intcon(1); + Node* opq = _gvn.transform(new (C, 2) Opaque1Node(C, cont)); + Node *bol = _gvn.transform(new (C, 2) Conv2BNode(opq)); + IfNode* iff = create_and_map_if(control(), bol, PROB_MAX, COUNT_UNKNOWN); + Node* iffalse = _gvn.transform(new (C, 1) IfFalseNode(iff)); + C->add_predicate_opaq(opq); + { + PreserveJVMState pjvms(this); + set_control(iffalse); + uncommon_trap(Deoptimization::Reason_predicate, + Deoptimization::Action_maybe_recompile); + } + Node* iftrue = _gvn.transform(new (C, 1) IfTrueNode(iff)); + set_control(iftrue); +} + #ifndef PRODUCT //------------------------show_parse_info-------------------------------------- void Parse::show_parse_info() {
--- a/src/share/vm/opto/parse2.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/parse2.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1998-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1998-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -278,6 +278,11 @@ if (len < 1) { // If this is a backward branch, add safepoint maybe_add_safepoint(default_dest); + if (should_add_predicate(default_dest)){ + _sp += 1; // set original stack for use by uncommon_trap + add_predicate(); + _sp -= 1; + } merge(default_dest); return; } @@ -324,6 +329,11 @@ if (len < 1) { // If this is a backward branch, add safepoint maybe_add_safepoint(default_dest); + if (should_add_predicate(default_dest)){ + _sp += 1; // set original stack for use by uncommon_trap + add_predicate(); + _sp -= 1; + } merge(default_dest); return; } @@ -731,6 +741,9 @@ push(_gvn.makecon(ret_addr)); // Flow to the jsr. + if (should_add_predicate(jsr_bci)){ + add_predicate(); + } merge(jsr_bci); } @@ -881,7 +894,7 @@ //-------------------------------repush_if_args-------------------------------- // Push arguments of an "if" bytecode back onto the stack by adjusting _sp. -inline void Parse::repush_if_args() { +inline int Parse::repush_if_args() { #ifndef PRODUCT if (PrintOpto && WizardMode) { tty->print("defending against excessive implicit null exceptions on %s @%d in ", @@ -895,6 +908,7 @@ assert(argument(0) != NULL, "must exist"); assert(bc_depth == 1 || argument(1) != NULL, "two must exist"); _sp += bc_depth; + return bc_depth; } //----------------------------------do_ifnull---------------------------------- @@ -954,8 +968,14 @@ // Update method data profile_taken_branch(target_bci); adjust_map_after_if(btest, c, prob, branch_block, next_block); - if (!stopped()) + if (!stopped()) { + if (should_add_predicate(target_bci)){ // add a predicate if it branches to a loop + int nargs = repush_if_args(); // set original stack for uncommon_trap + add_predicate(); + _sp -= nargs; + } merge(target_bci); + } } } @@ -1076,8 +1096,14 @@ // Update method data profile_taken_branch(target_bci); adjust_map_after_if(taken_btest, c, prob, branch_block, next_block); - if (!stopped()) + if (!stopped()) { + if (should_add_predicate(target_bci)){ // add a predicate if it branches to a loop + int nargs = repush_if_args(); // set original stack for the uncommon_trap + add_predicate(); + _sp -= nargs; + } merge(target_bci); + } } } @@ -2053,13 +2079,6 @@ // null exception oop throws NULL pointer exception do_null_check(peek(), T_OBJECT); if (stopped()) return; - if (env()->jvmti_can_post_exceptions()) { - // "Full-speed throwing" is not necessary here, - // since we're notifying the VM on every throw. - uncommon_trap(Deoptimization::Reason_unhandled, - Deoptimization::Action_none); - return; - } // Hook the thrown exception directly to subsequent handlers. if (BailoutToInterpreterForThrows) { // Keep method interpreted from now on. @@ -2067,6 +2086,11 @@ Deoptimization::Action_make_not_compilable); return; } + if (env()->jvmti_can_post_on_exceptions()) { + // check if we must post exception events, take uncommon trap if so (with must_throw = false) + uncommon_trap_if_should_post_on_exceptions(Deoptimization::Reason_unhandled, false); + } + // Here if either can_post_on_exceptions or should_post_on_exceptions is false add_exception_state(make_exception_state(peek())); break; @@ -2080,6 +2104,10 @@ // Update method data profile_taken_branch(target_bci); + // Add loop predicate if it goes to a loop + if (should_add_predicate(target_bci)){ + add_predicate(); + } // Merge the current control into the target basic block merge(target_bci);
--- a/src/share/vm/opto/parse3.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/parse3.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1998-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1998-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -125,7 +125,25 @@ void Parse::do_get_xxx(const TypePtr* obj_type, Node* obj, ciField* field, bool is_field) { // Does this field have a constant value? If so, just push the value. - if (field->is_constant() && push_constant(field->constant_value())) return; + if (field->is_constant()) { + if (field->is_static()) { + // final static field + if (push_constant(field->constant_value())) + return; + } + else { + // final non-static field of a trusted class ({java,sun}.dyn + // classes). + if (obj->is_Con()) { + const TypeOopPtr* oop_ptr = obj->bottom_type()->isa_oopptr(); + ciObject* constant_oop = oop_ptr->const_oop(); + ciConstant constant = field->constant_value_of(constant_oop); + + if (push_constant(constant, true)) + return; + } + } + } ciType* field_klass = field->type(); bool is_vol = field->is_volatile(); @@ -145,7 +163,7 @@ if (!field->type()->is_loaded()) { type = TypeInstPtr::BOTTOM; must_assert_null = true; - } else if (field->is_constant()) { + } else if (field->is_constant() && field->is_static()) { // This can happen if the constant oop is non-perm. ciObject* con = field->constant_value().as_object(); // Do not "join" in the previous type; it doesn't add value, @@ -240,19 +258,19 @@ // membar is dependent on the store, keeping any other membars generated // below from floating up past the store. int adr_idx = C->get_alias_index(adr_type); - insert_mem_bar_volatile(Op_MemBarVolatile, adr_idx); + insert_mem_bar_volatile(Op_MemBarVolatile, adr_idx, store); // Now place a membar for AliasIdxBot for the unknown yet-to-be-parsed // volatile alias indices. Skip this if the membar is redundant. if (adr_idx != Compile::AliasIdxBot) { - insert_mem_bar_volatile(Op_MemBarVolatile, Compile::AliasIdxBot); + insert_mem_bar_volatile(Op_MemBarVolatile, Compile::AliasIdxBot, store); } // Finally, place alias-index-specific membars for each volatile index // that isn't the adr_idx membar. Typically there's only 1 or 2. for( int i = Compile::AliasIdxRaw; i < C->num_alias_types(); i++ ) { if (i != adr_idx && C->alias_type(i)->is_volatile()) { - insert_mem_bar_volatile(Op_MemBarVolatile, i); + insert_mem_bar_volatile(Op_MemBarVolatile, i, store); } } } @@ -421,8 +439,18 @@ // Can use multianewarray instead of [a]newarray if only one dimension, // or if all non-final dimensions are small constants. - if (expand_count == 1 || (1 <= expand_count && expand_count <= expand_limit)) { - Node* obj = expand_multianewarray(array_klass, &length[0], ndimensions, ndimensions); + if (ndimensions == 1 || (1 <= expand_count && expand_count <= expand_limit)) { + Node* obj = NULL; + // Set the original stack and the reexecute bit for the interpreter + // to reexecute the multianewarray bytecode if deoptimization happens. + // Do it unconditionally even for one dimension multianewarray. + // Note: the reexecute bit will be set in GraphKit::add_safepoint_edges() + // when AllocateArray node for newarray is created. + { PreserveReexecuteState preexecs(this); + _sp += ndimensions; + // Pass 0 as nargs since uncommon trap code does not need to restore stack. + obj = expand_multianewarray(array_klass, &length[0], ndimensions, 0); + } //original reexecute and sp are set back here push(obj); return; }
--- a/src/share/vm/opto/parseHelper.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/parseHelper.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -221,6 +221,14 @@ // Push resultant oop onto stack push(obj); + + // Keep track of whether opportunities exist for StringBuilder + // optimizations. + if (OptimizeStringConcat && + (klass == C->env()->StringBuilder_klass() || + klass == C->env()->StringBuffer_klass())) { + C->set_has_stringbuilder(true); + } } #ifndef PRODUCT @@ -406,8 +414,6 @@ void Parse::profile_call(Node* receiver) { if (!method_data_update()) return; - profile_generic_call(); - switch (bc()) { case Bytecodes::_invokevirtual: case Bytecodes::_invokeinterface: @@ -416,6 +422,7 @@ case Bytecodes::_invokestatic: case Bytecodes::_invokedynamic: case Bytecodes::_invokespecial: + profile_generic_call(); break; default: fatal("unexpected call bytecode"); } @@ -436,13 +443,16 @@ void Parse::profile_receiver_type(Node* receiver) { assert(method_data_update(), "must be generating profile code"); - // Skip if we aren't tracking receivers - if (TypeProfileWidth < 1) return; - ciMethodData* md = method()->method_data(); assert(md != NULL, "expected valid ciMethodData"); ciProfileData* data = md->bci_to_data(bci()); assert(data->is_ReceiverTypeData(), "need ReceiverTypeData here"); + + // Skip if we aren't tracking receivers + if (TypeProfileWidth < 1) { + increment_md_counter_at(md, data, CounterData::count_offset()); + return; + } ciReceiverTypeData* rdata = (ciReceiverTypeData*)data->as_ReceiverTypeData(); Node* method_data = method_data_addressing(md, rdata, in_ByteSize(0));
--- a/src/share/vm/opto/phase.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/phase.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -44,6 +44,7 @@ BlockLayout, // Linear ordering of blocks Register_Allocation, // Register allocation, duh LIVE, // Dragon-book LIVE range problem + StringOpts, // StringBuilder related optimizations Interference_Graph, // Building the IFG Coalesce, // Coalescing copies Ideal_Loop, // Find idealized trip-counted loops
--- a/src/share/vm/opto/phaseX.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/phaseX.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -345,7 +345,11 @@ Node *hash_find(const Node *n) { return _table.hash_find(n); } // Used after parsing to eliminate values that are no longer in program - void remove_useless_nodes(VectorSet &useful) { _table.remove_useless_nodes(useful); } + void remove_useless_nodes(VectorSet &useful) { + _table.remove_useless_nodes(useful); + // this may invalidate cached cons so reset the cache + init_con_caches(); + } virtual ConNode* uncached_makecon(const Type* t); // override from PhaseTransform
--- a/src/share/vm/opto/runtime.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/runtime.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1998-2007 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1998-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -143,7 +143,7 @@ // We failed the fast-path allocation. Now we need to do a scavenge or GC // and try allocation again. -void OptoRuntime::maybe_defer_card_mark(JavaThread* thread) { +void OptoRuntime::new_store_pre_barrier(JavaThread* thread) { // After any safepoint, just before going back to compiled code, // we inform the GC that we will be doing initializing writes to // this object in the future without emitting card-marks, so @@ -156,7 +156,7 @@ assert(Universe::heap()->can_elide_tlab_store_barriers(), "compiler must check this first"); // GC may decide to give back a safer copy of new_obj. - new_obj = Universe::heap()->defer_store_barrier(thread, new_obj); + new_obj = Universe::heap()->new_store_pre_barrier(thread, new_obj); thread->set_vm_result(new_obj); } @@ -200,7 +200,7 @@ if (GraphKit::use_ReduceInitialCardMarks()) { // inform GC that we won't do card marks for initializing writes. - maybe_defer_card_mark(thread); + new_store_pre_barrier(thread); } JRT_END @@ -239,7 +239,7 @@ if (GraphKit::use_ReduceInitialCardMarks()) { // inform GC that we won't do card marks for initializing writes. - maybe_defer_card_mark(thread); + new_store_pre_barrier(thread); } JRT_END @@ -706,6 +706,11 @@ // vc->set_receiver_count(empty_row, DataLayout::counter_increment); int count_off = ReceiverTypeData::receiver_count_cell_index(empty_row); *(mdp + count_off) = DataLayout::counter_increment; + } else { + // Receiver did not match any saved receiver and there is no empty row for it. + // Increment total counter to indicate polymorphic case. + intptr_t* count_p = (intptr_t*)(((byte*)(data)) + in_bytes(CounterData::count_offset())); + *count_p += DataLayout::counter_increment; } JRT_END @@ -790,7 +795,7 @@ NOT_PRODUCT(Exceptions::debug_check_abort(exception)); #ifdef ASSERT - if (!(exception->is_a(SystemDictionary::throwable_klass()))) { + if (!(exception->is_a(SystemDictionary::Throwable_klass()))) { // should throw an exception here ShouldNotReachHere(); } @@ -810,7 +815,7 @@ // we are switching to old paradigm: search for exception handler in caller_frame // instead in exception handler of caller_frame.sender() - if (JvmtiExport::can_post_exceptions()) { + if (JvmtiExport::can_post_on_exceptions()) { // "Full-speed catching" is not necessary here, // since we're notifying the VM on every catch. // Force deoptimization and the rest of the lookup @@ -858,6 +863,9 @@ thread->set_exception_pc(pc); thread->set_exception_handler_pc(handler_address); thread->set_exception_stack_size(0); + + // Check if the exception PC is a MethodHandle call. + thread->set_is_method_handle_exception(nm->is_method_handle_return(pc)); } // Restore correct return pc. Was saved above. @@ -936,7 +944,7 @@ #endif assert (exception != NULL, "should have thrown a NULLPointerException"); #ifdef ASSERT - if (!(exception->is_a(SystemDictionary::throwable_klass()))) { + if (!(exception->is_a(SystemDictionary::Throwable_klass()))) { // should throw an exception here ShouldNotReachHere(); } @@ -972,8 +980,8 @@ assert(stub_frame.is_runtime_frame() || exception_blob()->contains(stub_frame.pc()), "sanity check"); frame caller_frame = stub_frame.sender(®_map); - VM_DeoptimizeFrame deopt(thread, caller_frame.id()); - VMThread::execute(&deopt); + // bypass VM_DeoptimizeFrame and deoptimize the frame directly + Deoptimization::deoptimize_frame(thread, caller_frame.id()); } }
--- a/src/share/vm/opto/runtime.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/runtime.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -133,8 +133,9 @@ // Allocate storage for a objArray or typeArray static void new_array_C(klassOopDesc* array_klass, int len, JavaThread *thread); - // Post-slow-path-allocation step for implementing ReduceInitialCardMarks: - static void maybe_defer_card_mark(JavaThread* thread); + // Post-slow-path-allocation, pre-initializing-stores step for + // implementing ReduceInitialCardMarks + static void new_store_pre_barrier(JavaThread* thread); // Allocate storage for a multi-dimensional arrays // Note: needs to be fixed for arbitrary number of dimensions
--- a/src/share/vm/opto/split_if.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/split_if.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -219,6 +219,7 @@ //------------------------------register_new_node------------------------------ void PhaseIdealLoop::register_new_node( Node *n, Node *blk ) { + assert(!n->is_CFG(), "must be data node"); _igvn.register_new_node_with_optimizer(n); set_ctrl(n, blk); IdealLoopTree *loop = get_loop(blk);
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/share/vm/opto/stringopts.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,1396 @@ +/* + * Copyright 2009-2010 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +#include "incls/_precompiled.incl" +#include "incls/_stringopts.cpp.incl" + +#define __ kit. + +class StringConcat : public ResourceObj { + private: + PhaseStringOpts* _stringopts; + Node* _string_alloc; + AllocateNode* _begin; // The allocation the begins the pattern + CallStaticJavaNode* _end; // The final call of the pattern. Will either be + // SB.toString or or String.<init>(SB.toString) + bool _multiple; // indicates this is a fusion of two or more + // separate StringBuilders + + Node* _arguments; // The list of arguments to be concatenated + GrowableArray<int> _mode; // into a String along with a mode flag + // indicating how to treat the value. + + Node_List _control; // List of control nodes that will be deleted + Node_List _uncommon_traps; // Uncommon traps that needs to be rewritten + // to restart at the initial JVMState. + public: + // Mode for converting arguments to Strings + enum { + StringMode, + IntMode, + CharMode + }; + + StringConcat(PhaseStringOpts* stringopts, CallStaticJavaNode* end): + _end(end), + _begin(NULL), + _multiple(false), + _string_alloc(NULL), + _stringopts(stringopts) { + _arguments = new (_stringopts->C, 1) Node(1); + _arguments->del_req(0); + } + + bool validate_control_flow(); + + void merge_add() { +#if 0 + // XXX This is place holder code for reusing an existing String + // allocation but the logic for checking the state safety is + // probably inadequate at the moment. + CallProjections endprojs; + sc->end()->extract_projections(&endprojs, false); + if (endprojs.resproj != NULL) { + for (SimpleDUIterator i(endprojs.resproj); i.has_next(); i.next()) { + CallStaticJavaNode *use = i.get()->isa_CallStaticJava(); + if (use != NULL && use->method() != NULL && + use->method()->holder() == C->env()->String_klass() && + use->method()->name() == ciSymbol::object_initializer_name() && + use->in(TypeFunc::Parms + 1) == endprojs.resproj) { + // Found useless new String(sb.toString()) so reuse the newly allocated String + // when creating the result instead of allocating a new one. + sc->set_string_alloc(use->in(TypeFunc::Parms)); + sc->set_end(use); + } + } + } +#endif + } + + StringConcat* merge(StringConcat* other, Node* arg); + + void set_allocation(AllocateNode* alloc) { + _begin = alloc; + } + + void append(Node* value, int mode) { + _arguments->add_req(value); + _mode.append(mode); + } + void push(Node* value, int mode) { + _arguments->ins_req(0, value); + _mode.insert_before(0, mode); + } + void push_string(Node* value) { + push(value, StringMode); + } + void push_int(Node* value) { + push(value, IntMode); + } + void push_char(Node* value) { + push(value, CharMode); + } + + Node* argument(int i) { + return _arguments->in(i); + } + void set_argument(int i, Node* value) { + _arguments->set_req(i, value); + } + int num_arguments() { + return _mode.length(); + } + int mode(int i) { + return _mode.at(i); + } + void add_control(Node* ctrl) { + assert(!_control.contains(ctrl), "only push once"); + _control.push(ctrl); + } + CallStaticJavaNode* end() { return _end; } + AllocateNode* begin() { return _begin; } + Node* string_alloc() { return _string_alloc; } + + void eliminate_unneeded_control(); + void eliminate_initialize(InitializeNode* init); + void eliminate_call(CallNode* call); + + void maybe_log_transform() { + CompileLog* log = _stringopts->C->log(); + if (log != NULL) { + log->head("replace_string_concat arguments='%d' string_alloc='%d' multiple='%d'", + num_arguments(), + _string_alloc != NULL, + _multiple); + JVMState* p = _begin->jvms(); + while (p != NULL) { + log->elem("jvms bci='%d' method='%d'", p->bci(), log->identify(p->method())); + p = p->caller(); + } + log->tail("replace_string_concat"); + } + } + + void convert_uncommon_traps(GraphKit& kit, const JVMState* jvms) { + for (uint u = 0; u < _uncommon_traps.size(); u++) { + Node* uct = _uncommon_traps.at(u); + + // Build a new call using the jvms state of the allocate + address call_addr = SharedRuntime::uncommon_trap_blob()->instructions_begin(); + const TypeFunc* call_type = OptoRuntime::uncommon_trap_Type(); + int size = call_type->domain()->cnt(); + const TypePtr* no_memory_effects = NULL; + Compile* C = _stringopts->C; + CallStaticJavaNode* call = new (C, size) CallStaticJavaNode(call_type, call_addr, "uncommon_trap", + jvms->bci(), no_memory_effects); + for (int e = 0; e < TypeFunc::Parms; e++) { + call->init_req(e, uct->in(e)); + } + // Set the trap request to record intrinsic failure if this trap + // is taken too many times. Ideally we would handle then traps by + // doing the original bookkeeping in the MDO so that if it caused + // the code to be thrown out we could still recompile and use the + // optimization. Failing the uncommon traps doesn't really mean + // that the optimization is a bad idea but there's no other way to + // do the MDO updates currently. + int trap_request = Deoptimization::make_trap_request(Deoptimization::Reason_intrinsic, + Deoptimization::Action_make_not_entrant); + call->init_req(TypeFunc::Parms, __ intcon(trap_request)); + kit.add_safepoint_edges(call); + + _stringopts->gvn()->transform(call); + C->gvn_replace_by(uct, call); + uct->disconnect_inputs(NULL); + } + } + + void cleanup() { + // disconnect the hook node + _arguments->disconnect_inputs(NULL); + } +}; + + +void StringConcat::eliminate_unneeded_control() { + eliminate_initialize(begin()->initialization()); + for (uint i = 0; i < _control.size(); i++) { + Node* n = _control.at(i); + if (n->is_Call()) { + if (n != _end) { + eliminate_call(n->as_Call()); + } + } else if (n->is_IfTrue()) { + Compile* C = _stringopts->C; + C->gvn_replace_by(n, n->in(0)->in(0)); + C->gvn_replace_by(n->in(0), C->top()); + } + } +} + + +StringConcat* StringConcat::merge(StringConcat* other, Node* arg) { + StringConcat* result = new StringConcat(_stringopts, _end); + for (uint x = 0; x < _control.size(); x++) { + Node* n = _control.at(x); + if (n->is_Call()) { + result->_control.push(n); + } + } + for (uint x = 0; x < other->_control.size(); x++) { + Node* n = other->_control.at(x); + if (n->is_Call()) { + result->_control.push(n); + } + } + assert(result->_control.contains(other->_end), "what?"); + assert(result->_control.contains(_begin), "what?"); + for (int x = 0; x < num_arguments(); x++) { + if (argument(x) == arg) { + // replace the toString result with the all the arguments that + // made up the other StringConcat + for (int y = 0; y < other->num_arguments(); y++) { + result->append(other->argument(y), other->mode(y)); + } + } else { + result->append(argument(x), mode(x)); + } + } + result->set_allocation(other->_begin); + result->_multiple = true; + return result; +} + + +void StringConcat::eliminate_call(CallNode* call) { + Compile* C = _stringopts->C; + CallProjections projs; + call->extract_projections(&projs, false); + if (projs.fallthrough_catchproj != NULL) { + C->gvn_replace_by(projs.fallthrough_catchproj, call->in(TypeFunc::Control)); + } + if (projs.fallthrough_memproj != NULL) { + C->gvn_replace_by(projs.fallthrough_memproj, call->in(TypeFunc::Memory)); + } + if (projs.catchall_memproj != NULL) { + C->gvn_replace_by(projs.catchall_memproj, C->top()); + } + if (projs.fallthrough_ioproj != NULL) { + C->gvn_replace_by(projs.fallthrough_ioproj, call->in(TypeFunc::I_O)); + } + if (projs.catchall_ioproj != NULL) { + C->gvn_replace_by(projs.catchall_ioproj, C->top()); + } + if (projs.catchall_catchproj != NULL) { + // EA can't cope with the partially collapsed graph this + // creates so put it on the worklist to be collapsed later. + for (SimpleDUIterator i(projs.catchall_catchproj); i.has_next(); i.next()) { + Node *use = i.get(); + int opc = use->Opcode(); + if (opc == Op_CreateEx || opc == Op_Region) { + _stringopts->record_dead_node(use); + } + } + C->gvn_replace_by(projs.catchall_catchproj, C->top()); + } + if (projs.resproj != NULL) { + C->gvn_replace_by(projs.resproj, C->top()); + } + C->gvn_replace_by(call, C->top()); +} + +void StringConcat::eliminate_initialize(InitializeNode* init) { + Compile* C = _stringopts->C; + + // Eliminate Initialize node. + assert(init->outcnt() <= 2, "only a control and memory projection expected"); + assert(init->req() <= InitializeNode::RawStores, "no pending inits"); + Node *ctrl_proj = init->proj_out(TypeFunc::Control); + if (ctrl_proj != NULL) { + C->gvn_replace_by(ctrl_proj, init->in(TypeFunc::Control)); + } + Node *mem_proj = init->proj_out(TypeFunc::Memory); + if (mem_proj != NULL) { + Node *mem = init->in(TypeFunc::Memory); + C->gvn_replace_by(mem_proj, mem); + } + C->gvn_replace_by(init, C->top()); + init->disconnect_inputs(NULL); +} + +Node_List PhaseStringOpts::collect_toString_calls() { + Node_List string_calls; + Node_List worklist; + + _visited.Clear(); + + // Prime the worklist + for (uint i = 1; i < C->root()->len(); i++) { + Node* n = C->root()->in(i); + if (n != NULL && !_visited.test_set(n->_idx)) { + worklist.push(n); + } + } + + while (worklist.size() > 0) { + Node* ctrl = worklist.pop(); + if (ctrl->is_CallStaticJava()) { + CallStaticJavaNode* csj = ctrl->as_CallStaticJava(); + ciMethod* m = csj->method(); + if (m != NULL && + (m->intrinsic_id() == vmIntrinsics::_StringBuffer_toString || + m->intrinsic_id() == vmIntrinsics::_StringBuilder_toString)) { + string_calls.push(csj); + } + } + if (ctrl->in(0) != NULL && !_visited.test_set(ctrl->in(0)->_idx)) { + worklist.push(ctrl->in(0)); + } + if (ctrl->is_Region()) { + for (uint i = 1; i < ctrl->len(); i++) { + if (ctrl->in(i) != NULL && !_visited.test_set(ctrl->in(i)->_idx)) { + worklist.push(ctrl->in(i)); + } + } + } + } + return string_calls; +} + + +StringConcat* PhaseStringOpts::build_candidate(CallStaticJavaNode* call) { + ciMethod* m = call->method(); + ciSymbol* string_sig; + ciSymbol* int_sig; + ciSymbol* char_sig; + if (m->holder() == C->env()->StringBuilder_klass()) { + string_sig = ciSymbol::String_StringBuilder_signature(); + int_sig = ciSymbol::int_StringBuilder_signature(); + char_sig = ciSymbol::char_StringBuilder_signature(); + } else if (m->holder() == C->env()->StringBuffer_klass()) { + string_sig = ciSymbol::String_StringBuffer_signature(); + int_sig = ciSymbol::int_StringBuffer_signature(); + char_sig = ciSymbol::char_StringBuffer_signature(); + } else { + return NULL; + } +#ifndef PRODUCT + if (PrintOptimizeStringConcat) { + tty->print("considering toString call in "); + call->jvms()->dump_spec(tty); tty->cr(); + } +#endif + + StringConcat* sc = new StringConcat(this, call); + + AllocateNode* alloc = NULL; + InitializeNode* init = NULL; + + // possible opportunity for StringBuilder fusion + CallStaticJavaNode* cnode = call; + while (cnode) { + Node* recv = cnode->in(TypeFunc::Parms)->uncast(); + if (recv->is_Proj()) { + recv = recv->in(0); + } + cnode = recv->isa_CallStaticJava(); + if (cnode == NULL) { + alloc = recv->isa_Allocate(); + if (alloc == NULL) { + break; + } + // Find the constructor call + Node* result = alloc->result_cast(); + if (result == NULL || !result->is_CheckCastPP()) { + // strange looking allocation +#ifndef PRODUCT + if (PrintOptimizeStringConcat) { + tty->print("giving up because allocation looks strange "); + alloc->jvms()->dump_spec(tty); tty->cr(); + } +#endif + break; + } + Node* constructor = NULL; + for (SimpleDUIterator i(result); i.has_next(); i.next()) { + CallStaticJavaNode *use = i.get()->isa_CallStaticJava(); + if (use != NULL && use->method() != NULL && + use->method()->name() == ciSymbol::object_initializer_name() && + use->method()->holder() == m->holder()) { + // Matched the constructor. + ciSymbol* sig = use->method()->signature()->as_symbol(); + if (sig == ciSymbol::void_method_signature() || + sig == ciSymbol::int_void_signature() || + sig == ciSymbol::string_void_signature()) { + if (sig == ciSymbol::string_void_signature()) { + // StringBuilder(String) so pick this up as the first argument + assert(use->in(TypeFunc::Parms + 1) != NULL, "what?"); + sc->push_string(use->in(TypeFunc::Parms + 1)); + } + // The int variant takes an initial size for the backing + // array so just treat it like the void version. + constructor = use; + } else { +#ifndef PRODUCT + if (PrintOptimizeStringConcat) { + tty->print("unexpected constructor signature: %s", sig->as_utf8()); + } +#endif + } + break; + } + } + if (constructor == NULL) { + // couldn't find constructor +#ifndef PRODUCT + if (PrintOptimizeStringConcat) { + tty->print("giving up because couldn't find constructor "); + alloc->jvms()->dump_spec(tty); + } +#endif + break; + } + + // Walked all the way back and found the constructor call so see + // if this call converted into a direct string concatenation. + sc->add_control(call); + sc->add_control(constructor); + sc->add_control(alloc); + sc->set_allocation(alloc); + if (sc->validate_control_flow()) { + return sc; + } else { + return NULL; + } + } else if (cnode->method() == NULL) { + break; + } else if (cnode->method()->holder() == m->holder() && + cnode->method()->name() == ciSymbol::append_name() && + (cnode->method()->signature()->as_symbol() == string_sig || + cnode->method()->signature()->as_symbol() == char_sig || + cnode->method()->signature()->as_symbol() == int_sig)) { + sc->add_control(cnode); + Node* arg = cnode->in(TypeFunc::Parms + 1); + if (cnode->method()->signature()->as_symbol() == int_sig) { + sc->push_int(arg); + } else if (cnode->method()->signature()->as_symbol() == char_sig) { + sc->push_char(arg); + } else { + if (arg->is_Proj() && arg->in(0)->is_CallStaticJava()) { + CallStaticJavaNode* csj = arg->in(0)->as_CallStaticJava(); + if (csj->method() != NULL && + csj->method()->holder() == C->env()->Integer_klass() && + csj->method()->name() == ciSymbol::toString_name()) { + sc->add_control(csj); + sc->push_int(csj->in(TypeFunc::Parms)); + continue; + } + } + sc->push_string(arg); + } + continue; + } else { + // some unhandled signature +#ifndef PRODUCT + if (PrintOptimizeStringConcat) { + tty->print("giving up because encountered unexpected signature "); + cnode->tf()->dump(); tty->cr(); + cnode->in(TypeFunc::Parms + 1)->dump(); + } +#endif + break; + } + } + return NULL; +} + + +PhaseStringOpts::PhaseStringOpts(PhaseGVN* gvn, Unique_Node_List*): + Phase(StringOpts), + _gvn(gvn), + _visited(Thread::current()->resource_area()) { + + assert(OptimizeStringConcat, "shouldn't be here"); + + size_table_field = C->env()->Integer_klass()->get_field_by_name(ciSymbol::make("sizeTable"), + ciSymbol::make("[I"), true); + if (size_table_field == NULL) { + // Something wrong so give up. + assert(false, "why can't we find Integer.sizeTable?"); + return; + } + + // Collect the types needed to talk about the various slices of memory + const TypeInstPtr* string_type = TypeInstPtr::make(TypePtr::NotNull, C->env()->String_klass(), + false, NULL, 0); + + const TypePtr* value_field_type = string_type->add_offset(java_lang_String::value_offset_in_bytes()); + const TypePtr* offset_field_type = string_type->add_offset(java_lang_String::offset_offset_in_bytes()); + const TypePtr* count_field_type = string_type->add_offset(java_lang_String::count_offset_in_bytes()); + + value_field_idx = C->get_alias_index(value_field_type); + count_field_idx = C->get_alias_index(count_field_type); + offset_field_idx = C->get_alias_index(offset_field_type); + char_adr_idx = C->get_alias_index(TypeAryPtr::CHARS); + + // For each locally allocated StringBuffer see if the usages can be + // collapsed into a single String construction. + + // Run through the list of allocation looking for SB.toString to see + // if it's possible to fuse the usage of the SB into a single String + // construction. + GrowableArray<StringConcat*> concats; + Node_List toStrings = collect_toString_calls(); + while (toStrings.size() > 0) { + StringConcat* sc = build_candidate(toStrings.pop()->as_CallStaticJava()); + if (sc != NULL) { + concats.push(sc); + } + } + + // try to coalesce separate concats + restart: + for (int c = 0; c < concats.length(); c++) { + StringConcat* sc = concats.at(c); + for (int i = 0; i < sc->num_arguments(); i++) { + Node* arg = sc->argument(i); + if (arg->is_Proj() && arg->in(0)->is_CallStaticJava()) { + CallStaticJavaNode* csj = arg->in(0)->as_CallStaticJava(); + if (csj->method() != NULL && + (csj->method()->holder() == C->env()->StringBuffer_klass() || + csj->method()->holder() == C->env()->StringBuilder_klass()) && + csj->method()->name() == ciSymbol::toString_name()) { + for (int o = 0; o < concats.length(); o++) { + if (c == o) continue; + StringConcat* other = concats.at(o); + if (other->end() == csj) { +#ifndef PRODUCT + if (PrintOptimizeStringConcat) { + tty->print_cr("considering stacked concats"); + } +#endif + + StringConcat* merged = sc->merge(other, arg); + if (merged->validate_control_flow()) { +#ifndef PRODUCT + if (PrintOptimizeStringConcat) { + tty->print_cr("stacking would succeed"); + } +#endif + if (c < o) { + concats.remove_at(o); + concats.at_put(c, merged); + } else { + concats.remove_at(c); + concats.at_put(o, merged); + } + goto restart; + } else { +#ifndef PRODUCT + if (PrintOptimizeStringConcat) { + tty->print_cr("stacking would fail"); + } +#endif + } + } + } + } + } + } + } + + + for (int c = 0; c < concats.length(); c++) { + StringConcat* sc = concats.at(c); + replace_string_concat(sc); + } + + remove_dead_nodes(); +} + +void PhaseStringOpts::record_dead_node(Node* dead) { + dead_worklist.push(dead); +} + +void PhaseStringOpts::remove_dead_nodes() { + // Delete any dead nodes to make things clean enough that escape + // analysis doesn't get unhappy. + while (dead_worklist.size() > 0) { + Node* use = dead_worklist.pop(); + int opc = use->Opcode(); + switch (opc) { + case Op_Region: { + uint i = 1; + for (i = 1; i < use->req(); i++) { + if (use->in(i) != C->top()) { + break; + } + } + if (i >= use->req()) { + for (SimpleDUIterator i(use); i.has_next(); i.next()) { + Node* m = i.get(); + if (m->is_Phi()) { + dead_worklist.push(m); + } + } + C->gvn_replace_by(use, C->top()); + } + break; + } + case Op_AddP: + case Op_CreateEx: { + // Recurisvely clean up references to CreateEx so EA doesn't + // get unhappy about the partially collapsed graph. + for (SimpleDUIterator i(use); i.has_next(); i.next()) { + Node* m = i.get(); + if (m->is_AddP()) { + dead_worklist.push(m); + } + } + C->gvn_replace_by(use, C->top()); + break; + } + case Op_Phi: + if (use->in(0) == C->top()) { + C->gvn_replace_by(use, C->top()); + } + break; + } + } +} + + +bool StringConcat::validate_control_flow() { + // We found all the calls and arguments now lets see if it's + // safe to transform the graph as we would expect. + + // Check to see if this resulted in too many uncommon traps previously + if (Compile::current()->too_many_traps(_begin->jvms()->method(), _begin->jvms()->bci(), + Deoptimization::Reason_intrinsic)) { + return false; + } + + // Walk backwards over the control flow from toString to the + // allocation and make sure all the control flow is ok. This + // means it's either going to be eliminated once the calls are + // removed or it can safely be transformed into an uncommon + // trap. + + int null_check_count = 0; + Unique_Node_List ctrl_path; + + assert(_control.contains(_begin), "missing"); + assert(_control.contains(_end), "missing"); + + // Collect the nodes that we know about and will eliminate into ctrl_path + for (uint i = 0; i < _control.size(); i++) { + // Push the call and it's control projection + Node* n = _control.at(i); + if (n->is_Allocate()) { + AllocateNode* an = n->as_Allocate(); + InitializeNode* init = an->initialization(); + ctrl_path.push(init); + ctrl_path.push(init->as_Multi()->proj_out(0)); + } + if (n->is_Call()) { + CallNode* cn = n->as_Call(); + ctrl_path.push(cn); + ctrl_path.push(cn->proj_out(0)); + ctrl_path.push(cn->proj_out(0)->unique_out()); + ctrl_path.push(cn->proj_out(0)->unique_out()->as_Catch()->proj_out(0)); + } else { + ShouldNotReachHere(); + } + } + + // Skip backwards through the control checking for unexpected contro flow + Node* ptr = _end; + bool fail = false; + while (ptr != _begin) { + if (ptr->is_Call() && ctrl_path.member(ptr)) { + ptr = ptr->in(0); + } else if (ptr->is_CatchProj() && ctrl_path.member(ptr)) { + ptr = ptr->in(0)->in(0)->in(0); + assert(ctrl_path.member(ptr), "should be a known piece of control"); + } else if (ptr->is_IfTrue()) { + IfNode* iff = ptr->in(0)->as_If(); + BoolNode* b = iff->in(1)->isa_Bool(); + Node* cmp = b->in(1); + Node* v1 = cmp->in(1); + Node* v2 = cmp->in(2); + Node* otherproj = iff->proj_out(1 - ptr->as_Proj()->_con); + + // Null check of the return of append which can simply be eliminated + if (b->_test._test == BoolTest::ne && + v2->bottom_type() == TypePtr::NULL_PTR && + v1->is_Proj() && ctrl_path.member(v1->in(0))) { + // NULL check of the return value of the append + null_check_count++; + if (otherproj->outcnt() == 1) { + CallStaticJavaNode* call = otherproj->unique_out()->isa_CallStaticJava(); + if (call != NULL && call->_name != NULL && strcmp(call->_name, "uncommon_trap") == 0) { + ctrl_path.push(call); + } + } + _control.push(ptr); + ptr = ptr->in(0)->in(0); + continue; + } + + // A test which leads to an uncommon trap which should be safe. + // Later this trap will be converted into a trap that restarts + // at the beginning. + if (otherproj->outcnt() == 1) { + CallStaticJavaNode* call = otherproj->unique_out()->isa_CallStaticJava(); + if (call != NULL && call->_name != NULL && strcmp(call->_name, "uncommon_trap") == 0) { + // control flow leads to uct so should be ok + _uncommon_traps.push(call); + ctrl_path.push(call); + ptr = ptr->in(0)->in(0); + continue; + } + } + +#ifndef PRODUCT + // Some unexpected control flow we don't know how to handle. + if (PrintOptimizeStringConcat) { + tty->print_cr("failing with unknown test"); + b->dump(); + cmp->dump(); + v1->dump(); + v2->dump(); + tty->cr(); + } +#endif + break; + } else if (ptr->is_Proj() && ptr->in(0)->is_Initialize()) { + ptr = ptr->in(0)->in(0); + } else if (ptr->is_Region()) { + Node* copy = ptr->as_Region()->is_copy(); + if (copy != NULL) { + ptr = copy; + continue; + } + if (ptr->req() == 3 && + ptr->in(1) != NULL && ptr->in(1)->is_Proj() && + ptr->in(2) != NULL && ptr->in(2)->is_Proj() && + ptr->in(1)->in(0) == ptr->in(2)->in(0) && + ptr->in(1)->in(0) != NULL && ptr->in(1)->in(0)->is_If()) { + // Simple diamond. + // XXX should check for possibly merging stores. simple data merges are ok. + ptr = ptr->in(1)->in(0)->in(0); + continue; + } +#ifndef PRODUCT + if (PrintOptimizeStringConcat) { + tty->print_cr("fusion would fail for region"); + _begin->dump(); + ptr->dump(2); + } +#endif + fail = true; + break; + } else { + // other unknown control + if (!fail) { +#ifndef PRODUCT + if (PrintOptimizeStringConcat) { + tty->print_cr("fusion would fail for"); + _begin->dump(); + } +#endif + fail = true; + } +#ifndef PRODUCT + if (PrintOptimizeStringConcat) { + ptr->dump(); + } +#endif + ptr = ptr->in(0); + } + } +#ifndef PRODUCT + if (PrintOptimizeStringConcat && fail) { + tty->cr(); + } +#endif + if (fail) return !fail; + + // Validate that all these results produced are contained within + // this cluster of objects. First collect all the results produced + // by calls in the region. + _stringopts->_visited.Clear(); + Node_List worklist; + Node* final_result = _end->proj_out(TypeFunc::Parms); + for (uint i = 0; i < _control.size(); i++) { + CallNode* cnode = _control.at(i)->isa_Call(); + if (cnode != NULL) { + _stringopts->_visited.test_set(cnode->_idx); + } + Node* result = cnode != NULL ? cnode->proj_out(TypeFunc::Parms) : NULL; + if (result != NULL && result != final_result) { + worklist.push(result); + } + } + + Node* last_result = NULL; + while (worklist.size() > 0) { + Node* result = worklist.pop(); + if (_stringopts->_visited.test_set(result->_idx)) + continue; + for (SimpleDUIterator i(result); i.has_next(); i.next()) { + Node *use = i.get(); + if (ctrl_path.member(use)) { + // already checked this + continue; + } + int opc = use->Opcode(); + if (opc == Op_CmpP || opc == Op_Node) { + ctrl_path.push(use); + continue; + } + if (opc == Op_CastPP || opc == Op_CheckCastPP) { + for (SimpleDUIterator j(use); j.has_next(); j.next()) { + worklist.push(j.get()); + } + worklist.push(use->in(1)); + ctrl_path.push(use); + continue; + } +#ifndef PRODUCT + if (PrintOptimizeStringConcat) { + if (result != last_result) { + last_result = result; + tty->print_cr("extra uses for result:"); + last_result->dump(); + } + use->dump(); + } +#endif + fail = true; + break; + } + } + +#ifndef PRODUCT + if (PrintOptimizeStringConcat && !fail) { + ttyLocker ttyl; + tty->cr(); + tty->print("fusion would succeed (%d %d) for ", null_check_count, _uncommon_traps.size()); + _begin->jvms()->dump_spec(tty); tty->cr(); + for (int i = 0; i < num_arguments(); i++) { + argument(i)->dump(); + } + _control.dump(); + tty->cr(); + } +#endif + + return !fail; +} + +Node* PhaseStringOpts::fetch_static_field(GraphKit& kit, ciField* field) { + const TypeKlassPtr* klass_type = TypeKlassPtr::make(field->holder()); + Node* klass_node = __ makecon(klass_type); + BasicType bt = field->layout_type(); + ciType* field_klass = field->type(); + + const Type *type; + if( bt == T_OBJECT ) { + if (!field->type()->is_loaded()) { + type = TypeInstPtr::BOTTOM; + } else if (field->is_constant()) { + // This can happen if the constant oop is non-perm. + ciObject* con = field->constant_value().as_object(); + // Do not "join" in the previous type; it doesn't add value, + // and may yield a vacuous result if the field is of interface type. + type = TypeOopPtr::make_from_constant(con)->isa_oopptr(); + assert(type != NULL, "field singleton type must be consistent"); + } else { + type = TypeOopPtr::make_from_klass(field_klass->as_klass()); + } + } else { + type = Type::get_const_basic_type(bt); + } + + return kit.make_load(NULL, kit.basic_plus_adr(klass_node, field->offset_in_bytes()), + type, T_OBJECT, + C->get_alias_index(klass_type->add_offset(field->offset_in_bytes()))); +} + +Node* PhaseStringOpts::int_stringSize(GraphKit& kit, Node* arg) { + RegionNode *final_merge = new (C, 3) RegionNode(3); + kit.gvn().set_type(final_merge, Type::CONTROL); + Node* final_size = new (C, 3) PhiNode(final_merge, TypeInt::INT); + kit.gvn().set_type(final_size, TypeInt::INT); + + IfNode* iff = kit.create_and_map_if(kit.control(), + __ Bool(__ CmpI(arg, __ intcon(0x80000000)), BoolTest::ne), + PROB_FAIR, COUNT_UNKNOWN); + Node* is_min = __ IfFalse(iff); + final_merge->init_req(1, is_min); + final_size->init_req(1, __ intcon(11)); + + kit.set_control(__ IfTrue(iff)); + if (kit.stopped()) { + final_merge->init_req(2, C->top()); + final_size->init_req(2, C->top()); + } else { + + // int size = (i < 0) ? stringSize(-i) + 1 : stringSize(i); + RegionNode *r = new (C, 3) RegionNode(3); + kit.gvn().set_type(r, Type::CONTROL); + Node *phi = new (C, 3) PhiNode(r, TypeInt::INT); + kit.gvn().set_type(phi, TypeInt::INT); + Node *size = new (C, 3) PhiNode(r, TypeInt::INT); + kit.gvn().set_type(size, TypeInt::INT); + Node* chk = __ CmpI(arg, __ intcon(0)); + Node* p = __ Bool(chk, BoolTest::lt); + IfNode* iff = kit.create_and_map_if(kit.control(), p, PROB_FAIR, COUNT_UNKNOWN); + Node* lessthan = __ IfTrue(iff); + Node* greaterequal = __ IfFalse(iff); + r->init_req(1, lessthan); + phi->init_req(1, __ SubI(__ intcon(0), arg)); + size->init_req(1, __ intcon(1)); + r->init_req(2, greaterequal); + phi->init_req(2, arg); + size->init_req(2, __ intcon(0)); + kit.set_control(r); + C->record_for_igvn(r); + C->record_for_igvn(phi); + C->record_for_igvn(size); + + // for (int i=0; ; i++) + // if (x <= sizeTable[i]) + // return i+1; + RegionNode *loop = new (C, 3) RegionNode(3); + loop->init_req(1, kit.control()); + kit.gvn().set_type(loop, Type::CONTROL); + + Node *index = new (C, 3) PhiNode(loop, TypeInt::INT); + index->init_req(1, __ intcon(0)); + kit.gvn().set_type(index, TypeInt::INT); + kit.set_control(loop); + Node* sizeTable = fetch_static_field(kit, size_table_field); + + Node* value = kit.load_array_element(NULL, sizeTable, index, TypeAryPtr::INTS); + C->record_for_igvn(value); + Node* limit = __ CmpI(phi, value); + Node* limitb = __ Bool(limit, BoolTest::le); + IfNode* iff2 = kit.create_and_map_if(kit.control(), limitb, PROB_MIN, COUNT_UNKNOWN); + Node* lessEqual = __ IfTrue(iff2); + Node* greater = __ IfFalse(iff2); + + loop->init_req(2, greater); + index->init_req(2, __ AddI(index, __ intcon(1))); + + kit.set_control(lessEqual); + C->record_for_igvn(loop); + C->record_for_igvn(index); + + final_merge->init_req(2, kit.control()); + final_size->init_req(2, __ AddI(__ AddI(index, size), __ intcon(1))); + } + + kit.set_control(final_merge); + C->record_for_igvn(final_merge); + C->record_for_igvn(final_size); + + return final_size; +} + +void PhaseStringOpts::int_getChars(GraphKit& kit, Node* arg, Node* char_array, Node* start, Node* end) { + RegionNode *final_merge = new (C, 4) RegionNode(4); + kit.gvn().set_type(final_merge, Type::CONTROL); + Node *final_mem = PhiNode::make(final_merge, kit.memory(char_adr_idx), Type::MEMORY, TypeAryPtr::CHARS); + kit.gvn().set_type(final_mem, Type::MEMORY); + + // need to handle Integer.MIN_VALUE specially because negating doesn't make it positive + { + // i == MIN_VALUE + IfNode* iff = kit.create_and_map_if(kit.control(), + __ Bool(__ CmpI(arg, __ intcon(0x80000000)), BoolTest::ne), + PROB_FAIR, COUNT_UNKNOWN); + + Node* old_mem = kit.memory(char_adr_idx); + + kit.set_control(__ IfFalse(iff)); + if (kit.stopped()) { + // Statically not equal to MIN_VALUE so this path is dead + final_merge->init_req(3, kit.control()); + } else { + copy_string(kit, __ makecon(TypeInstPtr::make(C->env()->the_min_jint_string())), + char_array, start); + final_merge->init_req(3, kit.control()); + final_mem->init_req(3, kit.memory(char_adr_idx)); + } + + kit.set_control(__ IfTrue(iff)); + kit.set_memory(old_mem, char_adr_idx); + } + + + // Simplified version of Integer.getChars + + // int q, r; + // int charPos = index; + Node* charPos = end; + + // char sign = 0; + + Node* i = arg; + Node* sign = __ intcon(0); + + // if (i < 0) { + // sign = '-'; + // i = -i; + // } + { + IfNode* iff = kit.create_and_map_if(kit.control(), + __ Bool(__ CmpI(arg, __ intcon(0)), BoolTest::lt), + PROB_FAIR, COUNT_UNKNOWN); + + RegionNode *merge = new (C, 3) RegionNode(3); + kit.gvn().set_type(merge, Type::CONTROL); + i = new (C, 3) PhiNode(merge, TypeInt::INT); + kit.gvn().set_type(i, TypeInt::INT); + sign = new (C, 3) PhiNode(merge, TypeInt::INT); + kit.gvn().set_type(sign, TypeInt::INT); + + merge->init_req(1, __ IfTrue(iff)); + i->init_req(1, __ SubI(__ intcon(0), arg)); + sign->init_req(1, __ intcon('-')); + merge->init_req(2, __ IfFalse(iff)); + i->init_req(2, arg); + sign->init_req(2, __ intcon(0)); + + kit.set_control(merge); + + C->record_for_igvn(merge); + C->record_for_igvn(i); + C->record_for_igvn(sign); + } + + // for (;;) { + // q = i / 10; + // r = i - ((q << 3) + (q << 1)); // r = i-(q*10) ... + // buf [--charPos] = digits [r]; + // i = q; + // if (i == 0) break; + // } + + { + RegionNode *head = new (C, 3) RegionNode(3); + head->init_req(1, kit.control()); + kit.gvn().set_type(head, Type::CONTROL); + Node *i_phi = new (C, 3) PhiNode(head, TypeInt::INT); + i_phi->init_req(1, i); + kit.gvn().set_type(i_phi, TypeInt::INT); + charPos = PhiNode::make(head, charPos); + kit.gvn().set_type(charPos, TypeInt::INT); + Node *mem = PhiNode::make(head, kit.memory(char_adr_idx), Type::MEMORY, TypeAryPtr::CHARS); + kit.gvn().set_type(mem, Type::MEMORY); + kit.set_control(head); + kit.set_memory(mem, char_adr_idx); + + Node* q = __ DivI(NULL, i_phi, __ intcon(10)); + Node* r = __ SubI(i_phi, __ AddI(__ LShiftI(q, __ intcon(3)), + __ LShiftI(q, __ intcon(1)))); + Node* m1 = __ SubI(charPos, __ intcon(1)); + Node* ch = __ AddI(r, __ intcon('0')); + + Node* st = __ store_to_memory(kit.control(), kit.array_element_address(char_array, m1, T_CHAR), + ch, T_CHAR, char_adr_idx); + + + IfNode* iff = kit.create_and_map_if(head, __ Bool(__ CmpI(q, __ intcon(0)), BoolTest::ne), + PROB_FAIR, COUNT_UNKNOWN); + Node* ne = __ IfTrue(iff); + Node* eq = __ IfFalse(iff); + + head->init_req(2, ne); + mem->init_req(2, st); + i_phi->init_req(2, q); + charPos->init_req(2, m1); + + charPos = m1; + + kit.set_control(eq); + kit.set_memory(st, char_adr_idx); + + C->record_for_igvn(head); + C->record_for_igvn(mem); + C->record_for_igvn(i_phi); + C->record_for_igvn(charPos); + } + + { + // if (sign != 0) { + // buf [--charPos] = sign; + // } + IfNode* iff = kit.create_and_map_if(kit.control(), + __ Bool(__ CmpI(sign, __ intcon(0)), BoolTest::ne), + PROB_FAIR, COUNT_UNKNOWN); + + final_merge->init_req(2, __ IfFalse(iff)); + final_mem->init_req(2, kit.memory(char_adr_idx)); + + kit.set_control(__ IfTrue(iff)); + if (kit.stopped()) { + final_merge->init_req(1, C->top()); + final_mem->init_req(1, C->top()); + } else { + Node* m1 = __ SubI(charPos, __ intcon(1)); + Node* st = __ store_to_memory(kit.control(), kit.array_element_address(char_array, m1, T_CHAR), + sign, T_CHAR, char_adr_idx); + + final_merge->init_req(1, kit.control()); + final_mem->init_req(1, st); + } + + kit.set_control(final_merge); + kit.set_memory(final_mem, char_adr_idx); + + C->record_for_igvn(final_merge); + C->record_for_igvn(final_mem); + } +} + + +Node* PhaseStringOpts::copy_string(GraphKit& kit, Node* str, Node* char_array, Node* start) { + Node* string = str; + Node* offset = kit.make_load(NULL, + kit.basic_plus_adr(string, string, java_lang_String::offset_offset_in_bytes()), + TypeInt::INT, T_INT, offset_field_idx); + Node* count = kit.make_load(NULL, + kit.basic_plus_adr(string, string, java_lang_String::count_offset_in_bytes()), + TypeInt::INT, T_INT, count_field_idx); + const TypeAryPtr* value_type = TypeAryPtr::make(TypePtr::NotNull, + TypeAry::make(TypeInt::CHAR,TypeInt::POS), + ciTypeArrayKlass::make(T_CHAR), true, 0); + Node* value = kit.make_load(NULL, + kit.basic_plus_adr(string, string, java_lang_String::value_offset_in_bytes()), + value_type, T_OBJECT, value_field_idx); + + // copy the contents + if (offset->is_Con() && count->is_Con() && value->is_Con() && count->get_int() < unroll_string_copy_length) { + // For small constant strings just emit individual stores. + // A length of 6 seems like a good space/speed tradeof. + int c = count->get_int(); + int o = offset->get_int(); + const TypeOopPtr* t = kit.gvn().type(value)->isa_oopptr(); + ciTypeArray* value_array = t->const_oop()->as_type_array(); + for (int e = 0; e < c; e++) { + __ store_to_memory(kit.control(), kit.array_element_address(char_array, start, T_CHAR), + __ intcon(value_array->char_at(o + e)), T_CHAR, char_adr_idx); + start = __ AddI(start, __ intcon(1)); + } + } else { + Node* src_ptr = kit.array_element_address(value, offset, T_CHAR); + Node* dst_ptr = kit.array_element_address(char_array, start, T_CHAR); + Node* c = count; + Node* extra = NULL; +#ifdef _LP64 + c = __ ConvI2L(c); + extra = C->top(); +#endif + Node* call = kit.make_runtime_call(GraphKit::RC_LEAF|GraphKit::RC_NO_FP, + OptoRuntime::fast_arraycopy_Type(), + CAST_FROM_FN_PTR(address, StubRoutines::jshort_disjoint_arraycopy()), + "jshort_disjoint_arraycopy", TypeAryPtr::CHARS, + src_ptr, dst_ptr, c, extra); + start = __ AddI(start, count); + } + return start; +} + + +void PhaseStringOpts::replace_string_concat(StringConcat* sc) { + // Log a little info about the transformation + sc->maybe_log_transform(); + + // pull the JVMState of the allocation into a SafePointNode to serve as + // as a shim for the insertion of the new code. + JVMState* jvms = sc->begin()->jvms()->clone_shallow(C); + uint size = sc->begin()->req(); + SafePointNode* map = new (C, size) SafePointNode(size, jvms); + + // copy the control and memory state from the final call into our + // new starting state. This allows any preceeding tests to feed + // into the new section of code. + for (uint i1 = 0; i1 < TypeFunc::Parms; i1++) { + map->init_req(i1, sc->end()->in(i1)); + } + // blow away old allocation arguments + for (uint i1 = TypeFunc::Parms; i1 < jvms->debug_start(); i1++) { + map->init_req(i1, C->top()); + } + // Copy the rest of the inputs for the JVMState + for (uint i1 = jvms->debug_start(); i1 < sc->begin()->req(); i1++) { + map->init_req(i1, sc->begin()->in(i1)); + } + // Make sure the memory state is a MergeMem for parsing. + if (!map->in(TypeFunc::Memory)->is_MergeMem()) { + map->set_req(TypeFunc::Memory, MergeMemNode::make(C, map->in(TypeFunc::Memory))); + } + + jvms->set_map(map); + map->ensure_stack(jvms, jvms->method()->max_stack()); + + + // disconnect all the old StringBuilder calls from the graph + sc->eliminate_unneeded_control(); + + // At this point all the old work has been completely removed from + // the graph and the saved JVMState exists at the point where the + // final toString call used to be. + GraphKit kit(jvms); + + // There may be uncommon traps which are still using the + // intermediate states and these need to be rewritten to point at + // the JVMState at the beginning of the transformation. + sc->convert_uncommon_traps(kit, jvms); + + // Now insert the logic to compute the size of the string followed + // by all the logic to construct array and resulting string. + + Node* null_string = __ makecon(TypeInstPtr::make(C->env()->the_null_string())); + + // Create a region for the overflow checks to merge into. + int args = MAX2(sc->num_arguments(), 1); + RegionNode* overflow = new (C, args) RegionNode(args); + kit.gvn().set_type(overflow, Type::CONTROL); + + // Create a hook node to hold onto the individual sizes since they + // are need for the copying phase. + Node* string_sizes = new (C, args) Node(args); + + Node* length = __ intcon(0); + for (int argi = 0; argi < sc->num_arguments(); argi++) { + Node* arg = sc->argument(argi); + switch (sc->mode(argi)) { + case StringConcat::IntMode: { + Node* string_size = int_stringSize(kit, arg); + + // accumulate total + length = __ AddI(length, string_size); + + // Cache this value for the use by int_toString + string_sizes->init_req(argi, string_size); + break; + } + case StringConcat::StringMode: { + const Type* type = kit.gvn().type(arg); + if (type == TypePtr::NULL_PTR) { + // replace the argument with the null checked version + arg = null_string; + sc->set_argument(argi, arg); + } else if (!type->higher_equal(TypeInstPtr::NOTNULL)) { + // s = s != null ? s : "null"; + // length = length + (s.count - s.offset); + RegionNode *r = new (C, 3) RegionNode(3); + kit.gvn().set_type(r, Type::CONTROL); + Node *phi = new (C, 3) PhiNode(r, type); + kit.gvn().set_type(phi, phi->bottom_type()); + Node* p = __ Bool(__ CmpP(arg, kit.null()), BoolTest::ne); + IfNode* iff = kit.create_and_map_if(kit.control(), p, PROB_MIN, COUNT_UNKNOWN); + Node* notnull = __ IfTrue(iff); + Node* isnull = __ IfFalse(iff); + kit.set_control(notnull); // set control for the cast_not_null + r->init_req(1, notnull); + phi->init_req(1, kit.cast_not_null(arg, false)); + r->init_req(2, isnull); + phi->init_req(2, null_string); + kit.set_control(r); + C->record_for_igvn(r); + C->record_for_igvn(phi); + // replace the argument with the null checked version + arg = phi; + sc->set_argument(argi, arg); + } + // Node* offset = kit.make_load(NULL, kit.basic_plus_adr(arg, arg, offset_offset), + // TypeInt::INT, T_INT, offset_field_idx); + Node* count = kit.make_load(NULL, kit.basic_plus_adr(arg, arg, java_lang_String::count_offset_in_bytes()), + TypeInt::INT, T_INT, count_field_idx); + length = __ AddI(length, count); + string_sizes->init_req(argi, NULL); + break; + } + case StringConcat::CharMode: { + // one character only + length = __ AddI(length, __ intcon(1)); + break; + } + default: + ShouldNotReachHere(); + } + if (argi > 0) { + // Check that the sum hasn't overflowed + IfNode* iff = kit.create_and_map_if(kit.control(), + __ Bool(__ CmpI(length, __ intcon(0)), BoolTest::lt), + PROB_MIN, COUNT_UNKNOWN); + kit.set_control(__ IfFalse(iff)); + overflow->set_req(argi, __ IfTrue(iff)); + } + } + + { + // Hook + PreserveJVMState pjvms(&kit); + kit.set_control(overflow); + kit.uncommon_trap(Deoptimization::Reason_intrinsic, + Deoptimization::Action_make_not_entrant); + } + + // length now contains the number of characters needed for the + // char[] so create a new AllocateArray for the char[] + Node* char_array = NULL; + { + PreserveReexecuteState preexecs(&kit); + // The original jvms is for an allocation of either a String or + // StringBuffer so no stack adjustment is necessary for proper + // reexecution. If we deoptimize in the slow path the bytecode + // will be reexecuted and the char[] allocation will be thrown away. + kit.jvms()->set_should_reexecute(true); + char_array = kit.new_array(__ makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_CHAR))), + length, 1); + } + + // Mark the allocation so that zeroing is skipped since the code + // below will overwrite the entire array + AllocateArrayNode* char_alloc = AllocateArrayNode::Ideal_array_allocation(char_array, _gvn); + char_alloc->maybe_set_complete(_gvn); + + // Now copy the string representations into the final char[] + Node* start = __ intcon(0); + for (int argi = 0; argi < sc->num_arguments(); argi++) { + Node* arg = sc->argument(argi); + switch (sc->mode(argi)) { + case StringConcat::IntMode: { + Node* end = __ AddI(start, string_sizes->in(argi)); + // getChars words backwards so pass the ending point as well as the start + int_getChars(kit, arg, char_array, start, end); + start = end; + break; + } + case StringConcat::StringMode: { + start = copy_string(kit, arg, char_array, start); + break; + } + case StringConcat::CharMode: { + __ store_to_memory(kit.control(), kit.array_element_address(char_array, start, T_CHAR), + arg, T_CHAR, char_adr_idx); + start = __ AddI(start, __ intcon(1)); + break; + } + default: + ShouldNotReachHere(); + } + } + + // If we're not reusing an existing String allocation then allocate one here. + Node* result = sc->string_alloc(); + if (result == NULL) { + PreserveReexecuteState preexecs(&kit); + // The original jvms is for an allocation of either a String or + // StringBuffer so no stack adjustment is necessary for proper + // reexecution. + kit.jvms()->set_should_reexecute(true); + result = kit.new_instance(__ makecon(TypeKlassPtr::make(C->env()->String_klass()))); + } + + // Intialize the string + kit.store_to_memory(kit.control(), kit.basic_plus_adr(result, java_lang_String::offset_offset_in_bytes()), + __ intcon(0), T_INT, offset_field_idx); + kit.store_to_memory(kit.control(), kit.basic_plus_adr(result, java_lang_String::count_offset_in_bytes()), + length, T_INT, count_field_idx); + kit.store_to_memory(kit.control(), kit.basic_plus_adr(result, java_lang_String::value_offset_in_bytes()), + char_array, T_OBJECT, value_field_idx); + + // hook up the outgoing control and result + kit.replace_call(sc->end(), result); + + // Unhook any hook nodes + string_sizes->disconnect_inputs(NULL); + sc->cleanup(); +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/share/vm/opto/stringopts.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,83 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +class StringConcat; + +class PhaseStringOpts : public Phase { + friend class StringConcat; + + private: + PhaseGVN* _gvn; + + // List of dead nodes to clean up aggressively at the end + Unique_Node_List dead_worklist; + + // Memory slices needed for code gen + int char_adr_idx; + int value_field_idx; + int count_field_idx; + int offset_field_idx; + + // Integer.sizeTable - used for int to String conversion + ciField* size_table_field; + + // A set for use by various stages + VectorSet _visited; + + // Collect a list of all SB.toString calls + Node_List collect_toString_calls(); + + // Examine the use of the SB alloc to see if it can be replace with + // a single string construction. + StringConcat* build_candidate(CallStaticJavaNode* call); + + // Replace all the SB calls in concat with an optimization String allocation + void replace_string_concat(StringConcat* concat); + + // Load the value of a static field, performing any constant folding. + Node* fetch_static_field(GraphKit& kit, ciField* field); + + // Compute the number of characters required to represent the int value + Node* int_stringSize(GraphKit& kit, Node* value); + + // Copy the characters representing value into char_array starting at start + void int_getChars(GraphKit& kit, Node* value, Node* char_array, Node* start, Node* end); + + // Copy of the contents of the String str into char_array starting at index start. + Node* copy_string(GraphKit& kit, Node* str, Node* char_array, Node* start); + + // Clean up any leftover nodes + void record_dead_node(Node* node); + void remove_dead_nodes(); + + PhaseGVN* gvn() { return _gvn; } + + enum { + // max length of constant string copy unrolling in copy_string + unroll_string_copy_length = 6 + }; + + public: + PhaseStringOpts(PhaseGVN* gvn, Unique_Node_List* worklist); +};
--- a/src/share/vm/opto/subnode.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/subnode.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -1244,8 +1244,7 @@ if( t1 == Type::TOP ) return Type::TOP; if( t1->base() != Type::DoubleCon ) return Type::DOUBLE; double d = t1->getd(); - if( d < 0.0 ) return Type::DOUBLE; - return TypeD::make( SharedRuntime::dcos( d ) ); + return TypeD::make( StubRoutines::intrinsic_cos( d ) ); } //============================================================================= @@ -1256,8 +1255,7 @@ if( t1 == Type::TOP ) return Type::TOP; if( t1->base() != Type::DoubleCon ) return Type::DOUBLE; double d = t1->getd(); - if( d < 0.0 ) return Type::DOUBLE; - return TypeD::make( SharedRuntime::dsin( d ) ); + return TypeD::make( StubRoutines::intrinsic_sin( d ) ); } //============================================================================= @@ -1268,8 +1266,7 @@ if( t1 == Type::TOP ) return Type::TOP; if( t1->base() != Type::DoubleCon ) return Type::DOUBLE; double d = t1->getd(); - if( d < 0.0 ) return Type::DOUBLE; - return TypeD::make( SharedRuntime::dtan( d ) ); + return TypeD::make( StubRoutines::intrinsic_tan( d ) ); } //============================================================================= @@ -1280,8 +1277,7 @@ if( t1 == Type::TOP ) return Type::TOP; if( t1->base() != Type::DoubleCon ) return Type::DOUBLE; double d = t1->getd(); - if( d < 0.0 ) return Type::DOUBLE; - return TypeD::make( SharedRuntime::dlog( d ) ); + return TypeD::make( StubRoutines::intrinsic_log( d ) ); } //============================================================================= @@ -1292,8 +1288,7 @@ if( t1 == Type::TOP ) return Type::TOP; if( t1->base() != Type::DoubleCon ) return Type::DOUBLE; double d = t1->getd(); - if( d < 0.0 ) return Type::DOUBLE; - return TypeD::make( SharedRuntime::dlog10( d ) ); + return TypeD::make( StubRoutines::intrinsic_log10( d ) ); } //============================================================================= @@ -1304,8 +1299,7 @@ if( t1 == Type::TOP ) return Type::TOP; if( t1->base() != Type::DoubleCon ) return Type::DOUBLE; double d = t1->getd(); - if( d < 0.0 ) return Type::DOUBLE; - return TypeD::make( SharedRuntime::dexp( d ) ); + return TypeD::make( StubRoutines::intrinsic_exp( d ) ); } @@ -1323,5 +1317,5 @@ double d2 = t2->getd(); if( d1 < 0.0 ) return Type::DOUBLE; if( d2 < 0.0 ) return Type::DOUBLE; - return TypeD::make( SharedRuntime::dpow( d1, d2 ) ); + return TypeD::make( StubRoutines::intrinsic_pow( d1, d2 ) ); }
--- a/src/share/vm/opto/superword.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/superword.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1921,6 +1921,11 @@ } // Match AddP(base, AddP(ptr, k*iv [+ invariant]), constant) Node* base = adr->in(AddPNode::Base); + //unsafe reference could not be aligned appropriately without runtime checking + if (base == NULL || base->bottom_type() == Type::TOP) { + assert(!valid(), "unsafe access"); + return; + } for (int i = 0; i < 3; i++) { if (!scaled_iv_plus_offset(adr->in(AddPNode::Offset))) { assert(!valid(), "too complex");
--- a/src/share/vm/opto/type.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/type.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -2431,7 +2431,7 @@ //------------------------------make_from_constant----------------------------- // Make a java pointer from an oop constant const TypeOopPtr* TypeOopPtr::make_from_constant(ciObject* o, bool require_constant) { - if (o->is_method_data() || o->is_method()) { + if (o->is_method_data() || o->is_method() || o->is_cpcache()) { // Treat much like a typeArray of bytes, like below, but fake the type... const Type* etype = (Type*)get_const_basic_type(T_BYTE); const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS); @@ -3969,7 +3969,7 @@ const TypeFunc* tf = C->last_tf(method); // check cache if (tf != NULL) return tf; // The hit rate here is almost 50%. const TypeTuple *domain; - if (method->flags().is_static()) { + if (method->is_static()) { domain = TypeTuple::make_domain(NULL, method->signature()); } else { domain = TypeTuple::make_domain(method->holder(), method->signature());
--- a/src/share/vm/opto/type.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/opto/type.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -847,9 +847,6 @@ // Constant pointer to array static const TypeAryPtr *make( PTR ptr, ciObject* o, const TypeAry *ary, ciKlass* k, bool xk, int offset, int instance_id = InstanceBot); - // Convenience - static const TypeAryPtr *make(ciObject* o); - // Return a 'ptr' version of this type virtual const Type *cast_to_ptr_type(PTR ptr) const;
--- a/src/share/vm/prims/jni.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/jni.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -396,11 +396,11 @@ oop mirror = NULL; int slot = 0; - if (reflected->klass() == SystemDictionary::reflect_constructor_klass()) { + if (reflected->klass() == SystemDictionary::reflect_Constructor_klass()) { mirror = java_lang_reflect_Constructor::clazz(reflected); slot = java_lang_reflect_Constructor::slot(reflected); } else { - assert(reflected->klass() == SystemDictionary::reflect_method_klass(), "wrong type"); + assert(reflected->klass() == SystemDictionary::reflect_Method_klass(), "wrong type"); mirror = java_lang_reflect_Method::clazz(reflected); slot = java_lang_reflect_Method::slot(reflected); } @@ -496,7 +496,7 @@ klassOop super = Klass::cast(k)->java_super(); // super2 is the value computed by the compiler's getSuperClass intrinsic: debug_only(klassOop super2 = ( Klass::cast(k)->oop_is_javaArray() - ? SystemDictionary::object_klass() + ? SystemDictionary::Object_klass() : Klass::cast(k)->super() ) ); assert(super == super2, "java_super computation depends on interface, array, other super"); @@ -584,7 +584,7 @@ if (thread->has_pending_exception()) { Handle ex(thread, thread->pending_exception()); thread->clear_pending_exception(); - if (ex->is_a(SystemDictionary::threaddeath_klass())) { + if (ex->is_a(SystemDictionary::ThreadDeath_klass())) { // Don't print anything if we are being killed. } else { jio_fprintf(defaultStream::error_stream(), "Exception "); @@ -593,12 +593,12 @@ jio_fprintf(defaultStream::error_stream(), "in thread \"%s\" ", thread->get_thread_name()); } - if (ex->is_a(SystemDictionary::throwable_klass())) { + if (ex->is_a(SystemDictionary::Throwable_klass())) { JavaValue result(T_VOID); JavaCalls::call_virtual(&result, ex, KlassHandle(THREAD, - SystemDictionary::throwable_klass()), + SystemDictionary::Throwable_klass()), vmSymbolHandles::printStackTrace_name(), vmSymbolHandles::void_method_signature(), THREAD); @@ -3231,6 +3231,21 @@ jint result = JNI_ERR; DT_RETURN_MARK(CreateJavaVM, jint, (const jint&)result); + // We're about to use Atomic::xchg for synchronization. Some Zero + // platforms use the GCC builtin __sync_lock_test_and_set for this, + // but __sync_lock_test_and_set is not guaranteed to do what we want + // on all architectures. So we check it works before relying on it. +#if defined(ZERO) && defined(ASSERT) + { + jint a = 0xcafebabe; + jint b = Atomic::xchg(0xdeadbeef, &a); + void *c = &a; + void *d = Atomic::xchg_ptr(&b, &c); + assert(a == (jint) 0xdeadbeef && b == (jint) 0xcafebabe, "Atomic::xchg() works"); + assert(c == &b && d == &a, "Atomic::xchg_ptr() works"); + } +#endif // ZERO && ASSERT + // At the moment it's only possible to have one Java VM, // since some of the runtime state is in global variables.
--- a/src/share/vm/prims/jniCheck.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/jniCheck.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -341,7 +341,7 @@ ReportJNIFatalError(thr, fatal_received_null_class); } - if (mirror->klass() != SystemDictionary::class_klass()) { + if (mirror->klass() != SystemDictionary::Class_klass()) { ReportJNIFatalError(thr, fatal_class_not_a_class); } @@ -358,7 +358,7 @@ assert(klass != NULL, "klass argument must have a value"); if (!Klass::cast(klass)->oop_is_instance() || - !instanceKlass::cast(klass)->is_subclass_of(SystemDictionary::throwable_klass())) { + !instanceKlass::cast(klass)->is_subclass_of(SystemDictionary::Throwable_klass())) { ReportJNIFatalError(thr, fatal_class_not_a_throwable_class); } }
--- a/src/share/vm/prims/jvm.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/jvm.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -80,7 +80,7 @@ while (!vfst.at_end()) { methodOop m = vfst.method(); - if (!vfst.method()->method_holder()->klass_part()->is_subclass_of(SystemDictionary::classloader_klass())&& + if (!vfst.method()->method_holder()->klass_part()->is_subclass_of(SystemDictionary::ClassLoader_klass())&& !vfst.method()->method_holder()->klass_part()->is_subclass_of(access_controller_klass) && !vfst.method()->method_holder()->klass_part()->is_subclass_of(privileged_action_klass)) { break; @@ -257,7 +257,7 @@ Handle value_str = java_lang_String::create_from_platform_dependent_str((value != NULL ? value : ""), CHECK); JavaCalls::call_virtual(&r, props, - KlassHandle(THREAD, SystemDictionary::properties_klass()), + KlassHandle(THREAD, SystemDictionary::Properties_klass()), vmSymbolHandles::put_name(), vmSymbolHandles::object_object_object_signature(), key_str, @@ -495,7 +495,7 @@ guarantee(klass->is_cloneable(), "all arrays are cloneable"); } else { guarantee(obj->is_instance(), "should be instanceOop"); - bool cloneable = klass->is_subtype_of(SystemDictionary::cloneable_klass()); + bool cloneable = klass->is_subtype_of(SystemDictionary::Cloneable_klass()); guarantee(cloneable == klass->is_cloneable(), "incorrect cloneable flag"); } #endif @@ -908,7 +908,7 @@ // Special handling for primitive objects if (java_lang_Class::is_primitive(mirror)) { // Primitive objects does not have any interfaces - objArrayOop r = oopFactory::new_objArray(SystemDictionary::class_klass(), 0, CHECK_NULL); + objArrayOop r = oopFactory::new_objArray(SystemDictionary::Class_klass(), 0, CHECK_NULL); return (jobjectArray) JNIHandles::make_local(env, r); } @@ -923,7 +923,7 @@ } // Allocate result array - objArrayOop r = oopFactory::new_objArray(SystemDictionary::class_klass(), size, CHECK_NULL); + objArrayOop r = oopFactory::new_objArray(SystemDictionary::Class_klass(), size, CHECK_NULL); objArrayHandle result (THREAD, r); // Fill in result if (klass->oop_is_instance()) { @@ -934,8 +934,8 @@ } } else { // All arrays implement java.lang.Cloneable and java.io.Serializable - result->obj_at_put(0, Klass::cast(SystemDictionary::cloneable_klass())->java_mirror()); - result->obj_at_put(1, Klass::cast(SystemDictionary::serializable_klass())->java_mirror()); + result->obj_at_put(0, Klass::cast(SystemDictionary::Cloneable_klass())->java_mirror()); + result->obj_at_put(1, Klass::cast(SystemDictionary::Serializable_klass())->java_mirror()); } return (jobjectArray) JNIHandles::make_local(env, result()); JVM_END @@ -1098,8 +1098,8 @@ pending_exception = Handle(THREAD, PENDING_EXCEPTION); CLEAR_PENDING_EXCEPTION; - if ( pending_exception->is_a(SystemDictionary::exception_klass()) && - !pending_exception->is_a(SystemDictionary::runtime_exception_klass())) { + if ( pending_exception->is_a(SystemDictionary::Exception_klass()) && + !pending_exception->is_a(SystemDictionary::RuntimeException_klass())) { // Throw a java.security.PrivilegedActionException(Exception e) exception JavaCallArguments args(pending_exception); THROW_ARG_0(vmSymbolHandles::java_security_PrivilegedActionException(), @@ -1190,7 +1190,7 @@ // the resource area must be registered in case of a gc RegisterArrayForGC ragc(thread, local_array); - objArrayOop context = oopFactory::new_objArray(SystemDictionary::protectionDomain_klass(), + objArrayOop context = oopFactory::new_objArray(SystemDictionary::ProtectionDomain_klass(), local_array->length(), CHECK_NULL); objArrayHandle h_context(thread, context); for (int index = 0; index < local_array->length(); index++) { @@ -1251,7 +1251,7 @@ if (java_lang_Class::is_primitive(JNIHandles::resolve_non_null(ofClass)) || ! Klass::cast(java_lang_Class::as_klassOop(JNIHandles::resolve_non_null(ofClass)))->oop_is_instance()) { - oop result = oopFactory::new_objArray(SystemDictionary::class_klass(), 0, CHECK_NULL); + oop result = oopFactory::new_objArray(SystemDictionary::Class_klass(), 0, CHECK_NULL); return (jobjectArray)JNIHandles::make_local(env, result); } @@ -1259,7 +1259,7 @@ if (k->inner_classes()->length() == 0) { // Neither an inner nor outer class - oop result = oopFactory::new_objArray(SystemDictionary::class_klass(), 0, CHECK_NULL); + oop result = oopFactory::new_objArray(SystemDictionary::Class_klass(), 0, CHECK_NULL); return (jobjectArray)JNIHandles::make_local(env, result); } @@ -1269,7 +1269,7 @@ int length = icls->length(); // Allocate temp. result array - objArrayOop r = oopFactory::new_objArray(SystemDictionary::class_klass(), length/4, CHECK_NULL); + objArrayOop r = oopFactory::new_objArray(SystemDictionary::Class_klass(), length/4, CHECK_NULL); objArrayHandle result (THREAD, r); int members = 0; @@ -1299,7 +1299,7 @@ if (members != length) { // Return array of right length - objArrayOop res = oopFactory::new_objArray(SystemDictionary::class_klass(), members, CHECK_NULL); + objArrayOop res = oopFactory::new_objArray(SystemDictionary::Class_klass(), members, CHECK_NULL); for(int i = 0; i < members; i++) { res->obj_at_put(i, result->obj_at(i)); } @@ -1318,19 +1318,20 @@ return NULL; } - symbolOop simple_name = NULL; + bool inner_is_member = false; klassOop outer_klass = instanceKlass::cast(java_lang_Class::as_klassOop(JNIHandles::resolve_non_null(ofClass)) - )->compute_enclosing_class(simple_name, CHECK_NULL); + )->compute_enclosing_class(&inner_is_member, CHECK_NULL); if (outer_klass == NULL) return NULL; // already a top-level class - if (simple_name == NULL) return NULL; // an anonymous class (inside a method) + if (!inner_is_member) return NULL; // an anonymous class (inside a method) return (jclass) JNIHandles::make_local(env, Klass::cast(outer_klass)->java_mirror()); } JVM_END // should be in instanceKlass.cpp, but is here for historical reasons klassOop instanceKlass::compute_enclosing_class_impl(instanceKlassHandle k, - symbolOop& simple_name_result, TRAPS) { + bool* inner_is_member, + TRAPS) { Thread* thread = THREAD; const int inner_class_info_index = inner_class_inner_class_info_offset; const int outer_class_info_index = inner_class_outer_class_info_offset; @@ -1347,8 +1348,7 @@ bool found = false; klassOop ok; instanceKlassHandle outer_klass; - bool inner_is_member = false; - int simple_name_index = 0; + *inner_is_member = false; // Find inner_klass attribute for (int i = 0; i < i_length && !found; i += inner_class_next_offset) { @@ -1364,8 +1364,7 @@ if (found && ooff != 0) { ok = i_cp->klass_at(ooff, CHECK_NULL); outer_klass = instanceKlassHandle(thread, ok); - simple_name_index = noff; - inner_is_member = true; + *inner_is_member = true; } } } @@ -1377,7 +1376,7 @@ if (encl_method_class_idx != 0) { ok = i_cp->klass_at(encl_method_class_idx, CHECK_NULL); outer_klass = instanceKlassHandle(thread, ok); - inner_is_member = false; + *inner_is_member = false; } } @@ -1387,9 +1386,7 @@ // Throws an exception if outer klass has not declared k as an inner klass // We need evidence that each klass knows about the other, or else // the system could allow a spoof of an inner class to gain access rights. - Reflection::check_for_inner_class(outer_klass, k, inner_is_member, CHECK_NULL); - - simple_name_result = (inner_is_member ? i_cp->symbol_at(simple_name_index) : symbolOop(NULL)); + Reflection::check_for_inner_class(outer_klass, k, *inner_is_member, CHECK_NULL); return outer_klass(); } @@ -1473,11 +1470,11 @@ oop mirror = NULL; int slot = 0; - if (reflected->klass() == SystemDictionary::reflect_constructor_klass()) { + if (reflected->klass() == SystemDictionary::reflect_Constructor_klass()) { mirror = java_lang_reflect_Constructor::clazz(reflected); slot = java_lang_reflect_Constructor::slot(reflected); } else { - assert(reflected->klass() == SystemDictionary::reflect_method_klass(), + assert(reflected->klass() == SystemDictionary::reflect_Method_klass(), "wrong type"); mirror = java_lang_reflect_Method::clazz(reflected); slot = java_lang_reflect_Method::slot(reflected); @@ -1533,7 +1530,7 @@ if (java_lang_Class::is_primitive(JNIHandles::resolve_non_null(ofClass)) || Klass::cast(java_lang_Class::as_klassOop(JNIHandles::resolve_non_null(ofClass)))->oop_is_javaArray()) { // Return empty array - oop res = oopFactory::new_objArray(SystemDictionary::reflect_field_klass(), 0, CHECK_NULL); + oop res = oopFactory::new_objArray(SystemDictionary::reflect_Field_klass(), 0, CHECK_NULL); return (jobjectArray) JNIHandles::make_local(env, res); } @@ -1561,13 +1558,13 @@ } else { num_fields = fields_len / instanceKlass::next_offset; - if (k() == SystemDictionary::throwable_klass()) { + if (k() == SystemDictionary::Throwable_klass()) { num_fields--; skip_backtrace = true; } } - objArrayOop r = oopFactory::new_objArray(SystemDictionary::reflect_field_klass(), num_fields, CHECK_NULL); + objArrayOop r = oopFactory::new_objArray(SystemDictionary::reflect_Field_klass(), num_fields, CHECK_NULL); objArrayHandle result (THREAD, r); int out_idx = 0; @@ -1601,7 +1598,7 @@ if (java_lang_Class::is_primitive(JNIHandles::resolve_non_null(ofClass)) || Klass::cast(java_lang_Class::as_klassOop(JNIHandles::resolve_non_null(ofClass)))->oop_is_javaArray()) { // Return empty array - oop res = oopFactory::new_objArray(SystemDictionary::reflect_method_klass(), 0, CHECK_NULL); + oop res = oopFactory::new_objArray(SystemDictionary::reflect_Method_klass(), 0, CHECK_NULL); return (jobjectArray) JNIHandles::make_local(env, res); } @@ -1625,7 +1622,7 @@ } // Allocate result - objArrayOop r = oopFactory::new_objArray(SystemDictionary::reflect_method_klass(), num_methods, CHECK_NULL); + objArrayOop r = oopFactory::new_objArray(SystemDictionary::reflect_Method_klass(), num_methods, CHECK_NULL); objArrayHandle result (THREAD, r); int out_idx = 0; @@ -1653,7 +1650,7 @@ if (java_lang_Class::is_primitive(JNIHandles::resolve_non_null(ofClass)) || Klass::cast(java_lang_Class::as_klassOop(JNIHandles::resolve_non_null(ofClass)))->oop_is_javaArray()) { // Return empty array - oop res = oopFactory::new_objArray(SystemDictionary::reflect_constructor_klass(), 0 , CHECK_NULL); + oop res = oopFactory::new_objArray(SystemDictionary::reflect_Constructor_klass(), 0 , CHECK_NULL); return (jobjectArray) JNIHandles::make_local(env, res); } @@ -1677,7 +1674,7 @@ } // Allocate result - objArrayOop r = oopFactory::new_objArray(SystemDictionary::reflect_constructor_klass(), num_constructors, CHECK_NULL); + objArrayOop r = oopFactory::new_objArray(SystemDictionary::reflect_Constructor_klass(), num_constructors, CHECK_NULL); objArrayHandle result(THREAD, r); int out_idx = 0; @@ -1890,7 +1887,7 @@ symbolHandle klass_name (THREAD, cp->klass_name_at(klass_ref)); symbolHandle member_name(THREAD, cp->uncached_name_ref_at(index)); symbolHandle member_sig (THREAD, cp->uncached_signature_ref_at(index)); - objArrayOop dest_o = oopFactory::new_objArray(SystemDictionary::string_klass(), 3, CHECK_NULL); + objArrayOop dest_o = oopFactory::new_objArray(SystemDictionary::String_klass(), 3, CHECK_NULL); objArrayHandle dest(THREAD, dest_o); Handle str = java_lang_String::create_from_symbol(klass_name, CHECK_NULL); dest->obj_at_put(0, str()); @@ -2257,10 +2254,8 @@ switch (cp->tag_at(cp_index).value()) { case JVM_CONSTANT_InterfaceMethodref: case JVM_CONSTANT_Methodref: + case JVM_CONSTANT_NameAndType: // for invokedynamic return cp->uncached_name_ref_at(cp_index)->as_utf8(); - case JVM_CONSTANT_NameAndType: - // for invokedynamic - return cp->nt_name_ref_at(cp_index)->as_utf8(); default: fatal("JVM_GetCPMethodNameUTF: illegal constant"); } @@ -2277,10 +2272,8 @@ switch (cp->tag_at(cp_index).value()) { case JVM_CONSTANT_InterfaceMethodref: case JVM_CONSTANT_Methodref: + case JVM_CONSTANT_NameAndType: // for invokedynamic return cp->uncached_signature_ref_at(cp_index)->as_utf8(); - case JVM_CONSTANT_NameAndType: - // for invokedynamic - return cp->nt_signature_ref_at(cp_index)->as_utf8(); default: fatal("JVM_GetCPMethodSignatureUTF: illegal constant"); } @@ -2582,7 +2575,7 @@ JavaValue result(T_VOID); JavaCalls::call_virtual(&result, obj, - KlassHandle(THREAD, SystemDictionary::thread_klass()), + KlassHandle(THREAD, SystemDictionary::Thread_klass()), vmSymbolHandles::run_method_name(), vmSymbolHandles::void_method_signature(), THREAD); @@ -2680,7 +2673,7 @@ // Fix for 4314342, 4145910, perhaps others: it now doesn't have // any effect on the "liveness" of a thread; see // JVM_IsThreadAlive, below. - if (java_throwable->is_a(SystemDictionary::threaddeath_klass())) { + if (java_throwable->is_a(SystemDictionary::ThreadDeath_klass())) { java_lang_Thread::set_stillborn(java_thread); } THROW_OOP(java_throwable); @@ -3035,7 +3028,7 @@ } // Create result array of type [Ljava/lang/Class; - objArrayOop result = oopFactory::new_objArray(SystemDictionary::class_klass(), depth, CHECK_NULL); + objArrayOop result = oopFactory::new_objArray(SystemDictionary::Class_klass(), depth, CHECK_NULL); // Fill in mirrors corresponding to method holders int index = 0; while (first != NULL) { @@ -4331,7 +4324,7 @@ JvmtiVMObjectAllocEventCollector oam; int num_threads = tle.num_threads(); - objArrayOop r = oopFactory::new_objArray(SystemDictionary::thread_klass(), num_threads, CHECK_NULL); + objArrayOop r = oopFactory::new_objArray(SystemDictionary::Thread_klass(), num_threads, CHECK_NULL); objArrayHandle threads_ah(THREAD, r); for (int i = 0; i < num_threads; i++) { @@ -4365,7 +4358,7 @@ // check if threads is not an array of objects of Thread class klassOop k = objArrayKlass::cast(ah->klass())->element_klass(); - if (k != SystemDictionary::thread_klass()) { + if (k != SystemDictionary::Thread_klass()) { THROW_(vmSymbols::java_lang_IllegalArgumentException(), 0); } @@ -4425,7 +4418,7 @@ if (encl_method_class_idx == 0) { return NULL; } - objArrayOop dest_o = oopFactory::new_objArray(SystemDictionary::object_klass(), 3, CHECK_NULL); + objArrayOop dest_o = oopFactory::new_objArray(SystemDictionary::Object_klass(), 3, CHECK_NULL); objArrayHandle dest(THREAD, dest_o); klassOop enc_k = ik_h->constants()->klass_at(encl_method_class_idx, CHECK_NULL); dest->obj_at_put(0, Klass::cast(enc_k)->java_mirror()); @@ -4539,7 +4532,7 @@ values_h->int_at(0) == java_lang_Thread::NEW, "Invalid threadStatus value"); - objArrayOop r = oopFactory::new_objArray(SystemDictionary::string_klass(), + objArrayOop r = oopFactory::new_objArray(SystemDictionary::String_klass(), 1, /* only 1 substate */ CHECK_NULL); names_h = objArrayHandle(THREAD, r); @@ -4552,7 +4545,7 @@ values_h->int_at(0) == java_lang_Thread::RUNNABLE, "Invalid threadStatus value"); - objArrayOop r = oopFactory::new_objArray(SystemDictionary::string_klass(), + objArrayOop r = oopFactory::new_objArray(SystemDictionary::String_klass(), 1, /* only 1 substate */ CHECK_NULL); names_h = objArrayHandle(THREAD, r); @@ -4565,7 +4558,7 @@ values_h->int_at(0) == java_lang_Thread::BLOCKED_ON_MONITOR_ENTER, "Invalid threadStatus value"); - objArrayOop r = oopFactory::new_objArray(SystemDictionary::string_klass(), + objArrayOop r = oopFactory::new_objArray(SystemDictionary::String_klass(), 1, /* only 1 substate */ CHECK_NULL); names_h = objArrayHandle(THREAD, r); @@ -4578,7 +4571,7 @@ values_h->int_at(0) == java_lang_Thread::IN_OBJECT_WAIT && values_h->int_at(1) == java_lang_Thread::PARKED, "Invalid threadStatus value"); - objArrayOop r = oopFactory::new_objArray(SystemDictionary::string_klass(), + objArrayOop r = oopFactory::new_objArray(SystemDictionary::String_klass(), 2, /* number of substates */ CHECK_NULL); names_h = objArrayHandle(THREAD, r); @@ -4596,7 +4589,7 @@ values_h->int_at(1) == java_lang_Thread::IN_OBJECT_WAIT_TIMED && values_h->int_at(2) == java_lang_Thread::PARKED_TIMED, "Invalid threadStatus value"); - objArrayOop r = oopFactory::new_objArray(SystemDictionary::string_klass(), + objArrayOop r = oopFactory::new_objArray(SystemDictionary::String_klass(), 3, /* number of substates */ CHECK_NULL); names_h = objArrayHandle(THREAD, r); @@ -4615,7 +4608,7 @@ assert(values_h->length() == 1 && values_h->int_at(0) == java_lang_Thread::TERMINATED, "Invalid threadStatus value"); - objArrayOop r = oopFactory::new_objArray(SystemDictionary::string_klass(), + objArrayOop r = oopFactory::new_objArray(SystemDictionary::String_klass(), 1, /* only 1 substate */ CHECK_NULL); names_h = objArrayHandle(THREAD, r); @@ -4650,4 +4643,3 @@ #endif // KERNEL } JVM_END -
--- a/src/share/vm/prims/jvmtiCodeBlobEvents.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/jvmtiCodeBlobEvents.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -402,7 +402,7 @@ address scopes_data = nm->scopes_data_begin(); for( pcd = nm->scopes_pcs_begin(); pcd < nm->scopes_pcs_end(); ++pcd ) { - ScopeDesc sc0(nm, pcd->scope_decode_offset(), pcd->should_reexecute()); + ScopeDesc sc0(nm, pcd->scope_decode_offset(), pcd->should_reexecute(), pcd->return_oop()); ScopeDesc *sd = &sc0; while( !sd->is_top() ) { sd = sd->sender(); } int bci = sd->bci();
--- a/src/share/vm/prims/jvmtiEnter.xsl Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/jvmtiEnter.xsl Wed Mar 24 17:16:33 2010 -0700 @@ -773,7 +773,7 @@ </xsl:apply-templates> <xsl:text> } - if (!thread_oop->is_a(SystemDictionary::thread_klass())) { + if (!thread_oop->is_a(SystemDictionary::Thread_klass())) { </xsl:text> <xsl:apply-templates select=".." mode="traceError"> <xsl:with-param name="err">JVMTI_ERROR_INVALID_THREAD</xsl:with-param> @@ -857,7 +857,7 @@ </xsl:apply-templates> <xsl:text> } - if (!k_mirror->is_a(SystemDictionary::class_klass())) { + if (!k_mirror->is_a(SystemDictionary::Class_klass())) { </xsl:text> <xsl:apply-templates select=".." mode="traceError"> <xsl:with-param name="err">JVMTI_ERROR_INVALID_CLASS</xsl:with-param>
--- a/src/share/vm/prims/jvmtiEnv.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/jvmtiEnv.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2003-2007 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2003-2009 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -32,15 +32,15 @@ // FIXLATER: hook into JvmtiTrace #define TraceJVMTICalls false -JvmtiEnv::JvmtiEnv() : JvmtiEnvBase() { +JvmtiEnv::JvmtiEnv(jint version) : JvmtiEnvBase(version) { } JvmtiEnv::~JvmtiEnv() { } JvmtiEnv* -JvmtiEnv::create_a_jvmti() { - return new JvmtiEnv(); +JvmtiEnv::create_a_jvmti(jint version) { + return new JvmtiEnv(version); } // VM operation class to copy jni function table at safepoint. @@ -133,7 +133,7 @@ if (thread_oop == NULL) { return JVMTI_ERROR_INVALID_THREAD; } - if (!thread_oop->is_a(SystemDictionary::thread_klass())) { + if (!thread_oop->is_a(SystemDictionary::Thread_klass())) { return JVMTI_ERROR_INVALID_THREAD; } JavaThread* java_thread = java_lang_Thread::thread(thread_oop); @@ -199,7 +199,7 @@ if (k_mirror == NULL) { return JVMTI_ERROR_INVALID_CLASS; } - if (!k_mirror->is_a(SystemDictionary::class_klass())) { + if (!k_mirror->is_a(SystemDictionary::Class_klass())) { return JVMTI_ERROR_INVALID_CLASS; } @@ -266,7 +266,7 @@ oop mirror = JNIHandles::resolve_external_guard(object); NULL_CHECK(mirror, JVMTI_ERROR_INVALID_OBJECT); - if (mirror->klass() == SystemDictionary::class_klass()) { + if (mirror->klass() == SystemDictionary::Class_klass()) { if (!java_lang_Class::is_primitive(mirror)) { mirror = java_lang_Class::as_klassOop(mirror); assert(mirror != NULL, "class for non-primitive mirror must exist"); @@ -327,7 +327,7 @@ if (thread_oop == NULL) { return JVMTI_ERROR_INVALID_THREAD; } - if (!thread_oop->is_a(SystemDictionary::thread_klass())) { + if (!thread_oop->is_a(SystemDictionary::Thread_klass())) { return JVMTI_ERROR_INVALID_THREAD; } java_thread = java_lang_Thread::thread(thread_oop); @@ -411,8 +411,15 @@ if (phase == JVMTI_PHASE_ONLOAD) { Arguments::append_sysclasspath(segment); return JVMTI_ERROR_NONE; - } else { - assert(phase == JVMTI_PHASE_LIVE, "sanity check"); + } else if (use_version_1_0_semantics()) { + // This JvmtiEnv requested version 1.0 semantics and this function + // is only allowed in the ONLOAD phase in version 1.0 so we need to + // return an error here. + return JVMTI_ERROR_WRONG_PHASE; + } else if (phase == JVMTI_PHASE_LIVE) { + // The phase is checked by the wrapper that called this function, + // but this thread could be racing with the thread that is + // terminating the VM so we check one more time. // create the zip entry ClassPathZipEntry* zip_entry = ClassLoader::create_class_path_zip_entry(segment); @@ -433,6 +440,8 @@ } ClassLoader::add_to_list(zip_entry); return JVMTI_ERROR_NONE; + } else { + return JVMTI_ERROR_WRONG_PHASE; } } /* end AddToBootstrapClassLoaderSearch */ @@ -451,11 +460,12 @@ } } return JVMTI_ERROR_NONE; - } else { + } else if (phase == JVMTI_PHASE_LIVE) { + // The phase is checked by the wrapper that called this function, + // but this thread could be racing with the thread that is + // terminating the VM so we check one more time. HandleMark hm; - assert(phase == JVMTI_PHASE_LIVE, "sanity check"); - // create the zip entry (which will open the zip file and hence // check that the segment is indeed a zip file). ClassPathZipEntry* zip_entry = ClassLoader::create_class_path_zip_entry(segment); @@ -504,6 +514,8 @@ } return JVMTI_ERROR_NONE; + } else { + return JVMTI_ERROR_WRONG_PHASE; } } /* end AddToSystemClassLoaderSearch */ @@ -580,7 +592,6 @@ break; case JVMTI_VERBOSE_GC: PrintGC = value != 0; - TraceClassUnloading = value != 0; break; case JVMTI_VERBOSE_JNI: PrintJNIResolving = value != 0; @@ -620,7 +631,7 @@ thread_oop = JNIHandles::resolve_external_guard(thread); } - if (thread_oop == NULL || !thread_oop->is_a(SystemDictionary::thread_klass())) { + if (thread_oop == NULL || !thread_oop->is_a(SystemDictionary::Thread_klass())) { return JVMTI_ERROR_INVALID_THREAD; } @@ -858,7 +869,7 @@ jvmtiError JvmtiEnv::InterruptThread(jthread thread) { oop thread_oop = JNIHandles::resolve_external_guard(thread); - if (thread_oop == NULL || !thread_oop->is_a(SystemDictionary::thread_klass())) + if (thread_oop == NULL || !thread_oop->is_a(SystemDictionary::Thread_klass())) return JVMTI_ERROR_INVALID_THREAD; JavaThread* current_thread = JavaThread::current(); @@ -895,7 +906,7 @@ } else { thread_oop = JNIHandles::resolve_external_guard(thread); } - if (thread_oop == NULL || !thread_oop->is_a(SystemDictionary::thread_klass())) + if (thread_oop == NULL || !thread_oop->is_a(SystemDictionary::Thread_klass())) return JVMTI_ERROR_INVALID_THREAD; Handle thread_obj(current_thread, thread_oop); @@ -1061,7 +1072,7 @@ jvmtiError JvmtiEnv::RunAgentThread(jthread thread, jvmtiStartFunction proc, const void* arg, jint priority) { oop thread_oop = JNIHandles::resolve_external_guard(thread); - if (thread_oop == NULL || !thread_oop->is_a(SystemDictionary::thread_klass())) { + if (thread_oop == NULL || !thread_oop->is_a(SystemDictionary::Thread_klass())) { return JVMTI_ERROR_INVALID_THREAD; } if (priority < JVMTI_THREAD_MIN_PRIORITY || priority > JVMTI_THREAD_MAX_PRIORITY) { @@ -2863,6 +2874,14 @@ // is_obsolete_ptr - pre-checked for NULL jvmtiError JvmtiEnv::IsMethodObsolete(methodOop method_oop, jboolean* is_obsolete_ptr) { + if (use_version_1_0_semantics() && + get_capabilities()->can_redefine_classes == 0) { + // This JvmtiEnv requested version 1.0 semantics and this function + // requires the can_redefine_classes capability in version 1.0 so + // we need to return an error here. + return JVMTI_ERROR_MUST_POSSESS_CAPABILITY; + } + if (method_oop == NULL || method_oop->is_obsolete()) { *is_obsolete_ptr = true; } else {
--- a/src/share/vm/prims/jvmtiEnvBase.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/jvmtiEnvBase.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -123,7 +123,26 @@ } -JvmtiEnvBase::JvmtiEnvBase() : _env_event_enable() { +bool +JvmtiEnvBase::use_version_1_0_semantics() { + int major, minor, micro; + + JvmtiExport::decode_version_values(_version, &major, &minor, µ); + return major == 1 && minor == 0; // micro version doesn't matter here +} + + +bool +JvmtiEnvBase::use_version_1_1_semantics() { + int major, minor, micro; + + JvmtiExport::decode_version_values(_version, &major, &minor, µ); + return major == 1 && minor == 1; // micro version doesn't matter here +} + + +JvmtiEnvBase::JvmtiEnvBase(jint version) : _env_event_enable() { + _version = version; _env_local_storage = NULL; _tag_map = NULL; _native_method_prefix_count = 0; @@ -508,7 +527,7 @@ JavaThread * JvmtiEnvBase::get_JavaThread(jthread jni_thread) { oop t = JNIHandles::resolve_external_guard(jni_thread); - if (t == NULL || !t->is_a(SystemDictionary::thread_klass())) { + if (t == NULL || !t->is_a(SystemDictionary::Thread_klass())) { return NULL; } // The following returns NULL if the thread has not yet run or is in @@ -1250,7 +1269,7 @@ for (int i = 0; i < _thread_count; ++i) { jthread jt = _thread_list[i]; oop thread_oop = JNIHandles::resolve_external_guard(jt); - if (thread_oop == NULL || !thread_oop->is_a(SystemDictionary::thread_klass())) { + if (thread_oop == NULL || !thread_oop->is_a(SystemDictionary::Thread_klass())) { set_result(JVMTI_ERROR_INVALID_THREAD); return; }
--- a/src/share/vm/prims/jvmtiEnvBase.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/jvmtiEnvBase.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2003-2006 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2003-2009 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -76,6 +76,7 @@ jvmtiEnv _jvmti_external; jint _magic; + jint _version; // version value passed to JNI GetEnv() JvmtiEnvBase* _next; bool _is_retransformable; const void *_env_local_storage; // per env agent allocated data. @@ -91,7 +92,7 @@ int _native_method_prefix_count; protected: - JvmtiEnvBase(); + JvmtiEnvBase(jint version); ~JvmtiEnvBase(); void dispose(); void env_dispose(); @@ -122,6 +123,9 @@ bool is_valid(); + bool use_version_1_0_semantics(); // agent asked for version 1.0 + bool use_version_1_1_semantics(); // agent asked for version 1.1 + bool is_retransformable() { return _is_retransformable; } static ByteSize jvmti_external_offset() {
--- a/src/share/vm/prims/jvmtiEventController.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/jvmtiEventController.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2003-2007 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2003-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -82,7 +82,7 @@ THREAD_START_BIT | THREAD_END_BIT | DYNAMIC_CODE_GENERATED_BIT; static const jlong GLOBAL_EVENT_BITS = ~THREAD_FILTERED_EVENT_BITS; - +static const jlong SHOULD_POST_ON_EXCEPTIONS_BITS = EXCEPTION_BITS | METHOD_EXIT_BIT | FRAME_POP_BIT; /////////////////////////////////////////////////////////////// // @@ -511,7 +511,12 @@ leave_interp_only_mode(state); } } + + // update the JavaThread cached value for thread-specific should_post_on_exceptions value + bool should_post_on_exceptions = (any_env_enabled & SHOULD_POST_ON_EXCEPTIONS_BITS) != 0; + state->set_should_post_on_exceptions(should_post_on_exceptions); } + return any_env_enabled; } @@ -615,6 +620,10 @@ // set global truly enabled, that is, any thread in any environment JvmtiEventController::_universal_global_event_enabled.set_bits(any_env_thread_enabled); + + // set global should_post_on_exceptions + JvmtiExport::set_should_post_on_exceptions((any_env_thread_enabled & SHOULD_POST_ON_EXCEPTIONS_BITS) != 0); + } EC_TRACE(("JVMTI [-] # recompute enabled - after %llx", any_env_thread_enabled));
--- a/src/share/vm/prims/jvmtiExport.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/jvmtiExport.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2003-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2003-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -319,7 +319,27 @@ jint JvmtiExport::get_jvmti_interface(JavaVM *jvm, void **penv, jint version) { - /* To Do: add version checks */ + // The JVMTI_VERSION_INTERFACE_JVMTI part of the version number + // has already been validated in JNI GetEnv(). + int major, minor, micro; + + // micro version doesn't matter here (yet?) + decode_version_values(version, &major, &minor, µ); + switch (major) { + case 1: + switch (minor) { + case 0: // version 1.0.<micro> is recognized + case 1: // version 1.1.<micro> is recognized + break; + + default: + return JNI_EVERSION; // unsupported minor version number + } + break; + + default: + return JNI_EVERSION; // unsupported major version number + } if (JvmtiEnv::get_phase() == JVMTI_PHASE_LIVE) { JavaThread* current_thread = (JavaThread*) ThreadLocalStorage::thread(); @@ -328,13 +348,13 @@ __ENTRY(jvmtiEnv*, JvmtiExport::get_jvmti_interface, current_thread) debug_only(VMNativeEntryWrapper __vew;) - JvmtiEnv *jvmti_env = JvmtiEnv::create_a_jvmti(); + JvmtiEnv *jvmti_env = JvmtiEnv::create_a_jvmti(version); *penv = jvmti_env->jvmti_external(); // actual type is jvmtiEnv* -- not to be confused with JvmtiEnv* return JNI_OK; } else if (JvmtiEnv::get_phase() == JVMTI_PHASE_ONLOAD) { // not live, no thread to transition - JvmtiEnv *jvmti_env = JvmtiEnv::create_a_jvmti(); + JvmtiEnv *jvmti_env = JvmtiEnv::create_a_jvmti(version); *penv = jvmti_env->jvmti_external(); // actual type is jvmtiEnv* -- not to be confused with JvmtiEnv* return JNI_OK; @@ -345,6 +365,15 @@ } } + +void +JvmtiExport::decode_version_values(jint version, int * major, int * minor, + int * micro) { + *major = (version & JVMTI_VERSION_MASK_MAJOR) >> JVMTI_VERSION_SHIFT_MAJOR; + *minor = (version & JVMTI_VERSION_MASK_MINOR) >> JVMTI_VERSION_SHIFT_MINOR; + *micro = (version & JVMTI_VERSION_MASK_MICRO) >> JVMTI_VERSION_SHIFT_MICRO; +} + void JvmtiExport::enter_primordial_phase() { JvmtiEnvBase::set_phase(JVMTI_PHASE_PRIMORDIAL); } @@ -627,7 +656,7 @@ klassOop k = obj->klass(); // if the object is a java.lang.Class then return the java mirror - if (k == SystemDictionary::class_klass()) { + if (k == SystemDictionary::Class_klass()) { if (!java_lang_Class::is_primitive(obj)) { k = java_lang_Class::as_klassOop(obj); assert(k != NULL, "class for non-primitive mirror must exist"); @@ -657,11 +686,11 @@ jvmtiAddrLocationMap *_map; const void *_compile_info; public: - JvmtiCompiledMethodLoadEventMark(JavaThread *thread, nmethod *nm) + JvmtiCompiledMethodLoadEventMark(JavaThread *thread, nmethod *nm, void* compile_info_ptr = NULL) : JvmtiMethodEventMark(thread,methodHandle(thread, nm->method())) { _code_data = nm->code_begin(); _code_size = nm->code_size(); - _compile_info = NULL; /* no info for our VM. */ + _compile_info = compile_info_ptr; // Set void pointer of compiledMethodLoad Event. Default value is NULL. JvmtiCodeBlobEvents::build_jvmti_addr_location_map(nm, &_map, &_map_length); } ~JvmtiCompiledMethodLoadEventMark() { @@ -848,7 +877,7 @@ bool JvmtiExport::_can_get_source_debug_extension = false; bool JvmtiExport::_can_maintain_original_method_order = false; bool JvmtiExport::_can_post_interpreter_events = false; -bool JvmtiExport::_can_post_exceptions = false; +bool JvmtiExport::_can_post_on_exceptions = false; bool JvmtiExport::_can_post_breakpoint = false; bool JvmtiExport::_can_post_field_access = false; bool JvmtiExport::_can_post_field_modification = false; @@ -879,6 +908,7 @@ bool JvmtiExport::_should_post_object_free = false; bool JvmtiExport::_should_post_resource_exhausted = false; bool JvmtiExport::_should_post_vm_object_alloc = false; +bool JvmtiExport::_should_post_on_exceptions = false; //////////////////////////////////////////////////////////////////////////////////////////////// @@ -1723,6 +1753,46 @@ } } +// Returns a record containing inlining information for the given nmethod +jvmtiCompiledMethodLoadInlineRecord* create_inline_record(nmethod* nm) { + jint numstackframes = 0; + jvmtiCompiledMethodLoadInlineRecord* record = (jvmtiCompiledMethodLoadInlineRecord*)NEW_RESOURCE_OBJ(jvmtiCompiledMethodLoadInlineRecord); + record->header.kind = JVMTI_CMLR_INLINE_INFO; + record->header.next = NULL; + record->header.majorinfoversion = JVMTI_CMLR_MAJOR_VERSION_1; + record->header.minorinfoversion = JVMTI_CMLR_MINOR_VERSION_0; + record->numpcs = 0; + for(PcDesc* p = nm->scopes_pcs_begin(); p < nm->scopes_pcs_end(); p++) { + if(p->scope_decode_offset() == DebugInformationRecorder::serialized_null) continue; + record->numpcs++; + } + record->pcinfo = (PCStackInfo*)(NEW_RESOURCE_ARRAY(PCStackInfo, record->numpcs)); + int scope = 0; + for(PcDesc* p = nm->scopes_pcs_begin(); p < nm->scopes_pcs_end(); p++) { + if(p->scope_decode_offset() == DebugInformationRecorder::serialized_null) continue; + void* pc_address = (void*)p->real_pc(nm); + assert(pc_address != NULL, "pc_address must be non-null"); + record->pcinfo[scope].pc = pc_address; + numstackframes=0; + for(ScopeDesc* sd = nm->scope_desc_at(p->real_pc(nm));sd != NULL;sd = sd->sender()) { + numstackframes++; + } + assert(numstackframes != 0, "numstackframes must be nonzero."); + record->pcinfo[scope].methods = (jmethodID *)NEW_RESOURCE_ARRAY(jmethodID, numstackframes); + record->pcinfo[scope].bcis = (jint *)NEW_RESOURCE_ARRAY(jint, numstackframes); + record->pcinfo[scope].numstackframes = numstackframes; + int stackframe = 0; + for(ScopeDesc* sd = nm->scope_desc_at(p->real_pc(nm));sd != NULL;sd = sd->sender()) { + // sd->method() can be NULL for stubs but not for nmethods. To be completely robust, include an assert that we should never see a null sd->method() + assert(!sd->method().is_null(), "sd->method() cannot be null."); + record->pcinfo[scope].methods[stackframe] = sd->method()->jmethod_id(); + record->pcinfo[scope].bcis[stackframe] = sd->bci(); + stackframe++; + } + scope++; + } + return record; +} void JvmtiExport::post_compiled_method_load(nmethod *nm) { // If there are pending CompiledMethodUnload events then these are @@ -1751,7 +1821,11 @@ (nm->method() == NULL) ? "NULL" : nm->method()->name()->as_C_string())); ResourceMark rm(thread); - JvmtiCompiledMethodLoadEventMark jem(thread, nm); + + // Add inlining information + jvmtiCompiledMethodLoadInlineRecord* inlinerecord = create_inline_record(nm); + // Pass inlining information through the void pointer + JvmtiCompiledMethodLoadEventMark jem(thread, nm, inlinerecord); JvmtiJavaThreadEventTransition jet(thread); jvmtiEventCompiledMethodLoad callback = env->callbacks()->CompiledMethodLoad; if (callback != NULL) { @@ -1896,7 +1970,7 @@ if (collector != NULL && collector->is_enabled()) { // Don't record classes as these will be notified via the ClassLoad // event. - if (obj->klass() != SystemDictionary::class_klass()) { + if (obj->klass() != SystemDictionary::Class_klass()) { collector->record_allocation(obj); } }
--- a/src/share/vm/prims/jvmtiExport.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/jvmtiExport.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1998-2006 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1998-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -66,7 +66,7 @@ JVMTI_SUPPORT_FLAG(can_get_source_debug_extension) JVMTI_SUPPORT_FLAG(can_maintain_original_method_order) JVMTI_SUPPORT_FLAG(can_post_interpreter_events) - JVMTI_SUPPORT_FLAG(can_post_exceptions) + JVMTI_SUPPORT_FLAG(can_post_on_exceptions) JVMTI_SUPPORT_FLAG(can_post_breakpoint) JVMTI_SUPPORT_FLAG(can_post_field_access) JVMTI_SUPPORT_FLAG(can_post_field_modification) @@ -93,6 +93,7 @@ JVMTI_SUPPORT_FLAG(should_post_data_dump) JVMTI_SUPPORT_FLAG(should_post_garbage_collection_start) JVMTI_SUPPORT_FLAG(should_post_garbage_collection_finish) + JVMTI_SUPPORT_FLAG(should_post_on_exceptions) // ------ the below maybe don't have to be (but are for now) // fixed conditions here ------------ @@ -236,6 +237,8 @@ static bool is_jvmti_version(jint version) { return (version & JVMTI_VERSION_MASK) == JVMTI_VERSION_VALUE; } static bool is_jvmdi_version(jint version) { return (version & JVMTI_VERSION_MASK) == JVMDI_VERSION_VALUE; } static jint get_jvmti_interface(JavaVM *jvm, void **penv, jint version); + static void decode_version_values(jint version, int * major, int * minor, + int * micro); // single stepping management methods static void at_single_stepping_point(JavaThread *thread, methodOop method, address location) KERNEL_RETURN;
--- a/src/share/vm/prims/jvmtiHpp.xsl Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/jvmtiHpp.xsl Wed Mar 24 17:16:33 2010 -0700 @@ -1,6 +1,6 @@ <?xml version="1.0"?> <!-- - Copyright 2002-2005 Sun Microsystems, Inc. All Rights Reserved. + Copyright 2002-2009 Sun Microsystems, Inc. All Rights Reserved. DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. This code is free software; you can redistribute it and/or modify it @@ -48,12 +48,12 @@ private: - JvmtiEnv(); + JvmtiEnv(jint version); ~JvmtiEnv(); public: - static JvmtiEnv* create_a_jvmti(); + static JvmtiEnv* create_a_jvmti(jint version); </xsl:text> <xsl:apply-templates select="functionsection"/>
--- a/src/share/vm/prims/jvmtiManageCapabilities.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/jvmtiManageCapabilities.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2003-2006 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2003-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -115,8 +115,10 @@ jvmtiCapabilities jc; memset(&jc, 0, sizeof(jc)); +#ifndef CC_INTERP jc.can_pop_frame = 1; jc.can_force_early_return = 1; +#endif // !CC_INTERP jc.can_get_source_debug_extension = 1; jc.can_access_local_variables = 1; jc.can_maintain_original_method_order = 1; @@ -355,7 +357,7 @@ avail.can_access_local_variables || avail.can_redefine_classes || avail.can_retransform_classes); - JvmtiExport::set_can_post_exceptions( + JvmtiExport::set_can_post_on_exceptions( avail.can_generate_exception_events || avail.can_generate_frame_pop_events || avail.can_generate_method_exit_events);
--- a/src/share/vm/prims/jvmtiRedefineClasses.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/jvmtiRedefineClasses.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -3214,7 +3214,7 @@ // - all instanceKlasses for redefined classes reused & contents updated the_class->vtable()->initialize_vtable(false, THREAD); the_class->itable()->initialize_itable(false, THREAD); - assert(!HAS_PENDING_EXCEPTION || (THREAD->pending_exception()->is_a(SystemDictionary::threaddeath_klass())), "redefine exception"); + assert(!HAS_PENDING_EXCEPTION || (THREAD->pending_exception()->is_a(SystemDictionary::ThreadDeath_klass())), "redefine exception"); } // Leave arrays of jmethodIDs and itable index cache unchanged
--- a/src/share/vm/prims/jvmtiTagMap.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/jvmtiTagMap.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -579,7 +579,7 @@ // If the object is a java.lang.Class then return the klassOop, // otherwise return the original object static inline oop klassOop_if_java_lang_Class(oop o) { - if (o->klass() == SystemDictionary::class_klass()) { + if (o->klass() == SystemDictionary::Class_klass()) { if (!java_lang_Class::is_primitive(o)) { o = (oop)java_lang_Class::as_klassOop(o); assert(o != NULL, "class for non-primitive mirror must exist"); @@ -644,7 +644,7 @@ } else { // if the object represents a runtime class then use the // tag for java.lang.Class - _klass = SystemDictionary::class_klass(); + _klass = SystemDictionary::Class_klass(); } _klass_tag = tag_for(tag_map, _klass); } @@ -747,7 +747,7 @@ // get referrer class tag. klassOop k = (_referrer == referrer) ? // Check if referrer is a class... _referrer->klass() // No, just get its class - : SystemDictionary::class_klass(); // Yes, its class is Class + : SystemDictionary::Class_klass(); // Yes, its class is Class _referrer_klass_tag = tag_for(tag_map, k); } } @@ -1126,7 +1126,7 @@ oop str, void* user_data) { - assert(str->klass() == SystemDictionary::string_klass(), "not a string"); + assert(str->klass() == SystemDictionary::String_klass(), "not a string"); // get the string value and length // (string value may be offset from the base) @@ -1186,7 +1186,7 @@ // for static fields only the index will be set static jvmtiHeapReferenceInfo reference_info = { 0 }; - assert(obj->klass() == SystemDictionary::class_klass(), "not a class"); + assert(obj->klass() == SystemDictionary::Class_klass(), "not a class"); if (java_lang_Class::is_primitive(obj)) { return 0; } @@ -1498,7 +1498,7 @@ if (callbacks()->primitive_field_callback != NULL && obj->is_instance()) { jint res; jvmtiPrimitiveFieldCallback cb = callbacks()->primitive_field_callback; - if (obj->klass() == SystemDictionary::class_klass()) { + if (obj->klass() == SystemDictionary::Class_klass()) { res = invoke_primitive_field_callback_for_static_fields(&wrapper, obj, cb, @@ -1515,7 +1515,7 @@ // string callback if (!is_array && callbacks()->string_primitive_value_callback != NULL && - obj->klass() == SystemDictionary::string_klass()) { + obj->klass() == SystemDictionary::String_klass()) { jint res = invoke_string_value_callback( callbacks()->string_primitive_value_callback, &wrapper, @@ -2381,7 +2381,7 @@ // invoke the string value callback inline bool CallbackInvoker::report_string_value(oop str) { - assert(str->klass() == SystemDictionary::string_klass(), "not a string"); + assert(str->klass() == SystemDictionary::String_klass(), "not a string"); AdvancedHeapWalkContext* context = advanced_context(); assert(context->string_primitive_value_callback() != NULL, "no callback"); @@ -2928,7 +2928,7 @@ // super (only if something more interesting than java.lang.Object) klassOop java_super = ik->java_super(); - if (java_super != NULL && java_super != SystemDictionary::object_klass()) { + if (java_super != NULL && java_super != SystemDictionary::Object_klass()) { oop super = Klass::cast(java_super)->java_mirror(); if (!CallbackInvoker::report_superclass_reference(mirror, super)) { return false; @@ -3070,7 +3070,7 @@ // if the object is a java.lang.String if (is_reporting_string_values() && - o->klass() == SystemDictionary::string_klass()) { + o->klass() == SystemDictionary::String_klass()) { if (!CallbackInvoker::report_string_value(o)) { return false; } @@ -3255,7 +3255,7 @@ // instance if (o->is_instance()) { - if (o->klass() == SystemDictionary::class_klass()) { + if (o->klass() == SystemDictionary::Class_klass()) { o = klassOop_if_java_lang_Class(o); if (o->is_klass()) { // a java.lang.Class
--- a/src/share/vm/prims/jvmtiThreadState.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/jvmtiThreadState.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2003-2006 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2003-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -381,6 +381,9 @@ static ByteSize earlyret_value_offset() { return byte_offset_of(JvmtiThreadState, _earlyret_value); } void oops_do(OopClosure* f); // GC support + +public: + void set_should_post_on_exceptions(bool val) { _thread->set_should_post_on_exceptions_flag(val ? JNI_TRUE : JNI_FALSE); } }; class RedefineVerifyMark : public StackObj {
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/share/vm/prims/methodHandleWalk.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,1401 @@ +/* + * Copyright 2008-2010 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/* + * JSR 292 reference implementation: method handle structure analysis + */ + +#include "incls/_precompiled.incl" +#include "incls/_methodHandleWalk.cpp.incl" + + +// ----------------------------------------------------------------------------- +// MethodHandleChain + +void MethodHandleChain::set_method_handle(Handle mh, TRAPS) { + if (!java_dyn_MethodHandle::is_instance(mh())) lose("bad method handle", CHECK); + + // set current method handle and unpack partially + _method_handle = mh; + _is_last = false; + _is_bound = false; + _arg_slot = -1; + _arg_type = T_VOID; + _conversion = -1; + _last_invoke = Bytecodes::_nop; //arbitrary non-garbage + + if (sun_dyn_DirectMethodHandle::is_instance(mh())) { + set_last_method(mh(), THREAD); + return; + } + if (sun_dyn_AdapterMethodHandle::is_instance(mh())) { + _conversion = AdapterMethodHandle_conversion(); + assert(_conversion != -1, "bad conv value"); + assert(sun_dyn_BoundMethodHandle::is_instance(mh()), "also BMH"); + } + if (sun_dyn_BoundMethodHandle::is_instance(mh())) { + if (!is_adapter()) // keep AMH and BMH separate in this model + _is_bound = true; + _arg_slot = BoundMethodHandle_vmargslot(); + oop target = MethodHandle_vmtarget_oop(); + if (!is_bound() || java_dyn_MethodHandle::is_instance(target)) { + _arg_type = compute_bound_arg_type(target, NULL, _arg_slot, CHECK); + } else if (target != NULL && target->is_method()) { + methodOop m = (methodOop) target; + _arg_type = compute_bound_arg_type(NULL, m, _arg_slot, CHECK); + set_last_method(mh(), CHECK); + } else { + _is_bound = false; // lose! + } + } + if (is_bound() && _arg_type == T_VOID) { + lose("bad vmargslot", CHECK); + } + if (!is_bound() && !is_adapter()) { + lose("unrecognized MH type", CHECK); + } +} + + +void MethodHandleChain::set_last_method(oop target, TRAPS) { + _is_last = true; + klassOop receiver_limit_oop = NULL; + int flags = 0; + methodOop m = MethodHandles::decode_method(target, receiver_limit_oop, flags); + _last_method = methodHandle(THREAD, m); + if ((flags & MethodHandles::_dmf_has_receiver) == 0) + _last_invoke = Bytecodes::_invokestatic; + else if ((flags & MethodHandles::_dmf_does_dispatch) == 0) + _last_invoke = Bytecodes::_invokespecial; + else if ((flags & MethodHandles::_dmf_from_interface) != 0) + _last_invoke = Bytecodes::_invokeinterface; + else + _last_invoke = Bytecodes::_invokevirtual; +} + + +BasicType MethodHandleChain::compute_bound_arg_type(oop target, methodOop m, int arg_slot, TRAPS) { + // There is no direct indication of whether the argument is primitive or not. + // It is implied by the _vmentry code, and by the MethodType of the target. + // FIXME: Make it explicit MethodHandleImpl refactors out from MethodHandle + BasicType arg_type = T_VOID; + if (target != NULL) { + oop mtype = java_dyn_MethodHandle::type(target); + int arg_num = MethodHandles::argument_slot_to_argnum(mtype, arg_slot); + if (arg_num >= 0) { + oop ptype = java_dyn_MethodType::ptype(mtype, arg_num); + arg_type = java_lang_Class::as_BasicType(ptype); + } + } else if (m != NULL) { + // figure out the argument type from the slot + // FIXME: make this explicit in the MH + int cur_slot = m->size_of_parameters(); + if (arg_slot >= cur_slot) + return T_VOID; + if (!m->is_static()) { + cur_slot -= type2size[T_OBJECT]; + if (cur_slot == arg_slot) + return T_OBJECT; + } + for (SignatureStream ss(m->signature()); !ss.is_done(); ss.next()) { + BasicType bt = ss.type(); + cur_slot -= type2size[bt]; + if (cur_slot <= arg_slot) { + if (cur_slot == arg_slot) + arg_type = bt; + break; + } + } + } + if (arg_type == T_ARRAY) + arg_type = T_OBJECT; + return arg_type; +} + + +void MethodHandleChain::lose(const char* msg, TRAPS) { + assert(false, "lose"); + _lose_message = msg; + if (!THREAD->is_Java_thread() || ((JavaThread*)THREAD)->thread_state() != _thread_in_vm) { + // throw a preallocated exception + THROW_OOP(Universe::virtual_machine_error_instance()); + } + THROW_MSG(vmSymbols::java_lang_InternalError(), msg); +} + + +// ----------------------------------------------------------------------------- +// MethodHandleWalker + +Bytecodes::Code MethodHandleWalker::conversion_code(BasicType src, BasicType dest) { + if (is_subword_type(src)) { + src = T_INT; // all subword src types act like int + } + if (src == dest) { + return Bytecodes::_nop; + } + +#define SRC_DEST(s,d) (((int)(s) << 4) + (int)(d)) + switch (SRC_DEST(src, dest)) { + case SRC_DEST(T_INT, T_LONG): return Bytecodes::_i2l; + case SRC_DEST(T_INT, T_FLOAT): return Bytecodes::_i2f; + case SRC_DEST(T_INT, T_DOUBLE): return Bytecodes::_i2d; + case SRC_DEST(T_INT, T_BYTE): return Bytecodes::_i2b; + case SRC_DEST(T_INT, T_CHAR): return Bytecodes::_i2c; + case SRC_DEST(T_INT, T_SHORT): return Bytecodes::_i2s; + + case SRC_DEST(T_LONG, T_INT): return Bytecodes::_l2i; + case SRC_DEST(T_LONG, T_FLOAT): return Bytecodes::_l2f; + case SRC_DEST(T_LONG, T_DOUBLE): return Bytecodes::_l2d; + + case SRC_DEST(T_FLOAT, T_INT): return Bytecodes::_f2i; + case SRC_DEST(T_FLOAT, T_LONG): return Bytecodes::_f2l; + case SRC_DEST(T_FLOAT, T_DOUBLE): return Bytecodes::_f2d; + + case SRC_DEST(T_DOUBLE, T_INT): return Bytecodes::_d2i; + case SRC_DEST(T_DOUBLE, T_LONG): return Bytecodes::_d2l; + case SRC_DEST(T_DOUBLE, T_FLOAT): return Bytecodes::_d2f; + } +#undef SRC_DEST + + // cannot do it in one step, or at all + return Bytecodes::_illegal; +} + + +// ----------------------------------------------------------------------------- +// MethodHandleWalker::walk +// +MethodHandleWalker::ArgToken +MethodHandleWalker::walk(TRAPS) { + ArgToken empty = ArgToken(); // Empty return value. + + walk_incoming_state(CHECK_(empty)); + + for (;;) { + set_method_handle(chain().method_handle_oop()); + + assert(_outgoing_argc == argument_count_slow(), "empty slots under control"); + + if (chain().is_adapter()) { + int conv_op = chain().adapter_conversion_op(); + int arg_slot = chain().adapter_arg_slot(); + SlotState* arg_state = slot_state(arg_slot); + if (arg_state == NULL + && conv_op > sun_dyn_AdapterMethodHandle::OP_RETYPE_RAW) { + lose("bad argument index", CHECK_(empty)); + } + + // perform the adapter action + switch (chain().adapter_conversion_op()) { + case sun_dyn_AdapterMethodHandle::OP_RETYPE_ONLY: + // No changes to arguments; pass the bits through. + break; + + case sun_dyn_AdapterMethodHandle::OP_RETYPE_RAW: { + // To keep the verifier happy, emit bitwise ("raw") conversions as needed. + // See MethodHandles::same_basic_type_for_arguments for allowed conversions. + Handle incoming_mtype(THREAD, chain().method_type_oop()); + oop outgoing_mh_oop = chain().vmtarget_oop(); + if (!java_dyn_MethodHandle::is_instance(outgoing_mh_oop)) + lose("outgoing target not a MethodHandle", CHECK_(empty)); + Handle outgoing_mtype(THREAD, java_dyn_MethodHandle::type(outgoing_mh_oop)); + outgoing_mh_oop = NULL; // GC safety + + int nptypes = java_dyn_MethodType::ptype_count(outgoing_mtype()); + if (nptypes != java_dyn_MethodType::ptype_count(incoming_mtype())) + lose("incoming and outgoing parameter count do not agree", CHECK_(empty)); + + for (int i = 0, slot = _outgoing.length() - 1; slot >= 0; slot--) { + SlotState* arg_state = slot_state(slot); + if (arg_state->_type == T_VOID) continue; + ArgToken arg = _outgoing.at(slot)._arg; + + klassOop in_klass = NULL; + klassOop out_klass = NULL; + BasicType inpbt = java_lang_Class::as_BasicType(java_dyn_MethodType::ptype(incoming_mtype(), i), &in_klass); + BasicType outpbt = java_lang_Class::as_BasicType(java_dyn_MethodType::ptype(outgoing_mtype(), i), &out_klass); + assert(inpbt == arg.basic_type(), "sanity"); + + if (inpbt != outpbt) { + vmIntrinsics::ID iid = vmIntrinsics::for_raw_conversion(inpbt, outpbt); + if (iid == vmIntrinsics::_none) { + lose("no raw conversion method", CHECK_(empty)); + } + ArgToken arglist[2]; + arglist[0] = arg; // outgoing 'this' + arglist[1] = ArgToken(); // sentinel + arg = make_invoke(NULL, iid, Bytecodes::_invokestatic, false, 1, &arglist[0], CHECK_(empty)); + change_argument(inpbt, slot, outpbt, arg); + } + + i++; // We need to skip void slots at the top of the loop. + } + + BasicType inrbt = java_lang_Class::as_BasicType(java_dyn_MethodType::rtype(incoming_mtype())); + BasicType outrbt = java_lang_Class::as_BasicType(java_dyn_MethodType::rtype(outgoing_mtype())); + if (inrbt != outrbt) { + if (inrbt == T_INT && outrbt == T_VOID) { + // See comments in MethodHandles::same_basic_type_for_arguments. + } else { + assert(false, "IMPLEMENT ME"); + lose("no raw conversion method", CHECK_(empty)); + } + } + break; + } + + case sun_dyn_AdapterMethodHandle::OP_CHECK_CAST: { + // checkcast the Nth outgoing argument in place + klassOop dest_klass = NULL; + BasicType dest = java_lang_Class::as_BasicType(chain().adapter_arg_oop(), &dest_klass); + assert(dest == T_OBJECT, ""); + assert(dest == arg_state->_type, ""); + ArgToken arg = arg_state->_arg; + ArgToken new_arg = make_conversion(T_OBJECT, dest_klass, Bytecodes::_checkcast, arg, CHECK_(empty)); + assert(arg.index() == new_arg.index(), "should be the same index"); + debug_only(dest_klass = (klassOop)badOop); + break; + } + + case sun_dyn_AdapterMethodHandle::OP_PRIM_TO_PRIM: { + // i2l, etc., on the Nth outgoing argument in place + BasicType src = chain().adapter_conversion_src_type(), + dest = chain().adapter_conversion_dest_type(); + Bytecodes::Code bc = conversion_code(src, dest); + ArgToken arg = arg_state->_arg; + if (bc == Bytecodes::_nop) { + break; + } else if (bc != Bytecodes::_illegal) { + arg = make_conversion(dest, NULL, bc, arg, CHECK_(empty)); + } else if (is_subword_type(dest)) { + bc = conversion_code(src, T_INT); + if (bc != Bytecodes::_illegal) { + arg = make_conversion(dest, NULL, bc, arg, CHECK_(empty)); + bc = conversion_code(T_INT, dest); + arg = make_conversion(dest, NULL, bc, arg, CHECK_(empty)); + } + } + if (bc == Bytecodes::_illegal) { + lose("bad primitive conversion", CHECK_(empty)); + } + change_argument(src, arg_slot, dest, arg); + break; + } + + case sun_dyn_AdapterMethodHandle::OP_REF_TO_PRIM: { + // checkcast to wrapper type & call intValue, etc. + BasicType dest = chain().adapter_conversion_dest_type(); + ArgToken arg = arg_state->_arg; + arg = make_conversion(T_OBJECT, SystemDictionary::box_klass(dest), + Bytecodes::_checkcast, arg, CHECK_(empty)); + vmIntrinsics::ID unboxer = vmIntrinsics::for_unboxing(dest); + if (unboxer == vmIntrinsics::_none) { + lose("no unboxing method", CHECK_(empty)); + } + ArgToken arglist[2]; + arglist[0] = arg; // outgoing 'this' + arglist[1] = ArgToken(); // sentinel + arg = make_invoke(NULL, unboxer, Bytecodes::_invokevirtual, false, 1, &arglist[0], CHECK_(empty)); + change_argument(T_OBJECT, arg_slot, dest, arg); + break; + } + + case sun_dyn_AdapterMethodHandle::OP_PRIM_TO_REF: { + // call wrapper type.valueOf + BasicType src = chain().adapter_conversion_src_type(); + ArgToken arg = arg_state->_arg; + vmIntrinsics::ID boxer = vmIntrinsics::for_boxing(src); + if (boxer == vmIntrinsics::_none) { + lose("no boxing method", CHECK_(empty)); + } + ArgToken arglist[2]; + arglist[0] = arg; // outgoing value + arglist[1] = ArgToken(); // sentinel + assert(false, "I think the argument count must be 1 instead of 0"); + arg = make_invoke(NULL, boxer, Bytecodes::_invokevirtual, false, 0, &arglist[0], CHECK_(empty)); + change_argument(src, arg_slot, T_OBJECT, arg); + break; + } + + case sun_dyn_AdapterMethodHandle::OP_SWAP_ARGS: { + int dest_arg_slot = chain().adapter_conversion_vminfo(); + if (!slot_has_argument(dest_arg_slot)) { + lose("bad swap index", CHECK_(empty)); + } + // a simple swap between two arguments + SlotState* dest_arg_state = slot_state(dest_arg_slot); + SlotState temp = (*dest_arg_state); + (*dest_arg_state) = (*arg_state); + (*arg_state) = temp; + break; + } + + case sun_dyn_AdapterMethodHandle::OP_ROT_ARGS: { + int dest_arg_slot = chain().adapter_conversion_vminfo(); + if (!slot_has_argument(dest_arg_slot) || arg_slot == dest_arg_slot) { + lose("bad rotate index", CHECK_(empty)); + } + SlotState* dest_arg_state = slot_state(dest_arg_slot); + // Rotate the source argument (plus following N slots) into the + // position occupied by the dest argument (plus following N slots). + int rotate_count = type2size[dest_arg_state->_type]; + // (no other rotate counts are currently supported) + if (arg_slot < dest_arg_slot) { + for (int i = 0; i < rotate_count; i++) { + SlotState temp = _outgoing.at(arg_slot); + _outgoing.remove_at(arg_slot); + _outgoing.insert_before(dest_arg_slot + rotate_count - 1, temp); + } + } else { // arg_slot > dest_arg_slot + for (int i = 0; i < rotate_count; i++) { + SlotState temp = _outgoing.at(arg_slot + rotate_count - 1); + _outgoing.remove_at(arg_slot + rotate_count - 1); + _outgoing.insert_before(dest_arg_slot, temp); + } + } + break; + } + + case sun_dyn_AdapterMethodHandle::OP_DUP_ARGS: { + int dup_slots = chain().adapter_conversion_stack_pushes(); + if (dup_slots <= 0) { + lose("bad dup count", CHECK_(empty)); + } + for (int i = 0; i < dup_slots; i++) { + SlotState* dup = slot_state(arg_slot + 2*i); + if (dup == NULL) break; // safety net + if (dup->_type != T_VOID) _outgoing_argc += 1; + _outgoing.insert_before(i, (*dup)); + } + break; + } + + case sun_dyn_AdapterMethodHandle::OP_DROP_ARGS: { + int drop_slots = -chain().adapter_conversion_stack_pushes(); + if (drop_slots <= 0) { + lose("bad drop count", CHECK_(empty)); + } + for (int i = 0; i < drop_slots; i++) { + SlotState* drop = slot_state(arg_slot); + if (drop == NULL) break; // safety net + if (drop->_type != T_VOID) _outgoing_argc -= 1; + _outgoing.remove_at(arg_slot); + } + break; + } + + case sun_dyn_AdapterMethodHandle::OP_COLLECT_ARGS: { //NYI, may GC + lose("unimplemented", CHECK_(empty)); + break; + } + + case sun_dyn_AdapterMethodHandle::OP_SPREAD_ARGS: { + klassOop array_klass_oop = NULL; + BasicType array_type = java_lang_Class::as_BasicType(chain().adapter_arg_oop(), + &array_klass_oop); + assert(array_type == T_OBJECT, ""); + assert(Klass::cast(array_klass_oop)->oop_is_array(), ""); + arrayKlassHandle array_klass(THREAD, array_klass_oop); + debug_only(array_klass_oop = (klassOop)badOop); + + klassOop element_klass_oop = NULL; + BasicType element_type = java_lang_Class::as_BasicType(array_klass->component_mirror(), + &element_klass_oop); + KlassHandle element_klass(THREAD, element_klass_oop); + debug_only(element_klass_oop = (klassOop)badOop); + + // Fetch the argument, which we will cast to the required array type. + assert(arg_state->_type == T_OBJECT, ""); + ArgToken array_arg = arg_state->_arg; + array_arg = make_conversion(T_OBJECT, array_klass(), Bytecodes::_checkcast, array_arg, CHECK_(empty)); + change_argument(T_OBJECT, arg_slot, T_VOID, ArgToken(tt_void)); + + // Check the required length. + int spread_slots = 1 + chain().adapter_conversion_stack_pushes(); + int spread_length = spread_slots; + if (type2size[element_type] == 2) { + if (spread_slots % 2 != 0) spread_slots = -1; // force error + spread_length = spread_slots / 2; + } + if (spread_slots < 0) { + lose("bad spread length", CHECK_(empty)); + } + + jvalue length_jvalue; length_jvalue.i = spread_length; + ArgToken length_arg = make_prim_constant(T_INT, &length_jvalue, CHECK_(empty)); + // Call a built-in method known to the JVM to validate the length. + ArgToken arglist[3]; + arglist[0] = array_arg; // value to check + arglist[1] = length_arg; // length to check + arglist[2] = ArgToken(); // sentinel + make_invoke(NULL, vmIntrinsics::_checkSpreadArgument, + Bytecodes::_invokestatic, false, 3, &arglist[0], CHECK_(empty)); + + // Spread out the array elements. + Bytecodes::Code aload_op = Bytecodes::_aaload; + if (element_type != T_OBJECT) { + lose("primitive array NYI", CHECK_(empty)); + } + int ap = arg_slot; + for (int i = 0; i < spread_length; i++) { + jvalue offset_jvalue; offset_jvalue.i = i; + ArgToken offset_arg = make_prim_constant(T_INT, &offset_jvalue, CHECK_(empty)); + ArgToken element_arg = make_fetch(element_type, element_klass(), aload_op, array_arg, offset_arg, CHECK_(empty)); + change_argument(T_VOID, ap, element_type, element_arg); + ap += type2size[element_type]; + } + break; + } + + case sun_dyn_AdapterMethodHandle::OP_FLYBY: //NYI, runs Java code + case sun_dyn_AdapterMethodHandle::OP_RICOCHET: //NYI, runs Java code + lose("unimplemented", CHECK_(empty)); + break; + + default: + lose("bad adapter conversion", CHECK_(empty)); + break; + } + } + + if (chain().is_bound()) { + // push a new argument + BasicType arg_type = chain().bound_arg_type(); + jint arg_slot = chain().bound_arg_slot(); + oop arg_oop = chain().bound_arg_oop(); + ArgToken arg; + if (arg_type == T_OBJECT) { + arg = make_oop_constant(arg_oop, CHECK_(empty)); + } else { + jvalue arg_value; + BasicType bt = java_lang_boxing_object::get_value(arg_oop, &arg_value); + if (bt == arg_type) { + arg = make_prim_constant(arg_type, &arg_value, CHECK_(empty)); + } else { + lose("bad bound value", CHECK_(empty)); + } + } + debug_only(arg_oop = badOop); + change_argument(T_VOID, arg_slot, arg_type, arg); + } + + // this test must come after the body of the loop + if (!chain().is_last()) { + chain().next(CHECK_(empty)); + } else { + break; + } + } + + // finish the sequence with a tail-call to the ultimate target + // parameters are passed in logical order (recv 1st), not slot order + ArgToken* arglist = NEW_RESOURCE_ARRAY(ArgToken, _outgoing.length() + 1); + int ap = 0; + for (int i = _outgoing.length() - 1; i >= 0; i--) { + SlotState* arg_state = slot_state(i); + if (arg_state->_type == T_VOID) continue; + arglist[ap++] = _outgoing.at(i)._arg; + } + assert(ap == _outgoing_argc, ""); + arglist[ap] = ArgToken(); // add a sentinel, for the sake of asserts + return make_invoke(chain().last_method_oop(), + vmIntrinsics::_none, + chain().last_invoke_code(), true, + ap, arglist, THREAD); +} + + +// ----------------------------------------------------------------------------- +// MethodHandleWalker::walk_incoming_state +// +void MethodHandleWalker::walk_incoming_state(TRAPS) { + Handle mtype(THREAD, chain().method_type_oop()); + int nptypes = java_dyn_MethodType::ptype_count(mtype()); + _outgoing_argc = nptypes; + int argp = nptypes - 1; + if (argp >= 0) { + _outgoing.at_grow(argp, make_state(T_VOID, ArgToken(tt_void))); // presize + } + for (int i = 0; i < nptypes; i++) { + klassOop arg_type_klass = NULL; + BasicType arg_type = java_lang_Class::as_BasicType( + java_dyn_MethodType::ptype(mtype(), i), &arg_type_klass); + int index = new_local_index(arg_type); + ArgToken arg = make_parameter(arg_type, arg_type_klass, index, CHECK); + debug_only(arg_type_klass = (klassOop) NULL); + _outgoing.at_put(argp, make_state(arg_type, arg)); + if (type2size[arg_type] == 2) { + // add the extra slot, so we can model the JVM stack + _outgoing.insert_before(argp+1, make_state(T_VOID, ArgToken(tt_void))); + } + --argp; + } + // call make_parameter at the end of the list for the return type + klassOop ret_type_klass = NULL; + BasicType ret_type = java_lang_Class::as_BasicType( + java_dyn_MethodType::rtype(mtype()), &ret_type_klass); + ArgToken ret = make_parameter(ret_type, ret_type_klass, -1, CHECK); + // ignore ret; client can catch it if needed +} + + +// ----------------------------------------------------------------------------- +// MethodHandleWalker::change_argument +// +// This is messy because some kinds of arguments are paired with +// companion slots containing an empty value. +void MethodHandleWalker::change_argument(BasicType old_type, int slot, BasicType new_type, + const ArgToken& new_arg) { + int old_size = type2size[old_type]; + int new_size = type2size[new_type]; + if (old_size == new_size) { + // simple case first + _outgoing.at_put(slot, make_state(new_type, new_arg)); + } else if (old_size > new_size) { + for (int i = old_size - 1; i >= new_size; i--) { + assert((i != 0) == (_outgoing.at(slot + i)._type == T_VOID), ""); + _outgoing.remove_at(slot + i); + } + if (new_size > 0) + _outgoing.at_put(slot, make_state(new_type, new_arg)); + else + _outgoing_argc -= 1; // deleted a real argument + } else { + for (int i = old_size; i < new_size; i++) { + _outgoing.insert_before(slot + i, make_state(T_VOID, ArgToken(tt_void))); + } + _outgoing.at_put(slot, make_state(new_type, new_arg)); + if (old_size == 0) + _outgoing_argc += 1; // inserted a real argument + } +} + + +#ifdef ASSERT +int MethodHandleWalker::argument_count_slow() { + int args_seen = 0; + for (int i = _outgoing.length() - 1; i >= 0; i--) { + if (_outgoing.at(i)._type != T_VOID) { + ++args_seen; + } + } + return args_seen; +} +#endif + + +// ----------------------------------------------------------------------------- +// MethodHandleCompiler + +MethodHandleCompiler::MethodHandleCompiler(Handle root, methodHandle callee, bool is_invokedynamic, TRAPS) + : MethodHandleWalker(root, is_invokedynamic, THREAD), + _callee(callee), + _thread(THREAD), + _bytecode(THREAD, 50), + _constants(THREAD, 10), + _cur_stack(0), + _max_stack(0), + _rtype(T_ILLEGAL) +{ + + // Element zero is always the null constant. + (void) _constants.append(NULL); + + // Set name and signature index. + _name_index = cpool_symbol_put(_callee->name()); + _signature_index = cpool_symbol_put(_callee->signature()); + + // Get return type klass. + Handle first_mtype(THREAD, chain().method_type_oop()); + // _rklass is NULL for primitives. + _rtype = java_lang_Class::as_BasicType(java_dyn_MethodType::rtype(first_mtype()), &_rklass); + if (_rtype == T_ARRAY) _rtype = T_OBJECT; + + int params = _callee->size_of_parameters(); // Incoming arguments plus receiver. + _num_params = for_invokedynamic() ? params - 1 : params; // XXX Check if callee is static? +} + + +// ----------------------------------------------------------------------------- +// MethodHandleCompiler::compile +// +// Compile this MethodHandle into a bytecode adapter and return a +// methodOop. +methodHandle MethodHandleCompiler::compile(TRAPS) { + assert(_thread == THREAD, "must be same thread"); + methodHandle nullHandle; + (void) walk(CHECK_(nullHandle)); + return get_method_oop(CHECK_(nullHandle)); +} + + +void MethodHandleCompiler::emit_bc(Bytecodes::Code op, int index) { + Bytecodes::check(op); // Are we legal? + + switch (op) { + // b + case Bytecodes::_aconst_null: + case Bytecodes::_iconst_m1: + case Bytecodes::_iconst_0: + case Bytecodes::_iconst_1: + case Bytecodes::_iconst_2: + case Bytecodes::_iconst_3: + case Bytecodes::_iconst_4: + case Bytecodes::_iconst_5: + case Bytecodes::_lconst_0: + case Bytecodes::_lconst_1: + case Bytecodes::_fconst_0: + case Bytecodes::_fconst_1: + case Bytecodes::_fconst_2: + case Bytecodes::_dconst_0: + case Bytecodes::_dconst_1: + case Bytecodes::_iload_0: + case Bytecodes::_iload_1: + case Bytecodes::_iload_2: + case Bytecodes::_iload_3: + case Bytecodes::_lload_0: + case Bytecodes::_lload_1: + case Bytecodes::_lload_2: + case Bytecodes::_lload_3: + case Bytecodes::_fload_0: + case Bytecodes::_fload_1: + case Bytecodes::_fload_2: + case Bytecodes::_fload_3: + case Bytecodes::_dload_0: + case Bytecodes::_dload_1: + case Bytecodes::_dload_2: + case Bytecodes::_dload_3: + case Bytecodes::_aload_0: + case Bytecodes::_aload_1: + case Bytecodes::_aload_2: + case Bytecodes::_aload_3: + case Bytecodes::_istore_0: + case Bytecodes::_istore_1: + case Bytecodes::_istore_2: + case Bytecodes::_istore_3: + case Bytecodes::_lstore_0: + case Bytecodes::_lstore_1: + case Bytecodes::_lstore_2: + case Bytecodes::_lstore_3: + case Bytecodes::_fstore_0: + case Bytecodes::_fstore_1: + case Bytecodes::_fstore_2: + case Bytecodes::_fstore_3: + case Bytecodes::_dstore_0: + case Bytecodes::_dstore_1: + case Bytecodes::_dstore_2: + case Bytecodes::_dstore_3: + case Bytecodes::_astore_0: + case Bytecodes::_astore_1: + case Bytecodes::_astore_2: + case Bytecodes::_astore_3: + case Bytecodes::_i2l: + case Bytecodes::_i2f: + case Bytecodes::_i2d: + case Bytecodes::_i2b: + case Bytecodes::_i2c: + case Bytecodes::_i2s: + case Bytecodes::_l2i: + case Bytecodes::_l2f: + case Bytecodes::_l2d: + case Bytecodes::_f2i: + case Bytecodes::_f2l: + case Bytecodes::_f2d: + case Bytecodes::_d2i: + case Bytecodes::_d2l: + case Bytecodes::_d2f: + case Bytecodes::_ireturn: + case Bytecodes::_lreturn: + case Bytecodes::_freturn: + case Bytecodes::_dreturn: + case Bytecodes::_areturn: + case Bytecodes::_return: + assert(strcmp(Bytecodes::format(op), "b") == 0, "wrong bytecode format"); + _bytecode.push(op); + break; + + // bi + case Bytecodes::_ldc: + case Bytecodes::_iload: + case Bytecodes::_lload: + case Bytecodes::_fload: + case Bytecodes::_dload: + case Bytecodes::_aload: + case Bytecodes::_istore: + case Bytecodes::_lstore: + case Bytecodes::_fstore: + case Bytecodes::_dstore: + case Bytecodes::_astore: + assert(strcmp(Bytecodes::format(op), "bi") == 0, "wrong bytecode format"); + assert((char) index == index, "index does not fit in 8-bit"); + _bytecode.push(op); + _bytecode.push(index); + break; + + // bii + case Bytecodes::_ldc2_w: + case Bytecodes::_checkcast: + assert(strcmp(Bytecodes::format(op), "bii") == 0, "wrong bytecode format"); + assert((short) index == index, "index does not fit in 16-bit"); + _bytecode.push(op); + _bytecode.push(index >> 8); + _bytecode.push(index); + break; + + // bjj + case Bytecodes::_invokestatic: + case Bytecodes::_invokespecial: + case Bytecodes::_invokevirtual: + assert(strcmp(Bytecodes::format(op), "bjj") == 0, "wrong bytecode format"); + assert((short) index == index, "index does not fit in 16-bit"); + _bytecode.push(op); + _bytecode.push(index >> 8); + _bytecode.push(index); + break; + + default: + ShouldNotReachHere(); + } +} + + +void MethodHandleCompiler::emit_load(BasicType bt, int index) { + if (index <= 3) { + switch (bt) { + case T_BOOLEAN: case T_BYTE: case T_CHAR: case T_SHORT: + case T_INT: emit_bc(Bytecodes::cast(Bytecodes::_iload_0 + index)); break; + case T_LONG: emit_bc(Bytecodes::cast(Bytecodes::_lload_0 + index)); break; + case T_FLOAT: emit_bc(Bytecodes::cast(Bytecodes::_fload_0 + index)); break; + case T_DOUBLE: emit_bc(Bytecodes::cast(Bytecodes::_dload_0 + index)); break; + case T_OBJECT: emit_bc(Bytecodes::cast(Bytecodes::_aload_0 + index)); break; + default: + ShouldNotReachHere(); + } + } + else { + switch (bt) { + case T_BOOLEAN: case T_BYTE: case T_CHAR: case T_SHORT: + case T_INT: emit_bc(Bytecodes::_iload, index); break; + case T_LONG: emit_bc(Bytecodes::_lload, index); break; + case T_FLOAT: emit_bc(Bytecodes::_fload, index); break; + case T_DOUBLE: emit_bc(Bytecodes::_dload, index); break; + case T_OBJECT: emit_bc(Bytecodes::_aload, index); break; + default: + ShouldNotReachHere(); + } + } + stack_push(bt); +} + +void MethodHandleCompiler::emit_store(BasicType bt, int index) { + if (index <= 3) { + switch (bt) { + case T_BOOLEAN: case T_BYTE: case T_CHAR: case T_SHORT: + case T_INT: emit_bc(Bytecodes::cast(Bytecodes::_istore_0 + index)); break; + case T_LONG: emit_bc(Bytecodes::cast(Bytecodes::_lstore_0 + index)); break; + case T_FLOAT: emit_bc(Bytecodes::cast(Bytecodes::_fstore_0 + index)); break; + case T_DOUBLE: emit_bc(Bytecodes::cast(Bytecodes::_dstore_0 + index)); break; + case T_OBJECT: emit_bc(Bytecodes::cast(Bytecodes::_astore_0 + index)); break; + default: + ShouldNotReachHere(); + } + } + else { + switch (bt) { + case T_BOOLEAN: case T_BYTE: case T_CHAR: case T_SHORT: + case T_INT: emit_bc(Bytecodes::_istore, index); break; + case T_LONG: emit_bc(Bytecodes::_lstore, index); break; + case T_FLOAT: emit_bc(Bytecodes::_fstore, index); break; + case T_DOUBLE: emit_bc(Bytecodes::_dstore, index); break; + case T_OBJECT: emit_bc(Bytecodes::_astore, index); break; + default: + ShouldNotReachHere(); + } + } + stack_pop(bt); +} + + +void MethodHandleCompiler::emit_load_constant(ArgToken arg) { + BasicType bt = arg.basic_type(); + switch (bt) { + case T_INT: { + jint value = arg.get_jint(); + if (-1 <= value && value <= 5) + emit_bc(Bytecodes::cast(Bytecodes::_iconst_0 + value)); + else + emit_bc(Bytecodes::_ldc, cpool_int_put(value)); + break; + } + case T_LONG: { + jlong value = arg.get_jlong(); + if (0 <= value && value <= 1) + emit_bc(Bytecodes::cast(Bytecodes::_lconst_0 + (int) value)); + else + emit_bc(Bytecodes::_ldc2_w, cpool_long_put(value)); + break; + } + case T_FLOAT: { + jfloat value = arg.get_jfloat(); + if (value == 0.0 || value == 1.0 || value == 2.0) + emit_bc(Bytecodes::cast(Bytecodes::_fconst_0 + (int) value)); + else + emit_bc(Bytecodes::_ldc, cpool_float_put(value)); + break; + } + case T_DOUBLE: { + jdouble value = arg.get_jdouble(); + if (value == 0.0 || value == 1.0) + emit_bc(Bytecodes::cast(Bytecodes::_dconst_0 + (int) value)); + else + emit_bc(Bytecodes::_ldc2_w, cpool_double_put(value)); + break; + } + case T_OBJECT: { + Handle value = arg.object(); + if (value.is_null()) + emit_bc(Bytecodes::_aconst_null); + else + emit_bc(Bytecodes::_ldc, cpool_object_put(value)); + break; + } + default: + ShouldNotReachHere(); + } + stack_push(bt); +} + + +MethodHandleWalker::ArgToken +MethodHandleCompiler::make_conversion(BasicType type, klassOop tk, Bytecodes::Code op, + const ArgToken& src, TRAPS) { + + BasicType srctype = src.basic_type(); + int index = src.index(); + + switch (op) { + case Bytecodes::_i2l: + case Bytecodes::_i2f: + case Bytecodes::_i2d: + case Bytecodes::_i2b: + case Bytecodes::_i2c: + case Bytecodes::_i2s: + + case Bytecodes::_l2i: + case Bytecodes::_l2f: + case Bytecodes::_l2d: + + case Bytecodes::_f2i: + case Bytecodes::_f2l: + case Bytecodes::_f2d: + + case Bytecodes::_d2i: + case Bytecodes::_d2l: + case Bytecodes::_d2f: + emit_load(srctype, index); + stack_pop(srctype); // pop the src type + emit_bc(op); + stack_push(type); // push the dest value + if (srctype != type) + index = new_local_index(type); + emit_store(type, index); + break; + + case Bytecodes::_checkcast: + emit_load(srctype, index); + emit_bc(op, cpool_klass_put(tk)); + emit_store(srctype, index); + break; + + default: + ShouldNotReachHere(); + } + + return make_parameter(type, tk, index, THREAD); +} + + +// ----------------------------------------------------------------------------- +// MethodHandleCompiler +// + +static jvalue zero_jvalue; + +// Emit bytecodes for the given invoke instruction. +MethodHandleWalker::ArgToken +MethodHandleCompiler::make_invoke(methodOop m, vmIntrinsics::ID iid, + Bytecodes::Code op, bool tailcall, + int argc, MethodHandleWalker::ArgToken* argv, + TRAPS) { + if (m == NULL) { + // Get the intrinsic methodOop. + m = vmIntrinsics::method_for(iid); + } + + klassOop klass = m->method_holder(); + symbolOop name = m->name(); + symbolOop signature = m->signature(); + + if (tailcall) { + // Actually, in order to make these methods more recognizable, + // let's put them in holder classes MethodHandle and InvokeDynamic. + // That way stack walkers and compiler heuristics can recognize them. + _target_klass = (for_invokedynamic() + ? SystemDictionary::InvokeDynamic_klass() + : SystemDictionary::MethodHandle_klass()); + } + + // instanceKlass* ik = instanceKlass::cast(klass); + // tty->print_cr("MethodHandleCompiler::make_invoke: %s %s.%s%s", Bytecodes::name(op), ik->external_name(), name->as_C_string(), signature->as_C_string()); + + // Inline the method. + InvocationCounter* ic = m->invocation_counter(); + ic->set_carry(); + + for (int i = 0; i < argc; i++) { + ArgToken arg = argv[i]; + TokenType tt = arg.token_type(); + BasicType bt = arg.basic_type(); + + switch (tt) { + case tt_parameter: + case tt_temporary: + emit_load(bt, arg.index()); + break; + case tt_constant: + emit_load_constant(arg); + break; + case tt_illegal: + // Sentinel. + assert(i == (argc - 1), "sentinel must be last entry"); + break; + case tt_void: + default: + ShouldNotReachHere(); + } + } + + // Populate constant pool. + int name_index = cpool_symbol_put(name); + int signature_index = cpool_symbol_put(signature); + int name_and_type_index = cpool_name_and_type_put(name_index, signature_index); + int klass_index = cpool_klass_put(klass); + int methodref_index = cpool_methodref_put(klass_index, name_and_type_index); + + // Generate invoke. + switch (op) { + case Bytecodes::_invokestatic: + case Bytecodes::_invokespecial: + case Bytecodes::_invokevirtual: + emit_bc(op, methodref_index); + break; + case Bytecodes::_invokeinterface: + Unimplemented(); + break; + default: + ShouldNotReachHere(); + } + + // If tailcall, we have walked all the way to a direct method handle. + // Otherwise, make a recursive call to some helper routine. + BasicType rbt = m->result_type(); + if (rbt == T_ARRAY) rbt = T_OBJECT; + ArgToken ret; + if (tailcall) { + if (rbt != _rtype) { + if (rbt == T_VOID) { + // push a zero of the right sort + ArgToken zero; + if (_rtype == T_OBJECT) { + zero = make_oop_constant(NULL, CHECK_(zero)); + } else { + zero = make_prim_constant(_rtype, &zero_jvalue, CHECK_(zero)); + } + emit_load_constant(zero); + } else if (_rtype == T_VOID) { + // We'll emit a _return with something on the stack. + // It's OK to ignore what's on the stack. + } else { + tty->print_cr("*** rbt=%d != rtype=%d", rbt, _rtype); + assert(false, "IMPLEMENT ME"); + } + } + switch (_rtype) { + case T_BOOLEAN: case T_BYTE: case T_CHAR: case T_SHORT: + case T_INT: emit_bc(Bytecodes::_ireturn); break; + case T_LONG: emit_bc(Bytecodes::_lreturn); break; + case T_FLOAT: emit_bc(Bytecodes::_freturn); break; + case T_DOUBLE: emit_bc(Bytecodes::_dreturn); break; + case T_VOID: emit_bc(Bytecodes::_return); break; + case T_OBJECT: + if (_rklass.not_null() && _rklass() != SystemDictionary::Object_klass()) + emit_bc(Bytecodes::_checkcast, cpool_klass_put(_rklass())); + emit_bc(Bytecodes::_areturn); + break; + default: ShouldNotReachHere(); + } + ret = ArgToken(); // Dummy return value. + } + else { + stack_push(rbt); // The return value is already pushed onto the stack. + int index = new_local_index(rbt); + switch (rbt) { + case T_BOOLEAN: case T_BYTE: case T_CHAR: case T_SHORT: + case T_INT: case T_LONG: case T_FLOAT: case T_DOUBLE: + case T_OBJECT: + emit_store(rbt, index); + ret = ArgToken(tt_temporary, rbt, index); + break; + case T_VOID: + ret = ArgToken(tt_void); + break; + default: + ShouldNotReachHere(); + } + } + + return ret; +} + +MethodHandleWalker::ArgToken +MethodHandleCompiler::make_fetch(BasicType type, klassOop tk, Bytecodes::Code op, + const MethodHandleWalker::ArgToken& base, + const MethodHandleWalker::ArgToken& offset, + TRAPS) { + Unimplemented(); + return ArgToken(); +} + + +int MethodHandleCompiler::cpool_primitive_put(BasicType bt, jvalue* con) { + jvalue con_copy; + assert(bt < T_OBJECT, ""); + if (type2aelembytes(bt) < jintSize) { + // widen to int + con_copy = (*con); + con = &con_copy; + switch (bt) { + case T_BOOLEAN: con->i = (con->z ? 1 : 0); break; + case T_BYTE: con->i = con->b; break; + case T_CHAR: con->i = con->c; break; + case T_SHORT: con->i = con->s; break; + default: ShouldNotReachHere(); + } + bt = T_INT; + } + +// for (int i = 1, imax = _constants.length(); i < imax; i++) { +// ConstantValue* con = _constants.at(i); +// if (con != NULL && con->is_primitive() && con->_type == bt) { +// bool match = false; +// switch (type2size[bt]) { +// case 1: if (pcon->_value.i == con->i) match = true; break; +// case 2: if (pcon->_value.j == con->j) match = true; break; +// } +// if (match) +// return i; +// } +// } + ConstantValue* cv = new ConstantValue(bt, *con); + int index = _constants.append(cv); + + // long and double entries take 2 slots, we add another empty entry. + if (type2size[bt] == 2) + (void) _constants.append(NULL); + + return index; +} + + +constantPoolHandle MethodHandleCompiler::get_constant_pool(TRAPS) const { + constantPoolHandle nullHandle; + bool is_conc_safe = true; + constantPoolOop cpool_oop = oopFactory::new_constantPool(_constants.length(), is_conc_safe, CHECK_(nullHandle)); + constantPoolHandle cpool(THREAD, cpool_oop); + + // Fill the real constant pool skipping the zero element. + for (int i = 1; i < _constants.length(); i++) { + ConstantValue* cv = _constants.at(i); + switch (cv->tag()) { + case JVM_CONSTANT_Utf8: cpool->symbol_at_put( i, cv->symbol_oop() ); break; + case JVM_CONSTANT_Integer: cpool->int_at_put( i, cv->get_jint() ); break; + case JVM_CONSTANT_Float: cpool->float_at_put( i, cv->get_jfloat() ); break; + case JVM_CONSTANT_Long: cpool->long_at_put( i, cv->get_jlong() ); break; + case JVM_CONSTANT_Double: cpool->double_at_put( i, cv->get_jdouble() ); break; + case JVM_CONSTANT_Class: cpool->klass_at_put( i, cv->klass_oop() ); break; + case JVM_CONSTANT_Methodref: cpool->method_at_put( i, cv->first_index(), cv->second_index()); break; + case JVM_CONSTANT_NameAndType: cpool->name_and_type_at_put(i, cv->first_index(), cv->second_index()); break; + case JVM_CONSTANT_Object: cpool->object_at_put( i, cv->object_oop() ); break; + default: ShouldNotReachHere(); + } + + switch (cv->tag()) { + case JVM_CONSTANT_Long: + case JVM_CONSTANT_Double: + i++; // Skip empty entry. + assert(_constants.at(i) == NULL, "empty entry"); + break; + } + } + + // Set the constant pool holder to the target method's class. + cpool->set_pool_holder(_target_klass()); + + return cpool; +} + + +methodHandle MethodHandleCompiler::get_method_oop(TRAPS) const { + methodHandle nullHandle; + // Create a method that holds the generated bytecode. invokedynamic + // has no receiver, normal MH calls do. + int flags_bits; + if (for_invokedynamic()) + flags_bits = (/*JVM_MH_INVOKE_BITS |*/ JVM_ACC_PUBLIC | JVM_ACC_FINAL | JVM_ACC_STATIC); + else + flags_bits = (/*JVM_MH_INVOKE_BITS |*/ JVM_ACC_PUBLIC | JVM_ACC_FINAL); + + bool is_conc_safe = true; + methodOop m_oop = oopFactory::new_method(bytecode_length(), + accessFlags_from(flags_bits), + 0, 0, 0, is_conc_safe, CHECK_(nullHandle)); + methodHandle m(THREAD, m_oop); + m_oop = NULL; // oop not GC safe + + constantPoolHandle cpool = get_constant_pool(CHECK_(nullHandle)); + m->set_constants(cpool()); + + m->set_name_index(_name_index); + m->set_signature_index(_signature_index); + + m->set_code((address) bytecode()); + + m->set_max_stack(_max_stack); + m->set_max_locals(max_locals()); + m->set_size_of_parameters(_num_params); + + typeArrayHandle exception_handlers(THREAD, Universe::the_empty_int_array()); + m->set_exception_table(exception_handlers()); + + // Set the carry bit of the invocation counter to force inlining of + // the adapter. + InvocationCounter* ic = m->invocation_counter(); + ic->set_carry(); + + // Rewrite the method and set up the constant pool cache. + objArrayOop m_array = oopFactory::new_system_objArray(1, CHECK_(nullHandle)); + objArrayHandle methods(THREAD, m_array); + methods->obj_at_put(0, m()); + Rewriter::rewrite(_target_klass(), cpool, methods, CHECK_(nullHandle)); // Use fake class. + +#ifndef PRODUCT + if (TraceMethodHandles) { + m->print(); + m->print_codes(); + } +#endif //PRODUCT + + return m; +} + + +#ifndef PRODUCT + +#if 0 +// MH printer for debugging. + +class MethodHandlePrinter : public MethodHandleWalker { +private: + outputStream* _out; + bool _verbose; + int _temp_num; + stringStream _strbuf; + const char* strbuf() { + const char* s = _strbuf.as_string(); + _strbuf.reset(); + return s; + } + ArgToken token(const char* str) { + return (ArgToken) str; + } + void start_params() { + _out->print("("); + } + void end_params() { + if (_verbose) _out->print("\n"); + _out->print(") => {"); + } + void put_type_name(BasicType type, klassOop tk, outputStream* s) { + const char* kname = NULL; + if (tk != NULL) + kname = Klass::cast(tk)->external_name(); + s->print("%s", (kname != NULL) ? kname : type2name(type)); + } + ArgToken maybe_make_temp(const char* statement_op, BasicType type, const char* temp_name) { + const char* value = strbuf(); + if (!_verbose) return token(value); + // make an explicit binding for each separate value + _strbuf.print("%s%d", temp_name, ++_temp_num); + const char* temp = strbuf(); + _out->print("\n %s %s %s = %s;", statement_op, type2name(type), temp, value); + return token(temp); + } + +public: + MethodHandlePrinter(Handle root, bool verbose, outputStream* out, TRAPS) + : MethodHandleWalker(root, THREAD), + _out(out), + _verbose(verbose), + _temp_num(0) + { + start_params(); + } + virtual ArgToken make_parameter(BasicType type, klassOop tk, int argnum, TRAPS) { + if (argnum < 0) { + end_params(); + return NULL; + } + if (argnum == 0) { + _out->print(_verbose ? "\n " : ""); + } else { + _out->print(_verbose ? ",\n " : ", "); + } + if (argnum >= _temp_num) + _temp_num = argnum; + // generate an argument name + _strbuf.print("a%d", argnum); + const char* arg = strbuf(); + put_type_name(type, tk, _out); + _out->print(" %s", arg); + return token(arg); + } + virtual ArgToken make_oop_constant(oop con, TRAPS) { + if (con == NULL) + _strbuf.print("null"); + else + con->print_value_on(&_strbuf); + if (_strbuf.size() == 0) { // yuck + _strbuf.print("(a "); + put_type_name(T_OBJECT, con->klass(), &_strbuf); + _strbuf.print(")"); + } + return maybe_make_temp("constant", T_OBJECT, "k"); + } + virtual ArgToken make_prim_constant(BasicType type, jvalue* con, TRAPS) { + java_lang_boxing_object::print(type, con, &_strbuf); + return maybe_make_temp("constant", type, "k"); + } + virtual ArgToken make_conversion(BasicType type, klassOop tk, Bytecodes::Code op, ArgToken src, TRAPS) { + _strbuf.print("%s(%s", Bytecodes::name(op), (const char*)src); + if (tk != NULL) { + _strbuf.print(", "); + put_type_name(type, tk, &_strbuf); + } + _strbuf.print(")"); + return maybe_make_temp("convert", type, "v"); + } + virtual ArgToken make_fetch(BasicType type, klassOop tk, Bytecodes::Code op, ArgToken base, ArgToken offset, TRAPS) { + _strbuf.print("%s(%s, %s", Bytecodes::name(op), (const char*)base, (const char*)offset); + if (tk != NULL) { + _strbuf.print(", "); + put_type_name(type, tk, &_strbuf); + } + _strbuf.print(")"); + return maybe_make_temp("fetch", type, "x"); + } + virtual ArgToken make_invoke(methodOop m, vmIntrinsics::ID iid, + Bytecodes::Code op, bool tailcall, + int argc, ArgToken* argv, TRAPS) { + symbolOop name, sig; + if (m != NULL) { + name = m->name(); + sig = m->signature(); + } else { + name = vmSymbols::symbol_at(vmIntrinsics::name_for(iid)); + sig = vmSymbols::symbol_at(vmIntrinsics::signature_for(iid)); + } + _strbuf.print("%s %s%s(", Bytecodes::name(op), name->as_C_string(), sig->as_C_string()); + for (int i = 0; i < argc; i++) { + _strbuf.print("%s%s", (i > 0 ? ", " : ""), (const char*)argv[i]); + } + _strbuf.print(")"); + if (!tailcall) { + BasicType rt = char2type(sig->byte_at(sig->utf8_length()-1)); + if (rt == T_ILLEGAL) rt = T_OBJECT; // ';' at the end of '(...)L...;' + return maybe_make_temp("invoke", rt, "x"); + } else { + const char* ret = strbuf(); + _out->print(_verbose ? "\n return " : " "); + _out->print("%s", ret); + _out->print(_verbose ? "\n}\n" : " }"); + } + return ArgToken(); + } + + virtual void set_method_handle(oop mh) { + if (WizardMode && Verbose) { + tty->print("\n--- next target: "); + mh->print(); + } + } + + static void print(Handle root, bool verbose, outputStream* out, TRAPS) { + ResourceMark rm; + MethodHandlePrinter printer(root, verbose, out, CHECK); + printer.walk(CHECK); + out->print("\n"); + } + static void print(Handle root, bool verbose = Verbose, outputStream* out = tty) { + EXCEPTION_MARK; + ResourceMark rm; + MethodHandlePrinter printer(root, verbose, out, THREAD); + if (!HAS_PENDING_EXCEPTION) + printer.walk(THREAD); + if (HAS_PENDING_EXCEPTION) { + oop ex = PENDING_EXCEPTION; + CLEAR_PENDING_EXCEPTION; + out->print("\n*** "); + if (ex != Universe::virtual_machine_error_instance()) + ex->print_on(out); + else + out->print("lose: %s", printer.lose_message()); + out->print("\n}\n"); + } + out->print("\n"); + } +}; +#endif // 0 + +extern "C" +void print_method_handle(oop mh) { + if (java_dyn_MethodHandle::is_instance(mh)) { + //MethodHandlePrinter::print(mh); + } else { + tty->print("*** not a method handle: "); + mh->print(); + } +} + +#endif // PRODUCT
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/share/vm/prims/methodHandleWalk.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,413 @@ +/* + * Copyright 2008-2010 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +// Low-level parser for method handle chains. +class MethodHandleChain : StackObj { +public: + typedef MethodHandles::EntryKind EntryKind; + +private: + Handle _root; // original target + Handle _method_handle; // current target + bool _is_last; // final guy in chain + bool _is_bound; // has a bound argument + BasicType _arg_type; // if is_bound, the bound argument type + int _arg_slot; // if is_bound or is_adapter, affected argument slot + jint _conversion; // conversion field of AMH or -1 + methodHandle _last_method; // if is_last, which method we target + Bytecodes::Code _last_invoke; // if is_last, type of invoke + const char* _lose_message; // saved argument to lose() + + void set_method_handle(Handle target, TRAPS); + void set_last_method(oop target, TRAPS); + static BasicType compute_bound_arg_type(oop target, methodOop m, int arg_slot, TRAPS); + + oop MethodHandle_type_oop() { return java_dyn_MethodHandle::type(method_handle_oop()); } + oop MethodHandle_vmtarget_oop() { return java_dyn_MethodHandle::vmtarget(method_handle_oop()); } + int MethodHandle_vmslots() { return java_dyn_MethodHandle::vmslots(method_handle_oop()); } + int DirectMethodHandle_vmindex() { return sun_dyn_DirectMethodHandle::vmindex(method_handle_oop()); } + oop BoundMethodHandle_argument_oop() { return sun_dyn_BoundMethodHandle::argument(method_handle_oop()); } + int BoundMethodHandle_vmargslot() { return sun_dyn_BoundMethodHandle::vmargslot(method_handle_oop()); } + int AdapterMethodHandle_conversion() { return sun_dyn_AdapterMethodHandle::conversion(method_handle_oop()); } + +public: + MethodHandleChain(Handle root, TRAPS) + : _root(root) + { set_method_handle(root, THREAD); } + + bool is_adapter() { return _conversion != -1; } + bool is_bound() { return _is_bound; } + bool is_last() { return _is_last; } + + void next(TRAPS) { + assert(!is_last(), ""); + set_method_handle(MethodHandle_vmtarget_oop(), THREAD); + } + + Handle method_handle() { return _method_handle; } + oop method_handle_oop() { return _method_handle(); } + oop method_type_oop() { return MethodHandle_type_oop(); } + oop vmtarget_oop() { return MethodHandle_vmtarget_oop(); } + + jint adapter_conversion() { assert(is_adapter(), ""); return _conversion; } + int adapter_conversion_op() { return MethodHandles::adapter_conversion_op(adapter_conversion()); } + BasicType adapter_conversion_src_type() + { return MethodHandles::adapter_conversion_src_type(adapter_conversion()); } + BasicType adapter_conversion_dest_type() + { return MethodHandles::adapter_conversion_dest_type(adapter_conversion()); } + int adapter_conversion_stack_move() + { return MethodHandles::adapter_conversion_stack_move(adapter_conversion()); } + int adapter_conversion_stack_pushes() + { return adapter_conversion_stack_move() / MethodHandles::stack_move_unit(); } + int adapter_conversion_vminfo() + { return MethodHandles::adapter_conversion_vminfo(adapter_conversion()); } + int adapter_arg_slot() { assert(is_adapter(), ""); return _arg_slot; } + oop adapter_arg_oop() { assert(is_adapter(), ""); return BoundMethodHandle_argument_oop(); } + + BasicType bound_arg_type() { assert(is_bound(), ""); return _arg_type; } + int bound_arg_slot() { assert(is_bound(), ""); return _arg_slot; } + oop bound_arg_oop() { assert(is_bound(), ""); return BoundMethodHandle_argument_oop(); } + + methodOop last_method_oop() { assert(is_last(), ""); return _last_method(); } + Bytecodes::Code last_invoke_code() { assert(is_last(), ""); return _last_invoke; } + + void lose(const char* msg, TRAPS); + const char* lose_message() { return _lose_message; } +}; + + +// Structure walker for method handles. +// Does abstract interpretation on top of low-level parsing. +// You supply the tokens shuffled by the abstract interpretation. +class MethodHandleWalker : StackObj { +public: + // Stack values: + enum TokenType { + tt_void, + tt_parameter, + tt_temporary, + tt_constant, + tt_illegal + }; + + // Argument token: + class ArgToken { + private: + TokenType _tt; + BasicType _bt; + jvalue _value; + Handle _handle; + + public: + ArgToken(TokenType tt = tt_illegal) : _tt(tt) {} + ArgToken(TokenType tt, BasicType bt, jvalue value) : _tt(tt), _bt(bt), _value(value) {} + + ArgToken(TokenType tt, BasicType bt, int index) : _tt(tt), _bt(bt) { + _value.i = index; + } + + ArgToken(TokenType tt, BasicType bt, Handle value) : _tt(tt), _bt(bt) { + _handle = value; + } + + TokenType token_type() const { return _tt; } + BasicType basic_type() const { return _bt; } + int index() const { return _value.i; } + Handle object() const { return _handle; } + + jint get_jint() const { return _value.i; } + jlong get_jlong() const { return _value.j; } + jfloat get_jfloat() const { return _value.f; } + jdouble get_jdouble() const { return _value.d; } + }; + + // Abstract interpretation state: + struct SlotState { + BasicType _type; + ArgToken _arg; + SlotState() : _type(), _arg() {} + }; + static SlotState make_state(BasicType type, ArgToken arg) { + SlotState ss; + ss._type = type; ss._arg = arg; + return ss; + } + +private: + MethodHandleChain _chain; + bool _for_invokedynamic; + int _local_index; + + GrowableArray<SlotState> _outgoing; // current outgoing parameter slots + int _outgoing_argc; // # non-empty outgoing slots + + // Replace a value of type old_type at slot (and maybe slot+1) with the new value. + // If old_type != T_VOID, remove the old argument at that point. + // If new_type != T_VOID, insert the new argument at that point. + // Insert or delete a second empty slot as needed. + void change_argument(BasicType old_type, int slot, BasicType new_type, const ArgToken& new_arg); + + SlotState* slot_state(int slot) { + if (slot < 0 || slot >= _outgoing.length()) + return NULL; + return _outgoing.adr_at(slot); + } + BasicType slot_type(int slot) { + SlotState* ss = slot_state(slot); + if (ss == NULL) + return T_ILLEGAL; + return ss->_type; + } + bool slot_has_argument(int slot) { + return slot_type(slot) < T_VOID; + } + +#ifdef ASSERT + int argument_count_slow(); +#endif + + // Return a bytecode for converting src to dest, if one exists. + Bytecodes::Code conversion_code(BasicType src, BasicType dest); + + void walk_incoming_state(TRAPS); + +public: + MethodHandleWalker(Handle root, bool for_invokedynamic, TRAPS) + : _chain(root, THREAD), + _for_invokedynamic(for_invokedynamic), + _outgoing(THREAD, 10), + _outgoing_argc(0) + { + _local_index = for_invokedynamic ? 0 : 1; + } + + MethodHandleChain& chain() { return _chain; } + + bool for_invokedynamic() const { return _for_invokedynamic; } + + int new_local_index(BasicType bt) { + //int index = _for_invokedynamic ? _local_index : _local_index - 1; + int index = _local_index; + _local_index += type2size[bt]; + return index; + } + + int max_locals() const { return _local_index; } + + // plug-in abstract interpretation steps: + virtual ArgToken make_parameter( BasicType type, klassOop tk, int argnum, TRAPS ) = 0; + virtual ArgToken make_prim_constant( BasicType type, jvalue* con, TRAPS ) = 0; + virtual ArgToken make_oop_constant( oop con, TRAPS ) = 0; + virtual ArgToken make_conversion( BasicType type, klassOop tk, Bytecodes::Code op, const ArgToken& src, TRAPS ) = 0; + virtual ArgToken make_fetch( BasicType type, klassOop tk, Bytecodes::Code op, const ArgToken& base, const ArgToken& offset, TRAPS ) = 0; + virtual ArgToken make_invoke( methodOop m, vmIntrinsics::ID iid, Bytecodes::Code op, bool tailcall, int argc, ArgToken* argv, TRAPS ) = 0; + + // For make_invoke, the methodOop can be NULL if the intrinsic ID + // is something other than vmIntrinsics::_none. + + // and in case anyone cares to related the previous actions to the chain: + virtual void set_method_handle(oop mh) { } + + void lose(const char* msg, TRAPS) { chain().lose(msg, THREAD); } + const char* lose_message() { return chain().lose_message(); } + + ArgToken walk(TRAPS); +}; + + +// An abstract interpreter for method handle chains. +// Produces an account of the semantics of a chain, in terms of a static IR. +// The IR happens to be JVM bytecodes. +class MethodHandleCompiler : public MethodHandleWalker { +private: + methodHandle _callee; + KlassHandle _rklass; // Return type for casting. + BasicType _rtype; + KlassHandle _target_klass; + Thread* _thread; + + // Fake constant pool entry. + class ConstantValue { + private: + int _tag; // Constant pool tag type. + JavaValue _value; + Handle _handle; + + public: + // Constructor for oop types. + ConstantValue(int tag, Handle con) : _tag(tag), _handle(con) { + assert(tag == JVM_CONSTANT_Utf8 || + tag == JVM_CONSTANT_Class || + tag == JVM_CONSTANT_String || + tag == JVM_CONSTANT_Object, "must be oop type"); + } + + // Constructor for oop reference types. + ConstantValue(int tag, int index) : _tag(tag) { + assert(JVM_CONSTANT_Fieldref <= tag && tag <= JVM_CONSTANT_NameAndType, "must be ref type"); + _value.set_jint(index); + } + ConstantValue(int tag, int first_index, int second_index) : _tag(tag) { + assert(JVM_CONSTANT_Fieldref <= tag && tag <= JVM_CONSTANT_NameAndType, "must be ref type"); + _value.set_jint(first_index << 16 | second_index); + } + + // Constructor for primitive types. + ConstantValue(BasicType bt, jvalue con) { + _value.set_type(bt); + switch (bt) { + case T_INT: _tag = JVM_CONSTANT_Integer; _value.set_jint( con.i); break; + case T_LONG: _tag = JVM_CONSTANT_Long; _value.set_jlong( con.j); break; + case T_FLOAT: _tag = JVM_CONSTANT_Float; _value.set_jfloat( con.f); break; + case T_DOUBLE: _tag = JVM_CONSTANT_Double; _value.set_jdouble(con.d); break; + default: ShouldNotReachHere(); + } + } + + int tag() const { return _tag; } + symbolOop symbol_oop() const { return (symbolOop) _handle(); } + klassOop klass_oop() const { return (klassOop) _handle(); } + oop object_oop() const { return _handle(); } + int index() const { return _value.get_jint(); } + int first_index() const { return _value.get_jint() >> 16; } + int second_index() const { return _value.get_jint() & 0x0000FFFF; } + + bool is_primitive() const { return is_java_primitive(_value.get_type()); } + jint get_jint() const { return _value.get_jint(); } + jlong get_jlong() const { return _value.get_jlong(); } + jfloat get_jfloat() const { return _value.get_jfloat(); } + jdouble get_jdouble() const { return _value.get_jdouble(); } + }; + + // Fake constant pool. + GrowableArray<ConstantValue*> _constants; + + // Accumulated compiler state: + GrowableArray<unsigned char> _bytecode; + + int _cur_stack; + int _max_stack; + int _num_params; + int _name_index; + int _signature_index; + + void stack_push(BasicType bt) { + _cur_stack += type2size[bt]; + if (_cur_stack > _max_stack) _max_stack = _cur_stack; + } + void stack_pop(BasicType bt) { + _cur_stack -= type2size[bt]; + assert(_cur_stack >= 0, "sanity"); + } + + unsigned char* bytecode() const { return _bytecode.adr_at(0); } + int bytecode_length() const { return _bytecode.length(); } + + // Fake constant pool. + int cpool_oop_put(int tag, Handle con) { + if (con.is_null()) return 0; + ConstantValue* cv = new ConstantValue(tag, con); + return _constants.append(cv); + } + + int cpool_oop_reference_put(int tag, int first_index, int second_index) { + if (first_index == 0 && second_index == 0) return 0; + assert(first_index != 0 && second_index != 0, "no zero indexes"); + ConstantValue* cv = new ConstantValue(tag, first_index, second_index); + return _constants.append(cv); + } + + int cpool_primitive_put(BasicType type, jvalue* con); + + int cpool_int_put(jint value) { + jvalue con; con.i = value; + return cpool_primitive_put(T_INT, &con); + } + int cpool_long_put(jlong value) { + jvalue con; con.j = value; + return cpool_primitive_put(T_LONG, &con); + } + int cpool_float_put(jfloat value) { + jvalue con; con.f = value; + return cpool_primitive_put(T_FLOAT, &con); + } + int cpool_double_put(jdouble value) { + jvalue con; con.d = value; + return cpool_primitive_put(T_DOUBLE, &con); + } + + int cpool_object_put(Handle obj) { + return cpool_oop_put(JVM_CONSTANT_Object, obj); + } + int cpool_symbol_put(symbolOop sym) { + return cpool_oop_put(JVM_CONSTANT_Utf8, sym); + } + int cpool_klass_put(klassOop klass) { + return cpool_oop_put(JVM_CONSTANT_Class, klass); + } + int cpool_methodref_put(int class_index, int name_and_type_index) { + return cpool_oop_reference_put(JVM_CONSTANT_Methodref, class_index, name_and_type_index); + } + int cpool_name_and_type_put(int name_index, int signature_index) { + return cpool_oop_reference_put(JVM_CONSTANT_NameAndType, name_index, signature_index); + } + + void emit_bc(Bytecodes::Code op, int index = 0); + void emit_load(BasicType bt, int index); + void emit_store(BasicType bt, int index); + void emit_load_constant(ArgToken arg); + + virtual ArgToken make_parameter(BasicType type, klassOop tk, int argnum, TRAPS) { + return ArgToken(tt_parameter, type, argnum); + } + virtual ArgToken make_oop_constant(oop con, TRAPS) { + Handle h(THREAD, con); + return ArgToken(tt_constant, T_OBJECT, h); + } + virtual ArgToken make_prim_constant(BasicType type, jvalue* con, TRAPS) { + return ArgToken(tt_constant, type, *con); + } + + virtual ArgToken make_conversion(BasicType type, klassOop tk, Bytecodes::Code op, const ArgToken& src, TRAPS); + virtual ArgToken make_fetch(BasicType type, klassOop tk, Bytecodes::Code op, const ArgToken& base, const ArgToken& offset, TRAPS); + virtual ArgToken make_invoke(methodOop m, vmIntrinsics::ID iid, Bytecodes::Code op, bool tailcall, int argc, ArgToken* argv, TRAPS); + + // Get a real constant pool. + constantPoolHandle get_constant_pool(TRAPS) const; + + // Get a real methodOop. + methodHandle get_method_oop(TRAPS) const; + +public: + MethodHandleCompiler(Handle root, methodHandle call_method, bool for_invokedynamic, TRAPS); + + // Compile the given MH chain into bytecode. + methodHandle compile(TRAPS); + + // Tests if the given class is a MH adapter holder. + static bool klass_is_method_handle_adapter_holder(klassOop klass) { + return (klass == SystemDictionary::MethodHandle_klass() || + klass == SystemDictionary::InvokeDynamic_klass()); + } +};
--- a/src/share/vm/prims/methodHandles.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/methodHandles.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -132,8 +132,9 @@ } return m; } else { + assert(vmtarget->is_klass(), "must be class or interface"); decode_flags_result |= MethodHandles::_dmf_does_dispatch; - assert(vmtarget->is_klass(), "must be class or interface"); + decode_flags_result |= MethodHandles::_dmf_has_receiver; receiver_limit_result = (klassOop)vmtarget; Klass* tk = Klass::cast((klassOop)vmtarget); if (tk->is_interface()) { @@ -142,7 +143,7 @@ return klassItable::method_for_itable_index((klassOop)vmtarget, vmindex); } else { if (!tk->oop_is_instance()) - tk = instanceKlass::cast(SystemDictionary::object_klass()); + tk = instanceKlass::cast(SystemDictionary::Object_klass()); return ((instanceKlass*)tk)->method_at_vtable(vmindex); } } @@ -179,8 +180,10 @@ // short-circuits directly to the methodOop. // (It might be another argument besides a receiver also.) assert(target->is_method(), "must be a simple method"); + decode_flags_result |= MethodHandles::_dmf_binds_method; methodOop m = (methodOop) target; - decode_flags_result |= MethodHandles::_dmf_binds_method; + if (!m->is_static()) + decode_flags_result |= MethodHandles::_dmf_has_receiver; return m; } } @@ -233,8 +236,8 @@ BasicType recv_bt = char2type(sig->byte_at(1)); // Note: recv_bt might be T_ILLEGAL if byte_at(2) is ')' assert(sig->byte_at(0) == '(', "must be method sig"); - if (recv_bt == T_OBJECT || recv_bt == T_ARRAY) - decode_flags_result |= _dmf_has_receiver; +// if (recv_bt == T_OBJECT || recv_bt == T_ARRAY) +// decode_flags_result |= _dmf_has_receiver; } else { // non-static method decode_flags_result |= _dmf_has_receiver; @@ -261,14 +264,14 @@ return decode_MemberName(x, receiver_limit_result, decode_flags_result); } else if (java_dyn_MethodHandle::is_subclass(xk)) { return decode_MethodHandle(x, receiver_limit_result, decode_flags_result); - } else if (xk == SystemDictionary::reflect_method_klass()) { + } else if (xk == SystemDictionary::reflect_Method_klass()) { oop clazz = java_lang_reflect_Method::clazz(x); int slot = java_lang_reflect_Method::slot(x); klassOop k = java_lang_Class::as_klassOop(clazz); if (k != NULL && Klass::cast(k)->oop_is_instance()) return decode_methodOop(instanceKlass::cast(k)->method_with_idnum(slot), decode_flags_result); - } else if (xk == SystemDictionary::reflect_constructor_klass()) { + } else if (xk == SystemDictionary::reflect_Constructor_klass()) { oop clazz = java_lang_reflect_Constructor::clazz(x); int slot = java_lang_reflect_Constructor::slot(x); klassOop k = java_lang_Class::as_klassOop(clazz); @@ -325,7 +328,7 @@ }; void MethodHandles::init_MemberName(oop mname_oop, oop target_oop) { - if (target_oop->klass() == SystemDictionary::reflect_field_klass()) { + if (target_oop->klass() == SystemDictionary::reflect_Field_klass()) { oop clazz = java_lang_reflect_Field::clazz(target_oop); // fd.field_holder() int slot = java_lang_reflect_Field::slot(target_oop); // fd.index() int mods = java_lang_reflect_Field::modifiers(target_oop); @@ -410,7 +413,7 @@ if (defc_klassOop == NULL) return; // a primitive; no resolution possible if (!Klass::cast(defc_klassOop)->oop_is_instance()) { if (!Klass::cast(defc_klassOop)->oop_is_array()) return; - defc_klassOop = SystemDictionary::object_klass(); + defc_klassOop = SystemDictionary::Object_klass(); } instanceKlassHandle defc(THREAD, defc_klassOop); defc_klassOop = NULL; // safety @@ -746,7 +749,7 @@ return NULL; // unformed MH } klassOop tklass = target->klass(); - if (Klass::cast(tklass)->is_subclass_of(SystemDictionary::object_klass())) { + if (Klass::cast(tklass)->is_subclass_of(SystemDictionary::Object_klass())) { return target; // target is another MH (or something else?) } } @@ -818,26 +821,26 @@ for (int i = 0; ; i++) { const char* test_name = always_null_names[i]; if (test_name == NULL) break; - if (name->equals(test_name, (int) strlen(test_name))) + if (name->equals(test_name)) return true; } return false; } bool MethodHandles::class_cast_needed(klassOop src, klassOop dst) { - if (src == dst || dst == SystemDictionary::object_klass()) + if (src == dst || dst == SystemDictionary::Object_klass()) return false; // quickest checks Klass* srck = Klass::cast(src); Klass* dstk = Klass::cast(dst); if (dstk->is_interface()) { // interface receivers can safely be viewed as untyped, // because interface calls always include a dynamic check - //dstk = Klass::cast(SystemDictionary::object_klass()); + //dstk = Klass::cast(SystemDictionary::Object_klass()); return false; } if (srck->is_interface()) { // interface arguments must be viewed as untyped - //srck = Klass::cast(SystemDictionary::object_klass()); + //srck = Klass::cast(SystemDictionary::Object_klass()); return true; } if (is_always_null_type(src)) { @@ -850,7 +853,7 @@ } static oop object_java_mirror() { - return Klass::cast(SystemDictionary::object_klass())->java_mirror(); + return Klass::cast(SystemDictionary::Object_klass())->java_mirror(); } bool MethodHandles::same_basic_type_for_arguments(BasicType src, @@ -1446,7 +1449,7 @@ break; } // check subrange of Integer.value, if necessary - if (argument == NULL || argument->klass() != SystemDictionary::int_klass()) { + if (argument == NULL || argument->klass() != SystemDictionary::Integer_klass()) { err = "bound integer argument must be of type java.lang.Integer"; break; } @@ -1469,7 +1472,7 @@ BasicType argbox = java_lang_boxing_object::basic_type(argument); if (argbox != ptype) { err = check_argument_type_change(T_OBJECT, (argument == NULL - ? SystemDictionary::object_klass() + ? SystemDictionary::Object_klass() : argument->klass()), ptype, ptype_klass(), argnum); assert(err != NULL, "this must be an error"); @@ -1487,8 +1490,9 @@ int target_pushes = decode_MethodHandle_stack_pushes(target()); assert(this_pushes == slots_pushed + target_pushes, "BMH stack motion must be correct"); // do not blow the stack; use a Java-based adapter if this limit is exceeded - if (slots_pushed + target_pushes > MethodHandlePushLimit) - err = "too many bound parameters"; + // FIXME + // if (slots_pushed + target_pushes > MethodHandlePushLimit) + // err = "too many bound parameters"; } } @@ -1518,6 +1522,11 @@ verify_vmslots(mh, CHECK); } + // Get bound type and required slots. + oop ptype_oop = java_dyn_MethodType::ptype(java_dyn_MethodHandle::type(target()), argnum); + BasicType ptype = java_lang_Class::as_BasicType(ptype_oop); + int slots_pushed = type2size[ptype]; + // If (a) the target is a direct non-dispatched method handle, // or (b) the target is a dispatched direct method handle and we // are binding the receiver, cut out the middle-man. @@ -1529,7 +1538,7 @@ int decode_flags = 0; klassOop receiver_limit_oop = NULL; methodHandle m(THREAD, decode_method(target(), receiver_limit_oop, decode_flags)); if (m.is_null()) { THROW_MSG(vmSymbols::java_lang_InternalError(), "DMH failed to decode"); } - DEBUG_ONLY(int m_vmslots = m->size_of_parameters() - 1); // pos. of 1st arg. + DEBUG_ONLY(int m_vmslots = m->size_of_parameters() - slots_pushed); // pos. of 1st arg. assert(sun_dyn_BoundMethodHandle::vmslots(mh()) == m_vmslots, "type w/ m sig"); if (argnum == 0 && (decode_flags & _dmf_has_receiver) != 0) { KlassHandle receiver_limit(THREAD, receiver_limit_oop); @@ -1554,10 +1563,6 @@ } // Next question: Is this a ref, int, or long bound value? - oop ptype_oop = java_dyn_MethodType::ptype(java_dyn_MethodHandle::type(target()), argnum); - BasicType ptype = java_lang_Class::as_BasicType(ptype_oop); - int slots_pushed = type2size[ptype]; - MethodHandleEntry* me = NULL; if (ptype == T_OBJECT) { if (direct_to_method) me = MethodHandles::entry(_bound_ref_direct_mh); @@ -2170,7 +2175,7 @@ symbolOop name = vmSymbols::toString_name(), sig = vmSymbols::void_string_signature(); JavaCallArguments args(Handle(THREAD, JNIHandles::resolve_non_null(erased_jh))); JavaValue result(T_OBJECT); - JavaCalls::call_virtual(&result, SystemDictionary::object_klass(), name, sig, + JavaCalls::call_virtual(&result, SystemDictionary::Object_klass(), name, sig, &args, CHECK); Handle str(THREAD, (oop)result.get_jobject()); java_lang_String::print(str, tty); @@ -2347,9 +2352,9 @@ JVM_ENTRY(void, MH_linkCallSite(JNIEnv *env, jobject igcls, jobject site_jh, jobject target_jh)) { // No special action required, yet. oop site_oop = JNIHandles::resolve(site_jh); - if (site_oop == NULL || site_oop->klass() != SystemDictionary::CallSiteImpl_klass()) + if (site_oop == NULL || site_oop->klass() != SystemDictionary::CallSite_klass()) THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(), "call site"); - sun_dyn_CallSiteImpl::set_target(site_oop, JNIHandles::resolve(target_jh)); + java_dyn_CallSite::set_target(site_oop, JNIHandles::resolve(target_jh)); } JVM_END @@ -2365,6 +2370,7 @@ #define OBJ LANG"Object;" #define CLS LANG"Class;" #define STRG LANG"String;" +#define CST JDYN"CallSite;" #define MT JDYN"MethodType;" #define MH JDYN"MethodHandle;" #define MHI IDYN"MethodHandleImpl;" @@ -2372,7 +2378,6 @@ #define AMH IDYN"AdapterMethodHandle;" #define BMH IDYN"BoundMethodHandle;" #define DMH IDYN"DirectMethodHandle;" -#define CSTI IDYN"CallSiteImpl;" #define CC (char*) /*cast a literal from (const char*)*/ #define FN_PTR(f) CAST_FROM_FN_PTR(void*, &f) @@ -2398,7 +2403,7 @@ // More entry points specifically for EnableInvokeDynamic. static JNINativeMethod methods2[] = { - {CC"linkCallSite", CC"("CSTI MH")V", FN_PTR(MH_linkCallSite)} + {CC"linkCallSite", CC"("CST MH")V", FN_PTR(MH_linkCallSite)} };
--- a/src/share/vm/prims/nativeLookup.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/prims/nativeLookup.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -137,7 +137,7 @@ } // Otherwise call static method findNative in ClassLoader - KlassHandle klass (THREAD, SystemDictionary::classloader_klass()); + KlassHandle klass (THREAD, SystemDictionary::ClassLoader_klass()); Handle name_arg = java_lang_String::create_from_str(jni_name, CHECK_NULL); JavaValue result(T_LONG);
--- a/src/share/vm/runtime/arguments.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/arguments.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -37,7 +37,6 @@ const char* Arguments::_gc_log_filename = NULL; bool Arguments::_has_profile = false; bool Arguments::_has_alloc_profile = false; -uintx Arguments::_initial_heap_size = 0; uintx Arguments::_min_heap_size = 0; Arguments::Mode Arguments::_mode = _mixed; bool Arguments::_java_compiler = false; @@ -182,6 +181,9 @@ { "ProcessingToTenuringRatio", JDK_Version::jdk(5), JDK_Version::jdk(7) }, { "MinTrainLength", JDK_Version::jdk(5), JDK_Version::jdk(7) }, { "AppendRatio", JDK_Version::jdk_update(6,10), JDK_Version::jdk(7) }, + { "DefaultMaxRAM", JDK_Version::jdk_update(6,18), JDK_Version::jdk(7) }, + { "DefaultInitialRAMFraction", + JDK_Version::jdk_update(6,18), JDK_Version::jdk(7) }, { NULL, JDK_Version(0), JDK_Version(0) } }; @@ -555,6 +557,10 @@ if (!is_neg && CommandLineFlags::uintxAtPut(name, &uintx_v, origin)) { return true; } + uint64_t uint64_t_v = (uint64_t) v; + if (!is_neg && CommandLineFlags::uint64_tAtPut(name, &uint64_t_v, origin)) { + return true; + } return false; } @@ -942,12 +948,13 @@ } } +#ifndef KERNEL // If the user has chosen ParallelGCThreads > 0, we set UseParNewGC // if it's not explictly set or unset. If the user has chosen // UseParNewGC and not explicitly set ParallelGCThreads we // set it, unless this is a single cpu machine. void Arguments::set_parnew_gc_flags() { - assert(!UseSerialGC && !UseParallelGC && !UseG1GC, + assert(!UseSerialGC && !UseParallelOldGC && !UseParallelGC && !UseG1GC, "control point invariant"); assert(UseParNewGC, "Error"); @@ -960,13 +967,13 @@ if (ParallelGCThreads == 0) { FLAG_SET_DEFAULT(ParallelGCThreads, Abstract_VM_Version::parallel_worker_threads()); - if (FLAG_IS_DEFAULT(ParallelGCThreads) && ParallelGCThreads == 1) { + if (ParallelGCThreads == 1) { FLAG_SET_DEFAULT(UseParNewGC, false); + FLAG_SET_DEFAULT(ParallelGCThreads, 0); } } - if (!UseParNewGC) { - FLAG_SET_DEFAULT(ParallelGCThreads, 0); - } else { + if (UseParNewGC) { + // CDS doesn't work with ParNew yet no_shared_spaces(); // By default YoungPLABSize and OldPLABSize are set to 4096 and 1024 respectively, @@ -980,7 +987,7 @@ FLAG_SET_DEFAULT(OldPLABSize, (intx)1024); } - // AlwaysTenure flag should make ParNew to promote all at first collection. + // AlwaysTenure flag should make ParNew promote all at first collection. // See CR 6362902. if (AlwaysTenure) { FLAG_SET_CMDLINE(intx, MaxTenuringThreshold, 0); @@ -1003,7 +1010,7 @@ // further optimization and tuning efforts, and would almost // certainly gain from analysis of platform and environment. void Arguments::set_cms_and_parnew_gc_flags() { - assert(!UseSerialGC && !UseParallelGC, "Error"); + assert(!UseSerialGC && !UseParallelOldGC && !UseParallelGC, "Error"); assert(UseConcMarkSweepGC, "CMS is expected to be on here"); // If we are using CMS, we prefer to UseParNewGC, @@ -1068,7 +1075,7 @@ } else { FLAG_SET_ERGO(uintx, MaxNewSize, preferred_max_new_size); } - if(PrintGCDetails && Verbose) { + if (PrintGCDetails && Verbose) { // Too early to use gclog_or_tty tty->print_cr("Ergo set MaxNewSize: " SIZE_FORMAT, MaxNewSize); } @@ -1097,15 +1104,15 @@ } else { min_new = NewSize; } - size_t prev_initial_size = initial_heap_size(); - if (prev_initial_size != 0 && prev_initial_size < min_new+OldSize) { - set_initial_heap_size(min_new+OldSize); + size_t prev_initial_size = InitialHeapSize; + if (prev_initial_size != 0 && prev_initial_size < min_new + OldSize) { + FLAG_SET_ERGO(uintx, InitialHeapSize, min_new + OldSize); // Currently minimum size and the initial heap sizes are the same. - set_min_heap_size(initial_heap_size()); + set_min_heap_size(InitialHeapSize); if (PrintGCDetails && Verbose) { warning("Initial heap size increased to " SIZE_FORMAT " M from " SIZE_FORMAT " M; use -XX:NewSize=... for finer control.", - initial_heap_size()/M, prev_initial_size/M); + InitialHeapSize/M, prev_initial_size/M); } } @@ -1114,12 +1121,12 @@ align_size_down(MaxHeapSize, CardTableRS::ct_max_alignment_constraint()); - if(PrintGCDetails && Verbose) { + if (PrintGCDetails && Verbose) { // Too early to use gclog_or_tty tty->print_cr("CMS set min_heap_size: " SIZE_FORMAT " initial_heap_size: " SIZE_FORMAT " max_heap: " SIZE_FORMAT, - min_heap_size(), initial_heap_size(), max_heap); + min_heap_size(), InitialHeapSize, max_heap); } if (max_heap > min_new) { // Unless explicitly requested otherwise, make young gen @@ -1127,7 +1134,7 @@ if (FLAG_IS_DEFAULT(NewSize)) { FLAG_SET_ERGO(uintx, NewSize, MAX2(NewSize, min_new)); FLAG_SET_ERGO(uintx, NewSize, MIN2(preferred_max_new_size, NewSize)); - if(PrintGCDetails && Verbose) { + if (PrintGCDetails && Verbose) { // Too early to use gclog_or_tty tty->print_cr("Ergo set NewSize: " SIZE_FORMAT, NewSize); } @@ -1137,8 +1144,8 @@ // later NewRatio will decide how it grows; see above. if (FLAG_IS_DEFAULT(OldSize)) { if (max_heap > NewSize) { - FLAG_SET_ERGO(uintx, OldSize, MIN2(3*NewSize, max_heap - NewSize)); - if(PrintGCDetails && Verbose) { + FLAG_SET_ERGO(uintx, OldSize, MIN2(3*NewSize, max_heap - NewSize)); + if (PrintGCDetails && Verbose) { // Too early to use gclog_or_tty tty->print_cr("Ergo set OldSize: " SIZE_FORMAT, OldSize); } @@ -1171,8 +1178,7 @@ // the value (either from the command line or ergonomics) of // OldPLABSize. Following OldPLABSize is an ergonomics decision. FLAG_SET_ERGO(uintx, CMSParPromoteBlocksToClaim, OldPLABSize); - } - else { + } else { // OldPLABSize and CMSParPromoteBlocksToClaim are both set. // CMSParPromoteBlocksToClaim is a collector-specific flag, so // we'll let it to take precedence. @@ -1182,11 +1188,27 @@ " CMSParPromoteBlocksToClaim will take precedence.\n"); } } + if (!FLAG_IS_DEFAULT(ResizeOldPLAB) && !ResizeOldPLAB) { + // OldPLAB sizing manually turned off: Use a larger default setting, + // unless it was manually specified. This is because a too-low value + // will slow down scavenges. + if (FLAG_IS_DEFAULT(CMSParPromoteBlocksToClaim)) { + FLAG_SET_ERGO(uintx, CMSParPromoteBlocksToClaim, 50); // default value before 6631166 + } + } + // Overwrite OldPLABSize which is the variable we will internally use everywhere. + FLAG_SET_ERGO(uintx, OldPLABSize, CMSParPromoteBlocksToClaim); + // If either of the static initialization defaults have changed, note this + // modification. + if (!FLAG_IS_DEFAULT(CMSParPromoteBlocksToClaim) || !FLAG_IS_DEFAULT(OldPLABWeight)) { + CFLS_LAB::modify_initialization(OldPLABSize, OldPLABWeight); + } } +#endif // KERNEL inline uintx max_heap_for_compressed_oops() { LP64_ONLY(return oopDesc::OopEncodingHeapMax - MaxPermSize - os::vm_page_size()); - NOT_LP64(return DefaultMaxRAM); + NOT_LP64(ShouldNotReachHere(); return 0); } bool Arguments::should_auto_select_low_pause_collector() { @@ -1205,7 +1227,7 @@ void Arguments::set_ergonomics_flags() { // Parallel GC is not compatible with sharing. If one specifies - // that they want sharing explicitly, do not set ergonmics flags. + // that they want sharing explicitly, do not set ergonomics flags. if (DumpSharedSpaces || ForceSharedSpaces) { return; } @@ -1234,9 +1256,11 @@ // Check that UseCompressedOops can be set with the max heap size allocated // by ergonomics. if (MaxHeapSize <= max_heap_for_compressed_oops()) { +#ifndef COMPILER1 if (FLAG_IS_DEFAULT(UseCompressedOops) && !UseG1GC) { FLAG_SET_ERGO(bool, UseCompressedOops, true); } +#endif #ifdef _WIN64 if (UseLargePages && UseCompressedOops) { // Cannot allocate guard pages for implicit checks in indexed addressing @@ -1271,8 +1295,6 @@ FLAG_SET_ERGO(uintx, ParallelGCThreads, Abstract_VM_Version::parallel_worker_threads()); - // PS is a server collector, setup the heap sizes accordingly. - set_server_heap_size(); // If InitialSurvivorRatio or MinSurvivorRatio were not specified, but the // SurvivorRatio has been set, reset their default values to SurvivorRatio + // 2. By doing this we make SurvivorRatio also work for Parallel Scavenger. @@ -1302,8 +1324,6 @@ void Arguments::set_g1_gc_flags() { assert(UseG1GC, "Error"); - // G1 is a server collector, setup the heap sizes accordingly. - set_server_heap_size(); #ifdef COMPILER1 FastTLABRefill = false; #endif @@ -1321,50 +1341,79 @@ } } -void Arguments::set_server_heap_size() { +void Arguments::set_heap_size() { + if (!FLAG_IS_DEFAULT(DefaultMaxRAMFraction)) { + // Deprecated flag + FLAG_SET_CMDLINE(uintx, MaxRAMFraction, DefaultMaxRAMFraction); + } + + const julong phys_mem = + FLAG_IS_DEFAULT(MaxRAM) ? MIN2(os::physical_memory(), (julong)MaxRAM) + : (julong)MaxRAM; + + // If the maximum heap size has not been set with -Xmx, + // then set it as fraction of the size of physical memory, + // respecting the maximum and minimum sizes of the heap. if (FLAG_IS_DEFAULT(MaxHeapSize)) { - const uint64_t reasonable_fraction = - os::physical_memory() / DefaultMaxRAMFraction; - const uint64_t maximum_size = (uint64_t) - (FLAG_IS_DEFAULT(DefaultMaxRAM) && UseCompressedOops ? - MIN2(max_heap_for_compressed_oops(), DefaultMaxRAM) : - DefaultMaxRAM); - size_t reasonable_max = - (size_t) os::allocatable_physical_memory(reasonable_fraction); - if (reasonable_max > maximum_size) { - reasonable_max = maximum_size; + julong reasonable_max = phys_mem / MaxRAMFraction; + + if (phys_mem <= MaxHeapSize * MinRAMFraction) { + // Small physical memory, so use a minimum fraction of it for the heap + reasonable_max = phys_mem / MinRAMFraction; + } else { + // Not-small physical memory, so require a heap at least + // as large as MaxHeapSize + reasonable_max = MAX2(reasonable_max, (julong)MaxHeapSize); + } + if (!FLAG_IS_DEFAULT(ErgoHeapSizeLimit) && ErgoHeapSizeLimit != 0) { + // Limit the heap size to ErgoHeapSizeLimit + reasonable_max = MIN2(reasonable_max, (julong)ErgoHeapSizeLimit); } + if (UseCompressedOops) { + // Limit the heap size to the maximum possible when using compressed oops + reasonable_max = MIN2(reasonable_max, (julong)max_heap_for_compressed_oops()); + } + reasonable_max = os::allocatable_physical_memory(reasonable_max); + + if (!FLAG_IS_DEFAULT(InitialHeapSize)) { + // An initial heap size was specified on the command line, + // so be sure that the maximum size is consistent. Done + // after call to allocatable_physical_memory because that + // method might reduce the allocation size. + reasonable_max = MAX2(reasonable_max, (julong)InitialHeapSize); + } + if (PrintGCDetails && Verbose) { // Cannot use gclog_or_tty yet. - tty->print_cr(" Max heap size for server class platform " - SIZE_FORMAT, reasonable_max); + tty->print_cr(" Maximum heap size " SIZE_FORMAT, reasonable_max); } - // If the initial_heap_size has not been set with -Xms, - // then set it as fraction of size of physical memory - // respecting the maximum and minimum sizes of the heap. - if (initial_heap_size() == 0) { - const uint64_t reasonable_initial_fraction = - os::physical_memory() / DefaultInitialRAMFraction; - const size_t reasonable_initial = - (size_t) os::allocatable_physical_memory(reasonable_initial_fraction); - const size_t minimum_size = NewSize + OldSize; - set_initial_heap_size(MAX2(MIN2(reasonable_initial, reasonable_max), - minimum_size)); - // Currently the minimum size and the initial heap sizes are the same. - set_min_heap_size(initial_heap_size()); - if (PrintGCDetails && Verbose) { - // Cannot use gclog_or_tty yet. - tty->print_cr(" Initial heap size for server class platform " - SIZE_FORMAT, initial_heap_size()); - } - } else { - // A minimum size was specified on the command line. Be sure - // that the maximum size is consistent. - if (initial_heap_size() > reasonable_max) { - reasonable_max = initial_heap_size(); - } + FLAG_SET_ERGO(uintx, MaxHeapSize, (uintx)reasonable_max); + } + + // If the initial_heap_size has not been set with InitialHeapSize + // or -Xms, then set it as fraction of the size of physical memory, + // respecting the maximum and minimum sizes of the heap. + if (FLAG_IS_DEFAULT(InitialHeapSize)) { + julong reasonable_minimum = (julong)(OldSize + NewSize); + + reasonable_minimum = MIN2(reasonable_minimum, (julong)MaxHeapSize); + + reasonable_minimum = os::allocatable_physical_memory(reasonable_minimum); + + julong reasonable_initial = phys_mem / InitialRAMFraction; + + reasonable_initial = MAX2(reasonable_initial, reasonable_minimum); + reasonable_initial = MIN2(reasonable_initial, (julong)MaxHeapSize); + + reasonable_initial = os::allocatable_physical_memory(reasonable_initial); + + if (PrintGCDetails && Verbose) { + // Cannot use gclog_or_tty yet. + tty->print_cr(" Initial heap size " SIZE_FORMAT, (uintx)reasonable_initial); + tty->print_cr(" Minimum heap size " SIZE_FORMAT, (uintx)reasonable_minimum); } - FLAG_SET_ERGO(uintx, MaxHeapSize, (uintx) reasonable_max); + FLAG_SET_ERGO(uintx, InitialHeapSize, (uintx)reasonable_initial); + set_min_heap_size((uintx)reasonable_minimum); } } @@ -1438,6 +1487,20 @@ //=========================================================================================================== // Parsing of main arguments +bool Arguments::verify_interval(uintx val, uintx min, + uintx max, const char* name) { + // Returns true iff value is in the inclusive interval [min..max] + // false, otherwise. + if (val >= min && val <= max) { + return true; + } + jio_fprintf(defaultStream::error_stream(), + "%s of " UINTX_FORMAT " is invalid; must be between " UINTX_FORMAT + " and " UINTX_FORMAT "\n", + name, val, min, max); + return false; +} + bool Arguments::verify_percentage(uintx value, const char* name) { if (value <= 100) { return true; @@ -1448,7 +1511,7 @@ return false; } -static void set_serial_gc_flags() { +static void force_serial_gc() { FLAG_SET_DEFAULT(UseSerialGC, true); FLAG_SET_DEFAULT(UseParNewGC, false); FLAG_SET_DEFAULT(UseConcMarkSweepGC, false); @@ -1584,15 +1647,15 @@ // force sharing off. if (DumpSharedSpaces || ForceSharedSpaces) { jio_fprintf(defaultStream::error_stream(), - "Reverting to Serial GC because of %s \n", + "Reverting to Serial GC because of %s\n", ForceSharedSpaces ? " -Xshare:on" : "-Xshare:dump"); - set_serial_gc_flags(); + force_serial_gc(); FLAG_SET_DEFAULT(SOLARIS_ONLY(UseISM) NOT_SOLARIS(UseLargePages), false); } else { - if (UseSharedSpaces) { + if (UseSharedSpaces && Verbose) { jio_fprintf(defaultStream::error_stream(), "Turning off use of shared archive because of " - "choice of garbage collector or large pages \n"); + "choice of garbage collector or large pages\n"); } no_shared_spaces(); } @@ -1674,6 +1737,16 @@ status = false; } + status = status && verify_interval(RefDiscoveryPolicy, + ReferenceProcessor::DiscoveryPolicyMin, + ReferenceProcessor::DiscoveryPolicyMax, + "RefDiscoveryPolicy"); + + // Limit the lower bound of this flag to 1 as it is used in a division + // expression. + status = status && verify_interval(TLABWasteTargetPercent, + 1, 100, "TLABWasteTargetPercent"); + return status; } @@ -1817,7 +1890,6 @@ FLAG_SET_CMDLINE(bool, TraceClassUnloading, true); } else if (!strcmp(tail, ":gc")) { FLAG_SET_CMDLINE(bool, PrintGC, true); - FLAG_SET_CMDLINE(bool, TraceClassUnloading, true); } else if (!strcmp(tail, ":jni")) { FLAG_SET_CMDLINE(bool, PrintJNIResolving, true); } @@ -1925,8 +1997,8 @@ describe_range_error(errcode); return JNI_EINVAL; } - FLAG_SET_CMDLINE(uintx, MaxNewSize, (size_t) long_initial_eden_size); - FLAG_SET_CMDLINE(uintx, NewSize, (size_t) long_initial_eden_size); + FLAG_SET_CMDLINE(uintx, MaxNewSize, (uintx)long_initial_eden_size); + FLAG_SET_CMDLINE(uintx, NewSize, (uintx)long_initial_eden_size); // -Xms } else if (match_option(option, "-Xms", &tail)) { julong long_initial_heap_size = 0; @@ -1937,9 +2009,9 @@ describe_range_error(errcode); return JNI_EINVAL; } - set_initial_heap_size((size_t) long_initial_heap_size); + FLAG_SET_CMDLINE(uintx, InitialHeapSize, (uintx)long_initial_heap_size); // Currently the minimum size and the initial heap sizes are the same. - set_min_heap_size(initial_heap_size()); + set_min_heap_size(InitialHeapSize); // -Xmx } else if (match_option(option, "-Xmx", &tail)) { julong long_max_heap_size = 0; @@ -1950,7 +2022,7 @@ describe_range_error(errcode); return JNI_EINVAL; } - FLAG_SET_CMDLINE(uintx, MaxHeapSize, (size_t) long_max_heap_size); + FLAG_SET_CMDLINE(uintx, MaxHeapSize, (uintx)long_max_heap_size); // Xmaxf } else if (match_option(option, "-Xmaxf", &tail)) { int maxf = (int)(atof(tail) * 100); @@ -2196,9 +2268,9 @@ if (FLAG_IS_DEFAULT(MaxHeapSize)) { FLAG_SET_CMDLINE(uintx, MaxHeapSize, initHeapSize); - set_initial_heap_size(MaxHeapSize); + FLAG_SET_CMDLINE(uintx, InitialHeapSize, initHeapSize); // Currently the minimum size and the initial heap sizes are the same. - set_min_heap_size(initial_heap_size()); + set_min_heap_size(initHeapSize); } if (FLAG_IS_DEFAULT(NewSize)) { // Make the young generation 3/8ths of the total heap. @@ -2337,22 +2409,25 @@ "ExtendedDTraceProbes flag is only applicable on Solaris\n"); return JNI_EINVAL; #endif // ndef SOLARIS - } else #ifdef ASSERT - if (match_option(option, "-XX:+FullGCALot", &tail)) { + } else if (match_option(option, "-XX:+FullGCALot", &tail)) { FLAG_SET_CMDLINE(bool, FullGCALot, true); // disable scavenge before parallel mark-compact FLAG_SET_CMDLINE(bool, ScavengeBeforeFullGC, false); - } else #endif - if (match_option(option, "-XX:ParCMSPromoteBlocksToClaim=", &tail)) { + } else if (match_option(option, "-XX:CMSParPromoteBlocksToClaim=", &tail)) { julong cms_blocks_to_claim = (julong)atol(tail); FLAG_SET_CMDLINE(uintx, CMSParPromoteBlocksToClaim, cms_blocks_to_claim); jio_fprintf(defaultStream::error_stream(), - "Please use -XX:CMSParPromoteBlocksToClaim in place of " + "Please use -XX:OldPLABSize in place of " + "-XX:CMSParPromoteBlocksToClaim in the future\n"); + } else if (match_option(option, "-XX:ParCMSPromoteBlocksToClaim=", &tail)) { + julong cms_blocks_to_claim = (julong)atol(tail); + FLAG_SET_CMDLINE(uintx, CMSParPromoteBlocksToClaim, cms_blocks_to_claim); + jio_fprintf(defaultStream::error_stream(), + "Please use -XX:OldPLABSize in place of " "-XX:ParCMSPromoteBlocksToClaim in the future\n"); - } else - if (match_option(option, "-XX:ParallelGCOldGenAllocBufferSize=", &tail)) { + } else if (match_option(option, "-XX:ParallelGCOldGenAllocBufferSize=", &tail)) { julong old_plab_size = 0; ArgsRange errcode = parse_memory_size(tail, &old_plab_size, 1); if (errcode != arg_in_range) { @@ -2365,8 +2440,7 @@ jio_fprintf(defaultStream::error_stream(), "Please use -XX:OldPLABSize in place of " "-XX:ParallelGCOldGenAllocBufferSize in the future\n"); - } else - if (match_option(option, "-XX:ParallelGCToSpaceAllocBufferSize=", &tail)) { + } else if (match_option(option, "-XX:ParallelGCToSpaceAllocBufferSize=", &tail)) { julong young_plab_size = 0; ArgsRange errcode = parse_memory_size(tail, &young_plab_size, 1); if (errcode != arg_in_range) { @@ -2379,8 +2453,7 @@ jio_fprintf(defaultStream::error_stream(), "Please use -XX:YoungPLABSize in place of " "-XX:ParallelGCToSpaceAllocBufferSize in the future\n"); - } else - if (match_option(option, "-XX:", &tail)) { // -XX:xxxx + } else if (match_option(option, "-XX:", &tail)) { // -XX:xxxx // Skip -XX:Flags= since that case has already been handled if (strncmp(tail, "Flags=", strlen("Flags=")) != 0) { if (!process_argument(tail, args->ignoreUnrecognized, origin)) { @@ -2451,6 +2524,9 @@ SOLARIS_ONLY(FLAG_SET_DEFAULT(UseISM, false)); } + // Tiered compilation is undefined with C1. + TieredCompilation = false; + #else if (!FLAG_IS_DEFAULT(OptoLoopAlignment) && FLAG_IS_DEFAULT(MaxLoopPad)) { FLAG_SET_DEFAULT(MaxLoopPad, OptoLoopAlignment-1); @@ -2600,6 +2676,10 @@ if (match_option(option, "-XX:-IgnoreUnrecognizedVMOptions", &tail)) { IgnoreUnrecognizedVMOptions = false; } + if (match_option(option, "-XX:+PrintFlagsInitial", &tail)) { + CommandLineFlags::printFlags(); + vm_exit(0); + } } if (IgnoreUnrecognizedVMOptions) { @@ -2666,17 +2746,27 @@ } ScavengeRootsInCode = 1; } +#ifdef COMPILER2 + if (EnableInvokeDynamic && DoEscapeAnalysis) { + // TODO: We need to find rules for invokedynamic and EA. For now, + // simply disable EA by default. + if (FLAG_IS_DEFAULT(DoEscapeAnalysis)) { + DoEscapeAnalysis = false; + } + } +#endif if (PrintGCDetails) { // Turn on -verbose:gc options as well PrintGC = true; - if (FLAG_IS_DEFAULT(TraceClassUnloading)) { - TraceClassUnloading = true; - } } +#if defined(_LP64) && defined(COMPILER1) + UseCompressedOops = false; +#endif + #ifdef SERIALGC - set_serial_gc_flags(); + force_serial_gc(); #endif // SERIALGC #ifdef KERNEL no_shared_spaces(); @@ -2685,24 +2775,39 @@ // Set flags based on ergonomics. set_ergonomics_flags(); +#ifdef _LP64 + // XXX JSR 292 currently does not support compressed oops. + if (EnableMethodHandles && UseCompressedOops) { + if (FLAG_IS_DEFAULT(UseCompressedOops) || FLAG_IS_ERGO(UseCompressedOops)) { + UseCompressedOops = false; + } + } +#endif // _LP64 + // Check the GC selections again. if (!check_gc_consistency()) { return JNI_EINVAL; } - if (UseParallelGC || UseParallelOldGC) { - // Set some flags for ParallelGC if needed. - set_parallel_gc_flags(); - } else if (UseConcMarkSweepGC) { - // Set some flags for CMS +#ifndef KERNEL + if (UseConcMarkSweepGC) { + // Set flags for CMS and ParNew. Check UseConcMarkSweep first + // to ensure that when both UseConcMarkSweepGC and UseParNewGC + // are true, we don't call set_parnew_gc_flags() as well. set_cms_and_parnew_gc_flags(); - } else if (UseParNewGC) { - // Set some flags for ParNew - set_parnew_gc_flags(); - } else if (UseG1GC) { - // Set some flags for garbage-first, if needed. - set_g1_gc_flags(); + } else { + // Set heap size based on available physical memory + set_heap_size(); + // Set per-collector flags + if (UseParallelGC || UseParallelOldGC) { + set_parallel_gc_flags(); + } else if (UseParNewGC) { + set_parnew_gc_flags(); + } else if (UseG1GC) { + set_g1_gc_flags(); + } } +#endif // KERNEL #ifdef SERIALGC assert(verify_serial_gc_flags(), "SerialGC unset"); @@ -2715,9 +2820,16 @@ set_aggressive_opts_flags(); #ifdef CC_INTERP - // Biased locking is not implemented with c++ interpreter + // Clear flags not supported by the C++ interpreter + FLAG_SET_DEFAULT(ProfileInterpreter, false); FLAG_SET_DEFAULT(UseBiasedLocking, false); -#endif /* CC_INTERP */ + LP64_ONLY(FLAG_SET_DEFAULT(UseCompressedOops, false)); +#endif // CC_INTERP + +#ifdef ZERO + // Clear flags not supported by Zero + FLAG_SET_DEFAULT(TaggedStackInterpreter, false); +#endif // ZERO #ifdef COMPILER2 if (!UseBiasedLocking || EmitSync != 0) { @@ -2725,15 +2837,27 @@ } #endif + if (PrintAssembly && FLAG_IS_DEFAULT(DebugNonSafepoints)) { + warning("PrintAssembly is enabled; turning on DebugNonSafepoints to gain additional output"); + DebugNonSafepoints = true; + } + +#ifndef PRODUCT + if (CompileTheWorld) { + // Force NmethodSweeper to sweep whole CodeCache each time. + if (FLAG_IS_DEFAULT(NmethodSweepFraction)) { + NmethodSweepFraction = 1; + } + } +#endif + if (PrintCommandLineFlags) { CommandLineFlags::printSetFlags(); } -#ifdef ASSERT if (PrintFlagsFinal) { CommandLineFlags::printFlags(); } -#endif return JNI_OK; }
--- a/src/share/vm/runtime/arguments.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/arguments.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -254,7 +254,6 @@ static bool _has_profile; static bool _has_alloc_profile; static const char* _gc_log_filename; - static uintx _initial_heap_size; static uintx _min_heap_size; // -Xrun arguments @@ -300,8 +299,8 @@ static void set_g1_gc_flags(); // GC ergonomics static void set_ergonomics_flags(); - // Setup heap size for a server platform - static void set_server_heap_size(); + // Setup heap size + static void set_heap_size(); // Based on automatic selection criteria, should the // low pause collector be used. static bool should_auto_select_low_pause_collector(); @@ -337,6 +336,8 @@ static bool is_bad_option(const JavaVMOption* option, jboolean ignore) { return is_bad_option(option, ignore, NULL); } + static bool verify_interval(uintx val, uintx min, + uintx max, const char* name); static bool verify_percentage(uintx value, const char* name); static void describe_range_error(ArgsRange errcode); static ArgsRange check_memory_size(julong size, julong min_size); @@ -434,9 +435,7 @@ static bool has_profile() { return _has_profile; } static bool has_alloc_profile() { return _has_alloc_profile; } - // -Xms , -Xmx - static uintx initial_heap_size() { return _initial_heap_size; } - static void set_initial_heap_size(uintx v) { _initial_heap_size = v; } + // -Xms, -Xmx static uintx min_heap_size() { return _min_heap_size; } static void set_min_heap_size(uintx v) { _min_heap_size = v; }
--- a/src/share/vm/runtime/compilationPolicy.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/compilationPolicy.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2000-2007 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2000-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -66,7 +66,7 @@ if (!canBeCompiled(m)) return false; return !UseInterpreter || // must compile all methods - (UseCompiler && AlwaysCompileLoopMethods && m->has_loops()); // eagerly compile loop methods + (UseCompiler && AlwaysCompileLoopMethods && m->has_loops() && CompileBroker::should_compile_new_jobs()); // eagerly compile loop methods } // Returns true if m is allowed to be compiled @@ -74,6 +74,16 @@ if (m->is_abstract()) return false; if (DontCompileHugeMethods && m->code_size() > HugeMethodLimit) return false; + // Math intrinsics should never be compiled as this can lead to + // monotonicity problems because the interpreter will prefer the + // compiled code to the intrinsic version. This can't happen in + // production because the invocation counter can't be incremented + // but we shouldn't expose the system to this problem in testing + // modes. + if (!AbstractInterpreter::can_be_compiled(m)) { + return false; + } + return !m->is_not_compilable(); } @@ -127,7 +137,7 @@ reset_counter_for_invocation_event(m); const char* comment = "count"; - if (!delayCompilationDuringStartup() && canBeCompiled(m) && UseCompiler) { + if (!delayCompilationDuringStartup() && canBeCompiled(m) && UseCompiler && CompileBroker::should_compile_new_jobs()) { nmethod* nm = m->code(); if (nm == NULL ) { const char* comment = "count"; @@ -152,7 +162,7 @@ int hot_count = m->backedge_count(); const char* comment = "backedge_count"; - if (!m->is_not_osr_compilable() && !delayCompilationDuringStartup() && canBeCompiled(m)) { + if (!m->is_not_osr_compilable() && !delayCompilationDuringStartup() && canBeCompiled(m) && CompileBroker::should_compile_new_jobs()) { CompileBroker::compile_method(m, loop_top_bci, m, hot_count, comment, CHECK); NOT_PRODUCT(trace_osr_completion(m->lookup_osr_nmethod_for(loop_top_bci));) @@ -194,7 +204,7 @@ reset_counter_for_invocation_event(m); const char* comment = "count"; - if (m->code() == NULL && !delayCompilationDuringStartup() && canBeCompiled(m) && UseCompiler) { + if (m->code() == NULL && !delayCompilationDuringStartup() && canBeCompiled(m) && UseCompiler && CompileBroker::should_compile_new_jobs()) { ResourceMark rm(THREAD); JavaThread *thread = (JavaThread*)THREAD; frame fr = thread->last_frame(); @@ -238,7 +248,7 @@ int hot_count = m->backedge_count(); const char* comment = "backedge_count"; - if (!m->is_not_osr_compilable() && !delayCompilationDuringStartup() && canBeCompiled(m)) { + if (!m->is_not_osr_compilable() && !delayCompilationDuringStartup() && canBeCompiled(m) && CompileBroker::should_compile_new_jobs()) { CompileBroker::compile_method(m, loop_top_bci, m, hot_count, comment, CHECK); NOT_PRODUCT(trace_osr_completion(m->lookup_osr_nmethod_for(loop_top_bci));)
--- a/src/share/vm/runtime/deoptimization.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/deoptimization.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -145,6 +145,27 @@ if (EliminateAllocations) { assert (chunk->at(0)->scope() != NULL,"expect only compiled java frames"); GrowableArray<ScopeValue*>* objects = chunk->at(0)->scope()->objects(); + + // The flag return_oop() indicates call sites which return oop + // in compiled code. Such sites include java method calls, + // runtime calls (for example, used to allocate new objects/arrays + // on slow code path) and any other calls generated in compiled code. + // It is not guaranteed that we can get such information here only + // by analyzing bytecode in deoptimized frames. This is why this flag + // is set during method compilation (see Compile::Process_OopMap_Node()). + bool save_oop_result = chunk->at(0)->scope()->return_oop(); + Handle return_value; + if (save_oop_result) { + // Reallocation may trigger GC. If deoptimization happened on return from + // call which returns oop we need to save it since it is not in oopmap. + oop result = deoptee.saved_oop_result(&map); + assert(result == NULL || result->is_oop(), "must be oop"); + return_value = Handle(thread, result); + assert(Universe::heap()->is_in_or_null(result), "must be heap pointer"); + if (TraceDeoptimization) { + tty->print_cr("SAVED OOP RESULT " INTPTR_FORMAT " in thread " INTPTR_FORMAT, result, thread); + } + } bool reallocated = false; if (objects != NULL) { JRT_BLOCK @@ -158,8 +179,12 @@ ttyLocker ttyl; tty->print_cr("REALLOC OBJECTS in thread " INTPTR_FORMAT, thread); print_objects(objects); + } +#endif } -#endif + if (save_oop_result) { + // Restore result. + deoptee.set_saved_oop_result(&map, return_value()); } } if (EliminateLocks) { @@ -235,6 +260,12 @@ assert(cb->frame_size() >= 0, "Unexpected frame size"); intptr_t* unpack_sp = stub_frame.sp() + cb->frame_size(); + // If the deopt call site is a MethodHandle invoke call site we have + // to adjust the unpack_sp. + nmethod* deoptee_nm = deoptee.cb()->as_nmethod_or_null(); + if (deoptee_nm != NULL && deoptee_nm->is_method_handle_return(deoptee.pc())) + unpack_sp = deoptee.unextended_sp(); + #ifdef ASSERT assert(cb->is_deoptimization_stub() || cb->is_uncommon_trap_stub(), "just checking"); Events::log("fetch unroll sp " INTPTR_FORMAT, unpack_sp); @@ -907,21 +938,6 @@ if (TraceDeoptimization) { ttyLocker ttyl; tty->print_cr(" Created vframeArray " INTPTR_FORMAT, array); - if (Verbose) { - int count = 0; - // this used to leak deoptimizedVFrame like it was going out of style!!! - for (int index = 0; index < array->frames(); index++ ) { - vframeArrayElement* e = array->element(index); - e->print(tty); - - /* - No printing yet. - array->vframe_at(index)->print_activation(count++); - // better as... - array->print_activation_for(index, count++); - */ - } - } } #endif // PRODUCT @@ -1332,13 +1348,14 @@ // Whether the interpreter is producing MDO data or not, we also need // to use the MDO to detect hot deoptimization points and control // aggressive optimization. + bool inc_recompile_count = false; + ProfileData* pdata = NULL; if (ProfileTraps && update_trap_state && trap_mdo.not_null()) { assert(trap_mdo() == get_method_data(thread, trap_method, false), "sanity"); uint this_trap_count = 0; bool maybe_prior_trap = false; bool maybe_prior_recompile = false; - ProfileData* pdata - = query_update_method_data(trap_mdo, trap_bci, reason, + pdata = query_update_method_data(trap_mdo, trap_bci, reason, //outputs: this_trap_count, maybe_prior_trap, @@ -1374,18 +1391,7 @@ // Detect repeated recompilation at the same BCI, and enforce a limit. if (make_not_entrant && maybe_prior_recompile) { // More than one recompile at this point. - trap_mdo->inc_overflow_recompile_count(); - if (maybe_prior_trap - && ((uint)trap_mdo->overflow_recompile_count() - > (uint)PerBytecodeRecompilationCutoff)) { - // Give up on the method containing the bad BCI. - if (trap_method() == nm->method()) { - make_not_compilable = true; - } else { - trap_method->set_not_compilable(); - // But give grace to the enclosing nm->method(). - } - } + inc_recompile_count = maybe_prior_trap; } } else { // For reasons which are not recorded per-bytecode, we simply @@ -1412,7 +1418,17 @@ reset_counters = true; } - if (make_not_entrant && pdata != NULL) { + } + + // Take requested actions on the method: + + // Recompile + if (make_not_entrant) { + if (!nm->make_not_entrant()) { + return; // the call did not change nmethod's state + } + + if (pdata != NULL) { // Record the recompilation event, if any. int tstate0 = pdata->trap_state(); int tstate1 = trap_state_set_recompiled(tstate0, true); @@ -1421,7 +1437,19 @@ } } - // Take requested actions on the method: + if (inc_recompile_count) { + trap_mdo->inc_overflow_recompile_count(); + if ((uint)trap_mdo->overflow_recompile_count() > + (uint)PerBytecodeRecompilationCutoff) { + // Give up on the method containing the bad BCI. + if (trap_method() == nm->method()) { + make_not_compilable = true; + } else { + trap_method->set_not_compilable(); + // But give grace to the enclosing nm->method(). + } + } + } // Reset invocation counters if (reset_counters) { @@ -1431,13 +1459,8 @@ reset_invocation_counter(trap_scope); } - // Recompile - if (make_not_entrant) { - nm->make_not_entrant(); - } - // Give up compiling - if (make_not_compilable) { + if (make_not_compilable && !nm->method()->is_not_compilable()) { assert(make_not_entrant, "consistent"); nm->method()->set_not_compilable(); } @@ -1510,9 +1533,11 @@ if (tstate1 != tstate0) pdata->set_trap_state(tstate1); } else { - if (LogCompilation && xtty != NULL) + if (LogCompilation && xtty != NULL) { + ttyLocker ttyl; // Missing MDP? Leave a small complaint in the log. xtty->elem("missing_mdp bci='%d'", trap_bci); + } } } @@ -1666,13 +1691,15 @@ "class_check", "array_check", "intrinsic", + "bimorphic", "unloaded", "uninitialized", "unreached", "unhandled", "constraint", "div0_check", - "age" + "age", + "predicate" }; const char* Deoptimization::_trap_action_name[Action_LIMIT] = { // Note: Keep this in sync. with enum DeoptAction.
--- a/src/share/vm/runtime/deoptimization.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/deoptimization.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -33,12 +33,15 @@ enum DeoptReason { Reason_many = -1, // indicates presence of several reasons Reason_none = 0, // indicates absence of a relevant deopt. + // Next 7 reasons are recorded per bytecode in DataLayout::trap_bits Reason_null_check, // saw unexpected null or zero divisor (@bci) Reason_null_assert, // saw unexpected non-null or non-zero (@bci) Reason_range_check, // saw unexpected array index (@bci) Reason_class_check, // saw unexpected object class (@bci) Reason_array_check, // saw unexpected array class (aastore @bci) Reason_intrinsic, // saw unexpected operand to intrinsic (@bci) + Reason_bimorphic, // saw unexpected object class in bimorphic inlining (@bci) + Reason_unloaded, // unloaded class or constant pool entry Reason_uninitialized, // bad class state (uninitialized) Reason_unreached, // code is not reached, compiler @@ -46,9 +49,10 @@ Reason_constraint, // arbitrary runtime constraint violated Reason_div0_check, // a null_check due to division by zero Reason_age, // nmethod too old; tier threshold reached + Reason_predicate, // compiler generated predicate failed Reason_LIMIT, // Note: Keep this enum in sync. with _trap_reason_name. - Reason_RECORDED_LIMIT = Reason_unloaded // some are not recorded per bc + Reason_RECORDED_LIMIT = Reason_bimorphic // some are not recorded per bc // Note: Reason_RECORDED_LIMIT should be < 8 to fit into 3 bits of // DataLayout::trap_bits. This dependency is enforced indirectly // via asserts, to avoid excessive direct header-to-header dependencies. @@ -278,7 +282,7 @@ int trap_state); static bool reason_is_recorded_per_bytecode(DeoptReason reason) { - return reason > Reason_none && reason < Reason_RECORDED_LIMIT; + return reason > Reason_none && reason <= Reason_RECORDED_LIMIT; } static DeoptReason reason_recorded_per_bytecode_if_any(DeoptReason reason) {
--- a/src/share/vm/runtime/frame.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/frame.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -107,7 +107,11 @@ address frame::raw_pc() const { if (is_deoptimized_frame()) { - return ((nmethod*) cb())->deopt_handler_begin() - pc_return_offset; + nmethod* nm = cb()->as_nmethod_or_null(); + if (nm->is_method_handle_return(pc())) + return nm->deopt_mh_handler_begin() - pc_return_offset; + else + return nm->deopt_handler_begin() - pc_return_offset; } else { return (pc() - pc_return_offset); } @@ -269,10 +273,16 @@ } // NeedsDeoptSuspend - address deopt = nm->deopt_handler_begin(); + // If the call site is a MethodHandle call site use the MH deopt + // handler. + address deopt = nm->is_method_handle_return(pc()) ? + nm->deopt_mh_handler_begin() : + nm->deopt_handler_begin(); + // Save the original pc before we patch in the new one nm->set_original_pc(this, pc()); patch_pc(thread, deopt); + #ifdef ASSERT { RegisterMap map(thread, false); @@ -596,12 +606,12 @@ for (BasicObjectLock* current = interpreter_frame_monitor_end(); current < interpreter_frame_monitor_begin(); current = next_monitor_in_interpreter_frame(current)) { - st->print_cr(" [ - obj "); + st->print(" - obj ["); current->obj()->print_value_on(st); - st->cr(); - st->print_cr(" - lock "); + st->print_cr("]"); + st->print(" - lock ["); current->lock()->print_on(st); - st->cr(); + st->print_cr("]"); } // monitor st->print_cr(" - monitor[" INTPTR_FORMAT "]", interpreter_frame_monitor_begin()); @@ -769,9 +779,9 @@ class InterpretedArgumentOopFinder: public SignatureInfo { private: - OopClosure* _f; // Closure to invoke - int _offset; // TOS-relative offset, decremented with each argument - bool _is_static; // true if the callee is a static method + OopClosure* _f; // Closure to invoke + int _offset; // TOS-relative offset, decremented with each argument + bool _has_receiver; // true if the callee has a receiver frame* _fr; void set(int size, BasicType type) { @@ -786,9 +796,9 @@ } public: - InterpretedArgumentOopFinder(symbolHandle signature, bool is_static, frame* fr, OopClosure* f) : SignatureInfo(signature) { + InterpretedArgumentOopFinder(symbolHandle signature, bool has_receiver, frame* fr, OopClosure* f) : SignatureInfo(signature), _has_receiver(has_receiver) { // compute size of arguments - int args_size = ArgumentSizeComputer(signature).size() + (is_static ? 0 : 1); + int args_size = ArgumentSizeComputer(signature).size() + (has_receiver ? 1 : 0); assert(!fr->is_interpreted_frame() || args_size <= fr->interpreter_frame_expression_stack_size(), "args cannot be on stack anymore"); @@ -796,11 +806,10 @@ _f = f; _fr = fr; _offset = args_size; - _is_static = is_static; } void oops_do() { - if (!_is_static) { + if (_has_receiver) { --_offset; oop_offset_do(); } @@ -912,7 +921,7 @@ int max_locals = m->is_native() ? m->size_of_parameters() : m->max_locals(); symbolHandle signature; - bool is_static = false; + bool has_receiver = false; // Process a callee's arguments if we are at a call site // (i.e., if we are at an invoke bytecode) @@ -922,7 +931,7 @@ Bytecode_invoke *call = Bytecode_invoke_at_check(m, bci); if (call != NULL) { signature = symbolHandle(thread, call->signature()); - is_static = call->is_invokestatic(); + has_receiver = call->has_receiver(); if (map->include_argument_oops() && interpreter_frame_expression_stack_size() > 0) { ResourceMark rm(thread); // is this right ??? @@ -936,7 +945,7 @@ // code in the interpreter calls a blocking runtime // routine which can cause this code to be executed). // (was bug gri 7/27/98) - oops_interpreted_arguments_do(signature, is_static, f); + oops_interpreted_arguments_do(signature, has_receiver, f); } } } @@ -950,7 +959,7 @@ mask = &oopmap_mask; #endif // ASSERT oops_interpreted_locals_do(f, max_locals, mask); - oops_interpreted_expressions_do(f, signature, is_static, + oops_interpreted_expressions_do(f, signature, has_receiver, m->max_stack(), max_locals, mask); } else { @@ -992,7 +1001,7 @@ void frame::oops_interpreted_expressions_do(OopClosure *f, symbolHandle signature, - bool is_static, + bool has_receiver, int max_stack, int max_locals, InterpreterOopMap *mask) { @@ -1005,7 +1014,7 @@ // arguments in callee's locals. int args_size = 0; if (!signature.is_null()) { - args_size = ArgumentSizeComputer(signature).size() + (is_static ? 0 : 1); + args_size = ArgumentSizeComputer(signature).size() + (has_receiver ? 1 : 0); } intptr_t *tos_addr = interpreter_frame_tos_at(args_size); @@ -1038,8 +1047,8 @@ } } -void frame::oops_interpreted_arguments_do(symbolHandle signature, bool is_static, OopClosure* f) { - InterpretedArgumentOopFinder finder(signature, is_static, this, f); +void frame::oops_interpreted_arguments_do(symbolHandle signature, bool has_receiver, OopClosure* f) { + InterpretedArgumentOopFinder finder(signature, has_receiver, this, f); finder.oops_do(); } @@ -1066,8 +1075,8 @@ class CompiledArgumentOopFinder: public SignatureInfo { protected: OopClosure* _f; - int _offset; // the current offset, incremented with each argument - bool _is_static; // true if the callee is a static method + int _offset; // the current offset, incremented with each argument + bool _has_receiver; // true if the callee has a receiver frame _fr; RegisterMap* _reg_map; int _arg_size; @@ -1087,24 +1096,24 @@ } public: - CompiledArgumentOopFinder(symbolHandle signature, bool is_static, OopClosure* f, frame fr, const RegisterMap* reg_map) + CompiledArgumentOopFinder(symbolHandle signature, bool has_receiver, OopClosure* f, frame fr, const RegisterMap* reg_map) : SignatureInfo(signature) { // initialize CompiledArgumentOopFinder _f = f; _offset = 0; - _is_static = is_static; + _has_receiver = has_receiver; _fr = fr; _reg_map = (RegisterMap*)reg_map; - _arg_size = ArgumentSizeComputer(signature).size() + (is_static ? 0 : 1); + _arg_size = ArgumentSizeComputer(signature).size() + (has_receiver ? 1 : 0); int arg_size; - _regs = SharedRuntime::find_callee_arguments(signature(), is_static, &arg_size); + _regs = SharedRuntime::find_callee_arguments(signature(), has_receiver, &arg_size); assert(arg_size == _arg_size, "wrong arg size"); } void oops_do() { - if (!_is_static) { + if (_has_receiver) { handle_oop_offset(); _offset++; } @@ -1112,9 +1121,9 @@ } }; -void frame::oops_compiled_arguments_do(symbolHandle signature, bool is_static, const RegisterMap* reg_map, OopClosure* f) { +void frame::oops_compiled_arguments_do(symbolHandle signature, bool has_receiver, const RegisterMap* reg_map, OopClosure* f) { ResourceMark rm; - CompiledArgumentOopFinder finder(signature, is_static, f, *this, reg_map); + CompiledArgumentOopFinder finder(signature, has_receiver, f, *this, reg_map); finder.oops_do(); } @@ -1190,9 +1199,19 @@ void frame::oops_do_internal(OopClosure* f, CodeBlobClosure* cf, RegisterMap* map, bool use_interpreter_oop_map_cache) { - if (is_interpreted_frame()) { oops_interpreted_do(f, map, use_interpreter_oop_map_cache); - } else if (is_entry_frame()) { oops_entry_do (f, map); - } else if (CodeCache::contains(pc())) { oops_code_blob_do (f, cf, map); +#ifndef PRODUCT + // simulate GC crash here to dump java thread in error report + if (CrashGCForDumpingJavaThread) { + char *t = NULL; + *t = 'c'; + } +#endif + if (is_interpreted_frame()) { + oops_interpreted_do(f, map, use_interpreter_oop_map_cache); + } else if (is_entry_frame()) { + oops_entry_do(f, map); + } else if (CodeCache::contains(pc())) { + oops_code_blob_do(f, cf, map); } else { ShouldNotReachHere(); }
--- a/src/share/vm/runtime/frame.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/frame.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -371,7 +371,7 @@ oop* oopmapreg_to_location(VMReg reg, const RegisterMap* regmap) const; // Oops-do's - void oops_compiled_arguments_do(symbolHandle signature, bool is_static, const RegisterMap* reg_map, OopClosure* f); + void oops_compiled_arguments_do(symbolHandle signature, bool has_receiver, const RegisterMap* reg_map, OopClosure* f); void oops_interpreted_do(OopClosure* f, const RegisterMap* map, bool query_oop_map_cache = true); private: @@ -379,9 +379,9 @@ int max_locals, InterpreterOopMap *mask); void oops_interpreted_expressions_do(OopClosure *f, symbolHandle signature, - bool is_static, int max_stack, int max_locals, + bool has_receiver, int max_stack, int max_locals, InterpreterOopMap *mask); - void oops_interpreted_arguments_do(symbolHandle signature, bool is_static, OopClosure* f); + void oops_interpreted_arguments_do(symbolHandle signature, bool has_receiver, OopClosure* f); // Iteration of oops void oops_do_internal(OopClosure* f, CodeBlobClosure* cf, RegisterMap* map, bool use_interpreter_oop_map_cache);
--- a/src/share/vm/runtime/globals.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/globals.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -46,7 +46,8 @@ bool Flag::is_unlocked() const { if (strcmp(kind, "{diagnostic}") == 0) { return UnlockDiagnosticVMOptions; - } else if (strcmp(kind, "{experimental}") == 0) { + } else if (strcmp(kind, "{experimental}") == 0 || + strcmp(kind, "{C2 experimental}") == 0) { return UnlockExperimentalVMOptions; } else { return true; @@ -69,9 +70,10 @@ void Flag::print_on(outputStream* st) { st->print("%5s %-35s %c= ", type, name, (origin != DEFAULT ? ':' : ' ')); - if (is_bool()) st->print("%-16s", get_bool() ? "true" : "false"); - if (is_intx()) st->print("%-16ld", get_intx()); - if (is_uintx()) st->print("%-16lu", get_uintx()); + if (is_bool()) st->print("%-16s", get_bool() ? "true" : "false"); + if (is_intx()) st->print("%-16ld", get_intx()); + if (is_uintx()) st->print("%-16lu", get_uintx()); + if (is_uint64_t()) st->print("%-16lu", get_uint64_t()); if (is_ccstr()) { const char* cp = get_ccstr(); if (cp != NULL) { @@ -100,6 +102,8 @@ st->print("-XX:%s=" INTX_FORMAT, name, get_intx()); } else if (is_uintx()) { st->print("-XX:%s=" UINTX_FORMAT, name, get_uintx()); + } else if (is_uint64_t()) { + st->print("-XX:%s=" UINT64_FORMAT, name, get_uint64_t()); } else if (is_ccstr()) { st->print("-XX:%s=", name); const char* cp = get_ccstr(); @@ -166,6 +170,7 @@ #define C2_PRODUCT_FLAG_STRUCT(type, name, value, doc) { #type, XSTR(name), &name, "{C2 product}", DEFAULT }, #define C2_PD_PRODUCT_FLAG_STRUCT(type, name, doc) { #type, XSTR(name), &name, "{C2 pd product}", DEFAULT }, #define C2_DIAGNOSTIC_FLAG_STRUCT(type, name, value, doc) { #type, XSTR(name), &name, "{C2 diagnostic}", DEFAULT }, +#define C2_EXPERIMENTAL_FLAG_STRUCT(type, name, value, doc) { #type, XSTR(name), &name, "{C2 experimental}", DEFAULT }, #ifdef PRODUCT #define C2_DEVELOP_FLAG_STRUCT(type, name, value, doc) /* flag is constant */ #define C2_PD_DEVELOP_FLAG_STRUCT(type, name, doc) /* flag is constant */ @@ -187,7 +192,7 @@ C1_FLAGS(C1_DEVELOP_FLAG_STRUCT, C1_PD_DEVELOP_FLAG_STRUCT, C1_PRODUCT_FLAG_STRUCT, C1_PD_PRODUCT_FLAG_STRUCT, C1_NOTPRODUCT_FLAG_STRUCT) #endif #ifdef COMPILER2 - C2_FLAGS(C2_DEVELOP_FLAG_STRUCT, C2_PD_DEVELOP_FLAG_STRUCT, C2_PRODUCT_FLAG_STRUCT, C2_PD_PRODUCT_FLAG_STRUCT, C2_DIAGNOSTIC_FLAG_STRUCT, C2_NOTPRODUCT_FLAG_STRUCT) + C2_FLAGS(C2_DEVELOP_FLAG_STRUCT, C2_PD_DEVELOP_FLAG_STRUCT, C2_PRODUCT_FLAG_STRUCT, C2_PD_PRODUCT_FLAG_STRUCT, C2_DIAGNOSTIC_FLAG_STRUCT, C2_EXPERIMENTAL_FLAG_STRUCT, C2_NOTPRODUCT_FLAG_STRUCT) #endif {0, NULL, NULL} }; @@ -324,6 +329,32 @@ faddr->origin = origin; } +bool CommandLineFlags::uint64_tAt(char* name, size_t len, uint64_t* value) { + Flag* result = Flag::find_flag(name, len); + if (result == NULL) return false; + if (!result->is_uint64_t()) return false; + *value = result->get_uint64_t(); + return true; +} + +bool CommandLineFlags::uint64_tAtPut(char* name, size_t len, uint64_t* value, FlagValueOrigin origin) { + Flag* result = Flag::find_flag(name, len); + if (result == NULL) return false; + if (!result->is_uint64_t()) return false; + uint64_t old_value = result->get_uint64_t(); + result->set_uint64_t(*value); + *value = old_value; + result->origin = origin; + return true; +} + +void CommandLineFlagsEx::uint64_tAtPut(CommandLineFlagWithType flag, uint64_t value, FlagValueOrigin origin) { + Flag* faddr = address_of_flag(flag); + guarantee(faddr != NULL && faddr->is_uint64_t(), "wrong flag type"); + faddr->set_uint64_t(value); + faddr->origin = origin; +} + bool CommandLineFlags::doubleAt(char* name, size_t len, double* value) { Flag* result = Flag::find_flag(name, len); if (result == NULL) return false; @@ -437,6 +468,8 @@ assert(Arguments::check_vm_args_consistency(), "Some flag settings conflict"); } +#endif // PRODUCT + void CommandLineFlags::printFlags() { // Print the flags sorted by name // note: this method is called before the thread structure is in place @@ -462,5 +495,3 @@ } FREE_C_HEAP_ARRAY(Flag*, array); } - -#endif
--- a/src/share/vm/runtime/globals.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/globals.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -47,8 +47,8 @@ define_pd_global(intx, OnStackReplacePercentage, 0); define_pd_global(bool, ResizeTLAB, false); define_pd_global(intx, FreqInlineSize, 0); +define_pd_global(intx, InlineSmallCode, 0); define_pd_global(intx, NewSizeThreadIncrease, 4*K); -define_pd_global(intx, NewRatio, 4); define_pd_global(intx, InlineClassNatives, true); define_pd_global(intx, InlineUnsafeOps, true); define_pd_global(intx, InitialCodeCacheSize, 160*K); @@ -58,7 +58,7 @@ define_pd_global(uintx,PermSize, ScaleForWordSize(4*M)); define_pd_global(uintx,MaxPermSize, ScaleForWordSize(64*M)); define_pd_global(bool, NeverActAsServerClassMachine, true); -define_pd_global(uintx, DefaultMaxRAM, 1*G); +define_pd_global(uint64_t,MaxRAM, 1ULL*G); #define CI_COMPILER_COUNT 0 #else @@ -113,6 +113,10 @@ uintx get_uintx() const { return *((uintx*) addr); } void set_uintx(uintx value) { *((uintx*) addr) = value; } + bool is_uint64_t() const { return strcmp(type, "uint64_t") == 0; } + uint64_t get_uint64_t() const { return *((uint64_t*) addr); } + void set_uint64_t(uint64_t value) { *((uint64_t*) addr) = value; } + bool is_double() const { return strcmp(type, "double") == 0; } double get_double() const { return *((double*) addr); } void set_double(double value) { *((double*) addr) = value; } @@ -188,6 +192,11 @@ static bool uintxAtPut(char* name, size_t len, uintx* value, FlagValueOrigin origin); static bool uintxAtPut(char* name, uintx* value, FlagValueOrigin origin) { return uintxAtPut(name, strlen(name), value, origin); } + static bool uint64_tAt(char* name, size_t len, uint64_t* value); + static bool uint64_tAt(char* name, uint64_t* value) { return uint64_tAt(name, strlen(name), value); } + static bool uint64_tAtPut(char* name, size_t len, uint64_t* value, FlagValueOrigin origin); + static bool uint64_tAtPut(char* name, uint64_t* value, FlagValueOrigin origin) { return uint64_tAtPut(name, strlen(name), value, origin); } + static bool doubleAt(char* name, size_t len, double* value); static bool doubleAt(char* name, double* value) { return doubleAt(name, strlen(name), value); } static bool doubleAtPut(char* name, size_t len, double* value, FlagValueOrigin origin); @@ -202,7 +211,7 @@ static bool wasSetOnCmdline(const char* name, bool* value); static void printSetFlags(); - static void printFlags() PRODUCT_RETURN; + static void printFlags(); static void verify() PRODUCT_RETURN; }; @@ -318,9 +327,6 @@ product(bool, UseMembar, false, \ "(Unstable) Issues membars on thread state transitions") \ \ - product(bool, PrintCommandLineFlags, false, \ - "Prints flags that appeared on the command line") \ - \ diagnostic(bool, UnlockDiagnosticVMOptions, trueInDebug, \ "Enable normal processing of flags relating to field diagnostics")\ \ @@ -736,6 +742,9 @@ diagnostic(bool, PrintAdapterHandlers, false, \ "Print code generated for i2c/c2i adapters") \ \ + develop(bool, VerifyAdapterSharing, false, \ + "Verify that the code for shared adapters is the equivalent") \ + \ diagnostic(bool, PrintAssembly, false, \ "Print assembly code (using external disassembler.so)") \ \ @@ -785,7 +794,7 @@ product(bool, ProfilerRecordPC, false, \ "Collects tick for each 16 byte interval of compiled code") \ \ - product(bool, ProfileVM, false, \ + product(bool, ProfileVM, false, \ "Profiles ticks that fall within VM (either in the VM Thread " \ "or VM code called through stubs)") \ \ @@ -815,7 +824,7 @@ \ product(bool, RegisterFinalizersAtInit, true, \ "Register finalizable objects at end of Object.<init> or " \ - "after allocation.") \ + "after allocation") \ \ develop(bool, RegisterReferences, true, \ "Tells whether the VM should register soft/weak/final/phantom " \ @@ -862,14 +871,14 @@ product(bool, AlwaysLockClassLoader, false, \ "Require the VM to acquire the class loader lock before calling " \ "loadClass() even for class loaders registering " \ - "as parallel capable. Default false. ") \ + "as parallel capable") \ \ product(bool, AllowParallelDefineClass, false, \ "Allow parallel defineClass requests for class loaders " \ - "registering as parallel capable. Default false") \ + "registering as parallel capable") \ \ product(bool, MustCallLoadClassInternal, false, \ - "Call loadClassInternal() rather than loadClass().Default false") \ + "Call loadClassInternal() rather than loadClass()") \ \ product_pd(bool, DontYieldALot, \ "Throw away obvious excess yield calls (for SOLARIS only)") \ @@ -921,9 +930,9 @@ "(Unstable, Linux-specific)" \ " avoid NPTL-FUTEX hang pthread_cond_timedwait" ) \ \ - product(bool, FilterSpuriousWakeups , true, \ - "Prevent spurious or premature wakeups from object.wait" \ - "(Solaris only)") \ + product(bool, FilterSpuriousWakeups, true, \ + "Prevent spurious or premature wakeups from object.wait " \ + "(Solaris only)") \ \ product(intx, NativeMonitorTimeout, -1, "(Unstable)" ) \ product(intx, NativeMonitorFlags, 0, "(Unstable)" ) \ @@ -973,7 +982,7 @@ \ product(bool, UseAltSigs, false, \ "Use alternate signals instead of SIGUSR1 & SIGUSR2 for VM " \ - "internal signals. (Solaris only)") \ + "internal signals (Solaris only)") \ \ product(bool, UseSpinning, false, \ "Use spinning in monitor inflation and before entry") \ @@ -1195,7 +1204,7 @@ product(bool, UseSerialGC, false, \ "Use the serial garbage collector") \ \ - experimental(bool, UseG1GC, false, \ + product(bool, UseG1GC, false, \ "Use the Garbage-First garbage collector") \ \ product(bool, UseParallelGC, false, \ @@ -1265,12 +1274,12 @@ "Always tenure objects in eden. (ParallelGC only)") \ \ product(bool, NeverTenure, false, \ - "Never tenure objects in eden, May tenure on overflow" \ - " (ParallelGC only)") \ + "Never tenure objects in eden, May tenure on overflow " \ + "(ParallelGC only)") \ \ product(bool, ScavengeBeforeFullGC, true, \ - "Scavenge youngest generation before each full GC," \ - " used with UseParallelGC") \ + "Scavenge youngest generation before each full GC, " \ + "used with UseParallelGC") \ \ develop(bool, ScavengeWithObjectsInToSpace, false, \ "Allow scavenges to occur when to_space contains objects.") \ @@ -1283,9 +1292,9 @@ " (effective only when UseConcMarkSweepGC)") \ \ product(bool, ExplicitGCInvokesConcurrentAndUnloadsClasses, false, \ - "A System.gc() request invokes a concurrent collection and" \ - " also unloads classes during such a concurrent gc cycle " \ - " (effective only when UseConcMarkSweepGC)") \ + "A System.gc() request invokes a concurrent collection and " \ + "also unloads classes during such a concurrent gc cycle " \ + "(effective only when UseConcMarkSweepGC)") \ \ develop(bool, UseCMSAdaptiveFreeLists, true, \ "Use Adaptive Free Lists in the CMS generation") \ @@ -1340,26 +1349,62 @@ "Whether we should simulate work queue overflow in ParNew") \ \ notproduct(uintx, ParGCWorkQueueOverflowInterval, 1000, \ - "An `interval' counter that determines how frequently" \ - " we simulate overflow; a smaller number increases frequency") \ + "An `interval' counter that determines how frequently " \ + "we simulate overflow; a smaller number increases frequency") \ \ product(uintx, ParGCDesiredObjsFromOverflowList, 20, \ "The desired number of objects to claim from the overflow list") \ \ - product(uintx, CMSParPromoteBlocksToClaim, 50, \ + product(uintx, CMSParPromoteBlocksToClaim, 16, \ "Number of blocks to attempt to claim when refilling CMS LAB for "\ "parallel GC.") \ \ + product(uintx, OldPLABWeight, 50, \ + "Percentage (0-100) used to weight the current sample when" \ + "computing exponentially decaying average for resizing CMSParPromoteBlocksToClaim.") \ + \ + product(bool, ResizeOldPLAB, true, \ + "Dynamically resize (old gen) promotion labs") \ + \ + product(bool, PrintOldPLAB, false, \ + "Print (old gen) promotion labs sizing decisions") \ + \ + product(uintx, CMSOldPLABMin, 16, \ + "Min size of CMS gen promotion lab caches per worker per blksize")\ + \ + product(uintx, CMSOldPLABMax, 1024, \ + "Max size of CMS gen promotion lab caches per worker per blksize")\ + \ + product(uintx, CMSOldPLABNumRefills, 4, \ + "Nominal number of refills of CMS gen promotion lab cache" \ + " per worker per block size") \ + \ + product(bool, CMSOldPLABResizeQuicker, false, \ + "Whether to react on-the-fly during a scavenge to a sudden" \ + " change in block demand rate") \ + \ + product(uintx, CMSOldPLABToleranceFactor, 4, \ + "The tolerance of the phase-change detector for on-the-fly" \ + " PLAB resizing during a scavenge") \ + \ + product(uintx, CMSOldPLABReactivityFactor, 2, \ + "The gain in the feedback loop for on-the-fly PLAB resizing" \ + " during a scavenge") \ + \ + product(uintx, CMSOldPLABReactivityCeiling, 10, \ + "The clamping of the gain in the feedback loop for on-the-fly" \ + " PLAB resizing during a scavenge") \ + \ product(bool, AlwaysPreTouch, false, \ "It forces all freshly committed pages to be pre-touched.") \ \ product(bool, CMSUseOldDefaults, false, \ - "A flag temporarily introduced to allow reverting to some older" \ - "default settings; older as of 6.0 ") \ + "A flag temporarily introduced to allow reverting to some " \ + "older default settings; older as of 6.0") \ \ product(intx, CMSYoungGenPerWorker, 16*M, \ "The amount of young gen chosen by default per GC worker " \ - "thread available ") \ + "thread available") \ \ product(bool, GCOverheadReporting, false, \ "Enables the GC overhead reporting facility") \ @@ -1380,43 +1425,71 @@ "automatically adjusted") \ \ product(uintx, CMSIncrementalDutyCycleMin, 0, \ - "Lower bound on the duty cycle when CMSIncrementalPacing is" \ - "enabled (a percentage, 0-100).") \ + "Lower bound on the duty cycle when CMSIncrementalPacing is " \ + "enabled (a percentage, 0-100)") \ \ product(uintx, CMSIncrementalSafetyFactor, 10, \ - "Percentage (0-100) used to add conservatism when computing the" \ - "duty cycle.") \ + "Percentage (0-100) used to add conservatism when computing the " \ + "duty cycle") \ \ product(uintx, CMSIncrementalOffset, 0, \ "Percentage (0-100) by which the CMS incremental mode duty cycle" \ - "is shifted to the right within the period between young GCs") \ - \ - product(uintx, CMSExpAvgFactor, 25, \ + " is shifted to the right within the period between young GCs") \ + \ + product(uintx, CMSExpAvgFactor, 50, \ "Percentage (0-100) used to weight the current sample when" \ "computing exponential averages for CMS statistics.") \ \ - product(uintx, CMS_FLSWeight, 50, \ + product(uintx, CMS_FLSWeight, 75, \ "Percentage (0-100) used to weight the current sample when" \ "computing exponentially decating averages for CMS FLS statistics.") \ \ - product(uintx, CMS_FLSPadding, 2, \ + product(uintx, CMS_FLSPadding, 1, \ "The multiple of deviation from mean to use for buffering" \ "against volatility in free list demand.") \ \ product(uintx, FLSCoalescePolicy, 2, \ "CMS: Aggression level for coalescing, increasing from 0 to 4") \ \ - product(uintx, CMS_SweepWeight, 50, \ - "Percentage (0-100) used to weight the current sample when" \ - "computing exponentially decaying average for inter-sweep duration.") \ - \ - product(uintx, CMS_SweepPadding, 2, \ - "The multiple of deviation from mean to use for buffering" \ + product(bool, FLSAlwaysCoalesceLarge, false, \ + "CMS: Larger free blocks are always available for coalescing") \ + \ + product(double, FLSLargestBlockCoalesceProximity, 0.99, \ + "CMS: the smaller the percentage the greater the coalition force")\ + \ + product(double, CMSSmallCoalSurplusPercent, 1.05, \ + "CMS: the factor by which to inflate estimated demand of small" \ + " block sizes to prevent coalescing with an adjoining block") \ + \ + product(double, CMSLargeCoalSurplusPercent, 0.95, \ + "CMS: the factor by which to inflate estimated demand of large" \ + " block sizes to prevent coalescing with an adjoining block") \ + \ + product(double, CMSSmallSplitSurplusPercent, 1.10, \ + "CMS: the factor by which to inflate estimated demand of small" \ + " block sizes to prevent splitting to supply demand for smaller" \ + " blocks") \ + \ + product(double, CMSLargeSplitSurplusPercent, 1.00, \ + "CMS: the factor by which to inflate estimated demand of large" \ + " block sizes to prevent splitting to supply demand for smaller" \ + " blocks") \ + \ + product(bool, CMSExtrapolateSweep, false, \ + "CMS: cushion for block demand during sweep") \ + \ + product(uintx, CMS_SweepWeight, 75, \ + "Percentage (0-100) used to weight the current sample when " \ + "computing exponentially decaying average for inter-sweep " \ + "duration") \ + \ + product(uintx, CMS_SweepPadding, 1, \ + "The multiple of deviation from mean to use for buffering " \ "against volatility in inter-sweep duration.") \ \ product(uintx, CMS_SweepTimerThresholdMillis, 10, \ "Skip block flux-rate sampling for an epoch unless inter-sweep " \ - " duration exceeds this threhold in milliseconds") \ + "duration exceeds this threhold in milliseconds") \ \ develop(bool, CMSTraceIncrementalMode, false, \ "Trace CMS incremental mode") \ @@ -1449,6 +1522,13 @@ product(uintx, CMSIndexedFreeListReplenish, 4, \ "Replenish and indexed free list with this number of chunks") \ \ + product(bool, CMSReplenishIntermediate, true, \ + "Replenish all intermediate free-list caches") \ + \ + product(bool, CMSSplitIndexedFreeListBlocks, true, \ + "When satisfying batched demand, splot blocks from the " \ + "IndexedFreeList whose size is a multiple of requested size") \ + \ product(bool, CMSLoopWarn, false, \ "Warn in case of excessive CMS looping") \ \ @@ -1583,6 +1663,18 @@ "Bitmap operations should process at most this many bits" \ "between yields") \ \ + product(bool, CMSDumpAtPromotionFailure, false, \ + "Dump useful information about the state of the CMS old " \ + " generation upon a promotion failure.") \ + \ + product(bool, CMSPrintChunksInDump, false, \ + "In a dump enabled by CMSDumpAtPromotionFailure, include " \ + " more detailed information about the free chunks.") \ + \ + product(bool, CMSPrintObjectsInDump, false, \ + "In a dump enabled by CMSDumpAtPromotionFailure, include " \ + " more detailed information about the allocated objects.") \ + \ diagnostic(bool, FLSVerifyAllHeapReferences, false, \ "Verify that all refs across the FLS boundary " \ " are to valid objects") \ @@ -1617,35 +1709,36 @@ \ product(intx, CMSTriggerRatio, 80, \ "Percentage of MinHeapFreeRatio in CMS generation that is " \ - " allocated before a CMS collection cycle commences") \ + "allocated before a CMS collection cycle commences") \ \ product(intx, CMSTriggerPermRatio, 80, \ - "Percentage of MinHeapFreeRatio in the CMS perm generation that" \ - " is allocated before a CMS collection cycle commences, that " \ - " also collects the perm generation") \ + "Percentage of MinHeapFreeRatio in the CMS perm generation that " \ + "is allocated before a CMS collection cycle commences, that " \ + "also collects the perm generation") \ \ product(uintx, CMSBootstrapOccupancy, 50, \ "Percentage CMS generation occupancy at which to " \ - " initiate CMS collection for bootstrapping collection stats") \ + "initiate CMS collection for bootstrapping collection stats") \ \ product(intx, CMSInitiatingOccupancyFraction, -1, \ "Percentage CMS generation occupancy to start a CMS collection " \ - " cycle (A negative value means that CMSTriggerRatio is used)") \ + "cycle. A negative value means that CMSTriggerRatio is used") \ \ product(intx, CMSInitiatingPermOccupancyFraction, -1, \ - "Percentage CMS perm generation occupancy to start a CMScollection"\ - " cycle (A negative value means that CMSTriggerPermRatio is used)")\ + "Percentage CMS perm generation occupancy to start a " \ + "CMScollection cycle. A negative value means that " \ + "CMSTriggerPermRatio is used") \ \ product(bool, UseCMSInitiatingOccupancyOnly, false, \ "Only use occupancy as a crierion for starting a CMS collection") \ \ product(intx, CMSIsTooFullPercentage, 98, \ - "An absolute ceiling above which CMS will always consider the" \ - " perm gen ripe for collection") \ + "An absolute ceiling above which CMS will always consider the " \ + "perm gen ripe for collection") \ \ develop(bool, CMSTestInFreeList, false, \ "Check if the coalesced range is already in the " \ - "free lists as claimed.") \ + "free lists as claimed") \ \ notproduct(bool, CMSVerifyReturnedBytes, false, \ "Check that all the garbage collected was returned to the " \ @@ -1663,8 +1756,12 @@ "Enforce ScavengeALot/GCALot at all potential safepoints") \ \ product(bool, HandlePromotionFailure, true, \ - "The youngest generation collection does not require" \ - " a guarantee of full promotion of all live objects.") \ + "The youngest generation collection does not require " \ + "a guarantee of full promotion of all live objects.") \ + \ + product(bool, PrintPromotionFailure, false, \ + "Print additional diagnostic information following " \ + " promotion failure") \ \ notproduct(bool, PromotionFailureALot, false, \ "Use promotion failure handling on every youngest generation " \ @@ -1692,7 +1789,7 @@ "Ratio of hard spins to calls to yield") \ \ product(uintx, PreserveMarkStackSize, 1024, \ - "Size for stack used in promotion failure handling") \ + "Size for stack used in promotion failure handling") \ \ product_pd(bool, UseTLAB, "Use thread-local object allocation") \ \ @@ -1720,14 +1817,27 @@ product(bool, AlwaysActAsServerClassMachine, false, \ "Always act like a server-class machine") \ \ - product_pd(uintx, DefaultMaxRAM, \ - "Maximum real memory size for setting server class heap size") \ + product_pd(uint64_t, MaxRAM, \ + "Real memory size (in bytes) used to set maximum heap size") \ + \ + product(uintx, ErgoHeapSizeLimit, 0, \ + "Maximum ergonomically set heap size (in bytes); zero means use " \ + "MaxRAM / MaxRAMFraction") \ + \ + product(uintx, MaxRAMFraction, 4, \ + "Maximum fraction (1/n) of real memory used for maximum heap " \ + "size") \ \ product(uintx, DefaultMaxRAMFraction, 4, \ - "Fraction (1/n) of real memory used for server class max heap") \ - \ - product(uintx, DefaultInitialRAMFraction, 64, \ - "Fraction (1/n) of real memory used for server class initial heap") \ + "Maximum fraction (1/n) of real memory used for maximum heap " \ + "size; deprecated: to be renamed to MaxRAMFraction") \ + \ + product(uintx, MinRAMFraction, 2, \ + "Minimum fraction (1/n) of real memory used for maxmimum heap " \ + "size on systems with small physical memory size") \ + \ + product(uintx, InitialRAMFraction, 64, \ + "Fraction (1/n) of real memory used for initial heap size") \ \ product(bool, UseAutoGCSelectPolicy, false, \ "Use automatic collection selection policy") \ @@ -1778,7 +1888,7 @@ "Number of collections before the adaptive sizing is started") \ \ product(uintx, AdaptiveSizePolicyOutputInterval, 0, \ - "Collecton interval for printing information, zero => never") \ + "Collecton interval for printing information; zero => never") \ \ product(bool, UseAdaptiveSizePolicyFootprintGoal, true, \ "Use adaptive minimum footprint as a goal") \ @@ -1808,7 +1918,8 @@ "Allowed collection cost difference between generations") \ \ product(uintx, AdaptiveSizePolicyCollectionCostMargin, 50, \ - "If collection costs are within margin, reduce both by full delta") \ + "If collection costs are within margin, reduce both by full " \ + "delta") \ \ product(uintx, YoungGenerationSizeIncrement, 20, \ "Adaptive size percentage change in young generation") \ @@ -1904,6 +2015,10 @@ diagnostic(bool, GCParallelVerificationEnabled, true, \ "Enable parallel memory system verification") \ \ + diagnostic(bool, DeferInitialCardMark, false, \ + "When +ReduceInitialCardMarks, explicitly defer any that " \ + "may arise from new_pre_store_barrier") \ + \ diagnostic(bool, VerifyRememberedSets, false, \ "Verify GC remembered sets") \ \ @@ -1942,9 +2057,6 @@ "number of times a GC thread (minus the coordinator) " \ "will sleep while yielding before giving up and resuming GC") \ \ - notproduct(bool, PrintFlagsFinal, false, \ - "Print all command line flags after argument processing") \ - \ /* gc tracing */ \ manageable(bool, PrintGC, false, \ "Print message at garbage collect") \ @@ -2244,11 +2356,20 @@ "If false, restricts profiled locations to the root method only") \ \ product(bool, PrintVMOptions, trueInDebug, \ - "print VM flag settings") \ + "Print flags that appeared on the command line") \ \ product(bool, IgnoreUnrecognizedVMOptions, false, \ "Ignore unrecognized VM options") \ \ + product(bool, PrintCommandLineFlags, false, \ + "Print flags specified on command line or set by ergonomics") \ + \ + product(bool, PrintFlagsInitial, false, \ + "Print all VM flags before argument processing and exit VM") \ + \ + product(bool, PrintFlagsFinal, false, \ + "Print all VM flags after argument and ergonomic processing") \ + \ diagnostic(bool, SerializeVMOutput, true, \ "Use a mutex to serialize output to tty and hotspot.log") \ \ @@ -2329,6 +2450,9 @@ notproduct(bool, CompileTheWorldIgnoreInitErrors, false, \ "Compile all methods although class initializer failed") \ \ + notproduct(intx, CompileTheWorldSafepointInterval, 100, \ + "Force a safepoint every n compiles so sweeper can keep up") \ + \ develop(bool, TraceIterativeGVN, false, \ "Print progress during Iterative Global Value Numbering") \ \ @@ -2527,8 +2651,11 @@ \ develop(bool, VerifyCompiledCode, false, \ "Include miscellaneous runtime verifications in nmethod code; " \ - "off by default because it disturbs nmethod size heuristics.") \ - \ + "default off because it disturbs nmethod size heuristics") \ + \ + notproduct(bool, CrashGCForDumpingJavaThread, false, \ + "Manually make GC thread crash then dump java stack trace; " \ + "Test only") \ \ /* compilation */ \ product(bool, UseCompiler, true, \ @@ -2644,10 +2771,10 @@ notproduct(intx, MaxSubklassPrintSize, 4, \ "maximum number of subklasses to print when printing klass") \ \ - develop(intx, MaxInlineLevel, 9, \ + product(intx, MaxInlineLevel, 9, \ "maximum number of nested calls that are inlined") \ \ - develop(intx, MaxRecursiveInlineLevel, 1, \ + product(intx, MaxRecursiveInlineLevel, 1, \ "maximum number of nested recursive calls that are inlined") \ \ product_pd(intx, InlineSmallCode, \ @@ -2660,10 +2787,10 @@ product_pd(intx, FreqInlineSize, \ "maximum bytecode size of a frequent method to be inlined") \ \ - develop(intx, MaxTrivialSize, 6, \ + product(intx, MaxTrivialSize, 6, \ "maximum bytecode size of a trivial method to be inlined") \ \ - develop(intx, MinInliningThreshold, 250, \ + product(intx, MinInliningThreshold, 250, \ "min. invocation count a method needs to have to be inlined") \ \ develop(intx, AlignEntryCode, 4, \ @@ -2740,7 +2867,7 @@ product(intx, PerMethodRecompilationCutoff, 400, \ "After recompiling N times, stay in the interpreter (-1=>'Inf')") \ \ - product(intx, PerBytecodeRecompilationCutoff, 100, \ + product(intx, PerBytecodeRecompilationCutoff, 200, \ "Per-BCI limit on repeated recompilation (-1=>'Inf')") \ \ product(intx, PerMethodTrapLimit, 100, \ @@ -2789,20 +2916,28 @@ "an OS lock") \ \ /* gc parameters */ \ - product(uintx, MaxHeapSize, ScaleForWordSize(64*M), \ - "Default maximum size for object heap (in bytes)") \ - \ - product_pd(uintx, NewSize, \ - "Default size of new generation (in bytes)") \ + product(uintx, InitialHeapSize, 0, \ + "Initial heap size (in bytes); zero means OldSize + NewSize") \ + \ + product(uintx, MaxHeapSize, ScaleForWordSize(96*M), \ + "Maximum heap size (in bytes)") \ + \ + product(uintx, OldSize, ScaleForWordSize(4*M), \ + "Initial tenured generation size (in bytes)") \ + \ + product(uintx, NewSize, ScaleForWordSize(4*M), \ + "Initial new generation size (in bytes)") \ \ product(uintx, MaxNewSize, max_uintx, \ - "Maximum size of new generation (in bytes)") \ + "Maximum new generation size (in bytes), max_uintx means set " \ + "ergonomically") \ \ product(uintx, PretenureSizeThreshold, 0, \ - "Max size in bytes of objects allocated in DefNew generation") \ - \ - product_pd(uintx, TLABSize, \ - "Default (or starting) size of TLAB (in bytes)") \ + "Maximum size in bytes of objects allocated in DefNew " \ + "generation; zero means no maximum") \ + \ + product(uintx, TLABSize, 0, \ + "Starting TLAB size (in bytes); zero means set ergonomically") \ \ product(uintx, MinTLABSize, 2*K, \ "Minimum allowed TLAB size (in bytes)") \ @@ -2819,10 +2954,10 @@ product(uintx, TLABWasteIncrement, 4, \ "Increment allowed waste at slow allocation") \ \ - product_pd(intx, SurvivorRatio, \ + product(intx, SurvivorRatio, 8, \ "Ratio of eden/survivor space size") \ \ - product_pd(intx, NewRatio, \ + product(intx, NewRatio, 2, \ "Ratio of new/old generation sizes") \ \ product(uintx, MaxLiveObjectEvacuationRatio, 100, \ @@ -2832,11 +2967,8 @@ "Additional size added to desired new generation size per " \ "non-daemon thread (in bytes)") \ \ - product(uintx, OldSize, ScaleForWordSize(4096*K), \ - "Default size of tenured generation (in bytes)") \ - \ product_pd(uintx, PermSize, \ - "Default size of permanent generation (in bytes)") \ + "Initial size of permanent generation (in bytes)") \ \ product_pd(uintx, MaxPermSize, \ "Maximum size of permanent generation (in bytes)") \ @@ -2988,6 +3120,15 @@ notproduct(bool, ExitOnFullCodeCache, false, \ "Exit the VM if we fill the code cache.") \ \ + product(bool, UseCodeCacheFlushing, false, \ + "Attempt to clean the code cache before shutting off compiler") \ + \ + product(intx, MinCodeCacheFlushingInterval, 30, \ + "Min number of seconds between code cache cleaning sessions") \ + \ + product(uintx, CodeCacheFlushingMinimumFreeSpace, 1500*K, \ + "When less than X space left, start code cache cleaning") \ + \ /* interpreter debugging */ \ develop(intx, BinarySwitchThreshold, 5, \ "Minimal number of lookupswitch entries for rewriting to binary " \ @@ -3334,6 +3475,9 @@ diagnostic(bool, OptimizeMethodHandles, true, \ "when constructing method handles, try to improve them") \ \ + experimental(bool, TrustFinalNonStaticFields, false, \ + "trust final non-static declarations for constant folding") \ + \ experimental(bool, EnableInvokeDynamic, false, \ "recognize the invokedynamic instruction") \ \
--- a/src/share/vm/runtime/globals_extension.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/globals_extension.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -64,6 +64,7 @@ #define C2_PRODUCT_FLAG_MEMBER(type, name, value, doc) FLAG_MEMBER(name), #define C2_PD_PRODUCT_FLAG_MEMBER(type, name, doc) FLAG_MEMBER(name), #define C2_DIAGNOSTIC_FLAG_MEMBER(type, name, value, doc) FLAG_MEMBER(name), +#define C2_EXPERIMENTAL_FLAG_MEMBER(type, name, value, doc) FLAG_MEMBER(name), #ifdef PRODUCT #define C2_DEVELOP_FLAG_MEMBER(type, name, value, doc) /* flag is constant */ #define C2_PD_DEVELOP_FLAG_MEMBER(type, name, doc) /* flag is constant */ @@ -84,7 +85,7 @@ C1_FLAGS(C1_DEVELOP_FLAG_MEMBER, C1_PD_DEVELOP_FLAG_MEMBER, C1_PRODUCT_FLAG_MEMBER, C1_PD_PRODUCT_FLAG_MEMBER, C1_NOTPRODUCT_FLAG_MEMBER) #endif #ifdef COMPILER2 - C2_FLAGS(C2_DEVELOP_FLAG_MEMBER, C2_PD_DEVELOP_FLAG_MEMBER, C2_PRODUCT_FLAG_MEMBER, C2_PD_PRODUCT_FLAG_MEMBER, C2_DIAGNOSTIC_FLAG_MEMBER, C2_NOTPRODUCT_FLAG_MEMBER) + C2_FLAGS(C2_DEVELOP_FLAG_MEMBER, C2_PD_DEVELOP_FLAG_MEMBER, C2_PRODUCT_FLAG_MEMBER, C2_PD_PRODUCT_FLAG_MEMBER, C2_DIAGNOSTIC_FLAG_MEMBER, C2_EXPERIMENTAL_FLAG_MEMBER, C2_NOTPRODUCT_FLAG_MEMBER) #endif NUM_CommandLineFlag } CommandLineFlag; @@ -130,6 +131,7 @@ #define C2_PRODUCT_FLAG_MEMBER_WITH_TYPE(type, name, value, doc) FLAG_MEMBER_WITH_TYPE(name,type), #define C2_PD_PRODUCT_FLAG_MEMBER_WITH_TYPE(type, name, doc) FLAG_MEMBER_WITH_TYPE(name,type), #define C2_DIAGNOSTIC_FLAG_MEMBER_WITH_TYPE(type, name, value, doc) FLAG_MEMBER_WITH_TYPE(name,type), +#define C2_EXPERIMENTAL_FLAG_MEMBER_WITH_TYPE(type, name, value, doc) FLAG_MEMBER_WITH_TYPE(name,type), #ifdef PRODUCT #define C2_DEVELOP_FLAG_MEMBER_WITH_TYPE(type, name, value, doc) /* flag is constant */ #define C2_PD_DEVELOP_FLAG_MEMBER_WITH_TYPE(type, name, doc) /* flag is constant */ @@ -181,6 +183,7 @@ C2_PRODUCT_FLAG_MEMBER_WITH_TYPE, C2_PD_PRODUCT_FLAG_MEMBER_WITH_TYPE, C2_DIAGNOSTIC_FLAG_MEMBER_WITH_TYPE, + C2_EXPERIMENTAL_FLAG_MEMBER_WITH_TYPE, C2_NOTPRODUCT_FLAG_MEMBER_WITH_TYPE) #endif NUM_CommandLineFlagWithType @@ -202,6 +205,7 @@ static void boolAtPut(CommandLineFlagWithType flag, bool value, FlagValueOrigin origin); static void intxAtPut(CommandLineFlagWithType flag, intx value, FlagValueOrigin origin); static void uintxAtPut(CommandLineFlagWithType flag, uintx value, FlagValueOrigin origin); + static void uint64_tAtPut(CommandLineFlagWithType flag, uint64_t value, FlagValueOrigin origin); static void doubleAtPut(CommandLineFlagWithType flag, double value, FlagValueOrigin origin); static void ccstrAtPut(CommandLineFlagWithType flag, ccstr value, FlagValueOrigin origin);
--- a/src/share/vm/runtime/jniHandles.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/jniHandles.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -144,7 +144,7 @@ EXCEPTION_MARK; // We will never reach the CATCH below since Exceptions::_throw will cause // the VM to exit if an exception is thrown during initialization - klassOop k = SystemDictionary::object_klass(); + klassOop k = SystemDictionary::Object_klass(); _deleted_handle = instanceKlass::cast(k)->allocate_permanent_instance(CATCH); }
--- a/src/share/vm/runtime/os.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/os.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -280,7 +280,7 @@ string, CHECK); - KlassHandle group(THREAD, SystemDictionary::threadGroup_klass()); + KlassHandle group(THREAD, SystemDictionary::ThreadGroup_klass()); JavaCalls::call_special(&result, thread_group, group,
--- a/src/share/vm/runtime/os.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/os.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -60,24 +60,26 @@ typedef void (*java_call_t)(JavaValue* value, methodHandle* method, JavaCallArguments* args, Thread* thread); class os: AllStatic { - private: + public: enum { page_sizes_max = 9 }; // Size of _page_sizes array (8 plus a sentinel) + private: static OSThread* _starting_thread; static address _polling_page; static volatile int32_t * _mem_serialize_page; static uintptr_t _serialize_page_mask; + public: static size_t _page_sizes[page_sizes_max]; + private: static void init_page_sizes(size_t default_page_size) { _page_sizes[0] = default_page_size; _page_sizes[1] = 0; // sentinel } public: - - static void init(void); // Called before command line parsing - static jint init_2(void); // Called after command line parsing + static void init(void); // Called before command line parsing + static jint init_2(void); // Called after command line parsing // File names are case-insensitive on windows only // Override me as needed @@ -141,6 +143,7 @@ static int processor_count() { return _processor_count; } + static void set_processor_count(int count) { _processor_count = count; } // Returns the number of CPUs this process is currently allowed to run on. // Note that on some OSes this can change dynamically. @@ -294,19 +297,16 @@ } static bool is_memory_serialize_page(JavaThread *thread, address addr) { - address thr_addr; if (UseMembar) return false; - // Calculate thread specific address + // Previously this function calculated the exact address of this + // thread's serialize page, and checked if the faulting address + // was equal. However, some platforms mask off faulting addresses + // to the page size, so now we just check that the address is + // within the page. This makes the thread argument unnecessary, + // but we retain the NULL check to preserve existing behaviour. if (thread == NULL) return false; - // TODO-FIXME: some platforms mask off faulting addresses to the base pagesize. - // Instead of using a test for equality we should probably use something - // of the form: - // return ((_mem_serialize_page ^ addr) & -pagesize) == 0 - // - thr_addr = (address)(((uintptr_t)thread >> - get_serialize_page_shift_count()) & - get_serialize_page_mask()) + (uintptr_t)_mem_serialize_page; - return (thr_addr == addr); + address page = (address) _mem_serialize_page; + return addr >= page && addr < (page + os::vm_page_size()); } static void block_on_serialize_page_trap();
--- a/src/share/vm/runtime/reflection.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/reflection.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -449,7 +449,7 @@ // sun/reflect/MagicAccessorImpl subclasses to succeed trivially. if ( JDK_Version::is_gte_jdk14x_version() && UseNewReflection - && Klass::cast(current_class)->is_subclass_of(SystemDictionary::reflect_magic_klass())) { + && Klass::cast(current_class)->is_subclass_of(SystemDictionary::reflect_MagicAccessorImpl_klass())) { return true; } @@ -482,6 +482,11 @@ under_host_klass(accessee_ik, accessor)) return true; + // Adapter frames can access anything. + if (MethodHandleCompiler::klass_is_method_handle_adapter_holder(accessor)) + // This is an internal adapter frame from the MethodHandleCompiler. + return true; + if (RelaxAccessControlCheck || (accessor_ik->major_version() < JAVA_1_5_VERSION && accessee_ik->major_version() < JAVA_1_5_VERSION)) { @@ -541,7 +546,7 @@ // sun/reflect/MagicAccessorImpl subclasses to succeed trivially. if ( JDK_Version::is_gte_jdk14x_version() && UseNewReflection - && Klass::cast(current_class)->is_subclass_of(SystemDictionary::reflect_magic_klass())) { + && Klass::cast(current_class)->is_subclass_of(SystemDictionary::reflect_MagicAccessorImpl_klass())) { return true; } @@ -631,7 +636,7 @@ objArrayHandle Reflection::get_parameter_types(methodHandle method, int parameter_count, oop* return_type, TRAPS) { // Allocate array holding parameter types (java.lang.Class instances) - objArrayOop m = oopFactory::new_objArray(SystemDictionary::class_klass(), parameter_count, CHECK_(objArrayHandle())); + objArrayOop m = oopFactory::new_objArray(SystemDictionary::Class_klass(), parameter_count, CHECK_(objArrayHandle())); objArrayHandle mirrors (THREAD, m); int index = 0; // Collect parameter types @@ -1308,7 +1313,7 @@ if (Klass::cast(klass)->oop_is_array() && which == MEMBER_DECLARED) return NULL; if (Klass::cast(java_lang_Class::as_klassOop(mirror))->oop_is_array()) { - klass = SystemDictionary::object_klass(); + klass = SystemDictionary::Object_klass(); } instanceKlassHandle h_k(THREAD, klass); @@ -1375,13 +1380,13 @@ // Exclude primitive types if (java_lang_Class::is_primitive(mirror) || (Klass::cast(java_lang_Class::as_klassOop(mirror))->oop_is_array() && (which == MEMBER_DECLARED))) { - klassOop klass = SystemDictionary::reflect_method_klass(); + klassOop klass = SystemDictionary::reflect_Method_klass(); return oopFactory::new_objArray(klass, 0, CHECK_NULL); // Return empty array } klassOop klass = java_lang_Class::as_klassOop(mirror); if (Klass::cast(java_lang_Class::as_klassOop(mirror))->oop_is_array()) { - klass = SystemDictionary::object_klass(); + klass = SystemDictionary::Object_klass(); } instanceKlassHandle h_k(THREAD, klass); @@ -1411,7 +1416,7 @@ } // Allocate result - klassOop klass = SystemDictionary::reflect_method_klass(); + klassOop klass = SystemDictionary::reflect_Method_klass(); objArrayOop r = oopFactory::new_objArray(klass, count, CHECK_NULL); objArrayHandle h_result (THREAD, r); @@ -1462,7 +1467,7 @@ } } // Allocate result - klassOop klass = SystemDictionary::reflect_method_klass(); + klassOop klass = SystemDictionary::reflect_Method_klass(); objArrayOop r = oopFactory::new_objArray(klass, count, CHECK_NULL); objArrayHandle h_result (THREAD, r); @@ -1523,7 +1528,7 @@ bool prim = java_lang_Class::is_primitive(mirror); Klass* k = prim ? NULL : Klass::cast(java_lang_Class::as_klassOop(mirror)); if (prim || k->is_interface() || k->oop_is_array()) { - return oopFactory::new_objArray(SystemDictionary::reflect_constructor_klass(), 0, CHECK_NULL); // Return empty array + return oopFactory::new_objArray(SystemDictionary::reflect_Constructor_klass(), 0, CHECK_NULL); // Return empty array } // Must be instanceKlass at this point
--- a/src/share/vm/runtime/reflectionUtils.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/reflectionUtils.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -63,15 +63,15 @@ void FilteredFieldsMap::initialize() { int offset; offset = java_lang_Throwable::get_backtrace_offset(); - _filtered_fields->append(new FilteredField(SystemDictionary::throwable_klass(), offset)); + _filtered_fields->append(new FilteredField(SystemDictionary::Throwable_klass(), offset)); // The latest version of vm may be used with old jdk. if (JDK_Version::is_gte_jdk16x_version()) { // The following class fields do not exist in // previous version of jdk. offset = sun_reflect_ConstantPool::cp_oop_offset(); - _filtered_fields->append(new FilteredField(SystemDictionary::reflect_constant_pool_klass(), offset)); + _filtered_fields->append(new FilteredField(SystemDictionary::reflect_ConstantPool_klass(), offset)); offset = sun_reflect_UnsafeStaticFieldAccessorImpl::base_offset(); - _filtered_fields->append(new FilteredField(SystemDictionary::reflect_unsafe_static_field_accessor_impl_klass(), offset)); + _filtered_fields->append(new FilteredField(SystemDictionary::reflect_UnsafeStaticFieldAccessorImpl_klass(), offset)); } }
--- a/src/share/vm/runtime/sharedRuntime.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/sharedRuntime.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -364,7 +364,7 @@ void SharedRuntime::throw_and_post_jvmti_exception(JavaThread *thread, Handle h_exception) { - if (JvmtiExport::can_post_exceptions()) { + if (JvmtiExport::can_post_on_exceptions()) { vframeStream vfst(thread, true); methodHandle method = methodHandle(thread, vfst.method()); address bcp = method()->bcp_from(vfst.bci()); @@ -607,7 +607,9 @@ _implicit_null_throws++; #endif target_pc = nm->continuation_for_implicit_exception(pc); - guarantee(target_pc != 0, "must have a continuation point"); + // If there's an unexpected fault, target_pc might be NULL, + // in which case we want to fall through into the normal + // error handling code. } break; // fall through @@ -621,14 +623,15 @@ _implicit_div0_throws++; #endif target_pc = nm->continuation_for_implicit_exception(pc); - guarantee(target_pc != 0, "must have a continuation point"); + // If there's an unexpected fault, target_pc might be NULL, + // in which case we want to fall through into the normal + // error handling code. break; // fall through } default: ShouldNotReachHere(); } - guarantee(target_pc != NULL, "must have computed destination PC for implicit exception"); assert(exception_kind == IMPLICIT_NULL || exception_kind == IMPLICIT_DIVIDE_BY_ZERO, "wrong implicit exception kind"); // for AbortVMOnException flag @@ -802,7 +805,7 @@ #ifdef ASSERT // Check that the receiver klass is of the right subtype and that it is initialized for virtual calls - if (bc != Bytecodes::_invokestatic) { + if (bc != Bytecodes::_invokestatic && bc != Bytecodes::_invokedynamic) { assert(receiver.not_null(), "should have thrown exception"); KlassHandle receiver_klass (THREAD, receiver->klass()); klassOop rk = constants->klass_ref_at(bytecode_index, CHECK_(nullHandle)); @@ -860,7 +863,7 @@ if (JvmtiExport::can_hotswap_or_post_breakpoint()) { int retry_count = 0; while (!HAS_PENDING_EXCEPTION && callee_method->is_old() && - callee_method->method_holder() != SystemDictionary::object_klass()) { + callee_method->method_holder() != SystemDictionary::Object_klass()) { // If has a pending exception then there is no need to re-try to // resolve this method. // If the method has been redefined, we need to try again. @@ -1027,7 +1030,26 @@ frame stub_frame = thread->last_frame(); assert(stub_frame.is_runtime_frame(), "sanity check"); frame caller_frame = stub_frame.sender(®_map); - if (caller_frame.is_interpreted_frame() || caller_frame.is_entry_frame() ) { + + // MethodHandle invokes don't have a CompiledIC and should always + // simply redispatch to the callee_target. + address sender_pc = caller_frame.pc(); + CodeBlob* sender_cb = caller_frame.cb(); + nmethod* sender_nm = sender_cb->as_nmethod_or_null(); + bool is_mh_invoke_via_adapter = false; // Direct c2c call or via adapter? + if (sender_nm != NULL && sender_nm->is_method_handle_return(sender_pc)) { + // If the callee_target is set, then we have come here via an i2c + // adapter. + methodOop callee = thread->callee_target(); + if (callee != NULL) { + assert(callee->is_method(), "sanity"); + is_mh_invoke_via_adapter = true; + } + } + + if (caller_frame.is_interpreted_frame() || + caller_frame.is_entry_frame() || + is_mh_invoke_via_adapter) { methodOop callee = thread->callee_target(); guarantee(callee != NULL && callee->is_method(), "bad handshake"); thread->set_vm_result(callee); @@ -1342,7 +1364,7 @@ // We are calling the interpreter via a c2i. Normally this would mean that // we were called by a compiled method. However we could have lost a race // where we went int -> i2c -> c2i and so the caller could in fact be -// interpreted. If the caller is compiled we attampt to patch the caller +// interpreted. If the caller is compiled we attempt to patch the caller // so he no longer calls into the interpreter. IRT_LEAF(void, SharedRuntime::fixup_callers_callsite(methodOopDesc* method, address caller_pc)) methodOop moop(method); @@ -1358,10 +1380,19 @@ // we did we'd leap into space because the callsite needs to use // "to interpreter" stub in order to load up the methodOop. Don't // ask me how I know this... - // CodeBlob* cb = CodeCache::find_blob(caller_pc); - if ( !cb->is_nmethod() || entry_point == moop->get_c2i_entry()) { + if (!cb->is_nmethod() || entry_point == moop->get_c2i_entry()) { + return; + } + + // The check above makes sure this is a nmethod. + nmethod* nm = cb->as_nmethod_or_null(); + assert(nm, "must be"); + + // Don't fixup MethodHandle call sites as c2i/i2c adapters are used + // to implement MethodHandle actions. + if (nm->is_method_handle_return(caller_pc)) { return; } @@ -1376,7 +1407,7 @@ if (moop->code() == NULL) return; - if (((nmethod*)cb)->is_in_use()) { + if (nm->is_in_use()) { // Expect to find a native call there (unless it was no-inline cache vtable dispatch) MutexLockerEx ml_patch(Patching_lock, Mutex::_no_safepoint_check_flag); @@ -1408,7 +1439,7 @@ if (callee == cb || callee->is_adapter_blob()) { // static call or optimized virtual if (TraceCallFixup) { - tty->print("fixup callsite at " INTPTR_FORMAT " to compiled code for", caller_pc); + tty->print("fixup callsite at " INTPTR_FORMAT " to compiled code for", caller_pc); moop->print_short_name(tty); tty->print_cr(" to " INTPTR_FORMAT, entry_point); } @@ -1424,7 +1455,7 @@ } } else { if (TraceCallFixup) { - tty->print("already patched callsite at " INTPTR_FORMAT " to compiled code for", caller_pc); + tty->print("already patched callsite at " INTPTR_FORMAT " to compiled code for", caller_pc); moop->print_short_name(tty); tty->print_cr(" to " INTPTR_FORMAT, entry_point); } @@ -1529,7 +1560,7 @@ oop SharedRuntime::wrong_method_type_is_for_single_argument(JavaThread* thr, oopDesc* required) { if (required == NULL) return NULL; - if (required->klass() == SystemDictionary::class_klass()) + if (required->klass() == SystemDictionary::Class_klass()) return required; if (required->is_klass()) return Klass::cast(klassOop(required))->java_mirror(); @@ -1671,6 +1702,8 @@ if( _find_handler_ctr ) tty->print_cr("%5d find exception handler", _find_handler_ctr ); if( _rethrow_ctr ) tty->print_cr("%5d rethrow handler", _rethrow_ctr ); + AdapterHandlerLibrary::print_statistics(); + if (xtty != NULL) xtty->tail("statistics"); } @@ -1771,11 +1804,282 @@ #endif +// A simple wrapper class around the calling convention information +// that allows sharing of adapters for the same calling convention. +class AdapterFingerPrint : public CHeapObj { + private: + union { + int _compact[3]; + int* _fingerprint; + } _value; + int _length; // A negative length indicates the fingerprint is in the compact form, + // Otherwise _value._fingerprint is the array. + + // Remap BasicTypes that are handled equivalently by the adapters. + // These are correct for the current system but someday it might be + // necessary to make this mapping platform dependent. + static BasicType adapter_encoding(BasicType in) { + assert((~0xf & in) == 0, "must fit in 4 bits"); + switch(in) { + case T_BOOLEAN: + case T_BYTE: + case T_SHORT: + case T_CHAR: + // There are all promoted to T_INT in the calling convention + return T_INT; + + case T_OBJECT: + case T_ARRAY: + if (!TaggedStackInterpreter) { +#ifdef _LP64 + return T_LONG; +#else + return T_INT; +#endif + } + return T_OBJECT; + + case T_INT: + case T_LONG: + case T_FLOAT: + case T_DOUBLE: + case T_VOID: + return in; + + default: + ShouldNotReachHere(); + return T_CONFLICT; + } + } + + public: + AdapterFingerPrint(int total_args_passed, BasicType* sig_bt) { + // The fingerprint is based on the BasicType signature encoded + // into an array of ints with four entries per int. + int* ptr; + int len = (total_args_passed + 3) >> 2; + if (len <= (int)(sizeof(_value._compact) / sizeof(int))) { + _value._compact[0] = _value._compact[1] = _value._compact[2] = 0; + // Storing the signature encoded as signed chars hits about 98% + // of the time. + _length = -len; + ptr = _value._compact; + } else { + _length = len; + _value._fingerprint = NEW_C_HEAP_ARRAY(int, _length); + ptr = _value._fingerprint; + } + + // Now pack the BasicTypes with 4 per int + int sig_index = 0; + for (int index = 0; index < len; index++) { + int value = 0; + for (int byte = 0; byte < 4; byte++) { + if (sig_index < total_args_passed) { + value = (value << 4) | adapter_encoding(sig_bt[sig_index++]); + } + } + ptr[index] = value; + } + } + + ~AdapterFingerPrint() { + if (_length > 0) { + FREE_C_HEAP_ARRAY(int, _value._fingerprint); + } + } + + int value(int index) { + if (_length < 0) { + return _value._compact[index]; + } + return _value._fingerprint[index]; + } + int length() { + if (_length < 0) return -_length; + return _length; + } + + bool is_compact() { + return _length <= 0; + } + + unsigned int compute_hash() { + int hash = 0; + for (int i = 0; i < length(); i++) { + int v = value(i); + hash = (hash << 8) ^ v ^ (hash >> 5); + } + return (unsigned int)hash; + } + + const char* as_string() { + stringStream st; + for (int i = 0; i < length(); i++) { + st.print(PTR_FORMAT, value(i)); + } + return st.as_string(); + } + + bool equals(AdapterFingerPrint* other) { + if (other->_length != _length) { + return false; + } + if (_length < 0) { + return _value._compact[0] == other->_value._compact[0] && + _value._compact[1] == other->_value._compact[1] && + _value._compact[2] == other->_value._compact[2]; + } else { + for (int i = 0; i < _length; i++) { + if (_value._fingerprint[i] != other->_value._fingerprint[i]) { + return false; + } + } + } + return true; + } +}; + + +// A hashtable mapping from AdapterFingerPrints to AdapterHandlerEntries +class AdapterHandlerTable : public BasicHashtable { + friend class AdapterHandlerTableIterator; + + private: + +#ifndef PRODUCT + static int _lookups; // number of calls to lookup + static int _buckets; // number of buckets checked + static int _equals; // number of buckets checked with matching hash + static int _hits; // number of successful lookups + static int _compact; // number of equals calls with compact signature +#endif + + AdapterHandlerEntry* bucket(int i) { + return (AdapterHandlerEntry*)BasicHashtable::bucket(i); + } + + public: + AdapterHandlerTable() + : BasicHashtable(293, sizeof(AdapterHandlerEntry)) { } + + // Create a new entry suitable for insertion in the table + AdapterHandlerEntry* new_entry(AdapterFingerPrint* fingerprint, address i2c_entry, address c2i_entry, address c2i_unverified_entry) { + AdapterHandlerEntry* entry = (AdapterHandlerEntry*)BasicHashtable::new_entry(fingerprint->compute_hash()); + entry->init(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry); + return entry; + } + + // Insert an entry into the table + void add(AdapterHandlerEntry* entry) { + int index = hash_to_index(entry->hash()); + add_entry(index, entry); + } + + void free_entry(AdapterHandlerEntry* entry) { + entry->deallocate(); + BasicHashtable::free_entry(entry); + } + + // Find a entry with the same fingerprint if it exists + AdapterHandlerEntry* lookup(int total_args_passed, BasicType* sig_bt) { + NOT_PRODUCT(_lookups++); + AdapterFingerPrint fp(total_args_passed, sig_bt); + unsigned int hash = fp.compute_hash(); + int index = hash_to_index(hash); + for (AdapterHandlerEntry* e = bucket(index); e != NULL; e = e->next()) { + NOT_PRODUCT(_buckets++); + if (e->hash() == hash) { + NOT_PRODUCT(_equals++); + if (fp.equals(e->fingerprint())) { +#ifndef PRODUCT + if (fp.is_compact()) _compact++; + _hits++; +#endif + return e; + } + } + } + return NULL; + } + +#ifndef PRODUCT + void print_statistics() { + ResourceMark rm; + int longest = 0; + int empty = 0; + int total = 0; + int nonempty = 0; + for (int index = 0; index < table_size(); index++) { + int count = 0; + for (AdapterHandlerEntry* e = bucket(index); e != NULL; e = e->next()) { + count++; + } + if (count != 0) nonempty++; + if (count == 0) empty++; + if (count > longest) longest = count; + total += count; + } + tty->print_cr("AdapterHandlerTable: empty %d longest %d total %d average %f", + empty, longest, total, total / (double)nonempty); + tty->print_cr("AdapterHandlerTable: lookups %d buckets %d equals %d hits %d compact %d", + _lookups, _buckets, _equals, _hits, _compact); + } +#endif +}; + + +#ifndef PRODUCT + +int AdapterHandlerTable::_lookups; +int AdapterHandlerTable::_buckets; +int AdapterHandlerTable::_equals; +int AdapterHandlerTable::_hits; +int AdapterHandlerTable::_compact; + +class AdapterHandlerTableIterator : public StackObj { + private: + AdapterHandlerTable* _table; + int _index; + AdapterHandlerEntry* _current; + + void scan() { + while (_index < _table->table_size()) { + AdapterHandlerEntry* a = _table->bucket(_index); + if (a != NULL) { + _current = a; + return; + } + _index++; + } + } + + public: + AdapterHandlerTableIterator(AdapterHandlerTable* table): _table(table), _index(0), _current(NULL) { + scan(); + } + bool has_next() { + return _current != NULL; + } + AdapterHandlerEntry* next() { + if (_current != NULL) { + AdapterHandlerEntry* result = _current; + _current = _current->next(); + if (_current == NULL) scan(); + return result; + } else { + return NULL; + } + } +}; +#endif + + // --------------------------------------------------------------------------- // Implementation of AdapterHandlerLibrary const char* AdapterHandlerEntry::name = "I2C/C2I adapters"; -GrowableArray<uint64_t>* AdapterHandlerLibrary::_fingerprints = NULL; -GrowableArray<AdapterHandlerEntry* >* AdapterHandlerLibrary::_handlers = NULL; +AdapterHandlerTable* AdapterHandlerLibrary::_adapters = NULL; +AdapterHandlerEntry* AdapterHandlerLibrary::_abstract_method_handler = NULL; const int AdapterHandlerLibrary_size = 16*K; BufferBlob* AdapterHandlerLibrary::_buffer = NULL; @@ -1787,28 +2091,31 @@ } void AdapterHandlerLibrary::initialize() { - if (_fingerprints != NULL) return; - _fingerprints = new(ResourceObj::C_HEAP)GrowableArray<uint64_t>(32, true); - _handlers = new(ResourceObj::C_HEAP)GrowableArray<AdapterHandlerEntry*>(32, true); - // Index 0 reserved for the slow path handler - _fingerprints->append(0/*the never-allowed 0 fingerprint*/); - _handlers->append(NULL); + if (_adapters != NULL) return; + _adapters = new AdapterHandlerTable(); // Create a special handler for abstract methods. Abstract methods // are never compiled so an i2c entry is somewhat meaningless, but // fill it in with something appropriate just in case. Pass handle // wrong method for the c2i transitions. address wrong_method = SharedRuntime::get_handle_wrong_method_stub(); - _fingerprints->append(0/*the never-allowed 0 fingerprint*/); - assert(_handlers->length() == AbstractMethodHandler, "in wrong slot"); - _handlers->append(new AdapterHandlerEntry(StubRoutines::throw_AbstractMethodError_entry(), - wrong_method, wrong_method)); + _abstract_method_handler = AdapterHandlerLibrary::new_entry(new AdapterFingerPrint(0, NULL), + StubRoutines::throw_AbstractMethodError_entry(), + wrong_method, wrong_method); } -int AdapterHandlerLibrary::get_create_adapter_index(methodHandle method) { - // Use customized signature handler. Need to lock around updates to the - // _fingerprints array (it is not safe for concurrent readers and a single - // writer: this can be fixed if it becomes a problem). +AdapterHandlerEntry* AdapterHandlerLibrary::new_entry(AdapterFingerPrint* fingerprint, + address i2c_entry, + address c2i_entry, + address c2i_unverified_entry) { + return _adapters->new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry); +} + +AdapterHandlerEntry* AdapterHandlerLibrary::get_adapter(methodHandle method) { + // Use customized signature handler. Need to lock around updates to + // the AdapterHandlerTable (it is not safe for concurrent readers + // and a single writer: this could be fixed if it becomes a + // problem). // Get the address of the ic_miss handlers before we grab the // AdapterHandlerLibrary_lock. This fixes bug 6236259 which @@ -1819,47 +2126,58 @@ address ic_miss = SharedRuntime::get_ic_miss_stub(); assert(ic_miss != NULL, "must have handler"); - int result; + ResourceMark rm; + NOT_PRODUCT(int code_size); BufferBlob *B = NULL; AdapterHandlerEntry* entry = NULL; - uint64_t fingerprint; + AdapterFingerPrint* fingerprint = NULL; { MutexLocker mu(AdapterHandlerLibrary_lock); // make sure data structure is initialized initialize(); if (method->is_abstract()) { - return AbstractMethodHandler; + return _abstract_method_handler; } + // Fill in the signature array, for the calling-convention call. + int total_args_passed = method->size_of_parameters(); // All args on stack + + BasicType* sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_args_passed); + VMRegPair* regs = NEW_RESOURCE_ARRAY(VMRegPair, total_args_passed); + int i = 0; + if (!method->is_static()) // Pass in receiver first + sig_bt[i++] = T_OBJECT; + for (SignatureStream ss(method->signature()); !ss.at_return_type(); ss.next()) { + sig_bt[i++] = ss.type(); // Collect remaining bits of signature + if (ss.type() == T_LONG || ss.type() == T_DOUBLE) + sig_bt[i++] = T_VOID; // Longs & doubles take 2 Java slots + } + assert(i == total_args_passed, ""); + // Lookup method signature's fingerprint - fingerprint = Fingerprinter(method).fingerprint(); - assert( fingerprint != CONST64( 0), "no zero fingerprints allowed" ); - // Fingerprints are small fixed-size condensed representations of - // signatures. If the signature is too large, it won't fit in a - // fingerprint. Signatures which cannot support a fingerprint get a new i2c - // adapter gen'd each time, instead of searching the cache for one. This -1 - // game can be avoided if I compared signatures instead of using - // fingerprints. However, -1 fingerprints are very rare. - if( fingerprint != UCONST64(-1) ) { // If this is a cache-able fingerprint - // Turns out i2c adapters do not care what the return value is. Mask it - // out so signatures that only differ in return type will share the same - // adapter. - fingerprint &= ~(SignatureIterator::result_feature_mask << SignatureIterator::static_feature_size); - // Search for a prior existing i2c/c2i adapter - int index = _fingerprints->find(fingerprint); - if( index >= 0 ) return index; // Found existing handlers? - } else { - // Annoyingly, I end up adding -1 fingerprints to the array of handlers, - // because I need a unique handler index. It cannot be scanned for - // because all -1's look alike. Instead, the matching index is passed out - // and immediately used to collect the 2 return values (the c2i and i2c - // adapters). + entry = _adapters->lookup(total_args_passed, sig_bt); + +#ifdef ASSERT + AdapterHandlerEntry* shared_entry = NULL; + if (VerifyAdapterSharing && entry != NULL) { + shared_entry = entry; + entry = NULL; + } +#endif + + if (entry != NULL) { + return entry; } + // Get a description of the compiled java calling convention and the largest used (VMReg) stack slot usage + int comp_args_on_stack = SharedRuntime::java_calling_convention(sig_bt, regs, total_args_passed, false); + + // Make a C heap allocated version of the fingerprint to store in the adapter + fingerprint = new AdapterFingerPrint(total_args_passed, sig_bt); + // Create I2C & C2I handlers - ResourceMark rm; BufferBlob* buf = buffer_blob(); // the temporary code buffer in CodeCache if (buf != NULL) { @@ -1869,32 +2187,26 @@ sizeof(buffer_locs)/sizeof(relocInfo)); MacroAssembler _masm(&buffer); - // Fill in the signature array, for the calling-convention call. - int total_args_passed = method->size_of_parameters(); // All args on stack - - BasicType* sig_bt = NEW_RESOURCE_ARRAY(BasicType,total_args_passed); - VMRegPair * regs = NEW_RESOURCE_ARRAY(VMRegPair ,total_args_passed); - int i=0; - if( !method->is_static() ) // Pass in receiver first - sig_bt[i++] = T_OBJECT; - for( SignatureStream ss(method->signature()); !ss.at_return_type(); ss.next()) { - sig_bt[i++] = ss.type(); // Collect remaining bits of signature - if( ss.type() == T_LONG || ss.type() == T_DOUBLE ) - sig_bt[i++] = T_VOID; // Longs & doubles take 2 Java slots - } - assert( i==total_args_passed, "" ); - - // Now get the re-packed compiled-Java layout. - int comp_args_on_stack; - - // Get a description of the compiled java calling convention and the largest used (VMReg) stack slot usage - comp_args_on_stack = SharedRuntime::java_calling_convention(sig_bt, regs, total_args_passed, false); - entry = SharedRuntime::generate_i2c2i_adapters(&_masm, total_args_passed, comp_args_on_stack, sig_bt, - regs); + regs, + fingerprint); + +#ifdef ASSERT + if (VerifyAdapterSharing) { + if (shared_entry != NULL) { + assert(shared_entry->compare_code(buf->instructions_begin(), buffer.code_size(), total_args_passed, sig_bt), + "code must match"); + // Release the one just created and return the original + _adapters->free_entry(entry); + return shared_entry; + } else { + entry->save_code(buf->instructions_begin(), buffer.code_size(), total_args_passed, sig_bt); + } + } +#endif B = BufferBlob::create(AdapterHandlerEntry::name, &buffer); NOT_PRODUCT(code_size = buffer.code_size()); @@ -1903,49 +2215,33 @@ // CodeCache is full, disable compilation // Ought to log this but compile log is only per compile thread // and we're some non descript Java thread. - UseInterpreter = true; - if (UseCompiler || AlwaysCompileLoopMethods ) { -#ifndef PRODUCT - warning("CodeCache is full. Compiler has been disabled"); - if (CompileTheWorld || ExitOnFullCodeCache) { - before_exit(JavaThread::current()); - exit_globals(); // will delete tty - vm_direct_exit(CompileTheWorld ? 0 : 1); - } -#endif - UseCompiler = false; - AlwaysCompileLoopMethods = false; - } - return 0; // Out of CodeCache space (_handlers[0] == NULL) + MutexUnlocker mu(AdapterHandlerLibrary_lock); + CompileBroker::handle_full_code_cache(); + return NULL; // Out of CodeCache space } entry->relocate(B->instructions_begin()); #ifndef PRODUCT // debugging suppport if (PrintAdapterHandlers) { tty->cr(); - tty->print_cr("i2c argument handler #%d for: %s %s (fingerprint = 0x%llx, %d bytes generated)", - _handlers->length(), (method->is_static() ? "static" : "receiver"), - method->signature()->as_C_string(), fingerprint, code_size ); + tty->print_cr("i2c argument handler #%d for: %s %s (fingerprint = %s, %d bytes generated)", + _adapters->number_of_entries(), (method->is_static() ? "static" : "receiver"), + method->signature()->as_C_string(), fingerprint->as_string(), code_size ); tty->print_cr("c2i argument handler starts at %p",entry->get_c2i_entry()); Disassembler::decode(entry->get_i2c_entry(), entry->get_i2c_entry() + code_size); } #endif - // add handlers to library - _fingerprints->append(fingerprint); - _handlers->append(entry); - // set handler index - assert(_fingerprints->length() == _handlers->length(), "sanity check"); - result = _fingerprints->length() - 1; + _adapters->add(entry); } // Outside of the lock if (B != NULL) { char blob_id[256]; jio_snprintf(blob_id, sizeof(blob_id), - "%s(" PTR64_FORMAT ")@" PTR_FORMAT, + "%s(%s)@" PTR_FORMAT, AdapterHandlerEntry::name, - fingerprint, + fingerprint->as_string(), B->instructions_begin()); VTune::register_stub(blob_id, B->instructions_begin(), B->instructions_end()); Forte::register_stub(blob_id, B->instructions_begin(), B->instructions_end()); @@ -1956,7 +2252,7 @@ B->instructions_end()); } } - return result; + return entry; } void AdapterHandlerEntry::relocate(address new_base) { @@ -1966,6 +2262,44 @@ _c2i_unverified_entry += delta; } + +void AdapterHandlerEntry::deallocate() { + delete _fingerprint; +#ifdef ASSERT + if (_saved_code) FREE_C_HEAP_ARRAY(unsigned char, _saved_code); + if (_saved_sig) FREE_C_HEAP_ARRAY(Basictype, _saved_sig); +#endif +} + + +#ifdef ASSERT +// Capture the code before relocation so that it can be compared +// against other versions. If the code is captured after relocation +// then relative instructions won't be equivalent. +void AdapterHandlerEntry::save_code(unsigned char* buffer, int length, int total_args_passed, BasicType* sig_bt) { + _saved_code = NEW_C_HEAP_ARRAY(unsigned char, length); + _code_length = length; + memcpy(_saved_code, buffer, length); + _total_args_passed = total_args_passed; + _saved_sig = NEW_C_HEAP_ARRAY(BasicType, _total_args_passed); + memcpy(_saved_sig, sig_bt, _total_args_passed * sizeof(BasicType)); +} + + +bool AdapterHandlerEntry::compare_code(unsigned char* buffer, int length, int total_args_passed, BasicType* sig_bt) { + if (length != _code_length) { + return false; + } + for (int i = 0; i < length; i++) { + if (buffer[i] != _saved_code[i]) { + return false; + } + } + return true; +} +#endif + + // Create a native wrapper for this native method. The wrapper converts the // java compiled calling convention to the native convention, handlizes // arguments, and transitions to native. On return from the native we transition @@ -2044,19 +2378,8 @@ // CodeCache is full, disable compilation // Ought to log this but compile log is only per compile thread // and we're some non descript Java thread. - UseInterpreter = true; - if (UseCompiler || AlwaysCompileLoopMethods ) { -#ifndef PRODUCT - warning("CodeCache is full. Compiler has been disabled"); - if (CompileTheWorld || ExitOnFullCodeCache) { - before_exit(JavaThread::current()); - exit_globals(); // will delete tty - vm_direct_exit(CompileTheWorld ? 0 : 1); - } -#endif - UseCompiler = false; - AlwaysCompileLoopMethods = false; - } + MutexUnlocker mu(AdapterHandlerLibrary_lock); + CompileBroker::handle_full_code_cache(); } return nm; } @@ -2136,7 +2459,7 @@ return regs.first(); } -VMRegPair *SharedRuntime::find_callee_arguments(symbolOop sig, bool is_static, int* arg_size) { +VMRegPair *SharedRuntime::find_callee_arguments(symbolOop sig, bool has_receiver, int* arg_size) { // This method is returning a data structure allocating as a // ResourceObject, so do not put any ResourceMarks in here. char *s = sig->as_C_string(); @@ -2148,7 +2471,7 @@ BasicType *sig_bt = NEW_RESOURCE_ARRAY( BasicType, 256 ); VMRegPair *regs = NEW_RESOURCE_ARRAY( VMRegPair, 256 ); int cnt = 0; - if (!is_static) { + if (has_receiver) { sig_bt[cnt++] = T_OBJECT; // Receiver is argument 0; not in signature } @@ -2299,30 +2622,31 @@ #ifndef PRODUCT bool AdapterHandlerLibrary::contains(CodeBlob* b) { - - if (_handlers == NULL) return false; - - for (int i = 0 ; i < _handlers->length() ; i++) { - AdapterHandlerEntry* a = get_entry(i); - if ( a != NULL && b == CodeCache::find_blob(a->get_i2c_entry()) ) return true; + AdapterHandlerTableIterator iter(_adapters); + while (iter.has_next()) { + AdapterHandlerEntry* a = iter.next(); + if ( b == CodeCache::find_blob(a->get_i2c_entry()) ) return true; } return false; } void AdapterHandlerLibrary::print_handler(CodeBlob* b) { - - for (int i = 0 ; i < _handlers->length() ; i++) { - AdapterHandlerEntry* a = get_entry(i); - if ( a != NULL && b == CodeCache::find_blob(a->get_i2c_entry()) ) { + AdapterHandlerTableIterator iter(_adapters); + while (iter.has_next()) { + AdapterHandlerEntry* a = iter.next(); + if ( b == CodeCache::find_blob(a->get_i2c_entry()) ) { tty->print("Adapter for signature: "); - // Fingerprinter::print(_fingerprints->at(i)); - tty->print("0x%" FORMAT64_MODIFIER "x", _fingerprints->at(i)); - tty->print_cr(" i2c: " INTPTR_FORMAT " c2i: " INTPTR_FORMAT " c2iUV: " INTPTR_FORMAT, + tty->print_cr("%s i2c: " INTPTR_FORMAT " c2i: " INTPTR_FORMAT " c2iUV: " INTPTR_FORMAT, + a->fingerprint()->as_string(), a->get_i2c_entry(), a->get_c2i_entry(), a->get_c2i_unverified_entry()); - return; } } assert(false, "Should have found handler"); } + +void AdapterHandlerLibrary::print_statistics() { + _adapters->print_statistics(); +} + #endif /* PRODUCT */
--- a/src/share/vm/runtime/sharedRuntime.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/sharedRuntime.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -23,6 +23,8 @@ */ class AdapterHandlerEntry; +class AdapterHandlerTable; +class AdapterFingerPrint; class vframeStream; // Runtime is the base class for various runtime interfaces @@ -337,7 +339,8 @@ int total_args_passed, int max_arg, const BasicType *sig_bt, - const VMRegPair *regs); + const VMRegPair *regs, + AdapterFingerPrint* fingerprint); // OSR support @@ -357,7 +360,7 @@ // Convert a sig into a calling convention register layout // and find interesting things about it. - static VMRegPair* find_callee_arguments(symbolOop sig, bool is_static, int *arg_size); + static VMRegPair* find_callee_arguments(symbolOop sig, bool has_receiver, int *arg_size); static VMReg name_for_receiver(); // "Top of Stack" slots that may be unused by the calling convention but must @@ -528,28 +531,64 @@ // used by the adapters. The code generation happens here because it's very // similar to what the adapters have to do. -class AdapterHandlerEntry : public CHeapObj { +class AdapterHandlerEntry : public BasicHashtableEntry { + friend class AdapterHandlerTable; + private: + AdapterFingerPrint* _fingerprint; address _i2c_entry; address _c2i_entry; address _c2i_unverified_entry; +#ifdef ASSERT + // Captures code and signature used to generate this adapter when + // verifing adapter equivalence. + unsigned char* _saved_code; + int _code_length; + BasicType* _saved_sig; + int _total_args_passed; +#endif + + void init(AdapterFingerPrint* fingerprint, address i2c_entry, address c2i_entry, address c2i_unverified_entry) { + _fingerprint = fingerprint; + _i2c_entry = i2c_entry; + _c2i_entry = c2i_entry; + _c2i_unverified_entry = c2i_unverified_entry; +#ifdef ASSERT + _saved_code = NULL; + _code_length = 0; + _saved_sig = NULL; + _total_args_passed = 0; +#endif + } + + void deallocate(); + + // should never be used + AdapterHandlerEntry(); + public: - // The name we give all buffer blobs static const char* name; - AdapterHandlerEntry(address i2c_entry, address c2i_entry, address c2i_unverified_entry): - _i2c_entry(i2c_entry), - _c2i_entry(c2i_entry), - _c2i_unverified_entry(c2i_unverified_entry) { - } - address get_i2c_entry() { return _i2c_entry; } address get_c2i_entry() { return _c2i_entry; } address get_c2i_unverified_entry() { return _c2i_unverified_entry; } void relocate(address new_base); + + AdapterFingerPrint* fingerprint() { return _fingerprint; } + + AdapterHandlerEntry* next() { + return (AdapterHandlerEntry*)BasicHashtableEntry::next(); + } + +#ifdef ASSERT + // Used to verify that code generated for shared adapters is equivalent + void save_code(unsigned char* code, int length, int total_args_passed, BasicType* sig_bt); + bool compare_code(unsigned char* code, int length, int total_args_passed, BasicType* sig_bt); +#endif + #ifndef PRODUCT void print(); #endif /* PRODUCT */ @@ -558,30 +597,18 @@ class AdapterHandlerLibrary: public AllStatic { private: static BufferBlob* _buffer; // the temporary code buffer in CodeCache - static GrowableArray<uint64_t>* _fingerprints; // the fingerprint collection - static GrowableArray<AdapterHandlerEntry*> * _handlers; // the corresponding handlers - enum { - AbstractMethodHandler = 1 // special handler for abstract methods - }; + static AdapterHandlerTable* _adapters; + static AdapterHandlerEntry* _abstract_method_handler; static BufferBlob* buffer_blob(); static void initialize(); - static int get_create_adapter_index(methodHandle method); - static address get_i2c_entry( int index ) { - return get_entry(index)->get_i2c_entry(); - } - static address get_c2i_entry( int index ) { - return get_entry(index)->get_c2i_entry(); - } - static address get_c2i_unverified_entry( int index ) { - return get_entry(index)->get_c2i_unverified_entry(); - } public: - static AdapterHandlerEntry* get_entry( int index ) { return _handlers->at(index); } + + static AdapterHandlerEntry* new_entry(AdapterFingerPrint* fingerprint, + address i2c_entry, address c2i_entry, address c2i_unverified_entry); static nmethod* create_native_wrapper(methodHandle method); - static AdapterHandlerEntry* get_adapter(methodHandle method) { - return get_entry(get_create_adapter_index(method)); - } + static AdapterHandlerEntry* get_adapter(methodHandle method); + #ifdef HAVE_DTRACE_H static nmethod* create_dtrace_nmethod (methodHandle method); #endif // HAVE_DTRACE_H @@ -589,6 +616,7 @@ #ifndef PRODUCT static void print_handler(CodeBlob* b); static bool contains(CodeBlob* b); + static void print_statistics(); #endif /* PRODUCT */ };
--- a/src/share/vm/runtime/statSampler.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/statSampler.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -177,7 +177,7 @@ // public static String getProperty(String key, String def); JavaCalls::call_static(&result, - KlassHandle(THREAD, SystemDictionary::system_klass()), + KlassHandle(THREAD, SystemDictionary::System_klass()), vmSymbolHandles::getProperty_name(), vmSymbolHandles::string_string_signature(), key_str,
--- a/src/share/vm/runtime/stubRoutines.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/stubRoutines.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2007 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -97,6 +97,14 @@ address StubRoutines::_unsafe_arraycopy = NULL; address StubRoutines::_generic_arraycopy = NULL; +double (* StubRoutines::_intrinsic_log )(double) = NULL; +double (* StubRoutines::_intrinsic_log10 )(double) = NULL; +double (* StubRoutines::_intrinsic_exp )(double) = NULL; +double (* StubRoutines::_intrinsic_pow )(double, double) = NULL; +double (* StubRoutines::_intrinsic_sin )(double) = NULL; +double (* StubRoutines::_intrinsic_cos )(double) = NULL; +double (* StubRoutines::_intrinsic_tan )(double) = NULL; + // Initialization // // Note: to break cycle with universe initialization, stubs are generated in two phases. @@ -188,11 +196,19 @@ // Default versions of arraycopy functions // +static void gen_arraycopy_barrier_pre(oop* dest, size_t count) { + assert(count != 0, "count should be non-zero"); + assert(count <= (size_t)max_intx, "count too large"); + BarrierSet* bs = Universe::heap()->barrier_set(); + assert(bs->has_write_ref_array_pre_opt(), "Must have pre-barrier opt"); + bs->write_ref_array_pre(dest, (int)count); +} + static void gen_arraycopy_barrier(oop* dest, size_t count) { assert(count != 0, "count should be non-zero"); BarrierSet* bs = Universe::heap()->barrier_set(); assert(bs->has_write_ref_array_opt(), "Barrier set must have ref array opt"); - bs->write_ref_array(MemRegion((HeapWord*)dest, (HeapWord*)(dest + count))); + bs->write_ref_array((HeapWord*)dest, count); } JRT_LEAF(void, StubRoutines::jbyte_copy(jbyte* src, jbyte* dest, size_t count)) @@ -232,6 +248,7 @@ SharedRuntime::_oop_array_copy_ctr++; // Slow-path oop array copy #endif // !PRODUCT assert(count != 0, "count should be non-zero"); + gen_arraycopy_barrier_pre(dest, count); Copy::conjoint_oops_atomic(src, dest, count); gen_arraycopy_barrier(dest, count); JRT_END @@ -273,6 +290,7 @@ SharedRuntime::_oop_array_copy_ctr++; // Slow-path oop array copy #endif // !PRODUCT assert(count != 0, "count should be non-zero"); + gen_arraycopy_barrier_pre((oop *) dest, count); Copy::arrayof_conjoint_oops(src, dest, count); gen_arraycopy_barrier((oop *) dest, count); JRT_END
--- a/src/share/vm/runtime/stubRoutines.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/stubRoutines.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2007 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -148,6 +148,20 @@ static address _unsafe_arraycopy; static address _generic_arraycopy; + // These are versions of the java.lang.Math methods which perform + // the same operations as the intrinsic version. They are used for + // constant folding in the compiler to ensure equivalence. If the + // intrinsic version returns the same result as the strict version + // then they can be set to the appropriate function from + // SharedRuntime. + static double (*_intrinsic_log)(double); + static double (*_intrinsic_log10)(double); + static double (*_intrinsic_exp)(double); + static double (*_intrinsic_pow)(double, double); + static double (*_intrinsic_sin)(double); + static double (*_intrinsic_cos)(double); + static double (*_intrinsic_tan)(double); + public: // Initialization/Testing static void initialize1(); // must happen before universe::genesis @@ -245,6 +259,35 @@ static address unsafe_arraycopy() { return _unsafe_arraycopy; } static address generic_arraycopy() { return _generic_arraycopy; } + static double intrinsic_log(double d) { + assert(_intrinsic_log != NULL, "must be defined"); + return _intrinsic_log(d); + } + static double intrinsic_log10(double d) { + assert(_intrinsic_log != NULL, "must be defined"); + return _intrinsic_log10(d); + } + static double intrinsic_exp(double d) { + assert(_intrinsic_exp != NULL, "must be defined"); + return _intrinsic_exp(d); + } + static double intrinsic_pow(double d, double d2) { + assert(_intrinsic_pow != NULL, "must be defined"); + return _intrinsic_pow(d, d2); + } + static double intrinsic_sin(double d) { + assert(_intrinsic_sin != NULL, "must be defined"); + return _intrinsic_sin(d); + } + static double intrinsic_cos(double d) { + assert(_intrinsic_cos != NULL, "must be defined"); + return _intrinsic_cos(d); + } + static double intrinsic_tan(double d) { + assert(_intrinsic_tan != NULL, "must be defined"); + return _intrinsic_tan(d); + } + // // Default versions of the above arraycopy functions for platforms which do // not have specialized versions
--- a/src/share/vm/runtime/sweeper.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/sweeper.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -33,6 +33,11 @@ jint NMethodSweeper::_locked_seen = 0; jint NMethodSweeper::_not_entrant_seen_on_stack = 0; bool NMethodSweeper::_rescan = false; +bool NMethodSweeper::_was_full = false; +jint NMethodSweeper::_advise_to_sweep = 0; +jlong NMethodSweeper::_last_was_full = 0; +uint NMethodSweeper::_highest_marked = 0; +long NMethodSweeper::_was_full_traversal = 0; class MarkActivationClosure: public CodeBlobClosure { public: @@ -114,6 +119,40 @@ tty->print_cr("### Couldn't make progress on some nmethods so stopping sweep"); } } + + if (UseCodeCacheFlushing) { + if (!CodeCache::needs_flushing()) { + // In a safepoint, no race with setters + _advise_to_sweep = 0; + } + + if (was_full()) { + // There was some progress so attempt to restart the compiler + jlong now = os::javaTimeMillis(); + jlong max_interval = (jlong)MinCodeCacheFlushingInterval * (jlong)1000; + jlong curr_interval = now - _last_was_full; + if ((!CodeCache::needs_flushing()) && (curr_interval > max_interval)) { + CompileBroker::set_should_compile_new_jobs(CompileBroker::run_compilation); + set_was_full(false); + + // Update the _last_was_full time so we can tell how fast the + // code cache is filling up + _last_was_full = os::javaTimeMillis(); + + if (PrintMethodFlushing) { + tty->print_cr("### sweeper: Live blobs:" UINT32_FORMAT "/Free code cache:" SIZE_FORMAT " bytes, restarting compiler", + CodeCache::nof_blobs(), CodeCache::unallocated_capacity()); + } + if (LogCompilation && (xtty != NULL)) { + ttyLocker ttyl; + xtty->begin_elem("restart_compiler live_blobs='" UINT32_FORMAT "' free_code_cache='" SIZE_FORMAT "'", + CodeCache::nof_blobs(), CodeCache::unallocated_capacity()); + xtty->stamp(); + xtty->end_elem(); + } + } + } + } } @@ -137,12 +176,12 @@ if (nm->is_marked_for_reclamation()) { assert(!nm->is_locked_by_vm(), "must not flush locked nmethods"); if (PrintMethodFlushing && Verbose) { - tty->print_cr("### Nmethod 0x%x (marked for reclamation) being flushed", nm); + tty->print_cr("### Nmethod %3d/" PTR_FORMAT " (marked for reclamation) being flushed", nm->compile_id(), nm); } nm->flush(); } else { if (PrintMethodFlushing && Verbose) { - tty->print_cr("### Nmethod 0x%x (zombie) being marked for reclamation", nm); + tty->print_cr("### Nmethod %3d/" PTR_FORMAT " (zombie) being marked for reclamation", nm->compile_id(), nm); } nm->mark_for_reclamation(); _rescan = true; @@ -152,7 +191,7 @@ // stack we can safely convert it to a zombie method if (nm->can_not_entrant_be_converted()) { if (PrintMethodFlushing && Verbose) { - tty->print_cr("### Nmethod 0x%x (not entrant) being made zombie", nm); + tty->print_cr("### Nmethod %3d/" PTR_FORMAT " (not entrant) being made zombie", nm->compile_id(), nm); } nm->make_zombie(); _rescan = true; @@ -167,7 +206,7 @@ } else if (nm->is_unloaded()) { // Unloaded code, just make it a zombie if (PrintMethodFlushing && Verbose) - tty->print_cr("### Nmethod 0x%x (unloaded) being made zombie", nm); + tty->print_cr("### Nmethod %3d/" PTR_FORMAT " (unloaded) being made zombie", nm->compile_id(), nm); if (nm->is_osr_method()) { // No inline caches will ever point to osr methods, so we can just remove it nm->flush(); @@ -177,7 +216,167 @@ } } else { assert(nm->is_alive(), "should be alive"); + + if (UseCodeCacheFlushing) { + if ((nm->method()->code() != nm) && !(nm->is_locked_by_vm()) && !(nm->is_osr_method()) && + (_traversals > _was_full_traversal+2) && (((uint)nm->compile_id()) < _highest_marked) && + CodeCache::needs_flushing()) { + // This method has not been called since the forced cleanup happened + nm->make_not_entrant(); + } + } + // Clean-up all inline caches that points to zombie/non-reentrant methods nm->cleanup_inline_caches(); } } + +// Code cache unloading: when compilers notice the code cache is getting full, +// they will call a vm op that comes here. This code attempts to speculatively +// unload the oldest half of the nmethods (based on the compile job id) by +// saving the old code in a list in the CodeCache. Then +// execution resumes. If a method so marked is not called by the second +// safepoint from the current one, the nmethod will be marked non-entrant and +// got rid of by normal sweeping. If the method is called, the methodOop's +// _code field is restored and the methodOop/nmethod +// go back to their normal state. +void NMethodSweeper::handle_full_code_cache(bool is_full) { + // Only the first one to notice can advise us to start early cleaning + if (!is_full){ + jint old = Atomic::cmpxchg( 1, &_advise_to_sweep, 0 ); + if (old != 0) { + return; + } + } + + if (is_full) { + // Since code cache is full, immediately stop new compiles + bool did_set = CompileBroker::set_should_compile_new_jobs(CompileBroker::stop_compilation); + if (!did_set) { + // only the first to notice can start the cleaning, + // others will go back and block + return; + } + set_was_full(true); + + // If we run out within MinCodeCacheFlushingInterval of the last unload time, give up + jlong now = os::javaTimeMillis(); + jlong max_interval = (jlong)MinCodeCacheFlushingInterval * (jlong)1000; + jlong curr_interval = now - _last_was_full; + if (curr_interval < max_interval) { + _rescan = true; + if (PrintMethodFlushing) { + tty->print_cr("### handle full too often, turning off compiler"); + } + if (LogCompilation && (xtty != NULL)) { + ttyLocker ttyl; + xtty->begin_elem("disable_compiler flushing_interval='" UINT64_FORMAT "' live_blobs='" UINT32_FORMAT "' free_code_cache='" SIZE_FORMAT "'", + curr_interval/1000, CodeCache::nof_blobs(), CodeCache::unallocated_capacity()); + xtty->stamp(); + xtty->end_elem(); + } + return; + } + } + + VM_HandleFullCodeCache op(is_full); + VMThread::execute(&op); + + // rescan again as soon as possible + _rescan = true; +} + +void NMethodSweeper::speculative_disconnect_nmethods(bool is_full) { + // If there was a race in detecting full code cache, only run + // one vm op for it or keep the compiler shut off + + debug_only(jlong start = os::javaTimeMillis();) + + if ((!was_full()) && (is_full)) { + if (!CodeCache::needs_flushing()) { + if (PrintMethodFlushing) { + tty->print_cr("### sweeper: Live blobs:" UINT32_FORMAT "/Free code cache:" SIZE_FORMAT " bytes, restarting compiler", + CodeCache::nof_blobs(), CodeCache::unallocated_capacity()); + } + if (LogCompilation && (xtty != NULL)) { + ttyLocker ttyl; + xtty->begin_elem("restart_compiler live_blobs='" UINT32_FORMAT "' free_code_cache='" SIZE_FORMAT "'", + CodeCache::nof_blobs(), CodeCache::unallocated_capacity()); + xtty->stamp(); + xtty->end_elem(); + } + CompileBroker::set_should_compile_new_jobs(CompileBroker::run_compilation); + return; + } + } + + // Traverse the code cache trying to dump the oldest nmethods + uint curr_max_comp_id = CompileBroker::get_compilation_id(); + uint flush_target = ((curr_max_comp_id - _highest_marked) >> 1) + _highest_marked; + if (PrintMethodFlushing && Verbose) { + tty->print_cr("### Cleaning code cache: Live blobs:" UINT32_FORMAT "/Free code cache:" SIZE_FORMAT " bytes", + CodeCache::nof_blobs(), CodeCache::unallocated_capacity()); + } + if (LogCompilation && (xtty != NULL)) { + ttyLocker ttyl; + xtty->begin_elem("start_cleaning_code_cache live_blobs='" UINT32_FORMAT "' free_code_cache='" SIZE_FORMAT "'", + CodeCache::nof_blobs(), CodeCache::unallocated_capacity()); + xtty->stamp(); + xtty->end_elem(); + } + + nmethod* nm = CodeCache::alive_nmethod(CodeCache::first()); + jint disconnected = 0; + jint made_not_entrant = 0; + while ((nm != NULL)){ + uint curr_comp_id = nm->compile_id(); + + // OSR methods cannot be flushed like this. Also, don't flush native methods + // since they are part of the JDK in most cases + if (nm->is_in_use() && (!nm->is_osr_method()) && (!nm->is_locked_by_vm()) && + (!nm->is_native_method()) && ((curr_comp_id < flush_target))) { + + if ((nm->method()->code() == nm)) { + // This method has not been previously considered for + // unloading or it was restored already + CodeCache::speculatively_disconnect(nm); + disconnected++; + } else if (nm->is_speculatively_disconnected()) { + // This method was previously considered for preemptive unloading and was not called since then + nm->method()->invocation_counter()->decay(); + nm->method()->backedge_counter()->decay(); + nm->make_not_entrant(); + made_not_entrant++; + } + + if (curr_comp_id > _highest_marked) { + _highest_marked = curr_comp_id; + } + } + nm = CodeCache::alive_nmethod(CodeCache::next(nm)); + } + + if (LogCompilation && (xtty != NULL)) { + ttyLocker ttyl; + xtty->begin_elem("stop_cleaning_code_cache disconnected='" UINT32_FORMAT "' made_not_entrant='" UINT32_FORMAT "' live_blobs='" UINT32_FORMAT "' free_code_cache='" SIZE_FORMAT "'", + disconnected, made_not_entrant, CodeCache::nof_blobs(), CodeCache::unallocated_capacity()); + xtty->stamp(); + xtty->end_elem(); + } + + // Shut off compiler. Sweeper will run exiting from this safepoint + // and turn it back on if it clears enough space + if (was_full()) { + _last_was_full = os::javaTimeMillis(); + CompileBroker::set_should_compile_new_jobs(CompileBroker::stop_compilation); + } + + // After two more traversals the sweeper will get rid of unrestored nmethods + _was_full_traversal = _traversals; +#ifdef ASSERT + jlong end = os::javaTimeMillis(); + if(PrintMethodFlushing && Verbose) { + tty->print_cr("### sweeper: unload time: " INT64_FORMAT, end-start); + } +#endif +}
--- a/src/share/vm/runtime/sweeper.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/sweeper.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -38,6 +38,11 @@ static int _locked_seen; // Number of locked nmethods encountered during the scan static int _not_entrant_seen_on_stack; // Number of not entrant nmethod were are still on stack + static bool _was_full; // remember if we did emergency unloading + static jint _advise_to_sweep; // flag to indicate code cache getting full + static jlong _last_was_full; // timestamp of last emergency unloading + static uint _highest_marked; // highest compile id dumped at last emergency unloading + static long _was_full_traversal; // trav number at last emergency unloading static void process_nmethod(nmethod *nm); public: @@ -51,4 +56,10 @@ // changes to false at safepoint so we can never overwrite it with false. _rescan = true; } + + static void handle_full_code_cache(bool is_full); // Called by compilers who fail to allocate + static void speculative_disconnect_nmethods(bool was_full); // Called by vm op to deal with alloc failure + + static void set_was_full(bool state) { _was_full = state; } + static bool was_full() { return _was_full; } };
--- a/src/share/vm/runtime/thread.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/thread.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -884,6 +884,22 @@ vmSymbolHandles::void_method_signature(), CHECK); } +#ifdef KERNEL +static void set_jkernel_boot_classloader_hook(TRAPS) { + klassOop k = SystemDictionary::sun_jkernel_DownloadManager_klass(); + instanceKlassHandle klass (THREAD, k); + + if (k == NULL) { + // sun.jkernel.DownloadManager may not present in the JDK; just return + return; + } + + JavaValue result(T_VOID); + JavaCalls::call_static(&result, klass, vmSymbolHandles::setBootClassLoaderHook_name(), + vmSymbolHandles::void_method_signature(), CHECK); +} +#endif // KERNEL + static void reset_vm_info_property(TRAPS) { // the vm info string ResourceMark rm(THREAD); @@ -957,7 +973,7 @@ return; } - KlassHandle group(this, SystemDictionary::threadGroup_klass()); + KlassHandle group(this, SystemDictionary::ThreadGroup_klass()); Handle threadObj(this, this->threadObj()); JavaCalls::call_special(&result, @@ -975,6 +991,7 @@ // uniquely named instances should derive from this. NamedThread::NamedThread() : Thread() { _name = NULL; + _processed_thread = NULL; } NamedThread::~NamedThread() { @@ -1156,6 +1173,7 @@ _exception_handler_pc = 0; _exception_stack_size = 0; _jvmti_thread_state= NULL; + _should_post_on_exceptions_flag = JNI_FALSE; _jvmti_get_loaded_classes_closure = NULL; _interp_only_mode = 0; _special_runtime_exit_condition = _no_async_condition; @@ -1451,7 +1469,7 @@ // so call ThreadGroup.uncaughtException() KlassHandle recvrKlass(THREAD, threadObj->klass()); CallInfo callinfo; - KlassHandle thread_klass(THREAD, SystemDictionary::thread_klass()); + KlassHandle thread_klass(THREAD, SystemDictionary::Thread_klass()); LinkResolver::resolve_virtual_call(callinfo, threadObj, recvrKlass, thread_klass, vmSymbolHandles::dispatchUncaughtException_name(), vmSymbolHandles::throwable_void_signature(), @@ -1467,7 +1485,7 @@ uncaught_exception, THREAD); } else { - KlassHandle thread_group(THREAD, SystemDictionary::threadGroup_klass()); + KlassHandle thread_group(THREAD, SystemDictionary::ThreadGroup_klass()); JavaValue result(T_VOID); JavaCalls::call_virtual(&result, group, thread_group, @@ -1488,7 +1506,7 @@ while (java_lang_Thread::threadGroup(threadObj()) != NULL && (count-- > 0)) { EXCEPTION_MARK; JavaValue result(T_VOID); - KlassHandle thread_klass(THREAD, SystemDictionary::thread_klass()); + KlassHandle thread_klass(THREAD, SystemDictionary::Thread_klass()); JavaCalls::call_virtual(&result, threadObj, thread_klass, vmSymbolHandles::exit_method_name(), @@ -1726,7 +1744,7 @@ // Check for pending async. exception if (_pending_async_exception != NULL) { // Only overwrite an already pending exception, if it is not a threadDeath. - if (!has_pending_exception() || !pending_exception()->is_a(SystemDictionary::threaddeath_klass())) { + if (!has_pending_exception() || !pending_exception()->is_a(SystemDictionary::ThreadDeath_klass())) { // We cannot call Exceptions::_throw(...) here because we cannot block set_pending_exception(_pending_async_exception, __FILE__, __LINE__); @@ -1835,14 +1853,14 @@ if (is_Compiler_thread()) return; // This is a change from JDK 1.1, but JDK 1.2 will also do it: - if (java_throwable->is_a(SystemDictionary::threaddeath_klass())) { + if (java_throwable->is_a(SystemDictionary::ThreadDeath_klass())) { java_lang_Thread::set_stillborn(threadObj()); } { // Actually throw the Throwable against the target Thread - however // only if there is no thread death exception installed already. - if (_pending_async_exception == NULL || !_pending_async_exception->is_a(SystemDictionary::threaddeath_klass())) { + if (_pending_async_exception == NULL || !_pending_async_exception->is_a(SystemDictionary::ThreadDeath_klass())) { // If the topmost frame is a runtime stub, then we are calling into // OptoRuntime from compiled code. Some runtime stubs (new, monitor_exit..) // must deoptimize the caller before continuing, as the compiled exception handler table @@ -2317,11 +2335,31 @@ frames_do(frame_gc_prologue); } +// If the caller is a NamedThread, then remember, in the current scope, +// the given JavaThread in its _processed_thread field. +class RememberProcessedThread: public StackObj { + NamedThread* _cur_thr; +public: + RememberProcessedThread(JavaThread* jthr) { + Thread* thread = Thread::current(); + if (thread->is_Named_thread()) { + _cur_thr = (NamedThread *)thread; + _cur_thr->set_processed_thread(jthr); + } else { + _cur_thr = NULL; + } + } + + ~RememberProcessedThread() { + if (_cur_thr) { + _cur_thr->set_processed_thread(NULL); + } + } +}; void JavaThread::oops_do(OopClosure* f, CodeBlobClosure* cf) { - // Flush deferred store-barriers, if any, associated with - // initializing stores done by this JavaThread in the current epoch. - Universe::heap()->flush_deferred_store_barrier(this); + // Verify that the deferred card marks have been flushed. + assert(deferred_card_mark().is_empty(), "Should be empty during GC"); // The ThreadProfiler oops_do is done from FlatProfiler::oops_do // since there may be more than one thread using each ThreadProfiler. @@ -2333,6 +2371,8 @@ (has_last_Java_frame() && java_call_counter() > 0), "wrong java_sp info!"); if (has_last_Java_frame()) { + // Record JavaThread to GC thread + RememberProcessedThread rpt(this); // Traverse the privileged stack if (_privileged_stack_top != NULL) { @@ -3055,6 +3095,12 @@ warning("java.lang.ArithmeticException has not been initialized"); warning("java.lang.StackOverflowError has not been initialized"); } + + if (EnableInvokeDynamic) { + // JSR 292: An intialized java.dyn.InvokeDynamic is required in + // the compiler. + initialize_class(vmSymbolHandles::java_dyn_InvokeDynamic(), CHECK_0); + } } // See : bugid 4211085. @@ -3102,6 +3148,12 @@ vm_exit_during_initialization(Handle(THREAD, PENDING_EXCEPTION)); } +#ifdef KERNEL + if (JDK_Version::is_gte_jdk17x_version()) { + set_jkernel_boot_classloader_hook(THREAD); + } +#endif // KERNEL + #ifndef SERIALGC // Support for ConcurrentMarkSweep. This should be cleaned up // and better encapsulated. The ugly nested if test would go away
--- a/src/share/vm/runtime/thread.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/thread.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -48,7 +48,12 @@ // Class hierarchy // - Thread -// - VMThread +// - NamedThread +// - VMThread +// - ConcurrentGCThread +// - WorkerThread +// - GangWorker +// - GCTaskThread // - JavaThread // - WatcherThread @@ -249,6 +254,7 @@ virtual bool is_GC_task_thread() const { return false; } virtual bool is_Watcher_thread() const { return false; } virtual bool is_ConcurrentGC_thread() const { return false; } + virtual bool is_Named_thread() const { return false; } virtual char* name() const { return (char*)"Unknown thread"; } @@ -568,12 +574,18 @@ }; private: char* _name; + // log JavaThread being processed by oops_do + JavaThread* _processed_thread; + public: NamedThread(); ~NamedThread(); // May only be called once per thread. void set_name(const char* format, ...); + virtual bool is_Named_thread() const { return true; } virtual char* name() const { return _name == NULL ? (char*)"Unknown Thread" : _name; } + JavaThread *processed_thread() { return _processed_thread; } + void set_processed_thread(JavaThread *thread) { _processed_thread = thread; } }; // Worker threads are named and have an id of an assigned work. @@ -760,6 +772,7 @@ volatile address _exception_pc; // PC where exception happened volatile address _exception_handler_pc; // PC for handler of exception volatile int _exception_stack_size; // Size of frame where exception happened + volatile int _is_method_handle_exception; // True if the current exception PC is at a MethodHandle call. // support for compilation bool _is_compiling; // is true if a compilation is active inthis thread (one compilation per thread possible) @@ -1095,11 +1108,13 @@ int exception_stack_size() const { return _exception_stack_size; } address exception_pc() const { return _exception_pc; } address exception_handler_pc() const { return _exception_handler_pc; } + int is_method_handle_exception() const { return _is_method_handle_exception; } void set_exception_oop(oop o) { _exception_oop = o; } void set_exception_pc(address a) { _exception_pc = a; } void set_exception_handler_pc(address a) { _exception_handler_pc = a; } void set_exception_stack_size(int size) { _exception_stack_size = size; } + void set_is_method_handle_exception(int value) { _is_method_handle_exception = value; } // Stack overflow support inline size_t stack_available(address cur_sp); @@ -1173,10 +1188,14 @@ static ByteSize exception_pc_offset() { return byte_offset_of(JavaThread, _exception_pc ); } static ByteSize exception_handler_pc_offset() { return byte_offset_of(JavaThread, _exception_handler_pc); } static ByteSize exception_stack_size_offset() { return byte_offset_of(JavaThread, _exception_stack_size); } + static ByteSize is_method_handle_exception_offset() { return byte_offset_of(JavaThread, _is_method_handle_exception); } static ByteSize stack_guard_state_offset() { return byte_offset_of(JavaThread, _stack_guard_state ); } static ByteSize suspend_flags_offset() { return byte_offset_of(JavaThread, _suspend_flags ); } static ByteSize do_not_unlock_if_synchronized_offset() { return byte_offset_of(JavaThread, _do_not_unlock_if_synchronized); } + static ByteSize should_post_on_exceptions_flag_offset() { + return byte_offset_of(JavaThread, _should_post_on_exceptions_flag); + } #ifndef SERIALGC static ByteSize satb_mark_queue_offset() { return byte_offset_of(JavaThread, _satb_mark_queue); } @@ -1416,6 +1435,16 @@ void increment_interp_only_mode() { ++_interp_only_mode; } void decrement_interp_only_mode() { --_interp_only_mode; } + // support for cached flag that indicates whether exceptions need to be posted for this thread + // if this is false, we can avoid deoptimizing when events are thrown + // this gets set to reflect whether jvmtiExport::post_exception_throw would actually do anything + private: + int _should_post_on_exceptions_flag; + + public: + int should_post_on_exceptions_flag() { return _should_post_on_exceptions_flag; } + void set_should_post_on_exceptions_flag(int val) { _should_post_on_exceptions_flag = val; } + private: ThreadStatistics *_thread_stat;
--- a/src/share/vm/runtime/vframe.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/vframe.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -124,7 +124,7 @@ static void print_locked_object_class_name(outputStream* st, Handle obj, const char* lock_state) { if (obj.not_null()) { st->print("\t- %s <" INTPTR_FORMAT "> ", lock_state, (address)obj()); - if (obj->klass() == SystemDictionary::class_klass()) { + if (obj->klass() == SystemDictionary::Class_klass()) { klassOop target_klass = java_lang_Class::as_klassOop(obj()); st->print_cr("(a java.lang.Class for %s)", instanceKlass::cast(target_klass)->external_name()); } else { @@ -430,8 +430,10 @@ // This is Method.invoke() -- skip it } else if (use_new_reflection && Klass::cast(method()->method_holder()) - ->is_subclass_of(SystemDictionary::reflect_method_accessor_klass())) { + ->is_subclass_of(SystemDictionary::reflect_MethodAccessorImpl_klass())) { // This is an auxilary frame -- skip it + } else if (method()->is_method_handle_adapter()) { + // This is an internal adapter frame from the MethodHandleCompiler -- skip it } else { // This is non-excluded frame, we need to count it against the depth if (depth-- <= 0) { @@ -490,8 +492,8 @@ void vframeStreamCommon::skip_reflection_related_frames() { while (!at_end() && (JDK_Version::is_gte_jdk14x_version() && UseNewReflection && - (Klass::cast(method()->method_holder())->is_subclass_of(SystemDictionary::reflect_method_accessor_klass()) || - Klass::cast(method()->method_holder())->is_subclass_of(SystemDictionary::reflect_constructor_accessor_klass())))) { + (Klass::cast(method()->method_holder())->is_subclass_of(SystemDictionary::reflect_MethodAccessorImpl_klass()) || + Klass::cast(method()->method_holder())->is_subclass_of(SystemDictionary::reflect_ConstructorAccessorImpl_klass())))) { next(); } }
--- a/src/share/vm/runtime/vframeArray.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/vframeArray.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -186,7 +186,7 @@ int popframe_preserved_args_size_in_bytes = 0; int popframe_preserved_args_size_in_words = 0; if (is_top_frame) { - JvmtiThreadState *state = thread->jvmti_thread_state(); + JvmtiThreadState *state = thread->jvmti_thread_state(); if (JvmtiExport::can_pop_frame() && (thread->has_pending_popframe() || thread->popframe_forcing_deopt_reexecution())) { if (thread->has_pending_popframe()) { @@ -381,7 +381,6 @@ RegisterMap map(thread); vframe* f = vframe::new_vframe(iframe(), &map, thread); f->print(); - iframe()->interpreter_frame_print_on(tty); tty->print_cr("locals size %d", locals()->size()); tty->print_cr("expression size %d", expressions()->size()); @@ -582,7 +581,7 @@ } void vframeArrayElement::print(outputStream* st) { - st->print_cr(" - interpreter_frame -> sp: ", INTPTR_FORMAT, iframe()->sp()); + st->print_cr(" - interpreter_frame -> sp: " INTPTR_FORMAT, iframe()->sp()); } void vframeArray::print_value_on(outputStream* st) const {
--- a/src/share/vm/runtime/vmStructs.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/vmStructs.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -309,6 +309,7 @@ nonstatic_field(CollectedHeap, _reserved, MemRegion) \ nonstatic_field(SharedHeap, _perm_gen, PermGen*) \ nonstatic_field(CollectedHeap, _barrier_set, BarrierSet*) \ + nonstatic_field(CollectedHeap, _defer_initial_card_mark, bool) \ nonstatic_field(CollectedHeap, _is_gc_active, bool) \ nonstatic_field(CompactibleSpace, _compaction_top, HeapWord*) \ nonstatic_field(CompactibleSpace, _first_dead, HeapWord*) \ @@ -455,40 +456,38 @@ static_field(SystemDictionary, _shared_dictionary, Dictionary*) \ static_field(SystemDictionary, _system_loader_lock_obj, oop) \ static_field(SystemDictionary, _loader_constraints, LoaderConstraintTable*) \ - static_field(SystemDictionary, WK_KLASS(object_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(string_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(class_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(cloneable_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(classloader_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(serializable_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(system_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(throwable_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(threaddeath_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(error_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(exception_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(runtime_exception_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(classNotFoundException_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(noClassDefFoundError_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(linkageError_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(Object_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(String_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(Class_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(Cloneable_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(ClassLoader_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(Serializable_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(System_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(Throwable_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(ThreadDeath_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(Error_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(Exception_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(RuntimeException_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(ClassNotFoundException_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(NoClassDefFoundError_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(LinkageError_klass), klassOop) \ static_field(SystemDictionary, WK_KLASS(ClassCastException_klass), klassOop) \ static_field(SystemDictionary, WK_KLASS(ArrayStoreException_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(virtualMachineError_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(VirtualMachineError_klass), klassOop) \ static_field(SystemDictionary, WK_KLASS(OutOfMemoryError_klass), klassOop) \ static_field(SystemDictionary, WK_KLASS(StackOverflowError_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(protectionDomain_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(ProtectionDomain_klass), klassOop) \ static_field(SystemDictionary, WK_KLASS(AccessControlContext_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(reference_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(soft_reference_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(weak_reference_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(final_reference_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(phantom_reference_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(finalizer_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(thread_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(threadGroup_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(properties_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(stringBuffer_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(vector_klass), klassOop) \ - static_field(SystemDictionary, WK_KLASS(hashtable_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(Reference_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(SoftReference_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(WeakReference_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(FinalReference_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(PhantomReference_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(Finalizer_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(Thread_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(ThreadGroup_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(Properties_klass), klassOop) \ + static_field(SystemDictionary, WK_KLASS(StringBuffer_klass), klassOop) \ static_field(SystemDictionary, _box_klasses[0], klassOop) \ static_field(SystemDictionary, _java_system_loader, oop) \ \ @@ -666,6 +665,7 @@ nonstatic_field(Thread, _current_pending_monitor_is_from_java, bool) \ nonstatic_field(Thread, _current_waiting_monitor, ObjectMonitor*) \ nonstatic_field(NamedThread, _name, char*) \ + nonstatic_field(NamedThread, _processed_thread, JavaThread*) \ nonstatic_field(JavaThread, _next, JavaThread*) \ nonstatic_field(JavaThread, _threadObj, oop) \ nonstatic_field(JavaThread, _anchor, JavaFrameAnchor) \
--- a/src/share/vm/runtime/vmThread.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/vmThread.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 1998-2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1998-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -204,8 +204,8 @@ } -VMThread::VMThread() : Thread() { - // nothing to do +VMThread::VMThread() : NamedThread() { + set_name("VM Thread"); } void VMThread::destroy() { @@ -426,11 +426,6 @@ // follow that also require a safepoint if (_cur_vm_operation->evaluate_at_safepoint()) { - if (PrintGCApplicationConcurrentTime) { - gclog_or_tty->print_cr("Application time: %3.7f seconds", - RuntimeService::last_application_time_sec()); - } - _vm_queue->set_drain_list(safepoint_ops); // ensure ops can be scanned SafepointSynchronize::begin(); @@ -477,12 +472,6 @@ // Complete safepoint synchronization SafepointSynchronize::end(); - if (PrintGCApplicationStoppedTime) { - gclog_or_tty->print_cr("Total time for which application threads " - "were stopped: %3.7f seconds", - RuntimeService::last_safepoint_time_sec()); - } - } else { // not a safepoint operation if (TraceLongCompiles) { elapsedTimer t;
--- a/src/share/vm/runtime/vmThread.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/vmThread.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -83,7 +83,7 @@ // like scavenge, garbage_collect etc. // -class VMThread: public Thread { +class VMThread: public NamedThread { private: static ThreadPriority _current_priority; @@ -101,8 +101,6 @@ bool is_VM_thread() const { return true; } bool is_GC_thread() const { return true; } - char* name() const { return (char*)"VM Thread"; } - // The ever running loop for the VMThread void loop();
--- a/src/share/vm/runtime/vm_operations.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/vm_operations.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -151,6 +151,10 @@ #endif // !PRODUCT +void VM_HandleFullCodeCache::doit() { + NMethodSweeper::speculative_disconnect_nmethods(_is_full); +} + void VM_Verify::doit() { Universe::verify(); }
--- a/src/share/vm/runtime/vm_operations.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/runtime/vm_operations.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -41,6 +41,7 @@ template(DeoptimizeFrame) \ template(DeoptimizeAll) \ template(ZombieAll) \ + template(HandleFullCodeCache) \ template(Verify) \ template(PrintJNI) \ template(HeapDumper) \ @@ -241,6 +242,16 @@ bool allow_nested_vm_operations() const { return true; } }; +class VM_HandleFullCodeCache: public VM_Operation { + private: + bool _is_full; + public: + VM_HandleFullCodeCache(bool is_full) { _is_full = is_full; } + VMOp_Type type() const { return VMOp_HandleFullCodeCache; } + void doit(); + bool allow_nested_vm_operations() const { return true; } +}; + #ifndef PRODUCT class VM_DeoptimizeAll: public VM_Operation { private:
--- a/src/share/vm/services/attachListener.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/services/attachListener.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -207,7 +207,7 @@ int tmp; int n = sscanf(arg1, "%d", &tmp); if (n != 1) { - out->print_cr("flag value has to be boolean (1 or 0)"); + out->print_cr("flag value must be a boolean (1 or 0)"); return JNI_ERR; } value = (tmp != 0); @@ -226,11 +226,11 @@ if ((arg1 = op->arg(1)) != NULL) { int n = sscanf(arg1, INTX_FORMAT, &value); if (n != 1) { - out->print_cr("flag value has to be integer"); + out->print_cr("flag value must be an integer"); return JNI_ERR; } } - bool res = CommandLineFlags::intxAtPut((char*)name, &value, ATTACH_ON_DEMAND); + bool res = CommandLineFlags::intxAtPut((char*)name, &value, ATTACH_ON_DEMAND); if (! res) { out->print_cr("setting flag %s failed", name); } @@ -245,11 +245,30 @@ if ((arg1 = op->arg(1)) != NULL) { int n = sscanf(arg1, UINTX_FORMAT, &value); if (n != 1) { - out->print_cr("flag value has to be integer"); + out->print_cr("flag value must be an unsigned integer"); return JNI_ERR; } } - bool res = CommandLineFlags::uintxAtPut((char*)name, &value, ATTACH_ON_DEMAND); + bool res = CommandLineFlags::uintxAtPut((char*)name, &value, ATTACH_ON_DEMAND); + if (! res) { + out->print_cr("setting flag %s failed", name); + } + + return res? JNI_OK : JNI_ERR; +} + +// set a uint64_t global flag using value from AttachOperation +static jint set_uint64_t_flag(const char* name, AttachOperation* op, outputStream* out) { + uint64_t value; + const char* arg1; + if ((arg1 = op->arg(1)) != NULL) { + int n = sscanf(arg1, UINT64_FORMAT, &value); + if (n != 1) { + out->print_cr("flag value must be an unsigned 64-bit integer"); + return JNI_ERR; + } + } + bool res = CommandLineFlags::uint64_tAtPut((char*)name, &value, ATTACH_ON_DEMAND); if (! res) { out->print_cr("setting flag %s failed", name); } @@ -261,10 +280,10 @@ static jint set_ccstr_flag(const char* name, AttachOperation* op, outputStream* out) { const char* value; if ((value = op->arg(1)) == NULL) { - out->print_cr("flag value has to be a string"); + out->print_cr("flag value must be a string"); return JNI_ERR; } - bool res = CommandLineFlags::ccstrAtPut((char*)name, &value, ATTACH_ON_DEMAND); + bool res = CommandLineFlags::ccstrAtPut((char*)name, &value, ATTACH_ON_DEMAND); if (res) { FREE_C_HEAP_ARRAY(char, value); } else { @@ -291,6 +310,8 @@ return set_intx_flag(name, op, out); } else if (f->is_uintx()) { return set_uintx_flag(name, op, out); + } else if (f->is_uint64_t()) { + return set_uint64_t_flag(name, op, out); } else if (f->is_ccstr()) { return set_ccstr_flag(name, op, out); } else { @@ -416,7 +437,7 @@ string, CHECK); - KlassHandle group(THREAD, SystemDictionary::threadGroup_klass()); + KlassHandle group(THREAD, SystemDictionary::ThreadGroup_klass()); JavaCalls::call_special(&result, thread_group, group,
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/share/vm/services/g1MemoryPool.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,162 @@ +/* + * Copyright (c) 2007 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +# include "incls/_precompiled.incl" +# include "incls/_g1MemoryPool.cpp.incl" + +G1MemoryPoolSuper::G1MemoryPoolSuper(G1CollectedHeap* g1h, + const char* name, + size_t init_size, + size_t max_size, + bool support_usage_threshold) : + _g1h(g1h), CollectedMemoryPool(name, + MemoryPool::Heap, + init_size, + max_size, + support_usage_threshold) { + assert(UseG1GC, "sanity"); +} + +// See the comment at the top of g1MemoryPool.hpp +size_t G1MemoryPoolSuper::eden_space_committed(G1CollectedHeap* g1h) { + return MAX2(eden_space_used(g1h), (size_t) HeapRegion::GrainBytes); +} + +// See the comment at the top of g1MemoryPool.hpp +size_t G1MemoryPoolSuper::eden_space_used(G1CollectedHeap* g1h) { + size_t young_list_length = g1h->young_list_length(); + size_t eden_used = young_list_length * HeapRegion::GrainBytes; + size_t survivor_used = survivor_space_used(g1h); + eden_used = subtract_up_to_zero(eden_used, survivor_used); + return eden_used; +} + +// See the comment at the top of g1MemoryPool.hpp +size_t G1MemoryPoolSuper::eden_space_max(G1CollectedHeap* g1h) { + // This should ensure that it returns a value no smaller than the + // region size. Currently, eden_space_committed() guarantees that. + return eden_space_committed(g1h); +} + +// See the comment at the top of g1MemoryPool.hpp +size_t G1MemoryPoolSuper::survivor_space_committed(G1CollectedHeap* g1h) { + return MAX2(survivor_space_used(g1h), (size_t) HeapRegion::GrainBytes); +} + +// See the comment at the top of g1MemoryPool.hpp +size_t G1MemoryPoolSuper::survivor_space_used(G1CollectedHeap* g1h) { + size_t survivor_num = g1h->g1_policy()->recorded_survivor_regions(); + size_t survivor_used = survivor_num * HeapRegion::GrainBytes; + return survivor_used; +} + +// See the comment at the top of g1MemoryPool.hpp +size_t G1MemoryPoolSuper::survivor_space_max(G1CollectedHeap* g1h) { + // This should ensure that it returns a value no smaller than the + // region size. Currently, survivor_space_committed() guarantees that. + return survivor_space_committed(g1h); +} + +// See the comment at the top of g1MemoryPool.hpp +size_t G1MemoryPoolSuper::old_space_committed(G1CollectedHeap* g1h) { + size_t committed = overall_committed(g1h); + size_t eden_committed = eden_space_committed(g1h); + size_t survivor_committed = survivor_space_committed(g1h); + committed = subtract_up_to_zero(committed, eden_committed); + committed = subtract_up_to_zero(committed, survivor_committed); + committed = MAX2(committed, (size_t) HeapRegion::GrainBytes); + return committed; +} + +// See the comment at the top of g1MemoryPool.hpp +size_t G1MemoryPoolSuper::old_space_used(G1CollectedHeap* g1h) { + size_t used = overall_used(g1h); + size_t eden_used = eden_space_used(g1h); + size_t survivor_used = survivor_space_used(g1h); + used = subtract_up_to_zero(used, eden_used); + used = subtract_up_to_zero(used, survivor_used); + return used; +} + +// See the comment at the top of g1MemoryPool.hpp +size_t G1MemoryPoolSuper::old_space_max(G1CollectedHeap* g1h) { + size_t max = overall_max(g1h); + size_t eden_max = eden_space_max(g1h); + size_t survivor_max = survivor_space_max(g1h); + max = subtract_up_to_zero(max, eden_max); + max = subtract_up_to_zero(max, survivor_max); + max = MAX2(max, (size_t) HeapRegion::GrainBytes); + return max; +} + +G1EdenPool::G1EdenPool(G1CollectedHeap* g1h) : + G1MemoryPoolSuper(g1h, + "G1 Eden", + eden_space_committed(g1h), /* init_size */ + eden_space_max(g1h), /* max_size */ + false /* support_usage_threshold */) { +} + +MemoryUsage G1EdenPool::get_memory_usage() { + size_t initial_sz = initial_size(); + size_t max_sz = max_size(); + size_t used = used_in_bytes(); + size_t committed = eden_space_committed(_g1h); + + return MemoryUsage(initial_sz, used, committed, max_sz); +} + +G1SurvivorPool::G1SurvivorPool(G1CollectedHeap* g1h) : + G1MemoryPoolSuper(g1h, + "G1 Survivor", + survivor_space_committed(g1h), /* init_size */ + survivor_space_max(g1h), /* max_size */ + false /* support_usage_threshold */) { +} + +MemoryUsage G1SurvivorPool::get_memory_usage() { + size_t initial_sz = initial_size(); + size_t max_sz = max_size(); + size_t used = used_in_bytes(); + size_t committed = survivor_space_committed(_g1h); + + return MemoryUsage(initial_sz, used, committed, max_sz); +} + +G1OldGenPool::G1OldGenPool(G1CollectedHeap* g1h) : + G1MemoryPoolSuper(g1h, + "G1 Old Gen", + old_space_committed(g1h), /* init_size */ + old_space_max(g1h), /* max_size */ + true /* support_usage_threshold */) { +} + +MemoryUsage G1OldGenPool::get_memory_usage() { + size_t initial_sz = initial_size(); + size_t max_sz = max_size(); + size_t used = used_in_bytes(); + size_t committed = old_space_committed(_g1h); + + return MemoryUsage(initial_sz, used, committed, max_sz); +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/share/vm/services/g1MemoryPool.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,197 @@ +/* + * Copyright (c) 2007 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +class G1CollectedHeap; + +// This file contains the three classes that represent the memory +// pools of the G1 spaces: G1EdenPool, G1SurvivorPool, and +// G1OldGenPool. In G1, unlike our other GCs, we do not have a +// physical space for each of those spaces. Instead, we allocate +// regions for all three spaces out of a single pool of regions (that +// pool basically covers the entire heap). As a result, the eden, +// survivor, and old gen are considered logical spaces in G1, as each +// is a set of non-contiguous regions. This is also reflected in the +// way we map them to memory pools here. The easiest way to have done +// this would have been to map the entire G1 heap to a single memory +// pool. However, it's helpful to show how large the eden and survivor +// get, as this does affect the performance and behavior of G1. Which +// is why we introduce the three memory pools implemented here. +// +// The above approach inroduces a couple of challenging issues in the +// implementation of the three memory pools: +// +// 1) The used space calculation for a pool is not necessarily +// independent of the others. We can easily get from G1 the overall +// used space in the entire heap, the number of regions in the young +// generation (includes both eden and survivors), and the number of +// survivor regions. So, from that we calculate: +// +// survivor_used = survivor_num * region_size +// eden_used = young_region_num * region_size - survivor_used +// old_gen_used = overall_used - eden_used - survivor_used +// +// Note that survivor_used and eden_used are upper bounds. To get the +// actual value we would have to iterate over the regions and add up +// ->used(). But that'd be expensive. So, we'll accept some lack of +// accuracy for those two. But, we have to be careful when calculating +// old_gen_used, in case we subtract from overall_used more then the +// actual number and our result goes negative. +// +// 2) Calculating the used space is straightforward, as described +// above. However, how do we calculate the committed space, given that +// we allocate space for the eden, survivor, and old gen out of the +// same pool of regions? One way to do this is to use the used value +// as also the committed value for the eden and survivor spaces and +// then calculate the old gen committed space as follows: +// +// old_gen_committed = overall_committed - eden_committed - survivor_committed +// +// Maybe a better way to do that would be to calculate used for eden +// and survivor as a sum of ->used() over their regions and then +// calculate committed as region_num * region_size (i.e., what we use +// to calculate the used space now). This is something to consider +// in the future. +// +// 3) Another decision that is again not straightforward is what is +// the max size that each memory pool can grow to. Right now, we set +// that the committed size for the eden and the survivors and +// calculate the old gen max as follows (basically, it's a similar +// pattern to what we use for the committed space, as described +// above): +// +// old_gen_max = overall_max - eden_max - survivor_max +// +// 4) Now, there is a very subtle issue with all the above. The +// framework will call get_memory_usage() on the three pools +// asynchronously. As a result, each call might get a different value +// for, say, survivor_num which will yield inconsistent values for +// eden_used, survivor_used, and old_gen_used (as survivor_num is used +// in the calculation of all three). This would normally be +// ok. However, it's possible that this might cause the sum of +// eden_used, survivor_used, and old_gen_used to go over the max heap +// size and this seems to sometimes cause JConsole (and maybe other +// clients) to get confused. There's not a really an easy / clean +// solution to this problem, due to the asynchrounous nature of the +// framework. + + +// This class is shared by the three G1 memory pool classes +// (G1EdenPool, G1SurvivorPool, G1OldGenPool). Given that the way we +// calculate used / committed bytes for these three pools is related +// (see comment above), we put the calculations in this class so that +// we can easily share them among the subclasses. +class G1MemoryPoolSuper : public CollectedMemoryPool { +private: + // It returns x - y if x > y, 0 otherwise. + // As described in the comment above, some of the inputs to the + // calculations we have to do are obtained concurrently and hence + // may be inconsistent with each other. So, this provides a + // defensive way of performing the subtraction and avoids the value + // going negative (which would mean a very large result, given that + // the parameter are size_t). + static size_t subtract_up_to_zero(size_t x, size_t y) { + if (x > y) { + return x - y; + } else { + return 0; + } + } + +protected: + G1CollectedHeap* _g1h; + + // Would only be called from subclasses. + G1MemoryPoolSuper(G1CollectedHeap* g1h, + const char* name, + size_t init_size, + size_t max_size, + bool support_usage_threshold); + + // The reason why all the code is in static methods is so that it + // can be safely called from the constructors of the subclasses. + + static size_t overall_committed(G1CollectedHeap* g1h) { + return g1h->capacity(); + } + static size_t overall_used(G1CollectedHeap* g1h) { + return g1h->used_unlocked(); + } + static size_t overall_max(G1CollectedHeap* g1h) { + return g1h->g1_reserved_obj_bytes(); + } + + static size_t eden_space_committed(G1CollectedHeap* g1h); + static size_t eden_space_used(G1CollectedHeap* g1h); + static size_t eden_space_max(G1CollectedHeap* g1h); + + static size_t survivor_space_committed(G1CollectedHeap* g1h); + static size_t survivor_space_used(G1CollectedHeap* g1h); + static size_t survivor_space_max(G1CollectedHeap* g1h); + + static size_t old_space_committed(G1CollectedHeap* g1h); + static size_t old_space_used(G1CollectedHeap* g1h); + static size_t old_space_max(G1CollectedHeap* g1h); +}; + +// Memory pool that represents the G1 eden. +class G1EdenPool : public G1MemoryPoolSuper { +public: + G1EdenPool(G1CollectedHeap* g1h); + + size_t used_in_bytes() { + return eden_space_used(_g1h); + } + size_t max_size() const { + return eden_space_max(_g1h); + } + MemoryUsage get_memory_usage(); +}; + +// Memory pool that represents the G1 survivor. +class G1SurvivorPool : public G1MemoryPoolSuper { +public: + G1SurvivorPool(G1CollectedHeap* g1h); + + size_t used_in_bytes() { + return survivor_space_used(_g1h); + } + size_t max_size() const { + return survivor_space_max(_g1h); + } + MemoryUsage get_memory_usage(); +}; + +// Memory pool that represents the G1 old gen. +class G1OldGenPool : public G1MemoryPoolSuper { +public: + G1OldGenPool(G1CollectedHeap* g1h); + + size_t used_in_bytes() { + return old_space_used(_g1h); + } + size_t max_size() const { + return old_space_max(_g1h); + } + MemoryUsage get_memory_usage(); +};
--- a/src/share/vm/services/heapDumper.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/services/heapDumper.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1274,7 +1274,7 @@ if (o->is_klass()) return; // skip classes as these emitted as HPROF_GC_CLASS_DUMP records - if (o->klass() == SystemDictionary::class_klass()) { + if (o->klass() == SystemDictionary::Class_klass()) { if (!java_lang_Class::is_primitive(o)) { return; }
--- a/src/share/vm/services/lowMemoryDetector.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/services/lowMemoryDetector.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -32,7 +32,7 @@ void LowMemoryDetector::initialize() { EXCEPTION_MARK; - instanceKlassHandle klass (THREAD, SystemDictionary::thread_klass()); + instanceKlassHandle klass (THREAD, SystemDictionary::Thread_klass()); instanceHandle thread_oop = klass->allocate_instance_handle(CHECK); const char thread_name[] = "Low Memory Detector";
--- a/src/share/vm/services/management.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/services/management.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -491,7 +491,7 @@ int num_flags = Arguments::num_jvm_flags(); int num_args = Arguments::num_jvm_args(); - instanceKlassHandle ik (THREAD, SystemDictionary::string_klass()); + instanceKlassHandle ik (THREAD, SystemDictionary::String_klass()); objArrayOop r = oopFactory::new_objArray(ik(), num_args + num_flags, CHECK_NULL); objArrayHandle result_h(THREAD, r); @@ -790,7 +790,7 @@ assert(!has_undefined_init_size, "Undefined init size"); assert(!has_undefined_max_size, "Undefined max size"); - MemoryUsage usage((heap ? Arguments::initial_heap_size() : total_init), + MemoryUsage usage((heap ? InitialHeapSize : total_init), total_used, total_committed, (heap ? Universe::heap()->max_capacity() : total_max)); @@ -1321,7 +1321,7 @@ LoadedClassesEnumerator lce(THREAD); // Pass current Thread as parameter int num_classes = lce.num_loaded_classes(); - objArrayOop r = oopFactory::new_objArray(SystemDictionary::class_klass(), num_classes, CHECK_0); + objArrayOop r = oopFactory::new_objArray(SystemDictionary::Class_klass(), num_classes, CHECK_0); objArrayHandle classes_ah(THREAD, r); for (int i = 0; i < num_classes; i++) { @@ -1481,7 +1481,7 @@ // last flag entry is always NULL, so subtract 1 int nFlags = (int) Flag::numFlags - 1; // allocate a temp array - objArrayOop r = oopFactory::new_objArray(SystemDictionary::string_klass(), + objArrayOop r = oopFactory::new_objArray(SystemDictionary::String_klass(), nFlags, CHECK_0); objArrayHandle flags_ah(THREAD, r); int num_entries = 0; @@ -1497,7 +1497,7 @@ if (num_entries < nFlags) { // Return array of right length - objArrayOop res = oopFactory::new_objArray(SystemDictionary::string_klass(), num_entries, CHECK_0); + objArrayOop res = oopFactory::new_objArray(SystemDictionary::String_klass(), num_entries, CHECK_0); for(int i = 0; i < num_entries; i++) { res->obj_at_put(i, flags_ah->obj_at(i)); } @@ -1507,16 +1507,17 @@ return (jobjectArray)JNIHandles::make_local(env, flags_ah()); JVM_END -// utility function used by jmm_GetVMGlobals -void add_global_entry(JNIEnv* env, Handle name, jmmVMGlobal *global, Flag *flag, TRAPS) { +// Utility function used by jmm_GetVMGlobals. Returns false if flag type +// can't be determined, true otherwise. If false is returned, then *global +// will be incomplete and invalid. +bool add_global_entry(JNIEnv* env, Handle name, jmmVMGlobal *global, Flag *flag, TRAPS) { Handle flag_name; if (name() == NULL) { - flag_name = java_lang_String::create_from_str(flag->name, CHECK); + flag_name = java_lang_String::create_from_str(flag->name, CHECK_false); } else { flag_name = name; } global->name = (jstring)JNIHandles::make_local(env, flag_name()); - global->type = JMM_VMGLOBAL_TYPE_UNKNOWN; if (flag->is_bool()) { global->value.z = flag->get_bool() ? JNI_TRUE : JNI_FALSE; @@ -1527,10 +1528,17 @@ } else if (flag->is_uintx()) { global->value.j = (jlong)flag->get_uintx(); global->type = JMM_VMGLOBAL_TYPE_JLONG; + } else if (flag->is_uint64_t()) { + global->value.j = (jlong)flag->get_uint64_t(); + global->type = JMM_VMGLOBAL_TYPE_JLONG; } else if (flag->is_ccstr()) { - Handle str = java_lang_String::create_from_str(flag->get_ccstr(), CHECK); + Handle str = java_lang_String::create_from_str(flag->get_ccstr(), CHECK_false); global->value.l = (jobject)JNIHandles::make_local(env, str()); global->type = JMM_VMGLOBAL_TYPE_JSTRING; + } else { + global->type = JMM_VMGLOBAL_TYPE_UNKNOWN; + assert(false, "Unsupported VMGlobal Type"); + return false; } global->writeable = flag->is_writeable(); @@ -1557,6 +1565,8 @@ default: global->origin = JMM_VMGLOBAL_ORIGIN_OTHER; } + + return true; } // Fill globals array of count length with jmmVMGlobal entries @@ -1583,7 +1593,7 @@ objArrayHandle names_ah(THREAD, ta); // Make sure we have a String array klassOop element_klass = objArrayKlass::cast(names_ah->klass())->element_klass(); - if (element_klass != SystemDictionary::string_klass()) { + if (element_klass != SystemDictionary::String_klass()) { THROW_MSG_(vmSymbols::java_lang_IllegalArgumentException(), "Array element type is not String class", 0); } @@ -1599,8 +1609,8 @@ Handle sh(THREAD, s); char* str = java_lang_String::as_utf8_string(s); Flag* flag = Flag::find_flag(str, strlen(str)); - if (flag != NULL) { - add_global_entry(env, sh, &globals[i], flag, THREAD); + if (flag != NULL && + add_global_entry(env, sh, &globals[i], flag, THREAD)) { num_entries++; } else { globals[i].name = NULL; @@ -1617,8 +1627,8 @@ for (int i = 0; i < nFlags && num_entries < count; i++) { Flag* flag = &Flag::flags[i]; // Exclude the locked (diagnostic, experimental) flags - if (flag->is_unlocked() || flag->is_unlocker()) { - add_global_entry(env, null_h, &globals[num_entries], flag, THREAD); + if ((flag->is_unlocked() || flag->is_unlocker()) && + add_global_entry(env, null_h, &globals[num_entries], flag, THREAD)) { num_entries++; } } @@ -1650,11 +1660,14 @@ bool bvalue = (new_value.z == JNI_TRUE ? true : false); succeed = CommandLineFlags::boolAtPut(name, &bvalue, MANAGEMENT); } else if (flag->is_intx()) { - intx ivalue = new_value.j; + intx ivalue = (intx)new_value.j; succeed = CommandLineFlags::intxAtPut(name, &ivalue, MANAGEMENT); } else if (flag->is_uintx()) { - uintx uvalue = new_value.j; + uintx uvalue = (uintx)new_value.j; succeed = CommandLineFlags::uintxAtPut(name, &uvalue, MANAGEMENT); + } else if (flag->is_uint64_t()) { + uint64_t uvalue = (uint64_t)new_value.j; + succeed = CommandLineFlags::uint64_tAtPut(name, &uvalue, MANAGEMENT); } else if (flag->is_ccstr()) { oop str = JNIHandles::resolve_external_guard(new_value.l); if (str == NULL) { @@ -1734,7 +1747,7 @@ // Make sure we have a String array klassOop element_klass = objArrayKlass::cast(names_ah->klass())->element_klass(); - if (element_klass != SystemDictionary::string_klass()) { + if (element_klass != SystemDictionary::String_klass()) { THROW_MSG_(vmSymbols::java_lang_IllegalArgumentException(), "Array element type is not String class", 0); } @@ -1769,7 +1782,7 @@ num_threads += cycle->num_threads(); } - objArrayOop r = oopFactory::new_objArray(SystemDictionary::thread_klass(), num_threads, CHECK_NH); + objArrayOop r = oopFactory::new_objArray(SystemDictionary::Thread_klass(), num_threads, CHECK_NH); objArrayHandle threads_ah(THREAD, r); int index = 0;
--- a/src/share/vm/services/memoryManager.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/services/memoryManager.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -72,6 +72,14 @@ return (GCMemoryManager*) new PSMarkSweepMemoryManager(); } +GCMemoryManager* MemoryManager::get_g1YoungGen_memory_manager() { + return (GCMemoryManager*) new G1YoungGenMemoryManager(); +} + +GCMemoryManager* MemoryManager::get_g1OldGen_memory_manager() { + return (GCMemoryManager*) new G1OldGenMemoryManager(); +} + instanceOop MemoryManager::get_memory_manager_instance(TRAPS) { // Must do an acquire so as to force ordering of subsequent // loads from anything _memory_mgr_obj points to or implies.
--- a/src/share/vm/services/memoryManager.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/services/memoryManager.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -54,7 +54,9 @@ ParNew, ConcurrentMarkSweep, PSScavenge, - PSMarkSweep + PSMarkSweep, + G1YoungGen, + G1OldGen }; MemoryManager(); @@ -85,6 +87,8 @@ static GCMemoryManager* get_cms_memory_manager(); static GCMemoryManager* get_psScavenge_memory_manager(); static GCMemoryManager* get_psMarkSweep_memory_manager(); + static GCMemoryManager* get_g1YoungGen_memory_manager(); + static GCMemoryManager* get_g1OldGen_memory_manager(); }; @@ -231,3 +235,21 @@ MemoryManager::Name kind() { return MemoryManager::PSMarkSweep; } const char* name() { return "PS MarkSweep"; } }; + +class G1YoungGenMemoryManager : public GCMemoryManager { +private: +public: + G1YoungGenMemoryManager() : GCMemoryManager() {} + + MemoryManager::Name kind() { return MemoryManager::G1YoungGen; } + const char* name() { return "G1 Young Generation"; } +}; + +class G1OldGenMemoryManager : public GCMemoryManager { +private: +public: + G1OldGenMemoryManager() : GCMemoryManager() {} + + MemoryManager::Name kind() { return MemoryManager::G1OldGen; } + const char* name() { return "G1 Old Generation"; } +};
--- a/src/share/vm/services/memoryService.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/services/memoryService.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -60,8 +60,8 @@ break; } case CollectedHeap::G1CollectedHeap : { - G1CollectedHeap::g1_unimplemented(); - return; + add_g1_heap_info(G1CollectedHeap::heap()); + break; } #endif // SERIALGC default: { @@ -164,6 +164,19 @@ add_psOld_memory_pool(heap->old_gen(), _major_gc_manager); add_psPerm_memory_pool(heap->perm_gen(), _major_gc_manager); } + +void MemoryService::add_g1_heap_info(G1CollectedHeap* g1h) { + assert(UseG1GC, "sanity"); + + _minor_gc_manager = MemoryManager::get_g1YoungGen_memory_manager(); + _major_gc_manager = MemoryManager::get_g1OldGen_memory_manager(); + _managers_list->append(_minor_gc_manager); + _managers_list->append(_major_gc_manager); + + add_g1YoungGen_memory_pool(g1h, _major_gc_manager, _minor_gc_manager); + add_g1OldGen_memory_pool(g1h, _major_gc_manager); + add_g1PermGen_memory_pool(g1h, _major_gc_manager); +} #endif // SERIALGC MemoryPool* MemoryService::add_gen(Generation* gen, @@ -384,6 +397,64 @@ mgr->add_pool(perm_gen); _pools_list->append(perm_gen); } + +void MemoryService::add_g1YoungGen_memory_pool(G1CollectedHeap* g1h, + MemoryManager* major_mgr, + MemoryManager* minor_mgr) { + assert(major_mgr != NULL && minor_mgr != NULL, "should have two managers"); + + G1EdenPool* eden = new G1EdenPool(g1h); + G1SurvivorPool* survivor = new G1SurvivorPool(g1h); + + major_mgr->add_pool(eden); + major_mgr->add_pool(survivor); + minor_mgr->add_pool(eden); + minor_mgr->add_pool(survivor); + _pools_list->append(eden); + _pools_list->append(survivor); +} + +void MemoryService::add_g1OldGen_memory_pool(G1CollectedHeap* g1h, + MemoryManager* mgr) { + assert(mgr != NULL, "should have one manager"); + + G1OldGenPool* old_gen = new G1OldGenPool(g1h); + mgr->add_pool(old_gen); + _pools_list->append(old_gen); +} + +void MemoryService::add_g1PermGen_memory_pool(G1CollectedHeap* g1h, + MemoryManager* mgr) { + assert(mgr != NULL, "should have one manager"); + + CompactingPermGenGen* perm_gen = (CompactingPermGenGen*) g1h->perm_gen(); + PermanentGenerationSpec* spec = perm_gen->spec(); + size_t max_size = spec->max_size() - spec->read_only_size() + - spec->read_write_size(); + MemoryPool* pool = add_space(perm_gen->unshared_space(), + "G1 Perm Gen", + false, /* is_heap */ + max_size, + true /* support_usage_threshold */); + mgr->add_pool(pool); + + // in case we support CDS in G1 + if (UseSharedSpaces) { + pool = add_space(perm_gen->ro_space(), + "G1 Perm Gen [shared-ro]", + false, /* is_heap */ + spec->read_only_size(), + true /* support_usage_threshold */); + mgr->add_pool(pool); + + pool = add_space(perm_gen->rw_space(), + "G1 Perm Gen [shared-rw]", + false, /* is_heap */ + spec->read_write_size(), + true /* support_usage_threshold */); + mgr->add_pool(pool); + } +} #endif // SERIALGC void MemoryService::add_code_heap_memory_pool(CodeHeap* heap) {
--- a/src/share/vm/services/memoryService.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/services/memoryService.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -40,6 +40,7 @@ class ParallelScavengeHeap; class CompactingPermGenGen; class CMSPermGenGen; +class G1CollectedHeap; // VM Monitoring and Management Support @@ -88,6 +89,13 @@ static void add_psPerm_memory_pool(PSPermGen* perm, MemoryManager* mgr); + static void add_g1YoungGen_memory_pool(G1CollectedHeap* g1h, + MemoryManager* major_mgr, + MemoryManager* minor_mgr); + static void add_g1OldGen_memory_pool(G1CollectedHeap* g1h, + MemoryManager* mgr); + static void add_g1PermGen_memory_pool(G1CollectedHeap* g1h, + MemoryManager* mgr); static MemoryPool* add_space(ContiguousSpace* space, const char* name, @@ -111,6 +119,7 @@ static void add_gen_collected_heap_info(GenCollectedHeap* heap); static void add_parallel_scavenge_heap_info(ParallelScavengeHeap* heap); + static void add_g1_heap_info(G1CollectedHeap* g1h); public: static void set_universe_heap(CollectedHeap* heap);
--- a/src/share/vm/services/runtimeService.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/services/runtimeService.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2003-2006 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2003-2010 Sun Microsystems, Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -104,6 +104,13 @@ void RuntimeService::record_safepoint_begin() { HS_DTRACE_PROBE(hs_private, safepoint__begin); + + // Print the time interval in which the app was executing + if (PrintGCApplicationConcurrentTime) { + gclog_or_tty->print_cr("Application time: %3.7f seconds", + last_application_time_sec()); + } + // update the time stamp to begin recording safepoint time _safepoint_timer.update(); if (UsePerfData) { @@ -122,6 +129,15 @@ void RuntimeService::record_safepoint_end() { HS_DTRACE_PROBE(hs_private, safepoint__end); + + // Print the time interval for which the app was stopped + // during the current safepoint operation. + if (PrintGCApplicationStoppedTime) { + gclog_or_tty->print_cr("Total time for which application threads " + "were stopped: %3.7f seconds", + last_safepoint_time_sec()); + } + // update the time stamp to begin recording app time _app_timer.update(); if (UsePerfData) {
--- a/src/share/vm/services/serviceUtil.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/services/serviceUtil.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -45,7 +45,7 @@ // instance if (o->is_instance()) { // instance objects are visible - if (o->klass() != SystemDictionary::class_klass()) { + if (o->klass() != SystemDictionary::Class_klass()) { return true; } if (java_lang_Class::is_primitive(o)) {
--- a/src/share/vm/services/threadService.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/services/threadService.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -540,7 +540,7 @@ } Handle ThreadStackTrace::allocate_fill_stack_trace_element_array(TRAPS) { - klassOop k = SystemDictionary::stackTraceElement_klass(); + klassOop k = SystemDictionary::StackTraceElement_klass(); assert(k != NULL, "must be loaded in 1.4+"); instanceKlassHandle ik(THREAD, k);
--- a/src/share/vm/utilities/constantTag.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/utilities/constantTag.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -36,7 +36,8 @@ JVM_CONSTANT_UnresolvedString = 102, // Temporary tag until actual use JVM_CONSTANT_StringIndex = 103, // Temporary tag while constructing constant pool JVM_CONSTANT_UnresolvedClassInError = 104, // Error tag due to resolution error - JVM_CONSTANT_InternalMax = 104 // Last implementation tag + JVM_CONSTANT_Object = 105, // Required for BoundMethodHandle arguments. + JVM_CONSTANT_InternalMax = 105 // Last implementation tag }; @@ -70,6 +71,8 @@ bool is_unresolved_string() const { return _tag == JVM_CONSTANT_UnresolvedString; } bool is_string_index() const { return _tag == JVM_CONSTANT_StringIndex; } + bool is_object() const { return _tag == JVM_CONSTANT_Object; } + bool is_klass_reference() const { return is_klass_index() || is_unresolved_klass(); } bool is_klass_or_reference() const{ return is_klass() || is_klass_reference(); } bool is_field_or_method() const { return is_field() || is_method() || is_interface_method(); }
--- a/src/share/vm/utilities/exceptions.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/utilities/exceptions.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -122,7 +122,7 @@ // Check for special boot-strapping/vm-thread handling if (special_exception(thread, file, line, h_exception)) return; - assert(h_exception->is_a(SystemDictionary::throwable_klass()), "exception is not a subclass of java/lang/Throwable"); + assert(h_exception->is_a(SystemDictionary::Throwable_klass()), "exception is not a subclass of java/lang/Throwable"); // set the pending exception thread->set_pending_exception(h_exception(), file, line); @@ -255,7 +255,7 @@ // Future: object initializer should take a cause argument if (h_cause() != NULL) { - assert(h_cause->is_a(SystemDictionary::throwable_klass()), + assert(h_cause->is_a(SystemDictionary::Throwable_klass()), "exception cause is not a subclass of java/lang/Throwable"); JavaValue result1(T_OBJECT); JavaCallArguments args1;
--- a/src/share/vm/utilities/globalDefinitions.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/utilities/globalDefinitions.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -139,6 +139,10 @@ const size_t G = M*K; const size_t HWperKB = K / sizeof(HeapWord); +const size_t LOG_K = 10; +const size_t LOG_M = 2 * LOG_K; +const size_t LOG_G = 2 * LOG_M; + const jint min_jint = (jint)1 << (sizeof(jint)*BitsPerByte-1); // 0x80000000 == smallest jint const jint max_jint = (juint)min_jint - 1; // 0x7FFFFFFF == largest jint
--- a/src/share/vm/utilities/growableArray.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/utilities/growableArray.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -278,6 +278,17 @@ _len--; } + // inserts the given element before the element at index i + void insert_before(const int idx, const E& elem) { + check_nesting(); + if (_len == _max) grow(_len); + for (int j = _len - 1; j >= idx; j--) { + _data[j + 1] = _data[j]; + } + _len++; + _data[idx] = elem; + } + void appendAll(const GrowableArray<E>* l) { for (int i = 0; i < l->_len; i++) { raw_at_put_grow(_len, l->_data[i], 0);
--- a/src/share/vm/utilities/numberSeq.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/utilities/numberSeq.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -241,3 +241,33 @@ return b0 + b1 * num; } + + +// Printing/Debugging Support + +void AbsSeq::dump() { dump_on(gclog_or_tty); } + +void AbsSeq::dump_on(outputStream* s) { + s->print_cr("\t _num = %d, _sum = %7.3f, _sum_of_squares = %7.3f", + _num, _sum, _sum_of_squares); + s->print_cr("\t _davg = %7.3f, _dvariance = %7.3f, _alpha = %7.3f", + _davg, _dvariance, _alpha); +} + +void NumberSeq::dump_on(outputStream* s) { + AbsSeq::dump_on(s); + s->print_cr("\t\t _last = %7.3f, _maximum = %7.3f"); +} + +void TruncatedSeq::dump_on(outputStream* s) { + AbsSeq::dump_on(s); + s->print_cr("\t\t _length = %d, _next = %d", _length, _next); + for (int i = 0; i < _length; i++) { + if (i%5 == 0) { + s->cr(); + s->print("\t"); + } + s->print("\t[%d]=%7.3f", i, _sequence[i]); + } + s->print_cr(""); +}
--- a/src/share/vm/utilities/numberSeq.hpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/utilities/numberSeq.hpp Wed Mar 24 17:16:33 2010 -0700 @@ -74,6 +74,10 @@ double davg() const; // decaying average double dvariance() const; // decaying variance double dsd() const; // decaying "standard deviation" + + // Debugging/Printing + virtual void dump(); + virtual void dump_on(outputStream* s); }; class NumberSeq: public AbsSeq { @@ -91,6 +95,9 @@ virtual void add(double val); virtual double maximum() const { return _maximum; } virtual double last() const { return _last; } + + // Debugging/Printing + virtual void dump_on(outputStream* s); }; class TruncatedSeq: public AbsSeq { @@ -114,4 +121,7 @@ double oldest() const; // the oldest valid value in the sequence double predict_next() const; // prediction based on linear regression + + // Debugging/Printing + virtual void dump_on(outputStream* s); };
--- a/src/share/vm/utilities/vmError.cpp Mon Mar 15 15:51:36 2010 -0400 +++ b/src/share/vm/utilities/vmError.cpp Wed Mar 24 17:16:33 2010 -0700 @@ -502,6 +502,23 @@ #endif // ZERO } + STEP(135, "(printing target Java thread stack)" ) + + // printing Java thread stack trace if it is involved in GC crash + if (_verbose && (_thread->is_Named_thread())) { + JavaThread* jt = ((NamedThread *)_thread)->processed_thread(); + if (jt != NULL) { + st->print_cr("JavaThread " PTR_FORMAT " (nid = " UINTX_FORMAT ") was being processed", jt, jt->osthread()->thread_id()); + if (jt->has_last_Java_frame()) { + st->print_cr("Java frames: (J=compiled Java code, j=interpreted, Vv=VM code)"); + for(StackFrameStream sfs(jt); !sfs.is_done(); sfs.next()) { + sfs.current()->print_on_error(st, buf, sizeof(buf), true); + st->cr(); + } + } + } + } + STEP(140, "(printing VM operation)" ) if (_verbose && _thread && _thread->is_VM_thread()) {
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6769124/TestArrayCopy6769124.java Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,53 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/** + * @test + * @bug 6769124 + * @summary arraycopy may crash the VM with c1 on 64 bit + */ + +public class TestArrayCopy6769124 { + + public static void main(String[] args) { + + int k = 1 << 31; + + + for(int j = 0; j <1000000; j++) { + int i = -1; + while(i < 10) { + i++; + } + + int m = k * i; + + int[] O1 = new int[20]; + int[] O2 = new int[20]; + + System.arraycopy(O1, i, O2, i, 1); //will crash on amd64 + System.arraycopy(O1, m, O2, m, 1); //will crash on sparcv9 + } + } +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6769124/TestDeoptInt6769124.java Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,56 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/** + * @test + * @bug 6769124 + * @summary int value might not be correctly decoded on deopt with c1 on 64 bit + * + * @run main/othervm -Xcomp -XX:CompileOnly=TestDeoptInt6769124.m TestDeoptInt6769124 + */ + +public class TestDeoptInt6769124 { + + static class A { + volatile int vl; + A(int v) { + vl = v; + } + } + + static void m(int b) { + A a = new A(10); + int c; + c = b + a.vl; //accessing volatile field of class not loaded at compile time forces a deopt + if(c != 20) { + System.out.println("a (= " + a.vl + ") + b (= " + b + ") = c (= " + c + ") != 20"); + throw new InternalError(); + } + } + + public static void main(String[] args) { + m(10); + } + +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6769124/TestUnalignedLoad6769124.java Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,69 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/** + * @test + * @bug 6769124 + * @summary unaligned load may fail with c1 on 64 bit + */ + +public class TestUnalignedLoad6769124 { + + static long l1v = 0x200000003L; + static long l2v = 0x400000005L; + static double d1v = Double.MAX_VALUE; + static double d2v = Double.MIN_VALUE; + + public static void main(String[] args) { + long l1 = l1v; + double d1 = d1v; + long l2 = l2v; + double d2 = d2v; + + // Run long enough to induce an OSR + for (int i = 0; i < 10000000; i++) { + } + boolean error = false; + + if (l1 != l1v) { + System.out.println(l1 + " != " + l1v); + error = true; + } + if (l2 != l2v) { + System.out.println(l2 + " != " + l2v); + error = true; + } + if (d1 != d1v) { + System.out.println(d1 + " != " + d1v); + error = true; + } + if (d2 != d2v) { + System.out.println(d2 + " != " + d2v); + error = true; + } + if (error) { + throw new InternalError(); + } + } +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6792161/Test6792161.java Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,49 @@ +/* + * Copyright 2010 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/** + * @test + * @bug 6792161 + * @summary assert("No dead instructions after post-alloc") + * + * @run main/othervm -Xcomp -XX:MaxInlineSize=120 Test6792161 + */ + +import java.lang.reflect.Constructor; +public class Test6792161 { + static Constructor test(Class cls) throws Exception { + Class[] args= { String.class }; + try { + return cls.getConstructor(args); + } catch (NoSuchMethodException e) {} + return cls.getConstructor(new Class[0]); + } + public static void main(final String[] args) throws Exception { + try { + for (int i = 0; i < 100000; i++) { + Constructor ctor = test(Class.forName("Test6792161")); + } + } catch (NoSuchMethodException e) {} + } +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6852078/Test6852078.java Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,57 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/** + * @test + * @bug 6852078 + * @summary Disable SuperWord optimization for unsafe read/write + * + * @run main/othervm Test6852078 + */ + +import java.util.*; +import java.nio.ByteBuffer; +import com.sun.corba.se.impl.encoding.ByteBufferWithInfo; +import com.sun.jndi.toolkit.corba.CorbaUtils; + +public class Test6852078 { + + public Test6852078(String [] args) { + + int capacity = 128; + ByteBuffer bb = ByteBuffer.allocateDirect(capacity); + ByteBufferWithInfo bbwi = new ByteBufferWithInfo( CorbaUtils.getOrb(null, -1, new Hashtable()), bb); + byte[] tmpBuf; + tmpBuf = new byte[bbwi.buflen]; + + for (int i = 0; i < capacity; i++) + tmpBuf[i] = bbwi.byteBuffer.get(i); + } + + public static void main(String [] args) { + for (int i=0; i<2000; i++) { + Test6852078 t = new Test6852078(args); + } + } +}
--- a/test/compiler/6877254/Test.java Mon Mar 15 15:51:36 2010 -0400 +++ b/test/compiler/6877254/Test.java Wed Mar 24 17:16:33 2010 -0700 @@ -26,7 +26,7 @@ * @bug 6877254 * @summary Implement StoreCMNode::Ideal to promote its OopStore above the MergeMem * - * @run main/othervm -server -Xcomp -XX:+UseConcMarkSweepGC Test + * @run main/othervm -Xcomp Test */ public class Test {
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6895383/Test.java Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,54 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/** + * @test + * @bug 6895383 + * @summary JCK test throws NPE for method compiled with Escape Analysis + * + * @run main/othervm -Xcomp Test + */ + +import java.util.*; +import java.util.concurrent.*; + +public class Test { + public static void main(String argv[]) { + Test test = new Test(); + test.testRemove1_IndexOutOfBounds(); + test.testAddAll1_IndexOutOfBoundsException(); + } + + public void testRemove1_IndexOutOfBounds() { + CopyOnWriteArrayList c = new CopyOnWriteArrayList(); + } + + public void testAddAll1_IndexOutOfBoundsException() { + try { + CopyOnWriteArrayList c = new CopyOnWriteArrayList(); + c.addAll(-1, new LinkedList()); // should throw IndexOutOfBoundsException + } catch (IndexOutOfBoundsException e) { + } + } +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6896727/Test.java Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,48 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/* + * @test + * @bug 6896727 + * @summary nsk/logging/LoggingPermission/LoggingPermission/logperm002 fails with G1, EscapeAnalisys w/o COOPs + * @run main/othervm -XX:+IgnoreUnrecognizedVMOptions -Xcomp -XX:+DoEscapeAnalysis Test + */ + +public class Test { + + final static String testString = "abracadabra"; + public static void main(String args[]) { + String params[][] = { + {"control", testString} + }; + for (int i=0; i<params.length; i++) { + try { + System.out.println("Params :" + testString + " and " + params[i][0] + ", " + params[i][1]); + if (params[i][1] == null) { + System.exit(97); + } + } catch (Exception e) {} + } + } +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6901572/Test.java Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,55 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + */ + +/** + * @test + * @bug 6901572 + * @summary JVM 1.6.16 crash on loops: assert(has_node(i),"") + * + * @run main/othervm Test + */ + + +public class Test { + + public static void main(String[] args) { + for (int i = 0; i < 2; i++) + NestedLoop(); + } + + public static long NestedLoop() { + final int n = 50; + long startTime = System.currentTimeMillis(); + int x = 0; + for(int a = 0; a < n; a++) + for(int b = 0; b < n; b++) + for(int c = 0; c < n; c++) + for(int d = 0; d < n; d++) + for(int e = 0; e < n; e++) + for(int f = 0; f < n; f++) + x++; + long stopTime = System.currentTimeMillis(); + + return stopTime - startTime; + } +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6909839/Test6909839.java Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,282 @@ +/* + * Copyright 2010 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/** + * @test + * @bug 6909839 + * @summary missing unsigned compare cases for some cmoves in sparc.ad + * + * @run main/othervm -XX:+AggressiveOpts -Xbatch Test6909839 + */ + +public class Test6909839 { + public static void main(String[] args) { + testi(); + testi(); + testi(); + testui(); + testui(); + testui(); + testdi(); + testdi(); + testdi(); + testfi(); + testfi(); + testfi(); + + testl(); + testl(); + testl(); + testul(); + testul(); + testul(); + testdl(); + testdl(); + testdl(); + testfl(); + testfl(); + testfl(); + + testf(); + testf(); + testf(); + testuf(); + testuf(); + testuf(); + testdf(); + testdf(); + testdf(); + testff(); + testff(); + testff(); + + testd(); + testd(); + testd(); + testud(); + testud(); + testud(); + testdd(); + testdd(); + testdd(); + testfd(); + testfd(); + testfd(); + + testp(); + testp(); + testp(); + testup(); + testup(); + testup(); + testdp(); + testdp(); + testdp(); + testfp(); + testfp(); + testfp(); + } + + static void testui() { + int total = 0; + for (int i = 0 ; i < 10000; i++) { + int v = i % 4; + total += ((v >= 1 && v < 3) ? 1 : 2); + } + System.out.println(total); + } + + static void testdi() { + int total = 0; + for (int i = 0 ; i < 10000; i++) { + int v = i % 4; + total += (v > 1.0) ? 1 : 2; + } + System.out.println(total); + } + + static void testfi() { + int total = 0; + for (int i = 0 ; i < 10000; i++) { + int v = i % 4; + total += (v > 1.0f) ? 1 : 2; + } + System.out.println(total); + } + + static void testi() { + int total = 0; + for (int i = 0 ; i < 10000; i++) { + total += (i % 4 != 0) ? 1 : 2; + } + System.out.println(total); + } + + static void testul() { + long total = 0; + for (int i = 0 ; i < 10000; i++) { + int v = i % 4; + total += ((v >= 1 && v < 3) ? 1L : 2L); + } + System.out.println(total); + } + + static void testdl() { + long total = 0; + for (int i = 0 ; i < 10000; i++) { + int v = i % 4; + total += (v > 1.0) ? 1L : 2L; + } + System.out.println(total); + } + + static void testfl() { + long total = 0; + for (int i = 0 ; i < 10000; i++) { + int v = i % 4; + total += (v > 1.0f) ? 1L : 2L; + } + System.out.println(total); + } + + static void testl() { + long total = 0; + for (int i = 0 ; i < 10000; i++) { + total += (i % 4 != 0) ? 1L : 2L; + } + System.out.println(total); + } + + static void testuf() { + float total = 0; + for (int i = 0 ; i < 10000; i++) { + int v = i % 4; + total += ((v >= 1 && v < 3) ? 1.0f : 2.0f); + } + System.out.println(total); + } + + static void testdf() { + float total = 0; + for (int i = 0 ; i < 10000; i++) { + int v = i % 4; + total += (v > 0.0) ? 1.0f : 2.0f; + } + System.out.println(total); + } + + static void testff() { + float total = 0; + for (int i = 0 ; i < 10000; i++) { + int v = i % 4; + total += (v > 0.0f) ? 1.0f : 2.0f; + } + System.out.println(total); + } + + static void testf() { + float total = 0; + for (int i = 0 ; i < 10000; i++) { + total += (i % 4 != 0) ? 1.0f : 2.0f; + } + System.out.println(total); + } + + static void testud() { + double total = 0; + for (int i = 0 ; i < 10000; i++) { + int v = i % 4; + total += ((v >= 1 && v < 3) ? 1.0d : 2.0d); + } + System.out.println(total); + } + + static void testdd() { + double total = 0; + for (int i = 0 ; i < 10000; i++) { + int v = i % 4; + total += (v > 1.0) ? 1.0d : 2.0d; + } + System.out.println(total); + } + + static void testfd() { + double total = 0; + for (int i = 0 ; i < 10000; i++) { + int v = i % 4; + total += (v > 1.0f) ? 1.0d : 2.0d; + } + System.out.println(total); + } + + static void testd() { + double total = 0; + for (int i = 0 ; i < 10000; i++) { + total += (i % 4 != 0) ? 1.0d : 2.0d; + } + System.out.println(total); + } + + static void testp() { + Object a = new Object(); + Object b = new Object();; + int total = 0; + for (int i = 0 ; i < 10000; i++) { + total += ((i % 4 != 0) ? a : b).hashCode(); + } + System.out.println(total); + } + + static void testup() { + Object a = new Object(); + Object b = new Object();; + int total = 0; + for (int i = 0 ; i < 10000; i++) { + int v = i % 4; + total += ((v >= 1 && v < 3) ? a : b).hashCode(); + } + System.out.println(total); + } + + static void testdp() { + Object a = new Object(); + Object b = new Object();; + int total = 0; + for (int i = 0 ; i < 10000; i++) { + int v = i % 4; + total += ((v > 1.0) ? a : b).hashCode(); + } + System.out.println(total); + } + static void testfp() { + Object a = new Object(); + Object b = new Object();; + int total = 0; + for (int i = 0 ; i < 10000; i++) { + int v = i % 4; + total += ((v > 1.0f) ? a : b).hashCode(); + } + System.out.println(total); + } +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6910484/Test.java Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,50 @@ +/* + * Copyright 2009 SAP. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + */ + +/** + * @test + * @bug 6910484 + * @summary incorrect integer optimization (loosing and op-r in a given example) + * + * @run main/othervm -Xbatch Test + */ + +public class Test { + + public static void main(String[] args) { + long iteration = 0; + for(int i = 0; i <11000; i++) { + iteration++; + int result = test(255); + if (result != 112) { + System.out.println("expected 112, but got " + result + " after iteration " + iteration); + System.exit(97); + } + } + } + + private static int test(int x) { + return (x & -32) / 2; + } + +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6910605/Test.java Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,78 @@ +/* + * Copyright 2010 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/** + * @test + * @bug 6910605 + * @summary C2: NullPointerException/ClassCaseException is thrown when C2 with DeoptimizeALot is used + * + * @run main/othervm -XX:+IgnoreUnrecognizedVMOptions -XX:+DeoptimizeALot -Xbatch Test + * + * original test: nsk/coverage/runtime/runtime007 + */ + +import java.io.*; + +public class Test { + public static int buf=0; + + public static void main( String argv[] ) { + System.exit(run(argv, System.out)+95); + } + + public static int run(String argv[],PrintStream out) { + int ret=0, retx=0, bad=0; + + for( int i=0; (i < 100000) && (bad < 10) ; i++ ) { + retx = OptoRuntime_f2i_Type(out); + ret += retx; + if( retx !=0 ) { + out.println("i="+i); + bad++; + } + } + return ret==0 ? 0 : 2 ; + } + + public static int OptoRuntime_f2i_Type(PrintStream out) { + int c1=2, c2=3, c3=4, c4=5, c5=6; + int j=0, k=0; + try { + int[][] iii=(int[][])(new int[c1][c2]); + + for( j=0; j<c1; j++ ) { + for( k=0; k<c2; k++ ) { + iii[j][k]=(int)((float)(j+1)/(float)(k+1)); + } + } + } catch (Throwable e) { + out.println("Unexpected exception " + e); + e.printStackTrace(out); + out.println("j="+j+", k="+k); + return 1; + } + return 0; + } + +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6910618/Test.java Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,74 @@ +/* + * Copyright 2010 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/** + * @test + * @bug 6910605 + * @summary C2: NullPointerException/ClassCaseException is thrown when C2 with DeoptimizeALot is used + * + * @run main/othervm -Xmx64m -XX:+IgnoreUnrecognizedVMOptions -XX:+DeoptimizeALot -XX:+DoEscapeAnalysis -Xbatch -XX:InlineSmallCode=2000 Test + * + */ + +/* + * Added InlineSmallCode=2000 to guaranty inlining of StringBuilder::append() to allow scalar replace StringBuilder object. + * + * original test: gc/gctests/StringGC + */ + +public class Test { + private final String toAdd = "0123456789abcdef"; + private int maxLength; + private static final int numberOfThreads = 8; + + private class StringAdder extends Thread { + private String s; + + public void test() { + s = s + toAdd; + } + public void run() { + do { + test(); + } while (s.length() < maxLength); + } + } + + public void test() throws InterruptedException { + maxLength = toAdd.length() * 15000/ numberOfThreads; + StringAdder[] sa = new StringAdder[numberOfThreads]; + for (int i = 0; i < numberOfThreads; i++) { + sa[i] = new StringAdder(); + sa[i].start(); + } + for (int i = 0; i < numberOfThreads; i++) { + sa[i].join(); + } + } + + public static void main(String[] args) throws InterruptedException { + Test t = new Test(); + t.test(); + } +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6912517/Test.java Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,113 @@ +/* + * Copyright 2009 D.E. Shaw. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + */ + +/** + * @test + * @bug 6912517 + * @summary JIT bug compiles out (and stops running) code that needs to be run. Causes NPE. + * + * @run main/othervm -Xbatch -XX:CompileThreshold=100 -XX:+IgnoreUnrecognizedVMOptions -XX:-UseCompressedOops Test + */ + +/** + * Highlights a bug with the JIT compiler. + * @author Matt Bruce m b r u c e __\at/__ g m a i l DOT c o m + */ +public class Test implements Runnable +{ + private final Thread myThread; + private Thread myInitialThread; + private boolean myShouldCheckThreads; + + /** + * Sets up the running thread, and starts it. + */ + public Test(int id) + { + myThread = new Thread(this); + myThread.setName("Runner: " + id); + myThread.start(); + myShouldCheckThreads = false; + } + + /** + * @param shouldCheckThreads the shouldCheckThreads to set + */ + public void setShouldCheckThreads(boolean shouldCheckThreads) + { + myShouldCheckThreads = shouldCheckThreads; + } + + /** + * Starts up the two threads with enough delay between them for JIT to + * kick in. + * @param args + * @throws InterruptedException + */ + public static void main(String[] args) throws InterruptedException + { + // let this run for a bit, so the "run" below is JITTed. + for (int id = 0; id < 20; id++) { + System.out.println("Starting thread: " + id); + Test bug = new Test(id); + bug.setShouldCheckThreads(true); + Thread.sleep(2500); + } + } + + /** + * @see java.lang.Runnable#run() + */ + public void run() + { + long runNumber = 0; + while (true) { + // run hot for a little while, give JIT time to kick in to this loop. + // then run less hot. + if (runNumber > 15000) { + try { + Thread.sleep(5); + } + catch (InterruptedException e) { + e.printStackTrace(); + } + } + runNumber++; + ensureProperCallingThread(); + } + } + + private void ensureProperCallingThread() + { + // this should never be null. but with the JIT bug, it will be. + // JIT BUG IS HERE ==>>>>> + if (myShouldCheckThreads) { + if (myInitialThread == null) { + myInitialThread = Thread.currentThread(); + } + else if (myInitialThread != Thread.currentThread()) { + System.out.println("Not working: " + myInitialThread.getName()); + } + } + } +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6916644/Test6916644.java Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,50 @@ +/* + * Copyright 2010 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/** + * @test + * @bug 6916644 + * @summary C2 compiler crash on x86 + * + * @run main/othervm -Xcomp -XX:CompileOnly=Test6916644.test Test6916644 + */ + +public class Test6916644 { + static int result; + static int i1; + static int i2; + + static public void test(double d) { + result = (d <= 0.0D) ? i1 : i2; + } + + public static void main(String[] args) { + for (int i = 0; i < 100000; i++) { + // use an alternating value so the test doesn't always go + // the same direction. Otherwise we won't transform it + // into a cmove. + test(i & 1); + } + } +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6921969/TestMultiplyLongHiZero.java Wed Mar 24 17:16:33 2010 -0700 @@ -0,0 +1,138 @@ +/* + * Copyright 2010 Google, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/* + * @test + * @bug 6921969 + * @summary Tests shorter long multiply sequences when the high 32 bits of long operands are known to be zero on x86_32 + * @run main/othervm -Xbatch -XX:-Inline -XX:CompileOnly=.testNormal,.testLeftOptimized,.testRightOptimized,.testOptimized,.testLeftOptimized_LoadUI2L,.testRightOptimized_LoadUI2L,.testOptimized_LoadUI2L TestMultiplyLongHiZero + */ + +// This test must run without any command line arguments. + +public class TestMultiplyLongHiZero { + + private static void check(long leftFactor, long rightFactor, long optimizedProduct, long constantProduct) { + long normalProduct = leftFactor * rightFactor; // unaffected by the new optimization + if (optimizedProduct != constantProduct || normalProduct != constantProduct) { + throw new RuntimeException("Not all three products are equal: " + + Long.toHexString(normalProduct) + ", " + + Long.toHexString(optimizedProduct) + ", " + + Long.toHexString(constantProduct)); + } + } + + private static int initInt(String[] args, int v) { + if (args.length > 0) { + try { + return Integer.valueOf(args[0]); + } catch (NumberFormatException e) { } + } + return v; + } + + private static final long mask32 = 0x00000000FFFFFFFFL; + + private static void testNormal(int leftFactor, int rightFactor, long constantProduct) { + check((long) leftFactor, + (long) rightFactor, + (long) leftFactor * (long) rightFactor, // unaffected by the new optimization + constantProduct); + } + + private static void testLeftOptimized(int leftFactor, int rightFactor, long constantProduct) { + check((leftFactor & mask32), + (long) rightFactor, + (leftFactor & mask32) * (long) rightFactor, // left factor optimized + constantProduct); + } + + private static void testRightOptimized(int leftFactor, int rightFactor, long constantProduct) { + check((long) leftFactor, + (rightFactor & mask32), + (long) leftFactor * (rightFactor & mask32), // right factor optimized + constantProduct); + } + + private static void testOptimized(int leftFactor, int rightFactor, long constantProduct) { + check((leftFactor & mask32), + (rightFactor & mask32), + (leftFactor & mask32) * (rightFactor & mask32), // both factors optimized + constantProduct); + } + + private static void testLeftOptimized_LoadUI2L(int leftFactor, int rightFactor, long constantProduct, int[] factors) { + check((leftFactor & mask32), + (long) rightFactor, + (factors[0] & mask32) * (long) rightFactor, // left factor optimized + constantProduct); + } + + private static void testRightOptimized_LoadUI2L(int leftFactor, int rightFactor, long constantProduct, int[] factors) { + check((long) leftFactor, + (rightFactor & mask32), + (long) leftFactor * (factors[1] & mask32), // right factor optimized + constantProduct); + } + + private static void testOptimized_LoadUI2L(int leftFactor, int rightFactor, long constantProduct, int[] factors) { + check((leftFactor & mask32), + (rightFactor & mask32), + (factors[0] & mask32) * (factors[1] & mask32), // both factors optimized + constantProduct); + } + + private static void test(int leftFactor, int rightFactor, + long normalConstantProduct, + long leftOptimizedConstantProduct, + long rightOptimizedConstantProduct, + long optimizedConstantProduct) { + int[] factors = new int[2]; + factors[0] = leftFactor; + factors[1] = rightFactor; + testNormal(leftFactor, rightFactor, normalConstantProduct); + testLeftOptimized(leftFactor, rightFactor, leftOptimizedConstantProduct); + testRightOptimized(leftFactor, rightFactor, rightOptimizedConstantProduct); + testOptimized(leftFactor, rightFactor, optimizedConstantProduct); + testLeftOptimized_LoadUI2L(leftFactor, rightFactor, leftOptimizedConstantProduct, factors); + testRightOptimized_LoadUI2L(leftFactor, rightFactor, rightOptimizedConstantProduct, factors); + testOptimized_LoadUI2L(leftFactor, rightFactor, optimizedConstantProduct, factors); + } + + public static void main(String[] args) { + for (int i = 0; i < 100000; ++i) { // Trigger compilation + int i0 = initInt(args, 1); + int i1 = initInt(args, 3); + int i2 = initInt(args, -1); + int i3 = initInt(args, 0x7FFFFFFF); + test(i0, i1, 3L, 3L, 3L, 3L); + test(i0, i2, -1L, -1L, 0xFFFFFFFFL, 0xFFFFFFFFL); + test(i0, i3, 0x7FFFFFFFL, 0x7FFFFFFFL, 0x7FFFFFFFL, 0x7FFFFFFFL); + test(i1, i2, -3L, -3L, 0x2FFFFFFFDL, 0x2FFFFFFFDL); + test(i1, i3, 0x17FFFFFFDL, 0x17FFFFFFDL, 0x17FFFFFFDL, 0x17FFFFFFDL); + test(i2, i3, 0xFFFFFFFF80000001L, 0x7FFFFFFE80000001L, + 0xFFFFFFFF80000001L, 0x7FFFFFFE80000001L); + } + } +}