# HG changeset patch
# User kvn
# Date 1302561031 25200
# Node ID 15c9a0e162697752443632bcb059121705a04063
# Parent  328926869b15a73c231b76afb5e64b7c41214f0f
7035713: 3DNow Prefetch Instruction Support
Summary: The upcoming processors from AMD are the first that support 3dnow prefetch without supporting the 3dnow instruction set.
Reviewed-by: kvn
Contributed-by: tom.deneau@amd.com

diff -r 328926869b15 -r 15c9a0e16269 src/cpu/x86/vm/assembler_x86.cpp
--- a/src/cpu/x86/vm/assembler_x86.cpp	Sat Apr 09 22:55:25 2011 -0700
+++ b/src/cpu/x86/vm/assembler_x86.cpp	Mon Apr 11 15:30:31 2011 -0700
@@ -2317,7 +2317,7 @@
 }
 
 void Assembler::prefetchr(Address src) {
-  NOT_LP64(assert(VM_Version::supports_3dnow(), "must support"));
+  NOT_LP64(assert(VM_Version::supports_3dnow_prefetch(), "must support"));
   InstructionMark im(this);
   prefetch_prefix(src);
   emit_byte(0x0D);
@@ -2349,7 +2349,7 @@
 }
 
 void Assembler::prefetchw(Address src) {
-  NOT_LP64(assert(VM_Version::supports_3dnow(), "must support"));
+  NOT_LP64(assert(VM_Version::supports_3dnow_prefetch(), "must support"));
   InstructionMark im(this);
   prefetch_prefix(src);
   emit_byte(0x0D);
diff -r 328926869b15 -r 15c9a0e16269 src/cpu/x86/vm/c1_LIRAssembler_x86.cpp
--- a/src/cpu/x86/vm/c1_LIRAssembler_x86.cpp	Sat Apr 09 22:55:25 2011 -0700
+++ b/src/cpu/x86/vm/c1_LIRAssembler_x86.cpp	Mon Apr 11 15:30:31 2011 -0700
@@ -1401,7 +1401,7 @@
       default:
         ShouldNotReachHere(); break;
     }
-  } else if (VM_Version::supports_3dnow()) {
+  } else if (VM_Version::supports_3dnow_prefetch()) {
     __ prefetchr(from_addr);
   }
 }
@@ -1424,7 +1424,7 @@
       default:
         ShouldNotReachHere(); break;
     }
-  } else if (VM_Version::supports_3dnow()) {
+  } else if (VM_Version::supports_3dnow_prefetch()) {
     __ prefetchw(from_addr);
   }
 }
diff -r 328926869b15 -r 15c9a0e16269 src/cpu/x86/vm/vm_version_x86.cpp
--- a/src/cpu/x86/vm/vm_version_x86.cpp	Sat Apr 09 22:55:25 2011 -0700
+++ b/src/cpu/x86/vm/vm_version_x86.cpp	Mon Apr 11 15:30:31 2011 -0700
@@ -348,7 +348,7 @@
   }
 
   char buf[256];
-  jio_snprintf(buf, sizeof(buf), "(%u cores per cpu, %u threads per core) family %d model %d stepping %d%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s",
+  jio_snprintf(buf, sizeof(buf), "(%u cores per cpu, %u threads per core) family %d model %d stepping %d%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s",
                cores_per_cpu(), threads_per_core(),
                cpu_family(), _model, _stepping,
                (supports_cmov() ? ", cmov" : ""),
@@ -363,8 +363,7 @@
                (supports_sse4_2() ? ", sse4.2" : ""),
                (supports_popcnt() ? ", popcnt" : ""),
                (supports_mmx_ext() ? ", mmxext" : ""),
-               (supports_3dnow()   ? ", 3dnow"  : ""),
-               (supports_3dnow2()  ? ", 3dnowext" : ""),
+               (supports_3dnow_prefetch() ? ", 3dnowpref" : ""),
                (supports_lzcnt()   ? ", lzcnt": ""),
                (supports_sse4a()   ? ", sse4a": ""),
                (supports_ht() ? ", ht": ""));
@@ -522,13 +521,13 @@
   // set valid Prefetch instruction
   if( ReadPrefetchInstr < 0 ) ReadPrefetchInstr = 0;
   if( ReadPrefetchInstr > 3 ) ReadPrefetchInstr = 3;
-  if( ReadPrefetchInstr == 3 && !supports_3dnow() ) ReadPrefetchInstr = 0;
-  if( !supports_sse() && supports_3dnow() ) ReadPrefetchInstr = 3;
+  if( ReadPrefetchInstr == 3 && !supports_3dnow_prefetch() ) ReadPrefetchInstr = 0;
+  if( !supports_sse() && supports_3dnow_prefetch() ) ReadPrefetchInstr = 3;
 
   if( AllocatePrefetchInstr < 0 ) AllocatePrefetchInstr = 0;
   if( AllocatePrefetchInstr > 3 ) AllocatePrefetchInstr = 3;
-  if( AllocatePrefetchInstr == 3 && !supports_3dnow() ) AllocatePrefetchInstr=0;
-  if( !supports_sse() && supports_3dnow() ) AllocatePrefetchInstr = 3;
+  if( AllocatePrefetchInstr == 3 && !supports_3dnow_prefetch() ) AllocatePrefetchInstr=0;
+  if( !supports_sse() && supports_3dnow_prefetch() ) AllocatePrefetchInstr = 3;
 
   // Allocation prefetch settings
   intx cache_line_size = L1_data_cache_line_size();
@@ -576,10 +575,10 @@
                   logical_processors_per_package());
     tty->print_cr("UseSSE=%d",UseSSE);
     tty->print("Allocation: ");
-    if (AllocatePrefetchStyle <= 0 || UseSSE == 0 && !supports_3dnow()) {
+    if (AllocatePrefetchStyle <= 0 || UseSSE == 0 && !supports_3dnow_prefetch()) {
       tty->print_cr("no prefetching");
     } else {
-      if (UseSSE == 0 && supports_3dnow()) {
+      if (UseSSE == 0 && supports_3dnow_prefetch()) {
         tty->print("PREFETCHW");
       } else if (UseSSE >= 1) {
         if (AllocatePrefetchInstr == 0) {
diff -r 328926869b15 -r 15c9a0e16269 src/cpu/x86/vm/vm_version_x86.hpp
--- a/src/cpu/x86/vm/vm_version_x86.hpp	Sat Apr 09 22:55:25 2011 -0700
+++ b/src/cpu/x86/vm/vm_version_x86.hpp	Mon Apr 11 15:30:31 2011 -0700
@@ -188,7 +188,8 @@
      CPU_FXSR   = (1 << 2),
      CPU_HT     = (1 << 3),
      CPU_MMX    = (1 << 4),
-     CPU_3DNOW  = (1 << 5), // 3DNow comes from cpuid 0x80000001 (EDX)
+     CPU_3DNOW_PREFETCH  = (1 << 5), // Processor supports 3dnow prefetch and prefetchw instructions
+                                     // may not necessarily support other 3dnow instructions
      CPU_SSE    = (1 << 6),
      CPU_SSE2   = (1 << 7),
      CPU_SSE3   = (1 << 8), // SSE3 comes from cpuid 1 (ECX)
@@ -328,8 +329,9 @@
 
     // AMD features.
     if (is_amd()) {
-      if (_cpuid_info.ext_cpuid1_edx.bits.tdnow != 0)
-        result |= CPU_3DNOW;
+      if ((_cpuid_info.ext_cpuid1_edx.bits.tdnow != 0) ||
+          (_cpuid_info.ext_cpuid1_ecx.bits.prefetchw != 0))
+        result |= CPU_3DNOW_PREFETCH;
       if (_cpuid_info.ext_cpuid1_ecx.bits.lzcnt != 0)
         result |= CPU_LZCNT;
       if (_cpuid_info.ext_cpuid1_ecx.bits.sse4a != 0)
@@ -446,9 +448,8 @@
   //
   // AMD features
   //
-  static bool supports_3dnow()    { return (_cpuFeatures & CPU_3DNOW) != 0; }
+  static bool supports_3dnow_prefetch()    { return (_cpuFeatures & CPU_3DNOW_PREFETCH) != 0; }
   static bool supports_mmx_ext()  { return is_amd() && _cpuid_info.ext_cpuid1_edx.bits.mmx_amd != 0; }
-  static bool supports_3dnow2()   { return is_amd() && _cpuid_info.ext_cpuid1_edx.bits.tdnow2 != 0; }
   static bool supports_lzcnt()    { return (_cpuFeatures & CPU_LZCNT) != 0; }
   static bool supports_sse4a()    { return (_cpuFeatures & CPU_SSE4A) != 0; }
 
diff -r 328926869b15 -r 15c9a0e16269 src/cpu/x86/vm/x86_32.ad
--- a/src/cpu/x86/vm/x86_32.ad	Sat Apr 09 22:55:25 2011 -0700
+++ b/src/cpu/x86/vm/x86_32.ad	Mon Apr 11 15:30:31 2011 -0700
@@ -3423,7 +3423,7 @@
          masm.movptr(boxReg, tmpReg);                   // consider: LEA box, [tmp-2] 
 
          // Using a prefetchw helps avoid later RTS->RTO upgrades and cache probes
-         if ((EmitSync & 2048) && VM_Version::supports_3dnow() && os::is_MP()) {
+         if ((EmitSync & 2048) && VM_Version::supports_3dnow_prefetch() && os::is_MP()) {
             // prefetchw [eax + Offset(_owner)-2]
             masm.prefetchw(Address(rax, ObjectMonitor::owner_offset_in_bytes()-2));
          }
@@ -3467,7 +3467,7 @@
          masm.movptr(boxReg, tmpReg) ; 
 
          // Using a prefetchw helps avoid later RTS->RTO upgrades and cache probes
-         if ((EmitSync & 2048) && VM_Version::supports_3dnow() && os::is_MP()) {
+         if ((EmitSync & 2048) && VM_Version::supports_3dnow_prefetch() && os::is_MP()) {
             // prefetchw [eax + Offset(_owner)-2]
             masm.prefetchw(Address(rax, ObjectMonitor::owner_offset_in_bytes()-2));
          }
@@ -3614,7 +3614,7 @@
       // See also http://gee.cs.oswego.edu/dl/jmm/cookbook.html.
 
       masm.get_thread (boxReg) ;
-      if ((EmitSync & 4096) && VM_Version::supports_3dnow() && os::is_MP()) {
+      if ((EmitSync & 4096) && VM_Version::supports_3dnow_prefetch() && os::is_MP()) {
         // prefetchw [ebx + Offset(_owner)-2]
         masm.prefetchw(Address(rbx, ObjectMonitor::owner_offset_in_bytes()-2));
       }
@@ -7333,7 +7333,7 @@
 // Must be safe to execute with invalid address (cannot fault).
 
 instruct prefetchr0( memory mem ) %{
-  predicate(UseSSE==0 && !VM_Version::supports_3dnow());
+  predicate(UseSSE==0 && !VM_Version::supports_3dnow_prefetch());
   match(PrefetchRead mem);
   ins_cost(0);
   size(0);
@@ -7343,7 +7343,7 @@
 %}
 
 instruct prefetchr( memory mem ) %{
-  predicate(UseSSE==0 && VM_Version::supports_3dnow() || ReadPrefetchInstr==3);
+  predicate(UseSSE==0 && VM_Version::supports_3dnow_prefetch() || ReadPrefetchInstr==3);
   match(PrefetchRead mem);
   ins_cost(100);
 
@@ -7387,7 +7387,7 @@
 %}
 
 instruct prefetchw0( memory mem ) %{
-  predicate(UseSSE==0 && !VM_Version::supports_3dnow());
+  predicate(UseSSE==0 && !VM_Version::supports_3dnow_prefetch());
   match(PrefetchWrite mem);
   ins_cost(0);
   size(0);
@@ -7397,7 +7397,7 @@
 %}
 
 instruct prefetchw( memory mem ) %{
-  predicate(UseSSE==0 && VM_Version::supports_3dnow() || AllocatePrefetchInstr==3);
+  predicate(UseSSE==0 && VM_Version::supports_3dnow_prefetch() || AllocatePrefetchInstr==3);
   match( PrefetchWrite mem );
   ins_cost(100);