Reverts refactoring of CPU feature detection for

runtime/vm/assembler_ia32.cc

 // Copyright (c) 2013, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 
 #include "vm/globals.h"
 #if defined(TARGET_ARCH_IA32)
 
 #include "vm/assembler.h"
 #include "vm/code_generator.h"
-#include "vm/cpu.h"
 #include "vm/heap.h"
 #include "vm/memory_region.h"
 #include "vm/runtime_entry.h"
 #include "vm/stack_frame.h"
 #include "vm/stub_code.h"
 
 namespace dart {
 
 DEFINE_FLAG(bool, print_stop_message, true, "Print stop message.");
+DEFINE_FLAG(bool, use_sse41, true, "Use SSE 4.1 if available");
 DECLARE_FLAG(bool, inline_alloc);
 
 
+bool CPUFeatures::sse2_supported_ = false;
+bool CPUFeatures::sse4_1_supported_ = false;
+#ifdef DEBUG
+bool CPUFeatures::initialized_ = false;
+#endif
+
+
+bool CPUFeatures::sse2_supported() {
+  DEBUG_ASSERT(initialized_);
+  return sse2_supported_;
+}
+
+
+bool CPUFeatures::sse4_1_supported() {
+  DEBUG_ASSERT(initialized_);
+  return sse4_1_supported_ && FLAG_use_sse41;
+}
+
+
+#define __ assembler.
+
+void CPUFeatures::InitOnce() {
+  Assembler assembler;
+  __ pushl(EBP);
+  __ pushl(EBX);
+  __ movl(EBP, ESP);
+  // Get feature information in ECX:EDX and return it in EDX:EAX.
+  __ movl(EAX, Immediate(1));
+  __ cpuid();
+  __ movl(EAX, EDX);
+  __ movl(EDX, ECX);
+  __ movl(ESP, EBP);
+  __ popl(EBX);
+  __ popl(EBP);
+  __ ret();
+
+  const Code& code =
+      Code::Handle(Code::FinalizeCode("DetectCPUFeatures", &assembler));
+  Instructions& instructions = Instructions::Handle(code.instructions());
+  typedef uint64_t (*DetectCPUFeatures)();
+  uint64_t features =
+      reinterpret_cast<DetectCPUFeatures>(instructions.EntryPoint())();
+  sse2_supported_ = (features & kSSE2BitMask) != 0;
+  sse4_1_supported_ = (features & kSSE4_1BitMask) != 0;
+#ifdef DEBUG
+  initialized_ = true;
+#endif
+}
+
+#undef __
+
+
 class DirectCallRelocation : public AssemblerFixup {
  public:
   void Process(const MemoryRegion& region, intptr_t position) {
     // Direct calls are relative to the following instruction on x86.
     int32_t pointer = region.Load<int32_t>(position);
     int32_t delta = region.start() + position + sizeof(int32_t);
     region.Store<int32_t>(position, pointer - delta);
   }
 
   virtual bool IsPointerOffset() const { return false; }
@@ skipping 1130 matching lines @@
 void Assembler::andpd(XmmRegister dst, XmmRegister src) {
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0x66);
   EmitUint8(0x0F);
   EmitUint8(0x54);
   EmitXmmRegisterOperand(dst, src);
 }
 
 
 void Assembler::pextrd(Register dst, XmmRegister src, const Immediate& imm) {
-  ASSERT(TargetCPUFeatures::sse4_1_supported());
+  ASSERT(CPUFeatures::sse4_1_supported());
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0x66);
   EmitUint8(0x0F);
   EmitUint8(0x3A);
   EmitUint8(0x16);
   EmitOperand(src, Operand(dst));
   ASSERT(imm.is_uint8());
   EmitUint8(imm.value());
 }
 
 
 void Assembler::pmovsxdq(XmmRegister dst, XmmRegister src) {
-  ASSERT(TargetCPUFeatures::sse4_1_supported());
+  ASSERT(CPUFeatures::sse4_1_supported());
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0x66);
   EmitUint8(0x0F);
   EmitUint8(0x38);
   EmitUint8(0x25);
   EmitXmmRegisterOperand(dst, src);
 }
 
 
 void Assembler::pcmpeqq(XmmRegister dst, XmmRegister src) {
-  ASSERT(TargetCPUFeatures::sse4_1_supported());
+  ASSERT(CPUFeatures::sse4_1_supported());
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0x66);
   EmitUint8(0x0F);
   EmitUint8(0x38);
   EmitUint8(0x29);
   EmitXmmRegisterOperand(dst, src);
 }
 
 
 void Assembler::pxor(XmmRegister dst, XmmRegister src) {
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0x66);
   EmitUint8(0x0F);
   EmitUint8(0xEF);
   EmitXmmRegisterOperand(dst, src);
 }
 
 
 void Assembler::roundsd(XmmRegister dst, XmmRegister src, RoundingMode mode) {
-  ASSERT(TargetCPUFeatures::sse4_1_supported());
+  ASSERT(CPUFeatures::sse4_1_supported());
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0x66);
   EmitUint8(0x0F);
   EmitUint8(0x3A);
   EmitUint8(0x0B);
   EmitXmmRegisterOperand(dst, src);
   // Mask precision exeption.
   EmitUint8(static_cast<uint8_t>(mode) | 0x8);
 }
 
@@ skipping 1420 matching lines @@
 
 const char* Assembler::FpuRegisterName(FpuRegister reg) {
   ASSERT((0 <= reg) && (reg < kNumberOfXmmRegisters));
   return xmm_reg_names[reg];
 }
 
 
 }  // namespace dart
 
 #endif  // defined TARGET_ARCH_IA32