[x86] Allow (v)divsd->(v)mulsd to execute in parallel.

src/compiler/ia32/code-generator-ia32.cc

 // Copyright 2013 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "src/compiler/code-generator.h"
 
 #include "src/compiler/code-generator-impl.h"
 #include "src/compiler/gap-resolver.h"
 #include "src/compiler/node-matchers.h"
 #include "src/ia32/assembler-ia32.h"
@@ skipping 450 matching lines @@
       __ addss(i.InputDoubleRegister(0), i.InputOperand(1));
       break;
     case kSSEFloat32Sub:
       __ subss(i.InputDoubleRegister(0), i.InputOperand(1));
       break;
     case kSSEFloat32Mul:
       __ mulss(i.InputDoubleRegister(0), i.InputOperand(1));
       break;
     case kSSEFloat32Div:
       __ divss(i.InputDoubleRegister(0), i.InputOperand(1));
+      // Don't delete this mov. It may improve performance on some CPUs,
+      // when there is a (v)mulss depending on the result.
+      __ movaps(i.OutputDoubleRegister(), i.OutputDoubleRegister());
       break;
     case kSSEFloat32Max:
       __ maxss(i.InputDoubleRegister(0), i.InputOperand(1));
       break;
     case kSSEFloat32Min:
       __ minss(i.InputDoubleRegister(0), i.InputOperand(1));
       break;
     case kSSEFloat32Sqrt:
       __ sqrtss(i.OutputDoubleRegister(), i.InputOperand(0));
       break;
@@ skipping 18 matching lines @@
       __ addsd(i.InputDoubleRegister(0), i.InputOperand(1));
       break;
     case kSSEFloat64Sub:
       __ subsd(i.InputDoubleRegister(0), i.InputOperand(1));
       break;
     case kSSEFloat64Mul:
       __ mulsd(i.InputDoubleRegister(0), i.InputOperand(1));
       break;
     case kSSEFloat64Div:
       __ divsd(i.InputDoubleRegister(0), i.InputOperand(1));
+      // Don't delete this mov. It may improve performance on some CPUs,
+      // when there is a (v)mulsd depending on the result.
+      __ movaps(i.OutputDoubleRegister(), i.OutputDoubleRegister());
       break;
     case kSSEFloat64Max:
       __ maxsd(i.InputDoubleRegister(0), i.InputOperand(1));
       break;
     case kSSEFloat64Min:
       __ minsd(i.InputDoubleRegister(0), i.InputOperand(1));
       break;
     case kSSEFloat64Mod: {
       // TODO(dcarney): alignment is wrong.
       __ sub(esp, Immediate(kDoubleSize));
@@ skipping 103 matching lines @@
     case kAVXFloat32Mul: {
       CpuFeatureScope avx_scope(masm(), AVX);
       __ vmulss(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
                 i.InputOperand(1));
       break;
     }
     case kAVXFloat32Div: {
       CpuFeatureScope avx_scope(masm(), AVX);
       __ vdivss(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
                 i.InputOperand(1));
+      // Don't delete this mov. It may improve performance on some CPUs,
+      // when there is a (v)mulss depending on the result.
+      __ movaps(i.OutputDoubleRegister(), i.OutputDoubleRegister());
       break;
     }
     case kAVXFloat32Max: {
       CpuFeatureScope avx_scope(masm(), AVX);
       __ vmaxss(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
                 i.InputOperand(1));
       break;
     }
     case kAVXFloat32Min: {
       CpuFeatureScope avx_scope(masm(), AVX);
@@ skipping 16 matching lines @@
     case kAVXFloat64Mul: {
       CpuFeatureScope avx_scope(masm(), AVX);
       __ vmulsd(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
                 i.InputOperand(1));
       break;
     }
     case kAVXFloat64Div: {
       CpuFeatureScope avx_scope(masm(), AVX);
       __ vdivsd(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
                 i.InputOperand(1));
+      // Don't delete this mov. It may improve performance on some CPUs,
+      // when there is a (v)mulsd depending on the result.
+      __ movaps(i.OutputDoubleRegister(), i.OutputDoubleRegister());
       break;
     }
     case kAVXFloat64Max: {
       CpuFeatureScope avx_scope(masm(), AVX);
       __ vmaxsd(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
                 i.InputOperand(1));
       break;
     }
     case kAVXFloat64Min: {
       CpuFeatureScope avx_scope(masm(), AVX);
@@ skipping 794 matching lines @@
     }
   }
   MarkLazyDeoptSite();
 }
 
 #undef __
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8