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

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

Index: src/compiler/ia32/code-generator-ia32.cc
diff --git a/src/compiler/ia32/code-generator-ia32.cc b/src/compiler/ia32/code-generator-ia32.cc
index aaa34dcfd66f3dcc58fc70edc50f044651074dbf..76a457b8f6645d82b8c97753511fe4777df111e4 100644
--- a/src/compiler/ia32/code-generator-ia32.cc
+++ b/src/compiler/ia32/code-generator-ia32.cc
@@ -468,6 +468,9 @@ void CodeGenerator::AssembleArchInstruction(Instruction* instr) {
       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));
@@ -506,6 +509,9 @@ void CodeGenerator::AssembleArchInstruction(Instruction* instr) {
       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));
@@ -629,6 +635,9 @@ void CodeGenerator::AssembleArchInstruction(Instruction* instr) {
       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: {
@@ -665,6 +674,9 @@ void CodeGenerator::AssembleArchInstruction(Instruction* instr) {
       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: {