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

src/compiler/x64/code-generator-x64.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/scopes.h"
@@ skipping 708 matching lines @@
       ASSEMBLE_SSE_BINOP(addss);
       break;
     case kSSEFloat32Sub:
       ASSEMBLE_SSE_BINOP(subss);
       break;
     case kSSEFloat32Mul:
       ASSEMBLE_SSE_BINOP(mulss);
       break;
     case kSSEFloat32Div:
       ASSEMBLE_SSE_BINOP(divss);
+      // 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 kSSEFloat32Abs: {
       // TODO(bmeurer): Use RIP relative 128-bit constants.
       __ pcmpeqd(kScratchDoubleReg, kScratchDoubleReg);
       __ psrlq(kScratchDoubleReg, 33);
       __ andps(i.OutputDoubleRegister(), kScratchDoubleReg);
       break;
     }
     case kSSEFloat32Neg: {
       // TODO(bmeurer): Use RIP relative 128-bit constants.
@@ skipping 21 matching lines @@
       ASSEMBLE_SSE_BINOP(addsd);
       break;
     case kSSEFloat64Sub:
       ASSEMBLE_SSE_BINOP(subsd);
       break;
     case kSSEFloat64Mul:
       ASSEMBLE_SSE_BINOP(mulsd);
       break;
     case kSSEFloat64Div:
       ASSEMBLE_SSE_BINOP(divsd);
+      // 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 kSSEFloat64Mod: {
       __ subq(rsp, Immediate(kDoubleSize));
       // Move values to st(0) and st(1).
       __ movsd(Operand(rsp, 0), i.InputDoubleRegister(1));
       __ fld_d(Operand(rsp, 0));
       __ movsd(Operand(rsp, 0), i.InputDoubleRegister(0));
       __ fld_d(Operand(rsp, 0));
       // Loop while fprem isn't done.
       Label mod_loop;
@@ skipping 132 matching lines @@
       ASSEMBLE_AVX_BINOP(vaddss);
       break;
     case kAVXFloat32Sub:
       ASSEMBLE_AVX_BINOP(vsubss);
       break;
     case kAVXFloat32Mul:
       ASSEMBLE_AVX_BINOP(vmulss);
       break;
     case kAVXFloat32Div:
       ASSEMBLE_AVX_BINOP(vdivss);
+      // 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:
       ASSEMBLE_AVX_BINOP(vmaxss);
       break;
     case kAVXFloat32Min:
       ASSEMBLE_AVX_BINOP(vminss);
       break;
     case kAVXFloat64Cmp: {
       CpuFeatureScope avx_scope(masm(), AVX);
       if (instr->InputAt(1)->IsDoubleRegister()) {
         __ vucomisd(i.InputDoubleRegister(0), i.InputDoubleRegister(1));
       } else {
         __ vucomisd(i.InputDoubleRegister(0), i.InputOperand(1));
       }
       break;
     }
     case kAVXFloat64Add:
       ASSEMBLE_AVX_BINOP(vaddsd);
       break;
     case kAVXFloat64Sub:
       ASSEMBLE_AVX_BINOP(vsubsd);
       break;
     case kAVXFloat64Mul:
       ASSEMBLE_AVX_BINOP(vmulsd);
       break;
     case kAVXFloat64Div:
       ASSEMBLE_AVX_BINOP(vdivsd);
+      // 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:
       ASSEMBLE_AVX_BINOP(vmaxsd);
       break;
     case kAVXFloat64Min:
       ASSEMBLE_AVX_BINOP(vminsd);
       break;
     case kAVXFloat32Abs: {
       // TODO(bmeurer): Use RIP relative 128-bit constants.
       __ pcmpeqd(kScratchDoubleReg, kScratchDoubleReg);
@@ skipping 699 matching lines @@
     }
   }
   MarkLazyDeoptSite();
 }
 
 #undef __
 
 }  // namespace internal
 }  // namespace compiler
 }  // namespace v8