[x64] Replace movaps with appropriate vmov* instructions when AVX is enabled.

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/compiler/osr.h"
@@ skipping 831 matching lines @@
     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());
+      __ Movapd(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 134 matching lines @@
     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());
+      __ 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());
+      __ Movapd(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 684 matching lines @@
       } else {
         DCHECK(destination->IsDoubleStackSlot());
         __ movq(kScratchRegister, src_const);
         __ movq(g.ToOperand(destination), kScratchRegister);
       }
     }
   } else if (source->IsDoubleRegister()) {
     XMMRegister src = g.ToDoubleRegister(source);
     if (destination->IsDoubleRegister()) {
       XMMRegister dst = g.ToDoubleRegister(destination);
-      __ movaps(dst, src);
+      __ Movapd(dst, src);
     } else {
       DCHECK(destination->IsDoubleStackSlot());
       Operand dst = g.ToOperand(destination);
       __ Movsd(dst, src);
     }
   } else if (source->IsDoubleStackSlot()) {
     DCHECK(destination->IsDoubleRegister() || destination->IsDoubleStackSlot());
     Operand src = g.ToOperand(source);
     if (destination->IsDoubleRegister()) {
       XMMRegister dst = g.ToDoubleRegister(destination);
@@ skipping 30 matching lines @@
     Operand src = g.ToOperand(source);
     Operand dst = g.ToOperand(destination);
     __ movq(tmp, dst);
     __ xchgq(tmp, src);
     __ movq(dst, tmp);
   } else if (source->IsDoubleRegister() && destination->IsDoubleRegister()) {
     // XMM register-register swap. We rely on having xmm0
     // available as a fixed scratch register.
     XMMRegister src = g.ToDoubleRegister(source);
     XMMRegister dst = g.ToDoubleRegister(destination);
-    __ movaps(xmm0, src);
-    __ movaps(src, dst);
-    __ movaps(dst, xmm0);
+    __ Movapd(xmm0, src);
+    __ Movapd(src, dst);
+    __ Movapd(dst, xmm0);
   } else if (source->IsDoubleRegister() && destination->IsDoubleStackSlot()) {
     // XMM register-memory swap.  We rely on having xmm0
     // available as a fixed scratch register.
     XMMRegister src = g.ToDoubleRegister(source);
     Operand dst = g.ToOperand(destination);
     __ Movsd(xmm0, src);
     __ Movsd(src, dst);
     __ Movsd(dst, xmm0);
   } else {
     // No other combinations are possible.
@@ skipping 25 matching lines @@
     int padding_size = last_lazy_deopt_pc_ + space_needed - current_pc;
     __ Nop(padding_size);
   }
 }
 
 #undef __
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8