[wasm] Add native x64 implementations for I32x4Splat, I32x4ExtractLane

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/ast/scopes.h"
 #include "src/compiler/code-generator-impl.h"
 #include "src/compiler/gap-resolver.h"
 #include "src/compiler/node-matchers.h"
@@ skipping 1728 matching lines @@
       } else {
         __ vxorpd(i.OutputDoubleRegister(), kScratchDoubleReg,
                   i.InputOperand(0));
       }
       break;
     }
     case kSSEFloat64SilenceNaN:
       __ Xorpd(kScratchDoubleReg, kScratchDoubleReg);
       __ Subsd(i.InputDoubleRegister(0), kScratchDoubleReg);
       break;
+    case kSSECreateInt32x4: {

[bbudge, 2016/08/22 19:09:52]

Would this be the AVX implementation too? In that case, it would be best if we
didn't have kSSE* values, i.e. this would be kX64CreateInt32x4.

[gdeepti, 2016/08/22 23:26:11]

I was conflicted about calling these kX64 ops because vshufps/vpextrd/vextractps
are not implemented in the x64 assembler, so these are essentially SSE ops. 

+      CpuFeatureScope sse_scope(masm(), SSE4_1);
+      XMMRegister dst = i.OutputSimd128Register();
+      __ movd(dst, i.InputRegister(0));
+      __ shufps(dst, dst, 0x0);
+      break;
+    }
+    case kSSEInt32x4ExtractLane: {
+      CpuFeatureScope sse_scope(masm(), SSE4_1);
+      __ pextrd(i.OutputRegister(), i.InputSimd128Register(0), i.InputInt8(1));

[bbudge, 2016/08/22 19:09:52]

Rather than defining kSSE and kAVX values, I think the desired pattern going
forward is to use the macro-assembler to handle this. So call MacroAssembler
Pextrd here:

https://cs.chromium.org/chromium/src/v8/src/x64/macro-assembler-x64.cc?rcl=0&...

Then rename the value kX64Int32x4ExtractLane.

[gdeepti, 2016/08/22 23:26:11]

Changed to kX64Int32x4ExtractLane, kX64CreateInt32x4, calling Pextrd, Movd in
the macro-assembler to fit the desired pattern. Shufps does not have a macro
assembler equivalent that switches between vshufps and shufps.  

+      break;
+    }
     case kX64Movsxbl:
       ASSEMBLE_MOVX(movsxbl);
       __ AssertZeroExtended(i.OutputRegister());
       break;
     case kX64Movzxbl:
       ASSEMBLE_MOVX(movzxbl);
       __ AssertZeroExtended(i.OutputRegister());
       break;
     case kX64Movsxbq:
       ASSEMBLE_MOVX(movsxbq);
@@ skipping 847 matching lines @@
     int padding_size = last_lazy_deopt_pc_ + space_needed - current_pc;
     __ Nop(padding_size);
   }
 }
 
 #undef __
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8