[wasm] Add native x64 implementations for I32x4Splat, I32x4ExtractLane

src/compiler/x64/code-generator-x64.cc

Index: src/compiler/x64/code-generator-x64.cc
diff --git a/src/compiler/x64/code-generator-x64.cc b/src/compiler/x64/code-generator-x64.cc
index 970f85d8d3dcc40cf63327d64c30ce0b816bb151..7b08d195911bf7f6d6d26a79b60eef6841702b63 100644
--- a/src/compiler/x64/code-generator-x64.cc
+++ b/src/compiler/x64/code-generator-x64.cc
@@ -1746,6 +1746,18 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       __ 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());