Add Int32x4 Wasm Simd Binops, compare ops, select

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 8f9b69d2c8b033f7654bbc6f53bd4b5d23934667..c73b03bf20c3c70fee082c42483e857434a7a0d1 100644
--- a/src/compiler/x64/code-generator-x64.cc
+++ b/src/compiler/x64/code-generator-x64.cc
@@ -2155,10 +2155,9 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       break;
     }
     case kX64Int32x4Splat: {
-      CpuFeatureScope sse_scope(masm(), SSE4_1);
       XMMRegister dst = i.OutputSimd128Register();
       __ Movd(dst, i.InputRegister(0));

[bbudge, 2017/02/28 19:34:39]

Still might use AVX.

[gdeepti, 2017/03/13 20:37:52]

Fixed now.

-      __ shufps(dst, dst, 0x0);
+      __ pshufd(dst, dst, 0x0);
       break;
     }
     case kX64Int32x4ExtractLane: {
@@ -2177,17 +2176,68 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       break;
     }
     case kX64Int32x4Add: {
-      CpuFeatureScope sse_scope(masm(), SSE4_1);
       __ paddd(i.OutputSimd128Register(), i.InputSimd128Register(1));
       break;
     }
     case kX64Int32x4Sub: {
-      CpuFeatureScope sse_scope(masm(), SSE4_1);
       __ psubd(i.OutputSimd128Register(), i.InputSimd128Register(1));
       break;
     }
-    case kX64Simd128Zero: {
+    case kX64Int32x4Mul: {
       CpuFeatureScope sse_scope(masm(), SSE4_1);
+      __ pmulld(i.OutputSimd128Register(), i.InputSimd128Register(1));
+      break;
+    }
+    case kX64Int32x4Min: {
+      CpuFeatureScope sse_scope(masm(), SSE4_1);
+      __ pminsd(i.OutputSimd128Register(), i.InputSimd128Register(1));
+      break;
+    }
+    case kX64Int32x4Max: {
+      CpuFeatureScope sse_scope(masm(), SSE4_1);
+      __ pmaxsd(i.OutputSimd128Register(), i.InputSimd128Register(1));
+      break;
+    }
+    case kX64Uint32x4Min: {
+      CpuFeatureScope sse_scope(masm(), SSE4_1);
+      __ pminud(i.OutputSimd128Register(), i.InputSimd128Register(1));
+      break;
+    }
+    case kX64Uint32x4Max: {
+      CpuFeatureScope sse_scope(masm(), SSE4_1);
+      __ pmaxud(i.OutputSimd128Register(), i.InputSimd128Register(1));
+      break;
+    }
+    case kX64Int32x4Equal: {
+      __ pcmpeqd(i.OutputSimd128Register(), i.InputSimd128Register(1));
+      break;
+    }
+    case kX64Int32x4NotEqual: {
+      __ pcmpeqd(i.OutputSimd128Register(), i.InputSimd128Register(1));
+      __ pcmpeqd(kScratchDoubleReg, kScratchDoubleReg);
+      __ pxor(i.OutputSimd128Register(), kScratchDoubleReg);
+      break;
+    }
+    case kX64Int32x4ShiftLeftByScalar: {
+      __ pslld(i.OutputSimd128Register(), i.InputInt8(1));
+      break;
+    }
+    case kX64Int32x4ShiftRightByScalar: {
+      __ psrad(i.OutputSimd128Register(), i.InputInt8(1));
+      break;
+    }
+    case kX64Uint32x4ShiftRightByScalar: {
+      __ psrld(i.OutputSimd128Register(), i.InputInt8(1));
+      break;
+    }
+    case kX64Simd32x4Select: {
+      XMMRegister dst = i.OutputSimd128Register();

[bbudge, 2017/02/28 19:34:39]

A comment here that dst contains the mask initially.

[gdeepti, 2017/03/13 20:37:52]

Done.

+      __ xorps(i.InputSimd128Register(1), i.InputSimd128Register(2));

[bbudge, 2017/02/28 19:34:39]

Maybe use the scratch FP register for this? Otherwise, we overwrite the first
input.

[gdeepti, 2017/03/13 20:37:52]

Done.

+      __ andps(dst, i.InputSimd128Register(1));
+      __ xorps(dst, i.InputSimd128Register(2));
+      break;
+    }
+    case kX64Simd128Zero: {
       XMMRegister dst = i.OutputSimd128Register();
       __ xorps(dst, dst);
       break;