[wasm-simd][x64] Optimize f32x4 splat and extract lanes

For splats, we can make use of vshufps to avoid a movss. Without
AVX, specific dst to be same as src in the instruction selector.

For extract lane, we can use vshufps to extract a float into a dst xmm,
and leave junk in the higher bits.

On the meshopt_decoder.js benchmark in linked bug, it removes about 7
movss instructions that did nothing. Hardware can do register renaming,
but let's not rely on that :)

R=bbudge@chromium.org

Bug: v8:10116
Change-Id: I4d68c10536a79659de673060d537d58113308477
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2481473
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Reviewed-by: Bill Budge <bbudge@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70628}

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

@@ -2494,21 +2494,29 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       }
       break;
     }
-    // TODO(gdeepti): Get rid of redundant moves for F32x4Splat/Extract below
     case kX64F32x4Splat: {
       XMMRegister dst = i.OutputSimd128Register();
-      if (instr->InputAt(0)->IsFPRegister()) {
-        __ Movss(dst, i.InputDoubleRegister(0));
+      XMMRegister src = i.InputDoubleRegister(0);
+      if (CpuFeatures::IsSupported(AVX)) {
+        CpuFeatureScope avx_scope(tasm(), AVX);
+        __ vshufps(dst, src, src, byte{0x0});
       } else {
-        __ Movss(dst, i.InputOperand(0));
-      }
+        DCHECK_EQ(dst, src);
         __ Shufps(dst, dst, byte{0x0});
+      }
       break;
     }
     case kX64F32x4ExtractLane: {
-      __ Extractps(kScratchRegister, i.InputSimd128Register(0),
+      if (CpuFeatures::IsSupported(AVX)) {
+        CpuFeatureScope avx_scope(tasm(), AVX);
+        XMMRegister src = i.InputSimd128Register(0);
+        // vshufps and leave junk in the 3 high lanes.
+        __ vshufps(i.OutputDoubleRegister(), src, src, i.InputInt8(1));
+      } else {
+        __ extractps(kScratchRegister, i.InputSimd128Register(0),
                      i.InputUint8(1));
-      __ Movd(i.OutputDoubleRegister(), kScratchRegister);
+        __ movd(i.OutputDoubleRegister(), kScratchRegister);
+      }
       break;
     }
     case kX64F32x4ReplaceLane: {

src/compiler/backend/x64/instruction-selector-x64.cc

@@ -2796,14 +2796,6 @@ VISIT_ATOMIC_BINOP(Or)
 VISIT_ATOMIC_BINOP(Xor)
 #undef VISIT_ATOMIC_BINOP
 
-#define SIMD_TYPES(V) \
-  V(F64x2)            \
-  V(F32x4)            \
-  V(I64x2)            \
-  V(I32x4)            \
-  V(I16x8)            \
-  V(I8x16)
-
 #define SIMD_BINOP_SSE_AVX_LIST(V) \
   V(F32x4Add)                      \
   V(F32x4Sub)                      \
@@ -2967,14 +2959,29 @@ void InstructionSelector::VisitS128Zero(Node* node) {
   Emit(kX64S128Zero, g.DefineAsRegister(node));
 }
 
+#define SIMD_TYPES_FOR_SPLAT(V) \
+  V(F64x2)                      \
+  V(I64x2)                      \
+  V(I32x4)                      \
+  V(I16x8)                      \
+  V(I8x16)
+
 #define VISIT_SIMD_SPLAT(Type)                               \
   void InstructionSelector::Visit##Type##Splat(Node* node) { \
     X64OperandGenerator g(this);                             \
     Emit(kX64##Type##Splat, g.DefineAsRegister(node),        \
          g.Use(node->InputAt(0)));                           \
   }
-SIMD_TYPES(VISIT_SIMD_SPLAT)
+SIMD_TYPES_FOR_SPLAT(VISIT_SIMD_SPLAT)
 #undef VISIT_SIMD_SPLAT
+#undef SIMD_TYPES_FOR_SPLAT
+
+void InstructionSelector::VisitF32x4Splat(Node* node) {
+  X64OperandGenerator g(this);
+  InstructionOperand dst =
+      IsSupported(AVX) ? g.DefineAsRegister(node) : g.DefineSameAsFirst(node);
+  Emit(kX64F32x4Splat, dst, g.UseRegister(node->InputAt(0)));
+}
 
 #define SIMD_VISIT_EXTRACT_LANE(Type, Sign)                              \
   void InstructionSelector::Visit##Type##ExtractLane##Sign(Node* node) { \
@@ -3124,7 +3131,6 @@ SIMD_ANYTRUE_LIST(VISIT_SIMD_ANYTRUE)
 SIMD_ALLTRUE_LIST(VISIT_SIMD_ALLTRUE)
 #undef VISIT_SIMD_ALLTRUE
 #undef SIMD_ALLTRUE_LIST
-#undef SIMD_TYPES
 
 void InstructionSelector::VisitS128Select(Node* node) {
   X64OperandGenerator g(this);