[wasm-simd][ia32] Unify SSE and AVX impl for i8x16.shl

Delegate to macro assembler functions to pick between SSE or AVX. This
simplifies the optimization of constant shifts later.

Bug: v8:10115
Change-Id: If0b7dc83a68f2d7839c65527a3b6a62310ace6dd
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2103443Reviewed-by: Deepti Gandluri <gdeepti@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#66753}

src/codegen/ia32/macro-assembler-ia32.h

@@ -305,6 +305,7 @@ class V8_EXPORT_PRIVATE TurboAssembler : public TurboAssemblerBase {
   AVX_OP3_XO(Packsswb, packsswb)
   AVX_OP3_XO(Packuswb, packuswb)
   AVX_OP3_XO(Paddusb, paddusb)
+  AVX_OP3_XO(Pand, pand)
   AVX_OP3_XO(Pcmpeqb, pcmpeqb)
   AVX_OP3_XO(Pcmpeqw, pcmpeqw)
   AVX_OP3_XO(Pcmpeqd, pcmpeqd)

src/compiler/backend/ia32/code-generator-ia32.cc

@@ -3146,7 +3146,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       }
       break;
     }
-    case kSSEI8x16Shl: {
+    case kIA32I8x16Shl: {
       XMMRegister dst = i.OutputSimd128Register();
       DCHECK_EQ(dst, i.InputSimd128Register(0));
       Register shift = i.InputRegister(1);
@@ -3155,36 +3155,15 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       // Take shift value modulo 8.
       __ and_(shift, 7);
       // Mask off the unwanted bits before word-shifting.
-      __ pcmpeqw(kScratchDoubleReg, kScratchDoubleReg);
-      __ mov(tmp, shift);
-      __ add(tmp, Immediate(8));
-      __ movd(tmp_simd, tmp);
-      __ psrlw(kScratchDoubleReg, tmp_simd);
-      __ packuswb(kScratchDoubleReg, kScratchDoubleReg);
-      __ pand(dst, kScratchDoubleReg);
-      __ movd(tmp_simd, shift);
-      __ psllw(dst, tmp_simd);
-      break;
-    }
-    case kAVXI8x16Shl: {
-      CpuFeatureScope avx_scope(tasm(), AVX);
-      XMMRegister dst = i.OutputSimd128Register();
-      XMMRegister src = i.InputSimd128Register(0);
-      Register shift = i.InputRegister(1);
-      Register tmp = i.ToRegister(instr->TempAt(0));
-      XMMRegister tmp_simd = i.TempSimd128Register(1);
-      // Take shift value modulo 8.
-      __ and_(shift, 7);
-      // Mask off the unwanted bits before word-shifting.
-      __ vpcmpeqw(kScratchDoubleReg, kScratchDoubleReg, kScratchDoubleReg);
+      __ Pcmpeqw(kScratchDoubleReg, kScratchDoubleReg);
       __ mov(tmp, shift);
       __ add(tmp, Immediate(8));
-      __ movd(tmp_simd, tmp);
-      __ vpsrlw(kScratchDoubleReg, kScratchDoubleReg, tmp_simd);
-      __ vpackuswb(kScratchDoubleReg, kScratchDoubleReg, kScratchDoubleReg);
-      __ vpand(dst, src, kScratchDoubleReg);
-      __ movd(tmp_simd, shift);
-      __ vpsllw(dst, dst, tmp_simd);
+      __ Movd(tmp_simd, tmp);
+      __ Psrlw(kScratchDoubleReg, kScratchDoubleReg, tmp_simd);
+      __ Packuswb(kScratchDoubleReg, kScratchDoubleReg);
+      __ Pand(dst, kScratchDoubleReg);
+      __ Movd(tmp_simd, shift);
+      __ Psllw(dst, dst, tmp_simd);
       break;
     }
     case kIA32I8x16ShrS: {

src/compiler/backend/ia32/instruction-codes-ia32.h

@@ -289,8 +289,7 @@ namespace compiler {
   V(SSEI8x16SConvertI16x8)         \
   V(AVXI8x16SConvertI16x8)         \
   V(IA32I8x16Neg)                  \
-  V(SSEI8x16Shl)                   \
-  V(AVXI8x16Shl)                   \
+  V(IA32I8x16Shl)                  \
   V(IA32I8x16ShrS)                 \
   V(SSEI8x16Add)                   \
   V(AVXI8x16Add)                   \

src/compiler/backend/ia32/instruction-scheduler-ia32.cc

@@ -270,8 +270,7 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kSSEI8x16SConvertI16x8:
     case kAVXI8x16SConvertI16x8:
     case kIA32I8x16Neg:
-    case kSSEI8x16Shl:
-    case kAVXI8x16Shl:
+    case kIA32I8x16Shl:
     case kIA32I8x16ShrS:
     case kSSEI8x16Add:
     case kAVXI8x16Add:

src/compiler/backend/ia32/instruction-selector-ia32.cc

@@ -2451,13 +2451,8 @@ void InstructionSelector::VisitI8x16Shl(Node* node) {
   InstructionOperand operand0 = g.UseUniqueRegister(node->InputAt(0));
   InstructionOperand operand1 = g.UseUniqueRegister(node->InputAt(1));
   InstructionOperand temps[] = {g.TempRegister(), g.TempSimd128Register()};
-  if (IsSupported(AVX)) {
-    Emit(kAVXI8x16Shl, g.DefineAsRegister(node), operand0, operand1,
-         arraysize(temps), temps);
-  } else {
-    Emit(kSSEI8x16Shl, g.DefineSameAsFirst(node), operand0, operand1,
-         arraysize(temps), temps);
-  }
+  Emit(kIA32I8x16Shl, g.DefineSameAsFirst(node), operand0, operand1,
+       arraysize(temps), temps);
 }
 
 void InstructionSelector::VisitInt32AbsWithOverflow(Node* node) {