[wasm-simd] Implement f64x2 sqrt for ia32

Bug: v8:9728
Change-Id: Ic15d793e6408af1ea2e1f7f71b9130300d359a95
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1808417
Reviewed-by: Deepti Gandluri <gdeepti@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#64073}

src/codegen/ia32/assembler-ia32.h

@@ -882,6 +882,10 @@ class V8_EXPORT_PRIVATE Assembler : public AssemblerBase {
   void rsqrtps(XMMRegister dst, XMMRegister src) { rsqrtps(dst, Operand(src)); }
   void haddps(XMMRegister dst, Operand src);
   void haddps(XMMRegister dst, XMMRegister src) { haddps(dst, Operand(src)); }
+  void sqrtpd(XMMRegister dst, Operand src) {
+    sse2_instr(dst, src, 0x66, 0x0F, 0x51);
+  }
+  void sqrtpd(XMMRegister dst, XMMRegister src) { sqrtpd(dst, Operand(src)); }
 
   void minps(XMMRegister dst, Operand src);
   void minps(XMMRegister dst, XMMRegister src) { minps(dst, Operand(src)); }
@@ -1318,6 +1322,10 @@ class V8_EXPORT_PRIVATE Assembler : public AssemblerBase {
   void vhaddps(XMMRegister dst, XMMRegister src1, Operand src2) {
     vinstr(0x7C, dst, src1, src2, kF2, k0F, kWIG);
   }
+  void vsqrtpd(XMMRegister dst, XMMRegister src) { vsqrtpd(dst, Operand(src)); }
+  void vsqrtpd(XMMRegister dst, Operand src) {
+    vinstr(0x51, dst, xmm0, src, k66, k0F, kWIG);
+  }
   void vmovaps(XMMRegister dst, XMMRegister src) { vmovaps(dst, Operand(src)); }
   void vmovaps(XMMRegister dst, Operand src) { vps(0x28, dst, xmm0, src); }
   void vmovapd(XMMRegister dst, XMMRegister src) { vmovapd(dst, Operand(src)); }

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

@@ -258,6 +258,7 @@ class V8_EXPORT_PRIVATE TurboAssembler : public TurboAssemblerBase {
   AVX_OP2_WITH_TYPE(Movd, movd, Register, XMMRegister)
   AVX_OP2_WITH_TYPE(Movd, movd, Operand, XMMRegister)
   AVX_OP2_WITH_TYPE(Cvtdq2ps, cvtdq2ps, XMMRegister, Operand)
+  AVX_OP2_WITH_TYPE(Sqrtpd, sqrtpd, XMMRegister, const Operand&)
 
 #undef AVX_OP2_WITH_TYPE
 

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

@@ -1900,6 +1900,10 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       }
       break;
     }
+    case kIA32F64x2Sqrt: {
+      __ Sqrtpd(i.OutputSimd128Register(), i.InputOperand(0));
+      break;
+    }
     case kSSEF32x4Splat: {
       DCHECK_EQ(i.OutputDoubleRegister(), i.InputDoubleRegister(0));
       XMMRegister dst = i.OutputSimd128Register();

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

@@ -122,6 +122,7 @@ namespace compiler {
   V(AVXF64x2ExtractLane)           \
   V(SSEF64x2ReplaceLane)           \
   V(AVXF64x2ReplaceLane)           \
+  V(IA32F64x2Sqrt)                 \
   V(SSEF32x4Splat)                 \
   V(AVXF32x4Splat)                 \
   V(SSEF32x4ExtractLane)           \

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

@@ -103,6 +103,7 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kAVXF64x2ExtractLane:
     case kSSEF64x2ReplaceLane:
     case kAVXF64x2ReplaceLane:
+    case kIA32F64x2Sqrt:
     case kSSEF32x4Splat:
     case kAVXF32x4Splat:
     case kSSEF32x4ExtractLane:

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

@@ -830,7 +830,8 @@ void InstructionSelector::VisitWord32Ror(Node* node) {
   V(Float64ExtractLowWord32, kSSEFloat64ExtractLowWord32)   \
   V(Float64ExtractHighWord32, kSSEFloat64ExtractHighWord32) \
   V(SignExtendWord8ToInt32, kIA32Movsxbl)                   \
-  V(SignExtendWord16ToInt32, kIA32Movsxwl)
+  V(SignExtendWord16ToInt32, kIA32Movsxwl)                  \
+  V(F64x2Sqrt, kIA32F64x2Sqrt)
 
 #define RO_WITH_TEMP_OP_LIST(V) V(ChangeUint32ToFloat64, kSSEUint32ToFloat64)
 

src/compiler/backend/instruction-selector.cc

@@ -2627,8 +2627,8 @@ void InstructionSelector::VisitF64x2ExtractLane(Node* node) { UNIMPLEMENTED(); }
 void InstructionSelector::VisitF64x2ReplaceLane(Node* node) { UNIMPLEMENTED(); }
 void InstructionSelector::VisitF64x2Abs(Node* node) { UNIMPLEMENTED(); }
 void InstructionSelector::VisitF64x2Neg(Node* node) { UNIMPLEMENTED(); }
-#endif  // !V8_TARGET_ARCH_IA32
 void InstructionSelector::VisitF64x2Sqrt(Node* node) { UNIMPLEMENTED(); }
+#endif  // !V8_TARGET_ARCH_IA32
 void InstructionSelector::VisitF64x2Add(Node* node) { UNIMPLEMENTED(); }
 void InstructionSelector::VisitF64x2Sub(Node* node) { UNIMPLEMENTED(); }
 void InstructionSelector::VisitF64x2Mul(Node* node) { UNIMPLEMENTED(); }

test/cctest/wasm/test-run-wasm-simd.cc

@@ -1285,11 +1285,11 @@ WASM_SIMD_TEST_NO_LOWERING(F64x2Neg) {
   RunF64x2UnOpTest(execution_tier, lower_simd, kExprF64x2Neg, Negate);
 }
 
-#if V8_TARGET_ARCH_X64 || V8_TARGET_ARCH_ARM64
 WASM_SIMD_TEST_NO_LOWERING(F64x2Sqrt) {
   RunF64x2UnOpTest(execution_tier, lower_simd, kExprF64x2Sqrt, Sqrt);
 }
 
+#if V8_TARGET_ARCH_X64 || V8_TARGET_ARCH_ARM64
 void RunF64x2BinOpTest(ExecutionTier execution_tier, LowerSimd lower_simd,
                        WasmOpcode opcode, DoubleBinOp expected_op) {
   WasmRunner<int32_t, double, double> r(execution_tier, lower_simd);