[wasm-simd][x64] Prototype i64x2.bitmask

Implement on interpreter and x64.

Bug: v8:10997
Change-Id: I3537ce54e1b56cc3b04d91cb07c430c35b88c3aa
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2459109
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Reviewed-by: Bill Budge <bbudge@chromium.org>
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70459}

src/compiler/backend/instruction-selector.cc

@@ -1994,6 +1994,8 @@ void InstructionSelector::VisitNode(Node* node) {
       return MarkAsSimd128(node), VisitI64x2UConvertI32x4Low(node);
     case IrOpcode::kI64x2UConvertI32x4High:
       return MarkAsSimd128(node), VisitI64x2UConvertI32x4High(node);
+    case IrOpcode::kI64x2BitMask:
+      return MarkAsWord32(node), VisitI64x2BitMask(node);
     case IrOpcode::kI64x2Shl:
       return MarkAsSimd128(node), VisitI64x2Shl(node);
     case IrOpcode::kI64x2ShrS:
@@ -2697,9 +2699,12 @@ void InstructionSelector::VisitI64x2UConvertI32x4High(Node* node) {
 }
 #endif  // !V8_TARGET_ARCH_ARM64
 
-// TODO(v8:10975): Prototyping load lane and store lane.
 #if !V8_TARGET_ARCH_X64
+// TODO(v8:10975): Prototyping load lane and store lane.
 void InstructionSelector::VisitLoadLane(Node* node) { UNIMPLEMENTED(); }
+
+// TODO(v8:10997) Prototype i64x2.bitmask.
+void InstructionSelector::VisitI64x2BitMask(Node* node) { UNIMPLEMENTED(); }
 #endif  // !V8_TARGET_ARCH_X64
 
 void InstructionSelector::VisitFinishRegion(Node* node) { EmitIdentity(node); }

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

@@ -2728,6 +2728,10 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       __ Psubq(dst, src);
       break;
     }
+    case kX64I64x2BitMask: {
+      __ Movmskpd(i.OutputRegister(), i.InputSimd128Register(0));
+      break;
+    }
     case kX64I64x2Shl: {
       // Take shift value modulo 2^6.
       ASSEMBLE_SIMD_SHIFT(Psllq, 6);

src/compiler/backend/x64/instruction-codes-x64.h

@@ -205,6 +205,7 @@ namespace compiler {
   V(X64I64x2ExtractLane)                  \
   V(X64I64x2ReplaceLane)                  \
   V(X64I64x2Neg)                          \
+  V(X64I64x2BitMask)                      \
   V(X64I64x2Shl)                          \
   V(X64I64x2ShrS)                         \
   V(X64I64x2Add)                          \

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

@@ -177,6 +177,7 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kX64I64x2ExtractLane:
     case kX64I64x2ReplaceLane:
     case kX64I64x2Neg:
+    case kX64I64x2BitMask:
     case kX64I64x2Shl:
     case kX64I64x2ShrS:
     case kX64I64x2Add:

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

@@ -2867,6 +2867,7 @@ VISIT_ATOMIC_BINOP(Xor)
   V(F32x4RecipApprox)       \
   V(F32x4RecipSqrtApprox)   \
   V(I64x2Neg)               \
+  V(I64x2BitMask)           \
   V(I32x4SConvertI16x8Low)  \
   V(I32x4SConvertI16x8High) \
   V(I32x4Neg)               \

src/compiler/machine-operator.cc

@@ -398,6 +398,7 @@ ShiftKind ShiftKindOf(Operator const* op) {
   V(I64x2SConvertI32x4High, Operator::kNoProperties, 1, 0, 1)              \
   V(I64x2UConvertI32x4Low, Operator::kNoProperties, 1, 0, 1)               \
   V(I64x2UConvertI32x4High, Operator::kNoProperties, 1, 0, 1)              \
+  V(I64x2BitMask, Operator::kNoProperties, 1, 0, 1)                        \
   V(I64x2Shl, Operator::kNoProperties, 2, 0, 1)                            \
   V(I64x2ShrS, Operator::kNoProperties, 2, 0, 1)                           \
   V(I64x2Add, Operator::kCommutative, 2, 0, 1)                             \

src/compiler/machine-operator.h

@@ -652,6 +652,7 @@ class V8_EXPORT_PRIVATE MachineOperatorBuilder final
   const Operator* I64x2SConvertI32x4High();
   const Operator* I64x2UConvertI32x4Low();
   const Operator* I64x2UConvertI32x4High();
+  const Operator* I64x2BitMask();
   const Operator* I64x2Shl();
   const Operator* I64x2ShrS();
   const Operator* I64x2Add();

src/compiler/opcodes.h

@@ -823,6 +823,7 @@
   V(I64x2SConvertI32x4High)     \
   V(I64x2UConvertI32x4Low)      \
   V(I64x2UConvertI32x4High)     \
+  V(I64x2BitMask)               \
   V(I64x2Shl)                   \
   V(I64x2ShrS)                  \
   V(I64x2Add)                   \

src/compiler/wasm-compiler.cc

@@ -4597,6 +4597,8 @@ Node* WasmGraphBuilder::SimdOp(wasm::WasmOpcode opcode, Node* const* inputs) {
     case wasm::kExprI64x2UConvertI32x4High:
       return graph()->NewNode(mcgraph()->machine()->I64x2UConvertI32x4High(),
                               inputs[0]);
+    case wasm::kExprI64x2BitMask:
+      return graph()->NewNode(mcgraph()->machine()->I64x2BitMask(), inputs[0]);
     case wasm::kExprI64x2Shl:
       return graph()->NewNode(mcgraph()->machine()->I64x2Shl(), inputs[0],
                               inputs[1]);

src/wasm/wasm-opcodes-inl.h

@@ -325,6 +325,7 @@ constexpr const char* WasmOpcodes::OpcodeName(WasmOpcode opcode) {
     CASE_I8x16_OP(BitMask, "bitmask")
     CASE_I16x8_OP(BitMask, "bitmask")
     CASE_I32x4_OP(BitMask, "bitmask")
+    CASE_I64x2_OP(BitMask, "bitmask")
 
     CASE_F32x4_OP(Pmin, "pmin")
     CASE_F32x4_OP(Pmax, "pmax")

src/wasm/wasm-opcodes.h

@@ -475,6 +475,7 @@ bool V8_EXPORT_PRIVATE IsJSCompatibleSignature(const FunctionSig* sig,
   V(I16x8Q15MulRSatS, 0xfd9c, s_ss)     \
   V(I64x2Eq, 0xfdc0, s_ss)              \
   V(F32x4Qfma, 0xfdb4, s_sss)           \
+  V(I64x2BitMask, 0xfdc4, i_s)          \
   V(F32x4Qfms, 0xfdd4, s_sss)           \
   V(F64x2Qfma, 0xfdfe, s_sss)           \
   V(F64x2Qfms, 0xfdff, s_sss)           \

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

@@ -1581,6 +1581,27 @@ WASM_SIMD_TEST(I32x4BitMask) {
   }
 }
 
+// TODO(v8:10997) Prototyping i64x2.bitmask.
+#if V8_TARGET_ARCH_X64
+WASM_SIMD_TEST_NO_LOWERING(I64x2BitMask) {
+  FLAG_SCOPE(wasm_simd_post_mvp);
+  WasmRunner<int32_t, int64_t> r(execution_tier, lower_simd);
+  byte value1 = r.AllocateLocal(kWasmS128);
+
+  BUILD(r, WASM_SET_LOCAL(value1, WASM_SIMD_I64x2_SPLAT(WASM_GET_LOCAL(0))),
+        WASM_SET_LOCAL(value1, WASM_SIMD_I64x2_REPLACE_LANE(
+                                   0, WASM_GET_LOCAL(value1), WASM_I64V_1(0))),
+        WASM_SIMD_UNOP(kExprI64x2BitMask, WASM_GET_LOCAL(value1)));
+
+  for (int64_t x : compiler::ValueHelper::GetVector<int64_t>()) {
+    int32_t actual = r.Call(x);
+    // Lane 0 is always 0 (positive).
+    int32_t expected = std::signbit(static_cast<double>(x)) ? 0x2 : 0x0;
+    CHECK_EQ(actual, expected);
+  }
+}
+#endif  // V8_TARGET_ARCH_X64
+
 WASM_SIMD_TEST(I8x16Splat) {
   WasmRunner<int32_t, int32_t> r(execution_tier, lower_simd);
   // Set up a global to hold output vector.

test/common/wasm/wasm-interpreter.cc

@@ -2293,6 +2293,7 @@ class WasmInterpreterInternals {
       BITMASK_CASE(I8x16BitMask, i8x16, int16, 16)
       BITMASK_CASE(I16x8BitMask, i16x8, int8, 8)
       BITMASK_CASE(I32x4BitMask, i32x4, int4, 4)
+      BITMASK_CASE(I64x2BitMask, i64x2, int2, 2)
 #undef BITMASK_CASE
 
 #define CMPOP_CASE(op, name, stype, out_stype, count, expr)   \