SIMD integer abs and bitmask instructions (#2703)

Adds full support for the {i8x16,i16x8,i32x4}.abs instructions merged
to the SIMD proposal in https://github.com/WebAssembly/simd/pull/128
as well as the {i8x16,i16x8,i32x4}.bitmask instructions proposed in
https://github.com/WebAssembly/simd/pull/201.

5 files changed, 96 insertions, 12 deletions

diff --git a/src/wasm/literal.cpp b/src/wasm/literal.cpp
index aa2acdcc1..b740ae520 100644
--- a/src/wasm/literal.cpp
+++ b/src/wasm/literal.cpp
@@ -565,9 +565,9 @@ Literal Literal::neg() const {
 Literal Literal::abs() const {
   switch (type.getSingle()) {
     case Type::i32:
-      return Literal(i32 & 0x7fffffff);
+      return Literal(std::abs(i32));
     case Type::i64:
-      return Literal(int64_t(i64 & 0x7fffffffffffffffULL));
+      return Literal(std::abs(i64));
     case Type::f32:
       return Literal(i32 & 0x7fffffff).castToF32();
     case Type::f64:
@@ -1442,6 +1442,15 @@ Literal Literal::notV128() const {
   ones.fill(0xff);
   return xorV128(Literal(ones.data()));
 }
+Literal Literal::absI8x16() const {
+  return unary<16, &Literal::getLanesSI8x16, &Literal::abs>(*this);
+}
+Literal Literal::absI16x8() const {
+  return unary<8, &Literal::getLanesSI16x8, &Literal::abs>(*this);
+}
+Literal Literal::absI32x4() const {
+  return unary<4, &Literal::getLanesI32x4, &Literal::abs>(*this);
+}
 Literal Literal::negI8x16() const {
   return unary<16, &Literal::getLanesUI8x16, &Literal::neg>(*this);
 }
@@ -1519,24 +1528,45 @@ static Literal all_true(const Literal& val) {
   return Literal(int32_t(1));
 }
 
+template<int Lanes, LaneArray<Lanes> (Literal::*IntoLanes)() const>
+static Literal bitmask(const Literal& val) {
+  uint32_t result = 0;
+  LaneArray<Lanes> lanes = (val.*IntoLanes)();
+  for (size_t i = 0; i < Lanes; ++i) {
+    if (lanes[i].geti32() & (1 << 31)) {
+      result = result | (1 << i);
+    }
+  }
+  return Literal(result);
+}
+
 Literal Literal::anyTrueI8x16() const {
   return any_true<16, &Literal::getLanesUI8x16>(*this);
 }
 Literal Literal::allTrueI8x16() const {
   return all_true<16, &Literal::getLanesUI8x16>(*this);
 }
+Literal Literal::bitmaskI8x16() const {
+  return bitmask<16, &Literal::getLanesSI8x16>(*this);
+}
 Literal Literal::anyTrueI16x8() const {
   return any_true<8, &Literal::getLanesUI16x8>(*this);
 }
 Literal Literal::allTrueI16x8() const {
   return all_true<8, &Literal::getLanesUI16x8>(*this);
 }
+Literal Literal::bitmaskI16x8() const {
+  return bitmask<8, &Literal::getLanesSI16x8>(*this);
+}
 Literal Literal::anyTrueI32x4() const {
   return any_true<4, &Literal::getLanesI32x4>(*this);
 }
 Literal Literal::allTrueI32x4() const {
   return all_true<4, &Literal::getLanesI32x4>(*this);
 }
+Literal Literal::bitmaskI32x4() const {
+  return bitmask<4, &Literal::getLanesI32x4>(*this);
+}
 Literal Literal::anyTrueI64x2() const {
   return any_true<2, &Literal::getLanesI64x2>(*this);
 }
diff --git a/src/wasm/wasm-binary.cpp b/src/wasm/wasm-binary.cpp
index 9fe046c27..dbd644a17 100644
--- a/src/wasm/wasm-binary.cpp
+++ b/src/wasm/wasm-binary.cpp
@@ -4053,6 +4053,10 @@ bool WasmBinaryBuilder::maybeVisitSIMDUnary(Expression*& out, uint32_t code) {
       curr = allocator.alloc<Unary>();
       curr->op = NotVec128;
       break;
+    case BinaryConsts::I8x16Abs:
+      curr = allocator.alloc<Unary>();
+      curr->op = AbsVecI8x16;
+      break;
     case BinaryConsts::I8x16Neg:
       curr = allocator.alloc<Unary>();
       curr->op = NegVecI8x16;
@@ -4065,6 +4069,14 @@ bool WasmBinaryBuilder::maybeVisitSIMDUnary(Expression*& out, uint32_t code) {
       curr = allocator.alloc<Unary>();
       curr->op = AllTrueVecI8x16;
       break;
+    case BinaryConsts::I8x16Bitmask:
+      curr = allocator.alloc<Unary>();
+      curr->op = BitmaskVecI8x16;
+      break;
+    case BinaryConsts::I16x8Abs:
+      curr = allocator.alloc<Unary>();
+      curr->op = AbsVecI16x8;
+      break;
     case BinaryConsts::I16x8Neg:
       curr = allocator.alloc<Unary>();
       curr->op = NegVecI16x8;
@@ -4077,6 +4089,14 @@ bool WasmBinaryBuilder::maybeVisitSIMDUnary(Expression*& out, uint32_t code) {
       curr = allocator.alloc<Unary>();
       curr->op = AllTrueVecI16x8;
       break;
+    case BinaryConsts::I16x8Bitmask:
+      curr = allocator.alloc<Unary>();
+      curr->op = BitmaskVecI16x8;
+      break;
+    case BinaryConsts::I32x4Abs:
+      curr = allocator.alloc<Unary>();
+      curr->op = AbsVecI32x4;
+      break;
     case BinaryConsts::I32x4Neg:
       curr = allocator.alloc<Unary>();
       curr->op = NegVecI32x4;
@@ -4089,6 +4109,10 @@ bool WasmBinaryBuilder::maybeVisitSIMDUnary(Expression*& out, uint32_t code) {
       curr = allocator.alloc<Unary>();
       curr->op = AllTrueVecI32x4;
       break;
+    case BinaryConsts::I32x4Bitmask:
+      curr = allocator.alloc<Unary>();
+      curr->op = BitmaskVecI32x4;
+      break;
     case BinaryConsts::I64x2Neg:
       curr = allocator.alloc<Unary>();
       curr->op = NegVecI64x2;
diff --git a/src/wasm/wasm-stack.cpp b/src/wasm/wasm-stack.cpp
index 3c162a9d5..9b5dc3f22 100644
--- a/src/wasm/wasm-stack.cpp
+++ b/src/wasm/wasm-stack.cpp
@@ -899,6 +899,9 @@ void BinaryInstWriter::visitUnary(Unary* curr) {
     case NotVec128:
       o << int8_t(BinaryConsts::SIMDPrefix) << U32LEB(BinaryConsts::V128Not);
       break;
+    case AbsVecI8x16:
+      o << int8_t(BinaryConsts::SIMDPrefix) << U32LEB(BinaryConsts::I8x16Abs);
+      break;
     case NegVecI8x16:
       o << int8_t(BinaryConsts::SIMDPrefix) << U32LEB(BinaryConsts::I8x16Neg);
       break;
@@ -910,6 +913,13 @@ void BinaryInstWriter::visitUnary(Unary* curr) {
       o << int8_t(BinaryConsts::SIMDPrefix)
         << U32LEB(BinaryConsts::I8x16AllTrue);
       break;
+    case BitmaskVecI8x16:
+      o << int8_t(BinaryConsts::SIMDPrefix)
+        << U32LEB(BinaryConsts::I8x16Bitmask);
+      break;
+    case AbsVecI16x8:
+      o << int8_t(BinaryConsts::SIMDPrefix) << U32LEB(BinaryConsts::I16x8Abs);
+      break;
     case NegVecI16x8:
       o << int8_t(BinaryConsts::SIMDPrefix) << U32LEB(BinaryConsts::I16x8Neg);
       break;
@@ -921,6 +931,13 @@ void BinaryInstWriter::visitUnary(Unary* curr) {
       o << int8_t(BinaryConsts::SIMDPrefix)
         << U32LEB(BinaryConsts::I16x8AllTrue);
       break;
+    case BitmaskVecI16x8:
+      o << int8_t(BinaryConsts::SIMDPrefix)
+        << U32LEB(BinaryConsts::I16x8Bitmask);
+      break;
+    case AbsVecI32x4:
+      o << int8_t(BinaryConsts::SIMDPrefix) << U32LEB(BinaryConsts::I32x4Abs);
+      break;
     case NegVecI32x4:
       o << int8_t(BinaryConsts::SIMDPrefix) << U32LEB(BinaryConsts::I32x4Neg);
       break;
@@ -932,6 +949,10 @@ void BinaryInstWriter::visitUnary(Unary* curr) {
       o << int8_t(BinaryConsts::SIMDPrefix)
         << U32LEB(BinaryConsts::I32x4AllTrue);
       break;
+    case BitmaskVecI32x4:
+      o << int8_t(BinaryConsts::SIMDPrefix)
+        << U32LEB(BinaryConsts::I32x4Bitmask);
+      break;
     case NegVecI64x2:
       o << int8_t(BinaryConsts::SIMDPrefix) << U32LEB(BinaryConsts::I64x2Neg);
       break;
diff --git a/src/wasm/wasm-validator.cpp b/src/wasm/wasm-validator.cpp
index 1db8ed7c0..984c25b3f 100644
--- a/src/wasm/wasm-validator.cpp
+++ b/src/wasm/wasm-validator.cpp
@@ -1689,6 +1689,9 @@ void FunctionValidator::visitUnary(Unary* curr) {
         curr->value->type, Type(Type::f64), curr, "expected f64 splat value");
       break;
     case NotVec128:
+    case AbsVecI8x16:
+    case AbsVecI16x8:
+    case AbsVecI32x4:
     case NegVecI8x16:
     case NegVecI16x8:
     case NegVecI32x4:
@@ -1720,17 +1723,17 @@ void FunctionValidator::visitUnary(Unary* curr) {
         curr->value->type, Type(Type::v128), curr, "expected v128 operand");
       break;
     case AnyTrueVecI8x16:
-    case AllTrueVecI8x16:
     case AnyTrueVecI16x8:
-    case AllTrueVecI16x8:
     case AnyTrueVecI32x4:
-    case AllTrueVecI32x4:
     case AnyTrueVecI64x2:
+    case AllTrueVecI8x16:
+    case AllTrueVecI16x8:
+    case AllTrueVecI32x4:
     case AllTrueVecI64x2:
-      shouldBeEqual(curr->type,
-                    Type(Type::i32),
-                    curr,
-                    "expected boolean reduction to have i32 type");
+    case BitmaskVecI8x16:
+    case BitmaskVecI16x8:
+    case BitmaskVecI32x4:
+      shouldBeEqual(curr->type, Type(Type::i32), curr, "expected i32 type");
       shouldBeEqual(
         curr->value->type, Type(Type::v128), curr, "expected v128 operand");
       break;
diff --git a/src/wasm/wasm.cpp b/src/wasm/wasm.cpp
index 35fa8b5c5..6f7ea4f86 100644
--- a/src/wasm/wasm.cpp
+++ b/src/wasm/wasm.cpp
@@ -737,6 +737,9 @@ void Unary::finalize() {
     case SplatVecF32x4:
     case SplatVecF64x2:
     case NotVec128:
+    case AbsVecI8x16:
+    case AbsVecI16x8:
+    case AbsVecI32x4:
     case NegVecI8x16:
     case NegVecI16x8:
     case NegVecI32x4:
@@ -766,13 +769,16 @@ void Unary::finalize() {
       type = Type::v128;
       break;
     case AnyTrueVecI8x16:
-    case AllTrueVecI8x16:
     case AnyTrueVecI16x8:
-    case AllTrueVecI16x8:
     case AnyTrueVecI32x4:
-    case AllTrueVecI32x4:
     case AnyTrueVecI64x2:
+    case AllTrueVecI8x16:
+    case AllTrueVecI16x8:
+    case AllTrueVecI32x4:
     case AllTrueVecI64x2:
+    case BitmaskVecI8x16:
+    case BitmaskVecI16x8:
+    case BitmaskVecI32x4:
       type = Type::i32;
       break;