Implement simd bitmask instructions (WebAssembly#1530)

ludoplex · Aug 31, 2020 · c3b9c32 · c3b9c32
1 parent ca7890a
commit c3b9c32
Show file tree

Hide file tree

Showing 10 changed files with 631 additions and 529 deletions.
diff --git a/src/binary-reader.cc b/src/binary-reader.cc
@@ -1189,6 +1189,9 @@ Result BinaryReader::ReadFunctionBody(Offset end_offset) {
       case Opcode::I8X16AnyTrue:
       case Opcode::I16X8AnyTrue:
       case Opcode::I32X4AnyTrue:
+      case Opcode::I8X16Bitmask:
+      case Opcode::I16X8Bitmask:
+      case Opcode::I32X4Bitmask:
       case Opcode::I8X16AllTrue:
       case Opcode::I16X8AllTrue:
       case Opcode::I32X4AllTrue:

diff --git a/src/interp/interp.cc b/src/interp/interp.cc
@@ -1497,6 +1497,7 @@ RunResult Thread::StepInternal(Trap::Ptr* out_trap) {
 
     case O::I8X16Neg:          return DoSimdUnop(IntNeg<u8>);
     case O::I8X16AnyTrue:      return DoSimdIsTrue<u8x16, 1>();
+    case O::I8X16Bitmask:      return DoSimdBitmask<s8x16>();
     case O::I8X16AllTrue:      return DoSimdIsTrue<u8x16, 16>();
     case O::I8X16Shl:          return DoSimdShift(IntShl<u8>);
     case O::I8X16ShrS:         return DoSimdShift(IntShr<s8>);
@@ -1514,6 +1515,7 @@ RunResult Thread::StepInternal(Trap::Ptr* out_trap) {
 
     case O::I16X8Neg:          return DoSimdUnop(IntNeg<u16>);
     case O::I16X8AnyTrue:      return DoSimdIsTrue<u16x8, 1>();
+    case O::I16X8Bitmask:      return DoSimdBitmask<s16x8>();
     case O::I16X8AllTrue:      return DoSimdIsTrue<u16x8, 8>();
     case O::I16X8Shl:          return DoSimdShift(IntShl<u16>);
     case O::I16X8ShrS:         return DoSimdShift(IntShr<s16>);
@@ -1532,6 +1534,7 @@ RunResult Thread::StepInternal(Trap::Ptr* out_trap) {
 
     case O::I32X4Neg:          return DoSimdUnop(IntNeg<u32>);
     case O::I32X4AnyTrue:      return DoSimdIsTrue<u32x4, 1>();
+    case O::I32X4Bitmask:      return DoSimdBitmask<s32x4>();
     case O::I32X4AllTrue:      return DoSimdIsTrue<u32x4, 4>();
     case O::I32X4Shl:          return DoSimdShift(IntShl<u32>);
     case O::I32X4ShrS:         return DoSimdShift(IntShr<s32>);
@@ -2022,6 +2025,19 @@ RunResult Thread::DoSimdIsTrue() {
   return RunResult::Ok;
 }
 
+template <typename S>
+RunResult Thread::DoSimdBitmask() {
+  auto val = Pop<S>();
+  u32 result = 0;
+  for (u8 i = 0; i < S::lanes; ++i) {
+    if (val.v[i] < 0) {
+      result |= 1 << i;
+    }
+  }
+  Push(result);
+  return RunResult::Ok;
+}
+
 template <typename R, typename T>
 RunResult Thread::DoSimdShift(BinopFunc<R, T> f) {
   using ST = typename Simd128<T>::Type;

diff --git a/src/interp/interp.h b/src/interp/interp.h
@@ -1129,6 +1129,8 @@ class Thread : public Object {
   RunResult DoSimdBitSelect();
   template <typename S, u8 count>
   RunResult DoSimdIsTrue();
+  template <typename S>
+  RunResult DoSimdBitmask();
   template <typename R, typename T>
   RunResult DoSimdShift(BinopFunc<R, T>);
   template <typename S, typename T>

diff --git a/src/interp/istream.cc b/src/interp/istream.cc
@@ -154,6 +154,7 @@ Instr Istream::Read(Offset* offset) const {
     case Opcode::F64X2Sqrt:
     case Opcode::I16X8AllTrue:
     case Opcode::I16X8AnyTrue:
+    case Opcode::I16X8Bitmask:
     case Opcode::I16X8Neg:
     case Opcode::I16X8Splat:
     case Opcode::I16X8WidenHighI8X16S:
@@ -178,6 +179,7 @@ Instr Istream::Read(Offset* offset) const {
     case Opcode::I32WrapI64:
     case Opcode::I32X4AllTrue:
     case Opcode::I32X4AnyTrue:
+    case Opcode::I32X4Bitmask:
     case Opcode::I32X4Neg:
     case Opcode::I32X4Splat:
     case Opcode::I32X4TruncSatF32X4S:
@@ -208,6 +210,7 @@ Instr Istream::Read(Offset* offset) const {
     case Opcode::I64X2Splat:
     case Opcode::I8X16AllTrue:
     case Opcode::I8X16AnyTrue:
+    case Opcode::I8X16Bitmask:
     case Opcode::I8X16Neg:
     case Opcode::I8X16Splat:
     case Opcode::RefIsNull:

diff --git a/src/lexer-keywords.txt b/src/lexer-keywords.txt
@@ -166,6 +166,7 @@ i16x8.add, TokenType::Binary, Opcode::I16X8Add
 i16x8.all_true, TokenType::Unary, Opcode::I16X8AllTrue
 i16x8.any_true, TokenType::Unary, Opcode::I16X8AnyTrue
 i16x8.avgr_u, TokenType::Binary, Opcode::I16X8AvgrU
+i16x8.bitmask, TokenType::Unary, Opcode::I16X8Bitmask
 i16x8.eq, TokenType::Compare, Opcode::I16X8Eq
 i16x8.extract_lane_s, TokenType::SimdLaneOp, Opcode::I16X8ExtractLaneS
 i16x8.extract_lane_u, TokenType::SimdLaneOp, Opcode::I16X8ExtractLaneU
@@ -283,6 +284,7 @@ i32x4.abs, TokenType::Unary, Opcode::I32X4Abs
 i32x4.add, TokenType::Binary, Opcode::I32X4Add
 i32x4.all_true, TokenType::Unary, Opcode::I32X4AllTrue
 i32x4.any_true, TokenType::Unary, Opcode::I32X4AnyTrue
+i32x4.bitmask, TokenType::Unary, Opcode::I32X4Bitmask
 i32x4.eq, TokenType::Compare, Opcode::I32X4Eq
 i32x4.extract_lane, TokenType::SimdLaneOp, Opcode::I32X4ExtractLane
 i32x4.ge_s, TokenType::Compare, Opcode::I32X4GeS
@@ -429,6 +431,7 @@ i8x16.add, TokenType::Binary, Opcode::I8X16Add
 i8x16.all_true, TokenType::Unary, Opcode::I8X16AllTrue
 i8x16.any_true, TokenType::Unary, Opcode::I8X16AnyTrue
 i8x16.avgr_u, TokenType::Binary, Opcode::I8X16AvgrU
+i8x16.bitmask, TokenType::Unary, Opcode::I8X16Bitmask
 i8x16.eq, TokenType::Compare, Opcode::I8X16Eq
 i8x16.extract_lane_s, TokenType::SimdLaneOp, Opcode::I8X16ExtractLaneS
 i8x16.extract_lane_u, TokenType::SimdLaneOp, Opcode::I8X16ExtractLaneU

diff --git a/src/opcode.cc b/src/opcode.cc
@@ -225,6 +225,9 @@ bool Opcode::IsEnabled(const Features& features) const {
     case Opcode::I8X16AnyTrue:
     case Opcode::I16X8AnyTrue:
     case Opcode::I32X4AnyTrue:
+    case Opcode::I8X16Bitmask:
+    case Opcode::I16X8Bitmask:
+    case Opcode::I32X4Bitmask:
     case Opcode::I8X16AllTrue:
     case Opcode::I16X8AllTrue:
     case Opcode::I32X4AllTrue:

diff --git a/src/opcode.def b/src/opcode.def
@@ -347,6 +347,7 @@ WABT_OPCODE(V128, V128,  ___, ___,  0,  0xfd, 0x60, I8X16Abs, "i8x16.abs", "")
 WABT_OPCODE(V128, V128, ___,  ___,  0,  0xfd, 0x61, I8X16Neg, "i8x16.neg", "")
 WABT_OPCODE(I32,  V128, ___,  ___,  0,  0xfd, 0x62, I8X16AnyTrue, "i8x16.any_true", "")
 WABT_OPCODE(I32,  V128, ___,  ___,  0,  0xfd, 0x63, I8X16AllTrue, "i8x16.all_true", "")
+WABT_OPCODE(I32,  V128, ___,  ___,  0,  0xfd, 0x64, I8X16Bitmask, "i8x16.bitmask", "")
 WABT_OPCODE(V128, V128, V128, ___,  0,  0xfd, 0x65, I8X16NarrowI16X8S, "i8x16.narrow_i16x8_s", "")
 WABT_OPCODE(V128, V128, V128, ___,  0,  0xfd, 0x66, I8X16NarrowI16X8U, "i8x16.narrow_i16x8_u", "")
 WABT_OPCODE(V128, V128, I32,  ___,  0,  0xfd, 0x6b, I8X16Shl, "i8x16.shl", "")
@@ -367,6 +368,7 @@ WABT_OPCODE(V128, V128,  ___, ___,  0,  0xfd, 0x80, I16X8Abs, "i16x8.abs", "")
 WABT_OPCODE(V128, V128, ___,  ___,  0,  0xfd, 0x81, I16X8Neg, "i16x8.neg", "")
 WABT_OPCODE(I32,  V128, ___,  ___,  0,  0xfd, 0x82, I16X8AnyTrue, "i16x8.any_true", "")
 WABT_OPCODE(I32,  V128, ___,  ___,  0,  0xfd, 0x83, I16X8AllTrue, "i16x8.all_true", "")
+WABT_OPCODE(I32,  V128, ___,  ___,  0,  0xfd, 0x84, I16X8Bitmask, "i16x8.bitmask", "")
 WABT_OPCODE(V128, V128, V128, ___,  0,  0xfd, 0x85, I16X8NarrowI32X4S, "i16x8.narrow_i32x4_s", "")
 WABT_OPCODE(V128, V128, V128, ___,  0,  0xfd, 0x86, I16X8NarrowI32X4U, "i16x8.narrow_i32x4_u", "")
 WABT_OPCODE(V128, V128,  ___, ___,  0,  0xfd, 0x87, I16X8WidenLowI8X16S, "i16x8.widen_low_i8x16_s", "")
@@ -392,6 +394,7 @@ WABT_OPCODE(V128, V128,  ___, ___,  0,  0xfd, 0xa0, I32X4Abs, "i32x4.abs", "")
 WABT_OPCODE(V128, V128, ___,  ___,  0,  0xfd, 0xa1, I32X4Neg, "i32x4.neg", "")
 WABT_OPCODE(I32,  V128, ___,  ___,  0,  0xfd, 0xa2, I32X4AnyTrue, "i32x4.any_true", "")
 WABT_OPCODE(I32,  V128, ___,  ___,  0,  0xfd, 0xa3, I32X4AllTrue, "i32x4.all_true", "")
+WABT_OPCODE(I32,  V128, ___,  ___,  0,  0xfd, 0xa4, I32X4Bitmask, "i32x4.bitmask", "")
 WABT_OPCODE(V128, V128,  ___, ___,  0,  0xfd, 0xa7, I32X4WidenLowI16X8S, "i32x4.widen_low_i16x8_s", "")
 WABT_OPCODE(V128, V128,  ___, ___,  0,  0xfd, 0xa8, I32X4WidenHighI16X8S, "i32x4.widen_high_i16x8_s", "")
 WABT_OPCODE(V128, V128,  ___, ___,  0,  0xfd, 0xa9, I32X4WidenLowI16X8U, "i32x4.widen_low_i16x8_u", "")