Implement simd bitmask instructions (#1530)

3 files changed, 21 insertions, 0 deletions

diff --git a/src/interp/interp.cc b/src/interp/interp.cc
index 5e5a6648..3beedcf8 100644
--- 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
index 5dfd42a9..32f6a73e 100644
--- 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
index 1063a37b..619a9844 100644
--- 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: