cranelift: Add bmask to interpreter

cranelift/filetests/filetests/runtests/bmask.clif

@@ -0,0 +1,161 @@
+test interpret
+
+function %bmask_b64_i64(b64) -> i64 {
+block0(v0: b64):
+  v1 = bmask.i64 v0
+  return v1
+}
+; run: %bmask_b64_i64(true) == -1
+; run: %bmask_b64_i64(false) == 0
+
+function %bmask_b64_i32(b64) -> i32 {
+block0(v0: b64):
+  v1 = bmask.i32 v0
+  return v1
+}
+; run: %bmask_b64_i32(true) == -1
+; run: %bmask_b64_i32(false) == 0
+
+function %bmask_b64_i16(b64) -> i16 {
+block0(v0: b64):
+  v1 = bmask.i16 v0
+  return v1
+}
+; run: %bmask_b64_i16(true) == -1
+; run: %bmask_b64_i16(false) == 0
+
+function %bmask_b64_i8(b64) -> i8 {
+block0(v0: b64):
+  v1 = bmask.i8 v0
+  return v1
+}
+; run: %bmask_b64_i8(true) == -1
+; run: %bmask_b64_i8(false) == 0
+
+function %bmask_b32_i64(b32) -> i64 {
+block0(v0: b32):
+  v1 = bmask.i64 v0
+  return v1
+}
+; run: %bmask_b32_i64(true) == -1
+; run: %bmask_b32_i64(false) == 0
+
+function %bmask_b32_i32(b32) -> i32 {
+block0(v0: b32):
+  v1 = bmask.i32 v0
+  return v1
+}
+; run: %bmask_b32_i32(true) == -1
+; run: %bmask_b32_i32(false) == 0
+
+function %bmask_b32_i16(b32) -> i16 {
+block0(v0: b32):
+  v1 = bmask.i16 v0
+  return v1
+}
+; run: %bmask_b32_i16(true) == -1
+; run: %bmask_b32_i16(false) == 0
+
+function %bmask_b32_i8(b32) -> i8 {
+block0(v0: b32):
+  v1 = bmask.i8 v0
+  return v1
+}
+; run: %bmask_b32_i8(true) == -1
+; run: %bmask_b32_i8(false) == 0
+
+function %bmask_b16_i64(b16) -> i64 {
+block0(v0: b16):
+  v1 = bmask.i64 v0
+  return v1
+}
+; run: %bmask_b16_i64(true) == -1
+; run: %bmask_b16_i64(false) == 0
+
+function %bmask_b16_i32(b16) -> i32 {
+block0(v0: b16):
+  v1 = bmask.i32 v0
+  return v1
+}
+; run: %bmask_b16_i32(true) == -1
+; run: %bmask_b16_i32(false) == 0
+
+function %bmask_b16_i16(b16) -> i16 {
+block0(v0: b16):
+  v1 = bmask.i16 v0
+  return v1
+}
+; run: %bmask_b16_i16(true) == -1
+; run: %bmask_b16_i16(false) == 0
+
+function %bmask_b16_i8(b16) -> i8 {
+block0(v0: b16):
+  v1 = bmask.i8 v0
+  return v1
+}
+; run: %bmask_b16_i8(true) == -1
+; run: %bmask_b16_i8(false) == 0
+
+function %bmask_b8_i64(b8) -> i64 {
+block0(v0: b8):
+  v1 = bmask.i64 v0
+  return v1
+}
+; run: %bmask_b8_i64(true) == -1
+; run: %bmask_b8_i64(false) == 0
+
+function %bmask_b8_i32(b8) -> i32 {
+block0(v0: b8):
+  v1 = bmask.i32 v0
+  return v1
+}
+; run: %bmask_b8_i32(true) == -1
+; run: %bmask_b8_i32(false) == 0
+
+function %bmask_b8_i16(b8) -> i16 {
+block0(v0: b8):
+  v1 = bmask.i16 v0
+  return v1
+}
+; run: %bmask_b8_i16(true) == -1
+; run: %bmask_b8_i16(false) == 0
+
+function %bmask_b8_i8(b8) -> i8 {
+block0(v0: b8):
+  v1 = bmask.i8 v0
+  return v1
+}
+; run: %bmask_b8_i8(true) == -1
+; run: %bmask_b8_i8(false) == 0
+
+function %bmask_b1_i64(b1) -> i64 {
+block0(v0: b1):
+  v1 = bmask.i64 v0
+  return v1
+}
+; run: %bmask_b1_i64(true) == -1
+; run: %bmask_b1_i64(false) == 0
+
+function %bmask_b1_i32(b1) -> i32 {
+block0(v0: b1):
+  v1 = bmask.i32 v0
+  return v1
+}
+; run: %bmask_b1_i32(true) == -1
+; run: %bmask_b1_i32(false) == 0
+
+function %bmask_b1_i16(b1) -> i16 {
+block0(v0: b1):
+  v1 = bmask.i16 v0
+  return v1
+}
+; run: %bmask_b1_i16(true) == -1
+; run: %bmask_b1_i16(false) == 0
+
+function %bmask_b1_i8(b1) -> i8 {
+block0(v0: b1):
+  v1 = bmask.i8 v0
+  return v1
+}
+; run: %bmask_b1_i8(true) == -1
+; run: %bmask_b1_i8(false) == 0

cranelift/filetests/filetests/runtests/i128-bmask.clif

@@ -0,0 +1,82 @@
+test interpret
+
+function %bmask_b128_i128(b128) -> i128 {
+block0(v0: b128):
+  v1 = bmask.i128 v0
+  return v1
+}
+; run: %bmask_b128_i128(true) == -1
+; run: %bmask_b128_i128(false) == 0
+
+function %bmask_b128_i64(b128) -> i64 {
+block0(v0: b128):
+  v1 = bmask.i64 v0
+  return v1
+}
+; run: %bmask_b128_i64(true) == -1
+; run: %bmask_b128_i64(false) == 0
+
+function %bmask_b128_i32(b128) -> i32 {
+block0(v0: b128):
+  v1 = bmask.i32 v0
+  return v1
+}
+; run: %bmask_b128_i32(true) == -1
+; run: %bmask_b128_i32(false) == 0
+
+function %bmask_b128_i16(b128) -> i16 {
+block0(v0: b128):
+  v1 = bmask.i16 v0
+  return v1
+}
+; run: %bmask_b128_i16(true) == -1
+; run: %bmask_b128_i16(false) == 0
+
+function %bmask_b128_i8(b128) -> i8 {
+block0(v0: b128):
+  v1 = bmask.i8 v0
+  return v1
+}
+; run: %bmask_b128_i8(true) == -1
+; run: %bmask_b128_i8(false) == 0
+
+
+function %bmask_b64_i128(b64) -> i128 {
+block0(v0: b64):
+  v1 = bmask.i128 v0
+  return v1
+}
+; run: %bmask_b64_i128(true) == -1
+; run: %bmask_b64_i128(false) == 0
+
+function %bmask_b32_i128(b32) -> i128 {
+block0(v0: b32):
+  v1 = bmask.i128 v0
+  return v1
+}
+; run: %bmask_b32_i128(true) == -1
+; run: %bmask_b32_i128(false) == 0
+
+function %bmask_b16_i128(b16) -> i128 {
+block0(v0: b16):
+  v1 = bmask.i128 v0
+  return v1
+}
+; run: %bmask_b16_i128(true) == -1
+; run: %bmask_b16_i128(false) == 0
+
+function %bmask_b8_i128(b8) -> i128 {
+block0(v0: b8):
+  v1 = bmask.i128 v0
+  return v1
+}
+; run: %bmask_b8_i128(true) == -1
+; run: %bmask_b8_i128(false) == 0
+
+function %bmask_b1_i128(b1) -> i128 {
+block0(v0: b1):
+  v1 = bmask.i128 v0
+  return v1
+}
+; run: %bmask_b1_i128(true) == -1
+; run: %bmask_b1_i128(false) == 0

cranelift/filetests/filetests/runtests/simd-bmask.clif

@@ -0,0 +1,30 @@
+test interpret
+
+
+function %bmask_i8x16(b8x16) -> i8x16 {
+block0(v0: b8x16):
+    v1 = bmask.i8x16 v0
+    return v1
+}
+; run: %bmask_i8x16([true false true false true false true false true false true false true false true false]) == [-1 0 -1 0 -1 0 -1 0 -1 0 -1 0 -1 0 -1 0]
+
+function %bmask_i16x8(b16x8) -> i16x8 {
+block0(v0: b16x8):
+    v1 = bmask.i16x8 v0
+    return v1
+}
+; run: %bmask_i16x8([true false true false true false true false]) == [-1 0 -1 0 -1 0 -1 0]
+
+function %bmask_i32x4(b32x4) -> i32x4 {
+block0(v0: b32x4):
+    v1 = bmask.i32x4 v0
+    return v1
+}
+; run: %bmask_i32x4([true false true false]) == [-1 0 -1 0]
+
+function %bmask_i64x2(b64x2) -> i64x2 {
+block0(v0: b64x2):
+    v1 = bmask.i64x2 v0
+    return v1
+}
+; run: %bmask_i64x2([true false]) == [-1 0]

cranelift/interpreter/src/step.rs

@@ -767,7 +767,6 @@ where
         | Opcode::Breduce
         | Opcode::Bextend
         | Opcode::Bint
-        | Opcode::Bmask
         | Opcode::Ireduce => assign(Value::convert(
             arg(0)?,
             ValueConversionKind::Exact(ctrl_ty),
@@ -802,6 +801,19 @@ where
                 .collect::<ValueResult<Vec<_>>>()?;
             assign(vectorizelanes(&new_vec, new_type)?)
         }
+        Opcode::Bmask => assign({
+            let bool = arg(0)?;
+            let bool_ty = ctrl_ty.as_bool_pedantic();
+            if ctrl_ty.is_vector() {
+                let lanes = extractlanes(&bool, bool_ty.lane_type())?
+                    .into_iter()
+                    .map(|lane| lane.convert(ValueConversionKind::Exact(ctrl_ty.lane_type())))
+                    .collect::<ValueResult<SimdVec<V>>>()?;
+                vectorizelanes(&lanes, ctrl_ty)?
+            } else {
+                bool.convert(ValueConversionKind::Exact(ctrl_ty))?
+            }
+        }),
         Opcode::Sextend => assign(Value::convert(
             arg(0)?,
             ValueConversionKind::SignExtend(ctrl_ty),

cranelift/interpreter/src/value.rs

@@ -277,11 +277,11 @@ impl Value for DataValue {
                 (DataValue::I64(n), types::I128) => DataValue::I128(n as i128),
                 (DataValue::B(b), t) if t.is_bool() => DataValue::B(b),
                 (DataValue::B(b), t) if t.is_int() => {
-                    let val = if b {
                     // Bools are represented in memory as all 1's
-                        (1i128 << t.bits()) - 1
-                    } else {
-                        0
+                    let val = match (b, t) {
+                        (true, types::I128) => -1,
+                        (true, t) => (1i128 << t.bits()) - 1,
+                        _ => 0,
                     };
                     DataValue::int(val, t)?
                 }