Merge pull request #3316 from dheaton-arm/implement-uwiden

Implement `UwidenLow` and `UwidenHigh` for the interpreter

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

@@ -0,0 +1,26 @@
+test interpret
+test run
+target aarch64
+set enable_simd
+target x86_64 machinst
+
+function %uwidenhigh_i8x16(i8x16) -> i16x8 {
+block0(v0: i8x16):
+    v1 = uwiden_high v0
+    return v1
+}
+; run: %uwidenhigh_i8x16([1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16]) == [9 10 11 12 13 14 15 16]
+
+function %uwidenhigh_i16x8(i16x8) -> i32x4 {
+block0(v0: i16x8):
+    v1 = uwiden_high v0
+    return v1
+}
+; run: %uwidenhigh_i16x8([1 2 3 4 5 6 7 8]) == [5 6 7 8]
+
+function %uwidenhigh_i32x4(i32x4) -> i64x2 {
+block0(v0: i32x4):
+    v1 = uwiden_high v0
+    return v1
+}
+; run: %uwidenhigh_i32x4([1 2 3 4]) == [3 4]

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

@@ -0,0 +1,26 @@
+test interpret
+test run
+target aarch64
+set enable_simd
+target x86_64 machinst
+
+function %uwidenlow_i8x16(i8x16) -> i16x8 {
+block0(v0: i8x16):
+    v1 = uwiden_low v0
+    return v1
+}
+; run: %uwidenlow_i8x16([1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16]) == [1 2 3 4 5 6 7 8]
+
+function %uwidenlow_i16x8(i16x8) -> i32x4 {
+block0(v0: i16x8):
+    v1 = uwiden_low v0
+    return v1
+}
+; run: %uwidenlow_i16x8([1 2 3 4 5 6 7 8]) == [1 2 3 4]
+
+function %uwidenlow_i32x4(i32x4) -> i64x2 {
+block0(v0: i32x4):
+    v1 = uwiden_low v0
+    return v1
+}
+; run: %uwidenlow_i32x4([1 2 3 4]) == [1 2]

cranelift/interpreter/src/step.rs

@@ -824,8 +824,24 @@ where
         )?),
         Opcode::SwidenLow => unimplemented!("SwidenLow"),
         Opcode::SwidenHigh => unimplemented!("SwidenHigh"),
-        Opcode::UwidenLow => unimplemented!("UwidenLow"),
-        Opcode::UwidenHigh => unimplemented!("UwidenHigh"),
+        Opcode::UwidenLow => {
+            let new_type = ctrl_ty.merge_lanes().unwrap();
+            let new_vec = extractlanes(&arg(0)?, ctrl_ty.lane_type())?
+                .into_iter()
+                .take(new_type.lane_count() as usize)
+                .map(|lane| lane.convert(ValueConversionKind::ZeroExtend(new_type.lane_type())))
+                .collect::<ValueResult<Vec<_>>>()?;
+            assign(vectorizelanes(&new_vec, new_type)?)
+        }
+        Opcode::UwidenHigh => {
+            let new_type = ctrl_ty.merge_lanes().unwrap();
+            let new_vec = extractlanes(&arg(0)?, ctrl_ty.lane_type())?
+                .into_iter()
+                .skip(new_type.lane_count() as usize)
+                .map(|lane| lane.convert(ValueConversionKind::ZeroExtend(new_type.lane_type())))
+                .collect::<ValueResult<Vec<_>>>()?;
+            assign(vectorizelanes(&new_vec, new_type)?)
+        }
         Opcode::FcvtToUint => unimplemented!("FcvtToUint"),
         Opcode::FcvtToUintSat => unimplemented!("FcvtToUintSat"),
         Opcode::FcvtToSint => unimplemented!("FcvtToSint"),

cranelift/interpreter/src/value.rs

@@ -290,9 +290,14 @@ impl Value for DataValue {
                 _ => unimplemented!("conversion: {} -> {:?}", self.ty(), kind),
             },
             ValueConversionKind::ZeroExtend(ty) => match (self, ty) {
+                (DataValue::U8(n), types::I16) => DataValue::U16(n as u16),
+                (DataValue::U8(n), types::I32) => DataValue::U32(n as u32),
+                (DataValue::U8(n), types::I64) => DataValue::U64(n as u64),
                 (DataValue::I8(n), types::I16) => DataValue::I16(n as u8 as i16),
                 (DataValue::I8(n), types::I32) => DataValue::I32(n as u8 as i32),
                 (DataValue::I8(n), types::I64) => DataValue::I64(n as u8 as i64),
+                (DataValue::U16(n), types::I32) => DataValue::U32(n as u32),
+                (DataValue::U16(n), types::I64) => DataValue::U64(n as u64),
                 (DataValue::I16(n), types::I32) => DataValue::I32(n as u16 as i32),
                 (DataValue::I16(n), types::I64) => DataValue::I64(n as u16 as i64),
                 (DataValue::U32(n), types::I64) => DataValue::U64(n as u64),