implement arith.bitcast operation

tests/dialects/test_arith.py

@@ -7,6 +7,7 @@
     AddiOp,
     AddUIExtendedOp,
     AndIOp,
+    BitcastOp,
     CeilDivSIOp,
     CeilDivUIOp,
     CmpfOp,
@@ -55,6 +56,7 @@
     IndexType,
     IntegerAttr,
     IntegerType,
+    Signedness,
     TensorType,
     VectorType,
     f32,
@@ -249,6 +251,27 @@ def test_select_op():
     assert select_f_op.result.type == f.result.type
 
 
+def test_bitcast_op():
+    signedness = [Signedness.SIGNED, Signedness.UNSIGNED, Signedness.SIGNLESS]
+
+    for bitwidth in [1, 2, 4, 5, 8, 16, 32, 64, 128]:
+        for from_sign in signedness:
+            for to_sign in signedness:
+                if from_sign == to_sign:
+                    continue
+
+                # here create variable of from type
+                value = 0 if bitwidth in [1, 2] else 5
+                a = ConstantOp(
+                    IntegerAttr(value, IntegerType(bitwidth, signedness=from_sign))
+                )
+                cast = BitcastOp(a, IntegerType(bitwidth, signedness=to_sign))
+
+                assert cast.result.type == IntegerType(bitwidth, signedness=to_sign)
+                assert cast.input.type == IntegerType(bitwidth, signedness=from_sign)
+                assert cast.input.owner == a
+
+
 def test_index_cast_op():
     a = ConstantOp.from_int_and_width(0, 32)
     cast = IndexCastOp(a, IndexType())

xdsl/dialects/arith.py

@@ -1223,6 +1223,40 @@ class MinnumfOp(FloatingPointLikeBinaryOperation):
     traits = traits_def(Pure())
 
 
+@irdl_op_definition
+class BitcastOp(IRDLOperation):
+    """
+    Reinterpret a value as another type of equal bitwidth without changing
+    the underlying representation.
+    """
+
+    name = "arith.bitcast"
+
+    input = operand_def(IntegerType | Float16Type | Float32Type | Float64Type)
+
+    result = result_def(IntegerType | Float16Type | Float32Type | Float64Type)
+
+    assembly_format = "$input attr-dict `:` type($input) `to` type($result)"
+
+    def __init__(self, input_arg: SSAValue | Operation, target_type: Attribute):
+        super().__init__(operands=[input_arg], result_types=[target_type])
+
+    def verify_(self) -> None:
+        assert isinstance(
+            self.input.type, IntegerType | Float16Type | Float32Type | Float64Type
+        )
+        assert isinstance(
+            self.result.type, IntegerType | Float16Type | Float32Type | Float64Type
+        )
+
+        ibw = self.input.type.width
+        obw = self.result.type.width
+        if ibw == obw:
+            raise VerifyException(
+                f"'arith.bitcast' can only be used on types with equal bitwidths, found {ibw} and {obw}"
+            )
+
+
 @irdl_op_definition
 class IndexCastOp(IRDLOperation):
     name = "arith.index_cast"