[Onnx] Pow support for other types (#8933)

* update pow

* update pow

* remove duplicate

Co-authored-by: Andrew Zhao Luo <andrewzhaoluo@system76-pc.localdomain>

python/tvm/relay/frontend/onnx.py

@@ -1021,6 +1021,32 @@ def _impl_v1(cls, inputs, attr, params):
         return _op.log(_op.exp(beta * inputs[0]) + _expr.const(1.0)) * alpha
 
 
+class Pow(OnnxOpConverter):
+    """Operator converter for Pow."""
+
+    @classmethod
+    def _impl_v13(cls, inputs, attr, params):
+        x = inputs[0]
+        y = inputs[1]
+
+        x_type = infer_type(x).checked_type.dtype
+        output_type = x_type
+        y_type = infer_type(y).checked_type.dtype
+
+        if not x_type.startswith("float"):
+            x_type = "float32"
+            x = _op.cast(x, x_type)
+
+        if x_type != y_type:
+            y = _op.cast(y, x_type)
+
+        # TODO: come up with good default integer pow() func for common backends
+        result = _op.power(x, y)
+        if x_type != output_type:
+            return _op.cast(result, output_type)
+        return result
+
+
 class Prelu(OnnxOpConverter):
     """Operator converter for Prelu."""
 
@@ -3654,7 +3680,7 @@ def _get_convert_map(opset):
         "Sinh": Renamer("sinh"),
         "Tan": Renamer("tan"),
         "Tanh": Renamer("tanh"),
-        "Pow": Renamer("power"),
+        "Pow": Pow.get_converter(opset),
         "PRelu": Prelu.get_converter(opset),
         "Sigmoid": Renamer("sigmoid"),
         "HardSigmoid": HardSigmoid.get_converter(opset),

tests/python/frontend/onnx/test_forward.py

@@ -4788,16 +4788,6 @@ def verify_eyelike(indata):
     # This nllloss test is flaky and sometimes gives NaNs
     # Investigate it here: https://github.com/apache/tvm/issues/8918
     "test_nllloss_NCd1d2d3_none_no_weight_negative_ii",
-    "test_pow_types_float",
-    "test_pow_types_float32_int32",
-    "test_pow_types_float32_int64",
-    "test_pow_types_float32_uint32",
-    "test_pow_types_float32_uint64",
-    "test_pow_types_int",
-    "test_pow_types_int32_float32",
-    "test_pow_types_int32_int32",
-    "test_pow_types_int64_float32",
-    "test_pow_types_int64_int64",
     "test_qlinearmatmul_2D",
     "test_qlinearmatmul_3D",
     "test_range_float_type_positive_delta_expanded",