[ONNX]Support Opset 13 split IFF the split is a constant (apache#9643)

python/tvm/relay/frontend/onnx.py

@@ -1493,6 +1493,31 @@ def _impl_v1(cls, inputs, attr, params):
             output = output[0]
         return output
 
+    @classmethod
+    def _impl_v13(cls, inputs, attr, params):
+        splits = inputs[1]
+        splits_rank = None
+        if splits is not None:
+            splits_rank = len(infer_shape(splits))
+        if splits is not None and splits_rank > 0:
+            if isinstance(splits, _expr.Constant):
+                splits = splits.data.asnumpy()
+                indices = []
+                index = 0
+                for i in splits[:-1]:
+                    index += i
+                    indices.append(index)
+            else:
+                raise ValueError("Dynamic Split not yet supported")
+        # When splits isnt specified divide evenly over axis.
+        else:
+            indices = attr["tvm_custom"]["num_outputs"]
+        output = _op.split(inputs[0], indices, attr.get("axis", 0))
+        # If the output of split is a single value, unpack if from the TupleWrapper
+        if len(output) == 1:
+            output = output[0]
+        return output
+
 
 class Slice(OnnxOpConverter):
     """Operator converter for Slice."""

tests/python/frontend/onnx/test_forward.py

@@ -61,7 +61,6 @@ def get_tvm_output_with_vm(
     if not isinstance(input_data, list):
         input_data = [input_data]
     _, shape_dict = get_input_data_shape_dict(graph_def, input_data)
-
     mod, params = relay.frontend.from_onnx(
         graph_def,
         shape_dict,
@@ -167,7 +166,6 @@ def verify_with_ort_with_inputs(
         model.opset_import[0].version = opset
 
     ort_out = get_onnxruntime_output(model, inputs)
-
     if use_vm:
         tvm_out = get_tvm_output_with_vm(
             model,
@@ -1954,7 +1952,9 @@ def verify_split(indata, outdatas, split, axis=0, pass_split=True, opset=11):
                 inputs.append(
                     helper.make_tensor_value_info("split", TensorProto.INT64, list(np_split.shape))
                 )
-                indata = [indata, np_split]
+                # TODO(mbrookhart): Support dynamic split, edit this test case to remove split from
+                # the initializer and add it back to the input data
+                indata = [indata]  # , np_split]
                 initializer.append(
                     helper.make_tensor("split", TensorProto.INT64, list(np_split.shape), np_split)
                 )
@@ -1989,6 +1989,8 @@ def verify_split(indata, outdatas, split, axis=0, pass_split=True, opset=11):
             opset=opset,
             target=target,
             dev=dev,
+            use_vm=True,
+            freeze_params=(opset >= 13),
         )
 
     # 1D
@@ -1997,13 +1999,23 @@ def verify_split(indata, outdatas, split, axis=0, pass_split=True, opset=11):
         [1.0, 2.0, 3.0, 4.0, 5.0, 6.0], [[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], [2, 2, 2], 0, False
     )
     verify_split([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], [[1.0, 2.0], [3.0], [4.0, 5.0, 6.0]], [2, 1, 3], 0)
+    verify_split(
+        [1.0, 2.0, 3.0, 4.0, 5.0, 6.0], [[1.0, 2.0], [3.0], [4.0, 5.0, 6.0]], [2, 1, 3], 0, opset=13
+    )
     # 2D
     verify_split(
         [[1.0, 2.0, 3.0, 4.0], [7.0, 8.0, 9.0, 10.0]],
         [[[1.0, 2.0], [7.0, 8.0]], [[3.0, 4.0], [9.0, 10.0]]],
         [2, 2],
         1,
     )
+    verify_split(
+        [[1.0, 2.0, 3.0, 4.0], [7.0, 8.0, 9.0, 10.0]],
+        [[[1.0, 2.0], [7.0, 8.0]], [[3.0, 4.0], [9.0, 10.0]]],
+        [2, 2],
+        1,
+        opset=13,
+    )
     # Split evenly (unstack)
     verify_split([1, 2, 3], [[1], [2], [3]], False, 0, False)
     # Split a single value to a single value