Dynamic support for split (#2871)

Author: cehongwang

py/torch_tensorrt/dynamo/conversion/aten_ops_converters.py

@@ -629,14 +629,19 @@ def aten_ops_softmax(
 
 
 @dynamo_tensorrt_converter(
-    torch.ops.aten.split.Tensor, capability_validator=has_static_shapes_in_args([1])
+    torch.ops.aten.split.Tensor,
+    capability_validator=has_static_shapes_in_args([1]),
+    supports_dynamic_shapes=True,
 )
 @dynamo_tensorrt_converter(
-    torch.ops.aten.split.sizes, capability_validator=has_static_shapes_in_args([1])
+    torch.ops.aten.split.sizes,
+    capability_validator=has_static_shapes_in_args([1]),
+    supports_dynamic_shapes=True,
 )
 @dynamo_tensorrt_converter(
     torch.ops.aten.split_with_sizes.default,
     capability_validator=has_static_shapes_in_args([1]),
+    supports_dynamic_shapes=True,
 )
 def aten_ops_split(
     ctx: ConversionContext,

tests/py/dynamo/conversion/test_convolution_aten.py

@@ -1,7 +1,6 @@
 import torch
 from parameterized import param, parameterized
 from torch.testing._internal.common_utils import run_tests
-
 from torch_tensorrt import Input
 
 from .harness import DispatchTestCase

tests/py/dynamo/conversion/test_split_aten.py

@@ -119,6 +119,7 @@ def forward(self, input):
     @parameterized.expand(
         [
             ("select_split_size_or_sections_dim_dynamic_shape", 2, 1),
+            ("select_split_size_or_sections_non_divisible_dim_dynamic_shape", 3, 1),
         ]
     )
     def test_split_dynamic(self, _, split_size_or_tensor, dim):
@@ -132,9 +133,37 @@ def forward(self, input):
 
         input_specs = [
             Input(
-                shape=(1, 10, -1),
                 dtype=torch.float32,
-                shape_ranges=[((1, 10, 1), (1, 10, 10), (1, 10, 10))],
+                min_shape=[1, 10, 1],
+                opt_shape=[1, 10, 10],
+                max_shape=[1, 10, 10],
+            ),
+        ]
+        self.run_test_with_dynamic_shape(
+            TestModule(),
+            input_specs,
+        )
+
+    @parameterized.expand(
+        [
+            ("select_split_size_or_sections_dim_dynamic_shape_on_first_axis", 2, 1),
+        ]
+    )
+    def test_split_dynamic_first_axis_dynamic(self, _, split_size_or_tensor, dim):
+        class TestModule(torch.nn.Module):
+            def __init__(self):
+                super().__init__()
+
+            def forward(self, input):
+                out = torch.ops.aten.split.Tensor(input, split_size_or_tensor, dim)
+                return out
+
+        input_specs = [
+            Input(
+                dtype=torch.float32,
+                min_shape=[1, 10, 10],
+                opt_shape=[3, 10, 10],
+                max_shape=[5, 10, 10],
             ),
         ]
         self.run_test_with_dynamic_shape(