Empty_like evaluator

Author: apbose

py/torch_tensorrt/dynamo/conversion/ops_evaluators.py

@@ -47,3 +47,14 @@ def aten_ops_arange_start_step(
     name: str,
 ) -> Union[TRTTensor, Sequence[TRTTensor]]:
     return np.arange(*args)
+
+
+@dynamo_tensorrt_converter(torch.ops.aten.empty_like.default)
+def aten_ops_empty_like(
+    ctx: ConversionContext,
+    target: Target,
+    args: Tuple[Argument, ...],
+    kwargs: Dict[str, Argument],
+    name: str,
+) -> TRTTensor:
+    return np.empty_like(*args)

tests/py/dynamo/conversion/test_empty_like.py

@@ -0,0 +1,40 @@
+import torch
+import torch.nn as nn
+from parameterized import parameterized
+from torch.testing._internal.common_utils import run_tests
+
+from .harness import DispatchTestCase
+
+
+class TestEmptyLikeConverter(DispatchTestCase):
+    @parameterized.expand(
+        [
+            ((1), torch.int),
+            ((1, 2), torch.int),
+            ((2, 1), torch.int),
+            ((2, 2), torch.int),
+            ((1, 2, 3), torch.int),
+            ((1, 3, 2), torch.int),
+            ((1), torch.int32),
+            ((1, 2), torch.int32),
+            ((2, 1), torch.int32),
+            ((2, 2), torch.int32),
+            ((1, 2, 3), torch.int32),
+            ((1, 3, 2), torch.int32),
+        ]
+    )
+    def test_empty_like(self, input_shape, dtype):
+        class Empty_Like(nn.Module):
+            def forward(self, x):
+                return torch.ops.aten.empty_like(x, dtype)
+
+        inputs = [torch.empty(input_shape)]
+        self.run_test(
+            Empty_Like(),
+            inputs,
+            use_dynamo_tracer=True,
+        )
+
+
+if __name__ == "__main__":
+    run_tests()