add converter registration

Author: bowang007

py/torch_tensorrt/dynamo/conversion/aten_ops_converters.py

@@ -357,14 +357,14 @@ def aten_ops_softmax(
 
 @dynamo_tensorrt_converter(
     torch.ops.aten.split.Tensor, capability_validator=dynamic_unsupported_with_args([1])
-)
+)  # type: ignore[misc]
 @dynamo_tensorrt_converter(
     torch.ops.aten.split.sizes, capability_validator=dynamic_unsupported_with_args([1])
-)
+)  # type: ignore[misc]
 @dynamo_tensorrt_converter(
     torch.ops.aten.split_with_sizes.default,
     capability_validator=dynamic_unsupported_with_args([1]),
-)
+)  # type: ignore[misc]
 def aten_ops_split(
     network: TRTNetwork,
     target: Target,
@@ -1378,3 +1378,22 @@ def aten_ops_linear(
         weight=args[1],
         bias=args_bounds_check(args, 2, None),
     )
+
+
+@dynamo_tensorrt_converter(torch.ops.aten.argmax.default)  # type: ignore[misc]
+def aten_ops_argmax(
+    network: TRTNetwork,
+    target: Target,
+    args: Tuple[Argument, ...],
+    kwargs: Dict[str, Argument],
+    name: str,
+) -> Union[TRTTensor, Sequence[TRTTensor]]:
+    return impl.argmax.argmax(
+        network,
+        target,
+        SourceIR.ATEN,
+        name,
+        input=args[0],
+        dim=args_bounds_check(args, 1),
+        keep_dim=args_bounds_check(args, 2),
+    )