Requires bool I/O but node can not be handled by Myelin

[549981178]

Description

Slice operator does not support bool input.

Environment

TensorRT Version: 8.2
GPU Type: A10
Nvidia Driver Version: 4.7
CUDA Version: 11.4

[spolisetty]

Hi,

Slice operator does support the bool data type. onnx-tensorrt/operators.md at main · onnx/onnx-tensorrt · GitHub

Please try on the latest TensorRT version 8.4.
https://developer.nvidia.com/nvidia-tensorrt-8x-download

If you still face this issue, we recommend you to please share with us the issue repro ONNX model and trtexec --verbose logs for better debugging.

Thank you.

[wangsenhao1]

I meet the same problem ,and my tensorrt is nvidia-tensorrt==8.4.0.6,and the information is :
[05/03/2022-20:19:31] [TRT] [E] 2: [myelinBuilderUtils.cpp::operator()::371] Error Code 2: Internal Error (Slice_79 requires bool I/O but node can not be handled by Myelin. Dynamic shapes are not equal.)
[05/03/2022-20:19:31] [TRT] [E] 2: [builder.cpp::buildSerializedNetwork::619] Error Code 2: Internal Error (Assertion engine != nullptr failed. )
Failed building engine!

[wangsenhao1]

verbose log is :
[TRT] [V] Parsing node: Slice_79 [Slice]
[TRT] [V] Searching for input: 552
[TRT] [V] Searching for input: 605
[TRT] [V] Searching for input: 606
[TRT] [V] Searching for input: 604
[TRT] [V] Searching for input: 607
[TRT] [V] Slice_79 [Slice] inputs: [552 → (-1, 1, -1)[BOOL]], [605 → (1)[INT32]], [606 → (1)[INT32]], [604 → (1)[INT32]], [607 → (1)[INT32]],
[TRT] [V] Registering layer: Slice_79 for ONNX node: Slice_79
[TRT] [V] Registering tensor: 608 for ONNX tensor: 608
[TRT] [V] Slice_79 [Slice] outputs: [608 → (-1, 1, -1)[BOOL]],

[spolisetty]

Hi,

Could you please share with us ONNX model and complete trtexec --verbose logs for better debugging.

Thank you.

[wangsenhao1]

encoder-result-FP32.txt (1.1 MB)
this is complete verbose logger

[spolisetty]

Hi,

We are looking into this.
Our team needs an ONNX model to try from our and work on the fix. Could you please share with us the minimal issue repro ONNX model. You can DM us as well to send privately.

Thank you.

[549981178]

I have solved the problem.
I added the cast node after the not node to cast to INT32 in onnx model.