Seg fault when converting model to INT8, DLA

I have a particular model that causes a seg fault when converting to INT8 for DLA on Jetson Xavier. The model works fine when converting to INT8 using GPU. My other models convert fine to GPU and DLA.

I am using the C++ API. I can share the model if needed.

Environment

TensorRT Version: Jetpack 4.5.1
GPU Type: Jetson Xavier AGX

Hi, Please refer to the below links to perform inference in INT8
https://github.com/NVIDIA/TensorRT/blob/master/samples/opensource/sampleINT8/README.md

Thanks!

I guess my question is that with 1 particular model I have, it seg faults when I convert it to INT8 targeting DLA. The model does not segfault when I target GPU.

I am able to convert my other models to both INT8 and GPU without problem.

Hi @jseng,

We recommend you to go through TensorRT user guide talks about working with DLA and running sample.

If you need further assistance, we request you to post your query on Jetson AGX Xavier forum. You may get better help here.

Thank you.