Seg fault when converting model to INT8, DLA

Robotics & Edge Computing Jetson & Embedded Systems Jetson AGX Xavier
1

I am trying to convert a particular model from ONNX to INT8 to run on DLA. I am getting a segmentation fault when running the conversion in C++. This model runs on the GPU fine. I have other models that run on DLA and GPU fine.

I am running Jetpack 4.5.1 on Xavier AGX.

3

Hi,

Did you run the conversion with trtexec?
If yes, could you share the trtexec log with --verbose?

Thanks.

4

I don’t use trtexec, but NVIDIA support on the TensorRT forum pointed me several times to this sample:

https://github.com/NVIDIA/TensorRT/blob/master/samples/opensource/sampleINT8/README.md

I have models which work with both GPU and DLA, but one particular model is seg faulting when I try DLA. I can share the ONNX model with you if you like.

5

Hi,

Sure. Please share the ONNX model so we can check it further.
More, trtexec with --verbose can also provide some useful information for us.

Thanks.