For DeepStream 6.1.1 GA and 6.2 GA, the TensorRT OSS plugin is not needed.
Below are the steps to build TensorRT OSS on Jetson device for Jetson libnvinfer_plugin.so. For cross-compiling, refer to TensorRT OSS README.
libnvinfer_plugin.so.8.6.2 provided in this folder was built with:
Jetson Xavier Jetpack GA (CUDA-12.2, cuDNN v8.6, TensorRT8.6.1.2 )
Note
You can get the prebuild lib by using "wget https://nvidia.box.com/shared/static/1ih810ui4z52nvwznfk1xaxvdej3mauw -O libnvinfer_plugin.so.8.6.2" if you met some LFS issue.
The TensorRT OSS source code does not contains the patch for 8.6 branch
| DeepStream Release | Jetpack Version | TRT Version | TRT_OSS_CHECKOUT_TAG |
|---|---|---|---|
| 5.0 | 4.4 GA - 4.5 | TRT 7.1.3 | release/7.1 |
| 5.0.1 | 4.4 GA - 4.5 | TRT 7.1.3 | release/7.1 |
| 5.1 | 4.5.1 | TRT 7.1.3 | release/7.1 |
| 6.0 EA | 4.5.1 | TRT 7.1.3 | release/7.1 |
| 6.0 GA | 4.6 | TRT 8.0.1 | release/8.0 |
| 6.0.1 | 4.6.1 / 4.6.2 | TRT 8.2.1 | release/8.2 |
| 6.1 | 5.0.1 | TRT 8.4.0.11 | release/8.4 |
| 6.1.1 | 5.0.2 | TRT 8.4.1 | OSS not needed |
| 6.2 | 5.1 | TRT 8.5.1 | OSS not needed |
| 6.3 | 5.1.2 | TRT 8.5.3 | OSS not needed |
| 6.4 | 6.0 | TRT 8.6.1.2 | binary only |
| 7.0 | 6.0 | TRT 8.6.1.2 | binary only |
### Replace "libnvinfer_plugin.so*"
sudo mv /usr/lib/aarch64-linux-gnu/libnvinfer_plugin.so.8.x.y ${HOME}/libnvinfer_plugin.so.8.x.y.bak // backup original libnvinfer_plugin.so.x.y sudo cp $DOWNLOAD_PATH/libnvinfer_plugin.so.8.m.n /usr/lib/aarch64-linux-gnu/libnvinfer_plugin.so.8.x.y sudo ldconfig
## GPU_ARCHS
1. GPU_ARCHs value can be got from "deviceQuery" CUDA sample
cd /usr/local/cuda/samples/1_Utilities/deviceQuery sudo make ./deviceQuery
2. Can also find GPU_ARCHs from below table
| Jetson Platform | GPU_ARCHS |
| --------------- | --------- |
| TX1 / NANO | 53 |
| TX2 | 62 |
| Xavier / NX | 72 |
| Orin | 87 |