[TensorRT] Jetson Orin NX에 TensorRT 설치

 

 

1. 아래 깃허브 레포 들어가서 소스 다운로드 및 TensorRT 설치

https://github.com/NVIDIA-AI-IOT/torch2trt

GitHub - NVIDIA-AI-IOT/torch2trt: An easy to use PyTorch to TensorRT converter

 

pip install packaging

git clone https://github.com/NVIDIA-AI-IOT/torch2trt
cd torch2trt
sudo -E python setup.py install

 

 

2. plugin 설치

# vi CMakeLists.txt

# 다음줄 바로 아래에 추가
# set_property(TARGET torch2trt_plugins PROPERTY CUDA_ARCHITECTURES ${CUDA_ARCHITECTURES}) 

target_compile_options(torch2trt_plugins PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:-gencode arch=compute_72,code=sm_72>)

 

sudo cmake -B build -DCMAKE_CUDA_COMPILER=/usr/local/cuda/bin/nvcc  . && sudo cmake --build build --target install && sudo ldconfig

 

 

* 참고

 

https://github.com/NVIDIA-AI-IOT/torch2trt

GitHub - NVIDIA-AI-IOT/torch2trt: An easy to use PyTorch to TensorRT converter

 

(디버그)

https://youtrack.jetbrains.com/issue/CPP-29089/No-CMAKECUDACOMPILER-could-be-found-but-nvcc-is-right-there

"No CMAKE_CUDA_COMPILER could be found" - but nvcc is right there : CPP-29089

 

https://github.com/NVIDIA-AI-IOT/torch2trt/issues/794

torch2trt/plugins/src/example_plugin.cu(27): error: identifier "__hmul" is undefined · Issue #794 · NVIDIA-AI-IOT/torch2trt