Arch Linux

Archer1412

Member

Registered: 2025-02-08

Posts: 15

CUDA TensorRT Configuration Help

Hi everyone, I'm new to the arch forum and I hope you can give me a solution that I can't find, I've tried everything. I'm a go player and I'm trying to compile katago tensor version, as for cuda, cudnn I managed to configure it without problems but I can't find the tensorRT libraries when I try to compile engine I get this error:

CMake Error at CMakeLists.txt:301 (message):
NvInfer.h was NOT found, specify TENSORRT_INCLUDE_DIR to indicate
where it is.

and I set the environment variables like this on ~/.bashrc:

export PATH=/opt/cuda/bin:$PATH
export LD_LIBRARY_PATH=/opt/cuda/lib64:$LD_LIBRARY_PATH
export LIBRARY_PATH=/opt/cuda/lib64:$LIBRARY_PATH

export PATH=/opt/TensorRT-10.8.0.43/bin:$PATH
export LD_LIBRARY_PATH=/opt/TensorRT-10.8.0.43/lib:$LD_LIBRARY_PATH
export TENSORRT_INCLUDE_DIR=/opt/TensorRT-10.8.0.43/include/NvInfer.h

what am I doing wrong? I installed cuda TensoRT from the official site and put the archive content in /opt/TensorRT. I don't know what else to do do you have any ideas? Thanks for the help

Last edited by Archer1412 (2025-02-17 20:54:45)

patchouli

Member

Registered: 2024-09-18

Posts: 22

Re: CUDA TensorRT Configuration Help

Try the tensorrt package from AUR. I personally use it, and it works fine.

lawmurray

Member

From: Bangkok

Registered: 2025-02-10

Posts: 2

Website

Re: CUDA TensorRT Configuration Help

This bit:

export TENSORRT_INCLUDE_DIR=/opt/TensorRT-10.8.0.43/include/NvInfer.h

Try with just the directory there, not the file, i.e.

export TENSORRT_INCLUDE_DIR=/opt/TensorRT-10.8.0.43/include

In case you're not aware also, rather than putting these in .bashrc at first, you can just put them in your terminal, make changes, iterate, to get it working before putting them in .bashrc. But this particular variable TENSORRT_INCLUDE_DIR I suspect you will never need in .bashrc. It's just a build time thing, so once built and installed it won't be needed.

Last edited by lawmurray (2025-02-18 01:10:31)

Archer1412

Member

Registered: 2025-02-08

Posts: 15

Re: CUDA TensorRT Configuration Help

So I tried to change the environment variable and put it on the terminal:

export TENSORRT_INCLUDE_DIR=/opt/TensorRT-10.8.0.43/include

but it doesn't work it always gives me this error:

CMake Error at CMakeLists.txt:301 (message):
NvInfer.h was NOT found, specify TENSORRT_INCLUDE_DIR to indicate
where it is.

As for the installation via AUR I had already tried but it didn't work it gave me a lot of errors.

UPDATE:

So I tried to install the tensorrt package from aur again and I don't know why now it did what it should strange before it didn't work but it gives me this error during installation:

Make Error at /usr/share/cmake/Modules/CMakeDetermineCXXCompiler.cmake:48 (message):
Could not find compiler set in environment variable CXX:

/opt/cuda/bin/g++.

Call Stack (most recent call first):
CMakeLists.txt:3 (project)

CMake Error: CMAKE_CXX_COMPILER not set, after EnableLanguage
-- Configuring incomplete, errors occurred!
==> ERROR: An error occurred in build().
The operation is about to be aborted...
-> error while building: python-pytorch-tensorrt-exit status 4

but despite everything it works. What is causing the error? Thanks for the help

Last edited by Archer1412 (2025-02-18 09:02:28)