[LightGBM] build v4.3.0 with OpenCL/CUDA support #8593
Conversation
|
@giordano ok to merge? |
|
I quickly skimmed through it and it looks ok, but is there an upstream issue to switch to FindCUDAToolkit? The deprecated FindCUDA module recently caused me lots of wasted time, it should really not be used. |
|
I will make a patch here and make an issue upstream |
|
can we check #8691? |
|
@giordano what do you think now? |
|
Are we ok to merge @giordano ? |
L/LightGBM/build_tarballs.jl
Outdated
| export CUDA_PATH=${WORKSPACE}/destdir/cuda | ||
| export PATH=$PATH:$CUDA_PATH/bin/ | ||
|
|
||
| ln -s $CUDA_PATH/lib $CUDA_PATH/lib64 |
There was a problem hiding this comment.
I think this leaves a broken symlink in the final tarballs?
There was a problem hiding this comment.
You are right, I did that so cmake could find the libs, I'll check a better approach
There was a problem hiding this comment.
I found that the issue was in the enable_language(CUDA) call, it was not using the library path provided, so I'm forcing it now with CMAKE_CUDA_FLAGS, is this a good approach?
* build LightGBM v4.3.0 with OpenCL/CUDA support * fix: replace FindCuda with FindCUDAToolkit * 😶 oops * patch switch to findcudatoolkit * boost pin compat * Update build_tarballs.jl * fix: add libs to CMAKE_CUDA_FLAGS * Update L/LightGBM/build_tarballs.jl --------- Co-authored-by: Mosè Giordano <[email protected]>
No description provided.