Closed henrypearce4D closed 4 months ago
Just a guess but the spaces in the path might be tripping it up? You could try adding speech marks to book-end your CUDA path. Also cmake and back slashes has caused me trouble in the past....
Thats odd. Which shell and OS are you using?
As a quick fix, you could probably exchange the $ENV:CUDA_PATH
part with C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8
in both -DCMAKE_CUDA_COMPILER
and DCUDA_P_PATH
Also maybe remove the build folder and retry, cmake stores compiler information there.
I'm using conda and windows 10.
Exchanging those parts it started to compile then shows a CuDNN error. That was mentioned as a software requirement in the Windows install but not instructions so I skipped past it. Installing CuDNN and that exchange to cuda_path its now compiling.
I've added a PR for those extra notes as part of the install process.
Just replace $ENV:CUDA_PATH
with %CUDA_PATH%
.
@tridemax thanks!
I had to also change $ENV:CONDA_PREFIX
with %CONDA_PREFIX%
and got it to compile!
cmake -Bbuild -DCMAKE_CUDA_COMPILER="%CUDA_PATH%\bin\nvcc.exe" -DCMAKE_PREFIX_PATH=".\External\libtorch" -DCONDA_P_PATH="%CONDA_PREFIX%" -DCUDA_P_PATH="%CUDA_PATH%" -DCMAKE_BUILD_TYPE=RelWithDebInfo .
@lfranke changes have been added to the PR
Getting an error when running;
cmake -Bbuild -DCMAKE_CUDA_COMPILER="$ENV:CUDA_PATH\bin\nvcc.exe" -DCMAKE_PREFIX_PATH=".\External\libtorch" -DCONDA_P_PATH="$ENV:CONDA_PREFIX" -DCUDA_P_PATH="$ENV:CUDA_PATH" -DCMAKE_BUILD_TYPE=RelWithDebInfo .
I have CUDA_PATH set as
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8
AndC:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\bin
set in my path variable.Any suggestions?