Error while installing

[ry85]

Hi, I am getting the following error "cannot find -lcuda", when installing via a local clone of the tcnn repo,

git clone https://github.com/NVlabs/tiny-cuda-nn.git cd bindings/torch python setup.py install

I had lcuda installed, but unfortunately it is not getting found. I find this thread https://stackoverflow.com/a/34256630 which suggests modifying the compile file to explicitly provide the path of the lcuda library (-L/us/lib/x86_64-linux-gnu , this directory contains all cuda libraries including lcuda).

Can you please help as to which compile file I have to modify. Sorry, I am completely new to this.

[aejion]

@ry85 have you solved this problem? i met the same problem as well

[RohanChacko]

I face the same error.

I had to install GCC 9.3.0 inside my conda env since the system GCC ver. is 5.4.0 and cannot compile tcnn. The libcuda.so file is in /usr/local/cuda-11.3/lib64/stubs/ and my LD_LIBRARY_PATH=/usr/local/cuda-11.3/lib64. But my conda GCC is not able to locate this file.

I am not sure how to modify the CMakelists.txt file to add an explicit path to the file. Any help would be appreciated. I am running on a cluster so do not have sudo permissions as well. Thanks!

[jzhangbs]

I solve this problem by putting the dir with libcuda.so in LIBRARY_PATH (instead of LD_LIBRARY_PATH).

[Talegqz]

I solve this problem by putting the dir with libcuda.so in LIBRARY_PATH (instead of LD_LIBRARY_PATH).

Could you please provide more detail steps? Thanks for your help!!!