not able to run pytorch-kernel about gpgpu-sim_distribution HOT 7 CLOSED

Hello, I have similar problem with you. (Issue #110)
Could you tell me which line of TCHApply.cuh makes the problem, exactly?

Thank you.

from gpgpu-sim_distribution.

Hi, I think your problem is different from my error. my error occurs from gpgpu-sim, not pytorch. cudaGetDeviceCount is a basic api defined in gpgpu-sim/libcuda/cuda_runtime_api.cc, so you should run it without any problem.

I think gpgpu-sim library link is failed in your environment....

from gpgpu-sim_distribution.

Yes, it was a problem related to library linking. After I set several environment values, it works well.
Thank you.

from gpgpu-sim_distribution.

Hi. This has been a known issue, and it has to do with kernels not being found in libcudnn.so. Please try the instructions in the link below and see if it helps.
https://docs.google.com/document/d/17fSM2vrWodP8rWR7ctpgaggVXEw0uD2VCAh0Gi4Gpb4/edit?usp=sharing

from gpgpu-sim_distribution.

@cng123 Thank you for sharing comment! I didn't try it yet and i will report the result after trial.. However, I have question. It seems it uses static CuDNN library. Then, did you build the cudnn as a static library not a shared library?

from gpgpu-sim_distribution.

I am not sure what you mean by build cudnn as a static library, since I downloaded the static and shared library "as is" from Nvidia's website.

from gpgpu-sim_distribution.

Aha, i forgot that there is official static library from Nvidia. Thank you for the precious comment!

from gpgpu-sim_distribution.