Problem with absolute path to nvcc in OpenMM 5.0.0

The functionality of OpenMM will (eventually) include everything that one would need to run modern molecular simulation.
POST REPLY
User avatar
Fabian Paul
Posts: 6
Joined: Wed Apr 11, 2012 10:41 am

Problem with absolute path to nvcc in OpenMM 5.0.0

Post by Fabian Paul » Thu Feb 14, 2013 4:12 am

Hi folks,

I just tried out the new 5.0 version of OpenMM. I used the binary 64bit package for Linux.
However, the testInstallation.py script failed with the message

| Error launching CUDA compiler: 32512
| sh: /usr/local/cuda/bin/nvcc: No such file or directory

Now it turned out that my CUDA installation was in a different directory.
I would suggest that you change the absolute path '/usr/local/cuda/bin/nvcc'
that is statically compiled into the libOpenMMCUDA.so plugin to something
more flexible like $CUDA_HOME/bin/nvcc.

Thanks for all the work on OpenMM and
regards,
Fabian

User avatar
Peter Eastman
Posts: 2593
Joined: Thu Aug 09, 2007 1:25 pm

Re: Problem with absolute path to nvcc in OpenMM 5.0.0

Post by Peter Eastman » Thu Feb 14, 2013 10:57 am

Hi Fabian,

Set the OPENMM_CUDA_COMPILER environment variable to point to nvcc. Alternatively, you can specify the location in your code by setting the CudaCompiler property when you create your Context. See chapter 11 of the Users Guide.

Peter

POST REPLY