Re: Quick question about GPU-CUDA

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view

Re: Quick question about GPU-CUDA

Andreas Kloeckner

Szu-Pei Fu <[hidden email]> writes:

> I followed your wiki page to run pycuda/test/python on porter
> and got the following error message.
> E           CompileError: nvcc compilation of /tmp/tmpImcdf6/
> failed
> E           [command: nvcc --cubin -arch sm_52
> -I/home/sf47/miniconda3/envs/inteq/lib/python3.5/site-packages/pycuda/cuda
> E           [stderr:
> E           /usr/include/c++/8/type_traits(1049): error: type name is not
> allowed
> E
> E           /usr/include/c++/8/type_traits(1049): error: type name is not
> allowed
> E
> E           /usr/include/c++/8/type_traits(1049): error: identifier
> "__is_assignable" is undefined
> E
> E           3 errors detected in the compilation of
> "/tmp/tmpxft_0000560f_00000000-6_kernel.cpp1.ii".
> E
> ]../../../.local/lib/python2.7/site-packages/pycuda/
> CompileError
> I'm wondering if I'm missing anything. Thank you for answering my question.
Something is very strange here. You traceback references both Python 2.7
and 3.5. Any clue why that might be the case? At any given moment, only
one version of Python should be active.

Next, the failure you're observing is nvcc refusing to compile some
source code. Could you try your nvcc in isolation? (e.g. with code from
here [1]) If that still fails, you should fix your CUDA installation and
then retry with PyCUDA.



CC'd the PyCUDA list for for archival.

PyCUDA mailing list
[hidden email]

signature.asc (847 bytes) Download Attachment