[gmx-developers] Building GROMACS 5.1/2016 with OpenCL on NVIDA CUDA 6
Szilárd Páll
pall.szilard at gmail.com
Fri May 13 01:57:53 CEST 2016
On Wed, May 11, 2016 at 8:27 PM, Vedran Miletić <rivanvx at gmail.com> wrote:
> 2016-05-11 18:41 GMT+02:00 Szilárd Páll <pall.szilard at gmail.com>:
> > clEnqueueWaitForEvents has been changed to clEnqueueBarrierWithWaitList
> in
> > OpenCL 1.2, but we do have both cases handled:
> >
> http://redmine.gromacs.org/projects/gromacs/repository/revisions/release-5-1/entry/src/gromacs/mdlib/nbnxn_ocl/nbnxn_ocl.cpp#L329
> >
> > So AFAICT you could be using a header that's OpenCL 1.2, but the old
> > libOpenCL.so from NVIDIA.
>
> Indeed [1], and this affects all the users of Debian 8 that use the
> packaged CUDA.
>
That seems to be a broken setup; the API needs to be compatible with the
libraries in the repo, so they need to provide a library that's v1.1 (or
perhaps even nvidia's own lib if
>
> Can I somehow force OpenCL 1.1 to be used, i.e. forcefully undefine
> CL_VERSION_1_2? I tried -U parameter but this does not work [2].
>
As far as I know you can only do that in the code with #undef. We could
manage internally our own CL_VERSION something like it's done with the PTX
version in CUDA (gpu_utils/cuda_arch_utils.cuh) and somehow work around the
Debian issue, but it would be good to know how widespread this is and
whether perhaps there is a better user-side solution.
--
Szilárd
>
> Regards,
> Vedran
>
> [1] https://packages.debian.org/jessie/opencl-headers
> [2]
> https://stackoverflow.com/questions/1978155/how-to-undefine-a-define-at-commandline-using-gcc
>
> --
> Vedran Miletić
> http://vedranmileti.ch/
> --
> Gromacs Developers mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List before
> posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers
> or send a mail to gmx-developers-request at gromacs.org.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20160512/f440a760/attachment-0001.html>
More information about the gromacs.org_gmx-developers
mailing list