[gmx-developers] Re: GROMACS 4.6 pre-release: new algorithms, parallelization schemes, & GPU acceleration

Szilárd Páll szilard.pall at cbr.su.se
Tue Mar 6 15:32:08 CET 2012


On Tue, Mar 6, 2012 at 2:49 PM, Mirco Wahab
<mirco.wahab at chemie.tu-freiberg.de> wrote:
> Am 06.03.2012 14:25, schrieb Szilárd Páll:
>
>> What was the issue with the FindCUDA script?
>
>
> On configuring CUDA, cmake (2.8.7) on Win64/VS2010 bails out w/error:
>
> CMake Error at C:/Apps/CMake/share/cmake-2.8/Modules/FindCUDA.cmake:877
> (message):
>  Invalid format flag passed to CUDA_WRAP_SRCS: 'OBJ'.  Use OBJ or PTX.
> Call Stack (most recent call first):
>  C:/Apps/CMake/share/cmake-2.8/Modules/FindCUDA.cmake:1198 (CUDA_WRAP_SRCS)
>  src/gmxlib/cuda_tools/CMakeLists.txt:5 (CUDA_ADD_LIBRARY)
>
>
> Interestingly, the parameter delivered to the macro in L.877
> is obviously 'OBJ', as can be seen here. Therefore, I made a
> quick temporary hack and explicitely set
>
>   set( compile_to_ptx OFF )
>
> in case of this error and turned the
> error into a "notice". Diff is attached. I was not able
> to deduce the real source of this error.

Thanks a bunch for the detailed information!

> BTW, there seems to have been a tremendous effort in
> enhancing Gromacs over the last months as can be
> learned from the sources. Also the new tMPI imple-
> mentation seems to be certainly faster when tested
> against the 4.5.x internal-MPI on my 10^6 particle
> test system.

Indeed, there has been quite a bit of work done to advance in the
direction of multi-level and heterogeneous parallelization.

Note that what you are referring to as "internal-MPI" is the same as
tMPI/thread-MPI so the performance improvement comes from somewhere
else, probably from the new cut-off scheme. You might want to check
out the wiki pages on the cut-off schemes and parallelization I linked
in my first mail.

Cheers,
--
Szilárd


> Thanks & regards
>
> M.
>
> --
> gmx-developers mailing list
> gmx-developers at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-developers
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-developers-request at gromacs.org.



More information about the gromacs.org_gmx-developers mailing list