[PyCUDA] compiler.get_nvcc_version() problems with mpi4py/mpich2 1.3

[Andreas Kloeckner]

On Fri, 28 Jan 2011 12:39:44 +0200, avidday <avidday at gmail.com> wrote:
> I have hit a wall moving some existing pycuda code to a distributed
> memory cluster and am hoping someone cleverer than I can suggest a
> work around.

Try calling pytools.prefork.enable_prefork before you call MPI_Init().
(PyCUDA is already prepared for this under the hood, and you already
have pytools if you have PyCUDA.)

http://git.tiker.net/pytools.git/blob/HEAD:/pytools/prefork.py

If you use mpi4py, you simply import pytools.mpiwrap in its stead, this
will take care of doing the prefork for you.  (mpi4py call MPI_Init
implicitly on import, which makes this a bit tricky.)

HTH,
Andreas
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: <http://lists.tiker.net/pipermail/pycuda/attachments/20110130/88fb0b04/attachment.pgp>