Hi,
I am trying to hook up PyCUDA's gpuarray with mpi4py for GPU to GPU message
passing built on MVAPICH2. I received advice from Lisandro Dalcin on how I
could do this in mpi4py's cython code by creating a send buffer from the
gpuarray ptr and nbytes member functions. Ex
def make_buffer(long address, ssize_t size, bint readonly=False):
return tobuffer(<void*>address,size,readonly)
Link to original post on mpi4py:
https://groups.google.com/forum/#!topic/mpi4py/qpM-ZcAtA_Y
However, when I use a send buffer created this way I get a segfault within
MPI_Send. What type should the gpuarray.ptr be in? I've tried long and
unsigned long and both segfault.
Thanks for your time,
Brody Huval
_______________________________________________
PyCUDA mailing list
[email protected]
http://lists.tiker.net/listinfo/pycuda
