parallel processing - Scattered indices in MPI -
i'm trying divide array between processors such each 1 takes points different parts in array. example, if
a = {1, 2, 3, 4, 5, 6, 7, 8}
and i'm using 2 processors, want p1 handle {1, 3, 5, 7}
, , p2 handle {2, 4, 6, 8}
.
when scaling large numbers of points (millions) , processors (128), tricky. in previous versions of function, gave p1 first chunk of points, p2 next chunk, , on (which easy mpi_gatherv
).
is there way use mpi_gatherv
make work, or way use mpi_send
, mpi_receive
achieve it? trouble mpi_gatherv
while can specify indices processors send to, still puts of p1 before p2 before p3 etc.
Comments
Post a Comment