int smpi_coll_tuned_alltoall_ompi( void *sbuf, int scount,
MPI_Datatype sdtype,
void* rbuf, int rcount,
MPI_Datatype rdtype,
MPI_Comm comm)
{
int communicator_size;
size_t dsize, block_dsize;
communicator_size = smpi_comm_size(comm);
/* Decision function based on measurement on Grig cluster at
the University of Tennessee (2GB MX) up to 64 nodes.
Has better performance for messages of intermediate sizes than the old one */
/* determine block size */
dsize = smpi_datatype_size(sdtype);
block_dsize = dsize * scount;
if ((block_dsize < 200) && (communicator_size > 12)) {
return smpi_coll_tuned_alltoall_bruck(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm);
} else if (block_dsize < 3000) {
return smpi_coll_tuned_alltoall_basic_linear(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm);
}
return smpi_coll_tuned_alltoall_ring (sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm);
}
int smpi_coll_tuned_alltoall_ompi2(void *sendbuf, int sendcount,
MPI_Datatype sendtype, void *recvbuf,
int recvcount, MPI_Datatype recvtype,
MPI_Comm comm)
{
int size, sendsize;
size = smpi_comm_size(comm);
sendsize = smpi_datatype_size(sendtype) * sendcount;
if (sendsize < 200 && size > 12) {
return
smpi_coll_tuned_alltoall_bruck(sendbuf, sendcount, sendtype,
recvbuf, recvcount, recvtype,
comm);
} else if (sendsize < 3000) {
return
smpi_coll_tuned_alltoall_basic_linear(sendbuf, sendcount,
sendtype, recvbuf,
recvcount, recvtype, comm);
} else {
return
smpi_coll_tuned_alltoall_ring(sendbuf, sendcount, sendtype,
recvbuf, recvcount, recvtype,
comm);
}
}
