int ompi_coll_tuned_alltoall_intra_do_this(void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void* rbuf, int rcount,
struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module,
int algorithm, int faninout, int segsize,
int max_requests)
{
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:alltoall_intra_do_this selected algorithm %d topo faninout %d segsize %d",
algorithm, faninout, segsize));
switch (algorithm) {
case (0): return ompi_coll_tuned_alltoall_intra_dec_fixed (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
case (1): return ompi_coll_tuned_alltoall_intra_basic_linear (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
case (2): return ompi_coll_tuned_alltoall_intra_pairwise (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
case (3): return ompi_coll_tuned_alltoall_intra_bruck (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
case (4): return ompi_coll_tuned_alltoall_intra_linear_sync (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module, max_requests);
case (5): return ompi_coll_tuned_alltoall_intra_two_procs (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
default:
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:alltoall_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",
algorithm, ompi_coll_tuned_forced_max_algorithms[ALLTOALL]));
return (MPI_ERR_ARG);
} /* switch */
}
int ompi_coll_tuned_alltoall_intra_do_forced(void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void* rbuf, int rcount,
struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
mca_coll_tuned_comm_t *data = tuned_module->tuned_data;
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:alltoall_intra_do_forced selected algorithm %d",
data->user_forced[ALLTOALL].algorithm));
switch (data->user_forced[ALLTOALL].algorithm) {
case (0): return ompi_coll_tuned_alltoall_intra_dec_fixed (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
case (1): return ompi_coll_tuned_alltoall_intra_basic_linear (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
case (2): return ompi_coll_tuned_alltoall_intra_pairwise (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
case (3): return ompi_coll_tuned_alltoall_intra_bruck (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
case (4): return ompi_coll_tuned_alltoall_intra_linear_sync (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module, data->user_forced[ALLTOALL].max_requests);
case (5): return ompi_coll_tuned_alltoall_intra_two_procs (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
default:
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:alltoall_intra_do_forced attempt to select algorithm %d when only 0-%d is valid?",
data->user_forced[ALLTOALL].algorithm, ompi_coll_tuned_forced_max_algorithms[ALLTOALL]));
return (MPI_ERR_ARG);
} /* switch */
}
int ompi_coll_tuned_alltoall_intra_do_forced(void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void* rbuf, int rcount,
struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm)
{
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:alltoall_intra_do_forced selected algorithm %d",
comm->c_coll_selected_data->user_forced[ALLTOALL].algorithm));
switch (comm->c_coll_selected_data->user_forced[ALLTOALL].algorithm) {
case (0): return ompi_coll_tuned_alltoall_intra_dec_fixed (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm);
case (1): return ompi_coll_tuned_alltoall_intra_basic_linear (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm);
case (2): return ompi_coll_tuned_alltoall_intra_pairwise (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm);
case (3): return ompi_coll_tuned_alltoall_intra_bruck (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm);
case (4): return ompi_coll_tuned_alltoall_intra_two_procs (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm);
default:
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:alltoall_intra_do_forced attempt to select algorithm %d when only 0-%d is valid?",
comm->c_coll_selected_data->user_forced[ALLTOALL].algorithm, ompi_coll_tuned_forced_max_algorithms[ALLTOALL]));
return (MPI_ERR_ARG);
} /* switch */
}
int ompi_coll_tuned_alltoall_intra_dec_fixed(void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void* rbuf, int rcount,
struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
int communicator_size;
size_t dsize, block_dsize;
#if 0
size_t total_dsize;
#endif
communicator_size = ompi_comm_size(comm);
/* special case */
if (communicator_size==2) {
return ompi_coll_tuned_alltoall_intra_two_procs(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
}
/* 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 */
ompi_ddt_type_size(sdtype, &dsize);
block_dsize = dsize * scount;
if ((block_dsize < 200) && (communicator_size > 12)) {
return ompi_coll_tuned_alltoall_intra_bruck(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
} else if (block_dsize < 3000) {
return ompi_coll_tuned_alltoall_intra_basic_linear(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
}
return ompi_coll_tuned_alltoall_intra_pairwise (sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
#if 0
/* previous decision */
/* else we need data size for decision function */
ompi_ddt_type_size(sdtype, &dsize);
total_dsize = dsize * scount * communicator_size; /* needed for decision */
OPAL_OUTPUT((ompi_coll_tuned_stream, "ompi_coll_tuned_alltoall_intra_dec_fixed rank %d com_size %d msg_length %ld",
ompi_comm_rank(comm), communicator_size, total_dsize));
if (communicator_size >= 12 && total_dsize <= 768) {
return ompi_coll_tuned_alltoall_intra_bruck (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
}
if (total_dsize <= 131072) {
return ompi_coll_tuned_alltoall_intra_basic_linear (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
}
return ompi_coll_tuned_alltoall_intra_pairwise (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
#endif
}
