int ompi_coll_tuned_gather_intra_dec_dynamic(const void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void* rbuf, int rcount,
struct ompi_datatype_t *rdtype,
int root,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
OPAL_OUTPUT((ompi_coll_tuned_stream,
"ompi_coll_tuned_gather_intra_dec_dynamic"));
/**
* check to see if we have some filebased rules.
*/
if (tuned_module->com_rules[GATHER]) {
int comsize, alg, faninout, segsize, max_requests;
size_t dsize;
comsize = ompi_comm_size(comm);
ompi_datatype_type_size (sdtype, &dsize);
dsize *= comsize;
alg = ompi_coll_tuned_get_target_method_params (tuned_module->com_rules[GATHER],
dsize, &faninout, &segsize, &max_requests);
if (alg) {
/* we have found a valid choice from the file based rules for this message size */
return ompi_coll_tuned_gather_intra_do_this (sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module,
alg, faninout, segsize);
} /* found a method */
} /*end if any com rules to check */
if (tuned_module->user_forced[GATHER].algorithm) {
return ompi_coll_tuned_gather_intra_do_this(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module,
tuned_module->user_forced[GATHER].algorithm,
tuned_module->user_forced[GATHER].tree_fanout,
tuned_module->user_forced[GATHER].segsize);
}
return ompi_coll_tuned_gather_intra_dec_fixed (sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module);
}
int
ompi_coll_tuned_gather_intra_do_forced(void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void* rbuf, int rcount,
struct ompi_datatype_t *rdtype,
int root,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:gather_intra_do_forced selected algorithm %d",
tuned_module->user_forced[GATHER].algorithm));
switch (tuned_module->user_forced[GATHER].algorithm) {
case (0):
return ompi_coll_tuned_gather_intra_dec_fixed(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module);
case (1):
return ompi_coll_base_gather_intra_basic_linear(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module);
case (2):
return ompi_coll_base_gather_intra_binomial(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module);
case (3):
return ompi_coll_base_gather_intra_linear_sync(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module,
tuned_module->user_forced[GATHER].segsize);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:gather_intra_do_forced attempt to select algorithm %d when only 0-%d is valid?",
tuned_module->user_forced[GATHER].algorithm,
ompi_coll_tuned_forced_max_algorithms[GATHER]));
return (MPI_ERR_ARG);
}
