/* If the user selects dynamic rules and specifies the algorithm to
* use, then this function is called. */
int ompi_coll_tuned_alltoallv_intra_do_this(void *sbuf, int *scounts, int *sdisps,
struct ompi_datatype_t *sdtype,
void* rbuf, int *rcounts, int *rdisps,
struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module,
int algorithm)
{
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:alltoallv_intra_do_this selected algorithm %d ",
algorithm));
switch (algorithm) {
case (0):
return ompi_coll_tuned_alltoallv_intra_dec_fixed(sbuf, scounts, sdisps, sdtype,
rbuf, rcounts, rdisps, rdtype,
comm, module);
case (1):
return ompi_coll_tuned_alltoallv_intra_basic_linear(sbuf, scounts, sdisps, sdtype,
rbuf, rcounts, rdisps, rdtype,
comm, module);
case (2):
return ompi_coll_tuned_alltoallv_intra_pairwise(sbuf, scounts, sdisps, sdtype,
rbuf, rcounts, rdisps, 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[ALLTOALLV]));
return (MPI_ERR_ARG);
}
}
int ompi_coll_tuned_alltoallv_intra_do_forced(void *sbuf, int *scounts, int *sdisps,
struct ompi_datatype_t *sdtype,
void* rbuf, int *rcounts, int *rdisps,
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;
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:alltoallv_intra_do_forced selected algorithm %d",
tuned_module->user_forced[ALLTOALLV].algorithm));
switch (tuned_module->user_forced[ALLTOALLV].algorithm) {
case (0):
return ompi_coll_tuned_alltoallv_intra_dec_fixed(sbuf, scounts, sdisps, sdtype,
rbuf, rcounts, rdisps, rdtype,
comm, module);
case (1):
return ompi_coll_base_alltoallv_intra_basic_linear(sbuf, scounts, sdisps, sdtype,
rbuf, rcounts, rdisps, rdtype,
comm, module);
case (2):
return ompi_coll_base_alltoallv_intra_pairwise(sbuf, scounts, sdisps, sdtype,
rbuf, rcounts, rdisps, rdtype,
comm, module);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:alltoallv_intra_do_forced attempt to "
"select algorithm %d when only 0-%d is valid.",
tuned_module->user_forced[ALLTOALLV].algorithm,
ompi_coll_tuned_forced_max_algorithms[ALLTOALLV]));
return (MPI_ERR_ARG);
}
int ompi_coll_tuned_alltoallv_intra_dec_dynamic(const void *sbuf, const int *scounts, const int *sdisps,
struct ompi_datatype_t *sdtype,
void* rbuf, const int *rcounts, const int *rdisps,
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;
OPAL_OUTPUT((ompi_coll_tuned_stream, "ompi_coll_tuned_alltoallv_intra_dec_dynamic"));
/**
* check to see if we have some filebased rules. As we don't have global
* knowledge about the total amount of data, use the first available rule.
* This allow the users to specify the alltoallv algorithm to be used only
* based on the communicator size.
*/
if (tuned_module->com_rules[ALLTOALLV]) {
int alg, faninout, segsize, max_requests;
alg = ompi_coll_tuned_get_target_method_params (tuned_module->com_rules[ALLTOALLV],
0, &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_alltoallv_intra_do_this (sbuf, scounts, sdisps, sdtype,
rbuf, rcounts, rdisps, rdtype,
comm, module,
alg);
} /* found a method */
} /*end if any com rules to check */
if (tuned_module->user_forced[ALLTOALLV].algorithm) {
return ompi_coll_tuned_alltoallv_intra_do_this(sbuf, scounts, sdisps, sdtype,
rbuf, rcounts, rdisps, rdtype,
comm, module,
tuned_module->user_forced[ALLTOALLV].algorithm);
}
return ompi_coll_tuned_alltoallv_intra_dec_fixed(sbuf, scounts, sdisps, sdtype,
rbuf, rcounts, rdisps, rdtype,
comm, module);
}
