int ompi_coll_tuned_reduce_intra_do_this(void *sbuf, void* rbuf, int count,
struct ompi_datatype_t *dtype,
struct ompi_op_t *op, int root,
struct ompi_communicator_t *comm,
int algorithm, int faninout, int segsize)
{
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_do_this selected algorithm %d topo faninout %d segsize %d",
algorithm, faninout, segsize));
switch (algorithm) {
case (0): return ompi_coll_tuned_reduce_intra_dec_fixed (sbuf, rbuf, count, dtype, op, root, comm);
case (1): return ompi_coll_tuned_reduce_intra_basic_linear (sbuf, rbuf, count, dtype, op, root, comm);
case (2): return ompi_coll_tuned_reduce_intra_chain (sbuf, rbuf, count, dtype, op, root, comm,
segsize, faninout);
case (3): return ompi_coll_tuned_reduce_intra_pipeline (sbuf, rbuf, count, dtype, op, root, comm,
segsize);
case (4): return ompi_coll_tuned_reduce_intra_binary (sbuf, rbuf, count, dtype, op, root, comm,
segsize);
case (5): return ompi_coll_tuned_reduce_intra_binomial (sbuf, rbuf, count, dtype, op, root, comm,
segsize);
default:
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",
algorithm, ompi_coll_tuned_forced_max_algorithms[REDUCE]));
return (MPI_ERR_ARG);
} /* switch */
}
/*
* reduce_intra_dec
*
* Function: - seletects reduce algorithm to use
* Accepts: - same arguments as MPI_reduce()
* Returns: - MPI_SUCCESS or error code (passed from the reduce implementation)
*
*/
int ompi_coll_tuned_reduce_intra_dec_dynamic( void *sendbuf, void *recvbuf,
int count, struct ompi_datatype_t* datatype,
struct ompi_op_t* op, 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;
mca_coll_tuned_comm_t *data = tuned_module->tuned_data;
OPAL_OUTPUT((ompi_coll_tuned_stream, "coll:tuned:reduce_intra_dec_dynamic"));
/* check to see if we have some filebased rules */
if (data->com_rules[REDUCE]) {
/* we do, so calc the message size or what ever we need and use this for the evaluation */
int alg, faninout, segsize, max_requests;
size_t dsize;
ompi_datatype_type_size (datatype, &dsize);
dsize *= count;
alg = ompi_coll_tuned_get_target_method_params (data->com_rules[REDUCE],
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_reduce_intra_do_this (sendbuf, recvbuf, count, datatype,
op, root,
comm, module,
alg, faninout,
segsize,
max_requests);
} /* found a method */
} /*end if any com rules to check */
if (data->user_forced[REDUCE].algorithm) {
return ompi_coll_tuned_reduce_intra_do_forced (sendbuf, recvbuf, count, datatype,
op, root,
comm, module);
}
return ompi_coll_tuned_reduce_intra_dec_fixed (sendbuf, recvbuf, count, datatype,
op, root,
comm, module);
}
int ompi_coll_tuned_reduce_intra_do_forced(void *sbuf, void* rbuf, int count,
struct ompi_datatype_t *dtype,
struct ompi_op_t *op, 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;
mca_coll_tuned_comm_t *data = tuned_module->tuned_data;
const int segsize = data->user_forced[REDUCE].segsize;
const int chain_fanout = data->user_forced[REDUCE].chain_fanout;
const int max_requests = data->user_forced[REDUCE].max_requests;
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_do_forced selected algorithm %d",
data->user_forced[REDUCE].algorithm));
switch (data->user_forced[REDUCE].algorithm) {
case (0): return ompi_coll_tuned_reduce_intra_dec_fixed (sbuf, rbuf, count, dtype,
op, root, comm, module);
case (1): return ompi_coll_tuned_reduce_intra_basic_linear (sbuf, rbuf, count, dtype,
op, root, comm, module);
case (2): return ompi_coll_tuned_reduce_intra_chain (sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, chain_fanout, max_requests);
case (3): return ompi_coll_tuned_reduce_intra_pipeline (sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
case (4): return ompi_coll_tuned_reduce_intra_binary (sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
case (5): return ompi_coll_tuned_reduce_intra_binomial (sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
case (6): return ompi_coll_tuned_reduce_intra_in_order_binary(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
default:
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_do_forced attempt to select algorithm %d when only 0-%d is valid?",
data->user_forced[REDUCE].algorithm, ompi_coll_tuned_forced_max_algorithms[REDUCE]));
return (MPI_ERR_ARG);
} /* switch */
}
