/*
* 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_alltoall_intra_dec_dynamic(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, "ompi_coll_tuned_alltoall_intra_dec_dynamic"));
/* check to see if we have some filebased rules */
if (data->com_rules[ALLTOALL]) {
/* we do, so calc the message size or what ever we need and use this for the evaluation */
int comsize;
int alg, faninout, segsize, max_requests;
size_t dsize;
ompi_datatype_type_size (sdtype, &dsize);
comsize = ompi_comm_size(comm);
dsize *= (ptrdiff_t)comsize * (ptrdiff_t)scount;
alg = ompi_coll_tuned_get_target_method_params (data->com_rules[ALLTOALL],
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_alltoall_intra_do_this (sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module,
alg, faninout, segsize, max_requests);
} /* found a method */
} /*end if any com rules to check */
if (data->user_forced[ALLTOALL].algorithm) {
return ompi_coll_tuned_alltoall_intra_do_forced (sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
}
return ompi_coll_tuned_alltoall_intra_dec_fixed (sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
}
/*
* reduce_scatter_intra_dec
*
* Function: - seletects reduce_scatter algorithm to use
* Accepts: - same arguments as MPI_Reduce_scatter()
* Returns: - MPI_SUCCESS or error code (passed from
* the reduce_scatter implementation)
*
*/
int ompi_coll_tuned_reduce_scatter_intra_dec_dynamic(const void *sbuf, void *rbuf,
const int *rcounts,
struct ompi_datatype_t *dtype,
struct ompi_op_t *op,
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:reduce_scatter_intra_dec_dynamic"));
/* check to see if we have some filebased rules */
if (tuned_module->com_rules[REDUCESCATTER]) {
/* we do, so calc the message size or what ever we need and use
this for the evaluation */
int alg, faninout, segsize, ignoreme, i, count, size;
size_t dsize;
size = ompi_comm_size(comm);
for (i = 0, count = 0; i < size; i++) { count += rcounts[i];}
ompi_datatype_type_size (dtype, &dsize);
dsize *= count;
alg = ompi_coll_tuned_get_target_method_params (tuned_module->com_rules[REDUCESCATTER],
dsize, &faninout,
&segsize, &ignoreme);
if (alg) {
/* we have found a valid choice from the file based rules for this message size */
return ompi_coll_tuned_reduce_scatter_intra_do_this (sbuf, rbuf, rcounts, dtype,
op, comm, module,
alg, faninout, segsize);
} /* found a method */
} /*end if any com rules to check */
if (tuned_module->user_forced[REDUCESCATTER].algorithm) {
return ompi_coll_tuned_reduce_scatter_intra_do_this(sbuf, rbuf, rcounts, dtype,
op, comm, module,
tuned_module->user_forced[REDUCESCATTER].algorithm,
tuned_module->user_forced[REDUCESCATTER].chain_fanout,
tuned_module->user_forced[REDUCESCATTER].segsize);
}
return ompi_coll_tuned_reduce_scatter_intra_dec_fixed (sbuf, rbuf, rcounts,
dtype, op, comm, module);
}
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);
}
/*
* bcast_intra_dec
*
* Function: - seletects broadcast algorithm to use
* Accepts: - same arguments as MPI_Bcast()
* Returns: - MPI_SUCCESS or error code (passed from the bcast implementation)
*/
int ompi_coll_tuned_bcast_intra_dec_dynamic(void *buf, int count,
struct ompi_datatype_t *dtype, 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:bcast_intra_dec_dynamic"));
/* check to see if we have some filebased rules */
if (tuned_module->com_rules[BCAST]) {
/* we do, so calc the message size or what ever we need and use this for the evaluation */
int alg, faninout, segsize, ignoreme;
size_t dsize;
ompi_datatype_type_size (dtype, &dsize);
dsize *= count;
alg = ompi_coll_tuned_get_target_method_params (tuned_module->com_rules[BCAST],
dsize, &faninout, &segsize, &ignoreme);
if (alg) {
/* we have found a valid choice from the file based rules for this message size */
return ompi_coll_tuned_bcast_intra_do_this (buf, count, dtype, root,
comm, module,
alg, faninout, segsize);
} /* found a method */
} /*end if any com rules to check */
if (tuned_module->user_forced[BCAST].algorithm) {
return ompi_coll_tuned_bcast_intra_do_this(buf, count, dtype,
root, comm, module,
tuned_module->user_forced[BCAST].algorithm,
tuned_module->user_forced[BCAST].chain_fanout,
tuned_module->user_forced[BCAST].segsize);
}
return ompi_coll_tuned_bcast_intra_dec_fixed (buf, count, dtype, root,
comm, module);
}
