int ompi_coll_tuned_reduce_intra_chain( void *sendbuf, void *recvbuf, int count,
ompi_datatype_t* datatype,
ompi_op_t* op, int root,
ompi_communicator_t* comm,
mca_coll_base_module_t *module,
uint32_t segsize, int fanout,
int max_outstanding_reqs )
{
int segcount = count;
size_t typelng;
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_chain rank %d fo %d ss %5d", ompi_comm_rank(comm), fanout, segsize));
COLL_TUNED_UPDATE_CHAIN( comm, tuned_module, root, fanout );
/**
* Determine number of segments and number of elements
* sent per operation
*/
ompi_datatype_type_size( datatype, &typelng );
COLL_TUNED_COMPUTED_SEGCOUNT( segsize, typelng, segcount );
return ompi_coll_tuned_reduce_generic( sendbuf, recvbuf, count, datatype,
op, root, comm, module,
data->cached_chain,
segcount, max_outstanding_reqs );
}
int
ompi_coll_tuned_bcast_intra_chain( void* buffer,
int count,
struct ompi_datatype_t* datatype,
int root,
struct ompi_communicator_t* comm,
mca_coll_base_module_t *module,
uint32_t segsize, int32_t chains )
{
int segcount = count;
size_t typelng;
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
mca_coll_tuned_comm_t *data = tuned_module->tuned_data;
COLL_TUNED_UPDATE_CHAIN( comm, tuned_module, root, chains );
/**
* Determine number of elements sent per operation.
*/
ompi_datatype_type_size( datatype, &typelng );
COLL_TUNED_COMPUTED_SEGCOUNT( segsize, typelng, segcount );
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:bcast_intra_chain rank %d fo %d ss %5d typelng %lu segcount %d",
ompi_comm_rank(comm), chains, segsize, (unsigned long)typelng, segcount));
return ompi_coll_tuned_bcast_intra_generic( buffer, count, datatype, root, comm, module,
segcount, data->cached_chain );
}
int ompi_coll_tuned_reduce_intra_chain( void *sendbuf, void *recvbuf, int count,
ompi_datatype_t* datatype, ompi_op_t* op,
int root, ompi_communicator_t* comm, uint32_t segsize,
int fanout)
{
int segcount = count;
size_t typelng;
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_chain rank %d fo %d ss %5d", ompi_comm_rank(comm), fanout, segsize));
COLL_TUNED_UPDATE_CHAIN( comm, root, fanout );
/**
* Determine number of segments and number of elements
* sent per operation
*/
ompi_ddt_type_size( datatype, &typelng );
COLL_TUNED_COMPUTED_SEGCOUNT( segsize, typelng, segcount );
return ompi_coll_tuned_reduce_generic( sendbuf, recvbuf, count, datatype, op, root, comm,
comm->c_coll_selected_data->cached_chain, segcount );
}