int mca_coll_monitoring_ineighbor_alltoall(const void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void *rbuf, int rcount,
struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm,
ompi_request_t ** request,
mca_coll_base_module_t *module)
{
mca_coll_monitoring_module_t*monitoring_module = (mca_coll_monitoring_module_t*) module;
size_t type_size, data_size, data_size_aggreg = 0;
const mca_topo_base_comm_cart_t *cart = comm->c_topo->mtc.cart;
int dim, srank, drank, world_rank;
ompi_datatype_type_size(sdtype, &type_size);
data_size = scount * type_size;
for( dim = 0; dim < cart->ndims; ++dim ) {
srank = MPI_PROC_NULL, drank = MPI_PROC_NULL;
if (cart->dims[dim] > 1) {
mca_topo_base_cart_shift (comm, dim, 1, &srank, &drank);
} else if (1 == cart->dims[dim] && cart->periods[dim]) {
/* Don't record exchanges with self */
continue;
}
if (MPI_PROC_NULL != srank) {
/**
* If this fails the destination is not part of my MPI_COM_WORLD
* Lookup its name in the rank hastable to get its MPI_COMM_WORLD rank
*/
if( OPAL_SUCCESS == mca_common_monitoring_get_world_rank(srank, comm, &world_rank) ) {
mca_common_monitoring_record_coll(world_rank, data_size);
data_size_aggreg += data_size;
}
}
if (MPI_PROC_NULL != drank) {
/**
* If this fails the destination is not part of my MPI_COM_WORLD
* Lookup its name in the rank hastable to get its MPI_COMM_WORLD rank
*/
if( OPAL_SUCCESS == mca_common_monitoring_get_world_rank(drank, comm, &world_rank) ) {
mca_common_monitoring_record_coll(world_rank, data_size);
data_size_aggreg += data_size;
}
}
}
mca_common_monitoring_coll_a2a(data_size_aggreg, monitoring_module->data);
return monitoring_module->real.coll_ineighbor_alltoall(sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, request, monitoring_module->real.coll_ineighbor_alltoall_module);
}
int mca_coll_monitoring_ialltoallv(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,
ompi_request_t ** request,
mca_coll_base_module_t *module)
{
mca_coll_monitoring_module_t*monitoring_module = (mca_coll_monitoring_module_t*) module;
size_t type_size, data_size, data_size_aggreg = 0;
const int comm_size = ompi_comm_size(comm);
const int my_rank = ompi_comm_rank(comm);
int i, rank;
ompi_datatype_type_size(sdtype, &type_size);
for( i = 0; i < comm_size; ++i ) {
if( my_rank == i ) continue; /* No communication for self */
data_size = scounts[i] * type_size;
/**
* If this fails the destination is not part of my MPI_COM_WORLD
* Lookup its name in the rank hastable to get its MPI_COMM_WORLD rank
*/
if( OPAL_SUCCESS == mca_common_monitoring_get_world_rank(i, comm->c_remote_group, &rank) ) {
mca_common_monitoring_record_coll(rank, data_size);
data_size_aggreg += data_size;
}
}
mca_common_monitoring_coll_a2a(data_size_aggreg, monitoring_module->data);
return monitoring_module->real.coll_ialltoallv(sbuf, scounts, sdisps, sdtype, rbuf, rcounts, rdisps, rdtype, comm, request, monitoring_module->real.coll_ialltoallv_module);
}
int mca_coll_monitoring_ireduce(const void *sbuf, void *rbuf, int count,
struct ompi_datatype_t *dtype,
struct ompi_op_t *op,
int root,
struct ompi_communicator_t *comm,
ompi_request_t ** request,
mca_coll_base_module_t *module)
{
mca_coll_monitoring_module_t*monitoring_module = (mca_coll_monitoring_module_t*) module;
if( root == ompi_comm_rank(comm) ) {
int i, rank;
size_t type_size, data_size;
const int comm_size = ompi_comm_size(comm);
ompi_datatype_type_size(dtype, &type_size);
data_size = count * type_size;
for( i = 0; i < comm_size; ++i ) {
if( root == i ) continue; /* No communication for self */
/**
* If this fails the destination is not part of my MPI_COM_WORLD
* Lookup its name in the rank hastable to get its MPI_COMM_WORLD rank
*/
if( OPAL_SUCCESS == mca_common_monitoring_get_world_rank(i, comm->c_remote_group, &rank) ) {
mca_common_monitoring_record_coll(rank, data_size);
}
}
mca_common_monitoring_coll_a2o(data_size * (comm_size - 1), monitoring_module->data);
}
return monitoring_module->real.coll_ireduce(sbuf, rbuf, count, dtype, op, root, comm, request, monitoring_module->real.coll_ireduce_module);
}
int mca_coll_monitoring_ibcast(void *buff, int count,
struct ompi_datatype_t *datatype,
int root,
struct ompi_communicator_t *comm,
ompi_request_t ** request,
mca_coll_base_module_t *module)
{
mca_coll_monitoring_module_t*monitoring_module = (mca_coll_monitoring_module_t*) module;
size_t type_size, data_size;
const int comm_size = ompi_comm_size(comm);
ompi_datatype_type_size(datatype, &type_size);
data_size = count * type_size;
if( root == ompi_comm_rank(comm) ) {
int i, rank;
mca_common_monitoring_coll_o2a(data_size * (comm_size - 1), monitoring_module->data);
for( i = 0; i < comm_size; ++i ) {
if( i == root ) continue; /* No self sending */
/**
* If this fails the destination is not part of my MPI_COM_WORLD
* Lookup its name in the rank hastable to get its MPI_COMM_WORLD rank
*/
if( OPAL_SUCCESS == mca_common_monitoring_get_world_rank(i, comm, &rank) ) {
mca_common_monitoring_record_coll(rank, data_size);
}
}
}
return monitoring_module->real.coll_ibcast(buff, count, datatype, root, comm, request, monitoring_module->real.coll_ibcast_module);
}
static inline void mca_common_monitoring_coll_cache(mca_monitoring_coll_data_t*data)
{
int world_rank;
if( NULL == data->comm_name && 0 < strlen(data->p_comm->c_name) ) {
data->comm_name = strdup(data->p_comm->c_name);
} else {
mca_common_monitoring_coll_check_name(data);
}
if( -1 == data->world_rank ) {
/* Get current process world_rank */
mca_common_monitoring_get_world_rank(ompi_comm_rank(data->p_comm), data->p_comm,
&data->world_rank);
}
/* Only list procs if the hashtable is already initialized, ie if the previous call worked */
if( (-1 != data->world_rank) && (NULL == data->procs || 0 == strlen(data->procs)) ) {
int i, pos = 0, size, world_size = -1, max_length;
char*tmp_procs;
size = ompi_comm_size(data->p_comm);
world_size = ompi_comm_size((ompi_communicator_t*)&ompi_mpi_comm_world) - 1;
assert( 0 < size );
/* Allocate enough space for list (add 1 to keep the final '\0' if already exact size) */
max_length = snprintf(NULL, 0, "%d,", world_size - 1) + 1;
tmp_procs = malloc((1 + max_length * size) * sizeof(char));
if( NULL == tmp_procs ) {
OPAL_MONITORING_PRINT_ERR("Cannot allocate memory for caching proc list.");
} else {
tmp_procs[0] = '\0';
/* Build procs list */
for(i = 0; i < size; ++i) {
mca_common_monitoring_get_world_rank(i, data->p_comm, &world_rank);
pos += sprintf(&tmp_procs[pos], "%d,", world_rank);
}
tmp_procs[pos - 1] = '\0'; /* Remove final coma */
data->procs = realloc(tmp_procs, pos * sizeof(char)); /* Adjust to size required */
}
}
}
int mca_pml_monitoring_send(const void *buf,
size_t count,
ompi_datatype_t *datatype,
int dst,
int tag,
mca_pml_base_send_mode_t mode,
struct ompi_communicator_t* comm)
{
int world_rank;
/* Are we sending to a peer from my own MPI_COMM_WORLD? */
if(OPAL_SUCCESS == mca_common_monitoring_get_world_rank(dst, comm->c_remote_group, &world_rank)) {
size_t type_size, data_size;
ompi_datatype_type_size(datatype, &type_size);
data_size = count*type_size;
mca_common_monitoring_record_pml(world_rank, data_size, tag);
}
return pml_selected_module.pml_send(buf, count, datatype,
dst, tag, mode, comm);
}
int mca_pml_monitoring_isend(const void *buf,
size_t count,
ompi_datatype_t *datatype,
int dst,
int tag,
mca_pml_base_send_mode_t mode,
struct ompi_communicator_t* comm,
struct ompi_request_t **request)
{
int world_rank;
/**
* If this fails the destination is not part of my MPI_COM_WORLD
* Lookup its name in the rank hastable to get its MPI_COMM_WORLD rank
*/
if(OPAL_SUCCESS == mca_common_monitoring_get_world_rank(dst, comm->c_remote_group, &world_rank)) {
size_t type_size, data_size;
ompi_datatype_type_size(datatype, &type_size);
data_size = count*type_size;
mca_common_monitoring_record_pml(world_rank, data_size, tag);
}
return pml_selected_module.pml_isend(buf, count, datatype,
dst, tag, mode, comm, request);
}
