coll/libnbc: revamp ibcast and use NBC_Schedule_request()
Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
Этот коммит содержится в:
Gilles Gouaillardet
KAWASHIMA Takahiro
родитель
84701cd2b0
Коммит
360a76f440
@ -26,17 +26,11 @@ int ompi_coll_libnbc_ibarrier(struct ompi_communicator_t *comm, ompi_request_t *
|
||||
{
|
||||
int rank, p, maxround, res, recvpeer, sendpeer;
|
||||
NBC_Schedule *schedule;
|
||||
NBC_Handle *handle;
|
||||
ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
|
||||
|
||||
rank = ompi_comm_rank (comm);
|
||||
p = ompi_comm_size (comm);
|
||||
|
||||
res = NBC_Init_handle(comm, &handle, libnbc_module);
|
||||
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
|
||||
return res;
|
||||
}
|
||||
|
||||
#ifdef NBC_CACHE_SCHEDULE
|
||||
/* there only one argument set per communicator -> hang it directly at
|
||||
* the tree-position, NBC_Dict_size[...] is 0 for not initialized and
|
||||
@ -47,13 +41,9 @@ int ompi_coll_libnbc_ibarrier(struct ompi_communicator_t *comm, ompi_request_t *
|
||||
#endif
|
||||
schedule = OBJ_NEW(NBC_Schedule);
|
||||
if (OPAL_UNLIKELY(NULL == schedule)) {
|
||||
NBC_Return_handle (handle);
|
||||
return OMPI_ERR_OUT_OF_RESOURCE;
|
||||
}
|
||||
|
||||
/* ensure the schedule is released with the handle on error */
|
||||
handle->schedule = schedule;
|
||||
|
||||
maxround = (int)ceil((log((double)p)/LOG2)-1);
|
||||
|
||||
for (int round = 0 ; round <= maxround ; ++round) {
|
||||
@ -64,52 +54,47 @@ int ompi_coll_libnbc_ibarrier(struct ompi_communicator_t *comm, ompi_request_t *
|
||||
/* send msg to sendpeer */
|
||||
res = NBC_Sched_send (NULL, false, 0, MPI_BYTE, sendpeer, schedule, false);
|
||||
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
|
||||
NBC_Return_handle (handle);
|
||||
return OMPI_ERR_OUT_OF_RESOURCE;
|
||||
OBJ_RELEASE(schedule);
|
||||
return res;
|
||||
}
|
||||
|
||||
/* recv msg from recvpeer */
|
||||
res = NBC_Sched_recv (NULL, false, 0, MPI_BYTE, recvpeer, schedule, false);
|
||||
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
|
||||
NBC_Return_handle (handle);
|
||||
return OMPI_ERR_OUT_OF_RESOURCE;
|
||||
OBJ_RELEASE(schedule);
|
||||
return res;
|
||||
}
|
||||
|
||||
/* end communication round */
|
||||
if (round < maxround) {
|
||||
res = NBC_Sched_barrier (schedule);
|
||||
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
|
||||
NBC_Return_handle (handle);
|
||||
return OMPI_ERR_OUT_OF_RESOURCE;
|
||||
OBJ_RELEASE(schedule);
|
||||
return res;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
res = NBC_Sched_commit (schedule);
|
||||
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
|
||||
NBC_Return_handle (handle);
|
||||
return OMPI_ERR_OUT_OF_RESOURCE;
|
||||
OBJ_RELEASE(schedule);
|
||||
return res;
|
||||
}
|
||||
|
||||
#ifdef NBC_CACHE_SCHEDULE
|
||||
/* add it */
|
||||
libnbc_module->NBC_Dict[NBC_BARRIER] = (hb_tree *) schedule;
|
||||
libnbc_module->NBC_Dict_size[NBC_BARRIER] = 1;
|
||||
} else {
|
||||
/* we found it */
|
||||
handle->schedule = schedule = (NBC_Schedule *) libnbc_module->NBC_Dict[NBC_BARRIER];
|
||||
}
|
||||
OBJ_RETAIN(schedule);
|
||||
#endif
|
||||
|
||||
res = NBC_Start (handle, schedule);
|
||||
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
|
||||
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
|
||||
NBC_Return_handle (handle);
|
||||
return OMPI_ERR_OUT_OF_RESOURCE;
|
||||
OBJ_RELEASE(schedule);
|
||||
return res;
|
||||
}
|
||||
|
||||
*request = (ompi_request_t *) handle;
|
||||
|
||||
return OMPI_SUCCESS;
|
||||
}
|
||||
|
||||
@ -118,32 +103,22 @@ int ompi_coll_libnbc_ibarrier_inter(struct ompi_communicator_t *comm, ompi_reque
|
||||
{
|
||||
int rank, res, rsize;
|
||||
NBC_Schedule *schedule;
|
||||
NBC_Handle *handle;
|
||||
ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
|
||||
|
||||
rank = ompi_comm_rank (comm);
|
||||
rsize = ompi_comm_remote_size (comm);
|
||||
|
||||
res = NBC_Init_handle(comm, &handle, libnbc_module);
|
||||
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
|
||||
return res;
|
||||
}
|
||||
|
||||
schedule = OBJ_NEW(NBC_Schedule);
|
||||
if (OPAL_UNLIKELY(NULL == schedule)) {
|
||||
NBC_Return_handle (handle);
|
||||
return OMPI_ERR_OUT_OF_RESOURCE;
|
||||
}
|
||||
|
||||
/* ensure the schedule is released with the handle on error */
|
||||
handle->schedule = schedule;
|
||||
|
||||
if (0 == rank) {
|
||||
for (int peer = 1 ; peer < rsize ; ++peer) {
|
||||
res = NBC_Sched_recv (NULL, false, 0, MPI_BYTE, peer, schedule, false);
|
||||
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
|
||||
NBC_Return_handle (handle);
|
||||
return OMPI_ERR_OUT_OF_RESOURCE;
|
||||
OBJ_RELEASE(schedule);
|
||||
return res;
|
||||
}
|
||||
}
|
||||
}
|
||||
@ -151,47 +126,44 @@ int ompi_coll_libnbc_ibarrier_inter(struct ompi_communicator_t *comm, ompi_reque
|
||||
/* synchronize with the remote root */
|
||||
res = NBC_Sched_recv (NULL, false, 0, MPI_BYTE, 0, schedule, false);
|
||||
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
|
||||
NBC_Return_handle (handle);
|
||||
return OMPI_ERR_OUT_OF_RESOURCE;
|
||||
OBJ_RELEASE(schedule);
|
||||
return res;
|
||||
}
|
||||
|
||||
res = NBC_Sched_send (NULL, false, 0, MPI_BYTE, 0, schedule, false);
|
||||
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
|
||||
NBC_Return_handle (handle);
|
||||
return OMPI_ERR_OUT_OF_RESOURCE;
|
||||
OBJ_RELEASE(schedule);
|
||||
return res;
|
||||
}
|
||||
|
||||
if (0 == rank) {
|
||||
/* wait for the remote root */
|
||||
res = NBC_Sched_barrier (schedule);
|
||||
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
|
||||
NBC_Return_handle (handle);
|
||||
return OMPI_ERR_OUT_OF_RESOURCE;
|
||||
OBJ_RELEASE(schedule);
|
||||
return res;
|
||||
}
|
||||
|
||||
/* inform remote peers that all local peers have entered the barrier */
|
||||
for (int peer = 1; peer < rsize ; ++peer) {
|
||||
res = NBC_Sched_send (NULL, false, 0, MPI_BYTE, peer, schedule, false);
|
||||
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
|
||||
NBC_Return_handle (handle);
|
||||
return OMPI_ERR_OUT_OF_RESOURCE;
|
||||
OBJ_RELEASE(schedule);
|
||||
return res;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
res = NBC_Sched_commit (schedule);
|
||||
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
|
||||
NBC_Return_handle (handle);
|
||||
return OMPI_ERR_OUT_OF_RESOURCE;
|
||||
OBJ_RELEASE(schedule);
|
||||
return res;
|
||||
}
|
||||
|
||||
res = NBC_Start (handle, schedule);
|
||||
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
|
||||
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
|
||||
NBC_Return_handle (handle);
|
||||
return OMPI_ERR_OUT_OF_RESOURCE;
|
||||
OBJ_RELEASE(schedule);
|
||||
return res;
|
||||
}
|
||||
|
||||
*request = (ompi_request_t *) handle;
|
||||
|
||||
return OMPI_SUCCESS;
|
||||
}
|
||||
|
||||
Загрузка…
x
Ссылка в новой задаче
Block a user