1
1

Correct fence logic in MPI_Init

The fence logic in MPI_Init got messed up somehow such that we were
always executing a fence, which is not desirable. The logic is supposed
to be:

* if async fence is requested and we are not collecting data, then do
not fence at all

* if async fence is requested and we are collecting data, then execute
the fence in the background - wait for completion at the end of MPI_Init.

* if async fence is not requested, then execute a blocking fence at that
point, collecting data as directed. Note that we cannot actually do a
blocking fence as we need to cycle the event library via opal_progress
as the PMIx progress thread is tied to the OMPI event base.

Signed-off-by: Ralph Castain <rhc@pmix.org>
Этот коммит содержится в:
Ralph Castain 2020-03-11 07:00:40 -07:00
родитель 11028d0322
Коммит dd623cec34
Не найден ключ, соответствующий данной подписи
Идентификатор ключа GPG: B63B630167D26BB5

Просмотреть файл

@ -679,38 +679,39 @@ int ompi_mpi_init(int argc, char **argv, int requested, int *provided,
#endif
if (!ompi_singleton) {
/* If we have a non-blocking fence:
* if we are doing an async modex, but we are collecting all
* data, then execute the non-blocking modex in the background.
* All calls to modex_recv will be cached until the background
* modex completes. If collect_all_data is false, then we skip
* the fence completely and retrieve data on-demand from the
* source node.
*
* If we do not have a non-blocking fence, then we must always
* execute the blocking fence as the system does not support
* later data retrieval. */
if (opal_pmix_base_async_modex) {
/* execute the fence_nb in the background to collect
* the data */
background_fence = true;
active = true;
OPAL_POST_OBJECT(&active);
PMIX_INFO_LOAD(&info[0], PMIX_COLLECT_DATA, &opal_pmix_collect_all_data, PMIX_BOOL);
if( PMIX_SUCCESS != (rc = PMIx_Fence_nb(NULL, 0, NULL, 0,
fence_release,
(void*)&active))) {
ret = opal_pmix_convert_status(rc);
error = "PMIx_Fence_nb() failed";
goto error;
/* if we are doing an async modex, but we are collecting all
* data, then execute the non-blocking modex in the background.
* All calls to modex_recv will be cached until the background
* modex completes. If collect_all_data is false, then we skip
* the fence completely and retrieve data on-demand from the
* source node.
*/
if (opal_pmix_collect_all_data) {
/* execute the fence_nb in the background to collect
* the data */
background_fence = true;
active = true;
OPAL_POST_OBJECT(&active);
PMIX_INFO_LOAD(&info[0], PMIX_COLLECT_DATA, &opal_pmix_collect_all_data, PMIX_BOOL);
if( PMIX_SUCCESS != (rc = PMIx_Fence_nb(NULL, 0, NULL, 0,
fence_release,
(void*)&active))) {
ret = opal_pmix_convert_status(rc);
error = "PMIx_Fence_nb() failed";
goto error;
}
}
} else if (!opal_pmix_base_async_modex) {
/* we want to do the modex */
} else {
/* we want to do the modex - we block at this point, but we must
* do so in a manner that allows us to call opal_progress so our
* event library can be cycled as we have tied PMIx to that
* event base */
active = true;
OPAL_POST_OBJECT(&active);
PMIX_INFO_LOAD(&info[0], PMIX_COLLECT_DATA, &opal_pmix_collect_all_data, PMIX_BOOL);
if( PMIX_SUCCESS != (rc = PMIx_Fence_nb(NULL, 0, info, 1, fence_release, (void*)&active))) {
rc = PMIx_Fence_nb(NULL, 0, info, 1, fence_release, (void*)&active);
if( PMIX_SUCCESS != rc) {
ret = opal_pmix_convert_status(rc);
error = "PMIx_Fence() failed";
goto error;