ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

Merge pull request #4618 from ggouaillardet/topic/pcoll

Просмотр исходного кода 
Add the persistent collectives feature
Этот коммит содержится в:
KAWASHIMA Takahiro 2018-06-26 12:36:34 +09:00 коммит произвёл GitHub
родитель 6c089518e7 3dce039886
Коммит a8da78eeaa
Не найден ключ, соответствующий данной подписи
Идентификатор ключа GPG: 4AEE18F83AFDEB23
73 изменённых файлов: 2976 добавлений и 610 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

2
.gitignore поставляемый
Просмотреть файл

@ -250,6 +250,8 @@ ompi/mpiext/example/tests/progress_usempif08
ompi/mpiext/cuda/c/MPIX_Query_cuda_support.3
ompi/mpiext/cuda/c/mpiext_cuda_c.h
ompi/mpiext/pcollreq/c/MPIX_*.3
ompi/tools/mpisync/mpisync
ompi/tools/mpisync/mpirun_prof
ompi/tools/mpisync/ompi_timing_post

8
README
Просмотреть файл

@ -717,6 +717,14 @@ Open MPI Extensions
- The following extensions are included in this version of Open MPI:
- pcollreq: Provides routines for persistent collective
communication operations and persistent neighborhood collective
communication operations, which are proposed in the MPI Forum as
of June 2018. The function names are prefixed with MPIX_ instead
of MPI_, like MPIX_Barrier_init, because they are not standardized
yet. Future versions of Open MPI will switch to the MPI_ prefix
once the MPI Standard which includes this feature is published.
See their man page for more details.
- affinity: Provides the OMPI_Affinity_str() routine on retrieving
a string that contains what resources a process is bound to. See
its man page for more details.

60
ompi/mca/coll/base/coll_base_comm_select.c
Просмотреть файл

@ -20,6 +20,7 @@
* Copyright (c) 2014 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2016-2017 IBM Corporation. All rights reserved.
* Copyright (c) 2017 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -52,7 +53,7 @@ struct avail_coll_t {
opal_list_item_t super;
int ac_priority;
mca_coll_base_module_2_2_0_t *ac_module;
mca_coll_base_module_2_3_0_t *ac_module;
const char * ac_component_name;
};
typedef struct avail_coll_t avail_coll_t;
@ -65,16 +66,16 @@ static opal_list_t *check_components(opal_list_t * components,
ompi_communicator_t * comm);
static int check_one_component(ompi_communicator_t * comm,
const mca_base_component_t * component,
mca_coll_base_module_2_2_0_t ** module);
mca_coll_base_module_2_3_0_t ** module);
static int query(const mca_base_component_t * component,
ompi_communicator_t * comm, int *priority,
mca_coll_base_module_2_2_0_t ** module);
mca_coll_base_module_2_3_0_t ** module);
static int query_2_0_0(const mca_coll_base_component_2_0_0_t *
coll_component, ompi_communicator_t * comm,
int *priority,
mca_coll_base_module_2_2_0_t ** module);
mca_coll_base_module_2_3_0_t ** module);
/*
* Stuff for the OBJ interface
@ -190,6 +191,24 @@ int mca_coll_base_comm_select(ompi_communicator_t * comm)
COPY(avail->ac_module, comm, iscatter);
COPY(avail->ac_module, comm, iscatterv);
COPY(avail->ac_module, comm, allgather_init);
COPY(avail->ac_module, comm, allgatherv_init);
COPY(avail->ac_module, comm, allreduce_init);
COPY(avail->ac_module, comm, alltoall_init);
COPY(avail->ac_module, comm, alltoallv_init);
COPY(avail->ac_module, comm, alltoallw_init);
COPY(avail->ac_module, comm, barrier_init);
COPY(avail->ac_module, comm, bcast_init);
COPY(avail->ac_module, comm, exscan_init);
COPY(avail->ac_module, comm, gather_init);
COPY(avail->ac_module, comm, gatherv_init);
COPY(avail->ac_module, comm, reduce_init);
COPY(avail->ac_module, comm, reduce_scatter_block_init);
COPY(avail->ac_module, comm, reduce_scatter_init);
COPY(avail->ac_module, comm, scan_init);
COPY(avail->ac_module, comm, scatter_init);
COPY(avail->ac_module, comm, scatterv_init);
/* We can not reliably check if this comm has a topology
* at this time. The flags are set *after* coll_select */
COPY(avail->ac_module, comm, neighbor_allgather);
@ -204,6 +223,12 @@ int mca_coll_base_comm_select(ompi_communicator_t * comm)
COPY(avail->ac_module, comm, ineighbor_alltoallv);
COPY(avail->ac_module, comm, ineighbor_alltoallw);
COPY(avail->ac_module, comm, neighbor_allgather_init);
COPY(avail->ac_module, comm, neighbor_allgatherv_init);
COPY(avail->ac_module, comm, neighbor_alltoall_init);
COPY(avail->ac_module, comm, neighbor_alltoallv_init);
COPY(avail->ac_module, comm, neighbor_alltoallw_init);
COPY(avail->ac_module, comm, reduce_local);
}
/* release the original module reference and the list item */
@ -249,6 +274,23 @@ int mca_coll_base_comm_select(ompi_communicator_t * comm)
((OMPI_COMM_IS_INTRA(comm)) && CHECK_NULL(which_func, comm, iscan)) ||
CHECK_NULL(which_func, comm, iscatter) ||
CHECK_NULL(which_func, comm, iscatterv) ||
CHECK_NULL(which_func, comm, allgather_init) ||
CHECK_NULL(which_func, comm, allgatherv_init) ||
CHECK_NULL(which_func, comm, allreduce_init) ||
CHECK_NULL(which_func, comm, alltoall_init) ||
CHECK_NULL(which_func, comm, alltoallv_init) ||
CHECK_NULL(which_func, comm, alltoallw_init) ||
CHECK_NULL(which_func, comm, barrier_init) ||
CHECK_NULL(which_func, comm, bcast_init) ||
((OMPI_COMM_IS_INTRA(comm)) && CHECK_NULL(which_func, comm, exscan_init)) ||
CHECK_NULL(which_func, comm, gather_init) ||
CHECK_NULL(which_func, comm, gatherv_init) ||
CHECK_NULL(which_func, comm, reduce_init) ||
CHECK_NULL(which_func, comm, reduce_scatter_block_init) ||
CHECK_NULL(which_func, comm, reduce_scatter_init) ||
((OMPI_COMM_IS_INTRA(comm)) && CHECK_NULL(which_func, comm, scan_init)) ||
CHECK_NULL(which_func, comm, scatter_init) ||
CHECK_NULL(which_func, comm, scatterv_init) ||
CHECK_NULL(which_func, comm, reduce_local) ) {
/* TODO -- Once the topology flags are set before coll_select then
* check if neighborhood collectives have been set. */
@ -288,7 +330,7 @@ static opal_list_t *check_components(opal_list_t * components,
int priority;
const mca_base_component_t *component;
mca_base_component_list_item_t *cli;
mca_coll_base_module_2_2_0_t *module;
mca_coll_base_module_2_3_0_t *module;
opal_list_t *selectable;
avail_coll_t *avail;
@ -344,7 +386,7 @@ static opal_list_t *check_components(opal_list_t * components,
*/
static int check_one_component(ompi_communicator_t * comm,
const mca_base_component_t * component,
mca_coll_base_module_2_2_0_t ** module)
mca_coll_base_module_2_3_0_t ** module)
{
int err;
int priority = -1;
@ -378,7 +420,7 @@ static int check_one_component(ompi_communicator_t * comm,
*/
static int query(const mca_base_component_t * component,
ompi_communicator_t * comm,
int *priority, mca_coll_base_module_2_2_0_t ** module)
int *priority, mca_coll_base_module_2_3_0_t ** module)
{
*module = NULL;
if (2 == component->mca_type_major_version &&
@ -398,9 +440,9 @@ static int query(const mca_base_component_t * component,
static int query_2_0_0(const mca_coll_base_component_2_0_0_t * component,
ompi_communicator_t * comm, int *priority,
mca_coll_base_module_2_2_0_t ** module)
mca_coll_base_module_2_3_0_t ** module)
{
mca_coll_base_module_2_2_0_t *ret;
mca_coll_base_module_2_3_0_t *ret;
/* There's currently no need for conversion */

25
ompi/mca/coll/base/coll_base_comm_unselect.c
Просмотреть файл

@ -15,6 +15,7 @@
* Copyright (c) 2014 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2017 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -85,6 +86,24 @@ int mca_coll_base_comm_unselect(ompi_communicator_t * comm)
CLOSE(comm, iscatter);
CLOSE(comm, iscatterv);
CLOSE(comm, allgather_init);
CLOSE(comm, allgatherv_init);
CLOSE(comm, allreduce_init);
CLOSE(comm, alltoall_init);
CLOSE(comm, alltoallv_init);
CLOSE(comm, alltoallw_init);
CLOSE(comm, barrier_init);
CLOSE(comm, bcast_init);
CLOSE(comm, exscan_init);
CLOSE(comm, gather_init);
CLOSE(comm, gatherv_init);
CLOSE(comm, reduce_init);
CLOSE(comm, reduce_scatter_block_init);
CLOSE(comm, reduce_scatter_init);
CLOSE(comm, scan_init);
CLOSE(comm, scatter_init);
CLOSE(comm, scatterv_init);
CLOSE(comm, neighbor_allgather);
CLOSE(comm, neighbor_allgatherv);
CLOSE(comm, neighbor_alltoall);
@ -97,6 +116,12 @@ int mca_coll_base_comm_unselect(ompi_communicator_t * comm)
CLOSE(comm, ineighbor_alltoallv);
CLOSE(comm, ineighbor_alltoallw);
CLOSE(comm, neighbor_allgather_init);
CLOSE(comm, neighbor_allgatherv_init);
CLOSE(comm, neighbor_alltoall_init);
CLOSE(comm, neighbor_alltoallv_init);
CLOSE(comm, neighbor_alltoallw_init);
CLOSE(comm, reduce_local);
free(comm->c_coll);

24
ompi/mca/coll/base/coll_base_functions.h
Просмотреть файл

@ -33,6 +33,7 @@
#include "ompi/mca/coll/base/base.h"
#include "ompi/mca/mca.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/info/info.h"
#include "ompi/request/request.h"
/* need to include our own topo prototypes so we can malloc data on the comm correctly */
@ -135,6 +136,29 @@ typedef enum COLLTYPE {
#define INEIGHBOR_ALLTOALLV_ARGS NEIGHBOR_ALLTOALLV_BASE_ARGS, ompi_request_t **request, mca_coll_base_module_t *module
#define INEIGHBOR_ALLTOALLW_ARGS NEIGHBOR_ALLTOALLW_BASE_ARGS, ompi_request_t **request, mca_coll_base_module_t *module
#define ALLGATHER_INIT_ARGS ALLGATHER_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define ALLGATHERV_INIT_ARGS ALLGATHERV_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define ALLREDUCE_INIT_ARGS ALLREDUCE_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define ALLTOALL_INIT_ARGS ALLTOALL_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define ALLTOALLV_INIT_ARGS ALLTOALLV_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define ALLTOALLW_INIT_ARGS ALLTOALLW_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define BARRIER_INIT_ARGS BARRIER_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define BCAST_INIT_ARGS BCAST_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define EXSCAN_INIT_ARGS EXSCAN_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define GATHER_INIT_ARGS GATHER_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define GATHERV_INIT_ARGS GATHERV_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define REDUCE_INIT_ARGS REDUCE_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define REDUCESCATTER_INIT_ARGS REDUCESCATTER_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define REDUCESCATTERBLOCK_INIT_ARGS REDUCESCATTERBLOCK_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define SCAN_INIT_ARGS SCAN_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define SCATTER_INIT_ARGS SCATTER_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define SCATTERV_INIT_ARGS SCATTERV_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define NEIGHBOR_ALLGATHER_INIT_ARGS NEIGHBOR_ALLGATHER_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define NEIGHBOR_ALLGATHERV_INIT_ARGS NEIGHBOR_ALLGATHERV_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define NEIGHBOR_ALLTOALL_INIT_ARGS NEIGHBOR_ALLTOALL_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define NEIGHBOR_ALLTOALLV_INIT_ARGS NEIGHBOR_ALLTOALLV_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define NEIGHBOR_ALLTOALLW_INIT_ARGS NEIGHBOR_ALLTOALLW_BASE_ARGS, ompi_info_t *info, ompi_request_t **request, mca_coll_base_module_t *module
#define ALLGATHER_BASE_ARG_NAMES sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm
#define ALLGATHERV_BASE_ARG_NAMES sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, comm
#define ALLREDUCE_BASE_ARG_NAMES sendbuf, recvbuf, count, datatype, op, comm

345
ompi/mca/coll/coll.h
Просмотреть файл

@ -18,6 +18,7 @@
* Copyright (c) 2014-2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2016-2017 IBM Corporation. All rights reserved.
* Copyright (c) 2017 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -137,7 +138,7 @@ typedef int (*mca_coll_base_component_init_query_fn_t)
* provide a module with the requested functionality or NULL if the
* component should not be used on the given communicator.
*/
typedef struct mca_coll_base_module_2_2_0_t *
typedef struct mca_coll_base_module_2_3_0_t *
(*mca_coll_base_component_comm_query_2_0_0_fn_t)
(struct ompi_communicator_t *comm, int *priority);
@ -177,7 +178,7 @@ typedef struct mca_coll_base_module_2_2_0_t *
* @param[in] comm Communicator being created
*/
typedef int
(*mca_coll_base_module_enable_1_1_0_fn_t)(struct mca_coll_base_module_2_2_0_t* module,
(*mca_coll_base_module_enable_1_1_0_fn_t)(struct mca_coll_base_module_2_3_0_t* module,
struct ompi_communicator_t *comm);
@ -192,147 +193,225 @@ typedef int
* @param[in] comm Communicator being disabled
*/
typedef int
(*mca_coll_base_module_disable_1_2_0_fn_t)(struct mca_coll_base_module_2_2_0_t* module,
(*mca_coll_base_module_disable_1_2_0_fn_t)(struct mca_coll_base_module_2_3_0_t* module,
struct ompi_communicator_t *comm);
/* blocking collectives */
typedef int (*mca_coll_base_module_allgather_fn_t)
(const void *sbuf, int scount, struct ompi_datatype_t *sdtype,
void *rbuf, int rcount, struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm, struct mca_coll_base_module_2_2_0_t *module);
struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_allgatherv_fn_t)
(const void *sbuf, int scount, struct ompi_datatype_t *sdtype,
void * rbuf, const int *rcounts, const int *disps, struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm, struct mca_coll_base_module_2_2_0_t *module);
struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_allreduce_fn_t)
(const void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_2_0_t *module);
struct ompi_op_t *op, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_alltoall_fn_t)
(const void *sbuf, int scount, struct ompi_datatype_t *sdtype,
void* rbuf, int rcount, struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm, struct mca_coll_base_module_2_2_0_t *module);
struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_alltoallv_fn_t)
(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, struct mca_coll_base_module_2_2_0_t *module);
struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_alltoallw_fn_t)
(const void *sbuf, const int *scounts, const int *sdisps, struct ompi_datatype_t * const *sdtypes,
void *rbuf, const int *rcounts, const int *rdisps, struct ompi_datatype_t * const *rdtypes,
struct ompi_communicator_t *comm, struct mca_coll_base_module_2_2_0_t *module);
struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_barrier_fn_t)
(struct ompi_communicator_t *comm, struct mca_coll_base_module_2_2_0_t *module);
(struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_bcast_fn_t)
(void *buff, int count, struct ompi_datatype_t *datatype, int root,
struct ompi_communicator_t *comm, struct mca_coll_base_module_2_2_0_t *module);
struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_exscan_fn_t)
(const void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_2_0_t *module);
struct ompi_op_t *op, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_gather_fn_t)
(const void *sbuf, int scount, struct ompi_datatype_t *sdtype,
void *rbuf, int rcount, struct ompi_datatype_t *rdtype,
int root, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_2_0_t *module);
int root, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_gatherv_fn_t)
(const void *sbuf, int scount, struct ompi_datatype_t *sdtype,
void *rbuf, const int *rcounts, const int *disps, struct ompi_datatype_t *rdtype,
int root, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_2_0_t *module);
int root, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_reduce_fn_t)
(const void *sbuf, void* rbuf, int count, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, int root, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_2_0_t *module);
struct ompi_op_t *op, int root, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_reduce_scatter_fn_t)
(const void *sbuf, void *rbuf, const int *rcounts, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_2_0_t *module);
struct ompi_op_t *op, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_reduce_scatter_block_fn_t)
(const void *sbuf, void *rbuf, int rcount, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_2_0_t *module);
struct ompi_op_t *op, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_scan_fn_t)
(const void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_2_0_t *module);
struct ompi_op_t *op, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_scatter_fn_t)
(const void *sbuf, int scount, struct ompi_datatype_t *sdtype,
void *rbuf, int rcount, struct ompi_datatype_t *rdtype,
int root, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_2_0_t *module);
int root, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_scatterv_fn_t)
(const void *sbuf, const int *scounts, const int *disps, struct ompi_datatype_t *sdtype,
void* rbuf, int rcount, struct ompi_datatype_t *rdtype,
int root, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_2_0_t *module);
int root, struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
/* nonblocking collectives */
typedef int (*mca_coll_base_module_iallgather_fn_t)
(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,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_iallgatherv_fn_t)
(const void *sbuf, int scount, struct ompi_datatype_t *sdtype,
void * rbuf, const int *rcounts, const int *disps, struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_iallreduce_fn_t)
(const void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm,
ompi_request_t ** request, struct mca_coll_base_module_2_2_0_t *module);
ompi_request_t ** request, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_ialltoall_fn_t)
(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,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_ialltoallv_fn_t)
(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,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_ialltoallw_fn_t)
(const void *sbuf, const int *scounts, const int *sdisps, struct ompi_datatype_t * const *sdtypes,
void *rbuf, const int *rcounts, const int *rdisps, struct ompi_datatype_t * const *rdtypes,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_ibarrier_fn_t)
(struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_ibcast_fn_t)
(void *buff, int count, struct ompi_datatype_t *datatype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_iexscan_fn_t)
(const void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_igather_fn_t)
(const void *sbuf, int scount, struct ompi_datatype_t *sdtype,
void *rbuf, int rcount, struct ompi_datatype_t *rdtype,
int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_igatherv_fn_t)
(const void *sbuf, int scount, struct ompi_datatype_t *sdtype,
void *rbuf, const int *rcounts, const int *disps, struct ompi_datatype_t *rdtype,
int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_ireduce_fn_t)
(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,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_ireduce_scatter_fn_t)
(const void *sbuf, void *rbuf, const int *rcounts, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_ireduce_scatter_block_fn_t)
(const void *sbuf, void *rbuf, int rcount, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_iscan_fn_t)
(const void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_iscatter_fn_t)
(const void *sbuf, int scount, struct ompi_datatype_t *sdtype,
void *rbuf, int rcount, struct ompi_datatype_t *rdtype,
int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_iscatterv_fn_t)
(const void *sbuf, const int *scounts, const int *disps, struct ompi_datatype_t *sdtype,
void* rbuf, int rcount, struct ompi_datatype_t *rdtype,
int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
/* persistent collectives */
typedef int (*mca_coll_base_module_allgather_init_fn_t)
(const void *sbuf, int scount, struct ompi_datatype_t *sdtype,
void *rbuf, int rcount, struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm, struct ompi_info_t *info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_allgatherv_init_fn_t)
(const void *sbuf, int scount, struct ompi_datatype_t *sdtype,
void * rbuf, const int *rcounts, const int *disps, struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm, struct ompi_info_t *info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_allreduce_init_fn_t)
(const void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, struct ompi_info_t *info,
ompi_request_t ** request, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_alltoall_init_fn_t)
(const void *sbuf, int scount, struct ompi_datatype_t *sdtype,
void* rbuf, int rcount, struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm, struct ompi_info_t *info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_alltoallv_init_fn_t)
(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, struct ompi_info_t *info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_alltoallw_init_fn_t)
(const void *sbuf, const int *scounts, const int *sdisps, struct ompi_datatype_t * const *sdtypes,
void *rbuf, const int *rcounts, const int *rdisps, struct ompi_datatype_t * const *rdtypes,
struct ompi_communicator_t *comm, struct ompi_info_t *info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_barrier_init_fn_t)
(struct ompi_communicator_t *comm, struct ompi_info_t *info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_bcast_init_fn_t)
(void *buff, int count, struct ompi_datatype_t *datatype, int root,
struct ompi_communicator_t *comm, struct ompi_info_t *info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_exscan_init_fn_t)
(const void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, struct ompi_info_t *info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_gather_init_fn_t)
(const void *sbuf, int scount, struct ompi_datatype_t *sdtype,
void *rbuf, int rcount, struct ompi_datatype_t *rdtype,
int root, struct ompi_communicator_t *comm, struct ompi_info_t *info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_gatherv_init_fn_t)
(const void *sbuf, int scount, struct ompi_datatype_t *sdtype,
void *rbuf, const int *rcounts, const int *disps, struct ompi_datatype_t *rdtype,
int root, struct ompi_communicator_t *comm, struct ompi_info_t *info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_reduce_init_fn_t)
(const void *sbuf, void* rbuf, int count, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, int root, struct ompi_communicator_t *comm, struct ompi_info_t *info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_reduce_scatter_init_fn_t)
(const void *sbuf, void *rbuf, const int *rcounts, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, struct ompi_info_t *info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_reduce_scatter_block_init_fn_t)
(const void *sbuf, void *rbuf, int rcount, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, struct ompi_info_t *info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_scan_init_fn_t)
(const void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, struct ompi_info_t *info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_scatter_init_fn_t)
(const void *sbuf, int scount, struct ompi_datatype_t *sdtype,
void *rbuf, int rcount, struct ompi_datatype_t *rdtype,
int root, struct ompi_communicator_t *comm, struct ompi_info_t *info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_scatterv_init_fn_t)
(const void *sbuf, const int *scounts, const int *disps, struct ompi_datatype_t *sdtype,
void* rbuf, int rcount, struct ompi_datatype_t *rdtype,
int root, struct ompi_communicator_t *comm, struct ompi_info_t *info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
/*
* The signature of the neighborhood alltoallw differs from alltoallw
@ -340,12 +419,17 @@ typedef int (*mca_coll_base_module_iscatterv_fn_t)
typedef int (*mca_coll_base_module_neighbor_alltoallw_fn_t)
(const void *sbuf, const int *scounts, const MPI_Aint *sdisps, struct ompi_datatype_t * const *sdtypes,
void *rbuf, const int *rcounts, const MPI_Aint *rdisps, struct ompi_datatype_t * const *rdtypes,
struct ompi_communicator_t *comm, struct mca_coll_base_module_2_2_0_t *module);
struct ompi_communicator_t *comm, struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_ineighbor_alltoallw_fn_t)
(const void *sbuf, const int *scounts, const MPI_Aint *sdisps, struct ompi_datatype_t * const *sdtypes,
void *rbuf, const int *rcounts, const MPI_Aint *rdisps, struct ompi_datatype_t * const *rdtypes,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
typedef int (*mca_coll_base_module_neighbor_alltoallw_init_fn_t)
(const void *sbuf, const int *scounts, const MPI_Aint *sdisps, struct ompi_datatype_t * const *sdtypes,
void *rbuf, const int *rcounts, const MPI_Aint *rdisps, struct ompi_datatype_t * const *rdtypes,
struct ompi_communicator_t *comm, struct ompi_info_t *info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
/*
* reduce_local
@ -356,7 +440,7 @@ typedef int (*mca_coll_base_module_ineighbor_alltoallw_fn_t)
typedef int (*mca_coll_base_module_reduce_local_fn_t)
(const void *inbuf, void *inoutbuf, int count,
struct ompi_datatype_t * dtype, struct ompi_op_t * op,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
/**
@ -417,7 +501,7 @@ typedef struct mca_coll_base_component_2_0_0_t mca_coll_base_component_t;
* function, so the component is free to create a structure that
* inherits from this one for use as the module structure.
*/
struct mca_coll_base_module_2_2_0_t {
struct mca_coll_base_module_2_3_0_t {
/** Collective modules all inherit from opal_object */
opal_object_t super;
@ -426,6 +510,7 @@ struct mca_coll_base_module_2_2_0_t {
mca_coll_base_module_enable_1_1_0_fn_t coll_module_enable;
/* Collective function pointers */
/* blocking functions */
mca_coll_base_module_allgather_fn_t coll_allgather;
mca_coll_base_module_allgatherv_fn_t coll_allgatherv;
@ -444,6 +529,7 @@ struct mca_coll_base_module_2_2_0_t {
mca_coll_base_module_scan_fn_t coll_scan;
mca_coll_base_module_scatter_fn_t coll_scatter;
mca_coll_base_module_scatterv_fn_t coll_scatterv;
/* nonblocking functions */
mca_coll_base_module_iallgather_fn_t coll_iallgather;
mca_coll_base_module_iallgatherv_fn_t coll_iallgatherv;
@ -463,6 +549,25 @@ struct mca_coll_base_module_2_2_0_t {
mca_coll_base_module_iscatter_fn_t coll_iscatter;
mca_coll_base_module_iscatterv_fn_t coll_iscatterv;
/* persistent functions */
mca_coll_base_module_allgather_init_fn_t coll_allgather_init;
mca_coll_base_module_allgatherv_init_fn_t coll_allgatherv_init;
mca_coll_base_module_allreduce_init_fn_t coll_allreduce_init;
mca_coll_base_module_alltoall_init_fn_t coll_alltoall_init;
mca_coll_base_module_alltoallv_init_fn_t coll_alltoallv_init;
mca_coll_base_module_alltoallw_init_fn_t coll_alltoallw_init;
mca_coll_base_module_barrier_init_fn_t coll_barrier_init;
mca_coll_base_module_bcast_init_fn_t coll_bcast_init;
mca_coll_base_module_exscan_init_fn_t coll_exscan_init;
mca_coll_base_module_gather_init_fn_t coll_gather_init;
mca_coll_base_module_gatherv_init_fn_t coll_gatherv_init;
mca_coll_base_module_reduce_init_fn_t coll_reduce_init;
mca_coll_base_module_reduce_scatter_init_fn_t coll_reduce_scatter_init;
mca_coll_base_module_reduce_scatter_block_init_fn_t coll_reduce_scatter_block_init;
mca_coll_base_module_scan_init_fn_t coll_scan_init;
mca_coll_base_module_scatter_init_fn_t coll_scatter_init;
mca_coll_base_module_scatterv_init_fn_t coll_scatterv_init;
/* neighborhood functions */
mca_coll_base_module_allgather_fn_t coll_neighbor_allgather;
mca_coll_base_module_allgatherv_fn_t coll_neighbor_allgatherv;
@ -470,12 +575,20 @@ struct mca_coll_base_module_2_2_0_t {
mca_coll_base_module_alltoallv_fn_t coll_neighbor_alltoallv;
mca_coll_base_module_neighbor_alltoallw_fn_t coll_neighbor_alltoallw;
/* nonblocking neighborhood functions */
mca_coll_base_module_iallgather_fn_t coll_ineighbor_allgather;
mca_coll_base_module_iallgatherv_fn_t coll_ineighbor_allgatherv;
mca_coll_base_module_ialltoall_fn_t coll_ineighbor_alltoall;
mca_coll_base_module_ialltoallv_fn_t coll_ineighbor_alltoallv;
mca_coll_base_module_ineighbor_alltoallw_fn_t coll_ineighbor_alltoallw;
/* persistent neighborhood functions */
mca_coll_base_module_allgather_init_fn_t coll_neighbor_allgather_init;
mca_coll_base_module_allgatherv_init_fn_t coll_neighbor_allgatherv_init;
mca_coll_base_module_alltoall_init_fn_t coll_neighbor_alltoall_init;
mca_coll_base_module_alltoallv_init_fn_t coll_neighbor_alltoallv_init;
mca_coll_base_module_neighbor_alltoallw_init_fn_t coll_neighbor_alltoallw_init;
/** Fault tolerance event trigger function */
mca_coll_base_module_ft_event_fn_t ft_event;
@ -489,12 +602,12 @@ struct mca_coll_base_module_2_2_0_t {
not be used by other modules */
struct mca_coll_base_comm_t* base_data;
};
typedef struct mca_coll_base_module_2_2_0_t mca_coll_base_module_2_2_0_t;
typedef struct mca_coll_base_module_2_3_0_t mca_coll_base_module_2_3_0_t;
/** Per guidence in mca.h, use the unversioned struct name if you just
want to always keep up with the most recent version of the
interace. */
typedef struct mca_coll_base_module_2_2_0_t mca_coll_base_module_t;
typedef struct mca_coll_base_module_2_3_0_t mca_coll_base_module_t;
OMPI_DECLSPEC OBJ_CLASS_DECLARATION(mca_coll_base_module_t);
/**
@ -507,101 +620,153 @@ OMPI_DECLSPEC OBJ_CLASS_DECLARATION(mca_coll_base_module_t);
* component.
*/
struct mca_coll_base_comm_coll_t {
/* blocking collectives */
mca_coll_base_module_allgather_fn_t coll_allgather;
mca_coll_base_module_2_2_0_t *coll_allgather_module;
mca_coll_base_module_2_3_0_t *coll_allgather_module;
mca_coll_base_module_allgatherv_fn_t coll_allgatherv;
mca_coll_base_module_2_2_0_t *coll_allgatherv_module;
mca_coll_base_module_2_3_0_t *coll_allgatherv_module;
mca_coll_base_module_allreduce_fn_t coll_allreduce;
mca_coll_base_module_2_2_0_t *coll_allreduce_module;
mca_coll_base_module_2_3_0_t *coll_allreduce_module;
mca_coll_base_module_alltoall_fn_t coll_alltoall;
mca_coll_base_module_2_2_0_t *coll_alltoall_module;
mca_coll_base_module_2_3_0_t *coll_alltoall_module;
mca_coll_base_module_alltoallv_fn_t coll_alltoallv;
mca_coll_base_module_2_2_0_t *coll_alltoallv_module;
mca_coll_base_module_2_3_0_t *coll_alltoallv_module;
mca_coll_base_module_alltoallw_fn_t coll_alltoallw;
mca_coll_base_module_2_2_0_t *coll_alltoallw_module;
mca_coll_base_module_2_3_0_t *coll_alltoallw_module;
mca_coll_base_module_barrier_fn_t coll_barrier;
mca_coll_base_module_2_2_0_t *coll_barrier_module;
mca_coll_base_module_2_3_0_t *coll_barrier_module;
mca_coll_base_module_bcast_fn_t coll_bcast;
mca_coll_base_module_2_2_0_t *coll_bcast_module;
mca_coll_base_module_2_3_0_t *coll_bcast_module;
mca_coll_base_module_exscan_fn_t coll_exscan;
mca_coll_base_module_2_2_0_t *coll_exscan_module;
mca_coll_base_module_2_3_0_t *coll_exscan_module;
mca_coll_base_module_gather_fn_t coll_gather;
mca_coll_base_module_2_2_0_t *coll_gather_module;
mca_coll_base_module_2_3_0_t *coll_gather_module;
mca_coll_base_module_gatherv_fn_t coll_gatherv;
mca_coll_base_module_2_2_0_t *coll_gatherv_module;
mca_coll_base_module_2_3_0_t *coll_gatherv_module;
mca_coll_base_module_reduce_fn_t coll_reduce;
mca_coll_base_module_2_2_0_t *coll_reduce_module;
mca_coll_base_module_2_3_0_t *coll_reduce_module;
mca_coll_base_module_reduce_scatter_fn_t coll_reduce_scatter;
mca_coll_base_module_2_2_0_t *coll_reduce_scatter_module;
mca_coll_base_module_2_3_0_t *coll_reduce_scatter_module;
mca_coll_base_module_reduce_scatter_block_fn_t coll_reduce_scatter_block;
mca_coll_base_module_2_2_0_t *coll_reduce_scatter_block_module;
mca_coll_base_module_2_3_0_t *coll_reduce_scatter_block_module;
mca_coll_base_module_scan_fn_t coll_scan;
mca_coll_base_module_2_2_0_t *coll_scan_module;
mca_coll_base_module_2_3_0_t *coll_scan_module;
mca_coll_base_module_scatter_fn_t coll_scatter;
mca_coll_base_module_2_2_0_t *coll_scatter_module;
mca_coll_base_module_2_3_0_t *coll_scatter_module;
mca_coll_base_module_scatterv_fn_t coll_scatterv;
mca_coll_base_module_2_2_0_t *coll_scatterv_module;
mca_coll_base_module_2_3_0_t *coll_scatterv_module;
/* nonblocking collectives */
mca_coll_base_module_iallgather_fn_t coll_iallgather;
mca_coll_base_module_2_2_0_t *coll_iallgather_module;
mca_coll_base_module_2_3_0_t *coll_iallgather_module;
mca_coll_base_module_iallgatherv_fn_t coll_iallgatherv;
mca_coll_base_module_2_2_0_t *coll_iallgatherv_module;
mca_coll_base_module_2_3_0_t *coll_iallgatherv_module;
mca_coll_base_module_iallreduce_fn_t coll_iallreduce;
mca_coll_base_module_2_2_0_t *coll_iallreduce_module;
mca_coll_base_module_2_3_0_t *coll_iallreduce_module;
mca_coll_base_module_ialltoall_fn_t coll_ialltoall;
mca_coll_base_module_2_2_0_t *coll_ialltoall_module;
mca_coll_base_module_2_3_0_t *coll_ialltoall_module;
mca_coll_base_module_ialltoallv_fn_t coll_ialltoallv;
mca_coll_base_module_2_2_0_t *coll_ialltoallv_module;
mca_coll_base_module_2_3_0_t *coll_ialltoallv_module;
mca_coll_base_module_ialltoallw_fn_t coll_ialltoallw;
mca_coll_base_module_2_2_0_t *coll_ialltoallw_module;
mca_coll_base_module_2_3_0_t *coll_ialltoallw_module;
mca_coll_base_module_ibarrier_fn_t coll_ibarrier;
mca_coll_base_module_2_2_0_t *coll_ibarrier_module;
mca_coll_base_module_2_3_0_t *coll_ibarrier_module;
mca_coll_base_module_ibcast_fn_t coll_ibcast;
mca_coll_base_module_2_2_0_t *coll_ibcast_module;
mca_coll_base_module_2_3_0_t *coll_ibcast_module;
mca_coll_base_module_iexscan_fn_t coll_iexscan;
mca_coll_base_module_2_2_0_t *coll_iexscan_module;
mca_coll_base_module_2_3_0_t *coll_iexscan_module;
mca_coll_base_module_igather_fn_t coll_igather;
mca_coll_base_module_2_2_0_t *coll_igather_module;
mca_coll_base_module_2_3_0_t *coll_igather_module;
mca_coll_base_module_igatherv_fn_t coll_igatherv;
mca_coll_base_module_2_2_0_t *coll_igatherv_module;
mca_coll_base_module_2_3_0_t *coll_igatherv_module;
mca_coll_base_module_ireduce_fn_t coll_ireduce;
mca_coll_base_module_2_2_0_t *coll_ireduce_module;
mca_coll_base_module_2_3_0_t *coll_ireduce_module;
mca_coll_base_module_ireduce_scatter_fn_t coll_ireduce_scatter;
mca_coll_base_module_2_2_0_t *coll_ireduce_scatter_module;
mca_coll_base_module_2_3_0_t *coll_ireduce_scatter_module;
mca_coll_base_module_ireduce_scatter_block_fn_t coll_ireduce_scatter_block;
mca_coll_base_module_2_2_0_t *coll_ireduce_scatter_block_module;
mca_coll_base_module_2_3_0_t *coll_ireduce_scatter_block_module;
mca_coll_base_module_iscan_fn_t coll_iscan;
mca_coll_base_module_2_2_0_t *coll_iscan_module;
mca_coll_base_module_2_3_0_t *coll_iscan_module;
mca_coll_base_module_iscatter_fn_t coll_iscatter;
mca_coll_base_module_2_2_0_t *coll_iscatter_module;
mca_coll_base_module_2_3_0_t *coll_iscatter_module;
mca_coll_base_module_iscatterv_fn_t coll_iscatterv;
mca_coll_base_module_2_2_0_t *coll_iscatterv_module;
mca_coll_base_module_2_3_0_t *coll_iscatterv_module;
/* neighborhood collectives */
/* persistent collectives */
mca_coll_base_module_allgather_init_fn_t coll_allgather_init;
mca_coll_base_module_2_3_0_t *coll_allgather_init_module;
mca_coll_base_module_allgatherv_init_fn_t coll_allgatherv_init;
mca_coll_base_module_2_3_0_t *coll_allgatherv_init_module;
mca_coll_base_module_allreduce_init_fn_t coll_allreduce_init;
mca_coll_base_module_2_3_0_t *coll_allreduce_init_module;
mca_coll_base_module_alltoall_init_fn_t coll_alltoall_init;
mca_coll_base_module_2_3_0_t *coll_alltoall_init_module;
mca_coll_base_module_alltoallv_init_fn_t coll_alltoallv_init;
mca_coll_base_module_2_3_0_t *coll_alltoallv_init_module;
mca_coll_base_module_alltoallw_init_fn_t coll_alltoallw_init;
mca_coll_base_module_2_3_0_t *coll_alltoallw_init_module;
mca_coll_base_module_barrier_init_fn_t coll_barrier_init;
mca_coll_base_module_2_3_0_t *coll_barrier_init_module;
mca_coll_base_module_bcast_init_fn_t coll_bcast_init;
mca_coll_base_module_2_3_0_t *coll_bcast_init_module;
mca_coll_base_module_exscan_init_fn_t coll_exscan_init;
mca_coll_base_module_2_3_0_t *coll_exscan_init_module;
mca_coll_base_module_gather_init_fn_t coll_gather_init;
mca_coll_base_module_2_3_0_t *coll_gather_init_module;
mca_coll_base_module_gatherv_init_fn_t coll_gatherv_init;
mca_coll_base_module_2_3_0_t *coll_gatherv_init_module;
mca_coll_base_module_reduce_init_fn_t coll_reduce_init;
mca_coll_base_module_2_3_0_t *coll_reduce_init_module;
mca_coll_base_module_reduce_scatter_init_fn_t coll_reduce_scatter_init;
mca_coll_base_module_2_3_0_t *coll_reduce_scatter_init_module;
mca_coll_base_module_reduce_scatter_block_init_fn_t coll_reduce_scatter_block_init;
mca_coll_base_module_2_3_0_t *coll_reduce_scatter_block_init_module;
mca_coll_base_module_scan_init_fn_t coll_scan_init;
mca_coll_base_module_2_3_0_t *coll_scan_init_module;
mca_coll_base_module_scatter_init_fn_t coll_scatter_init;
mca_coll_base_module_2_3_0_t *coll_scatter_init_module;
mca_coll_base_module_scatterv_init_fn_t coll_scatterv_init;
mca_coll_base_module_2_3_0_t *coll_scatterv_init_module;
/* blocking neighborhood collectives */
mca_coll_base_module_allgather_fn_t coll_neighbor_allgather;
mca_coll_base_module_2_2_0_t *coll_neighbor_allgather_module;
mca_coll_base_module_2_3_0_t *coll_neighbor_allgather_module;
mca_coll_base_module_allgatherv_fn_t coll_neighbor_allgatherv;
mca_coll_base_module_2_2_0_t *coll_neighbor_allgatherv_module;
mca_coll_base_module_2_3_0_t *coll_neighbor_allgatherv_module;
mca_coll_base_module_alltoall_fn_t coll_neighbor_alltoall;
mca_coll_base_module_2_2_0_t *coll_neighbor_alltoall_module;
mca_coll_base_module_2_3_0_t *coll_neighbor_alltoall_module;
mca_coll_base_module_alltoallv_fn_t coll_neighbor_alltoallv;
mca_coll_base_module_2_2_0_t *coll_neighbor_alltoallv_module;
mca_coll_base_module_2_3_0_t *coll_neighbor_alltoallv_module;
mca_coll_base_module_neighbor_alltoallw_fn_t coll_neighbor_alltoallw;
mca_coll_base_module_2_2_0_t *coll_neighbor_alltoallw_module;
mca_coll_base_module_2_3_0_t *coll_neighbor_alltoallw_module;
/* nonblocking neighborhood collectives */
mca_coll_base_module_iallgather_fn_t coll_ineighbor_allgather;
mca_coll_base_module_2_2_0_t *coll_ineighbor_allgather_module;
mca_coll_base_module_2_3_0_t *coll_ineighbor_allgather_module;
mca_coll_base_module_iallgatherv_fn_t coll_ineighbor_allgatherv;
mca_coll_base_module_2_2_0_t *coll_ineighbor_allgatherv_module;
mca_coll_base_module_2_3_0_t *coll_ineighbor_allgatherv_module;
mca_coll_base_module_ialltoall_fn_t coll_ineighbor_alltoall;
mca_coll_base_module_2_2_0_t *coll_ineighbor_alltoall_module;
mca_coll_base_module_2_3_0_t *coll_ineighbor_alltoall_module;
mca_coll_base_module_ialltoallv_fn_t coll_ineighbor_alltoallv;
mca_coll_base_module_2_2_0_t *coll_ineighbor_alltoallv_module;
mca_coll_base_module_2_3_0_t *coll_ineighbor_alltoallv_module;
mca_coll_base_module_ineighbor_alltoallw_fn_t coll_ineighbor_alltoallw;
mca_coll_base_module_2_2_0_t *coll_ineighbor_alltoallw_module;
mca_coll_base_module_2_3_0_t *coll_ineighbor_alltoallw_module;
/* persistent neighborhood collectives */
mca_coll_base_module_allgather_init_fn_t coll_neighbor_allgather_init;
mca_coll_base_module_2_3_0_t *coll_neighbor_allgather_init_module;
mca_coll_base_module_allgatherv_init_fn_t coll_neighbor_allgatherv_init;
mca_coll_base_module_2_3_0_t *coll_neighbor_allgatherv_init_module;
mca_coll_base_module_alltoall_init_fn_t coll_neighbor_alltoall_init;
mca_coll_base_module_2_3_0_t *coll_neighbor_alltoall_init_module;
mca_coll_base_module_alltoallv_init_fn_t coll_neighbor_alltoallv_init;
mca_coll_base_module_2_3_0_t *coll_neighbor_alltoallv_init_module;
mca_coll_base_module_neighbor_alltoallw_init_fn_t coll_neighbor_alltoallw_init;
mca_coll_base_module_2_3_0_t *coll_neighbor_alltoallw_init_module;
mca_coll_base_module_reduce_local_fn_t coll_reduce_local;
mca_coll_base_module_2_2_0_t *coll_reduce_local_module;
mca_coll_base_module_2_3_0_t *coll_reduce_local_module;
};
typedef struct mca_coll_base_comm_coll_t mca_coll_base_comm_coll_t;

210
ompi/mca/coll/libnbc/coll_libnbc.h
Просмотреть файл

@ -13,9 +13,10 @@
* Copyright (c) 2008 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2013-2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2014-2016 Research Organization for Information Science
* Copyright (c) 2014-2017 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2016-2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -116,6 +117,7 @@ struct ompi_coll_libnbc_request_t {
ompi_request_t super;
MPI_Comm comm;
long row_offset;
bool nbc_complete; /* status in libnbc level */
int tag;
volatile int req_count;
ompi_request_t **req_array;
@ -131,15 +133,13 @@ OBJ_CLASS_DECLARATION(ompi_coll_libnbc_request_t);
typedef ompi_coll_libnbc_request_t NBC_Handle;
#define OMPI_COLL_LIBNBC_REQUEST_ALLOC(comm, req) \
#define OMPI_COLL_LIBNBC_REQUEST_ALLOC(comm, persistent, req) \
do { \
opal_free_list_item_t *item; \
item = opal_free_list_wait (&mca_coll_libnbc_component.requests); \
req = (ompi_coll_libnbc_request_t*) item; \
OMPI_REQUEST_INIT(&req->super, false); \
OMPI_REQUEST_INIT(&req->super, persistent); \
req->super.req_mpi_object.comm = comm; \
req->super.req_complete = false; \
req->super.req_state = OMPI_REQUEST_ACTIVE; \
} while (0)
#define OMPI_COLL_LIBNBC_REQUEST_RETURN(req) \
@ -157,132 +157,258 @@ int NBC_Progress(NBC_Handle *handle);
int ompi_coll_libnbc_iallgather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_iallgatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *displs,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_iallreduce(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ialltoall(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ialltoallv(const void* sendbuf, const int *sendcounts, const int *sdispls,
MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *rdispls,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ialltoallw(const void *sbuf, const int *scounts, const int *sdisps, struct ompi_datatype_t * const *sdtypes,
void *rbuf, const int *rcounts, const int *rdisps, struct ompi_datatype_t * const *rdtypes,
struct ompi_communicator_t *comm, ompi_request_t **request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ibarrier(struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ibcast(void *buffer, int count, MPI_Datatype datatype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_iexscan(const void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, ompi_request_t **request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_igather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_igatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, const int *recvcounts, const int *displs, MPI_Datatype recvtype,
int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ireduce(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype,
MPI_Op op, int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ireduce_scatter(const void* sendbuf, void* recvbuf, const int *recvcounts, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ireduce_scatter_block(const void *sbuf, void *rbuf, int rcount, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm,
ompi_request_t **request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_iscan(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_iscatter(const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_iscatterv(const void* sendbuf, const int *sendcounts, const int *displs, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_iallgather_inter(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_iallgatherv_inter(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *displs,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_iallreduce_inter(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ialltoall_inter(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ialltoallv_inter(const void* sendbuf, const int *sendcounts, const int *sdispls,
MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *rdispls,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ialltoallw_inter(const void *sbuf, const int *scounts, const int *sdisps, struct ompi_datatype_t * const *sdtypes,
void *rbuf, const int *rcounts, const int *rdisps, struct ompi_datatype_t * const *rdtypes,
struct ompi_communicator_t *comm, ompi_request_t **request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ibarrier_inter(struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ibcast_inter(void *buffer, int count, MPI_Datatype datatype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_igather_inter(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_igatherv_inter(const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, const int *recvcounts, const int *displs, MPI_Datatype recvtype,
int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ireduce_inter(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype,
MPI_Op op, int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ireduce_scatter_inter(const void* sendbuf, void* recvbuf, const int *recvcounts, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ireduce_scatter_block_inter(const void *sbuf, void *rbuf, int rcount, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm,
ompi_request_t **request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_iscatter_inter(const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_iscatterv_inter(const void* sendbuf, const int *sendcounts, const int *displs, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ineighbor_allgather(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
int rcount, MPI_Datatype rtype, struct ompi_communicator_t *comm,
ompi_request_t ** request, struct mca_coll_base_module_2_2_0_t *module);
ompi_request_t ** request, struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ineighbor_allgatherv(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
const int *rcounts, const int *displs, MPI_Datatype rtype,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ineighbor_alltoall(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
int rcount, MPI_Datatype rtype, struct ompi_communicator_t *comm,
ompi_request_t ** request, struct mca_coll_base_module_2_2_0_t *module);
ompi_request_t ** request, struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ineighbor_alltoallv(const void *sbuf, const int *scounts, const int *sdispls, MPI_Datatype stype,
void *rbuf, const int *rcounts, const int *rdispls, MPI_Datatype rtype,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_ineighbor_alltoallw(const void *sbuf, const int *scounts, const MPI_Aint *sdisps, struct ompi_datatype_t * const *stypes,
void *rbuf, const int *rcounts, const MPI_Aint *rdisps, struct ompi_datatype_t * const *rtypes,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_allgather_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_allgatherv_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *displs,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_allreduce_init(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_alltoall_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_alltoallv_init(const void* sendbuf, const int *sendcounts, const int *sdispls,
MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *rdispls,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_alltoallw_init(const void *sbuf, const int *scounts, const int *sdisps, struct ompi_datatype_t * const *sdtypes,
void *rbuf, const int *rcounts, const int *rdisps, struct ompi_datatype_t * const *rdtypes,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t **request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_barrier_init(struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_bcast_init(void *buffer, int count, MPI_Datatype datatype, int root,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_exscan_init(const void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t **request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_gather_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, int root, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_gatherv_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, const int *recvcounts, const int *displs, MPI_Datatype recvtype,
int root, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_reduce_init(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype,
MPI_Op op, int root, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_reduce_scatter_init(const void* sendbuf, void* recvbuf, const int *recvcounts, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_reduce_scatter_block_init(const void *sbuf, void *rbuf, int rcount, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t **request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_scan_init(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_scatter_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_scatterv_init(const void* sendbuf, const int *sendcounts, const int *displs, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_allgather_inter_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_allgatherv_inter_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *displs,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_allreduce_inter_init(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_alltoall_inter_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_alltoallv_inter_init(const void* sendbuf, const int *sendcounts, const int *sdispls,
MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *rdispls,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_alltoallw_inter_init(const void *sbuf, const int *scounts, const int *sdisps, struct ompi_datatype_t * const *sdtypes,
void *rbuf, const int *rcounts, const int *rdisps, struct ompi_datatype_t * const *rdtypes,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t **request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_barrier_inter_init(struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_bcast_inter_init(void *buffer, int count, MPI_Datatype datatype, int root,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_gather_inter_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, int root, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_gatherv_inter_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, const int *recvcounts, const int *displs, MPI_Datatype recvtype,
int root, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_reduce_inter_init(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype,
MPI_Op op, int root, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_reduce_scatter_inter_init(const void* sendbuf, void* recvbuf, const int *recvcounts, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_reduce_scatter_block_inter_init(const void *sbuf, void *rbuf, int rcount, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm,
MPI_Info info, ompi_request_t **request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_scatter_inter_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_scatterv_inter_init(const void* sendbuf, const int *sendcounts, const int *displs, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_neighbor_allgather_init(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
int rcount, MPI_Datatype rtype, struct ompi_communicator_t *comm,
MPI_Info info, ompi_request_t ** request, struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_neighbor_allgatherv_init(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
const int *rcounts, const int *displs, MPI_Datatype rtype,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_neighbor_alltoall_init(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
int rcount, MPI_Datatype rtype, struct ompi_communicator_t *comm, MPI_Info info,
ompi_request_t ** request, struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_neighbor_alltoallv_init(const void *sbuf, const int *scounts, const int *sdispls, MPI_Datatype stype,
void *rbuf, const int *rcounts, const int *rdispls, MPI_Datatype rtype,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_libnbc_neighbor_alltoallw_init(const void *sbuf, const int *scounts, const MPI_Aint *sdisps, struct ompi_datatype_t * const *stypes,
void *rbuf, const int *rcounts, const MPI_Aint *rdisps, struct ompi_datatype_t * const *rtypes,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module);
END_C_DECLS

96
ompi/mca/coll/libnbc/coll_libnbc_component.c
Просмотреть файл

@ -13,9 +13,12 @@
* Copyright (c) 2008 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2013-2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2016 Research Organization for Information Science
* Copyright (c) 2016-2017 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2016 IBM Corporation. All rights reserved.
* Copyright (c) 2017 Ian Bradley Morgan and Anthony Skjellum. All
* rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -26,6 +29,7 @@
#include "ompi_config.h"
#include "coll_libnbc.h"
#include "nbc_internal.h"
#include "mpi.h"
#include "ompi/mca/coll/coll.h"
@ -207,6 +211,24 @@ libnbc_comm_query(struct ompi_communicator_t *comm,
module->super.coll_iscan = NULL;
module->super.coll_iscatter = ompi_coll_libnbc_iscatter_inter;
module->super.coll_iscatterv = ompi_coll_libnbc_iscatterv_inter;
module->super.coll_allgather_init = ompi_coll_libnbc_allgather_inter_init;
module->super.coll_allgatherv_init = ompi_coll_libnbc_allgatherv_inter_init;
module->super.coll_allreduce_init = ompi_coll_libnbc_allreduce_inter_init;
module->super.coll_alltoall_init = ompi_coll_libnbc_alltoall_inter_init;
module->super.coll_alltoallv_init = ompi_coll_libnbc_alltoallv_inter_init;
module->super.coll_alltoallw_init = ompi_coll_libnbc_alltoallw_inter_init;
module->super.coll_barrier_init = ompi_coll_libnbc_barrier_inter_init;
module->super.coll_bcast_init = ompi_coll_libnbc_bcast_inter_init;
module->super.coll_exscan_init = NULL;
module->super.coll_gather_init = ompi_coll_libnbc_gather_inter_init;
module->super.coll_gatherv_init = ompi_coll_libnbc_gatherv_inter_init;
module->super.coll_reduce_init = ompi_coll_libnbc_reduce_inter_init;
module->super.coll_reduce_scatter_init = ompi_coll_libnbc_reduce_scatter_inter_init;
module->super.coll_reduce_scatter_block_init = ompi_coll_libnbc_reduce_scatter_block_inter_init;
module->super.coll_scan_init = NULL;
module->super.coll_scatter_init = ompi_coll_libnbc_scatter_inter_init;
module->super.coll_scatterv_init = ompi_coll_libnbc_scatterv_inter_init;
} else {
module->super.coll_iallgather = ompi_coll_libnbc_iallgather;
module->super.coll_iallgatherv = ompi_coll_libnbc_iallgatherv;
@ -231,6 +253,30 @@ libnbc_comm_query(struct ompi_communicator_t *comm,
module->super.coll_ineighbor_alltoall = ompi_coll_libnbc_ineighbor_alltoall;
module->super.coll_ineighbor_alltoallv = ompi_coll_libnbc_ineighbor_alltoallv;
module->super.coll_ineighbor_alltoallw = ompi_coll_libnbc_ineighbor_alltoallw;
module->super.coll_allgather_init = ompi_coll_libnbc_allgather_init;
module->super.coll_allgatherv_init = ompi_coll_libnbc_allgatherv_init;
module->super.coll_allreduce_init = ompi_coll_libnbc_allreduce_init;
module->super.coll_alltoall_init = ompi_coll_libnbc_alltoall_init;
module->super.coll_alltoallv_init = ompi_coll_libnbc_alltoallv_init;
module->super.coll_alltoallw_init = ompi_coll_libnbc_alltoallw_init;
module->super.coll_barrier_init = ompi_coll_libnbc_barrier_init;
module->super.coll_bcast_init = ompi_coll_libnbc_bcast_init;
module->super.coll_exscan_init = ompi_coll_libnbc_exscan_init;
module->super.coll_gather_init = ompi_coll_libnbc_gather_init;
module->super.coll_gatherv_init = ompi_coll_libnbc_gatherv_init;
module->super.coll_reduce_init = ompi_coll_libnbc_reduce_init;
module->super.coll_reduce_scatter_init = ompi_coll_libnbc_reduce_scatter_init;
module->super.coll_reduce_scatter_block_init = ompi_coll_libnbc_reduce_scatter_block_init;
module->super.coll_scan_init = ompi_coll_libnbc_scan_init;
module->super.coll_scatter_init = ompi_coll_libnbc_scatter_init;
module->super.coll_scatterv_init = ompi_coll_libnbc_scatterv_init;
module->super.coll_neighbor_allgather_init = ompi_coll_libnbc_neighbor_allgather_init;
module->super.coll_neighbor_allgatherv_init = ompi_coll_libnbc_neighbor_allgatherv_init;
module->super.coll_neighbor_alltoall_init = ompi_coll_libnbc_neighbor_alltoall_init;
module->super.coll_neighbor_alltoallv_init = ompi_coll_libnbc_neighbor_alltoallv_init;
module->super.coll_neighbor_alltoallw_init = ompi_coll_libnbc_neighbor_alltoallw_init;
}
module->super.ft_event = NULL;
@ -291,7 +337,13 @@ ompi_coll_libnbc_progress(void)
else {
request->super.req_status.MPI_ERROR = res;
}
ompi_request_complete(&request->super, true);
if(request->super.req_persistent) {
/* reset for the next communication */
request->row_offset = 0;
}
if(!request->super.req_persistent || !REQUEST_COMPLETE(&request->super)) {
ompi_request_complete(&request->super, true);
}
}
OPAL_THREAD_LOCK(&mca_coll_libnbc_component.lock);
}
@ -333,6 +385,44 @@ OBJ_CLASS_INSTANCE(ompi_coll_libnbc_module_t,
libnbc_module_destruct);
static int
request_start(size_t count, ompi_request_t ** requests)
{
int i, res;
NBC_DEBUG(5, " ** request_start **\n");
for (i = 0; i < count; i++) {
NBC_Handle *handle = (NBC_Handle *) requests[i];
NBC_Schedule *schedule = handle->schedule;
NBC_DEBUG(5, "--------------------------------\n");
NBC_DEBUG(5, "schedule %p size %u\n", &schedule, sizeof(schedule));
NBC_DEBUG(5, "handle %p size %u\n", &handle, sizeof(handle));
NBC_DEBUG(5, "data %p size %u\n", &schedule->data, sizeof(schedule->data));
NBC_DEBUG(5, "req_array %p size %u\n", &handle->req_array, sizeof(handle->req_array));
NBC_DEBUG(5, "row_offset=%u address=%p size=%u\n", handle->row_offset, &handle->row_offset, sizeof(handle->row_offset));
NBC_DEBUG(5, "req_count=%u address=%p size=%u\n", handle->req_count, &handle->req_count, sizeof(handle->req_count));
NBC_DEBUG(5, "tmpbuf address=%p size=%u\n", handle->tmpbuf, sizeof(handle->tmpbuf));
NBC_DEBUG(5, "--------------------------------\n");
handle->super.req_complete = REQUEST_PENDING;
handle->nbc_complete = false;
res = NBC_Start(handle);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_DEBUG(5, " ** bad result from NBC_Start **\n");
return res;
}
}
NBC_DEBUG(5, " ** LEAVING request_start **\n");
return OMPI_SUCCESS;
}
static int
request_cancel(struct ompi_request_t *request, int complete)
{
@ -351,7 +441,6 @@ request_free(struct ompi_request_t **ompi_req)
}
OMPI_COLL_LIBNBC_REQUEST_RETURN(request);
*ompi_req = MPI_REQUEST_NULL;
return OMPI_SUCCESS;
@ -363,6 +452,7 @@ request_construct(ompi_coll_libnbc_request_t *request)
{
request->super.req_type = OMPI_REQUEST_COLL;
request->super.req_status._cancelled = 0;
request->super.req_start = request_start;
request->super.req_free = request_free;
request->super.req_cancel = request_cancel;
}

165
ompi/mca/coll/libnbc/nbc.c
Просмотреть файл

@ -17,7 +17,12 @@
*
* Copyright (c) 2012 Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2016 IBM Corporation. All rights reserved.
* Copyright (c) 2017 Ian Bradley Morgan and Anthony Skjellum. All
* rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*/
#include "nbc_internal.h"
#include "ompi/mca/coll/base/coll_tags.h"
@ -316,8 +321,7 @@ int NBC_Progress(NBC_Handle *handle) {
int i;
ompi_status_public_t status;
/* the handle is done if there is no schedule attached */
if (NULL == handle->schedule) {
if (handle->nbc_complete) {
return NBC_OK;
}
@ -383,7 +387,10 @@ int NBC_Progress(NBC_Handle *handle) {
/* this was the last round - we're done */
NBC_DEBUG(5, "NBC_Progress last round finished - we're done\n");
NBC_Free(handle);
handle->nbc_complete = true;
if (!handle->super.req_persistent) {
NBC_Free(handle);
}
return NBC_OK;
}
@ -581,60 +588,6 @@ static inline int NBC_Start_round(NBC_Handle *handle) {
return OMPI_SUCCESS;
}
int NBC_Init_handle(struct ompi_communicator_t *comm, ompi_coll_libnbc_request_t **request, ompi_coll_libnbc_module_t *comminfo)
{
int tmp_tag;
bool need_register = false;
ompi_coll_libnbc_request_t *handle;
OMPI_COLL_LIBNBC_REQUEST_ALLOC(comm, handle);
if (NULL == handle) return OMPI_ERR_OUT_OF_RESOURCE;
*request = handle;
handle->tmpbuf = NULL;
handle->req_count = 0;
handle->req_array = NULL;
handle->comm = comm;
handle->schedule = NULL;
handle->row_offset = 0;
/******************** Do the tag and shadow comm administration ... ***************/
OPAL_THREAD_LOCK(&comminfo->mutex);
tmp_tag = comminfo->tag--;
if (tmp_tag == MCA_COLL_BASE_TAG_NONBLOCKING_END) {
tmp_tag = comminfo->tag = MCA_COLL_BASE_TAG_NONBLOCKING_BASE;
NBC_DEBUG(2,"resetting tags ...\n");
}
if (true != comminfo->comm_registered) {
comminfo->comm_registered = true;
need_register = true;
}
OPAL_THREAD_UNLOCK(&comminfo->mutex);
handle->tag = tmp_tag;
/* register progress */
if (need_register) {
int32_t tmp =
OPAL_THREAD_ADD_FETCH32(&mca_coll_libnbc_component.active_comms, 1);
if (tmp == 1) {
opal_progress_register(ompi_coll_libnbc_progress);
}
}
handle->comm=comm;
/*printf("got comminfo: %lu tag: %i\n", comminfo, comminfo->tag);*/
/******************** end of tag and shadow comm administration ... ***************/
handle->comminfo = comminfo;
NBC_DEBUG(3, "got tag %i\n", handle->tag);
return OMPI_SUCCESS;
}
void NBC_Return_handle(ompi_coll_libnbc_request_t *request) {
NBC_Free (request);
OMPI_COLL_LIBNBC_REQUEST_RETURN(request);
@ -692,16 +645,21 @@ int NBC_Init_comm(MPI_Comm comm, NBC_Comminfo *comminfo) {
return OMPI_SUCCESS;
}
int NBC_Start(NBC_Handle *handle, NBC_Schedule *schedule) {
int NBC_Start(NBC_Handle *handle) {
int res;
handle->schedule = schedule;
/* bozo case */
if ((ompi_request_t *)handle == &ompi_request_empty) {
return OMPI_SUCCESS;
}
/* kick off first round */
handle->super.req_state = OMPI_REQUEST_ACTIVE;
res = NBC_Start_round(handle);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
OPAL_THREAD_LOCK(&mca_coll_libnbc_component.lock);
opal_list_append(&mca_coll_libnbc_component.active_requests, &(handle->super.super.super));
OPAL_THREAD_UNLOCK(&mca_coll_libnbc_component.lock);
@ -709,22 +667,85 @@ int NBC_Start(NBC_Handle *handle, NBC_Schedule *schedule) {
return OMPI_SUCCESS;
}
int NBC_Schedule_request(NBC_Schedule *schedule, ompi_communicator_t *comm, ompi_coll_libnbc_module_t *module, ompi_request_t **request, void *tmpbuf) {
int res;
NBC_Handle *handle;
res = NBC_Init_handle (comm, &handle, module);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
int NBC_Schedule_request(NBC_Schedule *schedule, ompi_communicator_t *comm,
ompi_coll_libnbc_module_t *module, bool persistent,
ompi_request_t **request, void *tmpbuf) {
int ret, tmp_tag;
bool need_register = false;
ompi_coll_libnbc_request_t *handle;
/* no operation (e.g. one process barrier)? */
if (((int *)schedule->data)[0] == 0 && schedule->data[sizeof(int)] == 0) {
ret = nbc_get_noop_request(persistent, request);
if (OMPI_SUCCESS != ret) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
/* update the module->tag here because other processes may have operations
* and they may update the module->tag */
OPAL_THREAD_LOCK(&module->mutex);
tmp_tag = module->tag--;
if (tmp_tag == MCA_COLL_BASE_TAG_NONBLOCKING_END) {
tmp_tag = module->tag = MCA_COLL_BASE_TAG_NONBLOCKING_BASE;
NBC_DEBUG(2,"resetting tags ...\n");
}
OPAL_THREAD_UNLOCK(&module->mutex);
OBJ_RELEASE(schedule);
free(tmpbuf);
return OMPI_SUCCESS;
}
OMPI_COLL_LIBNBC_REQUEST_ALLOC(comm, persistent, handle);
if (NULL == handle) return OMPI_ERR_OUT_OF_RESOURCE;
handle->tmpbuf = NULL;
handle->req_count = 0;
handle->req_array = NULL;
handle->comm = comm;
handle->schedule = NULL;
handle->row_offset = 0;
handle->nbc_complete = persistent ? true : false;
/******************** Do the tag and shadow comm administration ... ***************/
OPAL_THREAD_LOCK(&module->mutex);
tmp_tag = module->tag--;
if (tmp_tag == MCA_COLL_BASE_TAG_NONBLOCKING_END) {
tmp_tag = module->tag = MCA_COLL_BASE_TAG_NONBLOCKING_BASE;
NBC_DEBUG(2,"resetting tags ...\n");
}
if (true != module->comm_registered) {
module->comm_registered = true;
need_register = true;
}
OPAL_THREAD_UNLOCK(&module->mutex);
handle->tag = tmp_tag;
/* register progress */
if (need_register) {
int32_t tmp =
OPAL_THREAD_ADD_FETCH32(&mca_coll_libnbc_component.active_comms, 1);
if (tmp == 1) {
opal_progress_register(ompi_coll_libnbc_progress);
}
}
handle->comm=comm;
/*printf("got module: %lu tag: %i\n", module, module->tag);*/
/******************** end of tag and shadow comm administration ... ***************/
handle->comminfo = module;
NBC_DEBUG(3, "got tag %i\n", handle->tag);
handle->tmpbuf = tmpbuf;
res = NBC_Start (handle, schedule);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle (handle);
return res;
}
handle->schedule = schedule;
*request = (ompi_request_t *) handle;
return OMPI_SUCCESS;
}

98
ompi/mca/coll/libnbc/nbc_iallgather.c
Просмотреть файл

@ -10,6 +10,7 @@
* Copyright (c) 2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -43,9 +44,9 @@ int NBC_Allgather_args_compare(NBC_Allgather_args *a, NBC_Allgather_args *b, voi
* the algorithm uses p-1 rounds
* each node sends the packet it received last round (or has in round 0) to it's right neighbor (modulo p)
* each node receives from it's left (modulo p) neighbor */
int ompi_coll_libnbc_iallgather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module)
static int nbc_allgather_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent)
{
int rank, p, res;
MPI_Aint rcvext;
@ -69,7 +70,7 @@ int ompi_coll_libnbc_iallgather(const void* sendbuf, int sendcount, MPI_Datatype
if (inplace) {
sendtype = recvtype;
sendcount = recvcount;
} else {
} else if (!persistent) { /* for persistent, the copy must be scheduled */
/* copy my data to receive buffer */
rbuf = (char *) recvbuf + rank * recvcount * rcvext;
res = NBC_Copy (sendbuf, sendcount, sendtype, rbuf, recvcount, recvtype, comm);
@ -77,9 +78,8 @@ int ompi_coll_libnbc_iallgather(const void* sendbuf, int sendcount, MPI_Datatype
return res;
}
}
if (1 == p) {
*request = &ompi_request_empty;
return OMPI_SUCCESS;
if (1 == p && (!persistent || inplace)) {
return nbc_get_noop_request(persistent, request);
}
#ifdef NBC_CACHE_SCHEDULE
@ -99,6 +99,17 @@ int ompi_coll_libnbc_iallgather(const void* sendbuf, int sendcount, MPI_Datatype
}
sbuf = (char *)recvbuf + rank * recvcount * rcvext;
if (persistent && !inplace) { /* for nonblocking, data has been copied already */
/* copy my data to receive buffer (= send buffer of NBC_Sched_send) */
res = NBC_Sched_copy ((void *)sendbuf, false, sendcount, sendtype,
sbuf, false, recvcount, recvtype, schedule, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
}
}
/* do p-1 rounds */
for(int r = 0 ; r < p ; ++r) {
if(r != rank) {
@ -154,7 +165,7 @@ int ompi_coll_libnbc_iallgather(const void* sendbuf, int sendcount, MPI_Datatype
}
#endif
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -163,9 +174,29 @@ int ompi_coll_libnbc_iallgather(const void* sendbuf, int sendcount, MPI_Datatype
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_iallgather_inter(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module)
int ompi_coll_libnbc_iallgather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module)
{
int res = nbc_allgather_init(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle (*(ompi_coll_libnbc_request_t **)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
static int nbc_allgather_inter_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent)
{
int res, rsize;
MPI_Aint rcvext;
@ -211,7 +242,7 @@ int ompi_coll_libnbc_iallgather_inter(const void* sendbuf, int sendcount, MPI_Da
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -219,3 +250,46 @@ int ompi_coll_libnbc_iallgather_inter(const void* sendbuf, int sendcount, MPI_Da
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_iallgather_inter(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_allgather_inter_init(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle (*(ompi_coll_libnbc_request_t **)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_allgather_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_allgather_init(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_allgather_inter_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_allgather_inter_init(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

91
ompi/mca/coll/libnbc/nbc_iallgatherv.c
Просмотреть файл

@ -14,6 +14,7 @@
* Copyright (c) 2014-2017 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -33,9 +34,9 @@
* second round:
* each node sends to node (rank+2)%p sendcount elements
* each node receives from node (rank-2)%p recvcounts[(rank+2)%p] elements */
int ompi_coll_libnbc_iallgatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *displs,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module)
static int nbc_allgatherv_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *displs,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent)
{
int rank, p, res, speer, rpeer;
MPI_Aint rcvext;
@ -57,7 +58,7 @@ int ompi_coll_libnbc_iallgatherv(const void* sendbuf, int sendcount, MPI_Datatyp
if (inplace) {
sendtype = recvtype;
sendcount = recvcounts[rank];
} else {
} else if (!persistent) { /* for persistent, the copy must be scheduled */
/* copy my data to receive buffer */
rbuf = (char *) recvbuf + displs[rank] * rcvext;
res = NBC_Copy (sendbuf, sendcount, sendtype, rbuf, recvcounts[rank], recvtype, comm);
@ -73,6 +74,16 @@ int ompi_coll_libnbc_iallgatherv(const void* sendbuf, int sendcount, MPI_Datatyp
sbuf = (char *) recvbuf + displs[rank] * rcvext;
if (persistent && !inplace) { /* for nonblocking, data has been copied already */
/* copy my data to receive buffer (= send buffer of NBC_Sched_send) */
res = NBC_Sched_copy ((void *)sendbuf, false, sendcount, sendtype,
sbuf, false, recvcounts[rank], recvtype, schedule, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
}
}
/* do p-1 rounds */
for (int r = 1 ; r < p ; ++r) {
speer = (rank + r) % p;
@ -99,7 +110,7 @@ int ompi_coll_libnbc_iallgatherv(const void* sendbuf, int sendcount, MPI_Datatyp
return res;
}
res = NBC_Schedule_request (schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request (schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -108,9 +119,28 @@ int ompi_coll_libnbc_iallgatherv(const void* sendbuf, int sendcount, MPI_Datatyp
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_iallgatherv_inter(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *displs,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module)
int ompi_coll_libnbc_iallgatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *displs,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_allgatherv_init(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
static int nbc_allgatherv_inter_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *displs,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent)
{
int res, rsize;
MPI_Aint rcvext;
@ -159,7 +189,7 @@ int ompi_coll_libnbc_iallgatherv_inter(const void* sendbuf, int sendcount, MPI_D
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -167,3 +197,46 @@ int ompi_coll_libnbc_iallgatherv_inter(const void* sendbuf, int sendcount, MPI_D
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_iallgatherv_inter(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *displs,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_allgatherv_inter_init(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_allgatherv_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *displs,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_allgatherv_init(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_allgatherv_inter_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *displs,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_allgatherv_inter_init(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

107
ompi/mca/coll/libnbc/nbc_iallreduce.c
Просмотреть файл

@ -10,6 +10,7 @@
* Copyright (c) 2014-2017 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -52,9 +53,9 @@ int NBC_Allreduce_args_compare(NBC_Allreduce_args *a, NBC_Allreduce_args *b, voi
}
#endif
int ompi_coll_libnbc_iallreduce(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module)
static int nbc_allreduce_init(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent)
{
int rank, p, res;
ptrdiff_t ext, lb;
@ -86,7 +87,7 @@ int ompi_coll_libnbc_iallreduce(const void* sendbuf, void* recvbuf, int count, M
return res;
}
if (1 == p) {
if (1 == p && (!persistent || inplace)) {
if (!inplace) {
/* for a single node - copy data to receivebuf */
res = NBC_Copy(sendbuf, count, datatype, recvbuf, count, datatype, comm);
@ -94,8 +95,7 @@ int ompi_coll_libnbc_iallreduce(const void* sendbuf, void* recvbuf, int count, M
return res;
}
}
*request = &ompi_request_empty;
return OMPI_SUCCESS;
return nbc_get_noop_request(persistent, request);
}
span = opal_datatype_span(&datatype->super, count, &gap);
@ -127,13 +127,18 @@ int ompi_coll_libnbc_iallreduce(const void* sendbuf, void* recvbuf, int count, M
return OMPI_ERR_OUT_OF_RESOURCE;
}
switch(alg) {
case NBC_ARED_BINOMIAL:
res = allred_sched_diss(rank, p, count, datatype, gap, sendbuf, recvbuf, op, inplace, schedule, tmpbuf);
break;
case NBC_ARED_RING:
res = allred_sched_ring(rank, p, count, datatype, sendbuf, recvbuf, op, size, ext, schedule, tmpbuf);
break;
if (p == 1) {
res = NBC_Sched_copy((void *)sendbuf, false, count, datatype,
recvbuf, false, count, datatype, schedule, false);
} else {
switch(alg) {
case NBC_ARED_BINOMIAL:
res = allred_sched_diss(rank, p, count, datatype, gap, sendbuf, recvbuf, op, inplace, schedule, tmpbuf);
break;
case NBC_ARED_RING:
res = allred_sched_ring(rank, p, count, datatype, sendbuf, recvbuf, op, size, ext, schedule, tmpbuf);
break;
}
}
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
@ -180,7 +185,7 @@ int ompi_coll_libnbc_iallreduce(const void* sendbuf, void* recvbuf, int count, M
}
#endif
res = NBC_Schedule_request (schedule, comm, libnbc_module, request, tmpbuf);
res = NBC_Schedule_request (schedule, comm, libnbc_module, persistent, request, tmpbuf);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
free(tmpbuf);
@ -190,9 +195,28 @@ int ompi_coll_libnbc_iallreduce(const void* sendbuf, void* recvbuf, int count, M
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_iallreduce_inter(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module)
int ompi_coll_libnbc_iallreduce(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_allreduce_init(sendbuf, recvbuf, count, datatype, op,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
static int nbc_allreduce_inter_init(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent)
{
int rank, res, rsize;
size_t size;
@ -244,7 +268,7 @@ int ompi_coll_libnbc_iallreduce_inter(const void* sendbuf, void* recvbuf, int co
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, tmpbuf);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
free(tmpbuf);
@ -254,6 +278,24 @@ int ompi_coll_libnbc_iallreduce_inter(const void* sendbuf, void* recvbuf, int co
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_iallreduce_inter(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_allreduce_inter_init(sendbuf, recvbuf, count, datatype, op,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
/* binomial allreduce (binomial tree up and binomial bcast down)
* working principle:
@ -311,7 +353,9 @@ static inline int allred_sched_diss(int rank, int p, int count, MPI_Datatype dat
rbuf = recvbuf;
tmprbuf = false;
if (inplace) {
res = NBC_Copy(rbuf, count, datatype, ((char *)tmpbuf) - gap, count, datatype, MPI_COMM_SELF);
res = NBC_Sched_copy(rbuf, false, count, datatype,
((char *)tmpbuf) - gap, false, count, datatype,
schedule, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
@ -690,3 +734,28 @@ static inline int allred_sched_linear(int rank, int rsize, const void *sendbuf,
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_allreduce_init(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_allreduce_init(sendbuf, recvbuf, count, datatype, op,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_allreduce_inter_init(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_allreduce_inter_init(sendbuf, recvbuf, count, datatype, op,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

107
ompi/mca/coll/libnbc/nbc_ialltoall.c
Просмотреть файл

@ -11,6 +11,7 @@
* Copyright (c) 2014-2017 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -53,9 +54,9 @@ int NBC_Alltoall_args_compare(NBC_Alltoall_args *a, NBC_Alltoall_args *b, void *
#endif
/* simple linear MPI_Ialltoall the (simple) algorithm just sends to all nodes */
int ompi_coll_libnbc_ialltoall(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module)
static int nbc_alltoall_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent)
{
int rank, p, res, datasize;
size_t a2asize, sndsize;
@ -109,16 +110,6 @@ int ompi_coll_libnbc_ialltoall(const void* sendbuf, int sendcount, MPI_Datatype
} else
alg = NBC_A2A_LINEAR; /*NBC_A2A_PAIRWISE;*/
if (!inplace) {
/* copy my data to receive buffer */
rbuf = (char *) recvbuf + rank * recvcount * rcvext;
sbuf = (char *) sendbuf + rank * sendcount * sndext;
res = NBC_Copy (sbuf, sendcount, sendtype, rbuf, recvcount, recvtype, comm);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
}
/* allocate temp buffer if we need one */
if (alg == NBC_A2A_INPLACE) {
span = opal_datatype_span(&recvtype->super, recvcount, &gap);
@ -127,7 +118,10 @@ int ompi_coll_libnbc_ialltoall(const void* sendbuf, int sendcount, MPI_Datatype
return OMPI_ERR_OUT_OF_RESOURCE;
}
} else if (alg == NBC_A2A_DISS) {
/* only A2A_DISS needs buffers */
/* persistent operation is not supported currently for this algorithm;
* we need to replace PMPI_Pack, PMPI_Unpack, and mempcy */
assert(! persistent);
if(NBC_Type_intrinsic(sendtype)) {
datasize = sndext * sendcount;
} else {
@ -204,6 +198,19 @@ int ompi_coll_libnbc_ialltoall(const void* sendbuf, int sendcount, MPI_Datatype
return OMPI_ERR_OUT_OF_RESOURCE;
}
if (!inplace) {
/* copy my data to receive buffer */
rbuf = (char *) recvbuf + rank * recvcount * rcvext;
sbuf = (char *) sendbuf + rank * sendcount * sndext;
res = NBC_Sched_copy (sbuf, false, sendcount, sendtype,
rbuf, false, recvcount, recvtype, schedule, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
free(tmpbuf);
return res;
}
}
switch(alg) {
case NBC_A2A_INPLACE:
res = a2a_sched_inplace(rank, p, schedule, recvbuf, recvcount, recvtype, rcvext, gap, comm);
@ -264,7 +271,7 @@ int ompi_coll_libnbc_ialltoall(const void* sendbuf, int sendcount, MPI_Datatype
}
#endif
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, tmpbuf);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
free(tmpbuf);
@ -274,9 +281,28 @@ int ompi_coll_libnbc_ialltoall(const void* sendbuf, int sendcount, MPI_Datatype
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ialltoall_inter (const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module)
int ompi_coll_libnbc_ialltoall(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_alltoall_init(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
static int nbc_alltoall_inter_init (const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent)
{
int res, rsize;
MPI_Aint sndext, rcvext;
@ -330,7 +356,7 @@ int ompi_coll_libnbc_ialltoall_inter (const void* sendbuf, int sendcount, MPI_Da
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -339,6 +365,25 @@ int ompi_coll_libnbc_ialltoall_inter (const void* sendbuf, int sendcount, MPI_Da
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ialltoall_inter (const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_alltoall_inter_init(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
static inline int a2a_sched_pairwise(int rank, int p, MPI_Aint sndext, MPI_Aint rcvext, NBC_Schedule* schedule,
const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, MPI_Comm comm) {
@ -544,3 +589,27 @@ static inline int a2a_sched_inplace(int rank, int p, NBC_Schedule* schedule, voi
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_alltoall_init (const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_alltoall_init(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_alltoall_inter_init (const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_alltoall_inter_init(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

111
ompi/mca/coll/libnbc/nbc_ialltoallv.c
Просмотреть файл

@ -10,6 +10,7 @@
* Copyright (c) 2015-2017 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -40,10 +41,10 @@ static inline int a2av_sched_inplace(int rank, int p, NBC_Schedule *schedule,
* would not be sufficient ... we simply do not cache it */
/* simple linear Alltoallv */
int ompi_coll_libnbc_ialltoallv(const void* sendbuf, const int *sendcounts, const int *sdispls,
MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *rdispls,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module)
static int nbc_alltoallv_init(const void* sendbuf, const int *sendcounts, const int *sdispls,
MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *rdispls,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent)
{
int rank, p, res;
MPI_Aint sndext, rcvext;
@ -74,8 +75,7 @@ int ompi_coll_libnbc_ialltoallv(const void* sendbuf, const int *sendcounts, cons
}
span = opal_datatype_span(&recvtype->super, count, &gap);
if (OPAL_UNLIKELY(0 == span)) {
*request = &ompi_request_empty;
return MPI_SUCCESS;
return nbc_get_noop_request(persistent, request);
}
tmpbuf = malloc(span);
if (OPAL_UNLIKELY(NULL == tmpbuf)) {
@ -89,14 +89,6 @@ int ompi_coll_libnbc_ialltoallv(const void* sendbuf, const int *sendcounts, cons
NBC_Error("MPI Error in ompi_datatype_type_extent() (%i)", res);
return res;
}
if (sendcounts[rank] != 0) {
rbuf = (char *) recvbuf + rdispls[rank] * rcvext;
sbuf = (char *) sendbuf + sdispls[rank] * sndext;
res = NBC_Copy (sbuf, sendcounts[rank], sendtype, rbuf, recvcounts[rank], recvtype, comm);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
}
}
schedule = OBJ_NEW(NBC_Schedule);
@ -106,6 +98,17 @@ int ompi_coll_libnbc_ialltoallv(const void* sendbuf, const int *sendcounts, cons
}
if (!inplace && sendcounts[rank] != 0) {
rbuf = (char *) recvbuf + rdispls[rank] * rcvext;
sbuf = (char *) sendbuf + sdispls[rank] * sndext;
res = NBC_Sched_copy (sbuf, false, sendcounts[rank], sendtype,
rbuf, false, recvcounts[rank], recvtype, schedule, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
}
}
if (inplace) {
res = a2av_sched_inplace(rank, p, schedule, recvbuf, recvcounts,
rdispls, rcvext, recvtype, gap);
@ -127,7 +130,7 @@ int ompi_coll_libnbc_ialltoallv(const void* sendbuf, const int *sendcounts, cons
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, tmpbuf);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
free(tmpbuf);
@ -137,11 +140,32 @@ int ompi_coll_libnbc_ialltoallv(const void* sendbuf, const int *sendcounts, cons
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ialltoallv(const void* sendbuf, const int *sendcounts, const int *sdispls,
MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *rdispls,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_alltoallv_init(sendbuf, sendcounts, sdispls, sendtype,
recvbuf, recvcounts, rdispls, recvtype,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
/* simple linear Alltoallv */
int ompi_coll_libnbc_ialltoallv_inter (const void* sendbuf, const int *sendcounts, const int *sdispls,
MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *rdispls,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module)
static int nbc_alltoallv_inter_init (const void* sendbuf, const int *sendcounts, const int *sdispls,
MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *rdispls,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent)
{
int res, rsize;
MPI_Aint sndext, rcvext;
@ -195,7 +219,7 @@ int ompi_coll_libnbc_ialltoallv_inter (const void* sendbuf, const int *sendcount
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -204,6 +228,27 @@ int ompi_coll_libnbc_ialltoallv_inter (const void* sendbuf, const int *sendcount
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ialltoallv_inter (const void* sendbuf, const int *sendcounts, const int *sdispls,
MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *rdispls,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_alltoallv_inter_init(sendbuf, sendcounts, sdispls, sendtype,
recvbuf, recvcounts, rdispls, recvtype,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
__opal_attribute_unused__
static inline int a2av_sched_linear(int rank, int p, NBC_Schedule *schedule,
const void *sendbuf, const int *sendcounts, const int *sdispls,
@ -342,3 +387,29 @@ static inline int a2av_sched_inplace(int rank, int p, NBC_Schedule *schedule,
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_alltoallv_init(const void* sendbuf, const int *sendcounts, const int *sdispls,
MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *rdispls,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_alltoallv_init(sendbuf, sendcounts, sdispls, sendtype, recvbuf, recvcounts, rdispls, recvtype,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_alltoallv_inter_init(const void* sendbuf, const int *sendcounts, const int *sdispls,
MPI_Datatype sendtype, void* recvbuf, const int *recvcounts, const int *rdispls,
MPI_Datatype recvtype, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_alltoallv_inter_init(sendbuf, sendcounts, sdispls, sendtype, recvbuf, recvcounts, rdispls, recvtype,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

109
ompi/mca/coll/libnbc/nbc_ialltoallw.c
Просмотреть файл

@ -10,6 +10,7 @@
* Copyright (c) 2015-2017 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -40,10 +41,10 @@ static inline int a2aw_sched_inplace(int rank, int p, NBC_Schedule *schedule,
* would not be sufficient ... we simply do not cache it */
/* simple linear Alltoallw */
int ompi_coll_libnbc_ialltoallw(const void* sendbuf, const int *sendcounts, const int *sdispls,
struct ompi_datatype_t * const *sendtypes, void* recvbuf, const int *recvcounts, const int *rdispls,
struct ompi_datatype_t * const *recvtypes, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module)
static int nbc_alltoallw_init(const void* sendbuf, const int *sendcounts, const int *sdispls,
struct ompi_datatype_t * const *sendtypes, void* recvbuf, const int *recvcounts, const int *rdispls,
struct ompi_datatype_t * const *recvtypes, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent)
{
int rank, p, res;
NBC_Schedule *schedule;
@ -67,8 +68,7 @@ int ompi_coll_libnbc_ialltoallw(const void* sendbuf, const int *sendcounts, cons
}
}
if (OPAL_UNLIKELY(0 == span)) {
*request = &ompi_request_empty;
return OMPI_SUCCESS;
return nbc_get_noop_request(persistent, request);
}
tmpbuf = malloc(span);
if (OPAL_UNLIKELY(NULL == tmpbuf)) {
@ -77,13 +77,6 @@ int ompi_coll_libnbc_ialltoallw(const void* sendbuf, const int *sendcounts, cons
sendcounts = recvcounts;
sdispls = rdispls;
sendtypes = recvtypes;
} else if (sendcounts[rank] != 0) {
rbuf = (char *) recvbuf + rdispls[rank];
sbuf = (char *) sendbuf + sdispls[rank];
res = NBC_Copy(sbuf, sendcounts[rank], sendtypes[rank], rbuf, recvcounts[rank], recvtypes[rank], comm);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
}
schedule = OBJ_NEW(NBC_Schedule);
@ -92,6 +85,16 @@ int ompi_coll_libnbc_ialltoallw(const void* sendbuf, const int *sendcounts, cons
return OMPI_ERR_OUT_OF_RESOURCE;
}
if (!inplace && sendcounts[rank] != 0) {
rbuf = (char *) recvbuf + rdispls[rank];
sbuf = (char *) sendbuf + sdispls[rank];
res = NBC_Sched_copy(sbuf, false, sendcounts[rank], sendtypes[rank],
rbuf, false, recvcounts[rank], recvtypes[rank], schedule, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
}
if (inplace) {
res = a2aw_sched_inplace(rank, p, schedule, recvbuf,
recvcounts, rdispls, recvtypes);
@ -113,7 +116,7 @@ int ompi_coll_libnbc_ialltoallw(const void* sendbuf, const int *sendcounts, cons
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, tmpbuf);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
free(tmpbuf);
@ -123,11 +126,32 @@ int ompi_coll_libnbc_ialltoallw(const void* sendbuf, const int *sendcounts, cons
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ialltoallw(const void* sendbuf, const int *sendcounts, const int *sdispls,
struct ompi_datatype_t * const *sendtypes, void* recvbuf, const int *recvcounts, const int *rdispls,
struct ompi_datatype_t * const *recvtypes, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_alltoallw_init(sendbuf, sendcounts, sdispls, sendtypes,
recvbuf, recvcounts, rdispls, recvtypes,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
/* simple linear Alltoallw */
int ompi_coll_libnbc_ialltoallw_inter (const void* sendbuf, const int *sendcounts, const int *sdispls,
struct ompi_datatype_t * const *sendtypes, void* recvbuf, const int *recvcounts, const int *rdispls,
struct ompi_datatype_t * const *recvtypes, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module)
static int nbc_alltoallw_inter_init (const void* sendbuf, const int *sendcounts, const int *sdispls,
struct ompi_datatype_t * const *sendtypes, void* recvbuf, const int *recvcounts, const int *rdispls,
struct ompi_datatype_t * const *recvtypes, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent)
{
int res, rsize;
NBC_Schedule *schedule;
@ -168,7 +192,7 @@ int ompi_coll_libnbc_ialltoallw_inter (const void* sendbuf, const int *sendcount
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -177,6 +201,27 @@ int ompi_coll_libnbc_ialltoallw_inter (const void* sendbuf, const int *sendcount
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ialltoallw_inter(const void* sendbuf, const int *sendcounts, const int *sdispls,
struct ompi_datatype_t * const *sendtypes, void* recvbuf, const int *recvcounts, const int *rdispls,
struct ompi_datatype_t * const *recvtypes, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_alltoallw_inter_init(sendbuf, sendcounts, sdispls, sendtypes,
recvbuf, recvcounts, rdispls, recvtypes,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
static inline int a2aw_sched_linear(int rank, int p, NBC_Schedule *schedule,
const void *sendbuf, const int *sendcounts, const int *sdispls,
struct ompi_datatype_t * const * sendtypes,
@ -316,3 +361,29 @@ static inline int a2aw_sched_inplace(int rank, int p, NBC_Schedule *schedule,
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_alltoallw_init(const void* sendbuf, const int *sendcounts, const int *sdispls,
struct ompi_datatype_t * const *sendtypes, void* recvbuf, const int *recvcounts, const int *rdispls,
struct ompi_datatype_t * const *recvtypes, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_alltoallw_init(sendbuf, sendcounts, sdispls, sendtypes, recvbuf, recvcounts, rdispls, recvtypes,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_alltoallw_inter_init(const void* sendbuf, const int *sendcounts, const int *sdispls,
struct ompi_datatype_t * const *sendtypes, void* recvbuf, const int *recvcounts, const int *rdispls,
struct ompi_datatype_t * const *recvtypes, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_alltoallw_inter_init(sendbuf, sendcounts, sdispls, sendtypes, recvbuf, recvcounts, rdispls, recvtypes,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

143
ompi/mca/coll/libnbc/nbc_ibarrier.c
Просмотреть файл

@ -11,6 +11,7 @@
* and Technology (RIST). All rights reserved.
* Copyright (c) 2015 Mellanox Technologies. All rights reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -21,22 +22,16 @@
#include "nbc_internal.h"
/* Dissemination implementation of MPI_Ibarrier */
int ompi_coll_libnbc_ibarrier(struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module)
static int nbc_barrier_init(struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent)
{
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 +42,9 @@ 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) {
@ -64,86 +55,88 @@ 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, persistent, 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;
}
int ompi_coll_libnbc_ibarrier_inter(struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module)
int ompi_coll_libnbc_ibarrier(struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_barrier_init(comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
static int nbc_barrier_inter_init(struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent)
{
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 +144,81 @@ 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, persistent, 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;
}
int ompi_coll_libnbc_ibarrier_inter(struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_barrier_inter_init(comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_barrier_init(struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_barrier_init(comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_barrier_inter_init(struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_barrier_inter_init(comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

82
ompi/mca/coll/libnbc/nbc_ibcast.c
Просмотреть файл

@ -10,6 +10,7 @@
* Copyright (c) 2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2016-2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -44,9 +45,9 @@ int NBC_Bcast_args_compare(NBC_Bcast_args *a, NBC_Bcast_args *b, void *param) {
}
#endif
int ompi_coll_libnbc_ibcast(void *buffer, int count, MPI_Datatype datatype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module)
static int nbc_bcast_init(void *buffer, int count, MPI_Datatype datatype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent)
{
int rank, p, res, segsize;
size_t size;
@ -61,8 +62,7 @@ int ompi_coll_libnbc_ibcast(void *buffer, int count, MPI_Datatype datatype, int
p = ompi_comm_size (comm);
if (1 == p) {
*request = &ompi_request_empty;
return OMPI_SUCCESS;
return nbc_get_noop_request(persistent, request);
}
res = ompi_datatype_type_size(datatype, &size);
@ -162,7 +162,7 @@ int ompi_coll_libnbc_ibcast(void *buffer, int count, MPI_Datatype datatype, int
}
#endif
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -171,6 +171,25 @@ int ompi_coll_libnbc_ibcast(void *buffer, int count, MPI_Datatype datatype, int
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ibcast(void *buffer, int count, MPI_Datatype datatype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module)
{
int res = nbc_bcast_init(buffer, count, datatype, root,
comm, request, module, false);
if (OPAL_LIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
/* better binomial bcast
* working principle:
* - each node gets a virtual rank vrank
@ -323,9 +342,9 @@ static inline int bcast_sched_chain(int rank, int p, int root, NBC_Schedule *sch
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ibcast_inter(void *buffer, int count, MPI_Datatype datatype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
static int nbc_bcast_inter_init(void *buffer, int count, MPI_Datatype datatype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int res;
NBC_Schedule *schedule;
ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
@ -366,7 +385,7 @@ int ompi_coll_libnbc_ibcast_inter(void *buffer, int count, MPI_Datatype datatype
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -374,3 +393,46 @@ int ompi_coll_libnbc_ibcast_inter(void *buffer, int count, MPI_Datatype datatype
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ibcast_inter(void *buffer, int count, MPI_Datatype datatype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_bcast_inter_init(buffer, count, datatype, root,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_bcast_init(void *buffer, int count, MPI_Datatype datatype, int root,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_bcast_init(buffer, count, datatype, root,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_bcast_inter_init(void *buffer, int count, MPI_Datatype datatype, int root,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_bcast_inter_init(buffer, count, datatype, root,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

75
ompi/mca/coll/libnbc/nbc_iexscan.c
Просмотреть файл

@ -10,6 +10,7 @@
* Copyright (c) 2014-2017 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -45,9 +46,9 @@ int NBC_Scan_args_compare(NBC_Scan_args *a, NBC_Scan_args *b, void *param) {
* 3. all but rank p-1 do sends to it's right neigbor and exits
*
*/
int ompi_coll_libnbc_iexscan(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
static int nbc_exscan_init(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int rank, p, res;
ptrdiff_t gap, span;
NBC_Schedule *schedule;
@ -63,23 +64,6 @@ int ompi_coll_libnbc_iexscan(const void* sendbuf, void* recvbuf, int count, MPI_
rank = ompi_comm_rank (comm);
p = ompi_comm_size (comm);
span = opal_datatype_span(&datatype->super, count, &gap);
if (0 < rank) {
tmpbuf = malloc(span);
if (NULL == tmpbuf) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
if (inplace) {
res = NBC_Copy(recvbuf, count, datatype, (char *)tmpbuf-gap, count, datatype, comm);
} else {
res = NBC_Copy(sendbuf, count, datatype, (char *)tmpbuf-gap, count, datatype, comm);
}
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
free(tmpbuf);
return res;
}
}
#ifdef NBC_CACHE_SCHEDULE
/* search schedule in communicator specific tree */
search.sendbuf = sendbuf;
@ -97,6 +81,24 @@ int ompi_coll_libnbc_iexscan(const void* sendbuf, void* recvbuf, int count, MPI_
}
if (rank != 0) {
span = opal_datatype_span(&datatype->super, count, &gap);
tmpbuf = malloc(span);
if (NULL == tmpbuf) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
if (inplace) {
res = NBC_Sched_copy(recvbuf, false, count, datatype,
(char *)tmpbuf-gap, false, count, datatype, schedule, false);
} else {
res = NBC_Sched_copy((void *)sendbuf, false, count, datatype,
(char *)tmpbuf-gap, false, count, datatype, schedule, false);
}
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
free(tmpbuf);
return res;
}
res = NBC_Sched_recv (recvbuf, false, count, datatype, rank-1, schedule, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
@ -182,7 +184,7 @@ int ompi_coll_libnbc_iexscan(const void* sendbuf, void* recvbuf, int count, MPI_
}
#endif
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, tmpbuf);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
free(tmpbuf);
@ -191,3 +193,34 @@ int ompi_coll_libnbc_iexscan(const void* sendbuf, void* recvbuf, int count, MPI_
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_iexscan(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_exscan_init(sendbuf, recvbuf, count, datatype, op,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_exscan_init(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_exscan_init(sendbuf, recvbuf, count, datatype, op,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

106
ompi/mca/coll/libnbc/nbc_igather.c
Просмотреть файл

@ -13,6 +13,7 @@
* Copyright (c) 2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -43,10 +44,10 @@ int NBC_Gather_args_compare(NBC_Gather_args *a, NBC_Gather_args *b, void *param)
}
#endif
int ompi_coll_libnbc_igather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf,
int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
static int nbc_gather_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf,
int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int rank, p, res;
MPI_Aint rcvext = 0;
NBC_Schedule *schedule;
@ -70,13 +71,6 @@ int ompi_coll_libnbc_igather(const void* sendbuf, int sendcount, MPI_Datatype se
if (inplace) {
sendcount = recvcount;
sendtype = recvtype;
} else if (rank == root) {
rbuf = ((char *)recvbuf) + (rank*recvcount*rcvext);
/* if I am the root - just copy the message (only without MPI_IN_PLACE) */
res = NBC_Copy(sendbuf, sendcount, sendtype, rbuf, recvcount, recvtype, comm);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
}
#ifdef NBC_CACHE_SCHEDULE
@ -110,7 +104,17 @@ int ompi_coll_libnbc_igather(const void* sendbuf, int sendcount, MPI_Datatype se
} else {
for (int i = 0 ; i < p ; ++i) {
rbuf = (char *)recvbuf + i * recvcount * rcvext;
if (i != root) {
if (i == root) {
if (!inplace) {
/* if I am the root - just copy the message */
res = NBC_Sched_copy ((void *)sendbuf, false, sendcount, sendtype,
rbuf, false, recvcount, recvtype, schedule, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
}
}
} else {
/* root receives message to the right buffer */
res = NBC_Sched_recv (rbuf, false, recvcount, recvtype, i, schedule, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
@ -160,7 +164,7 @@ int ompi_coll_libnbc_igather(const void* sendbuf, int sendcount, MPI_Datatype se
}
#endif
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -169,10 +173,30 @@ int ompi_coll_libnbc_igather(const void* sendbuf, int sendcount, MPI_Datatype se
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_igather_inter (const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf,
int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
int ompi_coll_libnbc_igather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf,
int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_gather_init(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
static int nbc_gather_inter_init (const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf,
int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int res, rsize;
MPI_Aint rcvext = 0;
NBC_Schedule *schedule;
@ -220,7 +244,7 @@ int ompi_coll_libnbc_igather_inter (const void* sendbuf, int sendcount, MPI_Data
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -228,3 +252,49 @@ int ompi_coll_libnbc_igather_inter (const void* sendbuf, int sendcount, MPI_Data
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_igather_inter(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf,
int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_gather_inter_init(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_gather_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf,
int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_gather_init(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_gather_inter_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf,
int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_gather_inter_init(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

91
ompi/mca/coll/libnbc/nbc_igatherv.c
Просмотреть файл

@ -14,6 +14,7 @@
* reserved.
* Copyright (c) 2015 Mellanox Technologies. All rights reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -28,10 +29,10 @@
* would not be sufficient ... we simply do not cache it */
int ompi_coll_libnbc_igatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, const int *recvcounts, const int *displs, MPI_Datatype recvtype,
int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
static int nbc_gatherv_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, const int *recvcounts, const int *displs, MPI_Datatype recvtype,
int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int rank, p, res;
MPI_Aint rcvext = 0;
NBC_Schedule *schedule;
@ -71,8 +72,8 @@ int ompi_coll_libnbc_igatherv(const void* sendbuf, int sendcount, MPI_Datatype s
if (i == root) {
if (!inplace) {
/* if I am the root - just copy the message */
res = NBC_Copy (sendbuf, sendcount, sendtype, rbuf, recvcounts[i], recvtype,
comm);
res = NBC_Sched_copy ((void *)sendbuf, false, sendcount, sendtype,
rbuf, false, recvcounts[i], recvtype, schedule, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -95,7 +96,7 @@ int ompi_coll_libnbc_igatherv(const void* sendbuf, int sendcount, MPI_Datatype s
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -104,10 +105,30 @@ int ompi_coll_libnbc_igatherv(const void* sendbuf, int sendcount, MPI_Datatype s
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_igatherv_inter (const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, const int *recvcounts, const int *displs, MPI_Datatype recvtype,
int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
int ompi_coll_libnbc_igatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, const int *recvcounts, const int *displs, MPI_Datatype recvtype,
int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_gatherv_init(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, root,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
static int nbc_gatherv_inter_init (const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, const int *recvcounts, const int *displs, MPI_Datatype recvtype,
int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int res, rsize;
MPI_Aint rcvext;
NBC_Schedule *schedule;
@ -155,7 +176,7 @@ int ompi_coll_libnbc_igatherv_inter (const void* sendbuf, int sendcount, MPI_Dat
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -163,3 +184,49 @@ int ompi_coll_libnbc_igatherv_inter (const void* sendbuf, int sendcount, MPI_Dat
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_igatherv_inter(const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, const int *recvcounts, const int *displs, MPI_Datatype recvtype,
int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_gatherv_inter_init(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, root,
comm, request, module, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_gatherv_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, const int *recvcounts, const int *displs, MPI_Datatype recvtype,
int root, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_gatherv_init(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, root,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_gatherv_inter_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, const int *recvcounts, const int *displs, MPI_Datatype recvtype,
int root, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_gatherv_inter_init(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, root,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

192
ompi/mca/coll/libnbc/nbc_ineighbor_allgather.c
Просмотреть файл

@ -10,6 +10,7 @@
* Copyright (c) 2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -43,9 +44,10 @@ int NBC_Ineighbor_allgather_args_compare(NBC_Ineighbor_allgather_args *a, NBC_In
#endif
int ompi_coll_libnbc_ineighbor_allgather(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
int rcount, MPI_Datatype rtype, struct ompi_communicator_t *comm,
ompi_request_t ** request, struct mca_coll_base_module_2_2_0_t *module) {
static int nbc_neighbor_allgather_init(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
int rcount, MPI_Datatype rtype, struct ompi_communicator_t *comm,
ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int res, indegree, outdegree, *srcs, *dsts;
MPI_Aint rcvext;
ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
@ -152,7 +154,7 @@ int ompi_coll_libnbc_ineighbor_allgather(const void *sbuf, int scount, MPI_Datat
}
#endif
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -160,3 +162,185 @@ int ompi_coll_libnbc_ineighbor_allgather(const void *sbuf, int scount, MPI_Datat
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ineighbor_allgather(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
int rcount, MPI_Datatype rtype, struct ompi_communicator_t *comm,
ompi_request_t ** request, struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_neighbor_allgather_init(sbuf, scount, stype, rbuf, rcount, rtype,
comm, request, module, false);
if (OPAL_LIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
/* better binomial bcast
* working principle:
* - each node gets a virtual rank vrank
* - the 'root' node get vrank 0
* - node 0 gets the vrank of the 'root'
* - all other ranks stay identical (they do not matter)
*
* Algorithm:
* - each node with vrank > 2^r and vrank < 2^r+1 receives from node
* vrank - 2^r (vrank=1 receives from 0, vrank 0 receives never)
* - each node sends each round r to node vrank + 2^r
* - a node stops to send if 2^r > commsize
*/
#define RANK2VRANK(rank, vrank, root) \
{ \
vrank = rank; \
if (rank == 0) vrank = root; \
if (rank == root) vrank = 0; \
}
#define VRANK2RANK(rank, vrank, root) \
{ \
rank = vrank; \
if (vrank == 0) rank = root; \
if (vrank == root) rank = 0; \
}
static inline int bcast_sched_binomial(int rank, int p, int root, NBC_Schedule *schedule, void *buffer, int count, MPI_Datatype datatype) {
int maxr, vrank, peer, res;
maxr = (int)ceil((log((double)p)/LOG2));
RANK2VRANK(rank, vrank, root);
/* receive from the right hosts */
if (vrank != 0) {
for (int r = 0 ; r < maxr ; ++r) {
if ((vrank >= (1 << r)) && (vrank < (1 << (r + 1)))) {
VRANK2RANK(peer, vrank - (1 << r), root);
res = NBC_Sched_recv (buffer, false, count, datatype, peer, schedule, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
}
}
res = NBC_Sched_barrier (schedule);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
}
/* now send to the right hosts */
for (int r = 0 ; r < maxr ; ++r) {
if (((vrank + (1 << r) < p) && (vrank < (1 << r))) || (vrank == 0)) {
VRANK2RANK(peer, vrank + (1 << r), root);
res = NBC_Sched_send (buffer, false, count, datatype, peer, schedule, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
}
}
return OMPI_SUCCESS;
}
/* simple linear MPI_Ibcast */
static inline int bcast_sched_linear(int rank, int p, int root, NBC_Schedule *schedule, void *buffer, int count, MPI_Datatype datatype) {
int res;
/* send to all others */
if(rank == root) {
for (int peer = 0 ; peer < p ; ++peer) {
if (peer != root) {
/* send msg to peer */
res = NBC_Sched_send (buffer, false, count, datatype, peer, schedule, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
}
}
} else {
/* recv msg from root */
res = NBC_Sched_recv (buffer, false, count, datatype, root, schedule, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
}
return OMPI_SUCCESS;
}
/* simple chained MPI_Ibcast */
static inline int bcast_sched_chain(int rank, int p, int root, NBC_Schedule *schedule, void *buffer, int count, MPI_Datatype datatype, int fragsize, size_t size) {
int res, vrank, rpeer, speer, numfrag, fragcount, thiscount;
MPI_Aint ext;
char *buf;
RANK2VRANK(rank, vrank, root);
VRANK2RANK(rpeer, vrank-1, root);
VRANK2RANK(speer, vrank+1, root);
res = ompi_datatype_type_extent(datatype, &ext);
if (MPI_SUCCESS != res) {
NBC_Error("MPI Error in ompi_datatype_type_extent() (%i)", res);
return res;
}
if (count == 0) {
return OMPI_SUCCESS;
}
numfrag = count * size/fragsize;
if ((count * size) % fragsize != 0) {
numfrag++;
}
fragcount = count/numfrag;
for (int fragnum = 0 ; fragnum < numfrag ; ++fragnum) {
buf = (char *) buffer + fragnum * fragcount * ext;
thiscount = fragcount;
if (fragnum == numfrag-1) {
/* last fragment may not be full */
thiscount = count - fragcount * fragnum;
}
/* root does not receive */
if (vrank != 0) {
res = NBC_Sched_recv (buf, false, thiscount, datatype, rpeer, schedule, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
}
/* last rank does not send */
if (vrank != p-1) {
res = NBC_Sched_send (buf, false, thiscount, datatype, speer, schedule, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
/* this barrier here seems awaward but isn't!!!! */
if (vrank == 0) {
res = NBC_Sched_barrier (schedule);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
}
}
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_neighbor_allgather_init(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
int rcount, MPI_Datatype rtype, struct ompi_communicator_t *comm,
MPI_Info info, ompi_request_t ** request, struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_neighbor_allgather_init(sbuf, scount, stype, rbuf, rcount, rtype,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

43
ompi/mca/coll/libnbc/nbc_ineighbor_allgatherv.c
Просмотреть файл

@ -10,6 +10,7 @@
* Copyright (c) 2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -43,10 +44,10 @@ int NBC_Ineighbor_allgatherv_args_compare(NBC_Ineighbor_allgatherv_args *a, NBC_
#endif
int ompi_coll_libnbc_ineighbor_allgatherv(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
const int *rcounts, const int *displs, MPI_Datatype rtype,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
static int nbc_neighbor_allgatherv_init(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
const int *rcounts, const int *displs, MPI_Datatype rtype,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int res, indegree, outdegree, *srcs, *dsts;
MPI_Aint rcvext;
ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
@ -154,7 +155,7 @@ int ompi_coll_libnbc_ineighbor_allgatherv(const void *sbuf, int scount, MPI_Data
}
#endif
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -162,3 +163,35 @@ int ompi_coll_libnbc_ineighbor_allgatherv(const void *sbuf, int scount, MPI_Data
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ineighbor_allgatherv(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
const int *rcounts, const int *displs, MPI_Datatype rtype,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_neighbor_allgatherv_init(sbuf, scount, stype, rbuf, rcounts, displs, rtype,
comm, request, module, false);
if (OPAL_LIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_neighbor_allgatherv_init(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
const int *rcounts, const int *displs, MPI_Datatype rtype,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_neighbor_allgatherv_init(sbuf, scount, stype, rbuf, rcounts, displs, rtype,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

40
ompi/mca/coll/libnbc/nbc_ineighbor_alltoall.c
Просмотреть файл

@ -10,6 +10,7 @@
* Copyright (c) 2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -40,9 +41,10 @@ int NBC_Ineighbor_alltoall_args_compare(NBC_Ineighbor_alltoall_args *a, NBC_Inei
}
#endif
int ompi_coll_libnbc_ineighbor_alltoall(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
int rcount, MPI_Datatype rtype, struct ompi_communicator_t *comm,
ompi_request_t ** request, struct mca_coll_base_module_2_2_0_t *module) {
static int nbc_neighbor_alltoall_init(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
int rcount, MPI_Datatype rtype, struct ompi_communicator_t *comm,
ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int res, indegree, outdegree, *srcs, *dsts;
MPI_Aint sndext, rcvext;
ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
@ -156,7 +158,7 @@ int ompi_coll_libnbc_ineighbor_alltoall(const void *sbuf, int scount, MPI_Dataty
}
#endif
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -164,3 +166,33 @@ int ompi_coll_libnbc_ineighbor_alltoall(const void *sbuf, int scount, MPI_Dataty
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ineighbor_alltoall(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
int rcount, MPI_Datatype rtype, struct ompi_communicator_t *comm,
ompi_request_t ** request, struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_neighbor_alltoall_init(sbuf, scount, stype, rbuf, rcount, rtype,
comm, request, module, false);
if (OPAL_LIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_neighbor_alltoall_init(const void *sbuf, int scount, MPI_Datatype stype, void *rbuf,
int rcount, MPI_Datatype rtype, struct ompi_communicator_t *comm, MPI_Info info,
ompi_request_t ** request, struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_neighbor_alltoall_init(sbuf, scount, stype, rbuf, rcount, rtype,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

43
ompi/mca/coll/libnbc/nbc_ineighbor_alltoallv.c
Просмотреть файл

@ -10,6 +10,7 @@
* Copyright (c) 2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -43,10 +44,10 @@ int NBC_Ineighbor_alltoallv_args_compare(NBC_Ineighbor_alltoallv_args *a, NBC_In
#endif
int ompi_coll_libnbc_ineighbor_alltoallv(const void *sbuf, const int *scounts, const int *sdispls, MPI_Datatype stype,
void *rbuf, const int *rcounts, const int *rdispls, MPI_Datatype rtype,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
static int nbc_neighbor_alltoallv_init(const void *sbuf, const int *scounts, const int *sdispls, MPI_Datatype stype,
void *rbuf, const int *rcounts, const int *rdispls, MPI_Datatype rtype,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int res, indegree, outdegree, *srcs, *dsts;
MPI_Aint sndext, rcvext;
ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
@ -161,7 +162,7 @@ int ompi_coll_libnbc_ineighbor_alltoallv(const void *sbuf, const int *scounts, c
}
#endif
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -169,3 +170,35 @@ int ompi_coll_libnbc_ineighbor_alltoallv(const void *sbuf, const int *scounts, c
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ineighbor_alltoallv(const void *sbuf, const int *scounts, const int *sdispls, MPI_Datatype stype,
void *rbuf, const int *rcounts, const int *rdispls, MPI_Datatype rtype,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_neighbor_alltoallv_init(sbuf, scounts, sdispls, stype, rbuf, rcounts, rdispls, rtype,
comm, request, module, false);
if (OPAL_LIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_neighbor_alltoallv_init(const void *sbuf, const int *scounts, const int *sdispls, MPI_Datatype stype,
void *rbuf, const int *rcounts, const int *rdispls, MPI_Datatype rtype,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_neighbor_alltoallv_init(sbuf, scounts, sdispls, stype, rbuf, rcounts, rdispls, rtype,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

43
ompi/mca/coll/libnbc/nbc_ineighbor_alltoallw.c
Просмотреть файл

@ -10,6 +10,7 @@
* Copyright (c) 2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -42,10 +43,10 @@ int NBC_Ineighbor_alltoallw_args_compare(NBC_Ineighbor_alltoallw_args *a, NBC_In
}
#endif
int ompi_coll_libnbc_ineighbor_alltoallw(const void *sbuf, const int *scounts, const MPI_Aint *sdisps, struct ompi_datatype_t * const *stypes,
void *rbuf, const int *rcounts, const MPI_Aint *rdisps, struct ompi_datatype_t * const *rtypes,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
static int nbc_neighbor_alltoallw_init(const void *sbuf, const int *scounts, const MPI_Aint *sdisps, struct ompi_datatype_t * const *stypes,
void *rbuf, const int *rcounts, const MPI_Aint *rdisps, struct ompi_datatype_t * const *rtypes,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int res, indegree, outdegree, *srcs, *dsts;
ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
NBC_Schedule *schedule;
@ -146,7 +147,7 @@ int ompi_coll_libnbc_ineighbor_alltoallw(const void *sbuf, const int *scounts, c
}
#endif
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -154,3 +155,35 @@ int ompi_coll_libnbc_ineighbor_alltoallw(const void *sbuf, const int *scounts, c
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ineighbor_alltoallw(const void *sbuf, const int *scounts, const MPI_Aint *sdisps, struct ompi_datatype_t * const *stypes,
void *rbuf, const int *rcounts, const MPI_Aint *rdisps, struct ompi_datatype_t * const *rtypes,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_neighbor_alltoallw_init(sbuf, scounts, sdisps, stypes, rbuf, rcounts, rdisps, rtypes,
comm, request, module, false);
if (OPAL_LIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_neighbor_alltoallw_init(const void *sbuf, const int *scounts, const MPI_Aint *sdisps, struct ompi_datatype_t * const *stypes,
void *rbuf, const int *rcounts, const MPI_Aint *rdisps, struct ompi_datatype_t * const *rtypes,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_neighbor_alltoallw_init(sbuf, scounts, sdisps, stypes, rbuf, rcounts, rdisps, rtypes,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

20
ompi/mca/coll/libnbc/nbc_internal.h
Просмотреть файл

@ -14,7 +14,10 @@
* and Technology (RIST). All rights reserved.
* Copyright (c) 2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*/
#ifndef __NBC_INTERNAL_H__
#define __NBC_INTERNAL_H__
@ -259,9 +262,10 @@ void NBC_SchedCache_args_delete_key_dummy(void *k);
#endif
int NBC_Start(NBC_Handle *handle, NBC_Schedule *schedule);
int NBC_Init_handle(struct ompi_communicator_t *comm, ompi_coll_libnbc_request_t **request, ompi_coll_libnbc_module_t *module);
int NBC_Schedule_request(NBC_Schedule *schedule, ompi_communicator_t *comm, ompi_coll_libnbc_module_t *module, ompi_request_t **request, void *tmpbuf);
int NBC_Start(NBC_Handle *handle);
int NBC_Schedule_request(NBC_Schedule *schedule, ompi_communicator_t *comm,
ompi_coll_libnbc_module_t *module, bool persistent,
ompi_request_t **request, void *tmpbuf);
void NBC_Return_handle(ompi_coll_libnbc_request_t *request);
static inline int NBC_Type_intrinsic(MPI_Datatype type);
int NBC_Create_fortran_handle(int *fhandle, NBC_Handle **handle);
@ -365,6 +369,16 @@ static inline void nbc_schedule_inc_round (NBC_Schedule *schedule) {
memcpy (lastround, &last_round_num, sizeof (last_round_num));
}
/* returns a no-operation request (e.g. for one process barrier) */
static inline int nbc_get_noop_request(bool persistent, ompi_request_t **request) {
if (persistent) {
return ompi_request_persistent_noop_create(request);
} else {
*request = &ompi_request_empty;
return OMPI_SUCCESS;
}
}
/* NBC_PRINT_ROUND prints a round in a schedule. A round has the format:
* [num]{[type][type-args]} types: [int]{[enum][args-type]}
* e.g. [2][SEND][SEND-ARGS][RECV][RECV-ARGS] */

105
ompi/mca/coll/libnbc/nbc_ireduce.c
Просмотреть файл

@ -10,6 +10,7 @@
* Copyright (c) 2014-2017 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -52,9 +53,9 @@ int NBC_Reduce_args_compare(NBC_Reduce_args *a, NBC_Reduce_args *b, void *param)
#endif
/* the non-blocking reduce */
int ompi_coll_libnbc_ireduce(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype,
MPI_Op op, int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
static int nbc_reduce_init(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype,
MPI_Op op, int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int rank, p, res, segsize;
size_t size;
MPI_Aint ext;
@ -84,15 +85,14 @@ int ompi_coll_libnbc_ireduce(const void* sendbuf, void* recvbuf, int count, MPI_
}
/* only one node -> copy data */
if (p == 1) {
if (1 == p && (!persistent || inplace)) {
if (!inplace) {
res = NBC_Copy (sendbuf, count, datatype, recvbuf, count, datatype, comm);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
}
*request = &ompi_request_empty;
return OMPI_SUCCESS;
return nbc_get_noop_request(persistent, request);
}
span = opal_datatype_span(&datatype->super, count, &gap);
@ -140,13 +140,18 @@ int ompi_coll_libnbc_ireduce(const void* sendbuf, void* recvbuf, int count, MPI_
return OMPI_ERR_OUT_OF_RESOURCE;
}
switch(alg) {
case NBC_RED_BINOMIAL:
res = red_sched_binomial(rank, p, root, sendbuf, redbuf, tmpredbuf, count, datatype, op, inplace, schedule, tmpbuf);
break;
case NBC_RED_CHAIN:
res = red_sched_chain(rank, p, root, sendbuf, recvbuf, count, datatype, op, ext, size, schedule, tmpbuf, segsize);
break;
if (p == 1) {
res = NBC_Sched_copy ((void *)sendbuf, false, count, datatype,
recvbuf, false, count, datatype, schedule, false);
} else {
switch(alg) {
case NBC_RED_BINOMIAL:
res = red_sched_binomial(rank, p, root, sendbuf, redbuf, tmpredbuf, count, datatype, op, inplace, schedule, tmpbuf);
break;
case NBC_RED_CHAIN:
res = red_sched_chain(rank, p, root, sendbuf, recvbuf, count, datatype, op, ext, size, schedule, tmpbuf, segsize);
break;
}
}
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
@ -193,7 +198,7 @@ int ompi_coll_libnbc_ireduce(const void* sendbuf, void* recvbuf, int count, MPI_
}
#endif
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, tmpbuf);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
free(tmpbuf);
@ -203,9 +208,27 @@ int ompi_coll_libnbc_ireduce(const void* sendbuf, void* recvbuf, int count, MPI_
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ireduce_inter(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype,
MPI_Op op, int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
int ompi_coll_libnbc_ireduce(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype,
MPI_Op op, int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_reduce_init(sendbuf, recvbuf, count, datatype, op, root,
comm, request, module, false);
if (OPAL_LIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
static int nbc_reduce_inter_init(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype,
MPI_Op op, int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int rank, res, rsize;
NBC_Schedule *schedule;
ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
@ -241,7 +264,7 @@ int ompi_coll_libnbc_ireduce_inter(const void* sendbuf, void* recvbuf, int count
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, tmpbuf);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
free(tmpbuf);
@ -251,6 +274,24 @@ int ompi_coll_libnbc_ireduce_inter(const void* sendbuf, void* recvbuf, int count
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ireduce_inter(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype,
MPI_Op op, int root, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_reduce_inter_init(sendbuf, recvbuf, count, datatype, op, root,
comm, request, module, false);
if (OPAL_LIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
/* binomial reduce
* if op is not commutative, reduce on rank 0, and then send the result to root rank
@ -315,7 +356,9 @@ static inline int red_sched_binomial (int rank, int p, int root, const void *sen
rbuf = redbuf;
tmprbuf = tmpredbuf;
if (inplace) {
res = NBC_Copy(rbuf, count, datatype, ((char *)tmpbuf)-gap, count, datatype, MPI_COMM_SELF);
res = NBC_Sched_copy(rbuf, false, count, datatype,
((char *)tmpbuf)-gap, false, count, datatype,
schedule, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
@ -516,3 +559,27 @@ static inline int red_sched_linear (int rank, int rsize, int root, const void *s
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_reduce_init(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype,
MPI_Op op, int root, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_reduce_init(sendbuf, recvbuf, count, datatype, op, root,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_reduce_inter_init(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype,
MPI_Op op, int root, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_reduce_inter_init(sendbuf, recvbuf, count, datatype, op, root,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

91
ompi/mca/coll/libnbc/nbc_ireduce_scatter.c
Просмотреть файл

@ -13,6 +13,7 @@
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -41,9 +42,9 @@
*
*/
int ompi_coll_libnbc_ireduce_scatter(const void* sendbuf, void* recvbuf, const int *recvcounts, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
static int nbc_reduce_scatter_init(const void* sendbuf, void* recvbuf, const int *recvcounts, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int peer, rank, maxr, p, res, count;
MPI_Aint ext;
ptrdiff_t gap, span, span_align;
@ -69,7 +70,7 @@ int ompi_coll_libnbc_ireduce_scatter(const void* sendbuf, void* recvbuf, const i
count += recvcounts[r];
}
if (p == 1 || 0 == count) {
if ((1 == p && (!persistent || inplace)) || 0 == count) {
if (!inplace) {
/* single node not in_place: copy data to recvbuf */
res = NBC_Copy(sendbuf, recvcounts[0], datatype, recvbuf, recvcounts[0], datatype, comm);
@ -78,8 +79,7 @@ int ompi_coll_libnbc_ireduce_scatter(const void* sendbuf, void* recvbuf, const i
}
}
*request = &ompi_request_empty;
return OMPI_SUCCESS;
return nbc_get_noop_request(persistent, request);
}
maxr = (int) ceil ((log((double) p) / LOG2));
@ -174,8 +174,14 @@ int ompi_coll_libnbc_ireduce_scatter(const void* sendbuf, void* recvbuf, const i
}
}
res = NBC_Sched_copy (lbuf, true, recvcounts[0], datatype, recvbuf, false,
recvcounts[0], datatype, schedule, false);
if (p == 1) {
/* single node not in_place: copy data to recvbuf */
res = NBC_Sched_copy ((void *)sendbuf, false, recvcounts[0], datatype,
recvbuf, false, recvcounts[0], datatype, schedule, false);
} else {
res = NBC_Sched_copy (lbuf, true, recvcounts[0], datatype, recvbuf, false,
recvcounts[0], datatype, schedule, false);
}
} else {
res = NBC_Sched_recv (recvbuf, false, recvcounts[rank], datatype, 0, schedule, false);
}
@ -193,7 +199,7 @@ int ompi_coll_libnbc_ireduce_scatter(const void* sendbuf, void* recvbuf, const i
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, tmpbuf);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
free(tmpbuf);
@ -203,9 +209,26 @@ int ompi_coll_libnbc_ireduce_scatter(const void* sendbuf, void* recvbuf, const i
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ireduce_scatter_inter (const void* sendbuf, void* recvbuf, const int *recvcounts, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
int ompi_coll_libnbc_ireduce_scatter (const void* sendbuf, void* recvbuf, const int *recvcounts, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_reduce_scatter_init(sendbuf, recvbuf, recvcounts, datatype, op,
comm, request, module, false);
if (OPAL_LIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
static int nbc_reduce_scatter_inter_init (const void* sendbuf, void* recvbuf, const int *recvcounts, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int rank, res, count, lsize, rsize;
MPI_Aint ext;
ptrdiff_t gap, span, span_align;
@ -318,7 +341,7 @@ int ompi_coll_libnbc_ireduce_scatter_inter (const void* sendbuf, void* recvbuf,
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, tmpbuf);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
free(tmpbuf);
@ -327,3 +350,45 @@ int ompi_coll_libnbc_ireduce_scatter_inter (const void* sendbuf, void* recvbuf,
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ireduce_scatter_inter (const void* sendbuf, void* recvbuf, const int *recvcounts, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_reduce_scatter_inter_init(sendbuf, recvbuf, recvcounts, datatype, op,
comm, request, module, false);
if (OPAL_LIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_reduce_scatter_init(const void* sendbuf, void* recvbuf, const int *recvcounts, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_reduce_scatter_init(sendbuf, recvbuf, recvcounts, datatype, op,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_reduce_scatter_inter_init(const void* sendbuf, void* recvbuf, const int *recvcounts, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_reduce_scatter_inter_init(sendbuf, recvbuf, recvcounts, datatype, op,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

80
ompi/mca/coll/libnbc/nbc_ireduce_scatter_block.c
Просмотреть файл

@ -11,6 +11,7 @@
* Copyright (c) 2014-2017 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -39,9 +40,9 @@
*
*/
int ompi_coll_libnbc_ireduce_scatter_block(const void* sendbuf, void* recvbuf, int recvcount, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
static int nbc_reduce_scatter_block_init(const void* sendbuf, void* recvbuf, int recvcount, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int peer, rank, maxr, p, res, count;
MPI_Aint ext;
ptrdiff_t gap, span;
@ -88,7 +89,8 @@ int ompi_coll_libnbc_ireduce_scatter_block(const void* sendbuf, void* recvbuf, i
/* copy data to redbuf if we only have a single node */
if ((p == 1) && !inplace) {
res = NBC_Copy (sendbuf, count, datatype, redbuf, count, datatype, comm);
res = NBC_Sched_copy ((void *)sendbuf, false, count, datatype,
redbuf, false, count, datatype, schedule, false);
if (OMPI_SUCCESS != res) {
OBJ_RELEASE(schedule);
free(tmpbuf);
@ -195,7 +197,7 @@ int ompi_coll_libnbc_ireduce_scatter_block(const void* sendbuf, void* recvbuf, i
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, tmpbuf);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
free(tmpbuf);
@ -205,9 +207,27 @@ int ompi_coll_libnbc_ireduce_scatter_block(const void* sendbuf, void* recvbuf, i
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ireduce_scatter_block_inter(const void *sendbuf, void *recvbuf, int rcount, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm,
ompi_request_t **request, struct mca_coll_base_module_2_2_0_t *module) {
int ompi_coll_libnbc_ireduce_scatter_block(const void* sendbuf, void* recvbuf, int recvcount, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_reduce_scatter_block_init(sendbuf, recvbuf, recvcount, datatype, op,
comm, request, module, false);
if (OPAL_LIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
static int nbc_reduce_scatter_block_inter_init(const void *sendbuf, void *recvbuf, int rcount, struct ompi_datatype_t *dtype,
struct ompi_op_t *op, struct ompi_communicator_t *comm, ompi_request_t **request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int rank, res, count, lsize, rsize;
MPI_Aint ext;
ptrdiff_t gap, span, span_align;
@ -316,7 +336,7 @@ int ompi_coll_libnbc_ireduce_scatter_block_inter(const void *sendbuf, void *recv
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, tmpbuf);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
free(tmpbuf);
@ -325,3 +345,45 @@ int ompi_coll_libnbc_ireduce_scatter_block_inter(const void *sendbuf, void *recv
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_ireduce_scatter_block_inter(const void* sendbuf, void* recvbuf, int recvcount, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_reduce_scatter_block_inter_init(sendbuf, recvbuf, recvcount, datatype, op,
comm, request, module, false);
if (OPAL_LIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_reduce_scatter_block_init(const void* sendbuf, void* recvbuf, int recvcount, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_reduce_scatter_block_init(sendbuf, recvbuf, recvcount, datatype, op,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_reduce_scatter_block_inter_init(const void* sendbuf, void* recvbuf, int recvcount, MPI_Datatype datatype,
MPI_Op op, struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_reduce_scatter_block_inter_init(sendbuf, recvbuf, recvcount, datatype, op,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

57
ompi/mca/coll/libnbc/nbc_iscan.c
Просмотреть файл

@ -10,6 +10,7 @@
* Copyright (c) 2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -45,9 +46,9 @@ int NBC_Scan_args_compare(NBC_Scan_args *a, NBC_Scan_args *b, void *param) {
* 3. all but rank p-1 do sends to it's right neighbor and exits
*
*/
int ompi_coll_libnbc_iscan(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
static int nbc_scan_init(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int rank, p, res;
ptrdiff_t gap, span;
NBC_Schedule *schedule;
@ -60,14 +61,6 @@ int ompi_coll_libnbc_iscan(const void* sendbuf, void* recvbuf, int count, MPI_Da
rank = ompi_comm_rank (comm);
p = ompi_comm_size (comm);
if (!inplace) {
/* copy data to receivebuf */
res = NBC_Copy (sendbuf, count, datatype, recvbuf, count, datatype, comm);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
}
#ifdef NBC_CACHE_SCHEDULE
NBC_Scan_args *args, *found, search;
@ -85,6 +78,16 @@ int ompi_coll_libnbc_iscan(const void* sendbuf, void* recvbuf, int count, MPI_Da
return OMPI_ERR_OUT_OF_RESOURCE;
}
if (!inplace) {
/* copy data to receivebuf */
res = NBC_Sched_copy ((void *)sendbuf, false, count, datatype,
recvbuf, false, count, datatype, schedule, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
}
}
if(rank != 0) {
span = opal_datatype_span(&datatype->super, count, &gap);
tmpbuf = malloc (span);
@ -159,7 +162,7 @@ int ompi_coll_libnbc_iscan(const void* sendbuf, void* recvbuf, int count, MPI_Da
}
#endif
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, tmpbuf);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
free(tmpbuf);
@ -168,3 +171,33 @@ int ompi_coll_libnbc_iscan(const void* sendbuf, void* recvbuf, int count, MPI_Da
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_iscan(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_scan_init(sendbuf, recvbuf, count, datatype, op,
comm, request, module, false);
if (OPAL_LIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_scan_init(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_scan_init(sendbuf, recvbuf, count, datatype, op,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

106
ompi/mca/coll/libnbc/nbc_iscatter.c
Просмотреть файл

@ -13,6 +13,7 @@
* Copyright (c) 2014-2017 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -44,10 +45,10 @@ int NBC_Scatter_args_compare(NBC_Scatter_args *a, NBC_Scatter_args *b, void *par
#endif
/* simple linear MPI_Iscatter */
int ompi_coll_libnbc_iscatter (const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
static int nbc_scatter_init (const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int rank, p, res;
MPI_Aint sndext = 0;
NBC_Schedule *schedule;
@ -69,15 +70,6 @@ int ompi_coll_libnbc_iscatter (const void* sendbuf, int sendcount, MPI_Datatype
}
}
if ((rank == root) && (!inplace)) {
sbuf = (char *) sendbuf + rank * sendcount * sndext;
/* if I am the root - just copy the message (not for MPI_IN_PLACE) */
res = NBC_Copy (sbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
}
#ifdef NBC_CACHE_SCHEDULE
NBC_Scatter_args *args, *found, search;
@ -108,7 +100,17 @@ int ompi_coll_libnbc_iscatter (const void* sendbuf, int sendcount, MPI_Datatype
} else {
for (int i = 0 ; i < p ; ++i) {
sbuf = (char *) sendbuf + i * sendcount * sndext;
if (i != root) {
if (i == root) {
if (!inplace) {
/* if I am the root - just copy the message */
res = NBC_Sched_copy (sbuf, false, sendcount, sendtype,
recvbuf, false, recvcount, recvtype, schedule, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
}
}
} else {
/* root sends the right buffer to the right receiver */
res = NBC_Sched_send (sbuf, false, sendcount, sendtype, i, schedule, false);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
@ -157,7 +159,7 @@ int ompi_coll_libnbc_iscatter (const void* sendbuf, int sendcount, MPI_Datatype
}
#endif
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -166,10 +168,29 @@ int ompi_coll_libnbc_iscatter (const void* sendbuf, int sendcount, MPI_Datatype
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_iscatter_inter (const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
int ompi_coll_libnbc_iscatter (const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_scatter_init(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root,
comm, request, module, false);
if (OPAL_LIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
static int nbc_scatter_inter_init (const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int res, rsize;
MPI_Aint sndext;
NBC_Schedule *schedule;
@ -217,7 +238,7 @@ int ompi_coll_libnbc_iscatter_inter (const void* sendbuf, int sendcount, MPI_Dat
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -225,3 +246,48 @@ int ompi_coll_libnbc_iscatter_inter (const void* sendbuf, int sendcount, MPI_Dat
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_iscatter_inter (const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_scatter_inter_init(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root,
comm, request, module, false);
if (OPAL_LIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_scatter_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_scatter_init(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_scatter_inter_init(const void* sendbuf, int sendcount, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_scatter_inter_init(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

88
ompi/mca/coll/libnbc/nbc_iscatterv.c
Просмотреть файл

@ -13,6 +13,7 @@
* Copyright (c) 2014-2017 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2017 IBM Corporation. All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -27,10 +28,10 @@
* would not be sufficient ... we simply do not cache it */
/* simple linear MPI_Iscatterv */
int ompi_coll_libnbc_iscatterv(const void* sendbuf, const int *sendcounts, const int *displs, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
static int nbc_scatterv_init(const void* sendbuf, const int *sendcounts, const int *displs, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int rank, p, res;
MPI_Aint sndext;
NBC_Schedule *schedule;
@ -63,7 +64,8 @@ int ompi_coll_libnbc_iscatterv(const void* sendbuf, const int *sendcounts, const
if (i == root) {
if (!inplace) {
/* if I am the root - just copy the message */
res = NBC_Copy (sbuf, sendcounts[i], sendtype, recvbuf, recvcount, recvtype, comm);
res = NBC_Sched_copy (sbuf, false, sendcounts[i], sendtype,
recvbuf, false, recvcount, recvtype, schedule, false);
} else {
res = OMPI_SUCCESS;
}
@ -92,7 +94,7 @@ int ompi_coll_libnbc_iscatterv(const void* sendbuf, const int *sendcounts, const
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -101,10 +103,29 @@ int ompi_coll_libnbc_iscatterv(const void* sendbuf, const int *sendcounts, const
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_iscatterv_inter (const void* sendbuf, const int *sendcounts, const int *displs, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_2_0_t *module) {
int ompi_coll_libnbc_iscatterv(const void* sendbuf, const int *sendcounts, const int *displs, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_scatterv_init(sendbuf, sendcounts, displs, sendtype, recvbuf, recvcount, recvtype, root,
comm, request, module, false);
if (OPAL_LIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
static int nbc_scatterv_inter_init (const void* sendbuf, const int *sendcounts, const int *displs, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module, bool persistent) {
int res, rsize;
MPI_Aint sndext;
NBC_Schedule *schedule;
@ -151,7 +172,7 @@ int ompi_coll_libnbc_iscatterv_inter (const void* sendbuf, const int *sendcounts
return res;
}
res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
res = NBC_Schedule_request(schedule, comm, libnbc_module, persistent, request, NULL);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
OBJ_RELEASE(schedule);
return res;
@ -159,3 +180,48 @@ int ompi_coll_libnbc_iscatterv_inter (const void* sendbuf, const int *sendcounts
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_iscatterv_inter(const void* sendbuf, const int *sendcounts, const int *displs, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_scatterv_inter_init(sendbuf, sendcounts, displs, sendtype, recvbuf, recvcount, recvtype, root,
comm, request, module, false);
if (OPAL_LIKELY(OMPI_SUCCESS != res)) {
return res;
}
res = NBC_Start(*(ompi_coll_libnbc_request_t **)request);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
NBC_Return_handle ((ompi_coll_libnbc_request_t *)request);
*request = &ompi_request_null.request;
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_scatterv_init(const void* sendbuf, const int *sendcounts, const int *displs, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_scatterv_init(sendbuf, sendcounts, displs, sendtype, recvbuf, recvcount, recvtype, root,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}
int ompi_coll_libnbc_scatterv_inter_init(const void* sendbuf, const int *sendcounts, const int *displs, MPI_Datatype sendtype,
void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
struct ompi_communicator_t *comm, MPI_Info info, ompi_request_t ** request,
struct mca_coll_base_module_2_3_0_t *module) {
int res = nbc_scatterv_inter_init(sendbuf, sendcounts, displs, sendtype, recvbuf, recvcount, recvtype, root,
comm, request, module, true);
if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
return res;
}
return OMPI_SUCCESS;
}

6
ompi/mca/coll/portals4/coll_portals4.h
Просмотреть файл

@ -206,18 +206,18 @@ int ompi_coll_portals4_ireduce_intra(const void* sendbuf, void* recvbuf, int cou
int root,
struct ompi_communicator_t *comm,
ompi_request_t ** ompi_request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_portals4_ireduce_intra_fini(struct ompi_coll_portals4_request_t *request);
int ompi_coll_portals4_allreduce_intra(const void* sendbuf, void* recvbuf, int count,
MPI_Datatype dtype, MPI_Op op,
struct ompi_communicator_t *comm,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int ompi_coll_portals4_iallreduce_intra(const void* sendbuf, void* recvbuf, int count,
MPI_Datatype dtype, MPI_Op op,
struct ompi_communicator_t *comm,
ompi_request_t ** ompi_request,
struct mca_coll_base_module_2_2_0_t *module);
struct mca_coll_base_module_2_3_0_t *module);
int
ompi_coll_portals4_iallreduce_intra_fini(struct ompi_coll_portals4_request_t *request);

4
ompi/mca/coll/portals4/coll_portals4_allreduce.c
Просмотреть файл

@ -360,7 +360,7 @@ allreduce_kary_tree_bottom(ompi_coll_portals4_request_t *request)
int ompi_coll_portals4_allreduce_intra(const void* sendbuf, void* recvbuf, int count,
MPI_Datatype dtype, MPI_Op op,
struct ompi_communicator_t *comm,
struct mca_coll_base_module_2_2_0_t *module)
struct mca_coll_base_module_2_3_0_t *module)
{
mca_coll_portals4_module_t *portals4_module = (mca_coll_portals4_module_t*) module;
ompi_coll_portals4_request_t *request;
@ -390,7 +390,7 @@ int ompi_coll_portals4_iallreduce_intra(const void* sendbuf, void* recvbuf, int
MPI_Datatype dtype, MPI_Op op,
struct ompi_communicator_t *comm,
ompi_request_t ** ompi_request,
struct mca_coll_base_module_2_2_0_t *module)
struct mca_coll_base_module_2_3_0_t *module)
{
mca_coll_portals4_module_t *portals4_module = (mca_coll_portals4_module_t*) module;
ompi_coll_portals4_request_t *request;

2
ompi/mca/coll/portals4/coll_portals4_barrier.c
Просмотреть файл

@ -269,7 +269,7 @@ ompi_coll_portals4_barrier_intra(struct ompi_communicator_t *comm,
int
ompi_coll_portals4_ibarrier_intra(struct ompi_communicator_t *comm,
ompi_request_t **ompi_req,
struct mca_coll_base_module_2_2_0_t *module)
struct mca_coll_base_module_2_3_0_t *module)
{
int ret;
mca_coll_portals4_module_t *portals4_module = (mca_coll_portals4_module_t*) module;

2
ompi/mca/coll/portals4/coll_portals4_reduce.c
Просмотреть файл

@ -402,7 +402,7 @@ ompi_coll_portals4_ireduce_intra(const void* sendbuf, void* recvbuf, int count,
int root,
struct ompi_communicator_t *comm,
ompi_request_t ** ompi_request,
struct mca_coll_base_module_2_2_0_t *module)
struct mca_coll_base_module_2_3_0_t *module)
{
int ret;
mca_coll_portals4_module_t *portals4_module = (mca_coll_portals4_module_t*) module;

13
ompi/mpi/c/bsend_init.c
Просмотреть файл

@ -15,6 +15,7 @@
* reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -74,16 +75,8 @@ int MPI_Bsend_init(const void *buf, int count, MPI_Datatype type,
}
if (MPI_PROC_NULL == dest) {
*request = OBJ_NEW(ompi_request_t);
/* Other fields were initialized by the constructor for
ompi_request_t */
(*request)->req_type = OMPI_REQUEST_NOOP;
(*request)->req_status = ompi_request_empty.req_status;
(*request)->req_complete = REQUEST_COMPLETED;
(*request)->req_state = OMPI_REQUEST_INACTIVE;
(*request)->req_persistent = true;
(*request)->req_free = ompi_request_persistent_proc_null_free;
return MPI_SUCCESS;
rc = ompi_request_persistent_noop_create(request);
OMPI_ERRHANDLER_RETURN(rc, comm, rc, FUNC_NAME);
}
OPAL_CR_ENTER_LIBRARY();

13
ompi/mpi/c/recv_init.c
Просмотреть файл

@ -12,6 +12,7 @@
* Copyright (c) 2006 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -70,16 +71,8 @@ int MPI_Recv_init(void *buf, int count, MPI_Datatype type, int source,
}
if (MPI_PROC_NULL == source) {
*request = OBJ_NEW(ompi_request_t);
/* Other fields were initialized by the constructor for
ompi_request_t */
(*request)->req_type = OMPI_REQUEST_NOOP;
(*request)->req_status = ompi_request_empty.req_status;
(*request)->req_complete = REQUEST_COMPLETED;
(*request)->req_state = OMPI_REQUEST_INACTIVE;
(*request)->req_persistent = true;
(*request)->req_free = ompi_request_persistent_proc_null_free;
return MPI_SUCCESS;
rc = ompi_request_persistent_noop_create(request);
OMPI_ERRHANDLER_RETURN(rc, comm, rc, FUNC_NAME);
}
OPAL_CR_ENTER_LIBRARY();

13
ompi/mpi/c/rsend_init.c
Просмотреть файл

@ -15,6 +15,7 @@
* reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -75,16 +76,8 @@ int MPI_Rsend_init(const void *buf, int count, MPI_Datatype type,
}
if (MPI_PROC_NULL == dest) {
*request = OBJ_NEW(ompi_request_t);
/* Other fields were initialized by the constructor for
ompi_request_t */
(*request)->req_type = OMPI_REQUEST_NOOP;
(*request)->req_status = ompi_request_empty.req_status;
(*request)->req_complete = REQUEST_COMPLETED;
(*request)->req_state = OMPI_REQUEST_INACTIVE;
(*request)->req_persistent = true;
(*request)->req_free = ompi_request_persistent_proc_null_free;
return MPI_SUCCESS;
rc = ompi_request_persistent_noop_create(request);
OMPI_ERRHANDLER_RETURN(rc, comm, rc, FUNC_NAME);
}
OPAL_CR_ENTER_LIBRARY();

13
ompi/mpi/c/send_init.c
Просмотреть файл

@ -15,6 +15,7 @@
* reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -75,16 +76,8 @@ int MPI_Send_init(const void *buf, int count, MPI_Datatype type,
}
if (MPI_PROC_NULL == dest) {
*request = OBJ_NEW(ompi_request_t);
/* Other fields were initialized by the constructor for
ompi_request_t */
(*request)->req_type = OMPI_REQUEST_NOOP;
(*request)->req_status = ompi_request_empty.req_status;
(*request)->req_complete = REQUEST_COMPLETED;
(*request)->req_state = OMPI_REQUEST_INACTIVE;
(*request)->req_persistent = true;
(*request)->req_free = ompi_request_persistent_proc_null_free;
return MPI_SUCCESS;
rc = ompi_request_persistent_noop_create(request);
OMPI_ERRHANDLER_RETURN(rc, comm, rc, FUNC_NAME);
}
OPAL_CR_ENTER_LIBRARY();

13
ompi/mpi/c/ssend_init.c
Просмотреть файл

@ -15,6 +15,7 @@
* reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -75,16 +76,8 @@ int MPI_Ssend_init(const void *buf, int count, MPI_Datatype type,
}
if (MPI_PROC_NULL == dest) {
*request = OBJ_NEW(ompi_request_t);
/* Other fields were initialized by the constructor for
ompi_request_t */
(*request)->req_type = OMPI_REQUEST_NOOP;
(*request)->req_status = ompi_request_empty.req_status;
(*request)->req_complete = REQUEST_COMPLETED;
(*request)->req_state = OMPI_REQUEST_INACTIVE;
(*request)->req_persistent = true;
(*request)->req_free = ompi_request_persistent_proc_null_free;
return MPI_SUCCESS;
rc = ompi_request_persistent_noop_create(request);
OMPI_ERRHANDLER_RETURN(rc, comm, rc, FUNC_NAME);
}
OPAL_CR_ENTER_LIBRARY();

7
ompi/mpi/c/start.c
Просмотреть файл

@ -10,8 +10,9 @@
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2006 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2015 Research Organization for Information Science
* Copyright (c) 2015-2017 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2017 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -66,6 +67,10 @@ int MPI_Start(MPI_Request *request)
switch((*request)->req_type) {
case OMPI_REQUEST_PML:
case OMPI_REQUEST_COLL:
if ( MPI_PARAM_CHECK && !(*request)->req_persistent) {
return OMPI_ERRHANDLER_INVOKE(MPI_COMM_WORLD, MPI_ERR_REQUEST, FUNC_NAME);
}
OPAL_CR_ENTER_LIBRARY();
ret = (*request)->req_start(1, request);

7
ompi/mpi/c/startall.c
Просмотреть файл

@ -13,8 +13,9 @@
* Copyright (c) 2006 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2012-2016 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2014-2015 Research Organization for Information Science
* Copyright (c) 2014-2017 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2017-2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -64,7 +65,9 @@ int MPI_Startall(int count, MPI_Request requests[])
} else {
for (i = 0; i < count; ++i) {
if (NULL == requests[i] ||
(OMPI_REQUEST_PML != requests[i]->req_type &&
! requests[i]->req_persistent ||
(OMPI_REQUEST_PML != requests[i]->req_type &&
OMPI_REQUEST_COLL != requests[i]->req_type &&
OMPI_REQUEST_NOOP != requests[i]->req_type)) {
rc = MPI_ERR_REQUEST;
break;

10
ompi/mpiext/pcollreq/Makefile.am Обычный файл
Просмотреть файл

@ -0,0 +1,10 @@
#
# Copyright (c) 2017 FUJITSU LIMITED. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
SUBDIRS = c

13
ompi/mpiext/pcollreq/README.txt Обычный файл
Просмотреть файл

@ -0,0 +1,13 @@
Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
$COPYRIGHT$
This extension provides the feature of persistent collective communication
operations and persistent neighborhood collective communication operations,
which is proposed in the MPI Forum as of June 2018.
See MPIX_Barrier_init(3) for more details.
The code will be moved to the ompi/mpi directory and the MPIX_ prefix will
be switch to the MPI_ prefix once the MPI Standard which includes this
feature is published.

1
ompi/mpiext/pcollreq/c/MPIX_Allgather_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Allgatherv_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Allreduce_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Alltoall_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Alltoallv_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Alltoallw_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

157
ompi/mpiext/pcollreq/c/MPIX_Barrier_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1,157 @@
.\" -*- nroff -*-
.\" Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
.\" $COPYRIGHT$
.TH MPIX_Barrier_init 3 "#OMPI_DATE#" "#PACKAGE_VERSION#" "#PACKAGE_NAME#"
.SH NAME
\fBMPIX_Allgather_init, MPIX_Allgatherv_init, MPIX_Allreduce_init, MPIX_Alltoall_init, MPIX_Alltoallv_init, MPIX_Alltoallw_init, MPIX_Barrier_init, MPIX_Bcast_init, MPIX_Exscan_init, MPIX_Gather_init, MPIX_Gatherv_init, MPIX_Reduce_init, MPIX_Reduce_scatter_init, MPIX_Reduce_scatter_block_init, MPIX_Scan_init, MPIX_Scatter_init, MPIX_Scatterv_init, MPIX_Neighbor_allgather_init, MPIX_Neighbor_allgatherv_init, MPIX_Neighbor_alltoall_init, MPIX_Neighbor_alltoallv_init, MPIX_Neighbor_alltoallw_init\fP \- Builds a handle for a collective communication or neighborhood collective communication
.SH SYNTAX
.ft R
.SH C Syntax
.nf
#include <mpi.h>
#include <mpi-ext.h>
int MPIX_Allgather_init(const void *\fIsendbuf\fP, int \fIsendcount\fP,
MPI_Datatype \fIsendtype\fP, void *\fIrecvbuf\fP, int \fIrecvcount\fP,
MPI_Datatype \fIrecvtype\fP, MPI_Comm \fIcomm\fP, MPI_Info \fIinfo\fP,
MPI_Request *\fIrequest\fP)
int MPIX_Allgatherv_init(const void *\fIsendbuf\fP, int \fIsendcount\fP,
MPI_Datatype \fIsendtype\fP, void *\fIrecvbuf\fP, const int \fIrecvcounts\fP[],
const int \fIdispls\fP[], MPI_Datatype \fIrecvtype\fP, MPI_Comm \fIcomm\fP,
MPI_Info \fIinfo\fP, MPI_Request *\fIrequest\fP)
int MPIX_Allreduce_init(const void *\fIsendbuf\fP, void *\fIrecvbuf\fP, int \fIcount\fP,
MPI_Datatype \fIdatatype\fP, MPI_Op \fIop\fP, MPI_Comm \fIcomm\fP, MPI_Info \fIinfo\fP,
MPI_Request *\fIrequest\fP)
int MPIX_Alltoall_init(const void *\fIsendbuf\fP, int \fIsendcount\fP,
MPI_Datatype \fIsendtype\fP, void *\fIrecvbuf\fP, int \fIrecvcount\fP,
MPI_Datatype \fIrecvtype\fP, MPI_Comm \fIcomm\fP, MPI_Info \fIinfo\fP,
MPI_Request *\fIrequest\fP)
int MPIX_Alltoallv_init(const void *\fIsendbuf\fP, const int \fIsendcounts\fP[],
const int \fIsdispls\fP[], MPI_Datatype \fIsendtype\fP, void *\fIrecvbuf\fP,
const int \fIrecvcounts\fP[], const int \fIrdispls\fP[], MPI_Datatype \fIrecvtype\fP,
MPI_Comm \fIcomm\fP, MPI_Info \fIinfo\fP, MPI_Request *\fIrequest\fP)
int MPIX_Alltoallw_init(const void *\fIsendbuf\fP, const int \fIsendcounts\fP[],
const int \fIsdispls\fP[], const MPI_Datatype \fIsendtypes\fP[], void *\fIrecvbuf\fP,
const int \fIrecvcounts\fP[], const int \fIrdispls\fP[],
const MPI_Datatype \fIrecvtypes\fP[], MPI_Comm \fIcomm\fP, MPI_Info \fIinfo\fP,
MPI_Request *\fIrequest\fP)
int MPIX_Barrier_init(MPI_Comm \fIcomm\fP, MPI_Info \fIinfo\fP,
MPI_Request *\fIrequest\fP)
int MPIX_Bcast_init(void *\fIbuffer\fP, int \fIcount\fP, MPI_Datatype \fIdatatype\fP,
int \fIroot\fP, MPI_Comm \fIcomm\fP, MPI_Info \fIinfo\fP, MPI_Request *\fIrequest\fP)
int MPIX_Exscan_init(const void *\fIsendbuf\fP, void *\fIrecvbuf\fP, int \fIcount\fP,
MPI_Datatype \fIdatatype\fP, MPI_Op \fIop\fP, MPI_Comm \fIcomm\fP, MPI_Info \fIinfo\fP,
MPI_Request *\fIrequest\fP)
int MPIX_Gather_init(const void *\fIsendbuf\fP, int \fIsendcount\fP,
MPI_Datatype \fIsendtype\fP, void *\fIrecvbuf\fP, int \fIrecvcount\fP,
MPI_Datatype \fIrecvtype\fP, int \fIroot\fP, MPI_Comm \fIcomm\fP, MPI_Info \fIinfo\fP,
MPI_Request *\fIrequest\fP)
int MPIX_Gatherv_init(const void *\fIsendbuf\fP, int \fIsendcount\fP,
MPI_Datatype \fIsendtype\fP, void *\fIrecvbuf\fP, const int \fIrecvcounts\fP[],
const int \fIdispls\fP[], MPI_Datatype \fIrecvtype\fP, int \fIroot\fP, MPI_Comm \fIcomm\fP,
MPI_Info \fIinfo\fP, MPI_Request *\fIrequest\fP)
int MPIX_Reduce_init(const void *\fIsendbuf\fP, void *\fIrecvbuf\fP, int \fIcount\fP,
MPI_Datatype \fIdatatype\fP, MPI_Op \fIop\fP, int \fIroot\fP, MPI_Comm \fIcomm\fP,
MPI_Info \fIinfo\fP, MPI_Request *\fIrequest\fP)
int MPIX_Reduce_scatter_init(const void *\fIsendbuf\fP, void *\fIrecvbuf\fP,
const int \fIrecvcounts\fP[], MPI_Datatype \fIdatatype\fP, MPI_Op \fIop\fP,
MPI_Comm \fIcomm\fP, MPI_Info \fIinfo\fP, MPI_Request *\fIrequest\fP)
int MPIX_Reduce_scatter_block_init(const void *\fIsendbuf\fP, void *\fIrecvbuf\fP,
int \fIrecvcount\fP, MPI_Datatype \fIdatatype\fP, MPI_Op \fIop\fP, MPI_Comm \fIcomm\fP,
MPI_Info \fIinfo\fP, MPI_Request *\fIrequest\fP)
int MPIX_Scan_init(const void *\fIsendbuf\fP, void *\fIrecvbuf\fP, int \fIcount\fP,
MPI_Datatype \fIdatatype\fP, MPI_Op \fIop\fP, MPI_Comm \fIcomm\fP, MPI_Info \fIinfo\fP,
MPI_Request *\fIrequest\fP)
int MPIX_Scatter_init(const void *\fIsendbuf\fP, int \fIsendcount\fP,
MPI_Datatype \fIsendtype\fP, void *\fIrecvbuf\fP, int \fIrecvcount\fP,
MPI_Datatype \fIrecvtype\fP, int \fIroot\fP, MPI_Comm \fIcomm\fP, MPI_Info \fIinfo\fP,
MPI_Request *\fIrequest\fP)
int MPIX_Scatterv_init(const void *\fIsendbuf\fP, const int \fIsendcounts\fP[],
const int \fIdispls\fP[], MPI_Datatype \fIsendtype\fP, void *\fIrecvbuf\fP,
int \fIrecvcount\fP, MPI_Datatype \fIrecvtype\fP, int \fIroot\fP, MPI_Comm \fIcomm\fP,
MPI_Info \fIinfo\fP, MPI_Request *\fIrequest\fP)
int MPIX_Neighbor_allgather_init(const void *\fIsendbuf\fP, int \fIsendcount\fP,
MPI_Datatype \fIsendtype\fP, void *\fIrecvbuf\fP, int \fIrecvcount\fP,
MPI_Datatype \fIrecvtype\fP, MPI_Comm \fIcomm\fP, MPI_Info \fIinfo\fP,
MPI_Request *\fIrequest\fP)
int MPIX_Neighbor_allgatherv_init(const void *\fIsendbuf\fP, int \fIsendcount\fP,
MPI_Datatype \fIsendtype\fP, void *\fIrecvbuf\fP, const int \fIrecvcounts\fP[],
const int \fIdispls\fP[], MPI_Datatype \fIrecvtype\fP, MPI_Comm \fIcomm\fP,
MPI_Info \fIinfo\fP, MPI_Request *\fIrequest\fP)
int MPIX_Neighbor_alltoall_init(const void *\fIsendbuf\fP, int \fIsendcount\fP,
MPI_Datatype \fIsendtype\fP, void *\fIrecvbuf\fP, int \fIrecvcount\fP,
MPI_Datatype \fIrecvtype\fP, MPI_Comm \fIcomm\fP, MPI_Info \fIinfo\fP,
MPI_Request *\fIrequest\fP)
int MPIX_Neighbor_alltoallv_init(const void *\fIsendbuf\fP,
const int \fIsendcounts\fP[], const int \fIsdispls\fP[], MPI_Datatype \fIsendtype\fP,
void *\fIrecvbuf\fP, const int \fIrecvcounts\fP[], const int \fIrdispls\fP[],
MPI_Datatype \fIrecvtype\fP, MPI_Comm \fIcomm\fP, MPI_Info \fIinfo\fP,
MPI_Request *\fIrequest\fP)
int MPIX_Neighbor_alltoallw_init(const void *\fIsendbuf\fP,
const int \fIsendcounts\fP[], const MPI_Aint \fIsdispls\fP[],
const MPI_Datatype \fIsendtypes\fP[], void *\fIrecvbuf\fP,
const int \fIrecvcounts\fP[], const MPI_Aint \fIrdispls\fP[],
const MPI_Datatype \fIrecvtypes\fP[], MPI_Comm \fIcomm\fP, MPI_Info \fIinfo\fP,
MPI_Request *\fIrequest\fP)
.fi
.SH DESCRIPTION
.ft R
Creates a persistent communication request for a collective operation or neighborhood collective operation.
As of June 2018, the feature of persistent collective communication operations and persistent collective neighborhood communication operations is proposed in the MPI Forum.
.nf
https://github.com/mpi-forum/mpi-issues/issues/25
.fi
Open MPI implements its draft version shown in the following URL.
.nf
https://github.com/mpi-forum/mpi-issues/files/2078076/mpi32-report-ticket25-austin-vote-june2018.pdf
.fi
Because it is still in a draft stage, the interface may change in the standard. Therefore the prefix \fIMPIX_\fP is used instead of \fIMPI_\fP for these request creation functions. To start, complete, and free the created request, usual MPI functions (\fIMPI_Start\fP etc.) can be used. Only C bindings are available currently.
Future versions of Open MPI will switch to the \fIMPI_\fP prefix and will not require the header file \fImpi-ext.h\fP once the MPI Standard which includes this feature is published.
.SH EXAMPLE
.nf
MPI_Request req;
MPIX_Barrier_init(MPI_COMM_WORLD, MPI_INFO_NULL, &req);
MPI_Start(&req);
MPI_Wait(&req, MPI_STATUS_IGNORE);
MPI_Request_free(&req);
.fi
.SH SEE ALSO
.ft R
.sp
MPI_Start
.br
MPI_Startall
.br
MPI_Request_free

1
ompi/mpiext/pcollreq/c/MPIX_Bcast_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Exscan_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Gather_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Gatherv_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Neighbor_allgather_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Neighbor_allgatherv_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Neighbor_alltoall_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Neighbor_alltoallv_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Neighbor_alltoallw_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Reduce_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Reduce_scatter_block_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Reduce_scatter_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Scan_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Scatter_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

1
ompi/mpiext/pcollreq/c/MPIX_Scatterv_init.3in Обычный файл
Просмотреть файл

@ -0,0 +1 @@
.so man3/MPIX_Barrier_init.3

63
ompi/mpiext/pcollreq/c/Makefile.am Обычный файл
Просмотреть файл

@ -0,0 +1,63 @@
#
# Copyright (c) 2017 FUJITSU LIMITED. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# We must set these #defines so that the inner OMPI MPI prototype
# header files do the Right Thing.
AM_CPPFLAGS = -DOMPI_PROFILE_LAYER=0 -DOMPI_COMPILING_FORTRAN_WRAPPERS=1
include $(top_srcdir)/Makefile.ompi-rules
# Convenience libtool library that will be slurped up into libmpi.la.
noinst_LTLIBRARIES = libmpiext_pcollreq_c.la
# This is where the top-level header file (that is included in
# <mpi-ext.h>) must be installed.
ompidir = $(ompiincludedir)/ompi/mpiext/pcollreq/c
# This is the header file that is installed.
ompi_HEADERS = mpiext_pcollreq_c.h
# Sources for the convenience libtool library. Other than the one
# header file, all source files in the extension have no file naming
# conventions.
libmpiext_pcollreq_c_la_SOURCES = \
$(ompi_HEADERS) \
mpiext_pcollreq.c
libmpiext_pcollreq_c_la_LDFLAGS = -module -avoid-version
# Man page installation
nodist_man_MANS = \
MPIX_Allgather_init.3 \
MPIX_Allgatherv_init.3 \
MPIX_Allreduce_init.3 \
MPIX_Alltoall_init.3 \
MPIX_Alltoallv_init.3 \
MPIX_Alltoallw_init.3 \
MPIX_Barrier_init.3 \
MPIX_Bcast_init.3 \
MPIX_Exscan_init.3 \
MPIX_Gather_init.3 \
MPIX_Gatherv_init.3 \
MPIX_Reduce_init.3 \
MPIX_Reduce_scatter_block_init.3 \
MPIX_Reduce_scatter_init.3 \
MPIX_Scan_init.3 \
MPIX_Scatter_init.3 \
MPIX_Scatterv_init.3 \
MPIX_Neighbor_allgather_init.3 \
MPIX_Neighbor_allgatherv_init.3 \
MPIX_Neighbor_alltoall_init.3 \
MPIX_Neighbor_alltoallv_init.3 \
MPIX_Neighbor_alltoallw_init.3
# Man page sources
EXTRA_DIST = $(nodist_man_MANS:.3=.3in)
distclean-local:
rm -f $(nodist_man_MANS)

174
ompi/mpiext/pcollreq/c/mpiext_pcollreq.c Обычный файл
Просмотреть файл

@ -0,0 +1,174 @@
/*
* Copyright (c) 2017 FUJITSU LIMITED. All rights reserved.
* Copyright (c) 2017 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*
*/
#include "ompi_config.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/mca/coll/base/coll_base_functions.h"
#include "ompi/communicator/communicator.h"
#include "mpiext_pcollreq_c.h"
#define INFO_REQ_ARGS ompi_info_t *info, ompi_request_t **request
int MPIX_Allgather_init(ALLGATHER_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_allgather_init(
ALLGATHER_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_allgather_init_module);
}
int MPIX_Allgatherv_init(ALLGATHERV_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_allgatherv_init(
ALLGATHERV_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_allgatherv_init_module);
}
int MPIX_Allreduce_init(ALLREDUCE_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_allreduce_init(
ALLREDUCE_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_allreduce_init_module);
}
int MPIX_Alltoall_init(ALLTOALL_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_alltoall_init(
ALLTOALL_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_alltoall_init_module);
}
int MPIX_Alltoallv_init(ALLTOALLV_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_alltoallv_init(
ALLTOALLV_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_alltoallv_init_module);
}
int MPIX_Alltoallw_init(ALLTOALLW_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_alltoallw_init(
ALLTOALLW_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_alltoallw_init_module);
}
int MPIX_Barrier_init(BARRIER_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_barrier_init(
BARRIER_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_barrier_init_module);
}
int MPIX_Bcast_init(BCAST_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_bcast_init(
BCAST_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_bcast_init_module);
}
int MPIX_Exscan_init(EXSCAN_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_exscan_init(
EXSCAN_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_exscan_init_module);
}
int MPIX_Gather_init(GATHER_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_gather_init(
GATHER_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_gather_init_module);
}
int MPIX_Gatherv_init(GATHERV_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_gatherv_init(
GATHERV_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_gatherv_init_module);
}
int MPIX_Reduce_init(REDUCE_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_reduce_init(
REDUCE_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_reduce_init_module);
}
int MPIX_Reduce_scatter_init(REDUCESCATTER_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_reduce_scatter_init(
REDUCESCATTER_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_reduce_scatter_init_module);
}
int MPIX_Reduce_scatter_block_init(REDUCESCATTERBLOCK_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_reduce_scatter_block_init(
REDUCESCATTERBLOCK_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_reduce_scatter_block_init_module);
}
int MPIX_Scan_init(SCAN_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_scan_init(
SCAN_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_scan_init_module);
}
int MPIX_Scatter_init(SCATTER_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_scatter_init(
SCATTER_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_scatter_init_module);
}
int MPIX_Scatterv_init(SCATTERV_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_scatterv_init(
SCATTERV_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_scatterv_init_module);
}
int MPIX_Neighbor_allgather_init(NEIGHBOR_ALLGATHER_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_neighbor_allgather_init(
NEIGHBOR_ALLGATHER_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_neighbor_allgather_init_module);
}
int MPIX_Neighbor_allgatherv_init(NEIGHBOR_ALLGATHERV_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_neighbor_allgatherv_init(
NEIGHBOR_ALLGATHERV_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_neighbor_allgatherv_init_module);
}
int MPIX_Neighbor_alltoall_init(NEIGHBOR_ALLTOALL_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_neighbor_alltoall_init(
NEIGHBOR_ALLTOALL_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_neighbor_alltoall_init_module);
}
int MPIX_Neighbor_alltoallv_init(NEIGHBOR_ALLTOALLV_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_neighbor_alltoallv_init(
NEIGHBOR_ALLTOALLV_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_neighbor_alltoallv_init_module);
}
int MPIX_Neighbor_alltoallw_init(NEIGHBOR_ALLTOALLW_BASE_ARGS, INFO_REQ_ARGS)
{
return comm->c_coll->coll_neighbor_alltoallw_init(
NEIGHBOR_ALLTOALLW_BASE_ARG_NAMES, info, request,
comm->c_coll->coll_neighbor_alltoallw_init_module);
}

33
ompi/mpiext/pcollreq/c/mpiext_pcollreq_c.h Обычный файл
Просмотреть файл

@ -0,0 +1,33 @@
/*
* Copyright (c) 2017 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*
*/
OMPI_DECLSPEC int MPIX_Allgather_init(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Allgatherv_init(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, const int recvcounts[], const int displs[], MPI_Datatype recvtype, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Allreduce_init(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Alltoall_init(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Alltoallv_init(const void *sendbuf, const int sendcounts[], const int sdispls[], MPI_Datatype sendtype, void *recvbuf, const int recvcounts[], const int rdispls[], MPI_Datatype recvtype, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Alltoallw_init(const void *sendbuf, const int sendcounts[], const int sdispls[], const MPI_Datatype sendtypes[], void *recvbuf, const int recvcounts[], const int rdispls[], const MPI_Datatype recvtypes[], MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Barrier_init(MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Bcast_init(void *buffer, int count, MPI_Datatype datatype, int root, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Exscan_init(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Gather_init(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Gatherv_init(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, const int recvcounts[], const int displs[], MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Reduce_init(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Reduce_scatter_init(const void *sendbuf, void *recvbuf, const int recvcounts[], MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Reduce_scatter_block_init(const void *sendbuf, void *recvbuf, int recvcount, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Scan_init(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Scatter_init(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Scatterv_init(const void *sendbuf, const int sendcounts[], const int displs[], MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Neighbor_allgather_init(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Neighbor_allgatherv_init(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, const int recvcounts[], const int displs[], MPI_Datatype recvtype, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Neighbor_alltoall_init(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Neighbor_alltoallv_init(const void *sendbuf, const int sendcounts[], const int sdispls[], MPI_Datatype sendtype, void *recvbuf, const int recvcounts[], const int rdispls[], MPI_Datatype recvtype, MPI_Comm comm, MPI_Info info, MPI_Request *request);
OMPI_DECLSPEC int MPIX_Neighbor_alltoallw_init(const void *sendbuf, const int sendcounts[], const MPI_Aint sdispls[], const MPI_Datatype sendtypes[], void *recvbuf, const int recvcounts[], const MPI_Aint rdispls[], const MPI_Datatype recvtypes[], MPI_Comm comm, MPI_Info info, MPI_Request *request);

21
ompi/mpiext/pcollreq/configure.m4 Обычный файл
Просмотреть файл

@ -0,0 +1,21 @@
# -*- shell-script -*-
#
# Copyright (c) 2017 FUJITSU LIMITED. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# OMPI_MPIEXT_pcollreq_CONFIG([action-if-found], [action-if-not-found])
# -----------------------------------------------------------
AC_DEFUN([OMPI_MPIEXT_pcollreq_CONFIG],[
AC_CONFIG_FILES([ompi/mpiext/pcollreq/Makefile])
AC_CONFIG_FILES([ompi/mpiext/pcollreq/c/Makefile])
AS_IF([test "$ENABLE_pcollreq" = "1" || \
test "$ENABLE_EXT_ALL" = "1"],
[$1],
[$2])
])

26
ompi/request/request.c
Просмотреть файл

@ -17,6 +17,7 @@
* reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -86,7 +87,7 @@ static int ompi_request_empty_free(ompi_request_t** request)
return OMPI_SUCCESS;
}
int ompi_request_persistent_proc_null_free(ompi_request_t** request)
static int ompi_request_persistent_noop_free(ompi_request_t** request)
{
OMPI_REQUEST_FINI(*request);
(*request)->req_state = OMPI_REQUEST_INVALID;
@ -185,3 +186,26 @@ int ompi_request_finalize(void)
OBJ_DESTRUCT( &ompi_request_f_to_c_table );
return OMPI_SUCCESS;
}
int ompi_request_persistent_noop_create(ompi_request_t** request)
{
ompi_request_t *req;
req = OBJ_NEW(ompi_request_t);
if (NULL == req) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
/* Other fields were initialized by the constructor for
ompi_request_t */
req->req_type = OMPI_REQUEST_NOOP;
req->req_status = ompi_request_empty.req_status;
req->req_complete = REQUEST_COMPLETED;
req->req_state = OMPI_REQUEST_INACTIVE;
req->req_persistent = true;
req->req_free = ompi_request_persistent_noop_free;
*request = req;
return OMPI_SUCCESS;
}

12
ompi/request/request.h
Просмотреть файл

@ -15,6 +15,7 @@
* Copyright (c) 2012 Oak Ridge National Labs. All rights reserved.
* Copyright (c) 2015-2017 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2018 FUJITSU LIMITED. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -348,17 +349,16 @@ OMPI_DECLSPEC extern ompi_request_fns_t ompi_request_functions;
*/
int ompi_request_init(void);
/**
* Free a persistent request to a MPI_PROC_NULL peer (there's no
* freelist to put it back to, so we have to actually OBJ_RELEASE it).
*/
OMPI_DECLSPEC int ompi_request_persistent_proc_null_free(ompi_request_t **request);
/**
* Shut down the MPI_Request subsystem; invoked during MPI_FINALIZE.
*/
int ompi_request_finalize(void);
/**
* Create a persistent request that does nothing (e.g., to MPI_PROC_NULL).
*/
int ompi_request_persistent_noop_create(ompi_request_t **request);
/**
* Cancel a pending request.
*/