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

coll/basic: fix non standard ddt handling

Просмотр исходного кода 
- correctly handle non zero lower bound ddt
 - correctly handle ddt with size > extent

Thanks Yuki Matsumoto for the report
Этот коммит содержится в:
Gilles Gouaillardet 2016-07-05 13:32:36 +09:00
родитель c06fb04a9a
Коммит 488d037d51
3 изменённых файлов: 45 добавлений и 39 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

19
ompi/mca/coll/basic/coll_basic_allgather.c
Просмотреть файл

@ -9,7 +9,7 @@
* University of Stuttgart. All rights reserved. * University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California. * Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved. * All rights reserved.
* Copyright (c) 2014-2015 Research Organization for Information Science * Copyright (c) 2014-2016 Research Organization for Information Science
* and Technology (RIST). All rights reserved. * and Technology (RIST). All rights reserved.
* $COPYRIGHT$ * $COPYRIGHT$
* *
@ -48,8 +48,9 @@ mca_coll_basic_allgather_inter(const void *sbuf, int scount,
mca_coll_base_module_t *module) mca_coll_base_module_t *module)
{ {
int rank, root = 0, size, rsize, err, i, line; int rank, root = 0, size, rsize, err, i, line;
char *tmpbuf = NULL, *ptmp; char *tmpbuf_free = NULL, *tmpbuf, *ptmp;
ptrdiff_t rlb, slb, rextent, sextent, incr; ptrdiff_t rlb, rextent, incr;
ptrdiff_t gap, span;
ompi_request_t *req; ompi_request_t *req;
ompi_request_t **reqs = NULL; ompi_request_t **reqs = NULL;
@ -75,8 +76,6 @@ mca_coll_basic_allgather_inter(const void *sbuf, int scount,
/* receive a msg. from all other procs. */ /* receive a msg. from all other procs. */
err = ompi_datatype_get_extent(rdtype, &rlb, &rextent); err = ompi_datatype_get_extent(rdtype, &rlb, &rextent);
if (OMPI_SUCCESS != err) { line = __LINE__; goto exit; } if (OMPI_SUCCESS != err) { line = __LINE__; goto exit; }
err = ompi_datatype_get_extent(sdtype, &slb, &sextent);
if (OMPI_SUCCESS != err) { line = __LINE__; goto exit; }
/* Get a requests arrays of the right size */ /* Get a requests arrays of the right size */
reqs = coll_base_comm_get_reqs(module->base_data, rsize + 1); reqs = coll_base_comm_get_reqs(module->base_data, rsize + 1);
@ -107,8 +106,10 @@ mca_coll_basic_allgather_inter(const void *sbuf, int scount,
if (OMPI_SUCCESS != err) { line = __LINE__; goto exit; } if (OMPI_SUCCESS != err) { line = __LINE__; goto exit; }
/* Step 2: exchange the resuts between the root processes */ /* Step 2: exchange the resuts between the root processes */
tmpbuf = (char *) malloc(scount * size * sextent); span = opal_datatype_span(&sdtype->super, scount * size, &gap);
if (NULL == tmpbuf) { line = __LINE__; err = OMPI_ERR_OUT_OF_RESOURCE; goto exit; } tmpbuf_free = (char *) malloc(span);
if (NULL == tmpbuf_free) { line = __LINE__; err = OMPI_ERR_OUT_OF_RESOURCE; goto exit; }
tmpbuf = tmpbuf_free - gap;
err = MCA_PML_CALL(isend(rbuf, rsize * rcount, rdtype, 0, err = MCA_PML_CALL(isend(rbuf, rsize * rcount, rdtype, 0,
MCA_COLL_BASE_TAG_ALLGATHER, MCA_COLL_BASE_TAG_ALLGATHER,
@ -158,8 +159,8 @@ mca_coll_basic_allgather_inter(const void *sbuf, int scount,
(void)line; // silence compiler warning (void)line; // silence compiler warning
if( NULL != reqs ) ompi_coll_base_free_reqs(reqs, rsize+1); if( NULL != reqs ) ompi_coll_base_free_reqs(reqs, rsize+1);
} }
if (NULL != tmpbuf) { if (NULL != tmpbuf_free) {
free(tmpbuf); free(tmpbuf_free);
} }
return err; return err;

29
ompi/mca/coll/basic/coll_basic_reduce_scatter.c
Просмотреть файл

@ -14,7 +14,7 @@
* Copyright (c) 2012 Oak Ridge National Labs. All rights reserved. * Copyright (c) 2012 Oak Ridge National Labs. All rights reserved.
* Copyright (c) 2013 Los Alamos National Security, LLC. All rights * Copyright (c) 2013 Los Alamos National Security, LLC. All rights
* reserved. * reserved.
* Copyright (c) 2014-2015 Research Organization for Information Science * Copyright (c) 2014-2016 Research Organization for Information Science
* and Technology (RIST). All rights reserved. * and Technology (RIST). All rights reserved.
* $COPYRIGHT$ * $COPYRIGHT$
* *
@ -367,8 +367,9 @@ mca_coll_basic_reduce_scatter_inter(const void *sbuf, void *rbuf, const int *rco
{ {
int err, i, rank, root = 0, rsize, lsize; int err, i, rank, root = 0, rsize, lsize;
int totalcounts; int totalcounts;
ptrdiff_t lb, extent; ptrdiff_t gap, span;
char *tmpbuf = NULL, *tmpbuf2 = NULL; char *tmpbuf = NULL, *tmpbuf2 = NULL;
char *lbuf, *buf;
ompi_request_t *req; ompi_request_t *req;
int *disps = NULL; int *disps = NULL;
@ -399,10 +400,7 @@ mca_coll_basic_reduce_scatter_inter(const void *sbuf, void *rbuf, const int *rco
* its size is the same as the local communicator size. * its size is the same as the local communicator size.
*/ */
if (rank == root) { if (rank == root) {
err = ompi_datatype_get_extent(dtype, &lb, &extent); span = opal_datatype_span(&dtype->super, totalcounts, &gap);
if (OMPI_SUCCESS != err) {
return OMPI_ERROR;
}
/* Generate displacements for the scatterv part */ /* Generate displacements for the scatterv part */
disps = (int*) malloc(sizeof(int) * lsize); disps = (int*) malloc(sizeof(int) * lsize);
@ -414,12 +412,14 @@ mca_coll_basic_reduce_scatter_inter(const void *sbuf, void *rbuf, const int *rco
disps[i + 1] = disps[i] + rcounts[i]; disps[i + 1] = disps[i] + rcounts[i];
} }
tmpbuf = (char *) malloc(totalcounts * extent); tmpbuf = (char *) malloc(span);
tmpbuf2 = (char *) malloc(totalcounts * extent); tmpbuf2 = (char *) malloc(span);
if (NULL == tmpbuf || NULL == tmpbuf2) { if (NULL == tmpbuf || NULL == tmpbuf2) {
err = OMPI_ERR_OUT_OF_RESOURCE; err = OMPI_ERR_OUT_OF_RESOURCE;
goto exit; goto exit;
} }
lbuf = tmpbuf - gap;
buf = tmpbuf2 - gap;
/* Do a send-recv between the two root procs. to avoid deadlock */ /* Do a send-recv between the two root procs. to avoid deadlock */
err = MCA_PML_CALL(isend(sbuf, totalcounts, dtype, 0, err = MCA_PML_CALL(isend(sbuf, totalcounts, dtype, 0,
@ -429,7 +429,7 @@ mca_coll_basic_reduce_scatter_inter(const void *sbuf, void *rbuf, const int *rco
goto exit; goto exit;
} }
err = MCA_PML_CALL(recv(tmpbuf2, totalcounts, dtype, 0, err = MCA_PML_CALL(recv(lbuf, totalcounts, dtype, 0,
MCA_COLL_BASE_TAG_REDUCE_SCATTER, comm, MCA_COLL_BASE_TAG_REDUCE_SCATTER, comm,
MPI_STATUS_IGNORE)); MPI_STATUS_IGNORE));
if (OMPI_SUCCESS != err) { if (OMPI_SUCCESS != err) {
@ -444,10 +444,11 @@ mca_coll_basic_reduce_scatter_inter(const void *sbuf, void *rbuf, const int *rco
/* Loop receiving and calling reduction function (C or Fortran) /* Loop receiving and calling reduction function (C or Fortran)
* The result of this reduction operations is then in * The result of this reduction operations is then in
* tmpbuf2. * lbuf.
*/ */
for (i = 1; i < rsize; i++) { for (i = 1; i < rsize; i++) {
err = MCA_PML_CALL(recv(tmpbuf, totalcounts, dtype, i, char *tbuf;
err = MCA_PML_CALL(recv(buf, totalcounts, dtype, i,
MCA_COLL_BASE_TAG_REDUCE_SCATTER, comm, MCA_COLL_BASE_TAG_REDUCE_SCATTER, comm,
MPI_STATUS_IGNORE)); MPI_STATUS_IGNORE));
if (MPI_SUCCESS != err) { if (MPI_SUCCESS != err) {
@ -455,7 +456,9 @@ mca_coll_basic_reduce_scatter_inter(const void *sbuf, void *rbuf, const int *rco
} }
/* Perform the reduction */ /* Perform the reduction */
ompi_op_reduce(op, tmpbuf, tmpbuf2, totalcounts, dtype); ompi_op_reduce(op, lbuf, buf, totalcounts, dtype);
/* swap the buffers */
tbuf = lbuf; lbuf = buf; buf = tbuf;
} }
} else { } else {
/* If not root, send data to the root. */ /* If not root, send data to the root. */
@ -468,7 +471,7 @@ mca_coll_basic_reduce_scatter_inter(const void *sbuf, void *rbuf, const int *rco
} }
/* Now do a scatterv on the local communicator */ /* Now do a scatterv on the local communicator */
err = comm->c_local_comm->c_coll.coll_scatterv(tmpbuf2, rcounts, disps, dtype, err = comm->c_local_comm->c_coll.coll_scatterv(lbuf, rcounts, disps, dtype,
rbuf, rcounts[rank], dtype, 0, rbuf, rcounts[rank], dtype, 0,
comm->c_local_comm, comm->c_local_comm,
comm->c_local_comm->c_coll.coll_scatterv_module); comm->c_local_comm->c_coll.coll_scatterv_module);

36
ompi/mca/coll/basic/coll_basic_reduce_scatter_block.c
Просмотреть файл

@ -12,7 +12,7 @@
* Copyright (c) 2008 Sun Microsystems, Inc. All rights reserved. * Copyright (c) 2008 Sun Microsystems, Inc. All rights reserved.
* Copyright (c) 2012 Oak Ridge National Labs. All rights reserved. * Copyright (c) 2012 Oak Ridge National Labs. All rights reserved.
* Copyright (c) 2012 Sandia National Laboratories. All rights reserved. * Copyright (c) 2012 Sandia National Laboratories. All rights reserved.
* Copyright (c) 2014-2015 Research Organization for Information Science * Copyright (c) 2014-2016 Research Organization for Information Science
* and Technology (RIST). All rights reserved. * and Technology (RIST). All rights reserved.
* $COPYRIGHT$ * $COPYRIGHT$
* *
@ -58,7 +58,7 @@ mca_coll_basic_reduce_scatter_block_intra(const void *sbuf, void *rbuf, int rcou
mca_coll_base_module_t *module) mca_coll_base_module_t *module)
{ {
int rank, size, count, err = OMPI_SUCCESS; int rank, size, count, err = OMPI_SUCCESS;
ptrdiff_t extent, buf_size, gap; ptrdiff_t gap, span;
char *recv_buf = NULL, *recv_buf_free = NULL; char *recv_buf = NULL, *recv_buf_free = NULL;
/* Initialize */ /* Initialize */
@ -72,8 +72,7 @@ mca_coll_basic_reduce_scatter_block_intra(const void *sbuf, void *rbuf, int rcou
} }
/* get datatype information */ /* get datatype information */
ompi_datatype_type_extent(dtype, &extent); span = opal_datatype_span(&dtype->super, count, &gap);
buf_size = opal_datatype_span(&dtype->super, count, &gap);
/* Handle MPI_IN_PLACE */ /* Handle MPI_IN_PLACE */
if (MPI_IN_PLACE == sbuf) { if (MPI_IN_PLACE == sbuf) {
@ -83,12 +82,12 @@ mca_coll_basic_reduce_scatter_block_intra(const void *sbuf, void *rbuf, int rcou
if (0 == rank) { if (0 == rank) {
/* temporary receive buffer. See coll_basic_reduce.c for /* temporary receive buffer. See coll_basic_reduce.c for
details on sizing */ details on sizing */
recv_buf_free = (char*) malloc(buf_size); recv_buf_free = (char*) malloc(span);
recv_buf = recv_buf_free - gap;
if (NULL == recv_buf_free) { if (NULL == recv_buf_free) {
err = OMPI_ERR_OUT_OF_RESOURCE; err = OMPI_ERR_OUT_OF_RESOURCE;
goto cleanup; goto cleanup;
} }
recv_buf = recv_buf_free - gap;
} }
/* reduction */ /* reduction */
@ -126,8 +125,9 @@ mca_coll_basic_reduce_scatter_block_inter(const void *sbuf, void *rbuf, int rcou
{ {
int err, i, rank, root = 0, rsize, lsize; int err, i, rank, root = 0, rsize, lsize;
int totalcounts; int totalcounts;
ptrdiff_t lb, extent; ptrdiff_t gap, span;
char *tmpbuf = NULL, *tmpbuf2 = NULL; char *tmpbuf = NULL, *tmpbuf2 = NULL;
char *lbuf, *buf;
ompi_request_t *req; ompi_request_t *req;
rank = ompi_comm_rank(comm); rank = ompi_comm_rank(comm);
@ -151,16 +151,15 @@ mca_coll_basic_reduce_scatter_block_inter(const void *sbuf, void *rbuf, int rcou
* *
*/ */
if (rank == root) { if (rank == root) {
err = ompi_datatype_get_extent(dtype, &lb, &extent); span = opal_datatype_span(&dtype->super, totalcounts, &gap);
if (OMPI_SUCCESS != err) {
return OMPI_ERROR;
}
tmpbuf = (char *) malloc(totalcounts * extent); tmpbuf = (char *) malloc(span);
tmpbuf2 = (char *) malloc(totalcounts * extent); tmpbuf2 = (char *) malloc(span);
if (NULL == tmpbuf || NULL == tmpbuf2) { if (NULL == tmpbuf || NULL == tmpbuf2) {
return OMPI_ERR_OUT_OF_RESOURCE; return OMPI_ERR_OUT_OF_RESOURCE;
} }
lbuf = tmpbuf - gap;
buf = tmpbuf2 - gap;
/* Do a send-recv between the two root procs. to avoid deadlock */ /* Do a send-recv between the two root procs. to avoid deadlock */
err = MCA_PML_CALL(isend(sbuf, totalcounts, dtype, 0, err = MCA_PML_CALL(isend(sbuf, totalcounts, dtype, 0,
@ -170,7 +169,7 @@ mca_coll_basic_reduce_scatter_block_inter(const void *sbuf, void *rbuf, int rcou
goto exit; goto exit;
} }
err = MCA_PML_CALL(recv(tmpbuf2, totalcounts, dtype, 0, err = MCA_PML_CALL(recv(lbuf, totalcounts, dtype, 0,
MCA_COLL_BASE_TAG_REDUCE_SCATTER, comm, MCA_COLL_BASE_TAG_REDUCE_SCATTER, comm,
MPI_STATUS_IGNORE)); MPI_STATUS_IGNORE));
if (OMPI_SUCCESS != err) { if (OMPI_SUCCESS != err) {
@ -188,7 +187,8 @@ mca_coll_basic_reduce_scatter_block_inter(const void *sbuf, void *rbuf, int rcou
* tmpbuf2. * tmpbuf2.
*/ */
for (i = 1; i < rsize; i++) { for (i = 1; i < rsize; i++) {
err = MCA_PML_CALL(recv(tmpbuf, totalcounts, dtype, i, char *tbuf;
err = MCA_PML_CALL(recv(buf, totalcounts, dtype, i,
MCA_COLL_BASE_TAG_REDUCE_SCATTER, comm, MCA_COLL_BASE_TAG_REDUCE_SCATTER, comm,
MPI_STATUS_IGNORE)); MPI_STATUS_IGNORE));
if (MPI_SUCCESS != err) { if (MPI_SUCCESS != err) {
@ -196,7 +196,9 @@ mca_coll_basic_reduce_scatter_block_inter(const void *sbuf, void *rbuf, int rcou
} }
/* Perform the reduction */ /* Perform the reduction */
ompi_op_reduce(op, tmpbuf, tmpbuf2, totalcounts, dtype); ompi_op_reduce(op, lbuf, buf, totalcounts, dtype);
/* swap the buffers */
tbuf = lbuf; lbuf = buf; buf = tbuf;
} }
} else { } else {
/* If not root, send data to the root. */ /* If not root, send data to the root. */
@ -209,7 +211,7 @@ mca_coll_basic_reduce_scatter_block_inter(const void *sbuf, void *rbuf, int rcou
} }
/* Now do a scatterv on the local communicator */ /* Now do a scatterv on the local communicator */
err = comm->c_local_comm->c_coll.coll_scatter(tmpbuf2, rcount, dtype, err = comm->c_local_comm->c_coll.coll_scatter(lbuf, rcount, dtype,
rbuf, rcount, dtype, 0, rbuf, rcount, dtype, 0,
comm->c_local_comm, comm->c_local_comm,
comm->c_local_comm->c_coll.coll_scatter_module); comm->c_local_comm->c_coll.coll_scatter_module);