1 /* Copyright (c) 2013-2014. The SimGrid Team.
2 * All rights reserved. */
4 /* This program is free software; you can redistribute it and/or modify it
5 * under the terms of the license (GNU LGPL) which comes with this package. */
7 #include "colls_private.h"
9 * Linear functions are copied from the basic coll module. For
10 * some small number of nodes and/or small data sizes they are just as
11 * fast as tuned/tree based segmenting operations and as such may be
12 * selected by the decision functions. These are copied into this module
13 * due to the way we select modules in V1. i.e. in V2 we will handle this
14 * differently and so will not have to duplicate code.
15 * GEF Oct05 after asking Jeff.
18 smpi_coll_tuned_alltoallv_ompi_basic_linear(void *sbuf, int *scounts, int *sdisps,
20 void *rbuf, int *rcounts, int *rdisps,
29 size = smpi_comm_size(comm);
30 rank = smpi_comm_rank(comm);
31 MPI_Request *ireqs= xbt_malloc(sizeof(MPI_Request) * size * 2);
33 "coll:tuned:alltoallv_intra_basic_linear rank %d", rank);
35 sext=smpi_datatype_get_extent(sdtype);
36 rext=smpi_datatype_get_extent(rdtype);
38 /* Simple optimization - handle send to self first */
39 psnd = ((char *) sbuf) + (sdisps[rank] * sext);
40 prcv = ((char *) rbuf) + (rdisps[rank] * rext);
41 if (0 != scounts[rank]) {
42 smpi_datatype_copy(psnd, scounts[rank], sdtype,
43 prcv, rcounts[rank], rdtype);
46 /* If only one process, we're done. */
51 /* Now, initiate all send/recv to/from others. */
55 /* Post all receives first */
56 for (i = 0; i < size; ++i) {
57 if (i == rank || 0 == rcounts[i]) {
61 prcv = ((char *) rbuf) + (rdisps[i] * rext);
63 *preq = smpi_irecv_init(prcv, rcounts[i], rdtype,
64 i, COLL_TAG_ALLTOALLV, comm
71 /* Now post all sends */
72 for (i = 0; i < size; ++i) {
73 if (i == rank || 0 == scounts[i]) {
77 psnd = ((char *) sbuf) + (sdisps[i] * sext);
78 *preq=smpi_isend_init(psnd, scounts[i], sdtype,
79 i, COLL_TAG_ALLTOALLV, comm
85 /* Start your engines. This will never return an error. */
86 smpi_mpi_startall(nreqs, ireqs);
88 /* Wait for them all. If there's an error, note that we don't care
89 * what the error was -- just that there *was* an error. The PML
90 * will finish all requests, even if one or more of them fail.
91 * i.e., by the end of this call, all the requests are free-able.
92 * So free them anyway -- even if there was an error, and return the
93 * error after we free everything. */
94 smpi_mpi_waitall(nreqs, ireqs,
97 /* Free the requests. */
98 for (i = 0; i < nreqs; ++i) {
99 if(ireqs[i]!=MPI_REQUEST_NULL)smpi_mpi_request_free(&ireqs[i]);