1 /* Copyright (c) 2013-2019. 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.hpp"
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.
19 int alltoallv__ompi_basic_linear(const void *sbuf, const int *scounts, const int *sdisps,
21 void *rbuf, const int *rcounts, const int *rdisps,
32 MPI_Request* ireqs = new MPI_Request[size * 2];
34 "coll:tuned:alltoallv_intra_basic_linear rank %d", rank);
36 sext=sdtype->get_extent();
37 rext=rdtype->get_extent();
39 /* Simple optimization - handle send to self first */
40 psnd = ((char *) sbuf) + (sdisps[rank] * sext);
41 prcv = ((char *) rbuf) + (rdisps[rank] * rext);
42 if (0 != scounts[rank]) {
43 Datatype::copy(psnd, scounts[rank], sdtype,
44 prcv, rcounts[rank], rdtype);
47 /* If only one process, we're done. */
52 /* Now, initiate all send/recv to/from others. */
56 /* Post all receives first */
57 for (i = 0; i < size; ++i) {
62 prcv = ((char *) rbuf) + (rdisps[i] * rext);
64 *preq = Request::irecv_init(prcv, rcounts[i], rdtype,
65 i, COLL_TAG_ALLTOALLV, comm
72 /* Now post all sends */
73 for (i = 0; i < size; ++i) {
78 psnd = ((char *) sbuf) + (sdisps[i] * sext);
79 *preq=Request::isend_init(psnd, scounts[i], sdtype,
80 i, COLL_TAG_ALLTOALLV, comm
86 /* Start your engines. This will never return an error. */
87 Request::startall(nreqs, ireqs);
89 /* Wait for them all. If there's an error, note that we don't care
90 * what the error was -- just that there *was* an error. The PML
91 * will finish all requests, even if one or more of them fail.
92 * i.e., by the end of this call, all the requests are free-able.
93 * So free them anyway -- even if there was an error, and return the
94 * error after we free everything. */
95 Request::waitall(nreqs, ireqs,
98 /* Free the requests. */
99 for (i = 0; i < nreqs; ++i) {
100 if(ireqs[i]!=MPI_REQUEST_NULL)
101 Request::unref(&ireqs[i]);