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 /*****************************************************************************
11 Copyright (c) 2006, Ahmad Faraj & Xin Yuan,
14 Redistribution and use in source and binary forms, with or without
15 modification, are permitted provided that the following conditions are met:
17 * Redistributions of source code must retain the above copyright notice,
18 this list of conditions and the following disclaimer.
20 * Redistributions in binary form must reproduce the above copyright notice,
21 this list of conditions and the following disclaimer in the documentation
22 and/or other materials provided with the distribution.
24 * Neither the name of the Florida State University nor the names of its
25 contributors may be used to endorse or promote products derived from this
26 software without specific prior written permission.
28 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
29 ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
30 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
31 DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
32 ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
33 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
34 LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
35 ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
36 (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
37 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
39 *************************************************************************
40 * Any results obtained from executing this software require the *
41 * acknowledgment and citation of the software and its owners. *
42 * The full citation is given below: *
44 * A. Faraj and X. Yuan. "Automatic Generation and Tuning of MPI *
45 * Collective Communication Routines." The 19th ACM International *
46 * Conference on Supercomputing (ICS), Cambridge, Massachusetts, *
48 *************************************************************************
50 *****************************************************************************/
52 /*****************************************************************************
59 num: the number of processors in a communicator
63 * Descp: takes a number and tries to find a factoring of x, y mesh out of it
66 ****************************************************************************/
69 static int is_2dmesh(int num, int *i, int *j)
92 /*****************************************************************************
93 * Function: allgather_2dmesh_shoot
95 * send_buff: send input buffer
96 * send_count: number of elements to send
97 * send_type: data type of elements being sent
98 * recv_buff: receive output buffer
99 * recv_count: number of elements to received
100 * recv_type: data type of elements being received
101 * comm: communication
102 * Descrp: Function realizes the allgather operation using the 2dmesh
103 * algorithm. Allgather ommunication occurs first in the x dimension then in
104 * the y dimension. The communication in each dimension follows
106 * Auther: Ahmad Faraj
107 ****************************************************************************/
109 smpi_coll_tuned_allgather_2dmesh(void *send_buff, int send_count, MPI_Datatype
110 send_type, void *recv_buff, int recv_count,
111 MPI_Datatype recv_type, MPI_Comm comm)
113 MPI_Request *req, *req_ptr;
116 int i, src, dst, rank, num_procs;
117 int X, Y, send_offset, recv_offset;
118 int my_row_base, my_col_base, src_row_base, block_size, num_reqs;
119 int tag = COLL_TAG_ALLGATHER;
121 rank = smpi_comm_rank(comm);
122 num_procs = smpi_comm_size(comm);
124 extent = smpi_datatype_get_extent(send_type);
126 block_size = extent * send_count;
128 if (!is_2dmesh(num_procs, &X, &Y))
129 THROWF(arg_error,0, "allgather_2dmesh algorithm can't be used with this number of processes! ");
131 my_row_base = (rank / Y) * Y;
132 my_col_base = rank % Y;
138 req = (MPI_Request *) xbt_malloc(num_reqs * sizeof(MPI_Request));
142 // do local allgather/local copy
143 recv_offset = rank * block_size;
144 smpi_datatype_copy(send_buff, send_count, send_type, (char *)recv_buff + recv_offset,
145 recv_count, recv_type);
148 for (i = 0; i < Y; i++) {
149 src = i + my_row_base;
152 recv_offset = src * block_size;
153 *(req_ptr++) = smpi_mpi_irecv((char *)recv_buff + recv_offset, recv_count, recv_type, src, tag,
158 for (i = 0; i < Y; i++) {
159 dst = i + my_row_base;
162 smpi_mpi_send(send_buff, send_count, send_type, dst, tag, comm);
165 smpi_mpi_waitall(Y - 1, req, MPI_STATUSES_IGNORE);
170 for (i = 0; i < X; i++) {
171 src = (i * Y + my_col_base);
174 src_row_base = (src / Y) * Y;
175 recv_offset = src_row_base * block_size;
176 *(req_ptr++) = smpi_mpi_irecv((char *)recv_buff + recv_offset, recv_count * Y, recv_type, src, tag,
180 for (i = 0; i < X; i++) {
181 dst = (i * Y + my_col_base);
184 send_offset = my_row_base * block_size;
185 smpi_mpi_send((char *)recv_buff + send_offset, send_count * Y, send_type, dst, tag,
189 smpi_mpi_waitall(X - 1, req, MPI_STATUSES_IGNORE);