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 // now only work with power of two processes
12 smpi_coll_tuned_allgather_rhv(void *sbuf, int send_count,
13 MPI_Datatype send_type, void *rbuf,
14 int recv_count, MPI_Datatype recv_type,
18 MPI_Aint s_extent, r_extent;
20 // local int variables
21 int i, dst, send_base_offset, recv_base_offset, send_chunk, recv_chunk,
22 send_offset, recv_offset;
23 int tag = COLL_TAG_ALLGATHER;
27 // get size of the communicator, followed by rank
28 unsigned int num_procs = smpi_comm_size(comm);
30 if((num_procs&(num_procs-1)))
31 THROWF(arg_error,0, "allgather rhv algorithm can't be used with non power of two number of processes ! ");
33 unsigned int rank = smpi_comm_rank(comm);
35 // get size of single element's type for send buffer and recv buffer
36 s_extent = smpi_datatype_get_extent(send_type);
37 r_extent = smpi_datatype_get_extent(recv_type);
39 // multiply size of each element by number of elements to send or recv
40 send_chunk = s_extent * send_count;
41 recv_chunk = r_extent * recv_count;
43 if (send_chunk != recv_chunk) {
44 XBT_WARN("MPI_allgather_rhv use default MPI_allgather.");
45 smpi_mpi_allgather(sbuf, send_count, send_type, rbuf, recv_count,
50 // compute starting offset location to perform local copy
51 int size = num_procs / 2;
54 while (mask < num_procs) {
62 // printf("node %d base_offset %d\n",rank,base_offset);
64 //perform a remote copy
67 smpi_mpi_sendrecv(sbuf, send_count, send_type, dst, tag,
68 (char *)rbuf + base_offset * recv_chunk, recv_count, recv_type, dst, tag,
75 curr_count = recv_count;
77 // destination pair for both send and recv
81 send_base_offset = base_offset;
83 recv_base_offset = base_offset - i;
86 recv_base_offset = base_offset + i;
88 send_offset = send_base_offset * recv_chunk;
89 recv_offset = recv_base_offset * recv_chunk;
91 // printf("node %d send to %d in phase %d s_offset = %d r_offset = %d count = %d\n",rank,dst,phase, send_base_offset, recv_base_offset, curr_count);
93 smpi_mpi_sendrecv((char *)rbuf + send_offset, curr_count, recv_type, dst, tag,
94 (char *)rbuf + recv_offset, curr_count, recv_type, dst, tag,