1 /* Copyright (c) 2013-2017. The SimGrid Team. All rights reserved. */
3 /* This program is free software; you can redistribute it and/or modify it
4 * under the terms of the license (GNU LGPL) which comes with this package. */
6 #include "../colls_private.hpp"
7 #include "smpi_status.hpp"
13 Coll_allgather_rdb::allgather(void *sbuf, int send_count,
14 MPI_Datatype send_type, void *rbuf,
15 int recv_count, MPI_Datatype recv_type,
20 MPI_Aint send_chunk, recv_chunk;
22 // local int variables
23 unsigned int i, j, k, dst, send_offset, recv_offset, tree_root;
24 int dst_tree_root, rank_tree_root, last_recv_count = 0, num_procs_completed;
26 int tag = COLL_TAG_ALLGATHER;
27 unsigned int mask = 1;
29 int curr_count = recv_count;
31 // local string variables
32 char *send_ptr = (char *) sbuf;
33 char *recv_ptr = (char *) rbuf;
35 // get size of the communicator, followed by rank
36 unsigned int num_procs = comm->size();
37 unsigned int rank = comm->rank();
39 // get size of single element's type for send buffer and recv buffer
40 send_chunk = send_type->get_extent();
41 recv_chunk = recv_type->get_extent();
43 // multiply size of each element by number of elements to send or recv
44 send_chunk *= send_count;
45 recv_chunk *= recv_count;
47 // perform a local copy
48 Request::sendrecv(send_ptr, send_count, send_type, rank, tag,
49 recv_ptr + rank * recv_chunk, recv_count, recv_type, rank, tag,
53 while (mask < num_procs) {
55 dst_tree_root = dst >> i;
57 rank_tree_root = rank >> i;
59 send_offset = rank_tree_root * send_chunk;
60 recv_offset = dst_tree_root * recv_chunk;
62 if (dst < num_procs) {
63 Request::sendrecv(recv_ptr + send_offset, curr_count, send_type, dst,
64 tag, recv_ptr + recv_offset, mask * recv_count,
65 recv_type, dst, tag, comm, &status);
66 last_recv_count = Status::get_count(&status, recv_type);
67 curr_count += last_recv_count;
70 if (dst_tree_root + mask > num_procs) {
71 num_procs_completed = num_procs - rank_tree_root - mask;
72 /* num_procs_completed is the number of processes in this
73 subtree that have all the data. Send data to others
74 in a tree fashion. First find root of current tree
75 that is being divided into two. k is the number of
76 least-significant bits in this process's rank that
77 must be zeroed out to find the rank of the root */
87 offset = recv_chunk * (rank_tree_root + mask);
91 dst = rank ^ tmp_mask;
93 tree_root = rank >> k;
96 /* send only if this proc has data and destination
97 doesn't have data. at any step, multiple processes
98 can send if they have the data */
100 && (rank < tree_root + num_procs_completed)
101 && (dst >= tree_root + num_procs_completed)) {
102 Request::send(recv_ptr + offset, last_recv_count, recv_type, dst,
105 /* last_recv_cnt was set in the previous
106 receive. that's the amount of data to be
109 /* recv only if this proc. doesn't have data and sender
111 else if ((dst < rank)
112 && (dst < tree_root + num_procs_completed)
113 && (rank >= tree_root + num_procs_completed)) {
114 Request::recv(recv_ptr + offset,
115 recv_count * num_procs_completed,
116 recv_type, dst, tag, comm, &status);
117 // num_procs_completed is also equal to the no. of processes
118 // whose data we don't have
119 last_recv_count = Status::get_count(&status, recv_type);
120 curr_count += last_recv_count;