1 /* Copyright (c) 2013-2023. 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"
9 namespace simgrid::smpi {
12 allgather__rdb(const void *sbuf, int send_count,
13 MPI_Datatype send_type, void *rbuf,
14 int recv_count, MPI_Datatype recv_type,
19 MPI_Aint send_chunk, recv_chunk;
21 // local int variables
22 unsigned int i, j, k, dst, send_offset, recv_offset, tree_root;
23 int dst_tree_root, rank_tree_root, last_recv_count = 0, num_procs_completed;
25 int tag = COLL_TAG_ALLGATHER;
26 unsigned int mask = 1;
28 int curr_count = recv_count;
30 // local string variables
31 char *send_ptr = (char *) sbuf;
32 char *recv_ptr = (char *) rbuf;
34 // get size of the communicator, followed by rank
35 unsigned int num_procs = comm->size();
36 unsigned int rank = comm->rank();
38 // get size of single element's type for send buffer and recv buffer
39 send_chunk = send_type->get_extent();
40 recv_chunk = recv_type->get_extent();
42 // multiply size of each element by number of elements to send or recv
43 send_chunk *= send_count;
44 recv_chunk *= recv_count;
46 // perform a local copy
47 Request::sendrecv(send_ptr, send_count, send_type, rank, tag,
48 recv_ptr + rank * recv_chunk, recv_count, recv_type, rank, tag,
52 while (mask < num_procs) {
54 dst_tree_root = dst >> i;
56 rank_tree_root = rank >> i;
58 send_offset = rank_tree_root * send_chunk;
59 recv_offset = dst_tree_root * recv_chunk;
61 if (dst < num_procs) {
62 Request::sendrecv(recv_ptr + send_offset, curr_count, send_type, dst,
63 tag, recv_ptr + recv_offset, mask * recv_count,
64 recv_type, dst, tag, comm, &status);
65 last_recv_count = Status::get_count(&status, recv_type);
66 curr_count += last_recv_count;
69 if (dst_tree_root + mask > num_procs) {
70 num_procs_completed = num_procs - rank_tree_root - mask;
71 /* num_procs_completed is the number of processes in this
72 subtree that have all the data. Send data to others
73 in a tree fashion. First find root of current tree
74 that is being divided into two. k is the number of
75 least-significant bits in this process's rank that
76 must be zeroed out to find the rank of the root */
86 offset = recv_chunk * (rank_tree_root + mask);
90 dst = rank ^ tmp_mask;
92 tree_root = rank >> k;
95 /* send only if this proc has data and destination
96 doesn't have data. at any step, multiple processes
97 can send if they have the data */
99 && (rank < tree_root + num_procs_completed)
100 && (dst >= tree_root + num_procs_completed)) {
101 Request::send(recv_ptr + offset, last_recv_count, recv_type, dst,
104 /* last_recv_cnt was set in the previous
105 receive. that's the amount of data to be
108 /* recv only if this proc. doesn't have data and sender
110 else if ((dst < rank)
111 && (dst < tree_root + num_procs_completed)
112 && (rank >= tree_root + num_procs_completed)) {
113 Request::recv(recv_ptr + offset,
114 recv_count * num_procs_completed,
115 recv_type, dst, tag, comm, &status);
116 // num_procs_completed is also equal to the no. of processes
117 // whose data we don't have
118 last_recv_count = Status::get_count(&status, recv_type);
119 curr_count += last_recv_count;
133 } // namespace simgrid::smpi