-/* Copyright (c) 2013-2019. The SimGrid Team.
+/* Copyright (c) 2013-2020. The SimGrid Team.
* All rights reserved. */
/* This program is free software; you can redistribute it and/or modify it
* [02b] [12b] [22b]
*
* COMPUTATION PHASE 0 (a)
- * Step 0: rank r sends block ra to rank (r+1) and receives bloc (r-1)a
+ * Step 0: rank r sends block ra to rank (r+1) and receives block (r-1)a
* from rank (r-1) [with wraparound].
* # 0 1 2
* [00a] [00a+10a] [20a]
* [22a+02a] [12a] [22a]
* [02b] [12b] [22b]
*
- * Step 1: rank r sends block (r-1)a to rank (r+1) and receives bloc
+ * Step 1: rank r sends block (r-1)a to rank (r+1) and receives block
* (r-2)a from rank (r-1) [with wraparound].
* # 0 1 2
* [00a] [00a+10a] [00a+10a+20a]
* [02b] [12b] [22b]
*
* COMPUTATION PHASE 1 (b)
- * Step 0: rank r sends block rb to rank (r+1) and receives bloc (r-1)b
+ * Step 0: rank r sends block rb to rank (r+1) and receives block (r-1)b
* from rank (r-1) [with wraparound].
* # 0 1 2
* [00a] [00a+10a] [20a]
* [22a+02a] [12a] [22a]
* [22b+02b] [12b] [22b]
*
- * Step 1: rank r sends block (r-1)b to rank (r+1) and receives bloc
+ * Step 1: rank r sends block (r-1)b to rank (r+1) and receives block
* (r-2)b from rank (r-1) [with wraparound].
* # 0 1 2
* [00a] [00a+10a] [00a+10a+20a]
#include "../colls_private.hpp"
-namespace simgrid{
-namespace smpi{
-int
-Coll_allreduce_ompi_ring_segmented::allreduce(const void *sbuf, void *rbuf, int count,
- MPI_Datatype dtype,
- MPI_Op op,
- MPI_Comm comm)
+namespace simgrid {
+namespace smpi {
+int allreduce__ompi_ring_segmented(const void *sbuf, void *rbuf, int count,
+ MPI_Datatype dtype,
+ MPI_Op op,
+ MPI_Comm comm)
{
int ret = MPI_SUCCESS;
int line;
int block_count;
unsigned int inbi;
size_t typelng;
- char *tmpsend = NULL, *tmprecv = NULL;
+ char *tmpsend = nullptr, *tmprecv = nullptr;
unsigned char* inbuf[2] = {nullptr, nullptr};
ptrdiff_t true_extent, extent;
ptrdiff_t block_offset, max_real_segsize;
- MPI_Request reqs[2] = {NULL, NULL};
+ MPI_Request reqs[2] = {nullptr, nullptr};
const size_t segsize = 1 << 20; /* 1 MB */
int size = comm->size();
int rank = comm->rank();
/* Special case for count less than size * segcount - use regular ring */
if (count < size * segcount) {
XBT_DEBUG( "coll:tuned:allreduce_ring_segmented rank %d/%d, count %d, switching to regular ring", rank, size, count);
- return (Coll_allreduce_lr::allreduce(sbuf, rbuf, count, dtype, op,
- comm));
+ return (allreduce__lr(sbuf, rbuf, count, dtype, op, comm));
}
/* Determine the number of phases of the algorithm */
/* Allocate and initialize temporary buffers */
inbuf[0] = smpi_get_tmp_sendbuffer(max_real_segsize);
- if (NULL == inbuf[0]) { ret = -1; line = __LINE__; goto error_hndl; }
+ if (nullptr == inbuf[0]) {
+ ret = -1;
+ line = __LINE__;
+ goto error_hndl;
+ }
if (size > 2) {
inbuf[1] = smpi_get_tmp_recvbuffer(max_real_segsize);
if (nullptr == inbuf[1]) {
- wait on block (r + 1)
- compute on block (r + 1)
- send block (r + 1) to rank (r + 1)
- Note that we must be careful when computing the begining of buffers and
+ Note that we must be careful when computing the beginning of buffers and
for send operations and computation we must compute the exact block size.
*/
send_to = (rank + 1) % size;