-/* Copyright (c) 2013-2017. The SimGrid Team.
+/* Copyright (c) 2013-2019. The SimGrid Team.
* All rights reserved. */
/* This program is free software; you can redistribute it and/or modify it
* Additional copyrights may follow
*/
-#include "../colls_private.h"
-#include "../coll_tuned_topo.h"
-
+#include "../coll_tuned_topo.hpp"
+#include "../colls_private.hpp"
/*
- * Barrier is ment to be a synchronous operation, as some BTLs can mark
- * a request done before its passed to the NIC and progress might not be made
- * elsewhere we cannot allow a process to exit the barrier until its last
+ * Barrier is ment to be a synchronous operation, as some BTLs can mark
+ * a request done before its passed to the NIC and progress might not be made
+ * elsewhere we cannot allow a process to exit the barrier until its last
* [round of] sends are completed.
*
- * It is last round of sends rather than 'last' individual send as each pair of
- * peers can use different channels/devices/btls and the receiver of one of
+ * It is last round of sends rather than 'last' individual send as each pair of
+ * peers can use different channels/devices/btls and the receiver of one of
* these sends might be forced to wait as the sender
- * leaves the collective and does not make progress until the next mpi call
+ * leaves the collective and does not make progress until the next mpi call
*
*/
/*
* Simple double ring version of barrier
*
- * synchronous gurantee made by last ring of sends are synchronous
+ * synchronous guarantee made by last ring of sends are synchronous
*
*/
-namespace simgrid{
-namespace smpi{
-int Coll_barrier_ompi_doublering::barrier(MPI_Comm comm
- )
+namespace simgrid {
+namespace smpi {
+int barrier__ompi_doublering(MPI_Comm comm)
{
int rank, size;
int left, right;
right = ((rank+1)%size);
if (rank > 0) { /* receive message from the left */
- Request::recv((void*)NULL, 0, MPI_BYTE, left,
+ Request::recv((void*)NULL, 0, MPI_BYTE, left,
COLL_TAG_BARRIER, comm,
MPI_STATUS_IGNORE);
}
/* Send message to the right */
- Request::send((void*)NULL, 0, MPI_BYTE, right,
+ Request::send((void*)NULL, 0, MPI_BYTE, right,
COLL_TAG_BARRIER,
comm);
/* root needs to receive from the last node */
if (rank == 0) {
- Request::recv((void*)NULL, 0, MPI_BYTE, left,
+ Request::recv((void*)NULL, 0, MPI_BYTE, left,
COLL_TAG_BARRIER, comm,
MPI_STATUS_IGNORE);
}
/* Allow nodes to exit */
if (rank > 0) { /* post Receive from left */
- Request::recv((void*)NULL, 0, MPI_BYTE, left,
+ Request::recv((void*)NULL, 0, MPI_BYTE, left,
COLL_TAG_BARRIER, comm,
MPI_STATUS_IGNORE);
}
/* send message to the right one */
- Request::send((void*)NULL, 0, MPI_BYTE, right,
+ Request::send((void*)NULL, 0, MPI_BYTE, right,
COLL_TAG_BARRIER,
comm);
-
+
/* rank 0 post receive from the last node */
if (rank == 0) {
- Request::recv((void*)NULL, 0, MPI_BYTE, left,
+ Request::recv((void*)NULL, 0, MPI_BYTE, left,
COLL_TAG_BARRIER, comm,
MPI_STATUS_IGNORE);
}
* To make synchronous, uses sync sends and sync sendrecvs
*/
-int Coll_barrier_ompi_recursivedoubling::barrier(MPI_Comm comm
- )
+int barrier__ompi_recursivedoubling(MPI_Comm comm)
{
int rank, size, adjsize;
int mask, remote;
rank = comm->rank();
size = comm->size();
XBT_DEBUG(
- "ompi_coll_tuned_barrier_ompi_recursivedoubling rank %d",
+ "ompi_coll_tuned_barrier_ompi_recursivedoubling rank %d",
rank);
/* do nearest power of 2 less than size calc */
if (rank < (size - adjsize)) {
/* send enter message to higher ranked node */
remote = rank + adjsize;
- Request::send((void*)NULL, 0, MPI_BYTE, remote,
+ Request::send((void*)NULL, 0, MPI_BYTE, remote,
COLL_TAG_BARRIER,
comm);
* To make synchronous, uses sync sends and sync sendrecvs
*/
-int Coll_barrier_ompi_bruck::barrier(MPI_Comm comm
- )
+int barrier__ompi_bruck(MPI_Comm comm)
{
int rank, size;
int distance, to, from;
"ompi_coll_tuned_barrier_ompi_bruck rank %d", rank);
/* exchange data with rank-2^k and rank+2^k */
- for (distance = 1; distance < size; distance <<= 1) {
+ for (distance = 1; distance < size; distance <<= 1) {
from = (rank + size - distance) % size;
to = (rank + distance) % size;
/* send message to lower ranked node */
- Request::sendrecv(NULL, 0, MPI_BYTE, to,
+ Request::sendrecv(NULL, 0, MPI_BYTE, to,
COLL_TAG_BARRIER,
- NULL, 0, MPI_BYTE, from,
+ NULL, 0, MPI_BYTE, from,
COLL_TAG_BARRIER,
comm, MPI_STATUS_IGNORE);
}
* To make synchronous, uses sync sends and sync sendrecvs
*/
/* special case for two processes */
-int Coll_barrier_ompi_two_procs::barrier(MPI_Comm comm
- )
+int barrier__ompi_two_procs(MPI_Comm comm)
{
int remote;
"ompi_coll_tuned_barrier_ompi_two_procs rank %d", remote);
remote = (remote + 1) & 0x1;
- Request::sendrecv(NULL, 0, MPI_BYTE, remote,
+ Request::sendrecv(NULL, 0, MPI_BYTE, remote,
COLL_TAG_BARRIER,
- NULL, 0, MPI_BYTE, remote,
+ NULL, 0, MPI_BYTE, remote,
COLL_TAG_BARRIER,
comm, MPI_STATUS_IGNORE);
return (MPI_SUCCESS);
/* copied function (with appropriate renaming) starts here */
-int Coll_barrier_ompi_basic_linear::barrier(MPI_Comm comm)
+int barrier__ompi_basic_linear(MPI_Comm comm)
{
int i;
int size = comm->size();
/* All non-root send & receive zero-length message. */
if (rank > 0) {
- Request::send (NULL, 0, MPI_BYTE, 0,
+ Request::send (NULL, 0, MPI_BYTE, 0,
COLL_TAG_BARRIER,
comm);
- Request::recv (NULL, 0, MPI_BYTE, 0,
+ Request::recv (NULL, 0, MPI_BYTE, 0,
COLL_TAG_BARRIER,
comm, MPI_STATUS_IGNORE);
}
else {
MPI_Request* requests;
- requests = (MPI_Request*)malloc( size * sizeof(MPI_Request) );
+ requests = new MPI_Request[size];
for (i = 1; i < size; ++i) {
- requests[i] = Request::irecv(NULL, 0, MPI_BYTE, MPI_ANY_SOURCE,
- COLL_TAG_BARRIER, comm
- );
+ requests[i] = Request::irecv(NULL, 0, MPI_BYTE, i, COLL_TAG_BARRIER, comm);
}
Request::waitall( size-1, requests+1, MPI_STATUSES_IGNORE );
);
}
Request::waitall( size-1, requests+1, MPI_STATUSES_IGNORE );
- free( requests );
+ delete[] requests;
}
/* All done */
/*
* Another recursive doubling type algorithm, but in this case
- * we go up the tree and back down the tree.
+ * we go up the tree and back down the tree.
*/
-int Coll_barrier_ompi_tree::barrier(MPI_Comm comm)
+int barrier__ompi_tree(MPI_Comm comm)
{
int rank, size, depth;
int jump, partner;
rank = comm->rank();
size = comm->size();
XBT_DEBUG(
- "ompi_coll_tuned_barrier_ompi_tree %d",
+ "ompi_coll_tuned_barrier_ompi_tree %d",
rank);
/* Find the nearest power of 2 of the communicator size. */
partner = rank ^ jump;
if (!(partner & (jump-1)) && partner < size) {
if (partner > rank) {
- Request::recv (NULL, 0, MPI_BYTE, partner,
+ Request::recv (NULL, 0, MPI_BYTE, partner,
COLL_TAG_BARRIER, comm,
MPI_STATUS_IGNORE);
} else if (partner < rank) {
}
}
}
-
+
depth>>=1;
for (jump = depth; jump>0; jump>>=1) {
partner = rank ^ jump;
COLL_TAG_BARRIER,
comm);
} else if (partner < rank) {
- Request::recv (NULL, 0, MPI_BYTE, partner,
+ Request::recv (NULL, 0, MPI_BYTE, partner,
COLL_TAG_BARRIER, comm,
MPI_STATUS_IGNORE);
}
