d377a6b6f4
This code is the implementation of Software-base Performance Counters as described in the paper 'Using Software-Base Performance Counters to Expose Low-Level Open MPI Performance Information' in EuroMPI/USA '17 (http://icl.cs.utk.edu/news_pub/submissions/software-performance-counters.pdf). More practical usage information can be found here: https://github.com/davideberius/ompi/wiki/How-to-Use-Software-Based-Performance-Counters-(SPCs)-in-Open-MPI. All software events functions are put in macros that become no-ops when SOFTWARE_EVENTS_ENABLE is not defined. The internal timer units have been changed to cycles to avoid division operations which was a large source of overhead as discussed in the paper. Added a --with-spc configure option to enable SPCs in the Open MPI build. This defines SOFTWARE_EVENTS_ENABLE. Added an MCA parameter, mpi_spc_enable, for turning on specific counters. Added an MCA parameter, mpi_spc_dump_enabled, for turning on and off dumping SPC counters in MPI_Finalize. Added an SPC test and example. Signed-off-by: David Eberius <deberius@vols.utk.edu>
143 строки
4.9 KiB
C
143 строки
4.9 KiB
C
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
|
|
/*
|
|
* Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana
|
|
* University Research and Technology
|
|
* Corporation. All rights reserved.
|
|
* Copyright (c) 2004-2018 The University of Tennessee and The University
|
|
* of Tennessee Research Foundation. All rights
|
|
* reserved.
|
|
* Copyright (c) 2004-2008 High Performance Computing Center Stuttgart,
|
|
* University of Stuttgart. All rights reserved.
|
|
* Copyright (c) 2004-2005 The Regents of the University of California.
|
|
* All rights reserved.
|
|
* Copyright (c) 2006-2012 Cisco Systems, Inc. All rights reserved.
|
|
* Copyright (c) 2012-2013 Los Alamos National Security, LLC. All rights
|
|
* reserved.
|
|
* Copyright (c) 2015 Research Organization for Information Science
|
|
* and Technology (RIST). All rights reserved.
|
|
* Copyright (c) 2016 IBM Corporation. All rights reserved.
|
|
* $COPYRIGHT$
|
|
*
|
|
* Additional copyrights may follow
|
|
*
|
|
* $HEADER$
|
|
*/
|
|
#include "ompi_config.h"
|
|
#include <stdio.h>
|
|
|
|
#include "ompi/mpi/c/bindings.h"
|
|
#include "ompi/runtime/params.h"
|
|
#include "ompi/communicator/communicator.h"
|
|
#include "ompi/errhandler/errhandler.h"
|
|
#include "ompi/datatype/ompi_datatype.h"
|
|
#include "ompi/op/op.h"
|
|
#include "ompi/memchecker.h"
|
|
#include "ompi/runtime/ompi_spc.h"
|
|
|
|
#if OMPI_BUILD_MPI_PROFILING
|
|
#if OPAL_HAVE_WEAK_SYMBOLS
|
|
#pragma weak MPI_Ireduce_scatter = PMPI_Ireduce_scatter
|
|
#endif
|
|
#define MPI_Ireduce_scatter PMPI_Ireduce_scatter
|
|
#endif
|
|
|
|
static const char FUNC_NAME[] = "MPI_Ireduce_scatter";
|
|
|
|
|
|
int MPI_Ireduce_scatter(const void *sendbuf, void *recvbuf, const int recvcounts[],
|
|
MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request *request)
|
|
{
|
|
int i, err, size, count;
|
|
|
|
SPC_RECORD(OMPI_SPC_IREDUCE_SCATTER, 1);
|
|
|
|
MEMCHECKER(
|
|
int rank;
|
|
int count;
|
|
|
|
size = ompi_comm_size(comm);
|
|
rank = ompi_comm_rank(comm);
|
|
for (count = i = 0; i < size; ++i) {
|
|
if (0 == recvcounts[i]) {
|
|
count += recvcounts[i];
|
|
}
|
|
}
|
|
|
|
memchecker_comm(comm);
|
|
memchecker_datatype(datatype);
|
|
|
|
/* check receive buffer of current proccess, whether it's addressable. */
|
|
memchecker_call(&opal_memchecker_base_isaddressable, recvbuf,
|
|
recvcounts[rank], datatype);
|
|
|
|
/* check whether the actual send buffer is defined. */
|
|
if(MPI_IN_PLACE == sendbuf) {
|
|
memchecker_call(&opal_memchecker_base_isdefined, recvbuf, count, datatype);
|
|
} else {
|
|
memchecker_call(&opal_memchecker_base_isdefined, sendbuf, count, datatype);
|
|
|
|
}
|
|
);
|
|
|
|
if (MPI_PARAM_CHECK) {
|
|
char *msg;
|
|
err = MPI_SUCCESS;
|
|
OMPI_ERR_INIT_FINALIZE(FUNC_NAME);
|
|
if (ompi_comm_invalid(comm)) {
|
|
return OMPI_ERRHANDLER_INVOKE(MPI_COMM_WORLD, MPI_ERR_COMM,
|
|
FUNC_NAME);
|
|
}
|
|
|
|
/* Unrooted operation; same checks for all ranks on both
|
|
intracommunicators and intercommunicators */
|
|
|
|
else if (MPI_OP_NULL == op || NULL == op) {
|
|
err = MPI_ERR_OP;
|
|
} else if (!ompi_op_is_valid(op, datatype, &msg, FUNC_NAME)) {
|
|
int ret = OMPI_ERRHANDLER_INVOKE(comm, MPI_ERR_OP, msg);
|
|
free(msg);
|
|
return ret;
|
|
} else if (NULL == recvcounts) {
|
|
err = MPI_ERR_COUNT;
|
|
} else if (MPI_IN_PLACE == recvbuf) {
|
|
err = MPI_ERR_ARG;
|
|
}
|
|
OMPI_ERRHANDLER_CHECK(err, comm, err, FUNC_NAME);
|
|
|
|
/* Based on the standard each group has to provide the same total
|
|
number of elements, so the size of the recvcounts array depends
|
|
on the number of participants in the local group. */
|
|
size = ompi_comm_size(comm);
|
|
for (i = 0; i < size; ++i) {
|
|
OMPI_CHECK_DATATYPE_FOR_SEND(err, datatype, recvcounts[i]);
|
|
OMPI_ERRHANDLER_CHECK(err, comm, err, FUNC_NAME);
|
|
}
|
|
}
|
|
|
|
/* MPI standard says that reductions have to have a count of at least 1,
|
|
* but some benchmarks (e.g., IMB) calls this function with a count of 0.
|
|
* So handle that case.
|
|
*/
|
|
size = ompi_comm_size(comm);
|
|
for (count = i = 0; i < size; ++i) {
|
|
if (0 == recvcounts[i]) {
|
|
++count;
|
|
}
|
|
}
|
|
if (size == count) {
|
|
*request = &ompi_request_empty;
|
|
return MPI_SUCCESS;
|
|
}
|
|
|
|
OPAL_CR_ENTER_LIBRARY();
|
|
|
|
/* Invoke the coll component to perform the back-end operation */
|
|
|
|
OBJ_RETAIN(op);
|
|
err = comm->c_coll->coll_ireduce_scatter(sendbuf, recvbuf, recvcounts,
|
|
datatype, op, comm, request,
|
|
comm->c_coll->coll_ireduce_scatter_module);
|
|
OBJ_RELEASE(op);
|
|
OMPI_ERRHANDLER_RETURN(err, comm, err, FUNC_NAME);
|
|
}
|