6c5532072a
OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
239 строки
10 KiB
C
239 строки
10 KiB
C
/*
|
|
* Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana
|
|
* University Research and Technology
|
|
* Corporation. All rights reserved.
|
|
* Copyright (c) 2004-2005 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) 2009 Sun Microsystmes, Inc. 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/win/win.h"
|
|
#include "ompi/mca/osc/osc.h"
|
|
#include "ompi/op/op.h"
|
|
#include "ompi/datatype/ompi_datatype.h"
|
|
#include "ompi/datatype/ompi_datatype_internal.h"
|
|
#include "ompi/memchecker.h"
|
|
|
|
#if OPAL_HAVE_WEAK_SYMBOLS && OMPI_PROFILING_DEFINES
|
|
#pragma weak MPI_Accumulate = PMPI_Accumulate
|
|
#endif
|
|
|
|
#if OMPI_PROFILING_DEFINES
|
|
#include "ompi/mpi/c/profile/defines.h"
|
|
#endif
|
|
|
|
static const char FUNC_NAME[] = "MPI_Accumlate";
|
|
|
|
int MPI_Accumulate(void *origin_addr, int origin_count, MPI_Datatype origin_datatype,
|
|
int target_rank, MPI_Aint target_disp, int target_count,
|
|
MPI_Datatype target_datatype, MPI_Op op, MPI_Win win)
|
|
{
|
|
int rc;
|
|
ompi_win_t *ompi_win = (ompi_win_t*) win;
|
|
|
|
MEMCHECKER(
|
|
memchecker_datatype(origin_datatype);
|
|
memchecker_datatype(target_datatype);
|
|
memchecker_call(&opal_memchecker_base_isdefined, origin_addr, origin_count, origin_datatype);
|
|
);
|
|
|
|
if (MPI_PARAM_CHECK) {
|
|
rc = OMPI_SUCCESS;
|
|
|
|
OMPI_ERR_INIT_FINALIZE(FUNC_NAME);
|
|
|
|
if (ompi_win_invalid(win)) {
|
|
return OMPI_ERRHANDLER_INVOKE(MPI_COMM_WORLD, MPI_ERR_WIN, FUNC_NAME);
|
|
} else if (origin_count < 0 || target_count < 0) {
|
|
rc = MPI_ERR_COUNT;
|
|
} else if (ompi_win_peer_invalid(win, target_rank) &&
|
|
(MPI_PROC_NULL != target_rank)) {
|
|
rc = MPI_ERR_RANK;
|
|
} else if (MPI_OP_NULL == op) {
|
|
rc = MPI_ERR_OP;
|
|
} else if (!ompi_op_is_intrinsic(op)) {
|
|
rc = MPI_ERR_OP;
|
|
} else if (!ompi_win_comm_allowed(win)) {
|
|
rc = MPI_ERR_RMA_SYNC;
|
|
} else if ( target_disp < 0 ) {
|
|
rc = MPI_ERR_DISP;
|
|
} else {
|
|
OMPI_CHECK_DATATYPE_FOR_ONE_SIDED(rc, origin_datatype, origin_count);
|
|
if (OMPI_SUCCESS == rc) {
|
|
OMPI_CHECK_DATATYPE_FOR_ONE_SIDED(rc, target_datatype, target_count);
|
|
}
|
|
if (OMPI_SUCCESS == rc) {
|
|
/* While technically the standard probably requires that the
|
|
datatypes used with MPI_REPLACE conform to all the rules
|
|
for other reduction operators, we don't require such
|
|
behaivor, as checking for it is expensive here and we don't
|
|
care in implementation.. */
|
|
if (op != &ompi_mpi_op_replace.op) {
|
|
ompi_datatype_t *op_check_dt, *origin_check_dt;
|
|
char *msg;
|
|
|
|
if (opal_datatype_is_predefined(&(origin_datatype->super))) {
|
|
origin_check_dt = origin_datatype;
|
|
} else {
|
|
int i, found_index = -1, num_found = 0;
|
|
uint32_t mask = 1;
|
|
|
|
for (i = 0 ; i < OMPI_DATATYPE_MAX_PREDEFINED ; ++i) {
|
|
if (origin_datatype->super.bdt_used & mask) {
|
|
num_found++;
|
|
found_index = i;
|
|
}
|
|
mask *= 2;
|
|
}
|
|
if (found_index < 0 || num_found > 1) {
|
|
/* this is an erroneous datatype. Let
|
|
ompi_op_is_valid tell the user that */
|
|
OMPI_ERRHANDLER_RETURN(MPI_ERR_TYPE, win, MPI_ERR_TYPE, FUNC_NAME);
|
|
} else {
|
|
origin_check_dt = (ompi_datatype_t*)
|
|
/* XXX TODO This is not needed, I hope... */
|
|
ompi_datatype_basicDatatypes[found_index];
|
|
}
|
|
}
|
|
|
|
/* ACCUMULATE, unlike REDUCE, can use with derived
|
|
datatypes with predefinied operations, with some
|
|
restrictions outlined in MPI-2:6.3.4. The derived
|
|
datatype must be composed entirley from one predefined
|
|
datatype (so you can do all the construction you want,
|
|
but at the bottom, you can only use one datatype, say,
|
|
MPI_INT). If the datatype at the target isn't
|
|
predefined, then make sure it's composed of only one
|
|
datatype, and check that datatype against
|
|
ompi_op_is_valid(). */
|
|
if (opal_datatype_is_predefined(&(target_datatype->super))) {
|
|
op_check_dt = target_datatype;
|
|
} else {
|
|
int i, found_index = -1, num_found = 0;
|
|
uint32_t mask = 1;
|
|
|
|
for (i = 0 ; i < OMPI_DATATYPE_MAX_PREDEFINED ; ++i) {
|
|
if (target_datatype->super.bdt_used & mask) {
|
|
num_found++;
|
|
found_index = i;
|
|
}
|
|
mask *= 2;
|
|
}
|
|
if (found_index < 0 || num_found > 1) {
|
|
/* this is an erroneous datatype. Let
|
|
ompi_op_is_valid tell the user that */
|
|
OMPI_ERRHANDLER_RETURN(MPI_ERR_TYPE, win, MPI_ERR_TYPE, FUNC_NAME);
|
|
} else {
|
|
/* datatype passes muster as far as restrictions
|
|
in MPI-2:6.3.4. Is the primitive ok with the
|
|
op? Unfortunately have to cast away
|
|
constness... */
|
|
op_check_dt = (ompi_datatype_t*)
|
|
ompi_datatype_basicDatatypes[found_index];
|
|
}
|
|
}
|
|
|
|
/* check to make sure same primitive type */
|
|
if (op_check_dt != origin_check_dt) {
|
|
OMPI_ERRHANDLER_RETURN(MPI_ERR_ARG, win, MPI_ERR_ARG, FUNC_NAME);
|
|
}
|
|
|
|
/* check to make sure primitive type is valid for
|
|
reduction. Should do this on the target, but
|
|
then can't get the errcode back for this
|
|
call */
|
|
if (!ompi_op_is_valid(op, op_check_dt, &msg, FUNC_NAME)) {
|
|
int ret = OMPI_ERRHANDLER_INVOKE(win, MPI_ERR_OP, msg);
|
|
free(msg);
|
|
return ret;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
OMPI_ERRHANDLER_CHECK(rc, win, rc, FUNC_NAME);
|
|
|
|
/* While technically the standard probably requires that the
|
|
datatypes used with MPI_REPLACE conform to all the rules
|
|
for other reduction operators, we don't require such
|
|
behaivor, as checking for it is expensive here and we don't
|
|
care in implementation.. */
|
|
if (op != &ompi_mpi_op_replace.op) {
|
|
ompi_datatype_t *op_check_dt;
|
|
char *msg;
|
|
|
|
/* ACCUMULATE, unlike REDUCE, can use with derived
|
|
datatypes with predefinied operations, with some
|
|
restrictions outlined in MPI-2:6.3.4. The derived
|
|
datatype must be composed entirley from one predefined
|
|
datatype (so you can do all the construction you want,
|
|
but at the bottom, you can only use one datatype, say,
|
|
MPI_INT). If the datatype at the target isn't
|
|
predefined, then make sure it's composed of only one
|
|
datatype, and check that datatype against
|
|
ompi_op_is_valid(). */
|
|
if (opal_datatype_is_predefined(&(target_datatype->super))) {
|
|
op_check_dt = target_datatype;
|
|
} else {
|
|
int i, found_index = -1, num_found = 0;
|
|
uint32_t mask = 1;
|
|
|
|
for (i = 0 ; i < OMPI_DATATYPE_MAX_PREDEFINED ; ++i) {
|
|
if (target_datatype->super.bdt_used & mask) {
|
|
num_found++;
|
|
found_index = i;
|
|
}
|
|
mask *= 2;
|
|
}
|
|
if (found_index < 0 || num_found > 1) {
|
|
/* this is an erroneous datatype. Let
|
|
ompi_op_is_valid tell the user that */
|
|
op_check_dt = target_datatype;
|
|
} else {
|
|
/* datatype passes muster as far as restrictions
|
|
in MPI-2:6.3.4. Is the primitive ok with the
|
|
op? Unfortunately have to cast away
|
|
constness... */
|
|
op_check_dt = (ompi_datatype_t*)
|
|
ompi_datatype_basicDatatypes[found_index];
|
|
}
|
|
}
|
|
if (!ompi_op_is_valid(op, op_check_dt, &msg, FUNC_NAME)) {
|
|
int ret = OMPI_ERRHANDLER_INVOKE(win, MPI_ERR_OP, msg);
|
|
free(msg);
|
|
return ret;
|
|
}
|
|
}
|
|
}
|
|
|
|
if (MPI_PROC_NULL == target_rank) {
|
|
return MPI_SUCCESS;
|
|
}
|
|
|
|
OPAL_CR_ENTER_LIBRARY();
|
|
|
|
rc = ompi_win->w_osc_module->osc_accumulate(origin_addr,
|
|
origin_count,
|
|
origin_datatype,
|
|
target_rank,
|
|
target_disp,
|
|
target_count,
|
|
target_datatype,
|
|
op, win);
|
|
OMPI_ERRHANDLER_RETURN(rc, win, rc, FUNC_NAME);
|
|
}
|