2004-09-19 11:05:58 +04:00
|
|
|
/* -*- Mode: C; c-basic-offset:4 ; -*- */
|
|
|
|
|
2004-01-30 00:42:40 +03:00
|
|
|
/*
|
2007-03-17 02:11:45 +03:00
|
|
|
* Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana
|
2005-11-05 22:57:48 +03:00
|
|
|
* 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.
|
2008-02-12 11:46:27 +03:00
|
|
|
* Copyright (c) 2004-2008 High Performance Computing Center Stuttgart,
|
2004-11-28 23:09:25 +03:00
|
|
|
* University of Stuttgart. All rights reserved.
|
2005-03-24 15:43:37 +03:00
|
|
|
* Copyright (c) 2004-2005 The Regents of the University of California.
|
|
|
|
* All rights reserved.
|
2007-03-21 14:10:42 +03:00
|
|
|
* Copyright (c) 2006 Cisco Systems, Inc. All rights reserved.
|
2004-11-22 04:38:40 +03:00
|
|
|
* $COPYRIGHT$
|
|
|
|
*
|
|
|
|
* Additional copyrights may follow
|
|
|
|
*
|
2004-10-15 23:31:47 +04:00
|
|
|
* $HEADER$
|
2004-01-30 00:42:40 +03:00
|
|
|
*/
|
2004-06-07 19:33:53 +04:00
|
|
|
#include "ompi_config.h"
|
2004-01-30 00:42:40 +03:00
|
|
|
#include <stdio.h>
|
|
|
|
|
2006-02-12 04:33:29 +03:00
|
|
|
#include "ompi/mpi/c/bindings.h"
|
2009-04-29 05:32:14 +04:00
|
|
|
#include "ompi/runtime/params.h"
|
|
|
|
#include "ompi/communicator/communicator.h"
|
|
|
|
#include "ompi/errhandler/errhandler.h"
|
- Split the datatype engine into two parts: an MPI specific part in
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.
2009-07-13 08:56:31 +04:00
|
|
|
#include "ompi/datatype/ompi_datatype.h"
|
|
|
|
#include "opal/datatype/opal_convertor.h"
|
2008-02-12 21:01:17 +03:00
|
|
|
#include "ompi/memchecker.h"
|
2004-01-30 00:42:40 +03:00
|
|
|
|
2009-05-07 00:11:28 +04:00
|
|
|
#if OPAL_HAVE_WEAK_SYMBOLS && OMPI_PROFILING_DEFINES
|
2004-01-30 00:42:40 +03:00
|
|
|
#pragma weak MPI_Pack = PMPI_Pack
|
|
|
|
#endif
|
|
|
|
|
2004-06-07 19:33:53 +04:00
|
|
|
#if OMPI_PROFILING_DEFINES
|
2006-02-12 04:33:29 +03:00
|
|
|
#include "ompi/mpi/c/profile/defines.h"
|
2004-04-20 22:50:43 +04:00
|
|
|
#endif
|
|
|
|
|
2004-06-24 20:47:00 +04:00
|
|
|
static const char FUNC_NAME[] = "MPI_Pack";
|
2004-05-08 03:44:49 +04:00
|
|
|
|
|
|
|
|
2004-01-30 00:42:40 +03:00
|
|
|
int MPI_Pack(void *inbuf, int incount, MPI_Datatype datatype,
|
2004-05-08 03:44:49 +04:00
|
|
|
void *outbuf, int outsize, int *position, MPI_Comm comm)
|
|
|
|
{
|
2006-10-27 03:11:26 +04:00
|
|
|
int rc;
|
- Split the datatype engine into two parts: an MPI specific part in
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.
2009-07-13 08:56:31 +04:00
|
|
|
opal_convertor_t local_convertor;
|
2005-06-08 23:14:41 +04:00
|
|
|
struct iovec invec;
|
|
|
|
unsigned int iov_count;
|
|
|
|
size_t size;
|
2004-09-19 11:05:58 +04:00
|
|
|
|
2008-02-12 11:46:27 +03:00
|
|
|
MEMCHECKER(
|
|
|
|
memchecker_datatype(datatype);
|
|
|
|
memchecker_call(&opal_memchecker_base_isdefined, inbuf, incount, datatype);
|
|
|
|
memchecker_comm(comm);
|
|
|
|
);
|
2007-03-17 02:11:45 +03:00
|
|
|
|
2005-06-08 23:14:41 +04:00
|
|
|
if (MPI_PARAM_CHECK) {
|
|
|
|
OMPI_ERR_INIT_FINALIZE(FUNC_NAME);
|
2007-03-21 14:10:42 +03:00
|
|
|
if (ompi_comm_invalid(comm)) {
|
2005-06-08 23:14:41 +04:00
|
|
|
return OMPI_ERRHANDLER_INVOKE(MPI_COMM_WORLD, MPI_ERR_COMM, FUNC_NAME);
|
|
|
|
} else if ((NULL == outbuf) || (NULL == position)) { /* inbuf can be MPI_BOTTOM */
|
|
|
|
return OMPI_ERRHANDLER_INVOKE(comm, MPI_ERR_ARG, FUNC_NAME);
|
|
|
|
} else if (incount < 0) {
|
|
|
|
return OMPI_ERRHANDLER_INVOKE(comm, MPI_ERR_COUNT, FUNC_NAME);
|
|
|
|
} else if (outsize < 0) {
|
|
|
|
return OMPI_ERRHANDLER_INVOKE(comm, MPI_ERR_ARG, FUNC_NAME);
|
2007-03-21 14:10:42 +03:00
|
|
|
} else if (MPI_DATATYPE_NULL == datatype || NULL == datatype) {
|
2005-06-08 23:14:41 +04:00
|
|
|
return OMPI_ERRHANDLER_INVOKE(comm, MPI_ERR_TYPE, FUNC_NAME);
|
|
|
|
}
|
2004-05-08 03:44:49 +04:00
|
|
|
}
|
|
|
|
|
2008-02-20 01:15:52 +03:00
|
|
|
OPAL_CR_ENTER_LIBRARY();
|
|
|
|
|
- Split the datatype engine into two parts: an MPI specific part in
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.
2009-07-13 08:56:31 +04:00
|
|
|
OBJ_CONSTRUCT( &local_convertor, opal_convertor_t );
|
2005-06-08 23:14:41 +04:00
|
|
|
/* the resulting convertor will be set to the position ZERO */
|
- Split the datatype engine into two parts: an MPI specific part in
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.
2009-07-13 08:56:31 +04:00
|
|
|
opal_convertor_copy_and_prepare_for_send( ompi_mpi_local_convertor, &(datatype->super),
|
2006-03-17 21:46:48 +03:00
|
|
|
incount, inbuf, 0, &local_convertor );
|
2004-08-30 11:52:09 +04:00
|
|
|
|
2005-06-08 23:14:41 +04:00
|
|
|
/* Check for truncation */
|
- Split the datatype engine into two parts: an MPI specific part in
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.
2009-07-13 08:56:31 +04:00
|
|
|
opal_convertor_get_packed_size( &local_convertor, &size );
|
2005-06-08 23:14:41 +04:00
|
|
|
if( (*position + size) > (unsigned int)outsize ) { /* we can cast as we already checked for < 0 */
|
|
|
|
OBJ_DESTRUCT( &local_convertor );
|
2008-02-20 01:15:52 +03:00
|
|
|
OPAL_CR_EXIT_LIBRARY();
|
2005-06-08 23:14:41 +04:00
|
|
|
return OMPI_ERRHANDLER_INVOKE(comm, MPI_ERR_TRUNCATE, FUNC_NAME);
|
|
|
|
}
|
2004-05-08 03:44:49 +04:00
|
|
|
|
2005-06-08 23:14:41 +04:00
|
|
|
/* Prepare the iovec with all informations */
|
|
|
|
invec.iov_base = (char*) outbuf + (*position);
|
2006-08-29 21:17:35 +04:00
|
|
|
invec.iov_len = size;
|
2004-05-08 03:44:49 +04:00
|
|
|
|
2005-06-08 23:14:41 +04:00
|
|
|
/* Do the actual packing */
|
|
|
|
iov_count = 1;
|
- Split the datatype engine into two parts: an MPI specific part in
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.
2009-07-13 08:56:31 +04:00
|
|
|
rc = opal_convertor_pack( &local_convertor, &invec, &iov_count, &size );
|
2005-06-08 23:14:41 +04:00
|
|
|
*position += size;
|
|
|
|
OBJ_DESTRUCT( &local_convertor );
|
2004-08-30 02:45:06 +04:00
|
|
|
|
2005-06-08 23:14:41 +04:00
|
|
|
/* All done. Note that the convertor returns 1 upon success, not
|
|
|
|
OMPI_SUCCESS. */
|
|
|
|
OMPI_ERRHANDLER_RETURN((rc == 1) ? OMPI_SUCCESS : OMPI_ERROR,
|
|
|
|
comm, MPI_ERR_UNKNOWN, FUNC_NAME);
|
2004-01-30 00:42:40 +03:00
|
|
|
}
|