2005-09-01 03:38:09 +04:00
|
|
|
/*
|
2005-11-05 22:57:48 +03:00
|
|
|
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
|
|
|
|
* University Research and Technology
|
|
|
|
* Corporation. All rights reserved.
|
2009-04-01 03:44:59 +04:00
|
|
|
* Copyright (c) 2004-2009 The University of Tennessee and The University
|
2005-11-05 22:57:48 +03:00
|
|
|
* of Tennessee Research Foundation. All rights
|
|
|
|
* reserved.
|
2005-09-01 03:38:09 +04:00
|
|
|
* Copyright (c) 2004-2005 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.
|
|
|
|
* $COPYRIGHT$
|
|
|
|
*
|
|
|
|
* Additional copyrights may follow
|
|
|
|
*
|
|
|
|
* $HEADER$
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include "ompi_config.h"
|
|
|
|
#include "coll_tuned.h"
|
|
|
|
|
|
|
|
#include "mpi.h"
|
2006-02-12 04:33:29 +03:00
|
|
|
#include "ompi/constants.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"
|
2006-02-12 04:33:29 +03:00
|
|
|
#include "ompi/communicator/communicator.h"
|
|
|
|
#include "ompi/mca/coll/base/coll_tags.h"
|
|
|
|
#include "ompi/mca/pml/pml.h"
|
2005-11-22 06:42:15 +03:00
|
|
|
#include "coll_tuned_util.h"
|
2005-09-01 03:38:09 +04:00
|
|
|
|
2012-03-05 18:10:26 +04:00
|
|
|
int ompi_coll_tuned_sendrecv_actual( void* sendbuf, size_t scount,
|
2007-01-11 00:56:59 +03:00
|
|
|
ompi_datatype_t* sdatatype,
|
2006-10-18 06:00:46 +04:00
|
|
|
int dest, int stag,
|
2012-03-05 18:10:26 +04:00
|
|
|
void* recvbuf, size_t rcount,
|
2007-01-11 00:56:59 +03:00
|
|
|
ompi_datatype_t* rdatatype,
|
2006-10-18 06:00:46 +04:00
|
|
|
int source, int rtag,
|
|
|
|
struct ompi_communicator_t* comm,
|
|
|
|
ompi_status_public_t* status )
|
2005-09-01 03:38:09 +04:00
|
|
|
|
|
|
|
{ /* post receive first, then send, then waitall... should be fast (I hope) */
|
2006-10-18 06:00:46 +04:00
|
|
|
int err, line = 0;
|
|
|
|
ompi_request_t* reqs[2];
|
|
|
|
ompi_status_public_t statuses[2];
|
2005-09-01 03:38:09 +04:00
|
|
|
|
|
|
|
/* post new irecv */
|
2007-01-11 00:56:59 +03:00
|
|
|
err = MCA_PML_CALL(irecv( recvbuf, rcount, rdatatype, source, rtag,
|
|
|
|
comm, &reqs[0]));
|
2005-09-01 03:38:09 +04:00
|
|
|
if (err != MPI_SUCCESS) { line = __LINE__; goto error_handler; }
|
|
|
|
|
|
|
|
/* send data to children */
|
2007-01-11 00:56:59 +03:00
|
|
|
err = MCA_PML_CALL(isend( sendbuf, scount, sdatatype, dest, stag,
|
|
|
|
MCA_PML_BASE_SEND_STANDARD, comm, &reqs[1]));
|
2005-09-01 03:38:09 +04:00
|
|
|
if (err != MPI_SUCCESS) { line = __LINE__; goto error_handler; }
|
|
|
|
|
|
|
|
err = ompi_request_wait_all( 2, reqs, statuses );
|
2011-02-25 03:28:32 +03:00
|
|
|
if (err != MPI_SUCCESS) { line = __LINE__; goto error_handler; }
|
2005-09-01 03:38:09 +04:00
|
|
|
|
2009-04-01 03:44:59 +04:00
|
|
|
if (MPI_STATUS_IGNORE != status) {
|
2005-09-01 21:12:28 +04:00
|
|
|
*status = statuses[0];
|
|
|
|
}
|
2005-09-01 03:38:09 +04:00
|
|
|
|
|
|
|
return (MPI_SUCCESS);
|
|
|
|
|
2011-02-25 03:28:32 +03:00
|
|
|
error_handler:
|
2009-04-01 03:44:59 +04:00
|
|
|
/* As we use wait_all we will get MPI_ERR_IN_STATUS which is not an error
|
|
|
|
* code that we can propagate up the stack. Instead, look for the real
|
|
|
|
* error code from the MPI_ERROR in the status.
|
|
|
|
*/
|
|
|
|
if( MPI_ERR_IN_STATUS == err ) {
|
2011-02-25 03:28:32 +03:00
|
|
|
/* At least we know the error was detected during the wait_all */
|
2009-04-01 03:44:59 +04:00
|
|
|
int err_index = 0;
|
2011-02-25 03:28:32 +03:00
|
|
|
if( MPI_SUCCESS == statuses[0].MPI_ERROR ) {
|
2009-04-01 03:44:59 +04:00
|
|
|
err_index = 1;
|
|
|
|
}
|
|
|
|
if (MPI_STATUS_IGNORE != status) {
|
|
|
|
*status = statuses[err_index];
|
|
|
|
}
|
|
|
|
err = statuses[err_index].MPI_ERROR;
|
|
|
|
OPAL_OUTPUT ((ompi_coll_tuned_stream, "%s:%d: Error %d occurred (req index %d)\n",
|
|
|
|
__FILE__, line, err, err_index));
|
|
|
|
} else {
|
|
|
|
/* Error discovered during the posting of the irecv or isend,
|
|
|
|
* and no status is available.
|
|
|
|
*/
|
|
|
|
OPAL_OUTPUT ((ompi_coll_tuned_stream, "%s:%d: Error %d occurred\n",
|
|
|
|
__FILE__, line, err));
|
|
|
|
if (MPI_STATUS_IGNORE != status) {
|
|
|
|
status->MPI_ERROR = err;
|
|
|
|
}
|
|
|
|
}
|
2005-09-01 03:38:09 +04:00
|
|
|
return (err);
|
|
|
|
}
|
|
|
|
|
2006-02-01 02:21:46 +03:00
|
|
|
/*
|
|
|
|
* localcompleted version that makes sure the send has completed locally
|
2007-01-11 00:56:59 +03:00
|
|
|
* Currently this is a sync call, but will change to locally completed
|
|
|
|
* version when available
|
2006-02-01 02:21:46 +03:00
|
|
|
*/
|
|
|
|
|
2012-03-05 18:10:26 +04:00
|
|
|
int ompi_coll_tuned_sendrecv_actual_localcompleted( void* sendbuf, size_t scount,
|
2007-01-11 00:56:59 +03:00
|
|
|
ompi_datatype_t* sdatatype,
|
|
|
|
int dest, int stag,
|
2012-03-05 18:10:26 +04:00
|
|
|
void* recvbuf, size_t rcount,
|
2007-01-11 00:56:59 +03:00
|
|
|
ompi_datatype_t* rdatatype,
|
|
|
|
int source, int rtag,
|
|
|
|
struct ompi_communicator_t* comm,
|
|
|
|
ompi_status_public_t* status )
|
2006-02-01 02:21:46 +03:00
|
|
|
|
|
|
|
{ /* post receive first, then [local] sync send, then wait... should be fast (I hope) */
|
2006-10-18 06:00:46 +04:00
|
|
|
int err, line = 0;
|
2006-11-11 02:06:19 +03:00
|
|
|
ompi_request_t* req[2];
|
2009-04-01 03:44:59 +04:00
|
|
|
ompi_status_public_t statuses[2];
|
2006-02-01 02:21:46 +03:00
|
|
|
|
|
|
|
/* post new irecv */
|
2007-01-11 00:56:59 +03:00
|
|
|
err = MCA_PML_CALL(irecv( recvbuf, rcount, rdatatype, source, rtag,
|
|
|
|
comm, &(req[0])));
|
2006-02-01 02:21:46 +03:00
|
|
|
if (err != MPI_SUCCESS) { line = __LINE__; goto error_handler; }
|
|
|
|
|
|
|
|
/* send data to children */
|
2006-11-11 02:06:19 +03:00
|
|
|
err = MCA_PML_CALL(isend( sendbuf, scount, sdatatype, dest, stag,
|
|
|
|
MCA_PML_BASE_SEND_SYNCHRONOUS, comm, &(req[1])));
|
2006-02-01 02:21:46 +03:00
|
|
|
if (err != MPI_SUCCESS) { line = __LINE__; goto error_handler; }
|
2005-11-08 06:36:38 +03:00
|
|
|
|
2009-04-01 03:44:59 +04:00
|
|
|
err = ompi_request_wait_all( 2, req, statuses );
|
2006-02-01 02:21:46 +03:00
|
|
|
if (err != MPI_SUCCESS) { line = __LINE__; goto error_handler; }
|
|
|
|
|
2009-04-01 03:44:59 +04:00
|
|
|
if (MPI_STATUS_IGNORE != status) {
|
|
|
|
*status = statuses[0];
|
2006-02-01 02:21:46 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
return (MPI_SUCCESS);
|
|
|
|
|
|
|
|
error_handler:
|
2009-04-01 03:44:59 +04:00
|
|
|
/* As we use wait_all we will get MPI_ERR_IN_STATUS which is not an error
|
|
|
|
* code that we can propagate up the stack. Instead, look for the real
|
|
|
|
* error code from the MPI_ERROR in the status.
|
|
|
|
*/
|
|
|
|
if( MPI_ERR_IN_STATUS == err ) {
|
2011-02-25 03:28:32 +03:00
|
|
|
/* At least we know the error was detected during the wait_all */
|
2009-04-01 03:44:59 +04:00
|
|
|
int err_index = 0;
|
2011-02-25 03:28:32 +03:00
|
|
|
if( MPI_SUCCESS == statuses[0].MPI_ERROR ) {
|
2009-04-01 03:44:59 +04:00
|
|
|
err_index = 1;
|
|
|
|
}
|
|
|
|
if (MPI_STATUS_IGNORE != status) {
|
|
|
|
*status = statuses[err_index];
|
|
|
|
}
|
|
|
|
err = statuses[err_index].MPI_ERROR;
|
|
|
|
OPAL_OUTPUT ((ompi_coll_tuned_stream, "%s:%d: Error %d occurred (req index %d)\n",
|
|
|
|
__FILE__,line,err, err_index));
|
|
|
|
} else {
|
2011-02-25 03:28:32 +03:00
|
|
|
/* Error discovered during the posting of the irecv or isend,
|
|
|
|
* and no status is available.
|
|
|
|
*/
|
2009-04-01 03:44:59 +04:00
|
|
|
OPAL_OUTPUT ((ompi_coll_tuned_stream, "%s:%d: Error %d occurred\n",
|
2011-02-25 03:28:32 +03:00
|
|
|
__FILE__, line, err));
|
2009-04-01 03:44:59 +04:00
|
|
|
if (MPI_STATUS_IGNORE != status) {
|
|
|
|
status->MPI_ERROR = err;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2006-02-01 02:21:46 +03:00
|
|
|
return (err);
|
|
|
|
}
|
2007-01-18 00:29:13 +03:00
|
|
|
|