2004-01-14 03:42:02 +00:00
|
|
|
/*
|
2005-11-05 19:57:48 +00:00
|
|
|
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
|
|
|
|
* University Research and Technology
|
|
|
|
* Corporation. All rights reserved.
|
2016-05-24 18:20:51 -05:00
|
|
|
* Copyright (c) 2004-2016 The University of Tennessee and The University
|
2005-11-05 19:57:48 +00:00
|
|
|
* of Tennessee Research Foundation. All rights
|
|
|
|
* reserved.
|
2015-06-23 20:59:57 -07:00
|
|
|
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
2004-11-28 20:09:25 +00:00
|
|
|
* University of Stuttgart. All rights reserved.
|
2005-03-24 12:43:37 +00:00
|
|
|
* Copyright (c) 2004-2005 The Regents of the University of California.
|
|
|
|
* All rights reserved.
|
2006-10-27 12:32:36 +00:00
|
|
|
* Copyright (c) 2006 Cisco Systems, Inc. All rights reserved.
|
2015-08-28 14:54:23 +09:00
|
|
|
* Copyright (c) 2015 Research Organization for Information Science
|
|
|
|
* and Technology (RIST). All rights reserved.
|
2004-11-22 01:38:40 +00:00
|
|
|
* $COPYRIGHT$
|
2015-06-23 20:59:57 -07:00
|
|
|
*
|
2004-11-22 01:38:40 +00:00
|
|
|
* Additional copyrights may follow
|
2015-06-23 20:59:57 -07:00
|
|
|
*
|
2004-01-14 03:42:02 +00:00
|
|
|
* $HEADER$
|
|
|
|
*/
|
2004-03-26 14:15:20 +00:00
|
|
|
/**
|
|
|
|
* @file
|
|
|
|
*/
|
2004-01-14 03:42:02 +00:00
|
|
|
#ifndef MCA_PML_BASE_SEND_REQUEST_H
|
|
|
|
#define MCA_PML_BASE_SEND_REQUEST_H
|
|
|
|
|
2004-06-07 15:33:53 +00:00
|
|
|
#include "ompi_config.h"
|
2006-02-12 01:33:29 +00:00
|
|
|
#include "ompi/mca/pml/pml.h"
|
|
|
|
#include "ompi/mca/pml/base/pml_base_request.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 04:56:31 +00:00
|
|
|
#include "opal/datatype/opal_convertor.h"
|
2006-03-31 17:05:09 +00:00
|
|
|
#include "ompi/peruse/peruse-internal.h"
|
2004-01-14 03:42:02 +00:00
|
|
|
|
2007-12-21 06:19:46 +00:00
|
|
|
BEGIN_C_DECLS
|
2004-01-28 22:52:51 +00:00
|
|
|
|
2004-03-26 14:15:20 +00:00
|
|
|
/**
|
2015-06-23 20:59:57 -07:00
|
|
|
* Base type for send requests
|
2004-03-26 14:15:20 +00:00
|
|
|
*/
|
2004-06-24 19:28:30 +00:00
|
|
|
struct mca_pml_base_send_request_t {
|
2006-09-20 20:14:15 +00:00
|
|
|
mca_pml_base_request_t req_base; /**< base request type - common data structure for use by wait/test */
|
2015-08-28 14:54:23 +09:00
|
|
|
const void *req_addr; /**< pointer to send buffer - may not be application buffer */
|
2004-07-15 18:08:20 +00:00
|
|
|
size_t req_bytes_packed; /**< packed size of a message given the datatype and count */
|
|
|
|
mca_pml_base_send_mode_t req_send_mode; /**< type of send */
|
2004-01-14 03:42:02 +00:00
|
|
|
};
|
2004-06-24 19:28:30 +00:00
|
|
|
typedef struct mca_pml_base_send_request_t mca_pml_base_send_request_t;
|
2004-01-14 03:42:02 +00:00
|
|
|
|
2006-09-20 20:14:15 +00:00
|
|
|
OMPI_DECLSPEC OBJ_CLASS_DECLARATION( mca_pml_base_send_request_t );
|
2004-03-16 15:12:11 +00:00
|
|
|
|
2004-03-26 14:15:20 +00:00
|
|
|
/**
|
|
|
|
* Initialize a send request with call parameters.
|
|
|
|
*
|
2009-03-24 19:12:53 +00:00
|
|
|
* @param request (IN) Send request
|
|
|
|
* @param addr (IN) User buffer
|
|
|
|
* @param count (IN) Number of elements of indicated datatype.
|
|
|
|
* @param datatype (IN) User defined datatype
|
|
|
|
* @param peer (IN) Destination rank
|
|
|
|
* @param tag (IN) User defined tag
|
|
|
|
* @param comm (IN) Communicator
|
|
|
|
* @param mode (IN) Send mode (STANDARD,BUFFERED,SYNCHRONOUS,READY)
|
|
|
|
* @param persistent (IN) Is request persistent.
|
|
|
|
* @param convertor_flags (IN) Flags to pass to convertor
|
2004-10-19 22:00:19 +00:00
|
|
|
*
|
2005-02-11 20:32:10 +00:00
|
|
|
* Perform a any one-time initialization. Note that per-use initialization
|
2004-10-19 22:00:19 +00:00
|
|
|
* is done in the send request start routine.
|
2004-03-26 14:15:20 +00:00
|
|
|
*/
|
2004-10-12 15:50:01 +00:00
|
|
|
|
|
|
|
#define MCA_PML_BASE_SEND_REQUEST_INIT( request, \
|
|
|
|
addr, \
|
|
|
|
count, \
|
|
|
|
datatype, \
|
|
|
|
peer, \
|
|
|
|
tag, \
|
|
|
|
comm, \
|
|
|
|
mode, \
|
2009-03-24 19:12:53 +00:00
|
|
|
persistent, \
|
|
|
|
convertor_flags) \
|
2004-10-12 15:50:01 +00:00
|
|
|
{ \
|
2005-05-23 22:06:50 +00:00
|
|
|
/* increment reference counts */ \
|
2004-10-12 15:50:01 +00:00
|
|
|
OBJ_RETAIN(comm); \
|
|
|
|
\
|
2006-03-15 22:53:41 +00:00
|
|
|
OMPI_REQUEST_INIT(&(request)->req_base.req_ompi, persistent); \
|
2006-10-27 12:32:36 +00:00
|
|
|
(request)->req_base.req_ompi.req_mpi_object.comm = comm; \
|
2005-05-23 22:06:50 +00:00
|
|
|
(request)->req_addr = addr; \
|
|
|
|
(request)->req_send_mode = mode; \
|
2015-09-17 09:12:35 +09:00
|
|
|
(request)->req_base.req_addr = (void *)addr; \
|
2005-05-23 22:06:50 +00:00
|
|
|
(request)->req_base.req_count = count; \
|
|
|
|
(request)->req_base.req_datatype = datatype; \
|
|
|
|
(request)->req_base.req_peer = (int32_t)peer; \
|
|
|
|
(request)->req_base.req_tag = (int32_t)tag; \
|
|
|
|
(request)->req_base.req_comm = comm; \
|
2006-02-05 06:13:07 +00:00
|
|
|
/* (request)->req_base.req_proc is set on request allocation */ \
|
2016-05-24 18:20:51 -05:00
|
|
|
(request)->req_base.req_pml_complete = false; \
|
2005-05-23 22:06:50 +00:00
|
|
|
(request)->req_base.req_free_called = false; \
|
|
|
|
(request)->req_base.req_ompi.req_status._cancelled = 0; \
|
2006-09-20 20:14:15 +00:00
|
|
|
(request)->req_bytes_packed = 0; \
|
2004-10-12 15:50:01 +00:00
|
|
|
\
|
|
|
|
/* initialize datatype convertor for this request */ \
|
2006-09-20 20:14:15 +00:00
|
|
|
if( count > 0 ) { \
|
|
|
|
OBJ_RETAIN(datatype); \
|
2005-06-08 19:10:15 +00:00
|
|
|
/* We will create a convertor specialized for the */ \
|
|
|
|
/* remote architecture and prepared with the datatype. */ \
|
- 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 04:56:31 +00:00
|
|
|
opal_convertor_copy_and_prepare_for_send( \
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
(request)->req_base.req_proc->super.proc_convertor, \
|
- 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 04:56:31 +00:00
|
|
|
&((request)->req_base.req_datatype->super), \
|
2005-06-08 19:10:15 +00:00
|
|
|
(request)->req_base.req_count, \
|
|
|
|
(request)->req_base.req_addr, \
|
2009-03-24 19:12:53 +00:00
|
|
|
convertor_flags, \
|
2007-07-10 22:16:38 +00:00
|
|
|
&(request)->req_base.req_convertor ); \
|
- 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 04:56:31 +00:00
|
|
|
opal_convertor_get_packed_size( &(request)->req_base.req_convertor, \
|
2005-06-08 19:10:15 +00:00
|
|
|
&((request)->req_bytes_packed) );\
|
2004-10-12 15:50:01 +00:00
|
|
|
} \
|
2004-09-19 07:12:29 +00:00
|
|
|
}
|
2004-01-14 03:42:02 +00:00
|
|
|
|
2006-03-31 17:05:09 +00:00
|
|
|
/**
|
|
|
|
* Mark the request as started from the PML base point of view.
|
|
|
|
*
|
|
|
|
* @param request (IN) The send request.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#define MCA_PML_BASE_SEND_START( request ) \
|
|
|
|
do { \
|
|
|
|
(request)->req_pml_complete = false; \
|
2016-05-24 18:20:51 -05:00
|
|
|
(request)->req_ompi.req_complete = REQUEST_PENDING; \
|
2006-03-31 17:05:09 +00:00
|
|
|
(request)->req_ompi.req_state = OMPI_REQUEST_ACTIVE; \
|
|
|
|
(request)->req_ompi.req_status._cancelled = 0; \
|
|
|
|
} while (0)
|
|
|
|
|
2005-02-11 20:32:10 +00:00
|
|
|
/**
|
|
|
|
* Release the ref counts on the communicator and datatype.
|
|
|
|
*
|
|
|
|
* @param request (IN) The send request.
|
|
|
|
*/
|
|
|
|
|
2005-05-23 22:06:50 +00:00
|
|
|
#define MCA_PML_BASE_SEND_REQUEST_FINI( request ) \
|
2005-02-11 20:32:10 +00:00
|
|
|
do { \
|
2005-05-23 22:06:50 +00:00
|
|
|
OMPI_REQUEST_FINI(&(request)->req_base.req_ompi); \
|
2005-05-11 04:23:47 +00:00
|
|
|
OBJ_RELEASE((request)->req_base.req_comm); \
|
2006-09-20 20:14:15 +00:00
|
|
|
if( 0 != (request)->req_base.req_count ) \
|
|
|
|
OBJ_RELEASE((request)->req_base.req_datatype); \
|
- 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 04:56:31 +00:00
|
|
|
opal_convertor_cleanup( &((request)->req_base.req_convertor) ); \
|
2005-02-11 20:32:10 +00:00
|
|
|
} while (0)
|
2004-01-14 03:42:02 +00:00
|
|
|
|
2004-10-27 20:28:09 +00:00
|
|
|
|
2007-12-21 06:19:46 +00:00
|
|
|
END_C_DECLS
|
|
|
|
|
2004-01-14 03:42:02 +00:00
|
|
|
#endif
|
|
|
|
|