1
1
openmpi/ompi/op/op.c

403 строки
14 KiB
C
Исходник Обычный вид История

/*
* Copyright (c) 2004-2006 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2010 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2007 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) 2007-2010 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2009 Sun Microsystems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.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 "opal/class/opal_pointer_array.h"
#include "ompi/constants.h"
#include "ompi/op/op.h"
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
#include "ompi/mca/op/base/base.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_internal.h"
/*
* Table for Fortran <-> C op handle conversion
*/
opal_pointer_array_t *ompi_op_f_to_c_table;
/*
* Create intrinsic op
*/
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
static int add_intrinsic(ompi_op_t *op, int fort_handle, int flags,
const char *name);
/*
* Class information
*/
static void ompi_op_construct(ompi_op_t *eh);
static void ompi_op_destruct(ompi_op_t *eh);
/*
* Class instance
*/
OBJ_CLASS_INSTANCE(ompi_op_t, opal_object_t,
ompi_op_construct, ompi_op_destruct);
/*
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
* Intrinsic MPI_Op objects
*/
ompi_predefined_op_t ompi_mpi_op_null;
ompi_predefined_op_t ompi_mpi_op_max;
ompi_predefined_op_t ompi_mpi_op_min;
ompi_predefined_op_t ompi_mpi_op_sum;
ompi_predefined_op_t ompi_mpi_op_prod;
ompi_predefined_op_t ompi_mpi_op_land;
ompi_predefined_op_t ompi_mpi_op_band;
ompi_predefined_op_t ompi_mpi_op_lor;
ompi_predefined_op_t ompi_mpi_op_bor;
ompi_predefined_op_t ompi_mpi_op_lxor;
ompi_predefined_op_t ompi_mpi_op_bxor;
ompi_predefined_op_t ompi_mpi_op_maxloc;
ompi_predefined_op_t ompi_mpi_op_minloc;
ompi_predefined_op_t ompi_mpi_op_replace;
/*
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
* Map from ddt->id to position in op function pointer array
*/
- 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
int ompi_op_ddt_map[OMPI_DATATYPE_MAX_PREDEFINED];
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
#define FLAGS_NO_FLOAT \
(OMPI_OP_FLAGS_INTRINSIC | OMPI_OP_FLAGS_ASSOC | OMPI_OP_FLAGS_COMMUTE)
#define FLAGS \
(OMPI_OP_FLAGS_INTRINSIC | OMPI_OP_FLAGS_ASSOC | \
OMPI_OP_FLAGS_FLOAT_ASSOC | OMPI_OP_FLAGS_COMMUTE)
/*
* Initialize OMPI op infrastructure
*/
int ompi_op_init(void)
{
int i;
/* initialize ompi_op_f_to_c_table */
ompi_op_f_to_c_table = OBJ_NEW(opal_pointer_array_t);
if (NULL == ompi_op_f_to_c_table){
return OMPI_ERROR;
}
/* Fill in the ddt.id->op_position map */
- 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
for (i = 0; i < OMPI_DATATYPE_MAX_PREDEFINED; ++i) {
ompi_op_ddt_map[i] = -1;
}
ompi_op_ddt_map[OMPI_DATATYPE_MPI_INT8_T] = OMPI_OP_BASE_TYPE_INT8_T;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_UINT8_T] = OMPI_OP_BASE_TYPE_UINT8_T;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_INT16_T] = OMPI_OP_BASE_TYPE_INT16_T;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_UINT16_T] = OMPI_OP_BASE_TYPE_UINT16_T;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_INT32_T] = OMPI_OP_BASE_TYPE_INT32_T;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_UINT32_T] = OMPI_OP_BASE_TYPE_UINT32_T;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_INT64_T] = OMPI_OP_BASE_TYPE_INT64_T;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_UINT64_T] = OMPI_OP_BASE_TYPE_UINT64_T;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_FLOAT] = OMPI_OP_BASE_TYPE_FLOAT;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_DOUBLE] = OMPI_OP_BASE_TYPE_DOUBLE;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_LONG_DOUBLE] = OMPI_OP_BASE_TYPE_LONG_DOUBLE;
#if OMPI_SIZEOF_FORTRAN_COMPLEX8 == SIZEOF_FLOAT__COMPLEX
ompi_op_ddt_map[OMPI_DATATYPE_MPI_COMPLEX8] = OMPI_OP_BASE_TYPE_COMPLEX8;
#endif
#if OMPI_SIZEOF_FORTRAN_COMPLEX16 == SIZEOF_DOUBLE__COMPLEX
ompi_op_ddt_map[OMPI_DATATYPE_MPI_COMPLEX16] = OMPI_OP_BASE_TYPE_COMPLEX16;
#endif
#if OMPI_SIZEOF_FORTRAN_COMPLEX32 == SIZEOF_LONG_DOUBLE__COMPLEX
ompi_op_ddt_map[OMPI_DATATYPE_MPI_COMPLEX32] = OMPI_OP_BASE_TYPE_COMPLEX32;
#endif
ompi_op_ddt_map[OMPI_DATATYPE_MPI_WCHAR] = OMPI_OP_BASE_TYPE_WCHAR;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_BOOL] = OMPI_OP_BASE_TYPE_BOOL;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_LOGICAL] = OMPI_OP_BASE_TYPE_LOGICAL;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_CHARACTER] = OMPI_OP_BASE_TYPE_UINT8_T;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_INTEGER] = OMPI_OP_BASE_TYPE_INTEGER;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_REAL] = OMPI_OP_BASE_TYPE_REAL;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_DOUBLE_PRECISION] = OMPI_OP_BASE_TYPE_DOUBLE_PRECISION;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_COMPLEX] = OMPI_OP_BASE_TYPE_COMPLEX8;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_DOUBLE_COMPLEX] = OMPI_OP_BASE_TYPE_COMPLEX16;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_LONG_DOUBLE_COMPLEX] = OMPI_OP_BASE_TYPE_COMPLEX32;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_2INT] = OMPI_OP_BASE_TYPE_2INT;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_2INTEGER] = OMPI_OP_BASE_TYPE_2INTEGER;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_2REAL] = OMPI_OP_BASE_TYPE_2REAL;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_2DBLPREC] = OMPI_OP_BASE_TYPE_2DOUBLE_PRECISION;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_2COMPLEX] = -1; /* Not defined */
ompi_op_ddt_map[OMPI_DATATYPE_MPI_2DOUBLE_COMPLEX] = -1; /* Not defined */
ompi_op_ddt_map[OMPI_DATATYPE_MPI_FLOAT_INT] = OMPI_OP_BASE_TYPE_FLOAT_INT;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_DOUBLE_INT] = OMPI_OP_BASE_TYPE_DOUBLE_INT;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_LONG_DOUBLE_INT] = OMPI_OP_BASE_TYPE_LONG_DOUBLE_INT;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_LONG_INT] = OMPI_OP_BASE_TYPE_LONG_INT;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_SHORT_INT] = OMPI_OP_BASE_TYPE_SHORT_INT;
#if SIZEOF_PTRDIFF_T == 4
ompi_op_ddt_map[OMPI_DATATYPE_MPI_AINT] = OMPI_OP_BASE_TYPE_INT32_T;
#elif SIZEOF_PTRDIFF_T == 8
ompi_op_ddt_map[OMPI_DATATYPE_MPI_AINT] = OMPI_OP_BASE_TYPE_INT64_T;
#else
#warning Unsupported definition for MPI_AINT
#endif
#if OMPI_MPI_OFFSET_SIZE == 4
ompi_op_ddt_map[OMPI_DATATYPE_MPI_OFFSET] = OMPI_OP_BASE_TYPE_UINT32_T;
#elif OMPI_MPI_OFFSET_SIZE == 8
ompi_op_ddt_map[OMPI_DATATYPE_MPI_OFFSET] = OMPI_OP_BASE_TYPE_UINT64_T;
#else
#warning Unsupported definition for MPI_OFFSET
#endif
ompi_op_ddt_map[OMPI_DATATYPE_MPI_C_BOOL] = OMPI_OP_BASE_TYPE_BOOL;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_C_COMPLEX] = OMPI_OP_BASE_TYPE_COMPLEX8;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_C_FLOAT_COMPLEX] = OMPI_OP_BASE_TYPE_COMPLEX8;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_C_DOUBLE_COMPLEX] = OMPI_OP_BASE_TYPE_COMPLEX16;
ompi_op_ddt_map[OMPI_DATATYPE_MPI_C_LONG_DOUBLE_COMPLEX] = OMPI_OP_BASE_TYPE_COMPLEX32;
/* Create the intrinsic ops */
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
if (OMPI_SUCCESS !=
add_intrinsic(&ompi_mpi_op_null.op, OMPI_OP_BASE_FORTRAN_NULL,
FLAGS, "MPI_NULL") ||
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
OMPI_SUCCESS !=
add_intrinsic(&ompi_mpi_op_max.op, OMPI_OP_BASE_FORTRAN_MAX,
FLAGS, "MPI_MAX") ||
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
OMPI_SUCCESS !=
add_intrinsic(&ompi_mpi_op_min.op, OMPI_OP_BASE_FORTRAN_MIN,
FLAGS, "MPI_MIN") ||
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
OMPI_SUCCESS !=
add_intrinsic(&ompi_mpi_op_sum.op, OMPI_OP_BASE_FORTRAN_SUM,
FLAGS_NO_FLOAT, "MPI_SUM") ||
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
OMPI_SUCCESS !=
add_intrinsic(&ompi_mpi_op_prod.op, OMPI_OP_BASE_FORTRAN_PROD,
FLAGS_NO_FLOAT, "MPI_PROD") ||
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
OMPI_SUCCESS !=
add_intrinsic(&ompi_mpi_op_land.op, OMPI_OP_BASE_FORTRAN_LAND,
FLAGS, "MPI_LAND") ||
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
OMPI_SUCCESS !=
add_intrinsic(&ompi_mpi_op_band.op, OMPI_OP_BASE_FORTRAN_BAND,
FLAGS, "MPI_BAND") ||
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
OMPI_SUCCESS !=
add_intrinsic(&ompi_mpi_op_lor.op, OMPI_OP_BASE_FORTRAN_LOR,
FLAGS, "MPI_LOR") ||
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
OMPI_SUCCESS !=
add_intrinsic(&ompi_mpi_op_bor.op, OMPI_OP_BASE_FORTRAN_BOR,
FLAGS, "MPI_BOR") ||
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
OMPI_SUCCESS !=
add_intrinsic(&ompi_mpi_op_lxor.op, OMPI_OP_BASE_FORTRAN_LXOR,
FLAGS, "MPI_LXOR") ||
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
OMPI_SUCCESS !=
add_intrinsic(&ompi_mpi_op_bxor.op, OMPI_OP_BASE_FORTRAN_BXOR,
FLAGS, "MPI_BXOR") ||
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
OMPI_SUCCESS !=
add_intrinsic(&ompi_mpi_op_maxloc.op, OMPI_OP_BASE_FORTRAN_MAXLOC,
FLAGS, "MPI_MAXLOC") ||
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
OMPI_SUCCESS !=
add_intrinsic(&ompi_mpi_op_minloc.op, OMPI_OP_BASE_FORTRAN_MINLOC,
FLAGS, "MPI_MINLOC") ||
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
OMPI_SUCCESS !=
add_intrinsic(&ompi_mpi_op_replace.op, OMPI_OP_BASE_FORTRAN_REPLACE,
FLAGS, "MPI_REPLACE")) {
return OMPI_ERROR;
}
/* All done */
return OMPI_SUCCESS;
}
/*
* Clean up the op resources
*/
int ompi_op_finalize(void)
{
/* clean up the intrinsic ops */
OBJ_DESTRUCT(&ompi_mpi_op_replace);
OBJ_DESTRUCT(&ompi_mpi_op_minloc);
OBJ_DESTRUCT(&ompi_mpi_op_maxloc);
OBJ_DESTRUCT(&ompi_mpi_op_bxor);
OBJ_DESTRUCT(&ompi_mpi_op_lxor);
OBJ_DESTRUCT(&ompi_mpi_op_bor);
OBJ_DESTRUCT(&ompi_mpi_op_lor);
OBJ_DESTRUCT(&ompi_mpi_op_band);
OBJ_DESTRUCT(&ompi_mpi_op_land);
OBJ_DESTRUCT(&ompi_mpi_op_prod);
OBJ_DESTRUCT(&ompi_mpi_op_sum);
OBJ_DESTRUCT(&ompi_mpi_op_min);
OBJ_DESTRUCT(&ompi_mpi_op_max);
OBJ_DESTRUCT(&ompi_mpi_op_null);
/* Remove op F2C table */
OBJ_RELEASE(ompi_op_f_to_c_table);
/* All done */
return OMPI_SUCCESS;
}
/*
* Create a new MPI_Op
*/
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
ompi_op_t *ompi_op_create_user(bool commute,
ompi_op_fortran_handler_fn_t func)
{
ompi_op_t *new_op;
/* Create a new object and ensure that it's valid */
new_op = OBJ_NEW(ompi_op_t);
if (NULL == new_op) {
goto error;
}
if (0 > new_op->o_f_to_c_index) {
OBJ_RELEASE(new_op);
new_op = NULL;
goto error;
}
/*
* The new object is valid -- initialize it. If this is being
* created from fortran, the fortran MPI API wrapper function
* will override the o_flags field directly. We cast the
* function pointer type to the fortran type arbitrarily -- it
* only has to be a function pointer in order to store properly,
* it doesn't matter what type it is (we'll cast it to the Right
* type when we *use* it).
*/
new_op->o_flags = OMPI_OP_FLAGS_ASSOC;
if (commute) {
new_op->o_flags |= OMPI_OP_FLAGS_COMMUTE;
}
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
/* Set the user-defined callback function. The "fort_fn" member
is part of a union, so it doesn't matter if this is a C or
Fortan callback; we'll call the right flavor (per o_flags) at
invocation time. */
new_op->o_func.fort_fn = func;
error:
/* All done */
return new_op;
}
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
/*
* See lengthy comment in mpi/cxx/intercepts.cc for how the C++ MPI::Op
* callbacks work.
*/
void ompi_op_set_cxx_callback(ompi_op_t *op, MPI_User_function *fn)
{
op->o_flags |= OMPI_OP_FLAGS_CXX_FUNC;
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
/* The OMPI C++ intercept was previously stored in
op->o_func.fort_fn by ompi_op_create_user(). So save that in
cxx.intercept_fn and put the user's fn in cxx.user_fn. */
op->o_func.cxx_data.intercept_fn =
(ompi_op_cxx_handler_fn_t *) op->o_func.fort_fn;
op->o_func.cxx_data.user_fn = fn;
}
/**************************************************************************
*
* Static functions
*
**************************************************************************/
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
static int add_intrinsic(ompi_op_t *op, int fort_handle, int flags,
const char *name)
{
/* Add the op to the table */
OBJ_CONSTRUCT(op, ompi_op_t);
if (op->o_f_to_c_index != fort_handle) {
return OMPI_ERROR;
}
Two major things in this commit: * New "op" MPI layer framework * Addition of the MPI_REDUCE_LOCAL proposed function (for MPI-2.2) = Op framework = Add new "op" framework in the ompi layer. This framework replaces the hard-coded MPI_Op back-end functions for (MPI_Op, MPI_Datatype) tuples for pre-defined MPI_Ops, allowing components and modules to provide the back-end functions. The intent is that components can be written to take advantage of hardware acceleration (GPU, FPGA, specialized CPU instructions, etc.). Similar to other frameworks, components are intended to be able to discover at run-time if they can be used, and if so, elect themselves to be selected (or disqualify themselves from selection if they cannot run). If specialized hardware is not available, there is a default set of functions that will automatically be used. This framework is ''not'' used for user-defined MPI_Ops. The new op framework is similar to the existing coll framework, in that the final set of function pointers that are used on any given intrinsic MPI_Op can be a mixed bag of function pointers, potentially coming from multiple different op modules. This allows for hardware that only supports some of the operations, not all of them (e.g., a GPU that only supports single-precision operations). All the hard-coded back-end MPI_Op functions for (MPI_Op, MPI_Datatype) tuples still exist, but unlike coll, they're in the framework base (vs. being in a separate "basic" component) and are automatically used if no component is found at runtime that provides a module with the necessary function pointers. There is an "example" op component that will hopefully be useful to those writing meaningful op components. It is currently .ompi_ignore'd so that it doesn't impinge on other developers (it's somewhat chatty in terms of opal_output() so that you can tell when its functions have been invoked). See the README file in the example op component directory. Developers of new op components are encouraged to look at the following wiki pages: https://svn.open-mpi.org/trac/ompi/wiki/devel/Autogen https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateComponent https://svn.open-mpi.org/trac/ompi/wiki/devel/CreateFramework = MPI_REDUCE_LOCAL = Part of the MPI-2.2 proposal listed here: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/24 is to add a new function named MPI_REDUCE_LOCAL. It is very easy to implement, so I added it (also because it makes testing the op framework pretty easy -- you can do it in serial rather than via parallel reductions). There's even a man page! This commit was SVN r20280.
2009-01-15 02:44:31 +03:00
/* Set the members */
op->o_flags = flags;
strncpy(op->o_name, name, sizeof(op->o_name) - 1);
op->o_name[sizeof(op->o_name) - 1] = '\0';
/* Perform the selection on this op to fill in the function
pointers (except for NULL and REPLACE, which don't get
components) */
if (OMPI_OP_BASE_FORTRAN_NULL != op->o_f_to_c_index &&
OMPI_OP_BASE_FORTRAN_REPLACE != op->o_f_to_c_index) {
return ompi_op_base_op_select(op);
} else {
return OMPI_SUCCESS;
}
}
/*
* Op constructor
*/
static void ompi_op_construct(ompi_op_t *new_op)
{
int i;
/* assign entry in fortran <-> c translation array */
new_op->o_f_to_c_index =
opal_pointer_array_add(ompi_op_f_to_c_table, new_op);
/* Set everything to NULL so that we can intelligently free
non-NULL's in the destructor */
for (i = 0; i < OMPI_OP_BASE_TYPE_MAX; ++i) {
new_op->o_func.intrinsic.fns[i] = NULL;
new_op->o_func.intrinsic.modules[i] = NULL;
new_op->o_3buff_intrinsic.fns[i] = NULL;
new_op->o_3buff_intrinsic.modules[i] = NULL;
}
}
/*
* Op destructor
*/
static void ompi_op_destruct(ompi_op_t *op)
{
int i;
/* reset the ompi_op_f_to_c_table entry - make sure that the
entry is in the table */
if (NULL != opal_pointer_array_get_item(ompi_op_f_to_c_table,
op->o_f_to_c_index)) {
opal_pointer_array_set_item(ompi_op_f_to_c_table,
op->o_f_to_c_index, NULL);
}
for (i = 0; i < OMPI_OP_BASE_TYPE_MAX; ++i) {
op->o_func.intrinsic.fns[i] = NULL;
if( NULL != op->o_func.intrinsic.modules[i] ) {
OBJ_RELEASE(op->o_func.intrinsic.modules[i]);
op->o_func.intrinsic.modules[i] = NULL;
}
op->o_3buff_intrinsic.fns[i] = NULL;
if( NULL != op->o_3buff_intrinsic.modules[i] ) {
OBJ_RELEASE(op->o_3buff_intrinsic.modules[i]);
op->o_3buff_intrinsic.modules[i] = NULL;
}
}
}