1
1
openmpi/ompi/mca/coll/tuned/coll_tuned_bcast.c

858 строки
38 KiB
C
Исходник Обычный вид История

MCA/base: Add new MCA variable system Features: - Support for an override parameter file (openmpi-mca-param-override.conf). Variable values in this file can not be overridden by any file or environment value. - Support for boolean, unsigned, and unsigned long long variables. - Support for true/false values. - Support for enumerations on integer variables. - Support for MPIT scope, verbosity, and binding. - Support for command line source. - Support for setting variable source via the environment using OMPI_MCA_SOURCE_<var name>=source (either command or file:filename) - Cleaner API. - Support for variable groups (equivalent to MPIT categories). Notes: - Variables must be created with a backing store (char **, int *, or bool *) that must live at least as long as the variable. - Creating a variable with the MCA_BASE_VAR_FLAG_SETTABLE enables the use of mca_base_var_set_value() to change the value. - String values are duplicated when the variable is registered. It is up to the caller to free the original value if necessary. The new value will be freed by the mca_base_var system and must not be freed by the user. - Variables with constant scope may not be settable. - Variable groups (and all associated variables) are deregistered when the component is closed or the component repository item is freed. This prevents a segmentation fault from accessing a variable after its component is unloaded. - After some discussion we decided we should remove the automatic registration of component priority variables. Few component actually made use of this feature. - The enumerator interface was updated to be general enough to handle future uses of the interface. - The code to generate ompi_info output has been moved into the MCA variable system. See mca_base_var_dump(). opal: update core and components to mca_base_var system orte: update core and components to mca_base_var system ompi: update core and components to mca_base_var system This commit also modifies the rmaps framework. The following variables were moved from ppr and lama: rmaps_base_pernode, rmaps_base_n_pernode, rmaps_base_n_persocket. Both lama and ppr create synonyms for these variables. This commit was SVN r28236.
2013-03-28 01:09:41 +04:00
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2012 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2012 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "mpi.h"
#include "ompi/constants.h"
- Split the datatype engine into two parts: an MPI specific part in OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
2009-07-13 08:56:31 +04:00
#include "ompi/datatype/ompi_datatype.h"
#include "ompi/communicator/communicator.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/mca/coll/base/coll_tags.h"
#include "ompi/mca/pml/pml.h"
#include "coll_tuned.h"
#include "coll_tuned_topo.h"
#include "coll_tuned_util.h"
MCA/base: Add new MCA variable system Features: - Support for an override parameter file (openmpi-mca-param-override.conf). Variable values in this file can not be overridden by any file or environment value. - Support for boolean, unsigned, and unsigned long long variables. - Support for true/false values. - Support for enumerations on integer variables. - Support for MPIT scope, verbosity, and binding. - Support for command line source. - Support for setting variable source via the environment using OMPI_MCA_SOURCE_<var name>=source (either command or file:filename) - Cleaner API. - Support for variable groups (equivalent to MPIT categories). Notes: - Variables must be created with a backing store (char **, int *, or bool *) that must live at least as long as the variable. - Creating a variable with the MCA_BASE_VAR_FLAG_SETTABLE enables the use of mca_base_var_set_value() to change the value. - String values are duplicated when the variable is registered. It is up to the caller to free the original value if necessary. The new value will be freed by the mca_base_var system and must not be freed by the user. - Variables with constant scope may not be settable. - Variable groups (and all associated variables) are deregistered when the component is closed or the component repository item is freed. This prevents a segmentation fault from accessing a variable after its component is unloaded. - After some discussion we decided we should remove the automatic registration of component priority variables. Few component actually made use of this feature. - The enumerator interface was updated to be general enough to handle future uses of the interface. - The code to generate ompi_info output has been moved into the MCA variable system. See mca_base_var_dump(). opal: update core and components to mca_base_var system orte: update core and components to mca_base_var system ompi: update core and components to mca_base_var system This commit also modifies the rmaps framework. The following variables were moved from ppr and lama: rmaps_base_pernode, rmaps_base_n_pernode, rmaps_base_n_persocket. Both lama and ppr create synonyms for these variables. This commit was SVN r28236.
2013-03-28 01:09:41 +04:00
/* bcast algorithm variables */
static int coll_tuned_bcast_algorithm_count = 6;
static int coll_tuned_bcast_forced_algorithm = 0;
static int coll_tuned_bcast_segment_size = 0;
static int coll_tuned_bcast_tree_fanout;
static int coll_tuned_bcast_chain_fanout;
/* valid values for coll_tuned_bcast_forced_algorithm */
static mca_base_var_enum_value_t bcast_algorithms[] = {
{0, "ignore"},
{1, "basic_linear"},
{2, "chain"},
{3, "pipeline"},
{4, "split_binary_tree"},
{5, "binary_tree"},
{6, "binomial"},
{0, NULL}
};
int
ompi_coll_tuned_bcast_intra_generic( void* buffer,
int original_count,
struct ompi_datatype_t* datatype,
int root,
struct ompi_communicator_t* comm,
mca_coll_base_module_t *module,
uint32_t count_by_segment,
ompi_coll_tree_t* tree )
{
int err = 0, line, i, rank, size, segindex, req_index;
int num_segments; /* Number of segments */
int sendcount; /* number of elements sent in this segment */
size_t realsegsize, type_size;
char *tmpbuf;
ptrdiff_t extent, lb;
ompi_request_t *recv_reqs[2] = {MPI_REQUEST_NULL, MPI_REQUEST_NULL};
#if !defined(COLL_TUNED_BCAST_USE_BLOCKING)
ompi_request_t **send_reqs = NULL;
#endif
size = ompi_comm_size(comm);
rank = ompi_comm_rank(comm);
assert( size > 1 );
- Split the datatype engine into two parts: an MPI specific part in OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
2009-07-13 08:56:31 +04:00
ompi_datatype_get_extent (datatype, &lb, &extent);
ompi_datatype_type_size( datatype, &type_size );
num_segments = (original_count + count_by_segment - 1) / count_by_segment;
realsegsize = (ptrdiff_t)count_by_segment * extent;
/* Set the buffer pointers */
tmpbuf = (char *) buffer;
#if !defined(COLL_TUNED_BCAST_USE_BLOCKING)
if( tree->tree_nextsize != 0 ) {
send_reqs = (ompi_request_t**)malloc( (ptrdiff_t)tree->tree_nextsize *
sizeof(ompi_request_t*) );
}
#endif
/* Root code */
if( rank == root ) {
/*
For each segment:
- send segment to all children.
The last segment may have less elements than other segments.
*/
sendcount = count_by_segment;
for( segindex = 0; segindex < num_segments; segindex++ ) {
if( segindex == (num_segments - 1) ) {
sendcount = original_count - segindex * count_by_segment;
}
for( i = 0; i < tree->tree_nextsize; i++ ) {
#if defined(COLL_TUNED_BCAST_USE_BLOCKING)
err = MCA_PML_CALL(send(tmpbuf, sendcount, datatype,
tree->tree_next[i],
MCA_COLL_BASE_TAG_BCAST,
MCA_PML_BASE_SEND_STANDARD, comm));
#else
err = MCA_PML_CALL(isend(tmpbuf, sendcount, datatype,
tree->tree_next[i],
MCA_COLL_BASE_TAG_BCAST,
MCA_PML_BASE_SEND_STANDARD, comm,
&send_reqs[i]));
#endif /* COLL_TUNED_BCAST_USE_BLOCKING */
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
}
#if !defined(COLL_TUNED_BCAST_USE_BLOCKING)
/* complete the sends before starting the next sends */
err = ompi_request_wait_all( tree->tree_nextsize, send_reqs,
MPI_STATUSES_IGNORE );
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
#endif /* not COLL_TUNED_BCAST_USE_BLOCKING */
/* update tmp buffer */
tmpbuf += realsegsize;
}
}
/* Intermediate nodes code */
else if( tree->tree_nextsize > 0 ) {
/*
Create the pipeline.
1) Post the first receive
2) For segments 1 .. num_segments
- post new receive
- wait on the previous receive to complete
- send this data to children
3) Wait on the last segment
4) Compute number of elements in last segment.
5) Send the last segment to children
*/
req_index = 0;
err = MCA_PML_CALL(irecv(tmpbuf, count_by_segment, datatype,
tree->tree_prev, MCA_COLL_BASE_TAG_BCAST,
comm, &recv_reqs[req_index]));
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
for( segindex = 1; segindex < num_segments; segindex++ ) {
req_index = req_index ^ 0x1;
/* post new irecv */
err = MCA_PML_CALL(irecv( tmpbuf + realsegsize, count_by_segment,
datatype, tree->tree_prev,
MCA_COLL_BASE_TAG_BCAST,
comm, &recv_reqs[req_index]));
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
/* wait for and forward the previous segment to children */
err = ompi_request_wait( &recv_reqs[req_index ^ 0x1],
MPI_STATUSES_IGNORE );
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
for( i = 0; i < tree->tree_nextsize; i++ ) {
#if defined(COLL_TUNED_BCAST_USE_BLOCKING)
err = MCA_PML_CALL(send(tmpbuf, count_by_segment, datatype,
tree->tree_next[i],
MCA_COLL_BASE_TAG_BCAST,
MCA_PML_BASE_SEND_STANDARD, comm));
#else
err = MCA_PML_CALL(isend(tmpbuf, count_by_segment, datatype,
tree->tree_next[i],
MCA_COLL_BASE_TAG_BCAST,
MCA_PML_BASE_SEND_STANDARD, comm,
&send_reqs[i]));
#endif /* COLL_TUNED_BCAST_USE_BLOCKING */
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
}
#if !defined(COLL_TUNED_BCAST_USE_BLOCKING)
/* complete the sends before starting the next iteration */
err = ompi_request_wait_all( tree->tree_nextsize, send_reqs,
MPI_STATUSES_IGNORE );
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
#endif /* COLL_TUNED_BCAST_USE_BLOCKING */
/* Update the receive buffer */
tmpbuf += realsegsize;
}
/* Process the last segment */
err = ompi_request_wait( &recv_reqs[req_index], MPI_STATUSES_IGNORE );
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
sendcount = original_count - (ptrdiff_t)(num_segments - 1) * count_by_segment;
for( i = 0; i < tree->tree_nextsize; i++ ) {
#if defined(COLL_TUNED_BCAST_USE_BLOCKING)
err = MCA_PML_CALL(send(tmpbuf, sendcount, datatype,
tree->tree_next[i],
MCA_COLL_BASE_TAG_BCAST,
MCA_PML_BASE_SEND_STANDARD, comm));
#else
err = MCA_PML_CALL(isend(tmpbuf, sendcount, datatype,
tree->tree_next[i],
MCA_COLL_BASE_TAG_BCAST,
MCA_PML_BASE_SEND_STANDARD, comm,
&send_reqs[i]));
#endif /* COLL_TUNED_BCAST_USE_BLOCKING */
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
}
#if !defined(COLL_TUNED_BCAST_USE_BLOCKING)
err = ompi_request_wait_all( tree->tree_nextsize, send_reqs,
MPI_STATUSES_IGNORE );
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
#endif /* COLL_TUNED_BCAST_USE_BLOCKING */
}
/* Leaf nodes */
else {
/*
Receive all segments from parent in a loop:
1) post irecv for the first segment
2) for segments 1 .. num_segments
- post irecv for the next segment
- wait on the previous segment to arrive
3) wait for the last segment
*/
req_index = 0;
err = MCA_PML_CALL(irecv(tmpbuf, count_by_segment, datatype,
tree->tree_prev, MCA_COLL_BASE_TAG_BCAST,
comm, &recv_reqs[req_index]));
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
for( segindex = 1; segindex < num_segments; segindex++ ) {
req_index = req_index ^ 0x1;
tmpbuf += realsegsize;
/* post receive for the next segment */
err = MCA_PML_CALL(irecv(tmpbuf, count_by_segment, datatype,
tree->tree_prev, MCA_COLL_BASE_TAG_BCAST,
comm, &recv_reqs[req_index]));
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
/* wait on the previous segment */
err = ompi_request_wait( &recv_reqs[req_index ^ 0x1],
MPI_STATUS_IGNORE );
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
}
err = ompi_request_wait( &recv_reqs[req_index], MPI_STATUS_IGNORE );
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
}
#if !defined(COLL_TUNED_BCAST_USE_BLOCKING)
if( NULL != send_reqs ) free(send_reqs);
#endif
return (MPI_SUCCESS);
error_hndl:
OPAL_OUTPUT( (ompi_coll_tuned_stream,"%s:%4d\tError occurred %d, rank %2d",
__FILE__, line, err, rank) );
#if !defined(COLL_TUNED_BCAST_USE_BLOCKING)
if( NULL != send_reqs ) free(send_reqs);
#endif
return (err);
}
int
ompi_coll_tuned_bcast_intra_bintree ( void* buffer,
int count,
struct ompi_datatype_t* datatype,
int root,
struct ompi_communicator_t* comm,
mca_coll_base_module_t *module,
uint32_t segsize )
{
int segcount = count;
size_t typelng;
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
mca_coll_tuned_comm_t *data = tuned_module->tuned_data;
COLL_TUNED_UPDATE_BINTREE( comm, tuned_module, root );
/**
* Determine number of elements sent per operation.
*/
- 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
ompi_datatype_type_size( datatype, &typelng );
COLL_TUNED_COMPUTED_SEGCOUNT( segsize, typelng, segcount );
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:bcast_intra_binary rank %d ss %5d typelng %lu segcount %d",
ompi_comm_rank(comm), segsize, (unsigned long)typelng, segcount));
return ompi_coll_tuned_bcast_intra_generic( buffer, count, datatype, root, comm, module,
segcount, data->cached_bintree );
}
int
ompi_coll_tuned_bcast_intra_pipeline( void* buffer,
int count,
struct ompi_datatype_t* datatype,
int root,
struct ompi_communicator_t* comm,
mca_coll_base_module_t *module,
uint32_t segsize )
{
int segcount = count;
size_t typelng;
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
mca_coll_tuned_comm_t *data = tuned_module->tuned_data;
COLL_TUNED_UPDATE_PIPELINE( comm, tuned_module, root );
/**
* Determine number of elements sent per operation.
*/
- 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
ompi_datatype_type_size( datatype, &typelng );
COLL_TUNED_COMPUTED_SEGCOUNT( segsize, typelng, segcount );
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:bcast_intra_pipeline rank %d ss %5d typelng %lu segcount %d",
ompi_comm_rank(comm), segsize, (unsigned long)typelng, segcount));
return ompi_coll_tuned_bcast_intra_generic( buffer, count, datatype, root, comm, module,
segcount, data->cached_pipeline );
}
int
ompi_coll_tuned_bcast_intra_chain( void* buffer,
int count,
struct ompi_datatype_t* datatype,
int root,
struct ompi_communicator_t* comm,
mca_coll_base_module_t *module,
uint32_t segsize, int32_t chains )
{
int segcount = count;
size_t typelng;
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
mca_coll_tuned_comm_t *data = tuned_module->tuned_data;
COLL_TUNED_UPDATE_CHAIN( comm, tuned_module, root, chains );
/**
* Determine number of elements sent per operation.
*/
- 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
ompi_datatype_type_size( datatype, &typelng );
COLL_TUNED_COMPUTED_SEGCOUNT( segsize, typelng, segcount );
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:bcast_intra_chain rank %d fo %d ss %5d typelng %lu segcount %d",
ompi_comm_rank(comm), chains, segsize, (unsigned long)typelng, segcount));
return ompi_coll_tuned_bcast_intra_generic( buffer, count, datatype, root, comm, module,
segcount, data->cached_chain );
}
int
ompi_coll_tuned_bcast_intra_binomial( void* buffer,
int count,
struct ompi_datatype_t* datatype,
int root,
struct ompi_communicator_t* comm,
mca_coll_base_module_t *module,
uint32_t segsize )
{
int segcount = count;
size_t typelng;
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
mca_coll_tuned_comm_t *data = tuned_module->tuned_data;
COLL_TUNED_UPDATE_BMTREE( comm, tuned_module, root );
/**
* Determine number of elements sent per operation.
*/
- 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
ompi_datatype_type_size( datatype, &typelng );
COLL_TUNED_COMPUTED_SEGCOUNT( segsize, typelng, segcount );
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:bcast_intra_binomial rank %d ss %5d typelng %lu segcount %d",
ompi_comm_rank(comm), segsize, (unsigned long)typelng, segcount));
return ompi_coll_tuned_bcast_intra_generic( buffer, count, datatype, root, comm, module,
segcount, data->cached_bmtree );
}
int
ompi_coll_tuned_bcast_intra_split_bintree ( void* buffer,
int count,
struct ompi_datatype_t* datatype,
int root,
struct ompi_communicator_t* comm,
mca_coll_base_module_t *module,
uint32_t segsize )
{
int err=0, line, rank, size, segindex, i, lr, pair;
uint32_t counts[2];
int segcount[2]; /* Number of elements sent with each segment */
int num_segments[2]; /* Number of segmenets */
int sendcount[2]; /* the same like segcount, except for the last segment */
size_t realsegsize[2], type_size;
char *tmpbuf[2];
ptrdiff_t type_extent, lb;
ompi_request_t *base_req, *new_req;
ompi_coll_tree_t *tree;
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
mca_coll_tuned_comm_t *data = tuned_module->tuned_data;
size = ompi_comm_size(comm);
rank = ompi_comm_rank(comm);
OPAL_OUTPUT((ompi_coll_tuned_stream,"ompi_coll_tuned_bcast_intra_split_bintree rank %d root %d ss %5d", rank, root, segsize));
if (size == 1) {
return MPI_SUCCESS;
}
/* setup the binary tree topology. */
COLL_TUNED_UPDATE_BINTREE( comm, tuned_module, root );
tree = data->cached_bintree;
- 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
err = ompi_datatype_type_size( datatype, &type_size );
/* Determine number of segments and number of elements per segment */
counts[0] = count/2;
if (count % 2 != 0) counts[0]++;
counts[1] = count - counts[0];
if ( segsize > 0 ) {
- 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
/* Note that ompi_datatype_type_size() will never return a negative
value in typelng; it returns an int [vs. an unsigned type]
because of the MPI spec. */
if (segsize < ((uint32_t) type_size)) {
segsize = type_size; /* push segsize up to hold one type */
}
segcount[0] = segcount[1] = segsize / type_size;
num_segments[0] = counts[0]/segcount[0];
if ((counts[0] % segcount[0]) != 0) num_segments[0]++;
num_segments[1] = counts[1]/segcount[1];
if ((counts[1] % segcount[1]) != 0) num_segments[1]++;
} else {
segcount[0] = counts[0];
segcount[1] = counts[1];
num_segments[0] = num_segments[1] = 1;
}
/* if the message is too small to be split into segments */
if( (counts[0] == 0 || counts[1] == 0) ||
(segsize > ((ptrdiff_t)counts[0] * type_size)) ||
(segsize > ((ptrdiff_t)counts[1] * type_size)) ) {
/* call linear version here ! */
return (ompi_coll_tuned_bcast_intra_chain ( buffer, count, datatype,
root, comm, module,
segsize, 1 ));
}
- Split the datatype engine into two parts: an MPI specific part in OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
2009-07-13 08:56:31 +04:00
err = ompi_datatype_get_extent (datatype, &lb, &type_extent);
/* Determine real segment size */
realsegsize[0] = (ptrdiff_t)segcount[0] * type_extent;
realsegsize[1] = (ptrdiff_t)segcount[1] * type_extent;
/* set the buffer pointers */
tmpbuf[0] = (char *) buffer;
tmpbuf[1] = (char *) buffer + (ptrdiff_t)counts[0] * type_extent;
/* Step 1:
Root splits the buffer in 2 and sends segmented message down the branches.
Left subtree of the tree receives first half of the buffer, while right
subtree receives the remaining message.
*/
/* determine if I am left (0) or right (1), (root is right) */
lr = ((rank + size - root)%size + 1)%2;
/* root code */
if( rank == root ) {
/* determine segment count */
sendcount[0] = segcount[0];
sendcount[1] = segcount[1];
/* for each segment */
for (segindex = 0; segindex < num_segments[0]; segindex++) {
/* for each child */
for( i = 0; i < tree->tree_nextsize && i < 2; i++ ) {
if (segindex >= num_segments[i]) { /* no more segments */
continue;
}
/* determine how many elements are being sent in this round */
if(segindex == (num_segments[i] - 1))
sendcount[i] = counts[i] - segindex*segcount[i];
/* send data */
MCA_PML_CALL(send(tmpbuf[i], sendcount[i], datatype,
tree->tree_next[i], MCA_COLL_BASE_TAG_BCAST,
MCA_PML_BASE_SEND_STANDARD, comm));
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
/* update tmp buffer */
tmpbuf[i] += realsegsize[i];
}
}
}
/* intermediate nodes code */
else if( tree->tree_nextsize > 0 ) {
/* Intermediate nodes:
* It will receive segments only from one half of the data.
* Which one is determined by whether the node belongs to the "left" or "right"
* subtree. Topoloby building function builds binary tree such that
* odd "shifted ranks" ((rank + size - root)%size) are on the left subtree,
* and even on the right subtree.
*
* Create the pipeline. We first post the first receive, then in the loop we
* post the next receive and after that wait for the previous receive to complete
* and we disseminating the data to all children.
*/
sendcount[lr] = segcount[lr];
err = MCA_PML_CALL(irecv(tmpbuf[lr], sendcount[lr], datatype,
tree->tree_prev, MCA_COLL_BASE_TAG_BCAST,
comm, &base_req));
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
for( segindex = 1; segindex < num_segments[lr]; segindex++ ) {
/* determine how many elements to expect in this round */
if( segindex == (num_segments[lr] - 1))
sendcount[lr] = counts[lr] - (ptrdiff_t)segindex * (ptrdiff_t)segcount[lr];
/* post new irecv */
err = MCA_PML_CALL(irecv( tmpbuf[lr] + realsegsize[lr], sendcount[lr],
datatype, tree->tree_prev, MCA_COLL_BASE_TAG_BCAST,
comm, &new_req));
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
/* wait for and forward current segment */
err = ompi_request_wait_all( 1, &base_req, MPI_STATUSES_IGNORE );
for( i = 0; i < tree->tree_nextsize; i++ ) { /* send data to children (segcount[lr]) */
err = MCA_PML_CALL(send( tmpbuf[lr], segcount[lr], datatype,
tree->tree_next[i], MCA_COLL_BASE_TAG_BCAST,
MCA_PML_BASE_SEND_STANDARD, comm));
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
} /* end of for each child */
/* upate the base request */
base_req = new_req;
/* go to the next buffer (ie. the one corresponding to the next recv) */
tmpbuf[lr] += realsegsize[lr];
} /* end of for segindex */
/* wait for the last segment and forward current segment */
err = ompi_request_wait_all( 1, &base_req, MPI_STATUSES_IGNORE );
for( i = 0; i < tree->tree_nextsize; i++ ) { /* send data to children */
err = MCA_PML_CALL(send(tmpbuf[lr], sendcount[lr], datatype,
tree->tree_next[i], MCA_COLL_BASE_TAG_BCAST,
MCA_PML_BASE_SEND_STANDARD, comm));
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
} /* end of for each child */
}
/* leaf nodes */
else {
/* Just consume segments as fast as possible */
sendcount[lr] = segcount[lr];
for (segindex = 0; segindex < num_segments[lr]; segindex++) {
/* determine how many elements to expect in this round */
if (segindex == (num_segments[lr] - 1))
sendcount[lr] = counts[lr] - (ptrdiff_t)segindex * (ptrdiff_t)segcount[lr];
/* receive segments */
err = MCA_PML_CALL(recv(tmpbuf[lr], sendcount[lr], datatype,
tree->tree_prev, MCA_COLL_BASE_TAG_BCAST,
comm, MPI_STATUS_IGNORE));
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
/* update the initial pointer to the buffer */
tmpbuf[lr] += realsegsize[lr];
}
}
/* reset the buffer pointers */
tmpbuf[0] = (char *) buffer;
tmpbuf[1] = (char *) buffer + (ptrdiff_t)counts[0] * type_extent;
/* Step 2:
Find your immediate pair (identical node in opposite subtree) and SendRecv
data buffer with them.
The tree building function ensures that
if (we are not root)
if we are in the left subtree (lr == 0) our pair is (rank+1)%size.
if we are in the right subtree (lr == 1) our pair is (rank-1)%size
If we have even number of nodes the rank (size-1) will pair up with root.
*/
if (lr == 0) {
pair = (rank+1)%size;
} else {
pair = (rank+size-1)%size;
}
if ( (size%2) != 0 && rank != root) {
err = ompi_coll_tuned_sendrecv( tmpbuf[lr], counts[lr], datatype,
pair, MCA_COLL_BASE_TAG_BCAST,
tmpbuf[(lr+1)%2], counts[(lr+1)%2], datatype,
pair, MCA_COLL_BASE_TAG_BCAST,
comm, MPI_STATUS_IGNORE, rank);
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
} else if ( (size%2) == 0 ) {
/* root sends right buffer to the last node */
if( rank == root ) {
err = MCA_PML_CALL(send(tmpbuf[1], counts[1], datatype,
(root+size-1)%size, MCA_COLL_BASE_TAG_BCAST,
MCA_PML_BASE_SEND_STANDARD, comm));
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
}
/* last node receives right buffer from the root */
else if (rank == (root+size-1)%size) {
err = MCA_PML_CALL(recv(tmpbuf[1], counts[1], datatype,
root, MCA_COLL_BASE_TAG_BCAST,
comm, MPI_STATUS_IGNORE));
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
}
/* everyone else exchanges buffers */
else {
err = ompi_coll_tuned_sendrecv( tmpbuf[lr], counts[lr], datatype,
pair, MCA_COLL_BASE_TAG_BCAST,
tmpbuf[(lr+1)%2], counts[(lr+1)%2], datatype,
pair, MCA_COLL_BASE_TAG_BCAST,
comm, MPI_STATUS_IGNORE, rank);
if (err != MPI_SUCCESS) { line = __LINE__; goto error_hndl; }
}
}
return (MPI_SUCCESS);
error_hndl:
OPAL_OUTPUT((ompi_coll_tuned_stream,"%s:%4d\tError occurred %d, rank %2d", __FILE__,line,err,rank));
return (err);
}
/*
* Linear functions are copied from the BASIC coll module
* they do not segment the message and are simple implementations
* but for some small number of nodes and/or small data sizes they
* are just as fast as tuned/tree based segmenting operations
* and as such may be selected by the decision functions
* These are copied into this module due to the way we select modules
* in V1. i.e. in V2 we will handle this differently and so will not
* have to duplicate code.
* GEF Oct05 after asking Jeff.
*/
/* copied function (with appropriate renaming) starts here */
/*
* bcast_lin_intra
*
* Function: - broadcast using O(N) algorithm
* Accepts: - same arguments as MPI_Bcast()
* Returns: - MPI_SUCCESS or error code
*/
int
ompi_coll_tuned_bcast_intra_basic_linear (void *buff, int count,
struct ompi_datatype_t *datatype, int root,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
int i, size, rank, err;
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
mca_coll_tuned_comm_t *data = tuned_module->tuned_data;
ompi_request_t **preq, **reqs = data->mcct_reqs;
size = ompi_comm_size(comm);
rank = ompi_comm_rank(comm);
OPAL_OUTPUT((ompi_coll_tuned_stream,"ompi_coll_tuned_bcast_intra_basic_linear rank %d root %d", rank, root));
/* Non-root receive the data. */
if (rank != root) {
return MCA_PML_CALL(recv(buff, count, datatype, root,
MCA_COLL_BASE_TAG_BCAST, comm,
MPI_STATUS_IGNORE));
}
/* Root sends data to all others. */
for (i = 0, preq = reqs; i < size; ++i) {
if (i == rank) {
continue;
}
err = MCA_PML_CALL(isend_init(buff, count, datatype, i,
MCA_COLL_BASE_TAG_BCAST,
MCA_PML_BASE_SEND_STANDARD,
comm, preq++));
if (MPI_SUCCESS != err) {
return err;
}
}
--i;
/* Start your engines. This will never return an error. */
MCA_PML_CALL(start(i, reqs));
/* Wait for them all. If there's an error, note that we don't
* care what the error was -- just that there *was* an error. The
* PML will finish all requests, even if one or more of them fail.
* i.e., by the end of this call, all the requests are free-able.
* So free them anyway -- even if there was an error, and return
* the error after we free everything. */
err = ompi_request_wait_all(i, reqs, MPI_STATUSES_IGNORE);
/* Free the reqs */
ompi_coll_tuned_free_reqs(reqs, i);
/* All done */
return err;
}
/* copied function (with appropriate renaming) ends here */
/* The following are used by dynamic and forced rules */
/* publish details of each algorithm and if its forced/fixed/locked in */
/* as you add methods/algorithms you must update this and the query/map routines */
/* this routine is called by the component only */
/* this makes sure that the mca parameters are set to their initial values and perms */
/* module does not call this they call the forced_getvalues routine instead */
int ompi_coll_tuned_bcast_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices)
{
MCA/base: Add new MCA variable system Features: - Support for an override parameter file (openmpi-mca-param-override.conf). Variable values in this file can not be overridden by any file or environment value. - Support for boolean, unsigned, and unsigned long long variables. - Support for true/false values. - Support for enumerations on integer variables. - Support for MPIT scope, verbosity, and binding. - Support for command line source. - Support for setting variable source via the environment using OMPI_MCA_SOURCE_<var name>=source (either command or file:filename) - Cleaner API. - Support for variable groups (equivalent to MPIT categories). Notes: - Variables must be created with a backing store (char **, int *, or bool *) that must live at least as long as the variable. - Creating a variable with the MCA_BASE_VAR_FLAG_SETTABLE enables the use of mca_base_var_set_value() to change the value. - String values are duplicated when the variable is registered. It is up to the caller to free the original value if necessary. The new value will be freed by the mca_base_var system and must not be freed by the user. - Variables with constant scope may not be settable. - Variable groups (and all associated variables) are deregistered when the component is closed or the component repository item is freed. This prevents a segmentation fault from accessing a variable after its component is unloaded. - After some discussion we decided we should remove the automatic registration of component priority variables. Few component actually made use of this feature. - The enumerator interface was updated to be general enough to handle future uses of the interface. - The code to generate ompi_info output has been moved into the MCA variable system. See mca_base_var_dump(). opal: update core and components to mca_base_var system orte: update core and components to mca_base_var system ompi: update core and components to mca_base_var system This commit also modifies the rmaps framework. The following variables were moved from ppr and lama: rmaps_base_pernode, rmaps_base_n_pernode, rmaps_base_n_persocket. Both lama and ppr create synonyms for these variables. This commit was SVN r28236.
2013-03-28 01:09:41 +04:00
mca_base_var_enum_t *new_enum;
ompi_coll_tuned_forced_max_algorithms[BCAST] = coll_tuned_bcast_algorithm_count;
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"bcast_algorithm_count",
"Number of bcast algorithms available",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_DEFAULT_ONLY,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_CONSTANT,
&coll_tuned_bcast_algorithm_count);
/* MPI_T: This variable should eventually be bound to a communicator */
coll_tuned_bcast_forced_algorithm = 0;
(void) mca_base_var_enum_create("coll_tuned_bcast_algorithms", bcast_algorithms, &new_enum);
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"bcast_algorithm",
"Which bcast algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 chain, 3: pipeline, 4: split binary tree, 5: binary tree, 6: binomial tree.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_bcast_forced_algorithm);
OBJ_RELEASE(new_enum);
if (mca_param_indices->algorithm_param_index < 0) {
return mca_param_indices->algorithm_param_index;
}
2006-10-24 02:29:17 +04:00
MCA/base: Add new MCA variable system Features: - Support for an override parameter file (openmpi-mca-param-override.conf). Variable values in this file can not be overridden by any file or environment value. - Support for boolean, unsigned, and unsigned long long variables. - Support for true/false values. - Support for enumerations on integer variables. - Support for MPIT scope, verbosity, and binding. - Support for command line source. - Support for setting variable source via the environment using OMPI_MCA_SOURCE_<var name>=source (either command or file:filename) - Cleaner API. - Support for variable groups (equivalent to MPIT categories). Notes: - Variables must be created with a backing store (char **, int *, or bool *) that must live at least as long as the variable. - Creating a variable with the MCA_BASE_VAR_FLAG_SETTABLE enables the use of mca_base_var_set_value() to change the value. - String values are duplicated when the variable is registered. It is up to the caller to free the original value if necessary. The new value will be freed by the mca_base_var system and must not be freed by the user. - Variables with constant scope may not be settable. - Variable groups (and all associated variables) are deregistered when the component is closed or the component repository item is freed. This prevents a segmentation fault from accessing a variable after its component is unloaded. - After some discussion we decided we should remove the automatic registration of component priority variables. Few component actually made use of this feature. - The enumerator interface was updated to be general enough to handle future uses of the interface. - The code to generate ompi_info output has been moved into the MCA variable system. See mca_base_var_dump(). opal: update core and components to mca_base_var system orte: update core and components to mca_base_var system ompi: update core and components to mca_base_var system This commit also modifies the rmaps framework. The following variables were moved from ppr and lama: rmaps_base_pernode, rmaps_base_n_pernode, rmaps_base_n_persocket. Both lama and ppr create synonyms for these variables. This commit was SVN r28236.
2013-03-28 01:09:41 +04:00
coll_tuned_bcast_segment_size = 0;
mca_param_indices->segsize_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"bcast_algorithm_segmentsize",
"Segment size in bytes used by default for bcast algorithms. Only has meaning if algorithm is forced and supports segmenting. 0 bytes means no segmentation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_bcast_segment_size);
coll_tuned_bcast_tree_fanout = ompi_coll_tuned_init_tree_fanout; /* get system wide default */
mca_param_indices->tree_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"bcast_algorithm_tree_fanout",
"Fanout for n-tree used for bcast algorithms. Only has meaning if algorithm is forced and supports n-tree topo based operation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_bcast_tree_fanout);
coll_tuned_bcast_chain_fanout = ompi_coll_tuned_init_chain_fanout; /* get system wide default */
mca_param_indices->chain_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"bcast_algorithm_chain_fanout",
"Fanout for chains used for bcast algorithms. Only has meaning if algorithm is forced and supports chain topo based operation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_bcast_chain_fanout);
return (MPI_SUCCESS);
}
int ompi_coll_tuned_bcast_intra_do_forced(void *buf, int count,
struct ompi_datatype_t *dtype,
int root,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
mca_coll_tuned_comm_t *data = tuned_module->tuned_data;
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:bcast_intra_do_forced algorithm %d",
data->user_forced[BCAST].algorithm));
switch (data->user_forced[BCAST].algorithm) {
case (0): return ompi_coll_tuned_bcast_intra_dec_fixed( buf, count, dtype, root, comm, module );
case (1): return ompi_coll_tuned_bcast_intra_basic_linear( buf, count, dtype, root, comm, module );
case (2): return ompi_coll_tuned_bcast_intra_chain( buf, count, dtype, root, comm, module,
data->user_forced[BCAST].segsize,
data->user_forced[BCAST].chain_fanout );
case (3): return ompi_coll_tuned_bcast_intra_pipeline( buf, count, dtype, root, comm, module,
data->user_forced[BCAST].segsize );
case (4): return ompi_coll_tuned_bcast_intra_split_bintree( buf, count, dtype, root, comm, module,
data->user_forced[BCAST].segsize );
case (5): return ompi_coll_tuned_bcast_intra_bintree( buf, count, dtype, root, comm, module,
data->user_forced[BCAST].segsize );
case (6): return ompi_coll_tuned_bcast_intra_binomial( buf, count, dtype, root, comm, module,
data->user_forced[BCAST].segsize );
default:
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:bcast_intra_do_forced attempt to select algorithm %d when only 0-%d is valid?",
data->user_forced[BCAST].algorithm, ompi_coll_tuned_forced_max_algorithms[BCAST]));
} /* switch */
return (MPI_ERR_ARG);
}
int ompi_coll_tuned_bcast_intra_do_this(void *buf, int count,
struct ompi_datatype_t *dtype,
int root,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module,
int algorithm, int faninout, int segsize)
{
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:bcast_intra_do_this algorithm %d topo faninout %d segsize %d",
algorithm, faninout, segsize));
switch (algorithm) {
case (0): return ompi_coll_tuned_bcast_intra_dec_fixed( buf, count, dtype, root, comm, module );
case (1): return ompi_coll_tuned_bcast_intra_basic_linear( buf, count, dtype, root, comm, module );
case (2): return ompi_coll_tuned_bcast_intra_chain( buf, count, dtype, root, comm, module, segsize, faninout );
case (3): return ompi_coll_tuned_bcast_intra_pipeline( buf, count, dtype, root, comm, module, segsize );
case (4): return ompi_coll_tuned_bcast_intra_split_bintree( buf, count, dtype, root, comm, module, segsize );
case (5): return ompi_coll_tuned_bcast_intra_bintree( buf, count, dtype, root, comm, module, segsize );
case (6): return ompi_coll_tuned_bcast_intra_binomial( buf, count, dtype, root, comm, module, segsize );
default:
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:bcast_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",
algorithm, ompi_coll_tuned_forced_max_algorithms[BCAST]));
} /* switch */
return (MPI_ERR_ARG);
}