1
1
openmpi/ompi/mca/coll/sm/coll_sm_component.c

233 строки
7.9 KiB
C
Исходник Обычный вид История

/*
* Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2005 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* 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) 2008-2009 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/**
* @file
*
* Most of the description of the data layout is in the
* coll_sm_module.c file.
*/
#include "ompi_config.h"
#include "ompi/constants.h"
#include "ompi/mca/coll/coll.h"
#include "orte/util/show_help.h"
#include "coll_sm.h"
/*
* Public string showing the coll ompi_sm component version number
*/
const char *mca_coll_sm_component_version_string =
"Open MPI sm collective MCA component version " OMPI_VERSION;
/*
* Local functions
*/
static int sm_close(void);
Fixes trac:1988. The little bug that turned out to be huge. Yoinks. * Various cosmetic/style updates in the btl sm * Clean up concept of mpool module (I think that code was written way back when the concept of "modules" was fuzzy) * Bring over some old fixes from the /tmp/timattox-sm-coll/ tree to fix potential segv's when mmap'ed regions were at different addresses in different processes (thanks Tim!). * Change sm coll to no longer use mpool as its main source of shmem; rather, just mmap its own segment (because it's fixed size -- there was nothing to be gained by using mpool; shedding the use of mpool saved a lot of complexity in the sm coll setup). This effectively made Tim's fixes moot (because now everything is an offset into the mmap that is computed locally; there are no global pointers). :-) * Slightly updated common/sm to allow making mmap's for a specific set of procs (vs. ''all'' procs in the process). This potentially allows for same-host-inter-proc mmaps -- yay! * Fixed many, many things in the coll sm (particularly in reduce): * Fixed handling of MPI_IN_PLACE in reduce and allreduce * Fixed handling of non-contiguous datatypes in reduce * Changed the order of reductions to go from process (n-1)'s data to process 0's data, because that's how all other OMPI coll components work * Fixed lots of usage of ddt functions * When using a non-contiguous datatype, if the root process is not (n-1), now we used a 2nd convertor to copy from shmem to the rbuf (saves a memory copy vs. what was done before) * Lots and lots of little cleanups, clarifications, and minor optimizations (although still more could be done -- e.g., I think the use of write memory barriers is fairly sub-optimal; they could be ganged together at the root, for example) I'm marking this as "fixes trac:1988" and closing the ticket; if something is still broken, we can re-open the ticket. This commit was SVN r21967. The following Trac tickets were found above: Ticket 1988 --> https://svn.open-mpi.org/trac/ompi/ticket/1988
2009-09-15 04:25:21 +04:00
static int sm_register(void);
/*
* Instantiate the public struct with all of our public information
* and pointers to our public functions in it
*/
mca_coll_sm_component_t mca_coll_sm_component = {
/* First, fill in the super */
{
/* First, the mca_component_t struct containing meta
information about the component itself */
{
MCA_COLL_BASE_VERSION_2_0_0,
/* Component name and version */
"sm",
OMPI_MAJOR_VERSION,
OMPI_MINOR_VERSION,
OMPI_RELEASE_VERSION,
Fixes trac:1988. The little bug that turned out to be huge. Yoinks. * Various cosmetic/style updates in the btl sm * Clean up concept of mpool module (I think that code was written way back when the concept of "modules" was fuzzy) * Bring over some old fixes from the /tmp/timattox-sm-coll/ tree to fix potential segv's when mmap'ed regions were at different addresses in different processes (thanks Tim!). * Change sm coll to no longer use mpool as its main source of shmem; rather, just mmap its own segment (because it's fixed size -- there was nothing to be gained by using mpool; shedding the use of mpool saved a lot of complexity in the sm coll setup). This effectively made Tim's fixes moot (because now everything is an offset into the mmap that is computed locally; there are no global pointers). :-) * Slightly updated common/sm to allow making mmap's for a specific set of procs (vs. ''all'' procs in the process). This potentially allows for same-host-inter-proc mmaps -- yay! * Fixed many, many things in the coll sm (particularly in reduce): * Fixed handling of MPI_IN_PLACE in reduce and allreduce * Fixed handling of non-contiguous datatypes in reduce * Changed the order of reductions to go from process (n-1)'s data to process 0's data, because that's how all other OMPI coll components work * Fixed lots of usage of ddt functions * When using a non-contiguous datatype, if the root process is not (n-1), now we used a 2nd convertor to copy from shmem to the rbuf (saves a memory copy vs. what was done before) * Lots and lots of little cleanups, clarifications, and minor optimizations (although still more could be done -- e.g., I think the use of write memory barriers is fairly sub-optimal; they could be ganged together at the root, for example) I'm marking this as "fixes trac:1988" and closing the ticket; if something is still broken, we can re-open the ticket. This commit was SVN r21967. The following Trac tickets were found above: Ticket 1988 --> https://svn.open-mpi.org/trac/ompi/ticket/1988
2009-09-15 04:25:21 +04:00
/* Component functions */
NULL, /* open */
sm_close,
Fixes trac:1988. The little bug that turned out to be huge. Yoinks. * Various cosmetic/style updates in the btl sm * Clean up concept of mpool module (I think that code was written way back when the concept of "modules" was fuzzy) * Bring over some old fixes from the /tmp/timattox-sm-coll/ tree to fix potential segv's when mmap'ed regions were at different addresses in different processes (thanks Tim!). * Change sm coll to no longer use mpool as its main source of shmem; rather, just mmap its own segment (because it's fixed size -- there was nothing to be gained by using mpool; shedding the use of mpool saved a lot of complexity in the sm coll setup). This effectively made Tim's fixes moot (because now everything is an offset into the mmap that is computed locally; there are no global pointers). :-) * Slightly updated common/sm to allow making mmap's for a specific set of procs (vs. ''all'' procs in the process). This potentially allows for same-host-inter-proc mmaps -- yay! * Fixed many, many things in the coll sm (particularly in reduce): * Fixed handling of MPI_IN_PLACE in reduce and allreduce * Fixed handling of non-contiguous datatypes in reduce * Changed the order of reductions to go from process (n-1)'s data to process 0's data, because that's how all other OMPI coll components work * Fixed lots of usage of ddt functions * When using a non-contiguous datatype, if the root process is not (n-1), now we used a 2nd convertor to copy from shmem to the rbuf (saves a memory copy vs. what was done before) * Lots and lots of little cleanups, clarifications, and minor optimizations (although still more could be done -- e.g., I think the use of write memory barriers is fairly sub-optimal; they could be ganged together at the root, for example) I'm marking this as "fixes trac:1988" and closing the ticket; if something is still broken, we can re-open the ticket. This commit was SVN r21967. The following Trac tickets were found above: Ticket 1988 --> https://svn.open-mpi.org/trac/ompi/ticket/1988
2009-09-15 04:25:21 +04:00
NULL, /* query */
sm_register
},
{
/* The component is not checkpoint ready */
MCA_BASE_METADATA_PARAM_NONE
},
/* Initialization / querying functions */
mca_coll_sm_init_query,
mca_coll_sm_comm_query,
},
/* sm-component specifc information */
/* (default) priority */
/* JMS temporarily lowered until we can get more testing */
0,
/* (default) control size (bytes) */
4096,
/* (default) number of "in use" flags for each communicator's area
Fixes trac:1988. The little bug that turned out to be huge. Yoinks. * Various cosmetic/style updates in the btl sm * Clean up concept of mpool module (I think that code was written way back when the concept of "modules" was fuzzy) * Bring over some old fixes from the /tmp/timattox-sm-coll/ tree to fix potential segv's when mmap'ed regions were at different addresses in different processes (thanks Tim!). * Change sm coll to no longer use mpool as its main source of shmem; rather, just mmap its own segment (because it's fixed size -- there was nothing to be gained by using mpool; shedding the use of mpool saved a lot of complexity in the sm coll setup). This effectively made Tim's fixes moot (because now everything is an offset into the mmap that is computed locally; there are no global pointers). :-) * Slightly updated common/sm to allow making mmap's for a specific set of procs (vs. ''all'' procs in the process). This potentially allows for same-host-inter-proc mmaps -- yay! * Fixed many, many things in the coll sm (particularly in reduce): * Fixed handling of MPI_IN_PLACE in reduce and allreduce * Fixed handling of non-contiguous datatypes in reduce * Changed the order of reductions to go from process (n-1)'s data to process 0's data, because that's how all other OMPI coll components work * Fixed lots of usage of ddt functions * When using a non-contiguous datatype, if the root process is not (n-1), now we used a 2nd convertor to copy from shmem to the rbuf (saves a memory copy vs. what was done before) * Lots and lots of little cleanups, clarifications, and minor optimizations (although still more could be done -- e.g., I think the use of write memory barriers is fairly sub-optimal; they could be ganged together at the root, for example) I'm marking this as "fixes trac:1988" and closing the ticket; if something is still broken, we can re-open the ticket. This commit was SVN r21967. The following Trac tickets were found above: Ticket 1988 --> https://svn.open-mpi.org/trac/ompi/ticket/1988
2009-09-15 04:25:21 +04:00
in the per-communicator shmem segment */
2,
Fixes trac:1988. The little bug that turned out to be huge. Yoinks. * Various cosmetic/style updates in the btl sm * Clean up concept of mpool module (I think that code was written way back when the concept of "modules" was fuzzy) * Bring over some old fixes from the /tmp/timattox-sm-coll/ tree to fix potential segv's when mmap'ed regions were at different addresses in different processes (thanks Tim!). * Change sm coll to no longer use mpool as its main source of shmem; rather, just mmap its own segment (because it's fixed size -- there was nothing to be gained by using mpool; shedding the use of mpool saved a lot of complexity in the sm coll setup). This effectively made Tim's fixes moot (because now everything is an offset into the mmap that is computed locally; there are no global pointers). :-) * Slightly updated common/sm to allow making mmap's for a specific set of procs (vs. ''all'' procs in the process). This potentially allows for same-host-inter-proc mmaps -- yay! * Fixed many, many things in the coll sm (particularly in reduce): * Fixed handling of MPI_IN_PLACE in reduce and allreduce * Fixed handling of non-contiguous datatypes in reduce * Changed the order of reductions to go from process (n-1)'s data to process 0's data, because that's how all other OMPI coll components work * Fixed lots of usage of ddt functions * When using a non-contiguous datatype, if the root process is not (n-1), now we used a 2nd convertor to copy from shmem to the rbuf (saves a memory copy vs. what was done before) * Lots and lots of little cleanups, clarifications, and minor optimizations (although still more could be done -- e.g., I think the use of write memory barriers is fairly sub-optimal; they could be ganged together at the root, for example) I'm marking this as "fixes trac:1988" and closing the ticket; if something is still broken, we can re-open the ticket. This commit was SVN r21967. The following Trac tickets were found above: Ticket 1988 --> https://svn.open-mpi.org/trac/ompi/ticket/1988
2009-09-15 04:25:21 +04:00
/* (default) number of segments for each communicator in the
per-communicator shmem segment */
8,
/* (default) fragment size */
8192,
/* (default) degree of tree for tree-based operations (must be <=
control unit size) */
4,
/* (default) number of processes in coll_sm_shared_mem_size
information variable */
4,
/* default values for non-MCA parameters */
Fixes trac:1988. The little bug that turned out to be huge. Yoinks. * Various cosmetic/style updates in the btl sm * Clean up concept of mpool module (I think that code was written way back when the concept of "modules" was fuzzy) * Bring over some old fixes from the /tmp/timattox-sm-coll/ tree to fix potential segv's when mmap'ed regions were at different addresses in different processes (thanks Tim!). * Change sm coll to no longer use mpool as its main source of shmem; rather, just mmap its own segment (because it's fixed size -- there was nothing to be gained by using mpool; shedding the use of mpool saved a lot of complexity in the sm coll setup). This effectively made Tim's fixes moot (because now everything is an offset into the mmap that is computed locally; there are no global pointers). :-) * Slightly updated common/sm to allow making mmap's for a specific set of procs (vs. ''all'' procs in the process). This potentially allows for same-host-inter-proc mmaps -- yay! * Fixed many, many things in the coll sm (particularly in reduce): * Fixed handling of MPI_IN_PLACE in reduce and allreduce * Fixed handling of non-contiguous datatypes in reduce * Changed the order of reductions to go from process (n-1)'s data to process 0's data, because that's how all other OMPI coll components work * Fixed lots of usage of ddt functions * When using a non-contiguous datatype, if the root process is not (n-1), now we used a 2nd convertor to copy from shmem to the rbuf (saves a memory copy vs. what was done before) * Lots and lots of little cleanups, clarifications, and minor optimizations (although still more could be done -- e.g., I think the use of write memory barriers is fairly sub-optimal; they could be ganged together at the root, for example) I'm marking this as "fixes trac:1988" and closing the ticket; if something is still broken, we can re-open the ticket. This commit was SVN r21967. The following Trac tickets were found above: Ticket 1988 --> https://svn.open-mpi.org/trac/ompi/ticket/1988
2009-09-15 04:25:21 +04:00
/* Not specifying values here gives us all 0's */
};
/*
* Shut down the component
*/
static int sm_close(void)
{
mca_base_component_t *c = &mca_coll_sm_component.super.collm_version;
/* Common SM MCA params */
mca_common_sm_param_unregister(c);
return OMPI_SUCCESS;
}
/*
Fixes trac:1988. The little bug that turned out to be huge. Yoinks. * Various cosmetic/style updates in the btl sm * Clean up concept of mpool module (I think that code was written way back when the concept of "modules" was fuzzy) * Bring over some old fixes from the /tmp/timattox-sm-coll/ tree to fix potential segv's when mmap'ed regions were at different addresses in different processes (thanks Tim!). * Change sm coll to no longer use mpool as its main source of shmem; rather, just mmap its own segment (because it's fixed size -- there was nothing to be gained by using mpool; shedding the use of mpool saved a lot of complexity in the sm coll setup). This effectively made Tim's fixes moot (because now everything is an offset into the mmap that is computed locally; there are no global pointers). :-) * Slightly updated common/sm to allow making mmap's for a specific set of procs (vs. ''all'' procs in the process). This potentially allows for same-host-inter-proc mmaps -- yay! * Fixed many, many things in the coll sm (particularly in reduce): * Fixed handling of MPI_IN_PLACE in reduce and allreduce * Fixed handling of non-contiguous datatypes in reduce * Changed the order of reductions to go from process (n-1)'s data to process 0's data, because that's how all other OMPI coll components work * Fixed lots of usage of ddt functions * When using a non-contiguous datatype, if the root process is not (n-1), now we used a 2nd convertor to copy from shmem to the rbuf (saves a memory copy vs. what was done before) * Lots and lots of little cleanups, clarifications, and minor optimizations (although still more could be done -- e.g., I think the use of write memory barriers is fairly sub-optimal; they could be ganged together at the root, for example) I'm marking this as "fixes trac:1988" and closing the ticket; if something is still broken, we can re-open the ticket. This commit was SVN r21967. The following Trac tickets were found above: Ticket 1988 --> https://svn.open-mpi.org/trac/ompi/ticket/1988
2009-09-15 04:25:21 +04:00
* Register MCA params
*/
Fixes trac:1988. The little bug that turned out to be huge. Yoinks. * Various cosmetic/style updates in the btl sm * Clean up concept of mpool module (I think that code was written way back when the concept of "modules" was fuzzy) * Bring over some old fixes from the /tmp/timattox-sm-coll/ tree to fix potential segv's when mmap'ed regions were at different addresses in different processes (thanks Tim!). * Change sm coll to no longer use mpool as its main source of shmem; rather, just mmap its own segment (because it's fixed size -- there was nothing to be gained by using mpool; shedding the use of mpool saved a lot of complexity in the sm coll setup). This effectively made Tim's fixes moot (because now everything is an offset into the mmap that is computed locally; there are no global pointers). :-) * Slightly updated common/sm to allow making mmap's for a specific set of procs (vs. ''all'' procs in the process). This potentially allows for same-host-inter-proc mmaps -- yay! * Fixed many, many things in the coll sm (particularly in reduce): * Fixed handling of MPI_IN_PLACE in reduce and allreduce * Fixed handling of non-contiguous datatypes in reduce * Changed the order of reductions to go from process (n-1)'s data to process 0's data, because that's how all other OMPI coll components work * Fixed lots of usage of ddt functions * When using a non-contiguous datatype, if the root process is not (n-1), now we used a 2nd convertor to copy from shmem to the rbuf (saves a memory copy vs. what was done before) * Lots and lots of little cleanups, clarifications, and minor optimizations (although still more could be done -- e.g., I think the use of write memory barriers is fairly sub-optimal; they could be ganged together at the root, for example) I'm marking this as "fixes trac:1988" and closing the ticket; if something is still broken, we can re-open the ticket. This commit was SVN r21967. The following Trac tickets were found above: Ticket 1988 --> https://svn.open-mpi.org/trac/ompi/ticket/1988
2009-09-15 04:25:21 +04:00
static int sm_register(void)
{
Fixes trac:1988. The little bug that turned out to be huge. Yoinks. * Various cosmetic/style updates in the btl sm * Clean up concept of mpool module (I think that code was written way back when the concept of "modules" was fuzzy) * Bring over some old fixes from the /tmp/timattox-sm-coll/ tree to fix potential segv's when mmap'ed regions were at different addresses in different processes (thanks Tim!). * Change sm coll to no longer use mpool as its main source of shmem; rather, just mmap its own segment (because it's fixed size -- there was nothing to be gained by using mpool; shedding the use of mpool saved a lot of complexity in the sm coll setup). This effectively made Tim's fixes moot (because now everything is an offset into the mmap that is computed locally; there are no global pointers). :-) * Slightly updated common/sm to allow making mmap's for a specific set of procs (vs. ''all'' procs in the process). This potentially allows for same-host-inter-proc mmaps -- yay! * Fixed many, many things in the coll sm (particularly in reduce): * Fixed handling of MPI_IN_PLACE in reduce and allreduce * Fixed handling of non-contiguous datatypes in reduce * Changed the order of reductions to go from process (n-1)'s data to process 0's data, because that's how all other OMPI coll components work * Fixed lots of usage of ddt functions * When using a non-contiguous datatype, if the root process is not (n-1), now we used a 2nd convertor to copy from shmem to the rbuf (saves a memory copy vs. what was done before) * Lots and lots of little cleanups, clarifications, and minor optimizations (although still more could be done -- e.g., I think the use of write memory barriers is fairly sub-optimal; they could be ganged together at the root, for example) I'm marking this as "fixes trac:1988" and closing the ticket; if something is still broken, we can re-open the ticket. This commit was SVN r21967. The following Trac tickets were found above: Ticket 1988 --> https://svn.open-mpi.org/trac/ompi/ticket/1988
2009-09-15 04:25:21 +04:00
size_t size;
mca_base_component_t *c = &mca_coll_sm_component.super.collm_version;
mca_coll_sm_component_t *cs = &mca_coll_sm_component;
/* If we want to be selected (i.e., all procs on one node), then
we should have a high priority */
mca_base_param_reg_int(c, "priority",
"Priority of the sm coll component",
false, false,
cs->sm_priority,
&cs->sm_priority);
mca_base_param_reg_int(c, "control_size",
"Length of the control data -- should usually be either the length of a cache line on most SMPs, or the size of a page on machines that support direct memory affinity page placement (in bytes)",
false, false,
cs->sm_control_size,
&cs->sm_control_size);
mca_base_param_reg_int(c, "fragment_size",
"Fragment size (in bytes) used for passing data through shared memory (will be rounded up to the nearest control_size size)",
false, false,
cs->sm_fragment_size,
&cs->sm_fragment_size);
if (0 != (cs->sm_fragment_size % cs->sm_control_size)) {
cs->sm_fragment_size += cs->sm_control_size -
(cs->sm_fragment_size % cs->sm_control_size);
}
mca_base_param_reg_int(c, "comm_in_use_flags",
"Number of \"in use\" flags, used to mark a message passing area segment as currently being used or not (must be >= 2 and <= comm_num_segments)",
false, false,
cs->sm_comm_num_in_use_flags,
&cs->sm_comm_num_in_use_flags);
if (cs->sm_comm_num_in_use_flags < 2) {
cs->sm_comm_num_in_use_flags = 2;
}
mca_base_param_reg_int(c, "comm_num_segments",
"Number of segments in each communicator's shared memory message passing area (must be >= 2, and must be a multiple of comm_in_use_flags)",
false, false,
cs->sm_comm_num_segments,
&cs->sm_comm_num_segments);
if (cs->sm_comm_num_segments < cs->sm_comm_num_in_use_flags) {
cs->sm_comm_num_segments = cs->sm_comm_num_in_use_flags;
}
if (0 != (cs->sm_comm_num_segments % cs->sm_comm_num_in_use_flags)) {
cs->sm_comm_num_segments += cs->sm_comm_num_in_use_flags -
(cs->sm_comm_num_segments % cs->sm_comm_num_in_use_flags);
}
cs->sm_segs_per_inuse_flag =
cs->sm_comm_num_segments / cs->sm_comm_num_in_use_flags;
mca_base_param_reg_int(c, "tree_degree",
"Degree of the tree for tree-based operations (must be => 1 and <= min(control_size, 255))",
false, false,
cs->sm_tree_degree,
&cs->sm_tree_degree);
if (cs->sm_tree_degree > cs->sm_control_size) {
orte_show_help("help-mpi-coll-sm.txt",
"tree-degree-larger-than-control", true,
cs->sm_tree_degree, cs->sm_control_size);
cs->sm_tree_degree = cs->sm_control_size;
}
if (cs->sm_tree_degree > 255) {
orte_show_help("help-mpi-coll-sm.txt",
"tree-degree-larger-than-255", true,
cs->sm_tree_degree);
cs->sm_tree_degree = 255;
}
Fixes trac:1988. The little bug that turned out to be huge. Yoinks. * Various cosmetic/style updates in the btl sm * Clean up concept of mpool module (I think that code was written way back when the concept of "modules" was fuzzy) * Bring over some old fixes from the /tmp/timattox-sm-coll/ tree to fix potential segv's when mmap'ed regions were at different addresses in different processes (thanks Tim!). * Change sm coll to no longer use mpool as its main source of shmem; rather, just mmap its own segment (because it's fixed size -- there was nothing to be gained by using mpool; shedding the use of mpool saved a lot of complexity in the sm coll setup). This effectively made Tim's fixes moot (because now everything is an offset into the mmap that is computed locally; there are no global pointers). :-) * Slightly updated common/sm to allow making mmap's for a specific set of procs (vs. ''all'' procs in the process). This potentially allows for same-host-inter-proc mmaps -- yay! * Fixed many, many things in the coll sm (particularly in reduce): * Fixed handling of MPI_IN_PLACE in reduce and allreduce * Fixed handling of non-contiguous datatypes in reduce * Changed the order of reductions to go from process (n-1)'s data to process 0's data, because that's how all other OMPI coll components work * Fixed lots of usage of ddt functions * When using a non-contiguous datatype, if the root process is not (n-1), now we used a 2nd convertor to copy from shmem to the rbuf (saves a memory copy vs. what was done before) * Lots and lots of little cleanups, clarifications, and minor optimizations (although still more could be done -- e.g., I think the use of write memory barriers is fairly sub-optimal; they could be ganged together at the root, for example) I'm marking this as "fixes trac:1988" and closing the ticket; if something is still broken, we can re-open the ticket. This commit was SVN r21967. The following Trac tickets were found above: Ticket 1988 --> https://svn.open-mpi.org/trac/ompi/ticket/1988
2009-09-15 04:25:21 +04:00
/* INFO: Calculate how much space we need in the per-communicator
shmem data segment. This formula taken directly from
coll_sm_module.c. */
mca_base_param_reg_int(c, "info_num_procs",
"Number of processes to use for the calculation of the shared_mem_size MCA information parameter (must be => 2)",
false, false,
cs->sm_info_comm_size,
&cs->sm_info_comm_size);
Fixes trac:1988. The little bug that turned out to be huge. Yoinks. * Various cosmetic/style updates in the btl sm * Clean up concept of mpool module (I think that code was written way back when the concept of "modules" was fuzzy) * Bring over some old fixes from the /tmp/timattox-sm-coll/ tree to fix potential segv's when mmap'ed regions were at different addresses in different processes (thanks Tim!). * Change sm coll to no longer use mpool as its main source of shmem; rather, just mmap its own segment (because it's fixed size -- there was nothing to be gained by using mpool; shedding the use of mpool saved a lot of complexity in the sm coll setup). This effectively made Tim's fixes moot (because now everything is an offset into the mmap that is computed locally; there are no global pointers). :-) * Slightly updated common/sm to allow making mmap's for a specific set of procs (vs. ''all'' procs in the process). This potentially allows for same-host-inter-proc mmaps -- yay! * Fixed many, many things in the coll sm (particularly in reduce): * Fixed handling of MPI_IN_PLACE in reduce and allreduce * Fixed handling of non-contiguous datatypes in reduce * Changed the order of reductions to go from process (n-1)'s data to process 0's data, because that's how all other OMPI coll components work * Fixed lots of usage of ddt functions * When using a non-contiguous datatype, if the root process is not (n-1), now we used a 2nd convertor to copy from shmem to the rbuf (saves a memory copy vs. what was done before) * Lots and lots of little cleanups, clarifications, and minor optimizations (although still more could be done -- e.g., I think the use of write memory barriers is fairly sub-optimal; they could be ganged together at the root, for example) I'm marking this as "fixes trac:1988" and closing the ticket; if something is still broken, we can re-open the ticket. This commit was SVN r21967. The following Trac tickets were found above: Ticket 1988 --> https://svn.open-mpi.org/trac/ompi/ticket/1988
2009-09-15 04:25:21 +04:00
size = 4 * cs->sm_control_size +
(cs->sm_comm_num_in_use_flags * cs->sm_control_size) +
(cs->sm_comm_num_segments * (cs->sm_info_comm_size * cs->sm_control_size * 2)) +
(cs->sm_comm_num_segments * (cs->sm_info_comm_size * cs->sm_fragment_size));
mca_base_param_reg_int(c, "shared_mem_used_data",
Fixes trac:1988. The little bug that turned out to be huge. Yoinks. * Various cosmetic/style updates in the btl sm * Clean up concept of mpool module (I think that code was written way back when the concept of "modules" was fuzzy) * Bring over some old fixes from the /tmp/timattox-sm-coll/ tree to fix potential segv's when mmap'ed regions were at different addresses in different processes (thanks Tim!). * Change sm coll to no longer use mpool as its main source of shmem; rather, just mmap its own segment (because it's fixed size -- there was nothing to be gained by using mpool; shedding the use of mpool saved a lot of complexity in the sm coll setup). This effectively made Tim's fixes moot (because now everything is an offset into the mmap that is computed locally; there are no global pointers). :-) * Slightly updated common/sm to allow making mmap's for a specific set of procs (vs. ''all'' procs in the process). This potentially allows for same-host-inter-proc mmaps -- yay! * Fixed many, many things in the coll sm (particularly in reduce): * Fixed handling of MPI_IN_PLACE in reduce and allreduce * Fixed handling of non-contiguous datatypes in reduce * Changed the order of reductions to go from process (n-1)'s data to process 0's data, because that's how all other OMPI coll components work * Fixed lots of usage of ddt functions * When using a non-contiguous datatype, if the root process is not (n-1), now we used a 2nd convertor to copy from shmem to the rbuf (saves a memory copy vs. what was done before) * Lots and lots of little cleanups, clarifications, and minor optimizations (although still more could be done -- e.g., I think the use of write memory barriers is fairly sub-optimal; they could be ganged together at the root, for example) I'm marking this as "fixes trac:1988" and closing the ticket; if something is still broken, we can re-open the ticket. This commit was SVN r21967. The following Trac tickets were found above: Ticket 1988 --> https://svn.open-mpi.org/trac/ompi/ticket/1988
2009-09-15 04:25:21 +04:00
"Amount of shared memory used, per communicator, in the shared memory data area for info_num_procs processes (in bytes)",
false, true,
Fixes trac:1988. The little bug that turned out to be huge. Yoinks. * Various cosmetic/style updates in the btl sm * Clean up concept of mpool module (I think that code was written way back when the concept of "modules" was fuzzy) * Bring over some old fixes from the /tmp/timattox-sm-coll/ tree to fix potential segv's when mmap'ed regions were at different addresses in different processes (thanks Tim!). * Change sm coll to no longer use mpool as its main source of shmem; rather, just mmap its own segment (because it's fixed size -- there was nothing to be gained by using mpool; shedding the use of mpool saved a lot of complexity in the sm coll setup). This effectively made Tim's fixes moot (because now everything is an offset into the mmap that is computed locally; there are no global pointers). :-) * Slightly updated common/sm to allow making mmap's for a specific set of procs (vs. ''all'' procs in the process). This potentially allows for same-host-inter-proc mmaps -- yay! * Fixed many, many things in the coll sm (particularly in reduce): * Fixed handling of MPI_IN_PLACE in reduce and allreduce * Fixed handling of non-contiguous datatypes in reduce * Changed the order of reductions to go from process (n-1)'s data to process 0's data, because that's how all other OMPI coll components work * Fixed lots of usage of ddt functions * When using a non-contiguous datatype, if the root process is not (n-1), now we used a 2nd convertor to copy from shmem to the rbuf (saves a memory copy vs. what was done before) * Lots and lots of little cleanups, clarifications, and minor optimizations (although still more could be done -- e.g., I think the use of write memory barriers is fairly sub-optimal; they could be ganged together at the root, for example) I'm marking this as "fixes trac:1988" and closing the ticket; if something is still broken, we can re-open the ticket. This commit was SVN r21967. The following Trac tickets were found above: Ticket 1988 --> https://svn.open-mpi.org/trac/ompi/ticket/1988
2009-09-15 04:25:21 +04:00
(int)size, NULL);
/* Common SM MCA params */
mca_common_sm_param_register(c);
return OMPI_SUCCESS;
}