2015-04-18 18:36:05 +03:00
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
2004-08-21 04:49:07 +04:00
/*
2007-03-17 02:11:45 +03:00
* Copyright ( c ) 2004 - 2007 The Trustees of Indiana University and Indiana
2005-11-05 22:57:48 +03:00
* 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 .
2004-11-28 23:09:25 +03:00
* Copyright ( c ) 2004 - 2005 High Performance Computing Center Stuttgart ,
* University of Stuttgart . All rights reserved .
2005-03-24 15:43:37 +03:00
* Copyright ( c ) 2004 - 2005 The Regents of the University of California .
* All rights reserved .
2009-02-13 06:45:32 +03:00
* Copyright ( c ) 2008 - 2009 Cisco Systems , Inc . All rights reserved .
2015-04-18 18:36:05 +03:00
* Copyright ( c ) 2011 - 2015 Los Alamos National Security , LLC .
2011-06-21 19:41:57 +04:00
* All rights reserved .
2004-11-22 04:38:40 +03:00
* $ COPYRIGHT $
*
* Additional copyrights may follow
*
2004-08-21 04:49:07 +04:00
* $ HEADER $
2005-09-03 15:49:46 +04:00
*/
/**
* @ file
2004-08-21 04:49:07 +04:00
*
2005-09-03 15:49:46 +04:00
* Most of the description of the data layout is in the
* coll_sm_module . c file .
2004-08-21 04:49:07 +04:00
*/
# include "ompi_config.h"
2013-02-13 01:10:11 +04:00
# include "opal/util/show_help.h"
2006-02-12 04:33:29 +03:00
# include "ompi/constants.h"
2005-08-24 01:22:00 +04:00
# include "ompi/mca/coll/coll.h"
2004-08-21 04:49:07 +04:00
# include "coll_sm.h"
2005-01-30 04:42:57 +03:00
2004-08-21 04:49:07 +04:00
/*
* Public string showing the coll ompi_sm component version number
*/
const char * mca_coll_sm_component_version_string =
Major simplifications to component versioning:
- After long discussions and ruminations on how we run components in
LAM/MPI, made the decision that, by default, all components included
in Open MPI will use the version number of their parent project
(i.e., OMPI or ORTE). They are certaint free to use a different
number, but this simplification makes the common cases easy:
- components are only released when the parent project is released
- it is easy (trivial?) to distinguish which version component goes
with with version of the parent project
- removed all autogen/configure code for templating the version .h
file in components
- made all ORTE components use ORTE_*_VERSION for version numbers
- made all OMPI components use OMPI_*_VERSION for version numbers
- removed all VERSION files from components
- configure now displays OPAL, ORTE, and OMPI version numbers
- ditto for ompi_info
- right now, faking it -- OPAL and ORTE and OMPI will always have the
same version number (i.e., they all come from the same top-level
VERSION file). But this paves the way for the Great Configure
Reorganization, where, among other things, each project will have
its own version number.
So all in all, we went from a boatload of version numbers to
[effectively] three. That's pretty good. :-)
This commit was SVN r6344.
2005-07-05 00:12:36 +04:00
" Open MPI sm collective MCA component version " OMPI_VERSION ;
2005-01-30 04:42:57 +03:00
/*
* Local functions
*/
2011-03-08 20:36:59 +03:00
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 ) ;
2005-01-30 04:42:57 +03:00
2013-03-28 01:09:41 +04:00
static int coll_sm_shared_mem_used_data ;
2004-08-21 04:49:07 +04:00
/*
* Instantiate the public struct with all of our public information
* and pointers to our public functions in it
*/
2005-07-16 00:01:35 +04:00
mca_coll_sm_component_t mca_coll_sm_component = {
2004-08-21 04:49:07 +04:00
2008-07-29 02:40:57 +04:00
/* First, fill in the super */
2004-08-21 04:49:07 +04:00
2005-01-30 04:42:57 +03:00
{
2005-07-16 00:01:35 +04:00
/* First, the mca_component_t struct containing meta
information about the component itself */
2015-04-18 18:36:05 +03:00
. collm_version = {
2008-07-29 02:40:57 +04:00
MCA_COLL_BASE_VERSION_2_0_0 ,
2005-07-16 00:01:35 +04:00
/* Component name and version */
2015-04-18 18:36:05 +03:00
. mca_component_name = " sm " ,
MCA_BASE_MAKE_VERSION ( component , OMPI_MAJOR_VERSION , OMPI_MINOR_VERSION ,
OMPI_RELEASE_VERSION ) ,
2005-07-16 00:01:35 +04:00
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 */
2015-04-18 18:36:05 +03:00
. mca_close_component = sm_close ,
. mca_query_component = sm_register ,
2005-07-16 00:01:35 +04:00
} ,
2015-04-18 18:36:05 +03:00
. collm_data = {
2007-03-17 02:11:45 +03:00
/* The component is not checkpoint ready */
MCA_BASE_METADATA_PARAM_NONE
2005-07-16 00:01:35 +04:00
} ,
/* Initialization / querying functions */
2015-04-18 18:36:05 +03:00
. collm_init_query = mca_coll_sm_init_query ,
. collm_comm_query = mca_coll_sm_comm_query ,
2005-01-30 04:42:57 +03:00
} ,
2004-08-21 04:49:07 +04:00
2005-07-16 00:01:35 +04:00
/* sm-component specifc information */
2004-08-21 04:49:07 +04:00
2005-08-24 01:22:00 +04:00
/* (default) priority */
2005-09-29 22:05:47 +04:00
/* JMS temporarily lowered until we can get more testing */
0 ,
2004-08-21 04:49:07 +04:00
2005-09-03 15:49:46 +04:00
/* (default) control size (bytes) */
4096 ,
2005-08-24 01:22:00 +04:00
2005-09-03 15:49:46 +04:00
/* (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 */
2005-09-03 15:49:46 +04:00
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 */
2005-09-03 15:49:46 +04:00
8 ,
2005-08-24 01:22:00 +04:00
/* (default) fragment size */
8192 ,
/* (default) degree of tree for tree-based operations (must be <=
control unit size ) */
4 ,
2005-09-03 15:49:46 +04:00
/* (default) number of processes in coll_sm_shared_mem_size
information variable */
4 ,
2005-08-24 01:22:00 +04:00
/* 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 */
2004-08-21 04:49:07 +04:00
} ;
2005-01-30 04:42:57 +03:00
2011-03-08 20:36:59 +03:00
/*
* Shut down the component
*/
static int sm_close ( void )
{
return OMPI_SUCCESS ;
}
2013-03-28 01:09:41 +04:00
static int sm_verify_mca_variables ( void )
2005-01-30 04:42:57 +03:00
{
2005-08-24 01:22:00 +04:00
mca_coll_sm_component_t * cs = & mca_coll_sm_component ;
2005-07-16 00:01:35 +04:00
2005-08-24 01:22:00 +04:00
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 ) ;
}
2005-09-03 15:49:46 +04:00
if ( cs - > sm_comm_num_in_use_flags < 2 ) {
cs - > sm_comm_num_in_use_flags = 2 ;
}
2005-09-07 01:41:55 +04:00
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 ;
2005-09-03 15:49:46 +04:00
}
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 ) ;
2005-09-02 16:57:47 +04:00
}
2005-10-01 03:12:23 +04:00
cs - > sm_segs_per_inuse_flag =
cs - > sm_comm_num_segments / cs - > sm_comm_num_in_use_flags ;
2005-07-16 00:01:35 +04:00
2005-08-24 01:22:00 +04:00
if ( cs - > sm_tree_degree > cs - > sm_control_size ) {
2013-02-13 01:10:11 +04:00
opal_show_help ( " help-mpi-coll-sm.txt " ,
2005-08-24 01:22:00 +04:00
" 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 ) {
2013-02-13 01:10:11 +04:00
opal_show_help ( " help-mpi-coll-sm.txt " ,
2005-08-24 01:22:00 +04:00
" tree-degree-larger-than-255 " , true ,
cs - > sm_tree_degree ) ;
cs - > sm_tree_degree = 255 ;
}
2013-03-28 01:09:41 +04:00
coll_sm_shared_mem_used_data = ( int ) ( 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 ) ) ) ;
return OMPI_SUCCESS ;
}
/*
* Register MCA params
*/
static int sm_register ( void )
{
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 */
cs - > sm_priority = 0 ;
( void ) mca_base_component_var_register ( c , " priority " , " Priority of the sm coll component " ,
MCA_BASE_VAR_TYPE_INT , NULL , 0 , 0 ,
OPAL_INFO_LVL_9 ,
MCA_BASE_VAR_SCOPE_READONLY ,
& cs - > sm_priority ) ;
cs - > sm_control_size = 4096 ;
( void ) mca_base_component_var_register ( 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) " ,
MCA_BASE_VAR_TYPE_INT , NULL , 0 , 0 ,
OPAL_INFO_LVL_9 ,
MCA_BASE_VAR_SCOPE_READONLY ,
& cs - > sm_control_size ) ;
cs - > sm_fragment_size = 8192 ;
( void ) mca_base_component_var_register ( c , " fragment_size " ,
" Fragment size (in bytes) used for passing data through shared memory (will be rounded up to the nearest control_size size) " ,
MCA_BASE_VAR_TYPE_INT , NULL , 0 , 0 ,
OPAL_INFO_LVL_9 ,
MCA_BASE_VAR_SCOPE_READONLY ,
& cs - > sm_fragment_size ) ;
cs - > sm_comm_num_in_use_flags = 2 ;
( void ) mca_base_component_var_register ( 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) " ,
MCA_BASE_VAR_TYPE_INT , NULL , 0 , 0 ,
OPAL_INFO_LVL_9 ,
MCA_BASE_VAR_SCOPE_READONLY ,
& cs - > sm_comm_num_in_use_flags ) ;
cs - > sm_comm_num_segments = 8 ;
( void ) mca_base_component_var_register ( 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) " ,
MCA_BASE_VAR_TYPE_INT , NULL , 0 , 0 ,
OPAL_INFO_LVL_9 ,
MCA_BASE_VAR_SCOPE_READONLY ,
& cs - > sm_comm_num_segments ) ;
cs - > sm_tree_degree = 4 ;
( void ) mca_base_component_var_register ( c , " tree_degree " ,
" Degree of the tree for tree-based operations (must be => 1 and <= min(control_size, 255)) " ,
MCA_BASE_VAR_TYPE_INT , NULL , 0 , 0 ,
OPAL_INFO_LVL_9 ,
MCA_BASE_VAR_SCOPE_READONLY ,
& cs - > sm_tree_degree ) ;
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 . */
2013-03-28 01:09:41 +04:00
cs - > sm_info_comm_size = 4 ;
( void ) mca_base_component_var_register ( c , " info_num_procs " ,
" Number of processes to use for the calculation of the shared_mem_size MCA information parameter (must be => 2) " ,
MCA_BASE_VAR_TYPE_INT , NULL , 0 , 0 ,
OPAL_INFO_LVL_9 ,
MCA_BASE_VAR_SCOPE_READONLY ,
& cs - > sm_info_comm_size ) ;
coll_sm_shared_mem_used_data = ( int ) ( 4 * cs - > sm_control_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
( 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 ) ) +
2013-03-28 01:09:41 +04:00
( cs - > sm_comm_num_segments * ( cs - > sm_info_comm_size * cs - > sm_fragment_size ) ) ) ;
2005-08-24 01:22:00 +04:00
2013-03-28 01:09:41 +04:00
( void ) mca_base_component_var_register ( c , " shared_mem_used_data " ,
" Amount of shared memory used, per communicator, in the shared memory data area for info_num_procs processes (in bytes) " ,
MCA_BASE_VAR_TYPE_INT , NULL , 0 ,
MCA_BASE_VAR_FLAG_DEFAULT_ONLY ,
OPAL_INFO_LVL_9 ,
MCA_BASE_VAR_SCOPE_READONLY ,
& coll_sm_shared_mem_used_data ) ;
return sm_verify_mca_variables ( ) ;
2005-01-30 04:42:57 +03:00
}