1
1
Граф коммитов

571 Коммитов

Автор SHA1 Сообщение Дата
Pavel Shamis
b89f8fabc9 Adding Hierarchical Collectives project to the Open MPI trunk.
The project includes following components and frameworks: 
- ML Collective component
- NETPATTERNS and COMMPATTERNS common components
- BCOL framework
- SBGP framework

Note: By default the ML collective component is disabled. In order to enable
new collectives user should bump up the priority of ml component (coll_ml_priority)

=============================================

Primary Contributors (in alphabetical order):

Ishai Rabinovich (Mellanox)
Joshua S. Ladd (ORNL / Mellanox)
Manjunath Gorentla Venkata (ORNL)
Mike Dubman (Mellanox)
Noam Bloch (Mellanox)
Pavel (Pasha) Shamis (ORNL / Mellanox)
Richard Graham (ORNL / Mellanox)
Vasily Filipov (Mellanox)

This commit was SVN r27078.
2012-08-16 19:11:35 +00:00
Jeff Squyres
a4e97fb4c0 Ensure we assign "err" properly when invoking MCA_PML_CALLs. Although
technically this is a necessary thing to do, it wasn't a tragedy that
we didn't have it because err was initialize to 0 in the beginning of
the functions where this problem occurred.  Also, OMPI will likely
abort if one of the MCA_PML_CALLs actually incurs an error (or, even
if it doesn't, MPI doesn't define the behavior anyway ;-) ).  

But looking forward to an FT-aware world, fixing this issue is a Good
Thing.  Many thanks to Hristo Iliev for pointing out the issue.

This commit was SVN r27070.
2012-08-16 17:49:48 +00:00
Shiqing Fan
2f442799f8 fix several typecasts
This commit was SVN r26957.
2012-08-07 10:41:53 +00:00
Eugene Loh
10e3dc396b Add a missing return value.
This commit was SVN r26815.
2012-07-20 01:32:06 +00:00
Brian Barrett
2518014037 Fix a number of issues with IN_PLACE
This commit was SVN r26814.
2012-07-19 21:29:43 +00:00
Eugene Loh
a3e02fdaff With non-blocking collectives, a "round schedule" could fall on any address
alignment, which typically causes problems on SPARC.  Further, the pointer
manipulation to access elements in a round schedule was clumsy.  This change
introduces macros to facilitate addressing and make it more portable.

This commit was SVN r26802.
2012-07-18 17:08:24 +00:00
Brian Barrett
58413fa1e4 * properly setup communication infrastructure for libnbc.
* Prevent infinite recursion in progress loop.

Should fix improper barrier eugene was seeing.

This commit was SVN r26758.
2012-07-06 13:59:03 +00:00
Brian Barrett
e0ceabd486 Need to set MPI_ERROR in the status before calling ompi_request_complete.
This commit was SVN r26757.
2012-07-06 01:14:35 +00:00
Brian Barrett
27d45ad550 Implement reduce_scatter_block and ireduce_scatter_block, although possibly
not nearly as optimal as they should be.

This commit was SVN r26756.
2012-07-05 22:11:48 +00:00
George Bosilca
63278df92d Prevent the coll SM from looking for information about remote procs
during the init phase. This information is only available at a 
later stage.

This commit was SVN r26746.
2012-07-04 21:15:40 +00:00
Brian Barrett
d56de80b5d * Properly initialize handle variable as a request (since the coll_libnbc_request contains everything an NBC_Handle used to contain). Not sure how this slipped through...
This commit was SVN r26710.
2012-07-02 16:39:42 +00:00
Brian Barrett
7e67bfa175 Use OMPI's ops instead of the libnbc ops.
This commit was SVN r26708.
2012-07-02 15:47:22 +00:00
Brian Barrett
0b887ab5a1 * Remove unneeded prototype that was causing compile issues anyway
* Use proper tag space (the negatives below the blocking communicators)
  instead of the point-to-point space
* Use the PML interface instead of the MPI interface, since the MPI
  interface 1) shouldn't be used by components and 2) doesn't like
  negative tags

This commit was SVN r26693.
2012-06-28 16:52:03 +00:00
Ralph Castain
a1344bc5c0 Add missing header to tarball
This commit was SVN r26689.
2012-06-28 13:07:18 +00:00
Brian Barrett
32e70b691a Re-enable non-blocking collectives in libnbc after finding issue with the definition of
NBC_CACHE_SCHEDULE not being propogated to all uses.

This commit was SVN r26686.
2012-06-27 22:08:19 +00:00
Brian Barrett
d85fdd2605 temporarily back out r26682 and r26683 until I can figure out why they cause crashes during shutdown
This commit was SVN r26684.

The following SVN revision numbers were found above:
  r26682 --> open-mpi/ompi@15a30af11f
  r26683 --> open-mpi/ompi@f6ea4b7234
2012-06-27 19:32:53 +00:00
Brian Barrett
f6ea4b7234 Remove now unneeded header file
This commit was SVN r26683.
2012-06-27 18:43:40 +00:00
Brian Barrett
15a30af11f Turn on all the non-blocking collectives provided by libnbc...
This commit was SVN r26682.
2012-06-27 18:32:57 +00:00
Brian Barrett
3933d0a8f0 Ibarrier works! :)
This commit was SVN r26680.
2012-06-27 15:58:17 +00:00
Josh Hursey
28681deffa Backout the ORCA commit. :(
There is a linking issue on Mac OSX that needs to be addressed before this is able to come back into the trunk.

This commit was SVN r26676.
2012-06-27 01:28:28 +00:00
Josh Hursey
542330e3a7 Commit of ORCA: Open MPI Runtime Collaborative Abstraction
This is a runtime interposition project that sits between the OMPI and ORTE layers in Open MPI.

The project is described on the wiki:
  https://svn.open-mpi.org/trac/ompi/wiki/Runtime_Interposition

And on this email thread:
  http://www.open-mpi.org/community/lists/devel/2012/06/11109.php

This commit was SVN r26670.
2012-06-26 21:42:16 +00:00
Brian Barrett
7bdeafb772 Start bringing in libnbc. .ompi_ignored, as there's still a long way to go
This commit was SVN r26658.
2012-06-25 22:38:06 +00:00
Brian Barrett
b9e8e4aeb9 * Initial merge of the non-blocking collectives interface. No implementation of
the back-end yet, coming real soon now, need to solve some tag issues first.

This commit was SVN r26641.
2012-06-22 20:54:12 +00:00
Jeff Squyres
5451ee46bd Per r26575, the sync coll module is no longer necessary!
(the crowd goes wild)

This commit was SVN r26583.

The following SVN revision numbers were found above:
  r26575 --> open-mpi/ompi@59e529cf1d
2012-06-08 19:19:19 +00:00
Jeff Squyres
2ba10c37fe Per RFC, bring in the following changes:
* Remove paffinity, maffinity, and carto frameworks -- they've been
   wholly replaced by hwloc.
 * Move ompi_mpi_init() affinity-setting/checking code down to ORTE.
 * Update sm, smcuda, wv, and openib components to no longer use carto.
   Instead, use hwloc data.  There are still optimizations possible in
   the sm/smcuda BTLs (i.e., making multiple mpools).  Also, the old
   carto-based code found out how many NUMA nodes were ''available''
   -- not how many were used ''in this job''.  The new hwloc-using
   code computes the same value -- it was not updated to calculate how
   many NUMA nodes are used ''by this job.''
   * Note that I cannot compile the smcuda and wv BTLs -- I ''think''
     they're right, but they need to be verified by their owners.
 * The openib component now does a bunch of stuff to figure out where
   "near" OpenFabrics devices are.  '''THIS IS A CHANGE IN DEFAULT
   BEHAVIOR!!''' and still needs to be verified by OpenFabrics vendors
   (I do not have a NUMA machine with an OpenFabrics device that is a
   non-uniform distance from multiple different NUMA nodes).
 * Completely rewrite the OMPI_Affinity_str() routine from the
   "affinity" mpiext extension.  This extension now understands
   hyperthreads; the output format of it has changed a bit to reflect
   this new information.
 * Bunches of minor changes around the code base to update names/types
   from maffinity/paffinity-based names to hwloc-based names.
 * Add some helper functions into the hwloc base, mainly having to do
   with the fact that we have the hwloc data reporting ''all''
   topology information, but sometimes you really only want the
   (online | available) data.

This commit was SVN r26391.
2012-05-07 14:52:54 +00:00
George Bosilca
f09e3ce5a4 Spring cleanup. Nothing important.
This commit was SVN r26247.
2012-04-06 15:48:07 +00:00
George Bosilca
654c75ff24 As suggested on the mailing list a while back, switch the default alltoallv
algorithm to pairwise exchange instead of the default one. This might
improve the scheduling and relax the pressure on the network.

This commit was SVN r26246.
2012-04-06 15:47:29 +00:00
Josh Hursey
d1571b027a Fix a few error return paths
This commit was SVN r26233.
2012-04-04 15:11:03 +00:00
Mike Dubman
a45898ea9c fix support for fca 2.2, warning fixes on rhel 6.x
This commit was SVN r26166.
2012-03-20 10:00:52 +00:00
George Bosilca
a78a7bd8e8 The tuned collectives can now deal with more than 2Gb of data.
This commit was SVN r26103.
2012-03-05 22:23:44 +00:00
George Bosilca
762b3e13a9 Use the correct name for the datatype destruction function.
This commit was SVN r26100.
2012-03-05 15:54:53 +00:00
George Bosilca
7d523a8852 Avoid calling the bcast with counts larger than INT_MAX.
This commit was SVN r26098.
2012-03-05 14:30:30 +00:00
George Bosilca
e8c358c188 Allow Open MPI to deal with size_t internally.
This commit was SVN r26097.
2012-03-05 14:10:26 +00:00
George Bosilca
f83670211e Allow the user to define dynamic rules for messages larger than 2GB.
This commit was SVN r26084.
2012-03-02 21:16:23 +00:00
George Bosilca
8791ade293 Help he selection of the right algorithm for large data (> 2Gb). Thanks
to Fujitsu for the patch.

This commit was SVN r26080.
2012-03-02 19:12:22 +00:00
George Bosilca
72f731f25f The SM2 collective component has not been updated in a long
time. Rich, the original developer, agrees with this removal.

This commit was SVN r25368.
2011-10-25 22:07:09 +00:00
Rainer Keller
4e6a6fc146 - Check, whether the compiler supports __builtin_clz (count leading
zeroes);
   if so, use it for bit-operations like opal_cube_dim and opal_hibit.
   Implement two versions of power-of-two.
   In case of opal_next_poweroftwo, this reduces the average execution
   time from 83 cycles to 4 cycles (Intel Nehalem, icc, -O2, inlining,
   measured rdtsc, with loop over 2^27 values).
   Numbers for other functions are similar (but of course heavily depend
   on the usage, e.g. opal_hibit() with a start of 4 does not save
   much).  The bsr instruction on AMD Opteron is also not as fast.

 - Replace various places where the next power-of-two is computed.
   
   Tested on Intel Nehalem Cluster with openib, compilers GNU-4.6.1 and
   Intel-12.0.4 using mpi_testsuite -t "Collective" with 128 processes.

This commit was SVN r25270.
2011-10-11 22:49:01 +00:00
George Bosilca
2fefd3a928 Don't forget to move the pointer back by the true_lb.
This commit was SVN r25262.
2011-10-11 20:15:49 +00:00
George Bosilca
ce7935c8fa Obviously these were not needed.
This commit was SVN r25231.
2011-10-04 14:56:34 +00:00
George Bosilca
80c02647c8 Each level (OPAL/ORTE/OMPI) should only return it's own constants,
instead of the current mismatch.

This commit was SVN r25230.
2011-10-04 14:50:31 +00:00
Rolf vandeVaart
0749a220e8 Add support for MPI_IN_PLACE to MPI_Exscan. Required for MPI 2.2 compliance.
Reviewed by Jeff Squyres.  This fixes trac:2221.

This commit was SVN r25165.

The following Trac tickets were found above:
  Ticket 2221 --> https://svn.open-mpi.org/trac/ompi/ticket/2221
2011-09-20 14:54:41 +00:00
George Bosilca
9687e7f38e This commit fixes trac:2679 and should be added to cmr:v1.4:reviewer=jsquyres
and cmr:v1.5:reviewer=jsquyres

This commit was SVN r25155.

The following Trac tickets were found above:
  Ticket 2679 --> https://svn.open-mpi.org/trac/ompi/ticket/2679
2011-09-18 00:58:26 +00:00
Wesley Bland
4e7ff0bd5e By popular demand the epoch code is now disabled by default.
To enable the epochs and the resilient orte code, use the configure flag:

--enable-resilient-orte

This will define both:

ORTE_ENABLE_EPOCH
ORTE_RESIL_ORTE

This commit was SVN r25093.
2011-08-26 22:16:14 +00:00
Mike Dubman
96ef2fc0e4 fix handling datatypes which have a gap in the beginning
This commit was SVN r24936.
2011-07-25 06:30:09 +00:00
Jeff Squyres
b2b781e537 Fix a few miscelaneous memory leaks.
This commit was SVN r24865.
2011-07-08 16:39:58 +00:00
Wesley Bland
e1ba09ad51 Add a resilience to ORTE. Allows the runtime to continue after a process (or
ORTED) failure. Note that more work will be necessary to allow the MPI layer to
take advantage of this.

Per RFC:
http://www.open-mpi.org/community/lists/devel/2011/06/9299.php

This commit was SVN r24815.
2011-06-23 20:38:02 +00:00
Samuel Gutierrez
81f38b258a commit of new shared memory backing facility framework (shmem) and its components.
This commit was SVN r24795.
2011-06-21 15:41:57 +00:00
George Bosilca
65661a3cb4 Dont use a temporary string.
This commit was SVN r24786.
2011-06-20 09:29:19 +00:00
Mike Dubman
36db9c6233 * updated copyrights
* added support for non-contig data layout in FCA

This commit was SVN r24702.
2011-05-16 14:43:11 +00:00
Jeff Squyres
ec90a3ba6d Fix a few memory leaks, and ensure that coll sm is also registering
the common SM MCA params.

This commit was SVN r24497.
2011-03-08 17:36:59 +00:00