1
1
Граф коммитов

4768 Коммитов

Автор SHA1 Сообщение Дата
Avneesh Pant
8bdd334d95 Allow the PSM component to return ERR_NOT_AVAIL so it can be unloaded silently if executed on a node with no QLogic IB hardware. Also minor modifications to have the CM PML allow itself to be unloaded if no MTL components are available. The component selection logic can then continue to use other PMLs.
This commit was SVN r22410.
2010-01-14 19:39:35 +00:00
Shiqing Fan
872a4047ba Fix the bug that caused by ADD_DEPENDENCIES() from different version of CMake.
In CMake 2.6 and earlier, this function add dependencies for targets and also link the target libraries automatically, but in CMake 2.8,this behavior has been changed, i.e. it will only add the dependencies but no link, which will cause linking errors at compilation time.

This commit was SVN r22405.
2010-01-14 18:10:20 +00:00
Vasily Filipov
370b1c75c4 Added addition condition for create_srq
This commit was SVN r22403.
2010-01-14 16:09:10 +00:00
Jeff Squyres
b46628bf8d Reformat for 80-char width.
This commit was SVN r22402.
2010-01-14 13:31:11 +00:00
Avneesh Pant
774b965784 Add in support to specify IB path record query mechanism and IB Application/Service ID for PSM MTL. Also fix a minor bug in calculating the minimum connection timeout.
This commit was SVN r22397.
2010-01-13 18:58:00 +00:00
Jeff Squyres
2bdcb2a979 Move CM's MCA params into their own function (component.register).
This commit was SVN r22392.
2010-01-12 20:11:47 +00:00
Rainer Keller
3adde52707 - Sanity check initialization and finalization of PERUSE.
This commit was SVN r22390.
2010-01-12 16:36:24 +00:00
Jeff Squyres
40a8846b75 Disable the --enable-openib-ibcm switch. IBCM support is broken, so there's no point in even offering this config switch; it just confuses people. Closes trac:2158.
This commit was SVN r22383.

The following Trac tickets were found above:
  Ticket 2158 --> https://svn.open-mpi.org/trac/ompi/ticket/2158
2010-01-07 18:21:41 +00:00
Jeff Squyres
8f824177cd Closes trac:2158.
Special-case the before MPI_INIT / after MPI_FINALIZE error messages
so that they can be a bit more clear than the general "an error
occurred" messages that are displayed in the middle of MPI jobs.

This is not really a "bug fix", but it is helpful for usability.  I
leave it up to the v1.4 RM's to decide if they want it for the 1.4
series or not.

This commit was SVN r22382.

The following Trac tickets were found above:
  Ticket 2158 --> https://svn.open-mpi.org/trac/ompi/ticket/2158
2010-01-07 18:16:39 +00:00
Jeff Squyres
e96032dec9 Fix a type (otherwise we get a compiler warning).
This commit was SVN r22380.
2010-01-07 17:39:18 +00:00
Edgar Gabriel
5c6384e771 clean up the comm_cid code by removing everything related to the block_cid
algorithm. This makes it much easier to read again.

This commit was SVN r22379.
2010-01-07 16:26:30 +00:00
Ethan Mallove
f1f285c575 If CR is disabled, do not create ompi-checkpoint/restart symlinks and CR man pages.
This commit was SVN r22373.
2010-01-06 16:56:54 +00:00
Ralph Castain
fad1ba15b0 Move the test for case-sensitive file system from ompi to opal so that all layers can have that knowledge.
Use that for the orte wrapper compilers

This commit was SVN r22348.
2009-12-29 23:26:45 +00:00
Shiqing Fan
c37308b8eb Remove the deleted windows file from the tarball.
This commit was SVN r22347.
2009-12-29 16:11:32 +00:00
Shiqing Fan
b8555448b5 Remove the unnecessary/duplicated unistd.h.
This commit was SVN r22346.
2009-12-28 16:22:16 +00:00
Shiqing Fan
d0f85beaf3 Correctly include those header files.
This commit was SVN r22344.
2009-12-28 16:13:06 +00:00
Shiqing Fan
90e3092ce5 Fix a type cast.
This commit was SVN r22343.
2009-12-28 16:12:46 +00:00
Shiqing Fan
a2d00d4ab8 Exclude a pml component that is not necessary for Windows.
This commit was SVN r22342.
2009-12-28 16:12:28 +00:00
George Bosilca
e127b20038 Correct a type in the name of the help string.
This commit was SVN r22336.
2009-12-21 19:13:25 +00:00
Vasily Filipov
897b7c0aa8 Fix orte_show_help message type error.
This commit was SVN r22321.
2009-12-16 14:11:43 +00:00
Vasily Filipov
e73274f9a9 Disabling SRQ limit event for devices that doesn't support this feature.
This commit was SVN r22320.
2009-12-16 14:05:35 +00:00
Vasily Filipov
87e71b26fe Jeff Squyres fixes
This commit was SVN r22319.
2009-12-16 10:23:58 +00:00
George Bosilca
b3d3a8e7b3 Remove useless lines.
This commit was SVN r22316.
2009-12-15 23:55:14 +00:00
George Bosilca
b85c3ca081 Enable support for the INRIA
knem (http://runtime.bordeaux.inria.fr/knem/) kernel device. This
is part of Ma Teng's work on Open MPI.

This commit was SVN r22315.
2009-12-15 23:34:09 +00:00
Vasily Filipov
c036c6ef95 Adding support for on-demand SRQ pre-post (receive wqe allocation)
This commit was SVN r22313.
2009-12-15 15:52:10 +00:00
Vasily Filipov
354bfe527f Improving support for non homogeneous OpenFabrics network configurations
This commit was SVN r22312.
2009-12-15 14:25:07 +00:00
Pavel Shamis
4d02aea54c Enabling, by default, RDMACM connection manager for RDMAoE devices
This commit was SVN r22311.
2009-12-15 13:52:19 +00:00
Jeff Squyres
4f68dfb03c Remove some dead code (thanks to George for pointing it out).
This commit was SVN r22309.
2009-12-14 21:20:41 +00:00
Christopher Yeoh
27cc40e412 Fixes MPI errhandler set races
See #2103 for details

This commit was SVN r22300.
2009-12-14 03:38:01 +00:00
Christopher Yeoh
848bf0f5cd Fixes deadlock in osc rdma module
See #2102 for details

This commit was SVN r22299.
2009-12-14 01:52:57 +00:00
Christopher Yeoh
d5253aa0f1 Fixes multithread race which causes corruption of no_credits_pending_frags
list in the ib btl. See #2128 for details 

This commit was SVN r22298.
2009-12-14 01:41:45 +00:00
Jeff Squyres
ba886bcecd Strange how this wasn't needed on my mac -- but it's definitely needed
on linux!

This commit was SVN r22296.
2009-12-11 21:23:43 +00:00
Jeff Squyres
528a22d1eb Add a simple test inspired by Ashley Pittman to double check that the
msgq DLL is at least lt_dlopen-able.

This commit was SVN r22295.
2009-12-11 21:09:16 +00:00
Eugene Loh
8177d91835 Minor change so that if the number of shared-memory FIFOs is greater
than can be used (e.g., number of on-node peers), that no additional
room is set aside for those FIFOs that will never be created.  This
makes it easier to have dedicated FIFOs:  just set btl_sm_num_fifos
to be very large rather than setting it to be the local number of
procs.  In practice, we ask for extra headroom anyhow, so this change
generally won't matter.

This commit was SVN r22291.
2009-12-10 19:28:39 +00:00
Rainer Keller
499834bc6e - As Sylvain Jeaugey noted optional Fortran ddt ids are not properly taken
care of, see:
   http://www.open-mpi.org/community/lists/devel/2009/12/7193.php

   Assign them to proper ids:
   1. Proper Fortran type, if size matches, otherwise
   2. assign id of size-matching C-type.

   Refs trac:2133

   As stated in CMR #2133, this should move to v1.5, but '''not''' to v1.4.

This commit was SVN r22287.

The following Trac tickets were found above:
  Ticket 2133 --> https://svn.open-mpi.org/trac/ompi/ticket/2133
2009-12-08 22:26:04 +00:00
George Bosilca
76222eb869 Get rid of the useless mca_pml_base_endpoint_t and replace it by
[the well known and widely used!] mca_pml_endpoint_t.

This commit was SVN r22277.
2009-12-08 17:29:54 +00:00
Pavel Shamis
b024aee10c Removing unused lists from mca_btl_openib_qp_info_t. The lists were moved to device.
This commit was SVN r22271.
2009-12-07 17:42:09 +00:00
Matthias Jurenz
7a277b345d - repaired corrupt tools/opari/doc/lacsi01.pdf
- removed tools/opari/doc/lacsi01.ps.gz which is equivalent to tools/opari/doc/lacsi01.pdf
- corrected svn:mime-type of tools/opari/doc/opari-logo-100.gif

This commit was SVN r22267.
2009-12-04 15:08:40 +00:00
Jeff Squyres
16b100219d A patch from UTK to allow orte_init(), opal_init(), and associated
friends also receive &argc and &argv (George asked Jeff to Ralph to
review before committing).  The thought is that passing argv and argc
to opal/orte_init be useful to other projects outside of OMPI that are
using OPAL and/or ORTE (especially in conjunction with some other
bootstrapping code where it is helpful to modify argv).  It's such a
small thing that it's easy to apply here to make others' lives a
little easier.

Ask George for more details; I'm just the messenger.  :-)

Judging by the copyrights on this patch, it's been around for a
while.  :-)

This commit was SVN r22260.
2009-12-04 00:51:15 +00:00
George Bosilca
f0303a8b25 Indentation.
This commit was SVN r22254.
2009-12-02 22:03:52 +00:00
Ralph Castain
7c11f9d708 Remove last vestiges of fddp and sensor frameworks
This commit was SVN r22252.
2009-12-02 16:03:06 +00:00
Jeff Squyres
12520ca711 Just like we relaxed the error checking for MPI_CART_CREATE (r21816),
we should have also relaxed the error checking for MPI_GRAPH_CREATE.
Thanks to David Singleton for pointing this out.

This commit was SVN r22251.

The following SVN revision numbers were found above:
  r21816 --> open-mpi/ompi@b8332ea2b2
2009-12-01 21:50:39 +00:00
Pavel Shamis
7d46985096 Removing unneeded spaces
This commit was SVN r22246.
2009-12-01 11:15:40 +00:00
Pavel Shamis
75a48f4b3c Bugfix for possible race in rdmacm_destroy_dummy_qp
This commit was SVN r22245.
2009-12-01 08:09:43 +00:00
Shiqing Fan
7cf427c39b Include the missing thread header, which is needed when build with --enable-progress-thread.
This commit was SVN r22239.
2009-11-27 14:49:24 +00:00
Matthias Jurenz
f1c55df65c - create a process group with the real node name on CrayXT platforms
- updated version number of integrated VT to 5.4.11

This commit was SVN r22238.
2009-11-26 07:50:07 +00:00
Brian Barrett
b57b8c5b3f Clean up request handling in the I/O framework to be more consistent with
other request-using frameworks.

 - Rather than having mpi/c/* functions allocate requests explicitly,
   pass the MPI_Request* down to the I/O component and have it 
   perform the allocation.
 - While the I/O base provides a base request which can be used,
   it is not required and all request management occurs within
   the component.
 - Push progress management into the component, rather than having it
   happen in the base.  Progress functions are now easily registered,
   and not all (ie, the one existing) components use progress functions
   in any rational way.

ROMIO switched to generalized requests instead of MPIO_Requests many
moons ago, and Open MPI now uses ROMIO's generalized requests, so there
is no reason to wrap those requests (which are OMPI requests) in another
level of request.

Now the file function passes the MPI_Request* to the ROMIO component,
which passes it to the underlying ROMIO function, which calls 
MPI_Grequest_start to create an OMPI request, which is what gets set
as the request to the user.  Much cleaner.

This patch has two motivations.  One, a whole heck of a lot of code
just got removed, and request handling is now much cleaner for I/O
components.  Two, by adding support for Argonne's proposed generalized
request extensions, we can allow ROMIO to provide async I/O through
generalized requests, which we couldn't rationally do in the old
setup due to the crazy request completion rules.

This commit was SVN r22235.
2009-11-26 05:13:43 +00:00
Brian Barrett
8075640ef1 The tests are MPI programs and are built using mpicc, so including
OMPI headers won't work

This commit was SVN r22233.
2009-11-25 18:06:15 +00:00
George Bosilca
87fd85b17a Detach the user buffer prior to the orte_barrier in MPI_Finalize.
This patch fixes trac:2112.

This commit was SVN r22229.

The following Trac tickets were found above:
  Ticket 2112 --> https://svn.open-mpi.org/trac/ompi/ticket/2112
2009-11-24 02:33:13 +00:00
Jeff Squyres
766d56dc0a Minor typo fixes from Jed Brown.
This commit was SVN r22217.
2009-11-12 15:00:08 +00:00