1
1
Граф коммитов

4631 Коммитов

Автор SHA1 Сообщение Дата
Jeff Squyres
11d44cec1b Fix MPI_COMM_SPAWN[_MULTIPLE] to only check the info handles for
errors on the root.  Thanks to Federico Golfre Andreasi for reporting
the problem.

This commit was SVN r21838.
2009-08-19 13:24:12 +00:00
Rainer Keller
567e5c4342 - As described in RFC,
http://www.open-mpi.org/community/lists/devel/2009/08/6618.php
   lower the default priority of PML/cm to allow _defined_ behaviour
   for systems, where both MTLs and BTLs are available (Portals and MX).

   Keep the previous behaviour of favoring in case of PSM.
   Still, the user may select --mca pml cm for apps where applicable.

This commit was SVN r21834.
2009-08-18 19:12:43 +00:00
George Bosilca
23e8ce91ba Rework the selection logic for the tuned collectives. All supported collectives
now are able to use the dynamic rules. Moreover, these rules are loaded only once,
and stored at the component level. All communicators are able to use these rules
(not only MPI_COMM_WORLD as until now).
A lot of minor corrections, memory management issues and reduction in the amount
of memory used by the tuned collectives.

This commit was SVN r21825.
2009-08-14 21:06:23 +00:00
Jeff Squyres
c3afac1d50 Fix comment typo
This commit was SVN r21824.
2009-08-14 12:09:19 +00:00
Jeff Squyres
b8332ea2b2 Patch from Kiril to make the parameter checking on MPI_CART_CREATE a
bit more relaxed.

This commit was SVN r21816.
2009-08-13 22:06:38 +00:00
Ralph Castain
ded58ae483 Silence some compiler warnings about print statements
This commit was SVN r21814.
2009-08-13 13:45:38 +00:00
Rainer Keller
02a39a208d - Patch r18658 introduced NUMA awareness and memory affinity for
BTL/sm. This static variable needlessly ends up in the so.-file.
   init_maffinity is called once from sm_btl_first_time_init.

   Checked with lennyve, static here is not necessary.

This commit was SVN r21813.

The following SVN revision numbers were found above:
  r18658 --> open-mpi/ompi@f4811d6c4d
2009-08-13 13:08:39 +00:00
Avneesh Pant
261d34db3a Endpoint options port and outsl only appear post version 0x0107 so conditionally compile them in.
This commit was SVN r21812.
2009-08-12 19:59:15 +00:00
Ralph Castain
0c73aa6a97 Fix a couple of errors that are preventing this module from building in MTT.
NOTE: there are still two errors that I cannot fix - will send those to devel list

This commit was SVN r21809.
2009-08-12 13:18:04 +00:00
Shiqing Fan
bce2f44154 Update related .windows files with proper compiling properties, in order to have a successful DSO build.
This commit was SVN r21805.
2009-08-12 08:55:58 +00:00
Shiqing Fan
fb4be6fad7 First step to enable DSO build on Windows.
- add a cmake module for searching libltdl libraries and headers
  - a configure option to enable DSO build, default OFF.
  - update a few source files for including correct header
    and loading correct mca libraries path/suffix.

This commit was SVN r21804.
2009-08-12 08:52:48 +00:00
George Bosilca
cb6c9268ca Remove all unneeded defines.
This commit was SVN r21802.
2009-08-11 19:10:04 +00:00
George Bosilca
79f3f847ca Use the correct value for detecting packed data.
This commit was SVN r21801.
2009-08-11 19:09:45 +00:00
George Bosilca
52c722352d The COMPLEX are back. Due to some compilers flags right now the
support for _Complex is disabled until we figure out the correct
black magic. So instead of using this nice C99 feature, we use the
a strcture with a double type, the same approach that worked pretty
well for the last couple of years. 

Switching from one mode to the other is done using the 
OPAL_USE_[FLOAT|DOUBLE|LONG_DOUBLE]__COMPLEX macros defined in
opal_datatype_internal.h at line 442.

This commit was SVN r21800.
2009-08-11 18:44:06 +00:00
Pavel Shamis
31a88b149a Fixing thread deadlock flow in openib btl (mpi-thread enabled mode)
This commit was SVN r21793.
2009-08-11 10:43:52 +00:00
George Bosilca
51b2cfe40d This header is required to compile the FT.
This commit was SVN r21792.
2009-08-11 05:21:27 +00:00
Rainer Keller
76469ea64a - Change the property of a few files, that obviously
don't need to be svn:executable...

This commit was SVN r21786.
2009-08-11 01:40:00 +00:00
Rainer Keller
6050020c54 - Use OMPI_SUCCESS.
Fails to compile in environments with --disable-mpi

This commit was SVN r21785.
2009-08-10 17:46:25 +00:00
George Bosilca
9c2b993589 Complete r21778 by adding the missing headers.
This commit was SVN r21784.

The following SVN revision numbers were found above:
  r21778 --> open-mpi/ompi@e4d52b16b5
2009-08-10 17:07:43 +00:00
Rainer Keller
23e6ec1ebf - Update manpages of depracted functions.
Fix headers, move too long subjects into main text.
   Add See also sections.

This commit was SVN r21779.
2009-08-10 13:44:53 +00:00
Terry Dontje
e4d52b16b5 Add in eager limit checks in pmls.
This commit was SVN r21778.
2009-08-10 12:46:20 +00:00
Donald Kerr
de6a7f57b0 fix #1984; only decrement send request req_state when not equal to zero
This commit was SVN r21775.
2009-08-07 14:58:50 +00:00
Steve Wise
ed39853f41 add new device ids for the Chelsio T3 RNIC
This commit was SVN r21774.
2009-08-07 14:14:08 +00:00
Nysal Jan
cd9d1e6d1d Remove redundant code
This commit was SVN r21773.
2009-08-07 07:51:19 +00:00
Rolf vandeVaart
c82e468ede Undo revision r21767 - sorry folks
This commit was SVN r21769.

The following SVN revision numbers were found above:
  r21767 --> open-mpi/ompi@41f38110ff
2009-08-05 22:23:26 +00:00
Rolf vandeVaart
41f38110ff HCA failover support in openib BTL
This commit was SVN r21767.
2009-08-05 21:53:02 +00:00
George Bosilca
cf8bd2142a Various cleanups and typos.
This commit was SVN r21765.
2009-08-05 03:12:33 +00:00
Rainer Keller
1bd94f2d98 - When calling ompi_mtl_portals_finalize, when then pml/ob1 is used
(aka w/o  --mca pml cm), make sure PtlEQGet will actually work
   on ompi_mtl_portals.ptl_eq_h -- do so without adding code to 
   ompi_mtl_portals_progress.

   Otherwise we abort() with
[nid09979:32503] ompi_mtl_portals_finalize: Going to call ompi_mtl_portals_progress
[nid09979:32503]  Error returned from PtlEQGet.  Error code - 14
[nid09979:32502] Signal: Aborted (6)
[nid09979:32502] Signal code:  (-6) 

This commit was SVN r21761.
2009-08-04 22:48:07 +00:00
George Bosilca
98bdf5d17b Remove a compiler warning about missing braces around
initializer.

This commit was SVN r21760.
2009-08-04 21:41:14 +00:00
Jeff Squyres
846e6e4bad Fix borkedness from r21755:
* No need for OPAL_SIZEOF_BOOL and OPAL_SIZEOF_INT in comm_inln.h --
   just use sizeof()
 * Fix logic in ompi_setup_cxx.m4 to account for the case where we
   ''do'' have a C++ compiler (duh!!)
 * Fix spelling error in a shell variable that ended up making a
   bad/empty #define

This should bring the trunk back to being functional.  Sorry for the
interruption, folks...

This commit was SVN r21758.

The following SVN revision numbers were found above:
  r21755 --> open-mpi/ompi@90d6491737
2009-08-04 18:45:11 +00:00
Rainer Keller
5a4c6434e7 - Single request, so use single wait
Technically we should have used MPI_STATUSES_IGNORE, anyhow.
   (however both MPI_STATUS*_IGNORE are ((MPI_Status*)0) ;-)

This commit was SVN r21757.
2009-08-04 14:51:15 +00:00
Rainer Keller
2989b65e68 - Initialize the MPI_COMPLEX, to enable the pre-split behaviour.
This however may not be what we finally want to support MPI_COMPLEX:
   As with a possible difference of C99 _Bool and C++ Bool, we may want
   to have a base opal_datatype_complex (as was initially in the
   ompi-ddt branch), instead of assuming a struct of
   opal_datatype_float... Think of stricter alignment.

   The configure checks for (float/double/long double) complex have
   been added with the ompi-ddt branch.

This commit was SVN r21756.
2009-08-04 14:44:42 +00:00
Jeff Squyres
90d6491737 Since ompi_info is no longer written in C++, we no longer require a
C++ compiler in configure.  If we have a C++ compiler, then the MPI
C++ bindings are built by default.  If we don't have a C++ compiler,
then the MPI C++ bindings are not built by default.

--enable-mpi-cxx will now force an error if there is no C++ compiler
available.  --disable-mpi-cxx (or the lack of a C++ compiler) will now
disable many of the C++ compiler checks in configure.

Note that there are a few items to clean up regarding the difference
between C's _Bool type and C++'s bool type.  Right now, we assume that
they are the same.  But they aren't, and they shouldn't be treated as
such.  This cleanup will be forced in MPI-2.2 with the introduction of
the MPI_C_BOOL MPI datatype.

This commit was SVN r21755.
2009-08-04 11:54:01 +00:00
George Bosilca
32416761ce Make the open/close symmetric with regards to the local variable
contruction/destruction.

This commit was SVN r21753.
2009-08-03 16:45:18 +00:00
George Bosilca
b3ae54fb46 As some of our components do not play nicely with the open/close functionality
in the case of errors, close the opened components if and only if the whole
opening process complete succesfully.

This commit was SVN r21752.
2009-08-03 16:44:12 +00:00
Avneesh Pant
af09e7678c Convert a few opal_output() calls to instead use orte_show_help() as well as do some minor cosmetic changes dealing with tab spacing and c-blocks being enclosed with \{\}. There was also a long standing bug with the PSM mtl if the number of hardware contexts on adapter were less than the number of cores on a node (The default case is they are the same hence no issues were reported). For completeness we take care of this case as well but it requires us to tell PSM how many local processes are running on a node and the local rank of the process on a node so it can allocate the available hardware contexts appropriately.
This commit was SVN r21745.
2009-07-30 02:55:20 +00:00
Jeff Squyres
bff6a1be93 Gah! "vis", not "visi"
This commit was SVN r21741.
2009-07-29 14:22:46 +00:00
Jeff Squyres
3b9ca5a084 Reduce the label name so that it's not longer than 24 chars
This commit was SVN r21740.
2009-07-29 14:19:44 +00:00
George Bosilca
0bf381e931 This patch try to solve a issue on Leopard. The supposedly global
variables that are not initialized and are declared in a file that
doesn't export any globally visible function are marked as
non-initialized constants, i.e. uninitialized common symbols. For some
obscure reasons, they get removed from the object files on Mac OS X.

So far I found two solution to this problem. One require the addition
of "-c" to the linker command, the second one (corresponding to this
patch) force them to became a common initialized symbol.

This commit was SVN r21739.
2009-07-28 17:06:16 +00:00
Avneesh Pant
38e48d4e2f Add support for MCA parameters for PSM MTL to specify IB unit, port, IB service level and PSM debug level to use. Also specify in the openib btl params file that QLogic hardware supports a max inlined messages size of 0 only.
This commit was SVN r21734.
2009-07-24 20:09:39 +00:00
Jeff Squyres
c7376ae053 Start using Libtool's shared library versioning scheme. See lengthy
note in VERSION file.

NOTE: the versions will ''always'' be 0:0:0 on the SVN trunk and
developer branches.  They will only have meaningful values (starting
with 0:0:0 in 1.3.4) on release branches.  Only RM's will modify these
values immediately preceeding a release.

This commit was SVN r21729.
2009-07-23 21:35:17 +00:00
Edgar Gabriel
9a369d0fc1 - we accidentally decreased the counter for the number of dynamic
communicators twice, once in dpm.disconnect_wait, and once in
comm_free. The second location seems to be the right place for that (since a
communicator could be freed, and not disconnected), remove the instance in
disconnect_wait.

- add some error messages in case something goes wrong.

This commit was SVN r21720.
2009-07-20 19:54:24 +00:00
Terry Dontje
d432c9fdbc Add asserts to catch when btl_eager_limit is smaller than the pml headers.
This commit was SVN r21707.
2009-07-17 14:54:18 +00:00
George Bosilca
8275120656 Get rid of the ompi_convertor.h header file. Replace all references to ompi_convertor
by opal_convertor.
Cleanup the pcie BTL.

This commit was SVN r21703.
2009-07-16 19:13:30 +00:00
George Bosilca
2b57fc3835 Add the missing headers.
This commit was SVN r21701.
2009-07-16 18:42:14 +00:00
George Bosilca
3e971e61f3 The system headers are supposed to be protected by #ifdef and not by #if.
This commit was SVN r21700.
2009-07-16 18:27:33 +00:00
George Bosilca
d07ffedc54 No opal datatype functions in the BTL. The datatype attached to the
convertor is an ompi_datatype_t so calling the ompi level functions
is the way to go.

This commit was SVN r21698.
2009-07-16 18:25:08 +00:00
Jeff Squyres
9dc7f884b2 Fix yet another compile error from the great DDT split (r21641). Sigh.
This commit was SVN r21697.

The following SVN revision numbers were found above:
  r21641 --> open-mpi/ompi@6c5532072a
2009-07-16 18:08:03 +00:00
George Bosilca
e1383027e1 Correct a comment and cleanup/reorder the code.
This commit was SVN r21696.
2009-07-16 17:41:32 +00:00
George Bosilca
33e7cc864c The ops should be indexed based on the MPI datatype index (which is
actually the same as the Fortran value of the MPI type).

This commit was SVN r21689.
2009-07-15 21:30:09 +00:00