1
1
Граф коммитов

3933 Коммитов

Автор SHA1 Сообщение Дата
Jeff Squyres
90ef93b763 First cut of user docs (very rough; copied from LAM/MPI -- insertted
much of README.alpha in the release notes).

This commit was SVN r5145.
2005-04-02 15:11:28 +00:00
Jeff Squyres
7d9ea5b084 Add note to packagers
This commit was SVN r5144.
2005-04-02 13:59:08 +00:00
Jeff Squyres
693a742ec7 Include README.alpha in the tarball. Doh!
This commit was SVN r5143.
2005-04-02 00:49:19 +00:00
George Bosilca
84308a6e62 The is a relation between the first fragment and the GM segment size. Check to see that the user set it correctly. The
first fragment should always be less than the GM segment size (the difference should be the size of the rendez-vous header).
If the first fragment size is not set correctly force the correct value. However in the ompi_info we will still see the
default one.

This commit was SVN r5142.
2005-04-01 23:03:50 +00:00
Jeff Squyres
4107bdf17b Add some typedefs to enable the updated unit tests
This commit was SVN r5141.
2005-04-01 22:48:27 +00:00
Jeff Squyres
709a3e498e Convert gpr_internal_fns to use the new support functions and
internally call function pointers instead of direct calls

This commit was SVN r5140.
2005-04-01 22:18:24 +00:00
Jeff Squyres
842a4e904e Add some new support functions to the unit test library -- for
opening, examining, and closing components.

This commit was SVN r5139.
2005-04-01 22:13:05 +00:00
Ralph Castain
9cbb698ada Begin to add unit tests for the API layer on ras and rmaps.
Fix a compiler warning on dump.

This commit was SVN r5138.
2005-04-01 20:17:49 +00:00
Ralph Castain
d900d8f137 Plug one memory leak - still have one big one left here.
This commit was SVN r5137.
2005-04-01 19:51:47 +00:00
Jeff Squyres
b583c40102 Fix some compile errors, and make the test be skipped if there's no
thread support.

This commit was SVN r5136.
2005-04-01 19:50:17 +00:00
Ralph Castain
96508b3cb3 Continue expanding the range of unit tests. Fix a few memory leaks - orterun continues to leak like a sieve, but progress is being made.
This commit was SVN r5135.
2005-04-01 18:02:01 +00:00
Brian Barrett
d737000a5e * fix to make platform detection work on Linux Sparc
This commit was SVN r5134.
2005-04-01 15:49:27 +00:00
Ralph Castain
45e68f9714 Clean up a little. Still can't resolve the dump problem when multiple processes exist - letting it go for now.
This commit was SVN r5133.
2005-04-01 15:47:41 +00:00
Jeff Squyres
6b875da3b2 It makes more sense to have a version < 1.0 right now... Let's do
0.8, so that we can do 0.9 for a "beta".

This commit was SVN r5132.
2005-04-01 13:48:06 +00:00
Jeff Squyres
765250cb50 Fix compiler warnings and type comparisons
This commit was SVN r5131.
2005-04-01 13:35:57 +00:00
Graham Fagg
73a4044bff Forgot data unpacked in header (buffer might be misaligned?) + misc tidy up
This commit was SVN r5130.
2005-04-01 06:15:34 +00:00
Graham Fagg
c2420034aa Unpack alignment fix. Opposite of pack. Here we memcpy from the buffer to aligned memory before converting unless it is a char/byte/void pointer.
Final destination is always assumed to be aligned.

This commit was SVN r5129.
2005-04-01 06:00:24 +00:00
Graham Fagg
4f464e3d93 Alignment fix.
Packing. For packing we only need to change routines that assign to a typed (non void/char) pointer that might no longer be
word aligned etc. The source pointers are assumed to be aligned (and are kept or else the htonX calls become more complex).
Most routines not touched as they recurse and call base routines which have been changed.
memcpy pack passes dps_test with the pointer assignment unpack routines so buffers look to be packed as before :)

This commit was SVN r5128.
2005-04-01 04:56:25 +00:00
Brian Barrett
3b7d6acf2e now that there are two rmgr components this bug pops up.. Don't call
finalize() on the old best_module unless it actually has a finalize
function (which neither of the current ones have, btw).

Fixes a bus error on OS X.

This commit was SVN r5127.
2005-04-01 03:13:22 +00:00
Jeff Squyres
5733c5e0de Some minor updates / fixes
This commit was SVN r5125.
2005-04-01 01:18:35 +00:00
Jeff Squyres
1e4826e5b9 Remove all the .ompi_unignore files when making a nightly tarball so
that we don't accidentally include a directory in a tarball that we
shouldn't

This commit was SVN r5124.
2005-04-01 01:14:13 +00:00
Brian Barrett
f9673a8ba9 * add some missing header files so that it compiles on OS X
This commit was SVN r5123.
2005-04-01 00:30:37 +00:00
Jeff Squyres
62f4f7e82f Per Tim's advice, change oob to rml.
This commit was SVN r5122.
2005-03-31 23:29:42 +00:00
Jeff Squyres
d2eac6dab1 Move the rml close down right before the oob close. It doesn't really
matter, since we're not selecting/using any rml or oob components, but
it makes it closer to the ordering in ORTE.

This commit was SVN r5121.
2005-03-31 23:26:02 +00:00
Jeff Squyres
ba5dcfcb4a Move the initialization of the pls components into the selection --
not during the open.  This commit looks bigger than it really is --
most of this is just moved code, and added 2 bool's to indicate
whether the opened and available lists are valid for whether we
traverse them in the base close function or not. 

This commit was SVN r5120.
2005-03-31 23:25:24 +00:00
Jeff Squyres
fa81e66db8 Move the posting and canceling of the non-blocking receive from the
component to the module.

This commit was SVN r5119.
2005-03-31 23:22:44 +00:00
Jeff Squyres
4f461a874e First cut at a README.alpha file
This commit was SVN r5118.
2005-03-31 22:40:40 +00:00
Tim Woodall
53d51363ba move recv_cancel from close to finalize
This commit was SVN r5117.
2005-03-31 22:37:46 +00:00
Tim Woodall
7eb87a1822 - change registry call - need to look at why this didn't
work w/ KEYS_AND

This commit was SVN r5116.
2005-03-31 22:30:22 +00:00
Tim Woodall
341ad3c57c cancel any pending recvs
This commit was SVN r5115.
2005-03-31 22:11:54 +00:00
Ralph Castain
df7ef31b61 Plug a memory leak that crept into the registry.
Modify the locking scheme to try and resolve a problem with dump_triggers that only occurs with multiple processes. Didn't resolve the problem, but should be more robust anyway. Still tracking this one down.

This commit was SVN r5114.
2005-03-31 21:32:12 +00:00
Tim Woodall
74cc7ba381 on abnormal termination - dump info on the aborted process
This commit was SVN r5112.
2005-03-31 19:39:02 +00:00
Tim Woodall
81e9377c87 use nonblocking send/recv to send terminate message to the daemons
as they may have already exited

This commit was SVN r5111.
2005-03-31 18:53:35 +00:00
Tim Woodall
447d370905 - added proxy resource manager which is loaded when not the seed
- added support to pls fork/rsh modules for terminate_job

This commit was SVN r5110.
2005-03-31 15:47:37 +00:00
Ralph Castain
8686c60223 Fix the subscription system so it correctly deals with triggers at a level (as opposed to comparing two counters). Combined with Brian's latest checkin, this corrects the tendency for orterun to "hang" when one or more processes abnormally terminate.
This commit was SVN r5109.
2005-03-31 14:24:36 +00:00
Brian Barrett
5753c6a47f Only set the state of the processes the daemon was responsible for to
ABORTED if the ssh that started the daemon exited abnormally.  Otherwise,
bad things happen if all the processes on that node exit before the
processes on other nodes.

This patch is bigger than it should be because I had to indent a bunch of code 
when I moved the if statement.

This commit was SVN r5107.
2005-03-31 04:23:55 +00:00
Jeff Squyres
13a4aee1a5 - Add ignore file so that this component is not included in the
tarball
- Guess at usernames for unignore -- will send out mail shortly to
  verify

This commit was SVN r5106.
2005-03-31 03:24:16 +00:00
Jeff Squyres
f783edb272 Add ignore files so that these components are not included in the
tarball.

This commit was SVN r5105.
2005-03-31 03:23:22 +00:00
Brian Barrett
6c476d7822 * fix free() of NULL when no MPI applications are started
This commit was SVN r5104.
2005-03-30 18:03:08 +00:00
George Bosilca
c7fe83c845 The condition to be able to use the user buffer for the receive is to have a contiguous datatype AND to have
it's extent equal to the size.

This commit was SVN r5103.
2005-03-30 15:03:46 +00:00
George Bosilca
9100aa945d Do not forget to set the upper bound for predefined datatypes.
This commit was SVN r5102.
2005-03-30 15:02:34 +00:00
Ralph Castain
e3471e1342 Update the unit test matrix
This commit was SVN r5101.
2005-03-30 14:16:02 +00:00
Jeff Squyres
8a63a3d3d8 Make the test to find sched_yield() better so that we don't always
link in -lr on linux.

This commit was SVN r5100.
2005-03-30 01:43:16 +00:00
Brian Barrett
043e7682d2 * Make sure to free the callbacks array during finalize
* add MCA parameter (OMPI_MCA_mpi_yield_when_idle) to cause sched_yield()
  to be called when the progress engine is called and nothing happens.
  Default is to call sched_yield().
* add MCA parameter (OMPI_MCA_mpi_event_tick_rate) to adjust the rate
  at which the event library is called from ompi_progress.  When set
  to 0, the event library will never be ticked.  When set to 1, the
  event library will be progressed every time.  2 every other, etc.

The MCA parameters are only in effect from end of MPI_Init to start of
MPI_Finalize.

This commit was SVN r5099.
2005-03-30 01:40:26 +00:00
Brian Barrett
e00782ed7d * add missing string.h include and copyright header
This commit was SVN r5098.
2005-03-30 01:31:30 +00:00
Jeff Squyres
5eb51821c0 Somehow missed editing these bindings in the big f77 commit the other
day (to convert "void*" -> real function pointer types, and "char*
extra_state" to "MPI_Fint* extra_state").

George still gets to cleanup/finish MPI_REGISTER_DATAREP().  :-)

This commit was SVN r5097.
2005-03-29 22:24:51 +00:00
Ralph Castain
35b4fc02e2 Cleanup up some of the util unit tests. Fix a detected problem in orte_os_path for relative path names.
This commit was SVN r5096.
2005-03-29 19:58:32 +00:00
Tim Woodall
72b3a823c3 need to use integer when passing jobid on command line
This commit was SVN r5095.
2005-03-29 19:41:29 +00:00
Tim Woodall
f4f24cba9f cleanup, correct acks when forwarding through an intermediary
This commit was SVN r5094.
2005-03-29 19:40:38 +00:00
Ralph Castain
5f2051caad Fix it so that failure of getpwent defaults to using a string version of the uid - so session directories still work on bizarre systems that don't push the passwd file to the backend nodes...
This commit was SVN r5093.
2005-03-29 19:13:28 +00:00