ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность
2 730 коммитов 3 Ветки 0 Теги 130 MiB
b2cc35056c
Граф коммитов

2730 Коммитов

This Branch
This Branch
Все ветки
Автор SHA1 Сообщение Дата
Ralph Castain
b2cc35056c Remove a debug output. 
This commit was SVN r2911.
 2004-10-01 22:37:29 +00:00
Ralph Castain
3c92d18fc7 Consolidate the RTE startup sequence into a single function call for simpler maintenance. We seem to have this debugged enough now to commonize the startup across the various programs. Modify mpi_init, mpirun, openmpi, ompid, and ompiconsole accordingly. 
This commit was SVN r2910.
 2004-10-01 22:22:21 +00:00
Tim Woodall
a0542dd97c George --- this seems to fix a problem w/ calculating the correct offset 
for contigous datatypes.

This commit was SVN r2909.
 2004-10-01 20:08:39 +00:00
Tim Woodall
77460dc489 determine the correct offset for multi-fragment messages 
This commit was SVN r2908.
 2004-10-01 20:07:37 +00:00
Jeff Squyres
7f2b73a4e5 Fix for bug 989: parse command line properly. Good code to tweak 
around with while waiting for other things to compile.  :-)

Since there were some unit tests for the argv interface, took the
liberty of updating it for two new functions that were necessary:
ompi_argv_delete() and ompi_argv_insert().

This commit was SVN r2907.
 2004-10-01 18:38:16 +00:00
Jeff Squyres
3f8c5372c6 Fix header file name 
This commit was SVN r2906.
 2004-10-01 18:26:18 +00:00
Jeff Squyres
ff230303f3 Oops! This one is almost funny. :-) 
This commit was SVN r2905.
 2004-10-01 18:24:03 +00:00
Jeff Squyres
02cc1ccdc9 No need to do anything for sendcount==0 case 
This commit was SVN r2904.
 2004-10-01 18:23:39 +00:00
Sayantan Sur
ca2273ed7c Short & Long messages working. 
This commit was SVN r2903.
 2004-10-01 15:24:59 +00:00
Edgar Gabriel
5b03c18c07 minor fixes. If I start the spawned process manually, 
the spawning and retrieving the parent communicator
works already. Have to debug now, what is going wrong
when starting the processes.

This commit was SVN r2902.
 2004-10-01 14:06:23 +00:00
Jeff Squyres
11fc271e69 Based on a recent discussion with George, fix bug 977 to prevent 
reading from outside the allocated area.  This bug was *probably*
harmless, but valgrind correctly identified a read from outside the
alloca'ed area (although that value was subsequently never used).
It's safer to just never read it in the first place.

Added "JMS" comments in the code where I made changes.  Following up
with a mail to George...

This commit was SVN r2901.
 2004-10-01 12:16:45 +00:00
Jeff Squyres
cd6fd000df Oops -- I didn't check that *all* datatypes were added into the f_to_c 
table in my last commit.  This commit fixes that, and then re-patches
up mpif.h.in to match.

I'm not sure that we need *all* of the datatypes in there (e.g., do we
really need the C datatypes?), but since George isn't around, I
followed the "when in doubt, it doesn't hurt to add more entries to
the table" rule.  Following up with a detailed message to George...

This commit was SVN r2900.
 2004-09-30 22:14:35 +00:00
Ralph Castain
55fa43dbcf Move the definition of the OMPI_NAME_ARGS macro to the name server for general use - great shorthand for obtaining the three components when outputing diagnostic info! 
This commit was SVN r2899.
 2004-09-30 21:34:29 +00:00
Tim Woodall
da4b57d6cf - use loopback for oob/ptl if no other i/fs are present 
- force init to fail if no i/fs are available

This commit was SVN r2898.
 2004-09-30 21:23:10 +00:00
Tim Woodall
5452038f99 - re-implemented module exchange to use the registry - note this implementation 
needs to be revisited - as it is probably the most inefficient possible - but
  was the quickest to implement
- added downcalls into pml when new procs are added

This commit was SVN r2897.
 2004-09-30 20:54:26 +00:00
Brian Barrett
104e6ce2b7 * Don't use localhost for login - convert to machine's hostname 
* unregister the processes when we know they're dead.  This seems to still
  cause some hangs (but then, so did not doing this)

This commit was SVN r2896.
 2004-09-30 18:45:14 +00:00
Jeff Squyres
ffea11f88e Fix a few more bugs with fortran collectives. 
This commit was SVN r2895.
 2004-09-30 18:03:30 +00:00
Jeff Squyres
1b80a6325b Fix for bug 997 -- ensure that the d_f_to_c_index field is initialized 
properly.

This commit was SVN r2894.
 2004-09-30 17:59:24 +00:00
Tim Woodall
293bdf7f49 oops - fix for single threaded case 
This commit was SVN r2893.
 2004-09-30 16:46:44 +00:00
Tim Woodall
9dab6ab95b fix for publish/subscribe 
This commit was SVN r2892.
 2004-09-30 16:34:48 +00:00
Brian Barrett
550469cb0b * move wiatpid shutdown code into ompi_rte_finalize 
* remove now unneeded ompi_event_fini from mpirun

This commit was SVN r2891.
 2004-09-30 16:23:08 +00:00
Jeff Squyres
ff4c11368c No need for these anymore -- we won't be invoked if the count is zero. 
This commit was SVN r2890.
 2004-09-30 15:56:40 +00:00
Jeff Squyres
db04147154 Don't invoke back-end coll functions if the count is zero (ddt 
functions don't behave well when called with 0 counts)

This commit was SVN r2889.
 2004-09-30 15:56:08 +00:00
Jeff Squyres
1f668625d3 Update some fortran constants 
This commit was SVN r2888.
 2004-09-30 15:55:40 +00:00
Tim Woodall
a222c702ec cleanup of finalize code 
- unregister all event handlers from event library
- cancel pending non-blocking receives with oob

This commit was SVN r2887.
 2004-09-30 15:09:29 +00:00
Tim Woodall
9a4d4a6e73 added capability to enable/disable event processing 
This commit was SVN r2886.
 2004-09-30 14:45:25 +00:00
Ralph Castain
b42a361302 Patch a few things that were causing trouble for programs that re-entered the registry during a callback function. Also fixed a timing problem in rte_monitor - ensured that we were in fact already waiting on a condition before generating a wakeup signal. Adjusted the timing of mpirun to ensure that the synchro to alert mpirun of all-processes-done got registered before they completed. 
This commit was SVN r2885.
 2004-09-29 21:54:57 +00:00
Brian Barrett
d5f4ebde71 * add some comments about what the spawn selection constraints mean 
* memory leak cleanups
* implement rsh's kill_proc and kill_job for the case where we
  keep the ssh connections alive.  At least, I think this will work.
  Need to test some more.

This commit was SVN r2884.
 2004-09-29 21:29:51 +00:00
Jeff Squyres
f14cc9d4b9 Shame on me for not reviewing my diff before committing. :-( 
Remove some debugging junk.

This commit was SVN r2883.
 2004-09-29 21:29:32 +00:00
Jeff Squyres
5818adf6a4 Put in temporary workaround for non-communative reduction operations 
-- need to talk to Edgar/George about how to fix this properly.  But
now we at least give correct answers.

This commit was SVN r2882.
 2004-09-29 21:09:46 +00:00
Jeff Squyres
0a6069a613 Ditch weirdness for non-communative operations -- it wasn't necessary 
and caused incorrect answers :-)

This commit was SVN r2881.
 2004-09-29 21:09:08 +00:00
Jeff Squyres
f9e1c172c6 Fix some minor errors in the predefined MPI_Op's 
This commit was SVN r2880.
 2004-09-29 21:08:29 +00:00
Brian Barrett
d3adf8d816 * add abillity to notify mpirun that a process has died from rms. This just 
prints a warning now, since the infrastructure to do the notify doesn't
  work yet.

This commit was SVN r2879.
 2004-09-29 20:40:29 +00:00
Brian Barrett
0fb5abc6ab * add a whole bunch of headers to the list of headers to check for. These 
are all tests that will fail on Windows so we need to protect their
  inclusion in source code.

This commit was SVN r2878.
 2004-09-29 19:42:16 +00:00
Brian Barrett
452e5fd0f7 * want portable signal.h not non-portable sys/signal.h 
This commit was SVN r2877.
 2004-09-29 18:40:46 +00:00
Tim Woodall
f76e257172 - cleanup some debug output 
- ompi_rb_tree is broken - added temporary logic to look for peer
  in list if not found in tree

This commit was SVN r2876.
 2004-09-29 17:18:14 +00:00
Brian Barrett
e76e587990 * declare environ so that compiles on OS X 
This commit was SVN r2875.
 2004-09-29 13:11:29 +00:00
Edgar Gabriel
05a28efd1f first cut on the comm_spawn mechanism. It doesn't work yet 
(and I don't know why), but it also doesn't seem to break anything else...

This commit was SVN r2874.
 2004-09-29 12:41:55 +00:00
Prabhanjan Kambadur
20dda7d84f Reordering the steps in comm_destruct 
This commit was SVN r2873.
 2004-09-29 05:34:20 +00:00
Prabhanjan Kambadur
f4b05be986 Some really reall dumb typing mistakes. Hopefully this commit will not be seen :-) 
This commit was SVN r2872.
 2004-09-29 04:24:37 +00:00
Brian Barrett
21918bd59b * fix missing errno.h 
This commit was SVN r2871.
 2004-09-28 23:18:57 +00:00
Brian Barrett
84eb32ba31 * Move the pid tracking code out to the RMS pcm and into the pcm base 
so that the rsh pcm can use it as well.
* do the right things so that ssh exits if HIGH_QOS was not requested or
  stays alive and is monitored if HIGH_QOS is requested (which is the
  default for mpirun).
* Add code to support notifying people that procs have died, but turned 
  off since it will deadlock the gpr.

This commit was SVN r2870.
 2004-09-28 23:03:22 +00:00
Jeff Squyres
386336f50d Re-enable a commented out virtual destructor for MPI::Comm_null 
This commit was SVN r2869.
 2004-09-28 18:40:24 +00:00
Ralph Castain
b5e21eaac3 Fix a missing include file in ompi_rte_wait that caused the build to fail. 
Minor change to oob_base_init - point oob_name_self at correct name.

This commit was SVN r2868.
 2004-09-28 10:33:09 +00:00
Brian Barrett
a6963be12e * back out parts of r2864, moving calls to ompi_event_fini() back into 
MPI_Finalize and mpirun so that we shut down the event library before
  the TCP PTL.  This needs to change before release so that the RTE 
  components can deregister properly, but we need to run in the mean time

This commit was SVN r2867.

The following SVN revision numbers were found above:
  r2864 --> open-mpi/ompi@57ca18ce88
 2004-09-28 01:38:16 +00:00
Jeff Squyres
fd1e643ef1 Update svn:ignore 
This commit was SVN r2866.
 2004-09-28 00:08:54 +00:00
Jeff Squyres
e1ee08e632 Fix size mismatch that was causing assertion failures when compiled 
with -O3 (i.e., memory overwrites into memory that it shouldn't have).

This commit was SVN r2865.
 2004-09-27 23:02:37 +00:00
Brian Barrett
57ca18ce88 * move ompi_event_fini() from mpirun/MPI_Finalize to ompi_rte_finalize to 
match where ompi_event_init() lived
* initialize and shutdown the code to allow child process wait callbacks
* add comment about few() ussage in rte-enabled jobs (short answer:
  don't).

This commit was SVN r2864.
 2004-09-27 19:38:23 +00:00
Jeff Squyres
8bef16d214 - some OS's #define malloc, calloc, free, realloc -- ensure to #undef 
them when necessary
- we didn't have the re-definition for calloc

This commit was SVN r2863.
 2004-09-27 17:15:06 +00:00
Brian Barrett
4cd3d3aa56 * remove testing code - not really needed any more 
This commit was SVN r2862.
 2004-09-27 16:40:43 +00:00