ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность
3 344 коммитов 3 Ветки 0 Теги 130 MiB
7753253084
Граф коммитов

3344 Коммитов

This Branch
This Branch
Все ветки
Автор SHA1 Сообщение Дата
Jeff Squyres
7753253084 Eliminate dead code. 
This commit was SVN r3740.
 2004-12-07 22:35:35 +00:00
Jeff Squyres
b9f1975141 Eliminating warnings: per Edgar's advice, this code is definitely not 
being used.  I added some comments about why it's not being used
(because one would naievely think that increasing / decreasing the
refcount would be a Good Thing for the group constructor /
destructor).

This commit was SVN r3739.
 2004-12-07 22:35:10 +00:00
Ralph Castain
e259750296 Clean-up a few things - mostly adding a lot of diagnostics to chase down the comm_spawn issue. Believe this may fix that problem, but still checking that for sure. At least now it starts processes correctly again! 
This commit was SVN r3738.
 2004-12-07 19:38:42 +00:00
Jeff Squyres
03b9ed1ff5 Make all the f77 prototypes have proper MPI_Offset parameters (i.e., 
the integer size must be the same between Fortran and C).

This commit was SVN r3737.
 2004-12-07 19:12:40 +00:00
Brian Barrett
13bf550a0f *checkpoint some bproc pcm changes to move machines 
* add first take of mapper component for BJS

This commit was SVN r3736.
 2004-12-07 19:03:20 +00:00
Brian Barrett
947f166282 * abort if no bproc so that we don't try to build 
This commit was SVN r3735.
 2004-12-07 18:23:07 +00:00
Tim Woodall
ea5991f8fa protect creation of global lists 
This commit was SVN r3734.
 2004-12-07 18:16:26 +00:00
Tim Woodall
63935c0da4 corrected cleanup logic 
This commit was SVN r3733.
 2004-12-07 18:14:55 +00:00
Brian Barrett
6dc570c7a1 * Implement kill_job and kill_proc code 
This commit was SVN r3732.
 2004-12-07 15:41:40 +00:00
Tim Woodall
67a2d47a49 dont attempt to progress events for threaded case as this is handled in a seperate thread 
This commit was SVN r3731.
 2004-12-07 15:39:57 +00:00
Tim Woodall
448c6632ab first cut at support for threaded case - latency is bad 
This commit was SVN r3730.
 2004-12-07 15:38:01 +00:00
Brian Barrett
d8e1f396a7 * remove the ompi ignore. It's good enough for testing :/ 
This commit was SVN r3729.
 2004-12-07 15:30:29 +00:00
Brian Barrett
88c94f7439 * remove dead code 
This commit was SVN r3728.
 2004-12-07 15:16:06 +00:00
Brian Barrett
0853d49593 * remove one memory leak and move the setting of BPROC_RANK to where we 
know that it is actually useful

This commit was SVN r3727.
 2004-12-07 14:34:18 +00:00
Brian Barrett
fceeecd8c5 * process death notification support for the bproc pcm 
This commit was SVN r3726.
 2004-12-07 06:07:45 +00:00
Ralph Castain
be3facae87 This still doesn't fix the comm_spawn problem, but does fix an error in the proxy-to-replica communication when requesting startup/shutdown messages. 
This commit was SVN r3725.
 2004-12-07 03:57:15 +00:00
Brian Barrett
d51cfe0fc4 * duh. fix previous commit so that we don't free NULL pointers and all that 
This commit was SVN r3724.
 2004-12-07 03:17:22 +00:00
Brian Barrett
a5cc10e799 * fix up bproc to match change in delete function 
* fix unitialized value in llm selection code 

This commit was SVN r3723.
 2004-12-07 02:58:23 +00:00
Brian Barrett
785c3d1408 * adjust the ompi_argv_delete() function so that it shortens the passed 
argv array to match the deletion of elements in the array.

This commit was SVN r3722.
 2004-12-07 02:42:17 +00:00
Brian Barrett
555b261129 * checkpoint - We start processes :). Still need to implement: 
- stdio
   - death notification messages
   - killing
  and I think there are some places we leak memory way too much.  And a
  ton of printfs.  But we're making progress

This commit was SVN r3721.
 2004-12-07 00:39:44 +00:00
George Bosilca
fcb3bfa81c Typo ... 
This commit was SVN r3720.
 2004-12-07 00:18:37 +00:00
George Bosilca
857916a426 Indentation :) 
This commit was SVN r3719.
 2004-12-07 00:17:58 +00:00
George Bosilca
20ca7c2234 Check the boundaries for the memcpy using the correct length. 
This commit was SVN r3718.
 2004-12-07 00:17:28 +00:00
George Bosilca
071225f062 Correctly compute the displacement when we create a stack with a starting position. This displacement 
shoud be ZERO based (starting from the begining of the data) not stack related.

This commit was SVN r3717.
 2004-12-07 00:16:58 +00:00
Tim Woodall
76e6494de6 dont progress tcp if not selected for any 
This commit was SVN r3716.
 2004-12-06 20:53:22 +00:00
Tim Woodall
d595837d5d check for valid mmap file before cleaning up 
This commit was SVN r3715.
 2004-12-06 19:41:11 +00:00
Brian Barrett
0d68912417 * checkpoint. Now redirects I/O to file (if only BProc only did the right 
things with this!) and seems to get the right component loaded on the
  started app.  Unfortunately, it all goes down hill from there - the
  processes either die or hang in evil and twisted ways.

This commit was SVN r3714.
 2004-12-06 19:38:06 +00:00
Brian Barrett
f1e28f8f04 * make the arguments for show_help correct 
This commit was SVN r3713.
 2004-12-06 19:36:39 +00:00
Tim Woodall
171d50b0de release lock while in subscribe call 
This commit was SVN r3712.
 2004-12-06 18:37:14 +00:00
Tim Woodall
52ff849ecb fill in request status 
This commit was SVN r3711.
 2004-12-06 17:43:23 +00:00
Ralph Castain
ac0f63c3d8 Remove a debug message that isn't needed any more and was always "on". 
This commit was SVN r3710.
 2004-12-06 16:57:29 +00:00
Ralph Castain
1454832bf7 Mostly just added some diagnostic messages to help chase down a problem in comm_spawn. Fixed an error in the gpr notification system - not sure if that totally fixes the problem, but definitely made progress on it. 
This commit was SVN r3709.
 2004-12-06 16:50:45 +00:00
Craig E Rasmussen
8304d0c5fa Initial version. These stylesheets create the MPI F90 interfaces in the 
mpi module.

This commit was SVN r3708.
 2004-12-06 16:32:57 +00:00
Craig E Rasmussen
b9da854182 Initial version. These stylesheets create the F90 procedures containing 
void* parameters (e.g., MPI_Send) needed to forward the F90 call on to the
F77 implementation.

This commit was SVN r3707.
 2004-12-06 16:32:03 +00:00
Craig E Rasmussen
2c20ebba44 Initial version. This file creates the definitions for the F90 interfaces 
with void* parameters (e.g., MPI_Send).  These procedures just call
corresponding procedures in the F77 implementation.  For procedures without
void*, the F77 implementation is used directly and doesn't need an extra
call.

This commit was SVN r3706.
 2004-12-06 16:27:40 +00:00
Tim Woodall
5e043924ff corrections 
This commit was SVN r3705.
 2004-12-06 15:58:57 +00:00
Brian Barrett
611c594b16 * checkpoint. Still doesn't work (I think there's some memory badness in 
here somewhere), so still .ompi_ignored

This commit was SVN r3704.
 2004-12-06 03:59:51 +00:00
Ralph Castain
b9a1132450 Cleanup a bunch of warnings. Most were innocent enough (things not initialized that probably wouldn't be a problem), but a couple could be problematic. 
One warning on my list was left unfixed: the "mca_ptl_tcp_peer_dump" function is defined but not used. It is commented in the file as a diagnostic, so I left it alone since someone may choose to use it for debugging purposes. For now, it is uncalled anywhere in the code.

This commit was SVN r3703.
 2004-12-06 02:15:34 +00:00
Jeff Squyres
1adf5cb3a0 Remove type punning warnings; tell gcc that it cannot assume that 
strict anti-aliasing can occur

This commit was SVN r3702.
 2004-12-05 21:33:46 +00:00
Jeff Squyres
76c9f486e5 Remove useless file (it's ot used anywhere). 
This commit was SVN r3701.
 2004-12-05 21:32:43 +00:00
Ralph Castain
7287fe400c Fix two things: 
1. make it impossible to leave the thread locks set when errors are detected (couple of places had a path out of the function without clearing the lock)

2. redefine notify all so it doesn't trigger every time

This commit was SVN r3700.
 2004-12-05 16:31:51 +00:00
Brian Barrett
920fc1c494 * some code reorg to block into functional units 
This commit was SVN r3699.
 2004-12-05 00:14:14 +00:00
Brian Barrett
2a701cb6f4 * don't free() the return of getenv() - that's just silly 
This commit was SVN r3698.
 2004-12-05 00:07:10 +00:00
Brian Barrett
76d48e20f8 * checkpoint bproc changes 
This commit was SVN r3697.
 2004-12-04 22:13:32 +00:00
Brian Barrett
dd49abce62 * Handle case where multiple calls to bproc_vexecmove() where required to 
start this particular job by adding an offset that will be added to the
  BPROC_RANK assigned during the start
* Clean up comment a bit

This commit was SVN r3696.
 2004-12-04 20:06:50 +00:00
Brian Barrett
626d03745d * add BProc pcmclient code so that processes know who they are when started 
under BProc.  Of course, we can't start processes under BProc just yet,
  but that's coming real soon now.

This commit was SVN r3695.
 2004-12-04 19:49:35 +00:00
Edgar Gabriel
eaaa6fcd59 one minor fix for the proc reference counting in dynamic scenarios 
This commit was SVN r3694.
 2004-12-04 14:47:13 +00:00
Brian Barrett
5ac2b0988b * remove dead code that was causing warning 
This commit was SVN r3693.
 2004-12-04 00:42:05 +00:00
Ralph Castain
5e560cb148 Fix a problem that affected attributes - we were missing the vm_register function call. 
This commit was SVN r3692.
 2004-12-03 21:05:22 +00:00
Edgar Gabriel
a4c8e4f15c some further refcount fixes. Now the static scenario should really be correct... :-) 
This commit was SVN r3691.
 2004-12-03 14:45:15 +00:00