ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность
3 238 коммитов 3 Ветки 0 Теги 130 MiB
bcd498a2e1
Граф коммитов

3238 Коммитов

This Branch
This Branch
Все ветки
Автор SHA1 Сообщение Дата
Jeff Squyres
22f9115c82 Fix component close issue with topo, coll, and io. 
This commit was SVN r3556.
 2004-11-12 16:55:41 +00:00
Tim Woodall
73385f189e - use common macro for atomic add 
This commit was SVN r3554.
 2004-11-11 21:39:48 +00:00
Rich Graham
6590713a5c fix some bugs with new shared memory implementation. 
This commit was SVN r3553.
 2004-11-11 20:34:28 +00:00
Tim Woodall
19f0debe2f cleanup 
This commit was SVN r3552.
 2004-11-11 16:07:51 +00:00
Tim Woodall
2c7098b8cf add macro to conditionally use atomic operations if required 
This commit was SVN r3551.
 2004-11-11 15:41:07 +00:00
Tim Woodall
95d5165478 - move posting recv out of critical path 
This commit was SVN r3550.
 2004-11-11 15:01:48 +00:00
George Bosilca
f2c697170c Temporary checkpoint. 
This commit was SVN r3547.
 2004-11-09 16:01:53 +00:00
Tim Woodall
d88c20d724 fix make dist 
This commit was SVN r3546.
 2004-11-08 19:51:22 +00:00
Mitch Sukalski
ab7cfab702 changed address pointer casting from "unsigned int" 
to "unsigned long" so we don't truncate address on
64-bit machines...

This commit was SVN r3545.
 2004-11-08 19:33:29 +00:00
Jeff Squyres
6248540177 #$%^#$%#@$%!!! 
This commit was SVN r3544.
 2004-11-08 12:48:57 +00:00
Jeff Squyres
a1e7233dd1 Per MPI-2:88-89 -- if we are comm_free'ing the parent, then subsequent 
calls to comm_get_parent shuld return comm_null.  Additionally, if
it's freed, we don't want to free it again during ompi_comm_finalize.
So if we're freeing the user's handle to the parent comm (i.e., we're
not calling ompi_comm_free on ompi_mpi_comm_parent itself), then
explicitly set ompi_mpi_comm_parent to &ompi_mpi_comm_null.  This does
two things:

1. subsequent calls to comm_get_parent will return comm_null
2. we won't try to ompi_comm_free the ompi_mpi_comm_parent in
   ompi_comm_finalize() 

This commit was SVN r3543.
 2004-11-08 05:04:54 +00:00
Brian Barrett
6ea03e913d * remove debugging printf committed recently 
This commit was SVN r3539.
 2004-11-07 19:12:28 +00:00
Tim Woodall
46c70800a1 correct error returns - release locks 
This commit was SVN r3536.
 2004-11-07 03:30:37 +00:00
Tim Woodall
e391c90105 correct the return value to the allocated buffer - in the 
case of a partial read, the iovec no longer points to the
start of the allocated buffer or the entire length of the
message

This commit was SVN r3535.
 2004-11-07 03:18:00 +00:00
Rich Graham
2e6eaf6d5f code changes to support sm. 
This commit was SVN r3534.
 2004-11-06 22:00:24 +00:00
Jeff Squyres
4cc5d00c2f Use a safe tag to start with. Per bug 1047, this is still not the 
proper final solution.

This commit was SVN r3531.
 2004-11-06 15:43:31 +00:00
Brian Barrett
db54399f82 * Add all the show_help() commands instead of printf(). Probably want for 
SC, as some of the error messages mpirun could emmit were pretty terse
  and unfriendly

This commit was SVN r3530.
 2004-11-06 15:16:11 +00:00
Brian Barrett
f32d1309e1 * I don't understand what is going on and I don't think it's the RMS pcm's 
fault, but using rms causes the processes to abort.  Shutting off the
  rms pcm for now so that it doesn't accidently go insane during SC.

This commit was SVN r3529.
 2004-11-06 03:34:32 +00:00
Brian Barrett
2a1182e956 * try to slow the flood into the registry at startup by slowing startup. 
Yeah, it's a hack.  But it works for now...

This commit was SVN r3528.
 2004-11-06 00:39:08 +00:00
Edgar Gabriel
664567d7f7 start counting the ports with 1 instead of zero to avoid the wildcard problem with the oob 
This commit was SVN r3527.
 2004-11-05 19:58:44 +00:00
George Bosilca
39e7432ebf Just a checkpoint ... 
This commit was SVN r3526.
 2004-11-05 14:53:38 +00:00
Edgar Gabriel
b3d5c0a04f applying a diff-file from Jeff with one minor correction. Major problem was that I dereferenced comm_null accidentally four times instead of ones comm_null, twice group_null and once the error handler of comm_parent, in case a parent communicator other then comm_null has been established. 
This commit was SVN r3525.
 2004-11-05 12:58:14 +00:00
George Bosilca
9659288e74 I hate waiting on the airports. SO I start doing something usefull ... 
I remove a lot of inter-dependence, I use the struct_t type.
BEWARE not all the function are ready.

This commit was SVN r3524.
 2004-11-05 07:52:30 +00:00
George Bosilca
90d79bb1d1 Remove a useless define. 
This commit was SVN r3523.
 2004-11-05 07:39:57 +00:00
Tim Woodall
85490f00a5 restore previous matching logic until I have more time 
to validate the changes

This commit was SVN r3522.
 2004-11-05 03:13:21 +00:00
Tim Woodall
aef2983f15 eanup-This line, and those below, will be ignored-- 
M    base/ptl_base_match.c

This commit was SVN r3521.
 2004-11-05 01:43:58 +00:00
Tim Woodall
a84f9d0a7d correct revised logic 
This commit was SVN r3520.
 2004-11-04 23:36:48 +00:00
Brian Barrett
6b0b4e0aee * temporary workaround so that you don't have to have . in your path to 
launch an app.

This commit was SVN r3519.
 2004-11-04 23:27:59 +00:00
Tim Woodall
8e0b35ad99 correction 
This commit was SVN r3518.
 2004-11-04 21:48:37 +00:00
Brian Barrett
09bd66a636 * so it turns out that you actually have to increment the counter in order 
to fill in the entire array.

  This should fix the random badness seen at the end of mpirun if you
  launched on more than one machine

This commit was SVN r3517.
 2004-11-04 21:39:46 +00:00
Tim Woodall
705065888f remove some function calls in critical path 
This commit was SVN r3516.
 2004-11-04 21:39:14 +00:00
Tim Woodall
83281249e5 cleanup logic 
This commit was SVN r3515.
 2004-11-04 21:29:22 +00:00
George Bosilca
d4b62e320a That was really a very bad C statement ... and the compiler accept it ... 
This commit was SVN r3514.
 2004-11-04 21:06:47 +00:00
George Bosilca
dc9530bfa0 Remove a useless case in the pack function. Now this behaviour is handled by a separate function. 
This commit was SVN r3513.
 2004-11-04 20:59:04 +00:00
Brian Barrett
7b6dbc9dfa * don't start the oob recv until someone actually wants updates. This 
should fix the crash in ompi_info and give the oob one less thing to
  keep track of in the common case

This commit was SVN r3512.
 2004-11-04 18:31:08 +00:00
Jeff Squyres
502b8bcddf A newly revised os_create_dirpath algorithm, slightly simpler than the 
old one, and correctly take into account Windows-specific issues that
were discovered over the last week (particularly with respect to how
stat() functions on Windows vs. POSIX).  This new version seems to
work on both POSIX and Windows systems.

This commit was SVN r3510.
 2004-11-04 13:35:15 +00:00
Rich Graham
a47d32dcde change shared memory fifo queue 'stuff' to always be allocated in 
the memory scope of the sender.  There are still several bugs to
fix for the case that the sender and receiver don't have the
shared memory pool at the same virtual memory base address.

This commit was SVN r3509.
 2004-11-04 04:03:32 +00:00
Tim Woodall
e7046525d1 - optimizations for latency 
This commit was SVN r3508.
 2004-11-04 01:01:08 +00:00
George Bosilca
a6ad19e721 Removing some compiler warnings (thanks Jeff for the patch). Now we generate only the usefull functions for moving bytes around. 
This commit was SVN r3507.
 2004-11-03 23:52:44 +00:00
Mitch Sukalski
af6362ddb5 Fixed processing of remote IB QP OOB data by adding an 
extra char to the temporary string so atoi() doesn't
accidentally run off the string...

This commit was SVN r3505.
 2004-11-03 23:17:32 +00:00
George Bosilca
52c54906b9 Move the main pack/unpack functions in the datatype.h file and make them static inline. 
Move the logic for several iovec directly in each specialized pack/unpack function.
Disable the boundaries checking.
Improuve the packing of contiguous messages.

This commit was SVN r3504.
 2004-11-03 21:57:50 +00:00
Brian Barrett
818b06803f * just forward declare the struct, since all we use is pointers to the 
struct

This commit was SVN r3501.
 2004-11-03 19:33:31 +00:00
Brian Barrett
25abbb495f * back out include of communicator.h. resulted in weird compiler badness. 
This commit was SVN r3500.
 2004-11-03 19:14:05 +00:00
Brian Barrett
384cdbb351 * need communicator.h for declaration of ompi_communicator_t 
This commit was SVN r3499.
 2004-11-03 19:11:34 +00:00
Prabhanjan Kambadur
34b5d4bc15 simple event test 
This commit was SVN r3498.
 2004-11-03 18:59:50 +00:00
Brian Barrett
093cf8239b * change tmpdir back to /tmp 
This commit was SVN r3497.
 2004-11-03 18:59:27 +00:00
Prabhanjan Kambadur
c5ddf2d528 simple event test 
This commit was SVN r3496.
 2004-11-03 18:57:39 +00:00
Prabhanjan Kambadur
357891b26b simple event test 
This commit was SVN r3495.
 2004-11-03 18:57:20 +00:00
George Bosilca
973a924ea8 A lot of bug fixes and improuvements: 
- optimized pack/unpack for contiguous datatype is back.
 - improuve the latency for contiguous datatypes by minimizing all the initialization steps.
 - other datatypes use the new stack description.

TODO: redo all the va;lidations with intel and mpich testsuite...

This commit was SVN r3494.
 2004-11-03 16:55:02 +00:00
George Bosilca
56d8631709 Correctly initialize the mutex using the init function not the unlock one. 
This commit was SVN r3493.
 2004-11-03 16:52:49 +00:00