ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность
10 124 коммитов 3 Ветки 0 Теги
Граф коммитов

1871 Коммитов

Автор SHA1 Сообщение Дата
Jelena Pjesivac-Grbovic
d677db9b5f cleaning up alltoall implementation: 
- removing MPI_* calls from bruck implementation
- simplifying 2 process case
- identation, etc.

This commit was SVN r15301.
 2007-07-07 01:06:19 +00:00
Rainer Keller
cff1b6a71b - PERUSE_COMM_REQ_XFER_BEGIN should be emited for first fragment 
of larger message as well.

This commit was SVN r15299.
 2007-07-06 15:02:36 +00:00
Andrew Friedley
77038b65a8 Bring the UD BTL over to the trunk, named 'ofud'. 
This commit was SVN r15298.
 2007-07-05 23:42:54 +00:00
Brian Barrett
872623a527 Fix situation where we were not propogating error codes from ROMIO 
into non-blocking request status fields, so it was never being
relayed to the user

This commit was SVN r15296.
 2007-07-05 22:30:42 +00:00
Brian Barrett
25e52238ab add ability to buffer put/accumulate messages during an epoch 
This commit was SVN r15295.
 2007-07-05 21:40:06 +00:00
Brian Barrett
5bbee1482e make debugging output slightly more useful 
This commit was SVN r15294.
 2007-07-05 21:01:32 +00:00
Jelena Pjesivac-Grbovic
483222085e Fixing compiler warnings. 
In gather, the ptmp += incr is irrelevant, since ptmp is set within the loop.

This commit was SVN r15293.
 2007-07-05 20:40:50 +00:00
Brian Barrett
6c9de88d13 while the BTL semantics are sorted out, wait for completion of all rdma events before starting the ack 
This commit was SVN r15292.
 2007-07-05 16:50:05 +00:00
Sven Stork
21f12f29f8 - fix a sm bug that causes segfaults in the case of threaded builds. 
The problem is that in the case of threaded builds for every fifo
  a head and tail lock will be allocated inside the shared memory
  segment and the ptr is stored inside the fifo. In the case that the sm backend
  file will be mapped in all processes at the same address (mostly the
  case for non-thread builds) this is fine, but in the cases when the
  processes map the file at different addresses this addresses cause big
  trouble in other processes than the one that allocted the locks. 
  Therefore the send lock addresses have to be recalculated to match
  the local mapping of the processes that use them.

This commit was SVN r15291.
 2007-07-05 14:26:32 +00:00
Brian Barrett
74008aac53 Support real RDMA operations for networks that support it 
This commit was SVN r15288.
 2007-07-05 03:32:32 +00:00
Brian Barrett
41afd4ebee Clean up the MX configure test a bit. Use AC macros instead of hand 
writing them.  Better tests, less code, and caching.  Update the code
to match changes in configure defines.

This commit was SVN r15287.
 2007-07-04 22:07:30 +00:00
Jelena Pjesivac-Grbovic
3b0a52a104 adding tuned allgatherv implementation using bruck, ring, and neighbor-exchange algorithms. 
The implementations passed intel and imb tests up to 40 processes.

This commit was SVN r15280.
 2007-07-03 23:33:12 +00:00
Brian Barrett
f5c721d11c Wire up all the RDMA-capable BTLs. Still no RDMA communication, but the 
datastructures are finally all there

This commit was SVN r15271.
 2007-07-02 22:22:59 +00:00
George Bosilca
951e4929b9 Usually it's unlikely to have additional fragments. 
This commit was SVN r15253.
 2007-07-01 16:19:53 +00:00
George Bosilca
c435094639 Only trigger the PERUSE_COMM_REQ_XFER_BEGIN event on the initial fragment. 
This commit was SVN r15252.
 2007-07-01 16:19:13 +00:00
George Bosilca
60319f99ac Make sure in case of error what we return is clean (set to NULL). 
This commit was SVN r15251.
 2007-07-01 16:17:43 +00:00
George Bosilca
11656e20aa Remove few warnings. 
This commit was SVN r15250.
 2007-07-01 16:16:05 +00:00
Gleb Natapov
77e54ebc7e Schedule RDMA op on the last BTL that got completion. 
This commit was SVN r15249.
 2007-07-01 11:35:55 +00:00
Gleb Natapov
54b40aef91 Schedule SEND traffic of pipeline protocol between BTLs in accordance with 
relative bandwidths of each BTL. Precalculate what part of a message should
be send via each BTL in advance instead of doing it during scheduling.

This commit was SVN r15248.
 2007-07-01 11:34:23 +00:00
Gleb Natapov
e74aa6b295 Schedule RDMA traffic between BTLs in accordance with relative bandwidths of 
each BTL. Precalculate what part of a message should be send via each BTL in
advance instead of doing it during scheduling.

This commit was SVN r15247.
 2007-07-01 11:31:26 +00:00
George Bosilca
dfa5ae34e1 Per a discussion with Kees Verstoep and Reese Faucette add one more 
argument to the query for the line speed. This function is still not
documented, and it really look strange that we have to respecify the
nic_id (it's already attached to the endpoint).

This commit was SVN r15241.
 2007-06-28 20:58:00 +00:00
Jelena Pjesivac-Grbovic
d55b415bb0 fixing typo 
This commit was SVN r15240.
 2007-06-28 20:56:55 +00:00
Brian Barrett
f8fb1e9720 Fix some compile failures on Solaris 9 because it doesn't have V6ONLY. 
This commit was SVN r15237.
 2007-06-28 18:52:15 +00:00
George Bosilca
aec0b00f29 Get some hints about the network and propagate them to the upper level. 
This commit was SVN r15236.
 2007-06-28 18:51:48 +00:00
George Bosilca
98142263c6 These functions are potentially shared between multiple components so they should 
be visible.

This commit was SVN r15235.
 2007-06-28 18:50:33 +00:00
Gleb Natapov
1c7141df4d Remove unused struct. 
This commit was SVN r15228.
 2007-06-28 11:58:16 +00:00
Jelena Pjesivac-Grbovic
8fc8b44d11 Modifying reduce decision function for large, single element reduces (again). 
Binary algorithm without segmentation tends to outperform binomial algorithm 
in this case.

This commit was SVN r15226.
 2007-06-27 22:01:56 +00:00
Jelena Pjesivac-Grbovic
0ecef1750d Modifying the default reduce decision function to use binomial algorithm 
for single-element reduce (segmented algorithms make no sense in this case
and can cause performance degradation). 

This commit was SVN r15209.
 2007-06-26 20:14:03 +00:00
Jelena Pjesivac-Grbovic
567b40b9a9 Modifying the default broadcast decision function to use binomial algorithm 
for single-element broadcasts (segmented algorithms make no sense in this case
and can cause performance degradation).

This commit was SVN r15208.
 2007-06-26 20:08:31 +00:00
Josh Hursey
acae12d0bb Fix warning: 
stderr -> fileno(stderr)

This commit was SVN r15207.
 2007-06-26 19:28:40 +00:00
Josh Hursey
5199f4123d Add 2 new MCA parameters to set the size of the expected and unexpected 
queues.

This commit was SVN r15206.
 2007-06-26 17:31:43 +00:00
Rich Graham
aa2ffcfcd8 add some output before abort() is called. 
This commit was SVN r15204.
 2007-06-26 15:57:47 +00:00
Sven Stork
428f697542 - addition to r15198. Update also the prepare destintation functions. 
This commit was SVN r15199.

The following SVN revision numbers were found above:
  r15198 --> open-mpi/ompi@f63dd902cb
 2007-06-26 12:07:30 +00:00
Sven Stork
f63dd902cb - bring the order changes of r14768 also to the mvapi btl 
This commit was SVN r15198.

The following SVN revision numbers were found above:
  r14768 --> open-mpi/ompi@3401bd2b07
 2007-06-26 09:34:44 +00:00
Brian Barrett
e279192865 ug - fix some dumb copy-n-paste errors 
This commit was SVN r15188.
 2007-06-25 01:59:34 +00:00
Brian Barrett
42b2c4e1df * RELEASE not DETRUCT things created with OBJ_NEW to fix a memory leak 
* Fix potential race condition with starting a new lock epoch if we
   were releasing a lock
 * Increment the shared counter if we start a shared lock session during
   the unlock code

This commit was SVN r15186.
 2007-06-24 23:30:10 +00:00
Brian Barrett
2ed0548da8 * No need for waiting until exposure epochs are over in order to complete 
a WIN_FREE
  * Fix race condition in threaded builds with pending unlocks and
    finishing an epoch
  * Fix memory leak due to use of OBJ_DESTRUCT instead of OBJ_RELEASE
  * Fix race condition between releasing multiple shared locks and
    starting a new lock
  * Need to incremement the shared count if starting a new shared
    lock once an exclusive lock finishes

This commit was SVN r15185.
 2007-06-24 22:36:00 +00:00
Brian Barrett
5528e0ca60 Properly initialize variable for threaded case 
This commit was SVN r15174.
 2007-06-22 15:29:06 +00:00
Brian Barrett
5f16251808 revert r15167. I don't know what I was thinking, but it was most definitely 
"not right".

This commit was SVN r15172.

The following SVN revision numbers were found above:
  r15167 --> open-mpi/ompi@faa401dc47
 2007-06-22 15:25:39 +00:00
Brian Barrett
8031f6561e Make a bunch of debugging calls use the macro version 
This commit was SVN r15170.
 2007-06-21 22:24:40 +00:00
Brian Barrett
80c50120ad debugging output should be macro version 
This commit was SVN r15168.
 2007-06-21 22:09:37 +00:00
Brian Barrett
faa401dc47 * Need to OBJ_RELEASE, not OBJ_DESTRUCT things that were created with 
OBJ_NEW
  * Need to single when the passive unlock has left an expose epoch for
    the win_free case
  * Clean up some debugging output
  * fix missing variable initialization

This commit was SVN r15167.
 2007-06-21 22:08:30 +00:00
Jeff Squyres
022bd30558 Back out r15158 because it apparently breaks with recent versions of 
flex (which, incidentally, emit ''more'' warnings than earlier
versions).  Grumble.

This commit was SVN r15166.

The following SVN revision numbers were found above:
  r15158 --> open-mpi/ompi@57d09c10f7
 2007-06-21 21:14:10 +00:00
Jelena Pjesivac-Grbovic
3740640711 Modifying MPI_Gather in tuned module: 
- adding linear algorithm with synchronization for gather.
  This algorithm prevents congestion at root process, but introduces 
  synchronization (serializes non-root processes, but allows messages 
  to arrive from two processes at the same time).  
  It performed better than binomial and linear algorithms for large message, 
  and intermediate and large communicator sizes.
- Updating MPI_Gather decision function to reflect performance results
  from MX.  I will perform more measurements though - so this one can 
  change.

This commit was SVN r15165.
 2007-06-21 20:00:36 +00:00
Jeff Squyres
57d09c10f7 Avoid some compiler warnings that come up ''every day'' in MTT (and 
have been for eons): make a symbol be used in a dumb but harmless way.

This commit was SVN r15158.
 2007-06-21 15:42:06 +00:00
Gleb Natapov
b88b7dedfe Rename btl_rdma_offset to btl_pipeline_send_length. 
This commit was SVN r15153.
 2007-06-21 07:12:40 +00:00
Jeff Squyres
84487f5c4b Update and correct the help messages for the generic BTL MCA 
parameters.  Hopefully, they now make more sense to the mostly naieve
user...

This commit was SVN r15147.
 2007-06-20 16:37:50 +00:00
Jeff Squyres
930a9b7682 Make the help messages for if_include/if_exclude a little better. 
This commit was SVN r15134.
 2007-06-19 13:38:58 +00:00
Josh Hursey
7fd1805e97 Fix a couple of compile warnings that Tim P brought to by attention. 
This commit was SVN r15132.
 2007-06-19 00:46:16 +00:00
Gleb Natapov
643037907f Convert all #ifdef OMPI_ENABLE_DEBUG to #if. 
This commit was SVN r15117.
 2007-06-17 07:14:47 +00:00