1
1

1874 Коммитов

Автор SHA1 Сообщение Дата
Tim Woodall
3b5a9e7377 removed duplicate free
This commit was SVN r5312.
2005-04-13 17:51:43 +00:00
Tim Woodall
55f9f800b1 attach buffer in start and detach when the request completes - this adds some
overhead to the critical path - but ensures buffer reference counts are released
when requests are not released by the application

This commit was SVN r5311.
2005-04-13 17:28:06 +00:00
George Bosilca
682af1ee0f Make sure that the match is called just once after completly reading the header. This bug was raised if we didn't
manage to read any data from the message body, just the header. Then the match was done twice ...

This commit was SVN r5310.
2005-04-13 16:11:20 +00:00
George Bosilca
6c79cc3e04 Now we are able to cancel some requests. Not the sends only the receives if they are not matched.
This commit was SVN r5306.
2005-04-13 15:25:28 +00:00
Tim Woodall
d1327e616f fix proxy
This commit was SVN r5305.
2005-04-13 15:24:14 +00:00
Edgar Gabriel
4ef292717e minor fix in scatterv_inter. We should now pass all ibm inter-collective tests.
This commit was SVN r5302.
2005-04-13 14:34:57 +00:00
Tim Woodall
a2b1f98444 init tag
This commit was SVN r5301.
2005-04-13 14:09:40 +00:00
George Bosilca
6df8cb0f02 First step toward cancelation. We decided not to allow sends to be cancelled. For the others cases:
a completed request cannot be cancelled and a receive where the match was already realized cannot
be cancelled as well.

This commit was SVN r5295.
2005-04-13 05:20:11 +00:00
George Bosilca
86b0630d31 Update the macro used to return registered DMA fragments to the free list.
This commit was SVN r5294.
2005-04-13 05:15:18 +00:00
Brian Barrett
309ff000a6 * add --enable-mca-static to specify components that should be statically
linked into libmpi
* add --enable-mca-direct to specify components that should be directly
  called (instead of going through component structs and the like).  The
  components and component frameworks must explicitly support this.
  Currently, only the TEG PML does so.
* Updated all the calls to the PML to use a macro so that they can either
  be direct called or called through function pointer interfaces (aka
  the component infrastructure)

This commit was SVN r5291.
2005-04-13 03:19:48 +00:00
Josh Hursey
a283072b22 a couple of makefile changes for Windows, just to show how really broken the mca and tools directories are under Windows at the moment
This commit was SVN r5290.
2005-04-13 02:35:52 +00:00
Josh Hursey
b253023dbf A bunch of changes which enable the generation of libmpi under
Windows.
Added some functionality to replicate the waitpid and kill
under Windows.
Also created a ompi_environ.* file which contains the same 
funcationality as environ.*, but we can't use that filename
since it is already defined under Windows.

This commit was SVN r5289.
2005-04-13 02:29:33 +00:00
George Bosilca
3d5e2bc343 A LOT of cleanups and remove all useless code.
Fix a bug (the fragment came from a cache so we should reinitialize the ptl_peer before calling the match or
otherwise the match does not set the peer).
Add more user friendly log messages.

This commit was SVN r5288.
2005-04-12 22:46:48 +00:00
George Bosilca
eeb23c70c4 Send back the flags to the sender.
And keep a local directly to the PTL not to the fragment.

This commit was SVN r5287.
2005-04-12 22:44:47 +00:00
George Bosilca
0b89866fe6 We can initialize everything in this function as GM never use the MATCH type of header.
This commit was SVN r5286.
2005-04-12 22:42:41 +00:00
Brian Barrett
ce07c10a7e * fix error message formatting. resolves bug #1290.
This commit was SVN r5284.
2005-04-12 22:13:55 +00:00
George Bosilca
a9d8044b0a Automatically disable the rendez-vous protocol if the gm_get function is not available.
This commit was SVN r5282.
2005-04-12 21:50:26 +00:00
Jeff Squyres
2d98ac1bf4 Fix for bug 1285: made the code for re-registering a string MCA
parameter be simpler (and correct ;-) ).

This commit was SVN r5281.
2005-04-12 21:47:29 +00:00
Tim Woodall
f2acb6d66f restore subscribe for connect/accept
This commit was SVN r5280.
2005-04-12 21:25:51 +00:00
Jeff Squyres
31768b36de clean up some of the pointer arithmatic. Correct data
types.

This commit was SVN r5279.
2005-04-12 21:17:45 +00:00
Jeff Squyres
3ec8c614f7 Remove some redundant code
This commit was SVN r5278.
2005-04-12 21:11:40 +00:00
Edgar Gabriel
3f32a7b888 remove a comment line :-)
This commit was SVN r5274.
2005-04-12 19:15:25 +00:00
Edgar Gabriel
05624dda0c two minor fixes, reviewed by Tim
This commit was SVN r5273.
2005-04-12 19:10:37 +00:00
Thara Angskun
8a603291bc - define ompi_socklen_t to socket_t (not int) for AIX
- <netinet/tcp.h> (if necessary) should be included after <netinet/in.h> 
i.e. order is important for AIX.

This commit was SVN r5265.
2005-04-11 22:48:50 +00:00
Jeff Squyres
934f9513c0 add volatile declaration to fifo_tmp casts.
This commit was SVN r5263.
2005-04-11 19:27:06 +00:00
George Bosilca
04f8ea3f29 Make it compiling with errors or warnings.
This commit was SVN r5262.
2005-04-11 19:13:37 +00:00
Josh Hursey
092357aa9b Corrected a few unistd lib protection defines. Added a couple of new files to the dist for the windows build
This commit was SVN r5261.
2005-04-11 18:43:57 +00:00
George Bosilca
ff21a25ca9 Do not forget to construct the free_list before calling init.
This commit was SVN r5260.
2005-04-11 18:10:42 +00:00
George Bosilca
c11910c156 Allow GM to receive unexpected messages.
This commit was SVN r5259.
2005-04-11 17:54:53 +00:00
Jeff Squyres
9f286f83f6 Temporarily disable resfile for most developers, since it's causing
some build issues.  Still enabled for its main developer; e-mail to
follow...

This commit was SVN r5257.
2005-04-11 13:02:58 +00:00
Thara Angskun
3a27c53d19 Fixed aio_read/write/suspend for AIX w/o _AIO_AIX_SOURCE
This commit was SVN r5255.
2005-04-11 06:46:46 +00:00
Josh Hursey
9f5dea8dc3 added rmgr and rml to Windows build system
This commit was SVN r5253.
2005-04-11 05:07:57 +00:00
Josh Hursey
830d427a78 Windows build checkpoint. Added some more bits a peices for the Windows compile
This commit was SVN r5251.
2005-04-11 04:47:58 +00:00
Josh Hursey
9c4e9929f8 Fixed some include path problems, and some defines for windows build
This commit was SVN r5250.
2005-04-11 03:43:35 +00:00
George Bosilca
247e271b14 Several bugs have been fixed and the functionality of the GM PTL has been improved:
- Now we support GM 1.x
- We detect correctly the version of GM and activate some defines
- memory leaks fixes (dont release a fragment several times and release all fragments)
- protect the free_list while debug is activated
- change the place where we keep the informations about the local node as well as for the peers.
  It should be a little bit cleaner as we have less copies.
- first step toward supporting OSes without memory registration.
- correctly select the communication protocol depending on the capabilities of the GM library.
  There is still some work to be done as the detection is ... let's say user based at this moment (using
  MCA parameters).

This commit was SVN r5248.
2005-04-10 16:20:04 +00:00
George Bosilca
f854df3ec5 Dont print a message if the GM PTL cannot be initialized.
This commit was SVN r5246.
2005-04-10 00:31:23 +00:00
George Bosilca
1edade9afc If the nodes dont have the same number of interfaces try to match them all. Dont return
unreacheable from the PTL layer. The decision if a node is reacheable or not is taken
at the PML level.

This commit was SVN r5245.
2005-04-10 00:26:35 +00:00
George Bosilca
5ffe7b2697 Add a memory registration layer.
This commit was SVN r5244.
2005-04-09 20:27:42 +00:00
Brian Barrett
8ad4cae815 * some hacks to make progress threads compile. Still don't work..
This commit was SVN r5243.
2005-04-09 19:34:32 +00:00
George Bosilca
6a60aa3c2c Use the GM version as detected on the configure.stub ...
This commit was SVN r5240.
2005-04-09 06:12:49 +00:00
George Bosilca
6b052f54c0 Add the file containing the wrapper around the memory registration. The idea is that on OSes without the
possibility to register memory these functions will return some pre-allocated PTL buffers and pack the
data inside.

This commit was SVN r5239.
2005-04-09 06:10:16 +00:00
George Bosilca
fad48b5fcc Add more communication protocol. Transform the gm_registerPmemory on something more GM version unaware.
This commit was SVN r5238.
2005-04-09 06:07:59 +00:00
George Bosilca
3a1438b838 Add several others configure detection scripts.
This commit was SVN r5237.
2005-04-09 05:52:17 +00:00
George Bosilca
2e4a2084bd Add a new MCA parameter fixing the limit between 2 of the long message protocols (gm_rndv_burst_limit)
This commit was SVN r5236.
2005-04-09 05:51:31 +00:00
George Bosilca
344ee284f2 Add more GM PTL flags
This commit was SVN r5235.
2005-04-09 05:48:55 +00:00
Jeff Squyres
95d41c650b Patch recommended by Greg Watson for compiling on bproc4 machines.
Needs a review by Tim!

This commit was SVN r5233.
2005-04-08 20:02:50 +00:00
Jeff Squyres
8a7a171a44 Fix some bit rot; add a missing #include
This commit was SVN r5232.
2005-04-08 20:01:37 +00:00
George Bosilca
563d4f128d Add a test for broken gm_get from Paul H. Hargrove. Thanks.
This commit was SVN r5217.
2005-04-07 20:19:04 +00:00
Josh Hursey
c4ae276944 rollback ...
This commit was SVN r5215.
2005-04-07 20:01:46 +00:00
Josh Hursey
9068fc0395 rollback ...
This commit was SVN r5214.
2005-04-07 19:57:35 +00:00