ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность
7 772 коммитов 3 Ветки 0 Теги
Граф коммитов

1101 Коммитов

Автор SHA1 Сообщение Дата
George Bosilca
d00e6e29e8 Create a close function for the mpool SM module, in order to allow the cleanup. The 
mca_common_sm_mmap file was left over by the SM mpool, and there was nobody able
to unmap and unlink it.

This commit was SVN r10770.
 2006-07-12 22:12:07 +00:00
Brian Barrett
ca5bd805db * add missing continuation line 
This commit was SVN r10758.
 2006-07-12 14:33:08 +00:00
George Bosilca
fd39203262 As the self proc is marked as local, there will always be at least one local 
proc. Don't create the SM file until we really know there is someone lse on
the same node.

This commit was SVN r10740.
 2006-07-11 17:05:13 +00:00
George Bosilca
40f7d054f2 No more unused variables ... 
This commit was SVN r10735.
 2006-07-11 15:24:57 +00:00
George Bosilca
e2ebd1efcc Protect header file. 
This commit was SVN r10720.
 2006-07-11 05:38:12 +00:00
George Bosilca
ee6fab783d SwitchToThread is not defined by any library. Not even by the kernel32.lib 
as noted in the MSDN documentation. At least not on my WinXP Pro box.

This commit was SVN r10719.
 2006-07-11 05:36:04 +00:00
George Bosilca
14b3f141db Nothing relevant !!! 
This commit was SVN r10711.
 2006-07-11 00:30:26 +00:00
Galen Shipman
68ae99123d fix bsend completion.. 
This commit was SVN r10709.
 2006-07-10 22:27:32 +00:00
Galen Shipman
9a1221bf7d fix buffered sends (don't use blocking sends!) 
removed inaccurate comment.. 

This commit was SVN r10703.
 2006-07-10 16:11:14 +00:00
Andrew Friedley
b7e0484c37 Give up on dat_ep_query() and instead manually send our address information across the wire after connection establishment. 
I've introduced a race condition - seeing occasional LOCAL_LENGTH errors on the receive side.  I think I'm mixing up eager/max somehow - will look at it more on monday.

This commit was SVN r10690.
 2006-07-07 21:48:16 +00:00
Galen Shipman
5085061475 don't call unpack when we received directly into the user buffer.. the 
convertor doesn't handle it properly
continue peeking until we don't get anything else.. 
close the endpoint before closing the library.. 
add a blocking send that uses mx_test .. 

This commit was SVN r10684.
 2006-07-06 19:54:13 +00:00
Brian Barrett
cba9b1e6b7 * the POrtals MTL is now stable enough to not have it ompi ignored 
This commit was SVN r10682.
 2006-07-06 18:26:48 +00:00
Brian Barrett
58ce434292 * remove the broken, defunct portals PML. Not needed anymore, since we can 
do the same basic thing with the MTL design

This commit was SVN r10681.
 2006-07-06 18:24:08 +00:00
George Bosilca
476c9e64df Don't keep multiples copies of the datatype and count. The only one we really need 
is the one provided by the user. For the buffered send the real datatype used
for the communication is always MPI_BYTE and the count can be retrieved from
the req_bytes_packed field. This will decrease the size of the request by
one pointer and one size_t (8 bytes or 16 bytes depending on the architecture).

This commit was SVN r10680.
 2006-07-06 17:58:25 +00:00
Brian Barrett
b7b93e48f5 * can definitely be optimized more, but add code for calling send for MTL 
components that have a blocking send implementation

This commit was SVN r10679.
 2006-07-06 16:37:59 +00:00
Brian Barrett
ef6b7e170f * make mtl datatype wrapper code inline functions 
This commit was SVN r10678.
 2006-07-06 15:58:07 +00:00
Galen Shipman
2217fd4003 reset receive request convertor for persistent requests 
We can always call unpack.. 

This commit was SVN r10677.
 2006-07-06 15:13:26 +00:00
Brian Barrett
ef8c6a249b * Fix up some direct-calling issues for the PML/MTL 
This commit was SVN r10676.
 2006-07-06 15:12:38 +00:00
Brian Barrett
95118f83f6 * complete all outstanding Portals events before shutting down 
* Remove all knowledge of PML requests from the Portals MTL

This commit was SVN r10675.
 2006-07-06 14:33:29 +00:00
Brian Barrett
26eee59032 * turns out that you should only call bsend_request_alloc or 
bsend_request_init, but not both.  Otherwise, you don't free
  some buffer space and end up leaking buffers and ending in
  badness
* since you only call alloc() or init(), but not both, need to 
  restore reference counting in init()

This commit was SVN r10674.
 2006-07-06 14:02:51 +00:00
Gleb Natapov
e05ec69dc4 print "flush error" only once. 
This commit was SVN r10672.
 2006-07-06 08:03:01 +00:00
Gleb Natapov
9b0807e547 Put pending fragment on the right waiting list. 
This commit was SVN r10671.
 2006-07-06 07:51:23 +00:00
George Bosilca
01a59d68da Do not generate the XFER_BEGIN and XFER_END events if the length of 
the data is zero, for both the receives and the sends.

This commit was SVN r10670.
 2006-07-05 23:39:13 +00:00
Brian Barrett
c793ad0a3d unpack the amount received, not the amount we had space to receive. 
This commit was SVN r10669.
 2006-07-05 22:31:29 +00:00
Galen Shipman
c933c0f65f unpack the length actually received, not the length posted.. 
This commit was SVN r10668.
 2006-07-05 22:16:46 +00:00
Brian Barrett
3e29949cc8 * Fix shutdown code in utcp portals code 
* make all sends long sends for now in Portals MTL
* More optimized match check

This commit was SVN r10667.
 2006-07-05 21:46:45 +00:00
Galen Shipman
fe480cd003 change mask bits and don't call convertor if we received directly into the 
user buffer.. 

This commit was SVN r10665.
 2006-07-05 21:10:09 +00:00
George Bosilca
6265625983 Generate the XFER_CONTINUE PERUSE event (or the receive) before unpacking the data. 
This commit was SVN r10663.
 2006-07-05 19:45:00 +00:00
Brian Barrett
4ee4acb6a6 * ignore some Cray-only code when not on the Cray machine 
This commit was SVN r10660.
 2006-07-05 17:16:27 +00:00
Brian Barrett
043153dad3 * fix opal_list_item_t -> ompi_free_list_item_t type change 
This commit was SVN r10659.
 2006-07-05 17:02:16 +00:00
George Bosilca
d2bf3844e9 Include the header file which define opal_output. 
This commit was SVN r10648.
 2006-07-04 06:23:01 +00:00
George Bosilca
2bdb06b549 Force the request to NULL in order to avoid complaints from the compiler. 
This commit was SVN r10647.
 2006-07-04 06:20:13 +00:00
George Bosilca
402a03d229 Add a .h dependency in order to remove a warning when we compile without --enable-debug. 
This commit was SVN r10646.
 2006-07-04 04:53:38 +00:00
George Bosilca
9ac1a6cdb3 Remove the warnings. Now they are ompi_free_list_item not opal_list_item_t. 
This commit was SVN r10645.
 2006-07-04 04:21:16 +00:00
Brian Barrett
7d12f9119a * make sure to include post_configure.sh in the dist tarball, so that 
direct calling the ob1 pml works properly.

This commit was SVN r10644.
 2006-07-04 04:03:58 +00:00
Brian Barrett
47725c9b02 * Add new PML (CM) and network drivers (MTL) for high speed 
interconnects that provide matching logic in the library.
  Currently includes support for MX and some support for
  Portals
* Fix overuse of proc_pml pointer on the ompi_proc structuer, 
  splitting into proc_pml for pml data and proc_bml for
  the BML endpoint data
* bug fixes in bsend init code, which wasn't being used by
  the OB1 or DR PMLs...

This commit was SVN r10642.
 2006-07-04 01:20:20 +00:00
Graham Fagg
f10c21b746 corrected mca param description and algorithm count 
(now to find out why I have disallowed direct calling fo the bm tree)

This commit was SVN r10603.
 2006-06-30 23:22:49 +00:00
Graham Fagg
f64cbbe8f2 ops. some decisions used extent rather than size for decision making 
yes this means it WAS possible for two nodes to choice two different algorithms
(discovered by Doug Gregor and figured out by George)
Also changed some names like size to comsize so we know which sizes we are using where
This should be updated in al versions

This commit was SVN r10601.
 2006-06-30 21:49:04 +00:00
Brian Barrett
df9273587f * romio_cb_write should also be forced to enable when optimizations are 
requested

This commit was SVN r10584.
 2006-06-30 15:06:10 +00:00
Galen Shipman
7e079d20ab fix for stupid casting.. addresses issue on PPC64 where sizes get set 
improperly and badness ensues..

This commit was SVN r10574.
 2006-06-29 21:58:50 +00:00
George Bosilca
7d59a6885b Remove all references to the MRU list. Add back the repost list checks. For some reasons 
it decrease the latency by around 0.3 micro-seconds ...

This commit was SVN r10571.
 2006-06-29 19:25:44 +00:00
George Bosilca
78f0de127d Typo. 
This commit was SVN r10567.
 2006-06-29 15:16:25 +00:00
George Bosilca
4df58b5579 Latency is LATENCY as everybody understand it not some percentage of something. Now, we really 
order the BTL depending on the real latency for the eager protocol. Starting from now, the
latency one can specify for the devices will be in micro-second, while the bandwidth is in Mbs
(as it was before).

This commit was SVN r10566.
 2006-06-29 15:13:58 +00:00
George Bosilca
238147f576 Help the compiler to optimize the code. Now the order in the enum reflect the 
order we use them in the switch.

This commit was SVN r10565.
 2006-06-29 15:10:58 +00:00
George Bosilca
9bf281bca2 Remove the gm_mru_reg list as it is never used. Cleanup the repost logic. Now we repost 
a receive fragment only when we're done with the message from inside and we try to add it
to the list.

This commit was SVN r10564.
 2006-06-29 15:10:11 +00:00
George Bosilca
43b7b17033 Release the memory registration when the descriptors get freed. 
This commit was SVN r10540.
 2006-06-28 15:24:16 +00:00
George Bosilca
d9daa34a6c Set the registration field to NULL when we create a new fragment. 
This commit was SVN r10539.
 2006-06-28 15:23:36 +00:00
Gleb Natapov
c8f75c472a remove modulo op from fast path. Improvement 0.02-0.04ms. 
This commit was SVN r10538.
 2006-06-28 12:00:47 +00:00
Gleb Natapov
e58a89ef3e OMPI_ENABLE_DEBUG is always defined (to 0 or 1). Use #if and nto #ifdef. 
This commit was SVN r10537.
 2006-06-28 11:25:09 +00:00
Gleb Natapov
704a5eb645 Support for LMC (lid mask count) and multiple QPs per port. 
This commit was SVN r10536.
 2006-06-28 07:23:08 +00:00