ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность
4 646 коммитов 3 Ветки 0 Теги 130 MiB
098cc8cf3a
Граф коммитов

4646 Коммитов

This Branch
This Branch
Все ветки
Автор SHA1 Сообщение Дата
Ralph Castain
098cc8cf3a Bring the rest of the notification modes online. Update the unit test to cover notify-on-change. 
This commit was SVN r6043.
 2005-06-13 14:37:02 +00:00
Galen Shipman
e580c98365 flag for rdma/directed send in IB is now a mca param 
clear the descriptors on completion.. 

This commit was SVN r6042.
 2005-06-13 14:36:09 +00:00
Ralph Castain
25f0075ac3 Believe this should fix Friday's concerns over callback stacks. The registry was working just fine - the problem was that the modex exchange was asking to be notified any time an entry was added or changed, which the registry dutifully did. However, the initial "put" of information takes place during the compound command portion of mpi_init, during which time we do not allow messages to come back to the process. 
All that was required was to change the subscription request in module_exchange to specify "don't start notifying me until after the trigger". The trigger is set to fire when all processes have completed their initial stage_gate, so that means the process will receive notifications only of any changes from the base starting point.

Specifically:

1. I changed the "don't start" flag to ORTE_GPR_NOTIFY_STARTS_AFTER_TRIG to make it more obvious as to its function

2. Added that flag to the module_exchange subscription, and changed the name elsewhere it was used (I had already fixed this problem for the OOB)

3. Found an unrelated compiler warning of an unused variable in ns_proxy, so I fixed it too.

This commit was SVN r6041.
 2005-06-13 13:30:12 +00:00
Galen Shipman
66e0fb28dc bugfix.. 
This commit was SVN r6040.
 2005-06-12 22:55:00 +00:00
Galen Shipman
48677e4489 Changed recv to check des_dst instead od des_src 
other misc fixes. 

This commit was SVN r6039.
 2005-06-11 00:44:45 +00:00
Jeff Squyres
6477eb3c14 Need to have a virtual destructor since we have one virtual function 
(g++ 4.0 complains).  This function was commented out from many years
ago in the "bad old days" of new-not-conformant compilers.

This commit was SVN r6038.
 2005-06-10 21:02:52 +00:00
Galen Shipman
b3c8d47f62 More rdma work.. 
This commit was SVN r6036.
 2005-06-10 20:40:46 +00:00
Galen Shipman
215ba87116 Aligned the send and receive descriptors 
This commit was SVN r6034.
 2005-06-10 16:46:22 +00:00
Galen Shipman
0bdabafd9b Added RDMA interfaces to ib bmi. 
This commit was SVN r6033.
 2005-06-10 16:02:34 +00:00
Tim Woodall
dea537c9f7 only use rdma protocol for contigous data 
This commit was SVN r6032.
 2005-06-10 14:09:34 +00:00
Jeff Squyres
a1dd4e3173 Be a bit more social while spinning during init to setup the shared 
memory.

This commit was SVN r6031.
 2005-06-10 13:14:32 +00:00
Thara Angskun
5a7b0b598d prevent compilation error when compiled with xlc. 
This commit was SVN r6029.
 2005-06-10 08:06:07 +00:00
Jeff Squyres
9014ab2e86 mpio.h was incorrectly listed twice, and was incorrectly included in 
the distribution tarball (mpio.h is automatically generated and should
not be in the tarball).

This commit was SVN r6028.
 2005-06-10 04:20:57 +00:00
Ralph Castain
5a2c6a0b87 Fix the trunk so it runs again: 
1. Tell the oob subscription not to report on changes until AFTER the trigger fires (signifying that all procs have started)

2. Make sure tokens returned to the subscriber are from the container that housed the data being returned (previous mistake - tokens were from subscription, not from container)

Tested on the Mac running up to 10 processes

This commit was SVN r6025.
 2005-06-10 03:35:36 +00:00
Galen Shipman
ec1c1bae13 Fixed up sending large fragments and preposting of receive descriptors.. 
This commit was SVN r6023.
 2005-06-10 00:01:05 +00:00
George Bosilca
700ec8f999 Add one arguement to the memory allocation function. 
This commit was SVN r6022.
 2005-06-09 22:26:22 +00:00
George Bosilca
f11ec2560a The user provided memory allocation function take one more argument a void*. This can be 
set by the caller via the ompi_convertor_personalize call.

This commit was SVN r6021.
 2005-06-09 22:25:31 +00:00
Ralph Castain
1c57ae20b0 Checkpoint the notifier work - notify when something is added now works, need to simply turn on the other checks. 
Existing code shouldn't see any impacts. Tested on up to 125 processes.

This commit was SVN r6020.
 2005-06-09 20:37:25 +00:00
Tim Woodall
c848b9fb90 rollback change 
This commit was SVN r6019.
 2005-06-09 20:25:04 +00:00
Tim Woodall
3e7ffb6399 first cut at rdma protocol 
This commit was SVN r6018.
 2005-06-09 20:16:33 +00:00
George Bosilca
95cd062784 The convertor.h should be declared in the Makefile.am in order to get into the dist. 
This commit was SVN r6017.
 2005-06-09 19:18:04 +00:00
George Bosilca
3517c87fe3 UNIQ use now the new datatype engine. 
This commit was SVN r6016.
 2005-06-09 17:50:34 +00:00
George Bosilca
66e15238a6 The stack should contain some specific values not only ZEROS (get the code from the beta). 
This commit was SVN r6015.
 2005-06-09 17:13:34 +00:00
Galen Shipman
a8c48ac12c Rewrote qp creation/init 
Rewrote cq creation/init 
Removed all static constants for vapi init and added mca params 
Finished rewrite of component and module init 

This commit was SVN r6014.
 2005-06-09 16:36:51 +00:00
Tim Woodall
c930025531 move release of node out of claim_slots - as the upper level 
routine attempted to use the node after it had been released

This commit was SVN r6013.
 2005-06-09 16:00:41 +00:00
Tim Woodall
49563d411d file descriptor could be zero --- 
This commit was SVN r6012.
 2005-06-09 15:39:51 +00:00
Ralph Castain
51380eba13 Checkpoint the continuing re-enablement of the notifiers. 
Also added a check to protect the callback system from an error being seen by Tim P. - should help with debugging.

This commit was SVN r6010.
 2005-06-09 13:35:35 +00:00
George Bosilca
4d1b9d3f1d Convert to the new ddt engine. As the send and receive convertors are already created outside the PTL is useless to have an internal one. 
This commit was SVN r6009.
 2005-06-09 05:18:43 +00:00
George Bosilca
1d69091231 Set the stack to zero before trying to optimize the datatype. 
This commit was SVN r6008.
 2005-06-09 05:17:24 +00:00
Galen Shipman
b0eb765f22 - ib bmi now creates a seperate module for each ib port, instead of simply 
- using port 1 while ignoring others 
- added 2 free lists for varying size frags 
- query queue pair for max inline send for small fragments 

This commit was SVN r6007.
 2005-06-09 04:37:37 +00:00
Tim Woodall
98c0861d78 stub for missing function 
This commit was SVN r6006.
 2005-06-09 03:34:33 +00:00
Tim Woodall
b73fb5dcf2 rdma protocol - a work in progress 
This commit was SVN r6005.
 2005-06-09 03:11:51 +00:00
Brian Barrett
ce553cfec8 * technically, there should be a ; after the lock directive. GNU assembler 
doesn't care, but Apple's does
* Add x86 OS to the asm list

This commit was SVN r6004.
 2005-06-09 01:58:20 +00:00
Jeff Squyres
d582265c9f More minor doc updates. 
This commit was SVN r6001.
 2005-06-09 01:18:46 +00:00
Jeff Squyres
b38ea5eecb Doc updates. 
This commit was SVN r5999.
 2005-06-09 01:12:54 +00:00
George Bosilca
6886716763 Update the the new datatype engine. 
This commit was SVN r5998.
 2005-06-08 20:47:48 +00:00
Tim Woodall
567abb4a63 set segment length in bmi_prepare_src 
This commit was SVN r5997.
 2005-06-08 20:38:03 +00:00
Tim Woodall
fc9a84ec21 initial changes for rdma protocol 
This commit was SVN r5996.
 2005-06-08 20:37:19 +00:00
Ralph Castain
ba7673a83f Checkpoint the first step in re-enabling the notification for subscriptions that monitor value changes. Added a new array that stores the actions each time the registry is called via a function that modifies its values. Updated the dump function to output the action records. 
This commit was SVN r5995.
 2005-06-08 19:40:38 +00:00
George Bosilca
63b382ec6c First step toward the new datatype engine. 
This commit was SVN r5994.
 2005-06-08 19:15:26 +00:00
George Bosilca
b615be73d4 Update the MPI level to take advantage of the new datatype engine. 
This commit was SVN r5993.
 2005-06-08 19:14:41 +00:00
George Bosilca
5c3aa914b0 We depend now on the convertor.h not on the datatype.h 
This commit was SVN r5992.
 2005-06-08 19:14:07 +00:00
George Bosilca
de56467521 The max_size should be a size_t. I add myself to the ompi_unignore list :) 
This commit was SVN r5991.
 2005-06-08 19:13:28 +00:00
George Bosilca
7a41a6cd7a Now we have to include the convertor.h 
This commit was SVN r5990.
 2005-06-08 19:12:42 +00:00
George Bosilca
6d4624e5d4 Use the new datatype engine and the convertor attached to the requests (send and receive). 
This commit was SVN r5989.
 2005-06-08 19:12:14 +00:00
George Bosilca
8cc028fee9 Use the new datatype engine and the convertor attached to the requests (send and receive). 
This commit was SVN r5988.
 2005-06-08 19:11:32 +00:00
George Bosilca
a9f342ac7e Adapt to the new datatype engine. There are 2 big differences with the previous version: 
1. for the send operation: the convertor attached to the request is initialized by the
   _INIT macro and it will be used as is by the PTLs for the first fragment. For the
   others fragments it will be cloned by the PTLs.
2. for the receive operation: I create a macro MATCHED which create a convertor (attached
   to the receive PML request). The PTLs can use this convertor for the first fragment,
   and clone it forr the others fragments.

Playing with positions ... The position argument change it's type and prototype. Now it's
a size_t to be consistent with the rest of thee PML/PTL and it's a pointer to a size_t.
I need this pointer to correct the value of the position as some of the convertors are
not allowed to stop in the middle of a basic datatype (cannnot copy 2 bytes from an int).

This commit was SVN r5987.
 2005-06-08 19:10:15 +00:00
George Bosilca
3f8d210129 Add a newline at the end of the file. 
This commit was SVN r5986.
 2005-06-08 19:03:29 +00:00
George Bosilca
8e6223cb36 New datatype engine. Most of the work was on the initialization. Now there are several 
functions to create/clone a convertor, with the goal of reducing the time spend on
the convertor related functions, thus reducing the latency.

This commit was SVN r5985.
 2005-06-08 19:02:16 +00:00
George Bosilca
993982f748 Cleanup. Most of the convertor related functions moved from all the datatype file into these 2. 
This is the first commit for the new datatype engine. Please hold your updates until I send an email to the mailing list.

This commit was SVN r5984.
 2005-06-08 19:00:16 +00:00