Tim Woodall
b65dc08ab1
counters need to be signed as we check for <0
...
This commit was SVN r7155.
2005-09-02 18:26:07 +00:00
Tim Woodall
dfe52fceef
minor changes to thread locking
...
This commit was SVN r7154.
2005-09-02 16:27:01 +00:00
Galen Shipman
589b1b8b5a
Additional changes to add_proc and tokens
...
This commit was SVN r7152.
2005-09-02 15:18:36 +00:00
Galen Shipman
a7a4da4502
Scale the SRQ based on the log base 2 of the number of peers,
...
this assumes that the peers have all been added via add_procs up front.
Bad things will happen if add_procs is called again later on a new set of
procs to fix this we need to modify the srq which may wreck things.. looking
into this deeper..
This commit was SVN r7142.
2005-09-02 04:06:51 +00:00
Galen Shipman
c8a23106c0
More fixes for sq tokens,
...
Additional work on multi-rail support.
This commit was SVN r7139.
2005-09-02 03:04:28 +00:00
Tim Woodall
636ab23fdb
atomic increment/test
...
This commit was SVN r7130.
2005-09-01 15:09:50 +00:00
Jeff Squyres
3962c53e2e
- Add to AM_CPPFLAGS $(OPAL_LTDL_CPPFLAGS) where necessary in order to
...
add a -I to find the included ltdl.h (vs. a system-installed ltdl.h)
- Clean up kruft in a bunch of Makefile.am's to remove now-unnecessary
AM_CPPFLAGS settings to get static-components.h for each framework
- Move the component_repository API functions out of opal/mca/base/base.h
and into opal/mca/base/mca_base_component_repository.h in order to
decrease unnecessary dependencies (e.g., before this, almost
everything in the tree depended on ltdl.h, which is unnecessary --
only a small number of files really need ltdl.h)
This commit was SVN r7127.
2005-09-01 12:16:36 +00:00
Galen Shipman
29f7b4deda
Changed send tokens to both send/rdma tokens for both low and high priority
...
queue pairs. Tested on intel p2p with 16 procs - Passed.
This commit was SVN r7119.
2005-09-01 02:41:44 +00:00
Galen Shipman
c7e9563377
Added sender side per qp send tokens to limit the number of outstanding
...
sends.
This commit was SVN r7112.
2005-08-31 20:28:42 +00:00
Galen Shipman
09873f299f
Fixed a race in connection establishment..
...
This commit was SVN r7110.
2005-08-31 19:43:22 +00:00
Galen Shipman
00e0ff729d
intialize free list to rr_buf_max, report async errors to user.
...
This commit was SVN r7095.
2005-08-30 16:44:38 +00:00
Tim Woodall
d34e299829
correctly decrement progress_event if tcp is not being
...
used so that tcp doesn't impact progress loop
This commit was SVN r7078.
2005-08-29 17:29:58 +00:00
Brian Barrett
173e062fbb
* Spell LIBS as LIBS not LIBX ;)
...
This commit was SVN r7069.
2005-08-27 17:38:50 +00:00
Tim Woodall
5ed6f2c474
change flag to a sensible value
...
This commit was SVN r7056.
2005-08-26 20:21:07 +00:00
Tim Woodall
d57f3e1662
cleanup - handle request/prepare of zero bytes as special case
...
This commit was SVN r7055.
2005-08-26 20:19:11 +00:00
Galen Shipman
56f722c6c1
Removed all references to the old common/vapi stuff.
...
This commit was SVN r7029.
2005-08-25 15:04:22 +00:00
Galen Shipman
8c85aaf85a
tell the user no mvapi modules are found..
...
This commit was SVN r7016.
2005-08-24 21:59:55 +00:00
Brian Barrett
1241645166
* don't ignore Portals BTL anymore
...
This commit was SVN r6971.
2005-08-22 03:52:07 +00:00
Tim Woodall
205af3af0a
correct segment address
...
This commit was SVN r6942.
2005-08-19 20:20:27 +00:00
Galen Shipman
afdfa70f73
Added support for openib RDMA READ.. note that performance is currently an
...
issue so PUT is default.. We are determining if this is an openib issue or a
btl issue as we have seen performance increases on mvapi.
This commit was SVN r6928.
2005-08-18 17:08:27 +00:00
Galen Shipman
589d53e828
only poll high priority qp 1 per progress.
...
This commit was SVN r6918.
2005-08-17 22:52:56 +00:00
Tim Woodall
f274f524ab
- added get based protocol (if supported by btl) for pre-registered memory
...
- removed 8 bytes from the majority of the pml headers
This commit was SVN r6916.
2005-08-17 18:23:38 +00:00
Galen Shipman
ee6999fa90
typo in threaded build..
...
This commit was SVN r6898.
2005-08-16 13:22:08 +00:00
Tim Woodall
9a094ee3b4
- return corrected size
...
- set send inplace for eager send
This commit was SVN r6891.
2005-08-15 21:30:47 +00:00
Galen Shipman
f248db3789
misc fixes, changes to support multiple mvapi btl's
...
This commit was SVN r6890.
2005-08-15 19:39:56 +00:00
Brian Barrett
e5eba51e9f
* update rdma min size to be 64K. Seems to give best performance on RS.
...
This commit was SVN r6880.
2005-08-15 13:30:35 +00:00
Jeff Squyres
c465eb8567
Rename opal/threads/thread.h -> opal/threads/threads.h to avoid a
...
naming conflict with Solaris' <thread.h>
This commit was SVN r6879.
2005-08-15 11:02:01 +00:00
Brian Barrett
b9452e5afe
* finish switch to sending in place
...
This commit was SVN r6876.
2005-08-15 02:31:45 +00:00
Galen Shipman
8e1e2eec3d
Misc fixes for threaded builds..
...
This commit was SVN r6874.
2005-08-14 19:03:09 +00:00
Brian Barrett
f68ede1c93
* turn off debugging in the bowels of the Portals reference implementation
...
This commit was SVN r6864.
2005-08-14 02:05:23 +00:00
Brian Barrett
51531af9df
* implement Portals btl get
...
This commit was SVN r6862.
2005-08-13 20:55:29 +00:00
Brian Barrett
c83cb66bf2
* Portals can send "in place"
...
This commit was SVN r6860.
2005-08-13 19:43:08 +00:00
Jeff Squyres
cf16a521c8
Ensure to get ompi/include/constants.h
...
This commit was SVN r6845.
2005-08-12 21:42:07 +00:00
Tim Woodall
faf146ec4c
correct address count
...
This commit was SVN r6837.
2005-08-12 19:01:04 +00:00
Tim Woodall
5558c014b9
default TCP to only be used if self/sm/gm/mvapi.... are not available
...
This commit was SVN r6832.
2005-08-12 16:56:46 +00:00
Brian Barrett
2c44b2398d
* remove the src/ directory for Portals BTL to bring it in line with the
...
other BTLs
This commit was SVN r6820.
2005-08-12 14:33:45 +00:00
Galen Shipman
b01ebf45c9
Fixed build error related to direct call (bml_direct_call.h). Misc bug fixes
...
and compiler warning issues. Fixed threaded build issue.
This commit was SVN r6819.
2005-08-12 14:08:40 +00:00
Galen Shipman
c3c83aa3e1
BML (BTL Managment Layer). Allows BTL's to be used outside of the PML. See
...
bml.h and PML-OB1 for usage.
This commit was SVN r6815.
2005-08-12 02:41:14 +00:00
Brian Barrett
d9e5e3343d
* random code cleanup
...
This commit was SVN r6797.
2005-08-10 15:26:15 +00:00
Brian Barrett
d4aa9c702c
* remove unused files
...
This commit was SVN r6796.
2005-08-10 15:11:05 +00:00
Tim Woodall
d458daf437
support for priority flag
...
This commit was SVN r6795.
2005-08-10 14:32:10 +00:00
Galen Shipman
73757b300c
Added BTL_VERBOSE and OMPI_MCA_btl_base_debug , if set to 1 DEBUG output if
...
set to 2 VERBOSE output..
This commit was SVN r6783.
2005-08-09 17:49:39 +00:00
Tim Woodall
9b21413fe2
corrected newline
...
This commit was SVN r6781.
2005-08-09 16:24:48 +00:00
Tim Woodall
b4caa9f9f1
added verbose macro
...
This commit was SVN r6780.
2005-08-09 16:22:55 +00:00
Brian Barrett
88d316d443
* turn RDMA back on and wonder how I ever committed a patch turning it off...
...
This commit was SVN r6776.
2005-08-09 14:11:56 +00:00
Tim Woodall
078836c5b9
added put support for zero copy operation
...
This commit was SVN r6775.
2005-08-09 14:10:17 +00:00
Jeff Squyres
ba31fbf132
A better solution than r6672. If the caller passes in a data segment
...
alignment of 0, then assume there will be no data segment and don't do
the checks to see if it will be beyond the end of the file.
This commit was SVN r6773.
The following SVN revision numbers were found above:
r6672 --> open-mpi/ompi@8b56769307
2005-08-08 21:38:27 +00:00
Jeff Squyres
1c5382deac
- Fix a minor problem in alignment logic in sm common component
...
- Adjust btl sm to allocate just a few bytes extra to allow the common
sm component to assume that there will be a data segment (even though
the sm btl doesn't use the data segment in that portion of code)
This commit was SVN r6772.
2005-08-08 21:29:05 +00:00
Brian Barrett
6e2b07db91
* update Red Storm compat code to match today's changes
...
This commit was SVN r6771.
2005-08-08 21:22:15 +00:00
Brian Barrett
694bbc158f
* Set max tag for BTLs to 255 not 256
...
* Major rework of Portals to better match Red Storm and hopefully get
better performance:
- Always assume there is only one module (since there are no machines
on the planet with more than one Portals interface)
- make progress all one function rather than dispatching to other
functions and dispatch on event type, not comm type
- remove polling of unneeded events
This commit was SVN r6769.
2005-08-08 20:56:26 +00:00