charm.git
11 years agoAdding hook when resuming a normal thread of execution (i.e not the main thread or...
Filippo Gioachin [Thu, 4 Feb 2010 07:22:25 +0000 (01:22 -0600)]
Adding hook when resuming a normal thread of execution (i.e not the main thread or a scheduling thread).
This allows Converse or any higher language to decide if the thread switch should really be performed or not.

This has been added for the benefit of record/replay, so the replaying mechanism can reorder these messages as well.

11 years agoNo need to send it the long way, just put it at the beginning of the scheduler's...
Filippo Gioachin [Thu, 4 Feb 2010 07:21:13 +0000 (01:21 -0600)]
No need to send it the long way, just put it at the beginning of the scheduler's local queue (we are waiting for this messages, so might as well place it as the first pick)

11 years agoAdding a new field to CthThreadToken data structure. This is useful for record/replay...
Filippo Gioachin [Thu, 4 Feb 2010 07:15:39 +0000 (01:15 -0600)]
Adding a new field to CthThreadToken data structure. This is useful for record/replay to keep track of when threads are switched (this way each thread is associated with a unique ID at creation time)

11 years agoFor some reason it complains to me about pointer conversion. Since really there shoul...
Filippo Gioachin [Thu, 4 Feb 2010 07:13:57 +0000 (01:13 -0600)]
For some reason it complains to me about pointer conversion. Since really there should be a cast from void* to Queue, I'm adding it

11 years agoIntercepting all calls to Cmi* message sending functions. Before calling the underlying
Filippo Gioachin [Wed, 3 Feb 2010 03:18:26 +0000 (21:18 -0600)]
Intercepting all calls to Cmi* message sending functions. Before calling the underlying
send routines, check if we are in a state for which the call should be skipped. Right
now the only condition is message replay (as part of Record/Replay). Later other conditions
could be added.

11 years agoChanging order in which message watchers are parsed: must add +replay after
Filippo Gioachin [Wed, 3 Feb 2010 03:15:38 +0000 (21:15 -0600)]
Changing order in which message watchers are parsed: must add +replay after
+record-detail, otherwise we'll end up reading the wrong log file (since
+record-detail changes the MyPe().
Also, the decision to process a message or not is a joint decision of all the message
watchers: if one of them says no (CmiFalse), then the message is not processed.

11 years agoFixing CkMessageDetailedReplay:
Filippo Gioachin [Wed, 3 Feb 2010 03:13:38 +0000 (21:13 -0600)]
Fixing CkMessageDetailedReplay:
- handling end-of-file (return NULL)
- missing return statement

11 years agoAdded new flag for marking when we are replaying
Filippo Gioachin [Wed, 3 Feb 2010 00:08:56 +0000 (18:08 -0600)]
Added new flag for marking when we are replaying

11 years agoError in definition of reset function
Filippo Gioachin [Wed, 3 Feb 2010 00:06:19 +0000 (18:06 -0600)]
Error in definition of reset function

11 years agoMerge branch 'charm' into development
Filippo Gioachin [Tue, 2 Feb 2010 04:01:06 +0000 (22:01 -0600)]
Merge branch 'charm' into development

11 years agoDecrementing refernces to objects otherwise leaking
Filippo Gioachin [Tue, 2 Feb 2010 03:05:53 +0000 (21:05 -0600)]
Decrementing refernces to objects otherwise leaking

11 years agoFixing a race condition between array manager and location manager.
Filippo Gioachin [Tue, 2 Feb 2010 02:08:02 +0000 (20:08 -0600)]
Fixing a race condition between array manager and location manager.
Given that message ordring is not guaranteed in Charm, the array manager cannot
rely on the fact that the location manager (whose message is sent first) has
already been created when it (array manager) is constructed.
Thus, buffer the initMsg and all the broadcasts until the location manager is
created. When created, the location manager notifies all waiting array managers.

11 years ago1. for charmrun, in the smp mode, the tasks_per_node keyword has some special usage
Chao Mei [Mon, 1 Feb 2010 20:21:00 +0000 (14:21 -0600)]
1. for charmrun, in the smp mode, the tasks_per_node keyword has some special usage

2. conv-common.h: change the basic msg header to make the message root as "int" instead of "unsigned short" to fit the case there are more than 64K procs.

3. machine.c: made changes for lapi smp layer where node-level broadcasting schemes (including spanning tree and hypercube) are implemented.

11 years agoAdding a check for smp/pxshm. If these options are not provided by ./build then they...
Isaac Dooley [Mon, 1 Feb 2010 19:01:19 +0000 (13:01 -0600)]
Adding a check for smp/pxshm. If these options are not provided by ./build then they will not be displayed to the user.

11 years agoask user to cd to build specific tests dir to test simplearrayhello.
Gengbin Zheng [Sun, 31 Jan 2010 15:41:55 +0000 (09:41 -0600)]
ask user to cd to build specific tests dir to test simplearrayhello.

11 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Sun, 31 Jan 2010 14:44:17 +0000 (08:44 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agodisabled skipping CmiGetNonLocalNodeQ for SMP when numpes = 1.
Gengbin Zheng [Sun, 31 Jan 2010 14:43:24 +0000 (08:43 -0600)]
disabled skipping CmiGetNonLocalNodeQ for SMP when numpes = 1.
Hmm, I thought this was checked in yesterday, but was actually not.

11 years agoremove the check for OBJECT_MODE = 32. this check is invalid if "mpcc64" build option...
Gengbin Zheng [Sun, 31 Jan 2010 04:15:25 +0000 (22:15 -0600)]
remove the check for OBJECT_MODE = 32. this check is invalid if "mpcc64" build option presents.

11 years agoskip CmiGetNonLocal only in non-smp version. In SMP CmiGetNonLocal does not poll...
Gengbin Zheng [Sun, 31 Jan 2010 03:47:50 +0000 (21:47 -0600)]
skip CmiGetNonLocal only in non-smp version. In SMP CmiGetNonLocal does not poll anyway, only comm thread does. CmiGetNonLocal does need to poll recv queue in case there is message received from comm thread (that executes immediate mesg)

11 years agoreverted some unwanted change in previous checkin.
Gengbin Zheng [Sun, 31 Jan 2010 02:41:43 +0000 (20:41 -0600)]
reverted some unwanted change in previous checkin.

11 years agonote that bitwise AND (&) has lower priority than !=. That breaks puTest on various...
Gengbin Zheng [Sat, 30 Jan 2010 05:15:01 +0000 (23:15 -0600)]
note that bitwise AND (&) has lower priority than !=. That breaks puTest on various platforms.

11 years agowhen run on only one processor, skip GetNonLocal (and hence polling of network)
Gengbin Zheng [Sat, 30 Jan 2010 02:57:49 +0000 (20:57 -0600)]
when run on only one processor, skip GetNonLocal (and hence polling of network)

11 years agoforce 32bit compilation. (so that if incidentally set OBJECT_MODE=64, quickthread...
Gengbin Zheng [Sat, 30 Jan 2010 02:55:57 +0000 (20:55 -0600)]
force 32bit compilation. (so that if incidentally set OBJECT_MODE=64, quickthread won't work under this setting)

11 years agorca_lib is only needed if we want network topology information
Abhinav Bhatele [Fri, 29 Jan 2010 23:22:06 +0000 (18:22 -0500)]
rca_lib is only needed if we want network topology information

11 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Fri, 29 Jan 2010 20:52:48 +0000 (14:52 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoAdding instruction of template load leveler file
Filippo Gioachin [Fri, 29 Jan 2010 20:52:00 +0000 (14:52 -0600)]
Adding instruction of template load leveler file

11 years agochange to use flag _chareRestored to disable use of chare after restart. This fixes...
Gengbin Zheng [Fri, 29 Jan 2010 20:51:43 +0000 (14:51 -0600)]
change to use flag _chareRestored to disable use of chare after restart. This fixes bigsim crash.

11 years agoanother flag to indicate if chares are restored or not.
Gengbin Zheng [Fri, 29 Jan 2010 20:50:59 +0000 (14:50 -0600)]
another flag to indicate if chares are restored or not.

11 years agoAMPI mpich-test: run charmrun within the binary directory
Phil Miller [Fri, 29 Jan 2010 01:28:45 +0000 (19:28 -0600)]
AMPI mpich-test: run charmrun within the binary directory

11 years agoAMPI mpich-test: make output more useful for bughunting
Phil Miller [Fri, 29 Jan 2010 01:28:09 +0000 (19:28 -0600)]
AMPI mpich-test: make output more useful for bughunting

11 years agoAMPI mpich-test: add missing headers from upstream tarball
Phil Miller [Fri, 29 Jan 2010 00:26:19 +0000 (18:26 -0600)]
AMPI mpich-test: add missing headers from upstream tarball

11 years agoAMPI mpich-test: make paths in scripts generic
Phil Miller [Fri, 29 Jan 2010 00:25:19 +0000 (18:25 -0600)]
AMPI mpich-test: make paths in scripts generic

11 years agoMerge branch 'charm' of charmgit:charm into charm
Abhinav Bhatele [Thu, 28 Jan 2010 21:57:17 +0000 (15:57 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agofixed indentations
Abhinav Bhatele [Thu, 28 Jan 2010 21:40:27 +0000 (15:40 -0600)]
fixed indentations

11 years agoAMPI: Formatting fix, to keep output variables together
Phil Miller [Thu, 28 Jan 2010 20:39:07 +0000 (14:39 -0600)]
AMPI: Formatting fix, to keep output variables together

11 years agoAMPI: Cleanup duplication in Cart_shift bugfix
Phil Miller [Thu, 28 Jan 2010 20:28:59 +0000 (14:28 -0600)]
AMPI: Cleanup duplication in Cart_shift bugfix

11 years agoeven with the latest fix, the implementation of AMPI_Cart_shift is still buggy...
Gengbin Zheng [Thu, 28 Jan 2010 19:56:56 +0000 (13:56 -0600)]
even with the latest fix, the implementation  of AMPI_Cart_shift  is still buggy when "disp" contain some extreme values (+ or -, or very big value). Should be a lot robust now.

11 years agofixed a bug in calculating rank_source in AMPI_Cart_shift
Gengbin Zheng [Thu, 28 Jan 2010 19:45:15 +0000 (13:45 -0600)]
fixed a bug in calculating rank_source in AMPI_Cart_shift

11 years agofixed a typo in previous checks that is caught only by xlc compiler.
Gengbin Zheng [Thu, 28 Jan 2010 01:50:02 +0000 (19:50 -0600)]
fixed a typo in previous checks that is caught only by xlc compiler.

11 years agoMerge branch 'charm' of charmgit:charm into charm
Phil Miller [Thu, 28 Jan 2010 00:48:01 +0000 (18:48 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agofix for sqrt in pwr7 and use safer cross platform cpp syntax
Eric Bohm [Thu, 28 Jan 2010 00:43:44 +0000 (18:43 -0600)]
fix for sqrt in pwr7 and use safer cross platform cpp syntax

11 years agoNote fixed version of PGI compiler, for later cleanup
Phil Miller [Thu, 28 Jan 2010 00:23:19 +0000 (18:23 -0600)]
Note fixed version of PGI compiler, for later cleanup

11 years agoFixed a bug in the line that compiles pingpong 2way.
Forrest Iandola [Wed, 27 Jan 2010 23:44:55 +0000 (17:44 -0600)]
Fixed a bug in the line that compiles pingpong 2way.

11 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Wed, 27 Jan 2010 21:03:09 +0000 (15:03 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoadded a little more testing for a plain chare
Gengbin Zheng [Wed, 27 Jan 2010 21:00:16 +0000 (15:00 -0600)]
added a little more testing for a plain chare

11 years agoNow ChaNGa requires also liveViz library
Filippo Gioachin [Wed, 27 Jan 2010 20:15:50 +0000 (14:15 -0600)]
Now ChaNGa requires also liveViz library

11 years agoadded two more global variables, one indicate that if current run is after restart...
Gengbin Zheng [Wed, 27 Jan 2010 20:47:57 +0000 (14:47 -0600)]
added two more global variables, one indicate that if current run is after restart, another variable is the number of PEs in the checkpoints that used to restart the current run.

11 years agoTypo
Filippo Gioachin [Wed, 27 Jan 2010 20:17:32 +0000 (14:17 -0600)]
Typo

11 years agoNow ChaNGa requires also liveViz library
Filippo Gioachin [Wed, 27 Jan 2010 20:15:50 +0000 (14:15 -0600)]
Now ChaNGa requires also liveViz library

11 years agoA little bit more logic in CkMessageDetailReplay
Filippo Gioachin [Wed, 27 Jan 2010 19:22:14 +0000 (13:22 -0600)]
A little bit more logic in CkMessageDetailReplay

11 years agoMaking CmiPushPE non static in all machine layers, since it must be usable by everybody
Filippo Gioachin [Wed, 27 Jan 2010 19:21:23 +0000 (13:21 -0600)]
Making CmiPushPE non static in all machine layers, since it must be usable by everybody

11 years agoAdding first sketch of CkMessageDetailReplay for single processor replay
Filippo Gioachin [Wed, 27 Jan 2010 17:30:09 +0000 (11:30 -0600)]
Adding first sketch of CkMessageDetailReplay for single processor replay

11 years agofixed the subroutine that pups plain chares. Chare migratable constructor needs t...
Gengbin Zheng [Wed, 27 Jan 2010 16:53:11 +0000 (10:53 -0600)]
fixed the subroutine that pups plain chares. Chare migratable constructor needs t be called to properly assign virtual tables of a chare class in order to use virtual function pup later.

11 years agoadded chare_types table for plain chares which store the chare types.
Gengbin Zheng [Wed, 27 Jan 2010 16:51:57 +0000 (10:51 -0600)]
added chare_types table for plain chares which store the chare types.
Added a new external function CkCreateLocalChare to allocate a new chare locally.

11 years agoRemoving dead code
Filippo Gioachin [Wed, 27 Jan 2010 04:52:22 +0000 (22:52 -0600)]
Removing dead code

11 years agoRemoving a few useless extern
Filippo Gioachin [Wed, 27 Jan 2010 04:42:02 +0000 (22:42 -0600)]
Removing a few useless extern

11 years agoMaking _initHandler record messages through record-replay
Filippo Gioachin [Wed, 27 Jan 2010 04:11:00 +0000 (22:11 -0600)]
Making _initHandler record messages through record-replay

11 years agoAdding a mechanism for SafePool to reset a data structure when this is placed
Filippo Gioachin [Wed, 27 Jan 2010 04:09:33 +0000 (22:09 -0600)]
Adding a mechanism for SafePool to reset a data structure when this is placed
back into the pool for later reuse.
Using this mechanism to reset the envelope eventID.

11 years agoMerge branch 'charm' of charmgit:charm into charm
Isaac Dooley [Tue, 26 Jan 2010 23:07:04 +0000 (17:07 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoImproving the performance steering planning routine in the control point framework.
Isaac Dooley [Tue, 26 Jan 2010 23:06:38 +0000 (17:06 -0600)]
Improving the performance steering planning routine in the control point framework.

11 years agoIncluding a header for stl algorithms.
Isaac Dooley [Tue, 26 Jan 2010 18:47:59 +0000 (13:47 -0500)]
Including a header for stl algorithms.

11 years agoMerge branch 'charm' of charmgit:charm into charm
Isaac Dooley [Tue, 26 Jan 2010 18:06:46 +0000 (13:06 -0500)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoMaking memory tuning happen faster.
Isaac Dooley [Tue, 26 Jan 2010 18:06:17 +0000 (13:06 -0500)]
Making memory tuning happen faster.

11 years agoImproving performance data gathering & recording for control point framework.
Isaac Dooley [Tue, 26 Jan 2010 17:46:28 +0000 (11:46 -0600)]
Improving performance data gathering & recording for control point framework.

11 years agoMerge branch 'charm' of charmgit:charm into charm
Isaac Dooley [Tue, 26 Jan 2010 17:44:52 +0000 (11:44 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Tue, 26 Jan 2010 16:03:46 +0000 (10:03 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoMerge branch 'charm' of charmgit:charm into charm
Isaac Dooley [Tue, 26 Jan 2010 16:01:29 +0000 (10:01 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoAdding an echo command to indicate that LIBS built successfully. Its previous last...
Isaac Dooley [Tue, 26 Jan 2010 16:00:42 +0000 (10:00 -0600)]
Adding an echo command to indicate that LIBS built successfully. Its previous last printed line is not very informative.

11 years agoMEMCHECK on some O(P^2) memory allocation in comlib.
Gengbin Zheng [Tue, 26 Jan 2010 07:46:27 +0000 (01:46 -0600)]
MEMCHECK on some O(P^2) memory allocation in comlib.

11 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Tue, 26 Jan 2010 06:41:03 +0000 (00:41 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoJust comments
Filippo Gioachin [Tue, 26 Jan 2010 03:46:14 +0000 (21:46 -0600)]
Just comments

11 years agoGroups and Nodegroups created after mainchares have finished (i.e as consequence
Filippo Gioachin [Tue, 26 Jan 2010 03:43:25 +0000 (21:43 -0600)]
Groups and Nodegroups created after mainchares have finished (i.e as consequence
of regular messages) are now buffered and processed as all other messages after
the initialization process has finished.

11 years agoDeleting a few handler never used
Filippo Gioachin [Tue, 26 Jan 2010 03:40:49 +0000 (21:40 -0600)]
Deleting a few handler never used

11 years agoremove the problematic CmiAbort when restoring chares when number of pes changes.
Gengbin Zheng [Tue, 26 Jan 2010 02:08:07 +0000 (20:08 -0600)]
remove the problematic CmiAbort when restoring chares when number of pes changes.

11 years agoalways using the old chare implementation (using obj ptr) for bigsim to avoid memory...
Gengbin Zheng [Tue, 26 Jan 2010 00:38:19 +0000 (18:38 -0600)]
always using the old chare implementation (using obj ptr) for bigsim to avoid memory needed to keep track of them.

11 years agoMerge branch 'charm' of charmgit:charm into charm
Eric Bohm [Mon, 25 Jan 2010 23:16:51 +0000 (17:16 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoMerge branch 'charm' of charmgit:charm into charm
Eric ebohm [Mon, 25 Jan 2010 23:16:22 +0000 (17:16 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoAdd VSX support for Power 7
Eric Bohm [Mon, 25 Jan 2010 23:10:02 +0000 (17:10 -0600)]
Add VSX support for Power 7

11 years agoMerge commit 'charm/charm' into charm
Filippo Gioachin [Mon, 25 Jan 2010 23:09:58 +0000 (17:09 -0600)]
Merge commit 'charm/charm' into charm

11 years agoMerge branch 'charm' of charmgit:charm into charm
Eric bohm [Mon, 25 Jan 2010 23:07:52 +0000 (17:07 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoFixing a bug in startup related to SMP: the node-visible variable _numInitNodeMsgs
Filippo Gioachin [Sun, 24 Jan 2010 06:07:46 +0000 (00:07 -0600)]
Fixing a bug in startup related to SMP: the node-visible variable _numInitNodeMsgs
is incremented even after the RODataMsg with the number of init messages has been
sent out. This causes a race condition on node zero.
With this, making the inequality a strict equality.
Placing back the RODataMsg to increment only of one message (there is no reason for having +2)

11 years agoMerge branch 'charm' of ssh://gitosis@charm.cs.uiuc.edu/charm into charm
Chee Wai Lee [Mon, 25 Jan 2010 22:59:25 +0000 (16:59 -0600)]
Merge branch 'charm' of ssh://gitosis@charm.cs.uiuc.edu/charm into charm

11 years agoUpdates to ARMCI support for armci 1.4. Examples now supported are:
Chee Wai Lee [Mon, 25 Jan 2010 22:53:00 +0000 (16:53 -0600)]
Updates to ARMCI support for armci 1.4. Examples now supported are:

examples/benchmark/lu/armci_blocking/*
examples/benchmark/lu/armci_nonblocking/*

11 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Mon, 25 Jan 2010 22:25:13 +0000 (16:25 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agocheckpoint/restore chares to disk files.
Gengbin Zheng [Mon, 25 Jan 2010 22:24:42 +0000 (16:24 -0600)]
checkpoint/restore chares to disk files.

11 years agoMerge branch 'charm' of git://charm.cs.uiuc.edu/charm into charm
Jonathan Lifflander [Mon, 25 Jan 2010 21:54:56 +0000 (15:54 -0600)]
Merge branch 'charm' of git://charm.cs.uiuc.edu/charm into charm

11 years agoI discovered a bug in threaded/sync entry methods, where if the chare that the sync...
Jonathan Lifflander [Mon, 25 Jan 2010 18:19:51 +0000 (12:19 -0600)]
I discovered a bug in threaded/sync entry methods, where if the chare that the sync entry method is part of does not reside on the same processor and the message is forwarded the line of code that I commented out will actually change the source PE, which inhibits the returning  message from the sync entry method from getting back to the right PE, causing the program to hang. Filippo Gioachin tracked this problem down to this location when I showed him the incorrect behavior.

11 years agoMoving the definition of the destructor to the .C file.
Filippo Gioachin [Sun, 24 Jan 2010 06:07:46 +0000 (00:07 -0600)]
Moving the definition of the destructor to the .C file.
This is because of a problem with xlC that fails when compiling optimized code.

11 years agoMerge commit 'charm/charm' into charm
Filippo Gioachin [Sat, 23 Jan 2010 01:12:53 +0000 (19:12 -0600)]
Merge commit 'charm/charm' into charm

11 years agoChanging the value indicating no data returned yet from NULL to -1 (it can be NULL...
Filippo Gioachin [Sat, 23 Jan 2010 01:10:04 +0000 (19:10 -0600)]
Changing the value indicating no data returned yet from NULL to -1 (it can be NULL if no data is passed in to send)

11 years agoAdding ability to specify initial values for control points via command line.
Isaac Dooley [Fri, 22 Jan 2010 23:33:30 +0000 (17:33 -0600)]
Adding ability to specify initial values for control points via command line.

11 years agodetect is mpicxx is a wrapper for pgCC, in which case CMK_CC_PGCC must be defined...
Gengbin Zheng [Fri, 22 Jan 2010 17:06:07 +0000 (11:06 -0600)]
detect is mpicxx is a wrapper for pgCC, in which case CMK_CC_PGCC must be defined as it is important to activate some STL hack in comlib for pgcc.

11 years agoMerge commit 'charm/charm' into charm
Filippo Gioachin [Fri, 22 Jan 2010 00:00:34 +0000 (18:00 -0600)]
Merge commit 'charm/charm' into charm

11 years agoHorrible! In one place pup callback with pup operator, in the other as bytes!!!
Filippo Gioachin [Thu, 21 Jan 2010 23:46:12 +0000 (17:46 -0600)]
Horrible! In one place pup callback with pup operator, in the other as bytes!!!

11 years agoMerge branch 'development' into charm
Filippo Gioachin [Thu, 21 Jan 2010 21:59:45 +0000 (15:59 -0600)]
Merge branch 'development' into charm

11 years agoIf the LB wants to send charm messages bypassing the Charm++ system (i.e as converse...
Filippo Gioachin [Thu, 21 Jan 2010 21:39:11 +0000 (15:39 -0600)]
If the LB wants to send charm messages bypassing the Charm++ system (i.e as converse messages), then it must obey to the CkMessageWatcher when delivering them

11 years agoApparently the even having a pup routine for CkCallaback does not solve the problem:
Filippo Gioachin [Thu, 21 Jan 2010 21:35:41 +0000 (15:35 -0600)]
Apparently the even having a pup routine for CkCallaback does not solve the problem:
there are many places where a CkCallback struct is placed verbose inside a message.
This means that the operator= is invoked to basically memcopy the structure as a whole.
Thus the solution to blank out the whole memory used by the callback in every constructor (disabled when CMK_OPTIMIZE is turned on).

11 years agoThe previous commit adding quiescence detection didn't actually work (it didn't do...
Filippo Gioachin [Thu, 21 Jan 2010 21:35:10 +0000 (15:35 -0600)]
The previous commit adding quiescence detection didn't actually work (it didn't do anything)
Now quiescence is really on

11 years agoRemoving print statements
Filippo Gioachin [Thu, 21 Jan 2010 21:33:53 +0000 (15:33 -0600)]
Removing print statements

11 years agofixed the previous fix on bzero - need to include charm config files for CMK_HAS_STRE...
Gengbin Zheng [Thu, 21 Jan 2010 16:59:22 +0000 (10:59 -0600)]
fixed the previous fix on bzero - need to include charm config files for CMK_HAS_STREINGS to be defined.