charm.git
14 years agobug fix
Chao Huang [Sat, 10 Feb 2007 11:37:01 +0000 (11:37 +0000)]
bug fix

14 years agoFixed silly memory bug.
Chee Wai Lee [Sat, 10 Feb 2007 00:50:22 +0000 (00:50 +0000)]
Fixed silly memory bug.

14 years ago*** empty log message ***
Chao Huang [Fri, 9 Feb 2007 22:19:29 +0000 (22:19 +0000)]
*** empty log message ***

14 years agocomplete make clean target
Chao Huang [Fri, 9 Feb 2007 17:21:49 +0000 (17:21 +0000)]
complete make clean target

14 years agoc++ version of jacobi
Chao Huang [Fri, 9 Feb 2007 17:20:27 +0000 (17:20 +0000)]
c++ version of jacobi

14 years agoSome more TOPS modifications
Isaac Dooley [Fri, 9 Feb 2007 01:36:34 +0000 (01:36 +0000)]
Some more TOPS modifications

14 years agofixed a CmiReleaseSentMessage() call in CmiGetNonLocal() (a bug introduced by Sameer...
Gengbin Zheng [Thu, 1 Feb 2007 22:13:05 +0000 (22:13 +0000)]
fixed a CmiReleaseSentMessage() call in CmiGetNonLocal() (a bug introduced by Sameer), which broke mpi-linux-smp.

14 years agoUpdated many objects to use initializers.
Eric Bohm [Thu, 1 Feb 2007 09:11:05 +0000 (09:11 +0000)]
Updated many objects to use initializers.
Tweaked UnsetTS logic to match use of POSE_UnsetTS
instead of a mix of that and -1.

Changed value of POSE_UnsetTS from -1 to -20, so misuse of uninitialized
timestamps is more likely to trigger an assert failue.

This version should be slightly faster than the old version.  It is notably
more correct in that BigNetSim is now processing large trace runs in
parallel when using LONG_TIMESTAMPS.

14 years agofixed a bug in AsyncSend, also add "++ppn" parameter.
Gengbin Zheng [Thu, 1 Feb 2007 03:28:32 +0000 (03:28 +0000)]
fixed a bug in AsyncSend, also add "++ppn" parameter.

14 years agosome uninitialized variables.
Gengbin Zheng [Thu, 1 Feb 2007 02:53:17 +0000 (02:53 +0000)]
some uninitialized variables.

14 years agofixed a possible string overflow and memory leak.
Gengbin Zheng [Fri, 26 Jan 2007 09:58:53 +0000 (09:58 +0000)]
fixed a possible string overflow and memory leak.

14 years agoAdded new feature in the mpi machine layer to add user events on MPI_Barrier,
Chee Wai Lee [Thu, 25 Jan 2007 22:00:12 +0000 (22:00 +0000)]
Added new feature in the mpi machine layer to add user events on MPI_Barrier,
MPI_Send, MPI_Recv, MPI_Isend and MPI_Irecv calls when the trace-projections
module is active.

Fixed 2 bugs associated with this:

  1) Machine layers used to (technically incorrect) register user events
     at ConverseExit. This is no longer possible as trace-projections is
     now a charm++ module. I've created a callback framework for machine
     layers to register their user event registration function which are
     then called by the (interested) trace module at the appropriate time.
     The MPI machine layer now uses this, ELAN still does not as I cannot
     test this easily and we don't typically do perf analysis on lemieux
     anymore, so there are no immediate problems with not updating the
     ELAN layer to do this.

  2) Performance analysis modules have *never* been tested properly with
     -DCMK_OPTIMIZE. trace-summary and trace-projections registers a
     callback function at module-exit time (called at CkExit) which, if
     simple like trace-summary's, escapes notice. trace-projections, on
     the other hand makes extensive use of the trace structure on close.
     This dies horribly since those structures are never created when the
     runtime is compiled with -DCMK_OPTIMIZE. My change is to simply call
     CkExit for the callback function (required to allow other modules to
     exercise their exit-callbacks) when CMK_OPTIMIZE is defined.

A third bug is left unfixed: Not all projections options are consumed even
if CMK_OPTIMIZE is defined. This has a not-so-intuitive effect of getting
a projections option passed down to the application (which may barf on it).
The proper behavior should have all options processed by the tracing
framework and OPTIMIZE_WARNING issued if CMK_OPTIMIZE is defined.

14 years agofixed a bug in puping projections multicast log
Gengbin Zheng [Thu, 25 Jan 2007 18:39:42 +0000 (18:39 +0000)]
fixed a bug in puping projections multicast log

14 years agofixed another compilaton error for BgMark
Gengbin Zheng [Wed, 24 Jan 2007 19:53:18 +0000 (19:53 +0000)]
fixed another compilaton error for BgMark

14 years agofixed a compiler error.
Gengbin Zheng [Wed, 24 Jan 2007 19:10:06 +0000 (19:10 +0000)]
fixed a compiler error.

14 years agoadded event marker.
Gengbin Zheng [Wed, 24 Jan 2007 05:30:15 +0000 (05:30 +0000)]
added event marker.

14 years agoremoved +bgprintoff, add BgMark() call to set a marker event.
Gengbin Zheng [Wed, 24 Jan 2007 05:29:35 +0000 (05:29 +0000)]
removed +bgprintoff, add BgMark() call to set a marker event.

14 years agoprint execTime in ascii format.
Gengbin Zheng [Wed, 24 Jan 2007 05:08:21 +0000 (05:08 +0000)]
print execTime in ascii format.

14 years agoNow bracketed times are correctly interpolated
Isaac Dooley [Wed, 24 Jan 2007 05:05:17 +0000 (05:05 +0000)]
Now bracketed times are correctly interpolated

14 years agofixing execTime when endTIme is 0.
Gengbin Zheng [Wed, 24 Jan 2007 04:54:28 +0000 (04:54 +0000)]
fixing execTime when endTIme is 0.

14 years agoprint relative time in ascii format of the log file
Gengbin Zheng [Wed, 24 Jan 2007 02:40:48 +0000 (02:40 +0000)]
print relative time in ascii format of the log file

14 years agoa little sanity check of the logs
Gengbin Zheng [Tue, 23 Jan 2007 23:39:49 +0000 (23:39 +0000)]
a little sanity check of the logs

14 years agominor change in comments
Gengbin Zheng [Tue, 23 Jan 2007 23:12:12 +0000 (23:12 +0000)]
minor change in comments

14 years agofixed the last log on PE0 where the endTime is 0.
Gengbin Zheng [Tue, 23 Jan 2007 23:11:04 +0000 (23:11 +0000)]
fixed the last log on PE0 where the endTime is 0.

14 years agoadd -fPIC so it can compile dynamic lib for csar.
Gengbin Zheng [Tue, 23 Jan 2007 22:04:30 +0000 (22:04 +0000)]
add -fPIC so it can compile dynamic lib for csar.

14 years agoadded a dummy network definition, which then disable the emulating network latency.
Gengbin Zheng [Tue, 23 Jan 2007 21:05:46 +0000 (21:05 +0000)]
added a dummy network definition, which then disable the emulating network latency.

14 years agoOutput fewer bgTrace files. Number should now match the input ones.
Isaac Dooley [Tue, 23 Jan 2007 20:03:42 +0000 (20:03 +0000)]
Output fewer bgTrace files. Number should now match the input ones.

14 years agovafriable name change for global-elf
Gengbin Zheng [Tue, 23 Jan 2007 19:25:15 +0000 (19:25 +0000)]
vafriable name change for global-elf

14 years agochange bgSize to _bgSize
Gengbin Zheng [Tue, 23 Jan 2007 19:22:53 +0000 (19:22 +0000)]
change bgSize to _bgSize

14 years ago+bgprintoff to turn screen output off.
Gengbin Zheng [Tue, 23 Jan 2007 19:21:53 +0000 (19:21 +0000)]
+bgprintoff to turn screen output off.

14 years agochanges for printing out durations
Isaac Dooley [Tue, 23 Jan 2007 05:47:56 +0000 (05:47 +0000)]
changes for printing out durations

14 years agoadded a new network
Gengbin Zheng [Tue, 23 Jan 2007 04:25:13 +0000 (04:25 +0000)]
added a new network

14 years agoNow added a scale factor for all unknown event times.
Isaac Dooley [Mon, 22 Jan 2007 23:46:46 +0000 (23:46 +0000)]
Now added a scale factor for all unknown event times.

14 years agoRemoving some extra output.
Isaac Dooley [Mon, 22 Jan 2007 21:51:16 +0000 (21:51 +0000)]
Removing some extra output.

14 years agoFixed memory issue.
Isaac Dooley [Mon, 22 Jan 2007 21:40:13 +0000 (21:40 +0000)]
Fixed memory issue.

14 years agoRemoved unused code
Isaac Dooley [Mon, 22 Jan 2007 19:25:17 +0000 (19:25 +0000)]
Removed unused code

14 years agoWrites a separate file for each simulated processor
Isaac Dooley [Mon, 22 Jan 2007 19:20:37 +0000 (19:20 +0000)]
Writes a separate file for each simulated processor

14 years agominot change
Gengbin Zheng [Sun, 21 Jan 2007 04:43:56 +0000 (04:43 +0000)]
minot change

14 years agoFunctions now can be separated into cases/classes
Isaac Dooley [Sat, 20 Jan 2007 20:02:15 +0000 (20:02 +0000)]
Functions now can be separated into cases/classes

14 years agoMore diagnostic functions
Isaac Dooley [Thu, 18 Jan 2007 22:40:01 +0000 (22:40 +0000)]
More diagnostic functions

14 years agoMinor changes.
Isaac Dooley [Thu, 18 Jan 2007 21:33:17 +0000 (21:33 +0000)]
Minor changes.

14 years agoFixed some problems. Now we identify when a predicted value is less than zero.
Isaac Dooley [Thu, 18 Jan 2007 03:57:07 +0000 (03:57 +0000)]
Fixed some problems. Now we identify when a predicted value is less than zero.

14 years agoFirst possibly correct/working version
Isaac Dooley [Thu, 18 Jan 2007 03:20:34 +0000 (03:20 +0000)]
First possibly correct/working version

14 years agoNow accepts wider range of input files
Isaac Dooley [Thu, 18 Jan 2007 02:29:27 +0000 (02:29 +0000)]
Now accepts wider range of input files

14 years agoNow I can read in the parameter file. The next step will be to glue the parameter...
Isaac Dooley [Thu, 18 Jan 2007 01:32:44 +0000 (01:32 +0000)]
Now I can read in the parameter file. The next step will be to glue the parameter file access to the rewriting of the logs.

14 years agostartTraceBigSim -> beginTraceBigSim ChaNGa_1-0
Chao Huang [Wed, 17 Jan 2007 23:09:25 +0000 (23:09 +0000)]
startTraceBigSim -> beginTraceBigSim

14 years agoadded flags for scripts (like configure) that require them
Filippo Gioachin [Wed, 17 Jan 2007 23:02:15 +0000 (23:02 +0000)]
added flags for scripts (like configure) that require them

14 years agoRemove debugging output
Chao Huang [Wed, 17 Jan 2007 22:09:29 +0000 (22:09 +0000)]
Remove debugging output

14 years agoAdd PUP::zdisk. Now AMPIMSGLOG controls if msg logging is enabled,
Chao Huang [Wed, 17 Jan 2007 21:50:17 +0000 (21:50 +0000)]
Add PUP::zdisk. Now AMPIMSGLOG controls if msg logging is enabled,
and CMK_PROJECTIONS_USE_ZLIB controls if zlib is used.

14 years agoPut PUP::zdisk to ampi library, just like projection compression pup'er
Chao Huang [Wed, 17 Jan 2007 21:48:47 +0000 (21:48 +0000)]
Put PUP::zdisk to ampi library, just like projection compression pup'er

14 years agoAdded two things
Chao Huang [Wed, 17 Jan 2007 09:17:12 +0000 (09:17 +0000)]
Added two things
1) AMPI msg logging mechanism (msg logging actually a lie. We replace
the output parameters including buffers). It's turned on by defining
AMPIMSGLOG
2) Interface (C/C++ and Fortran) for start/end tracing BigSim markers.

14 years agoAdded compress disk pup'er using zlib. It's turned on with definition of
Chao Huang [Wed, 17 Jan 2007 09:14:09 +0000 (09:14 +0000)]
Added compress disk pup'er using zlib. It's turned on with definition of
AMPIMSGLOG

14 years agoNow the interpolation works for multiple events. The traces still need to be
Isaac Dooley [Wed, 17 Jan 2007 05:29:41 +0000 (05:29 +0000)]
Now the interpolation works for multiple events. The traces still need to be
appropriately rewritten.

14 years agoRestructuring code to read in cycle accurate files with multiple functions timed...
Isaac Dooley [Wed, 17 Jan 2007 04:25:40 +0000 (04:25 +0000)]
Restructuring code to read in cycle accurate files with multiple functions timed in each.
Unfortunately the interpolation is not yet producing non-zero values.

14 years agoMore changes. Probably doesn't build yet.
Isaac Dooley [Wed, 17 Jan 2007 00:38:59 +0000 (00:38 +0000)]
More changes. Probably doesn't build yet.

14 years agoupdate sequential compiler to be for the compute node
Filippo Gioachin [Tue, 16 Jan 2007 22:10:13 +0000 (22:10 +0000)]
update sequential compiler to be for the compute node

14 years agoAdding some sample input files which will look more like what we are trying to do.
Isaac Dooley [Tue, 16 Jan 2007 21:20:33 +0000 (21:20 +0000)]
Adding some sample input files which will look more like what we are trying to do.

14 years agorevised manual to version 0.2
Filippo Gioachin [Mon, 15 Jan 2007 21:12:45 +0000 (21:12 +0000)]
revised manual to version 0.2
added Memory views description

14 years agochange to comments
Isaac Dooley [Mon, 15 Jan 2007 19:48:50 +0000 (19:48 +0000)]
change to comments

14 years agoAdded a class that handles the interpolation and wraps calls to gsl.
Isaac Dooley [Mon, 15 Jan 2007 19:45:12 +0000 (19:45 +0000)]
Added a class that handles the interpolation and wraps calls to gsl.

14 years agoMore Interpolation is working correctly
Isaac Dooley [Mon, 15 Jan 2007 19:44:44 +0000 (19:44 +0000)]
More Interpolation is working correctly

14 years agoAdding a new example/test case showing 2d dense array initialization.
Robert Blake [Thu, 11 Jan 2007 19:10:58 +0000 (19:10 +0000)]
Adding a new example/test case showing 2d dense array initialization.

14 years agoChanging the manual to talk about the new naive way to initialize arrays.
Robert Blake [Thu, 11 Jan 2007 19:07:40 +0000 (19:07 +0000)]
Changing the manual to talk about the new naive way to initialize arrays.

14 years agofixed a bug in handling broadcast, may resulted in out of order messages ot length...
Gengbin Zheng [Thu, 11 Jan 2007 03:33:38 +0000 (03:33 +0000)]
fixed a bug in handling broadcast, may resulted in out of order messages ot length mismatch.

14 years agofor getrusage implementation of cmiwalltimer, make sure that current time returned...
Gengbin Zheng [Wed, 10 Jan 2007 18:21:14 +0000 (18:21 +0000)]
for getrusage implementation of cmiwalltimer, make sure that current time returned increasedlook monotonically, which otherwise breaks simulation when timer goes backwards.

14 years agoRemoved some unused local malloc/free code. Moved some defines from spert.cpp to...
David Kunzman [Mon, 8 Jan 2007 05:18:15 +0000 (05:18 +0000)]
Removed some unused local malloc/free code.  Moved some defines from spert.cpp to spert.h.

14 years agoSwitched malloc and free to a local version instead of the malloc and free provided...
David Kunzman [Mon, 8 Jan 2007 05:00:42 +0000 (05:00 +0000)]
Switched malloc and free to a local version instead of the malloc and free provided by the SDK.

14 years agoUpdated spert.s.timing rule for SDK v2.0.
David Kunzman [Mon, 8 Jan 2007 04:58:52 +0000 (04:58 +0000)]
Updated spert.s.timing rule for SDK v2.0.

14 years agoA new program for interpolating values to a desired parameterization. params-time...
Isaac Dooley [Mon, 8 Jan 2007 03:18:22 +0000 (03:18 +0000)]
A new program for interpolating values to a desired parameterization. params-time-table would be generated by something like a cycle accurate simulator.

14 years agoMake clean now removes the directory newtraces
Isaac Dooley [Sat, 6 Jan 2007 21:29:43 +0000 (21:29 +0000)]
Make clean now removes the directory newtraces

14 years agoNow sent message times are corrected to be linearly mapped into the new time range...
Isaac Dooley [Sat, 6 Jan 2007 21:22:08 +0000 (21:22 +0000)]
Now sent message times are corrected to be linearly mapped into the new time range for an event. Also the output directory
is created, and an explanation at the top of rewritelog.C is more informative

14 years agoInitial version of a program which loads bgTrace* files and then outputs them. Next...
Isaac Dooley [Sat, 6 Jan 2007 19:34:58 +0000 (19:34 +0000)]
Initial version of a program which loads bgTrace* files and then outputs them. Next the code to modify the
trace data structures will be added.

14 years agoAdded more detailed timing data for the SPE_TIMING option (start and end times for...
David Kunzman [Fri, 5 Jan 2007 22:57:17 +0000 (22:57 +0000)]
Added more detailed timing data for the SPE_TIMING option (start and end times for the states a standard work request goes through are now collected).

14 years agoRemoved macros for InitOffloadAPI().
David Kunzman [Fri, 5 Jan 2007 22:47:19 +0000 (22:47 +0000)]
Removed macros for InitOffloadAPI().

14 years agoFound another POSE_TimeType bug. Works now, I guess the POSE_TimeMax
Terry L. Wilmarth [Fri, 5 Jan 2007 19:17:29 +0000 (19:17 +0000)]
Found another POSE_TimeType bug.  Works now, I guess the POSE_TimeMax
isn't a problem.  Maybe the debugger just doesn't print it properly.

14 years agoMade several POSE virtual time variables into POSE_TimeType instead of
Terry L. Wilmarth [Fri, 5 Jan 2007 18:58:53 +0000 (18:58 +0000)]
Made several POSE virtual time variables into POSE_TimeType instead of
int to avoid bad conversion problems.  When
USE_LONG_TIMESTAMPS=1, POSE_TimeMax appears to be -1 in
the debugger, so this is an outstanding bug in pose_config.h.

14 years agoFixed a bug with memory management. However, default mem_usage limit is
Terry L. Wilmarth [Wed, 3 Jan 2007 21:37:33 +0000 (21:37 +0000)]
Fixed a bug with memory management.  However, default mem_usage limit is
very low, 5 or 10.  Not yet sure where this is being set.  This will
slow down the code dramatically.  Setting it higher with +memman_pose
<val> will speed things up if memory is not a huge concern.

14 years agoFound a couple more stray INT_MAXs used in setting priorities.
Terry L. Wilmarth [Tue, 2 Jan 2007 22:50:59 +0000 (22:50 +0000)]
Found a couple more stray INT_MAXs used in setting priorities.

14 years agoFixed a bug with setting the timestamp. Used POSE_TimeMax instead of
Terry L. Wilmarth [Tue, 2 Jan 2007 22:06:09 +0000 (22:06 +0000)]
Fixed a bug with setting the timestamp.  Used POSE_TimeMax instead of
INT_MAX.

14 years agoChanged bikvs from static to Ckpv to avoid problem on SMP.
Chao Huang [Mon, 1 Jan 2007 02:49:51 +0000 (02:49 +0000)]
Changed bikvs from static to Ckpv to avoid problem on SMP.

14 years agofixed return statement for net-sol-cc
Filippo Gioachin [Fri, 29 Dec 2006 22:38:30 +0000 (22:38 +0000)]
fixed return statement for net-sol-cc

14 years agoImplemented a different way for dclock to be used with the mpi-crayxt3
Chee Wai Lee [Fri, 29 Dec 2006 19:03:07 +0000 (19:03 +0000)]
Implemented a different way for dclock to be used with the mpi-crayxt3
build based on Shawn Brown's (from PSC) modifications.

14 years agoMore spelling errors fixed
Isaac Dooley [Tue, 26 Dec 2006 18:38:17 +0000 (18:38 +0000)]
More spelling errors fixed

14 years agoMore spelling errors
Isaac Dooley [Tue, 26 Dec 2006 18:35:11 +0000 (18:35 +0000)]
More spelling errors

14 years agoFixed spelling errors
Isaac Dooley [Tue, 26 Dec 2006 18:33:33 +0000 (18:33 +0000)]
Fixed spelling errors

14 years agoAdded line to set mesh to "get" mode.
Terry L. Wilmarth [Thu, 21 Dec 2006 20:08:53 +0000 (20:08 +0000)]
Added line to set mesh to "get" mode.

14 years agoFixed a few small ampi bugs
Sayantan Chakravorty [Thu, 21 Dec 2006 09:13:55 +0000 (09:13 +0000)]
Fixed a few small ampi bugs
1. In  getAttr attribute_value should have been the address of kv_builtin_storage
2. MPI_BXOR needed to be implemented for unsigned int

14 years agocopyglobal change
Gengbin Zheng [Wed, 20 Dec 2006 15:34:46 +0000 (15:34 +0000)]
copyglobal change

14 years agoFixed the type of output parameter attr_val in MPI_Attr_get for built-in
Chao Huang [Wed, 20 Dec 2006 04:58:50 +0000 (04:58 +0000)]
Fixed the type of output parameter attr_val in MPI_Attr_get for built-in
types.

14 years agoFixed the type of output parameter attr_val in MPI_Attr_get for built-in
Chao Huang [Wed, 20 Dec 2006 04:53:48 +0000 (04:53 +0000)]
Fixed the type of output parameter attr_val in MPI_Attr_get for built-in
types.

14 years agoMinor bug fixes. Updated for Cell SDK 2.0
David Kunzman [Mon, 18 Dec 2006 22:31:27 +0000 (22:31 +0000)]
Minor bug fixes.  Updated for Cell SDK 2.0

14 years agoCalling InitOffloadAPI() directly (not using macro)
David Kunzman [Mon, 18 Dec 2006 22:29:30 +0000 (22:29 +0000)]
Calling InitOffloadAPI() directly (not using macro)

14 years agoUpdated for Cell SDK 2.0
David Kunzman [Mon, 18 Dec 2006 22:27:28 +0000 (22:27 +0000)]
Updated for Cell SDK 2.0

14 years agoOptimized some code for the SPE Runtime. Added timing data to the Notification
David Kunzman [Mon, 18 Dec 2006 19:31:33 +0000 (19:31 +0000)]
Optimized some code for the SPE Runtime.  Added timing data to the Notification
Queue.

14 years agoupdated charm and converse modules to support the new memory inspector in
Filippo Gioachin [Sun, 17 Dec 2006 16:50:01 +0000 (16:50 +0000)]
updated charm and converse modules to support the new memory inspector in
charmdebug. This includes a ccs handler to get the machine architecture
information from the running application, and new debug ccs handlers to
transport raw memory data from the application to charmdebug.

14 years agoChanging a tag for the posted receives. My changes are still only active when a speci...
Isaac Dooley [Thu, 14 Dec 2006 22:02:40 +0000 (22:02 +0000)]
Changing a tag for the posted receives. My changes are still only active when a special define is used.

14 years agoMade a few more refinements to the algorithm for determining the "best"
Greg Koenig [Thu, 14 Dec 2006 19:14:50 +0000 (19:14 +0000)]
Made a few more refinements to the algorithm for determining the "best"
PE to place an object onto.  Also, changed the default mode of the balancer
to use the discrete algorithm instead of the scaled load algorithm since
this mode usually works better in more cases.

14 years agoFixed a bug in using ckhashtableiterator. Its next object returns
Sayantan Chakravorty [Thu, 14 Dec 2006 09:40:26 +0000 (09:40 +0000)]
Fixed a bug in using ckhashtableiterator. Its next object returns
a pointer to the object not the object itself

14 years agoan error checking for file open
Gengbin Zheng [Thu, 14 Dec 2006 07:49:52 +0000 (07:49 +0000)]
an error checking for file open

14 years agoreadme for Bigsim
Gengbin Zheng [Thu, 14 Dec 2006 07:22:32 +0000 (07:22 +0000)]
readme for Bigsim