charm.git
17 years agoUpdated build script to support the papi build option.
Chee Wai Lee [Fri, 12 Mar 2004 02:18:13 +0000 (02:18 +0000)]
Updated build script to support the papi build option.

17 years agoFixed bug with newNbrRef initialization.
Terry L. Wilmarth [Thu, 11 Mar 2004 17:44:29 +0000 (17:44 +0000)]
Fixed bug with newNbrRef initialization.

17 years agothe minimum message size was set a little bit too high, set it back to 2^5.
Gengbin Zheng [Thu, 11 Mar 2004 04:07:27 +0000 (04:07 +0000)]
the minimum message size was set a little bit too high, set it back to 2^5.

17 years agonetwork version of CmiListSend of processor multicast.
Gengbin Zheng [Thu, 11 Mar 2004 01:58:22 +0000 (01:58 +0000)]
network version of CmiListSend of processor multicast.
It uses memory reference counter to reuse the buffer and avoid memory copying.
This is possible because CmiSyncSendAndFree() did not change the message buffer when it returns.

17 years agoone simple optimization for CmiListSend of multicast, reducing one last memory copy...
Gengbin Zheng [Thu, 11 Mar 2004 01:44:52 +0000 (01:44 +0000)]
one simple optimization for CmiListSend of multicast, reducing one last memory copy by using CmiSyncSendAndFree().

17 years agouse mkstemp instead of tmpnam if mkstemp is available.
Gengbin Zheng [Thu, 11 Mar 2004 00:35:47 +0000 (00:35 +0000)]
use mkstemp instead of tmpnam if mkstemp is available.

17 years agoupdated.
Gengbin Zheng [Thu, 11 Mar 2004 00:30:12 +0000 (00:30 +0000)]
updated.

17 years agotesting if mkstemp exists, turing new gcc3.3 starts to not accept tmpnam any more.
Gengbin Zheng [Thu, 11 Mar 2004 00:28:19 +0000 (00:28 +0000)]
testing if mkstemp exists, turing new gcc3.3 starts to not accept tmpnam any more.

17 years agore-merged my previous optimization back (lost in last commit), which when pending...
Gengbin Zheng [Wed, 10 Mar 2004 22:58:00 +0000 (22:58 +0000)]
re-merged my previous optimization back (lost in last commit), which when pending messages are too many, brake the sending until there is a sending token recylced. This surprisingly improve the performance by 2 seconds when sending a 200MB message.

17 years agoAdd label for reference from othercalls.tex.
Orion Lawlor [Wed, 10 Mar 2004 20:15:13 +0000 (20:15 +0000)]
Add label for reference from othercalls.tex.

17 years agoAdd description of new CkExitAfterQuiescence routine.
Orion Lawlor [Wed, 10 Mar 2004 20:13:45 +0000 (20:13 +0000)]
Add description of new CkExitAfterQuiescence routine.

17 years agoAdded CkExitAfterQuiescence, which shuts the program down as soon
Orion Lawlor [Wed, 10 Mar 2004 20:05:49 +0000 (20:05 +0000)]
Added CkExitAfterQuiescence, which shuts the program down as soon
as quiescence is reached.

17 years agoTest if charmrun machine's hostname does not resolve
Orion Lawlor [Wed, 10 Mar 2004 19:57:38 +0000 (19:57 +0000)]
Test if charmrun machine's hostname does not resolve
(this is common on people's home Linux boxes).

If the hostname doesn't resolve, warn the user and use 127.0.0.1.

To fix the problem, use a DNS hostname or add hostname to /etc/hosts.

17 years agoCheck printf (and fflush) return code for errors.
Orion Lawlor [Wed, 10 Mar 2004 19:42:45 +0000 (19:42 +0000)]
Check printf (and fflush) return code for errors.
This will prevent peoples' output silently disappearing
if, e.g., they redirect output to a full disk.

Currently only prints a warning to stderr; in some
cases e-mail may be our only hope of actually alerting
the user!

17 years agoAdded a simple serial test of converse user-level threads
Orion Lawlor [Wed, 10 Mar 2004 19:31:33 +0000 (19:31 +0000)]
Added a simple serial test of converse user-level threads
(Cth) routines.  Use "make test" to run the test.

17 years agoRenamed
Orion Lawlor [Wed, 10 Mar 2004 19:30:26 +0000 (19:30 +0000)]
Renamed
CMK_STACK_GROWDOWN as CMK_CONTEXT_STACKEND
CMK_STACK_GROWUP as the silent default CMK_CONTEXT_STACKBEGIN
CMK_STACK_GROWMIDDLE as CMK_CONTEXT_STACKMIDDLE

The new names are clearer, because *no* machine's stack
actually grows up; but *lots* of machines context routines
expect a pointer to the beginning of the stack.

The only user of these #defines is src/conv-core/threads.c.

17 years agoTwo small changes:
Orion Lawlor [Wed, 10 Mar 2004 19:28:09 +0000 (19:28 +0000)]
Two small changes:
  1.) Added support for uJcontext threads, which look very much like
setcontext/getcontext, but are implemented using setjmp/longjmp.
  2.) Renamed confusing CMK_STACK_GROWUP as CMK_CONTEXT_STACKEND.
This is clearer because the stack *always* grows down, the #defines
were controlling how we *passed* the stack to the context routines.

17 years agoAdded new CMK_HAS_ALLOCA_H #define, used by src/util/uJcontext.c
Orion Lawlor [Wed, 10 Mar 2004 19:26:23 +0000 (19:26 +0000)]
Added new CMK_HAS_ALLOCA_H #define, used by src/util/uJcontext.c

17 years agoRemove the old CMK_STACK_GROWDOWN define, which was never needed.
Orion Lawlor [Wed, 10 Mar 2004 19:25:15 +0000 (19:25 +0000)]
Remove the old CMK_STACK_GROWDOWN define, which was never needed.

17 years agoCreated a new threads package based on setjmp/alloca/longjmp
Orion Lawlor [Wed, 10 Mar 2004 19:24:38 +0000 (19:24 +0000)]
Created a new threads package based on setjmp/alloca/longjmp
(just like QuickThreads), but providing the more modern
SYSV getcontext/setcontext interface.

This is needed because:
  - Quickthreads is truly nasty code, and is beginning to be a portability
bottleneck.
  - The getcontext/setcontext interface is cleaner than Quickthreads'.
  - uJcontext works under Windows, while quickthreads won't.

17 years agoWhen adding new elements, this sample program uses the same technique as
Terry L. Wilmarth [Wed, 10 Mar 2004 19:24:22 +0000 (19:24 +0000)]
When adding new elements, this sample program uses the same technique as
tri.C to both preserve orientation and result in identical connectivity.

17 years agoCheck orientations of triangles coming out of refinement.
Orion Lawlor [Wed, 10 Mar 2004 16:56:35 +0000 (16:56 +0000)]
Check orientations of triangles coming out of refinement.

17 years agoBug fix: set keyval to invalid in MPI_Keyval_delete
Chao Huang [Wed, 10 Mar 2004 05:29:23 +0000 (05:29 +0000)]
Bug fix: set keyval to invalid in MPI_Keyval_delete

17 years agoAdded files to build with Intel compiler for x86.
Greg Koenig [Tue, 9 Mar 2004 23:32:55 +0000 (23:32 +0000)]
Added files to build with Intel compiler for x86.

17 years agoBug fix: add "stdin" on whitelist
Chao Huang [Tue, 9 Mar 2004 21:54:33 +0000 (21:54 +0000)]
Bug fix: add "stdin" on whitelist

17 years agoInitial checkin of Jade.
Jayant DeSouza [Tue, 9 Mar 2004 21:00:38 +0000 (21:00 +0000)]
Initial checkin of Jade.

17 years agoFirst set of bug fixes for yesterday's (big) commit.
Greg Koenig [Tue, 9 Mar 2004 19:44:27 +0000 (19:44 +0000)]
First set of bug fixes for yesterday's (big) commit.

17 years agoTrimmed spaces and tabs at end of line.
Jayant DeSouza [Tue, 9 Mar 2004 18:51:01 +0000 (18:51 +0000)]
Trimmed spaces and tabs at end of line.

17 years agoMake ghost layer addition into a subroutine.
Orion Lawlor [Tue, 9 Mar 2004 18:41:33 +0000 (18:41 +0000)]
Make ghost layer addition into a subroutine.

17 years agoOptimizations to fossil collection added.
Terry L. Wilmarth [Tue, 9 Mar 2004 17:29:46 +0000 (17:29 +0000)]
Optimizations to fossil collection added.

17 years agoModifications to make femtest work in fem_alone mode:
Orion Lawlor [Mon, 8 Mar 2004 23:35:45 +0000 (23:35 +0000)]
Modifications to make femtest work in fem_alone mode:
  - Replace CkPrintf with just printf
  - Replace CkAbort with "bad" routine

17 years agoPrint an error message if GM cannot be initialized.
Orion Lawlor [Mon, 8 Mar 2004 23:30:47 +0000 (23:30 +0000)]
Print an error message if GM cannot be initialized.
Without this printout, trying to run net-linux-gm programs
on the Turing front end shows *nothing*.

17 years agomake cpu and walltime calles better nested for begin/end execute, i.e. wall|cpu ...
Gengbin Zheng [Mon, 8 Mar 2004 23:14:30 +0000 (23:14 +0000)]
make cpu and walltime calles better nested for begin/end execute, i.e. wall|cpu => cpu|wall.

17 years agoThese are changes to unify the data structures for send, receive, and
Greg Koenig [Mon, 8 Mar 2004 22:17:32 +0000 (22:17 +0000)]
These are changes to unify the data structures for send, receive, and
persistent handles.  These changes either directly allow, or will allow
in the future, code paths to be optimized particularly for the case of
persistent communications handles.

Instead of allocating handles from the VMI memory manager, we simply
allocate an array of handles when the program starts up and use these.

17 years agoskip mkdir if the directory already exists.
Gengbin Zheng [Mon, 8 Mar 2004 22:10:30 +0000 (22:10 +0000)]
skip mkdir if the directory already exists.

17 years agoElement additions now preserve triangle orientation.
Terry L. Wilmarth [Mon, 8 Mar 2004 21:29:57 +0000 (21:29 +0000)]
Element additions now preserve triangle orientation.

17 years agoget ride of a //
Gengbin Zheng [Mon, 8 Mar 2004 20:55:07 +0000 (20:55 +0000)]
get ride of a //

17 years agoin MPI timer initialization, try to synchronize timers on all processors:
Gengbin Zheng [Mon, 8 Mar 2004 20:52:51 +0000 (20:52 +0000)]
in MPI timer initialization, try to synchronize timers on all processors:
1. check MPI environment variable if the clock is already globally synchronized, if so, we are fine, just find out the least time on all processors and use it as start time;
2. if no synchronized clock, try to make things better by doing several MPI barrier to synchronize time (it is not perfect though).

17 years agodefine fucntion CmiStartTimer(). In case the timer is absolute time, it returns the...
Gengbin Zheng [Mon, 8 Mar 2004 19:52:08 +0000 (19:52 +0000)]
define fucntion CmiStartTimer(). In case the timer is absolute time, it returns the program init time, if the timer is relative time, return 0.0 always.

17 years agoset trace start timer to the beginning of the program, depending on if the timer...
Gengbin Zheng [Mon, 8 Mar 2004 19:49:22 +0000 (19:49 +0000)]
set trace start timer to the beginning of the program, depending on if the timer is an absolute time or not, which is from CmiStartTimer().

17 years agoremoved the double reset of trace start timer in BeginComputation event.
Gengbin Zheng [Mon, 8 Mar 2004 19:48:15 +0000 (19:48 +0000)]
removed the double reset of trace start timer in BeginComputation event.

17 years agomake TRACE_CPUTIMER to call cputimer.
Gengbin Zheng [Mon, 8 Mar 2004 18:28:21 +0000 (18:28 +0000)]
make TRACE_CPUTIMER to call cputimer.

17 years agofixed a bug in neighborhood load balancing. It could invoke migration when calling...
Gengbin Zheng [Mon, 8 Mar 2004 07:02:36 +0000 (07:02 +0000)]
fixed a bug in neighborhood load balancing. It could invoke migration when calling AtSync() which tcharm is not happy with.

17 years agoimplemented fault tolerant gm layer. When GM detects a catastrophic error,
Gengbin Zheng [Mon, 8 Mar 2004 05:31:07 +0000 (05:31 +0000)]
implemented fault tolerant gm layer. When GM detects a catastrophic error,
it temporarily disables the delivery of all messages with the same sender
port, target port, and priority as the message that experienced the error.
This layer needs to properly handle the error message of GM and resume
the port.

17 years agochanged -native to -host
Gengbin Zheng [Sat, 6 Mar 2004 02:27:07 +0000 (02:27 +0000)]
changed -native to -host

17 years agoAdded \xspace, \jade.
Jayant DeSouza [Fri, 5 Mar 2004 21:18:01 +0000 (21:18 +0000)]
Added \xspace, \jade.

17 years agoFix for net-sol-cc
Jayant DeSouza [Fri, 5 Mar 2004 20:58:14 +0000 (20:58 +0000)]
Fix for net-sol-cc

17 years agonet-sol-cc bug bix
Jayant DeSouza [Fri, 5 Mar 2004 20:53:15 +0000 (20:53 +0000)]
net-sol-cc bug bix

17 years agoBugfix of +LBDump option flag.
Filippo Gioachin [Fri, 5 Mar 2004 20:41:45 +0000 (20:41 +0000)]
Bugfix of +LBDump option flag.

17 years agoCorrected a few misunderstandings about the PUP contract.
Orion Lawlor [Fri, 5 Mar 2004 20:27:19 +0000 (20:27 +0000)]
Corrected a few misunderstandings about the PUP contract.

17 years agoBug fix: \tt{foo} enters the typewriter mode for the rest of the
Orion Lawlor [Fri, 5 Mar 2004 19:32:40 +0000 (19:32 +0000)]
Bug fix: \tt{foo} enters the typewriter mode for the rest of the
document.  What you want is either {\tt foo} or \kw{foo}.

17 years agofixed bug in miscalculating the estimation of memory usage by the centralLBs.
Gengbin Zheng [Fri, 5 Mar 2004 07:20:40 +0000 (07:20 +0000)]
fixed bug in miscalculating the estimation of memory usage by the centralLBs.
also after sequential simulation, let it print the memory usage.

17 years agodefine os malloc as default;
Gengbin Zheng [Fri, 5 Mar 2004 05:17:33 +0000 (05:17 +0000)]
define os malloc as default;
enable CMK_MEMORY_PREALLOCATE_HACK for testing.

17 years agoAdded the implementation of the interface for FEM_Register_entity and FEM_Register_array
Sayantan Chakravorty [Fri, 5 Mar 2004 03:22:40 +0000 (03:22 +0000)]
Added the implementation of the interface for FEM_Register_entity and FEM_Register_array
FEM_Register_entity is used to register an entity with a present and maximum possible size
as well as a function pointer to be called when the number of elements of that entity exceeds
the max specified.
FEM_Register_array is used to register a user array as the data for an attribute of some entity.

17 years agoMoved the header files for Jonathan's search library to $(CDIR)/include/cklibs
Chee Wai Lee [Fri, 5 Mar 2004 00:27:04 +0000 (00:27 +0000)]
Moved the header files for Jonathan's search library to $(CDIR)/include/cklibs
instead of the original $(CDIR)/include. This avoids a conflict with the
system's search.h (it will always assume the use of the library version) by
requiring the programmer to explicitly use #include "cklibs/search.h" when
refering to our implementation.

Tests on Cjacobi3D/jacobi.C indicate that the correct behavior is being
observed.

17 years agoSmall addition to the explanation of CmiAlloc
Filippo Gioachin [Thu, 4 Mar 2004 22:44:04 +0000 (22:44 +0000)]
Small addition to the explanation of CmiAlloc

17 years agoModified behaviour of +LBDump option: now a negative number translated to 0 (first...
Filippo Gioachin [Thu, 4 Mar 2004 22:32:45 +0000 (22:32 +0000)]
Modified behaviour of +LBDump option: now a negative number translated to 0 (first step).

17 years agoSimulation DumpStep better explanation
Filippo Gioachin [Thu, 4 Mar 2004 22:30:00 +0000 (22:30 +0000)]
Simulation DumpStep better explanation

17 years agoAdded empty pup with error essage. Need full pup for Jade.
Jayant DeSouza [Thu, 4 Mar 2004 20:20:36 +0000 (20:20 +0000)]
Added empty pup with error essage.  Need full pup for Jade.

17 years agoFixed Makefile bug; added constructors needed for Jade.
Jayant DeSouza [Thu, 4 Mar 2004 20:06:04 +0000 (20:06 +0000)]
Fixed Makefile bug; added constructors needed for Jade.

17 years agodefined F90 compiler.
Gengbin Zheng [Thu, 4 Mar 2004 04:37:16 +0000 (04:37 +0000)]
defined F90 compiler.

17 years agofixed several compiler errors with bluegene and icc on Tungsten.
Gengbin Zheng [Thu, 4 Mar 2004 04:33:48 +0000 (04:33 +0000)]
fixed several compiler errors with bluegene and icc on Tungsten.

17 years agonew option "elan" for -lelan option.
Gengbin Zheng [Wed, 3 Mar 2004 23:04:58 +0000 (23:04 +0000)]
new option "elan" for -lelan option.
also use -lmpi instead of -lpmpi.

17 years agoadd -Aa to turn on new ansi C features.
Gengbin Zheng [Wed, 3 Mar 2004 23:03:05 +0000 (23:03 +0000)]
add -Aa to turn on new ansi C features.

17 years agoported to GM 2.0 which is slightly incompatiable with the previous 1.0.
Gengbin Zheng [Wed, 3 Mar 2004 19:59:21 +0000 (19:59 +0000)]
ported to GM 2.0 which is slightly incompatiable with the previous 1.0.

17 years agoClarified use of | in pups and made use of it in the manual more consistent.
Eric Bohm [Wed, 3 Mar 2004 18:44:33 +0000 (18:44 +0000)]
Clarified use of | in pups and made use of it in the manual more consistent.

17 years agoMade various modes of PUP usage more clear.
Eric Bohm [Wed, 3 Mar 2004 17:16:58 +0000 (17:16 +0000)]
Made various modes of PUP usage more clear.

17 years agoadded cputime to trace projections logs for beginExecute and endExecute events.
Gengbin Zheng [Tue, 2 Mar 2004 21:38:27 +0000 (21:38 +0000)]
added cputime to trace projections logs for beginExecute and endExecute events.

17 years agoFixed assert warning on origin compilers.
Jayant DeSouza [Tue, 2 Mar 2004 18:15:51 +0000 (18:15 +0000)]
Fixed assert warning on origin compilers.

17 years agoCommented out unused variable.
Jayant DeSouza [Tue, 2 Mar 2004 18:13:01 +0000 (18:13 +0000)]
Commented out unused variable.

17 years agoReplaced unportable bzero with portable memset.
Orion Lawlor [Tue, 2 Mar 2004 17:16:52 +0000 (17:16 +0000)]
Replaced unportable bzero with portable memset.

17 years agoremoved bufferred broadcast messages in CkArrayBroadcaster::flushState().
Lixia Shi [Tue, 2 Mar 2004 04:51:34 +0000 (04:51 +0000)]
removed bufferred broadcast messages in CkArrayBroadcaster::flushState().

17 years agolatest version with fault tolerant support (__FAULT__)
Lixia Shi [Tue, 2 Mar 2004 04:14:30 +0000 (04:14 +0000)]
latest version with fault tolerant support (__FAULT__)

17 years agothe latest version of fault tolerant support with bug fix for a race condition.
Lixia Shi [Tue, 2 Mar 2004 04:03:33 +0000 (04:03 +0000)]
the latest version of fault tolerant support with bug fix for a race condition.
It only happens frequently on very big machines - restarted process A sends a message to B which hasn't received the updated socket table. B holds A's old UDP socket, thus the UDP ack for the message could not go through to A. When A timeout, it resent the message to B, and the message now is executed twice! This could repeat until B receives a updated node table from charmrun.

17 years agoError checking added to eventQueue code, and a bug in the InternalCommitPrintf code...
Terry L. Wilmarth [Mon, 1 Mar 2004 23:07:07 +0000 (23:07 +0000)]
Error checking added to eventQueue code, and a bug in the InternalCommitPrintf code fixed, as well as a related bug in CommitError.

17 years agoCorrected indentation.
Jayant DeSouza [Mon, 1 Mar 2004 17:51:59 +0000 (17:51 +0000)]
Corrected indentation.

17 years agoMoved msa header files to include/msa subdirectory.
Jayant DeSouza [Mon, 1 Mar 2004 17:06:25 +0000 (17:06 +0000)]
Moved msa header files to include/msa subdirectory.

17 years agomoved msa header files to include/msa/ subdirectory.
Jayant DeSouza [Mon, 1 Mar 2004 17:00:21 +0000 (17:00 +0000)]
moved msa header files to include/msa/ subdirectory.

17 years agoAdded build instructions for msa.
Jayant DeSouza [Mon, 1 Mar 2004 16:37:28 +0000 (16:37 +0000)]
Added build instructions for msa.

17 years agomultiphase shared arrays documentation.
Jayant DeSouza [Mon, 1 Mar 2004 16:05:38 +0000 (16:05 +0000)]
multiphase shared arrays documentation.

17 years agoAdded multiphase shared arrays documentation.
Jayant DeSouza [Mon, 1 Mar 2004 16:05:19 +0000 (16:05 +0000)]
Added multiphase shared arrays documentation.

17 years agoInitial checkin of multiphase shared arrays.
Jayant DeSouza [Mon, 1 Mar 2004 15:35:29 +0000 (15:35 +0000)]
Initial checkin of multiphase shared arrays.

17 years agoInitial checkin of multiphase shared arrays.
Jayant DeSouza [Mon, 1 Mar 2004 15:35:15 +0000 (15:35 +0000)]
Initial checkin of multiphase shared arrays.
Implemented by Rahul and Jay DeSouza
To use, #include "msa.h", and compile using charmc with -module msa.

17 years agochanged CentralLB and BaseLB from irreduciable group to normal group.
Gengbin Zheng [Mon, 1 Mar 2004 07:21:55 +0000 (07:21 +0000)]
changed CentralLB and BaseLB from irreduciable group to normal group.
Collected all command line options flags into a class instead of individual flags flying around.
Implemented a true barrier after load balancing and migrations of objects
finish on *all* processors. This can be turned on by +LBSynResume command line option. The default behavior is still that each processor waits only for objects that will migrate on it.

17 years agofixed compiler error when CMK_MEM_CHECKPOINT is off.
Gengbin Zheng [Sun, 29 Feb 2004 04:19:01 +0000 (04:19 +0000)]
fixed compiler error when CMK_MEM_CHECKPOINT is off.

17 years agolatest stable version, with a lot of bug fixes.
Lixia Shi [Sun, 29 Feb 2004 04:13:11 +0000 (04:13 +0000)]
latest stable version, with a lot of bug fixes.

17 years agoflushStates() for BaseLB to clear load balancing database after restart from crash.
Lixia Shi [Sat, 28 Feb 2004 17:20:10 +0000 (17:20 +0000)]
flushStates() for BaseLB to clear load balancing database after restart from crash.

17 years agoUnset commitErr flag in UndoEvent. If the event is undone, we don't want it to
Terry L. Wilmarth [Fri, 27 Feb 2004 21:19:17 +0000 (21:19 +0000)]
Unset commitErr flag in UndoEvent.  If the event is undone, we don't want it to
flag an error when it gets redone and committed.

17 years agoFixed a bug in the updating of neighbor faces to point to new elements that are creat...
Terry L. Wilmarth [Fri, 27 Feb 2004 21:16:38 +0000 (21:16 +0000)]
Fixed a bug in the updating of neighbor faces to point to new elements that are created.  Runs on 2 PEs now!

17 years agoAll three refinements work together now. Fixed a bug in one of the refinements that...
Terry L. Wilmarth [Fri, 27 Feb 2004 20:15:49 +0000 (20:15 +0000)]
All three refinements work together now.  Fixed a bug in one of the refinements that did not pass on the target volume to newly created elements.

17 years agoSome adjustments to locking/unlocking pattern. Improving how far along the LF and...
Terry L. Wilmarth [Fri, 27 Feb 2004 18:27:05 +0000 (18:27 +0000)]
Some adjustments to locking/unlocking pattern.  Improving how far along the LF and LE strategies get when working together.

17 years agoadded sndump so eventids can be output within commitprintfs sanely.
Eric Bohm [Thu, 26 Feb 2004 17:16:40 +0000 (17:16 +0000)]
added sndump so eventids can be output within commitprintfs sanely.

17 years agoThere were still some glitches with the GVT when running with BigNetSim. I think...
Terry L. Wilmarth [Thu, 26 Feb 2004 05:21:06 +0000 (05:21 +0000)]
There were still some glitches with the GVT when running with BigNetSim.  I think these are now worked out, and the GVT is even faster.  It may have troubles as the number of processors increases though...

17 years agoWhoops! Some little bugs in CmiAsserts fixed.
Terry L. Wilmarth [Wed, 25 Feb 2004 18:19:38 +0000 (18:19 +0000)]
Whoops!  Some little bugs in CmiAsserts fixed.

17 years agoDramatic GVT improvements. It's much faster now.
Terry L. Wilmarth [Wed, 25 Feb 2004 17:38:58 +0000 (17:38 +0000)]
Dramatic GVT improvements.  It's much faster now.

17 years agoRemoved some outdated dumping code.
Terry L. Wilmarth [Wed, 25 Feb 2004 16:00:11 +0000 (16:00 +0000)]
Removed some outdated dumping code.

17 years agoFor in-memory checkpoint at restarting phase, ignore contributorArriving
Gengbin Zheng [Wed, 25 Feb 2004 03:57:47 +0000 (03:57 +0000)]
For in-memory checkpoint at restarting phase, ignore contributorArriving
when modifying the adj(r).lcount. This is because it is really not an
arriving event from another processor. The array element is created from
old states.

17 years agoBug fix: Thanks to Sandhya, the f90 interface finally *works* as well
Orion Lawlor [Tue, 24 Feb 2004 22:10:40 +0000 (22:10 +0000)]
Bug fix: Thanks to Sandhya, the f90 interface finally *works* as well
as compiles.

The problem is that f90 passes in a "f90 pointer" object when you
give a subroutine prototype that includes
integer, pointer :: conn(:,:)
Without the prototype, f90 passes in a pointer to the actual data.

17 years agoIn IF statement of SDAG, the CCode matching rule defined in xi-grammar.y
Lixia Shi [Tue, 24 Feb 2004 21:59:21 +0000 (21:59 +0000)]
In IF statement of SDAG, the CCode matching rule defined in xi-grammar.y
can not recoginize a C function which has more than one paramters, ie. with ',' in between. Found this when making changes for LeanMD. This is fixed now.

17 years agoLF strategy doesn't want to place nice with LE strategy. BUT, it is
Terry L. Wilmarth [Tue, 24 Feb 2004 01:00:30 +0000 (01:00 +0000)]
LF strategy doesn't want to place nice with LE strategy.  BUT, it is
converted to the new locking mechanism, and it does work with the CP
refinements.  I also got rid of some unused code.