charm.git
16 years agoChanging CharmStrategy and DirectMulticastStrategy to use the strategy_handler
Sameer Kumar [Tue, 22 Mar 2005 15:14:03 +0000 (15:14 +0000)]
Changing CharmStrategy and DirectMulticastStrategy to use the strategy_handler
code.

16 years agoAdding a new handler which makes it easy for strategies to
Sameer Kumar [Tue, 22 Mar 2005 15:10:59 +0000 (15:10 +0000)]
Adding a new handler which makes it easy for strategies to
communicate with each other.

16 years agoAdded the node to node adjacency information
Sayantan Chakravorty [Tue, 22 Mar 2005 04:24:21 +0000 (04:24 +0000)]
Added the node to node adjacency information

16 years agoimplement change for LBDump format 1 which has a less field in procStat.
Gengbin Zheng [Tue, 22 Mar 2005 01:10:56 +0000 (01:10 +0000)]
implement change for LBDump format 1 which has a less field in procStat.
By default, dump version number into file after version 1.

16 years agoa runtime option +LBVersion to specify a version number for LBDump file.
Gengbin Zheng [Tue, 22 Mar 2005 00:42:06 +0000 (00:42 +0000)]
a runtime option +LBVersion to specify a version number for LBDump file.

16 years agointroduced a LBDump file version number to keep backward compatibilities.
Gengbin Zheng [Tue, 22 Mar 2005 00:41:24 +0000 (00:41 +0000)]
introduced  a LBDump file version number to keep backward compatibilities.

16 years agorearranged header inclusion so that now it allows LBDatabase to directly (inline...
Gengbin Zheng [Mon, 21 Mar 2005 20:38:09 +0000 (20:38 +0000)]
rearranged header inclusion so that now it allows LBDatabase to directly (inline) call to LBDBManager without paying a function call to LDfoo() in lbdb.C.

16 years agoenabled BigSim simualtor specific converse streaming to reduce of the overhead on...
Gengbin Zheng [Mon, 21 Mar 2005 20:27:26 +0000 (20:27 +0000)]
enabled BigSim simualtor specific converse streaming to reduce of the overhead on pe 0 when tnes of thousands of VPs send stats message to it.

16 years agowhen clearing LBStats database, call CkVec::clear instead of resize(0) to actually...
Gengbin Zheng [Mon, 21 Mar 2005 20:26:23 +0000 (20:26 +0000)]
when clearing LBStats database, call CkVec::clear instead of resize(0) to actually free the memory because resize(0) won't do anything if existing capacity is enough for the request which is 0.

16 years agoreplaced call to LBDatabaseObj() by the_lbdb which is already pre-assigned.
Gengbin Zheng [Mon, 21 Mar 2005 19:37:36 +0000 (19:37 +0000)]
replaced call to LBDatabaseObj() by the_lbdb which is already pre-assigned.

16 years agoadded some more recasts for the double<--complex cases. Should help with alignment...
Eric Bohm [Mon, 21 Mar 2005 17:54:14 +0000 (17:54 +0000)]
added some more recasts for the double<--complex cases.  Should help with alignment issues.

16 years agoupdated bluegene header
Gengbin Zheng [Mon, 21 Mar 2005 17:12:35 +0000 (17:12 +0000)]
updated bluegene header

16 years agoupdate bluegene header
Gengbin Zheng [Mon, 21 Mar 2005 17:11:06 +0000 (17:11 +0000)]
update bluegene header

16 years agoupdate bluegene header to use geenric header defined in converse.h.
Gengbin Zheng [Mon, 21 Mar 2005 17:09:06 +0000 (17:09 +0000)]
update bluegene header to use geenric header defined in converse.h.

16 years agoBug in the copy constructor for Array4D.
Eric Bohm [Mon, 21 Mar 2005 16:13:04 +0000 (16:13 +0000)]
Bug in the copy constructor for Array4D.
Noted by Gengbin, should be fixed.

16 years agoConverting CkpvAccess(_qd) back to CpvAccess(_qd). No idea why that broke bluegene...
Sameer Kumar [Mon, 21 Mar 2005 14:44:43 +0000 (14:44 +0000)]
Converting CkpvAccess(_qd) back to CpvAccess(_qd). No idea why that broke bluegene build. Gengbin, please verify.

16 years agoThe correct fix for setup complete flag. It takes into account migration and atsync...
Sameer Kumar [Mon, 21 Mar 2005 14:33:20 +0000 (14:33 +0000)]
The correct fix for setup complete flag. It takes into account migration and atsync calls between comlib setup of the previous step.

16 years agoswitch to use generic bluegene header CMK_BLUEGENE_FIELDS
Gengbin Zheng [Mon, 21 Mar 2005 08:17:46 +0000 (08:17 +0000)]
switch to use generic bluegene header CMK_BLUEGENE_FIELDS

16 years agoreversing change to stop using the receivedTable flag. Now all setup completion check...
Sameer Kumar [Mon, 21 Mar 2005 06:59:33 +0000 (06:59 +0000)]
reversing change to stop using the receivedTable flag. Now all setup completion checks use receiveTable flag. Will have to investigate this later.

16 years agoimplemented inline message for array.
Gengbin Zheng [Mon, 21 Mar 2005 05:57:47 +0000 (05:57 +0000)]
implemented inline message for array.
Fix inline deliver so that timing of the current object is temporarily stopped and resumed after inline message processed.

16 years agoCode cleanup. Shorter header for streaming and enable array caching again.
Sameer Kumar [Mon, 21 Mar 2005 05:21:51 +0000 (05:21 +0000)]
Code cleanup. Shorter header for streaming and enable array caching again.

16 years agoReducing iterations further.
Sameer Kumar [Mon, 21 Mar 2005 05:17:01 +0000 (05:17 +0000)]
Reducing iterations further.

16 years agoa minor name change CK_MSG_SKIPSCHEDULER => CK_MSG_EXPEDITED
Gengbin Zheng [Mon, 21 Mar 2005 05:13:19 +0000 (05:13 +0000)]
a minor name change CK_MSG_SKIPSCHEDULER => CK_MSG_EXPEDITED

16 years agoa minor name change CK_MSG_SKIPSCHEDULER => CK_MSG_EXPEDITED
Gengbin Zheng [Mon, 21 Mar 2005 05:07:14 +0000 (05:07 +0000)]
a minor name change CK_MSG_SKIPSCHEDULER => CK_MSG_EXPEDITED

16 years agoadded a new entry attribute [inline].
Gengbin Zheng [Mon, 21 Mar 2005 05:06:35 +0000 (05:06 +0000)]
added a new entry attribute [inline].

16 years agoRemoving unnecessary prints.
Sameer Kumar [Mon, 21 Mar 2005 05:04:00 +0000 (05:04 +0000)]
Removing unnecessary prints.

16 years agomake decl file depends on charmxi, so that it gets recompiled with cahrmxi is changed.
Gengbin Zheng [Mon, 21 Mar 2005 05:03:21 +0000 (05:03 +0000)]
make decl file depends on charmxi, so that it gets recompiled with cahrmxi is changed.

16 years agoreducing iterations for these two too!
Sameer Kumar [Sun, 20 Mar 2005 23:40:21 +0000 (23:40 +0000)]
reducing iterations for these two too!

16 years agoreducing iterations by 10 times.
Sameer Kumar [Sun, 20 Mar 2005 23:39:41 +0000 (23:39 +0000)]
reducing iterations by 10 times.

16 years agowhen array listener delivering a broadcast message to an element, check to make sure...
Gengbin Zheng [Sun, 20 Mar 2005 23:36:13 +0000 (23:36 +0000)]
when array listener delivering a broadcast message to an element, check to make sure if this array element has already received it before to prevent a duplication. Surprisingly this bug was not discovered until now.

16 years agochanging the start() broadcast to point-to-point messages. For some reason broadcast...
Sameer Kumar [Sun, 20 Mar 2005 22:10:38 +0000 (22:10 +0000)]
changing the start() broadcast to point-to-point messages. For some reason broadcast leads to a duplication which results in forwarding and hence an exception in the strategy on the destination processor.

16 years agoa fix for compiler error.
Gengbin Zheng [Sun, 20 Mar 2005 21:36:07 +0000 (21:36 +0000)]
a fix for compiler error.

16 years agofixed a minor thing in parsing traceRoot.
Gengbin Zheng [Sun, 20 Mar 2005 21:34:29 +0000 (21:34 +0000)]
fixed a minor thing in parsing traceRoot.

16 years agocleanup bluegene header.
Gengbin Zheng [Sun, 20 Mar 2005 21:01:53 +0000 (21:01 +0000)]
cleanup bluegene header.

16 years agocleanup bluegene header
Gengbin Zheng [Sun, 20 Mar 2005 21:00:10 +0000 (21:00 +0000)]
cleanup bluegene header

16 years agoclean up bluegene header
Gengbin Zheng [Sun, 20 Mar 2005 20:57:36 +0000 (20:57 +0000)]
clean up bluegene header

16 years agocleanup bluegene header.
Gengbin Zheng [Sun, 20 Mar 2005 20:56:15 +0000 (20:56 +0000)]
cleanup bluegene header.

16 years agoan optimization for memory on message broadcast.
Gengbin Zheng [Sun, 20 Mar 2005 20:53:29 +0000 (20:53 +0000)]
an optimization for memory on message broadcast.
The old scheme copies message on each simulated virtual processors on each
node, which can easily run out of memory. The new scheme only copies message
header and make a token, which saves memory dramatically.

16 years agowhen run charmrun under a job shell, apply -machinefile $PBS_NODEFILE in command...
Gengbin Zheng [Sun, 20 Mar 2005 17:34:08 +0000 (17:34 +0000)]
when run charmrun under a job shell, apply -machinefile $PBS_NODEFILE in command line.

16 years agoreverted change in batsync which caused a program hang when no load balancer. Reduced...
Gengbin Zheng [Sun, 20 Mar 2005 17:20:18 +0000 (17:20 +0000)]
reverted change in batsync which caused a program hang when no load balancer. Reduced lb period time from 1s to 0.5 second.

16 years agoDGEMM in calculatePairs_gemm is now the default.
Eric Bohm [Sun, 20 Mar 2005 04:27:06 +0000 (04:27 +0000)]
DGEMM in calculatePairs_gemm is now the default.
Set NOGEMM to return to the old (probably slower) behavior.
This checkin also strips out the zgemm version from calculatePairs_gemm
It was slower than dgemm and didn't produce correct results for noncomplex output.

If we ever need a calculatePairs which produces complex output we can dredge it
out of CVS.

16 years agoa helper function to return obj handle from index.
Gengbin Zheng [Sun, 20 Mar 2005 03:41:33 +0000 (03:41 +0000)]
a helper function to return obj handle from index.

16 years agoa simple function to return a LBObj from an index of the LBObj array.
Gengbin Zheng [Sun, 20 Mar 2005 03:34:48 +0000 (03:34 +0000)]
a simple function to return a LBObj from an index of the LBObj array.

16 years agono need to batsync for starting LB periodically if numpes is 1.
Gengbin Zheng [Sun, 20 Mar 2005 03:31:41 +0000 (03:31 +0000)]
no need to batsync for starting LB periodically if numpes is 1.

16 years agodefine bluegene msg header with generic macro defined in CMK_BLUEGENE_FIELDS of conve...
Gengbin Zheng [Sun, 20 Mar 2005 01:59:53 +0000 (01:59 +0000)]
define bluegene msg header with generic macro defined in CMK_BLUEGENE_FIELDS of converse.h.

16 years agoadded a reference counter to bluegene msg header for memory optimization.
Gengbin Zheng [Sun, 20 Mar 2005 01:55:09 +0000 (01:55 +0000)]
added a reference counter to bluegene msg header for memory optimization.

16 years agoThe DGEMM version for calculatePairs
Eric Bohm [Sun, 20 Mar 2005 00:07:29 +0000 (00:07 +0000)]
The DGEMM version for calculatePairs
Enable with _PAIRCALC_USE_DGEMM_

16 years agoadded a "flag" field in bluegene header for future use.
Gengbin Zheng [Sat, 19 Mar 2005 21:12:42 +0000 (21:12 +0000)]
added a "flag" field in bluegene header for future use.

16 years agosince bluegene converse msg header is common to all platforms, use the common definit...
Gengbin Zheng [Sat, 19 Mar 2005 21:11:16 +0000 (21:11 +0000)]
since bluegene converse msg header is common to all platforms, use the common definition defined in converse.h (CMK_BLUEGENE_FIELDS).

16 years agoprint number of real pes used for simualtion at startup.
Gengbin Zheng [Sat, 19 Mar 2005 20:37:06 +0000 (20:37 +0000)]
print number of real pes used for simualtion at startup.

16 years agosome reorgnization of timer subroutines.
Gengbin Zheng [Sat, 19 Mar 2005 20:33:08 +0000 (20:33 +0000)]
some reorgnization of timer subroutines.
Implemented a special purpose streaming for reducing messages for example in the case of collecting all load balancing stats into processor 0. After streaming, there will be one message out from one physical processor instead of as many as virtual processors on each physical node.

16 years agoDisabling benchmark for now. Will fix it when I get a chance.
Sameer Kumar [Sat, 19 Mar 2005 20:00:56 +0000 (20:00 +0000)]
Disabling benchmark for now. Will fix it when I get a chance.

16 years agofixed a compiler error for ibm compiler.
Gengbin Zheng [Sat, 19 Mar 2005 08:39:40 +0000 (08:39 +0000)]
fixed a compiler error for ibm compiler.

16 years agoAddition of calculatePairs_gemm which contains the GEMM version of the calculation.
Eric Bohm [Sat, 19 Mar 2005 04:54:13 +0000 (04:54 +0000)]
Addition of calculatePairs_gemm which contains the GEMM version of the calculation.
Enable it by turning on the -DGEMM flag.

Current implementation uses ZGEMM and strips out the imaginary results.

Works, solution seems equivalent.  Doesn't have much performance impact on cool cluster.

Implemented version using the DGEMM scheme, but it segfaults.  I also completely fail to comprehend how it could possibly work correctly.

16 years agoanother workaround
Sameer Kumar [Fri, 18 Mar 2005 23:22:14 +0000 (23:22 +0000)]
another workaround

16 years agocasting to satisfy rs6k xlc compiler
Eric Bohm [Fri, 18 Mar 2005 17:26:25 +0000 (17:26 +0000)]
casting to satisfy rs6k xlc compiler

16 years agoPutting a work around for the autobuild failures. For some reason I cannot remprduce...
Sameer Kumar [Fri, 18 Mar 2005 15:41:27 +0000 (15:41 +0000)]
Putting a work around for the autobuild failures. For some reason I cannot remprduce them. So cacheing of array ids seems to be my frist suspect.

16 years agoadded free() to CkVec to actually resize to 0.
Gengbin Zheng [Fri, 18 Mar 2005 06:02:46 +0000 (06:02 +0000)]
added free() to CkVec to actually resize to 0.

16 years agoMight have forgotten to pack the message. Adding an extra pack.
Sameer Kumar [Fri, 18 Mar 2005 04:15:26 +0000 (04:15 +0000)]
Might have forgotten to pack the message. Adding an extra pack.

16 years agoAdded RefineKLB
Tarun Agarwal [Fri, 18 Mar 2005 03:53:29 +0000 (03:53 +0000)]
Added RefineKLB
(Checking in again because forgot to add log comment in the last checkin)

16 years agoSeveral new changes
Sameer Kumar [Fri, 18 Mar 2005 03:21:55 +0000 (03:21 +0000)]
Several new changes

     Learning framework will now be tested in autobuild
     Fixed cached index bug.

     EachToManyMulticast strategy now does not work with sections. This makes it a lot simpler.
     The strategy should now be explicitly be used for all-to-all operations

     Cleaned up direct and ring multicast strategies. Isolated and abstracted out a lot
     of common code.

16 years agoFixing bug in the router strategy. MyPe was not initialized.
Sameer Kumar [Fri, 18 Mar 2005 03:19:25 +0000 (03:19 +0000)]
Fixing bug in the router strategy. MyPe was not initialized.

16 years agoFixing bug in linear.C. The benchmark was not storing the bucketsize correctly.
Sameer Kumar [Fri, 18 Mar 2005 03:18:26 +0000 (03:18 +0000)]
Fixing bug in linear.C. The benchmark was not storing the bucketsize correctly.
Now bucketsize is readonly.

16 years agoNew benchmarks to test the learning framework. Several new tests are now enabled.
Sameer Kumar [Fri, 18 Mar 2005 03:17:16 +0000 (03:17 +0000)]
New benchmarks to test the learning framework. Several new tests are now enabled.

16 years agoAdded RefineKLB to the Makefiles
Tarun Agarwal [Fri, 18 Mar 2005 02:17:50 +0000 (02:17 +0000)]
Added RefineKLB to the Makefiles

16 years ago*** empty log message ***
Tarun Agarwal [Fri, 18 Mar 2005 01:57:44 +0000 (01:57 +0000)]
*** empty log message ***

16 years agoFixed a locking problem, and one of many weird element problems.
Terry L. Wilmarth [Thu, 17 Mar 2005 22:55:51 +0000 (22:55 +0000)]
Fixed a locking problem, and one of many weird element problems.

16 years agoFixed make test
Eric Bohm [Thu, 17 Mar 2005 22:31:41 +0000 (22:31 +0000)]
Fixed make test

16 years agofixed an infinite loop in the rehash code
Nilesh Choudhury [Thu, 17 Mar 2005 21:56:57 +0000 (21:56 +0000)]
fixed an infinite loop in the rehash code

16 years agosomehow log message skipped last time ...
Gengbin Zheng [Thu, 17 Mar 2005 20:01:28 +0000 (20:01 +0000)]
somehow log message skipped last time ...
Implemented CmiMemoryUsage() that could print current memory usage. Only meant for gnu-malloc.

16 years ago*** empty log message ***
Gengbin Zheng [Thu, 17 Mar 2005 19:57:38 +0000 (19:57 +0000)]
*** empty log message ***

16 years agotweak for better compiler compat
Eric Bohm [Thu, 17 Mar 2005 19:39:54 +0000 (19:39 +0000)]
tweak for better compiler compat

16 years agotweaks for better C++ compatability and inlining with CMK_OPTIMIZE
Eric Bohm [Thu, 17 Mar 2005 19:12:17 +0000 (19:12 +0000)]
tweaks for better C++ compatability and inlining with CMK_OPTIMIZE

16 years agoMore stats collection. Moved initialization to after memory subsystem.
Eric Bohm [Thu, 17 Mar 2005 18:10:01 +0000 (18:10 +0000)]
More stats collection. Moved initialization to after memory subsystem.
Still unstable.

16 years agoFixed another locking bug, added derived boundaries. Boundaries work great, but...
Terry L. Wilmarth [Thu, 17 Mar 2005 00:50:55 +0000 (00:50 +0000)]
Fixed another locking bug, added derived boundaries.  Boundaries work great, but there are still a few locking problems when running sequentially.

16 years agoAdd cmipool to converse.
Eric Bohm [Wed, 16 Mar 2005 18:42:29 +0000 (18:42 +0000)]
Add cmipool to converse.

16 years agoBasic message pooling at the converse layer.
Eric Bohm [Wed, 16 Mar 2005 18:41:29 +0000 (18:41 +0000)]
Basic message pooling at the converse layer.
Set CONVERSE_POOL to enable.  Currently unstable.

16 years agoMore refinements.
Eric Bohm [Wed, 16 Mar 2005 18:04:13 +0000 (18:04 +0000)]
More refinements.
Still has some sort of race condition.  Probably need to add locking.

16 years agoFixed sequential deadlock problem... I hope.
Terry L. Wilmarth [Wed, 16 Mar 2005 17:53:31 +0000 (17:53 +0000)]
Fixed sequential deadlock problem... I hope.

16 years agoShould have been modified along with corresponding changes to other
Chee Wai Lee [Wed, 16 Mar 2005 04:15:04 +0000 (04:15 +0000)]
Should have been modified along with corresponding changes to other
name changes from registerFunction etc ... to traceRegisterFunction

16 years agoAdded operator= for 4D analogous to 6D.
Eric Bohm [Tue, 15 Mar 2005 23:05:05 +0000 (23:05 +0000)]
Added operator= for 4D analogous to 6D.
Should help with leanCP issues.

16 years agoFixes to not pool large messages.
Eric Bohm [Tue, 15 Mar 2005 22:43:22 +0000 (22:43 +0000)]
Fixes to not pool large messages.

16 years agoA standalone NetFEM to Paraview mesh file converter. Build with "make". Run executabl...
Isaac Dooley [Tue, 15 Mar 2005 17:36:33 +0000 (17:36 +0000)]
A standalone NetFEM to Paraview mesh file converter. Build with "make". Run executable from parent
directory of the "NetFEM" containing the chunk/timestep meshes.

16 years agoCleaned up my changes to make them look nicer, modified documentation. No functional...
Isaac Dooley [Tue, 15 Mar 2005 17:33:48 +0000 (17:33 +0000)]
Cleaned up my changes to make them look nicer, modified documentation. No functional changes.

16 years agoremoved an unused obj handle in computeInfo to save space.
Gengbin Zheng [Tue, 15 Mar 2005 17:16:23 +0000 (17:16 +0000)]
removed an unused obj handle in computeInfo to save space.

16 years agoForgot to check this in with the header file. Switching GAMMA to GAMMA_MEM
Sameer Kumar [Mon, 14 Mar 2005 20:12:57 +0000 (20:12 +0000)]
Forgot to check this in with the header file. Switching GAMMA to GAMMA_MEM

16 years agoRenaming GAMMA to GAMMA_MEM for the sp compilers.
Sameer Kumar [Mon, 14 Mar 2005 16:04:38 +0000 (16:04 +0000)]
Renaming GAMMA to GAMMA_MEM for the sp compilers.

16 years agoNew version to test the learning frmework.
Sameer Kumar [Sun, 13 Mar 2005 22:09:35 +0000 (22:09 +0000)]
New version to test the learning frmework.

16 years agoNew AAPC learner which uses a smarter technique. It now takes max of two sets of...
Sameer Kumar [Sun, 13 Mar 2005 21:55:37 +0000 (21:55 +0000)]
New AAPC learner which uses a smarter technique. It now takes max of two sets of parameter values.

16 years agoadded LBProfit into Makefile.
Gengbin Zheng [Sun, 13 Mar 2005 21:07:51 +0000 (21:07 +0000)]
added LBProfit into Makefile.

16 years agoa simple class for determining if a new load balancing cycle is profitable or not.
Gengbin Zheng [Sun, 13 Mar 2005 21:07:16 +0000 (21:07 +0000)]
a simple class for determining if a new load balancing cycle is profitable or not.

16 years agoshrank stats data structure size.
Gengbin Zheng [Sat, 12 Mar 2005 21:57:36 +0000 (21:57 +0000)]
shrank stats data structure size.

16 years agochange avail_vector to be static class member (this helps same memory for bigsim...
Gengbin Zheng [Sat, 12 Mar 2005 21:07:35 +0000 (21:07 +0000)]
change avail_vector to be static class member (this helps same memory for bigsim simulation)

16 years agofixed a memory leak.
Gengbin Zheng [Sat, 12 Mar 2005 08:24:34 +0000 (08:24 +0000)]
fixed a memory leak.

16 years agoAdded netfem_data.h to the headers. We want it to be included so we can build
Isaac Dooley [Sat, 12 Mar 2005 06:00:31 +0000 (06:00 +0000)]
Added netfem_data.h to the headers. We want it to be included so we can build
the standalone NetFEM -> Paraview converter against our version of charm/netfem.

16 years agoA minor change needed for the standalone NetFEM -> Paraview converter. Basically...
Isaac Dooley [Sat, 12 Mar 2005 05:57:07 +0000 (05:57 +0000)]
A minor change needed for the standalone NetFEM -> Paraview converter. Basically we add two accessor functions to get connectivity data. This
might be a hack, but was the easiest way I saw of making it work.

16 years agoAdded more debug output
Sayantan Chakravorty [Sat, 12 Mar 2005 01:06:44 +0000 (01:06 +0000)]
Added more debug output

16 years ago*** empty log message ***
Terry L. Wilmarth [Sat, 12 Mar 2005 00:38:27 +0000 (00:38 +0000)]
*** empty log message ***

16 years ago*** empty log message ***
Terry L. Wilmarth [Sat, 12 Mar 2005 00:23:45 +0000 (00:23 +0000)]
*** empty log message ***

16 years agoThis code is not working less than the previous checkin.
Terry L. Wilmarth [Fri, 11 Mar 2005 22:05:44 +0000 (22:05 +0000)]
This code is not working less than the previous checkin.