charm.git
16 years agoadded a callback functionality at the end of an iteration
Filippo Gioachin [Thu, 14 Apr 2005 00:18:55 +0000 (00:18 +0000)]
added a callback functionality at the end of an iteration

16 years agoChecking in adaptivity module. Not yet built.
Terry L. Wilmarth [Wed, 13 Apr 2005 23:33:43 +0000 (23:33 +0000)]
Checking in adaptivity module.  Not yet built.

16 years agoEven more stubs.
Terry L. Wilmarth [Wed, 13 Apr 2005 23:32:16 +0000 (23:32 +0000)]
Even more stubs.

16 years agoadded missing CmiAlloc and CmiFree.
Gengbin Zheng [Wed, 13 Apr 2005 22:35:23 +0000 (22:35 +0000)]
added missing CmiAlloc and CmiFree.

16 years agoupdated for new blue_standalone.C
Gengbin Zheng [Wed, 13 Apr 2005 21:26:05 +0000 (21:26 +0000)]
updated for new blue_standalone.C

16 years agoa standalone file for using blue_logs for a pure C program, it has implementations...
Gengbin Zheng [Wed, 13 Apr 2005 21:25:09 +0000 (21:25 +0000)]
a standalone file for using blue_logs for a pure C program, it has implementations of some converse functions such as CmiAbort (shamelessly stolen from ckstandalone.C).

16 years agolink with blue-standalone
Gengbin Zheng [Wed, 13 Apr 2005 21:24:00 +0000 (21:24 +0000)]
link with blue-standalone

16 years agotake off libckstandalone.a for now.
Gengbin Zheng [Wed, 13 Apr 2005 17:26:20 +0000 (17:26 +0000)]
take off libckstandalone.a for now.

16 years agoMore stubs for mesh modification added.
Terry L. Wilmarth [Wed, 13 Apr 2005 16:54:29 +0000 (16:54 +0000)]
More stubs for mesh modification added.

16 years agocompile ckstandalone.C into a library.
Gengbin Zheng [Wed, 13 Apr 2005 16:01:52 +0000 (16:01 +0000)]
compile ckstandalone.C into a library.

16 years agoa better way to fix it - compiling into pure c++ program with new ckstandalone lib.
Gengbin Zheng [Wed, 13 Apr 2005 16:01:21 +0000 (16:01 +0000)]
a better way to fix it - compiling into pure c++ program with new ckstandalone lib.

16 years agomoved Cmi message calls into blue_timing.C so that blue_logs library can be used...
Gengbin Zheng [Wed, 13 Apr 2005 15:58:10 +0000 (15:58 +0000)]
moved Cmi message calls  into blue_timing.C so that blue_logs library can be used in a sequential c program.

16 years agoreverted previous change
Gengbin Zheng [Wed, 13 Apr 2005 15:32:05 +0000 (15:32 +0000)]
reverted previous change

16 years agouse charmrun to run it.
Gengbin Zheng [Wed, 13 Apr 2005 14:43:06 +0000 (14:43 +0000)]
use charmrun to run it.

16 years agoadded return for non-void functions.
Gengbin Zheng [Wed, 13 Apr 2005 14:41:42 +0000 (14:41 +0000)]
added return for non-void functions.

16 years agopgm.C writes NetFEM data to file.
Terry L. Wilmarth [Tue, 12 Apr 2005 21:21:28 +0000 (21:21 +0000)]
pgm.C writes NetFEM data to file.

16 years agoAdded empty adjacency operations.
Terry L. Wilmarth [Tue, 12 Apr 2005 20:29:28 +0000 (20:29 +0000)]
Added empty adjacency operations.

16 years agorevert previous change because it has to be a parallel compilation (not pure c++...
Gengbin Zheng [Tue, 12 Apr 2005 19:10:14 +0000 (19:10 +0000)]
revert previous change because it has to be a parallel compilation (not pure c++ code), although it has to be run on one processor.

16 years agofixed seq compilers and f90main.
Gengbin Zheng [Tue, 12 Apr 2005 18:13:30 +0000 (18:13 +0000)]
fixed seq compilers and f90main.

16 years agoUpdated refinement edge split operation to take node indices local to the edge itself...
Terry L. Wilmarth [Tue, 12 Apr 2005 17:07:36 +0000 (17:07 +0000)]
Updated refinement edge split operation to take node indices local to the edge itself.  This is now consistent with how the edge collapse parameters are passed. Added idxl list interface to nodeUpdate, but not yet passing in meaning ful data.

16 years agochanged the coarsenclient so that instead of queuing up operations it performs them...
Sayantan Chakravorty [Tue, 12 Apr 2005 06:47:18 +0000 (06:47 +0000)]
changed the coarsenclient so that instead of queuing up operations it performs them immediately

16 years agosupport idb.
Gengbin Zheng [Tue, 12 Apr 2005 05:42:12 +0000 (05:42 +0000)]
support idb.

16 years agosmarter version, able to iterate over LoadLeveler errors (255) but signalling
Filippo Gioachin [Tue, 12 Apr 2005 02:14:15 +0000 (02:14 +0000)]
smarter version, able to iterate over LoadLeveler errors (255) but signalling
user errors (not 255)

16 years agobugfix in the new implementation of cklists. Also, changed back to non-allocation
Filippo Gioachin [Tue, 12 Apr 2005 00:47:15 +0000 (00:47 +0000)]
bugfix in the new implementation of cklists. Also, changed back to non-allocation
of memory if the queue is never used

16 years agocorrected comment section error
Filippo Gioachin [Mon, 11 Apr 2005 23:31:09 +0000 (23:31 +0000)]
corrected comment section error

16 years agoFixed hanging problem with grainSize != S
Eric Bohm [Mon, 11 Apr 2005 18:49:02 +0000 (18:49 +0000)]
Fixed hanging problem with grainSize != S

16 years agomore dead code removal
Eric Bohm [Mon, 11 Apr 2005 18:35:08 +0000 (18:35 +0000)]
more dead code removal

16 years agostripping out the ifdef SPARSE_CONTINUOUS, its been the default for months.
Eric Bohm [Mon, 11 Apr 2005 18:31:08 +0000 (18:31 +0000)]
stripping out the ifdef SPARSE_CONTINUOUS, its been the default for months.
the non case has been removed.

Stopped sending the always zero othernewdata in the backward path.
Theoretically this might cause a behavior change somewhere.
But as Glenn's email indicated no use for it, and we're not creating one,
there seems little value in sending it around so its zeros can be
ignored and take up space.

Moved the nogemm calculatePairs to the end of the file.

16 years agoInitial addition of this file to charm.
David Kunzman [Sun, 10 Apr 2005 22:32:46 +0000 (22:32 +0000)]
Initial addition of this file to charm.

16 years agoInitial creation of file. Not fully functional yet.
David Kunzman [Sun, 10 Apr 2005 22:31:49 +0000 (22:31 +0000)]
Initial creation of file.  Not fully functional yet.

16 years agoAdded Ammasso hooks.
David Kunzman [Sun, 10 Apr 2005 22:28:44 +0000 (22:28 +0000)]
Added Ammasso hooks.

16 years agoAdded element to element adjacency functions.
Isaac Dooley [Sun, 10 Apr 2005 18:25:49 +0000 (18:25 +0000)]
Added element to element adjacency functions.

16 years agoAdded Element->Element adjacency functions. I may have modified this file, but I...
Isaac Dooley [Sat, 9 Apr 2005 22:49:36 +0000 (22:49 +0000)]
Added Element->Element adjacency functions. I may have modified this file, but I'm not sure.

16 years agoAdded Element->Element Adjacency functions.
Isaac Dooley [Sat, 9 Apr 2005 22:32:03 +0000 (22:32 +0000)]
Added Element->Element Adjacency functions.

16 years agoAdded map.h
Isaac Dooley [Sat, 9 Apr 2005 22:26:05 +0000 (22:26 +0000)]
Added map.h

16 years agoRemoved some items from here, putting them in map.h. No fundamental
Isaac Dooley [Sat, 9 Apr 2005 22:23:35 +0000 (22:23 +0000)]
Removed some items from here, putting them in map.h. No fundamental
changes other than this.

16 years agoMoved some items out of map.C that are needed by elem->elem adjacency
Isaac Dooley [Sat, 9 Apr 2005 22:22:56 +0000 (22:22 +0000)]
Moved some items out of map.C that are needed by elem->elem adjacency
computations. The tupleTable class is one of these.

16 years agotweaked header file to use C++ initization instead of assignment.
Eric Bohm [Sat, 9 Apr 2005 20:30:03 +0000 (20:30 +0000)]
tweaked header file to use C++ initization instead of assignment.
Speeds up construction.

16 years agoadded the boundary flag
Sayantan Chakravorty [Sat, 9 Apr 2005 07:53:25 +0000 (07:53 +0000)]
added the boundary flag

16 years agodocumentation for ListSend functions
Filippo Gioachin [Sat, 9 Apr 2005 07:37:42 +0000 (07:37 +0000)]
documentation for ListSend functions

16 years agofix in cleanup
Filippo Gioachin [Sat, 9 Apr 2005 07:10:17 +0000 (07:10 +0000)]
fix in cleanup

16 years agoFixed bug introduced by change of CMK_INCDIR and CMK_LIBDIR with OPTS_CXX
Chee Wai Lee [Sat, 9 Apr 2005 04:01:43 +0000 (04:01 +0000)]
Fixed bug introduced by change of CMK_INCDIR and CMK_LIBDIR with OPTS_CXX
and OPTS_LD respectively. They were originally omitted from the test scripts
in configure.in, preventing the detection of desired system libraries like
PAPI at build time. OPTS_CXX and OPTS_LD are now included alongside the
original CMK_INCDIR and CMK_LIBDIR in configure.in, although I suspect the
latter is not required because CMK_CXX and CMK_LD seem to usually include
CMK_INCDIR and CMK_LIBDIR in their respective machine layers.

The configure.in script has also been fixed to only test for PAPI_TOT_CYC
which applies to all machine architectures. The other test for PAPI_FP_INS
does not apply to the itanium 2 architecture.

Also updated the tracing module such that papiEventSet is initialized to
PAPI_NULL as mandated by the PAPI 3 specification.

16 years agoAdded a local cell/element number field so we can look at individual cells. This...
Isaac Dooley [Sat, 9 Apr 2005 03:48:22 +0000 (03:48 +0000)]
Added a local cell/element number field so we can look at individual cells. This attribute
may not correspond with the original FEM element id's. It is just a sequential numbering from
0 to however many elements exist in each chunk.

----------------------------------------------------------------------

16 years agodeleted % operator (expensive) in favor of a mask operation. This imply that
Filippo Gioachin [Sat, 9 Apr 2005 00:42:04 +0000 (00:42 +0000)]
deleted % operator (expensive) in favor of a mask operation. This imply that
the table is now forced to be power of 2

16 years agoHack: should not break anything else in charm except codes that use PAPI.
Chee Wai Lee [Fri, 8 Apr 2005 23:15:33 +0000 (23:15 +0000)]
Hack: should not break anything else in charm except codes that use PAPI.
Tentatively modified to track PAPI_L3_DCM, PAPI_FP_OPS and PAPI_TOT_INS
for the itanium 2 architecture as a default. A correct implementation would
be a change to the machine layers of each architecture to provide a baseline
"default" for each machine layer.

16 years agoAdded interface to pass new boundary flag on node to FEM framework.
Terry L. Wilmarth [Fri, 8 Apr 2005 22:28:47 +0000 (22:28 +0000)]
Added interface to pass new boundary flag on node to FEM framework.

16 years agoUpdated the fixed node detection given boundary flags.
Terry L. Wilmarth [Fri, 8 Apr 2005 22:12:18 +0000 (22:12 +0000)]
Updated the fixed node detection given boundary flags.

16 years ago*** empty log message ***
Terry L. Wilmarth [Fri, 8 Apr 2005 06:32:53 +0000 (06:32 +0000)]
*** empty log message ***

16 years agoNew locking scheme, works on 1 PE. Still debugging parallel case.
Terry L. Wilmarth [Fri, 8 Apr 2005 05:26:29 +0000 (05:26 +0000)]
New locking scheme, works on 1 PE.  Still debugging parallel case.

16 years agoif it runs without charmrun it needs to be compiled in a sequential (non parallel...
Filippo Gioachin [Thu, 7 Apr 2005 16:52:43 +0000 (16:52 +0000)]
if it runs without charmrun it needs to be compiled in a sequential (non parallel) way

16 years agoTake off Cjacobi3D when I'm messing it up with benchmarking,
Chao Huang [Thu, 7 Apr 2005 16:08:35 +0000 (16:08 +0000)]
Take off Cjacobi3D when I'm messing it up with benchmarking,
will rollback and resume testing when i'm done.

16 years ago*** empty log message ***
Chao Huang [Thu, 7 Apr 2005 05:43:03 +0000 (05:43 +0000)]
*** empty log message ***

16 years ago*** empty log message ***
Chao Huang [Thu, 7 Apr 2005 04:47:41 +0000 (04:47 +0000)]
*** empty log message ***

16 years agoSwitching to new locking scheme in progress. This version does not yet run.
Terry L. Wilmarth [Wed, 6 Apr 2005 23:51:33 +0000 (23:51 +0000)]
Switching to new locking scheme in progress.  This version does not yet run.

16 years agodgemm replacement for zgemm in the backward path.
Eric Bohm [Wed, 6 Apr 2005 22:59:27 +0000 (22:59 +0000)]
dgemm replacement for zgemm in the backward path.
Is currently the default and notably faster for S=grainSize=128.
Set ZG_BACK to return to the previous zgemm implementation.

16 years agofem_mesh.h accessible from refinement
Sayantan Chakravorty [Wed, 6 Apr 2005 22:49:26 +0000 (22:49 +0000)]
fem_mesh.h accessible from refinement

16 years agoPAss in meshid to refinement code.. starts off merger of fem and refine inc
Sayantan Chakravorty [Wed, 6 Apr 2005 22:11:44 +0000 (22:11 +0000)]
PAss in meshid to refinement code.. starts off merger of fem and refine inc

16 years agoadded boundary flag to NewMesh in Fortran interface.
Terry L. Wilmarth [Wed, 6 Apr 2005 16:22:55 +0000 (16:22 +0000)]
added boundary flag to NewMesh in Fortran interface.

16 years agoaddint template LoadLeveler job for submission with charmrun
Filippo Gioachin [Wed, 6 Apr 2005 09:08:29 +0000 (09:08 +0000)]
addint template LoadLeveler job for submission with charmrun

16 years agoCompletely new version of LAPI. Most of the files has been completely rewritten,
Filippo Gioachin [Wed, 6 Apr 2005 08:22:44 +0000 (08:22 +0000)]
Completely new version of LAPI. Most of the files has been completely rewritten,
so a diff with the previous is useless. Code cleanup and optimization have to be
performed. It runs megatest correctly, both in normal and SMP version.

16 years agoprevent deadlock when more than one thread (or the same more than once) is
Filippo Gioachin [Wed, 6 Apr 2005 08:20:45 +0000 (08:20 +0000)]
prevent deadlock when more than one thread (or the same more than once) is
trying to grab the lock of immediate messages. If the grab fails it basically
give up: this means someone is already executing.

16 years agoallow non-smp version to safely include this file. In non-smp, CmiMyNodeSize
Filippo Gioachin [Wed, 6 Apr 2005 08:15:06 +0000 (08:15 +0000)]
allow non-smp version to safely include this file. In non-smp, CmiMyNodeSize
is hardcoded to 1 and the other variable is not defined. In smp they are
hardcoded as the same thing.

16 years agofixed missing module
Filippo Gioachin [Tue, 5 Apr 2005 23:55:15 +0000 (23:55 +0000)]
fixed missing module

16 years agoFixed the strategy to use for alltoall
Chao Huang [Tue, 5 Apr 2005 21:27:39 +0000 (21:27 +0000)]
Fixed the strategy to use for alltoall

16 years agoSayantan: checking in fem_valid fortran definition
Sayantan Chakravorty [Tue, 5 Apr 2005 19:27:33 +0000 (19:27 +0000)]
Sayantan: checking in fem_valid fortran definition

16 years agoupdated for new files added.
Gengbin Zheng [Tue, 5 Apr 2005 16:40:58 +0000 (16:40 +0000)]
updated for new files added.

16 years agoadded a file with some lb utils.
Gengbin Zheng [Tue, 5 Apr 2005 16:40:23 +0000 (16:40 +0000)]
added a file with some lb utils.

16 years agoFixed the DEBUGREF statements and added some new locking code that is not yet used.
Terry L. Wilmarth [Mon, 4 Apr 2005 23:23:04 +0000 (23:23 +0000)]
Fixed the DEBUGREF statements and added some new locking code that is not yet used.

16 years agofixed a compiler error.
Gengbin Zheng [Sun, 3 Apr 2005 20:51:53 +0000 (20:51 +0000)]
fixed a compiler error.

16 years agothe top level strategy now can switch to vector strategy to save memory, which only...
Gengbin Zheng [Sun, 3 Apr 2005 20:16:15 +0000 (20:16 +0000)]
the top level strategy now can switch to vector strategy to save memory, which only collects processor load without detailed object load data. Each group leader is responsible for selecting objects to migrate.

16 years agodefine a different migration message that only specify total load to migrate between...
Gengbin Zheng [Sun, 3 Apr 2005 18:18:44 +0000 (18:18 +0000)]
define a different migration message that only specify total load to migrate between processors but not which object.

16 years agochange CkExit to CkAbort when a unknown load balancer specified.
Gengbin Zheng [Sun, 3 Apr 2005 18:04:02 +0000 (18:04 +0000)]
change CkExit to CkAbort when a unknown load balancer specified.

16 years agofine tuning
Eric Bohm [Sun, 3 Apr 2005 02:16:02 +0000 (02:16 +0000)]
fine tuning

16 years agoslight memory tuning to trim unneeded array of pointers
Eric Bohm [Sun, 3 Apr 2005 01:51:18 +0000 (01:51 +0000)]
slight memory tuning to trim unneeded array of pointers

16 years ago*** empty log message ***
Chao Huang [Sat, 2 Apr 2005 22:44:46 +0000 (22:44 +0000)]
*** empty log message ***

16 years ago*** empty log message ***
Chao Huang [Fri, 1 Apr 2005 21:49:45 +0000 (21:49 +0000)]
*** empty log message ***

16 years agoChanges for the _SECONDPHASE_LOADBAL_ flag. Default is set to on now. It works for
Yan Shi [Fri, 1 Apr 2005 20:58:36 +0000 (20:58 +0000)]
Changes for the _SECONDPHASE_LOADBAL_ flag. Default is set to on now. It works for
non-symm version.

16 years agoChecked in an alltoall benchmark
Chao Huang [Fri, 1 Apr 2005 17:02:57 +0000 (17:02 +0000)]
Checked in an alltoall benchmark

16 years agochange
Sayantan Chakravorty [Fri, 1 Apr 2005 02:42:19 +0000 (02:42 +0000)]
change

16 years agoInhibit warnings by using a -w flag with gcc.
Isaac Dooley [Fri, 1 Apr 2005 01:25:01 +0000 (01:25 +0000)]
Inhibit warnings by using a -w flag with gcc.

16 years agoFixed cell partition/chunk attributes. Was a Float64, now a Int32 as it ought to be.
Isaac Dooley [Fri, 1 Apr 2005 01:24:34 +0000 (01:24 +0000)]
Fixed cell partition/chunk attributes. Was a Float64, now a Int32 as it ought to be.

16 years agoremove VERSION sym link before create it.
Gengbin Zheng [Fri, 1 Apr 2005 00:42:49 +0000 (00:42 +0000)]
remove VERSION sym link before create it.

16 years ago_SECONDPHASE_LOADBAL_ flag: fixed for non-symm case. This part is now load balanced...
Yan Shi [Thu, 31 Mar 2005 23:36:40 +0000 (23:36 +0000)]
_SECONDPHASE_LOADBAL_ flag: fixed for non-symm case. This part is now load balanced on sumparitialresult
calls.

16 years agoFEM_VALID checked in
Sayantan Chakravorty [Thu, 31 Mar 2005 21:07:22 +0000 (21:07 +0000)]
FEM_VALID checked in

16 years agolink VERSION file to top dir.
Gengbin Zheng [Thu, 31 Mar 2005 20:40:49 +0000 (20:40 +0000)]
link VERSION file to top dir.

16 years agoRemoved an unneeded debug print statement.
Isaac Dooley [Thu, 31 Mar 2005 17:13:21 +0000 (17:13 +0000)]
Removed an unneeded debug print statement.

16 years agowhen +LBPrintSummary and +LBSyncResume, take advantage of the reduction to get maximu...
Gengbin Zheng [Thu, 31 Mar 2005 04:30:33 +0000 (04:30 +0000)]
when +LBPrintSummary and +LBSyncResume, take advantage of the reduction to get maximum predicted load on all processors.

16 years agoLatest coarsening code. Parallel works up to about 3 chunks.
Terry L. Wilmarth [Wed, 30 Mar 2005 20:59:47 +0000 (20:59 +0000)]
Latest coarsening code.  Parallel works up to about 3 chunks.

16 years agotweaked a little a debug print for object migration to print tis original obejct...
Gengbin Zheng [Wed, 30 Mar 2005 19:48:08 +0000 (19:48 +0000)]
tweaked a little a debug print for object migration to print tis original obejct *index* instead of the order it is taken out of a heap.

16 years agoA change of some included headers. We need these for OSX
Isaac Dooley [Wed, 30 Mar 2005 18:26:16 +0000 (18:26 +0000)]
A change of some included headers. We need these for OSX

16 years agoprint a sumamry of memory usage by hybrid strategies.
Gengbin Zheng [Tue, 29 Mar 2005 22:33:15 +0000 (22:33 +0000)]
print a sumamry of memory usage by hybrid strategies.

16 years agoa fix for LB dump file version 1 format.
Gengbin Zheng [Tue, 29 Mar 2005 20:06:55 +0000 (20:06 +0000)]
a fix for LB dump file version 1 format.

16 years agoimplement the functionality to print summary of load balancing after calling strategy...
Gengbin Zheng [Tue, 29 Mar 2005 06:34:07 +0000 (06:34 +0000)]
implement the functionality to print summary of load balancing after calling strategy for HybridLB.
also hybridLB can ignore background load upon request

16 years agoadded a note for greedyLB used in hybridLB. The greedy LB needs to be able to take...
Gengbin Zheng [Tue, 29 Mar 2005 06:12:45 +0000 (06:12 +0000)]
added a note for greedyLB used in hybridLB. The greedy LB needs to be able to take a LDStats that has objects with invalid to_pe , representing that the objects are from outside of the processor group.

16 years agoadded new runtime option +LBPrintSummary to print a summary of load balancing that...
Gengbin Zheng [Tue, 29 Mar 2005 06:09:52 +0000 (06:09 +0000)]
added new runtime option +LBPrintSummary to print a summary of load balancing that shows the max load and average representing the quality of load balance.

16 years agochange to allow option +LBNoBackground to ignore background load.
Gengbin Zheng [Tue, 29 Mar 2005 05:57:08 +0000 (05:57 +0000)]
change to allow option +LBNoBackground to ignore background load.

16 years agoChanging more deletes to delete []
Sameer Kumar [Mon, 28 Mar 2005 19:15:48 +0000 (19:15 +0000)]
Changing more deletes to delete []

16 years agoMemory errors
Sameer Kumar [Mon, 28 Mar 2005 06:51:35 +0000 (06:51 +0000)]
Memory errors

16 years agoNeed to delete the elements of hashtable before deleting it.
Sameer Kumar [Mon, 28 Mar 2005 06:51:13 +0000 (06:51 +0000)]
Need to delete the elements of hashtable before deleting it.

16 years agofixed a file dependency that could break parallel make.
Gengbin Zheng [Sun, 27 Mar 2005 16:50:38 +0000 (16:50 +0000)]
fixed a file dependency that could break parallel make.