charm.git
16 years agorouterstrategy.C uses static definitions of AllToAllHdr, it should instead use
Filippo Gioachin [Mon, 25 Apr 2005 20:16:45 +0000 (20:16 +0000)]
routerstrategy.C uses static definitions of AllToAllHdr, it should instead use
the one in petable.h --> To be fixed

16 years agotemporary fix, until elan machine layer is fixed. Now vectorize is disabled
Filippo Gioachin [Mon, 25 Apr 2005 19:33:16 +0000 (19:33 +0000)]
temporary fix, until elan machine layer is fixed. Now vectorize is disabled

16 years agoproper changes for inline call
Eric Bohm [Mon, 25 Apr 2005 10:15:50 +0000 (10:15 +0000)]
proper changes for inline call
and for partial summing symmetric in gspace

16 years agodon't try to print a non simple type
Eric Bohm [Mon, 25 Apr 2005 10:04:08 +0000 (10:04 +0000)]
don't try to print a non simple type

16 years agoshift the sumpartial for symmetric to gspace acceptpsi.
Eric Bohm [Mon, 25 Apr 2005 09:37:19 +0000 (09:37 +0000)]
shift the sumpartial for symmetric to gspace acceptpsi.
convert the broadcastentire result to using inline messages
so they can be detected by load balancing

16 years agofor pthread version, call pthread_attr_setstacksize() to set stack size.
Gengbin Zheng [Mon, 25 Apr 2005 01:23:08 +0000 (01:23 +0000)]
for pthread version, call pthread_attr_setstacksize() to set stack size.

16 years agoUncommented the line that drains the Send Queue while the getQPSendBuffer is looping.
Isaac Dooley [Sun, 24 Apr 2005 21:22:17 +0000 (21:22 +0000)]
Uncommented the line that drains the Send Queue while the getQPSendBuffer is looping.

16 years agoChecking in all the changes that were needed to change the machine layer to use RDMA...
David Kunzman [Sun, 24 Apr 2005 20:54:38 +0000 (20:54 +0000)]
Checking in all the changes that were needed to change the machine layer to use RDMA instread of the Sends and Receives.

16 years agoAdded some data members to the OtherNode structure to support our RMDA changes.
David Kunzman [Sun, 24 Apr 2005 20:53:34 +0000 (20:53 +0000)]
Added some data members to the OtherNode structure to support our RMDA changes.

16 years agoChanging the converse version elan flag.
Sameer Kumar [Sun, 24 Apr 2005 06:04:10 +0000 (06:04 +0000)]
Changing the converse version elan flag.

16 years agonothing significant
Tarun Agarwal [Sun, 24 Apr 2005 03:56:42 +0000 (03:56 +0000)]
nothing significant

16 years agoAdded getHopBytes() for performance stats
Tarun Agarwal [Sun, 24 Apr 2005 03:53:49 +0000 (03:53 +0000)]
Added getHopBytes() for performance stats

16 years agoSwitching broadcast strategy to hypercube. Will be faster on larger number
Sameer Kumar [Sun, 24 Apr 2005 03:24:02 +0000 (03:24 +0000)]
Switching broadcast strategy to hypercube. Will be faster on larger number
of processors.

16 years agoFixed a bug where "fixed" status of a node was not properly updated across chunks.
Terry L. Wilmarth [Sat, 23 Apr 2005 17:11:31 +0000 (17:11 +0000)]
Fixed a bug where "fixed" status of a node was not properly updated across chunks.

16 years agoReports corners.
Terry L. Wilmarth [Sat, 23 Apr 2005 16:26:18 +0000 (16:26 +0000)]
Reports corners.

16 years agoFixed a corner detecting bug.
Terry L. Wilmarth [Sat, 23 Apr 2005 16:23:30 +0000 (16:23 +0000)]
Fixed a corner detecting bug.

16 years agoforgot ci file
Filippo Gioachin [Sat, 23 Apr 2005 02:31:59 +0000 (02:31 +0000)]
forgot ci file

16 years agonew converse strategy for message pipelining (point-to-point communication)
Filippo Gioachin [Sat, 23 Apr 2005 00:09:52 +0000 (00:09 +0000)]
new converse strategy for message pipelining (point-to-point communication)

16 years agoUses BFS function LBTopology::get_pairwise_hop_count() now.
Tarun Agarwal [Fri, 22 Apr 2005 23:55:14 +0000 (23:55 +0000)]
Uses BFS function LBTopology::get_pairwise_hop_count() now.
Initializing of pairwise distance is O(nE).

16 years agoAdded get_pairwise_hop_count(double **distance) function.
Tarun Agarwal [Fri, 22 Apr 2005 23:53:09 +0000 (23:53 +0000)]
Added get_pairwise_hop_count(double **distance) function.

16 years agoTuring off prints.
Sameer Kumar [Fri, 22 Apr 2005 22:46:22 +0000 (22:46 +0000)]
Turing off prints.

16 years agobe sure that usesatsync is set regardless
Eric Bohm [Fri, 22 Apr 2005 21:13:12 +0000 (21:13 +0000)]
be sure that usesatsync is set regardless

16 years agoruntime controls for atsync activation
Eric Bohm [Fri, 22 Apr 2005 18:36:34 +0000 (18:36 +0000)]
runtime controls for atsync activation

16 years agoPrinting both received and reported boundaries on nodes.
Terry L. Wilmarth [Fri, 22 Apr 2005 17:01:12 +0000 (17:01 +0000)]
Printing both received and reported boundaries on nodes.

16 years agoPrinting node boundary changes reported to FEM framework.
Terry L. Wilmarth [Fri, 22 Apr 2005 16:57:24 +0000 (16:57 +0000)]
Printing node boundary changes reported to FEM framework.

16 years agoCode spiffying.
Terry L. Wilmarth [Fri, 22 Apr 2005 16:04:47 +0000 (16:04 +0000)]
Code spiffying.

16 years agoAdding pumpmsgs and release sent messages to CmiGetNonlocal whether messages areceive...
Sameer Kumar [Fri, 22 Apr 2005 15:22:35 +0000 (15:22 +0000)]
Adding pumpmsgs and release sent messages to CmiGetNonlocal whether messages areceived or not. Dramatically increases performance on the altix. Should also help bluegene, IBM SP and many other versions. Fine grained applications should minimize this poll frequency through the use of +csdLocalMax.

16 years agoBug fix in the ring building algorithms
Sameer Kumar [Fri, 22 Apr 2005 13:37:19 +0000 (13:37 +0000)]
Bug fix in the ring building algorithms

16 years agoremoved --allow-multiple-definition which does not appear to work for older redhat...
Gengbin Zheng [Fri, 22 Apr 2005 01:14:55 +0000 (01:14 +0000)]
removed --allow-multiple-definition which does not appear to work for older redhat. Let configure script test it.

16 years agocheck if linker command really works.
Gengbin Zheng [Fri, 22 Apr 2005 01:14:12 +0000 (01:14 +0000)]
check if linker command really works.
for net-linux, check if --allow-multiple-definition works or not.

16 years agoAdding new strategy into comlib.
Sameer Kumar [Thu, 21 Apr 2005 19:26:54 +0000 (19:26 +0000)]
Adding new strategy into comlib.

16 years agoNew comlib strategies for CPMD. Propagates faster than a ring and is largely contenti...
Sameer Kumar [Thu, 21 Apr 2005 19:23:01 +0000 (19:23 +0000)]
New comlib strategies for CPMD. Propagates faster than a ring and is largely contention free.

16 years agorecover of wrong flag checked it
Filippo Gioachin [Thu, 21 Apr 2005 02:03:33 +0000 (02:03 +0000)]
recover of wrong flag checked it

16 years agosupport load balancing
Eric Bohm [Wed, 20 Apr 2005 21:33:52 +0000 (21:33 +0000)]
support load balancing

16 years agofix of the last checking, (together with the uncommented checkins)
Filippo Gioachin [Wed, 20 Apr 2005 20:50:05 +0000 (20:50 +0000)]
fix of the last checking, (together with the uncommented checkins)

16 years ago*** empty log message ***
Filippo Gioachin [Wed, 20 Apr 2005 19:01:34 +0000 (19:01 +0000)]
*** empty log message ***

16 years ago*** empty log message ***
Filippo Gioachin [Wed, 20 Apr 2005 18:42:09 +0000 (18:42 +0000)]
*** empty log message ***

16 years ago*** empty log message ***
Filippo Gioachin [Wed, 20 Apr 2005 18:42:09 +0000 (18:42 +0000)]
*** empty log message ***

16 years agoMinor corrections in comments
Tarun Agarwal [Wed, 20 Apr 2005 18:38:08 +0000 (18:38 +0000)]
Minor corrections in comments

16 years agonew implementation of AllToAll routines. now an option is present to enable
Filippo Gioachin [Wed, 20 Apr 2005 18:29:21 +0000 (18:29 +0000)]
new implementation of AllToAll routines. now an option is present to enable
the zero-copy functions: CMK_COMLIB_USE_VECTORIZE

16 years agomodified the CmiVectorSend functions, now a negative number means that the packets...
Filippo Gioachin [Wed, 20 Apr 2005 18:28:22 +0000 (18:28 +0000)]
modified the CmiVectorSend functions, now a negative number means that the packets need to be padded.

16 years agoAdded interface and code to keep track of edgeBoundaries in the refinement framework.
Terry L. Wilmarth [Wed, 20 Apr 2005 17:00:13 +0000 (17:00 +0000)]
Added interface and code to keep track of edgeBoundaries in the refinement framework.

16 years agoAdditions for TopoLB
Tarun Agarwal [Wed, 20 Apr 2005 02:52:09 +0000 (02:52 +0000)]
Additions for TopoLB

16 years agoAdding TopoLB: a new Topology-Aware loadbalancer
Tarun Agarwal [Wed, 20 Apr 2005 02:45:01 +0000 (02:45 +0000)]
Adding TopoLB: a new Topology-Aware loadbalancer

16 years agoNothing significant.
Terry L. Wilmarth [Wed, 20 Apr 2005 00:10:13 +0000 (00:10 +0000)]
Nothing significant.

16 years agoActually, if I make sure every element reaches target area properly, the element...
Terry L. Wilmarth [Tue, 19 Apr 2005 23:43:44 +0000 (23:43 +0000)]
Actually, if I make sure every element reaches target area properly, the element quality is not so bad.

16 years agoMany bug fixes to both refinement and coarsening to get the mesh to fully refine...
Terry L. Wilmarth [Tue, 19 Apr 2005 23:32:30 +0000 (23:32 +0000)]
Many bug fixes to both refinement and coarsening to get the mesh to fully refine and follow up with a full coarsening.  Works, exposes the fact that shortest edge collapse is limited in its ability to improve mesh quality.  Need to look at angles instead perhaps.  Tried a second pass, but refinement following the coarsening is much less happy.  Crashes in fenrefine.C.

16 years agoremoved rehash for fastq
Nilesh Choudhury [Tue, 19 Apr 2005 23:24:29 +0000 (23:24 +0000)]
removed rehash for fastq

16 years agoFixed the #elif for the AMMASSO_STATS macros.
David Kunzman [Tue, 19 Apr 2005 21:07:59 +0000 (21:07 +0000)]
Fixed the #elif for the AMMASSO_STATS macros.

16 years agoChange data types for send_UseCounter and send_InUseCounter from char to int.
David Kunzman [Tue, 19 Apr 2005 21:01:53 +0000 (21:01 +0000)]
Change data types for send_UseCounter and send_InUseCounter from char to int.

16 years agoAdded more stat gathering code.
David Kunzman [Tue, 19 Apr 2005 20:30:27 +0000 (20:30 +0000)]
Added more stat gathering code.

16 years agoA bug concerning makeCommHash has been fixed
Amit Sharma [Tue, 19 Apr 2005 20:02:04 +0000 (20:02 +0000)]
A bug concerning makeCommHash has been fixed

16 years agobugfix
Filippo Gioachin [Tue, 19 Apr 2005 19:17:45 +0000 (19:17 +0000)]
bugfix

16 years agofixed problem with pgm-remesh. it was passing the wrong iteration number
Sayantan Chakravorty [Tue, 19 Apr 2005 17:42:34 +0000 (17:42 +0000)]
fixed problem with pgm-remesh. it was passing the wrong iteration number
to Netfem_begin

16 years agowrong /www directory!
Filippo Gioachin [Tue, 19 Apr 2005 17:21:52 +0000 (17:21 +0000)]
wrong /www directory!

16 years agoAdded a more user friendly paragraph about using Poll mode.
Isaac Dooley [Tue, 19 Apr 2005 17:06:35 +0000 (17:06 +0000)]
Added a more user friendly paragraph about using Poll mode.

16 years agoAdded a comment that serverpush is currently unused. Filippo and I found that the...
Isaac Dooley [Tue, 19 Apr 2005 16:35:29 +0000 (16:35 +0000)]
Added a comment that serverpush is currently unused. Filippo and I found that the flag in
the config is not ever used in the server. The get method that would return the flag is
never called.

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

16 years agoAdded sorted list of elements to coarsen to the chunk. This speeds coarsening up...
Terry L. Wilmarth [Tue, 19 Apr 2005 15:01:56 +0000 (15:01 +0000)]
Added sorted list of elements to coarsen to the chunk.  This speeds coarsening up dramatically, in spite of using a silly bubble sort.  The sort is only needed oncewhen multipleCoarsen is called from the FEM framework, and it only sorts the elements that need to be coarsened, via a "sinking bubble" from the top of the stack.  The stack is freed and reallocated before each coarsening phase, so it never has space for more elements than will be needed.

16 years agoguarded the new code with CMK_SHARED_VARS_UNAVAILABLE.
Gengbin Zheng [Tue, 19 Apr 2005 14:48:47 +0000 (14:48 +0000)]
guarded the new code with CMK_SHARED_VARS_UNAVAILABLE.

16 years agoAdded statistics code to help understand where the code is spending its time. Also...
David Kunzman [Tue, 19 Apr 2005 05:43:19 +0000 (05:43 +0000)]
Added statistics code to help understand where the code is spending its time.  Also, added some calls to CompletionEventHandler() so it is used more
for polling the completion queue rather than as a callback (though I'm not sure that the callback nature is entirely gone yet).

16 years agoMigration and load balancing work for the paircalculator now.
Eric Bohm [Tue, 19 Apr 2005 04:40:57 +0000 (04:40 +0000)]
Migration and load balancing work for the paircalculator now.

16 years agoPassing comm list modifications to FEM. Also some bug fixes.
Terry L. Wilmarth [Tue, 19 Apr 2005 03:25:34 +0000 (03:25 +0000)]
Passing comm list modifications to FEM.  Also some bug fixes.

16 years agodefault to no load balancing
Eric Bohm [Tue, 19 Apr 2005 00:22:34 +0000 (00:22 +0000)]
default to no load balancing

16 years agoload balancing changes
Eric Bohm [Tue, 19 Apr 2005 00:11:04 +0000 (00:11 +0000)]
load balancing changes

16 years agoin netpoll mode, when calling GetNonLocal() and with NULL msg in the queue,
Gengbin Zheng [Mon, 18 Apr 2005 21:07:42 +0000 (21:07 +0000)]
in netpoll mode, when calling GetNonLocal() and with NULL msg in the queue,
call communicationServer() to poll network.

16 years agoFixed PUP. Flipped flag for usesAtSync
Yan Shi [Mon, 18 Apr 2005 20:50:05 +0000 (20:50 +0000)]
Fixed PUP. Flipped flag for usesAtSync

16 years agoModified send function for messages of length greater than AMMASSO_BUFSIZE.
David Kunzman [Mon, 18 Apr 2005 16:57:37 +0000 (16:57 +0000)]
Modified send function for messages of length greater than AMMASSO_BUFSIZE.

16 years agoupdated syntax of the AllToAll message, in order to prepare for the two new
Filippo Gioachin [Mon, 18 Apr 2005 09:12:00 +0000 (09:12 +0000)]
updated syntax of the AllToAll message, in order to prepare for the two new
zero-copy routines

16 years agoAdding persistent messages to ring
Sameer Kumar [Mon, 18 Apr 2005 02:54:48 +0000 (02:54 +0000)]
Adding persistent messages to ring

16 years agoAdded a couple of locks in the OtherNode structures for the sequence number members...
David Kunzman [Mon, 18 Apr 2005 01:17:38 +0000 (01:17 +0000)]
Added a couple of locks in the OtherNode structures for the sequence number members (send_next and recv_expect) for the Ammasso code.

16 years agoModified several lines including getQPSendBuffer() so some code is included in a...
David Kunzman [Mon, 18 Apr 2005 01:16:21 +0000 (01:16 +0000)]
Modified several lines including getQPSendBuffer() so some code is included in a lock that it was not included in before.

16 years agofix for CmiMultipleSend
Filippo Gioachin [Sun, 17 Apr 2005 20:51:16 +0000 (20:51 +0000)]
fix for CmiMultipleSend

16 years agonon gcc compiler want a return value even for useless functions...
Filippo Gioachin [Sun, 17 Apr 2005 19:41:09 +0000 (19:41 +0000)]
non gcc compiler want a return value even for useless functions...

16 years agofixed alignment problem
Filippo Gioachin [Sun, 17 Apr 2005 19:39:27 +0000 (19:39 +0000)]
fixed alignment problem

16 years agoupdated.
Gengbin Zheng [Sat, 16 Apr 2005 19:26:45 +0000 (19:26 +0000)]
updated.

16 years agoa new LB from Amit that load balancing each array separately.
Gengbin Zheng [Sat, 16 Apr 2005 19:26:09 +0000 (19:26 +0000)]
a new LB from Amit that load balancing each array separately.

16 years agoFixed the symmetric version hang.
Yan Shi [Sat, 16 Apr 2005 17:45:37 +0000 (17:45 +0000)]
Fixed the symmetric version hang.

16 years agosection multicast to both sides
Eric Bohm [Sat, 16 Apr 2005 16:37:34 +0000 (16:37 +0000)]
section multicast to both sides

16 years agoa little cleanup and conformation of usage of CmiChunkHeader. It will prevent in
Filippo Gioachin [Sat, 16 Apr 2005 08:15:20 +0000 (08:15 +0000)]
a little cleanup and conformation of usage of CmiChunkHeader. It will prevent in
future breaks of charm if this header is modified.
Moreover, fixed a bug (memory leak) for which a vector was not emptied at the
end of an iteration. This have performance impact since every insert will move all
the data present in the array many times.

16 years agomodified the interface, added a new possibility of Cmi VectorSends to pack inside
Filippo Gioachin [Sat, 16 Apr 2005 08:12:10 +0000 (08:12 +0000)]
modified the interface, added a new possibility of Cmi VectorSends to pack inside
the message also the CmiChunkHeader allocated with every CmiAlloc

16 years agoadded a small functionality of removeAll
Filippo Gioachin [Sat, 16 Apr 2005 08:09:05 +0000 (08:09 +0000)]
added a small functionality of removeAll

16 years agoproperly working messages for calculatePairs
Eric Bohm [Sat, 16 Apr 2005 03:33:34 +0000 (03:33 +0000)]
properly working messages for calculatePairs

16 years agoconvert calculatePairs to messages
Eric Bohm [Fri, 15 Apr 2005 23:29:38 +0000 (23:29 +0000)]
convert calculatePairs to messages

16 years agoMade several minor changes.
David Kunzman [Fri, 15 Apr 2005 22:39:51 +0000 (22:39 +0000)]
Made several minor changes.

16 years agoAdded ACKs control messages.
David Kunzman [Fri, 15 Apr 2005 20:30:34 +0000 (20:30 +0000)]
Added ACKs control messages.

16 years agoAdded ACK control messages. Now each connection can have AMMASSO_NUMMSGBUFS_PER_QP...
David Kunzman [Fri, 15 Apr 2005 20:22:01 +0000 (20:22 +0000)]
Added ACK control messages.  Now each connection can have AMMASSO_NUMMSGBUFS_PER_QP messages in-flight per QP connection.

16 years agochanges to support load balancing
Eric Bohm [Fri, 15 Apr 2005 19:09:27 +0000 (19:09 +0000)]
changes to support load balancing

16 years agochanges to support load balancing
Eric Bohm [Fri, 15 Apr 2005 02:18:38 +0000 (02:18 +0000)]
changes to support load balancing

16 years agominor changes, removing dead code.
Gengbin Zheng [Thu, 14 Apr 2005 22:18:53 +0000 (22:18 +0000)]
minor changes, removing dead code.

16 years agoFixed compilation bug.
Terry L. Wilmarth [Thu, 14 Apr 2005 19:56:51 +0000 (19:56 +0000)]
Fixed compilation bug.

16 years agoadding verbose CkVec::pushback_v which returns the position so we can
Eric Bohm [Thu, 14 Apr 2005 19:20:07 +0000 (19:20 +0000)]
adding verbose CkVec::pushback_v which returns the position so we can
use that in a subsequent remove call.

16 years agoFixed a bug in the coarsening code, now both refinement and coarsening run for at...
Terry L. Wilmarth [Thu, 14 Apr 2005 16:33:16 +0000 (16:33 +0000)]
Fixed a bug in the coarsening code, now both refinement and coarsening run for at least a single processor.  Adding a remeshing program whose first goal is to test if refinement and coarsening can work together.  So far, no.  The remeshing program works on a stretched out mesh, so ultimately, it will show us if combinations of refinement and coarsening can be used to improve mesh quality, thereby acting as a parallel remeshing tool.

16 years agoFinished element_bisect. Given this is sequential code, and that edge_bisect was...
Terry L. Wilmarth [Thu, 14 Apr 2005 15:51:15 +0000 (15:51 +0000)]
Finished element_bisect.  Given this is sequential code, and that edge_bisect was already done, a recursive algorithm for this turned out to be remarkably easy.

16 years agodisabling deletion of persistent section data.
Sameer Kumar [Thu, 14 Apr 2005 02:39:45 +0000 (02:39 +0000)]
disabling deletion of persistent section data.

16 years agoFEM_Adapt code added and built in FEM Makefile. Currently it builds on net-linux...
Terry L. Wilmarth [Thu, 14 Apr 2005 00:31:25 +0000 (00:31 +0000)]
FEM_Adapt code added and built in FEM Makefile.  Currently it builds on net-linux. It is not used anywhere.

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