charm.git
13 years agofixed debug print to a long type.
Gengbin Zheng [Fri, 16 May 2008 06:57:13 +0000 (06:57 +0000)]
fixed debug print to a long type.

13 years agosuport x86_64 and multicore-linux64 and the reordering of darwin
Eric Bohm [Fri, 16 May 2008 02:45:48 +0000 (02:45 +0000)]
suport x86_64 and multicore-linux64 and the reordering of darwin

13 years agoaliasing charm6_0
Gengbin Zheng [Thu, 15 May 2008 21:39:29 +0000 (21:39 +0000)]
aliasing

13 years agorename ampi.h when doing mpi tests
Gengbin Zheng [Thu, 15 May 2008 21:37:17 +0000 (21:37 +0000)]
rename ampi.h when doing mpi tests

13 years agoallows arch name aliasing.
Gengbin Zheng [Thu, 15 May 2008 21:33:40 +0000 (21:33 +0000)]
allows arch name aliasing.

13 years agofixed a compilation error for multicore-cygwin
Gengbin Zheng [Wed, 14 May 2008 03:36:35 +0000 (03:36 +0000)]
fixed a compilation error for multicore-cygwin

13 years agofixed a bug in ++local that drops the last environ var.
Gengbin Zheng [Wed, 14 May 2008 03:09:14 +0000 (03:09 +0000)]
fixed a bug in ++local that drops the last environ var.

13 years agowhen setting Cmi_NodeSize environment (for VMI version?), there is a bug in accessing...
Gengbin Zheng [Tue, 13 May 2008 17:08:57 +0000 (17:08 +0000)]
when setting Cmi_NodeSize environment (for VMI version?), there is a bug in accessing array with out of bound index.

13 years agoNETFEM_init call needs to be initproc instead of initnode because it is meant to...
Gengbin Zheng [Tue, 13 May 2008 04:31:49 +0000 (04:31 +0000)]
NETFEM_init call needs to be initproc instead of initnode because it is meant to be called on each PE.

13 years agoadded a note to CmiBarrier()/
Gengbin Zheng [Tue, 13 May 2008 03:37:38 +0000 (03:37 +0000)]
added a note to CmiBarrier()/

13 years agofixed two subtle bugs:
Gengbin Zheng [Tue, 13 May 2008 03:36:54 +0000 (03:36 +0000)]
fixed two subtle bugs:
1. race condition when using CmiNodeBarrier after CmiNodeAllBarrier. The communication thread may never be able to get out of the barrier.
2. for idle locking, program may hang when acquiring lock, seemingly due to improper setting of timeout.
CS: ----------------------------------------------------------------------

13 years agofixed a subtle race condition when pending BOC and nodegroup messages are handled...
Gengbin Zheng [Tue, 13 May 2008 03:32:00 +0000 (03:32 +0000)]
fixed a subtle race condition when pending BOC and nodegroup messages are handled when a local group is created.

13 years agoavail_vector is a per node array, which cause race condition in smp version. Use...
Gengbin Zheng [Mon, 12 May 2008 05:37:24 +0000 (05:37 +0000)]
avail_vector is a per node array, which cause race condition in smp version. Use initnode call and lock to protect it.

13 years agoanytime migration test
Pritish Jetley [Sun, 11 May 2008 23:41:05 +0000 (23:41 +0000)]
anytime migration test

13 years agoadded a function to clear pending messages queue (set to NULL) for GroupTable.
Gengbin Zheng [Sun, 11 May 2008 04:17:55 +0000 (04:17 +0000)]
added a function to clear pending messages queue (set to NULL) for GroupTable.

13 years agoAdding high-level refinement code. Currently exposes a bug in region locking.
Terry L. Wilmarth [Fri, 9 May 2008 22:10:47 +0000 (22:10 +0000)]
Adding high-level refinement code.  Currently exposes a bug in region locking.

13 years agoautomatically define macro CMK_64BIT when CMK_AMD64 is defined
Gengbin Zheng [Fri, 9 May 2008 18:15:36 +0000 (18:15 +0000)]
automatically define macro CMK_64BIT when CMK_AMD64 is defined

13 years agoset optimized config
Gengbin Zheng [Fri, 9 May 2008 18:13:02 +0000 (18:13 +0000)]
set optimized config

13 years agofixed compiler warnings for gcc 4.2.3
Gengbin Zheng [Fri, 9 May 2008 18:12:00 +0000 (18:12 +0000)]
fixed compiler warnings for gcc 4.2.3

13 years agofix the previous checkin: the locking was too finer grain in the case of broadcast...
Gengbin Zheng [Fri, 9 May 2008 18:09:00 +0000 (18:09 +0000)]
fix the previous checkin: the locking was too finer grain in the case of broadcast. I suspect somewhere in megatest, we rely on the fact that the converse broadcast needs to be ordered.

13 years agoadded missing return statement.
Gengbin Zheng [Fri, 9 May 2008 14:43:04 +0000 (14:43 +0000)]
added missing return statement.

13 years agoignore immeidate msg for QD purpose
Gengbin Zheng [Fri, 9 May 2008 07:31:25 +0000 (07:31 +0000)]
ignore immeidate msg for QD purpose

13 years agoset CMK_USE_TLS_THREAD to 1 by default now. This however is only activated when charm...
Gengbin Zheng [Thu, 8 May 2008 21:02:53 +0000 (21:02 +0000)]
set CMK_USE_TLS_THREAD to 1 by default now. This however is only activated when charm config check if __thread does compile.
Use PCQUEUE_LOCK by default.
Leave the OS malloc default COMMENTED for now. Open the comment to use OS malloc as default.

13 years agomake locking finer grain.
Gengbin Zheng [Thu, 8 May 2008 20:55:49 +0000 (20:55 +0000)]
make locking finer grain.

13 years agofixed compiler warnings for gcc 4.2.3
Gengbin Zheng [Thu, 8 May 2008 07:21:42 +0000 (07:21 +0000)]
fixed compiler warnings for gcc 4.2.3

13 years agomake sure only comm thread calls handleImmediateMsg in smp mode.
Gengbin Zheng [Thu, 8 May 2008 07:11:52 +0000 (07:11 +0000)]
make sure only comm thread calls handleImmediateMsg in smp mode.

13 years agomake sure only comm thread executes immediate message in smp mode.
Gengbin Zheng [Thu, 8 May 2008 06:33:57 +0000 (06:33 +0000)]
make sure only comm thread executes immediate message in smp mode.

13 years agofixed a subtle bug in QD in smp mode. now immediate messages are invisible to QD.
Gengbin Zheng [Thu, 8 May 2008 06:32:56 +0000 (06:32 +0000)]
fixed a subtle bug in QD in smp mode. now immediate messages are invisible to QD.

13 years agotwo init calls changed to initproc, because CpvInitialize needs to be called on every...
Gengbin Zheng [Thu, 8 May 2008 06:16:04 +0000 (06:16 +0000)]
two init calls changed to initproc, because CpvInitialize needs to be called on every rank for SMP.

13 years agorename mpi.h under charm/include when testing native MPI.
Gengbin Zheng [Wed, 7 May 2008 15:46:08 +0000 (15:46 +0000)]
rename mpi.h under charm/include when testing native MPI.

13 years agofix bad delete
Eric Bohm [Wed, 7 May 2008 07:19:13 +0000 (07:19 +0000)]
fix bad delete

13 years agomacro for atomic increment and decrement.
Gengbin Zheng [Wed, 7 May 2008 04:03:12 +0000 (04:03 +0000)]
macro for atomic increment and decrement.

13 years agoNoises off in parallel_part. First bulk_adapt refinement algorithm (almost) ready.
Terry L. Wilmarth [Tue, 6 May 2008 20:29:02 +0000 (20:29 +0000)]
Noises off in parallel_part.  First bulk_adapt refinement algorithm (almost) ready.

13 years agoTurning off noise.
Terry L. Wilmarth [Tue, 6 May 2008 20:26:01 +0000 (20:26 +0000)]
Turning off noise.

13 years agofixed a bug in multiqueue scheme, comm thread has its own queue as well for immediate...
Gengbin Zheng [Tue, 6 May 2008 19:57:46 +0000 (19:57 +0000)]
fixed a bug in multiqueue scheme, comm thread has its own queue as well for immediate message.
fixed a backward compatibility issue with MPI_Init_thread call.

13 years agomistakenly turned memory fence on by default, use pcqueue lock now.
Gengbin Zheng [Tue, 6 May 2008 19:18:53 +0000 (19:18 +0000)]
mistakenly turned memory fence on by default, use pcqueue lock now.

13 years agoRemoved duplicate line for IDLE_POLL
Isaac Dooley [Tue, 6 May 2008 18:21:54 +0000 (18:21 +0000)]
Removed duplicate line for IDLE_POLL

13 years agoNow an error message only prints after 10 unsuccessful calls to shm_open, instead...
Isaac Dooley [Tue, 6 May 2008 18:17:42 +0000 (18:17 +0000)]
Now an error message only prints after 10 unsuccessful calls to shm_open, instead of 1.

13 years agocheck if building charm on MPI, and check if MPI_Init_thread is supported.
Gengbin Zheng [Tue, 6 May 2008 17:10:52 +0000 (17:10 +0000)]
check if building charm on MPI, and check if MPI_Init_thread is supported.

13 years agooops, fixed the error in previous checkin.
Gengbin Zheng [Tue, 6 May 2008 15:32:04 +0000 (15:32 +0000)]
oops, fixed the error in previous checkin.

13 years agorevert the previous change temporarily so that user defined arraymap works again.
Gengbin Zheng [Tue, 6 May 2008 07:14:41 +0000 (07:14 +0000)]
revert the previous change temporarily so that user defined arraymap works again.

13 years agoAdding support for OSSpinLock in the POSIX shared memory layer. Now there are three...
Isaac Dooley [Mon, 5 May 2008 23:31:43 +0000 (23:31 +0000)]
Adding support for OSSpinLock in the POSIX shared memory layer. Now there are three options for synchronization.

13 years agochanged the definition of setMemoryStatus. Now it supports recursive entry method...
Filippo Gioachin [Mon, 5 May 2008 23:06:09 +0000 (23:06 +0000)]
changed the definition of setMemoryStatus. Now it supports recursive entry method calls (like inlined methods)

13 years agofixed memory leak when using CkCallbackResumeThread.
Filippo Gioachin [Mon, 5 May 2008 23:04:43 +0000 (23:04 +0000)]
fixed memory leak when using CkCallbackResumeThread.
Now the incoming message is not retrieved by the user. The only way to retrieve a message from a CkCallbackResumeThread is by passing in a pointer; this pointer will be pointing at the data upon return.
When using directly a CkCallback of CkCallback::resumeThread type, thread_delay can still be used

13 years agofix bug in AllocationTree
Filippo Gioachin [Mon, 5 May 2008 23:01:31 +0000 (23:01 +0000)]
fix bug in AllocationTree

13 years agoswitch to gettimeofday instead of using CmiWallTimer() for MPI.
Gengbin Zheng [Mon, 5 May 2008 20:46:54 +0000 (20:46 +0000)]
switch to gettimeofday instead of using CmiWallTimer() for MPI.
Using MPI call to CmiWallTimer may need lock (when mpich does not support multithreading), which is not sufficient.

13 years agoOptimization for Mac
Isaac Dooley [Mon, 5 May 2008 20:44:11 +0000 (20:44 +0000)]
Optimization for Mac

13 years agoAdding better support for Mac
Isaac Dooley [Mon, 5 May 2008 20:32:46 +0000 (20:32 +0000)]
Adding better support for Mac

13 years agodefine MULTI_SENDQUEUE, when used, use sendmsg queue for each worker thread.
Gengbin Zheng [Mon, 5 May 2008 20:29:12 +0000 (20:29 +0000)]
define MULTI_SENDQUEUE, when used, use sendmsg queue for each worker thread.

Use MPI_Init_thread instead of MPI_Init to retrieve the level the multi-threaded support by the MPI library.

13 years agoadd include of sfftw.h
Eric Bohm [Mon, 5 May 2008 16:08:06 +0000 (16:08 +0000)]
add include of sfftw.h

13 years agoadded a placeholder for CMK_USE_TLS_THREAD
Gengbin Zheng [Sun, 4 May 2008 15:06:42 +0000 (15:06 +0000)]
added a placeholder for CMK_USE_TLS_THREAD

13 years agomove CmiState to __thread when using TLS. This change how worker/comm threads are...
Gengbin Zheng [Sun, 4 May 2008 15:05:33 +0000 (15:05 +0000)]
move CmiState to __thread when using TLS. This change how worker/comm threads are initialized.

13 years agofixed gcc 4.2 compiler warnings.
Gengbin Zheng [Sun, 4 May 2008 14:51:57 +0000 (14:51 +0000)]
fixed gcc 4.2 compiler warnings.

13 years agoget rid of most compiler warnings for gcc 4.2
Gengbin Zheng [Sun, 4 May 2008 14:50:53 +0000 (14:50 +0000)]
get rid of most compiler warnings for gcc 4.2

13 years agoremove Sayantan's change in scheduling using macro definition of Cpv directly, which...
Gengbin Zheng [Sun, 4 May 2008 04:12:18 +0000 (04:12 +0000)]
remove Sayantan's change in scheduling using macro definition of Cpv directly, which does not allow different implementation of Cpv.

13 years agoa complete working version of CpvAccess using __thread. also support CpvAccessOther...
Gengbin Zheng [Sat, 3 May 2008 04:38:17 +0000 (04:38 +0000)]
a complete working version of CpvAccess using __thread. also support CpvAccessOther(). To use this Cpv, __thread must be detected at autoconf time, and user must specify  CMK_USE_TLS_THREAD.

13 years agoFixed a small bug in the fence based code
Sayantan Chakravorty [Fri, 2 May 2008 20:06:51 +0000 (20:06 +0000)]
Fixed a small bug in the fence based code

13 years agoFixed the lock less fence based version
Sayantan Chakravorty [Fri, 2 May 2008 19:18:12 +0000 (19:18 +0000)]
Fixed the lock less fence based version

13 years agoreduced file size
Gengbin Zheng [Fri, 2 May 2008 16:33:50 +0000 (16:33 +0000)]
reduced file size

13 years agoprint MPI version number
Gengbin Zheng [Fri, 2 May 2008 14:44:33 +0000 (14:44 +0000)]
print MPI version number

13 years agoFence-based synchronization code added.
Esteban Meneses [Fri, 2 May 2008 03:11:38 +0000 (03:11 +0000)]
Fence-based synchronization code added.

13 years agoskip "sync" call, as it seems to hang cobalt.ncsa.
Gengbin Zheng [Thu, 1 May 2008 22:17:03 +0000 (22:17 +0000)]
skip "sync" call, as it seems to hang cobalt.ncsa.

13 years agorecording messages allocated with CkCopyMsg as messages in charmdebug
Filippo Gioachin [Wed, 30 Apr 2008 15:24:07 +0000 (15:24 +0000)]
recording messages allocated with CkCopyMsg as messages in charmdebug

13 years agochange it so that it does not make parfum libs every time typing make, this is needed...
Gengbin Zheng [Wed, 30 Apr 2008 07:03:13 +0000 (07:03 +0000)]
change it so that it does not make parfum libs every time typing make, this is needed for autobuild when libs have to be built on frontend node instead of on compute nodes.

13 years agoChanges to Physics part by Sanjay.
Kumaresh P [Wed, 30 Apr 2008 05:13:26 +0000 (05:13 +0000)]
Changes to Physics part by Sanjay.
Working Demo!

13 years agoupdate CmiBarrier to match changes in converse.h
Eric Bohm [Wed, 30 Apr 2008 03:47:45 +0000 (03:47 +0000)]
update CmiBarrier to match changes in converse.h

13 years agoported 64 bit version to new turing linux partition.
Gengbin Zheng [Tue, 29 Apr 2008 20:29:20 +0000 (20:29 +0000)]
ported 64 bit version to new turing linux partition.

13 years agoadded test for __thread
Gengbin Zheng [Tue, 29 Apr 2008 19:18:33 +0000 (19:18 +0000)]
added test for __thread

13 years agofixed problem when using CMK_OPTIMIZE
Filippo Gioachin [Tue, 29 Apr 2008 19:02:50 +0000 (19:02 +0000)]
fixed problem when using CMK_OPTIMIZE

13 years agofixed include problem
Filippo Gioachin [Tue, 29 Apr 2008 18:51:51 +0000 (18:51 +0000)]
fixed include problem

13 years agoAdd CmiDirect_readyMark and CmiDirect_readyPollq as split versions of CmiDirect_ready
Eric Bohm [Mon, 28 Apr 2008 22:05:03 +0000 (22:05 +0000)]
Add CmiDirect_readyMark and CmiDirect_readyPollq as split versions of CmiDirect_ready

13 years agoadd split version of CmiDirect_ready -> CmiDirect_readyMark and CmiDirect_readyPollQ
Eric Bohm [Mon, 28 Apr 2008 22:04:03 +0000 (22:04 +0000)]
add split version of CmiDirect_ready -> CmiDirect_readyMark and CmiDirect_readyPollQ
which mark the OOB byte pattern, and insert the handle into the polling queue respectively.
This allows the marking to be done as early as is valid,  while allowing us to shift the
polling to as late as is sensible.  It also erases a race condition wherein puts could
occur before a late ready call was made and not be detected.  Now the put can occur any time
after Mark, and be detected by readyPollQ.

13 years agoput latency test
Pritish Jetley [Mon, 28 Apr 2008 22:03:49 +0000 (22:03 +0000)]
put latency test

13 years agocheck
Sayantan Chakravorty [Fri, 25 Apr 2008 20:05:04 +0000 (20:05 +0000)]
check

13 years agoadded ParFUM dependency on msa, added ParFUM-Tops to LIBS
Aaron Becker [Fri, 25 Apr 2008 19:35:15 +0000 (19:35 +0000)]
added ParFUM dependency on msa, added ParFUM-Tops to LIBS

13 years ago*** empty log message ***
Abhinav Bhatele [Fri, 25 Apr 2008 02:07:54 +0000 (02:07 +0000)]
*** empty log message ***

13 years agofixed missing return statements
Filippo Gioachin [Thu, 24 Apr 2008 22:17:08 +0000 (22:17 +0000)]
fixed missing return statements

13 years agodgemm support
Abhinav Bhatele [Thu, 24 Apr 2008 18:37:19 +0000 (18:37 +0000)]
dgemm support

13 years agoiterative version
Abhinav Bhatele [Thu, 24 Apr 2008 18:34:24 +0000 (18:34 +0000)]
iterative version

13 years ago*** empty log message ***
Abhinav Bhatele [Thu, 24 Apr 2008 18:28:49 +0000 (18:28 +0000)]
*** empty log message ***

13 years agosynced with the CkDirect version
Abhinav Bhatele [Thu, 24 Apr 2008 18:13:07 +0000 (18:13 +0000)]
synced with the CkDirect version

13 years agocleaning up of code
Abhinav Bhatele [Thu, 24 Apr 2008 18:05:55 +0000 (18:05 +0000)]
cleaning up of code

13 years agofixed the previous checkin for uJcontext
Gengbin Zheng [Thu, 24 Apr 2008 07:32:14 +0000 (07:32 +0000)]
fixed the previous checkin for uJcontext

13 years agoReady for demo - Final MD Example
Kumaresh P [Thu, 24 Apr 2008 07:03:16 +0000 (07:03 +0000)]
Ready for demo - Final MD Example
Functionality split into many files

13 years agoan important fix for context version of thread implementation for amd64.
Gengbin Zheng [Thu, 24 Apr 2008 07:02:51 +0000 (07:02 +0000)]
an important fix for context version of thread implementation for amd64.
makecontext call only takes INTEGER arguments, instead we pass pointers. had to fix this to allow makecontext to work with memory os.

13 years agomissed a space
Filippo Gioachin [Thu, 24 Apr 2008 03:10:19 +0000 (03:10 +0000)]
missed a space

13 years agoMD Examples - Particles assigned initial velocity
Kumaresh P [Thu, 24 Apr 2008 03:07:44 +0000 (03:07 +0000)]
MD Examples - Particles assigned initial velocity

13 years agoAdding a 64-bit version for Intel Macs.
Isaac Dooley [Thu, 24 Apr 2008 03:06:26 +0000 (03:06 +0000)]
Adding a 64-bit version for Intel Macs.

13 years agoAdding routines to compute crc checksums on memory regions.
Filippo Gioachin [Thu, 24 Apr 2008 02:20:07 +0000 (02:20 +0000)]
Adding routines to compute crc checksums on memory regions.

Fixed bug in charmxi for debugging puppers, when pupping parameter marshalled arrays.
Adding compile time option to charmxi generated files to compute and check CRC fields for all memory allocation. Works only in conjuction with "-memory charmdebug".
Modified definition of "user allocated memory". Before it was all the memory allocated inside an entry method, now it only refers to user-defined entry methods.

13 years agoWrapping memory-os with the other memory libraries available: paranoid, charmdebug...
Filippo Gioachin [Thu, 24 Apr 2008 02:15:37 +0000 (02:15 +0000)]
Wrapping memory-os with the other memory libraries available: paranoid, charmdebug, verbose, leak, isomalloc.
The new libraries are called libmemory-os-<wrapper_library>
An extra file and library, libmemory-os-wrapper, is compiled to provide runtime support to this feature. charmc automatically includes this library if the user requests a library wrapping the OS malloc (like "-memory os-leak"

Charm build process does not depend on these libraries to successfully build.

13 years agoadding support for CRC checked memory.
Filippo Gioachin [Thu, 24 Apr 2008 02:10:50 +0000 (02:10 +0000)]
adding support for CRC checked memory.
added a few pointer casts.
fixed problem when collecting stack traces, it crashed if there was not a single starting function for all stacks. this is not true when user-level threads are present.

13 years agomodified definition of "user allocated memory". Before it referred to the memory...
Filippo Gioachin [Thu, 24 Apr 2008 02:08:21 +0000 (02:08 +0000)]
modified definition of "user allocated memory". Before it referred to the memory allocate while delivering a message. Now it refers to memory allocated while delivering messages to user-defined entry methods.

13 years agoMolecular Dynamics Example - With LiveViz
Kumaresh P [Wed, 23 Apr 2008 20:56:59 +0000 (20:56 +0000)]
Molecular Dynamics Example - With LiveViz

13 years agoa dummy implementation of CmiBarrier()
Gengbin Zheng [Wed, 23 Apr 2008 18:15:17 +0000 (18:15 +0000)]
a dummy implementation of CmiBarrier()

13 years agoMulti-Step working - NAN needs to be fixed
Kumaresh P [Wed, 23 Apr 2008 08:51:52 +0000 (08:51 +0000)]
Multi-Step working - NAN needs to be fixed

13 years agoon some machines, isomalloc memory regions on different nodes
Gengbin Zheng [Wed, 23 Apr 2008 04:33:14 +0000 (04:33 +0000)]
on some machines, isomalloc memory regions on different nodes
can be different. use +isomalloc_sync to calculate the
intersect of all memory regions on all nodes.
This simple-minded implementation is to use file (on NFS) to calculate the intersect across all processors.

13 years agofinally a udp version of CmiBarrier() implemented.
Gengbin Zheng [Wed, 23 Apr 2008 04:26:57 +0000 (04:26 +0000)]
finally a udp version of CmiBarrier() implemented.
However, this only works when +netpoll, this is because otherwise an interrupt-based implementation is tricky.

13 years agoCmiBarrier() now returns a status to tell if it is implemnented.
Gengbin Zheng [Wed, 23 Apr 2008 04:23:34 +0000 (04:23 +0000)]
CmiBarrier() now returns a status to tell if it is implemnented.

13 years agoCmiBarrier() now returns a status.
Gengbin Zheng [Wed, 23 Apr 2008 04:22:45 +0000 (04:22 +0000)]
CmiBarrier() now returns a status.

13 years agoMD Version - Fixed a Bug.
Kumaresh P [Wed, 23 Apr 2008 01:32:12 +0000 (01:32 +0000)]
MD Version - Fixed a Bug.