charm.git
9 years agoMake event id to be macro defined, and warned that those user event ids should
Chao Mei [Thu, 1 Mar 2012 17:31:11 +0000 (11:31 -0600)]
Make event id to be macro defined, and warned that those user event ids should
be globally unique.

9 years agoMinor fix to get rid of warnings
Chao Mei [Wed, 29 Feb 2012 17:46:24 +0000 (11:46 -0600)]
Minor fix to get rid of warnings

9 years agoBetter support for affinity setting in the pthread version.
Chao Mei [Wed, 29 Feb 2012 17:41:56 +0000 (11:41 -0600)]
Better support for affinity setting in the pthread version.

9 years agoAdded the initial support of using pthreads for parallelization (a special SMP mode...
Chao Mei [Wed, 29 Feb 2012 05:11:55 +0000 (23:11 -0600)]
Added the initial support of using pthreads for parallelization (a special SMP mode for charm++)
There's one problem that remains as the cpu affinity setting for those pthreads

9 years agoMoved the registration of converse handlers to initproc function for consistency...
Chao Mei [Mon, 27 Feb 2012 23:21:15 +0000 (17:21 -0600)]
Moved the registration of converse handlers to initproc function for consistency. It's important the same handler value on every proc corresponds to the same handler function! If the application itself has new handlers to register, then the order between registering the nodehelper handlers and the app handlers has to be consistent across all procs. Otherwise, it will lead to wierd bug.

9 years agoAdded msg buffer to allow the combination of unsync and sync parallelization
Chao Mei [Sat, 25 Feb 2012 22:14:09 +0000 (16:14 -0600)]
Added msg buffer to allow the combination of unsync and sync parallelization
tasks

9 years agoReset "inited" in the unsync mode so as to avoid expensive atomic operation by
Chao Mei [Fri, 24 Feb 2012 02:28:53 +0000 (20:28 -0600)]
Reset "inited" in the unsync mode so as to avoid expensive atomic operation by
helpers that are not able to do the work in time.

9 years agoAdded a parameter to indicate whether the loop needs to be synced after parallelizati...
Chao Mei [Fri, 24 Feb 2012 02:04:36 +0000 (20:04 -0600)]
Added a parameter to indicate whether the loop needs to be synced after parallelization (i.e. whether the main thread could continue unless all partions have finished)

9 years ago1. Removed unnecessary assignment
Chao Mei [Thu, 23 Feb 2012 07:48:33 +0000 (01:48 -0600)]
1. Removed unnecessary assignment
2. Add a comment for rare data-racing case

9 years agoA better spin-waiting scheme. Using compare_and_set will cause bus-broadcast
Chao Mei [Thu, 23 Feb 2012 07:05:53 +0000 (01:05 -0600)]
A better spin-waiting scheme. Using compare_and_set will cause bus-broadcast
each time.

9 years agoreduced the frequency of calling atomic increment.
Chao Mei [Thu, 23 Feb 2012 06:11:57 +0000 (00:11 -0600)]
reduced the frequency of calling atomic increment.

9 years agoSimplified initialization
Chao Mei [Thu, 23 Feb 2012 04:12:37 +0000 (22:12 -0600)]
Simplified initialization

9 years agoMake notifyMsg local to each helper instead of being local to the node helper
Chao Mei [Thu, 23 Feb 2012 00:11:34 +0000 (18:11 -0600)]
Make notifyMsg local to each helper instead of being local to the node helper

9 years agoClean up the codes so that only the best implementation is kept. The old
Chao Mei [Wed, 22 Feb 2012 20:28:43 +0000 (14:28 -0600)]
Clean up the codes so that only the best implementation is kept. The old
one is moved to "alloptions" directory for performance testing reference.

9 years agochanged to allow multiple installations at the same time. use install.sh to
Chao Mei [Wed, 22 Feb 2012 00:29:13 +0000 (18:29 -0600)]
changed to allow multiple installations at the same time. use install.sh to
install the library

9 years agoFixed a bug in the case when a busy helper cannot run the previous
Chao Mei [Wed, 22 Feb 2012 00:02:17 +0000 (18:02 -0600)]
Fixed a bug in the case when a busy helper cannot run the previous
loop task in time, but then happens to run stealWork (the msg from
the previous task) during the initialization of a new loop task.

9 years agoFixed a bug in using the macro for the number of test iterations.
Chao Mei [Thu, 2 Feb 2012 22:12:33 +0000 (16:12 -0600)]
Fixed a bug in using the macro for the number of test iterations.

9 years agoAdded the implicit tree broadcast for notifying other helpers about the parallel...
Chao Mei [Thu, 2 Feb 2012 22:11:10 +0000 (16:11 -0600)]
Added the implicit tree broadcast for notifying other helpers about the parallel jobs

9 years agoFixed a bug in picking which notification msg to be used.
Chao Mei [Thu, 2 Feb 2012 20:16:16 +0000 (14:16 -0600)]
Fixed a bug in picking which notification msg to be used.

9 years agoA bug fix which has assumed the #chunks a job is partitioned into is the same with...
Chao Mei [Thu, 2 Feb 2012 18:49:00 +0000 (12:49 -0600)]
A bug fix which has assumed the #chunks a job is partitioned into is the same with the number of helpers.

9 years agoAdded a new scheme (work-stealing) which significantly reduced the overhead. E.g...
Chao Mei [Fri, 27 Jan 2012 04:22:14 +0000 (22:22 -0600)]
Added a new scheme (work-stealing) which significantly reduced the overhead. E.g. with ++ppn 12, the overhead of using the library is now only 3.7 us (on finesse)!! For more details, please refer to TMS.

9 years agoDid two optimizations:
Chao Mei [Thu, 26 Jan 2012 23:12:51 +0000 (17:12 -0600)]
Did two optimizations:
1. Replaced the notification scheme with converse-level msg instead of a charm-level entry method.
2. Using gcc built-ins (atomic increment, compare_and_swap) to test whether the job finishes

9 years agoAdded output to explain the mode of using this library.
Chao Mei [Thu, 26 Jan 2012 04:31:01 +0000 (22:31 -0600)]
Added output to explain the mode of using this library.

9 years ago1. Fixed the fftw plan creation so that the program now works.
Chao Mei [Tue, 24 Jan 2012 04:49:17 +0000 (22:49 -0600)]
1. Fixed the fftw plan creation so that the program now works.
2. Changed numThreads to numTasks for better understanding of the program.
3. Added a comment on what to parallelize for fftw computation.

9 years agoSome fix for using the nodeHelper lib. It compiles but seg fault when executing fftw...
Chao Mei [Mon, 23 Jan 2012 23:42:01 +0000 (17:42 -0600)]
Some fix for using the nodeHelper lib. It compiles but seg fault when executing fftw plans

9 years agoFixed a bug in setting the range of task iterations
Chao Mei [Mon, 23 Jan 2012 23:33:49 +0000 (17:33 -0600)]
Fixed a bug in setting the range of task iterations

9 years agoAdded the openmp test for performance comparison of the simple loop benchmark.
Chao Mei [Mon, 23 Jan 2012 19:36:18 +0000 (13:36 -0600)]
Added the openmp test for performance comparison of the simple loop benchmark.

9 years agoAdapted the fft-trans to the newly polished API.
Chao Mei [Mon, 23 Jan 2012 18:44:34 +0000 (12:44 -0600)]
Adapted the fft-trans to the newly polished API.

9 years agoA tweak in order to disable compiler optimization when calibrating the loop consecuti...
Chao Mei [Sat, 21 Jan 2012 22:55:48 +0000 (16:55 -0600)]
A tweak in order to disable compiler optimization when calibrating the loop consecutively.

9 years agoFixed a bug in computing average execution time.
Chao Mei [Sat, 21 Jan 2012 22:48:49 +0000 (16:48 -0600)]
Fixed a bug in computing average execution time.

9 years agoPolished/Updated the library and changed the example codes accordingly.
Chao Mei [Sat, 21 Jan 2012 00:13:20 +0000 (18:13 -0600)]
Polished/Updated the library and changed the example codes accordingly.

9 years agocopy of fft-trans modified to use nodeHelperLib
Eric Bohm [Wed, 4 Jan 2012 19:40:18 +0000 (13:40 -0600)]
copy of fft-trans modified to use nodeHelperLib

9 years agocomments and portability
Eric Bohm [Wed, 4 Jan 2012 19:37:03 +0000 (13:37 -0600)]
comments and portability

10 years agoinitial checkin
Chao Mei [Fri, 21 Oct 2011 18:49:50 +0000 (13:49 -0500)]
initial checkin

9 years agoREADME: Last-minute update to reflect current architectures, options v6.4.0
Phil Miller [Tue, 20 Mar 2012 21:05:32 +0000 (16:05 -0500)]
README: Last-minute update to reflect current architectures, options

9 years agoChangelog: mention BGQ PAMI and elaborate some details
Phil Miller [Tue, 20 Mar 2012 21:00:46 +0000 (16:00 -0500)]
Changelog: mention BGQ PAMI and elaborate some details

9 years agobuild: Add missing dependency (commitid.c) to make target 'depends'
Ramprasad Venkataraman [Tue, 20 Mar 2012 17:20:12 +0000 (12:20 -0500)]
build: Add missing dependency (commitid.c) to make target 'depends'

9 years agofix memory registeration bug with persistent message
Gengbin Zheng [Tue, 20 Mar 2012 07:27:44 +0000 (02:27 -0500)]
fix memory registeration bug with persistent message

9 years agofixed persistent put_done_tag in sendbuf
Yanhua Sun [Tue, 20 Mar 2012 05:30:29 +0000 (00:30 -0500)]
fixed persistent put_done_tag in sendbuf

9 years agoWrap BG/Q-specific topology code in the appropriate ifdef
Phil Miller [Tue, 20 Mar 2012 02:32:40 +0000 (21:32 -0500)]
Wrap BG/Q-specific topology code in the appropriate ifdef

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Tue, 20 Mar 2012 02:26:25 +0000 (19:26 -0700)]
Merge branch 'charm' of charmgit:charm into charm

9 years agosupport REMOTE_EVENT in persistent
Gengbin Zheng [Mon, 19 Mar 2012 22:49:40 +0000 (15:49 -0700)]
support REMOTE_EVENT in persistent

9 years agoRe-remove mentions of the archaic 'exemplar' architecture
Phil Miller [Mon, 19 Mar 2012 22:28:56 +0000 (17:28 -0500)]
Re-remove mentions of the archaic 'exemplar' architecture

9 years agoMerge branch 'charm' of charmgit:charm into charm
Eric Bohm [Mon, 19 Mar 2012 22:07:20 +0000 (22:07 +0000)]
Merge branch 'charm' of charmgit:charm into charm

9 years agotopo changes for BG/Q and path/config changes for ACLF VEAC BG/Q
Eric Bohm [Mon, 19 Mar 2012 22:05:24 +0000 (22:05 +0000)]
topo changes for BG/Q and path/config changes for ACLF VEAC BG/Q

9 years agoNDMeshStreamer: instead of recording time of last send, keep track of whether
Lukasz Wesolowski [Mon, 19 Mar 2012 21:12:16 +0000 (16:12 -0500)]
NDMeshStreamer: instead of recording time of last send, keep track of whether
sends have been performed using a simple flag

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Mon, 19 Mar 2012 19:14:00 +0000 (12:14 -0700)]
Merge branch 'charm' of charmgit:charm into charm

9 years agomake ackpool general
Gengbin Zheng [Mon, 19 Mar 2012 19:13:15 +0000 (12:13 -0700)]
make ackpool general
fix a bug in registerMessage that mistakenly check memhandle in chunk header

9 years agoDelete some generated files, and make sure they're cleaned up
Phil Miller [Mon, 19 Mar 2012 19:08:13 +0000 (14:08 -0500)]
Delete some generated files, and make sure they're cleaned up

9 years agopami prototype
Eric Bohm [Mon, 19 Mar 2012 18:06:18 +0000 (18:06 +0000)]
pami prototype
Paths updated for cetus.anl.alcf.gov

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Mon, 19 Mar 2012 18:02:20 +0000 (11:02 -0700)]
Merge branch 'charm' of charmgit:charm into charm

9 years agosome minor clean up
Gengbin Zheng [Mon, 19 Mar 2012 18:01:57 +0000 (11:01 -0700)]
some minor clean up

9 years agonet-sol-x86_64: Support buggy STLPort as in PGI and Fujitsu
Phil Miller [Mon, 19 Mar 2012 16:29:12 +0000 (11:29 -0500)]
net-sol-x86_64: Support buggy STLPort as in PGI and Fujitsu

9 years agoMerge branch 'charm' of charmgit:charm into charm
Eric Bohm [Mon, 19 Mar 2012 14:32:40 +0000 (14:32 +0000)]
Merge branch 'charm' of charmgit:charm into charm

9 years agoseem to be a typo
Gengbin Zheng [Mon, 19 Mar 2012 07:26:19 +0000 (00:26 -0700)]
seem to be a typo

9 years agoreverte
Gengbin Zheng [Mon, 19 Mar 2012 07:22:41 +0000 (00:22 -0700)]
reverte

9 years agoMerge branch 'charm' of charmgit:charm into charm
Yanhua Sun [Mon, 19 Mar 2012 07:11:22 +0000 (02:11 -0500)]
Merge branch 'charm' of charmgit:charm into charm

9 years agoadded CQWrite support for persistent message
Yanhua Sun [Mon, 19 Mar 2012 07:10:59 +0000 (02:10 -0500)]
added CQWrite support for persistent message

9 years agodisable FMA if using remote events
Gengbin Zheng [Mon, 19 Mar 2012 07:02:01 +0000 (00:02 -0700)]
disable FMA if using remote events

9 years agofix message registration bug in persistent
Gengbin Zheng [Mon, 19 Mar 2012 06:54:14 +0000 (23:54 -0700)]
fix message registration bug in persistent

9 years agoonly when print_stat is used, output counters
Yanhua Sun [Mon, 19 Mar 2012 04:09:15 +0000 (23:09 -0500)]
only when print_stat is used, output counters

9 years agoprint statistics counters into files
Yanhua Sun [Mon, 19 Mar 2012 03:42:32 +0000 (22:42 -0500)]
print statistics counters into files

9 years agoadded RDMA statistic information
Yanhua Sun [Mon, 19 Mar 2012 02:07:50 +0000 (21:07 -0500)]
added RDMA statistic information

9 years agopersistent comm works again with the new mem registration scheme.
Gengbin Zheng [Sun, 18 Mar 2012 23:13:13 +0000 (16:13 -0700)]
persistent comm works again with the new mem registration scheme.
still using ack_msg, can be optimized later
tested on simple tests with both SMP and  non-SMP.

9 years agoonly add padding for 32 bit
Gengbin Zheng [Sun, 18 Mar 2012 23:12:06 +0000 (16:12 -0700)]
only add padding for 32 bit

9 years agofix for persistent on SMP
Gengbin Zheng [Sun, 18 Mar 2012 22:31:52 +0000 (15:31 -0700)]
fix for persistent on SMP

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Sun, 18 Mar 2012 21:51:07 +0000 (14:51 -0700)]
Merge branch 'charm' of charmgit:charm into charm

9 years agosrcpe is used for two purpose, one for plain chare creation, another mainly by projec...
Gengbin Zheng [Sun, 18 Mar 2012 21:50:11 +0000 (14:50 -0700)]
srcpe is used for two purpose, one for plain chare creation, another mainly by projections. split them into two separate fields.

9 years agominor
Yanhua Sun [Sun, 18 Mar 2012 21:37:20 +0000 (16:37 -0500)]
minor

9 years agouncomment trace BTE Init tracing
Yanhua Sun [Sun, 18 Mar 2012 21:08:26 +0000 (14:08 -0700)]
uncomment trace BTE Init tracing

9 years agoadded trace FMA init
Yanhua Sun [Sun, 18 Mar 2012 20:18:45 +0000 (15:18 -0500)]
added trace FMA init

9 years agoAMPI FIX: Alltoall recursive doubling algo for small message requires in-order AMPI...
Gengbin Zheng [Sun, 18 Mar 2012 17:13:12 +0000 (10:13 -0700)]
AMPI FIX: Alltoall recursive doubling algo for small message requires in-order AMPI message delivery.

9 years agoMerge branch 'charm' of charmgit:charm into charm
Yanhua Sun [Sun, 18 Mar 2012 07:05:01 +0000 (02:05 -0500)]
Merge branch 'charm' of charmgit:charm into charm

9 years agoadded statistic informatin of gemini
Yanhua Sun [Sun, 18 Mar 2012 06:50:03 +0000 (01:50 -0500)]
added statistic informatin of gemini

9 years agoturn CMK_MEMORY_PROTECTABLE off. When mixing with hugetlbfs, the thread page ma be...
Gengbin Zheng [Sun, 18 Mar 2012 06:43:48 +0000 (23:43 -0700)]
turn CMK_MEMORY_PROTECTABLE off. When mixing with hugetlbfs, the thread page ma be large page and so mprotect seem not working.

9 years agoNDMeshStreamer: fixed a bug where timeOfLastSend_ was being set when
Lukasz Wesolowski [Sun, 18 Mar 2012 04:20:17 +0000 (23:20 -0500)]
NDMeshStreamer: fixed a bug where timeOfLastSend_ was being set when
no send had taken place

9 years agoNDMeshStreamer: initialize timeOfLastSend_ to prevent comparison
Lukasz Wesolowski [Sun, 18 Mar 2012 04:08:51 +0000 (23:08 -0500)]
NDMeshStreamer: initialize timeOfLastSend_ to prevent comparison
against unitialized value

9 years agoimportant fix: correctly keep track of non-NULL element when doing broadcast when...
Gengbin Zheng [Sat, 17 Mar 2012 16:11:26 +0000 (11:11 -0500)]
important fix: correctly keep track of non-NULL element when doing broadcast when stableLocation is true.

9 years agoMakefiles: consistently pass $(TESTOPTS) to bgtest runs
Phil Miller [Fri, 16 Mar 2012 15:26:51 +0000 (10:26 -0500)]
Makefiles: consistently pass $(TESTOPTS) to bgtest runs

9 years agofix for smp
Gengbin Zheng [Fri, 16 Mar 2012 06:45:56 +0000 (23:45 -0700)]
fix for smp

9 years agoerror check for overflow
Gengbin Zheng [Fri, 16 Mar 2012 05:56:35 +0000 (00:56 -0500)]
error check for overflow

9 years agofixed a bug of CQWrite memory registration
Yanhua Sun [Thu, 15 Mar 2012 21:27:11 +0000 (16:27 -0500)]
fixed a bug of CQWrite memory registration

9 years agore-fix the previous commit
Gengbin Zheng [Thu, 15 Mar 2012 20:09:06 +0000 (15:09 -0500)]
re-fix the previous commit

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Thu, 15 Mar 2012 20:04:00 +0000 (15:04 -0500)]
Merge branch 'charm' of charmgit:charm into charm

9 years agofix a bug in error checking for overflow
Gengbin Zheng [Thu, 15 Mar 2012 20:03:41 +0000 (15:03 -0500)]
fix a bug in error checking for overflow

9 years agoRelease note: describe AMPI's improved MPI_IN_PLACE support
Phil Miller [Thu, 15 Mar 2012 03:11:27 +0000 (22:11 -0500)]
Release note: describe AMPI's improved MPI_IN_PLACE support

9 years agoadded abort because of too much smsg mailbox
Yanhua Sun [Thu, 15 Mar 2012 04:14:27 +0000 (23:14 -0500)]
added abort because of too much smsg mailbox

9 years agoAMPI: checkBuf should never see MPI_IN_PLACE - all routines should handle that first
Phil Miller [Thu, 15 Mar 2012 02:41:04 +0000 (21:41 -0500)]
AMPI: checkBuf should never see MPI_IN_PLACE - all routines should handle that first

9 years agoAMPI_Sendrecv: Reject MPI_IN_PLACE where the standard doesn't call for it
Phil Miller [Thu, 15 Mar 2012 02:36:18 +0000 (21:36 -0500)]
AMPI_Sendrecv: Reject MPI_IN_PLACE where the standard doesn't call for it

9 years agoAMPI: Handle MPI_IN_PLACE uniformly where it already works, and error out where it...
Phil Miller [Thu, 15 Mar 2012 02:25:51 +0000 (21:25 -0500)]
AMPI: Handle MPI_IN_PLACE uniformly where it already works, and error out where it should but doesn't

9 years agomlogft AMPI_Reduce: Handle MPI_IN_PLACE before reading results out of the message log
Phil Miller [Wed, 14 Mar 2012 23:20:00 +0000 (18:20 -0500)]
mlogft AMPI_Reduce: Handle MPI_IN_PLACE before reading results out of the message log

9 years agoAMPI: tighten up error checking on arguments to many routines
Phil Miller [Thu, 15 Mar 2012 02:59:17 +0000 (21:59 -0500)]
AMPI: tighten up error checking on arguments to many routines

9 years agoAMPI errorCheck: take an optional 2nd buffer, to ease checking collectives
Phil Miller [Wed, 14 Mar 2012 23:25:09 +0000 (18:25 -0500)]
AMPI errorCheck: take an optional 2nd buffer, to ease checking collectives

9 years agoAMPI: Remove pointless prototype for errorCheck right before its definition
Phil Miller [Wed, 14 Mar 2012 23:24:10 +0000 (18:24 -0500)]
AMPI: Remove pointless prototype for errorCheck right before its definition

9 years agofix a bug in error checking for MPI_Gather, only root rank has a valid recvBuf pointe...
Gengbin Zheng [Wed, 14 Mar 2012 22:30:51 +0000 (17:30 -0500)]
fix a bug in error checking for MPI_Gather, only root rank has a valid recvBuf pointer. (need to be careful about error checking).
Also changed deprecated CMK_OPTIMIZE to CMK_ERROR_CHECKING

9 years agoAMPI: always assume no anytime migration and static insertion.
Gengbin Zheng [Wed, 14 Mar 2012 07:29:44 +0000 (00:29 -0700)]
AMPI:   always assume no anytime migration and static insertion.
since AMPI collectives are synchronous, so there is no anytime migration.

9 years agoadd a command line option for staticInsertion
Gengbin Zheng [Wed, 14 Mar 2012 07:12:35 +0000 (00:12 -0700)]
add a command line option for staticInsertion

9 years ago[warning] Bigsim: const-correctness fix
Phil Miller [Wed, 14 Mar 2012 04:05:52 +0000 (23:05 -0500)]
[warning] Bigsim: const-correctness fix

9 years ago[warning] METIS: Don't pass arbitrary strings as the format argument to fprintf
Phil Miller [Tue, 13 Mar 2012 22:59:02 +0000 (17:59 -0500)]
[warning] METIS: Don't pass arbitrary strings as the format argument to fprintf

9 years agoWarnings in projections: Don't assume that printf %lld formats 8-byte ints
Phil Miller [Tue, 13 Mar 2012 22:02:01 +0000 (17:02 -0500)]
Warnings in projections: Don't assume that printf %lld formats 8-byte ints