charm.git
9 years agoChanges to maintain the older scheme for isomalloc
Nikhil Jain [Wed, 11 Jan 2012 22:48:45 +0000 (16:48 -0600)]
Changes to maintain the older scheme for isomalloc

9 years agofix a compilation error
Gengbin Zheng [Wed, 7 Dec 2011 05:44:28 +0000 (23:44 -0600)]
fix a compilation error

9 years agoFixing the cause for armci crash - nothing wrong with implementation,
Nikhil Jain [Wed, 11 Jan 2012 16:49:22 +0000 (10:49 -0600)]
Fixing the cause for armci crash - nothing wrong with implementation,
the example tried to free a local memory allocation after migration
which should result in error

9 years agohandle nested events
Gengbin Zheng [Wed, 11 Jan 2012 15:43:31 +0000 (09:43 -0600)]
handle nested events

9 years agoadd missing END_EXECUTION trace in the beginning of a SDAG entry
Gengbin Zheng [Wed, 11 Jan 2012 15:41:04 +0000 (09:41 -0600)]
add missing END_EXECUTION trace in the beginning of a SDAG entry
which leads to wrong trace-projections/summary logs with mismatched begin/end execution.

9 years agoremove one line of print
Yanhua Sun [Wed, 11 Jan 2012 03:10:41 +0000 (21:10 -0600)]
remove one line of print

9 years agoBug fix for armci with mempool based isomalloc, as modifying
Nikhil Jain [Tue, 10 Jan 2012 09:32:09 +0000 (03:32 -0600)]
Bug fix for armci with mempool based isomalloc, as modifying
the example code to test migration

9 years agoBug fix for globals enabling the use of mempool based isomalloc
Nikhil Jain [Tue, 10 Jan 2012 07:23:52 +0000 (01:23 -0600)]
Bug fix for globals enabling the use of mempool based isomalloc

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Mon, 9 Jan 2012 22:31:23 +0000 (14:31 -0800)]
Merge branch 'charm' of charmgit:charm into charm

9 years agodisable RDMA test when SMP (since there seems to be bug for ibverbs layer with SMP...
Gengbin Zheng [Mon, 9 Jan 2012 22:28:56 +0000 (14:28 -0800)]
disable RDMA test when SMP (since there seems to be bug for ibverbs layer with SMP to fix)
Also change the originally rather bizarre use of node group element proxies.

9 years agominor change to suppress compiler warnings
Gengbin Zheng [Mon, 9 Jan 2012 22:27:16 +0000 (14:27 -0800)]
minor change to suppress compiler warnings

9 years agocharmxi: mark commit ID string as coming from C
Phil Miller [Mon, 9 Jan 2012 16:55:01 +0000 (10:55 -0600)]
charmxi: mark commit ID string as coming from C

9 years agorollback the changes to make mpi ft_restart work
Xiang Ni [Mon, 9 Jan 2012 02:07:59 +0000 (20:07 -0600)]
rollback the changes to make mpi ft_restart work

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Sun, 8 Jan 2012 21:25:20 +0000 (15:25 -0600)]
Merge branch 'charm' of charmgit:charm into charm

9 years agocharmxi: Use host compiler to embed commit ID
Phil Miller [Sun, 8 Jan 2012 17:25:05 +0000 (11:25 -0600)]
charmxi: Use host compiler to embed commit ID

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Sun, 8 Jan 2012 06:04:50 +0000 (00:04 -0600)]
Merge branch 'charm' of charmgit:charm into charm

9 years agoa CkCompactVec, which grows at front, and remove at the end linearly
Gengbin Zheng [Sun, 8 Jan 2012 05:52:39 +0000 (23:52 -0600)]
a CkCompactVec, which grows at front, and remove at the end linearly
the vector keeps an offset, which represents all elements before offset
are T(0)

9 years agochange insideTraceBracket to static
Gengbin Zheng [Sun, 8 Jan 2012 05:50:40 +0000 (23:50 -0600)]
change insideTraceBracket to static

9 years agofix a memory leak
Gengbin Zheng [Sun, 8 Jan 2012 05:16:06 +0000 (23:16 -0600)]
fix a memory leak

9 years agocharmxi: print commit ID when erroring out
Phil Miller [Sat, 7 Jan 2012 23:25:14 +0000 (17:25 -0600)]
charmxi: print commit ID when erroring out

9 years agoRevert "Drop requirement for user code to call CBase_foo::pup(p)"
Phil Miller [Sat, 7 Jan 2012 08:04:59 +0000 (02:04 -0600)]
Revert "Drop requirement for user code to call CBase_foo::pup(p)"

This reverts commit 0bd6105523351c97b7d6424c3b0227d6e5326498.

Conflicts:

doc/charm++/pup.tex
src/ck-core/cklocation.C

9 years agoRevert "TCharm & AMPI: Adapt to automatic pup of base class"
Phil Miller [Sat, 7 Jan 2012 07:53:06 +0000 (01:53 -0600)]
Revert "TCharm &  AMPI: Adapt to automatic pup of base class"

This reverts commit b9446be574ca6dce6048e5796ebdabb5a52787ca.

9 years agoRevert "Adapt checkpointing code to call base_pup methods"
Phil Miller [Sat, 7 Jan 2012 07:51:24 +0000 (01:51 -0600)]
Revert "Adapt checkpointing code to call base_pup methods"

This reverts commit 60eaf3ad33cf4d6777d6dcb5f54509afb612f571.

9 years agoReinsert call to CBase::pup in place of deleted ArrayElement1D::pup
Phil Miller [Sat, 7 Jan 2012 07:50:37 +0000 (01:50 -0600)]
Reinsert call to CBase::pup in place of deleted ArrayElement1D::pup

9 years agoMerge branch 'charm' of charmgit:charm into charm
Chao Mei [Fri, 6 Jan 2012 23:17:53 +0000 (17:17 -0600)]
Merge branch 'charm' of charmgit:charm into charm

9 years agoImplemented an alternative to use MPI_Irecv for msgs that use rzv protocol.
Chao Mei [Fri, 6 Jan 2012 23:15:43 +0000 (17:15 -0600)]
Implemented an alternative to use MPI_Irecv for msgs that use rzv protocol.

9 years agoGPUManager: minor changes to prevent most warnings.
Lukasz Wesolowski [Fri, 6 Jan 2012 04:30:41 +0000 (22:30 -0600)]
GPUManager: minor changes to prevent most warnings.

9 years agoCleaning up charm++ CUDA examples.
Lukasz Wesolowski [Fri, 6 Jan 2012 04:29:12 +0000 (22:29 -0600)]
Cleaning up charm++ CUDA examples.

9 years agomanual fix of priority and mapping
Ehsan [Fri, 6 Jan 2012 03:01:58 +0000 (21:01 -0600)]
manual fix of priority and mapping

9 years agoGPU Manager: remove assumed location of CUDA toolkit
Lukasz Wesolowski [Fri, 6 Jan 2012 00:53:22 +0000 (18:53 -0600)]
GPU Manager: remove assumed location of CUDA toolkit

9 years agoadd magic number to chare base class.
Gengbin Zheng [Thu, 5 Jan 2012 23:03:47 +0000 (17:03 -0600)]
add magic number to chare base class.
sanity check after pup to make sure super class is passed.
fixed  migration constructor calling chain.

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Thu, 5 Jan 2012 23:02:20 +0000 (17:02 -0600)]
Merge branch 'charm' of charmgit:charm into charm

9 years agoadd test for GreedyLB
Gengbin Zheng [Thu, 5 Jan 2012 23:01:34 +0000 (17:01 -0600)]
add test for GreedyLB

9 years agocharmxi: factor out CK_TEMPLATES_ONLY guard generation
Phil Miller [Wed, 4 Jan 2012 20:28:34 +0000 (14:28 -0600)]
charmxi: factor out CK_TEMPLATES_ONLY guard generation

9 years agoMerge commit '30f31a362b23bc054c1e3244d8c6c98758dcfe57' into charm
Phil Miller [Thu, 5 Jan 2012 01:05:58 +0000 (19:05 -0600)]
Merge commit '30f31a362b23bc054c1e3244d8c6c98758dcfe57' into charm

9 years agoAdding GPU Manager support for cray-xe build of charm.
Lukasz Wesolowski [Wed, 4 Jan 2012 23:20:29 +0000 (17:20 -0600)]
Adding GPU Manager support for cray-xe build of charm.

9 years agominor change about indentation
Gengbin Zheng [Wed, 4 Jan 2012 21:30:53 +0000 (15:30 -0600)]
minor change about indentation

9 years agominor change
Gengbin Zheng [Wed, 4 Jan 2012 21:29:31 +0000 (15:29 -0600)]
minor change

9 years agoadd a flag CMK_MALLINFO_IS_BROKEN to allow disabling using of mallinfo if it is broken.
Gengbin Zheng [Wed, 4 Jan 2012 21:12:12 +0000 (13:12 -0800)]
add a flag CMK_MALLINFO_IS_BROKEN to allow disabling using of mallinfo if it is broken.

9 years agohandle pgcc properly
Gengbin Zheng [Wed, 4 Jan 2012 21:10:20 +0000 (13:10 -0800)]
handle pgcc properly

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Wed, 4 Jan 2012 21:09:55 +0000 (13:09 -0800)]
Merge branch 'charm' of charmgit:charm into charm

9 years agominor clean up of flags
Gengbin Zheng [Wed, 4 Jan 2012 21:09:29 +0000 (13:09 -0800)]
minor clean up of flags

9 years agoComlib: Manager should inherit from some Chare-ish base class
Phil Miller [Wed, 4 Jan 2012 07:32:04 +0000 (01:32 -0600)]
Comlib: Manager should inherit from some Chare-ish base class

9 years agoA fresh implementation of post-recv scheme for MPI machine layer, including a dynamic...
Chao Mei [Wed, 4 Jan 2012 05:42:05 +0000 (23:42 -0600)]
A fresh implementation of post-recv scheme for MPI machine layer, including a dynamic post-recv scheme.

9 years agoMeshStreamer: removed one '%' operation, leading to a slight performance
Lukasz Wesolowski [Wed, 4 Jan 2012 00:28:21 +0000 (18:28 -0600)]
MeshStreamer: removed one '%' operation, leading to a slight performance
improvement.

9 years agoshould not even ask for thread level of support of MULTIPLE if the smp mode is comm_s...
Gengbin Zheng [Tue, 3 Jan 2012 21:25:50 +0000 (15:25 -0600)]
should not even ask for thread level of support of MULTIPLE if the smp mode is comm_send_recv.

9 years agoMake the case that work threads send msgs optional if MPI_THREAD_MULTIPLE is provided...
Chao Mei [Tue, 3 Jan 2012 06:36:29 +0000 (00:36 -0600)]
Make the case that work threads send msgs optional if MPI_THREAD_MULTIPLE is provided. The default option is still to let comm. thread send and recv.

9 years agoMinor changes for code format and naming convention
Chao Mei [Tue, 3 Jan 2012 05:45:06 +0000 (23:45 -0600)]
Minor changes for code format and naming convention

9 years agoSimplified broadcast root setting with a minor bug fix when spanning tree is not...
Chao Mei [Tue, 3 Jan 2012 04:08:25 +0000 (22:08 -0600)]
Simplified broadcast root setting with a minor bug fix when spanning tree is not used in SMP case.

9 years agoMeshStreamer: rewrote location caching scheme using arrays instead of a map
Lukasz Wesolowski [Mon, 2 Jan 2012 07:21:11 +0000 (01:21 -0600)]
MeshStreamer: rewrote location caching scheme using arrays instead of a map
to reduce overhead. With the optimization RandomAccess still runs a little
slower than without it so I am leaving it off by default.

9 years agoFix for static use in inline - getting rid of static usage
Nikhil Jain [Sun, 1 Jan 2012 21:08:00 +0000 (15:08 -0600)]
Fix for static use in inline - getting rid of static usage

9 years agoMinor change in mempool - saving size as power of two in slots
Nikhil Jain [Sun, 1 Jan 2012 08:19:14 +0000 (02:19 -0600)]
Minor change in mempool - saving size as power of two in slots
rather in terms of absolute number, corresponding changes
done in isomalloc

9 years agoAPI change to AdvanceCommunication
Gengbin Zheng [Sun, 1 Jan 2012 04:10:48 +0000 (22:10 -0600)]
API change to AdvanceCommunication

9 years agominor change
Gengbin Zheng [Sat, 31 Dec 2011 23:50:52 +0000 (15:50 -0800)]
minor change

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Sat, 31 Dec 2011 17:50:03 +0000 (09:50 -0800)]
Merge branch 'charm' of charmgit:charm into charm

9 years agocall PumpDatagramConnection every few millisecond (not just when idling) to make...
Gengbin Zheng [Sat, 31 Dec 2011 17:49:18 +0000 (09:49 -0800)]
call PumpDatagramConnection every few millisecond (not just when idling) to make sure connections are established.

9 years agochange due to API change of AdvanceCommunication()
Gengbin Zheng [Sat, 31 Dec 2011 17:44:13 +0000 (17:44 +0000)]
change due to API change of AdvanceCommunication()

9 years agoAPI change, LrtsAdvanceCommunication() now takes an extra parameter "whenidle"
Gengbin Zheng [Sat, 31 Dec 2011 17:34:19 +0000 (09:34 -0800)]
API change, LrtsAdvanceCommunication() now takes an extra parameter "whenidle"

9 years agocall PumpDatagramConnection only at idle time to avoid performance degradation.
Gengbin Zheng [Sat, 31 Dec 2011 17:23:28 +0000 (09:23 -0800)]
call PumpDatagramConnection only at idle time to avoid performance degradation.

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Sat, 31 Dec 2011 08:13:34 +0000 (00:13 -0800)]
Merge branch 'charm' of charmgit:charm into charm

9 years agocode clean up and a smsg buffer memory leak fix
Gengbin Zheng [Sat, 31 Dec 2011 08:12:52 +0000 (00:12 -0800)]
code clean up and a smsg buffer memory leak fix

9 years agoModified CUDA hello example to use current version of GPU Manager.
Lukasz Wesolowski [Sat, 31 Dec 2011 07:14:33 +0000 (01:14 -0600)]
Modified CUDA hello example to use current version of GPU Manager.

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Sat, 31 Dec 2011 06:22:45 +0000 (22:22 -0800)]
Merge branch 'charm' of charmgit:charm into charm

9 years agomake dynamic smsgs a runtime option, instead of compile time one.
Gengbin Zheng [Sat, 31 Dec 2011 06:21:15 +0000 (22:21 -0800)]
make dynamic smsgs a runtime option, instead of compile time one.
Also +smsgConnection to specify initial smsgs connections.

9 years agoReducing number of iterations to fit in small memory systems
Nikhil Jain [Sat, 31 Dec 2011 04:54:41 +0000 (22:54 -0600)]
Reducing number of iterations to fit in small memory systems

9 years agochanges due to API change of AdvanceCommunication
Gengbin Zheng [Fri, 30 Dec 2011 23:41:36 +0000 (17:41 -0600)]
changes due to API change of AdvanceCommunication

9 years agoremove the line that connecting to PE 0 always atr startup.
Gengbin Zheng [Fri, 30 Dec 2011 22:31:35 +0000 (14:31 -0800)]
remove the line that connecting to PE 0 always atr startup.

9 years agoturn useDynamicSMSG to 0
Gengbin Zheng [Fri, 30 Dec 2011 21:32:12 +0000 (13:32 -0800)]
turn useDynamicSMSG to 0

9 years agocode clean up.
Gengbin Zheng [Fri, 30 Dec 2011 21:31:38 +0000 (13:31 -0800)]
code clean up.

9 years agosimplify the handling of GNI_RC_ERROR_RESOURCE when making smsgs connections.
Gengbin Zheng [Fri, 30 Dec 2011 21:10:23 +0000 (13:10 -0800)]
simplify the handling of GNI_RC_ERROR_RESOURCE when making smsgs connections.
it can dafely ignore the error, and let unbound endpoint to handle it (passively).

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Fri, 30 Dec 2011 17:04:31 +0000 (09:04 -0800)]
Merge branch 'charm' of charmgit:charm into charm

9 years agodynamic smsgs connection setup seems to be working now after fixing a race condition...
Gengbin Zheng [Fri, 30 Dec 2011 17:03:40 +0000 (09:03 -0800)]
dynamic smsgs connection setup seems to be working now after fixing a race condition of receiving SMSG message and setting up the smsgs connection.

9 years agoremove -g from OPTS
Gengbin Zheng [Fri, 30 Dec 2011 16:49:28 +0000 (10:49 -0600)]
remove -g from OPTS

9 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Fri, 30 Dec 2011 01:15:10 +0000 (17:15 -0800)]
Merge branch 'charm' of charmgit:charm into charm

9 years agofix race condition in setting dynamic connections.
Gengbin Zheng [Fri, 30 Dec 2011 01:14:02 +0000 (17:14 -0800)]
fix race condition in setting dynamic connections.

9 years agoGit ignore: generated documentation
Phil Miller [Wed, 28 Dec 2011 20:29:45 +0000 (14:29 -0600)]
Git ignore: generated documentation

9 years agofix variable declaration for VC++
Gengbin Zheng [Thu, 29 Dec 2011 16:35:20 +0000 (10:35 -0600)]
fix variable declaration for VC++

9 years agoremove lb_test which prevents any checkin under that directory
Gengbin Zheng [Thu, 29 Dec 2011 16:34:38 +0000 (10:34 -0600)]
remove lb_test which prevents any checkin under that directory

9 years agofix the same bug when expanding smsgs buffer at dynamic setup.
Gengbin Zheng [Wed, 28 Dec 2011 05:48:59 +0000 (21:48 -0800)]
fix the same bug when expanding smsgs buffer at dynamic setup.

9 years agofixed smsg send atr setup (including bzero smsg buffer) for dynamic smsg setup.
Gengbin Zheng [Wed, 28 Dec 2011 05:47:14 +0000 (21:47 -0800)]
fixed smsg send atr setup (including bzero smsg buffer) for dynamic smsg setup.

9 years agoDynamic Smsg fix
Yanhua Sun [Wed, 28 Dec 2011 01:53:39 +0000 (17:53 -0800)]
Dynamic Smsg fix

9 years agominor
Yanhua Sun [Tue, 27 Dec 2011 22:47:13 +0000 (14:47 -0800)]
minor

9 years agofixed a memory leak problem in commbench broadcast
Yanhua Sun [Tue, 27 Dec 2011 22:40:23 +0000 (14:40 -0800)]
fixed a memory leak problem in commbench broadcast

9 years agoFix in commbench reduction and broadcast benchmark
Nikhil Jain [Tue, 27 Dec 2011 18:41:31 +0000 (12:41 -0600)]
Fix in commbench reduction and broadcast benchmark

9 years agoDoc: Fix typo 'c alled' -> 'called'
Phil Miller [Tue, 27 Dec 2011 15:49:35 +0000 (09:49 -0600)]
Doc: Fix typo 'c alled' -> 'called'

9 years agoDoc: Fix breakage
Phil Miller [Tue, 27 Dec 2011 15:49:11 +0000 (09:49 -0600)]
Doc: Fix breakage

9 years agoMerge branch 'charm' of charmgit:charm into charm
Yanhua Sun [Tue, 27 Dec 2011 06:32:28 +0000 (22:32 -0800)]
Merge branch 'charm' of charmgit:charm into charm

9 years agomodify mempool structure
Yanhua Sun [Tue, 27 Dec 2011 06:32:09 +0000 (22:32 -0800)]
modify mempool structure

9 years agoadded a point-point broadcast example
YanhuaSun [Tue, 27 Dec 2011 06:07:45 +0000 (00:07 -0600)]
added a point-point broadcast example

9 years agoMerge branch 'charm' of charmgit:charm into charm
Phil Miller [Tue, 27 Dec 2011 05:04:02 +0000 (23:04 -0600)]
Merge branch 'charm' of charmgit:charm into charm

9 years agoConvert libraries to inherit from CBase classes
Phil Miller [Tue, 27 Dec 2011 04:46:27 +0000 (22:46 -0600)]
Convert libraries to inherit from CBase classes

9 years agoAdding performance benchmarks for CmiReduce and Broadcast in commbench
Nikhil Jain [Tue, 27 Dec 2011 00:09:47 +0000 (18:09 -0600)]
Adding performance benchmarks for CmiReduce and Broadcast in commbench

9 years agoexample kNeighbor: Don't inherit from CBase and then call a system class PUP routine
Phil Miller [Mon, 26 Dec 2011 21:07:54 +0000 (15:07 -0600)]
example kNeighbor: Don't inherit from CBase and then call a system class PUP routine

9 years agoMeshStreamer: Inherit from CBase classes
Phil Miller [Mon, 26 Dec 2011 19:55:58 +0000 (13:55 -0600)]
MeshStreamer: Inherit from CBase classes

9 years agoAdapt checkpointing code to call base_pup methods
Phil Miller [Mon, 26 Dec 2011 17:30:14 +0000 (11:30 -0600)]
Adapt checkpointing code to call base_pup methods

9 years agoMegatest templates: Fix ambiguity for older compilers
Phil Miller [Mon, 26 Dec 2011 16:55:05 +0000 (10:55 -0600)]
Megatest templates: Fix ambiguity for older compilers

9 years agoTCharm & AMPI: Adapt to automatic pup of base class
Phil Miller [Fri, 2 Dec 2011 09:05:25 +0000 (03:05 -0600)]
TCharm &  AMPI: Adapt to automatic pup of base class

9 years agoDrop requirement for user code to call CBase_foo::pup(p)
Phil Miller [Fri, 2 Dec 2011 08:11:21 +0000 (02:11 -0600)]
Drop requirement for user code to call CBase_foo::pup(p)

Requiring users to explicitly call CBase_foo::pup(p) at the beginning
of any chare's pup routine was easy to forget and led to substantial
frustration. Thus, make the system responsible for this, and take the
burden off users. Update tests and documentation to match.

Add an additional virtual method base_pup(PUP::er &p) to take over the
responsibility of code that was previously in the user-called base
class pup() methods. Insert calls to base_pup() in the main
array-migration code paths.

Currently *not* modified are the checkpoint code and migration for
plain chares. The fault-tolerance code paths under various #ifdef
guards are also not changed.

After consideration, I'm fairly confident that this change will not
break existing code that directly inherits from
Chare/Group/NodeGroup/ArrayElementND, so long as its pup routine stays
old-school to match (i.e. calls Chare::pup directly). The
implementation of base_pup is empty in those classes, and the
generated class's base_pup would call the base class pup routines.

Existing code with user-defined inheritance is a tricky case where
this change may break. In the simplest variants, some data might run
through pup twice. More disconcerting is the possibility of double
allocation on unpacking and memory leaks. If there is any actual code
that encounters this problem, I'll be happy to provide the labor
necessary to repair it.

9 years agolb_test: Use relative path to charmc, not Filippo's
Phil Miller [Fri, 2 Dec 2011 07:50:10 +0000 (01:50 -0600)]
lb_test: Use relative path to charmc, not Filippo's

9 years agoTests: Always inherit from generated CBase_foo class
Phil Miller [Fri, 2 Dec 2011 07:42:47 +0000 (01:42 -0600)]
Tests: Always inherit from generated CBase_foo class