charm.git
11 years agoMerge branch 'charm' of charmgit:charm into charm
Esteban Meneses [Thu, 11 Mar 2010 19:58:10 +0000 (13:58 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoChanges to introduce team-based message logging support.
Esteban Meneses [Thu, 11 Mar 2010 19:57:48 +0000 (13:57 -0600)]
Changes to introduce team-based message logging support.

11 years agoMerge branch 'charm' of charmgit:charm into charm
Yanhua YanhuaSunLaptop [Thu, 11 Mar 2010 18:47:33 +0000 (12:47 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agomodification in satisfiability example. Add grain size control
Yanhua YanhuaSunLaptop [Thu, 11 Mar 2010 18:47:16 +0000 (12:47 -0600)]
modification in satisfiability example. Add grain size control

11 years agoSetting variable CMK_CHARE_USE_PTR to 1 for the mlogft versions.
Esteban Meneses [Thu, 11 Mar 2010 17:10:08 +0000 (11:10 -0600)]
Setting variable CMK_CHARE_USE_PTR to 1 for the mlogft versions.
Eventually, we have to turn this flag off to provide singleton chare migration.

11 years agoprotect the use of ETIMEDOUT, VC++ does not define it.
Gengbin Zheng [Thu, 11 Mar 2010 16:18:26 +0000 (10:18 -0600)]
protect the use of ETIMEDOUT, VC++ does not define it.

via git-CVS emulator

11 years agoscatter the n^2 connection setup by shift the loop start.
Gengbin Zheng [Wed, 10 Mar 2010 16:31:12 +0000 (10:31 -0600)]
scatter the n^2 connection setup by shift the loop start.

11 years agocan't seem to ignore timieout error blindly. skt_send when timeout should terminate...
Gengbin Zheng [Wed, 10 Mar 2010 16:04:36 +0000 (10:04 -0600)]
can't seem to ignore timieout error blindly. skt_send when timeout should terminate the node program. Only skt_connect now ignores time out error, since it takes another timeout argument anyway to protect the case that charmrun may die prematurally at startup.

11 years agoat errno of ETIMEDOUT, should retry.
Gengbin Zheng [Wed, 10 Mar 2010 01:23:50 +0000 (19:23 -0600)]
at errno of ETIMEDOUT, should retry.
This solves the problem when connect() takes a long time at startup.

11 years agoMerge branch 'charm' of charmgit:charm into charm
Isaac Dooley [Tue, 9 Mar 2010 21:13:28 +0000 (16:13 -0500)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoInclude cur_restart_phase for all FT Charm++ versions.
Esteban Meneses [Tue, 9 Mar 2010 17:49:50 +0000 (11:49 -0600)]
Include cur_restart_phase for all FT Charm++ versions.

11 years agoUpdating Make.depends, because the old one was causing me some build errors on Jaguar.
Isaac Dooley [Tue, 9 Mar 2010 17:44:23 +0000 (12:44 -0500)]
Updating Make.depends, because the old one was causing me some build errors on Jaguar.

11 years agouse assert instead of CmiAssert.
Gengbin Zheng [Tue, 9 Mar 2010 14:01:17 +0000 (08:01 -0600)]
use assert instead of CmiAssert.

11 years agoslight modification about grain size control in satisfiability program
Yanhua Yanhua [Tue, 9 Mar 2010 03:27:44 +0000 (21:27 -0600)]
slight modification about grain size control in satisfiability program

11 years agoadd a satisfiability program
Yanhua YanhuaSunLaptop [Mon, 8 Mar 2010 22:50:30 +0000 (16:50 -0600)]
add a satisfiability program

11 years agoAdding smart build option for cray-xt build.
Isaac Dooley [Mon, 8 Mar 2010 21:57:28 +0000 (16:57 -0500)]
Adding smart build option for cray-xt build.

11 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Mon, 8 Mar 2010 20:14:37 +0000 (14:14 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoreport how long charmrun takes to launch all node programs.
Gengbin Zheng [Mon, 8 Mar 2010 20:13:59 +0000 (14:13 -0600)]
report how long charmrun takes to launch all node programs.
Right now it always prints, later can change to only when ++verbose.

11 years agoskt_select1 now calls poll() if possible.
Gengbin Zheng [Mon, 8 Mar 2010 20:11:19 +0000 (14:11 -0600)]
skt_select1 now calls poll() if possible.

11 years agoMerge branch 'charm' of charmgit:charm into charm
Phil Miller [Mon, 8 Mar 2010 19:12:01 +0000 (13:12 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoReport a sensible error when building 32-bit charm on 64-bit system
Phil Miller [Mon, 8 Mar 2010 19:08:03 +0000 (13:08 -0600)]
Report a sensible error when building 32-bit charm on 64-bit system

If a user tries to make a net-linux build of charm on a normal 64-bit
system, QuickThreads will fail to build in a relatively
incomprehensible way. When this build failure occurs, offer a
suggestion of the likely cause.

11 years agoCUDA: Modified conv-mach.h to prevent sleep calls in converse if there is no CPU...
Lukasz Wesolowski [Mon, 8 Mar 2010 19:05:16 +0000 (13:05 -0600)]
CUDA: Modified conv-mach.h to prevent sleep calls in converse if there is no CPU work available.
The sleep calls caused poor performance when GPU was running while the CPU was idle.

11 years agospanning tree: fix trivial typos in comments
Ramprasad Venkataraman [Mon, 8 Mar 2010 16:47:37 +0000 (10:47 -0600)]
spanning tree: fix trivial typos in comments

11 years agoMerge branch 'pm/sptree' into charm.
Ramprasad Venkataraman [Mon, 8 Mar 2010 16:40:20 +0000 (10:40 -0600)]
Merge branch 'pm/sptree' into charm.

Use overloading for cleaner implementation of topo::buildSpanningTree*()
Changes are as discussed
Trivial additions to make comments doxygen-friendly

11 years agoselect, unless configured specially, has a limit on number of sockets (which is 1024...
Gengbin Zheng [Mon, 8 Mar 2010 14:18:59 +0000 (08:18 -0600)]
select, unless configured specially, has a limit on number of sockets (which is 1024). This prevent charmrun starts a parallel job greater than 1k cores, for example on ranger.
Use poll instead select whenever it is possible, which is scalable.
If select is the only choice, a CmiAssert is added to detect the misuse. When fd is greater than FD_SETSIZE(1024), the behavior is not defined.

11 years agouninitialized parModulesRemaining caused program to hang at end during proj analysis...
Gengbin Zheng [Sun, 7 Mar 2010 19:57:36 +0000 (13:57 -0600)]
uninitialized parModulesRemaining caused program to hang at end during proj analysis when using projections.

11 years agoadded a function skt_tcp_no_nagle in sockRoutines.
Gengbin Zheng [Sun, 7 Mar 2010 17:11:26 +0000 (11:11 -0600)]
added a function skt_tcp_no_nagle in sockRoutines.
disabling nagle alg. in charmrun tcp sockets, this makes CmiBarrier working reliably.

11 years agormeove extra ; for pedantic build
Gengbin Zheng [Sat, 6 Mar 2010 13:48:21 +0000 (07:48 -0600)]
rmeove extra ; for pedantic build

11 years agoremoved trailing things after #endif
Gengbin Zheng [Sat, 6 Mar 2010 06:58:42 +0000 (00:58 -0600)]
removed trailing things after #endif

11 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Sat, 6 Mar 2010 06:53:49 +0000 (00:53 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agofixed non-constant size array declaration which VC++ does not like
Gengbin Zheng [Sat, 6 Mar 2010 06:52:58 +0000 (00:52 -0600)]
fixed non-constant size array declaration which VC++ does not like

11 years agoPrevious commit fc2d08771f broke record-replay. Fixing it
Filippo Gioachin [Sat, 6 Mar 2010 06:03:27 +0000 (00:03 -0600)]
Previous commit fc2d08771f broke record-replay. Fixing it

11 years agoTurning off allocationHistory in memory-charmdebug as it is an overhead for all appli...
Filippo Gioachin [Sat, 6 Mar 2010 06:01:28 +0000 (00:01 -0600)]
Turning off allocationHistory in memory-charmdebug as it is an overhead for all applications using charmdebug, without having any benefit for normal users

11 years agoRevert "//define PROJ_ANALYSIS"
Gengbin Zheng [Sat, 6 Mar 2010 05:22:50 +0000 (23:22 -0600)]
Revert "//define PROJ_ANALYSIS"

This reverts commit a37dc4898e2d613c4f5c381ee5b4393d5ddb840f.

11 years agocomment a debug print
Gengbin Zheng [Sat, 6 Mar 2010 05:21:36 +0000 (23:21 -0600)]
comment a debug print

11 years agoa new implementation of cpu topology using CmiReduce.
Gengbin Zheng [Sat, 6 Mar 2010 04:55:04 +0000 (22:55 -0600)]
a new implementation of cpu topology using CmiReduce.

11 years agominor: fixed a print in comment
Gengbin Zheng [Sat, 6 Mar 2010 04:48:49 +0000 (22:48 -0600)]
minor: fixed a print in comment

11 years agofixed a bug in new CmiBarrier for SMP.
Gengbin Zheng [Sat, 6 Mar 2010 04:47:26 +0000 (22:47 -0600)]
fixed a bug in new CmiBarrier for SMP.
barrier is at node level, so use node number (req_nClients) instead of pe count to determine

11 years agoMerge branch 'charm' of charmgit:charm into charm
Eric Bohm [Sat, 6 Mar 2010 02:16:02 +0000 (20:16 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoFix for const ** problem with cross sections.
Eric Bohm [Sat, 6 Mar 2010 02:11:15 +0000 (20:11 -0600)]
Fix for const ** problem with cross sections.
const type **thing cannot be initialized dynamically. const type * const * thing is fine.
FYI: type const * const * thing, is arguably best of all in terms of clarity, but the compiler doesn't care about the order of type and const in that syntax.

This fix is necessary for the new version of openatom.

11 years agoN Queens Example: modify main.c
Yanhua Yanhua [Fri, 5 Mar 2010 23:02:26 +0000 (17:02 -0600)]
N Queens Example: modify main.c

11 years agoAdd the NQueen example
Yanhua Yanhua [Tue, 23 Feb 2010 21:50:28 +0000 (15:50 -0600)]
Add the NQueen example

11 years ago//define PROJ_ANALYSIS
Yanhua Yanhua [Tue, 2 Feb 2010 20:57:39 +0000 (14:57 -0600)]
//define PROJ_ANALYSIS

11 years agoenable PROJ_ANALYSIS macro to get the global end time and write it into rc file
Yanhua Yanhua [Tue, 2 Feb 2010 20:55:00 +0000 (14:55 -0600)]
enable PROJ_ANALYSIS macro to get the global end time and write it into rc file

11 years agoa minor change so that ampi rank can migrate during block in MPI_Wait for Issend()
Gengbin Zheng [Fri, 5 Mar 2010 20:35:09 +0000 (14:35 -0600)]
a minor change so that ampi rank can migrate during block in MPI_Wait for Issend()

11 years agoMerge branch 'charm' of charmgit:charm into charm
Eric Bohm [Fri, 5 Mar 2010 20:32:25 +0000 (14:32 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoupdated the direct_config for the new DCMF release.
Eric Bohm [Fri, 5 Mar 2010 20:30:24 +0000 (14:30 -0600)]
updated the direct_config for the new DCMF release.
Fixes the bug in charm++/pingpong RDMA.

11 years agoa test for MPI_Issend()
Gengbin Zheng [Fri, 5 Mar 2010 20:28:03 +0000 (14:28 -0600)]
a test for MPI_Issend()

11 years agoimplemented MPI_Issend() - nonblocking synchronous send.
Gengbin Zheng [Fri, 5 Mar 2010 20:26:52 +0000 (14:26 -0600)]
implemented MPI_Issend() - nonblocking synchronous send.

11 years agonotify_crash defined in ckmemcheckpoint.C however is needed by machine.c to notify...
Gengbin Zheng [Fri, 5 Mar 2010 16:30:23 +0000 (10:30 -0600)]
notify_crash defined in ckmemcheckpoint.C however is needed by machine.c to notify a crash. This does not link if compiling a converse program.
Fixed this by using a function pointer.

11 years agofixed another bug in the new CmiBarrier0. charmrun's req_client[0] may not necessaril...
Gengbin Zheng [Fri, 5 Mar 2010 15:13:00 +0000 (09:13 -0600)]
fixed another bug in the new CmiBarrier0. charmrun's req_client[0] may not necessarily be node 0. Keep track of Pe number explicitly.

via git-CVS emulator

11 years agofixed a bug in the new CMiBarrier to allow interrupt happening during the process...
Gengbin Zheng [Fri, 5 Mar 2010 14:40:33 +0000 (08:40 -0600)]
fixed a bug in the new CMiBarrier  to allow interrupt happening during the process of barrier.

via git-CVS emulator

11 years agoremoved hardcoded -memory gnu
Gengbin Zheng [Fri, 5 Mar 2010 06:21:02 +0000 (00:21 -0600)]
removed hardcoded -memory gnu

11 years agoimplemented a reliable CmiBarrier using charmrun's tcp socket.
Gengbin Zheng [Fri, 5 Mar 2010 04:55:41 +0000 (22:55 -0600)]
implemented a reliable CmiBarrier using charmrun's tcp socket.

11 years agoCmiBarrier() for net version never worked. Disable it.
Gengbin Zheng [Fri, 5 Mar 2010 04:24:54 +0000 (22:24 -0600)]
CmiBarrier() for net version never worked. Disable it.

11 years agoincreased timeout in select.
Gengbin Zheng [Fri, 5 Mar 2010 03:32:04 +0000 (21:32 -0600)]
increased timeout in select.

11 years agoSpanning Trees: Overload instead of weird ownership logic
Phil Miller [Fri, 5 Mar 2010 00:37:42 +0000 (18:37 -0600)]
Spanning Trees: Overload instead of weird ownership logic

Spanning trees generated for multicasts, reductions, and so forth can
be generated with either a user-supplied strategy or a default one
(selected by a factory knowledgeable of machien characteristics and so
forth). The logic to do this was based on checking if the argument to
buildSpanningTree[Generation] was NULL, and constructing a strategy if
it was. This strategy then had to be deleted afterward only if it was
locally constructed, a somewhat clunky arrangement.

Instead, split the "use the default strategy" logic into an overloaded
function that doesn't take the pointer argument, and passes through to
the (now-simplified) original version. Incidentally, some conditionals
are replaced by compile-time overload resolution.

11 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Thu, 4 Mar 2010 22:36:09 +0000 (16:36 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agovarious fixes and optimizations for in-mem checkpointing/restart.
Gengbin Zheng [Thu, 4 Mar 2010 22:31:41 +0000 (16:31 -0600)]
various fixes and optimizations for in-mem checkpointing/restart.
1. at restart, a more robust code is written to discard old messages from older phase, and enforce QD.  (make QD worked in the first place)
2. uses several streaming to combine small messages to one bigger message to optimize communication at restart.
3. reduce the number of times CkStartQD has to be used

11 years agoCUDA: (1) Added a call to the GPU Manager progress function immediately after executi...
Lukasz Wesolowski [Thu, 4 Mar 2010 20:42:39 +0000 (14:42 -0600)]
CUDA: (1) Added a call to the GPU Manager progress function immediately after execution of
callbacks to increase GPU throughput and (2) modified the condition on executing cudaFree calls
to also wait for completion of pending GPU transfer in requests

11 years agoUpdate variable name and comments to reflect reality
Phil Miller [Thu, 4 Mar 2010 15:11:25 +0000 (09:11 -0600)]
Update variable name and comments to reflect reality

Change ArrayElement::_numElements to numInitialElements to reflect its actual
role and newfound privacy.

11 years agoanother fix for the _numElements, make it now a private variable, and a new function...
Gengbin Zheng [Thu, 4 Mar 2010 05:48:19 +0000 (23:48 -0600)]
another fix for the _numElements, make it now a private variable, and a new function ckGetArraySize() to access it.

11 years agoPedantic leak fix in spanning tree strategy code
Ramprasad Venkataraman [Wed, 3 Mar 2010 21:04:49 +0000 (15:04 -0600)]
Pedantic leak fix in spanning tree strategy code

Delete strategy object if the spanning tree code allocates one
Probably not of real consequence as these objects dont hold any data

11 years agoremoved the trailing ; in CmiGetRestartPhase
Gengbin Zheng [Thu, 4 Mar 2010 02:25:10 +0000 (20:25 -0600)]
removed the trailing ; in CmiGetRestartPhase

11 years agoupdated dependence
Gengbin Zheng [Thu, 4 Mar 2010 02:17:48 +0000 (20:17 -0600)]
updated dependence

11 years agoadded pup for checkpointing the ckmulticast group.
Gengbin Zheng [Thu, 4 Mar 2010 02:17:37 +0000 (20:17 -0600)]
added pup for checkpointing the ckmulticast group.

11 years agodisable when built for FT, because a restarted process can not do global comm to...
Gengbin Zheng [Thu, 4 Mar 2010 02:10:26 +0000 (20:10 -0600)]
disable when built for FT, because a restarted process can not do global comm to get cpu topology info.

11 years agoAdded new flag into memory layer to mark all the allocated block as "clean". This...
Filippo Gioachin [Thu, 4 Mar 2010 02:09:51 +0000 (20:09 -0600)]
Added new flag into memory layer to mark all the allocated block as "clean". This way leak views can filter initial stuff. Added also CCS functions

11 years agoremoved some trailing things after #endif
Gengbin Zheng [Thu, 4 Mar 2010 01:05:29 +0000 (19:05 -0600)]
removed some trailing things after #endif

11 years agoAdded new argument to specify job name
Filippo Gioachin [Thu, 4 Mar 2010 01:01:57 +0000 (19:01 -0600)]
Added new argument to specify job name

11 years agoMACHSTATE_I may print inside another print (interrupt), protect it using memlock.
Gengbin Zheng [Thu, 4 Mar 2010 01:04:19 +0000 (19:04 -0600)]
MACHSTATE_I may print inside another print (interrupt), protect it using memlock.

11 years agoMerge branch 'charm' of charmgit:charm into charm
Lukasz Wesolowski [Wed, 3 Mar 2010 22:31:19 +0000 (16:31 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoCUDA: fixed a bug in pinnedMallocHost and fixed the inclusion directory for CUDA...
Lukasz Wesolowski [Wed, 3 Mar 2010 22:25:24 +0000 (16:25 -0600)]
CUDA: fixed a bug in pinnedMallocHost and fixed the inclusion directory for CUDA libraries

11 years agorenamed numElements in ArrayElement to _numElements.
Gengbin Zheng [Wed, 3 Mar 2010 22:00:46 +0000 (16:00 -0600)]
renamed numElements in ArrayElement to _numElements.
numElements is a populate variable name, which is error prone to be a class member. e.g. when there is a global variable with same name.

11 years agoremoved parentheses surrounding CUDA_DIR and NVIDIA_CUDA_SDK
Pritish Jetley [Wed, 3 Mar 2010 20:25:12 +0000 (14:25 -0600)]
removed parentheses surrounding CUDA_DIR and NVIDIA_CUDA_SDK

11 years agoswitch to gigabit setting as default now. Performance-wise, it has been shown (for...
Gengbin Zheng [Wed, 3 Mar 2010 19:41:45 +0000 (13:41 -0600)]
switch to gigabit setting as default now. Performance-wise, it has been shown (for NAMD), and UDP/giga > TCP > UDP/eth

11 years agocharmrun keep track of restart count, and pass the counter to restarted process in...
Gengbin Zheng [Wed, 3 Mar 2010 00:57:21 +0000 (18:57 -0600)]
charmrun keep track of restart count, and pass the counter to restarted process in the +restartaftercrash argument.

11 years agoseveral changes related to FT:
Gengbin Zheng [Wed, 3 Mar 2010 00:53:12 +0000 (18:53 -0600)]
several changes related to FT:
1. CkLocRec_buffering destructor actually free up messages
2. change resume() to not to notify home processor as default, which allows combining messages as an optimization.
3. when reclaiming an object at restart time, do not send a message to remote processor to free remote Rec. Those remote recs will be free'ed by the remote processor anyway in the removeArrayElement phase of restart.

11 years agoMerge branch 'charm' of charmgit:charm into charm
Pritish Jetley [Tue, 2 Mar 2010 22:27:09 +0000 (16:27 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoFixed instrument bug that was causing NaN start times. Renamed startTime to phaseStar...
Pritish Jetley [Tue, 2 Mar 2010 22:27:02 +0000 (16:27 -0600)]
Fixed instrument bug that was causing NaN start times. Renamed startTime to phaseStartTime for struct workRequest.

11 years agoadded +netint which use sigio (opposite to +netpoll)
Gengbin Zheng [Tue, 2 Mar 2010 07:45:44 +0000 (01:45 -0600)]
added +netint which use sigio (opposite to +netpoll)

11 years agoadded CcdPERIODIC_5minute
Gengbin Zheng [Tue, 2 Mar 2010 07:38:25 +0000 (01:38 -0600)]
added CcdPERIODIC_5minute

11 years agoadded dependence to machine-recover.c
Gengbin Zheng [Tue, 2 Mar 2010 06:39:22 +0000 (00:39 -0600)]
added dependence to machine-recover.c

11 years agoRemoved the trailing backslash from the macros SCHEDULE_TOP,
Lukasz Wesolowski [Tue, 2 Mar 2010 02:02:16 +0000 (20:02 -0600)]
Removed the trailing backslash from the macros SCHEDULE_TOP,
SCHEDULE_MESSAGE, and SCHEDULE_IDLE

11 years agoFix msg leak in entry method CkMulticastMgr::setup()
Ramprasad Venkataraman [Mon, 1 Mar 2010 19:24:17 +0000 (13:24 -0600)]
Fix msg leak in entry method CkMulticastMgr::setup()

11 years agoInserting empty line after macro definition
Filippo Gioachin [Mon, 1 Mar 2010 20:40:50 +0000 (14:40 -0600)]
Inserting empty line after macro definition

11 years agoProtect createPool definition with GPU_MEMPOOL macro.
Pritish Jetley [Fri, 26 Feb 2010 19:19:52 +0000 (13:19 -0600)]
Protect createPool definition with GPU_MEMPOOL macro.

11 years agoMoved "#include cklists.h" file to .cu from .h. Otherwise, charmc complains
Pritish Jetley [Fri, 26 Feb 2010 19:10:24 +0000 (13:10 -0600)]
Moved "#include cklists.h" file to .cu from .h. Otherwise, charmc complains
that pup_c.h should be used in C programs instead of pup.h.

11 years agominor change to fix WR instrumentation code (GPU_INSTRUMENT_WRS).
Pritish Jetley [Fri, 26 Feb 2010 18:35:02 +0000 (12:35 -0600)]
minor change to fix WR instrumentation code (GPU_INSTRUMENT_WRS).

11 years agoBuild script prints erroneous information to the user
Filippo Gioachin [Fri, 26 Feb 2010 04:21:29 +0000 (22:21 -0600)]
Build script prints erroneous information to the user

11 years agoCUDA: Modified the Makefile and configuration files to use environment
Lukasz Wesolowski [Thu, 25 Feb 2010 22:29:30 +0000 (16:29 -0600)]
CUDA: Modified the Makefile and configuration files to use environment
variables NVIDIA_CUDA_SDK and CUDA_DIR as locations of the SDK and the
toolkit, respectively.

11 years agoMerge branch 'charm' of charmgit:charm into charm
Pritish Jetley [Wed, 24 Feb 2010 23:59:47 +0000 (17:59 -0600)]
Merge branch 'charm' of charmgit:charm into charm

11 years agoMinor change to fix the HybridAPI instrumentation mechanism.
Pritish Jetley [Wed, 24 Feb 2010 23:55:52 +0000 (17:55 -0600)]
Minor change to fix the HybridAPI instrumentation mechanism.
Checks request phase against length of CkVec before trying to
return data.

11 years agoCUDA: Changed the overlap test to work with the current version of the API.
Lukasz Wesolowski [Wed, 24 Feb 2010 22:14:19 +0000 (16:14 -0600)]
CUDA: Changed the overlap test to work with the current version of the API.

11 years agoCUDA: Small fix in the delayed free.
Lukasz Wesolowski [Wed, 24 Feb 2010 22:05:12 +0000 (16:05 -0600)]
CUDA: Small fix in the delayed free.

11 years agoCUDA: Simplified the delayed free code.
Lukasz Wesolowski [Wed, 24 Feb 2010 21:27:10 +0000 (15:27 -0600)]
CUDA: Simplified the delayed free code.

11 years agoFixed keyword conflicts (total_tasks and tasks_per_node cannot be used together)...
Chao Mei [Wed, 24 Feb 2010 19:53:36 +0000 (13:53 -0600)]
Fixed keyword conflicts (total_tasks and tasks_per_node cannot be used together) for submitting smp jobs.

11 years agoMake polling as the default mode now for both smp and nonsmp because real apps (such...
Chao Mei [Wed, 24 Feb 2010 19:51:34 +0000 (13:51 -0600)]
Make polling as the default mode now for both smp and nonsmp because real apps (such as NAMD) show better performance in the polling mode.

11 years agoRevert "Trace TCharm function calls as bracketed events"
Phil Miller [Wed, 24 Feb 2010 06:30:04 +0000 (00:30 -0600)]
Revert "Trace TCharm function calls as bracketed events"

Adding a new char[n] member to TCharmAPIRoutine, where n>8, causes
crashes after migration. This is mysterious, but that mystery should not
impede other development.

This reverts commit 18dde97d470814b3acc6f2ce2a1cca7d39b9de04.

11 years agoMerge branch 'charm' of charmgit:charm into charm
Pritish Jetley [Wed, 24 Feb 2010 05:12:43 +0000 (23:12 -0600)]
Merge branch 'charm' of charmgit:charm into charm