charm.git
12 years agoFixed broken build with control points on bluegenep.
Jonathan Lifflander [Tue, 10 Nov 2009 23:08:07 +0000 (23:08 +0000)]
Fixed broken build with control points on bluegenep.

12 years agoModified...
Jonathan Lifflander [Tue, 10 Nov 2009 22:11:13 +0000 (22:11 +0000)]
Modified...

12 years agoDocumentation: Some comments for our converse callbacks implementation
Ramprasad Venkataraman [Mon, 9 Nov 2009 20:17:58 +0000 (20:17 +0000)]
Documentation: Some comments for our converse callbacks implementation

12 years agoUpdate examples to show how custom array indices should use placement new to construc...
Ramprasad Venkataraman [Mon, 9 Nov 2009 19:31:42 +0000 (19:31 +0000)]
Update examples to show how custom array indices should use placement new to construct their indices in memory already allocated by the CkArrayIndex base class.

12 years agoIntermediate fix for the recurring issues with the funky data layout assumptions...
Ramprasad Venkataraman [Mon, 9 Nov 2009 19:17:26 +0000 (19:17 +0000)]
Intermediate fix for the recurring issues with the funky data layout assumptions inside the CkArrayIndex class hierarchy.

- Actual index data is now stored as a union within the base class (CkArrayIndex)
- Base class constructor zeroes the index bits
- ArrayIndices 4D-6D now use base-class union member indexShorts to access the data as shorts

Repurcussions:
- No effect on CkArrayIndex Max, 1D, 2D and 3D
- Any user or mapping code that directly uses data members of 4D-6D indices (eg. CkArrayIndex4D.index), instead of the member functions, needs to be fixed
- Custom array indices defined by user code have to use placement-new to construct their custom index

12 years agoupdated from kranken
Gengbin Zheng [Sun, 8 Nov 2009 06:13:06 +0000 (06:13 +0000)]
updated from kranken

12 years agoc++ style comments in c
Gengbin Zheng [Sat, 7 Nov 2009 20:00:36 +0000 (20:00 +0000)]
c++ style comments in c

12 years agoadded -qcpluscmt
Gengbin Zheng [Sat, 7 Nov 2009 19:57:52 +0000 (19:57 +0000)]
added -qcpluscmt

12 years agowhen mmap for a slot failed (which should not), try 5 times with random delay.
Gengbin Zheng [Fri, 6 Nov 2009 18:16:20 +0000 (18:16 +0000)]
when mmap for a slot failed (which should not), try 5 times with random delay.
I have observed some contention behavior when multiple processors on a same node
 compete for calling mmap for very large trunks.

12 years agotested for usleep
Gengbin Zheng [Fri, 6 Nov 2009 18:13:42 +0000 (18:13 +0000)]
tested for usleep

12 years agominor change in info output
Gengbin Zheng [Fri, 6 Nov 2009 17:11:34 +0000 (17:11 +0000)]
minor change in info output

12 years agowhen mmap fails, print errno
Gengbin Zheng [Fri, 6 Nov 2009 17:08:24 +0000 (17:08 +0000)]
when mmap fails, print errno

12 years agoJust turn off a debug flag.
Esteban Meneses [Fri, 6 Nov 2009 14:15:42 +0000 (14:15 +0000)]
Just turn off a debug flag.

12 years agoDocumentation: Describe CmiAlloc's limited purpose
Phil Miller [Fri, 6 Nov 2009 01:36:50 +0000 (01:36 +0000)]
Documentation: Describe CmiAlloc's limited purpose

12 years agoThe meaning of +p in the smp version has now changed and is similar to
Abhinav Bhatele [Thu, 5 Nov 2009 23:05:40 +0000 (23:05 +0000)]
The meaning of +p in the smp version has now changed and is similar to
other charm builds.

12 years agofor mmap probe scheme, if multiple processes on a same physical node trying to probe...
Gengbin Zheng [Thu, 5 Nov 2009 20:12:26 +0000 (20:12 +0000)]
for mmap probe scheme, if multiple processes on a same physical node trying to probe mmap region simultaneously for very large region, mmap may fail. Partially remedied this by retrying for a feel times with some random delay.

12 years agoCleaning up headers so that cp_effects.h is included in user program indirectly when...
Isaac Dooley [Thu, 5 Nov 2009 16:43:06 +0000 (16:43 +0000)]
Cleaning up headers so that cp_effects.h is included in user program indirectly when controlPoints.h is included in the user program.

12 years agoFixing the control point optimization schemes so they can be used without requiring...
Isaac Dooley [Thu, 5 Nov 2009 16:00:31 +0000 (16:00 +0000)]
Fixing the control point optimization schemes so they can be used without requiring previous data to be recorded.

12 years agoqueue not long from CmiAlloc
Gengbin Zheng [Thu, 5 Nov 2009 15:47:09 +0000 (15:47 +0000)]
queue not long from CmiAlloc

12 years agonot just doing the previous fix for ibverbs, but for all
Gengbin Zheng [Thu, 5 Nov 2009 15:46:13 +0000 (15:46 +0000)]
not just doing the previous fix for ibverbs, but for all

12 years agoIn BigSim, the incoming message holds an event corresponding to its receipt. The...
Gengbin Zheng [Thu, 5 Nov 2009 02:23:45 +0000 (02:23 +0000)]
In BigSim, the incoming message holds an event corresponding to its receipt. The event log become invalid after migrates. Store its original PE number to identify the migration.

12 years agoin TCHARM_Api_trace which use constructor/destructor to mark an entry into a tcharm...
Gengbin Zheng [Thu, 5 Nov 2009 02:20:39 +0000 (02:20 +0000)]
in TCHARM_Api_trace which use constructor/destructor to mark an entry into a tcharm/ampi function, the bglog saved at the entry can become valid after migrating to a new processor.
Store a processor number, so that it can easily identify this scenario when migration occur

12 years agoFixing some bugs with how phase data is accessed in controlPoint().
Isaac Dooley [Wed, 4 Nov 2009 21:43:22 +0000 (21:43 +0000)]
Fixing some bugs with how phase data is accessed in controlPoint().

12 years agoAdded new constructors for multi-sections
Filippo Gioachin [Wed, 4 Nov 2009 20:41:28 +0000 (20:41 +0000)]
Added new constructors for multi-sections

12 years agoExposing a function that flushes logs. This will allow for logs to be dumped together...
Isaac Dooley [Wed, 4 Nov 2009 20:14:15 +0000 (20:14 +0000)]
Exposing a function that flushes logs. This will allow for logs to be dumped together on all processors, minimizing the performance impact in the program.

12 years agoI don't any point using CmiAlloc for malloc queue structure. In fact, allocating...
Gengbin Zheng [Wed, 4 Nov 2009 16:11:49 +0000 (16:11 +0000)]
I don't any point using CmiAlloc for malloc queue structure. In fact, allocating from pinned memory can be expensive, and limited in the total amount allowed.
Using malloc for now at least for ibverbs.

12 years agofree message after doing broadcast.
Gengbin Zheng [Wed, 4 Nov 2009 02:28:25 +0000 (02:28 +0000)]
free message after doing broadcast.

12 years agoAdding a controlPointTimingStamp() function in the control point framework.
Isaac Dooley [Tue, 3 Nov 2009 21:55:37 +0000 (21:55 +0000)]
Adding a controlPointTimingStamp() function in the control point framework.

12 years agoBigsim loadlog utility: Minor cleanup
Phil Miller [Tue, 3 Nov 2009 20:29:01 +0000 (20:29 +0000)]
Bigsim loadlog utility: Minor cleanup

12 years agoFixing bug that occurs if an empty control point log is loaded at startup.
Isaac Dooley [Tue, 3 Nov 2009 20:14:44 +0000 (20:14 +0000)]
Fixing bug that occurs if an empty control point log is loaded at startup.

12 years agoModifying the control point command line arguments.
Isaac Dooley [Tue, 3 Nov 2009 19:32:27 +0000 (19:32 +0000)]
Modifying the control point command line arguments.

12 years agoformatting of printed data and cleanup
Abhinav Bhatele [Tue, 3 Nov 2009 18:11:12 +0000 (18:11 +0000)]
formatting of printed data and cleanup

12 years agoAdding command line arguments to enable loading/saving of control point data.
Isaac Dooley [Tue, 3 Nov 2009 16:33:26 +0000 (16:33 +0000)]
Adding command line arguments to enable loading/saving of control point data.

12 years agoAdding command line arguments that control which control point tuning scheme is used.
Isaac Dooley [Tue, 3 Nov 2009 15:43:51 +0000 (15:43 +0000)]
Adding command line arguments that control which control point tuning scheme is used.

12 years agoConverting some comments to doxygen comments.
Isaac Dooley [Tue, 3 Nov 2009 15:42:13 +0000 (15:42 +0000)]
Converting some comments to doxygen comments.

12 years agoadded TESTOPTS
Gengbin Zheng [Sun, 1 Nov 2009 15:13:46 +0000 (15:13 +0000)]
added TESTOPTS

12 years agohdlList[i]->op & BLOCKING_MASK == 1 is dangerous with -O, removed the comprison to 1
Gengbin Zheng [Sat, 31 Oct 2009 02:33:57 +0000 (02:33 +0000)]
hdlList[i]->op & BLOCKING_MASK == 1 is dangerous with -O, removed the comprison to 1

12 years agoAdded missing <string> header for std::string usage.
Orion Lawlor [Fri, 30 Oct 2009 19:41:30 +0000 (19:41 +0000)]
Added missing <string> header for std::string usage.
Curious this worked at all on normal linux gcc machines...

12 years agotypo
Gengbin Zheng [Fri, 30 Oct 2009 17:31:40 +0000 (17:31 +0000)]
typo

12 years agoAMPI, BigSim: Let AMPI events record a size in the BigSim timeline
Phil Miller [Fri, 30 Oct 2009 16:59:38 +0000 (16:59 +0000)]
AMPI, BigSim: Let AMPI events record a size in the BigSim timeline

Modeling MPI collectives explicitly in the simulator requires
knowledge of how much data the collective is moving around. There was
previously no way to record this information explicitly.

- Add an interface in trace-bluegene to set the 'size' field of the
  current timeline entry.
- Add a call to this interface in AMPI_Allreduce.
- Refactor some duplication in AMPI_Allreduce.
- Fix a typo in AMPI_Iallreduce that would record it as
  AMPI_Allreduce.

12 years agoBigsim: groundwork for size recording of MPI collective operations
Phil Miller [Fri, 30 Oct 2009 14:56:41 +0000 (14:56 +0000)]
Bigsim: groundwork for size recording of MPI collective operations

12 years agoSwapglobals: print out how much padding there actually is
Phil Miller [Fri, 30 Oct 2009 05:07:26 +0000 (05:07 +0000)]
Swapglobals: print out how much padding there actually is

12 years agofix another bug in previous fix.
Gengbin Zheng [Fri, 30 Oct 2009 01:17:31 +0000 (01:17 +0000)]
fix another bug in previous fix.

12 years agoadded armci
Gengbin Zheng [Fri, 30 Oct 2009 00:16:55 +0000 (00:16 +0000)]
added armci

12 years agoadded bgtest
Gengbin Zheng [Fri, 30 Oct 2009 00:16:33 +0000 (00:16 +0000)]
added bgtest

12 years agoCore: don't put extra ';' where they aren't needed
Phil Miller [Thu, 29 Oct 2009 23:51:53 +0000 (23:51 +0000)]
Core: don't put extra ';' where they aren't needed

12 years agotweaked a little on finding a buddy processor
Gengbin Zheng [Thu, 29 Oct 2009 22:15:49 +0000 (22:15 +0000)]
tweaked a little on finding a buddy processor

12 years agoAdded image resize to work around JPEG 64K image height limit,
Orion Lawlor [Thu, 29 Oct 2009 21:56:19 +0000 (21:56 +0000)]
Added image resize to work around JPEG 64K image height limit,
and added runlength encoding for new client.

12 years agominor changes to previous fix
Gengbin Zheng [Thu, 29 Oct 2009 19:13:42 +0000 (19:13 +0000)]
minor changes to previous fix

12 years agofixed a divide by 0 bug in case the object size is 0 in the recent changes.
Gengbin Zheng [Thu, 29 Oct 2009 18:39:25 +0000 (18:39 +0000)]
fixed a divide by 0 bug in case the object size is 0 in the recent changes.

12 years agoremoved a debug print
Gengbin Zheng [Thu, 29 Oct 2009 16:48:32 +0000 (16:48 +0000)]
removed a debug print

12 years agoMoving a function to the ControlPoint namespace.
Isaac Dooley [Thu, 29 Oct 2009 16:09:46 +0000 (16:09 +0000)]
Moving a function to the ControlPoint namespace.

12 years agoreduced iteration number
Gengbin Zheng [Thu, 29 Oct 2009 15:54:50 +0000 (15:54 +0000)]
reduced iteration number

12 years agoChanges due to the new type for ChareType struct.
Esteban Meneses [Thu, 29 Oct 2009 14:56:45 +0000 (14:56 +0000)]
Changes due to the new type for ChareType struct.

12 years agoAdded support for team-based message logging.
Esteban Meneses [Thu, 29 Oct 2009 14:42:31 +0000 (14:42 +0000)]
Added support for team-based message logging.

12 years agoAdded some support for team-based message logging.
Esteban Meneses [Thu, 29 Oct 2009 14:42:11 +0000 (14:42 +0000)]
Added some support for team-based message logging.

12 years agoAdded a new flag for message logging and team-based support.
Esteban Meneses [Thu, 29 Oct 2009 14:41:41 +0000 (14:41 +0000)]
Added a new flag for message logging and team-based support.

12 years agoadded armci test
Gengbin Zheng [Thu, 29 Oct 2009 07:19:04 +0000 (07:19 +0000)]
added armci test

12 years agoadded
Gengbin Zheng [Thu, 29 Oct 2009 07:18:09 +0000 (07:18 +0000)]
added

12 years agofixed makefiles.
Gengbin Zheng [Thu, 29 Oct 2009 06:57:44 +0000 (06:57 +0000)]
fixed makefiles.

12 years agogeneralzie the previous fix: make sure only those pthreads that spawned by charm...
Gengbin Zheng [Thu, 29 Oct 2009 03:48:28 +0000 (03:48 +0000)]
generalzie the previous fix: make sure only those pthreads that spawned by charm can call isomalloc, other system spawned pthreads should not call isomalloc.

12 years agoperftest
Gengbin Zheng [Thu, 29 Oct 2009 03:19:30 +0000 (03:19 +0000)]
perftest

12 years agowhen not enabled, it treats cpu topology just the same as the charm/SMP node (CmiNumN...
Gengbin Zheng [Wed, 28 Oct 2009 21:23:14 +0000 (21:23 +0000)]
when not enabled, it treats cpu topology just the same as the charm/SMP node (CmiNumNodes(), CmiRankOf(pe), etc)

12 years agolog changes
Gengbin Zheng [Wed, 28 Oct 2009 21:04:02 +0000 (21:04 +0000)]
log changes

12 years agoupdated due to change in cputopology
Gengbin Zheng [Wed, 28 Oct 2009 20:45:27 +0000 (20:45 +0000)]
updated due to change in cputopology

12 years agochanged cputopology function API (some parameter from pe to node id)
Gengbin Zheng [Wed, 28 Oct 2009 20:44:54 +0000 (20:44 +0000)]
changed cputopology function API (some parameter from pe to node id)
added two more functions

12 years agowhen picking buddy processors, always find a processor on next *physical* processors...
Gengbin Zheng [Wed, 28 Oct 2009 20:43:31 +0000 (20:43 +0000)]
when picking buddy processors, always find a processor on next *physical* processors. Implemented using cpu topology API

12 years agofixed syntax error in IA64's atomic operations
Gengbin Zheng [Wed, 28 Oct 2009 03:31:49 +0000 (03:31 +0000)]
fixed syntax error in IA64's atomic operations

12 years agoswitch to memory fence for IA64
Gengbin Zheng [Wed, 28 Oct 2009 03:30:56 +0000 (03:30 +0000)]
switch to memory fence for IA64

12 years agocharmxi: Work around failings of ancient Sun compilers
Phil Miller [Tue, 27 Oct 2009 21:54:45 +0000 (21:54 +0000)]
charmxi: Work around failings of ancient Sun compilers

Sun C++ 5.4, included with Sun Studio 7, can't handle default
arguments to template functions correctly. Overload the templates in
question to have versions with and without the argument, and work the
argument into the body of the overload.

12 years agoAdded control point stuff to the Makefile...
Jonathan Lifflander [Tue, 27 Oct 2009 20:41:26 +0000 (20:41 +0000)]
Added control point stuff to the Makefile...

12 years agoAdded the generated file for control points.
Jonathan Lifflander [Tue, 27 Oct 2009 19:50:29 +0000 (19:50 +0000)]
Added the generated file for control points.

12 years agofixed a compiler error
Gengbin Zheng [Tue, 27 Oct 2009 13:55:07 +0000 (13:55 +0000)]
fixed a compiler error

12 years agocall CmiMachineExit() in machine_exit()
Gengbin Zheng [Tue, 27 Oct 2009 06:32:26 +0000 (06:32 +0000)]
call CmiMachineExit() in machine_exit()

12 years agosome fix for SMP mode, and clean shutdown.
Gengbin Zheng [Tue, 27 Oct 2009 06:31:51 +0000 (06:31 +0000)]
some fix for SMP mode, and clean shutdown.

12 years agofixed a typo in previous change
Gengbin Zheng [Mon, 26 Oct 2009 20:48:55 +0000 (20:48 +0000)]
fixed a typo in previous change

12 years agoUpdating Control Point API generator.
Isaac Dooley [Mon, 26 Oct 2009 17:50:11 +0000 (17:50 +0000)]
Updating Control Point API generator.

12 years agoRenaming Makefile because the current charm build scripts will overwrite the main...
Isaac Dooley [Mon, 26 Oct 2009 17:29:13 +0000 (17:29 +0000)]
Renaming Makefile because the current charm build scripts will overwrite the main Makefile with this one.

12 years agoCleaning up the main doxygen page. Stripping out a bunch of unnecessary HTML code...
Isaac Dooley [Mon, 26 Oct 2009 17:20:03 +0000 (17:20 +0000)]
Cleaning up the main doxygen page. Stripping out a bunch of unnecessary HTML code to replace with simpler doxygen commands. I also added a section for experimental features such as critical path detection.

12 years agoA script that generates parts of the control point framework's API and implementation.
Isaac Dooley [Mon, 26 Oct 2009 16:42:16 +0000 (16:42 +0000)]
A script that generates parts of the control point framework's API and implementation.

12 years agoprint MPI_Init_thread level in characters for some missing ones.
Gengbin Zheng [Mon, 26 Oct 2009 16:17:29 +0000 (16:17 +0000)]
print MPI_Init_thread level in characters for some missing ones.

12 years agoadded missing TypeMainChare
Gengbin Zheng [Sun, 25 Oct 2009 03:30:35 +0000 (03:30 +0000)]
added missing TypeMainChare

12 years agoremoved POSIX_SOURCE macro
Gengbin Zheng [Sat, 24 Oct 2009 14:40:58 +0000 (14:40 +0000)]
removed POSIX_SOURCE macro

12 years agoprint num core infop on cray
Gengbin Zheng [Sat, 24 Oct 2009 05:52:28 +0000 (05:52 +0000)]
print num core infop on cray

12 years agoturned off signal
Gengbin Zheng [Sat, 24 Oct 2009 03:42:46 +0000 (03:42 +0000)]
turned off signal

12 years agoenable topology for bluegenep, which the nodeID can be an arbitary number (not ordere...
Gengbin Zheng [Sat, 24 Oct 2009 03:41:42 +0000 (03:41 +0000)]
enable topology for bluegenep, which the nodeID can be an arbitary number (not ordered in [0-numNodes].

12 years agoread --mode
Gengbin Zheng [Sat, 24 Oct 2009 03:30:18 +0000 (03:30 +0000)]
read --mode

12 years agoRemove spurious include of <features.h>; it's a Linux-ism
Phil Miller [Fri, 23 Oct 2009 22:53:02 +0000 (22:53 +0000)]
Remove spurious include of <features.h>; it's a Linux-ism

12 years agoArch: Add a pedantic mode
Phil Miller [Fri, 23 Oct 2009 19:53:33 +0000 (19:53 +0000)]
Arch: Add a pedantic mode

12 years agoPedantry: Core changes to compile with -pedantic
Phil Miller [Fri, 23 Oct 2009 19:49:15 +0000 (19:49 +0000)]
Pedantry: Core changes to compile with -pedantic

12 years agoPedantry: The last extraneous ';'s
Phil Miller [Fri, 23 Oct 2009 19:21:21 +0000 (19:21 +0000)]
Pedantry: The last extraneous ';'s

12 years agoturn off shmem_swap which seems to be very slow
Gengbin Zheng [Fri, 23 Oct 2009 03:00:43 +0000 (03:00 +0000)]
turn off shmem_swap which seems to be very slow
Added condition callback for idle

12 years agofixed a syntax error in previous checkin
Gengbin Zheng [Fri, 23 Oct 2009 02:58:56 +0000 (02:58 +0000)]
fixed a syntax error in previous checkin

12 years agofixed previous checkin
Gengbin Zheng [Fri, 23 Oct 2009 02:48:38 +0000 (02:48 +0000)]
fixed previous checkin

12 years agouse %g to print float
Gengbin Zheng [Fri, 23 Oct 2009 02:24:10 +0000 (02:24 +0000)]
use %g to print float

12 years agoThe initial implementation of prefetch optimization for bigsim's ooc emulation by...
Chao Mei [Fri, 23 Oct 2009 02:22:37 +0000 (02:22 +0000)]
The initial implementation of prefetch optimization for bigsim's ooc emulation by using asynchronous IO (use callback in a newly created pthread for notifying the IO completion). To enable this, modify conv-mach-bigsim.h to turn on BIGSIM_OOC_PREFETCH.

12 years ago1. Rewrote the part of using/initializing PAPI counter
Chao Mei [Fri, 23 Oct 2009 02:20:02 +0000 (02:20 +0000)]
1. Rewrote the part of using/initializing PAPI counter
2. Changes for prefetch optimization for ooc emulation

12 years agoAdded a thread private variable for the sake of prefetch optimization (using AIO...
Chao Mei [Fri, 23 Oct 2009 02:17:34 +0000 (02:17 +0000)]
Added a thread private variable for the sake of prefetch optimization (using AIO) in bigsim's ooc emulation.

12 years agoAdded a macro for enabling prefetch optimization (using AIO) in bigsim ooc emulation
Chao Mei [Fri, 23 Oct 2009 02:16:34 +0000 (02:16 +0000)]
Added a macro for enabling prefetch optimization (using AIO) in bigsim ooc emulation

12 years agopath to charm bin fixed
Abhinav Bhatele [Thu, 22 Oct 2009 23:28:57 +0000 (23:28 +0000)]
path to charm bin fixed