charm.git
16 years agomake sure -nobs not overrided by -shared option.
Gengbin Zheng [Wed, 2 Feb 2005 05:25:50 +0000 (05:25 +0000)]
make sure -nobs not overrided by -shared option.

16 years agofixed a bug in detecting the underlying mpiCC compiler.
Gengbin Zheng [Wed, 2 Feb 2005 05:22:04 +0000 (05:22 +0000)]
fixed a bug in detecting the underlying mpiCC compiler.

16 years agogenerate .so before .a to make sure when shared lib fails, no .a is generated.
Gengbin Zheng [Wed, 2 Feb 2005 05:06:24 +0000 (05:06 +0000)]
generate .so before .a to make sure when shared lib fails, no .a is generated.

16 years agoFixed a bug in pipelined broadcast comlib support
Chao Huang [Wed, 2 Feb 2005 04:59:52 +0000 (04:59 +0000)]
Fixed a bug in pipelined broadcast comlib support

16 years agomake sure only one copy of trace-common.o at link time. For ibm compiler, different...
Gengbin Zheng [Wed, 2 Feb 2005 04:47:30 +0000 (04:47 +0000)]
make sure only one copy of trace-common.o at link time. For ibm compiler, different copies of trace-common.o in each trace library results in different copies of data.

16 years agocharmrun by default write script to /tmp, and without error check!
Gengbin Zheng [Wed, 2 Feb 2005 03:57:16 +0000 (03:57 +0000)]
charmrun by default write script to /tmp, and without error check!
fixed by adding error checking in case file system is full, and try current directory in case /tmp is full.

16 years agoAdded support for broadcast comlib support. By default it's turned off.
Chao Huang [Tue, 1 Feb 2005 22:28:28 +0000 (22:28 +0000)]
Added support for broadcast comlib support. By default it's turned off.
The default strategy is USE_HYPERCUBE.

16 years agocall HypercubeGetBcastDestinations with virtual num pes.
Gengbin Zheng [Tue, 1 Feb 2005 16:50:17 +0000 (16:50 +0000)]
call HypercubeGetBcastDestinations with virtual num pes.

16 years agoa small change in HypercubeGetBcastDestinations() so that it takes any numpes instead...
Gengbin Zheng [Tue, 1 Feb 2005 16:48:26 +0000 (16:48 +0000)]
a small change in HypercubeGetBcastDestinations() so that it takes any numpes instead of CmiNumPes(). This is useful in BigSim.

16 years agoadded a bgtest.
Gengbin Zheng [Tue, 1 Feb 2005 04:26:39 +0000 (04:26 +0000)]
added a bgtest.

16 years agouse CkRegisterHandler instead of CmiRegisterHandler for bigsim
Gengbin Zheng [Tue, 1 Feb 2005 04:21:23 +0000 (04:21 +0000)]
use CkRegisterHandler instead of CmiRegisterHandler for bigsim

16 years agouse CkRegisterHandler instead of CmiRegisterHandler
Gengbin Zheng [Tue, 1 Feb 2005 04:20:55 +0000 (04:20 +0000)]
use CkRegisterHandler instead of CmiRegisterHandler

16 years ago*** empty log message ***
Filippo Gioachin [Mon, 31 Jan 2005 23:36:17 +0000 (23:36 +0000)]
*** empty log message ***

16 years agoFixing pipe bcast
Filippo Gioachin [Mon, 31 Jan 2005 23:22:09 +0000 (23:22 +0000)]
Fixing pipe bcast

16 years agoFixing pipe broadcast
Filippo Gioachin [Mon, 31 Jan 2005 23:21:39 +0000 (23:21 +0000)]
Fixing pipe broadcast

16 years ago*** empty log message ***
Sameer Kumar [Mon, 31 Jan 2005 18:35:10 +0000 (18:35 +0000)]
*** empty log message ***

16 years agoTEmporarily turning off learning
Sameer Kumar [Mon, 31 Jan 2005 17:21:00 +0000 (17:21 +0000)]
TEmporarily turning off learning

16 years agofixed a compilation error.
Gengbin Zheng [Mon, 31 Jan 2005 08:15:03 +0000 (08:15 +0000)]
fixed a compilation error.

16 years agoremoved.
Gengbin Zheng [Mon, 31 Jan 2005 05:58:08 +0000 (05:58 +0000)]
removed.

16 years agoadded back stream test.
Gengbin Zheng [Mon, 31 Jan 2005 05:28:16 +0000 (05:28 +0000)]
added back stream test.

16 years agoadded a simple helper function to change processor in comm. database.
Gengbin Zheng [Mon, 31 Jan 2005 04:51:19 +0000 (04:51 +0000)]
added a simple helper function to change processor in comm. database.

16 years agoa modification in LDStats database, changed pointer to array to CkVec for easier...
Gengbin Zheng [Mon, 31 Jan 2005 02:33:58 +0000 (02:33 +0000)]
a modification in LDStats database, changed pointer to array to CkVec for easier operations.

16 years agofixed a potential bug in puper of CkVec when vector is NULL.
Gengbin Zheng [Sun, 30 Jan 2005 23:35:30 +0000 (23:35 +0000)]
fixed a potential bug in puper of CkVec when vector is NULL.

16 years agorollback to the version before the change of fragmentation/buffer reuse.
Gengbin Zheng [Sun, 30 Jan 2005 18:59:30 +0000 (18:59 +0000)]
rollback to the version before the change of fragmentation/buffer reuse.

16 years agoadded -qthreaded to f90 compiler
Gengbin Zheng [Sun, 30 Jan 2005 03:27:06 +0000 (03:27 +0000)]
added -qthreaded to f90 compiler

16 years agoadded -qthreaded to make sure temporary vars are allocated on stack.
Gengbin Zheng [Sun, 30 Jan 2005 03:26:13 +0000 (03:26 +0000)]
added -qthreaded to make sure temporary vars are allocated on stack.

16 years agowhen parse error occurs for textfile, give a line number for the error.
Gengbin Zheng [Sat, 29 Jan 2005 23:51:57 +0000 (23:51 +0000)]
when parse error occurs for textfile, give a line number for the error.

16 years agocall ifort with -auto to place variable on stack.
Gengbin Zheng [Sat, 29 Jan 2005 19:36:48 +0000 (19:36 +0000)]
call ifort with -auto  to place variable on stack.

16 years agoadded -qextname to f77
Gengbin Zheng [Fri, 28 Jan 2005 21:13:15 +0000 (21:13 +0000)]
added -qextname to f77

16 years agoadded -qextname for fortran name mangling.
Gengbin Zheng [Fri, 28 Jan 2005 21:12:35 +0000 (21:12 +0000)]
added -qextname for fortran name mangling.

16 years agoPup the requests now. It works only for -memory isomalloc
Sayantan Chakravorty [Thu, 27 Jan 2005 22:14:12 +0000 (22:14 +0000)]
Pup the requests now. It works only for -memory isomalloc

16 years agoguess default gm dir.
Gengbin Zheng [Thu, 27 Jan 2005 17:22:16 +0000 (17:22 +0000)]
guess default gm dir.

16 years agofor teragrid always use ppn=2 to improve the chance of running autobuild timely.
Gengbin Zheng [Thu, 27 Jan 2005 04:09:54 +0000 (04:09 +0000)]
for teragrid always use ppn=2 to improve the chance of running autobuild timely.

16 years agocalling -gm_long <size> to let cmpirun switch to long protocol for larger messages...
Gengbin Zheng [Wed, 26 Jan 2005 17:15:29 +0000 (17:15 +0000)]
calling -gm_long <size> to let cmpirun switch to long protocol for larger messages. this seems to improve the situation where megatest crash on long messages on tungsten.

16 years agofound a workaround for megatest crash, turn back QT thread
Gengbin Zheng [Wed, 26 Jan 2005 17:10:00 +0000 (17:10 +0000)]
found a workaround for megatest crash, turn back QT thread

16 years agoa minor change
Gengbin Zheng [Wed, 26 Jan 2005 07:46:14 +0000 (07:46 +0000)]
a minor change

16 years agodefine CMK_F90MAINLIBS (for_main.o)
Gengbin Zheng [Wed, 26 Jan 2005 07:45:58 +0000 (07:45 +0000)]
define CMK_F90MAINLIBS (for_main.o)

16 years agoskip call to CmiGmConvertMachineID() to translate GM node id to global id if running...
Gengbin Zheng [Wed, 26 Jan 2005 05:06:05 +0000 (05:06 +0000)]
skip call to CmiGmConvertMachineID() to translate GM node id to global id if running in charmrun standalone mode.

16 years agoneed -nof90main as well to override the -f90main in some rare cases.
Gengbin Zheng [Tue, 25 Jan 2005 23:13:30 +0000 (23:13 +0000)]
need -nof90main as well to override the -f90main in some rare cases.

16 years agoadded new option -f90main which simply tells charmc that main is in f90 code, and...
Gengbin Zheng [Tue, 25 Jan 2005 21:22:40 +0000 (21:22 +0000)]
added new option -f90main which simply tells charmc that main is in f90 code, and for_main needs to be linked.

16 years agofor_main can only be linked when there is main() in fortran, for those main() in...
Gengbin Zheng [Tue, 25 Jan 2005 21:21:29 +0000 (21:21 +0000)]
for_main can only be linked when there is main() in fortran, for those main() in c/c++, it gave link error.

16 years agofor f90 linking, for_main.o is important for intel compilers.
Gengbin Zheng [Tue, 25 Jan 2005 19:57:32 +0000 (19:57 +0000)]
for f90 linking, for_main.o is important for intel compilers.

16 years agoswitch back to context threads since QT thread breaks megatest marshall test.
Gengbin Zheng [Tue, 25 Jan 2005 19:56:58 +0000 (19:56 +0000)]
switch back to context threads since QT thread breaks megatest marshall test.
thread migration then won't work, need to find a fix

16 years agoremoved c++ style comments, fixed dangerous // at end of #define!
Gengbin Zheng [Tue, 25 Jan 2005 17:40:23 +0000 (17:40 +0000)]
removed c++ style comments, fixed dangerous // at end of #define!

16 years agoreduce job wallclock limit to 10 minutes so that it may have better chance to be...
Gengbin Zheng [Tue, 25 Jan 2005 17:25:33 +0000 (17:25 +0000)]
reduce job wallclock limit to 10 minutes so that it may have better chance to be schduled.

16 years agofixed c++ stye comment.
Gengbin Zheng [Tue, 25 Jan 2005 17:14:58 +0000 (17:14 +0000)]
fixed c++ stye comment.

16 years agofixed broken line.
Gengbin Zheng [Tue, 25 Jan 2005 17:13:13 +0000 (17:13 +0000)]
fixed broken line.

16 years agodefined int8 to be long long
Gengbin Zheng [Tue, 25 Jan 2005 06:36:36 +0000 (06:36 +0000)]
defined int8 to be long long

16 years agoThere have been two features added: fragmented AMPI message and reusing
Chao Huang [Tue, 25 Jan 2005 06:19:30 +0000 (06:19 +0000)]
There have been two features added: fragmented AMPI message and reusing
AmpiMsg buffers. They are independent and can be turned on/off by
#define AMPI_ENABLE_FRAG_MSG and AMPI_ENABLE_BUFFER_POOLING to 1/0.
By default they are both turned off. The default msg fragment size
is 64KB. There are 4 queues for msg reuse: 1KB, 4KB, 16KB and 64KB.

16 years agoupdated
Gengbin Zheng [Tue, 25 Jan 2005 04:14:23 +0000 (04:14 +0000)]
updated

16 years agofixed ck_libs to compile on csar sun workstations.
Gengbin Zheng [Mon, 24 Jan 2005 23:48:31 +0000 (23:48 +0000)]
fixed ck_libs to compile on csar sun workstations.

16 years agoprototype code for hybrid LB, not in the official make yet.
Gengbin Zheng [Mon, 24 Jan 2005 16:16:13 +0000 (16:16 +0000)]
prototype code for hybrid LB, not in the official make yet.

16 years agoreturn NULL in dummy my_preferred_procs
Gengbin Zheng [Mon, 24 Jan 2005 16:07:47 +0000 (16:07 +0000)]
return NULL in dummy my_preferred_procs

16 years agouse queue myr2 on new turing.
Gengbin Zheng [Mon, 24 Jan 2005 04:03:45 +0000 (04:03 +0000)]
use queue myr2 on new turing.

16 years agoa minor change using macro.
Gengbin Zheng [Sun, 23 Jan 2005 22:32:47 +0000 (22:32 +0000)]
a minor change using macro.

16 years agoadded LBAgent.[hC] into charm official compilation.
Gengbin Zheng [Sun, 23 Jan 2005 21:31:18 +0000 (21:31 +0000)]
added LBAgent.[hC] into charm official compilation.

16 years agoadded a multicast LB agent.
Gengbin Zheng [Sun, 23 Jan 2005 21:27:42 +0000 (21:27 +0000)]
added a multicast LB agent.

16 years agoswitch to RotateLB.
Gengbin Zheng [Sun, 23 Jan 2005 17:35:03 +0000 (17:35 +0000)]
switch to RotateLB.

16 years agowhen hit "ctrl^c", cleanup job.
Gengbin Zheng [Sun, 23 Jan 2005 07:27:50 +0000 (07:27 +0000)]
when hit "ctrl^c", cleanup job.

16 years agoconverted AmpiMsg to varsize message
Gengbin Zheng [Sun, 23 Jan 2005 01:47:22 +0000 (01:47 +0000)]
converted AmpiMsg to varsize message

16 years agotested isomalloc (allow failure though)
Gengbin Zheng [Sat, 22 Jan 2005 18:23:55 +0000 (18:23 +0000)]
tested isomalloc (allow failure though)

16 years agomake qstat more robust to tolerate trasient errors by returing a few more times.
Gengbin Zheng [Sat, 22 Jan 2005 18:09:27 +0000 (18:09 +0000)]
make qstat more robust to tolerate trasient errors by returing a few more times.

16 years agoput a fixme comment at thread listen which will break after thread migrates.
Gengbin Zheng [Sat, 22 Jan 2005 03:54:03 +0000 (03:54 +0000)]
put a fixme comment at thread listen which will break after thread migrates.

16 years agoswitch to QT threads, context thread migration broken on tungsten.
Gengbin Zheng [Sat, 22 Jan 2005 03:20:20 +0000 (03:20 +0000)]
switch to QT threads, context thread migration broken on tungsten.

16 years agonow support LSF as on tun.ncsa.
Gengbin Zheng [Fri, 21 Jan 2005 20:07:36 +0000 (20:07 +0000)]
now support LSF as on tun.ncsa.

16 years agormeove '#' in command line
Gengbin Zheng [Fri, 21 Jan 2005 08:53:48 +0000 (08:53 +0000)]
rmeove '#' in command line

16 years agoswitched back to gnuold, since new gnu malloc does not work.
Gengbin Zheng [Fri, 21 Jan 2005 08:33:25 +0000 (08:33 +0000)]
switched back to gnuold, since new gnu malloc does not work.

16 years agofixed "help" print.
Gengbin Zheng [Fri, 21 Jan 2005 08:29:15 +0000 (08:29 +0000)]
fixed "help" print.

16 years agoFixed a bug in the interaction between tcharm and isomalloc.
Sayantan Chakravorty [Fri, 21 Jan 2005 07:31:51 +0000 (07:31 +0000)]
Fixed a bug in the interaction between tcharm and isomalloc.
In ampi programs ampi_init called TCHARM_ELEMENT and caused
deactivateThread to be called for an already deactivated thread.
The activate corresponding to the second deactivate, activated
the thread, whereas it should have restored the old state, ie
deactivated. This caused memory chunks to be wrongly identified
as belonging to a thread in isomalloc. So when a thread migrated
the wrong chunk of memory went along with it. This was fixed by
checking if a thread was already deactivated before deactivating it
and storing its heapBlock and nulling out the heap associated
with the thread. At activation the old value of heapBlock is restored.

A similar problem might still exist for swap globals

16 years agoAdded in member function for creation fftw plans, to be called in constructor and...
Yan Shi [Fri, 21 Jan 2005 04:17:32 +0000 (04:17 +0000)]
Added in member function for creation fftw plans, to be called in constructor and PUP.

16 years agoWhoops! One commented out line resulted in a devastating bug. I now have the defini...
Terry L. Wilmarth [Fri, 21 Jan 2005 01:14:01 +0000 (01:14 +0000)]
Whoops!  One commented out line resulted in a devastating bug.  I now have the definitive answer to why {} should always be used even for one-line blocks.

16 years agoa minor tweak in CMK_LD definitions.
Gengbin Zheng [Fri, 21 Jan 2005 00:19:44 +0000 (00:19 +0000)]
a minor tweak in CMK_LD definitions.

16 years agofixed gnu/intel mixed cxx libraries.
Gengbin Zheng [Fri, 21 Jan 2005 00:16:25 +0000 (00:16 +0000)]
fixed gnu/intel mixed cxx libraries.
program built with shared lib didn't run with old config.

16 years agoFixed similar bugs in rfftlib: size taken wrong, forgotten perslab in
Yan Shi [Thu, 20 Jan 2005 16:43:30 +0000 (16:43 +0000)]
Fixed similar bugs in rfftlib: size taken wrong, forgotten perslab in
copying. And make a private function to create fft plans.

16 years agoreduced number of iterations
Gengbin Zheng [Thu, 20 Jan 2005 08:43:04 +0000 (08:43 +0000)]
reduced number of iterations

16 years agofixed for bgtest
Gengbin Zheng [Thu, 20 Jan 2005 08:25:59 +0000 (08:25 +0000)]
fixed for bgtest

16 years agomake test depend on hello
Gengbin Zheng [Thu, 20 Jan 2005 08:15:53 +0000 (08:15 +0000)]
make test depend on hello

16 years agoadd delegation
Gengbin Zheng [Thu, 20 Jan 2005 08:04:08 +0000 (08:04 +0000)]
add delegation

16 years agoadded Makefile for test
Gengbin Zheng [Thu, 20 Jan 2005 08:03:41 +0000 (08:03 +0000)]
added Makefile for test

16 years agoFixed a bug!
Terry L. Wilmarth [Thu, 20 Jan 2005 01:10:15 +0000 (01:10 +0000)]
Fixed a bug!

16 years agoFixed a bug in the PVT reduction that only showed up for odd # PEs.
Terry L. Wilmarth [Thu, 20 Jan 2005 00:51:17 +0000 (00:51 +0000)]
Fixed a bug in the PVT reduction that only showed up for odd # PEs.
Added a COMM_TIMER to stats to separate it from DO_TIMER which the grainsize is based on.

16 years agofixed verious bugs due to wrong array index, wrong size in creating plans.
Gengbin Zheng [Wed, 19 Jan 2005 21:46:41 +0000 (21:46 +0000)]
fixed verious bugs due to wrong array index, wrong size in creating plans.

16 years agofixed shared lib suffix for max.
Gengbin Zheng [Wed, 19 Jan 2005 09:27:26 +0000 (09:27 +0000)]
fixed shared lib suffix for max.

16 years agoAdded in assertion check for validation of tranform flag.
Yan Shi [Tue, 18 Jan 2005 23:06:29 +0000 (23:06 +0000)]
Added in assertion check for validation of tranform flag.

16 years agoTest to monitor speed of message sends.
Terry L. Wilmarth [Tue, 18 Jan 2005 22:58:20 +0000 (22:58 +0000)]
Test to monitor speed of message sends.

16 years agoSwitched to use a UNIFORM object distribution with locality for closer array indices.
Terry L. Wilmarth [Tue, 18 Jan 2005 22:10:06 +0000 (22:10 +0000)]
Switched to use a UNIFORM object distribution with locality for closer array indices.

16 years agoSwitched to use adapt4, commented out some output.
Terry L. Wilmarth [Tue, 18 Jan 2005 22:08:53 +0000 (22:08 +0000)]
Switched to use adapt4, commented out some output.

16 years agoTwo changes:
Terry L. Wilmarth [Tue, 18 Jan 2005 22:07:18 +0000 (22:07 +0000)]
Two changes:
1 - Finished and debugged adapt4; this took some time to debug...
2 - Commented out a tiny part of POSE_invoke that is currently unused; it would be useful for load balancing one day, but is not used directly right now:  it was calling "lastKnown" to make a map of processor communication.  LB current approach keeps track of object comm instead of this, which is probably more accurate, and, at least in terms of lastKnown calls, more efficient.

16 years agoadd -lrfftw so things link correctly
Eric Bohm [Tue, 18 Jan 2005 21:05:18 +0000 (21:05 +0000)]
add -lrfftw so things link correctly

16 years agomisnamed variable in pup
Eric Bohm [Tue, 18 Jan 2005 20:53:04 +0000 (20:53 +0000)]
misnamed variable in pup

16 years agoready for 5.9 release 1
Gengbin Zheng [Tue, 18 Jan 2005 07:40:56 +0000 (07:40 +0000)]
ready for 5.9 release 1

16 years agoin case of ECONNRESET (Connection reset by peer) or ENOBUFS(No buffer space available...
Gengbin Zheng [Mon, 17 Jan 2005 21:40:24 +0000 (21:40 +0000)]
in case of ECONNRESET (Connection reset by peer) or ENOBUFS(No buffer space available), operations can retry.

16 years agoNo more QD.
Terry L. Wilmarth [Mon, 17 Jan 2005 20:52:24 +0000 (20:52 +0000)]
No more QD.

16 years agowith "help" only print uniq options.
Gengbin Zheng [Mon, 17 Jan 2005 04:03:44 +0000 (04:03 +0000)]
with "help" only print uniq options.

16 years agoupdate to the new topology class creation interface
Gengbin Zheng [Sun, 16 Jan 2005 22:57:40 +0000 (22:57 +0000)]
update to the new topology class creation interface

16 years agomerged with Amit's new code with topology info. Fixed include LBSimulation.h. The...
Gengbin Zheng [Sun, 16 Jan 2005 22:57:05 +0000 (22:57 +0000)]
merged with Amit's new code with topology info. Fixed include LBSimulation.h. The creation of a topology is thus changed.

16 years agoincreased job walltime limit to 20min
Gengbin Zheng [Sun, 16 Jan 2005 17:18:51 +0000 (17:18 +0000)]
increased job walltime limit to 20min

16 years agocheck if program correctly finished by grep'ing "End of Program"
Gengbin Zheng [Sun, 16 Jan 2005 17:09:39 +0000 (17:09 +0000)]
check if program correctly finished by grep'ing "End of Program"

16 years agosome minor code cleanup.
Gengbin Zheng [Sun, 16 Jan 2005 06:41:25 +0000 (06:41 +0000)]
some minor code cleanup.

16 years agopreserve OPTS_LDXX when update its value in conv-mach-pre.sh
Gengbin Zheng [Sun, 16 Jan 2005 06:35:23 +0000 (06:35 +0000)]
preserve OPTS_LDXX when update its value in conv-mach-pre.sh