charm.git
10 years agoprint if it is using absolute timer.
Gengbin Zheng [Tue, 12 Apr 2011 04:23:56 +0000 (23:23 -0500)]
print if it is using absolute timer.

10 years agofixed an unitialized variable to please Valgrind.
Gengbin Zheng [Tue, 12 Apr 2011 04:21:54 +0000 (23:21 -0500)]
fixed an unitialized variable to please Valgrind.

10 years agoMerge branch 'charm' of charmgit:charm into charm
Aaron Becker [Mon, 11 Apr 2011 19:50:56 +0000 (14:50 -0500)]
Merge branch 'charm' of charmgit:charm into charm

10 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Sun, 10 Apr 2011 20:09:54 +0000 (15:09 -0500)]
Merge branch 'charm' of charmgit:charm into charm

10 years agocomm thread should not block when it is is bluegenep smp.
Gengbin Zheng [Sun, 10 Apr 2011 20:08:16 +0000 (15:08 -0500)]
comm thread should not block when it is is bluegenep smp.

10 years ago+pemap does not print mapping by default, only print when +show_cpu_affinity
Gengbin Zheng [Sat, 9 Apr 2011 02:17:22 +0000 (21:17 -0500)]
+pemap does not print mapping by default, only print when +show_cpu_affinity

10 years agoreview the code for generating dependencies for MPI collectives.
Gengbin Zheng [Thu, 7 Apr 2011 16:32:30 +0000 (11:32 -0500)]
review the code for generating dependencies for MPI collectives.
various minor changes. added mpiOp for MPI_Alltoall.

10 years agouse macro to define name string of the log
Gengbin Zheng [Thu, 7 Apr 2011 15:06:58 +0000 (10:06 -0500)]
use macro to define name string of the log

10 years agofixed a silly bug of charm *.<junk> removes all files.
Gengbin Zheng [Wed, 6 Apr 2011 20:30:00 +0000 (15:30 -0500)]
fixed a silly bug of charm *.<junk> removes all files.

10 years agoCharj: Added testing script to check for successful compilation
Aaron Becker [Tue, 5 Apr 2011 22:08:56 +0000 (17:08 -0500)]
Charj: Added testing script to check for successful compilation

10 years agofor CRAY, setcpuaffinity does not have to involve global communication to find the...
Gengbin Zheng [Tue, 5 Apr 2011 15:21:41 +0000 (10:21 -0500)]
for CRAY, setcpuaffinity does not have to involve global communication to find the charm PE to core mapping.

10 years agominor changes for BG/Q which it seems not supported.
Gengbin Zheng [Tue, 5 Apr 2011 14:58:34 +0000 (09:58 -0500)]
minor changes for BG/Q which it seems not supported.

10 years agoMerge branch 'charm' of charmgit:charm into charm
Chao Mei [Mon, 4 Apr 2011 20:27:09 +0000 (15:27 -0500)]
Merge branch 'charm' of charmgit:charm into charm

10 years agoAdded to use cap when there're many messages to be released.
Chao Mei [Mon, 4 Apr 2011 20:25:52 +0000 (15:25 -0500)]
Added to use cap when there're many messages to be released.

10 years agoadded missing new timer functions
Gengbin Zheng [Sun, 3 Apr 2011 15:39:40 +0000 (15:39 +0000)]
added missing new timer functions

10 years agofix a syntax error.
Gengbin Zheng [Sun, 3 Apr 2011 15:27:45 +0000 (10:27 -0500)]
fix a syntax error.

10 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Sat, 2 Apr 2011 22:59:16 +0000 (17:59 -0500)]
Merge branch 'charm' of charmgit:charm into charm

10 years agoCmiTimerInit() takes argv
Gengbin Zheng [Sat, 2 Apr 2011 22:58:51 +0000 (17:58 -0500)]
CmiTimerInit() takes argv

10 years agoadded some of the missing CmiTimerAbsolute()
Gengbin Zheng [Sat, 2 Apr 2011 22:52:10 +0000 (17:52 -0500)]
added some of the missing CmiTimerAbsolute()

10 years agoCmiTimerInit() now takes argv
Gengbin Zheng [Sat, 2 Apr 2011 22:51:11 +0000 (17:51 -0500)]
CmiTimerInit() now takes argv

10 years agofix for silly bug which gives the wrong number of cores per node
Abhinav S Bhatele [Sat, 2 Apr 2011 22:37:57 +0000 (17:37 -0500)]
fix for silly bug which gives the wrong number of cores per node

10 years ago Reset the send/recv cap when re-entering the send/recv functions.
Chao Mei [Sat, 2 Apr 2011 05:20:37 +0000 (00:20 -0500)]
 Reset the send/recv cap when re-entering the send/recv functions.

10 years agoAdded CMK_TRACE_ENABLED to guard the tracing of comm thread
Chao Mei [Fri, 1 Apr 2011 23:36:02 +0000 (18:36 -0500)]
Added CMK_TRACE_ENABLED to guard the tracing of comm thread

10 years agoadded semi-dynamic cap control for send/recv for smp mode. What semi-dynamic means...
Chao Mei [Fri, 1 Apr 2011 22:49:27 +0000 (17:49 -0500)]
added semi-dynamic cap control for send/recv for smp mode. What semi-dynamic means here is
in the sense  that the cap size for send/recv is fixed while the condition for exerting
cap is dynamic.

10 years agofixed a problem with the new +useAbsoluteTime, where the start bin time needs to...
Gengbin Zheng [Fri, 1 Apr 2011 01:43:40 +0000 (20:43 -0500)]
fixed a problem with the new +useAbsoluteTime, where the start bin time needs to be the absolute time as well (instead of 0).
also fixed a bug when begin/end idle and execute do not match. A checking is enforced.

10 years agoLB load+comm
Emmanuel Jeannot [Thu, 31 Mar 2011 22:43:49 +0000 (00:43 +0200)]
LB load+comm

10 years agoLB load+comm
Emmanuel Jeannot [Thu, 31 Mar 2011 22:43:37 +0000 (00:43 +0200)]
LB load+comm

10 years agoTimings debugged
Emmanuel Jeannot [Thu, 31 Mar 2011 18:48:29 +0000 (13:48 -0500)]
Timings debugged

10 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Thu, 31 Mar 2011 18:40:47 +0000 (13:40 -0500)]
Merge branch 'charm' of charmgit:charm into charm

10 years agoa new option +useAbsoluteTime time to converse timing module to force it to use...
Gengbin Zheng [Thu, 31 Mar 2011 18:38:15 +0000 (13:38 -0500)]
a new option +useAbsoluteTime time  to converse timing module to force it to use absolute time.
When +useAbsoluteTime is specified, projections will find global least init time automatically at the end of tracing. This retired the option +traceStartTime

10 years agodebug code
Emmanuel Jeannot [Thu, 31 Mar 2011 16:55:36 +0000 (18:55 +0200)]
debug code

10 years agoMerge branch 'charm' of charmgit:charm into charm
Emmanuel Jeannot [Thu, 31 Mar 2011 16:51:21 +0000 (18:51 +0200)]
Merge branch 'charm' of charmgit:charm into charm

10 years agotime_diff with no parameter
Emmanuel Jeannot [Thu, 31 Mar 2011 16:50:49 +0000 (18:50 +0200)]
time_diff with no parameter

10 years agoa new command line option: +traceStartTime for trace-projections
Gengbin Zheng [Thu, 31 Mar 2011 15:34:38 +0000 (10:34 -0500)]
a new command line option: +traceStartTime for trace-projections
at CkExit, it finds the min of start time on all processors and adjust logs before writing them to disks. This is useful if we use absolute wall timer (w/o substracting the artifical init_time)

10 years agoMerge branch 'charm' of charmgit:charm into charm
Emmanuel Jeannot [Wed, 30 Mar 2011 21:02:14 +0000 (23:02 +0200)]
Merge branch 'charm' of charmgit:charm into charm

10 years agofix retruned value of comm_speed
Emmanuel Jeannot [Wed, 30 Mar 2011 21:01:53 +0000 (23:01 +0200)]
fix retruned value of comm_speed

10 years agofixed a dependence to .def.h
Gengbin Zheng [Wed, 30 Mar 2011 20:47:24 +0000 (15:47 -0500)]
fixed a dependence to .def.h

10 years agoMerge branch 'charm' of charmgit:charm into charm
Emmanuel Jeannot [Wed, 30 Mar 2011 20:34:51 +0000 (22:34 +0200)]
Merge branch 'charm' of charmgit:charm into charm

10 years agofabs instead of abs
Emmanuel Jeannot [Wed, 30 Mar 2011 20:34:30 +0000 (22:34 +0200)]
fabs instead of abs

10 years agosplit CmiLog2 to CmiILog2 for int and CmiLog2 for double.
Gengbin Zheng [Wed, 30 Mar 2011 17:27:44 +0000 (12:27 -0500)]
split CmiLog2 to CmiILog2 for int and CmiLog2 for double.

10 years agochange so that new files can be compiled outside of charm.
Gengbin Zheng [Wed, 30 Mar 2011 17:11:29 +0000 (12:11 -0500)]
change so that new files can be compiled outside of charm.

10 years agomore minor changes
Gengbin Zheng [Wed, 30 Mar 2011 17:09:17 +0000 (12:09 -0500)]
more minor changes

10 years agofixed VC++ compilation errors
Gengbin Zheng [Wed, 30 Mar 2011 16:25:18 +0000 (11:25 -0500)]
fixed VC++ compilation errors

10 years agoMerge branch 'charm' of charmgit:charm into charm
Emmanuel Jeannot [Wed, 30 Mar 2011 16:09:42 +0000 (18:09 +0200)]
Merge branch 'charm' of charmgit:charm into charm

10 years agoDebug memory end segfault
Emmanuel Jeannot [Wed, 30 Mar 2011 16:07:49 +0000 (18:07 +0200)]
Debug memory end segfault

10 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Wed, 30 Mar 2011 14:49:10 +0000 (09:49 -0500)]
Merge branch 'charm' of charmgit:charm into charm

10 years agobigsimified variable startedEvac (which only affect smp version)
Gengbin Zheng [Wed, 30 Mar 2011 14:48:24 +0000 (09:48 -0500)]
bigsimified variable startedEvac (which only affect smp version)

10 years agoMerge branch 'charm' of charmgit:charm into charm
Chao Mei [Tue, 29 Mar 2011 22:06:00 +0000 (17:06 -0500)]
Merge branch 'charm' of charmgit:charm into charm

10 years agoTo reduce the overhead of tracing commthread by only recording very long MPI_Iprobe...
Chao Mei [Tue, 29 Mar 2011 22:04:55 +0000 (17:04 -0500)]
To reduce the overhead of tracing commthread by only recording very long MPI_Iprobe and MPI_Test (i.e. longer than 1ms)

10 years agofixed for VC++ windows
Gengbin Zheng [Tue, 29 Mar 2011 19:44:14 +0000 (14:44 -0500)]
fixed for VC++ windows

10 years agoupdated dependence rules for TreeMatchLB
Gengbin Zheng [Tue, 29 Mar 2011 19:38:45 +0000 (14:38 -0500)]
updated dependence rules for TreeMatchLB

10 years agoOverwrite changes
Emmanuel Jeannot [Tue, 29 Mar 2011 19:36:05 +0000 (21:36 +0200)]
Overwrite changes

10 years agoi
Emmanuel Jeannot [Tue, 29 Mar 2011 19:33:34 +0000 (21:33 +0200)]
i

10 years agosyntax error --
Emmanuel Jeannot [Tue, 29 Mar 2011 19:30:45 +0000 (21:30 +0200)]
syntax error --

10 years agoTreematch load balancer ++
Emmanuel Jeannot [Tue, 29 Mar 2011 19:11:48 +0000 (21:11 +0200)]
Treematch load balancer ++

10 years agoMerge branch 'charm' of charmgit:charm into charm
Emmanuel Jeannot [Tue, 29 Mar 2011 18:53:26 +0000 (20:53 +0200)]
Merge branch 'charm' of charmgit:charm into charm

10 years agoload balancer with taking inti account load of objects
Emmanuel Jeannot [Tue, 29 Mar 2011 18:52:46 +0000 (20:52 +0200)]
load balancer with taking inti account load of objects

10 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Tue, 29 Mar 2011 18:47:42 +0000 (13:47 -0500)]
Merge branch 'charm' of charmgit:charm into charm

10 years agofixed VC++.
Gengbin Zheng [Tue, 29 Mar 2011 18:47:19 +0000 (13:47 -0500)]
fixed VC++.

10 years agoturned off RDTSC timer.
Gengbin Zheng [Tue, 29 Mar 2011 16:07:17 +0000 (11:07 -0500)]
turned off RDTSC timer.

10 years agovarious fix for windows build.
Gengbin Zheng [Tue, 29 Mar 2011 01:04:35 +0000 (20:04 -0500)]
various fix for windows build.

10 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Mon, 28 Mar 2011 20:05:57 +0000 (15:05 -0500)]
Merge branch 'charm' of charmgit:charm into charm

10 years agoadded Prof Emmanuel Jeannot's tree match load balancer.
Gengbin Zheng [Mon, 28 Mar 2011 20:04:44 +0000 (15:04 -0500)]
added Prof Emmanuel Jeannot's tree match load balancer.

10 years agoMerge branch 'charm' of charmgit:charm into charm
Chao Mei [Mon, 28 Mar 2011 07:19:01 +0000 (02:19 -0500)]
Merge branch 'charm' of charmgit:charm into charm

10 years agoAdd CMK_TRACE_ENABLED for the previous check-in
Chao Mei [Mon, 28 Mar 2011 07:18:17 +0000 (02:18 -0500)]
Add CMK_TRACE_ENABLED for the previous check-in

10 years agoupdated charm version minor number due to the change in the previous commit in hybrid...
Gengbin Zheng [Mon, 28 Mar 2011 06:16:43 +0000 (01:16 -0500)]
updated charm version minor number due to the change in the previous commit in hybrid tree argument

10 years agoThreeLevelTree takes an argument to set group size.
Gengbin Zheng [Mon, 28 Mar 2011 06:01:09 +0000 (01:01 -0500)]
ThreeLevelTree takes an argument to set group size.

10 years agoRemove CMK_OPTIMIZE if tracing user events is enabled
Chao Mei [Mon, 28 Mar 2011 04:08:50 +0000 (23:08 -0500)]
Remove CMK_OPTIMIZE if tracing user events is enabled

10 years agoOnly print out info regarding cpuaffinity mapping on node 0.
Chao Mei [Sun, 27 Mar 2011 00:52:12 +0000 (19:52 -0500)]
Only print out info regarding cpuaffinity mapping on node 0.

10 years agoAdded codes to trace the MPI calls in the machine layer for non-smp run
Chao Mei [Sat, 26 Mar 2011 04:20:06 +0000 (23:20 -0500)]
Added codes to trace the MPI calls in the machine layer for non-smp run

10 years agodo nothing if adding a backward dependence.
Gengbin Zheng [Thu, 24 Mar 2011 16:50:08 +0000 (11:50 -0500)]
do nothing if adding a backward dependence.

10 years agofixed compiler warnings
Gengbin Zheng [Thu, 24 Mar 2011 16:49:35 +0000 (11:49 -0500)]
fixed compiler warnings

10 years agofixed a compiler warning
Gengbin Zheng [Thu, 24 Mar 2011 16:49:16 +0000 (11:49 -0500)]
fixed a compiler warning

10 years agomore load balancing time to use LBRealType.
Gengbin Zheng [Thu, 24 Mar 2011 06:48:16 +0000 (01:48 -0500)]
more load balancing time to use LBRealType.

10 years agovariable to store the original timer call is double always
Gengbin Zheng [Thu, 24 Mar 2011 06:45:00 +0000 (01:45 -0500)]
variable to store the original timer call is double always

10 years agoadd --with-lbtime-type in help
Gengbin Zheng [Thu, 24 Mar 2011 06:36:51 +0000 (01:36 -0500)]
add --with-lbtime-type in help

10 years agoa workaround for clang compiler, which crash at declaring array of size of negative...
Gengbin Zheng [Thu, 24 Mar 2011 06:24:06 +0000 (01:24 -0500)]
a workaround for clang compiler, which crash at declaring array of size of negative value.

10 years agoTopoMgrAPI: update the max nid on Jaguar
Abhinav S Bhatele [Thu, 24 Mar 2011 03:41:46 +0000 (22:41 -0500)]
TopoMgrAPI: update the max nid on Jaguar

10 years agoAdd explicit "this->" to comply with strict C++ two-level lookup rules
Aaron Becker [Wed, 23 Mar 2011 22:35:53 +0000 (17:35 -0500)]
Add explicit "this->" to comply with strict C++ two-level lookup rules

10 years agoUse CmiAbort instead of writing directly to null pointer.
Aaron Becker [Wed, 23 Mar 2011 22:32:19 +0000 (17:32 -0500)]
Use CmiAbort instead of writing directly to null pointer.

Clang complains about "*((void*)0) = 0;", and may optimize it away
unless the write is made to a volatile variable. Using CmiAbort
is more informative in any case.

10 years agoMerge branch 'charm' of charmgit:charm into charm
Aaron Becker [Wed, 23 Mar 2011 20:08:43 +0000 (15:08 -0500)]
Merge branch 'charm' of charmgit:charm into charm

10 years agoAdd clang to supported compilers
Aaron Becker [Wed, 23 Mar 2011 20:05:20 +0000 (15:05 -0500)]
Add clang to supported compilers

10 years agoUse explicit "this" to comply with strict C++ two-level naming rules
Aaron Becker [Wed, 23 Mar 2011 20:02:25 +0000 (15:02 -0500)]
Use explicit "this" to comply with strict C++ two-level naming rules

10 years agominor change
Gengbin Zheng [Tue, 22 Mar 2011 23:16:43 +0000 (18:16 -0500)]
minor change

10 years agoMinor change so that the pemap info is only printed out on proc 0 when it is loaded...
Chao Mei [Wed, 23 Mar 2011 06:54:04 +0000 (01:54 -0500)]
Minor change so that the pemap info is only printed out on proc 0 when it is loaded from a file.

10 years agoa new cmmand line option +pemapfile to specify a file name that contains the pemap.
Gengbin Zheng [Tue, 22 Mar 2011 21:39:36 +0000 (16:39 -0500)]
a new cmmand line option +pemapfile to specify a file name that contains the pemap.

10 years agoadded bigsim version of CkNodeSize and CkNodeFirst
Gengbin Zheng [Sat, 19 Mar 2011 06:15:52 +0000 (01:15 -0500)]
added bigsim version of CkNodeSize and CkNodeFirst

10 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Fri, 18 Mar 2011 21:29:11 +0000 (16:29 -0500)]
Merge branch 'charm' of charmgit:charm into charm

10 years agodefine SEQ compiler properly, instead of being like using icc and gcc at same time.
Gengbin Zheng [Fri, 18 Mar 2011 21:28:08 +0000 (16:28 -0500)]
define SEQ compiler properly, instead of being like using icc and gcc at same time.

10 years agofixed that bug that prevent charm converse broadcast to work on num of pes greater...
Gengbin Zheng [Fri, 18 Mar 2011 04:39:23 +0000 (23:39 -0500)]
fixed that bug that prevent charm converse broadcast to work on num of pes greater than 65536 (size of short). This was due to the use of Uint2 as "root" in converser header.
Fixed it by expanding it to a UInt4.
Prevent the header size from increasing by remove the error checking fields (magic and checksum) to be only used when --enable-error-checking.
Production build of mpi layer should not be affected.

10 years agomake the test shorter
Gengbin Zheng [Fri, 18 Mar 2011 04:34:36 +0000 (23:34 -0500)]
make the test shorter

10 years agominor changes
Gengbin Zheng [Fri, 18 Mar 2011 04:20:38 +0000 (23:20 -0500)]
minor changes

10 years agoMerge branch 'charm' of charmgit:charm into charm
Gengbin Zheng [Thu, 17 Mar 2011 07:37:34 +0000 (07:37 +0000)]
Merge branch 'charm' of charmgit:charm into charm

10 years agoremoved unnecessary reference of paper.
Gengbin Zheng [Thu, 17 Mar 2011 07:33:27 +0000 (07:33 +0000)]
removed unnecessary reference of paper.
We don't want to encourage this practice, otherwise we can have many papers crowded into charm source code.

10 years agoadded new reduction built-in type: nop
Gengbin Zheng [Thu, 17 Mar 2011 01:23:53 +0000 (20:23 -0500)]
added new reduction built-in type: nop

10 years agoadded a reduction operation "nop" which does nothing.
Gengbin Zheng [Thu, 17 Mar 2011 01:20:23 +0000 (20:20 -0500)]
added a reduction operation "nop" which does nothing.

10 years agoMerge branch 'charm' of charmgit:charm into charm
Chao Mei [Wed, 16 Mar 2011 22:30:05 +0000 (17:30 -0500)]
Merge branch 'charm' of charmgit:charm into charm

10 years agoAdded a flag to control whether to notify the children elements that a reduction...
Chao Mei [Wed, 16 Mar 2011 22:28:38 +0000 (17:28 -0500)]
Added a flag to control whether to notify the children elements that a reduction has started. Enabling this flag when users are sure that there's at least one element on each processor per chare array created. In such case, the reduction overhead will be reduced.

10 years agoEnable GROUP_LEVEL_TREE in ckreduction when the build is multicore or non-smp to...
Chao Mei [Wed, 16 Mar 2011 21:23:07 +0000 (16:23 -0500)]
Enable GROUP_LEVEL_TREE in ckreduction when the build is multicore or non-smp to reduce the overhead in reduction

10 years agoa minor fix
Gengbin Zheng [Wed, 16 Mar 2011 06:54:31 +0000 (01:54 -0500)]
a minor fix