charm.git
17 years agoadded separate make target for testing various components
Gengbin Zheng [Fri, 22 Oct 2004 16:27:24 +0000 (16:27 +0000)]
added separate make target for testing various components

17 years agoreplaced pgms by tests and examples.
Gengbin Zheng [Fri, 22 Oct 2004 16:26:52 +0000 (16:26 +0000)]
replaced pgms by tests and examples.

17 years agofixed a compiler error when timestamp is long
Gengbin Zheng [Fri, 22 Oct 2004 07:11:02 +0000 (07:11 +0000)]
fixed a compiler error when timestamp is long

17 years agoremoved libmoduleseqpose.dep
Gengbin Zheng [Fri, 22 Oct 2004 05:52:30 +0000 (05:52 +0000)]
removed libmoduleseqpose.dep
changes make target pose to POSE due to confliction with default rules.

17 years agoAdded a build case for pgm.seq
Terry L. Wilmarth [Fri, 22 Oct 2004 04:38:26 +0000 (04:38 +0000)]
Added a build case for pgm.seq

17 years agofixed make depends.
Gengbin Zheng [Fri, 22 Oct 2004 04:30:14 +0000 (04:30 +0000)]
fixed make depends.

17 years agoMore inlining, whoop-de-do.
Terry L. Wilmarth [Thu, 21 Oct 2004 21:16:48 +0000 (21:16 +0000)]
More inlining, whoop-de-do.

17 years agochange to use -module seqpose if seq mode
Gengbin Zheng [Thu, 21 Oct 2004 21:15:55 +0000 (21:15 +0000)]
change to use -module seqpose if seq mode

17 years agofixed mode with seq pose
Gengbin Zheng [Thu, 21 Oct 2004 21:15:20 +0000 (21:15 +0000)]
fixed mode with seq pose

17 years agowrite a fake registerseqpose() for seq pose. This is a hack because the moudle libmod...
Gengbin Zheng [Thu, 21 Oct 2004 21:06:29 +0000 (21:06 +0000)]
write a fake registerseqpose() for seq pose. This is a hack because the moudle libmoduleseqpose actually is just a rename of module and is not defined in ci file.

17 years agofixed make rules for compiling both pose and seqpose.
Gengbin Zheng [Thu, 21 Oct 2004 19:31:25 +0000 (19:31 +0000)]
fixed make rules for compiling both pose and seqpose.

17 years agoset CMK_LD_SHARED_ABSOLUTE_PATH to true
Gengbin Zheng [Thu, 21 Oct 2004 18:20:27 +0000 (18:20 +0000)]
set CMK_LD_SHARED_ABSOLUTE_PATH to true

17 years agoif CMK_LD_SHARED_ABSOLUTE_PATH is set, change libso path to absolute path
Gengbin Zheng [Thu, 21 Oct 2004 18:19:52 +0000 (18:19 +0000)]
if CMK_LD_SHARED_ABSOLUTE_PATH is set, change libso path to absolute path

17 years agoWas using different strategies in same simulation! Now all adapt3...
Terry L. Wilmarth [Thu, 21 Oct 2004 00:48:47 +0000 (00:48 +0000)]
Was using different strategies in same simulation!  Now all adapt3...

17 years agoAdded print out to check cordiantes of shared node
Sayantan Chakravorty [Wed, 20 Oct 2004 23:31:54 +0000 (23:31 +0000)]
Added print out to check cordiantes of shared node
Temporary

17 years agotwo changes for compiling csar code:
Gengbin Zheng [Wed, 20 Oct 2004 22:40:43 +0000 (22:40 +0000)]
two changes for compiling csar code:
1. apparently passing -D___CHARMC___ to both c and fortran compilers is not a good idea because f90 compiler does not like it often time, so cahnge it to send to c and cxx compiler only
2. for user provides parameter like  -framework vecLib, what happened to charmc was that it takes -framework as unrecognizable parameter and pass to real compiler, while abort since it does not know how to deal with "vecLib". Change it to pass to compiler anyway if this case also.

17 years agoFixing the IDXL sort2d. The old sort sorted points on the basis of distance from...
Sayantan Chakravorty [Wed, 20 Oct 2004 21:04:49 +0000 (21:04 +0000)]
Fixing the IDXL sort2d. The old sort sorted points on the basis of distance from the origin only. However this meant the sort was not unique if 2 points had the exact same distance from origin. To fix that points which have the same distance to an accuracy of 1e-100 are sorted by their x value

17 years agoAdded mapping from chunk to paneID for Roccom pconn routines.
Orion Lawlor [Wed, 20 Oct 2004 19:51:50 +0000 (19:51 +0000)]
Added mapping from chunk to paneID for Roccom pconn routines.

Roccom's "PaneID"'s are allocated in rather strange ways for
many applications, such as being numbered as
100*<1-based processor number> + <1-based patch number>
Hence we should just take this mapping as a parameter instead of
trying to hardcode (or postmortem modify) the mapping.

17 years agoMissing -Dcompile
Eric Bohm [Wed, 20 Oct 2004 15:13:43 +0000 (15:13 +0000)]
Missing -Dcompile

17 years agoForgot return statement in makeIDXLside.
Orion Lawlor [Wed, 20 Oct 2004 13:17:13 +0000 (13:17 +0000)]
Forgot return statement in makeIDXLside.

17 years agomodifed to recognize .dylib
Gengbin Zheng [Wed, 20 Oct 2004 05:03:58 +0000 (05:03 +0000)]
modifed to recognize .dylib

17 years agoadded support for dynamic shared lib.
Gengbin Zheng [Wed, 20 Oct 2004 04:53:59 +0000 (04:53 +0000)]
added support for dynamic shared lib.

17 years agompi version for Mac.
Gengbin Zheng [Wed, 20 Oct 2004 04:45:42 +0000 (04:45 +0000)]
mpi version for Mac.

17 years agoremove double declaration of traceCoreOn.
Gengbin Zheng [Wed, 20 Oct 2004 04:23:50 +0000 (04:23 +0000)]
remove double declaration of traceCoreOn.

17 years agoshould use $(MAKE) always to avoid gmake hassles on lemieux
Eric Bohm [Wed, 20 Oct 2004 03:47:35 +0000 (03:47 +0000)]
should use $(MAKE) always to avoid gmake hassles on lemieux

17 years agoAdded sequential support.
Eric Bohm [Wed, 20 Oct 2004 03:46:30 +0000 (03:46 +0000)]
Added sequential support.
make SEQUENTIAL=1 now builds the sequential version of the simulator.
It turns on the -DSEQUENTIAL_POSE flag, adds the -s to etrans, and
copies libmoduleseqpose to libmodulepose for linking so you get the
sequential pose.
Building without sequential copies the libmoduleorigpose back in so
parallel builds shouldn't be broken.

17 years agoquoted opts line takes off the curse on arch and lemieux
Eric Bohm [Wed, 20 Oct 2004 03:02:31 +0000 (03:02 +0000)]
quoted opts line takes off the curse on arch and lemieux

17 years agoAdded two new API routines:
Orion Lawlor [Wed, 20 Oct 2004 03:02:12 +0000 (03:02 +0000)]
Added two new API routines:
   void FEM_Make_node_globalno(int fem_mesh,FEM_Comm_t comm_context);
   void FEM_Set_roccom_pconn(int fem_mesh,const int *src,int total_len,int ghost_len);

These are needed by the CSAR Roccom remeshing tools.
The globalno procedure, in particular, could be useful
in other contexts.

17 years agoChange default mesh entity length and width from -1 to 0.
Orion Lawlor [Wed, 20 Oct 2004 03:00:17 +0000 (03:00 +0000)]
Change default mesh entity length and width from -1 to 0.
-1 causes a whole list of weird special casing inside FEM,
and isn't any better than 0 as a sentinal value.

The proximate cause of this checkin is that
FEM_Mesh_data(mesh,FEM_ELEM, ... , 0,0, ...);
(i.e., adding a zero-length element type) caused a
segfault during partitioning, because partitioning says
foo=new bar[mesh.elem[t].length()];
which is bad if length()==-1.  With this checkin,
zero-length elements are fine.

17 years agoAllow IDXL_Side::add to be used to add new communication
Orion Lawlor [Wed, 20 Oct 2004 02:57:23 +0000 (02:57 +0000)]
Allow IDXL_Side::add to be used to add new communication
partners, in addition to simple nodes.  This is needed by
FEM to upload a new IDXL list from outside.

17 years agoa quick fix (may have better solution - when fortran name has no score, in fact,...
Gengbin Zheng [Tue, 19 Oct 2004 21:52:56 +0000 (21:52 +0000)]
a quick fix (may have better solution - when fortran name has no score, in fact, driver() is choosen in fortran code which happen to conflict with the one defined in compat_driver.c. Simply skip the driver definition in case name is no scroe here.

17 years agopassing CMK_SHARED_SUF into making loibqt.so
Gengbin Zheng [Tue, 19 Oct 2004 18:49:23 +0000 (18:49 +0000)]
passing CMK_SHARED_SUF into making loibqt.so

17 years agoMoved -Dcompile out of OPTS and onto the appropriate lines so we can build without...
Terry L. Wilmarth [Tue, 19 Oct 2004 02:04:22 +0000 (02:04 +0000)]
Moved -Dcompile out of OPTS and onto the appropriate lines so we can build without -g and with -O for a change!!! Augh!

17 years agoSequential should not be the default!!! Augh!!!
Terry L. Wilmarth [Tue, 19 Oct 2004 02:03:03 +0000 (02:03 +0000)]
Sequential should not be the default!!! Augh!!!

17 years agoset up for hisim runs
Terry L. Wilmarth [Tue, 19 Oct 2004 01:57:17 +0000 (01:57 +0000)]
set up for hisim runs

17 years agoadd includes for math.h
Gengbin Zheng [Mon, 18 Oct 2004 18:29:16 +0000 (18:29 +0000)]
add includes for math.h

17 years agoturing off sorting message by timestamp in queue when not in timestamp correction...
Gengbin Zheng [Mon, 18 Oct 2004 06:25:36 +0000 (06:25 +0000)]
turing off sorting message by timestamp in queue when not in timestamp correction mode. This is for comparison of the prediction with and without timestamp correction.

17 years agofixed a bug in generating bgtimelog that has no dependence at all, in which case...
Gengbin Zheng [Mon, 18 Oct 2004 05:58:20 +0000 (05:58 +0000)]
fixed a bug in generating bgtimelog that has no dependence at all, in which case the event can float in timestamp correction.

17 years agoerrot check for BgLoadTraceSummary
Gengbin Zheng [Mon, 18 Oct 2004 05:55:09 +0000 (05:55 +0000)]
errot check for BgLoadTraceSummary

17 years agoadded error checking for BgLoadTraceSummary
Gengbin Zheng [Mon, 18 Oct 2004 05:43:24 +0000 (05:43 +0000)]
added error checking for BgLoadTraceSummary

17 years agowith a minor bug, when netsim off, turn of BgElapse of cpu time.
Gengbin Zheng [Mon, 18 Oct 2004 05:34:22 +0000 (05:34 +0000)]
with a minor bug, when netsim off, turn of BgElapse of cpu time.

17 years agoturning the flag off
Sameer Kumar [Mon, 18 Oct 2004 04:35:29 +0000 (04:35 +0000)]
turning the flag off

17 years agoUndo my earlier comment. sorry sameer...
Chao Huang [Mon, 18 Oct 2004 03:40:51 +0000 (03:40 +0000)]
Undo my earlier comment. sorry sameer...

17 years agoModified speculative tolerance in adapt3.
Terry L. Wilmarth [Mon, 18 Oct 2004 03:36:13 +0000 (03:36 +0000)]
Modified speculative tolerance in adapt3.

17 years agoadded a sanity check for existence of a special bglog that is floatable in timeline.
Gengbin Zheng [Mon, 18 Oct 2004 03:08:31 +0000 (03:08 +0000)]
added a sanity check for existence of a special bglog that is floatable in timeline.

17 years agoadded +bgelapse command line option to overwrite config file.
Gengbin Zheng [Mon, 18 Oct 2004 03:03:36 +0000 (03:03 +0000)]
added +bgelapse command line option to overwrite config file.

17 years agocomment out a printf in comlib that makes AMPI output look crazy
Chao Huang [Mon, 18 Oct 2004 02:58:41 +0000 (02:58 +0000)]
comment out a printf in comlib that makes AMPI output look crazy

17 years agogetting rid of calls to cbrt
Sameer Kumar [Sun, 17 Oct 2004 23:30:17 +0000 (23:30 +0000)]
getting rid of calls to cbrt

17 years agoadding a new cbrt function
Sameer Kumar [Sun, 17 Oct 2004 23:29:36 +0000 (23:29 +0000)]
adding a new cbrt function

17 years agoreverted my previous change since Sameer added the missing macro defintion back.
Gengbin Zheng [Sun, 17 Oct 2004 23:04:06 +0000 (23:04 +0000)]
reverted my previous change since Sameer added the missing macro defintion back.

17 years agoadding the persistent tag
Sameer Kumar [Sun, 17 Oct 2004 23:00:58 +0000 (23:00 +0000)]
adding the persistent tag

17 years agofixed compilation errors in Sameer's half abandoned comlib interface calling persiste...
Gengbin Zheng [Sun, 17 Oct 2004 22:48:13 +0000 (22:48 +0000)]
fixed compilation errors in Sameer's half abandoned comlib interface calling persistent interface.

17 years agoAdded libmoduleseqpose
Eric Bohm [Sun, 17 Oct 2004 22:40:36 +0000 (22:40 +0000)]
Added libmoduleseqpose
A regular make will create both libmodulepose and libmoduleseqpose.
libmoduleseqpose is just pose compiled with -DSEQUENTIAL_POSE=1.
Which means you can't simply link via -module seqpose because the module
linking system wants various registration functions to exist with the same
name as the module.  Attempts to create standalone seqpose modules failed
due to the bewildering variety of decl, def, etrans, and pose.h include file
namespace agony which result in irreconcilable netsim link failures.

But this cheap hack work as long as you copy it over.
Therefore we also create a libmoduleorigpose copy of the libmodulepose
during the make so we can restore normality later.

The Makefiles for the netsims will have SEQUENTIAL options to take care
of this jiggery pokery (including etrans -s) behind the scenes.
BigNetSim is already done.  HiSim is a bit more complicated.
"Evil hidden from the user is indistinguishable from Good." - Jim Phillips

17 years agofixed broken elan converse header for persistent comm.
Gengbin Zheng [Sun, 17 Oct 2004 22:26:23 +0000 (22:26 +0000)]
fixed broken elan converse header for persistent comm.

17 years agoapply -expect_unresolved to linker for suppressing warnings.
Gengbin Zheng [Sun, 17 Oct 2004 21:35:22 +0000 (21:35 +0000)]
apply -expect_unresolved to linker for suppressing warnings.

17 years agofixed test for malloc.h, only test if it presents and not check if malloc is actually...
Gengbin Zheng [Sun, 17 Oct 2004 21:22:16 +0000 (21:22 +0000)]
fixed test for malloc.h, only test if it presents and not check if malloc is actually defined in it.

17 years agorearrange converse header so that new comlib fields are placed same position in basic...
Gengbin Zheng [Sun, 17 Oct 2004 19:34:43 +0000 (19:34 +0000)]
rearrange converse header so that new comlib fields are placed same position in basic and ext headers, also fixed bluegene header for the latest change.

17 years agoInstead of abort immediate when send timeout, now retry it for 4 times, then abort
Gengbin Zheng [Sun, 17 Oct 2004 01:32:48 +0000 (01:32 +0000)]
Instead of abort immediate when send timeout, now retry it for 4 times, then abort

17 years agochanged to adapt2 because adapt3 is hanging.
Gengbin Zheng [Sat, 16 Oct 2004 23:16:24 +0000 (23:16 +0000)]
changed to adapt2 because adapt3 is hanging.

17 years agoadded tproj_on in config.
Gengbin Zheng [Sat, 16 Oct 2004 22:11:58 +0000 (22:11 +0000)]
added tproj_on in config.
remove commlib module in Makefile

17 years agosupported -tproj option, improved warning prints when task is not found.
Gengbin Zheng [Sat, 16 Oct 2004 22:11:12 +0000 (22:11 +0000)]
supported -tproj option, improved warning prints when task is not found.

17 years agowith -tproj to output projections from bgTrace
Gengbin Zheng [Sat, 16 Oct 2004 22:05:38 +0000 (22:05 +0000)]
with -tproj to output projections from bgTrace

17 years agoremoved commlib.
Gengbin Zheng [Sat, 16 Oct 2004 21:36:29 +0000 (21:36 +0000)]
removed commlib.

17 years agofixed charmc path
Gengbin Zheng [Sat, 16 Oct 2004 21:05:15 +0000 (21:05 +0000)]
fixed charmc path

17 years agoAdded MeshStreaming to pose.
Terry L. Wilmarth [Sat, 16 Oct 2004 20:34:43 +0000 (20:34 +0000)]
Added MeshStreaming to pose.

17 years agofixing smp and immediate message.
Gengbin Zheng [Sat, 16 Oct 2004 16:46:43 +0000 (16:46 +0000)]
fixing smp and immediate message.
Immediate message can arrive before nodegroup has been created, need to delay immediate message in this case.
another example of errornous removing old code without really know what it is doing.

17 years agosymlink to libmodulepose.dep
Eric Bohm [Fri, 15 Oct 2004 21:48:07 +0000 (21:48 +0000)]
symlink to libmodulepose.dep

17 years agoupdated.
Gengbin Zheng [Fri, 15 Oct 2004 20:00:53 +0000 (20:00 +0000)]
updated.

17 years agoextended a few command line options to overwrite bgconfig file definition.
Gengbin Zheng [Fri, 15 Oct 2004 19:42:58 +0000 (19:42 +0000)]
extended a few command line options to overwrite bgconfig file definition.
Also added new "projections" config in bgconfig file to choose processors with trace on.

17 years agocleanup the old hack for bigsim for selectively choosing subset of processors withe...
Gengbin Zheng [Fri, 15 Oct 2004 19:39:37 +0000 (19:39 +0000)]
cleanup the old hack for bigsim for selectively choosing subset of processors withe trace on, that require source code change to hack in the processor numbers.
Now work for both bigsim and normal charm++, with a new +traceprocessors 1,5-10
runtime options. In additionally, bigsim can also use bgconfig file to define these processor that generates trace logs.

17 years agoadded an additional parameter for num of iterations
Gengbin Zheng [Fri, 15 Oct 2004 18:43:55 +0000 (18:43 +0000)]
added an additional parameter for num of iterations
also added BgPrintf for bigsim (protected by BLUEGENE macro)

17 years agoadded a bgtest target
Gengbin Zheng [Fri, 15 Oct 2004 18:43:00 +0000 (18:43 +0000)]
added a bgtest target

17 years agofor bigsim
Gengbin Zheng [Fri, 15 Oct 2004 17:46:06 +0000 (17:46 +0000)]
for bigsim

17 years agocalled gathertree for examples.
Gengbin Zheng [Fri, 15 Oct 2004 06:37:43 +0000 (06:37 +0000)]
called gathertree for examples.
symbolic link ckliststring.h

17 years agofor BgPrint output, print target processor number for each line.
Gengbin Zheng [Fri, 15 Oct 2004 04:22:34 +0000 (04:22 +0000)]
for BgPrint output, print target processor number for each line.

17 years agodelegation bug fix for commlib
Sameer Kumar [Fri, 15 Oct 2004 02:27:19 +0000 (02:27 +0000)]
delegation bug fix for commlib

17 years agoimplemented CmiBarrier using elan_sync call.
Gengbin Zheng [Fri, 15 Oct 2004 02:26:40 +0000 (02:26 +0000)]
implemented CmiBarrier using elan_sync call.

17 years agoMore delegation changes
Sameer Kumar [Fri, 15 Oct 2004 02:26:34 +0000 (02:26 +0000)]
More delegation changes

17 years agoAdded CmiMemoryCheck definition
Sayantan Chakravorty [Thu, 14 Oct 2004 21:32:36 +0000 (21:32 +0000)]
Added CmiMemoryCheck definition

17 years agoAdded function definition for sparse edges
Sayantan Chakravorty [Thu, 14 Oct 2004 21:32:08 +0000 (21:32 +0000)]
Added function definition for sparse edges

17 years agoupdated the error output in building shared lib.
Gengbin Zheng [Thu, 14 Oct 2004 20:42:20 +0000 (20:42 +0000)]
updated the error output in building shared lib.

17 years agoadded name to atomic
Gengbin Zheng [Thu, 14 Oct 2004 20:05:09 +0000 (20:05 +0000)]
added name to atomic

17 years agoAdded some default definitions for methods of the refineClient.
Sayantan Chakravorty [Thu, 14 Oct 2004 19:49:40 +0000 (19:49 +0000)]
Added some default definitions for methods of the refineClient.
Their absence was causing crashes

17 years agosmall compiler fix in = operator
Sameer Kumar [Thu, 14 Oct 2004 19:26:13 +0000 (19:26 +0000)]
small compiler fix in = operator

17 years agonew interface to comlib, comlibinstancehandle now inherits from CkDelegateData
Sameer Kumar [Thu, 14 Oct 2004 17:42:52 +0000 (17:42 +0000)]
new interface to comlib, comlibinstancehandle now inherits from CkDelegateData

17 years agoAdding the new interface to access comm lib.
Sameer Kumar [Thu, 14 Oct 2004 17:42:19 +0000 (17:42 +0000)]
Adding the new interface to access comm lib.

17 years agoChanged all _nodeLock to _nodeGroupTableImmLock
Chao Huang [Thu, 14 Oct 2004 10:13:36 +0000 (10:13 +0000)]
Changed all _nodeLock to _nodeGroupTableImmLock

17 years agoupdated.
Gengbin Zheng [Thu, 14 Oct 2004 06:34:55 +0000 (06:34 +0000)]
updated.

17 years agoupdated bluegene to bigsim
Gengbin Zheng [Thu, 14 Oct 2004 06:12:10 +0000 (06:12 +0000)]
updated bluegene to bigsim

17 years agoupdated with some place holder sections.
Gengbin Zheng [Thu, 14 Oct 2004 06:10:26 +0000 (06:10 +0000)]
updated with some place holder sections.

17 years agoCarried over the changes related to sparse elements from the old tmr to
Sayantan Chakravorty [Wed, 13 Oct 2004 23:28:26 +0000 (23:28 +0000)]
Carried over the changes related to sparse elements from the old tmr to
the new TMR2D.

17 years agoa simple readme
Gengbin Zheng [Wed, 13 Oct 2004 21:52:22 +0000 (21:52 +0000)]
a simple readme

17 years agoAdded (probably misnamed) "elan" version, for use on
Orion Lawlor [Wed, 13 Oct 2004 21:20:51 +0000 (21:20 +0000)]
Added (probably misnamed) "elan" version, for use on
LLNL "alc" machine.  This is actually:
   - Xeon processors (ia32)
   - Quadrics Elan version of MPICH
   - Intel C/F90 compilers

The default mpi-linux version fails here, because
the "mpiicc" wrapper script puts -I/usr/local/mpi
*before* the user include files.  This makes it impossible
for AMPI to insert its own mpi.h; so here we call the
underlying compilers directly.

17 years agoCompleted the earlier fix by making sure that the turning off of the trace
Chee Wai Lee [Wed, 13 Oct 2004 06:25:51 +0000 (06:25 +0000)]
Completed the earlier fix by making sure that the turning off of the trace
module by the user affects the call to addThreadListener.
This is done by applying _TRACE_ONLY to ALLDO. It, in turn, required that
the definition of the _TRACE_ONLY and _TRACE_ALWAYS macros be moved further
up the code.

17 years agoa simple class that contains a list of integers in such a format: 1,5,100-200.
Gengbin Zheng [Wed, 13 Oct 2004 05:30:53 +0000 (05:30 +0000)]
a simple class that contains a list of integers in such a format: 1,5,100-200.
This is useful in couple places in trace and bigsim for selective subsets of processors.

17 years agoFixed crash error when thread listeners are used in conjunction with
Chee Wai Lee [Wed, 13 Oct 2004 05:09:30 +0000 (05:09 +0000)]
Fixed crash error when thread listeners are used in conjunction with
CMK_OPTIMIZE

17 years agonew config
Gengbin Zheng [Wed, 13 Oct 2004 04:17:38 +0000 (04:17 +0000)]
new config

17 years agoadded bgtest
Gengbin Zheng [Wed, 13 Oct 2004 04:17:16 +0000 (04:17 +0000)]
added bgtest

17 years agoAdded "test" targets to all Makefiles.
Orion Lawlor [Wed, 13 Oct 2004 03:28:06 +0000 (03:28 +0000)]
Added "test" targets to all Makefiles.

17 years agoAdded test targets to Makefiles.
Orion Lawlor [Wed, 13 Oct 2004 02:58:08 +0000 (02:58 +0000)]
Added test targets to Makefiles.