charm.git
18 years agoSubstantial change to how isomalloc searches for memory--
Orion Lawlor [Mon, 5 May 2003 22:06:05 +0000 (22:06 +0000)]
Substantial change to how isomalloc searches for memory--
instead of looking at just the ends of the block, check
the middle of the range as well.  Recursively breaks down
big bad blocks into smaller blocks when unmappable memory
is hit.

18 years agoadded +LBObjOnly option to turn on the mode in which LB only balance considering...
Gengbin Zheng [Fri, 2 May 2003 17:36:27 +0000 (17:36 +0000)]
added +LBObjOnly option to turn on the mode in which LB only balance considering obj load only, ignore all background load and nonmigratable load. (specifically for CPAIMD)

18 years agoreimplement sumonly mode(geenrate only one .sum that summarize across all pes) by...
Gengbin Zheng [Thu, 1 May 2003 06:06:21 +0000 (06:06 +0000)]
reimplement sumonly mode(geenrate only one .sum that summarize across all pes)  by group reduction.

18 years agoFix typos and extra routines. F90 actually tries to link
Orion Lawlor [Wed, 30 Apr 2003 00:09:55 +0000 (00:09 +0000)]
Fix typos and extra routines.  F90 actually tries to link
to these, even if they're never used in the source!

18 years agoAdded node numbering figure.
Orion Lawlor [Tue, 29 Apr 2003 23:53:10 +0000 (23:53 +0000)]
Added node numbering figure.

18 years agoAdded description of new MBLK_Get_nodelocs routine, with figure.
Orion Lawlor [Tue, 29 Apr 2003 23:53:02 +0000 (23:53 +0000)]
Added description of new MBLK_Get_nodelocs routine, with figure.

18 years agoAdded MBLK_Get_nodelocs, to get the node locations;
Orion Lawlor [Tue, 29 Apr 2003 23:26:29 +0000 (23:26 +0000)]
Added MBLK_Get_nodelocs, to get the node locations;
and changed .mblk to be a portable ASCII file.

18 years agoquick implementation to make projections to have object id for array elements. last...
Gengbin Zheng [Tue, 29 Apr 2003 22:52:07 +0000 (22:52 +0000)]
quick implementation to make projections to have object id for array elements. last minute change, hope don't break anything.

18 years agourgent fix to add array index as object id for projections.
Gengbin Zheng [Tue, 29 Apr 2003 22:47:59 +0000 (22:47 +0000)]
urgent fix to add array index as object id for projections.

18 years agolet Refineer deal with nonmigratables.
Gengbin Zheng [Tue, 29 Apr 2003 20:12:47 +0000 (20:12 +0000)]
let Refineer deal with nonmigratables.

18 years agofixed a bug when removing nonmigratable objects.
Gengbin Zheng [Tue, 29 Apr 2003 02:23:22 +0000 (02:23 +0000)]
fixed a bug when removing nonmigratable objects.

18 years agoConverted // comment into /* */ comment for some C compiler.
Chao Huang [Mon, 28 Apr 2003 19:10:22 +0000 (19:10 +0000)]
Converted // comment into /* */ comment for some C compiler.

18 years agoFix heap-corrupting bug: allocate a ProcState for the commserver
Orion Lawlor [Mon, 28 Apr 2003 18:21:17 +0000 (18:21 +0000)]
Fix heap-corrupting bug: allocate a ProcState for the commserver
thread, since the loop immediately below uses it.

18 years agoupdate GreedyLB
Gengbin Zheng [Sun, 27 Apr 2003 19:38:04 +0000 (19:38 +0000)]
update GreedyLB

18 years ago1. rename HeapCentLB to GreedyLB to be consistent with GreedyRefLB
Gengbin Zheng [Sun, 27 Apr 2003 17:51:20 +0000 (17:51 +0000)]
1. rename HeapCentLB to GreedyLB to be consistent with GreedyRefLB
2. LB SImulation reports biggest and lightest object load
3. CommLB cleanup

18 years agoa fix for running lb simulation mode for a fake number of processors using a fixed...
Gengbin Zheng [Sat, 26 Apr 2003 22:56:47 +0000 (22:56 +0000)]
a fix for running lb simulation mode for a fake number of processors using a fixed pes data.

18 years agoreflect change of file name
Gengbin Zheng [Fri, 25 Apr 2003 22:18:58 +0000 (22:18 +0000)]
reflect change of file name

18 years agoTiny change: give the number of chunks we're running on.
Orion Lawlor [Fri, 25 Apr 2003 21:17:53 +0000 (21:17 +0000)]
Tiny change: give the number of chunks we're running on.

18 years agosimulation now can read LBStats from disk file to M processors and call Strategy...
Gengbin Zheng [Fri, 25 Apr 2003 20:22:12 +0000 (20:22 +0000)]
simulation now can read LBStats from disk file to M processors and call Strategy() to balance a different N processors.

18 years agoPretend "envelope" is a struct, so C can understand it.
Orion Lawlor [Fri, 25 Apr 2003 14:54:45 +0000 (14:54 +0000)]
Pretend "envelope" is a struct, so C can understand it.
It's actually a class, in envelope.h; hopefully this duplicate
declaration won't cause any major problems.

18 years agoAdded ability to use 64-bit MPI on IBM SP's.
Orion Lawlor [Thu, 24 Apr 2003 20:52:55 +0000 (20:52 +0000)]
Added ability to use 64-bit MPI on IBM SP's.

18 years agoWhoops-- just use CkPointer, since the translator doesn't
Orion Lawlor [Thu, 24 Apr 2003 15:54:07 +0000 (15:54 +0000)]
Whoops-- just use CkPointer, since the translator doesn't
understand "foo *" as the first parameter (thinks it's a message!).

18 years agoModified the constructor that Orion wrote.
Ramkumar Vadali [Thu, 24 Apr 2003 02:57:10 +0000 (02:57 +0000)]
Modified the constructor that Orion wrote.

18 years agoSubstantially change both policy and implementation:
Orion Lawlor [Wed, 23 Apr 2003 22:08:00 +0000 (22:08 +0000)]
Substantially change both policy and implementation:
  - Added "idle" global flush criterion.
  - Added "bufferMax" local flush criterion.
Also substantially cleaned up code.

18 years agoAllow the user to create an actual Strategy and pass it to
Orion Lawlor [Wed, 23 Apr 2003 22:06:19 +0000 (22:06 +0000)]
Allow the user to create an actual Strategy and pass it to
the ComlibManager constructor.  This avoids the rather
ridiculous limitation that strategies can only take
one "int" as a constructor argument.

18 years agoFix memory leak:
Orion Lawlor [Wed, 23 Apr 2003 16:31:10 +0000 (16:31 +0000)]
Fix memory leak:
   allocated "result" and "ret", but only deleted "ret".
the fix is to only use "result".

18 years agoAdd "CHARM" define to all source code. This lets programs easily
Orion Lawlor [Tue, 22 Apr 2003 20:33:36 +0000 (20:33 +0000)]
Add "CHARM" define to all source code.  This lets programs easily
tell if they're being built under charm, so, e.g.:

#ifdef CHARM
CmiAbort("foo.");
#else
myAbort("foo.");
#endif

18 years agoLook for intel compiler in /opt, too.
Orion Lawlor [Tue, 22 Apr 2003 20:32:10 +0000 (20:32 +0000)]
Look for intel compiler in /opt, too.

18 years agoFix bug for ATAReqs. There was a memory leak.
Chao Huang [Tue, 22 Apr 2003 05:37:54 +0000 (05:37 +0000)]
Fix bug for ATAReqs. There was a memory leak.

18 years agofixed an array bound write bug and did a small optimization
Vikas Mehta [Mon, 21 Apr 2003 23:31:35 +0000 (23:31 +0000)]
fixed an array bound write bug and did a small optimization

18 years agoUpdated and expanded Rashmi's changes for modern tcharm.
Orion Lawlor [Mon, 21 Apr 2003 16:05:49 +0000 (16:05 +0000)]
Updated and expanded Rashmi's changes for modern tcharm.

18 years agoAdded MPI_Ialltoall, hack version only.
Chao Huang [Sat, 19 Apr 2003 05:39:23 +0000 (05:39 +0000)]
Added MPI_Ialltoall, hack version only.

18 years agoAdded description of makemblock.
Orion Lawlor [Fri, 18 Apr 2003 17:12:20 +0000 (17:12 +0000)]
Added description of makemblock.

18 years agoSkip the constructor for the registeredInfo table,
Orion Lawlor [Fri, 18 Apr 2003 16:21:40 +0000 (16:21 +0000)]
Skip the constructor for the registeredInfo table,
because it may already be initialized by the time
the constructor gets called:
  - Milind's bizarre Readonly<foo> class calls CkRegisterReadonly
    from its constructor.  This class's constructor gets called
    while setting up global variables.
  - If _readonlyTable gets initialized *after* Readonly<foo>,
    foo's registration in the table will be overwritten by
    the call to the constructor.  This happens on Solaris CC
    and on the IBM SP.

This (horrific) solution--disable the constructor for
_readonlyTable-- relies on the zero initialization for
global variables being appropriate for CkVec's.
The right solution is probably to remove the silly Readonly
class, or have it delay calling CkRegisterReadonly until
after all globals have been set up.

18 years agoAdded "CkSkipInitialization" class, which is used to
Orion Lawlor [Fri, 18 Apr 2003 16:17:54 +0000 (16:17 +0000)]
Added "CkSkipInitialization" class, which is used to
indicate the constructor *shouldn't* touch the class's
values.  This is a workaround for a global variable
initialization ordering bug described in more
detail in the comments on register.h.

18 years agoUpdated this file to match the new _entryTable object.
Orion Lawlor [Fri, 18 Apr 2003 14:10:12 +0000 (14:10 +0000)]
Updated this file to match the new _entryTable object.

18 years agoMinor changes to match with _entryTable now being a real class,
Orion Lawlor [Thu, 17 Apr 2003 21:15:19 +0000 (21:15 +0000)]
Minor changes to match with _entryTable now being a real class,
so _numEntries becomes _entryTable.size().  Also factored out
hideous duplicated .sts-writing code from trace-projections, projector,
and summary into trace-common.

18 years agoMajor registration cleanup:
Orion Lawlor [Thu, 17 Apr 2003 21:14:11 +0000 (21:14 +0000)]
Major registration cleanup:
 - Made _entryTable (& friends) an actual class, which
   avoids the problem of running out of a fixed allocation
   and provides a cleaner interface.
 - Moved call to _REGISTER_DONE into init.C
   from translated code.
 - Renamed _REGISTER_BASE as CkRegisterBase
   to match with the other routines.
 - Removed obsolete _GET_IDX, _CHECK_CID, and classIdx.
   These used to be used for runtime CkChareID typecast
   checking, now obsolete because of proxies.
 - Removed obsolete CkCoerceFn for messages.

18 years agoChanged checkpoint/restart to create 1 file per PE.
Chao Huang [Wed, 16 Apr 2003 04:59:55 +0000 (04:59 +0000)]
Changed checkpoint/restart to create 1 file per PE.

18 years agoWork around a crashing bug on win32--somehow the destructor
Orion Lawlor [Tue, 15 Apr 2003 21:49:12 +0000 (21:49 +0000)]
Work around a crashing bug on win32--somehow the destructor
for mpi_comm_world is getting called twice, so we have to
zero out our "name" field after deleting it (the first time!)
in the destructor.

18 years agoFix hanging bug on win32-- condition on sleep is different
Orion Lawlor [Tue, 15 Apr 2003 18:32:36 +0000 (18:32 +0000)]
Fix hanging bug on win32-- condition on sleep is different
from termination condition.

18 years agoChanged name of "install" target to "web", just like manuals.
Orion Lawlor [Tue, 15 Apr 2003 17:03:43 +0000 (17:03 +0000)]
Changed name of "install" target to "web", just like manuals.

18 years ago*** empty log message ***
Rashmi Jyothi [Tue, 15 Apr 2003 15:03:01 +0000 (15:03 +0000)]
*** empty log message ***

18 years ago*** empty log message ***
Rashmi Jyothi [Tue, 15 Apr 2003 13:21:23 +0000 (13:21 +0000)]
*** empty log message ***

18 years agolibrary instrns
Rashmi Jyothi [Tue, 15 Apr 2003 12:36:22 +0000 (12:36 +0000)]
library  instrns

18 years agochanged syntax for dynamic array allocation in cksparsecontiguousreducer.h also
Vikas Mehta [Tue, 15 Apr 2003 03:11:21 +0000 (03:11 +0000)]
changed syntax for dynamic array allocation in cksparsecontiguousreducer.h also

18 years agochanged array allocation syntax at all the places so that it compiles on lemieux
Vikas Mehta [Tue, 15 Apr 2003 03:00:15 +0000 (03:00 +0000)]
changed array allocation syntax at all the places so that it compiles on lemieux

18 years agochanged dynamic array allocation syntax for 'data' so that it complies on lemieux
Vikas Mehta [Tue, 15 Apr 2003 02:57:12 +0000 (02:57 +0000)]
changed dynamic array allocation syntax for 'data' so that it complies on lemieux

18 years agoAdd a non-mmap'able region to the IBM SP address space. I'm not sure
Orion Lawlor [Mon, 14 Apr 2003 21:40:04 +0000 (21:40 +0000)]
Add a non-mmap'able region to the IBM SP address space.  I'm not sure
what the heck goes in this hole (kernel data structures?):

3200 MB: |                                                                |
3264 MB: |                                                                |
3328 MB: |................................................................|
3392 MB: |................................................................|
3456 MB: |................................................................|
3520 MB: |................................................................|
3584 MB: |                                                                |
3648 MB: |                                                                |

This should eventually be determined automatically, but I'm not quite
sure how to best do that yet.

18 years agoAdded ability to define CMK_BAD_MMAP_ADDRESS in the conv-mach.h files,
Orion Lawlor [Mon, 14 Apr 2003 21:36:35 +0000 (21:36 +0000)]
Added ability to define CMK_BAD_MMAP_ADDRESS in the conv-mach.h files,
which lets me work around a bizarre hole in the address space of
the IBM SP.  Eventually, these holes should be automatically determined,
for example, via a "mmap survey" either at configure or run time.

18 years agoMake debugLog per-node, and created earlier.
Orion Lawlor [Mon, 14 Apr 2003 20:57:53 +0000 (20:57 +0000)]
Make debugLog per-node, and created earlier.

18 years agoTweak DgramHeader declaration to make it 8 bytes long on win32,
Orion Lawlor [Mon, 14 Apr 2003 20:57:30 +0000 (20:57 +0000)]
Tweak DgramHeader declaration to make it 8 bytes long on win32,
and add a check to make sure it *stays* 8 bytes long.

18 years agoUse MAP_FIXED even for MAP_ANON. Solaris requires MAP_FIXED,
Orion Lawlor [Mon, 14 Apr 2003 15:23:03 +0000 (15:23 +0000)]
Use MAP_FIXED even for MAP_ANON.  Solaris requires MAP_FIXED,
or else it ignores the "addr" argument you pass in.

18 years ago1. fixed an ancient bug in load balancer. ClearLoad() which clears all load balancing...
Gengbin Zheng [Mon, 14 Apr 2003 05:34:31 +0000 (05:34 +0000)]
1. fixed an ancient bug in load balancer. ClearLoad() which clears all load balancing data and reset the wall clock timer should be called after load balancing finishs instead of at the point of AtSync holds. This is because otherwise the load on processor 0 may be very high because the time spent in load balancing will be counted as its background load.
2. sequential load balancer(in simulation mode) now works in the way that one application can read dumped load data from another application. It didn't work because when unpack a load stats message, the msgIdx in envelope from another application may be different with the one in the current application.

18 years agoAdded documentation for CommLib support for MPI_Alltoall
Chao Huang [Sat, 12 Apr 2003 16:46:42 +0000 (16:46 +0000)]
Added documentation for CommLib support for MPI_Alltoall

18 years agoAdded USE_HYPERCUBE for CommLib support
Chao Huang [Sat, 12 Apr 2003 16:44:24 +0000 (16:44 +0000)]
Added USE_HYPERCUBE for CommLib support

18 years agoChange QT from solaris-cc, which segfaults on the first
Orion Lawlor [Fri, 11 Apr 2003 21:40:18 +0000 (21:40 +0000)]
Change QT from solaris-cc, which segfaults on the first
context switch under Sun Workshop 6, to generic_alloca.

18 years agoJust like "generic", but define "ALLOCA_H".
Orion Lawlor [Fri, 11 Apr 2003 21:37:46 +0000 (21:37 +0000)]
Just like "generic", but define "ALLOCA_H".
This is needed on Suns to get alloca().

18 years agoRemove duplicate "int i" from for loop.
Orion Lawlor [Fri, 11 Apr 2003 20:39:40 +0000 (20:39 +0000)]
Remove duplicate "int i" from for loop.

18 years agoAdded a tiny coding guidelines section.
Orion Lawlor [Fri, 11 Apr 2003 19:30:05 +0000 (19:30 +0000)]
Added a tiny coding guidelines section.

18 years agoAdded (almost useless) doxygen header.
Orion Lawlor [Fri, 11 Apr 2003 17:48:09 +0000 (17:48 +0000)]
Added (almost useless) doxygen header.

18 years agoAdded zillions of doxygen comments.
Orion Lawlor [Fri, 11 Apr 2003 17:47:53 +0000 (17:47 +0000)]
Added zillions of doxygen comments.

18 years agoMany tweaks to config file:
Orion Lawlor [Fri, 11 Apr 2003 17:46:55 +0000 (17:46 +0000)]
Many tweaks to config file:
  - Less annoying, and more useful output.
  - Skip useless collaboration graph generation with "dot" because:
      1.) Dot isn't installed on any Linux box by default.
      2.) Dot, when installed, is very very slow (30min with; 2 min without!)
      3.) Collaboration diagrams don't tell you very much.
  - Use autobrief.
  - Include src/util directory.

18 years agoMore details on how to run the thing.
Orion Lawlor [Fri, 11 Apr 2003 17:44:45 +0000 (17:44 +0000)]
More details on how to run the thing.

18 years agoAdded this master page, to point to various sub-pages.
Orion Lawlor [Fri, 11 Apr 2003 17:44:19 +0000 (17:44 +0000)]
Added this master page, to point to various sub-pages.

18 years agoFixed name of TCHARM_Create routine.
Orion Lawlor [Fri, 11 Apr 2003 16:59:06 +0000 (16:59 +0000)]
Fixed name of TCHARM_Create routine.

18 years agoFixed checkpoint problem with LBDB.
Chao Huang [Fri, 11 Apr 2003 01:45:20 +0000 (01:45 +0000)]
Fixed checkpoint problem with LBDB.

18 years agostopped the creation of an extra nodegroup for irrgroups
Sayantan Chakravorty [Thu, 10 Apr 2003 23:40:48 +0000 (23:40 +0000)]
stopped the creation of an extra nodegroup for irrgroups

18 years agoadded prototype of new decompressor methods in cksparsecontiguousreducer.h
Vikas Mehta [Thu, 10 Apr 2003 21:41:39 +0000 (21:41 +0000)]
added prototype of new decompressor methods in cksparsecontiguousreducer.h

18 years agoBack out std::ostream-based ckout.
Orion Lawlor [Thu, 10 Apr 2003 17:53:55 +0000 (17:53 +0000)]
Back out std::ostream-based ckout.

18 years agoBack out std::ostream-based ckout, because:
Orion Lawlor [Thu, 10 Apr 2003 17:53:32 +0000 (17:53 +0000)]
Back out std::ostream-based ckout, because:
  - On DEC Alpha cxx, std::streambuf isn't in <iostream>,
    but <streambuf> isn't everywhere, so we'd need a
    configuration test.
  - On the SGI Origin2000, std::ostream results in a
    segfault.
  - Under Intel's icc, <iostream> screws up routines
    named "init".
  - Under Portland Group's pgcc, I get a link error.

Basically, it seems C++ compilers aren't yet mature enough to
reliably use custom ostreams.  Too bad...

18 years agoWrite bitvectors to regular std::ostreams instead of
Orion Lawlor [Thu, 10 Apr 2003 15:38:50 +0000 (15:38 +0000)]
Write bitvectors to regular std::ostreams instead of
bizarre CkOutStream.  This is more portable, and
required for stuff like:
ckout<<"Foo: "<<myBitvector;
because the operator<<(std::ostream,char *) returns
a std::ostream, not a CkOutStream.

18 years agoJust "using std::endl", because so many people
Orion Lawlor [Thu, 10 Apr 2003 15:37:54 +0000 (15:37 +0000)]
Just "using std::endl", because so many people
forget to include the (nonportable!) "std::".

18 years agoMoved ckout/ckerr setup into ckstream.C.
Orion Lawlor [Thu, 10 Apr 2003 15:35:56 +0000 (15:35 +0000)]
Moved ckout/ckerr setup into ckstream.C.

18 years agoCalling PUP::able::pup is a good idea, but it
Orion Lawlor [Thu, 10 Apr 2003 03:39:11 +0000 (03:39 +0000)]
Calling PUP::able::pup is a good idea, but it
confuses Visual C++ 6, and isn't actually needed.

18 years agoUse new ckstream.C source file.
Orion Lawlor [Thu, 10 Apr 2003 03:35:27 +0000 (03:35 +0000)]
Use new ckstream.C source file.

18 years agoUse regular C++ ostream for ckout, instead of
Orion Lawlor [Thu, 10 Apr 2003 03:34:56 +0000 (03:34 +0000)]
Use regular C++ ostream for ckout, instead of
defining our own mutant CkOStream type.
This lets people write operator<<'s for regular
std::ostream and still use ckout.

18 years agoFix win32 version.
Orion Lawlor [Thu, 10 Apr 2003 02:24:40 +0000 (02:24 +0000)]
Fix win32 version.

18 years agoRemoved call to PUP::able::pup. This is normally
Orion Lawlor [Wed, 9 Apr 2003 22:41:32 +0000 (22:41 +0000)]
Removed call to PUP::able::pup.  This is normally
OK, but it severely confuses Visual C++ 6.0 and
can safely be removed.

18 years agoRename .C files .cpp for easy use with Windows IDE's.
Orion Lawlor [Wed, 9 Apr 2003 22:40:51 +0000 (22:40 +0000)]
Rename .C files .cpp for easy use with Windows IDE's.

18 years agoRemove incorrect "const" on message.
Orion Lawlor [Wed, 9 Apr 2003 22:39:48 +0000 (22:39 +0000)]
Remove incorrect "const" on message.

18 years agoUse PUP::able's builtin "clone" method, instead of building our own.
Orion Lawlor [Wed, 9 Apr 2003 22:39:37 +0000 (22:39 +0000)]
Use PUP::able's builtin "clone" method, instead of building our own.

18 years agoRemoved unportable duplicate initialization
Orion Lawlor [Wed, 9 Apr 2003 22:39:18 +0000 (22:39 +0000)]
Removed unportable duplicate initialization
of variables in for loops.  Even though this
is legal under ISO, many compilers, including
MS Visual C++ 6, don't allow it.

18 years agoRemoved call to pupCkVec, since it didn't actually
Orion Lawlor [Wed, 9 Apr 2003 22:38:48 +0000 (22:38 +0000)]
Removed call to pupCkVec, since it didn't actually
fix the problem it was intended to solve.

18 years agoDefine pup routine for CkVec under MS Visual C++ 6.0.
Orion Lawlor [Wed, 9 Apr 2003 22:38:36 +0000 (22:38 +0000)]
Define pup routine for CkVec under MS Visual C++ 6.0.
This is needed to get around a "ambiguous operator|"
problem present only on that compiler.

18 years agoAdded search library into the list of targets to be built.
Jonathan Booth [Wed, 9 Apr 2003 18:50:10 +0000 (18:50 +0000)]
Added search library into the list of targets to be built.

18 years agoFixed the delete character pointers in the print functions to be a
Jonathan Booth [Wed, 9 Apr 2003 18:46:28 +0000 (18:46 +0000)]
Fixed the delete character pointers in the print functions to be a
delete [] pointer like it should be.

18 years agoChecking in my idaStar search tree library. Liberally commented, not
Jonathan Booth [Wed, 9 Apr 2003 18:27:14 +0000 (18:27 +0000)]
Checking in my idaStar search tree library. Liberally commented, not
yet documented. I'll check in an example in the pgms directory along
with documentation later, once I can do that while waiting on data
runs.

18 years agoVersion 2.0 of collision library:
Orion Lawlor [Tue, 8 Apr 2003 21:47:14 +0000 (21:47 +0000)]
Version 2.0 of collision library:
  - Cleanly separated utility classes from API implementation.
  - Provides clean Charm and TCharm API's.
  - Allows collision priority to be specified (useful for,
    among other things, data transfer).
  - No explicit assumption about a small number of collisions.
  - Massive source reorganization.

18 years agoAdded "TCharm Semaphores", a bizarre way of moving pointers
Orion Lawlor [Tue, 8 Apr 2003 21:34:23 +0000 (21:34 +0000)]
Added "TCharm Semaphores", a bizarre way of moving pointers
between threads.  I'm not happy with the interface, so
this one's undocumented for now.

18 years agocorrected the name of reducer
Vikas Mehta [Tue, 8 Apr 2003 20:54:38 +0000 (20:54 +0000)]
corrected the name of reducer

18 years agoRemove duplicate loop counter declarations, which
Orion Lawlor [Tue, 8 Apr 2003 20:42:06 +0000 (20:42 +0000)]
Remove duplicate loop counter declarations, which
confuse Visual C++ 6.0 (since it uses pre-ISO for
loop declaration scoping).

18 years agoadded overloaded contribute function for cpaimd.
Vikas Mehta [Tue, 8 Apr 2003 19:56:21 +0000 (19:56 +0000)]
added overloaded contribute function for cpaimd.

18 years agodefined new reducer types!
Vikas Mehta [Tue, 8 Apr 2003 19:43:35 +0000 (19:43 +0000)]
defined new reducer types!

18 years agoadded new reducers which can be used for complex data types.
Vikas Mehta [Tue, 8 Apr 2003 18:57:45 +0000 (18:57 +0000)]
added new reducers which can be used for complex data types.

18 years agoSeveral portability and correctness fixes:
Orion Lawlor [Tue, 8 Apr 2003 18:45:41 +0000 (18:45 +0000)]
Several portability and correctness fixes:
  - Pass entire, terminated directory string rather
    than overwriting memory to add terminator on receive side.

  - Replace error-prone and unportable "chdir" call with
    explicit directory paths.

  - Replace "mkdir" call with new, win32-compatible "CmiMkdir".

  - Replace unportable ISO "int i" declarations in for loop
    with portable single declaration.

18 years agoCorrectness fix: pass entire, null-terminated string in
Orion Lawlor [Tue, 8 Apr 2003 18:43:25 +0000 (18:43 +0000)]
Correctness fix: pass entire, null-terminated string in
message rather than (wrongly) pasting in null termination
afterwards.

18 years agoAdded CmiMkdir, a portable directory creation call.
Orion Lawlor [Tue, 8 Apr 2003 18:41:14 +0000 (18:41 +0000)]
Added CmiMkdir, a portable directory creation call.
This really only exists because Win32 is so stupid.

18 years agoChange constants to unsigned to eliminate compiler warnings.
Orion Lawlor [Tue, 8 Apr 2003 18:40:28 +0000 (18:40 +0000)]
Change constants to unsigned to eliminate compiler warnings.

18 years agofixed a bug in tracemode all
Gengbin Zheng [Tue, 8 Apr 2003 04:30:24 +0000 (04:30 +0000)]
fixed a bug in tracemode all