charm.git
17 years agoAdded note about the jade package.
Jayant DeSouza [Wed, 31 Mar 2004 18:27:12 +0000 (18:27 +0000)]
Added note about the jade package.

17 years agoAdded a new pass to the compiler for the for-loop manipulation.
Jayant DeSouza [Wed, 31 Mar 2004 18:23:19 +0000 (18:23 +0000)]
Added a new pass to the compiler for the for-loop manipulation.
Moved the default-package code into a package by making jade the package name.

17 years agofixed compiler error for mpcc64, fixed indents.
Gengbin Zheng [Wed, 31 Mar 2004 16:48:32 +0000 (16:48 +0000)]
fixed compiler error for mpcc64, fixed indents.

17 years agoAdded a function to parse a string and return an AST.
Jayant DeSouza [Wed, 31 Mar 2004 16:21:27 +0000 (16:21 +0000)]
Added a function to parse a string and return an AST.

17 years agoAdded comments.
Jayant DeSouza [Wed, 31 Mar 2004 16:20:14 +0000 (16:20 +0000)]
Added comments.

17 years agoFixed.
Yan Shi [Wed, 31 Mar 2004 16:11:27 +0000 (16:11 +0000)]
Fixed.

17 years agoChanged name for util.h for avoiding conflicts. Optimized inline function for complex...
Yan Shi [Wed, 31 Mar 2004 15:53:12 +0000 (15:53 +0000)]
Changed name for util.h for avoiding conflicts. Optimized inline function for complex class.

17 years agoOptimized loop. Still using double for reduction.
Yan Shi [Wed, 31 Mar 2004 15:51:03 +0000 (15:51 +0000)]
Optimized loop. Still using double for reduction.

17 years agochanged SEQ to NATIVE definitions to fix a bug that g++ was used for compiling charmxi.
Gengbin Zheng [Wed, 31 Mar 2004 08:10:37 +0000 (08:10 +0000)]
changed SEQ to NATIVE definitions to fix a bug that g++ was used for compiling charmxi.

17 years agoexamples in help page for using --incdir and --libdir.
Gengbin Zheng [Wed, 31 Mar 2004 03:11:24 +0000 (03:11 +0000)]
examples in help page for using --incdir and --libdir.

17 years agoadded writests for tracing functions in ampi
Sayantan Chakravorty [Tue, 30 Mar 2004 23:31:17 +0000 (23:31 +0000)]
added writests for tracing functions in ampi

17 years agoConfigure script updated to remove echo statement.
Chee Wai Lee [Tue, 30 Mar 2004 21:25:02 +0000 (21:25 +0000)]
Configure script updated to remove echo statement.

17 years agoUpdated trace modules to generate papi traces and accept AMPI user function
Chee Wai Lee [Tue, 30 Mar 2004 21:24:07 +0000 (21:24 +0000)]
Updated trace modules to generate papi traces and accept AMPI user function
tracing. The new version of projections is now 6.6

17 years agoAdded headers for registering and tracing ampi functions
Sayantan Chakravorty [Tue, 30 Mar 2004 21:10:42 +0000 (21:10 +0000)]
Added headers for registering and tracing ampi functions

17 years agoChanges for registering and tracing function calls
Sayantan Chakravorty [Tue, 30 Mar 2004 21:10:06 +0000 (21:10 +0000)]
Changes for registering and tracing function calls

17 years agoFlip back to sparseContigurous lib.
Yan Shi [Tue, 30 Mar 2004 06:00:45 +0000 (06:00 +0000)]
Flip back to sparseContigurous lib.

17 years agominor changes. added a command line option +gm_maxsize to set the the maximum packet...
Gengbin Zheng [Mon, 29 Mar 2004 20:32:12 +0000 (20:32 +0000)]
minor changes. added a command line option +gm_maxsize to set the the maximum packet size for packetization.

17 years agoadd a complete graph topology.
Gengbin Zheng [Mon, 29 Mar 2004 07:21:40 +0000 (07:21 +0000)]
add a complete graph topology.

17 years agoremoved debug output
Sayantan Chakravorty [Sun, 28 Mar 2004 02:37:47 +0000 (02:37 +0000)]
removed debug output

17 years agoinlining to get rid of the if in ComlibPrintf
Sameer Kumar [Fri, 26 Mar 2004 23:50:48 +0000 (23:50 +0000)]
inlining to get rid of the if in ComlibPrintf

17 years agoFix removal of comlibprintf under CMK_OPTIMIZE
Eric Bohm [Fri, 26 Mar 2004 22:07:57 +0000 (22:07 +0000)]
Fix removal of comlibprintf under CMK_OPTIMIZE

17 years agosetting size in envelope. Funny that I had to do it. Fixed bug anyway
Sameer Kumar [Fri, 26 Mar 2004 18:23:31 +0000 (18:23 +0000)]
setting size in envelope. Funny that I had to do it. Fixed bug anyway

17 years agofixed compiler errors due to default parameter in a function typedef, which is not...
Gengbin Zheng [Fri, 26 Mar 2004 17:23:39 +0000 (17:23 +0000)]
fixed compiler errors due to default parameter in a function typedef, which is not portable.

17 years agoDisabling get based send by default.
Sameer Kumar [Fri, 26 Mar 2004 16:35:34 +0000 (16:35 +0000)]
Disabling get based send by default.

17 years agoModified to use double for reduction instead of complex.
Yan Shi [Thu, 25 Mar 2004 23:22:35 +0000 (23:22 +0000)]
Modified to use double for reduction instead of complex.

17 years agosupport for async lb. TCHARM_Async_Migrate() starts a asyn lb without suspend. The...
Gengbin Zheng [Thu, 25 Mar 2004 21:45:08 +0000 (21:45 +0000)]
support for async lb.  TCHARM_Async_Migrate() starts a asyn lb without suspend. The actual migration happens when TCHARM_Allow_Migrate() is called.

17 years agochanges made for doing asynchronous laod balancing - when some array elements says...
Gengbin Zheng [Thu, 25 Mar 2004 20:48:46 +0000 (20:48 +0000)]
changes made for doing asynchronous laod balancing - when some array elements says AtSync(0), it tells load balancer that it doesnot care about the migration so that the end of load balancing does not depend on it. So that the program can run without waiting for its migration in the barrier.

17 years agosupport asynchronous load balancing, in which migration decisions from the last decis...
Gengbin Zheng [Thu, 25 Mar 2004 20:46:03 +0000 (20:46 +0000)]
support asynchronous load balancing, in which migration decisions from the last decision making phase is buffered and which an application to continue running without waiting for the migration messages barrier.

17 years agosome function/variable names change for aync load balancing.
Gengbin Zheng [Thu, 25 Mar 2004 20:18:58 +0000 (20:18 +0000)]
some function/variable names change for aync load balancing.

17 years agoseveral function name changes.
Gengbin Zheng [Thu, 25 Mar 2004 20:17:26 +0000 (20:17 +0000)]
several function name changes.

17 years agoa few features.
Gengbin Zheng [Thu, 25 Mar 2004 20:11:55 +0000 (20:11 +0000)]
a few features.
local barrier can now be control by anonymous client.
support for asynchronous migration when a migration request comes when an object just migrates off the processor already.

17 years agoThis modification fixes a race condition where two processes can enter
Greg Koenig [Thu, 25 Mar 2004 18:04:09 +0000 (18:04 +0000)]
This modification fixes a race condition where two processes can enter
CmiSyncSendFn() simultaneously and send RDMA messages to each other.  To
do this, they first send rendezvous messages and then VMI_Poll() to wait
for the RDMA to complete.  The error in this is that the Converse handler
in the rendezvous message never gets a chance to execute.

One fix was to call CsdScheduler(1) in the loop where VMI_Poll() runs.
This caused problems because you could end up cascading multiple calls to
CsdScheduler.

The solution I used was to go back to having CMI_VMI_Stream_Receive_Handler()
check directly to see if the message is a rendezvous setup and handle the
request itself.  This avoids the livelock condition since forward progress
can always happen.

There is a bug still in this code when running Orion's Charm++ marshall test
in megatest.  A fix to that problem is coming.

17 years agogetpid() is not quite portable, missing on several platforms. using mkstemp to genera...
Gengbin Zheng [Thu, 25 Mar 2004 05:44:27 +0000 (05:44 +0000)]
getpid() is not quite portable, missing on several platforms. using mkstemp to generate disk checkpoint filename.

17 years agoAdded a simple implementation of the "memory aliasing" idea for
Orion Lawlor [Thu, 25 Mar 2004 03:07:40 +0000 (03:07 +0000)]
Added a simple implementation of the "memory aliasing" idea for
thread stacks.  This is exactly like stack-copying threads,
except the copy happens only in virtual address space-- the
physical pages representing the stack never get moved.

This means memory aliased stacks:
   - Can be swapped very quickly (under 5us total penalty in AMPI)
   - Don't slow down with larger stacks
   - Don't consume large amounts of virtual address space like isomalloc stacks

Memory aliased stacks, however, will *NOT* work in SMP mode (because
all stacks get mapped to the same set of addresses), and
*will* break code that, e.g., swaps directly from one migratable
thread to the next without going through the scheduler, or creates
threads from inside a migratable thread (because the running stack
gets replaced with a new stack).

Because of the drawbacks, memory aliased stacks are off by default;
to enable them compile threads.c with -DCMK_THREADS_ALIAS_STACK=1.

17 years agoRemoving if statement surrounding ComlibPrintf when CML_OPTIMIZE is set.
Sameer Kumar [Thu, 25 Mar 2004 00:12:46 +0000 (00:12 +0000)]
Removing if statement surrounding ComlibPrintf when CML_OPTIMIZE is set.

17 years agofixed a typo, and move a global variable for timer into class.
Gengbin Zheng [Wed, 24 Mar 2004 23:57:46 +0000 (23:57 +0000)]
fixed a typo, and move a global variable for timer into class.

17 years agoAdded CMK_CF90_FIXED configuration target, for fixed-format
Orion Lawlor [Wed, 24 Mar 2004 23:37:30 +0000 (23:37 +0000)]
Added CMK_CF90_FIXED configuration target, for fixed-format
fortran source code.  Without this, the crucial "-stackvar"
flag doesn't get passed in, which causes all threaded f90
code to give wrong results.

17 years agoAdded in group object for section broadcast and reduction.
Yan Shi [Wed, 24 Mar 2004 22:47:16 +0000 (22:47 +0000)]
Added in group object for section broadcast and reduction.

17 years agofixed a bug in parsing +LBDump, tried to pass it twice.
Gengbin Zheng [Wed, 24 Mar 2004 19:48:46 +0000 (19:48 +0000)]
fixed a bug in parsing +LBDump, tried to pass it twice.

17 years agofixed getpid() which does not exist on win32.
Gengbin Zheng [Wed, 24 Mar 2004 19:31:17 +0000 (19:31 +0000)]
fixed getpid() which does not exist on win32.

17 years agoModified without the sparseContiguous library.
Yan Shi [Wed, 24 Mar 2004 18:02:35 +0000 (18:02 +0000)]
Modified without the sparseContiguous library.

17 years agomany greate changes. Restructured the checkpoint table to have separate class of...
Gengbin Zheng [Wed, 24 Mar 2004 07:01:59 +0000 (07:01 +0000)]
many greate changes. Restructured the checkpoint table to have separate class of in-memory and in-disk entries. calling fsync for in-disk checkpointing for correct timing.

17 years agofixed this silly bug: CkStartCheckpoint was only allowed to be called on PE 0, and...
Gengbin Zheng [Wed, 24 Mar 2004 06:54:08 +0000 (06:54 +0000)]
fixed this silly bug: CkStartCheckpoint was only allowed to be called on PE 0, and no error check.

17 years agofortran interface for the fem mesh refinement framework
Sayantan Chakravorty [Wed, 24 Mar 2004 00:33:02 +0000 (00:33 +0000)]
fortran interface for the fem mesh refinement framework

17 years agoreplaced FEM_NODE with the parameter nodeID in the split call
Sayantan Chakravorty [Tue, 23 Mar 2004 21:35:24 +0000 (21:35 +0000)]
replaced FEM_NODE with the parameter nodeID in the split call

17 years agoThe FEM_REFINE2D methods to refine the FEM mesh
Sayantan Chakravorty [Tue, 23 Mar 2004 21:31:21 +0000 (21:31 +0000)]
The FEM_REFINE2D methods to refine the FEM mesh

17 years agoExpose fem_mesh.C classes for use in the femrefine interface
Sayantan Chakravorty [Tue, 23 Mar 2004 21:30:29 +0000 (21:30 +0000)]
Expose fem_mesh.C classes for use in the femrefine interface

17 years agoRename MPI_Setup for Fortran binding to AMPI_Setup to avoid naming
Chao Huang [Tue, 23 Mar 2004 21:19:33 +0000 (21:19 +0000)]
Rename MPI_Setup for Fortran binding to AMPI_Setup to avoid naming
conflict with user code.

17 years agoSome decent default parameters.
Terry L. Wilmarth [Tue, 23 Mar 2004 17:28:47 +0000 (17:28 +0000)]
Some decent default parameters.

17 years agoAdded a few more ComlibPrintfs so users can figure out what is going on more easily
Eric Bohm [Tue, 23 Mar 2004 16:11:57 +0000 (16:11 +0000)]
Added a few more ComlibPrintfs so users can figure out what is going on more easily

17 years agoBug fixes for CMI_VMI_AsyncMsgCount, among others.
Greg Koenig [Tue, 23 Mar 2004 00:38:53 +0000 (00:38 +0000)]
Bug fixes for CMI_VMI_AsyncMsgCount, among others.

17 years agoChanged GVT, checkpointing strategy and rollback behavior.
Terry L. Wilmarth [Tue, 23 Mar 2004 00:01:21 +0000 (00:01 +0000)]
Changed GVT, checkpointing strategy and rollback behavior.

17 years agoFixing make files in pair calc
Sameer Kumar [Fri, 19 Mar 2004 21:16:17 +0000 (21:16 +0000)]
Fixing make files in pair calc

17 years agoMinor reworking of LiveViz:
Orion Lawlor [Fri, 19 Mar 2004 17:04:45 +0000 (17:04 +0000)]
Minor reworking of LiveViz:
  1.) Allow arbitrary extra data to be sent from client.
      This data is sent along in the liveVizRequestMsg, and
      can be extracted into a user data type via liveVizRequestUnpack.
      In particular, the 3d data is now treated as generic
      "extra client data", not as something built into liveViz.
  2.) Cleaned up liveVizPoll implementation.  In particular,
        a.) AtSync won't be needed if we just don't set usesAtSync.
        b.) liveVizPollRequestMsg is now the same as liveVizRequestMsg.

17 years agoBring along new pup headers and source files.
Orion Lawlor [Fri, 19 Mar 2004 16:53:01 +0000 (16:53 +0000)]
Bring along new pup headers and source files.

17 years agoMinor rearrangement of pup headers: added pup_toNetwork4.C
Orion Lawlor [Fri, 19 Mar 2004 16:52:37 +0000 (16:52 +0000)]
Minor rearrangement of pup headers: added pup_toNetwork4.C
and pup_toNetwork.[hC]

17 years agoMoved pup_toNetwork4 implementation to a separate file, and
Orion Lawlor [Fri, 19 Mar 2004 16:51:57 +0000 (16:51 +0000)]
Moved pup_toNetwork4 implementation to a separate file, and
added a new "pup_toNetwork" class that includes the ability
to pack 8-byte ints and doubles.

17 years ago*** empty log message ***
Chao Huang [Fri, 19 Mar 2004 08:48:55 +0000 (08:48 +0000)]
*** empty log message ***

17 years agoneeds to use pthreads.
Gengbin Zheng [Fri, 19 Mar 2004 07:10:26 +0000 (07:10 +0000)]
needs to use pthreads.

17 years agooption for mpi-gm with gm version 2.
Gengbin Zheng [Fri, 19 Mar 2004 07:09:43 +0000 (07:09 +0000)]
option for mpi-gm with gm version 2.

17 years agoAdd correct F90 libraries for linking with IBM xl f90.
Orion Lawlor [Fri, 19 Mar 2004 02:30:54 +0000 (02:30 +0000)]
Add correct F90 libraries for linking with IBM xl f90.
I'm not sure what fortran compiler the old linker commands
were going for, but the IBM compiler seems to be the most common.

17 years agoReplace error-prone "const int" paramters with
Orion Lawlor [Fri, 19 Mar 2004 01:06:18 +0000 (01:06 +0000)]
Replace error-prone "const int" paramters with
more portable (and more idomatic) enums.

17 years agoReplace *undefined* DEFAULT_MAX_BYTES with MSA_DEFAULT_MAX_BYTES.
Orion Lawlor [Fri, 19 Mar 2004 01:05:25 +0000 (01:05 +0000)]
Replace *undefined* DEFAULT_MAX_BYTES with MSA_DEFAULT_MAX_BYTES.
The only reason it even compiled before (on most machines)
was because the default parameter was never needed.
Compilers must be *terrified* of templates to not complain
about this; only IBM SP screwed up.

17 years agofix for CMK_MULTICAST_LIST_USE_COMMON_CODE.
Gengbin Zheng [Thu, 18 Mar 2004 18:11:01 +0000 (18:11 +0000)]
fix for CMK_MULTICAST_LIST_USE_COMMON_CODE.

17 years agoChanged the data field to public for ckcallback class, for the matrix pairwair librar...
Yan Shi [Thu, 18 Mar 2004 17:22:39 +0000 (17:22 +0000)]
Changed the data field to public for ckcallback class, for the matrix pairwair library used by CPMD. Just tempararily...

17 years agoturn on optimization for CmiListSend for GM version because it is not i/o interrupt...
Gengbin Zheng [Wed, 17 Mar 2004 23:14:32 +0000 (23:14 +0000)]
turn on optimization for CmiListSend for GM version because it is not i/o interrupt-based.

17 years agodefined CMK_MULTICAST_LIST_USE_COMMON_CODE back to 1 to disable the memory copy optim...
Gengbin Zheng [Wed, 17 Mar 2004 23:12:55 +0000 (23:12 +0000)]
defined CMK_MULTICAST_LIST_USE_COMMON_CODE back to 1 to disable the memory copy optimization as default for CmiListSend.

17 years agoMinor header rearrangement to work around MIPSpro 7.4
Orion Lawlor [Wed, 17 Mar 2004 21:06:05 +0000 (21:06 +0000)]
Minor header rearrangement to work around MIPSpro 7.4
compiler bug: #include <sys/time.h> as the first header
of a C++ program fails with bizarre compiler errors;
the fix is to #include <sys/types.h> first.

17 years agoFixed typo in comment.
Jayant DeSouza [Wed, 17 Mar 2004 19:19:54 +0000 (19:19 +0000)]
Fixed typo in comment.

17 years agoCreated a new block on the localStack when parsing a compound
Jayant DeSouza [Wed, 17 Mar 2004 19:19:01 +0000 (19:19 +0000)]
Created a new block on the localStack when parsing a compound
statement.
Also added method parameters to the localStack.

17 years agoSimplified testing for "chare", "chareArray1D" etc.
Jayant DeSouza [Wed, 17 Mar 2004 18:42:16 +0000 (18:42 +0000)]
Simplified testing for "chare", "chareArray1D" etc.

17 years agoSimplified testing for "MSA1D"/"MSA2D" strings.
Jayant DeSouza [Wed, 17 Mar 2004 18:34:17 +0000 (18:34 +0000)]
Simplified testing for "MSA1D"/"MSA2D" strings.

17 years ago*** empty log message ***
Jayant DeSouza [Wed, 17 Mar 2004 17:47:12 +0000 (17:47 +0000)]
*** empty log message ***

17 years agoAdded target for msa.
Jayant DeSouza [Wed, 17 Mar 2004 17:24:05 +0000 (17:24 +0000)]
Added target for msa.

17 years agoSupport for MSA in Jade.
Jayant DeSouza [Wed, 17 Mar 2004 17:21:40 +0000 (17:21 +0000)]
Support for MSA in Jade.

17 years agoa fix for queens example.
Gengbin Zheng [Tue, 16 Mar 2004 05:57:11 +0000 (05:57 +0000)]
a fix for queens example.

17 years agowhen CK_EP_TRACEDIABLE is set, the tracing of entry function is disabled.
Gengbin Zheng [Sun, 14 Mar 2004 22:27:07 +0000 (22:27 +0000)]
when CK_EP_TRACEDIABLE is set, the tracing of entry function is disabled.

17 years agofor an immediate message entry function, set CK_EP_TRACEDIABLE attrib bit.
Gengbin Zheng [Sun, 14 Mar 2004 22:25:59 +0000 (22:25 +0000)]
for an immediate message entry function, set CK_EP_TRACEDIABLE attrib bit.

17 years agominor changes in function names.
Gengbin Zheng [Sat, 13 Mar 2004 23:02:55 +0000 (23:02 +0000)]
minor changes in function names.

17 years agoimplemented two helper functions in CkLocMgr,
Gengbin Zheng [Sat, 13 Mar 2004 23:00:37 +0000 (23:00 +0000)]
implemented two helper functions in CkLocMgr,
(1) CkLocMgr::flushAllRecs()
    remove all records, the local record also will remove all CkMigratables.
(2) CkLocMgr::migratableList(CkLocRec_local *rec, ...)
    returns a list of CkMigratables belong to the same local Rec.

17 years agooops, mistakenly checked in ck.C with some garbage files. Revert these files.
Gengbin Zheng [Sat, 13 Mar 2004 22:34:20 +0000 (22:34 +0000)]
oops, mistakenly checked in ck.C with some garbage files. Revert these files.

17 years agoModified destructor of IrrGroup to clear the objPtr in the groupTable.
Gengbin Zheng [Sat, 13 Mar 2004 22:30:23 +0000 (22:30 +0000)]
Modified destructor of IrrGroup to clear the objPtr in the groupTable.
function CkDeleteChares()-odified

17 years agoBug fixes for my most recent changes.
Greg Koenig [Sat, 13 Mar 2004 22:03:48 +0000 (22:03 +0000)]
Bug fixes for my most recent changes.

Unfortunately there is a big problem involving either threads or memory
allocation in conjunction with VMI 2.0 Beta 3.  Somehow the stack gets
completely corrupted in a non-obvious way.  Switching to use pthreads in
Converse seems to help this greatly.  It is unknown whether this is related
directly to recent changes in Charm threads, recent changes in Charm memory
allocation (isomalloc?), or whether it's just an unfortunate clash with
changes in Charm + changes in VMI between Beta 2 and Beta 3.

This is going to be painful to fix.

17 years agofixed destructor of CentralLB.
Gengbin Zheng [Sat, 13 Mar 2004 22:01:06 +0000 (22:01 +0000)]
fixed destructor of CentralLB.

17 years agofixed destructors. Deleting of a load balancer may be after the deletion of LBDatabas...
Gengbin Zheng [Sat, 13 Mar 2004 21:37:17 +0000 (21:37 +0000)]
fixed destructors. Deleting of a load balancer may be after the deletion of LBDatabase group, that is the de-registering of callbacks in destructors may be invalid.

17 years agoBack out Chao's "bug fix" which:
Orion Lawlor [Sat, 13 Mar 2004 17:46:19 +0000 (17:46 +0000)]
Back out Chao's "bug fix" which:
  - Includes a debugging printout.
  - Hangs when MPI_Reduce is called from a split communicator.
  - Doesn't actually fix any bugs, because MPI_Barrier and
     MPI_Allreduce are both collective operations.

17 years agoclean up from last checkin.
Gengbin Zheng [Sat, 13 Mar 2004 17:32:16 +0000 (17:32 +0000)]
clean up from last checkin.

17 years agoClarified comments on which classes to use;
Orion Lawlor [Sat, 13 Mar 2004 06:19:07 +0000 (06:19 +0000)]
Clarified comments on which classes to use;
CkHashtableLayout is painfully difficult to
use properly, so we steer people to CkHashtableT.

17 years agoupdated.
Gengbin Zheng [Sat, 13 Mar 2004 05:00:48 +0000 (05:00 +0000)]
updated.

17 years agomoved common conv-mach-gm.h, conv-mach-gm.sh to mpi
Gengbin Zheng [Sat, 13 Mar 2004 04:59:58 +0000 (04:59 +0000)]
moved common conv-mach-gm.h, conv-mach-gm.sh to mpi

17 years agoFixed a potential bug: Barrier now does allreduce on one integer
Chao Huang [Fri, 12 Mar 2004 21:44:02 +0000 (21:44 +0000)]
Fixed a potential bug: Barrier now does allreduce on one integer
instead of NULL. Unmatched sendraw and recv was possible.

17 years agoGVT improvements and timer accuracy improvements.
Terry L. Wilmarth [Fri, 12 Mar 2004 21:31:05 +0000 (21:31 +0000)]
GVT improvements and timer accuracy improvements.

17 years agoreverted to previous version of using fixed mpi TAG. the dynamic tag version seem...
Gengbin Zheng [Fri, 12 Mar 2004 07:32:37 +0000 (07:32 +0000)]
reverted to previous version of using fixed mpi TAG. the dynamic tag version seem to have no advantage and increased some wierd crash due to invalid tag.

17 years agothe checkpoint/restart generation number is set starting from 1. 0 has special meanin...
Lixia Shi [Fri, 12 Mar 2004 05:22:02 +0000 (05:22 +0000)]
the checkpoint/restart generation number is set starting from 1. 0 has special meanings when one processor temporarily does not want to process any incoming messages.

17 years agoUpdated configure scripts to test for a build-time flag CMK_USE_PAPI before
Chee Wai Lee [Fri, 12 Mar 2004 02:23:06 +0000 (02:23 +0000)]
Updated configure scripts to test for a build-time flag CMK_USE_PAPI before
actually testing for the presence of papi. This allows charm++ to ignore
the papi library by default.

The user can explicitly build a new version of charm++ that uses papi by
the papi build option in the build script. This can be done orthogonal to
the simultaneous presence of a non-papi version of charm.

17 years agoSupport for papi option in build script. Unfortunately because of the way
Chee Wai Lee [Fri, 12 Mar 2004 02:20:51 +0000 (02:20 +0000)]
Support for papi option in build script. Unfortunately because of the way
build is written, the intuitive solution of putting #include <papi.h> in
conv-mach-papi.h cannot be done. This line will be included only when
the configure script has run, detecting the presence of the papi library.

conv-mach-papi.sh defines CMK_USE_PAPI which will be the switch that tells
the configure whether it should go ahead and test for papi.

17 years agoUpdated build script to support the papi build option.
Chee Wai Lee [Fri, 12 Mar 2004 02:18:13 +0000 (02:18 +0000)]
Updated build script to support the papi build option.

17 years agoFixed bug with newNbrRef initialization.
Terry L. Wilmarth [Thu, 11 Mar 2004 17:44:29 +0000 (17:44 +0000)]
Fixed bug with newNbrRef initialization.

17 years agothe minimum message size was set a little bit too high, set it back to 2^5.
Gengbin Zheng [Thu, 11 Mar 2004 04:07:27 +0000 (04:07 +0000)]
the minimum message size was set a little bit too high, set it back to 2^5.

17 years agonetwork version of CmiListSend of processor multicast.
Gengbin Zheng [Thu, 11 Mar 2004 01:58:22 +0000 (01:58 +0000)]
network version of CmiListSend of processor multicast.
It uses memory reference counter to reuse the buffer and avoid memory copying.
This is possible because CmiSyncSendAndFree() did not change the message buffer when it returns.