charm.git
18 years agoAdd LOGCONDITIONAL protection around RegisterLanguage and
Orion Lawlor [Sat, 30 Aug 2003 02:42:03 +0000 (02:42 +0000)]
Add LOGCONDITIONAL protection around RegisterLanguage and
RegisterEvent calls (before they would segfault if called
under CMK_OPTIMIZE).

18 years agoAdd funny do{}while(0) protection to LOGCONDITIONAL macro.
Orion Lawlor [Sat, 30 Aug 2003 02:41:13 +0000 (02:41 +0000)]
Add funny do{}while(0) protection to LOGCONDITIONAL macro.
This lets LOGCONDITIONAL be used properly inside an "if"
statement (before it might screw up a subseqent "else").

18 years agoLink in all icc libraries dynamically; this works properly with
Orion Lawlor [Sat, 30 Aug 2003 02:33:49 +0000 (02:33 +0000)]
Link in all icc libraries dynamically; this works properly with
RedHat 9.0, makes smaller executables, and because icc links in
some libraries anyway, doesn't actually diminish portability.

18 years agoinstead of creating LBMigrateMsg (migration decision) message each strategy itself...
Gengbin Zheng [Fri, 29 Aug 2003 22:40:04 +0000 (22:40 +0000)]
instead of creating LBMigrateMsg (migration decision) message each strategy itself, let CentralLB do it since they essentially are doing same thing creating migrateInfo etc. Another advantage is that when LBMigrateMsg data structure changes I don;t have to go through each strategy to make changes any more. The migrate info is now encoded in LDStats itself.

18 years agoinstead of creating LBMigrateMsg (migration decision) message each strategy itself...
Gengbin Zheng [Fri, 29 Aug 2003 22:40:04 +0000 (22:40 +0000)]
instead of creating LBMigrateMsg (migration decision) message each strategy itself, let CentralLB do it since they essentially are doing same thing creating migrateInfo etc. Another advantage is that when LBMigrateMsg data structure changes I don;t have to go through each strategy to make changes any more. The migrate info is not encoded in LDStats itself.

18 years agofor -tracemode option, check for duplcaites and ignore the duplciated options.
Gengbin Zheng [Fri, 29 Aug 2003 16:05:59 +0000 (16:05 +0000)]
for -tracemode option, check for duplcaites and ignore the duplciated options.

18 years agoChange prototype for "main" routine to *exactly* match
Orion Lawlor [Fri, 29 Aug 2003 16:05:48 +0000 (16:05 +0000)]
Change prototype for "main" routine to *exactly* match
what AMPI is expecting.  Leaving off the return type
compiles, but is then silently ignored by ampi, which
calls all the default mains and then immediately exits!

18 years agoForgot "inline" keyword in template function. Without "inline",
Orion Lawlor [Fri, 29 Aug 2003 15:33:15 +0000 (15:33 +0000)]
Forgot "inline" keyword in template function.  Without "inline",
certain stupid compilers like Compaq cxx can't figure out to
instantiate the template, and FEM dies with a link error.

18 years agofixed compiler error on some compilers in CkpvInitialize.
Gengbin Zheng [Fri, 29 Aug 2003 14:35:35 +0000 (14:35 +0000)]
fixed compiler error on some compilers in CkpvInitialize.

18 years agominor change to fix a compiler warning.
Gengbin Zheng [Fri, 29 Aug 2003 03:27:13 +0000 (03:27 +0000)]
minor change to fix a compiler warning.

18 years agominor change in comment.
Gengbin Zheng [Fri, 29 Aug 2003 00:40:13 +0000 (00:40 +0000)]
minor change in comment.

18 years agoenabled trace AMPI program in bluegene simualtor and potentially better in trace...
Gengbin Zheng [Thu, 28 Aug 2003 22:39:49 +0000 (22:39 +0000)]
enabled trace AMPI program in bluegene simualtor and potentially better in trace projections. To make it happen, ampi::recv() is splitted into two versions recv() and recvNoTrace(). Tracing code is only added for recv() one.

18 years agominor change, moving varaible traceBluegeneLinked definition to trace-common.C.
Gengbin Zheng [Thu, 28 Aug 2003 22:35:11 +0000 (22:35 +0000)]
minor change, moving varaible traceBluegeneLinked definition to trace-common.C.

18 years agoupdated due to macro name changes in trace bluegene.
Gengbin Zheng [Thu, 28 Aug 2003 22:33:19 +0000 (22:33 +0000)]
updated due to macro name changes in trace bluegene.

18 years agocleanup trace bluegene code. moved all bluegene specific tracing calls and macros...
Gengbin Zheng [Thu, 28 Aug 2003 22:32:03 +0000 (22:32 +0000)]
cleanup trace bluegene code. moved all bluegene specific tracing calls and macros into trace-bluegene.[hC].

18 years agowhen compile with bgcharm++, trace-bluegene is always enabled and linked.
Gengbin Zheng [Thu, 28 Aug 2003 22:28:48 +0000 (22:28 +0000)]
when compile with bgcharm++, trace-bluegene is always enabled and linked.

18 years agoadded trace-bluegene.h into charm header files.
Gengbin Zheng [Thu, 28 Aug 2003 22:28:02 +0000 (22:28 +0000)]
added trace-bluegene.h into charm header files.

18 years agoinstead of letting it check illegal nested begin/end execute events by default, only...
Gengbin Zheng [Thu, 28 Aug 2003 22:13:26 +0000 (22:13 +0000)]
instead of letting it check illegal nested begin/end execute events by default, only do so when use "+checknested". This is because all AMPI programs currently has these illegal nested events.

18 years agoadding latest version. The seapping will not work for SMP's.
Sameer Kumar [Thu, 28 Aug 2003 17:37:53 +0000 (17:37 +0000)]
adding latest version. The seapping will not work for SMP's.

18 years agoIf the mesh has no geometric information,
Orion Lawlor [Thu, 28 Aug 2003 17:35:49 +0000 (17:35 +0000)]
If the mesh has no geometric information,
calling the geometric partitioner won't work;
so we have to just copy the mesh.

This is needed by, e.g., programs that only set
global mesh data during init().

18 years agoAdding a clean implemenation of blocking receives. To enable it use
Sameer Kumar [Thu, 28 Aug 2003 15:52:06 +0000 (15:52 +0000)]
Adding a clean implemenation of blocking receives. To enable it use

+enableBlockingReceives

18 years agofixed online timing correction so that it can handle node group broadcast and broadca...
Gengbin Zheng [Thu, 28 Aug 2003 05:44:41 +0000 (05:44 +0000)]
fixed online timing correction so that it can handle node group broadcast and broadcast with exception messages.

18 years agosimple checking for nested begin/end execute events.
Gengbin Zheng [Thu, 28 Aug 2003 04:56:16 +0000 (04:56 +0000)]
simple checking for nested begin/end execute events.

18 years agoglobal offset table manager for AMPI
Sameer Kumar [Wed, 27 Aug 2003 17:24:15 +0000 (17:24 +0000)]
global offset table manager for AMPI

18 years agoswitch to old gnu malloc for mpi version also.
Gengbin Zheng [Tue, 26 Aug 2003 18:09:35 +0000 (18:09 +0000)]
switch to old gnu malloc for mpi version also.

18 years agochanged thread tracing so that it won't trace Blue Gene (converse) threads that simul...
Gengbin Zheng [Tue, 26 Aug 2003 00:08:23 +0000 (00:08 +0000)]
changed thread tracing so that it won't trace Blue Gene (converse) threads that simulating the processors, etc. User aplication threads in AMPI will be explicitly traced.

18 years agoMinor bug fix: instead of relying on the initialization
Orion Lawlor [Mon, 25 Aug 2003 23:33:26 +0000 (23:33 +0000)]
Minor bug fix: instead of relying on the initialization
order of global variables to set the initial stack size,
check for a zero stack size explicitly.

Without this fix, all tcharm threads were created with
the default (32K!) stack size...

18 years agoset back batsyncer's period time to 1s for now.
Gengbin Zheng [Mon, 25 Aug 2003 06:34:16 +0000 (06:34 +0000)]
set back batsyncer's period time to 1s for now.
it needs a better fix in both usesAtSync and not usesAtSync cases.

18 years agoadd pup for bgMsgEntry to clean up guna's code a little bit.
Gengbin Zheng [Fri, 22 Aug 2003 22:03:57 +0000 (22:03 +0000)]
add pup for bgMsgEntry to clean up guna's code a little bit.

18 years agoadded set_avail_vector() back, which was lost because it became a class function.
Gengbin Zheng [Thu, 21 Aug 2003 21:55:40 +0000 (21:55 +0000)]
added set_avail_vector() back, which was lost because it became a class function.

18 years agoRemoved an obsolete (and scary) global variable called "flag".
Chao Huang [Thu, 21 Aug 2003 20:44:02 +0000 (20:44 +0000)]
Removed an obsolete (and scary) global variable called "flag".

18 years agofixed for uth, added missing functions when CMK_LBDB is false.
Gengbin Zheng [Thu, 21 Aug 2003 15:44:08 +0000 (15:44 +0000)]
fixed for uth, added missing functions when CMK_LBDB is false.

18 years agominor change, suppress warnings.
Gengbin Zheng [Thu, 21 Aug 2003 15:43:29 +0000 (15:43 +0000)]
minor change, suppress warnings.

18 years agominor changes, added a destructor for BaseLB.
Gengbin Zheng [Thu, 21 Aug 2003 06:02:32 +0000 (06:02 +0000)]
minor changes, added a destructor for BaseLB.

18 years agofixed broken uth version.
Gengbin Zheng [Wed, 20 Aug 2003 14:46:52 +0000 (14:46 +0000)]
fixed broken uth version.

18 years agoupdated
Gengbin Zheng [Wed, 20 Aug 2003 07:01:47 +0000 (07:01 +0000)]
updated

18 years agofixed a bug for NAMD in the case that application writes its own load balancer, load...
Gengbin Zheng [Wed, 20 Aug 2003 06:20:38 +0000 (06:20 +0000)]
fixed a bug for NAMD in the case that application writes its own load balancer, load balance should always be turned on in this case.

18 years agoadded new features and bug fixes:
Gengbin Zheng [Wed, 20 Aug 2003 05:23:02 +0000 (05:23 +0000)]
added new features and bug fixes:

1. allow users to specify multiple load balancers at compile time or runtime
by using multiple "-balancer" at compile time and "+balancer" at runtime.
For example, if foo1LB, foo2LB amd foo3LB are specified, the first
load balancing phase will invoke foo1LB, second phase uses foo2LB,
all afterwards load balancing phases will use foo3LB.

2. bug fixes and improvements in NullLB: when no real load balancer exists,
NullLB now doesnot bother clearing the LB database, also it disables the
builtin batsyncer which original calls atSync in every second.
These reduce unnecessary overhead of NullLB.

3. batsyncer itself is not strictly needed in the AtSync mode load balancing
(useful only when there is no object on one processor to call AtSync),
however, it is very important in automatic (non atSync) mode to set the
load balancing interval time. For AtSync mode, batsyncer should call
AtSync immediately to avoid any delay; for non AtSync mode however, the
period time is the interval between load balancing phases, it still can be
set to 0 second but better to be some larger value. Thus, for the default
batsync period time for both cases, 0 second is a better choice which means
the batsyncer barrier is always ready. In the case of non atSync
mode, users should use +LBPeriod to set the load balancing interval anyway.

18 years agoupdated.
Gengbin Zheng [Wed, 20 Aug 2003 05:00:59 +0000 (05:00 +0000)]
updated.

18 years agoincrease the charm version; change LB calls in moduleinit.C.
Gengbin Zheng [Wed, 20 Aug 2003 04:59:27 +0000 (04:59 +0000)]
increase the charm version; change LB calls in moduleinit.C.

18 years agominor changes, assigning variable theLbdb in the super class.
Gengbin Zheng [Tue, 19 Aug 2003 05:45:10 +0000 (05:45 +0000)]
minor changes, assigning variable theLbdb in the super class.

18 years agoTwo fixes:
Orion Lawlor [Tue, 19 Aug 2003 01:07:42 +0000 (01:07 +0000)]
Two fixes:
  1.) Remove wrong "if (CmiMyRank()==0)" nodegroup
      constructor (anti) idiom.

  2.) Handle nodeMgr startup race conditions--
     a.) we're running before our nodegroup's been created
     b.) we're running *during* our nodegroup's constructor
  Sadly, both these happen-- Graeme reported (a),
  and I've seen (b) on megatest +p4, on a cool with
  ++cpus 2 in my nodelist.

18 years agoMinor tweak on Nodegroup creation: zero out memory of
Orion Lawlor [Tue, 19 Aug 2003 01:04:42 +0000 (01:04 +0000)]
Minor tweak on Nodegroup creation: zero out memory of
the nodegroup, which makes it possible to use a nodegroup
variable as an initialization flag.

The right fix, which could break existing code, is to
keep a newly-created nodegroup *out* of the nodegrouptable
until its constructor is *finished*.

18 years agoTwo bug fixes:
Orion Lawlor [Tue, 19 Aug 2003 01:03:32 +0000 (01:03 +0000)]
Two bug fixes:
  1.) Remove wrong "if (CmiMyRank()==0)" test.
      Nodegroup constructors only happen on *one*
      processor.
  2.) Add hack "ctorDoneFlag" so ckreduction can
      be sure I've finished initialization.  We need
      to fix the nodegroup reduction semantics to
      make this unnecessary.  (But for now, here it is...)

18 years agoMinor tweaks to enable FEM framework to be built without
Orion Lawlor [Mon, 18 Aug 2003 23:22:50 +0000 (23:22 +0000)]
Minor tweaks to enable FEM framework to be built without
the rest of Charm++, on the machines' native MPI.

See fem/make_fem_alone.sh for details on how this is done.

18 years agoAdded this (bizarre) source file to allow certain
Orion Lawlor [Mon, 18 Aug 2003 23:21:49 +0000 (23:21 +0000)]
Added this (bizarre) source file to allow certain
AMPI-based libraries like FEM to be built without
the rest of the Charm++ runtime on the machine's native
MPI.

See
charm/src/libs/ck-libs/fem/make_fem_alone.sh
for an example of how this file might be used.

18 years agobatsyncer atsync period time set to 1s. inadvertently set it to 0 when I checked...
Gengbin Zheng [Mon, 18 Aug 2003 22:44:32 +0000 (22:44 +0000)]
batsyncer atsync period time set to 1s. inadvertently set it to 0 when I checked in, which caused lots of problems.

18 years agocc64 works for xlC as well.
Orion Lawlor [Mon, 18 Aug 2003 21:04:18 +0000 (21:04 +0000)]
cc64 works for xlC as well.

18 years agoAdd (commented-out) true volume transfer.
Orion Lawlor [Mon, 18 Aug 2003 19:16:30 +0000 (19:16 +0000)]
Add (commented-out) true volume transfer.

18 years ago1. added command line options +LBPeriod to set lb's built in load balancing barrier...
Gengbin Zheng [Mon, 18 Aug 2003 17:34:08 +0000 (17:34 +0000)]
1. added command line options +LBPeriod to set lb's built in load balancing barrier period.
2. all centralized load balancers now can be dynamically enabled or disabled on the fly. that is one can switch to a different balancers at runtime.

18 years agojust added a warning about turning MACHINE_DEBUG
Sayantan Chakravorty [Fri, 15 Aug 2003 21:44:42 +0000 (21:44 +0000)]
just added a warning about turning MACHINE_DEBUG

18 years agoCorrect MPI_WTICK by adding the return type.
Chao Huang [Fri, 15 Aug 2003 20:13:14 +0000 (20:13 +0000)]
Correct MPI_WTICK by adding the return type.

18 years agoAdd useful prototype for MPI_Wtime. Functions returning values
Orion Lawlor [Fri, 15 Aug 2003 15:38:34 +0000 (15:38 +0000)]
Add useful prototype for MPI_Wtime.  Functions returning values
need to be declared as such-- the ampimod.f90
external MPI_Wtime
is not equivalent to
double, external MPI_Wtime
because it doesn't declare the return type; *and* if you just
include the header, you don't get the prototype.

18 years agoNeeded function prototypes for CMI_VMI_CmiAlloc() and CMI_VMI_CmiFree()
Greg Koenig [Fri, 15 Aug 2003 03:34:20 +0000 (03:34 +0000)]
Needed function prototypes for CMI_VMI_CmiAlloc() and CMI_VMI_CmiFree()
for the CONVERSE_VERSION_VMI case.  Without these, the compiler assumes
int as the return type for CMI_VMI_CmiAlloc() when in fact it is a
(void *).  On IA-64, these are not the same size.

This bug may be present for the Elan version but not visible because
int and pointer sizes are the same on that architecture (?).

18 years agoAdded hook into CmiAlloc() and CmiFree() for CONVERSE_VERSION_VMI to support
Greg Koenig [Fri, 15 Aug 2003 00:38:35 +0000 (00:38 +0000)]
Added hook into CmiAlloc() and CmiFree() for CONVERSE_VERSION_VMI to support
memory pool.

18 years agoFixed RDMA bug where RDMA context was expected to be 64-bits.
Greg Koenig [Fri, 15 Aug 2003 00:37:35 +0000 (00:37 +0000)]
Fixed RDMA bug where RDMA context was expected to be 64-bits.
Added memory pool hooks into CmiAlloc().

18 years agoMore cleanup for IA-32 and IA-64 building.
Greg Koenig [Thu, 14 Aug 2003 06:58:17 +0000 (06:58 +0000)]
More cleanup for IA-32 and IA-64 building.

18 years agomove processor bit vector to LBDatabaase from CentralLB.
Gengbin Zheng [Thu, 14 Aug 2003 06:32:29 +0000 (06:32 +0000)]
move processor bit vector to LBDatabaase from CentralLB.

18 years agoModifications for IA-64 support.
Greg Koenig [Thu, 14 Aug 2003 04:02:51 +0000 (04:02 +0000)]
Modifications for IA-64 support.

18 years agoadded clustermatic in help page.
Gengbin Zheng [Thu, 14 Aug 2003 03:22:13 +0000 (03:22 +0000)]
added clustermatic in help page.
not show vmi in arch list.

18 years agoBug fix in termination.
Greg Koenig [Thu, 14 Aug 2003 01:16:26 +0000 (01:16 +0000)]
Bug fix in termination.

18 years agofixed a typo and path to invoke the scyld one.
Gengbin Zheng [Wed, 13 Aug 2003 22:57:18 +0000 (22:57 +0000)]
fixed a typo and path to invoke the scyld one.

18 years agochanged CMK_SCYLD to CMK_BPROC
Gengbin Zheng [Wed, 13 Aug 2003 22:30:25 +0000 (22:30 +0000)]
changed CMK_SCYLD to CMK_BPROC

18 years agoadded clustermatic conv-machs. It is just scyld but with a different name for clarifi...
Gengbin Zheng [Wed, 13 Aug 2003 22:29:44 +0000 (22:29 +0000)]
added clustermatic conv-machs. It is just scyld but with a different name for clarification.
also changed CMK_SCYLD to CMK_BPROC since bproc is not unique to Scyld, and it is used by both Beowulf.

18 years agoAdded VMI machine layer (koenig's thesis) to repository.
Greg Koenig [Wed, 13 Aug 2003 21:13:13 +0000 (21:13 +0000)]
Added VMI machine layer (koenig's thesis) to repository.

18 years agoRemove duplicate constants, since they freak out
Orion Lawlor [Tue, 12 Aug 2003 17:22:57 +0000 (17:22 +0000)]
Remove duplicate constants, since they freak out
the f90 compiler.

18 years agoMinor reorganization of partitioning data-- store all data in
Orion Lawlor [Tue, 12 Aug 2003 17:05:13 +0000 (17:05 +0000)]
Minor reorganization of partitioning data-- store all data in
a new "FEM_Partition" object, rather than having a separate
global variable for the ghost layers, symmetries, and elem->chunk
mapping.

18 years agoTwo small additions:
Orion Lawlor [Tue, 12 Aug 2003 17:04:03 +0000 (17:04 +0000)]
Two small additions:
  - Pass a default MPI communicator to IDXL_Init.
    This lets us use IDXL from split communicators more easily.

  - New interface to grab the indices out of an IDXL list.

18 years agoallow compilation for SMP version of Charm. to compile a net-linux-smp version, chang...
Gengbin Zheng [Tue, 12 Aug 2003 06:24:24 +0000 (06:24 +0000)]
allow compilation for SMP version of Charm. to compile a net-linux-smp version, change buildsmp to 1.

18 years agochanged to build AMPI binary.
Gengbin Zheng [Tue, 12 Aug 2003 04:48:49 +0000 (04:48 +0000)]
changed to build AMPI binary.

18 years agofirst prelimiary (working however) version of rpmbuild spec file.
Gengbin Zheng [Tue, 12 Aug 2003 04:11:33 +0000 (04:11 +0000)]
first prelimiary (working however) version of rpmbuild spec file.

18 years agoadded a simple target "install" which can be used in rpmbuild.
Gengbin Zheng [Tue, 12 Aug 2003 04:00:17 +0000 (04:00 +0000)]
added a simple target "install" which can be used in rpmbuild.

18 years agoadded -h, --help option to display a much detailed help page.
Gengbin Zheng [Sun, 10 Aug 2003 18:09:08 +0000 (18:09 +0000)]
added -h, --help option to display a much detailed help page.

18 years agowhen num of pe is 1, skip load balancing.
Gengbin Zheng [Sat, 9 Aug 2003 05:24:00 +0000 (05:24 +0000)]
when num of pe is 1, skip load balancing.
GreedyRefLB - refine after greedy now use the better api for interfacing different strategies.

18 years agodefine HAVE_USR_INCLUDE_MALLOC_H to 1 magically make new gnu malloc works on Opteron.
Gengbin Zheng [Sat, 9 Aug 2003 01:21:19 +0000 (01:21 +0000)]
define HAVE_USR_INCLUDE_MALLOC_H to 1 magically make new gnu malloc works on Opteron.

18 years agoAdded manual for checkpoint/restart.
Chao Huang [Fri, 8 Aug 2003 22:57:42 +0000 (22:57 +0000)]
Added manual for checkpoint/restart.

18 years agoAdded Fortran support for latest functions.
Chao Huang [Fri, 8 Aug 2003 22:56:33 +0000 (22:56 +0000)]
Added Fortran support for latest functions.

18 years agochanged back to old gnu malloc as default.
Gengbin Zheng [Fri, 8 Aug 2003 21:03:02 +0000 (21:03 +0000)]
changed back to old gnu malloc as default.

18 years agoremoved the redundent test rule. Charm-5_8b1
Gengbin Zheng [Fri, 8 Aug 2003 20:54:07 +0000 (20:54 +0000)]
removed the redundent test rule.

18 years agodisable new gnu and use gnuold by default.
Gengbin Zheng [Fri, 8 Aug 2003 19:12:57 +0000 (19:12 +0000)]
disable new gnu and use gnuold by default.
add config for smp version.

18 years agofunction that dump all bglogs to disk according to physical processors.
Gengbin Zheng [Fri, 8 Aug 2003 05:41:19 +0000 (05:41 +0000)]
function that dump all bglogs to disk according to physical processors.

18 years agofixed for aCC on HP. check for _STRUCT_MALLINFO macro.
Gengbin Zheng [Fri, 8 Aug 2003 01:20:33 +0000 (01:20 +0000)]
fixed for aCC on HP. check for _STRUCT_MALLINFO macro.

18 years agofixed the algorithm of calculating the obj key when the obj id is -1.
Gengbin Zheng [Thu, 7 Aug 2003 21:20:08 +0000 (21:20 +0000)]
fixed the algorithm of calculating the obj key when the obj id is -1.

18 years agofixed alignment in MigrateMsg.
Gengbin Zheng [Thu, 7 Aug 2003 18:32:13 +0000 (18:32 +0000)]
fixed alignment in MigrateMsg.

18 years agoAdded Physical Processor info in MPI_Get_proc_name.
Chao Huang [Thu, 7 Aug 2003 17:29:14 +0000 (17:29 +0000)]
Added Physical Processor info in MPI_Get_proc_name.
Now it returns a string "AMPI_VP[vp#]_PE[pp#]

18 years agoremoved c++ comments.
Gengbin Zheng [Thu, 7 Aug 2003 16:47:45 +0000 (16:47 +0000)]
removed c++ comments.

18 years agochanged c++ style comment
Gengbin Zheng [Thu, 7 Aug 2003 16:43:52 +0000 (16:43 +0000)]
changed c++ style comment

18 years agochanged the way objects are registed in LBDBManager. This way, when an object migrate...
Gengbin Zheng [Thu, 7 Aug 2003 16:07:08 +0000 (16:07 +0000)]
changed the way objects are registed in LBDBManager. This way, when an object migrates and unregisters itself, the memory of LBObj (which is big) can be free up. Also fixed a bug in generating migration message when in "work" scheme for RandCentLB, moved createMigrateMessage() up to CentralLB from GreedyLB.C because it can be commonly used by strategies in "work" scheme.

18 years agoChanged code so that if __FAULT__ is defined then if a processor
Sayantan Chakravorty [Thu, 7 Aug 2003 07:32:24 +0000 (07:32 +0000)]
Changed code so that if __FAULT__ is defined then if a processor
dies charmrun doesn't kill the remaining processors.

18 years agoChanged code so that if __FAULT__ is defined then if a processor
Sayantan Chakravorty [Thu, 7 Aug 2003 07:31:24 +0000 (07:31 +0000)]
Changed code so that if __FAULT__ is defined then if a processor
receives a SIGKILL SIGSTOP etc it doesn't kill the other processors
in that program

18 years agodefine CMK_MALLOC_USE_GNUOLD_MALLOC to allow conv-mach.h to set gnuold as default...
Gengbin Zheng [Thu, 7 Aug 2003 06:17:20 +0000 (06:17 +0000)]
define CMK_MALLOC_USE_GNUOLD_MALLOC to allow conv-mach.h to set gnuold as default malloc.

18 years agoswitch to gnu old as default malloc for now.
Gengbin Zheng [Thu, 7 Aug 2003 06:13:35 +0000 (06:13 +0000)]
switch to gnu old as default malloc for now.

18 years agomove fucntions around to compile with POSE.
Gengbin Zheng [Wed, 6 Aug 2003 16:41:59 +0000 (16:41 +0000)]
move fucntions around to compile with POSE.

18 years agoRemove unportable declaration of "sbrk", since it's incompatible
Orion Lawlor [Wed, 6 Aug 2003 16:23:53 +0000 (16:23 +0000)]
Remove unportable declaration of "sbrk", since it's incompatible
with the ibmsp UNISTD.h version.

18 years agoRemoved unnecssary and unportable <features.h>
Orion Lawlor [Wed, 6 Aug 2003 15:34:31 +0000 (15:34 +0000)]
Removed unnecssary and unportable <features.h>
declaration.

18 years agoAdded puper for KeyvalNode to remove the compile error.
Chao Huang [Wed, 6 Aug 2003 14:54:53 +0000 (14:54 +0000)]
Added puper for KeyvalNode to remove the compile error.
But really KeyvalNode is not meant to be pup'ed.

18 years agodecreased PACKET_MAX, and increased socket buffer.
Gengbin Zheng [Wed, 6 Aug 2003 06:57:14 +0000 (06:57 +0000)]
decreased PACKET_MAX, and increased socket buffer.

18 years agoset send/recv buffer size to be at least the maximum packet size.
Gengbin Zheng [Wed, 6 Aug 2003 06:09:26 +0000 (06:09 +0000)]
set send/recv buffer size to be at least the maximum packet size.

18 years agoadded function skt_setSockBuf() - set send and recv buffer of a socket.
Gengbin Zheng [Wed, 6 Aug 2003 06:07:57 +0000 (06:07 +0000)]
added function skt_setSockBuf() - set send and recv buffer of a socket.

18 years agoAdded Caching support: MPI_Keyval_X and MPI_Attr_X.
Chao Huang [Wed, 6 Aug 2003 01:34:12 +0000 (01:34 +0000)]
Added Caching support: MPI_Keyval_X and MPI_Attr_X.