charm.git
21 years ago*** empty log message ***
Terry L. Wilmarth [Wed, 13 Sep 2000 19:31:09 +0000 (19:31 +0000)]
*** empty log message ***

21 years agoFixed pup_er. Most C compilers do not allow empty structs.
Milind Bhandarkar [Wed, 13 Sep 2000 15:42:07 +0000 (15:42 +0000)]
Fixed pup_er. Most C compilers do not allow empty structs.

21 years agoTiny change-- lineno is now "unsigned int", which caused a
Orion Lawlor [Mon, 11 Sep 2000 21:54:35 +0000 (21:54 +0000)]
Tiny change-- lineno is now "unsigned int", which caused a
link error on NT.

21 years agoSlightly reorganized header structure. Now no headers should
Orion Lawlor [Mon, 11 Sep 2000 21:36:55 +0000 (21:36 +0000)]
Slightly reorganized header structure.  Now no headers should
need to be listed twice (once in ALLHEADERS, again in CKHEADERS).
Now headers are soft-linked instead of copied.  This makes
development much easier.
Added support for the new Common/util directory.

21 years agoMoved these files to ../util.
Orion Lawlor [Mon, 11 Sep 2000 21:34:26 +0000 (21:34 +0000)]
Moved these files to ../util.

21 years agoMoved these files from conv-core and ck-core.
Orion Lawlor [Mon, 11 Sep 2000 21:33:31 +0000 (21:33 +0000)]
Moved these files from conv-core and ck-core.

sockRoutines is new-- it provides a common interface to the
skt_* berkely sockets wrappers that are used in conv-ccs.c,
net/machine.c, net/conv-host.c, and ccs.c.

pup_c is also new-- it's a C-langugage binding for the
PUP::er routines.

21 years agoModified to suit the new interface for registering and retrieving user's data.
Milind Bhandarkar [Sun, 10 Sep 2000 08:54:32 +0000 (08:54 +0000)]
Modified to suit the new interface for registering and retrieving user's data.

21 years agoAllowed registering more than one datatypes.
Milind Bhandarkar [Sun, 10 Sep 2000 08:52:28 +0000 (08:52 +0000)]
Allowed registering more than one datatypes.

21 years agoMade soft links use relative pathnames instead of absolute.
Orion Lawlor [Fri, 8 Sep 2000 22:03:36 +0000 (22:03 +0000)]
Made soft links use relative pathnames instead of absolute.
This lets you move a charm++ installation without having
to recompile anything.

It also stops the confusing oddness that happens when you
move, e.g., charm to old_charm, and install a new charm.
Then the old_charm (absolute) soft links point into the
new charm directory!

21 years agoCombined the getBuf and advance methods of PUP::er.
Milind Bhandarkar [Fri, 8 Sep 2000 19:48:00 +0000 (19:48 +0000)]
Combined the getBuf and advance methods of PUP::er.

21 years agoAdded a pupOpaqueObject function. This function was born out of the observation
Milind Bhandarkar [Fri, 8 Sep 2000 08:09:14 +0000 (08:09 +0000)]
Added a pupOpaqueObject function. This function was born out of the observation
that many Converse-level entities have similar things to do for pack/unpack.
As it turned out, in FEM Framework, the user-supplied data is also an opaque
object, and so could be packed using the same function.

21 years agoClarified a few points, after answering Neelam's reasonable
Orion Lawlor [Thu, 7 Sep 2000 18:37:34 +0000 (18:37 +0000)]
Clarified a few points, after answering Neelam's reasonable
questions about Array syntax and semantics.

21 years agoFixed CMK_AR.
Milind Bhandarkar [Tue, 5 Sep 2000 15:27:47 +0000 (15:27 +0000)]
Fixed CMK_AR.

21 years agoRemoved the multiple declaration errors.
Milind Bhandarkar [Tue, 5 Sep 2000 13:46:29 +0000 (13:46 +0000)]
Removed the multiple declaration errors.

21 years agoHad broken cmk_lbdb_off option when making the_lbdb public. Fixed that.
Milind Bhandarkar [Tue, 5 Sep 2000 11:53:19 +0000 (11:53 +0000)]
Had broken cmk_lbdb_off option when making the_lbdb public. Fixed that.

21 years agoFixed the threaded-entry-migrate-freemsg bug.
Milind Bhandarkar [Tue, 5 Sep 2000 07:22:51 +0000 (07:22 +0000)]
Fixed the threaded-entry-migrate-freemsg bug.

21 years agoA very fast O(n) algorithm, uses more memory, but the tradeoff was
Milind Bhandarkar [Tue, 5 Sep 2000 07:22:18 +0000 (07:22 +0000)]
A very fast O(n) algorithm, uses more memory, but the tradeoff was
worth it. Coded by Karthik Mahesh, minor optimizations by Milind.

21 years agoAdded the class static readonly variables to ci file syntax.
Milind Bhandarkar [Mon, 4 Sep 2000 17:57:32 +0000 (17:57 +0000)]
Added the class static readonly variables to ci file syntax.

21 years agoMinor fix-fixes:
Orion Lawlor [Mon, 4 Sep 2000 01:33:55 +0000 (01:33 +0000)]
Minor fix-fixes:
1.) Change to LBObjid hash function would fail for >4-int object
indices.  Replaced with proper function, which also preserves
the 1-int case.

2.) Array element sends must go via the message queue to prevent
stack build-up for deep single-processor call chains. These might
happen, e.g., in a driver element calling itself for the main
time loop.  Messages are now properly noted as sent, then wait
through the queue for delivery.  This entailed minor reorganization
of the message delivery subsystem.

21 years agoFixed a bug in code generated for threaded (void) entry methods of array
Milind Bhandarkar [Sat, 2 Sep 2000 07:04:46 +0000 (07:04 +0000)]
Fixed a bug in code generated for threaded (void) entry methods of array
elements. The dummy message that is passed to that method in a thread
has to be deleted before calling the object method, because upon object
method's return, the thread might have migrated.

21 years agoPaving way for splitting conv-host code into multiple files. Instead of just
Milind Bhandarkar [Fri, 1 Sep 2000 18:02:25 +0000 (18:02 +0000)]
Paving way for splitting conv-host code into multiple files. Instead of just
a single file conv-host.c, now the conv-host belongs to its separate
directory under Common.* directory. The Makefile checks for existence
of this directory
rather than the file, and proceeds to make conv-host in its own directory.

21 years agoAdded code to check directories before issuing the mkdir command.
Milind Bhandarkar [Fri, 1 Sep 2000 17:58:08 +0000 (17:58 +0000)]
Added code to check directories before issuing the mkdir command.

21 years agoAdded FEM_Timer function. This is mainly for use with fortran programs.
Milind Bhandarkar [Fri, 1 Sep 2000 06:16:15 +0000 (06:16 +0000)]
Added FEM_Timer function. This is mainly for use with fortran programs.

21 years agoAdded Makefile.
Milind Bhandarkar [Fri, 1 Sep 2000 05:19:56 +0000 (05:19 +0000)]
Added Makefile.

21 years agoFixed an initialization bug in fem_reduce_field.
Milind Bhandarkar [Fri, 1 Sep 2000 04:57:07 +0000 (04:57 +0000)]
Fixed an initialization bug in fem_reduce_field.

21 years agoCompleted integration of documentation into charm cvs tree.
Milind Bhandarkar [Thu, 31 Aug 2000 22:02:37 +0000 (22:02 +0000)]
Completed integration of documentation into charm cvs tree.
Added appropriate targets for super-install for making different types
(ps, pdf, html) of documentation.

21 years agochmod html-manual properly
Terry L. Wilmarth [Thu, 31 Aug 2000 18:22:16 +0000 (18:22 +0000)]
chmod html-manual properly

21 years agoAdded the thread-migrate-exit-freemsg bug.
Milind Bhandarkar [Thu, 31 Aug 2000 17:40:37 +0000 (17:40 +0000)]
Added the thread-migrate-exit-freemsg bug.

21 years agoAdded timing data to crack2D.
Milind Bhandarkar [Thu, 31 Aug 2000 16:43:30 +0000 (16:43 +0000)]
Added timing data to crack2D.

21 years agoMade crack2D threads to deposit LB Data on behalf of the chunks.
Milind Bhandarkar [Tue, 29 Aug 2000 01:29:17 +0000 (01:29 +0000)]
Made crack2D threads to deposit LB Data on behalf of the chunks.
Made a few members of the ArrayElement, and CkArray public in order
to do that. Also found a source of inefficiency (and also a bug) where
CkArray::Send was used as an entry method, and not an ordinary method
of CkArray for sending messages ot other objects from array elements.
Also, the id[0] of LDHandle did not contain the array element index in 1D
case as was the case earlier. MetisLB was depending on that to collect
communication data for the graph partitioner, and was segfaulting.

21 years agoFixed numerous bugs in the chunk pupper. Now crack2D runs with
Milind Bhandarkar [Sun, 27 Aug 2000 13:18:20 +0000 (13:18 +0000)]
Fixed numerous bugs in the chunk pupper. Now crack2D runs with
isomalloc threads with migration.

21 years agoEliminated migrator group from fem framework. Replaced it with a method
Milind Bhandarkar [Sun, 27 Aug 2000 10:43:57 +0000 (10:43 +0000)]
Eliminated migrator group from fem framework. Replaced it with a method
in chunk object itself. Fixed a small bug in pupper that caused it not
to advance in the isSizing case.
Also, fixed a small typo in mesh2graph.C.

21 years agoSeparated mesh to graph conversion capability into a separate program.
Milind Bhandarkar [Sun, 27 Aug 2000 08:41:11 +0000 (08:41 +0000)]
Separated mesh to graph conversion capability into a separate program.
This way, the generated graph can be partitioned repeatedly.

21 years agoOptimized mesh2graph conversion.
Milind Bhandarkar [Sun, 27 Aug 2000 08:03:44 +0000 (08:03 +0000)]
Optimized mesh2graph conversion.

21 years agoRenamed TRUECRASH to CMK_TRUECRASH. And made it possible to set it while
Milind Bhandarkar [Sat, 26 Aug 2000 23:21:36 +0000 (23:21 +0000)]
Renamed TRUECRASH to CMK_TRUECRASH. And made it possible to set it while
super-installing, without editing the source.

21 years agoAdded a macro CtvAccessOther that lets you get/set a Ctv variable of any
Milind Bhandarkar [Sat, 26 Aug 2000 23:12:41 +0000 (23:12 +0000)]
Added a macro CtvAccessOther that lets you get/set a Ctv variable of any
thread. It should be invoked as CtvAccessOther(thread,varname);
Added CthGetData function to each of the threads implementation.
This function is used in the CtvAccessOther macro.

21 years agoAdded a migrator group to aid in migration, and coded the FEM_Migrate function.
Milind Bhandarkar [Sat, 26 Aug 2000 23:06:44 +0000 (23:06 +0000)]
Added a migrator group to aid in migration, and coded the FEM_Migrate function.
This is really a placeholder right now to test the migration capabilities of
isomalloc threads.

21 years agoAdded packsize, pack and unpack functions for dealing with GlobalData.
Milind Bhandarkar [Sat, 26 Aug 2000 22:28:46 +0000 (22:28 +0000)]
Added packsize, pack and unpack functions for dealing with GlobalData.

21 years agoAdded dummy packsize, pack and unpack methods, plus code to register them
Milind Bhandarkar [Sat, 26 Aug 2000 21:40:07 +0000 (21:40 +0000)]
Added dummy packsize, pack and unpack methods, plus code to register them
with the framework alongwith GlobalData.

21 years agoAdded a pupper for fem chunk object. Also added several utility functions to
Milind Bhandarkar [Sat, 26 Aug 2000 21:39:22 +0000 (21:39 +0000)]
Added a pupper for fem chunk object. Also added several utility functions to
aid in migration. Added functions for registering user's object associated with
a chunk, and functions to packsize, pack and unpack them. A placeholder for
migrate is added. Outstanding problem: Ctv variable of the migrated thread
needs to be set somehow to point to the migrated chunk.

21 years agoAdded pupper for tempoarray.
Milind Bhandarkar [Sat, 26 Aug 2000 08:31:40 +0000 (08:31 +0000)]
Added pupper for tempoarray.

21 years agoReduced the memory requirements by storing free slots as block
Milind Bhandarkar [Sat, 26 Aug 2000 08:28:04 +0000 (08:28 +0000)]
Reduced the memory requirements by storing free slots as block
of consecutive slots. Also, allowed threads to have stacks with
user-specified sizes (internally adjusted to multiples of default stacksize.)
Essentially, the specified stacksize is the slotsize.

21 years agoAdded two virtual methods getBuf and advance to PUP::er.
Milind Bhandarkar [Sat, 26 Aug 2000 07:28:06 +0000 (07:28 +0000)]
Added two virtual methods getBuf and advance to PUP::er.
Provided implementation for getBuf and advance in toMem and fromMem
classes. And advance has been implemented in sizer class.
This was done to avoid yucky (shivani's favorite word) code in
the puppers for array elements that have threads to be packed.
Accordingly rewrote the pupper in the tempo/migrate test.

21 years agoChanged the page alignment to reflect the mmap alignment.
Milind Bhandarkar [Sat, 26 Aug 2000 06:53:59 +0000 (06:53 +0000)]
Changed the page alignment to reflect the mmap alignment.
The mmap man page specifically states that it is not the same as
page size.

21 years agoAdded packing and unpacking to isomalloc threads. tested on net-sol-cc
Milind Bhandarkar [Fri, 25 Aug 2000 22:50:18 +0000 (22:50 +0000)]
Added packing and unpacking to isomalloc threads. tested on net-sol-cc
with tempo migrate test.

21 years agoAdded code to check for exhaustion of slots on a processor. Now, the
Milind Bhandarkar [Fri, 25 Aug 2000 20:08:15 +0000 (20:08 +0000)]
Added code to check for exhaustion of slots on a processor. Now, the
thread creation is split into two phases if the number of stack slots
are exhausted. The thread is not awakened unless it has a valid stack
slot. The free slot is requested from the next processor in a ring-like
manner.

21 years agoInitial implementation of isomalloc-based threads.
Milind Bhandarkar [Fri, 25 Aug 2000 19:16:49 +0000 (19:16 +0000)]
Initial implementation of isomalloc-based threads.
This implementation uses a fixed stack size for all threads
(can be set at runtime.)
Also, it does not check for cases such as exhaustion of stack slots
on a processor. Also, it does not have necessary pack-unpack routines.
megacon, and megatest work on net-sol-cc.

21 years agoFixed some more bugs. Now crack2D runs on multiple processors.
Milind Bhandarkar [Thu, 24 Aug 2000 22:26:18 +0000 (22:26 +0000)]
Fixed some more bugs. Now crack2D runs on multiple processors.

21 years agoFixed most of the bugs that caused sigsegv on multiple chunks.
Milind Bhandarkar [Thu, 24 Aug 2000 21:53:52 +0000 (21:53 +0000)]
Fixed most of the bugs that caused sigsegv on multiple chunks.
now the bugs are limited to fp exceptions.

21 years agoFixed the standalone programs for 1 partition special case that cannot be
Milind Bhandarkar [Thu, 24 Aug 2000 20:09:50 +0000 (20:09 +0000)]
Fixed the standalone programs for 1 partition special case that cannot be
handled by metis.

21 years agoAdded the 2D crack propagation program. Works on 1 partition, to be tested with
Milind Bhandarkar [Thu, 24 Aug 2000 20:05:33 +0000 (20:05 +0000)]
Added the 2D crack propagation program. Works on 1 partition, to be tested with
multiple partitions.

21 years agoHandled numPartitions=1 case for reduce without suspending.
Milind Bhandarkar [Thu, 17 Aug 2000 17:29:29 +0000 (17:29 +0000)]
Handled numPartitions=1 case for reduce without suspending.

21 years agoAdded capabilities to handle esoteric meshes to standalone offline programs.
Milind Bhandarkar [Mon, 14 Aug 2000 22:16:40 +0000 (22:16 +0000)]
Added capabilities to handle esoteric meshes to standalone offline programs.
Makefile now produces gmap and fgmap programs, which are used
for this purpose.
They convert the mesh to a graph before partitioning it using Metis.

21 years agoFixed +vp option parsing from command line. Now Both +vp4 and +vp 4
Milind Bhandarkar [Fri, 11 Aug 2000 18:55:50 +0000 (18:55 +0000)]
Fixed +vp option parsing from command line. Now Both +vp4 and +vp 4
are accepted.

21 years agoCthInit now takes in argv as the only parameter, and sets up the default
Milind Bhandarkar [Fri, 11 Aug 2000 18:42:40 +0000 (18:42 +0000)]
CthInit now takes in argv as the only parameter, and sets up the default
stacksize for converse threads using +stacksize parameter.

21 years agoremoved a "variable declared but not used" warning
Milind Bhandarkar [Fri, 11 Aug 2000 18:34:15 +0000 (18:34 +0000)]
removed a "variable declared but not used" warning

21 years agoremoved a "variable declared but not used warning."
Milind Bhandarkar [Fri, 11 Aug 2000 18:32:59 +0000 (18:32 +0000)]
removed a "variable declared but not used warning."

21 years agoAdded f90objs.
Milind Bhandarkar [Fri, 11 Aug 2000 17:52:46 +0000 (17:52 +0000)]
Added f90objs.

21 years agoAdded f90objs to all the fortran-based languages.
Milind Bhandarkar [Fri, 11 Aug 2000 17:47:24 +0000 (17:47 +0000)]
Added f90objs to all the fortran-based languages.

21 years agoFixed array ordering problems, and numbering problems for C/F90
Milind Bhandarkar [Tue, 8 Aug 2000 20:48:42 +0000 (20:48 +0000)]
Fixed array ordering problems, and numbering problems for C/F90
interoperability.

21 years agoremoved call to offsetof, because it used to crash by rangechecking.
Milind Bhandarkar [Tue, 8 Aug 2000 20:47:49 +0000 (20:47 +0000)]
removed call to offsetof, because it used to crash by rangechecking.

21 years agoAdded a test program in fortran90.
Milind Bhandarkar [Tue, 8 Aug 2000 20:41:28 +0000 (20:41 +0000)]
Added a test program in fortran90.

21 years agoModified the C/C++ interface. init_, driver_ and finalize_ have been changed
Milind Bhandarkar [Tue, 8 Aug 2000 04:25:00 +0000 (04:25 +0000)]
Modified the C/C++ interface. init_, driver_ and finalize_ have been changed
to init, driver, and finalize. Also, all the FEM_* functions are now extern "C"
so that one can use them in C.

21 years agoModified to match new interface to C/C++.
Milind Bhandarkar [Tue, 8 Aug 2000 04:23:37 +0000 (04:23 +0000)]
Modified to match new interface to C/C++.

21 years agoAdded update test, reduce_field test, and reduce test.
Milind Bhandarkar [Mon, 7 Aug 2000 21:35:25 +0000 (21:35 +0000)]
Added update test, reduce_field test, and reduce test.

21 years agoAdded FEM_Set_Mesh_Transform, that converts the connectivity matrix for
Milind Bhandarkar [Mon, 7 Aug 2000 20:10:25 +0000 (20:10 +0000)]
Added FEM_Set_Mesh_Transform, that converts the connectivity matrix for
hexahedral and quadrilateral elements to the format that METIS requires
using a permutation vector given as argument. Also fixed a longstanding bug
where FEM_Done was NOT idempotent. Now it can be called any number of
times within a chunk, with only the first call succeeding.

21 years agochamrc is made to pass -DCMK_SEQUENTIAL=1 to all the compilers when -seq
Milind Bhandarkar [Mon, 7 Aug 2000 17:00:48 +0000 (17:00 +0000)]
chamrc is made to pass -DCMK_SEQUENTIAL=1 to all the compilers when -seq
option is given. This is so that the programs (or libraries) know which
functions (CkPrintf/printf etc.) to use.

21 years agoMade map into a self-contained program, that takes in the sequential mesh
Milind Bhandarkar [Mon, 7 Aug 2000 16:55:56 +0000 (16:55 +0000)]
Made map into a self-contained program, that takes in the sequential mesh
description, and outputs the partitioned meshes readable by the framework.

21 years agoAdded a function that prints partition information. This function proved out
Milind Bhandarkar [Sat, 5 Aug 2000 07:55:27 +0000 (07:55 +0000)]
Added a function that prints partition information. This function proved out
to be very useful in debugging. It detected several bugs in the fem_map
as well as in the readComm method of chunk. Fixed those bugs as well.
One of the bug in the fem_map function was based on the false assumption
that Metis assigns partition to nodes based on whether the node is connected
to an element, which is assigned to that partition. It turns out not to be.
So, fixed the algorithm for primary partition finder.

21 years agoRemoved the debugging prints for each partition, and instead made it into
Milind Bhandarkar [Sat, 5 Aug 2000 07:51:57 +0000 (07:51 +0000)]
Removed the debugging prints for each partition, and instead made it into
a function in the fem framework.

21 years agoAdded -lmetis to -language fem and femf.
Milind Bhandarkar [Fri, 4 Aug 2000 22:12:37 +0000 (22:12 +0000)]
Added -lmetis to -language fem and femf.

21 years agoHandled case for partitioning in 1 chunks (trivial), as Metis does
Milind Bhandarkar [Fri, 4 Aug 2000 22:12:14 +0000 (22:12 +0000)]
Handled case for partitioning in 1 chunks (trivial), as Metis does
not handle this case. Fixed the prototype of FEM_Set_Mesh. Fixed a bug in
fem_map where there were writes to unallocated arrays.

21 years agoAdded a test program for fem framework.
Milind Bhandarkar [Fri, 4 Aug 2000 22:09:01 +0000 (22:09 +0000)]
Added a test program for fem framework.

21 years agoRewrote Rui Liu's program to convert output of metis to the format
Milind Bhandarkar [Fri, 4 Aug 2000 09:28:02 +0000 (09:28 +0000)]
Rewrote Rui Liu's program to convert output of metis to the format
requested by the FEM framework. Made it available as a library function
that can be called from the init_ routine via FEM_Set_Mesh. So, now if
the init_ routine calles FEM_Set_Mesh to specify the initial sequential mesh,
the driver gets the partition information via message argument to run method,
and does not read from files. If this routine is not called from init_
then the chunk has to read this from the files. (This is right now
for backward compatibility only, and will go away. The only method will
be to specify a complete mesh via FEM_Set_Mesh.)

21 years agoAdded an option to let the user specify which "xterm" to use.
Milind Bhandarkar [Wed, 2 Aug 2000 18:04:50 +0000 (18:04 +0000)]
Added an option to let the user specify which "xterm" to use.
For example, on some systems (CDE), only dtterm is installed.
So, by putting ++xterm dtterm on the conv-host commandline, one can
use dtterm when ++in-xterm option is specified.

21 years agoFurther fixes for AMPI fortran bindings on linux.
Milind Bhandarkar [Tue, 1 Aug 2000 14:17:32 +0000 (14:17 +0000)]
Further fixes for AMPI fortran bindings on linux.

21 years agoFixed Makefile for proper operation with gmake on turing.
Milind Bhandarkar [Tue, 1 Aug 2000 13:41:17 +0000 (13:41 +0000)]
Fixed Makefile for proper operation with gmake on turing.
Also, fixed f90 binding error for AMPI_Sendrecv.

21 years agoAdded FEM_Print. And made fortran bindings compatible with linux.
Milind Bhandarkar [Tue, 1 Aug 2000 09:54:21 +0000 (09:54 +0000)]
Added FEM_Print. And made fortran bindings compatible with linux.

21 years agoAdded AMPI_Print. This is obviously non-MPI, but is very useful for debugging
Milind Bhandarkar [Tue, 1 Aug 2000 09:24:44 +0000 (09:24 +0000)]
Added AMPI_Print. This is obviously non-MPI, but is very useful for debugging
AMPI programs written using Fortran on network of workstations, because
write(*,*) does not print anything to the host. Currently AMPI_Print takes only
a single string, but it can be modified later to include a printf like
specification.
Also, added fortran bindings that use all caps. This is controlled by
CMK_FORTRAN_USES_ALLCAPS flag. Added this flag to all linux conv-mach.h files.

21 years agoAdded libfai.so to f90 link line, because it contains a function reshape.
Milind Bhandarkar [Fri, 28 Jul 2000 20:35:01 +0000 (20:35 +0000)]
Added libfai.so to f90 link line, because it contains a function reshape.

21 years agoFixed read_field function for fortran array-base change.
Milind Bhandarkar [Fri, 28 Jul 2000 20:34:17 +0000 (20:34 +0000)]
Fixed read_field function for fortran array-base change.

21 years agoFixed the nIndex=0 bug.
Milind Bhandarkar [Fri, 28 Jul 2000 20:06:26 +0000 (20:06 +0000)]
Fixed the nIndex=0 bug.

21 years agoFixed AMPI_Reduce.
Milind Bhandarkar [Fri, 28 Jul 2000 09:26:53 +0000 (09:26 +0000)]
Fixed AMPI_Reduce.

21 years agoAdded bcastraw, and fixed allreduce function.
Milind Bhandarkar [Fri, 28 Jul 2000 07:57:46 +0000 (07:57 +0000)]
Added bcastraw, and fixed allreduce function.

21 years agoChanged so that ampif.h can now be included both with cpp and fpp.
Milind Bhandarkar [Thu, 27 Jul 2000 22:07:39 +0000 (22:07 +0000)]
Changed so that ampif.h can now be included both with cpp and fpp.
Also, fixed a bug with nindex, which can be 0.

21 years agoBUGFIX: thishandle for the arrayelement was not being correctly set.
Milind Bhandarkar [Thu, 27 Jul 2000 19:01:46 +0000 (19:01 +0000)]
BUGFIX: thishandle for the arrayelement was not being correctly set.
Bug was reported by Neelam. The buggy code belonged to Orion.

21 years agoFixed the manual name.
Terry L. Wilmarth [Thu, 27 Jul 2000 18:19:18 +0000 (18:19 +0000)]
Fixed the manual name.

21 years agoampic and ampif are now dependent on libs.
Milind Bhandarkar [Thu, 27 Jul 2000 16:57:44 +0000 (16:57 +0000)]
ampic and ampif are now dependent on libs.

21 years agoMade ArgsInfo message a packed message, so that it can safely be sent
Milind Bhandarkar [Wed, 26 Jul 2000 22:04:06 +0000 (22:04 +0000)]
Made ArgsInfo message a packed message, so that it can safely be sent
across address spaces.

21 years agoFixed some stupid dependencies.
Milind Bhandarkar [Wed, 26 Jul 2000 20:14:18 +0000 (20:14 +0000)]
Fixed some stupid dependencies.

21 years agoSeparated charm++ and charmlibs.
Milind Bhandarkar [Wed, 26 Jul 2000 20:07:04 +0000 (20:07 +0000)]
Separated charm++ and charmlibs.
Similarly, separated converse and convlibs.
So, super-installing charm++ does not try to make the libraries.

21 years agoAdded a sendraw method, that allows one to send message to ampi threads
Milind Bhandarkar [Wed, 26 Jul 2000 19:44:23 +0000 (19:44 +0000)]
Added a sendraw method, that allows one to send message to ampi threads
from outside ampi.

21 years agoAdded -C -stackvar and -g to f90 commandline. This is a temporary change.
Milind Bhandarkar [Wed, 26 Jul 2000 19:37:54 +0000 (19:37 +0000)]
Added -C -stackvar and -g to f90 commandline. This is a temporary change.
charmc will be modified later to handle more F90 opts.

21 years agoMade Fortran conn array start with element 1 instead of 0 in readElems.
Tim Hinrichs [Wed, 26 Jul 2000 18:52:44 +0000 (18:52 +0000)]
Made Fortran conn array start with element 1 instead of 0 in readElems.

21 years agofixed for gmake
Terry L. Wilmarth [Tue, 25 Jul 2000 16:04:05 +0000 (16:04 +0000)]
fixed for gmake

21 years agoFixed the AMPI_FORTRAN stuff. Gengbin had made temporary provisions for using
Milind Bhandarkar [Tue, 25 Jul 2000 11:35:05 +0000 (11:35 +0000)]
Fixed the AMPI_FORTRAN stuff. Gengbin had made temporary provisions for using
AMPI in NAMD. Refined most of those.
Separated Fortran AMPI interface completely. It now exists only in the
library to be linked with fortran. So, one does not have to wait
until running the program to discover that he/she has linked the wrong
ampi library.
removed the unnecessary ddt_send function from ddt.C.
renamed the Type_* methods of DDT to new* since they denote creation of
a new derived data type.
Fixed the AMPI_Type_* function names. They did not match the MPI standard
in proper capitalization. Fortran binding is obviously not affected.
Added a parameter "num" to the ddt serialize method. This indicates "external"
count.
made the dir parameter for serialize non-optional.
merged the needed functionality from tempo into ampi.
Reduction needs to be carried to the level of ddt.

21 years agoA small CentralLB change. When the stats are collected, a message
Robert Brunner [Mon, 24 Jul 2000 22:29:10 +0000 (22:29 +0000)]
A small CentralLB change.  When the stats are collected, a message
is sent, so that projections does not count the load balancer activity
as work for the last method executed.

21 years agoChnaged name of the copyBuffer method. It is now called "serialize". It takes
Milind Bhandarkar [Mon, 24 Jul 2000 21:54:43 +0000 (21:54 +0000)]
Chnaged name of the copyBuffer method. It is now called "serialize". It takes
an optional parameter called "direction", which when not specified, or if
set to "1", serializes the dtype. If it is set to 0, it deserializes
the ddt. Also, cleaned up code for serialize. Modified AMPI to make call to
serialize.

21 years agoAdded code to output the column-major connectivity matrix in case
Milind Bhandarkar [Mon, 24 Jul 2000 20:24:25 +0000 (20:24 +0000)]
Added code to output the column-major connectivity matrix in case
FEM_FORTRAN is defined to 1. Modified Makefile to make libfem.a and libfemf.a
from the same source fem.C by specifying different commandline flags to
charmc.

21 years agoAdded code to output the column-major connectivity matrix in case
Milind Bhandarkar [Mon, 24 Jul 2000 20:18:17 +0000 (20:18 +0000)]
Added code to output the column-major connectivity matrix in case
FEM_FORTRAN is defined to 1. Modified Makefile to make libfem.a and libfemf.a
from the same source fem.C by specifying different commandline flags to
charmc.