charm.git
19 years agoextend trace-summary +sumonly to be working when only partial log files are generated.
Gengbin Zheng [Tue, 13 Aug 2002 17:11:45 +0000 (17:11 +0000)]
extend trace-summary +sumonly to be working when only partial log files are generated.

19 years agoAdded language "charm++f", which is just charm++ but with the
Orion Lawlor [Mon, 12 Aug 2002 23:51:52 +0000 (23:51 +0000)]
Added language "charm++f", which is just charm++ but with the
Fortran runtime libraries.  This is needed by Charm++ programs
that happen to include a small amount of (serial) fortran.

19 years agoAdded fixed-format f90 compiler.
Orion Lawlor [Mon, 12 Aug 2002 20:10:16 +0000 (20:10 +0000)]
Added fixed-format f90 compiler.

19 years agoAdded CF90_FIXED, for fixed-format f90 source files.
Orion Lawlor [Mon, 12 Aug 2002 20:09:38 +0000 (20:09 +0000)]
Added CF90_FIXED, for fixed-format f90 source files.

19 years agodisable trace user events in bluegene mode.
Gengbin Zheng [Sun, 11 Aug 2002 22:57:14 +0000 (22:57 +0000)]
disable trace user events in bluegene mode.

19 years agoregister module for trace-summary in trace all case.
Gengbin Zheng [Fri, 9 Aug 2002 20:30:21 +0000 (20:30 +0000)]
register module for trace-summary in trace all case.

19 years agoAdded inlineable version of find.
Orion Lawlor [Fri, 9 Aug 2002 20:12:50 +0000 (20:12 +0000)]
Added inlineable version of find.

19 years agoclean up blue gene timing correction.
Gengbin Zheng [Fri, 9 Aug 2002 06:18:13 +0000 (06:18 +0000)]
clean up blue gene timing correction.

19 years agofixed a buf when only 1 pe, broadcast should be treated as send to CkMyPe().
Gengbin Zheng [Thu, 8 Aug 2002 23:29:53 +0000 (23:29 +0000)]
fixed a buf when only 1 pe, broadcast should be treated as send to CkMyPe().

19 years agocall endCompute before send summary data to 0.
Gengbin Zheng [Thu, 8 Aug 2002 17:58:53 +0000 (17:58 +0000)]
call endCompute before send summary data to 0.

19 years agofully implement some missing features for some special cases since now skipCld also...
Gengbin Zheng [Thu, 8 Aug 2002 06:12:42 +0000 (06:12 +0000)]
fully implement some missing features for some special cases since now skipCld also go to this.

19 years agofixed a timing bug caused by skipCldEnqueue send a message to itself.
Gengbin Zheng [Thu, 8 Aug 2002 04:00:05 +0000 (04:00 +0000)]
fixed a timing bug caused by skipCldEnqueue send a message to itself.

19 years agomake a CkExit() wrapper and added CkExitFn call back, which will do some task and...
Gengbin Zheng [Wed, 7 Aug 2002 21:05:05 +0000 (21:05 +0000)]
make a CkExit() wrapper and added CkExitFn call back, which will do some task and come back to CkExit. This allow, for example, trace-summary can get control before real CkExit and collects summary data into one node before CkExit shutdown the program.

19 years agoadded +sumonly which will combine summary bin data from all pes to pe0 and write...
Gengbin Zheng [Wed, 7 Aug 2002 21:02:39 +0000 (21:02 +0000)]
added +sumonly which will combine summary bin data from all pes to pe0 and write only one file. This use registerCkExit() callback in init.C so that when CkExit is called, this collection procedure starts.

19 years agoupdated Makefile for new trace-summary.ci, etc.
Gengbin Zheng [Wed, 7 Aug 2002 21:00:59 +0000 (21:00 +0000)]
updated Makefile for new trace-summary.ci, etc.
charmc to register the module.

19 years agoadding 1us as default time for an entry in bgelapse mode.
Gengbin Zheng [Wed, 7 Aug 2002 21:00:07 +0000 (21:00 +0000)]
adding 1us as default time for an entry in bgelapse mode.

19 years agominor change in the info print .
Gengbin Zheng [Wed, 7 Aug 2002 06:51:09 +0000 (06:51 +0000)]
minor change in the info print .

19 years agoIf conv-mach.sh didn't define CMK_F90_OPTIMIZE, charmc would
Orion Lawlor [Tue, 6 Aug 2002 19:54:27 +0000 (19:54 +0000)]
If conv-mach.sh didn't define CMK_F90_OPTIMIZE, charmc would
silently ignore -O for f90 files.

19 years agoGM version doesn't have a SIGIO handler, so I can't register
Orion Lawlor [Tue, 6 Aug 2002 18:32:26 +0000 (18:32 +0000)]
GM version doesn't have a SIGIO handler, so I can't register
stdout for a SIGIO.  The (probably better) alternative is to
add a SIGIO handler for the GM version that checks stdout.

19 years agoadded BgElapse(deltaT) user interface to let user update the thread time. It is the...
Gengbin Zheng [Tue, 6 Aug 2002 05:37:46 +0000 (05:37 +0000)]
added BgElapse(deltaT) user interface to let user update the thread time. It is the default timing method now. use +bgwalltime to switch to the old method.

19 years agoUpdated to use the armci "preferred" startup interface with argc, argv
Chee Wai Lee [Tue, 6 Aug 2002 04:34:24 +0000 (04:34 +0000)]
Updated to use the armci "preferred" startup interface with argc, argv

19 years agofix a case when trace user event is called before traceInit.
Gengbin Zheng [Mon, 5 Aug 2002 19:53:38 +0000 (19:53 +0000)]
fix a case when trace user event is called before traceInit.

19 years agoAdded CkMsgQ, a pup-able queue of Charm++ messages.
Orion Lawlor [Mon, 5 Aug 2002 19:21:33 +0000 (19:21 +0000)]
Added CkMsgQ, a pup-able queue of Charm++ messages.

19 years agouse elan_poll instead of elan_wait for the first elan_put.
Gengbin Zheng [Sat, 3 Aug 2002 13:51:30 +0000 (13:51 +0000)]
use elan_poll instead of elan_wait for the first elan_put.

19 years agoadded LDTurnManualLBOn() for manual load balancing mode. call StartLB() to start...
Gengbin Zheng [Fri, 2 Aug 2002 20:19:51 +0000 (20:19 +0000)]
added LDTurnManualLBOn() for manual load balancing mode. call StartLB() to start LB and avoid using local barrier.

19 years agoRedefining "exit" is dangerous on machines where the system
Orion Lawlor [Fri, 2 Aug 2002 19:37:57 +0000 (19:37 +0000)]
Redefining "exit" is dangerous on machines where the system
headers decorate "exit", so instead call MPI_Exit from MPI_Finalize.

19 years agoFixed 4 bugs:
Orion Lawlor [Fri, 2 Aug 2002 18:47:31 +0000 (18:47 +0000)]
Fixed 4 bugs:

- Fortran version ampif.C referred to mpi_ncomms, which I
renamed mpi_nworlds

- MPI_Bcast specified "root" as the source, while corresponding
receives specified nbcasts.

- Several methods (MPI_Alltoall, etc.) had "if (comm!=MPI_COMM_WORLD)"
tests.

- The MPI idiom:
MPI_Finalize();
exit(1);
would cause charmrun to die with "socket closed before recv".
Solution is to redefine exit to call MPI_Exit; although a cleaner
solution might be to redefine MPI_Finalize to call TCharmDone.

19 years agofixes to build
Jim Phillips [Fri, 2 Aug 2002 18:40:46 +0000 (18:40 +0000)]
fixes to build

19 years agoupdated Charm version number, update conv-autoconfig.h.in which should be done last...
Gengbin Zheng [Fri, 2 Aug 2002 18:27:25 +0000 (18:27 +0000)]
updated Charm version number, update conv-autoconfig.h.in which should be done last time for long double type.

19 years agoAdded sanity check to PUP::sizer. This check made debugging a
Orion Lawlor [Fri, 2 Aug 2002 16:24:40 +0000 (16:24 +0000)]
Added sanity check to PUP::sizer.  This check made debugging a
pup-routine problem infinitely easier.

19 years agoClarify most common case-- Linux PC.
Orion Lawlor [Fri, 2 Aug 2002 16:11:25 +0000 (16:11 +0000)]
Clarify most common case-- Linux PC.

19 years agoRelatively minor portability and migration-correctness fixes:
Orion Lawlor [Fri, 2 Aug 2002 15:50:27 +0000 (15:50 +0000)]
Relatively minor portability and migration-correctness fixes:

1.) &(void *)messagePtr isn't actually valid C++, because & only
applies to Lvalues (which the result of a typecast isn't).
You want (void **)&messagePtr, which is valid C++ and will be
accepted by compilers other than the (promiscuous) g++.

2.) Casting addresses to int's doesn't work on machines where
addresses are 8 bytes and ints are 4; you aught to just marshall
"pointer"s instead of "int"s.

3.) CtvStaticDeclare(ArmciVirtualProcessor *, _armci_ptr)
declares a separate _armci_ptr for each file it's included in.
"Static" globals should never, ever go in header files;
if it's used by multiple files, you need CtvExtern in the
header and a CtvDeclare in exactly one source file.

4.) ArmciVirtualProcessor didn't have a destructor, so it leaked
(isomalloc'd!) memory when it migrated.

5.) addressReply isn't initialized to NULL in the constructor,
so pup will crash if you hadn't already done a put.  Similarly,
put didn't reset addressReply to NULL after deleting the message,
so the pup routine wouldn't have worked.

19 years agoadded a prototype of a function.
Gengbin Zheng [Fri, 2 Aug 2002 14:06:18 +0000 (14:06 +0000)]
added a prototype of a function.

19 years agochanged a function name
Gengbin Zheng [Fri, 2 Aug 2002 14:06:01 +0000 (14:06 +0000)]
changed a function name

19 years agopersistent communication for elan.
Gengbin Zheng [Fri, 2 Aug 2002 13:51:43 +0000 (13:51 +0000)]
persistent communication for elan.

19 years agoupdated for persistent comm.
Gengbin Zheng [Fri, 2 Aug 2002 13:50:05 +0000 (13:50 +0000)]
updated for persistent comm.

19 years agopersistent communication interface(now only for elan version.
Gengbin Zheng [Fri, 2 Aug 2002 13:49:34 +0000 (13:49 +0000)]
persistent communication interface(now only for elan version.

19 years agofixed couple compiler warnings.
Gengbin Zheng [Fri, 2 Aug 2002 05:19:06 +0000 (05:19 +0000)]
fixed couple compiler warnings.

19 years agoAdded armci "skeleton" library and example test program.
Chee Wai Lee [Fri, 2 Aug 2002 00:09:35 +0000 (00:09 +0000)]
Added armci "skeleton" library and example test program.

19 years agobring back the fix for grow down stack in makecontext version that Jim removed.
Gengbin Zheng [Thu, 1 Aug 2002 21:28:33 +0000 (21:28 +0000)]
bring back the fix for grow down stack in makecontext version that Jim removed.

19 years agoUse explicit casts to CkCallback for stupid Win32 compiler.
Orion Lawlor [Thu, 1 Aug 2002 20:16:38 +0000 (20:16 +0000)]
Use explicit casts to CkCallback for stupid Win32 compiler.

19 years agoWin32 compiler gets confused by PUP::able::pup; so use a typedef.
Orion Lawlor [Thu, 1 Aug 2002 20:16:10 +0000 (20:16 +0000)]
Win32 compiler gets confused by PUP::able::pup; so use a typedef.

19 years ago"union" is a keyword in C/C++!
Orion Lawlor [Thu, 1 Aug 2002 16:23:10 +0000 (16:23 +0000)]
"union" is a keyword in C/C++!

19 years agoAdded net-hp-ia64-acc port.
Jim Phillips [Thu, 1 Aug 2002 15:25:54 +0000 (15:25 +0000)]
Added net-hp-ia64-acc port.

19 years agoHack-free context stack creation, now works on hpux-ia64 as well.
Jim Phillips [Thu, 1 Aug 2002 15:20:54 +0000 (15:20 +0000)]
Hack-free context stack creation, now works on hpux-ia64 as well.

19 years agoPortability for Win32.
Orion Lawlor [Wed, 31 Jul 2002 22:55:33 +0000 (22:55 +0000)]
Portability for Win32.

19 years agoBy default, clip the pixels of images as they're added. This prevents
Orion Lawlor [Wed, 31 Jul 2002 22:22:13 +0000 (22:22 +0000)]
By default, clip the pixels of images as they're added.  This prevents
overflow for cases like the Astro code where several processors may
contribute large values to each pixel.

(Need to write the image-combine function using MMX instructions on Intel)

19 years agoAllow CkVector's of different types to be converted among one another.
Orion Lawlor [Wed, 31 Jul 2002 22:21:09 +0000 (22:21 +0000)]
Allow CkVector's of different types to be converted among one another.

19 years agoi/o interrupt was using cputimer, fixed.
Gengbin Zheng [Wed, 31 Jul 2002 19:33:15 +0000 (19:33 +0000)]
i/o interrupt was using cputimer, fixed.

19 years agoReplaced call to cbrt with call to pow(), for win32.
Orion Lawlor [Wed, 31 Jul 2002 18:07:57 +0000 (18:07 +0000)]
Replaced call to cbrt with call to pow(), for win32.

From cbrt man page:
"cbrt() is a GNU extension".

19 years agoBug fix-- I copied-and-pasted the writeback-to-array code from
Orion Lawlor [Wed, 31 Jul 2002 15:26:46 +0000 (15:26 +0000)]
Bug fix-- I copied-and-pasted the writeback-to-array code from
the long double case, and forgot to adjust one datatype.

19 years agoCleaned up command-line option handling, and added
Orion Lawlor [Tue, 30 Jul 2002 23:15:45 +0000 (23:15 +0000)]
Cleaned up command-line option handling, and added
OPTS_CPP to f90 command line.  (f90 runs preprocessor
on files that end in .F or .F90)

19 years agoAdded MPI_Comm_Split, and the support needed for all sorts of communicators.
Orion Lawlor [Tue, 30 Jul 2002 22:38:17 +0000 (22:38 +0000)]
Added MPI_Comm_Split, and the support needed for all sorts of communicators.

A communicator is now implemented as a new shadow array, bound to the
TCharm threads an a "holder" called ampiParent.  This allows the messaging,
broadcasts, and reductions that happen on one communicator to be independent
of other communicators.

This implementation has not been tested well, and there are likely bugs
remaining.  See the FIXME's for details.

19 years agoConsume command-line arguments on *all* procesors, not just processor 0.
Orion Lawlor [Tue, 30 Jul 2002 22:35:35 +0000 (22:35 +0000)]
Consume command-line arguments on *all* procesors, not just processor 0.

19 years agoRearranged command-line argument handling.
Orion Lawlor [Tue, 30 Jul 2002 22:35:00 +0000 (22:35 +0000)]
Rearranged command-line argument handling.

Command-line arguments need to be stripped on *all* processors,
not just processor 0.  This is because the user can now access
the command line on any processor, not just processor 0.

19 years agoAdded comment warning about dangerous uses of CkVec.
Orion Lawlor [Tue, 30 Jul 2002 22:30:06 +0000 (22:30 +0000)]
Added comment warning about dangerous uses of CkVec.

The right solution probably would have been to change CkVec
to be less dangerous (by calling copy constructors); but this
is quite difficult to do right in C++, and std::vector does it
properly already.

19 years agoSupport for new types.
Orion Lawlor [Tue, 30 Jul 2002 20:22:34 +0000 (20:22 +0000)]
Support for new types.

19 years agoRemoved reliance on "CMK_BOOL_UNDEFINED", changed to "!CMK_BOOL_DEFINED".
Orion Lawlor [Tue, 30 Jul 2002 20:22:14 +0000 (20:22 +0000)]
Removed reliance on "CMK_BOOL_UNDEFINED", changed to "!CMK_BOOL_DEFINED".

19 years agoAdded support for "long long", "long double" pupping.
Orion Lawlor [Tue, 30 Jul 2002 20:20:27 +0000 (20:20 +0000)]
Added support for "long long", "long double" pupping.

19 years agoAdded CMK_LONG_LONG_DEFINED, CMK___int64_DEFINED, and CMK_LONG_DOUBLE_DEFINED.
Orion Lawlor [Tue, 30 Jul 2002 20:19:37 +0000 (20:19 +0000)]
Added CMK_LONG_LONG_DEFINED, CMK___int64_DEFINED, and CMK_LONG_DOUBLE_DEFINED.

19 years agoDon't use AMPI version of mpi.h-- it doesn't have the same
Orion Lawlor [Tue, 30 Jul 2002 19:09:43 +0000 (19:09 +0000)]
Don't use AMPI version of mpi.h-- it doesn't have the same
data type, source, and tag constants as the system mpi, and
MPI_Status is layed out differently.

We need to come up with some automatic way of grabbing the
system's mpi, not ampi...

19 years agoFixed ancient bug-- Construct::external is never initialized if
Orion Lawlor [Tue, 30 Jul 2002 18:29:02 +0000 (18:29 +0000)]
Fixed ancient bug-- Construct::external is never initialized if
setExtern isn't called.  setExtern isn't called for the inner
construct in a templated class.

This has been fixed two ways:
  -external is always initialized (to 0).
  -setExtern is always called, even from a templated class.

This actually fixes a bug on the IBM SP3 b80login.sdsc.edu.

19 years agoMinor compiler fix for windows.
Orion Lawlor [Tue, 30 Jul 2002 17:15:51 +0000 (17:15 +0000)]
Minor compiler fix for windows.

19 years agoMinor compiler compatability fixes for mpi-sp.
Orion Lawlor [Tue, 30 Jul 2002 17:12:21 +0000 (17:12 +0000)]
Minor compiler compatability fixes for mpi-sp.

19 years agosmall update to manual.tex
Chao Huang [Mon, 29 Jul 2002 21:47:33 +0000 (21:47 +0000)]
small update to manual.tex

19 years agoDeclaring a "static const int SIZE=..." in a class has several
Orion Lawlor [Mon, 29 Jul 2002 18:05:51 +0000 (18:05 +0000)]
Declaring a "static const int SIZE=..." in a class has several
portability problems:
  -in-class static initializers aren't portable C++ (yet)
  -"SIZE" is #defined by the (cursed) MS headers
Changed to HIST_SIZE.

19 years agoHoisted nonportable uses of "for(int i=...".
Orion Lawlor [Mon, 29 Jul 2002 17:29:48 +0000 (17:29 +0000)]
Hoisted nonportable uses of "for(int i=...".

19 years agoUse "PF_UNIX" instead of (linux-ism) "PF_LOCAL", and disable the
Orion Lawlor [Mon, 29 Jul 2002 17:25:41 +0000 (17:25 +0000)]
Use "PF_UNIX" instead of (linux-ism) "PF_LOCAL", and disable the
printf SIGIO if we're on an SMP system.

19 years agochanged variable names.
Gengbin Zheng [Sun, 28 Jul 2002 21:47:37 +0000 (21:47 +0000)]
changed variable names.

19 years agoa minor cleanup.
Gengbin Zheng [Sun, 28 Jul 2002 21:39:03 +0000 (21:39 +0000)]
a minor cleanup.

19 years agooutput distribution of entry time.
Gengbin Zheng [Sun, 28 Jul 2002 21:25:07 +0000 (21:25 +0000)]
output distribution of entry time.

19 years agofixed a bug in msg size.
Gengbin Zheng [Sun, 28 Jul 2002 19:58:40 +0000 (19:58 +0000)]
fixed a bug in msg size.

19 years agocall the ccd callbacks in reverse order, so that the NodifyIdle is the last one to...
Gengbin Zheng [Sun, 28 Jul 2002 19:34:06 +0000 (19:34 +0000)]
call the ccd callbacks in reverse order, so that the NodifyIdle is the last one to be called.

19 years agoadd +idleblocking, call blocking version of recv in idle(for testing on Lemieux).
Gengbin Zheng [Sun, 28 Jul 2002 18:58:39 +0000 (18:58 +0000)]
add +idleblocking, call blocking version of recv in idle(for testing on Lemieux).

19 years agoUse a UNIX socket to trap stdout output, which lets us grab a SIGIO,
Orion Lawlor [Fri, 26 Jul 2002 20:02:36 +0000 (20:02 +0000)]
Use a UNIX socket to trap stdout output, which lets us grab a SIGIO,
which means no more error-prone "Output discontinuity" messages!

19 years agoadded static class method LBDatabase::Object() to return the load branch.
Gengbin Zheng [Fri, 26 Jul 2002 16:47:30 +0000 (16:47 +0000)]
added static class method LBDatabase::Object() to return the load branch.
also add StartLB() interface.

19 years agoTest out mesh_updated after running time loop.
Orion Lawlor [Fri, 26 Jul 2002 14:56:44 +0000 (14:56 +0000)]
Test out mesh_updated after running time loop.

19 years agoWorkaround for Intel C++ 6.0 compiler.
Orion Lawlor [Fri, 26 Jul 2002 14:22:04 +0000 (14:22 +0000)]
Workaround for Intel C++ 6.0 compiler.

19 years agofixed int i.
Gengbin Zheng [Fri, 26 Jul 2002 13:31:44 +0000 (13:31 +0000)]
fixed int i.

19 years agoSilly "if (x=1)" bug.
Orion Lawlor [Thu, 25 Jul 2002 22:08:15 +0000 (22:08 +0000)]
Silly "if (x=1)" bug.

19 years agoFixed FEM_Update_Mesh-- it's been broken since TCharm.
Orion Lawlor [Thu, 25 Jul 2002 21:52:29 +0000 (21:52 +0000)]
Fixed FEM_Update_Mesh-- it's been broken since TCharm.

19 years agoDetermine symmetries from a (permuted) list of matching faces.
Orion Lawlor [Thu, 25 Jul 2002 21:48:22 +0000 (21:48 +0000)]
Determine symmetries from a (permuted) list of matching faces.

19 years agoAdded CkPupAblePtrVec, a vector of pointers to PUP::able's.
Orion Lawlor [Thu, 25 Jul 2002 20:58:50 +0000 (20:58 +0000)]
Added CkPupAblePtrVec, a vector of pointers to PUP::able's.
I really need to start using templated pointer classes
instead of longer and longer type names here.

19 years agoAdded symmetry determination based on matching up (permuted)
Orion Lawlor [Thu, 25 Jul 2002 20:57:53 +0000 (20:57 +0000)]
Added symmetry determination based on matching up (permuted)
face lists.  This amounts to calling FEM_Add_Linear_Periodicity
in init, and FEM_Sym_Coordinates in driver.

19 years agorecovered the fix for the striping Cmi_argv.
Gengbin Zheng [Thu, 25 Jul 2002 20:53:33 +0000 (20:53 +0000)]
recovered the fix for the striping Cmi_argv.

19 years agofix for striping Cmi_argv, works for SMP version.
Gengbin Zheng [Thu, 25 Jul 2002 20:49:57 +0000 (20:49 +0000)]
fix for striping Cmi_argv, works for SMP version.

19 years agoMaking CldEnqueue call CmiFreeListSendFn which can then be implemenetd as a multicast...
Sameer Kumar [Thu, 25 Jul 2002 19:44:00 +0000 (19:44 +0000)]
Making CldEnqueue call CmiFreeListSendFn which can then be implemenetd as a multicast library in the machine layer.

19 years agoEnabling list send functions.
Sameer Kumar [Thu, 25 Jul 2002 19:42:44 +0000 (19:42 +0000)]
Enabling list send functions.

19 years agofixed compilation bug!
Sameer Kumar [Thu, 25 Jul 2002 19:35:52 +0000 (19:35 +0000)]
fixed compilation bug!

19 years agoEnabling the CmiListSend functions in machine.c
Sameer Kumar [Thu, 25 Jul 2002 19:34:02 +0000 (19:34 +0000)]
Enabling the CmiListSend functions in machine.c

19 years agoMessages are now being pulled by the receiver instead of being pushed by the sender!
Sameer Kumar [Thu, 25 Jul 2002 19:17:29 +0000 (19:17 +0000)]
Messages are now being pulled by the receiver instead of being pushed by the sender!

19 years agofixed argv stripping bug.
Gengbin Zheng [Thu, 25 Jul 2002 19:12:01 +0000 (19:12 +0000)]
fixed argv stripping bug.

19 years agoAccept -E, -M, -S, and -P options, which supress the normal
Orion Lawlor [Wed, 24 Jul 2002 20:50:13 +0000 (20:50 +0000)]
Accept -E, -M, -S, and -P options, which supress the normal
.o output and create various bizarre preprocessor or
assembler outputs.  -M is needed by the Rocket center codes.

19 years agoSlightly rearranged startup sequence-- MPI_Setup now gets
Orion Lawlor [Wed, 24 Jul 2002 20:48:34 +0000 (20:48 +0000)]
Slightly rearranged startup sequence-- MPI_Setup now gets
called for *both* C and f90, with the non-defined version
supplied as a compat_ routine.  If they're both non-defined,
use a new Main that calls all *three* of C, C++, and f90
main routines.

This is all needed for calling MPI main routines defined in
C++, which is needed by the Rocket center codes.

19 years agoAdded "include"-able version of AMPI headers, and
Orion Lawlor [Wed, 24 Jul 2002 20:45:56 +0000 (20:45 +0000)]
Added "include"-able version of AMPI headers, and
made the MPIINCL module use that version.

19 years agoRenamed "DDT_..." data types "CkDDT_...". This is cleaner,
Orion Lawlor [Wed, 24 Jul 2002 20:44:08 +0000 (20:44 +0000)]
Renamed "DDT_..." data types "CkDDT_...".  This is cleaner,
and avoids a namespace collision with the genx routines of the
same names.

19 years agoLarger receive buffer and some cleanup.
Sameer Kumar [Wed, 24 Jul 2002 17:02:16 +0000 (17:02 +0000)]
Larger receive buffer and some cleanup.

19 years agoNeed 'mpif.h' because some fortran guys just include it directly.
Orion Lawlor [Wed, 24 Jul 2002 15:22:46 +0000 (15:22 +0000)]
Need 'mpif.h' because some fortran guys just include it directly.
Other fortran people "use MPIINCL", though, so ampimod.f90 is still needed.

19 years agocorrected a mistake
Chao Huang [Wed, 24 Jul 2002 13:59:51 +0000 (13:59 +0000)]
corrected a mistake

19 years agoadded seperate projections manual
Chao Huang [Wed, 24 Jul 2002 03:54:53 +0000 (03:54 +0000)]
added seperate projections manual