reported as a problem on the Tianhe-1A machine by Hongjia Cao @ NUDT.

Reviewed by buntinas.

to Yauheni Zelenko for reporting this issue.

exited. So, all the information with respect to the nodes on which the
proxies were launched was lost. Fixes ticket #1480.

processes. Also allow it to contain the number of processes on that
node.

well this sure turned out to be more involved than I imagined.  We do in fact
need to look at more than just the start and end offsets of the file domains.
Comparing MPI datatypes (file views) among all mpi processes requires some
work. In this approach we marshal the datatype into a serialized
representation, hash that representation, then compare that hash among all
processors.

utilities if anything in mpich2 ever decides to sha1-hash something.

RMK, use the hostname we detected for the machine, rather than using
"localhost" directly. This was causing problems for dynamic processes
as the business card contained "localhost" in that case.

Fixes ticket #1480.

[svn-r8463] Fixed configure test that erroneous allows initthread.c to call Fortran constant initialization function defined in setbotf.f.  This fixes bugreport #1476.

another workaroudn for bluegene out of memory conditions: when all processes
wish to read the same data, just do a read-and-broadcast instead of the full
two-phase algorithm.

The src/binding/f77/buildiface script can now handle an arbitrary number
of function prefixes, although "MPI_" and "MPIX_" are the only ones
enabled by default.  This commit also moves the MPIX_Icollective and
PMPIX_I* prototypes between "Begin Prototypes"/"End Prototypes" in
mpi.h so that the buildiface script can see them.

Reviewed by gropp@.

Reviewed by buntinas@.

This includes Rabenseifner's algorithm (reduce-scatter then allgather),
a recursive-doubling algorithm, and the SMP algorithm.

Reviewed by buntinas@.

While the CHKPMEM call is itself explicit, the schedule ops that are
added are easy to forget.  Including this barrier will occasionally
prevent some parallelism, but will frequently prevent incorrect code
that contains a non-obvious premature free.

Reviewed by buntinas@.

This mainly includes a few build system improvements compared to r112,
the previously used version.

No reviewer.

No reviewer.

This should hopefully fix the old nightly tests after r8383, wherein the
MPD and SMPD tests all failed because the daemon was never started.

No reviewer.

No reviewer.

No reviewer.

It caused problems when error checking was disabled.

No reviewer.

executables created in standalone mode and fixes ticket #1469.

installed in a specific path to confdb, and use it for PLPA as well.

The old one was stale and yielded a 404.

No reviewer.

This was the only remaining non-SMP algorithm to port from MPI_Reduce.

Reviewed by buntinas@.

This fixes MPIX_Ialltoallw as used in the coll/icalltoallw test.

No reviewer.

No reviewer.

The bug fix was a devilish stack corruption bug that was caused by using
a stack variable as the broadcast buffer.  After MPIR_Ibarrier_inter
returned, the memory for the variable was reused for other purposes,
resulting in erratic corruption and segfaults in the progress engine.
Mallocing the buffer instead resolves this problem.

The modernization includes switching to the MPID_SCHED_BARRIER macro and
dispatching sub-collectives via the coll_fns table instead of statically.

No reviewer.

Also, dispatch to the localcomm's coll_fns table instead of statically
selecting an intracomm bcast routine.

No reviewer.

No reviewer.