GitLab

some cases.

No reviewer.

No reviewer.

No reviewer.

On Blue Gene (and some other platforms) the compiler occasionally generates
suboptimal code if it guesses the likelihood of a branch incorrectly.
These macros allow us to help the compiler out in cases that it guesses
poorly.

No reviewer.

No reviewer.

[svn-r7366] fixed bug in spawn where a bad port caused non root processes to hang.  reviewed by goodell@

This will hopefully be helpful once MPI-3 is official.

No reviewer.

This isn't a good test, it just helps to make sure that all 17 new
nonblocking collective functions have sensible function prototypes,
exist in the library, and don't explode when you call them with simple
arguments.

The test is currently disabled in the testlist since we will currently
always fail at runtime.

No reviewer.

These functions build but will yield an error at runtime if invoked.
They will correctly call any collective function overrides specified in
comm_ptr->coll_fns, permitting devices to experiment with NBC
implementation before a complete implementation is available above the
ADI.

Reviewed by buntinas@.

Prior to this commit, the time that MPI_Group_translate_ranks took was
dependent on the _values_ of the input ranks, the size and pattern of
group2, as well as the number of input ranks.  Now for comm_world's
group (and simple subsets thereof) the translation time only depends on
the number of ranks that are translated.

The MPI_COMM_WORLD use case is essential for tools like Scalasca that
must translate communicator ranks to canonical ranks in MPI_COMM_WORLD.

Reviewed by buntinas@.

The error threshold ratio, NUM_LOOPS, or number of processes may need to
be tweaked in order to make this test pass/fail reliably on different
systems.

Reviewed by buntinas@.

This test was accidentally not enabled in the testlist for some reason.

Reviewed by buntinas@.

This should make it much easier to tune the collective algorithm
selection on a given platform.  It also provides some knobs to expose
to apps/tools via the upcoming MPIT interface.

Reviewed by buntinas@.

This adds convenience and isolates the caller from changes in the MPIR_Param
interface's implementation details.

Reviewed by buntinas@.

the number of bytes read via the status object.  Should have been either
MPI_STATUS_IGNORE or (as in this cs) a real status

they do is the same.

[svn-r7342] Explicitly elevate the permissions of the windows installer. Also see ticket 1070 for details

[svn-r7341] replaced the runtime test that determine sizeof($MPI_OFFSET_TYPE) where MPI_OFFSET_TYPE is determined by ROMIO by AC_COMPUTE_INT of sizeof(MPI_Offset).  This fix is related to req#1121.

Logging the depth value before asserting on it is more helpful
for debugging...

No reviewer.

The INIT CS wasn't recursive, so it didn't update the depth, so it would
blow up when assertions were enabled.  This only caused a problem for the
spawning code because the non-spawning MPID_Init didn't actually perform
any communication, so it never attempted to YIELD in progress.

No reviewer.

An unintended consequence of eliminating the NMPI_ interface is that
multithreaded ROMIO usage began to fail because of recursively locking a
non-recursive mutex.  Making ALLFUNC recursive is a bit heavy handed,
but it effectively restores this aspect the NMPI_ behavior and fixes
ROMIO+MPI_THREAD_MULTIPLE.

No reviewer.

[svn-r7337] Ignore duplicate resource info from system. At least with Win7 on VirtualBox we see these dup infos. See ticket #1115 for details

[svn-r7336] Increase the host desc length in SMPD - reqd for current systems with multiple interfaces, ipv6 addrs, bridges etc

No reviewer.

[svn-r7334] Fixed bug where a cookie was allocated on the stack, but the send didn't complete immediately.  This is a work-around fix for to bug listed in #1039 just for knem

On its face, this value seems like it should only affect multithreaded
performance.  However it also can impact single-threaded, yet oversubscribed
(via processes) where MPI_THREAD_MULTIPLE is used anyway.

Reviewed by buntinas@.

On Darwin at least, performance of the yield implementation without this
fix is very poor in the face of oversubscription.

Reviewed by buntinas@.

Reviewed by buntinas@.

Fix for Finalize sequence