GitLab

recommendation.

released formally in the mpich2-1.3 series after more testing.

thread waiting around for a request that only completes at
MPI_Finalize time. Fixes ticket #917.

Reviewed by goodell.

[svn-r5689] Making sure that we don't have the ESCAPE character at the end of a command argument in SMPD. See ticket 915 for details

[svn-r5684] Changing *VERSION to MPICH2_*VERSION in windows configure - this change is the result of r5681

[svn-r5682] 1) Adding manually defined types and types defined from other types into the dictionary so that routines defining fortran types can use them. (2) Also added sanity checks when defining types from other types - Without the fix, types unavailable in the type dictionary where defined to be 0 in fortran/fortran90. This fixes ticket 922

pick up (this will be displayed in mpiexec -info). This required
replacing the "VERSION" string to "MPICH2_VERSION" since VERSION seems
to be overwritten by autoconf.

Reviewed by buntinas.

Reviewed by goodell.

[svn-r5676] fixed reduce_local.c and added fortran version, reducelocalf.f, and added these tests into testlist.  Resolving req#798.

Anthony noticed multiple serious problems in the reduce_local test.  The
most important was that it wasn't even being built and run!

No reviewer.

been replaced with the new pipelined algorithm for a few releases.

Reviewed by goodell.

settings. We currently use this to switch between PMI-1 and
PMI-1.1. PMI-2 is still a configure option, but with the recent
changes to use non-conflicting symbols, it can use this code in the
future as well.

Reviewed by goodell.

When MPICH2 is configured with --enable-strict it has the effect of
removing the u_char, u_short, u_int, and u_long types from sys/types.h.
Unfortunately, sys/stat.h still uses these types whether or not they are
defined.  So we define them ourselves if they are not present so that
our checks involving sys/stat.h will succeed later.

Reviewed by balaji@.

This commit adds a finalize callback to cleanup all of the classes
created by MPIX_Grequest_class_create.  This currently isn't thread-safe
under fine-grained threading.

Reviewed by balaji@.

Reviewed by goodell.

Fixes ticket #913.

No reviewer.

installation; this is needed on FreeBSD. Thanks to Thierry Thomas for
reporting it.

(this is needed for mpd).

reporting it.

future proof. Also, extend the UI to allow users to make binding
requests more intelligently.

The HYDT_bind_info structure itself should provide enough flexibility
for hwloc (barring any comments from Guillaume). The user still cannot
specify arbitrarily complex binding requests. Not sure how this can be
done without overly complicating the command line interface.

This should fix ticket #911.

expecting the signal handler to handle it (this is more reliable).

will automatically pick the hostname specified by the user in the
hostfile as the preferred communication interface. With this, Hydra
should work more correctly as described in
http://wiki.mcs.anl.gov/mpich2/index.php/Using_the_Hydra_Process_Manager#Hydra_with_Non-Ethernet_Networks

This is yet another band-aid on mpd's process mapping support.  The
right fix is probably just to use hydra and remove this functionality
from mpd.

Reviewed by balaji@.

[svn-r5634] Handling the case where we don't have info about logical procs on windows - eg: Win2K systems. If info about logical procs is not available we don't set proc affinity and allow these systems to continue working. See tickets 832,776

[svn-r5633] Define thread specific abstractions on windows as in unix. This change should be reverted back/re-visited after we figure out the *right* defaults for these abstractions