GitLab

[svn-r7012] replaced NMPI_[i]Send, [i]Recv, Wait[all] and Test with MPIR__impl, or directly called MPID_ functions.  Reviewed by jayesh@

[svn-r7011] Added testing for MPI version in configure; runtest now implements an mpiversion=2.2 option for tests so that tests that require new routines are not run on older MPI implementations, and enhanced error reporting from many of the tests to help in giving more specific error messages for MPI implementations that fail the tests (these particular tests chosen because they failed on one or more versions of the IBM MPI implementation)

[svn-r7009] Re-introduce the fixed-width int tests that were removed in r6956. Without these checks tests that use fixed-width ints fail to compile on windows

process group; we were using the value for PG0 for all process groups.

Due to the tangled mess of header include ordering, a convenience #ifdef was
checked too early in another header when it was not yet defined.  This led to a
null implementation of MPIDI_CH3I_PROGRESS_WAKEUP in ch3:sock, even though this
is essential functionality for sock.

No reviewer.

[svn-r7005] Remove support for the MPI-1.0 only datatypes MPI_2COMPLEX and MPI_2DOUBLE_COMPLEX (support and be reenable by defining MPICH_DEFINE_2COMPLEX in case we have any users that really need it)

[svn-r7004] Add support to check for 2COMPLEX and 2DOUBLE_COMPLEX, which only existed in MPI 1.0 (and was removed in 1.1)

based on the suggestion by Yauheni Zelenko @ Cadence.

No reviewer.

No reviewer.

[svn-r6998] Specify that MPI_2COMPLEX is available on windows - The check for availability was added in r6990

[svn-r6997] Due to changes in the autoconf macros, the PAC macro to select the compilers to use with MPI programs required multiple changes, as autoconf would prematurely run AC_PROG_CC.  These factor the MPI checks into provide AC_ARG options, select compilers, and check libraries

[svn-r6996] Due to changes in the autoconf macros, the PAC_PROG_CC version no longer does all that AC_PROG_CC does, and thus can no longer assert that it provides AC_PROG_CC

initcxx.cxx was needlessly including mpihandlemem.h

No reviewer.

snprintf.

restrictive model that functions under the following constraints
(mainly because of the inflexibility of POE):

1. The load-leveler scripts cannot specify more than one tasks per
node. The number of tasks per node need to be specified to mpiexec
using the -ppn option.

2. The number of processes being launched need to cover all of the
allocated nodes.

No reviewer.

There are probably more uses out there that I am missing at this point.

No reviewer.

No reviewer.

This is necessary because there is no portable way to statically initialize
mutexes.  Windows doesn't support it, and some older POSIX impls don't either.

No reviewer.

No reviewer.

demonstrates a problem on Blue Gene/P.

user.

[svn-r6974] Fix to make async.c not get added to libpmpich.a in order to avoid warnings about empty archives on mac

per node as well. Based on the suggestion by Yauheni Zelenko @
Cadence.

it. Based on the suggestion by Yauheni Zelenko @ Cadence.

count needs to be per process group, not global.