Pavan Balaji authored Apr 08, 2015 and Kenneth Raffenetti committed Apr 10, 2015

The current number of combinations we are checking are too many,
causing the test to take too long on some platforms.  This patch
simplifies the test, so we build two versions of the test.  In the
first version, we run only on COMM_WORLD but go through all datatypes.
In the second version, we run on all communicators, but go through
only a small subset of datatypes.

Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

Pavan Balaji authored Apr 08, 2015 and Kenneth Raffenetti committed Apr 10, 2015

1. Renamed bcast2 to bcast.

2. White-space cleanup for bcast.c

Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

Pavan Balaji authored Apr 08, 2015 and Kenneth Raffenetti committed Apr 10, 2015

This test is exactly the same as bcast2.  Originally these two tests
were different, but over time they have become essentially the same.
There's no point testing the same thing twice.

Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

FreeBSD seems to take slightly more than 20 minutes to run these tests
primarily because the poll() call is somehow superslow on that OS.
This new timeout should be sufficient to cover that case.

No reviewer.

Put tests that check if MPI correctly detects aliased buffers in
collective operations into the errors section of the testsuite.

Fixes #2211

Signed-off-by: Wesley Bland <wbland@anl.gov>

Some overloaded nightly test nodes use almost 20 minutes for running
these tests. We increase their time limit for now to easily figure out
other bugs reported by nightly test.

Nightly testing reported timeout on octopus with 12 timelimit. Each of
them took 12:05 ~ 12:10 mins.

Timeout is reported on some overloaded machines with 10 minutes time
limitation.

Signed-off-by: Xin Zhao <xinzhao3@illinois.edu>

Signed-off-by: Wesley Bland <wbland@anl.gov>

Added tests to the existing test suite. Usually these tests are next to
existing MPI_IN_PLACE tests to make sure that the IN_PLACE checks are correct.
For the nonblocking tests, they are all in one big test (nonblocking4) since
the other nonblocking tests were already in that format.

Fixes #2049

Signed-off-by: Antonio J. Pena <apenya@mcs.anl.gov>

We made progress for MPI_Wait and friends, but not
MPI_Test/MPI_Iprobe/etc.  Thanks to Jed Brown for letting us know about
this issue.

fixes #1785

Reviewed-by: buntinas

No reviewer.

strict=false.

Reviewed by dinan.

[svn-r9918] Added simple intercomm reduce_scatter_block test to test code that the coverage tests show is currently uncovered

See http://lists.mcs.anl.gov/pipermail/mpich2-dev/2012-April/000927.html

No reviewer.

[svn-r9808] Ensured that the master_top_srcdir is set so that the f90 tests will not fail due to unresolved dependencies.  Added strict=value as runtests option.  Added some versions of the collective tests using the pmpi adapter to use the nonblocking collective routines

The problem is that the (previously on by default) collective error
propagation functionality is implemented in a way that is fundamentally
incompatible with the way that nonblocking collectives are implemented.

Thanks to Jed Brown for the bug report and reproducer.

No reviewer.

I thought 420 seconds was long enough, but it still ran too long on crank last
night.  If it's still failing after this we'll need to consider other options
besides just increasing the time limit.

No reviewer.

The nonblocking3 test, in particular, was failing with timeouts in the nightly
tests.  420 seconds is the value we use for bcast2 and bcast3, so it seems
reasonable to use the same value here.  If nonblocking3 continues to fail with
timeouts with the new value then there is probably a hang occurring in the test
or the MPI library.

No reviewer.

The test "randomly" selects NBC requests to post and ways to complete
those requests, over and over again in a loop.  This test is intended to
stress the NBC implementation by keeping many collectives in flight
simultaneously on multiple communicators, and by completing them with
multiple different mechanisms.  The current NBC ops that are actually
posted are derived from the nonblocking2 test.

Reviewed by buntinas@.

Also fix a small bug in the "nonblocking" test.

Reviewed by buntinas@.

[svn-r8277] Added a test for intercomm reduce scatter, and corrected a few bugs in error reporting in reduce-scatter tests

[svn-r7556] Added some additional tests for the collective routines to address coverage gaps in testing

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.

[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)

The send and recv buffers aliased each other in an "in place" fashion,
yet MPI_IN_PLACE was not used.  Thanks to Joe Ratterman @ IBM for the
bug report.

No reviewer.

[svn-r6680] added reduce_scatter_block version of redscat2.c test in preparation for optimized version of reduce_scatter_block that uses the same algorithms as reduce_scatter

We weren't actually running the new op_commutative test in the test
suite, so it wasn't even getting built on a "make testing".  As it turns
out, there were compilation errors.

Thanks to Joe Ratterman for catching this.

No reviewer.

[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.

This fixes the cases in MPI_Alltoallv/MPI_Alltoallw where the send
counts are 0 with a non-zero-sized send type and the recv counts are
non-zero with a zero-sized recv type.  Prior to this change MPICH2 would
hang when this situation was encountered.  This commit also includes a
regression test for this behavior.

Thanks to Michael Hofmann for reporting the bug and providing a succinct
test case that illustrates the problem.

No reviewer.

This commit implements MPI Forum ticket #27, which adds a new collective
function named MPI_Reduce_scatter_block.  This is equivalent to
MPI_Reduce_scatter with a consistent block size argument.

NOTE: this particular implementation of MPI_Reduce_scatter block is
inefficient.  It just performs an MPI_Reduce into a temporary buffer on
rank 0 and then scatters that buffer to the rest of the communicator.

For more history and information about this feature, please see the
following URL:

https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/27

No reviewer.

This corrects a bug that was introduced in r5249.  When the is_low_group
field isn't set deadlocks will typically occur for "full-duplex"
collectives over intercommunicators (such as MPI_Allreduce).

This commit also runs each of the intercomm collective tests with a
higher process count to ensure that more intercommunicator code paths
are exercised.  In this particular case running at higher process counts
is an effective regression test.

No reviewer.

[svn-r3988] Reduce the number of processes used and the maximum message size for the allgatherv4 test.