The numbering of MPI_COMBINER constants was changed when we added
MPI-2.2 support.  This patch reverts that numbering and adds the
MPI-2.2 specific combiners to the end, to maintain ABI compatibility
with older mpich releases.

An initial version of this patch was contributed by Intel.

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

This was changed to 1024 to give space for larger error strings.  But
this caused an ABI breakage for a number of MPICH partners.  This
patch reverts the size back the size to what we had in older mpich
releases.

An initial version of this patch was contributed by Intel.

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

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

Jan Bierbaum authored Oct 14, 2013 and Kenneth Raffenetti committed Oct 15, 2013

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

xfails are primarily for our consumption, so we disable them by
default. This will result in cleaner test output for the average
user.

Also note that if the testsuite is configured in strict MPI mode,
all strict tests, regardless of xfail, will run.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

MPIDU_Datatype_debug is really helpful when trying to figure out what a
library or someone else's datatype code is feeding to MPICH.  For some
reason it didn't know how to dump subarray types.  now it does.

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

clang seems to be unhappy when we use "0" instead of "NULL" for buffer
addresses.

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

bglockless feature file should not define anything pvfs2 related.

this guy no longer gets used now that there's common hint processing
code

I end up looking at this code only about once every six months.  Add
more comments so the next time I'm here, I don't have to re-learn the
teeny-tiny details.

drivers updated:
- ad_bgl
- ad_bg
- ad_lustre
- ad_panfs
- ad_pvfs2

Reviewed-by: Rajeev Thakur <thakur@mcs.anl.gov>

continue the slow and intermittent process of adding error checking to
romio test cases

ROMIO tests were not cleanly buildig with --enable-strict, mostly due to
missing 'const' for strings.   One test uses strdup, which is
incompatible with strict ansi.

for years we've added hints by cutting and pasting.  Turn all this
dupilcated code into common hint processing funcitons.   Still need to
do this for all drivers, too

All functions in these tests return errors to all processes when
there is a failure in the communicator. Mark failing tests as xfail.

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

pt2pt tests for blocking and non-blocking sends and recvs within
a communicator with a failed process. Mark failing tests as xfail.

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

Fault tolerance tests should test for program correctness even in
cases where mpiexec returns a non-zero exit code.

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

Jenkins counts a passing test with a TODO as an error.

No reviewer.

Signed-off-by: Rob Latham <robl@mcs.anl.gov>

Some builtin kinds (e.g., MPI_DOUBLE_INT), can have extra padding when
a vector of them is copied.  So we need to use the extent for such
types, not just the size.  In this commit, we just simplify the
optimization to only work on BUILTIN types, which do not have this
issue.

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

No reviewer.

Optimize the case where the origin and target both use basic
datatypes.  In this case, we assume that the data is aligned correctly
for the appropriate datatype and perform a direct assignment instead
of a memory copy.

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

i know the ft stuff is kind of broken but come on this didn't even
compile!

In this case, we don't need to find the lower-bound.  It's a simple
copy.

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

The check was originally in the ch3 layer, but doesn't seem to use any
ch3 specific information.  This macro will be useful at the upper
layers for optimizations, e.g., in the localcopy routine.

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

We already do a check for shared memory before calling the
shared-memory specific functions.  This patch simplifies some of those
redundant checks.

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

Xin Zhao authored Sep 26, 2013 and Pavan Balaji committed Sep 26, 2013

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

The memory barrier ensures that all load/store operations issued
directly to shared memory are complete.

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

Xin Zhao authored Sep 26, 2013 and Pavan Balaji committed Sep 26, 2013

Because when SHM is allocated, it is possible that orig rank and target rank
are on different nodes, in such situation operations are not done yet and
win_flush cannot exit.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Sep 26, 2013 and Pavan Balaji committed Sep 26, 2013

Check shm_allocated flag in win_flush to determine if do full
memory barrier or not.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Sep 26, 2013 and Pavan Balaji committed Sep 26, 2013

Check both shm_allocated flag and VC's node_id for
req-based operations.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

During a win_flush_all, if a target does not have any operations to
flush out, don't call the win_flush function at all.  This reduces the
number of function calls on large systems where the RMA operations are
sparsely issued.

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

We changed the default from FALSE to TRUE for the alloc_shm info flag
for MPI_WIN_ALLOCATE.

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

Xin Zhao authored Sep 26, 2013 and Pavan Balaji committed Sep 26, 2013

Flip the default shared memory allocation info argument for
MPI_WIN_ALLOCATE.  We were alredy doing this for
MPI_WIN_ALLOCATE_SHARED, but not for WIN_ALLOCATE.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

No reviewer.

Xin Zhao authored Sep 25, 2013 and Pavan Balaji committed Sep 26, 2013

If SHM is allocated by MPI_Win_allocate and target is on the same node
with origin, origin needs to acquire lock eagerly before it can
perform any SHM RMA operations immediately on target's SHM region.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Sep 25, 2013 and Pavan Balaji committed Sep 26, 2013

If SHM is allocated, perform RMA operations immediately after it is
issued, not queuing them up any more.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>