GitLab

This fixes the performance regression that was introduced by concatenation of
per-target lists.

Reviewer: goodell

Moved fence_issued and start_assert MPID_Win members into CH3.  These should
not be a required part of the ADI, since they are not needed above the ADI and
implementations should be free to choose different mechanisms for tracking
the state of synchronization operations.

Reviewer: buntinas

The fence_cnt field in MPID_Win is not a counter, it's a flag that indicates if
fence has been called.

Reviewer: buntinas

Added tests for improper RMA synchronization.

reviewer: balaji

The implementation of lock/unlock_all was updated with the addition of epoch
tracking.  This patch adds a corresponding update of the memory fences for
shared memory windows.

Reviewer: balaji

This patch adds code to track the RMA epoch state of the local process.
Currently, we are tracking the synchronization states that are allowed by
MPICH; in the future, we may want to restrict this to only states that are
allowed by the standard.  The addition of epoch tracking has several benefits:

 * It allows us to detect synchronization errors (implemented in this patch).
 * It allows us to implement lock_all more efficiently (implemented in this
   patch).
 * It will allow us to distinguish between active and passive target epochs and
   avoid O(p) op list concatenation (future patch).

Reviewer: balaji

This change will set us up to acquire the local lock inside of other
synchronization calls, specifically Win_lock_all.

Reviewer: buntinas

Reviewer: goodell

I thought I fixed it before, but I misunderstood the script and didn't
check the output before committing last time.

No reviewer.

programs.

No reviewer.

The dist_graph implementation itself doesn't need any changes, since we
already switched away from an impl where MPI_UNWEIGHTED==NULL.

Reviewed by dinan@.

Don't litter a ".tmp" file in case of failure.  Notice changes to
the "maint/errmsgdirs" file.

Reviewed by dinan@.

And add "src/mpi_t" to the list of special directories that is
searched for "errnames.txt" files.

Reviewed by dinan@.

Reviewed by dinan@.

Reviewed by dinan@.

Should simplify numerous datatype handle conversions that we currently
do with `expr` and manual hex-->dec conversion in configure.ac.

Reviewed by dinan@.

Forgotten from the previous MPI_Count-related commits (oops).

Reviewed by dinan@.

Reviewed by buntinas.

Reviewed by buntinas.

This is a very lame implementation that creates this object at Init
time and does not populate it with any keys (the MPI standard does not
require that the keys be populated).

Reviewed by buntinas.

Reviewed by buntinas.

arguments are defined.  They are entirely ignored.

Reviewed by buntinas.

Reviewed by buntinas.

is completely ignored.

Reviewed by buntinas.

There are other instances of this warning that definitely need to get
fixed, but this one is especially annoying because it is in a header
and therefore causes a warning in nearly all nemesis compilation
units.

Reviewed by balaji@.

Any code that causes this new warning to trigger is likely to be a bug,
or at least a case that ought to be checked/asserted and then explicitly
casted.  This warning is essential to tracking down problems related to
the new MPI_Count datatype/interfaces added in MPI-3.

Reviewed by balaji@.

This macro checks for positive overflow due to multiplication and
asserts if the product would exceed the given limit.  Useful for
guarding older, sloppier code from before the MPI_Count transition.

Reviewed by balaji@.

Attempts to validate that we have successfully implemented MPI-3's new
"large count" routines and updated behavior for "legacy" count routines:

MPI_Get_count
MPI_Get_elements_x
MPI_Status_set_elements_x
MPI_Type_get_extent_x
MPI_Type_get_true_extent_x
MPI_Type_size_x

Because we only use contigs and vectors in this test, we are actually
missing quite a bit of coverage in terms of checking for MPI_Count-clean
size/extent calculations.  At some point we should strengthen this test
with a wider variety of tests using more complex datatypes.

Reviewed by balaji@.

Also update MPI_Type_get_true_extent to have proper
MPI_UNDEFINED-clamping per MPI-3.

Reviewed by balaji@.

Also update MPI_Type_get_extent to have proper MPI_UNDEFINED-clamping
per MPI-3.

Reviewed by balaji@.

Also update MPI_Status_set_elements to have proper MPI_UNDEFINED-clamping per
MPI-3.

Reviewed by balaji@.

Also update MPI_Get_elements to have proper MPI_UNDEFINED-clamping
per MPI-3.

Reviewed by balaji@.

Also update MPI_Type_size to have proper MPI_UNDEFINED-clamping per
MPI-3.

Reviewed by balaji@.

MPI_UNDEFINED is now returned in case the count cannot be represented
by an "int".  Work may still be needed for the case when
sizeof(int)!=sizeof(MPI_Fint).

Reviewed by balaji@.

This commit actually does a few things to the status object:

(1) `int count` becomes `MPI_Count count`;
(2) `count` & `cancelled` have been reordered after the "public"
fields (e.g., MPI_TAG) to ensure that direct access from Fortran is
not somehow broken by the (potentially) non-INTEGER count field;
(3) and a new future-proofing field, `abi_slush_fund`, has been added to
attempt to allow us to add fields to the status object in the future
without having to break ABI compatibility.

There's still too much duplication of the status object's contents/layout
in multiple pieces of the code, but I don't see a reasonable way to fix
this in the near term.

Warning: large count breaks the experimental Fortran binding code
generated for handling the "sizeof(int)!=sizeof(MPI_Fint)" case.

Reviewed by balaji@.

This commit is largely auto-generated boilerplate for the following
"large count" routines:

* MPI_Get_elements_x
* MPI_Type_get_extent_x
* MPI_Type_get_true_extent_x
* MPI_Type_size_x
* MPI_Status_set_elements_x

None of these routines actually _do_ anything (not even assert/error
out) as implemented.  But this makes it very easy to review the logic of
the implementation later on in the succeeding smaller commits.

Reviewed by balaji@.

This commit only contains the build system and mpi.h modifications
necessary for `MPI_Count` to be a valid new type.  It does *not*
actually use this new type anywhere, such as in MPI_Status.

Reviewed by balaji@.

Currently unused, but we should probably be using these in more places
for overflow checks.

Reviewed by balaji@.