GitLab

This patch uses packet header flags to piggyback the unlock operation on other
RMA operations.  For most operations, there is no net change.  However, FOP and
GACC, unlock piggybacking was previously not implemented.

Reviewer: goodell

Added a flags field to MPID_Request that we can use to stash flags from
suspended RMA ops and retrieve them later when we complete the operation.

Reviewer: goodell

This change extends RMA op processing to pass around flags as needed.  It
doesn't yet utilize the flags.

Reviewer: goodell

Added flags field to RMA operation packets that are sent from origin to target.
This will be used to piggyback RMA synchronization operations.

Reviewer: goodell

Convert the packet type field from an enum to a uint16.  This change was also
applied to packet types defined by Nemesis.  Downstream netmod developers will
also have to make this change if they defined new packet types.

Reviewer: goodell

This patch consolidates the synchronization and tracking of RMA operations into
a single routine that is called whenever we complete an operation.  The only
exception are lock-op-unlock operations that are completed from within the lock
operation processing code.

This code is pretty ugly, but it will get cleaner once packet flags are been
added.

Reviewer: goodell

Partially reverted [0b364068] in preparation for incorporating new
piggybacking infrastructure.  This temporarily re-introduces that bug
and it will be fixed again with the new piggybacking patch.

Reviwer: goodell

Cleaned up a variety of compiler warnings relating to header files and
other such bugs.  Most of this patch was provided by Jeff Hammond.

Reviewer: none

This patch fixes several warnings when compiling the lockcontention3
test.  PUT_VAL was reduced because the old value overflowed signed
integer comparisons.

Reviewer: none

This function was missing thread safety CS_ENTER/EXIT locks.  The global
lock is necessary for this function because it accesses multiple parts
of the comm object, calls several internal routines that assume the
global lock is held, and upates the ref counts on the groups of each
communicator.

Reported by Steve Oyanagi @ Cray.  This closes ticket #1796.

Reviewer: goodell

Reviewer: none

No reviewer.

0e247915.

No reviewer.

libpci, which works around the GPL license of libpci.

No reviewer.

No reviewer.

No reviewer.

The linked list benchmarks were reportin the sum of the timings across
all nodes, rather than the maximum.

Reviewer: none

Caused problems for the "mpich-osx" jenkins build.

No reviewer.

Added support for passing the length of a Fortran CHARACTER as size_t instead of int.  Added a configure option to set; note that despite documentation to the contrary, it looks like some compilers still pass the length as an int (the common practice for decades), so changing to size_t must be made with extreme caution

Patterns that end in non-word chars (such as `:`) will not match a `\b`
atom when followed by a non-word like ` ` or `\n`.

No reviewer.

Caught by the cool new type checking annotations in our `mpi.h` header
that work with modern clang versions.

No reviewer.

`-Wempty-body` will cause this, and is enabled by default in modern
clang when configuring MPICH with `--enable-strict`.

No reviewer.

We were not communicating the MPI_WEIGHTS_EMPTY common block value to C
via the `mpirinitc_` routine.  Also fix a minor warning about the C struct
static initializer when `HAVE_C_MULTI_ATTR_ALIAS` is defined, as it
usually is on Linux.

The existing `dgraph_unwgtf` test does not catch this case.  AFAICS,
there is no black-box way to test this behavior using only the MPI
interface provided that the MPI library correctly implements
`MPI_UNWEIGHTED` handling, since the MPI implementation is forbidden to
inspect the pointer value other than to check whether it is
`MPI_UNWEIGHTED`.  The best idea I've come up with so far is to create a
sanity check routine that we can call from Fortran that ensures all
Fortran-C interoperability constants are correctly recognized, but even
that's a pretty weak test, since it must also be updated whenever a new
Fortran common block is added.  This will have to do for now...

Follow-up to [d6ceef97].

Reviewed-by: dinan

Handle the case where GNU coreutils are installed in some odd `bin`
directory, such as `gnubin`.

Also fix a bad regex pattern (`[^\w]` means match a single character
that is not `\` and is not `w`, rather than the opposite of `\w`, which
is `\W`) by using the `\b` "boundary" atom.

No reviewer.

Send out all lock requests before waiting for lock acquisition.  This
should improve the performance of lock-all operations on shared memory
windows.

Reviewer: goodell

This fix enables flush to ignore non-local targets that have not been targeted
by an RMA operation.  Prior to this fix, a call to flush-all would result in
acquiring the lock at all targets, which is extremely inefficient.

Reviewer: goodell

Prior to this patch, a lock entry was enqueued in the RMA ops list when
Win_lock was called.  This patch adds a new state tracking mechanism, which we
use to record the synchronization state with respect to each RMA target.  This
new mechanism absorbs tracking of lock operation and the lock state at the
target.  It significantly simplifies the RMA synchronization and ops list
processing.

Reviewer: goodell

references ticket #1754.

No reviewer.

This doesn't fix any of the serious bugs or inefficiencies present in
the current external32 implementation.  But it at least fixes some very
valid warnings related to const-ness and passing incorrect pointer
types.

References ticket #1754

Reviewed-by: robl

GACC operations were both piggybacking the unlock message to the origin, and
sending back a PT done packet.  This was causing the origin to be unlocked
twice.  When another lock operation was performed between the GACC and PT done
unlock operations, there was a synchronization race.

NTFS does not have the POSIX "file exists until everyone closes it"
property, so introduce the feature mechanism to NTFS and let ADIO know
NTFS cannot rely on that property.

This patch updates BSend debugging messages to be compatible with new size_t
arguments added in [7fa7e2cf].  Format strings are updated and size_t args are
cast to unsigned long long.  Upconverting to the largest unsinged integer type
is the only safe way to deal with this in pre-C99.  If we were coding to C99,
we could have avoided the type cast and used '%z' directly in the format
strings.  This closes ticket #1786.

Reviewer: goodell

ABI string is bumped from c:r:a of 10:1:0 to 10:2:1.  Though changes
were made to mpi.h.in, none of them affect "interfaces" in the libtool
sense.  They merely affect some clang compile-time checking.

No reviewer.