Probably not too harmful to omit the guards, but better not to jumble
more constants together in the compiler's type-checking machinery.

No reviewer.

Obvious bug.

No reviewer.

It was possible to have a working `long double _Complex` even if the MPI
library was pretending that `long double` should not work.

Ditto for the C++ version.

No reviewer.

This helps Cray out (and conceivably other vendors), since they ship
several compilers which have different interpretations of what `long
double` should mean (80-bit vs. 128-bit, being most common).

Note that exporting `MPID_NO_LONG_DOUBLE=yes` in the environment before
running `configure` does work, but may not work as intended in the face
of an automake-initiated reconfigure.  Autoconf does not treat this
variable as precious (and we don't ask it to, since it's for
inside-configure-use only), so its value will not be preserved for the
reconfigure.  The same holds for any `MPID_` variables mentioned at the
top of `configure.ac`.

No reviewer.

Otherwise `oputil.h` barfs on a bad preprocessor comparison when
`MPID_NO_LONG_DOUBLE` is set.

Fix reported by and suggested by Steve Oyanagi from Cray.

No reviewer.

In particular, the clang type checking helpers can cause
`MPI_DATATYPE_NULL` to be used before it has been `#define`d.

No reviewer.

Added a new test to check that 0-byte transfers are handled correctly.

Reviewer: goodell

Added a check for 0-byte transfers in request-generating operations.
When a 0-byte transfer is encountered, the user is returned a completed
handle and no communication is performed.

Reviewer: goodell

Extended the RMA PROC_NULL test to cover new operations and also to test
passive target synchronization.

Reviewer: goodell

Moved handling of MPI_PROC_NULL down to the device and cleaned up
handling of PROC_NULL in the CH3 device.

Reviewer: goodell

Updated request-generating RMA operations to correctly handle
communication with MPI_PROC_NULL.

Reviewer: goodell

Fix bug #884. The routine was returning wrong results for derived
datatypes. Two comparisons have been fixed in PMPI_LOCAL MPI_Count
MPIR_Type_get_elements, src/mpi/datatype/get_elements_x.c. Two new
test cases, get_elems_d and get_elems_u have been provided in f90.

Reviewer: goodell

MPI_Type_commit was missing CS ENTER/EXIT guards, which may be causing a
conflict with RMA code that creates derived datatypes at the target
process, when MPICH_ASYNC_PROGRESS is specified.  This may have fixed #1766.

Reviewer: goodell

This merge rounds out shared memory window support in CH3.  Note that CH3
currently supports shared memory windows on MPI_COMM_SELF (and dups).

Hoisted the window flavor error check in Win_shared_query from Nemesis
up to the public API where it belongs.

Reviewer: apenya

This adds a default shared memory window implementation for CH3 (used
by e.g. sock), which works only for MPI_COMM_SELF (this is what the
default comm_split_type provides).  This closes ticket #1666.

Reviewer: apenya

Moved the MPI_Comm_split_type override from CH3 down to Nemesis. CH3
should now use the default implementation, which returns MPI_COMM_SELF.

Reviewer: apenya

Added missing boilerplate declarations before
finalize_failed_procs_group.

Reviewer: apenya

This patch adds basic error descriptions for the new RMA error classes
that were introduced in MPI 3.0.

Reviewer: none

This merge adds a flags field to the packet header in RMA packets.  The flags
are used to piggyback RMA synchronization operations.

Reviewer: goodell

I initially added a conservative flush message for empty epochs (mostly
for documentation purposes).  This is not needed in the current
implementation, since ops are not issued eagerly.  If/when eager ops are
implemented, this patch should be reverted and additional window state
tracking for this case should be added.  In the meantime, I am removing
this code to improve performance.

Reviewer: goodell

This patch adds a few missing memory fences to the window
synchronization operations for shared memory windows.  This closes
ticket #1729.

Reviewer: goodell

Moved RMA errors used in the CH3 RMA implementation into the ch3
errnames.txt file.

Reviewer: goodell

Additional testing for shared lock with MODE_NOCHECK.

Reviewer: goodell

When the MPI_MODE_NOCHECK assertion is given to a passive target lock
operation, we defer acquisition of the lock and piggyback the request on
the first RMA op to the target.  This eliminates a round-trip
lock-request message.

Reviewer: goodell

This patch adds piggybacking of flush synchronization on top of the last
operation in an RMA epoch.

Reviewer: goodell

The single_op_opt flag in the request object was previously used to
track whether an operation is a lock-op-unlock type, for the purposes of
completion.  Tracking this state has been merged into the packet header
flags, so the single_op_opt flag is no longer needed.

Reviewer: goodell

Removed source_win_handle from the packet header, since it's no longer
needed.

Reviewer: goodell

We can simplify completion of get operations at the target, since
Finish_rma_op should now be called unconditionally.  Flags are now used
to determing whether an acknowledgement should be sent, so the
source_window_handle check is removed.

Reviewer: goodell

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