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

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

Added immediate locking (needed to support load/store access) and memory fences
(to support third-party load/store communication).

Reviewer: balaji

1. Make the manpage entries consistent with the function prototype.

2. Consistently use "Parameters" instead of the inconsistent
"Parameter" and "Parameters" usage.

3. Correct the manpages for MPI_Type_get_contents and
MPI_Type_get_envelope.

Reviewed by dinan.

Updated RMA code to remove trailing "_e" and "_s" on enum and struct type
names to match the MPICH style.

Reviewer: goodell

Reviewer: balaji

Code review cleanups that couldn't be merged into earlier patches because of
later changes.

Reviewer: goodell

Updated get_accumulate implementation to properly handle NO_OP operations;
currently, we dispatch these as get operations.  As a part of this change, the
op issue code was also gathered into a macro to reduce copy-paste across all
of the synchronization functions.

Reviewer: goodell

Removed scattered initializations of the RMA op dataloop field that are
redundant, since the dataloop field is now initialized when the op is
allocated.

Reviewer: goodell

Fetch-and-op has been updated to ignore the origin data and perform no
operation when the given op is MPI_NO_OP.

Reviewer: goodell

Reviewer: goodell

Removed calls to higher-level MPIU_RMA_CALL() macro.  This macro dispatches
either from the RMAFns table or directly to the MPID_ function; since CH3
always uses the RMAFns table, this macro shouldn't be used inside of CH3.

Reviewer: balaji

If flush is used as a part of polling for incoming data, we can deadlock, since
local RMA calls never poke the progress engine.  Extra progress was added to
local flushes to better fake passive progress.

Reviewer: balaji

Reviewer: buntinas

Updated RMA implementation to track the passive target status individually, for
each target.  Includes new implementation for lock/unlock_all.  Lock_all is
currently unoptimized, see #1734 for future plans.

Reviewer: buntinas

Remove ugly rank 0 hack and updated the active target code to maintain and
process a proper RMA ops list.

Reviewer: goodell

The use of a dense array is a temporary measure to support the reference
implementation.  This will be much improved by ticket #1735.

Reviewer: goodell

A couple of the RMA synchronization error checks were returning the wrong error
class.

reviewer: buntinas

The old "lockRank" error checking is no longer sufficient in MPI 3.0 and must
be removed to add support for locking multiple targets.

Reviewer: balaji

Removed the use of lockRank in the passive target RMA code.  This was a hack to
start with, since lockRank belonged to the the front-end error checking and
should not have been used below the ADI.

Reviewer: buntinas

This is a temporary measure for the reference implementation that should be
improved by ticket #1733.

Reviewer: buntinas

In this patch, I have refactored the RMA ops list again to use the MPL UTList
doubly-linked list and to treat the list as a proper object.  This should set
us up to work with multiple lists, as we will soon have one list per target.
Doubly-linking the list is a big help in terms of maintainability (no more
prevNext pointers) and flexibility (better implementation of request-based
ops and other optimizations).

Reviewer: goodell

Request refs at the target were incorrectly set to two.  Only one ref is held,
by the progress engine.

Reviewer: none

Reviewer: buntinas

Reviewer: balaji

Implementation of MPI-3 window info query routines.

Reviewer: balaji

Info argument was never used during window initialization.  Move the
responsibility to look at the info argument to the window creation function.

Reviewer: balaji

This adds a simple implementation for request-based RMA operations.  The
implementation uses generalized requests as follows:

 1. Insert operation into the RMA ops list
 2. Create a generalized request
 3. When the request is polled, flush the window and mark completed

Reviewer: buntinas