Fixes scoll_basic failures with shmem_verifier, caused by recent changes
in handling of zero-size collectives.
- Check for zero-size length only for fixed size collect (shmem_fcollect),
but not for variable-size collect (shmem_collect)
- Add 'nlong_type' parameter to internal broadcast function, to indicate
whether the 'nlong' parameter is valid on non-root PEs, since it's
used by shmem_collect algorithm. Before this change, some components
assumed it's true (scoll_mpi) while others assumed it's false
(scoll_basic).
- In scoll_basic, if nlong_type==false, do not exit if nlong==0, since
this parameter may not be the same on all PEs.
- In scoll_mpi, fallback to scoll_basic if nlong_type==false, since MPI
requires the 'count' argument of MPI_Bcast to be valid on all ranks.
(Picked from master 939162e)
Signed-off-by: Yossi Itigin <yosefe@mellanox.com>
- according spec 1.4, annex C shmem collectives should process
calls where number of elements is zero independently from pointer
value
- added zero-count processing - it just call barrier to
sync ranks
Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit 9de128afaf)
Deadlock when using the shmem_collect32()/shmem_collect64() routines and any of the non-root PEs pass 0 as the number of elements.
Algorithm in _algorithm_central_collector() does use 0 as a special value, and thus does not break out of the loop.
fixed by IgorI, reviewed by MikeD
cmr=v1.8.2:reviewer=ompi-rm1.8
This commit was SVN r31814.
1. fix in oshmem scoll component: basic algorithms should
call basic collectives since their implementation
incompatible with others (fca, hcoll).
2. Set OPAL_EVLOOP_ONCE flag ON for libevent in the case
of yoda smpl. Otherwise there is possible deadlock in
atomic_basic_lock call
fixed by Val, Igor, reviewed by Miked
cmr=v1.7.5:reviewer=ompi-rm1.7
This commit was SVN r30762.
