- 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 9de128afaf5224193a88f00c897f5d5c94336f99)
prefer #include vs include in order to correctly handle long Fortran lines.
We use the full path, and it can be very long, this is why
it cannot be passed to the Fortran compiler.
Thanks Igor Andriyash and Axel Huebl for reporting this issue.
Refs open-mpi/ompi#6106
Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
(cherry picked from commit 475d0355d7ce2d31b582944c685fae13d711b3c9)
without this fix, an error handler invoked on pml_ucx request would
segfault while trying to dereference requests[i]->req_mpi_object.comm
(picked from master f36eeef)
Signed-off-by: Yossi Itigin <yosefe@mellanox.com>
The feature of persistent collectives is approved in the Sept. 2018
MPI Forum meeting and 2018 Draft Specification of the MPI standard is
published during SC18.
Signed-off-by: KAWASHIMA Takahiro <t-kawashima@jp.fujitsu.com>
(cherry picked from commit 5f0fcf0f456aaa43f36215859412f33f79ac91f9)
- added UCX version evaluation to set OSC UCX priority
Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit e91f214982391b8e1b26be39147c357d32b8380e)
- OSC/UCX module set priority to 200 to be used by default
Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit 36934a8bb2484c3d27d14683d65012ff422334f4)
* Needed to properly read PMIx job data like the following
- `OPAL_PMIX_LOCALLDR`
- `OPAL_PMIX_RANK`
- `OPAL_PMIX_GLOBAL_RANK`
- `OPAL_PMIX_APPLDR`
- `OPAL_PMIX_APP_RANK`
Signed-off-by: Joshua Hursey <jhursey@us.ibm.com>
(cherry picked from commit a557c4130c42a5a41aba5c08e606e7129d0bcb6d)
ompi/mpiext/cuda/c/mpiext_cuda_c.h is automatically generated from
ompi/mpiext/cuda/c/mpiext_cuda_c.h.in at configure time.
Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
(cherry picked from commit open-mpi/ompi@f8318f0a8f)
(cherry picked from commit open-mpi/ompi@b3ce25af95)
Also remove a now-outdated LSF reference.
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
(cherry picked from commit 419852ab433e578a790e8882711a20f2f570f0a2)
Follow on to 430c659908: clarify the help message and fix one typo.
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
(cherry picked from commit e9bf318dcb2f337267211f37e6d59c9f8bf5d8be)
Update the show_help message for when there are not enough slots to
run an application.
Also, remove a bunch of copies of this message in various show_help
text files that aren't used/referred to anywhere in the code.
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
(cherry picked from commit 430c659908f9c1ba1ff652379a694314718ff3d8)
Per feedback from https://github.com/open-mpi/ompi/pull/6028, remove
"+ob1" from the sentence to emphasize that it's only IB usage through
openib that is deprecated/superceded (i.e., ob1 is definitely not
deprecated).
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
(cherry picked from commit 6cb415982615c504b97748d8c4007a352edd7246)
Move the UCX and MXM text up to flow better with the rest of the
text+content. Also emphasize that MXM is deprecated.
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
(cherry picked from commit 4ec8e6fe2250f67b8a2213f0699a9ee18c3d1a91)
Several names are now no longer returned by MPI_Type_get_envelope.
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
(cherry picked from commit 65eb118e087b0bdaa9c92a12eba151eb30994590)
It seems in some cases (gcc older than v6.0.0) the __atomic_thread_fence is a
no-op with __ATOMIC_ACQUIRE. This appears to be the case with X86_64 so go
ahead and use __ATOMIC_SEQ_CST for the x86_64 read memory barrier. This should
not cause any performance issues as it is equivalent to the memory barrier
in the hand-written atomics.
References #6014
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
(cherry picked from commit 30119ee339eea086f43e3392352899187a4a73c7)
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
Commit 89da9651b inadvertantly #if'ed out both deprecated *and*
removed items from mpi.h. The intent was only to #if out items that
have been *removed* from the MPI specification and leave all items
that are merely deprecated.
This commit also re-orders the deleted typedef+functions to be in the
same order as they are listed in MPI-3.1 chapter 17, just to make
verifying/checking the code easier.
Note that --enable-mpi1-compatibility can still be used to restore
prototypes for the items that have been removed from the MPI
specification (e.g., MPI_Address()).
Signed-off-by: Geoffrey Paulsen <gpaulsen@us.ibm.com>
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
(cherry picked from commit b03a39d359b019d2d7803d194fd03b2fcdffddce)
Under certain circumstances, ibv_exp_query_device was
returning an error due to uninitialized fields in the
extended attributes struct.
Fixes: #5810Fixes: #5914
Signed-off-by: Howard Pritchard <howardp@lanl.gov>
(cherry picked from commit 8126779a354b3e0c720d3e1790f7b936dd5b93b2)
- added <cr> to split API groups to simplify human processing
Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit 6e7810208966d73e0a56f74b536aa5c56b9a8d1c)
- added missing file to profile makefile
- constants SHMEM_CTX_* are shifted into public header
Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit 4a3e83780c0303e7e4d0ff92d7ba85d3a2239737)
PR #5450 addresses MPI_IN_PLACE processing for basic collective algorithms.
But in conjunction with that, we need to check for MPI_IN_PLACE in tuned paths
as well before calling ompi_datatype_type_size() as otherwise we segfault.
MPI spec also stipulates to ignore sendcount and sendtype for Alltoall and
Allgatherv operations. So, extending the check to these algorithms as well.
Signed-off-by: Aravind Gopalakrishnan <Aravind.Gopalakrishnan@intel.com>
(cherry picked from commit 88d781056f43934a93e16db556b340e72cdd3742)
mca_scoll_basic_alltoall() passed (pSync + 1) to barrier function, but
the value of _SHMEM_ALLTOALL_SYNC_SIZE is 1, which made the barrier
function use an invalid memory location. In particular, this location
was not initialized to _SHMEM_SYNC_VALUE, which broke the barrier
algorithm and it did not complete: One PE could read 0 from its peer and
assume the peer already started the barrier, and then write 1 to the
peer. Then, the peer entered the barrier and overwrote the 1 with 0, and
then it waited forever to see '1' in its pSync.
Found with shmem_verifier test suite.
(picked from master 6754bf1)
Signed-off-by: Yossi Itigin <yosefe@mellanox.com>
