1
1

617 Коммитов

Автор SHA1 Сообщение Дата
Sergey Oblomov
1f9fce8955 SPML/UCX: fixed comment
Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit 01dacaa6a42b35c1b7538d8ff0036bded913c847)
2019-08-22 11:42:03 +03:00
Sergey Oblomov
66e18563bf SPML/UCX: fixed hang in SHMEM_FINALIZE
- used MPI _Barrier to synchronize processes

Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit 182023febb6f8f31ce34dc54c8aa409ad7e44fa2)
2019-08-22 11:41:52 +03:00
Sergey Oblomov
2fa112c0a6 UCX: added PPN hint for UCX context
- added PPN hint for UCX context init

Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit 43186e494b47ca29e8d5e7a864b6b98b8e873195)

Conflicts:
	opal/mca/common/ucx/common_ucx_wpool.c
2019-08-09 11:51:30 +03:00
Geoff Paulsen
105bfed176
Merge pull request #6741 from open-mpi/smiller_shmem_wait_types_v4.0.x
v4.0.x: shmem/c: Fix shmem type for calls to shmem_test and shmem_wait_until …
2019-06-07 14:06:28 -05:00
Howard Pritchard
a42977f1c2
Merge pull request #6707 from hoopoepg/topic/alloc-with-hint-realloc-inplace-v4.0
ALLOC_WITH_HINT: added inplace realloc - v4.0
2019-06-06 10:16:57 -07:00
Geoff Paulsen
bd602cc3a0
Merge pull request #6701 from hoopoepg/topic/sshmem-mpi-coll-collect-v4.0
SSHMEM/COLL: added sshmem/mpi implementation for shmem_collect call - v4.0
2019-06-05 13:02:43 -05:00
Scott Miller
e6e09c6cba shmem/c: Fix shmem type for calls to shmem_test and shmem_wait_until with [u]int32_t and [u]int64_t
Signed-off-by: Scott Miller <scott.miller1@ibm.com>
(cherry picked from commit ca59cabc679ebdf1decdcf75f3da0766b35a34f7)
2019-06-05 13:42:39 -04:00
Sergey Oblomov
69923e78c7 SPML/UCX: added synchronized flush on quiet
- added synchronized flush operation on quiet call.
- flush is implemented using get operation

Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit 0b108411f89727a68cd622f3b04c783efa359b8e)
2019-05-30 18:08:33 +03:00
Sergey Oblomov
456c5b90ae OSHMEM: minor optimization of realloc in shadow allocator
Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit d6a09120244be36d870e791146b5baed93659754)
2019-05-27 11:44:32 +03:00
Sergey Oblomov
748a5f5e73 SHADOW ALLOCATOR: minor code optimization
Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit a51badd627c5cdd3212cd6bedd3daa236cd6c8db)
2019-05-27 11:44:26 +03:00
Sergey Oblomov
f75d46faa9 ALLOC_WITH_HINT: added implace realloc
- in some cases realloc operation may be completed without
  allocation of new buffer (and without additional data copy)
- added logic to reallocate buffer inplace if possible

Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit 277c2a9e5c7711098be826e6c154253747fdad9a)
2019-05-27 11:44:18 +03:00
Sergey Oblomov
c142605566 SSHMEM/COLL: added sshmem/mpi implementation for shmem_collect call
- added MPI based implementation of shmem_collect call

Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit 7d8cb75b2e344f867021cffd4e656ff08d3945d8)
2019-05-23 15:34:12 +03:00
Geoff Paulsen
c22326e59a
Merge pull request #6652 from yosefe/topic/alloc-with-hint-impl-master-v4.0.x
OSHMEM: Add support for shmemx_malloc_with_hint() - v4.0.x
2019-05-17 15:48:35 -05:00
Yossi Itigin
fbd6798bf8 OSHMEM/MMAP/SYSV: Return ERR_NOT_IMPLEMENTED if segment hint != 0
(picked from master f708674)

Signed-off-by: Yossi Itigin <yosefe@mellanox.com>
2019-05-15 17:11:24 +03:00
Mikhail Brinskii
ff9ecc183f SPML/UCX: Fix coverity error
Signed-off-by: Mikhail Brinskii <mikhailb@mellanox.com>
(cherry picked from commit d81dc533f6f7ebba0b00c1652190975f0aca9e06)
2019-05-15 14:20:05 +03:00
Sergey Oblomov
e6cb5b02e8 OSHMEM/free: suppressed coverity issue
- removed dead code

(cherry picked from master 4df8c1b)

Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
Signed-off-by: Yossi Itigin <yosefe@mellanox.com>
2019-05-13 16:38:57 +03:00
Yossi Itigin
fc41c16134 OSHMEM: Add support for shmemx_malloc_with_hint()
- added multiple segments processing
- added shmemx_malloc_with_hint call + set of hints

(picked from master 94b5e91)

Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
Signed-off-by: Yossi Itigin <yosefe@mellanox.com>
2019-05-12 11:42:59 +03:00
Mikhail Brinskii
6861a68de6 SPML/UCS: CR comments p2
Signed-off-by: Mikhail Brinskii <mikhailb@mellanox.com>
(cherry picked from commit d4843b165172a9cffbdc967ec92ec1d60d053903)
2019-05-02 21:27:15 +03:00
Mikhail Brinskii
1c56f49a44 SPML/UCX: CR comments p1
Signed-off-by: Mikhail Brinskii <mikhailb@mellanox.com>
(cherry picked from commit c4c99457db6577f85b5b1c4c6849b5a04c226a84)
2019-05-02 21:26:55 +03:00
Mikhail Brinskii
e4ee56d1f3 SPML/UCX: Add shmemx_alltoall_global_nb routine to shmemx.h
The new routine transfers the data asynchronously from the source PE to all
PEs in the OpenSHMEM job. The routine returns immediately. The source and
target buffers are reusable only after the completion of the routine.
After the data is transferred to the target buffers, the counter object
is updated atomically. The counter object can be read either using atomic
operations such as shmem_atomic_fetch or can use point-to-point synchronization
routines such as shmem_wait_until and shmem_test.

Signed-off-by: Mikhail Brinskii <mikhailb@mellanox.com>
(cherry picked from commit 2ef5bd8b3671f1e10caf00d06d66d120eac9c5be)
2019-05-02 21:25:59 +03:00
Valentin Petrov
281f78c6e4 Fixes the O(N^2) loop in the mca_scoll_mpi_comm_query
The new proc group is created from the "world_group" based on the
      ranks mapping which can be directly taken from proc_name->vpid.

Signed-off-by: Valentin Petrov <valentinp@mellanox.com>
2019-04-15 08:43:09 +03:00
Geoff Paulsen
b3492090c2
Merge pull request #6561 from benmenadue/fix-scoll-fca
v4.0.x: scoll/fca: add missing argument to call to original broadcast
2019-04-08 14:05:47 -05:00
Ben Menadue
173192a6f4 Add missing #include to oshmem/shmem/c/shmem_context.c.
Signed-off-by: Ben Menadue <ben.menadue@nci.org.au>
(cherry picked from commit 063596b82837cf0b07926e2c696cd8f48a59143d)
2019-04-03 16:02:58 +11:00
Ben Menadue
001fa5b6ce Add missing nlong_type parameter to call to original broadcast in scoll/fca broadcast.
Signed-off-by: Ben Menadue <ben.menadue@nci.org.au>
2019-04-03 14:01:41 +11:00
Howard Pritchard
976cc1e07f
Merge pull request #6509 from janjust/oshmem-multiple-contexts-v4.0.x
v4.0.x: Oshmem multiple contexts
2019-04-01 13:15:38 -06:00
Scott Miller
5f4f5d45b3 shmem/fortran: Fix invalid datatype size in call to atomic cswap
Signed-off-by: Scott Miller <scott.miller1@ibm.com>
(cherry picked from commit 6b294e064150d26dfc68ec307cf4cd2e40891a1b)
2019-03-25 12:38:04 -04:00
Xin Zhao
69a80fce9f ompi/oshmem/spml/ucx: use lockfree array to optimize spml_ucx_progress/delete oshmem_barrier in shmem_ctx_destroy
ompi/oshmem/spml/ucx: optimize spml ucx progress

Signed-off-by: Tomislav Janjusic <tomislavj@mellanox.com>
(cherry picked from commit 9c3d00b144641d2929f830279dcc9d163c38e9e1)
2019-03-21 23:59:58 +02:00
Xin Zhao
580b584179 ompi/oshmem/spml/ucx:delete oob path of getting rkeys in spml ucx
Signed-off-by: Tomislav Janjusic <tomislavj@mellanox.com>
(cherry picked from commit e0414006b0c0a8e9918a4cf8ac4bb819b977ec91)
2019-03-21 23:59:46 +02:00
Xin Zhao
596997c194 ompi/oshmem/spml/ucx: defer clean up shmem_ctx to shmem_finalize
Signed-off-by: Tomislav Janjusic <tomislavj@mellanox.com>
(cherry picked from commit e1c1ab020227fc18d145379ab29ea86a3cdb66b1)
2019-03-21 23:58:23 +02:00
Xin Zhao
ce54b63b90 ompi/oshmem: add spml_context back to sshmem_type in memheap, to keep track of ucx_ctx_default's rkeys
Signed-off-by: Tomislav Janjusic <tomislavj@mellanox.com>
(cherry picked from commit 48033ac1f43159c053241b65e74a39777e5e31e4)
2019-03-20 23:30:21 +02:00
Xin Zhao
06183a7bec ompi/oshmem/spml/ucx: let shmem_finalize to clean up any ctx left
Signed-off-by: Tomislav Janjusic <tomislavj@mellanox.com>
(cherry picked from commit 9a060009622e9220d6332d9b63f6a1a7328418a0)
2019-03-20 23:30:09 +02:00
Xin Zhao
91793484ed OMPI/OSHMEM: bug-fix: store mkeys for each oshmem ctx.
Signed-off-by: Xin Zhao <xinz@mellanox.com>
Signed-off-by: Tomislav Janjusic <tomislavj@mellanox.com>
(cherry picked from commit 289595e45dc3ebfe5ae1a9dc6f347b1b2d569c4a)
2019-03-20 23:29:53 +02:00
Xin Zhao
f666d75322 ompi/oshmem/spml/ucx: fix eps destroy in shmem_ctx_destroy().
Signed-off-by: Tomislav Janjusic <tomislavj@mellanox.com>
(cherry picked from commit 79ba7526677bd1641239bb77559a2999c8cd3a4a)
2019-03-20 23:29:38 +02:00
Sergey Oblomov
14c271f993 PML/SPML/UCX: added evaluation of mmap events
- there was a set of UCX related issues reported which caused
  by mmap API hooks conflicts. We added diagnostic of such
  problems to simplify bug-resolving pipeline

Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit d8e3562bae700d84873c1d5ca9c45c846d7387ed)
2019-03-14 16:48:25 +02:00
Gilles Gouaillardet
9c5d4bb3b9 oshmem: fix macro usage in pshmem.h
pshmem.h now includes shmem.h (since open-mpi/ompi@f46130cd20) and some macros were removed at that time.

Use the OSHMEM_HAVE_C11 macro (defined in shmem.h) instead of the
previous OSHMEMP_HAVE_C11 macrso previously defined in pshmem.h

Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
(cherry picked from commit 5ea939aa542605503b7926a44df10e23a344d063)
2019-03-06 00:14:19 +01:00
Bert Wesarg
d174e46dc1 OSHMEM: Let pshmem.h include shmem.h to be stand-alone again
See #6093

Signed-off-by: Bert Wesarg <bert.wesarg@tu-dresden.de>
(cherry picked from commit f46130cd20914a3c60b19c3b2ed7c4b380319ee9)
2019-03-06 00:13:34 +01:00
Sergey Oblomov
ef7ab9bf96 OSHMEM: added missing API for get/put operations
- added calls for datatypes int/uint/8/16/32/size/ptrdiff
  for shmem_g/get/iget/get_nbi/_p/put/iput/put_nbi

Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit cfa9150934481ff19cbd9f2454da6d40a290f08a)
2019-01-09 11:52:26 +02:00
Yossi Itigin
ad4b33336d oshmem/scoll: fix shmem_collect32/64 for zero-size length
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>
2019-01-02 12:15:01 +02:00
Howard Pritchard
552834b9a6
Merge pull request #6110 from hoopoepg/topic/amo-int32-64-datatypes-v4.0
OSHMEM/AMO: added int/uint/32/64 atomics calls - v4.0
2018-12-11 06:43:05 -07:00
Sergey Oblomov
5838760a3a OSHMEM/COLL/BCAST: removed unnecessary bcast call
- removed unnecessary bcast call on zero-length request

Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit c93927e27a8e4241236d82c0d34ea445aa619aff)
2018-11-27 14:26:56 +02:00
Sergey Oblomov
0a064d8c8d OSHMEM/COLL: optimization on zero-length ops
- removed barrier call on zero-length operations

Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit ff2fd0679eb4b31bfd840395d73746612e2670f4)
2018-11-27 14:26:52 +02:00
Sergey Oblomov
dea9cf6b63 OSHMEM: added processing of zero-length collectives
- 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)
2018-11-27 14:26:44 +02:00
Sergey Oblomov
63cbe36cab OSHMEM/AMO: added int/uint/32/64 atomics calls
- added int/uint/32/64 atomics calls
- added SHMEM_SYNC_SIZE macro

Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit 4c071da565b152dbceb8f48368327bd0ae218942)
2018-11-26 09:29:55 +02:00
Geoff Paulsen
6898de01a7
Merge pull request #6012 from hoopoepg/topic/added-missing-amo-datatypes-v4.0
OSHMEM/AMO: added missing C11 macro datatypes - v4.0
2018-11-19 14:20:15 -06:00
Howard Pritchard
3369b0d10f
Merge pull request #6011 from hoopoepg/topic/fixed-oshmem-profile-build-v4.0
OSHMEM/PROFILE: fixed profile build - v4.0
2018-11-19 13:15:09 -07:00
Sergey Oblomov
b416c8afe2 OSHMEM/AMO: code beautify
- added <cr> to split API groups to simplify human processing

Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit 6e7810208966d73e0a56f74b536aa5c56b9a8d1c)
2018-11-01 16:52:01 +02:00
Sergey Oblomov
d3f08d010c OSHMEM/AMO: added missing C11 macro datatypes
- added signed datatypes for atomic_add calls
- added unsigned datatypes for atomic put/inc/get/fetch calls
- fixed incorrect SHMEM_CTX_DEFAULT macro, added
  external declaration of oshmem_ctx_default variable

Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com>
(cherry picked from commit f63d6da6d733e263385188b5470df195f092d041)
2018-11-01 16:51:56 +02:00
Sergey Oblomov
aaf15a6c17 OSHMEM/PROFILE: fixed profile build
- 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)
2018-11-01 16:37:49 +02:00
Yossi Itigin
8a329a797c SCOLL/BASIC: Fix invalid pSync pointer passed to barrier func
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>
2018-10-31 12:22:19 +02:00
Yossi Itigin
cda310733f shmem/lock: progress communications while waiting for shmem_lock
(cherry picked from commit 4101150)

Signed-off-by: Yossi Itigin <yosefe@mellanox.com>
2018-09-27 11:42:34 +03:00