If UCX is available, then pml/ucx will be used instead of
pml/ob1 + btl/openib, so there is no need to warn about
btl/openib not supporting Infiniband.

Signed-off-by: Gilles Gouaillardet <[email protected]>

(cherry picked from commit open-mpi/ompi@0a2ce58)

…led.

Fixes an issue introduced in open-mpi/ompi@0a2ce58

This is a one-off commit for the v4.0.x branch since btl/openib has been removed from master.

Refs. open-mpi#6137

Signed-off-by: Gilles Gouaillardet <[email protected]>

Many thanks to Sergey Oblomov for reporting this issue
and the countless traces provided when troubleshooting it.

This is a one-off commit for the v4.0.x branch since btl/openib has been removed
 from master.

Refs. open-mpi#6137

Signed-off-by: Gilles Gouaillardet <[email protected]>

In case of using a btl_put in ob1, the handle of the locally registered
memory is sent with a PUT control message. In the current master code
the sent handle is necessary the handle in the frag but if the handle
has been successfully registered in the request, the frag structure does
not have any valid handle and all fragments use the request one.

I suggest to check if the handle in the fragment is valid and if not to
send the handle from the request.

Signed-off-by: Brelle Emmanuel <[email protected]>
(cherry picked from commit e630046)

The rdma_frag attached to the send request was not correctly released
upon request completion, leaking until MPI_Finalize. A quick solution
would have been to add RDMA_FRAG_RETURN at different locations on the
send request completion, but it would have unnecessarily made the
sendreq completion path more complex. Instead, I added the length to
the RDMA fragment so that it can be completed during the remote ack.

Be more explicit on the comment.

The rdma_frag can only be freed once when the peer forced a protocol
change (from RDMA GET to send/recv). Otherwise the fragment will be
returned once all data pertaining to it has been trasnferred.

NOTE: Had to add a typedef for "opal_atomic_size_t" from master into
opal/threads/thread_usage.h into this cherry pick (it is in
opal/include/opal_stdatomic.h on master, but that file does not exist
here on the v4.0.x branch).

Signed-off-by: George Bosilca <[email protected]>
(cherry picked from commit a16cf0e)
Signed-off-by: Jeff Squyres <[email protected]>

…user

    If user sets HCOLL_EXTERNAL_UCM_EVENTS=1 then we try init opal
    memory framework and register a mem release cb. Otherwise, rely on ucx.

Signed-off-by: Valentin Petrov <[email protected]>

Signed-off-by: Nathan Hjelm <[email protected]>
(cherry picked from commit 3e1dd36)

- added MPI based implementation of shmem_collect call

Signed-off-by: Sergey Oblomov <[email protected]>
(cherry picked from commit 7d8cb75)

- 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 <[email protected]>
(cherry picked from commit 277c2a9)

Signed-off-by: Sergey Oblomov <[email protected]>
(cherry picked from commit a51badd)

Signed-off-by: Sergey Oblomov <[email protected]>
(cherry picked from commit d6a0912)

- added synchronized flush operation on quiet call.
- flush is implemented using get operation

Signed-off-by: Sergey Oblomov <[email protected]>
(cherry picked from commit 0b10841)

This is mostly based off recent UCX additions to their patcher:
    openucx/ucx#2703

They added triggers for
* mmap when (flags & MAP_FIXED) && (addr != NULL)
* shmat when (shmflg & SHM_REMAP) && (shmaddr != NULL)

Beyond that I noticed they already had a trigger for
* madvise when (advice == MADV_FREE)
that we didn't so I added that.

And the other main thing is we didn't really have shmat/shmdt
active for some systems because we only had a path for
syscall(SYS_shmdt, ) but we needed to also have a path for
syscall(SYS_ipc, IPCOP_shmdt, ) and same for shmat.

Signed-off-by: Mark Allen <[email protected]>
(cherry picked from commit eb88811)

Signed-off-by: Yong Qin <[email protected]>

…et-v4.0

SPML/UCX: added synchronized flush on quiet - v4.0

Use the PVAR ctx to save the SPC index, so that no lookup nor
restriction on the SPC vars position is imposed.
Make sure the PVAR are always registered.

Signed-off-by: George Bosilca <[email protected]>

Signed-off-by: George Bosilca <[email protected]>

Signed-off-by: George Bosilca <[email protected]>

…rom_6668

btl/uct: check for support before disabling UCX memory hooks

shmat/shmdt additions for patcher

V4.0.x Coll/hcoll: don't init opal memhooks unless explicitely requested

…warning

btl/openib: delay UCX warning to add_procs()

Remove the debruijn component as it changes the daemon's parent
process ID, thus breaking the other routed components

Signed-off-by: Ralph Castain <[email protected]>

Signed-off-by: Ralph Castain <[email protected]>

…with [u]int32_t and [u]int64_t

Signed-off-by: Scott Miller <[email protected]>
(cherry picked from commit ca59cab)

…collect-v4.0

SSHMEM/COLL: added sshmem/mpi implementation for shmem_collect call - v4.0

use strtol() instead of atoi() in order to handle hostnames
containing a large number.

This is a one-off commit for the release branches since
the regx framework has already been removed from master.

Refs. open-mpi#6729

Signed-off-by: perrynzhou <[email protected]>

Updating NEWS for v4.0.2

Refs openpmix/openpmix#1413

Signed-off-by: Ralph Castain <[email protected]>

    open-mpi/ompi@0fe756d Introduced
    a bug in coll/hcoll component. The ompi_requests allocated by
    libhcoll would be treated as coll_base_nbc_request during
    ompi_coll_base_retain_<> call. Afterwards this would lead to a
    segv in the request cleanup.

    Fix: since libhcoll interface does not distinguish between the
    blocling/non-blocking requests use coll_base_nbc_request all the
    time and initialize it properly in
    coll/hcoll/get_coll_handle(). It is still within 2 cache lines.

Signed-off-by: Valentin Petrov <[email protected]>

These variables were renamed in
904276b; update them to use the new
names.

Signed-off-by: Jeff Squyres <[email protected]>
(cherry picked from commit 2ab8109)

…request_bugfix

V4.0.x Coll/hcoll: fixes hcoll non-blocking colls support

Signed-off-by: Joseph Schuchart <[email protected]>
(cherry picked from commit 08cb638)

…able-names-in-make-check

v4.0.x: Update OPAL DDT variable names

UCX osc: properly release exclusive lock to avoid lockup (v4.0.x)

 * Fix open-mpi#6618
   - See comments on Issue open-mpi#6618 for finer details.
 * The `plm/rsh` component uses the highest priority `routed` component
   to construct the launch tree. The remote orted's will activate all
   available `routed` components when updating routes. This allows the
   opportunity for the parent vpid on the remote `orted` to not match
   that which was expected in the tree launch. The result is that the
   remote orted tries to contact their parent with the wrong contact
   information and orted wireup will fail.
 * This fix forces the orteds to use the same `routed` component as
   the HNP used when contructing the tree, if tree launch is enabled.

Signed-off-by: Joshua Hursey <[email protected]>

This commit fixes issue open-mpi#6853 by removing
MacOS/Darwin-specific logic from intercept_mmap.

Signed-off-by: Harumi Kuno <[email protected]>

v4.0.x: regx/naive: add regx/naive component

v4.0.x: Fix mmap infinite recurse in memory patcher

Remove unnecessary error log

Signed-off-by: guserav <[email protected]>
(cherry picked from commit 3c9f4e6)

…it-atomics

v4.0.x: Fix osc sm posts when only 32 bit atomics support v4.0.x

This patch fixes the merge of contiguous elements into larger but more
compact datatypes, and allows for contiguous elements to have thir
blocklen increasing instead of the count. The idea is to always maximize
the blocklen, aka. the contiguous part of the datatype.

Signed-off-by: George Bosilca <[email protected]>
(cherry picked from commit 41e6f55)

elements that can be merged into a larger UINT1 type.

Signed-off-by: George Bosilca <[email protected]>
(cherry picked from commit 82d6322)

If both types of interfaces are enabled, don't error out if one of them
isn't able to open listener sockets. Only one interface family may be
available on some machines, but someone might want to build the code to
run more generally.

Refs openpmix/prrte#249

Signed-off-by: Ralph Castain <[email protected]>
(cherry picked from commit 06d188e)

v4.0.x: Be a little less restrictive on interface requirements

v4.0.x: Datatype optimization and fix

Fix tree spawn routed component issue

Reving VERSION to v4.0.2rc2

Signed-off-by: Geoffrey Paulsen <[email protected]>

Reving VERSION to v4.0.2rc2

See https://github.com/ParaToolsInc/ompi.git

There was a restructuring of the comm/win/group objects
that shifted the fields down for getting at the class name.
This appears to relate to opal_infosubscriber_t changes from 50aa143.

NOTE: The opal_infosubscriber_t changes were not applied to the
oshmem_group_t structure, so no need for changes there.

Guard these using existing OPAL_ENABLE_MEM_PROFILE option
that is enabled via '--enable-mem-profile' configury.