1
1
Граф коммитов

456 Коммитов

Автор SHA1 Сообщение Дата
Nathan Hjelm
fad0803920 osc/rdma: fix typo in atomic code
Fixes #3267

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-04-03 15:54:28 -06:00
Nathan Hjelm
c72fb30eb5 osc/pt2pt: fix typo
Signed-off-by: Nathan Hjelm <hjelmn@me.com>
2017-03-23 09:00:21 -06:00
Jeff Squyres
760db0d5ce osc/pt2pt: fix compiler warning
Remove unused variable.

Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
2017-03-16 05:46:11 -07:00
Valentin Petrov
fe069c9570 Fixes the coll_allgather usage bug
One should use the correct module object when calling
      c_coll.coll_allgather. Otherwise there will be a segfault in the
      case, for example, when hcoll is used. In that case
      c_coll.coll_allgather = mca_coll_hcoll_allgather while
      c_coll.coll_gather_module = tuned.

Signed-off-by: Valentin Petrov <valentinp@mellanox.com>
2017-03-14 09:47:39 +02:00
Nathan Hjelm
0195d15401 osc/pt2pt: flush pending fragments on lock ack
This commit addresses an issue that can occur in cases where a lot of
fragments are outstanding.

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-06 13:58:46 -07:00
Josh Hursey
b1c4e50500 Merge pull request #2934 from jjhursey/topic/coll-comm-restructure
Move coll structure outside of the communicator
2017-02-28 08:45:18 -06:00
Nathan Hjelm
032bcf915a osc/rdma: fix compile warning
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-02-27 16:26:00 -07:00
George Bosilca
366d64b7e5 Move the collective structure outside the communicator.
As we changed the ABI (forcing a major release), we can limit
the size of the predefined communicators by moving the collective
structure outside the communicator. This might have a minimal,
but unnoticeable, impact on performance. This approach has been
discussed during the January 2017 devel meeting.

Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
Signed-off-by: Joshua Hursey <jhursey@us.ibm.com>
2017-02-27 11:54:17 -06:00
Nathan Hjelm
581bff9871 Merge pull request #3034 from hjelmn/osc_rdma_atomic
osc/rdma: make locking code more robust
2017-02-27 08:46:52 -07:00
Nathan Hjelm
4707c7c5e0 osc/rdma: make locking code more robust
Under heavy load the locking code could fail if the underlying btl
module started to return OPAL_ERR_OUT_OF_RESOURCE on atomic
operations. This commit updates the code to gracefully handle btl
errors.

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-02-27 00:01:26 -07:00
Gilles Gouaillardet
af0b5cffb4 asm: rename the AMD64 into X86_64
in this context, AMD64 really means amd64 or em64t, so let's
rename this into X86_64 in order to avoid any confusion

Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
2017-02-27 15:10:50 +09:00
Gilles Gouaillardet
4184c01be5 Merge pull request #2393 from bosilca/topic/no_predefined_ddt_refcount
Don't refcount the predefined datatypes.
2017-02-21 09:38:11 +09:00
Todd Kordenbrock
048f757d9f osc-portals4: add support for noncontiguous datatypes
This commit implements onesided operations for noncontiguous
datatypes using two different algorithms.

 * If the result and/or origin datatype is noncontiguous and the
   target datatype is contiguous, then an iovec MD is created for
   the result and origin.  The operation is performed using a
   single Portals4 call (unless it exceeds the max message size).
 * If the target datatype is noncontigous, then an algorithm
   similar to the one in osc-rdma is used to loop over the
   contiguous blocks of each datatype.  The operation is
   performed using multiple Portals4 calls.

This commit ensures that individual operations do not exceed the
max atomic size or the max message size supported by the device.

Signed-off-by: Todd Kordenbrock <thkgcode@gmail.com>
2017-02-15 16:17:13 -06:00
Gilles Gouaillardet
cd4537193c osc/sm: fix MPI_Win_allocate_shared() alignment
add padding so the memory allocated by MPI_Win_allocate_shared()
is 64 bytes aligned.

Thanks Joseph Schuchart for the bug report

Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
2017-02-15 13:40:48 +09:00
Nathan Hjelm
cc4a0fabcf Merge pull request #2727 from hjelmn/osc_rdma
osc/rdma: fix typo in check for MPI_MODE_NOCHECK
2017-02-14 10:50:33 -07:00
Geoff Paulsen
4917e44a7d Merge pull request #2832 from jjhursey/topic/ibm/osc-base-dt-abort
osc/base: Detect unsupported data types and abort
2017-02-05 04:26:04 -06:00
Nathan Hjelm
362ac8b87e osc/pt2pt: fix threading issues
This commit fixes a number of threading issues discovered in
osc/pt2pt. This includes:

 - Lock the synchronization object not the module in osc_pt2pt_start.
   This fixes a race between the start function and processing post
   messages.

 - Always lock before calling cond_broadcast. Fixes a race between
   the waiting thread and signaling thread.

 - Make all atomically updated values volatile.

 - Make the module lock recursive to protect against some deadlock
   conditions. Will roll this back once the locks have been
   re-designed.

 - Mark incoming complete *after* completing an accumulate not
   before. This was causing an incorrect answer under certain
   conditions.

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-02-01 10:33:01 -07:00
Nysal Jan K.A
94f92f6b49 osc/base: Detect unsupported data types and abort
Using MPI_MINLOC or MPI_MAXLOC with the following data types
leads to data corruption:
 * MPI_DOUBLE_INT
 * MPI_LONG_INT
 * MPI_SHORT_INT
 * MPI_LONG_DOUBLE_INT

Detect this print a error message and abort.
This workaround should be removed once the following issue is resolved:
 * https://github.com/open-mpi/ompi/issues/1666

Signed-off-by: Joshua Hursey <jhursey@us.ibm.com>
2017-01-25 15:28:28 -06:00
Nathan Hjelm
0497ec0b70 osc/rdma: fix typo in check for MPI_MODE_NOCHECK
This commit fixes two typos in the lock_all path that inverted the
MPI_MODE_NOCHECK flag.

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-01-12 11:28:11 -07:00
George Bosilca
c2cd717f82 Don't refcount the predefined datatypes.
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
2017-01-11 16:48:59 -05:00
Ralph Castain
dadc6fbaf6 Merge pull request #2448 from thananon/remove_request_lock
Completely removed ompi_request_lock and ompi_request_cond
2017-01-03 19:31:46 -08:00
Mark Allen
eec1d5bf2e osc/pt2pt: Fix hang with Put and Win_lock_all
* When using `MPI_Put` with `MPI_Win_lock_all` a hang is possible since
   the `put` is waiting on `eager_send_active` to become `true` but
   that variable might not be reset in the case of `MPI_Win_lock_all`
   depending on other incoming events (e.g., `post` or ACKs of lock
   requests.

Signed-off-by: Joshua Hursey <jhursey@us.ibm.com>
2016-12-16 11:52:53 -05:00
Mark Allen
0d1336b4a8 osc/pt2pt: Fix Lock/Unlock and Get wrong answer
* When using `MPI_Lock`/`MPI_Unlock` with `MPI_Get` and non-contiguous
   datatypes is is possible that the unlock finishes too early before
   the data is actually present in the recv buffer.
 * We need to wait for the irecv to complete before unlocking the target.
   This commit waits for the outgoing fragment counts to become equal
   before unlocking.

Signed-off-by: Joshua Hursey <jhursey@us.ibm.com>
2016-12-16 11:52:51 -05:00
Mark Allen
1ebf9fd3a4 osc/pt2pt: Fix PSCW after Fence wrong answer.
* If the user uses PSCW synchronization after a Fence then the previous
   epoch is not reset which can cause the PSCW to transfer data before
   it is ready leading to wrong answers.
 * This commit resets the `eager_send_active` in the start call.

Signed-off-by: Joshua Hursey <jhursey@us.ibm.com>
2016-12-16 11:52:49 -05:00
Ralph Castain
585540bcee Reduce the flood of warnings due to uninitialized variables, mismatched types, and unused things to a more bearable trickle
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
2016-12-14 16:33:50 -08:00
Ralph Castain
1e2019ce2a Revert "Update to sync with OMPI master and cleanup to build"
This reverts commit cb55c88a8b.
2016-11-22 15:03:20 -08:00
Thananon Patinyasakdikul
b25a8c3fa5 Completely removed ompi_request_lock and ompi_request_cond as we dont need them anymore.
Signed-off-by: Thananon Patinyasakdikul <tpatinya@utk.edu>
2016-11-22 17:58:31 -05:00
Ralph Castain
cb55c88a8b Update to sync with OMPI master and cleanup to build
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
2016-11-22 14:24:54 -08:00
Gilles Gouaillardet
bd364d29f7 osc/sm: plug an other memory leak in ompi_osc_sm_free
Fixes open-mpi/ompi@f1b473ee63

Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
2016-11-14 23:19:07 -07:00
Gilles Gouaillardet
f1b473ee63 osc/sm: plug a memory leak in ompi_osc_sm_free
Thanks Joseph Schuchart for the report.

Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
2016-11-14 22:22:43 -07:00
Gilles Gouaillardet
958e29f929 osc/rdma: silence a warning
declare a local variable volatile and silence CID 1372692
2016-10-13 16:10:07 +09:00
Nathan Hjelm
e8ef503bee osc/rdma: fix warnings
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2016-10-12 10:17:25 -06:00
Nathan Hjelm
432d79046b Merge pull request #2197 from tkordenbrock/topic/master/osc-rdma.put.use.true_extent
osc-rdma: fix datatype lower bound errors in ompi_osc_rdma_master()
2016-10-11 10:42:02 -06:00
Todd Kordenbrock
05f86b5df7 osc-rdma: fix datatype lower bound errors in ompi_osc_rdma_master()
Instead of ompi_datatype_get_extent(), use ompi_datatype_get_true_extent()
to get the local and remote lower bound.  For derived types like
subarray, true_lb is the correct offset for RDMA operations.
2016-10-10 06:45:28 -05:00
Todd Kordenbrock
cc863ff9fb osc-portals4: fix datatype errors in put()
Instead of ompi_datatype_get_extent(), use ompi_datatype_get_true_extent()
to get the origin and target lower bound.  For derived types like
subarray, true_lb is the correct offset for RDMA operations.  Also,
instead of the extent use the size of the datatype.
2016-10-10 06:45:14 -05:00
Todd Kordenbrock
c536e11cf3 osc-portals4: fix offset bug in raccumulate()
This commit fixes a bug where the remote offset was used as both
the local and remote offset.

Thanks to @PDeveze for the patch.
2016-10-04 09:09:17 -05:00
Nathan Hjelm
59bae1a330 osc/rdma: fix typo in compare-and-swap
This commit fixes a typo in compare-and-swap when retrieving the
memory region associated with a displacement. It was erroneously 8
bytes instead of the datatype size. This can cause an incorrect RMA
range error when the compare-and-swap is less than 4 bytes from the
end of the region.

Fixed open-mpi/ompi#2080

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2016-09-14 16:49:42 -06:00
Nathan Hjelm
7c8e7691a7 Merge pull request #2045 from hjelmn/osc_rdma_atomics
osc/rdma: add support for network AMOs
2016-09-08 11:21:49 -06:00
Nathan Hjelm
1ce5847e8b osc/rdma: add support for network AMOs
This commit adds support for using network AMOs for MPI_Accumulate,
MPI_Fetch_and_op, and MPI_Compare_and_swap. This support is only
enabled if the ompi_single_intrinsic info key is specified or the
acc_single_interinsic MCA variable is set. This configuration
indicates to this implementation that no long accumulates will be
performed since these do not currently mix with the AMO
implementation.

This commit also cleans up the code somwhat. This includes removing
unnecessary struct keywords where the type is also typedef'd.

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2016-09-01 15:47:33 -06:00
Nathan Hjelm
cb1cb5ffed osc/pt2pt: do not use frag send to send lock request
This commit cleans up some code in the passive target path. The code
used the buffered frag control send path but it is more appropriate to
use the unbuffered one. This avoids checking structures that are
should not be in use in this path.

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2016-09-01 09:57:27 -06:00
Nathan Hjelm
99b26644c1 Merge pull request #2011 from hjelmn/osc_pt2pt_fix
osc/pt2pt: fix possible race in peer locking
2016-08-29 09:17:36 -06:00
Nathan Hjelm
e53de7ecbe osc/rdma: fix bug in dynamic memory window tracking code
This commit fixes an ordering bug in the code that keeps track of all
attached memory windows. The code is intended to keep the memory
regions sorted but was often inserting at the wrong index. Thanks to
Christoph Niethammer for reporting the issue. The reproducer will be
added to nightly MTT testing.

Fixes open-mpi/ompi#2012

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2016-08-25 12:08:46 -06:00
Nathan Hjelm
7af138f83b osc/pt2pt: fix possible race in peer locking
It is possible for another thread to process a lock ack before the
peer is set as locked. In this case either setting the locked or the
eager active flag might clobber the other thread. To address this the
flags have been made volatile and are set atomically. Since there is
no a opal_atomic_or or opal_atomic_and function just use cmpset for
now.

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2016-08-25 09:28:25 -06:00
Nathan Hjelm
70f8a6e792 osc/pt2pt: fix several bugs
This commit fixes some bugs uncovered during thread testing of
2.0.1rc1. With these fixes the component is running cleanly with
threads.

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2016-08-24 14:35:45 -06:00
Ralph Castain
6549c878a9 Silence the warnings 2016-08-22 15:35:27 -07:00
Nathan Hjelm
e5c7512692 Merge pull request #1983 from hjelmn/request_cb
ompi/request: change semantics of ompi request callbacks
2016-08-18 08:31:56 -06:00
Nathan Hjelm
6aa658ae33 ompi/request: change semantics of ompi request callbacks
This commit changes the sematics of ompi request callbacks. If a
request's callback has freed or re-posted (using start) a request
the callback must return 1 instead of OMPI_SUCCESS. This indicates
to ompi_request_complete that the request should not be modified
further. This fixes a race condition in osc/pt2pt that could lead
to the req_state being inconsistent if a request is freed between
the callback and setting the request as complete.

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2016-08-17 20:14:01 -06:00
Nathan Hjelm
40b70889e5 osc/pt2pt: make receive count an unsigned int
This receive_count MCA variable should never be negative. Change it
to an unsigned int.

Signed-off-by: Nathan Hjelm <hjelmn@me.com>
2016-08-17 08:14:24 -06:00
Gilles Gouaillardet
8faa1edafa osc/pt2pt: silence misc warnings 2016-08-17 14:24:14 +09:00
Nathan Hjelm
9444df1eb7 osc/pt2pt: make lock_all locking on-demand
The original lock_all algorithm in osc/pt2pt sent a lock message to
each peer in the communicator even if the peer is never the target of
an operation. Since this scales very poorly the implementation has
been replaced by one that locks the remote peer on first communication
after a call to MPI_Win_lock_all.

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2016-08-11 15:33:07 -06:00