1
1
Граф коммитов

5770 Коммитов

Автор SHA1 Сообщение Дата
Nathan Hjelm
0de9445fc7 osc/rdma: fix bugs when running more than one process per node
A previous commit updated the one-sided code to register the state
region only once. This created an issue when using the scratch lock
with fetching atomics. In this case on any rank that isn't local rank
0 the module->state_handle is NULL. This commit fixes the issue by
removing the scratch lock and using a fragment pointer instead.

Fixes open-mpi/ompi#1290

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2015-12-15 11:25:25 -07:00
Nathan Hjelm
b7ba301310 Merge pull request #1165 from hjelmn/add_procs_group
ompi/group: release ompi_proc_t's at group destruction
2015-12-14 13:53:42 -08:00
Nathan Hjelm
9d659465b7 Merge pull request #1210 from artpol84/icbarrier_fix
Fix NBC iBarrier for inter-communicators.
2015-12-14 13:52:38 -08:00
Nathan Hjelm
4b3dac5933 Merge pull request #1216 from artpol84/icgatherv_fix
Fix NBC iGatherv for inter-communicators.
2015-12-14 13:51:58 -08:00
Matias Cabral
7cfd7d50b9 Merge pull request #1219 from matcabral/PSM2_tag_hashing
Support for PSM2 hashing lookup in message queue.
2015-12-14 12:01:55 -08:00
matcabral
9a1f9be146 A new internal feature in PSM2 will use hash tables to
accelerate message queue lookups if the lookups have
the proper tag&mask layout. OpenMPI should follow
PSM2's preferred tag&mask spec, so that PSM2 can provide
a performance benefit.
2015-12-14 10:13:39 -08:00
Artem Polyakov
2d0919dbdc Fix NBC iGatherv for inter-communicators.
We need to use remote size to form a schedule.
2015-12-14 12:19:10 +06:00
Artem Polyakov
fc17deca43 Fix NBC iBarrier for inter-communicators.
Remove send of the extra message. This bug hase triggered on
MPICH/coll/nbicbarrier test. In this test a series of communicators
are created.
This extre-message was reseived after original communicator was destroyed
and queued into non_existing_communicator_pending. When new completely
unrelated communicator with the same id as original was created this message
was pushed into the frags_cant_match queue and caused seq numbers skew and hang.
2015-12-12 13:27:31 +06:00
Gilles Gouaillardet
3a3b13ea12 coll/base: fix an integer overflow in ompi_coll_base_reduce_generic
Refs open-mpi/ompi#1198
2015-12-11 13:55:59 +09:00
Alina Sklarevich
3ffd8dcd20 PML UCX: fix typo (following 7becc54d). 2015-12-10 13:51:10 +02:00
Nathan Hjelm
dae3746d2f Merge pull request #1190 from kawashima-fj/pr/sm-win-test-fix
osc/sm: Fix a bug that `MPI_WIN_TEST` does not update `flag` to 0
2015-12-08 06:39:16 -07:00
KAWASHIMA Takahiro
9c7b6a4352 osc/sm: Fix a bug that MPI_WIN_TEST does not update flag to 0.
`MPI_WIN_TEST` must update the `flag` parameter to 0 when not all
origin processes called `MPI_WIN_COMPLETE`. But sm OSC doesn't.
If the caller initialize the `flag` argument to a non-0 value,
the caller will receive the non-0 `flag` value.
2015-12-08 19:23:21 +09:00
Gilles Gouaillardet
59a361b781 ompio: correctly handle zero f_cc_size in mca_io_ompio_simple_grouping 2015-12-08 17:00:11 +09:00
Nathan Hjelm
f68c315188 pml/ob1: add missing ompi_request_wait_completion for buffered sends
This commit adds a call to ompi_request_wait_completion for buffered
sends. Without this line it is possible to get into a state where the
data is never sent.

Fixes open-mpi/ompi#1185

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2015-12-07 22:28:07 -07:00
Gilles Gouaillardet
bfe8e03d9d fcoll/two_phase: use ompi_mpi_abort instead of PMPI_Abort
Thanks Jeff for the review
2015-12-07 11:34:36 +09:00
Gilles Gouaillardet
37c978f5e9 coll/libnbc: correctly handle changed types.
this fixes open-mpi/ompi@d816d1c194
thanks Jeff for the review
2015-12-07 10:13:43 +09:00
George Bosilca
3a9664ac9d Fix Coverity CIDs 1341584-1341589. 2015-12-06 14:06:36 -05:00
bosilca
8fee96c086 Merge pull request #1091 from bosilca/topic/datatype_span
Cleanup the temporary memory allocation in collectives
2015-12-03 19:25:04 -05:00
Gilles Gouaillardet
a5440ade5f topo/treematch: do not invoke hwloc_topology_{init,load}
* this is not necessary
 * this overwrites existing topology, that could be different if hwloc_base_topo_file is used
2015-12-03 11:24:32 +09:00
George Bosilca
688108cf7f Patch submitted by @ggouaillardet on ticket #1091. 2015-12-02 20:42:18 -05:00
George Bosilca
4d00c59b2e Cleanup the memory handling for temporary buffers in
some of the collective modules. Added a new function
opan_datatype_span, to compute the memory span of
count number of datatype, excluding the gaps in the
beginning and at the end. If a memory allocation is
made using the returned value, the gap (also returned)
should be removed from the allocated pointer.
2015-12-02 20:42:18 -05:00
Jeff Squyres
15325c8094 op/x86: change the owner to Ralph
Cisco no longer cares about this component, but Intel might.
Transferring ownership to Ralph.
2015-12-01 15:08:07 -08:00
Nathan Hjelm
5334d22a37 ompi/group: release ompi_proc_t's at group destruction
This commit changes the way ompi_proc_t's are retained/released by
ompi_group_t's. Before this change ompi_proc_t's were retained once
for the group and then once for each retain of a group. This method
adds unnecessary overhead (need to traverse the group list each time
the group is retained) and causes problems when using an async
add_procs.

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2015-11-30 23:03:47 -07:00
Ryan Grant
324534b191 Merge pull request #1161 from tkordenbrock/topic/add.triggered.scatter
coll-portals4: add scatter and iscatter implementations that use Portals4 triggered operations
2015-11-30 16:53:47 -07:00
Todd Kordenbrock
4721b70dd5 coll-portals4: add scatter and iscatter implementations that use Portals4 triggered operations
This commit adds implementations of scatter and iscatter using
Portals4 triggered operations.  Currently, the only algorithm
is linear.
2015-11-30 15:07:18 -06:00
Todd Kordenbrock
f6f525e0d8 coll-portals4: remove unneeded code from gather
This commit removes two pieces of unneeded code from gather.  First
it removes destroy_tree() calls from linear_top(), because the
linear algorithm does not create a tree, so there is no need to
destroy it.  Second it removes unpack_bytes from the gather request
because it was calculated but never used.
2015-11-30 10:38:51 -06:00
Gilles Gouaillardet
80f02518ff topo/base: correctly free the topo object in mca_topo_base_dist_graph_create_adjacent 2015-11-30 15:33:59 +09:00
Ryan Grant
81d482dca6 Merge pull request #1137 from francois-wellenreiter/trig_mtl_rdv
MTL portals4 : improve the rendez-vous protocol using PtlTriggeredGet…
2015-11-24 17:31:31 -07:00
Ryan Grant
219581e87e Merge pull request #1090 from tkordenbrock/topic/check.for.invalid.handles.in.finalize
mtl-portals4: test for valid handle before releasing resources
2015-11-20 07:54:44 -06:00
Mike Dubman
c544620a7c Merge pull request #1138 from igor-ivanov/pr/yalla-valgrind
yalla: fix valgrind error due to uninitialized status field.
2015-11-20 07:19:11 -05:00
Gilles Gouaillardet
002c7b8b3a fcoll/two_phase: use PMPI_* insted of MPI_* 2015-11-20 13:46:19 +09:00
Gilles Gouaillardet
561e7f6647 vprotocol/pessimist: use internal ompi_* insted of MPI_* 2015-11-20 13:46:19 +09:00
Gilles Gouaillardet
025fd8a9fc osc: use PMPI_* insted of MPI_* 2015-11-20 13:46:19 +09:00
Gilles Gouaillardet
d816d1c194 coll/libnbc: use PMPI_* and internal ompi_* insted of MPI_* 2015-11-20 13:46:19 +09:00
yosefe
3bb1270715 yalla: fix valgrind error due to uninitialized status field. 2015-11-19 10:59:31 +02:00
Francois WELLENREITER
9126ea5e82 MTL portals4 : improve the rendez-vous protocol using PtlTriggeredGet operation 2015-11-19 09:52:53 +01:00
Edgar Gabriel
9e5ade4e8b argh, a debugging sleep statement got into the source code. 2015-11-16 13:26:57 -06:00
Edgar Gabriel
dbfbcdecd5 make adjustments for the default settings of grouping parameters and the default contiguous group size option.
minor bug fix in the simple grouping strategy.
2015-11-16 08:17:27 -06:00
Edgar Gabriel
27628774c7 add a new option for a simple aggregator selection which has zero communication costs. 2015-11-16 08:17:26 -06:00
Edgar Gabriel
66c1ea5fcb change the default value of the grouping option. Also add new grouping option which skips the refinement step in the aggregator selection. 2015-11-16 08:17:23 -06:00
Edgar Gabriel
e8e117503d reduce the communication volume during MPI_File_set_view 2015-11-16 08:17:22 -06:00
yohann
005400a937 mtl/ofi: Make sure the resources are managed by the provider. 2015-11-13 16:16:58 -08:00
Nathan Hjelm
9ef0821856 osc/rdma: fix some threading bugs
There were two bugs in osc/rdma when using threads:

 - Deadlock is ompi_osc_rdma_start_atomic. This occurs because
   ompi_osc_rdma_frag_alloc is called with the module lock. To fix the
   issue the module lock is now recursive. In the future I will add a
   new lock to protect just the current rdma fragment.

 - Do not drop the lock in ompi_osc_rdma_frag_alloc when calling
   ompi_osc_rdma_frag_complete. Not only is it not needed but dropping
   the lock at this point can cause a competing thread to mess up the
   state.

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2015-11-12 20:25:57 -07:00
Yossi
b750b72a81 Merge pull request #1127 from yosefe/topic/pml-ucx-implement-cancel
pml_ucx: implement cancel, and add small optimizations.
2015-11-12 10:50:48 +02:00
yosefe
7becc54d67 pml_ucx: fix typo. 2015-11-12 09:57:41 +02:00
Todd Kordenbrock
b9630f802b Merge pull request #1120 from francois-wellenreiter/mtl_min_mdbind
mtl-portals4 : remove useless PtlMDBind PtlMDRelease calls for rendez-vous messages
2015-11-10 14:34:19 -06:00
yosefe
d66b01d380 pml_ucx: implement cancel, and add small optimizations. 2015-11-10 17:40:06 +02:00
Gilles Gouaillardet
d6ff25b9a2 pml/monitoring: initialize common symbols 2015-11-10 13:58:54 +09:00
Jeff Squyres
e89ecac83c bml r2: fix exclusivity comparison
Fixes open-mpi/ompi#1106
2015-11-06 13:26:32 -08:00
Francois WELLENREITER
b301b49a40 MTL portals4 : remove useless PtlMDBind PtlMDRelease calls for rendez-vous messages 2015-11-06 15:55:44 +01:00