This commit fixes a bug that disabled both the RDMA pipeline and RDMA
protocols in ob1. ob1 was internally caching the values of
opal_leave_pinned and opal_leave_pinned_pipeline at init time. This is
no longer valid as opal_leave_pinned may be set by any call to a btl's
add_procs.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
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>
this fixes the issue reported by Nicolas Joly on the mailing: the sharedfp/lockedfile component does not support right now a scenario where multiple jobs read from the same input file, due to a collision of the filenames utilized for the sharedfp handle. Although not part of the oroginal report, the same occurs for the sharedfp/sm component. Add therefore the jobid to be part of the lockedfilename/sm file name.
use the OMPI_CAST_RTE_NAME macro to determine jobid
Fixes: #3098
Signed-off-by: Edgar Gabriel <egabriel@central.uh.edu>
this commit brings over the behavior from the 2.x series to master, mostly with the fork for the 3.x series in mind.
Also, use strncasecmp instead of two strncmps
Signed-off-by: Edgar Gabriel <egabriel@central.uh.edu>
- Support MPI-2.2 and MPI-3.0 COLL features.
* `MPI_REDUCE_SCATTER_BLOCK`
* neighborhood collective communication
* nonblocking collective communication
- Add `*_BASE_ARGS` and `*_BASE_ARG_NAMES` for convenience.
- Use parameter names used in the MPI Standard.
Signed-off-by: KAWASHIMA Takahiro <t-kawashima@jp.fujitsu.com>
Memory hooks are now set-up on demand. pml/yalla, mtl/mxm and
coll/hcoll need the memory hooks, so make sure those are installed.
Signed-off-by: Yossi Itigin <yosefe@mellanox.com>
Per a prior commit, the presence of "hwloc.h" can cause ambiguity when
using --with-hwloc=external (i.e., whether to include
opal/mca/hwloc/hwloc.h or whether to include the system-installed
hwloc.h).
This commit:
1. Renames opal/mca/hwloc/hwloc.h to hwloc-internal.h.
2. Adds opal/mca/hwloc/autogen.options to tell autogen.pl to expect to
find hwloc-internal.h (instead of hwloc.h) in opal/mca/hwloc.
3. s@opal/mca/hwloc/hwloc.h@opal/mca/hwloc/hwloc-internal.h@g in the
rest of the code base.
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
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>
* Include a 'demo' component that shows some of the features.
* Currently has hooks for:
- MPI_Initialized
- top, bottom
- MPI_Init_thread
- top, bottom
- MPI_Finalized
- top, bottom
- MPI_Init
- top (pre-opal_init), top (post-opal_init), error, bottom
- MPI_Finalize
- top, bottom
* Other places in ompi can 'register' to hook into any one of these places
by passing back a component structure filled with function pointers.
* Add a `MCA_BASE_COMPONENT_FLAG_REQUIRED` flag to the MCA structure that
is checked by the `hook` framework. If a required, static component has
been excluded then the `hook` framework will fail to initialize.
- See note in `opal/mca/mca.h` as to why this is checked in the `hook`
framework and not in `opal/mca/base/mca_base_component_find.c`
Signed-off-by: Joshua Hursey <jhursey@us.ibm.com>
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>
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>
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>
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>
* Since we are adding a new function to `mca_coll_base_module_2_1_0_t`
we need to increase the version of the module structure to `2_2_0`.
* Add a comment just above the PREDEFINED_COMMUNICATOR_PAD describing
it's purpose and when it should change. To help future developers
trying to answer the question noted in the comment.
Signed-off-by: Joshua Hursey <jhursey@us.ibm.com>
* Negative values are parameter errors for neighborhood collectives
- Add checks to the mpi/c interface `MPI_PARAM_CHECK`
* Fix a success check for neighbor_alltoallw with dist_graph
Signed-off-by: Joshua Hursey <jhursey@us.ibm.com>
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>
Fixes a wrong answer from MPI_Ireduce when the red_sched_chain()
path was taken (which only happens for np<=4 and mesgsize>=64k).
The way libnbc treats MPI_IN_PLACE is to set sbuf == rbuf, and
whether an algorithm will work cleanly or not after that depends on the
details.
In this case the last steps of the algorithm amounted to
(right neighbor is sending us reduction results from ranks 1..n-1)
recv into rbuf from right neighbor
add the contribution from our sbuf into rbuf
this would be fine in general, but if sbuf==rbuf, that recv overwrites
the sbuf. I changed it to recv into a tmpbuf if MPI_IN_PLACE was used.
Signed-off-by: Geoffrey Paulsen <gpaulsen@us.ibm.com>
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>
MPI_Allgatherv with MPI_IN_PLACE reads data from wrong location.
They were locating the MPI_IN_PLACE send buffer as
```c
send_buf = (char*)rbuf;
for (i = 0; i < rank; ++i) {
send_buf += ((ptrdiff_t)rcounts[i] * extent);
}
```
when it should be
```c
send_buf = (char*)rbuf;
send_buf += ((ptrdiff_t)disps[rank] * extent);
```
because disps[] specifies where things are in the v-style buffers.
Signed-off-by: Joshua Hursey <jhursey@us.ibm.com>
when a file is opened a second time for shared file pointer operations,
avoid setting the create and exclusive flag.
Signed-off-by: Edgar Gabriel <egabriel@central.uh.edu>
it looks like disabling the lazy_open flag for sharedfp components
revealead a bug that lead to a crash in file_close in some tests. Make
sure the SHAREDFP_IS_SET flag is correctly set (and not overwritten again),
and we use that to avoid a double-free of the communicator.
Signed-off-by: Edgar Gabriel <egabriel@central.uh.edu>
Revert the logic of io_ompio_sharedfp_lazy_open. The user now has to explicitely
disable shared fp in order for the structures not to be allocated.
Otherwise, resetting the shared fp e.g. in case the file was opened
in append mode will not work correctly, the code could deadlock.
Signed-off-by: Edgar Gabriel <egabriel@central.uh.edu>
Fixes a bug reported on the mailing list. ompio did only reposition the individual
file pointer when the file was opened in append mode. Set the shared file
pointer also to point to the end of the file, similarly to the individual
file pointer.
Signed-off-by: Edgar Gabriel <egabriel@central.uh.edu>
There are only five places in the non-daemon code paths where opal_hwloc_topology is currently referenced:
* shared memory BTLs (sm, smcuda). I have added a code path to those components that uses the location string
instead of the topology itself, if available, thus avoiding instantiating the topology
* openib BTL. This uses the distance matrix. At present, I haven't developed a method
for replacing that reference. Thus, this component will instantiate the topology
* usnic BTL. Uses the distance matrix.
* treematch TOPO component. Does some complex tree-based algorithm, so it will instantiate
the topology
* ess base functions. If a process is direct launched and not bound at launch, this
code attempts to bind it. Thus, procs in this scenario will instantiate the
topology
Note that instantiating the topology on complex chips such as KNL can consume
megabytes of memory.
Fix pernode binding policy
Properly handle the unbound case
Correct pointer usage
Do not free static error messages!
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
* 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>
* 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>
* 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>
According to the MPI-3.1 p.52 and p.53 (cited below), a request
created by `MPI_*_INIT` but not yet started by `MPI_START` or
`MPI_STARTALL` is inactive therefore `MPI_WAIT` or its friends
must return immediately if such a request is passed.
The current implementation hangs in `MPI_WAIT` and its friends
in such case because a persistent request is initialized as
`req_complete = REQUEST_PENDING`. This commit fixes the
initialization.
Also, this commit fixes internal requests used in `MPI_PROBE`
and `MPI_IPROBE` which was marked wrongly as persistent.
MPI-3.1 p.52:
We shall use the following terminology: A null handle is a handle
with value MPI_REQUEST_NULL. A persistent request and the handle
to it are inactive if the request is not associated with any ongoing
communication (see Section 3.9). A handle is active if it is neither
null nor inactive. An empty status is a status which is set to return
tag = MPI_ANY_TAG, source = MPI_ANY_SOURCE, error = MPI_SUCCESS, and
is also internally configured so that calls to MPI_GET_COUNT,
MPI_GET_ELEMENTS, and MPI_GET_ELEMENTS_X return count = 0 and
MPI_TEST_CANCELLED returns false. We set a status variable to empty
when the value returned by it is not significant. Status is set in
this way so as to prevent errors due to accesses of stale information.
MPI-3.1 p.53:
One is allowed to call MPI_WAIT with a null or inactive request
argument. In this case the operation returns immediately with empty
status.
Signed-off-by: KAWASHIMA Takahiro <t-kawashima@jp.fujitsu.com>
Adds the new API hcoll_conetxt_free that resolves the issues
observed with the ctx cache and group_destroy_notify.
Signed-off-by: Valentin Petrov <valentinp@mellanox.com>
`sturct mca_pml_ob1_comm_proc_t`, which is allocated per
connected rank in a communicator, had two paddings after
`expected_sequence` and `send_sequence` by alignments.
By changing the order of the members, the size of
`mca_pml_ob1_comm_proc_t` is reduced by 8 bytes on 64-bit
architectures.
Signed-off-by: KAWASHIMA Takahiro <t-kawashima@jp.fujitsu.com>
This fixes a bug reported in-house occuring with this component. It is triggered if the data assigned to different aggregators is highly differing, leading to different number of internal iterations required to handle it.
Signed-off-by: Edgar Gabriel <egabriel@central.uh.edu>
protect the mca_coll_libnbc_component.active_requests list with
the new mca_coll_libnbc_component.lock mutex.
Thanks Jie Hu for the report
Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
change the default value of the mca_io_ompio_cycle_buffer_size parameter in order to avoid accidental truncation of a file for very large individual operations.
Thanks to @cniethammer for reporting it.
Signed-off-by: Edgar Gabriel <egabriel@central.uh.edu>
- instead of coll_base_comm_get_reqs(2) for irecv/isend, use only
one request allocated in the stack and do a irecv/send
- instead of ompi_request_wait_all(2), simpy ompi_request_wait
Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
this is generally done in mca_pml_ob1_recv_request_free(), but this is not invoked
in via mca_pml_ob1_recv(), so do it manually
Thanks Yvan Fournier for the report
Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
* If (legal) non-uniform data type signatures are used in ibcast
then the chosen algorithm may fail on the request, and worst case
it could produce wrong answers.
* Add an MCA parameter that, by default, protects the user from this
scenario. If the user really wants to use it then they have to
'opt-in' by setting the following parameter to false:
- `-mca coll_libnbc_ibcast_skip_dt_decision f`
* Once the following Issues are resolved then this parameter can
be removed.
- https://github.com/open-mpi/ompi/issues/2256
- https://github.com/open-mpi/ompi/issues/1763
Signed-off-by: Joshua Hursey <jhursey@us.ibm.com>
Adds mapping of the MPI Fortran pair types (2INTEGER, 2REAL, 2DBLPREC)
to the corresponding hcoll dtypes.
Signed-off-by: Valentin Petrov <valentinp@mellanox.com>
recvreq->req_recv.req_base.req_type should always be set before invoking
MCA_PML_OB1_RECV_REQUEST_INIT(recvreq, ...) otherwise, the previous type
might be set, and you could end up with MPC_PML_REQUEST_IMPROBE when
MCA_PML_REQUEST_RECV is expected.
Thanks Chris Pattison for the report and test case.
Fixesopen-mpi/ompi#2275
* If an error is detected internal to libnbc (e.g., PML truncation error)
this patch makes sure that the request is completed and the `MPI_ERROR`
field is set approprately.
* Make an attempt to cleanup outstanding requests before returning.
- This is a "best attempt" since not all PMLs support canceling requests.
In order to optimize for MPI_IN_PLACE, data is sent from the receive buffer.
consequently, it should be sent with the receive type and count.
Thanks Josh Hursey for the report and test case
Refs open-mpi/ompi#2256
- pass field_mask to ucp_init().
- use non-blocking disconnect.
- recv() with pre-allocated request.
- call opal_progress() from iprobe() and improbe().
- use shift pattern in connect/disconnect.
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.
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.
* Add a configure time option to rename libmpi(_FOO).*
- `--with-libmpi-name=STRING`
* This commit only impacts the installed libraries.
Internal, temporary libraries have not been renamed to limit the
scope of the patch to only what is needed.
For example:
```shell
shell$ ./configure --with-libmpi-name=wookie
...
shell$ find . -name "libmpi*"
shell$ find . -name "libwookie*"
./lib/libwookie.so.0.0.0
./lib/libwookie.so.0
./lib/libwookie.so
./lib/libwookie.la
./lib/libwookie_mpifh.so.0.0.0
./lib/libwookie_mpifh.so.0
./lib/libwookie_mpifh.so
./lib/libwookie_mpifh.la
./lib/libwookie_usempi.so.0.0.0
./lib/libwookie_usempi.so.0
./lib/libwookie_usempi.so
./lib/libwookie_usempi.la
shell$
```
--disable-io-ompio is a shortcut that disable the following
frameworks and components
- fbtl
- fcoll
- sharedfp
- common/ompio
- io/ompio
Fixesopen-mpi/ompi#1934
- move the mpi-io configury option into config/ompi_configure_options.m4
- add ompi/mca/common/ompio/configure.m4 so this component is not built when
Open MPI is configure'd with --disable-mpi-io
Fixesopen-mpi/ompi#2009
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.
Fixedopen-mpi/ompi#2080
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
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>
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>
if sendbuf is equal to recvbuf, that should not be interpreted
as equivalent to MPI_IN_PLACE on the non root rank(s)
Thanks Valentin Petrov for the report
predefined datatypes such as MPI_LONG_DOUBLE_INT are not really contiguous,
so use span as returned by opal_datatype_span() instead of type extent,
otherwise data might be written above allocated memory.
Thanks Valentin Petrov for the report
protect the remaining functions where necessary by a mutex lock
to avoid problems in multi-threaded executions. Some functions
do not require that in my opinion, and I provided an explanation
in those cases.
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.
Fixesopen-mpi/ompi#2012
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
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>
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>
Also, remove the lock/unlock in the file_open ompi-interface routines of romio314.
The global lock in the romio component does probably not work, it is easy to construct a testcase where two threads perform collective I/O operations on different file handles. With a global lock it is easy to deadlock. THe lock has to be at least on the file handle basis.
move the mutex to file/file.c to avoid duplicate symbol problem in file_open.c pfile_open.c
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>
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>
This commit fixes an issue that can occur if a target gets overwhelmed with
requests. This can cause osc/pt2pt to go into deep recursion with a stack
like req_complete_cb -> ompi_osc_pt2pt_callback -> start -> req_complete_cb
-> ... . At small scale this is fine as the recursion depth stays small but
at larger scale we can quickly exhaust the stack processing frag requests.
To fix the issue the request callback now simply puts the request on a
list and returns. The osc/pt2pt progress function then handles the
processing and reposting of the request.
As part of this change osc/pt2pt can now post multiple fragment receive
requests per window. This should help prevent a target from being overwhelmed.
Signed-off-by: Nathan Hjelm <hjelmn@me.com>
This commit updates the btl selection logic for the RDMA and RDMA
pipeline protocols to use a btl iff: 1) the btl is also used for eager
messages (high exclusivity), or 2) no other RDMA btl is available on
an endpoint and the pml_ob1_use_all_rdma MCA variable is true. This
fixes a performance regression with shared memory when an RDMA capable
network is available.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
Clang 5.1 on my mac was a sad panda compiling a couple
of files, complaining about uninitialized stack variables.
This commit makes clang a happier panda (or at least not so sad).
Signed-off-by: Howard Pritchard <howardp@lanl.gov>
Based on current implementation it is faster to use a blocking
send than the non-blocking version. Switch the exchange function
used in the barrier to use the blocking version combined with
the non-blocking version of the receive.
This is similar to open-mpi/ompi@223d75595d
It is possible for the start call to complete the requests. For this
reason the module rdma_frag field should be filled in before start is
called. If the request completes the completion callback will reset
the rdma_frag field to NULL. Fixes a bug discovered by @tkordenbrock.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
On start we were not correctly resetting all request fields. This was
leading to a double-completion on persistent receives. This commit
updates the base start code to reset the receive req_bytes_packed and
the send request convertor.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
- move the sort_iovec operations to fcoll/base
- move set_view_internal to common/ompio
- move set_file_default to common/ompio
- remove io_ompio_sort, not used anymore.
This commit expands the OPAL_THREAD macros to include 32- and 64-bit
atomic swap. Additionally, macro declararations have been updated to
include both OPAL_THREAD_* and OPAL_ATOMIC_*. Before this commit the
former was used with add and the later with cmpset.
Signed-off-by: Nathan Hjelm <hjelmn@me.com>
allow for toggling of both control/data progress models.
allow for using FI_AV_TABLE or FI_AV_MAP for av type.
Signed-off-by: Howard Pritchard <howardp@lanl.gov>
