Open MPI v4.1.1rc2 is now available at
https://www.open-mpi.org/software/ompi/v4.1/
Changes since the 4.1.0 release include:
- Reverted temporary solution that worked around launch issues in
SLURM v20.11.{0,1,2}. SchedMD encourages users to avoid these
versions and to upgrade to v20.11.3 or newer.
- Updated PMIx to v3.2.2.
- Fixed configuration issue on Apple Silicon observed with
Homebrew. Thanks to François-Xavier Coudert for reporting the issue.
- Disabled gcc built-in atomics by default on aarch64 platforms.
- Disabled UCX PML when UCX v1.8.0 is detected. UCX version 1.8.0 has a bug that
may cause data corruption when its TCP transport is used in conjunction with
the shared memory transport. UCX versions prior to v1.8.0 are not affected by
this issue. Thanks to @ksiazekm for reporting the issue.
- Fixed detection of available UCX transports/devices to better inform PML
prioritization.
- Fixed SLURM support to mark ORTE daemons as non-MPI tasks.
- Improved AVX detection to more accurately detect supported
platforms. Also improved the generated AVX code, and switched to
using word-based MCA params for the op/avx component (vs. numeric
big flags).
- Improved OFI compatibility support and fixed memory leaks in error
handling paths.
- Improved HAN collectives with support for Barrier and Scatter. Thanks
to @EmmanuelBRELLE for these changes and the relevant bug fixes.
- Fixed MPI debugger support (i.e., the MPIR_Breakpoint() symbol).
Thanks to @louisespellacy-arm for reporting the issue.
- Fixed ORTE bug that prevented debuggers from reading MPIR_Proctable.
- Removed PML uniformity check from the UCX PML to address performance
regression.
- Fixed MPI_Init_thread(3) statement about C++ binding and update
references about MPI_THREAD_MULTIPLE. Thanks to Andreas Lösel for
bringing the outdated docs to our attention.
- Added fence_nb to Flux PMIx support to address segmentation faults.
- Ensured progress of AIO requests in the POSIX FBTL component to
prevent exceeding maximum number of pending requests on MacOS.
- Used OPAL's mutli-thread support in the orted to leverage atomic
operations for object refcounting.
- Fixed segv when launching with static TCP ports.
- Fixed --debug-daemons mpirun CLI option.
- Fixed bug where mpirun did not honor --host in a managed job
allocation.
- Made a managed allocation filter a hostfile/hostlist.
- Fixed bug to marked a generalized request as pending once initiated.
- Fixed external PMIx v4.x check.
- Fixed OSHMEM build with `--enable-mem-debug`.
- Fixed a performance regression observed with older versions of GCC when
__ATOMIC_SEQ_CST is used. Thanks to @BiplabRaut for reporting the issue.
- Fixed buffer allocation bug in the binomial tree scatter algorithm when
non-contiguous datatypes are used. Thanks to @sadcat11 for reporting the
issue.
- Fixed bugs related to the accumulate and atomics functionality in the
osc/rdma component.
- Fixed race condition in MPI group operations observed with
MPI_THREAD_MULTIPLE threading level.
- Fixed a deadlock in the TCP BTL's connection matching logic.
- Fixed pml/ob1 compilation error when CUDA support is enabled.
- Fixed a build issue with Lustre caused by unnecessary header includes.
- Fixed a build issue with IMB LSF workload manager.
- Fixed linker error with UCX SPML.
We are waiting for resolution on some critical issues, but we are close to a
final release. Please do test this release candidate and report issues via
email to mailto:devel@lists.open-mpi.org, or open a Github issue at
https://github.com/open-mpi/ompi/issues/.
Raghu
