-
Notifications
You must be signed in to change notification settings - Fork 144
Conference call notes 20220511
Kenneth Hoste edited this page Jul 6, 2022
·
7 revisions
(back to Conference calls)
Notes on the 197th EasyBuild conference call, Wednesday 11 May 2022 (15:00 UTC)
Alphabetical list of attendees (12):
- Sebastian Achilles (JSC, Germany)
- Simon Branford (Univ. of Birmingham, UK)
- Fotis Georgatos (EPFL, Switzerland)
- Jasper Grimm (University of York, UK)
- Kenneth Hoste (HPC-UGent, Belgium)
- Adam Huffman (Big Data Institute, Oxford, UK)
- Kurt Lust (Univ. of Antwerpen, Belgium + LUMI User Support Team)
- Sebastien Moretti (SIB, Switzerland)
- Jurij Pečar (EMBL, Germany)
- Bart Oldeman (ComputeCanada)
- Davide Vanzo (Microsoft)
- Lars Viklund (Umeå University, Sweden)
- overview of recent developments + outlook to next EasyBuild release
- progress on 2022a update of common toolchains
- Q&A
- release timeline
- latest release: EasyBuild v4.5.4 (31 March 2022)
- ETA next release: mid May 2022
- recent changes
-
framework
- bug fixes
-
enhancements
- add definition for gfbf toolchain (GCC, FlexiBLAS, FFTW) (PR #4006)
- changes
-
easyblocks
- bug fixes
-
enhancements
- ...
- updates
-
changes
- don't use
gold
linker by default for GCC >= 11.3 (PR #2711-
ld.gold
is no longer actively developed, so back to classicld.bfd
linker (by default)
-
- don't use
- new software
- ...
-
easyconfigs
- over 100 easyconfig PRs merged since last conf call
- bug fixes
- enhancements
-
(noteworthy) new software
- see also overview at https://github.com/easybuilders/easybuild-easyconfigs/issues?q=is%3Amerged+label%3Anew
- ...
-
noteworthy software updates
- GCC 11.3.0 (PR #15311) + GCC 12.1.0 (PR #15470)
- intel-compilers v2022.1.0 (PR #15483)
- impi v2021.6.0 (PR #15453)
- imkl v2022.1.0 (PR #15484)
- FlexiBLAS v3.1.3 with
GCC/11.3.0
(PR #15399) - Python v3.10.4 with
GCC/11.3.0
((PR #15404) - R v4.2.0 with
foss/2021b
(PR #15356) - R-bundle-Bioconductor v3.15 (PR #15444)
- ReFrame v3.11.0 (PR #15368)
- changes
- tweak find command used in preconfigopts in easyconfig for pkg-config v0.29.2 with system toolchain to avoid descending into other filesystems (PR #15313)
-
framework
- to merge/fix/tackle soon
-
framework
-
reported bugs / bug fixes
- ...
-
enhancements
- add
Generation
module naming scheme (PR #3547) - improve error finding (draft PR #4001)
- copies
ctest_log_parser.py
from Spack, contains patterns and filters that an be used to recognize error messages
- copies
- update
prepare_rpath_wrappers
to enable wrapper shipping with a module (WIP) PR #4003- see also companion PR for
GCC
easyblock (PR #2638)
- see also companion PR for
- tarball update when working with cloned repositories (issue #4002)
- crash when using string value for 'sources' in extension (issue #4004)
-
testoutput
option is ignored (issue #4007)
- add
-
changes
- ...
-
reported bugs / bug fixes
-
easyblocks
- bug reports/fixes
- enhancements
-
updates
- update sanity check in OpenMPI easyblock to support OpenMPI v5.0.0 (PR #2709)
- new software
- ...
-
changes
- ...
-
easyconfigs
- now over 600 open easyconfig PRs, we're way overdue a significant cleanup round...
- bug fixes
- enhancements
- ...
-
new software
- see https://github.com/easybuilders/easybuild-easyconfigs/labels/new
- MPItrampoline-based toolchain (WIP) (PR #14607)
- VisIt (PR #15109)
-
noteworthy software updates
- RoseTTAFold v1.0.0 (PR #13795)
- TensorFlow v2.7.1 (WIP) (PR #14990)
- failing tests are for new features of TensorFlow
- ignore these failing tests for now, report upstream
- hipSYCL v0.9.2 (PR #15074)
- Qt6 (PR #15096)
- AlphaFold v2.2.0 (PR #15129)
- failing
jax
tests on A100? - Sam: easy way out, just use older
jax
version that is still supported
- failing
- PyTorch v1.11.0 (PR #15137)
- CP2K v9.1 with
foss/2021a
(PR #15146) andintel/2021a
(PR #15147)- call for testing by others, and keep an eye on result of test suite...
- changes
- ...
-
framework
- see (
2022a
project](https://github.com/easybuilders/easybuild-easyconfigs/projects/4) in easyconfigs repo - GCC 11.3.0 as base (see merged PR #15311)
- with
ld.bfd
as default linker (rather thanld.gold
, which is no longer actively maintained) - see easyblocks PR #2711
- with
-
foss/2022a
:- GCC 11.3.0 (vs 11.2.0 in
foss/2021b
) - latest OpenMPI 4.x: v4.1.4 (PR #15426)
- latest FlexiBLAS/OpenBLAS(/BLIS) (see PR #15399)
- FlexiBLAS v3.1.3 (vs 3.0.4 in
foss/2021b
)- with OpenBLAS + BLIS backends
- OpenBLAS v0.3.20 (vs 0.3.18 in
foss/2021b
) - BLIS v0.9.0 (vs 0.8.1 in
foss/2021b
)
- FlexiBLAS v3.1.3 (vs 3.0.4 in
- FFTW: 3.3.10 (same as in
foss/2021b
)- split up in "serial" FFTW and FFTW.MPI
- requires small change in framework
- ScaLAPACK: 2.2.0 (update from 2.1.0 vs
foss/2021b
)
- GCC 11.3.0 (vs 11.2.0 in
-
intel/2022a
- other software versions
- Perl 5.34.1
- Python 3.10.4
- LLVM 14.0.3
- Mesa 22.0.3
- software to test on top of candidate versions for
foss/2022a
andintel/2022a
:- SciPy-bundle, CP2K, OpenFOAM, ...
- Bart has been looking into the OpenMPI + CUDA mess
- alternative approach to Åke's PR #14919
- could use CUDA as a build dependency to OpenMPI, but then we "break" the policy to only have FOSS in
foss
toolchain - alternative can be to include a small CUDA header file to build OpenMPI GPU-aware
- also fixes the problematic case reported by Caspar (cfr. issue #14801)
- some performance measurements were also done, shows that this approach introduces very little overhead; an overhead of max 10% is visible on very small messages (<1K) on MPI_Alltoall for 64 cores on shared memory but not on small point-to-point messages or MPI_Allreduce for example. We are not aware of real-world problematic overheads.
- PR by Bart coming up soon...