A list of ELPA releases in reverse chronological order¶
2023.05.001: See Changelog
2022.11.001: See Changelog
2022.11.001.rc2: Fixes to rc1; Store GPU setup information per ELPA object
2022.11.001.rc1: Production ready support of AMD GPUs, better scalability for ELPA 2stage, new C, and C++ interfaces. For details please see the Changelog.
2022.05.001: Better support (but still experimental) for AMD and Intel GPUs. For details please see the Changelog.
2021.11.002: Fixes a bug with choosing the Nvidia GPU kernels, if both the standard and the new kernel for V100 have been build
2021.11.001: Provides support for NVIDIA cusolver library and offers a GPU offloading of invert_trm and cholesky. Delievers improved performance on Nvidia A100 GPUs. Offers a new version of the autotuning mechanism.
2021.11.001.rc1: Release candidate rc1 for upcoming ELPA 2021.11.001 release
2021.05.002_bugfix: Backport of a fix of a potential MPI hanging error
2021.05.002: No changes for end-users. Fixes an SO version error in release 2021.05.001
2021.05.001_bugfix: Backport of a fix of a potential MPI hanging error
2021.05.001: Several bugfixes. Better maping of GPUs to tasks. Experimental support for AMD and INTEL GPUs. Support for FUGAKU. It is recommended to use ELPA 2021.05.001 due to an SO version error
2021.05.001.rc2: Several bugfixes. Better maping of GPUs to tasks. Experimental support for AMD and INTEL GPUs. Support for FUGAKU
2021.05.001.rc1: Several bugfixes. Better maping of GPUs to tasks. Experimental support for AMD and INTEL GPUs. Support for FUGAKU.
2020.11.001_bugfix: Backport of a fix of a potential MPI hanging error
2020.11.001: Several bugfixes. Experimental features for ARM SVE support. Switch internally from python2 to python3.
2020.05.001: Improved GPU version for ELPA 2 stage. Experimental feature of internal matrix redistribution. Support for gcc v10.
2019.11.001: Some bugfixes, and support of skew-symmetric matrices.
2019.05.002: Include, src files forgotten in previous release.
2019.05.001: Improve performance and some bugfixes.
2018.11.001: Improved autotuning functionality. Significantly improved performance during the transformation of the generalized eigenvalue problem.
2018.05.001: Provides better autotuning functionality and increased performance on the K-computer.
2017.11.001: Provide compute kernels for Power8 and Fujitsu Sparc64 processors. First implementation of ELPA autotuning capability. Improved GPU performance.
2017.05.003: Some bugfixes for ELPA 2017.05.002.
2017.05.002: Some bugfixes for ELPA 2017.05.001.
2017.05.001: Improved calculations on GPU / KNL and X86.Most importantly, this release candidate introduces the new re-designed API of ELPA, which is more flexible and should render future API changes mostly unnecessary. The API as defined in release ELPA 2016.11.001 is still available, however, flagged as “legacy”. The online documentation is available here.
2016.11.001: Introduce GPU support and single/double precision calculations. This release comes with an experimental API, which will most likely change in the next stable release!
2016.05.004: Fixes problems encountered in ELPA 2016.05.003. Cleanup of the install process.
2016.05.003: Fixes problems encountered in ELPA 2016.05.002. Cleanup of the install process.
2016.05.002: Fixes a problem with the test scripts in ELPA 2016.05.001. Cleanup of the install process.
2016.05.001: Among others new SSE gcc intrinsic kernels have been introduced, a bug due to unaligned memory in version elpa-2015.11.001 has been removed. Building ELPA without MPI parallelisation (one compute node only) is supported.
2015.11.001: This version featues support of Intel AVX2 and FMA instructions for the latest Intel processors. Intel, we thank you for your efforts!
2015.05.001: This version removes a potential memory leak, erroneously introduced in elpa-2013.08.006.tar.gz. A better build documentation is provided.
2015.02.002: This version provides C interfaces for the ELPA library.
2015.02.001: This version provides an optional QR-decomposition for real valued matrices. It can be used via setting an environment variable or via calling ELPA2 with an additional argument. Also it possible to obtain some debugging information by setting the environment variable “ELPA_DEBUG_MESSAGES” to “yes”.
2014.06.001: This version provides a new API and ABI for the ELPA library: ELPA-2 kernels do not have to be specified at build-time anymore. Instead, all kernels possible for the build machine are packed into the library and the user can choose at run-time the desired kernel from all available kernels. Also, the library does not abort on error anymore, but rather returns an error flag which the calling program can catch and act upon it. As always, the functionality of all previous versions is included
2013.11.008: This version provides a hybrid MPI/OpenMP parallelisation. The functionality of all previous versions is included.
2013.08.006: This versions, adds the support of several ELPA-2 kernels, which are optimized for different architectures.
2011.12.002: First release of the ELPA library