[opensuse-factory] BLAS implementations
BLAS (Basic Linear Algebra Subprograms) is a library that provides standard blocks for performing basic vector and matrix operations. Currently openSUSE uses reference BLAS/LAPACK implementations from netlib by default. This versions are very stable, but not fast. There are optimized implementations: opensource - openBLAS, Atlas and proprietary - Intel MKL, AMD CML. openSUSE repositories contain only openBLAS. We have update-alternatives mechanism for choosing BLAS implementations: /usr/sbin/update-alternatives --config libblas.so.3 All implementations have the same API/ABI, so user can choose system BLAS on local machine easily. And user can use proprietary BLAS library. We don't have policy for package maintainers. Most part of packages use default reference implementation, but some scientific packages are configured to use openBLAS directly: opencv, suitesparce, scipy. And only a little amount of openSUSE users know about update-alternatives configuration. Unfortunately openBLAS not so stable and contains bugs. Additionally we still have problems with AMD CPUs: https://bugzilla.opensuse.org/show_bug.cgi? id=1006743 Are there any ideas about the best solution? Do we need to use reference BLAS in all cases or maintainers can use openBLAS directly? -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org
participants (1)
-
Dmitry Roshchin