Bug ID 1112768
Summary science/openblas: request clean up openblas update-alternatives
Classification openSUSE
Product openSUSE.org
Version unspecified
Hardware Other
OS Other
Status NEW
Severity Normal
Priority P5 - None
Component 3rd party software
Assignee dmitry@roshchin.org
Reporter chrisbcoutinho@gmail.com
QA Contact bnc-team-screening@forge.provo.novell.com
Found By ---
Blocker ---

I think the BLAS/LAPACK alternatives are inconsistent. I'm not sure if this is
necessarily a bug or not - some guidance would be appreciated.

I have multiple versions of the BLAS/LAPACK libraries installed on my system,
and I'm having trouble selecting the right ones using `update-alternatives`. I
think the alternatives to the BLAS/LAPACK libraries could be cleaned up a bit.

For example, /usr/lib64/libblas.so doesn't point to /usr/lib64/libblas.so.3 (or
vice-versa), which causes problems if the two are not in sync. The basic BLAS
library /usr/lib64/libblas.so points to /usr/lib64/blas/libblas.so.3.8.0,
whereas /usr/lib64/libblas.so.3 is an alternative that can be set to any number
of BLAS implementations. Instructions can be seen on the wiki regarding
libblas.so.3:

https://en.opensuse.org/openSUSE:Science_Linear_algebra_libraries

I'm able to set libblas.so.3 using update-alternatives, but that change is not
seen by cmake because of the precedence it uses to find BLAS/LAPACK (namely
defaulting to /usr/lib64/openblas.so).

Further, when LAPACK is required using cmake and openblas is installed, LAPACK
and BLAS both default to /usr/lib64/libopenblas.so (based on
/usr/lib64/cmake/openblas/OpenBLASConfig.cmake), but that in turn points to an
alternative for openblas_pthreads.so.0 (why?). Shouldn't it be the other way
around, where /usr/lib64/openblas.so is the alternative with options for
openblas_{pthreads,openmp}.so? That way even custom (open)blas installations
could be set using update-alternatives and picked up by other software using
cmake by default.

What is the recommended way to cleanly switch between BLAS/LAPACK
implementations?


You are receiving this mail because: