Bug#1126688: openblas causes slepc4py test failure on ppc64el: Error in LAPACK subroutine gesvd: info=127
Drew Parsons
dparsons at debian.org
Fri Jan 30 17:51:42 GMT 2026
Source: openblas
Version: 0.3.30+ds-3
Severity: normal
X-Debbugs-Cc: debian-powerpc at lists.debian.org, Trupti <trupti at linux.ibm.com>
User: debian-powerpc at lists.debian.org
Usertags: ppc64el
Control: affects -1 src:slepc4py
slepc4py 3.24.2-1 has been failing ppc64el tests in testing, while
bizarrely passing in unstable.
unstable passes: https://ci.debian.net/packages/s/slepc4py/unstable/ppc64el/68218387/
testing fails: https://ci.debian.net/packages/s/slepc4py/testing/ppc64el/68217024/
The error comes from LAPACK:
225s python3.13 ex10.py
...
225s Solve the problem with POD
225s The L2-norm of the error is: 4.323608292134194
225s NORMAL END
225s python3.13 ex11.py
226s 2-D Laplacian Eigenproblem solved with contour integral, N=1024 (32x32 grid)
226s
226s petsc4py.PETSc.Error: error code 76
226s [0] EPSSolve() at /build/reproducible-path/slepc-3.24.2+dfsg1/src/eps/interface/epssolve.c:162
226s [0] EPSSolve_CISS() at /build/reproducible-path/slepc-3.24.2+dfsg1/src/eps/impls/ciss/ciss.c:493
226s [0] BVSVDAndRank() at /build/reproducible-path/slepc-3.24.2+dfsg1/src/sys/classes/bv/interface/bvcontour.c:556
226s [0] BVSVDAndRank_Refine() at /build/reproducible-path/slepc-3.24.2+dfsg1/src/sys/classes/bv/interface/bvcontour.c:349
226s [0] Error in external library
226s [0] Error in LAPACK subroutine gesvd: info=127
226s
226s The above exception was the direct cause of the following exception:
226s
226s Traceback (most recent call last):
226s File "/tmp/autopkgtest-lxc.igtag2q7/downtmp/build.LE7/src/test-demos-py3.13/ex11.py", line 83, in <module>
226s main()
226s ~~~~^^
226s File "/tmp/autopkgtest-lxc.igtag2q7/downtmp/build.LE7/src/test-demos-py3.13/ex11.py", line 75, in main
226s E.solve()
226s ~~~~~~~^^
226s SystemError: <cyfunction EPS.solve at 0x3ffe9018cdc0> returned a result with an exception set
226s make: Leaving directory '/tmp/autopkgtest-lxc.igtag2q7/downtmp/build.LE7/src/test-demos-py3.13'
226s make: *** [makefile:13: run_ex11] Error 1
Comparing the two logs, the difference that stands out is that
unstable uses generic liblapack3 3.12.1-7+b1, and passes,
while testing uses openblas libopenblas0-pthread 0.3.30+ds-3+b1, and
fails.
My tentative conclusion is that liblapack.so from libopenblas0-pthread
is generating the reported error 127 in gesvd
It's not clear to me why testing uses openblas while unstable uses
generic lapack, but it's formally permitted by alternative libpetsc3.24-dev-common
Depends: liblapack-dev | libopenblas-dev | libatlas-base-dev | liblapack.so
which provides the lapack dependency for slepc.
There was a previous recent ppc64el-specific bug in openblas, Bug#1121177 (also #1114016).
Trupti was helpful in triaging that bug, so I'm cc:ing you here.
There have been some recent ppc64 patches in openblas upstream,
perhaps they address the issue
https://github.com/OpenMathLib/OpenBLAS/pull/5628
see also https://github.com/OpenMathLib/OpenBLAS/issues/5627
But not clear if that's the same issue though.
More information about the debian-science-maintainers
mailing list