Re: Bug#1121177: openblas breaks multiple autopkgtest on ppc64el

Sébastien Villemot <[email protected]> Wed, 14 Jan 2026 20:05:26 +0100

Newsgroups	gmane.linux.debian.ports.powerpc
Organization	Debian
Message-ID	<[email protected]>

Le mercredi 07 janvier 2026 à 23:35 +0530, Trupti a écrit :
> On 2026-01-07 11:29, Simon Richter wrote:
> > On 1/7/26 5:57 AM, Sébastien Villemot wrote:
> > 
> > > If you have the time, could you possibly also check that the two other
> > > autopkgtest regressions (in src:gemma and src:openmolcas) are also
> > > tolerance-related? (see https://tracker.debian.org/pkg/openblas for 
> > > the
> > > list of autopkgtest regressions)
> 
> For src:gemma, the autopkgtest failure on ppc64el occurs during the 
> eigen-decomposition step.
> The run reports a warning about many eigenvalues close to zero, followed 
> by an LU decomposition failure in GSL/LAPACK.
> The failure is triggered in the following code path:
> 
> // LU decomposition.
> void LUDecomp(gsl_matrix *LU, gsl_permutation *p, int *signum) {
>    // debug_msg("entering");
>    enforce_gsl(gsl_linalg_LU_decomp(LU, p, signum));
>    return;
> }
> 
> 
> 
> For src:openmolcas, the autopkgtest failures on ppc64el are limited to 
> CASPT2 tests (standard tests 009, 010 and hdf5 test 601).
> The logs show floating-point exceptions (IEEE invalid, divide-by-zero, 
> underflow) followed by CASPT2 convergence failures (_NOT_CONVERGED_ / 
> _INTERNAL_ERROR_).
> All non-CASPT2 tests complete successfully.  The CASPT2 output itself 
> indicates numerical instability and suggests increasing 
> linear-dependence thresholds.

Thanks Trupti.

Do you consider the following as a good summary of your analysis: there
is no structural problem in the new OpenBLAS on ppc64el (just slightly
numerically different results, within the usual tolerance of numerical
software), and as a consequence the adjustment needs to be done in the
testsuite of the affected reverse dependencies (xtensor-blas, gemma and
openmolcas)? (that seems clear from what you said of src:xtensor-blas
and src:openmolcas, less so for the case of src:gemma, hence my
question)

Paul: if my statement above is correct, what would be the right course
of action?

-- 
⢀⣴⠾⠻⢶⣦⠀  Sébastien Villemot
⣾⠁⢠⠒⠀⣿⡁  Debian Developer
⢿⡄⠘⠷⠚⠋⠀  https://sebastien.villemot.name
⠈⠳⣄⠀⠀⠀⠀  https://www.debian.org

signature.asc (application/pgp-signature, 833 B)