Le samedi 29 décembre 2018 à 23:53 +0100, Lucas Nussbaum a écrit : > Source: octave-interval > Version: 3.2.0-3 > Severity: serious > Justification: FTBFS on amd64 > Tags: buster sid
> During a rebuild of all packages in sid, your package failed to build
> on amd64.
> Relevant part (hopefully):
> > [src/mpfr_matrix_mul_d.cc]
> > > > > > > /<<PKGBUILDDIR>>/src/mpfr_matrix_mul_d.cc
> >
> > ***** test;
> > [l, u] = mpfr_matrix_mul_d (magic (3), magic (3)', magic (3) + 1,
> > magic (3)' + 1);
> > assert (l, [101, 71, 53; 71, 83, 71; 53, 71, 101]);
> > assert (u, [134, 104, 86; 104, 116, 104; 86, 104, 134]);
> > ***** test;
> > if (((exist ("__octave_config_info__") && ...
> > __octave_config_info__.ENABLE_OPENMP) || ...
> > (not (exist ("__octave_config_info__")) && ...
> > octave_config_info ("features").OPENMP)) ...
> > && nproc ("overridable") > 1)
> > ## OpenMP is enabled and may use more than one thread.
> > ## We should observe speed-ups by that.
> > A = vec (magic (2000));
> > tic;
> > mpfr_matrix_mul_d (A', A, A', A);
> > time_per_element = toc;
> > A = [A A];
> > tic;
> > mpfr_matrix_mul_d (A', A, A', A);
> > time_per_element_parallel = toc / numel (ans);
> >
> > ## Ideally, the runtime would be cut in half (at least),
> > ## however, there is an overhead, so let's be pessimistic and
> > ## assume that the parallel execution is at least 20% faster.
> > assert (time_per_element > 1.2 * time_per_element_parallel)
> > endif
> > !!!!! test failed
> > assert (time_per_element > 1.2 * time_per_element_parallel) failed
> > 2 tests, 1 passed, 0 known failure, 0 skipped
The problem comes from the 2nd test in mpfr_matrix_mul_d.cc, which
fails if the buildd is heavily loaded (because the speedup of
parallelization does not materialize).
I think this test should be patched out for Debian.
Oliver, what do you think?
--
⢀⣴⠾⠻⢶⣦⠀ Sébastien Villemot
⣾⠁⢠⠒⠀⣿⡁ Debian Developer
⢿⡄⠘⠷⠚⠋⠀ http://sebastien.villemot.name
⠈⠳⣄⠀⠀⠀⠀ http://www.debian.org
signature.asc
Description: This is a digitally signed message part
