On Wed, 4 Apr 2018 21:13:43 +0300 Jussi Pakkanen <jpakk...@gmail.com> wrote: > On Wed, Apr 4, 2018 at 10:25 AM, Matthias Klose <d...@debian.org> wrote: > > > meson fails it's autopkg tests on i386. simd is not in the base architecture of > > the i386 target, so you probably should skip that test. > > There are two things to note here: > > 1) the test is written so that it compiles all the things that the > compiler supports but then, at runtime, checks what the current > processor supports and runs that. If the current processor does not > have any SIMD support then it falls back to plain C code. The test > should run on every platform and CPU regardless of whether they have > SIMD or not. > > 2) as part of Meson's release gating, I run the full package build and > tests for Debian using a sid i386 pbuilder (under VMware currently). > The tests have passed on my machine. It does work as expected, at > least under some circumstances. >
is it just i386 pbuilder, or actually an i686 kernel? because it segfaults for us when running on an i686 kernel, in an openstack cloud VM. it appears to be an alignment issue, as identified by mwhudson, and affects most simd implementation in the test. https://patch-diff.githubusercontent.com/raw/mesonbuild/meson/pull/4482.patch fixes it for us, and tests that all simd variants can correctly execute on an i686 kernel. Regards, Dimitri.
