Bug#767138: libfftw3 SIGILL on armhf

Edmund Grimley Evans edmund.grimley.evans at gmail.com
Wed Nov 19 20:37:15 UTC 2014


Kamal Mostafa <kamal at debian.org>:

> Edmund et al., I would appreciate your expertise here... Does it also appear to
> you that my minimodem (0.20-1) failure[0] could be caused by the fftw3 "NEON is
> perhaps broken" bug?   Does it make sense that abel.d.o would be affected by
> it, but harris and asachi would not?

Finding out which armhf machines have NEON is depressingly difficult,
but I think that abel ("Marvell Armada 370/XP CPU @ 1.6GHz on a
Marvell MV78460 SoC Development Board (ARM v7)") doesn't have NEON and
the others do, so it makes sense.

> All: Should the severity of the libfftw3 bug #767138 be elevated to "serious"?

I'd say it's serious.

And I'd recommend throwing out the signal-catching code and just using
auxv.

Edmund



More information about the debian-science-maintainers mailing list