Bug#799905: gcc-4.7: generates broken SSE2 code for -ftree-vectorize/-O3 for unaligned dword access
On Thu, 24 Sep 2015, Matthias Klose wrote:
> which is EOL upstream for more than a year, and which targets non-default
> flags. You don't even try to reproduce with current GCC versions in
-O3 is not exactly a "uncommon" flag. I just tracked it down to
-ftree-vectorize (enabled by -O3) to _help_.
As for -msse2, that's DEFAULT on amd64/x86-64. I listed it explicitly so
that I can also check for the very same issue on i686 (the Pentium M from
2005 already has SSE2).
> unstable/testing. Feel free to supply better information, testing with
> gcc-snapshot, and then maybe reporting this upstream. Not spending any time
> on this myself.
I figured I didn't have to add a sentence "I will test stable and unstable
tomorrow" since I was going to reply within 24h... evidently, I was wrong.
It still exists on gcc-4.9 in stable (gcc Debian 4.9.2-10). I am going to
test on an unstable chroot and gcc-snapshot in a few moments.
Also, since it triggers on -O3 on amd64/x86-64, it potentially affects a lot
of packages in Debian (we have lots of packages doing -O3). We are lucky
that unaligned 32-bit access is not something people do just for fun, OTOH,
it can easily be input-dependent, so the bug might not trigger always...
--
"One disk to rule them all, One disk to find them. One disk to bring
them all and in the darkness grind them. In the Land of Redmond
where the shadows lie." -- The Silicon Valley Tarot
Henrique Holschuh
Reply to: