Bug#799905: gcc-4.7: generates broken SSE2 code for -ftree-vectorize/-O3 for unaligned dword access
Matthias,
Given the new information I just sent to the bug report, and since all
current versions of gcc have this issue, it should not remain as "wishlist"
in gcc-4.7 only.
I have tagged it "upstream" for now, and removed the "moreinfo" tag.
Should I raise severity back to important, or to normal? Should I mark the
other gcc packages as also affected by this bug (might require cloning the
bug)?
codesearch.debian.org reports lots of packages potentially using -O3 in
CFLAGS: approximately 1800 packages, and some of them are present on nearly
every Debian system and very important (like perl5), or extremely security
sensitive (like krb5).
Not all of these packages will actually build with -O3, and I assume the
vast majority of them tries to stay away from unaligned accesses, anyway.
Still, this crash-on-unaligned issue is quite unexpected on x86 and amd64,
and I can easily imagine it being a boundary condition for an application
that would only trigger in specially-crafted/corrupt input files.
As for forwarding the report upstream myself, I never dealt with gcc
upstream before. If you don't feel it might be better if someone in the
Debian gcc team did it, I will do it.
--
"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: