This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug target/65369] [5 Regression] nettle test failure on powerpc64le-linux-gnu when built with -O3
- From: "rguenth at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: Fri, 13 Mar 2015 09:45:55 +0000
- Subject: [Bug target/65369] [5 Regression] nettle test failure on powerpc64le-linux-gnu when built with -O3
- Auto-submitted: auto-generated
- References: <bug-65369-4 at http dot gcc dot gnu dot org/bugzilla/>
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65369
--- Comment #30 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Alan Modra from comment #28)
> Created attachment 35024 [details]
> modified testcase without bswap optimization
>
> This modified testcase avoids triggering the bswap optimization but still
> shows a failure at -O3. So definitely not a problem caused by Thomas' patch.
>
> -O3 -fno-tree-slp-vectorize is OK
>
> -O3 slp dump shows weird offset of +12 between vector loads rather than +16
> as is usual
Well, that's the realign-load sequence. Load ptr & ~15 and then load (ptr +
12) & ~15. If ptr is already aligned both loads should load from the same
location.
SLP happens in md4_update only (but it's quite trivial).