This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug ipa/65334] [5 Regression] r221099 caused: FAIL: gfortran.fortran-torture/execute/in-pack.f90 execution, -O2 -ftree-vectorize -msse2
- From: "hubicka at ucw dot cz" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: Fri, 06 Mar 2015 23:36:28 +0000
- Subject: [Bug ipa/65334] [5 Regression] r221099 caused: FAIL: gfortran.fortran-torture/execute/in-pack.f90 execution, -O2 -ftree-vectorize -msse2
- Auto-submitted: auto-generated
- References: <bug-65334-4 at http dot gcc dot gnu dot org/bugzilla/>
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65334
--- Comment #3 from Jan Hubicka <hubicka at ucw dot cz> ---
This is patch I am testing. It turns out that vectorizer does update of
alignment twice and both are wrong in aliases in different ways.
In increase_alignment it decides to bump up alignment of alias target that is
unsafe because target may be interposable while alias is not. Bumping up
DECL_ALIGN will make us to produce invalid aligned accesses to the interposable
references.
In vectorizer itself it updates only DECL_ALIGN of alias and ignore the fact
that target definition needs to be aligned.
This patch moves the heavy lifting into symbol table and adds extra abstraction
around alignments.
Honza