[Bug middle-end/27770] [4.2 Regression] wrong code in spec tests for -ftree-vectorize -maltivec

[richard at codesourcery dot com]

------- Comment #18 from richard at codesourcery dot com  2006-07-02 13:39 -------
Subject: Re:  [4.2 Regression] wrong code in spec tests for -ftree-vectorize
-maltivec

"hubicka at gcc dot gnu dot org" <gcc-bugzilla@gcc.gnu.org> writes:
> Increasing alignment of static variable during tree optimization should be as
> easy as increasing the DECL_ALIGN. All variables are output after all
> functions.

This is what already happens.  The problem is _when_ it happens.

In the testcase, we first compile a simple function that has no
vectorisable parts.  The function accesses a variable and we decide to
use section anchors to address it.  We therefore place the variable in
the section based on its current DECL_ALIGN.

We then compile a different function that has vectorisable accesses
to the same variable.  We increase the variable's DECL_ALIGN accordingly.
This new DECL_ALIGN throws off the assumptions made for the first function.

The fix we've agreed is best in principle is to speculatively increase
the DECL_ALIGN of vectorisable variables before compiling functions.
Dorit says that there is a patch related to this on the autovect branch,
which I'll look at when I get back from Ottawa.

Richard


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27770