This is the mail archive of the
mailing list for the GCC project.
Re: [lno] [patch] vectorizer update - loop bound
- From: Dorit Naishlos <DORIT at il dot ibm dot com>
- To: Daniel Berlin <dberlin at dberlin dot org>
- Cc: gcc-patches at gcc dot gnu dot org, pop at gauvain dot u-strasbg dot fr, Toon Moene <toon at moene dot indiv dot nluug dot nl>
- Date: Tue, 20 Jan 2004 19:10:47 +0200
- Subject: Re: [lno] [patch] vectorizer update - loop bound
> I could rewrite the normalization pass if you want, it was done right
> before major SSA changes, and i never got around to rewriting it.
I'm not sure we're talking about the same thing. After reading the
responses from Richard and Paul - do you still think your loop
normalization solves this problem?
<dberlin@dberlin. To: Dorit Naishlos/Haifa/IBM@IBMIL
org> cc: email@example.com, firstname.lastname@example.org, Geoff Keating
<email@example.com>, Toon Moene <firstname.lastname@example.org>
19/01/2004 18:22 Subject: Re: [lno] [patch] vectorizer update - loop bound
> I figured out what the problem was, but now that the vectorizer passes
> stage, it fails on an alignment check, which currently tries to make
> that the array base and the first access to the array are aligned
> (which is
> over conservative on purpose). The first access however is at index 1,
> which leads the vectorizer to conclude that the accesses to this array
> not aligned. Shouldn't the array accesses be normalized to start from 0
> when the Fortran program is translated to the trees? If not, can I
> what was the source language and treat Fortran programs differently
> than C
> (e.g., consider accesses to index i as accesses to index (i-1) in case
I could rewrite the normalization pass if you want, it was done right
before major SSA changes, and i never got around to rewriting it.
My only concern is whether it will make the evolutions too complex to
be "simple" evolutions.