This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: PR 11319 analyzed (Re: Patch fixing 3.3 bug PR 9745 and PR 10021)
- From: David Edelsohn <dje at watson dot ibm dot com>
- To: Dale Johannesen <dalej at apple dot com>
- Cc: Jim Wilson <wilson at tuliptree dot org>, Franz Sirl <Franz dot Sirl-kernel at lauterbach dot com>, Mark Mitchell <mark at codesourcery dot com>, Marcus Meissner <meissner at suse dot de>, Michael Matz <matz at suse dot de>, gcc at gcc dot gnu dot org, Olaf Hering <olh at suse dot de>
- Date: Tue, 15 Jul 2003 21:48:19 -0400
- Subject: Re: PR 11319 analyzed (Re: Patch fixing 3.3 bug PR 9745 and PR 10021)
>>>>> Dale Johannesen writes:
Dale> It is not necessarily true that fixing the bug will introduce
Dale> performance regressions.
Dale> If it is the case that the buggy code is moving some memory references
Dale> out of
Dale> loops in cases where the compiler cannot tell that it is safe, and is
Dale> getting lucky,
Dale> then yes, fixing the bug will cause regressions. But I don't think we
Dale> have
Dale> evidence of that. (In practice, I mean - certainly that can happen in
Dale> theory.)
Compile the testcase from PR 9745 dumping out the GCC loop phase
with and without Jim's patch. With Jim's patch about 2/3 of the
replaceable givs and Hoisted references disappear. Those loop
transformations were partially more aggressive loop optimization as Jim
has suggested restoring and partially "getting lucky".
David