This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Your change to expmed.c on 3/13/2002
- From: David Edelsohn <dje at watson dot ibm dot com>
- To: kenner at vlsi1 dot ultra dot nyu dot edu (Richard Kenner)
- Cc: aldy at redhat dot com, gcc at gcc dot gnu dot org
- Date: Mon, 21 Oct 2002 14:44:48 -0400
- Subject: Re: Your change to expmed.c on 3/13/2002
>>>>> Richard Kenner writes:
Richard> This must have gone by when I wasn't looking, but the way it's coded,
Richard> it essentially disables all aliasing detection when there is a
Richard> component reference.
Richard> Can you explain more about exactly how this can occur and why some
Richard> more precise test isn't more appropriate?
If you review the discussion on the GCC mailinglist, you will see
that this patch arose out of the strct-pack-1.c failure. You were
intimately involved in the discussions. See the discussion beginning 13
March 2002.
Yes, GCC *should* do better, but at the moment GCC cannot
represent the more detailed information necessary to do a better job. The
choice was to have something that is non-optimal but works correctly.
Also, any other change was too drastic during that stage of development.
If you want to develop a more correct and complete implementation
for merge during Stage 1 of GCC 3.4 development, I am sure that we all
would appreciate that. You said that you were interested in the issue in
December 2001.
David