This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [4.5] Better support for widening multiplies
- From: Dave Korn <dave dot korn dot cygwin at googlemail dot com>
- To: Bernd Schmidt <bernds_cb1 at t-online dot de>
- Cc: Dave Korn <dave dot korn dot cygwin at googlemail dot com>, GCC Patches <gcc-patches at gcc dot gnu dot org>
- Date: Mon, 16 Feb 2009 23:26:33 +0000
- Subject: Re: [4.5] Better support for widening multiplies
- References: <4999B411.6010902@t-online.de> <4999CE9C.2020508@gmail.com> <4999E204.4040808@t-online.de>
Bernd Schmidt wrote:
> Dave Korn wrote:
>> Bernd Schmidt wrote:
>>> This adds another small pass
>> Is this a win for all backends [...]
>
> Most of them would be my guess:
>
> $ grep -l '^.define.*mul[qhs]i[hsd]i' config/*/*md
That's a fairly persuasive argument you have there, but I suppose it doesn't
tell us anything about how well utilised those patterns already are? Is it
possible that on other backends than bfin, they might scoop up most of the
available opportunities already, by other means?
It's not clear to me if the problem that you identified on bfin arises to
the same extent elsewhere. (It would be great if you could somehow instrument
your testing to show the gains when you do the linux bootstrap/testing, but I
guess that would add a lot of work.) Do you have a testcase that would show a
missed optimisation opportunity on i386, for example?
cheers,
DaveK