[Mesa-dev] [PATCH] nir: Copy-propagate vecN operations that are actually moves
Jason Ekstrand
jason at jlekstrand.net
Fri Feb 20 09:23:31 PST 2015
On Thu, Feb 19, 2015 at 11:03 PM, Matt Turner <mattst88 at gmail.com> wrote:
> On Thu, Feb 19, 2015 at 11:01 PM, Connor Abbott <cwabbott0 at gmail.com>
> wrote:
> > I agree with Ken that the regressions are small enough, and it seems
> > they're mostly stuff we can prevent by being smarter when doing the
> > sel peephole, so it seems like the cleanup that will probably help
> > other passes is worth it.
>
> So, usually we do that as a preparatory patch. Why aren't we doing that
> here?
>
Do what in a preparatory patch? Fix up the sel peephole to be able to
handle "if (foo) bar = baz;"? Sure, I can put that patch together.
> NIR instruction counts is not the metric we care about.
>
No, but cleaning things up means that we can do other optimizations
better. Also, in each of those cases, the non-ssa NIR code was better it
was just less cleanable by the backend. We need to work on that, but I
don't think it's an indicator of a problem.
--Jason
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20150220/ebed8daa/attachment.html>
More information about the mesa-dev
mailing list