[Mesa-dev] [PATCH 07/15] i965/blorp: Use 16 pixel dispatch on Gen7 when there is a WM program.

Paul Berry stereotype441 at gmail.com
Wed May 23 12:12:29 PDT 2012


On 23 May 2012 11:15, Chad Versace <chad.versace at linux.intel.com> wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On 05/23/2012 10:22 AM, Kenneth Graunke wrote:
> > On 05/23/2012 07:45 AM, Paul Berry wrote:
> >> On 22 May 2012 13:02, Eric Anholt <eric at anholt.net
> >> <mailto:eric at anholt.net>> wrote:
> >>
> >>     On Fri, 11 May 2012 11:03:50 -0700, Paul Berry
> >>     <stereotype441 at gmail.com <mailto:stereotype441 at gmail.com>> wrote:
> >>      > When executing a blorp operation on Gen7 that does not need a WM
> >>      > program (i.e. a HiZ operation), we use 32-pixel dispatch mode
> because
> >>      > it's faster.  However, when executing a blorp operation that does
> >>     need
> >>      > a WM program, we need to use 16-pixel dispatch, because blorp WM
> >>      > programs are compiled assuming 16-pixel dispatch.
> >>
> >>     I didn't think dispatch mode means anything unless you're actually
> >>     dispatching a thread (though I could see how having no dispatch
> mode set
> >>     might anger things).  Basically, I'm thinking just always set 16,
> unless
> >>     there is some efficiency reason I don't know of.
> >>
> >>
> >> I don't honestly know.  Chad, I inherited the 32-pixel dispatch from
> >> your HiZ code, and assumed you did it on purpose because it was faster.
> >> But I didn't actually do any tests.  Do you have any thoughts about
> this?
> >
> > I recall Chad saying that not setting thread dispatch broke horribly, so
> since he had enable thread dispatch, he picked the widest mode.
> >
>
> What Ken said.
>

Thanks, Chad.  I'll send out a v2 patch that just uses 16-wide dispatch
unconditionally.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20120523/fc29b84d/attachment.html>


More information about the mesa-dev mailing list