[Mesa-dev] [PATCH 2/7] i965: Allow fast depth clears if scissoring doesn't do anything.

[Oliver McFadden]

On Fri, Sep 28, 2012 at 08:41:51AM -0700, Ian Romanick wrote:
> On 09/19/2012 01:27 PM, Kenneth Graunke wrote:
> > A game we're working with leaves scissoring enabled, but frequently sets
> > the scissor rectangle to the size of the whole screen.  In that case,
> > scissoring has no effect, so it's safe to go ahead with a fast clear.
> >
> > Chad believe this should help with Oliver McFadden's "Dante" as well.
> >
> > v2/Chad: Use the drawbuffer dimensions rather than the miptree slice
> > dimensions.  The miptree slice may be slightly larger due to alignment
> > restrictions.
> 
> There was some question of whether this actually gave any performance 
> improvement or just made the INTEL_DEBUG=perf warning go away.  Can we 
> see any actual performance delta in Dante?

I will try to get you some numbers during this week.  If I have time; my
schedule is fairly packed.

> 
> > Signed-off-by: Kenneth Graunke <kenneth at whitecape.org>
> > Signed-off-by: Chad Versace <chad.versace at linux.intel.com>
> > Cc: Oliver McFadden <oliver.mcfadden at linux.intel.com>
> > ---
> >   src/mesa/drivers/dri/i965/brw_clear.c | 14 +++++++++++++-
> >   1 file changed, 13 insertions(+), 1 deletion(-)
> >
> > diff --git a/src/mesa/drivers/dri/i965/brw_clear.c b/src/mesa/drivers/dri/i965/brw_clear.c
> > index c3c75df..625cc85 100644
> > --- a/src/mesa/drivers/dri/i965/brw_clear.c
> > +++ b/src/mesa/drivers/dri/i965/brw_clear.c
> > @@ -77,6 +77,18 @@ debug_mask(const char *name, GLbitfield mask)
> >   }
> >
> >   /**
> > + * Returns true if the scissor is a noop (cuts out nothing).
> > + */
> > +static bool
> > +noop_scissor(struct gl_context *ctx, struct gl_framebuffer *fb)
> > +{
> > +   return ctx->Scissor.X <= 0 &&
> > +          ctx->Scissor.Y <= 0 &&
> > +          ctx->Scissor.Width >= fb->Width &&
> > +          ctx->Scissor.Height >= fb->Height;
> > +}
> > +
> > +/**
> >    * Implements fast depth clears on gen6+.
> >    *
> >    * Fast clears basically work by setting a flag in each of the subspans
> > @@ -110,7 +122,7 @@ brw_fast_clear_depth(struct gl_context *ctx)
> >       * a previous clear had happened at a different clear value and resolve it
> >       * first.
> >       */
> > -   if (ctx->Scissor.Enabled) {
> > +   if (ctx->Scissor.Enabled && !noop_scissor(ctx, fb)) {
> >         perf_debug("Failed to fast clear depth due to scissor being enabled.  "
> >                    "Possible 5%% performance win if avoided.\n");
> >         return false;
> >
> 

-- 
Oliver McFadden.