[Mesa-dev] [PATCH 4/4] intel: Allocate s8_z24 non-texture renderbuffers when using separate stencil

[Chad Versace]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Sun, 19 Jun 2011 22:58:18 -0700, Eric Anholt <eric at anholt.net> wrote:
Non-text part: multipart/signed
> On Sun, 19 Jun 2011 14:51:28 -0700, Chad Versace <chad at chad-versace.us> wrote:
> > On Sun, 19 Jun 2011 10:15:49 -0700, Eric Anholt <eric at anholt.net> wrote:
> > Non-text part: multipart/signed
> > > On Fri, 17 Jun 2011 15:43:43 -0700, Chad Versace <chad at chad-versace.us> wrote:
> > > > Now all infrastructure is in place to support s8_z24 non-texture
> > > > renderbuffers for gen7.
> > > > 
> > > > Signed-off-by: Chad Versace <chad at chad-versace.us>
> > > > ---
> > > >  src/mesa/drivers/dri/intel/intel_fbo.c |   84 ++++++++++++++++++++++++++++++-
> > > >  1 files changed, 81 insertions(+), 3 deletions(-)
> > > > 
> > > > diff --git a/src/mesa/drivers/dri/intel/intel_fbo.c b/src/mesa/drivers/dri/intel/intel_fbo.c
> > > > index 0d49a55..72dcf8f 100644
> > > > --- a/src/mesa/drivers/dri/intel/intel_fbo.c
> > > > +++ b/src/mesa/drivers/dri/intel/intel_fbo.c
> > > > @@ -107,6 +107,12 @@ intel_get_pointer(struct gl_context * ctx, struct gl_renderbuffer *rb,
> > > >  }
> > > >  
> > > >  
> > > > +static struct gl_renderbuffer *
> > > > +intel_create_renderbuffer_unwrapper(struct gl_context * ctx,
> > > > +				    struct gl_renderbuffer *wrapper,
> > > > +				    gl_format format);
> > > > +
> > > 
> > > I'd probably call this intel_create_wrapped_renderbuffer now.
> > 
> > Oops... thanks for catching that. I forgot to rename this func.
> > 
> > > 
> > > > @@ -196,6 +204,38 @@ intel_alloc_renderbuffer_storage(struct gl_context * ctx, struct gl_renderbuffer
> > > >  				       width,
> > > >  				       height / 2,
> > > >  				       GL_TRUE);
> > > > +   } else if (irb->Base.Format == MESA_FORMAT_S8_Z24
> > > > +	      && intel->has_separate_stencil) {
> > > 
> > > itym must_use_separate_stencil?
> > 
> > I meant has_separate_stencil.
> > 
> > Gen5 and gen6 will *eventually* have:
> >      .has_hiz = true
> >      .has_separate_stencil = true
> >      .must_use_separate_stencil = false
> > 
> > Gen7 *now* has:
> >      .has_hiz = false
> >      .has_separate_stencil = true
> >      .must_use_separate_stencil = true
> 
> I got that part, I just didn't actually expect us to go separate stencil
> when asked for S8Z24 unless the hardware has no other option.  I guess
> it's down to whether the extra memory bw for separate stencil ends up
> being more or less than that saved by hiz including the cost of resolves
> -- my first guess would have been "no", but I don't actually have
> data.

Ah, I didn't consider the possible bandwidth savings there. My guess
here is "no" too, so I'll change the condition to
must_use_separate_stencil. We can change it later if empirical data
makes it case.

- ----
Chad Versace
chad at chad-versace.us
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)

iQIcBAEBAgAGBQJN/4GEAAoJEAIvNt057x8irB4P/2BGrAkSpcFhgq1GIodDdudw
M8SzpwtUjmJYZ/XSuaM4dtPa7SLeDgO3InjThYt/CrzTs2fsGncCjS7xbgB3EInL
eIn3d84k4f48r5Y6VZ38gaccIKw3ymHET9eKt4L88s0+14Ztciwt0AP1VuMZizRY
UuL9k3kw4BNLrMst82xQOvr4mQayGYH+4NYVankTnQ8v6X2PIBXMf10pcnqXSVnb
Qu3SBPDOHFBDZdEFF3Ph+liakzolM97lZgHHq2tx94UNBTgEwhFQmIMVEfc4UnZA
hC8WF2mL9Evwt/8ssqU0RBiCUUpq+iBWyKFlNlliRKbxtWPxZn6fyjw542uU1jGy
Gp6hUfuEyMsMCE0s//pU7+UAQP71KrXKR9K8ET8o/aoSrpjIboO5oZ5O1sB7xGaW
3ZiVAZZ/TaJgDmW88kyvb9GKNFUcOz4J956SDmbAJ4WchOqrbgUl6+xB/jJhwzZP
qXbDqx038Z2aV1RcSLOior48o3AiGz91YdivzKYeP9vhqg8HJx4QVcHrVXgfZrhC
8cUnO504NYGnFMemHRQSz1ccDeZXXopUnBSUP3Khli1BwgV5u6XQwzrprXrFdCik
M7N1ms6YKEWQenK0b6Y40P5/pTuUfdYMomwkyhtVkeROLwUaSeLMPwTD6cZS9QwN
lMF/isQQRAAc4MOikH0z
=YUoH
-----END PGP SIGNATURE-----