[Mesa-dev] [RFC 1/6] dri: Support 64 bit rgba masks
Strasser, Kevin
kevin.strasser at intel.com
Wed Jan 16 21:11:10 UTC 2019
Adam Jackson wrote:
> On Wed, 2019-01-16 at 17:58 +0000, Strasser, Kevin wrote:
> > Adam Jackson wrote:
> >
> > > Probably what I'd do is only define those masks to non-zero for <=32bpp
> > > formats,
> >
> > Not sure if I understood this statement. Do you mean if we keep the _HI and
> > _LO attributes, only make them valid for formats that are <=64bpp, like
> > fp16?
>
> I was thinking:
>
> a) Leave __DRI_ATTRIB_XXX_MASK in place, and simply set it to zero for
> configs that don't fit in 32bpp
>
> b) New attribute(s) for __DRI_ATTRIB_XXX_SHIFT, whose value is the
> right-hand side of the << operator you'd use to find that channel
>
> c) Fix dri2_add_config() and friends to work in terms of shifts not
> masks
>
> This nicely avoids worrying about just how wide a pixel will ever get
> (ie how many _HI attribs you'd need for fp32, fp64, etc), because the
> left-shift is a whole 32-bit integer, and we're almost certainly not
> worried about needing a billion bits per channel. The only other
> flexibility masks would give you is allowing discontiguous bit ranges
> for color channels, but that's not a feature of any hardware we do or
> will care about (and I'm not sure was ever really a thing, tbh).
>
> Setting the MASK attribs to zero appears to magically DTRT when loading
> a newer driver (with fp16 format support) on an older libEGL or
> xserver; since the masks of the winsys' visuals will be non-zero, fp
> formats will never match.
OK, thanks for explaining, I think I get it now. I'll see if I can get this
working.
Thanks,
Kevin
More information about the mesa-dev
mailing list