[Mesa-dev] [PATCH] anv: remove support for d32_sfloat_s8_uint

[Jason Ekstrand]

On Fri, Aug 5, 2016 at 10:23 AM, Lionel Landwerlin <
lionel.g.landwerlin at intel.com> wrote:

> On 05/08/16 18:08, Jason Ekstrand wrote:
>
> On Aug 5, 2016 9:49 AM, "Lionel Landwerlin" <lionel.g.landwerlin at intel.com>
> wrote:
> >
> > It doesn't seem like we have any support for 32bits of something + 8bits
> of
> > something else.
>
> Not quite. Depth-stencil formats are weird because they describe two
> different things at the same time. In our hardware (post Sandy Bridge) the
> depth and stencil components are actually stored in different surfaces but
> in the same VkImage.  We certainly can handle 32-bit depth with stencil, we
> just need to get the blits working.Th
>
>
> Thanks for the explanation. Would that be ISL_FORMAT_R32_FLOAT_X8X24_TYPELESS
> then?
>

No.  Back in the old days (Sandy Bridge and earlier) the hardware supported
combined depth-stencil buffers where the depth and stencil were stored
together in the same surface with each pixel containing both depth and
stencil components.  Iron Lake and Sandy Bridge supported both combined
depth-stencil separate stencil.  On Ivy Bridge and later, only separate
stencil is supported.  With separate stencil, the depth is stored as
R32_FLOAT and the stencil is stored as R8_UINT in separate surfaces.  For
depth-stencil formats, the format table only stores the depth format so we
really do want R32_FLOAT.  The stencil part of the format is handled by
anv_get_isl_format.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20160805/60c4e2c1/attachment.html>