[PATCH 4/5] wined3d: Pass depth and stencil separately down the CPU fill path.

[Zebediah Figura]

On 5/22/22 04:16, Stefan Dösinger wrote:
> From: Stefan Dösinger <stefan at codeweavers.com>
> 
> Signed-off-by: Stefan Dösinger <stefan at codeweavers.com>
> 
> ---
> 
> Love it or leave it. Shouldn't affect the test behind it.
> 
> Eventually we can merge depth formats into the UNORM or FLOAT codepaths,
> but right now the UNORM flag is specific to color formats. I'm also not
> sure how we'd ever realistically clear a depth buffer to anything other
> than zero in the CPU codepath. Tests show that depth locking is
> unreliable, and even DEPTHFILL value vs actual Z test value in d3d7 is
> vendor specific.
> 
> Merging depth-unorm and color-unorm as well as depth-float and
> color-float handling works without this patch too. Just multiply and OR
> color.r with mask_to_size(red) and mask_to_size(depth), as long as we
> don't encounter a format that has both red and depth channels.
> ---
>   dlls/wined3d/device.c          |  4 +--
>   dlls/wined3d/resource.c        |  8 +++---
>   dlls/wined3d/surface.c         | 12 +++++----
>   dlls/wined3d/texture.c         | 19 ++++----------
>   dlls/wined3d/utils.c           | 47 +++++++++++++++++++---------------
>   dlls/wined3d/wined3d_private.h |  5 ++--
>   6 files changed, 48 insertions(+), 47 deletions(-)
> 

I don't dislike this, although on a second look I think using the union 
would work just as well? Note that because a format is either float or 
depth-stencil (and we have zero multipliers accordingly) we don't 
actually need to check the depth/stencil size in 
wined3d_format_convert_from_float(). Otherwise this patch would be 
broken wrt wined3d_texture_load_location().