Re: [PATCH 2/7] wined3d: Use ARB_internalformat_query2 to check for texture format rendering and blending support, where available.

[Henri Verbeet]

On 18 April 2013 21:19, Matteo Bruni <mbruni(a)codeweavers.com> wrote:
> +            if (!format->glInternal) continue;
> +
> +            gl_info->gl_ops.ext.p_glGetInternalformativ(GL_TEXTURE_2D, format->glInternal,
> +                    GL_FRAMEBUFFER_RENDERABLE, 1, &value);
> +            if (value == GL_FULL_SUPPORT)
> +            {
> +                TRACE("Format %s is supported as FBO color attachment.\n", debug_d3dformat(format->id));
> +                format->flags |= WINED3DFMT_FLAG_RENDERTARGET | WINED3DFMT_FLAG_FBO_ATTACHABLE;
> +                format->rtInternal = format->glInternal;
> +
> +                if (!(format->flags & (WINED3DFMT_FLAG_DEPTH | WINED3DFMT_FLAG_STENCIL)))
> +                    query_format_flag(gl_info, format, format->glInternal, GL_FRAMEBUFFER_BLEND,
> +                            WINED3DFMT_FLAG_POSTPIXELSHADER_BLENDING, "post-pixelshader blending");
> +            }

This may be a bit dangerous, since you're now potentially adding
WINED3DFMT_FLAG_RENDERTARGET for formats like e.g.
WINED3DFMT_R8G8B8A8_SNORM that didn't previously had it, and where
it's a bit questionable if it's really going to work as intended.
Also, the depth/stencil check that follows is either redundant, or
we're now also potentially setting WINED3DFMT_FLAG_RENDERTARGET for
depth/stencil formats. At least initially we'll probably want to stick
with formats that already have WINED3DFMT_FLAG_RENDERTARGET set, and
only clear it if unsupported. At a later point we may want to do
something more centered around the information we get from
internalformat_query2, but I suspect it will be a bit harder than just
checking if there's no fixup, no conversion function, etc.

[PATCH 2/7] wined3d: Use ARB_internalformat_query2 to check for texture format rendering and blending support, where available.

[Matteo Bruni]

Hrm, I sent the wrong version of the patch, I had actually moved the
depth/stencil check earlier (so yeah, totally agree on that part).
Also you're right about WINED3DFMT_FLAG_RENDERTARGET, for some reason
I thought we were setting that in check_fbo_compat() but it's clearly
not the case.

I'm sending the updated patches in a bit (also dropping patches 5-7
since they are superseded by yours).