Re: [PATCH 12/12] wined3d: Use ARB_fragment_coord_conventions for vpos fixups when supported.

[Henri Verbeet]

On 20 May 2016 at 19:56, Matteo Bruni <mbruni(a)codeweavers.com> wrote:
> if (reg_maps->vpos)
>      {
> -        if (shader->device->wined3d->flags & WINED3D_PIXEL_CENTER_INTEGER)
> +        if (gl_info->supported[ARB_FRAGMENT_COORD_CONVENTIONS])
> +            shader_addline(buffer, "vpos = gl_FragCoord;\n");
> +        else if (shader->device->wined3d->flags & WINED3D_PIXEL_CENTER_INTEGER)

Isn't it harder than that? You're dropping the y-flip completely now.
I think we can actually do that regardless of
ARB_fragment_coord_conventions if we have always_offscreen, but not in
general.

It does look like the benefits of ARB_fragment_coord_conventions are
somewhat limited. I probably mixed it up with ARB_clip_control when I
initially suggested it. I think that would allow us to get rid of the
position fixup, but we'd have to make it a requirement for shader
model 4 if we wanted to avoid patch 6/12.

[PATCH 12/12] wined3d: Use ARB_fragment_coord_conventions for vpos fixups when supported.

[Matteo Bruni]

2016-05-22 18:23 GMT+02:00 Henri Verbeet <hverbeet(a)gmail.com>:
> On 20 May 2016 at 19:56, Matteo Bruni <mbruni(a)codeweavers.com> wrote:
> > if (reg_maps->vpos)
> >      {
> > -        if (shader->device->wined3d->flags & WINED3D_PIXEL_CENTER_INTEGER)
> > +        if (gl_info->supported[ARB_FRAGMENT_COORD_CONVENTIONS])
> > +            shader_addline(buffer, "vpos = gl_FragCoord;\n");
> > +        else if (shader->device->wined3d->flags & WINED3D_PIXEL_CENTER_INTEGER)
>
> Isn't it harder than that? You're dropping the y-flip completely now.
> I think we can actually do that regardless of
> ARB_fragment_coord_conventions if we have always_offscreen, but not in
> general.

You're correct of course... That probably means we need onscreen /
offscreen shader variants in the AlwaysOffscreen = disabled case,
which might be okay. Although...
> It does look like the benefits of ARB_fragment_coord_conventions are
> somewhat limited. I probably mixed it up with ARB_clip_control when I
> initially suggested it. I think that would allow us to get rid of the
> position fixup, but we'd have to make it a requirement for shader
> model 4 if we wanted to avoid patch 6/12.

Right. It looks like ARB_clip_control is available in Mesa for AMD,
Nvidia and Intel since 10.6 so I think it might be reasonable to make
use of it when supported (I'd keep it optional). I'll have a look.