Re: [3/3] WineD3D: Update the scissor rect on a render target change

[Henri Verbeet]

2009/7/20 Stefan Dösinger :
> -
>    scissor.top = 0;
> -
>    scissor.left = 0;
> -
>    scissor.right = viewport.Width;
> -
>    scissor.bottom = viewport.Height;
> -
>    IWineD3DDevice_SetScissorRect(iface, &scissor);
> -
>    IWineD3DDeviceImpl_MarkStateDirty(This, STATE_SCISSORRECT);

The MarkStateDirty() call looks redundant to me. Also, does this do
the right thing when we're recording a stateblock?

[3/3] WineD3D: Update the scissor rect on a render target change

[Stefan Dösinger]

Am Monday 20 July 2009 12:58:16 schrieb Henri Verbeet:
> The MarkStateDirty() call looks redundant to me. Also, does this do
> the right thing when we're recording a stateblock?

It's not redundant because the scissor rect needs updating on a RT switch 
anyway - I think it wasn't visible in this diff, but there is a SetViewport 
call above the SetScissorRect which explains the situation.

The stateblock recording is a good question. The SetScissorRect will update 
the update stateblock, not the device's stateblock if a recording happens, 
and SetRenderTarget is not recorded in a stateblock. I'll extend the test.

[3/3] WineD3D: Update the scissor rect on a render target change

[Henri Verbeet]

2009/7/20 Stefan Dösinger :
> Am Monday 20 July 2009 12:58:16 schrieb Henri Verbeet:
> > The MarkStateDirty() call looks redundant to me. Also, does this do
> > the right thing when we're recording a stateblock?
>
> It's not redundant because the scissor rect needs updating on a RT switch
> anyway - I think it wasn't visible in this diff, but there is a SetViewport
> call above the SetScissorRect which explains the situation.

Ok, but that's for a different reason then. Also, the wording of that
comment could use some work.

[3/3] WineD3D: Update the scissor rect on a render target change

[Henri Verbeet]

2009/7/20 Henri Verbeet :
> 2009/7/20 Stefan Dösinger :
> > Am Monday 20 July 2009 12:58:16 schrieb Henri Verbeet:
> > > The MarkStateDirty() call looks redundant to me. Also, does this do
> > > the right thing when we're recording a stateblock?
> >
> > It's not redundant because the scissor rect needs updating on a RT switch
> > anyway - I think it wasn't visible in this diff, but there is a SetViewport
> > call above the SetScissorRect which explains the situation.
>
> Ok, but that's for a different reason then. Also, the wording of that
> comment could use some work.

Actually, that is already handled by FindContext(), although perhaps
"render_offscreen" should be tracked by the context rather than the
device.