Re: [2/4] wined3d: Track when the FBO is dirty.

[Henri Verbeet]

On 21 March 2011 21:57, Adam Martinson  wrote:
> @@ -2102,6 +2121,7 @@ BOOL context_apply_clear_state(struct wined3d_context *context, IWineD3DDeviceIm
>              if (rt_count < context->gl_info->limits.buffers)
>                  memset(context->blit_targets + rt_count, 0, (context->gl_info->limits.buffers - rt_count) * sizeof(*rts));
>              context_apply_fbo_state(context, GL_FRAMEBUFFER, rt_count, context->blit_targets, depth_stencil, SFLAG_INTEXTURE);
> -
>        context->device_fbo = context->current_fbo;

You can't do that.
> @@ -4801,6 +4801,7 @@ static HRESULT WINAPI IWineD3DDeviceImpl_Clear(IWineD3DDevice *iface, DWORD rect
>      }
>  device_get_draw_rect(device, &draw_rect);
> - IWineD3DDeviceImpl_MarkFBODirty(device);

What is this for?
> +static inline void IWineD3DDeviceImpl_MarkFBODirty(IWineD3DDeviceImpl *This)

This doesn't need to be in wined3d_private.h, it's only used from device.c.
> - int i;
> - for (i = 0; i < This->numContexts; ++i)

numContexts is unsigned.

What would happen when the current draw fbo entry would get evicted
from the cache? Also, while the patch avoids redundant lookups for
draws, it doesn't do a lot for clears. Is that intentional because
it's not worth it, or is there a different reason for that?

[2/4] wined3d: Track when the FBO is dirty.

[Adam Martinson]

On 03/27/2011 06:45 AM, Henri Verbeet wrote:
> On 21 March 2011 21:57, Adam Martinson  wrote:
> > @@ -2102,6 +2121,7 @@ BOOL context_apply_clear_state(struct wined3d_context *context, IWineD3DDeviceIm
> >               if (rt_count<  context->gl_info->limits.buffers)
> >                   memset(context->blit_targets + rt_count, 0, (context->gl_info->limits.buffers - rt_count) * sizeof(*rts));
> >               context_apply_fbo_state(context, GL_FRAMEBUFFER, rt_count, context->blit_targets, depth_stencil, SFLAG_INTEXTURE);
> > -
> >        context->device_fbo = context->current_fbo;
>
> You can't do that.
> > @@ -4801,6 +4801,7 @@ static HRESULT WINAPI IWineD3DDeviceImpl_Clear(IWineD3DDevice *iface, DWORD rect
> >       }
> >   device_get_draw_rect(device,&draw_rect);
> > - IWineD3DDeviceImpl_MarkFBODirty(device);
>
> What is this for?
> > +static inline void IWineD3DDeviceImpl_MarkFBODirty(IWineD3DDeviceImpl *This)
>
> This doesn't need to be in wined3d_private.h, it's only used from device.c.
> > - int i;
> > - for (i = 0; i<  This->numContexts; ++i)
>
> numContexts is unsigned.
>
> What would happen when the current draw fbo entry would get evicted
> from the cache? Also, while the patch avoids redundant lookups for
> draws, it doesn't do a lot for clears. Is that intentional because
> it's not worth it, or is there a different reason for that?

I think I've addressed all of these... attached try #2.

[2/4] wined3d: Track when the FBO is dirty.

[Henri Verbeet]

On 7 April 2011 17:40, Adam Martinson  wrote:
> I think I've addressed all of these... attached try #2.

It doesn't compile.

[2/4] wined3d: Track when the FBO is dirty.

[Adam Martinson]

On 04/07/2011 10:54 AM, Henri Verbeet wrote:
> On 7 April 2011 17:40, Adam Martinson  wrote:
> > I think I've addressed all of these... attached try #2.
>
> It doesn't compile.

Heh... someone changed a variable name on me, sorry.

[2/4] wined3d: Track when the FBO is dirty.

[Henri Verbeet]

On 7 April 2011 18:55, Adam Martinson  wrote:
> Heh... someone changed a variable name on me, sorry.

Well yeah, things change. But how did you run the tests if the patch
didn't compile?