[PATCH 1/2] wined3d: Handle scale with multisampled destination in texture2d_blt_fbo().

[Henri Verbeet]

On Mon, 9 Mar 2020 at 14:08, Paul Gofman <gofmanp at gmail.com> wrote:
> @@ -188,9 +194,14 @@ void texture2d_blt_fbo(struct wined3d_device *device, struct wined3d_context *co
>       * the purpose of loading the destination surface.) */
>      wined3d_texture_load_location(src_texture, src_sub_resource_idx, context, src_location);
>      if (!texture2d_is_full_rect(dst_texture, dst_sub_resource_idx % dst_texture->level_count, dst_rect))
> +    {
>          wined3d_texture_load_location(dst_texture, dst_sub_resource_idx, context, dst_location);
> +    }
>      else
> +    {
>          wined3d_texture_prepare_location(dst_texture, dst_sub_resource_idx, context, dst_location);
> +        wined3d_texture_validate_location(dst_texture, dst_sub_resource_idx, dst_location);
> +    }
...
> -    if (dst_location == WINED3D_LOCATION_DRAWABLE && dst_texture->swapchain->front_buffer == dst_texture)
> +    if (dst_location != orig_dst_location)
> +    {
> +        wined3d_texture_invalidate_location(dst_texture, dst_sub_resource_idx, ~dst_location);
> +        wined3d_texture_load_location(dst_texture, dst_sub_resource_idx, context, orig_dst_location);
> +    }

As a general rule, it's not ideal to manipulate the valid locations
here, for various reasons. It would probably be best to do any
resolving in fbo_blitter_blit() instead of texture2d_blt_fbo(), and
then return the valid locations to the caller. (Compare e.g.
cpu_blitter_blit().)