[PATCH 09/10] d3dx9: Use intermediate surfaces for loading from unmappable source surfaces in D3DXLoadSurfaceFromSurface().

[Matteo Bruni]

On Sun, Nov 18, 2018 at 7:40 PM Henri Verbeet <hverbeet at codeweavers.com> wrote:
> @@ -1937,23 +1939,37 @@ HRESULT WINAPI D3DXLoadSurfaceFromSurface(IDirect3DSurface9 *dst_surface,
>      if (!dst_surface || !src_surface)
>          return D3DERR_INVALIDCALL;
>
> -    IDirect3DSurface9_GetDesc(src_surface, &SrcDesc);
> +    IDirect3DSurface9_GetDesc(src_surface, &src_desc);
> +    if (src_desc.Pool == D3DPOOL_DEFAULT && !(src_desc.Usage & D3DUSAGE_DYNAMIC))
> +    {
> +        IDirect3DSurface9_GetDevice(src_surface, &device);
> +        if (FAILED(IDirect3DDevice9_CreateRenderTarget(device, src_desc.Width, src_desc.Height,
> +                src_desc.Format, D3DMULTISAMPLE_NONE, 0, TRUE, &surface, NULL))
> +                || FAILED(IDirect3DDevice9_StretchRect(device, src_surface, NULL, surface, NULL, D3DTEXF_NONE)))
> +            surface = src_surface;
> +        IDirect3DDevice9_Release(device);
> +    }

Doesn't this leak the staging RT if StretchRect() fails?
I'm also unsure whether StretchRect() covers all the cases of our
interest. After staring at the StretchRect() tests + docs for a while,
my current understanding is that it should always work for us, we're
probably doing an extra copy for off-screen plain surfaces but that's
no big deal.

I guess GetRenderTargetData() might be a bit better / faster for the
cases where it can be used, although it might complicate the function
somewhat and it should probably be in a separate patch anyway.