On Wed, 11 Mar 2020 at 16:25, Paul Gofman <gofmanp@gmail.com> wrote:
> On 3/11/20 15:22, Henri Verbeet wrote:
> > On Tue, 10 Mar 2020 at 15:01, Paul Gofman <gofmanp@gmail.com> wrote:
> >> @@ -2629,8 +2622,14 @@ HRESULT texture2d_blt(struct wined3d_texture *dst_texture, unsigned int dst_sub_
> >>          dst_location = dst_texture->resource.map_binding;
> >>
> >>      context = context_acquire(device, dst_texture, dst_sub_resource_idx);
> >> +    scale = abs(src_box->right - src_box->left) != abs(dst_box->right - dst_box->left)
> >> +            || abs(src_box->bottom - src_box->top) != abs(dst_box->bottom - dst_box->top);
> > "scale" is already initialised earlier in the function. It's true that
> > this variant would allow flips and mirrors whereas the variant earlier
> > in the texture2d_blt() doesn't, but flipped/mirrored boxes aren't
> > valid for texture2d_blt(). (In particular, wined3d_texture_blt() calls
> > wined3d_texture_check_box_dimensions(); callers would use
> > WINEDDBLTFX_MIRROR* to do flipped/mirrored blits.)
> >
> >> +    src_location = (scale || convert || blit_op != WINED3D_BLIT_OP_COLOR_BLIT)
> >> +            && wined3d_texture_is_multisample_location(src_texture,
> >> +            src_texture->resource.draw_binding) ? WINED3D_LOCATION_RB_RESOLVED
> >> +            : src_texture->resource.draw_binding;
> > is the "blit_op" check needed?
> I think it is not strictly necessary as colour key or alpha test blits
> won't end up in FBO blitter anyway, but I thought it is less fragile and
> more clear to be explicit here. Should I remove that?

To the extent that it makes a difference for those blit types, I think
using the resolved location makes sense there as well.