This helps significantly for large rectangles.

This is only done for the 32-bpp case as typically the lower bit
depths would more likely be used to draw small icons and the like,
where the extra fixed cost of this optimization will potentially
outweigh the savings.
Signed-off-by: Huw Davies <huw@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>