Spent some time last night coding a circle-drawing "Rect" with anti-aliased edges. Rolled in a super fast LUT-based integer square root method, and optimized the daylights out of the Pixel::Blend() code. I now have a 64-bit wide calculation in there that handles all three color channels in single vector operations. Should be even faster on XS-200, which has some native 64-bit math. Even so, the 64-bit variant outperformed by best 32-bit incarnation on XS-1 by a small margin.