@NomDuClavier mostly still correct, I think, though new optimizations have improved coalescing and skipping work that's not needed