>>179
MMX registers are already 64 bits and have NOT AND and such. The only optimization that comes to mind would be vectorizing and doing two at once in SSE2 128bit registers