Without general theories, the only reasonable question is "Faster for What"? What are generalizable theories of 'fast'? Fewer bits moving less with fewer instr

Without general theories, the only reasonable question is "Faster for What"? What are generalizable theories of 'fast'? Fewer bits moving less with fewer instructions, and using the most suitable, specialized hardware available. Even once you have a reasonably well-defined goal, if it's complex enough you may start hitting subsidiary questions, like numerical precision and stability.