12 ms·
Now run the superoptimizer on itself (or at least parts of it.) Yes I realize this would quickly hit a point of diminishing returns since it would only try to c
by Houshalter 14y ago
Now run the superoptimizer on itself (or at least parts of it.) Yes I realize this would quickly hit a point of diminishing returns since it would only try to create code that produces the exact same output, not novel code that produces even better output, but it's still an interesting idea to think about.
And maybe you could judge the output code by it's utility rather than how closely it matches the output of the original algorithm, but that's a much harder problem.