[Shootout-list] A different way of measuring

[Peter Hinely]

Hi folks,

I think what the shootout really should do is measure the slope of the
graph of CPU time vs. N iterations. The number of iterations for a
particular test by a particular language could be binary searched until
the timing of N iterations was approximately X seconds, say 0.5 second.
Then execution time of N*2, N*3, N*4, N*5 iterations would be measured.
The shootout score for a language for a test would be the slope of the
graph of those points.

That way both slow and fast language implementations could be measured
accurately.  It would also factor out start-up times. (The startup-time
would be the y-intercept on the graph.)

Regards,
Peter