Benchmark: Savage

[StephenG1CMZ]

In the other thread, Parisse said CAS floats had 48 bits not 53 for the mantissa.
I work on the Android version, so don't have any absolute timings, but the Pythonesque version seems about twice as fast as the PPL.

It's not the same as in your reference, in which the FOR loop is followed by a division by 2500, if I read it correctly.
I don't recall that in the original. (I do a similar calculation for relative error, but outside the timed code).
My benchmark is meant to implement the original Savage, apart from:
I don't change the mode to Radians (I don't like benchmarks that leave the device in a different mode than it started in)
And I wrap the original code in a procedure call, which wasn't in the original.