Summation based benchmark for calculators

[pier4r]

(12-23-2017 07:56 AM)AlexFekken Wrote:  You do realise that you are basically repeating the same calculation thousands of times?

The atan will quickly converge to pi/2 and so the summand to exp(1/3).

Yes, but the calculator should not know (although I am not sure what happens when x gets large and tan^-1 may report always the same value). Moreover as I wrote, the formula is pretty arbitrary and is one picked quickly that does not converge but also does not explode with larger values.

I was thinking about alternate series (using sin or cos) but at the end it would be similar. It is repeating thousands of times the same functions. At it is also what is wanted, so one gets an idea how those functions performs together, likely avoiding possible internal optimizations such as "a function and its inverse composed".

What is interesting is that with the 50g, in userRPL, I was not yet able to get a program faster than the summation. And the faster value is a summation, that is an algebraic object that is notoriously slower than pure userRPL!

Anyone with a 48 or other devices? Is it difficult to produce a RPN equivalent (for 41 and other systems) for the formula?

@Pauli: is it difficult to run an internal program on the 34s for this summation? If not, could you try?


edit: also newrpl should be interesting. Because the first second is done at 6 mhz and then the 50g switches to full 192 mhz according to what I read in the newRPL threads.