Mersenne Twister implentation

[grsbanks]

(02-06-2017 10:18 PM)Tim Wessman Wrote:  There is also the TEVAL function.

I've only been using the Prime for less than a week and hadn't found out about TEVAL by the time I wrote these tests If I were to rewrite the benchmark now, that's what I'd use!

(02-06-2017 11:23 PM)rprosperi Wrote:  Tim/Cyril - why would the relative speed differences for these 2 types of commands vary this much between the device and virtual Prime? Internally, if they are presumably running the same code, the relative speeds should also be the same, right?

Higher ratios on the emulator implies the FOR...LOOP generated code is better optimized on the PC?

To be honest, I don 't know exactly how the Prime emulator works. If it's similar to Emu48 et al. then it emulates the hardware and uses the original machine's ROM. The precise manner in which the emulator handles the ARM instruction set will therefore have a direct influence on execution time and, more to the point, relative execution time between various algorithms.

It's one possible explanation anyway...

Edit: surely a higher FOR-to-MAKELIST execution time ratio on the PC means that the FOR loop is better implemented on the Prime (and/or MAKELIST not so well implemented) if it's closer to the MAKELIST execution time?