http://d.puremagic.com/issues/show_bug.cgi?id=6594
--- Comment #3 from Joseph Rushton Wakeling <[email protected]> 2013-07-05 08:37:38 PDT --- (In reply to comment #2) > The performance penalty most probably comes from passing it (by value) to > algorithms or whatnot. If you're passing it by value, you have bigger problems than slowdown ... :-) > I know for a fact this *will* cause significant slowdown. I observed this > while > implementing a Lagged Fibonacci PRNG, whose size varies a lot. The bigger > flavors of the PRNG would make the runtime grind to a halt, while the biggest > versions simply stack overflowed. > > ---- > All this to say, I think it is not the algorithm of mersenne itself that is > slow, but rather the range implementation. Until this is fixed, we *may* want > to change the default. Fully agree that the real solution here is to change the RNGs to reference types. I don't think switching default to Xorshift is a good idea, though, even as a short term method. Apart from any other consideration, Issue #10550 should give us pause on that idea. -- Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email ------- You are receiving this mail because: -------
