It is 2*10 lookups. 10 lookups for the global variable + 10 lookups to find the correct compute method. When the correct method is found, inference works nicely again. Compared to the timing of the whole method (on my computer about 3.5 seconds), that seems insignificant.
It definitely does not hurt to avoid non const globals though. Ivar kl. 18:52:53 UTC+2 torsdag 31. juli 2014 skrev Patrick O'Leary følgende: > > There are two points: (1) the problem isn't the lookup, it's type > inference (see https://github.com/JuliaLang/julia/issues/524) which can > prevent Julia from generating fast code, and (2) even if in this case it > doesn't matter, it might later on, so let's recommend some good habits. > > Daniel points out using `const` with the global definition, which is > another valid approach. > > On Thursday, July 31, 2014 11:28:26 AM UTC-5, Ivar Nesje wrote: >> >> Patrick, that is wrong. You don't spend a significant portion of 3 >> seconds looking up a global variable 10 times. >> >> kl. 18:25:25 UTC+2 torsdag 31. juli 2014 skrev Patrick O'Leary følgende: >>> >>> On Thursday, July 31, 2014 10:38:33 AM UTC-5, Neal Becker wrote: >>>> >>>> Attached is my 1st attempt at julia, it is a simple FIR filter, which I >>>> translated from my c++ version. >>>> >>>> It is benchmarking about 10x slower than python wrapped c++ version. >>>> >>>> Any suggestions? >>> >>> >>> Your variable `g` is in global scope, and used in the `timeit()` >>> function, so that's a likely candidate (see >>> http://julia.readthedocs.org/en/latest/manual/performance-tips/#avoid-global-variables). >>> >>> Try defining `g` inside `timeit()`. >>> >>
