As an aside, most JIT runtimes actually don't compile code until after it has been run a few thousand times and seems hot – before that they interpret it. Julia is unusual in that it fully JIT compiles almost all code (except for some top-level expressions which are interpreted) before running it at all. That's technically what JIT originally meant, but not what most systems do since they often *need* profile information from execution in order to compile the code, whereas Julia can infer this information in advance.
On Tue, Sep 13, 2016 at 3:09 AM, Matjaz Licer <matjaz.li...@gmail.com> wrote: > Makes sense, thanks :-) > > On 12 September 2016 at 17:17, Stefan Karpinski <ste...@karpinski.org> > wrote: > >> JIT = Just In Time, i.e. the first time you use the code. >> >> On Mon, Sep 12, 2016 at 6:52 AM, MLicer <matjaz.li...@gmail.com> wrote: >> >>> Indeed it does! I thought JIT compilation takes place prior to execution >>> of the script. Thanks so much, this makes sense now! >>> >>> Output: >>> first call: 0.804573 seconds (1.18 M allocations: 53.183 MB, 1.43% gc >>> time) >>> repeated call: 0.000472 seconds (217 allocations: 402.938 KB) >>> >>> Thanks again, >>> >>> Cheers! >>> >>> >>> On Monday, September 12, 2016 at 12:48:30 PM UTC+2, randm...@gmail.com >>> wrote: >>>> >>>> The Julia code takes 0.000535 seconds for me on the second run -- >>>> during the first run, Julia has to compile the method you're timing. Have a >>>> look at the performance tips >>>> <http://docs.julialang.org/en/latest/manual/performance-tips/#measure-performance-with-time-and-pay-attention-to-memory-allocation> >>>> for a more in depth explanation. >>>> >>>> Am Montag, 12. September 2016 11:53:01 UTC+2 schrieb MLicer: >>>>> >>>>> Dear all, >>>>> >>>>> i've written a low-pass filter in Julia and Python and the code in >>>>> Julia seems to be much slower (*0.800 sec in Julia vs 0.000 sec in >>>>> Python*). I *must* be coding ineffieciently, can anyone comment on >>>>> the two codes below? >>>>> >>>>> *Julia:* >>>>> >>>>> >>>>> <https://lh3.googleusercontent.com/-pEyZjPDcmP8/V9Z6hJdXKfI/AAAAAAAAKLc/iiefpFSzY88S333QndnbnnAnKUBQvzTkACLcB/s1600/filter_julia.png> >>>>> using PyPlot, DSP >>>>> >>>>> # generate data: >>>>> x = linspace(0,30,1e4) >>>>> sin_noise(arr) = sin(arr) + rand(length(arr)) >>>>> >>>>> # create filter: >>>>> designmethod = Butterworth(5) >>>>> ff = digitalfilter(Lowpass(0.02),designmethod) >>>>> @time yl = filtfilt(ff, sin_noise(x)) >>>>> >>>>> Python: >>>>> >>>>> from scipy import signal >>>>> import numpy as np >>>>> import cProfile, pstats >>>>> >>>>> def sin_noise(arr): >>>>> return np.sin(arr) + np.random.rand(len(arr)) >>>>> >>>>> def filterSignal(b,a,x): >>>>> return signal.filtfilt(b, a, x, axis=-1) >>>>> >>>>> def main(): >>>>> # generate data: >>>>> x = np.linspace(0,30,1e4) >>>>> y = sin_noise(x) >>>>> b, a = signal.butter(5, 0.02, "lowpass", analog=False) >>>>> ff = filterSignal(b,a,y) >>>>> >>>>> cProfile.runctx('filterSignal(b,a,y)',globals(),{'b':b,'a':a,'y':y >>>>> },filename='profileStatistics') >>>>> >>>>> p = pstats.Stats('profileStatistics') >>>>> printFirstN = 5 >>>>> p.sort_stats('cumtime').print_stats(printFirstN) >>>>> >>>>> if __name__=="__main__": >>>>> main() >>>>> >>>>> >>>>> Thanks very much for any replies! >>>>> >>>> >> >
