Hello, Some random notes from a quick reading,
First, you are using 2000x2000 piddle of double (~30Mb) so may be using 'inplace' in several places could help in this area (allocating memory is generally an expensive operation). From my point of view temporary array(s) should to be used also but this is out of purpose if the code is intended to be didactic. Second, you're not using (pdl) thread and apparently operation with $u and $t can be threaded. Something like (at the beginning) $re = zeroes($npts)->xlinvals; $im = zeroes($npts)->dummy()->ylinvals; $t = $re; $u = $im; sounds more pdl'ish and you'll save memory too. Hope that helps, Greg Le jeudi 01 mars 2007 à 07:32 +0000, Xavier Calbet a écrit : > Dear all, > > I am writing an article for Free Software Magazine. As some > of you know this is an online magazine with relatively high impact. > I will be publishing some PDL benchmarks when compared with > other languages. As you know benchmarks can be very > tricky and some misdoing can be easily done. > I am showing here the results for criticism > or comments. > Please send me your comments before 15 March if you think > there is something severly flawed in these results. Any other > comments are obviously also welcome. > I am mostly interested in the comparison of IDL and PDL. > I am running code to calculate the madelbrot fractal. > They are fairly simple and short programs. > I attach the code for your review. If there is anything > you think should be modified to make the code faster > in any of the languages please let me know. > I also attach the matlab program since I am > not sure I implemented the "clip" function in the > right way. Any comments from expert matlab programmers > welcome. > > I have measured the times and run the codes with > date;./mandel.pl;date > date;idl < mandel.bat;date > date;matlab -nodisplay -nojvm < mandel.m;date > > where mandel.bat is a batch file that compiles and runs mandel.pro > > I have run these benchmarks without running anything else on the computer. > I have done this several times and have obtained consistent times. > Is there a better way to measure the time without resorting to date? > > The results are the following: > > gcc 201 sec > g77 201 sec > PDL 651 sec > IDL 694 sec > MatLab 2738 sec > Octave 2031 sec > > Do these times seem reasonable to you? > > Many thanks, > > Xavier > _______________________________________________ > PDL-porters mailing list > [email protected] > http://mailman.jach.hawaii.edu/mailman/listinfo/pdl-porters ___________________________________________________________________________ Yahoo! Mail réinvente le mail ! Découvrez le nouveau Yahoo! Mail et son interface révolutionnaire. http://fr.mail.yahoo.com _______________________________________________ Perldl mailing list [email protected] http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
