Viral, Thanks for the information about possible improvements in the future.
Gray, if you look at the 'getAlgs' function in my code you'll see that I do include an alternate (but unused by default) way of encapsulating each algorithm in a function. This doesn't seem to improve the compilation speed. chris On Tuesday, January 20, 2015 at 1:02:17 PM UTC-8, Viral Shah wrote: > > With Jameson's work on static compilation and very early work towards > interpreting Julia code, this will get addressed, but not in the near > future. > > -viral > > On Tuesday, January 20, 2015 at 12:11:45 AM UTC+5:30, Gray Calhoun wrote: >> >> Hi Christian, you can certainly make the code-writing process >> much faster by splitting large functions into many distinct >> smaller functions. In your example (which I've only skimmed), >> `mimosimu` looks like it directly implements and annotates >> several similar but different algorithms which could all be >> separate functions. Compilation happens at the function level, >> not the file level, and shorter functions compile (and usually >> run) much faster. >> >> Also, loading part of a file is something that editors can handle >> as well, so if there aren't satisfactory packages (I haven't >> tried the Autoreload.jl package recommended in the last thread, >> but it looks promising), using an editor like Emacs (with ESS) or >> Light Table (with Juno) could help. I'm sure there are lots of >> other editors that do this too, but those are the two I'm >> familiar with. >> >> On Sunday, January 18, 2015 at 10:04:58 PM UTC-6, Christian Peel wrote: >>> >>> I'm enjoying learning Julia. >>> >>> I have the same toy script written in Matlab and Julia at the following >>> URL: >>> https://github.com/ChristianPeel/toySims/tree/master/mimoToys >>> Running the following in matlab >>> tic; mimoUPtoy(200,4,4,4,0,12,12,[-10:5:30]); toc >>> takes about 0.82 seconds on a specific machine. In contrast, with Julia >>> I first need to 'include' it, then the first time through the compiler >>> takes around 7 seconds to compile the code and execute it. Subsequent >>> executions of the code have speed similar to Matlab. >>> julia> tic(); include("mimoUPtoy.jl"); toc() >>> elapsed time: 0.348886314 seconds >>> 0.348886314 >>> julia> tic(); mimoUPtoy(200,4,4,4,0,12,12,[-10.0:5:30]); toq() >>> ...output... >>> 7.002885787 >>> julia> tic(); mimoUPtoy(200,4,4,4,0,12,12,[-10.0:5:30]); toq() >>> ...output... >>> 0.860070365 >>> Some questions are (1) is there is some improvement in 0.4 or otherwise >>> which would improve the initial JIT time? (2) is there any way to only >>> recompile the parts of a file that have changed? Say by using a hash on a >>> function to see if it has changed? (3) I'm used to the Matlab development >>> cycle in which I don't need to 'include' anything (it's done automatically) >>> and also any JIT compilation that Matlab does is very fast and is >>> essentially not noticable. So even though Matlab may be slower for code >>> execution, it feels faster for code development. Is there anything that >>> can be done to Julia to make the code writing process quicker? Is there a >>> way to automate the 'include' process? >>> >>> I acknowledge that one option to speed compilation time is to break >>> mimoUPtoy.jl into separate files. I'm also sure that there are things that >>> can be improved in both my Matlab and Julia code. Finally; this function >>> is similar to that which I refered to in a previous post (I was busy at the >>> time and didn't get the code posted) >>> >>> https://groups.google.com/forum/?fromgroups=#!searchin/julia-users/peel/julia-users/thR_80jtE2Q/ymV5i-AXmKkJ >>> >>>
