Comments for Code movement in Java programs with pure functions http://purejava.wordpress.com A journal of my research for a BSc (Hons) project at VUW Fri, 22 Aug 2008 09:00:17 +0000 http://wordpress.com/ hourly 1 Comment on You are doing it wrong… by qwandor http://purejava.wordpress.com/2008/08/21/you-are-doing-it-wrong/#comment-58 qwandor Fri, 22 Aug 2008 09:00:17 +0000 http://purejava.wordpress.com/?p=43#comment-58 If I were trying to optimise a particular program, then I would agree with you Dmitry. However, the aim of this project is to prototype a new compiler optimisation based on using knowledge of pure methods (and other properties of classes and methods) to do loop invariant code movement that would not otherwise be possible. As such I am trying at the moment to find cases where the existing optimisations in HotSpot's JIT (for example) cannot be applied as it does not know enough about the structure of the program, but my optimisations can safely be applied and improve performance. It seems that such cases are not as common as I had hoped. However, I have just this evening been looking at other JVMs that do not do so much optimisation at runtime, and in these cases my compile-time optimisations do seem to be worthwhile. I will post some more about this shortly. If I were trying to optimise a particular program, then I would agree with you Dmitry. However, the aim of this project is to prototype a new compiler optimisation based on using knowledge of pure methods (and other properties of classes and methods) to do loop invariant code movement that would not otherwise be possible. As such I am trying at the moment to find cases where the existing optimisations in HotSpot’s JIT (for example) cannot be applied as it does not know enough about the structure of the program, but my optimisations can safely be applied and improve performance.

It seems that such cases are not as common as I had hoped. However, I have just this evening been looking at other JVMs that do not do so much optimisation at runtime, and in these cases my compile-time optimisations do seem to be worthwhile. I will post some more about this shortly.

]]> Comment on You are doing it wrong… by Dmitry http://purejava.wordpress.com/2008/08/21/you-are-doing-it-wrong/#comment-57 Dmitry Fri, 22 Aug 2008 08:33:01 +0000 http://purejava.wordpress.com/?p=43#comment-57 You should first profile the application to identify the bottlenecks, then optimize those bottlenecks, then repeat. This is much more effective than mechanically applying a trivial optimization throughout the entire code base. Two good Java profilers are JProfiler and Yourkit. You should first profile the application to identify the bottlenecks, then optimize those bottlenecks, then repeat. This is much more effective than mechanically applying a trivial optimization throughout the entire code base.

Two good Java profilers are JProfiler and Yourkit.

]]> Comment on You are doing it wrong… by qwandor http://purejava.wordpress.com/2008/08/21/you-are-doing-it-wrong/#comment-56 qwandor Fri, 22 Aug 2008 00:17:25 +0000 http://purejava.wordpress.com/?p=43#comment-56 Okay, by reducing the amount of 'optimisation' I do (only moving expressions out of loops if they actually contain method calls or field dereferences), I can get more reasonable times, but still not any faster than the unoptimised build: <code> real 0m3.441s user 0m3.282s sys 0m0.051s real 0m3.442s user 0m3.295s sys 0m0.040s </code> Okay, by reducing the amount of ‘optimisation’ I do (only moving expressions out of loops if they actually contain method calls or field dereferences), I can get more reasonable times, but still not any faster than the unoptimised build:

real 0m3.441s
user 0m3.282s
sys 0m0.051s

real 0m3.442s
user 0m3.295s
sys 0m0.040s

]]> Comment on Testing manual optimisation by More tests « Code movement in Java programs with pure functions http://purejava.wordpress.com/2008/08/19/testing-manual-optimisation/#comment-54 More tests « Code movement in Java programs with pure functions Thu, 21 Aug 2008 02:24:32 +0000 http://purejava.wordpress.com/?p=36#comment-54 [...] Code movement in Java programs with pure functions A journal of my research for a BSc (Hons) project at VUW « Testing manual optimisation [...] [...] Code movement in Java programs with pure functions A journal of my research for a BSc (Hons) project at VUW « Testing manual optimisation [...]

]]> Comment on Testing manual optimisation by qwandor http://purejava.wordpress.com/2008/08/19/testing-manual-optimisation/#comment-53 qwandor Thu, 21 Aug 2008 01:05:27 +0000 http://purejava.wordpress.com/?p=36#comment-53 Thanks Dmitry, that confirms what I was suspecting. I assume that HotSpot is inlining short methods when it can work out statically what methods are actually called. Thanks Dmitry, that confirms what I was suspecting. I assume that HotSpot is inlining short methods when it can work out statically what methods are actually called.

]]> Comment on Testing manual optimisation by Dmitry http://purejava.wordpress.com/2008/08/19/testing-manual-optimisation/#comment-52 Dmitry Wed, 20 Aug 2008 07:12:22 +0000 http://purejava.wordpress.com/?p=36#comment-52 First, HotSpot Server optimizes your loops away as they do not compute any result. You should do something inside the loops, and make sure the result is not thrown away, otherwise the results are meaningless. Google for something like "common microbenchmarking mistakes". Second, HotSpot Server is capable of detecting that array.length and list.size() are loop invariants. If you have some code inside the loops that _may_ potentially change 'array' or 'list', your manual optimization would come into play. First, HotSpot Server optimizes your loops away as they do not compute any result. You should do something inside the loops, and make sure the result is not thrown away, otherwise the results are meaningless. Google for something like “common microbenchmarking mistakes”.

Second, HotSpot Server is capable of detecting that array.length and list.size() are loop invariants. If you have some code inside the loops that _may_ potentially change ‘array’ or ‘list’, your manual optimization would come into play.

]]> Comment on Milestone 1 by Presentation « Code movement in Java programs with pure functions http://purejava.wordpress.com/2008/05/09/milestone-1/#comment-32 Presentation « Code movement in Java programs with pure functions Fri, 16 May 2008 00:12:38 +0000 http://purejava.wordpress.com/?p=12#comment-32 [...] Code movement in Java programs with pure functions A journal of my research for a BSc (Hons) project at VUW « Milestone 1 [...] [...] Code movement in Java programs with pure functions A journal of my research for a BSc (Hons) project at VUW « Milestone 1 [...]

]]> Comment on A first simple optimisation by This time, maybe? « Code movement in Java programs with pure functions http://purejava.wordpress.com/2008/04/25/a-first-simple-optimisation/#comment-16 This time, maybe? « Code movement in Java programs with pure functions Wed, 30 Apr 2008 03:55:52 +0000 http://purejava.wordpress.com/?p=10#comment-16 [...] Code movement in Java programs with pure functions A journal of my research for a BSc (Hons) project at VUW « A first simple optimisation [...] [...] Code movement in Java programs with pure functions A journal of my research for a BSc (Hons) project at VUW « A first simple optimisation [...]

]]> Comment on Introduction by qwandor http://purejava.wordpress.com/2008/03/25/introduction/#comment-3 qwandor Fri, 28 Mar 2008 05:34:41 +0000 http://purejava.wordpress.com/?p=4#comment-3 It looks like I forgot. Oops. It looks like I forgot. Oops.

]]> Comment on Introduction by Neil Walbran http://purejava.wordpress.com/2008/03/25/introduction/#comment-2 Neil Walbran Thu, 27 Mar 2008 22:35:27 +0000 http://purejava.wordpress.com/?p=4#comment-2 How come objective 4 of your research proposal doesn't have any time allocated to it? How come objective 4 of your research proposal doesn’t have any time allocated to it?

]]>