Comments on: On Optimisations http://www.vectorstorm.org/2009/06/21/on-optimisations/ Creating games, one brightly glowing line at a time. Mon, 26 Jul 2010 12:30:57 +0000 hourly 1 http://wordpress.org/?v=3.0.1 By: trevor http://www.vectorstorm.org/2009/06/21/on-optimisations/comment-page-1/#comment-924 trevor Tue, 23 Jun 2009 05:00:54 +0000 http://www.vectorstorm.org/?p=580#comment-924 Yep, I totally agree with that forum post. The Introversion guys tend to know what they're talking about. Except when it comes to their devotion toward Multiwinia, which has always sort of baffled me. Yep, I totally agree with that forum post. The Introversion guys tend to know what they’re talking about. Except when it comes to their devotion toward Multiwinia, which has always sort of baffled me.

]]> By: Paul http://www.vectorstorm.org/2009/06/21/on-optimisations/comment-page-1/#comment-923 Paul Tue, 23 Jun 2009 01:03:17 +0000 http://www.vectorstorm.org/?p=580#comment-923 here's something you may find interesting on optimization, and fuel your feelings of justification on the matter: http://forums.introversion.co.uk/introversion/viewtopic.php?t=2003 here’s something you may find interesting on optimization, and fuel your feelings of justification on the matter:

http://forums.introversion.co.uk/introversion/viewtopic.php?t=2003

]]> By: trevor http://www.vectorstorm.org/2009/06/21/on-optimisations/comment-page-1/#comment-922 trevor Mon, 22 Jun 2009 00:53:15 +0000 http://www.vectorstorm.org/?p=580#comment-922 All good points! (And thank you for the nice comment on my memory code!) My point wasn't that you should ignore performance until after having finished your code; rather, I was suggesting that once you've written a block of code, it's probably not worth going back and trying to improve its speed until (a) your program is complete enough that you're certain that code will actually be in the final version and (b) you've done profiling tests and verified that the code actually is slow enough that it's worth optimising. Which I think is more or less what you're saying as well. :) All good points! (And thank you for the nice comment on my memory code!)

My point wasn’t that you should ignore performance until after having finished your code; rather, I was suggesting that once you’ve written a block of code, it’s probably not worth going back and trying to improve its speed until (a) your program is complete enough that you’re certain that code will actually be in the final version and (b) you’ve done profiling tests and verified that the code actually is slow enough that it’s worth optimising.

Which I think is more or less what you’re saying as well. :)

]]> By: Dan Haraj http://www.vectorstorm.org/2009/06/21/on-optimisations/comment-page-1/#comment-921 Dan Haraj Sun, 21 Jun 2009 20:34:49 +0000 http://www.vectorstorm.org/?p=580#comment-921 While I know that premature optimization is evil, and you shouldn't optimize without measuring, I think it is still very important to write code in the first place conscious of performance. I think that a lot of the optimizations you can do in C++ are just idiomatic, or should be idiomatic. Things like copying memory should be done the way the hardware wants it done. Like copying resources from system to video memory. DX10, for example, finishes a copy at the start of the second frame after the copy command is issued. So a piece of software should be aware of that and try to access resources at the most optimal times. Otherwise, it will block needlessly. There are other things, like being cache aware. You could write an inner loop naively, measure later, and optimize later, or you could keep in mind cache effects and make sure you're iterating through elements in a sane manner when you write it :) But things like your memory code, which is very well written might I add, definitely could only be caught with proper measurement and optimization. While I know that premature optimization is evil, and you shouldn’t optimize without measuring, I think it is still very important to write code in the first place conscious of performance. I think that a lot of the optimizations you can do in C++ are just idiomatic, or should be idiomatic. Things like copying memory should be done the way the hardware wants it done.

Like copying resources from system to video memory. DX10, for example, finishes a copy at the start of the second frame after the copy command is issued. So a piece of software should be aware of that and try to access resources at the most optimal times. Otherwise, it will block needlessly. There are other things, like being cache aware. You could write an inner loop naively, measure later, and optimize later, or you could keep in mind cache effects and make sure you’re iterating through elements in a sane manner when you write it :)

But things like your memory code, which is very well written might I add, definitely could only be caught with proper measurement and optimization.

]]>