Jump to content

  • Log In with Google      Sign In   
  • Create Account


Simple & Robust OOP vs Performance


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • This topic is locked This topic is locked
3 replies to this topic

#1 rockseller   Members   -  Reputation: 121

Like
0Likes
Like

Posted 06 September 2012 - 10:23 AM

Good day peeps,

I'm going to talk about a project I started, a 2D Game, aimed for Android 2.2 and above, OpenGL ES 2.0, and being coded with the default Android Java SDK (up-to date), rather than the NDK framework.

Ever since I started developing a platform RPG game, I took special care of the micro code optimization, because I felt it would have a big impact in the long term, specially because It's a serious project, involving help from other designers and creative guys;

In terms of performance, before micro-code-optimizing, the experts say that you should profile your game, and keep measuring it on real devices in order to see where your bottleneck really is.

In my experience, and please flame me if you feel I'm wrong, for a time consuming and serious project, this is not 100% true, since you spend time developing tools for other guys to work within your engine framework, and layout the way Bitmaps should be drawn in order to be used as textures (standardizing them), and most of the time you need to save time by optimizing parts of the code (assuming you have a good code design) by the very beginning, in parts where you do know that you will have a performance boost

(I.E. reducing the possibilities for the Garbage Collector to run)


As a project leader and coder, I have realized that in the end is better to have a robust and well designed Class, rather than an "optimized" all-in class.


For instance, I had a class called:

Monster
That had 100 methods

Sure it had a lot of micro optimization inside those methods (good bye readability, I know, good practices, still!!)



But after some time, I figured out that iwas better to have a class:

Monster
With 5 methods, and 3 objects inside
MonsterAttackModule
MonsterAIModule
MonsterBodyModule



With the second case, I still have some micro code optimization, more readability, but more objects
That's more memory, isn't it?



So my point is that in the end is better to have a balance of readability, and a robust OOP Code, micro-code optimize up and there, and profile as experts says, making the original sentence, not completely true, right? Posted Image


What you guys think?

Attached Thumbnails

  • device-2012-08-04-141939.png

Edited by rockseller, 06 September 2012 - 10:27 AM.


Sponsor:

#2 frob   Moderators   -  Reputation: 19757

Like
0Likes
Like

Posted 06 September 2012 - 04:21 PM

Basically you summed it up.

Language features do have a cost, you need to be aware of those costs.

Make it work (first pass development). Make it work well (debug). Then make it work fast (optimize). In that order.


You point out in your example that you were able to fit a tiny bit more data into your game. But what is the cost? It takes longer to develop initially, is more difficult to debug, and more time consuming to maintain.

You need to know and understand the tradeoff. Is the very slight reduction of gameplay performance worth the cost to develop the software? That's a business decision, but generally the tradeoff is made in favor of faster/cheaper development.

For example, virtual functions were scoffed when they were introduced in the 1980s. People (correctly) pointed out their performance cost, and (incorrectly) calculated the performance cost as though every function were virtual. Hopefully it is obvious that it is unwise to make every function virtual because there is overhead (roughly 15ns per function call on current hardware), but that doesn't mean virtual functions are bad. Used properly virtual functions solve a problem and generally offer better performance than trying to write your own solution.

Know the cost and use the feature properly.


you need to save time by optimizing parts of the code (assuming you have a good code design) by the very beginning, in parts where you do know that you will have a performance boost


Knuth's saying was that premature optimization is the root of all evil. Wait until you have written it and profiled it before changing it, and make sure you measure afterword to verify your changes improved things.

But that doesn't mean you should intentionally write pessimistic code when you know something is performance-intensive. If you know something will be resource heavy you should plan for it accordingly. That is NOT a premature optimization.
Check out my personal indie blog at bryanwagstaff.com.

#3 rockseller   Members   -  Reputation: 121

Like
0Likes
Like

Posted 06 September 2012 - 04:32 PM

Basically you summed it up.

Language features do have a cost, you need to be aware of those costs.

Make it work (first pass development). Make it work well (debug). Then make it work fast (optimize). In that order.


You point out in your example that you were able to fit a tiny bit more data into your game. But what is the cost? It takes longer to develop initially, is more difficult to debug, and more time consuming to maintain.

You need to know and understand the tradeoff. Is the very slight reduction of gameplay performance worth the cost to develop the software? That's a business decision, but generally the tradeoff is made in favor of faster/cheaper development.

For example, virtual functions were scoffed when they were introduced in the 1980s. People (correctly) pointed out their performance cost, and (incorrectly) calculated the performance cost as though every function were virtual. Hopefully it is obvious that it is unwise to make every function virtual because there is overhead (roughly 15ns per function call on current hardware), but that doesn't mean virtual functions are bad. Used properly virtual functions solve a problem and generally offer better performance than trying to write your own solution.

Know the cost and use the feature properly.



you need to save time by optimizing parts of the code (assuming you have a good code design) by the very beginning, in parts where you do know that you will have a performance boost


Knuth's saying was that premature optimization is the root of all evil. Wait until you have written it and profiled it before changing it, and make sure you measure afterword to verify your changes improved things.

But that doesn't mean you should intentionally write pessimistic code when you know something is performance-intensive. If you know something will be resource heavy you should plan for it accordingly. That is NOT a premature optimization.



I agree. And I assume you code in C# :)

#4 swiftcoder   Senior Moderators   -  Reputation: 9760

Like
0Likes
Like

Posted 06 September 2012 - 05:17 PM

Please don't cross-post. I'm leaving the other one open, any further discussion should occur there.

Tristam MacDonald - Software Engineer @Amazon - [swiftcoding]





Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



PARTNERS