Optimization Tips

[random10122]

Ahhh... right, sorry. So this only offers a performance increase if the state of a key has to be checked more than once in a frame.

Gotya.

[Frase69]

I'm not sure if this has already been explained in thread but if it has, I don't think I got it, plz try help.

If I have a bullet MC, I need to store the hitTest action in it for when it hits people MC's.

say the people MC's are named:

man1
man2
man3

etc...

man29
man30

Well instead of 30 If hitTest actions can't I simply have one which will be triggered if the bullet hits any of them?

Plz try explain this as best you can.

Thanks for your help.

[random10122]

ripX tells me that multi-dimensional arrays are slower than just the one dimension... But, is a 3D array slower than a 2D and faster than a 4D?

[iopred]

of course, because you have to look up the first dimension, then the second, then the third..

However I would say the speed loss is negligable..

Also here is a little speed boost I have got..

lets say we have a 2d array which is rather wide:


map = Array100x100();
for(var y = 0; y < 100 ;y++){
var dmap = map[y];
for(var x = 0; x < 100 ; x++){
trace(dmap[x]);
}
}

Bit faster. Would cut off the difference between dimensions IMHO.

[random10122]

cheers preddy.

[Squize]

Drunken post, which I really shouldn't do in this thread...

If you look at multi dim arrays in byte code it's frightening how much data has to be pulled up before actually getting the value you want.

Where poss. you should try and pre-calc the multi-dim values you want ( ie 99 times out of a 100 it's going to be map data ) and transfer it to a single dim array.

Squize.

PS. As ever, sexy code pred. Just once post something lame

[fantasio]

hello everyone,
i just lurked so far and read all of those nice tips, some of which I already knew, some new ones.

Now, a question:
does it make sense to incorporate all those tips in my current project with flash 2004 around the corner and with that possible performance increases ???

In my last projects I tried to go further down the road of OOP vs. my so far chaotic, but effective scripting.

Now a lot of performance issues are due this style of scripting.
For example the "dot" syntax is very cool to work with but a lot of times, the tell target is much faster.
But I think it's a pain in the butt to use so many different approaches to coding all the time....

Whaddaya think, is MM getting their act together or do I have to re-code every given project after i finished it, to make it perform better (very time consuming that...).

Also, maybe we should collect all known and working optimisations in a FAQ ?

[RipX]

With regard to multi-dimentional arrays being slow, I quickly wrote this which I was talking about to Random.

var mapL = 3;
var mapH = 3;
var l = 0;
map = [1, 2, 1, 2, 1, 2, 1, 2, 1];
for (var i = 0; i<mapL; i++) {
	for (var j = 0; j<mapH; j++) {
		t = attachMovie("tile", "tile"+i+"_"+j, d++);
		t._x = i*32;
		t._y = j*32;
		t.gotoAndStop(map[l]);
		l++;
	}
}

With a map editor this information can be produced easily and keep the array fast and 1 dimentional.

RipX

[strille]

Yes, 1 dimensional arrays are generally a bit faster. Since you have the map in one array, you can loop through it using only one loop:

code:

---

var mapL = 3;

map = [1, 2, 1, 2, 1, 2, 1, 2, 1];



for (var i = 0; i<map.length; i++) {

		t = attachMovie("tile", "tile"+i, i);

		t._x = (i % mapL) * 32;

		t._y = int(i/mapL) * 32;

		t.gotoAndStop(map[i]);

}

---

The advantage is that you can remove the variables l and d and use i instead. And you remove one nested for loop, which is good. You have to do some more calculations to place the tile, but the overall code runs a bit faster than the code RipX posted above.

[RipX]

I'd actually thought of that way but I didn't want to complicate things too much since I wanted to show just the one optimation! But nice one Strille for posting that

RipX

[Squize]

fantasio good question.

I'm still of the mindset that if a game works, then don't kill yourself trying to optimise it.

Use tellTarget and other straight forward optimisations where poss. but I've not sat in front of a routine for a while and tried to skim a few ms off the execution time.
I find having a think about the best way to actually write a routine before hand is a million times easier than writing something and then trying to optimise it. May be a really blatant statement, but it's something I've never really worried about too much before 'cause I always had the time to optimise things, but things have changed recently ( I've got into the habit of using quite long var names, which is a speed no-no, but it saves me so much time when it comes to debugging that I can cope with losing the odd ms ).
If I do actually optimise something then it'll be the always called inner loops.

As to the faq, mBenney is working on one atm ( I've stupidly said I'd do the OOP section ).

Going back to the 1 dim arrays for maps, how do you guys plan to include the attributes ? Skip every other byte or have a seperate attribute array ?
To be honest I think it'll be easier to just pre-calculate the current map data you need and store that in a temp. 1 dim array ( Using it in effect like a stack ).

Squize.

[iopred]

Hey guys,

Just doing some testing...

timer = getTimer();
var temp = 0;
for (var i = 0; i<100000; i++) {
	if(i && i && i){
		temp++;
	}
}
timer = getTimer()-timer;

A simple if statement, seeing if i is not 0, 3 times. This runs at about 5600ms in Flash's inbuilt player.

timer = getTimer();
var temp = 0;
for (var i = 0; i<100000; i++) {
	if(i){
		if(i){
			if(i){
				temp++;
			}
		}
	}
}
timer = getTimer()-timer;

The EXACT same code, but with the if's expanded results in a cost of 4800ms.. Thats a fair bit!

My conclusion, Flash's compiler stinks so hard, I want to puke.

[BlinkOk]

is that with the "Omit trace actions" set or clear in the publish settings?

[iopred]

Urg, kill me.. Didnt have it checked =|

[BlinkOk]

Still it's important to have it checked. i, for one, never bothered with it but you can bet anything i make now will have it checked.

[Squize]

It's always quicker to split your If statements up ( And avoid and,or ) because Flash has to get all the data needed for ALL the checks before it can perform even the first one, so with an if statement with 3 conditionals to check, even if the first one fails to meet your condition it's still got another 2 lots of data first ( And if you throw some array[offset] checks into that as well it's a stupid amount of code generated ).

As a follow up question, I've seen a lot of people use the switch statement a lot recently, although there's not meant to be any performance improvement from it. Anyone know for sure ? Cheers

Squize.

[strille]

Squize, I made a comparison between if-else and switch, and it appear that they are in fact equally fast (or slow, this is Flash we're talking about )

This is a bit surprising, one would think that switch would be faster since you evaluate the value once and then jump to the right code, while if-else has to evaluate each if-statement until one is true or there are no more. Could it be that the switch-statement is compiled as a series of if-else statements?

Anyway, here's the code I used:

code:

---

a = 4;



time = getTimer();

for (var n=0;n<10000;n++) {

	if (a == 1) {

		b = a;

	} else if (a == 2) {

		b = a;

	} else if (a == 3) {

		b = a;

	} else if (a == 4) {

		b = a;

	}

}

trace("if-else: "+(getTimer()-time));



time = getTimer();

for (var n=0;n<10000;n++) {

	switch(a) {

		case 1:

			b = a;

			break;

		case 2:

			b = a;

			break;

		case 3:

			b = a;

			break;

		case 4:

			b = a;

			break;

	}

}

trace("switch: "+(getTimer()-time));

---

In either case, I think switch looks better...

[Squize]

Thanks mate!

You're just a sexy actionscript writing machine atm

Squize.

[random10122]

Just thought i would drop a note to say that i have collected the most usefull tips from this thread in a page on my webspace as a list split into three sections: graphics/code/misc

Just thought i would make it easier to search through. (Link in sig)

[strille]

random10122, that's a very good idea. Reading through this whole thread can probably be a bit dry...

Squize, I'm trying my best to be as sexy as humanly possible.

Btw Squize, I'm naming you the current champion of the use of the word "sexy" in your posts:

Squize: 16 (887 posts)
BlinkOk: 5 (2785 posts)
Strille: 2 (620 posts)
I_am_TheFlasher: 1 (400 posts)
nGFX: 1 (159 posts)
Saphua: 1 (381 posts)
Tonypa: 1 (2425 posts)
marmotte: 0 (1562 posts)
iopred: 0 (64 posts)

If someone can find a Games member with a higher count before this post was posted, let me know