Avoiding Denormals
From ActionScriptWiki
Denormal numbers are floating point numbers that are really small. When such a number occurs the CPU switches into denormal mode and calculations become very slow.
Contents |
[edit] Example
In this example the framerate will jump from approximately 40fps to 1fps. The reason are denormal numbers.
private var _denormals: Vector.<Number>;
private function init(): void
{
_denormals = new Vector.<Number>( 0xfffff, true );
var n: int = 0xfffff;
while( --n > -1 )
_denormals[ n ] = 1e-256;//very small number!
EnterFrameProvider.connect( enterFrame );
}
private function enterFrame(): void
{
var n: int = 0xfffff;
while( --n > -1 )
_denormals[ n ] *= 0.5;//do some calculations with veery small numbers
}
[edit] ActionScript Solution
There are a couple of solutions to deal with denormal numbers. Some try to avoid them by adding always a very small value or by adding some white noise. In ActionScript the best and most simple solution is by adding and subtracting an anti-denormal constant.
[edit] Solution
This solution is easy to implement in ActionScript and will result in real zero values. It has been tested successfully on
- Mac OS X (Intel CPU)
- Windows/Linux (Intel/AMD CPU)
x = x + 1e-18 - 1e-18;
[edit] Example
This is the same example with the fix for denormal numbers. The framerate will now stay at approximately 40fps.
private var _denormals: Vector.<Number>;
private function init(): void
{
_denormals = new Vector.<Number>( 0xfffff, true );
var n: int = 0xfffff;
while( --n > -1 )
_denormals[ n ] = 1e-256;//very small number!
EnterFrameProvider.connect( enterFrame );
}
private function enterFrame(): void
{
var n: int = 0xfffff;
while( --n > -1 )
_denormals[ n ] = _denormals[ n ] * 0.5 + 1e-18 - 1e-18;//avoiding denormals
}
