Re: [Enigma-devel] Maemo Port for Nokia Internet Tablets

[Ronald Lamprecht]

Hi,

Pipeline wrote:
> After some playtesting, i did manage to find a few areas (although these 
> are more rare now) where performance is an issue.  A simple example 
> would be Enigma II - 20 Meditation.  On this level it begins very 
> smoothly, only after setting four balls into the hollows does a 
> performance issue become apparent, at which time the performance drops 
> to around 3 fps and setting a fifth ball reduces performance to around 
> 2fps and a sixth perhaps 1fps.  So the holes (hollows) seem to be adding 
> a significant amount of computation overhead which the 400mhz ARM cpu 
> can't keep up with.  I tried looking into the source code however this 
> is rather complicated to determine what might be causing.  I am guessing 
> this is some sort of math calculation which adds forces when in hollows 
> but i can't pinpoint.  I have tried compiling with flags to enable 
> hardware floating point ( -mcpu=arm1136j-s -mfpu=vfp -mfloat-abi=softfp) 
> which has worked well on other ports but makes no difference when added 
> to src and lua project makefiles.
>
> Any ideas on whether this is avoidable?  Or could it be the default 
> maemo libm (which, i believe, does not take advantage of floating point) 
> is the culprit?

I thought I had eliminated all algorithms with quadratic behaviour on 
the number of actors :-(

But I never checked the hollows thoroughly! Indeed I just dected by a 
quick code review a totally unecessary quadratic algorithm that can be 
replaced by a linear one. Thanks for the report!

Greets,

Ronald