Floats vs Integer Scaling

[pmprog]

One of my projects is vector graphic based, and am wanting to implement scaling so the player can zoom in and out nicely. Everybody always says floats are "expensive" on ARM chips; so I was looking at other integer methods (like this one).


Which would be better? What if I used floating point, but cached the results, so I'd only need to recalculate if the scale changed?


Cheers

 

[mcobit]

There are neon shaders already available.


You just have to search here on the forums.


Edit: Found the thread:


http://boards.openpa.../page__hl__neon


Edit2: Just saw, that this is not really what you want, so disregard me

 

[sebt3]

just playing with the gles vewport would be easier

 

[pmprog]

Well, I'm currently using Sparrow3D to draw primitives (lines, ellipsis, and rectangles). Managed to implement a Q16.8 integer maths library.


But then GLES uses floating point math, so is it that floating point isn't that big a deal on ARM? Is it the rendering that's the bottleneck?