Re: [pygame] Fwded Post

[Pete Shinners <shredwheat@mediaone.net>]

Guillaume Proux wrote:
> Talking about floating point values, my background (game,demo) is the 
> Amiga, the little
> one that did not come with a MMU nor a FPU. So floating point 
> computations were VERY
> expensive and we would most of the time do 32 bits fixed point 
> arithmetics for that.
>
> Nobody uses this technique apart from me?

integer arithmetic is good like this for small tight loops. in fact many 
of pygame's internal functions like rotate/scale/draw all use integer 
arithmetic for extra speed.

i've always been very curious if integer arithmetic in python was really 
any faster than regular float/int conversions. since i'm in the 
benchmarking mood apparantly, i did a quick test to see what's up..

my initial thought would be they come out pretty close in speed. to my 
surprise, the all-integer version comes out twice as fast. wow.

FLOAT TIME:	2.81134796143
INT TIME:	1.44705092907
<<with python-2.1 on a 800mhz celeron2>>
the test just increments a numeric value by "1/8"th many times. i don't 
think this means everyone should go out and change their games from 
using floating point positions to integer. in a game, the amount of time 
spend moving the objects is so tiny, that cutting it in half probably 
won't even show up on the frames-per-second.

on the other hand, if you really do have a lot of game objects (like a 
particle system especially) keeping track of the positions this way 
could bring some speed to you. it might be worth investigating there.

in any event, i'd recommend _not_ doing the math with integer arithmetic 
like this until you are happy with how everything works with regular 
floating point. the code can be a lot harder to read and work with.

#test speed between int shifting and float-int conversion



def float_test():
    inc = .125
    pos = 0.0
    for x in range(800000):
        pos += inc
        int_pos = int(pos)
    print "  final pos = ", int_pos

def int_test():
    inc = 1
    pos = 0<<3
    for x in range(800000):
        pos += inc
        int_pos = pos>>3
    print "  final pos = ", int_pos


import time
start = time.time()
float_test()
print 'FLOAT TIME:\t', time.time() - start

start = time.time()
int_test()
print 'INT TIME:\t', time.time() - start