Quote:Original post by AnitraZ
Are you telling me to "simulate" a larger IEEE float functioning?

Yes thats basically it, 'cept you dont have to mess with the dirty details (such as an implied bit, a biased exponent, or obsessive normalization)

While it might seem intimidating at first, its really not. Its only slightly more complex than 'fixed point' math.

You can work it out on paper (or in notepad) by using scientific notation (base-10 rather than base-2) and a bit of reasoning ability.. for instance:

x = 4E+12
y = 6E+14

multiplication:

x * y = (4 * 6)E+(12+14) = 24E+26

having worked it out, if you define a structure with '.m' being the mantissa and '.e' being the exponent, then multiplication in general:

z.m = x.m * y.m;
z.e = x.e + y.e;

From here its just a matter of avoiding (or handling) overflows in the mantissa (its highly unlikely that you will overflow a 64-bit exponent in any reasonable calculation)

How about rather than having events based on the absolute position of the frame you could do amount of frames until the next event. That could result in smaller managable numbers.

I wrote a large number class in C# which can count up to 1e3000, though I wouldn't recommend starting from 1 ;). I really ought to write it up and post it on CodeProject or something, but if you're interested I'll post the code.