binary fractions (floats)

BlitzMax Forums/BlitzMax Programming/binary fractions (floats)

Chris C

(Posted 2005) [#1]

0.600000024 is the nearest to 0.6 for a float type but is this the same on all platforms?

ie is this going to work okay on all platforms?
is binary rounding a reliable "feature" ?

if a>0.600000024 then
if a<0.600000024 then
if a=0.600000024 then

where I need 3 distinct states

just for fun!

Local a:Float=0.0
While a<>0.600000024
	a:+0.1
	Print a
Wend

ImaginaryHuman

(Posted 2005) [#2]

Maybe you can use an Integer where the upper 16 bits are the main part of the number, and the lower 16 bits are the decimal part. It wouldn't give you great accuracy either but it would be predictable.

Robert

(Posted 2005) [#3]

I really wouldn't recommend making assumptions like that. With floats you should always check that the number lies within certain bounds, ie. in the above case, test that the float equals 0.6 to a given level of accuracy (say 3dp)

fredborg

(Posted 2005) [#4]

The best way to handle stuff like that is to use an epsilon value, ie:

Const EPSILON:Float = 0.001
Local a:Float = 6.0
Print "a is really: "+a
If Abs(a-6.0)<EPSILON
  Print "a=6.0"
ElseIf a-6.0<0.0
  Print "a<6.0"
Else
  Print "a>6.0"
EndIf

Incidentally "a is really 6.0" on my machine :)

RexRhino

(Posted 2005) [#5]

Also, depending on what you are doing, I have seen source code where people use two ints to store a numerator and denominator.