🔒 Locked

FLT_EPSILON

Started by papa Apr 27, 2004 at 5:19 AM 2 replies 7.4k views

Original Post

Apr 27, 2004 at 5:19 AM

Hi I dont quite understand why FLT_EPSILON contains 1.19209290E-07F. It says that its the smallest positive value so why it starts with 1.1 I've done a simple meaningless test which fails. why? float smallest = FLT_EPSILON; if( smallest < 0.0000001f ) { // It never gets here? why I thought 0.0000001f is still positive and smaller value could be 0.00000001f (one more zero) } so if FLT_EPSILON is the smallest positive value why the if() fails? Is it something to do with the float precission. Obviously I dont understand this clearly can you please explain. I am basically trying to determine if some float X value is not 0.0f but can be near zero before rounding errors occur. Second question.. Would it be actually correct to use epsilon for this case. Is it right test I am doing? Thank you. [edited by - papa on April 27, 2004 6:22:09 AM]

vicviper

Apr 27, 2004 at 5:24 AM

I''m guessing the reason for this is that Epsilon is commonly used as a test previous an operation that could fail if the values are too low.

So, if you want to do a division, you can test the values against epsilon, and know if the result of the operation will be reasonably accurate or not.

papa

Author

papa

Apr 27, 2004 at 5:43 AM

ok. Yes that could be one of the example. But to me 0.0000001f isn't complete 0.0f value. Why the above if statement fails?

Question:
What kind of representation 1.19209290E-07F is that suggests that is smaller than foe eg. 0.0001 or 0.00001?

why it fails once is below 0.0000001f?

[edited by - papa on April 27, 2004 6:44:16 AM]

Jingo

Apr 27, 2004 at 5:49 AM

Your number is

0.0000001

smallest is

0.00000011920929

Which is bigger than your value. So smallest < 0.0000001f is false.

Topic Locked

This topic has been locked by a moderator. New replies are not allowed.