C2000 floating point issue: x/x not always 1

Johannes von Löwis

I'm using cgtools 5.2.10 and rts2800_fpu32_fast_supplement.lib on an f28335.

I observe that the following code

float q, x = 0.0434;
int y;
q = x/x;
y = q;

gives q slightly less than 1 and consequently y == 0.

Now the question: Is this a bug in the division routine or is it wrong to assume that the quotient of two equal floats (aside from 0, NaN and so on) is exactly 1?

Regards,

Johannes

over 12 years ago

0 Archaeologist over 12 years ago

TI__Guru* 84285 points

Aside from the special cases, you should be guaranteed that x/x==1. What result do you get when you do not use rts2800_fpu32_fast_supplement.lib, and instead use rts2800_fpu32.lib? If you get the correct result with rts2800_fpu32.lib, this is not a compiler bug, and we'll need to move this thread to the correct forum.

0 Johannes von Löwis over 12 years ago in reply to Archaeologist

Expert 1806 points

Archaeologist said:

Aside from the special cases, you should be guaranteed that x/x==1.

Good to know...

Archaeologist said:

What result do you get when you do not use rts2800_fpu32_fast_supplement.lib, and instead use rts2800_fpu32.lib? If you get the correct result with rts2800_fpu32.lib, this is not a compiler bug, and we'll need to move this thread to the correct forum.

I just checked: With rts2800_fpu32.lib and x=0.0434 I get 1.0 for x/x.

BTW, I'm using Version 1.00 of rts2800_fpu32_fast_supplement.lib.

Regards,

Johannes