fixed fract from float data (separation) on SHARC

[kirgizz]

Hello,

I've got a problem with data format conversion on ADSP 21262 which can process both formats, fixed and floating point data. I give you an example for it:

The floating point number is: 200.15 (f15=200.25).
The DSP represents it as (1).564453125*2^^7 (S=0, E=134, M=.564453125) in floating point format.
I can get the integer part by r12=fix f15.

Here is the most exciting thing: how can I get the real fractional part (=.25) of the number?
Doing
r4= mant f15; //manissa from f15
r5= lshift r4 by 1; //removing hidden bit

results in r5=0.564453125 (as unsigned fractional)

Do anyone have an idea?

Thank you in advance

 

[brmadhukar]

200.25 = 1.5.. times 2^(something)
unless you use 2^something to multiply == shifting by all available exponents we will not get the proper result. you are just taking the fractional part of the normalized number. YOU ARE EXPECTED TO GET A WRONG RESULT.

B R Madhukar

 

[kirgizz]

thank you, brmadhukar.

Excuse me, my question was not correct. It means how can I get the decimal fraction (.25) from normalized floating? The hidden "1." causes trouble to me.

So I need a fast algorithm to separate the decimal fraction from a floating point data.
Anyway, the compiler can do this (from decimal in normalized float).