Thread: Floating-Point (BCD)

Hi there!

I'm trying to implement fp numbers that can be approximated in C with this structure:

Code:

struct fp
{
     char            sign;          // 1 = negative
     unsigned short  exponent;      // 32768 = 10**0
     unsigned char   mantissa[8];   // Packed-BCD (16 digits)
}

E.g. p [pi] would be stored as:

Code:

fp.sign = 0;
fp.exponent = 32768
fp.mantissa = {0x31, 0x41, 0x59, 0x26, 0x53, 0x58, 0x97, 0x93};

And what I would like to know is how to implement fast multiplication, division, and exponentiation functions.

P.S. There isn't any FPU, in case you were wondering.