In my Polynomial RootSolver there are a few expressions in an iteration loop of the form r = s - a*b -c*d.

I see 3 possibilities to calculate this expression:

1) without any tricks in the given form s - a*b -c*d

2) using complex multiplication (cpx*) in the following way: first a*b + c*d (complex) and then subtract from s

3) using cpx* twice: 1*(1*s-a*b) - c*d, i.e. first calculate 1*s-a*b and then 1*result-c*d

Currently I'm using method 2), i.e. (a*b+c*d) complex and then s - (result).

Now I would like to know if method 3) would give a higher precision than 2), or if even none of both complex tricks is better than the standard way 1) ?

Method 1) of course needs less (internal) calculations (no real+imaginary part!), so if there's no difference than I would switch back to 1).

And then one more (principle) question: is there any difference between the precision (i.e. used/stored digits of a number) of the stack and the registers?

I guess there's none (?), but then I don't really see the advantage of using cpx* : what is an internal precision of 34 digits worth, when I only have 16 digit numbers on the stack?

Franz

*Edited: 28 Oct 2011, 8:15 a.m. *