Tuning

[Josh Liu]

I have a working multiple precision arithmetic library that I have 
written over the last six months. It is based upon the threshold system 
that GMP uses to partition the effective algorithm used for certain 
input sets. What is the best way to find these thresholds? I know that 
a binary search is definitely out of the question as the inaccuracies 
make the comparison function poorly defined. How does tuneup work? I 
want to know how the precision parameter comes into play as well. The 
other possibility is to do curve fitting of the time function and solve 
for the thresholds. This possibility would reduce the inherent 
inaccuracies by working with well behaved functions, instead of the 
inaccurate and seemingly random nature of empirical timing.

http://www.student.gsu.edu/~zliu2/centrinia.html
-Josh Liu