Wed Jul 30 19:18:35 EDT 2014

Linear interpolation

It seems best to encode the tables to make the computation as simple
as possible, i.e. as   a x + b,   where x is the fractional component

does that make sense?

number of table components doubles, but that isn't really an issue.

the problem is really the fractional point.
how many bits are needed anyway?  say theres a 12-step power of 2 table.

wait, what about picking the table size based on the needed precision?

i.e. say we want to just keep the top 8 bits to add to a 16bit number
that's a 1/256 accuracy.

not the right track..