These are chat archives for symengine/symengine
mpz_classfor storing that packed result for now
So the plan would be:
1) send a PR for Piranha in cmake
2) code for packing exponents into machine int64 (send a PR and we'll discuss details on it as we go)
3) try to use Piranha's integer to see how it performs -- we'll write our own class Integer, where in it we switch between int and mpz_class automatically, and we should use it everywhere in SymEngine, instead of mpz_class.
4) I would tackle faster hash table later, but if you are interested, we can do it as well.
We'll compare against Piranha, and if we get similar timings, then we can start putting it into the Polynomial class.
Polynomial, right? the value of the hashmap
@certik @thilinarmtb For some functions, like atanh, atan, I've added more than one algorithms to compute the series( formula based or trick based). I tried to benchmark them, but have not yet been able to make a proper comparison (some work better in certain cases and so on).
This paper says Newton's method works better in certain cases. I think it is better to keep a single algo for now. We can add other algos once we figure the conditions in which they are more efficient. What do you suggest?
rs_atanh. Now, we can test them separately.