From: Stephan F. <cri...@cs...> - 2013-04-24 21:34:17
|
Hi Akshay, no I don't know a good heuristic for switching to GMP yet and coming up with some benchmarks is one of the next steps, though not getting cheated by the compiler and still knowing what is going on performance-wise is not so easy (as you have just seen yourself :-) Waldek switches for a result bignum-length of 40 (which is the number of words/limbs) for addition, though I find that rather large. Also exercising some random tests to check, whether all the upper bounds for the bignum result pre-allocations are correct, would be useful (since GMP does not always behave as expected wrt. to the required memory size; see e.g. my comment for fib2). Currently, I am integrating ratios, though progress is a bit slow as my work projects drain most of my mental energy. Regs, Stephan |