On Feb 3, 2008 1:28 PM, Bruno Lalande <bruno.lalande@gmail.com> wrote:
I can write tests and docs for you, as well as study the different performance / optimization aspects with some other compilers. As I've never done this before, I'll have to read the different guidelines describing the way in which such contributions have to be made, so it can take a few days. But I think we're not in a hurry.
Steven, could you tell me if some aspects of the implementation you made for Boost.Unit are better than mine? This would permit to be sure to have the best implementation in Boost.Math.
I've usually seen it described like this: template <int N> struct positive_power { template <typename T> static float result(T base) { return (N%2) ? positive_power<1>::result(base) * positive_power<N-1>::result(base) : positive_power<2>::result( positive_power<N/2>::result(base) ); } }; I have no idea whether that would work better or not. It would mean more template instantiations, of course, but avoiding storing it in a named temporary might make the optimizer happier. (Are there provisions for allowing different temporary sizes in floating point math? I seem to recall a flag for gcc that forces storing all the intermediate results to prevent this, so there might be.) I'd also wonder whether -ffast-math or similar flags make a difference.