Boris Mansencal wrote:
indeed, return (size_t)(p.x()*P_height + p.y()) seems to give slightly better results... (but actually I still do not really grasp why...)
If P_height happens to be a multiple of the number of buckets in the container and the container uses a plain modulus of the hash value then the x value will have no effect on the chosen bucket. If you want your program to be portable to different container implementations this could be a problem. A possible way to work around this is to apply something like Thomas Wang's integer hash function to the result. http://www.cris.com/~Ttwang/tech/inthash.htm Although the function is specific to the number of bits. But I'm surprised that boost::hash is slow. With full optimization, I'd expect it to work okay. Maybe a simpler hash_value might be better: std::size_t hash_value(point const& p) { std::size_t seed = p.x(); seed ^= (std::size_t) p.y() + (seed<<6) + (seed>>2); return seed; } If possible, I'd be interested in looking at your tests (you can email me off list about them). Daniel