switched from float to double the performance advantage from float over double is negligible (if it exit at all - there are tons of discussion on this issue in the internet) on 64bit architectures but we run much to often into numerical problems when using float.