Many of the functions in this library exist in specially optimized versions written in machine language for various CPU/FPU architectures, for maximum performance. There is also the distinction between machines using IEEE floating point format internally and these that use a different format, which is reflected in the source.