All listed functions from Math387 unit support real and complex numbers (where appropriate). Most functions for math in Delphi are now overloaded, to minimize the number of parameters required and function names to remember. All functions are capable of handling NAN and INF values and use of FWAIT instruction is therefore omitted, thus providing further performance gains. The unit also declares a large subset of typed constants for real and complex arithmetics. It supports both double and single floating point precision. Most of the function's were rewritten in assembly language (as the name 387 suggests). The performance gains are especially large for functions taking complex arguments. This unit does not provide any vector/matrix arithmetics (routines working on arrays of elements) and therefore can not make use of MMX/MMX2/SSE/SSE2/3DNOW. Most of the routines do not save intermediate results to memory, thus preserving full 80 bit (20 decimal places) precision. This improves the precision accuracy of the final result by up to 5 decimal places with double precision. (for example from 10 to 15).