To judge results of functions coded in double precision against precise results (as they may be given by Maple) one has to respect decimal presentations on one side and IEEE on the other side.
For that one can use routines developed by Florent de Dinechin, which are worth to be better known.
Here is a Maple sheet showing how one can do it (looking at the complex valued power function using evalhf versus using MS VC2005 as an example).
Download 102_checking_double-precision_results__convert_to_nearest_IEEE.mws
Download 102_checking_double-precision_results__convert_to_nearest_IEEE.pdf
