It has nothing to do with the accuracy of the math libraries, the question is if the constant needs to be converted to real*4 before assignment in real*8 x=1.7 as done in most compilers, or if it is OK to assign the real*8 approximation. Joost