Double precision square root function

Hi there,

I’m porting a code which contains quite a lot of double precision square roots dsqrt.

I’ve trawled the cuda fortran programming guide and it seems there is no supported dsqrt or I just can’t find it. Does anyone know anyway around this problem or if it is supported, as it would seriously effect the accuracy of my code if i was to downgrade to single precision.


Problem solved: I read through the FAQ on precision problems and realised that the compiler treats constants as real4 even when declared as real8. Adding the compiler flag -r8 solved the problem.