Double precision power function error on GPU (CUDA Fortran)

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

Hi I’m working on some fluid dynamics simulation using NVIDIA GTX-750 Ti on Windows and NVIDIA GPUs, GTX-1080 on Linux (Ubuntu 16.04).
My problem is, when I calculate double precision power function, like 8756.3425^(-0.43), it works well on Windows. However, when I compile the same source code on Linux with the same compiler options, and run the program, the code gives NaN value instead of true one.
I wonder why calculation breaks down on Linux, while it works well on Windows.
In addition, I found some information about ‘LIBM’ module and tried using ‘POW’ function in the module, but it also gave NaN value.

Thanks.