Disrepancy in results between OpenMP/OpenACC implementation and gcc/PGI compilers

Question

I have a larger Fortran program that I am trying to convert so that the computationally intensive part will run on an NVidia GPU using OpenMP and/or OpenACC. During development