While using OpenACC with a GPU accelerator, the coding style affects the performance.
When accelerating an OpenACC kernel with a subroutine, it is performance critical to compile with inlining (Fortran).
However, if the subroutine comes from an external library, the (Fortran) compiler doesn't always know how to perform inlining.
This is an example comparing the performance difference caused by the inlining problem mentioned above, as well as how to solve the problem.