Changes

Openacc tweaks and cleanup 2 (#1099)

* Making all end parallel directives specify end loop

* Replacing last acc declare in a procedure with acc enter/exit data commands

* Removing pure declarations, turns out they dont really improve performance and openmp isn't allowed within them.

* Adding explicit directives to penta_lu and tridiag_lu, as opposed to the previously used kernels directive. Also splitting up a loop to improve GPU performance.

* Splitting up a couple loops for performance reasons.

* Missed one somehow