Compile with ifx on Derecho

[scrasmussen]

TYPE: bug fix

KEYWORDS: ifx, compiler, Derecho

SOURCE: Soren Rasmussen, NCAR

DESCRIPTION OF CHANGES:

• Intel compilers use non-standard argument inquire(directory=dir,...) to check if a directory exists.
• ifx needed additional use statement for certain functions.
• coscalar allocatable character array causes ICE in ifx.
• 'module subroutine water_simple' needed to be 'subroutine water_simple', seemingly so 'module module_water_simple' doesn't clash in compiler symbol table
• Added -O0 to Intel's debugslow so the following message won't be shown everytime a file is compiled: ifx: remark #10440: Note that use of a debug option without any optimization-level option will turnoff most compiler optimizations similar to use of '-O0'

TESTS CONDUCTED: built and ran test case on Derecho with ifx (distributed and shared arguments), GNU, and Cray compilers

NOTE: Results from testcase with the different compilers on 1 Derecho node:

compiler	num_images	total (s)	init	input	output	physics
ifx distributed	128	100.1	65.3	6.6	0.1	28.1
ifx shared	128	99.7	65.7	6.4	0.2	27.5
gnu	128	117.2	71.7	6.7	0.2	38.5
cray	128	81.8	65.9	7.0	0.2	9.0

Checklist

• [ ] Closes issue #xxxx (An issue must exist or be created to be closed. The issue describes and documents the problem and general solution, the PR describes the technical details of the solution.)
• [ ] Tests added (unit tests and/or regression/integration tests)
• [X] Backwards compatible
• [ ] Requires new files? If so, how to generate them.
• [ ] Fully documented

[gutmann]

Thanks Soren, any chance you can run those simulations longer so that the "physics" part is 99% of the runtime? That is the important part to document. Currently the total runtime is dominated by the init, which in a "real" (~12hr) run will be a tiny fraction of compute.

[gutmann]

can you also run those tests on Cheyenne (same number of cores) with gfortran for reference?