Add quiet nan and signaling nan options to FieldFill.

[danrosen25]

Questions

• If we add nan (a.k.a. quiet nan) and snan (a.k.a. signaling nan) options to the FieldFill will it help developers and scientists initialize fields or background data and see data issues? Yes, we believe this to be helpful.
• Is the ieee_arithmetic intrinsic module available with all compilers? IEEE arithmetic is supported for all Fortran compilers listed there except g95 https://fortranwiki.org/fortran/show/Fortran+2003+status

Testing

I've added a test to ESMF_FieldRegridUTest. There's also a snan (signaling nan) test that will fail IF floating point exceptions are added to the build; this test is turned off.
Intel fpe flags

+ESMF_F90OPTFLAG_G       += -traceback -check arg_temp_created,bounds,format,output_conversion,stack,uninit -fpe0
+ESMF_CXXOPTFLAG_G       += -traceback -Wcheck -fp-trap=common

Testing complete with gfortran (mac os) and intel (linux)

[billsacks]

Thanks for this. I like the idea of providing a mechanism for initializing fields to NaN.

According to https://fortranwiki.org/fortran/show/Fortran+2003+status, IEEE arithmetic is supported for all Fortran compilers listed there except g95 (is that compiler actually still used anywhere?)... though there may be some other compilers not listed in the table, and we have definitely found instances where this table lists a compiler as supporting something when in fact the support is incomplete / buggy.

For what it's worth, here is the inf/nan module we use in CESM:

https://github.com/escomp/CESM_share/blob/main/src/shr_infnan_mod.F90.in

It's quite possible that it is out of date in terms of its list of compilers that support ieee arithmetic - e.g., you found that gnu supports it, though it's possible that its support isn't complete (see also my comment in the next paragraph).

Also, as a side-note: When working with NaNs in gfortran, I ran into this issue: ESMCI/ccs_config_cesm#4. I don't think this should affect the changes here, but is something to be aware of when working with NaNs.

I'd suggest that, if we include this, then we include at least one unit test that can indicate whether this is working for a given machine/compiler – e.g., setting a field to NaN and then checking that the value is NaN.

[danrosen25]

@theurich @billsacks
I cleaned up the code and I wanted to build a test that would fail due to a signaling NaN (see link) but when I run it on a Mac it doesn't fail. I learned that Macs treat signaling NaNs like quiet NaNs. I don't really know where this test belongs but I think it could eventually end up in the esmx-app-prototypes. I could also set a desired exit code for a test in ctest but that's a problem because this test won't fail on Mac OS.
https://github.com/esmf-org/nuopc-app-prototypes/tree/feature/snan/ESMX_NaN

[danrosen25]

@danrosen25 Add unit test to FieldFill for filling nan and snan.

[theurich]

Code changes look good, and results under https://earthsystemmodeling.org/esmf-test-summary/feature_fill_nan/index.html are as expected.

This is ready to be merged into develop.

[theurich]

@danrosen25 after this PR is closed, please open a new one for looking at adding additional compiler flags under under the ESMF_BOPT=g option. Thanks.

[danrosen25]

@theurich Should we worry about any of these results?

Combo No.	Machine	OS	Compiler	Version	MPI	Version	BOPT	NetCDF	v8.5.0b18-9-g15f3b88
18	breve	Linux	aocc	3.2.0	mpiuni	None	O	None	36
24	cheyenne	Linux	nvhpc	21.11	openmpi	4.1.1	O	4.8.1	916
26	cheyenne	Linux	nvhpc	22.2	mpt	2.25	O	4.8.1	914
29	cheyenne	Linux	nvhpc	22.2	mpiuni	None	O	4.8.1	643
52	discover	Linux	pgi	20.4	mpiuni	None	O	None	631
87	grits	Darwin	gfortranclang	12.2.0_14.0.0	mpich	4.1.1	O	4.9.2	3
89	gust	Unicos	cce	15.0.1	mpi	8.1.25	O	4.9.1	81
92	gust	Unicos	cce	15.0.1	mpiuni	None	O	4.9.1	78
93	gust	Unicos	nvhpc	23.1	mpi	8.1.25	O	4.9.1	10
96	gust	Unicos	nvhpc	23.1	mpiuni	None	O	4.9.1	2
109	hera	Linux	pgi	19.1	intelmpi	2019.0.5	O	None	854
114	hera	Linux	pgi	19.1	mpiuni	None	O	None	631
134	jet	Linux	nvhpc	22.2	intelmpi	2022.3.0	O	None	856
135	jet	Linux	nvhpc	22.2	mpiuni	None	O	None	631
142	orion	Linux	pgi	2019	openmpi	4.0.2	O	4.7.4	914
149	perlmutter	Unicos	aocc	3.2.0	mpiuni	None	O	4.9.0	36
150	perlmutter	Unicos	aocc	3.2.0	mpi	8.1.22	O	4.9.0	94

[theurich]

No, they are as expected. The branch is behind a fix that went into develop recently to bring many of those counts down. I did not see anything in the results I looked at that would indicate there is any issue with the NaNs, especially not on a fundamental level, like not finding the ieee module, etc. Please merge into develop at your convenience. Thanks.