Fix a build issue with nvidia compiler

[ndkeen]

With nvidia compiler on pm-cpu, we see new build fails looking for definition of isnan.

Fixes #7820

[bfb]

[bishtgautam]

@ndkeen The label should be pm-gpu instead of pm-cpu. Right?

[ndkeen]

Well the e3sm_developer tests that we run every other day are on pm-cpu. So these are all CPU builds. There may be same issue on pm-gpu, but we still only use GNU compiler there.

[peterdschwartz]

Yeah, these NaN checks need to be removed. There's no valid reason for putting them inside hot loops like so. shr_infnan_isnan is a pure elemental procedure: meaning it can work on scalars and arrays of any dimension:

! For scalars:
if (isnan(x)) call endrun("x is NaN")

! For arrays
if ( any(isnan(arr1d)) ) call endrun("arr1d has NaN")
bad_cols => any(isnan(arr2d), dim=1)   ! per-column NaN flags
nbad = count(isnan(arr3d))             ! total NaNs

Not to mention debugging flags can already cause a backtrace when encountering a NaN.

[peterdschwartz]

merged to next