[Bug] Adding profiling to a 'PURE' routine should remove the 'PURE' attribute

[arporter]

When adding profiling to ORCA2_ICE_PISCES, there are numerous warnings about pure routines that now have calls to profiling calipers. The code still builds but we should fix this - adding profiling should remove the 'PURE' attribute on a routine.

[sergisiso]

Agreed, but probably we should not add profiling to functions? This are often call from inside the parallel/target region and adding profiling would make them not parallel/offloadable.

[arporter]

True. This is what I used to do (and maybe still do) in the kernels transformation script. I can extend the PR I've just opened (#3033) to refuse to do functions if you like?

[sergisiso]

Yeah it would make your PR be about any function, I had this problem a couple of times in the past which I always had to go and modify that method.

[arporter]

In testing #3032 I've found that gfortran refuses to compile an ELEMENTAL routine that has had profiling calipers added to it.
I think probably we have to refuse to add profiling calipers to such a routine as otherwise we might have to modify the call site (if we removed the elemental attribute)?