Well, you think is a bug, but without stating as such, Intel folks via @greenrongreen appear to indicate to you it is a “feature”!!
See this comment of mine from weeks ago to @greenrongreen :
So my view is what Intel Fortran does with arrays of default REAL type with floating point divisions violates the principle of least astonishment (POLA) with software design by Intel, they are asking you to change your compiler options here to get things right rather than getting expected behavior by default or with the ones you have (some of which do make no sense, that’s a separate matter):
-prec-div is another compiler option
you can consider per Intel.
But the lack of response by @greenrongreen to my earlier inquiry re: a bug here suggests Intel is not bothered by POLA here, so that is that!
So you can take it or leave it
Posting multiple cases of the same underlying aspect is not kosher, if you want a change, pay for a support subscription to Intel, submit a support request on this, and wait for Intel’s response.
In the meantime, use the suggested compiler option(s) and move on.