bfloat16 proposal

[certik]

The latest proposal is at https://j3-fortran.org/doc/year/19/19-221r1.txt.

[certik]

Regarding the vote from the proposal:

Straw votes:
  1) Y/N  Should we proceed with with specs and syntax?
  2) If (1) is YES:  how should BFLOAT16 be described?
     (a) unspecified (aside from 16 bits)
     (b) reference non iso (probably not allowed)
     (c) explicit definition in terms of bits?

The committee voted on 10/15/2019 as No on 1), with the reasoning that the arithmetic is not well specified yet (each hardware vendor can implement it differently), so currently the bfloat16 would act as a "placeholder" in Fortran. And that is not a good enough reason. As bfloat16 usage is more common, the committee can revisit adding it to the language.

[certik]

The committee discussed https://j3-fortran.org/doc/year/20/20-118.txt on Thursday 2/27/2020.

[zjibben]

As per the paper, JoR recommends not pursuing BFLOAT16 for now. The sentiment was echoed, especially concerning lack of a standardized definition for BFLOAT16, and the paper was passed unanimously. The conclusion (to my understanding) is that the committee will not proceed with BFLOAT16.

[klausler]

If your environment supports bfloat16, you should be able to get at it via SELECTED_REAL_KIND(2,37) and confirm it via DIGITS() == 8.