18980, "bradcray", "Should Chapel support promoted on-clauses?", "2022-01-12T23:09:54Z"
While working on slides today, I found myself wondering whether Chapel should support
promoted on-clauses as follows:
on Locales[0..<numLocales-2] do ...
which we might imagine to most literally be interpreted as:
forall loc in Locales[0..<numLocales/2] do
on loc do ...
but we'd probably prefer to define it as:
coforall loc in Locales[0..<numLocales/2] do
on loc do ...
To the best of my memory, this hasn't been kicked around before. But is
there any reason not to consider it? It makes explicit parallelism implicit,
but in that sense is not so different from other forms of promotion.