[flang-commits] [flang] Parallel runtime library design doc (PRIF) (PR #76088)

[Peter Klausler via flang-commits]

klausler wrote:

> @klausler wrote:
> 
> > Does the latest revision support APIs for the two features that I mentioned above (remote address calculation and split asynchronous transactions)? If so, great, ship it.
> 
> Not yet. As explained in [this comment](https://github.com/llvm/llvm-project/pull/76088#issuecomment-2420482329), both planned refinements are explicitly listed as future work in section 7 of the document.
> 
> Our current goal is to reach consensus on the initial revision of a compilation target interface that is _sufficient_ to correctly express the entire parallel feature set of Fortran 2023 (as outlined in Section 2). We believe PRIF revision 0.5 achieves that goal. We fully expect this API to evolve in future revisions to adopt whatever extensions are needed to express explicit optimization capabilities, but we need to start somewhere.

Ok, thanks for the confirmation.  But perhaps you can see how this puts me into a difficult spot as I consider the implementation of coarrays in lowering and runtime in the upcoming months once I have semantics cleaned up.  I'll need to either (1) do something that doesn't get top performance on our own gear but is upstreamable to LLVM because it uses this API, or something that can perform well and either(2) isn't going to be upstreamed or (3) is upstreamable but uses this API as one of several transport options behind a new lowering/runtime API.  None of these is ideal, obviously.

https://github.com/llvm/llvm-project/pull/76088