To: J3                                                     J3/19-217
From: Bill Long
Subject: HPC Feature recommendations to WG5
Date: 2019-August-08



HPC subgroup considered 5 proposed features.

1) UK01 - A request from the UK to add an optional argument to the
          C_F_POINTER procedure in the iso_c_binding module to enable
          specification of lower bounds for the created pointer.

          HPC subgroup recommends accepting this feature.
          Edits to implement the feature are provided in 19-194r1.


2) US12 - A request to relax existing coarray rules to allow an
          allocatable array of a type that an ultimate coarray
          component. Paper 19-208r1 and possible revisions.

          HPC subgroup recommends accepting this feature.


3) US13 - PUT_with_NOTIFY feature. See paper 18-277r1.  This proposal
          is the HPC feature that is entirely focused on performance.

          HPC subgroup recommends accepting this feature.


4) US23 - BITS type feature. Alternatives offered are detailed in
          19-211, 19-212, and 19-213. Part 1 is discussed in 19-211
          and is very small. Part 3 is discussed in 19-213. If Part 3
          is accepted, Part 2 (19-212) is redundant. Part 2 is a
          relatively small feature.

          HPC subgroup recommends accepting Part 1 of this feature.
          Subgroup recommends accepting either Part 2 or Part 3, but
          defers to WG5 to choose which one.


5) US28 - Request to specify a standard layout of the C descriptor
          used to represent objects for C interoperability.  Currently
          each compiler and its corresponding C companion compiler
          have flexibility to add struct members between the mandated
          members at the beginning of the struct and the dimension
          triples at the end.  HPC subgroup discussed the various
          problems implementation of this proposal would cause,
          including vendor resistance, causing backward
          incompatibility, and the consequential requirement of
          recompiling codes.  Support for the feature was weak.

          HPC subgroup recommendation is to defer this to some future
          revision of the standard, but not include it in F202X.