On Mon, Nov 16, 2026 at 11:43 PM Suvarchal K Cheedela
<[email protected]> wrote:
>
> Thank you both for the comments.  I reply to both here.
>
> On 11/16/26 22:00, Harald Anlauf wrote:
> > Am 16.11.26 um 9:18 PM schrieb Janne Blomqvist:
> >> On Mon, Nov 16, 2026 at 5:46 PM Suvarchal Kumar Cheedela
> >> <[email protected]> wrote:

> I agree that conforming programs must not pay for it, so I withdraw the
> proposal to make the copy the default.  I would still ask you to
> consider it as an option that is off by default, so that the default
> path stays exactly as it is.  My reasons:
>
> - Few programs seem to use Fortran asynchronous I/O for large data.
>    WAIT with ID= has returned too early since 2018 (PR128719), and an
>    asynchronous WRITE of a pointer array queued one item per element
>    (PR128722, 120 s instead of 4 s for a FESOM2 restart).  As far as I
>    can find, neither had been reported.

I think you're right here that few programs actually use async I/O,
and thus there are bugs lurking in the implementation. And thank you
for reporting those bugs along with patches to fix them.

But if anything, isn't that an argument for fixing the bugs in the
implementation? Not to add even more complexity in order to make some
classes of non-conforming behavior work differently?

> - Climate and weather models built asynchronous output outside the
>    language instead: XIOS (NEMO), CDI-PIO (ICON, MPI-ESM), PIO (CESM,
>    E3SM), FESOM2's own C++ writer threads.  Most of them use dedicated
>    I/O processes: every field is sent over the network to processes or
>    nodes that do nothing but write, and those cores are taken from the
>    computation.  An I/O thread in each process writes from where the
>    data already is, with no extra network traffic, and can use the cores
>    or hardware threads that a node often has left over.

This might very well be true, depending on the application, but
large-scale parallel application architecture seems rather far out of
the purview of the GFortran runtime library?




-- 
Janne Blomqvist

Reply via email to