(j3.2006) (SC22WG5.4947) [ukfortran] AW: WG5 ballot on first draft TS 18508, Additional Parallel Features in Fortran (Update)
N.M. Maclaren
nmm1
Sat Mar 30 05:18:33 EDT 2013
On Mar 29 2013, Bill Long wrote:
>
>If we added this, it would be CO_PRODUCT since the local one is PRODUCT.
> The previous proposal had CO_PRODUCT. It was removed because the
>corresponding MPI_REDUCE for that operation almost never occurs in real
>codes. Is there any common usage of this operation? Also, is there
>hardware support in network hardware for a multiply reduction?
Yes. It's a common reduction in anything that uses statistical models,
especially generalised linear ones (by which I mean the original and
general form, as formalised by John Nelder et al.) The point is that
likelihoods multiply. And, of course, statistics includes information
theory in this context (or conversely, if you prefer).
There is no support in most hardware for most reductions or, indeed,
most of modern Fortran. We shouldn't be adopting a C/C++ mindset and
imagining that the primary purpose of a high-level language is to give
the programmer access to the hardware facilities. Fortran has lasted
over more variations in those than most people can believe!
Regards,
Nick Maclaren.
More information about the J3
mailing list