(j3.2006) (SC22WG5.5906) [ukfortran] 3 levels of parallelism?
Bill Long
longb
Thu Jul 6 14:31:04 EDT 2017
> On Jul 6, 2017, at 5:22 AM, N.M. Maclaren <nmm1 at cam.ac.uk> wrote:
>
> OpenAcc has effectively been subsumed into OpenMP.
The latest OpenMP, particularly with the DEVICE clause, can basically do what OpenACC can do. Whether existing OpenACC codes will get converted, or just stagnate, remains to be seen. At least one vendor is still actively promoting OpenACC. Of course, had vendors been able to agree on GPU-focused changes to OpenMP several years ago, OpenACC would have likely never existed in the first place.
> Whether the OpenMP
> 'SIMD' features (i.e. OpenAcc) are practical is unclear, and the
> implementation status less so. My very limited feedback from people who
> have tried to use it was highly negative. As far as I know, CUDA looks
> likely to dominate for the forseeable future.
>
> Also, while languages are traditionally descriptive, not prescriptive,
> explicit threading (as in C++ etc.) and coarrays are both prescriptive
> models. In particular, they can't take advantage of SIMD, though Fortran as
> a whole can.
SIMD is really a vectorization directive. Vectorization is tightly linked to hardware. It is the ability to issue one hardware instruction and have it act on multiple data items. (Hence the acronym SIMD). The particular SIMD concept in OpenMP is focused in on one vendor?s hardware design of vector registers and associated instructions. Fortunately, other vendors can basically ignore it and just tell the optimizer to vectorize the loop. The more general concept of vectorization predates the OpenMP SIMD concept by decades. As Nick points out, Fortran is pretty well equipped to enable general vectorization.
Cheers,
Bill
>
> Regards,
> Nick Maclaren.
>
> _______________________________________________
> J3 mailing list
> J3 at mailman.j3-fortran.org
> http://mailman.j3-fortran.org/mailman/listinfo/j3
Bill Long longb at cray.com
Principal Engineer, Fortran Technical Support & voice: 651-605-9024
Bioinformatics Software Development fax: 651-605-9143
Cray Inc./ 2131 Lindau Lane/ Suite 1000/ Bloomington, MN 55425
More information about the J3
mailing list