[Mpi3-rma] MPI_Fetch_and_op and predefined datatypes
htor at illinois.edu
Sat Oct 27 08:56:59 CDT 2012
On Sat, Oct 27, 2012 at 09:05:06AM -0500, Jeff Hammond wrote:
> If someone wants MPI_NO_OP or MPI_REPLACE on MPI_WCHAR, can't they
> just do it with MPI_Get_accumulate on the appropriate width using
> MPI_BYTE and MPI_SIZEOF?
Not really, because:
1) MPI_SIZEOF is only defined for Fortran ;-)
2) atomicity and ordering are only guaranteed for each basic datatype
(see 11.7.1 and 11.7.2). Thus, a read of many bytes can read those bytes
in different order, while a WCHAR read has to be atomic.
> I see no performance-critical justification for MPI_WCHAR atomic reads
> and writes.
To cite myself, I "dreamt up" a use-case :-).
> On Fri, Oct 26, 2012 at 4:07 PM, Torsten Hoefler <htor at illinois.edu> wrote:
> > On Fri, Oct 26, 2012 at 04:22:02PM -0500, Jeff Hammond wrote:
> >> I don't want to wait for 3.1. Are we seriously going to require
> >> MPI_WCHAR support in F&Op to be MPI-3 compliant? That makes no sense.
> > I think the intent of "Any of the predefined operations for MPI_REDUCE,
> > as well as MPI_NO_OP or MPI_REPLACE, can be specified as op;
> > user-defined functions cannot be used." is clear. One could even dream
> > up a use-case for MPI_WCHAR if one wants to emulate an atomic read
> > (using MPI_NO_OP).
> > I will not vote against an errata if you want to propose text and
> > initiate the process.
> > Best,
> > Torsten
> > --
> > ### qreharg rug ebs fv crryF ------------- http://www.unixer.de/ -----
> > Torsten Hoefler | Assistant Professor
> > Dept. of Computer Science | ETH Zürich
> > Universitätsstrasse 6 | Zurich-8092, Switzerland
> > CAB E 64.1 | Phone: +41 76 309 79 29
> > _______________________________________________
> > mpi3-rma mailing list
> > mpi3-rma at lists.mpi-forum.org
> > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma
> Jeff Hammond
> Argonne Leadership Computing Facility
> University of Chicago Computation Institute
> jhammond at alcf.anl.gov / (630) 252-5381
> mpi3-rma mailing list
> mpi3-rma at lists.mpi-forum.org
### qreharg rug ebs fv crryF ------------- http://www.unixer.de/ -----
Torsten Hoefler | Assistant Professor
Dept. of Computer Science | ETH Zürich
Universitätsstrasse 6 | Zurich-8092, Switzerland
CAB E 64.1 | Phone: +41 76 309 79 29
More information about the mpiwg-rma