Boost logo

Boost-MPI :

Subject: Re: [Boost-mpi] Fwd: mpi: MPI_MAX_LOC
From: MM (finjulhich_at_[hidden])
Date: 2013-10-14 08:36:29


Ok,

One of the possibilities is this alogrithm:

1. Construct list of all n-tuples in the root process as a vector of
pair<int, ntuple>.( this feel a bit redudant as std::vector's indices is
naturally what I want)
2. use reduce with the type T = pair< int, ntuple >, and have my function
object evaluate f at 2 ntuples and return the pair that gives the better
f(.) evaluation of the two.

Where do the f evaluations take place in this case? All at the root
process,or distributed?

MM

On 14 October 2013 10:39, Matthias Troyer <troyer_at_[hidden]> wrote:

> Look at the documentation of MPI_MAX_LOC. The datatype of to be reduced
> needs to be a pair, consisting of value and index. You can do the same with
> your custom function.
>
>
> On 14 Oct 2013, at 11:02, MM <finjulhich_at_[hidden]> wrote:
>
> > That will give me the optimum value of f(x1, ..., xn) but not _which_
> (x1,....xn) gives that optimum evaluation of f.
> >
> > MM
> >
> >
> > On 14 October 2013 09:57, Matthias Troyer <troyer_at_[hidden]> wrote:
> >
> > On 14 Oct 2013, at 10:53, MM <finjulhich_at_[hidden]> wrote:
> >
> >> Hello
> >> In a problem of getting the maximum of a multi-variable function, I
> would use mpi::reduce and provide it with a function object that compares
> the evaluations of this function.
> >>
> >> Each process evaluates a part of the space, ie a number of n-tuples of
> the n-dimension space.
> >>
> >> How do I get the n-tuple that yields the max of f(), given that
> MPI_MAXLOC is not available in boost mpi?
> >>
> >> Thanks
> >>
> >> MM
> >
> >
> > Hi MM,
> >
> > you can just provide your own function object to the boost::mpi::reduce
> function
> >
> > Matthias
> >
> >
> > _______________________________________________
> > Boost-mpi mailing list
> > Boost-mpi_at_[hidden]
> > http://lists.boost.org/mailman/listinfo.cgi/boost-mpi
> >
> >
> > _______________________________________________
> > Boost-mpi mailing list
> > Boost-mpi_at_[hidden]
> > http://lists.boost.org/mailman/listinfo.cgi/boost-mpi
>
> _______________________________________________
> Boost-mpi mailing list
> Boost-mpi_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-mpi
>



Boost-Commit list run by troyer at boostpro.com