Boost logo

Boost Users :

Subject: Re: [Boost-users] mpi all reduce with std::vector and back_inserter
From: Júlio Hoffimann (julio.hoffimann_at_[hidden])
Date: 2011-05-05 19:03:04


Your first post referred to other implementation. Anyway, if you have two
vectors of strings:

CPU 0: vec = [ a b c d e ]
CPU 1: vec = [ f g h i j ]

then all_gather will produce a vector of strings:

result = [ a b c d e f g h i j ]

If you have a vector<vector<string> >, then all_gather will produce
vector<vector<string> > it's not possible to change this behavior.

You begin with a bidimensional vector, then you end with a bidimensional
vector. I don't know if it's possible to concatenate the matrix lines with
other Boost library, but the collective operations in Boost.MPI are like in
the MPI 1.1 standard, nothing else.

Regards,
Júlio.

2011/5/5 Kraus Philipp <philipp.kraus_at_[hidden]>

>
> Am 06.05.2011 um 00:24 schrieb Júlio Hoffimann:
>
> Hi Kraus,
>
> If i understood well, it's not the case for `all_reduce`, you need
> `all_gather` instead:
>
> http://www.boost.org/doc/libs/1_46_1/doc/html/boost/mpi/all_gather.html
>
> <http://www.boost.org/doc/libs/1_46_1/doc/html/boost/mpi/all_gather.html>The
> vectors has the same size?
>
>
> No, they have different sizes. I use all_gather, but I must iterate over
> each element in the result vector and put the elements into a new vector. I
> do this at the time:
>
> std::vector< std::vector<std::string> > result;
> mpi::all_gather( mpi object, vec, result);
>
> std::vector<std::string> myvec;
> foreach( result as res )
> myvec.push_back( foreach in res );
>
> I need the loop after all_gather and my question is: Can I do the loop call
> into the mpi command, so that I don't have a loop after the gather?
>
> Thx
>
> Phil
>
>
> _______________________________________________
> Boost-users mailing list
> Boost-users_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-users
>



Boost-users list run by williamkempf at hotmail.com, kalb at libertysoft.com, bjorn.karlsson at readsoft.com, gregod at cs.rpi.edu, wekempf at cox.net