Re: [Boost-users] boost.mpi possible bug with mpi::scatter

Subject: Re: [Boost-users] boost.mpi possible bug with mpi::scatter
From: Matthias Troyer (troyer_at_[hidden])
Date: 2010-06-08 13:50:10

This is corrrect behavior since scatter needs exactly ax many elements as there are processes you scatter to.

Matthias

Sent from my iPad

On Jun 8, 2010, at 1:41 AM, raj <rajnesh.jindel_at_[hidden]> wrote:

> Ok I'm new to mpi so maybe this is expected behaviour. But when using scatter, if number the of threads running is greater than number of elements in the array\vector sent to scatter, then there are the following issues
> 	1) if the array contains a user-defined data then an out-of-bounds error will occur
> 	2) otherwise (if is_mpi_datatype is true) the recieved data for threads that are "out-of-bound", will contain garbage and as such validity of the data cannot be tested.
> 
> I assume this not correct behaviour?
> 
> Raj
> 
> _______________________________________________
> Boost-users mailing list
> Boost-users_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-users
>