Re: mpi 1.1.1 released

[Sukanta Basu]

On 1/3/2014 8:04 AM, Sukanta Basu wrote:
> On Fri, Jan 3, 2014 at 7:57 AM, c. <address@hidden> wrote:
> > On 3 Jan 2014, at 13:42, c. <address@hidden> wrote:
> >
> > > Sukanta,
> > >
> > > Could you please avoid top-posting in your messages?
> > > It makes it really hard to follow the discussion.
> > > In particular I must have missed some message as I don't
> > > understand what you refer to in your mail above.
> > >
> > > c.
> >
> > OK,
> >
> > I now found Michael's message (which for some reason
> > was downloaded by my client AFTER your reply) and I understand
> > what you are talking about.
> >
> > > On 3 Jan 2014, at 13:25, Sukanta Basu <address@hidden> wrote:
> > >
> > > > Hi Michael,
> > > >
> > > > Thanks for catching the typo in the sample file. I quickly created
> > > > this sample code for testing. The outer loop should have a different
> > > > index.
> > I'm going to reformat the code and post it in the bug traker in order
> > to keep track of the progress.
> >
> > > > I have to send multi-dimensional arrays in my work. There is no
> > > > alternative.
> > well, you could actually try to reshape your arrays before MPI_Send and
> > after MPI_Receive. This causes a very small overhead and, if it avoids
> > the emory leak it could help identify the origin of the problem.
> >
> > > > Best regards,
> > > > Sukanta
> > Thanks,
> > c.
> Hi Carlo,
>
> Thanks in advance for posting the code in bug-tracker. My code deals
> with many arrays. Reshaping them for every communication will
> definitely slow down the computation. Today, I am going to check if
> one-dimensional array solves the problem.
>
> Cheers,
> Sukanta

Hi Carlo and Michael,

I added reshape in the sample code and tested it quickly. The memory 
leak still persists.

Specifically, I used:
        u = utot(:,:,nzb*i+1:nzb*(i+1));
        v = reshape(u,numel(u),1);
        MPI_Send(v,i,i*mpitag(27),comm);
and
        u = ZeRo; v = MPI_Recv(0,me*mpitag(27),comm);
        [u(:,:,2:nzb+1)] = reshape(v,nx,ny,nzb);

Best regards,
Sukanta

--
Sukanta Basu
Associate Professor
North Carolina State University
http://www4.ncsu.edu/~sbasu5/