dolfin team mailing list archive
-
dolfin team
-
Mailing list archive
-
Message #20293
Re: [noreply@xxxxxxxxxxxxx: [Branch ~dolfin-core/dolfin/main] Rev 5334: Turn on renumbering for multi-core bench. Seg-faults for 2 processes.]
-
To:
DOLFIN Mailing List <dolfin@xxxxxxxxxxxxxxxxxxx>
-
From:
Anders Logg <logg@xxxxxxxxx>
-
Date:
Mon, 29 Nov 2010 15:26:22 +0100
-
In-reply-to:
<20101129124550.GL2030@eowyn>
-
User-agent:
Mutt/1.5.21 (2010-09-15)
Looks like it's working now. We get a factor 1.9 speedup with 2
threads. It's somewhat worse if we consider that the multithreaded
assembly with 1 thread is itself a factor 1.08 faster than regular
assembly.
--
Anders
On Mon, Nov 29, 2010 at 01:45:52PM +0100, Anders Logg wrote:
> I know, that's why I didn't turn it on. :-)
>
> Currently debugging.
>
> Date: Mon, 29 Nov 2010 12:36:40 -0000
> From: noreply@xxxxxxxxxxxxx
> To: Anders Logg <logg@xxxxxxxxx>
> Subject: [Branch ~dolfin-core/dolfin/main] Rev 5334: Turn on renumbering
> for multi-core bench. Seg-faults for 2 processes.
>
> ------------------------------------------------------------
> revno: 5334
> committer: Garth N. Wells <gnw20@xxxxxxxxx>
> branch nick: dolfin-all
> timestamp: Mon 2010-11-29 12:32:27 +0000
> message:
> Turn on renumbering for multi-core bench. Seg-faults for 2 processes.
> modified:
> bench/fem/multicore/cpp/main.cpp
>
>
> === modified file 'bench/fem/multicore/cpp/main.cpp'
> --- bench/fem/multicore/cpp/main.cpp 2010-11-27 23:44:11 +0000
> +++ bench/fem/multicore/cpp/main.cpp 2010-11-29 12:32:27 +0000
> @@ -12,6 +12,7 @@
>
> #include <dolfin.h>
> #include <dolfin/fem/AssemblerTools.h>
> +#include <dolfin/mesh/MeshRenumbering.h>
> #include "Poisson.h"
> #include "NavierStokes.h"
>
> @@ -106,6 +107,7 @@
> // Create mesh
> UnitCube mesh(SIZE, SIZE, SIZE);
> mesh.color("vertex");
> + MeshRenumbering::renumber_by_color(mesh);
> //mesh.init(1);
>
> // Test cases
>
References