On Tue, 19 Jan 2016, Roy Stogner wrote:
> In the short term the workaround ought to be to use
> WRITE_SERIAL_FILES; I'm not sure that's working but if it's not I'll
> try to fix it ASAP.
WRITE_SERIAL_FILES | WRITE_DATA works.
Parallel xda mesh support ought to be relatively easy to add, but I'm
On Thu, 14 Jan 2016, Roy Stogner wrote:
> You know what? It doesn't for me either, my previous reports to the
> contrary. But that use case *should* work, and does work on lots of
> other meshes I've tried. I'll keep looking into it.
And now I think I understand the problem. Our solution res
You may be interested in the "Checkpoint" mesh format:
https://github.com/libMesh/libmesh/blob/master/include/mesh/checkpoint_io.h
It is still XDR based... but basically just writes out the raw mesh as it
sits on each processor and then reads it back in during a restart. Because
of this you have
I didn’t know the numbering could change. Now it makes sense. Thanks.
Sincerely,
Miguel A. Salazar de Troya
Graduate Scholar, Lawrence Livermore National Laboratory
Ph.D. Candidate, University of Illinois at Urbana-Champaign
T1878
Rm: 112
Ext: 2-6411
L-code: 288
On 1/14/16, 11:13 AM, "Roy Sto
On Thu, 14 Jan 2016, Salazar De Troya, Miguel wrote:
Sorry the last statement is wrong. Leaving partition_agnostic=true with a
SerialMesh doesn’t return the same assertion error, but the vector written
and the vector read are not the same when I print them. The values seem to
be in different or
Sorry the last statement is wrong. Leaving partition_agnostic=true with a
SerialMesh doesn’t return the same assertion error, but the vector written
and the vector read are not the same when I print them. The values seem to
be in different order.
Miguel A. Salazar de Troya
Graduate Scholar, Lawr
Leaving partition_agnostic=true gives me the same assertion error on
SerialMesh as well.
Sincerely,
Miguel A. Salazar de Troya
Graduate Scholar, Lawrence Livermore National Laboratory
Ph.D. Candidate, University of Illinois at Urbana-Champaign
T1878
Rm: 112
Ext: 2-6411
L-code: 288
On 1/14/16,
In that case, what would be the compatible way to restart the mesh if I
still leave partition_agnostic=false? What would be the way to fix the
code? Leaving partition_agnostic=true doesn¹t avoid the assertion either
for me.
Sincerely,
Miguel A. Salazar de Troya
Graduate Scholar, Lawrence Livermor
On Wed, 13 Jan 2016, Roy Stogner wrote:
On Tue, 12 Jan 2016, Salazar De Troya, Miguel wrote:
I’m attaching a simple code that illustrates the problem. I open a
mesh (l_problem.xda) and perform a uniform refinement (if I don’t,
there are no problems), then I create an ExplicitSystem and ass
On Tue, 12 Jan 2016, Salazar De Troya, Miguel wrote:
I’m attaching a simple code that illustrates the problem. I open a
mesh (l_problem.xda) and perform a uniform refinement (if I don’t,
there are no problems), then I create an ExplicitSystem and assign
random values to its solution. I print th
Sorry I sent the email to only John Peterson by mistake. The code is here
http://paste.ofcode.org/XqU3mKRdgGnQQZRveVwKRr and the l_problem.xda file:
http://justpaste.it/qe3i
Is it possible to attach files on this mail list? I did that for my
original message by it looks like it didn¹t get through.
On Tue, 12 Jan 2016, John Peterson wrote:
>> I get this assertion error:
>>
>>
>> Assertion `cnt < io_buffer.size()' failed.
>>
>> cnt = 12755
>>
>> io_buffer.size() = 12755
Hmm... I'm seeing the same assertion error on a ParallelMesh restart
of a "slit mesh", with overlapping nodes, but I haven
alazar mailto:salazardet...@llnl.gov>>
Cc:
"libmesh-users@lists.sourceforge.net<mailto:libmesh-users@lists.sourceforge.net>"
mailto:libmesh-users@lists.sourceforge.net>>
Subject: Re: [Libmesh-users] ParallelMesh, SerialMesh and simulation restarts
On Tue, Jan 12, 2016 at
On Tue, Jan 12, 2016 at 11:17 AM, Salazar De Troya, Miguel <
salazardet...@llnl.gov> wrote:
> Hello
>
> I’m attaching a simple code that illustrates the problem. I open a mesh
> (l_problem.xda) and perform a uniform refinement (if I don’t, there are no
> problems), then I create an ExplicitSystem
Hello
I’m attaching a simple code that illustrates the problem. I open a mesh
(l_problem.xda) and perform a uniform refinement (if I don’t, there are no
problems), then I create an ExplicitSystem and assign random values to its
solution. I print the solution for verification purposes. I save bo
>
> I've run meshes with several million elements on my desktop with several
> non-linear variables per DOF without memory issues. The mesh data structure
> is relatively small compared to your equation systems (unless you are
> running really simple "single" physics problems). If it fits in memory
I've run meshes with several million elements on my desktop with several
non-linear variables per DOF without memory issues. The mesh data structure
is relatively small compared to your equation systems (unless you are
running really simple "single" physics problems). If it fits in memory,
you'll a
>
> At that point there's little reason to bother using a ParallelMesh at
> all; after an allgather it's become a less-efficient-than-SerialMesh
> serial mesh.
> You can use a point locator on the distributed mesh without
> serializing first; you just have to account for the fact that it will
>
>
> For now I just keep all the elements on all processes (basically
> SerialMesh).
>
> Why not just use SerialMesh?
>
I was initially trying to have only the required elements on each process,
since the mesh could get big (100,000+ elements). I would like to be able
to do that, so I haven't cha
Why not just use SerialMesh?
On Tue, Sep 22, 2015 at 3:33 PM Harshad Sahasrabudhe
wrote:
> >
> > I initially use the Poisson mesh point locator on quadrature points
> > of the second mesh to find the elements of Poisson mesh
> >
>
> Hmm.. assuming the point locator is up to date, that should giv
On Tue, 22 Sep 2015, Harshad Sahasrabudhe wrote:
> I forgot to mention that I call allgather() on the Poisson mesh and
> reinitialize the point locator before calling it.
At that point there's little reason to bother using a ParallelMesh at
all; after an allgather it's become a less-efficient-th
>
> I initially use the Poisson mesh point locator on quadrature points
> of the second mesh to find the elements of Poisson mesh
>
Hmm.. assuming the point locator is up to date, that should give you
> semilocal elements or NULL, but shouldn't give you anything with an
> invalid_id. I'm stumped.
On Tue, 22 Sep 2015, Harshad Sahasrabudhe wrote:
> I initially use the Poisson mesh point locator on quadrature points
> of the second mesh to find the elements of Poisson mesh
Hmm.. assuming the point locator is up to date, that should give you
semilocal elements or NULL, but shouldn't give you
I don't understand this. An element (and therefore all quadrature
points on that element) is either entirely owned by a processor or is
entirely not. Do you have a coarser grid that you're integrating on?
> Each element is entirely on one process, but the quantity that I'm
>
> I don't understand this. An element (and therefore all quadrature
> points on that element) is either entirely owned by a processor or is
> entirely not. Do you have a coarser grid that you're integrating on?
>
> Each element is entirely on one process, but the quantity that
On Tue, 22 Sep 2015, Harshad Sahasrabudhe wrote:
I don't understand this. An element (and therefore all quadrature
points on that element) is either entirely owned by a processor or is
entirely not. Do you have a coarser grid that you're integrating on?
Each element is enti
>
> Not sure what this means: "quantity is not available for all quadrature
> points of some elements on one process"
You shouldn't be computing anything on ghosted elements... they will be the
> responsibility of the processor that owns them.
So the reason why that quantity isn't available on t
>
> I don't understand this. An element (and therefore all quadrature
> points on that element) is either entirely owned by a processor or is
> entirely not. Do you have a coarser grid that you're integrating on?
Each element is entirely on one process, but the quantity that I'm
integrating isn
On Tue, Sep 22, 2015 at 2:42 PM, Harshad Sahasrabudhe
wrote:
> Hi,
>
> I'm using ParallelMesh to discretize the Poisson equation. I want to
> integrate a certain quantity over each element. That quantity is not
> available for all quadrature points of some elements on one process (such
> elements
On Tue, 22 Sep 2015, Harshad Sahasrabudhe wrote:
> I'm using ParallelMesh to discretize the Poisson equation. I want to
> integrate a certain quantity over each element. That quantity is not
> available for all quadrature points of some elements on one process (such
> elements are ghosted), so I
Hi,
I'm using ParallelMesh to discretize the Poisson equation. I want to
integrate a certain quantity over each element. That quantity is not
available for all quadrature points of some elements on one process (such
elements are ghosted), so I sum the parts which are available in respective
proces
On Tue, 24 Mar 2009, Andrea Hawkins wrote:
> SerialMesh::point has a check libmesh_assert (i < this->n_nodes()).
>
> I believe n_nodes() returns the number of nodes local to the processor.
No, n_nodes() is the number of global nodes.
I'm actually not sure why we've removed that assert from Par
On Tue, 24 Mar 2009, Andrea Hawkins wrote:
> I notice in my libmesh_config.h file that ParallelMesh is referred to as
> "experimental." How experimental is it?
There are cases when using adaptive refinement and coarsening where an
assert fails in debug mode and the mesh topology "tears" in opt m
Hello-
I notice in my libmesh_config.h file that ParallelMesh is referred to as
"experimental." How experimental is it?
I'm trying to use mesh.point(), but when on computing on multiple processors
I don't see an easy way to figure out the appropriate unsigned int that
should be passed in when the
34 matches
Mail list logo
