Dear all,
Matrix multiplication requires resulting matrix to be initialized. Even if
rebuild_sparsity_C = true .
template
template
void
SparseMatrix::mmult (SparseMatrix ,
const SparseMatrix ,
const Vector,
Dear all,
In following line:
line 4545 of /include/deal.II/matrix_free/fe_evaluation.h:
VectorizedArray half = make_vectorized_array (0.5);
0.5 is interpreted as double and it leads to using
make_vectorized_array. That is OK if Number=double but for any
other type (eg. float) it results is
W dniu niedziela, 8 października 2017 12:55:02 UTC+2 użytkownik Martin
Kronbichler napisał:
>
> Dear Michal,
>
> Could you please be a bit more specific? What was the error that you
> obtained? What vector did you pass into
> FEEvaluation::read_dof_values/distribute_local_to_global? I guess
I've tried to create pull request, but I've got error message:
mwichro@Preludio:~/lib/dealii$ git push origin fix_feevaluation_float_inst
Username for 'https://github.com': mwichro
Password for 'https://mwic...@github.com':
remote: Permission to dealii/dealii.git denied to mwichro.
fatal:
on one block? For example I would like
to compute approximation of A^-1 using multigrid.
Michał Wichrowski
--
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see
https://groups.google.com/d/forum/dealii?hl=en
---
You received this message because you
W dniu wtorek, 26 września 2017 18:42:33 UTC+2 użytkownik Wolfgang Bangerth
napisał:
>
> On 09/26/2017 09:42 AM, Michał Wichrowski wrote:
> >
> > 3) Is there a way to use multigrid on one block? For example I would
> > like to compute approximation of A^-1 using mu
>
>
> I've started working on new version MGTransferBlockMatrixFree and I now
> see the problem. MGTransferBlockMatrixFree is easy to rewrite using vector
> of dof handres etc, but PreconditionMG requires MGTransfer that have
> copy_to_mg and copy_from_mg that work on single dof handre. I am
W dniu wtorek, 26 września 2017 20:37:59 UTC+2 użytkownik Daniel Arndt
napisał:
>
> Michael,
>
> Adding to Wolfgang's response:
>
> Am Dienstag, 26. September 2017 17:42:41 UTC+2 schrieb Michał Wichrowski:
>>
>> Dear all,
>> I'm dealing with Stokes problem th
operator (B from Stokes). However, vmult checks
requires src and dst to have same size and throws an error. I'm not sure if
pre and post processing of constraints will work on non-square matrix.
Michał Wichrowski
The Divergence operator is defined as follows:
template
class DivergenceOperator
W dniu niedziela, 1 października 2017 21:20:06 UTC+2 użytkownik Daniel
Arndt napisał:
>
> Michael,
>
> I implemented something similar to what you proposed. You might want to
> have a look at https://github.com/dealii/dealii/pull/5175
>
> Best,
> Daniel
>
> Thanks, for now I'm using mine
. The program worked and even
covered in reasonable number of iterations. I think FEEvaluation is right
place to put this assertion.
Michał Wichrowski
--
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see
https://groups.google.com/d/forum/dealii?hl=en
W dniu sobota, 28 października 2017 18:52:57 UTC+2 użytkownik Martin
Kronbichler napisał:
>
> Dear Michal,
>
> You mean because once you apply an inhomogeneous Dirichlet condition on
> the velocity you also get a contribution because the divergence is not
> zero?
Exactly.
> You could
That's very bad news for me. This workaround will work for Laplace problem
but not for my case. I'm dealing with Stokes problem and my method requires
that second block of right hand side is zero, otherwise MinRes crashes and
GMRES have to be used (that is far less effective).
Can You give me
I think that the problem can be avoided by splitting operator A into two
parts: dirichlet nodes and other. The structure will be:
A1 0
0I
and the operator A1 applies matrix-vector product using fixed values on
Dirichlet nodes.
If I got it right, the problem is with this part of code:
Line:
Dear Martin,
I finally managed to make it work by following Your idea. Thanks!
Michał
--
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see
https://groups.google.com/d/forum/dealii?hl=en
---
You received this message because you are subscribed to the
Dear all,
I'm trying to use iterative inverse with distributed deal.II vector. From
following lines
const auto S= linear_operator (mg_sc_matrices[level]);
const auto Shat =
inverse_operator(S,solver,SCPreconditionMGs[level]);
I get compilation errors:
;).
> The example should be as short as possible and doesn't need to be
> functional / or do anything at all - we essentially only need to be able
> to figure out the types you try to wrap into the linear_operator object.
>
> Best,
> Matthias
>
>
> On Tue, Feb 13, 201
Dear all,
I had some trouble while compiling deal.II on my cluster:
[ 70%] Building CXX object
source/fe/CMakeFiles/obj_fe_release.dir/fe_tools.cc.o
/home/mwichro/lib/dealii-9.0.0/source/base/mpi.cc: In function 'int
dealii::Utilities::MPI::create_group(ompi_communicator_t* const&,
W dniu czwartek, 18 października 2018 17:59:40 UTC+2 użytkownik Bruno
Turcksin napisał:
>
> Michal,
>
> Have you tried to use ARPACK to get the smallest eigenvalues? That's what
> we are doing in one of our code and it works pretty well. We have a patch
> here
>
Dear all,
I need to compute the minimum and maximum eigenvalues of the matrix-free
operator (vmult is only available ). For the largest eigenvalue, I'm using
the power method, but I have some problems with minimal eigenvalue. The
condition number of the operator is close to 1, so I tried using
Dear Bruno
>
> > 2) The largest eigenvalue is computed instead of the largest one, and, I
> guess, the smallest eigenvalue is computed instead of the largest one.
> If you use ARPACK through deal without patching it, then the only
> method available is shift-inverse. The thing is that when
PS.
I noticed that in some cases the "smallest" eigenvalue does not
match maximum eigenvalue obtained from power method.
--
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see
https://groups.google.com/d/forum/dealii?hl=en
---
You received this
Dear Bruno,
I've used ARPACK and the computation time has dropped signifficantly. But I
have some strange results, I have following code:
IterationNumberControl arpack_control(30, 1e-8);
typename ArpackSolver::AdditionalData arpack_data_smallest(30,
Dear all,
I have errors when compiling my code with the devellpment version of
deal.II . The same problems occurs also with 9.0.0, the code works well
with 9.0.0-pre. I have no idea how to resolve these problems.
Michał
/home/mwichro/lib/deal.II/include/deal.II/multigrid/multigrid.h:137:7:
>
> Michal,
> the error is in your code, but since I don't have it, I can't tell
> where. I suspect that you are trying to copy an object of type
> Multigrid, in StokeMatrixFree/StokesMatrixFree.cc:34 of your project.
> This copy may happen implicitly if you call a function that takes a
>
W dniu wtorek, 25 września 2018 15:30:06 UTC+2 użytkownik Wolfgang Bangerth
napisał:
>
> On 09/25/2018 05:56 AM, Michał Wichrowski wrote:
> >
> > Yep, I am trying to copy Multigrid object on purpose, I found the line
> > corresponding to the problem:
> >
>
> On 09/25/2018 07:38 AM, Michał Wichrowski wrote:
> >
> >
> > Yes, I need triangulation.n_global_levels() copies of multigrid. I
> modify each
> > of them in next line:
> >
> >std::vector >
> > multigrids_sc(triangulation.n_
Dear all,
I am able to reproduce the problem with modified step-37, I expect the
similar problem in case of multigrid with sparse matrices. The problem
occurs only if mesh is refined adaptively.
I added refine_mesh() function and modified solve():
//
//work on level 3, that is not the
Done: https://github.com/dealii/dealii/issues/7581
W dniu wtorek, 8 stycznia 2019 23:42:06 UTC+1 użytkownik Daniel Arndt
napisał:
>
> Michal,
>
> not setting `limit_level_difference_at_vertices` when copying a
> parallel::distributed::Triangulation can be considered an oversight/bug.
> Do you
Dear all,
I noticed that if I do the following:
parallel::distributed::Triangulation
triangulation(MPI_COMM_WORLD,
parallel::distributed::Triangulation::limit_level_difference_at_vertices,
parallel::distributed::Triangulation::construct_multigrid_hierarchy);
Triangulation
W dniu poniedziałek, 17 września 2018 16:10:32 UTC+2 użytkownik Jean-Paul
Pelteret napisał:
>
> Dear Michał,
>
> I think that calling *git checkout* is insufficient. You probably want to
> call* git pull *(or* git fetch && git rebase origin/master*) to fetch and
> merge the most current state
Dear all,
I had a problem while compiling deal.II source code from the git repository:
[ 72%] Building CXX object
source/algorithms/CMakeFiles/obj_algorithms_debug.dir/operator.cc.o
In file included from
/home/mwichro/lib/dealii/include/deal.II/algorithms/any_data.h:21:0,
from
I've just double checked, I'm using the latest version:
mwichro@major:~/lib/dealii$ git checkout
Your branch is up-to-date with 'origin/master'.
mwichro@major:~/lib/dealii$ make
[ 0%] Built target expand_instantiations_exe
[ 0%] Built target doxygen_headers
[ 3%] Built target
Dear all,
It nothing big, but I thing adding method fill_random(const double& a=-1.,
const double =1.) to vector classes that will fill a vector with random
values (from range [a,b]) would be useful. I'm using it in most of my
programs, it's great for testing solvers, especially multigrid
>
>
> So you want to a separate function that calls the current interface with
> default constructed MGConstraindedDoFs objects?
> We are happy to have a look and discuss if you create a pull request for
> this.
>
> Thanks, for reporting the problem with the default constructor. I will
> have
Dear all,
I found out that calling
MGTransferBlockMatrixFree< dim, Number > ::clear
sets the number of blocks to 0/ It is followed directly by build(dofs);
will thow an exeption:
Additional information:
Dimension 0 not equal to 2.
There is no way to set number of block properly other than
Dear all,
I'm trying to use matrix-free with discontinuous Galerkin elements for
variable coefficient problem. The face integral require access to value of
coefficient. For the cell I've done the following thing:
//Vectors of values, stored in object:
AlignedVector< VectorizedArray > mu;
> Dear Martin
>
> We do not offer convenience access for all data structures on faces yet,
> so feel free to suggest what you need.
>
1) functions similar to get_cell_iterator for getting iterator to
interior/exterior cell + face index:
std::pair::cell_iterator, unstigned int>
Dear all,
I need to construct a sparse matrix from MatrixFree operator. I want to do
it in a similar way to present in step-37 for computing diagonal. The only
difference will be that the computed values from FEEValuation and
FEFaceEvaluation have to be distributed to the specified row of a
W dniu sobota, 12 października 2019 14:29:23 UTC+2 użytkownik Daniel Arndt
napisał:
>
> MIchaĺ,
>
> The documentation in the developer version or the current version (
> https://www.dealii.org/developer/doxygen/deal.II/classPreconditionChebyshev.html)
>
> says:
> "[...] Note that if the
Dear all,
The documentation of Chebyshev precondition says that:
"Note that if the degree variable is set to *zero*, the Chebyshev iteration
corresponds to a Jacobi preconditioner"
while actually setting degree to 0 triggers Assertion at line 2211 of
precondition.h. Moreover, the
I tried to fix that by myself, but I could not locate the source code of
MatrixFree::internal_reinit
W dniu poniedziałek, 9 marca 2020 14:20:16 UTC+1 użytkownik Michał
Wichrowski napisał:
>
> Deal all,
> I want to weakly impose boundary conditions on, let's say, Laplace problem
>
way to assemble the
matrix with vmult?
Best,
Michał Wichrowski
--
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see
https://groups.google.com/d/forum/dealii?hl=en
---
You received this message because you are subscribed to the Google Groups
"deal.
default
proceed as in standard case (no boundary integrals) - the boundary
integrals does not generate additional DoF coupling.
Best,
Michał Wichrowski
An error occurred in line <786> of file
in function
dealii::types::global_dof_index
dealii::Utilities::MPI::Partitioner::local_to_glo
assembly of sparse matrices.
>
> Thanks,
> Peter
>
> On Monday, 9 March 2020 15:26:10 UTC+1, Michał Wichrowski wrote:
>>
>> Dear all,
>> I've got matrix-free multigrid solver for Stokes problem. The main
>> bottleneck is solution of coarse problem, so I tri
Dear all,
I was trying to implement MatrixFree biharmonic solver. We already have
most of functionality needed to do it (DG, computing laplacian on both
faces and cells), but the one thing missing is extending FEEvaluation and
FEFaceEvaluation to support integration of 2nd order formulae:
a few weeks for deal.II
> (scheduled feature freeze May 8), so the goal would be to try to achieve
> this soon after the release.
>
> Best,
> Martin
> On 25.04.20 12:28, Michał Wichrowski wrote:
>
> Dear all,
>
> I was trying to implement MatrixFree biharmonic
with sparsity
pattern, it could be optimized to reduce the number of vmults.
Michał
W dniu wtorek, 10 marca 2020 18:28:15 UTC+1 użytkownik Michał Wichrowski
napisał:
>
>
>
> W dniu wtorek, 10 marca 2020 17:33:43 UTC+1 użytkownik Wolfgang Bangerth
> napisał:
>>
>> On 3/9/20
Dear all,
I've wrote a interface for Trilinos direct solver so that it may be used as
coarse solver for MG. It assembles the matrix by using N vmults (so this
part is very inefficient), so it will only pay off if solver is used
multiple times. I've got also block version. It is possible to
Dear all,
I've got multigrid solver matrix-free for FSI in ALE formulation. For
deforming grid I'm using MappingFEField, stored
in std::shared_ptr>. On the coarsest level I'm assembling the
sparse matrix in standard way, I then initialize direct solver (MUMPS from
Trilinos) and than I'm using
Dear all,
I'm considering a multigrid method for the Stokes problem. I need to have
an approximation of Schur complement at each level that is represented by a
matrix-free operator.
The problem is that in case of adaptive mesh refinement MGConstrainedDoFs
assumes zero boundary condition at
he suggested modifications. Everyone is welcome to contribute, and I think
> your information would be very valuable to other users as well.
>
> Best,
> Luca.
>
> > On 30 Jan 2021, at 16:34, Michał Wichrowski wrote:
> >
> > Dear all,
> > The Cmake g
Dear all,
The Cmake generator of Eclipse project does not work very well, it was not
developed for years. As a consequence, the Eclipse project generated by
CMake (as recommended in dealii Wiki) sometimes results with annoying
issues with indexer.
There is a useful plugin: cmake4eclipse that
>
> @Peter
Thanks!
@Bruno,
I've sent you an email last week.
--
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see
https://groups.google.com/d/forum/dealii?hl=en
---
You received this message because you are subscribed to the Google Groups
"deal.II
Jest overwrite error estimate in one domain by zeros:
unsigned int cell_n = 0;
for (const auto & cell : triangulation.active_cell_iterators ){
if (cell-> material_id() == 0 )
estimated_error_per_cell(i) = 0;
++i;
}
sobota, 3 kwietnia 2021 o 16:48:48 UTC+2 hkch...@gmail.com
Dear all,
Have anyone managed to compile and *run* CUDA code on non-nVidia device?
I've found HIPify: https://github.com/ROCm-Developer-Tools/HIPIFY
that translates CUDA code to HIP code that can be then compiled and ran on
both AMD and nVidia GPUs.
The thing is, I'm looking for a new laptop.
Michal,
>
> Support for AMD through HIP is very high on my todo list but the compiler
> is not that great right now, i.e. the compiler can (and will) generate
> buggy code.
>
Ok, so even if Deal.II will be ported I can assume so for other libraries.
Good to know that HIP does require
Dear all,
I am trying to set up deal.II on a computer with no system-wide zlib. I do
not have sudo privileges, thus I cannot just install, so I compiled it from
the source and installed in my home directory. Now I am trying to build
p4est using it. According to the message thrown by the scrip
@Wolfgang:
I commented out that lines, compiled deal.II and it looks OK, at least
`make test` passes and step-37 works in both release and debug mode.
@Marc
I had the same idea, but there was nothing in `configure --help` to put
custom zlib patch.
Thank you for the answers. For reference, I
The simple answer is multigrid. But depending on how the problem looks like
it will differ. If it is far from being incompressible standard MG will be
fine. Trilinos AMG should also work, just take look at its settings (there
is an option to specify that you are dealing with vector problem).
Dear all,
While I was setting up deal.II compilation on another machine I got stuck
for weeks: it was crashing when running with MPI. It turned out that
Deal.II managed to auto-detect PETSc (I did not turn it on, I am not using
it), which was causing issues with the initialization of any MPI
,
Michał Wichrowski
--
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see
https://groups.google.com/d/forum/dealii?hl=en
---
You received this message because you are subscribed to the Google Groups
"deal.II User Group" group.
To unsubs
62 matches
Mail list logo
