Partitioned heat conduction (G+Smo participant)

[Crazy-Rich-Meghan]

Summary: Partitioned heat conduction G+Smo (IsoGeometric Analysis) participant
NOTE: G+Smo adapter is a submodule of the G+Smo library, to run the tutorial you need to configure G+Smo main library.

• Updated README.md file, which contains instructions on how to configure G+Smo library with preCICE adapter.
• Added gismo-executable folder to contain the symbolic link for the G+Smo executable.

This PR is marked as a draft because the documentation for the G+Smo adapter requires updates.

[uekerman]

Let's first iterate on #603 and then here, OK?

[MakisH]

@Crazy-Rich-Meghan do you plan to continue this development? It seems to currently be in draft state. We are happy to help get it in a good state!

[Crazy-Rich-Meghan]

Hi everyone,

Sorry for the delay! I have updated the partitioned heat conduction example from our side, this version supports the direct mesh access feature. To run the benchmark you need to do the following:
1. Clone G+Smo
2. cmake .. -DCMAKE_BUILD_TYPE=Release -DGISMO_OPTIONAL="gsPreCICE"
3. cmake --build build --target partitioned-heat-conduction -j 8
4. Install this example systemwide make install partitioned-heat-conduction
5. Go to the neumann-gismo folder run the Neumann participant with the run script.
6. Go to the dirichlet-gismo folder run the Dirichlet participant with the run script.

Let me know if you get stuck somewhere!

[MakisH]

Hi @Crazy-Rich-Meghan,

thanks for the updates! I am now trying to build on my system, based on the stable branch (Commit ef4c754785cf4949d3a89b916eed807b74b97a9b, October 2).

I do get a building error, though (g++ (Ubuntu 12.3.0-1ubuntu1~22.04.2) 12.3.0):

[ 97%] Building CXX object optional/gsPreCICE/examples/CMakeFiles/partitioned-heat-conduction.dir/partitioned-heat-conduction.cpp.o
/home/gc/repos/misc/gismo/optional/gsPreCICE/examples/partitioned-heat-conduction.cpp: In function ‘int main(int, char**)’:
/home/gc/repos/misc/gismo/optional/gsPreCICE/examples/partitioned-heat-conduction.cpp:110:17: error: base operand of ‘->’ has non-pointer type ‘gismo::gsBasis<double>::domainIter’ {aka ‘gismo::gsDomainIteratorWrapper<double>’}
  110 |     for (; domIt->good(); domIt->next(), k++ )
      |                 ^~
/home/gc/repos/misc/gismo/optional/gsPreCICE/examples/partitioned-heat-conduction.cpp:110:32: error: base operand of ‘->’ has non-pointer type ‘gismo::gsBasis<double>::domainIter’ {aka ‘gismo::gsDomainIteratorWrapper<double>’}
  110 |     for (; domIt->good(); domIt->next(), k++ )
      |                                ^~
/home/gc/repos/misc/gismo/optional/gsPreCICE/examples/partitioned-heat-conduction.cpp:120:15: error: base operand of ‘->’ has non-pointer type ‘gismo::gsBasis<double>::domainIter’ {aka ‘gismo::gsDomainIteratorWrapper<double>’}
  120 |     for (domIt->reset(); domIt->good(); domIt->next(), k++ )
      |               ^~
/home/gc/repos/misc/gismo/optional/gsPreCICE/examples/partitioned-heat-conduction.cpp:120:31: error: base operand of ‘->’ has non-pointer type ‘gismo::gsBasis<double>::domainIter’ {aka ‘gismo::gsDomainIteratorWrapper<double>’}
  120 |     for (domIt->reset(); domIt->good(); domIt->next(), k++ )
      |                               ^~
/home/gc/repos/misc/gismo/optional/gsPreCICE/examples/partitioned-heat-conduction.cpp:120:46: error: base operand of ‘->’ has non-pointer type ‘gismo::gsBasis<double>::domainIter’ {aka ‘gismo::gsDomainIteratorWrapper<double>’}
  120 |     for (domIt->reset(); domIt->good(); domIt->next(), k++ )
      |                                              ^~
/home/gc/repos/misc/gismo/optional/gsPreCICE/examples/partitioned-heat-conduction.cpp:124:29: error: base operand of ‘->’ has non-pointer type ‘gismo::gsBasis<double>::domainIter’ {aka ‘gismo::gsDomainIteratorWrapper<double>’}
  124 |         QuRule->mapTo( domIt->lowerCorner(), domIt->upperCorner(),
      |                             ^~
/home/gc/repos/misc/gismo/optional/gsPreCICE/examples/partitioned-heat-conduction.cpp:124:51: error: base operand of ‘->’ has non-pointer type ‘gismo::gsBasis<double>::domainIter’ {aka ‘gismo::gsDomainIteratorWrapper<double>’}
  124 |         QuRule->mapTo( domIt->lowerCorner(), domIt->upperCorner(),
      |                                                   ^~
gmake[3]: *** [optional/gsPreCICE/examples/CMakeFiles/partitioned-heat-conduction.dir/build.make:76: optional/gsPreCICE/examples/CMakeFiles/partitioned-heat-conduction.dir/partitioned-heat-conduction.cpp.o] Error 1
gmake[2]: *** [CMakeFiles/Makefile2:1940: optional/gsPreCICE/examples/CMakeFiles/partitioned-heat-conduction.dir/all] Error 2
gmake[1]: *** [CMakeFiles/Makefile2:1947: optional/gsPreCICE/examples/CMakeFiles/partitioned-heat-conduction.dir/rule] Error 2
gmake: *** [Makefile:907: partitioned-heat-conduction] Error 2

By the way, I edited your instructions above to add a number next to the -j flag. Without a number, it will use as many threads as parallelizable build targets, which might fill the memory completely and make a system crash.

[Crazy-Rich-Meghan]

Hi @Crazy-Rich-Meghan,

thanks for the updates! I am now trying to build on my system, based on the stable branch (Commit ef4c754785cf4949d3a89b916eed807b74b97a9b, October 2).

I do get a building error, though (g++ (Ubuntu 12.3.0-1ubuntu1~22.04.2) 12.3.0):

[ 97%] Building CXX object optional/gsPreCICE/examples/CMakeFiles/partitioned-heat-conduction.dir/partitioned-heat-conduction.cpp.o

/home/gc/repos/misc/gismo/optional/gsPreCICE/examples/partitioned-heat-conduction.cpp: In function ‘int main(int, char**)’:

/home/gc/repos/misc/gismo/optional/gsPreCICE/examples/partitioned-heat-conduction.cpp:110:17: error: base operand of ‘->’ has non-pointer type ‘gismo::gsBasis<double>::domainIter’ {aka ‘gismo::gsDomainIteratorWrapper<double>’}

  110 |     for (; domIt->good(); domIt->next(), k++ )

      |                 ^~

/home/gc/repos/misc/gismo/optional/gsPreCICE/examples/partitioned-heat-conduction.cpp:110:32: error: base operand of ‘->’ has non-pointer type ‘gismo::gsBasis<double>::domainIter’ {aka ‘gismo::gsDomainIteratorWrapper<double>’}

  110 |     for (; domIt->good(); domIt->next(), k++ )

      |                                ^~

/home/gc/repos/misc/gismo/optional/gsPreCICE/examples/partitioned-heat-conduction.cpp:120:15: error: base operand of ‘->’ has non-pointer type ‘gismo::gsBasis<double>::domainIter’ {aka ‘gismo::gsDomainIteratorWrapper<double>’}

  120 |     for (domIt->reset(); domIt->good(); domIt->next(), k++ )

      |               ^~

/home/gc/repos/misc/gismo/optional/gsPreCICE/examples/partitioned-heat-conduction.cpp:120:31: error: base operand of ‘->’ has non-pointer type ‘gismo::gsBasis<double>::domainIter’ {aka ‘gismo::gsDomainIteratorWrapper<double>’}

  120 |     for (domIt->reset(); domIt->good(); domIt->next(), k++ )

      |                               ^~

/home/gc/repos/misc/gismo/optional/gsPreCICE/examples/partitioned-heat-conduction.cpp:120:46: error: base operand of ‘->’ has non-pointer type ‘gismo::gsBasis<double>::domainIter’ {aka ‘gismo::gsDomainIteratorWrapper<double>’}

  120 |     for (domIt->reset(); domIt->good(); domIt->next(), k++ )

      |                                              ^~

/home/gc/repos/misc/gismo/optional/gsPreCICE/examples/partitioned-heat-conduction.cpp:124:29: error: base operand of ‘->’ has non-pointer type ‘gismo::gsBasis<double>::domainIter’ {aka ‘gismo::gsDomainIteratorWrapper<double>’}

  124 |         QuRule->mapTo( domIt->lowerCorner(), domIt->upperCorner(),

      |                             ^~

/home/gc/repos/misc/gismo/optional/gsPreCICE/examples/partitioned-heat-conduction.cpp:124:51: error: base operand of ‘->’ has non-pointer type ‘gismo::gsBasis<double>::domainIter’ {aka ‘gismo::gsDomainIteratorWrapper<double>’}

  124 |         QuRule->mapTo( domIt->lowerCorner(), domIt->upperCorner(),

      |                                                   ^~

gmake[3]: *** [optional/gsPreCICE/examples/CMakeFiles/partitioned-heat-conduction.dir/build.make:76: optional/gsPreCICE/examples/CMakeFiles/partitioned-heat-conduction.dir/partitioned-heat-conduction.cpp.o] Error 1

gmake[2]: *** [CMakeFiles/Makefile2:1940: optional/gsPreCICE/examples/CMakeFiles/partitioned-heat-conduction.dir/all] Error 2

gmake[1]: *** [CMakeFiles/Makefile2:1947: optional/gsPreCICE/examples/CMakeFiles/partitioned-heat-conduction.dir/rule] Error 2

gmake: *** [Makefile:907: partitioned-heat-conduction] Error 2

By the way, I edited your instructions above to add a number next to the -j flag. Without a number, it will use as many threads as parallelizable build targets, which might fill the memory completely and make a system crash.

Hi @makis, have you pulled the main Gismo? We couldn't replicate the issue (on both MacOS and Ubuntu).

[MakisH]

@Crazy-Rich-Meghan I have not forgotten about this PR, and I see there are still some easy points to close. I added a log file with my building issue.

[Crazy-Rich-Meghan]

Hi @MakisH,

The domIt->good() issue has been fixed in gsPreCICE (commit e557b51). Please update your gsPreCICE to the latest version:

rm -rf optional/gsPreCICE
cmake .. -DGISMO_OPTIONAL="gsPreCICE"

[MakisH]

Thanks, @Crazy-Rich-Meghan, this worked! The perpendicular-flap-vertex-gismo does not build for me anymore, but I will have another look.

If you have no objections, I will (later) push a few changes directly to this branch, to make the tutorial more consistent with https://precice.org/tutorials-perpendicular-flap-stress.html

[Crazy-Rich-Meghan]

Hi @MakisH,

Thanks for the review! I've addressed your comments:

• Updated clean.sh scripts to use the standard clean_gismo() function
• Simplified the README.md, linking to the adapter documentation
• Updated executable paths in run.sh scripts
• Removed the redundant gismo.gif (same result as FEniCS)

Please let me know if there's anything else that needs to be changed.

[MakisH]

@Crazy-Rich-Meghan I am trying to run this case again, but it is failing immediately with:

Patches: 1, degree: 2
Active options:
Options (12):
* DirichletStrategy    (int) = 11       Method for enforcement of Dirichlet BCs [11..14]
* DirichletValues      (int) = 101      Method for computation of Dirichlet DoF values [100..103]
* SameElement         (bool) = 1        Activates optimization if all quadrature points are located in the same element
* bdA                 (real) = 2        Estimated nonzeros per column of the matrix: bdA*deg + bdB
* bdB                  (int) = 1        Estimated nonzeros per column of the matrix: bdA*deg + bdB
* bdO                 (real) = 0.333    Overhead of sparse mem. allocation: (1+bdO)(bdA*deg + bdB) [0..1]
* flipSide            (bool) = 0        Flip side of interface where integration is performed.
* movingInterface     (bool) = 0        Used in interface assembly when interface is not stationary.
* overInt             (bool) = 0        Apply over-integration on boundary elements or not?
* quA                 (real) = 1        Number of quadrature points: quA*deg + quB; For patchRule: Regularity of the target space
* quB                  (int) = 1        Number of quadrature points: quA*deg + quB; For patchRule: Degree of the target space
* quRule               (int) = 1        Quadrature rule used (1) Gauss-Legendre; (2) Gauss-Lobatto; (3) Patch-Rule

Assert `index >= 0 && index < size()` 
/home/gc/repos/misc/gismo/external/gsEigen/src/Core/DenseCoeffsBase.h, line 163 (operator[])
terminate called after throwing an instance of 'std::logic_error'
  what():  GISMO_ASSERT
./run.sh: line 7: 90127 Aborted                 (core dumped) ./gismo-executable -c ../precice-config.xml --plot -s 0

I am using a clean build of the latest dev branch (but an earlier commit was also failing) - commit gismo/gismo@a43e434

I have configured it with the modules gsElasticity;gsKLShell;gsPreCICE;gsStructuralAnalysis.

Does the same state of G+Smo work for you?

[MakisH]

@Crazy-Rich-Meghan it would be great to be able to include this in the upcoming distribution release. I think it is very close to merging; I only want to make sure that I can run it, for which I need some help.

[MakisH]

I tried again in an Ubuntu 24.04 container and the partitioned-heat-conduction-direct case works (note to self: container ID dfc17ae89aca, name affectionate_sinoussi).

However, the partitioned-heat-conduction still seems to be calling the direct mesh access API. @Crazy-Rich-Meghan am I missing some option? Am I building the wrong solver?

[Crazy-Rich-Meghan]

Hi @MakisH, changed:

• Auto detect mesh access gismo/gsPreCICE#26 (merged): the example now auto-detects from the
  config whether to use direct mesh access. Same binary handles both
  tutorials.
• This PR: renamed direct-access -> api-access in the direct config,
  and documented the gsPreCICE submodule dependency in the README.

[MakisH]

Everything works as expected now. Thanks a lot for the (many) updates, @Crazy-Rich-Meghan!