Hello, I am trying to parallelize a snippet of code which computes quadrature rules for CutFEM in every cell. I have understood that FEniCS automatically partitions the mesh when mpirun -n nproc is executed. This makes me wonder how one can build a global list of quadrature points and weights from …

Please make a minimal reproducible example. The code below has several unspecified functions, variables and imports, and it is not clear from the question if you use legacy dolfin or dolfinx

Hello @dokken , Thanks for your answer. You are right, sorry, my idea was to provide more of a pseudocode, as the above function uses stuff coming from other files. The minimal reproducible example which I can think of is below. I am using dolfinx and not legacy dolfin. import dolfinx from mpi4py i…

[image] federicobetti: was mainly interested in understanding whether it is possible to parallelize the computation of quadrature rules for CutFEM over multiple processors. I understand that the mesh cells are already divided among the processes, so is it actually feasible to do this? If you …

I am currently getting a weird error when trying to call dofmap.links() for a certain cell. But besides this, if you look into the code I am currently splitting cells among processors before starting the loop. I wonder whether this is useless, since dolfinx already splits the mesh automatically, an…

I’ll have a look at your code later today.

Thanks a lot @dokken , I wait to hear from you. I might also need your help on this other thread that I opened yesterday Ordering of cells via mesh.topology.original_cell_index

Your code doesn’t quite make sense to me, as it seems like you are trying to assign cells to processes after the partitioning has been done by dolfinx. See my reply in your other post for context. It would greatly help if you could sketch out what you expect as input to your algorithm. You could a…

Hi @dokken , Thanks for all of your help. I would like to divide the computation of quadrature rules for CutFEM (which is done with the C++ wrapper cppyy) among processors. Then, I could simply gather all the local information into a unique stacked list of quadrature rules, but I am not sure how th…

[image] federicobetti: You are right that it is useless to split cells among processors if this is automatically done by dolfinx. However, it seems to me that the local numbering in each rank is quite random with regards to the global one. Which command allows me to get this? The ordering of …

Hello @dokken , Thanks for your answer. I was able to find the bug in the dofmap.links() call above, hence now by running in parallel the code the code works and gives decent speedups. Basically what I miss now to continue is a map between local cells in each processor and the global cells in the …

Parallel computation of quadrature rules

mesh

dokken October 4, 2023, 12:00pm 12

I’ve already showed you this map in: Ordering of cells via mesh.topology.original_cell_index - #2 by dokken

Topic		Replies	Views
Why does custom quadrature give different answers in serial and parallel? dolfinx	3	143	January 11, 2025
Questions about local dofs and ordering General	8	1467	February 15, 2023
Regarding the vector reordering problem caused by parallel grid partitioning? General	3	104	January 9, 2025
Problems paralelizing a mesh generated using gmsh python api General mesh , dolfinx , mpi , gmsh	9	331	August 5, 2024
In parall computation(mpirun), how to communicate(scatter & gather) the data between local processes and some global process? dolfinx	5	1097	October 26, 2022

Parallel computation of quadrature rules

Related topics