Will Pazner

I think this was an alias synchronization issue; it should be fixed in 1c90b1fd4fc358d7eeb900e50ab0202af7c8a87b.

@v-dobrev @artv3, see https://github.com/mfem/mfem/pull/4408#discussion_r1821860778, this will crash on CUDA, it should be fixed before merging.

Partial assembly only current supports tensor-product elements (quadrilaterals and hexahedra). The square-disc mesh is a triangular mesh, so it won't work with PA. You can generate a quad mesh of...

Probably related to #4633. Potential solution in [this comment](https://github.com/mfem/mfem/issues/4633#issuecomment-2558361499). One workflow is to load the mesh with big int support, partition it, and then save the partitions, which can then...

The partitioned mesh can be saved and loaded as follows (thought as @v-dobrev mentioned, perhaps this is not the cause of your problem). ```cpp void SavePartitionedMesh(const string &prefix, Mesh &mesh,...

@v-dobrev, is there a solution for very large serial meshes (e.g. those for which the mesh data structures may have size greater than $2^{31}$)?

@v-dobrev, does this look OK to you?

@v-dobrev, @adam-sim-dev, the latest commits (7f263221dd567e42f46d1d8343eee8baaef8f3f8 and e0c1018c9cebaae5c344bfb9a08a2fbd2b1dd4d6) include a workaround for a bug in cuSPARSE (CUSPARSE-1897, [see here](https://docs.nvidia.com/cuda/archive/12.6.3/cuda-toolkit-release-notes/index.html#cusparse-release-12-6-update-2)). Does that look OK to you?

Done, thank you @helloworld922!

This might be a good case for [dynamic shared memory](https://developer.nvidia.com/blog/using-shared-memory-cuda-cc/#dynamic_shared_memory). What do you think @YohannDudouit?