Julian Andrej

The new mesh reader is more strict to ensure correctness.

All elements need to belong to an element block and all boundaries need to be assigned to a side set. If you define 2D meshes you have to make sure...

No requirement on having a nodeset (not supported in mfem AFAIK). You have to assign all faces that are "in contact with void" to a sideset.

See #4487

> 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? Also see https://www.smish.dev/programming/dynamic_kernel_refactor.

> @v-dobrev, have we looked into supporting dynamic shared memory with `mfem::forall` kernels? Maybe we could add an optional parameter to the `mfem::forall` functions (defaulting to zero) with the size...

This might be compiler specific but the right way to deal with this is probably to guard certain data format support on double precision. I'm not confident that truncation will...

The error is right here ``` CMake Error at config/cmake/modules/FindPETSc.cmake:147 (message): The pair PETSC_DIR=/work2/00434/eijkhout/mfem/mfem-4.7/../petsc PETSC_ARCH=arch-linux2-c-debug do not specify a valid PETSc installation Call Stack (most recent call first): CMakeLists.txt:271 (find_package)...

Sharing the failing code through compiler explorer https://fwd.gymni.ch/PE2qjx. This still doesn't give the correct answer.

If both function arguments are const you can workaround the issue with some evil casts https://fwd.gymni.ch/HTJOzZ