Add reflecting boundary conditions

[pgrete]

What is says on the tin (and the docs).

Bonus: LW implosion test problem.

[pgrete]

Also, I just ran some quick test wrt symmetry preservation. RK3 with PPM is not that great (see also discussion around Athena++ and that method paper),but VL2+PLM is (density at t=2.4):

[BenWibking]

Did you disable FMA instructions with a compiler flag? I don't think the reconstruction method should matter. In Quokka, RK2+PPM gives perfect symmetry preservation, as long as you disable FMA instructions.

[pgrete]

Did you disable FMA instructions with a compiler flag? I don't think the reconstruction method should matter. In Quokka, RK2+PPM gives perfect symmetry preservation, as long as you disable FMA instructions.

I have not tried disabling FMA.

[BenWibking]

Minor comments below. The major concern I have is that the symmetry does not appear to be preserved when there are multiple MeshBlocks used for the LW problem.

[BenWibking]

Did you disable FMA instructions with a compiler flag? I don't think the reconstruction method should matter. In Quokka, RK2+PPM gives perfect symmetry preservation, as long as you disable FMA instructions.

I have not tried disabling FMA.

For CUDA, I disabled FMA in Quokka with these CMake flags:

if(DISABLE_FMAD)
  message(STATUS "Fused multiply-add (FMAD) is disabled for device code.")

  if(CMAKE_CUDA_COMPILER_ID STREQUAL "NVIDIA")
    add_compile_options($<$<COMPILE_LANGUAGE:CUDA>:--fmad=false>)
  endif()

  if(CMAKE_CUDA_COMPILER_ID STREQUAL "Clang")
    add_compile_options($<$<COMPILE_LANGUAGE:CUDA>:-ffp-contract=off>)
    add_compile_options($<$<COMPILE_LANGUAGE:C>:-ffp-contract=off>)
    add_compile_options($<$<COMPILE_LANGUAGE:CXX>:-ffp-contract=off>)
  endif()

  if(CMAKE_CUDA_COMPILER_ID STREQUAL "NVHPC")
    add_compile_options($<$<COMPILE_LANGUAGE:C>:-Mnofma>)
    add_compile_options($<$<COMPILE_LANGUAGE:CXX>:-Mnofma>)
    add_compile_options($<$<COMPILE_LANGUAGE:CUDA>:-gpu=nofma>)
  endif()
else()
  # fused MAD causes directional asymmetry -- enabled by default by nvcc and clang
  message(STATUS "Fused multiply-add (FMAD) is *enabled* for device code. Exact direction symmetry will NOT be preserved.")
endif(DISABLE_FMAD)

[pgrete]

Alright, I think this is good to go now.