Add Preliminary MHD Integrator

[bcaddy]

All non-MHD tests pass except those noted in issues #218 and #219 which also don't pass in dev.

This commit primarily adds support for MHD to the Van Leer 3D integrator along with myriad small fixes and modifications. All new features have tests as well.

The MHD implementation still has bugs but needs to be merged in before we reformat the code so that the merging doesn't provoke too many merge conflicts.

MHD

• Add MHD support to Van Leer 3D integrator
• All MHD code is in the mhd namespace with sub-namespaces when relevant (e.g. mhd::utils)
• Add kernel to compute the CT electric fields and tests for it
• Add kernel to update the magnetic field and tests for it
• Add MHD support to the PCM reconstruction
• Add new Linear_Wave initial condition type
• Add ctElectricFields global device array
• Set all MHD input parameters to default to zero
• Remove incorrect MHD specializations in the boundary conditions
• Add check to make sure MHD has at least 3 ghost cells
• Add functions to compute the maximum magnetic divergence, report it, and exit if it's too large
• Clean up make.type.mhd and enable cuda error checking
• Add example parameter files for
  • Alfven wave
  • Fast magnetosonic wave
  • Slow magnetosonic wave
  • Ryu & Jones 1995 shock tubes 1a and 2a
• Add the following input parameters for generalized linear waves
  • EigenVec_rho
  • EigenVec_MomentumX
  • EigenVec_MomentumY
  • EigenVec_MomentumZ
  • EigenVec_E
  • EigenVec_Bx
  • EigenVec_By
  • EigenVec_Bz
  • pitch
  • yaw
• Added MHD system tests for:
  • Hydro only constant state test
  • MHD constant state
  • Sod test that works with PCM (this will be merged with the hydro test when MHD supports PPMC)
  • All 4 MHD linear waves
  • MHD Einfeldt strong rarefaction

HLLD Riemann Solver

• Updated interface state indexing for new format
• Added documentation
• Moved dot product function to math_utils

Testing

• Add new method, runL1ErrorTest, to SystemTestRunner that computes the L1 error compared to the initial conditions. Ideal for wave tests
• Updated hydro wave tests to use new runL1ErrorTest method

Reductions

• Add new FP atomic max that works on CUDA or ROCm
• Update time step calculation to use the grid reduction with atomic and DeviceVector
• Updated tests to match
• Remove dev_dti, host_dti_array, and dev_dti_array global variables

Utilities

Cuda Utilities

• Add function initGpuMemory which initializes GPU memory so the CUDA compute sanitizer doesn't complain about it. Used to initialize all GPU arrays that I know of.
• Add struct AutomaticLaunchParams which is a thin wrapper over the occupancy API. Primarily intended for reductions where performance is sensitive to the number of blocks but could be used for any kernel launch

DeviceVector

• New option to initialize memory

Math Utilities

New namespace and file for math utilities. Currently contains a semi-general rotation and dot product functions

MHD Utilities

• The mhd::utils::computeEnergy function now works properly with either MHD or hydro and returns the appropriate energy

Other

• Add a function to_string_exact to convert floating point numbers to a string such that it can be exactly deserialized back from a string to the same floating point number. Only used in tests currently
• Commit 379b86e in this PR should Fix Issue #219 as well.

[bcaddy]

This is currently a draft and not ready for merging but wanted to make the changes available for code review as the things that are left to do are minor and should be done shortly.

To do

• [x] Rebase off of dev again now that PR #213 is merged
• [x] Replace current indexing in MHD code with grid enums

[bcaddy]

Documentation for the runL1Error method has been added to the wiki

[bcaddy]

Converting MHD to using grid_enum is done so I'm converting this from a draft to a full PR.

@alwinm, switching to using the enum was pretty easy. Could you take a look at my additions to the enum and my general usage? I think it's what we talked about but I'd like confirmation and feedback.

[bcaddy]

FYI, I think github actions are having some issues at the moment. Some jobs are getting shutdown signals from somewhere. I double checked and everything builds fine on C-3PO and the containers running on my machine. For now I'm just restarting failed jobs until they pass.

[bcaddy]

I think I figured out the issue with builds randomly failing. The error ended up being too many resources consumed. I reduced the number of jobs that Make launched and it seems to have fixed it