Open alanedelman opened 5 years ago
Note, the discretization in shallow_water.jl
is not the final target discretization. We are planning on using entropy stable discretizations (such as [1] and [2]) on unstructured grids and may need to include control over oscillations in the solution to get the desired results [3].
This usually means the right level of abstraction for us is matrices and vectors but feel free to prove me wrong. If approxfun can provide an entropy stable discretization with control over the wiggles then great!
Is it possible to refactor the shallow water code with approxfun and put the gpu efforts into approxfun rather than the shallow water code.
This may seem like a hard question, and I know there are always gotchya's, but if this could work then the multiplier would be valuable.
I think we can start with the non approxfun direct approach, as it will be less risky and valuable anyway, but if maybe just maybe my thinking is right, the multiplier would be huge. (It's not just this one code, etc etc, ....)