Implement sumfact kernels in separate functions
This has two main advantages:
- The scalability issue of the loopy scheduler is under control
- Having a small kernel object is better suited for optimization of the kernel
The following things remain to do:
-
Implement basic infrastructure -
Jacobians -
Instrumentation -
Vectorization of Quadrature Loop -
Vectorization of sum factorization kernels -
FastDG -
Systems -
Hi/Lo Vectorization in FastDG -
Cleanup
This fixes #75 (closed) .
Edited by Dominic Kempf