Add ExaHype kernel benchmarks

[tkoskela]

Tobias Weinzierl has given us access to ExaHype kernel benchmarks in Peano (Benchmarks described around here).

ExaHyPE 2’s implementation of the Euler equations is shipped with a benchmark to assess the speed of the compute kernels. The benchmark solely computes the throughput on one core or one GPU and thus provides key information how efficient these core kernels perform on your system

The benchmark gives you the total runtime and cost per degree of freedom update (throughput) on one core or GPU, respectively. Data is provided for various patch sizes batched (fused) into one large patch assembly, for patches which exploit stateless operations, and different kernel realisations.

To include these in the framework, we need to

• [x] Contact Tobias
• [x] Create a spack recipe
  • I've written a spack recipe using their autoconf build system. It needed a bit of customisation so to do autoreconf correctly, and the final build stage requires you to call another python script. They have also developed a cmake build system which might make things more straightforward.
• [ ] Create a ReFrame script
  • They have some ReFrame tests in the repository but those are not maintained and won't work.
  • They are working on a new CI pipeline using Docker images which will eventually involve new ReFrame scripts
  • They use Docker to give you the right compiler toolchain for GPU offloading

[tkoskela]

Spoke with Sean Baccas yesterday. He is working on CI for Peano and will work on creating ReFrame scripts.

[tkoskela]

@tkoskela needs to talk to Sean about this again

[tkoskela]

There is a spack package for ExaHype 2.

However, the Spack package currently works only for the branch "cmake-install" as it requires some restructuring in the CMake build system which is being done in one of the open draft MRs.

All is documented here if you scroll down to the containers.