16 setup benchmarking infrastructure

[lorenzovarese]

#16

• Separate in different modules the mesh definitions in order to retrieve them in benchmarks
• Create a generic suite of benchmarks in benchmark/benchmarks.jl
• Create a specific suite for advection in benchmark/benchmarks_advection.jl
• Make the suite compatible with AirspeedVelocity and document the usage of AirspeedVelocity on the README.md of the benchmark folder.

[lorenzovarese]

@tehrengruber
Main changes:

• Split the benchmarks into two parts/files:

  1. benchmark.jl: includes all the benchmarks for broadcast cases, trigonometric functions, neighbor sum, and remapping.
  2. benchmark_advection.jl: contains all (supported) benchmarks for advection, including the Upwind scheme for both embedded and Python backends, and the Mpdata program, embedded only.

• The benchmarks are supported by running them directly or using the AirspeedVelocity package. I have added a README.md to the benchmark folder that contains additional instructions useful for running the benchmarks.

• Some modifications that were left from issue Improve Atlas #15 are now implemented here. This includes the separation of the simulation loop from the advection setup of the miniapp and the creation of a new module for AtlasMeshes.

There are no optimizations present in this branch. The objective is to merge this into the origin and then tag the merge commit as a baseline for all other optimizations.

P.s. the plot is working. After running the benchmark with benchpkg you can generate easily the plot with the command benchpkgplot. Here a basic example on two commits: