4 The Weno benchmark

The Weno benchmark

Results obtained on the reference machine:

Computing time / Computing time in C++:

• Ju : 1.095
• PyVec : 2.620
• Numba : 14.24
• Py : 338.6

We see that:

1. Only Julia competes with C++, but optimization was not so straightforward (see the Optimizing Julia benchmarks page on this wiki).
2. For Python, the PyVec/ version is quite fast, but the coding style is quite far from the naive implementation in Py/
3. Pythran is not here! Actually implementing Weno needs to compose functions: Weno is the RHS of the Runge-Kutta method, but Weno receives the Numerical Flux as parameter, and the Numerical Flux must be built with the flux of the hyperbolic equation. This 4 levels composition is something I did not succeed to implement (note that we try to keep everything simple).