Just to jump on the numba train, I've generally found it to reliably obtain C like performance from C-like Python code. This property also holds when you use Py

Just to jump on the numba train, I've generally found it to reliably obtain C like performance from C-like Python code. This property also holds when you use Python as a preprocessor language for generating computational kernels, which provides a lot of flexibility not evident in the documentation. It also has simple-to-use openmp-like multicore parallelization, limited class support, AOT compilation and CUDA & AMD HSAIL support.