Question

• Using C, C++ or Fortran give essentially identcial performance
• Of the JIT solutions:
  • Cython is the fastest but needs the extra work of type annotations
  • numba is almost as fast and simplest to use - just say jit(functiion)
  • numexpr is slightly slower and works best for small numpy expressions but is also very convenient
• A pure numpy solution also perfroms reasonably and will be the shortest solutoin (a one-liner in this case)
• The pure python approach is very slow, but serves as a useful template for converting to native langauge directly or via a JIT compiler
• Note that the fsatest alternatives are approximately 1000 times faster than the pure python version for the test problem with n=1000 and p=3.