Supports Pythran, Cython, Numba, …

Transonic can use different tools to accelerate the code. We use the name backend.

The default backend is Pythran and currently, there are 3 other backends (“cython”, “numba” or “python”). The “python” backend is mainly used for internal testing.

There are different methods to choose which backend is used:

• An environment variable TRANSONIC_BACKEND (which has to be “pythran”, “cython”, “numba” or “python”) should be used to change the backend globally for one process.

• The transonic command-line has an option -b (--backend).

• The functions transonic.backends.make_backend_files() and transonic.dist.init_transonic_extensions() have an optional argument backend. Note that these functions should be imported from the transonic.dist package.

• There are two functions transonic.config.set_backend() and transonic.backends.set_backend_for_this_module(). Note that these functions can be imported from the transonic package.

• The boost and jit decorators have an optional argument backend.

• Unified code for different backends
  • Ahead-of-time compilation
  • Just-in-time compilation
• Pythran backend
  • Why Pythran is great for Python users doing numerical computing?
  • Why Transonic is great for Pythran users?
• Cython backend
  • Cython syntaxes already supported
  • Cython syntaxes partly supported
  • Cython syntaxes that can be supported quite easily
  • Cython syntaxes that will be difficult to support
• Numba backend