mma.rst

Mixed Martial Arts

Jonas Juselius <[email protected]>
HPC@UiT

Why

• There exists no perfect language for all tasks

• Languages have different stenghts:

  • Python is excellent for driving computations and pre- and post-processing results
  • Fortran is suitable for high-performance linear algebra and mathematical programming
  • C++ supports advanced data structures, generic programming and more
  • C is the mother tongue of UNIX
  • Haskell is purely functional and lazy

• Coupling functionality from existing programs

How

• C is the common denominator
• Export needed functionality to C

• Often necessary to modularize the host code:

  • init()
  • step()
  • finalize()
  • + accessors

• Couple either directly or via Cython
• Use CMake to build and link

Fortran

• Use iso_c_binding

• Calling conventions:

  • Pass by value
  • bind(C)
  • c_ptr, c_loc, c_f_pointer
  • Column vs. row major matrices

Fortran calling C

C calling Fortran

C++

C++ cannot be directly linked due to name mangling.

Cython

• Compile Python to C!

  • Speedup Pyhton
  • Call external C/C++ functions directly
  • Pass arrays via numpy
  • Use CMake!

• https://github.com/bakerjonas/python-tutorials/tree/master/Cython

Cython for speed

Cython calling C

Calling Cython from Python

Cython CMake

Haskell

• FFI (Foreign Functions Interface):

  • Call C functions from Haskell
  • Export Haskell functions to C