Talk PyPy's JIT for scientific python computations

A common problem when developing scientific applications is that CPython is too slow. This can mitigated by usage of FORTRAN, C, or Cython in performance-critical parts of the code, but it would be much nicer if python ran faster.

Creating a JIT has been one of the main objectives of the PyPy project for quite a while now. As of March 2009, we're able to speed up simple examples in Python about 20-30x over the speed of CPython. Since PyPy's JIT is much more flexible than Pysco we expect even greater speedups in the future, as well as support for floats and 64bit architectures, which Psyco will not achieve in forseeable future.

In this talk, we will discuss the design principles and future of PyPy's JIT but won't go into too many architectural details.

We will also present how JIT can make the life easier for simple postprocessing of data generated by nonhydrostatic model of geophysical flows - EULAG for the case of boundary layer clouds observed during various experimental campaign.
tagged by
no related entity