ENH: Performance and practicality tests with various Python speed up alternatives (cython, numba, ...) about pycopancore HOT 4 CLOSED

Just to through it in the round: PyDSTool (which Jan seems to have used already)can seemingly compile code in-time to C and is precisely made to solve any kinds of dynamical systems with a lot of speed.
http://www.ni.gsu.edu/~rclewley/PyDSTool/FrontPage.html

from pycopancore.

This is mainly for bifurcation analysis for odes, but we have a more general model class.

from pycopancore.

Any results in this so far?

from pycopancore.

Just a short note:
We should check whether Cython is compatible with ctypes, but I couldn't find anything online. As numba has the cfunc decorator as an interface to C, this would mean one could code in cython but extensions in numba could be included, too. I attached a small archive containing a comparison of pure (unoptimized) C-code and numba using the cfunc decorator interfacing to C and the speed is about the same. For compilation, use gcc -fPIC -shared -o libreduce.so reduce.c . Maybe somebody with more experience than me could could extend this example Cython? (@wbarfuss?)

PS1: nb-ctypes.tar.gz
PS2: Note that the fully-optimized C-Code can be a factor of 10 faster than numba, because it can use inlining and direct cpu directives. (Interestingly, the reduce.so produced by gcc is quite a bit slower than using the (afaik somewhat more modern) clang compiler when using the -O3 flag for both.)

from pycopancore.