Infrastructure

gravitation accuracy (analyze kernels' relative numerical accuracy)
Support for conda
Add tests (pytest is desired)

Benchmark dimensions

half, single and double precision floating point (where possible)
Python interpreters for benchmark workers (where applicable) - CPython 3.x, pypy, different compiler versions, different compile-time optimizations
C- / Fortan-compilers - clang/LLWM, GCC >= 4.x, Intel, ...

Kernels

See "Desired Kernels" in README.md for further ideas. Prepared but not yet integrated code:

C with SSE2 intrinsics single-thread
Plain C multi-thread openMP
Visual Basic / VBA

Kernel Design & Implementation Issues

make gravitation worker pypy-compatible (again ...)
np4: improve joblib worker initialization
oc4: improve joblib worker initialization
C code: translate comments into English
pc3: figure out what goes wrong (bodies keep "disappearing")
gravitation.lib.simulation: translate comments into English
gravitation.lib.simulation.create_galaxy: test against original C implementation
np3, np4 and other parallel kernels: code for computing batch size is redundant and broken for small numbers of bodies ("line index tuples for evenly sized batches")