Timing Clebsch-Gordan-coefficients calls in Julia (beating c++ again)

Specific Domains Numerics
3

Hi Steven, these are great suggestions!
The amount of zeros is ≈ 17%, and 14% are equal to 1.0. I tried to exclude obviosly-trivial Clebsches in the rand_clebsch() function (+fixed a typo).

Just checked the other suggestions:

  • type stability + logfact2 gives 1-2 μs gain
  • @inbounds removes extra 2-4 μs for the same tests.

Not so much, but still cool. Thanks a lot for pointing out!

Did you notice if I am doing something wrong with SymPy calls? My colleagues running tensorflow used to calculate both the Clebsch-Gordan coefficientss and the Wigner d using the sympy.physics.wigner. It is hard to believe that the call time which they are working with is x1000. I have not benchmarked it thoroughly, tho.

1 Like