mdmaas/julia-numba

Simple benchmark of Julia vs Numba

julia-numba

Simple benchmark of Julia vs Numba

julia-numba's Issues

Faster Numba script

Very interesting comparison! I really like it so I tried it my own. What I found is, if you improve the code a little bit and enable the fastmath option, you can see an approximate 10X speed up.

@nb.jit(nopython=True, fastmath=True)   
def quad_trap(f,a,b,N):
    h = (b-a)/N
    integral = h * ( f(a) + f(b) ) / 2
    for k in range(N):
        xk = h * k + a
        integral += h*f(xk)
    return integral

I am using an Intel i9-9880H CPU with Numba version 0.55. The above script could finish quad_trap in 0.000028 seconds.
BTW, Numba also supports SIMD, so in your loop function you may want to have computations that can be run in parallel. I am not an expert on both Numba and Julia, but I am able to write a sparse matrix multiplication function in numba that is more than 5 times faster than scipy sparse.