Poor performance for large numpy arrays

[gimp-landg]

Given two sizeable numpy arrays with dtype=float64 and all differences below tolerance, recursive_eq is 100x slower than numpy.allclose, whereas they were supposed to exhibit comparable performance:

>>> a = numpy.zeros(int(1e7))
>>> %time numpy.allclose(a, a, atol=1, rtol=0)
CPU times: user 126 ms, sys: 60 ms, total: 186 ms
Wall time: 187 ms
>>> %time recursive_eq(a, a, abs_tol=1, rel_tol=0)
CPU times: user 11.3 s, sys: 2.08 s, total: 13.4 s
Wall time: 13.5 s