Implement np.lcm

[ivirshup]

Building on #4780, this PR implements np.lcm as a ufunc. It's very similar to that PR, and uses much of the same boiler-plate.

[ivirshup]

@esc, I think this is ready for review – though that might be easier once the gcd branch merges. The only relevant commits here are the most recent two. Two notes:

• This doesn't have overflow guards, but neither does numpy.
• I've only tested this like the other gufuncs are tested, but this method also doesn't exist in math, so there are fewer tests running on the underlying implementation.

Also, here is a quick benchmark:

import numpy as np
from numba import njit

def lcm(a, b):
    np.lcm(a, b)

nlcm = njit(lcm)
plcm = njit(parallel=True)(lcm)

a = np.random.randint(0, 10000000, 10000)  
b = np.random.randint(0, 10000000, 10000)

# warmups
nlcm(a, b) 
plcm(a, b)

%timeit lcm(a, b)
1.45 ms ± 22.4 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

%timeit nlcm(a, b)
591 µs ± 6.99 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

%timeit plcm(a, b)
133 µs ± 14.4 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)

[esc]

@ivirshup thanks for updating this, I have changed the label of the PR accordingly. The benchmarks look impressive from a first glance. 👍

[stuartarchibald]

/AzurePipelines run

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[stuartarchibald]

@ivirshup #4780 is merged if you want to rebase/merge in?

[ivirshup]

@stuartarchibald, done rebasing

[stuartarchibald]

Thanks for the PR. Looks good.