torch.norm(x-y, 2, 1) is significantly slower than torch.sqrt((x - y).pow(2).sum(1))

[zou3519]

I've tested a few other norms as well:

x = torch.randn(1024, 256)
y = torch.randn(1024, 256)


In [12]: %timeit torch.norm(x-y, 1, 1)
2.55 ms ± 253 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

In [13]: %timeit (x-y).sum(1)
339 µs ± 699 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each)

In [14]: %timeit torch.norm(x-y, 2, 1)
2.42 ms ± 33.7 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

In [15]: %timeit torch.sqrt((x-y).pow(2).sum(1))
736 µs ± 2.3 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

In [16]: %timeit torch.pow((x - y).pow(3).sum(1), 1/3)
700 µs ± 571 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each)

In [17]: %timeit torch.norm(x-y, 3, 1)
16.1 ms ± 31.5 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

Reported here on the forums