Performance regression on 0.52.0 wrt accumulation

Reporting a bug

• I have tried using the latest released version of Numba (most recent is visible in the change log (https://github.com/numba/numba/blob/master/CHANGE_LOG).
• I have included a self contained code sample to reproduce the problem. i.e. it’s possible to run as ‘python bug.py’.

Commit f4000439daac2aaefc20994c80e8fe9d903ce31a slows down my workload from ~60s to ~200s. I’ve slimmed down a repo to:

from numba import njit
from numpy import random, uint64, empty_like
from timeit import timeit

@njit
def fn(x, out):
    for i in range(len(x)):
        for j in range(1, 63):
            out[i] = x[i] << j  # no accumulation here but the shift exhibits the perf problem

rng = random.default_rng(0)
x = rng.integers(~uint64(0), size=500, dtype=uint64)
out = empty_like(x)
print(timeit(lambda: fn(x, out)))

on my server, f4000439daac2aaefc20994c80e8fe9d903ce31a gives ~11s and 9381206a03a3e7231d417f147474577745f6a720 gives ~3s.

[brandonwillard]

It seems like this issue might overlap with this example from the Numba Discourse.

[stuartarchibald]

@cmartinezdem Thanks for the report. Do you have a reproducer please?