BUG: AVX-512 path for `ufunc.accumulate` is broken

We use np.minimum.accumulate in statsmodels. Recent pre-release tests have started failing on after calls to np.minimum.accumulate.

This code only fails on systems with AVX-512. Passes on systems with AVX and AVX2.

Reproducing code example:

import numpy as np
ntests = 10
pvals = np.random.random_sample(10)
pvals_corrected_raw = np.arange(ntests, 0, -1) * pvals
pvals_corrected = np.minimum.accumulate(pvals_corrected_raw[::-1])[::-1]

Failing job:

https://travis-ci.org/statsmodels/statsmodels/jobs/651893167

Error message:

No error, but wrong results. Likely strides are off since we often seem empty-like values in places.

Numpy/Python version information:

Python: 3.7.5.final.0
cython: 3.0a0
numpy: 1.19.0.dev0+4d2b585
scipy: 1.5.0.dev0+f614064 
pandas: 1.1.0.dev0+500.g06eb8db97 

[r-devulap]

yup, taking a look …

[r-devulap]

I brought up this question when I submitted the patch (see comments https://github.com/numpy/numpy/pull/14867#issuecomment-560984763). There are no tests covering np.maximum.accumulate. Can someone help me understand under what scenarios do we expect input and output arrays to overlap in memory to ensure we haven’t missed all the use cases.