scipy: BUG: hypergeom.cdf slower in 1.8.0 than 1.7.3

Describe your issue.

While using the fisher_exact test in a loop on some 2x2 tables, I noticed a significant slowdown between Scipy versions 1.7.3 and 1.8.0 (1.8.0 is about 20x slower than 1.7.3). I narrowed it down to the call to hypergeom.cdf, and found a specific set of arguments with which the slowdown can be reproduced (see code example below).

Reproducing Code Example

import time
import scipy
from scipy.stats import distributions

ts = time.time()
for _ in range(10000):
    distributions.hypergeom.cdf(0, 48127, 57, 35775)
te = time.time()
print(scipy.__version__, '%.2fs' % (te-ts))

# Output for version info 1.7.3 1.22.2 sys.version_info(major=3, minor=8, micro=10, releaselevel='final', serial=0)
# 1.7.3 1.84s

# Output for version info 1.8.0 1.22.2 sys.version_info(major=3, minor=8, micro=10, releaselevel='final', serial=0)
# 1.8.0 41.11s

Error message

-

SciPy/NumPy/Python version information

1.8.0 1.22.2 sys.version_info(major=3, minor=8, micro=10, releaselevel=‘final’, serial=0)