n_jobs greater than 1 causes segmentation fault

Describe the bug

Setting n_jobs for sklearn.manifold.TSNE to anything greater than 1 (including -1) causes a segmentation fault

Steps/Code to Reproduce

Example:

import numpy as np
from sklearn.manifold import TSNE
TSNE(n_components=2,n_jobs=1).fit_transform(np.zeros((100,512)) # Runs fine
TSNE(n_components=2,n_jobs=2).fit_transform(np.zeros((100,512)) # Segmentation fault

Expected Results

Would expect it to work with any amount of n_jobs

Actual Results

Segmentation fault

Versions

System: python: 3.7.3 (default, Mar 26 2019, 01:59:45) [GCC 5.4.0 20160609] executable: /home/<ME>/.virtualenvs/reid/bin/python3.7 machine: Linux-4.15.0-39-generic-x86_64-with-Ubuntu-16.04-xenial

Python dependencies:
       pip: 20.0.1
setuptools: 41.0.1
   sklearn: 0.22.1
     numpy: 1.16.3
     scipy: 1.2.1
    Cython: 0.29.12
    pandas: 0.25.1
matplotlib: 3.0.3
    joblib: 0.13.2

Built with OpenMP: True

N.B. Could very well be related to #12922 but I’m not sure.

[JakeCowton]

My bad. Updated.

[lesteve]

@JakeCowton it would be great if you could try to reduce the problem further:

• can you get the segmentation fault with pairwise_distances?
• can you get the segmentation fault with only joblib.Parallel?