OpenBLAS randomly crashes / deadlocks 2

[mderickx]

The first of the two bugs at #22021 popped up again in sage 8.1.beta6 on ​arando

sage -t --long src/sage/matrix/matrix_integer_dense.pyx
    Killed due to segmentation fault
**********************************************************************
Tests run before process (pid=3193) failed:
sage: a = matrix(ZZ, 3,3, range(9)); a ## line 23 ##
[0 1 2]
[3 4 5]
[6 7 8]
sage: a.det() ## line 27 ##
0
...
12
sage: A*C == d*B ## line 4265 ##
True
sage: A = random_matrix(ZZ, 4, 4) ## line 4270 ##
sage: B = random_matrix(ZZ, 2, 3) ## line 4271 ##
sage: B._solve_iml(A) ## line 4272 ##
sage: A._solve_iml(B, right=False) ## line 4276 ##
sage: A._solve_iml(B, right=True) ## line 4280 ##
sage: A = random_matrix(ZZ, 2000, 2000) ## line 4287 ##
sage: B = random_matrix(ZZ, 2000, 2000) ## line 4288 ##
sage: t0 = walltime() ## line 4289 ##
sage: alarm(2); A._solve_iml(B)  # long time ## line 4290 ##
------------------------------------------------------------------------
sage: t = walltime(t0) ## line 4294 ##

[embray]

I think maybe this has been fixed upstream by ​https://github.com/xianyi/OpenBLAS/pull/1234 . I'm going to test it.

[jdemeyer]

Beware that #24638 also patches openblas.

[embray]

Apparently we've already upgraded to OpenBLAS 0.2.20 which appears to include the fix. Then the only thing to do would be to drop the USE_THREAD=0 from the OpenBLAS build.

[embray]

Seems to work on my Ubuntu machine. Indeed, this ticket was reported against 8.1.beta6 which was still on OpenBLAS 0.2.19, but this issue was only fixed in OpenBLAS 0.2.20

[embray]

If #24669 gets merged and works for people then this can just be closed as worksforme.