julia windows 64 openblas segfault

[vtjnash]

The following code snippet (from Color.jl) cause julia-win64 to segfault (probably stack corruption in DLANSP)

# Chromatic adaptation from CIECAM97s
const CAT97s = [ 0.8562  0.3372 -0.1934
                -0.8360  1.8327  0.0033
                 0.0357 -0.0469  1.0112 ]

z = lufact(CAT97s)

edit: as has been the case in the past, calling openblas_num_threads(1) fixes the issue (2 or more threads fails)
edit2: wine reports this results in stack overflow, after some delay. is that meaningful or helpful?

[vtjnash]

I think I need to switch to LIB64=0 and turn off threading for a while :(

Possibly related: OpenMathLib/OpenBLAS#214

[vtjnash]

Alternatively, setting LIB64=0 and -Wl,--stack,0x40000000 helps the test code from the gist in #2124 pass (for a small number of threads, limited by the amount of memory wine is willing to allocate me on julia.mit.edu for a given run)
This, by the way, is a bit excessive: https://www.google.com/search?q=0x40000000+bytes+in+gigabytes

(the required stack appears to grow proportionally to the number of openblas threads. 256MB each was sometimes enough for 10 threads)