Failures in threaded_replicate test

[dmbates]

I am pretty consistently getting about a 50% failure rate on that test. @palday Do you also get failures on that? If so, can we suppress it until we find out why it is failing?

[palday]

That test does occasionally fail for me, but almost never nowadays. (An earlier version of the test was a little bit too strict and failed a fair amount.)

• Which Julia version?
• Which BLAS? If OpenBLAS, which build? (This is displayed at the start of the testsuite.)
• How many threads do you run Julia with?
• What's your precise CPU architecture? (Trying to figure out which SIMD instructions are available)

I suspect there's an interaction between threading and SIMD that can fail in certain configurations.