How to use LoopVectorization.jl?

[skleinbo]

I can’t tell you how to fix it, but you might be hitting a (current) limitation of LoopVectorization. From the
README

The macro assumes that loop iterations can be reordered. It also currently supports simple nested loops, where loop bounds of inner loops are constant across iterations of the outer loop, and only a single loop at each level of loop nest. These limitations should be removed in a future version.

Also, “thread-local state”, i.e. indexing arrays by thread id, is bound to break. Please see

PSA: Thread-local state is no longer recommended; Common misconceptions about threadid() and nthreads()

Here’s a new blogpost we’ve been working on for some weeks now intended to explain and help fix a common and incorrect pattern that has been showing up in multithreaded julia code The TL;DR is that if you’ve written code with things like state[Threads.threadid()] you’ve probably created a race condition and you should fix it immediately Shout out to @jling for pointing out the scale of this problem, and for encouraging us to think hard about how it should be approached and fixed.