Chris Elrod

No, doing that, and I do see the warning. You can't nest `@turbo`s. Only one layer can SIMD. You could write out loops manually and place an `@turbo` on the...

Presumably caused by: https://github.com/JuliaSIMD/ThreadingUtilities.jl/commit/290f4a2478f7ad06d869876c4949b0ae356a76f2 I wanted cleaner JET reports.

The discussion links don't work for me.

> Overall, IIUC, Polyester really just creates, manages, and re-uses a bunch of sticky tasks and just runs them on existing Julia threads (similar to @threads :static). `@threads :static` does...

Eh, I suspect it's a bit broken. Performance also isn't great. Would be great if I (or someone else) can get around to reimplementing. I described the approach I'd take...

I see the same thing with `Threads.@threads`. ```julia julia> using Polyester julia> function reset_x!(x) x .= 0 end reset_x! (generic function with 1 method) julia> function without_batch(x) for i in...

It may be interesting to uncover why. Possible ideas: 1. Threads take a while to go to sleep; they can hog system resources until then. 2. Clock speeds. Could mess...

> Wouldn't that rule out your first idea? Yes.

You could try counting cpu cycles with [cpucycle](https://github.com/m-j-w/CpuId.jl).

Note that by default, Polyester only uses one thread per physical core. You can change this (`@batch per=thread`). You could also get `@threads` to only use one per core, by...