Chris Elrod

Thanks for the issue. That error was fixed here: https://github.com/JuliaSIMD/CheapThreads.jl/commit/4120d333183b455796f423fb1271d492754b164a But I still need to add a test. As an aside, `@inbounds` doesn't penetrate closures, so it won't work with...

Actually, this isn't fixed because the iteration splitting code assumes `step` is defined. Making anything `Iterators.product`-like work using `CheapThreads`'s current behavior would require some hacky overloads. Should make try and...

Even though you fixed the crashes in #21, I'll leave either this or the original issue about allocations open until `NamedTuple`s work well.

Are you by any chance on a 4 core/8 thread laptop? The jump is quite large here: ```julia julia> @btime send_to_workers(:f1!, p, inchannels, outchannels,3) 68.613 μs (8008 allocations: 375.28 KiB)...

No reason for this to be `@generated`: ```julia function _stop_working_threads(threadmask, nthread, torelease) threads = CheapThreads.UnsignedIteratorEarlyStop(threadmask, nthread) tm = CheapThreads.mask(threads) tid = 0x00000000 while true VectorizationBase.assume(tm ≠ zero(tm)) tz = trailing_zeros(tm)...

> Ah I don't quite understand when things need to use `@generated` - why does `_launch_working_threads` need it? They need to be `@generated` when you want to change the function's...

> Yes a copy and paste works, except `stop_workers` should not be run before the benchmarks Ha ha. Just to confirm, I was definitely running `stop_workers` before running the benchmarks....

> I can't find anything about pauses anywhere - do you know of an example I could look at for this? [This](https://github.com/JuliaLinearAlgebra/Octavian.jl/blob/3a95eebe3e238aee589b8a20c93d165a2724197e/src/matmul.jl#L441-L451) is probably not the best way to do...

This is probably the usual problem of Polyester not playing well with base threads, given that FFTW seems to use them https://github.com/JuliaMath/FFTW.jl/pull/105 Does the second `@time` also show that FFTW...

> > Does the second @time also show that FFTW is slower? > > No thats way I make two calls. Its just the first that is affected. That's what...