RFC: Track queued FFTW thread count to aid profiling

[IanButterworth]

This demo tracks the number of queued FFTW threads (#234).

The idea being that it could be monitored in profiling tooling to identify poor balancing of nthreads() vs. FFTW.get_num_threads(), for instance within a Threads.@threads for loop.

julia> FFTW.set_num_threads(Threads.nthreads())

julia> @time begin
	         @show Threads.nthreads() FFTW.get_num_threads()
	         t = Timer(0, interval = 0.01) do t
	             print(FFTW.num_queued_threads[], " ")
	         end
	         x = rand(1000,1000,10)
	         Threads.@threads for i in 1:(Threads.nthreads() * 10)
	             fft(x)
	         end
	         close(t); println()
       end
Threads.nthreads() = 6
FFTW.get_num_threads() = 6
10 20 22 25 23 21 20 11 11 15 19 16 12 13 14 8 4 12 21 26 25 16 3 3 1 12 28 27 25 20 20 13 2 1 0 16 25 26 25 22 19 17 12 10 0 28 23 16 11 10 8 7 6 3 2 2 1 1 0 29 26 23 19 15 14 11 5 9 11 22 22 21 19 6 7 6 8 19 24 20 15 11 10 7 6 3 2 2 1 0 18 23 23 18 13 0 3 
  7.885598 seconds (48.46 k allocations: 17.959 GiB, 10.47% gc time, 0.51% compilation time)

In a simple loop example like above, it should be pretty obvious to the user that they should balance nthreads() and FFTW threads.

But in the case where FFTW is used deep inside a library function, it may not be obvious when FFTW is being used, and poor balancing might be impeding performance.

One idea might be to do a @warn "..." maxlog=1 with advice about checking load balancing if significantly more queued FFTW threads are detected than Sys.CPU_THREADS, but it seems @warn doesn't work within a @cfunction. It would probably be too opinionated though.

[codecov]

Codecov Report

Merging #235 (19f7f37) into master (17bc81a) will increase coverage by 0.31%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #235      +/-   ##
==========================================
+ Coverage   61.07%   61.39%   +0.31%     
==========================================
  Files           5        5              
  Lines         483      487       +4     
==========================================
+ Hits          295      299       +4     
  Misses        188      188              

Impacted Files	Coverage Δ
src/providers.jl	45.16% <100.00%> (+8.12%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 17bc81a...19f7f37. Read the comment docs.