most of the profile runtime is in task_done_hook unless threads = 1 #41713
Description
When the number of threads is not 1, a lot of time is reported in task_done_hook. This makes profiling results confusing. julia -t 1 is a workaround.
julia> VERSION
v"1.8.0-DEV.271"
julia> using Profile
julia> f(n) = [sum(randn(100)) for _ in 1:n]
f (generic function with 1 method)
julia> f(1);
julia> @profile f(10^5);
julia> Profile.print(; maxdepth = 3)
Overhead ╎ [+additional indent] Count File:Line; Function
=========================================================
╎261 @Base/client.jl:495; _start()
╎ 261 @Base/client.jl:309; exec_options(opts::Base.JLOptions)
╎ 261 @Base/client.jl:379; run_main_repl(interactive::Bool, quiet::Bool, banner::Bool, history_file::Bool, color_set::Bool)
╎ 261 @Base/essentials.jl:716; invokelatest
╎530 @Base/task.jl:531; task_done_hook(t::Task)
╎ 530 @Base/task.jl:823; wait()
529╎ 530 @Base/task.jl:814; poptask(W::Base.InvasiveLinkedListSynchronized{Task})
╎265 @Base/threadingconstructs.jl:178; (::Distributed.var"#145#147")()
╎ 265 @Distributed/src/remotecall.jl:260; macro expansion
╎ 265 @Base/condition.jl:78; lock
╎ 265 @Base/lock.jl:187; lock(f::Distributed.var"#146#148", l::ReentrantLock)
1╎1 @Random/src/normal.jl:38; randn(rng::Random.TaskLocalRNG)
2╎2 @Random/src/normal.jl:70; randn_unlikely(rng::Random.TaskLocalRNG, idx::Int64, rabs::Int64, x::Float64)
Total snapshots: 1060