@threads is really working in parallel?

Specific Domains Julia at Scale
1

Hi,

on my system:

julia> versioninfo()
Julia Version 1.1.1
Commit 55e36cc308 (2019-05-16 04:10 UTC)
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: Intel(R) Core(TM) i5-4210M CPU @ 2.60GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-6.0.1 (ORCJIT, haswell)
Environment:
  JULIA_EDITOR = "C:\Users\WildKev\AppData\Local\atom\app-1.38.2\atom.exe"  -a
  JULIA_NUM_THREADS = 4

the following code

using Plots

a = falses(1000)
a[100] = true

function tfunc(a)
   stop = Threads.Atomic{Bool}(false)
   threadid = zeros(size(a))

   Threads.@threads for i = 1:length(a)
     threadid[i] = Threads.threadid()
     stop[] && break
     a[i] && Threads.atomic_xchg!(stop, true)
   end

   return threadid
end

plot(tfunc(a))

produces the following threadid over element plot.

image
image1366×728 21.5 KB

It looks like the splitted iteration space runs sequential.

2

what does Threads.nthreads() say?

3 
julia> Threads.nthreads()
4

In the plot the y-axis also indicates that they are in use.

4

Which part makes you think “It looks like the splitted iteration space runs sequential.” Did you expect everything to start exactly at the same time?

5

You are hitting thread start up time…it takes time to start the 3 additional threads.

If you change it to:

   a[600] = true

You will get some interesting graphs and on my machine (4 core) the result is different every run.

This might be more what you are looking for. Basically I give each thread 3 seconds to start and become stable before processing the array:

using Plots
using Dates

a = falses(1000)
a[100] = true

function tfunc(a)
   stop = Threads.Atomic{Bool}(false)
   threadid = zeros(size(a))

   start = (second(now()) + 3) % 60
   Threads.@threads for i = 1:length(a)
     while second(now()) != start
     end

     threadid[i] = Threads.threadid()
     stop[] && break
     a[i] && Threads.atomic_xchg!(stop, true)
   end

   return threadid
end

plot(tfunc(a))