Home/ Questions/Q 6950875
In Process

The Archive Base Latest Questions

Editorial Team
  • 0
Asked: 2026-05-27T14:08:50+00:00

In this code sample I benchmark scoped vs manual locking of a pthread mutex.

  • 0

In this code sample I benchmark scoped vs manual locking of a pthread mutex. I expected equal perfomance for both approaches. But much to my surprise the scoped lock solution seems a little faster.

Can anyone explain why this happens?

I’m using using g++ 4.5.2 and compile with the following options:

g++ -std=c++0x -O2 -o test main.cpp

Update

When doing 2 billion iterations on my local PC (not Ideone) I get a larger difference:

g++ -std=c++0x -O2 -o test main.cpp
scoped_lock: 10530ms
normal: 11290ms
scoped_lock: 10530ms
normal: 11280ms
scoped_lock: 10530ms
normal: 11280ms
scoped_lock: 10530ms
normal: 11290ms

Update 2

I improved the program with a higher resolution clock: http://ideone.com/CMbuw. This time the scoped lock is a little slower.

I think it’s safe to conclude that it’s a measuring anomaly.

Share

Leave an answer

You must login to add an answer.

Need An Account,

1 Answer

  1. Editorial Team

    I changed the iteration count to 50 millions (was 20 millions in your code) and the results are now

    result: Time limit exceeded     time: 5s    memory: 2828 kB     signal: 24 (SIGXCPU)
input: no
output:
scoped_lock: 740ms
normal: 740ms
scoped_lock: 710ms
normal: 710ms
scoped_lock: 720ms
normal: 720ms

    perfectly identical. So I’d blame measurement errors in your test case – just not enough iterations of code. If you really care you should look into emitted machine code.

    • 0