Why tasks_count_ decrease 1 out of side the Execute for loop?

[XimingCheng]

scheduling/include/scheduling/scheduling.hpp

Line 542 in ef687f5

if (tasks_count_.fetch_sub(1) == 1) {

Above code each loop in Execute function will done a task, but tasks_count_ decrease 1 is out of side the Execute for loop, I do not know why. Could you tell me the reason, thanks!

[dpuyda]

Hi @XimingCheng !
Thank you for your interest in this small repository, and especially for looking into the code! I do not remember all the details right now to be honest, but as I can see in the code, task_count_ is used to know when there are no tasks so that the scheduler can be put into sleep and for the Wait purposes. Therefore, we only care whether it reaches zero or not. Under these circumstances, we may decrement it when a task and some of its sub-tasks are completed, it is not necessary to decrement it as soon as each individual task completes. I think I've put this check outside the loop mostly to avoid unnecessary ifs in the loop.

[XimingCheng]

Thanks, I got it. @dpuyda !

[dpuyda]

Thanks again for your interest! 👍🏻

[XimingCheng]

Analysis with CGraph, I will test CGraph and find out why scheduling is slower than CGraph as my Mac also run very slow.

[dpuyda]

Yeah, I recall the discussion with @ChunelFeng, but I don't have the answer yet🤷🏻‍♂️ We did some experiments back then, but did not come up with the explanation...

[ChunelFeng]

Analysis with CGraph, I will test CGraph and find out why scheduling is slower than CGraph as my Mac also run very slow.

It is not recommended to perform performance testing on a MacBook. Consider using a clean Linux system instead.

[XimingCheng]

Analysis with CGraph, I will test CGraph and find out why scheduling is slower than CGraph as my Mac also run very slow.

It is not recommended to perform performance testing on a MacBook. Consider using a clean Linux system instead.

Why?