R&D for TaskScheduler to have multiple queues

[hcedwar]

Potential reduce contention. Given current use of atomics for queue modification is open question as to the performance impact.

[nmhamster]

Are you considering smaller local queues for threads to reduce contention in the normal case? If their local is exhausted pull from others/master queue? If atomics are local to the queue they should complete very quickly (because zero contention except for task insertion).

[hcedwar]

Another dimension of this R&D is if cores have variable performance then try to match task priority level to core performance.

[dhollman]

Now that the tasking work from earlier this year is merged, I'm going to call this done and close this issue.

[ibaned]

Reopening with InDevelop, which is basically the same as closed but follows our workflow of not closing until a develop->master promotion.