What is Test-and-Set used for?

Question

After reading the Test-and-Set Wikipedia entry, I am still left with the question "What would a Test-and-Set be used for?"

I realize that you can use it to implement Mutex (as described in wikipedia), but what other uses does it have?

Answer 1

A good example is "increment."

Say two threads execute a = a + 1. Say a starts with the value 100. If both threads are running at the same time (multi-core), both would load a as 100, increment to 101, and store that back in a. Wrong!

With test-and-set, you are saying "Set a to 101, but only if it currently has the value 100." In this case, one thread will pass that test but the other will fail. In the failure case, the thread can retry the entire statement, this time loading a as 101. Success.

This is generally faster than using a mutex because:

1. Most of the time there isn't a race condition, so the update happens without having to acquire some sort of mutex.
2. Even during collision, one thread isn't blocked at all, and it's faster for the other thread to just spin and retry than it would be to suspend itself in line for some mutex.

Answer 2

You use it any time you want to write data to memory after doing some work and make sure another thread hasn't overwritten the destination since you started. A lot of lock/mutex-free algorithms take this form.