Re: [PATCH v4 0/8] MR6031: Draft: wine: Use atomic locks insted pthread_mutex.

[Alfred Agrell (＠Alcaro)]

On Wed Jul 10 20:47:56 2024 +0000, Grigory Vasilyev wrote:
> v4. The fastest implementation that is theoretically possible and can be
> implemented on major platforms. Almost twice as fast as pthread mutex.
> **TODO: Recursive Mutex.**
> Time elapsed: t1=3.191000ms, t2=5.786000ms
>
> Thanks to the article `Futexes Are Tricky` by `Ulrich Drepper` from `Red
> Hat, Inc.`
> https://dept-info.labri.fr/~denis/Enseignement/2008-IR/Articles/01-futex.pdf

Before trying to implement recursive mutexes, you should create a benchmark showing that this noticably speeds up Wine, not just a microbenchmark. As mentioned, futexes are tricky, and we need proof that this MR improves things enough to be worth the added complexity.