https://www.youtube.com/watch?v=UdKqfQ3a_sY&t=36s

Fastest/best locking mechanism depends also on the parallel workload to be done:

• low contention: use atomics
  • aviod CAS loops
• high contenction: use spinlocks
  • tune the spinlock for high contention
  • do not put the spinlock and its payload on the same cache line (!)
• read ops on shared vars create high latency for writers