The article in the first link is pretty bad, and the entire code is undefined behavior. First of all, using inline asm + volatile to emulate well-ordered releas

The article in the first link is pretty bad, and the entire code is undefined behavior. First of all, using inline asm + volatile to emulate well-ordered release & acquire on x86 is unnecessary. That's exactly what atomics are supposed to provide for you. > In this article, I concentrate on x86, as it is the most widespread architecture rather than write generic (but slower) code. Any reasonable C++ compiler will generate simple loads and stores for atomic r&a. There's no penalty for writing generic code.