Summary
In this chapter, we have learned about the C++ memory model and the guarantees it gives to the programmer. The result is a thorough understanding of the low level of what happens when multiple threads interact through shared data.
In multi-threaded programs, unsynchronized and unordered access to memory leads to undefined behavior and must be avoided at any cost. The cost, however, is usually paid in performance. While we always value a correct program over an incorrect but fast one, when it comes to memory synchronization, it is easy to overpay for correctness. We have seen different ways to manage concurrent memory accesses, their advantages, and tradeoffs. The simplest option is to lock all accesses to the shared data. The most elaborate implementation, on the other hand, uses atomic operations and restricts memory order as little as possible.
The first rule of performance is in full force here: performance must be measured, not guessed. This is even more important...
