Beyond Sequential Consistency - Leveraging Atomics for Fun and Profit

Explore advanced C++ memory models and atomic operations in this comprehensive conference talk that moves beyond the default sequential consistency guarantees to unlock significant performance improvements. Learn how C++'s formally defined memory model, introduced in 2011, provides the foundation for portable multi-threaded code, then dive deep into weaker memory orderings including acquire/release and relaxed semantics. Examine practical implementations through a ring buffer example that demonstrates how strategic use of weaker synchronization can maintain correctness while boosting performance. Discover how the C++ memory model maps to real hardware architectures, with detailed comparisons between x86's native guarantees and less coherent platforms like ARM. Master the techniques for leveraging atomic operations effectively in high-performance applications while understanding the trade-offs between safety and speed in concurrent programming scenarios.