atomic_queue

atomic_queue is a header-only C++ library by Maxim Egorshin providing several lock-free MPMC queue implementations. Its OptimistAtomicQueue consistently wins C++ benchmarks for highest throughput and lowest latency: sub-100 ns round-trip push-to-pop in MPMC mode, SPSC throughput exceeding 200M ops/s, and 200–500M+ ops/s in 1P/1C configuration.

What's inside

The library provides a family of fixed-capacity, ring-buffer-backed queues:

• AtomicQueue — the conservative MPMC implementation (CAS-retry on a per-slot state)
• OptimistAtomicQueue — the headline, optimized hot path; busy-waits on slot state
• AtomicQueueB / OptimistAtomicQueueB — bounded variants using element-as-sentinel
• SPSC specializations that drop the atomic RMW entirely

All use power-of-two capacity for cheap modular arithmetic and apply cache-line padding aggressively.

Why it wins on latency

OptimistAtomicQueue is engineered for the uncontended case: when the next slot is in the expected state, the operation completes in two atomic loads plus a release store, no RMW. Under contention it falls back to a brief busy-wait. This makes it ideal for low-latency systems where the queue is rarely actually contended (HFT, real-time DSP, telemetry pipelines) but absolutely needs sub-100 ns latency on the common path.

When to use it vs alternatives

• atomic_queue — lowest latency, strict FIFO, x86 and ARM. The default for new C++ code that needs raw speed.
• moodycamel-concurrent-queue — higher absolute throughput in many-producer scenarios, but no strict FIFO across producers.
• LCRQ C++ port (Pedro Ramalhete) — higher MPMC throughput at very high thread counts on x86, but more complex and CAS2-dependent.

See also

• mpmc-queue — broader landscape
• spsc-queue — SPSC tier
• the-fastest-queue — full hierarchy