ReadModifyWrite

Read-modify-write (RMW) refers to an operation that reads a memory location, computes a new value from the read data, and writes the result back, ideally as an indivisible action from the perspective of other threads. Many computer architectures provide atomic RMW instructions that perform the read, compute, and write in a single, uninterruptible step, guaranteeing that concurrent operations do not observe intermediate states.

Common atomic RMW operations include fetch-and-add, test-and-set, compare-and-swap (CAS), and exchange (swap). These primitives enable lock-free synchronization and are foundational for building concurrent data structures such as counters, pointers, queues, and stacks without explicit locks. High-level programming languages offer atomic types and memory-ordering controls that map to these primitives, allowing developers to implement safe, concurrent updates.

Practical use cases include implementing atomic counters, pointer updates in lock-free linked structures, reference counting, and certain forms of memory reclamation. RMW operations are also relevant in systems programming, databases, and operating systems where synchronized modification of shared state is required.

Challenges and considerations include memory ordering and visibility across cores, which may require fences or stronger memory guarantees beyond the atomic primitive itself. The ABA problem—where a value changes and later reverts, potentially misleading CAS-based logic—can undermine correctness, and is mitigated with version tags or tagged pointers. Performance concerns arise from contention and cache-coherence traffic, and in persistent memory environments, crash-consistency requires careful sequencing of writes and flushes.

See also: atomic operations, memory ordering, lock-free programming, ABA problem.