A single-header, high-performance shared mutex (reader-writer lock) implementation using C11 atomics, designed for efficient concurrent access patterns.
- Lock-free implementation: Uses atomic operations instead of traditional mutexes
- Spin-then-yield strategy: Optimal balance between performance and CPU utilization
- Cache-friendly design: Optional padding to prevent false sharing
- Architecture-specific optimizations: Uses CPU pause instructions when available
- Comprehensive API: Complete set of lock operations including timed variants
- Customizable: Configurable spinning behavior, backoff strategy, and yielding
- Debugging support: Runtime consistency validation and assertions
- Include the header in your project
- Define
SMTX_IMPLEMENTATIONin exactly one source file
Define these before including the header to customize behavior:
SMTX_NDEBUG: Disable debug assertions and checksSMTX_ASSERT(expr): Override default assert implementationSMTX_NEXT_SPINS(curr_spins): Override spin count progression strategy (default: exponential backoff)SMTX_MAX_WRITER_WAIT_SPINS: Maximum spin count when waiting for writers (default: 1024)SMTX_MAX_READER_WAIT_SPINS: Maximum spin count when waiting for readers (default: 1024)SMTX_YIELD_THRESHOLD: Spin count threshold before yielding the thread (default: 512)SMTX_YIELD: Override thread yielding mechanism (default: thrd_yield())SMTX_CLOCK_ID: Clock ID to use for timeouts (default: CLOCK_MONOTONIC)SMTX_CACHE_LINE_SIZE: Set cache line size in bytes (default: 64)SMTX_PREVENT_FALSE_SHARING: Add padding and enforce alignment to prevent false sharing
smtx_init: Initialize a shared mutex
smtx_lock_shared: Acquire a shared lock (multiple readers allowed)smtx_trylock_shared: Try to acquire a shared lock without blockingsmtx_timedlock_shared: Try to acquire a shared lock with timeoutsmtx_unlock_shared: Release a shared lock
smtx_lock_exclusive: Acquire an exclusive lock (one writer, no readers)smtx_trylock_exclusive: Try to acquire an exclusive lock without blockingsmtx_timedlock_exclusive: Try to acquire an exclusive lock with timeoutsmtx_unlock_exclusive: Release an exclusive lock
- Best performance for short-duration critical sections
- For high-contention workloads, tune spin count parameters
- Enable
SMTX_PREVENT_FALSE_SHARINGfor multi-socket systems - Use trylock variants for non-blocking operations when possible
All SMTX functions are thread-safe when used correctly:
- Each lock operation must have a corresponding unlock operation
- Shared locks are held by multiple reader threads simultaneously
- Exclusive locks are held by exactly one writer thread with no active readers
MIT License. See the LICENSE file for details.