atomic_exchange

Live Demo Pending

This intrinsic does not yet have a self-contained interactive preview. Compute-only and external-resource intrinsics need a different demo path than the current fragment and 3D showcases.

The atomic_exchange function atomically replaces the value at a memory location and returns the original value.

Signature

bwsl

atomic_exchange(dest, value)

Parameters

Parameter	Description
`dest`	Integer lvalue to modify atomically
`value`	New value to store

Return Value

Returns the original value at dest before the exchange.

Example

bwsl

pipeline SpinLock {
  pass "Main" {
    compute "Main" [64, 1, 1] {
    // Try to acquire lock
    int lockIndex = getTileLock(tileId);
    int expected = 0;

    // Spin until we acquire the lock
    while (atomic_exchange(lockBuffer[lockIndex], 1) != 0) {
        // Lock was held, try again
    }

    // Critical section - we have the lock
    processExclusive(tileId);

    // Release lock
    atomic_exchange(lockBuffer[lockIndex], 0);
    }
  }
}

Common Use Cases

Spin Lock

bwsl

// Acquire lock
while (atomic_exchange(lock, 1) != 0) {
    // spin
}
// ... critical section ...
atomic_exchange(lock, 0);  // Release

Swap Values

bwsl

// Atomically swap in new value
int oldValue = atomic_exchange(data, newValue);
processOldValue(oldValue);

Double Buffering

bwsl

// Swap buffer index
int oldBuffer = atomic_exchange(activeBuffer, newBuffer);
// oldBuffer is now safe to write to

State Machine

bwsl

// Atomically transition state
int previousState = atomic_exchange(state, newState);
if (previousState != expectedState) {
    // Handle unexpected state
}

One-Shot Triggers

bwsl

// First thread to arrive triggers action
int wasTriggered = atomic_exchange(trigger, 1);
if (wasTriggered == 0) {
    // This thread triggered - do one-time action
    performOneTimeAction();
}

Spin Lock Caution

Spin locks in GPU shaders can cause severe performance issues or hangs if not all threads make progress. Prefer lock-free algorithms when possible.

Compare-Exchange Alternative

For conditional updates (only exchange if current value matches expected), use atomic_cmp_exchange instead.

Compiled Output

When compiled to GLSL:

glsl

atomicExchange(dest, value)

When compiled to HLSL:

hlsl

InterlockedExchange(dest, value, originalValue)

When compiled to SPIR-V:

Uses OpAtomicExchange instruction.

Test atomic_exchange in a live shader

Signature

Parameters

Return Value

Example

Common Use Cases

Spin Lock

Swap Values

Double Buffering

State Machine

One-Shot Triggers

Compiled Output

See Also