Expand description
Context scheduling flags passed to Context::with_flags.
These control how the CPU thread behaves while waiting for GPU operations.
Constantsยง
- LMEM_
RESIZE_ TO_ MAX - Keep local memory allocation after launch (deprecated flag kept for completeness).
- MAP_
HOST - Enable mapped pinned allocations in this context.
- SCHED_
AUTO - Let the driver choose the optimal scheduling policy.
- SCHED_
BLOCKING_ SYNC - Block the calling thread on a synchronisation primitive. Lowest CPU usage but slightly higher latency.
- SCHED_
SPIN - Actively spin (busy-wait) while waiting for GPU results. Lowest latency but consumes a full CPU core.
- SCHED_
YIELD - Yield the CPU time-slice to other threads while waiting. Good for multi-threaded applications.