Mojo package
GPU synchronization primitives package.
This package provides GPU synchronization operations including:
- barrier: Block-level synchronization barriers
- syncwarp: Warp-level synchronization
- mbarrier: Memory barrier operations (arrive/wait)
- named_barrier: Named barriers for flexible synchronization
- schedule_barrier: AMD instruction scheduling barriers
- Semaphore: Device-wide semaphore implementation
- cp_async_bulk: Bulk async copy synchronization
These primitives enable coordination of execution and memory operations across threads, warps, and blocks in GPU kernels.
Modules
-
semaphore: This module provides a device-wide semaphore implementation for NVIDIA GPUs. -
sync: This module provides GPU synchronization primitives and barriers.