Synchronization

The synchronization part of IR APIs.

tvm.aipu.script.ir.synchronization.barrier()

All work-items in a work-group executing the kernel on a processor must execute this function before any are allowed to continue execution beyond the barrier. The barrier function will queue a memory fence to ensure correct ordering of memory operations to private memory and wait for events to finish between TECs.

Examples

S.barrier()

See Also

Zhouyi Compass OpenCL Programming Guide: barrier

tvm.aipu.script.ir.synchronization.alloc_events(count)

Allocates the specified number of events. The total number of events that the whole DSL program can allocate up to the concrete Zhouyi NPU target.

Parameters

countint: The number of events you want to allocate for this invocation.

Returns

retUnion[PrimExpr, Tuple[PrimExpr]]: The events that already allocated.

Examples

ev0, ev1, ev2, ev3 = S.alloc_events(4)

See Also

How to Use DMA

tvm.aipu.script.ir.synchronization.wait_events(*events)

Blocks the current DSL program until all specified events are occurred.

Parameters

eventsTuple[PrimExpr]: The events need to be waited.

Examples

S.wait_events(ev0)
S.wait_events(ev0, ev1, ev2)

See Also

How to Use DMA

tvm.aipu.script.ir.synchronization.free_events(*events)

Release the allocated events, will call S.wait_events first automatically.

Parameters

eventsTuple[PrimExpr]: The events need to be released.

Examples

S.free_events(ev0)
S.free_events(ev0, ev1, ev2)

See Also

How to Use DMA