kernel.async.cluster.gather

← kernel dialect

Initiate a workgroup-cluster asynchronous load from a global-like source view into a workgroup/shared-memory destination view. The required i32 cluster_mask is a semantic participant set: bit N names flat cluster rank N, with x as the minor dimension. Source and destination must have the same static byte footprint, and that footprint must be exactly 1, 4, 8, or 16 bytes. Every named participant must execute the operation in the same dynamic order with corresponding lane-local source and destination addresses. Target lowering maps the participant set, addresses, and cache policy to the selected machine protocol. The returned token must be committed to exactly one kernel.async.group.

Operation contract

Property	Value
Semantic phase	—
Target contracts	kernel.async
Memory effects	read source, write dest

Signature

Kind	Name	Type	Cardinality	Description
Operand	source	view	required	Global-like source fragment broadcast across the cluster.
Operand	dest	view	required	Workgroup/LDS destination fragment for this workgroup.
Operand	cluster_mask	integer	required	i32 set of participating flat workgroup-cluster ranks.
Result	token	any	required	Opaque async-copy token for the cluster gather.
Attribute	cache_scope	enum CacheScope	required	Required cache/coherency scope for the transfer.
Attribute	cache_temporal	enum CacheTemporal	required	Required temporal cache hint for the transfer.

Examples

%copy = kernel.async.cluster.gather %src to %lds using %mask {cache_scope = se, cache_temporal = high_temporal} : view<16xi8> to view<16xi8>, i32 -> kernel.async.token