kernel.workgroup.scan

← kernel dialect

Prefix-scan a scalar or rank-1 vector value across the current workgroup. Workitems are ordered by x + size_x * (y + size_y * z), independently of their physical subgroup or lane assignment. Forward scans use increasing order; reverse scans use decreasing order. Vector components scan independently. An exclusive scan's empty prefix has the combining kind's empty value, such as zero for addition or one for multiplication; that value is not an additional contribution to nonempty prefixes. Each active contribution participates once per component; association is unspecified. Floating sums and products default to native target arithmetic, which may change signed zeros, flush subnormals, or relax nonfinite handling. The optional preserve requirement instead evaluates each combine with the source format's arithmetic semantics. Neither mode promises serial-reference bits, reproducibility, or a correctly rounded real-number sum. Integer arithmetic and the selected floating-extremum kind keep their exact semantics.

Operation contract

Property	Value
Semantic phase	executable
Traits	Convergent, HasAnyAncestor(kernel.def, pipeline.strand)
Target contracts	kernel.synchronization

Signature

Kind	Name	Type	Cardinality	Description
Operand	value	any	required	Per-invocation value to scan.
Result	result	any	required	Scanned value.
Attribute	kind	enum CombiningKind	required	Combining operation.
Attribute	math	flags CollectiveMathFlags	optional	Optional floating-point preservation requirement; omission selects native arithmetic.
Attribute	mode	enum KernelWorkgroupScanMode	required	Inclusive or exclusive scan.
Attribute	direction	enum KernelWorkgroupScanDirection	required	Workitem order to scan.

Verification constraints

• SameType(value, result)

Examples

%prefix = kernel.workgroup.scan<addf> %v {direction = forward, mode = inclusive} : f32

%offset = kernel.workgroup.scan<addi> %count {direction = forward, mode = exclusive} : i32