vector.dotf

← vector dialect

Compute a same-element floating-point dot product with an explicit scalar accumulator. Semantics are equivalent to accumulating scalar.fmaf(lhs_lane, rhs_lane, acc) over lanes in logical lane order; use vector.mulf followed by vector.reduce when separately rounded products and additions are required. The source vectors must have the same shape and element type, and the init/result scalar type matches that element type. Zero-lane inputs return init. Optional fastmath flags carry the same floating-point permissions as scalar arithmetic, including reassociation of the fused dot terms.

Operation contract

Property	Value
Semantic phase	—
Traits	Pure
Target contracts	vector.contraction

Signature

Kind	Name	Type	Cardinality	Description
Operand	lhs	vector	required	Floating-point source lanes.
Operand	rhs	vector	required	Floating-point source lanes.
Operand	init	scalar	required	Scalar accumulator seed.
Result	result	scalar	required	Scalar dot-product accumulator result.
Attribute	fastmath	flags FastMathFlags	optional	—

Verification constraints

• HasFloatElement(lhs)
• SameShape(lhs, rhs)
• SameType(init, result)
• SameElementType(lhs, rhs, init)

Examples

%r = vector.dotf %lhs, %rhs, %acc : vector<16xf32>, vector<16xf32>, f32

%r = vector.dotf<reassoc|contract> %lhs, %rhs, %acc : vector<16xf32>, vector<16xf32>, f32