vector.fmaf
← vector dialect
Lanewise fused multiply-add of same-typed floating-point vectors. Each result lane computes a*b + c with one final rounding; use separate vector.mulf/vector.addf when unfused rounding is required. AFN permits target-native fused forms with weaker subnormal handling.
Operation contract
| Property |
Value |
| Semantic phase |
executable |
| Traits |
Pure, Elementwise, SafeToSpeculate |
Signature
| Kind |
Name |
Type |
Cardinality |
Description |
| Operand |
a |
vector |
required |
— |
| Operand |
b |
vector |
required |
— |
| Operand |
c |
vector |
required |
— |
| Result |
result |
vector |
required |
— |
| Attribute |
fastmath |
flags FastMathFlags |
optional |
— |
Verification constraints
HasFloatElement(result)SameType(a, b, c, result)
Examples
%r = vector.fmaf %a, %b, %c : vector<16xf32>