Expand description
Kahan compensated summation — bit-identical results regardless of platform.
Every floating-point reduction in this library uses Kahan summation to guarantee deterministic results. The compensation term captures rounding error that naive left-to-right summation would lose.
Structs§
- Kahan
Accumulator - Kahan compensated accumulator for f64 values.
Functions§
- kahan_
sum - One-shot Kahan summation for a slice of f64 values.