pub fn compute_priorities_simd(losses: &[f64], epsilon: f64) -> Vec<f64>Expand description
SIMD-friendly LAP priority: priority[i] = |loss[i]| + epsilon
The abs + add portion is trivially vectorizable. The subsequent
powf(alpha) is left to the caller (not SIMD-friendly).
Processes 4 f64s at a time.