Module statistics 

SIMD-accelerated statistical operations

This module provides high-performance statistical computations on raster data using architecture-specific SIMD intrinsics for horizontal reductions and aggregations.

Architecture Support

• aarch64: NEON (128-bit) for parallel accumulation and comparison
• x86-64: SSE2 (baseline), AVX2 (runtime detected) for wider operations
• Other: Scalar fallback with auto-vectorization hints

Supported Operations

• Reductions: sum, mean, variance, standard deviation
• Extrema: min, max, argmin, argmax, minmax (single-pass)
• Percentiles: median, quartiles, arbitrary percentiles
• Histograms: Fast histogram computation with SIMD bucketing

Performance

Expected speedup over scalar: 4-8x for most operations

Example

use oxigdal_algorithms::simd::statistics::{sum_f32, mean_f32, minmax_f32};

let data = vec![1.0, 2.0, 3.0, 4.0, 5.0];

let sum = sum_f32(&data);
let mean = mean_f32(&data)?;
let (min, max) = minmax_f32(&data)?;

assert_eq!(sum, 15.0);
assert_eq!(mean, 3.0);
assert_eq!(min, 1.0);
assert_eq!(max, 5.0);

Functions

argmax_f32

Find the index of the maximum value

argmin_f32

Find the index of the minimum value

covariance_f32

Compute the covariance between two slices

histogram_auto_f32

Compute histogram with automatic range detection

histogram_f32

Compute histogram with specified number of bins

max_f32

Find the maximum value in the slice using SIMD comparison

mean_f32

Compute the mean (average) of all elements

mean_f64

Compute the mean (average) of all elements (f64 version)

min_f32

Find the minimum value in the slice using SIMD comparison

minmax_f32

Find both minimum and maximum values in a single pass using SIMD

std_dev_f32

Compute standard deviation using SIMD-accelerated variance

sum_f32

Compute the sum of all elements using SIMD horizontal reduction

sum_f64

Compute the sum of all elements using SIMD (f64 version)

variance_f32

Compute variance using two-pass algorithm with SIMD acceleration

welford_variance_f32

Compute Welford’s online variance (single-pass, numerically stable)