tetration 0.1.6

Chunked tensor .tet format, tet CLI, and optional C ABI (tetration-ffi)
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136

//! Bulk variance accumulators vs elementwise Welford reference.

use crate::query::fold::reduction::{
    ReductionKind, ScalarReductionResult, ValueAccum, WelfordAccum,
};
use crate::query::fold::variance_simd;

fn var_from(vals: &[f32]) -> f64 {
    let mut acc = ValueAccum::default();
    acc.push_f32_le_bytes(bytemuck::cast_slice(vals), ReductionKind::Var);
    acc.finish_f64(ReductionKind::Var)
}

fn var_elementwise(vals: &[f32]) -> f64 {
    let mut w = WelfordAccum::default();
    for &v in vals {
        w.push(f64::from(v));
    }
    w.population_variance()
}

#[test]
fn f32_population_variance_large_slice_matches_welford() {
    // Bench-scale element count; f32 GPU tree reduce was wrong here, f64 SIMD must hold.
    let n = 2_000_000usize;
    let vals: Vec<f32> = (0..n).map(|i| (i % 1000) as f32 / 999.0).collect();
    let (sum, sumsq) = variance_simd::f32_sum_sumsq(&vals);
    let nf = n as f64;
    let mean = sum / nf;
    let simd_var = (sumsq / nf - mean * mean).max(0.0);
    let welford_var = var_elementwise(&vals);
    assert!(
        (simd_var - welford_var).abs() < 1e-9,
        "simd={simd_var} welford={welford_var}"
    );
    // Uniform on [0, 1] → population variance 1/12.
    assert!((simd_var - 1.0 / 12.0).abs() < 0.01, "simd={simd_var}");
}

#[test]
fn scalar_reduction_merge_partial_sum_and_mean() {
    let mut acc = ScalarReductionResult::default_fields(0);
    acc.merge_partial(
        &ScalarReductionResult {
            element_count: 3,
            sum_scalar: Some(6.0),
            ..ScalarReductionResult::default_fields(3)
        },
        ReductionKind::Sum,
    );
    acc.merge_partial(
        &ScalarReductionResult {
            element_count: 3,
            sum_scalar: Some(15.0),
            ..ScalarReductionResult::default_fields(3)
        },
        ReductionKind::Sum,
    );
    let done = acc.finalize_merged(ReductionKind::Sum);
    assert_eq!(done.element_count, 6);
    assert!((done.sum_scalar.unwrap() - 21.0).abs() < 1e-9);

    let mut mean_acc = ScalarReductionResult::default_fields(0);
    mean_acc.merge_partial(
        &ScalarReductionResult {
            element_count: 2,
            sum_scalar: Some(3.0),
            ..ScalarReductionResult::default_fields(2)
        },
        ReductionKind::Mean,
    );
    mean_acc.merge_partial(
        &ScalarReductionResult {
            element_count: 2,
            sum_scalar: Some(7.0),
            ..ScalarReductionResult::default_fields(2)
        },
        ReductionKind::Mean,
    );
    let mean_done = mean_acc.finalize_merged(ReductionKind::Mean);
    assert_eq!(mean_done.element_count, 4);
    assert!((mean_done.mean_scalar.unwrap() - 2.5).abs() < 1e-9);
}

#[test]
fn bulk_f32_var_matches_elementwise_welford() {
    let vals: Vec<f32> = (0..10_000).map(|i| (i as f32) * 0.001).collect();
    let bulk = var_from(&vals);
    let elem = var_elementwise(&vals);
    assert!((bulk - elem).abs() < 1e-6, "bulk={bulk} elem={elem}");
}

#[test]
fn bulk_f64_var_matches_elementwise_welford() {
    let vals: Vec<f64> = (0..10_000).map(|i| i as f64 * 0.001).collect();
    let mut bulk = ValueAccum::default();
    bulk.push_f64_le_bytes(bytemuck::cast_slice(&vals), ReductionKind::Var);
    let bulk_v = bulk.finish_f64(ReductionKind::Var);

    let mut w = WelfordAccum::default();
    for &v in &vals {
        w.push(v);
    }
    let elem = w.population_variance();
    assert!((bulk_v - elem).abs() < 1e-9, "bulk={bulk_v} elem={elem}");
}

#[test]
fn bulk_i32_var_matches_elementwise_welford() {
    let vals: Vec<i32> = (0..10_000).map(|i| i - 5_000).collect();
    let mut bulk = ValueAccum::default();
    bulk.push_i32_le_bytes(bytemuck::cast_slice(&vals), ReductionKind::Var);
    let bulk_v = bulk.finish_f64(ReductionKind::Var);

    let mut w = WelfordAccum::default();
    for &v in &vals {
        w.push(f64::from(v));
    }
    let elem = w.population_variance();
    assert!((bulk_v - elem).abs() < 1e-6, "bulk={bulk_v} elem={elem}");
}

#[test]
fn bulk_u8_var_matches_elementwise_welford() {
    let vals: Vec<u8> = (0..10_000).map(|i| (i % 256) as u8).collect();
    let mut bulk = ValueAccum::default();
    bulk.push_u8_le_bytes(&vals, ReductionKind::Var);
    let bulk_v = bulk.finish_f64(ReductionKind::Var);

    let mut w = WelfordAccum::default();
    for &v in &vals {
        w.push(f64::from(v));
    }
    let elem = w.population_variance();
    assert!((bulk_v - elem).abs() < 1e-6, "bulk={bulk_v} elem={elem}");
}