velesdb-core 3.1.0

High-performance vector database engine written in Rust
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

#![allow(
    clippy::cast_precision_loss,
    clippy::cast_possible_truncation,
    clippy::cast_sign_loss,
    clippy::float_cmp
)]
//! Tests for OnceLock warmup (EPIC-052/US-004)
//!
//! Tests that warmup_simd_cache correctly initializes SIMD caches.

use super::{cosine_similarity_native, dot_product_native, warmup_simd_cache};

// ============================================================================
// Warmup Tests
// ============================================================================

/// Test that warmup reduces latency (performance test, may be flaky in CI).
/// Run with: cargo test -- --ignored
#[test]
#[ignore = "performance test - run with --ignored or PERF_TESTS=1"]
fn test_warmup_reduces_first_request_latency() {
    // Call warmup
    warmup_simd_cache();

    // First request after warmup should be fast
    let size = 768;
    let a: Vec<f32> = (0..size).map(|i| ((i * 7) % 100) as f32 * 0.01).collect();
    let b: Vec<f32> = (0..size)
        .map(|i| (((size - i) * 13) % 100) as f32 * 0.01)
        .collect();

    // Measure first call after warmup
    let start = std::time::Instant::now();
    let _ = dot_product_native(&a, &b);
    let first_call_ns = start.elapsed().as_nanos() as f64;

    // Should be < 250ns (not cold start - cold start can be 2-3x slower)
    assert!(
        first_call_ns < 250.0,
        "First call after warmup too slow: {:.2}ns (should be < 250ns)",
        first_call_ns
    );

    // Verify correctness
    let result = dot_product_native(&a, &b);
    let expected: f32 = a.iter().zip(b.iter()).map(|(x, y)| x * y).sum();
    assert!(
        (result - expected).abs() < 1e-3,
        "Warmup affected correctness: got {}, expected {}",
        result,
        expected
    );
}

#[test]
fn test_warmup_idempotent() {
    // Multiple warmups should be safe
    warmup_simd_cache();
    warmup_simd_cache();
    warmup_simd_cache();

    // Should still work correctly
    let a: Vec<f32> = vec![1.0, 2.0, 3.0, 4.0];
    let b: Vec<f32> = vec![5.0, 6.0, 7.0, 8.0];

    let result = dot_product_native(&a, &b);
    assert!(
        (result - 70.0).abs() < 1e-6,
        "Multiple warmups broke correctness"
    );
}

#[test]
fn test_warmup_all_functions() {
    warmup_simd_cache();

    // Test dot product
    let a: Vec<f32> = (0..768).map(|i| i as f32 * 0.01).collect();
    let b: Vec<f32> = (0..768).map(|i| (767 - i) as f32 * 0.01).collect();

    let dot_result = dot_product_native(&a, &b);
    let dot_expected: f32 = a.iter().zip(b.iter()).map(|(x, y)| x * y).sum();
    // Relative tolerance: the accumulated sum is in the hundreds, so SIMD vs
    // scalar reduction order diverges by more than a tight absolute epsilon.
    assert!(
        (dot_result - dot_expected).abs() <= dot_expected.abs() * 1e-4,
        "Dot product after warmup: got {dot_result}, expected {dot_expected}"
    );

    // Test cosine
    let cos_result = cosine_similarity_native(&a, &b);
    let cos_expected = crate::simd_native::scalar::cosine_scalar(&a, &b);
    assert!(
        (cos_result - cos_expected).abs() < 1e-3,
        "Cosine after warmup: got {cos_result}, expected {cos_expected}"
    );
}