xlog-cuda 0.9.2

CUDA kernel provider, buffers, and interop for XLOG
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
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154

// crates/xlog-cuda/tests/sort_tests.rs
mod common;
use common::setup_provider;
use xlog_core::{ScalarType, Schema};

#[test]
fn test_sort_single_column() {
    let Some(provider) = setup_provider() else {
        eprintln!("Skipping: no CUDA device");
        return;
    };

    // Unsorted: [5, 2, 8, 1, 9, 3]
    // Sorted:   [1, 2, 3, 5, 8, 9]
    let keys: Vec<u32> = vec![5, 2, 8, 1, 9, 3];
    let schema = Schema::new(vec![("key".to_string(), ScalarType::U32)]);

    let buffer = provider
        .create_buffer_from_slice::<u32>(&keys, schema.clone())
        .unwrap();
    let sorted = provider.sort(&buffer, &[0]).unwrap();

    let result = provider.download_column::<u32>(&sorted, 0).unwrap();
    assert_eq!(result, vec![1, 2, 3, 5, 8, 9]);
}

#[test]
fn test_sort_preserves_other_columns() {
    let Some(provider) = setup_provider() else {
        eprintln!("Skipping: no CUDA device");
        return;
    };

    // Sort by col0, col1 should follow
    // col0: [3, 1, 2] -> [1, 2, 3]
    // col1: [30, 10, 20] -> [10, 20, 30]
    let col0: Vec<u32> = vec![3, 1, 2];
    let col1: Vec<u32> = vec![30, 10, 20];
    let schema = Schema::new(vec![
        ("key".to_string(), ScalarType::U32),
        ("value".to_string(), ScalarType::U32),
    ]);

    let buffer = provider
        .create_buffer_from_u32_columns(&[&col0, &col1], schema)
        .unwrap();
    let sorted = provider.sort(&buffer, &[0]).unwrap();

    let result0 = provider.download_column::<u32>(&sorted, 0).unwrap();
    let result1 = provider.download_column::<u32>(&sorted, 1).unwrap();

    assert_eq!(result0, vec![1, 2, 3]);
    assert_eq!(result1, vec![10, 20, 30]);
}

#[test]
fn test_sort_empty() {
    let Some(provider) = setup_provider() else {
        eprintln!("Skipping: no CUDA device");
        return;
    };

    let schema = Schema::new(vec![("key".to_string(), ScalarType::U32)]);
    let buffer = provider.create_empty_buffer(schema.clone()).unwrap();
    let sorted = provider.sort(&buffer, &[0]).unwrap();

    assert_eq!(sorted.num_rows(), 0);
}

#[test]
fn test_sort_already_sorted() {
    let Some(provider) = setup_provider() else {
        eprintln!("Skipping: no CUDA device");
        return;
    };

    let keys: Vec<u32> = vec![1, 2, 3, 4, 5];
    let schema = Schema::new(vec![("key".to_string(), ScalarType::U32)]);

    let buffer = provider
        .create_buffer_from_slice::<u32>(&keys, schema)
        .unwrap();
    let sorted = provider.sort(&buffer, &[0]).unwrap();

    let result = provider.download_column::<u32>(&sorted, 0).unwrap();
    assert_eq!(result, vec![1, 2, 3, 4, 5]);
}

#[test]
fn test_sort_duplicates() {
    let Some(provider) = setup_provider() else {
        eprintln!("Skipping: no CUDA device");
        return;
    };

    // Input with duplicates: [3, 1, 2, 1, 3, 2]
    // Sorted: [1, 1, 2, 2, 3, 3]
    let keys: Vec<u32> = vec![3, 1, 2, 1, 3, 2];
    let schema = Schema::new(vec![("key".to_string(), ScalarType::U32)]);

    let buffer = provider
        .create_buffer_from_slice::<u32>(&keys, schema)
        .unwrap();
    let sorted = provider.sort(&buffer, &[0]).unwrap();

    let result = provider.download_column::<u32>(&sorted, 0).unwrap();
    assert_eq!(result, vec![1, 1, 2, 2, 3, 3]);
}

#[test]
fn test_sort_respects_device_row_count_after_filter() {
    let Some(provider) = setup_provider() else {
        eprintln!("Skipping: no CUDA device");
        return;
    };

    let schema = Schema::new(vec![("key".to_string(), ScalarType::U32)]);
    let keys: Vec<u32> = vec![3, 1, 2];
    let buffer = provider
        .create_buffer_from_slice::<u32>(&keys, schema)
        .unwrap();

    let mask: Vec<u8> = vec![0, 1, 1];
    let d_mask = provider.device().inner().htod_sync_copy(&mask).unwrap();
    let filtered = provider.filter_by_device_mask(&buffer, &d_mask).unwrap();

    let filtered_vals = provider.download_column::<u32>(&filtered, 0).unwrap();
    assert_eq!(filtered_vals, vec![1, 2]);

    let sorted = provider.sort(&filtered, &[0]).unwrap();
    let sorted_vals = provider.download_column::<u32>(&sorted, 0).unwrap();
    assert_eq!(sorted_vals, vec![1, 2]);
}

#[test]
fn test_sort_reverse() {
    let Some(provider) = setup_provider() else {
        eprintln!("Skipping: no CUDA device");
        return;
    };

    // Reverse sorted: [5, 4, 3, 2, 1]
    // Sorted: [1, 2, 3, 4, 5]
    let keys: Vec<u32> = vec![5, 4, 3, 2, 1];
    let schema = Schema::new(vec![("key".to_string(), ScalarType::U32)]);

    let buffer = provider
        .create_buffer_from_slice::<u32>(&keys, schema)
        .unwrap();
    let sorted = provider.sort(&buffer, &[0]).unwrap();

    let result = provider.download_column::<u32>(&sorted, 0).unwrap();
    assert_eq!(result, vec![1, 2, 3, 4, 5]);
}