lgpu 0.1.1

Ultra-lightweight, bindless-first cross-platform GPU abstraction
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

//! Integration tests verifying GPU memory suballocation and pipeline caching.

use lgpu::*;

#[test]
fn test_suballocator_exceeding_4096_allocations() {
    let instance = Instance::new().expect("Failed to create instance");
    let adapters = instance
        .enumerate_adapters()
        .expect("Failed to enumerate adapters");
    if adapters.is_empty() {
        return;
    }
    let (device, _) = adapters[0]
        .request_device(&instance)
        .expect("Failed to request device");

    // Allocate 5,000 small buffers to verify suballocation.
    // Without a suballocator, this would exceed the Vulkan maxMemoryAllocationCount
    // limit (typically 4096) and fail with VK_ERROR_TOO_MANY_OBJECTS.
    const BUFFER_COUNT: usize = 5000;
    let mut buffers = Vec::with_capacity(BUFFER_COUNT);

    for i in 0..BUFFER_COUNT {
        let buf = device
            .create_buffer(&BufferDesc {
                label: Some("SuballocatedBuffer"),
                size: 256,
                usage: BufferUsage::STORAGE | BufferUsage::TRANSFER_SRC | BufferUsage::TRANSFER_DST,
                memory_location: MemoryLocation::CpuToGpu,
            })
            .expect("create_buffer failed under suballocation");

        // Write unique pattern
        let pattern = [(i & 0xFF) as u8; 4];
        buf.write_at(0, &pattern).expect("write failed");
        buffers.push(buf);
    }

    assert_eq!(buffers.len(), BUFFER_COUNT);

    // Verify samples across the entire allocated range
    for i in (0..BUFFER_COUNT).step_by(250) {
        let mut read_back = [0u8; 4];
        buffers[i].read_at(0, &mut read_back).expect("read failed");
        assert_eq!(read_back, [(i & 0xFF) as u8; 4]);
    }

    // Drop all buffers to verify clean free list recycling
    drop(buffers);

    // Reallocate 1,000 buffers into the recycled space
    let mut recycled_buffers = Vec::with_capacity(1000);
    for _ in 0..1000 {
        let buf = device
            .create_buffer(&BufferDesc {
                label: Some("RecycledBuffer"),
                size: 512,
                usage: BufferUsage::STORAGE | BufferUsage::TRANSFER_SRC | BufferUsage::TRANSFER_DST,
                memory_location: MemoryLocation::CpuToGpu,
            })
            .expect("reallocation in recycled suballocator block failed");
        recycled_buffers.push(buf);
    }
    assert_eq!(recycled_buffers.len(), 1000);
}

#[test]
fn test_pipeline_cache_reuse() {
    let instance = Instance::new().expect("Failed to create instance");
    let adapters = instance
        .enumerate_adapters()
        .expect("Failed to enumerate adapters");
    if adapters.is_empty() {
        return;
    }
    let (device, queue) = adapters[0]
        .request_device(&instance)
        .expect("Failed to request device");

    pub const VECTOR_ADD_SPV: &[u32] = &[
        119734787, 65536, 28, 36, 0, 131089, 1, 720906, 1599492179, 1599227979, 1919906931,
        1600481121, 1717990754, 1935635045, 1634889588, 1667196263, 1936941420, 0, 393227, 1,
        1280527431, 1685353262, 808793134, 0, 196622, 0, 1, 393231, 5, 20, 1852399981, 0, 17,
        393232, 20, 17, 64, 1, 1, 262215, 4, 6, 4, 262215, 7, 34, 0, 262215, 7, 33, 0, 196679, 8,
        2, 327752, 8, 0, 35, 0, 262215, 10, 34, 0, 262215, 10, 33, 1, 196679, 11, 2, 327752, 11, 0,
        35, 0, 262215, 13, 34, 0, 262215, 13, 33, 2, 196679, 14, 2, 327752, 14, 0, 35, 0, 262215,
        17, 11, 28, 131091, 2, 196630, 3, 32, 196637, 4, 3, 262165, 6, 32, 0, 262167, 5, 6, 3,
        196638, 8, 4, 262176, 9, 12, 8, 262203, 9, 7, 12, 196638, 11, 4, 262176, 12, 12, 11,
        262203, 12, 10, 12, 196638, 14, 4, 262176, 15, 12, 14, 262203, 15, 13, 12, 262176, 18, 1,
        5, 262203, 18, 17, 1, 196641, 21, 2, 262176, 22, 12, 4, 262187, 6, 23, 0, 262176, 29, 12,
        3, 327734, 2, 20, 0, 21, 131320, 16, 262205, 5, 19, 17, 327745, 22, 24, 7, 23, 327745, 22,
        25, 10, 23, 327745, 22, 26, 13, 23, 131321, 27, 131320, 27, 327761, 6, 28, 19, 0, 327745,
        29, 30, 24, 28, 262205, 3, 31, 30, 327745, 29, 32, 25, 28, 262205, 3, 33, 32, 327809, 3,
        34, 31, 33, 327745, 29, 35, 26, 28, 196670, 35, 34, 65789, 65592,
    ];

    let shader = device
        .create_shader_spirv(VECTOR_ADD_SPV)
        .expect("shader creation failed");

    let buffer_desc = BufferDesc {
        label: Some("PipelineTestBuffer"),
        size: 256,
        usage: BufferUsage::STORAGE | BufferUsage::TRANSFER_SRC | BufferUsage::TRANSFER_DST,
        memory_location: MemoryLocation::CpuToGpu,
    };
    let buf_a = device
        .create_buffer(&buffer_desc)
        .expect("create buffer A failed");
    let buf_b = device
        .create_buffer(&buffer_desc)
        .expect("create buffer B failed");
    let buf_c = device
        .create_buffer(&buffer_desc)
        .expect("create buffer C failed");

    // Create 10 compute pipelines consecutively using the same shared pipeline cache
    for _ in 0..10 {
        let pipeline = device
            .create_compute_pipeline(&shader, 3)
            .expect("create_compute_pipeline failed");

        let mut encoder = device
            .create_command_encoder()
            .expect("create encoder failed");
        encoder.dispatch_compute(&pipeline, &[&buf_a, &buf_b, &buf_c], 1, 1, 1);
        encoder.memory_barrier();
        let cmd = encoder.finish();
        queue.submit(cmd).expect("submit failed");
    }
}