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

use blade_graphics as gpu;
use std::mem;

struct ReusableBuffer {
    raw: gpu::Buffer,
    size: u64,
}

/// Configuration of the Blade belt.
pub struct BufferBeltDescriptor {
    /// Kind of memory to allocate from.
    pub memory: gpu::Memory,
    pub min_chunk_size: u64,
    pub alignment: u64,
}

/// A belt of reusable buffer space.
/// Could be useful for temporary data, such as texture staging areas.
pub struct BufferBelt {
    desc: BufferBeltDescriptor,
    buffers: Vec<(ReusableBuffer, gpu::SyncPoint)>,
    active: Vec<(ReusableBuffer, u64)>,
}

impl BufferBelt {
    /// Create a new belt.
    pub fn new(desc: BufferBeltDescriptor) -> Self {
        assert_ne!(desc.alignment, 0);
        Self {
            desc,
            buffers: Vec::new(),
            active: Vec::new(),
        }
    }

    /// Destroy this belt.
    pub fn destroy(&mut self, gpu: &gpu::Context) {
        for (buffer, _) in self.buffers.drain(..) {
            gpu.destroy_buffer(buffer.raw);
        }
        for (buffer, _) in self.active.drain(..) {
            gpu.destroy_buffer(buffer.raw);
        }
    }

    /// Allocate a region of `size` bytes.
    #[profiling::function]
    pub fn alloc(&mut self, size: u64, gpu: &gpu::Context) -> gpu::BufferPiece {
        for &mut (ref rb, ref mut offset) in self.active.iter_mut() {
            let aligned = offset.next_multiple_of(self.desc.alignment);
            if aligned + size <= rb.size {
                let piece = rb.raw.at(aligned);
                *offset = aligned + size;
                return piece;
            }
        }

        let index_maybe = self
            .buffers
            .iter()
            .position(|(rb, sp)| size <= rb.size && gpu.wait_for(sp, 0));
        if let Some(index) = index_maybe {
            let (rb, _) = self.buffers.remove(index);
            let piece = rb.raw.into();
            self.active.push((rb, size));
            return piece;
        }

        let chunk_index = self.buffers.len() + self.active.len();
        let chunk_size = size.max(self.desc.min_chunk_size);
        let chunk = gpu.create_buffer(gpu::BufferDesc {
            name: &format!("chunk-{}", chunk_index),
            size: chunk_size,
            memory: self.desc.memory,
        });
        let rb = ReusableBuffer {
            raw: chunk,
            size: chunk_size,
        };
        self.active.push((rb, size));
        chunk.into()
    }

    /// Allocate a region to hold the byte `data` slice contents.
    pub fn alloc_bytes(&mut self, data: &[u8], gpu: &gpu::Context) -> gpu::BufferPiece {
        assert!(!data.is_empty());
        let bp = self.alloc(data.len() as u64, gpu);
        unsafe {
            std::ptr::copy_nonoverlapping(data.as_ptr(), bp.data(), data.len());
        }
        bp
    }

    // SAFETY: T should be zeroable and ordinary data, no references, pointers, cells or other complicated data type.
    /// Allocate a region to hold the typed `data` slice contents.
    pub unsafe fn alloc_typed<T>(&mut self, data: &[T], gpu: &gpu::Context) -> gpu::BufferPiece {
        assert!(!data.is_empty());
        let type_alignment = mem::align_of::<T>() as u64;
        debug_assert_eq!(
            self.desc.alignment % type_alignment,
            0,
            "Type alignment {} is too big",
            type_alignment
        );
        let total_bytes = std::mem::size_of_val(data);
        let bp = self.alloc(total_bytes as u64, gpu);
        unsafe {
            std::ptr::copy_nonoverlapping(data.as_ptr() as *const u8, bp.data(), total_bytes);
        }
        bp
    }

    /// Allocate a region to hold the POD `data` slice contents.
    pub fn alloc_pod<T: bytemuck::Pod>(
        &mut self,
        data: &[T],
        gpu: &gpu::Context,
    ) -> gpu::BufferPiece {
        unsafe { self.alloc_typed(data, gpu) }
    }

    /// Mark the actively used buffers as used by GPU with a given sync point.
    pub fn flush(&mut self, sp: &gpu::SyncPoint) {
        self.buffers
            .extend(self.active.drain(..).map(|(rb, _)| (rb, sp.clone())));
    }
}