oximedia-gpu 0.1.1

GPU compute pipeline using WGPU for OxiMedia - cross-platform acceleration
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
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156

// Color space conversion shaders for RGB <-> YUV transformations
// Using BT.601 coefficients for standard definition video

// RGB to YUV conversion kernel
@group(0) @binding(0) var<storage, read> input_rgb: array<u32>;
@group(0) @binding(1) var<storage, read_write> output_yuv: array<u32>;
@group(0) @binding(2) var<uniform> params: ConversionParams;

struct ConversionParams {
    width: u32,
    height: u32,
    stride: u32,
    format: u32, // 0=BT.601, 1=BT.709, 2=BT.2020
}

// BT.601 coefficients
const BT601_KR: f32 = 0.299;
const BT601_KB: f32 = 0.114;
const BT601_KG: f32 = 0.587;

// BT.709 coefficients
const BT709_KR: f32 = 0.2126;
const BT709_KB: f32 = 0.0722;
const BT709_KG: f32 = 0.7152;

// BT.2020 coefficients
const BT2020_KR: f32 = 0.2627;
const BT2020_KB: f32 = 0.0593;
const BT2020_KG: f32 = 0.6780;

fn unpack_rgba(packed: u32) -> vec4<f32> {
    let r = f32((packed >> 24u) & 0xFFu) / 255.0;
    let g = f32((packed >> 16u) & 0xFFu) / 255.0;
    let b = f32((packed >> 8u) & 0xFFu) / 255.0;
    let a = f32(packed & 0xFFu) / 255.0;
    return vec4<f32>(r, g, b, a);
}

fn pack_yuva(y: f32, u: f32, v: f32, a: f32) -> u32 {
    let yi = u32(clamp(y * 255.0, 0.0, 255.0));
    let ui = u32(clamp(u * 255.0, 0.0, 255.0));
    let vi = u32(clamp(v * 255.0, 0.0, 255.0));
    let ai = u32(clamp(a * 255.0, 0.0, 255.0));
    return (yi << 24u) | (ui << 16u) | (vi << 8u) | ai;
}

fn rgb_to_yuv_bt601(rgb: vec3<f32>) -> vec3<f32> {
    let y = BT601_KR * rgb.r + BT601_KG * rgb.g + BT601_KB * rgb.b;
    let u = (rgb.b - y) / (2.0 * (1.0 - BT601_KB)) + 0.5;
    let v = (rgb.r - y) / (2.0 * (1.0 - BT601_KR)) + 0.5;
    return vec3<f32>(y, u, v);
}

fn rgb_to_yuv_bt709(rgb: vec3<f32>) -> vec3<f32> {
    let y = BT709_KR * rgb.r + BT709_KG * rgb.g + BT709_KB * rgb.b;
    let u = (rgb.b - y) / (2.0 * (1.0 - BT709_KB)) + 0.5;
    let v = (rgb.r - y) / (2.0 * (1.0 - BT709_KR)) + 0.5;
    return vec3<f32>(y, u, v);
}

fn rgb_to_yuv_bt2020(rgb: vec3<f32>) -> vec3<f32> {
    let y = BT2020_KR * rgb.r + BT2020_KG * rgb.g + BT2020_KB * rgb.b;
    let u = (rgb.b - y) / (2.0 * (1.0 - BT2020_KB)) + 0.5;
    let v = (rgb.r - y) / (2.0 * (1.0 - BT2020_KR)) + 0.5;
    return vec3<f32>(y, u, v);
}

@compute @workgroup_size(16, 16, 1)
fn rgb_to_yuv_main(@builtin(global_invocation_id) global_id: vec3<u32>) {
    let x = global_id.x;
    let y = global_id.y;

    if (x >= params.width || y >= params.height) {
        return;
    }

    let idx = y * params.stride + x;
    let rgba = unpack_rgba(input_rgb[idx]);

    var yuv: vec3<f32>;
    if (params.format == 0u) {
        yuv = rgb_to_yuv_bt601(rgba.rgb);
    } else if (params.format == 1u) {
        yuv = rgb_to_yuv_bt709(rgba.rgb);
    } else {
        yuv = rgb_to_yuv_bt2020(rgba.rgb);
    }

    output_yuv[idx] = pack_yuva(yuv.x, yuv.y, yuv.z, rgba.a);
}

// YUV to RGB conversion kernel
fn yuv_to_rgb_bt601(yuv: vec3<f32>) -> vec3<f32> {
    let y = yuv.x;
    let u = yuv.y - 0.5;
    let v = yuv.z - 0.5;

    let r = y + 2.0 * (1.0 - BT601_KR) * v;
    let b = y + 2.0 * (1.0 - BT601_KB) * u;
    let g = (y - BT601_KR * r - BT601_KB * b) / BT601_KG;

    return vec3<f32>(r, g, b);
}

fn yuv_to_rgb_bt709(yuv: vec3<f32>) -> vec3<f32> {
    let y = yuv.x;
    let u = yuv.y - 0.5;
    let v = yuv.z - 0.5;

    let r = y + 2.0 * (1.0 - BT709_KR) * v;
    let b = y + 2.0 * (1.0 - BT709_KB) * u;
    let g = (y - BT709_KR * r - BT709_KB * b) / BT709_KG;

    return vec3<f32>(r, g, b);
}

fn yuv_to_rgb_bt2020(yuv: vec3<f32>) -> vec3<f32> {
    let y = yuv.x;
    let u = yuv.y - 0.5;
    let v = yuv.z - 0.5;

    let r = y + 2.0 * (1.0 - BT2020_KR) * v;
    let b = y + 2.0 * (1.0 - BT2020_KB) * u;
    let g = (y - BT2020_KR * r - BT2020_KB * b) / BT2020_KG;

    return vec3<f32>(r, g, b);
}

@compute @workgroup_size(16, 16, 1)
fn yuv_to_rgb_main(@builtin(global_invocation_id) global_id: vec3<u32>) {
    let x = global_id.x;
    let y = global_id.y;

    if (x >= params.width || y >= params.height) {
        return;
    }

    let idx = y * params.stride + x;
    let yuva = unpack_rgba(input_rgb[idx]); // Reusing input buffer name

    var rgb: vec3<f32>;
    if (params.format == 0u) {
        rgb = yuv_to_rgb_bt601(yuva.rgb);
    } else if (params.format == 1u) {
        rgb = yuv_to_rgb_bt709(yuva.rgb);
    } else {
        rgb = yuv_to_rgb_bt2020(yuva.rgb);
    }

    let r = u32(clamp(rgb.r * 255.0, 0.0, 255.0));
    let g = u32(clamp(rgb.g * 255.0, 0.0, 255.0));
    let b = u32(clamp(rgb.b * 255.0, 0.0, 255.0));
    let a = u32(clamp(yuva.a * 255.0, 0.0, 255.0));

    output_yuv[idx] = (r << 24u) | (g << 16u) | (b << 8u) | a;
}