vaam-image-webp 0.1.0

Pure-Rust WebP codec with a lossy (VP8) encoder - no C, no libwebp bindings
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
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184

/// 16 bit fixed point version of cos(PI/8) * sqrt(2) - 1
const CONST1: i64 = 20091;
/// 16 bit fixed point version of sin(PI/8) * sqrt(2)
const CONST2: i64 = 35468;

// inverse discrete cosine transform, used in decoding
pub(crate) fn idct4x4(block: &mut [i32]) {
    // The intermediate results may overflow the types, so we stretch the type.
    fn fetch(block: &[i32], idx: usize) -> i64 {
        i64::from(block[idx])
    }

    // Perform one length check up front to avoid subsequent bounds checks in this function
    assert!(block.len() >= 16);

    for i in 0usize..4 {
        let a1 = fetch(block, i) + fetch(block, 8 + i);
        let b1 = fetch(block, i) - fetch(block, 8 + i);

        let t1 = (fetch(block, 4 + i) * CONST2) >> 16;
        let t2 = fetch(block, 12 + i) + ((fetch(block, 12 + i) * CONST1) >> 16);
        let c1 = t1 - t2;

        let t1 = fetch(block, 4 + i) + ((fetch(block, 4 + i) * CONST1) >> 16);
        let t2 = (fetch(block, 12 + i) * CONST2) >> 16;
        let d1 = t1 + t2;

        block[i] = (a1 + d1) as i32;
        block[4 + i] = (b1 + c1) as i32;
        block[4 * 3 + i] = (a1 - d1) as i32;
        block[4 * 2 + i] = (b1 - c1) as i32;
    }

    for i in 0usize..4 {
        let a1 = fetch(block, 4 * i) + fetch(block, 4 * i + 2);
        let b1 = fetch(block, 4 * i) - fetch(block, 4 * i + 2);

        let t1 = (fetch(block, 4 * i + 1) * CONST2) >> 16;
        let t2 = fetch(block, 4 * i + 3) + ((fetch(block, 4 * i + 3) * CONST1) >> 16);
        let c1 = t1 - t2;

        let t1 = fetch(block, 4 * i + 1) + ((fetch(block, 4 * i + 1) * CONST1) >> 16);
        let t2 = (fetch(block, 4 * i + 3) * CONST2) >> 16;
        let d1 = t1 + t2;

        block[4 * i] = ((a1 + d1 + 4) >> 3) as i32;
        block[4 * i + 3] = ((a1 - d1 + 4) >> 3) as i32;
        block[4 * i + 1] = ((b1 + c1 + 4) >> 3) as i32;
        block[4 * i + 2] = ((b1 - c1 + 4) >> 3) as i32;
    }
}

// 14.3 inverse walsh-hadamard transform, used in decoding
pub(crate) fn iwht4x4(block: &mut [i32]) {
    // Perform one length check up front to avoid subsequent bounds checks in this function
    assert!(block.len() >= 16);

    for i in 0usize..4 {
        let a1 = block[i] + block[12 + i];
        let b1 = block[4 + i] + block[8 + i];
        let c1 = block[4 + i] - block[8 + i];
        let d1 = block[i] - block[12 + i];

        block[i] = a1 + b1;
        block[4 + i] = c1 + d1;
        block[8 + i] = a1 - b1;
        block[12 + i] = d1 - c1;
    }

    for block in block.chunks_exact_mut(4) {
        let a1 = block[0] + block[3];
        let b1 = block[1] + block[2];
        let c1 = block[1] - block[2];
        let d1 = block[0] - block[3];

        let a2 = a1 + b1;
        let b2 = c1 + d1;
        let c2 = a1 - b1;
        let d2 = d1 - c1;

        block[0] = (a2 + 3) >> 3;
        block[1] = (b2 + 3) >> 3;
        block[2] = (c2 + 3) >> 3;
        block[3] = (d2 + 3) >> 3;
    }
}

pub(crate) fn wht4x4(block: &mut [i32; 16]) {
    // The intermediate results may overflow the types, so we stretch the type.
    fn fetch(block: &[i32], idx: usize) -> i64 {
        i64::from(block[idx])
    }

    // vertical
    for i in 0..4 {
        let a = fetch(block, i * 4) + fetch(block, i * 4 + 3);
        let b = fetch(block, i * 4 + 1) + fetch(block, i * 4 + 2);
        let c = fetch(block, i * 4 + 1) - fetch(block, i * 4 + 2);
        let d = fetch(block, i * 4) - fetch(block, i * 4 + 3);

        block[i * 4] = (a + b) as i32;
        block[i * 4 + 1] = (c + d) as i32;
        block[i * 4 + 2] = (a - b) as i32;
        block[i * 4 + 3] = (d - c) as i32;
    }

    // horizontal
    for i in 0..4 {
        let a1 = fetch(block, i) + fetch(block, i + 12);
        let b1 = fetch(block, i + 4) + fetch(block, i + 8);
        let c1 = fetch(block, i + 4) - fetch(block, i + 8);
        let d1 = fetch(block, i) - fetch(block, i + 12);

        let a2 = a1 + b1;
        let b2 = c1 + d1;
        let c2 = a1 - b1;
        let d2 = d1 - c1;

        let a3 = (a2 + if a2 > 0 { 1 } else { 0 }) / 2;
        let b3 = (b2 + if b2 > 0 { 1 } else { 0 }) / 2;
        let c3 = (c2 + if c2 > 0 { 1 } else { 0 }) / 2;
        let d3 = (d2 + if d2 > 0 { 1 } else { 0 }) / 2;

        block[i] = a3 as i32;
        block[i + 4] = b3 as i32;
        block[i + 8] = c3 as i32;
        block[i + 12] = d3 as i32;
    }
}

pub(crate) fn dct4x4(block: &mut [i32; 16]) {
    // The intermediate results may overflow the types, so we stretch the type.
    fn fetch(block: &[i32], idx: usize) -> i64 {
        i64::from(block[idx])
    }

    // vertical
    for i in 0..4 {
        let a = (fetch(block, i * 4) + fetch(block, i * 4 + 3)) * 8;
        let b = (fetch(block, i * 4 + 1) + fetch(block, i * 4 + 2)) * 8;
        let c = (fetch(block, i * 4 + 1) - fetch(block, i * 4 + 2)) * 8;
        let d = (fetch(block, i * 4) - fetch(block, i * 4 + 3)) * 8;

        block[i * 4] = (a + b) as i32;
        block[i * 4 + 2] = (a - b) as i32;
        block[i * 4 + 1] = ((c * 2217 + d * 5352 + 14500) >> 12) as i32;
        block[i * 4 + 3] = ((d * 2217 - c * 5352 + 7500) >> 12) as i32;
    }

    // horizontal
    for i in 0..4 {
        let a = fetch(block, i) + fetch(block, i + 12);
        let b = fetch(block, i + 4) + fetch(block, i + 8);
        let c = fetch(block, i + 4) - fetch(block, i + 8);
        let d = fetch(block, i) - fetch(block, i + 12);

        block[i] = ((a + b + 7) >> 4) as i32;
        block[i + 8] = ((a - b + 7) >> 4) as i32;
        block[i + 4] = (((c * 2217 + d * 5352 + 12000) >> 16) + if d != 0 { 1 } else { 0 }) as i32;
        block[i + 12] = ((d * 2217 - c * 5352 + 51000) >> 16) as i32;
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_dct_inverse() {
        const BLOCK: [i32; 16] = [
            38, 6, 210, 107, 42, 125, 185, 151, 241, 224, 125, 233, 227, 8, 57, 96,
        ];

        let mut dct_block = BLOCK.clone();

        dct4x4(&mut dct_block);

        let mut inverse_dct_block = dct_block.clone();

        idct4x4(&mut inverse_dct_block);

        assert_eq!(BLOCK, inverse_dct_block);
    }
}