imgproc-rs 0.3.0

Image processing library for Rust
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
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271

//! A module for SIMD operations

use crate::image::{BaseImage, Image};

#[cfg(all(feature = "simd", target_arch = "x86_64"))]
use std::arch::x86_64::*;

#[cfg(all(feature = "simd", target_arch = "x86"))]
use std::arch::x86::*;

/// Adds `val` to each 8-bit channel of `input` using saturation, ignoring the alpha channel
/// if present
#[cfg(all(feature = "simd", any(target_arch = "x86", target_arch = "x86_64")))]
#[target_feature(enable = "avx2")]
pub unsafe fn adds_256_u8(input: &Image<u8>, val: i16) -> Image<u8> {
    let num_bytes = input.info().full_size();
    let mut data: Vec<u8> = vec![0; num_bytes as usize];

    // Load val into 256-bit SIMD register
    let mut val_256 = _mm256_set1_epi8(val.abs() as i8);

    // Mask the alpha bit if present
    if input.info().alpha {
        let zeroes_256 = _mm256_setzero_si256();
        let alpha_mask = _mm256_set1_epi32(-2 * 2_i32.pow(30));
        val_256 = _mm256_blendv_epi8(val_256, zeroes_256, alpha_mask);
    }

    // Process 32 bytes at a time
    let mut i = 0;
    while (i + 32) <= num_bytes {
        // Load 32-byte chunk into 256-bit SIMD register
        let chunk = _mm256_loadu_si256(input.data().as_ptr().
            offset(i as isize) as *const _);

        // Add or subtract val from each 8-bit integer with saturation
        let res = if val > 0 {
            _mm256_adds_epu8(chunk, val_256)
        } else {
            _mm256_subs_epu8(chunk, val_256)
        };

        // Store 32 bytes in output vector
        _mm256_storeu_si256(data.as_mut_ptr().offset(i as isize) as *mut _, res);

        i += 32;
    }

    // Process the remaining bytes normally
    if i > num_bytes {
        for j in (i - 32)..num_bytes {
            data[j as usize] = (input.data()[j as usize] as i16 + val).clamp(0, 255) as u8;
        }
    }

    Image::from_vec(input.info().width, input.info().height, input.info().channels,
                    input.info().alpha, data)
}

/// Adds `val` to every `n`th 8-bit channel of `input` using saturation.
/// If `n` is an invalid channel number, adds `val` to all channels.
#[cfg(all(feature = "simd", any(target_arch = "x86", target_arch = "x86_64")))]
#[target_feature(enable = "avx2")]
pub unsafe fn adds_n_256_u8(input: &Image<u8>, val: i16, n: u8) -> Image<u8> {
    let num_bytes = input.info().full_size();
    let mut data: Vec<u8> = vec![0; num_bytes as usize];

    let mut val_256 = _mm256_set1_epi8(val.abs() as i8);
    let zeroes_256 = _mm256_setzero_si256();

    // Create masks for the channels we want to ignore
    let mask = if input.info().alpha {
        match n {
            0 => _mm256_set1_epi32(0x80),
            1 => _mm256_set1_epi32(0x8000),
            2 => _mm256_set1_epi32(0x800000),
            3 => _mm256_set1_epi32(-0x80000000),
            _ => _mm256_setzero_si256()
        }
    } else {
        match n {
            0 => _mm256_set_epi64x(0x80000080, 0x800000800000,
                                   -0x7FFFFF7FFFFF8000, 0x80000080000080),
            1 => _mm256_set_epi64x(0x8000008000, 0x80000080000080,
                                   0x800000800000, -0x7FFFFF7FFFFF8000),
            2 => _mm256_set_epi64x(0x800000800000, -0x7FFFFF7FFFFF8000,
                                   0x80000080000080, 0x800000800000),
            _ => _mm256_setzero_si256()
        }
    };
    val_256 = _mm256_blendv_epi8(zeroes_256, val_256, mask);

    // In order to reuse the mask for every chunk in a 3-channel pixel, we can only process chunks
    // that are multiples of 8 * 3 = 24 bits. The largest such chunk that fits in 256 bits is
    // 24 * 10 = 240 bits = 30 bytes.
    let mut step = 30;

    // 4-channel pixels fit evenly within 32-byte chunks
    if input.info().alpha {
        step = 32;
    }

    let mut i = 0;
    while (i + step) <= num_bytes {
        let chunk = _mm256_loadu_si256(input.data().as_ptr().
            offset(i as isize) as *const _);

        let res = if val > 0 {
            _mm256_adds_epu8(chunk, val_256)
        } else {
            _mm256_subs_epu8(chunk, val_256)
        };

        _mm256_storeu_si256(data.as_mut_ptr().offset(i as isize) as *mut _, res);

        i += step;
    }

    if i > num_bytes {
        for j in ((i - step)..num_bytes).step_by(input.info().channels as usize) {
            let index = j as usize + n as usize;
            data[index] = (input.data()[index] as i16 + val).clamp(0, 255) as u8;
        }
    }

    Image::from_vec(input.info().width, input.info().height, input.info().channels,
                       input.info().alpha, data)
}

/// Separates a 3-channel input image into 3 256-bit wide integer vectors, starting at the channel
/// denoted by `offset`. Does not check if `offset` is valid.
#[cfg(all(feature = "simd", any(target_arch = "x86", target_arch = "x86_64")))]
#[target_feature(enable = "avx2")]
pub unsafe fn deinterleave_3_256_u8(input: &Image<u8>, offset: usize) -> (__m256i, __m256i, __m256i) {
    let step = input.info().channels as usize;
    let range = 32 * step;
    let mut r = Vec::with_capacity(256);
    let mut g = Vec::with_capacity(256);
    let mut b = Vec::with_capacity(256);

    // Separate channels into different vectors
    for i in (offset..(offset + range)).step_by(step) {
        r.push(input.data()[i]);
        g.push(input.data()[i + 1]);
        b.push(input.data()[i + 2]);
    }

    // Load each vector into a 256-bit SIMD register
    let r_256 = _mm256_loadu_si256(r.as_mut_ptr() as *mut _);
    let g_256 = _mm256_loadu_si256(g.as_mut_ptr() as *mut _);
    let b_256 = _mm256_loadu_si256(b.as_mut_ptr() as *mut _);

    (r_256, g_256, b_256)
}

/// Separates a 4-channel input image into 4 256-bit wide integer vectors, starting at the channel
/// denoted by `offset`. Does not check if `offset` is valid.
#[cfg(all(feature = "simd", any(target_arch = "x86", target_arch = "x86_64")))]
#[target_feature(enable = "avx2")]
pub unsafe fn deinterleave_4_256_u8(input: &Image<u8>, offset: usize) -> (__m256i, __m256i, __m256i, __m256i) {
    let step = input.info().channels as usize;
    let range = 32 * step;
    let mut r = Vec::with_capacity(256);
    let mut g = Vec::with_capacity(256);
    let mut b = Vec::with_capacity(256);
    let mut a = Vec::with_capacity(256);

    for i in (offset..(offset + range)).step_by(step) {
        r.push(input.data()[i]);
        g.push(input.data()[i + 1]);
        b.push(input.data()[i + 2]);
        a.push(input.data()[i + 3]);
    }

    let r_256 = _mm256_loadu_si256(r.as_mut_ptr() as *mut _);
    let g_256 = _mm256_loadu_si256(g.as_mut_ptr() as *mut _);
    let b_256 = _mm256_loadu_si256(b.as_mut_ptr() as *mut _);
    let a_256 = _mm256_loadu_si256(a.as_mut_ptr() as *mut _);

    (r_256, g_256, b_256, a_256)
}

/// Computes the average of each 3-channel pixel of `input` and returns a grayscale image
#[cfg(all(feature = "simd", any(target_arch = "x86", target_arch = "x86_64")))]
#[target_feature(enable = "avx2")]
pub unsafe fn avg_256_u8(input: &Image<u8>) -> Image<u8> {
    let channels = input.info().channels as usize;
    let chunk_size = 32 * channels;
    let num_bytes = input.info().full_size() as usize;
    let mut data: Vec<u8> = vec![0; (num_bytes / 3) as usize];

    let mut i: usize= 0;
    while (i + chunk_size) <= num_bytes {
        let (r, g, b) = deinterleave_3_256_u8(input, i);

        // Approximate average
        let avg_rg = _mm256_avg_epu8(r, g);
        let avg_rgb = _mm256_avg_epu8(avg_rg, b);

        _mm256_storeu_si256(data.as_mut_ptr().offset((i / channels) as isize)
                                as *mut _, avg_rgb);

        i += chunk_size;
    }

    if i > num_bytes {
        for j in ((i - chunk_size)..num_bytes).step_by(3) {
            let sum = (input.data()[i] + input.data()[i + 1] + input.data()[i + 2]) as f32;
            data[j / channels] = (sum / 3.0).round() as u8;
        }
    }

    Image::from_vec(input.info().width, input.info().height, 1,
                       input.info().alpha, data)
}

/// Computes the average of each 4-channel pixel of `input`, ignoring the alpha channel, and
/// returns a grayscale image
#[cfg(all(feature = "simd", any(target_arch = "x86", target_arch = "x86_64")))]
#[target_feature(enable = "avx2")]
pub unsafe fn avg_alpha_256_u8(input: &Image<u8>) -> Image<u8> {
    let channels = input.info().channels as usize;
    let chunk_size = 32 * channels;
    let num_bytes = input.info().full_size() as usize;
    let mut data: Vec<u8> = vec![0; (num_bytes / 2) as usize];

    let mut i: usize= 0;
    while (i + chunk_size) <= num_bytes {
        let (r, g, b, a) = deinterleave_4_256_u8(input, i);

        // Approximate average
        let avg_rg = _mm256_avg_epu8(r, g);
        let avg_rgb = _mm256_avg_epu8(avg_rg, b);

        // Interleave the alpha and grayscale channels
        let unpacked_lo = _mm256_unpacklo_epi8(avg_rgb, a);
        let unpacked_hi = _mm256_unpackhi_epi8(avg_rgb, a);

        // Store the interleaved channels in the output vector
        _mm256_storeu2_m128i(data.as_mut_ptr().offset((i / 2) as isize + 32) as *mut _,
                            data.as_mut_ptr().offset((i / 2) as isize) as *mut _,
                            unpacked_lo);
        _mm256_storeu2_m128i(data.as_mut_ptr().offset((i / 2) as isize + 48) as *mut _,
                             data.as_mut_ptr().offset((i / 2) as isize + 16) as *mut _,
                             unpacked_hi);

        i += chunk_size;
    }

    if i > num_bytes {
        for j in ((i - chunk_size)..num_bytes).step_by(channels) {
            let sum = (input.data()[i] + input.data()[i + 1] + input.data()[i + 2]) as f32;
            data[j / 2] = (sum / 3.0).round() as u8;
            data[j / 2 + 1] = input.data()[i + 3];
        }
    }

    Image::from_vec(input.info().width, input.info().height, 2,
                    input.info().alpha, data)
}

/// Computes the average of each of `input`, ignoring the alpha channel if present, and
/// returns a grayscale image
#[cfg(feature = "simd")]
pub fn avg_checked_256_u8(input: &Image<u8>) -> Image<u8> {
    if input.info().alpha {
        unsafe { avg_alpha_256_u8(input) }
    } else {
        unsafe { avg_256_u8(input) }
    }
}