fast_image_resize 6.0.0

Library for fast image resizing with using of SIMD instructions
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

use core::arch::x86_64::*;

use super::native;
use crate::pixels::F32x4;
use crate::{ImageView, ImageViewMut};

#[target_feature(enable = "avx2")]
pub(crate) unsafe fn multiply_alpha(
    src_view: &impl ImageView<Pixel = F32x4>,
    dst_view: &mut impl ImageViewMut<Pixel = F32x4>,
) {
    let src_rows = src_view.iter_rows(0);
    let dst_rows = dst_view.iter_rows_mut(0);

    for (src_row, dst_row) in src_rows.zip(dst_rows) {
        multiply_alpha_row(src_row, dst_row);
    }
}

#[target_feature(enable = "avx2")]
pub(crate) unsafe fn multiply_alpha_inplace(image_view: &mut impl ImageViewMut<Pixel = F32x4>) {
    for row in image_view.iter_rows_mut(0) {
        multiply_alpha_row_inplace(row);
    }
}

#[inline]
#[target_feature(enable = "avx2")]
pub(crate) unsafe fn multiply_alpha_row(src_row: &[F32x4], dst_row: &mut [F32x4]) {
    let src_chunks = src_row.chunks_exact(8);
    let src_remainder = src_chunks.remainder();
    let mut dst_chunks = dst_row.chunks_exact_mut(8);
    for (src_chunk, dst_chunk) in src_chunks.zip(&mut dst_chunks) {
        let src_pixels = load_8_pixels(src_chunk);
        multiply_alpha_8_pixels(src_pixels, dst_chunk);
    }

    if !src_remainder.is_empty() {
        let dst_reminder = dst_chunks.into_remainder();
        native::multiply_alpha_row(src_remainder, dst_reminder);
    }
}

#[inline]
#[target_feature(enable = "avx2")]
pub(crate) unsafe fn multiply_alpha_row_inplace(row: &mut [F32x4]) {
    let mut chunks = row.chunks_exact_mut(8);
    for chunk in &mut chunks {
        let src_pixels = load_8_pixels(chunk);
        multiply_alpha_8_pixels(src_pixels, chunk);
    }

    let reminder = chunks.into_remainder();
    if !reminder.is_empty() {
        native::multiply_alpha_row_inplace(reminder);
    }
}

#[inline]
#[target_feature(enable = "avx2")]
unsafe fn multiply_alpha_8_pixels(pixels: [__m256; 4], dst_chunk: &mut [F32x4]) {
    let r_f32x8 = _mm256_mul_ps(pixels[0], pixels[3]);
    let g_f32x8 = _mm256_mul_ps(pixels[1], pixels[3]);
    let b_f32x8 = _mm256_mul_ps(pixels[2], pixels[3]);
    store_8_pixels([r_f32x8, g_f32x8, b_f32x8, pixels[3]], dst_chunk);
}

// Divide

#[target_feature(enable = "avx2")]
pub(crate) unsafe fn divide_alpha(
    src_view: &impl ImageView<Pixel = F32x4>,
    dst_view: &mut impl ImageViewMut<Pixel = F32x4>,
) {
    let src_rows = src_view.iter_rows(0);
    let dst_rows = dst_view.iter_rows_mut(0);

    for (src_row, dst_row) in src_rows.zip(dst_rows) {
        divide_alpha_row(src_row, dst_row);
    }
}

#[target_feature(enable = "avx2")]
pub(crate) unsafe fn divide_alpha_inplace(image_view: &mut impl ImageViewMut<Pixel = F32x4>) {
    for row in image_view.iter_rows_mut(0) {
        divide_alpha_row_inplace(row);
    }
}

#[target_feature(enable = "avx2")]
pub(crate) unsafe fn divide_alpha_row(src_row: &[F32x4], dst_row: &mut [F32x4]) {
    let src_chunks = src_row.chunks_exact(8);
    let src_remainder = src_chunks.remainder();
    let mut dst_chunks = dst_row.chunks_exact_mut(8);

    for (src_chunk, dst_chunk) in src_chunks.zip(&mut dst_chunks) {
        let src_pixels = load_8_pixels(src_chunk);
        divide_alpha_8_pixels(src_pixels, dst_chunk);
    }

    if !src_remainder.is_empty() {
        let dst_reminder = dst_chunks.into_remainder();
        native::divide_alpha_row(src_remainder, dst_reminder);
    }
}

#[target_feature(enable = "avx2")]
pub(crate) unsafe fn divide_alpha_row_inplace(row: &mut [F32x4]) {
    let mut chunks = row.chunks_exact_mut(8);
    for chunk in &mut chunks {
        let src_pixels = load_8_pixels(chunk);
        divide_alpha_8_pixels(src_pixels, chunk);
    }

    let reminder = chunks.into_remainder();
    if !reminder.is_empty() {
        native::divide_alpha_row_inplace(reminder);
    }
}

#[inline]
#[target_feature(enable = "avx2")]
unsafe fn divide_alpha_8_pixels(pixels: [__m256; 4], dst_chunk: &mut [F32x4]) {
    let mut r_f32x8 = _mm256_div_ps(pixels[0], pixels[3]);
    let mut g_f32x8 = _mm256_div_ps(pixels[1], pixels[3]);
    let mut b_f32x8 = _mm256_div_ps(pixels[2], pixels[3]);

    let zero = _mm256_setzero_ps();
    let mask_zero = _mm256_cmp_ps::<_CMP_NEQ_UQ>(pixels[3], zero);
    r_f32x8 = _mm256_and_ps(mask_zero, r_f32x8);
    g_f32x8 = _mm256_and_ps(mask_zero, g_f32x8);
    b_f32x8 = _mm256_and_ps(mask_zero, b_f32x8);

    store_8_pixels([r_f32x8, g_f32x8, b_f32x8, pixels[3]], dst_chunk);
}

#[inline]
#[target_feature(enable = "avx2")]
unsafe fn load_8_pixels(pixels: &[F32x4]) -> [__m256; 4] {
    let ptr = pixels.as_ptr() as *const f32;
    cols_into_rows([
        _mm256_loadu_ps(ptr),
        _mm256_loadu_ps(ptr.add(8)),
        _mm256_loadu_ps(ptr.add(16)),
        _mm256_loadu_ps(ptr.add(24)),
    ])
}

#[inline]
#[target_feature(enable = "avx2")]
unsafe fn store_8_pixels(pixels: [__m256; 4], dst_chunk: &mut [F32x4]) {
    let pixels = cols_into_rows(pixels);
    let mut dst_ptr = dst_chunk.as_mut_ptr() as *mut f32;
    for rgba in pixels {
        _mm256_storeu_ps(dst_ptr, rgba);
        dst_ptr = dst_ptr.add(8)
    }
}

#[inline]
#[target_feature(enable = "avx2")]
unsafe fn cols_into_rows(pixels: [__m256; 4]) -> [__m256; 4] {
    let rrgg02_rrgg13 = _mm256_unpacklo_ps(pixels[0], pixels[1]);
    let rrgg46_rrgg57 = _mm256_unpacklo_ps(pixels[2], pixels[3]);
    let r0246_r1357 = _mm256_castsi256_ps(_mm256_unpacklo_epi64(
        _mm256_castps_si256(rrgg02_rrgg13),
        _mm256_castps_si256(rrgg46_rrgg57),
    ));
    let g0246_g1357 = _mm256_castsi256_ps(_mm256_unpackhi_epi64(
        _mm256_castps_si256(rrgg02_rrgg13),
        _mm256_castps_si256(rrgg46_rrgg57),
    ));

    let bbaa02_bbaa13 = _mm256_unpackhi_ps(pixels[0], pixels[1]);
    let bbaa46_bbaa57 = _mm256_unpackhi_ps(pixels[2], pixels[3]);
    let b0246_b1357 = _mm256_castsi256_ps(_mm256_unpacklo_epi64(
        _mm256_castps_si256(bbaa02_bbaa13),
        _mm256_castps_si256(bbaa46_bbaa57),
    ));
    let a0246_a1357 = _mm256_castsi256_ps(_mm256_unpackhi_epi64(
        _mm256_castps_si256(bbaa02_bbaa13),
        _mm256_castps_si256(bbaa46_bbaa57),
    ));
    [r0246_r1357, g0246_g1357, b0246_b1357, a0246_a1357]
}