libblur 0.24.0

Fast image blurring in pure 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

/*
 * // Copyright (c) Radzivon Bartoshyk 6/2025. All rights reserved.
 * //
 * // Redistribution and use in source and binary forms, with or without modification,
 * // are permitted provided that the following conditions are met:
 * //
 * // 1.  Redistributions of source code must retain the above copyright notice, this
 * // list of conditions and the following disclaimer.
 * //
 * // 2.  Redistributions in binary form must reproduce the above copyright notice,
 * // this list of conditions and the following disclaimer in the documentation
 * // and/or other materials provided with the distribution.
 * //
 * // 3.  Neither the name of the copyright holder nor the names of its
 * // contributors may be used to endorse or promote products derived from
 * // this software without specific prior written permission.
 * //
 * // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
 * // AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 * // DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
 * // FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * // DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
 * // SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
 * // CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
 * // OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 * // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
#![allow(clippy::manual_clamp)]
use crate::BilateralBlurParams;
use crate::bilateral::bp8::{BilateralStore, BilateralUnit};
use crate::filter1d::Arena;
use std::arch::aarch64::*;

pub(crate) struct BilateralExecutionUnitNeon<'a, const N: usize> {
    pub(crate) arena: Arena,
    pub(crate) params: BilateralBlurParams,
    pub(crate) store: &'a BilateralStore,
    pub(crate) src_width: usize,
}

#[inline(always)]
fn v_lut(v: uint16x8_t, lut: &[f32; 65536]) -> (float32x4_t, float32x4_t) {
    unsafe {
        let v0 = vld1q_lane_f32::<0>(
            lut.get_unchecked(vgetq_lane_u16::<0>(v) as usize),
            vdupq_n_f32(0.),
        );
        let v1 = vld1q_lane_f32::<1>(lut.get_unchecked(vgetq_lane_u16::<1>(v) as usize), v0);
        let v2 = vld1q_lane_f32::<2>(lut.get_unchecked(vgetq_lane_u16::<2>(v) as usize), v1);
        let v3 = vld1q_lane_f32::<3>(lut.get_unchecked(vgetq_lane_u16::<3>(v) as usize), v2);

        let v4 = vld1q_lane_f32::<0>(
            lut.get_unchecked(vgetq_lane_u16::<4>(v) as usize),
            vdupq_n_f32(0.),
        );
        let v5 = vld1q_lane_f32::<1>(lut.get_unchecked(vgetq_lane_u16::<5>(v) as usize), v4);
        let v6 = vld1q_lane_f32::<2>(lut.get_unchecked(vgetq_lane_u16::<6>(v) as usize), v5);
        let v7 = vld1q_lane_f32::<3>(lut.get_unchecked(vgetq_lane_u16::<7>(v) as usize), v6);
        (v3, v7)
    }
}

#[inline(always)]
fn replace_zeros_with_ones(input: float32x4_t) -> float32x4_t {
    unsafe {
        let is_zero = vceqq_f32(input, vdupq_n_f32(0.0));
        vbslq_f32(is_zero, vdupq_n_f32(1.0), input)
    }
}

impl<const N: usize> BilateralUnit<u8> for BilateralExecutionUnitNeon<'_, N> {
    fn execute(&self, a_src: &[u8], y: usize, dst_row: &mut [u8], src_row: &[u8]) {
        let sliced_range = &self.store.range[..self.params.kernel_size * self.params.kernel_size];
        let ss = &self.store.spatial;
        let useful_width = self.src_width * N;
        let a_stride = self.arena.width * self.arena.components;
        let mut dst_row = &mut dst_row[..useful_width];
        let mut src_row = &src_row[..useful_width];
        let mut offset = 0usize;

        unsafe {
            for (wx, (dst, center)) in dst_row
                .chunks_exact_mut(16)
                .zip(src_row.chunks_exact(16))
                .enumerate()
            {
                let mut sum0 = vdupq_n_f32(0.);
                let mut sum1 = vdupq_n_f32(0.);
                let mut sum2 = vdupq_n_f32(0.);
                let mut sum3 = vdupq_n_f32(0.);

                let mut iw0 = vdupq_n_f32(0.);
                let mut iw1 = vdupq_n_f32(0.);
                let mut iw2 = vdupq_n_f32(0.);
                let mut iw3 = vdupq_n_f32(0.);

                let cxz = vld1q_u8(center.as_ptr().cast());
                let cx0 = vshll_n_u8::<8>(vget_low_u8(cxz));
                let cx1 = vshll_high_n_u8::<8>(cxz);

                let x = wx * 16;

                for (ky, ky_row) in sliced_range
                    .chunks_exact(self.params.kernel_size)
                    .enumerate()
                {
                    let c_slice = (y + ky) * a_stride + x;

                    for (w, &rwz) in ky_row.iter().enumerate() {
                        let v_rwz = vdupq_n_f32(rwz);
                        let pxz = vld1q_u8(
                            a_src
                                .get_unchecked(c_slice + w * N..c_slice + w * N + 16)
                                .as_ptr()
                                .cast(),
                        );

                        let px16l = vmovl_u8(vget_low_u8(pxz));
                        let px16h = vmovl_high_u8(pxz);

                        let px0_lo = vcvtq_f32_u32(vmovl_u16(vget_low_u16(px16l)));
                        let px0_hi = vcvtq_f32_u32(vmovl_high_u16(px16l));
                        let px1_lo = vcvtq_f32_u32(vmovl_u16(vget_low_u16(px16h)));
                        let px1_hi = vcvtq_f32_u32(vmovl_high_u16(px16h));

                        let w0 = vaddq_u16(cx0, px16l);
                        let w1 = vaddq_u16(cx1, px16h);

                        let (mut z0, mut z1) = v_lut(w0, ss);
                        let (mut z2, mut z3) = v_lut(w1, ss);

                        z0 = vmulq_f32(z0, v_rwz);
                        z1 = vmulq_f32(z1, v_rwz);
                        z2 = vmulq_f32(z2, v_rwz);
                        z3 = vmulq_f32(z3, v_rwz);

                        sum0 = vfmaq_f32(sum0, z0, px0_lo);
                        sum1 = vfmaq_f32(sum1, z1, px0_hi);
                        sum2 = vfmaq_f32(sum2, z2, px1_lo);
                        sum3 = vfmaq_f32(sum3, z3, px1_hi);

                        iw0 = vaddq_f32(iw0, z0);
                        iw1 = vaddq_f32(iw1, z1);
                        iw2 = vaddq_f32(iw2, z2);
                        iw3 = vaddq_f32(iw3, z3);
                    }
                }

                iw0 = replace_zeros_with_ones(iw0);
                iw1 = replace_zeros_with_ones(iw1);
                iw2 = replace_zeros_with_ones(iw2);
                iw3 = replace_zeros_with_ones(iw3);

                let s0 = vcvtaq_u32_f32(vdivq_f32(sum0, iw0));
                let s1 = vcvtaq_u32_f32(vdivq_f32(sum1, iw1));
                let s2 = vcvtaq_u32_f32(vdivq_f32(sum2, iw2));
                let s3 = vcvtaq_u32_f32(vdivq_f32(sum3, iw3));

                let ss0 = vmovn_u32(s0);
                let ss1 = vmovn_u32(s1);
                let ss2 = vmovn_u32(s2);
                let ss3 = vmovn_u32(s3);

                let vss0 = vqmovn_u16(vcombine_u16(ss0, ss1));
                let vss1 = vqmovn_u16(vcombine_u16(ss2, ss3));
                vst1q_u8(dst.as_mut_ptr().cast(), vcombine_u8(vss0, vss1));
                offset += 16;
            }

            dst_row = dst_row.chunks_exact_mut(16).into_remainder();
            src_row = src_row.chunks_exact(16).remainder();
        }

        unsafe {
            for (wx, (dst, center)) in dst_row
                .chunks_exact_mut(8)
                .zip(src_row.chunks_exact(8))
                .enumerate()
            {
                let mut sum0 = vdupq_n_f32(0.);
                let mut sum1 = vdupq_n_f32(0.);

                let mut iw0 = vdupq_n_f32(0.);
                let mut iw1 = vdupq_n_f32(0.);

                let cx = vshll_n_u8::<8>(vld1_u8(center.as_ptr().cast()));

                let x = offset + wx * 8;

                for (ky, ky_row) in sliced_range
                    .chunks_exact(self.params.kernel_size)
                    .enumerate()
                {
                    let c_slice = (y + ky) * a_stride + x;

                    for (w, &rwz) in ky_row.iter().enumerate() {
                        let v_rwz = vdupq_n_f32(rwz);
                        let px = vld1_u8(
                            a_src
                                .get_unchecked(c_slice + w * N..c_slice + w * N + 8)
                                .as_ptr()
                                .cast(),
                        );

                        let px16 = vmovl_u8(px);
                        let px_lo = vcvtq_f32_u32(vmovl_u16(vget_low_u16(px16)));
                        let px_hi = vcvtq_f32_u32(vmovl_high_u16(px16));

                        let w0 = vaddq_u16(cx, px16);
                        let (mut z0, mut z1) = v_lut(w0, ss);

                        z0 = vmulq_f32(z0, v_rwz);
                        z1 = vmulq_f32(z1, v_rwz);

                        sum0 = vfmaq_f32(sum0, z0, px_lo);
                        sum1 = vfmaq_f32(sum1, z1, px_hi);

                        iw0 = vaddq_f32(iw0, z0);
                        iw1 = vaddq_f32(iw1, z1);
                    }
                }

                iw0 = replace_zeros_with_ones(iw0);
                iw1 = replace_zeros_with_ones(iw1);

                let s0 = vcvtaq_u32_f32(vdivq_f32(sum0, iw0));
                let s1 = vcvtaq_u32_f32(vdivq_f32(sum1, iw1));

                let ss0 = vmovn_u32(s0);
                let ss1 = vmovn_u32(s1);

                let vss0 = vqmovn_u16(vcombine_u16(ss0, ss1));
                vst1_u8(dst.as_mut_ptr().cast(), vss0);
                offset += 8;
            }
        }

        let dst_row = dst_row.chunks_exact_mut(8).into_remainder();
        let src_row = src_row.chunks_exact(8).remainder();

        for (x, (dst, &center)) in dst_row.iter_mut().zip(src_row.iter()).enumerate() {
            let mut sum0 = 0f32;
            let mut iw0 = 0f32;

            let x = x + offset;

            let sx = x % N;
            for (ky, ky_row) in sliced_range
                .chunks_exact(self.params.kernel_size)
                .enumerate()
            {
                let c_slice = (y + ky) * a_stride + x - sx;
                let c_px_slice =
                    &a_src[(c_slice + sx)..(c_slice + N * self.params.kernel_size - sx)];
                for (&c_px, &rwz) in c_px_slice.iter().step_by(N).zip(ky_row.iter()) {
                    let z0 = rwz * ss[(center as u16 * 256 + c_px as u16) as usize];
                    sum0 += z0 * c_px as f32;
                    iw0 += z0;
                }
            }

            iw0 = if iw0 == 0. { 1. } else { iw0 };

            *dst = (sum0 / iw0).round().min(255.).max(0.) as u8;
            offset += 1;
        }
    }
}