yuv 0.8.13

High performance utilities for YUV format handling and conversion.
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
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348

/*
 * Copyright (c) Radzivon Bartoshyk, 10/2024. 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.
 */

use crate::internals::ProcessedOffset;
use crate::yuv_support::{CbCrForwardTransform, YuvChromaRange, YuvNVOrder, YuvSourceChannels};
use std::arch::aarch64::*;

#[target_feature(enable = "i8mm")]
pub(crate) unsafe fn neon_rgba_to_nv_dot_rgba420<const ORIGIN_CHANNELS: u8, const UV_ORDER: u8>(
    y_plane0: &mut [u8],
    y_plane1: &mut [u8],
    uv_plane: &mut [u8],
    rgba0: &[u8],
    rgba1: &[u8],
    width: u32,
    range: &YuvChromaRange,
    transform: &CbCrForwardTransform<i32>,
    start_cx: usize,
    start_ux: usize,
) -> ProcessedOffset {
    let source_channels: YuvSourceChannels = ORIGIN_CHANNELS.into();
    let order: YuvNVOrder = UV_ORDER.into();
    assert!(
        source_channels == YuvSourceChannels::Rgba || source_channels == YuvSourceChannels::Bgra
    );
    let channels = source_channels.get_channels_count();

    const A_E: i32 = 7;
    let y_bias = vdupq_n_s32(range.bias_y as i32 * (1 << A_E) + (1 << (A_E - 1)) - 1);
    let uv_bias = vdupq_n_s32(range.bias_uv as i32 * (1 << A_E) + (1 << (A_E - 1)) - 1);

    let weights_yr: i32 = i32::from_ne_bytes([
        transform.yr as u8,
        transform.yg as u8,
        transform.yb as u8,
        0,
    ]);
    let weights_yr_bgra: i32 = i32::from_ne_bytes([
        transform.yb as u8,
        transform.yg as u8,
        transform.yr as u8,
        0,
    ]);

    let weights_cb_rgba: i32 = i32::from_ne_bytes([
        transform.cb_r as u8,
        transform.cb_g as u8,
        transform.cb_b as u8,
        0,
    ]);

    let weights_cb_bgra: i32 = i32::from_ne_bytes([
        transform.cb_b as u8,
        transform.cb_g as u8,
        transform.cb_r as u8,
        0,
    ]);

    let weights_cr_rgba: i32 = i32::from_ne_bytes([
        transform.cr_r as u8,
        transform.cr_g as u8,
        transform.cr_b as u8,
        0,
    ]);
    let weights_cr_bgra: i32 = i32::from_ne_bytes([
        transform.cr_b as u8,
        transform.cr_g as u8,
        transform.cr_r as u8,
        0,
    ]);

    let y_weights =
        vreinterpretq_s8_s32(vdupq_n_s32(if source_channels == YuvSourceChannels::Rgba {
            weights_yr
        } else {
            weights_yr_bgra
        }));
    let cb_weights =
        vreinterpretq_s8_s32(vdupq_n_s32(if source_channels == YuvSourceChannels::Rgba {
            weights_cb_rgba
        } else {
            weights_cb_bgra
        }));
    let cr_weights =
        vreinterpretq_s8_s32(vdupq_n_s32(if source_channels == YuvSourceChannels::Rgba {
            weights_cr_rgba
        } else {
            weights_cr_bgra
        }));

    let mut cx = start_cx;
    let mut ux = start_ux;

    while cx + 16 <= width as usize {
        let src0 = rgba0.get_unchecked(cx * channels..).as_ptr();
        let src1 = rgba1.get_unchecked(cx * channels..).as_ptr();

        let v0 = vld1q_u8(src0);
        let v1 = vld1q_u8(src0.add(16));
        let v2 = vld1q_u8(src0.add(32));
        let v3 = vld1q_u8(src0.add(48));

        let v4 = vld1q_u8(src1);
        let v5 = vld1q_u8(src1.add(16));
        let v6 = vld1q_u8(src1.add(32));
        let v7 = vld1q_u8(src1.add(48));

        let y0 = vusdotq_s32(y_bias, v0, y_weights);
        let y1 = vusdotq_s32(y_bias, v1, y_weights);
        let y2 = vusdotq_s32(y_bias, v2, y_weights);
        let y3 = vusdotq_s32(y_bias, v3, y_weights);

        let y4 = vusdotq_s32(y_bias, v4, y_weights);
        let y5 = vusdotq_s32(y_bias, v5, y_weights);
        let y6 = vusdotq_s32(y_bias, v6, y_weights);
        let y7 = vusdotq_s32(y_bias, v7, y_weights);

        let yn_0 = vqshrun_n_s32::<A_E>(y0);
        let yn_1 = vqshrun_n_s32::<A_E>(y1);
        let yn_2 = vqshrun_n_s32::<A_E>(y2);
        let yn_3 = vqshrun_n_s32::<A_E>(y3);

        let yn_4 = vqshrun_n_s32::<A_E>(y4);
        let yn_5 = vqshrun_n_s32::<A_E>(y5);
        let yn_6 = vqshrun_n_s32::<A_E>(y6);
        let yn_7 = vqshrun_n_s32::<A_E>(y7);

        let j0 = vqmovn_u16(vcombine_u16(yn_0, yn_1));
        let j1 = vqmovn_u16(vcombine_u16(yn_2, yn_3));
        let j2 = vqmovn_u16(vcombine_u16(yn_4, yn_5));
        let j3 = vqmovn_u16(vcombine_u16(yn_6, yn_7));

        let y_vl0 = vcombine_u8(j0, j1);
        let y_vl1 = vcombine_u8(j2, j3);

        vst1q_u8(y_plane0.get_unchecked_mut(cx..).as_mut_ptr(), y_vl0);
        vst1q_u8(y_plane1.get_unchecked_mut(cx..).as_mut_ptr(), y_vl1);

        let v0 = vhaddq_u8(v0, v4);
        let v1 = vhaddq_u8(v1, v5);
        let v2 = vhaddq_u8(v2, v6);
        let v3 = vhaddq_u8(v3, v7);

        let uzp0 = vuzpq_u32(vreinterpretq_u32_u8(v0), vreinterpretq_u32_u8(v1));
        let uzp1 = vuzpq_u32(vreinterpretq_u32_u8(v2), vreinterpretq_u32_u8(v3));

        let v0_s = vreinterpretq_u8_u32(uzp0.0);
        let v1_s = vreinterpretq_u8_u32(uzp0.1);
        let v2_s = vreinterpretq_u8_u32(uzp1.0);
        let v3_s = vreinterpretq_u8_u32(uzp1.1);

        let v0_f = vhaddq_u8(v0_s, v1_s);
        let v1_f = vhaddq_u8(v2_s, v3_s);

        let cb0 = vusdotq_s32(uv_bias, v0_f, cb_weights);
        let cb1 = vusdotq_s32(uv_bias, v1_f, cb_weights);

        let cr0 = vusdotq_s32(uv_bias, v0_f, cr_weights);
        let cr1 = vusdotq_s32(uv_bias, v1_f, cr_weights);

        let cb_0 = vqshrun_n_s32::<A_E>(cb0);
        let cb_1 = vqshrun_n_s32::<A_E>(cb1);

        let cr_0 = vqshrun_n_s32::<A_E>(cr0);
        let cr_1 = vqshrun_n_s32::<A_E>(cr1);

        let mut cb_vl = vqmovn_u16(vcombine_u16(cb_0, cb_1));
        let mut cr_vl = vqmovn_u16(vcombine_u16(cr_0, cr_1));

        if order == YuvNVOrder::VU {
            std::mem::swap(&mut cb_vl, &mut cr_vl);
        }

        vst2_u8(
            uv_plane.get_unchecked_mut(ux..).as_mut_ptr(),
            uint8x8x2_t(cb_vl, cr_vl),
        );

        ux += 16;
        cx += 16;
    }

    if cx < width as usize {
        let diff = width as usize - cx;
        assert!(diff <= 16);

        let mut src_buffer0: [u8; 16 * 4] = [0; 16 * 4];
        let mut src_buffer1: [u8; 16 * 4] = [0; 16 * 4];
        let mut y_buffer0: [u8; 16] = [0; 16];
        let mut y_buffer1: [u8; 16] = [0; 16];
        let mut uv_buffer: [u8; 16] = [0; 16];

        std::ptr::copy_nonoverlapping(
            rgba0.get_unchecked(cx * channels..).as_ptr(),
            src_buffer0.as_mut_ptr().cast(),
            diff * channels,
        );
        std::ptr::copy_nonoverlapping(
            rgba1.get_unchecked(cx * channels..).as_ptr(),
            src_buffer1.as_mut_ptr().cast(),
            diff * channels,
        );

        // Replicate last item to one more position for subsampling
        if diff % 2 != 0 {
            let lst = (width as usize - 1) * channels;
            let last_items0 = rgba0.get_unchecked(lst..(lst + channels));
            let last_items1 = rgba1.get_unchecked(lst..(lst + channels));
            let dvb = diff * channels;
            let dst0 = src_buffer0.get_unchecked_mut(dvb..(dvb + channels));
            let dst1 = src_buffer1.get_unchecked_mut(dvb..(dvb + channels));
            for (dst, src) in dst0.iter_mut().zip(last_items0) {
                *dst = *src;
            }
            for (dst, src) in dst1.iter_mut().zip(last_items1) {
                *dst = *src;
            }
        }

        let v0 = vld1q_u8(src_buffer0.as_ptr().cast());
        let v1 = vld1q_u8(src_buffer0.as_ptr().add(16).cast());
        let v2 = vld1q_u8(src_buffer0.as_ptr().add(32).cast());
        let v3 = vld1q_u8(src_buffer0.as_ptr().add(48).cast());

        let v4 = vld1q_u8(src_buffer1.as_ptr().cast());
        let v5 = vld1q_u8(src_buffer1.as_ptr().add(16).cast());
        let v6 = vld1q_u8(src_buffer1.as_ptr().add(32).cast());
        let v7 = vld1q_u8(src_buffer1.as_ptr().add(48).cast());

        let y0 = vusdotq_s32(y_bias, v0, y_weights);
        let y1 = vusdotq_s32(y_bias, v1, y_weights);
        let y2 = vusdotq_s32(y_bias, v2, y_weights);
        let y3 = vusdotq_s32(y_bias, v3, y_weights);

        let y4 = vusdotq_s32(y_bias, v4, y_weights);
        let y5 = vusdotq_s32(y_bias, v5, y_weights);
        let y6 = vusdotq_s32(y_bias, v6, y_weights);
        let y7 = vusdotq_s32(y_bias, v7, y_weights);

        let yn_0 = vqshrun_n_s32::<A_E>(y0);
        let yn_1 = vqshrun_n_s32::<A_E>(y1);
        let yn_2 = vqshrun_n_s32::<A_E>(y2);
        let yn_3 = vqshrun_n_s32::<A_E>(y3);

        let yn_4 = vqshrun_n_s32::<A_E>(y4);
        let yn_5 = vqshrun_n_s32::<A_E>(y5);
        let yn_6 = vqshrun_n_s32::<A_E>(y6);
        let yn_7 = vqshrun_n_s32::<A_E>(y7);

        let j0 = vqmovn_u16(vcombine_u16(yn_0, yn_1));
        let j1 = vqmovn_u16(vcombine_u16(yn_2, yn_3));
        let j2 = vqmovn_u16(vcombine_u16(yn_4, yn_5));
        let j3 = vqmovn_u16(vcombine_u16(yn_6, yn_7));

        let y_vl0 = vcombine_u8(j0, j1);
        let y_vl1 = vcombine_u8(j2, j3);

        vst1q_u8(y_buffer0.as_mut_ptr().cast(), y_vl0);
        vst1q_u8(y_buffer1.as_mut_ptr().cast(), y_vl1);

        let v0 = vhaddq_u8(v0, v4);
        let v1 = vhaddq_u8(v1, v5);
        let v2 = vhaddq_u8(v2, v6);
        let v3 = vhaddq_u8(v3, v7);

        let uzp0 = vuzpq_u32(vreinterpretq_u32_u8(v0), vreinterpretq_u32_u8(v1));
        let uzp1 = vuzpq_u32(vreinterpretq_u32_u8(v2), vreinterpretq_u32_u8(v3));

        let v0_s = vreinterpretq_u8_u32(uzp0.0);
        let v1_s = vreinterpretq_u8_u32(uzp0.1);
        let v2_s = vreinterpretq_u8_u32(uzp1.0);
        let v3_s = vreinterpretq_u8_u32(uzp1.1);

        let v0_f = vhaddq_u8(v0_s, v1_s);
        let v1_f = vhaddq_u8(v2_s, v3_s);

        let cb0 = vusdotq_s32(uv_bias, v0_f, cb_weights);
        let cb1 = vusdotq_s32(uv_bias, v1_f, cb_weights);

        let cr0 = vusdotq_s32(uv_bias, v0_f, cr_weights);
        let cr1 = vusdotq_s32(uv_bias, v1_f, cr_weights);

        let cb_0 = vqshrun_n_s32::<A_E>(cb0);
        let cb_1 = vqshrun_n_s32::<A_E>(cb1);

        let cr_0 = vqshrun_n_s32::<A_E>(cr0);
        let cr_1 = vqshrun_n_s32::<A_E>(cr1);

        let mut cb_vl = vqmovn_u16(vcombine_u16(cb_0, cb_1));
        let mut cr_vl = vqmovn_u16(vcombine_u16(cr_0, cr_1));

        if order == YuvNVOrder::VU {
            std::mem::swap(&mut cb_vl, &mut cr_vl);
        }

        vst2_u8(uv_buffer.as_mut_ptr().cast(), uint8x8x2_t(cb_vl, cr_vl));

        std::ptr::copy_nonoverlapping(
            y_buffer0.as_ptr().cast(),
            y_plane0.get_unchecked_mut(cx..).as_mut_ptr(),
            diff,
        );
        std::ptr::copy_nonoverlapping(
            y_buffer1.as_ptr().cast(),
            y_plane1.get_unchecked_mut(cx..).as_mut_ptr(),
            diff,
        );

        let hv = diff.div_ceil(2) * 2;

        std::ptr::copy_nonoverlapping(
            uv_buffer.as_ptr().cast(),
            uv_plane.get_unchecked_mut(ux..).as_mut_ptr(),
            hv,
        );

        cx += diff;
        ux += hv;
    }

    ProcessedOffset { cx, ux }
}