threers-codec 0.0.10

Pure-Rust media codecs: H.264/HEVC/VP9 encoders, H.264/HEVC/VP9/AV1 decoders, AAC/MP3/Opus/Vorbis/FLAC/WAV decoders, MP4/WebM/Matroska/Ogg, GIF/APNG, captions, and VideoToolbox acceleration on macOS
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
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
//! VP9 block reconstruction (ffmpeg `vp9recon.c`, `vp9_mc_template.c`).

use super::block::FrameDec;
use super::data::{BWH_TAB, INTRA_TXFM_TYPE};
use super::dsp;
use super::*;

/// `mode_conv[mode][have_left][have_top]`.
const MODE_CONV: [[[usize; 2]; 2]; 10] = [
    [[DC_127_PRED, VERT_PRED], [DC_127_PRED, VERT_PRED]],
    [[DC_129_PRED, DC_129_PRED], [HOR_PRED, HOR_PRED]],
    [[DC_128_PRED, TOP_DC_PRED], [LEFT_DC_PRED, DC_PRED]],
    [[DC_127_PRED, DIAG_DOWN_LEFT_PRED], [DC_127_PRED, DIAG_DOWN_LEFT_PRED]],
    [[DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_RIGHT_PRED], [DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_RIGHT_PRED]],
    [[VERT_RIGHT_PRED, VERT_RIGHT_PRED], [VERT_RIGHT_PRED, VERT_RIGHT_PRED]],
    [[HOR_DOWN_PRED, HOR_DOWN_PRED], [HOR_DOWN_PRED, HOR_DOWN_PRED]],
    [[DC_127_PRED, VERT_LEFT_PRED], [DC_127_PRED, VERT_LEFT_PRED]],
    [[DC_129_PRED, DC_129_PRED], [HOR_UP_PRED, HOR_UP_PRED]],
    [[DC_129_PRED, VERT_PRED], [HOR_PRED, TM_VP8_PRED]],
];

/// `edges[mode]`: (needs_left, needs_top, needs_topleft, needs_topright,
/// invert_left).
const EDGES: [(bool, bool, bool, bool, bool); 15] = [
    (false, true, false, false, false), // VERT
    (true, false, false, false, false), // HOR
    (true, true, false, false, false),  // DC
    (false, true, false, true, false),  // DIAG_DOWN_LEFT
    (true, true, true, false, false),   // DIAG_DOWN_RIGHT
    (true, true, true, false, false),   // VERT_RIGHT
    (true, true, true, false, false),   // HOR_DOWN
    (false, true, false, true, false),  // VERT_LEFT
    (true, false, false, false, true),  // HOR_UP
    (true, true, true, false, false),   // TM
    (true, false, false, false, false), // LEFT_DC
    (false, true, false, false, false), // TOP_DC
    (false, false, false, false, false),
    (false, false, false, false, false),
    (false, false, false, false, false),
];

fn rounded_div(a: i32, b: i32) -> i32 {
    if a >= 0 {
        (a + (b >> 1)) / b
    } else {
        (a - (b >> 1)) / b
    }
}

fn avg2(a: Mv, b: Mv) -> Mv {
    Mv { x: rounded_div(a.x as i32 + b.x as i32, 2) as i16, y: rounded_div(a.y as i32 + b.y as i32, 2) as i16 }
}

fn avg4(a: Mv, b: Mv, c: Mv, d: Mv) -> Mv {
    Mv {
        x: rounded_div(a.x as i32 + b.x as i32 + c.x as i32 + d.x as i32, 4) as i16,
        y: rounded_div(a.y as i32 + b.y as i32 + c.y as i32 + d.y as i32, 4) as i16,
    }
}

impl FrameDec<'_> {
    /// `check_intra_mode` + the prediction: predict the `4 << tx` block at
    /// `(px, py)` (samples) of plane `p`. `x`, `y` are its 4×4 position in
    /// the block, `w` the block width in 4×4 units.
    #[allow(clippy::too_many_arguments)]
    fn intra_block(&mut self, mode: usize, p: usize, px: usize, py: usize, x: usize, y: usize, w: usize, tx: u8) {
        let (ss_h, ss_v) = if p == 0 { (0, 0) } else { (self.ss_h, self.ss_v) };
        let (row, col) = (self.row, self.col);
        let bpp = self.bpp;
        let have_top = row > 0 || y > 0;
        let have_left = col > self.tile_col_start || x > 0;
        let have_right = x < w - 1;
        let mode = MODE_CONV[mode][have_left as usize][have_top as usize];
        let (needs_left, needs_top, needs_topleft, needs_topright, invert_left) = EDGES[mode];
        let n = 4usize << tx;
        let plane = &self.cur.planes[p];
        let stride = plane.width;
        let base = 128i32 << (bpp - 8);
        let mut top = [0i32; 65];
        let mut left = [0i32; 32];
        if needs_top {
            let n_px_need = n;
            let n_px_have = ((((self.cols - col) << (1 - ss_h)) as isize - x as isize) * 4).max(0) as usize;
            let n_px_need_tr = if tx == TX_4X4 && needs_topright && have_right { 4 } else { 0 };
            let at = |xx: usize| plane.data[(py - 1) * stride + xx] as i32;
            if have_top && (!needs_topleft || have_left) && (tx != TX_4X4 || !needs_topright || have_right) && n_px_need + n_px_need_tr <= n_px_have {
                // ffmpeg reads the row above in place (`*a = top`).
                for i in 0..n + n_px_need_tr {
                    top[1 + i] = at(px + i);
                }
                if needs_topleft {
                    top[0] = at(px - 1);
                }
                if tx == TX_4X4 && needs_topright && n_px_need_tr == 0 {
                    // Not reached: the in-place path requires have_right.
                    for i in 4..8 {
                        top[1 + i] = top[4];
                    }
                }
            } else {
                if have_top {
                    if n_px_need <= n_px_have {
                        for i in 0..n_px_need {
                            top[1 + i] = at(px + i);
                        }
                    } else {
                        for i in 0..n_px_have {
                            top[1 + i] = at(px + i);
                        }
                        let v = top[n_px_have];
                        for i in n_px_have..n_px_need {
                            top[1 + i] = v;
                        }
                    }
                } else {
                    for i in 0..n_px_need {
                        top[1 + i] = base - 1;
                    }
                }
                if needs_topleft {
                    top[0] = if have_left && have_top { at(px - 1) } else { base + if have_top { 1 } else { -1 } };
                }
                if tx == TX_4X4 && needs_topright {
                    if have_top && have_right && n_px_need + n_px_need_tr <= n_px_have {
                        for i in 4..8 {
                            top[1 + i] = at(px + i);
                        }
                    } else {
                        for i in 4..8 {
                            top[1 + i] = top[4];
                        }
                    }
                }
            }
        }
        if needs_left {
            if have_left {
                let n_px_need = n;
                let n_px_have = ((((self.rows - row) << (1 - ss_v)) as isize - y as isize) * 4).max(0) as usize;
                let lv = |i: usize| plane.data[(py + i) * stride + px - 1] as i32;
                if invert_left {
                    let k = n_px_need.min(n_px_have);
                    for i in 0..k {
                        left[i] = lv(i);
                    }
                    for i in k..n_px_need {
                        left[i] = left[k - 1];
                    }
                } else if n_px_need <= n_px_have {
                    for i in 0..n_px_need {
                        left[n_px_need - 1 - i] = lv(i);
                    }
                } else {
                    for i in 0..n_px_have {
                        left[n_px_need - 1 - i] = lv(i);
                    }
                    let v = left[n_px_need - n_px_have];
                    for i in 0..n_px_need - n_px_have {
                        left[i] = v;
                    }
                }
            } else {
                for v in left.iter_mut().take(n) {
                    *v = base + 1;
                }
            }
        }
        let plane = &mut self.cur.planes[p];
        let stride = plane.width;
        dsp::intra_pred(tx, mode, &mut plane.data, py * stride + px, stride, &left, &top, bpp);
    }

    /// `intra_recon`.
    pub(super) fn intra_recon(&mut self) {
        let b = self.b;
        let (row, col) = (self.row, self.col);
        let bs = b.bs as usize;
        let lossless = self.h.lossless as usize;
        let mut w4 = (BWH_TAB[1][bs][0] as usize) << 1;
        let step1d = 1usize << b.tx;
        let step = 1usize << (b.tx * 2);
        let mut end_x = (2 * (self.cols - col)).min(w4);
        let mut end_y = (2 * (self.rows - row)).min((BWH_TAB[1][bs][1] as usize) << 1);
        let tx = 4 * lossless + b.tx as usize;
        let uvtx = b.uvtx as usize + 4 * lossless;
        let bpp = self.bpp;
        let (x0, y0) = (col * 8, row * 8);
        let mut n = 0;
        let mut y = 0;
        while y < end_y {
            let mut x = 0;
            while x < end_x {
                let mode = b.mode[if b.bs > BS_8X8 && b.tx == TX_4X4 { y * 2 + x } else { 0 }] as usize;
                let txtp = INTRA_TXFM_TYPE[mode] as usize;
                let eob = if b.skip { 0 } else { self.eob[n] as usize };
                let (px, py) = (x0 + 4 * x, y0 + 4 * y);
                self.intra_block(mode, 0, px, py, x, y, w4, b.tx);
                if eob != 0 {
                    let pl = &mut self.cur.planes[0];
                    let stride = pl.width;
                    dsp::itxfm_add(tx, txtp, &mut pl.data, py * stride + px, stride, &mut self.block[16 * n..], eob, bpp);
                }
                x += step1d;
                n += step;
            }
            y += step1d;
        }
        // U/V.
        w4 >>= self.ss_h;
        end_x >>= self.ss_h;
        end_y >>= self.ss_v;
        let uvstep1d = 1usize << b.uvtx;
        let uvstep = 1usize << (b.uvtx * 2);
        let (cx0, cy0) = ((col * 8) >> self.ss_h, (row * 8) >> self.ss_v);
        for p in 0..2 {
            let mut n = 0;
            let mut y = 0;
            while y < end_y {
                let mut x = 0;
                while x < end_x {
                    let eob = if b.skip { 0 } else { self.uveob[p][n] as usize };
                    let (px, py) = (cx0 + 4 * x, cy0 + 4 * y);
                    self.intra_block(b.uvmode as usize, p + 1, px, py, x, y, w4, b.uvtx);
                    if eob != 0 {
                        let pl = &mut self.cur.planes[1 + p];
                        let stride = pl.width;
                        dsp::itxfm_add(uvtx, DCT_DCT, &mut pl.data, py * stride + px, stride, &mut self.uvblock[p][16 * n..], eob, bpp);
                    }
                    x += uvstep1d;
                    n += uvstep;
                }
                y += uvstep1d;
            }
        }
    }

    /// One luma prediction: `mc_luma_unscaled` / `mc_luma_scaled`. `(x, y)`
    /// is the block's sample position, `(px, py, pw, ph)` its sub-block
    /// offset and the whole block size (for the scaled MV clamp).
    #[allow(clippy::too_many_arguments)]
    fn mc_luma(&mut self, refi: usize, x: usize, y: usize, mv: Mv, px: i32, py: i32, pw: i32, ph: i32, bw: usize, bh: usize, avg: bool) {
        let rf = self.refs[self.b.r#ref[refi] as usize].clone().unwrap();
        let (w, h) = (rf.width as usize, rf.height as usize);
        let filter = self.b.filter;
        let bpp = self.bpp;
        let dst = &mut self.cur.planes[0];
        let dstride = dst.width;
        let doff = y * dstride + x;
        let scale = self.mvscale[self.b.r#ref[refi] as usize];
        let mut win = Vec::new();
        if scale[0] == 0 {
            let (mx, my) = (mv.x as i32, mv.y as i32);
            let yy = y as isize + (my >> 3) as isize;
            let xx = x as isize + (mx >> 3) as isize;
            let (mx, my) = ((mx & 7) << 1, (my & 7) << 1);
            dsp::fetch(&rf.planes[0], xx - 3, yy - 3, bw + 8, bh + 8, w, h, &mut win);
            dsp::mc_put(&mut dst.data, doff, dstride, &win, bw + 8, bw, bh, filter, mx, my, avg, bpp);
        } else {
            let step = self.mvstep[self.b.r#ref[refi] as usize];
            let sm = |n: i64, d: usize| ((n * scale[d] as i64) >> 14) as i32;
            let (x, y) = (x as i32, y as i32);
            let mvx = (mv.x as i32).clamp(-(x + pw - px + 4) * 8, (self.cols as i32 * 8 - x + px + 3) * 8);
            let mvy = (mv.y as i32).clamp(-(y + ph - py + 4) * 8, (self.rows as i32 * 8 - y + py + 3) * 8);
            // BUG (libvpx-compatible): the components scale separately.
            let mx = sm(mvx as i64 * 2, 0) + sm(x as i64 * 16, 0);
            let my = sm(mvy as i64 * 2, 1) + sm(y as i64 * 16, 1);
            let (xx, yy) = (mx >> 4, my >> 4);
            let (mx, my) = (mx & 15, my & 15);
            let refbw_m1 = (((bw as i32 - 1) * step[0] as i32 + mx) >> 4) as usize;
            let refbh_m1 = (((bh as i32 - 1) * step[1] as i32 + my) >> 4) as usize;
            dsp::fetch(&rf.planes[0], xx as isize - 3, yy as isize - 3, refbw_m1 + 8, refbh_m1 + 8, w, h, &mut win);
            dsp::mc_scaled(&mut dst.data, doff, dstride, &win, refbw_m1 + 8, bw, bh, filter, mx, my, step[0] as i32, step[1] as i32, avg, bpp);
        }
    }

    /// One chroma prediction (both planes): `mc_chroma_unscaled` /
    /// `mc_chroma_scaled`. `(x, y)` in chroma samples; `w`, `h` the
    /// reference's chroma size.
    #[allow(clippy::too_many_arguments)]
    fn mc_chroma(&mut self, refi: usize, x: usize, y: usize, mv: Mv, px: i32, py: i32, pw: i32, ph: i32, bw: usize, bh: usize, avg: bool) {
        let rf = self.refs[self.b.r#ref[refi] as usize].clone().unwrap();
        let (ss_h, ss_v) = (self.ss_h, self.ss_v);
        let w = ((rf.width as usize) + ss_h) >> ss_h;
        let h = ((rf.height as usize) + ss_v) >> ss_v;
        let filter = self.b.filter;
        let bpp = self.bpp;
        let scale = self.mvscale[self.b.r#ref[refi] as usize];
        let mut win = Vec::new();
        if scale[0] == 0 {
            let mx = mv.x as i32 * (1 << (1 - ss_h));
            let my = mv.y as i32 * (1 << (1 - ss_v));
            let yy = y as isize + (my >> 4) as isize;
            let xx = x as isize + (mx >> 4) as isize;
            let (mx, my) = (mx & 15, my & 15);
            for p in 1..3 {
                dsp::fetch(&rf.planes[p], xx - 3, yy - 3, bw + 8, bh + 8, w, h, &mut win);
                let dst = &mut self.cur.planes[p];
                let dstride = dst.width;
                dsp::mc_put(&mut dst.data, y * dstride + x, dstride, &win, bw + 8, bw, bh, filter, mx, my, avg, bpp);
            }
        } else {
            let step = self.mvstep[self.b.r#ref[refi] as usize];
            let sm = |n: i64, d: usize| ((n * scale[d] as i64) >> 14) as i32;
            let (x, y) = (x as i32, y as i32);
            let mx = if ss_h != 0 {
                // BUG (libvpx-compatible, webm issue 820).
                let mvx = (mv.x as i32).clamp(-(x + pw - px + 4) * 16, (self.cols as i32 * 4 - x + px + 3) * 16);
                sm(mvx as i64, 0) + (sm(x as i64 * 16, 0) & !15) + (sm(x as i64 * 32, 0) & 15)
            } else {
                let mvx = (mv.x as i32).clamp(-(x + pw - px + 4) * 8, (self.cols as i32 * 8 - x + px + 3) * 8);
                sm(mvx as i64 * 2, 0) + sm(x as i64 * 16, 0)
            };
            let my = if ss_v != 0 {
                let mvy = (mv.y as i32).clamp(-(y + ph - py + 4) * 16, (self.rows as i32 * 4 - y + py + 3) * 16);
                sm(mvy as i64, 1) + (sm(y as i64 * 16, 1) & !15) + (sm(y as i64 * 32, 1) & 15)
            } else {
                let mvy = (mv.y as i32).clamp(-(y + ph - py + 4) * 8, (self.rows as i32 * 8 - y + py + 3) * 8);
                sm(mvy as i64 * 2, 1) + sm(y as i64 * 16, 1)
            };
            let (xx, yy) = (mx >> 4, my >> 4);
            let (mx, my) = (mx & 15, my & 15);
            let refbw_m1 = (((bw as i32 - 1) * step[0] as i32 + mx) >> 4) as usize;
            let refbh_m1 = (((bh as i32 - 1) * step[1] as i32 + my) >> 4) as usize;
            for p in 1..3 {
                dsp::fetch(&rf.planes[p], xx as isize - 3, yy as isize - 3, refbw_m1 + 8, refbh_m1 + 8, w, h, &mut win);
                let dst = &mut self.cur.planes[p];
                let dstride = dst.width;
                dsp::mc_scaled(&mut dst.data, y as usize * dstride + x as usize, dstride, &win, refbw_m1 + 8, bw, bh, filter, mx, my, step[0] as i32, step[1] as i32, avg, bpp);
            }
        }
    }

    /// `inter_pred` (unscaled and scaled templates).
    fn inter_pred(&mut self) {
        let b = self.b;
        let (row, col) = (self.row, self.col);
        let (ss_h, ss_v) = (self.ss_h, self.ss_v);
        let scaled = self.mvscale[b.r#ref[0] as usize][0] != 0 || (b.comp && self.mvscale[b.r#ref[1] as usize][0] != 0);
        let refs = if b.comp { 2 } else { 1 };
        if b.bs > BS_8X8 {
            for i in 0..refs {
                let avg = i == 1;
                let mv = |k: usize| b.mv[k][i];
                if b.bs == BS_8X4 && !scaled {
                    self.mc_luma(i, col * 8, row * 8, mv(0), 0, 0, 0, 0, 8, 4, avg);
                    self.mc_luma(i, col * 8, row * 8 + 4, mv(2), 0, 0, 0, 0, 8, 4, avg);
                    if ss_v != 0 {
                        let uv = avg2(mv(0), mv(2));
                        self.mc_chroma(i, (col * 8) >> ss_h, row * 4, uv, 0, 0, 0, 0, 8 >> ss_h, 4, avg);
                    } else {
                        self.mc_chroma(i, (col * 8) >> ss_h, row * 8, mv(0), 0, 0, 0, 0, 8 >> ss_h, 4, avg);
                        // BUG (libvpx, webm issue 993): 4:2:2 8x8 bottom
                        // block's chroma uses the wrong motion vector.
                        let uv = if ss_h == 0 { mv(2) } else { avg2(mv(0), mv(2)) };
                        self.mc_chroma(i, (col * 8) >> ss_h, row * 8 + 4, uv, 0, 0, 0, 0, 8 >> ss_h, 4, avg);
                    }
                } else if b.bs == BS_4X8 && !scaled {
                    self.mc_luma(i, col * 8, row * 8, mv(0), 0, 0, 0, 0, 4, 8, avg);
                    self.mc_luma(i, col * 8 + 4, row * 8, mv(1), 0, 0, 0, 0, 4, 8, avg);
                    if ss_h != 0 {
                        let uv = avg2(mv(0), mv(1));
                        self.mc_chroma(i, col * 4, (row * 8) >> ss_v, uv, 0, 0, 0, 0, 4, 8 >> ss_v, avg);
                    } else {
                        self.mc_chroma(i, col * 8, (row * 8) >> ss_v, mv(0), 0, 0, 0, 0, 4, 8 >> ss_v, avg);
                        self.mc_chroma(i, col * 8 + 4, (row * 8) >> ss_v, mv(1), 0, 0, 0, 0, 4, 8 >> ss_v, avg);
                    }
                } else {
                    // 4×4 (and, when scaled, 8×4 / 4×8).
                    let (x8, y8) = (col * 8, row * 8);
                    self.mc_luma(i, x8, y8, mv(0), 0, 0, 8, 8, 4, 4, avg);
                    self.mc_luma(i, x8 + 4, y8, mv(1), 4, 0, 8, 8, 4, 4, avg);
                    self.mc_luma(i, x8, y8 + 4, mv(2), 0, 4, 8, 8, 4, 4, avg);
                    self.mc_luma(i, x8 + 4, y8 + 4, mv(3), 4, 4, 8, 8, 4, 4, avg);
                    if ss_v != 0 {
                        if ss_h != 0 {
                            let uv = avg4(mv(0), mv(1), mv(2), mv(3));
                            self.mc_chroma(i, col * 4, row * 4, uv, 0, 0, 4, 4, 4, 4, avg);
                        } else {
                            let uv = avg2(mv(0), mv(2));
                            self.mc_chroma(i, col * 8, row * 4, uv, 0, 0, 8, 4, 4, 4, avg);
                            let uv = avg2(mv(1), mv(3));
                            self.mc_chroma(i, col * 8 + 4, row * 4, uv, 4, 0, 8, 4, 4, 4, avg);
                        }
                    } else if ss_h != 0 {
                        let uv = avg2(mv(0), mv(1));
                        self.mc_chroma(i, col * 4, row * 8, uv, 0, 0, 4, 8, 4, 4, avg);
                        // BUG (libvpx, webm issue 993).
                        let uv = avg2(mv(1), mv(2));
                        self.mc_chroma(i, col * 4, row * 8 + 4, uv, 0, 4, 4, 8, 4, 4, avg);
                    } else {
                        self.mc_chroma(i, col * 8, row * 8, mv(0), 0, 0, 8, 8, 4, 4, avg);
                        self.mc_chroma(i, col * 8 + 4, row * 8, mv(1), 4, 0, 8, 8, 4, 4, avg);
                        self.mc_chroma(i, col * 8, row * 8 + 4, mv(2), 0, 4, 8, 8, 4, 4, avg);
                        self.mc_chroma(i, col * 8 + 4, row * 8 + 4, mv(3), 4, 4, 8, 8, 4, 4, avg);
                    }
                }
            }
        } else {
            let bs = b.bs as usize;
            let bw = BWH_TAB[0][bs][0] as usize * 4;
            let bh = BWH_TAB[0][bs][1] as usize * 4;
            let uvbw = BWH_TAB[ss_h][bs][0] as usize * 4;
            let uvbh = BWH_TAB[ss_v][bs][1] as usize * 4;
            for i in 0..refs {
                let avg = i == 1;
                let mv = b.mv[0][i];
                self.mc_luma(i, col * 8, row * 8, mv, 0, 0, bw as i32, bh as i32, bw, bh, avg);
                self.mc_chroma(i, (col * 8) >> ss_h, (row * 8) >> ss_v, mv, 0, 0, uvbw as i32, uvbh as i32, uvbw, uvbh, avg);
            }
        }
    }

    /// `inter_recon`.
    pub(super) fn inter_recon(&mut self) {
        let b = self.b;
        if self.mvscale[b.r#ref[0] as usize][0] == 0xFFFF || (b.comp && self.mvscale[b.r#ref[1] as usize][0] == 0xFFFF) {
            return;
        }
        self.inter_pred();
        if b.skip {
            return;
        }
        let (row, col) = (self.row, self.col);
        let bs = b.bs as usize;
        let w4 = (BWH_TAB[1][bs][0] as usize) << 1;
        let h4 = (BWH_TAB[1][bs][1] as usize) << 1;
        let step1d = 1usize << b.tx;
        let step = 1usize << (b.tx * 2);
        let mut end_x = (2 * (self.cols - col)).min(w4);
        let mut end_y = (2 * (self.rows - row)).min(h4);
        let lossless = self.h.lossless as usize;
        let tx = 4 * lossless + b.tx as usize;
        let uvtx = b.uvtx as usize + 4 * lossless;
        let bpp = self.bpp;
        let (x0, y0) = (col * 8, row * 8);
        let mut n = 0;
        let mut y = 0;
        while y < end_y {
            let mut x = 0;
            while x < end_x {
                let eob = self.eob[n] as usize;
                if eob != 0 {
                    let pl = &mut self.cur.planes[0];
                    let stride = pl.width;
                    dsp::itxfm_add(tx, DCT_DCT, &mut pl.data, (y0 + 4 * y) * stride + x0 + 4 * x, stride, &mut self.block[16 * n..], eob, bpp);
                }
                x += step1d;
                n += step;
            }
            y += step1d;
        }
        end_x >>= self.ss_h;
        end_y >>= self.ss_v;
        let uvstep1d = 1usize << b.uvtx;
        let uvstep = 1usize << (b.uvtx * 2);
        let (cx0, cy0) = ((col * 8) >> self.ss_h, (row * 8) >> self.ss_v);
        for p in 0..2 {
            let mut n = 0;
            let mut y = 0;
            while y < end_y {
                let mut x = 0;
                while x < end_x {
                    let eob = self.uveob[p][n] as usize;
                    if eob != 0 {
                        let pl = &mut self.cur.planes[1 + p];
                        let stride = pl.width;
                        dsp::itxfm_add(uvtx, DCT_DCT, &mut pl.data, (cy0 + 4 * y) * stride + cx0 + 4 * x, stride, &mut self.uvblock[p][16 * n..], eob, bpp);
                    }
                    x += uvstep1d;
                    n += uvstep;
                }
                y += uvstep1d;
            }
        }
    }
}