film_grain_tmpl.c 21.6 KB
Newer Older
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
/*
 * Copyright © 2018, Niklas Haas
 * Copyright © 2018, VideoLAN and dav1d authors
 * Copyright © 2018, Two Orioles, LLC
 * 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.
 *
 * 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 OWNER 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.
 */

29
30
#include "config.h"

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
#include <assert.h>
#include <stdint.h>

#include "common.h"
#include "common/intops.h"
#include "common/bitdepth.h"
#include "tables.h"

#include "film_grain.h"

#if BITDEPTH == 8
typedef int8_t entry;
#else
typedef int16_t entry;
#endif

enum {
    GRAIN_WIDTH  = 82,
    GRAIN_HEIGHT = 73,
    SUB_GRAIN_WIDTH = 44,
    SUB_GRAIN_HEIGHT = 38,
    SUB_GRAIN_OFFSET = 6,
    BLOCK_SIZE = 32,
    SCALING_SIZE = 1 << BITDEPTH,
};

static inline int get_random_number(const int bits, unsigned *state) {
    const int r = *state;
    unsigned bit = ((r >> 0) ^ (r >> 1) ^ (r >> 3) ^ (r >> 12)) & 1;
    *state = (r >> 1) | (bit << 15);

    return (*state >> (16 - bits)) & ((1 << bits) - 1);
}

static inline int round2(const int x, const int shift) {
    return (x + ((1 << shift) >> 1)) >> shift;
}

enum {
    GRAIN_CENTER = 128 << (BITDEPTH - 8),
    GRAIN_MIN = -GRAIN_CENTER,
    GRAIN_MAX = (256 << (BITDEPTH - 8)) - 1 - GRAIN_CENTER,
};

static void generate_grain_y(const Dav1dPicture *const in,
                             entry buf[GRAIN_HEIGHT][GRAIN_WIDTH])
{
    const Dav1dFilmGrainData *data = &in->p.film_grain;
    unsigned seed = data->seed;
    const int shift = 12 - BITDEPTH + data->grain_scale_shift;

    for (int y = 0; y < GRAIN_HEIGHT; y++) {
        for (int x = 0; x < GRAIN_WIDTH; x++) {
            const int value = get_random_number(11, &seed);
            buf[y][x] = round2(dav1d_gaussian_sequence[ value ], shift);
        }
    }

    const int ar_pad = 3;
    const int ar_lag = data->ar_coeff_lag;

    for (int y = ar_pad; y < GRAIN_HEIGHT; y++) {
        for (int x = ar_pad; x < GRAIN_WIDTH - ar_pad; x++) {
            const int8_t *coeff = data->ar_coeffs_y;
            int sum = 0;
            for (int dy = -ar_lag; dy <= 0; dy++) {
                for (int dx = -ar_lag; dx <= ar_lag; dx++) {
                    if (!dx && !dy)
                        break;
                    sum += *(coeff++) * buf[y + dy][x + dx];
                }
            }

            int grain = buf[y][x] + round2(sum, data->ar_coeff_shift);
            buf[y][x] = iclip(grain, GRAIN_MIN, GRAIN_MAX);
        }
    }
}

static void generate_grain_uv(const Dav1dPicture *const in, int uv,
                              entry buf[GRAIN_HEIGHT][GRAIN_WIDTH],
                              entry buf_y[GRAIN_HEIGHT][GRAIN_WIDTH])
{
    const Dav1dFilmGrainData *data = &in->p.film_grain;
    unsigned seed = data->seed ^ (uv ? 0x49d8 : 0xb524);
    const int shift = 12 - BITDEPTH + data->grain_scale_shift;

    const int subx = in->p.layout != DAV1D_PIXEL_LAYOUT_I444;
    const int suby = in->p.layout == DAV1D_PIXEL_LAYOUT_I420;

    const int chromaW = subx ? SUB_GRAIN_WIDTH  : GRAIN_WIDTH;
    const int chromaH = suby ? SUB_GRAIN_HEIGHT : GRAIN_HEIGHT;

    for (int y = 0; y < chromaH; y++) {
        for (int x = 0; x < chromaW; x++) {
            const int value = get_random_number(11, &seed);
            buf[y][x] = round2(dav1d_gaussian_sequence[ value ], shift);
        }
    }

    const int ar_pad = 3;
    const int ar_lag = data->ar_coeff_lag;

    for (int y = ar_pad; y < chromaH; y++) {
        for (int x = ar_pad; x < chromaW - ar_pad; x++) {
            const int8_t *coeff = data->ar_coeffs_uv[uv];
            int sum = 0;
            for (int dy = -ar_lag; dy <= 0; dy++) {
                for (int dx = -ar_lag; dx <= ar_lag; dx++) {
                    // For the final (current) pixel, we need to add in the
                    // contribution from the luma grain texture
                    if (!dx && !dy) {
                        if (!data->num_y_points)
                            break;
                        int luma = 0;
                        const int lumaX = ((x - ar_pad) << subx) + ar_pad;
                        const int lumaY = ((y - ar_pad) << suby) + ar_pad;
                        for (int i = 0; i <= suby; i++) {
                            for (int j = 0; j <= subx; j++) {
                                luma += buf_y[lumaY + i][lumaX + j];
                            }
                        }
                        luma = round2(luma, subx + suby);
                        sum += luma * (*coeff);
                        break;
                    }

                    sum += *(coeff++) * buf[y + dy][x + dx];
                }
            }

            const int grain = buf[y][x] + round2(sum, data->ar_coeff_shift);
            buf[y][x] = iclip(grain, GRAIN_MIN, GRAIN_MAX);
        }
    }
}

static void generate_scaling(const uint8_t points[][2], int num,
                             uint8_t scaling[SCALING_SIZE])
{
    const int shift_x = BITDEPTH - 8;

    // Fill up the preceding entries with the initial value
    for (int i = 0; i < points[0][0] << shift_x; i++)
        scaling[i] = points[0][1];

    // Linearly interpolate the values in the middle
    for (int i = 0; i < num - 1; i++) {
        const int bx = points[i][0] << shift_x;
        const int by = points[i][1];
        const int ex = points[i+1][0] << shift_x;
        const int ey = points[i+1][1];
        const int dx = ex - bx;
        const int dy = ey - by;
        const int delta = dy * ((0xFFFF + (dx >> 1))) / dx;
        for (int x = 0; x < dx; x++) {
            const int v = by + ((x * delta + 0x8000) >> 16);
            scaling[bx + x] = v;
        }
    }

    // Fill up the remaining entries with the final value
    for (int i = points[num - 1][0] << shift_x; i < SCALING_SIZE; i++)
        scaling[i] = points[num - 1][1];
}

// samples from the correct block of a grain LUT, while taking into account the
// offsets provided by the offsets cache
static inline entry sample_lut(entry grain_lut[GRAIN_HEIGHT][GRAIN_WIDTH],
                               int offsets[2][2], int subx, int suby,
                               int bx, int by, int x, int y)
{
    const int randval = offsets[bx][by];
    const int offx = 3 + (2 >> subx) * (3 + (randval >> 4));
    const int offy = 3 + (2 >> suby) * (3 + (randval & 0xF));
    return grain_lut[offy + y + (BLOCK_SIZE >> suby) * by]
                    [offx + x + (BLOCK_SIZE >> subx) * bx];
}

static void apply_to_row_y(Dav1dPicture *const out, const Dav1dPicture *const in,
                           entry grain_lut[GRAIN_HEIGHT][GRAIN_WIDTH],
                           uint8_t scaling[SCALING_SIZE], int row_num)
{
    const Dav1dFilmGrainData *const data = &out->p.film_grain;
    const int rows = 1 + (data->overlap_flag && row_num > 0);

    int min_value, max_value;
    if (data->clip_to_restricted_range) {
        min_value = 16 << (BITDEPTH - 8);
        max_value = 235 << (BITDEPTH - 8);
    } else {
        min_value = 0;
        max_value = (1 << BITDEPTH) - 1;
    }

    // seed[0] contains the current row, seed[1] contains the previous
    unsigned seed[2];
    for (int i = 0; i < rows; i++) {
        seed[i] = data->seed;
        seed[i] ^= (((row_num - i) * 37  + 178) & 0xFF) << 8;
        seed[i] ^= (((row_num - i) * 173 + 105) & 0xFF);
    }

    const ptrdiff_t stride = out->stride[0];
    assert(stride % (BLOCK_SIZE * sizeof(pixel)) == 0);
    assert(stride == in->stride[0]);
237
238
    pixel *const src_row = (pixel *)  in->data[0] + PXSTRIDE(stride) * row_num * BLOCK_SIZE;
    pixel *const dst_row = (pixel *) out->data[0] + PXSTRIDE(stride) * row_num * BLOCK_SIZE;
239
240
241
242
243

    // edge extend source pixels
    const int row_len = (out->p.w + BLOCK_SIZE - 1) & ~(BLOCK_SIZE - 1);
    for (int x = out->p.w; x < row_len; x++) {
        for (int y = 0; y < BLOCK_SIZE; y++) {
244
            pixel *src = src_row + y * PXSTRIDE(stride) + x;
245
246
247
248
249
250
            *src = 0;
        }
    }

    const int row_h = (row_num + 1) * BLOCK_SIZE;
    for (int y = out->p.h; y < row_h; y++)
251
        memset((pixel *) in->data[0] + PXSTRIDE(stride) * y, 0, row_len * sizeof(pixel));
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273

    int offsets[2 /* col offset */][2 /* row offset */];

    // process this row in BLOCK_SIZE^2 blocks
    for (int bx = 0; bx < out->p.w; bx += BLOCK_SIZE) {
        if (data->overlap_flag && bx) {
            // shift previous offsets left
            for (int i = 0; i < rows; i++)
                offsets[1][i] = offsets[0][i];
        }

        // update current offsets
        for (int i = 0; i < rows; i++)
            offsets[0][i] = get_random_number(8, &seed[i]);

        // x/y block offsets to compensate for overlapped regions
        const int ystart = data->overlap_flag && row_num ? 2 : 0;
        const int xstart = data->overlap_flag && bx      ? 2 : 0;

        static const int w[2][2] = { { 27, 17 }, { 17, 27 } };

#define add_noise_y(x, y, grain)                                                \
274
275
            pixel *src = src_row + (y) * PXSTRIDE(stride) + (bx + (x));         \
            pixel *dst = dst_row + (y) * PXSTRIDE(stride) + (bx + (x));         \
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
            int noise = round2(scaling[ *src ] * (grain), data->scaling_shift); \
            *dst = iclip(*src + noise, min_value, max_value);

        for (int y = ystart; y < BLOCK_SIZE; y++) {
            // Non-overlapped image region (straightforward)
            for (int x = xstart; x < BLOCK_SIZE; x++) {
                int grain = sample_lut(grain_lut, offsets, 0, 0, 0, 0, x, y);
                add_noise_y(x, y, grain);
            }

            // Special case for overlapped column
            for (int x = 0; x < xstart; x++) {
                int grain = sample_lut(grain_lut, offsets, 0, 0, 0, 0, x, y);
                int old   = sample_lut(grain_lut, offsets, 0, 0, 1, 0, x, y);
                grain = round2(old * w[x][0] + grain * w[x][1], 5);
                grain = iclip(grain, GRAIN_MIN, GRAIN_MAX);
                add_noise_y(x, y, grain);
            }
        }

        for (int y = 0; y < ystart; y++) {
            // Special case for overlapped row (sans corner)
            for (int x = xstart; x < BLOCK_SIZE; x++) {
                int grain = sample_lut(grain_lut, offsets, 0, 0, 0, 0, x, y);
                int old   = sample_lut(grain_lut, offsets, 0, 0, 0, 1, x, y);
                grain = round2(old * w[y][0] + grain * w[y][1], 5);
                grain = iclip(grain, GRAIN_MIN, GRAIN_MAX);
                add_noise_y(x, y, grain);
            }

            // Special case for doubly-overlapped corner
            for (int x = 0; x < xstart; x++) {
                // Blend the top pixel with the top left block
                int top = sample_lut(grain_lut, offsets, 0, 0, 0, 1, x, y);
                int old = sample_lut(grain_lut, offsets, 0, 0, 1, 1, x, y);
                top = round2(old * w[x][0] + top * w[x][1], 5);
                top = iclip(top, GRAIN_MIN, GRAIN_MAX);

                // Blend the current pixel with the left block
                int grain = sample_lut(grain_lut, offsets, 0, 0, 0, 0, x, y);
                old = sample_lut(grain_lut, offsets, 0, 0, 1, 0, x, y);
                grain = round2(old * w[x][0] + grain * w[x][1], 5);
                grain = iclip(grain, GRAIN_MIN, GRAIN_MAX);

                // Mix the row rows together and apply grain
                grain = round2(top * w[y][0] + grain * w[y][1], 5);
                grain = iclip(grain, GRAIN_MIN, GRAIN_MAX);
                add_noise_y(x, y, grain);
            }
        }
    }
}

static void apply_to_row_uv(Dav1dPicture *const out, const Dav1dPicture *const in,
                            entry grain_lut[GRAIN_HEIGHT][GRAIN_WIDTH],
                            uint8_t scaling[SCALING_SIZE], int uv, int row_num)
{
    const Dav1dFilmGrainData *const data = &out->p.film_grain;
    const int rows = 1 + (data->overlap_flag && row_num > 0);

    int min_value, max_value;
    if (data->clip_to_restricted_range) {
        min_value = 16 << (BITDEPTH - 8);
        if (out->p.mtrx == DAV1D_MC_IDENTITY) {
            max_value = 235 << (BITDEPTH - 8);
        } else {
            max_value = 240 << (BITDEPTH - 8);
        }
    } else {
        min_value = 0;
        max_value = (1 << BITDEPTH) - 1;
    }

    const int sx = in->p.layout != DAV1D_PIXEL_LAYOUT_I444;
    const int sy = in->p.layout == DAV1D_PIXEL_LAYOUT_I420;

    // seed[0] contains the current row, seed[1] contains the previous
    unsigned seed[2];
    for (int i = 0; i < rows; i++) {
        seed[i] = data->seed;
        seed[i] ^= (((row_num - i) * 37  + 178) & 0xFF) << 8;
        seed[i] ^= (((row_num - i) * 173 + 105) & 0xFF);
    }

    const ptrdiff_t stride = out->stride[1];
    assert(stride % (BLOCK_SIZE * sizeof(pixel)) == 0);
    assert(stride == in->stride[1]);

    const int by = row_num * (BLOCK_SIZE >> sy);
365
366
367
    pixel *const dst_row = (pixel *) out->data[1 + uv] + PXSTRIDE(stride) * by;
    pixel *const src_row = (pixel *)  in->data[1 + uv] + PXSTRIDE(stride) * by;
    pixel *const luma_row = (pixel *) out->data[0] + PXSTRIDE(out->stride[0]) * row_num * BLOCK_SIZE;
368
369

    // edge extend source pixels
370
    const int row_len = (((out->p.w + sx) >> sx) + (BLOCK_SIZE >> sx) - 1)
371
                        & ~((BLOCK_SIZE >> sx) - 1);
372
    for (int x = (out->p.w + sx) >> sx; x < row_len; x++) {
373
        for (int y = 0; y < BLOCK_SIZE >> sy; y++) {
374
            pixel *src = src_row + y * PXSTRIDE(stride) + x;
375
376
377
378
379
            *src = 0;
        }
    }

    const int row_h = (row_num + 1) * (BLOCK_SIZE >> sy);
380
    for (int y = (out->p.h + sy) >> sy; y < row_h; y++)
381
        memset((pixel *) in->data[1 + uv] + PXSTRIDE(stride) * y, 0, row_len * sizeof(pixel));
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

    int offsets[2 /* col offset */][2 /* row offset */];

    // process this row in BLOCK_SIZE^2 blocks (subsampled)
    for (int bx = 0; bx < (out->p.w + sx) >> sx; bx += BLOCK_SIZE >> sx) {
        if (data->overlap_flag && bx) {
            // shift previous offsets left
            for (int i = 0; i < rows; i++)
                offsets[1][i] = offsets[0][i];
        }

        // update current offsets
        for (int i = 0; i < rows; i++)
            offsets[0][i] = get_random_number(8, &seed[i]);

        // x/y block offsets to compensate for overlapped regions
        const int ystart = data->overlap_flag && row_num ? (2 >> sy) : 0;
        const int xstart = data->overlap_flag && bx      ? (2 >> sx) : 0;

        static const int w[2 /* sub */][2 /* off */][2] = {
            { { 27, 17 }, { 17, 27 } },
            { { 23, 22 } },
        };

#define add_noise_uv(x, y, grain)                                               \
            const int lx = (bx + x) << sx;                                      \
            const int ly = y << sy;                                             \
409
            pixel *luma = luma_row + ly * PXSTRIDE(out->stride[0]) + lx;        \
410
411
412
413
            pixel avg = luma[0];                                                \
            if (sx && lx + 1 < out->p.w)                                        \
                avg = (avg + luma[1] + 1) >> 1;                                 \
                                                                                \
414
415
            pixel *src = src_row + (y) * PXSTRIDE(stride) + (bx + (x));         \
            pixel *dst = dst_row + (y) * PXSTRIDE(stride) + (bx + (x));         \
416
417
418
419
420
            int val = avg;                                                      \
            if (!data->chroma_scaling_from_luma) {                              \
                int combined = avg * data->uv_luma_mult[uv] +                   \
                               *src * data->uv_mult[uv];                        \
                val = iclip_pixel( (combined >> 6) +                            \
421
                                   (data->uv_offset[uv] * (1 << (BITDEPTH - 8))) );   \
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
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
            }                                                                   \
                                                                                \
            int noise = round2(scaling[ val ] * (grain), data->scaling_shift);  \
            *dst = iclip(*src + noise, min_value, max_value);

        for (int y = ystart; y < BLOCK_SIZE >> sy; y++) {
            // Non-overlapped image region (straightforward)
            for (int x = xstart; x < BLOCK_SIZE >> sx; x++) {
                int grain = sample_lut(grain_lut, offsets, sx, sy, 0, 0, x, y);
                add_noise_uv(x, y, grain);
            }

            // Special case for overlapped column
            for (int x = 0; x < xstart; x++) {
                int grain = sample_lut(grain_lut, offsets, sx, sy, 0, 0, x, y);
                int old   = sample_lut(grain_lut, offsets, sx, sy, 1, 0, x, y);
                grain = (old * w[sx][x][0] + grain * w[sx][x][1] + 16) >> 5;
                grain = iclip(grain, GRAIN_MIN, GRAIN_MAX);
                add_noise_uv(x, y, grain);
            }
        }

        for (int y = 0; y < ystart; y++) {
            // Special case for overlapped row (sans corner)
            for (int x = xstart; x < BLOCK_SIZE >> sx; x++) {
                int grain = sample_lut(grain_lut, offsets, sx, sy, 0, 0, x, y);
                int old   = sample_lut(grain_lut, offsets, sx, sy, 0, 1, x, y);
                grain = (old * w[sy][y][0] + grain * w[sy][y][1] + 16) >> 5;
                grain = iclip(grain, GRAIN_MIN, GRAIN_MAX);
                add_noise_uv(x, y, grain);
            }

            // Special case for doubly-overlapped corner
            for (int x = 0; x < xstart; x++) {
                // Blend the top pixel with the top left block
                int top = sample_lut(grain_lut, offsets, sx, sy, 0, 1, x, y);
                int old = sample_lut(grain_lut, offsets, sx, sy, 1, 1, x, y);
                top = (old * w[sx][x][0] + top * w[sx][x][1] + 16) >> 5;
                top = iclip(top, GRAIN_MIN, GRAIN_MAX);

                // Blend the current pixel with the left block
                int grain = sample_lut(grain_lut, offsets, sx, sy, 0, 0, x, y);
                old = sample_lut(grain_lut, offsets, sx, sy, 1, 0, x, y);
                grain = (old * w[sx][x][0] + grain * w[sx][x][1] + 16) >> 5;
                grain = iclip(grain, GRAIN_MIN, GRAIN_MAX);

                // Mix the row rows together and apply to image
                grain = (top * w[sy][y][0] + grain * w[sy][y][1] + 16) >> 5;
                grain = iclip(grain, GRAIN_MIN, GRAIN_MAX);
                add_noise_uv(x, y, grain);
            }
        }
    }
}

void bitfn(dav1d_apply_grain)(Dav1dPicture *const out,
                              const Dav1dPicture *const in)
{
    const Dav1dFilmGrainData *const data = &out->p.film_grain;

    entry grain_lut[3][GRAIN_HEIGHT][GRAIN_WIDTH];
    uint8_t scaling[3][SCALING_SIZE];

    // Generate grain LUTs as needed
    generate_grain_y(out, grain_lut[0]); // always needed
    if (data->num_uv_points[0] || data->chroma_scaling_from_luma)
        generate_grain_uv(out, 0, grain_lut[1], grain_lut[0]);
    if (data->num_uv_points[1] || data->chroma_scaling_from_luma)
        generate_grain_uv(out, 1, grain_lut[2], grain_lut[0]);

    // Generate scaling LUTs as needed
    if (data->num_y_points)
        generate_scaling(data->y_points, data->num_y_points, scaling[0]);
    if (data->num_uv_points[0])
        generate_scaling(data->uv_points[0], data->num_uv_points[0], scaling[1]);
    if (data->num_uv_points[1])
        generate_scaling(data->uv_points[1], data->num_uv_points[1], scaling[2]);

    // Synthesize grain for the affected planes
    int rows = (out->p.h + 16) >> 5;
    for (int row = 0; row < rows; row++) {
        if (data->num_y_points)
            apply_to_row_y(out, in, grain_lut[0], scaling[0], row);

        if (data->chroma_scaling_from_luma) {
            apply_to_row_uv(out, in, grain_lut[1], scaling[0], 0, row);
            apply_to_row_uv(out, in, grain_lut[2], scaling[0], 1, row);
        } else {
            if (data->num_uv_points[0])
                apply_to_row_uv(out, in, grain_lut[1], scaling[1], 0, row);
            if (data->num_uv_points[1])
                apply_to_row_uv(out, in, grain_lut[2], scaling[2], 1, row);
        }
    }

    // Copy over the non-modified planes
    // TODO: eliminate in favor of per-plane refs
    if (!data->num_y_points) {
        assert(out->stride[0] == in->stride[0]);
        memcpy(out->data[0], in->data[0], out->p.h * out->stride[0]);
    }

    for (int i = 0; i < 2; i++) {
        if (!data->num_uv_points[i] && !data->chroma_scaling_from_luma) {
            const int suby = in->p.layout == DAV1D_PIXEL_LAYOUT_I420;
            assert(out->stride[1] == in->stride[1]);
            memcpy(out->data[1+i], in->data[1+i],
                   (out->p.h >> suby) * out->stride[1]);
        }
    }
}