xref: /third_party/ffmpeg/libavcodec/vc2enc.c (revision cabdff1a)
1/*
2 * Copyright (C) 2016 Open Broadcast Systems Ltd.
3 * Author        2016 Rostislav Pehlivanov <atomnuker@gmail.com>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22#include "libavutil/pixdesc.h"
23#include "libavutil/opt.h"
24#include "libavutil/version.h"
25#include "codec_internal.h"
26#include "dirac.h"
27#include "encode.h"
28#include "put_bits.h"
29#include "version.h"
30
31#include "vc2enc_dwt.h"
32#include "diractab.h"
33
34/* The limited size resolution of each slice forces us to do this */
35#define SSIZE_ROUND(b) (FFALIGN((b), s->size_scaler) + 4 + s->prefix_bytes)
36
37/* Decides the cutoff point in # of slices to distribute the leftover bytes */
38#define SLICE_REDIST_TOTAL 150
39
40typedef struct VC2BaseVideoFormat {
41    enum AVPixelFormat pix_fmt;
42    AVRational time_base;
43    int width, height, interlaced, level;
44    const char *name;
45} VC2BaseVideoFormat;
46
47static const VC2BaseVideoFormat base_video_fmts[] = {
48    { 0 }, /* Custom format, here just to make indexing equal to base_vf */
49    { AV_PIX_FMT_YUV420P,   { 1001, 15000 },  176,  120, 0, 1,     "QSIF525" },
50    { AV_PIX_FMT_YUV420P,   {    2,    25 },  176,  144, 0, 1,     "QCIF"    },
51    { AV_PIX_FMT_YUV420P,   { 1001, 15000 },  352,  240, 0, 1,     "SIF525"  },
52    { AV_PIX_FMT_YUV420P,   {    2,    25 },  352,  288, 0, 1,     "CIF"     },
53    { AV_PIX_FMT_YUV420P,   { 1001, 15000 },  704,  480, 0, 1,     "4SIF525" },
54    { AV_PIX_FMT_YUV420P,   {    2,    25 },  704,  576, 0, 1,     "4CIF"    },
55
56    { AV_PIX_FMT_YUV422P10, { 1001, 30000 },  720,  480, 1, 2,   "SD480I-60" },
57    { AV_PIX_FMT_YUV422P10, {    1,    25 },  720,  576, 1, 2,   "SD576I-50" },
58
59    { AV_PIX_FMT_YUV422P10, { 1001, 60000 }, 1280,  720, 0, 3,  "HD720P-60"  },
60    { AV_PIX_FMT_YUV422P10, {    1,    50 }, 1280,  720, 0, 3,  "HD720P-50"  },
61    { AV_PIX_FMT_YUV422P10, { 1001, 30000 }, 1920, 1080, 1, 3,  "HD1080I-60" },
62    { AV_PIX_FMT_YUV422P10, {    1,    25 }, 1920, 1080, 1, 3,  "HD1080I-50" },
63    { AV_PIX_FMT_YUV422P10, { 1001, 60000 }, 1920, 1080, 0, 3,  "HD1080P-60" },
64    { AV_PIX_FMT_YUV422P10, {    1,    50 }, 1920, 1080, 0, 3,  "HD1080P-50" },
65
66    { AV_PIX_FMT_YUV444P12, {    1,    24 }, 2048, 1080, 0, 4,        "DC2K" },
67    { AV_PIX_FMT_YUV444P12, {    1,    24 }, 4096, 2160, 0, 5,        "DC4K" },
68
69    { AV_PIX_FMT_YUV422P10, { 1001, 60000 }, 3840, 2160, 0, 6, "UHDTV 4K-60" },
70    { AV_PIX_FMT_YUV422P10, {    1,    50 }, 3840, 2160, 0, 6, "UHDTV 4K-50" },
71
72    { AV_PIX_FMT_YUV422P10, { 1001, 60000 }, 7680, 4320, 0, 7, "UHDTV 8K-60" },
73    { AV_PIX_FMT_YUV422P10, {    1,    50 }, 7680, 4320, 0, 7, "UHDTV 8K-50" },
74
75    { AV_PIX_FMT_YUV422P10, { 1001, 24000 }, 1920, 1080, 0, 3,  "HD1080P-24" },
76    { AV_PIX_FMT_YUV422P10, { 1001, 30000 },  720,  486, 1, 2,  "SD Pro486"  },
77};
78static const int base_video_fmts_len = FF_ARRAY_ELEMS(base_video_fmts);
79
80enum VC2_QM {
81    VC2_QM_DEF = 0,
82    VC2_QM_COL,
83    VC2_QM_FLAT,
84
85    VC2_QM_NB
86};
87
88typedef struct SubBand {
89    dwtcoef *buf;
90    ptrdiff_t stride;
91    int width;
92    int height;
93} SubBand;
94
95typedef struct Plane {
96    SubBand band[MAX_DWT_LEVELS][4];
97    dwtcoef *coef_buf;
98    int width;
99    int height;
100    int dwt_width;
101    int dwt_height;
102    ptrdiff_t coef_stride;
103} Plane;
104
105typedef struct SliceArgs {
106    PutBitContext pb;
107    int cache[DIRAC_MAX_QUANT_INDEX];
108    void *ctx;
109    int x;
110    int y;
111    int quant_idx;
112    int bits_ceil;
113    int bits_floor;
114    int bytes;
115} SliceArgs;
116
117typedef struct TransformArgs {
118    void *ctx;
119    Plane *plane;
120    void *idata;
121    ptrdiff_t istride;
122    int field;
123    VC2TransformContext t;
124} TransformArgs;
125
126typedef struct VC2EncContext {
127    AVClass *av_class;
128    PutBitContext pb;
129    Plane plane[3];
130    AVCodecContext *avctx;
131    DiracVersionInfo ver;
132
133    SliceArgs *slice_args;
134    TransformArgs transform_args[3];
135
136    /* For conversion from unsigned pixel values to signed */
137    int diff_offset;
138    int bpp;
139    int bpp_idx;
140
141    /* Picture number */
142    uint32_t picture_number;
143
144    /* Base video format */
145    int base_vf;
146    int level;
147    int profile;
148
149    /* Quantization matrix */
150    uint8_t quant[MAX_DWT_LEVELS][4];
151    int custom_quant_matrix;
152
153    /* Division LUT */
154    uint32_t qmagic_lut[116][2];
155
156    int num_x; /* #slices horizontally */
157    int num_y; /* #slices vertically */
158    int prefix_bytes;
159    int size_scaler;
160    int chroma_x_shift;
161    int chroma_y_shift;
162
163    /* Rate control stuff */
164    int frame_max_bytes;
165    int slice_max_bytes;
166    int slice_min_bytes;
167    int q_ceil;
168    int q_avg;
169
170    /* Options */
171    double tolerance;
172    int wavelet_idx;
173    int wavelet_depth;
174    int strict_compliance;
175    int slice_height;
176    int slice_width;
177    int interlaced;
178    enum VC2_QM quant_matrix;
179
180    /* Parse code state */
181    uint32_t next_parse_offset;
182    enum DiracParseCodes last_parse_code;
183} VC2EncContext;
184
185static av_always_inline void put_vc2_ue_uint(PutBitContext *pb, uint32_t val)
186{
187    int i;
188    int pbits = 0, bits = 0, topbit = 1, maxval = 1;
189
190    if (!val++) {
191        put_bits(pb, 1, 1);
192        return;
193    }
194
195    while (val > maxval) {
196        topbit <<= 1;
197        maxval <<= 1;
198        maxval |=  1;
199    }
200
201    bits = ff_log2(topbit);
202
203    for (i = 0; i < bits; i++) {
204        topbit >>= 1;
205        pbits <<= 2;
206        if (val & topbit)
207            pbits |= 0x1;
208    }
209
210    put_bits(pb, bits*2 + 1, (pbits << 1) | 1);
211}
212
213static av_always_inline int count_vc2_ue_uint(uint32_t val)
214{
215    int topbit = 1, maxval = 1;
216
217    if (!val++)
218        return 1;
219
220    while (val > maxval) {
221        topbit <<= 1;
222        maxval <<= 1;
223        maxval |=  1;
224    }
225
226    return ff_log2(topbit)*2 + 1;
227}
228
229/* VC-2 10.4 - parse_info() */
230static void encode_parse_info(VC2EncContext *s, enum DiracParseCodes pcode)
231{
232    uint32_t cur_pos, dist;
233
234    align_put_bits(&s->pb);
235
236    cur_pos = put_bits_count(&s->pb) >> 3;
237
238    /* Magic string */
239    ff_put_string(&s->pb, "BBCD", 0);
240
241    /* Parse code */
242    put_bits(&s->pb, 8, pcode);
243
244    /* Next parse offset */
245    dist = cur_pos - s->next_parse_offset;
246    AV_WB32(s->pb.buf + s->next_parse_offset + 5, dist);
247    s->next_parse_offset = cur_pos;
248    put_bits32(&s->pb, pcode == DIRAC_PCODE_END_SEQ ? 13 : 0);
249
250    /* Last parse offset */
251    put_bits32(&s->pb, s->last_parse_code == DIRAC_PCODE_END_SEQ ? 13 : dist);
252
253    s->last_parse_code = pcode;
254}
255
256/* VC-2 11.1 - parse_parameters()
257 * The level dictates what the decoder should expect in terms of resolution
258 * and allows it to quickly reject whatever it can't support. Remember,
259 * this codec kinda targets cheapo FPGAs without much memory. Unfortunately
260 * it also limits us greatly in our choice of formats, hence the flag to disable
261 * strict_compliance */
262static void encode_parse_params(VC2EncContext *s)
263{
264    put_vc2_ue_uint(&s->pb, s->ver.major); /* VC-2 demands this to be 2 */
265    put_vc2_ue_uint(&s->pb, s->ver.minor); /* ^^ and this to be 0       */
266    put_vc2_ue_uint(&s->pb, s->profile);   /* 3 to signal HQ profile    */
267    put_vc2_ue_uint(&s->pb, s->level);     /* 3 - 1080/720, 6 - 4K      */
268}
269
270/* VC-2 11.3 - frame_size() */
271static void encode_frame_size(VC2EncContext *s)
272{
273    put_bits(&s->pb, 1, !s->strict_compliance);
274    if (!s->strict_compliance) {
275        AVCodecContext *avctx = s->avctx;
276        put_vc2_ue_uint(&s->pb, avctx->width);
277        put_vc2_ue_uint(&s->pb, avctx->height);
278    }
279}
280
281/* VC-2 11.3.3 - color_diff_sampling_format() */
282static void encode_sample_fmt(VC2EncContext *s)
283{
284    put_bits(&s->pb, 1, !s->strict_compliance);
285    if (!s->strict_compliance) {
286        int idx;
287        if (s->chroma_x_shift == 1 && s->chroma_y_shift == 0)
288            idx = 1; /* 422 */
289        else if (s->chroma_x_shift == 1 && s->chroma_y_shift == 1)
290            idx = 2; /* 420 */
291        else
292            idx = 0; /* 444 */
293        put_vc2_ue_uint(&s->pb, idx);
294    }
295}
296
297/* VC-2 11.3.4 - scan_format() */
298static void encode_scan_format(VC2EncContext *s)
299{
300    put_bits(&s->pb, 1, !s->strict_compliance);
301    if (!s->strict_compliance)
302        put_vc2_ue_uint(&s->pb, s->interlaced);
303}
304
305/* VC-2 11.3.5 - frame_rate() */
306static void encode_frame_rate(VC2EncContext *s)
307{
308    put_bits(&s->pb, 1, !s->strict_compliance);
309    if (!s->strict_compliance) {
310        AVCodecContext *avctx = s->avctx;
311        put_vc2_ue_uint(&s->pb, 0);
312        put_vc2_ue_uint(&s->pb, avctx->time_base.den);
313        put_vc2_ue_uint(&s->pb, avctx->time_base.num);
314    }
315}
316
317/* VC-2 11.3.6 - aspect_ratio() */
318static void encode_aspect_ratio(VC2EncContext *s)
319{
320    put_bits(&s->pb, 1, !s->strict_compliance);
321    if (!s->strict_compliance) {
322        AVCodecContext *avctx = s->avctx;
323        put_vc2_ue_uint(&s->pb, 0);
324        put_vc2_ue_uint(&s->pb, avctx->sample_aspect_ratio.num);
325        put_vc2_ue_uint(&s->pb, avctx->sample_aspect_ratio.den);
326    }
327}
328
329/* VC-2 11.3.7 - clean_area() */
330static void encode_clean_area(VC2EncContext *s)
331{
332    put_bits(&s->pb, 1, 0);
333}
334
335/* VC-2 11.3.8 - signal_range() */
336static void encode_signal_range(VC2EncContext *s)
337{
338    put_bits(&s->pb, 1, !s->strict_compliance);
339    if (!s->strict_compliance)
340        put_vc2_ue_uint(&s->pb, s->bpp_idx);
341}
342
343/* VC-2 11.3.9 - color_spec() */
344static void encode_color_spec(VC2EncContext *s)
345{
346    AVCodecContext *avctx = s->avctx;
347    put_bits(&s->pb, 1, !s->strict_compliance);
348    if (!s->strict_compliance) {
349        int val;
350        put_vc2_ue_uint(&s->pb, 0);
351
352        /* primaries */
353        put_bits(&s->pb, 1, 1);
354        if (avctx->color_primaries == AVCOL_PRI_BT470BG)
355            val = 2;
356        else if (avctx->color_primaries == AVCOL_PRI_SMPTE170M)
357            val = 1;
358        else if (avctx->color_primaries == AVCOL_PRI_SMPTE240M)
359            val = 1;
360        else
361            val = 0;
362        put_vc2_ue_uint(&s->pb, val);
363
364        /* color matrix */
365        put_bits(&s->pb, 1, 1);
366        if (avctx->colorspace == AVCOL_SPC_RGB)
367            val = 3;
368        else if (avctx->colorspace == AVCOL_SPC_YCOCG)
369            val = 2;
370        else if (avctx->colorspace == AVCOL_SPC_BT470BG)
371            val = 1;
372        else
373            val = 0;
374        put_vc2_ue_uint(&s->pb, val);
375
376        /* transfer function */
377        put_bits(&s->pb, 1, 1);
378        if (avctx->color_trc == AVCOL_TRC_LINEAR)
379            val = 2;
380        else if (avctx->color_trc == AVCOL_TRC_BT1361_ECG)
381            val = 1;
382        else
383            val = 0;
384        put_vc2_ue_uint(&s->pb, val);
385    }
386}
387
388/* VC-2 11.3 - source_parameters() */
389static void encode_source_params(VC2EncContext *s)
390{
391    encode_frame_size(s);
392    encode_sample_fmt(s);
393    encode_scan_format(s);
394    encode_frame_rate(s);
395    encode_aspect_ratio(s);
396    encode_clean_area(s);
397    encode_signal_range(s);
398    encode_color_spec(s);
399}
400
401/* VC-2 11 - sequence_header() */
402static void encode_seq_header(VC2EncContext *s)
403{
404    align_put_bits(&s->pb);
405    encode_parse_params(s);
406    put_vc2_ue_uint(&s->pb, s->base_vf);
407    encode_source_params(s);
408    put_vc2_ue_uint(&s->pb, s->interlaced); /* Frames or fields coding */
409}
410
411/* VC-2 12.1 - picture_header() */
412static void encode_picture_header(VC2EncContext *s)
413{
414    align_put_bits(&s->pb);
415    put_bits32(&s->pb, s->picture_number++);
416}
417
418/* VC-2 12.3.4.1 - slice_parameters() */
419static void encode_slice_params(VC2EncContext *s)
420{
421    put_vc2_ue_uint(&s->pb, s->num_x);
422    put_vc2_ue_uint(&s->pb, s->num_y);
423    put_vc2_ue_uint(&s->pb, s->prefix_bytes);
424    put_vc2_ue_uint(&s->pb, s->size_scaler);
425}
426
427/* 1st idx = LL, second - vertical, third - horizontal, fourth - total */
428static const uint8_t vc2_qm_col_tab[][4] = {
429    {20,  9, 15,  4},
430    { 0,  6,  6,  4},
431    { 0,  3,  3,  5},
432    { 0,  3,  5,  1},
433    { 0, 11, 10, 11}
434};
435
436static const uint8_t vc2_qm_flat_tab[][4] = {
437    { 0,  0,  0,  0},
438    { 0,  0,  0,  0},
439    { 0,  0,  0,  0},
440    { 0,  0,  0,  0},
441    { 0,  0,  0,  0}
442};
443
444static void init_quant_matrix(VC2EncContext *s)
445{
446    int level, orientation;
447
448    if (s->wavelet_depth <= 4 && s->quant_matrix == VC2_QM_DEF) {
449        s->custom_quant_matrix = 0;
450        for (level = 0; level < s->wavelet_depth; level++) {
451            s->quant[level][0] = ff_dirac_default_qmat[s->wavelet_idx][level][0];
452            s->quant[level][1] = ff_dirac_default_qmat[s->wavelet_idx][level][1];
453            s->quant[level][2] = ff_dirac_default_qmat[s->wavelet_idx][level][2];
454            s->quant[level][3] = ff_dirac_default_qmat[s->wavelet_idx][level][3];
455        }
456        return;
457    }
458
459    s->custom_quant_matrix = 1;
460
461    if (s->quant_matrix == VC2_QM_DEF) {
462        for (level = 0; level < s->wavelet_depth; level++) {
463            for (orientation = 0; orientation < 4; orientation++) {
464                if (level <= 3)
465                    s->quant[level][orientation] = ff_dirac_default_qmat[s->wavelet_idx][level][orientation];
466                else
467                    s->quant[level][orientation] = vc2_qm_col_tab[level][orientation];
468            }
469        }
470    } else if (s->quant_matrix == VC2_QM_COL) {
471        for (level = 0; level < s->wavelet_depth; level++) {
472            for (orientation = 0; orientation < 4; orientation++) {
473                s->quant[level][orientation] = vc2_qm_col_tab[level][orientation];
474            }
475        }
476    } else {
477        for (level = 0; level < s->wavelet_depth; level++) {
478            for (orientation = 0; orientation < 4; orientation++) {
479                s->quant[level][orientation] = vc2_qm_flat_tab[level][orientation];
480            }
481        }
482    }
483}
484
485/* VC-2 12.3.4.2 - quant_matrix() */
486static void encode_quant_matrix(VC2EncContext *s)
487{
488    int level;
489    put_bits(&s->pb, 1, s->custom_quant_matrix);
490    if (s->custom_quant_matrix) {
491        put_vc2_ue_uint(&s->pb, s->quant[0][0]);
492        for (level = 0; level < s->wavelet_depth; level++) {
493            put_vc2_ue_uint(&s->pb, s->quant[level][1]);
494            put_vc2_ue_uint(&s->pb, s->quant[level][2]);
495            put_vc2_ue_uint(&s->pb, s->quant[level][3]);
496        }
497    }
498}
499
500/* VC-2 12.3 - transform_parameters() */
501static void encode_transform_params(VC2EncContext *s)
502{
503    put_vc2_ue_uint(&s->pb, s->wavelet_idx);
504    put_vc2_ue_uint(&s->pb, s->wavelet_depth);
505
506    encode_slice_params(s);
507    encode_quant_matrix(s);
508}
509
510/* VC-2 12.2 - wavelet_transform() */
511static void encode_wavelet_transform(VC2EncContext *s)
512{
513    encode_transform_params(s);
514    align_put_bits(&s->pb);
515}
516
517/* VC-2 12 - picture_parse() */
518static void encode_picture_start(VC2EncContext *s)
519{
520    align_put_bits(&s->pb);
521    encode_picture_header(s);
522    align_put_bits(&s->pb);
523    encode_wavelet_transform(s);
524}
525
526#define QUANT(c, mul, add, shift) (((mul) * (c) + (add)) >> (shift))
527
528/* VC-2 13.5.5.2 - slice_band() */
529static void encode_subband(VC2EncContext *s, PutBitContext *pb, int sx, int sy,
530                           SubBand *b, int quant)
531{
532    int x, y;
533
534    const int left   = b->width  * (sx+0) / s->num_x;
535    const int right  = b->width  * (sx+1) / s->num_x;
536    const int top    = b->height * (sy+0) / s->num_y;
537    const int bottom = b->height * (sy+1) / s->num_y;
538
539    dwtcoef *coeff = b->buf + top * b->stride;
540    const uint64_t q_m = ((uint64_t)(s->qmagic_lut[quant][0])) << 2;
541    const uint64_t q_a = s->qmagic_lut[quant][1];
542    const int q_s = av_log2(ff_dirac_qscale_tab[quant]) + 32;
543
544    for (y = top; y < bottom; y++) {
545        for (x = left; x < right; x++) {
546            uint32_t c_abs = QUANT(FFABS(coeff[x]), q_m, q_a, q_s);
547            put_vc2_ue_uint(pb, c_abs);
548            if (c_abs)
549                put_bits(pb, 1, coeff[x] < 0);
550        }
551        coeff += b->stride;
552    }
553}
554
555static int count_hq_slice(SliceArgs *slice, int quant_idx)
556{
557    int x, y;
558    uint8_t quants[MAX_DWT_LEVELS][4];
559    int bits = 0, p, level, orientation;
560    VC2EncContext *s = slice->ctx;
561
562    if (slice->cache[quant_idx])
563        return slice->cache[quant_idx];
564
565    bits += 8*s->prefix_bytes;
566    bits += 8; /* quant_idx */
567
568    for (level = 0; level < s->wavelet_depth; level++)
569        for (orientation = !!level; orientation < 4; orientation++)
570            quants[level][orientation] = FFMAX(quant_idx - s->quant[level][orientation], 0);
571
572    for (p = 0; p < 3; p++) {
573        int bytes_start, bytes_len, pad_s, pad_c;
574        bytes_start = bits >> 3;
575        bits += 8;
576        for (level = 0; level < s->wavelet_depth; level++) {
577            for (orientation = !!level; orientation < 4; orientation++) {
578                SubBand *b = &s->plane[p].band[level][orientation];
579
580                const int q_idx = quants[level][orientation];
581                const uint64_t q_m = ((uint64_t)s->qmagic_lut[q_idx][0]) << 2;
582                const uint64_t q_a = s->qmagic_lut[q_idx][1];
583                const int q_s = av_log2(ff_dirac_qscale_tab[q_idx]) + 32;
584
585                const int left   = b->width  * slice->x    / s->num_x;
586                const int right  = b->width  *(slice->x+1) / s->num_x;
587                const int top    = b->height * slice->y    / s->num_y;
588                const int bottom = b->height *(slice->y+1) / s->num_y;
589
590                dwtcoef *buf = b->buf + top * b->stride;
591
592                for (y = top; y < bottom; y++) {
593                    for (x = left; x < right; x++) {
594                        uint32_t c_abs = QUANT(FFABS(buf[x]), q_m, q_a, q_s);
595                        bits += count_vc2_ue_uint(c_abs);
596                        bits += !!c_abs;
597                    }
598                    buf += b->stride;
599                }
600            }
601        }
602        bits += FFALIGN(bits, 8) - bits;
603        bytes_len = (bits >> 3) - bytes_start - 1;
604        pad_s = FFALIGN(bytes_len, s->size_scaler)/s->size_scaler;
605        pad_c = (pad_s*s->size_scaler) - bytes_len;
606        bits += pad_c*8;
607    }
608
609    slice->cache[quant_idx] = bits;
610
611    return bits;
612}
613
614/* Approaches the best possible quantizer asymptotically, its kinda exaustive
615 * but we have a LUT to get the coefficient size in bits. Guaranteed to never
616 * overshoot, which is apparently very important when streaming */
617static int rate_control(AVCodecContext *avctx, void *arg)
618{
619    SliceArgs *slice_dat = arg;
620    VC2EncContext *s = slice_dat->ctx;
621    const int top = slice_dat->bits_ceil;
622    const int bottom = slice_dat->bits_floor;
623    int quant_buf[2] = {-1, -1};
624    int quant = slice_dat->quant_idx, step = 1;
625    int bits_last, bits = count_hq_slice(slice_dat, quant);
626    while ((bits > top) || (bits < bottom)) {
627        const int signed_step = bits > top ? +step : -step;
628        quant  = av_clip(quant + signed_step, 0, s->q_ceil-1);
629        bits   = count_hq_slice(slice_dat, quant);
630        if (quant_buf[1] == quant) {
631            quant = FFMAX(quant_buf[0], quant);
632            bits  = quant == quant_buf[0] ? bits_last : bits;
633            break;
634        }
635        step         = av_clip(step/2, 1, (s->q_ceil-1)/2);
636        quant_buf[1] = quant_buf[0];
637        quant_buf[0] = quant;
638        bits_last    = bits;
639    }
640    slice_dat->quant_idx = av_clip(quant, 0, s->q_ceil-1);
641    slice_dat->bytes = SSIZE_ROUND(bits >> 3);
642    return 0;
643}
644
645static int calc_slice_sizes(VC2EncContext *s)
646{
647    int i, j, slice_x, slice_y, bytes_left = 0;
648    int bytes_top[SLICE_REDIST_TOTAL] = {0};
649    int64_t total_bytes_needed = 0;
650    int slice_redist_range = FFMIN(SLICE_REDIST_TOTAL, s->num_x*s->num_y);
651    SliceArgs *enc_args = s->slice_args;
652    SliceArgs *top_loc[SLICE_REDIST_TOTAL] = {NULL};
653
654    init_quant_matrix(s);
655
656    for (slice_y = 0; slice_y < s->num_y; slice_y++) {
657        for (slice_x = 0; slice_x < s->num_x; slice_x++) {
658            SliceArgs *args = &enc_args[s->num_x*slice_y + slice_x];
659            args->ctx = s;
660            args->x   = slice_x;
661            args->y   = slice_y;
662            args->bits_ceil  = s->slice_max_bytes << 3;
663            args->bits_floor = s->slice_min_bytes << 3;
664            memset(args->cache, 0, s->q_ceil*sizeof(*args->cache));
665        }
666    }
667
668    /* First pass - determine baseline slice sizes w.r.t. max_slice_size */
669    s->avctx->execute(s->avctx, rate_control, enc_args, NULL, s->num_x*s->num_y,
670                      sizeof(SliceArgs));
671
672    for (i = 0; i < s->num_x*s->num_y; i++) {
673        SliceArgs *args = &enc_args[i];
674        bytes_left += args->bytes;
675        for (j = 0; j < slice_redist_range; j++) {
676            if (args->bytes > bytes_top[j]) {
677                bytes_top[j] = args->bytes;
678                top_loc[j]   = args;
679                break;
680            }
681        }
682    }
683
684    bytes_left = s->frame_max_bytes - bytes_left;
685
686    /* Second pass - distribute leftover bytes */
687    while (bytes_left > 0) {
688        int distributed = 0;
689        for (i = 0; i < slice_redist_range; i++) {
690            SliceArgs *args;
691            int bits, bytes, diff, prev_bytes, new_idx;
692            if (bytes_left <= 0)
693                break;
694            if (!top_loc[i] || !top_loc[i]->quant_idx)
695                break;
696            args = top_loc[i];
697            prev_bytes = args->bytes;
698            new_idx = FFMAX(args->quant_idx - 1, 0);
699            bits  = count_hq_slice(args, new_idx);
700            bytes = SSIZE_ROUND(bits >> 3);
701            diff  = bytes - prev_bytes;
702            if ((bytes_left - diff) > 0) {
703                args->quant_idx = new_idx;
704                args->bytes = bytes;
705                bytes_left -= diff;
706                distributed++;
707            }
708        }
709        if (!distributed)
710            break;
711    }
712
713    for (i = 0; i < s->num_x*s->num_y; i++) {
714        SliceArgs *args = &enc_args[i];
715        total_bytes_needed += args->bytes;
716        s->q_avg = (s->q_avg + args->quant_idx)/2;
717    }
718
719    return total_bytes_needed;
720}
721
722/* VC-2 13.5.3 - hq_slice */
723static int encode_hq_slice(AVCodecContext *avctx, void *arg)
724{
725    SliceArgs *slice_dat = arg;
726    VC2EncContext *s = slice_dat->ctx;
727    PutBitContext *pb = &slice_dat->pb;
728    const int slice_x = slice_dat->x;
729    const int slice_y = slice_dat->y;
730    const int quant_idx = slice_dat->quant_idx;
731    const int slice_bytes_max = slice_dat->bytes;
732    uint8_t quants[MAX_DWT_LEVELS][4];
733    int p, level, orientation;
734
735    /* The reference decoder ignores it, and its typical length is 0 */
736    memset(put_bits_ptr(pb), 0, s->prefix_bytes);
737    skip_put_bytes(pb, s->prefix_bytes);
738
739    put_bits(pb, 8, quant_idx);
740
741    /* Slice quantization (slice_quantizers() in the specs) */
742    for (level = 0; level < s->wavelet_depth; level++)
743        for (orientation = !!level; orientation < 4; orientation++)
744            quants[level][orientation] = FFMAX(quant_idx - s->quant[level][orientation], 0);
745
746    /* Luma + 2 Chroma planes */
747    for (p = 0; p < 3; p++) {
748        int bytes_start, bytes_len, pad_s, pad_c;
749        bytes_start = put_bits_count(pb) >> 3;
750        put_bits(pb, 8, 0);
751        for (level = 0; level < s->wavelet_depth; level++) {
752            for (orientation = !!level; orientation < 4; orientation++) {
753                encode_subband(s, pb, slice_x, slice_y,
754                               &s->plane[p].band[level][orientation],
755                               quants[level][orientation]);
756            }
757        }
758        align_put_bits(pb);
759        bytes_len = (put_bits_count(pb) >> 3) - bytes_start - 1;
760        if (p == 2) {
761            int len_diff = slice_bytes_max - (put_bits_count(pb) >> 3);
762            pad_s = FFALIGN((bytes_len + len_diff), s->size_scaler)/s->size_scaler;
763            pad_c = (pad_s*s->size_scaler) - bytes_len;
764        } else {
765            pad_s = FFALIGN(bytes_len, s->size_scaler)/s->size_scaler;
766            pad_c = (pad_s*s->size_scaler) - bytes_len;
767        }
768        pb->buf[bytes_start] = pad_s;
769        flush_put_bits(pb);
770        /* vc2-reference uses that padding that decodes to '0' coeffs */
771        memset(put_bits_ptr(pb), 0xFF, pad_c);
772        skip_put_bytes(pb, pad_c);
773    }
774
775    return 0;
776}
777
778/* VC-2 13.5.1 - low_delay_transform_data() */
779static int encode_slices(VC2EncContext *s)
780{
781    uint8_t *buf;
782    int slice_x, slice_y, skip = 0;
783    SliceArgs *enc_args = s->slice_args;
784
785    flush_put_bits(&s->pb);
786    buf = put_bits_ptr(&s->pb);
787
788    for (slice_y = 0; slice_y < s->num_y; slice_y++) {
789        for (slice_x = 0; slice_x < s->num_x; slice_x++) {
790            SliceArgs *args = &enc_args[s->num_x*slice_y + slice_x];
791            init_put_bits(&args->pb, buf + skip, args->bytes+s->prefix_bytes);
792            skip += args->bytes;
793        }
794    }
795
796    s->avctx->execute(s->avctx, encode_hq_slice, enc_args, NULL, s->num_x*s->num_y,
797                      sizeof(SliceArgs));
798
799    skip_put_bytes(&s->pb, skip);
800
801    return 0;
802}
803
804/*
805 * Transform basics for a 3 level transform
806 * |---------------------------------------------------------------------|
807 * |  LL-0  | HL-0  |                 |                                  |
808 * |--------|-------|      HL-1       |                                  |
809 * |  LH-0  | HH-0  |                 |                                  |
810 * |----------------|-----------------|              HL-2                |
811 * |                |                 |                                  |
812 * |     LH-1       |      HH-1       |                                  |
813 * |                |                 |                                  |
814 * |----------------------------------|----------------------------------|
815 * |                                  |                                  |
816 * |                                  |                                  |
817 * |                                  |                                  |
818 * |              LH-2                |              HH-2                |
819 * |                                  |                                  |
820 * |                                  |                                  |
821 * |                                  |                                  |
822 * |---------------------------------------------------------------------|
823 *
824 * DWT transforms are generally applied by splitting the image in two vertically
825 * and applying a low pass transform on the left part and a corresponding high
826 * pass transform on the right hand side. This is known as the horizontal filter
827 * stage.
828 * After that, the same operation is performed except the image is divided
829 * horizontally, with the high pass on the lower and the low pass on the higher
830 * side.
831 * Therefore, you're left with 4 subdivisions - known as  low-low, low-high,
832 * high-low and high-high. They're referred to as orientations in the decoder
833 * and encoder.
834 *
835 * The LL (low-low) area contains the original image downsampled by the amount
836 * of levels. The rest of the areas can be thought as the details needed
837 * to restore the image perfectly to its original size.
838 */
839static int dwt_plane(AVCodecContext *avctx, void *arg)
840{
841    TransformArgs *transform_dat = arg;
842    VC2EncContext *s = transform_dat->ctx;
843    const void *frame_data = transform_dat->idata;
844    const ptrdiff_t linesize = transform_dat->istride;
845    const int field = transform_dat->field;
846    const Plane *p = transform_dat->plane;
847    VC2TransformContext *t = &transform_dat->t;
848    dwtcoef *buf = p->coef_buf;
849    const int idx = s->wavelet_idx;
850    const int skip = 1 + s->interlaced;
851
852    int x, y, level, offset;
853    ptrdiff_t pix_stride = linesize >> (s->bpp - 1);
854
855    if (field == 1) {
856        offset = 0;
857        pix_stride <<= 1;
858    } else if (field == 2) {
859        offset = pix_stride;
860        pix_stride <<= 1;
861    } else {
862        offset = 0;
863    }
864
865    if (s->bpp == 1) {
866        const uint8_t *pix = (const uint8_t *)frame_data + offset;
867        for (y = 0; y < p->height*skip; y+=skip) {
868            for (x = 0; x < p->width; x++) {
869                buf[x] = pix[x] - s->diff_offset;
870            }
871            memset(&buf[x], 0, (p->coef_stride - p->width)*sizeof(dwtcoef));
872            buf += p->coef_stride;
873            pix += pix_stride;
874        }
875    } else {
876        const uint16_t *pix = (const uint16_t *)frame_data + offset;
877        for (y = 0; y < p->height*skip; y+=skip) {
878            for (x = 0; x < p->width; x++) {
879                buf[x] = pix[x] - s->diff_offset;
880            }
881            memset(&buf[x], 0, (p->coef_stride - p->width)*sizeof(dwtcoef));
882            buf += p->coef_stride;
883            pix += pix_stride;
884        }
885    }
886
887    memset(buf, 0, p->coef_stride * (p->dwt_height - p->height) * sizeof(dwtcoef));
888
889    for (level = s->wavelet_depth-1; level >= 0; level--) {
890        const SubBand *b = &p->band[level][0];
891        t->vc2_subband_dwt[idx](t, p->coef_buf, p->coef_stride,
892                                b->width, b->height);
893    }
894
895    return 0;
896}
897
898static int encode_frame(VC2EncContext *s, AVPacket *avpkt, const AVFrame *frame,
899                        const char *aux_data, const int header_size, int field)
900{
901    int i, ret;
902    int64_t max_frame_bytes;
903
904     /* Threaded DWT transform */
905    for (i = 0; i < 3; i++) {
906        s->transform_args[i].ctx   = s;
907        s->transform_args[i].field = field;
908        s->transform_args[i].plane = &s->plane[i];
909        s->transform_args[i].idata = frame->data[i];
910        s->transform_args[i].istride = frame->linesize[i];
911    }
912    s->avctx->execute(s->avctx, dwt_plane, s->transform_args, NULL, 3,
913                      sizeof(TransformArgs));
914
915    /* Calculate per-slice quantizers and sizes */
916    max_frame_bytes = header_size + calc_slice_sizes(s);
917
918    if (field < 2) {
919        ret = ff_get_encode_buffer(s->avctx, avpkt,
920                                   max_frame_bytes << s->interlaced, 0);
921        if (ret) {
922            av_log(s->avctx, AV_LOG_ERROR, "Error getting output packet.\n");
923            return ret;
924        }
925        init_put_bits(&s->pb, avpkt->data, avpkt->size);
926    }
927
928    /* Sequence header */
929    encode_parse_info(s, DIRAC_PCODE_SEQ_HEADER);
930    encode_seq_header(s);
931
932    /* Encoder version */
933    if (aux_data) {
934        encode_parse_info(s, DIRAC_PCODE_AUX);
935        ff_put_string(&s->pb, aux_data, 1);
936    }
937
938    /* Picture header */
939    encode_parse_info(s, DIRAC_PCODE_PICTURE_HQ);
940    encode_picture_start(s);
941
942    /* Encode slices */
943    encode_slices(s);
944
945    /* End sequence */
946    encode_parse_info(s, DIRAC_PCODE_END_SEQ);
947
948    return 0;
949}
950
951static av_cold int vc2_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
952                                      const AVFrame *frame, int *got_packet)
953{
954    int ret = 0;
955    int slice_ceil, sig_size = 256;
956    VC2EncContext *s = avctx->priv_data;
957    const int bitexact = avctx->flags & AV_CODEC_FLAG_BITEXACT;
958    const char *aux_data = bitexact ? "Lavc" : LIBAVCODEC_IDENT;
959    const int aux_data_size = bitexact ? sizeof("Lavc") : sizeof(LIBAVCODEC_IDENT);
960    const int header_size = 100 + aux_data_size;
961    int64_t r_bitrate = avctx->bit_rate >> (s->interlaced);
962
963    s->avctx = avctx;
964    s->size_scaler = 2;
965    s->prefix_bytes = 0;
966    s->last_parse_code = 0;
967    s->next_parse_offset = 0;
968
969    /* Rate control */
970    s->frame_max_bytes = (av_rescale(r_bitrate, s->avctx->time_base.num,
971                                     s->avctx->time_base.den) >> 3) - header_size;
972    s->slice_max_bytes = slice_ceil = av_rescale(s->frame_max_bytes, 1, s->num_x*s->num_y);
973
974    /* Find an appropriate size scaler */
975    while (sig_size > 255) {
976        int r_size = SSIZE_ROUND(s->slice_max_bytes);
977        if (r_size > slice_ceil) {
978            s->slice_max_bytes -= r_size - slice_ceil;
979            r_size = SSIZE_ROUND(s->slice_max_bytes);
980        }
981        sig_size = r_size/s->size_scaler; /* Signalled slize size */
982        s->size_scaler <<= 1;
983    }
984
985    s->slice_min_bytes = s->slice_max_bytes - s->slice_max_bytes*(s->tolerance/100.0f);
986    if (s->slice_min_bytes < 0)
987        return AVERROR(EINVAL);
988
989    ret = encode_frame(s, avpkt, frame, aux_data, header_size, s->interlaced);
990    if (ret)
991        return ret;
992    if (s->interlaced) {
993        ret = encode_frame(s, avpkt, frame, aux_data, header_size, 2);
994        if (ret)
995            return ret;
996    }
997
998    flush_put_bits(&s->pb);
999    av_shrink_packet(avpkt, put_bytes_output(&s->pb));
1000
1001    *got_packet = 1;
1002
1003    return 0;
1004}
1005
1006static av_cold int vc2_encode_end(AVCodecContext *avctx)
1007{
1008    int i;
1009    VC2EncContext *s = avctx->priv_data;
1010
1011    av_log(avctx, AV_LOG_INFO, "Qavg: %i\n", s->q_avg);
1012
1013    for (i = 0; i < 3; i++) {
1014        ff_vc2enc_free_transforms(&s->transform_args[i].t);
1015        av_freep(&s->plane[i].coef_buf);
1016    }
1017
1018    av_freep(&s->slice_args);
1019
1020    return 0;
1021}
1022
1023static av_cold int vc2_encode_init(AVCodecContext *avctx)
1024{
1025    Plane *p;
1026    SubBand *b;
1027    int i, level, o, shift, ret;
1028    const AVPixFmtDescriptor *fmt = av_pix_fmt_desc_get(avctx->pix_fmt);
1029    const int depth = fmt->comp[0].depth;
1030    VC2EncContext *s = avctx->priv_data;
1031
1032    s->picture_number = 0;
1033
1034    /* Total allowed quantization range */
1035    s->q_ceil    = DIRAC_MAX_QUANT_INDEX;
1036
1037    s->ver.major = 2;
1038    s->ver.minor = 0;
1039    s->profile   = 3;
1040    s->level     = 3;
1041
1042    s->base_vf   = -1;
1043    s->strict_compliance = 1;
1044
1045    s->q_avg = 0;
1046    s->slice_max_bytes = 0;
1047    s->slice_min_bytes = 0;
1048
1049    /* Mark unknown as progressive */
1050    s->interlaced = !((avctx->field_order == AV_FIELD_UNKNOWN) ||
1051                      (avctx->field_order == AV_FIELD_PROGRESSIVE));
1052
1053    for (i = 0; i < base_video_fmts_len; i++) {
1054        const VC2BaseVideoFormat *fmt = &base_video_fmts[i];
1055        if (avctx->pix_fmt != fmt->pix_fmt)
1056            continue;
1057        if (avctx->time_base.num != fmt->time_base.num)
1058            continue;
1059        if (avctx->time_base.den != fmt->time_base.den)
1060            continue;
1061        if (avctx->width != fmt->width)
1062            continue;
1063        if (avctx->height != fmt->height)
1064            continue;
1065        if (s->interlaced != fmt->interlaced)
1066            continue;
1067        s->base_vf = i;
1068        s->level   = base_video_fmts[i].level;
1069        break;
1070    }
1071
1072    if (s->interlaced)
1073        av_log(avctx, AV_LOG_WARNING, "Interlacing enabled!\n");
1074
1075    if ((s->slice_width  & (s->slice_width  - 1)) ||
1076        (s->slice_height & (s->slice_height - 1))) {
1077        av_log(avctx, AV_LOG_ERROR, "Slice size is not a power of two!\n");
1078        return AVERROR_UNKNOWN;
1079    }
1080
1081    if ((s->slice_width > avctx->width) ||
1082        (s->slice_height > avctx->height)) {
1083        av_log(avctx, AV_LOG_ERROR, "Slice size is bigger than the image!\n");
1084        return AVERROR_UNKNOWN;
1085    }
1086
1087    if (s->base_vf <= 0) {
1088        if (avctx->strict_std_compliance < FF_COMPLIANCE_STRICT) {
1089            s->strict_compliance = s->base_vf = 0;
1090            av_log(avctx, AV_LOG_WARNING, "Format does not strictly comply with VC2 specs\n");
1091        } else {
1092            av_log(avctx, AV_LOG_WARNING, "Given format does not strictly comply with "
1093                   "the specifications, decrease strictness to use it.\n");
1094            return AVERROR_UNKNOWN;
1095        }
1096    } else {
1097        av_log(avctx, AV_LOG_INFO, "Selected base video format = %i (%s)\n",
1098               s->base_vf, base_video_fmts[s->base_vf].name);
1099    }
1100
1101    /* Chroma subsampling */
1102    ret = av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
1103    if (ret)
1104        return ret;
1105
1106    /* Bit depth and color range index */
1107    if (depth == 8 && avctx->color_range == AVCOL_RANGE_JPEG) {
1108        s->bpp = 1;
1109        s->bpp_idx = 1;
1110        s->diff_offset = 128;
1111    } else if (depth == 8 && (avctx->color_range == AVCOL_RANGE_MPEG ||
1112               avctx->color_range == AVCOL_RANGE_UNSPECIFIED)) {
1113        s->bpp = 1;
1114        s->bpp_idx = 2;
1115        s->diff_offset = 128;
1116    } else if (depth == 10) {
1117        s->bpp = 2;
1118        s->bpp_idx = 3;
1119        s->diff_offset = 512;
1120    } else {
1121        s->bpp = 2;
1122        s->bpp_idx = 4;
1123        s->diff_offset = 2048;
1124    }
1125
1126    /* Planes initialization */
1127    for (i = 0; i < 3; i++) {
1128        int w, h;
1129        p = &s->plane[i];
1130        p->width      = avctx->width  >> (i ? s->chroma_x_shift : 0);
1131        p->height     = avctx->height >> (i ? s->chroma_y_shift : 0);
1132        if (s->interlaced)
1133            p->height >>= 1;
1134        p->dwt_width  = w = FFALIGN(p->width,  (1 << s->wavelet_depth));
1135        p->dwt_height = h = FFALIGN(p->height, (1 << s->wavelet_depth));
1136        p->coef_stride = FFALIGN(p->dwt_width, 32);
1137        p->coef_buf = av_mallocz(p->coef_stride*p->dwt_height*sizeof(dwtcoef));
1138        if (!p->coef_buf)
1139            return AVERROR(ENOMEM);
1140        for (level = s->wavelet_depth-1; level >= 0; level--) {
1141            w = w >> 1;
1142            h = h >> 1;
1143            for (o = 0; o < 4; o++) {
1144                b = &p->band[level][o];
1145                b->width  = w;
1146                b->height = h;
1147                b->stride = p->coef_stride;
1148                shift = (o > 1)*b->height*b->stride + (o & 1)*b->width;
1149                b->buf = p->coef_buf + shift;
1150            }
1151        }
1152
1153        /* DWT init */
1154        if (ff_vc2enc_init_transforms(&s->transform_args[i].t,
1155                                      s->plane[i].coef_stride,
1156                                      s->plane[i].dwt_height,
1157                                      s->slice_width, s->slice_height))
1158            return AVERROR(ENOMEM);
1159    }
1160
1161    /* Slices */
1162    s->num_x = s->plane[0].dwt_width/s->slice_width;
1163    s->num_y = s->plane[0].dwt_height/s->slice_height;
1164
1165    s->slice_args = av_calloc(s->num_x*s->num_y, sizeof(SliceArgs));
1166    if (!s->slice_args)
1167        return AVERROR(ENOMEM);
1168
1169    for (i = 0; i < 116; i++) {
1170        const uint64_t qf = ff_dirac_qscale_tab[i];
1171        const uint32_t m = av_log2(qf);
1172        const uint32_t t = (1ULL << (m + 32)) / qf;
1173        const uint32_t r = (t*qf + qf) & UINT32_MAX;
1174        if (!(qf & (qf - 1))) {
1175            s->qmagic_lut[i][0] = 0xFFFFFFFF;
1176            s->qmagic_lut[i][1] = 0xFFFFFFFF;
1177        } else if (r <= 1 << m) {
1178            s->qmagic_lut[i][0] = t + 1;
1179            s->qmagic_lut[i][1] = 0;
1180        } else {
1181            s->qmagic_lut[i][0] = t;
1182            s->qmagic_lut[i][1] = t;
1183        }
1184    }
1185
1186    return 0;
1187}
1188
1189#define VC2ENC_FLAGS (AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_VIDEO_PARAM)
1190static const AVOption vc2enc_options[] = {
1191    {"tolerance",     "Max undershoot in percent", offsetof(VC2EncContext, tolerance), AV_OPT_TYPE_DOUBLE, {.dbl = 5.0f}, 0.0f, 45.0f, VC2ENC_FLAGS, "tolerance"},
1192    {"slice_width",   "Slice width",  offsetof(VC2EncContext, slice_width), AV_OPT_TYPE_INT, {.i64 = 32}, 32, 1024, VC2ENC_FLAGS, "slice_width"},
1193    {"slice_height",  "Slice height", offsetof(VC2EncContext, slice_height), AV_OPT_TYPE_INT, {.i64 = 16}, 8, 1024, VC2ENC_FLAGS, "slice_height"},
1194    {"wavelet_depth", "Transform depth", offsetof(VC2EncContext, wavelet_depth), AV_OPT_TYPE_INT, {.i64 = 4}, 1, 5, VC2ENC_FLAGS, "wavelet_depth"},
1195    {"wavelet_type",  "Transform type",  offsetof(VC2EncContext, wavelet_idx), AV_OPT_TYPE_INT, {.i64 = VC2_TRANSFORM_9_7}, 0, VC2_TRANSFORMS_NB, VC2ENC_FLAGS, "wavelet_idx"},
1196        {"9_7",          "Deslauriers-Dubuc (9,7)", 0, AV_OPT_TYPE_CONST, {.i64 = VC2_TRANSFORM_9_7},    INT_MIN, INT_MAX, VC2ENC_FLAGS, "wavelet_idx"},
1197        {"5_3",          "LeGall (5,3)",            0, AV_OPT_TYPE_CONST, {.i64 = VC2_TRANSFORM_5_3},    INT_MIN, INT_MAX, VC2ENC_FLAGS, "wavelet_idx"},
1198        {"haar",         "Haar (with shift)",       0, AV_OPT_TYPE_CONST, {.i64 = VC2_TRANSFORM_HAAR_S}, INT_MIN, INT_MAX, VC2ENC_FLAGS, "wavelet_idx"},
1199        {"haar_noshift", "Haar (without shift)",    0, AV_OPT_TYPE_CONST, {.i64 = VC2_TRANSFORM_HAAR},   INT_MIN, INT_MAX, VC2ENC_FLAGS, "wavelet_idx"},
1200    {"qm", "Custom quantization matrix", offsetof(VC2EncContext, quant_matrix), AV_OPT_TYPE_INT, {.i64 = VC2_QM_DEF}, 0, VC2_QM_NB, VC2ENC_FLAGS, "quant_matrix"},
1201        {"default",   "Default from the specifications", 0, AV_OPT_TYPE_CONST, {.i64 = VC2_QM_DEF}, INT_MIN, INT_MAX, VC2ENC_FLAGS, "quant_matrix"},
1202        {"color",     "Prevents low bitrate discoloration", 0, AV_OPT_TYPE_CONST, {.i64 = VC2_QM_COL}, INT_MIN, INT_MAX, VC2ENC_FLAGS, "quant_matrix"},
1203        {"flat",      "Optimize for PSNR", 0, AV_OPT_TYPE_CONST, {.i64 = VC2_QM_FLAT}, INT_MIN, INT_MAX, VC2ENC_FLAGS, "quant_matrix"},
1204    {NULL}
1205};
1206
1207static const AVClass vc2enc_class = {
1208    .class_name = "SMPTE VC-2 encoder",
1209    .category = AV_CLASS_CATEGORY_ENCODER,
1210    .option = vc2enc_options,
1211    .item_name = av_default_item_name,
1212    .version = LIBAVUTIL_VERSION_INT
1213};
1214
1215static const FFCodecDefault vc2enc_defaults[] = {
1216    { "b",              "600000000"   },
1217    { NULL },
1218};
1219
1220static const enum AVPixelFormat allowed_pix_fmts[] = {
1221    AV_PIX_FMT_YUV420P,   AV_PIX_FMT_YUV422P,   AV_PIX_FMT_YUV444P,
1222    AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
1223    AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12,
1224    AV_PIX_FMT_NONE
1225};
1226
1227const FFCodec ff_vc2_encoder = {
1228    .p.name         = "vc2",
1229    .p.long_name    = NULL_IF_CONFIG_SMALL("SMPTE VC-2"),
1230    .p.type         = AVMEDIA_TYPE_VIDEO,
1231    .p.id           = AV_CODEC_ID_DIRAC,
1232    .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_SLICE_THREADS,
1233    .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
1234    .priv_data_size = sizeof(VC2EncContext),
1235    .init           = vc2_encode_init,
1236    .close          = vc2_encode_end,
1237    FF_CODEC_ENCODE_CB(vc2_encode_frame),
1238    .p.priv_class   = &vc2enc_class,
1239    .defaults       = vc2enc_defaults,
1240    .p.pix_fmts     = allowed_pix_fmts
1241};
1242