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 "dirac.h"
25 #include "put_bits.h"
26 #include "internal.h"
27 #include "version.h"
28
29 #include "vc2enc_dwt.h"
30 #include "diractab.h"
31
32 /* The limited size resolution of each slice forces us to do this */
33 #define SSIZE_ROUND(b) (FFALIGN((b), s->size_scaler) + 4 + s->prefix_bytes)
34
35 /* Decides the cutoff point in # of slices to distribute the leftover bytes */
36 #define SLICE_REDIST_TOTAL 150
37
38 typedef struct VC2BaseVideoFormat {
39 enum AVPixelFormat pix_fmt;
40 AVRational time_base;
41 int width, height, interlaced, level;
42 const char *name;
43 } VC2BaseVideoFormat;
44
45 static const VC2BaseVideoFormat base_video_fmts[] = {
46 { 0 }, /* Custom format, here just to make indexing equal to base_vf */
47 { AV_PIX_FMT_YUV420P, { 1001, 15000 }, 176, 120, 0, 1, "QSIF525" },
48 { AV_PIX_FMT_YUV420P, { 2, 25 }, 176, 144, 0, 1, "QCIF" },
49 { AV_PIX_FMT_YUV420P, { 1001, 15000 }, 352, 240, 0, 1, "SIF525" },
50 { AV_PIX_FMT_YUV420P, { 2, 25 }, 352, 288, 0, 1, "CIF" },
51 { AV_PIX_FMT_YUV420P, { 1001, 15000 }, 704, 480, 0, 1, "4SIF525" },
52 { AV_PIX_FMT_YUV420P, { 2, 25 }, 704, 576, 0, 1, "4CIF" },
53
54 { AV_PIX_FMT_YUV422P10, { 1001, 30000 }, 720, 480, 1, 2, "SD480I-60" },
55 { AV_PIX_FMT_YUV422P10, { 1, 25 }, 720, 576, 1, 2, "SD576I-50" },
56
57 { AV_PIX_FMT_YUV422P10, { 1001, 60000 }, 1280, 720, 0, 3, "HD720P-60" },
58 { AV_PIX_FMT_YUV422P10, { 1, 50 }, 1280, 720, 0, 3, "HD720P-50" },
59 { AV_PIX_FMT_YUV422P10, { 1001, 30000 }, 1920, 1080, 1, 3, "HD1080I-60" },
60 { AV_PIX_FMT_YUV422P10, { 1, 25 }, 1920, 1080, 1, 3, "HD1080I-50" },
61 { AV_PIX_FMT_YUV422P10, { 1001, 60000 }, 1920, 1080, 0, 3, "HD1080P-60" },
62 { AV_PIX_FMT_YUV422P10, { 1, 50 }, 1920, 1080, 0, 3, "HD1080P-50" },
63
64 { AV_PIX_FMT_YUV444P12, { 1, 24 }, 2048, 1080, 0, 4, "DC2K" },
65 { AV_PIX_FMT_YUV444P12, { 1, 24 }, 4096, 2160, 0, 5, "DC4K" },
66
67 { AV_PIX_FMT_YUV422P10, { 1001, 60000 }, 3840, 2160, 0, 6, "UHDTV 4K-60" },
68 { AV_PIX_FMT_YUV422P10, { 1, 50 }, 3840, 2160, 0, 6, "UHDTV 4K-50" },
69
70 { AV_PIX_FMT_YUV422P10, { 1001, 60000 }, 7680, 4320, 0, 7, "UHDTV 8K-60" },
71 { AV_PIX_FMT_YUV422P10, { 1, 50 }, 7680, 4320, 0, 7, "UHDTV 8K-50" },
72
73 { AV_PIX_FMT_YUV422P10, { 1001, 24000 }, 1920, 1080, 0, 3, "HD1080P-24" },
74 { AV_PIX_FMT_YUV422P10, { 1001, 30000 }, 720, 486, 1, 2, "SD Pro486" },
75 };
76 static const int base_video_fmts_len = FF_ARRAY_ELEMS(base_video_fmts);
77
78 enum VC2_QM {
79 VC2_QM_DEF = 0,
80 VC2_QM_COL,
81 VC2_QM_FLAT,
82
83 VC2_QM_NB
84 };
85
86 typedef struct SubBand {
87 dwtcoef *buf;
88 ptrdiff_t stride;
89 int width;
90 int height;
91 } SubBand;
92
93 typedef struct Plane {
94 SubBand band[MAX_DWT_LEVELS][4];
95 dwtcoef *coef_buf;
96 int width;
97 int height;
98 int dwt_width;
99 int dwt_height;
100 ptrdiff_t coef_stride;
101 } Plane;
102
103 typedef struct SliceArgs {
104 PutBitContext pb;
105 int cache[DIRAC_MAX_QUANT_INDEX];
106 void *ctx;
107 int x;
108 int y;
109 int quant_idx;
110 int bits_ceil;
111 int bits_floor;
112 int bytes;
113 } SliceArgs;
114
115 typedef struct TransformArgs {
116 void *ctx;
117 Plane *plane;
118 void *idata;
119 ptrdiff_t istride;
120 int field;
121 VC2TransformContext t;
122 } TransformArgs;
123
124 typedef struct VC2EncContext {
125 AVClass *av_class;
126 PutBitContext pb;
127 Plane plane[3];
128 AVCodecContext *avctx;
129 DiracVersionInfo ver;
130
131 SliceArgs *slice_args;
132 TransformArgs transform_args[3];
133
134 /* For conversion from unsigned pixel values to signed */
135 int diff_offset;
136 int bpp;
137 int bpp_idx;
138
139 /* Picture number */
140 uint32_t picture_number;
141
142 /* Base video format */
143 int base_vf;
144 int level;
145 int profile;
146
147 /* Quantization matrix */
148 uint8_t quant[MAX_DWT_LEVELS][4];
149 int custom_quant_matrix;
150
151 /* Division LUT */
152 uint32_t qmagic_lut[116][2];
153
154 int num_x; /* #slices horizontally */
155 int num_y; /* #slices vertically */
156 int prefix_bytes;
157 int size_scaler;
158 int chroma_x_shift;
159 int chroma_y_shift;
160
161 /* Rate control stuff */
162 int frame_max_bytes;
163 int slice_max_bytes;
164 int slice_min_bytes;
165 int q_ceil;
166 int q_avg;
167
168 /* Options */
169 double tolerance;
170 int wavelet_idx;
171 int wavelet_depth;
172 int strict_compliance;
173 int slice_height;
174 int slice_width;
175 int interlaced;
176 enum VC2_QM quant_matrix;
177
178 /* Parse code state */
179 uint32_t next_parse_offset;
180 enum DiracParseCodes last_parse_code;
181 } VC2EncContext;
182
put_vc2_ue_uint(PutBitContext * pb,uint32_t val)183 static av_always_inline void put_vc2_ue_uint(PutBitContext *pb, uint32_t val)
184 {
185 int i;
186 int pbits = 0, bits = 0, topbit = 1, maxval = 1;
187
188 if (!val++) {
189 put_bits(pb, 1, 1);
190 return;
191 }
192
193 while (val > maxval) {
194 topbit <<= 1;
195 maxval <<= 1;
196 maxval |= 1;
197 }
198
199 bits = ff_log2(topbit);
200
201 for (i = 0; i < bits; i++) {
202 topbit >>= 1;
203 pbits <<= 2;
204 if (val & topbit)
205 pbits |= 0x1;
206 }
207
208 put_bits(pb, bits*2 + 1, (pbits << 1) | 1);
209 }
210
count_vc2_ue_uint(uint32_t val)211 static av_always_inline int count_vc2_ue_uint(uint32_t val)
212 {
213 int topbit = 1, maxval = 1;
214
215 if (!val++)
216 return 1;
217
218 while (val > maxval) {
219 topbit <<= 1;
220 maxval <<= 1;
221 maxval |= 1;
222 }
223
224 return ff_log2(topbit)*2 + 1;
225 }
226
227 /* VC-2 10.4 - parse_info() */
encode_parse_info(VC2EncContext * s,enum DiracParseCodes pcode)228 static void encode_parse_info(VC2EncContext *s, enum DiracParseCodes pcode)
229 {
230 uint32_t cur_pos, dist;
231
232 avpriv_align_put_bits(&s->pb);
233
234 cur_pos = put_bits_count(&s->pb) >> 3;
235
236 /* Magic string */
237 avpriv_put_string(&s->pb, "BBCD", 0);
238
239 /* Parse code */
240 put_bits(&s->pb, 8, pcode);
241
242 /* Next parse offset */
243 dist = cur_pos - s->next_parse_offset;
244 AV_WB32(s->pb.buf + s->next_parse_offset + 5, dist);
245 s->next_parse_offset = cur_pos;
246 put_bits32(&s->pb, pcode == DIRAC_PCODE_END_SEQ ? 13 : 0);
247
248 /* Last parse offset */
249 put_bits32(&s->pb, s->last_parse_code == DIRAC_PCODE_END_SEQ ? 13 : dist);
250
251 s->last_parse_code = pcode;
252 }
253
254 /* VC-2 11.1 - parse_parameters()
255 * The level dictates what the decoder should expect in terms of resolution
256 * and allows it to quickly reject whatever it can't support. Remember,
257 * this codec kinda targets cheapo FPGAs without much memory. Unfortunately
258 * it also limits us greatly in our choice of formats, hence the flag to disable
259 * strict_compliance */
encode_parse_params(VC2EncContext * s)260 static void encode_parse_params(VC2EncContext *s)
261 {
262 put_vc2_ue_uint(&s->pb, s->ver.major); /* VC-2 demands this to be 2 */
263 put_vc2_ue_uint(&s->pb, s->ver.minor); /* ^^ and this to be 0 */
264 put_vc2_ue_uint(&s->pb, s->profile); /* 3 to signal HQ profile */
265 put_vc2_ue_uint(&s->pb, s->level); /* 3 - 1080/720, 6 - 4K */
266 }
267
268 /* VC-2 11.3 - frame_size() */
encode_frame_size(VC2EncContext * s)269 static void encode_frame_size(VC2EncContext *s)
270 {
271 put_bits(&s->pb, 1, !s->strict_compliance);
272 if (!s->strict_compliance) {
273 AVCodecContext *avctx = s->avctx;
274 put_vc2_ue_uint(&s->pb, avctx->width);
275 put_vc2_ue_uint(&s->pb, avctx->height);
276 }
277 }
278
279 /* VC-2 11.3.3 - color_diff_sampling_format() */
encode_sample_fmt(VC2EncContext * s)280 static void encode_sample_fmt(VC2EncContext *s)
281 {
282 put_bits(&s->pb, 1, !s->strict_compliance);
283 if (!s->strict_compliance) {
284 int idx;
285 if (s->chroma_x_shift == 1 && s->chroma_y_shift == 0)
286 idx = 1; /* 422 */
287 else if (s->chroma_x_shift == 1 && s->chroma_y_shift == 1)
288 idx = 2; /* 420 */
289 else
290 idx = 0; /* 444 */
291 put_vc2_ue_uint(&s->pb, idx);
292 }
293 }
294
295 /* VC-2 11.3.4 - scan_format() */
encode_scan_format(VC2EncContext * s)296 static void encode_scan_format(VC2EncContext *s)
297 {
298 put_bits(&s->pb, 1, !s->strict_compliance);
299 if (!s->strict_compliance)
300 put_vc2_ue_uint(&s->pb, s->interlaced);
301 }
302
303 /* VC-2 11.3.5 - frame_rate() */
encode_frame_rate(VC2EncContext * s)304 static void encode_frame_rate(VC2EncContext *s)
305 {
306 put_bits(&s->pb, 1, !s->strict_compliance);
307 if (!s->strict_compliance) {
308 AVCodecContext *avctx = s->avctx;
309 put_vc2_ue_uint(&s->pb, 0);
310 put_vc2_ue_uint(&s->pb, avctx->time_base.den);
311 put_vc2_ue_uint(&s->pb, avctx->time_base.num);
312 }
313 }
314
315 /* VC-2 11.3.6 - aspect_ratio() */
encode_aspect_ratio(VC2EncContext * s)316 static void encode_aspect_ratio(VC2EncContext *s)
317 {
318 put_bits(&s->pb, 1, !s->strict_compliance);
319 if (!s->strict_compliance) {
320 AVCodecContext *avctx = s->avctx;
321 put_vc2_ue_uint(&s->pb, 0);
322 put_vc2_ue_uint(&s->pb, avctx->sample_aspect_ratio.num);
323 put_vc2_ue_uint(&s->pb, avctx->sample_aspect_ratio.den);
324 }
325 }
326
327 /* VC-2 11.3.7 - clean_area() */
encode_clean_area(VC2EncContext * s)328 static void encode_clean_area(VC2EncContext *s)
329 {
330 put_bits(&s->pb, 1, 0);
331 }
332
333 /* VC-2 11.3.8 - signal_range() */
encode_signal_range(VC2EncContext * s)334 static void encode_signal_range(VC2EncContext *s)
335 {
336 put_bits(&s->pb, 1, !s->strict_compliance);
337 if (!s->strict_compliance)
338 put_vc2_ue_uint(&s->pb, s->bpp_idx);
339 }
340
341 /* VC-2 11.3.9 - color_spec() */
encode_color_spec(VC2EncContext * s)342 static void encode_color_spec(VC2EncContext *s)
343 {
344 AVCodecContext *avctx = s->avctx;
345 put_bits(&s->pb, 1, !s->strict_compliance);
346 if (!s->strict_compliance) {
347 int val;
348 put_vc2_ue_uint(&s->pb, 0);
349
350 /* primaries */
351 put_bits(&s->pb, 1, 1);
352 if (avctx->color_primaries == AVCOL_PRI_BT470BG)
353 val = 2;
354 else if (avctx->color_primaries == AVCOL_PRI_SMPTE170M)
355 val = 1;
356 else if (avctx->color_primaries == AVCOL_PRI_SMPTE240M)
357 val = 1;
358 else
359 val = 0;
360 put_vc2_ue_uint(&s->pb, val);
361
362 /* color matrix */
363 put_bits(&s->pb, 1, 1);
364 if (avctx->colorspace == AVCOL_SPC_RGB)
365 val = 3;
366 else if (avctx->colorspace == AVCOL_SPC_YCOCG)
367 val = 2;
368 else if (avctx->colorspace == AVCOL_SPC_BT470BG)
369 val = 1;
370 else
371 val = 0;
372 put_vc2_ue_uint(&s->pb, val);
373
374 /* transfer function */
375 put_bits(&s->pb, 1, 1);
376 if (avctx->color_trc == AVCOL_TRC_LINEAR)
377 val = 2;
378 else if (avctx->color_trc == AVCOL_TRC_BT1361_ECG)
379 val = 1;
380 else
381 val = 0;
382 put_vc2_ue_uint(&s->pb, val);
383 }
384 }
385
386 /* VC-2 11.3 - source_parameters() */
encode_source_params(VC2EncContext * s)387 static void encode_source_params(VC2EncContext *s)
388 {
389 encode_frame_size(s);
390 encode_sample_fmt(s);
391 encode_scan_format(s);
392 encode_frame_rate(s);
393 encode_aspect_ratio(s);
394 encode_clean_area(s);
395 encode_signal_range(s);
396 encode_color_spec(s);
397 }
398
399 /* VC-2 11 - sequence_header() */
encode_seq_header(VC2EncContext * s)400 static void encode_seq_header(VC2EncContext *s)
401 {
402 avpriv_align_put_bits(&s->pb);
403 encode_parse_params(s);
404 put_vc2_ue_uint(&s->pb, s->base_vf);
405 encode_source_params(s);
406 put_vc2_ue_uint(&s->pb, s->interlaced); /* Frames or fields coding */
407 }
408
409 /* VC-2 12.1 - picture_header() */
encode_picture_header(VC2EncContext * s)410 static void encode_picture_header(VC2EncContext *s)
411 {
412 avpriv_align_put_bits(&s->pb);
413 put_bits32(&s->pb, s->picture_number++);
414 }
415
416 /* VC-2 12.3.4.1 - slice_parameters() */
encode_slice_params(VC2EncContext * s)417 static void encode_slice_params(VC2EncContext *s)
418 {
419 put_vc2_ue_uint(&s->pb, s->num_x);
420 put_vc2_ue_uint(&s->pb, s->num_y);
421 put_vc2_ue_uint(&s->pb, s->prefix_bytes);
422 put_vc2_ue_uint(&s->pb, s->size_scaler);
423 }
424
425 /* 1st idx = LL, second - vertical, third - horizontal, fourth - total */
426 const uint8_t vc2_qm_col_tab[][4] = {
427 {20, 9, 15, 4},
428 { 0, 6, 6, 4},
429 { 0, 3, 3, 5},
430 { 0, 3, 5, 1},
431 { 0, 11, 10, 11}
432 };
433
434 const uint8_t vc2_qm_flat_tab[][4] = {
435 { 0, 0, 0, 0},
436 { 0, 0, 0, 0},
437 { 0, 0, 0, 0},
438 { 0, 0, 0, 0},
439 { 0, 0, 0, 0}
440 };
441
init_quant_matrix(VC2EncContext * s)442 static void init_quant_matrix(VC2EncContext *s)
443 {
444 int level, orientation;
445
446 if (s->wavelet_depth <= 4 && s->quant_matrix == VC2_QM_DEF) {
447 s->custom_quant_matrix = 0;
448 for (level = 0; level < s->wavelet_depth; level++) {
449 s->quant[level][0] = ff_dirac_default_qmat[s->wavelet_idx][level][0];
450 s->quant[level][1] = ff_dirac_default_qmat[s->wavelet_idx][level][1];
451 s->quant[level][2] = ff_dirac_default_qmat[s->wavelet_idx][level][2];
452 s->quant[level][3] = ff_dirac_default_qmat[s->wavelet_idx][level][3];
453 }
454 return;
455 }
456
457 s->custom_quant_matrix = 1;
458
459 if (s->quant_matrix == VC2_QM_DEF) {
460 for (level = 0; level < s->wavelet_depth; level++) {
461 for (orientation = 0; orientation < 4; orientation++) {
462 if (level <= 3)
463 s->quant[level][orientation] = ff_dirac_default_qmat[s->wavelet_idx][level][orientation];
464 else
465 s->quant[level][orientation] = vc2_qm_col_tab[level][orientation];
466 }
467 }
468 } else if (s->quant_matrix == VC2_QM_COL) {
469 for (level = 0; level < s->wavelet_depth; level++) {
470 for (orientation = 0; orientation < 4; orientation++) {
471 s->quant[level][orientation] = vc2_qm_col_tab[level][orientation];
472 }
473 }
474 } else {
475 for (level = 0; level < s->wavelet_depth; level++) {
476 for (orientation = 0; orientation < 4; orientation++) {
477 s->quant[level][orientation] = vc2_qm_flat_tab[level][orientation];
478 }
479 }
480 }
481 }
482
483 /* VC-2 12.3.4.2 - quant_matrix() */
encode_quant_matrix(VC2EncContext * s)484 static void encode_quant_matrix(VC2EncContext *s)
485 {
486 int level;
487 put_bits(&s->pb, 1, s->custom_quant_matrix);
488 if (s->custom_quant_matrix) {
489 put_vc2_ue_uint(&s->pb, s->quant[0][0]);
490 for (level = 0; level < s->wavelet_depth; level++) {
491 put_vc2_ue_uint(&s->pb, s->quant[level][1]);
492 put_vc2_ue_uint(&s->pb, s->quant[level][2]);
493 put_vc2_ue_uint(&s->pb, s->quant[level][3]);
494 }
495 }
496 }
497
498 /* VC-2 12.3 - transform_parameters() */
encode_transform_params(VC2EncContext * s)499 static void encode_transform_params(VC2EncContext *s)
500 {
501 put_vc2_ue_uint(&s->pb, s->wavelet_idx);
502 put_vc2_ue_uint(&s->pb, s->wavelet_depth);
503
504 encode_slice_params(s);
505 encode_quant_matrix(s);
506 }
507
508 /* VC-2 12.2 - wavelet_transform() */
encode_wavelet_transform(VC2EncContext * s)509 static void encode_wavelet_transform(VC2EncContext *s)
510 {
511 encode_transform_params(s);
512 avpriv_align_put_bits(&s->pb);
513 }
514
515 /* VC-2 12 - picture_parse() */
encode_picture_start(VC2EncContext * s)516 static void encode_picture_start(VC2EncContext *s)
517 {
518 avpriv_align_put_bits(&s->pb);
519 encode_picture_header(s);
520 avpriv_align_put_bits(&s->pb);
521 encode_wavelet_transform(s);
522 }
523
524 #define QUANT(c, mul, add, shift) (((mul) * (c) + (add)) >> (shift))
525
526 /* VC-2 13.5.5.2 - slice_band() */
encode_subband(VC2EncContext * s,PutBitContext * pb,int sx,int sy,SubBand * b,int quant)527 static void encode_subband(VC2EncContext *s, PutBitContext *pb, int sx, int sy,
528 SubBand *b, int quant)
529 {
530 int x, y;
531
532 const int left = b->width * (sx+0) / s->num_x;
533 const int right = b->width * (sx+1) / s->num_x;
534 const int top = b->height * (sy+0) / s->num_y;
535 const int bottom = b->height * (sy+1) / s->num_y;
536
537 dwtcoef *coeff = b->buf + top * b->stride;
538 const uint64_t q_m = ((uint64_t)(s->qmagic_lut[quant][0])) << 2;
539 const uint64_t q_a = s->qmagic_lut[quant][1];
540 const int q_s = av_log2(ff_dirac_qscale_tab[quant]) + 32;
541
542 for (y = top; y < bottom; y++) {
543 for (x = left; x < right; x++) {
544 uint32_t c_abs = QUANT(FFABS(coeff[x]), q_m, q_a, q_s);
545 put_vc2_ue_uint(pb, c_abs);
546 if (c_abs)
547 put_bits(pb, 1, coeff[x] < 0);
548 }
549 coeff += b->stride;
550 }
551 }
552
count_hq_slice(SliceArgs * slice,int quant_idx)553 static int count_hq_slice(SliceArgs *slice, int quant_idx)
554 {
555 int x, y;
556 uint8_t quants[MAX_DWT_LEVELS][4];
557 int bits = 0, p, level, orientation;
558 VC2EncContext *s = slice->ctx;
559
560 if (slice->cache[quant_idx])
561 return slice->cache[quant_idx];
562
563 bits += 8*s->prefix_bytes;
564 bits += 8; /* quant_idx */
565
566 for (level = 0; level < s->wavelet_depth; level++)
567 for (orientation = !!level; orientation < 4; orientation++)
568 quants[level][orientation] = FFMAX(quant_idx - s->quant[level][orientation], 0);
569
570 for (p = 0; p < 3; p++) {
571 int bytes_start, bytes_len, pad_s, pad_c;
572 bytes_start = bits >> 3;
573 bits += 8;
574 for (level = 0; level < s->wavelet_depth; level++) {
575 for (orientation = !!level; orientation < 4; orientation++) {
576 SubBand *b = &s->plane[p].band[level][orientation];
577
578 const int q_idx = quants[level][orientation];
579 const uint64_t q_m = ((uint64_t)s->qmagic_lut[q_idx][0]) << 2;
580 const uint64_t q_a = s->qmagic_lut[q_idx][1];
581 const int q_s = av_log2(ff_dirac_qscale_tab[q_idx]) + 32;
582
583 const int left = b->width * slice->x / s->num_x;
584 const int right = b->width *(slice->x+1) / s->num_x;
585 const int top = b->height * slice->y / s->num_y;
586 const int bottom = b->height *(slice->y+1) / s->num_y;
587
588 dwtcoef *buf = b->buf + top * b->stride;
589
590 for (y = top; y < bottom; y++) {
591 for (x = left; x < right; x++) {
592 uint32_t c_abs = QUANT(FFABS(buf[x]), q_m, q_a, q_s);
593 bits += count_vc2_ue_uint(c_abs);
594 bits += !!c_abs;
595 }
596 buf += b->stride;
597 }
598 }
599 }
600 bits += FFALIGN(bits, 8) - bits;
601 bytes_len = (bits >> 3) - bytes_start - 1;
602 pad_s = FFALIGN(bytes_len, s->size_scaler)/s->size_scaler;
603 pad_c = (pad_s*s->size_scaler) - bytes_len;
604 bits += pad_c*8;
605 }
606
607 slice->cache[quant_idx] = bits;
608
609 return bits;
610 }
611
612 /* Approaches the best possible quantizer asymptotically, its kinda exaustive
613 * but we have a LUT to get the coefficient size in bits. Guaranteed to never
614 * overshoot, which is apparently very important when streaming */
rate_control(AVCodecContext * avctx,void * arg)615 static int rate_control(AVCodecContext *avctx, void *arg)
616 {
617 SliceArgs *slice_dat = arg;
618 VC2EncContext *s = slice_dat->ctx;
619 const int top = slice_dat->bits_ceil;
620 const int bottom = slice_dat->bits_floor;
621 int quant_buf[2] = {-1, -1};
622 int quant = slice_dat->quant_idx, step = 1;
623 int bits_last, bits = count_hq_slice(slice_dat, quant);
624 while ((bits > top) || (bits < bottom)) {
625 const int signed_step = bits > top ? +step : -step;
626 quant = av_clip(quant + signed_step, 0, s->q_ceil-1);
627 bits = count_hq_slice(slice_dat, quant);
628 if (quant_buf[1] == quant) {
629 quant = FFMAX(quant_buf[0], quant);
630 bits = quant == quant_buf[0] ? bits_last : bits;
631 break;
632 }
633 step = av_clip(step/2, 1, (s->q_ceil-1)/2);
634 quant_buf[1] = quant_buf[0];
635 quant_buf[0] = quant;
636 bits_last = bits;
637 }
638 slice_dat->quant_idx = av_clip(quant, 0, s->q_ceil-1);
639 slice_dat->bytes = SSIZE_ROUND(bits >> 3);
640 return 0;
641 }
642
calc_slice_sizes(VC2EncContext * s)643 static int calc_slice_sizes(VC2EncContext *s)
644 {
645 int i, j, slice_x, slice_y, bytes_left = 0;
646 int bytes_top[SLICE_REDIST_TOTAL] = {0};
647 int64_t total_bytes_needed = 0;
648 int slice_redist_range = FFMIN(SLICE_REDIST_TOTAL, s->num_x*s->num_y);
649 SliceArgs *enc_args = s->slice_args;
650 SliceArgs *top_loc[SLICE_REDIST_TOTAL] = {NULL};
651
652 init_quant_matrix(s);
653
654 for (slice_y = 0; slice_y < s->num_y; slice_y++) {
655 for (slice_x = 0; slice_x < s->num_x; slice_x++) {
656 SliceArgs *args = &enc_args[s->num_x*slice_y + slice_x];
657 args->ctx = s;
658 args->x = slice_x;
659 args->y = slice_y;
660 args->bits_ceil = s->slice_max_bytes << 3;
661 args->bits_floor = s->slice_min_bytes << 3;
662 memset(args->cache, 0, s->q_ceil*sizeof(*args->cache));
663 }
664 }
665
666 /* First pass - determine baseline slice sizes w.r.t. max_slice_size */
667 s->avctx->execute(s->avctx, rate_control, enc_args, NULL, s->num_x*s->num_y,
668 sizeof(SliceArgs));
669
670 for (i = 0; i < s->num_x*s->num_y; i++) {
671 SliceArgs *args = &enc_args[i];
672 bytes_left += args->bytes;
673 for (j = 0; j < slice_redist_range; j++) {
674 if (args->bytes > bytes_top[j]) {
675 bytes_top[j] = args->bytes;
676 top_loc[j] = args;
677 break;
678 }
679 }
680 }
681
682 bytes_left = s->frame_max_bytes - bytes_left;
683
684 /* Second pass - distribute leftover bytes */
685 while (bytes_left > 0) {
686 int distributed = 0;
687 for (i = 0; i < slice_redist_range; i++) {
688 SliceArgs *args;
689 int bits, bytes, diff, prev_bytes, new_idx;
690 if (bytes_left <= 0)
691 break;
692 if (!top_loc[i] || !top_loc[i]->quant_idx)
693 break;
694 args = top_loc[i];
695 prev_bytes = args->bytes;
696 new_idx = FFMAX(args->quant_idx - 1, 0);
697 bits = count_hq_slice(args, new_idx);
698 bytes = SSIZE_ROUND(bits >> 3);
699 diff = bytes - prev_bytes;
700 if ((bytes_left - diff) > 0) {
701 args->quant_idx = new_idx;
702 args->bytes = bytes;
703 bytes_left -= diff;
704 distributed++;
705 }
706 }
707 if (!distributed)
708 break;
709 }
710
711 for (i = 0; i < s->num_x*s->num_y; i++) {
712 SliceArgs *args = &enc_args[i];
713 total_bytes_needed += args->bytes;
714 s->q_avg = (s->q_avg + args->quant_idx)/2;
715 }
716
717 return total_bytes_needed;
718 }
719
720 /* VC-2 13.5.3 - hq_slice */
encode_hq_slice(AVCodecContext * avctx,void * arg)721 static int encode_hq_slice(AVCodecContext *avctx, void *arg)
722 {
723 SliceArgs *slice_dat = arg;
724 VC2EncContext *s = slice_dat->ctx;
725 PutBitContext *pb = &slice_dat->pb;
726 const int slice_x = slice_dat->x;
727 const int slice_y = slice_dat->y;
728 const int quant_idx = slice_dat->quant_idx;
729 const int slice_bytes_max = slice_dat->bytes;
730 uint8_t quants[MAX_DWT_LEVELS][4];
731 int p, level, orientation;
732
733 /* The reference decoder ignores it, and its typical length is 0 */
734 memset(put_bits_ptr(pb), 0, s->prefix_bytes);
735 skip_put_bytes(pb, s->prefix_bytes);
736
737 put_bits(pb, 8, quant_idx);
738
739 /* Slice quantization (slice_quantizers() in the specs) */
740 for (level = 0; level < s->wavelet_depth; level++)
741 for (orientation = !!level; orientation < 4; orientation++)
742 quants[level][orientation] = FFMAX(quant_idx - s->quant[level][orientation], 0);
743
744 /* Luma + 2 Chroma planes */
745 for (p = 0; p < 3; p++) {
746 int bytes_start, bytes_len, pad_s, pad_c;
747 bytes_start = put_bits_count(pb) >> 3;
748 put_bits(pb, 8, 0);
749 for (level = 0; level < s->wavelet_depth; level++) {
750 for (orientation = !!level; orientation < 4; orientation++) {
751 encode_subband(s, pb, slice_x, slice_y,
752 &s->plane[p].band[level][orientation],
753 quants[level][orientation]);
754 }
755 }
756 avpriv_align_put_bits(pb);
757 bytes_len = (put_bits_count(pb) >> 3) - bytes_start - 1;
758 if (p == 2) {
759 int len_diff = slice_bytes_max - (put_bits_count(pb) >> 3);
760 pad_s = FFALIGN((bytes_len + len_diff), s->size_scaler)/s->size_scaler;
761 pad_c = (pad_s*s->size_scaler) - bytes_len;
762 } else {
763 pad_s = FFALIGN(bytes_len, s->size_scaler)/s->size_scaler;
764 pad_c = (pad_s*s->size_scaler) - bytes_len;
765 }
766 pb->buf[bytes_start] = pad_s;
767 flush_put_bits(pb);
768 /* vc2-reference uses that padding that decodes to '0' coeffs */
769 memset(put_bits_ptr(pb), 0xFF, pad_c);
770 skip_put_bytes(pb, pad_c);
771 }
772
773 return 0;
774 }
775
776 /* VC-2 13.5.1 - low_delay_transform_data() */
encode_slices(VC2EncContext * s)777 static int encode_slices(VC2EncContext *s)
778 {
779 uint8_t *buf;
780 int slice_x, slice_y, skip = 0;
781 SliceArgs *enc_args = s->slice_args;
782
783 avpriv_align_put_bits(&s->pb);
784 flush_put_bits(&s->pb);
785 buf = put_bits_ptr(&s->pb);
786
787 for (slice_y = 0; slice_y < s->num_y; slice_y++) {
788 for (slice_x = 0; slice_x < s->num_x; slice_x++) {
789 SliceArgs *args = &enc_args[s->num_x*slice_y + slice_x];
790 init_put_bits(&args->pb, buf + skip, args->bytes+s->prefix_bytes);
791 skip += args->bytes;
792 }
793 }
794
795 s->avctx->execute(s->avctx, encode_hq_slice, enc_args, NULL, s->num_x*s->num_y,
796 sizeof(SliceArgs));
797
798 skip_put_bytes(&s->pb, skip);
799
800 return 0;
801 }
802
803 /*
804 * Transform basics for a 3 level transform
805 * |---------------------------------------------------------------------|
806 * | LL-0 | HL-0 | | |
807 * |--------|-------| HL-1 | |
808 * | LH-0 | HH-0 | | |
809 * |----------------|-----------------| HL-2 |
810 * | | | |
811 * | LH-1 | HH-1 | |
812 * | | | |
813 * |----------------------------------|----------------------------------|
814 * | | |
815 * | | |
816 * | | |
817 * | LH-2 | HH-2 |
818 * | | |
819 * | | |
820 * | | |
821 * |---------------------------------------------------------------------|
822 *
823 * DWT transforms are generally applied by splitting the image in two vertically
824 * and applying a low pass transform on the left part and a corresponding high
825 * pass transform on the right hand side. This is known as the horizontal filter
826 * stage.
827 * After that, the same operation is performed except the image is divided
828 * horizontally, with the high pass on the lower and the low pass on the higher
829 * side.
830 * Therefore, you're left with 4 subdivisions - known as low-low, low-high,
831 * high-low and high-high. They're referred to as orientations in the decoder
832 * and encoder.
833 *
834 * The LL (low-low) area contains the original image downsampled by the amount
835 * of levels. The rest of the areas can be thought as the details needed
836 * to restore the image perfectly to its original size.
837 */
dwt_plane(AVCodecContext * avctx,void * arg)838 static int dwt_plane(AVCodecContext *avctx, void *arg)
839 {
840 TransformArgs *transform_dat = arg;
841 VC2EncContext *s = transform_dat->ctx;
842 const void *frame_data = transform_dat->idata;
843 const ptrdiff_t linesize = transform_dat->istride;
844 const int field = transform_dat->field;
845 const Plane *p = transform_dat->plane;
846 VC2TransformContext *t = &transform_dat->t;
847 dwtcoef *buf = p->coef_buf;
848 const int idx = s->wavelet_idx;
849 const int skip = 1 + s->interlaced;
850
851 int x, y, level, offset;
852 ptrdiff_t pix_stride = linesize >> (s->bpp - 1);
853
854 if (field == 1) {
855 offset = 0;
856 pix_stride <<= 1;
857 } else if (field == 2) {
858 offset = pix_stride;
859 pix_stride <<= 1;
860 } else {
861 offset = 0;
862 }
863
864 if (s->bpp == 1) {
865 const uint8_t *pix = (const uint8_t *)frame_data + offset;
866 for (y = 0; y < p->height*skip; y+=skip) {
867 for (x = 0; x < p->width; x++) {
868 buf[x] = pix[x] - s->diff_offset;
869 }
870 memset(&buf[x], 0, (p->coef_stride - p->width)*sizeof(dwtcoef));
871 buf += p->coef_stride;
872 pix += pix_stride;
873 }
874 } else {
875 const uint16_t *pix = (const uint16_t *)frame_data + offset;
876 for (y = 0; y < p->height*skip; y+=skip) {
877 for (x = 0; x < p->width; x++) {
878 buf[x] = pix[x] - s->diff_offset;
879 }
880 memset(&buf[x], 0, (p->coef_stride - p->width)*sizeof(dwtcoef));
881 buf += p->coef_stride;
882 pix += pix_stride;
883 }
884 }
885
886 memset(buf, 0, p->coef_stride * (p->dwt_height - p->height) * sizeof(dwtcoef));
887
888 for (level = s->wavelet_depth-1; level >= 0; level--) {
889 const SubBand *b = &p->band[level][0];
890 t->vc2_subband_dwt[idx](t, p->coef_buf, p->coef_stride,
891 b->width, b->height);
892 }
893
894 return 0;
895 }
896
encode_frame(VC2EncContext * s,AVPacket * avpkt,const AVFrame * frame,const char * aux_data,const int header_size,int field)897 static int encode_frame(VC2EncContext *s, AVPacket *avpkt, const AVFrame *frame,
898 const char *aux_data, const int header_size, int field)
899 {
900 int i, ret;
901 int64_t max_frame_bytes;
902
903 /* Threaded DWT transform */
904 for (i = 0; i < 3; i++) {
905 s->transform_args[i].ctx = s;
906 s->transform_args[i].field = field;
907 s->transform_args[i].plane = &s->plane[i];
908 s->transform_args[i].idata = frame->data[i];
909 s->transform_args[i].istride = frame->linesize[i];
910 }
911 s->avctx->execute(s->avctx, dwt_plane, s->transform_args, NULL, 3,
912 sizeof(TransformArgs));
913
914 /* Calculate per-slice quantizers and sizes */
915 max_frame_bytes = header_size + calc_slice_sizes(s);
916
917 if (field < 2) {
918 ret = ff_alloc_packet2(s->avctx, avpkt,
919 max_frame_bytes << s->interlaced,
920 max_frame_bytes << s->interlaced);
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 avpriv_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
vc2_encode_frame(AVCodecContext * avctx,AVPacket * avpkt,const AVFrame * frame,int * got_packet)951 static 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
987 ret = encode_frame(s, avpkt, frame, aux_data, header_size, s->interlaced);
988 if (ret)
989 return ret;
990 if (s->interlaced) {
991 ret = encode_frame(s, avpkt, frame, aux_data, header_size, 2);
992 if (ret)
993 return ret;
994 }
995
996 flush_put_bits(&s->pb);
997 avpkt->size = put_bits_count(&s->pb) >> 3;
998
999 *got_packet = 1;
1000
1001 return 0;
1002 }
1003
vc2_encode_end(AVCodecContext * avctx)1004 static av_cold int vc2_encode_end(AVCodecContext *avctx)
1005 {
1006 int i;
1007 VC2EncContext *s = avctx->priv_data;
1008
1009 av_log(avctx, AV_LOG_INFO, "Qavg: %i\n", s->q_avg);
1010
1011 for (i = 0; i < 3; i++) {
1012 ff_vc2enc_free_transforms(&s->transform_args[i].t);
1013 av_freep(&s->plane[i].coef_buf);
1014 }
1015
1016 av_freep(&s->slice_args);
1017
1018 return 0;
1019 }
1020
vc2_encode_init(AVCodecContext * avctx)1021 static av_cold int vc2_encode_init(AVCodecContext *avctx)
1022 {
1023 Plane *p;
1024 SubBand *b;
1025 int i, level, o, shift, ret;
1026 const AVPixFmtDescriptor *fmt = av_pix_fmt_desc_get(avctx->pix_fmt);
1027 const int depth = fmt->comp[0].depth;
1028 VC2EncContext *s = avctx->priv_data;
1029
1030 s->picture_number = 0;
1031
1032 /* Total allowed quantization range */
1033 s->q_ceil = DIRAC_MAX_QUANT_INDEX;
1034
1035 s->ver.major = 2;
1036 s->ver.minor = 0;
1037 s->profile = 3;
1038 s->level = 3;
1039
1040 s->base_vf = -1;
1041 s->strict_compliance = 1;
1042
1043 s->q_avg = 0;
1044 s->slice_max_bytes = 0;
1045 s->slice_min_bytes = 0;
1046
1047 /* Mark unknown as progressive */
1048 s->interlaced = !((avctx->field_order == AV_FIELD_UNKNOWN) ||
1049 (avctx->field_order == AV_FIELD_PROGRESSIVE));
1050
1051 for (i = 0; i < base_video_fmts_len; i++) {
1052 const VC2BaseVideoFormat *fmt = &base_video_fmts[i];
1053 if (avctx->pix_fmt != fmt->pix_fmt)
1054 continue;
1055 if (avctx->time_base.num != fmt->time_base.num)
1056 continue;
1057 if (avctx->time_base.den != fmt->time_base.den)
1058 continue;
1059 if (avctx->width != fmt->width)
1060 continue;
1061 if (avctx->height != fmt->height)
1062 continue;
1063 if (s->interlaced != fmt->interlaced)
1064 continue;
1065 s->base_vf = i;
1066 s->level = base_video_fmts[i].level;
1067 break;
1068 }
1069
1070 if (s->interlaced)
1071 av_log(avctx, AV_LOG_WARNING, "Interlacing enabled!\n");
1072
1073 if ((s->slice_width & (s->slice_width - 1)) ||
1074 (s->slice_height & (s->slice_height - 1))) {
1075 av_log(avctx, AV_LOG_ERROR, "Slice size is not a power of two!\n");
1076 return AVERROR_UNKNOWN;
1077 }
1078
1079 if ((s->slice_width > avctx->width) ||
1080 (s->slice_height > avctx->height)) {
1081 av_log(avctx, AV_LOG_ERROR, "Slice size is bigger than the image!\n");
1082 return AVERROR_UNKNOWN;
1083 }
1084
1085 if (s->base_vf <= 0) {
1086 if (avctx->strict_std_compliance < FF_COMPLIANCE_STRICT) {
1087 s->strict_compliance = s->base_vf = 0;
1088 av_log(avctx, AV_LOG_WARNING, "Format does not strictly comply with VC2 specs\n");
1089 } else {
1090 av_log(avctx, AV_LOG_WARNING, "Given format does not strictly comply with "
1091 "the specifications, decrease strictness to use it.\n");
1092 return AVERROR_UNKNOWN;
1093 }
1094 } else {
1095 av_log(avctx, AV_LOG_INFO, "Selected base video format = %i (%s)\n",
1096 s->base_vf, base_video_fmts[s->base_vf].name);
1097 }
1098
1099 /* Chroma subsampling */
1100 ret = av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
1101 if (ret)
1102 return ret;
1103
1104 /* Bit depth and color range index */
1105 if (depth == 8 && avctx->color_range == AVCOL_RANGE_JPEG) {
1106 s->bpp = 1;
1107 s->bpp_idx = 1;
1108 s->diff_offset = 128;
1109 } else if (depth == 8 && (avctx->color_range == AVCOL_RANGE_MPEG ||
1110 avctx->color_range == AVCOL_RANGE_UNSPECIFIED)) {
1111 s->bpp = 1;
1112 s->bpp_idx = 2;
1113 s->diff_offset = 128;
1114 } else if (depth == 10) {
1115 s->bpp = 2;
1116 s->bpp_idx = 3;
1117 s->diff_offset = 512;
1118 } else {
1119 s->bpp = 2;
1120 s->bpp_idx = 4;
1121 s->diff_offset = 2048;
1122 }
1123
1124 /* Planes initialization */
1125 for (i = 0; i < 3; i++) {
1126 int w, h;
1127 p = &s->plane[i];
1128 p->width = avctx->width >> (i ? s->chroma_x_shift : 0);
1129 p->height = avctx->height >> (i ? s->chroma_y_shift : 0);
1130 if (s->interlaced)
1131 p->height >>= 1;
1132 p->dwt_width = w = FFALIGN(p->width, (1 << s->wavelet_depth));
1133 p->dwt_height = h = FFALIGN(p->height, (1 << s->wavelet_depth));
1134 p->coef_stride = FFALIGN(p->dwt_width, 32);
1135 p->coef_buf = av_mallocz(p->coef_stride*p->dwt_height*sizeof(dwtcoef));
1136 if (!p->coef_buf)
1137 goto alloc_fail;
1138 for (level = s->wavelet_depth-1; level >= 0; level--) {
1139 w = w >> 1;
1140 h = h >> 1;
1141 for (o = 0; o < 4; o++) {
1142 b = &p->band[level][o];
1143 b->width = w;
1144 b->height = h;
1145 b->stride = p->coef_stride;
1146 shift = (o > 1)*b->height*b->stride + (o & 1)*b->width;
1147 b->buf = p->coef_buf + shift;
1148 }
1149 }
1150
1151 /* DWT init */
1152 if (ff_vc2enc_init_transforms(&s->transform_args[i].t,
1153 s->plane[i].coef_stride,
1154 s->plane[i].dwt_height,
1155 s->slice_width, s->slice_height))
1156 goto alloc_fail;
1157 }
1158
1159 /* Slices */
1160 s->num_x = s->plane[0].dwt_width/s->slice_width;
1161 s->num_y = s->plane[0].dwt_height/s->slice_height;
1162
1163 s->slice_args = av_calloc(s->num_x*s->num_y, sizeof(SliceArgs));
1164 if (!s->slice_args)
1165 goto alloc_fail;
1166
1167 for (i = 0; i < 116; i++) {
1168 const uint64_t qf = ff_dirac_qscale_tab[i];
1169 const uint32_t m = av_log2(qf);
1170 const uint32_t t = (1ULL << (m + 32)) / qf;
1171 const uint32_t r = (t*qf + qf) & UINT32_MAX;
1172 if (!(qf & (qf - 1))) {
1173 s->qmagic_lut[i][0] = 0xFFFFFFFF;
1174 s->qmagic_lut[i][1] = 0xFFFFFFFF;
1175 } else if (r <= 1 << m) {
1176 s->qmagic_lut[i][0] = t + 1;
1177 s->qmagic_lut[i][1] = 0;
1178 } else {
1179 s->qmagic_lut[i][0] = t;
1180 s->qmagic_lut[i][1] = t;
1181 }
1182 }
1183
1184 return 0;
1185
1186 alloc_fail:
1187 vc2_encode_end(avctx);
1188 av_log(avctx, AV_LOG_ERROR, "Unable to allocate memory!\n");
1189 return AVERROR(ENOMEM);
1190 }
1191
1192 #define VC2ENC_FLAGS (AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_VIDEO_PARAM)
1193 static const AVOption vc2enc_options[] = {
1194 {"tolerance", "Max undershoot in percent", offsetof(VC2EncContext, tolerance), AV_OPT_TYPE_DOUBLE, {.dbl = 5.0f}, 0.0f, 45.0f, VC2ENC_FLAGS, "tolerance"},
1195 {"slice_width", "Slice width", offsetof(VC2EncContext, slice_width), AV_OPT_TYPE_INT, {.i64 = 32}, 32, 1024, VC2ENC_FLAGS, "slice_width"},
1196 {"slice_height", "Slice height", offsetof(VC2EncContext, slice_height), AV_OPT_TYPE_INT, {.i64 = 16}, 8, 1024, VC2ENC_FLAGS, "slice_height"},
1197 {"wavelet_depth", "Transform depth", offsetof(VC2EncContext, wavelet_depth), AV_OPT_TYPE_INT, {.i64 = 4}, 1, 5, VC2ENC_FLAGS, "wavelet_depth"},
1198 {"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"},
1199 {"9_7", "Deslauriers-Dubuc (9,7)", 0, AV_OPT_TYPE_CONST, {.i64 = VC2_TRANSFORM_9_7}, INT_MIN, INT_MAX, VC2ENC_FLAGS, "wavelet_idx"},
1200 {"5_3", "LeGall (5,3)", 0, AV_OPT_TYPE_CONST, {.i64 = VC2_TRANSFORM_5_3}, INT_MIN, INT_MAX, VC2ENC_FLAGS, "wavelet_idx"},
1201 {"haar", "Haar (with shift)", 0, AV_OPT_TYPE_CONST, {.i64 = VC2_TRANSFORM_HAAR_S}, INT_MIN, INT_MAX, VC2ENC_FLAGS, "wavelet_idx"},
1202 {"haar_noshift", "Haar (without shift)", 0, AV_OPT_TYPE_CONST, {.i64 = VC2_TRANSFORM_HAAR}, INT_MIN, INT_MAX, VC2ENC_FLAGS, "wavelet_idx"},
1203 {"qm", "Custom quantization matrix", offsetof(VC2EncContext, quant_matrix), AV_OPT_TYPE_INT, {.i64 = VC2_QM_DEF}, 0, VC2_QM_NB, VC2ENC_FLAGS, "quant_matrix"},
1204 {"default", "Default from the specifications", 0, AV_OPT_TYPE_CONST, {.i64 = VC2_QM_DEF}, INT_MIN, INT_MAX, VC2ENC_FLAGS, "quant_matrix"},
1205 {"color", "Prevents low bitrate discoloration", 0, AV_OPT_TYPE_CONST, {.i64 = VC2_QM_COL}, INT_MIN, INT_MAX, VC2ENC_FLAGS, "quant_matrix"},
1206 {"flat", "Optimize for PSNR", 0, AV_OPT_TYPE_CONST, {.i64 = VC2_QM_FLAT}, INT_MIN, INT_MAX, VC2ENC_FLAGS, "quant_matrix"},
1207 {NULL}
1208 };
1209
1210 static const AVClass vc2enc_class = {
1211 .class_name = "SMPTE VC-2 encoder",
1212 .category = AV_CLASS_CATEGORY_ENCODER,
1213 .option = vc2enc_options,
1214 .item_name = av_default_item_name,
1215 .version = LIBAVUTIL_VERSION_INT
1216 };
1217
1218 static const AVCodecDefault vc2enc_defaults[] = {
1219 { "b", "600000000" },
1220 { NULL },
1221 };
1222
1223 static const enum AVPixelFormat allowed_pix_fmts[] = {
1224 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P,
1225 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
1226 AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12,
1227 AV_PIX_FMT_NONE
1228 };
1229
1230 AVCodec ff_vc2_encoder = {
1231 .name = "vc2",
1232 .long_name = NULL_IF_CONFIG_SMALL("SMPTE VC-2"),
1233 .type = AVMEDIA_TYPE_VIDEO,
1234 .id = AV_CODEC_ID_DIRAC,
1235 .priv_data_size = sizeof(VC2EncContext),
1236 .init = vc2_encode_init,
1237 .close = vc2_encode_end,
1238 .capabilities = AV_CODEC_CAP_SLICE_THREADS,
1239 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
1240 .encode2 = vc2_encode_frame,
1241 .priv_class = &vc2enc_class,
1242 .defaults = vc2enc_defaults,
1243 .pix_fmts = allowed_pix_fmts
1244 };
1245