1 /*
2 * AV1 video decoder
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21 #include "libavutil/film_grain_params.h"
22 #include "libavutil/pixdesc.h"
23 #include "libavutil/opt.h"
24 #include "avcodec.h"
25 #include "av1dec.h"
26 #include "bytestream.h"
27 #include "hwconfig.h"
28 #include "internal.h"
29 #include "profiles.h"
30
inverse_recenter(int r,uint32_t v)31 static uint32_t inverse_recenter(int r, uint32_t v)
32 {
33 if (v > 2 * r)
34 return v;
35 else if (v & 1)
36 return r - ((v + 1) >> 1);
37 else
38 return r + (v >> 1);
39 }
40
decode_unsigned_subexp_with_ref(uint32_t sub_exp,int mx,int r)41 static uint32_t decode_unsigned_subexp_with_ref(uint32_t sub_exp,
42 int mx, int r)
43 {
44 if ((r << 1) <= mx) {
45 return inverse_recenter(r, sub_exp);
46 } else {
47 return mx - 1 - inverse_recenter(mx - 1 - r, sub_exp);
48 }
49 }
50
decode_signed_subexp_with_ref(uint32_t sub_exp,int low,int high,int r)51 static int32_t decode_signed_subexp_with_ref(uint32_t sub_exp, int low,
52 int high, int r)
53 {
54 int32_t x = decode_unsigned_subexp_with_ref(sub_exp, high - low, r - low);
55 return x + low;
56 }
57
read_global_param(AV1DecContext * s,int type,int ref,int idx)58 static void read_global_param(AV1DecContext *s, int type, int ref, int idx)
59 {
60 uint8_t primary_frame, prev_frame;
61 uint32_t abs_bits, prec_bits, round, prec_diff, sub, mx;
62 int32_t r, prev_gm_param;
63
64 primary_frame = s->raw_frame_header->primary_ref_frame;
65 prev_frame = s->raw_frame_header->ref_frame_idx[primary_frame];
66 abs_bits = AV1_GM_ABS_ALPHA_BITS;
67 prec_bits = AV1_GM_ALPHA_PREC_BITS;
68
69 /* setup_past_independence() sets PrevGmParams to default values. We can
70 * simply point to the current's frame gm_params as they will be initialized
71 * with defaults at this point.
72 */
73 if (s->raw_frame_header->primary_ref_frame == AV1_PRIMARY_REF_NONE)
74 prev_gm_param = s->cur_frame.gm_params[ref][idx];
75 else
76 prev_gm_param = s->ref[prev_frame].gm_params[ref][idx];
77
78 if (idx < 2) {
79 if (type == AV1_WARP_MODEL_TRANSLATION) {
80 abs_bits = AV1_GM_ABS_TRANS_ONLY_BITS -
81 !s->raw_frame_header->allow_high_precision_mv;
82 prec_bits = AV1_GM_TRANS_ONLY_PREC_BITS -
83 !s->raw_frame_header->allow_high_precision_mv;
84 } else {
85 abs_bits = AV1_GM_ABS_TRANS_BITS;
86 prec_bits = AV1_GM_TRANS_PREC_BITS;
87 }
88 }
89 round = (idx % 3) == 2 ? (1 << AV1_WARPEDMODEL_PREC_BITS) : 0;
90 prec_diff = AV1_WARPEDMODEL_PREC_BITS - prec_bits;
91 sub = (idx % 3) == 2 ? (1 << prec_bits) : 0;
92 mx = 1 << abs_bits;
93 r = (prev_gm_param >> prec_diff) - sub;
94
95 s->cur_frame.gm_params[ref][idx] =
96 (decode_signed_subexp_with_ref(s->raw_frame_header->gm_params[ref][idx],
97 -mx, mx + 1, r) << prec_diff) + round;
98 }
99
100 /**
101 * update gm type/params, since cbs already implemented part of this funcation,
102 * so we don't need to full implement spec.
103 */
global_motion_params(AV1DecContext * s)104 static void global_motion_params(AV1DecContext *s)
105 {
106 const AV1RawFrameHeader *header = s->raw_frame_header;
107 int type, ref;
108
109 for (ref = AV1_REF_FRAME_LAST; ref <= AV1_REF_FRAME_ALTREF; ref++) {
110 s->cur_frame.gm_type[ref] = AV1_WARP_MODEL_IDENTITY;
111 for (int i = 0; i < 6; i++)
112 s->cur_frame.gm_params[ref][i] = (i % 3 == 2) ?
113 1 << AV1_WARPEDMODEL_PREC_BITS : 0;
114 }
115 if (header->frame_type == AV1_FRAME_KEY ||
116 header->frame_type == AV1_FRAME_INTRA_ONLY)
117 return;
118
119 for (ref = AV1_REF_FRAME_LAST; ref <= AV1_REF_FRAME_ALTREF; ref++) {
120 if (header->is_global[ref]) {
121 if (header->is_rot_zoom[ref]) {
122 type = AV1_WARP_MODEL_ROTZOOM;
123 } else {
124 type = header->is_translation[ref] ? AV1_WARP_MODEL_TRANSLATION
125 : AV1_WARP_MODEL_AFFINE;
126 }
127 } else {
128 type = AV1_WARP_MODEL_IDENTITY;
129 }
130 s->cur_frame.gm_type[ref] = type;
131
132 if (type >= AV1_WARP_MODEL_ROTZOOM) {
133 read_global_param(s, type, ref, 2);
134 read_global_param(s, type, ref, 3);
135 if (type == AV1_WARP_MODEL_AFFINE) {
136 read_global_param(s, type, ref, 4);
137 read_global_param(s, type, ref, 5);
138 } else {
139 s->cur_frame.gm_params[ref][4] = -s->cur_frame.gm_params[ref][3];
140 s->cur_frame.gm_params[ref][5] = s->cur_frame.gm_params[ref][2];
141 }
142 }
143 if (type >= AV1_WARP_MODEL_TRANSLATION) {
144 read_global_param(s, type, ref, 0);
145 read_global_param(s, type, ref, 1);
146 }
147 }
148 }
149
get_relative_dist(const AV1RawSequenceHeader * seq,unsigned int a,unsigned int b)150 static int get_relative_dist(const AV1RawSequenceHeader *seq,
151 unsigned int a, unsigned int b)
152 {
153 unsigned int diff = a - b;
154 unsigned int m = 1 << seq->order_hint_bits_minus_1;
155 return (diff & (m - 1)) - (diff & m);
156 }
157
skip_mode_params(AV1DecContext * s)158 static void skip_mode_params(AV1DecContext *s)
159 {
160 const AV1RawFrameHeader *header = s->raw_frame_header;
161 const AV1RawSequenceHeader *seq = s->raw_seq;
162
163 int forward_idx, backward_idx;
164 int forward_hint, backward_hint;
165 int second_forward_idx, second_forward_hint;
166 int ref_hint, dist, i;
167
168 if (!header->skip_mode_present)
169 return;
170
171 forward_idx = -1;
172 backward_idx = -1;
173 for (i = 0; i < AV1_REFS_PER_FRAME; i++) {
174 ref_hint = s->ref[header->ref_frame_idx[i]].raw_frame_header->order_hint;
175 dist = get_relative_dist(seq, ref_hint, header->order_hint);
176 if (dist < 0) {
177 if (forward_idx < 0 ||
178 get_relative_dist(seq, ref_hint, forward_hint) > 0) {
179 forward_idx = i;
180 forward_hint = ref_hint;
181 }
182 } else if (dist > 0) {
183 if (backward_idx < 0 ||
184 get_relative_dist(seq, ref_hint, backward_hint) < 0) {
185 backward_idx = i;
186 backward_hint = ref_hint;
187 }
188 }
189 }
190
191 if (forward_idx < 0) {
192 return;
193 } else if (backward_idx >= 0) {
194 s->cur_frame.skip_mode_frame_idx[0] =
195 AV1_REF_FRAME_LAST + FFMIN(forward_idx, backward_idx);
196 s->cur_frame.skip_mode_frame_idx[1] =
197 AV1_REF_FRAME_LAST + FFMAX(forward_idx, backward_idx);
198 return;
199 }
200
201 second_forward_idx = -1;
202 for (i = 0; i < AV1_REFS_PER_FRAME; i++) {
203 ref_hint = s->ref[header->ref_frame_idx[i]].raw_frame_header->order_hint;
204 if (get_relative_dist(seq, ref_hint, forward_hint) < 0) {
205 if (second_forward_idx < 0 ||
206 get_relative_dist(seq, ref_hint, second_forward_hint) > 0) {
207 second_forward_idx = i;
208 second_forward_hint = ref_hint;
209 }
210 }
211 }
212
213 if (second_forward_idx < 0)
214 return;
215
216 s->cur_frame.skip_mode_frame_idx[0] =
217 AV1_REF_FRAME_LAST + FFMIN(forward_idx, second_forward_idx);
218 s->cur_frame.skip_mode_frame_idx[1] =
219 AV1_REF_FRAME_LAST + FFMAX(forward_idx, second_forward_idx);
220 }
221
coded_lossless_param(AV1DecContext * s)222 static void coded_lossless_param(AV1DecContext *s)
223 {
224 const AV1RawFrameHeader *header = s->raw_frame_header;
225 int i;
226
227 if (header->delta_q_y_dc || header->delta_q_u_ac ||
228 header->delta_q_u_dc || header->delta_q_v_ac ||
229 header->delta_q_v_dc) {
230 s->cur_frame.coded_lossless = 0;
231 return;
232 }
233
234 s->cur_frame.coded_lossless = 1;
235 for (i = 0; i < AV1_MAX_SEGMENTS; i++) {
236 int qindex;
237 if (header->feature_enabled[i][AV1_SEG_LVL_ALT_Q]) {
238 qindex = (header->base_q_idx +
239 header->feature_value[i][AV1_SEG_LVL_ALT_Q]);
240 } else {
241 qindex = header->base_q_idx;
242 }
243 qindex = av_clip_uintp2(qindex, 8);
244
245 if (qindex) {
246 s->cur_frame.coded_lossless = 0;
247 return;
248 }
249 }
250 }
251
load_grain_params(AV1DecContext * s)252 static void load_grain_params(AV1DecContext *s)
253 {
254 const AV1RawFrameHeader *header = s->raw_frame_header;
255 const AV1RawFilmGrainParams *film_grain = &header->film_grain, *src;
256 AV1RawFilmGrainParams *dst = &s->cur_frame.film_grain;
257
258 if (!film_grain->apply_grain)
259 return;
260
261 if (film_grain->update_grain) {
262 memcpy(dst, film_grain, sizeof(*dst));
263 return;
264 }
265
266 src = &s->ref[film_grain->film_grain_params_ref_idx].film_grain;
267
268 memcpy(dst, src, sizeof(*dst));
269 dst->grain_seed = film_grain->grain_seed;
270 }
271
init_tile_data(AV1DecContext * s)272 static int init_tile_data(AV1DecContext *s)
273
274 {
275 int cur_tile_num =
276 s->raw_frame_header->tile_cols * s->raw_frame_header->tile_rows;
277 if (s->tile_num < cur_tile_num) {
278 int ret = av_reallocp_array(&s->tile_group_info, cur_tile_num,
279 sizeof(TileGroupInfo));
280 if (ret < 0) {
281 s->tile_num = 0;
282 return ret;
283 }
284 }
285 s->tile_num = cur_tile_num;
286
287 return 0;
288 }
289
get_tiles_info(AVCodecContext * avctx,const AV1RawTileGroup * tile_group)290 static int get_tiles_info(AVCodecContext *avctx, const AV1RawTileGroup *tile_group)
291 {
292 AV1DecContext *s = avctx->priv_data;
293 GetByteContext gb;
294 uint16_t tile_num, tile_row, tile_col;
295 uint32_t size = 0, size_bytes = 0;
296
297 bytestream2_init(&gb, tile_group->tile_data.data,
298 tile_group->tile_data.data_size);
299 s->tg_start = tile_group->tg_start;
300 s->tg_end = tile_group->tg_end;
301
302 for (tile_num = tile_group->tg_start; tile_num <= tile_group->tg_end; tile_num++) {
303 tile_row = tile_num / s->raw_frame_header->tile_cols;
304 tile_col = tile_num % s->raw_frame_header->tile_cols;
305
306 if (tile_num == tile_group->tg_end) {
307 s->tile_group_info[tile_num].tile_size = bytestream2_get_bytes_left(&gb);
308 s->tile_group_info[tile_num].tile_offset = bytestream2_tell(&gb);
309 s->tile_group_info[tile_num].tile_row = tile_row;
310 s->tile_group_info[tile_num].tile_column = tile_col;
311 return 0;
312 }
313 size_bytes = s->raw_frame_header->tile_size_bytes_minus1 + 1;
314 if (bytestream2_get_bytes_left(&gb) < size_bytes)
315 return AVERROR_INVALIDDATA;
316 size = 0;
317 for (int i = 0; i < size_bytes; i++)
318 size |= bytestream2_get_byteu(&gb) << 8 * i;
319 if (bytestream2_get_bytes_left(&gb) <= size)
320 return AVERROR_INVALIDDATA;
321 size++;
322
323 s->tile_group_info[tile_num].tile_size = size;
324 s->tile_group_info[tile_num].tile_offset = bytestream2_tell(&gb);
325 s->tile_group_info[tile_num].tile_row = tile_row;
326 s->tile_group_info[tile_num].tile_column = tile_col;
327
328 bytestream2_skipu(&gb, size);
329 }
330
331 return 0;
332
333 }
334
get_pixel_format(AVCodecContext * avctx)335 static int get_pixel_format(AVCodecContext *avctx)
336 {
337 AV1DecContext *s = avctx->priv_data;
338 const AV1RawSequenceHeader *seq = s->raw_seq;
339 uint8_t bit_depth;
340 int ret;
341 enum AVPixelFormat pix_fmt = AV_PIX_FMT_NONE;
342 #define HWACCEL_MAX (CONFIG_AV1_DXVA2_HWACCEL + \
343 CONFIG_AV1_D3D11VA_HWACCEL * 2 + \
344 CONFIG_AV1_NVDEC_HWACCEL + \
345 CONFIG_AV1_VAAPI_HWACCEL)
346 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmtp = pix_fmts;
347
348 if (seq->seq_profile == 2 && seq->color_config.high_bitdepth)
349 bit_depth = seq->color_config.twelve_bit ? 12 : 10;
350 else if (seq->seq_profile <= 2)
351 bit_depth = seq->color_config.high_bitdepth ? 10 : 8;
352 else {
353 av_log(avctx, AV_LOG_ERROR,
354 "Unknown AV1 profile %d.\n", seq->seq_profile);
355 return -1;
356 }
357
358 if (!seq->color_config.mono_chrome) {
359 // 4:4:4 x:0 y:0, 4:2:2 x:1 y:0, 4:2:0 x:1 y:1
360 if (seq->color_config.subsampling_x == 0 &&
361 seq->color_config.subsampling_y == 0) {
362 if (bit_depth == 8)
363 pix_fmt = AV_PIX_FMT_YUV444P;
364 else if (bit_depth == 10)
365 pix_fmt = AV_PIX_FMT_YUV444P10;
366 else if (bit_depth == 12)
367 pix_fmt = AV_PIX_FMT_YUV444P12;
368 else
369 av_log(avctx, AV_LOG_WARNING, "Unknown AV1 pixel format.\n");
370 } else if (seq->color_config.subsampling_x == 1 &&
371 seq->color_config.subsampling_y == 0) {
372 if (bit_depth == 8)
373 pix_fmt = AV_PIX_FMT_YUV422P;
374 else if (bit_depth == 10)
375 pix_fmt = AV_PIX_FMT_YUV422P10;
376 else if (bit_depth == 12)
377 pix_fmt = AV_PIX_FMT_YUV422P12;
378 else
379 av_log(avctx, AV_LOG_WARNING, "Unknown AV1 pixel format.\n");
380 } else if (seq->color_config.subsampling_x == 1 &&
381 seq->color_config.subsampling_y == 1) {
382 if (bit_depth == 8)
383 pix_fmt = AV_PIX_FMT_YUV420P;
384 else if (bit_depth == 10)
385 pix_fmt = AV_PIX_FMT_YUV420P10;
386 else if (bit_depth == 12)
387 pix_fmt = AV_PIX_FMT_YUV420P12;
388 else
389 av_log(avctx, AV_LOG_WARNING, "Unknown AV1 pixel format.\n");
390 }
391 } else {
392 if (bit_depth == 8)
393 pix_fmt = AV_PIX_FMT_GRAY8;
394 else if (bit_depth == 10)
395 pix_fmt = AV_PIX_FMT_GRAY10;
396 else if (bit_depth == 12)
397 pix_fmt = AV_PIX_FMT_GRAY12;
398 else
399 av_log(avctx, AV_LOG_WARNING, "Unknown AV1 pixel format.\n");
400 }
401
402 av_log(avctx, AV_LOG_DEBUG, "AV1 decode get format: %s.\n",
403 av_get_pix_fmt_name(pix_fmt));
404
405 if (pix_fmt == AV_PIX_FMT_NONE)
406 return -1;
407 s->pix_fmt = pix_fmt;
408
409 switch (s->pix_fmt) {
410 case AV_PIX_FMT_YUV420P:
411 #if CONFIG_AV1_DXVA2_HWACCEL
412 *fmtp++ = AV_PIX_FMT_DXVA2_VLD;
413 #endif
414 #if CONFIG_AV1_D3D11VA_HWACCEL
415 *fmtp++ = AV_PIX_FMT_D3D11VA_VLD;
416 *fmtp++ = AV_PIX_FMT_D3D11;
417 #endif
418 #if CONFIG_AV1_NVDEC_HWACCEL
419 *fmtp++ = AV_PIX_FMT_CUDA;
420 #endif
421 #if CONFIG_AV1_VAAPI_HWACCEL
422 *fmtp++ = AV_PIX_FMT_VAAPI;
423 #endif
424 break;
425 case AV_PIX_FMT_YUV420P10:
426 #if CONFIG_AV1_DXVA2_HWACCEL
427 *fmtp++ = AV_PIX_FMT_DXVA2_VLD;
428 #endif
429 #if CONFIG_AV1_D3D11VA_HWACCEL
430 *fmtp++ = AV_PIX_FMT_D3D11VA_VLD;
431 *fmtp++ = AV_PIX_FMT_D3D11;
432 #endif
433 #if CONFIG_AV1_NVDEC_HWACCEL
434 *fmtp++ = AV_PIX_FMT_CUDA;
435 #endif
436 #if CONFIG_AV1_VAAPI_HWACCEL
437 *fmtp++ = AV_PIX_FMT_VAAPI;
438 #endif
439 break;
440 case AV_PIX_FMT_GRAY8:
441 #if CONFIG_AV1_NVDEC_HWACCEL
442 *fmtp++ = AV_PIX_FMT_CUDA;
443 #endif
444 break;
445 case AV_PIX_FMT_GRAY10:
446 #if CONFIG_AV1_NVDEC_HWACCEL
447 *fmtp++ = AV_PIX_FMT_CUDA;
448 #endif
449 break;
450 }
451
452 *fmtp++ = s->pix_fmt;
453 *fmtp = AV_PIX_FMT_NONE;
454
455 ret = ff_thread_get_format(avctx, pix_fmts);
456 if (ret < 0)
457 return ret;
458
459 /**
460 * check if the HW accel is inited correctly. If not, return un-implemented.
461 * Since now the av1 decoder doesn't support native decode, if it will be
462 * implemented in the future, need remove this check.
463 */
464 if (!avctx->hwaccel) {
465 av_log(avctx, AV_LOG_ERROR, "Your platform doesn't suppport"
466 " hardware accelerated AV1 decoding.\n");
467 return AVERROR(ENOSYS);
468 }
469
470 avctx->pix_fmt = ret;
471
472 return 0;
473 }
474
av1_frame_unref(AVCodecContext * avctx,AV1Frame * f)475 static void av1_frame_unref(AVCodecContext *avctx, AV1Frame *f)
476 {
477 ff_thread_release_buffer(avctx, &f->tf);
478 av_buffer_unref(&f->hwaccel_priv_buf);
479 f->hwaccel_picture_private = NULL;
480 av_buffer_unref(&f->header_ref);
481 f->raw_frame_header = NULL;
482 f->spatial_id = f->temporal_id = 0;
483 memset(f->skip_mode_frame_idx, 0,
484 2 * sizeof(uint8_t));
485 memset(&f->film_grain, 0, sizeof(f->film_grain));
486 f->coded_lossless = 0;
487 }
488
av1_frame_ref(AVCodecContext * avctx,AV1Frame * dst,const AV1Frame * src)489 static int av1_frame_ref(AVCodecContext *avctx, AV1Frame *dst, const AV1Frame *src)
490 {
491 int ret;
492
493 ret = ff_thread_ref_frame(&dst->tf, &src->tf);
494 if (ret < 0)
495 return ret;
496
497 dst->header_ref = av_buffer_ref(src->header_ref);
498 if (!dst->header_ref)
499 goto fail;
500
501 dst->raw_frame_header = src->raw_frame_header;
502
503 if (src->hwaccel_picture_private) {
504 dst->hwaccel_priv_buf = av_buffer_ref(src->hwaccel_priv_buf);
505 if (!dst->hwaccel_priv_buf)
506 goto fail;
507 dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data;
508 }
509
510 dst->spatial_id = src->spatial_id;
511 dst->temporal_id = src->temporal_id;
512 memcpy(dst->gm_type,
513 src->gm_type,
514 AV1_NUM_REF_FRAMES * sizeof(uint8_t));
515 memcpy(dst->gm_params,
516 src->gm_params,
517 AV1_NUM_REF_FRAMES * 6 * sizeof(int32_t));
518 memcpy(dst->skip_mode_frame_idx,
519 src->skip_mode_frame_idx,
520 2 * sizeof(uint8_t));
521 memcpy(&dst->film_grain,
522 &src->film_grain,
523 sizeof(dst->film_grain));
524 dst->coded_lossless = src->coded_lossless;
525
526 return 0;
527
528 fail:
529 av1_frame_unref(avctx, dst);
530 return AVERROR(ENOMEM);
531 }
532
av1_decode_free(AVCodecContext * avctx)533 static av_cold int av1_decode_free(AVCodecContext *avctx)
534 {
535 AV1DecContext *s = avctx->priv_data;
536
537 for (int i = 0; i < FF_ARRAY_ELEMS(s->ref); i++) {
538 av1_frame_unref(avctx, &s->ref[i]);
539 av_frame_free(&s->ref[i].tf.f);
540 }
541 av1_frame_unref(avctx, &s->cur_frame);
542 av_frame_free(&s->cur_frame.tf.f);
543
544 av_buffer_unref(&s->seq_ref);
545 av_buffer_unref(&s->header_ref);
546 av_freep(&s->tile_group_info);
547
548 ff_cbs_fragment_free(&s->current_obu);
549 ff_cbs_close(&s->cbc);
550
551 return 0;
552 }
553
set_context_with_sequence(AVCodecContext * avctx,const AV1RawSequenceHeader * seq)554 static int set_context_with_sequence(AVCodecContext *avctx,
555 const AV1RawSequenceHeader *seq)
556 {
557 int width = seq->max_frame_width_minus_1 + 1;
558 int height = seq->max_frame_height_minus_1 + 1;
559
560 avctx->profile = seq->seq_profile;
561 avctx->level = seq->seq_level_idx[0];
562
563 avctx->color_range =
564 seq->color_config.color_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
565 avctx->color_primaries = seq->color_config.color_primaries;
566 avctx->colorspace = seq->color_config.color_primaries;
567 avctx->color_trc = seq->color_config.transfer_characteristics;
568
569 switch (seq->color_config.chroma_sample_position) {
570 case AV1_CSP_VERTICAL:
571 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
572 break;
573 case AV1_CSP_COLOCATED:
574 avctx->chroma_sample_location = AVCHROMA_LOC_TOPLEFT;
575 break;
576 }
577
578 if (avctx->width != width || avctx->height != height) {
579 int ret = ff_set_dimensions(avctx, width, height);
580 if (ret < 0)
581 return ret;
582 }
583 avctx->sample_aspect_ratio = (AVRational) { 1, 1 };
584
585 if (seq->timing_info.num_units_in_display_tick &&
586 seq->timing_info.time_scale) {
587 av_reduce(&avctx->framerate.den, &avctx->framerate.num,
588 seq->timing_info.num_units_in_display_tick,
589 seq->timing_info.time_scale,
590 INT_MAX);
591 if (seq->timing_info.equal_picture_interval)
592 avctx->ticks_per_frame = seq->timing_info.num_ticks_per_picture_minus_1 + 1;
593 }
594
595 return 0;
596 }
597
update_context_with_frame_header(AVCodecContext * avctx,const AV1RawFrameHeader * header)598 static int update_context_with_frame_header(AVCodecContext *avctx,
599 const AV1RawFrameHeader *header)
600 {
601 AVRational aspect_ratio;
602 int width = header->frame_width_minus_1 + 1;
603 int height = header->frame_height_minus_1 + 1;
604 int r_width = header->render_width_minus_1 + 1;
605 int r_height = header->render_height_minus_1 + 1;
606 int ret;
607
608 if (avctx->width != width || avctx->height != height) {
609 ret = ff_set_dimensions(avctx, width, height);
610 if (ret < 0)
611 return ret;
612 }
613
614 av_reduce(&aspect_ratio.num, &aspect_ratio.den,
615 (int64_t)height * r_width,
616 (int64_t)width * r_height,
617 INT_MAX);
618
619 if (av_cmp_q(avctx->sample_aspect_ratio, aspect_ratio)) {
620 ret = ff_set_sar(avctx, aspect_ratio);
621 if (ret < 0)
622 return ret;
623 }
624
625 return 0;
626 }
627
av1_decode_init(AVCodecContext * avctx)628 static av_cold int av1_decode_init(AVCodecContext *avctx)
629 {
630 AV1DecContext *s = avctx->priv_data;
631 AV1RawSequenceHeader *seq;
632 int ret;
633
634 s->avctx = avctx;
635 s->pix_fmt = AV_PIX_FMT_NONE;
636
637 for (int i = 0; i < FF_ARRAY_ELEMS(s->ref); i++) {
638 s->ref[i].tf.f = av_frame_alloc();
639 if (!s->ref[i].tf.f) {
640 av_log(avctx, AV_LOG_ERROR,
641 "Failed to allocate reference frame buffer %d.\n", i);
642 return AVERROR(ENOMEM);
643 }
644 }
645
646 s->cur_frame.tf.f = av_frame_alloc();
647 if (!s->cur_frame.tf.f) {
648 av_log(avctx, AV_LOG_ERROR,
649 "Failed to allocate current frame buffer.\n");
650 return AVERROR(ENOMEM);
651 }
652
653 ret = ff_cbs_init(&s->cbc, AV_CODEC_ID_AV1, avctx);
654 if (ret < 0)
655 return ret;
656
657 av_opt_set_int(s->cbc->priv_data, "operating_point", s->operating_point, 0);
658
659 if (avctx->extradata && avctx->extradata_size) {
660 ret = ff_cbs_read_extradata_from_codec(s->cbc,
661 &s->current_obu,
662 avctx);
663 if (ret < 0) {
664 av_log(avctx, AV_LOG_WARNING, "Failed to read extradata.\n");
665 return ret;
666 }
667
668 seq = ((CodedBitstreamAV1Context *)(s->cbc->priv_data))->sequence_header;
669 if (!seq) {
670 av_log(avctx, AV_LOG_WARNING, "No sequence header available.\n");
671 goto end;
672 }
673
674 ret = set_context_with_sequence(avctx, seq);
675 if (ret < 0) {
676 av_log(avctx, AV_LOG_WARNING, "Failed to set decoder context.\n");
677 goto end;
678 }
679
680 end:
681 ff_cbs_fragment_reset(&s->current_obu);
682 }
683
684 return ret;
685 }
686
av1_frame_alloc(AVCodecContext * avctx,AV1Frame * f)687 static int av1_frame_alloc(AVCodecContext *avctx, AV1Frame *f)
688 {
689 AV1DecContext *s = avctx->priv_data;
690 AV1RawFrameHeader *header= s->raw_frame_header;
691 AVFrame *frame;
692 int ret;
693
694 ret = update_context_with_frame_header(avctx, header);
695 if (ret < 0) {
696 av_log(avctx, AV_LOG_ERROR, "Failed to update context with frame header\n");
697 return ret;
698 }
699
700 f->header_ref = av_buffer_ref(s->header_ref);
701 if (!f->header_ref)
702 return AVERROR(ENOMEM);
703
704 f->raw_frame_header = s->raw_frame_header;
705
706 if ((ret = ff_thread_get_buffer(avctx, &f->tf, AV_GET_BUFFER_FLAG_REF)) < 0)
707 goto fail;
708
709 frame = f->tf.f;
710 frame->key_frame = header->frame_type == AV1_FRAME_KEY;
711
712 switch (header->frame_type) {
713 case AV1_FRAME_KEY:
714 case AV1_FRAME_INTRA_ONLY:
715 frame->pict_type = AV_PICTURE_TYPE_I;
716 break;
717 case AV1_FRAME_INTER:
718 frame->pict_type = AV_PICTURE_TYPE_P;
719 break;
720 case AV1_FRAME_SWITCH:
721 frame->pict_type = AV_PICTURE_TYPE_SP;
722 break;
723 }
724
725 if (avctx->hwaccel) {
726 const AVHWAccel *hwaccel = avctx->hwaccel;
727 if (hwaccel->frame_priv_data_size) {
728 f->hwaccel_priv_buf =
729 av_buffer_allocz(hwaccel->frame_priv_data_size);
730 if (!f->hwaccel_priv_buf) {
731 ret = AVERROR(ENOMEM);
732 goto fail;
733 }
734 f->hwaccel_picture_private = f->hwaccel_priv_buf->data;
735 }
736 }
737 return 0;
738
739 fail:
740 av1_frame_unref(avctx, f);
741 return ret;
742 }
743
export_film_grain(AVCodecContext * avctx,AVFrame * frame)744 static int export_film_grain(AVCodecContext *avctx, AVFrame *frame)
745 {
746 AV1DecContext *s = avctx->priv_data;
747 const AV1RawFilmGrainParams *film_grain = &s->cur_frame.film_grain;
748 AVFilmGrainParams *fgp;
749 AVFilmGrainAOMParams *aom;
750
751 if (!film_grain->apply_grain)
752 return 0;
753
754 fgp = av_film_grain_params_create_side_data(frame);
755 if (!fgp)
756 return AVERROR(ENOMEM);
757
758 fgp->type = AV_FILM_GRAIN_PARAMS_AV1;
759 fgp->seed = film_grain->grain_seed;
760
761 aom = &fgp->codec.aom;
762 aom->chroma_scaling_from_luma = film_grain->chroma_scaling_from_luma;
763 aom->scaling_shift = film_grain->grain_scaling_minus_8 + 8;
764 aom->ar_coeff_lag = film_grain->ar_coeff_lag;
765 aom->ar_coeff_shift = film_grain->ar_coeff_shift_minus_6 + 6;
766 aom->grain_scale_shift = film_grain->grain_scale_shift;
767 aom->overlap_flag = film_grain->overlap_flag;
768 aom->limit_output_range = film_grain->clip_to_restricted_range;
769
770 aom->num_y_points = film_grain->num_y_points;
771 for (int i = 0; i < film_grain->num_y_points; i++) {
772 aom->y_points[i][0] = film_grain->point_y_value[i];
773 aom->y_points[i][1] = film_grain->point_y_scaling[i];
774 }
775 aom->num_uv_points[0] = film_grain->num_cb_points;
776 for (int i = 0; i < film_grain->num_cb_points; i++) {
777 aom->uv_points[0][i][0] = film_grain->point_cb_value[i];
778 aom->uv_points[0][i][1] = film_grain->point_cb_scaling[i];
779 }
780 aom->num_uv_points[1] = film_grain->num_cr_points;
781 for (int i = 0; i < film_grain->num_cr_points; i++) {
782 aom->uv_points[1][i][0] = film_grain->point_cr_value[i];
783 aom->uv_points[1][i][1] = film_grain->point_cr_scaling[i];
784 }
785
786 for (int i = 0; i < 24; i++) {
787 aom->ar_coeffs_y[i] = film_grain->ar_coeffs_y_plus_128[i] - 128;
788 }
789 for (int i = 0; i < 25; i++) {
790 aom->ar_coeffs_uv[0][i] = film_grain->ar_coeffs_cb_plus_128[i] - 128;
791 aom->ar_coeffs_uv[1][i] = film_grain->ar_coeffs_cr_plus_128[i] - 128;
792 }
793
794 aom->uv_mult[0] = film_grain->cb_mult;
795 aom->uv_mult[1] = film_grain->cr_mult;
796 aom->uv_mult_luma[0] = film_grain->cb_luma_mult;
797 aom->uv_mult_luma[1] = film_grain->cr_luma_mult;
798 aom->uv_offset[0] = film_grain->cb_offset;
799 aom->uv_offset[1] = film_grain->cr_offset;
800
801 return 0;
802 }
803
set_output_frame(AVCodecContext * avctx,AVFrame * frame,const AVPacket * pkt,int * got_frame)804 static int set_output_frame(AVCodecContext *avctx, AVFrame *frame,
805 const AVPacket *pkt, int *got_frame)
806 {
807 AV1DecContext *s = avctx->priv_data;
808 const AVFrame *srcframe = s->cur_frame.tf.f;
809 int ret;
810
811 // TODO: all layers
812 if (s->operating_point_idc &&
813 av_log2(s->operating_point_idc >> 8) > s->cur_frame.spatial_id)
814 return 0;
815
816 ret = av_frame_ref(frame, srcframe);
817 if (ret < 0)
818 return ret;
819
820 if (avctx->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN) {
821 ret = export_film_grain(avctx, frame);
822 if (ret < 0) {
823 av_frame_unref(frame);
824 return ret;
825 }
826 }
827
828 frame->pts = pkt->pts;
829 frame->pkt_dts = pkt->dts;
830 frame->pkt_size = pkt->size;
831
832 *got_frame = 1;
833
834 return 0;
835 }
836
update_reference_list(AVCodecContext * avctx)837 static int update_reference_list(AVCodecContext *avctx)
838 {
839 AV1DecContext *s = avctx->priv_data;
840 const AV1RawFrameHeader *header = s->raw_frame_header;
841 int ret;
842
843 for (int i = 0; i < AV1_NUM_REF_FRAMES; i++) {
844 if (header->refresh_frame_flags & (1 << i)) {
845 if (s->ref[i].tf.f->buf[0])
846 av1_frame_unref(avctx, &s->ref[i]);
847 if ((ret = av1_frame_ref(avctx, &s->ref[i], &s->cur_frame)) < 0) {
848 av_log(avctx, AV_LOG_ERROR,
849 "Failed to update frame %d in reference list\n", i);
850 return ret;
851 }
852 }
853 }
854 return 0;
855 }
856
get_current_frame(AVCodecContext * avctx)857 static int get_current_frame(AVCodecContext *avctx)
858 {
859 AV1DecContext *s = avctx->priv_data;
860 int ret;
861
862 if (s->cur_frame.tf.f->buf[0])
863 av1_frame_unref(avctx, &s->cur_frame);
864
865 ret = av1_frame_alloc(avctx, &s->cur_frame);
866 if (ret < 0) {
867 av_log(avctx, AV_LOG_ERROR,
868 "Failed to allocate space for current frame.\n");
869 return ret;
870 }
871
872 ret = init_tile_data(s);
873 if (ret < 0) {
874 av_log(avctx, AV_LOG_ERROR, "Failed to init tile data.\n");
875 return ret;
876 }
877
878 global_motion_params(s);
879 skip_mode_params(s);
880 coded_lossless_param(s);
881 load_grain_params(s);
882
883 return ret;
884 }
885
av1_decode_frame(AVCodecContext * avctx,void * frame,int * got_frame,AVPacket * pkt)886 static int av1_decode_frame(AVCodecContext *avctx, void *frame,
887 int *got_frame, AVPacket *pkt)
888 {
889 AV1DecContext *s = avctx->priv_data;
890 AV1RawTileGroup *raw_tile_group = NULL;
891 int ret;
892
893 ret = ff_cbs_read_packet(s->cbc, &s->current_obu, pkt);
894 if (ret < 0) {
895 av_log(avctx, AV_LOG_ERROR, "Failed to read packet.\n");
896 goto end;
897 }
898 av_log(avctx, AV_LOG_DEBUG, "Total obu for this frame:%d.\n",
899 s->current_obu.nb_units);
900
901 for (int i = 0; i < s->current_obu.nb_units; i++) {
902 CodedBitstreamUnit *unit = &s->current_obu.units[i];
903 AV1RawOBU *obu = unit->content;
904 const AV1RawOBUHeader *header;
905
906 if (!obu)
907 continue;
908
909 header = &obu->header;
910 av_log(avctx, AV_LOG_DEBUG, "Obu idx:%d, obu type:%d.\n", i, unit->type);
911
912 switch (unit->type) {
913 case AV1_OBU_SEQUENCE_HEADER:
914 av_buffer_unref(&s->seq_ref);
915 s->seq_ref = av_buffer_ref(unit->content_ref);
916 if (!s->seq_ref) {
917 ret = AVERROR(ENOMEM);
918 goto end;
919 }
920
921 s->raw_seq = &obu->obu.sequence_header;
922
923 ret = set_context_with_sequence(avctx, s->raw_seq);
924 if (ret < 0) {
925 av_log(avctx, AV_LOG_ERROR, "Failed to set context.\n");
926 s->raw_seq = NULL;
927 goto end;
928 }
929
930 s->operating_point_idc = s->raw_seq->operating_point_idc[s->operating_point];
931
932 if (s->pix_fmt == AV_PIX_FMT_NONE) {
933 ret = get_pixel_format(avctx);
934 if (ret < 0) {
935 av_log(avctx, AV_LOG_ERROR,
936 "Failed to get pixel format.\n");
937 s->raw_seq = NULL;
938 goto end;
939 }
940 }
941
942 if (avctx->hwaccel && avctx->hwaccel->decode_params) {
943 ret = avctx->hwaccel->decode_params(avctx, unit->type, unit->data,
944 unit->data_size);
945 if (ret < 0) {
946 av_log(avctx, AV_LOG_ERROR, "HW accel decode params fail.\n");
947 s->raw_seq = NULL;
948 goto end;
949 }
950 }
951 break;
952 case AV1_OBU_REDUNDANT_FRAME_HEADER:
953 if (s->raw_frame_header)
954 break;
955 // fall-through
956 case AV1_OBU_FRAME:
957 case AV1_OBU_FRAME_HEADER:
958 if (!s->raw_seq) {
959 av_log(avctx, AV_LOG_ERROR, "Missing Sequence Header.\n");
960 ret = AVERROR_INVALIDDATA;
961 goto end;
962 }
963
964 av_buffer_unref(&s->header_ref);
965 s->header_ref = av_buffer_ref(unit->content_ref);
966 if (!s->header_ref) {
967 ret = AVERROR(ENOMEM);
968 goto end;
969 }
970
971 if (unit->type == AV1_OBU_FRAME)
972 s->raw_frame_header = &obu->obu.frame.header;
973 else
974 s->raw_frame_header = &obu->obu.frame_header;
975
976 if (s->raw_frame_header->show_existing_frame) {
977 if (s->cur_frame.tf.f->buf[0])
978 av1_frame_unref(avctx, &s->cur_frame);
979
980 ret = av1_frame_ref(avctx, &s->cur_frame,
981 &s->ref[s->raw_frame_header->frame_to_show_map_idx]);
982 if (ret < 0) {
983 av_log(avctx, AV_LOG_ERROR, "Failed to get reference frame.\n");
984 goto end;
985 }
986
987 ret = update_reference_list(avctx);
988 if (ret < 0) {
989 av_log(avctx, AV_LOG_ERROR, "Failed to update reference list.\n");
990 goto end;
991 }
992
993 ret = set_output_frame(avctx, frame, pkt, got_frame);
994 if (ret < 0)
995 av_log(avctx, AV_LOG_ERROR, "Set output frame error.\n");
996
997 s->raw_frame_header = NULL;
998
999 goto end;
1000 }
1001
1002 ret = get_current_frame(avctx);
1003 if (ret < 0) {
1004 av_log(avctx, AV_LOG_ERROR, "Get current frame error\n");
1005 goto end;
1006 }
1007
1008 s->cur_frame.spatial_id = header->spatial_id;
1009 s->cur_frame.temporal_id = header->temporal_id;
1010
1011 if (avctx->hwaccel) {
1012 ret = avctx->hwaccel->start_frame(avctx, unit->data,
1013 unit->data_size);
1014 if (ret < 0) {
1015 av_log(avctx, AV_LOG_ERROR, "HW accel start frame fail.\n");
1016 goto end;
1017 }
1018 }
1019 if (unit->type != AV1_OBU_FRAME)
1020 break;
1021 // fall-through
1022 case AV1_OBU_TILE_GROUP:
1023 if (!s->raw_frame_header) {
1024 av_log(avctx, AV_LOG_ERROR, "Missing Frame Header.\n");
1025 ret = AVERROR_INVALIDDATA;
1026 goto end;
1027 }
1028
1029 if (unit->type == AV1_OBU_FRAME)
1030 raw_tile_group = &obu->obu.frame.tile_group;
1031 else
1032 raw_tile_group = &obu->obu.tile_group;
1033
1034 ret = get_tiles_info(avctx, raw_tile_group);
1035 if (ret < 0)
1036 goto end;
1037
1038 if (avctx->hwaccel) {
1039 ret = avctx->hwaccel->decode_slice(avctx,
1040 raw_tile_group->tile_data.data,
1041 raw_tile_group->tile_data.data_size);
1042 if (ret < 0) {
1043 av_log(avctx, AV_LOG_ERROR,
1044 "HW accel decode slice fail.\n");
1045 goto end;
1046 }
1047 }
1048 break;
1049 case AV1_OBU_TILE_LIST:
1050 case AV1_OBU_TEMPORAL_DELIMITER:
1051 case AV1_OBU_PADDING:
1052 case AV1_OBU_METADATA:
1053 break;
1054 default:
1055 av_log(avctx, AV_LOG_DEBUG,
1056 "Unknown obu type: %d (%"SIZE_SPECIFIER" bits).\n",
1057 unit->type, unit->data_size);
1058 }
1059
1060 if (raw_tile_group && (s->tile_num == raw_tile_group->tg_end + 1)) {
1061 if (avctx->hwaccel) {
1062 ret = avctx->hwaccel->end_frame(avctx);
1063 if (ret < 0) {
1064 av_log(avctx, AV_LOG_ERROR, "HW accel end frame fail.\n");
1065 goto end;
1066 }
1067 }
1068
1069 ret = update_reference_list(avctx);
1070 if (ret < 0) {
1071 av_log(avctx, AV_LOG_ERROR, "Failed to update reference list.\n");
1072 goto end;
1073 }
1074
1075 if (s->raw_frame_header->show_frame) {
1076 ret = set_output_frame(avctx, frame, pkt, got_frame);
1077 if (ret < 0) {
1078 av_log(avctx, AV_LOG_ERROR, "Set output frame error\n");
1079 goto end;
1080 }
1081 }
1082 raw_tile_group = NULL;
1083 s->raw_frame_header = NULL;
1084 }
1085 }
1086
1087 end:
1088 ff_cbs_fragment_reset(&s->current_obu);
1089 if (ret < 0)
1090 s->raw_frame_header = NULL;
1091 return ret;
1092 }
1093
av1_decode_flush(AVCodecContext * avctx)1094 static void av1_decode_flush(AVCodecContext *avctx)
1095 {
1096 AV1DecContext *s = avctx->priv_data;
1097
1098 for (int i = 0; i < FF_ARRAY_ELEMS(s->ref); i++)
1099 av1_frame_unref(avctx, &s->ref[i]);
1100
1101 av1_frame_unref(avctx, &s->cur_frame);
1102 s->operating_point_idc = 0;
1103 s->raw_frame_header = NULL;
1104 s->raw_seq = NULL;
1105
1106 ff_cbs_flush(s->cbc);
1107 }
1108
1109 #define OFFSET(x) offsetof(AV1DecContext, x)
1110 #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
1111 static const AVOption av1_options[] = {
1112 { "operating_point", "Select an operating point of the scalable bitstream",
1113 OFFSET(operating_point), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, AV1_MAX_OPERATING_POINTS - 1, VD },
1114 { NULL }
1115 };
1116
1117 static const AVClass av1_class = {
1118 .class_name = "AV1 decoder",
1119 .item_name = av_default_item_name,
1120 .option = av1_options,
1121 .version = LIBAVUTIL_VERSION_INT,
1122 };
1123
1124 AVCodec ff_av1_decoder = {
1125 .name = "av1",
1126 .long_name = NULL_IF_CONFIG_SMALL("Alliance for Open Media AV1"),
1127 .type = AVMEDIA_TYPE_VIDEO,
1128 .id = AV_CODEC_ID_AV1,
1129 .priv_data_size = sizeof(AV1DecContext),
1130 .init = av1_decode_init,
1131 .close = av1_decode_free,
1132 .decode = av1_decode_frame,
1133 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_AVOID_PROBING,
1134 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE |
1135 FF_CODEC_CAP_INIT_CLEANUP |
1136 FF_CODEC_CAP_SETS_PKT_DTS,
1137 .flush = av1_decode_flush,
1138 .profiles = NULL_IF_CONFIG_SMALL(ff_av1_profiles),
1139 .priv_class = &av1_class,
1140 .hw_configs = (const AVCodecHWConfigInternal *const []) {
1141 #if CONFIG_AV1_DXVA2_HWACCEL
1142 HWACCEL_DXVA2(av1),
1143 #endif
1144 #if CONFIG_AV1_D3D11VA_HWACCEL
1145 HWACCEL_D3D11VA(av1),
1146 #endif
1147 #if CONFIG_AV1_D3D11VA2_HWACCEL
1148 HWACCEL_D3D11VA2(av1),
1149 #endif
1150 #if CONFIG_AV1_NVDEC_HWACCEL
1151 HWACCEL_NVDEC(av1),
1152 #endif
1153 #if CONFIG_AV1_VAAPI_HWACCEL
1154 HWACCEL_VAAPI(av1),
1155 #endif
1156 NULL
1157 },
1158 };
1159