1 /*
2 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
3 *
4 * This source code is subject to the terms of the BSD 2 Clause License and
5 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6 * was not distributed with this source code in the LICENSE file, you can
7 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8 * Media Patent License 1.0 was not distributed with this source code in the
9 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10 */
11
12 #include <assert.h>
13 #include <limits.h>
14 #include <stdio.h>
15
16 #include "aom/aom_encoder.h"
17 #include "aom_dsp/aom_dsp_common.h"
18 #include "aom_dsp/binary_codes_writer.h"
19 #include "aom_dsp/bitwriter_buffer.h"
20 #include "aom_mem/aom_mem.h"
21 #include "aom_ports/bitops.h"
22 #include "aom_ports/mem_ops.h"
23 #if CONFIG_BITSTREAM_DEBUG
24 #include "aom_util/debug_util.h"
25 #endif // CONFIG_BITSTREAM_DEBUG
26
27 #include "av1/common/cdef.h"
28 #include "av1/common/cfl.h"
29 #include "av1/common/entropy.h"
30 #include "av1/common/entropymode.h"
31 #include "av1/common/entropymv.h"
32 #include "av1/common/mvref_common.h"
33 #include "av1/common/pred_common.h"
34 #include "av1/common/reconinter.h"
35 #include "av1/common/reconintra.h"
36 #include "av1/common/seg_common.h"
37 #include "av1/common/tile_common.h"
38
39 #include "av1/encoder/bitstream.h"
40 #include "av1/encoder/cost.h"
41 #include "av1/encoder/encodemv.h"
42 #include "av1/encoder/encodetxb.h"
43 #include "av1/encoder/ethread.h"
44 #include "av1/encoder/mcomp.h"
45 #include "av1/encoder/palette.h"
46 #include "av1/encoder/pickrst.h"
47 #include "av1/encoder/segmentation.h"
48 #include "av1/encoder/tokenize.h"
49
50 #define ENC_MISMATCH_DEBUG 0
51 #define SETUP_TIME_OH_CONST 5 // Setup time overhead constant per worker
52 #define JOB_DISP_TIME_OH_CONST 1 // Job dispatch time overhead per tile
53
write_uniform(aom_writer * w,int n,int v)54 static INLINE void write_uniform(aom_writer *w, int n, int v) {
55 const int l = get_unsigned_bits(n);
56 const int m = (1 << l) - n;
57 if (l == 0) return;
58 if (v < m) {
59 aom_write_literal(w, v, l - 1);
60 } else {
61 aom_write_literal(w, m + ((v - m) >> 1), l - 1);
62 aom_write_literal(w, (v - m) & 1, 1);
63 }
64 }
65
66 #if !CONFIG_REALTIME_ONLY
67 static AOM_INLINE void loop_restoration_write_sb_coeffs(
68 const AV1_COMMON *const cm, MACROBLOCKD *xd, int runit_idx,
69 aom_writer *const w, int plane, FRAME_COUNTS *counts);
70 #endif
71
write_intra_y_mode_kf(FRAME_CONTEXT * frame_ctx,const MB_MODE_INFO * mi,const MB_MODE_INFO * above_mi,const MB_MODE_INFO * left_mi,PREDICTION_MODE mode,aom_writer * w)72 static AOM_INLINE void write_intra_y_mode_kf(FRAME_CONTEXT *frame_ctx,
73 const MB_MODE_INFO *mi,
74 const MB_MODE_INFO *above_mi,
75 const MB_MODE_INFO *left_mi,
76 PREDICTION_MODE mode,
77 aom_writer *w) {
78 assert(!is_intrabc_block(mi));
79 (void)mi;
80 aom_write_symbol(w, mode, get_y_mode_cdf(frame_ctx, above_mi, left_mi),
81 INTRA_MODES);
82 }
83
write_inter_mode(aom_writer * w,PREDICTION_MODE mode,FRAME_CONTEXT * ec_ctx,const int16_t mode_ctx)84 static AOM_INLINE void write_inter_mode(aom_writer *w, PREDICTION_MODE mode,
85 FRAME_CONTEXT *ec_ctx,
86 const int16_t mode_ctx) {
87 const int16_t newmv_ctx = mode_ctx & NEWMV_CTX_MASK;
88
89 aom_write_symbol(w, mode != NEWMV, ec_ctx->newmv_cdf[newmv_ctx], 2);
90
91 if (mode != NEWMV) {
92 const int16_t zeromv_ctx =
93 (mode_ctx >> GLOBALMV_OFFSET) & GLOBALMV_CTX_MASK;
94 aom_write_symbol(w, mode != GLOBALMV, ec_ctx->zeromv_cdf[zeromv_ctx], 2);
95
96 if (mode != GLOBALMV) {
97 int16_t refmv_ctx = (mode_ctx >> REFMV_OFFSET) & REFMV_CTX_MASK;
98 aom_write_symbol(w, mode != NEARESTMV, ec_ctx->refmv_cdf[refmv_ctx], 2);
99 }
100 }
101 }
102
write_drl_idx(FRAME_CONTEXT * ec_ctx,const MB_MODE_INFO * mbmi,const MB_MODE_INFO_EXT_FRAME * mbmi_ext_frame,aom_writer * w)103 static AOM_INLINE void write_drl_idx(
104 FRAME_CONTEXT *ec_ctx, const MB_MODE_INFO *mbmi,
105 const MB_MODE_INFO_EXT_FRAME *mbmi_ext_frame, aom_writer *w) {
106 assert(mbmi->ref_mv_idx < 3);
107
108 const int new_mv = mbmi->mode == NEWMV || mbmi->mode == NEW_NEWMV;
109 if (new_mv) {
110 int idx;
111 for (idx = 0; idx < 2; ++idx) {
112 if (mbmi_ext_frame->ref_mv_count > idx + 1) {
113 uint8_t drl_ctx = av1_drl_ctx(mbmi_ext_frame->weight, idx);
114
115 aom_write_symbol(w, mbmi->ref_mv_idx != idx, ec_ctx->drl_cdf[drl_ctx],
116 2);
117 if (mbmi->ref_mv_idx == idx) return;
118 }
119 }
120 return;
121 }
122
123 if (have_nearmv_in_inter_mode(mbmi->mode)) {
124 int idx;
125 // TODO(jingning): Temporary solution to compensate the NEARESTMV offset.
126 for (idx = 1; idx < 3; ++idx) {
127 if (mbmi_ext_frame->ref_mv_count > idx + 1) {
128 uint8_t drl_ctx = av1_drl_ctx(mbmi_ext_frame->weight, idx);
129 aom_write_symbol(w, mbmi->ref_mv_idx != (idx - 1),
130 ec_ctx->drl_cdf[drl_ctx], 2);
131 if (mbmi->ref_mv_idx == (idx - 1)) return;
132 }
133 }
134 return;
135 }
136 }
137
write_inter_compound_mode(MACROBLOCKD * xd,aom_writer * w,PREDICTION_MODE mode,const int16_t mode_ctx)138 static AOM_INLINE void write_inter_compound_mode(MACROBLOCKD *xd, aom_writer *w,
139 PREDICTION_MODE mode,
140 const int16_t mode_ctx) {
141 assert(is_inter_compound_mode(mode));
142 aom_write_symbol(w, INTER_COMPOUND_OFFSET(mode),
143 xd->tile_ctx->inter_compound_mode_cdf[mode_ctx],
144 INTER_COMPOUND_MODES);
145 }
146
write_tx_size_vartx(MACROBLOCKD * xd,const MB_MODE_INFO * mbmi,TX_SIZE tx_size,int depth,int blk_row,int blk_col,aom_writer * w)147 static AOM_INLINE void write_tx_size_vartx(MACROBLOCKD *xd,
148 const MB_MODE_INFO *mbmi,
149 TX_SIZE tx_size, int depth,
150 int blk_row, int blk_col,
151 aom_writer *w) {
152 FRAME_CONTEXT *const ec_ctx = xd->tile_ctx;
153 const int max_blocks_high = max_block_high(xd, mbmi->bsize, 0);
154 const int max_blocks_wide = max_block_wide(xd, mbmi->bsize, 0);
155
156 if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
157
158 if (depth == MAX_VARTX_DEPTH) {
159 txfm_partition_update(xd->above_txfm_context + blk_col,
160 xd->left_txfm_context + blk_row, tx_size, tx_size);
161 return;
162 }
163
164 const int ctx = txfm_partition_context(xd->above_txfm_context + blk_col,
165 xd->left_txfm_context + blk_row,
166 mbmi->bsize, tx_size);
167 const int txb_size_index =
168 av1_get_txb_size_index(mbmi->bsize, blk_row, blk_col);
169 const int write_txfm_partition =
170 tx_size == mbmi->inter_tx_size[txb_size_index];
171 if (write_txfm_partition) {
172 aom_write_symbol(w, 0, ec_ctx->txfm_partition_cdf[ctx], 2);
173
174 txfm_partition_update(xd->above_txfm_context + blk_col,
175 xd->left_txfm_context + blk_row, tx_size, tx_size);
176 // TODO(yuec): set correct txfm partition update for qttx
177 } else {
178 const TX_SIZE sub_txs = sub_tx_size_map[tx_size];
179 const int bsw = tx_size_wide_unit[sub_txs];
180 const int bsh = tx_size_high_unit[sub_txs];
181
182 aom_write_symbol(w, 1, ec_ctx->txfm_partition_cdf[ctx], 2);
183
184 if (sub_txs == TX_4X4) {
185 txfm_partition_update(xd->above_txfm_context + blk_col,
186 xd->left_txfm_context + blk_row, sub_txs, tx_size);
187 return;
188 }
189
190 assert(bsw > 0 && bsh > 0);
191 for (int row = 0; row < tx_size_high_unit[tx_size]; row += bsh) {
192 const int offsetr = blk_row + row;
193 for (int col = 0; col < tx_size_wide_unit[tx_size]; col += bsw) {
194 const int offsetc = blk_col + col;
195 write_tx_size_vartx(xd, mbmi, sub_txs, depth + 1, offsetr, offsetc, w);
196 }
197 }
198 }
199 }
200
write_selected_tx_size(const MACROBLOCKD * xd,aom_writer * w)201 static AOM_INLINE void write_selected_tx_size(const MACROBLOCKD *xd,
202 aom_writer *w) {
203 const MB_MODE_INFO *const mbmi = xd->mi[0];
204 const BLOCK_SIZE bsize = mbmi->bsize;
205 FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
206 if (block_signals_txsize(bsize)) {
207 const TX_SIZE tx_size = mbmi->tx_size;
208 const int tx_size_ctx = get_tx_size_context(xd);
209 const int depth = tx_size_to_depth(tx_size, bsize);
210 const int max_depths = bsize_to_max_depth(bsize);
211 const int32_t tx_size_cat = bsize_to_tx_size_cat(bsize);
212
213 assert(depth >= 0 && depth <= max_depths);
214 assert(!is_inter_block(mbmi));
215 assert(IMPLIES(is_rect_tx(tx_size), is_rect_tx_allowed(xd, mbmi)));
216
217 aom_write_symbol(w, depth, ec_ctx->tx_size_cdf[tx_size_cat][tx_size_ctx],
218 max_depths + 1);
219 }
220 }
221
write_skip(const AV1_COMMON * cm,const MACROBLOCKD * xd,uint8_t segment_id,const MB_MODE_INFO * mi,aom_writer * w)222 static int write_skip(const AV1_COMMON *cm, const MACROBLOCKD *xd,
223 uint8_t segment_id, const MB_MODE_INFO *mi,
224 aom_writer *w) {
225 if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP)) {
226 return 1;
227 } else {
228 const int skip_txfm = mi->skip_txfm;
229 const int ctx = av1_get_skip_txfm_context(xd);
230 FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
231 aom_write_symbol(w, skip_txfm, ec_ctx->skip_txfm_cdfs[ctx], 2);
232 return skip_txfm;
233 }
234 }
235
write_skip_mode(const AV1_COMMON * cm,const MACROBLOCKD * xd,uint8_t segment_id,const MB_MODE_INFO * mi,aom_writer * w)236 static int write_skip_mode(const AV1_COMMON *cm, const MACROBLOCKD *xd,
237 uint8_t segment_id, const MB_MODE_INFO *mi,
238 aom_writer *w) {
239 if (!cm->current_frame.skip_mode_info.skip_mode_flag) return 0;
240 if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP)) {
241 return 0;
242 }
243 const int skip_mode = mi->skip_mode;
244 if (!is_comp_ref_allowed(mi->bsize)) {
245 assert(!skip_mode);
246 return 0;
247 }
248 if (segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME) ||
249 segfeature_active(&cm->seg, segment_id, SEG_LVL_GLOBALMV)) {
250 // These features imply single-reference mode, while skip mode implies
251 // compound reference. Hence, the two are mutually exclusive.
252 // In other words, skip_mode is implicitly 0 here.
253 assert(!skip_mode);
254 return 0;
255 }
256 const int ctx = av1_get_skip_mode_context(xd);
257 aom_write_symbol(w, skip_mode, xd->tile_ctx->skip_mode_cdfs[ctx], 2);
258 return skip_mode;
259 }
260
write_is_inter(const AV1_COMMON * cm,const MACROBLOCKD * xd,uint8_t segment_id,aom_writer * w,const int is_inter)261 static AOM_INLINE void write_is_inter(const AV1_COMMON *cm,
262 const MACROBLOCKD *xd, uint8_t segment_id,
263 aom_writer *w, const int is_inter) {
264 if (!segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME)) {
265 if (segfeature_active(&cm->seg, segment_id, SEG_LVL_GLOBALMV)) {
266 assert(is_inter);
267 return;
268 }
269 const int ctx = av1_get_intra_inter_context(xd);
270 FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
271 aom_write_symbol(w, is_inter, ec_ctx->intra_inter_cdf[ctx], 2);
272 }
273 }
274
write_motion_mode(const AV1_COMMON * cm,MACROBLOCKD * xd,const MB_MODE_INFO * mbmi,aom_writer * w)275 static AOM_INLINE void write_motion_mode(const AV1_COMMON *cm, MACROBLOCKD *xd,
276 const MB_MODE_INFO *mbmi,
277 aom_writer *w) {
278 MOTION_MODE last_motion_mode_allowed =
279 cm->features.switchable_motion_mode
280 ? motion_mode_allowed(cm->global_motion, xd, mbmi,
281 cm->features.allow_warped_motion)
282 : SIMPLE_TRANSLATION;
283 assert(mbmi->motion_mode <= last_motion_mode_allowed);
284 switch (last_motion_mode_allowed) {
285 case SIMPLE_TRANSLATION: break;
286 case OBMC_CAUSAL:
287 aom_write_symbol(w, mbmi->motion_mode == OBMC_CAUSAL,
288 xd->tile_ctx->obmc_cdf[mbmi->bsize], 2);
289 break;
290 default:
291 aom_write_symbol(w, mbmi->motion_mode,
292 xd->tile_ctx->motion_mode_cdf[mbmi->bsize],
293 MOTION_MODES);
294 }
295 }
296
write_delta_qindex(const MACROBLOCKD * xd,int delta_qindex,aom_writer * w)297 static AOM_INLINE void write_delta_qindex(const MACROBLOCKD *xd,
298 int delta_qindex, aom_writer *w) {
299 int sign = delta_qindex < 0;
300 int abs = sign ? -delta_qindex : delta_qindex;
301 int rem_bits, thr;
302 int smallval = abs < DELTA_Q_SMALL ? 1 : 0;
303 FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
304
305 aom_write_symbol(w, AOMMIN(abs, DELTA_Q_SMALL), ec_ctx->delta_q_cdf,
306 DELTA_Q_PROBS + 1);
307
308 if (!smallval) {
309 rem_bits = get_msb(abs - 1);
310 thr = (1 << rem_bits) + 1;
311 aom_write_literal(w, rem_bits - 1, 3);
312 aom_write_literal(w, abs - thr, rem_bits);
313 }
314 if (abs > 0) {
315 aom_write_bit(w, sign);
316 }
317 }
318
write_delta_lflevel(const AV1_COMMON * cm,const MACROBLOCKD * xd,int lf_id,int delta_lflevel,int delta_lf_multi,aom_writer * w)319 static AOM_INLINE void write_delta_lflevel(const AV1_COMMON *cm,
320 const MACROBLOCKD *xd, int lf_id,
321 int delta_lflevel,
322 int delta_lf_multi, aom_writer *w) {
323 int sign = delta_lflevel < 0;
324 int abs = sign ? -delta_lflevel : delta_lflevel;
325 int rem_bits, thr;
326 int smallval = abs < DELTA_LF_SMALL ? 1 : 0;
327 FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
328 (void)cm;
329
330 if (delta_lf_multi) {
331 assert(lf_id >= 0 && lf_id < (av1_num_planes(cm) > 1 ? FRAME_LF_COUNT
332 : FRAME_LF_COUNT - 2));
333 aom_write_symbol(w, AOMMIN(abs, DELTA_LF_SMALL),
334 ec_ctx->delta_lf_multi_cdf[lf_id], DELTA_LF_PROBS + 1);
335 } else {
336 aom_write_symbol(w, AOMMIN(abs, DELTA_LF_SMALL), ec_ctx->delta_lf_cdf,
337 DELTA_LF_PROBS + 1);
338 }
339
340 if (!smallval) {
341 rem_bits = get_msb(abs - 1);
342 thr = (1 << rem_bits) + 1;
343 aom_write_literal(w, rem_bits - 1, 3);
344 aom_write_literal(w, abs - thr, rem_bits);
345 }
346 if (abs > 0) {
347 aom_write_bit(w, sign);
348 }
349 }
350
pack_map_tokens(aom_writer * w,const TokenExtra ** tp,int n,int num,MapCdf map_pb_cdf)351 static AOM_INLINE void pack_map_tokens(aom_writer *w, const TokenExtra **tp,
352 int n, int num, MapCdf map_pb_cdf) {
353 const TokenExtra *p = *tp;
354 const int palette_size_idx = n - PALETTE_MIN_SIZE;
355 write_uniform(w, n, p->token); // The first color index.
356 ++p;
357 --num;
358 for (int i = 0; i < num; ++i) {
359 assert((p->color_ctx >= 0) &&
360 (p->color_ctx < PALETTE_COLOR_INDEX_CONTEXTS));
361 aom_cdf_prob *color_map_cdf = map_pb_cdf[palette_size_idx][p->color_ctx];
362 aom_write_symbol(w, p->token, color_map_cdf, n);
363 ++p;
364 }
365 *tp = p;
366 }
367
pack_txb_tokens(aom_writer * w,AV1_COMMON * cm,MACROBLOCK * const x,const TokenExtra ** tp,const TokenExtra * const tok_end,MACROBLOCKD * xd,MB_MODE_INFO * mbmi,int plane,BLOCK_SIZE plane_bsize,aom_bit_depth_t bit_depth,int block,int blk_row,int blk_col,TX_SIZE tx_size,TOKEN_STATS * token_stats)368 static AOM_INLINE void pack_txb_tokens(
369 aom_writer *w, AV1_COMMON *cm, MACROBLOCK *const x, const TokenExtra **tp,
370 const TokenExtra *const tok_end, MACROBLOCKD *xd, MB_MODE_INFO *mbmi,
371 int plane, BLOCK_SIZE plane_bsize, aom_bit_depth_t bit_depth, int block,
372 int blk_row, int blk_col, TX_SIZE tx_size, TOKEN_STATS *token_stats) {
373 const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
374 const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
375
376 if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
377
378 const struct macroblockd_plane *const pd = &xd->plane[plane];
379 const TX_SIZE plane_tx_size =
380 plane ? av1_get_max_uv_txsize(mbmi->bsize, pd->subsampling_x,
381 pd->subsampling_y)
382 : mbmi->inter_tx_size[av1_get_txb_size_index(plane_bsize, blk_row,
383 blk_col)];
384
385 if (tx_size == plane_tx_size || plane) {
386 av1_write_coeffs_txb(cm, x, w, blk_row, blk_col, plane, block, tx_size);
387 #if CONFIG_RD_DEBUG
388 TOKEN_STATS tmp_token_stats;
389 init_token_stats(&tmp_token_stats);
390 token_stats->cost += tmp_token_stats.cost;
391 #endif
392 } else {
393 const TX_SIZE sub_txs = sub_tx_size_map[tx_size];
394 const int bsw = tx_size_wide_unit[sub_txs];
395 const int bsh = tx_size_high_unit[sub_txs];
396 const int step = bsh * bsw;
397 const int row_end =
398 AOMMIN(tx_size_high_unit[tx_size], max_blocks_high - blk_row);
399 const int col_end =
400 AOMMIN(tx_size_wide_unit[tx_size], max_blocks_wide - blk_col);
401
402 assert(bsw > 0 && bsh > 0);
403
404 for (int r = 0; r < row_end; r += bsh) {
405 const int offsetr = blk_row + r;
406 for (int c = 0; c < col_end; c += bsw) {
407 const int offsetc = blk_col + c;
408 pack_txb_tokens(w, cm, x, tp, tok_end, xd, mbmi, plane, plane_bsize,
409 bit_depth, block, offsetr, offsetc, sub_txs,
410 token_stats);
411 block += step;
412 }
413 }
414 }
415 }
416
set_spatial_segment_id(const CommonModeInfoParams * const mi_params,uint8_t * segment_ids,BLOCK_SIZE bsize,int mi_row,int mi_col,uint8_t segment_id)417 static INLINE void set_spatial_segment_id(
418 const CommonModeInfoParams *const mi_params, uint8_t *segment_ids,
419 BLOCK_SIZE bsize, int mi_row, int mi_col, uint8_t segment_id) {
420 const int mi_offset = mi_row * mi_params->mi_cols + mi_col;
421 const int bw = mi_size_wide[bsize];
422 const int bh = mi_size_high[bsize];
423 const int xmis = AOMMIN(mi_params->mi_cols - mi_col, bw);
424 const int ymis = AOMMIN(mi_params->mi_rows - mi_row, bh);
425
426 const int mi_stride = mi_params->mi_cols;
427
428 set_segment_id(segment_ids, mi_offset, xmis, ymis, mi_stride, segment_id);
429 }
430
av1_neg_interleave(int x,int ref,int max)431 int av1_neg_interleave(int x, int ref, int max) {
432 assert(x < max);
433 const int diff = x - ref;
434 if (!ref) return x;
435 if (ref >= (max - 1)) return -x + max - 1;
436 if (2 * ref < max) {
437 if (abs(diff) <= ref) {
438 if (diff > 0)
439 return (diff << 1) - 1;
440 else
441 return ((-diff) << 1);
442 }
443 return x;
444 } else {
445 if (abs(diff) < (max - ref)) {
446 if (diff > 0)
447 return (diff << 1) - 1;
448 else
449 return ((-diff) << 1);
450 }
451 return (max - x) - 1;
452 }
453 }
454
write_segment_id(AV1_COMP * cpi,MACROBLOCKD * const xd,const MB_MODE_INFO * const mbmi,aom_writer * w,const struct segmentation * seg,struct segmentation_probs * segp,int skip_txfm)455 static AOM_INLINE void write_segment_id(AV1_COMP *cpi, MACROBLOCKD *const xd,
456 const MB_MODE_INFO *const mbmi,
457 aom_writer *w,
458 const struct segmentation *seg,
459 struct segmentation_probs *segp,
460 int skip_txfm) {
461 if (!seg->enabled || !seg->update_map) return;
462
463 AV1_COMMON *const cm = &cpi->common;
464 int cdf_num;
465 const uint8_t pred = av1_get_spatial_seg_pred(
466 cm, xd, &cdf_num, cpi->cyclic_refresh->skip_over4x4);
467 const int mi_row = xd->mi_row;
468 const int mi_col = xd->mi_col;
469
470 if (skip_txfm) {
471 // Still need to transmit tx size for intra blocks even if skip_txfm is
472 // true. Changing segment_id may make the tx size become invalid, e.g
473 // changing from lossless to lossy.
474 assert(is_inter_block(mbmi) || !cpi->enc_seg.has_lossless_segment);
475
476 set_spatial_segment_id(&cm->mi_params, cm->cur_frame->seg_map, mbmi->bsize,
477 mi_row, mi_col, pred);
478 set_spatial_segment_id(&cm->mi_params, cpi->enc_seg.map, mbmi->bsize,
479 mi_row, mi_col, pred);
480 /* mbmi is read only but we need to update segment_id */
481 ((MB_MODE_INFO *)mbmi)->segment_id = pred;
482 return;
483 }
484
485 const int coded_id =
486 av1_neg_interleave(mbmi->segment_id, pred, seg->last_active_segid + 1);
487 aom_cdf_prob *pred_cdf = segp->spatial_pred_seg_cdf[cdf_num];
488 aom_write_symbol(w, coded_id, pred_cdf, MAX_SEGMENTS);
489 set_spatial_segment_id(&cm->mi_params, cm->cur_frame->seg_map, mbmi->bsize,
490 mi_row, mi_col, mbmi->segment_id);
491 }
492
493 #define WRITE_REF_BIT(bname, pname) \
494 aom_write_symbol(w, bname, av1_get_pred_cdf_##pname(xd), 2)
495
496 // This function encodes the reference frame
write_ref_frames(const AV1_COMMON * cm,const MACROBLOCKD * xd,aom_writer * w)497 static AOM_INLINE void write_ref_frames(const AV1_COMMON *cm,
498 const MACROBLOCKD *xd, aom_writer *w) {
499 const MB_MODE_INFO *const mbmi = xd->mi[0];
500 const int is_compound = has_second_ref(mbmi);
501 const uint8_t segment_id = mbmi->segment_id;
502
503 // If segment level coding of this signal is disabled...
504 // or the segment allows multiple reference frame options
505 if (segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME)) {
506 assert(!is_compound);
507 assert(mbmi->ref_frame[0] ==
508 get_segdata(&cm->seg, segment_id, SEG_LVL_REF_FRAME));
509 } else if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP) ||
510 segfeature_active(&cm->seg, segment_id, SEG_LVL_GLOBALMV)) {
511 assert(!is_compound);
512 assert(mbmi->ref_frame[0] == LAST_FRAME);
513 } else {
514 // does the feature use compound prediction or not
515 // (if not specified at the frame/segment level)
516 if (cm->current_frame.reference_mode == REFERENCE_MODE_SELECT) {
517 if (is_comp_ref_allowed(mbmi->bsize))
518 aom_write_symbol(w, is_compound, av1_get_reference_mode_cdf(xd), 2);
519 } else {
520 assert((!is_compound) ==
521 (cm->current_frame.reference_mode == SINGLE_REFERENCE));
522 }
523
524 if (is_compound) {
525 const COMP_REFERENCE_TYPE comp_ref_type = has_uni_comp_refs(mbmi)
526 ? UNIDIR_COMP_REFERENCE
527 : BIDIR_COMP_REFERENCE;
528 aom_write_symbol(w, comp_ref_type, av1_get_comp_reference_type_cdf(xd),
529 2);
530
531 if (comp_ref_type == UNIDIR_COMP_REFERENCE) {
532 const int bit = mbmi->ref_frame[0] == BWDREF_FRAME;
533 WRITE_REF_BIT(bit, uni_comp_ref_p);
534
535 if (!bit) {
536 assert(mbmi->ref_frame[0] == LAST_FRAME);
537 const int bit1 = mbmi->ref_frame[1] == LAST3_FRAME ||
538 mbmi->ref_frame[1] == GOLDEN_FRAME;
539 WRITE_REF_BIT(bit1, uni_comp_ref_p1);
540 if (bit1) {
541 const int bit2 = mbmi->ref_frame[1] == GOLDEN_FRAME;
542 WRITE_REF_BIT(bit2, uni_comp_ref_p2);
543 }
544 } else {
545 assert(mbmi->ref_frame[1] == ALTREF_FRAME);
546 }
547
548 return;
549 }
550
551 assert(comp_ref_type == BIDIR_COMP_REFERENCE);
552
553 const int bit = (mbmi->ref_frame[0] == GOLDEN_FRAME ||
554 mbmi->ref_frame[0] == LAST3_FRAME);
555 WRITE_REF_BIT(bit, comp_ref_p);
556
557 if (!bit) {
558 const int bit1 = mbmi->ref_frame[0] == LAST2_FRAME;
559 WRITE_REF_BIT(bit1, comp_ref_p1);
560 } else {
561 const int bit2 = mbmi->ref_frame[0] == GOLDEN_FRAME;
562 WRITE_REF_BIT(bit2, comp_ref_p2);
563 }
564
565 const int bit_bwd = mbmi->ref_frame[1] == ALTREF_FRAME;
566 WRITE_REF_BIT(bit_bwd, comp_bwdref_p);
567
568 if (!bit_bwd) {
569 WRITE_REF_BIT(mbmi->ref_frame[1] == ALTREF2_FRAME, comp_bwdref_p1);
570 }
571
572 } else {
573 const int bit0 = (mbmi->ref_frame[0] <= ALTREF_FRAME &&
574 mbmi->ref_frame[0] >= BWDREF_FRAME);
575 WRITE_REF_BIT(bit0, single_ref_p1);
576
577 if (bit0) {
578 const int bit1 = mbmi->ref_frame[0] == ALTREF_FRAME;
579 WRITE_REF_BIT(bit1, single_ref_p2);
580
581 if (!bit1) {
582 WRITE_REF_BIT(mbmi->ref_frame[0] == ALTREF2_FRAME, single_ref_p6);
583 }
584 } else {
585 const int bit2 = (mbmi->ref_frame[0] == LAST3_FRAME ||
586 mbmi->ref_frame[0] == GOLDEN_FRAME);
587 WRITE_REF_BIT(bit2, single_ref_p3);
588
589 if (!bit2) {
590 const int bit3 = mbmi->ref_frame[0] != LAST_FRAME;
591 WRITE_REF_BIT(bit3, single_ref_p4);
592 } else {
593 const int bit4 = mbmi->ref_frame[0] != LAST3_FRAME;
594 WRITE_REF_BIT(bit4, single_ref_p5);
595 }
596 }
597 }
598 }
599 }
600
write_filter_intra_mode_info(const AV1_COMMON * cm,const MACROBLOCKD * xd,const MB_MODE_INFO * const mbmi,aom_writer * w)601 static AOM_INLINE void write_filter_intra_mode_info(
602 const AV1_COMMON *cm, const MACROBLOCKD *xd, const MB_MODE_INFO *const mbmi,
603 aom_writer *w) {
604 if (av1_filter_intra_allowed(cm, mbmi)) {
605 aom_write_symbol(w, mbmi->filter_intra_mode_info.use_filter_intra,
606 xd->tile_ctx->filter_intra_cdfs[mbmi->bsize], 2);
607 if (mbmi->filter_intra_mode_info.use_filter_intra) {
608 const FILTER_INTRA_MODE mode =
609 mbmi->filter_intra_mode_info.filter_intra_mode;
610 aom_write_symbol(w, mode, xd->tile_ctx->filter_intra_mode_cdf,
611 FILTER_INTRA_MODES);
612 }
613 }
614 }
615
write_angle_delta(aom_writer * w,int angle_delta,aom_cdf_prob * cdf)616 static AOM_INLINE void write_angle_delta(aom_writer *w, int angle_delta,
617 aom_cdf_prob *cdf) {
618 aom_write_symbol(w, angle_delta + MAX_ANGLE_DELTA, cdf,
619 2 * MAX_ANGLE_DELTA + 1);
620 }
621
write_mb_interp_filter(AV1_COMMON * const cm,ThreadData * td,aom_writer * w)622 static AOM_INLINE void write_mb_interp_filter(AV1_COMMON *const cm,
623 ThreadData *td, aom_writer *w) {
624 const MACROBLOCKD *xd = &td->mb.e_mbd;
625 const MB_MODE_INFO *const mbmi = xd->mi[0];
626 FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
627
628 if (!av1_is_interp_needed(xd)) {
629 int_interpfilters filters = av1_broadcast_interp_filter(
630 av1_unswitchable_filter(cm->features.interp_filter));
631 assert(mbmi->interp_filters.as_int == filters.as_int);
632 (void)filters;
633 return;
634 }
635 if (cm->features.interp_filter == SWITCHABLE) {
636 int dir;
637 for (dir = 0; dir < 2; ++dir) {
638 const int ctx = av1_get_pred_context_switchable_interp(xd, dir);
639 InterpFilter filter =
640 av1_extract_interp_filter(mbmi->interp_filters, dir);
641 aom_write_symbol(w, filter, ec_ctx->switchable_interp_cdf[ctx],
642 SWITCHABLE_FILTERS);
643 ++td->interp_filter_selected[filter];
644 if (cm->seq_params->enable_dual_filter == 0) return;
645 }
646 }
647 }
648
649 // Transmit color values with delta encoding. Write the first value as
650 // literal, and the deltas between each value and the previous one. "min_val" is
651 // the smallest possible value of the deltas.
delta_encode_palette_colors(const int * colors,int num,int bit_depth,int min_val,aom_writer * w)652 static AOM_INLINE void delta_encode_palette_colors(const int *colors, int num,
653 int bit_depth, int min_val,
654 aom_writer *w) {
655 if (num <= 0) return;
656 assert(colors[0] < (1 << bit_depth));
657 aom_write_literal(w, colors[0], bit_depth);
658 if (num == 1) return;
659 int max_delta = 0;
660 int deltas[PALETTE_MAX_SIZE];
661 memset(deltas, 0, sizeof(deltas));
662 for (int i = 1; i < num; ++i) {
663 assert(colors[i] < (1 << bit_depth));
664 const int delta = colors[i] - colors[i - 1];
665 deltas[i - 1] = delta;
666 assert(delta >= min_val);
667 if (delta > max_delta) max_delta = delta;
668 }
669 const int min_bits = bit_depth - 3;
670 int bits = AOMMAX(av1_ceil_log2(max_delta + 1 - min_val), min_bits);
671 assert(bits <= bit_depth);
672 int range = (1 << bit_depth) - colors[0] - min_val;
673 aom_write_literal(w, bits - min_bits, 2);
674 for (int i = 0; i < num - 1; ++i) {
675 aom_write_literal(w, deltas[i] - min_val, bits);
676 range -= deltas[i];
677 bits = AOMMIN(bits, av1_ceil_log2(range));
678 }
679 }
680
681 // Transmit luma palette color values. First signal if each color in the color
682 // cache is used. Those colors that are not in the cache are transmitted with
683 // delta encoding.
write_palette_colors_y(const MACROBLOCKD * const xd,const PALETTE_MODE_INFO * const pmi,int bit_depth,aom_writer * w)684 static AOM_INLINE void write_palette_colors_y(
685 const MACROBLOCKD *const xd, const PALETTE_MODE_INFO *const pmi,
686 int bit_depth, aom_writer *w) {
687 const int n = pmi->palette_size[0];
688 uint16_t color_cache[2 * PALETTE_MAX_SIZE];
689 const int n_cache = av1_get_palette_cache(xd, 0, color_cache);
690 int out_cache_colors[PALETTE_MAX_SIZE];
691 uint8_t cache_color_found[2 * PALETTE_MAX_SIZE];
692 const int n_out_cache =
693 av1_index_color_cache(color_cache, n_cache, pmi->palette_colors, n,
694 cache_color_found, out_cache_colors);
695 int n_in_cache = 0;
696 for (int i = 0; i < n_cache && n_in_cache < n; ++i) {
697 const int found = cache_color_found[i];
698 aom_write_bit(w, found);
699 n_in_cache += found;
700 }
701 assert(n_in_cache + n_out_cache == n);
702 delta_encode_palette_colors(out_cache_colors, n_out_cache, bit_depth, 1, w);
703 }
704
705 // Write chroma palette color values. U channel is handled similarly to the luma
706 // channel. For v channel, either use delta encoding or transmit raw values
707 // directly, whichever costs less.
write_palette_colors_uv(const MACROBLOCKD * const xd,const PALETTE_MODE_INFO * const pmi,int bit_depth,aom_writer * w)708 static AOM_INLINE void write_palette_colors_uv(
709 const MACROBLOCKD *const xd, const PALETTE_MODE_INFO *const pmi,
710 int bit_depth, aom_writer *w) {
711 const int n = pmi->palette_size[1];
712 const uint16_t *colors_u = pmi->palette_colors + PALETTE_MAX_SIZE;
713 const uint16_t *colors_v = pmi->palette_colors + 2 * PALETTE_MAX_SIZE;
714 // U channel colors.
715 uint16_t color_cache[2 * PALETTE_MAX_SIZE];
716 const int n_cache = av1_get_palette_cache(xd, 1, color_cache);
717 int out_cache_colors[PALETTE_MAX_SIZE];
718 uint8_t cache_color_found[2 * PALETTE_MAX_SIZE];
719 const int n_out_cache = av1_index_color_cache(
720 color_cache, n_cache, colors_u, n, cache_color_found, out_cache_colors);
721 int n_in_cache = 0;
722 for (int i = 0; i < n_cache && n_in_cache < n; ++i) {
723 const int found = cache_color_found[i];
724 aom_write_bit(w, found);
725 n_in_cache += found;
726 }
727 delta_encode_palette_colors(out_cache_colors, n_out_cache, bit_depth, 0, w);
728
729 // V channel colors. Don't use color cache as the colors are not sorted.
730 const int max_val = 1 << bit_depth;
731 int zero_count = 0, min_bits_v = 0;
732 int bits_v =
733 av1_get_palette_delta_bits_v(pmi, bit_depth, &zero_count, &min_bits_v);
734 const int rate_using_delta =
735 2 + bit_depth + (bits_v + 1) * (n - 1) - zero_count;
736 const int rate_using_raw = bit_depth * n;
737 if (rate_using_delta < rate_using_raw) { // delta encoding
738 assert(colors_v[0] < (1 << bit_depth));
739 aom_write_bit(w, 1);
740 aom_write_literal(w, bits_v - min_bits_v, 2);
741 aom_write_literal(w, colors_v[0], bit_depth);
742 for (int i = 1; i < n; ++i) {
743 assert(colors_v[i] < (1 << bit_depth));
744 if (colors_v[i] == colors_v[i - 1]) { // No need to signal sign bit.
745 aom_write_literal(w, 0, bits_v);
746 continue;
747 }
748 const int delta = abs((int)colors_v[i] - colors_v[i - 1]);
749 const int sign_bit = colors_v[i] < colors_v[i - 1];
750 if (delta <= max_val - delta) {
751 aom_write_literal(w, delta, bits_v);
752 aom_write_bit(w, sign_bit);
753 } else {
754 aom_write_literal(w, max_val - delta, bits_v);
755 aom_write_bit(w, !sign_bit);
756 }
757 }
758 } else { // Transmit raw values.
759 aom_write_bit(w, 0);
760 for (int i = 0; i < n; ++i) {
761 assert(colors_v[i] < (1 << bit_depth));
762 aom_write_literal(w, colors_v[i], bit_depth);
763 }
764 }
765 }
766
write_palette_mode_info(const AV1_COMMON * cm,const MACROBLOCKD * xd,const MB_MODE_INFO * const mbmi,aom_writer * w)767 static AOM_INLINE void write_palette_mode_info(const AV1_COMMON *cm,
768 const MACROBLOCKD *xd,
769 const MB_MODE_INFO *const mbmi,
770 aom_writer *w) {
771 const int num_planes = av1_num_planes(cm);
772 const BLOCK_SIZE bsize = mbmi->bsize;
773 assert(av1_allow_palette(cm->features.allow_screen_content_tools, bsize));
774 const PALETTE_MODE_INFO *const pmi = &mbmi->palette_mode_info;
775 const int bsize_ctx = av1_get_palette_bsize_ctx(bsize);
776
777 if (mbmi->mode == DC_PRED) {
778 const int n = pmi->palette_size[0];
779 const int palette_y_mode_ctx = av1_get_palette_mode_ctx(xd);
780 aom_write_symbol(
781 w, n > 0,
782 xd->tile_ctx->palette_y_mode_cdf[bsize_ctx][palette_y_mode_ctx], 2);
783 if (n > 0) {
784 aom_write_symbol(w, n - PALETTE_MIN_SIZE,
785 xd->tile_ctx->palette_y_size_cdf[bsize_ctx],
786 PALETTE_SIZES);
787 write_palette_colors_y(xd, pmi, cm->seq_params->bit_depth, w);
788 }
789 }
790
791 const int uv_dc_pred =
792 num_planes > 1 && mbmi->uv_mode == UV_DC_PRED && xd->is_chroma_ref;
793 if (uv_dc_pred) {
794 const int n = pmi->palette_size[1];
795 const int palette_uv_mode_ctx = (pmi->palette_size[0] > 0);
796 aom_write_symbol(w, n > 0,
797 xd->tile_ctx->palette_uv_mode_cdf[palette_uv_mode_ctx], 2);
798 if (n > 0) {
799 aom_write_symbol(w, n - PALETTE_MIN_SIZE,
800 xd->tile_ctx->palette_uv_size_cdf[bsize_ctx],
801 PALETTE_SIZES);
802 write_palette_colors_uv(xd, pmi, cm->seq_params->bit_depth, w);
803 }
804 }
805 }
806
av1_write_tx_type(const AV1_COMMON * const cm,const MACROBLOCKD * xd,TX_TYPE tx_type,TX_SIZE tx_size,aom_writer * w)807 void av1_write_tx_type(const AV1_COMMON *const cm, const MACROBLOCKD *xd,
808 TX_TYPE tx_type, TX_SIZE tx_size, aom_writer *w) {
809 MB_MODE_INFO *mbmi = xd->mi[0];
810 const FeatureFlags *const features = &cm->features;
811 const int is_inter = is_inter_block(mbmi);
812 if (get_ext_tx_types(tx_size, is_inter, features->reduced_tx_set_used) > 1 &&
813 ((!cm->seg.enabled && cm->quant_params.base_qindex > 0) ||
814 (cm->seg.enabled && xd->qindex[mbmi->segment_id] > 0)) &&
815 !mbmi->skip_txfm &&
816 !segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP)) {
817 FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
818 const TX_SIZE square_tx_size = txsize_sqr_map[tx_size];
819 const TxSetType tx_set_type = av1_get_ext_tx_set_type(
820 tx_size, is_inter, features->reduced_tx_set_used);
821 const int eset =
822 get_ext_tx_set(tx_size, is_inter, features->reduced_tx_set_used);
823 // eset == 0 should correspond to a set with only DCT_DCT and there
824 // is no need to send the tx_type
825 assert(eset > 0);
826 assert(av1_ext_tx_used[tx_set_type][tx_type]);
827 if (is_inter) {
828 aom_write_symbol(w, av1_ext_tx_ind[tx_set_type][tx_type],
829 ec_ctx->inter_ext_tx_cdf[eset][square_tx_size],
830 av1_num_ext_tx_set[tx_set_type]);
831 } else {
832 PREDICTION_MODE intra_dir;
833 if (mbmi->filter_intra_mode_info.use_filter_intra)
834 intra_dir =
835 fimode_to_intradir[mbmi->filter_intra_mode_info.filter_intra_mode];
836 else
837 intra_dir = mbmi->mode;
838 aom_write_symbol(
839 w, av1_ext_tx_ind[tx_set_type][tx_type],
840 ec_ctx->intra_ext_tx_cdf[eset][square_tx_size][intra_dir],
841 av1_num_ext_tx_set[tx_set_type]);
842 }
843 }
844 }
845
write_intra_y_mode_nonkf(FRAME_CONTEXT * frame_ctx,BLOCK_SIZE bsize,PREDICTION_MODE mode,aom_writer * w)846 static AOM_INLINE void write_intra_y_mode_nonkf(FRAME_CONTEXT *frame_ctx,
847 BLOCK_SIZE bsize,
848 PREDICTION_MODE mode,
849 aom_writer *w) {
850 aom_write_symbol(w, mode, frame_ctx->y_mode_cdf[size_group_lookup[bsize]],
851 INTRA_MODES);
852 }
853
write_intra_uv_mode(FRAME_CONTEXT * frame_ctx,UV_PREDICTION_MODE uv_mode,PREDICTION_MODE y_mode,CFL_ALLOWED_TYPE cfl_allowed,aom_writer * w)854 static AOM_INLINE void write_intra_uv_mode(FRAME_CONTEXT *frame_ctx,
855 UV_PREDICTION_MODE uv_mode,
856 PREDICTION_MODE y_mode,
857 CFL_ALLOWED_TYPE cfl_allowed,
858 aom_writer *w) {
859 aom_write_symbol(w, uv_mode, frame_ctx->uv_mode_cdf[cfl_allowed][y_mode],
860 UV_INTRA_MODES - !cfl_allowed);
861 }
862
write_cfl_alphas(FRAME_CONTEXT * const ec_ctx,uint8_t idx,int8_t joint_sign,aom_writer * w)863 static AOM_INLINE void write_cfl_alphas(FRAME_CONTEXT *const ec_ctx,
864 uint8_t idx, int8_t joint_sign,
865 aom_writer *w) {
866 aom_write_symbol(w, joint_sign, ec_ctx->cfl_sign_cdf, CFL_JOINT_SIGNS);
867 // Magnitudes are only signaled for nonzero codes.
868 if (CFL_SIGN_U(joint_sign) != CFL_SIGN_ZERO) {
869 aom_cdf_prob *cdf_u = ec_ctx->cfl_alpha_cdf[CFL_CONTEXT_U(joint_sign)];
870 aom_write_symbol(w, CFL_IDX_U(idx), cdf_u, CFL_ALPHABET_SIZE);
871 }
872 if (CFL_SIGN_V(joint_sign) != CFL_SIGN_ZERO) {
873 aom_cdf_prob *cdf_v = ec_ctx->cfl_alpha_cdf[CFL_CONTEXT_V(joint_sign)];
874 aom_write_symbol(w, CFL_IDX_V(idx), cdf_v, CFL_ALPHABET_SIZE);
875 }
876 }
877
write_cdef(AV1_COMMON * cm,MACROBLOCKD * const xd,aom_writer * w,int skip)878 static AOM_INLINE void write_cdef(AV1_COMMON *cm, MACROBLOCKD *const xd,
879 aom_writer *w, int skip) {
880 if (cm->features.coded_lossless || cm->features.allow_intrabc) return;
881
882 // At the start of a superblock, mark that we haven't yet written CDEF
883 // strengths for any of the CDEF units contained in this superblock.
884 const int sb_mask = (cm->seq_params->mib_size - 1);
885 const int mi_row_in_sb = (xd->mi_row & sb_mask);
886 const int mi_col_in_sb = (xd->mi_col & sb_mask);
887 if (mi_row_in_sb == 0 && mi_col_in_sb == 0) {
888 xd->cdef_transmitted[0] = xd->cdef_transmitted[1] =
889 xd->cdef_transmitted[2] = xd->cdef_transmitted[3] = false;
890 }
891
892 // CDEF unit size is 64x64 irrespective of the superblock size.
893 const int cdef_size = 1 << (6 - MI_SIZE_LOG2);
894
895 // Find index of this CDEF unit in this superblock.
896 const int index_mask = cdef_size;
897 const int cdef_unit_row_in_sb = ((xd->mi_row & index_mask) != 0);
898 const int cdef_unit_col_in_sb = ((xd->mi_col & index_mask) != 0);
899 const int index = (cm->seq_params->sb_size == BLOCK_128X128)
900 ? cdef_unit_col_in_sb + 2 * cdef_unit_row_in_sb
901 : 0;
902
903 // Write CDEF strength to the first non-skip coding block in this CDEF unit.
904 if (!xd->cdef_transmitted[index] && !skip) {
905 // CDEF strength for this CDEF unit needs to be stored in the MB_MODE_INFO
906 // of the 1st block in this CDEF unit.
907 const int first_block_mask = ~(cdef_size - 1);
908 const CommonModeInfoParams *const mi_params = &cm->mi_params;
909 const int grid_idx =
910 get_mi_grid_idx(mi_params, xd->mi_row & first_block_mask,
911 xd->mi_col & first_block_mask);
912 const MB_MODE_INFO *const mbmi = mi_params->mi_grid_base[grid_idx];
913 aom_write_literal(w, mbmi->cdef_strength, cm->cdef_info.cdef_bits);
914 xd->cdef_transmitted[index] = true;
915 }
916 }
917
write_inter_segment_id(AV1_COMP * cpi,MACROBLOCKD * const xd,aom_writer * w,const struct segmentation * const seg,struct segmentation_probs * const segp,int skip,int preskip)918 static AOM_INLINE void write_inter_segment_id(
919 AV1_COMP *cpi, MACROBLOCKD *const xd, aom_writer *w,
920 const struct segmentation *const seg, struct segmentation_probs *const segp,
921 int skip, int preskip) {
922 MB_MODE_INFO *const mbmi = xd->mi[0];
923 AV1_COMMON *const cm = &cpi->common;
924 const int mi_row = xd->mi_row;
925 const int mi_col = xd->mi_col;
926
927 if (seg->update_map) {
928 if (preskip) {
929 if (!seg->segid_preskip) return;
930 } else {
931 if (seg->segid_preskip) return;
932 if (skip) {
933 write_segment_id(cpi, xd, mbmi, w, seg, segp, 1);
934 if (seg->temporal_update) mbmi->seg_id_predicted = 0;
935 return;
936 }
937 }
938 if (seg->temporal_update) {
939 const int pred_flag = mbmi->seg_id_predicted;
940 aom_cdf_prob *pred_cdf = av1_get_pred_cdf_seg_id(segp, xd);
941 aom_write_symbol(w, pred_flag, pred_cdf, 2);
942 if (!pred_flag) {
943 write_segment_id(cpi, xd, mbmi, w, seg, segp, 0);
944 }
945 if (pred_flag) {
946 set_spatial_segment_id(&cm->mi_params, cm->cur_frame->seg_map,
947 mbmi->bsize, mi_row, mi_col, mbmi->segment_id);
948 }
949 } else {
950 write_segment_id(cpi, xd, mbmi, w, seg, segp, 0);
951 }
952 }
953 }
954
955 // If delta q is present, writes delta_q index.
956 // Also writes delta_q loop filter levels, if present.
write_delta_q_params(AV1_COMMON * const cm,MACROBLOCKD * const xd,int skip,aom_writer * w)957 static AOM_INLINE void write_delta_q_params(AV1_COMMON *const cm,
958 MACROBLOCKD *const xd, int skip,
959 aom_writer *w) {
960 const DeltaQInfo *const delta_q_info = &cm->delta_q_info;
961
962 if (delta_q_info->delta_q_present_flag) {
963 const MB_MODE_INFO *const mbmi = xd->mi[0];
964 const BLOCK_SIZE bsize = mbmi->bsize;
965 const int super_block_upper_left =
966 ((xd->mi_row & (cm->seq_params->mib_size - 1)) == 0) &&
967 ((xd->mi_col & (cm->seq_params->mib_size - 1)) == 0);
968
969 if ((bsize != cm->seq_params->sb_size || skip == 0) &&
970 super_block_upper_left) {
971 assert(mbmi->current_qindex > 0);
972 const int reduced_delta_qindex =
973 (mbmi->current_qindex - xd->current_base_qindex) /
974 delta_q_info->delta_q_res;
975 write_delta_qindex(xd, reduced_delta_qindex, w);
976 xd->current_base_qindex = mbmi->current_qindex;
977 if (delta_q_info->delta_lf_present_flag) {
978 if (delta_q_info->delta_lf_multi) {
979 const int frame_lf_count =
980 av1_num_planes(cm) > 1 ? FRAME_LF_COUNT : FRAME_LF_COUNT - 2;
981 for (int lf_id = 0; lf_id < frame_lf_count; ++lf_id) {
982 int reduced_delta_lflevel =
983 (mbmi->delta_lf[lf_id] - xd->delta_lf[lf_id]) /
984 delta_q_info->delta_lf_res;
985 write_delta_lflevel(cm, xd, lf_id, reduced_delta_lflevel, 1, w);
986 xd->delta_lf[lf_id] = mbmi->delta_lf[lf_id];
987 }
988 } else {
989 int reduced_delta_lflevel =
990 (mbmi->delta_lf_from_base - xd->delta_lf_from_base) /
991 delta_q_info->delta_lf_res;
992 write_delta_lflevel(cm, xd, -1, reduced_delta_lflevel, 0, w);
993 xd->delta_lf_from_base = mbmi->delta_lf_from_base;
994 }
995 }
996 }
997 }
998 }
999
write_intra_prediction_modes(const AV1_COMMON * cm,MACROBLOCKD * const xd,int is_keyframe,aom_writer * w)1000 static AOM_INLINE void write_intra_prediction_modes(const AV1_COMMON *cm,
1001 MACROBLOCKD *const xd,
1002 int is_keyframe,
1003 aom_writer *w) {
1004 FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
1005 const MB_MODE_INFO *const mbmi = xd->mi[0];
1006 const PREDICTION_MODE mode = mbmi->mode;
1007 const BLOCK_SIZE bsize = mbmi->bsize;
1008
1009 // Y mode.
1010 if (is_keyframe) {
1011 const MB_MODE_INFO *const above_mi = xd->above_mbmi;
1012 const MB_MODE_INFO *const left_mi = xd->left_mbmi;
1013 write_intra_y_mode_kf(ec_ctx, mbmi, above_mi, left_mi, mode, w);
1014 } else {
1015 write_intra_y_mode_nonkf(ec_ctx, bsize, mode, w);
1016 }
1017
1018 // Y angle delta.
1019 const int use_angle_delta = av1_use_angle_delta(bsize);
1020 if (use_angle_delta && av1_is_directional_mode(mode)) {
1021 write_angle_delta(w, mbmi->angle_delta[PLANE_TYPE_Y],
1022 ec_ctx->angle_delta_cdf[mode - V_PRED]);
1023 }
1024
1025 // UV mode and UV angle delta.
1026 if (!cm->seq_params->monochrome && xd->is_chroma_ref) {
1027 const UV_PREDICTION_MODE uv_mode = mbmi->uv_mode;
1028 write_intra_uv_mode(ec_ctx, uv_mode, mode, is_cfl_allowed(xd), w);
1029 if (uv_mode == UV_CFL_PRED)
1030 write_cfl_alphas(ec_ctx, mbmi->cfl_alpha_idx, mbmi->cfl_alpha_signs, w);
1031 const PREDICTION_MODE intra_mode = get_uv_mode(uv_mode);
1032 if (use_angle_delta && av1_is_directional_mode(intra_mode)) {
1033 write_angle_delta(w, mbmi->angle_delta[PLANE_TYPE_UV],
1034 ec_ctx->angle_delta_cdf[intra_mode - V_PRED]);
1035 }
1036 }
1037
1038 // Palette.
1039 if (av1_allow_palette(cm->features.allow_screen_content_tools, bsize)) {
1040 write_palette_mode_info(cm, xd, mbmi, w);
1041 }
1042
1043 // Filter intra.
1044 write_filter_intra_mode_info(cm, xd, mbmi, w);
1045 }
1046
mode_context_analyzer(const int16_t mode_context,const MV_REFERENCE_FRAME * const rf)1047 static INLINE int16_t mode_context_analyzer(
1048 const int16_t mode_context, const MV_REFERENCE_FRAME *const rf) {
1049 if (rf[1] <= INTRA_FRAME) return mode_context;
1050
1051 const int16_t newmv_ctx = mode_context & NEWMV_CTX_MASK;
1052 const int16_t refmv_ctx = (mode_context >> REFMV_OFFSET) & REFMV_CTX_MASK;
1053
1054 const int16_t comp_ctx = compound_mode_ctx_map[refmv_ctx >> 1][AOMMIN(
1055 newmv_ctx, COMP_NEWMV_CTXS - 1)];
1056 return comp_ctx;
1057 }
1058
get_ref_mv_from_stack(int ref_idx,const MV_REFERENCE_FRAME * ref_frame,int ref_mv_idx,const MB_MODE_INFO_EXT_FRAME * mbmi_ext_frame)1059 static INLINE int_mv get_ref_mv_from_stack(
1060 int ref_idx, const MV_REFERENCE_FRAME *ref_frame, int ref_mv_idx,
1061 const MB_MODE_INFO_EXT_FRAME *mbmi_ext_frame) {
1062 const int8_t ref_frame_type = av1_ref_frame_type(ref_frame);
1063 const CANDIDATE_MV *curr_ref_mv_stack = mbmi_ext_frame->ref_mv_stack;
1064
1065 if (ref_frame[1] > INTRA_FRAME) {
1066 assert(ref_idx == 0 || ref_idx == 1);
1067 return ref_idx ? curr_ref_mv_stack[ref_mv_idx].comp_mv
1068 : curr_ref_mv_stack[ref_mv_idx].this_mv;
1069 }
1070
1071 assert(ref_idx == 0);
1072 return ref_mv_idx < mbmi_ext_frame->ref_mv_count
1073 ? curr_ref_mv_stack[ref_mv_idx].this_mv
1074 : mbmi_ext_frame->global_mvs[ref_frame_type];
1075 }
1076
get_ref_mv(const MACROBLOCK * x,int ref_idx)1077 static INLINE int_mv get_ref_mv(const MACROBLOCK *x, int ref_idx) {
1078 const MACROBLOCKD *xd = &x->e_mbd;
1079 const MB_MODE_INFO *mbmi = xd->mi[0];
1080 int ref_mv_idx = mbmi->ref_mv_idx;
1081 if (mbmi->mode == NEAR_NEWMV || mbmi->mode == NEW_NEARMV) {
1082 assert(has_second_ref(mbmi));
1083 ref_mv_idx += 1;
1084 }
1085 return get_ref_mv_from_stack(ref_idx, mbmi->ref_frame, ref_mv_idx,
1086 x->mbmi_ext_frame);
1087 }
1088
pack_inter_mode_mvs(AV1_COMP * cpi,ThreadData * const td,aom_writer * w)1089 static AOM_INLINE void pack_inter_mode_mvs(AV1_COMP *cpi, ThreadData *const td,
1090 aom_writer *w) {
1091 AV1_COMMON *const cm = &cpi->common;
1092 MACROBLOCK *const x = &td->mb;
1093 MACROBLOCKD *const xd = &x->e_mbd;
1094 FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
1095 const struct segmentation *const seg = &cm->seg;
1096 struct segmentation_probs *const segp = &ec_ctx->seg;
1097 const MB_MODE_INFO *const mbmi = xd->mi[0];
1098 const MB_MODE_INFO_EXT_FRAME *const mbmi_ext_frame = x->mbmi_ext_frame;
1099 const PREDICTION_MODE mode = mbmi->mode;
1100 const uint8_t segment_id = mbmi->segment_id;
1101 const BLOCK_SIZE bsize = mbmi->bsize;
1102 const int allow_hp = cm->features.allow_high_precision_mv;
1103 const int is_inter = is_inter_block(mbmi);
1104 const int is_compound = has_second_ref(mbmi);
1105 int ref;
1106
1107 write_inter_segment_id(cpi, xd, w, seg, segp, 0, 1);
1108
1109 write_skip_mode(cm, xd, segment_id, mbmi, w);
1110
1111 assert(IMPLIES(mbmi->skip_mode, mbmi->skip_txfm));
1112 const int skip =
1113 mbmi->skip_mode ? 1 : write_skip(cm, xd, segment_id, mbmi, w);
1114
1115 write_inter_segment_id(cpi, xd, w, seg, segp, skip, 0);
1116
1117 write_cdef(cm, xd, w, skip);
1118
1119 write_delta_q_params(cm, xd, skip, w);
1120
1121 if (!mbmi->skip_mode) write_is_inter(cm, xd, mbmi->segment_id, w, is_inter);
1122
1123 if (mbmi->skip_mode) return;
1124
1125 if (!is_inter) {
1126 write_intra_prediction_modes(cm, xd, 0, w);
1127 } else {
1128 int16_t mode_ctx;
1129
1130 av1_collect_neighbors_ref_counts(xd);
1131
1132 write_ref_frames(cm, xd, w);
1133
1134 mode_ctx =
1135 mode_context_analyzer(mbmi_ext_frame->mode_context, mbmi->ref_frame);
1136
1137 // If segment skip is not enabled code the mode.
1138 if (!segfeature_active(seg, segment_id, SEG_LVL_SKIP)) {
1139 if (is_inter_compound_mode(mode))
1140 write_inter_compound_mode(xd, w, mode, mode_ctx);
1141 else if (is_inter_singleref_mode(mode))
1142 write_inter_mode(w, mode, ec_ctx, mode_ctx);
1143
1144 if (mode == NEWMV || mode == NEW_NEWMV || have_nearmv_in_inter_mode(mode))
1145 write_drl_idx(ec_ctx, mbmi, mbmi_ext_frame, w);
1146 else
1147 assert(mbmi->ref_mv_idx == 0);
1148 }
1149
1150 if (mode == NEWMV || mode == NEW_NEWMV) {
1151 for (ref = 0; ref < 1 + is_compound; ++ref) {
1152 nmv_context *nmvc = &ec_ctx->nmvc;
1153 const int_mv ref_mv = get_ref_mv(x, ref);
1154 av1_encode_mv(cpi, w, td, &mbmi->mv[ref].as_mv, &ref_mv.as_mv, nmvc,
1155 allow_hp);
1156 }
1157 } else if (mode == NEAREST_NEWMV || mode == NEAR_NEWMV) {
1158 nmv_context *nmvc = &ec_ctx->nmvc;
1159 const int_mv ref_mv = get_ref_mv(x, 1);
1160 av1_encode_mv(cpi, w, td, &mbmi->mv[1].as_mv, &ref_mv.as_mv, nmvc,
1161 allow_hp);
1162 } else if (mode == NEW_NEARESTMV || mode == NEW_NEARMV) {
1163 nmv_context *nmvc = &ec_ctx->nmvc;
1164 const int_mv ref_mv = get_ref_mv(x, 0);
1165 av1_encode_mv(cpi, w, td, &mbmi->mv[0].as_mv, &ref_mv.as_mv, nmvc,
1166 allow_hp);
1167 }
1168
1169 if (cpi->common.current_frame.reference_mode != COMPOUND_REFERENCE &&
1170 cpi->common.seq_params->enable_interintra_compound &&
1171 is_interintra_allowed(mbmi)) {
1172 const int interintra = mbmi->ref_frame[1] == INTRA_FRAME;
1173 const int bsize_group = size_group_lookup[bsize];
1174 aom_write_symbol(w, interintra, ec_ctx->interintra_cdf[bsize_group], 2);
1175 if (interintra) {
1176 aom_write_symbol(w, mbmi->interintra_mode,
1177 ec_ctx->interintra_mode_cdf[bsize_group],
1178 INTERINTRA_MODES);
1179 if (av1_is_wedge_used(bsize)) {
1180 aom_write_symbol(w, mbmi->use_wedge_interintra,
1181 ec_ctx->wedge_interintra_cdf[bsize], 2);
1182 if (mbmi->use_wedge_interintra) {
1183 aom_write_symbol(w, mbmi->interintra_wedge_index,
1184 ec_ctx->wedge_idx_cdf[bsize], MAX_WEDGE_TYPES);
1185 }
1186 }
1187 }
1188 }
1189
1190 if (mbmi->ref_frame[1] != INTRA_FRAME) write_motion_mode(cm, xd, mbmi, w);
1191
1192 // First write idx to indicate current compound inter prediction mode group
1193 // Group A (0): dist_wtd_comp, compound_average
1194 // Group B (1): interintra, compound_diffwtd, wedge
1195 if (has_second_ref(mbmi)) {
1196 const int masked_compound_used = is_any_masked_compound_used(bsize) &&
1197 cm->seq_params->enable_masked_compound;
1198
1199 if (masked_compound_used) {
1200 const int ctx_comp_group_idx = get_comp_group_idx_context(xd);
1201 aom_write_symbol(w, mbmi->comp_group_idx,
1202 ec_ctx->comp_group_idx_cdf[ctx_comp_group_idx], 2);
1203 } else {
1204 assert(mbmi->comp_group_idx == 0);
1205 }
1206
1207 if (mbmi->comp_group_idx == 0) {
1208 if (mbmi->compound_idx)
1209 assert(mbmi->interinter_comp.type == COMPOUND_AVERAGE);
1210
1211 if (cm->seq_params->order_hint_info.enable_dist_wtd_comp) {
1212 const int comp_index_ctx = get_comp_index_context(cm, xd);
1213 aom_write_symbol(w, mbmi->compound_idx,
1214 ec_ctx->compound_index_cdf[comp_index_ctx], 2);
1215 } else {
1216 assert(mbmi->compound_idx == 1);
1217 }
1218 } else {
1219 assert(cpi->common.current_frame.reference_mode != SINGLE_REFERENCE &&
1220 is_inter_compound_mode(mbmi->mode) &&
1221 mbmi->motion_mode == SIMPLE_TRANSLATION);
1222 assert(masked_compound_used);
1223 // compound_diffwtd, wedge
1224 assert(mbmi->interinter_comp.type == COMPOUND_WEDGE ||
1225 mbmi->interinter_comp.type == COMPOUND_DIFFWTD);
1226
1227 if (is_interinter_compound_used(COMPOUND_WEDGE, bsize))
1228 aom_write_symbol(w, mbmi->interinter_comp.type - COMPOUND_WEDGE,
1229 ec_ctx->compound_type_cdf[bsize],
1230 MASKED_COMPOUND_TYPES);
1231
1232 if (mbmi->interinter_comp.type == COMPOUND_WEDGE) {
1233 assert(is_interinter_compound_used(COMPOUND_WEDGE, bsize));
1234 aom_write_symbol(w, mbmi->interinter_comp.wedge_index,
1235 ec_ctx->wedge_idx_cdf[bsize], MAX_WEDGE_TYPES);
1236 aom_write_bit(w, mbmi->interinter_comp.wedge_sign);
1237 } else {
1238 assert(mbmi->interinter_comp.type == COMPOUND_DIFFWTD);
1239 aom_write_literal(w, mbmi->interinter_comp.mask_type,
1240 MAX_DIFFWTD_MASK_BITS);
1241 }
1242 }
1243 }
1244 write_mb_interp_filter(cm, td, w);
1245 }
1246 }
1247
write_intrabc_info(MACROBLOCKD * xd,const MB_MODE_INFO_EXT_FRAME * mbmi_ext_frame,aom_writer * w)1248 static AOM_INLINE void write_intrabc_info(
1249 MACROBLOCKD *xd, const MB_MODE_INFO_EXT_FRAME *mbmi_ext_frame,
1250 aom_writer *w) {
1251 const MB_MODE_INFO *const mbmi = xd->mi[0];
1252 int use_intrabc = is_intrabc_block(mbmi);
1253 FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
1254 aom_write_symbol(w, use_intrabc, ec_ctx->intrabc_cdf, 2);
1255 if (use_intrabc) {
1256 assert(mbmi->mode == DC_PRED);
1257 assert(mbmi->uv_mode == UV_DC_PRED);
1258 assert(mbmi->motion_mode == SIMPLE_TRANSLATION);
1259 int_mv dv_ref = mbmi_ext_frame->ref_mv_stack[0].this_mv;
1260 av1_encode_dv(w, &mbmi->mv[0].as_mv, &dv_ref.as_mv, &ec_ctx->ndvc);
1261 }
1262 }
1263
write_mb_modes_kf(AV1_COMP * cpi,MACROBLOCKD * xd,const MB_MODE_INFO_EXT_FRAME * mbmi_ext_frame,aom_writer * w)1264 static AOM_INLINE void write_mb_modes_kf(
1265 AV1_COMP *cpi, MACROBLOCKD *xd,
1266 const MB_MODE_INFO_EXT_FRAME *mbmi_ext_frame, aom_writer *w) {
1267 AV1_COMMON *const cm = &cpi->common;
1268 FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
1269 const struct segmentation *const seg = &cm->seg;
1270 struct segmentation_probs *const segp = &ec_ctx->seg;
1271 const MB_MODE_INFO *const mbmi = xd->mi[0];
1272
1273 if (seg->segid_preskip && seg->update_map)
1274 write_segment_id(cpi, xd, mbmi, w, seg, segp, 0);
1275
1276 const int skip = write_skip(cm, xd, mbmi->segment_id, mbmi, w);
1277
1278 if (!seg->segid_preskip && seg->update_map)
1279 write_segment_id(cpi, xd, mbmi, w, seg, segp, skip);
1280
1281 write_cdef(cm, xd, w, skip);
1282
1283 write_delta_q_params(cm, xd, skip, w);
1284
1285 if (av1_allow_intrabc(cm)) {
1286 write_intrabc_info(xd, mbmi_ext_frame, w);
1287 if (is_intrabc_block(mbmi)) return;
1288 }
1289
1290 write_intra_prediction_modes(cm, xd, 1, w);
1291 }
1292
1293 #if CONFIG_RD_DEBUG
dump_mode_info(MB_MODE_INFO * mi)1294 static AOM_INLINE void dump_mode_info(MB_MODE_INFO *mi) {
1295 printf("\nmi->mi_row == %d\n", mi->mi_row);
1296 printf("&& mi->mi_col == %d\n", mi->mi_col);
1297 printf("&& mi->bsize == %d\n", mi->bsize);
1298 printf("&& mi->tx_size == %d\n", mi->tx_size);
1299 printf("&& mi->mode == %d\n", mi->mode);
1300 }
1301
rd_token_stats_mismatch(RD_STATS * rd_stats,TOKEN_STATS * token_stats,int plane)1302 static int rd_token_stats_mismatch(RD_STATS *rd_stats, TOKEN_STATS *token_stats,
1303 int plane) {
1304 if (rd_stats->txb_coeff_cost[plane] != token_stats->cost) {
1305 printf("\nplane %d rd_stats->txb_coeff_cost %d token_stats->cost %d\n",
1306 plane, rd_stats->txb_coeff_cost[plane], token_stats->cost);
1307 return 1;
1308 }
1309 return 0;
1310 }
1311 #endif
1312
1313 #if ENC_MISMATCH_DEBUG
enc_dump_logs(const AV1_COMMON * const cm,const MBMIExtFrameBufferInfo * const mbmi_ext_info,int mi_row,int mi_col)1314 static AOM_INLINE void enc_dump_logs(
1315 const AV1_COMMON *const cm,
1316 const MBMIExtFrameBufferInfo *const mbmi_ext_info, int mi_row, int mi_col) {
1317 const MB_MODE_INFO *const mbmi = *(
1318 cm->mi_params.mi_grid_base + (mi_row * cm->mi_params.mi_stride + mi_col));
1319 const MB_MODE_INFO_EXT_FRAME *const mbmi_ext_frame =
1320 mbmi_ext_info->frame_base + get_mi_ext_idx(mi_row, mi_col,
1321 cm->mi_params.mi_alloc_bsize,
1322 mbmi_ext_info->stride);
1323 if (is_inter_block(mbmi)) {
1324 #define FRAME_TO_CHECK 11
1325 if (cm->current_frame.frame_number == FRAME_TO_CHECK &&
1326 cm->show_frame == 1) {
1327 const BLOCK_SIZE bsize = mbmi->bsize;
1328
1329 int_mv mv[2] = { 0 };
1330 const int is_comp_ref = has_second_ref(mbmi);
1331
1332 for (int ref = 0; ref < 1 + is_comp_ref; ++ref)
1333 mv[ref].as_mv = mbmi->mv[ref].as_mv;
1334
1335 if (!is_comp_ref) {
1336 mv[1].as_int = 0;
1337 }
1338
1339 const int16_t mode_ctx =
1340 is_comp_ref ? 0
1341 : mode_context_analyzer(mbmi_ext_frame->mode_context,
1342 mbmi->ref_frame);
1343
1344 const int16_t newmv_ctx = mode_ctx & NEWMV_CTX_MASK;
1345 int16_t zeromv_ctx = -1;
1346 int16_t refmv_ctx = -1;
1347
1348 if (mbmi->mode != NEWMV) {
1349 zeromv_ctx = (mode_ctx >> GLOBALMV_OFFSET) & GLOBALMV_CTX_MASK;
1350 if (mbmi->mode != GLOBALMV)
1351 refmv_ctx = (mode_ctx >> REFMV_OFFSET) & REFMV_CTX_MASK;
1352 }
1353
1354 printf(
1355 "=== ENCODER ===: "
1356 "Frame=%d, (mi_row,mi_col)=(%d,%d), skip_mode=%d, mode=%d, bsize=%d, "
1357 "show_frame=%d, mv[0]=(%d,%d), mv[1]=(%d,%d), ref[0]=%d, "
1358 "ref[1]=%d, motion_mode=%d, mode_ctx=%d, "
1359 "newmv_ctx=%d, zeromv_ctx=%d, refmv_ctx=%d, tx_size=%d\n",
1360 cm->current_frame.frame_number, mi_row, mi_col, mbmi->skip_mode,
1361 mbmi->mode, bsize, cm->show_frame, mv[0].as_mv.row, mv[0].as_mv.col,
1362 mv[1].as_mv.row, mv[1].as_mv.col, mbmi->ref_frame[0],
1363 mbmi->ref_frame[1], mbmi->motion_mode, mode_ctx, newmv_ctx,
1364 zeromv_ctx, refmv_ctx, mbmi->tx_size);
1365 }
1366 }
1367 }
1368 #endif // ENC_MISMATCH_DEBUG
1369
write_mbmi_b(AV1_COMP * cpi,ThreadData * const td,aom_writer * w)1370 static AOM_INLINE void write_mbmi_b(AV1_COMP *cpi, ThreadData *const td,
1371 aom_writer *w) {
1372 AV1_COMMON *const cm = &cpi->common;
1373 MACROBLOCKD *const xd = &td->mb.e_mbd;
1374 MB_MODE_INFO *m = xd->mi[0];
1375
1376 if (frame_is_intra_only(cm)) {
1377 write_mb_modes_kf(cpi, xd, td->mb.mbmi_ext_frame, w);
1378 } else {
1379 // has_subpel_mv_component needs the ref frame buffers set up to look
1380 // up if they are scaled. has_subpel_mv_component is in turn needed by
1381 // write_switchable_interp_filter, which is called by pack_inter_mode_mvs.
1382 set_ref_ptrs(cm, xd, m->ref_frame[0], m->ref_frame[1]);
1383
1384 #if ENC_MISMATCH_DEBUG
1385 enc_dump_logs(cm, &cpi->mbmi_ext_info, xd->mi_row, xd->mi_col);
1386 #endif // ENC_MISMATCH_DEBUG
1387
1388 pack_inter_mode_mvs(cpi, td, w);
1389 }
1390 }
1391
write_inter_txb_coeff(AV1_COMMON * const cm,MACROBLOCK * const x,MB_MODE_INFO * const mbmi,aom_writer * w,const TokenExtra ** tok,const TokenExtra * const tok_end,TOKEN_STATS * token_stats,const int row,const int col,int * block,const int plane)1392 static AOM_INLINE void write_inter_txb_coeff(
1393 AV1_COMMON *const cm, MACROBLOCK *const x, MB_MODE_INFO *const mbmi,
1394 aom_writer *w, const TokenExtra **tok, const TokenExtra *const tok_end,
1395 TOKEN_STATS *token_stats, const int row, const int col, int *block,
1396 const int plane) {
1397 MACROBLOCKD *const xd = &x->e_mbd;
1398 const struct macroblockd_plane *const pd = &xd->plane[plane];
1399 const BLOCK_SIZE bsize = mbmi->bsize;
1400 assert(bsize < BLOCK_SIZES_ALL);
1401 const int ss_x = pd->subsampling_x;
1402 const int ss_y = pd->subsampling_y;
1403 const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, ss_x, ss_y);
1404 assert(plane_bsize < BLOCK_SIZES_ALL);
1405 const TX_SIZE max_tx_size = get_vartx_max_txsize(xd, plane_bsize, plane);
1406 const int step =
1407 tx_size_wide_unit[max_tx_size] * tx_size_high_unit[max_tx_size];
1408 const int bkw = tx_size_wide_unit[max_tx_size];
1409 const int bkh = tx_size_high_unit[max_tx_size];
1410 const BLOCK_SIZE max_unit_bsize =
1411 get_plane_block_size(BLOCK_64X64, ss_x, ss_y);
1412 const int num_4x4_w = mi_size_wide[plane_bsize];
1413 const int num_4x4_h = mi_size_high[plane_bsize];
1414 const int mu_blocks_wide = mi_size_wide[max_unit_bsize];
1415 const int mu_blocks_high = mi_size_high[max_unit_bsize];
1416 const int unit_height = AOMMIN(mu_blocks_high + (row >> ss_y), num_4x4_h);
1417 const int unit_width = AOMMIN(mu_blocks_wide + (col >> ss_x), num_4x4_w);
1418 for (int blk_row = row >> ss_y; blk_row < unit_height; blk_row += bkh) {
1419 for (int blk_col = col >> ss_x; blk_col < unit_width; blk_col += bkw) {
1420 pack_txb_tokens(w, cm, x, tok, tok_end, xd, mbmi, plane, plane_bsize,
1421 cm->seq_params->bit_depth, *block, blk_row, blk_col,
1422 max_tx_size, token_stats);
1423 *block += step;
1424 }
1425 }
1426 }
1427
write_tokens_b(AV1_COMP * cpi,MACROBLOCK * const x,aom_writer * w,const TokenExtra ** tok,const TokenExtra * const tok_end)1428 static AOM_INLINE void write_tokens_b(AV1_COMP *cpi, MACROBLOCK *const x,
1429 aom_writer *w, const TokenExtra **tok,
1430 const TokenExtra *const tok_end) {
1431 AV1_COMMON *const cm = &cpi->common;
1432 MACROBLOCKD *const xd = &x->e_mbd;
1433 MB_MODE_INFO *const mbmi = xd->mi[0];
1434 const BLOCK_SIZE bsize = mbmi->bsize;
1435
1436 assert(!mbmi->skip_txfm);
1437
1438 const int is_inter = is_inter_block(mbmi);
1439 if (!is_inter) {
1440 av1_write_intra_coeffs_mb(cm, x, w, bsize);
1441 } else {
1442 int block[MAX_MB_PLANE] = { 0 };
1443 assert(bsize == get_plane_block_size(bsize, xd->plane[0].subsampling_x,
1444 xd->plane[0].subsampling_y));
1445 const int num_4x4_w = mi_size_wide[bsize];
1446 const int num_4x4_h = mi_size_high[bsize];
1447 TOKEN_STATS token_stats;
1448 init_token_stats(&token_stats);
1449
1450 const BLOCK_SIZE max_unit_bsize = BLOCK_64X64;
1451 assert(max_unit_bsize == get_plane_block_size(BLOCK_64X64,
1452 xd->plane[0].subsampling_x,
1453 xd->plane[0].subsampling_y));
1454 int mu_blocks_wide = mi_size_wide[max_unit_bsize];
1455 int mu_blocks_high = mi_size_high[max_unit_bsize];
1456 mu_blocks_wide = AOMMIN(num_4x4_w, mu_blocks_wide);
1457 mu_blocks_high = AOMMIN(num_4x4_h, mu_blocks_high);
1458
1459 const int num_planes = av1_num_planes(cm);
1460 for (int row = 0; row < num_4x4_h; row += mu_blocks_high) {
1461 for (int col = 0; col < num_4x4_w; col += mu_blocks_wide) {
1462 for (int plane = 0; plane < num_planes; ++plane) {
1463 if (plane && !xd->is_chroma_ref) break;
1464 write_inter_txb_coeff(cm, x, mbmi, w, tok, tok_end, &token_stats, row,
1465 col, &block[plane], plane);
1466 }
1467 }
1468 }
1469 #if CONFIG_RD_DEBUG
1470 for (int plane = 0; plane < num_planes; ++plane) {
1471 if (mbmi->bsize >= BLOCK_8X8 &&
1472 rd_token_stats_mismatch(&mbmi->rd_stats, &token_stats, plane)) {
1473 dump_mode_info(mbmi);
1474 assert(0);
1475 }
1476 }
1477 #endif // CONFIG_RD_DEBUG
1478 }
1479 }
1480
write_modes_b(AV1_COMP * cpi,ThreadData * const td,const TileInfo * const tile,aom_writer * w,const TokenExtra ** tok,const TokenExtra * const tok_end,int mi_row,int mi_col)1481 static AOM_INLINE void write_modes_b(AV1_COMP *cpi, ThreadData *const td,
1482 const TileInfo *const tile, aom_writer *w,
1483 const TokenExtra **tok,
1484 const TokenExtra *const tok_end,
1485 int mi_row, int mi_col) {
1486 const AV1_COMMON *cm = &cpi->common;
1487 const CommonModeInfoParams *const mi_params = &cm->mi_params;
1488 MACROBLOCKD *xd = &td->mb.e_mbd;
1489 FRAME_CONTEXT *tile_ctx = xd->tile_ctx;
1490 const int grid_idx = mi_row * mi_params->mi_stride + mi_col;
1491 xd->mi = mi_params->mi_grid_base + grid_idx;
1492 td->mb.mbmi_ext_frame =
1493 cpi->mbmi_ext_info.frame_base +
1494 get_mi_ext_idx(mi_row, mi_col, cm->mi_params.mi_alloc_bsize,
1495 cpi->mbmi_ext_info.stride);
1496 xd->tx_type_map = mi_params->tx_type_map + grid_idx;
1497 xd->tx_type_map_stride = mi_params->mi_stride;
1498
1499 const MB_MODE_INFO *mbmi = xd->mi[0];
1500 const BLOCK_SIZE bsize = mbmi->bsize;
1501 assert(bsize <= cm->seq_params->sb_size ||
1502 (bsize >= BLOCK_SIZES && bsize < BLOCK_SIZES_ALL));
1503
1504 const int bh = mi_size_high[bsize];
1505 const int bw = mi_size_wide[bsize];
1506 set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, mi_params->mi_rows,
1507 mi_params->mi_cols);
1508
1509 xd->above_txfm_context = cm->above_contexts.txfm[tile->tile_row] + mi_col;
1510 xd->left_txfm_context =
1511 xd->left_txfm_context_buffer + (mi_row & MAX_MIB_MASK);
1512
1513 write_mbmi_b(cpi, td, w);
1514
1515 for (int plane = 0; plane < AOMMIN(2, av1_num_planes(cm)); ++plane) {
1516 const uint8_t palette_size_plane =
1517 mbmi->palette_mode_info.palette_size[plane];
1518 assert(!mbmi->skip_mode || !palette_size_plane);
1519 if (palette_size_plane > 0) {
1520 assert(mbmi->use_intrabc == 0);
1521 assert(av1_allow_palette(cm->features.allow_screen_content_tools,
1522 mbmi->bsize));
1523 assert(!plane || xd->is_chroma_ref);
1524 int rows, cols;
1525 av1_get_block_dimensions(mbmi->bsize, plane, xd, NULL, NULL, &rows,
1526 &cols);
1527 assert(*tok < tok_end);
1528 MapCdf map_pb_cdf = plane ? tile_ctx->palette_uv_color_index_cdf
1529 : tile_ctx->palette_y_color_index_cdf;
1530 pack_map_tokens(w, tok, palette_size_plane, rows * cols, map_pb_cdf);
1531 }
1532 }
1533
1534 const int is_inter_tx = is_inter_block(mbmi);
1535 const int skip_txfm = mbmi->skip_txfm;
1536 const uint8_t segment_id = mbmi->segment_id;
1537 if (cm->features.tx_mode == TX_MODE_SELECT && block_signals_txsize(bsize) &&
1538 !(is_inter_tx && skip_txfm) && !xd->lossless[segment_id]) {
1539 if (is_inter_tx) { // This implies skip flag is 0.
1540 const TX_SIZE max_tx_size = get_vartx_max_txsize(xd, bsize, 0);
1541 const int txbh = tx_size_high_unit[max_tx_size];
1542 const int txbw = tx_size_wide_unit[max_tx_size];
1543 const int width = mi_size_wide[bsize];
1544 const int height = mi_size_high[bsize];
1545 for (int idy = 0; idy < height; idy += txbh) {
1546 for (int idx = 0; idx < width; idx += txbw) {
1547 write_tx_size_vartx(xd, mbmi, max_tx_size, 0, idy, idx, w);
1548 }
1549 }
1550 } else {
1551 write_selected_tx_size(xd, w);
1552 set_txfm_ctxs(mbmi->tx_size, xd->width, xd->height, 0, xd);
1553 }
1554 } else {
1555 set_txfm_ctxs(mbmi->tx_size, xd->width, xd->height,
1556 skip_txfm && is_inter_tx, xd);
1557 }
1558
1559 if (!mbmi->skip_txfm) {
1560 int start = aom_tell_size(w);
1561
1562 write_tokens_b(cpi, &td->mb, w, tok, tok_end);
1563
1564 const int end = aom_tell_size(w);
1565 td->coefficient_size += end - start;
1566 }
1567 }
1568
write_partition(const AV1_COMMON * const cm,const MACROBLOCKD * const xd,int hbs,int mi_row,int mi_col,PARTITION_TYPE p,BLOCK_SIZE bsize,aom_writer * w)1569 static AOM_INLINE void write_partition(const AV1_COMMON *const cm,
1570 const MACROBLOCKD *const xd, int hbs,
1571 int mi_row, int mi_col, PARTITION_TYPE p,
1572 BLOCK_SIZE bsize, aom_writer *w) {
1573 const int is_partition_point = bsize >= BLOCK_8X8;
1574
1575 if (!is_partition_point) return;
1576
1577 const int has_rows = (mi_row + hbs) < cm->mi_params.mi_rows;
1578 const int has_cols = (mi_col + hbs) < cm->mi_params.mi_cols;
1579 const int ctx = partition_plane_context(xd, mi_row, mi_col, bsize);
1580 FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
1581
1582 if (!has_rows && !has_cols) {
1583 assert(p == PARTITION_SPLIT);
1584 return;
1585 }
1586
1587 if (has_rows && has_cols) {
1588 aom_write_symbol(w, p, ec_ctx->partition_cdf[ctx],
1589 partition_cdf_length(bsize));
1590 } else if (!has_rows && has_cols) {
1591 assert(p == PARTITION_SPLIT || p == PARTITION_HORZ);
1592 assert(bsize > BLOCK_8X8);
1593 aom_cdf_prob cdf[2];
1594 partition_gather_vert_alike(cdf, ec_ctx->partition_cdf[ctx], bsize);
1595 aom_write_cdf(w, p == PARTITION_SPLIT, cdf, 2);
1596 } else {
1597 assert(has_rows && !has_cols);
1598 assert(p == PARTITION_SPLIT || p == PARTITION_VERT);
1599 assert(bsize > BLOCK_8X8);
1600 aom_cdf_prob cdf[2];
1601 partition_gather_horz_alike(cdf, ec_ctx->partition_cdf[ctx], bsize);
1602 aom_write_cdf(w, p == PARTITION_SPLIT, cdf, 2);
1603 }
1604 }
1605
write_modes_sb(AV1_COMP * const cpi,ThreadData * const td,const TileInfo * const tile,aom_writer * const w,const TokenExtra ** tok,const TokenExtra * const tok_end,int mi_row,int mi_col,BLOCK_SIZE bsize)1606 static AOM_INLINE void write_modes_sb(
1607 AV1_COMP *const cpi, ThreadData *const td, const TileInfo *const tile,
1608 aom_writer *const w, const TokenExtra **tok,
1609 const TokenExtra *const tok_end, int mi_row, int mi_col, BLOCK_SIZE bsize) {
1610 const AV1_COMMON *const cm = &cpi->common;
1611 const CommonModeInfoParams *const mi_params = &cm->mi_params;
1612 MACROBLOCKD *const xd = &td->mb.e_mbd;
1613 assert(bsize < BLOCK_SIZES_ALL);
1614 const int hbs = mi_size_wide[bsize] / 2;
1615 const int quarter_step = mi_size_wide[bsize] / 4;
1616 int i;
1617 const PARTITION_TYPE partition = get_partition(cm, mi_row, mi_col, bsize);
1618 const BLOCK_SIZE subsize = get_partition_subsize(bsize, partition);
1619
1620 if (mi_row >= mi_params->mi_rows || mi_col >= mi_params->mi_cols) return;
1621
1622 #if !CONFIG_REALTIME_ONLY
1623 const int num_planes = av1_num_planes(cm);
1624 for (int plane = 0; plane < num_planes; ++plane) {
1625 int rcol0, rcol1, rrow0, rrow1;
1626
1627 // Skip some unnecessary work if loop restoration is disabled
1628 if (cm->rst_info[plane].frame_restoration_type == RESTORE_NONE) continue;
1629
1630 if (av1_loop_restoration_corners_in_sb(cm, plane, mi_row, mi_col, bsize,
1631 &rcol0, &rcol1, &rrow0, &rrow1)) {
1632 const int rstride = cm->rst_info[plane].horz_units;
1633 for (int rrow = rrow0; rrow < rrow1; ++rrow) {
1634 for (int rcol = rcol0; rcol < rcol1; ++rcol) {
1635 const int runit_idx = rcol + rrow * rstride;
1636 loop_restoration_write_sb_coeffs(cm, xd, runit_idx, w, plane,
1637 td->counts);
1638 }
1639 }
1640 }
1641 }
1642 #endif
1643
1644 write_partition(cm, xd, hbs, mi_row, mi_col, partition, bsize, w);
1645 switch (partition) {
1646 case PARTITION_NONE:
1647 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row, mi_col);
1648 break;
1649 case PARTITION_HORZ:
1650 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row, mi_col);
1651 if (mi_row + hbs < mi_params->mi_rows)
1652 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row + hbs, mi_col);
1653 break;
1654 case PARTITION_VERT:
1655 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row, mi_col);
1656 if (mi_col + hbs < mi_params->mi_cols)
1657 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row, mi_col + hbs);
1658 break;
1659 case PARTITION_SPLIT:
1660 write_modes_sb(cpi, td, tile, w, tok, tok_end, mi_row, mi_col, subsize);
1661 write_modes_sb(cpi, td, tile, w, tok, tok_end, mi_row, mi_col + hbs,
1662 subsize);
1663 write_modes_sb(cpi, td, tile, w, tok, tok_end, mi_row + hbs, mi_col,
1664 subsize);
1665 write_modes_sb(cpi, td, tile, w, tok, tok_end, mi_row + hbs, mi_col + hbs,
1666 subsize);
1667 break;
1668 case PARTITION_HORZ_A:
1669 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row, mi_col);
1670 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row, mi_col + hbs);
1671 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row + hbs, mi_col);
1672 break;
1673 case PARTITION_HORZ_B:
1674 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row, mi_col);
1675 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row + hbs, mi_col);
1676 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row + hbs, mi_col + hbs);
1677 break;
1678 case PARTITION_VERT_A:
1679 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row, mi_col);
1680 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row + hbs, mi_col);
1681 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row, mi_col + hbs);
1682 break;
1683 case PARTITION_VERT_B:
1684 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row, mi_col);
1685 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row, mi_col + hbs);
1686 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row + hbs, mi_col + hbs);
1687 break;
1688 case PARTITION_HORZ_4:
1689 for (i = 0; i < 4; ++i) {
1690 int this_mi_row = mi_row + i * quarter_step;
1691 if (i > 0 && this_mi_row >= mi_params->mi_rows) break;
1692
1693 write_modes_b(cpi, td, tile, w, tok, tok_end, this_mi_row, mi_col);
1694 }
1695 break;
1696 case PARTITION_VERT_4:
1697 for (i = 0; i < 4; ++i) {
1698 int this_mi_col = mi_col + i * quarter_step;
1699 if (i > 0 && this_mi_col >= mi_params->mi_cols) break;
1700
1701 write_modes_b(cpi, td, tile, w, tok, tok_end, mi_row, this_mi_col);
1702 }
1703 break;
1704 default: assert(0);
1705 }
1706
1707 // update partition context
1708 update_ext_partition_context(xd, mi_row, mi_col, subsize, bsize, partition);
1709 }
1710
1711 // Populate token pointers appropriately based on token_info.
get_token_pointers(const TokenInfo * token_info,const int tile_row,int tile_col,const int sb_row_in_tile,const TokenExtra ** tok,const TokenExtra ** tok_end)1712 static AOM_INLINE void get_token_pointers(const TokenInfo *token_info,
1713 const int tile_row, int tile_col,
1714 const int sb_row_in_tile,
1715 const TokenExtra **tok,
1716 const TokenExtra **tok_end) {
1717 if (!is_token_info_allocated(token_info)) {
1718 *tok = NULL;
1719 *tok_end = NULL;
1720 return;
1721 }
1722 *tok = token_info->tplist[tile_row][tile_col][sb_row_in_tile].start;
1723 *tok_end =
1724 *tok + token_info->tplist[tile_row][tile_col][sb_row_in_tile].count;
1725 }
1726
write_modes(AV1_COMP * const cpi,ThreadData * const td,const TileInfo * const tile,aom_writer * const w,int tile_row,int tile_col)1727 static AOM_INLINE void write_modes(AV1_COMP *const cpi, ThreadData *const td,
1728 const TileInfo *const tile,
1729 aom_writer *const w, int tile_row,
1730 int tile_col) {
1731 AV1_COMMON *const cm = &cpi->common;
1732 MACROBLOCKD *const xd = &td->mb.e_mbd;
1733 const int mi_row_start = tile->mi_row_start;
1734 const int mi_row_end = tile->mi_row_end;
1735 const int mi_col_start = tile->mi_col_start;
1736 const int mi_col_end = tile->mi_col_end;
1737 const int num_planes = av1_num_planes(cm);
1738
1739 av1_zero_above_context(cm, xd, mi_col_start, mi_col_end, tile->tile_row);
1740 av1_init_above_context(&cm->above_contexts, num_planes, tile->tile_row, xd);
1741
1742 if (cpi->common.delta_q_info.delta_q_present_flag) {
1743 xd->current_base_qindex = cpi->common.quant_params.base_qindex;
1744 if (cpi->common.delta_q_info.delta_lf_present_flag) {
1745 av1_reset_loop_filter_delta(xd, num_planes);
1746 }
1747 }
1748
1749 for (int mi_row = mi_row_start; mi_row < mi_row_end;
1750 mi_row += cm->seq_params->mib_size) {
1751 const int sb_row_in_tile =
1752 (mi_row - tile->mi_row_start) >> cm->seq_params->mib_size_log2;
1753 const TokenInfo *token_info = &cpi->token_info;
1754 const TokenExtra *tok;
1755 const TokenExtra *tok_end;
1756 get_token_pointers(token_info, tile_row, tile_col, sb_row_in_tile, &tok,
1757 &tok_end);
1758
1759 av1_zero_left_context(xd);
1760
1761 for (int mi_col = mi_col_start; mi_col < mi_col_end;
1762 mi_col += cm->seq_params->mib_size) {
1763 td->mb.cb_coef_buff = av1_get_cb_coeff_buffer(cpi, mi_row, mi_col);
1764 write_modes_sb(cpi, td, tile, w, &tok, tok_end, mi_row, mi_col,
1765 cm->seq_params->sb_size);
1766 }
1767 assert(tok == tok_end);
1768 }
1769 }
1770
encode_restoration_mode(AV1_COMMON * cm,struct aom_write_bit_buffer * wb)1771 static AOM_INLINE void encode_restoration_mode(
1772 AV1_COMMON *cm, struct aom_write_bit_buffer *wb) {
1773 assert(!cm->features.all_lossless);
1774 if (!cm->seq_params->enable_restoration) return;
1775 if (cm->features.allow_intrabc) return;
1776 const int num_planes = av1_num_planes(cm);
1777 int all_none = 1, chroma_none = 1;
1778 for (int p = 0; p < num_planes; ++p) {
1779 RestorationInfo *rsi = &cm->rst_info[p];
1780 if (rsi->frame_restoration_type != RESTORE_NONE) {
1781 all_none = 0;
1782 chroma_none &= p == 0;
1783 }
1784 switch (rsi->frame_restoration_type) {
1785 case RESTORE_NONE:
1786 aom_wb_write_bit(wb, 0);
1787 aom_wb_write_bit(wb, 0);
1788 break;
1789 case RESTORE_WIENER:
1790 aom_wb_write_bit(wb, 1);
1791 aom_wb_write_bit(wb, 0);
1792 break;
1793 case RESTORE_SGRPROJ:
1794 aom_wb_write_bit(wb, 1);
1795 aom_wb_write_bit(wb, 1);
1796 break;
1797 case RESTORE_SWITCHABLE:
1798 aom_wb_write_bit(wb, 0);
1799 aom_wb_write_bit(wb, 1);
1800 break;
1801 default: assert(0);
1802 }
1803 }
1804 if (!all_none) {
1805 assert(cm->seq_params->sb_size == BLOCK_64X64 ||
1806 cm->seq_params->sb_size == BLOCK_128X128);
1807 const int sb_size = cm->seq_params->sb_size == BLOCK_128X128 ? 128 : 64;
1808
1809 RestorationInfo *rsi = &cm->rst_info[0];
1810
1811 assert(rsi->restoration_unit_size >= sb_size);
1812 assert(RESTORATION_UNITSIZE_MAX == 256);
1813
1814 if (sb_size == 64) {
1815 aom_wb_write_bit(wb, rsi->restoration_unit_size > 64);
1816 }
1817 if (rsi->restoration_unit_size > 64) {
1818 aom_wb_write_bit(wb, rsi->restoration_unit_size > 128);
1819 }
1820 }
1821
1822 if (num_planes > 1) {
1823 int s =
1824 AOMMIN(cm->seq_params->subsampling_x, cm->seq_params->subsampling_y);
1825 if (s && !chroma_none) {
1826 aom_wb_write_bit(wb, cm->rst_info[1].restoration_unit_size !=
1827 cm->rst_info[0].restoration_unit_size);
1828 assert(cm->rst_info[1].restoration_unit_size ==
1829 cm->rst_info[0].restoration_unit_size ||
1830 cm->rst_info[1].restoration_unit_size ==
1831 (cm->rst_info[0].restoration_unit_size >> s));
1832 assert(cm->rst_info[2].restoration_unit_size ==
1833 cm->rst_info[1].restoration_unit_size);
1834 } else if (!s) {
1835 assert(cm->rst_info[1].restoration_unit_size ==
1836 cm->rst_info[0].restoration_unit_size);
1837 assert(cm->rst_info[2].restoration_unit_size ==
1838 cm->rst_info[1].restoration_unit_size);
1839 }
1840 }
1841 }
1842
1843 #if !CONFIG_REALTIME_ONLY
write_wiener_filter(int wiener_win,const WienerInfo * wiener_info,WienerInfo * ref_wiener_info,aom_writer * wb)1844 static AOM_INLINE void write_wiener_filter(int wiener_win,
1845 const WienerInfo *wiener_info,
1846 WienerInfo *ref_wiener_info,
1847 aom_writer *wb) {
1848 if (wiener_win == WIENER_WIN)
1849 aom_write_primitive_refsubexpfin(
1850 wb, WIENER_FILT_TAP0_MAXV - WIENER_FILT_TAP0_MINV + 1,
1851 WIENER_FILT_TAP0_SUBEXP_K,
1852 ref_wiener_info->vfilter[0] - WIENER_FILT_TAP0_MINV,
1853 wiener_info->vfilter[0] - WIENER_FILT_TAP0_MINV);
1854 else
1855 assert(wiener_info->vfilter[0] == 0 &&
1856 wiener_info->vfilter[WIENER_WIN - 1] == 0);
1857 aom_write_primitive_refsubexpfin(
1858 wb, WIENER_FILT_TAP1_MAXV - WIENER_FILT_TAP1_MINV + 1,
1859 WIENER_FILT_TAP1_SUBEXP_K,
1860 ref_wiener_info->vfilter[1] - WIENER_FILT_TAP1_MINV,
1861 wiener_info->vfilter[1] - WIENER_FILT_TAP1_MINV);
1862 aom_write_primitive_refsubexpfin(
1863 wb, WIENER_FILT_TAP2_MAXV - WIENER_FILT_TAP2_MINV + 1,
1864 WIENER_FILT_TAP2_SUBEXP_K,
1865 ref_wiener_info->vfilter[2] - WIENER_FILT_TAP2_MINV,
1866 wiener_info->vfilter[2] - WIENER_FILT_TAP2_MINV);
1867 if (wiener_win == WIENER_WIN)
1868 aom_write_primitive_refsubexpfin(
1869 wb, WIENER_FILT_TAP0_MAXV - WIENER_FILT_TAP0_MINV + 1,
1870 WIENER_FILT_TAP0_SUBEXP_K,
1871 ref_wiener_info->hfilter[0] - WIENER_FILT_TAP0_MINV,
1872 wiener_info->hfilter[0] - WIENER_FILT_TAP0_MINV);
1873 else
1874 assert(wiener_info->hfilter[0] == 0 &&
1875 wiener_info->hfilter[WIENER_WIN - 1] == 0);
1876 aom_write_primitive_refsubexpfin(
1877 wb, WIENER_FILT_TAP1_MAXV - WIENER_FILT_TAP1_MINV + 1,
1878 WIENER_FILT_TAP1_SUBEXP_K,
1879 ref_wiener_info->hfilter[1] - WIENER_FILT_TAP1_MINV,
1880 wiener_info->hfilter[1] - WIENER_FILT_TAP1_MINV);
1881 aom_write_primitive_refsubexpfin(
1882 wb, WIENER_FILT_TAP2_MAXV - WIENER_FILT_TAP2_MINV + 1,
1883 WIENER_FILT_TAP2_SUBEXP_K,
1884 ref_wiener_info->hfilter[2] - WIENER_FILT_TAP2_MINV,
1885 wiener_info->hfilter[2] - WIENER_FILT_TAP2_MINV);
1886 memcpy(ref_wiener_info, wiener_info, sizeof(*wiener_info));
1887 }
1888
write_sgrproj_filter(const SgrprojInfo * sgrproj_info,SgrprojInfo * ref_sgrproj_info,aom_writer * wb)1889 static AOM_INLINE void write_sgrproj_filter(const SgrprojInfo *sgrproj_info,
1890 SgrprojInfo *ref_sgrproj_info,
1891 aom_writer *wb) {
1892 aom_write_literal(wb, sgrproj_info->ep, SGRPROJ_PARAMS_BITS);
1893 const sgr_params_type *params = &av1_sgr_params[sgrproj_info->ep];
1894
1895 if (params->r[0] == 0) {
1896 assert(sgrproj_info->xqd[0] == 0);
1897 aom_write_primitive_refsubexpfin(
1898 wb, SGRPROJ_PRJ_MAX1 - SGRPROJ_PRJ_MIN1 + 1, SGRPROJ_PRJ_SUBEXP_K,
1899 ref_sgrproj_info->xqd[1] - SGRPROJ_PRJ_MIN1,
1900 sgrproj_info->xqd[1] - SGRPROJ_PRJ_MIN1);
1901 } else if (params->r[1] == 0) {
1902 aom_write_primitive_refsubexpfin(
1903 wb, SGRPROJ_PRJ_MAX0 - SGRPROJ_PRJ_MIN0 + 1, SGRPROJ_PRJ_SUBEXP_K,
1904 ref_sgrproj_info->xqd[0] - SGRPROJ_PRJ_MIN0,
1905 sgrproj_info->xqd[0] - SGRPROJ_PRJ_MIN0);
1906 } else {
1907 aom_write_primitive_refsubexpfin(
1908 wb, SGRPROJ_PRJ_MAX0 - SGRPROJ_PRJ_MIN0 + 1, SGRPROJ_PRJ_SUBEXP_K,
1909 ref_sgrproj_info->xqd[0] - SGRPROJ_PRJ_MIN0,
1910 sgrproj_info->xqd[0] - SGRPROJ_PRJ_MIN0);
1911 aom_write_primitive_refsubexpfin(
1912 wb, SGRPROJ_PRJ_MAX1 - SGRPROJ_PRJ_MIN1 + 1, SGRPROJ_PRJ_SUBEXP_K,
1913 ref_sgrproj_info->xqd[1] - SGRPROJ_PRJ_MIN1,
1914 sgrproj_info->xqd[1] - SGRPROJ_PRJ_MIN1);
1915 }
1916
1917 memcpy(ref_sgrproj_info, sgrproj_info, sizeof(*sgrproj_info));
1918 }
1919
loop_restoration_write_sb_coeffs(const AV1_COMMON * const cm,MACROBLOCKD * xd,int runit_idx,aom_writer * const w,int plane,FRAME_COUNTS * counts)1920 static AOM_INLINE void loop_restoration_write_sb_coeffs(
1921 const AV1_COMMON *const cm, MACROBLOCKD *xd, int runit_idx,
1922 aom_writer *const w, int plane, FRAME_COUNTS *counts) {
1923 const RestorationUnitInfo *rui = &cm->rst_info[plane].unit_info[runit_idx];
1924 const RestorationInfo *rsi = cm->rst_info + plane;
1925 RestorationType frame_rtype = rsi->frame_restoration_type;
1926 assert(frame_rtype != RESTORE_NONE);
1927
1928 (void)counts;
1929 assert(!cm->features.all_lossless);
1930
1931 const int wiener_win = (plane > 0) ? WIENER_WIN_CHROMA : WIENER_WIN;
1932 WienerInfo *ref_wiener_info = &xd->wiener_info[plane];
1933 SgrprojInfo *ref_sgrproj_info = &xd->sgrproj_info[plane];
1934 RestorationType unit_rtype = rui->restoration_type;
1935
1936 if (frame_rtype == RESTORE_SWITCHABLE) {
1937 aom_write_symbol(w, unit_rtype, xd->tile_ctx->switchable_restore_cdf,
1938 RESTORE_SWITCHABLE_TYPES);
1939 #if CONFIG_ENTROPY_STATS
1940 ++counts->switchable_restore[unit_rtype];
1941 #endif
1942 switch (unit_rtype) {
1943 case RESTORE_WIENER:
1944 #if DEBUG_LR_COSTING
1945 assert(!memcmp(
1946 ref_wiener_info,
1947 &lr_ref_params[RESTORE_SWITCHABLE][plane][runit_idx].wiener_info,
1948 sizeof(*ref_wiener_info)));
1949 #endif
1950 write_wiener_filter(wiener_win, &rui->wiener_info, ref_wiener_info, w);
1951 break;
1952 case RESTORE_SGRPROJ:
1953 #if DEBUG_LR_COSTING
1954 assert(!memcmp(&ref_sgrproj_info->xqd,
1955 &lr_ref_params[RESTORE_SWITCHABLE][plane][runit_idx]
1956 .sgrproj_info.xqd,
1957 sizeof(ref_sgrproj_info->xqd)));
1958 #endif
1959 write_sgrproj_filter(&rui->sgrproj_info, ref_sgrproj_info, w);
1960 break;
1961 default: assert(unit_rtype == RESTORE_NONE); break;
1962 }
1963 } else if (frame_rtype == RESTORE_WIENER) {
1964 aom_write_symbol(w, unit_rtype != RESTORE_NONE,
1965 xd->tile_ctx->wiener_restore_cdf, 2);
1966 #if CONFIG_ENTROPY_STATS
1967 ++counts->wiener_restore[unit_rtype != RESTORE_NONE];
1968 #endif
1969 if (unit_rtype != RESTORE_NONE) {
1970 #if DEBUG_LR_COSTING
1971 assert(
1972 !memcmp(ref_wiener_info,
1973 &lr_ref_params[RESTORE_WIENER][plane][runit_idx].wiener_info,
1974 sizeof(*ref_wiener_info)));
1975 #endif
1976 write_wiener_filter(wiener_win, &rui->wiener_info, ref_wiener_info, w);
1977 }
1978 } else if (frame_rtype == RESTORE_SGRPROJ) {
1979 aom_write_symbol(w, unit_rtype != RESTORE_NONE,
1980 xd->tile_ctx->sgrproj_restore_cdf, 2);
1981 #if CONFIG_ENTROPY_STATS
1982 ++counts->sgrproj_restore[unit_rtype != RESTORE_NONE];
1983 #endif
1984 if (unit_rtype != RESTORE_NONE) {
1985 #if DEBUG_LR_COSTING
1986 assert(!memcmp(
1987 &ref_sgrproj_info->xqd,
1988 &lr_ref_params[RESTORE_SGRPROJ][plane][runit_idx].sgrproj_info.xqd,
1989 sizeof(ref_sgrproj_info->xqd)));
1990 #endif
1991 write_sgrproj_filter(&rui->sgrproj_info, ref_sgrproj_info, w);
1992 }
1993 }
1994 }
1995 #endif // !CONFIG_REALTIME_ONLY
1996
1997 // Only write out the ref delta section if any of the elements
1998 // will signal a delta.
is_mode_ref_delta_meaningful(AV1_COMMON * cm)1999 static bool is_mode_ref_delta_meaningful(AV1_COMMON *cm) {
2000 struct loopfilter *lf = &cm->lf;
2001 if (!lf->mode_ref_delta_update) {
2002 return 0;
2003 }
2004 const RefCntBuffer *buf = get_primary_ref_frame_buf(cm);
2005 int8_t last_ref_deltas[REF_FRAMES];
2006 int8_t last_mode_deltas[MAX_MODE_LF_DELTAS];
2007 if (buf == NULL) {
2008 av1_set_default_ref_deltas(last_ref_deltas);
2009 av1_set_default_mode_deltas(last_mode_deltas);
2010 } else {
2011 memcpy(last_ref_deltas, buf->ref_deltas, REF_FRAMES);
2012 memcpy(last_mode_deltas, buf->mode_deltas, MAX_MODE_LF_DELTAS);
2013 }
2014 for (int i = 0; i < REF_FRAMES; i++) {
2015 if (lf->ref_deltas[i] != last_ref_deltas[i]) {
2016 return true;
2017 }
2018 }
2019 for (int i = 0; i < MAX_MODE_LF_DELTAS; i++) {
2020 if (lf->mode_deltas[i] != last_mode_deltas[i]) {
2021 return true;
2022 }
2023 }
2024 return false;
2025 }
2026
encode_loopfilter(AV1_COMMON * cm,struct aom_write_bit_buffer * wb)2027 static AOM_INLINE void encode_loopfilter(AV1_COMMON *cm,
2028 struct aom_write_bit_buffer *wb) {
2029 assert(!cm->features.coded_lossless);
2030 if (cm->features.allow_intrabc) return;
2031 const int num_planes = av1_num_planes(cm);
2032 struct loopfilter *lf = &cm->lf;
2033
2034 // Encode the loop filter level and type
2035 aom_wb_write_literal(wb, lf->filter_level[0], 6);
2036 aom_wb_write_literal(wb, lf->filter_level[1], 6);
2037 if (num_planes > 1) {
2038 if (lf->filter_level[0] || lf->filter_level[1]) {
2039 aom_wb_write_literal(wb, lf->filter_level_u, 6);
2040 aom_wb_write_literal(wb, lf->filter_level_v, 6);
2041 }
2042 }
2043 aom_wb_write_literal(wb, lf->sharpness_level, 3);
2044
2045 aom_wb_write_bit(wb, lf->mode_ref_delta_enabled);
2046
2047 // Write out loop filter deltas applied at the MB level based on mode or
2048 // ref frame (if they are enabled), only if there is information to write.
2049 int meaningful = is_mode_ref_delta_meaningful(cm);
2050 aom_wb_write_bit(wb, meaningful);
2051 if (!meaningful) {
2052 return;
2053 }
2054
2055 const RefCntBuffer *buf = get_primary_ref_frame_buf(cm);
2056 int8_t last_ref_deltas[REF_FRAMES];
2057 int8_t last_mode_deltas[MAX_MODE_LF_DELTAS];
2058 if (buf == NULL) {
2059 av1_set_default_ref_deltas(last_ref_deltas);
2060 av1_set_default_mode_deltas(last_mode_deltas);
2061 } else {
2062 memcpy(last_ref_deltas, buf->ref_deltas, REF_FRAMES);
2063 memcpy(last_mode_deltas, buf->mode_deltas, MAX_MODE_LF_DELTAS);
2064 }
2065 for (int i = 0; i < REF_FRAMES; i++) {
2066 const int delta = lf->ref_deltas[i];
2067 const int changed = delta != last_ref_deltas[i];
2068 aom_wb_write_bit(wb, changed);
2069 if (changed) aom_wb_write_inv_signed_literal(wb, delta, 6);
2070 }
2071 for (int i = 0; i < MAX_MODE_LF_DELTAS; i++) {
2072 const int delta = lf->mode_deltas[i];
2073 const int changed = delta != last_mode_deltas[i];
2074 aom_wb_write_bit(wb, changed);
2075 if (changed) aom_wb_write_inv_signed_literal(wb, delta, 6);
2076 }
2077 }
2078
encode_cdef(const AV1_COMMON * cm,struct aom_write_bit_buffer * wb)2079 static AOM_INLINE void encode_cdef(const AV1_COMMON *cm,
2080 struct aom_write_bit_buffer *wb) {
2081 assert(!cm->features.coded_lossless);
2082 if (!cm->seq_params->enable_cdef) return;
2083 if (cm->features.allow_intrabc) return;
2084 const int num_planes = av1_num_planes(cm);
2085 int i;
2086 aom_wb_write_literal(wb, cm->cdef_info.cdef_damping - 3, 2);
2087 aom_wb_write_literal(wb, cm->cdef_info.cdef_bits, 2);
2088 for (i = 0; i < cm->cdef_info.nb_cdef_strengths; i++) {
2089 aom_wb_write_literal(wb, cm->cdef_info.cdef_strengths[i],
2090 CDEF_STRENGTH_BITS);
2091 if (num_planes > 1)
2092 aom_wb_write_literal(wb, cm->cdef_info.cdef_uv_strengths[i],
2093 CDEF_STRENGTH_BITS);
2094 }
2095 }
2096
write_delta_q(struct aom_write_bit_buffer * wb,int delta_q)2097 static AOM_INLINE void write_delta_q(struct aom_write_bit_buffer *wb,
2098 int delta_q) {
2099 if (delta_q != 0) {
2100 aom_wb_write_bit(wb, 1);
2101 aom_wb_write_inv_signed_literal(wb, delta_q, 6);
2102 } else {
2103 aom_wb_write_bit(wb, 0);
2104 }
2105 }
2106
encode_quantization(const CommonQuantParams * const quant_params,int num_planes,bool separate_uv_delta_q,struct aom_write_bit_buffer * wb)2107 static AOM_INLINE void encode_quantization(
2108 const CommonQuantParams *const quant_params, int num_planes,
2109 bool separate_uv_delta_q, struct aom_write_bit_buffer *wb) {
2110 aom_wb_write_literal(wb, quant_params->base_qindex, QINDEX_BITS);
2111 write_delta_q(wb, quant_params->y_dc_delta_q);
2112 if (num_planes > 1) {
2113 int diff_uv_delta =
2114 (quant_params->u_dc_delta_q != quant_params->v_dc_delta_q) ||
2115 (quant_params->u_ac_delta_q != quant_params->v_ac_delta_q);
2116 if (separate_uv_delta_q) aom_wb_write_bit(wb, diff_uv_delta);
2117 write_delta_q(wb, quant_params->u_dc_delta_q);
2118 write_delta_q(wb, quant_params->u_ac_delta_q);
2119 if (diff_uv_delta) {
2120 write_delta_q(wb, quant_params->v_dc_delta_q);
2121 write_delta_q(wb, quant_params->v_ac_delta_q);
2122 }
2123 }
2124 aom_wb_write_bit(wb, quant_params->using_qmatrix);
2125 if (quant_params->using_qmatrix) {
2126 aom_wb_write_literal(wb, quant_params->qmatrix_level_y, QM_LEVEL_BITS);
2127 aom_wb_write_literal(wb, quant_params->qmatrix_level_u, QM_LEVEL_BITS);
2128 if (!separate_uv_delta_q)
2129 assert(quant_params->qmatrix_level_u == quant_params->qmatrix_level_v);
2130 else
2131 aom_wb_write_literal(wb, quant_params->qmatrix_level_v, QM_LEVEL_BITS);
2132 }
2133 }
2134
encode_segmentation(AV1_COMMON * cm,struct aom_write_bit_buffer * wb)2135 static AOM_INLINE void encode_segmentation(AV1_COMMON *cm,
2136 struct aom_write_bit_buffer *wb) {
2137 int i, j;
2138 struct segmentation *seg = &cm->seg;
2139
2140 aom_wb_write_bit(wb, seg->enabled);
2141 if (!seg->enabled) return;
2142
2143 // Write update flags
2144 if (cm->features.primary_ref_frame != PRIMARY_REF_NONE) {
2145 aom_wb_write_bit(wb, seg->update_map);
2146 if (seg->update_map) aom_wb_write_bit(wb, seg->temporal_update);
2147 aom_wb_write_bit(wb, seg->update_data);
2148 }
2149
2150 // Segmentation data
2151 if (seg->update_data) {
2152 for (i = 0; i < MAX_SEGMENTS; i++) {
2153 for (j = 0; j < SEG_LVL_MAX; j++) {
2154 const int active = segfeature_active(seg, i, j);
2155 aom_wb_write_bit(wb, active);
2156 if (active) {
2157 const int data_max = av1_seg_feature_data_max(j);
2158 const int data_min = -data_max;
2159 const int ubits = get_unsigned_bits(data_max);
2160 const int data = clamp(get_segdata(seg, i, j), data_min, data_max);
2161
2162 if (av1_is_segfeature_signed(j)) {
2163 aom_wb_write_inv_signed_literal(wb, data, ubits);
2164 } else {
2165 aom_wb_write_literal(wb, data, ubits);
2166 }
2167 }
2168 }
2169 }
2170 }
2171 }
2172
write_frame_interp_filter(InterpFilter filter,struct aom_write_bit_buffer * wb)2173 static AOM_INLINE void write_frame_interp_filter(
2174 InterpFilter filter, struct aom_write_bit_buffer *wb) {
2175 aom_wb_write_bit(wb, filter == SWITCHABLE);
2176 if (filter != SWITCHABLE)
2177 aom_wb_write_literal(wb, filter, LOG_SWITCHABLE_FILTERS);
2178 }
2179
2180 // Same function as write_uniform but writing to uncompresses header wb
wb_write_uniform(struct aom_write_bit_buffer * wb,int n,int v)2181 static AOM_INLINE void wb_write_uniform(struct aom_write_bit_buffer *wb, int n,
2182 int v) {
2183 const int l = get_unsigned_bits(n);
2184 const int m = (1 << l) - n;
2185 if (l == 0) return;
2186 if (v < m) {
2187 aom_wb_write_literal(wb, v, l - 1);
2188 } else {
2189 aom_wb_write_literal(wb, m + ((v - m) >> 1), l - 1);
2190 aom_wb_write_literal(wb, (v - m) & 1, 1);
2191 }
2192 }
2193
write_tile_info_max_tile(const AV1_COMMON * const cm,struct aom_write_bit_buffer * wb)2194 static AOM_INLINE void write_tile_info_max_tile(
2195 const AV1_COMMON *const cm, struct aom_write_bit_buffer *wb) {
2196 int width_sb =
2197 CEIL_POWER_OF_TWO(cm->mi_params.mi_cols, cm->seq_params->mib_size_log2);
2198 int height_sb =
2199 CEIL_POWER_OF_TWO(cm->mi_params.mi_rows, cm->seq_params->mib_size_log2);
2200 int size_sb, i;
2201 const CommonTileParams *const tiles = &cm->tiles;
2202
2203 aom_wb_write_bit(wb, tiles->uniform_spacing);
2204
2205 if (tiles->uniform_spacing) {
2206 int ones = tiles->log2_cols - tiles->min_log2_cols;
2207 while (ones--) {
2208 aom_wb_write_bit(wb, 1);
2209 }
2210 if (tiles->log2_cols < tiles->max_log2_cols) {
2211 aom_wb_write_bit(wb, 0);
2212 }
2213
2214 // rows
2215 ones = tiles->log2_rows - tiles->min_log2_rows;
2216 while (ones--) {
2217 aom_wb_write_bit(wb, 1);
2218 }
2219 if (tiles->log2_rows < tiles->max_log2_rows) {
2220 aom_wb_write_bit(wb, 0);
2221 }
2222 } else {
2223 // Explicit tiles with configurable tile widths and heights
2224 // columns
2225 for (i = 0; i < tiles->cols; i++) {
2226 size_sb = tiles->col_start_sb[i + 1] - tiles->col_start_sb[i];
2227 wb_write_uniform(wb, AOMMIN(width_sb, tiles->max_width_sb), size_sb - 1);
2228 width_sb -= size_sb;
2229 }
2230 assert(width_sb == 0);
2231
2232 // rows
2233 for (i = 0; i < tiles->rows; i++) {
2234 size_sb = tiles->row_start_sb[i + 1] - tiles->row_start_sb[i];
2235 wb_write_uniform(wb, AOMMIN(height_sb, tiles->max_height_sb),
2236 size_sb - 1);
2237 height_sb -= size_sb;
2238 }
2239 assert(height_sb == 0);
2240 }
2241 }
2242
write_tile_info(const AV1_COMMON * const cm,struct aom_write_bit_buffer * saved_wb,struct aom_write_bit_buffer * wb)2243 static AOM_INLINE void write_tile_info(const AV1_COMMON *const cm,
2244 struct aom_write_bit_buffer *saved_wb,
2245 struct aom_write_bit_buffer *wb) {
2246 write_tile_info_max_tile(cm, wb);
2247
2248 *saved_wb = *wb;
2249 if (cm->tiles.rows * cm->tiles.cols > 1) {
2250 // tile id used for cdf update
2251 aom_wb_write_literal(wb, 0, cm->tiles.log2_cols + cm->tiles.log2_rows);
2252 // Number of bytes in tile size - 1
2253 aom_wb_write_literal(wb, 3, 2);
2254 }
2255 }
2256
write_ext_tile_info(const AV1_COMMON * const cm,struct aom_write_bit_buffer * saved_wb,struct aom_write_bit_buffer * wb)2257 static AOM_INLINE void write_ext_tile_info(
2258 const AV1_COMMON *const cm, struct aom_write_bit_buffer *saved_wb,
2259 struct aom_write_bit_buffer *wb) {
2260 // This information is stored as a separate byte.
2261 int mod = wb->bit_offset % CHAR_BIT;
2262 if (mod > 0) aom_wb_write_literal(wb, 0, CHAR_BIT - mod);
2263 assert(aom_wb_is_byte_aligned(wb));
2264
2265 *saved_wb = *wb;
2266 if (cm->tiles.rows * cm->tiles.cols > 1) {
2267 // Note that the last item in the uncompressed header is the data
2268 // describing tile configuration.
2269 // Number of bytes in tile column size - 1
2270 aom_wb_write_literal(wb, 0, 2);
2271 // Number of bytes in tile size - 1
2272 aom_wb_write_literal(wb, 0, 2);
2273 }
2274 }
2275
find_identical_tile(const int tile_row,const int tile_col,TileBufferEnc (* const tile_buffers)[MAX_TILE_COLS])2276 static INLINE int find_identical_tile(
2277 const int tile_row, const int tile_col,
2278 TileBufferEnc (*const tile_buffers)[MAX_TILE_COLS]) {
2279 const MV32 candidate_offset[1] = { { 1, 0 } };
2280 const uint8_t *const cur_tile_data =
2281 tile_buffers[tile_row][tile_col].data + 4;
2282 const size_t cur_tile_size = tile_buffers[tile_row][tile_col].size;
2283
2284 int i;
2285
2286 if (tile_row == 0) return 0;
2287
2288 // (TODO: yunqingwang) For now, only above tile is checked and used.
2289 // More candidates such as left tile can be added later.
2290 for (i = 0; i < 1; i++) {
2291 int row_offset = candidate_offset[0].row;
2292 int col_offset = candidate_offset[0].col;
2293 int row = tile_row - row_offset;
2294 int col = tile_col - col_offset;
2295 const uint8_t *tile_data;
2296 TileBufferEnc *candidate;
2297
2298 if (row < 0 || col < 0) continue;
2299
2300 const uint32_t tile_hdr = mem_get_le32(tile_buffers[row][col].data);
2301
2302 // Read out tile-copy-mode bit:
2303 if ((tile_hdr >> 31) == 1) {
2304 // The candidate is a copy tile itself: the offset is stored in bits
2305 // 30 through 24 inclusive.
2306 row_offset += (tile_hdr >> 24) & 0x7f;
2307 row = tile_row - row_offset;
2308 }
2309
2310 candidate = &tile_buffers[row][col];
2311
2312 if (row_offset >= 128 || candidate->size != cur_tile_size) continue;
2313
2314 tile_data = candidate->data + 4;
2315
2316 if (memcmp(tile_data, cur_tile_data, cur_tile_size) != 0) continue;
2317
2318 // Identical tile found
2319 assert(row_offset > 0);
2320 return row_offset;
2321 }
2322
2323 // No identical tile found
2324 return 0;
2325 }
2326
write_render_size(const AV1_COMMON * cm,struct aom_write_bit_buffer * wb)2327 static AOM_INLINE void write_render_size(const AV1_COMMON *cm,
2328 struct aom_write_bit_buffer *wb) {
2329 const int scaling_active = av1_resize_scaled(cm);
2330 aom_wb_write_bit(wb, scaling_active);
2331 if (scaling_active) {
2332 aom_wb_write_literal(wb, cm->render_width - 1, 16);
2333 aom_wb_write_literal(wb, cm->render_height - 1, 16);
2334 }
2335 }
2336
write_superres_scale(const AV1_COMMON * const cm,struct aom_write_bit_buffer * wb)2337 static AOM_INLINE void write_superres_scale(const AV1_COMMON *const cm,
2338 struct aom_write_bit_buffer *wb) {
2339 const SequenceHeader *const seq_params = cm->seq_params;
2340 if (!seq_params->enable_superres) {
2341 assert(cm->superres_scale_denominator == SCALE_NUMERATOR);
2342 return;
2343 }
2344
2345 // First bit is whether to to scale or not
2346 if (cm->superres_scale_denominator == SCALE_NUMERATOR) {
2347 aom_wb_write_bit(wb, 0); // no scaling
2348 } else {
2349 aom_wb_write_bit(wb, 1); // scaling, write scale factor
2350 assert(cm->superres_scale_denominator >= SUPERRES_SCALE_DENOMINATOR_MIN);
2351 assert(cm->superres_scale_denominator <
2352 SUPERRES_SCALE_DENOMINATOR_MIN + (1 << SUPERRES_SCALE_BITS));
2353 aom_wb_write_literal(
2354 wb, cm->superres_scale_denominator - SUPERRES_SCALE_DENOMINATOR_MIN,
2355 SUPERRES_SCALE_BITS);
2356 }
2357 }
2358
write_frame_size(const AV1_COMMON * cm,int frame_size_override,struct aom_write_bit_buffer * wb)2359 static AOM_INLINE void write_frame_size(const AV1_COMMON *cm,
2360 int frame_size_override,
2361 struct aom_write_bit_buffer *wb) {
2362 const int coded_width = cm->superres_upscaled_width - 1;
2363 const int coded_height = cm->superres_upscaled_height - 1;
2364
2365 if (frame_size_override) {
2366 const SequenceHeader *seq_params = cm->seq_params;
2367 int num_bits_width = seq_params->num_bits_width;
2368 int num_bits_height = seq_params->num_bits_height;
2369 aom_wb_write_literal(wb, coded_width, num_bits_width);
2370 aom_wb_write_literal(wb, coded_height, num_bits_height);
2371 }
2372
2373 write_superres_scale(cm, wb);
2374 write_render_size(cm, wb);
2375 }
2376
write_frame_size_with_refs(const AV1_COMMON * const cm,struct aom_write_bit_buffer * wb)2377 static AOM_INLINE void write_frame_size_with_refs(
2378 const AV1_COMMON *const cm, struct aom_write_bit_buffer *wb) {
2379 int found = 0;
2380
2381 MV_REFERENCE_FRAME ref_frame;
2382 for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
2383 const YV12_BUFFER_CONFIG *cfg = get_ref_frame_yv12_buf(cm, ref_frame);
2384
2385 if (cfg != NULL) {
2386 found = cm->superres_upscaled_width == cfg->y_crop_width &&
2387 cm->superres_upscaled_height == cfg->y_crop_height;
2388 found &= cm->render_width == cfg->render_width &&
2389 cm->render_height == cfg->render_height;
2390 }
2391 aom_wb_write_bit(wb, found);
2392 if (found) {
2393 write_superres_scale(cm, wb);
2394 break;
2395 }
2396 }
2397
2398 if (!found) {
2399 int frame_size_override = 1; // Always equal to 1 in this function
2400 write_frame_size(cm, frame_size_override, wb);
2401 }
2402 }
2403
write_profile(BITSTREAM_PROFILE profile,struct aom_write_bit_buffer * wb)2404 static AOM_INLINE void write_profile(BITSTREAM_PROFILE profile,
2405 struct aom_write_bit_buffer *wb) {
2406 assert(profile >= PROFILE_0 && profile < MAX_PROFILES);
2407 aom_wb_write_literal(wb, profile, PROFILE_BITS);
2408 }
2409
write_bitdepth(const SequenceHeader * const seq_params,struct aom_write_bit_buffer * wb)2410 static AOM_INLINE void write_bitdepth(const SequenceHeader *const seq_params,
2411 struct aom_write_bit_buffer *wb) {
2412 // Profile 0/1: [0] for 8 bit, [1] 10-bit
2413 // Profile 2: [0] for 8 bit, [10] 10-bit, [11] - 12-bit
2414 aom_wb_write_bit(wb, seq_params->bit_depth == AOM_BITS_8 ? 0 : 1);
2415 if (seq_params->profile == PROFILE_2 && seq_params->bit_depth != AOM_BITS_8) {
2416 aom_wb_write_bit(wb, seq_params->bit_depth == AOM_BITS_10 ? 0 : 1);
2417 }
2418 }
2419
write_color_config(const SequenceHeader * const seq_params,struct aom_write_bit_buffer * wb)2420 static AOM_INLINE void write_color_config(
2421 const SequenceHeader *const seq_params, struct aom_write_bit_buffer *wb) {
2422 write_bitdepth(seq_params, wb);
2423 const int is_monochrome = seq_params->monochrome;
2424 // monochrome bit
2425 if (seq_params->profile != PROFILE_1)
2426 aom_wb_write_bit(wb, is_monochrome);
2427 else
2428 assert(!is_monochrome);
2429 if (seq_params->color_primaries == AOM_CICP_CP_UNSPECIFIED &&
2430 seq_params->transfer_characteristics == AOM_CICP_TC_UNSPECIFIED &&
2431 seq_params->matrix_coefficients == AOM_CICP_MC_UNSPECIFIED) {
2432 aom_wb_write_bit(wb, 0); // No color description present
2433 } else {
2434 aom_wb_write_bit(wb, 1); // Color description present
2435 aom_wb_write_literal(wb, seq_params->color_primaries, 8);
2436 aom_wb_write_literal(wb, seq_params->transfer_characteristics, 8);
2437 aom_wb_write_literal(wb, seq_params->matrix_coefficients, 8);
2438 }
2439 if (is_monochrome) {
2440 // 0: [16, 235] (i.e. xvYCC), 1: [0, 255]
2441 aom_wb_write_bit(wb, seq_params->color_range);
2442 return;
2443 }
2444 if (seq_params->color_primaries == AOM_CICP_CP_BT_709 &&
2445 seq_params->transfer_characteristics == AOM_CICP_TC_SRGB &&
2446 seq_params->matrix_coefficients == AOM_CICP_MC_IDENTITY) {
2447 assert(seq_params->subsampling_x == 0 && seq_params->subsampling_y == 0);
2448 assert(seq_params->profile == PROFILE_1 ||
2449 (seq_params->profile == PROFILE_2 &&
2450 seq_params->bit_depth == AOM_BITS_12));
2451 } else {
2452 // 0: [16, 235] (i.e. xvYCC), 1: [0, 255]
2453 aom_wb_write_bit(wb, seq_params->color_range);
2454 if (seq_params->profile == PROFILE_0) {
2455 // 420 only
2456 assert(seq_params->subsampling_x == 1 && seq_params->subsampling_y == 1);
2457 } else if (seq_params->profile == PROFILE_1) {
2458 // 444 only
2459 assert(seq_params->subsampling_x == 0 && seq_params->subsampling_y == 0);
2460 } else if (seq_params->profile == PROFILE_2) {
2461 if (seq_params->bit_depth == AOM_BITS_12) {
2462 // 420, 444 or 422
2463 aom_wb_write_bit(wb, seq_params->subsampling_x);
2464 if (seq_params->subsampling_x == 0) {
2465 assert(seq_params->subsampling_y == 0 &&
2466 "4:4:0 subsampling not allowed in AV1");
2467 } else {
2468 aom_wb_write_bit(wb, seq_params->subsampling_y);
2469 }
2470 } else {
2471 // 422 only
2472 assert(seq_params->subsampling_x == 1 &&
2473 seq_params->subsampling_y == 0);
2474 }
2475 }
2476 if (seq_params->matrix_coefficients == AOM_CICP_MC_IDENTITY) {
2477 assert(seq_params->subsampling_x == 0 && seq_params->subsampling_y == 0);
2478 }
2479 if (seq_params->subsampling_x == 1 && seq_params->subsampling_y == 1) {
2480 aom_wb_write_literal(wb, seq_params->chroma_sample_position, 2);
2481 }
2482 }
2483 aom_wb_write_bit(wb, seq_params->separate_uv_delta_q);
2484 }
2485
write_timing_info_header(const aom_timing_info_t * const timing_info,struct aom_write_bit_buffer * wb)2486 static AOM_INLINE void write_timing_info_header(
2487 const aom_timing_info_t *const timing_info,
2488 struct aom_write_bit_buffer *wb) {
2489 aom_wb_write_unsigned_literal(wb, timing_info->num_units_in_display_tick, 32);
2490 aom_wb_write_unsigned_literal(wb, timing_info->time_scale, 32);
2491 aom_wb_write_bit(wb, timing_info->equal_picture_interval);
2492 if (timing_info->equal_picture_interval) {
2493 aom_wb_write_uvlc(wb, timing_info->num_ticks_per_picture - 1);
2494 }
2495 }
2496
write_decoder_model_info(const aom_dec_model_info_t * const decoder_model_info,struct aom_write_bit_buffer * wb)2497 static AOM_INLINE void write_decoder_model_info(
2498 const aom_dec_model_info_t *const decoder_model_info,
2499 struct aom_write_bit_buffer *wb) {
2500 aom_wb_write_literal(
2501 wb, decoder_model_info->encoder_decoder_buffer_delay_length - 1, 5);
2502 aom_wb_write_unsigned_literal(
2503 wb, decoder_model_info->num_units_in_decoding_tick, 32);
2504 aom_wb_write_literal(wb, decoder_model_info->buffer_removal_time_length - 1,
2505 5);
2506 aom_wb_write_literal(
2507 wb, decoder_model_info->frame_presentation_time_length - 1, 5);
2508 }
2509
write_dec_model_op_parameters(const aom_dec_model_op_parameters_t * op_params,int buffer_delay_length,struct aom_write_bit_buffer * wb)2510 static AOM_INLINE void write_dec_model_op_parameters(
2511 const aom_dec_model_op_parameters_t *op_params, int buffer_delay_length,
2512 struct aom_write_bit_buffer *wb) {
2513 aom_wb_write_unsigned_literal(wb, op_params->decoder_buffer_delay,
2514 buffer_delay_length);
2515 aom_wb_write_unsigned_literal(wb, op_params->encoder_buffer_delay,
2516 buffer_delay_length);
2517 aom_wb_write_bit(wb, op_params->low_delay_mode_flag);
2518 }
2519
write_tu_pts_info(AV1_COMMON * const cm,struct aom_write_bit_buffer * wb)2520 static AOM_INLINE void write_tu_pts_info(AV1_COMMON *const cm,
2521 struct aom_write_bit_buffer *wb) {
2522 aom_wb_write_unsigned_literal(
2523 wb, cm->frame_presentation_time,
2524 cm->seq_params->decoder_model_info.frame_presentation_time_length);
2525 }
2526
write_film_grain_params(const AV1_COMP * const cpi,struct aom_write_bit_buffer * wb)2527 static AOM_INLINE void write_film_grain_params(
2528 const AV1_COMP *const cpi, struct aom_write_bit_buffer *wb) {
2529 const AV1_COMMON *const cm = &cpi->common;
2530 const aom_film_grain_t *const pars = &cm->cur_frame->film_grain_params;
2531 aom_wb_write_bit(wb, pars->apply_grain);
2532 if (!pars->apply_grain) return;
2533
2534 aom_wb_write_literal(wb, pars->random_seed, 16);
2535
2536 if (cm->current_frame.frame_type == INTER_FRAME)
2537 aom_wb_write_bit(wb, pars->update_parameters);
2538
2539 if (!pars->update_parameters) {
2540 int ref_frame, ref_idx;
2541 for (ref_frame = LAST_FRAME; ref_frame < REF_FRAMES; ref_frame++) {
2542 ref_idx = get_ref_frame_map_idx(cm, ref_frame);
2543 assert(ref_idx != INVALID_IDX);
2544 const RefCntBuffer *const buf = cm->ref_frame_map[ref_idx];
2545 if (buf->film_grain_params_present &&
2546 aom_check_grain_params_equiv(pars, &buf->film_grain_params)) {
2547 break;
2548 }
2549 }
2550 assert(ref_frame < REF_FRAMES);
2551 aom_wb_write_literal(wb, ref_idx, 3);
2552 return;
2553 }
2554
2555 // Scaling functions parameters
2556 aom_wb_write_literal(wb, pars->num_y_points, 4); // max 14
2557 for (int i = 0; i < pars->num_y_points; i++) {
2558 aom_wb_write_literal(wb, pars->scaling_points_y[i][0], 8);
2559 aom_wb_write_literal(wb, pars->scaling_points_y[i][1], 8);
2560 }
2561
2562 if (!cm->seq_params->monochrome) {
2563 aom_wb_write_bit(wb, pars->chroma_scaling_from_luma);
2564 } else {
2565 assert(!pars->chroma_scaling_from_luma);
2566 }
2567
2568 if (cm->seq_params->monochrome || pars->chroma_scaling_from_luma ||
2569 ((cm->seq_params->subsampling_x == 1) &&
2570 (cm->seq_params->subsampling_y == 1) && (pars->num_y_points == 0))) {
2571 assert(pars->num_cb_points == 0 && pars->num_cr_points == 0);
2572 } else {
2573 aom_wb_write_literal(wb, pars->num_cb_points, 4); // max 10
2574 for (int i = 0; i < pars->num_cb_points; i++) {
2575 aom_wb_write_literal(wb, pars->scaling_points_cb[i][0], 8);
2576 aom_wb_write_literal(wb, pars->scaling_points_cb[i][1], 8);
2577 }
2578
2579 aom_wb_write_literal(wb, pars->num_cr_points, 4); // max 10
2580 for (int i = 0; i < pars->num_cr_points; i++) {
2581 aom_wb_write_literal(wb, pars->scaling_points_cr[i][0], 8);
2582 aom_wb_write_literal(wb, pars->scaling_points_cr[i][1], 8);
2583 }
2584 }
2585
2586 aom_wb_write_literal(wb, pars->scaling_shift - 8, 2); // 8 + value
2587
2588 // AR coefficients
2589 // Only sent if the corresponsing scaling function has
2590 // more than 0 points
2591
2592 aom_wb_write_literal(wb, pars->ar_coeff_lag, 2);
2593
2594 int num_pos_luma = 2 * pars->ar_coeff_lag * (pars->ar_coeff_lag + 1);
2595 int num_pos_chroma = num_pos_luma;
2596 if (pars->num_y_points > 0) ++num_pos_chroma;
2597
2598 if (pars->num_y_points)
2599 for (int i = 0; i < num_pos_luma; i++)
2600 aom_wb_write_literal(wb, pars->ar_coeffs_y[i] + 128, 8);
2601
2602 if (pars->num_cb_points || pars->chroma_scaling_from_luma)
2603 for (int i = 0; i < num_pos_chroma; i++)
2604 aom_wb_write_literal(wb, pars->ar_coeffs_cb[i] + 128, 8);
2605
2606 if (pars->num_cr_points || pars->chroma_scaling_from_luma)
2607 for (int i = 0; i < num_pos_chroma; i++)
2608 aom_wb_write_literal(wb, pars->ar_coeffs_cr[i] + 128, 8);
2609
2610 aom_wb_write_literal(wb, pars->ar_coeff_shift - 6, 2); // 8 + value
2611
2612 aom_wb_write_literal(wb, pars->grain_scale_shift, 2);
2613
2614 if (pars->num_cb_points) {
2615 aom_wb_write_literal(wb, pars->cb_mult, 8);
2616 aom_wb_write_literal(wb, pars->cb_luma_mult, 8);
2617 aom_wb_write_literal(wb, pars->cb_offset, 9);
2618 }
2619
2620 if (pars->num_cr_points) {
2621 aom_wb_write_literal(wb, pars->cr_mult, 8);
2622 aom_wb_write_literal(wb, pars->cr_luma_mult, 8);
2623 aom_wb_write_literal(wb, pars->cr_offset, 9);
2624 }
2625
2626 aom_wb_write_bit(wb, pars->overlap_flag);
2627
2628 aom_wb_write_bit(wb, pars->clip_to_restricted_range);
2629 }
2630
write_sb_size(const SequenceHeader * const seq_params,struct aom_write_bit_buffer * wb)2631 static AOM_INLINE void write_sb_size(const SequenceHeader *const seq_params,
2632 struct aom_write_bit_buffer *wb) {
2633 (void)seq_params;
2634 (void)wb;
2635 assert(seq_params->mib_size == mi_size_wide[seq_params->sb_size]);
2636 assert(seq_params->mib_size == 1 << seq_params->mib_size_log2);
2637 assert(seq_params->sb_size == BLOCK_128X128 ||
2638 seq_params->sb_size == BLOCK_64X64);
2639 aom_wb_write_bit(wb, seq_params->sb_size == BLOCK_128X128 ? 1 : 0);
2640 }
2641
write_sequence_header(const SequenceHeader * const seq_params,struct aom_write_bit_buffer * wb)2642 static AOM_INLINE void write_sequence_header(
2643 const SequenceHeader *const seq_params, struct aom_write_bit_buffer *wb) {
2644 aom_wb_write_literal(wb, seq_params->num_bits_width - 1, 4);
2645 aom_wb_write_literal(wb, seq_params->num_bits_height - 1, 4);
2646 aom_wb_write_literal(wb, seq_params->max_frame_width - 1,
2647 seq_params->num_bits_width);
2648 aom_wb_write_literal(wb, seq_params->max_frame_height - 1,
2649 seq_params->num_bits_height);
2650
2651 if (!seq_params->reduced_still_picture_hdr) {
2652 aom_wb_write_bit(wb, seq_params->frame_id_numbers_present_flag);
2653 if (seq_params->frame_id_numbers_present_flag) {
2654 // We must always have delta_frame_id_length < frame_id_length,
2655 // in order for a frame to be referenced with a unique delta.
2656 // Avoid wasting bits by using a coding that enforces this restriction.
2657 aom_wb_write_literal(wb, seq_params->delta_frame_id_length - 2, 4);
2658 aom_wb_write_literal(
2659 wb,
2660 seq_params->frame_id_length - seq_params->delta_frame_id_length - 1,
2661 3);
2662 }
2663 }
2664
2665 write_sb_size(seq_params, wb);
2666
2667 aom_wb_write_bit(wb, seq_params->enable_filter_intra);
2668 aom_wb_write_bit(wb, seq_params->enable_intra_edge_filter);
2669
2670 if (!seq_params->reduced_still_picture_hdr) {
2671 aom_wb_write_bit(wb, seq_params->enable_interintra_compound);
2672 aom_wb_write_bit(wb, seq_params->enable_masked_compound);
2673 aom_wb_write_bit(wb, seq_params->enable_warped_motion);
2674 aom_wb_write_bit(wb, seq_params->enable_dual_filter);
2675
2676 aom_wb_write_bit(wb, seq_params->order_hint_info.enable_order_hint);
2677
2678 if (seq_params->order_hint_info.enable_order_hint) {
2679 aom_wb_write_bit(wb, seq_params->order_hint_info.enable_dist_wtd_comp);
2680 aom_wb_write_bit(wb, seq_params->order_hint_info.enable_ref_frame_mvs);
2681 }
2682 if (seq_params->force_screen_content_tools == 2) {
2683 aom_wb_write_bit(wb, 1);
2684 } else {
2685 aom_wb_write_bit(wb, 0);
2686 aom_wb_write_bit(wb, seq_params->force_screen_content_tools);
2687 }
2688 if (seq_params->force_screen_content_tools > 0) {
2689 if (seq_params->force_integer_mv == 2) {
2690 aom_wb_write_bit(wb, 1);
2691 } else {
2692 aom_wb_write_bit(wb, 0);
2693 aom_wb_write_bit(wb, seq_params->force_integer_mv);
2694 }
2695 } else {
2696 assert(seq_params->force_integer_mv == 2);
2697 }
2698 if (seq_params->order_hint_info.enable_order_hint)
2699 aom_wb_write_literal(
2700 wb, seq_params->order_hint_info.order_hint_bits_minus_1, 3);
2701 }
2702
2703 aom_wb_write_bit(wb, seq_params->enable_superres);
2704 aom_wb_write_bit(wb, seq_params->enable_cdef);
2705 aom_wb_write_bit(wb, seq_params->enable_restoration);
2706 }
2707
write_global_motion_params(const WarpedMotionParams * params,const WarpedMotionParams * ref_params,struct aom_write_bit_buffer * wb,int allow_hp)2708 static AOM_INLINE void write_global_motion_params(
2709 const WarpedMotionParams *params, const WarpedMotionParams *ref_params,
2710 struct aom_write_bit_buffer *wb, int allow_hp) {
2711 const TransformationType type = params->wmtype;
2712
2713 // As a workaround for an AV1 spec bug, we avoid choosing TRANSLATION
2714 // type models. Check here that we don't accidentally pick one somehow.
2715 // See comments in gm_get_motion_vector() for details on the bug we're
2716 // working around here
2717 assert(type != TRANSLATION);
2718
2719 aom_wb_write_bit(wb, type != IDENTITY);
2720 if (type != IDENTITY) {
2721 aom_wb_write_bit(wb, type == ROTZOOM);
2722 if (type != ROTZOOM) aom_wb_write_bit(wb, type == TRANSLATION);
2723 }
2724
2725 if (type >= ROTZOOM) {
2726 aom_wb_write_signed_primitive_refsubexpfin(
2727 wb, GM_ALPHA_MAX + 1, SUBEXPFIN_K,
2728 (ref_params->wmmat[2] >> GM_ALPHA_PREC_DIFF) -
2729 (1 << GM_ALPHA_PREC_BITS),
2730 (params->wmmat[2] >> GM_ALPHA_PREC_DIFF) - (1 << GM_ALPHA_PREC_BITS));
2731 aom_wb_write_signed_primitive_refsubexpfin(
2732 wb, GM_ALPHA_MAX + 1, SUBEXPFIN_K,
2733 (ref_params->wmmat[3] >> GM_ALPHA_PREC_DIFF),
2734 (params->wmmat[3] >> GM_ALPHA_PREC_DIFF));
2735 }
2736
2737 if (type >= AFFINE) {
2738 aom_wb_write_signed_primitive_refsubexpfin(
2739 wb, GM_ALPHA_MAX + 1, SUBEXPFIN_K,
2740 (ref_params->wmmat[4] >> GM_ALPHA_PREC_DIFF),
2741 (params->wmmat[4] >> GM_ALPHA_PREC_DIFF));
2742 aom_wb_write_signed_primitive_refsubexpfin(
2743 wb, GM_ALPHA_MAX + 1, SUBEXPFIN_K,
2744 (ref_params->wmmat[5] >> GM_ALPHA_PREC_DIFF) -
2745 (1 << GM_ALPHA_PREC_BITS),
2746 (params->wmmat[5] >> GM_ALPHA_PREC_DIFF) - (1 << GM_ALPHA_PREC_BITS));
2747 }
2748
2749 if (type >= TRANSLATION) {
2750 const int trans_bits = (type == TRANSLATION)
2751 ? GM_ABS_TRANS_ONLY_BITS - !allow_hp
2752 : GM_ABS_TRANS_BITS;
2753 const int trans_prec_diff = (type == TRANSLATION)
2754 ? GM_TRANS_ONLY_PREC_DIFF + !allow_hp
2755 : GM_TRANS_PREC_DIFF;
2756 aom_wb_write_signed_primitive_refsubexpfin(
2757 wb, (1 << trans_bits) + 1, SUBEXPFIN_K,
2758 (ref_params->wmmat[0] >> trans_prec_diff),
2759 (params->wmmat[0] >> trans_prec_diff));
2760 aom_wb_write_signed_primitive_refsubexpfin(
2761 wb, (1 << trans_bits) + 1, SUBEXPFIN_K,
2762 (ref_params->wmmat[1] >> trans_prec_diff),
2763 (params->wmmat[1] >> trans_prec_diff));
2764 }
2765 }
2766
write_global_motion(AV1_COMP * cpi,struct aom_write_bit_buffer * wb)2767 static AOM_INLINE void write_global_motion(AV1_COMP *cpi,
2768 struct aom_write_bit_buffer *wb) {
2769 AV1_COMMON *const cm = &cpi->common;
2770 int frame;
2771 for (frame = LAST_FRAME; frame <= ALTREF_FRAME; ++frame) {
2772 const WarpedMotionParams *ref_params =
2773 cm->prev_frame ? &cm->prev_frame->global_motion[frame]
2774 : &default_warp_params;
2775 write_global_motion_params(&cm->global_motion[frame], ref_params, wb,
2776 cm->features.allow_high_precision_mv);
2777 // TODO(sarahparker, debargha): The logic in the commented out code below
2778 // does not work currently and causes mismatches when resize is on.
2779 // Fix it before turning the optimization back on.
2780 /*
2781 YV12_BUFFER_CONFIG *ref_buf = get_ref_frame_yv12_buf(cpi, frame);
2782 if (cpi->source->y_crop_width == ref_buf->y_crop_width &&
2783 cpi->source->y_crop_height == ref_buf->y_crop_height) {
2784 write_global_motion_params(&cm->global_motion[frame],
2785 &cm->prev_frame->global_motion[frame], wb,
2786 cm->features.allow_high_precision_mv);
2787 } else {
2788 assert(cm->global_motion[frame].wmtype == IDENTITY &&
2789 "Invalid warp type for frames of different resolutions");
2790 }
2791 */
2792 /*
2793 printf("Frame %d/%d: Enc Ref %d: %d %d %d %d\n",
2794 cm->current_frame.frame_number, cm->show_frame, frame,
2795 cm->global_motion[frame].wmmat[0],
2796 cm->global_motion[frame].wmmat[1], cm->global_motion[frame].wmmat[2],
2797 cm->global_motion[frame].wmmat[3]);
2798 */
2799 }
2800 }
2801
check_frame_refs_short_signaling(AV1_COMMON * const cm,bool enable_ref_short_signaling)2802 static int check_frame_refs_short_signaling(AV1_COMMON *const cm,
2803 bool enable_ref_short_signaling) {
2804 // In rtc case when res < 360p and speed >= 9, we turn on
2805 // frame_refs_short_signaling if it won't break the decoder.
2806 if (enable_ref_short_signaling) {
2807 const int gld_map_idx = get_ref_frame_map_idx(cm, GOLDEN_FRAME);
2808 const int base =
2809 1 << (cm->seq_params->order_hint_info.order_hint_bits_minus_1 + 1);
2810
2811 const int order_hint_group_cur =
2812 cm->current_frame.display_order_hint / base;
2813 const int order_hint_group_gld =
2814 cm->ref_frame_map[gld_map_idx]->display_order_hint / base;
2815 const int relative_dist = cm->current_frame.order_hint -
2816 cm->ref_frame_map[gld_map_idx]->order_hint;
2817
2818 // If current frame and GOLDEN frame are in the same order_hint group, and
2819 // they are not far apart (i.e., > 64 frames), then return 1.
2820 if (order_hint_group_cur == order_hint_group_gld && relative_dist >= 0 &&
2821 relative_dist <= 64) {
2822 return 1;
2823 }
2824 return 0;
2825 }
2826
2827 // Check whether all references are distinct frames.
2828 const RefCntBuffer *seen_bufs[INTER_REFS_PER_FRAME] = { NULL };
2829 int num_refs = 0;
2830 for (int ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
2831 const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame);
2832 if (buf != NULL) {
2833 int seen = 0;
2834 for (int i = 0; i < num_refs; i++) {
2835 if (seen_bufs[i] == buf) {
2836 seen = 1;
2837 break;
2838 }
2839 }
2840 if (!seen) seen_bufs[num_refs++] = buf;
2841 }
2842 }
2843
2844 // We only turn on frame_refs_short_signaling when all references are
2845 // distinct.
2846 if (num_refs < INTER_REFS_PER_FRAME) {
2847 // It indicates that there exist more than one reference frame pointing to
2848 // the same reference buffer, i.e. two or more references are duplicate.
2849 return 0;
2850 }
2851
2852 // Check whether the encoder side ref frame choices are aligned with that to
2853 // be derived at the decoder side.
2854 int remapped_ref_idx_decoder[REF_FRAMES];
2855
2856 const int lst_map_idx = get_ref_frame_map_idx(cm, LAST_FRAME);
2857 const int gld_map_idx = get_ref_frame_map_idx(cm, GOLDEN_FRAME);
2858
2859 // Set up the frame refs mapping indexes according to the
2860 // frame_refs_short_signaling policy.
2861 av1_set_frame_refs(cm, remapped_ref_idx_decoder, lst_map_idx, gld_map_idx);
2862
2863 // We only turn on frame_refs_short_signaling when the encoder side decision
2864 // on ref frames is identical to that at the decoder side.
2865 int frame_refs_short_signaling = 1;
2866 for (int ref_idx = 0; ref_idx < INTER_REFS_PER_FRAME; ++ref_idx) {
2867 // Compare the buffer index between two reference frames indexed
2868 // respectively by the encoder and the decoder side decisions.
2869 RefCntBuffer *ref_frame_buf_new = NULL;
2870 if (remapped_ref_idx_decoder[ref_idx] != INVALID_IDX) {
2871 ref_frame_buf_new = cm->ref_frame_map[remapped_ref_idx_decoder[ref_idx]];
2872 }
2873 if (get_ref_frame_buf(cm, LAST_FRAME + ref_idx) != ref_frame_buf_new) {
2874 frame_refs_short_signaling = 0;
2875 break;
2876 }
2877 }
2878
2879 #if 0 // For debug
2880 printf("\nFrame=%d: \n", cm->current_frame.frame_number);
2881 printf("***frame_refs_short_signaling=%d\n", frame_refs_short_signaling);
2882 for (int ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
2883 printf("enc_ref(map_idx=%d)=%d, vs. "
2884 "dec_ref(map_idx=%d)=%d\n",
2885 get_ref_frame_map_idx(cm, ref_frame), ref_frame,
2886 cm->remapped_ref_idx[ref_frame - LAST_FRAME],
2887 ref_frame);
2888 }
2889 #endif // 0
2890
2891 return frame_refs_short_signaling;
2892 }
2893
2894 // New function based on HLS R18
write_uncompressed_header_obu(AV1_COMP * cpi,MACROBLOCKD * const xd,struct aom_write_bit_buffer * saved_wb,struct aom_write_bit_buffer * wb)2895 static AOM_INLINE void write_uncompressed_header_obu(
2896 AV1_COMP *cpi, MACROBLOCKD *const xd, struct aom_write_bit_buffer *saved_wb,
2897 struct aom_write_bit_buffer *wb) {
2898 AV1_COMMON *const cm = &cpi->common;
2899 const SequenceHeader *const seq_params = cm->seq_params;
2900 const CommonQuantParams *quant_params = &cm->quant_params;
2901 CurrentFrame *const current_frame = &cm->current_frame;
2902 FeatureFlags *const features = &cm->features;
2903
2904 if (!cpi->sf.rt_sf.enable_ref_short_signaling ||
2905 !seq_params->order_hint_info.enable_order_hint ||
2906 seq_params->order_hint_info.enable_ref_frame_mvs) {
2907 current_frame->frame_refs_short_signaling = 0;
2908 } else {
2909 current_frame->frame_refs_short_signaling = 1;
2910 }
2911
2912 if (seq_params->still_picture) {
2913 assert(cm->show_existing_frame == 0);
2914 assert(cm->show_frame == 1);
2915 assert(current_frame->frame_type == KEY_FRAME);
2916 }
2917 if (!seq_params->reduced_still_picture_hdr) {
2918 if (encode_show_existing_frame(cm)) {
2919 aom_wb_write_bit(wb, 1); // show_existing_frame
2920 aom_wb_write_literal(wb, cpi->existing_fb_idx_to_show, 3);
2921
2922 if (seq_params->decoder_model_info_present_flag &&
2923 seq_params->timing_info.equal_picture_interval == 0) {
2924 write_tu_pts_info(cm, wb);
2925 }
2926 if (seq_params->frame_id_numbers_present_flag) {
2927 int frame_id_len = seq_params->frame_id_length;
2928 int display_frame_id = cm->ref_frame_id[cpi->existing_fb_idx_to_show];
2929 aom_wb_write_literal(wb, display_frame_id, frame_id_len);
2930 }
2931 return;
2932 } else {
2933 aom_wb_write_bit(wb, 0); // show_existing_frame
2934 }
2935
2936 aom_wb_write_literal(wb, current_frame->frame_type, 2);
2937
2938 aom_wb_write_bit(wb, cm->show_frame);
2939 if (cm->show_frame) {
2940 if (seq_params->decoder_model_info_present_flag &&
2941 seq_params->timing_info.equal_picture_interval == 0)
2942 write_tu_pts_info(cm, wb);
2943 } else {
2944 aom_wb_write_bit(wb, cm->showable_frame);
2945 }
2946 if (frame_is_sframe(cm)) {
2947 assert(features->error_resilient_mode);
2948 } else if (!(current_frame->frame_type == KEY_FRAME && cm->show_frame)) {
2949 aom_wb_write_bit(wb, features->error_resilient_mode);
2950 }
2951 }
2952 aom_wb_write_bit(wb, features->disable_cdf_update);
2953
2954 if (seq_params->force_screen_content_tools == 2) {
2955 aom_wb_write_bit(wb, features->allow_screen_content_tools);
2956 } else {
2957 assert(features->allow_screen_content_tools ==
2958 seq_params->force_screen_content_tools);
2959 }
2960
2961 if (features->allow_screen_content_tools) {
2962 if (seq_params->force_integer_mv == 2) {
2963 aom_wb_write_bit(wb, features->cur_frame_force_integer_mv);
2964 } else {
2965 assert(features->cur_frame_force_integer_mv ==
2966 seq_params->force_integer_mv);
2967 }
2968 } else {
2969 assert(features->cur_frame_force_integer_mv == 0);
2970 }
2971
2972 int frame_size_override_flag = 0;
2973
2974 if (seq_params->reduced_still_picture_hdr) {
2975 assert(cm->superres_upscaled_width == seq_params->max_frame_width &&
2976 cm->superres_upscaled_height == seq_params->max_frame_height);
2977 } else {
2978 if (seq_params->frame_id_numbers_present_flag) {
2979 int frame_id_len = seq_params->frame_id_length;
2980 aom_wb_write_literal(wb, cm->current_frame_id, frame_id_len);
2981 }
2982
2983 if (cm->superres_upscaled_width > seq_params->max_frame_width ||
2984 cm->superres_upscaled_height > seq_params->max_frame_height) {
2985 aom_internal_error(cm->error, AOM_CODEC_UNSUP_BITSTREAM,
2986 "Frame dimensions are larger than the maximum values");
2987 }
2988
2989 frame_size_override_flag =
2990 frame_is_sframe(cm)
2991 ? 1
2992 : (cm->superres_upscaled_width != seq_params->max_frame_width ||
2993 cm->superres_upscaled_height != seq_params->max_frame_height);
2994 if (!frame_is_sframe(cm)) aom_wb_write_bit(wb, frame_size_override_flag);
2995
2996 if (seq_params->order_hint_info.enable_order_hint)
2997 aom_wb_write_literal(
2998 wb, current_frame->order_hint,
2999 seq_params->order_hint_info.order_hint_bits_minus_1 + 1);
3000
3001 if (!features->error_resilient_mode && !frame_is_intra_only(cm)) {
3002 aom_wb_write_literal(wb, features->primary_ref_frame, PRIMARY_REF_BITS);
3003 }
3004 }
3005
3006 if (seq_params->decoder_model_info_present_flag) {
3007 aom_wb_write_bit(wb, cpi->ppi->buffer_removal_time_present);
3008 if (cpi->ppi->buffer_removal_time_present) {
3009 for (int op_num = 0;
3010 op_num < seq_params->operating_points_cnt_minus_1 + 1; op_num++) {
3011 if (seq_params->op_params[op_num].decoder_model_param_present_flag) {
3012 if (seq_params->operating_point_idc[op_num] == 0 ||
3013 ((seq_params->operating_point_idc[op_num] >>
3014 cm->temporal_layer_id) &
3015 0x1 &&
3016 (seq_params->operating_point_idc[op_num] >>
3017 (cm->spatial_layer_id + 8)) &
3018 0x1)) {
3019 aom_wb_write_unsigned_literal(
3020 wb, cm->buffer_removal_times[op_num],
3021 seq_params->decoder_model_info.buffer_removal_time_length);
3022 cm->buffer_removal_times[op_num]++;
3023 if (cm->buffer_removal_times[op_num] == 0) {
3024 aom_internal_error(cm->error, AOM_CODEC_UNSUP_BITSTREAM,
3025 "buffer_removal_time overflowed");
3026 }
3027 }
3028 }
3029 }
3030 }
3031 }
3032
3033 // Shown keyframes and switch-frames automatically refreshes all reference
3034 // frames. For all other frame types, we need to write refresh_frame_flags.
3035 if ((current_frame->frame_type == KEY_FRAME && !cm->show_frame) ||
3036 current_frame->frame_type == INTER_FRAME ||
3037 current_frame->frame_type == INTRA_ONLY_FRAME)
3038 aom_wb_write_literal(wb, current_frame->refresh_frame_flags, REF_FRAMES);
3039
3040 if (!frame_is_intra_only(cm) || current_frame->refresh_frame_flags != 0xff) {
3041 // Write all ref frame order hints if error_resilient_mode == 1
3042 if (features->error_resilient_mode &&
3043 seq_params->order_hint_info.enable_order_hint) {
3044 for (int ref_idx = 0; ref_idx < REF_FRAMES; ref_idx++) {
3045 aom_wb_write_literal(
3046 wb, cm->ref_frame_map[ref_idx]->order_hint,
3047 seq_params->order_hint_info.order_hint_bits_minus_1 + 1);
3048 }
3049 }
3050 }
3051
3052 if (current_frame->frame_type == KEY_FRAME) {
3053 write_frame_size(cm, frame_size_override_flag, wb);
3054 assert(!av1_superres_scaled(cm) || !features->allow_intrabc);
3055 if (features->allow_screen_content_tools && !av1_superres_scaled(cm))
3056 aom_wb_write_bit(wb, features->allow_intrabc);
3057 } else {
3058 if (current_frame->frame_type == INTRA_ONLY_FRAME) {
3059 write_frame_size(cm, frame_size_override_flag, wb);
3060 assert(!av1_superres_scaled(cm) || !features->allow_intrabc);
3061 if (features->allow_screen_content_tools && !av1_superres_scaled(cm))
3062 aom_wb_write_bit(wb, features->allow_intrabc);
3063 } else if (current_frame->frame_type == INTER_FRAME ||
3064 frame_is_sframe(cm)) {
3065 MV_REFERENCE_FRAME ref_frame;
3066
3067 // NOTE: Error resilient mode turns off frame_refs_short_signaling
3068 // automatically.
3069 #define FRAME_REFS_SHORT_SIGNALING 0
3070 #if FRAME_REFS_SHORT_SIGNALING
3071 current_frame->frame_refs_short_signaling =
3072 seq_params->order_hint_info.enable_order_hint;
3073 #endif // FRAME_REFS_SHORT_SIGNALING
3074
3075 if (current_frame->frame_refs_short_signaling) {
3076 // In rtc case when cpi->sf.rt_sf.enable_ref_short_signaling is true,
3077 // we turn on frame_refs_short_signaling when the current frame and
3078 // golden frame are in the same order_hint group, and their relative
3079 // distance is <= 64 (in order to be decodable).
3080
3081 // For other cases, an example solution for encoder-side
3082 // implementation on frame_refs_short_signaling is also provided in
3083 // this function, where frame_refs_short_signaling is only turned on
3084 // when the encoder side decision on ref frames is identical to that
3085 // at the decoder side.
3086
3087 current_frame->frame_refs_short_signaling =
3088 check_frame_refs_short_signaling(
3089 cm, cpi->sf.rt_sf.enable_ref_short_signaling);
3090 }
3091
3092 if (seq_params->order_hint_info.enable_order_hint)
3093 aom_wb_write_bit(wb, current_frame->frame_refs_short_signaling);
3094
3095 if (current_frame->frame_refs_short_signaling) {
3096 const int lst_ref = get_ref_frame_map_idx(cm, LAST_FRAME);
3097 aom_wb_write_literal(wb, lst_ref, REF_FRAMES_LOG2);
3098
3099 const int gld_ref = get_ref_frame_map_idx(cm, GOLDEN_FRAME);
3100 aom_wb_write_literal(wb, gld_ref, REF_FRAMES_LOG2);
3101 }
3102
3103 for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
3104 assert(get_ref_frame_map_idx(cm, ref_frame) != INVALID_IDX);
3105 if (!current_frame->frame_refs_short_signaling)
3106 aom_wb_write_literal(wb, get_ref_frame_map_idx(cm, ref_frame),
3107 REF_FRAMES_LOG2);
3108 if (seq_params->frame_id_numbers_present_flag) {
3109 int i = get_ref_frame_map_idx(cm, ref_frame);
3110 int frame_id_len = seq_params->frame_id_length;
3111 int diff_len = seq_params->delta_frame_id_length;
3112 int delta_frame_id_minus_1 =
3113 ((cm->current_frame_id - cm->ref_frame_id[i] +
3114 (1 << frame_id_len)) %
3115 (1 << frame_id_len)) -
3116 1;
3117 if (delta_frame_id_minus_1 < 0 ||
3118 delta_frame_id_minus_1 >= (1 << diff_len)) {
3119 aom_internal_error(cm->error, AOM_CODEC_ERROR,
3120 "Invalid delta_frame_id_minus_1");
3121 }
3122 aom_wb_write_literal(wb, delta_frame_id_minus_1, diff_len);
3123 }
3124 }
3125
3126 if (!features->error_resilient_mode && frame_size_override_flag) {
3127 write_frame_size_with_refs(cm, wb);
3128 } else {
3129 write_frame_size(cm, frame_size_override_flag, wb);
3130 }
3131
3132 if (!features->cur_frame_force_integer_mv)
3133 aom_wb_write_bit(wb, features->allow_high_precision_mv);
3134 write_frame_interp_filter(features->interp_filter, wb);
3135 aom_wb_write_bit(wb, features->switchable_motion_mode);
3136 if (frame_might_allow_ref_frame_mvs(cm)) {
3137 aom_wb_write_bit(wb, features->allow_ref_frame_mvs);
3138 } else {
3139 assert(features->allow_ref_frame_mvs == 0);
3140 }
3141 }
3142 }
3143
3144 const int might_bwd_adapt = !(seq_params->reduced_still_picture_hdr) &&
3145 !(features->disable_cdf_update);
3146 if (cm->tiles.large_scale)
3147 assert(features->refresh_frame_context == REFRESH_FRAME_CONTEXT_DISABLED);
3148
3149 if (might_bwd_adapt) {
3150 aom_wb_write_bit(
3151 wb, features->refresh_frame_context == REFRESH_FRAME_CONTEXT_DISABLED);
3152 }
3153
3154 write_tile_info(cm, saved_wb, wb);
3155 encode_quantization(quant_params, av1_num_planes(cm),
3156 cm->seq_params->separate_uv_delta_q, wb);
3157 encode_segmentation(cm, wb);
3158
3159 const DeltaQInfo *const delta_q_info = &cm->delta_q_info;
3160 if (delta_q_info->delta_q_present_flag) assert(quant_params->base_qindex > 0);
3161 if (quant_params->base_qindex > 0) {
3162 aom_wb_write_bit(wb, delta_q_info->delta_q_present_flag);
3163 if (delta_q_info->delta_q_present_flag) {
3164 aom_wb_write_literal(wb, get_msb(delta_q_info->delta_q_res), 2);
3165 xd->current_base_qindex = quant_params->base_qindex;
3166 if (features->allow_intrabc)
3167 assert(delta_q_info->delta_lf_present_flag == 0);
3168 else
3169 aom_wb_write_bit(wb, delta_q_info->delta_lf_present_flag);
3170 if (delta_q_info->delta_lf_present_flag) {
3171 aom_wb_write_literal(wb, get_msb(delta_q_info->delta_lf_res), 2);
3172 aom_wb_write_bit(wb, delta_q_info->delta_lf_multi);
3173 av1_reset_loop_filter_delta(xd, av1_num_planes(cm));
3174 }
3175 }
3176 }
3177
3178 if (features->all_lossless) {
3179 assert(!av1_superres_scaled(cm));
3180 } else {
3181 if (!features->coded_lossless) {
3182 encode_loopfilter(cm, wb);
3183 encode_cdef(cm, wb);
3184 }
3185 encode_restoration_mode(cm, wb);
3186 }
3187
3188 // Write TX mode
3189 if (features->coded_lossless)
3190 assert(features->tx_mode == ONLY_4X4);
3191 else
3192 aom_wb_write_bit(wb, features->tx_mode == TX_MODE_SELECT);
3193
3194 if (!frame_is_intra_only(cm)) {
3195 const int use_hybrid_pred =
3196 current_frame->reference_mode == REFERENCE_MODE_SELECT;
3197
3198 aom_wb_write_bit(wb, use_hybrid_pred);
3199 }
3200
3201 if (current_frame->skip_mode_info.skip_mode_allowed)
3202 aom_wb_write_bit(wb, current_frame->skip_mode_info.skip_mode_flag);
3203
3204 if (frame_might_allow_warped_motion(cm))
3205 aom_wb_write_bit(wb, features->allow_warped_motion);
3206 else
3207 assert(!features->allow_warped_motion);
3208
3209 aom_wb_write_bit(wb, features->reduced_tx_set_used);
3210
3211 if (!frame_is_intra_only(cm)) write_global_motion(cpi, wb);
3212
3213 if (seq_params->film_grain_params_present &&
3214 (cm->show_frame || cm->showable_frame))
3215 write_film_grain_params(cpi, wb);
3216
3217 if (cm->tiles.large_scale) write_ext_tile_info(cm, saved_wb, wb);
3218 }
3219
choose_size_bytes(uint32_t size,int spare_msbs)3220 static int choose_size_bytes(uint32_t size, int spare_msbs) {
3221 // Choose the number of bytes required to represent size, without
3222 // using the 'spare_msbs' number of most significant bits.
3223
3224 // Make sure we will fit in 4 bytes to start with..
3225 if (spare_msbs > 0 && size >> (32 - spare_msbs) != 0) return -1;
3226
3227 // Normalise to 32 bits
3228 size <<= spare_msbs;
3229
3230 if (size >> 24 != 0)
3231 return 4;
3232 else if (size >> 16 != 0)
3233 return 3;
3234 else if (size >> 8 != 0)
3235 return 2;
3236 else
3237 return 1;
3238 }
3239
mem_put_varsize(uint8_t * const dst,const int sz,const int val)3240 static AOM_INLINE void mem_put_varsize(uint8_t *const dst, const int sz,
3241 const int val) {
3242 switch (sz) {
3243 case 1: dst[0] = (uint8_t)(val & 0xff); break;
3244 case 2: mem_put_le16(dst, val); break;
3245 case 3: mem_put_le24(dst, val); break;
3246 case 4: mem_put_le32(dst, val); break;
3247 default: assert(0 && "Invalid size"); break;
3248 }
3249 }
3250
remux_tiles(const CommonTileParams * const tiles,uint8_t * dst,const uint32_t data_size,const uint32_t max_tile_size,const uint32_t max_tile_col_size,int * const tile_size_bytes,int * const tile_col_size_bytes)3251 static int remux_tiles(const CommonTileParams *const tiles, uint8_t *dst,
3252 const uint32_t data_size, const uint32_t max_tile_size,
3253 const uint32_t max_tile_col_size,
3254 int *const tile_size_bytes,
3255 int *const tile_col_size_bytes) {
3256 // Choose the tile size bytes (tsb) and tile column size bytes (tcsb)
3257 int tsb;
3258 int tcsb;
3259
3260 if (tiles->large_scale) {
3261 // The top bit in the tile size field indicates tile copy mode, so we
3262 // have 1 less bit to code the tile size
3263 tsb = choose_size_bytes(max_tile_size, 1);
3264 tcsb = choose_size_bytes(max_tile_col_size, 0);
3265 } else {
3266 tsb = choose_size_bytes(max_tile_size, 0);
3267 tcsb = 4; // This is ignored
3268 (void)max_tile_col_size;
3269 }
3270
3271 assert(tsb > 0);
3272 assert(tcsb > 0);
3273
3274 *tile_size_bytes = tsb;
3275 *tile_col_size_bytes = tcsb;
3276 if (tsb == 4 && tcsb == 4) return data_size;
3277
3278 uint32_t wpos = 0;
3279 uint32_t rpos = 0;
3280
3281 if (tiles->large_scale) {
3282 int tile_row;
3283 int tile_col;
3284
3285 for (tile_col = 0; tile_col < tiles->cols; tile_col++) {
3286 // All but the last column has a column header
3287 if (tile_col < tiles->cols - 1) {
3288 uint32_t tile_col_size = mem_get_le32(dst + rpos);
3289 rpos += 4;
3290
3291 // Adjust the tile column size by the number of bytes removed
3292 // from the tile size fields.
3293 tile_col_size -= (4 - tsb) * tiles->rows;
3294
3295 mem_put_varsize(dst + wpos, tcsb, tile_col_size);
3296 wpos += tcsb;
3297 }
3298
3299 for (tile_row = 0; tile_row < tiles->rows; tile_row++) {
3300 // All, including the last row has a header
3301 uint32_t tile_header = mem_get_le32(dst + rpos);
3302 rpos += 4;
3303
3304 // If this is a copy tile, we need to shift the MSB to the
3305 // top bit of the new width, and there is no data to copy.
3306 if (tile_header >> 31 != 0) {
3307 if (tsb < 4) tile_header >>= 32 - 8 * tsb;
3308 mem_put_varsize(dst + wpos, tsb, tile_header);
3309 wpos += tsb;
3310 } else {
3311 mem_put_varsize(dst + wpos, tsb, tile_header);
3312 wpos += tsb;
3313
3314 tile_header += AV1_MIN_TILE_SIZE_BYTES;
3315 memmove(dst + wpos, dst + rpos, tile_header);
3316 rpos += tile_header;
3317 wpos += tile_header;
3318 }
3319 }
3320 }
3321
3322 assert(rpos > wpos);
3323 assert(rpos == data_size);
3324
3325 return wpos;
3326 }
3327 const int n_tiles = tiles->cols * tiles->rows;
3328 int n;
3329
3330 for (n = 0; n < n_tiles; n++) {
3331 int tile_size;
3332
3333 if (n == n_tiles - 1) {
3334 tile_size = data_size - rpos;
3335 } else {
3336 tile_size = mem_get_le32(dst + rpos);
3337 rpos += 4;
3338 mem_put_varsize(dst + wpos, tsb, tile_size);
3339 tile_size += AV1_MIN_TILE_SIZE_BYTES;
3340 wpos += tsb;
3341 }
3342
3343 memmove(dst + wpos, dst + rpos, tile_size);
3344
3345 rpos += tile_size;
3346 wpos += tile_size;
3347 }
3348
3349 assert(rpos > wpos);
3350 assert(rpos == data_size);
3351
3352 return wpos;
3353 }
3354
av1_write_obu_header(AV1LevelParams * const level_params,int * frame_header_count,OBU_TYPE obu_type,int obu_extension,uint8_t * const dst)3355 uint32_t av1_write_obu_header(AV1LevelParams *const level_params,
3356 int *frame_header_count, OBU_TYPE obu_type,
3357 int obu_extension, uint8_t *const dst) {
3358 if (level_params->keep_level_stats &&
3359 (obu_type == OBU_FRAME || obu_type == OBU_FRAME_HEADER))
3360 ++(*frame_header_count);
3361
3362 struct aom_write_bit_buffer wb = { dst, 0 };
3363 uint32_t size = 0;
3364
3365 aom_wb_write_literal(&wb, 0, 1); // forbidden bit.
3366 aom_wb_write_literal(&wb, (int)obu_type, 4);
3367 aom_wb_write_literal(&wb, obu_extension ? 1 : 0, 1);
3368 aom_wb_write_literal(&wb, 1, 1); // obu_has_size_field
3369 aom_wb_write_literal(&wb, 0, 1); // reserved
3370
3371 if (obu_extension) {
3372 aom_wb_write_literal(&wb, obu_extension & 0xFF, 8);
3373 }
3374
3375 size = aom_wb_bytes_written(&wb);
3376 return size;
3377 }
3378
av1_write_uleb_obu_size(size_t obu_header_size,size_t obu_payload_size,uint8_t * dest)3379 int av1_write_uleb_obu_size(size_t obu_header_size, size_t obu_payload_size,
3380 uint8_t *dest) {
3381 const size_t offset = obu_header_size;
3382 size_t coded_obu_size = 0;
3383 const uint32_t obu_size = (uint32_t)obu_payload_size;
3384 assert(obu_size == obu_payload_size);
3385
3386 if (aom_uleb_encode(obu_size, sizeof(obu_size), dest + offset,
3387 &coded_obu_size) != 0) {
3388 return AOM_CODEC_ERROR;
3389 }
3390
3391 return AOM_CODEC_OK;
3392 }
3393
av1_obu_memmove(size_t obu_header_size,size_t obu_payload_size,uint8_t * data)3394 size_t av1_obu_memmove(size_t obu_header_size, size_t obu_payload_size,
3395 uint8_t *data) {
3396 const size_t length_field_size = aom_uleb_size_in_bytes(obu_payload_size);
3397 const size_t move_dst_offset = length_field_size + obu_header_size;
3398 const size_t move_src_offset = obu_header_size;
3399 const size_t move_size = obu_payload_size;
3400 memmove(data + move_dst_offset, data + move_src_offset, move_size);
3401 return length_field_size;
3402 }
3403
add_trailing_bits(struct aom_write_bit_buffer * wb)3404 static AOM_INLINE void add_trailing_bits(struct aom_write_bit_buffer *wb) {
3405 if (aom_wb_is_byte_aligned(wb)) {
3406 aom_wb_write_literal(wb, 0x80, 8);
3407 } else {
3408 // assumes that the other bits are already 0s
3409 aom_wb_write_bit(wb, 1);
3410 }
3411 }
3412
write_bitstream_level(AV1_LEVEL seq_level_idx,struct aom_write_bit_buffer * wb)3413 static AOM_INLINE void write_bitstream_level(AV1_LEVEL seq_level_idx,
3414 struct aom_write_bit_buffer *wb) {
3415 assert(is_valid_seq_level_idx(seq_level_idx));
3416 aom_wb_write_literal(wb, seq_level_idx, LEVEL_BITS);
3417 }
3418
av1_write_sequence_header_obu(const SequenceHeader * seq_params,uint8_t * const dst)3419 uint32_t av1_write_sequence_header_obu(const SequenceHeader *seq_params,
3420 uint8_t *const dst) {
3421 struct aom_write_bit_buffer wb = { dst, 0 };
3422 uint32_t size = 0;
3423
3424 write_profile(seq_params->profile, &wb);
3425
3426 // Still picture or not
3427 aom_wb_write_bit(&wb, seq_params->still_picture);
3428 assert(IMPLIES(!seq_params->still_picture,
3429 !seq_params->reduced_still_picture_hdr));
3430 // whether to use reduced still picture header
3431 aom_wb_write_bit(&wb, seq_params->reduced_still_picture_hdr);
3432
3433 if (seq_params->reduced_still_picture_hdr) {
3434 assert(seq_params->timing_info_present == 0);
3435 assert(seq_params->decoder_model_info_present_flag == 0);
3436 assert(seq_params->display_model_info_present_flag == 0);
3437 write_bitstream_level(seq_params->seq_level_idx[0], &wb);
3438 } else {
3439 aom_wb_write_bit(
3440 &wb, seq_params->timing_info_present); // timing info present flag
3441
3442 if (seq_params->timing_info_present) {
3443 // timing_info
3444 write_timing_info_header(&seq_params->timing_info, &wb);
3445 aom_wb_write_bit(&wb, seq_params->decoder_model_info_present_flag);
3446 if (seq_params->decoder_model_info_present_flag) {
3447 write_decoder_model_info(&seq_params->decoder_model_info, &wb);
3448 }
3449 }
3450 aom_wb_write_bit(&wb, seq_params->display_model_info_present_flag);
3451 aom_wb_write_literal(&wb, seq_params->operating_points_cnt_minus_1,
3452 OP_POINTS_CNT_MINUS_1_BITS);
3453 int i;
3454 for (i = 0; i < seq_params->operating_points_cnt_minus_1 + 1; i++) {
3455 aom_wb_write_literal(&wb, seq_params->operating_point_idc[i],
3456 OP_POINTS_IDC_BITS);
3457 write_bitstream_level(seq_params->seq_level_idx[i], &wb);
3458 if (seq_params->seq_level_idx[i] >= SEQ_LEVEL_4_0)
3459 aom_wb_write_bit(&wb, seq_params->tier[i]);
3460 if (seq_params->decoder_model_info_present_flag) {
3461 aom_wb_write_bit(
3462 &wb, seq_params->op_params[i].decoder_model_param_present_flag);
3463 if (seq_params->op_params[i].decoder_model_param_present_flag) {
3464 write_dec_model_op_parameters(
3465 &seq_params->op_params[i],
3466 seq_params->decoder_model_info
3467 .encoder_decoder_buffer_delay_length,
3468 &wb);
3469 }
3470 }
3471 if (seq_params->display_model_info_present_flag) {
3472 aom_wb_write_bit(
3473 &wb, seq_params->op_params[i].display_model_param_present_flag);
3474 if (seq_params->op_params[i].display_model_param_present_flag) {
3475 assert(seq_params->op_params[i].initial_display_delay >= 1);
3476 assert(seq_params->op_params[i].initial_display_delay <= 10);
3477 aom_wb_write_literal(
3478 &wb, seq_params->op_params[i].initial_display_delay - 1, 4);
3479 }
3480 }
3481 }
3482 }
3483 write_sequence_header(seq_params, &wb);
3484
3485 write_color_config(seq_params, &wb);
3486
3487 aom_wb_write_bit(&wb, seq_params->film_grain_params_present);
3488
3489 add_trailing_bits(&wb);
3490
3491 size = aom_wb_bytes_written(&wb);
3492 return size;
3493 }
3494
write_frame_header_obu(AV1_COMP * cpi,MACROBLOCKD * const xd,struct aom_write_bit_buffer * saved_wb,uint8_t * const dst,int append_trailing_bits)3495 static uint32_t write_frame_header_obu(AV1_COMP *cpi, MACROBLOCKD *const xd,
3496 struct aom_write_bit_buffer *saved_wb,
3497 uint8_t *const dst,
3498 int append_trailing_bits) {
3499 struct aom_write_bit_buffer wb = { dst, 0 };
3500 write_uncompressed_header_obu(cpi, xd, saved_wb, &wb);
3501 if (append_trailing_bits) add_trailing_bits(&wb);
3502 return aom_wb_bytes_written(&wb);
3503 }
3504
write_tile_group_header(uint8_t * const dst,int start_tile,int end_tile,int tiles_log2,int tile_start_and_end_present_flag)3505 static uint32_t write_tile_group_header(uint8_t *const dst, int start_tile,
3506 int end_tile, int tiles_log2,
3507 int tile_start_and_end_present_flag) {
3508 struct aom_write_bit_buffer wb = { dst, 0 };
3509 uint32_t size = 0;
3510
3511 if (!tiles_log2) return size;
3512
3513 aom_wb_write_bit(&wb, tile_start_and_end_present_flag);
3514
3515 if (tile_start_and_end_present_flag) {
3516 aom_wb_write_literal(&wb, start_tile, tiles_log2);
3517 aom_wb_write_literal(&wb, end_tile, tiles_log2);
3518 }
3519
3520 size = aom_wb_bytes_written(&wb);
3521 return size;
3522 }
3523
3524 extern void av1_print_uncompressed_frame_header(const uint8_t *data, int size,
3525 const char *filename);
3526
3527 typedef struct {
3528 uint32_t tg_hdr_size;
3529 uint32_t frame_header_size;
3530 } LargeTileFrameOBU;
3531
3532 // Initialize OBU header for large scale tile case.
init_large_scale_tile_obu_header(AV1_COMP * const cpi,uint8_t ** data,struct aom_write_bit_buffer * saved_wb,LargeTileFrameOBU * lst_obu)3533 static uint32_t init_large_scale_tile_obu_header(
3534 AV1_COMP *const cpi, uint8_t **data, struct aom_write_bit_buffer *saved_wb,
3535 LargeTileFrameOBU *lst_obu) {
3536 AV1LevelParams *const level_params = &cpi->ppi->level_params;
3537 CurrentFrame *const current_frame = &cpi->common.current_frame;
3538 // For large_scale_tile case, we always have only one tile group, so it can
3539 // be written as an OBU_FRAME.
3540 const OBU_TYPE obu_type = OBU_FRAME;
3541 lst_obu->tg_hdr_size = av1_write_obu_header(
3542 level_params, &cpi->frame_header_count, obu_type, 0, *data);
3543 *data += lst_obu->tg_hdr_size;
3544
3545 const uint32_t frame_header_size =
3546 write_frame_header_obu(cpi, &cpi->td.mb.e_mbd, saved_wb, *data, 0);
3547 *data += frame_header_size;
3548 lst_obu->frame_header_size = frame_header_size;
3549 // (yunqing) This test ensures the correctness of large scale tile coding.
3550 if (cpi->oxcf.tile_cfg.enable_ext_tile_debug) {
3551 char fn[20] = "./fh";
3552 fn[4] = current_frame->frame_number / 100 + '0';
3553 fn[5] = (current_frame->frame_number % 100) / 10 + '0';
3554 fn[6] = (current_frame->frame_number % 10) + '0';
3555 fn[7] = '\0';
3556 av1_print_uncompressed_frame_header(*data - frame_header_size,
3557 frame_header_size, fn);
3558 }
3559 return frame_header_size;
3560 }
3561
3562 // Write total buffer size and related information into the OBU header for large
3563 // scale tile case.
write_large_scale_tile_obu_size(const CommonTileParams * const tiles,uint8_t * const dst,uint8_t * data,struct aom_write_bit_buffer * saved_wb,LargeTileFrameOBU * const lst_obu,int have_tiles,uint32_t * total_size,int max_tile_size,int max_tile_col_size)3564 static void write_large_scale_tile_obu_size(
3565 const CommonTileParams *const tiles, uint8_t *const dst, uint8_t *data,
3566 struct aom_write_bit_buffer *saved_wb, LargeTileFrameOBU *const lst_obu,
3567 int have_tiles, uint32_t *total_size, int max_tile_size,
3568 int max_tile_col_size) {
3569 int tile_size_bytes = 0;
3570 int tile_col_size_bytes = 0;
3571 if (have_tiles) {
3572 *total_size = remux_tiles(
3573 tiles, data, *total_size - lst_obu->frame_header_size, max_tile_size,
3574 max_tile_col_size, &tile_size_bytes, &tile_col_size_bytes);
3575 *total_size += lst_obu->frame_header_size;
3576 }
3577
3578 // In EXT_TILE case, only use 1 tile group. Follow the obu syntax, write
3579 // current tile group size before tile data(include tile column header).
3580 // Tile group size doesn't include the bytes storing tg size.
3581 *total_size += lst_obu->tg_hdr_size;
3582 const uint32_t obu_payload_size = *total_size - lst_obu->tg_hdr_size;
3583 const size_t length_field_size =
3584 av1_obu_memmove(lst_obu->tg_hdr_size, obu_payload_size, dst);
3585 if (av1_write_uleb_obu_size(lst_obu->tg_hdr_size, obu_payload_size, dst) !=
3586 AOM_CODEC_OK)
3587 assert(0);
3588
3589 *total_size += (uint32_t)length_field_size;
3590 saved_wb->bit_buffer += length_field_size;
3591
3592 // Now fill in the gaps in the uncompressed header.
3593 if (have_tiles) {
3594 assert(tile_col_size_bytes >= 1 && tile_col_size_bytes <= 4);
3595 aom_wb_overwrite_literal(saved_wb, tile_col_size_bytes - 1, 2);
3596
3597 assert(tile_size_bytes >= 1 && tile_size_bytes <= 4);
3598 aom_wb_overwrite_literal(saved_wb, tile_size_bytes - 1, 2);
3599 }
3600 }
3601
3602 // Store information on each large scale tile in the OBU header.
write_large_scale_tile_obu(AV1_COMP * const cpi,uint8_t * const dst,LargeTileFrameOBU * const lst_obu,int * const largest_tile_id,uint32_t * total_size,const int have_tiles,unsigned int * const max_tile_size,unsigned int * const max_tile_col_size)3603 static void write_large_scale_tile_obu(
3604 AV1_COMP *const cpi, uint8_t *const dst, LargeTileFrameOBU *const lst_obu,
3605 int *const largest_tile_id, uint32_t *total_size, const int have_tiles,
3606 unsigned int *const max_tile_size, unsigned int *const max_tile_col_size) {
3607 AV1_COMMON *const cm = &cpi->common;
3608 const CommonTileParams *const tiles = &cm->tiles;
3609
3610 TileBufferEnc tile_buffers[MAX_TILE_ROWS][MAX_TILE_COLS];
3611 const int tile_cols = tiles->cols;
3612 const int tile_rows = tiles->rows;
3613 unsigned int tile_size = 0;
3614
3615 av1_reset_pack_bs_thread_data(&cpi->td);
3616 for (int tile_col = 0; tile_col < tile_cols; tile_col++) {
3617 TileInfo tile_info;
3618 const int is_last_col = (tile_col == tile_cols - 1);
3619 const uint32_t col_offset = *total_size;
3620
3621 av1_tile_set_col(&tile_info, cm, tile_col);
3622
3623 // The last column does not have a column header
3624 if (!is_last_col) *total_size += 4;
3625
3626 for (int tile_row = 0; tile_row < tile_rows; tile_row++) {
3627 TileBufferEnc *const buf = &tile_buffers[tile_row][tile_col];
3628 const int data_offset = have_tiles ? 4 : 0;
3629 const int tile_idx = tile_row * tile_cols + tile_col;
3630 TileDataEnc *this_tile = &cpi->tile_data[tile_idx];
3631 av1_tile_set_row(&tile_info, cm, tile_row);
3632 aom_writer mode_bc;
3633
3634 buf->data = dst + *total_size + lst_obu->tg_hdr_size;
3635
3636 // Is CONFIG_EXT_TILE = 1, every tile in the row has a header,
3637 // even for the last one, unless no tiling is used at all.
3638 *total_size += data_offset;
3639 cpi->td.mb.e_mbd.tile_ctx = &this_tile->tctx;
3640 mode_bc.allow_update_cdf = !tiles->large_scale;
3641 mode_bc.allow_update_cdf =
3642 mode_bc.allow_update_cdf && !cm->features.disable_cdf_update;
3643 aom_start_encode(&mode_bc, buf->data + data_offset);
3644 write_modes(cpi, &cpi->td, &tile_info, &mode_bc, tile_row, tile_col);
3645 if (aom_stop_encode(&mode_bc) < 0) {
3646 aom_internal_error(cm->error, AOM_CODEC_ERROR, "Error writing modes");
3647 }
3648 tile_size = mode_bc.pos;
3649 buf->size = tile_size;
3650
3651 // Record the maximum tile size we see, so we can compact headers later.
3652 if (tile_size > *max_tile_size) {
3653 *max_tile_size = tile_size;
3654 *largest_tile_id = tile_cols * tile_row + tile_col;
3655 }
3656
3657 if (have_tiles) {
3658 // tile header: size of this tile, or copy offset
3659 uint32_t tile_header = tile_size - AV1_MIN_TILE_SIZE_BYTES;
3660 const int tile_copy_mode =
3661 ((AOMMAX(tiles->width, tiles->height) << MI_SIZE_LOG2) <= 256) ? 1
3662 : 0;
3663
3664 // If tile_copy_mode = 1, check if this tile is a copy tile.
3665 // Very low chances to have copy tiles on the key frames, so don't
3666 // search on key frames to reduce unnecessary search.
3667 if (cm->current_frame.frame_type != KEY_FRAME && tile_copy_mode) {
3668 const int identical_tile_offset =
3669 find_identical_tile(tile_row, tile_col, tile_buffers);
3670
3671 // Indicate a copy-tile by setting the most significant bit.
3672 // The row-offset to copy from is stored in the highest byte.
3673 // remux_tiles will move these around later
3674 if (identical_tile_offset > 0) {
3675 tile_size = 0;
3676 tile_header = identical_tile_offset | 0x80;
3677 tile_header <<= 24;
3678 }
3679 }
3680
3681 mem_put_le32(buf->data, (MEM_VALUE_T)tile_header);
3682 }
3683
3684 *total_size += tile_size;
3685 }
3686 if (!is_last_col) {
3687 uint32_t col_size = *total_size - col_offset - 4;
3688 mem_put_le32(dst + col_offset + lst_obu->tg_hdr_size, col_size);
3689
3690 // Record the maximum tile column size we see.
3691 *max_tile_col_size = AOMMAX(*max_tile_col_size, col_size);
3692 }
3693 }
3694 av1_accumulate_pack_bs_thread_data(cpi, &cpi->td);
3695 }
3696
3697 // Packs information in the obu header for large scale tiles.
pack_large_scale_tiles_in_tg_obus(AV1_COMP * const cpi,uint8_t * const dst,struct aom_write_bit_buffer * saved_wb,int * const largest_tile_id)3698 static INLINE uint32_t pack_large_scale_tiles_in_tg_obus(
3699 AV1_COMP *const cpi, uint8_t *const dst,
3700 struct aom_write_bit_buffer *saved_wb, int *const largest_tile_id) {
3701 AV1_COMMON *const cm = &cpi->common;
3702 const CommonTileParams *const tiles = &cm->tiles;
3703 uint32_t total_size = 0;
3704 unsigned int max_tile_size = 0;
3705 unsigned int max_tile_col_size = 0;
3706 const int have_tiles = tiles->cols * tiles->rows > 1;
3707 uint8_t *data = dst;
3708
3709 LargeTileFrameOBU lst_obu;
3710
3711 total_size +=
3712 init_large_scale_tile_obu_header(cpi, &data, saved_wb, &lst_obu);
3713
3714 write_large_scale_tile_obu(cpi, dst, &lst_obu, largest_tile_id, &total_size,
3715 have_tiles, &max_tile_size, &max_tile_col_size);
3716
3717 write_large_scale_tile_obu_size(tiles, dst, data, saved_wb, &lst_obu,
3718 have_tiles, &total_size, max_tile_size,
3719 max_tile_col_size);
3720
3721 return total_size;
3722 }
3723
3724 // Writes obu, tile group and uncompressed headers to bitstream.
av1_write_obu_tg_tile_headers(AV1_COMP * const cpi,MACROBLOCKD * const xd,PackBSParams * const pack_bs_params,const int tile_idx)3725 void av1_write_obu_tg_tile_headers(AV1_COMP *const cpi, MACROBLOCKD *const xd,
3726 PackBSParams *const pack_bs_params,
3727 const int tile_idx) {
3728 AV1_COMMON *const cm = &cpi->common;
3729 const CommonTileParams *const tiles = &cm->tiles;
3730 int *const curr_tg_hdr_size = &pack_bs_params->curr_tg_hdr_size;
3731 const int tg_size =
3732 (tiles->rows * tiles->cols + cpi->num_tg - 1) / cpi->num_tg;
3733
3734 // Write Tile group, frame and OBU header
3735 // A new tile group begins at this tile. Write the obu header and
3736 // tile group header
3737 const OBU_TYPE obu_type = (cpi->num_tg == 1) ? OBU_FRAME : OBU_TILE_GROUP;
3738 *curr_tg_hdr_size = av1_write_obu_header(
3739 &cpi->ppi->level_params, &cpi->frame_header_count, obu_type,
3740 pack_bs_params->obu_extn_header, pack_bs_params->tile_data_curr);
3741 pack_bs_params->obu_header_size = *curr_tg_hdr_size;
3742
3743 if (cpi->num_tg == 1)
3744 *curr_tg_hdr_size += write_frame_header_obu(
3745 cpi, xd, pack_bs_params->saved_wb,
3746 pack_bs_params->tile_data_curr + *curr_tg_hdr_size, 0);
3747 *curr_tg_hdr_size += write_tile_group_header(
3748 pack_bs_params->tile_data_curr + *curr_tg_hdr_size, tile_idx,
3749 AOMMIN(tile_idx + tg_size - 1, tiles->cols * tiles->rows - 1),
3750 (tiles->log2_rows + tiles->log2_cols), cpi->num_tg > 1);
3751 *pack_bs_params->total_size += *curr_tg_hdr_size;
3752 }
3753
3754 // Pack tile data in the bitstream with tile_group, frame
3755 // and OBU header.
av1_pack_tile_info(AV1_COMP * const cpi,ThreadData * const td,PackBSParams * const pack_bs_params)3756 void av1_pack_tile_info(AV1_COMP *const cpi, ThreadData *const td,
3757 PackBSParams *const pack_bs_params) {
3758 aom_writer mode_bc;
3759 AV1_COMMON *const cm = &cpi->common;
3760 int tile_row = pack_bs_params->tile_row;
3761 int tile_col = pack_bs_params->tile_col;
3762 uint32_t *const total_size = pack_bs_params->total_size;
3763 TileInfo tile_info;
3764 av1_tile_set_col(&tile_info, cm, tile_col);
3765 av1_tile_set_row(&tile_info, cm, tile_row);
3766 mode_bc.allow_update_cdf = 1;
3767 mode_bc.allow_update_cdf =
3768 mode_bc.allow_update_cdf && !cm->features.disable_cdf_update;
3769
3770 unsigned int tile_size;
3771
3772 const int num_planes = av1_num_planes(cm);
3773 av1_reset_loop_restoration(&td->mb.e_mbd, num_planes);
3774
3775 pack_bs_params->buf.data = pack_bs_params->dst + *total_size;
3776
3777 // The last tile of the tile group does not have a header.
3778 if (!pack_bs_params->is_last_tile_in_tg) *total_size += 4;
3779
3780 // Pack tile data
3781 aom_start_encode(&mode_bc, pack_bs_params->dst + *total_size);
3782 write_modes(cpi, td, &tile_info, &mode_bc, tile_row, tile_col);
3783 if (aom_stop_encode(&mode_bc) < 0) {
3784 aom_internal_error(td->mb.e_mbd.error_info, AOM_CODEC_ERROR,
3785 "Error writing modes");
3786 }
3787 tile_size = mode_bc.pos;
3788 assert(tile_size >= AV1_MIN_TILE_SIZE_BYTES);
3789
3790 pack_bs_params->buf.size = tile_size;
3791
3792 // Write tile size
3793 if (!pack_bs_params->is_last_tile_in_tg) {
3794 // size of this tile
3795 mem_put_le32(pack_bs_params->buf.data, tile_size - AV1_MIN_TILE_SIZE_BYTES);
3796 }
3797 }
3798
av1_write_last_tile_info(AV1_COMP * const cpi,const FrameHeaderInfo * fh_info,struct aom_write_bit_buffer * saved_wb,size_t * curr_tg_data_size,uint8_t * curr_tg_start,uint32_t * const total_size,uint8_t ** tile_data_start,int * const largest_tile_id,int * const is_first_tg,uint32_t obu_header_size,uint8_t obu_extn_header)3799 void av1_write_last_tile_info(
3800 AV1_COMP *const cpi, const FrameHeaderInfo *fh_info,
3801 struct aom_write_bit_buffer *saved_wb, size_t *curr_tg_data_size,
3802 uint8_t *curr_tg_start, uint32_t *const total_size,
3803 uint8_t **tile_data_start, int *const largest_tile_id,
3804 int *const is_first_tg, uint32_t obu_header_size, uint8_t obu_extn_header) {
3805 // write current tile group size
3806 const uint32_t obu_payload_size =
3807 (uint32_t)(*curr_tg_data_size) - obu_header_size;
3808 const size_t length_field_size =
3809 av1_obu_memmove(obu_header_size, obu_payload_size, curr_tg_start);
3810 if (av1_write_uleb_obu_size(obu_header_size, obu_payload_size,
3811 curr_tg_start) != AOM_CODEC_OK) {
3812 assert(0);
3813 }
3814 *curr_tg_data_size += (int)length_field_size;
3815 *total_size += (uint32_t)length_field_size;
3816 *tile_data_start += length_field_size;
3817 if (cpi->num_tg == 1) {
3818 // if this tg is combined with the frame header then update saved
3819 // frame header base offset according to length field size
3820 saved_wb->bit_buffer += length_field_size;
3821 }
3822
3823 if (!(*is_first_tg) && cpi->common.features.error_resilient_mode) {
3824 // Make room for a duplicate Frame Header OBU.
3825 memmove(curr_tg_start + fh_info->total_length, curr_tg_start,
3826 *curr_tg_data_size);
3827
3828 // Insert a copy of the Frame Header OBU.
3829 memcpy(curr_tg_start, fh_info->frame_header, fh_info->total_length);
3830
3831 // Force context update tile to be the first tile in error
3832 // resilient mode as the duplicate frame headers will have
3833 // context_update_tile_id set to 0
3834 *largest_tile_id = 0;
3835
3836 // Rewrite the OBU header to change the OBU type to Redundant Frame
3837 // Header.
3838 av1_write_obu_header(&cpi->ppi->level_params, &cpi->frame_header_count,
3839 OBU_REDUNDANT_FRAME_HEADER, obu_extn_header,
3840 &curr_tg_start[fh_info->obu_header_byte_offset]);
3841
3842 *curr_tg_data_size += (int)(fh_info->total_length);
3843 *total_size += (uint32_t)(fh_info->total_length);
3844 }
3845 *is_first_tg = 0;
3846 }
3847
av1_reset_pack_bs_thread_data(ThreadData * const td)3848 void av1_reset_pack_bs_thread_data(ThreadData *const td) {
3849 td->coefficient_size = 0;
3850 td->max_mv_magnitude = 0;
3851 av1_zero(td->interp_filter_selected);
3852 }
3853
av1_accumulate_pack_bs_thread_data(AV1_COMP * const cpi,ThreadData const * td)3854 void av1_accumulate_pack_bs_thread_data(AV1_COMP *const cpi,
3855 ThreadData const *td) {
3856 int do_max_mv_magnitude_update = 1;
3857 cpi->rc.coefficient_size += td->coefficient_size;
3858
3859 // Disable max_mv_magnitude update for parallel frames based on update flag.
3860 if (!cpi->do_frame_data_update) do_max_mv_magnitude_update = 0;
3861
3862 if (cpi->sf.mv_sf.auto_mv_step_size && do_max_mv_magnitude_update)
3863 cpi->mv_search_params.max_mv_magnitude =
3864 AOMMAX(cpi->mv_search_params.max_mv_magnitude, td->max_mv_magnitude);
3865
3866 for (InterpFilter filter = EIGHTTAP_REGULAR; filter < SWITCHABLE; filter++)
3867 cpi->common.cur_frame->interp_filter_selected[filter] +=
3868 td->interp_filter_selected[filter];
3869 }
3870
3871 // Store information related to each default tile in the OBU header.
write_tile_obu(AV1_COMP * const cpi,uint8_t * const dst,uint32_t * total_size,struct aom_write_bit_buffer * saved_wb,uint8_t obu_extn_header,const FrameHeaderInfo * fh_info,int * const largest_tile_id,unsigned int * max_tile_size,uint32_t * const obu_header_size,uint8_t ** tile_data_start)3872 static void write_tile_obu(
3873 AV1_COMP *const cpi, uint8_t *const dst, uint32_t *total_size,
3874 struct aom_write_bit_buffer *saved_wb, uint8_t obu_extn_header,
3875 const FrameHeaderInfo *fh_info, int *const largest_tile_id,
3876 unsigned int *max_tile_size, uint32_t *const obu_header_size,
3877 uint8_t **tile_data_start) {
3878 AV1_COMMON *const cm = &cpi->common;
3879 MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
3880 const CommonTileParams *const tiles = &cm->tiles;
3881 const int tile_cols = tiles->cols;
3882 const int tile_rows = tiles->rows;
3883 // Fixed size tile groups for the moment
3884 const int num_tg_hdrs = cpi->num_tg;
3885 const int tg_size = (tile_rows * tile_cols + num_tg_hdrs - 1) / num_tg_hdrs;
3886 int tile_count = 0;
3887 size_t curr_tg_data_size = 0;
3888 uint8_t *tile_data_curr = dst;
3889 int new_tg = 1;
3890 int is_first_tg = 1;
3891
3892 av1_reset_pack_bs_thread_data(&cpi->td);
3893 for (int tile_row = 0; tile_row < tile_rows; tile_row++) {
3894 for (int tile_col = 0; tile_col < tile_cols; tile_col++) {
3895 const int tile_idx = tile_row * tile_cols + tile_col;
3896 TileDataEnc *this_tile = &cpi->tile_data[tile_idx];
3897
3898 int is_last_tile_in_tg = 0;
3899 if (new_tg) {
3900 tile_data_curr = dst + *total_size;
3901 tile_count = 0;
3902 }
3903 tile_count++;
3904
3905 if (tile_count == tg_size || tile_idx == (tile_cols * tile_rows - 1))
3906 is_last_tile_in_tg = 1;
3907
3908 xd->tile_ctx = &this_tile->tctx;
3909
3910 // PackBSParams stores all parameters required to pack tile and header
3911 // info.
3912 PackBSParams pack_bs_params;
3913 pack_bs_params.dst = dst;
3914 pack_bs_params.curr_tg_hdr_size = 0;
3915 pack_bs_params.is_last_tile_in_tg = is_last_tile_in_tg;
3916 pack_bs_params.new_tg = new_tg;
3917 pack_bs_params.obu_extn_header = obu_extn_header;
3918 pack_bs_params.obu_header_size = 0;
3919 pack_bs_params.saved_wb = saved_wb;
3920 pack_bs_params.tile_col = tile_col;
3921 pack_bs_params.tile_row = tile_row;
3922 pack_bs_params.tile_data_curr = tile_data_curr;
3923 pack_bs_params.total_size = total_size;
3924
3925 if (new_tg)
3926 av1_write_obu_tg_tile_headers(cpi, xd, &pack_bs_params, tile_idx);
3927
3928 av1_pack_tile_info(cpi, &cpi->td, &pack_bs_params);
3929
3930 if (new_tg) {
3931 curr_tg_data_size = pack_bs_params.curr_tg_hdr_size;
3932 *tile_data_start += pack_bs_params.curr_tg_hdr_size;
3933 *obu_header_size = pack_bs_params.obu_header_size;
3934 new_tg = 0;
3935 }
3936 if (is_last_tile_in_tg) new_tg = 1;
3937
3938 curr_tg_data_size +=
3939 (pack_bs_params.buf.size + (is_last_tile_in_tg ? 0 : 4));
3940
3941 if (pack_bs_params.buf.size > *max_tile_size) {
3942 *largest_tile_id = tile_idx;
3943 *max_tile_size = (unsigned int)pack_bs_params.buf.size;
3944 }
3945
3946 if (is_last_tile_in_tg)
3947 av1_write_last_tile_info(cpi, fh_info, saved_wb, &curr_tg_data_size,
3948 tile_data_curr, total_size, tile_data_start,
3949 largest_tile_id, &is_first_tg,
3950 *obu_header_size, obu_extn_header);
3951 *total_size += (uint32_t)pack_bs_params.buf.size;
3952 }
3953 }
3954 av1_accumulate_pack_bs_thread_data(cpi, &cpi->td);
3955 }
3956
3957 // Write total buffer size and related information into the OBU header for
3958 // default tile case.
write_tile_obu_size(AV1_COMP * const cpi,uint8_t * const dst,struct aom_write_bit_buffer * saved_wb,int largest_tile_id,uint32_t * const total_size,unsigned int max_tile_size,uint32_t obu_header_size,uint8_t * tile_data_start)3959 static void write_tile_obu_size(AV1_COMP *const cpi, uint8_t *const dst,
3960 struct aom_write_bit_buffer *saved_wb,
3961 int largest_tile_id, uint32_t *const total_size,
3962 unsigned int max_tile_size,
3963 uint32_t obu_header_size,
3964 uint8_t *tile_data_start) {
3965 const CommonTileParams *const tiles = &cpi->common.tiles;
3966
3967 // Fill in context_update_tile_id indicating the tile to use for the
3968 // cdf update. The encoder currently sets it to the largest tile
3969 // (but is up to the encoder)
3970 aom_wb_overwrite_literal(saved_wb, largest_tile_id,
3971 (tiles->log2_cols + tiles->log2_rows));
3972 // If more than one tile group. tile_size_bytes takes the default value 4
3973 // and does not need to be set. For a single tile group it is set in the
3974 // section below.
3975 if (cpi->num_tg != 1) return;
3976 int tile_size_bytes = 4, unused;
3977 const uint32_t tile_data_offset = (uint32_t)(tile_data_start - dst);
3978 const uint32_t tile_data_size = *total_size - tile_data_offset;
3979
3980 *total_size = remux_tiles(tiles, tile_data_start, tile_data_size,
3981 max_tile_size, 0, &tile_size_bytes, &unused);
3982 *total_size += tile_data_offset;
3983 assert(tile_size_bytes >= 1 && tile_size_bytes <= 4);
3984
3985 aom_wb_overwrite_literal(saved_wb, tile_size_bytes - 1, 2);
3986
3987 // Update the OBU length if remux_tiles() reduced the size.
3988 uint64_t payload_size;
3989 size_t length_field_size;
3990 int res =
3991 aom_uleb_decode(dst + obu_header_size, *total_size - obu_header_size,
3992 &payload_size, &length_field_size);
3993 assert(res == 0);
3994 (void)res;
3995
3996 const uint64_t new_payload_size =
3997 *total_size - obu_header_size - length_field_size;
3998 if (new_payload_size != payload_size) {
3999 size_t new_length_field_size;
4000 res = aom_uleb_encode(new_payload_size, length_field_size,
4001 dst + obu_header_size, &new_length_field_size);
4002 assert(res == 0);
4003 if (new_length_field_size < length_field_size) {
4004 const size_t src_offset = obu_header_size + length_field_size;
4005 const size_t dst_offset = obu_header_size + new_length_field_size;
4006 memmove(dst + dst_offset, dst + src_offset, (size_t)payload_size);
4007 *total_size -= (int)(length_field_size - new_length_field_size);
4008 }
4009 }
4010 }
4011
4012 // As per the experiments, single-thread bitstream packing is better for
4013 // frames with a smaller bitstream size. This behavior is due to setup time
4014 // overhead of multithread function would be more than that of time required
4015 // to pack the smaller bitstream of such frames. This function computes the
4016 // number of required number of workers based on setup time overhead and job
4017 // dispatch time overhead for given tiles and available workers.
calc_pack_bs_mt_workers(const TileDataEnc * tile_data,int num_tiles,int avail_workers,bool pack_bs_mt_enabled)4018 int calc_pack_bs_mt_workers(const TileDataEnc *tile_data, int num_tiles,
4019 int avail_workers, bool pack_bs_mt_enabled) {
4020 if (!pack_bs_mt_enabled) return 1;
4021
4022 uint64_t frame_abs_sum_level = 0;
4023
4024 for (int idx = 0; idx < num_tiles; idx++)
4025 frame_abs_sum_level += tile_data[idx].abs_sum_level;
4026
4027 int ideal_num_workers = 1;
4028 const float job_disp_time_const = (float)num_tiles * JOB_DISP_TIME_OH_CONST;
4029 float max_sum = 0.0;
4030
4031 for (int num_workers = avail_workers; num_workers > 1; num_workers--) {
4032 const float fas_per_worker_const =
4033 ((float)(num_workers - 1) / num_workers) * frame_abs_sum_level;
4034 const float setup_time_const = (float)num_workers * SETUP_TIME_OH_CONST;
4035 const float this_sum = fas_per_worker_const - setup_time_const -
4036 job_disp_time_const / num_workers;
4037
4038 if (this_sum > max_sum) {
4039 max_sum = this_sum;
4040 ideal_num_workers = num_workers;
4041 }
4042 }
4043 return ideal_num_workers;
4044 }
4045
pack_tiles_in_tg_obus(AV1_COMP * const cpi,uint8_t * const dst,struct aom_write_bit_buffer * saved_wb,uint8_t obu_extension_header,const FrameHeaderInfo * fh_info,int * const largest_tile_id)4046 static INLINE uint32_t pack_tiles_in_tg_obus(
4047 AV1_COMP *const cpi, uint8_t *const dst,
4048 struct aom_write_bit_buffer *saved_wb, uint8_t obu_extension_header,
4049 const FrameHeaderInfo *fh_info, int *const largest_tile_id) {
4050 const CommonTileParams *const tiles = &cpi->common.tiles;
4051 uint32_t total_size = 0;
4052 unsigned int max_tile_size = 0;
4053 uint32_t obu_header_size = 0;
4054 uint8_t *tile_data_start = dst;
4055 const int tile_cols = tiles->cols;
4056 const int tile_rows = tiles->rows;
4057 const int num_tiles = tile_rows * tile_cols;
4058
4059 const int num_workers = calc_pack_bs_mt_workers(
4060 cpi->tile_data, num_tiles, cpi->mt_info.num_mod_workers[MOD_PACK_BS],
4061 cpi->mt_info.pack_bs_mt_enabled);
4062
4063 if (num_workers > 1) {
4064 av1_write_tile_obu_mt(cpi, dst, &total_size, saved_wb, obu_extension_header,
4065 fh_info, largest_tile_id, &max_tile_size,
4066 &obu_header_size, &tile_data_start, num_workers);
4067 } else {
4068 write_tile_obu(cpi, dst, &total_size, saved_wb, obu_extension_header,
4069 fh_info, largest_tile_id, &max_tile_size, &obu_header_size,
4070 &tile_data_start);
4071 }
4072
4073 if (num_tiles > 1)
4074 write_tile_obu_size(cpi, dst, saved_wb, *largest_tile_id, &total_size,
4075 max_tile_size, obu_header_size, tile_data_start);
4076 return total_size;
4077 }
4078
write_tiles_in_tg_obus(AV1_COMP * const cpi,uint8_t * const dst,struct aom_write_bit_buffer * saved_wb,uint8_t obu_extension_header,const FrameHeaderInfo * fh_info,int * const largest_tile_id)4079 static uint32_t write_tiles_in_tg_obus(AV1_COMP *const cpi, uint8_t *const dst,
4080 struct aom_write_bit_buffer *saved_wb,
4081 uint8_t obu_extension_header,
4082 const FrameHeaderInfo *fh_info,
4083 int *const largest_tile_id) {
4084 AV1_COMMON *const cm = &cpi->common;
4085 const CommonTileParams *const tiles = &cm->tiles;
4086 *largest_tile_id = 0;
4087
4088 // Select the coding strategy (temporal or spatial)
4089 if (cm->seg.enabled && cm->seg.update_map) {
4090 if (cm->features.primary_ref_frame == PRIMARY_REF_NONE) {
4091 cm->seg.temporal_update = 0;
4092 } else {
4093 cm->seg.temporal_update = 1;
4094 if (cpi->td.rd_counts.seg_tmp_pred_cost[0] <
4095 cpi->td.rd_counts.seg_tmp_pred_cost[1])
4096 cm->seg.temporal_update = 0;
4097 }
4098 }
4099
4100 if (tiles->large_scale)
4101 return pack_large_scale_tiles_in_tg_obus(cpi, dst, saved_wb,
4102 largest_tile_id);
4103
4104 return pack_tiles_in_tg_obus(cpi, dst, saved_wb, obu_extension_header,
4105 fh_info, largest_tile_id);
4106 }
4107
av1_write_metadata_obu(const aom_metadata_t * metadata,uint8_t * const dst)4108 static size_t av1_write_metadata_obu(const aom_metadata_t *metadata,
4109 uint8_t *const dst) {
4110 size_t coded_metadata_size = 0;
4111 const uint64_t metadata_type = (uint64_t)metadata->type;
4112 if (aom_uleb_encode(metadata_type, sizeof(metadata_type), dst,
4113 &coded_metadata_size) != 0) {
4114 return 0;
4115 }
4116 memcpy(dst + coded_metadata_size, metadata->payload, metadata->sz);
4117 // Add trailing bits.
4118 dst[coded_metadata_size + metadata->sz] = 0x80;
4119 return (uint32_t)(coded_metadata_size + metadata->sz + 1);
4120 }
4121
av1_write_metadata_array(AV1_COMP * const cpi,uint8_t * dst)4122 static size_t av1_write_metadata_array(AV1_COMP *const cpi, uint8_t *dst) {
4123 if (!cpi->source) return 0;
4124 AV1_COMMON *const cm = &cpi->common;
4125 aom_metadata_array_t *arr = cpi->source->metadata;
4126 if (!arr) return 0;
4127 size_t obu_header_size = 0;
4128 size_t obu_payload_size = 0;
4129 size_t total_bytes_written = 0;
4130 size_t length_field_size = 0;
4131 for (size_t i = 0; i < arr->sz; i++) {
4132 aom_metadata_t *current_metadata = arr->metadata_array[i];
4133 if (current_metadata && current_metadata->payload) {
4134 if ((cm->current_frame.frame_type == KEY_FRAME &&
4135 current_metadata->insert_flag == AOM_MIF_KEY_FRAME) ||
4136 (cm->current_frame.frame_type != KEY_FRAME &&
4137 current_metadata->insert_flag == AOM_MIF_NON_KEY_FRAME) ||
4138 current_metadata->insert_flag == AOM_MIF_ANY_FRAME) {
4139 obu_header_size = av1_write_obu_header(&cpi->ppi->level_params,
4140 &cpi->frame_header_count,
4141 OBU_METADATA, 0, dst);
4142 obu_payload_size =
4143 av1_write_metadata_obu(current_metadata, dst + obu_header_size);
4144 length_field_size =
4145 av1_obu_memmove(obu_header_size, obu_payload_size, dst);
4146 if (av1_write_uleb_obu_size(obu_header_size, obu_payload_size, dst) ==
4147 AOM_CODEC_OK) {
4148 const size_t obu_size = obu_header_size + obu_payload_size;
4149 dst += obu_size + length_field_size;
4150 total_bytes_written += obu_size + length_field_size;
4151 } else {
4152 aom_internal_error(cpi->common.error, AOM_CODEC_ERROR,
4153 "Error writing metadata OBU size");
4154 }
4155 }
4156 }
4157 }
4158 return total_bytes_written;
4159 }
4160
av1_pack_bitstream(AV1_COMP * const cpi,uint8_t * dst,size_t * size,int * const largest_tile_id)4161 int av1_pack_bitstream(AV1_COMP *const cpi, uint8_t *dst, size_t *size,
4162 int *const largest_tile_id) {
4163 uint8_t *data = dst;
4164 uint32_t data_size;
4165 AV1_COMMON *const cm = &cpi->common;
4166 AV1LevelParams *const level_params = &cpi->ppi->level_params;
4167 uint32_t obu_header_size = 0;
4168 uint32_t obu_payload_size = 0;
4169 FrameHeaderInfo fh_info = { NULL, 0, 0 };
4170 const uint8_t obu_extension_header =
4171 cm->temporal_layer_id << 5 | cm->spatial_layer_id << 3 | 0;
4172
4173 // If no non-zero delta_q has been used, reset delta_q_present_flag
4174 if (cm->delta_q_info.delta_q_present_flag && cpi->deltaq_used == 0) {
4175 cm->delta_q_info.delta_q_present_flag = 0;
4176 }
4177
4178 #if CONFIG_BITSTREAM_DEBUG
4179 bitstream_queue_reset_write();
4180 #endif
4181
4182 cpi->frame_header_count = 0;
4183
4184 // The TD is now written outside the frame encode loop
4185
4186 // write sequence header obu at each key frame or intra_only frame,
4187 // preceded by 4-byte size
4188 if (cm->current_frame.frame_type == INTRA_ONLY_FRAME ||
4189 cm->current_frame.frame_type == KEY_FRAME) {
4190 obu_header_size = av1_write_obu_header(
4191 level_params, &cpi->frame_header_count, OBU_SEQUENCE_HEADER, 0, data);
4192 obu_payload_size =
4193 av1_write_sequence_header_obu(cm->seq_params, data + obu_header_size);
4194 const size_t length_field_size =
4195 av1_obu_memmove(obu_header_size, obu_payload_size, data);
4196 if (av1_write_uleb_obu_size(obu_header_size, obu_payload_size, data) !=
4197 AOM_CODEC_OK) {
4198 return AOM_CODEC_ERROR;
4199 }
4200
4201 data += obu_header_size + obu_payload_size + length_field_size;
4202 }
4203
4204 // write metadata obus before the frame obu that has the show_frame flag set
4205 if (cm->show_frame) data += av1_write_metadata_array(cpi, data);
4206
4207 const int write_frame_header =
4208 (cpi->num_tg > 1 || encode_show_existing_frame(cm));
4209 struct aom_write_bit_buffer saved_wb = { NULL, 0 };
4210 size_t length_field = 0;
4211 if (write_frame_header) {
4212 // Write Frame Header OBU.
4213 fh_info.frame_header = data;
4214 obu_header_size =
4215 av1_write_obu_header(level_params, &cpi->frame_header_count,
4216 OBU_FRAME_HEADER, obu_extension_header, data);
4217 obu_payload_size = write_frame_header_obu(cpi, &cpi->td.mb.e_mbd, &saved_wb,
4218 data + obu_header_size, 1);
4219
4220 length_field = av1_obu_memmove(obu_header_size, obu_payload_size, data);
4221 if (av1_write_uleb_obu_size(obu_header_size, obu_payload_size, data) !=
4222 AOM_CODEC_OK) {
4223 return AOM_CODEC_ERROR;
4224 }
4225
4226 fh_info.obu_header_byte_offset = 0;
4227 fh_info.total_length = obu_header_size + obu_payload_size + length_field;
4228 data += fh_info.total_length;
4229 }
4230
4231 if (encode_show_existing_frame(cm)) {
4232 data_size = 0;
4233 } else {
4234 // Since length_field is determined adaptively after frame header
4235 // encoding, saved_wb must be adjusted accordingly.
4236 if (saved_wb.bit_buffer != NULL) {
4237 saved_wb.bit_buffer += length_field;
4238 }
4239
4240 // Each tile group obu will be preceded by 4-byte size of the tile group
4241 // obu
4242 data_size = write_tiles_in_tg_obus(
4243 cpi, data, &saved_wb, obu_extension_header, &fh_info, largest_tile_id);
4244 }
4245 data += data_size;
4246 *size = data - dst;
4247 return AOM_CODEC_OK;
4248 }
4249