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 <limits.h>
13 #include <float.h>
14 #include <math.h>
15 #include <stdio.h>
16 #include <time.h>
17 #include <stdlib.h>
18
19 #include "config/aom_config.h"
20 #include "config/aom_dsp_rtcd.h"
21
22 #include "aom/aomcx.h"
23
24 #if CONFIG_DENOISE
25 #include "aom_dsp/grain_table.h"
26 #include "aom_dsp/noise_util.h"
27 #include "aom_dsp/noise_model.h"
28 #endif
29 #include "aom_dsp/psnr.h"
30 #if CONFIG_INTERNAL_STATS
31 #include "aom_dsp/ssim.h"
32 #endif
33 #include "aom_ports/aom_timer.h"
34 #include "aom_ports/mem.h"
35 #include "aom_ports/system_state.h"
36 #include "aom_scale/aom_scale.h"
37 #if CONFIG_BITSTREAM_DEBUG
38 #include "aom_util/debug_util.h"
39 #endif // CONFIG_BITSTREAM_DEBUG
40
41 #include "av1/common/alloccommon.h"
42 #include "av1/common/filter.h"
43 #include "av1/common/idct.h"
44 #include "av1/common/reconinter.h"
45 #include "av1/common/reconintra.h"
46 #include "av1/common/resize.h"
47 #include "av1/common/tile_common.h"
48
49 #include "av1/encoder/aq_complexity.h"
50 #include "av1/encoder/aq_cyclicrefresh.h"
51 #include "av1/encoder/aq_variance.h"
52 #include "av1/encoder/bitstream.h"
53 #include "av1/encoder/context_tree.h"
54 #include "av1/encoder/encodeframe.h"
55 #include "av1/encoder/encodemv.h"
56 #include "av1/encoder/encode_strategy.h"
57 #include "av1/encoder/encoder.h"
58 #include "av1/encoder/encoder_alloc.h"
59 #include "av1/encoder/encoder_utils.h"
60 #include "av1/encoder/encodetxb.h"
61 #include "av1/encoder/ethread.h"
62 #include "av1/encoder/firstpass.h"
63 #include "av1/encoder/hash_motion.h"
64 #include "av1/encoder/intra_mode_search.h"
65 #include "av1/encoder/mv_prec.h"
66 #include "av1/encoder/pass2_strategy.h"
67 #include "av1/encoder/pickcdef.h"
68 #include "av1/encoder/picklpf.h"
69 #include "av1/encoder/pickrst.h"
70 #include "av1/encoder/random.h"
71 #include "av1/encoder/ratectrl.h"
72 #include "av1/encoder/rc_utils.h"
73 #include "av1/encoder/rd.h"
74 #include "av1/encoder/rdopt.h"
75 #include "av1/encoder/segmentation.h"
76 #include "av1/encoder/speed_features.h"
77 #include "av1/encoder/superres_scale.h"
78 #include "av1/encoder/tpl_model.h"
79 #include "av1/encoder/reconinter_enc.h"
80 #include "av1/encoder/var_based_part.h"
81
82 #define DEFAULT_EXPLICIT_ORDER_HINT_BITS 7
83
84 #if CONFIG_ENTROPY_STATS
85 FRAME_COUNTS aggregate_fc;
86 #endif // CONFIG_ENTROPY_STATS
87
88 // #define OUTPUT_YUV_REC
89 #ifdef OUTPUT_YUV_REC
90 FILE *yuv_rec_file;
91 #define FILE_NAME_LEN 100
92 #endif
93
94 #ifdef OUTPUT_YUV_DENOISED
95 FILE *yuv_denoised_file = NULL;
96 #endif
97
Scale2Ratio(AOM_SCALING mode,int * hr,int * hs)98 static INLINE void Scale2Ratio(AOM_SCALING mode, int *hr, int *hs) {
99 switch (mode) {
100 case NORMAL:
101 *hr = 1;
102 *hs = 1;
103 break;
104 case FOURFIVE:
105 *hr = 4;
106 *hs = 5;
107 break;
108 case THREEFIVE:
109 *hr = 3;
110 *hs = 5;
111 break;
112 case THREEFOUR:
113 *hr = 3;
114 *hs = 4;
115 break;
116 case ONEFOUR:
117 *hr = 1;
118 *hs = 4;
119 break;
120 case ONEEIGHT:
121 *hr = 1;
122 *hs = 8;
123 break;
124 case ONETWO:
125 *hr = 1;
126 *hs = 2;
127 break;
128 default:
129 *hr = 1;
130 *hs = 1;
131 assert(0);
132 break;
133 }
134 }
135
av1_set_active_map(AV1_COMP * cpi,unsigned char * new_map_16x16,int rows,int cols)136 int av1_set_active_map(AV1_COMP *cpi, unsigned char *new_map_16x16, int rows,
137 int cols) {
138 const CommonModeInfoParams *const mi_params = &cpi->common.mi_params;
139 if (rows == mi_params->mb_rows && cols == mi_params->mb_cols) {
140 unsigned char *const active_map_8x8 = cpi->active_map.map;
141 const int mi_rows = mi_params->mi_rows;
142 const int mi_cols = mi_params->mi_cols;
143 const int row_scale = mi_size_high[BLOCK_16X16] == 2 ? 1 : 2;
144 const int col_scale = mi_size_wide[BLOCK_16X16] == 2 ? 1 : 2;
145 cpi->active_map.update = 1;
146 if (new_map_16x16) {
147 int r, c;
148 for (r = 0; r < mi_rows; ++r) {
149 for (c = 0; c < mi_cols; ++c) {
150 active_map_8x8[r * mi_cols + c] =
151 new_map_16x16[(r >> row_scale) * cols + (c >> col_scale)]
152 ? AM_SEGMENT_ID_ACTIVE
153 : AM_SEGMENT_ID_INACTIVE;
154 }
155 }
156 cpi->active_map.enabled = 1;
157 } else {
158 cpi->active_map.enabled = 0;
159 }
160 return 0;
161 } else {
162 return -1;
163 }
164 }
165
av1_get_active_map(AV1_COMP * cpi,unsigned char * new_map_16x16,int rows,int cols)166 int av1_get_active_map(AV1_COMP *cpi, unsigned char *new_map_16x16, int rows,
167 int cols) {
168 const CommonModeInfoParams *const mi_params = &cpi->common.mi_params;
169 if (rows == mi_params->mb_rows && cols == mi_params->mb_cols &&
170 new_map_16x16) {
171 unsigned char *const seg_map_8x8 = cpi->enc_seg.map;
172 const int mi_rows = mi_params->mi_rows;
173 const int mi_cols = mi_params->mi_cols;
174 const int row_scale = mi_size_high[BLOCK_16X16] == 2 ? 1 : 2;
175 const int col_scale = mi_size_wide[BLOCK_16X16] == 2 ? 1 : 2;
176
177 memset(new_map_16x16, !cpi->active_map.enabled, rows * cols);
178 if (cpi->active_map.enabled) {
179 int r, c;
180 for (r = 0; r < mi_rows; ++r) {
181 for (c = 0; c < mi_cols; ++c) {
182 // Cyclic refresh segments are considered active despite not having
183 // AM_SEGMENT_ID_ACTIVE
184 new_map_16x16[(r >> row_scale) * cols + (c >> col_scale)] |=
185 seg_map_8x8[r * mi_cols + c] != AM_SEGMENT_ID_INACTIVE;
186 }
187 }
188 }
189 return 0;
190 } else {
191 return -1;
192 }
193 }
194
av1_initialize_enc(void)195 void av1_initialize_enc(void) {
196 av1_rtcd();
197 aom_dsp_rtcd();
198 aom_scale_rtcd();
199 av1_init_intra_predictors();
200 av1_init_me_luts();
201 av1_rc_init_minq_luts();
202 av1_init_wedge_masks();
203 }
204
update_reference_segmentation_map(AV1_COMP * cpi)205 static void update_reference_segmentation_map(AV1_COMP *cpi) {
206 AV1_COMMON *const cm = &cpi->common;
207 const CommonModeInfoParams *const mi_params = &cm->mi_params;
208 MB_MODE_INFO **mi_4x4_ptr = mi_params->mi_grid_base;
209 uint8_t *cache_ptr = cm->cur_frame->seg_map;
210
211 for (int row = 0; row < mi_params->mi_rows; row++) {
212 MB_MODE_INFO **mi_4x4 = mi_4x4_ptr;
213 uint8_t *cache = cache_ptr;
214 for (int col = 0; col < mi_params->mi_cols; col++, mi_4x4++, cache++)
215 cache[0] = mi_4x4[0]->segment_id;
216 mi_4x4_ptr += mi_params->mi_stride;
217 cache_ptr += mi_params->mi_cols;
218 }
219 }
220
av1_new_framerate(AV1_COMP * cpi,double framerate)221 void av1_new_framerate(AV1_COMP *cpi, double framerate) {
222 cpi->framerate = framerate < 0.1 ? 30 : framerate;
223 av1_rc_update_framerate(cpi, cpi->common.width, cpi->common.height);
224 }
225
av1_get_compression_ratio(const AV1_COMMON * const cm,size_t encoded_frame_size)226 double av1_get_compression_ratio(const AV1_COMMON *const cm,
227 size_t encoded_frame_size) {
228 const int upscaled_width = cm->superres_upscaled_width;
229 const int height = cm->height;
230 const int luma_pic_size = upscaled_width * height;
231 const SequenceHeader *const seq_params = &cm->seq_params;
232 const BITSTREAM_PROFILE profile = seq_params->profile;
233 const int pic_size_profile_factor =
234 profile == PROFILE_0 ? 15 : (profile == PROFILE_1 ? 30 : 36);
235 encoded_frame_size =
236 (encoded_frame_size > 129 ? encoded_frame_size - 128 : 1);
237 const size_t uncompressed_frame_size =
238 (luma_pic_size * pic_size_profile_factor) >> 3;
239 return uncompressed_frame_size / (double)encoded_frame_size;
240 }
241
set_tile_info(AV1_COMMON * const cm,const TileConfig * const tile_cfg)242 static void set_tile_info(AV1_COMMON *const cm,
243 const TileConfig *const tile_cfg) {
244 const CommonModeInfoParams *const mi_params = &cm->mi_params;
245 const SequenceHeader *const seq_params = &cm->seq_params;
246 CommonTileParams *const tiles = &cm->tiles;
247 int i, start_sb;
248
249 av1_get_tile_limits(cm);
250
251 // configure tile columns
252 if (tile_cfg->tile_width_count == 0 || tile_cfg->tile_height_count == 0) {
253 tiles->uniform_spacing = 1;
254 tiles->log2_cols = AOMMAX(tile_cfg->tile_columns, tiles->min_log2_cols);
255 tiles->log2_cols = AOMMIN(tiles->log2_cols, tiles->max_log2_cols);
256 } else {
257 int mi_cols =
258 ALIGN_POWER_OF_TWO(mi_params->mi_cols, seq_params->mib_size_log2);
259 int sb_cols = mi_cols >> seq_params->mib_size_log2;
260 int size_sb, j = 0;
261 tiles->uniform_spacing = 0;
262 for (i = 0, start_sb = 0; start_sb < sb_cols && i < MAX_TILE_COLS; i++) {
263 tiles->col_start_sb[i] = start_sb;
264 size_sb = tile_cfg->tile_widths[j++];
265 if (j >= tile_cfg->tile_width_count) j = 0;
266 start_sb += AOMMIN(size_sb, tiles->max_width_sb);
267 }
268 tiles->cols = i;
269 tiles->col_start_sb[i] = sb_cols;
270 }
271 av1_calculate_tile_cols(seq_params, mi_params->mi_rows, mi_params->mi_cols,
272 tiles);
273
274 // configure tile rows
275 if (tiles->uniform_spacing) {
276 tiles->log2_rows = AOMMAX(tile_cfg->tile_rows, tiles->min_log2_rows);
277 tiles->log2_rows = AOMMIN(tiles->log2_rows, tiles->max_log2_rows);
278 } else {
279 int mi_rows =
280 ALIGN_POWER_OF_TWO(mi_params->mi_rows, seq_params->mib_size_log2);
281 int sb_rows = mi_rows >> seq_params->mib_size_log2;
282 int size_sb, j = 0;
283 for (i = 0, start_sb = 0; start_sb < sb_rows && i < MAX_TILE_ROWS; i++) {
284 tiles->row_start_sb[i] = start_sb;
285 size_sb = tile_cfg->tile_heights[j++];
286 if (j >= tile_cfg->tile_height_count) j = 0;
287 start_sb += AOMMIN(size_sb, tiles->max_height_sb);
288 }
289 tiles->rows = i;
290 tiles->row_start_sb[i] = sb_rows;
291 }
292 av1_calculate_tile_rows(seq_params, mi_params->mi_rows, tiles);
293 }
294
av1_update_frame_size(AV1_COMP * cpi)295 void av1_update_frame_size(AV1_COMP *cpi) {
296 AV1_COMMON *const cm = &cpi->common;
297 MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
298
299 // We need to reallocate the context buffers here in case we need more mis.
300 if (av1_alloc_context_buffers(cm, cm->width, cm->height)) {
301 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
302 "Failed to allocate context buffers");
303 }
304 av1_init_mi_buffers(&cm->mi_params);
305
306 av1_init_macroblockd(cm, xd);
307
308 if (!is_stat_generation_stage(cpi))
309 alloc_context_buffers_ext(cm, &cpi->mbmi_ext_info);
310
311 if (!cpi->seq_params_locked)
312 set_sb_size(&cm->seq_params, av1_select_sb_size(cpi));
313
314 set_tile_info(cm, &cpi->oxcf.tile_cfg);
315 }
316
does_level_match(int width,int height,double fps,int lvl_width,int lvl_height,double lvl_fps,int lvl_dim_mult)317 static INLINE int does_level_match(int width, int height, double fps,
318 int lvl_width, int lvl_height,
319 double lvl_fps, int lvl_dim_mult) {
320 const int64_t lvl_luma_pels = lvl_width * lvl_height;
321 const double lvl_display_sample_rate = lvl_luma_pels * lvl_fps;
322 const int64_t luma_pels = width * height;
323 const double display_sample_rate = luma_pels * fps;
324 return luma_pels <= lvl_luma_pels &&
325 display_sample_rate <= lvl_display_sample_rate &&
326 width <= lvl_width * lvl_dim_mult &&
327 height <= lvl_height * lvl_dim_mult;
328 }
329
set_bitstream_level_tier(SequenceHeader * seq,AV1_COMMON * cm,int width,int height,double init_framerate)330 static void set_bitstream_level_tier(SequenceHeader *seq, AV1_COMMON *cm,
331 int width, int height,
332 double init_framerate) {
333 // TODO(any): This is a placeholder function that only addresses dimensions
334 // and max display sample rates.
335 // Need to add checks for max bit rate, max decoded luma sample rate, header
336 // rate, etc. that are not covered by this function.
337 AV1_LEVEL level = SEQ_LEVEL_MAX;
338 if (does_level_match(width, height, init_framerate, 512, 288, 30.0, 4)) {
339 level = SEQ_LEVEL_2_0;
340 } else if (does_level_match(width, height, init_framerate, 704, 396, 30.0,
341 4)) {
342 level = SEQ_LEVEL_2_1;
343 } else if (does_level_match(width, height, init_framerate, 1088, 612, 30.0,
344 4)) {
345 level = SEQ_LEVEL_3_0;
346 } else if (does_level_match(width, height, init_framerate, 1376, 774, 30.0,
347 4)) {
348 level = SEQ_LEVEL_3_1;
349 } else if (does_level_match(width, height, init_framerate, 2048, 1152, 30.0,
350 3)) {
351 level = SEQ_LEVEL_4_0;
352 } else if (does_level_match(width, height, init_framerate, 2048, 1152, 60.0,
353 3)) {
354 level = SEQ_LEVEL_4_1;
355 } else if (does_level_match(width, height, init_framerate, 4096, 2176, 30.0,
356 2)) {
357 level = SEQ_LEVEL_5_0;
358 } else if (does_level_match(width, height, init_framerate, 4096, 2176, 60.0,
359 2)) {
360 level = SEQ_LEVEL_5_1;
361 } else if (does_level_match(width, height, init_framerate, 4096, 2176, 120.0,
362 2)) {
363 level = SEQ_LEVEL_5_2;
364 } else if (does_level_match(width, height, init_framerate, 8192, 4352, 30.0,
365 2)) {
366 level = SEQ_LEVEL_6_0;
367 } else if (does_level_match(width, height, init_framerate, 8192, 4352, 60.0,
368 2)) {
369 level = SEQ_LEVEL_6_1;
370 } else if (does_level_match(width, height, init_framerate, 8192, 4352, 120.0,
371 2)) {
372 level = SEQ_LEVEL_6_2;
373 }
374
375 SequenceHeader *const seq_params = &cm->seq_params;
376 for (int i = 0; i < MAX_NUM_OPERATING_POINTS; ++i) {
377 seq->seq_level_idx[i] = level;
378 // Set the maximum parameters for bitrate and buffer size for this profile,
379 // level, and tier
380 seq_params->op_params[i].bitrate = av1_max_level_bitrate(
381 cm->seq_params.profile, seq->seq_level_idx[i], seq->tier[i]);
382 // Level with seq_level_idx = 31 returns a high "dummy" bitrate to pass the
383 // check
384 if (seq_params->op_params[i].bitrate == 0)
385 aom_internal_error(
386 &cm->error, AOM_CODEC_UNSUP_BITSTREAM,
387 "AV1 does not support this combination of profile, level, and tier.");
388 // Buffer size in bits/s is bitrate in bits/s * 1 s
389 seq_params->op_params[i].buffer_size = seq_params->op_params[i].bitrate;
390 }
391 }
392
av1_init_seq_coding_tools(SequenceHeader * seq,AV1_COMMON * cm,const AV1EncoderConfig * oxcf,int use_svc)393 void av1_init_seq_coding_tools(SequenceHeader *seq, AV1_COMMON *cm,
394 const AV1EncoderConfig *oxcf, int use_svc) {
395 const FrameDimensionCfg *const frm_dim_cfg = &oxcf->frm_dim_cfg;
396 const ToolCfg *const tool_cfg = &oxcf->tool_cfg;
397
398 seq->still_picture =
399 !tool_cfg->force_video_mode && (oxcf->input_cfg.limit == 1);
400 seq->reduced_still_picture_hdr =
401 seq->still_picture && !tool_cfg->full_still_picture_hdr;
402 seq->force_screen_content_tools = (oxcf->mode == REALTIME) ? 0 : 2;
403 seq->force_integer_mv = 2;
404 seq->order_hint_info.enable_order_hint = tool_cfg->enable_order_hint;
405 seq->frame_id_numbers_present_flag =
406 !seq->reduced_still_picture_hdr &&
407 !oxcf->tile_cfg.enable_large_scale_tile &&
408 tool_cfg->error_resilient_mode && !use_svc;
409 if (seq->reduced_still_picture_hdr) {
410 seq->order_hint_info.enable_order_hint = 0;
411 seq->force_screen_content_tools = 2;
412 seq->force_integer_mv = 2;
413 }
414 seq->order_hint_info.order_hint_bits_minus_1 =
415 seq->order_hint_info.enable_order_hint
416 ? DEFAULT_EXPLICIT_ORDER_HINT_BITS - 1
417 : -1;
418
419 seq->max_frame_width = frm_dim_cfg->forced_max_frame_width
420 ? frm_dim_cfg->forced_max_frame_width
421 : frm_dim_cfg->width;
422 seq->max_frame_height = frm_dim_cfg->forced_max_frame_height
423 ? frm_dim_cfg->forced_max_frame_height
424 : frm_dim_cfg->height;
425 seq->num_bits_width =
426 (seq->max_frame_width > 1) ? get_msb(seq->max_frame_width - 1) + 1 : 1;
427 seq->num_bits_height =
428 (seq->max_frame_height > 1) ? get_msb(seq->max_frame_height - 1) + 1 : 1;
429 assert(seq->num_bits_width <= 16);
430 assert(seq->num_bits_height <= 16);
431
432 seq->frame_id_length = FRAME_ID_LENGTH;
433 seq->delta_frame_id_length = DELTA_FRAME_ID_LENGTH;
434
435 seq->enable_dual_filter = tool_cfg->enable_dual_filter;
436 seq->order_hint_info.enable_dist_wtd_comp =
437 oxcf->comp_type_cfg.enable_dist_wtd_comp;
438 seq->order_hint_info.enable_dist_wtd_comp &=
439 seq->order_hint_info.enable_order_hint;
440 seq->order_hint_info.enable_ref_frame_mvs = tool_cfg->ref_frame_mvs_present;
441 seq->order_hint_info.enable_ref_frame_mvs &=
442 seq->order_hint_info.enable_order_hint;
443 seq->enable_superres = oxcf->superres_cfg.enable_superres;
444 seq->enable_cdef = tool_cfg->enable_cdef;
445 seq->enable_restoration = tool_cfg->enable_restoration;
446 seq->enable_warped_motion = oxcf->motion_mode_cfg.enable_warped_motion;
447 seq->enable_interintra_compound = tool_cfg->enable_interintra_comp;
448 seq->enable_masked_compound = oxcf->comp_type_cfg.enable_masked_comp;
449 seq->enable_intra_edge_filter = oxcf->intra_mode_cfg.enable_intra_edge_filter;
450 seq->enable_filter_intra = oxcf->intra_mode_cfg.enable_filter_intra;
451
452 set_bitstream_level_tier(seq, cm, frm_dim_cfg->width, frm_dim_cfg->height,
453 oxcf->input_cfg.init_framerate);
454
455 if (seq->operating_points_cnt_minus_1 == 0) {
456 seq->operating_point_idc[0] = 0;
457 } else {
458 // Set operating_point_idc[] such that the i=0 point corresponds to the
459 // highest quality operating point (all layers), and subsequent
460 // operarting points (i > 0) are lower quality corresponding to
461 // skip decoding enhancement layers (temporal first).
462 int i = 0;
463 assert(seq->operating_points_cnt_minus_1 ==
464 (int)(cm->number_spatial_layers * cm->number_temporal_layers - 1));
465 for (unsigned int sl = 0; sl < cm->number_spatial_layers; sl++) {
466 for (unsigned int tl = 0; tl < cm->number_temporal_layers; tl++) {
467 seq->operating_point_idc[i] =
468 (~(~0u << (cm->number_spatial_layers - sl)) << 8) |
469 ~(~0u << (cm->number_temporal_layers - tl));
470 i++;
471 }
472 }
473 }
474 }
475
init_config(struct AV1_COMP * cpi,AV1EncoderConfig * oxcf)476 static void init_config(struct AV1_COMP *cpi, AV1EncoderConfig *oxcf) {
477 AV1_COMMON *const cm = &cpi->common;
478 SequenceHeader *const seq_params = &cm->seq_params;
479 ResizePendingParams *resize_pending_params = &cpi->resize_pending_params;
480 const DecoderModelCfg *const dec_model_cfg = &oxcf->dec_model_cfg;
481 const ColorCfg *const color_cfg = &oxcf->color_cfg;
482 cpi->oxcf = *oxcf;
483 cpi->framerate = oxcf->input_cfg.init_framerate;
484
485 seq_params->profile = oxcf->profile;
486 seq_params->bit_depth = oxcf->tool_cfg.bit_depth;
487 seq_params->use_highbitdepth = oxcf->use_highbitdepth;
488 seq_params->color_primaries = color_cfg->color_primaries;
489 seq_params->transfer_characteristics = color_cfg->transfer_characteristics;
490 seq_params->matrix_coefficients = color_cfg->matrix_coefficients;
491 seq_params->monochrome = oxcf->tool_cfg.enable_monochrome;
492 seq_params->chroma_sample_position = color_cfg->chroma_sample_position;
493 seq_params->color_range = color_cfg->color_range;
494 seq_params->timing_info_present = dec_model_cfg->timing_info_present;
495 seq_params->timing_info.num_units_in_display_tick =
496 dec_model_cfg->timing_info.num_units_in_display_tick;
497 seq_params->timing_info.time_scale = dec_model_cfg->timing_info.time_scale;
498 seq_params->timing_info.equal_picture_interval =
499 dec_model_cfg->timing_info.equal_picture_interval;
500 seq_params->timing_info.num_ticks_per_picture =
501 dec_model_cfg->timing_info.num_ticks_per_picture;
502
503 seq_params->display_model_info_present_flag =
504 dec_model_cfg->display_model_info_present_flag;
505 seq_params->decoder_model_info_present_flag =
506 dec_model_cfg->decoder_model_info_present_flag;
507 if (dec_model_cfg->decoder_model_info_present_flag) {
508 // set the decoder model parameters in schedule mode
509 seq_params->decoder_model_info.num_units_in_decoding_tick =
510 dec_model_cfg->num_units_in_decoding_tick;
511 cm->buffer_removal_time_present = 1;
512 av1_set_aom_dec_model_info(&seq_params->decoder_model_info);
513 av1_set_dec_model_op_parameters(&seq_params->op_params[0]);
514 } else if (seq_params->timing_info_present &&
515 seq_params->timing_info.equal_picture_interval &&
516 !seq_params->decoder_model_info_present_flag) {
517 // set the decoder model parameters in resource availability mode
518 av1_set_resource_availability_parameters(&seq_params->op_params[0]);
519 } else {
520 seq_params->op_params[0].initial_display_delay =
521 10; // Default value (not signaled)
522 }
523
524 if (seq_params->monochrome) {
525 seq_params->subsampling_x = 1;
526 seq_params->subsampling_y = 1;
527 } else if (seq_params->color_primaries == AOM_CICP_CP_BT_709 &&
528 seq_params->transfer_characteristics == AOM_CICP_TC_SRGB &&
529 seq_params->matrix_coefficients == AOM_CICP_MC_IDENTITY) {
530 seq_params->subsampling_x = 0;
531 seq_params->subsampling_y = 0;
532 } else {
533 if (seq_params->profile == 0) {
534 seq_params->subsampling_x = 1;
535 seq_params->subsampling_y = 1;
536 } else if (seq_params->profile == 1) {
537 seq_params->subsampling_x = 0;
538 seq_params->subsampling_y = 0;
539 } else {
540 if (seq_params->bit_depth == AOM_BITS_12) {
541 seq_params->subsampling_x = oxcf->input_cfg.chroma_subsampling_x;
542 seq_params->subsampling_y = oxcf->input_cfg.chroma_subsampling_y;
543 } else {
544 seq_params->subsampling_x = 1;
545 seq_params->subsampling_y = 0;
546 }
547 }
548 }
549
550 cm->width = oxcf->frm_dim_cfg.width;
551 cm->height = oxcf->frm_dim_cfg.height;
552 set_sb_size(seq_params,
553 av1_select_sb_size(cpi)); // set sb size before allocations
554 alloc_compressor_data(cpi);
555
556 av1_update_film_grain_parameters(cpi, oxcf);
557
558 // Single thread case: use counts in common.
559 cpi->td.counts = &cpi->counts;
560
561 // Set init SVC parameters.
562 cpi->use_svc = 0;
563 cpi->svc.external_ref_frame_config = 0;
564 cpi->svc.non_reference_frame = 0;
565 cpi->svc.number_spatial_layers = 1;
566 cpi->svc.number_temporal_layers = 1;
567 cm->number_spatial_layers = 1;
568 cm->number_temporal_layers = 1;
569 cm->spatial_layer_id = 0;
570 cm->temporal_layer_id = 0;
571
572 // change includes all joint functionality
573 av1_change_config(cpi, oxcf);
574
575 cpi->ref_frame_flags = 0;
576
577 // Reset resize pending flags
578 resize_pending_params->width = 0;
579 resize_pending_params->height = 0;
580
581 init_buffer_indices(&cpi->force_intpel_info, cm->remapped_ref_idx);
582
583 av1_noise_estimate_init(&cpi->noise_estimate, cm->width, cm->height);
584 }
585
av1_change_config(struct AV1_COMP * cpi,const AV1EncoderConfig * oxcf)586 void av1_change_config(struct AV1_COMP *cpi, const AV1EncoderConfig *oxcf) {
587 AV1_COMMON *const cm = &cpi->common;
588 SequenceHeader *const seq_params = &cm->seq_params;
589 RATE_CONTROL *const rc = &cpi->rc;
590 MACROBLOCK *const x = &cpi->td.mb;
591 AV1LevelParams *const level_params = &cpi->level_params;
592 InitialDimensions *const initial_dimensions = &cpi->initial_dimensions;
593 RefreshFrameFlagsInfo *const refresh_frame_flags = &cpi->refresh_frame;
594 const FrameDimensionCfg *const frm_dim_cfg = &cpi->oxcf.frm_dim_cfg;
595 const DecoderModelCfg *const dec_model_cfg = &oxcf->dec_model_cfg;
596 const ColorCfg *const color_cfg = &oxcf->color_cfg;
597 const RateControlCfg *const rc_cfg = &oxcf->rc_cfg;
598 // in case of LAP, lag in frames is set according to number of lap buffers
599 // calculated at init time. This stores and restores LAP's lag in frames to
600 // prevent override by new cfg.
601 int lap_lag_in_frames = -1;
602 if (cpi->lap_enabled && cpi->compressor_stage == LAP_STAGE) {
603 lap_lag_in_frames = cpi->oxcf.gf_cfg.lag_in_frames;
604 }
605
606 if (seq_params->profile != oxcf->profile) seq_params->profile = oxcf->profile;
607 seq_params->bit_depth = oxcf->tool_cfg.bit_depth;
608 seq_params->color_primaries = color_cfg->color_primaries;
609 seq_params->transfer_characteristics = color_cfg->transfer_characteristics;
610 seq_params->matrix_coefficients = color_cfg->matrix_coefficients;
611 seq_params->monochrome = oxcf->tool_cfg.enable_monochrome;
612 seq_params->chroma_sample_position = color_cfg->chroma_sample_position;
613 seq_params->color_range = color_cfg->color_range;
614
615 assert(IMPLIES(seq_params->profile <= PROFILE_1,
616 seq_params->bit_depth <= AOM_BITS_10));
617
618 seq_params->timing_info_present = dec_model_cfg->timing_info_present;
619 seq_params->timing_info.num_units_in_display_tick =
620 dec_model_cfg->timing_info.num_units_in_display_tick;
621 seq_params->timing_info.time_scale = dec_model_cfg->timing_info.time_scale;
622 seq_params->timing_info.equal_picture_interval =
623 dec_model_cfg->timing_info.equal_picture_interval;
624 seq_params->timing_info.num_ticks_per_picture =
625 dec_model_cfg->timing_info.num_ticks_per_picture;
626
627 seq_params->display_model_info_present_flag =
628 dec_model_cfg->display_model_info_present_flag;
629 seq_params->decoder_model_info_present_flag =
630 dec_model_cfg->decoder_model_info_present_flag;
631 if (dec_model_cfg->decoder_model_info_present_flag) {
632 // set the decoder model parameters in schedule mode
633 seq_params->decoder_model_info.num_units_in_decoding_tick =
634 dec_model_cfg->num_units_in_decoding_tick;
635 cm->buffer_removal_time_present = 1;
636 av1_set_aom_dec_model_info(&seq_params->decoder_model_info);
637 av1_set_dec_model_op_parameters(&seq_params->op_params[0]);
638 } else if (seq_params->timing_info_present &&
639 seq_params->timing_info.equal_picture_interval &&
640 !seq_params->decoder_model_info_present_flag) {
641 // set the decoder model parameters in resource availability mode
642 av1_set_resource_availability_parameters(&seq_params->op_params[0]);
643 } else {
644 seq_params->op_params[0].initial_display_delay =
645 10; // Default value (not signaled)
646 }
647
648 av1_update_film_grain_parameters(cpi, oxcf);
649
650 cpi->oxcf = *oxcf;
651 // When user provides superres_mode = AOM_SUPERRES_AUTO, we still initialize
652 // superres mode for current encoding = AOM_SUPERRES_NONE. This is to ensure
653 // that any analysis (e.g. TPL) happening outside the main encoding loop still
654 // happens at full resolution.
655 // This value will later be set appropriately just before main encoding loop.
656 cpi->superres_mode = oxcf->superres_cfg.superres_mode == AOM_SUPERRES_AUTO
657 ? AOM_SUPERRES_NONE
658 : oxcf->superres_cfg.superres_mode; // default
659 x->e_mbd.bd = (int)seq_params->bit_depth;
660 x->e_mbd.global_motion = cm->global_motion;
661
662 memcpy(level_params->target_seq_level_idx, cpi->oxcf.target_seq_level_idx,
663 sizeof(level_params->target_seq_level_idx));
664 level_params->keep_level_stats = 0;
665 for (int i = 0; i < MAX_NUM_OPERATING_POINTS; ++i) {
666 if (level_params->target_seq_level_idx[i] <= SEQ_LEVELS) {
667 level_params->keep_level_stats |= 1u << i;
668 if (!level_params->level_info[i]) {
669 CHECK_MEM_ERROR(cm, level_params->level_info[i],
670 aom_calloc(1, sizeof(*level_params->level_info[i])));
671 }
672 }
673 }
674
675 // TODO(huisu@): level targeting currently only works for the 0th operating
676 // point, so scalable coding is not supported yet.
677 if (level_params->target_seq_level_idx[0] < SEQ_LEVELS) {
678 // Adjust encoder config in order to meet target level.
679 config_target_level(cpi, level_params->target_seq_level_idx[0],
680 seq_params->tier[0]);
681 }
682
683 if ((has_no_stats_stage(cpi)) && (rc_cfg->mode == AOM_Q)) {
684 rc->baseline_gf_interval = FIXED_GF_INTERVAL;
685 } else {
686 rc->baseline_gf_interval = (MIN_GF_INTERVAL + MAX_GF_INTERVAL) / 2;
687 }
688
689 refresh_frame_flags->golden_frame = false;
690 refresh_frame_flags->bwd_ref_frame = false;
691
692 cm->features.refresh_frame_context =
693 (oxcf->tool_cfg.frame_parallel_decoding_mode)
694 ? REFRESH_FRAME_CONTEXT_DISABLED
695 : REFRESH_FRAME_CONTEXT_BACKWARD;
696 if (oxcf->tile_cfg.enable_large_scale_tile)
697 cm->features.refresh_frame_context = REFRESH_FRAME_CONTEXT_DISABLED;
698
699 if (x->palette_buffer == NULL) {
700 CHECK_MEM_ERROR(cm, x->palette_buffer,
701 aom_memalign(16, sizeof(*x->palette_buffer)));
702 }
703
704 if (x->comp_rd_buffer.pred0 == NULL) {
705 alloc_compound_type_rd_buffers(cm, &x->comp_rd_buffer);
706 }
707
708 if (x->tmp_conv_dst == NULL) {
709 CHECK_MEM_ERROR(
710 cm, x->tmp_conv_dst,
711 aom_memalign(32, MAX_SB_SIZE * MAX_SB_SIZE * sizeof(*x->tmp_conv_dst)));
712 x->e_mbd.tmp_conv_dst = x->tmp_conv_dst;
713 }
714 for (int i = 0; i < 2; ++i) {
715 if (x->tmp_pred_bufs[i] == NULL) {
716 CHECK_MEM_ERROR(cm, x->tmp_pred_bufs[i],
717 aom_memalign(32, 2 * MAX_MB_PLANE * MAX_SB_SQUARE *
718 sizeof(*x->tmp_pred_bufs[i])));
719 x->e_mbd.tmp_obmc_bufs[i] = x->tmp_pred_bufs[i];
720 }
721 }
722
723 av1_reset_segment_features(cm);
724
725 av1_set_high_precision_mv(cpi, 1, 0);
726
727 set_rc_buffer_sizes(rc, rc_cfg);
728
729 // Under a configuration change, where maximum_buffer_size may change,
730 // keep buffer level clipped to the maximum allowed buffer size.
731 rc->bits_off_target = AOMMIN(rc->bits_off_target, rc->maximum_buffer_size);
732 rc->buffer_level = AOMMIN(rc->buffer_level, rc->maximum_buffer_size);
733
734 // Set up frame rate and related parameters rate control values.
735 av1_new_framerate(cpi, cpi->framerate);
736
737 // Set absolute upper and lower quality limits
738 rc->worst_quality = rc_cfg->worst_allowed_q;
739 rc->best_quality = rc_cfg->best_allowed_q;
740
741 cm->features.interp_filter =
742 oxcf->tile_cfg.enable_large_scale_tile ? EIGHTTAP_REGULAR : SWITCHABLE;
743 cm->features.switchable_motion_mode = 1;
744
745 if (frm_dim_cfg->render_width > 0 && frm_dim_cfg->render_height > 0) {
746 cm->render_width = frm_dim_cfg->render_width;
747 cm->render_height = frm_dim_cfg->render_height;
748 } else {
749 cm->render_width = frm_dim_cfg->width;
750 cm->render_height = frm_dim_cfg->height;
751 }
752 cm->width = frm_dim_cfg->width;
753 cm->height = frm_dim_cfg->height;
754
755 int sb_size = seq_params->sb_size;
756 // Superblock size should not be updated after the first key frame.
757 if (!cpi->seq_params_locked) {
758 set_sb_size(&cm->seq_params, av1_select_sb_size(cpi));
759 for (int i = 0; i < MAX_NUM_OPERATING_POINTS; ++i)
760 seq_params->tier[i] = (oxcf->tier_mask >> i) & 1;
761 }
762
763 if (initial_dimensions->width || sb_size != seq_params->sb_size) {
764 if (cm->width > initial_dimensions->width ||
765 cm->height > initial_dimensions->height ||
766 seq_params->sb_size != sb_size) {
767 av1_free_context_buffers(cm);
768 av1_free_shared_coeff_buffer(&cpi->td.shared_coeff_buf);
769 av1_free_sms_tree(&cpi->td);
770 av1_free_pmc(cpi->td.firstpass_ctx, av1_num_planes(cm));
771 cpi->td.firstpass_ctx = NULL;
772 alloc_compressor_data(cpi);
773 realloc_segmentation_maps(cpi);
774 initial_dimensions->width = initial_dimensions->height = 0;
775 }
776 }
777 av1_update_frame_size(cpi);
778
779 rc->is_src_frame_alt_ref = 0;
780
781 set_tile_info(cm, &cpi->oxcf.tile_cfg);
782
783 if (!cpi->svc.external_ref_frame_config)
784 cpi->ext_flags.refresh_frame.update_pending = 0;
785 cpi->ext_flags.refresh_frame_context_pending = 0;
786
787 #if CONFIG_AV1_HIGHBITDEPTH
788 highbd_set_var_fns(cpi);
789 #endif
790
791 // Init sequence level coding tools
792 // This should not be called after the first key frame.
793 if (!cpi->seq_params_locked) {
794 seq_params->operating_points_cnt_minus_1 =
795 (cm->number_spatial_layers > 1 || cm->number_temporal_layers > 1)
796 ? cm->number_spatial_layers * cm->number_temporal_layers - 1
797 : 0;
798 av1_init_seq_coding_tools(&cm->seq_params, cm, oxcf, cpi->use_svc);
799 }
800
801 if (cpi->use_svc)
802 av1_update_layer_context_change_config(cpi, rc_cfg->target_bandwidth);
803
804 // restore the value of lag_in_frame for LAP stage.
805 if (lap_lag_in_frames != -1) {
806 cpi->oxcf.gf_cfg.lag_in_frames = lap_lag_in_frames;
807 }
808 }
809
init_frame_info(FRAME_INFO * frame_info,const AV1_COMMON * const cm)810 static INLINE void init_frame_info(FRAME_INFO *frame_info,
811 const AV1_COMMON *const cm) {
812 const CommonModeInfoParams *const mi_params = &cm->mi_params;
813 const SequenceHeader *const seq_params = &cm->seq_params;
814 frame_info->frame_width = cm->width;
815 frame_info->frame_height = cm->height;
816 frame_info->mi_cols = mi_params->mi_cols;
817 frame_info->mi_rows = mi_params->mi_rows;
818 frame_info->mb_cols = mi_params->mb_cols;
819 frame_info->mb_rows = mi_params->mb_rows;
820 frame_info->num_mbs = mi_params->MBs;
821 frame_info->bit_depth = seq_params->bit_depth;
822 frame_info->subsampling_x = seq_params->subsampling_x;
823 frame_info->subsampling_y = seq_params->subsampling_y;
824 }
825
init_frame_index_set(FRAME_INDEX_SET * frame_index_set)826 static INLINE void init_frame_index_set(FRAME_INDEX_SET *frame_index_set) {
827 frame_index_set->show_frame_count = 0;
828 }
829
update_frame_index_set(FRAME_INDEX_SET * frame_index_set,int is_show_frame)830 static INLINE void update_frame_index_set(FRAME_INDEX_SET *frame_index_set,
831 int is_show_frame) {
832 if (is_show_frame) {
833 frame_index_set->show_frame_count++;
834 }
835 }
836
av1_create_compressor(AV1EncoderConfig * oxcf,BufferPool * const pool,FIRSTPASS_STATS * frame_stats_buf,COMPRESSOR_STAGE stage,int num_lap_buffers,int lap_lag_in_frames,STATS_BUFFER_CTX * stats_buf_context)837 AV1_COMP *av1_create_compressor(AV1EncoderConfig *oxcf, BufferPool *const pool,
838 FIRSTPASS_STATS *frame_stats_buf,
839 COMPRESSOR_STAGE stage, int num_lap_buffers,
840 int lap_lag_in_frames,
841 STATS_BUFFER_CTX *stats_buf_context) {
842 AV1_COMP *volatile const cpi = aom_memalign(32, sizeof(AV1_COMP));
843 AV1_COMMON *volatile const cm = cpi != NULL ? &cpi->common : NULL;
844
845 if (!cm) return NULL;
846
847 av1_zero(*cpi);
848
849 // The jmp_buf is valid only for the duration of the function that calls
850 // setjmp(). Therefore, this function must reset the 'setjmp' field to 0
851 // before it returns.
852 if (setjmp(cm->error.jmp)) {
853 cm->error.setjmp = 0;
854 av1_remove_compressor(cpi);
855 return 0;
856 }
857
858 cm->error.setjmp = 1;
859 cpi->lap_enabled = num_lap_buffers > 0;
860 cpi->compressor_stage = stage;
861
862 CommonModeInfoParams *const mi_params = &cm->mi_params;
863 mi_params->free_mi = enc_free_mi;
864 mi_params->setup_mi = enc_setup_mi;
865 mi_params->set_mb_mi = (oxcf->pass == 1 || cpi->compressor_stage == LAP_STAGE)
866 ? stat_stage_set_mb_mi
867 : enc_set_mb_mi;
868
869 mi_params->mi_alloc_bsize = BLOCK_4X4;
870
871 CHECK_MEM_ERROR(cm, cm->fc,
872 (FRAME_CONTEXT *)aom_memalign(32, sizeof(*cm->fc)));
873 CHECK_MEM_ERROR(
874 cm, cm->default_frame_context,
875 (FRAME_CONTEXT *)aom_memalign(32, sizeof(*cm->default_frame_context)));
876 memset(cm->fc, 0, sizeof(*cm->fc));
877 memset(cm->default_frame_context, 0, sizeof(*cm->default_frame_context));
878
879 cpi->common.buffer_pool = pool;
880
881 init_config(cpi, oxcf);
882 if (cpi->compressor_stage == LAP_STAGE) {
883 cpi->oxcf.gf_cfg.lag_in_frames = lap_lag_in_frames;
884 }
885
886 cpi->frames_left = cpi->oxcf.input_cfg.limit;
887
888 av1_rc_init(&cpi->oxcf, oxcf->pass, &cpi->rc);
889
890 // For two pass and lag_in_frames > 33 in LAP.
891 cpi->rc.enable_scenecut_detection = ENABLE_SCENECUT_MODE_2;
892 if (cpi->lap_enabled) {
893 if ((num_lap_buffers <
894 (MAX_GF_LENGTH_LAP + SCENE_CUT_KEY_TEST_INTERVAL + 1)) &&
895 num_lap_buffers >= (MAX_GF_LENGTH_LAP + 3)) {
896 /*
897 * For lag in frames >= 19 and <33, enable scenecut
898 * with limited future frame prediction.
899 */
900 cpi->rc.enable_scenecut_detection = ENABLE_SCENECUT_MODE_1;
901 } else if (num_lap_buffers < (MAX_GF_LENGTH_LAP + 3)) {
902 // Disable scenecut when lag_in_frames < 19.
903 cpi->rc.enable_scenecut_detection = DISABLE_SCENECUT;
904 }
905 }
906 init_frame_info(&cpi->frame_info, cm);
907 init_frame_index_set(&cpi->frame_index_set);
908
909 cm->current_frame.frame_number = 0;
910 cm->current_frame_id = -1;
911 cpi->seq_params_locked = 0;
912 cpi->partition_search_skippable_frame = 0;
913 cpi->tile_data = NULL;
914 cpi->last_show_frame_buf = NULL;
915 realloc_segmentation_maps(cpi);
916
917 cpi->refresh_frame.alt_ref_frame = false;
918
919 cpi->b_calculate_psnr = CONFIG_INTERNAL_STATS;
920 #if CONFIG_INTERNAL_STATS
921 cpi->b_calculate_blockiness = 1;
922 cpi->b_calculate_consistency = 1;
923 cpi->total_inconsistency = 0;
924 cpi->psnr[0].worst = 100.0;
925 cpi->psnr[1].worst = 100.0;
926 cpi->worst_ssim = 100.0;
927 cpi->worst_ssim_hbd = 100.0;
928
929 cpi->count[0] = 0;
930 cpi->count[1] = 0;
931 cpi->bytes = 0;
932 #if CONFIG_SPEED_STATS
933 cpi->tx_search_count = 0;
934 #endif // CONFIG_SPEED_STATS
935
936 if (cpi->b_calculate_psnr) {
937 cpi->total_sq_error[0] = 0;
938 cpi->total_samples[0] = 0;
939 cpi->total_sq_error[1] = 0;
940 cpi->total_samples[1] = 0;
941 cpi->tot_recode_hits = 0;
942 cpi->summed_quality = 0;
943 cpi->summed_weights = 0;
944 cpi->summed_quality_hbd = 0;
945 cpi->summed_weights_hbd = 0;
946 }
947
948 cpi->fastssim.worst = 100.0;
949 cpi->psnrhvs.worst = 100.0;
950
951 if (cpi->b_calculate_blockiness) {
952 cpi->total_blockiness = 0;
953 cpi->worst_blockiness = 0.0;
954 }
955
956 if (cpi->b_calculate_consistency) {
957 CHECK_MEM_ERROR(
958 cm, cpi->ssim_vars,
959 aom_malloc(sizeof(*cpi->ssim_vars) * 4 * cpi->common.mi_params.mi_rows *
960 cpi->common.mi_params.mi_cols));
961 cpi->worst_consistency = 100.0;
962 }
963 #endif
964 #if CONFIG_ENTROPY_STATS
965 av1_zero(aggregate_fc);
966 #endif // CONFIG_ENTROPY_STATS
967
968 cpi->time_stamps.first_ts_start = INT64_MAX;
969
970 #ifdef OUTPUT_YUV_REC
971 yuv_rec_file = fopen("rec.yuv", "wb");
972 #endif
973 #ifdef OUTPUT_YUV_DENOISED
974 yuv_denoised_file = fopen("denoised.yuv", "wb");
975 #endif
976
977 assert(MAX_LAP_BUFFERS >= MAX_LAG_BUFFERS);
978 int size = get_stats_buf_size(num_lap_buffers, MAX_LAG_BUFFERS);
979 for (int i = 0; i < size; i++)
980 cpi->twopass.frame_stats_arr[i] = &frame_stats_buf[i];
981
982 cpi->twopass.stats_buf_ctx = stats_buf_context;
983 cpi->twopass.stats_in = cpi->twopass.stats_buf_ctx->stats_in_start;
984
985 #if !CONFIG_REALTIME_ONLY
986 if (is_stat_consumption_stage(cpi)) {
987 const size_t packet_sz = sizeof(FIRSTPASS_STATS);
988 const int packets = (int)(oxcf->twopass_stats_in.sz / packet_sz);
989
990 if (!cpi->lap_enabled) {
991 /*Re-initialize to stats buffer, populated by application in the case of
992 * two pass*/
993 cpi->twopass.stats_buf_ctx->stats_in_start = oxcf->twopass_stats_in.buf;
994 cpi->twopass.stats_in = cpi->twopass.stats_buf_ctx->stats_in_start;
995 cpi->twopass.stats_buf_ctx->stats_in_end =
996 &cpi->twopass.stats_buf_ctx->stats_in_start[packets - 1];
997
998 av1_init_second_pass(cpi);
999 } else {
1000 av1_init_single_pass_lap(cpi);
1001 }
1002 }
1003 #endif
1004
1005 alloc_obmc_buffers(&cpi->td.mb.obmc_buffer, cm);
1006
1007 CHECK_MEM_ERROR(
1008 cm, cpi->td.mb.inter_modes_info,
1009 (InterModesInfo *)aom_malloc(sizeof(*cpi->td.mb.inter_modes_info)));
1010
1011 for (int x = 0; x < 2; x++)
1012 for (int y = 0; y < 2; y++)
1013 CHECK_MEM_ERROR(
1014 cm, cpi->td.mb.intrabc_hash_info.hash_value_buffer[x][y],
1015 (uint32_t *)aom_malloc(
1016 AOM_BUFFER_SIZE_FOR_BLOCK_HASH *
1017 sizeof(*cpi->td.mb.intrabc_hash_info.hash_value_buffer[0][0])));
1018
1019 cpi->td.mb.intrabc_hash_info.g_crc_initialized = 0;
1020
1021 av1_set_speed_features_framesize_independent(cpi, oxcf->speed);
1022 av1_set_speed_features_framesize_dependent(cpi, oxcf->speed);
1023
1024 CHECK_MEM_ERROR(cm, cpi->consec_zero_mv,
1025 aom_calloc((mi_params->mi_rows * mi_params->mi_cols) >> 2,
1026 sizeof(*cpi->consec_zero_mv)));
1027
1028 {
1029 const int bsize = BLOCK_16X16;
1030 const int w = mi_size_wide[bsize];
1031 const int h = mi_size_high[bsize];
1032 const int num_cols = (mi_params->mi_cols + w - 1) / w;
1033 const int num_rows = (mi_params->mi_rows + h - 1) / h;
1034 CHECK_MEM_ERROR(cm, cpi->tpl_rdmult_scaling_factors,
1035 aom_calloc(num_rows * num_cols,
1036 sizeof(*cpi->tpl_rdmult_scaling_factors)));
1037 CHECK_MEM_ERROR(cm, cpi->tpl_sb_rdmult_scaling_factors,
1038 aom_calloc(num_rows * num_cols,
1039 sizeof(*cpi->tpl_sb_rdmult_scaling_factors)));
1040 }
1041
1042 {
1043 const int bsize = BLOCK_16X16;
1044 const int w = mi_size_wide[bsize];
1045 const int h = mi_size_high[bsize];
1046 const int num_cols = (mi_params->mi_cols + w - 1) / w;
1047 const int num_rows = (mi_params->mi_rows + h - 1) / h;
1048 CHECK_MEM_ERROR(cm, cpi->ssim_rdmult_scaling_factors,
1049 aom_calloc(num_rows * num_cols,
1050 sizeof(*cpi->ssim_rdmult_scaling_factors)));
1051 }
1052
1053 #if CONFIG_TUNE_VMAF
1054 {
1055 const int bsize = BLOCK_64X64;
1056 const int w = mi_size_wide[bsize];
1057 const int h = mi_size_high[bsize];
1058 const int num_cols = (mi_params->mi_cols + w - 1) / w;
1059 const int num_rows = (mi_params->mi_rows + h - 1) / h;
1060 CHECK_MEM_ERROR(cm, cpi->vmaf_info.rdmult_scaling_factors,
1061 aom_calloc(num_rows * num_cols,
1062 sizeof(*cpi->vmaf_info.rdmult_scaling_factors)));
1063 for (int i = 0; i < MAX_ARF_LAYERS; i++) {
1064 cpi->vmaf_info.last_frame_unsharp_amount[i] = -1.0;
1065 cpi->vmaf_info.last_frame_ysse[i] = -1.0;
1066 cpi->vmaf_info.last_frame_vmaf[i] = -1.0;
1067 }
1068 cpi->vmaf_info.original_qindex = -1;
1069
1070 #if CONFIG_USE_VMAF_RC
1071 cpi->vmaf_info.vmaf_model = NULL;
1072 #endif
1073 }
1074 #endif
1075
1076 if (!is_stat_generation_stage(cpi)) {
1077 setup_tpl_buffers(cm, &cpi->tpl_data, cpi->oxcf.gf_cfg.lag_in_frames);
1078 }
1079
1080 #if CONFIG_COLLECT_PARTITION_STATS
1081 av1_zero(cpi->partition_stats);
1082 #endif // CONFIG_COLLECT_PARTITION_STATS
1083
1084 #define BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX4DF, JSDAF, JSVAF) \
1085 cpi->fn_ptr[BT].sdf = SDF; \
1086 cpi->fn_ptr[BT].sdaf = SDAF; \
1087 cpi->fn_ptr[BT].vf = VF; \
1088 cpi->fn_ptr[BT].svf = SVF; \
1089 cpi->fn_ptr[BT].svaf = SVAF; \
1090 cpi->fn_ptr[BT].sdx4df = SDX4DF; \
1091 cpi->fn_ptr[BT].jsdaf = JSDAF; \
1092 cpi->fn_ptr[BT].jsvaf = JSVAF;
1093
1094 // Realtime mode doesn't use 4x rectangular blocks.
1095 #if !CONFIG_REALTIME_ONLY
1096 BFP(BLOCK_4X16, aom_sad4x16, aom_sad4x16_avg, aom_variance4x16,
1097 aom_sub_pixel_variance4x16, aom_sub_pixel_avg_variance4x16,
1098 aom_sad4x16x4d, aom_dist_wtd_sad4x16_avg,
1099 aom_dist_wtd_sub_pixel_avg_variance4x16)
1100
1101 BFP(BLOCK_16X4, aom_sad16x4, aom_sad16x4_avg, aom_variance16x4,
1102 aom_sub_pixel_variance16x4, aom_sub_pixel_avg_variance16x4,
1103 aom_sad16x4x4d, aom_dist_wtd_sad16x4_avg,
1104 aom_dist_wtd_sub_pixel_avg_variance16x4)
1105
1106 BFP(BLOCK_8X32, aom_sad8x32, aom_sad8x32_avg, aom_variance8x32,
1107 aom_sub_pixel_variance8x32, aom_sub_pixel_avg_variance8x32,
1108 aom_sad8x32x4d, aom_dist_wtd_sad8x32_avg,
1109 aom_dist_wtd_sub_pixel_avg_variance8x32)
1110
1111 BFP(BLOCK_32X8, aom_sad32x8, aom_sad32x8_avg, aom_variance32x8,
1112 aom_sub_pixel_variance32x8, aom_sub_pixel_avg_variance32x8,
1113 aom_sad32x8x4d, aom_dist_wtd_sad32x8_avg,
1114 aom_dist_wtd_sub_pixel_avg_variance32x8)
1115
1116 BFP(BLOCK_16X64, aom_sad16x64, aom_sad16x64_avg, aom_variance16x64,
1117 aom_sub_pixel_variance16x64, aom_sub_pixel_avg_variance16x64,
1118 aom_sad16x64x4d, aom_dist_wtd_sad16x64_avg,
1119 aom_dist_wtd_sub_pixel_avg_variance16x64)
1120
1121 BFP(BLOCK_64X16, aom_sad64x16, aom_sad64x16_avg, aom_variance64x16,
1122 aom_sub_pixel_variance64x16, aom_sub_pixel_avg_variance64x16,
1123 aom_sad64x16x4d, aom_dist_wtd_sad64x16_avg,
1124 aom_dist_wtd_sub_pixel_avg_variance64x16)
1125 #endif // !CONFIG_REALTIME_ONLY
1126
1127 BFP(BLOCK_128X128, aom_sad128x128, aom_sad128x128_avg, aom_variance128x128,
1128 aom_sub_pixel_variance128x128, aom_sub_pixel_avg_variance128x128,
1129 aom_sad128x128x4d, aom_dist_wtd_sad128x128_avg,
1130 aom_dist_wtd_sub_pixel_avg_variance128x128)
1131
1132 BFP(BLOCK_128X64, aom_sad128x64, aom_sad128x64_avg, aom_variance128x64,
1133 aom_sub_pixel_variance128x64, aom_sub_pixel_avg_variance128x64,
1134 aom_sad128x64x4d, aom_dist_wtd_sad128x64_avg,
1135 aom_dist_wtd_sub_pixel_avg_variance128x64)
1136
1137 BFP(BLOCK_64X128, aom_sad64x128, aom_sad64x128_avg, aom_variance64x128,
1138 aom_sub_pixel_variance64x128, aom_sub_pixel_avg_variance64x128,
1139 aom_sad64x128x4d, aom_dist_wtd_sad64x128_avg,
1140 aom_dist_wtd_sub_pixel_avg_variance64x128)
1141
1142 BFP(BLOCK_32X16, aom_sad32x16, aom_sad32x16_avg, aom_variance32x16,
1143 aom_sub_pixel_variance32x16, aom_sub_pixel_avg_variance32x16,
1144 aom_sad32x16x4d, aom_dist_wtd_sad32x16_avg,
1145 aom_dist_wtd_sub_pixel_avg_variance32x16)
1146
1147 BFP(BLOCK_16X32, aom_sad16x32, aom_sad16x32_avg, aom_variance16x32,
1148 aom_sub_pixel_variance16x32, aom_sub_pixel_avg_variance16x32,
1149 aom_sad16x32x4d, aom_dist_wtd_sad16x32_avg,
1150 aom_dist_wtd_sub_pixel_avg_variance16x32)
1151
1152 BFP(BLOCK_64X32, aom_sad64x32, aom_sad64x32_avg, aom_variance64x32,
1153 aom_sub_pixel_variance64x32, aom_sub_pixel_avg_variance64x32,
1154 aom_sad64x32x4d, aom_dist_wtd_sad64x32_avg,
1155 aom_dist_wtd_sub_pixel_avg_variance64x32)
1156
1157 BFP(BLOCK_32X64, aom_sad32x64, aom_sad32x64_avg, aom_variance32x64,
1158 aom_sub_pixel_variance32x64, aom_sub_pixel_avg_variance32x64,
1159 aom_sad32x64x4d, aom_dist_wtd_sad32x64_avg,
1160 aom_dist_wtd_sub_pixel_avg_variance32x64)
1161
1162 BFP(BLOCK_32X32, aom_sad32x32, aom_sad32x32_avg, aom_variance32x32,
1163 aom_sub_pixel_variance32x32, aom_sub_pixel_avg_variance32x32,
1164 aom_sad32x32x4d, aom_dist_wtd_sad32x32_avg,
1165 aom_dist_wtd_sub_pixel_avg_variance32x32)
1166
1167 BFP(BLOCK_64X64, aom_sad64x64, aom_sad64x64_avg, aom_variance64x64,
1168 aom_sub_pixel_variance64x64, aom_sub_pixel_avg_variance64x64,
1169 aom_sad64x64x4d, aom_dist_wtd_sad64x64_avg,
1170 aom_dist_wtd_sub_pixel_avg_variance64x64)
1171
1172 BFP(BLOCK_16X16, aom_sad16x16, aom_sad16x16_avg, aom_variance16x16,
1173 aom_sub_pixel_variance16x16, aom_sub_pixel_avg_variance16x16,
1174 aom_sad16x16x4d, aom_dist_wtd_sad16x16_avg,
1175 aom_dist_wtd_sub_pixel_avg_variance16x16)
1176
1177 BFP(BLOCK_16X8, aom_sad16x8, aom_sad16x8_avg, aom_variance16x8,
1178 aom_sub_pixel_variance16x8, aom_sub_pixel_avg_variance16x8,
1179 aom_sad16x8x4d, aom_dist_wtd_sad16x8_avg,
1180 aom_dist_wtd_sub_pixel_avg_variance16x8)
1181
1182 BFP(BLOCK_8X16, aom_sad8x16, aom_sad8x16_avg, aom_variance8x16,
1183 aom_sub_pixel_variance8x16, aom_sub_pixel_avg_variance8x16,
1184 aom_sad8x16x4d, aom_dist_wtd_sad8x16_avg,
1185 aom_dist_wtd_sub_pixel_avg_variance8x16)
1186
1187 BFP(BLOCK_8X8, aom_sad8x8, aom_sad8x8_avg, aom_variance8x8,
1188 aom_sub_pixel_variance8x8, aom_sub_pixel_avg_variance8x8, aom_sad8x8x4d,
1189 aom_dist_wtd_sad8x8_avg, aom_dist_wtd_sub_pixel_avg_variance8x8)
1190
1191 BFP(BLOCK_8X4, aom_sad8x4, aom_sad8x4_avg, aom_variance8x4,
1192 aom_sub_pixel_variance8x4, aom_sub_pixel_avg_variance8x4, aom_sad8x4x4d,
1193 aom_dist_wtd_sad8x4_avg, aom_dist_wtd_sub_pixel_avg_variance8x4)
1194
1195 BFP(BLOCK_4X8, aom_sad4x8, aom_sad4x8_avg, aom_variance4x8,
1196 aom_sub_pixel_variance4x8, aom_sub_pixel_avg_variance4x8, aom_sad4x8x4d,
1197 aom_dist_wtd_sad4x8_avg, aom_dist_wtd_sub_pixel_avg_variance4x8)
1198
1199 BFP(BLOCK_4X4, aom_sad4x4, aom_sad4x4_avg, aom_variance4x4,
1200 aom_sub_pixel_variance4x4, aom_sub_pixel_avg_variance4x4, aom_sad4x4x4d,
1201 aom_dist_wtd_sad4x4_avg, aom_dist_wtd_sub_pixel_avg_variance4x4)
1202
1203 #if !CONFIG_REALTIME_ONLY
1204 #define OBFP(BT, OSDF, OVF, OSVF) \
1205 cpi->fn_ptr[BT].osdf = OSDF; \
1206 cpi->fn_ptr[BT].ovf = OVF; \
1207 cpi->fn_ptr[BT].osvf = OSVF;
1208
1209 OBFP(BLOCK_128X128, aom_obmc_sad128x128, aom_obmc_variance128x128,
1210 aom_obmc_sub_pixel_variance128x128)
1211 OBFP(BLOCK_128X64, aom_obmc_sad128x64, aom_obmc_variance128x64,
1212 aom_obmc_sub_pixel_variance128x64)
1213 OBFP(BLOCK_64X128, aom_obmc_sad64x128, aom_obmc_variance64x128,
1214 aom_obmc_sub_pixel_variance64x128)
1215 OBFP(BLOCK_64X64, aom_obmc_sad64x64, aom_obmc_variance64x64,
1216 aom_obmc_sub_pixel_variance64x64)
1217 OBFP(BLOCK_64X32, aom_obmc_sad64x32, aom_obmc_variance64x32,
1218 aom_obmc_sub_pixel_variance64x32)
1219 OBFP(BLOCK_32X64, aom_obmc_sad32x64, aom_obmc_variance32x64,
1220 aom_obmc_sub_pixel_variance32x64)
1221 OBFP(BLOCK_32X32, aom_obmc_sad32x32, aom_obmc_variance32x32,
1222 aom_obmc_sub_pixel_variance32x32)
1223 OBFP(BLOCK_32X16, aom_obmc_sad32x16, aom_obmc_variance32x16,
1224 aom_obmc_sub_pixel_variance32x16)
1225 OBFP(BLOCK_16X32, aom_obmc_sad16x32, aom_obmc_variance16x32,
1226 aom_obmc_sub_pixel_variance16x32)
1227 OBFP(BLOCK_16X16, aom_obmc_sad16x16, aom_obmc_variance16x16,
1228 aom_obmc_sub_pixel_variance16x16)
1229 OBFP(BLOCK_16X8, aom_obmc_sad16x8, aom_obmc_variance16x8,
1230 aom_obmc_sub_pixel_variance16x8)
1231 OBFP(BLOCK_8X16, aom_obmc_sad8x16, aom_obmc_variance8x16,
1232 aom_obmc_sub_pixel_variance8x16)
1233 OBFP(BLOCK_8X8, aom_obmc_sad8x8, aom_obmc_variance8x8,
1234 aom_obmc_sub_pixel_variance8x8)
1235 OBFP(BLOCK_4X8, aom_obmc_sad4x8, aom_obmc_variance4x8,
1236 aom_obmc_sub_pixel_variance4x8)
1237 OBFP(BLOCK_8X4, aom_obmc_sad8x4, aom_obmc_variance8x4,
1238 aom_obmc_sub_pixel_variance8x4)
1239 OBFP(BLOCK_4X4, aom_obmc_sad4x4, aom_obmc_variance4x4,
1240 aom_obmc_sub_pixel_variance4x4)
1241 OBFP(BLOCK_4X16, aom_obmc_sad4x16, aom_obmc_variance4x16,
1242 aom_obmc_sub_pixel_variance4x16)
1243 OBFP(BLOCK_16X4, aom_obmc_sad16x4, aom_obmc_variance16x4,
1244 aom_obmc_sub_pixel_variance16x4)
1245 OBFP(BLOCK_8X32, aom_obmc_sad8x32, aom_obmc_variance8x32,
1246 aom_obmc_sub_pixel_variance8x32)
1247 OBFP(BLOCK_32X8, aom_obmc_sad32x8, aom_obmc_variance32x8,
1248 aom_obmc_sub_pixel_variance32x8)
1249 OBFP(BLOCK_16X64, aom_obmc_sad16x64, aom_obmc_variance16x64,
1250 aom_obmc_sub_pixel_variance16x64)
1251 OBFP(BLOCK_64X16, aom_obmc_sad64x16, aom_obmc_variance64x16,
1252 aom_obmc_sub_pixel_variance64x16)
1253 #endif // !CONFIG_REALTIME_ONLY
1254
1255 #define MBFP(BT, MCSDF, MCSVF) \
1256 cpi->fn_ptr[BT].msdf = MCSDF; \
1257 cpi->fn_ptr[BT].msvf = MCSVF;
1258
1259 MBFP(BLOCK_128X128, aom_masked_sad128x128,
1260 aom_masked_sub_pixel_variance128x128)
1261 MBFP(BLOCK_128X64, aom_masked_sad128x64, aom_masked_sub_pixel_variance128x64)
1262 MBFP(BLOCK_64X128, aom_masked_sad64x128, aom_masked_sub_pixel_variance64x128)
1263 MBFP(BLOCK_64X64, aom_masked_sad64x64, aom_masked_sub_pixel_variance64x64)
1264 MBFP(BLOCK_64X32, aom_masked_sad64x32, aom_masked_sub_pixel_variance64x32)
1265 MBFP(BLOCK_32X64, aom_masked_sad32x64, aom_masked_sub_pixel_variance32x64)
1266 MBFP(BLOCK_32X32, aom_masked_sad32x32, aom_masked_sub_pixel_variance32x32)
1267 MBFP(BLOCK_32X16, aom_masked_sad32x16, aom_masked_sub_pixel_variance32x16)
1268 MBFP(BLOCK_16X32, aom_masked_sad16x32, aom_masked_sub_pixel_variance16x32)
1269 MBFP(BLOCK_16X16, aom_masked_sad16x16, aom_masked_sub_pixel_variance16x16)
1270 MBFP(BLOCK_16X8, aom_masked_sad16x8, aom_masked_sub_pixel_variance16x8)
1271 MBFP(BLOCK_8X16, aom_masked_sad8x16, aom_masked_sub_pixel_variance8x16)
1272 MBFP(BLOCK_8X8, aom_masked_sad8x8, aom_masked_sub_pixel_variance8x8)
1273 MBFP(BLOCK_4X8, aom_masked_sad4x8, aom_masked_sub_pixel_variance4x8)
1274 MBFP(BLOCK_8X4, aom_masked_sad8x4, aom_masked_sub_pixel_variance8x4)
1275 MBFP(BLOCK_4X4, aom_masked_sad4x4, aom_masked_sub_pixel_variance4x4)
1276
1277 #if !CONFIG_REALTIME_ONLY
1278 MBFP(BLOCK_4X16, aom_masked_sad4x16, aom_masked_sub_pixel_variance4x16)
1279 MBFP(BLOCK_16X4, aom_masked_sad16x4, aom_masked_sub_pixel_variance16x4)
1280 MBFP(BLOCK_8X32, aom_masked_sad8x32, aom_masked_sub_pixel_variance8x32)
1281 MBFP(BLOCK_32X8, aom_masked_sad32x8, aom_masked_sub_pixel_variance32x8)
1282 MBFP(BLOCK_16X64, aom_masked_sad16x64, aom_masked_sub_pixel_variance16x64)
1283 MBFP(BLOCK_64X16, aom_masked_sad64x16, aom_masked_sub_pixel_variance64x16)
1284 #endif
1285
1286 #define SDSFP(BT, SDSF, SDSX4DF) \
1287 cpi->fn_ptr[BT].sdsf = SDSF; \
1288 cpi->fn_ptr[BT].sdsx4df = SDSX4DF;
1289
1290 SDSFP(BLOCK_128X128, aom_sad_skip_128x128, aom_sad_skip_128x128x4d);
1291 SDSFP(BLOCK_128X64, aom_sad_skip_128x64, aom_sad_skip_128x64x4d);
1292 SDSFP(BLOCK_64X128, aom_sad_skip_64x128, aom_sad_skip_64x128x4d);
1293 SDSFP(BLOCK_64X64, aom_sad_skip_64x64, aom_sad_skip_64x64x4d);
1294 SDSFP(BLOCK_64X32, aom_sad_skip_64x32, aom_sad_skip_64x32x4d);
1295
1296 SDSFP(BLOCK_32X64, aom_sad_skip_32x64, aom_sad_skip_32x64x4d);
1297 SDSFP(BLOCK_32X32, aom_sad_skip_32x32, aom_sad_skip_32x32x4d);
1298 SDSFP(BLOCK_32X16, aom_sad_skip_32x16, aom_sad_skip_32x16x4d);
1299
1300 SDSFP(BLOCK_16X32, aom_sad_skip_16x32, aom_sad_skip_16x32x4d);
1301 SDSFP(BLOCK_16X16, aom_sad_skip_16x16, aom_sad_skip_16x16x4d);
1302 SDSFP(BLOCK_16X8, aom_sad_skip_16x8, aom_sad_skip_16x8x4d);
1303 SDSFP(BLOCK_8X16, aom_sad_skip_8x16, aom_sad_skip_8x16x4d);
1304 SDSFP(BLOCK_8X8, aom_sad_skip_8x8, aom_sad_skip_8x8x4d);
1305
1306 SDSFP(BLOCK_4X8, aom_sad_skip_4x8, aom_sad_skip_4x8x4d);
1307
1308 #if !CONFIG_REALTIME_ONLY
1309 SDSFP(BLOCK_64X16, aom_sad_skip_64x16, aom_sad_skip_64x16x4d);
1310 SDSFP(BLOCK_16X64, aom_sad_skip_16x64, aom_sad_skip_16x64x4d);
1311 SDSFP(BLOCK_32X8, aom_sad_skip_32x8, aom_sad_skip_32x8x4d);
1312 SDSFP(BLOCK_8X32, aom_sad_skip_8x32, aom_sad_skip_8x32x4d);
1313 SDSFP(BLOCK_4X16, aom_sad_skip_4x16, aom_sad_skip_4x16x4d);
1314 #endif
1315 #undef SDSFP
1316
1317 #if CONFIG_AV1_HIGHBITDEPTH
1318 highbd_set_var_fns(cpi);
1319 #endif
1320
1321 /* av1_init_quantizer() is first called here. Add check in
1322 * av1_frame_init_quantizer() so that av1_init_quantizer is only
1323 * called later when needed. This will avoid unnecessary calls of
1324 * av1_init_quantizer() for every frame.
1325 */
1326 av1_init_quantizer(&cpi->enc_quant_dequant_params, &cm->quant_params,
1327 cm->seq_params.bit_depth);
1328 av1_qm_init(&cm->quant_params, av1_num_planes(cm));
1329
1330 av1_loop_filter_init(cm);
1331 cm->superres_scale_denominator = SCALE_NUMERATOR;
1332 cm->superres_upscaled_width = oxcf->frm_dim_cfg.width;
1333 cm->superres_upscaled_height = oxcf->frm_dim_cfg.height;
1334 #if !CONFIG_REALTIME_ONLY
1335 av1_loop_restoration_precal();
1336 #endif
1337 cm->error.setjmp = 0;
1338
1339 return cpi;
1340 }
1341
1342 #if CONFIG_INTERNAL_STATS
1343 #define SNPRINT(H, T) snprintf((H) + strlen(H), sizeof(H) - strlen(H), (T))
1344
1345 #define SNPRINT2(H, T, V) \
1346 snprintf((H) + strlen(H), sizeof(H) - strlen(H), (T), (V))
1347 #endif // CONFIG_INTERNAL_STATS
1348
1349 // This function will change the state and free the mutex of corresponding
1350 // workers and terminate the object. The object can not be re-used unless a call
1351 // to reset() is made.
terminate_worker_data(AV1_COMP * cpi)1352 static AOM_INLINE void terminate_worker_data(AV1_COMP *cpi) {
1353 MultiThreadInfo *const mt_info = &cpi->mt_info;
1354 for (int t = mt_info->num_workers - 1; t >= 0; --t) {
1355 AVxWorker *const worker = &mt_info->workers[t];
1356 aom_get_worker_interface()->end(worker);
1357 }
1358 }
1359
1360 // Deallocate allocated thread_data.
free_thread_data(AV1_COMP * cpi)1361 static AOM_INLINE void free_thread_data(AV1_COMP *cpi) {
1362 MultiThreadInfo *const mt_info = &cpi->mt_info;
1363 AV1_COMMON *cm = &cpi->common;
1364 for (int t = 0; t < mt_info->num_workers; ++t) {
1365 EncWorkerData *const thread_data = &mt_info->tile_thr_data[t];
1366 aom_free(thread_data->td->tctx);
1367 if (t == 0) continue;
1368 aom_free(thread_data->td->palette_buffer);
1369 aom_free(thread_data->td->tmp_conv_dst);
1370 release_compound_type_rd_buffers(&thread_data->td->comp_rd_buffer);
1371 for (int j = 0; j < 2; ++j) {
1372 aom_free(thread_data->td->tmp_pred_bufs[j]);
1373 }
1374 release_obmc_buffers(&thread_data->td->obmc_buffer);
1375 aom_free(thread_data->td->vt64x64);
1376
1377 aom_free(thread_data->td->inter_modes_info);
1378 for (int x = 0; x < 2; x++) {
1379 for (int y = 0; y < 2; y++) {
1380 aom_free(thread_data->td->hash_value_buffer[x][y]);
1381 thread_data->td->hash_value_buffer[x][y] = NULL;
1382 }
1383 }
1384 aom_free(thread_data->td->counts);
1385 av1_free_pmc(thread_data->td->firstpass_ctx, av1_num_planes(cm));
1386 thread_data->td->firstpass_ctx = NULL;
1387 av1_free_shared_coeff_buffer(&thread_data->td->shared_coeff_buf);
1388 av1_free_sms_tree(thread_data->td);
1389 aom_free(thread_data->td);
1390 }
1391 }
1392
av1_remove_compressor(AV1_COMP * cpi)1393 void av1_remove_compressor(AV1_COMP *cpi) {
1394 if (!cpi) return;
1395
1396 AV1_COMMON *cm = &cpi->common;
1397 if (cm->current_frame.frame_number > 0) {
1398 #if CONFIG_ENTROPY_STATS
1399 if (!is_stat_generation_stage(cpi)) {
1400 fprintf(stderr, "Writing counts.stt\n");
1401 FILE *f = fopen("counts.stt", "wb");
1402 fwrite(&aggregate_fc, sizeof(aggregate_fc), 1, f);
1403 fclose(f);
1404 }
1405 #endif // CONFIG_ENTROPY_STATS
1406 #if CONFIG_INTERNAL_STATS
1407 aom_clear_system_state();
1408
1409 if (!is_stat_generation_stage(cpi)) {
1410 char headings[512] = { 0 };
1411 char results[512] = { 0 };
1412 FILE *f = fopen("opsnr.stt", "a");
1413 double time_encoded =
1414 (cpi->time_stamps.prev_ts_end - cpi->time_stamps.first_ts_start) /
1415 10000000.000;
1416 double total_encode_time =
1417 (cpi->time_receive_data + cpi->time_compress_data) / 1000.000;
1418 const double dr =
1419 (double)cpi->bytes * (double)8 / (double)1000 / time_encoded;
1420 const double peak =
1421 (double)((1 << cpi->oxcf.input_cfg.input_bit_depth) - 1);
1422 const double target_rate =
1423 (double)cpi->oxcf.rc_cfg.target_bandwidth / 1000;
1424 const double rate_err = ((100.0 * (dr - target_rate)) / target_rate);
1425
1426 if (cpi->b_calculate_psnr) {
1427 const double total_psnr =
1428 aom_sse_to_psnr((double)cpi->total_samples[0], peak,
1429 (double)cpi->total_sq_error[0]);
1430 const double total_ssim =
1431 100 * pow(cpi->summed_quality / cpi->summed_weights, 8.0);
1432 snprintf(headings, sizeof(headings),
1433 "Bitrate\tAVGPsnr\tGLBPsnr\tAVPsnrP\tGLPsnrP\t"
1434 "AOMSSIM\tVPSSIMP\tFASTSIM\tPSNRHVS\t"
1435 "WstPsnr\tWstSsim\tWstFast\tWstHVS\t"
1436 "AVPsrnY\tAPsnrCb\tAPsnrCr");
1437 snprintf(results, sizeof(results),
1438 "%7.2f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t"
1439 "%7.3f\t%7.3f\t%7.3f\t%7.3f\t"
1440 "%7.3f\t%7.3f\t%7.3f\t%7.3f\t"
1441 "%7.3f\t%7.3f\t%7.3f",
1442 dr, cpi->psnr[0].stat[STAT_ALL] / cpi->count[0], total_psnr,
1443 cpi->psnr[0].stat[STAT_ALL] / cpi->count[0], total_psnr,
1444 total_ssim, total_ssim,
1445 cpi->fastssim.stat[STAT_ALL] / cpi->count[0],
1446 cpi->psnrhvs.stat[STAT_ALL] / cpi->count[0],
1447 cpi->psnr[0].worst, cpi->worst_ssim, cpi->fastssim.worst,
1448 cpi->psnrhvs.worst, cpi->psnr[0].stat[STAT_Y] / cpi->count[0],
1449 cpi->psnr[0].stat[STAT_U] / cpi->count[0],
1450 cpi->psnr[0].stat[STAT_V] / cpi->count[0]);
1451
1452 if (cpi->b_calculate_blockiness) {
1453 SNPRINT(headings, "\t Block\tWstBlck");
1454 SNPRINT2(results, "\t%7.3f", cpi->total_blockiness / cpi->count[0]);
1455 SNPRINT2(results, "\t%7.3f", cpi->worst_blockiness);
1456 }
1457
1458 if (cpi->b_calculate_consistency) {
1459 double consistency =
1460 aom_sse_to_psnr((double)cpi->total_samples[0], peak,
1461 (double)cpi->total_inconsistency);
1462
1463 SNPRINT(headings, "\tConsist\tWstCons");
1464 SNPRINT2(results, "\t%7.3f", consistency);
1465 SNPRINT2(results, "\t%7.3f", cpi->worst_consistency);
1466 }
1467
1468 SNPRINT(headings, "\t Time\tRcErr\tAbsErr");
1469 SNPRINT2(results, "\t%8.0f", total_encode_time);
1470 SNPRINT2(results, " %7.2f", rate_err);
1471 SNPRINT2(results, " %7.2f", fabs(rate_err));
1472
1473 SNPRINT(headings, "\tAPsnr611");
1474 SNPRINT2(results, " %7.3f",
1475 (6 * cpi->psnr[0].stat[STAT_Y] + cpi->psnr[0].stat[STAT_U] +
1476 cpi->psnr[0].stat[STAT_V]) /
1477 (cpi->count[0] * 8));
1478
1479 #if CONFIG_AV1_HIGHBITDEPTH
1480 const uint32_t in_bit_depth = cpi->oxcf.input_cfg.input_bit_depth;
1481 const uint32_t bit_depth = cpi->td.mb.e_mbd.bd;
1482 if ((cpi->source->flags & YV12_FLAG_HIGHBITDEPTH) &&
1483 (in_bit_depth < bit_depth)) {
1484 const double peak_hbd = (double)((1 << bit_depth) - 1);
1485 const double total_psnr_hbd =
1486 aom_sse_to_psnr((double)cpi->total_samples[1], peak_hbd,
1487 (double)cpi->total_sq_error[1]);
1488 const double total_ssim_hbd =
1489 100 * pow(cpi->summed_quality_hbd / cpi->summed_weights_hbd, 8.0);
1490 SNPRINT(headings,
1491 "\t AVGPsnrH GLBPsnrH AVPsnrPH GLPsnrPH"
1492 " AVPsnrYH APsnrCbH APsnrCrH WstPsnrH"
1493 " AOMSSIMH VPSSIMPH WstSsimH");
1494 SNPRINT2(results, "\t%7.3f",
1495 cpi->psnr[1].stat[STAT_ALL] / cpi->count[1]);
1496 SNPRINT2(results, " %7.3f", total_psnr_hbd);
1497 SNPRINT2(results, " %7.3f",
1498 cpi->psnr[1].stat[STAT_ALL] / cpi->count[1]);
1499 SNPRINT2(results, " %7.3f", total_psnr_hbd);
1500 SNPRINT2(results, " %7.3f",
1501 cpi->psnr[1].stat[STAT_Y] / cpi->count[1]);
1502 SNPRINT2(results, " %7.3f",
1503 cpi->psnr[1].stat[STAT_U] / cpi->count[1]);
1504 SNPRINT2(results, " %7.3f",
1505 cpi->psnr[1].stat[STAT_V] / cpi->count[1]);
1506 SNPRINT2(results, " %7.3f", cpi->psnr[1].worst);
1507 SNPRINT2(results, " %7.3f", total_ssim_hbd);
1508 SNPRINT2(results, " %7.3f", total_ssim_hbd);
1509 SNPRINT2(results, " %7.3f", cpi->worst_ssim_hbd);
1510 }
1511 #endif
1512 fprintf(f, "%s\n", headings);
1513 fprintf(f, "%s\n", results);
1514 }
1515
1516 fclose(f);
1517 }
1518 #endif // CONFIG_INTERNAL_STATS
1519 #if CONFIG_SPEED_STATS
1520 if (!is_stat_generation_stage(cpi)) {
1521 fprintf(stdout, "tx_search_count = %d\n", cpi->tx_search_count);
1522 }
1523 #endif // CONFIG_SPEED_STATS
1524
1525 #if CONFIG_COLLECT_PARTITION_STATS == 2
1526 if (!is_stat_generation_stage(cpi)) {
1527 av1_print_fr_partition_timing_stats(&cpi->partition_stats,
1528 "fr_part_timing_data.csv");
1529 }
1530 #endif
1531 }
1532
1533 #if CONFIG_AV1_TEMPORAL_DENOISING
1534 av1_denoiser_free(&(cpi->denoiser));
1535 #endif
1536
1537 TplParams *const tpl_data = &cpi->tpl_data;
1538 for (int frame = 0; frame < MAX_LAG_BUFFERS; ++frame) {
1539 aom_free(tpl_data->tpl_stats_pool[frame]);
1540 aom_free_frame_buffer(&tpl_data->tpl_rec_pool[frame]);
1541 }
1542
1543 if (cpi->compressor_stage != LAP_STAGE) {
1544 terminate_worker_data(cpi);
1545 free_thread_data(cpi);
1546 }
1547
1548 MultiThreadInfo *const mt_info = &cpi->mt_info;
1549 #if CONFIG_MULTITHREAD
1550 pthread_mutex_t *const enc_row_mt_mutex_ = mt_info->enc_row_mt.mutex_;
1551 pthread_mutex_t *const gm_mt_mutex_ = mt_info->gm_sync.mutex_;
1552 if (enc_row_mt_mutex_ != NULL) {
1553 pthread_mutex_destroy(enc_row_mt_mutex_);
1554 aom_free(enc_row_mt_mutex_);
1555 }
1556 if (gm_mt_mutex_ != NULL) {
1557 pthread_mutex_destroy(gm_mt_mutex_);
1558 aom_free(gm_mt_mutex_);
1559 }
1560 #endif
1561 av1_row_mt_mem_dealloc(cpi);
1562 if (cpi->compressor_stage != LAP_STAGE) {
1563 aom_free(mt_info->tile_thr_data);
1564 aom_free(mt_info->workers);
1565 }
1566
1567 #if !CONFIG_REALTIME_ONLY
1568 av1_tpl_dealloc(&tpl_data->tpl_mt_sync);
1569 #endif
1570 if (mt_info->num_workers > 1) {
1571 av1_loop_filter_dealloc(&mt_info->lf_row_sync);
1572 av1_cdef_mt_dealloc(&mt_info->cdef_sync);
1573 #if !CONFIG_REALTIME_ONLY
1574 av1_loop_restoration_dealloc(&mt_info->lr_row_sync,
1575 mt_info->num_mod_workers[MOD_LR]);
1576 av1_gm_dealloc(&mt_info->gm_sync);
1577 av1_tf_mt_dealloc(&mt_info->tf_sync);
1578 #endif
1579 }
1580
1581 dealloc_compressor_data(cpi);
1582
1583 #if CONFIG_INTERNAL_STATS
1584 aom_free(cpi->ssim_vars);
1585 cpi->ssim_vars = NULL;
1586 #endif // CONFIG_INTERNAL_STATS
1587
1588 av1_remove_common(cm);
1589 av1_free_ref_frame_buffers(cm->buffer_pool);
1590
1591 aom_free(cpi);
1592
1593 #ifdef OUTPUT_YUV_REC
1594 fclose(yuv_rec_file);
1595 #endif
1596
1597 #ifdef OUTPUT_YUV_DENOISED
1598 fclose(yuv_denoised_file);
1599 #endif
1600 }
1601
generate_psnr_packet(AV1_COMP * cpi)1602 static void generate_psnr_packet(AV1_COMP *cpi) {
1603 struct aom_codec_cx_pkt pkt;
1604 int i;
1605 PSNR_STATS psnr;
1606 #if CONFIG_AV1_HIGHBITDEPTH
1607 const uint32_t in_bit_depth = cpi->oxcf.input_cfg.input_bit_depth;
1608 const uint32_t bit_depth = cpi->td.mb.e_mbd.bd;
1609 aom_calc_highbd_psnr(cpi->source, &cpi->common.cur_frame->buf, &psnr,
1610 bit_depth, in_bit_depth);
1611 #else
1612 aom_calc_psnr(cpi->source, &cpi->common.cur_frame->buf, &psnr);
1613 #endif
1614
1615 for (i = 0; i < 4; ++i) {
1616 pkt.data.psnr.samples[i] = psnr.samples[i];
1617 pkt.data.psnr.sse[i] = psnr.sse[i];
1618 pkt.data.psnr.psnr[i] = psnr.psnr[i];
1619 }
1620
1621 #if CONFIG_AV1_HIGHBITDEPTH
1622 if ((cpi->source->flags & YV12_FLAG_HIGHBITDEPTH) &&
1623 (in_bit_depth < bit_depth)) {
1624 for (i = 0; i < 4; ++i) {
1625 pkt.data.psnr.samples_hbd[i] = psnr.samples_hbd[i];
1626 pkt.data.psnr.sse_hbd[i] = psnr.sse_hbd[i];
1627 pkt.data.psnr.psnr_hbd[i] = psnr.psnr_hbd[i];
1628 }
1629 }
1630 #endif
1631
1632 pkt.kind = AOM_CODEC_PSNR_PKT;
1633 aom_codec_pkt_list_add(cpi->output_pkt_list, &pkt);
1634 }
1635
av1_use_as_reference(int * ext_ref_frame_flags,int ref_frame_flags)1636 int av1_use_as_reference(int *ext_ref_frame_flags, int ref_frame_flags) {
1637 if (ref_frame_flags > ((1 << INTER_REFS_PER_FRAME) - 1)) return -1;
1638
1639 *ext_ref_frame_flags = ref_frame_flags;
1640 return 0;
1641 }
1642
av1_copy_reference_enc(AV1_COMP * cpi,int idx,YV12_BUFFER_CONFIG * sd)1643 int av1_copy_reference_enc(AV1_COMP *cpi, int idx, YV12_BUFFER_CONFIG *sd) {
1644 AV1_COMMON *const cm = &cpi->common;
1645 const int num_planes = av1_num_planes(cm);
1646 YV12_BUFFER_CONFIG *cfg = get_ref_frame(cm, idx);
1647 if (cfg) {
1648 aom_yv12_copy_frame(cfg, sd, num_planes);
1649 return 0;
1650 } else {
1651 return -1;
1652 }
1653 }
1654
av1_set_reference_enc(AV1_COMP * cpi,int idx,YV12_BUFFER_CONFIG * sd)1655 int av1_set_reference_enc(AV1_COMP *cpi, int idx, YV12_BUFFER_CONFIG *sd) {
1656 AV1_COMMON *const cm = &cpi->common;
1657 const int num_planes = av1_num_planes(cm);
1658 YV12_BUFFER_CONFIG *cfg = get_ref_frame(cm, idx);
1659 if (cfg) {
1660 aom_yv12_copy_frame(sd, cfg, num_planes);
1661 return 0;
1662 } else {
1663 return -1;
1664 }
1665 }
1666
1667 #ifdef OUTPUT_YUV_REC
aom_write_one_yuv_frame(AV1_COMMON * cm,YV12_BUFFER_CONFIG * s)1668 void aom_write_one_yuv_frame(AV1_COMMON *cm, YV12_BUFFER_CONFIG *s) {
1669 uint8_t *src = s->y_buffer;
1670 int h = cm->height;
1671 if (yuv_rec_file == NULL) return;
1672 if (s->flags & YV12_FLAG_HIGHBITDEPTH) {
1673 uint16_t *src16 = CONVERT_TO_SHORTPTR(s->y_buffer);
1674
1675 do {
1676 fwrite(src16, s->y_width, 2, yuv_rec_file);
1677 src16 += s->y_stride;
1678 } while (--h);
1679
1680 src16 = CONVERT_TO_SHORTPTR(s->u_buffer);
1681 h = s->uv_height;
1682
1683 do {
1684 fwrite(src16, s->uv_width, 2, yuv_rec_file);
1685 src16 += s->uv_stride;
1686 } while (--h);
1687
1688 src16 = CONVERT_TO_SHORTPTR(s->v_buffer);
1689 h = s->uv_height;
1690
1691 do {
1692 fwrite(src16, s->uv_width, 2, yuv_rec_file);
1693 src16 += s->uv_stride;
1694 } while (--h);
1695
1696 fflush(yuv_rec_file);
1697 return;
1698 }
1699
1700 do {
1701 fwrite(src, s->y_width, 1, yuv_rec_file);
1702 src += s->y_stride;
1703 } while (--h);
1704
1705 src = s->u_buffer;
1706 h = s->uv_height;
1707
1708 do {
1709 fwrite(src, s->uv_width, 1, yuv_rec_file);
1710 src += s->uv_stride;
1711 } while (--h);
1712
1713 src = s->v_buffer;
1714 h = s->uv_height;
1715
1716 do {
1717 fwrite(src, s->uv_width, 1, yuv_rec_file);
1718 src += s->uv_stride;
1719 } while (--h);
1720
1721 fflush(yuv_rec_file);
1722 }
1723 #endif // OUTPUT_YUV_REC
1724
set_mv_search_params(AV1_COMP * cpi)1725 static void set_mv_search_params(AV1_COMP *cpi) {
1726 const AV1_COMMON *const cm = &cpi->common;
1727 MotionVectorSearchParams *const mv_search_params = &cpi->mv_search_params;
1728 const int max_mv_def = AOMMAX(cm->width, cm->height);
1729
1730 // Default based on max resolution.
1731 mv_search_params->mv_step_param = av1_init_search_range(max_mv_def);
1732
1733 if (cpi->sf.mv_sf.auto_mv_step_size) {
1734 if (frame_is_intra_only(cm)) {
1735 // Initialize max_mv_magnitude for use in the first INTER frame
1736 // after a key/intra-only frame.
1737 mv_search_params->max_mv_magnitude = max_mv_def;
1738 } else {
1739 // Use cpi->max_mv_magnitude == -1 to exclude first pass case.
1740 if (cm->show_frame && mv_search_params->max_mv_magnitude != -1) {
1741 // Allow mv_steps to correspond to twice the max mv magnitude found
1742 // in the previous frame, capped by the default max_mv_magnitude based
1743 // on resolution.
1744 mv_search_params->mv_step_param = av1_init_search_range(
1745 AOMMIN(max_mv_def, 2 * mv_search_params->max_mv_magnitude));
1746 }
1747 mv_search_params->max_mv_magnitude = -1;
1748 }
1749 }
1750 }
1751
av1_set_screen_content_options(AV1_COMP * cpi,FeatureFlags * features)1752 void av1_set_screen_content_options(AV1_COMP *cpi, FeatureFlags *features) {
1753 const AV1_COMMON *const cm = &cpi->common;
1754
1755 if (cm->seq_params.force_screen_content_tools != 2) {
1756 features->allow_screen_content_tools = features->allow_intrabc =
1757 cm->seq_params.force_screen_content_tools;
1758 return;
1759 }
1760
1761 if (cpi->oxcf.mode == REALTIME) {
1762 assert(cm->seq_params.reduced_still_picture_hdr);
1763 features->allow_screen_content_tools = features->allow_intrabc = 0;
1764 return;
1765 }
1766
1767 if (cpi->oxcf.tune_cfg.content == AOM_CONTENT_SCREEN) {
1768 features->allow_screen_content_tools = features->allow_intrabc = 1;
1769 return;
1770 }
1771
1772 // Estimate if the source frame is screen content, based on the portion of
1773 // blocks that have few luma colors.
1774 const uint8_t *src = cpi->unfiltered_source->y_buffer;
1775 assert(src != NULL);
1776 const int use_hbd = cpi->unfiltered_source->flags & YV12_FLAG_HIGHBITDEPTH;
1777 const int stride = cpi->unfiltered_source->y_stride;
1778 const int width = cpi->unfiltered_source->y_width;
1779 const int height = cpi->unfiltered_source->y_height;
1780 const int bd = cm->seq_params.bit_depth;
1781 const int blk_w = 16;
1782 const int blk_h = 16;
1783 // These threshold values are selected experimentally.
1784 const int color_thresh = 4;
1785 const unsigned int var_thresh = 0;
1786 // Counts of blocks with no more than color_thresh colors.
1787 int counts_1 = 0;
1788 // Counts of blocks with no more than color_thresh colors and variance larger
1789 // than var_thresh.
1790 int counts_2 = 0;
1791
1792 for (int r = 0; r + blk_h <= height; r += blk_h) {
1793 for (int c = 0; c + blk_w <= width; c += blk_w) {
1794 int count_buf[1 << 8]; // Maximum (1 << 8) bins for hbd path.
1795 const uint8_t *const this_src = src + r * stride + c;
1796 int n_colors;
1797 if (use_hbd)
1798 av1_count_colors_highbd(this_src, stride, blk_w, blk_h, bd, NULL,
1799 count_buf, &n_colors, NULL);
1800 else
1801 av1_count_colors(this_src, stride, blk_w, blk_h, count_buf, &n_colors);
1802 if (n_colors > 1 && n_colors <= color_thresh) {
1803 ++counts_1;
1804 struct buf_2d buf;
1805 buf.stride = stride;
1806 buf.buf = (uint8_t *)this_src;
1807 const unsigned int var =
1808 use_hbd
1809 ? av1_high_get_sby_perpixel_variance(cpi, &buf, BLOCK_16X16, bd)
1810 : av1_get_sby_perpixel_variance(cpi, &buf, BLOCK_16X16);
1811 if (var > var_thresh) ++counts_2;
1812 }
1813 }
1814 }
1815
1816 // The threshold values are selected experimentally.
1817 features->allow_screen_content_tools =
1818 counts_1 * blk_h * blk_w * 10 > width * height;
1819 // IntraBC would force loop filters off, so we use more strict rules that also
1820 // requires that the block has high variance.
1821 features->allow_intrabc = features->allow_screen_content_tools &&
1822 counts_2 * blk_h * blk_w * 12 > width * height;
1823 cpi->use_screen_content_tools = features->allow_screen_content_tools;
1824 cpi->is_screen_content_type =
1825 features->allow_intrabc ||
1826 (counts_1 * blk_h * blk_w * 10 > width * height * 4 &&
1827 counts_2 * blk_h * blk_w * 30 > width * height);
1828 }
1829
1830 // Function pointer to search site config initialization
1831 // of different search method functions.
1832 typedef void (*av1_init_search_site_config)(search_site_config *cfg, int stride,
1833 int level);
1834
1835 av1_init_search_site_config
1836 av1_init_motion_compensation[NUM_DISTINCT_SEARCH_METHODS] = {
1837 av1_init_dsmotion_compensation, av1_init_motion_compensation_nstep,
1838 av1_init_motion_compensation_nstep, av1_init_dsmotion_compensation,
1839 av1_init_motion_compensation_hex, av1_init_motion_compensation_bigdia,
1840 av1_init_motion_compensation_square
1841 };
1842
init_motion_estimation(AV1_COMP * cpi)1843 static void init_motion_estimation(AV1_COMP *cpi) {
1844 AV1_COMMON *const cm = &cpi->common;
1845 MotionVectorSearchParams *const mv_search_params = &cpi->mv_search_params;
1846 const int y_stride = cpi->scaled_source.y_stride;
1847 const int y_stride_src = ((cpi->oxcf.frm_dim_cfg.width != cm->width ||
1848 cpi->oxcf.frm_dim_cfg.height != cm->height) ||
1849 av1_superres_scaled(cm))
1850 ? y_stride
1851 : cpi->lookahead->buf->img.y_stride;
1852 int fpf_y_stride = cm->cur_frame != NULL ? cm->cur_frame->buf.y_stride
1853 : cpi->scaled_source.y_stride;
1854
1855 // Update if search_site_cfg is uninitialized or the current frame has a new
1856 // stride
1857 const int should_update =
1858 !mv_search_params->search_site_cfg[SS_CFG_SRC][DIAMOND].stride ||
1859 !mv_search_params->search_site_cfg[SS_CFG_LOOKAHEAD][DIAMOND].stride ||
1860 (y_stride !=
1861 mv_search_params->search_site_cfg[SS_CFG_SRC][DIAMOND].stride);
1862
1863 if (!should_update) {
1864 return;
1865 }
1866
1867 // Initialization of search_site_cfg for NUM_DISTINCT_SEARCH_METHODS.
1868 for (SEARCH_METHODS i = DIAMOND; i < NUM_DISTINCT_SEARCH_METHODS; i++) {
1869 const int level = ((i == NSTEP_8PT) || (i == CLAMPED_DIAMOND)) ? 1 : 0;
1870 av1_init_motion_compensation[i](
1871 &mv_search_params->search_site_cfg[SS_CFG_SRC][i], y_stride, level);
1872 av1_init_motion_compensation[i](
1873 &mv_search_params->search_site_cfg[SS_CFG_LOOKAHEAD][i], y_stride_src,
1874 level);
1875 }
1876
1877 // First pass search site config initialization.
1878 av1_init_motion_fpf(&mv_search_params->search_site_cfg[SS_CFG_FPF][DIAMOND],
1879 fpf_y_stride);
1880 for (SEARCH_METHODS i = NSTEP; i < NUM_DISTINCT_SEARCH_METHODS; i++) {
1881 memcpy(&mv_search_params->search_site_cfg[SS_CFG_FPF][i],
1882 &mv_search_params->search_site_cfg[SS_CFG_FPF][DIAMOND],
1883 sizeof(search_site_config));
1884 }
1885 }
1886
1887 #if !CONFIG_REALTIME_ONLY
1888 #define COUPLED_CHROMA_FROM_LUMA_RESTORATION 0
set_restoration_unit_size(int width,int height,int sx,int sy,RestorationInfo * rst)1889 static void set_restoration_unit_size(int width, int height, int sx, int sy,
1890 RestorationInfo *rst) {
1891 (void)width;
1892 (void)height;
1893 (void)sx;
1894 (void)sy;
1895 #if COUPLED_CHROMA_FROM_LUMA_RESTORATION
1896 int s = AOMMIN(sx, sy);
1897 #else
1898 int s = 0;
1899 #endif // !COUPLED_CHROMA_FROM_LUMA_RESTORATION
1900
1901 if (width * height > 352 * 288)
1902 rst[0].restoration_unit_size = RESTORATION_UNITSIZE_MAX;
1903 else
1904 rst[0].restoration_unit_size = (RESTORATION_UNITSIZE_MAX >> 1);
1905 rst[1].restoration_unit_size = rst[0].restoration_unit_size >> s;
1906 rst[2].restoration_unit_size = rst[1].restoration_unit_size;
1907 }
1908 #endif
1909
init_ref_frame_bufs(AV1_COMP * cpi)1910 static void init_ref_frame_bufs(AV1_COMP *cpi) {
1911 AV1_COMMON *const cm = &cpi->common;
1912 int i;
1913 BufferPool *const pool = cm->buffer_pool;
1914 cm->cur_frame = NULL;
1915 for (i = 0; i < REF_FRAMES; ++i) {
1916 cm->ref_frame_map[i] = NULL;
1917 }
1918 for (i = 0; i < FRAME_BUFFERS; ++i) {
1919 pool->frame_bufs[i].ref_count = 0;
1920 }
1921 }
1922
av1_check_initial_width(AV1_COMP * cpi,int use_highbitdepth,int subsampling_x,int subsampling_y)1923 void av1_check_initial_width(AV1_COMP *cpi, int use_highbitdepth,
1924 int subsampling_x, int subsampling_y) {
1925 AV1_COMMON *const cm = &cpi->common;
1926 SequenceHeader *const seq_params = &cm->seq_params;
1927 InitialDimensions *const initial_dimensions = &cpi->initial_dimensions;
1928
1929 if (!initial_dimensions->width ||
1930 seq_params->use_highbitdepth != use_highbitdepth ||
1931 seq_params->subsampling_x != subsampling_x ||
1932 seq_params->subsampling_y != subsampling_y) {
1933 seq_params->subsampling_x = subsampling_x;
1934 seq_params->subsampling_y = subsampling_y;
1935 seq_params->use_highbitdepth = use_highbitdepth;
1936
1937 av1_set_speed_features_framesize_independent(cpi, cpi->oxcf.speed);
1938 av1_set_speed_features_framesize_dependent(cpi, cpi->oxcf.speed);
1939
1940 if (!is_stat_generation_stage(cpi)) {
1941 alloc_altref_frame_buffer(cpi);
1942 alloc_util_frame_buffers(cpi);
1943 }
1944 init_ref_frame_bufs(cpi);
1945
1946 init_motion_estimation(cpi); // TODO(agrange) This can be removed.
1947
1948 initial_dimensions->width = cm->width;
1949 initial_dimensions->height = cm->height;
1950 cpi->initial_mbs = cm->mi_params.MBs;
1951 }
1952 }
1953
1954 #if CONFIG_AV1_TEMPORAL_DENOISING
setup_denoiser_buffer(AV1_COMP * cpi)1955 static void setup_denoiser_buffer(AV1_COMP *cpi) {
1956 AV1_COMMON *const cm = &cpi->common;
1957 if (cpi->oxcf.noise_sensitivity > 0 &&
1958 !cpi->denoiser.frame_buffer_initialized) {
1959 if (av1_denoiser_alloc(
1960 cm, &cpi->svc, &cpi->denoiser, cpi->use_svc,
1961 cpi->oxcf.noise_sensitivity, cm->width, cm->height,
1962 cm->seq_params.subsampling_x, cm->seq_params.subsampling_y,
1963 cm->seq_params.use_highbitdepth, AOM_BORDER_IN_PIXELS))
1964 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
1965 "Failed to allocate denoiser");
1966 }
1967 }
1968 #endif
1969
1970 // Returns 1 if the assigned width or height was <= 0.
av1_set_size_literal(AV1_COMP * cpi,int width,int height)1971 int av1_set_size_literal(AV1_COMP *cpi, int width, int height) {
1972 AV1_COMMON *cm = &cpi->common;
1973 InitialDimensions *const initial_dimensions = &cpi->initial_dimensions;
1974 av1_check_initial_width(cpi, cm->seq_params.use_highbitdepth,
1975 cm->seq_params.subsampling_x,
1976 cm->seq_params.subsampling_y);
1977
1978 if (width <= 0 || height <= 0) return 1;
1979
1980 cm->width = width;
1981 cm->height = height;
1982
1983 #if CONFIG_AV1_TEMPORAL_DENOISING
1984 setup_denoiser_buffer(cpi);
1985 #endif
1986
1987 if (initial_dimensions->width && initial_dimensions->height &&
1988 (cm->width > initial_dimensions->width ||
1989 cm->height > initial_dimensions->height)) {
1990 av1_free_context_buffers(cm);
1991 av1_free_shared_coeff_buffer(&cpi->td.shared_coeff_buf);
1992 av1_free_sms_tree(&cpi->td);
1993 av1_free_pmc(cpi->td.firstpass_ctx, av1_num_planes(cm));
1994 cpi->td.firstpass_ctx = NULL;
1995 alloc_compressor_data(cpi);
1996 realloc_segmentation_maps(cpi);
1997 initial_dimensions->width = initial_dimensions->height = 0;
1998 }
1999 av1_update_frame_size(cpi);
2000
2001 return 0;
2002 }
2003
av1_set_frame_size(AV1_COMP * cpi,int width,int height)2004 void av1_set_frame_size(AV1_COMP *cpi, int width, int height) {
2005 AV1_COMMON *const cm = &cpi->common;
2006 const SequenceHeader *const seq_params = &cm->seq_params;
2007 const int num_planes = av1_num_planes(cm);
2008 MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
2009 int ref_frame;
2010
2011 if (width != cm->width || height != cm->height) {
2012 // There has been a change in the encoded frame size
2013 av1_set_size_literal(cpi, width, height);
2014 // Recalculate 'all_lossless' in case super-resolution was (un)selected.
2015 cm->features.all_lossless =
2016 cm->features.coded_lossless && !av1_superres_scaled(cm);
2017
2018 av1_noise_estimate_init(&cpi->noise_estimate, cm->width, cm->height);
2019 #if CONFIG_AV1_TEMPORAL_DENOISING
2020 // Reset the denoiser on the resized frame.
2021 if (cpi->oxcf.noise_sensitivity > 0) {
2022 av1_denoiser_free(&(cpi->denoiser));
2023 setup_denoiser_buffer(cpi);
2024 }
2025 #endif
2026 }
2027 set_mv_search_params(cpi);
2028
2029 if (is_stat_consumption_stage(cpi)) {
2030 av1_set_target_rate(cpi, cm->width, cm->height);
2031 }
2032
2033 alloc_frame_mvs(cm, cm->cur_frame);
2034
2035 // Allocate above context buffers
2036 CommonContexts *const above_contexts = &cm->above_contexts;
2037 if (above_contexts->num_planes < av1_num_planes(cm) ||
2038 above_contexts->num_mi_cols < cm->mi_params.mi_cols ||
2039 above_contexts->num_tile_rows < cm->tiles.rows) {
2040 av1_free_above_context_buffers(above_contexts);
2041 if (av1_alloc_above_context_buffers(above_contexts, cm->tiles.rows,
2042 cm->mi_params.mi_cols,
2043 av1_num_planes(cm)))
2044 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
2045 "Failed to allocate context buffers");
2046 }
2047
2048 // Reset the frame pointers to the current frame size.
2049 if (aom_realloc_frame_buffer(
2050 &cm->cur_frame->buf, cm->width, cm->height, seq_params->subsampling_x,
2051 seq_params->subsampling_y, seq_params->use_highbitdepth,
2052 cpi->oxcf.border_in_pixels, cm->features.byte_alignment, NULL, NULL,
2053 NULL))
2054 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
2055 "Failed to allocate frame buffer");
2056
2057 #if !CONFIG_REALTIME_ONLY
2058 const int use_restoration = cm->seq_params.enable_restoration &&
2059 !cm->features.all_lossless &&
2060 !cm->tiles.large_scale;
2061 if (use_restoration) {
2062 const int frame_width = cm->superres_upscaled_width;
2063 const int frame_height = cm->superres_upscaled_height;
2064 set_restoration_unit_size(frame_width, frame_height,
2065 seq_params->subsampling_x,
2066 seq_params->subsampling_y, cm->rst_info);
2067 for (int i = 0; i < num_planes; ++i)
2068 cm->rst_info[i].frame_restoration_type = RESTORE_NONE;
2069
2070 av1_alloc_restoration_buffers(cm);
2071 }
2072 #endif
2073 if (!is_stat_generation_stage(cpi)) alloc_util_frame_buffers(cpi);
2074 init_motion_estimation(cpi);
2075
2076 for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
2077 RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame);
2078 if (buf != NULL) {
2079 struct scale_factors *sf = get_ref_scale_factors(cm, ref_frame);
2080 av1_setup_scale_factors_for_frame(sf, buf->buf.y_crop_width,
2081 buf->buf.y_crop_height, cm->width,
2082 cm->height);
2083 if (av1_is_scaled(sf)) aom_extend_frame_borders(&buf->buf, num_planes);
2084 }
2085 }
2086
2087 av1_setup_scale_factors_for_frame(&cm->sf_identity, cm->width, cm->height,
2088 cm->width, cm->height);
2089
2090 set_ref_ptrs(cm, xd, LAST_FRAME, LAST_FRAME);
2091 }
2092
2093 /*!\brief Select and apply cdef filters and switchable restoration filters
2094 *
2095 * \ingroup high_level_algo
2096 */
cdef_restoration_frame(AV1_COMP * cpi,AV1_COMMON * cm,MACROBLOCKD * xd,int use_restoration,int use_cdef)2097 static void cdef_restoration_frame(AV1_COMP *cpi, AV1_COMMON *cm,
2098 MACROBLOCKD *xd, int use_restoration,
2099 int use_cdef) {
2100 #if !CONFIG_REALTIME_ONLY
2101 if (use_restoration)
2102 av1_loop_restoration_save_boundary_lines(&cm->cur_frame->buf, cm, 0);
2103 #else
2104 (void)use_restoration;
2105 #endif
2106
2107 if (use_cdef) {
2108 #if CONFIG_COLLECT_COMPONENT_TIMING
2109 start_timing(cpi, cdef_time);
2110 #endif
2111 // Find CDEF parameters
2112 av1_cdef_search(&cpi->mt_info, &cm->cur_frame->buf, cpi->source, cm, xd,
2113 cpi->sf.lpf_sf.cdef_pick_method, cpi->td.mb.rdmult);
2114
2115 // Apply the filter
2116 if (!cpi->sf.rt_sf.skip_loopfilter_non_reference)
2117 av1_cdef_frame(&cm->cur_frame->buf, cm, xd);
2118 #if CONFIG_COLLECT_COMPONENT_TIMING
2119 end_timing(cpi, cdef_time);
2120 #endif
2121 } else {
2122 cm->cdef_info.cdef_bits = 0;
2123 cm->cdef_info.cdef_strengths[0] = 0;
2124 cm->cdef_info.nb_cdef_strengths = 1;
2125 cm->cdef_info.cdef_uv_strengths[0] = 0;
2126 }
2127
2128 av1_superres_post_encode(cpi);
2129
2130 #if !CONFIG_REALTIME_ONLY
2131 #if CONFIG_COLLECT_COMPONENT_TIMING
2132 start_timing(cpi, loop_restoration_time);
2133 #endif
2134 if (use_restoration) {
2135 MultiThreadInfo *const mt_info = &cpi->mt_info;
2136 const int num_workers = mt_info->num_mod_workers[MOD_LR];
2137 av1_loop_restoration_save_boundary_lines(&cm->cur_frame->buf, cm, 1);
2138 av1_pick_filter_restoration(cpi->source, cpi);
2139 if (cm->rst_info[0].frame_restoration_type != RESTORE_NONE ||
2140 cm->rst_info[1].frame_restoration_type != RESTORE_NONE ||
2141 cm->rst_info[2].frame_restoration_type != RESTORE_NONE) {
2142 if (num_workers > 1)
2143 av1_loop_restoration_filter_frame_mt(
2144 &cm->cur_frame->buf, cm, 0, mt_info->workers, num_workers,
2145 &mt_info->lr_row_sync, &cpi->lr_ctxt);
2146 else
2147 av1_loop_restoration_filter_frame(&cm->cur_frame->buf, cm, 0,
2148 &cpi->lr_ctxt);
2149 }
2150 } else {
2151 cm->rst_info[0].frame_restoration_type = RESTORE_NONE;
2152 cm->rst_info[1].frame_restoration_type = RESTORE_NONE;
2153 cm->rst_info[2].frame_restoration_type = RESTORE_NONE;
2154 }
2155 #if CONFIG_COLLECT_COMPONENT_TIMING
2156 end_timing(cpi, loop_restoration_time);
2157 #endif
2158 #endif // !CONFIG_REALTIME_ONLY
2159 }
2160
2161 /*!\brief Select and apply in-loop deblocking filters, cdef filters, and
2162 * restoration filters
2163 *
2164 * \ingroup high_level_algo
2165 */
loopfilter_frame(AV1_COMP * cpi,AV1_COMMON * cm)2166 static void loopfilter_frame(AV1_COMP *cpi, AV1_COMMON *cm) {
2167 MultiThreadInfo *const mt_info = &cpi->mt_info;
2168 const int num_workers = mt_info->num_mod_workers[MOD_LPF];
2169 const int num_planes = av1_num_planes(cm);
2170 MACROBLOCKD *xd = &cpi->td.mb.e_mbd;
2171
2172 assert(IMPLIES(is_lossless_requested(&cpi->oxcf.rc_cfg),
2173 cm->features.coded_lossless && cm->features.all_lossless));
2174
2175 const int use_loopfilter =
2176 !cm->features.coded_lossless && !cm->tiles.large_scale;
2177 const int use_cdef = cm->seq_params.enable_cdef &&
2178 !cm->features.coded_lossless && !cm->tiles.large_scale;
2179 const int use_restoration = cm->seq_params.enable_restoration &&
2180 !cm->features.all_lossless &&
2181 !cm->tiles.large_scale;
2182
2183 struct loopfilter *lf = &cm->lf;
2184
2185 #if CONFIG_COLLECT_COMPONENT_TIMING
2186 start_timing(cpi, loop_filter_time);
2187 #endif
2188 if (use_loopfilter) {
2189 aom_clear_system_state();
2190 av1_pick_filter_level(cpi->source, cpi, cpi->sf.lpf_sf.lpf_pick);
2191 } else {
2192 lf->filter_level[0] = 0;
2193 lf->filter_level[1] = 0;
2194 }
2195
2196 if ((lf->filter_level[0] || lf->filter_level[1]) &&
2197 !cpi->sf.rt_sf.skip_loopfilter_non_reference) {
2198 if (num_workers > 1)
2199 av1_loop_filter_frame_mt(&cm->cur_frame->buf, cm, xd, 0, num_planes, 0,
2200 #if CONFIG_LPF_MASK
2201 0,
2202 #endif
2203 mt_info->workers, num_workers,
2204 &mt_info->lf_row_sync);
2205 else
2206 av1_loop_filter_frame(&cm->cur_frame->buf, cm, xd,
2207 #if CONFIG_LPF_MASK
2208 0,
2209 #endif
2210 0, num_planes, 0);
2211 }
2212 #if CONFIG_COLLECT_COMPONENT_TIMING
2213 end_timing(cpi, loop_filter_time);
2214 #endif
2215
2216 cdef_restoration_frame(cpi, cm, xd, use_restoration, use_cdef);
2217 }
2218
2219 /*!\brief Encode a frame without the recode loop, usually used in one-pass
2220 * encoding and realtime coding.
2221 *
2222 * \ingroup high_level_algo
2223 *
2224 * \param[in] cpi Top-level encoder structure
2225 *
2226 * \return Returns a value to indicate if the encoding is done successfully.
2227 * \retval #AOM_CODEC_OK
2228 * \retval #AOM_CODEC_ERROR
2229 */
encode_without_recode(AV1_COMP * cpi)2230 static int encode_without_recode(AV1_COMP *cpi) {
2231 AV1_COMMON *const cm = &cpi->common;
2232 const QuantizationCfg *const q_cfg = &cpi->oxcf.q_cfg;
2233 SVC *const svc = &cpi->svc;
2234 ResizePendingParams *const resize_pending_params =
2235 &cpi->resize_pending_params;
2236 const int resize_pending =
2237 (resize_pending_params->width && resize_pending_params->height &&
2238 (cpi->common.width != resize_pending_params->width ||
2239 cpi->common.height != resize_pending_params->height));
2240
2241 int top_index = 0, bottom_index = 0, q = 0;
2242 YV12_BUFFER_CONFIG *unscaled = cpi->unscaled_source;
2243 InterpFilter filter_scaler =
2244 cpi->use_svc ? svc->downsample_filter_type[svc->spatial_layer_id]
2245 : EIGHTTAP_SMOOTH;
2246 int phase_scaler =
2247 cpi->use_svc ? svc->downsample_filter_phase[svc->spatial_layer_id] : 0;
2248
2249 set_size_independent_vars(cpi);
2250 av1_setup_frame_size(cpi);
2251 av1_set_size_dependent_vars(cpi, &q, &bottom_index, &top_index);
2252
2253 if (!cpi->use_svc) {
2254 phase_scaler = 8;
2255 // 2:1 scaling.
2256 if ((cm->width << 1) == unscaled->y_crop_width &&
2257 (cm->height << 1) == unscaled->y_crop_height) {
2258 filter_scaler = BILINEAR;
2259 // For lower resolutions use eighttap_smooth.
2260 if (cm->width * cm->height <= 320 * 180) filter_scaler = EIGHTTAP_SMOOTH;
2261 } else if ((cm->width << 2) == unscaled->y_crop_width &&
2262 (cm->height << 2) == unscaled->y_crop_height) {
2263 // 4:1 scaling.
2264 filter_scaler = EIGHTTAP_SMOOTH;
2265 } else if ((cm->width << 2) == 3 * unscaled->y_crop_width &&
2266 (cm->height << 2) == 3 * unscaled->y_crop_height) {
2267 // 4:3 scaling.
2268 filter_scaler = EIGHTTAP_REGULAR;
2269 }
2270 }
2271
2272 if (cpi->sf.part_sf.partition_search_type == VAR_BASED_PARTITION)
2273 variance_partition_alloc(cpi);
2274
2275 if (cm->current_frame.frame_type == KEY_FRAME) copy_frame_prob_info(cpi);
2276
2277 #if CONFIG_COLLECT_COMPONENT_TIMING
2278 printf("\n Encoding a frame:");
2279 #endif
2280
2281 aom_clear_system_state();
2282
2283 cpi->source = av1_scale_if_required(cm, unscaled, &cpi->scaled_source,
2284 filter_scaler, phase_scaler, true, false);
2285 if (frame_is_intra_only(cm) || resize_pending != 0) {
2286 memset(cpi->consec_zero_mv, 0,
2287 ((cm->mi_params.mi_rows * cm->mi_params.mi_cols) >> 2) *
2288 sizeof(*cpi->consec_zero_mv));
2289 }
2290
2291 if (cpi->unscaled_last_source != NULL) {
2292 cpi->last_source = av1_scale_if_required(
2293 cm, cpi->unscaled_last_source, &cpi->scaled_last_source, filter_scaler,
2294 phase_scaler, true, false);
2295 }
2296
2297 if (cpi->sf.rt_sf.use_temporal_noise_estimate) {
2298 av1_update_noise_estimate(cpi);
2299 }
2300
2301 #if CONFIG_AV1_TEMPORAL_DENOISING
2302 if (cpi->oxcf.noise_sensitivity > 0 && cpi->use_svc)
2303 av1_denoiser_reset_on_first_frame(cpi);
2304 #endif
2305
2306 // For 1 spatial layer encoding: if the (non-LAST) reference has different
2307 // resolution from the source then disable that reference. This is to avoid
2308 // significant increase in encode time from scaling the references in
2309 // av1_scale_references. Note GOLDEN is forced to update on the (first/tigger)
2310 // resized frame and ALTREF will be refreshed ~4 frames later, so both
2311 // references become available again after few frames.
2312 if (svc->number_spatial_layers == 1) {
2313 if (cpi->ref_frame_flags & av1_ref_frame_flag_list[GOLDEN_FRAME]) {
2314 const YV12_BUFFER_CONFIG *const ref =
2315 get_ref_frame_yv12_buf(cm, GOLDEN_FRAME);
2316 if (ref->y_crop_width != cm->width || ref->y_crop_height != cm->height)
2317 cpi->ref_frame_flags ^= AOM_GOLD_FLAG;
2318 }
2319 if (cpi->ref_frame_flags & av1_ref_frame_flag_list[ALTREF_FRAME]) {
2320 const YV12_BUFFER_CONFIG *const ref =
2321 get_ref_frame_yv12_buf(cm, ALTREF_FRAME);
2322 if (ref->y_crop_width != cm->width || ref->y_crop_height != cm->height)
2323 cpi->ref_frame_flags ^= AOM_ALT_FLAG;
2324 }
2325 }
2326
2327 // For SVC the inter-layer/spatial prediction is not done for newmv
2328 // (zero_mode is forced), and since the scaled references are only
2329 // use for newmv search, we can avoid scaling here.
2330 if (!frame_is_intra_only(cm) &&
2331 !(cpi->use_svc && cpi->svc.force_zero_mode_spatial_ref))
2332 av1_scale_references(cpi, filter_scaler, phase_scaler, 1);
2333
2334 av1_set_quantizer(cm, q_cfg->qm_minlevel, q_cfg->qm_maxlevel, q,
2335 q_cfg->enable_chroma_deltaq);
2336 av1_set_speed_features_qindex_dependent(cpi, cpi->oxcf.speed);
2337 if ((q_cfg->deltaq_mode != NO_DELTA_Q) || q_cfg->enable_chroma_deltaq)
2338 av1_init_quantizer(&cpi->enc_quant_dequant_params, &cm->quant_params,
2339 cm->seq_params.bit_depth);
2340 av1_set_variance_partition_thresholds(cpi, q, 0);
2341 av1_setup_frame(cpi);
2342
2343 // Check if this high_source_sad (scene/slide change) frame should be
2344 // encoded at high/max QP, and if so, set the q and adjust some rate
2345 // control parameters.
2346 if (cpi->sf.rt_sf.overshoot_detection_cbr == FAST_DETECTION_MAXQ &&
2347 cpi->rc.high_source_sad) {
2348 if (av1_encodedframe_overshoot_cbr(cpi, &q)) {
2349 av1_set_quantizer(cm, q_cfg->qm_minlevel, q_cfg->qm_maxlevel, q,
2350 q_cfg->enable_chroma_deltaq);
2351 av1_set_speed_features_qindex_dependent(cpi, cpi->oxcf.speed);
2352 if (q_cfg->deltaq_mode != NO_DELTA_Q || q_cfg->enable_chroma_deltaq)
2353 av1_init_quantizer(&cpi->enc_quant_dequant_params, &cm->quant_params,
2354 cm->seq_params.bit_depth);
2355 av1_set_variance_partition_thresholds(cpi, q, 0);
2356 if (frame_is_intra_only(cm) || cm->features.error_resilient_mode)
2357 av1_setup_frame(cpi);
2358 }
2359 }
2360
2361 if (q_cfg->aq_mode == CYCLIC_REFRESH_AQ) {
2362 suppress_active_map(cpi);
2363 av1_cyclic_refresh_setup(cpi);
2364 av1_apply_active_map(cpi);
2365 }
2366 if (cm->seg.enabled) {
2367 if (!cm->seg.update_data && cm->prev_frame) {
2368 segfeatures_copy(&cm->seg, &cm->prev_frame->seg);
2369 cm->seg.enabled = cm->prev_frame->seg.enabled;
2370 } else {
2371 av1_calculate_segdata(&cm->seg);
2372 }
2373 } else {
2374 memset(&cm->seg, 0, sizeof(cm->seg));
2375 }
2376 segfeatures_copy(&cm->cur_frame->seg, &cm->seg);
2377 cm->cur_frame->seg.enabled = cm->seg.enabled;
2378
2379 #if CONFIG_COLLECT_COMPONENT_TIMING
2380 start_timing(cpi, av1_encode_frame_time);
2381 #endif
2382
2383 // Set the motion vector precision based on mv stats from the last coded
2384 // frame.
2385 if (!frame_is_intra_only(cm)) av1_pick_and_set_high_precision_mv(cpi, q);
2386
2387 // transform / motion compensation build reconstruction frame
2388 av1_encode_frame(cpi);
2389
2390 // Update some stats from cyclic refresh.
2391 if (q_cfg->aq_mode == CYCLIC_REFRESH_AQ && !frame_is_intra_only(cm))
2392 av1_cyclic_refresh_postencode(cpi);
2393
2394 #if CONFIG_COLLECT_COMPONENT_TIMING
2395 end_timing(cpi, av1_encode_frame_time);
2396 #endif
2397 #if CONFIG_INTERNAL_STATS
2398 ++cpi->tot_recode_hits;
2399 #endif
2400
2401 aom_clear_system_state();
2402
2403 return AOM_CODEC_OK;
2404 }
2405
2406 #if !CONFIG_REALTIME_ONLY
2407
2408 /*!\brief Recode loop for encoding one frame. the purpose of encoding one frame
2409 * for multiple times can be approaching a target bitrate or adjusting the usage
2410 * of global motions.
2411 *
2412 * \ingroup high_level_algo
2413 *
2414 * \param[in] cpi Top-level encoder structure
2415 * \param[in] size Bitstream size
2416 * \param[in] dest Bitstream output
2417 *
2418 * \return Returns a value to indicate if the encoding is done successfully.
2419 * \retval #AOM_CODEC_OK
2420 * \retval -1
2421 * \retval #AOM_CODEC_ERROR
2422 */
encode_with_recode_loop(AV1_COMP * cpi,size_t * size,uint8_t * dest)2423 static int encode_with_recode_loop(AV1_COMP *cpi, size_t *size, uint8_t *dest) {
2424 AV1_COMMON *const cm = &cpi->common;
2425 RATE_CONTROL *const rc = &cpi->rc;
2426 GlobalMotionInfo *const gm_info = &cpi->gm_info;
2427 const AV1EncoderConfig *const oxcf = &cpi->oxcf;
2428 const QuantizationCfg *const q_cfg = &oxcf->q_cfg;
2429 const int allow_recode = (cpi->sf.hl_sf.recode_loop != DISALLOW_RECODE);
2430 // Must allow recode if minimum compression ratio is set.
2431 assert(IMPLIES(oxcf->rc_cfg.min_cr > 0, allow_recode));
2432
2433 set_size_independent_vars(cpi);
2434 if (is_stat_consumption_stage_twopass(cpi) &&
2435 cpi->sf.interp_sf.adaptive_interp_filter_search)
2436 cpi->interp_search_flags.interp_filter_search_mask =
2437 av1_setup_interp_filter_search_mask(cpi);
2438 cpi->source->buf_8bit_valid = 0;
2439
2440 av1_setup_frame_size(cpi);
2441
2442 if (av1_superres_in_recode_allowed(cpi) &&
2443 cpi->superres_mode != AOM_SUPERRES_NONE &&
2444 cm->superres_scale_denominator == SCALE_NUMERATOR) {
2445 // Superres mode is currently enabled, but the denominator selected will
2446 // disable superres. So no need to continue, as we will go through another
2447 // recode loop for full-resolution after this anyway.
2448 return -1;
2449 }
2450
2451 int top_index = 0, bottom_index = 0;
2452 int q = 0, q_low = 0, q_high = 0;
2453 av1_set_size_dependent_vars(cpi, &q, &bottom_index, &top_index);
2454 q_low = bottom_index;
2455 q_high = top_index;
2456
2457 if (cpi->sf.part_sf.partition_search_type == VAR_BASED_PARTITION)
2458 variance_partition_alloc(cpi);
2459
2460 if (cm->current_frame.frame_type == KEY_FRAME) copy_frame_prob_info(cpi);
2461
2462 #if CONFIG_COLLECT_COMPONENT_TIMING
2463 printf("\n Encoding a frame:");
2464 #endif
2465
2466 // Determine whether to use screen content tools using two fast encoding.
2467 if (!cpi->sf.hl_sf.disable_extra_sc_testing)
2468 av1_determine_sc_tools_with_encoding(cpi, q);
2469
2470 #if CONFIG_USE_VMAF_RC
2471 if (oxcf->tune_cfg.tuning == AOM_TUNE_VMAF_NEG_MAX_GAIN) {
2472 av1_vmaf_neg_preprocessing(cpi, cpi->unscaled_source);
2473 }
2474 #endif
2475
2476 // Loop variables
2477 int loop = 0;
2478 int loop_count = 0;
2479 int overshoot_seen = 0;
2480 int undershoot_seen = 0;
2481 int low_cr_seen = 0;
2482 int last_loop_allow_hp = 0;
2483
2484 do {
2485 loop = 0;
2486 aom_clear_system_state();
2487
2488 // if frame was scaled calculate global_motion_search again if already
2489 // done
2490 if (loop_count > 0 && cpi->source && gm_info->search_done) {
2491 if (cpi->source->y_crop_width != cm->width ||
2492 cpi->source->y_crop_height != cm->height) {
2493 gm_info->search_done = 0;
2494 }
2495 }
2496 cpi->source =
2497 av1_scale_if_required(cm, cpi->unscaled_source, &cpi->scaled_source,
2498 EIGHTTAP_REGULAR, 0, false, false);
2499
2500 if (cpi->unscaled_last_source != NULL) {
2501 cpi->last_source = av1_scale_if_required(
2502 cm, cpi->unscaled_last_source, &cpi->scaled_last_source,
2503 EIGHTTAP_REGULAR, 0, false, false);
2504 }
2505
2506 if (!frame_is_intra_only(cm)) {
2507 if (loop_count > 0) {
2508 release_scaled_references(cpi);
2509 }
2510 av1_scale_references(cpi, EIGHTTAP_REGULAR, 0, 0);
2511 }
2512 #if CONFIG_TUNE_VMAF
2513 if (oxcf->tune_cfg.tuning >= AOM_TUNE_VMAF_WITH_PREPROCESSING &&
2514 oxcf->tune_cfg.tuning <= AOM_TUNE_VMAF_NEG_MAX_GAIN) {
2515 cpi->vmaf_info.original_qindex = q;
2516 q = av1_get_vmaf_base_qindex(cpi, q);
2517 }
2518 #endif
2519 av1_set_quantizer(cm, q_cfg->qm_minlevel, q_cfg->qm_maxlevel, q,
2520 q_cfg->enable_chroma_deltaq);
2521 av1_set_speed_features_qindex_dependent(cpi, oxcf->speed);
2522
2523 if (q_cfg->deltaq_mode != NO_DELTA_Q || q_cfg->enable_chroma_deltaq)
2524 av1_init_quantizer(&cpi->enc_quant_dequant_params, &cm->quant_params,
2525 cm->seq_params.bit_depth);
2526
2527 av1_set_variance_partition_thresholds(cpi, q, 0);
2528
2529 // printf("Frame %d/%d: q = %d, frame_type = %d superres_denom = %d\n",
2530 // cm->current_frame.frame_number, cm->show_frame, q,
2531 // cm->current_frame.frame_type, cm->superres_scale_denominator);
2532
2533 if (loop_count == 0) {
2534 av1_setup_frame(cpi);
2535 } else if (get_primary_ref_frame_buf(cm) == NULL) {
2536 // Base q-index may have changed, so we need to assign proper default coef
2537 // probs before every iteration.
2538 av1_default_coef_probs(cm);
2539 av1_setup_frame_contexts(cm);
2540 }
2541
2542 if (q_cfg->aq_mode == VARIANCE_AQ) {
2543 av1_vaq_frame_setup(cpi);
2544 } else if (q_cfg->aq_mode == COMPLEXITY_AQ) {
2545 av1_setup_in_frame_q_adj(cpi);
2546 }
2547
2548 if (cm->seg.enabled) {
2549 if (!cm->seg.update_data && cm->prev_frame) {
2550 segfeatures_copy(&cm->seg, &cm->prev_frame->seg);
2551 cm->seg.enabled = cm->prev_frame->seg.enabled;
2552 } else {
2553 av1_calculate_segdata(&cm->seg);
2554 }
2555 } else {
2556 memset(&cm->seg, 0, sizeof(cm->seg));
2557 }
2558 segfeatures_copy(&cm->cur_frame->seg, &cm->seg);
2559 cm->cur_frame->seg.enabled = cm->seg.enabled;
2560
2561 #if CONFIG_COLLECT_COMPONENT_TIMING
2562 start_timing(cpi, av1_encode_frame_time);
2563 #endif
2564 // Set the motion vector precision based on mv stats from the last coded
2565 // frame.
2566 if (!frame_is_intra_only(cm)) {
2567 av1_pick_and_set_high_precision_mv(cpi, q);
2568
2569 // If the precision has changed during different iteration of the loop,
2570 // then we need to reset the global motion vectors
2571 if (loop_count > 0 &&
2572 cm->features.allow_high_precision_mv != last_loop_allow_hp) {
2573 gm_info->search_done = 0;
2574 }
2575 last_loop_allow_hp = cm->features.allow_high_precision_mv;
2576 }
2577
2578 // transform / motion compensation build reconstruction frame
2579 av1_encode_frame(cpi);
2580
2581 // Reset the mv_stats in case we are interrupted by an intraframe or an
2582 // overlay frame.
2583 if (cpi->mv_stats.valid) {
2584 av1_zero(cpi->mv_stats);
2585 }
2586 // Gather the mv_stats for the next frame
2587 if (cpi->sf.hl_sf.high_precision_mv_usage == LAST_MV_DATA &&
2588 av1_frame_allows_smart_mv(cpi)) {
2589 av1_collect_mv_stats(cpi, q);
2590 }
2591
2592 #if CONFIG_COLLECT_COMPONENT_TIMING
2593 end_timing(cpi, av1_encode_frame_time);
2594 #endif
2595
2596 aom_clear_system_state();
2597
2598 // Dummy pack of the bitstream using up to date stats to get an
2599 // accurate estimate of output frame size to determine if we need
2600 // to recode.
2601 const int do_dummy_pack =
2602 (cpi->sf.hl_sf.recode_loop >= ALLOW_RECODE_KFARFGF &&
2603 oxcf->rc_cfg.mode != AOM_Q) ||
2604 oxcf->rc_cfg.min_cr > 0;
2605 if (do_dummy_pack) {
2606 av1_finalize_encoded_frame(cpi);
2607 int largest_tile_id = 0; // Output from bitstream: unused here
2608 rc->coefficient_size = 0;
2609 if (av1_pack_bitstream(cpi, dest, size, &largest_tile_id) !=
2610 AOM_CODEC_OK) {
2611 return AOM_CODEC_ERROR;
2612 }
2613
2614 rc->projected_frame_size = (int)(*size) << 3;
2615 }
2616
2617 #if CONFIG_TUNE_VMAF
2618 if (oxcf->tune_cfg.tuning >= AOM_TUNE_VMAF_WITH_PREPROCESSING &&
2619 oxcf->tune_cfg.tuning <= AOM_TUNE_VMAF_NEG_MAX_GAIN) {
2620 q = cpi->vmaf_info.original_qindex;
2621 }
2622 #endif
2623 if (allow_recode) {
2624 // Update q and decide whether to do a recode loop
2625 recode_loop_update_q(cpi, &loop, &q, &q_low, &q_high, top_index,
2626 bottom_index, &undershoot_seen, &overshoot_seen,
2627 &low_cr_seen, loop_count);
2628 }
2629
2630 if (loop) {
2631 ++loop_count;
2632
2633 #if CONFIG_INTERNAL_STATS
2634 ++cpi->tot_recode_hits;
2635 #endif
2636 }
2637 #if CONFIG_COLLECT_COMPONENT_TIMING
2638 if (loop) printf("\n Recoding:");
2639 #endif
2640 } while (loop);
2641
2642 return AOM_CODEC_OK;
2643 }
2644 #endif // !CONFIG_REALTIME_ONLY
2645
2646 // TODO(jingning, paulwilkins): Set up high grain level to test
2647 // hardware decoders. Need to adapt the actual noise variance
2648 // according to the difference between reconstructed frame and the
2649 // source signal.
set_grain_syn_params(AV1_COMMON * cm)2650 static void set_grain_syn_params(AV1_COMMON *cm) {
2651 aom_film_grain_t *film_grain_params = &cm->film_grain_params;
2652 film_grain_params->apply_grain = 1;
2653 film_grain_params->update_parameters = 1;
2654 film_grain_params->random_seed = rand() & 0xffff;
2655
2656 film_grain_params->num_y_points = 1;
2657 film_grain_params->scaling_points_y[0][0] = 128;
2658 film_grain_params->scaling_points_y[0][1] = 100;
2659
2660 film_grain_params->num_cb_points = 1;
2661 film_grain_params->scaling_points_cb[0][0] = 128;
2662 film_grain_params->scaling_points_cb[0][1] = 100;
2663
2664 film_grain_params->num_cr_points = 1;
2665 film_grain_params->scaling_points_cr[0][0] = 128;
2666 film_grain_params->scaling_points_cr[0][1] = 100;
2667
2668 film_grain_params->chroma_scaling_from_luma = 0;
2669 film_grain_params->scaling_shift = 1;
2670 film_grain_params->ar_coeff_lag = 0;
2671 film_grain_params->ar_coeff_shift = 1;
2672 film_grain_params->overlap_flag = 1;
2673 film_grain_params->grain_scale_shift = 0;
2674 }
2675
2676 /*!\brief Recode loop or a single loop for encoding one frame, followed by
2677 * in-loop deblocking filters, CDEF filters, and restoration filters.
2678 *
2679 * \ingroup high_level_algo
2680 * \callgraph
2681 * \callergraph
2682 *
2683 * \param[in] cpi Top-level encoder structure
2684 * \param[in] size Bitstream size
2685 * \param[in] dest Bitstream output
2686 * \param[in] sse Total distortion of the frame
2687 * \param[in] rate Total rate of the frame
2688 * \param[in] largest_tile_id Tile id of the last tile
2689 *
2690 * \return Returns a value to indicate if the encoding is done successfully.
2691 * \retval #AOM_CODEC_OK
2692 * \retval #AOM_CODEC_ERROR
2693 */
encode_with_recode_loop_and_filter(AV1_COMP * cpi,size_t * size,uint8_t * dest,int64_t * sse,int64_t * rate,int * largest_tile_id)2694 static int encode_with_recode_loop_and_filter(AV1_COMP *cpi, size_t *size,
2695 uint8_t *dest, int64_t *sse,
2696 int64_t *rate,
2697 int *largest_tile_id) {
2698 #if CONFIG_COLLECT_COMPONENT_TIMING
2699 start_timing(cpi, encode_with_recode_loop_time);
2700 #endif
2701 int err;
2702 #if CONFIG_REALTIME_ONLY
2703 err = encode_without_recode(cpi);
2704 #else
2705 if (cpi->sf.hl_sf.recode_loop == DISALLOW_RECODE)
2706 err = encode_without_recode(cpi);
2707 else
2708 err = encode_with_recode_loop(cpi, size, dest);
2709 #endif
2710 #if CONFIG_COLLECT_COMPONENT_TIMING
2711 end_timing(cpi, encode_with_recode_loop_time);
2712 #endif
2713 if (err != AOM_CODEC_OK) {
2714 if (err == -1) {
2715 // special case as described in encode_with_recode_loop().
2716 // Encoding was skipped.
2717 err = AOM_CODEC_OK;
2718 if (sse != NULL) *sse = INT64_MAX;
2719 if (rate != NULL) *rate = INT64_MAX;
2720 *largest_tile_id = 0;
2721 }
2722 return err;
2723 }
2724
2725 #ifdef OUTPUT_YUV_DENOISED
2726 const AV1EncoderConfig *const oxcf = &cpi->oxcf;
2727 if (oxcf->noise_sensitivity > 0 && denoise_svc(cpi)) {
2728 aom_write_yuv_frame(yuv_denoised_file,
2729 &cpi->denoiser.running_avg_y[INTRA_FRAME]);
2730 }
2731 #endif
2732
2733 AV1_COMMON *const cm = &cpi->common;
2734 SequenceHeader *const seq_params = &cm->seq_params;
2735
2736 // Special case code to reduce pulsing when key frames are forced at a
2737 // fixed interval. Note the reconstruction error if it is the frame before
2738 // the force key frame
2739 if (cpi->rc.next_key_frame_forced && cpi->rc.frames_to_key == 1) {
2740 #if CONFIG_AV1_HIGHBITDEPTH
2741 if (seq_params->use_highbitdepth) {
2742 cpi->ambient_err = aom_highbd_get_y_sse(cpi->source, &cm->cur_frame->buf);
2743 } else {
2744 cpi->ambient_err = aom_get_y_sse(cpi->source, &cm->cur_frame->buf);
2745 }
2746 #else
2747 cpi->ambient_err = aom_get_y_sse(cpi->source, &cm->cur_frame->buf);
2748 #endif
2749 }
2750
2751 cm->cur_frame->buf.color_primaries = seq_params->color_primaries;
2752 cm->cur_frame->buf.transfer_characteristics =
2753 seq_params->transfer_characteristics;
2754 cm->cur_frame->buf.matrix_coefficients = seq_params->matrix_coefficients;
2755 cm->cur_frame->buf.monochrome = seq_params->monochrome;
2756 cm->cur_frame->buf.chroma_sample_position =
2757 seq_params->chroma_sample_position;
2758 cm->cur_frame->buf.color_range = seq_params->color_range;
2759 cm->cur_frame->buf.render_width = cm->render_width;
2760 cm->cur_frame->buf.render_height = cm->render_height;
2761
2762 // Pick the loop filter level for the frame.
2763 if (!cm->features.allow_intrabc) {
2764 loopfilter_frame(cpi, cm);
2765 } else {
2766 cm->lf.filter_level[0] = 0;
2767 cm->lf.filter_level[1] = 0;
2768 cm->cdef_info.cdef_bits = 0;
2769 cm->cdef_info.cdef_strengths[0] = 0;
2770 cm->cdef_info.nb_cdef_strengths = 1;
2771 cm->cdef_info.cdef_uv_strengths[0] = 0;
2772 cm->rst_info[0].frame_restoration_type = RESTORE_NONE;
2773 cm->rst_info[1].frame_restoration_type = RESTORE_NONE;
2774 cm->rst_info[2].frame_restoration_type = RESTORE_NONE;
2775 }
2776
2777 // TODO(debargha): Fix mv search range on encoder side
2778 // aom_extend_frame_inner_borders(&cm->cur_frame->buf, av1_num_planes(cm));
2779 aom_extend_frame_borders(&cm->cur_frame->buf, av1_num_planes(cm));
2780
2781 #ifdef OUTPUT_YUV_REC
2782 aom_write_one_yuv_frame(cm, &cm->cur_frame->buf);
2783 #endif
2784
2785 if (cpi->oxcf.tune_cfg.content == AOM_CONTENT_FILM) {
2786 set_grain_syn_params(cm);
2787 }
2788
2789 av1_finalize_encoded_frame(cpi);
2790 // Build the bitstream
2791 #if CONFIG_COLLECT_COMPONENT_TIMING
2792 start_timing(cpi, av1_pack_bitstream_final_time);
2793 #endif
2794 cpi->rc.coefficient_size = 0;
2795 if (av1_pack_bitstream(cpi, dest, size, largest_tile_id) != AOM_CODEC_OK)
2796 return AOM_CODEC_ERROR;
2797 #if CONFIG_COLLECT_COMPONENT_TIMING
2798 end_timing(cpi, av1_pack_bitstream_final_time);
2799 #endif
2800
2801 // Compute sse and rate.
2802 if (sse != NULL) {
2803 #if CONFIG_AV1_HIGHBITDEPTH
2804 *sse = (seq_params->use_highbitdepth)
2805 ? aom_highbd_get_y_sse(cpi->source, &cm->cur_frame->buf)
2806 : aom_get_y_sse(cpi->source, &cm->cur_frame->buf);
2807 #else
2808 *sse = aom_get_y_sse(cpi->source, &cm->cur_frame->buf);
2809 #endif
2810 }
2811 if (rate != NULL) {
2812 const int64_t bits = (*size << 3);
2813 *rate = (bits << 5); // To match scale.
2814 }
2815 return AOM_CODEC_OK;
2816 }
2817
encode_with_and_without_superres(AV1_COMP * cpi,size_t * size,uint8_t * dest,int * largest_tile_id)2818 static int encode_with_and_without_superres(AV1_COMP *cpi, size_t *size,
2819 uint8_t *dest,
2820 int *largest_tile_id) {
2821 const AV1_COMMON *const cm = &cpi->common;
2822 assert(cm->seq_params.enable_superres);
2823 assert(av1_superres_in_recode_allowed(cpi));
2824 aom_codec_err_t err = AOM_CODEC_OK;
2825 av1_save_all_coding_context(cpi);
2826
2827 int64_t sse1 = INT64_MAX;
2828 int64_t rate1 = INT64_MAX;
2829 int largest_tile_id1 = 0;
2830 int64_t sse2 = INT64_MAX;
2831 int64_t rate2 = INT64_MAX;
2832 int largest_tile_id2;
2833 double proj_rdcost1 = DBL_MAX;
2834
2835 // Encode with superres.
2836 if (cpi->sf.hl_sf.superres_auto_search_type == SUPERRES_AUTO_ALL) {
2837 SuperResCfg *const superres_cfg = &cpi->oxcf.superres_cfg;
2838 int64_t superres_sses[SCALE_NUMERATOR];
2839 int64_t superres_rates[SCALE_NUMERATOR];
2840 int superres_largest_tile_ids[SCALE_NUMERATOR];
2841 // Use superres for Key-frames and Alt-ref frames only.
2842 const GF_GROUP *const gf_group = &cpi->gf_group;
2843 if (gf_group->update_type[gf_group->index] != OVERLAY_UPDATE &&
2844 gf_group->update_type[gf_group->index] != INTNL_OVERLAY_UPDATE) {
2845 for (int denom = SCALE_NUMERATOR + 1; denom <= 2 * SCALE_NUMERATOR;
2846 ++denom) {
2847 superres_cfg->superres_scale_denominator = denom;
2848 superres_cfg->superres_kf_scale_denominator = denom;
2849 const int this_index = denom - (SCALE_NUMERATOR + 1);
2850
2851 cpi->superres_mode = AOM_SUPERRES_AUTO; // Super-res on for this loop.
2852 err = encode_with_recode_loop_and_filter(
2853 cpi, size, dest, &superres_sses[this_index],
2854 &superres_rates[this_index],
2855 &superres_largest_tile_ids[this_index]);
2856 cpi->superres_mode = AOM_SUPERRES_NONE; // Reset to default (full-res).
2857 if (err != AOM_CODEC_OK) return err;
2858 restore_all_coding_context(cpi);
2859 }
2860 // Reset.
2861 superres_cfg->superres_scale_denominator = SCALE_NUMERATOR;
2862 superres_cfg->superres_kf_scale_denominator = SCALE_NUMERATOR;
2863 } else {
2864 for (int denom = SCALE_NUMERATOR + 1; denom <= 2 * SCALE_NUMERATOR;
2865 ++denom) {
2866 const int this_index = denom - (SCALE_NUMERATOR + 1);
2867 superres_sses[this_index] = INT64_MAX;
2868 superres_rates[this_index] = INT64_MAX;
2869 }
2870 }
2871 // Encode without superres.
2872 assert(cpi->superres_mode == AOM_SUPERRES_NONE);
2873 err = encode_with_recode_loop_and_filter(cpi, size, dest, &sse2, &rate2,
2874 &largest_tile_id2);
2875 if (err != AOM_CODEC_OK) return err;
2876
2877 // Note: Both use common rdmult based on base qindex of fullres.
2878 const int64_t rdmult =
2879 av1_compute_rd_mult_based_on_qindex(cpi, cm->quant_params.base_qindex);
2880
2881 // Find the best rdcost among all superres denoms.
2882 int best_denom = -1;
2883 for (int denom = SCALE_NUMERATOR + 1; denom <= 2 * SCALE_NUMERATOR;
2884 ++denom) {
2885 const int this_index = denom - (SCALE_NUMERATOR + 1);
2886 const int64_t this_sse = superres_sses[this_index];
2887 const int64_t this_rate = superres_rates[this_index];
2888 const int this_largest_tile_id = superres_largest_tile_ids[this_index];
2889 const double this_rdcost = RDCOST_DBL_WITH_NATIVE_BD_DIST(
2890 rdmult, this_rate, this_sse, cm->seq_params.bit_depth);
2891 if (this_rdcost < proj_rdcost1) {
2892 sse1 = this_sse;
2893 rate1 = this_rate;
2894 largest_tile_id1 = this_largest_tile_id;
2895 proj_rdcost1 = this_rdcost;
2896 best_denom = denom;
2897 }
2898 }
2899 const double proj_rdcost2 = RDCOST_DBL_WITH_NATIVE_BD_DIST(
2900 rdmult, rate2, sse2, cm->seq_params.bit_depth);
2901 // Re-encode with superres if it's better.
2902 if (proj_rdcost1 < proj_rdcost2) {
2903 restore_all_coding_context(cpi);
2904 // TODO(urvang): We should avoid rerunning the recode loop by saving
2905 // previous output+state, or running encode only for the selected 'q' in
2906 // previous step.
2907 // Again, temporarily force the best denom.
2908 superres_cfg->superres_scale_denominator = best_denom;
2909 superres_cfg->superres_kf_scale_denominator = best_denom;
2910 int64_t sse3 = INT64_MAX;
2911 int64_t rate3 = INT64_MAX;
2912 cpi->superres_mode =
2913 AOM_SUPERRES_AUTO; // Super-res on for this recode loop.
2914 err = encode_with_recode_loop_and_filter(cpi, size, dest, &sse3, &rate3,
2915 largest_tile_id);
2916 cpi->superres_mode = AOM_SUPERRES_NONE; // Reset to default (full-res).
2917 assert(sse1 == sse3);
2918 assert(rate1 == rate3);
2919 assert(largest_tile_id1 == *largest_tile_id);
2920 // Reset.
2921 superres_cfg->superres_scale_denominator = SCALE_NUMERATOR;
2922 superres_cfg->superres_kf_scale_denominator = SCALE_NUMERATOR;
2923 } else {
2924 *largest_tile_id = largest_tile_id2;
2925 }
2926 } else {
2927 assert(cpi->sf.hl_sf.superres_auto_search_type == SUPERRES_AUTO_DUAL);
2928 cpi->superres_mode =
2929 AOM_SUPERRES_AUTO; // Super-res on for this recode loop.
2930 err = encode_with_recode_loop_and_filter(cpi, size, dest, &sse1, &rate1,
2931 &largest_tile_id1);
2932 cpi->superres_mode = AOM_SUPERRES_NONE; // Reset to default (full-res).
2933 if (err != AOM_CODEC_OK) return err;
2934 restore_all_coding_context(cpi);
2935 // Encode without superres.
2936 assert(cpi->superres_mode == AOM_SUPERRES_NONE);
2937 err = encode_with_recode_loop_and_filter(cpi, size, dest, &sse2, &rate2,
2938 &largest_tile_id2);
2939 if (err != AOM_CODEC_OK) return err;
2940
2941 // Note: Both use common rdmult based on base qindex of fullres.
2942 const int64_t rdmult =
2943 av1_compute_rd_mult_based_on_qindex(cpi, cm->quant_params.base_qindex);
2944 proj_rdcost1 = RDCOST_DBL_WITH_NATIVE_BD_DIST(rdmult, rate1, sse1,
2945 cm->seq_params.bit_depth);
2946 const double proj_rdcost2 = RDCOST_DBL_WITH_NATIVE_BD_DIST(
2947 rdmult, rate2, sse2, cm->seq_params.bit_depth);
2948 // Re-encode with superres if it's better.
2949 if (proj_rdcost1 < proj_rdcost2) {
2950 restore_all_coding_context(cpi);
2951 // TODO(urvang): We should avoid rerunning the recode loop by saving
2952 // previous output+state, or running encode only for the selected 'q' in
2953 // previous step.
2954 int64_t sse3 = INT64_MAX;
2955 int64_t rate3 = INT64_MAX;
2956 cpi->superres_mode =
2957 AOM_SUPERRES_AUTO; // Super-res on for this recode loop.
2958 err = encode_with_recode_loop_and_filter(cpi, size, dest, &sse3, &rate3,
2959 largest_tile_id);
2960 cpi->superres_mode = AOM_SUPERRES_NONE; // Reset to default (full-res).
2961 assert(sse1 == sse3);
2962 assert(rate1 == rate3);
2963 assert(largest_tile_id1 == *largest_tile_id);
2964 } else {
2965 *largest_tile_id = largest_tile_id2;
2966 }
2967 }
2968
2969 return err;
2970 }
2971
2972 extern void av1_print_frame_contexts(const FRAME_CONTEXT *fc,
2973 const char *filename);
2974
2975 /*!\brief Run the final pass encoding for 1-pass/2-pass encoding mode, and pack
2976 * the bitstream
2977 *
2978 * \ingroup high_level_algo
2979 * \callgraph
2980 * \callergraph
2981 *
2982 * \param[in] cpi Top-level encoder structure
2983 * \param[in] size Bitstream size
2984 * \param[in] dest Bitstream output
2985 *
2986 * \return Returns a value to indicate if the encoding is done successfully.
2987 * \retval #AOM_CODEC_OK
2988 * \retval #AOM_CODEC_ERROR
2989 */
encode_frame_to_data_rate(AV1_COMP * cpi,size_t * size,uint8_t * dest)2990 static int encode_frame_to_data_rate(AV1_COMP *cpi, size_t *size,
2991 uint8_t *dest) {
2992 AV1_COMMON *const cm = &cpi->common;
2993 SequenceHeader *const seq_params = &cm->seq_params;
2994 CurrentFrame *const current_frame = &cm->current_frame;
2995 const AV1EncoderConfig *const oxcf = &cpi->oxcf;
2996 struct segmentation *const seg = &cm->seg;
2997 FeatureFlags *const features = &cm->features;
2998 const TileConfig *const tile_cfg = &oxcf->tile_cfg;
2999
3000 #if CONFIG_COLLECT_COMPONENT_TIMING
3001 start_timing(cpi, encode_frame_to_data_rate_time);
3002 #endif
3003
3004 if (frame_is_intra_only(cm)) {
3005 av1_set_screen_content_options(cpi, features);
3006 }
3007
3008 // frame type has been decided outside of this function call
3009 cm->cur_frame->frame_type = current_frame->frame_type;
3010
3011 cm->tiles.large_scale = tile_cfg->enable_large_scale_tile;
3012 cm->tiles.single_tile_decoding = tile_cfg->enable_single_tile_decoding;
3013
3014 features->allow_ref_frame_mvs &= frame_might_allow_ref_frame_mvs(cm);
3015 // features->allow_ref_frame_mvs needs to be written into the frame header
3016 // while cm->tiles.large_scale is 1, therefore, "cm->tiles.large_scale=1" case
3017 // is separated from frame_might_allow_ref_frame_mvs().
3018 features->allow_ref_frame_mvs &= !cm->tiles.large_scale;
3019
3020 features->allow_warped_motion = oxcf->motion_mode_cfg.allow_warped_motion &&
3021 frame_might_allow_warped_motion(cm);
3022
3023 cpi->last_frame_type = current_frame->frame_type;
3024
3025 if (frame_is_sframe(cm)) {
3026 GF_GROUP *gf_group = &cpi->gf_group;
3027 // S frame will wipe out any previously encoded altref so we cannot place
3028 // an overlay frame
3029 gf_group->update_type[gf_group->size] = GF_UPDATE;
3030 }
3031
3032 if (encode_show_existing_frame(cm)) {
3033 av1_finalize_encoded_frame(cpi);
3034 // Build the bitstream
3035 int largest_tile_id = 0; // Output from bitstream: unused here
3036 cpi->rc.coefficient_size = 0;
3037 if (av1_pack_bitstream(cpi, dest, size, &largest_tile_id) != AOM_CODEC_OK)
3038 return AOM_CODEC_ERROR;
3039
3040 if (seq_params->frame_id_numbers_present_flag &&
3041 current_frame->frame_type == KEY_FRAME) {
3042 // Displaying a forward key-frame, so reset the ref buffer IDs
3043 int display_frame_id = cm->ref_frame_id[cpi->existing_fb_idx_to_show];
3044 for (int i = 0; i < REF_FRAMES; i++)
3045 cm->ref_frame_id[i] = display_frame_id;
3046 }
3047
3048 cpi->seq_params_locked = 1;
3049
3050 #if DUMP_RECON_FRAMES == 1
3051 // NOTE(zoeliu): For debug - Output the filtered reconstructed video.
3052 av1_dump_filtered_recon_frames(cpi);
3053 #endif // DUMP_RECON_FRAMES
3054
3055 // NOTE: Save the new show frame buffer index for --test-code=warn, i.e.,
3056 // for the purpose to verify no mismatch between encoder and decoder.
3057 if (cm->show_frame) cpi->last_show_frame_buf = cm->cur_frame;
3058
3059 #if CONFIG_AV1_TEMPORAL_DENOISING
3060 av1_denoiser_update_ref_frame(cpi);
3061 #endif
3062
3063 refresh_reference_frames(cpi);
3064
3065 // Since we allocate a spot for the OVERLAY frame in the gf group, we need
3066 // to do post-encoding update accordingly.
3067 av1_set_target_rate(cpi, cm->width, cm->height);
3068 av1_rc_postencode_update(cpi, *size);
3069
3070 if (is_psnr_calc_enabled(cpi)) {
3071 cpi->source =
3072 realloc_and_scale_source(cpi, cm->cur_frame->buf.y_crop_width,
3073 cm->cur_frame->buf.y_crop_height);
3074 }
3075
3076 ++current_frame->frame_number;
3077 update_frame_index_set(&cpi->frame_index_set, cm->show_frame);
3078 return AOM_CODEC_OK;
3079 }
3080
3081 // Work out whether to force_integer_mv this frame
3082 if (!is_stat_generation_stage(cpi) &&
3083 cpi->common.features.allow_screen_content_tools &&
3084 !frame_is_intra_only(cm)) {
3085 if (cpi->common.seq_params.force_integer_mv == 2) {
3086 // Adaptive mode: see what previous frame encoded did
3087 if (cpi->unscaled_last_source != NULL) {
3088 features->cur_frame_force_integer_mv = av1_is_integer_mv(
3089 cpi->source, cpi->unscaled_last_source, &cpi->force_intpel_info);
3090 } else {
3091 cpi->common.features.cur_frame_force_integer_mv = 0;
3092 }
3093 } else {
3094 cpi->common.features.cur_frame_force_integer_mv =
3095 cpi->common.seq_params.force_integer_mv;
3096 }
3097 } else {
3098 cpi->common.features.cur_frame_force_integer_mv = 0;
3099 }
3100
3101 // Set default state for segment based loop filter update flags.
3102 cm->lf.mode_ref_delta_update = 0;
3103
3104 // Set various flags etc to special state if it is a key frame.
3105 if (frame_is_intra_only(cm) || frame_is_sframe(cm)) {
3106 // Reset the loop filter deltas and segmentation map.
3107 av1_reset_segment_features(cm);
3108
3109 // If segmentation is enabled force a map update for key frames.
3110 if (seg->enabled) {
3111 seg->update_map = 1;
3112 seg->update_data = 1;
3113 }
3114 }
3115 if (tile_cfg->mtu == 0) {
3116 cpi->num_tg = tile_cfg->num_tile_groups;
3117 } else {
3118 // Use a default value for the purposes of weighting costs in probability
3119 // updates
3120 cpi->num_tg = DEFAULT_MAX_NUM_TG;
3121 }
3122
3123 // For 1 pass CBR, check if we are dropping this frame.
3124 // Never drop on key frame.
3125 if (has_no_stats_stage(cpi) && oxcf->rc_cfg.mode == AOM_CBR &&
3126 current_frame->frame_type != KEY_FRAME) {
3127 if (av1_rc_drop_frame(cpi)) {
3128 av1_setup_frame_size(cpi);
3129 av1_rc_postencode_update_drop_frame(cpi);
3130 release_scaled_references(cpi);
3131 return AOM_CODEC_OK;
3132 }
3133 }
3134
3135 if (oxcf->tune_cfg.tuning == AOM_TUNE_SSIM)
3136 av1_set_mb_ssim_rdmult_scaling(cpi);
3137
3138 #if CONFIG_TUNE_VMAF
3139 if (oxcf->tune_cfg.tuning == AOM_TUNE_VMAF_WITHOUT_PREPROCESSING ||
3140 oxcf->tune_cfg.tuning == AOM_TUNE_VMAF_MAX_GAIN ||
3141 oxcf->tune_cfg.tuning == AOM_TUNE_VMAF_NEG_MAX_GAIN) {
3142 av1_set_mb_vmaf_rdmult_scaling(cpi);
3143 }
3144 #endif
3145
3146 aom_clear_system_state();
3147
3148 #if CONFIG_INTERNAL_STATS
3149 memset(cpi->mode_chosen_counts, 0,
3150 MAX_MODES * sizeof(*cpi->mode_chosen_counts));
3151 #endif
3152
3153 if (seq_params->frame_id_numbers_present_flag) {
3154 /* Non-normative definition of current_frame_id ("frame counter" with
3155 * wraparound) */
3156 if (cm->current_frame_id == -1) {
3157 int lsb, msb;
3158 /* quasi-random initialization of current_frame_id for a key frame */
3159 if (cpi->source->flags & YV12_FLAG_HIGHBITDEPTH) {
3160 lsb = CONVERT_TO_SHORTPTR(cpi->source->y_buffer)[0] & 0xff;
3161 msb = CONVERT_TO_SHORTPTR(cpi->source->y_buffer)[1] & 0xff;
3162 } else {
3163 lsb = cpi->source->y_buffer[0] & 0xff;
3164 msb = cpi->source->y_buffer[1] & 0xff;
3165 }
3166 cm->current_frame_id =
3167 ((msb << 8) + lsb) % (1 << seq_params->frame_id_length);
3168
3169 // S_frame is meant for stitching different streams of different
3170 // resolutions together, so current_frame_id must be the
3171 // same across different streams of the same content current_frame_id
3172 // should be the same and not random. 0x37 is a chosen number as start
3173 // point
3174 if (oxcf->kf_cfg.sframe_dist != 0) cm->current_frame_id = 0x37;
3175 } else {
3176 cm->current_frame_id =
3177 (cm->current_frame_id + 1 + (1 << seq_params->frame_id_length)) %
3178 (1 << seq_params->frame_id_length);
3179 }
3180 }
3181
3182 switch (oxcf->algo_cfg.cdf_update_mode) {
3183 case 0: // No CDF update for any frames(4~6% compression loss).
3184 features->disable_cdf_update = 1;
3185 break;
3186 case 1: // Enable CDF update for all frames.
3187 features->disable_cdf_update = 0;
3188 break;
3189 case 2:
3190 // Strategically determine at which frames to do CDF update.
3191 // Currently only enable CDF update for all-intra and no-show frames(1.5%
3192 // compression loss).
3193 // TODO(huisu@google.com): design schemes for various trade-offs between
3194 // compression quality and decoding speed.
3195 if (oxcf->mode == GOOD) {
3196 features->disable_cdf_update =
3197 (frame_is_intra_only(cm) || !cm->show_frame) ? 0 : 1;
3198 } else {
3199 if (cpi->svc.number_spatial_layers == 1 &&
3200 cpi->svc.number_temporal_layers == 1)
3201 features->disable_cdf_update = cm->current_frame.frame_number & 1;
3202 else if (cpi->svc.number_temporal_layers > 1)
3203 // Disable only on top temporal enhancement layer for now.
3204 features->disable_cdf_update = (cpi->svc.temporal_layer_id ==
3205 cpi->svc.number_temporal_layers - 1);
3206 }
3207 break;
3208 }
3209 seq_params->timing_info_present &= !seq_params->reduced_still_picture_hdr;
3210
3211 int largest_tile_id = 0;
3212 if (av1_superres_in_recode_allowed(cpi)) {
3213 if (encode_with_and_without_superres(cpi, size, dest, &largest_tile_id) !=
3214 AOM_CODEC_OK) {
3215 return AOM_CODEC_ERROR;
3216 }
3217 } else {
3218 const aom_superres_mode orig_superres_mode = cpi->superres_mode; // save
3219 cpi->superres_mode = cpi->oxcf.superres_cfg.superres_mode;
3220 if (encode_with_recode_loop_and_filter(cpi, size, dest, NULL, NULL,
3221 &largest_tile_id) != AOM_CODEC_OK) {
3222 return AOM_CODEC_ERROR;
3223 }
3224 cpi->superres_mode = orig_superres_mode; // restore
3225 }
3226
3227 cpi->seq_params_locked = 1;
3228
3229 // Update reference frame ids for reference frames this frame will overwrite
3230 if (seq_params->frame_id_numbers_present_flag) {
3231 for (int i = 0; i < REF_FRAMES; i++) {
3232 if ((current_frame->refresh_frame_flags >> i) & 1) {
3233 cm->ref_frame_id[i] = cm->current_frame_id;
3234 }
3235 }
3236 }
3237
3238 if (cpi->svc.spatial_layer_id == cpi->svc.number_spatial_layers - 1)
3239 cpi->svc.num_encoded_top_layer++;
3240
3241 #if DUMP_RECON_FRAMES == 1
3242 // NOTE(zoeliu): For debug - Output the filtered reconstructed video.
3243 av1_dump_filtered_recon_frames(cpi);
3244 #endif // DUMP_RECON_FRAMES
3245
3246 if (cm->seg.enabled) {
3247 if (cm->seg.update_map) {
3248 update_reference_segmentation_map(cpi);
3249 } else if (cm->last_frame_seg_map) {
3250 memcpy(cm->cur_frame->seg_map, cm->last_frame_seg_map,
3251 cm->cur_frame->mi_cols * cm->cur_frame->mi_rows *
3252 sizeof(*cm->cur_frame->seg_map));
3253 }
3254 }
3255
3256 if (frame_is_intra_only(cm) == 0) {
3257 release_scaled_references(cpi);
3258 }
3259 #if CONFIG_AV1_TEMPORAL_DENOISING
3260 av1_denoiser_update_ref_frame(cpi);
3261 #endif
3262
3263 // NOTE: Save the new show frame buffer index for --test-code=warn, i.e.,
3264 // for the purpose to verify no mismatch between encoder and decoder.
3265 if (cm->show_frame) cpi->last_show_frame_buf = cm->cur_frame;
3266
3267 refresh_reference_frames(cpi);
3268
3269 #if CONFIG_ENTROPY_STATS
3270 av1_accumulate_frame_counts(&aggregate_fc, &cpi->counts);
3271 #endif // CONFIG_ENTROPY_STATS
3272
3273 if (features->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) {
3274 *cm->fc = cpi->tile_data[largest_tile_id].tctx;
3275 av1_reset_cdf_symbol_counters(cm->fc);
3276 }
3277 if (!cm->tiles.large_scale) {
3278 cm->cur_frame->frame_context = *cm->fc;
3279 }
3280
3281 if (tile_cfg->enable_ext_tile_debug) {
3282 // (yunqing) This test ensures the correctness of large scale tile coding.
3283 if (cm->tiles.large_scale && is_stat_consumption_stage(cpi)) {
3284 char fn[20] = "./fc";
3285 fn[4] = current_frame->frame_number / 100 + '0';
3286 fn[5] = (current_frame->frame_number % 100) / 10 + '0';
3287 fn[6] = (current_frame->frame_number % 10) + '0';
3288 fn[7] = '\0';
3289 av1_print_frame_contexts(cm->fc, fn);
3290 }
3291 }
3292
3293 cpi->last_frame_type = current_frame->frame_type;
3294
3295 av1_rc_postencode_update(cpi, *size);
3296
3297 // Clear the one shot update flags for segmentation map and mode/ref loop
3298 // filter deltas.
3299 cm->seg.update_map = 0;
3300 cm->seg.update_data = 0;
3301 cm->lf.mode_ref_delta_update = 0;
3302
3303 // A droppable frame might not be shown but it always
3304 // takes a space in the gf group. Therefore, even when
3305 // it is not shown, we still need update the count down.
3306 if (cm->show_frame) {
3307 update_frame_index_set(&cpi->frame_index_set, cm->show_frame);
3308 ++current_frame->frame_number;
3309 }
3310
3311 #if CONFIG_COLLECT_COMPONENT_TIMING
3312 end_timing(cpi, encode_frame_to_data_rate_time);
3313 #endif
3314
3315 return AOM_CODEC_OK;
3316 }
3317
av1_encode(AV1_COMP * const cpi,uint8_t * const dest,const EncodeFrameInput * const frame_input,const EncodeFrameParams * const frame_params,EncodeFrameResults * const frame_results)3318 int av1_encode(AV1_COMP *const cpi, uint8_t *const dest,
3319 const EncodeFrameInput *const frame_input,
3320 const EncodeFrameParams *const frame_params,
3321 EncodeFrameResults *const frame_results) {
3322 AV1_COMMON *const cm = &cpi->common;
3323 CurrentFrame *const current_frame = &cm->current_frame;
3324
3325 cpi->unscaled_source = frame_input->source;
3326 cpi->source = frame_input->source;
3327 cpi->unscaled_last_source = frame_input->last_source;
3328
3329 current_frame->refresh_frame_flags = frame_params->refresh_frame_flags;
3330 cm->features.error_resilient_mode = frame_params->error_resilient_mode;
3331 cm->features.primary_ref_frame = frame_params->primary_ref_frame;
3332 cm->current_frame.frame_type = frame_params->frame_type;
3333 cm->show_frame = frame_params->show_frame;
3334 cpi->ref_frame_flags = frame_params->ref_frame_flags;
3335 cpi->speed = frame_params->speed;
3336 cm->show_existing_frame = frame_params->show_existing_frame;
3337 cpi->existing_fb_idx_to_show = frame_params->existing_fb_idx_to_show;
3338
3339 memcpy(cm->remapped_ref_idx, frame_params->remapped_ref_idx,
3340 REF_FRAMES * sizeof(*cm->remapped_ref_idx));
3341
3342 memcpy(&cpi->refresh_frame, &frame_params->refresh_frame,
3343 sizeof(cpi->refresh_frame));
3344
3345 if (current_frame->frame_type == KEY_FRAME && !cpi->no_show_fwd_kf) {
3346 current_frame->frame_number = 0;
3347 }
3348
3349 current_frame->order_hint =
3350 current_frame->frame_number + frame_params->order_offset;
3351
3352 current_frame->display_order_hint = current_frame->order_hint;
3353 current_frame->order_hint %=
3354 (1 << (cm->seq_params.order_hint_info.order_hint_bits_minus_1 + 1));
3355
3356 if (is_stat_generation_stage(cpi)) {
3357 #if !CONFIG_REALTIME_ONLY
3358 av1_first_pass(cpi, frame_input->ts_duration);
3359 #endif
3360 } else if (cpi->oxcf.pass == 0 || cpi->oxcf.pass == 2) {
3361 if (encode_frame_to_data_rate(cpi, &frame_results->size, dest) !=
3362 AOM_CODEC_OK) {
3363 return AOM_CODEC_ERROR;
3364 }
3365 } else {
3366 return AOM_CODEC_ERROR;
3367 }
3368
3369 return AOM_CODEC_OK;
3370 }
3371
3372 #if CONFIG_DENOISE
apply_denoise_2d(AV1_COMP * cpi,YV12_BUFFER_CONFIG * sd,int block_size,float noise_level,int64_t time_stamp,int64_t end_time)3373 static int apply_denoise_2d(AV1_COMP *cpi, YV12_BUFFER_CONFIG *sd,
3374 int block_size, float noise_level,
3375 int64_t time_stamp, int64_t end_time) {
3376 AV1_COMMON *const cm = &cpi->common;
3377 if (!cpi->denoise_and_model) {
3378 cpi->denoise_and_model = aom_denoise_and_model_alloc(
3379 cm->seq_params.bit_depth, block_size, noise_level);
3380 if (!cpi->denoise_and_model) {
3381 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
3382 "Error allocating denoise and model");
3383 return -1;
3384 }
3385 }
3386 if (!cpi->film_grain_table) {
3387 cpi->film_grain_table = aom_malloc(sizeof(*cpi->film_grain_table));
3388 if (!cpi->film_grain_table) {
3389 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
3390 "Error allocating grain table");
3391 return -1;
3392 }
3393 memset(cpi->film_grain_table, 0, sizeof(*cpi->film_grain_table));
3394 }
3395 if (aom_denoise_and_model_run(cpi->denoise_and_model, sd,
3396 &cm->film_grain_params,
3397 cpi->oxcf.enable_dnl_denoising)) {
3398 if (cm->film_grain_params.apply_grain) {
3399 aom_film_grain_table_append(cpi->film_grain_table, time_stamp, end_time,
3400 &cm->film_grain_params);
3401 }
3402 }
3403 return 0;
3404 }
3405 #endif
3406
av1_receive_raw_frame(AV1_COMP * cpi,aom_enc_frame_flags_t frame_flags,YV12_BUFFER_CONFIG * sd,int64_t time_stamp,int64_t end_time)3407 int av1_receive_raw_frame(AV1_COMP *cpi, aom_enc_frame_flags_t frame_flags,
3408 YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
3409 int64_t end_time) {
3410 AV1_COMMON *const cm = &cpi->common;
3411 const SequenceHeader *const seq_params = &cm->seq_params;
3412 int res = 0;
3413 const int subsampling_x = sd->subsampling_x;
3414 const int subsampling_y = sd->subsampling_y;
3415 const int use_highbitdepth = (sd->flags & YV12_FLAG_HIGHBITDEPTH) != 0;
3416
3417 #if CONFIG_TUNE_VMAF
3418 if (!is_stat_generation_stage(cpi) &&
3419 cpi->oxcf.tune_cfg.tuning == AOM_TUNE_VMAF_WITH_PREPROCESSING) {
3420 av1_vmaf_frame_preprocessing(cpi, sd);
3421 }
3422 if (!is_stat_generation_stage(cpi) &&
3423 cpi->oxcf.tune_cfg.tuning == AOM_TUNE_VMAF_MAX_GAIN) {
3424 av1_vmaf_blk_preprocessing(cpi, sd);
3425 }
3426 #endif
3427
3428 #if CONFIG_INTERNAL_STATS
3429 struct aom_usec_timer timer;
3430 aom_usec_timer_start(&timer);
3431 #endif
3432
3433 #if CONFIG_AV1_TEMPORAL_DENOISING
3434 setup_denoiser_buffer(cpi);
3435 #endif
3436
3437 #if CONFIG_DENOISE
3438 // even if denoise_noise_level is > 0, we don't need need to denoise on pass
3439 // 1 of 2 if enable_dnl_denoising is disabled since the 2nd pass will be
3440 // encoding the original (non-denoised) frame
3441 if (cpi->oxcf.noise_level > 0 &&
3442 !(cpi->oxcf.pass == AOM_RC_FIRST_PASS && !cpi->oxcf.enable_dnl_denoising))
3443 if (apply_denoise_2d(cpi, sd, cpi->oxcf.noise_block_size,
3444 cpi->oxcf.noise_level, time_stamp, end_time) < 0)
3445 res = -1;
3446 #endif // CONFIG_DENOISE
3447
3448 if (av1_lookahead_push(cpi->lookahead, sd, time_stamp, end_time,
3449 use_highbitdepth, frame_flags))
3450 res = -1;
3451 #if CONFIG_INTERNAL_STATS
3452 aom_usec_timer_mark(&timer);
3453 cpi->time_receive_data += aom_usec_timer_elapsed(&timer);
3454 #endif
3455
3456 // Note: Regarding profile setting, the following checks are added to help
3457 // choose a proper profile for the input video. The criterion is that all
3458 // bitstreams must be designated as the lowest profile that match its content.
3459 // E.G. A bitstream that contains 4:4:4 video must be designated as High
3460 // Profile in the seq header, and likewise a bitstream that contains 4:2:2
3461 // bitstream must be designated as Professional Profile in the sequence
3462 // header.
3463 if ((seq_params->profile == PROFILE_0) && !seq_params->monochrome &&
3464 (subsampling_x != 1 || subsampling_y != 1)) {
3465 aom_internal_error(&cm->error, AOM_CODEC_INVALID_PARAM,
3466 "Non-4:2:0 color format requires profile 1 or 2");
3467 res = -1;
3468 }
3469 if ((seq_params->profile == PROFILE_1) &&
3470 !(subsampling_x == 0 && subsampling_y == 0)) {
3471 aom_internal_error(&cm->error, AOM_CODEC_INVALID_PARAM,
3472 "Profile 1 requires 4:4:4 color format");
3473 res = -1;
3474 }
3475 if ((seq_params->profile == PROFILE_2) &&
3476 (seq_params->bit_depth <= AOM_BITS_10) &&
3477 !(subsampling_x == 1 && subsampling_y == 0)) {
3478 aom_internal_error(&cm->error, AOM_CODEC_INVALID_PARAM,
3479 "Profile 2 bit-depth <= 10 requires 4:2:2 color format");
3480 res = -1;
3481 }
3482
3483 return res;
3484 }
3485
3486 #if CONFIG_INTERNAL_STATS
3487 extern double av1_get_blockiness(const unsigned char *img1, int img1_pitch,
3488 const unsigned char *img2, int img2_pitch,
3489 int width, int height);
3490
adjust_image_stat(double y,double u,double v,double all,ImageStat * s)3491 static void adjust_image_stat(double y, double u, double v, double all,
3492 ImageStat *s) {
3493 s->stat[STAT_Y] += y;
3494 s->stat[STAT_U] += u;
3495 s->stat[STAT_V] += v;
3496 s->stat[STAT_ALL] += all;
3497 s->worst = AOMMIN(s->worst, all);
3498 }
3499
compute_internal_stats(AV1_COMP * cpi,int frame_bytes)3500 static void compute_internal_stats(AV1_COMP *cpi, int frame_bytes) {
3501 AV1_COMMON *const cm = &cpi->common;
3502 double samples = 0.0;
3503 const uint32_t in_bit_depth = cpi->oxcf.input_cfg.input_bit_depth;
3504 const uint32_t bit_depth = cpi->td.mb.e_mbd.bd;
3505
3506 #if CONFIG_INTER_STATS_ONLY
3507 if (cm->current_frame.frame_type == KEY_FRAME) return; // skip key frame
3508 #endif
3509 cpi->bytes += frame_bytes;
3510 if (cm->show_frame) {
3511 const YV12_BUFFER_CONFIG *orig = cpi->source;
3512 const YV12_BUFFER_CONFIG *recon = &cpi->common.cur_frame->buf;
3513 double y, u, v, frame_all;
3514
3515 cpi->count[0]++;
3516 cpi->count[1]++;
3517 if (cpi->b_calculate_psnr) {
3518 PSNR_STATS psnr;
3519 double weight[2] = { 0.0, 0.0 };
3520 double frame_ssim2[2] = { 0.0, 0.0 };
3521 aom_clear_system_state();
3522 #if CONFIG_AV1_HIGHBITDEPTH
3523 aom_calc_highbd_psnr(orig, recon, &psnr, bit_depth, in_bit_depth);
3524 #else
3525 aom_calc_psnr(orig, recon, &psnr);
3526 #endif
3527 adjust_image_stat(psnr.psnr[1], psnr.psnr[2], psnr.psnr[3], psnr.psnr[0],
3528 &(cpi->psnr[0]));
3529 cpi->total_sq_error[0] += psnr.sse[0];
3530 cpi->total_samples[0] += psnr.samples[0];
3531 samples = psnr.samples[0];
3532
3533 // TODO(yaowu): unify these two versions into one.
3534 if (cm->seq_params.use_highbitdepth)
3535 aom_highbd_calc_ssim(orig, recon, weight, bit_depth, in_bit_depth,
3536 frame_ssim2);
3537 else
3538 aom_calc_ssim(orig, recon, &weight[0], &frame_ssim2[0]);
3539
3540 cpi->worst_ssim = AOMMIN(cpi->worst_ssim, frame_ssim2[0]);
3541 cpi->summed_quality += frame_ssim2[0] * weight[0];
3542 cpi->summed_weights += weight[0];
3543
3544 #if CONFIG_AV1_HIGHBITDEPTH
3545 // Compute PSNR based on stream bit depth
3546 if ((cpi->source->flags & YV12_FLAG_HIGHBITDEPTH) &&
3547 (in_bit_depth < bit_depth)) {
3548 adjust_image_stat(psnr.psnr_hbd[1], psnr.psnr_hbd[2], psnr.psnr_hbd[3],
3549 psnr.psnr_hbd[0], &cpi->psnr[1]);
3550 cpi->total_sq_error[1] += psnr.sse_hbd[0];
3551 cpi->total_samples[1] += psnr.samples_hbd[0];
3552
3553 cpi->worst_ssim_hbd = AOMMIN(cpi->worst_ssim_hbd, frame_ssim2[1]);
3554 cpi->summed_quality_hbd += frame_ssim2[1] * weight[1];
3555 cpi->summed_weights_hbd += weight[1];
3556 }
3557 #endif
3558
3559 #if 0
3560 {
3561 FILE *f = fopen("q_used.stt", "a");
3562 double y2 = psnr.psnr[1];
3563 double u2 = psnr.psnr[2];
3564 double v2 = psnr.psnr[3];
3565 double frame_psnr2 = psnr.psnr[0];
3566 fprintf(f, "%5d : Y%f7.3:U%f7.3:V%f7.3:F%f7.3:S%7.3f\n",
3567 cm->current_frame.frame_number, y2, u2, v2,
3568 frame_psnr2, frame_ssim2);
3569 fclose(f);
3570 }
3571 #endif
3572 }
3573 if (cpi->b_calculate_blockiness) {
3574 if (!cm->seq_params.use_highbitdepth) {
3575 const double frame_blockiness =
3576 av1_get_blockiness(orig->y_buffer, orig->y_stride, recon->y_buffer,
3577 recon->y_stride, orig->y_width, orig->y_height);
3578 cpi->worst_blockiness = AOMMAX(cpi->worst_blockiness, frame_blockiness);
3579 cpi->total_blockiness += frame_blockiness;
3580 }
3581
3582 if (cpi->b_calculate_consistency) {
3583 if (!cm->seq_params.use_highbitdepth) {
3584 const double this_inconsistency = aom_get_ssim_metrics(
3585 orig->y_buffer, orig->y_stride, recon->y_buffer, recon->y_stride,
3586 orig->y_width, orig->y_height, cpi->ssim_vars, &cpi->metrics, 1);
3587
3588 const double peak = (double)((1 << in_bit_depth) - 1);
3589 const double consistency =
3590 aom_sse_to_psnr(samples, peak, cpi->total_inconsistency);
3591 if (consistency > 0.0)
3592 cpi->worst_consistency =
3593 AOMMIN(cpi->worst_consistency, consistency);
3594 cpi->total_inconsistency += this_inconsistency;
3595 }
3596 }
3597 }
3598
3599 frame_all =
3600 aom_calc_fastssim(orig, recon, &y, &u, &v, bit_depth, in_bit_depth);
3601 adjust_image_stat(y, u, v, frame_all, &cpi->fastssim);
3602 frame_all = aom_psnrhvs(orig, recon, &y, &u, &v, bit_depth, in_bit_depth);
3603 adjust_image_stat(y, u, v, frame_all, &cpi->psnrhvs);
3604 }
3605 }
3606 #endif // CONFIG_INTERNAL_STATS
3607
av1_get_compressed_data(AV1_COMP * cpi,unsigned int * frame_flags,size_t * size,uint8_t * dest,int64_t * time_stamp,int64_t * time_end,int flush,const aom_rational64_t * timestamp_ratio)3608 int av1_get_compressed_data(AV1_COMP *cpi, unsigned int *frame_flags,
3609 size_t *size, uint8_t *dest, int64_t *time_stamp,
3610 int64_t *time_end, int flush,
3611 const aom_rational64_t *timestamp_ratio) {
3612 const AV1EncoderConfig *const oxcf = &cpi->oxcf;
3613 AV1_COMMON *const cm = &cpi->common;
3614
3615 #if CONFIG_BITSTREAM_DEBUG
3616 assert(cpi->oxcf.max_threads <= 1 &&
3617 "bitstream debug tool does not support multithreading");
3618 bitstream_queue_record_write();
3619 aom_bitstream_queue_set_frame_write(cm->current_frame.order_hint * 2 +
3620 cm->show_frame);
3621 #endif
3622 if (cpi->use_svc && cm->number_spatial_layers > 1) {
3623 av1_one_pass_cbr_svc_start_layer(cpi);
3624 }
3625
3626 cm->showable_frame = 0;
3627 *size = 0;
3628 #if CONFIG_INTERNAL_STATS
3629 struct aom_usec_timer cmptimer;
3630 aom_usec_timer_start(&cmptimer);
3631 #endif
3632 av1_set_high_precision_mv(cpi, 1, 0);
3633
3634 // Normal defaults
3635 cm->features.refresh_frame_context =
3636 oxcf->tool_cfg.frame_parallel_decoding_mode
3637 ? REFRESH_FRAME_CONTEXT_DISABLED
3638 : REFRESH_FRAME_CONTEXT_BACKWARD;
3639 if (oxcf->tile_cfg.enable_large_scale_tile)
3640 cm->features.refresh_frame_context = REFRESH_FRAME_CONTEXT_DISABLED;
3641
3642 // Initialize fields related to forward keyframes
3643 cpi->no_show_fwd_kf = 0;
3644
3645 if (assign_cur_frame_new_fb(cm) == NULL) return AOM_CODEC_ERROR;
3646
3647 #if CONFIG_COLLECT_COMPONENT_TIMING
3648 // Only accumulate 2nd pass time.
3649 if (cpi->oxcf.pass == 2) start_timing(cpi, av1_encode_strategy_time);
3650 #endif
3651
3652 const int result =
3653 av1_encode_strategy(cpi, size, dest, frame_flags, time_stamp, time_end,
3654 timestamp_ratio, flush);
3655
3656 #if CONFIG_COLLECT_COMPONENT_TIMING
3657 if (cpi->oxcf.pass == 2) end_timing(cpi, av1_encode_strategy_time);
3658
3659 // Print out timing information.
3660 // Note: Use "cpi->frame_component_time[0] > 100 us" to avoid showing of
3661 // show_existing_frame and lag-in-frames.
3662 if (cpi->oxcf.pass == 2 && cpi->frame_component_time[0] > 100) {
3663 int i;
3664 uint64_t frame_total = 0, total = 0;
3665
3666 fprintf(stderr, "\n Frame number: %d, Frame type: %s, Show Frame: %d\n",
3667 cm->current_frame.frame_number,
3668 get_frame_type_enum(cm->current_frame.frame_type), cm->show_frame);
3669 for (i = 0; i < kTimingComponents; i++) {
3670 cpi->component_time[i] += cpi->frame_component_time[i];
3671 // Use av1_encode_strategy_time (i = 0) as the total time.
3672 if (i == 0) {
3673 frame_total = cpi->frame_component_time[0];
3674 total = cpi->component_time[0];
3675 }
3676 fprintf(stderr,
3677 " %50s: %15" PRId64 " us [%6.2f%%] (total: %15" PRId64
3678 " us [%6.2f%%])\n",
3679 get_component_name(i), cpi->frame_component_time[i],
3680 (float)((float)cpi->frame_component_time[i] * 100.0 /
3681 (float)frame_total),
3682 cpi->component_time[i],
3683 (float)((float)cpi->component_time[i] * 100.0 / (float)total));
3684 cpi->frame_component_time[i] = 0;
3685 }
3686 }
3687 #endif
3688
3689 if (result == -1) {
3690 // Returning -1 indicates no frame encoded; more input is required
3691 return -1;
3692 }
3693 if (result != AOM_CODEC_OK) {
3694 return AOM_CODEC_ERROR;
3695 }
3696 #if CONFIG_INTERNAL_STATS
3697 aom_usec_timer_mark(&cmptimer);
3698 cpi->time_compress_data += aom_usec_timer_elapsed(&cmptimer);
3699 #endif // CONFIG_INTERNAL_STATS
3700 // Note *size = 0 indicates a dropped frame for which psnr is not calculated
3701 if (cpi->b_calculate_psnr && *size > 0) {
3702 if (cm->show_existing_frame ||
3703 (!is_stat_generation_stage(cpi) && cm->show_frame)) {
3704 generate_psnr_packet(cpi);
3705 }
3706 }
3707
3708 if (cpi->level_params.keep_level_stats && !is_stat_generation_stage(cpi)) {
3709 // Initialize level info. at the beginning of each sequence.
3710 if (cm->current_frame.frame_type == KEY_FRAME && !cpi->no_show_fwd_kf) {
3711 av1_init_level_info(cpi);
3712 }
3713 av1_update_level_info(cpi, *size, *time_stamp, *time_end);
3714 }
3715
3716 #if CONFIG_INTERNAL_STATS
3717 if (!is_stat_generation_stage(cpi)) {
3718 compute_internal_stats(cpi, (int)(*size));
3719 }
3720 #endif // CONFIG_INTERNAL_STATS
3721 #if CONFIG_SPEED_STATS
3722 if (!is_stat_generation_stage(cpi) && !cm->show_existing_frame) {
3723 cpi->tx_search_count += cpi->td.mb.txfm_search_info.tx_search_count;
3724 cpi->td.mb.txfm_search_info.tx_search_count = 0;
3725 }
3726 #endif // CONFIG_SPEED_STATS
3727
3728 aom_clear_system_state();
3729
3730 return AOM_CODEC_OK;
3731 }
3732
av1_get_preview_raw_frame(AV1_COMP * cpi,YV12_BUFFER_CONFIG * dest)3733 int av1_get_preview_raw_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *dest) {
3734 AV1_COMMON *cm = &cpi->common;
3735 if (!cm->show_frame) {
3736 return -1;
3737 } else {
3738 int ret;
3739 if (cm->cur_frame != NULL) {
3740 *dest = cm->cur_frame->buf;
3741 dest->y_width = cm->width;
3742 dest->y_height = cm->height;
3743 dest->uv_width = cm->width >> cm->seq_params.subsampling_x;
3744 dest->uv_height = cm->height >> cm->seq_params.subsampling_y;
3745 ret = 0;
3746 } else {
3747 ret = -1;
3748 }
3749 aom_clear_system_state();
3750 return ret;
3751 }
3752 }
3753
av1_get_last_show_frame(AV1_COMP * cpi,YV12_BUFFER_CONFIG * frame)3754 int av1_get_last_show_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *frame) {
3755 if (cpi->last_show_frame_buf == NULL) return -1;
3756
3757 *frame = cpi->last_show_frame_buf->buf;
3758 return 0;
3759 }
3760
av1_copy_new_frame_enc(AV1_COMMON * cm,YV12_BUFFER_CONFIG * new_frame,YV12_BUFFER_CONFIG * sd)3761 aom_codec_err_t av1_copy_new_frame_enc(AV1_COMMON *cm,
3762 YV12_BUFFER_CONFIG *new_frame,
3763 YV12_BUFFER_CONFIG *sd) {
3764 const int num_planes = av1_num_planes(cm);
3765 if (!equal_dimensions_and_border(new_frame, sd))
3766 aom_internal_error(&cm->error, AOM_CODEC_ERROR,
3767 "Incorrect buffer dimensions");
3768 else
3769 aom_yv12_copy_frame(new_frame, sd, num_planes);
3770
3771 return cm->error.error_code;
3772 }
3773
av1_set_internal_size(AV1EncoderConfig * const oxcf,ResizePendingParams * resize_pending_params,AOM_SCALING horiz_mode,AOM_SCALING vert_mode)3774 int av1_set_internal_size(AV1EncoderConfig *const oxcf,
3775 ResizePendingParams *resize_pending_params,
3776 AOM_SCALING horiz_mode, AOM_SCALING vert_mode) {
3777 int hr = 0, hs = 0, vr = 0, vs = 0;
3778
3779 if (horiz_mode > ONETWO || vert_mode > ONETWO) return -1;
3780
3781 Scale2Ratio(horiz_mode, &hr, &hs);
3782 Scale2Ratio(vert_mode, &vr, &vs);
3783
3784 // always go to the next whole number
3785 resize_pending_params->width = (hs - 1 + oxcf->frm_dim_cfg.width * hr) / hs;
3786 resize_pending_params->height = (vs - 1 + oxcf->frm_dim_cfg.height * vr) / vs;
3787
3788 if (horiz_mode != NORMAL || vert_mode != NORMAL) {
3789 oxcf->resize_cfg.resize_mode = RESIZE_FIXED;
3790 oxcf->algo_cfg.enable_tpl_model = 0;
3791 }
3792 return 0;
3793 }
3794
av1_get_quantizer(AV1_COMP * cpi)3795 int av1_get_quantizer(AV1_COMP *cpi) {
3796 return cpi->common.quant_params.base_qindex;
3797 }
3798
av1_convert_sect5obus_to_annexb(uint8_t * buffer,size_t * frame_size)3799 int av1_convert_sect5obus_to_annexb(uint8_t *buffer, size_t *frame_size) {
3800 size_t output_size = 0;
3801 size_t total_bytes_read = 0;
3802 size_t remaining_size = *frame_size;
3803 uint8_t *buff_ptr = buffer;
3804
3805 // go through each OBUs
3806 while (total_bytes_read < *frame_size) {
3807 uint8_t saved_obu_header[2];
3808 uint64_t obu_payload_size;
3809 size_t length_of_payload_size;
3810 size_t length_of_obu_size;
3811 uint32_t obu_header_size = (buff_ptr[0] >> 2) & 0x1 ? 2 : 1;
3812 size_t obu_bytes_read = obu_header_size; // bytes read for current obu
3813
3814 // save the obu header (1 or 2 bytes)
3815 memmove(saved_obu_header, buff_ptr, obu_header_size);
3816 // clear the obu_has_size_field
3817 saved_obu_header[0] = saved_obu_header[0] & (~0x2);
3818
3819 // get the payload_size and length of payload_size
3820 if (aom_uleb_decode(buff_ptr + obu_header_size, remaining_size,
3821 &obu_payload_size, &length_of_payload_size) != 0) {
3822 return AOM_CODEC_ERROR;
3823 }
3824 obu_bytes_read += length_of_payload_size;
3825
3826 // calculate the length of size of the obu header plus payload
3827 length_of_obu_size =
3828 aom_uleb_size_in_bytes((uint64_t)(obu_header_size + obu_payload_size));
3829
3830 // move the rest of data to new location
3831 memmove(buff_ptr + length_of_obu_size + obu_header_size,
3832 buff_ptr + obu_bytes_read, remaining_size - obu_bytes_read);
3833 obu_bytes_read += (size_t)obu_payload_size;
3834
3835 // write the new obu size
3836 const uint64_t obu_size = obu_header_size + obu_payload_size;
3837 size_t coded_obu_size;
3838 if (aom_uleb_encode(obu_size, sizeof(obu_size), buff_ptr,
3839 &coded_obu_size) != 0) {
3840 return AOM_CODEC_ERROR;
3841 }
3842
3843 // write the saved (modified) obu_header following obu size
3844 memmove(buff_ptr + length_of_obu_size, saved_obu_header, obu_header_size);
3845
3846 total_bytes_read += obu_bytes_read;
3847 remaining_size -= obu_bytes_read;
3848 buff_ptr += length_of_obu_size + obu_size;
3849 output_size += length_of_obu_size + (size_t)obu_size;
3850 }
3851
3852 *frame_size = output_size;
3853 return AOM_CODEC_OK;
3854 }
3855
svc_set_updates_external_ref_frame_config(ExtRefreshFrameFlagsInfo * const ext_refresh_frame_flags,SVC * const svc)3856 static void svc_set_updates_external_ref_frame_config(
3857 ExtRefreshFrameFlagsInfo *const ext_refresh_frame_flags, SVC *const svc) {
3858 ext_refresh_frame_flags->update_pending = 1;
3859 ext_refresh_frame_flags->last_frame = svc->refresh[svc->ref_idx[0]];
3860 ext_refresh_frame_flags->golden_frame = svc->refresh[svc->ref_idx[3]];
3861 ext_refresh_frame_flags->bwd_ref_frame = svc->refresh[svc->ref_idx[4]];
3862 ext_refresh_frame_flags->alt2_ref_frame = svc->refresh[svc->ref_idx[5]];
3863 ext_refresh_frame_flags->alt_ref_frame = svc->refresh[svc->ref_idx[6]];
3864 svc->non_reference_frame = 1;
3865 for (int i = 0; i < REF_FRAMES; i++) {
3866 if (svc->refresh[i] == 1) {
3867 svc->non_reference_frame = 0;
3868 break;
3869 }
3870 }
3871 }
3872
svc_set_references_external_ref_frame_config(AV1_COMP * cpi)3873 static int svc_set_references_external_ref_frame_config(AV1_COMP *cpi) {
3874 // LAST_FRAME (0), LAST2_FRAME(1), LAST3_FRAME(2), GOLDEN_FRAME(3),
3875 // BWDREF_FRAME(4), ALTREF2_FRAME(5), ALTREF_FRAME(6).
3876 int ref = AOM_REFFRAME_ALL;
3877 for (int i = 0; i < INTER_REFS_PER_FRAME; i++) {
3878 if (!cpi->svc.reference[i]) ref ^= (1 << i);
3879 }
3880 return ref;
3881 }
3882
av1_apply_encoding_flags(AV1_COMP * cpi,aom_enc_frame_flags_t flags)3883 void av1_apply_encoding_flags(AV1_COMP *cpi, aom_enc_frame_flags_t flags) {
3884 // TODO(yunqingwang): For what references to use, external encoding flags
3885 // should be consistent with internal reference frame selection. Need to
3886 // ensure that there is not conflict between the two. In AV1 encoder, the
3887 // priority rank for 7 reference frames are: LAST, ALTREF, LAST2, LAST3,
3888 // GOLDEN, BWDREF, ALTREF2.
3889
3890 ExternalFlags *const ext_flags = &cpi->ext_flags;
3891 ExtRefreshFrameFlagsInfo *const ext_refresh_frame_flags =
3892 &ext_flags->refresh_frame;
3893 ext_flags->ref_frame_flags = AOM_REFFRAME_ALL;
3894 if (flags &
3895 (AOM_EFLAG_NO_REF_LAST | AOM_EFLAG_NO_REF_LAST2 | AOM_EFLAG_NO_REF_LAST3 |
3896 AOM_EFLAG_NO_REF_GF | AOM_EFLAG_NO_REF_ARF | AOM_EFLAG_NO_REF_BWD |
3897 AOM_EFLAG_NO_REF_ARF2)) {
3898 int ref = AOM_REFFRAME_ALL;
3899
3900 if (flags & AOM_EFLAG_NO_REF_LAST) ref ^= AOM_LAST_FLAG;
3901 if (flags & AOM_EFLAG_NO_REF_LAST2) ref ^= AOM_LAST2_FLAG;
3902 if (flags & AOM_EFLAG_NO_REF_LAST3) ref ^= AOM_LAST3_FLAG;
3903
3904 if (flags & AOM_EFLAG_NO_REF_GF) ref ^= AOM_GOLD_FLAG;
3905
3906 if (flags & AOM_EFLAG_NO_REF_ARF) {
3907 ref ^= AOM_ALT_FLAG;
3908 ref ^= AOM_BWD_FLAG;
3909 ref ^= AOM_ALT2_FLAG;
3910 } else {
3911 if (flags & AOM_EFLAG_NO_REF_BWD) ref ^= AOM_BWD_FLAG;
3912 if (flags & AOM_EFLAG_NO_REF_ARF2) ref ^= AOM_ALT2_FLAG;
3913 }
3914
3915 av1_use_as_reference(&ext_flags->ref_frame_flags, ref);
3916 } else {
3917 if (cpi->svc.external_ref_frame_config) {
3918 int ref = svc_set_references_external_ref_frame_config(cpi);
3919 av1_use_as_reference(&ext_flags->ref_frame_flags, ref);
3920 }
3921 }
3922
3923 if (flags &
3924 (AOM_EFLAG_NO_UPD_LAST | AOM_EFLAG_NO_UPD_GF | AOM_EFLAG_NO_UPD_ARF)) {
3925 int upd = AOM_REFFRAME_ALL;
3926
3927 // Refreshing LAST/LAST2/LAST3 is handled by 1 common flag.
3928 if (flags & AOM_EFLAG_NO_UPD_LAST) upd ^= AOM_LAST_FLAG;
3929
3930 if (flags & AOM_EFLAG_NO_UPD_GF) upd ^= AOM_GOLD_FLAG;
3931
3932 if (flags & AOM_EFLAG_NO_UPD_ARF) {
3933 upd ^= AOM_ALT_FLAG;
3934 upd ^= AOM_BWD_FLAG;
3935 upd ^= AOM_ALT2_FLAG;
3936 }
3937
3938 ext_refresh_frame_flags->last_frame = (upd & AOM_LAST_FLAG) != 0;
3939 ext_refresh_frame_flags->golden_frame = (upd & AOM_GOLD_FLAG) != 0;
3940 ext_refresh_frame_flags->alt_ref_frame = (upd & AOM_ALT_FLAG) != 0;
3941 ext_refresh_frame_flags->bwd_ref_frame = (upd & AOM_BWD_FLAG) != 0;
3942 ext_refresh_frame_flags->alt2_ref_frame = (upd & AOM_ALT2_FLAG) != 0;
3943 ext_refresh_frame_flags->update_pending = 1;
3944 } else {
3945 if (cpi->svc.external_ref_frame_config)
3946 svc_set_updates_external_ref_frame_config(ext_refresh_frame_flags,
3947 &cpi->svc);
3948 else
3949 ext_refresh_frame_flags->update_pending = 0;
3950 }
3951
3952 ext_flags->use_ref_frame_mvs = cpi->oxcf.tool_cfg.enable_ref_frame_mvs &
3953 ((flags & AOM_EFLAG_NO_REF_FRAME_MVS) == 0);
3954 ext_flags->use_error_resilient = cpi->oxcf.tool_cfg.error_resilient_mode |
3955 ((flags & AOM_EFLAG_ERROR_RESILIENT) != 0);
3956 ext_flags->use_s_frame =
3957 cpi->oxcf.kf_cfg.enable_sframe | ((flags & AOM_EFLAG_SET_S_FRAME) != 0);
3958 ext_flags->use_primary_ref_none =
3959 (flags & AOM_EFLAG_SET_PRIMARY_REF_NONE) != 0;
3960
3961 if (flags & AOM_EFLAG_NO_UPD_ENTROPY) {
3962 update_entropy(&ext_flags->refresh_frame_context,
3963 &ext_flags->refresh_frame_context_pending, 0);
3964 }
3965 }
3966
av1_get_global_headers(AV1_COMP * cpi)3967 aom_fixed_buf_t *av1_get_global_headers(AV1_COMP *cpi) {
3968 if (!cpi) return NULL;
3969
3970 uint8_t header_buf[512] = { 0 };
3971 const uint32_t sequence_header_size =
3972 av1_write_sequence_header_obu(&cpi->common.seq_params, &header_buf[0]);
3973 assert(sequence_header_size <= sizeof(header_buf));
3974 if (sequence_header_size == 0) return NULL;
3975
3976 const size_t obu_header_size = 1;
3977 const size_t size_field_size = aom_uleb_size_in_bytes(sequence_header_size);
3978 const size_t payload_offset = obu_header_size + size_field_size;
3979
3980 if (payload_offset + sequence_header_size > sizeof(header_buf)) return NULL;
3981 memmove(&header_buf[payload_offset], &header_buf[0], sequence_header_size);
3982
3983 if (av1_write_obu_header(&cpi->level_params, OBU_SEQUENCE_HEADER, 0,
3984 &header_buf[0]) != obu_header_size) {
3985 return NULL;
3986 }
3987
3988 size_t coded_size_field_size = 0;
3989 if (aom_uleb_encode(sequence_header_size, size_field_size,
3990 &header_buf[obu_header_size],
3991 &coded_size_field_size) != 0) {
3992 return NULL;
3993 }
3994 assert(coded_size_field_size == size_field_size);
3995
3996 aom_fixed_buf_t *global_headers =
3997 (aom_fixed_buf_t *)malloc(sizeof(*global_headers));
3998 if (!global_headers) return NULL;
3999
4000 const size_t global_header_buf_size =
4001 obu_header_size + size_field_size + sequence_header_size;
4002
4003 global_headers->buf = malloc(global_header_buf_size);
4004 if (!global_headers->buf) {
4005 free(global_headers);
4006 return NULL;
4007 }
4008
4009 memcpy(global_headers->buf, &header_buf[0], global_header_buf_size);
4010 global_headers->sz = global_header_buf_size;
4011 return global_headers;
4012 }
4013