AOMedia AV1 Codec
svc_encoder_rtc
1/*
2 * Copyright (c) 2019, 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// This is an example demonstrating how to implement a multi-layer AOM
13// encoding scheme for RTC video applications.
14
15#include <assert.h>
16#include <inttypes.h>
17#include <limits.h>
18#include <math.h>
19#include <stdio.h>
20#include <stdlib.h>
21#include <string.h>
22
23#include <memory>
24
25#include "config/aom_config.h"
26
27#if CONFIG_AV1_DECODER
28#include "aom/aom_decoder.h"
29#endif
30#include "aom/aom_encoder.h"
31#include "aom/aom_image.h"
32#include "aom/aom_integer.h"
33#include "aom/aomcx.h"
34#include "aom_dsp/bitwriter_buffer.h"
35#include "aom_ports/aom_timer.h"
36#include "av1/ratectrl_rtc.h"
37#include "common/args.h"
38#include "common/tools_common.h"
39#include "common/video_writer.h"
40#include "examples/encoder_util.h"
41#include "examples/multilayer_metadata.h"
42
43#define OPTION_BUFFER_SIZE 1024
44#define MAX_NUM_SPATIAL_LAYERS 4
45
46#define GOOD_QUALITY 0
47
48typedef struct {
49 const char *output_filename;
50 char options[OPTION_BUFFER_SIZE];
51 struct AvxInputContext input_ctx[MAX_NUM_SPATIAL_LAYERS];
52 int speed;
53 int aq_mode;
54 int layering_mode;
55 int output_obu;
56 int decode;
57 int tune_content;
58 int show_psnr;
59 bool use_external_rc;
60 bool scale_factors_explicitly_set;
61 const char *multilayer_metadata_file;
62} AppInput;
63
64typedef enum {
65 QUANTIZER = 0,
66 BITRATE,
67 SCALE_FACTOR,
68 AUTO_ALT_REF,
69 ALL_OPTION_TYPES
70} LAYER_OPTION_TYPE;
71
72enum { kSkip = 0, kDeltaQ = 1, kDeltaLF = 2, kReference = 3 };
73
74static const arg_def_t outputfile =
75 ARG_DEF("o", "output", 1, "Output filename");
76static const arg_def_t frames_arg =
77 ARG_DEF("f", "frames", 1, "Number of frames to encode");
78static const arg_def_t threads_arg =
79 ARG_DEF("th", "threads", 1, "Number of threads to use");
80static const arg_def_t width_arg = ARG_DEF("w", "width", 1, "Source width");
81static const arg_def_t height_arg = ARG_DEF("h", "height", 1, "Source height");
82static const arg_def_t timebase_arg =
83 ARG_DEF("t", "timebase", 1, "Timebase (num/den)");
84static const arg_def_t bitrate_arg = ARG_DEF(
85 "b", "target-bitrate", 1, "Encoding bitrate, in kilobits per second");
86static const arg_def_t spatial_layers_arg =
87 ARG_DEF("sl", "spatial-layers", 1, "Number of spatial SVC layers");
88static const arg_def_t temporal_layers_arg =
89 ARG_DEF("tl", "temporal-layers", 1, "Number of temporal SVC layers");
90static const arg_def_t layering_mode_arg =
91 ARG_DEF("lm", "layering-mode", 1, "Temporal layering scheme.");
92static const arg_def_t kf_dist_arg =
93 ARG_DEF("k", "kf-dist", 1, "Number of frames between keyframes");
94static const arg_def_t scale_factors_arg =
95 ARG_DEF("r", "scale-factors", 1, "Scale factors (lowest to highest layer)");
96static const arg_def_t min_q_arg =
97 ARG_DEF(NULL, "min-q", 1, "Minimum quantizer");
98static const arg_def_t max_q_arg =
99 ARG_DEF(NULL, "max-q", 1, "Maximum quantizer");
100static const arg_def_t speed_arg =
101 ARG_DEF("sp", "speed", 1, "Speed configuration");
102static const arg_def_t aqmode_arg =
103 ARG_DEF("aq", "aqmode", 1, "AQ mode off/on");
104static const arg_def_t bitrates_arg =
105 ARG_DEF("bl", "bitrates", 1,
106 "Bitrates[spatial_layer * num_temporal_layer + temporal_layer]");
107static const arg_def_t dropframe_thresh_arg =
108 ARG_DEF(NULL, "drop-frame", 1, "Temporal resampling threshold (buf %)");
109static const arg_def_t error_resilient_arg =
110 ARG_DEF(NULL, "error-resilient", 1, "Error resilient flag");
111static const arg_def_t output_obu_arg =
112 ARG_DEF(NULL, "output-obu", 1,
113 "Write OBUs when set to 1. Otherwise write IVF files.");
114static const arg_def_t test_decode_arg =
115 ARG_DEF(NULL, "test-decode", 1,
116 "Attempt to test decoding the output when set to 1. Default is 1.");
117static const arg_def_t psnr_arg =
118 ARG_DEF(NULL, "psnr", -1, "Show PSNR in status line.");
119static const arg_def_t ext_rc_arg =
120 ARG_DEF(NULL, "use-ext-rc", 0, "Use external rate control.");
121static const struct arg_enum_list tune_content_enum[] = {
122 { "default", AOM_CONTENT_DEFAULT },
123 { "screen", AOM_CONTENT_SCREEN },
124 { "film", AOM_CONTENT_FILM },
125 { NULL, 0 }
126};
127static const arg_def_t tune_content_arg = ARG_DEF_ENUM(
128 NULL, "tune-content", 1, "Tune content type", tune_content_enum);
129#if CONFIG_CWG_E050
130static const arg_def_t multilayer_metadata_file_arg =
131 ARG_DEF("ml", "multilayer_metadata_file", 1,
132 "Experimental: path to multilayer metadata file");
133#endif
134
135#if CONFIG_AV1_HIGHBITDEPTH
136static const struct arg_enum_list bitdepth_enum[] = { { "8", AOM_BITS_8 },
137 { "10", AOM_BITS_10 },
138 { NULL, 0 } };
139
140static const arg_def_t bitdepth_arg = ARG_DEF_ENUM(
141 "d", "bit-depth", 1, "Bit depth for codec 8 or 10. ", bitdepth_enum);
142#endif // CONFIG_AV1_HIGHBITDEPTH
143
144static const arg_def_t *svc_args[] = {
145 &frames_arg,
146 &outputfile,
147 &width_arg,
148 &height_arg,
149 &timebase_arg,
150 &bitrate_arg,
151 &spatial_layers_arg,
152 &kf_dist_arg,
153 &scale_factors_arg,
154 &min_q_arg,
155 &max_q_arg,
156 &temporal_layers_arg,
157 &layering_mode_arg,
158 &threads_arg,
159 &aqmode_arg,
160#if CONFIG_AV1_HIGHBITDEPTH
161 &bitdepth_arg,
162#endif
163 &speed_arg,
164 &bitrates_arg,
165 &dropframe_thresh_arg,
166 &error_resilient_arg,
167 &output_obu_arg,
168 &test_decode_arg,
169 &tune_content_arg,
170 &psnr_arg,
171#if CONFIG_CWG_E050
172 &multilayer_metadata_file_arg,
173#endif
174 NULL,
175};
176
177#define zero(Dest) memset(&(Dest), 0, sizeof(Dest))
178
179static const char *exec_name;
180
181void usage_exit(void) {
182 fprintf(stderr,
183 "Usage: %s <options> input_filename [input_filename ...] -o "
184 "output_filename\n",
185 exec_name);
186 fprintf(stderr, "Options:\n");
187 arg_show_usage(stderr, svc_args);
188 fprintf(
189 stderr,
190 "Input files must be y4m or yuv.\n"
191 "If multiple input files are specified, they correspond to spatial "
192 "layers, and there should be as many as there are spatial layers.\n"
193 "All input files must have the same width, height, frame rate and number "
194 "of frames.\n"
195 "If only one file is specified, it is used for all spatial layers.\n");
196 exit(EXIT_FAILURE);
197}
198
199static int file_is_y4m(const char detect[4]) {
200 return memcmp(detect, "YUV4", 4) == 0;
201}
202
203static int fourcc_is_ivf(const char detect[4]) {
204 if (memcmp(detect, "DKIF", 4) == 0) {
205 return 1;
206 }
207 return 0;
208}
209
210static const int option_max_values[ALL_OPTION_TYPES] = { 63, INT_MAX, INT_MAX,
211 1 };
212
213static const int option_min_values[ALL_OPTION_TYPES] = { 0, 0, 1, 0 };
214
215static void open_input_file(struct AvxInputContext *input,
217 /* Parse certain options from the input file, if possible */
218 input->file = strcmp(input->filename, "-") ? fopen(input->filename, "rb")
219 : set_binary_mode(stdin);
220
221 if (!input->file) fatal("Failed to open input file");
222
223 if (!fseeko(input->file, 0, SEEK_END)) {
224 /* Input file is seekable. Figure out how long it is, so we can get
225 * progress info.
226 */
227 input->length = ftello(input->file);
228 rewind(input->file);
229 }
230
231 /* Default to 1:1 pixel aspect ratio. */
232 input->pixel_aspect_ratio.numerator = 1;
233 input->pixel_aspect_ratio.denominator = 1;
234
235 /* For RAW input sources, these bytes will applied on the first frame
236 * in read_frame().
237 */
238 input->detect.buf_read = fread(input->detect.buf, 1, 4, input->file);
239 input->detect.position = 0;
240
241 if (input->detect.buf_read == 4 && file_is_y4m(input->detect.buf)) {
242 if (y4m_input_open(&input->y4m, input->file, input->detect.buf, 4, csp,
243 input->only_i420) >= 0) {
244 input->file_type = FILE_TYPE_Y4M;
245 input->width = input->y4m.pic_w;
246 input->height = input->y4m.pic_h;
247 input->pixel_aspect_ratio.numerator = input->y4m.par_n;
248 input->pixel_aspect_ratio.denominator = input->y4m.par_d;
249 input->framerate.numerator = input->y4m.fps_n;
250 input->framerate.denominator = input->y4m.fps_d;
251 input->fmt = input->y4m.aom_fmt;
252 input->bit_depth = static_cast<aom_bit_depth_t>(input->y4m.bit_depth);
253 } else {
254 fatal("Unsupported Y4M stream.");
255 }
256 } else if (input->detect.buf_read == 4 && fourcc_is_ivf(input->detect.buf)) {
257 fatal("IVF is not supported as input.");
258 } else {
259 input->file_type = FILE_TYPE_RAW;
260 }
261}
262
263static aom_codec_err_t extract_option(LAYER_OPTION_TYPE type, char *input,
264 int *value0, int *value1) {
265 if (type == SCALE_FACTOR) {
266 *value0 = (int)strtol(input, &input, 10);
267 if (*input++ != '/') return AOM_CODEC_INVALID_PARAM;
268 *value1 = (int)strtol(input, &input, 10);
269
270 if (*value0 < option_min_values[SCALE_FACTOR] ||
271 *value1 < option_min_values[SCALE_FACTOR] ||
272 *value0 > option_max_values[SCALE_FACTOR] ||
273 *value1 > option_max_values[SCALE_FACTOR] ||
274 *value0 > *value1) // num shouldn't be greater than den
276 } else {
277 *value0 = atoi(input);
278 if (*value0 < option_min_values[type] || *value0 > option_max_values[type])
280 }
281 return AOM_CODEC_OK;
282}
283
284static aom_codec_err_t parse_layer_options_from_string(
285 aom_svc_params_t *svc_params, LAYER_OPTION_TYPE type, const char *input,
286 int *option0, int *option1) {
288 char *input_string;
289 char *token;
290 const char *delim = ",";
291 int num_layers = svc_params->number_spatial_layers;
292 int i = 0;
293
294 if (type == BITRATE)
295 num_layers =
296 svc_params->number_spatial_layers * svc_params->number_temporal_layers;
297
298 if (input == NULL || option0 == NULL ||
299 (option1 == NULL && type == SCALE_FACTOR))
301
302 const size_t input_length = strlen(input);
303 input_string = reinterpret_cast<char *>(malloc(input_length + 1));
304 if (input_string == NULL) return AOM_CODEC_MEM_ERROR;
305 memcpy(input_string, input, input_length + 1);
306 token = strtok(input_string, delim); // NOLINT
307 for (i = 0; i < num_layers; ++i) {
308 if (token != NULL) {
309 res = extract_option(type, token, option0 + i, option1 + i);
310 if (res != AOM_CODEC_OK) break;
311 token = strtok(NULL, delim); // NOLINT
312 } else {
314 break;
315 }
316 }
317 free(input_string);
318 return res;
319}
320
321static void parse_command_line(int argc, const char **argv_,
322 AppInput *app_input,
323 aom_svc_params_t *svc_params,
324 aom_codec_enc_cfg_t *enc_cfg) {
325 struct arg arg;
326 char **argv = NULL;
327 char **argi = NULL;
328 char **argj = NULL;
329 char string_options[1024] = { 0 };
330
331 // Default settings
332 svc_params->number_spatial_layers = 1;
333 svc_params->number_temporal_layers = 1;
334 app_input->layering_mode = 0;
335 app_input->output_obu = 0;
336 app_input->decode = 1;
337 enc_cfg->g_threads = 1;
338 enc_cfg->rc_end_usage = AOM_CBR;
339
340 // process command line options
341 argv = argv_dup(argc - 1, argv_ + 1);
342 if (!argv) {
343 fprintf(stderr, "Error allocating argument list\n");
344 exit(EXIT_FAILURE);
345 }
346 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) {
347 arg.argv_step = 1;
348
349 if (arg_match(&arg, &outputfile, argi)) {
350 app_input->output_filename = arg.val;
351 } else if (arg_match(&arg, &width_arg, argi)) {
352 enc_cfg->g_w = arg_parse_uint(&arg);
353 } else if (arg_match(&arg, &height_arg, argi)) {
354 enc_cfg->g_h = arg_parse_uint(&arg);
355 } else if (arg_match(&arg, &timebase_arg, argi)) {
356 enc_cfg->g_timebase = arg_parse_rational(&arg);
357 } else if (arg_match(&arg, &bitrate_arg, argi)) {
358 enc_cfg->rc_target_bitrate = arg_parse_uint(&arg);
359 } else if (arg_match(&arg, &spatial_layers_arg, argi)) {
360 svc_params->number_spatial_layers = arg_parse_uint(&arg);
361 } else if (arg_match(&arg, &temporal_layers_arg, argi)) {
362 svc_params->number_temporal_layers = arg_parse_uint(&arg);
363 } else if (arg_match(&arg, &speed_arg, argi)) {
364 app_input->speed = arg_parse_uint(&arg);
365 if (app_input->speed > 11) {
366 aom_tools_warn("Mapping speed %d to speed 11.\n", app_input->speed);
367 }
368 } else if (arg_match(&arg, &aqmode_arg, argi)) {
369 app_input->aq_mode = arg_parse_uint(&arg);
370 } else if (arg_match(&arg, &threads_arg, argi)) {
371 enc_cfg->g_threads = arg_parse_uint(&arg);
372 } else if (arg_match(&arg, &layering_mode_arg, argi)) {
373 app_input->layering_mode = arg_parse_int(&arg);
374 } else if (arg_match(&arg, &kf_dist_arg, argi)) {
375 enc_cfg->kf_min_dist = arg_parse_uint(&arg);
376 enc_cfg->kf_max_dist = enc_cfg->kf_min_dist;
377 } else if (arg_match(&arg, &scale_factors_arg, argi)) {
378 aom_codec_err_t res = parse_layer_options_from_string(
379 svc_params, SCALE_FACTOR, arg.val, svc_params->scaling_factor_num,
380 svc_params->scaling_factor_den);
381 app_input->scale_factors_explicitly_set = true;
382 if (res != AOM_CODEC_OK) {
383 die("Failed to parse scale factors: %s\n",
385 }
386 } else if (arg_match(&arg, &min_q_arg, argi)) {
387 enc_cfg->rc_min_quantizer = arg_parse_uint(&arg);
388 } else if (arg_match(&arg, &max_q_arg, argi)) {
389 enc_cfg->rc_max_quantizer = arg_parse_uint(&arg);
390#if CONFIG_AV1_HIGHBITDEPTH
391 } else if (arg_match(&arg, &bitdepth_arg, argi)) {
392 enc_cfg->g_bit_depth =
393 static_cast<aom_bit_depth_t>(arg_parse_enum_or_int(&arg));
394 switch (enc_cfg->g_bit_depth) {
395 case AOM_BITS_8:
396 enc_cfg->g_input_bit_depth = 8;
397 enc_cfg->g_profile = 0;
398 break;
399 case AOM_BITS_10:
400 enc_cfg->g_input_bit_depth = 10;
401 enc_cfg->g_profile = 0;
402 break;
403 default:
404 die("Error: Invalid bit depth selected (%d)\n", enc_cfg->g_bit_depth);
405 }
406#endif // CONFIG_VP9_HIGHBITDEPTH
407 } else if (arg_match(&arg, &dropframe_thresh_arg, argi)) {
408 enc_cfg->rc_dropframe_thresh = arg_parse_uint(&arg);
409 } else if (arg_match(&arg, &error_resilient_arg, argi)) {
410 enc_cfg->g_error_resilient = arg_parse_uint(&arg);
411 if (enc_cfg->g_error_resilient != 0 && enc_cfg->g_error_resilient != 1)
412 die("Invalid value for error resilient (0, 1): %d.",
413 enc_cfg->g_error_resilient);
414 } else if (arg_match(&arg, &output_obu_arg, argi)) {
415 app_input->output_obu = arg_parse_uint(&arg);
416 if (app_input->output_obu != 0 && app_input->output_obu != 1)
417 die("Invalid value for obu output flag (0, 1): %d.",
418 app_input->output_obu);
419 } else if (arg_match(&arg, &test_decode_arg, argi)) {
420 app_input->decode = arg_parse_uint(&arg);
421 if (app_input->decode != 0 && app_input->decode != 1)
422 die("Invalid value for test decode flag (0, 1): %d.",
423 app_input->decode);
424 } else if (arg_match(&arg, &tune_content_arg, argi)) {
425 app_input->tune_content = arg_parse_enum_or_int(&arg);
426 printf("tune content %d\n", app_input->tune_content);
427 } else if (arg_match(&arg, &psnr_arg, argi)) {
428 app_input->show_psnr = 1;
429 } else if (arg_match(&arg, &ext_rc_arg, argi)) {
430 app_input->use_external_rc = true;
431#if CONFIG_CWG_E050
432 } else if (arg_match(&arg, &multilayer_metadata_file_arg, argi)) {
433 app_input->multilayer_metadata_file = arg.val;
434#endif
435 } else {
436 ++argj;
437 }
438 }
439
440 // Total bitrate needs to be parsed after the number of layers.
441 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) {
442 arg.argv_step = 1;
443 if (arg_match(&arg, &bitrates_arg, argi)) {
444 aom_codec_err_t res = parse_layer_options_from_string(
445 svc_params, BITRATE, arg.val, svc_params->layer_target_bitrate, NULL);
446 if (res != AOM_CODEC_OK) {
447 die("Failed to parse bitrates: %s\n", aom_codec_err_to_string(res));
448 }
449 } else {
450 ++argj;
451 }
452 }
453
454 // There will be a space in front of the string options
455 if (strlen(string_options) > 0)
456 strncpy(app_input->options, string_options, OPTION_BUFFER_SIZE);
457
458 // Check for unrecognized options
459 for (argi = argv; *argi; ++argi)
460 if (argi[0][0] == '-' && strlen(argi[0]) > 1)
461 die("Error: Unrecognized option %s\n", *argi);
462
463 if (argv[0] == NULL) {
464 usage_exit();
465 }
466
467 int input_count = 0;
468 while (argv[input_count] != NULL && input_count < MAX_NUM_SPATIAL_LAYERS) {
469 app_input->input_ctx[input_count].filename = argv[input_count];
470 ++input_count;
471 }
472 if (input_count > 1 && input_count != svc_params->number_spatial_layers) {
473 die("Error: Number of input files does not match number of spatial layers");
474 }
475 if (argv[input_count] != NULL) {
476 die("Error: Too many input files specified, there should be at most %d",
477 MAX_NUM_SPATIAL_LAYERS);
478 }
479
480 free(argv);
481
482 for (int i = 0; i < input_count; ++i) {
483 open_input_file(&app_input->input_ctx[i], AOM_CSP_UNKNOWN);
484 if (app_input->input_ctx[i].file_type == FILE_TYPE_Y4M) {
485 if (enc_cfg->g_w == 0 || enc_cfg->g_h == 0) {
486 // Override these settings with the info from Y4M file.
487 enc_cfg->g_w = app_input->input_ctx[i].width;
488 enc_cfg->g_h = app_input->input_ctx[i].height;
489 // g_timebase is the reciprocal of frame rate.
490 enc_cfg->g_timebase.num = app_input->input_ctx[i].framerate.denominator;
491 enc_cfg->g_timebase.den = app_input->input_ctx[i].framerate.numerator;
492 } else if (enc_cfg->g_w != app_input->input_ctx[i].width ||
493 enc_cfg->g_h != app_input->input_ctx[i].height ||
494 enc_cfg->g_timebase.num !=
495 app_input->input_ctx[i].framerate.denominator ||
496 enc_cfg->g_timebase.den !=
497 app_input->input_ctx[i].framerate.numerator) {
498 die("Error: Input file dimensions and/or frame rate mismatch");
499 }
500 }
501 }
502 if (enc_cfg->g_w == 0 || enc_cfg->g_h == 0) {
503 die("Error: Input file dimensions not set, use -w and -h");
504 }
505
506 if (enc_cfg->g_w < 16 || enc_cfg->g_w % 2 || enc_cfg->g_h < 16 ||
507 enc_cfg->g_h % 2)
508 die("Invalid resolution: %d x %d\n", enc_cfg->g_w, enc_cfg->g_h);
509
510 printf(
511 "Codec %s\n"
512 "layers: %d\n"
513 "width %u, height: %u\n"
514 "num: %d, den: %d, bitrate: %u\n"
515 "gop size: %u\n",
517 svc_params->number_spatial_layers, enc_cfg->g_w, enc_cfg->g_h,
518 enc_cfg->g_timebase.num, enc_cfg->g_timebase.den,
519 enc_cfg->rc_target_bitrate, enc_cfg->kf_max_dist);
520}
521
522static const int mode_to_num_temporal_layers[12] = {
523 1, 2, 3, 3, 2, 1, 1, 3, 3, 3, 3, 3,
524};
525static const int mode_to_num_spatial_layers[12] = {
526 1, 1, 1, 1, 1, 2, 3, 2, 3, 3, 3, 3,
527};
528
529// For rate control encoding stats.
530struct RateControlMetrics {
531 // Number of input frames per layer.
532 int layer_input_frames[AOM_MAX_TS_LAYERS];
533 // Number of encoded non-key frames per layer.
534 int layer_enc_frames[AOM_MAX_TS_LAYERS];
535 // Framerate per layer layer (cumulative).
536 double layer_framerate[AOM_MAX_TS_LAYERS];
537 // Target average frame size per layer (per-frame-bandwidth per layer).
538 double layer_pfb[AOM_MAX_LAYERS];
539 // Actual average frame size per layer.
540 double layer_avg_frame_size[AOM_MAX_LAYERS];
541 // Average rate mismatch per layer (|target - actual| / target).
542 double layer_avg_rate_mismatch[AOM_MAX_LAYERS];
543 // Actual encoding bitrate per layer (cumulative across temporal layers).
544 double layer_encoding_bitrate[AOM_MAX_LAYERS];
545 // Average of the short-time encoder actual bitrate.
546 // TODO(marpan): Should we add these short-time stats for each layer?
547 double avg_st_encoding_bitrate;
548 // Variance of the short-time encoder actual bitrate.
549 double variance_st_encoding_bitrate;
550 // Window (number of frames) for computing short-timee encoding bitrate.
551 int window_size;
552 // Number of window measurements.
553 int window_count;
554 int layer_target_bitrate[AOM_MAX_LAYERS];
555};
556
557static const int REF_FRAMES = 8;
558
559static const int INTER_REFS_PER_FRAME = 7;
560
561// Reference frames used in this example encoder.
562enum {
563 SVC_LAST_FRAME = 0,
564 SVC_LAST2_FRAME,
565 SVC_LAST3_FRAME,
566 SVC_GOLDEN_FRAME,
567 SVC_BWDREF_FRAME,
568 SVC_ALTREF2_FRAME,
569 SVC_ALTREF_FRAME
570};
571
572static int read_frame(struct AvxInputContext *input_ctx, aom_image_t *img) {
573 FILE *f = input_ctx->file;
574 y4m_input *y4m = &input_ctx->y4m;
575 int shortread = 0;
576
577 if (input_ctx->file_type == FILE_TYPE_Y4M) {
578 if (y4m_input_fetch_frame(y4m, f, img) < 1) return 0;
579 } else {
580 shortread = read_yuv_frame(input_ctx, img);
581 }
582
583 return !shortread;
584}
585
586static void close_input_file(struct AvxInputContext *input) {
587 fclose(input->file);
588 if (input->file_type == FILE_TYPE_Y4M) y4m_input_close(&input->y4m);
589}
590
591// Note: these rate control metrics assume only 1 key frame in the
592// sequence (i.e., first frame only). So for temporal pattern# 7
593// (which has key frame for every frame on base layer), the metrics
594// computation will be off/wrong.
595// TODO(marpan): Update these metrics to account for multiple key frames
596// in the stream.
597static void set_rate_control_metrics(struct RateControlMetrics *rc,
598 double framerate, int ss_number_layers,
599 int ts_number_layers) {
600 int ts_rate_decimator[AOM_MAX_TS_LAYERS] = { 1 };
601 ts_rate_decimator[0] = 1;
602 if (ts_number_layers == 2) {
603 ts_rate_decimator[0] = 2;
604 ts_rate_decimator[1] = 1;
605 }
606 if (ts_number_layers == 3) {
607 ts_rate_decimator[0] = 4;
608 ts_rate_decimator[1] = 2;
609 ts_rate_decimator[2] = 1;
610 }
611 // Set the layer (cumulative) framerate and the target layer (non-cumulative)
612 // per-frame-bandwidth, for the rate control encoding stats below.
613 for (int sl = 0; sl < ss_number_layers; ++sl) {
614 int i = sl * ts_number_layers;
615 rc->layer_framerate[0] = framerate / ts_rate_decimator[0];
616 rc->layer_pfb[i] =
617 1000.0 * rc->layer_target_bitrate[i] / rc->layer_framerate[0];
618 for (int tl = 0; tl < ts_number_layers; ++tl) {
619 i = sl * ts_number_layers + tl;
620 if (tl > 0) {
621 rc->layer_framerate[tl] = framerate / ts_rate_decimator[tl];
622 rc->layer_pfb[i] =
623 1000.0 *
624 (rc->layer_target_bitrate[i] - rc->layer_target_bitrate[i - 1]) /
625 (rc->layer_framerate[tl] - rc->layer_framerate[tl - 1]);
626 }
627 rc->layer_input_frames[tl] = 0;
628 rc->layer_enc_frames[tl] = 0;
629 rc->layer_encoding_bitrate[i] = 0.0;
630 rc->layer_avg_frame_size[i] = 0.0;
631 rc->layer_avg_rate_mismatch[i] = 0.0;
632 }
633 }
634 rc->window_count = 0;
635 rc->window_size = 15;
636 rc->avg_st_encoding_bitrate = 0.0;
637 rc->variance_st_encoding_bitrate = 0.0;
638}
639
640static void printout_rate_control_summary(struct RateControlMetrics *rc,
641 int frame_cnt, int ss_number_layers,
642 int ts_number_layers) {
643 int tot_num_frames = 0;
644 double perc_fluctuation = 0.0;
645 printf("Total number of processed frames: %d\n\n", frame_cnt - 1);
646 printf("Rate control layer stats for %d layer(s):\n\n", ts_number_layers);
647 for (int sl = 0; sl < ss_number_layers; ++sl) {
648 tot_num_frames = 0;
649 for (int tl = 0; tl < ts_number_layers; ++tl) {
650 int i = sl * ts_number_layers + tl;
651 const int num_dropped =
652 tl > 0 ? rc->layer_input_frames[tl] - rc->layer_enc_frames[tl]
653 : rc->layer_input_frames[tl] - rc->layer_enc_frames[tl] - 1;
654 tot_num_frames += rc->layer_input_frames[tl];
655 rc->layer_encoding_bitrate[i] = 0.001 * rc->layer_framerate[tl] *
656 rc->layer_encoding_bitrate[i] /
657 tot_num_frames;
658 rc->layer_avg_frame_size[i] =
659 rc->layer_avg_frame_size[i] / rc->layer_enc_frames[tl];
660 rc->layer_avg_rate_mismatch[i] =
661 100.0 * rc->layer_avg_rate_mismatch[i] / rc->layer_enc_frames[tl];
662 printf("For layer#: %d %d \n", sl, tl);
663 printf("Bitrate (target vs actual): %d %f\n", rc->layer_target_bitrate[i],
664 rc->layer_encoding_bitrate[i]);
665 printf("Average frame size (target vs actual): %f %f\n", rc->layer_pfb[i],
666 rc->layer_avg_frame_size[i]);
667 printf("Average rate_mismatch: %f\n", rc->layer_avg_rate_mismatch[i]);
668 printf(
669 "Number of input frames, encoded (non-key) frames, "
670 "and perc dropped frames: %d %d %f\n",
671 rc->layer_input_frames[tl], rc->layer_enc_frames[tl],
672 100.0 * num_dropped / rc->layer_input_frames[tl]);
673 printf("\n");
674 }
675 }
676 rc->avg_st_encoding_bitrate = rc->avg_st_encoding_bitrate / rc->window_count;
677 rc->variance_st_encoding_bitrate =
678 rc->variance_st_encoding_bitrate / rc->window_count -
679 (rc->avg_st_encoding_bitrate * rc->avg_st_encoding_bitrate);
680 perc_fluctuation = 100.0 * sqrt(rc->variance_st_encoding_bitrate) /
681 rc->avg_st_encoding_bitrate;
682 printf("Short-time stats, for window of %d frames:\n", rc->window_size);
683 printf("Average, rms-variance, and percent-fluct: %f %f %f\n",
684 rc->avg_st_encoding_bitrate, sqrt(rc->variance_st_encoding_bitrate),
685 perc_fluctuation);
686 if (frame_cnt - 1 != tot_num_frames)
687 die("Error: Number of input frames not equal to output!\n");
688}
689
690// Layer pattern configuration.
691static void set_layer_pattern(
692 int layering_mode, int superframe_cnt, aom_svc_layer_id_t *layer_id,
693 aom_svc_ref_frame_config_t *ref_frame_config,
694 aom_svc_ref_frame_comp_pred_t *ref_frame_comp_pred, int *use_svc_control,
695 int spatial_layer_id, int is_key_frame, int ksvc_mode, int speed,
696 int *reference_updated, int test_roi_map) {
697 // Setting this flag to 1 enables simplex example of
698 // RPS (Reference Picture Selection) for 1 layer.
699 int use_rps_example = 0;
700 int i;
701 int enable_longterm_temporal_ref = 1;
702 int shift = (layering_mode == 8) ? 2 : 0;
703 int simulcast_mode = (layering_mode == 11);
704 *use_svc_control = 1;
705 layer_id->spatial_layer_id = spatial_layer_id;
706 int lag_index = 0;
707 int base_count = superframe_cnt >> 2;
708 ref_frame_comp_pred->use_comp_pred[0] = 0; // GOLDEN_LAST
709 ref_frame_comp_pred->use_comp_pred[1] = 0; // LAST2_LAST
710 ref_frame_comp_pred->use_comp_pred[2] = 0; // ALTREF_LAST
711 // Set the reference map buffer idx for the 7 references:
712 // LAST_FRAME (0), LAST2_FRAME(1), LAST3_FRAME(2), GOLDEN_FRAME(3),
713 // BWDREF_FRAME(4), ALTREF2_FRAME(5), ALTREF_FRAME(6).
714 for (i = 0; i < INTER_REFS_PER_FRAME; i++) ref_frame_config->ref_idx[i] = i;
715 for (i = 0; i < INTER_REFS_PER_FRAME; i++) ref_frame_config->reference[i] = 0;
716 for (i = 0; i < REF_FRAMES; i++) ref_frame_config->refresh[i] = 0;
717
718 if (ksvc_mode) {
719 // Same pattern as case 9, but the reference strucutre will be constrained
720 // below.
721 layering_mode = 9;
722 }
723 switch (layering_mode) {
724 case 0:
725 if (use_rps_example == 0) {
726 // 1-layer: update LAST on every frame, reference LAST.
727 layer_id->temporal_layer_id = 0;
728 layer_id->spatial_layer_id = 0;
729 ref_frame_config->refresh[0] = 1;
730 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
731 // Add additional reference (GOLDEN) if test_roi_map is set,
732 // to test reference frame feature on segment.
733 if (test_roi_map) ref_frame_config->reference[SVC_GOLDEN_FRAME] = 1;
734 } else {
735 // Pattern of 2 references (ALTREF and GOLDEN) trailing
736 // LAST by 4 and 8 frames, with some switching logic to
737 // sometimes only predict from the longer-term reference
738 //(golden here). This is simple example to test RPS
739 // (reference picture selection).
740 int last_idx = 0;
741 int last_idx_refresh = 0;
742 int gld_idx = 0;
743 int alt_ref_idx = 0;
744 int lag_alt = 4;
745 int lag_gld = 8;
746 layer_id->temporal_layer_id = 0;
747 layer_id->spatial_layer_id = 0;
748 int sh = 8; // slots 0 - 7.
749 // Moving index slot for last: 0 - (sh - 1)
750 if (superframe_cnt > 1) last_idx = (superframe_cnt - 1) % sh;
751 // Moving index for refresh of last: one ahead for next frame.
752 last_idx_refresh = superframe_cnt % sh;
753 // Moving index for gld_ref, lag behind current by lag_gld
754 if (superframe_cnt > lag_gld) gld_idx = (superframe_cnt - lag_gld) % sh;
755 // Moving index for alt_ref, lag behind LAST by lag_alt frames.
756 if (superframe_cnt > lag_alt)
757 alt_ref_idx = (superframe_cnt - lag_alt) % sh;
758 // Set the ref_idx.
759 // Default all references to slot for last.
760 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
761 ref_frame_config->ref_idx[i] = last_idx;
762 // Set the ref_idx for the relevant references.
763 ref_frame_config->ref_idx[SVC_LAST_FRAME] = last_idx;
764 ref_frame_config->ref_idx[SVC_LAST2_FRAME] = last_idx_refresh;
765 ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = gld_idx;
766 ref_frame_config->ref_idx[SVC_ALTREF_FRAME] = alt_ref_idx;
767 // Refresh this slot, which will become LAST on next frame.
768 ref_frame_config->refresh[last_idx_refresh] = 1;
769 // Reference LAST, ALTREF, and GOLDEN
770 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
771 ref_frame_config->reference[SVC_ALTREF_FRAME] = 1;
772 ref_frame_config->reference[SVC_GOLDEN_FRAME] = 1;
773 // Switch to only GOLDEN every 300 frames.
774 if (superframe_cnt % 200 == 0 && superframe_cnt > 0) {
775 ref_frame_config->reference[SVC_LAST_FRAME] = 0;
776 ref_frame_config->reference[SVC_ALTREF_FRAME] = 0;
777 ref_frame_config->reference[SVC_GOLDEN_FRAME] = 1;
778 // Test if the long-term is LAST instead, this is just a renaming
779 // but its tests if encoder behaves the same, whether its
780 // LAST or GOLDEN.
781 if (superframe_cnt % 400 == 0 && superframe_cnt > 0) {
782 ref_frame_config->ref_idx[SVC_LAST_FRAME] = gld_idx;
783 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
784 ref_frame_config->reference[SVC_ALTREF_FRAME] = 0;
785 ref_frame_config->reference[SVC_GOLDEN_FRAME] = 0;
786 }
787 }
788 }
789 break;
790 case 1:
791 // 2-temporal layer.
792 // 1 3 5
793 // 0 2 4
794 // Keep golden fixed at slot 3.
795 base_count = superframe_cnt >> 1;
796 ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 3;
797 // Cyclically refresh slots 5, 6, 7, for lag alt ref.
798 lag_index = 5;
799 if (base_count > 0) {
800 lag_index = 5 + (base_count % 3);
801 if (superframe_cnt % 2 != 0) lag_index = 5 + ((base_count + 1) % 3);
802 }
803 // Set the altref slot to lag_index.
804 ref_frame_config->ref_idx[SVC_ALTREF_FRAME] = lag_index;
805 if (superframe_cnt % 2 == 0) {
806 layer_id->temporal_layer_id = 0;
807 // Update LAST on layer 0, reference LAST.
808 ref_frame_config->refresh[0] = 1;
809 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
810 // Refresh lag_index slot, needed for lagging golen.
811 ref_frame_config->refresh[lag_index] = 1;
812 // Refresh GOLDEN every x base layer frames.
813 if (base_count % 32 == 0) ref_frame_config->refresh[3] = 1;
814 } else {
815 layer_id->temporal_layer_id = 1;
816 // No updates on layer 1, reference LAST (TL0).
817 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
818 }
819 // Always reference golden and altref on TL0.
820 if (layer_id->temporal_layer_id == 0) {
821 ref_frame_config->reference[SVC_GOLDEN_FRAME] = 1;
822 ref_frame_config->reference[SVC_ALTREF_FRAME] = 1;
823 }
824 break;
825 case 2:
826 // 3-temporal layer:
827 // 1 3 5 7
828 // 2 6
829 // 0 4 8
830 if (superframe_cnt % 4 == 0) {
831 // Base layer.
832 layer_id->temporal_layer_id = 0;
833 // Update LAST on layer 0, reference LAST.
834 ref_frame_config->refresh[0] = 1;
835 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
836 } else if ((superframe_cnt - 1) % 4 == 0) {
837 layer_id->temporal_layer_id = 2;
838 // First top layer: no updates, only reference LAST (TL0).
839 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
840 } else if ((superframe_cnt - 2) % 4 == 0) {
841 layer_id->temporal_layer_id = 1;
842 // Middle layer (TL1): update LAST2, only reference LAST (TL0).
843 ref_frame_config->refresh[1] = 1;
844 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
845 } else if ((superframe_cnt - 3) % 4 == 0) {
846 layer_id->temporal_layer_id = 2;
847 // Second top layer: no updates, only reference LAST.
848 // Set buffer idx for LAST to slot 1, since that was the slot
849 // updated in previous frame. So LAST is TL1 frame.
850 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
851 ref_frame_config->ref_idx[SVC_LAST2_FRAME] = 0;
852 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
853 }
854 break;
855 case 3:
856 // 3 TL, same as above, except allow for predicting
857 // off 2 more references (GOLDEN and ALTREF), with
858 // GOLDEN updated periodically, and ALTREF lagging from
859 // LAST from ~4 frames. Both GOLDEN and ALTREF
860 // can only be updated on base temporal layer.
861
862 // Keep golden fixed at slot 3.
863 ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 3;
864 // Cyclically refresh slots 5, 6, 7, for lag altref.
865 lag_index = 5;
866 if (base_count > 0) {
867 lag_index = 5 + (base_count % 3);
868 if (superframe_cnt % 4 != 0) lag_index = 5 + ((base_count + 1) % 3);
869 }
870 // Set the altref slot to lag_index.
871 ref_frame_config->ref_idx[SVC_ALTREF_FRAME] = lag_index;
872 if (superframe_cnt % 4 == 0) {
873 // Base layer.
874 layer_id->temporal_layer_id = 0;
875 // Update LAST on layer 0, reference LAST.
876 ref_frame_config->refresh[0] = 1;
877 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
878 // Refresh GOLDEN every x ~10 base layer frames.
879 if (base_count % 10 == 0) ref_frame_config->refresh[3] = 1;
880 // Refresh lag_index slot, needed for lagging altref.
881 ref_frame_config->refresh[lag_index] = 1;
882 } else if ((superframe_cnt - 1) % 4 == 0) {
883 layer_id->temporal_layer_id = 2;
884 // First top layer: no updates, only reference LAST (TL0).
885 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
886 } else if ((superframe_cnt - 2) % 4 == 0) {
887 layer_id->temporal_layer_id = 1;
888 // Middle layer (TL1): update LAST2, only reference LAST (TL0).
889 ref_frame_config->refresh[1] = 1;
890 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
891 } else if ((superframe_cnt - 3) % 4 == 0) {
892 layer_id->temporal_layer_id = 2;
893 // Second top layer: no updates, only reference LAST.
894 // Set buffer idx for LAST to slot 1, since that was the slot
895 // updated in previous frame. So LAST is TL1 frame.
896 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
897 ref_frame_config->ref_idx[SVC_LAST2_FRAME] = 0;
898 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
899 }
900 // Every frame can reference GOLDEN AND ALTREF.
901 ref_frame_config->reference[SVC_GOLDEN_FRAME] = 1;
902 ref_frame_config->reference[SVC_ALTREF_FRAME] = 1;
903 // Allow for compound prediction for LAST-ALTREF and LAST-GOLDEN.
904 if (speed >= 7) {
905 ref_frame_comp_pred->use_comp_pred[2] = 1;
906 ref_frame_comp_pred->use_comp_pred[0] = 1;
907 }
908 break;
909 case 4:
910 // 3-temporal layer: but middle layer updates GF, so 2nd TL2 will
911 // only reference GF (not LAST). Other frames only reference LAST.
912 // 1 3 5 7
913 // 2 6
914 // 0 4 8
915 if (superframe_cnt % 4 == 0) {
916 // Base layer.
917 layer_id->temporal_layer_id = 0;
918 // Update LAST on layer 0, only reference LAST.
919 ref_frame_config->refresh[0] = 1;
920 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
921 } else if ((superframe_cnt - 1) % 4 == 0) {
922 layer_id->temporal_layer_id = 2;
923 // First top layer: no updates, only reference LAST (TL0).
924 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
925 } else if ((superframe_cnt - 2) % 4 == 0) {
926 layer_id->temporal_layer_id = 1;
927 // Middle layer (TL1): update GF, only reference LAST (TL0).
928 ref_frame_config->refresh[3] = 1;
929 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
930 } else if ((superframe_cnt - 3) % 4 == 0) {
931 layer_id->temporal_layer_id = 2;
932 // Second top layer: no updates, only reference GF.
933 ref_frame_config->reference[SVC_GOLDEN_FRAME] = 1;
934 }
935 break;
936
937 case 5:
938 /*
939 // 2 spatial layers, 1 temporal, without temporal prediction on SL1.
940 layer_id->temporal_layer_id = 0;
941 if (layer_id->spatial_layer_id == 0) {
942 // Reference LAST, update LAST.
943 ref_frame_config->refresh[0] = 1;
944 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
945 } else if (layer_id->spatial_layer_id == 1) {
946 // Reference LAST, which is SL0, and no refresh.
947 ref_frame_config->refresh[0] = 0;
948 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
949 }
950 break;
951 */
952 // 2 spatial layers, 1 temporal.
953 layer_id->temporal_layer_id = 0;
954 if (layer_id->spatial_layer_id == 0) {
955 // Reference LAST, update LAST.
956 ref_frame_config->refresh[0] = 1;
957 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 0;
958 ref_frame_config->ref_idx[SVC_LAST2_FRAME] = 2;
959 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
960 } else if (layer_id->spatial_layer_id == 1) {
961 // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 1
962 // and GOLDEN to slot 0. Update slot 1 (LAST).
963 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
964 ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 0;
965 ref_frame_config->ref_idx[SVC_LAST2_FRAME] = 2;
966 ref_frame_config->refresh[1] = 1;
967 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
968 ref_frame_config->reference[SVC_GOLDEN_FRAME] = 1;
969 }
970 break;
971
972 case 6:
973 // 3 spatial layers, 1 temporal.
974 // Note for this case, we set the buffer idx for all references to be
975 // either LAST or GOLDEN, which are always valid references, since decoder
976 // will check if any of the 7 references is valid scale in
977 // valid_ref_frame_size().
978 layer_id->temporal_layer_id = 0;
979 if (layer_id->spatial_layer_id == 0) {
980 // Reference LAST, update LAST. Set all buffer_idx to 0.
981 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
982 ref_frame_config->ref_idx[i] = 0;
983 ref_frame_config->refresh[0] = 1;
984 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
985 } else if (layer_id->spatial_layer_id == 1) {
986 // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 1
987 // and GOLDEN (and all other refs) to slot 0.
988 // Update slot 1 (LAST).
989 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
990 ref_frame_config->ref_idx[i] = 0;
991 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
992 ref_frame_config->refresh[1] = 1;
993 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
994 ref_frame_config->reference[SVC_GOLDEN_FRAME] = 1;
995 } else if (layer_id->spatial_layer_id == 2) {
996 // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 2
997 // and GOLDEN (and all other refs) to slot 1.
998 // Update slot 2 (LAST).
999 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1000 ref_frame_config->ref_idx[i] = 1;
1001 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 2;
1002 ref_frame_config->refresh[2] = 1;
1003 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
1004 ref_frame_config->reference[SVC_GOLDEN_FRAME] = 1;
1005 // For 3 spatial layer case: allow for top spatial layer to use
1006 // additional temporal reference. Update every 10 frames.
1007 if (enable_longterm_temporal_ref) {
1008 ref_frame_config->ref_idx[SVC_ALTREF_FRAME] = REF_FRAMES - 1;
1009 ref_frame_config->reference[SVC_ALTREF_FRAME] = 1;
1010 if (base_count % 10 == 0)
1011 ref_frame_config->refresh[REF_FRAMES - 1] = 1;
1012 }
1013 }
1014 break;
1015 case 7:
1016 // 2 spatial and 3 temporal layer.
1017 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
1018 if (superframe_cnt % 4 == 0) {
1019 // Base temporal layer
1020 layer_id->temporal_layer_id = 0;
1021 if (layer_id->spatial_layer_id == 0) {
1022 // Reference LAST, update LAST
1023 // Set all buffer_idx to 0
1024 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1025 ref_frame_config->ref_idx[i] = 0;
1026 ref_frame_config->refresh[0] = 1;
1027 } else if (layer_id->spatial_layer_id == 1) {
1028 // Reference LAST and GOLDEN.
1029 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1030 ref_frame_config->ref_idx[i] = 0;
1031 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
1032 ref_frame_config->refresh[1] = 1;
1033 }
1034 } else if ((superframe_cnt - 1) % 4 == 0) {
1035 // First top temporal enhancement layer.
1036 layer_id->temporal_layer_id = 2;
1037 if (layer_id->spatial_layer_id == 0) {
1038 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1039 ref_frame_config->ref_idx[i] = 0;
1040 ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 3;
1041 ref_frame_config->refresh[3] = 1;
1042 } else if (layer_id->spatial_layer_id == 1) {
1043 // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 1,
1044 // GOLDEN (and all other refs) to slot 3.
1045 // No update.
1046 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1047 ref_frame_config->ref_idx[i] = 3;
1048 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
1049 }
1050 } else if ((superframe_cnt - 2) % 4 == 0) {
1051 // Middle temporal enhancement layer.
1052 layer_id->temporal_layer_id = 1;
1053 if (layer_id->spatial_layer_id == 0) {
1054 // Reference LAST.
1055 // Set all buffer_idx to 0.
1056 // Set GOLDEN to slot 5 and update slot 5.
1057 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1058 ref_frame_config->ref_idx[i] = 0;
1059 ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 5 - shift;
1060 ref_frame_config->refresh[5 - shift] = 1;
1061 } else if (layer_id->spatial_layer_id == 1) {
1062 // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 1,
1063 // GOLDEN (and all other refs) to slot 5.
1064 // Set LAST3 to slot 6 and update slot 6.
1065 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1066 ref_frame_config->ref_idx[i] = 5 - shift;
1067 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
1068 ref_frame_config->ref_idx[SVC_LAST3_FRAME] = 6 - shift;
1069 ref_frame_config->refresh[6 - shift] = 1;
1070 }
1071 } else if ((superframe_cnt - 3) % 4 == 0) {
1072 // Second top temporal enhancement layer.
1073 layer_id->temporal_layer_id = 2;
1074 if (layer_id->spatial_layer_id == 0) {
1075 // Set LAST to slot 5 and reference LAST.
1076 // Set GOLDEN to slot 3 and update slot 3.
1077 // Set all other buffer_idx to 0.
1078 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1079 ref_frame_config->ref_idx[i] = 0;
1080 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 5 - shift;
1081 ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 3;
1082 ref_frame_config->refresh[3] = 1;
1083 } else if (layer_id->spatial_layer_id == 1) {
1084 // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 6,
1085 // GOLDEN to slot 3. No update.
1086 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1087 ref_frame_config->ref_idx[i] = 0;
1088 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 6 - shift;
1089 ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 3;
1090 }
1091 }
1092 break;
1093 case 8:
1094 // 3 spatial and 3 temporal layer.
1095 // Same as case 9 but overalap in the buffer slot updates.
1096 // (shift = 2). The slots 3 and 4 updated by first TL2 are
1097 // reused for update in TL1 superframe.
1098 // Note for this case, frame order hint must be disabled for
1099 // lower resolutios (operating points > 0) to be decoedable.
1100 case 9:
1101 // 3 spatial and 3 temporal layer.
1102 // No overlap in buffer updates between TL2 and TL1.
1103 // TL2 updates slot 3 and 4, TL1 updates 5, 6, 7.
1104 // Set the references via the svc_ref_frame_config control.
1105 // Always reference LAST.
1106 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
1107 if (superframe_cnt % 4 == 0) {
1108 // Base temporal layer.
1109 layer_id->temporal_layer_id = 0;
1110 if (layer_id->spatial_layer_id == 0) {
1111 // Reference LAST, update LAST.
1112 // Set all buffer_idx to 0.
1113 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1114 ref_frame_config->ref_idx[i] = 0;
1115 ref_frame_config->refresh[0] = 1;
1116 } else if (layer_id->spatial_layer_id == 1) {
1117 // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 1,
1118 // GOLDEN (and all other refs) to slot 0.
1119 // Update slot 1 (LAST).
1120 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1121 ref_frame_config->ref_idx[i] = 0;
1122 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
1123 ref_frame_config->refresh[1] = 1;
1124 } else if (layer_id->spatial_layer_id == 2) {
1125 // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 2,
1126 // GOLDEN (and all other refs) to slot 1.
1127 // Update slot 2 (LAST).
1128 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1129 ref_frame_config->ref_idx[i] = 1;
1130 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 2;
1131 ref_frame_config->refresh[2] = 1;
1132 }
1133 } else if ((superframe_cnt - 1) % 4 == 0) {
1134 // First top temporal enhancement layer.
1135 layer_id->temporal_layer_id = 2;
1136 if (layer_id->spatial_layer_id == 0) {
1137 // Reference LAST (slot 0).
1138 // Set GOLDEN to slot 3 and update slot 3.
1139 // Set all other buffer_idx to slot 0.
1140 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1141 ref_frame_config->ref_idx[i] = 0;
1142 ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 3;
1143 ref_frame_config->refresh[3] = 1;
1144 } else if (layer_id->spatial_layer_id == 1) {
1145 // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 1,
1146 // GOLDEN (and all other refs) to slot 3.
1147 // Set LAST2 to slot 4 and Update slot 4.
1148 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1149 ref_frame_config->ref_idx[i] = 3;
1150 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
1151 ref_frame_config->ref_idx[SVC_LAST2_FRAME] = 4;
1152 ref_frame_config->refresh[4] = 1;
1153 } else if (layer_id->spatial_layer_id == 2) {
1154 // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 2,
1155 // GOLDEN (and all other refs) to slot 4.
1156 // No update.
1157 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1158 ref_frame_config->ref_idx[i] = 4;
1159 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 2;
1160 }
1161 } else if ((superframe_cnt - 2) % 4 == 0) {
1162 // Middle temporal enhancement layer.
1163 layer_id->temporal_layer_id = 1;
1164 if (layer_id->spatial_layer_id == 0) {
1165 // Reference LAST.
1166 // Set all buffer_idx to 0.
1167 // Set GOLDEN to slot 5 and update slot 5.
1168 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1169 ref_frame_config->ref_idx[i] = 0;
1170 ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 5 - shift;
1171 ref_frame_config->refresh[5 - shift] = 1;
1172 } else if (layer_id->spatial_layer_id == 1) {
1173 // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 1,
1174 // GOLDEN (and all other refs) to slot 5.
1175 // Set LAST3 to slot 6 and update slot 6.
1176 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1177 ref_frame_config->ref_idx[i] = 5 - shift;
1178 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
1179 ref_frame_config->ref_idx[SVC_LAST3_FRAME] = 6 - shift;
1180 ref_frame_config->refresh[6 - shift] = 1;
1181 } else if (layer_id->spatial_layer_id == 2) {
1182 // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 2,
1183 // GOLDEN (and all other refs) to slot 6.
1184 // Set LAST3 to slot 7 and update slot 7.
1185 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1186 ref_frame_config->ref_idx[i] = 6 - shift;
1187 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 2;
1188 ref_frame_config->ref_idx[SVC_LAST3_FRAME] = 7 - shift;
1189 ref_frame_config->refresh[7 - shift] = 1;
1190 }
1191 } else if ((superframe_cnt - 3) % 4 == 0) {
1192 // Second top temporal enhancement layer.
1193 layer_id->temporal_layer_id = 2;
1194 if (layer_id->spatial_layer_id == 0) {
1195 // Set LAST to slot 5 and reference LAST.
1196 // Set GOLDEN to slot 3 and update slot 3.
1197 // Set all other buffer_idx to 0.
1198 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1199 ref_frame_config->ref_idx[i] = 0;
1200 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 5 - shift;
1201 ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 3;
1202 ref_frame_config->refresh[3] = 1;
1203 } else if (layer_id->spatial_layer_id == 1) {
1204 // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 6,
1205 // GOLDEN to slot 3. Set LAST2 to slot 4 and update slot 4.
1206 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1207 ref_frame_config->ref_idx[i] = 0;
1208 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 6 - shift;
1209 ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 3;
1210 ref_frame_config->ref_idx[SVC_LAST2_FRAME] = 4;
1211 ref_frame_config->refresh[4] = 1;
1212 } else if (layer_id->spatial_layer_id == 2) {
1213 // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 7,
1214 // GOLDEN to slot 4. No update.
1215 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1216 ref_frame_config->ref_idx[i] = 0;
1217 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 7 - shift;
1218 ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 4;
1219 }
1220 }
1221 break;
1222 case 11:
1223 // Simulcast mode for 3 spatial and 3 temporal layers.
1224 // No inter-layer predicton, only prediction is temporal and single
1225 // reference (LAST).
1226 // No overlap in buffer slots between spatial layers. So for example,
1227 // SL0 only uses slots 0 and 1.
1228 // SL1 only uses slots 2 and 3.
1229 // SL2 only uses slots 4 and 5.
1230 // All 7 references for each inter-frame must only access buffer slots
1231 // for that spatial layer.
1232 // On key (super)frames: SL1 and SL2 must have no references set
1233 // and must refresh all the slots for that layer only (so 2 and 3
1234 // for SL1, 4 and 5 for SL2). The base SL0 will be labelled internally
1235 // as a Key frame (refresh all slots). SL1/SL2 will be labelled
1236 // internally as Intra-only frames that allow that stream to be decoded.
1237 // These conditions will allow for each spatial stream to be
1238 // independently decodeable.
1239
1240 // Initialize all references to 0 (don't use reference).
1241 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1242 ref_frame_config->reference[i] = 0;
1243 // Initialize as no refresh/update for all slots.
1244 for (i = 0; i < REF_FRAMES; i++) ref_frame_config->refresh[i] = 0;
1245 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1246 ref_frame_config->ref_idx[i] = 0;
1247
1248 if (is_key_frame) {
1249 if (layer_id->spatial_layer_id == 0) {
1250 // Assign LAST/GOLDEN to slot 0/1.
1251 // Refesh slots 0 and 1 for SL0.
1252 // SL0: this will get set to KEY frame internally.
1253 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 0;
1254 ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 1;
1255 ref_frame_config->refresh[0] = 1;
1256 ref_frame_config->refresh[1] = 1;
1257 } else if (layer_id->spatial_layer_id == 1) {
1258 // Assign LAST/GOLDEN to slot 2/3.
1259 // Refesh slots 2 and 3 for SL1.
1260 // This will get set to Intra-only frame internally.
1261 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 2;
1262 ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 3;
1263 ref_frame_config->refresh[2] = 1;
1264 ref_frame_config->refresh[3] = 1;
1265 } else if (layer_id->spatial_layer_id == 2) {
1266 // Assign LAST/GOLDEN to slot 4/5.
1267 // Refresh slots 4 and 5 for SL2.
1268 // This will get set to Intra-only frame internally.
1269 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 4;
1270 ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 5;
1271 ref_frame_config->refresh[4] = 1;
1272 ref_frame_config->refresh[5] = 1;
1273 }
1274 } else if (superframe_cnt % 4 == 0) {
1275 // Base temporal layer: TL0
1276 layer_id->temporal_layer_id = 0;
1277 if (layer_id->spatial_layer_id == 0) { // SL0
1278 // Reference LAST. Assign all references to either slot
1279 // 0 or 1. Here we assign LAST to slot 0, all others to 1.
1280 // Update slot 0 (LAST).
1281 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
1282 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1283 ref_frame_config->ref_idx[i] = 1;
1284 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 0;
1285 ref_frame_config->refresh[0] = 1;
1286 } else if (layer_id->spatial_layer_id == 1) { // SL1
1287 // Reference LAST. Assign all references to either slot
1288 // 2 or 3. Here we assign LAST to slot 2, all others to 3.
1289 // Update slot 2 (LAST).
1290 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
1291 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1292 ref_frame_config->ref_idx[i] = 3;
1293 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 2;
1294 ref_frame_config->refresh[2] = 1;
1295 } else if (layer_id->spatial_layer_id == 2) { // SL2
1296 // Reference LAST. Assign all references to either slot
1297 // 4 or 5. Here we assign LAST to slot 4, all others to 5.
1298 // Update slot 4 (LAST).
1299 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
1300 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1301 ref_frame_config->ref_idx[i] = 5;
1302 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 4;
1303 ref_frame_config->refresh[4] = 1;
1304 }
1305 } else if ((superframe_cnt - 1) % 4 == 0) {
1306 // First top temporal enhancement layer: TL2
1307 layer_id->temporal_layer_id = 2;
1308 if (layer_id->spatial_layer_id == 0) { // SL0
1309 // Reference LAST (slot 0). Assign other references to slot 1.
1310 // No update/refresh on any slots.
1311 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
1312 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1313 ref_frame_config->ref_idx[i] = 1;
1314 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 0;
1315 } else if (layer_id->spatial_layer_id == 1) { // SL1
1316 // Reference LAST (slot 2). Assign other references to slot 3.
1317 // No update/refresh on any slots.
1318 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
1319 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1320 ref_frame_config->ref_idx[i] = 3;
1321 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 2;
1322 } else if (layer_id->spatial_layer_id == 2) { // SL2
1323 // Reference LAST (slot 4). Assign other references to slot 4.
1324 // No update/refresh on any slots.
1325 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
1326 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1327 ref_frame_config->ref_idx[i] = 5;
1328 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 4;
1329 }
1330 } else if ((superframe_cnt - 2) % 4 == 0) {
1331 // Middle temporal enhancement layer: TL1
1332 layer_id->temporal_layer_id = 1;
1333 if (layer_id->spatial_layer_id == 0) { // SL0
1334 // Reference LAST (slot 0).
1335 // Set GOLDEN to slot 1 and update slot 1.
1336 // This will be used as reference for next TL2.
1337 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
1338 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1339 ref_frame_config->ref_idx[i] = 1;
1340 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 0;
1341 ref_frame_config->refresh[1] = 1;
1342 } else if (layer_id->spatial_layer_id == 1) { // SL1
1343 // Reference LAST (slot 2).
1344 // Set GOLDEN to slot 3 and update slot 3.
1345 // This will be used as reference for next TL2.
1346 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
1347 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1348 ref_frame_config->ref_idx[i] = 3;
1349 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 2;
1350 ref_frame_config->refresh[3] = 1;
1351 } else if (layer_id->spatial_layer_id == 2) { // SL2
1352 // Reference LAST (slot 4).
1353 // Set GOLDEN to slot 5 and update slot 5.
1354 // This will be used as reference for next TL2.
1355 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
1356 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1357 ref_frame_config->ref_idx[i] = 5;
1358 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 4;
1359 ref_frame_config->refresh[5] = 1;
1360 }
1361 } else if ((superframe_cnt - 3) % 4 == 0) {
1362 // Second top temporal enhancement layer: TL2
1363 layer_id->temporal_layer_id = 2;
1364 if (layer_id->spatial_layer_id == 0) { // SL0
1365 // Reference LAST (slot 1). Assign other references to slot 0.
1366 // No update/refresh on any slots.
1367 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
1368 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1369 ref_frame_config->ref_idx[i] = 0;
1370 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
1371 } else if (layer_id->spatial_layer_id == 1) { // SL1
1372 // Reference LAST (slot 3). Assign other references to slot 2.
1373 // No update/refresh on any slots.
1374 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
1375 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1376 ref_frame_config->ref_idx[i] = 2;
1377 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 3;
1378 } else if (layer_id->spatial_layer_id == 2) { // SL2
1379 // Reference LAST (slot 5). Assign other references to slot 4.
1380 // No update/refresh on any slots.
1381 ref_frame_config->reference[SVC_LAST_FRAME] = 1;
1382 for (i = 0; i < INTER_REFS_PER_FRAME; i++)
1383 ref_frame_config->ref_idx[i] = 4;
1384 ref_frame_config->ref_idx[SVC_LAST_FRAME] = 5;
1385 }
1386 }
1387 if (!simulcast_mode && layer_id->spatial_layer_id > 0) {
1388 // Always reference GOLDEN (inter-layer prediction).
1389 ref_frame_config->reference[SVC_GOLDEN_FRAME] = 1;
1390 if (ksvc_mode) {
1391 // KSVC: only keep the inter-layer reference (GOLDEN) for
1392 // superframes whose base is key.
1393 if (!is_key_frame) ref_frame_config->reference[SVC_GOLDEN_FRAME] = 0;
1394 }
1395 if (is_key_frame && layer_id->spatial_layer_id > 1) {
1396 // On superframes whose base is key: remove LAST to avoid prediction
1397 // off layer two levels below.
1398 ref_frame_config->reference[SVC_LAST_FRAME] = 0;
1399 }
1400 }
1401 // For 3 spatial layer case 8 (where there is free buffer slot):
1402 // allow for top spatial layer to use additional temporal reference.
1403 // Additional reference is only updated on base temporal layer, every
1404 // 10 TL0 frames here.
1405 if (!simulcast_mode && enable_longterm_temporal_ref &&
1406 layer_id->spatial_layer_id == 2 && layering_mode == 8) {
1407 ref_frame_config->ref_idx[SVC_ALTREF_FRAME] = REF_FRAMES - 1;
1408 if (!is_key_frame) ref_frame_config->reference[SVC_ALTREF_FRAME] = 1;
1409 if (base_count % 10 == 0 && layer_id->temporal_layer_id == 0)
1410 ref_frame_config->refresh[REF_FRAMES - 1] = 1;
1411 }
1412 break;
1413 default: assert(0); die("Error: Unsupported temporal layering mode!\n");
1414 }
1415 for (i = 0; i < REF_FRAMES; i++) {
1416 if (ref_frame_config->refresh[i] == 1) {
1417 *reference_updated = 1;
1418 break;
1419 }
1420 }
1421}
1422
1423static void write_literal(struct aom_write_bit_buffer *wb, uint32_t data,
1424 uint8_t bits, uint32_t offset = 0) {
1425 if (bits > 32) {
1426 die("Invalid bits value %d > 32\n", bits);
1427 }
1428 const uint32_t max = static_cast<uint32_t>(((uint64_t)1 << bits) - 1);
1429 if (data < offset || (data - offset) > max) {
1430 die("Invalid data, value %u out of range [%u, %" PRIu64 "]\n", data, offset,
1431 (uint64_t)max + offset);
1432 }
1433 aom_wb_write_unsigned_literal(wb, data - offset, bits);
1434}
1435
1436static void write_depth_representation_element(
1437 struct aom_write_bit_buffer *buffer,
1438 const std::pair<libaom_examples::DepthRepresentationElement, bool>
1439 &element) {
1440 if (!element.second) {
1441 return;
1442 }
1443 write_literal(buffer, element.first.sign_flag, 1);
1444 write_literal(buffer, element.first.exponent, 7);
1445 if (element.first.mantissa_len == 0 || element.first.mantissa_len > 32) {
1446 die("Invalid mantissan_len %d\n", element.first.mantissa_len);
1447 }
1448 write_literal(buffer, element.first.mantissa_len - 1, 5);
1449 write_literal(buffer, element.first.mantissa, element.first.mantissa_len);
1450}
1451
1452static void write_color_properties(
1453 struct aom_write_bit_buffer *buffer,
1454 const std::pair<libaom_examples::ColorProperties, bool> &color_properties) {
1455 write_literal(buffer, color_properties.second, 1);
1456 if (color_properties.second) {
1457 write_literal(buffer, color_properties.first.color_range, 1);
1458 write_literal(buffer, color_properties.first.color_primaries, 8);
1459 write_literal(buffer, color_properties.first.transfer_characteristics, 8);
1460 write_literal(buffer, color_properties.first.matrix_coefficients, 8);
1461 } else {
1462 write_literal(buffer, 0, 1); // reserved_1bit
1463 }
1464}
1465
1466static void write_alpha_information(
1467 struct aom_write_bit_buffer *buffer,
1468 const libaom_examples::AlphaInformation &alpha_info) {
1469 write_literal(buffer, alpha_info.alpha_use_idc, 2);
1470 write_literal(buffer, alpha_info.alpha_simple_flag, 1);
1471 if (!alpha_info.alpha_simple_flag) {
1472 write_literal(buffer, alpha_info.alpha_bit_depth, 3, /*offset=*/8);
1473 write_literal(buffer, alpha_info.alpha_clip_idc, 2);
1474 write_literal(buffer, alpha_info.alpha_incr_flag, 1);
1475 write_literal(buffer, alpha_info.alpha_transparent_value,
1476 alpha_info.alpha_bit_depth + 1);
1477 write_literal(buffer, alpha_info.alpha_opaque_value,
1478 alpha_info.alpha_bit_depth + 1);
1479 if (buffer->bit_offset % 8 != 0) {
1480 // ai_byte_alignment_bits
1481 write_literal(buffer, 0, 8 - (buffer->bit_offset % 8));
1482 }
1483 assert(buffer->bit_offset % 8 == 0);
1484
1485 write_literal(buffer, 0, 6); // ai_reserved_6bits
1486 write_color_properties(buffer, alpha_info.alpha_color_description);
1487 } else {
1488 write_literal(buffer, 0, 5); // ai_reserved_5bits
1489 }
1490}
1491
1492static void write_depth_information(
1493 struct aom_write_bit_buffer *buffer,
1494 const libaom_examples::DepthInformation &depth_info) {
1495 write_literal(buffer, depth_info.z_near.second, 1);
1496 write_literal(buffer, depth_info.z_far.second, 1);
1497 write_literal(buffer, depth_info.d_min.second, 1);
1498 write_literal(buffer, depth_info.d_max.second, 1);
1499 write_literal(buffer, depth_info.depth_representation_type, 4);
1500 if (depth_info.d_min.second || depth_info.d_max.second) {
1501 write_literal(buffer, depth_info.disparity_ref_view_id, 2);
1502 }
1503 write_depth_representation_element(buffer, depth_info.z_near);
1504 write_depth_representation_element(buffer, depth_info.z_far);
1505 write_depth_representation_element(buffer, depth_info.d_min);
1506 write_depth_representation_element(buffer, depth_info.d_max);
1507 if (buffer->bit_offset % 8 != 0) {
1508 write_literal(buffer, 0, 8 - (buffer->bit_offset % 8));
1509 }
1510}
1511
1512static void add_multilayer_metadata(
1513 aom_image_t *frame, const libaom_examples::MultilayerMetadata &multilayer,
1514 int frame_idx, int spatial_id) {
1515 // Large enough buffer for the multilayer metadata.
1516 // Each layer's metadata is less than 100 bytes and there are at most 4
1517 // layers.
1518 std::vector<uint8_t> data(1024);
1519 struct aom_write_bit_buffer buffer = { data.data(), 0 };
1520
1521 write_literal(&buffer, multilayer.use_case, 6);
1522 if (multilayer.layers.empty()) {
1523 die("Invalid multilayer metadata, no layers found\n");
1524 } else if (multilayer.layers.size() > MAX_NUM_SPATIAL_LAYERS) {
1525 die("Invalid multilayer metadata, too many layers (max is %d)\n",
1526 MAX_NUM_SPATIAL_LAYERS);
1527 }
1528 write_literal(&buffer, (int)multilayer.layers.size() - 1, 2);
1529 assert(buffer.bit_offset % 8 == 0);
1530 for (size_t i = 0; i < multilayer.layers.size(); ++i) {
1531 const libaom_examples::LayerMetadata &layer = multilayer.layers[i];
1532 // Alpha info with segmentation with labels can be up to about 66k bytes,
1533 // which requires 3 bytes to encode in leb128.
1534 const int bytes_reserved_for_size = 3;
1535 // Placeholder for layer_metadata_size which will be written later.
1536 write_literal(&buffer, 0, bytes_reserved_for_size * 8);
1537 const uint32_t metadata_start = buffer.bit_offset;
1538 write_literal(&buffer, (int)i, 2); // ml_spatial_id
1539 write_literal(&buffer, layer.layer_type, 5);
1540 write_literal(&buffer, layer.luma_plane_only_flag, 1);
1541 write_literal(&buffer, layer.layer_view_type, 3);
1542 write_literal(&buffer, layer.group_id, 2);
1543 write_literal(&buffer, layer.layer_dependency_idc, 3);
1544 write_literal(&buffer, layer.layer_metadata_scope, 2);
1545 write_literal(&buffer, 0, 4); // ml_reserved_4bits
1546
1547 if (i > 0) {
1548 write_color_properties(&buffer, layer.layer_color_description);
1549 } else {
1550 write_literal(&buffer, 0, 2); // ml_reserved_2bits
1551 }
1552 assert(buffer.bit_offset % 8 == 0);
1553
1554 if (layer.layer_type == libaom_examples::MULTILAYER_LAYER_TYPE_ALPHA &&
1555 layer.layer_metadata_scope >= libaom_examples::SCOPE_GLOBAL) {
1556 write_alpha_information(&buffer, layer.alpha);
1557 assert(buffer.bit_offset % 8 == 0);
1558 } else if (layer.layer_type ==
1559 libaom_examples::MULTILAYER_LAYER_TYPE_DEPTH &&
1560 layer.layer_metadata_scope >= libaom_examples::SCOPE_GLOBAL) {
1561 write_depth_information(&buffer, layer.depth);
1562 assert(buffer.bit_offset % 8 == 0);
1563 }
1564
1565 assert(buffer.bit_offset % 8 == 0);
1566
1567 const int metadata_size_bytes = (buffer.bit_offset - metadata_start) / 8;
1568 const uint8_t size_pos = metadata_start / 8 - bytes_reserved_for_size;
1569 size_t coded_size;
1570 if (aom_uleb_encode_fixed_size(metadata_size_bytes, bytes_reserved_for_size,
1571 bytes_reserved_for_size,
1572 &buffer.bit_buffer[size_pos], &coded_size)) {
1573 // Need to increase bytes_reserved_for_size in the code above.
1574 die("Error: Failed to write metadata size\n");
1575 }
1576 }
1577 assert(buffer.bit_offset % 8 == 0);
1578 if (aom_img_add_metadata(frame, 33 /*METADATA_TYPE_MULTILAYER*/,
1579 buffer.bit_buffer, buffer.bit_offset / 8,
1581 die("Error: Failed to add metadata\n");
1582 }
1583
1584 if ((int)multilayer.layers.size() > spatial_id) {
1585 const libaom_examples::LayerMetadata &layer = multilayer.layers[spatial_id];
1586 for (const libaom_examples::FrameLocalMetadata &local_metadata :
1587 layer.local_metadata) {
1588 if (local_metadata.frame_idx == frame_idx) {
1589 if (layer.layer_type == libaom_examples::MULTILAYER_LAYER_TYPE_ALPHA) {
1590 buffer = { data.data(), 0 };
1591 write_alpha_information(&buffer, local_metadata.alpha);
1592 if (aom_img_add_metadata(frame,
1593 34 /*METADATA_TYPE_ALPHA_INFORMATION*/,
1594 buffer.bit_buffer, buffer.bit_offset / 8,
1596 die("Error: Failed to add metadata\n");
1597 }
1598 } else if (layer.layer_type ==
1599 libaom_examples::MULTILAYER_LAYER_TYPE_DEPTH) {
1600 buffer = { data.data(), 0 };
1601 write_depth_information(&buffer, local_metadata.depth);
1602 if (aom_img_add_metadata(frame,
1603 35 /*METADATA_TYPE_DEPTH_INFORMATION*/,
1604 buffer.bit_buffer, buffer.bit_offset / 8,
1606 die("Error: Failed to add metadata\n");
1607 }
1608 }
1609 break;
1610 }
1611 }
1612 }
1613}
1614
1615#if CONFIG_AV1_DECODER
1616// Returns whether there is a mismatch between the encoder's new frame and the
1617// decoder's new frame.
1618static int test_decode(aom_codec_ctx_t *encoder, aom_codec_ctx_t *decoder,
1619 const int frames_out) {
1620 aom_image_t enc_img, dec_img;
1621 int mismatch = 0;
1622
1623 /* Get the internal new frame */
1626
1627#if CONFIG_AV1_HIGHBITDEPTH
1628 if ((enc_img.fmt & AOM_IMG_FMT_HIGHBITDEPTH) !=
1629 (dec_img.fmt & AOM_IMG_FMT_HIGHBITDEPTH)) {
1630 if (enc_img.fmt & AOM_IMG_FMT_HIGHBITDEPTH) {
1631 aom_image_t enc_hbd_img;
1633 &enc_hbd_img,
1634 static_cast<aom_img_fmt_t>(enc_img.fmt - AOM_IMG_FMT_HIGHBITDEPTH),
1635 enc_img.d_w, enc_img.d_h, 16);
1636 aom_img_truncate_16_to_8(&enc_hbd_img, &enc_img);
1637 enc_img = enc_hbd_img;
1638 }
1639 if (dec_img.fmt & AOM_IMG_FMT_HIGHBITDEPTH) {
1640 aom_image_t dec_hbd_img;
1642 &dec_hbd_img,
1643 static_cast<aom_img_fmt_t>(dec_img.fmt - AOM_IMG_FMT_HIGHBITDEPTH),
1644 dec_img.d_w, dec_img.d_h, 16);
1645 aom_img_truncate_16_to_8(&dec_hbd_img, &dec_img);
1646 dec_img = dec_hbd_img;
1647 }
1648 }
1649#endif
1650
1651 if (!aom_compare_img(&enc_img, &dec_img)) {
1652 int y[4], u[4], v[4];
1653#if CONFIG_AV1_HIGHBITDEPTH
1654 if (enc_img.fmt & AOM_IMG_FMT_HIGHBITDEPTH) {
1655 aom_find_mismatch_high(&enc_img, &dec_img, y, u, v);
1656 } else {
1657 aom_find_mismatch(&enc_img, &dec_img, y, u, v);
1658 }
1659#else
1660 aom_find_mismatch(&enc_img, &dec_img, y, u, v);
1661#endif
1662 fprintf(stderr,
1663 "Encode/decode mismatch on frame %d at"
1664 " Y[%d, %d] {%d/%d},"
1665 " U[%d, %d] {%d/%d},"
1666 " V[%d, %d] {%d/%d}\n",
1667 frames_out, y[0], y[1], y[2], y[3], u[0], u[1], u[2], u[3], v[0],
1668 v[1], v[2], v[3]);
1669 mismatch = 1;
1670 }
1671
1672 aom_img_free(&enc_img);
1673 aom_img_free(&dec_img);
1674 return mismatch;
1675}
1676#endif // CONFIG_AV1_DECODER
1677
1678struct psnr_stats {
1679 // The second element of these arrays is reserved for high bitdepth.
1680 uint64_t psnr_sse_total[2];
1681 uint64_t psnr_samples_total[2];
1682 double psnr_totals[2][4];
1683 int psnr_count[2];
1684};
1685
1686static void show_psnr(struct psnr_stats *psnr_stream, double peak) {
1687 double ovpsnr;
1688
1689 if (!psnr_stream->psnr_count[0]) return;
1690
1691 fprintf(stderr, "\nPSNR (Overall/Avg/Y/U/V)");
1692 ovpsnr = sse_to_psnr((double)psnr_stream->psnr_samples_total[0], peak,
1693 (double)psnr_stream->psnr_sse_total[0]);
1694 fprintf(stderr, " %.3f", ovpsnr);
1695
1696 for (int i = 0; i < 4; i++) {
1697 fprintf(stderr, " %.3f",
1698 psnr_stream->psnr_totals[0][i] / psnr_stream->psnr_count[0]);
1699 }
1700 fprintf(stderr, "\n");
1701}
1702
1703static aom::AV1RateControlRtcConfig create_rtc_rc_config(
1704 const aom_codec_enc_cfg_t &cfg, const AppInput &app_input) {
1705 aom::AV1RateControlRtcConfig rc_cfg;
1706 rc_cfg.width = cfg.g_w;
1707 rc_cfg.height = cfg.g_h;
1708 rc_cfg.max_quantizer = cfg.rc_max_quantizer;
1709 rc_cfg.min_quantizer = cfg.rc_min_quantizer;
1710 rc_cfg.target_bandwidth = cfg.rc_target_bitrate;
1711 rc_cfg.buf_initial_sz = cfg.rc_buf_initial_sz;
1712 rc_cfg.buf_optimal_sz = cfg.rc_buf_optimal_sz;
1713 rc_cfg.buf_sz = cfg.rc_buf_sz;
1714 rc_cfg.overshoot_pct = cfg.rc_overshoot_pct;
1715 rc_cfg.undershoot_pct = cfg.rc_undershoot_pct;
1716 // This is hardcoded as AOME_SET_MAX_INTRA_BITRATE_PCT
1717 rc_cfg.max_intra_bitrate_pct = 300;
1718 rc_cfg.framerate = cfg.g_timebase.den;
1719 // TODO(jianj): Add suppor for SVC.
1720 rc_cfg.ss_number_layers = 1;
1721 rc_cfg.ts_number_layers = 1;
1722 rc_cfg.scaling_factor_num[0] = 1;
1723 rc_cfg.scaling_factor_den[0] = 1;
1724 rc_cfg.layer_target_bitrate[0] = static_cast<int>(rc_cfg.target_bandwidth);
1725 rc_cfg.max_quantizers[0] = rc_cfg.max_quantizer;
1726 rc_cfg.min_quantizers[0] = rc_cfg.min_quantizer;
1727 rc_cfg.aq_mode = app_input.aq_mode;
1728
1729 return rc_cfg;
1730}
1731
1732static int qindex_to_quantizer(int qindex) {
1733 // Table that converts 0-63 range Q values passed in outside to the 0-255
1734 // range Qindex used internally.
1735 static const int quantizer_to_qindex[] = {
1736 0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48,
1737 52, 56, 60, 64, 68, 72, 76, 80, 84, 88, 92, 96, 100,
1738 104, 108, 112, 116, 120, 124, 128, 132, 136, 140, 144, 148, 152,
1739 156, 160, 164, 168, 172, 176, 180, 184, 188, 192, 196, 200, 204,
1740 208, 212, 216, 220, 224, 228, 232, 236, 240, 244, 249, 255,
1741 };
1742 for (int quantizer = 0; quantizer < 64; ++quantizer)
1743 if (quantizer_to_qindex[quantizer] >= qindex) return quantizer;
1744
1745 return 63;
1746}
1747
1748static void set_active_map(const aom_codec_enc_cfg_t *cfg,
1749 aom_codec_ctx_t *codec, int frame_cnt) {
1750 aom_active_map_t map = { 0, 0, 0 };
1751
1752 map.rows = (cfg->g_h + 15) / 16;
1753 map.cols = (cfg->g_w + 15) / 16;
1754
1755 map.active_map = (uint8_t *)malloc(map.rows * map.cols);
1756 if (!map.active_map) die("Failed to allocate active map");
1757
1758 // Example map for testing.
1759 for (unsigned int i = 0; i < map.rows; ++i) {
1760 for (unsigned int j = 0; j < map.cols; ++j) {
1761 int index = map.cols * i + j;
1762 map.active_map[index] = 1;
1763 if (frame_cnt < 300) {
1764 if (i < map.rows / 2 && j < map.cols / 2) map.active_map[index] = 0;
1765 } else if (frame_cnt >= 300) {
1766 if (i < map.rows / 2 && j >= map.cols / 2) map.active_map[index] = 0;
1767 }
1768 }
1769 }
1770
1771 if (aom_codec_control(codec, AOME_SET_ACTIVEMAP, &map))
1772 die_codec(codec, "Failed to set active map");
1773
1774 free(map.active_map);
1775}
1776
1777static void set_roi_map(const aom_codec_enc_cfg_t *cfg, aom_codec_ctx_t *codec,
1778 int roi_feature) {
1780 const int block_size = 4;
1781 roi.rows = (cfg->g_h + block_size - 1) / block_size;
1782 roi.cols = (cfg->g_w + block_size - 1) / block_size;
1783 memset(&roi.skip, 0, sizeof(roi.skip));
1784 memset(&roi.delta_q, 0, sizeof(roi.delta_q));
1785 memset(&roi.delta_lf, 0, sizeof(roi.delta_lf));
1786 memset(roi.ref_frame, -1, sizeof(roi.ref_frame));
1787 // Set ROI map to be 1 (segment #1) in middle square of image,
1788 // 0 elsewhere.
1789 roi.enabled = 1;
1790 roi.roi_map = (uint8_t *)calloc(roi.rows * roi.cols, sizeof(*roi.roi_map));
1791 for (unsigned int i = 0; i < roi.rows; ++i) {
1792 for (unsigned int j = 0; j < roi.cols; ++j) {
1793 const int idx = i * roi.cols + j;
1794 if (i > roi.rows / 4 && i < (3 * roi.rows) / 4 && j > roi.cols / 4 &&
1795 j < (3 * roi.cols) / 4)
1796 roi.roi_map[idx] = 1;
1797 else
1798 roi.roi_map[idx] = 0;
1799 }
1800 }
1801 // Set the ROI feature, on segment #1.
1802 if (roi_feature == kSkip)
1803 roi.skip[1] = 1;
1804 else if (roi_feature == kDeltaQ)
1805 roi.delta_q[1] = -40;
1806 else if (roi_feature == kDeltaLF)
1807 roi.delta_lf[1] = 40;
1808 else if (roi_feature == kReference)
1809 roi.ref_frame[1] = 4; // GOLDEN_FRAME
1810
1811 if (aom_codec_control(codec, AOME_SET_ROI_MAP, &roi))
1812 die_codec(codec, "Failed to set roi map");
1813
1814 free(roi.roi_map);
1815}
1816int main(int argc, const char **argv) {
1817 AppInput app_input;
1818 AvxVideoWriter *outfile[AOM_MAX_LAYERS] = { NULL };
1819 FILE *obu_files[AOM_MAX_LAYERS] = { NULL };
1820 AvxVideoWriter *total_layer_file = NULL;
1821 FILE *total_layer_obu_file = NULL;
1823 int frame_cnt = 0;
1824 aom_image_t raw;
1825 int frame_avail;
1826 int got_data = 0;
1827 int flags = 0;
1828 int i;
1829 int pts = 0; // PTS starts at 0.
1830 int frame_duration = 1; // 1 timebase tick per frame.
1831 aom_svc_layer_id_t layer_id;
1832 aom_svc_params_t svc_params;
1833 aom_svc_ref_frame_config_t ref_frame_config;
1834 aom_svc_ref_frame_comp_pred_t ref_frame_comp_pred;
1835
1836#if CONFIG_INTERNAL_STATS
1837 FILE *stats_file = fopen("opsnr.stt", "a");
1838 if (stats_file == NULL) {
1839 die("Cannot open opsnr.stt\n");
1840 }
1841#endif
1842#if CONFIG_AV1_DECODER
1843 aom_codec_ctx_t decoder;
1844#endif
1845
1846 struct RateControlMetrics rc;
1847 int64_t cx_time = 0;
1848 int64_t cx_time_layer[AOM_MAX_LAYERS]; // max number of layers.
1849 int frame_cnt_layer[AOM_MAX_LAYERS];
1850 double sum_bitrate = 0.0;
1851 double sum_bitrate2 = 0.0;
1852 double framerate = 30.0;
1853 int use_svc_control = 1;
1854 int set_err_resil_frame = 0;
1855 int test_changing_bitrate = 0;
1856 zero(rc.layer_target_bitrate);
1857 memset(&layer_id, 0, sizeof(aom_svc_layer_id_t));
1858 memset(&app_input, 0, sizeof(AppInput));
1859 memset(&svc_params, 0, sizeof(svc_params));
1860
1861 // Flag to test dynamic scaling of source frames for single
1862 // spatial stream, using the scaling_mode control.
1863 const int test_dynamic_scaling_single_layer = 0;
1864
1865 // Flag to test setting speed per layer.
1866 const int test_speed_per_layer = 0;
1867
1868 // Flag for testing active maps.
1869 const int test_active_maps = 0;
1870
1871 // Flag for testing roi map.
1872 const int test_roi_map = 0;
1873
1874 /* Setup default input stream settings */
1875 for (i = 0; i < MAX_NUM_SPATIAL_LAYERS; ++i) {
1876 app_input.input_ctx[i].framerate.numerator = 30;
1877 app_input.input_ctx[i].framerate.denominator = 1;
1878 app_input.input_ctx[i].only_i420 = 0;
1879 app_input.input_ctx[i].bit_depth = AOM_BITS_8;
1880 }
1881 app_input.speed = 7;
1882 exec_name = argv[0];
1883
1884 // start with default encoder configuration
1885#if GOOD_QUALITY
1888#else
1891#endif
1892 if (res != AOM_CODEC_OK) {
1893 die("Failed to get config: %s\n", aom_codec_err_to_string(res));
1894 }
1895
1896#if GOOD_QUALITY
1898#else
1899 // Real time parameters.
1901#endif
1902
1903 cfg.rc_end_usage = AOM_CBR;
1904 cfg.rc_min_quantizer = 2;
1905 cfg.rc_max_quantizer = 52;
1906 cfg.rc_undershoot_pct = 50;
1907 cfg.rc_overshoot_pct = 50;
1908 cfg.rc_buf_initial_sz = 600;
1909 cfg.rc_buf_optimal_sz = 600;
1910 cfg.rc_buf_sz = 1000;
1911 cfg.rc_resize_mode = 0; // Set to RESIZE_DYNAMIC for dynamic resize.
1912 cfg.g_lag_in_frames = 0;
1913 cfg.kf_mode = AOM_KF_AUTO;
1914 cfg.g_w = 0; // Force user to specify width and height for raw input.
1915 cfg.g_h = 0;
1916
1917 parse_command_line(argc, argv, &app_input, &svc_params, &cfg);
1918
1919 int ts_number_layers = svc_params.number_temporal_layers;
1920 int ss_number_layers = svc_params.number_spatial_layers;
1921
1922 unsigned int width = cfg.g_w;
1923 unsigned int height = cfg.g_h;
1924
1925 if (app_input.layering_mode >= 0) {
1926 if (ts_number_layers !=
1927 mode_to_num_temporal_layers[app_input.layering_mode] ||
1928 ss_number_layers !=
1929 mode_to_num_spatial_layers[app_input.layering_mode]) {
1930 die("Number of layers doesn't match layering mode.");
1931 }
1932 }
1933
1934 bool has_non_y4m_input = false;
1935 for (i = 0; i < AOM_MAX_LAYERS; ++i) {
1936 if (app_input.input_ctx[i].file_type != FILE_TYPE_Y4M) {
1937 has_non_y4m_input = true;
1938 break;
1939 }
1940 }
1941 // Y4M reader has its own allocation.
1942 if (has_non_y4m_input) {
1943 if (!aom_img_alloc(&raw, AOM_IMG_FMT_I420, width, height, 32)) {
1944 die("Failed to allocate image (%dx%d)", width, height);
1945 }
1946 }
1947
1949
1950 memcpy(&rc.layer_target_bitrate[0], &svc_params.layer_target_bitrate[0],
1951 sizeof(svc_params.layer_target_bitrate));
1952
1953 unsigned int total_rate = 0;
1954 for (i = 0; i < ss_number_layers; i++) {
1955 total_rate +=
1956 svc_params
1957 .layer_target_bitrate[i * ts_number_layers + ts_number_layers - 1];
1958 }
1959 if (total_rate != cfg.rc_target_bitrate) {
1960 die("Incorrect total target bitrate, expected: %d", total_rate);
1961 }
1962
1963 svc_params.framerate_factor[0] = 1;
1964 if (ts_number_layers == 2) {
1965 svc_params.framerate_factor[0] = 2;
1966 svc_params.framerate_factor[1] = 1;
1967 } else if (ts_number_layers == 3) {
1968 svc_params.framerate_factor[0] = 4;
1969 svc_params.framerate_factor[1] = 2;
1970 svc_params.framerate_factor[2] = 1;
1971 }
1972
1973 libaom_examples::MultilayerMetadata multilayer_metadata;
1974 if (app_input.multilayer_metadata_file != NULL) {
1975 if (!libaom_examples::parse_multilayer_file(
1976 app_input.multilayer_metadata_file, &multilayer_metadata)) {
1977 die("Failed to parse multilayer metadata");
1978 }
1979 libaom_examples::print_multilayer_metadata(multilayer_metadata);
1980 }
1981
1982 framerate = cfg.g_timebase.den / cfg.g_timebase.num;
1983 set_rate_control_metrics(&rc, framerate, ss_number_layers, ts_number_layers);
1984
1985 AvxVideoInfo info;
1986 info.codec_fourcc = get_fourcc_by_aom_encoder(encoder);
1987 info.frame_width = cfg.g_w;
1988 info.frame_height = cfg.g_h;
1989 info.time_base.numerator = cfg.g_timebase.num;
1990 info.time_base.denominator = cfg.g_timebase.den;
1991 // Open an output file for each stream.
1992 for (int sl = 0; sl < ss_number_layers; ++sl) {
1993 for (int tl = 0; tl < ts_number_layers; ++tl) {
1994 i = sl * ts_number_layers + tl;
1995 char file_name[PATH_MAX];
1996 snprintf(file_name, sizeof(file_name), "%s_%d.av1",
1997 app_input.output_filename, i);
1998 if (app_input.output_obu) {
1999 obu_files[i] = fopen(file_name, "wb");
2000 if (!obu_files[i]) die("Failed to open %s for writing", file_name);
2001 } else {
2002 outfile[i] = aom_video_writer_open(file_name, kContainerIVF, &info);
2003 if (!outfile[i]) die("Failed to open %s for writing", file_name);
2004 }
2005 }
2006 }
2007 if (app_input.output_obu) {
2008 total_layer_obu_file = fopen(app_input.output_filename, "wb");
2009 if (!total_layer_obu_file)
2010 die("Failed to open %s for writing", app_input.output_filename);
2011 } else {
2012 total_layer_file =
2013 aom_video_writer_open(app_input.output_filename, kContainerIVF, &info);
2014 if (!total_layer_file)
2015 die("Failed to open %s for writing", app_input.output_filename);
2016 }
2017
2018 // Initialize codec.
2019 aom_codec_ctx_t codec;
2020 aom_codec_flags_t flag = 0;
2022 flag |= app_input.show_psnr ? AOM_CODEC_USE_PSNR : 0;
2023 if (aom_codec_enc_init(&codec, encoder, &cfg, flag))
2024 die_codec(&codec, "Failed to initialize encoder");
2025
2026#if CONFIG_AV1_DECODER
2027 if (app_input.decode) {
2028 if (aom_codec_dec_init(&decoder, get_aom_decoder_by_index(0), NULL, 0))
2029 die_codec(&decoder, "Failed to initialize decoder");
2030 }
2031#endif
2032
2033 aom_codec_control(&codec, AOME_SET_CPUUSED, app_input.speed);
2034 aom_codec_control(&codec, AV1E_SET_AQ_MODE, app_input.aq_mode ? 3 : 0);
2044#if GOOD_QUALITY
2049#else
2054#endif
2056
2057 // Settings to reduce key frame encoding time.
2063
2065
2066 aom_codec_control(&codec, AV1E_SET_TUNE_CONTENT, app_input.tune_content);
2067 if (app_input.tune_content == AOM_CONTENT_SCREEN) {
2069 // INTRABC is currently disabled for rt mode, as it's too slow.
2071 }
2072
2073 if (app_input.use_external_rc) {
2075 }
2076
2078
2081
2083
2084 svc_params.number_spatial_layers = ss_number_layers;
2085 svc_params.number_temporal_layers = ts_number_layers;
2086 for (i = 0; i < ss_number_layers * ts_number_layers; ++i) {
2087 svc_params.max_quantizers[i] = cfg.rc_max_quantizer;
2088 svc_params.min_quantizers[i] = cfg.rc_min_quantizer;
2089 }
2090 // SET QUANTIZER PER LAYER, E.G FOR 2 SPATIAL LAYERS:
2091 // svc_params.max_quantizers[0] = 40;
2092 // svc_params.min_quantizers[0] = 40;
2093 // svc_params.max_quantizers[1] = 50;
2094 // svc_params.min_quantizers[1] = 50;
2095
2096 if (!app_input.scale_factors_explicitly_set) {
2097 for (i = 0; i < ss_number_layers; ++i) {
2098 svc_params.scaling_factor_num[i] = 1;
2099 svc_params.scaling_factor_den[i] = 1;
2100 }
2101 if (ss_number_layers == 2) {
2102 svc_params.scaling_factor_num[0] = 1;
2103 svc_params.scaling_factor_den[0] = 2;
2104 } else if (ss_number_layers == 3) {
2105 svc_params.scaling_factor_num[0] = 1;
2106 svc_params.scaling_factor_den[0] = 4;
2107 svc_params.scaling_factor_num[1] = 1;
2108 svc_params.scaling_factor_den[1] = 2;
2109 }
2110 }
2111 aom_codec_control(&codec, AV1E_SET_SVC_PARAMS, &svc_params);
2112 // TODO(aomedia:3032): Configure KSVC in fixed mode.
2113
2114 // This controls the maximum target size of the key frame.
2115 // For generating smaller key frames, use a smaller max_intra_size_pct
2116 // value, like 100 or 200.
2117 {
2118 const int max_intra_size_pct = 300;
2120 max_intra_size_pct);
2121 }
2122
2123 for (int lx = 0; lx < ts_number_layers * ss_number_layers; lx++) {
2124 cx_time_layer[lx] = 0;
2125 frame_cnt_layer[lx] = 0;
2126 }
2127
2128 std::unique_ptr<aom::AV1RateControlRTC> rc_api;
2129 if (app_input.use_external_rc) {
2130 const aom::AV1RateControlRtcConfig rc_cfg =
2131 create_rtc_rc_config(cfg, app_input);
2132 rc_api = aom::AV1RateControlRTC::Create(rc_cfg);
2133 }
2134
2135 frame_avail = 1;
2136 struct psnr_stats psnr_stream;
2137 memset(&psnr_stream, 0, sizeof(psnr_stream));
2138 while (frame_avail || got_data) {
2139 struct aom_usec_timer timer;
2140 frame_avail = read_frame(&(app_input.input_ctx[0]), &raw);
2141 // Loop over spatial layers.
2142 for (int slx = 0; slx < ss_number_layers; slx++) {
2143 if (slx > 0 && app_input.input_ctx[slx].filename != NULL) {
2144 const int previous_layer_frame_avail = frame_avail;
2145 frame_avail = read_frame(&(app_input.input_ctx[slx]), &raw);
2146 if (previous_layer_frame_avail != frame_avail) {
2147 die("Mismatch in number of frames between spatial layer input files");
2148 }
2149 }
2150
2151 aom_codec_iter_t iter = NULL;
2152 const aom_codec_cx_pkt_t *pkt;
2153 int reference_updated = 0;
2154 int layer = 0;
2155 // Flag for superframe whose base is key.
2156 int is_key_frame = (frame_cnt % cfg.kf_max_dist) == 0;
2157 // For flexible mode:
2158 if (app_input.layering_mode >= 0) {
2159 // Set the reference/update flags, layer_id, and reference_map
2160 // buffer index.
2161 set_layer_pattern(app_input.layering_mode, frame_cnt, &layer_id,
2162 &ref_frame_config, &ref_frame_comp_pred,
2163 &use_svc_control, slx, is_key_frame,
2164 (app_input.layering_mode == 10), app_input.speed,
2165 &reference_updated, test_roi_map);
2166 aom_codec_control(&codec, AV1E_SET_SVC_LAYER_ID, &layer_id);
2167 if (use_svc_control) {
2169 &ref_frame_config);
2171 &ref_frame_comp_pred);
2172 }
2173 if (app_input.multilayer_metadata_file != NULL) {
2174 add_multilayer_metadata(&raw, multilayer_metadata, frame_cnt, slx);
2175 }
2176 // Set the speed per layer.
2177 if (test_speed_per_layer) {
2178 int speed_per_layer = 10;
2179 if (layer_id.spatial_layer_id == 0) {
2180 if (layer_id.temporal_layer_id == 0) speed_per_layer = 6;
2181 if (layer_id.temporal_layer_id == 1) speed_per_layer = 7;
2182 if (layer_id.temporal_layer_id == 2) speed_per_layer = 8;
2183 } else if (layer_id.spatial_layer_id == 1) {
2184 if (layer_id.temporal_layer_id == 0) speed_per_layer = 7;
2185 if (layer_id.temporal_layer_id == 1) speed_per_layer = 8;
2186 if (layer_id.temporal_layer_id == 2) speed_per_layer = 9;
2187 } else if (layer_id.spatial_layer_id == 2) {
2188 if (layer_id.temporal_layer_id == 0) speed_per_layer = 8;
2189 if (layer_id.temporal_layer_id == 1) speed_per_layer = 9;
2190 if (layer_id.temporal_layer_id == 2) speed_per_layer = 10;
2191 }
2192 aom_codec_control(&codec, AOME_SET_CPUUSED, speed_per_layer);
2193 }
2194 } else {
2195 // Only up to 3 temporal layers supported in fixed mode.
2196 // Only need to set spatial and temporal layer_id: reference
2197 // prediction, refresh, and buffer_idx are set internally.
2198 layer_id.spatial_layer_id = slx;
2199 layer_id.temporal_layer_id = 0;
2200 if (ts_number_layers == 2) {
2201 layer_id.temporal_layer_id = (frame_cnt % 2) != 0;
2202 } else if (ts_number_layers == 3) {
2203 if (frame_cnt % 2 != 0)
2204 layer_id.temporal_layer_id = 2;
2205 else if ((frame_cnt > 1) && ((frame_cnt - 2) % 4 == 0))
2206 layer_id.temporal_layer_id = 1;
2207 }
2208 aom_codec_control(&codec, AV1E_SET_SVC_LAYER_ID, &layer_id);
2209 }
2210
2211 if (set_err_resil_frame && cfg.g_error_resilient == 0) {
2212 // Set error_resilient per frame: off/0 for base layer and
2213 // on/1 for enhancement layer frames.
2214 // Note that this is can only be done on the fly/per-frame/layer
2215 // if the config error_resilience is off/0. See the logic for updating
2216 // in set_encoder_config():
2217 // tool_cfg->error_resilient_mode =
2218 // cfg->g_error_resilient | extra_cfg->error_resilient_mode;
2219 const int err_resil_mode =
2220 layer_id.spatial_layer_id > 0 || layer_id.temporal_layer_id > 0;
2222 err_resil_mode);
2223 }
2224
2225 layer = slx * ts_number_layers + layer_id.temporal_layer_id;
2226 if (frame_avail && slx == 0) ++rc.layer_input_frames[layer];
2227
2228 if (test_dynamic_scaling_single_layer) {
2229 // Example to scale source down by 2x2, then 4x4, and then back up to
2230 // 2x2, and then back to original.
2231 int frame_2x2 = 200;
2232 int frame_4x4 = 400;
2233 int frame_2x2up = 600;
2234 int frame_orig = 800;
2235 if (frame_cnt >= frame_2x2 && frame_cnt < frame_4x4) {
2236 // Scale source down by 2x2.
2237 struct aom_scaling_mode mode = { AOME_ONETWO, AOME_ONETWO };
2238 aom_codec_control(&codec, AOME_SET_SCALEMODE, &mode);
2239 } else if (frame_cnt >= frame_4x4 && frame_cnt < frame_2x2up) {
2240 // Scale source down by 4x4.
2241 struct aom_scaling_mode mode = { AOME_ONEFOUR, AOME_ONEFOUR };
2242 aom_codec_control(&codec, AOME_SET_SCALEMODE, &mode);
2243 } else if (frame_cnt >= frame_2x2up && frame_cnt < frame_orig) {
2244 // Source back up to 2x2.
2245 struct aom_scaling_mode mode = { AOME_ONETWO, AOME_ONETWO };
2246 aom_codec_control(&codec, AOME_SET_SCALEMODE, &mode);
2247 } else if (frame_cnt >= frame_orig) {
2248 // Source back up to original resolution (no scaling).
2249 struct aom_scaling_mode mode = { AOME_NORMAL, AOME_NORMAL };
2250 aom_codec_control(&codec, AOME_SET_SCALEMODE, &mode);
2251 }
2252 if (frame_cnt == frame_2x2 || frame_cnt == frame_4x4 ||
2253 frame_cnt == frame_2x2up || frame_cnt == frame_orig) {
2254 // For dynamic resize testing on single layer: refresh all references
2255 // on the resized frame: this is to avoid decode error:
2256 // if resize goes down by >= 4x4 then libaom decoder will throw an
2257 // error that some reference (even though not used) is beyond the
2258 // limit size (must be smaller than 4x4).
2259 for (i = 0; i < REF_FRAMES; i++) ref_frame_config.refresh[i] = 1;
2260 if (use_svc_control) {
2262 &ref_frame_config);
2264 &ref_frame_comp_pred);
2265 }
2266 }
2267 }
2268
2269 // Change target_bitrate every other frame.
2270 if (test_changing_bitrate && frame_cnt % 2 == 0) {
2271 if (frame_cnt < 500)
2272 cfg.rc_target_bitrate += 10;
2273 else
2274 cfg.rc_target_bitrate -= 10;
2275 // Do big increase and decrease.
2276 if (frame_cnt == 100) cfg.rc_target_bitrate <<= 1;
2277 if (frame_cnt == 600) cfg.rc_target_bitrate >>= 1;
2278 if (cfg.rc_target_bitrate < 100) cfg.rc_target_bitrate = 100;
2279 // Call change_config, or bypass with new control.
2280 // res = aom_codec_enc_config_set(&codec, &cfg);
2282 cfg.rc_target_bitrate))
2283 die_codec(&codec, "Failed to SET_BITRATE_ONE_PASS_CBR");
2284 }
2285
2286 if (rc_api) {
2287 aom::AV1FrameParamsRTC frame_params;
2288 // TODO(jianj): Add support for SVC.
2289 frame_params.spatial_layer_id = 0;
2290 frame_params.temporal_layer_id = 0;
2291 frame_params.frame_type =
2292 is_key_frame ? aom::kKeyFrame : aom::kInterFrame;
2293 rc_api->ComputeQP(frame_params);
2294 const int current_qp = rc_api->GetQP();
2296 qindex_to_quantizer(current_qp))) {
2297 die_codec(&codec, "Failed to SET_QUANTIZER_ONE_PASS");
2298 }
2299 }
2300
2301 if (test_active_maps) set_active_map(&cfg, &codec, frame_cnt);
2302
2303 if (test_roi_map) set_roi_map(&cfg, &codec, kDeltaQ);
2304
2305 // Do the layer encode.
2306 aom_usec_timer_start(&timer);
2307 if (aom_codec_encode(&codec, frame_avail ? &raw : NULL, pts, 1, flags))
2308 die_codec(&codec, "Failed to encode frame");
2309 aom_usec_timer_mark(&timer);
2310 cx_time += aom_usec_timer_elapsed(&timer);
2311 cx_time_layer[layer] += aom_usec_timer_elapsed(&timer);
2312 frame_cnt_layer[layer] += 1;
2313
2314 // Get the high motion content flag.
2315 int content_flag = 0;
2317 &content_flag)) {
2318 die_codec(&codec, "Failed to GET_HIGH_MOTION_CONTENT_SCREEN_RTC");
2319 }
2320
2321 got_data = 0;
2322 // For simulcast (mode 11): write out each spatial layer to the file.
2323 int ss_layers_write = (app_input.layering_mode == 11)
2324 ? layer_id.spatial_layer_id + 1
2325 : ss_number_layers;
2326 while ((pkt = aom_codec_get_cx_data(&codec, &iter))) {
2327 switch (pkt->kind) {
2329 for (int sl = layer_id.spatial_layer_id; sl < ss_layers_write;
2330 ++sl) {
2331 for (int tl = layer_id.temporal_layer_id; tl < ts_number_layers;
2332 ++tl) {
2333 int j = sl * ts_number_layers + tl;
2334 if (app_input.output_obu) {
2335 fwrite(pkt->data.frame.buf, 1, pkt->data.frame.sz,
2336 obu_files[j]);
2337 } else {
2338 aom_video_writer_write_frame(
2339 outfile[j],
2340 reinterpret_cast<const uint8_t *>(pkt->data.frame.buf),
2341 pkt->data.frame.sz, pts);
2342 }
2343 if (sl == layer_id.spatial_layer_id)
2344 rc.layer_encoding_bitrate[j] += 8.0 * pkt->data.frame.sz;
2345 }
2346 }
2347 got_data = 1;
2348 // Write everything into the top layer.
2349 if (app_input.output_obu) {
2350 fwrite(pkt->data.frame.buf, 1, pkt->data.frame.sz,
2351 total_layer_obu_file);
2352 } else {
2353 aom_video_writer_write_frame(
2354 total_layer_file,
2355 reinterpret_cast<const uint8_t *>(pkt->data.frame.buf),
2356 pkt->data.frame.sz, pts);
2357 }
2358 // Keep count of rate control stats per layer (for non-key).
2359 if (!(pkt->data.frame.flags & AOM_FRAME_IS_KEY)) {
2360 int j = layer_id.spatial_layer_id * ts_number_layers +
2361 layer_id.temporal_layer_id;
2362 assert(j >= 0);
2363 rc.layer_avg_frame_size[j] += 8.0 * pkt->data.frame.sz;
2364 rc.layer_avg_rate_mismatch[j] +=
2365 fabs(8.0 * pkt->data.frame.sz - rc.layer_pfb[j]) /
2366 rc.layer_pfb[j];
2367 if (slx == 0) ++rc.layer_enc_frames[layer_id.temporal_layer_id];
2368 }
2369
2370 if (rc_api) {
2371 rc_api->PostEncodeUpdate(pkt->data.frame.sz);
2372 }
2373 // Update for short-time encoding bitrate states, for moving window
2374 // of size rc->window, shifted by rc->window / 2.
2375 // Ignore first window segment, due to key frame.
2376 // For spatial layers: only do this for top/highest SL.
2377 if (frame_cnt > rc.window_size && slx == ss_number_layers - 1) {
2378 sum_bitrate += 0.001 * 8.0 * pkt->data.frame.sz * framerate;
2379 rc.window_size = (rc.window_size <= 0) ? 1 : rc.window_size;
2380 if (frame_cnt % rc.window_size == 0) {
2381 rc.window_count += 1;
2382 rc.avg_st_encoding_bitrate += sum_bitrate / rc.window_size;
2383 rc.variance_st_encoding_bitrate +=
2384 (sum_bitrate / rc.window_size) *
2385 (sum_bitrate / rc.window_size);
2386 sum_bitrate = 0.0;
2387 }
2388 }
2389 // Second shifted window.
2390 if (frame_cnt > rc.window_size + rc.window_size / 2 &&
2391 slx == ss_number_layers - 1) {
2392 sum_bitrate2 += 0.001 * 8.0 * pkt->data.frame.sz * framerate;
2393 if (frame_cnt > 2 * rc.window_size &&
2394 frame_cnt % rc.window_size == 0) {
2395 rc.window_count += 1;
2396 rc.avg_st_encoding_bitrate += sum_bitrate2 / rc.window_size;
2397 rc.variance_st_encoding_bitrate +=
2398 (sum_bitrate2 / rc.window_size) *
2399 (sum_bitrate2 / rc.window_size);
2400 sum_bitrate2 = 0.0;
2401 }
2402 }
2403
2404#if CONFIG_AV1_DECODER
2405 if (app_input.decode) {
2406 if (aom_codec_decode(
2407 &decoder,
2408 reinterpret_cast<const uint8_t *>(pkt->data.frame.buf),
2409 pkt->data.frame.sz, NULL))
2410 die_codec(&decoder, "Failed to decode frame");
2411 }
2412#endif
2413
2414 break;
2415 case AOM_CODEC_PSNR_PKT:
2416 if (app_input.show_psnr) {
2417 psnr_stream.psnr_sse_total[0] += pkt->data.psnr.sse[0];
2418 psnr_stream.psnr_samples_total[0] += pkt->data.psnr.samples[0];
2419 for (int plane = 0; plane < 4; plane++) {
2420 psnr_stream.psnr_totals[0][plane] += pkt->data.psnr.psnr[plane];
2421 }
2422 psnr_stream.psnr_count[0]++;
2423 }
2424 break;
2425 default: break;
2426 }
2427 }
2428#if CONFIG_AV1_DECODER
2429 if (got_data && app_input.decode) {
2430 // Don't look for mismatch on non reference frames.
2431 if (reference_updated) {
2432 if (test_decode(&codec, &decoder, frame_cnt)) {
2433#if CONFIG_INTERNAL_STATS
2434 fprintf(stats_file, "First mismatch occurred in frame %d\n",
2435 frame_cnt);
2436 fclose(stats_file);
2437#endif
2438 fatal("Mismatch seen");
2439 }
2440 }
2441 }
2442#endif
2443 } // loop over spatial layers
2444 ++frame_cnt;
2445 pts += frame_duration;
2446 }
2447
2448 for (i = 0; i < MAX_NUM_SPATIAL_LAYERS; ++i) {
2449 if (app_input.input_ctx[i].filename == NULL) {
2450 break;
2451 }
2452 close_input_file(&(app_input.input_ctx[i]));
2453 }
2454 printout_rate_control_summary(&rc, frame_cnt, ss_number_layers,
2455 ts_number_layers);
2456
2457 printf("\n");
2458 for (int slx = 0; slx < ss_number_layers; slx++)
2459 for (int tlx = 0; tlx < ts_number_layers; tlx++) {
2460 int lx = slx * ts_number_layers + tlx;
2461 printf("Per layer encoding time/FPS stats for encoder: %d %d %d %f %f \n",
2462 slx, tlx, frame_cnt_layer[lx],
2463 (float)cx_time_layer[lx] / (double)(frame_cnt_layer[lx] * 1000),
2464 1000000 * (double)frame_cnt_layer[lx] / (double)cx_time_layer[lx]);
2465 }
2466
2467 printf("\n");
2468 printf("Frame cnt and encoding time/FPS stats for encoding: %d %f %f\n",
2469 frame_cnt, 1000 * (float)cx_time / (double)(frame_cnt * 1000000),
2470 1000000 * (double)frame_cnt / (double)cx_time);
2471
2472 if (app_input.show_psnr) {
2473 show_psnr(&psnr_stream, 255.0);
2474 }
2475
2476 if (aom_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy encoder");
2477
2478#if CONFIG_AV1_DECODER
2479 if (app_input.decode) {
2480 if (aom_codec_destroy(&decoder))
2481 die_codec(&decoder, "Failed to destroy decoder");
2482 }
2483#endif
2484
2485#if CONFIG_INTERNAL_STATS
2486 fprintf(stats_file, "No mismatch detected in recon buffers\n");
2487 fclose(stats_file);
2488#endif
2489
2490 // Try to rewrite the output file headers with the actual frame count.
2491 for (i = 0; i < ss_number_layers * ts_number_layers; ++i)
2492 aom_video_writer_close(outfile[i]);
2493 aom_video_writer_close(total_layer_file);
2494
2495 if (has_non_y4m_input) {
2496 aom_img_free(&raw);
2497 }
2498 return EXIT_SUCCESS;
2499}
Describes the decoder algorithm interface to applications.
Describes the encoder algorithm interface to applications.
Describes the aom image descriptor and associated operations.
@ AOM_MIF_KEY_FRAME
Definition aom_image.h:176
@ AOM_MIF_ANY_FRAME_LAYER_SPECIFIC
Definition aom_image.h:183
@ AOM_CSP_UNKNOWN
Definition aom_image.h:143
enum aom_chroma_sample_position aom_chroma_sample_position_t
List of chroma sample positions.
#define AOM_IMG_FMT_HIGHBITDEPTH
Definition aom_image.h:38
aom_image_t * aom_img_alloc(aom_image_t *img, aom_img_fmt_t fmt, unsigned int d_w, unsigned int d_h, unsigned int align)
Open a descriptor, allocating storage for the underlying image.
@ AOM_IMG_FMT_I420
Definition aom_image.h:45
enum aom_img_fmt aom_img_fmt_t
List of supported image formats.
int aom_img_add_metadata(aom_image_t *img, uint32_t type, const uint8_t *data, size_t sz, aom_metadata_insert_flags_t insert_flag)
Add metadata to image.
struct aom_image aom_image_t
Image Descriptor.
void aom_img_free(aom_image_t *img)
Close an image descriptor.
Provides definitions for using AOM or AV1 encoder algorithm within the aom Codec Interface.
#define AOM_MAX_LAYERS
Definition aomcx.h:1777
struct aom_svc_params aom_svc_params_t
Parameter type for SVC.
#define AOM_MAX_TS_LAYERS
Definition aomcx.h:1779
aom_codec_iface_t * aom_codec_av1_cx(void)
The interface to the AV1 encoder.
struct aom_roi_map aom_roi_map_t
aom region of interest map
struct aom_svc_layer_id aom_svc_layer_id_t
Struct for spatial and temporal layer ID.
struct aom_active_map aom_active_map_t
aom active region map
struct aom_svc_ref_frame_comp_pred aom_svc_ref_frame_comp_pred_t
Parameters for setting ref frame compound prediction.
struct aom_svc_ref_frame_config aom_svc_ref_frame_config_t
Parameters for setting ref frame config.
@ AOM_FULL_SUPERFRAME_DROP
Definition aomcx.h:1851
@ AV1E_SET_BITRATE_ONE_PASS_CBR
Codec control to set the target bitrate in kilobits per second, unsigned int parameter....
Definition aomcx.h:1544
@ AV1E_SET_ENABLE_SMOOTH_INTRA
Codec control function to turn on / off smooth intra modes usage, int parameter.
Definition aomcx.h:1081
@ AV1E_SET_ENABLE_TPL_MODEL
Codec control function to enable RDO modulated by frame temporal dependency, unsigned int parameter.
Definition aomcx.h:418
@ AV1E_SET_AQ_MODE
Codec control function to set adaptive quantization mode, unsigned int parameter.
Definition aomcx.h:478
@ AV1E_SET_SVC_LAYER_ID
Codec control function to set the layer id, aom_svc_layer_id_t* parameter.
Definition aomcx.h:1293
@ AV1E_SET_SVC_REF_FRAME_CONFIG
Codec control function to set the reference frame config, aom_svc_ref_frame_config_t* parameter.
Definition aomcx.h:1303
@ AV1E_SET_TUNE_CONTENT
Codec control function to set content type, aom_tune_content parameter.
Definition aomcx.h:507
@ AOME_SET_ROI_MAP
Codec control function to pass an ROI map to encoder, aom_roi_map_t* parameter.
Definition aomcx.h:185
@ AV1E_SET_CDF_UPDATE_MODE
Codec control function to set CDF update mode, unsigned int parameter.
Definition aomcx.h:516
@ AV1E_SET_ENABLE_ANGLE_DELTA
Codec control function to turn on/off intra angle delta, int parameter.
Definition aomcx.h:1128
@ AV1E_SET_MV_COST_UPD_FREQ
Control to set frequency of the cost updates for motion vectors, unsigned int parameter.
Definition aomcx.h:1271
@ AV1E_SET_INTRA_DEFAULT_TX_ONLY
Control to use default tx type only for intra modes, int parameter.
Definition aomcx.h:1220
@ AV1E_SET_SVC_REF_FRAME_COMP_PRED
Codec control function to set reference frame compound prediction. aom_svc_ref_frame_comp_pred_t* par...
Definition aomcx.h:1408
@ AV1E_SET_ENABLE_INTRABC
Codec control function to turn on/off intra block copy mode, int parameter.
Definition aomcx.h:1124
@ AV1E_SET_ENABLE_WARPED_MOTION
Codec control function to turn on / off warped motion usage at sequence level, int parameter.
Definition aomcx.h:1049
@ AV1E_SET_RTC_EXTERNAL_RC
Codec control function to set flag for rate control used by external encoders.
Definition aomcx.h:1443
@ AV1E_SET_COEFF_COST_UPD_FREQ
Control to set frequency of the cost updates for coefficients, unsigned int parameter.
Definition aomcx.h:1251
@ AV1E_SET_ENABLE_CDEF
Codec control function to encode with CDEF, unsigned int parameter.
Definition aomcx.h:681
@ AOME_SET_ACTIVEMAP
Codec control function to pass an Active map to encoder, aom_active_map_t* parameter.
Definition aomcx.h:190
@ AV1E_SET_DV_COST_UPD_FREQ
Control to set frequency of the cost updates for intrabc motion vectors, unsigned int parameter.
Definition aomcx.h:1374
@ AV1E_SET_SVC_FRAME_DROP_MODE
Codec control to set the frame drop mode for SVC, unsigned int parameter. The valid values are consta...
Definition aomcx.h:1557
@ AV1E_SET_SVC_PARAMS
Codec control function to set SVC parameters, aom_svc_params_t* parameter.
Definition aomcx.h:1298
@ AV1E_SET_ENABLE_FILTER_INTRA
Codec control function to turn on / off filter intra usage at sequence level, int parameter.
Definition aomcx.h:1070
@ AV1E_SET_ENABLE_PALETTE
Codec control function to turn on/off palette mode, int parameter.
Definition aomcx.h:1120
@ AV1E_SET_ENABLE_CFL_INTRA
Codec control function to turn on / off CFL uv intra mode usage, int parameter.
Definition aomcx.h:1099
@ AOME_SET_MAX_INTRA_BITRATE_PCT
Codec control function to set max data rate for intra frames, unsigned int parameter.
Definition aomcx.h:312
@ AV1E_SET_ERROR_RESILIENT_MODE
Codec control function to enable error_resilient_mode, int parameter.
Definition aomcx.h:452
@ AV1E_SET_ENABLE_OBMC
Codec control function to predict with OBMC mode, unsigned int parameter.
Definition aomcx.h:708
@ AV1E_SET_AUTO_TILES
Codec control to set auto tiling, unsigned int parameter. Value of 1 means encoder will set number of...
Definition aomcx.h:1565
@ AV1E_SET_LOOPFILTER_CONTROL
Codec control to control loop filter.
Definition aomcx.h:1423
@ AOME_SET_SCALEMODE
Codec control function to set encoder scaling mode for the next frame to be coded,...
Definition aomcx.h:197
@ AV1E_SET_ENABLE_ORDER_HINT
Codec control function to turn on / off frame order hint (int parameter). Affects: joint compound mod...
Definition aomcx.h:876
@ AV1E_SET_DELTAQ_MODE
Codec control function to set the delta q mode, unsigned int parameter.
Definition aomcx.h:1148
@ AV1E_SET_POSTENCODE_DROP_RTC
Codec control to enable post encode frame drop for RTC encoding, int parameter.
Definition aomcx.h:1581
@ AV1E_SET_ENABLE_GLOBAL_MOTION
Codec control function to turn on / off global motion usage for a sequence, int parameter.
Definition aomcx.h:1039
@ AOME_SET_CPUUSED
Codec control function to set encoder internal speed settings, int parameter.
Definition aomcx.h:220
@ AV1E_GET_HIGH_MOTION_CONTENT_SCREEN_RTC
Codec control to get the high motion content flag, used for screen content realtime (RTC) encoding,...
Definition aomcx.h:1572
@ AV1E_SET_GF_CBR_BOOST_PCT
Boost percentage for Golden Frame in CBR mode, unsigned int parameter.
Definition aomcx.h:349
@ AV1E_SET_QUANTIZER_ONE_PASS
Codec control to set quantizer for the next frame, int parameter.
Definition aomcx.h:1506
@ AV1E_SET_MODE_COST_UPD_FREQ
Control to set frequency of the cost updates for mode, unsigned int parameter.
Definition aomcx.h:1261
@ AV1E_SET_MAX_CONSEC_FRAME_DROP_MS_CBR
Codec control to set the maximum number of consecutive frame drops, in units of time (milliseconds),...
Definition aomcx.h:1587
@ AV1_GET_NEW_FRAME_IMAGE
Codec control function to get a pointer to the new frame.
Definition aom.h:70
const char * aom_codec_iface_name(aom_codec_iface_t *iface)
Return the name for a given interface.
enum aom_bit_depth aom_bit_depth_t
Bit depth for codecThis enumeration determines the bit depth of the codec.
aom_codec_err_t aom_codec_control(aom_codec_ctx_t *ctx, int ctrl_id,...)
Algorithm Control.
long aom_codec_flags_t
Initialization-time Feature Enabling.
Definition aom_codec.h:232
struct aom_codec_ctx aom_codec_ctx_t
Codec context structure.
const struct aom_codec_iface aom_codec_iface_t
Codec interface structure.
Definition aom_codec.h:271
aom_codec_err_t aom_codec_destroy(aom_codec_ctx_t *ctx)
Destroy a codec instance.
const char * aom_codec_err_to_string(aom_codec_err_t err)
Convert error number to printable string.
aom_codec_err_t
Algorithm return codes.
Definition aom_codec.h:155
#define AOM_CODEC_CONTROL_TYPECHECKED(ctx, id, data)
aom_codec_control wrapper macro (adds type-checking, less flexible)
Definition aom_codec.h:542
const void * aom_codec_iter_t
Iterator.
Definition aom_codec.h:305
#define AOM_FRAME_IS_KEY
Definition aom_codec.h:288
@ AOM_BITS_8
Definition aom_codec.h:336
@ AOM_BITS_10
Definition aom_codec.h:337
@ AOM_CODEC_INVALID_PARAM
An application-supplied parameter is not valid.
Definition aom_codec.h:200
@ AOM_CODEC_MEM_ERROR
Memory operation failed.
Definition aom_codec.h:163
@ AOM_CODEC_OK
Operation completed without error.
Definition aom_codec.h:157
aom_codec_err_t aom_codec_decode(aom_codec_ctx_t *ctx, const uint8_t *data, size_t data_sz, void *user_priv)
Decode data.
#define aom_codec_dec_init(ctx, iface, cfg, flags)
Convenience macro for aom_codec_dec_init_ver()
Definition aom_decoder.h:129
#define AOM_USAGE_GOOD_QUALITY
usage parameter analogous to AV1 GOOD QUALITY mode.
Definition aom_encoder.h:1014
const aom_codec_cx_pkt_t * aom_codec_get_cx_data(aom_codec_ctx_t *ctx, aom_codec_iter_t *iter)
Encoded data iterator.
struct aom_codec_cx_pkt aom_codec_cx_pkt_t
Encoder output packet.
aom_codec_err_t aom_codec_encode(aom_codec_ctx_t *ctx, const aom_image_t *img, aom_codec_pts_t pts, unsigned long duration, aom_enc_frame_flags_t flags)
Encode a frame.
#define aom_codec_enc_init(ctx, iface, cfg, flags)
Convenience macro for aom_codec_enc_init_ver()
Definition aom_encoder.h:943
aom_codec_err_t aom_codec_enc_config_default(aom_codec_iface_t *iface, aom_codec_enc_cfg_t *cfg, unsigned int usage)
Get the default configuration for a usage.
struct aom_codec_enc_cfg aom_codec_enc_cfg_t
Encoder configuration structure.
#define AOM_USAGE_REALTIME
usage parameter analogous to AV1 REALTIME mode.
Definition aom_encoder.h:1016
#define AOM_CODEC_USE_HIGHBITDEPTH
Definition aom_encoder.h:80
#define AOM_CODEC_USE_PSNR
Initialization-time Feature Enabling.
Definition aom_encoder.h:79
@ AOM_CBR
Definition aom_encoder.h:187
@ AOM_KF_AUTO
Definition aom_encoder.h:202
@ AOM_CODEC_PSNR_PKT
Definition aom_encoder.h:113
@ AOM_CODEC_CX_FRAME_PKT
Definition aom_encoder.h:110
unsigned int rows
Definition aomcx.h:1678
unsigned int cols
Definition aomcx.h:1679
unsigned char * active_map
specify an on (1) or off (0) each 16x16 region within a frame
Definition aomcx.h:1677
size_t sz
Definition aom_encoder.h:127
enum aom_codec_cx_pkt_kind kind
Definition aom_encoder.h:123
double psnr[4]
Definition aom_encoder.h:145
union aom_codec_cx_pkt::@202210014045072156205127107315337341215221351166 data
aom_codec_frame_flags_t flags
Definition aom_encoder.h:132
struct aom_codec_cx_pkt::@202210014045072156205127107315337341215221351166::@052232317104146204273007241322037340334334344046 frame
void * buf
Definition aom_encoder.h:126
unsigned int g_input_bit_depth
Bit-depth of the input frames.
Definition aom_encoder.h:475
unsigned int rc_dropframe_thresh
Temporal resampling configuration, if supported by the codec.
Definition aom_encoder.h:540
struct aom_rational g_timebase
Stream timebase units.
Definition aom_encoder.h:489
unsigned int g_usage
Algorithm specific "usage" value.
Definition aom_encoder.h:399
unsigned int rc_buf_sz
Decoder Buffer Size.
Definition aom_encoder.h:705
unsigned int g_h
Height of the frame.
Definition aom_encoder.h:435
enum aom_kf_mode kf_mode
Keyframe placement mode.
Definition aom_encoder.h:768
enum aom_rc_mode rc_end_usage
Rate control algorithm to use.
Definition aom_encoder.h:623
unsigned int g_threads
Maximum number of threads to use.
Definition aom_encoder.h:407
unsigned int kf_min_dist
Keyframe minimum interval.
Definition aom_encoder.h:777
unsigned int g_lag_in_frames
Allow lagged encoding.
Definition aom_encoder.h:518
unsigned int rc_buf_initial_sz
Decoder Buffer Initial Size.
Definition aom_encoder.h:714
unsigned int g_profile
Bitstream profile to use.
Definition aom_encoder.h:417
aom_bit_depth_t g_bit_depth
Bit-depth of the codec.
Definition aom_encoder.h:467
unsigned int g_w
Width of the frame.
Definition aom_encoder.h:426
unsigned int rc_undershoot_pct
Rate control adaptation undershoot control.
Definition aom_encoder.h:681
unsigned int kf_max_dist
Keyframe maximum interval.
Definition aom_encoder.h:786
aom_codec_er_flags_t g_error_resilient
Enable error resilient modes.
Definition aom_encoder.h:497
unsigned int rc_max_quantizer
Maximum (Worst Quality) Quantizer.
Definition aom_encoder.h:668
unsigned int rc_buf_optimal_sz
Decoder Buffer Optimal Size.
Definition aom_encoder.h:723
unsigned int rc_min_quantizer
Minimum (Best Quality) Quantizer.
Definition aom_encoder.h:658
unsigned int rc_target_bitrate
Target data rate.
Definition aom_encoder.h:644
unsigned int rc_resize_mode
Mode for spatial resampling, if supported by the codec.
Definition aom_encoder.h:549
unsigned int rc_overshoot_pct
Rate control adaptation overshoot control.
Definition aom_encoder.h:690
aom_img_fmt_t fmt
Definition aom_image.h:199
unsigned int d_w
Definition aom_image.h:213
unsigned int d_h
Definition aom_image.h:214
int num
Definition aom_encoder.h:165
int den
Definition aom_encoder.h:166
unsigned int cols
Definition aomcx.h:1659
int delta_lf[8]
Definition aomcx.h:1661
int ref_frame[8]
Definition aomcx.h:1663
unsigned int rows
Definition aomcx.h:1658
unsigned char * roi_map
Definition aomcx.h:1657
int delta_q[8]
Definition aomcx.h:1660
uint8_t enabled
Definition aomcx.h:1655
int skip[8]
Definition aomcx.h:1662
aom image scaling mode
Definition aomcx.h:1687
int temporal_layer_id
Definition aomcx.h:1784
int spatial_layer_id
Definition aomcx.h:1783
int max_quantizers[32]
Definition aomcx.h:1808
int number_spatial_layers
Definition aomcx.h:1800
int layer_target_bitrate[32]
Definition aomcx.h:1813
int framerate_factor[8]
Definition aomcx.h:1815
int min_quantizers[32]
Definition aomcx.h:1809
int scaling_factor_den[4]
Definition aomcx.h:1811
int number_temporal_layers
Definition aomcx.h:1807
int scaling_factor_num[4]
Definition aomcx.h:1810
int use_comp_pred[3]
Definition aomcx.h:1845
int reference[7]
Definition aomcx.h:1835
int refresh[8]
Definition aomcx.h:1838
int ref_idx[7]
Definition aomcx.h:1837