AOMedia AV1 Codec
svc_encoder_rtc
1 /*
2  * Copyright (c) 2019, Alliance for Open Media. All Rights Reserved.
3  *
4  * Use of this source code is governed by a BSD-style license
5  * that can be found in the LICENSE file in the root of the source
6  * tree. An additional intellectual property rights grant can be found
7  * in the file PATENTS. All contributing project authors may
8  * be found in the AUTHORS file in the root of the source tree.
9  */
10 
11 // This is an example demonstrating how to implement a multi-layer AOM
12 // encoding scheme for RTC video applications.
13 
14 #include <assert.h>
15 #include <math.h>
16 #include <stdio.h>
17 #include <stdlib.h>
18 #include <string.h>
19 
20 #include "aom/aom_encoder.h"
21 #include "aom/aomcx.h"
22 #include "av1/common/enums.h"
23 #include "av1/encoder/encoder.h"
24 #include "common/args.h"
25 #include "common/tools_common.h"
26 #include "common/video_writer.h"
27 #include "examples/encoder_util.h"
28 #include "aom_ports/aom_timer.h"
29 
30 #define OPTION_BUFFER_SIZE 1024
31 
32 typedef struct {
33  const char *output_filename;
34  char options[OPTION_BUFFER_SIZE];
35  struct AvxInputContext input_ctx;
36  int speed;
37  int aq_mode;
38  int layering_mode;
39  int output_obu;
40 } AppInput;
41 
42 typedef enum {
43  QUANTIZER = 0,
44  BITRATE,
45  SCALE_FACTOR,
46  AUTO_ALT_REF,
47  ALL_OPTION_TYPES
48 } LAYER_OPTION_TYPE;
49 
50 static const arg_def_t outputfile =
51  ARG_DEF("o", "output", 1, "Output filename");
52 static const arg_def_t frames_arg =
53  ARG_DEF("f", "frames", 1, "Number of frames to encode");
54 static const arg_def_t threads_arg =
55  ARG_DEF("th", "threads", 1, "Number of threads to use");
56 static const arg_def_t width_arg = ARG_DEF("w", "width", 1, "Source width");
57 static const arg_def_t height_arg = ARG_DEF("h", "height", 1, "Source height");
58 static const arg_def_t timebase_arg =
59  ARG_DEF("t", "timebase", 1, "Timebase (num/den)");
60 static const arg_def_t bitrate_arg = ARG_DEF(
61  "b", "target-bitrate", 1, "Encoding bitrate, in kilobits per second");
62 static const arg_def_t spatial_layers_arg =
63  ARG_DEF("sl", "spatial-layers", 1, "Number of spatial SVC layers");
64 static const arg_def_t temporal_layers_arg =
65  ARG_DEF("tl", "temporal-layers", 1, "Number of temporal SVC layers");
66 static const arg_def_t layering_mode_arg =
67  ARG_DEF("lm", "layering-mode", 1, "Temporal layering scheme.");
68 static const arg_def_t kf_dist_arg =
69  ARG_DEF("k", "kf-dist", 1, "Number of frames between keyframes");
70 static const arg_def_t scale_factors_arg =
71  ARG_DEF("r", "scale-factors", 1, "Scale factors (lowest to highest layer)");
72 static const arg_def_t min_q_arg =
73  ARG_DEF(NULL, "min-q", 1, "Minimum quantizer");
74 static const arg_def_t max_q_arg =
75  ARG_DEF(NULL, "max-q", 1, "Maximum quantizer");
76 static const arg_def_t speed_arg =
77  ARG_DEF("sp", "speed", 1, "Speed configuration");
78 static const arg_def_t aqmode_arg =
79  ARG_DEF("aq", "aqmode", 1, "AQ mode off/on");
80 static const arg_def_t bitrates_arg =
81  ARG_DEF("bl", "bitrates", 1,
82  "Bitrates[spatial_layer * num_temporal_layer + temporal_layer]");
83 static const arg_def_t dropframe_thresh_arg =
84  ARG_DEF(NULL, "drop-frame", 1, "Temporal resampling threshold (buf %)");
85 static const arg_def_t error_resilient_arg =
86  ARG_DEF(NULL, "error-resilient", 1, "Error resilient flag");
87 static const arg_def_t output_obu_arg =
88  ARG_DEF(NULL, "output-obu", 1,
89  "Write OBUs when set to 1. Otherwise write IVF files.");
90 
91 #if CONFIG_AV1_HIGHBITDEPTH
92 static const struct arg_enum_list bitdepth_enum[] = {
93  { "8", AOM_BITS_8 }, { "10", AOM_BITS_10 }, { "12", AOM_BITS_12 }, { NULL, 0 }
94 };
95 
96 static const arg_def_t bitdepth_arg = ARG_DEF_ENUM(
97  "d", "bit-depth", 1, "Bit depth for codec 8, 10 or 12. ", bitdepth_enum);
98 #endif // CONFIG_AV1_HIGHBITDEPTH
99 
100 static const arg_def_t *svc_args[] = {
101  &frames_arg, &outputfile, &width_arg,
102  &height_arg, &timebase_arg, &bitrate_arg,
103  &spatial_layers_arg, &kf_dist_arg, &scale_factors_arg,
104  &min_q_arg, &max_q_arg, &temporal_layers_arg,
105  &layering_mode_arg, &threads_arg, &aqmode_arg,
106 #if CONFIG_AV1_HIGHBITDEPTH
107  &bitdepth_arg,
108 #endif
109  &speed_arg, &bitrates_arg, &dropframe_thresh_arg,
110  &error_resilient_arg, &output_obu_arg, NULL
111 };
112 
113 #define zero(Dest) memset(&(Dest), 0, sizeof(Dest));
114 
115 static const char *exec_name;
116 
117 void usage_exit(void) {
118  fprintf(stderr, "Usage: %s <options> input_filename -o output_filename\n",
119  exec_name);
120  fprintf(stderr, "Options:\n");
121  arg_show_usage(stderr, svc_args);
122  exit(EXIT_FAILURE);
123 }
124 
125 static int file_is_y4m(const char detect[4]) {
126  return memcmp(detect, "YUV4", 4) == 0;
127 }
128 
129 static int fourcc_is_ivf(const char detect[4]) {
130  if (memcmp(detect, "DKIF", 4) == 0) {
131  return 1;
132  }
133  return 0;
134 }
135 
136 static const int option_max_values[ALL_OPTION_TYPES] = { 63, INT_MAX, INT_MAX,
137  1 };
138 
139 static const int option_min_values[ALL_OPTION_TYPES] = { 0, 0, 1, 0 };
140 
141 static void open_input_file(struct AvxInputContext *input,
143  /* Parse certain options from the input file, if possible */
144  input->file = strcmp(input->filename, "-") ? fopen(input->filename, "rb")
145  : set_binary_mode(stdin);
146 
147  if (!input->file) fatal("Failed to open input file");
148 
149  if (!fseeko(input->file, 0, SEEK_END)) {
150  /* Input file is seekable. Figure out how long it is, so we can get
151  * progress info.
152  */
153  input->length = ftello(input->file);
154  rewind(input->file);
155  }
156 
157  /* Default to 1:1 pixel aspect ratio. */
158  input->pixel_aspect_ratio.numerator = 1;
159  input->pixel_aspect_ratio.denominator = 1;
160 
161  /* For RAW input sources, these bytes will applied on the first frame
162  * in read_frame().
163  */
164  input->detect.buf_read = fread(input->detect.buf, 1, 4, input->file);
165  input->detect.position = 0;
166 
167  if (input->detect.buf_read == 4 && file_is_y4m(input->detect.buf)) {
168  if (y4m_input_open(&input->y4m, input->file, input->detect.buf, 4, csp,
169  input->only_i420) >= 0) {
170  input->file_type = FILE_TYPE_Y4M;
171  input->width = input->y4m.pic_w;
172  input->height = input->y4m.pic_h;
173  input->pixel_aspect_ratio.numerator = input->y4m.par_n;
174  input->pixel_aspect_ratio.denominator = input->y4m.par_d;
175  input->framerate.numerator = input->y4m.fps_n;
176  input->framerate.denominator = input->y4m.fps_d;
177  input->fmt = input->y4m.aom_fmt;
178  input->bit_depth = input->y4m.bit_depth;
179  } else {
180  fatal("Unsupported Y4M stream.");
181  }
182  } else if (input->detect.buf_read == 4 && fourcc_is_ivf(input->detect.buf)) {
183  fatal("IVF is not supported as input.");
184  } else {
185  input->file_type = FILE_TYPE_RAW;
186  }
187 }
188 
189 static aom_codec_err_t extract_option(LAYER_OPTION_TYPE type, char *input,
190  int *value0, int *value1) {
191  if (type == SCALE_FACTOR) {
192  *value0 = (int)strtol(input, &input, 10);
193  if (*input++ != '/') return AOM_CODEC_INVALID_PARAM;
194  *value1 = (int)strtol(input, &input, 10);
195 
196  if (*value0 < option_min_values[SCALE_FACTOR] ||
197  *value1 < option_min_values[SCALE_FACTOR] ||
198  *value0 > option_max_values[SCALE_FACTOR] ||
199  *value1 > option_max_values[SCALE_FACTOR] ||
200  *value0 > *value1) // num shouldn't be greater than den
202  } else {
203  *value0 = atoi(input);
204  if (*value0 < option_min_values[type] || *value0 > option_max_values[type])
206  }
207  return AOM_CODEC_OK;
208 }
209 
210 static aom_codec_err_t parse_layer_options_from_string(
211  aom_svc_params_t *svc_params, LAYER_OPTION_TYPE type, const char *input,
212  int *option0, int *option1) {
214  char *input_string;
215  char *token;
216  const char *delim = ",";
217  int num_layers = svc_params->number_spatial_layers;
218  int i = 0;
219 
220  if (type == BITRATE)
221  num_layers =
222  svc_params->number_spatial_layers * svc_params->number_temporal_layers;
223 
224  if (input == NULL || option0 == NULL ||
225  (option1 == NULL && type == SCALE_FACTOR))
227 
228  input_string = malloc(strlen(input));
229  memcpy(input_string, input, strlen(input));
230  if (input_string == NULL) return AOM_CODEC_MEM_ERROR;
231  token = strtok(input_string, delim); // NOLINT
232  for (i = 0; i < num_layers; ++i) {
233  if (token != NULL) {
234  res = extract_option(type, token, option0 + i, option1 + i);
235  if (res != AOM_CODEC_OK) break;
236  token = strtok(NULL, delim); // NOLINT
237  } else {
238  break;
239  }
240  }
241  if (res == AOM_CODEC_OK && i != num_layers) {
243  }
244  free(input_string);
245  return res;
246 }
247 
248 static void parse_command_line(int argc, const char **argv_,
249  AppInput *app_input,
250  aom_svc_params_t *svc_params,
251  aom_codec_enc_cfg_t *enc_cfg) {
252  struct arg arg;
253  char **argv = NULL;
254  char **argi = NULL;
255  char **argj = NULL;
256  char string_options[1024] = { 0 };
257 
258  // Default settings
259  svc_params->number_spatial_layers = 1;
260  svc_params->number_temporal_layers = 1;
261  app_input->layering_mode = 0;
262  app_input->output_obu = 0;
263  enc_cfg->g_threads = 1;
264  enc_cfg->rc_end_usage = AOM_CBR;
265 
266  // process command line options
267  argv = argv_dup(argc - 1, argv_ + 1);
268  for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) {
269  arg.argv_step = 1;
270 
271  if (arg_match(&arg, &outputfile, argi)) {
272  app_input->output_filename = arg.val;
273  } else if (arg_match(&arg, &width_arg, argi)) {
274  enc_cfg->g_w = arg_parse_uint(&arg);
275  } else if (arg_match(&arg, &height_arg, argi)) {
276  enc_cfg->g_h = arg_parse_uint(&arg);
277  } else if (arg_match(&arg, &timebase_arg, argi)) {
278  enc_cfg->g_timebase = arg_parse_rational(&arg);
279  } else if (arg_match(&arg, &bitrate_arg, argi)) {
280  enc_cfg->rc_target_bitrate = arg_parse_uint(&arg);
281  } else if (arg_match(&arg, &spatial_layers_arg, argi)) {
282  svc_params->number_spatial_layers = arg_parse_uint(&arg);
283  } else if (arg_match(&arg, &temporal_layers_arg, argi)) {
284  svc_params->number_temporal_layers = arg_parse_uint(&arg);
285  } else if (arg_match(&arg, &speed_arg, argi)) {
286  app_input->speed = arg_parse_uint(&arg);
287  if (app_input->speed > 9) {
288  aom_tools_warn("Mapping speed %d to speed 9.\n", app_input->speed);
289  }
290  } else if (arg_match(&arg, &aqmode_arg, argi)) {
291  app_input->aq_mode = arg_parse_uint(&arg);
292  } else if (arg_match(&arg, &threads_arg, argi)) {
293  enc_cfg->g_threads = arg_parse_uint(&arg);
294  } else if (arg_match(&arg, &layering_mode_arg, argi)) {
295  app_input->layering_mode = arg_parse_int(&arg);
296  } else if (arg_match(&arg, &kf_dist_arg, argi)) {
297  enc_cfg->kf_min_dist = arg_parse_uint(&arg);
298  enc_cfg->kf_max_dist = enc_cfg->kf_min_dist;
299  } else if (arg_match(&arg, &scale_factors_arg, argi)) {
300  parse_layer_options_from_string(svc_params, SCALE_FACTOR, arg.val,
301  svc_params->scaling_factor_num,
302  svc_params->scaling_factor_den);
303  } else if (arg_match(&arg, &min_q_arg, argi)) {
304  enc_cfg->rc_min_quantizer = arg_parse_uint(&arg);
305  } else if (arg_match(&arg, &max_q_arg, argi)) {
306  enc_cfg->rc_max_quantizer = arg_parse_uint(&arg);
307 #if CONFIG_AV1_HIGHBITDEPTH
308  } else if (arg_match(&arg, &bitdepth_arg, argi)) {
309  enc_cfg->g_bit_depth = arg_parse_enum_or_int(&arg);
310  switch (enc_cfg->g_bit_depth) {
311  case AOM_BITS_8:
312  enc_cfg->g_input_bit_depth = 8;
313  enc_cfg->g_profile = 0;
314  break;
315  case AOM_BITS_10:
316  enc_cfg->g_input_bit_depth = 10;
317  enc_cfg->g_profile = 2;
318  break;
319  case AOM_BITS_12:
320  enc_cfg->g_input_bit_depth = 12;
321  enc_cfg->g_profile = 2;
322  break;
323  default:
324  die("Error: Invalid bit depth selected (%d)\n", enc_cfg->g_bit_depth);
325  break;
326  }
327 #endif // CONFIG_VP9_HIGHBITDEPTH
328  } else if (arg_match(&arg, &dropframe_thresh_arg, argi)) {
329  enc_cfg->rc_dropframe_thresh = arg_parse_uint(&arg);
330  } else if (arg_match(&arg, &error_resilient_arg, argi)) {
331  enc_cfg->g_error_resilient = arg_parse_uint(&arg);
332  if (enc_cfg->g_error_resilient != 0 && enc_cfg->g_error_resilient != 1)
333  die("Invalid value for error resilient (0, 1): %d.",
334  enc_cfg->g_error_resilient);
335  } else if (arg_match(&arg, &output_obu_arg, argi)) {
336  app_input->output_obu = arg_parse_uint(&arg);
337  if (app_input->output_obu != 0 && app_input->output_obu != 1)
338  die("Invalid value for obu output flag (0, 1): %d.",
339  app_input->output_obu);
340  } else {
341  ++argj;
342  }
343  }
344 
345  // Total bitrate needs to be parsed after the number of layers.
346  for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) {
347  arg.argv_step = 1;
348  if (arg_match(&arg, &bitrates_arg, argi)) {
349  parse_layer_options_from_string(svc_params, BITRATE, arg.val,
350  svc_params->layer_target_bitrate, NULL);
351  } else {
352  ++argj;
353  }
354  }
355 
356  // There will be a space in front of the string options
357  if (strlen(string_options) > 0)
358  strncpy(app_input->options, string_options, OPTION_BUFFER_SIZE);
359 
360  // Check for unrecognized options
361  for (argi = argv; *argi; ++argi)
362  if (argi[0][0] == '-' && strlen(argi[0]) > 1)
363  die("Error: Unrecognized option %s\n", *argi);
364 
365  if (argv[0] == NULL) {
366  usage_exit();
367  }
368 
369  app_input->input_ctx.filename = argv[0];
370  free(argv);
371 
372  open_input_file(&app_input->input_ctx, 0);
373  if (app_input->input_ctx.file_type == FILE_TYPE_Y4M) {
374  enc_cfg->g_w = app_input->input_ctx.width;
375  enc_cfg->g_h = app_input->input_ctx.height;
376  }
377 
378  if (enc_cfg->g_w < 16 || enc_cfg->g_w % 2 || enc_cfg->g_h < 16 ||
379  enc_cfg->g_h % 2)
380  die("Invalid resolution: %d x %d\n", enc_cfg->g_w, enc_cfg->g_h);
381 
382  printf(
383  "Codec %s\n"
384  "layers: %d\n"
385  "width %u, height: %u\n"
386  "num: %d, den: %d, bitrate: %u\n"
387  "gop size: %u\n",
389  svc_params->number_spatial_layers, enc_cfg->g_w, enc_cfg->g_h,
390  enc_cfg->g_timebase.num, enc_cfg->g_timebase.den,
391  enc_cfg->rc_target_bitrate, enc_cfg->kf_max_dist);
392 }
393 
394 static unsigned int mode_to_num_temporal_layers[11] = { 1, 2, 3, 3, 2, 1,
395  1, 3, 3, 3, 3 };
396 static unsigned int mode_to_num_spatial_layers[11] = { 1, 1, 1, 1, 1, 2,
397  3, 2, 3, 3, 3 };
398 
399 // For rate control encoding stats.
400 struct RateControlMetrics {
401  // Number of input frames per layer.
402  int layer_input_frames[AOM_MAX_TS_LAYERS];
403  // Number of encoded non-key frames per layer.
404  int layer_enc_frames[AOM_MAX_TS_LAYERS];
405  // Framerate per layer layer (cumulative).
406  double layer_framerate[AOM_MAX_TS_LAYERS];
407  // Target average frame size per layer (per-frame-bandwidth per layer).
408  double layer_pfb[AOM_MAX_LAYERS];
409  // Actual average frame size per layer.
410  double layer_avg_frame_size[AOM_MAX_LAYERS];
411  // Average rate mismatch per layer (|target - actual| / target).
412  double layer_avg_rate_mismatch[AOM_MAX_LAYERS];
413  // Actual encoding bitrate per layer (cumulative across temporal layers).
414  double layer_encoding_bitrate[AOM_MAX_LAYERS];
415  // Average of the short-time encoder actual bitrate.
416  // TODO(marpan): Should we add these short-time stats for each layer?
417  double avg_st_encoding_bitrate;
418  // Variance of the short-time encoder actual bitrate.
419  double variance_st_encoding_bitrate;
420  // Window (number of frames) for computing short-timee encoding bitrate.
421  int window_size;
422  // Number of window measurements.
423  int window_count;
424  int layer_target_bitrate[AOM_MAX_LAYERS];
425 };
426 
427 // Reference frames used in this example encoder.
428 enum {
429  SVC_LAST_FRAME = 0,
430  SVC_LAST2_FRAME,
431  SVC_LAST3_FRAME,
432  SVC_GOLDEN_FRAME,
433  SVC_BWDREF_FRAME,
434  SVC_ALTREF2_FRAME,
435  SVC_ALTREF_FRAME
436 };
437 
438 static int read_frame(struct AvxInputContext *input_ctx, aom_image_t *img) {
439  FILE *f = input_ctx->file;
440  y4m_input *y4m = &input_ctx->y4m;
441  int shortread = 0;
442 
443  if (input_ctx->file_type == FILE_TYPE_Y4M) {
444  if (y4m_input_fetch_frame(y4m, f, img) < 1) return 0;
445  } else {
446  shortread = read_yuv_frame(input_ctx, img);
447  }
448 
449  return !shortread;
450 }
451 
452 static void close_input_file(struct AvxInputContext *input) {
453  fclose(input->file);
454  if (input->file_type == FILE_TYPE_Y4M) y4m_input_close(&input->y4m);
455 }
456 
457 // Note: these rate control metrics assume only 1 key frame in the
458 // sequence (i.e., first frame only). So for temporal pattern# 7
459 // (which has key frame for every frame on base layer), the metrics
460 // computation will be off/wrong.
461 // TODO(marpan): Update these metrics to account for multiple key frames
462 // in the stream.
463 static void set_rate_control_metrics(struct RateControlMetrics *rc,
464  double framerate,
465  unsigned int ss_number_layers,
466  unsigned int ts_number_layers) {
467  int ts_rate_decimator[AOM_MAX_TS_LAYERS] = { 1 };
468  ts_rate_decimator[0] = 1;
469  if (ts_number_layers == 2) {
470  ts_rate_decimator[0] = 2;
471  ts_rate_decimator[1] = 1;
472  }
473  if (ts_number_layers == 3) {
474  ts_rate_decimator[0] = 4;
475  ts_rate_decimator[1] = 2;
476  ts_rate_decimator[2] = 1;
477  }
478  // Set the layer (cumulative) framerate and the target layer (non-cumulative)
479  // per-frame-bandwidth, for the rate control encoding stats below.
480  for (unsigned int sl = 0; sl < ss_number_layers; ++sl) {
481  unsigned int i = sl * ts_number_layers;
482  rc->layer_framerate[0] = framerate / ts_rate_decimator[0];
483  rc->layer_pfb[i] =
484  1000.0 * rc->layer_target_bitrate[i] / rc->layer_framerate[0];
485  for (unsigned int tl = 0; tl < ts_number_layers; ++tl) {
486  i = sl * ts_number_layers + tl;
487  if (tl > 0) {
488  rc->layer_framerate[tl] = framerate / ts_rate_decimator[tl];
489  rc->layer_pfb[i] =
490  1000.0 *
491  (rc->layer_target_bitrate[i] - rc->layer_target_bitrate[i - 1]) /
492  (rc->layer_framerate[tl] - rc->layer_framerate[tl - 1]);
493  }
494  rc->layer_input_frames[tl] = 0;
495  rc->layer_enc_frames[tl] = 0;
496  rc->layer_encoding_bitrate[i] = 0.0;
497  rc->layer_avg_frame_size[i] = 0.0;
498  rc->layer_avg_rate_mismatch[i] = 0.0;
499  }
500  }
501  rc->window_count = 0;
502  rc->window_size = 15;
503  rc->avg_st_encoding_bitrate = 0.0;
504  rc->variance_st_encoding_bitrate = 0.0;
505 }
506 
507 static void printout_rate_control_summary(struct RateControlMetrics *rc,
508  int frame_cnt,
509  unsigned int ss_number_layers,
510  unsigned int ts_number_layers) {
511  int tot_num_frames = 0;
512  double perc_fluctuation = 0.0;
513  printf("Total number of processed frames: %d\n\n", frame_cnt - 1);
514  printf("Rate control layer stats for %u layer(s):\n\n", ts_number_layers);
515  for (unsigned int sl = 0; sl < ss_number_layers; ++sl) {
516  tot_num_frames = 0;
517  for (unsigned int tl = 0; tl < ts_number_layers; ++tl) {
518  unsigned int i = sl * ts_number_layers + tl;
519  const int num_dropped =
520  tl > 0 ? rc->layer_input_frames[tl] - rc->layer_enc_frames[tl]
521  : rc->layer_input_frames[tl] - rc->layer_enc_frames[tl] - 1;
522  tot_num_frames += rc->layer_input_frames[tl];
523  rc->layer_encoding_bitrate[i] = 0.001 * rc->layer_framerate[tl] *
524  rc->layer_encoding_bitrate[i] /
525  tot_num_frames;
526  rc->layer_avg_frame_size[i] =
527  rc->layer_avg_frame_size[i] / rc->layer_enc_frames[tl];
528  rc->layer_avg_rate_mismatch[i] =
529  100.0 * rc->layer_avg_rate_mismatch[i] / rc->layer_enc_frames[tl];
530  printf("For layer#: %u %u \n", sl, tl);
531  printf("Bitrate (target vs actual): %d %f\n", rc->layer_target_bitrate[i],
532  rc->layer_encoding_bitrate[i]);
533  printf("Average frame size (target vs actual): %f %f\n", rc->layer_pfb[i],
534  rc->layer_avg_frame_size[i]);
535  printf("Average rate_mismatch: %f\n", rc->layer_avg_rate_mismatch[i]);
536  printf(
537  "Number of input frames, encoded (non-key) frames, "
538  "and perc dropped frames: %d %d %f\n",
539  rc->layer_input_frames[tl], rc->layer_enc_frames[tl],
540  100.0 * num_dropped / rc->layer_input_frames[tl]);
541  printf("\n");
542  }
543  }
544  rc->avg_st_encoding_bitrate = rc->avg_st_encoding_bitrate / rc->window_count;
545  rc->variance_st_encoding_bitrate =
546  rc->variance_st_encoding_bitrate / rc->window_count -
547  (rc->avg_st_encoding_bitrate * rc->avg_st_encoding_bitrate);
548  perc_fluctuation = 100.0 * sqrt(rc->variance_st_encoding_bitrate) /
549  rc->avg_st_encoding_bitrate;
550  printf("Short-time stats, for window of %d frames:\n", rc->window_size);
551  printf("Average, rms-variance, and percent-fluct: %f %f %f\n",
552  rc->avg_st_encoding_bitrate, sqrt(rc->variance_st_encoding_bitrate),
553  perc_fluctuation);
554  if (frame_cnt - 1 != tot_num_frames)
555  die("Error: Number of input frames not equal to output!\n");
556 }
557 
558 // Layer pattern configuration.
559 static void set_layer_pattern(
560  int layering_mode, int superframe_cnt, aom_svc_layer_id_t *layer_id,
561  aom_svc_ref_frame_config_t *ref_frame_config,
562  aom_svc_ref_frame_comp_pred_t *ref_frame_comp_pred, int *use_svc_control,
563  int spatial_layer_id, int is_key_frame, int ksvc_mode, int speed) {
564  int i;
565  int enable_longterm_temporal_ref = 1;
566  int shift = (layering_mode == 8) ? 2 : 0;
567  *use_svc_control = 1;
568  layer_id->spatial_layer_id = spatial_layer_id;
569  int lag_index = 0;
570  int base_count = superframe_cnt >> 2;
571  ref_frame_comp_pred->use_comp_pred[0] = 0; // GOLDEN_LAST
572  ref_frame_comp_pred->use_comp_pred[1] = 0; // LAST2_LAST
573  ref_frame_comp_pred->use_comp_pred[2] = 0; // ALTREF_LAST
574  // Set the reference map buffer idx for the 7 references:
575  // LAST_FRAME (0), LAST2_FRAME(1), LAST3_FRAME(2), GOLDEN_FRAME(3),
576  // BWDREF_FRAME(4), ALTREF2_FRAME(5), ALTREF_FRAME(6).
577  for (i = 0; i < INTER_REFS_PER_FRAME; i++) ref_frame_config->ref_idx[i] = i;
578  for (i = 0; i < INTER_REFS_PER_FRAME; i++) ref_frame_config->reference[i] = 0;
579  for (i = 0; i < REF_FRAMES; i++) ref_frame_config->refresh[i] = 0;
580 
581  if (ksvc_mode) {
582  // Same pattern as case 9, but the reference strucutre will be constrained
583  // below.
584  layering_mode = 9;
585  }
586  switch (layering_mode) {
587  case 0:
588  // 1-layer: update LAST on every frame, reference LAST.
589  layer_id->temporal_layer_id = 0;
590  ref_frame_config->refresh[0] = 1;
591  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
592  break;
593  case 1:
594  // 2-temporal layer.
595  // 1 3 5
596  // 0 2 4
597  if (superframe_cnt % 2 == 0) {
598  layer_id->temporal_layer_id = 0;
599  // Update LAST on layer 0, reference LAST.
600  ref_frame_config->refresh[0] = 1;
601  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
602  } else {
603  layer_id->temporal_layer_id = 1;
604  // No updates on layer 1, only reference LAST (TL0).
605  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
606  }
607  break;
608  case 2:
609  // 3-temporal layer:
610  // 1 3 5 7
611  // 2 6
612  // 0 4 8
613  if (superframe_cnt % 4 == 0) {
614  // Base layer.
615  layer_id->temporal_layer_id = 0;
616  // Update LAST on layer 0, reference LAST.
617  ref_frame_config->refresh[0] = 1;
618  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
619  } else if ((superframe_cnt - 1) % 4 == 0) {
620  layer_id->temporal_layer_id = 2;
621  // First top layer: no updates, only reference LAST (TL0).
622  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
623  } else if ((superframe_cnt - 2) % 4 == 0) {
624  layer_id->temporal_layer_id = 1;
625  // Middle layer (TL1): update LAST2, only reference LAST (TL0).
626  ref_frame_config->refresh[1] = 1;
627  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
628  } else if ((superframe_cnt - 3) % 4 == 0) {
629  layer_id->temporal_layer_id = 2;
630  // Second top layer: no updates, only reference LAST.
631  // Set buffer idx for LAST to slot 1, since that was the slot
632  // updated in previous frame. So LAST is TL1 frame.
633  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
634  ref_frame_config->ref_idx[SVC_LAST2_FRAME] = 0;
635  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
636  }
637  break;
638  case 3:
639  // 3 TL, same as above, except allow for predicting
640  // off 2 more references (GOLDEN and ALTREF), with
641  // GOLDEN updated periodically, and ALTREF lagging from
642  // LAST from ~4 frames. Both GOLDEN and ALTREF
643  // can only be updated on base temporal layer.
644 
645  // Keep golden fixed at slot 3.
646  ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 3;
647  // Cyclically refresh slots 5, 6, 7, for lag altref.
648  lag_index = 5;
649  if (base_count > 0) {
650  lag_index = 5 + (base_count % 3);
651  if (superframe_cnt % 4 != 0) lag_index = 5 + ((base_count + 1) % 3);
652  }
653  // Set the altref slot to lag_index.
654  ref_frame_config->ref_idx[SVC_ALTREF_FRAME] = lag_index;
655  if (superframe_cnt % 4 == 0) {
656  // Base layer.
657  layer_id->temporal_layer_id = 0;
658  // Update LAST on layer 0, reference LAST.
659  ref_frame_config->refresh[0] = 1;
660  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
661  // Refresh GOLDEN every x ~10 base layer frames.
662  if (base_count % 10 == 0) ref_frame_config->refresh[3] = 1;
663  // Refresh lag_index slot, needed for lagging altref.
664  ref_frame_config->refresh[lag_index] = 1;
665  } else if ((superframe_cnt - 1) % 4 == 0) {
666  layer_id->temporal_layer_id = 2;
667  // First top layer: no updates, only reference LAST (TL0).
668  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
669  } else if ((superframe_cnt - 2) % 4 == 0) {
670  layer_id->temporal_layer_id = 1;
671  // Middle layer (TL1): update LAST2, only reference LAST (TL0).
672  ref_frame_config->refresh[1] = 1;
673  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
674  } else if ((superframe_cnt - 3) % 4 == 0) {
675  layer_id->temporal_layer_id = 2;
676  // Second top layer: no updates, only reference LAST.
677  // Set buffer idx for LAST to slot 1, since that was the slot
678  // updated in previous frame. So LAST is TL1 frame.
679  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
680  ref_frame_config->ref_idx[SVC_LAST2_FRAME] = 0;
681  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
682  }
683  // Every frame can reference GOLDEN AND ALTREF.
684  ref_frame_config->reference[SVC_GOLDEN_FRAME] = 1;
685  ref_frame_config->reference[SVC_ALTREF_FRAME] = 1;
686  // Allow for compound prediction using LAST and ALTREF.
687  if (speed >= 7) ref_frame_comp_pred->use_comp_pred[2] = 1;
688  break;
689  case 4:
690  // 3-temporal layer: but middle layer updates GF, so 2nd TL2 will
691  // only reference GF (not LAST). Other frames only reference LAST.
692  // 1 3 5 7
693  // 2 6
694  // 0 4 8
695  if (superframe_cnt % 4 == 0) {
696  // Base layer.
697  layer_id->temporal_layer_id = 0;
698  // Update LAST on layer 0, only reference LAST.
699  ref_frame_config->refresh[0] = 1;
700  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
701  } else if ((superframe_cnt - 1) % 4 == 0) {
702  layer_id->temporal_layer_id = 2;
703  // First top layer: no updates, only reference LAST (TL0).
704  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
705  } else if ((superframe_cnt - 2) % 4 == 0) {
706  layer_id->temporal_layer_id = 1;
707  // Middle layer (TL1): update GF, only reference LAST (TL0).
708  ref_frame_config->refresh[3] = 1;
709  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
710  } else if ((superframe_cnt - 3) % 4 == 0) {
711  layer_id->temporal_layer_id = 2;
712  // Second top layer: no updates, only reference GF.
713  ref_frame_config->reference[SVC_GOLDEN_FRAME] = 1;
714  }
715  break;
716  case 5:
717  // 2 spatial layers, 1 temporal.
718  layer_id->temporal_layer_id = 0;
719  if (layer_id->spatial_layer_id == 0) {
720  // Reference LAST, update LAST.
721  ref_frame_config->refresh[0] = 1;
722  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
723  } else if (layer_id->spatial_layer_id == 1) {
724  // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 1
725  // and GOLDEN to slot 0. Update slot 1 (LAST).
726  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
727  ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 0;
728  ref_frame_config->refresh[1] = 1;
729  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
730  ref_frame_config->reference[SVC_GOLDEN_FRAME] = 1;
731  }
732  break;
733  case 6:
734  // 3 spatial layers, 1 temporal.
735  // Note for this case, we set the buffer idx for all references to be
736  // either LAST or GOLDEN, which are always valid references, since decoder
737  // will check if any of the 7 references is valid scale in
738  // valid_ref_frame_size().
739  layer_id->temporal_layer_id = 0;
740  if (layer_id->spatial_layer_id == 0) {
741  // Reference LAST, update LAST. Set all buffer_idx to 0.
742  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
743  ref_frame_config->ref_idx[i] = 0;
744  ref_frame_config->refresh[0] = 1;
745  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
746  } else if (layer_id->spatial_layer_id == 1) {
747  // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 1
748  // and GOLDEN (and all other refs) to slot 0.
749  // Update slot 1 (LAST).
750  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
751  ref_frame_config->ref_idx[i] = 0;
752  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
753  ref_frame_config->refresh[1] = 1;
754  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
755  ref_frame_config->reference[SVC_GOLDEN_FRAME] = 1;
756  } else if (layer_id->spatial_layer_id == 2) {
757  // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 2
758  // and GOLDEN (and all other refs) to slot 1.
759  // Update slot 2 (LAST).
760  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
761  ref_frame_config->ref_idx[i] = 1;
762  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 2;
763  ref_frame_config->refresh[2] = 1;
764  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
765  ref_frame_config->reference[SVC_GOLDEN_FRAME] = 1;
766  // For 3 spatial layer case: allow for top spatial layer to use
767  // additional temporal reference. Update every 10 frames.
768  if (enable_longterm_temporal_ref) {
769  ref_frame_config->ref_idx[SVC_ALTREF_FRAME] = REF_FRAMES - 1;
770  ref_frame_config->reference[SVC_ALTREF_FRAME] = 1;
771  if (base_count % 10 == 0)
772  ref_frame_config->refresh[REF_FRAMES - 1] = 1;
773  }
774  }
775  break;
776  case 7:
777  // 2 spatial and 3 temporal layer.
778  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
779  if (superframe_cnt % 4 == 0) {
780  // Base temporal layer
781  layer_id->temporal_layer_id = 0;
782  if (layer_id->spatial_layer_id == 0) {
783  // Reference LAST, update LAST
784  // Set all buffer_idx to 0
785  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
786  ref_frame_config->ref_idx[i] = 0;
787  ref_frame_config->refresh[0] = 1;
788  } else if (layer_id->spatial_layer_id == 1) {
789  // Reference LAST and GOLDEN.
790  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
791  ref_frame_config->ref_idx[i] = 0;
792  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
793  ref_frame_config->refresh[1] = 1;
794  }
795  } else if ((superframe_cnt - 1) % 4 == 0) {
796  // First top temporal enhancement layer.
797  layer_id->temporal_layer_id = 2;
798  if (layer_id->spatial_layer_id == 0) {
799  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
800  ref_frame_config->ref_idx[i] = 0;
801  ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 3;
802  ref_frame_config->refresh[3] = 1;
803  } else if (layer_id->spatial_layer_id == 1) {
804  // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 1,
805  // GOLDEN (and all other refs) to slot 3.
806  // No update.
807  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
808  ref_frame_config->ref_idx[i] = 3;
809  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
810  }
811  } else if ((superframe_cnt - 2) % 4 == 0) {
812  // Middle temporal enhancement layer.
813  layer_id->temporal_layer_id = 1;
814  if (layer_id->spatial_layer_id == 0) {
815  // Reference LAST.
816  // Set all buffer_idx to 0.
817  // Set GOLDEN to slot 5 and update slot 5.
818  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
819  ref_frame_config->ref_idx[i] = 0;
820  ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 5 - shift;
821  ref_frame_config->refresh[5 - shift] = 1;
822  } else if (layer_id->spatial_layer_id == 1) {
823  // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 1,
824  // GOLDEN (and all other refs) to slot 5.
825  // Set LAST3 to slot 6 and update slot 6.
826  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
827  ref_frame_config->ref_idx[i] = 5 - shift;
828  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
829  ref_frame_config->ref_idx[SVC_LAST3_FRAME] = 6 - shift;
830  ref_frame_config->refresh[6 - shift] = 1;
831  }
832  } else if ((superframe_cnt - 3) % 4 == 0) {
833  // Second top temporal enhancement layer.
834  layer_id->temporal_layer_id = 2;
835  if (layer_id->spatial_layer_id == 0) {
836  // Set LAST to slot 5 and reference LAST.
837  // Set GOLDEN to slot 3 and update slot 3.
838  // Set all other buffer_idx to 0.
839  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
840  ref_frame_config->ref_idx[i] = 0;
841  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 5 - shift;
842  ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 3;
843  ref_frame_config->refresh[3] = 1;
844  } else if (layer_id->spatial_layer_id == 1) {
845  // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 6,
846  // GOLDEN to slot 3. No update.
847  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
848  ref_frame_config->ref_idx[i] = 0;
849  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 6 - shift;
850  ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 3;
851  }
852  }
853  break;
854  case 8:
855  // 3 spatial and 3 temporal layer.
856  // Same as case 9 but overalap in the buffer slot updates.
857  // (shift = 2). The slots 3 and 4 updated by first TL2 are
858  // reused for update in TL1 superframe.
859  // Note for this case, frame order hint must be disabled for
860  // lower resolutios (operating points > 0) to be decoedable.
861  case 9:
862  // 3 spatial and 3 temporal layer.
863  // No overlap in buffer updates between TL2 and TL1.
864  // TL2 updates slot 3 and 4, TL1 updates 5, 6, 7.
865  // Set the references via the svc_ref_frame_config control.
866  // Always reference LAST.
867  ref_frame_config->reference[SVC_LAST_FRAME] = 1;
868  if (superframe_cnt % 4 == 0) {
869  // Base temporal layer.
870  layer_id->temporal_layer_id = 0;
871  if (layer_id->spatial_layer_id == 0) {
872  // Reference LAST, update LAST.
873  // Set all buffer_idx to 0.
874  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
875  ref_frame_config->ref_idx[i] = 0;
876  ref_frame_config->refresh[0] = 1;
877  } else if (layer_id->spatial_layer_id == 1) {
878  // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 1,
879  // GOLDEN (and all other refs) to slot 0.
880  // Update slot 1 (LAST).
881  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
882  ref_frame_config->ref_idx[i] = 0;
883  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
884  ref_frame_config->refresh[1] = 1;
885  } else if (layer_id->spatial_layer_id == 2) {
886  // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 2,
887  // GOLDEN (and all other refs) to slot 1.
888  // Update slot 2 (LAST).
889  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
890  ref_frame_config->ref_idx[i] = 1;
891  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 2;
892  ref_frame_config->refresh[2] = 1;
893  }
894  } else if ((superframe_cnt - 1) % 4 == 0) {
895  // First top temporal enhancement layer.
896  layer_id->temporal_layer_id = 2;
897  if (layer_id->spatial_layer_id == 0) {
898  // Reference LAST (slot 0).
899  // Set GOLDEN to slot 3 and update slot 3.
900  // Set all other buffer_idx to slot 0.
901  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
902  ref_frame_config->ref_idx[i] = 0;
903  ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 3;
904  ref_frame_config->refresh[3] = 1;
905  } else if (layer_id->spatial_layer_id == 1) {
906  // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 1,
907  // GOLDEN (and all other refs) to slot 3.
908  // Set LAST2 to slot 4 and Update slot 4.
909  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
910  ref_frame_config->ref_idx[i] = 3;
911  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
912  ref_frame_config->ref_idx[SVC_LAST2_FRAME] = 4;
913  ref_frame_config->refresh[4] = 1;
914  } else if (layer_id->spatial_layer_id == 2) {
915  // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 2,
916  // GOLDEN (and all other refs) to slot 4.
917  // No update.
918  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
919  ref_frame_config->ref_idx[i] = 4;
920  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 2;
921  }
922  } else if ((superframe_cnt - 2) % 4 == 0) {
923  // Middle temporal enhancement layer.
924  layer_id->temporal_layer_id = 1;
925  if (layer_id->spatial_layer_id == 0) {
926  // Reference LAST.
927  // Set all buffer_idx to 0.
928  // Set GOLDEN to slot 5 and update slot 5.
929  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
930  ref_frame_config->ref_idx[i] = 0;
931  ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 5 - shift;
932  ref_frame_config->refresh[5 - shift] = 1;
933  } else if (layer_id->spatial_layer_id == 1) {
934  // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 1,
935  // GOLDEN (and all other refs) to slot 5.
936  // Set LAST3 to slot 6 and update slot 6.
937  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
938  ref_frame_config->ref_idx[i] = 5 - shift;
939  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 1;
940  ref_frame_config->ref_idx[SVC_LAST3_FRAME] = 6 - shift;
941  ref_frame_config->refresh[6 - shift] = 1;
942  } else if (layer_id->spatial_layer_id == 2) {
943  // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 2,
944  // GOLDEN (and all other refs) to slot 6.
945  // Set LAST3 to slot 7 and update slot 7.
946  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
947  ref_frame_config->ref_idx[i] = 6 - shift;
948  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 2;
949  ref_frame_config->ref_idx[SVC_LAST3_FRAME] = 7 - shift;
950  ref_frame_config->refresh[7 - shift] = 1;
951  }
952  } else if ((superframe_cnt - 3) % 4 == 0) {
953  // Second top temporal enhancement layer.
954  layer_id->temporal_layer_id = 2;
955  if (layer_id->spatial_layer_id == 0) {
956  // Set LAST to slot 5 and reference LAST.
957  // Set GOLDEN to slot 3 and update slot 3.
958  // Set all other buffer_idx to 0.
959  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
960  ref_frame_config->ref_idx[i] = 0;
961  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 5 - shift;
962  ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 3;
963  ref_frame_config->refresh[3] = 1;
964  } else if (layer_id->spatial_layer_id == 1) {
965  // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 6,
966  // GOLDEN to slot 3. Set LAST2 to slot 4 and update slot 4.
967  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
968  ref_frame_config->ref_idx[i] = 0;
969  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 6 - shift;
970  ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 3;
971  ref_frame_config->ref_idx[SVC_LAST2_FRAME] = 4;
972  ref_frame_config->refresh[4] = 1;
973  } else if (layer_id->spatial_layer_id == 2) {
974  // Reference LAST and GOLDEN. Set buffer_idx for LAST to slot 7,
975  // GOLDEN to slot 4. No update.
976  for (i = 0; i < INTER_REFS_PER_FRAME; i++)
977  ref_frame_config->ref_idx[i] = 0;
978  ref_frame_config->ref_idx[SVC_LAST_FRAME] = 7 - shift;
979  ref_frame_config->ref_idx[SVC_GOLDEN_FRAME] = 4;
980  }
981  }
982  if (layer_id->spatial_layer_id > 0) {
983  // Always reference GOLDEN (inter-layer prediction).
984  ref_frame_config->reference[SVC_GOLDEN_FRAME] = 1;
985  if (ksvc_mode) {
986  // KSVC: only keep the inter-layer reference (GOLDEN) for
987  // superframes whose base is key.
988  if (!is_key_frame) ref_frame_config->reference[SVC_GOLDEN_FRAME] = 0;
989  }
990  if (is_key_frame && layer_id->spatial_layer_id > 1) {
991  // On superframes whose base is key: remove LAST to avoid prediction
992  // off layer two levels below.
993  ref_frame_config->reference[SVC_LAST_FRAME] = 0;
994  }
995  }
996  // For 3 spatial layer case 8 (where there is free buffer slot):
997  // allow for top spatial layer to use additional temporal reference.
998  // Additional reference is only updated on base temporal layer, every
999  // 10 TL0 frames here.
1000  if (enable_longterm_temporal_ref && layer_id->spatial_layer_id == 2 &&
1001  layering_mode == 8) {
1002  ref_frame_config->ref_idx[SVC_ALTREF_FRAME] = REF_FRAMES - 1;
1003  ref_frame_config->reference[SVC_ALTREF_FRAME] = 1;
1004  if (base_count % 10 == 0 && layer_id->temporal_layer_id == 0)
1005  ref_frame_config->refresh[REF_FRAMES - 1] = 1;
1006  }
1007  break;
1008  default: assert(0); die("Error: Unsupported temporal layering mode!\n");
1009  }
1010 }
1011 
1012 #if CONFIG_AV1_DECODER
1013 static void test_decode(aom_codec_ctx_t *encoder, aom_codec_ctx_t *decoder,
1014  const int frames_out, int *mismatch_seen) {
1015  aom_image_t enc_img, dec_img;
1016 
1017  if (*mismatch_seen) return;
1018 
1019  /* Get the internal reference frame */
1022 
1023 #if CONFIG_AV1_HIGHBITDEPTH
1024  if ((enc_img.fmt & AOM_IMG_FMT_HIGHBITDEPTH) !=
1025  (dec_img.fmt & AOM_IMG_FMT_HIGHBITDEPTH)) {
1026  if (enc_img.fmt & AOM_IMG_FMT_HIGHBITDEPTH) {
1027  aom_image_t enc_hbd_img;
1028  aom_img_alloc(&enc_hbd_img, enc_img.fmt - AOM_IMG_FMT_HIGHBITDEPTH,
1029  enc_img.d_w, enc_img.d_h, 16);
1030  aom_img_truncate_16_to_8(&enc_hbd_img, &enc_img);
1031  enc_img = enc_hbd_img;
1032  }
1033  if (dec_img.fmt & AOM_IMG_FMT_HIGHBITDEPTH) {
1034  aom_image_t dec_hbd_img;
1035  aom_img_alloc(&dec_hbd_img, dec_img.fmt - AOM_IMG_FMT_HIGHBITDEPTH,
1036  dec_img.d_w, dec_img.d_h, 16);
1037  aom_img_truncate_16_to_8(&dec_hbd_img, &dec_img);
1038  dec_img = dec_hbd_img;
1039  }
1040  }
1041 #endif
1042 
1043  if (!aom_compare_img(&enc_img, &dec_img)) {
1044  int y[4], u[4], v[4];
1045 #if CONFIG_AV1_HIGHBITDEPTH
1046  if (enc_img.fmt & AOM_IMG_FMT_HIGHBITDEPTH) {
1047  aom_find_mismatch_high(&enc_img, &dec_img, y, u, v);
1048  } else {
1049  aom_find_mismatch(&enc_img, &dec_img, y, u, v);
1050  }
1051 #else
1052  aom_find_mismatch(&enc_img, &dec_img, y, u, v);
1053 #endif
1054  decoder->err = 1;
1055  printf(
1056  "Encode/decode mismatch on frame %d at"
1057  " Y[%d, %d] {%d/%d},"
1058  " U[%d, %d] {%d/%d},"
1059  " V[%d, %d] {%d/%d}",
1060  frames_out, y[0], y[1], y[2], y[3], u[0], u[1], u[2], u[3], v[0], v[1],
1061  v[2], v[3]);
1062  *mismatch_seen = frames_out;
1063  }
1064 
1065  aom_img_free(&enc_img);
1066  aom_img_free(&dec_img);
1067 }
1068 #endif // CONFIG_AV1_DECODER
1069 
1070 int main(int argc, const char **argv) {
1071  AppInput app_input;
1072  AvxVideoWriter *outfile[AOM_MAX_LAYERS] = { NULL };
1073  FILE *obu_files[AOM_MAX_LAYERS] = { NULL };
1074  AvxVideoWriter *total_layer_file = NULL;
1075  FILE *total_layer_obu_file = NULL;
1076  aom_codec_enc_cfg_t cfg;
1077  int frame_cnt = 0;
1078  aom_image_t raw;
1079  int frame_avail;
1080  int got_data = 0;
1081  int flags = 0;
1082  unsigned i;
1083  int pts = 0; // PTS starts at 0.
1084  int frame_duration = 1; // 1 timebase tick per frame.
1085  aom_svc_layer_id_t layer_id;
1086  aom_svc_params_t svc_params;
1087  aom_svc_ref_frame_config_t ref_frame_config;
1088  aom_svc_ref_frame_comp_pred_t ref_frame_comp_pred;
1089 
1090 #if CONFIG_INTERNAL_STATS
1091  FILE *stats_file = fopen("opsnr.stt", "a");
1092  if (stats_file == NULL) {
1093  die("Cannot open opsnr.stt\n");
1094  }
1095 #endif
1096 #if CONFIG_AV1_DECODER
1097  int mismatch_seen = 0;
1098  aom_codec_ctx_t decoder;
1099 #endif
1100 
1101  struct RateControlMetrics rc;
1102  int64_t cx_time = 0;
1103  int64_t cx_time_sl[3]; // max number of spatial layers.
1104  double sum_bitrate = 0.0;
1105  double sum_bitrate2 = 0.0;
1106  double framerate = 30.0;
1107  int use_svc_control = 1;
1108  int set_err_resil_frame = 0;
1109  zero(rc.layer_target_bitrate);
1110  memset(&layer_id, 0, sizeof(aom_svc_layer_id_t));
1111  memset(&app_input, 0, sizeof(AppInput));
1112  memset(&svc_params, 0, sizeof(svc_params));
1113 
1114  // Flag to test dynamic scaling of source frames for single
1115  // spatial stream, using the scaling_mode control.
1116  const int test_dynamic_scaling_single_layer = 0;
1117 
1118  /* Setup default input stream settings */
1119  app_input.input_ctx.framerate.numerator = 30;
1120  app_input.input_ctx.framerate.denominator = 1;
1121  app_input.input_ctx.only_i420 = 1;
1122  app_input.input_ctx.bit_depth = 0;
1123  app_input.speed = 7;
1124  exec_name = argv[0];
1125 
1126  // start with default encoder configuration
1129  if (res) {
1130  die("Failed to get config: %s\n", aom_codec_err_to_string(res));
1131  }
1132 
1133  // Real time parameters.
1135 
1136  cfg.rc_end_usage = AOM_CBR;
1137  cfg.rc_min_quantizer = 2;
1138  cfg.rc_max_quantizer = 52;
1139  cfg.rc_undershoot_pct = 50;
1140  cfg.rc_overshoot_pct = 50;
1141  cfg.rc_buf_initial_sz = 600;
1142  cfg.rc_buf_optimal_sz = 600;
1143  cfg.rc_buf_sz = 1000;
1144  cfg.rc_resize_mode = 0; // Set to RESIZE_DYNAMIC for dynamic resize.
1145  cfg.g_lag_in_frames = 0;
1146  cfg.kf_mode = AOM_KF_AUTO;
1147 
1148  parse_command_line(argc, argv, &app_input, &svc_params, &cfg);
1149 
1150  unsigned int ts_number_layers = svc_params.number_temporal_layers;
1151  unsigned int ss_number_layers = svc_params.number_spatial_layers;
1152 
1153  unsigned int width = cfg.g_w;
1154  unsigned int height = cfg.g_h;
1155 
1156  if (app_input.layering_mode >= 0) {
1157  if (ts_number_layers !=
1158  mode_to_num_temporal_layers[app_input.layering_mode] ||
1159  ss_number_layers !=
1160  mode_to_num_spatial_layers[app_input.layering_mode]) {
1161  die("Number of layers doesn't match layering mode.");
1162  }
1163  }
1164 
1165  // Y4M reader has its own allocation.
1166  if (app_input.input_ctx.file_type != FILE_TYPE_Y4M) {
1167  if (!aom_img_alloc(&raw, AOM_IMG_FMT_I420, width, height, 32)) {
1168  die("Failed to allocate image (%dx%d)", width, height);
1169  }
1170  }
1171 
1172  aom_codec_iface_t *encoder = get_aom_encoder_by_short_name("av1");
1173 
1174  memcpy(&rc.layer_target_bitrate[0], &svc_params.layer_target_bitrate[0],
1175  sizeof(svc_params.layer_target_bitrate));
1176 
1177  unsigned int total_rate = 0;
1178  for (i = 0; i < ss_number_layers; i++) {
1179  total_rate +=
1180  svc_params
1181  .layer_target_bitrate[i * ts_number_layers + ts_number_layers - 1];
1182  }
1183  if (total_rate != cfg.rc_target_bitrate) {
1184  die("Incorrect total target bitrate");
1185  }
1186 
1187  svc_params.framerate_factor[0] = 1;
1188  if (ts_number_layers == 2) {
1189  svc_params.framerate_factor[0] = 2;
1190  svc_params.framerate_factor[1] = 1;
1191  } else if (ts_number_layers == 3) {
1192  svc_params.framerate_factor[0] = 4;
1193  svc_params.framerate_factor[1] = 2;
1194  svc_params.framerate_factor[2] = 1;
1195  }
1196 
1197  if (app_input.input_ctx.file_type == FILE_TYPE_Y4M) {
1198  // Override these settings with the info from Y4M file.
1199  cfg.g_w = app_input.input_ctx.width;
1200  cfg.g_h = app_input.input_ctx.height;
1201  // g_timebase is the reciprocal of frame rate.
1202  cfg.g_timebase.num = app_input.input_ctx.framerate.denominator;
1203  cfg.g_timebase.den = app_input.input_ctx.framerate.numerator;
1204  }
1205  framerate = cfg.g_timebase.den / cfg.g_timebase.num;
1206  set_rate_control_metrics(&rc, framerate, ss_number_layers, ts_number_layers);
1207 
1208  AvxVideoInfo info;
1209  info.codec_fourcc = get_fourcc_by_aom_encoder(encoder);
1210  info.frame_width = cfg.g_w;
1211  info.frame_height = cfg.g_h;
1212  info.time_base.numerator = cfg.g_timebase.num;
1213  info.time_base.denominator = cfg.g_timebase.den;
1214  // Open an output file for each stream.
1215  for (unsigned int sl = 0; sl < ss_number_layers; ++sl) {
1216  for (unsigned tl = 0; tl < ts_number_layers; ++tl) {
1217  i = sl * ts_number_layers + tl;
1218  char file_name[PATH_MAX];
1219  snprintf(file_name, sizeof(file_name), "%s_%u.av1",
1220  app_input.output_filename, i);
1221  if (app_input.output_obu) {
1222  obu_files[i] = fopen(file_name, "wb");
1223  if (!obu_files[i]) die("Failed to open %s for writing", file_name);
1224  } else {
1225  outfile[i] = aom_video_writer_open(file_name, kContainerIVF, &info);
1226  if (!outfile[i]) die("Failed to open %s for writing", file_name);
1227  }
1228  }
1229  }
1230  if (app_input.output_obu) {
1231  total_layer_obu_file = fopen(app_input.output_filename, "wb");
1232  if (!total_layer_obu_file)
1233  die("Failed to open %s for writing", app_input.output_filename);
1234  } else {
1235  total_layer_file =
1236  aom_video_writer_open(app_input.output_filename, kContainerIVF, &info);
1237  if (!total_layer_file)
1238  die("Failed to open %s for writing", app_input.output_filename);
1239  }
1240 
1241  // Initialize codec.
1242  aom_codec_ctx_t codec;
1243  if (aom_codec_enc_init(&codec, encoder, &cfg, 0))
1244  die("Failed to initialize encoder");
1245 
1246 #if CONFIG_AV1_DECODER
1247  if (aom_codec_dec_init(&decoder, get_aom_decoder_by_index(0), NULL, 0)) {
1248  die("Failed to initialize decoder");
1249  }
1250 #endif
1251 
1252  aom_codec_control(&codec, AOME_SET_CPUUSED, app_input.speed);
1253  aom_codec_control(&codec, AV1E_SET_AQ_MODE, app_input.aq_mode ? 3 : 0);
1269  cfg.g_threads ? get_msb(cfg.g_threads) : 0);
1270  if (cfg.g_threads > 1) aom_codec_control(&codec, AV1E_SET_ROW_MT, 1);
1271 
1272  svc_params.number_spatial_layers = ss_number_layers;
1273  svc_params.number_temporal_layers = ts_number_layers;
1274  for (i = 0; i < ss_number_layers * ts_number_layers; ++i) {
1275  svc_params.max_quantizers[i] = cfg.rc_max_quantizer;
1276  svc_params.min_quantizers[i] = cfg.rc_min_quantizer;
1277  }
1278  for (i = 0; i < ss_number_layers; ++i) {
1279  svc_params.scaling_factor_num[i] = 1;
1280  svc_params.scaling_factor_den[i] = 1;
1281  }
1282  if (ss_number_layers == 2) {
1283  svc_params.scaling_factor_num[0] = 1;
1284  svc_params.scaling_factor_den[0] = 2;
1285  } else if (ss_number_layers == 3) {
1286  svc_params.scaling_factor_num[0] = 1;
1287  svc_params.scaling_factor_den[0] = 4;
1288  svc_params.scaling_factor_num[1] = 1;
1289  svc_params.scaling_factor_den[1] = 2;
1290  }
1291  aom_codec_control(&codec, AV1E_SET_SVC_PARAMS, &svc_params);
1292  // TODO(aomedia:3032): Configure KSVC in fixed mode.
1293 
1294  // This controls the maximum target size of the key frame.
1295  // For generating smaller key frames, use a smaller max_intra_size_pct
1296  // value, like 100 or 200.
1297  {
1298  const int max_intra_size_pct = 300;
1300  max_intra_size_pct);
1301  }
1302 
1303  for (unsigned int slx = 0; slx < ss_number_layers; slx++) cx_time_sl[slx] = 0;
1304  frame_avail = 1;
1305  while (frame_avail || got_data) {
1306  struct aom_usec_timer timer;
1307  frame_avail = read_frame(&(app_input.input_ctx), &raw);
1308  // Loop over spatial layers.
1309  for (unsigned int slx = 0; slx < ss_number_layers; slx++) {
1310  aom_codec_iter_t iter = NULL;
1311  const aom_codec_cx_pkt_t *pkt;
1312  int layer = 0;
1313  // Flag for superframe whose base is key.
1314  int is_key_frame = (frame_cnt % cfg.kf_max_dist) == 0;
1315  // For flexible mode:
1316  if (app_input.layering_mode >= 0) {
1317  // Set the reference/update flags, layer_id, and reference_map
1318  // buffer index.
1319  set_layer_pattern(app_input.layering_mode, frame_cnt, &layer_id,
1320  &ref_frame_config, &ref_frame_comp_pred,
1321  &use_svc_control, slx, is_key_frame,
1322  (app_input.layering_mode == 10), app_input.speed);
1323  aom_codec_control(&codec, AV1E_SET_SVC_LAYER_ID, &layer_id);
1324  if (use_svc_control) {
1326  &ref_frame_config);
1328  &ref_frame_comp_pred);
1329  }
1330  } else {
1331  // Only up to 3 temporal layers supported in fixed mode.
1332  // Only need to set spatial and temporal layer_id: reference
1333  // prediction, refresh, and buffer_idx are set internally.
1334  layer_id.spatial_layer_id = slx;
1335  layer_id.temporal_layer_id = 0;
1336  if (ts_number_layers == 2) {
1337  layer_id.temporal_layer_id = (frame_cnt % 2) != 0;
1338  } else if (ts_number_layers == 3) {
1339  if (frame_cnt % 2 != 0)
1340  layer_id.temporal_layer_id = 2;
1341  else if ((frame_cnt > 1) && ((frame_cnt - 2) % 4 == 0))
1342  layer_id.temporal_layer_id = 1;
1343  }
1344  aom_codec_control(&codec, AV1E_SET_SVC_LAYER_ID, &layer_id);
1345  }
1346 
1347  if (set_err_resil_frame) {
1348  // Set error_resilient per frame: off/0 for base layer and
1349  // on/1 for enhancement layer frames.
1350  int err_resil_mode =
1351  (layer_id.spatial_layer_id > 0 || layer_id.temporal_layer_id > 0);
1353  err_resil_mode);
1354  }
1355 
1356  layer = slx * ts_number_layers + layer_id.temporal_layer_id;
1357  if (frame_avail && slx == 0) ++rc.layer_input_frames[layer];
1358 
1359  if (test_dynamic_scaling_single_layer) {
1360  if (frame_cnt >= 200 && frame_cnt <= 400) {
1361  // Scale source down by 2x2.
1362  struct aom_scaling_mode mode = { AOME_ONETWO, AOME_ONETWO };
1363  aom_codec_control(&codec, AOME_SET_SCALEMODE, &mode);
1364  } else {
1365  // Source back up to original resolution (no scaling).
1366  struct aom_scaling_mode mode = { AOME_NORMAL, AOME_NORMAL };
1367  aom_codec_control(&codec, AOME_SET_SCALEMODE, &mode);
1368  }
1369  }
1370 
1371  // Do the layer encode.
1372  aom_usec_timer_start(&timer);
1373  if (aom_codec_encode(&codec, frame_avail ? &raw : NULL, pts, 1, flags))
1374  die_codec(&codec, "Failed to encode frame");
1375  aom_usec_timer_mark(&timer);
1376  cx_time += aom_usec_timer_elapsed(&timer);
1377  cx_time_sl[slx] += aom_usec_timer_elapsed(&timer);
1378 
1379  got_data = 0;
1380  while ((pkt = aom_codec_get_cx_data(&codec, &iter))) {
1381  got_data = 1;
1382  switch (pkt->kind) {
1384  for (unsigned int sl = layer_id.spatial_layer_id;
1385  sl < ss_number_layers; ++sl) {
1386  for (unsigned tl = layer_id.temporal_layer_id;
1387  tl < ts_number_layers; ++tl) {
1388  unsigned int j = sl * ts_number_layers + tl;
1389  if (app_input.output_obu) {
1390  fwrite(pkt->data.frame.buf, 1, pkt->data.frame.sz,
1391  obu_files[j]);
1392  } else {
1393  aom_video_writer_write_frame(outfile[j], pkt->data.frame.buf,
1394  pkt->data.frame.sz, pts);
1395  }
1396  if (sl == (unsigned int)layer_id.spatial_layer_id)
1397  rc.layer_encoding_bitrate[j] += 8.0 * pkt->data.frame.sz;
1398  }
1399  }
1400  // Write everything into the top layer.
1401  if (app_input.output_obu) {
1402  fwrite(pkt->data.frame.buf, 1, pkt->data.frame.sz,
1403  total_layer_obu_file);
1404  } else {
1405  aom_video_writer_write_frame(total_layer_file,
1406  pkt->data.frame.buf,
1407  pkt->data.frame.sz, pts);
1408  }
1409  // Keep count of rate control stats per layer (for non-key).
1410  if (!(pkt->data.frame.flags & AOM_FRAME_IS_KEY)) {
1411  unsigned int j = layer_id.spatial_layer_id * ts_number_layers +
1412  layer_id.temporal_layer_id;
1413  rc.layer_avg_frame_size[j] += 8.0 * pkt->data.frame.sz;
1414  rc.layer_avg_rate_mismatch[j] +=
1415  fabs(8.0 * pkt->data.frame.sz - rc.layer_pfb[j]) /
1416  rc.layer_pfb[j];
1417  if (slx == 0) ++rc.layer_enc_frames[layer_id.temporal_layer_id];
1418  }
1419 
1420  // Update for short-time encoding bitrate states, for moving window
1421  // of size rc->window, shifted by rc->window / 2.
1422  // Ignore first window segment, due to key frame.
1423  // For spatial layers: only do this for top/highest SL.
1424  if (frame_cnt > rc.window_size && slx == ss_number_layers - 1) {
1425  sum_bitrate += 0.001 * 8.0 * pkt->data.frame.sz * framerate;
1426  rc.window_size = (rc.window_size <= 0) ? 1 : rc.window_size;
1427  if (frame_cnt % rc.window_size == 0) {
1428  rc.window_count += 1;
1429  rc.avg_st_encoding_bitrate += sum_bitrate / rc.window_size;
1430  rc.variance_st_encoding_bitrate +=
1431  (sum_bitrate / rc.window_size) *
1432  (sum_bitrate / rc.window_size);
1433  sum_bitrate = 0.0;
1434  }
1435  }
1436  // Second shifted window.
1437  if (frame_cnt > rc.window_size + rc.window_size / 2 &&
1438  slx == ss_number_layers - 1) {
1439  sum_bitrate2 += 0.001 * 8.0 * pkt->data.frame.sz * framerate;
1440  if (frame_cnt > 2 * rc.window_size &&
1441  frame_cnt % rc.window_size == 0) {
1442  rc.window_count += 1;
1443  rc.avg_st_encoding_bitrate += sum_bitrate2 / rc.window_size;
1444  rc.variance_st_encoding_bitrate +=
1445  (sum_bitrate2 / rc.window_size) *
1446  (sum_bitrate2 / rc.window_size);
1447  sum_bitrate2 = 0.0;
1448  }
1449  }
1450 
1451 #if CONFIG_AV1_DECODER
1452  if (aom_codec_decode(&decoder, pkt->data.frame.buf,
1453  (unsigned int)pkt->data.frame.sz, NULL))
1454  die_codec(&decoder, "Failed to decode frame.");
1455 #endif
1456 
1457  break;
1458  default: break;
1459  }
1460  }
1461 #if CONFIG_AV1_DECODER
1462  // Don't look for mismatch on top spatial and top temporal layers as they
1463  // are non reference frames.
1464  if ((ss_number_layers > 1 || ts_number_layers > 1) &&
1465  !(layer_id.temporal_layer_id > 0 &&
1466  layer_id.temporal_layer_id == (int)ts_number_layers - 1)) {
1467  test_decode(&codec, &decoder, frame_cnt, &mismatch_seen);
1468  }
1469 #endif
1470  } // loop over spatial layers
1471  ++frame_cnt;
1472  pts += frame_duration;
1473  }
1474 
1475  close_input_file(&(app_input.input_ctx));
1476  printout_rate_control_summary(&rc, frame_cnt, ss_number_layers,
1477  ts_number_layers);
1478  printf("\n");
1479  printf("Frame cnt and encoding time/FPS stats for encoding: %d %f %f\n",
1480  frame_cnt, 1000 * (float)cx_time / (double)(frame_cnt * 1000000),
1481  1000000 * (double)frame_cnt / (double)cx_time);
1482 
1483  if (ss_number_layers > 1) {
1484  printf("Per spatial layer: \n");
1485  for (unsigned int slx = 0; slx < ss_number_layers; slx++)
1486  printf("Frame cnt and encoding time/FPS stats for encoding: %d %f %f\n",
1487  frame_cnt, (float)cx_time_sl[slx] / (double)(frame_cnt * 1000),
1488  1000000 * (double)frame_cnt / (double)cx_time_sl[slx]);
1489  }
1490 
1491  if (aom_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec");
1492 
1493 #if CONFIG_INTERNAL_STATS
1494  if (mismatch_seen) {
1495  fprintf(stats_file, "First mismatch occurred in frame %d\n", mismatch_seen);
1496  } else {
1497  fprintf(stats_file, "No mismatch detected in recon buffers\n");
1498  }
1499  fclose(stats_file);
1500 #endif
1501 
1502  // Try to rewrite the output file headers with the actual frame count.
1503  for (i = 0; i < ss_number_layers * ts_number_layers; ++i)
1504  aom_video_writer_close(outfile[i]);
1505  aom_video_writer_close(total_layer_file);
1506 
1507  if (app_input.input_ctx.file_type != FILE_TYPE_Y4M) {
1508  aom_img_free(&raw);
1509  }
1510  return EXIT_SUCCESS;
1511 }
Describes the encoder algorithm interface to applications.
enum aom_chroma_sample_position aom_chroma_sample_position_t
List of chroma sample positions.
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.
#define AOM_IMG_FMT_HIGHBITDEPTH
Definition: aom_image.h:38
@ AOM_IMG_FMT_I420
Definition: aom_image.h:45
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.
Declares top-level encoder structures and functions.
#define AOM_MAX_LAYERS
Definition: aomcx.h:1504
aom_codec_iface_t * aom_codec_av1_cx(void)
The interface to the AV1 encoder.
#define AOM_MAX_TS_LAYERS
Definition: aomcx.h:1506
@ AV1E_SET_ROW_MT
Codec control function to enable the row based multi-threading of the encoder, unsigned int parameter...
Definition: aomcx.h:359
@ AV1E_SET_ENABLE_TPL_MODEL
Codec control function to enable RDO modulated by frame temporal dependency, unsigned int parameter.
Definition: aomcx.h:406
@ AV1E_SET_AQ_MODE
Codec control function to set adaptive quantization mode, unsigned int parameter.
Definition: aomcx.h:466
@ AV1E_SET_SVC_LAYER_ID
Codec control function to set the layer id, aom_svc_layer_id_t* parameter.
Definition: aomcx.h:1270
@ AV1E_SET_SVC_REF_FRAME_CONFIG
Codec control function to set reference frame config: the ref_idx and the refresh flags for each buff...
Definition: aomcx.h:1281
@ AV1E_SET_CDF_UPDATE_MODE
Codec control function to set CDF update mode, unsigned int parameter.
Definition: aomcx.h:504
@ AV1E_SET_MV_COST_UPD_FREQ
Control to set frequency of the cost updates for motion vectors, unsigned int parameter.
Definition: aomcx.h:1248
@ 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:1383
@ AV1E_SET_ENABLE_WARPED_MOTION
Codec control function to turn on / off warped motion usage at sequence level, int parameter.
Definition: aomcx.h:1032
@ AV1E_SET_COEFF_COST_UPD_FREQ
Control to set frequency of the cost updates for coefficients, unsigned int parameter.
Definition: aomcx.h:1228
@ AV1E_SET_ENABLE_CDEF
Codec control function to encode with CDEF, unsigned int parameter.
Definition: aomcx.h:664
@ AV1E_SET_DV_COST_UPD_FREQ
Control to set frequency of the cost updates for intrabc motion vectors, unsigned int parameter.
Definition: aomcx.h:1352
@ AV1E_SET_SVC_PARAMS
Codec control function to set SVC paramaeters, aom_svc_params_t* parameter.
Definition: aomcx.h:1275
@ AOME_SET_MAX_INTRA_BITRATE_PCT
Codec control function to set max data rate for intra frames, unsigned int parameter.
Definition: aomcx.h:304
@ AV1E_SET_ERROR_RESILIENT_MODE
Codec control function to enable error_resilient_mode, int parameter.
Definition: aomcx.h:440
@ AV1E_SET_ENABLE_OBMC
Codec control function to predict with OBMC mode, unsigned int parameter.
Definition: aomcx.h:691
@ AV1E_SET_LOOPFILTER_CONTROL
Codec control to control loop filter.
Definition: aomcx.h:1398
@ AOME_SET_SCALEMODE
Codec control function to set encoder scaling mode, aom_scaling_mode_t* parameter.
Definition: aomcx.h:196
@ AV1E_SET_TILE_COLUMNS
Codec control function to set number of tile columns. unsigned int parameter.
Definition: aomcx.h:378
@ AV1E_SET_ENABLE_ORDER_HINT
Codec control function to turn on / off frame order hint (int parameter). Affects: joint compound mod...
Definition: aomcx.h:859
@ AV1E_SET_DELTAQ_MODE
Codec control function to set the delta q mode, unsigned int parameter.
Definition: aomcx.h:1125
@ AV1E_SET_ENABLE_GLOBAL_MOTION
Codec control function to turn on / off global motion usage for a sequence, int parameter.
Definition: aomcx.h:1022
@ AOME_SET_CPUUSED
Codec control function to set encoder internal speed settings, int parameter.
Definition: aomcx.h:218
@ AV1E_SET_GF_CBR_BOOST_PCT
Boost percentage for Golden Frame in CBR mode, unsigned int parameter.
Definition: aomcx.h:337
@ AV1E_SET_MODE_COST_UPD_FREQ
Control to set frequency of the cost updates for mode, unsigned int parameter.
Definition: aomcx.h:1238
@ 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.
aom_codec_err_t aom_codec_control(aom_codec_ctx_t *ctx, int ctrl_id,...)
Algorithm Control.
const struct aom_codec_iface aom_codec_iface_t
Codec interface structure.
Definition: aom_codec.h:254
const char * aom_codec_err_to_string(aom_codec_err_t err)
Convert error number to printable string.
aom_codec_err_t aom_codec_destroy(aom_codec_ctx_t *ctx)
Destroy a codec instance.
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:521
const void * aom_codec_iter_t
Iterator.
Definition: aom_codec.h:288
#define AOM_FRAME_IS_KEY
Definition: aom_codec.h:271
@ AOM_BITS_12
Definition: aom_codec.h:321
@ AOM_BITS_8
Definition: aom_codec.h:319
@ AOM_BITS_10
Definition: aom_codec.h:320
@ 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
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:950
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.
#define AOM_USAGE_REALTIME
usage parameter analogous to AV1 REALTIME mode.
Definition: aom_encoder.h:1023
const aom_codec_cx_pkt_t * aom_codec_get_cx_data(aom_codec_ctx_t *ctx, aom_codec_iter_t *iter)
Encoded data iterator.
@ AOM_CBR
Definition: aom_encoder.h:186
@ AOM_KF_AUTO
Definition: aom_encoder.h:201
@ AOM_CODEC_CX_FRAME_PKT
Definition: aom_encoder.h:109
Codec context structure.
Definition: aom_codec.h:298
aom_codec_err_t err
Definition: aom_codec.h:301
Encoder output packet.
Definition: aom_encoder.h:121
enum aom_codec_cx_pkt_kind kind
Definition: aom_encoder.h:122
union aom_codec_cx_pkt::@1 data
struct aom_codec_cx_pkt::@1::@2 frame
Encoder configuration structure.
Definition: aom_encoder.h:386
unsigned int g_input_bit_depth
Bit-depth of the input frames.
Definition: aom_encoder.h:469
unsigned int rc_dropframe_thresh
Temporal resampling configuration, if supported by the codec.
Definition: aom_encoder.h:534
struct aom_rational g_timebase
Stream timebase units.
Definition: aom_encoder.h:483
unsigned int g_usage
Algorithm specific "usage" value.
Definition: aom_encoder.h:398
unsigned int rc_buf_sz
Decoder Buffer Size.
Definition: aom_encoder.h:698
unsigned int g_h
Height of the frame.
Definition: aom_encoder.h:434
enum aom_kf_mode kf_mode
Keyframe placement mode.
Definition: aom_encoder.h:761
enum aom_rc_mode rc_end_usage
Rate control algorithm to use.
Definition: aom_encoder.h:617
unsigned int g_threads
Maximum number of threads to use.
Definition: aom_encoder.h:406
unsigned int kf_min_dist
Keyframe minimum interval.
Definition: aom_encoder.h:770
unsigned int g_lag_in_frames
Allow lagged encoding.
Definition: aom_encoder.h:512
unsigned int rc_buf_initial_sz
Decoder Buffer Initial Size.
Definition: aom_encoder.h:707
unsigned int g_profile
Bitstream profile to use.
Definition: aom_encoder.h:416
aom_bit_depth_t g_bit_depth
Bit-depth of the codec.
Definition: aom_encoder.h:461
unsigned int g_w
Width of the frame.
Definition: aom_encoder.h:425
unsigned int rc_undershoot_pct
Rate control adaptation undershoot control.
Definition: aom_encoder.h:674
unsigned int kf_max_dist
Keyframe maximum interval.
Definition: aom_encoder.h:779
aom_codec_er_flags_t g_error_resilient
Enable error resilient modes.
Definition: aom_encoder.h:491
unsigned int rc_max_quantizer
Maximum (Worst Quality) Quantizer.
Definition: aom_encoder.h:661
unsigned int rc_buf_optimal_sz
Decoder Buffer Optimal Size.
Definition: aom_encoder.h:716
unsigned int rc_min_quantizer
Minimum (Best Quality) Quantizer.
Definition: aom_encoder.h:651
unsigned int rc_target_bitrate
Target data rate.
Definition: aom_encoder.h:637
unsigned int rc_resize_mode
Mode for spatial resampling, if supported by the codec.
Definition: aom_encoder.h:543
unsigned int rc_overshoot_pct
Rate control adaptation overshoot control.
Definition: aom_encoder.h:683
Image Descriptor.
Definition: aom_image.h:171
aom_img_fmt_t fmt
Definition: aom_image.h:172
unsigned int d_w
Definition: aom_image.h:186
unsigned int d_h
Definition: aom_image.h:187
int num
Definition: aom_encoder.h:164
int den
Definition: aom_encoder.h:165
aom image scaling mode
Definition: aomcx.h:1468
Definition: aomcx.h:1509
int temporal_layer_id
Definition: aomcx.h:1511
int spatial_layer_id
Definition: aomcx.h:1510
Definition: aomcx.h:1515
int max_quantizers[32]
Definition: aomcx.h:1518
int number_spatial_layers
Definition: aomcx.h:1516
int layer_target_bitrate[32]
Definition: aomcx.h:1523
int framerate_factor[8]
Definition: aomcx.h:1525
int min_quantizers[32]
Definition: aomcx.h:1519
int scaling_factor_den[4]
Definition: aomcx.h:1521
int number_temporal_layers
Definition: aomcx.h:1517
int scaling_factor_num[4]
Definition: aomcx.h:1520
Definition: aomcx.h:1539
int use_comp_pred[3]
Definition: aomcx.h:1542
Definition: aomcx.h:1529
int reference[7]
Definition: aomcx.h:1532
int refresh[8]
Definition: aomcx.h:1535
int ref_idx[7]
Definition: aomcx.h:1534