WebM Codec SDK
vp9_spatial_svc_encoder
1 /*
2  * Copyright (c) 2012 The WebM project authors. 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 /*
12  * This is an example demonstrating how to implement a multi-layer
13  * VP9 encoding scheme based on spatial scalability for video applications
14  * that benefit from a scalable bitstream.
15  */
16 
17 #include <math.h>
18 #include <stdarg.h>
19 #include <stdlib.h>
20 #include <string.h>
21 #include <time.h>
22 
23 #include "../args.h"
24 #include "../tools_common.h"
25 #include "../video_writer.h"
26 
27 #include "../vpx_ports/vpx_timer.h"
28 #include "./svc_context.h"
29 #include "vpx/vp8cx.h"
30 #include "vpx/vpx_encoder.h"
31 #include "../vpxstats.h"
32 #include "vp9/encoder/vp9_encoder.h"
33 #include "./y4minput.h"
34 
35 #define OUTPUT_RC_STATS 1
36 
37 #define SIMULCAST_MODE 0
38 
39 static const arg_def_t outputfile =
40  ARG_DEF("o", "output", 1, "Output filename");
41 static const arg_def_t skip_frames_arg =
42  ARG_DEF("s", "skip-frames", 1, "input frames to skip");
43 static const arg_def_t frames_arg =
44  ARG_DEF("f", "frames", 1, "number of frames to encode");
45 static const arg_def_t threads_arg =
46  ARG_DEF("th", "threads", 1, "number of threads to use");
47 #if OUTPUT_RC_STATS
48 static const arg_def_t output_rc_stats_arg =
49  ARG_DEF("rcstat", "output_rc_stats", 1, "output rc stats");
50 #endif
51 static const arg_def_t width_arg = ARG_DEF("w", "width", 1, "source width");
52 static const arg_def_t height_arg = ARG_DEF("h", "height", 1, "source height");
53 static const arg_def_t timebase_arg =
54  ARG_DEF("t", "timebase", 1, "timebase (num/den)");
55 static const arg_def_t bitrate_arg = ARG_DEF(
56  "b", "target-bitrate", 1, "encoding bitrate, in kilobits per second");
57 static const arg_def_t spatial_layers_arg =
58  ARG_DEF("sl", "spatial-layers", 1, "number of spatial SVC layers");
59 static const arg_def_t temporal_layers_arg =
60  ARG_DEF("tl", "temporal-layers", 1, "number of temporal SVC layers");
61 static const arg_def_t temporal_layering_mode_arg =
62  ARG_DEF("tlm", "temporal-layering-mode", 1,
63  "temporal layering scheme."
64  "VP9E_TEMPORAL_LAYERING_MODE");
65 static const arg_def_t kf_dist_arg =
66  ARG_DEF("k", "kf-dist", 1, "number of frames between keyframes");
67 static const arg_def_t scale_factors_arg =
68  ARG_DEF("r", "scale-factors", 1, "scale factors (lowest to highest layer)");
69 static const arg_def_t min_q_arg =
70  ARG_DEF(NULL, "min-q", 1, "Minimum quantizer");
71 static const arg_def_t max_q_arg =
72  ARG_DEF(NULL, "max-q", 1, "Maximum quantizer");
73 static const arg_def_t min_bitrate_arg =
74  ARG_DEF(NULL, "min-bitrate", 1, "Minimum bitrate");
75 static const arg_def_t max_bitrate_arg =
76  ARG_DEF(NULL, "max-bitrate", 1, "Maximum bitrate");
77 static const arg_def_t lag_in_frame_arg =
78  ARG_DEF(NULL, "lag-in-frames", 1,
79  "Number of frame to input before "
80  "generating any outputs");
81 static const arg_def_t rc_end_usage_arg =
82  ARG_DEF(NULL, "rc-end-usage", 1, "0 - 3: VBR, CBR, CQ, Q");
83 static const arg_def_t speed_arg =
84  ARG_DEF("sp", "speed", 1, "speed configuration");
85 static const arg_def_t aqmode_arg =
86  ARG_DEF("aq", "aqmode", 1, "aq-mode off/on");
87 static const arg_def_t bitrates_arg =
88  ARG_DEF("bl", "bitrates", 1, "bitrates[sl * num_tl + tl]");
89 static const arg_def_t dropframe_thresh_arg =
90  ARG_DEF(NULL, "drop-frame", 1, "Temporal resampling threshold (buf %)");
91 static const struct arg_enum_list tune_content_enum[] = {
92  { "default", VP9E_CONTENT_DEFAULT },
93  { "screen", VP9E_CONTENT_SCREEN },
94  { "film", VP9E_CONTENT_FILM },
95  { NULL, 0 }
96 };
97 
98 static const arg_def_t tune_content_arg = ARG_DEF_ENUM(
99  NULL, "tune-content", 1, "Tune content type", tune_content_enum);
100 static const arg_def_t inter_layer_pred_arg = ARG_DEF(
101  NULL, "inter-layer-pred", 1, "0 - 3: On, Off, Key-frames, Constrained");
102 
103 #if CONFIG_VP9_HIGHBITDEPTH
104 static const struct arg_enum_list bitdepth_enum[] = {
105  { "8", VPX_BITS_8 }, { "10", VPX_BITS_10 }, { "12", VPX_BITS_12 }, { NULL, 0 }
106 };
107 
108 static const arg_def_t bitdepth_arg = ARG_DEF_ENUM(
109  "d", "bit-depth", 1, "Bit depth for codec 8, 10 or 12. ", bitdepth_enum);
110 #endif // CONFIG_VP9_HIGHBITDEPTH
111 
112 static const arg_def_t *svc_args[] = { &frames_arg,
113  &outputfile,
114  &width_arg,
115  &height_arg,
116  &timebase_arg,
117  &bitrate_arg,
118  &skip_frames_arg,
119  &spatial_layers_arg,
120  &kf_dist_arg,
121  &scale_factors_arg,
122  &min_q_arg,
123  &max_q_arg,
124  &min_bitrate_arg,
125  &max_bitrate_arg,
126  &temporal_layers_arg,
127  &temporal_layering_mode_arg,
128  &lag_in_frame_arg,
129  &threads_arg,
130  &aqmode_arg,
131 #if OUTPUT_RC_STATS
132  &output_rc_stats_arg,
133 #endif
134 
135 #if CONFIG_VP9_HIGHBITDEPTH
136  &bitdepth_arg,
137 #endif
138  &speed_arg,
139  &rc_end_usage_arg,
140  &bitrates_arg,
141  &dropframe_thresh_arg,
142  &tune_content_arg,
143  &inter_layer_pred_arg,
144  NULL };
145 
146 static const uint32_t default_frames_to_skip = 0;
147 static const uint32_t default_frames_to_code = 60 * 60;
148 static const uint32_t default_width = 1920;
149 static const uint32_t default_height = 1080;
150 static const uint32_t default_timebase_num = 1;
151 static const uint32_t default_timebase_den = 60;
152 static const uint32_t default_bitrate = 1000;
153 static const uint32_t default_spatial_layers = 5;
154 static const uint32_t default_temporal_layers = 1;
155 static const uint32_t default_kf_dist = 100;
156 static const uint32_t default_temporal_layering_mode = 0;
157 static const uint32_t default_output_rc_stats = 0;
158 static const int32_t default_speed = -1; // -1 means use library default.
159 static const uint32_t default_threads = 0; // zero means use library default.
160 
161 typedef struct {
162  const char *output_filename;
163  uint32_t frames_to_code;
164  uint32_t frames_to_skip;
165  struct VpxInputContext input_ctx;
166  stats_io_t rc_stats;
167  int tune_content;
168  int inter_layer_pred;
169 } AppInput;
170 
171 static const char *exec_name;
172 
173 void usage_exit(void) {
174  fprintf(stderr, "Usage: %s <options> input_filename -o output_filename\n",
175  exec_name);
176  fprintf(stderr, "Options:\n");
177  arg_show_usage(stderr, svc_args);
178  exit(EXIT_FAILURE);
179 }
180 
181 static void parse_command_line(int argc, const char **argv_,
182  AppInput *app_input, SvcContext *svc_ctx,
183  vpx_codec_enc_cfg_t *enc_cfg) {
184  struct arg arg;
185  char **argv = NULL;
186  char **argi = NULL;
187  char **argj = NULL;
188  vpx_codec_err_t res;
189  unsigned int min_bitrate = 0;
190  unsigned int max_bitrate = 0;
191  char string_options[1024] = { 0 };
192 
193  // initialize SvcContext with parameters that will be passed to vpx_svc_init
194  svc_ctx->log_level = SVC_LOG_DEBUG;
195  svc_ctx->spatial_layers = default_spatial_layers;
196  svc_ctx->temporal_layers = default_temporal_layers;
197  svc_ctx->temporal_layering_mode = default_temporal_layering_mode;
198 #if OUTPUT_RC_STATS
199  svc_ctx->output_rc_stat = default_output_rc_stats;
200 #endif
201  svc_ctx->speed = default_speed;
202  svc_ctx->threads = default_threads;
203 
204  // start with default encoder configuration
205  res = vpx_codec_enc_config_default(vpx_codec_vp9_cx(), enc_cfg, 0);
206  if (res) {
207  die("Failed to get config: %s\n", vpx_codec_err_to_string(res));
208  }
209  // update enc_cfg with app default values
210  enc_cfg->g_w = default_width;
211  enc_cfg->g_h = default_height;
212  enc_cfg->g_timebase.num = default_timebase_num;
213  enc_cfg->g_timebase.den = default_timebase_den;
214  enc_cfg->rc_target_bitrate = default_bitrate;
215  enc_cfg->kf_min_dist = default_kf_dist;
216  enc_cfg->kf_max_dist = default_kf_dist;
217  enc_cfg->rc_end_usage = VPX_CQ;
218 
219  // initialize AppInput with default values
220  app_input->frames_to_code = default_frames_to_code;
221  app_input->frames_to_skip = default_frames_to_skip;
222 
223  // process command line options
224  argv = argv_dup(argc - 1, argv_ + 1);
225  if (!argv) {
226  fprintf(stderr, "Error allocating argument list\n");
227  exit(EXIT_FAILURE);
228  }
229  for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) {
230  arg.argv_step = 1;
231 
232  if (arg_match(&arg, &frames_arg, argi)) {
233  app_input->frames_to_code = arg_parse_uint(&arg);
234  } else if (arg_match(&arg, &outputfile, argi)) {
235  app_input->output_filename = arg.val;
236  } else if (arg_match(&arg, &width_arg, argi)) {
237  enc_cfg->g_w = arg_parse_uint(&arg);
238  } else if (arg_match(&arg, &height_arg, argi)) {
239  enc_cfg->g_h = arg_parse_uint(&arg);
240  } else if (arg_match(&arg, &timebase_arg, argi)) {
241  enc_cfg->g_timebase = arg_parse_rational(&arg);
242  } else if (arg_match(&arg, &bitrate_arg, argi)) {
243  enc_cfg->rc_target_bitrate = arg_parse_uint(&arg);
244  } else if (arg_match(&arg, &skip_frames_arg, argi)) {
245  app_input->frames_to_skip = arg_parse_uint(&arg);
246  } else if (arg_match(&arg, &spatial_layers_arg, argi)) {
247  svc_ctx->spatial_layers = arg_parse_uint(&arg);
248  } else if (arg_match(&arg, &temporal_layers_arg, argi)) {
249  svc_ctx->temporal_layers = arg_parse_uint(&arg);
250 #if OUTPUT_RC_STATS
251  } else if (arg_match(&arg, &output_rc_stats_arg, argi)) {
252  svc_ctx->output_rc_stat = arg_parse_uint(&arg);
253 #endif
254  } else if (arg_match(&arg, &speed_arg, argi)) {
255  svc_ctx->speed = arg_parse_uint(&arg);
256  if (svc_ctx->speed > 9) {
257  warn("Mapping speed %d to speed 9.\n", svc_ctx->speed);
258  }
259  } else if (arg_match(&arg, &aqmode_arg, argi)) {
260  svc_ctx->aqmode = arg_parse_uint(&arg);
261  } else if (arg_match(&arg, &threads_arg, argi)) {
262  svc_ctx->threads = arg_parse_uint(&arg);
263  } else if (arg_match(&arg, &temporal_layering_mode_arg, argi)) {
264  svc_ctx->temporal_layering_mode = enc_cfg->temporal_layering_mode =
265  arg_parse_int(&arg);
266  if (svc_ctx->temporal_layering_mode) {
267  enc_cfg->g_error_resilient = 1;
268  }
269  } else if (arg_match(&arg, &kf_dist_arg, argi)) {
270  enc_cfg->kf_min_dist = arg_parse_uint(&arg);
271  enc_cfg->kf_max_dist = enc_cfg->kf_min_dist;
272  } else if (arg_match(&arg, &scale_factors_arg, argi)) {
273  strncat(string_options, " scale-factors=",
274  sizeof(string_options) - strlen(string_options) - 1);
275  strncat(string_options, arg.val,
276  sizeof(string_options) - strlen(string_options) - 1);
277  } else if (arg_match(&arg, &bitrates_arg, argi)) {
278  strncat(string_options, " bitrates=",
279  sizeof(string_options) - strlen(string_options) - 1);
280  strncat(string_options, arg.val,
281  sizeof(string_options) - strlen(string_options) - 1);
282  } else if (arg_match(&arg, &min_q_arg, argi)) {
283  strncat(string_options, " min-quantizers=",
284  sizeof(string_options) - strlen(string_options) - 1);
285  strncat(string_options, arg.val,
286  sizeof(string_options) - strlen(string_options) - 1);
287  } else if (arg_match(&arg, &max_q_arg, argi)) {
288  strncat(string_options, " max-quantizers=",
289  sizeof(string_options) - strlen(string_options) - 1);
290  strncat(string_options, arg.val,
291  sizeof(string_options) - strlen(string_options) - 1);
292  } else if (arg_match(&arg, &min_bitrate_arg, argi)) {
293  min_bitrate = arg_parse_uint(&arg);
294  } else if (arg_match(&arg, &max_bitrate_arg, argi)) {
295  max_bitrate = arg_parse_uint(&arg);
296  } else if (arg_match(&arg, &lag_in_frame_arg, argi)) {
297  enc_cfg->g_lag_in_frames = arg_parse_uint(&arg);
298  } else if (arg_match(&arg, &rc_end_usage_arg, argi)) {
299  enc_cfg->rc_end_usage = arg_parse_uint(&arg);
300 #if CONFIG_VP9_HIGHBITDEPTH
301  } else if (arg_match(&arg, &bitdepth_arg, argi)) {
302  enc_cfg->g_bit_depth = arg_parse_enum_or_int(&arg);
303  switch (enc_cfg->g_bit_depth) {
304  case VPX_BITS_8:
305  enc_cfg->g_input_bit_depth = 8;
306  enc_cfg->g_profile = 0;
307  break;
308  case VPX_BITS_10:
309  enc_cfg->g_input_bit_depth = 10;
310  enc_cfg->g_profile = 2;
311  break;
312  case VPX_BITS_12:
313  enc_cfg->g_input_bit_depth = 12;
314  enc_cfg->g_profile = 2;
315  break;
316  default:
317  die("Error: Invalid bit depth selected (%d)\n", enc_cfg->g_bit_depth);
318  break;
319  }
320 #endif // CONFIG_VP9_HIGHBITDEPTH
321  } else if (arg_match(&arg, &dropframe_thresh_arg, argi)) {
322  enc_cfg->rc_dropframe_thresh = arg_parse_uint(&arg);
323  } else if (arg_match(&arg, &tune_content_arg, argi)) {
324  app_input->tune_content = arg_parse_uint(&arg);
325  } else if (arg_match(&arg, &inter_layer_pred_arg, argi)) {
326  app_input->inter_layer_pred = arg_parse_uint(&arg);
327  } else {
328  ++argj;
329  }
330  }
331 
332  // There will be a space in front of the string options
333  if (strlen(string_options) > 0)
334  vpx_svc_set_options(svc_ctx, string_options + 1);
335 
336  enc_cfg->g_pass = VPX_RC_ONE_PASS;
337 
338  if (enc_cfg->rc_target_bitrate > 0) {
339  if (min_bitrate > 0) {
340  enc_cfg->rc_2pass_vbr_minsection_pct =
341  min_bitrate * 100 / enc_cfg->rc_target_bitrate;
342  }
343  if (max_bitrate > 0) {
344  enc_cfg->rc_2pass_vbr_maxsection_pct =
345  max_bitrate * 100 / enc_cfg->rc_target_bitrate;
346  }
347  }
348 
349  // Check for unrecognized options
350  for (argi = argv; *argi; ++argi)
351  if (argi[0][0] == '-' && strlen(argi[0]) > 1)
352  die("Error: Unrecognized option %s\n", *argi);
353 
354  if (argv[0] == NULL) {
355  usage_exit();
356  }
357  app_input->input_ctx.filename = argv[0];
358  free(argv);
359 
360  open_input_file(&app_input->input_ctx);
361  if (app_input->input_ctx.file_type == FILE_TYPE_Y4M) {
362  enc_cfg->g_w = app_input->input_ctx.width;
363  enc_cfg->g_h = app_input->input_ctx.height;
364  enc_cfg->g_timebase.den = app_input->input_ctx.framerate.numerator;
365  enc_cfg->g_timebase.num = app_input->input_ctx.framerate.denominator;
366  }
367 
368  if (enc_cfg->g_w < 16 || enc_cfg->g_w % 2 || enc_cfg->g_h < 16 ||
369  enc_cfg->g_h % 2)
370  die("Invalid resolution: %d x %d\n", enc_cfg->g_w, enc_cfg->g_h);
371 
372  printf(
373  "Codec %s\nframes: %d, skip: %d\n"
374  "layers: %d\n"
375  "width %d, height: %d,\n"
376  "num: %d, den: %d, bitrate: %d,\n"
377  "gop size: %d\n",
378  vpx_codec_iface_name(vpx_codec_vp9_cx()), app_input->frames_to_code,
379  app_input->frames_to_skip, svc_ctx->spatial_layers, enc_cfg->g_w,
380  enc_cfg->g_h, enc_cfg->g_timebase.num, enc_cfg->g_timebase.den,
381  enc_cfg->rc_target_bitrate, enc_cfg->kf_max_dist);
382 }
383 
384 #if OUTPUT_RC_STATS
385 // For rate control encoding stats.
386 struct RateControlStats {
387  // Number of input frames per layer.
388  int layer_input_frames[VPX_MAX_LAYERS];
389  // Total (cumulative) number of encoded frames per layer.
390  int layer_tot_enc_frames[VPX_MAX_LAYERS];
391  // Number of encoded non-key frames per layer.
392  int layer_enc_frames[VPX_MAX_LAYERS];
393  // Framerate per layer (cumulative).
394  double layer_framerate[VPX_MAX_LAYERS];
395  // Target average frame size per layer (per-frame-bandwidth per layer).
396  double layer_pfb[VPX_MAX_LAYERS];
397  // Actual average frame size per layer.
398  double layer_avg_frame_size[VPX_MAX_LAYERS];
399  // Average rate mismatch per layer (|target - actual| / target).
400  double layer_avg_rate_mismatch[VPX_MAX_LAYERS];
401  // Actual encoding bitrate per layer (cumulative).
402  double layer_encoding_bitrate[VPX_MAX_LAYERS];
403  // Average of the short-time encoder actual bitrate.
404  // TODO(marpan): Should we add these short-time stats for each layer?
405  double avg_st_encoding_bitrate;
406  // Variance of the short-time encoder actual bitrate.
407  double variance_st_encoding_bitrate;
408  // Window (number of frames) for computing short-time encoding bitrate.
409  int window_size;
410  // Number of window measurements.
411  int window_count;
412 };
413 
414 // Note: these rate control stats assume only 1 key frame in the
415 // sequence (i.e., first frame only).
416 static void set_rate_control_stats(struct RateControlStats *rc,
417  vpx_codec_enc_cfg_t *cfg) {
418  unsigned int sl, tl;
419  // Set the layer (cumulative) framerate and the target layer (non-cumulative)
420  // per-frame-bandwidth, for the rate control encoding stats below.
421  const double framerate = cfg->g_timebase.den / cfg->g_timebase.num;
422 
423  for (sl = 0; sl < cfg->ss_number_layers; ++sl) {
424  for (tl = 0; tl < cfg->ts_number_layers; ++tl) {
425  const int layer = sl * cfg->ts_number_layers + tl;
426  if (cfg->ts_number_layers == 1)
427  rc->layer_framerate[layer] = framerate;
428  else
429  rc->layer_framerate[layer] = framerate / cfg->ts_rate_decimator[tl];
430  if (tl > 0) {
431  rc->layer_pfb[layer] =
432  1000.0 *
433  (cfg->layer_target_bitrate[layer] -
434  cfg->layer_target_bitrate[layer - 1]) /
435  (rc->layer_framerate[layer] - rc->layer_framerate[layer - 1]);
436  } else {
437  rc->layer_pfb[layer] = 1000.0 * cfg->layer_target_bitrate[layer] /
438  rc->layer_framerate[layer];
439  }
440  rc->layer_input_frames[layer] = 0;
441  rc->layer_enc_frames[layer] = 0;
442  rc->layer_tot_enc_frames[layer] = 0;
443  rc->layer_encoding_bitrate[layer] = 0.0;
444  rc->layer_avg_frame_size[layer] = 0.0;
445  rc->layer_avg_rate_mismatch[layer] = 0.0;
446  }
447  }
448  rc->window_count = 0;
449  rc->window_size = 15;
450  rc->avg_st_encoding_bitrate = 0.0;
451  rc->variance_st_encoding_bitrate = 0.0;
452 }
453 
454 static void printout_rate_control_summary(struct RateControlStats *rc,
455  vpx_codec_enc_cfg_t *cfg,
456  int frame_cnt) {
457  unsigned int sl, tl;
458  double perc_fluctuation = 0.0;
459  int tot_num_frames = 0;
460  printf("Total number of processed frames: %d\n\n", frame_cnt - 1);
461  printf("Rate control layer stats for sl%d tl%d layer(s):\n\n",
463  for (sl = 0; sl < cfg->ss_number_layers; ++sl) {
464  tot_num_frames = 0;
465  for (tl = 0; tl < cfg->ts_number_layers; ++tl) {
466  const int layer = sl * cfg->ts_number_layers + tl;
467  const int num_dropped =
468  (tl > 0)
469  ? (rc->layer_input_frames[layer] - rc->layer_enc_frames[layer])
470  : (rc->layer_input_frames[layer] - rc->layer_enc_frames[layer] -
471  1);
472  tot_num_frames += rc->layer_input_frames[layer];
473  rc->layer_encoding_bitrate[layer] = 0.001 * rc->layer_framerate[layer] *
474  rc->layer_encoding_bitrate[layer] /
475  tot_num_frames;
476  rc->layer_avg_frame_size[layer] =
477  rc->layer_avg_frame_size[layer] / rc->layer_enc_frames[layer];
478  rc->layer_avg_rate_mismatch[layer] = 100.0 *
479  rc->layer_avg_rate_mismatch[layer] /
480  rc->layer_enc_frames[layer];
481  printf("For layer#: sl%d tl%d \n", sl, tl);
482  printf("Bitrate (target vs actual): %d %f.0 kbps\n",
483  cfg->layer_target_bitrate[layer],
484  rc->layer_encoding_bitrate[layer]);
485  printf("Average frame size (target vs actual): %f %f bits\n",
486  rc->layer_pfb[layer], rc->layer_avg_frame_size[layer]);
487  printf("Average rate_mismatch: %f\n", rc->layer_avg_rate_mismatch[layer]);
488  printf(
489  "Number of input frames, encoded (non-key) frames, "
490  "and percent dropped frames: %d %d %f.0 \n",
491  rc->layer_input_frames[layer], rc->layer_enc_frames[layer],
492  100.0 * num_dropped / rc->layer_input_frames[layer]);
493  printf("\n");
494  }
495  }
496  rc->avg_st_encoding_bitrate = rc->avg_st_encoding_bitrate / rc->window_count;
497  rc->variance_st_encoding_bitrate =
498  rc->variance_st_encoding_bitrate / rc->window_count -
499  (rc->avg_st_encoding_bitrate * rc->avg_st_encoding_bitrate);
500  perc_fluctuation = 100.0 * sqrt(rc->variance_st_encoding_bitrate) /
501  rc->avg_st_encoding_bitrate;
502  printf("Short-time stats, for window of %d frames: \n", rc->window_size);
503  printf("Average, rms-variance, and percent-fluct: %f %f %f \n",
504  rc->avg_st_encoding_bitrate, sqrt(rc->variance_st_encoding_bitrate),
505  perc_fluctuation);
506  printf("Num of input, num of encoded (super) frames: %d %d \n", frame_cnt,
507  tot_num_frames);
508 }
509 
510 static vpx_codec_err_t parse_superframe_index(const uint8_t *data,
511  size_t data_sz, uint64_t sizes[8],
512  int *count) {
513  // A chunk ending with a byte matching 0xc0 is an invalid chunk unless
514  // it is a super frame index. If the last byte of real video compression
515  // data is 0xc0 the encoder must add a 0 byte. If we have the marker but
516  // not the associated matching marker byte at the front of the index we have
517  // an invalid bitstream and need to return an error.
518 
519  uint8_t marker;
520 
521  marker = *(data + data_sz - 1);
522  *count = 0;
523 
524  if ((marker & 0xe0) == 0xc0) {
525  const uint32_t frames = (marker & 0x7) + 1;
526  const uint32_t mag = ((marker >> 3) & 0x3) + 1;
527  const size_t index_sz = 2 + mag * frames;
528 
529  // This chunk is marked as having a superframe index but doesn't have
530  // enough data for it, thus it's an invalid superframe index.
531  if (data_sz < index_sz) return VPX_CODEC_CORRUPT_FRAME;
532 
533  {
534  const uint8_t marker2 = *(data + data_sz - index_sz);
535 
536  // This chunk is marked as having a superframe index but doesn't have
537  // the matching marker byte at the front of the index therefore it's an
538  // invalid chunk.
539  if (marker != marker2) return VPX_CODEC_CORRUPT_FRAME;
540  }
541 
542  {
543  // Found a valid superframe index.
544  uint32_t i, j;
545  const uint8_t *x = &data[data_sz - index_sz + 1];
546 
547  for (i = 0; i < frames; ++i) {
548  uint32_t this_sz = 0;
549 
550  for (j = 0; j < mag; ++j) this_sz |= (*x++) << (j * 8);
551  sizes[i] = this_sz;
552  }
553  *count = frames;
554  }
555  }
556  return VPX_CODEC_OK;
557 }
558 #endif
559 
560 // Example pattern for spatial layers and 2 temporal layers used in the
561 // bypass/flexible mode. The pattern corresponds to the pattern
562 // VP9E_TEMPORAL_LAYERING_MODE_0101 (temporal_layering_mode == 2) used in
563 // non-flexible mode.
564 static void set_frame_flags_bypass_mode_ex0(
565  int tl, int num_spatial_layers, int is_key_frame,
566  vpx_svc_ref_frame_config_t *ref_frame_config) {
567  int sl;
568  for (sl = 0; sl < num_spatial_layers; ++sl)
569  ref_frame_config->update_buffer_slot[sl] = 0;
570 
571  for (sl = 0; sl < num_spatial_layers; ++sl) {
572  // Set the buffer idx.
573  if (tl == 0) {
574  ref_frame_config->lst_fb_idx[sl] = sl;
575  if (sl) {
576  if (is_key_frame) {
577  ref_frame_config->lst_fb_idx[sl] = sl - 1;
578  ref_frame_config->gld_fb_idx[sl] = sl;
579  } else {
580  ref_frame_config->gld_fb_idx[sl] = sl - 1;
581  }
582  } else {
583  ref_frame_config->gld_fb_idx[sl] = 0;
584  }
585  ref_frame_config->alt_fb_idx[sl] = 0;
586  } else if (tl == 1) {
587  ref_frame_config->lst_fb_idx[sl] = sl;
588  ref_frame_config->gld_fb_idx[sl] =
589  (sl == 0) ? 0 : num_spatial_layers + sl - 1;
590  ref_frame_config->alt_fb_idx[sl] = num_spatial_layers + sl;
591  }
592  // Set the reference and update flags.
593  if (!tl) {
594  if (!sl) {
595  // Base spatial and base temporal (sl = 0, tl = 0)
596  ref_frame_config->reference_last[sl] = 1;
597  ref_frame_config->reference_golden[sl] = 0;
598  ref_frame_config->reference_alt_ref[sl] = 0;
599  ref_frame_config->update_buffer_slot[sl] |=
600  1 << ref_frame_config->lst_fb_idx[sl];
601  } else {
602  if (is_key_frame) {
603  ref_frame_config->reference_last[sl] = 1;
604  ref_frame_config->reference_golden[sl] = 0;
605  ref_frame_config->reference_alt_ref[sl] = 0;
606  ref_frame_config->update_buffer_slot[sl] |=
607  1 << ref_frame_config->gld_fb_idx[sl];
608  } else {
609  // Non-zero spatiall layer.
610  ref_frame_config->reference_last[sl] = 1;
611  ref_frame_config->reference_golden[sl] = 1;
612  ref_frame_config->reference_alt_ref[sl] = 1;
613  ref_frame_config->update_buffer_slot[sl] |=
614  1 << ref_frame_config->lst_fb_idx[sl];
615  }
616  }
617  } else if (tl == 1) {
618  if (!sl) {
619  // Base spatial and top temporal (tl = 1)
620  ref_frame_config->reference_last[sl] = 1;
621  ref_frame_config->reference_golden[sl] = 0;
622  ref_frame_config->reference_alt_ref[sl] = 0;
623  ref_frame_config->update_buffer_slot[sl] |=
624  1 << ref_frame_config->alt_fb_idx[sl];
625  } else {
626  // Non-zero spatial.
627  if (sl < num_spatial_layers - 1) {
628  ref_frame_config->reference_last[sl] = 1;
629  ref_frame_config->reference_golden[sl] = 1;
630  ref_frame_config->reference_alt_ref[sl] = 0;
631  ref_frame_config->update_buffer_slot[sl] |=
632  1 << ref_frame_config->alt_fb_idx[sl];
633  } else if (sl == num_spatial_layers - 1) {
634  // Top spatial and top temporal (non-reference -- doesn't update any
635  // reference buffers)
636  ref_frame_config->reference_last[sl] = 1;
637  ref_frame_config->reference_golden[sl] = 1;
638  ref_frame_config->reference_alt_ref[sl] = 0;
639  }
640  }
641  }
642  }
643 }
644 
645 // Example pattern for 2 spatial layers and 2 temporal layers used in the
646 // bypass/flexible mode, except only 1 spatial layer when temporal_layer_id = 1.
647 static void set_frame_flags_bypass_mode_ex1(
648  int tl, int num_spatial_layers, int is_key_frame,
649  vpx_svc_ref_frame_config_t *ref_frame_config) {
650  int sl;
651  for (sl = 0; sl < num_spatial_layers; ++sl)
652  ref_frame_config->update_buffer_slot[sl] = 0;
653 
654  if (tl == 0) {
655  if (is_key_frame) {
656  ref_frame_config->lst_fb_idx[1] = 0;
657  ref_frame_config->gld_fb_idx[1] = 1;
658  } else {
659  ref_frame_config->lst_fb_idx[1] = 1;
660  ref_frame_config->gld_fb_idx[1] = 0;
661  }
662  ref_frame_config->alt_fb_idx[1] = 0;
663 
664  ref_frame_config->lst_fb_idx[0] = 0;
665  ref_frame_config->gld_fb_idx[0] = 0;
666  ref_frame_config->alt_fb_idx[0] = 0;
667  }
668  if (tl == 1) {
669  ref_frame_config->lst_fb_idx[0] = 0;
670  ref_frame_config->gld_fb_idx[0] = 1;
671  ref_frame_config->alt_fb_idx[0] = 2;
672 
673  ref_frame_config->lst_fb_idx[1] = 1;
674  ref_frame_config->gld_fb_idx[1] = 2;
675  ref_frame_config->alt_fb_idx[1] = 3;
676  }
677  // Set the reference and update flags.
678  if (tl == 0) {
679  // Base spatial and base temporal (sl = 0, tl = 0)
680  ref_frame_config->reference_last[0] = 1;
681  ref_frame_config->reference_golden[0] = 0;
682  ref_frame_config->reference_alt_ref[0] = 0;
683  ref_frame_config->update_buffer_slot[0] |=
684  1 << ref_frame_config->lst_fb_idx[0];
685 
686  if (is_key_frame) {
687  ref_frame_config->reference_last[1] = 1;
688  ref_frame_config->reference_golden[1] = 0;
689  ref_frame_config->reference_alt_ref[1] = 0;
690  ref_frame_config->update_buffer_slot[1] |=
691  1 << ref_frame_config->gld_fb_idx[1];
692  } else {
693  // Non-zero spatiall layer.
694  ref_frame_config->reference_last[1] = 1;
695  ref_frame_config->reference_golden[1] = 1;
696  ref_frame_config->reference_alt_ref[1] = 1;
697  ref_frame_config->update_buffer_slot[1] |=
698  1 << ref_frame_config->lst_fb_idx[1];
699  }
700  }
701  if (tl == 1) {
702  // Top spatial and top temporal (non-reference -- doesn't update any
703  // reference buffers)
704  ref_frame_config->reference_last[1] = 1;
705  ref_frame_config->reference_golden[1] = 0;
706  ref_frame_config->reference_alt_ref[1] = 0;
707  }
708 }
709 
710 #if CONFIG_VP9_DECODER && !SIMULCAST_MODE
711 static void test_decode(vpx_codec_ctx_t *encoder, vpx_codec_ctx_t *decoder,
712  const int frames_out, int *mismatch_seen) {
713  vpx_image_t enc_img, dec_img;
714  struct vp9_ref_frame ref_enc, ref_dec;
715  if (*mismatch_seen) return;
716  /* Get the internal reference frame */
717  ref_enc.idx = 0;
718  ref_dec.idx = 0;
719  vpx_codec_control(encoder, VP9_GET_REFERENCE, &ref_enc);
720  enc_img = ref_enc.img;
721  vpx_codec_control(decoder, VP9_GET_REFERENCE, &ref_dec);
722  dec_img = ref_dec.img;
723 #if CONFIG_VP9_HIGHBITDEPTH
724  if ((enc_img.fmt & VPX_IMG_FMT_HIGHBITDEPTH) !=
725  (dec_img.fmt & VPX_IMG_FMT_HIGHBITDEPTH)) {
726  if (enc_img.fmt & VPX_IMG_FMT_HIGHBITDEPTH) {
727  vpx_img_alloc(&enc_img, enc_img.fmt - VPX_IMG_FMT_HIGHBITDEPTH,
728  enc_img.d_w, enc_img.d_h, 16);
729  vpx_img_truncate_16_to_8(&enc_img, &ref_enc.img);
730  }
731  if (dec_img.fmt & VPX_IMG_FMT_HIGHBITDEPTH) {
732  vpx_img_alloc(&dec_img, dec_img.fmt - VPX_IMG_FMT_HIGHBITDEPTH,
733  dec_img.d_w, dec_img.d_h, 16);
734  vpx_img_truncate_16_to_8(&dec_img, &ref_dec.img);
735  }
736  }
737 #endif
738 
739  if (!compare_img(&enc_img, &dec_img)) {
740  int y[4], u[4], v[4];
741 #if CONFIG_VP9_HIGHBITDEPTH
742  if (enc_img.fmt & VPX_IMG_FMT_HIGHBITDEPTH) {
743  find_mismatch_high(&enc_img, &dec_img, y, u, v);
744  } else {
745  find_mismatch(&enc_img, &dec_img, y, u, v);
746  }
747 #else
748  find_mismatch(&enc_img, &dec_img, y, u, v);
749 #endif
750  decoder->err = 1;
751  printf(
752  "Encode/decode mismatch on frame %d at"
753  " Y[%d, %d] {%d/%d},"
754  " U[%d, %d] {%d/%d},"
755  " V[%d, %d] {%d/%d}\n",
756  frames_out, y[0], y[1], y[2], y[3], u[0], u[1], u[2], u[3], v[0], v[1],
757  v[2], v[3]);
758  *mismatch_seen = frames_out;
759  }
760 
761  vpx_img_free(&enc_img);
762  vpx_img_free(&dec_img);
763 }
764 #endif
765 
766 #if OUTPUT_RC_STATS
767 static void svc_output_rc_stats(
768  vpx_codec_ctx_t *codec, vpx_codec_enc_cfg_t *enc_cfg,
769  vpx_svc_layer_id_t *layer_id, const vpx_codec_cx_pkt_t *cx_pkt,
770  struct RateControlStats *rc, VpxVideoWriter **outfile,
771  const uint32_t frame_cnt, const double framerate) {
772  int num_layers_encoded = 0;
773  unsigned int sl, tl;
774  uint64_t sizes[8];
775  uint64_t sizes_parsed[8];
776  int count = 0;
777  double sum_bitrate = 0.0;
778  double sum_bitrate2 = 0.0;
779  vp9_zero(sizes);
780  vp9_zero(sizes_parsed);
781  vpx_codec_control(codec, VP9E_GET_SVC_LAYER_ID, layer_id);
782  parse_superframe_index(cx_pkt->data.frame.buf, cx_pkt->data.frame.sz,
783  sizes_parsed, &count);
784  if (enc_cfg->ss_number_layers == 1) {
785  sizes[0] = cx_pkt->data.frame.sz;
786  } else {
787  for (sl = 0; sl < enc_cfg->ss_number_layers; ++sl) {
788  sizes[sl] = 0;
789  if (cx_pkt->data.frame.spatial_layer_encoded[sl]) {
790  sizes[sl] = sizes_parsed[num_layers_encoded];
791  num_layers_encoded++;
792  }
793  }
794  }
795  for (sl = 0; sl < enc_cfg->ss_number_layers; ++sl) {
796  unsigned int sl2;
797  uint64_t tot_size = 0;
798 #if SIMULCAST_MODE
799  for (sl2 = 0; sl2 < sl; ++sl2) {
800  if (cx_pkt->data.frame.spatial_layer_encoded[sl2]) tot_size += sizes[sl2];
801  }
802  vpx_video_writer_write_frame(outfile[sl],
803  (uint8_t *)(cx_pkt->data.frame.buf) + tot_size,
804  (size_t)(sizes[sl]), cx_pkt->data.frame.pts);
805 #else
806  for (sl2 = 0; sl2 <= sl; ++sl2) {
807  if (cx_pkt->data.frame.spatial_layer_encoded[sl2]) tot_size += sizes[sl2];
808  }
809  if (tot_size > 0)
810  vpx_video_writer_write_frame(outfile[sl], cx_pkt->data.frame.buf,
811  (size_t)(tot_size), cx_pkt->data.frame.pts);
812 #endif // SIMULCAST_MODE
813  }
814  for (sl = 0; sl < enc_cfg->ss_number_layers; ++sl) {
815  if (cx_pkt->data.frame.spatial_layer_encoded[sl]) {
816  for (tl = layer_id->temporal_layer_id; tl < enc_cfg->ts_number_layers;
817  ++tl) {
818  const int layer = sl * enc_cfg->ts_number_layers + tl;
819  ++rc->layer_tot_enc_frames[layer];
820  rc->layer_encoding_bitrate[layer] += 8.0 * sizes[sl];
821  // Keep count of rate control stats per layer, for non-key
822  // frames.
823  if (tl == (unsigned int)layer_id->temporal_layer_id &&
824  !(cx_pkt->data.frame.flags & VPX_FRAME_IS_KEY)) {
825  rc->layer_avg_frame_size[layer] += 8.0 * sizes[sl];
826  rc->layer_avg_rate_mismatch[layer] +=
827  fabs(8.0 * sizes[sl] - rc->layer_pfb[layer]) /
828  rc->layer_pfb[layer];
829  ++rc->layer_enc_frames[layer];
830  }
831  }
832  }
833  }
834 
835  // Update for short-time encoding bitrate states, for moving
836  // window of size rc->window, shifted by rc->window / 2.
837  // Ignore first window segment, due to key frame.
838  if (frame_cnt > (unsigned int)rc->window_size) {
839  for (sl = 0; sl < enc_cfg->ss_number_layers; ++sl) {
840  if (cx_pkt->data.frame.spatial_layer_encoded[sl])
841  sum_bitrate += 0.001 * 8.0 * sizes[sl] * framerate;
842  }
843  if (frame_cnt % rc->window_size == 0) {
844  rc->window_count += 1;
845  rc->avg_st_encoding_bitrate += sum_bitrate / rc->window_size;
846  rc->variance_st_encoding_bitrate +=
847  (sum_bitrate / rc->window_size) * (sum_bitrate / rc->window_size);
848  }
849  }
850 
851  // Second shifted window.
852  if (frame_cnt > (unsigned int)(rc->window_size + rc->window_size / 2)) {
853  for (sl = 0; sl < enc_cfg->ss_number_layers; ++sl) {
854  sum_bitrate2 += 0.001 * 8.0 * sizes[sl] * framerate;
855  }
856 
857  if (frame_cnt > (unsigned int)(2 * rc->window_size) &&
858  frame_cnt % rc->window_size == 0) {
859  rc->window_count += 1;
860  rc->avg_st_encoding_bitrate += sum_bitrate2 / rc->window_size;
861  rc->variance_st_encoding_bitrate +=
862  (sum_bitrate2 / rc->window_size) * (sum_bitrate2 / rc->window_size);
863  }
864  }
865 }
866 #endif
867 
868 int main(int argc, const char **argv) {
869  AppInput app_input;
870  VpxVideoWriter *writer = NULL;
871  VpxVideoInfo info;
872  vpx_codec_ctx_t encoder;
873  vpx_codec_enc_cfg_t enc_cfg;
874  SvcContext svc_ctx;
875  vpx_svc_frame_drop_t svc_drop_frame;
876  uint32_t i;
877  uint32_t frame_cnt = 0;
878  vpx_image_t raw;
879  vpx_codec_err_t res;
880  int pts = 0; /* PTS starts at 0 */
881  int frame_duration = 1; /* 1 timebase tick per frame */
882  int end_of_stream = 0;
883  int frames_received = 0;
884 #if OUTPUT_RC_STATS
885  VpxVideoWriter *outfile[VPX_SS_MAX_LAYERS] = { NULL };
886  struct RateControlStats rc;
887  vpx_svc_layer_id_t layer_id;
888  vpx_svc_ref_frame_config_t ref_frame_config;
889  unsigned int sl;
890  double framerate = 30.0;
891 #endif
892  struct vpx_usec_timer timer;
893  int64_t cx_time = 0;
894 #if CONFIG_INTERNAL_STATS
895  FILE *f = fopen("opsnr.stt", "a");
896 #endif
897 #if CONFIG_VP9_DECODER && !SIMULCAST_MODE
898  int mismatch_seen = 0;
899  vpx_codec_ctx_t decoder;
900 #endif
901  memset(&svc_ctx, 0, sizeof(svc_ctx));
902  memset(&app_input, 0, sizeof(AppInput));
903  memset(&info, 0, sizeof(VpxVideoInfo));
904  memset(&layer_id, 0, sizeof(vpx_svc_layer_id_t));
905  memset(&rc, 0, sizeof(struct RateControlStats));
906  exec_name = argv[0];
907 
908  /* Setup default input stream settings */
909  app_input.input_ctx.framerate.numerator = 30;
910  app_input.input_ctx.framerate.denominator = 1;
911  app_input.input_ctx.only_i420 = 1;
912  app_input.input_ctx.bit_depth = 0;
913 
914  parse_command_line(argc, argv, &app_input, &svc_ctx, &enc_cfg);
915 
916  // Y4M reader handles its own allocation.
917  if (app_input.input_ctx.file_type != FILE_TYPE_Y4M) {
918 // Allocate image buffer
919 #if CONFIG_VP9_HIGHBITDEPTH
920  if (!vpx_img_alloc(&raw,
921  enc_cfg.g_input_bit_depth == 8 ? VPX_IMG_FMT_I420
923  enc_cfg.g_w, enc_cfg.g_h, 32)) {
924  die("Failed to allocate image %dx%d\n", enc_cfg.g_w, enc_cfg.g_h);
925  }
926 #else
927  if (!vpx_img_alloc(&raw, VPX_IMG_FMT_I420, enc_cfg.g_w, enc_cfg.g_h, 32)) {
928  die("Failed to allocate image %dx%d\n", enc_cfg.g_w, enc_cfg.g_h);
929  }
930 #endif // CONFIG_VP9_HIGHBITDEPTH
931  }
932 
933  // Initialize codec
934  if (vpx_svc_init(&svc_ctx, &encoder, vpx_codec_vp9_cx(), &enc_cfg) !=
935  VPX_CODEC_OK)
936  die("Failed to initialize encoder\n");
937 #if CONFIG_VP9_DECODER && !SIMULCAST_MODE
938  if (vpx_codec_dec_init(
939  &decoder, get_vpx_decoder_by_name("vp9")->codec_interface(), NULL, 0))
940  die("Failed to initialize decoder\n");
941 #endif
942 
943 #if OUTPUT_RC_STATS
944  rc.window_count = 1;
945  rc.window_size = 15; // Silence a static analysis warning.
946  rc.avg_st_encoding_bitrate = 0.0;
947  rc.variance_st_encoding_bitrate = 0.0;
948  if (svc_ctx.output_rc_stat) {
949  set_rate_control_stats(&rc, &enc_cfg);
950  framerate = enc_cfg.g_timebase.den / enc_cfg.g_timebase.num;
951  }
952 #endif
953 
954  info.codec_fourcc = VP9_FOURCC;
955  info.frame_width = enc_cfg.g_w;
956  info.frame_height = enc_cfg.g_h;
957  info.time_base.numerator = enc_cfg.g_timebase.num;
958  info.time_base.denominator = enc_cfg.g_timebase.den;
959 
960  writer =
961  vpx_video_writer_open(app_input.output_filename, kContainerIVF, &info);
962  if (!writer)
963  die("Failed to open %s for writing\n", app_input.output_filename);
964 
965 #if OUTPUT_RC_STATS
966  // Write out spatial layer stream.
967  // TODO(marpan/jianj): allow for writing each spatial and temporal stream.
968  if (svc_ctx.output_rc_stat) {
969  for (sl = 0; sl < enc_cfg.ss_number_layers; ++sl) {
970  char file_name[PATH_MAX];
971 
972  snprintf(file_name, sizeof(file_name), "%s_s%d.ivf",
973  app_input.output_filename, sl);
974  outfile[sl] = vpx_video_writer_open(file_name, kContainerIVF, &info);
975  if (!outfile[sl]) die("Failed to open %s for writing", file_name);
976  }
977  }
978 #endif
979 
980  // skip initial frames
981  for (i = 0; i < app_input.frames_to_skip; ++i)
982  read_frame(&app_input.input_ctx, &raw);
983 
984  if (svc_ctx.speed != -1)
985  vpx_codec_control(&encoder, VP8E_SET_CPUUSED, svc_ctx.speed);
986  if (svc_ctx.threads) {
988  get_msb(svc_ctx.threads));
989  if (svc_ctx.threads > 1)
990  vpx_codec_control(&encoder, VP9E_SET_ROW_MT, 1);
991  else
992  vpx_codec_control(&encoder, VP9E_SET_ROW_MT, 0);
993  }
994  if (svc_ctx.speed >= 5 && svc_ctx.aqmode == 1)
995  vpx_codec_control(&encoder, VP9E_SET_AQ_MODE, 3);
996  if (svc_ctx.speed >= 5)
999 
1001  app_input.inter_layer_pred);
1002 
1004 
1005  vpx_codec_control(&encoder, VP9E_SET_TUNE_CONTENT, app_input.tune_content);
1006 
1009 
1010  svc_drop_frame.framedrop_mode = FULL_SUPERFRAME_DROP;
1011  for (sl = 0; sl < (unsigned int)svc_ctx.spatial_layers; ++sl)
1012  svc_drop_frame.framedrop_thresh[sl] = enc_cfg.rc_dropframe_thresh;
1013  svc_drop_frame.max_consec_drop = INT_MAX;
1014  vpx_codec_control(&encoder, VP9E_SET_SVC_FRAME_DROP_LAYER, &svc_drop_frame);
1015 
1016  // Encode frames
1017  while (!end_of_stream) {
1018  vpx_codec_iter_t iter = NULL;
1019  const vpx_codec_cx_pkt_t *cx_pkt;
1020  // Example patterns for bypass/flexible mode:
1021  // example_pattern = 0: 2 temporal layers, and spatial_layers = 1,2,3. Exact
1022  // to fixed SVC patterns. example_pattern = 1: 2 spatial and 2 temporal
1023  // layers, with SL0 only has TL0, and SL1 has both TL0 and TL1. This example
1024  // uses the extended API.
1025  int example_pattern = 0;
1026  if (frame_cnt >= app_input.frames_to_code ||
1027  !read_frame(&app_input.input_ctx, &raw)) {
1028  // We need one extra vpx_svc_encode call at end of stream to flush
1029  // encoder and get remaining data
1030  end_of_stream = 1;
1031  }
1032 
1033  // For BYPASS/FLEXIBLE mode, set the frame flags (reference and updates)
1034  // and the buffer indices for each spatial layer of the current
1035  // (super)frame to be encoded. The spatial and temporal layer_id for the
1036  // current frame also needs to be set.
1037  // TODO(marpan): Should rename the "VP9E_TEMPORAL_LAYERING_MODE_BYPASS"
1038  // mode to "VP9E_LAYERING_MODE_BYPASS".
1039  if (svc_ctx.temporal_layering_mode == VP9E_TEMPORAL_LAYERING_MODE_BYPASS) {
1040  layer_id.spatial_layer_id = 0;
1041  // Example for 2 temporal layers.
1042  if (frame_cnt % 2 == 0) {
1043  layer_id.temporal_layer_id = 0;
1044  for (i = 0; i < VPX_SS_MAX_LAYERS; i++)
1045  layer_id.temporal_layer_id_per_spatial[i] = 0;
1046  } else {
1047  layer_id.temporal_layer_id = 1;
1048  for (i = 0; i < VPX_SS_MAX_LAYERS; i++)
1049  layer_id.temporal_layer_id_per_spatial[i] = 1;
1050  }
1051  if (example_pattern == 1) {
1052  // example_pattern 1 is hard-coded for 2 spatial and 2 temporal layers.
1053  assert(svc_ctx.spatial_layers == 2);
1054  assert(svc_ctx.temporal_layers == 2);
1055  if (frame_cnt % 2 == 0) {
1056  // Spatial layer 0 and 1 are encoded.
1057  layer_id.temporal_layer_id_per_spatial[0] = 0;
1058  layer_id.temporal_layer_id_per_spatial[1] = 0;
1059  layer_id.spatial_layer_id = 0;
1060  } else {
1061  // Only spatial layer 1 is encoded here.
1062  layer_id.temporal_layer_id_per_spatial[1] = 1;
1063  layer_id.spatial_layer_id = 1;
1064  }
1065  }
1066  vpx_codec_control(&encoder, VP9E_SET_SVC_LAYER_ID, &layer_id);
1067  // TODO(jianj): Fix the parameter passing for "is_key_frame" in
1068  // set_frame_flags_bypass_model() for case of periodic key frames.
1069  if (example_pattern == 0) {
1070  set_frame_flags_bypass_mode_ex0(layer_id.temporal_layer_id,
1071  svc_ctx.spatial_layers, frame_cnt == 0,
1072  &ref_frame_config);
1073  } else if (example_pattern == 1) {
1074  set_frame_flags_bypass_mode_ex1(layer_id.temporal_layer_id,
1075  svc_ctx.spatial_layers, frame_cnt == 0,
1076  &ref_frame_config);
1077  }
1078  ref_frame_config.duration[0] = frame_duration * 1;
1079  ref_frame_config.duration[1] = frame_duration * 1;
1080 
1082  &ref_frame_config);
1083  // Keep track of input frames, to account for frame drops in rate control
1084  // stats/metrics.
1085  for (sl = 0; sl < enc_cfg.ss_number_layers; ++sl) {
1086  ++rc.layer_input_frames[sl * enc_cfg.ts_number_layers +
1087  layer_id.temporal_layer_id];
1088  }
1089  } else {
1090  // For the fixed pattern SVC, temporal layer is given by superframe count.
1091  unsigned int tl = 0;
1092  if (enc_cfg.ts_number_layers == 2)
1093  tl = (frame_cnt % 2 != 0);
1094  else if (enc_cfg.ts_number_layers == 3) {
1095  if (frame_cnt % 2 != 0) tl = 2;
1096  if ((frame_cnt > 1) && ((frame_cnt - 2) % 4 == 0)) tl = 1;
1097  }
1098  for (sl = 0; sl < enc_cfg.ss_number_layers; ++sl)
1099  ++rc.layer_input_frames[sl * enc_cfg.ts_number_layers + tl];
1100  }
1101 
1102  vpx_usec_timer_start(&timer);
1103  res = vpx_svc_encode(
1104  &svc_ctx, &encoder, (end_of_stream ? NULL : &raw), pts, frame_duration,
1105  svc_ctx.speed >= 5 ? VPX_DL_REALTIME : VPX_DL_GOOD_QUALITY);
1106  vpx_usec_timer_mark(&timer);
1107  cx_time += vpx_usec_timer_elapsed(&timer);
1108 
1109  fflush(stdout);
1110  if (res != VPX_CODEC_OK) {
1111  die_codec(&encoder, "Failed to encode frame");
1112  }
1113 
1114  while ((cx_pkt = vpx_codec_get_cx_data(&encoder, &iter)) != NULL) {
1115  switch (cx_pkt->kind) {
1116  case VPX_CODEC_CX_FRAME_PKT: {
1117  SvcInternal_t *const si = (SvcInternal_t *)svc_ctx.internal;
1118  if (cx_pkt->data.frame.sz > 0) {
1119  vpx_video_writer_write_frame(writer, cx_pkt->data.frame.buf,
1120  cx_pkt->data.frame.sz,
1121  cx_pkt->data.frame.pts);
1122 #if OUTPUT_RC_STATS
1123  if (svc_ctx.output_rc_stat) {
1124  svc_output_rc_stats(&encoder, &enc_cfg, &layer_id, cx_pkt, &rc,
1125  outfile, frame_cnt, framerate);
1126  }
1127 #endif
1128  }
1129  /*
1130  printf("SVC frame: %d, kf: %d, size: %d, pts: %d\n", frames_received,
1131  !!(cx_pkt->data.frame.flags & VPX_FRAME_IS_KEY),
1132  (int)cx_pkt->data.frame.sz, (int)cx_pkt->data.frame.pts);
1133  */
1134  if (enc_cfg.ss_number_layers == 1 && enc_cfg.ts_number_layers == 1)
1135  si->bytes_sum[0] += (int)cx_pkt->data.frame.sz;
1136  ++frames_received;
1137 #if CONFIG_VP9_DECODER && !SIMULCAST_MODE
1138  if (vpx_codec_decode(&decoder, cx_pkt->data.frame.buf,
1139  (unsigned int)cx_pkt->data.frame.sz, NULL, 0))
1140  die_codec(&decoder, "Failed to decode frame.");
1141 #endif
1142  break;
1143  }
1144  case VPX_CODEC_STATS_PKT: {
1145  stats_write(&app_input.rc_stats, cx_pkt->data.twopass_stats.buf,
1146  cx_pkt->data.twopass_stats.sz);
1147  break;
1148  }
1149  default: { break; }
1150  }
1151 
1152 #if CONFIG_VP9_DECODER && !SIMULCAST_MODE
1153  vpx_codec_control(&encoder, VP9E_GET_SVC_LAYER_ID, &layer_id);
1154  // Don't look for mismatch on top spatial and top temporal layers as they
1155  // are non reference frames.
1156  if ((enc_cfg.ss_number_layers > 1 || enc_cfg.ts_number_layers > 1) &&
1157  !(layer_id.temporal_layer_id > 0 &&
1158  layer_id.temporal_layer_id == (int)enc_cfg.ts_number_layers - 1 &&
1159  cx_pkt->data.frame
1160  .spatial_layer_encoded[enc_cfg.ss_number_layers - 1])) {
1161  test_decode(&encoder, &decoder, frame_cnt, &mismatch_seen);
1162  }
1163 #endif
1164  }
1165 
1166  if (!end_of_stream) {
1167  ++frame_cnt;
1168  pts += frame_duration;
1169  }
1170  }
1171 
1172  printf("Processed %d frames\n", frame_cnt);
1173 
1174  close_input_file(&app_input.input_ctx);
1175 
1176 #if OUTPUT_RC_STATS
1177  if (svc_ctx.output_rc_stat) {
1178  printout_rate_control_summary(&rc, &enc_cfg, frame_cnt);
1179  printf("\n");
1180  }
1181 #endif
1182  if (vpx_codec_destroy(&encoder))
1183  die_codec(&encoder, "Failed to destroy codec");
1184  if (writer) {
1185  vpx_video_writer_close(writer);
1186  }
1187 #if OUTPUT_RC_STATS
1188  if (svc_ctx.output_rc_stat) {
1189  for (sl = 0; sl < enc_cfg.ss_number_layers; ++sl) {
1190  vpx_video_writer_close(outfile[sl]);
1191  }
1192  }
1193 #endif
1194 #if CONFIG_INTERNAL_STATS
1195  if (mismatch_seen) {
1196  fprintf(f, "First mismatch occurred in frame %d\n", mismatch_seen);
1197  } else {
1198  fprintf(f, "No mismatch detected in recon buffers\n");
1199  }
1200  fclose(f);
1201 #endif
1202  printf("Frame cnt and encoding time/FPS stats for encoding: %d %f %f \n",
1203  frame_cnt, 1000 * (float)cx_time / (double)(frame_cnt * 1000000),
1204  1000000 * (double)frame_cnt / (double)cx_time);
1205  if (app_input.input_ctx.file_type != FILE_TYPE_Y4M) {
1206  vpx_img_free(&raw);
1207  }
1208  // display average size, psnr
1209  vpx_svc_dump_statistics(&svc_ctx);
1210  vpx_svc_release(&svc_ctx);
1211  return EXIT_SUCCESS;
1212 }
vpx_codec_err_t vpx_codec_destroy(vpx_codec_ctx_t *ctx)
Destroy a codec instance.
const void * vpx_codec_iter_t
Iterator.
Definition: vpx_codec.h:190
const char * vpx_codec_iface_name(vpx_codec_iface_t *iface)
Return the name for a given interface.
const char * vpx_codec_err_to_string(vpx_codec_err_t err)
Convert error number to printable string.
#define vpx_codec_control(ctx, id, data)
vpx_codec_control wrapper macro
Definition: vpx_codec.h:407
vpx_codec_err_t
Algorithm return codes.
Definition: vpx_codec.h:93
@ VPX_CODEC_CORRUPT_FRAME
The coded data for this stream is corrupt or incomplete.
Definition: vpx_codec.h:133
@ VPX_CODEC_OK
Operation completed without error.
Definition: vpx_codec.h:95
@ VPX_BITS_8
Definition: vpx_codec.h:221
@ VPX_BITS_12
Definition: vpx_codec.h:223
@ VPX_BITS_10
Definition: vpx_codec.h:222
vpx_codec_err_t vpx_codec_decode(vpx_codec_ctx_t *ctx, const uint8_t *data, unsigned int data_sz, void *user_priv, long deadline)
Decode data.
#define vpx_codec_dec_init(ctx, iface, cfg, flags)
Convenience macro for vpx_codec_dec_init_ver()
Definition: vpx_decoder.h:143
#define VPX_DL_REALTIME
deadline parameter analogous to VPx REALTIME mode.
Definition: vpx_encoder.h:978
#define VPX_DL_GOOD_QUALITY
deadline parameter analogous to VPx GOOD QUALITY mode.
Definition: vpx_encoder.h:980
#define VPX_MAX_LAYERS
Definition: vpx_encoder.h:44
#define VPX_FRAME_IS_KEY
Definition: vpx_encoder.h:118
#define VPX_SS_MAX_LAYERS
Definition: vpx_encoder.h:47
vpx_codec_err_t vpx_codec_enc_config_default(vpx_codec_iface_t *iface, vpx_codec_enc_cfg_t *cfg, unsigned int usage)
Get a default configuration.
const vpx_codec_cx_pkt_t * vpx_codec_get_cx_data(vpx_codec_ctx_t *ctx, vpx_codec_iter_t *iter)
Encoded data iterator.
@ VPX_CODEC_CX_FRAME_PKT
Definition: vpx_encoder.h:149
@ VPX_CODEC_STATS_PKT
Definition: vpx_encoder.h:150
@ VPX_RC_ONE_PASS
Definition: vpx_encoder.h:227
@ VPX_CQ
Definition: vpx_encoder.h:236
vpx_codec_iface_t * vpx_codec_vp9_cx(void)
The interface to the VP9 encoder.
@ FULL_SUPERFRAME_DROP
Definition: vp8cx.h:922
@ VP9E_SET_SVC_LAYER_ID
Codec control function to set svc layer for spatial and temporal.
Definition: vp8cx.h:470
@ VP8E_SET_MAX_INTRA_BITRATE_PCT
Codec control function to set Max data rate for Intra frames.
Definition: vp8cx.h:274
@ VP9E_SET_SVC_INTER_LAYER_PRED
Codec control function to constrain the inter-layer prediction (prediction of lower spatial resolutio...
Definition: vp8cx.h:624
@ VP9E_SET_AQ_MODE
Codec control function to set adaptive quantization mode.
Definition: vp8cx.h:415
@ VP9E_SET_DISABLE_OVERSHOOT_MAXQ_CBR
Codec control function to disable increase Q on overshoot in CBR.
Definition: vp8cx.h:699
@ VP9E_SET_TUNE_CONTENT
Codec control function to set content type.
Definition: vp8cx.h:480
@ VP9E_SET_ROW_MT
Codec control function to set row level multi-threading.
Definition: vp8cx.h:575
@ VP8E_SET_CPUUSED
Codec control function to set encoder internal speed settings.
Definition: vp8cx.h:172
@ VP9E_SET_TILE_COLUMNS
Codec control function to set number of tile columns.
Definition: vp8cx.h:368
@ VP9E_SET_SVC_FRAME_DROP_LAYER
Codec control function to set mode and thresholds for frame dropping in SVC. Drop frame thresholds ar...
Definition: vp8cx.h:633
@ VP9E_SET_SVC_REF_FRAME_CONFIG
Codec control function to set the frame flags and buffer indices for spatial layers....
Definition: vp8cx.h:550
@ VP8E_SET_STATIC_THRESHOLD
Codec control function to set the threshold for MBs treated static.
Definition: vp8cx.h:205
@ VP9E_SET_DISABLE_LOOPFILTER
Codec control function to disable loopfilter.
Definition: vp8cx.h:708
@ VP9E_SET_NOISE_SENSITIVITY
Codec control function to set noise sensitivity.
Definition: vp8cx.h:438
@ VP9E_GET_SVC_LAYER_ID
Codec control function to get svc layer ID.
Definition: vp8cx.h:488
@ VP9E_TEMPORAL_LAYERING_MODE_BYPASS
Bypass mode. Used when application needs to control temporal layering. This will only work when the n...
Definition: vp8cx.h:789
@ VP9_GET_REFERENCE
Definition: vp8.h:55
VP9 specific reference frame data struct.
Definition: vp8.h:110
int idx
Definition: vp8.h:111
Codec context structure.
Definition: vpx_codec.h:200
vpx_codec_err_t err
Definition: vpx_codec.h:203
Encoder output packet.
Definition: vpx_encoder.h:161
vpx_fixed_buf_t twopass_stats
Definition: vpx_encoder.h:184
enum vpx_codec_cx_pkt_kind kind
Definition: vpx_encoder.h:162
struct vpx_codec_cx_pkt::@1::@2 frame
union vpx_codec_cx_pkt::@1 data
Encoder configuration structure.
Definition: vpx_encoder.h:270
int temporal_layering_mode
Temporal layering mode indicating which temporal layering scheme to use.
Definition: vpx_encoder.h:695
unsigned int kf_min_dist
Keyframe minimum interval.
Definition: vpx_encoder.h:607
unsigned int ts_number_layers
Number of temporal coding layers.
Definition: vpx_encoder.h:646
unsigned int ss_number_layers
Number of spatial coding layers.
Definition: vpx_encoder.h:626
unsigned int rc_2pass_vbr_minsection_pct
Two-pass mode per-GOP minimum bitrate.
Definition: vpx_encoder.h:572
unsigned int g_profile
Bitstream profile to use.
Definition: vpx_encoder.h:297
unsigned int layer_target_bitrate[12]
Target bitrate for each spatial/temporal layer.
Definition: vpx_encoder.h:686
unsigned int g_h
Height of the frame.
Definition: vpx_encoder.h:315
vpx_codec_er_flags_t g_error_resilient
Enable error resilient modes.
Definition: vpx_encoder.h:353
unsigned int g_w
Width of the frame.
Definition: vpx_encoder.h:306
unsigned int rc_dropframe_thresh
Temporal resampling configuration, if supported by the codec.
Definition: vpx_encoder.h:393
struct vpx_rational g_timebase
Stream timebase units.
Definition: vpx_encoder.h:345
enum vpx_enc_pass g_pass
Multi-pass Encoding Mode.
Definition: vpx_encoder.h:360
unsigned int g_lag_in_frames
Allow lagged encoding.
Definition: vpx_encoder.h:374
enum vpx_rc_mode rc_end_usage
Rate control algorithm to use.
Definition: vpx_encoder.h:442
vpx_bit_depth_t g_bit_depth
Bit-depth of the codec.
Definition: vpx_encoder.h:323
unsigned int rc_2pass_vbr_maxsection_pct
Two-pass mode per-GOP maximum bitrate.
Definition: vpx_encoder.h:579
unsigned int rc_target_bitrate
Target data rate.
Definition: vpx_encoder.h:462
unsigned int g_input_bit_depth
Bit-depth of the input frames.
Definition: vpx_encoder.h:331
unsigned int ts_rate_decimator[5]
Frame rate decimation factor for each temporal layer.
Definition: vpx_encoder.h:660
unsigned int kf_max_dist
Keyframe maximum interval.
Definition: vpx_encoder.h:616
size_t sz
Definition: vpx_encoder.h:100
void * buf
Definition: vpx_encoder.h:99
Image Descriptor.
Definition: vpx_image.h:72
vpx_img_fmt_t fmt
Definition: vpx_image.h:73
unsigned int d_h
Definition: vpx_image.h:84
unsigned int d_w
Definition: vpx_image.h:83
int den
Definition: vpx_encoder.h:222
int num
Definition: vpx_encoder.h:221
vp9 svc frame dropping parameters.
Definition: vp8cx.h:934
int framedrop_thresh[5]
Definition: vp8cx.h:935
SVC_LAYER_DROP_MODE framedrop_mode
Definition: vp8cx.h:937
int max_consec_drop
Definition: vp8cx.h:938
vp9 svc layer parameters
Definition: vp8cx.h:883
int temporal_layer_id
Definition: vp8cx.h:886
vp9 svc frame flag parameters.
Definition: vp8cx.h:898
int lst_fb_idx[5]
Definition: vp8cx.h:899
int update_buffer_slot[5]
Definition: vp8cx.h:902
int gld_fb_idx[5]
Definition: vp8cx.h:900
int reference_last[5]
Definition: vp8cx.h:907
int reference_golden[5]
Definition: vp8cx.h:908
int reference_alt_ref[5]
Definition: vp8cx.h:909
int64_t duration[5]
Definition: vp8cx.h:910
int alt_fb_idx[5]
Definition: vp8cx.h:901
Provides definitions for using VP8 or VP9 encoder algorithm within the vpx Codec Interface.
Describes the encoder algorithm interface to applications.
#define VPX_IMG_FMT_HIGHBITDEPTH
Definition: vpx_image.h:35
@ VPX_IMG_FMT_I42016
Definition: vpx_image.h:47
@ VPX_IMG_FMT_I420
Definition: vpx_image.h:42
vpx_image_t * vpx_img_alloc(vpx_image_t *img, vpx_img_fmt_t fmt, unsigned int d_w, unsigned int d_h, unsigned int align)
Open a descriptor, allocating storage for the underlying image.
void vpx_img_free(vpx_image_t *img)
Close an image descriptor.