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 passes_arg =
70  ARG_DEF("p", "passes", 1, "Number of passes (1/2)");
71 static const arg_def_t pass_arg =
72  ARG_DEF(NULL, "pass", 1, "Pass to execute (1/2)");
73 static const arg_def_t fpf_name_arg =
74  ARG_DEF(NULL, "fpf", 1, "First pass statistics file name");
75 static const arg_def_t min_q_arg =
76  ARG_DEF(NULL, "min-q", 1, "Minimum quantizer");
77 static const arg_def_t max_q_arg =
78  ARG_DEF(NULL, "max-q", 1, "Maximum quantizer");
79 static const arg_def_t min_bitrate_arg =
80  ARG_DEF(NULL, "min-bitrate", 1, "Minimum bitrate");
81 static const arg_def_t max_bitrate_arg =
82  ARG_DEF(NULL, "max-bitrate", 1, "Maximum bitrate");
83 static const arg_def_t lag_in_frame_arg =
84  ARG_DEF(NULL, "lag-in-frames", 1,
85  "Number of frame to input before "
86  "generating any outputs");
87 static const arg_def_t rc_end_usage_arg =
88  ARG_DEF(NULL, "rc-end-usage", 1, "0 - 3: VBR, CBR, CQ, Q");
89 static const arg_def_t speed_arg =
90  ARG_DEF("sp", "speed", 1, "speed configuration");
91 static const arg_def_t aqmode_arg =
92  ARG_DEF("aq", "aqmode", 1, "aq-mode off/on");
93 static const arg_def_t bitrates_arg =
94  ARG_DEF("bl", "bitrates", 1, "bitrates[sl * num_tl + tl]");
95 static const arg_def_t dropframe_thresh_arg =
96  ARG_DEF(NULL, "drop-frame", 1, "Temporal resampling threshold (buf %)");
97 static const struct arg_enum_list tune_content_enum[] = {
98  { "default", VP9E_CONTENT_DEFAULT },
99  { "screen", VP9E_CONTENT_SCREEN },
100  { "film", VP9E_CONTENT_FILM },
101  { NULL, 0 }
102 };
103 
104 static const arg_def_t tune_content_arg = ARG_DEF_ENUM(
105  NULL, "tune-content", 1, "Tune content type", tune_content_enum);
106 static const arg_def_t inter_layer_pred_arg = ARG_DEF(
107  NULL, "inter-layer-pred", 1, "0 - 3: On, Off, Key-frames, Constrained");
108 
109 #if CONFIG_VP9_HIGHBITDEPTH
110 static const struct arg_enum_list bitdepth_enum[] = {
111  { "8", VPX_BITS_8 }, { "10", VPX_BITS_10 }, { "12", VPX_BITS_12 }, { NULL, 0 }
112 };
113 
114 static const arg_def_t bitdepth_arg = ARG_DEF_ENUM(
115  "d", "bit-depth", 1, "Bit depth for codec 8, 10 or 12. ", bitdepth_enum);
116 #endif // CONFIG_VP9_HIGHBITDEPTH
117 
118 static const arg_def_t *svc_args[] = { &frames_arg,
119  &outputfile,
120  &width_arg,
121  &height_arg,
122  &timebase_arg,
123  &bitrate_arg,
124  &skip_frames_arg,
125  &spatial_layers_arg,
126  &kf_dist_arg,
127  &scale_factors_arg,
128  &passes_arg,
129  &pass_arg,
130  &fpf_name_arg,
131  &min_q_arg,
132  &max_q_arg,
133  &min_bitrate_arg,
134  &max_bitrate_arg,
135  &temporal_layers_arg,
136  &temporal_layering_mode_arg,
137  &lag_in_frame_arg,
138  &threads_arg,
139  &aqmode_arg,
140 #if OUTPUT_RC_STATS
141  &output_rc_stats_arg,
142 #endif
143 
144 #if CONFIG_VP9_HIGHBITDEPTH
145  &bitdepth_arg,
146 #endif
147  &speed_arg,
148  &rc_end_usage_arg,
149  &bitrates_arg,
150  &dropframe_thresh_arg,
151  &tune_content_arg,
152  &inter_layer_pred_arg,
153  NULL };
154 
155 static const uint32_t default_frames_to_skip = 0;
156 static const uint32_t default_frames_to_code = 60 * 60;
157 static const uint32_t default_width = 1920;
158 static const uint32_t default_height = 1080;
159 static const uint32_t default_timebase_num = 1;
160 static const uint32_t default_timebase_den = 60;
161 static const uint32_t default_bitrate = 1000;
162 static const uint32_t default_spatial_layers = 5;
163 static const uint32_t default_temporal_layers = 1;
164 static const uint32_t default_kf_dist = 100;
165 static const uint32_t default_temporal_layering_mode = 0;
166 static const uint32_t default_output_rc_stats = 0;
167 static const int32_t default_speed = -1; // -1 means use library default.
168 static const uint32_t default_threads = 0; // zero means use library default.
169 
170 typedef struct {
171  const char *output_filename;
172  uint32_t frames_to_code;
173  uint32_t frames_to_skip;
174  struct VpxInputContext input_ctx;
175  stats_io_t rc_stats;
176  int passes;
177  int pass;
178  int tune_content;
179  int inter_layer_pred;
180 } AppInput;
181 
182 static const char *exec_name;
183 
184 void usage_exit(void) {
185  fprintf(stderr, "Usage: %s <options> input_filename -o output_filename\n",
186  exec_name);
187  fprintf(stderr, "Options:\n");
188  arg_show_usage(stderr, svc_args);
189  exit(EXIT_FAILURE);
190 }
191 
192 static void parse_command_line(int argc, const char **argv_,
193  AppInput *app_input, SvcContext *svc_ctx,
194  vpx_codec_enc_cfg_t *enc_cfg) {
195  struct arg arg;
196  char **argv = NULL;
197  char **argi = NULL;
198  char **argj = NULL;
199  vpx_codec_err_t res;
200  int passes = 0;
201  int pass = 0;
202  const char *fpf_file_name = NULL;
203  unsigned int min_bitrate = 0;
204  unsigned int max_bitrate = 0;
205  char string_options[1024] = { 0 };
206 
207  // initialize SvcContext with parameters that will be passed to vpx_svc_init
208  svc_ctx->log_level = SVC_LOG_DEBUG;
209  svc_ctx->spatial_layers = default_spatial_layers;
210  svc_ctx->temporal_layers = default_temporal_layers;
211  svc_ctx->temporal_layering_mode = default_temporal_layering_mode;
212 #if OUTPUT_RC_STATS
213  svc_ctx->output_rc_stat = default_output_rc_stats;
214 #endif
215  svc_ctx->speed = default_speed;
216  svc_ctx->threads = default_threads;
217 
218  // start with default encoder configuration
219  res = vpx_codec_enc_config_default(vpx_codec_vp9_cx(), enc_cfg, 0);
220  if (res) {
221  die("Failed to get config: %s\n", vpx_codec_err_to_string(res));
222  }
223  // update enc_cfg with app default values
224  enc_cfg->g_w = default_width;
225  enc_cfg->g_h = default_height;
226  enc_cfg->g_timebase.num = default_timebase_num;
227  enc_cfg->g_timebase.den = default_timebase_den;
228  enc_cfg->rc_target_bitrate = default_bitrate;
229  enc_cfg->kf_min_dist = default_kf_dist;
230  enc_cfg->kf_max_dist = default_kf_dist;
231  enc_cfg->rc_end_usage = VPX_CQ;
232 
233  // initialize AppInput with default values
234  app_input->frames_to_code = default_frames_to_code;
235  app_input->frames_to_skip = default_frames_to_skip;
236 
237  // process command line options
238  argv = argv_dup(argc - 1, argv_ + 1);
239  for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) {
240  arg.argv_step = 1;
241 
242  if (arg_match(&arg, &frames_arg, argi)) {
243  app_input->frames_to_code = arg_parse_uint(&arg);
244  } else if (arg_match(&arg, &outputfile, argi)) {
245  app_input->output_filename = arg.val;
246  } else if (arg_match(&arg, &width_arg, argi)) {
247  enc_cfg->g_w = arg_parse_uint(&arg);
248  } else if (arg_match(&arg, &height_arg, argi)) {
249  enc_cfg->g_h = arg_parse_uint(&arg);
250  } else if (arg_match(&arg, &timebase_arg, argi)) {
251  enc_cfg->g_timebase = arg_parse_rational(&arg);
252  } else if (arg_match(&arg, &bitrate_arg, argi)) {
253  enc_cfg->rc_target_bitrate = arg_parse_uint(&arg);
254  } else if (arg_match(&arg, &skip_frames_arg, argi)) {
255  app_input->frames_to_skip = arg_parse_uint(&arg);
256  } else if (arg_match(&arg, &spatial_layers_arg, argi)) {
257  svc_ctx->spatial_layers = arg_parse_uint(&arg);
258  } else if (arg_match(&arg, &temporal_layers_arg, argi)) {
259  svc_ctx->temporal_layers = arg_parse_uint(&arg);
260 #if OUTPUT_RC_STATS
261  } else if (arg_match(&arg, &output_rc_stats_arg, argi)) {
262  svc_ctx->output_rc_stat = arg_parse_uint(&arg);
263 #endif
264  } else if (arg_match(&arg, &speed_arg, argi)) {
265  svc_ctx->speed = arg_parse_uint(&arg);
266  if (svc_ctx->speed > 9) {
267  warn("Mapping speed %d to speed 9.\n", svc_ctx->speed);
268  }
269  } else if (arg_match(&arg, &aqmode_arg, argi)) {
270  svc_ctx->aqmode = arg_parse_uint(&arg);
271  } else if (arg_match(&arg, &threads_arg, argi)) {
272  svc_ctx->threads = arg_parse_uint(&arg);
273  } else if (arg_match(&arg, &temporal_layering_mode_arg, argi)) {
274  svc_ctx->temporal_layering_mode = enc_cfg->temporal_layering_mode =
275  arg_parse_int(&arg);
276  if (svc_ctx->temporal_layering_mode) {
277  enc_cfg->g_error_resilient = 1;
278  }
279  } else if (arg_match(&arg, &kf_dist_arg, argi)) {
280  enc_cfg->kf_min_dist = arg_parse_uint(&arg);
281  enc_cfg->kf_max_dist = enc_cfg->kf_min_dist;
282  } else if (arg_match(&arg, &scale_factors_arg, argi)) {
283  strncat(string_options, " scale-factors=",
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, &bitrates_arg, argi)) {
288  strncat(string_options, " bitrates=",
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, &passes_arg, argi)) {
293  passes = arg_parse_uint(&arg);
294  if (passes < 1 || passes > 2) {
295  die("Error: Invalid number of passes (%d)\n", passes);
296  }
297  } else if (arg_match(&arg, &pass_arg, argi)) {
298  pass = arg_parse_uint(&arg);
299  if (pass < 1 || pass > 2) {
300  die("Error: Invalid pass selected (%d)\n", pass);
301  }
302  } else if (arg_match(&arg, &fpf_name_arg, argi)) {
303  fpf_file_name = arg.val;
304  } else if (arg_match(&arg, &min_q_arg, argi)) {
305  strncat(string_options, " min-quantizers=",
306  sizeof(string_options) - strlen(string_options) - 1);
307  strncat(string_options, arg.val,
308  sizeof(string_options) - strlen(string_options) - 1);
309  } else if (arg_match(&arg, &max_q_arg, argi)) {
310  strncat(string_options, " max-quantizers=",
311  sizeof(string_options) - strlen(string_options) - 1);
312  strncat(string_options, arg.val,
313  sizeof(string_options) - strlen(string_options) - 1);
314  } else if (arg_match(&arg, &min_bitrate_arg, argi)) {
315  min_bitrate = arg_parse_uint(&arg);
316  } else if (arg_match(&arg, &max_bitrate_arg, argi)) {
317  max_bitrate = arg_parse_uint(&arg);
318  } else if (arg_match(&arg, &lag_in_frame_arg, argi)) {
319  enc_cfg->g_lag_in_frames = arg_parse_uint(&arg);
320  } else if (arg_match(&arg, &rc_end_usage_arg, argi)) {
321  enc_cfg->rc_end_usage = arg_parse_uint(&arg);
322 #if CONFIG_VP9_HIGHBITDEPTH
323  } else if (arg_match(&arg, &bitdepth_arg, argi)) {
324  enc_cfg->g_bit_depth = arg_parse_enum_or_int(&arg);
325  switch (enc_cfg->g_bit_depth) {
326  case VPX_BITS_8:
327  enc_cfg->g_input_bit_depth = 8;
328  enc_cfg->g_profile = 0;
329  break;
330  case VPX_BITS_10:
331  enc_cfg->g_input_bit_depth = 10;
332  enc_cfg->g_profile = 2;
333  break;
334  case VPX_BITS_12:
335  enc_cfg->g_input_bit_depth = 12;
336  enc_cfg->g_profile = 2;
337  break;
338  default:
339  die("Error: Invalid bit depth selected (%d)\n", enc_cfg->g_bit_depth);
340  break;
341  }
342 #endif // CONFIG_VP9_HIGHBITDEPTH
343  } else if (arg_match(&arg, &dropframe_thresh_arg, argi)) {
344  enc_cfg->rc_dropframe_thresh = arg_parse_uint(&arg);
345  } else if (arg_match(&arg, &tune_content_arg, argi)) {
346  app_input->tune_content = arg_parse_uint(&arg);
347  } else if (arg_match(&arg, &inter_layer_pred_arg, argi)) {
348  app_input->inter_layer_pred = arg_parse_uint(&arg);
349  } else {
350  ++argj;
351  }
352  }
353 
354  // There will be a space in front of the string options
355  if (strlen(string_options) > 0)
356  vpx_svc_set_options(svc_ctx, string_options + 1);
357 
358  if (passes == 0 || passes == 1) {
359  if (pass) {
360  fprintf(stderr, "pass is ignored since there's only one pass\n");
361  }
362  enc_cfg->g_pass = VPX_RC_ONE_PASS;
363  } else {
364  if (pass == 0) {
365  die("pass must be specified when passes is 2\n");
366  }
367 
368  if (fpf_file_name == NULL) {
369  die("fpf must be specified when passes is 2\n");
370  }
371 
372  if (pass == 1) {
373  enc_cfg->g_pass = VPX_RC_FIRST_PASS;
374  if (!stats_open_file(&app_input->rc_stats, fpf_file_name, 0)) {
375  fatal("Failed to open statistics store");
376  }
377  } else {
378  enc_cfg->g_pass = VPX_RC_LAST_PASS;
379  if (!stats_open_file(&app_input->rc_stats, fpf_file_name, 1)) {
380  fatal("Failed to open statistics store");
381  }
382  enc_cfg->rc_twopass_stats_in = stats_get(&app_input->rc_stats);
383  }
384  app_input->passes = passes;
385  app_input->pass = pass;
386  }
387 
388  if (enc_cfg->rc_target_bitrate > 0) {
389  if (min_bitrate > 0) {
390  enc_cfg->rc_2pass_vbr_minsection_pct =
391  min_bitrate * 100 / enc_cfg->rc_target_bitrate;
392  }
393  if (max_bitrate > 0) {
394  enc_cfg->rc_2pass_vbr_maxsection_pct =
395  max_bitrate * 100 / enc_cfg->rc_target_bitrate;
396  }
397  }
398 
399  // Check for unrecognized options
400  for (argi = argv; *argi; ++argi)
401  if (argi[0][0] == '-' && strlen(argi[0]) > 1)
402  die("Error: Unrecognized option %s\n", *argi);
403 
404  if (argv[0] == NULL) {
405  usage_exit();
406  }
407  app_input->input_ctx.filename = argv[0];
408  free(argv);
409 
410  open_input_file(&app_input->input_ctx);
411  if (app_input->input_ctx.file_type == FILE_TYPE_Y4M) {
412  enc_cfg->g_w = app_input->input_ctx.width;
413  enc_cfg->g_h = app_input->input_ctx.height;
414  }
415 
416  if (enc_cfg->g_w < 16 || enc_cfg->g_w % 2 || enc_cfg->g_h < 16 ||
417  enc_cfg->g_h % 2)
418  die("Invalid resolution: %d x %d\n", enc_cfg->g_w, enc_cfg->g_h);
419 
420  printf(
421  "Codec %s\nframes: %d, skip: %d\n"
422  "layers: %d\n"
423  "width %d, height: %d,\n"
424  "num: %d, den: %d, bitrate: %d,\n"
425  "gop size: %d\n",
426  vpx_codec_iface_name(vpx_codec_vp9_cx()), app_input->frames_to_code,
427  app_input->frames_to_skip, svc_ctx->spatial_layers, enc_cfg->g_w,
428  enc_cfg->g_h, enc_cfg->g_timebase.num, enc_cfg->g_timebase.den,
429  enc_cfg->rc_target_bitrate, enc_cfg->kf_max_dist);
430 }
431 
432 #if OUTPUT_RC_STATS
433 // For rate control encoding stats.
434 struct RateControlStats {
435  // Number of input frames per layer.
436  int layer_input_frames[VPX_MAX_LAYERS];
437  // Total (cumulative) number of encoded frames per layer.
438  int layer_tot_enc_frames[VPX_MAX_LAYERS];
439  // Number of encoded non-key frames per layer.
440  int layer_enc_frames[VPX_MAX_LAYERS];
441  // Framerate per layer (cumulative).
442  double layer_framerate[VPX_MAX_LAYERS];
443  // Target average frame size per layer (per-frame-bandwidth per layer).
444  double layer_pfb[VPX_MAX_LAYERS];
445  // Actual average frame size per layer.
446  double layer_avg_frame_size[VPX_MAX_LAYERS];
447  // Average rate mismatch per layer (|target - actual| / target).
448  double layer_avg_rate_mismatch[VPX_MAX_LAYERS];
449  // Actual encoding bitrate per layer (cumulative).
450  double layer_encoding_bitrate[VPX_MAX_LAYERS];
451  // Average of the short-time encoder actual bitrate.
452  // TODO(marpan): Should we add these short-time stats for each layer?
453  double avg_st_encoding_bitrate;
454  // Variance of the short-time encoder actual bitrate.
455  double variance_st_encoding_bitrate;
456  // Window (number of frames) for computing short-time encoding bitrate.
457  int window_size;
458  // Number of window measurements.
459  int window_count;
460 };
461 
462 // Note: these rate control stats assume only 1 key frame in the
463 // sequence (i.e., first frame only).
464 static void set_rate_control_stats(struct RateControlStats *rc,
465  vpx_codec_enc_cfg_t *cfg) {
466  unsigned int sl, tl;
467  // Set the layer (cumulative) framerate and the target layer (non-cumulative)
468  // per-frame-bandwidth, for the rate control encoding stats below.
469  const double framerate = cfg->g_timebase.den / cfg->g_timebase.num;
470 
471  for (sl = 0; sl < cfg->ss_number_layers; ++sl) {
472  for (tl = 0; tl < cfg->ts_number_layers; ++tl) {
473  const int layer = sl * cfg->ts_number_layers + tl;
474  if (cfg->ts_number_layers == 1)
475  rc->layer_framerate[layer] = framerate;
476  else
477  rc->layer_framerate[layer] = framerate / cfg->ts_rate_decimator[tl];
478  if (tl > 0) {
479  rc->layer_pfb[layer] =
480  1000.0 *
481  (cfg->layer_target_bitrate[layer] -
482  cfg->layer_target_bitrate[layer - 1]) /
483  (rc->layer_framerate[layer] - rc->layer_framerate[layer - 1]);
484  } else {
485  rc->layer_pfb[layer] = 1000.0 * cfg->layer_target_bitrate[layer] /
486  rc->layer_framerate[layer];
487  }
488  rc->layer_input_frames[layer] = 0;
489  rc->layer_enc_frames[layer] = 0;
490  rc->layer_tot_enc_frames[layer] = 0;
491  rc->layer_encoding_bitrate[layer] = 0.0;
492  rc->layer_avg_frame_size[layer] = 0.0;
493  rc->layer_avg_rate_mismatch[layer] = 0.0;
494  }
495  }
496  rc->window_count = 0;
497  rc->window_size = 15;
498  rc->avg_st_encoding_bitrate = 0.0;
499  rc->variance_st_encoding_bitrate = 0.0;
500 }
501 
502 static void printout_rate_control_summary(struct RateControlStats *rc,
503  vpx_codec_enc_cfg_t *cfg,
504  int frame_cnt) {
505  unsigned int sl, tl;
506  double perc_fluctuation = 0.0;
507  int tot_num_frames = 0;
508  printf("Total number of processed frames: %d\n\n", frame_cnt - 1);
509  printf("Rate control layer stats for sl%d tl%d layer(s):\n\n",
511  for (sl = 0; sl < cfg->ss_number_layers; ++sl) {
512  tot_num_frames = 0;
513  for (tl = 0; tl < cfg->ts_number_layers; ++tl) {
514  const int layer = sl * cfg->ts_number_layers + tl;
515  const int num_dropped =
516  (tl > 0)
517  ? (rc->layer_input_frames[layer] - rc->layer_enc_frames[layer])
518  : (rc->layer_input_frames[layer] - rc->layer_enc_frames[layer] -
519  1);
520  tot_num_frames += rc->layer_input_frames[layer];
521  rc->layer_encoding_bitrate[layer] = 0.001 * rc->layer_framerate[layer] *
522  rc->layer_encoding_bitrate[layer] /
523  tot_num_frames;
524  rc->layer_avg_frame_size[layer] =
525  rc->layer_avg_frame_size[layer] / rc->layer_enc_frames[layer];
526  rc->layer_avg_rate_mismatch[layer] = 100.0 *
527  rc->layer_avg_rate_mismatch[layer] /
528  rc->layer_enc_frames[layer];
529  printf("For layer#: sl%d tl%d \n", sl, tl);
530  printf("Bitrate (target vs actual): %d %f.0 kbps\n",
531  cfg->layer_target_bitrate[layer],
532  rc->layer_encoding_bitrate[layer]);
533  printf("Average frame size (target vs actual): %f %f bits\n",
534  rc->layer_pfb[layer], rc->layer_avg_frame_size[layer]);
535  printf("Average rate_mismatch: %f\n", rc->layer_avg_rate_mismatch[layer]);
536  printf(
537  "Number of input frames, encoded (non-key) frames, "
538  "and percent dropped frames: %d %d %f.0 \n",
539  rc->layer_input_frames[layer], rc->layer_enc_frames[layer],
540  100.0 * num_dropped / rc->layer_input_frames[layer]);
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  printf("Num of input, num of encoded (super) frames: %d %d \n", frame_cnt,
555  tot_num_frames);
556 }
557 
558 static vpx_codec_err_t parse_superframe_index(const uint8_t *data,
559  size_t data_sz, uint64_t sizes[8],
560  int *count) {
561  // A chunk ending with a byte matching 0xc0 is an invalid chunk unless
562  // it is a super frame index. If the last byte of real video compression
563  // data is 0xc0 the encoder must add a 0 byte. If we have the marker but
564  // not the associated matching marker byte at the front of the index we have
565  // an invalid bitstream and need to return an error.
566 
567  uint8_t marker;
568 
569  marker = *(data + data_sz - 1);
570  *count = 0;
571 
572  if ((marker & 0xe0) == 0xc0) {
573  const uint32_t frames = (marker & 0x7) + 1;
574  const uint32_t mag = ((marker >> 3) & 0x3) + 1;
575  const size_t index_sz = 2 + mag * frames;
576 
577  // This chunk is marked as having a superframe index but doesn't have
578  // enough data for it, thus it's an invalid superframe index.
579  if (data_sz < index_sz) return VPX_CODEC_CORRUPT_FRAME;
580 
581  {
582  const uint8_t marker2 = *(data + data_sz - index_sz);
583 
584  // This chunk is marked as having a superframe index but doesn't have
585  // the matching marker byte at the front of the index therefore it's an
586  // invalid chunk.
587  if (marker != marker2) return VPX_CODEC_CORRUPT_FRAME;
588  }
589 
590  {
591  // Found a valid superframe index.
592  uint32_t i, j;
593  const uint8_t *x = &data[data_sz - index_sz + 1];
594 
595  for (i = 0; i < frames; ++i) {
596  uint32_t this_sz = 0;
597 
598  for (j = 0; j < mag; ++j) this_sz |= (*x++) << (j * 8);
599  sizes[i] = this_sz;
600  }
601  *count = frames;
602  }
603  }
604  return VPX_CODEC_OK;
605 }
606 #endif
607 
608 // Example pattern for spatial layers and 2 temporal layers used in the
609 // bypass/flexible mode. The pattern corresponds to the pattern
610 // VP9E_TEMPORAL_LAYERING_MODE_0101 (temporal_layering_mode == 2) used in
611 // non-flexible mode.
612 static void set_frame_flags_bypass_mode_ex0(
613  int tl, int num_spatial_layers, int is_key_frame,
614  vpx_svc_ref_frame_config_t *ref_frame_config) {
615  int sl;
616  for (sl = 0; sl < num_spatial_layers; ++sl)
617  ref_frame_config->update_buffer_slot[sl] = 0;
618 
619  for (sl = 0; sl < num_spatial_layers; ++sl) {
620  // Set the buffer idx.
621  if (tl == 0) {
622  ref_frame_config->lst_fb_idx[sl] = sl;
623  if (sl) {
624  if (is_key_frame) {
625  ref_frame_config->lst_fb_idx[sl] = sl - 1;
626  ref_frame_config->gld_fb_idx[sl] = sl;
627  } else {
628  ref_frame_config->gld_fb_idx[sl] = sl - 1;
629  }
630  } else {
631  ref_frame_config->gld_fb_idx[sl] = 0;
632  }
633  ref_frame_config->alt_fb_idx[sl] = 0;
634  } else if (tl == 1) {
635  ref_frame_config->lst_fb_idx[sl] = sl;
636  ref_frame_config->gld_fb_idx[sl] = num_spatial_layers + sl - 1;
637  ref_frame_config->alt_fb_idx[sl] = num_spatial_layers + sl;
638  }
639  // Set the reference and update flags.
640  if (!tl) {
641  if (!sl) {
642  // Base spatial and base temporal (sl = 0, tl = 0)
643  ref_frame_config->reference_last[sl] = 1;
644  ref_frame_config->reference_golden[sl] = 0;
645  ref_frame_config->reference_alt_ref[sl] = 0;
646  ref_frame_config->update_buffer_slot[sl] |=
647  1 << ref_frame_config->lst_fb_idx[sl];
648  } else {
649  if (is_key_frame) {
650  ref_frame_config->reference_last[sl] = 1;
651  ref_frame_config->reference_golden[sl] = 0;
652  ref_frame_config->reference_alt_ref[sl] = 0;
653  ref_frame_config->update_buffer_slot[sl] |=
654  1 << ref_frame_config->gld_fb_idx[sl];
655  } else {
656  // Non-zero spatiall layer.
657  ref_frame_config->reference_last[sl] = 1;
658  ref_frame_config->reference_golden[sl] = 1;
659  ref_frame_config->reference_alt_ref[sl] = 1;
660  ref_frame_config->update_buffer_slot[sl] |=
661  1 << ref_frame_config->lst_fb_idx[sl];
662  }
663  }
664  } else if (tl == 1) {
665  if (!sl) {
666  // Base spatial and top temporal (tl = 1)
667  ref_frame_config->reference_last[sl] = 1;
668  ref_frame_config->reference_golden[sl] = 0;
669  ref_frame_config->reference_alt_ref[sl] = 0;
670  ref_frame_config->update_buffer_slot[sl] |=
671  1 << ref_frame_config->alt_fb_idx[sl];
672  } else {
673  // Non-zero spatial.
674  if (sl < num_spatial_layers - 1) {
675  ref_frame_config->reference_last[sl] = 1;
676  ref_frame_config->reference_golden[sl] = 1;
677  ref_frame_config->reference_alt_ref[sl] = 0;
678  ref_frame_config->update_buffer_slot[sl] |=
679  1 << ref_frame_config->alt_fb_idx[sl];
680  } else if (sl == num_spatial_layers - 1) {
681  // Top spatial and top temporal (non-reference -- doesn't update any
682  // reference buffers)
683  ref_frame_config->reference_last[sl] = 1;
684  ref_frame_config->reference_golden[sl] = 1;
685  ref_frame_config->reference_alt_ref[sl] = 0;
686  }
687  }
688  }
689  }
690 }
691 
692 // Example pattern for 2 spatial layers and 2 temporal layers used in the
693 // bypass/flexible mode, except only 1 spatial layer when temporal_layer_id = 1.
694 static void set_frame_flags_bypass_mode_ex1(
695  int tl, int num_spatial_layers, int is_key_frame,
696  vpx_svc_ref_frame_config_t *ref_frame_config) {
697  int sl;
698  for (sl = 0; sl < num_spatial_layers; ++sl)
699  ref_frame_config->update_buffer_slot[sl] = 0;
700 
701  if (tl == 0) {
702  if (is_key_frame) {
703  ref_frame_config->lst_fb_idx[1] = 0;
704  ref_frame_config->gld_fb_idx[1] = 1;
705  } else {
706  ref_frame_config->lst_fb_idx[1] = 1;
707  ref_frame_config->gld_fb_idx[1] = 0;
708  }
709  ref_frame_config->alt_fb_idx[1] = 0;
710 
711  ref_frame_config->lst_fb_idx[0] = 0;
712  ref_frame_config->gld_fb_idx[0] = 0;
713  ref_frame_config->alt_fb_idx[0] = 0;
714  }
715  if (tl == 1) {
716  ref_frame_config->lst_fb_idx[0] = 0;
717  ref_frame_config->gld_fb_idx[0] = 1;
718  ref_frame_config->alt_fb_idx[0] = 2;
719 
720  ref_frame_config->lst_fb_idx[1] = 1;
721  ref_frame_config->gld_fb_idx[1] = 2;
722  ref_frame_config->alt_fb_idx[1] = 3;
723  }
724  // Set the reference and update flags.
725  if (tl == 0) {
726  // Base spatial and base temporal (sl = 0, tl = 0)
727  ref_frame_config->reference_last[0] = 1;
728  ref_frame_config->reference_golden[0] = 0;
729  ref_frame_config->reference_alt_ref[0] = 0;
730  ref_frame_config->update_buffer_slot[0] |=
731  1 << ref_frame_config->lst_fb_idx[0];
732 
733  if (is_key_frame) {
734  ref_frame_config->reference_last[1] = 1;
735  ref_frame_config->reference_golden[1] = 0;
736  ref_frame_config->reference_alt_ref[1] = 0;
737  ref_frame_config->update_buffer_slot[1] |=
738  1 << ref_frame_config->gld_fb_idx[1];
739  } else {
740  // Non-zero spatiall layer.
741  ref_frame_config->reference_last[1] = 1;
742  ref_frame_config->reference_golden[1] = 1;
743  ref_frame_config->reference_alt_ref[1] = 1;
744  ref_frame_config->update_buffer_slot[1] |=
745  1 << ref_frame_config->lst_fb_idx[1];
746  }
747  }
748  if (tl == 1) {
749  // Top spatial and top temporal (non-reference -- doesn't update any
750  // reference buffers)
751  ref_frame_config->reference_last[1] = 1;
752  ref_frame_config->reference_golden[1] = 0;
753  ref_frame_config->reference_alt_ref[1] = 0;
754  }
755 }
756 
757 #if CONFIG_VP9_DECODER && !SIMULCAST_MODE
758 static void test_decode(vpx_codec_ctx_t *encoder, vpx_codec_ctx_t *decoder,
759  const int frames_out, int *mismatch_seen) {
760  vpx_image_t enc_img, dec_img;
761  struct vp9_ref_frame ref_enc, ref_dec;
762  if (*mismatch_seen) return;
763  /* Get the internal reference frame */
764  ref_enc.idx = 0;
765  ref_dec.idx = 0;
766  vpx_codec_control(encoder, VP9_GET_REFERENCE, &ref_enc);
767  enc_img = ref_enc.img;
768  vpx_codec_control(decoder, VP9_GET_REFERENCE, &ref_dec);
769  dec_img = ref_dec.img;
770 #if CONFIG_VP9_HIGHBITDEPTH
771  if ((enc_img.fmt & VPX_IMG_FMT_HIGHBITDEPTH) !=
772  (dec_img.fmt & VPX_IMG_FMT_HIGHBITDEPTH)) {
773  if (enc_img.fmt & VPX_IMG_FMT_HIGHBITDEPTH) {
774  vpx_img_alloc(&enc_img, enc_img.fmt - VPX_IMG_FMT_HIGHBITDEPTH,
775  enc_img.d_w, enc_img.d_h, 16);
776  vpx_img_truncate_16_to_8(&enc_img, &ref_enc.img);
777  }
778  if (dec_img.fmt & VPX_IMG_FMT_HIGHBITDEPTH) {
779  vpx_img_alloc(&dec_img, dec_img.fmt - VPX_IMG_FMT_HIGHBITDEPTH,
780  dec_img.d_w, dec_img.d_h, 16);
781  vpx_img_truncate_16_to_8(&dec_img, &ref_dec.img);
782  }
783  }
784 #endif
785 
786  if (!compare_img(&enc_img, &dec_img)) {
787  int y[4], u[4], v[4];
788 #if CONFIG_VP9_HIGHBITDEPTH
789  if (enc_img.fmt & VPX_IMG_FMT_HIGHBITDEPTH) {
790  find_mismatch_high(&enc_img, &dec_img, y, u, v);
791  } else {
792  find_mismatch(&enc_img, &dec_img, y, u, v);
793  }
794 #else
795  find_mismatch(&enc_img, &dec_img, y, u, v);
796 #endif
797  decoder->err = 1;
798  printf(
799  "Encode/decode mismatch on frame %d at"
800  " Y[%d, %d] {%d/%d},"
801  " U[%d, %d] {%d/%d},"
802  " V[%d, %d] {%d/%d}\n",
803  frames_out, y[0], y[1], y[2], y[3], u[0], u[1], u[2], u[3], v[0], v[1],
804  v[2], v[3]);
805  *mismatch_seen = frames_out;
806  }
807 
808  vpx_img_free(&enc_img);
809  vpx_img_free(&dec_img);
810 }
811 #endif
812 
813 #if OUTPUT_RC_STATS
814 static void svc_output_rc_stats(
815  vpx_codec_ctx_t *codec, vpx_codec_enc_cfg_t *enc_cfg,
816  vpx_svc_layer_id_t *layer_id, const vpx_codec_cx_pkt_t *cx_pkt,
817  struct RateControlStats *rc, VpxVideoWriter **outfile,
818  const uint32_t frame_cnt, const double framerate) {
819  int num_layers_encoded = 0;
820  unsigned int sl, tl;
821  uint64_t sizes[8];
822  uint64_t sizes_parsed[8];
823  int count = 0;
824  double sum_bitrate = 0.0;
825  double sum_bitrate2 = 0.0;
826  vp9_zero(sizes);
827  vp9_zero(sizes_parsed);
828  vpx_codec_control(codec, VP9E_GET_SVC_LAYER_ID, layer_id);
829  parse_superframe_index(cx_pkt->data.frame.buf, cx_pkt->data.frame.sz,
830  sizes_parsed, &count);
831  if (enc_cfg->ss_number_layers == 1) sizes[0] = cx_pkt->data.frame.sz;
832  for (sl = 0; sl < enc_cfg->ss_number_layers; ++sl) {
833  sizes[sl] = 0;
834  if (cx_pkt->data.frame.spatial_layer_encoded[sl]) {
835  sizes[sl] = sizes_parsed[num_layers_encoded];
836  num_layers_encoded++;
837  }
838  }
839  for (sl = 0; sl < enc_cfg->ss_number_layers; ++sl) {
840  unsigned int sl2;
841  uint64_t tot_size = 0;
842 #if SIMULCAST_MODE
843  for (sl2 = 0; sl2 < sl; ++sl2) {
844  if (cx_pkt->data.frame.spatial_layer_encoded[sl2]) tot_size += sizes[sl2];
845  }
846  vpx_video_writer_write_frame(outfile[sl],
847  (uint8_t *)(cx_pkt->data.frame.buf) + tot_size,
848  (size_t)(sizes[sl]), cx_pkt->data.frame.pts);
849 #else
850  for (sl2 = 0; sl2 <= sl; ++sl2) {
851  if (cx_pkt->data.frame.spatial_layer_encoded[sl2]) tot_size += sizes[sl2];
852  }
853  if (tot_size > 0)
854  vpx_video_writer_write_frame(outfile[sl], cx_pkt->data.frame.buf,
855  (size_t)(tot_size), cx_pkt->data.frame.pts);
856 #endif // SIMULCAST_MODE
857  }
858  for (sl = 0; sl < enc_cfg->ss_number_layers; ++sl) {
859  if (cx_pkt->data.frame.spatial_layer_encoded[sl]) {
860  for (tl = layer_id->temporal_layer_id; tl < enc_cfg->ts_number_layers;
861  ++tl) {
862  const int layer = sl * enc_cfg->ts_number_layers + tl;
863  ++rc->layer_tot_enc_frames[layer];
864  rc->layer_encoding_bitrate[layer] += 8.0 * sizes[sl];
865  // Keep count of rate control stats per layer, for non-key
866  // frames.
867  if (tl == (unsigned int)layer_id->temporal_layer_id &&
868  !(cx_pkt->data.frame.flags & VPX_FRAME_IS_KEY)) {
869  rc->layer_avg_frame_size[layer] += 8.0 * sizes[sl];
870  rc->layer_avg_rate_mismatch[layer] +=
871  fabs(8.0 * sizes[sl] - rc->layer_pfb[layer]) /
872  rc->layer_pfb[layer];
873  ++rc->layer_enc_frames[layer];
874  }
875  }
876  }
877  }
878 
879  // Update for short-time encoding bitrate states, for moving
880  // window of size rc->window, shifted by rc->window / 2.
881  // Ignore first window segment, due to key frame.
882  if (frame_cnt > (unsigned int)rc->window_size) {
883  for (sl = 0; sl < enc_cfg->ss_number_layers; ++sl) {
884  if (cx_pkt->data.frame.spatial_layer_encoded[sl])
885  sum_bitrate += 0.001 * 8.0 * sizes[sl] * framerate;
886  }
887  if (frame_cnt % rc->window_size == 0) {
888  rc->window_count += 1;
889  rc->avg_st_encoding_bitrate += sum_bitrate / rc->window_size;
890  rc->variance_st_encoding_bitrate +=
891  (sum_bitrate / rc->window_size) * (sum_bitrate / rc->window_size);
892  }
893  }
894 
895  // Second shifted window.
896  if (frame_cnt > (unsigned int)(rc->window_size + rc->window_size / 2)) {
897  for (sl = 0; sl < enc_cfg->ss_number_layers; ++sl) {
898  sum_bitrate2 += 0.001 * 8.0 * sizes[sl] * framerate;
899  }
900 
901  if (frame_cnt > (unsigned int)(2 * rc->window_size) &&
902  frame_cnt % rc->window_size == 0) {
903  rc->window_count += 1;
904  rc->avg_st_encoding_bitrate += sum_bitrate2 / rc->window_size;
905  rc->variance_st_encoding_bitrate +=
906  (sum_bitrate2 / rc->window_size) * (sum_bitrate2 / rc->window_size);
907  }
908  }
909 }
910 #endif
911 
912 int main(int argc, const char **argv) {
913  AppInput app_input;
914  VpxVideoWriter *writer = NULL;
915  VpxVideoInfo info;
916  vpx_codec_ctx_t encoder;
917  vpx_codec_enc_cfg_t enc_cfg;
918  SvcContext svc_ctx;
919  vpx_svc_frame_drop_t svc_drop_frame;
920  uint32_t i;
921  uint32_t frame_cnt = 0;
922  vpx_image_t raw;
923  vpx_codec_err_t res;
924  int pts = 0; /* PTS starts at 0 */
925  int frame_duration = 1; /* 1 timebase tick per frame */
926  int end_of_stream = 0;
927  int frames_received = 0;
928 #if OUTPUT_RC_STATS
929  VpxVideoWriter *outfile[VPX_SS_MAX_LAYERS] = { NULL };
930  struct RateControlStats rc;
931  vpx_svc_layer_id_t layer_id;
932  vpx_svc_ref_frame_config_t ref_frame_config;
933  unsigned int sl;
934  double framerate = 30.0;
935 #endif
936  struct vpx_usec_timer timer;
937  int64_t cx_time = 0;
938 #if CONFIG_INTERNAL_STATS
939  FILE *f = fopen("opsnr.stt", "a");
940 #endif
941 #if CONFIG_VP9_DECODER && !SIMULCAST_MODE
942  int mismatch_seen = 0;
943  vpx_codec_ctx_t decoder;
944 #endif
945  memset(&svc_ctx, 0, sizeof(svc_ctx));
946  memset(&app_input, 0, sizeof(AppInput));
947  memset(&info, 0, sizeof(VpxVideoInfo));
948  memset(&layer_id, 0, sizeof(vpx_svc_layer_id_t));
949  memset(&rc, 0, sizeof(struct RateControlStats));
950  exec_name = argv[0];
951 
952  /* Setup default input stream settings */
953  app_input.input_ctx.framerate.numerator = 30;
954  app_input.input_ctx.framerate.denominator = 1;
955  app_input.input_ctx.only_i420 = 1;
956  app_input.input_ctx.bit_depth = 0;
957 
958  parse_command_line(argc, argv, &app_input, &svc_ctx, &enc_cfg);
959 
960  // Y4M reader handles its own allocation.
961  if (app_input.input_ctx.file_type != FILE_TYPE_Y4M) {
962 // Allocate image buffer
963 #if CONFIG_VP9_HIGHBITDEPTH
964  if (!vpx_img_alloc(&raw,
965  enc_cfg.g_input_bit_depth == 8 ? VPX_IMG_FMT_I420
967  enc_cfg.g_w, enc_cfg.g_h, 32)) {
968  die("Failed to allocate image %dx%d\n", enc_cfg.g_w, enc_cfg.g_h);
969  }
970 #else
971  if (!vpx_img_alloc(&raw, VPX_IMG_FMT_I420, enc_cfg.g_w, enc_cfg.g_h, 32)) {
972  die("Failed to allocate image %dx%d\n", enc_cfg.g_w, enc_cfg.g_h);
973  }
974 #endif // CONFIG_VP9_HIGHBITDEPTH
975  }
976 
977  // Initialize codec
978  if (vpx_svc_init(&svc_ctx, &encoder, vpx_codec_vp9_cx(), &enc_cfg) !=
979  VPX_CODEC_OK)
980  die("Failed to initialize encoder\n");
981 #if CONFIG_VP9_DECODER && !SIMULCAST_MODE
982  if (vpx_codec_dec_init(
983  &decoder, get_vpx_decoder_by_name("vp9")->codec_interface(), NULL, 0))
984  die("Failed to initialize decoder\n");
985 #endif
986 
987 #if OUTPUT_RC_STATS
988  rc.window_count = 1;
989  rc.window_size = 15; // Silence a static analysis warning.
990  rc.avg_st_encoding_bitrate = 0.0;
991  rc.variance_st_encoding_bitrate = 0.0;
992  if (svc_ctx.output_rc_stat) {
993  set_rate_control_stats(&rc, &enc_cfg);
994  framerate = enc_cfg.g_timebase.den / enc_cfg.g_timebase.num;
995  }
996 #endif
997 
998  info.codec_fourcc = VP9_FOURCC;
999  info.frame_width = enc_cfg.g_w;
1000  info.frame_height = enc_cfg.g_h;
1001  info.time_base.numerator = enc_cfg.g_timebase.num;
1002  info.time_base.denominator = enc_cfg.g_timebase.den;
1003 
1004  if (!(app_input.passes == 2 && app_input.pass == 1)) {
1005  // We don't save the bitstream for the 1st pass on two pass rate control
1006  writer =
1007  vpx_video_writer_open(app_input.output_filename, kContainerIVF, &info);
1008  if (!writer)
1009  die("Failed to open %s for writing\n", app_input.output_filename);
1010  }
1011 #if OUTPUT_RC_STATS
1012  // Write out spatial layer stream.
1013  // TODO(marpan/jianj): allow for writing each spatial and temporal stream.
1014  if (svc_ctx.output_rc_stat) {
1015  for (sl = 0; sl < enc_cfg.ss_number_layers; ++sl) {
1016  char file_name[PATH_MAX];
1017 
1018  snprintf(file_name, sizeof(file_name), "%s_s%d.ivf",
1019  app_input.output_filename, sl);
1020  outfile[sl] = vpx_video_writer_open(file_name, kContainerIVF, &info);
1021  if (!outfile[sl]) die("Failed to open %s for writing", file_name);
1022  }
1023  }
1024 #endif
1025 
1026  // skip initial frames
1027  for (i = 0; i < app_input.frames_to_skip; ++i)
1028  read_frame(&app_input.input_ctx, &raw);
1029 
1030  if (svc_ctx.speed != -1)
1031  vpx_codec_control(&encoder, VP8E_SET_CPUUSED, svc_ctx.speed);
1032  if (svc_ctx.threads) {
1034  get_msb(svc_ctx.threads));
1035  if (svc_ctx.threads > 1)
1036  vpx_codec_control(&encoder, VP9E_SET_ROW_MT, 1);
1037  else
1038  vpx_codec_control(&encoder, VP9E_SET_ROW_MT, 0);
1039  }
1040  if (svc_ctx.speed >= 5 && svc_ctx.aqmode == 1)
1041  vpx_codec_control(&encoder, VP9E_SET_AQ_MODE, 3);
1042  if (svc_ctx.speed >= 5)
1045 
1047  app_input.inter_layer_pred);
1048 
1050 
1051  vpx_codec_control(&encoder, VP9E_SET_TUNE_CONTENT, app_input.tune_content);
1052 
1053  svc_drop_frame.framedrop_mode = FULL_SUPERFRAME_DROP;
1054  for (sl = 0; sl < (unsigned int)svc_ctx.spatial_layers; ++sl)
1055  svc_drop_frame.framedrop_thresh[sl] = enc_cfg.rc_dropframe_thresh;
1056  svc_drop_frame.max_consec_drop = INT_MAX;
1057  vpx_codec_control(&encoder, VP9E_SET_SVC_FRAME_DROP_LAYER, &svc_drop_frame);
1058 
1059  // Encode frames
1060  while (!end_of_stream) {
1061  vpx_codec_iter_t iter = NULL;
1062  const vpx_codec_cx_pkt_t *cx_pkt;
1063  // Example patterns for bypass/flexible mode:
1064  // example_pattern = 0: 2 temporal layers, and spatial_layers = 1,2,3. Exact
1065  // to fixed SVC patterns. example_pattern = 1: 2 spatial and 2 temporal
1066  // layers, with SL0 only has TL0, and SL1 has both TL0 and TL1. This example
1067  // uses the extended API.
1068  int example_pattern = 0;
1069  if (frame_cnt >= app_input.frames_to_code ||
1070  !read_frame(&app_input.input_ctx, &raw)) {
1071  // We need one extra vpx_svc_encode call at end of stream to flush
1072  // encoder and get remaining data
1073  end_of_stream = 1;
1074  }
1075 
1076  // For BYPASS/FLEXIBLE mode, set the frame flags (reference and updates)
1077  // and the buffer indices for each spatial layer of the current
1078  // (super)frame to be encoded. The spatial and temporal layer_id for the
1079  // current frame also needs to be set.
1080  // TODO(marpan): Should rename the "VP9E_TEMPORAL_LAYERING_MODE_BYPASS"
1081  // mode to "VP9E_LAYERING_MODE_BYPASS".
1082  if (svc_ctx.temporal_layering_mode == VP9E_TEMPORAL_LAYERING_MODE_BYPASS) {
1083  layer_id.spatial_layer_id = 0;
1084  // Example for 2 temporal layers.
1085  if (frame_cnt % 2 == 0) {
1086  layer_id.temporal_layer_id = 0;
1087  for (i = 0; i < VPX_SS_MAX_LAYERS; i++)
1088  layer_id.temporal_layer_id_per_spatial[i] = 0;
1089  } else {
1090  layer_id.temporal_layer_id = 1;
1091  for (i = 0; i < VPX_SS_MAX_LAYERS; i++)
1092  layer_id.temporal_layer_id_per_spatial[i] = 1;
1093  }
1094  if (example_pattern == 1) {
1095  // example_pattern 1 is hard-coded for 2 spatial and 2 temporal layers.
1096  assert(svc_ctx.spatial_layers == 2);
1097  assert(svc_ctx.temporal_layers == 2);
1098  if (frame_cnt % 2 == 0) {
1099  // Spatial layer 0 and 1 are encoded.
1100  layer_id.temporal_layer_id_per_spatial[0] = 0;
1101  layer_id.temporal_layer_id_per_spatial[1] = 0;
1102  layer_id.spatial_layer_id = 0;
1103  } else {
1104  // Only spatial layer 1 is encoded here.
1105  layer_id.temporal_layer_id_per_spatial[1] = 1;
1106  layer_id.spatial_layer_id = 1;
1107  }
1108  }
1109  vpx_codec_control(&encoder, VP9E_SET_SVC_LAYER_ID, &layer_id);
1110  // TODO(jianj): Fix the parameter passing for "is_key_frame" in
1111  // set_frame_flags_bypass_model() for case of periodic key frames.
1112  if (example_pattern == 0) {
1113  set_frame_flags_bypass_mode_ex0(layer_id.temporal_layer_id,
1114  svc_ctx.spatial_layers, frame_cnt == 0,
1115  &ref_frame_config);
1116  } else if (example_pattern == 1) {
1117  set_frame_flags_bypass_mode_ex1(layer_id.temporal_layer_id,
1118  svc_ctx.spatial_layers, frame_cnt == 0,
1119  &ref_frame_config);
1120  }
1121  ref_frame_config.duration[0] = frame_duration * 1;
1122  ref_frame_config.duration[1] = frame_duration * 1;
1123 
1125  &ref_frame_config);
1126  // Keep track of input frames, to account for frame drops in rate control
1127  // stats/metrics.
1128  for (sl = 0; sl < enc_cfg.ss_number_layers; ++sl) {
1129  ++rc.layer_input_frames[sl * enc_cfg.ts_number_layers +
1130  layer_id.temporal_layer_id];
1131  }
1132  } else {
1133  // For the fixed pattern SVC, temporal layer is given by superframe count.
1134  unsigned int tl = 0;
1135  if (enc_cfg.ts_number_layers == 2)
1136  tl = (frame_cnt % 2 != 0);
1137  else if (enc_cfg.ts_number_layers == 3) {
1138  if (frame_cnt % 2 != 0) tl = 2;
1139  if ((frame_cnt > 1) && ((frame_cnt - 2) % 4 == 0)) tl = 1;
1140  }
1141  for (sl = 0; sl < enc_cfg.ss_number_layers; ++sl)
1142  ++rc.layer_input_frames[sl * enc_cfg.ts_number_layers + tl];
1143  }
1144 
1145  vpx_usec_timer_start(&timer);
1146  res = vpx_svc_encode(
1147  &svc_ctx, &encoder, (end_of_stream ? NULL : &raw), pts, frame_duration,
1148  svc_ctx.speed >= 5 ? VPX_DL_REALTIME : VPX_DL_GOOD_QUALITY);
1149  vpx_usec_timer_mark(&timer);
1150  cx_time += vpx_usec_timer_elapsed(&timer);
1151 
1152  fflush(stdout);
1153  if (res != VPX_CODEC_OK) {
1154  die_codec(&encoder, "Failed to encode frame");
1155  }
1156 
1157  while ((cx_pkt = vpx_codec_get_cx_data(&encoder, &iter)) != NULL) {
1158  switch (cx_pkt->kind) {
1159  case VPX_CODEC_CX_FRAME_PKT: {
1160  SvcInternal_t *const si = (SvcInternal_t *)svc_ctx.internal;
1161  if (cx_pkt->data.frame.sz > 0) {
1162  vpx_video_writer_write_frame(writer, cx_pkt->data.frame.buf,
1163  cx_pkt->data.frame.sz,
1164  cx_pkt->data.frame.pts);
1165 #if OUTPUT_RC_STATS
1166  if (svc_ctx.output_rc_stat) {
1167  svc_output_rc_stats(&encoder, &enc_cfg, &layer_id, cx_pkt, &rc,
1168  outfile, frame_cnt, framerate);
1169  }
1170 #endif
1171  }
1172  /*
1173  printf("SVC frame: %d, kf: %d, size: %d, pts: %d\n", frames_received,
1174  !!(cx_pkt->data.frame.flags & VPX_FRAME_IS_KEY),
1175  (int)cx_pkt->data.frame.sz, (int)cx_pkt->data.frame.pts);
1176  */
1177  if (enc_cfg.ss_number_layers == 1 && enc_cfg.ts_number_layers == 1)
1178  si->bytes_sum[0] += (int)cx_pkt->data.frame.sz;
1179  ++frames_received;
1180 #if CONFIG_VP9_DECODER && !SIMULCAST_MODE
1181  if (vpx_codec_decode(&decoder, cx_pkt->data.frame.buf,
1182  (unsigned int)cx_pkt->data.frame.sz, NULL, 0))
1183  die_codec(&decoder, "Failed to decode frame.");
1184 #endif
1185  break;
1186  }
1187  case VPX_CODEC_STATS_PKT: {
1188  stats_write(&app_input.rc_stats, cx_pkt->data.twopass_stats.buf,
1189  cx_pkt->data.twopass_stats.sz);
1190  break;
1191  }
1192  default: { break; }
1193  }
1194 
1195 #if CONFIG_VP9_DECODER && !SIMULCAST_MODE
1196  vpx_codec_control(&encoder, VP9E_GET_SVC_LAYER_ID, &layer_id);
1197  // Don't look for mismatch on top spatial and top temporal layers as they
1198  // are non reference frames.
1199  if ((enc_cfg.ss_number_layers > 1 || enc_cfg.ts_number_layers > 1) &&
1200  !(layer_id.temporal_layer_id > 0 &&
1201  layer_id.temporal_layer_id == (int)enc_cfg.ts_number_layers - 1 &&
1202  cx_pkt->data.frame
1203  .spatial_layer_encoded[enc_cfg.ss_number_layers - 1])) {
1204  test_decode(&encoder, &decoder, frame_cnt, &mismatch_seen);
1205  }
1206 #endif
1207  }
1208 
1209  if (!end_of_stream) {
1210  ++frame_cnt;
1211  pts += frame_duration;
1212  }
1213  }
1214 
1215  printf("Processed %d frames\n", frame_cnt);
1216 
1217  close_input_file(&app_input.input_ctx);
1218 
1219 #if OUTPUT_RC_STATS
1220  if (svc_ctx.output_rc_stat) {
1221  printout_rate_control_summary(&rc, &enc_cfg, frame_cnt);
1222  printf("\n");
1223  }
1224 #endif
1225  if (vpx_codec_destroy(&encoder))
1226  die_codec(&encoder, "Failed to destroy codec");
1227  if (app_input.passes == 2) stats_close(&app_input.rc_stats, 1);
1228  if (writer) {
1229  vpx_video_writer_close(writer);
1230  }
1231 #if OUTPUT_RC_STATS
1232  if (svc_ctx.output_rc_stat) {
1233  for (sl = 0; sl < enc_cfg.ss_number_layers; ++sl) {
1234  vpx_video_writer_close(outfile[sl]);
1235  }
1236  }
1237 #endif
1238 #if CONFIG_INTERNAL_STATS
1239  if (mismatch_seen) {
1240  fprintf(f, "First mismatch occurred in frame %d\n", mismatch_seen);
1241  } else {
1242  fprintf(f, "No mismatch detected in recon buffers\n");
1243  }
1244  fclose(f);
1245 #endif
1246  printf("Frame cnt and encoding time/FPS stats for encoding: %d %f %f \n",
1247  frame_cnt, 1000 * (float)cx_time / (double)(frame_cnt * 1000000),
1248  1000000 * (double)frame_cnt / (double)cx_time);
1249  if (app_input.input_ctx.file_type != FILE_TYPE_Y4M) {
1250  vpx_img_free(&raw);
1251  }
1252  // display average size, psnr
1253  vpx_svc_dump_statistics(&svc_ctx);
1254  vpx_svc_release(&svc_ctx);
1255  return EXIT_SUCCESS;
1256 }
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:830
#define VPX_DL_GOOD_QUALITY
deadline parameter analogous to VPx GOOD QUALITY mode.
Definition: vpx_encoder.h:832
#define VPX_MAX_LAYERS
Definition: vpx_encoder.h:43
#define VPX_FRAME_IS_KEY
Definition: vpx_encoder.h:116
#define VPX_SS_MAX_LAYERS
Definition: vpx_encoder.h:46
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:147
@ VPX_CODEC_STATS_PKT
Definition: vpx_encoder.h:148
@ VPX_RC_LAST_PASS
Definition: vpx_encoder.h:227
@ VPX_RC_ONE_PASS
Definition: vpx_encoder.h:225
@ VPX_RC_FIRST_PASS
Definition: vpx_encoder.h:226
@ VPX_CQ
Definition: vpx_encoder.h:234
@ FULL_SUPERFRAME_DROP
Definition: vp8cx.h:849
@ VP9E_SET_SVC_LAYER_ID
Codec control function to set svc layer for spatial and temporal.
Definition: vp8cx.h:453
@ VP8E_SET_MAX_INTRA_BITRATE_PCT
Codec control function to set Max data rate for Intra frames.
Definition: vp8cx.h:257
@ VP9E_SET_SVC_INTER_LAYER_PRED
Codec control function to constrain the inter-layer prediction (prediction of lower spatial resolutio...
Definition: vp8cx.h:619
@ VP9E_SET_AQ_MODE
Codec control function to set adaptive quantization mode.
Definition: vp8cx.h:398
@ VP9E_SET_TUNE_CONTENT
Codec control function to set content type.
Definition: vp8cx.h:463
@ VP9E_SET_ROW_MT
Codec control function to set row level multi-threading.
Definition: vp8cx.h:570
@ VP8E_SET_CPUUSED
Codec control function to set encoder internal speed settings.
Definition: vp8cx.h:155
@ VP9E_SET_TILE_COLUMNS
Codec control function to set number of tile columns.
Definition: vp8cx.h:351
@ 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:628
@ VP9E_SET_SVC_REF_FRAME_CONFIG
Codec control function to set the frame flags and buffer indices for spatial layers....
Definition: vp8cx.h:545
@ VP8E_SET_STATIC_THRESHOLD
Codec control function to set the threshold for MBs treated static.
Definition: vp8cx.h:188
@ VP9E_SET_NOISE_SENSITIVITY
Codec control function to set noise sensitivity.
Definition: vp8cx.h:421
@ VP9E_GET_SVC_LAYER_ID
Codec control function to get svc layer ID.
Definition: vp8cx.h:471
@ 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:716
@ 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:159
vpx_fixed_buf_t twopass_stats
Definition: vpx_encoder.h:182
enum vpx_codec_cx_pkt_kind kind
Definition: vpx_encoder.h:160
struct vpx_codec_cx_pkt::@1::@2 frame
union vpx_codec_cx_pkt::@1 data
Encoder configuration structure.
Definition: vpx_encoder.h:268
int temporal_layering_mode
Temporal layering mode indicating which temporal layering scheme to use.
Definition: vpx_encoder.h:693
unsigned int kf_min_dist
Keyframe minimum interval.
Definition: vpx_encoder.h:605
unsigned int ts_number_layers
Number of temporal coding layers.
Definition: vpx_encoder.h:644
vpx_fixed_buf_t rc_twopass_stats_in
Two-pass stats buffer.
Definition: vpx_encoder.h:447
unsigned int ss_number_layers
Number of spatial coding layers.
Definition: vpx_encoder.h:624
unsigned int rc_2pass_vbr_minsection_pct
Two-pass mode per-GOP minimum bitrate.
Definition: vpx_encoder.h:570
unsigned int g_profile
Bitstream profile to use.
Definition: vpx_encoder.h:295
unsigned int layer_target_bitrate[12]
Target bitrate for each spatial/temporal layer.
Definition: vpx_encoder.h:684
unsigned int g_h
Height of the frame.
Definition: vpx_encoder.h:313
vpx_codec_er_flags_t g_error_resilient
Enable error resilient modes.
Definition: vpx_encoder.h:351
unsigned int g_w
Width of the frame.
Definition: vpx_encoder.h:304
unsigned int rc_dropframe_thresh
Temporal resampling configuration, if supported by the codec.
Definition: vpx_encoder.h:391
struct vpx_rational g_timebase
Stream timebase units.
Definition: vpx_encoder.h:343
enum vpx_enc_pass g_pass
Multi-pass Encoding Mode.
Definition: vpx_encoder.h:358
unsigned int g_lag_in_frames
Allow lagged encoding.
Definition: vpx_encoder.h:372
enum vpx_rc_mode rc_end_usage
Rate control algorithm to use.
Definition: vpx_encoder.h:440
vpx_bit_depth_t g_bit_depth
Bit-depth of the codec.
Definition: vpx_encoder.h:321
unsigned int rc_2pass_vbr_maxsection_pct
Two-pass mode per-GOP maximum bitrate.
Definition: vpx_encoder.h:577
unsigned int rc_target_bitrate
Target data rate.
Definition: vpx_encoder.h:460
unsigned int g_input_bit_depth
Bit-depth of the input frames.
Definition: vpx_encoder.h:329
unsigned int ts_rate_decimator[5]
Frame rate decimation factor for each temporal layer.
Definition: vpx_encoder.h:658
unsigned int kf_max_dist
Keyframe maximum interval.
Definition: vpx_encoder.h:614
size_t sz
Definition: vpx_encoder.h:98
void * buf
Definition: vpx_encoder.h:97
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:220
int num
Definition: vpx_encoder.h:219
vp9 svc frame dropping parameters.
Definition: vp8cx.h:861
int framedrop_thresh[5]
Definition: vp8cx.h:862
SVC_LAYER_DROP_MODE framedrop_mode
Definition: vp8cx.h:864
int max_consec_drop
Definition: vp8cx.h:865
vp9 svc layer parameters
Definition: vp8cx.h:810
int temporal_layer_id
Definition: vp8cx.h:813
vp9 svc frame flag parameters.
Definition: vp8cx.h:825
int lst_fb_idx[5]
Definition: vp8cx.h:826
int update_buffer_slot[5]
Definition: vp8cx.h:829
int gld_fb_idx[5]
Definition: vp8cx.h:827
int reference_last[5]
Definition: vp8cx.h:834
int reference_golden[5]
Definition: vp8cx.h:835
int reference_alt_ref[5]
Definition: vp8cx.h:836
int64_t duration[5]
Definition: vp8cx.h:837
int alt_fb_idx[5]
Definition: vp8cx.h:828
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.