WebM Codec SDK
vp8_multi_resolution_encoder
1 /*
2  * Copyright (c) 2010 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 multi-resolution encoding in VP8.
13  * High-resolution input video is down-sampled to lower-resolutions. The
14  * encoder then encodes the video and outputs multiple bitstreams with
15  * different resolutions.
16  *
17  * This test also allows for settings temporal layers for each spatial layer.
18  * Different number of temporal layers per spatial stream may be used.
19  * Currently up to 3 temporal layers per spatial stream (encoder) are supported
20  * in this test.
21  */
22 
23 #include "./vpx_config.h"
24 
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <stdarg.h>
28 #include <string.h>
29 #include <math.h>
30 #include <assert.h>
31 #include <sys/time.h>
32 #if USE_POSIX_MMAP
33 #include <sys/types.h>
34 #include <sys/stat.h>
35 #include <sys/mman.h>
36 #include <fcntl.h>
37 #include <unistd.h>
38 #endif
39 #include "vpx_ports/vpx_timer.h"
40 #define VPX_CODEC_DISABLE_COMPAT 1
41 #include "vpx/vpx_encoder.h"
42 #include "vpx/vp8cx.h"
43 #include "vpx_ports/mem_ops.h"
44 #include "./tools_common.h"
45 #define interface (vpx_codec_vp8_cx())
46 #define fourcc 0x30385056
47 
48 void usage_exit() {
49  exit(EXIT_FAILURE);
50 }
51 
52 /*
53  * The input video frame is downsampled several times to generate a multi-level
54  * hierarchical structure. NUM_ENCODERS is defined as the number of encoding
55  * levels required. For example, if the size of input video is 1280x720,
56  * NUM_ENCODERS is 3, and down-sampling factor is 2, the encoder outputs 3
57  * bitstreams with resolution of 1280x720(level 0), 640x360(level 1), and
58  * 320x180(level 2) respectively.
59  */
60 
61 /* Number of encoders (spatial resolutions) used in this test. */
62 #define NUM_ENCODERS 3
63 
64 /* Maximum number of temporal layers allowed for this test. */
65 #define MAX_NUM_TEMPORAL_LAYERS 3
66 
67 /* This example uses the scaler function in libyuv. */
68 #include "third_party/libyuv/include/libyuv/basic_types.h"
69 #include "third_party/libyuv/include/libyuv/scale.h"
70 #include "third_party/libyuv/include/libyuv/cpu_id.h"
71 
72 int (*read_frame_p)(FILE *f, vpx_image_t *img);
73 
74 static int read_frame(FILE *f, vpx_image_t *img) {
75  size_t nbytes, to_read;
76  int res = 1;
77 
78  to_read = img->w*img->h*3/2;
79  nbytes = fread(img->planes[0], 1, to_read, f);
80  if(nbytes != to_read) {
81  res = 0;
82  if(nbytes > 0)
83  printf("Warning: Read partial frame. Check your width & height!\n");
84  }
85  return res;
86 }
87 
88 static int read_frame_by_row(FILE *f, vpx_image_t *img) {
89  size_t nbytes, to_read;
90  int res = 1;
91  int plane;
92 
93  for (plane = 0; plane < 3; plane++)
94  {
95  unsigned char *ptr;
96  int w = (plane ? (1 + img->d_w) / 2 : img->d_w);
97  int h = (plane ? (1 + img->d_h) / 2 : img->d_h);
98  int r;
99 
100  /* Determine the correct plane based on the image format. The for-loop
101  * always counts in Y,U,V order, but this may not match the order of
102  * the data on disk.
103  */
104  switch (plane)
105  {
106  case 1:
107  ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U];
108  break;
109  case 2:
110  ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V];
111  break;
112  default:
113  ptr = img->planes[plane];
114  }
115 
116  for (r = 0; r < h; r++)
117  {
118  to_read = w;
119 
120  nbytes = fread(ptr, 1, to_read, f);
121  if(nbytes != to_read) {
122  res = 0;
123  if(nbytes > 0)
124  printf("Warning: Read partial frame. Check your width & height!\n");
125  break;
126  }
127 
128  ptr += img->stride[plane];
129  }
130  if (!res)
131  break;
132  }
133 
134  return res;
135 }
136 
137 static void write_ivf_file_header(FILE *outfile,
138  const vpx_codec_enc_cfg_t *cfg,
139  int frame_cnt) {
140  char header[32];
141 
142  if(cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS)
143  return;
144  header[0] = 'D';
145  header[1] = 'K';
146  header[2] = 'I';
147  header[3] = 'F';
148  mem_put_le16(header+4, 0); /* version */
149  mem_put_le16(header+6, 32); /* headersize */
150  mem_put_le32(header+8, fourcc); /* headersize */
151  mem_put_le16(header+12, cfg->g_w); /* width */
152  mem_put_le16(header+14, cfg->g_h); /* height */
153  mem_put_le32(header+16, cfg->g_timebase.den); /* rate */
154  mem_put_le32(header+20, cfg->g_timebase.num); /* scale */
155  mem_put_le32(header+24, frame_cnt); /* length */
156  mem_put_le32(header+28, 0); /* unused */
157 
158  (void) fwrite(header, 1, 32, outfile);
159 }
160 
161 static void write_ivf_frame_header(FILE *outfile,
162  const vpx_codec_cx_pkt_t *pkt)
163 {
164  char header[12];
165  vpx_codec_pts_t pts;
166 
167  if(pkt->kind != VPX_CODEC_CX_FRAME_PKT)
168  return;
169 
170  pts = pkt->data.frame.pts;
171  mem_put_le32(header, pkt->data.frame.sz);
172  mem_put_le32(header+4, pts&0xFFFFFFFF);
173  mem_put_le32(header+8, pts >> 32);
174 
175  (void) fwrite(header, 1, 12, outfile);
176 }
177 
178 /* Temporal scaling parameters */
179 /* This sets all the temporal layer parameters given |num_temporal_layers|,
180  * including the target bit allocation across temporal layers. Bit allocation
181  * parameters will be passed in as user parameters in another version.
182  */
183 static void set_temporal_layer_pattern(int num_temporal_layers,
184  vpx_codec_enc_cfg_t *cfg,
185  int bitrate,
186  int *layer_flags)
187 {
188  assert(num_temporal_layers <= MAX_NUM_TEMPORAL_LAYERS);
189  switch (num_temporal_layers)
190  {
191  case 1:
192  {
193  /* 1-layer */
194  cfg->ts_number_layers = 1;
195  cfg->ts_periodicity = 1;
196  cfg->ts_rate_decimator[0] = 1;
197  cfg->ts_layer_id[0] = 0;
198  cfg->ts_target_bitrate[0] = bitrate;
199 
200  // Update L only.
201  layer_flags[0] = VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
202  break;
203  }
204 
205  case 2:
206  {
207  /* 2-layers, with sync point at first frame of layer 1. */
208  cfg->ts_number_layers = 2;
209  cfg->ts_periodicity = 2;
210  cfg->ts_rate_decimator[0] = 2;
211  cfg->ts_rate_decimator[1] = 1;
212  cfg->ts_layer_id[0] = 0;
213  cfg->ts_layer_id[1] = 1;
214  // Use 60/40 bit allocation as example.
215  cfg->ts_target_bitrate[0] = 0.6f * bitrate;
216  cfg->ts_target_bitrate[1] = bitrate;
217 
218  /* 0=L, 1=GF */
219  // ARF is used as predictor for all frames, and is only updated on
220  // key frame. Sync point every 8 frames.
221 
222  // Layer 0: predict from L and ARF, update L and G.
223  layer_flags[0] = VP8_EFLAG_NO_REF_GF |
225 
226  // Layer 1: sync point: predict from L and ARF, and update G.
227  layer_flags[1] = VP8_EFLAG_NO_REF_GF |
230 
231  // Layer 0, predict from L and ARF, update L.
232  layer_flags[2] = VP8_EFLAG_NO_REF_GF |
235 
236  // Layer 1: predict from L, G and ARF, and update G.
237  layer_flags[3] = VP8_EFLAG_NO_UPD_ARF |
240 
241  // Layer 0
242  layer_flags[4] = layer_flags[2];
243 
244  // Layer 1
245  layer_flags[5] = layer_flags[3];
246 
247  // Layer 0
248  layer_flags[6] = layer_flags[4];
249 
250  // Layer 1
251  layer_flags[7] = layer_flags[5];
252  break;
253  }
254 
255  case 3:
256  default:
257  {
258  // 3-layers structure where ARF is used as predictor for all frames,
259  // and is only updated on key frame.
260  // Sync points for layer 1 and 2 every 8 frames.
261  cfg->ts_number_layers = 3;
262  cfg->ts_periodicity = 4;
263  cfg->ts_rate_decimator[0] = 4;
264  cfg->ts_rate_decimator[1] = 2;
265  cfg->ts_rate_decimator[2] = 1;
266  cfg->ts_layer_id[0] = 0;
267  cfg->ts_layer_id[1] = 2;
268  cfg->ts_layer_id[2] = 1;
269  cfg->ts_layer_id[3] = 2;
270  // Use 40/20/40 bit allocation as example.
271  cfg->ts_target_bitrate[0] = 0.4f * bitrate;
272  cfg->ts_target_bitrate[1] = 0.6f * bitrate;
273  cfg->ts_target_bitrate[2] = bitrate;
274 
275  /* 0=L, 1=GF, 2=ARF */
276 
277  // Layer 0: predict from L and ARF; update L and G.
278  layer_flags[0] = VP8_EFLAG_NO_UPD_ARF |
280 
281  // Layer 2: sync point: predict from L and ARF; update none.
282  layer_flags[1] = VP8_EFLAG_NO_REF_GF |
287 
288  // Layer 1: sync point: predict from L and ARF; update G.
289  layer_flags[2] = VP8_EFLAG_NO_REF_GF |
292 
293  // Layer 2: predict from L, G, ARF; update none.
294  layer_flags[3] = VP8_EFLAG_NO_UPD_GF |
298 
299  // Layer 0: predict from L and ARF; update L.
300  layer_flags[4] = VP8_EFLAG_NO_UPD_GF |
303 
304  // Layer 2: predict from L, G, ARF; update none.
305  layer_flags[5] = layer_flags[3];
306 
307  // Layer 1: predict from L, G, ARF; update G.
308  layer_flags[6] = VP8_EFLAG_NO_UPD_ARF |
310 
311  // Layer 2: predict from L, G, ARF; update none.
312  layer_flags[7] = layer_flags[3];
313  break;
314  }
315  }
316 }
317 
318 /* The periodicity of the pattern given the number of temporal layers. */
319 static int periodicity_to_num_layers[MAX_NUM_TEMPORAL_LAYERS] = {1, 8, 8};
320 
321 int main(int argc, char **argv)
322 {
323  FILE *infile, *outfile[NUM_ENCODERS];
324  FILE *downsampled_input[NUM_ENCODERS - 1];
325  char filename[50];
326  vpx_codec_ctx_t codec[NUM_ENCODERS];
327  vpx_codec_enc_cfg_t cfg[NUM_ENCODERS];
328  int frame_cnt = 0;
329  vpx_image_t raw[NUM_ENCODERS];
330  vpx_codec_err_t res[NUM_ENCODERS];
331 
332  int i;
333  long width;
334  long height;
335  int length_frame;
336  int frame_avail;
337  int got_data;
338  int flags = 0;
339  int layer_id = 0;
340 
341  int layer_flags[VPX_TS_MAX_PERIODICITY * NUM_ENCODERS]
342  = {0};
343  int flag_periodicity;
344 
345  /*Currently, only realtime mode is supported in multi-resolution encoding.*/
346  int arg_deadline = VPX_DL_REALTIME;
347 
348  /* Set show_psnr to 1/0 to show/not show PSNR. Choose show_psnr=0 if you
349  don't need to know PSNR, which will skip PSNR calculation and save
350  encoding time. */
351  int show_psnr = 0;
352  int key_frame_insert = 0;
353  uint64_t psnr_sse_total[NUM_ENCODERS] = {0};
354  uint64_t psnr_samples_total[NUM_ENCODERS] = {0};
355  double psnr_totals[NUM_ENCODERS][4] = {{0,0}};
356  int psnr_count[NUM_ENCODERS] = {0};
357 
358  double cx_time = 0;
359  struct timeval tv1, tv2, difftv;
360 
361  /* Set the required target bitrates for each resolution level.
362  * If target bitrate for highest-resolution level is set to 0,
363  * (i.e. target_bitrate[0]=0), we skip encoding at that level.
364  */
365  unsigned int target_bitrate[NUM_ENCODERS]={1000, 500, 100};
366 
367  /* Enter the frame rate of the input video */
368  int framerate = 30;
369 
370  /* Set down-sampling factor for each resolution level.
371  dsf[0] controls down sampling from level 0 to level 1;
372  dsf[1] controls down sampling from level 1 to level 2;
373  dsf[2] is not used. */
374  vpx_rational_t dsf[NUM_ENCODERS] = {{2, 1}, {2, 1}, {1, 1}};
375 
376  /* Set the number of temporal layers for each encoder/resolution level,
377  * starting from highest resoln down to lowest resoln. */
378  unsigned int num_temporal_layers[NUM_ENCODERS] = {3, 3, 3};
379 
380  if(argc!= (7 + 3 * NUM_ENCODERS))
381  die("Usage: %s <width> <height> <frame_rate> <infile> <outfile(s)> "
382  "<rate_encoder(s)> <temporal_layer(s)> <key_frame_insert> <output psnr?> \n",
383  argv[0]);
384 
385  printf("Using %s\n",vpx_codec_iface_name(interface));
386 
387  width = strtol(argv[1], NULL, 0);
388  height = strtol(argv[2], NULL, 0);
389  framerate = strtol(argv[3], NULL, 0);
390 
391  if(width < 16 || width%2 || height <16 || height%2)
392  die("Invalid resolution: %ldx%ld", width, height);
393 
394  /* Open input video file for encoding */
395  if(!(infile = fopen(argv[4], "rb")))
396  die("Failed to open %s for reading", argv[4]);
397 
398  /* Open output file for each encoder to output bitstreams */
399  for (i=0; i< NUM_ENCODERS; i++)
400  {
401  if(!target_bitrate[i])
402  {
403  outfile[i] = NULL;
404  continue;
405  }
406 
407  if(!(outfile[i] = fopen(argv[i+5], "wb")))
408  die("Failed to open %s for writing", argv[i+4]);
409  }
410 
411  // Bitrates per spatial layer: overwrite default rates above.
412  for (i=0; i< NUM_ENCODERS; i++)
413  {
414  target_bitrate[i] = strtol(argv[NUM_ENCODERS + 5 + i], NULL, 0);
415  }
416 
417  // Temporal layers per spatial layers: overwrite default settings above.
418  for (i=0; i< NUM_ENCODERS; i++)
419  {
420  num_temporal_layers[i] = strtol(argv[2 * NUM_ENCODERS + 5 + i], NULL, 0);
421  if (num_temporal_layers[i] < 1 || num_temporal_layers[i] > 3)
422  die("Invalid temporal layers: %d, Must be 1, 2, or 3. \n",
423  num_temporal_layers);
424  }
425 
426  /* Open file to write out each spatially downsampled input stream. */
427  for (i=0; i< NUM_ENCODERS - 1; i++)
428  {
429  // Highest resoln is encoder 0.
430  if (sprintf(filename,"ds%d.yuv",NUM_ENCODERS - i) < 0)
431  {
432  return EXIT_FAILURE;
433  }
434  downsampled_input[i] = fopen(filename,"wb");
435  }
436 
437  key_frame_insert = strtol(argv[3 * NUM_ENCODERS + 5], NULL, 0);
438 
439  show_psnr = strtol(argv[3 * NUM_ENCODERS + 6], NULL, 0);
440 
441 
442  /* Populate default encoder configuration */
443  for (i=0; i< NUM_ENCODERS; i++)
444  {
445  res[i] = vpx_codec_enc_config_default(interface, &cfg[i], 0);
446  if(res[i]) {
447  printf("Failed to get config: %s\n", vpx_codec_err_to_string(res[i]));
448  return EXIT_FAILURE;
449  }
450  }
451 
452  /*
453  * Update the default configuration according to needs of the application.
454  */
455  /* Highest-resolution encoder settings */
456  cfg[0].g_w = width;
457  cfg[0].g_h = height;
458  cfg[0].rc_dropframe_thresh = 0;
459  cfg[0].rc_end_usage = VPX_CBR;
460  cfg[0].rc_resize_allowed = 0;
461  cfg[0].rc_min_quantizer = 2;
462  cfg[0].rc_max_quantizer = 56;
463  cfg[0].rc_undershoot_pct = 100;
464  cfg[0].rc_overshoot_pct = 15;
465  cfg[0].rc_buf_initial_sz = 500;
466  cfg[0].rc_buf_optimal_sz = 600;
467  cfg[0].rc_buf_sz = 1000;
468  cfg[0].g_error_resilient = 1; /* Enable error resilient mode */
469  cfg[0].g_lag_in_frames = 0;
470 
471  /* Disable automatic keyframe placement */
472  /* Note: These 3 settings are copied to all levels. But, except the lowest
473  * resolution level, all other levels are set to VPX_KF_DISABLED internally.
474  */
475  cfg[0].kf_mode = VPX_KF_AUTO;
476  cfg[0].kf_min_dist = 3000;
477  cfg[0].kf_max_dist = 3000;
478 
479  cfg[0].rc_target_bitrate = target_bitrate[0]; /* Set target bitrate */
480  cfg[0].g_timebase.num = 1; /* Set fps */
481  cfg[0].g_timebase.den = framerate;
482 
483  /* Other-resolution encoder settings */
484  for (i=1; i< NUM_ENCODERS; i++)
485  {
486  memcpy(&cfg[i], &cfg[0], sizeof(vpx_codec_enc_cfg_t));
487 
488  cfg[i].rc_target_bitrate = target_bitrate[i];
489 
490  /* Note: Width & height of other-resolution encoders are calculated
491  * from the highest-resolution encoder's size and the corresponding
492  * down_sampling_factor.
493  */
494  {
495  unsigned int iw = cfg[i-1].g_w*dsf[i-1].den + dsf[i-1].num - 1;
496  unsigned int ih = cfg[i-1].g_h*dsf[i-1].den + dsf[i-1].num - 1;
497  cfg[i].g_w = iw/dsf[i-1].num;
498  cfg[i].g_h = ih/dsf[i-1].num;
499  }
500 
501  /* Make width & height to be multiplier of 2. */
502  // Should support odd size ???
503  if((cfg[i].g_w)%2)cfg[i].g_w++;
504  if((cfg[i].g_h)%2)cfg[i].g_h++;
505  }
506 
507 
508  // Set the number of threads per encode/spatial layer.
509  // (1, 1, 1) means no encoder threading.
510  cfg[0].g_threads = 2;
511  cfg[1].g_threads = 1;
512  cfg[2].g_threads = 1;
513 
514  /* Allocate image for each encoder */
515  for (i=0; i< NUM_ENCODERS; i++)
516  if(!vpx_img_alloc(&raw[i], VPX_IMG_FMT_I420, cfg[i].g_w, cfg[i].g_h, 32))
517  die("Failed to allocate image", cfg[i].g_w, cfg[i].g_h);
518 
519  if (raw[0].stride[VPX_PLANE_Y] == raw[0].d_w)
520  read_frame_p = read_frame;
521  else
522  read_frame_p = read_frame_by_row;
523 
524  for (i=0; i< NUM_ENCODERS; i++)
525  if(outfile[i])
526  write_ivf_file_header(outfile[i], &cfg[i], 0);
527 
528  /* Temporal layers settings */
529  for ( i=0; i<NUM_ENCODERS; i++)
530  {
531  set_temporal_layer_pattern(num_temporal_layers[i],
532  &cfg[i],
533  cfg[i].rc_target_bitrate,
534  &layer_flags[i * VPX_TS_MAX_PERIODICITY]);
535  }
536 
537  /* Initialize multi-encoder */
538  if(vpx_codec_enc_init_multi(&codec[0], interface, &cfg[0], NUM_ENCODERS,
539  (show_psnr ? VPX_CODEC_USE_PSNR : 0), &dsf[0]))
540  die_codec(&codec[0], "Failed to initialize encoder");
541 
542  /* The extra encoding configuration parameters can be set as follows. */
543  /* Set encoding speed */
544  for ( i=0; i<NUM_ENCODERS; i++)
545  {
546  int speed = -6;
547  /* Lower speed for the lowest resolution. */
548  if (i == NUM_ENCODERS - 1) speed = -4;
549  if(vpx_codec_control(&codec[i], VP8E_SET_CPUUSED, speed))
550  die_codec(&codec[i], "Failed to set cpu_used");
551  }
552 
553  /* Set static threshold = 1 for all encoders */
554  for ( i=0; i<NUM_ENCODERS; i++)
555  {
557  die_codec(&codec[i], "Failed to set static threshold");
558  }
559 
560  /* Set NOISE_SENSITIVITY to do TEMPORAL_DENOISING */
561  /* Enable denoising for the highest-resolution encoder. */
563  die_codec(&codec[0], "Failed to set noise_sensitivity");
564  for ( i=1; i< NUM_ENCODERS; i++)
565  {
567  die_codec(&codec[i], "Failed to set noise_sensitivity");
568  }
569 
570  /* Set the number of token partitions */
571  for ( i=0; i<NUM_ENCODERS; i++)
572  {
574  die_codec(&codec[i], "Failed to set static threshold");
575  }
576 
577  /* Set the max intra target bitrate */
578  for ( i=0; i<NUM_ENCODERS; i++)
579  {
580  unsigned int max_intra_size_pct =
581  (int)(((double)cfg[0].rc_buf_optimal_sz * 0.5) * framerate / 10);
583  max_intra_size_pct))
584  die_codec(&codec[i], "Failed to set static threshold");
585  //printf("%d %d \n",i,max_intra_size_pct);
586  }
587 
588  frame_avail = 1;
589  got_data = 0;
590 
591  while(frame_avail || got_data)
592  {
593  vpx_codec_iter_t iter[NUM_ENCODERS]={NULL};
594  const vpx_codec_cx_pkt_t *pkt[NUM_ENCODERS];
595 
596  flags = 0;
597  frame_avail = read_frame_p(infile, &raw[0]);
598 
599  if(frame_avail)
600  {
601  for ( i=1; i<NUM_ENCODERS; i++)
602  {
603  /*Scale the image down a number of times by downsampling factor*/
604  /* FilterMode 1 or 2 give better psnr than FilterMode 0. */
605  I420Scale(raw[i-1].planes[VPX_PLANE_Y], raw[i-1].stride[VPX_PLANE_Y],
606  raw[i-1].planes[VPX_PLANE_U], raw[i-1].stride[VPX_PLANE_U],
607  raw[i-1].planes[VPX_PLANE_V], raw[i-1].stride[VPX_PLANE_V],
608  raw[i-1].d_w, raw[i-1].d_h,
609  raw[i].planes[VPX_PLANE_Y], raw[i].stride[VPX_PLANE_Y],
610  raw[i].planes[VPX_PLANE_U], raw[i].stride[VPX_PLANE_U],
611  raw[i].planes[VPX_PLANE_V], raw[i].stride[VPX_PLANE_V],
612  raw[i].d_w, raw[i].d_h, 1);
613  /* Write out down-sampled input. */
614  length_frame = cfg[i].g_w * cfg[i].g_h *3/2;
615  if (fwrite(raw[i].planes[0], 1, length_frame,
616  downsampled_input[NUM_ENCODERS - i - 1]) !=
617  length_frame)
618  {
619  return EXIT_FAILURE;
620  }
621  }
622  }
623 
624  /* Set the flags (reference and update) for all the encoders.*/
625  for ( i=0; i<NUM_ENCODERS; i++)
626  {
627  layer_id = cfg[i].ts_layer_id[frame_cnt % cfg[i].ts_periodicity];
628  flags = 0;
629  flag_periodicity = periodicity_to_num_layers
630  [num_temporal_layers[i] - 1];
631  flags = layer_flags[i * VPX_TS_MAX_PERIODICITY +
632  frame_cnt % flag_periodicity];
633  // Key frame flag for first frame.
634  if (frame_cnt == 0)
635  {
636  flags |= VPX_EFLAG_FORCE_KF;
637  }
638  if (frame_cnt > 0 && frame_cnt == key_frame_insert)
639  {
640  flags = VPX_EFLAG_FORCE_KF;
641  }
642 
643  vpx_codec_control(&codec[i], VP8E_SET_FRAME_FLAGS, flags);
644  vpx_codec_control(&codec[i], VP8E_SET_TEMPORAL_LAYER_ID, layer_id);
645  }
646 
647  gettimeofday(&tv1, NULL);
648  /* Encode each frame at multi-levels */
649  /* Note the flags must be set to 0 in the encode call if they are set
650  for each frame with the vpx_codec_control(), as done above. */
651  if(vpx_codec_encode(&codec[0], frame_avail? &raw[0] : NULL,
652  frame_cnt, 1, 0, arg_deadline))
653  {
654  die_codec(&codec[0], "Failed to encode frame");
655  }
656  gettimeofday(&tv2, NULL);
657  timersub(&tv2, &tv1, &difftv);
658  cx_time += (double)(difftv.tv_sec * 1000000 + difftv.tv_usec);
659  for (i=NUM_ENCODERS-1; i>=0 ; i--)
660  {
661  got_data = 0;
662  while( (pkt[i] = vpx_codec_get_cx_data(&codec[i], &iter[i])) )
663  {
664  got_data = 1;
665  switch(pkt[i]->kind) {
667  write_ivf_frame_header(outfile[i], pkt[i]);
668  (void) fwrite(pkt[i]->data.frame.buf, 1,
669  pkt[i]->data.frame.sz, outfile[i]);
670  break;
671  case VPX_CODEC_PSNR_PKT:
672  if (show_psnr)
673  {
674  int j;
675 
676  psnr_sse_total[i] += pkt[i]->data.psnr.sse[0];
677  psnr_samples_total[i] += pkt[i]->data.psnr.samples[0];
678  for (j = 0; j < 4; j++)
679  {
680  psnr_totals[i][j] += pkt[i]->data.psnr.psnr[j];
681  }
682  psnr_count[i]++;
683  }
684 
685  break;
686  default:
687  break;
688  }
689  printf(pkt[i]->kind == VPX_CODEC_CX_FRAME_PKT
690  && (pkt[i]->data.frame.flags & VPX_FRAME_IS_KEY)? "K":"");
691  fflush(stdout);
692  }
693  }
694  frame_cnt++;
695  }
696  printf("\n");
697  printf("FPS for encoding %d %f %f \n", frame_cnt, (float)cx_time / 1000000,
698  1000000 * (double)frame_cnt / (double)cx_time);
699 
700  fclose(infile);
701 
702  printf("Processed %ld frames.\n",(long int)frame_cnt-1);
703  for (i=0; i< NUM_ENCODERS; i++)
704  {
705  /* Calculate PSNR and print it out */
706  if ( (show_psnr) && (psnr_count[i]>0) )
707  {
708  int j;
709  double ovpsnr = sse_to_psnr(psnr_samples_total[i], 255.0,
710  psnr_sse_total[i]);
711 
712  fprintf(stderr, "\n ENC%d PSNR (Overall/Avg/Y/U/V)", i);
713 
714  fprintf(stderr, " %.3lf", ovpsnr);
715  for (j = 0; j < 4; j++)
716  {
717  fprintf(stderr, " %.3lf", psnr_totals[i][j]/psnr_count[i]);
718  }
719  }
720 
721  if(vpx_codec_destroy(&codec[i]))
722  die_codec(&codec[i], "Failed to destroy codec");
723 
724  vpx_img_free(&raw[i]);
725 
726  if(!outfile[i])
727  continue;
728 
729  /* Try to rewrite the file header with the actual frame count */
730  if(!fseek(outfile[i], 0, SEEK_SET))
731  write_ivf_file_header(outfile[i], &cfg[i], frame_cnt-1);
732  fclose(outfile[i]);
733  }
734  printf("\n");
735 
736  return EXIT_SUCCESS;
737 }
Rational Number.
Definition: vpx_encoder.h:256
unsigned int rc_buf_initial_sz
Decoder Buffer Initial Size.
Definition: vpx_encoder.h:607
unsigned int ts_number_layers
Number of temporal coding layers.
Definition: vpx_encoder.h:712
Codec control function to set encoder internal speed settings.
Definition: vp8cx.h:183
#define VP8_EFLAG_NO_UPD_GF
Don't update the golden frame.
Definition: vp8cx.h:93
Image Descriptor.
Definition: vpx_image.h:82
Describes the encoder algorithm interface to applications.
const char * vpx_codec_iface_name(vpx_codec_iface_t *iface)
Return the name for a given interface.
Definition: vpx_image.h:55
const char * vpx_codec_err_to_string(vpx_codec_err_t err)
Convert error number to printable string.
struct vpx_rational g_timebase
Stream timebase units.
Definition: vpx_encoder.h:394
Definition: vpx_encoder.h:273
unsigned int rc_buf_sz
Decoder Buffer Size.
Definition: vpx_encoder.h:597
#define VP8_EFLAG_NO_REF_GF
Don't reference the golden frame.
Definition: vp8cx.h:68
Codec control function to set reference and update frame flags.
Definition: vp8cx.h:285
enum vpx_kf_mode kf_mode
Keyframe placement mode.
Definition: vpx_encoder.h:662
int den
Definition: vpx_encoder.h:258
vpx_codec_err_t vpx_codec_encode(vpx_codec_ctx_t *ctx, const vpx_image_t *img, vpx_codec_pts_t pts, unsigned long duration, vpx_enc_frame_flags_t flags, unsigned long deadline)
Encode a frame.
unsigned int rc_max_quantizer
Maximum (Worst Quality) Quantizer.
Definition: vpx_encoder.h:549
unsigned int rc_min_quantizer
Minimum (Best Quality) Quantizer.
Definition: vpx_encoder.h:538
unsigned int kf_max_dist
Keyframe maximum interval.
Definition: vpx_encoder.h:682
unsigned int g_lag_in_frames
Allow lagged encoding.
Definition: vpx_encoder.h:426
Encoder configuration structure.
Definition: vpx_encoder.h:311
Definition: vpx_encoder.h:176
Definition: vpx_encoder.h:289
Codec control function to set Max data rate for Intra frames.
Definition: vp8cx.h:279
Encoder output packet.
Definition: vpx_encoder.h:192
unsigned int rc_overshoot_pct
Rate control adaptation overshoot control.
Definition: vpx_encoder.h:580
unsigned int ts_rate_decimator[5]
Frame rate decimation factor for each temporal layer.
Definition: vpx_encoder.h:726
unsigned int rc_buf_optimal_sz
Decoder Buffer Optimal Size.
Definition: vpx_encoder.h:617
#define VPX_PLANE_V
Definition: vpx_image.h:103
unsigned int kf_min_dist
Keyframe minimum interval.
Definition: vpx_encoder.h:672
Definition: vpx_encoder.h:266
unsigned int ts_layer_id[16]
Template defining the membership of frames to temporal layers.
Definition: vpx_encoder.h:744
struct vpx_codec_cx_pkt::@1::@2 frame
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.
Definition: vpx_image.h:56
unsigned int d_w
Definition: vpx_image.h:92
unsigned int g_w
Width of the frame.
Definition: vpx_encoder.h:354
unsigned int ts_target_bitrate[5]
Target bitrate for each temporal layer.
Definition: vpx_encoder.h:719
unsigned int rc_undershoot_pct
Rate control adaptation undershoot control.
Definition: vpx_encoder.h:567
unsigned int g_h
Height of the frame.
Definition: vpx_encoder.h:364
int stride[4]
Definition: vpx_image.h:106
enum vpx_codec_cx_pkt_kind kind
Definition: vpx_encoder.h:193
unsigned int rc_dropframe_thresh
Temporal resampling configuration, if supported by the codec.
Definition: vpx_encoder.h:449
Codec control function to set the temporal layer id.
Definition: vp8cx.h:326
#define VP8_EFLAG_NO_UPD_LAST
Don't update the last frame.
Definition: vp8cx.h:85
void vpx_img_free(vpx_image_t *img)
Close an image descriptor.
vpx_img_fmt_t fmt
Definition: vpx_image.h:83
unsigned char * planes[4]
Definition: vpx_image.h:105
Codec control function to set the number of token partitions.
Definition: vp8cx.h:216
unsigned int rc_target_bitrate
Target data rate.
Definition: vpx_encoder.h:522
#define VPX_DL_REALTIME
Definition: vpx_encoder.h:891
int num
Definition: vpx_encoder.h:257
control function to set noise sensitivity
Definition: vp8cx.h:198
enum vpx_enc_pass g_pass
Multi-pass Encoding Mode.
Definition: vpx_encoder.h:411
double psnr[4]
Definition: vpx_encoder.h:216
unsigned int g_threads
Maximum number of threads to use.
Definition: vpx_encoder.h:332
Provides definitions for using VP8 or VP9 encoder algorithm within the vpx Codec Interface.
#define VPX_PLANE_U
Definition: vpx_image.h:102
unsigned int rc_resize_allowed
Enable/disable spatial resampling, if supported by the codec.
Definition: vpx_encoder.h:459
unsigned int h
Definition: vpx_image.h:88
vpx_codec_err_t
Algorithm return codes.
Definition: vpx_codec.h:89
const vpx_codec_cx_pkt_t * vpx_codec_get_cx_data(vpx_codec_ctx_t *ctx, vpx_codec_iter_t *iter)
Encoded data iterator.
union vpx_codec_cx_pkt::@1 data
#define vpx_codec_enc_init_multi(ctx, iface, cfg, num_enc, flags, dsf)
Convenience macro for vpx_codec_enc_init_multi_ver()
Definition: vpx_encoder.h:830
int64_t vpx_codec_pts_t
Time Stamp Type.
Definition: vpx_encoder.h:116
vpx_codec_err_t vpx_codec_enc_config_default(vpx_codec_iface_t *iface, vpx_codec_enc_cfg_t *cfg, unsigned int reserved)
Get a default configuration.
#define VPX_TS_MAX_PERIODICITY
Definition: vpx_encoder.h:37
#define vpx_codec_control(ctx, id, data)
vpx_codec_control wrapper macro
Definition: vpx_codec.h:407
unsigned int ts_periodicity
Length of the sequence defining frame temporal layer membership.
Definition: vpx_encoder.h:735
vpx_codec_err_t vpx_codec_destroy(vpx_codec_ctx_t *ctx)
Destroy a codec instance.
unsigned int d_h
Definition: vpx_image.h:93
unsigned int w
Definition: vpx_image.h:87
Codec control function to set the threshold for MBs treated static.
Definition: vp8cx.h:210
#define VPX_FRAME_IS_KEY
Definition: vpx_encoder.h:127
#define VPX_EFLAG_FORCE_KF
Definition: vpx_encoder.h:302
const void * vpx_codec_iter_t
Iterator.
Definition: vpx_codec.h:188
Definition: vpx_encoder.h:173
vpx_codec_er_flags_t g_error_resilient
Enable error resilient modes.
Definition: vpx_encoder.h:403
#define VP8_EFLAG_NO_UPD_ARF
Don't update the alternate reference frame.
Definition: vp8cx.h:101
#define VP8_EFLAG_NO_UPD_ENTROPY
Disable entropy update.
Definition: vp8cx.h:125
enum vpx_rc_mode rc_end_usage
Rate control algorithm to use.
Definition: vpx_encoder.h:501
Definition: vpx_encoder.h:264
Codec context structure.
Definition: vpx_codec.h:199