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 #include "vpx/vpx_encoder.h"
41 #include "vpx/vp8cx.h"
42 #include "vpx_ports/mem_ops.h"
43 #include "../tools_common.h"
44 #define interface (vpx_codec_vp8_cx())
45 #define fourcc 0x30385056
46 
47 void usage_exit(void) {
48  exit(EXIT_FAILURE);
49 }
50 
51 /*
52  * The input video frame is downsampled several times to generate a multi-level
53  * hierarchical structure. NUM_ENCODERS is defined as the number of encoding
54  * levels required. For example, if the size of input video is 1280x720,
55  * NUM_ENCODERS is 3, and down-sampling factor is 2, the encoder outputs 3
56  * bitstreams with resolution of 1280x720(level 0), 640x360(level 1), and
57  * 320x180(level 2) respectively.
58  */
59 
60 /* Number of encoders (spatial resolutions) used in this test. */
61 #define NUM_ENCODERS 3
62 
63 /* Maximum number of temporal layers allowed for this test. */
64 #define MAX_NUM_TEMPORAL_LAYERS 3
65 
66 /* This example uses the scaler function in libyuv. */
67 #include "third_party/libyuv/include/libyuv/basic_types.h"
68 #include "third_party/libyuv/include/libyuv/scale.h"
69 #include "third_party/libyuv/include/libyuv/cpu_id.h"
70 
71 int (*read_frame_p)(FILE *f, vpx_image_t *img);
72 
73 static int read_frame(FILE *f, vpx_image_t *img) {
74  size_t nbytes, to_read;
75  int res = 1;
76 
77  to_read = img->w*img->h*3/2;
78  nbytes = fread(img->planes[0], 1, to_read, f);
79  if(nbytes != to_read) {
80  res = 0;
81  if(nbytes > 0)
82  printf("Warning: Read partial frame. Check your width & height!\n");
83  }
84  return res;
85 }
86 
87 static int read_frame_by_row(FILE *f, vpx_image_t *img) {
88  size_t nbytes, to_read;
89  int res = 1;
90  int plane;
91 
92  for (plane = 0; plane < 3; plane++)
93  {
94  unsigned char *ptr;
95  int w = (plane ? (1 + img->d_w) / 2 : img->d_w);
96  int h = (plane ? (1 + img->d_h) / 2 : img->d_h);
97  int r;
98 
99  /* Determine the correct plane based on the image format. The for-loop
100  * always counts in Y,U,V order, but this may not match the order of
101  * the data on disk.
102  */
103  switch (plane)
104  {
105  case 1:
106  ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U];
107  break;
108  case 2:
109  ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V];
110  break;
111  default:
112  ptr = img->planes[plane];
113  }
114 
115  for (r = 0; r < h; r++)
116  {
117  to_read = w;
118 
119  nbytes = fread(ptr, 1, to_read, f);
120  if(nbytes != to_read) {
121  res = 0;
122  if(nbytes > 0)
123  printf("Warning: Read partial frame. Check your width & height!\n");
124  break;
125  }
126 
127  ptr += img->stride[plane];
128  }
129  if (!res)
130  break;
131  }
132 
133  return res;
134 }
135 
136 static void write_ivf_file_header(FILE *outfile,
137  const vpx_codec_enc_cfg_t *cfg,
138  int frame_cnt) {
139  char header[32];
140 
141  if(cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS)
142  return;
143  header[0] = 'D';
144  header[1] = 'K';
145  header[2] = 'I';
146  header[3] = 'F';
147  mem_put_le16(header+4, 0); /* version */
148  mem_put_le16(header+6, 32); /* headersize */
149  mem_put_le32(header+8, fourcc); /* headersize */
150  mem_put_le16(header+12, cfg->g_w); /* width */
151  mem_put_le16(header+14, cfg->g_h); /* height */
152  mem_put_le32(header+16, cfg->g_timebase.den); /* rate */
153  mem_put_le32(header+20, cfg->g_timebase.num); /* scale */
154  mem_put_le32(header+24, frame_cnt); /* length */
155  mem_put_le32(header+28, 0); /* unused */
156 
157  (void) fwrite(header, 1, 32, outfile);
158 }
159 
160 static void write_ivf_frame_header(FILE *outfile,
161  const vpx_codec_cx_pkt_t *pkt)
162 {
163  char header[12];
164  vpx_codec_pts_t pts;
165 
166  if(pkt->kind != VPX_CODEC_CX_FRAME_PKT)
167  return;
168 
169  pts = pkt->data.frame.pts;
170  mem_put_le32(header, pkt->data.frame.sz);
171  mem_put_le32(header+4, pts&0xFFFFFFFF);
172  mem_put_le32(header+8, pts >> 32);
173 
174  (void) fwrite(header, 1, 12, outfile);
175 }
176 
177 /* Temporal scaling parameters */
178 /* This sets all the temporal layer parameters given |num_temporal_layers|,
179  * including the target bit allocation across temporal layers. Bit allocation
180  * parameters will be passed in as user parameters in another version.
181  */
182 static void set_temporal_layer_pattern(int num_temporal_layers,
183  vpx_codec_enc_cfg_t *cfg,
184  int bitrate,
185  int *layer_flags)
186 {
187  assert(num_temporal_layers <= MAX_NUM_TEMPORAL_LAYERS);
188  switch (num_temporal_layers)
189  {
190  case 1:
191  {
192  /* 1-layer */
193  cfg->ts_number_layers = 1;
194  cfg->ts_periodicity = 1;
195  cfg->ts_rate_decimator[0] = 1;
196  cfg->ts_layer_id[0] = 0;
197  cfg->ts_target_bitrate[0] = bitrate;
198 
199  // Update L only.
200  layer_flags[0] = VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
201  break;
202  }
203 
204  case 2:
205  {
206  /* 2-layers, with sync point at first frame of layer 1. */
207  cfg->ts_number_layers = 2;
208  cfg->ts_periodicity = 2;
209  cfg->ts_rate_decimator[0] = 2;
210  cfg->ts_rate_decimator[1] = 1;
211  cfg->ts_layer_id[0] = 0;
212  cfg->ts_layer_id[1] = 1;
213  // Use 60/40 bit allocation as example.
214  cfg->ts_target_bitrate[0] = 0.6f * bitrate;
215  cfg->ts_target_bitrate[1] = bitrate;
216 
217  /* 0=L, 1=GF */
218  // ARF is used as predictor for all frames, and is only updated on
219  // key frame. Sync point every 8 frames.
220 
221  // Layer 0: predict from L and ARF, update L and G.
222  layer_flags[0] = VP8_EFLAG_NO_REF_GF |
224 
225  // Layer 1: sync point: predict from L and ARF, and update G.
226  layer_flags[1] = VP8_EFLAG_NO_REF_GF |
229 
230  // Layer 0, predict from L and ARF, update L.
231  layer_flags[2] = VP8_EFLAG_NO_REF_GF |
234 
235  // Layer 1: predict from L, G and ARF, and update G.
236  layer_flags[3] = VP8_EFLAG_NO_UPD_ARF |
239 
240  // Layer 0
241  layer_flags[4] = layer_flags[2];
242 
243  // Layer 1
244  layer_flags[5] = layer_flags[3];
245 
246  // Layer 0
247  layer_flags[6] = layer_flags[4];
248 
249  // Layer 1
250  layer_flags[7] = layer_flags[5];
251  break;
252  }
253 
254  case 3:
255  default:
256  {
257  // 3-layers structure where ARF is used as predictor for all frames,
258  // and is only updated on key frame.
259  // Sync points for layer 1 and 2 every 8 frames.
260  cfg->ts_number_layers = 3;
261  cfg->ts_periodicity = 4;
262  cfg->ts_rate_decimator[0] = 4;
263  cfg->ts_rate_decimator[1] = 2;
264  cfg->ts_rate_decimator[2] = 1;
265  cfg->ts_layer_id[0] = 0;
266  cfg->ts_layer_id[1] = 2;
267  cfg->ts_layer_id[2] = 1;
268  cfg->ts_layer_id[3] = 2;
269  // Use 40/20/40 bit allocation as example.
270  cfg->ts_target_bitrate[0] = 0.4f * bitrate;
271  cfg->ts_target_bitrate[1] = 0.6f * bitrate;
272  cfg->ts_target_bitrate[2] = bitrate;
273 
274  /* 0=L, 1=GF, 2=ARF */
275 
276  // Layer 0: predict from L and ARF; update L and G.
277  layer_flags[0] = VP8_EFLAG_NO_UPD_ARF |
279 
280  // Layer 2: sync point: predict from L and ARF; update none.
281  layer_flags[1] = VP8_EFLAG_NO_REF_GF |
286 
287  // Layer 1: sync point: predict from L and ARF; update G.
288  layer_flags[2] = VP8_EFLAG_NO_REF_GF |
291 
292  // Layer 2: predict from L, G, ARF; update none.
293  layer_flags[3] = VP8_EFLAG_NO_UPD_GF |
297 
298  // Layer 0: predict from L and ARF; update L.
299  layer_flags[4] = VP8_EFLAG_NO_UPD_GF |
302 
303  // Layer 2: predict from L, G, ARF; update none.
304  layer_flags[5] = layer_flags[3];
305 
306  // Layer 1: predict from L, G, ARF; update G.
307  layer_flags[6] = VP8_EFLAG_NO_UPD_ARF |
309 
310  // Layer 2: predict from L, G, ARF; update none.
311  layer_flags[7] = layer_flags[3];
312  break;
313  }
314  }
315 }
316 
317 /* The periodicity of the pattern given the number of temporal layers. */
318 static int periodicity_to_num_layers[MAX_NUM_TEMPORAL_LAYERS] = {1, 8, 8};
319 
320 int main(int argc, char **argv)
321 {
322  FILE *infile, *outfile[NUM_ENCODERS];
323  FILE *downsampled_input[NUM_ENCODERS - 1];
324  char filename[50];
325  vpx_codec_ctx_t codec[NUM_ENCODERS];
326  vpx_codec_enc_cfg_t cfg[NUM_ENCODERS];
327  int frame_cnt = 0;
328  vpx_image_t raw[NUM_ENCODERS];
329  vpx_codec_err_t res[NUM_ENCODERS];
330 
331  int i;
332  long width;
333  long height;
334  int length_frame;
335  int frame_avail;
336  int got_data;
337  int flags = 0;
338  int layer_id = 0;
339 
340  int layer_flags[VPX_TS_MAX_PERIODICITY * NUM_ENCODERS]
341  = {0};
342  int flag_periodicity;
343 
344  /*Currently, only realtime mode is supported in multi-resolution encoding.*/
345  int arg_deadline = VPX_DL_REALTIME;
346 
347  /* Set show_psnr to 1/0 to show/not show PSNR. Choose show_psnr=0 if you
348  don't need to know PSNR, which will skip PSNR calculation and save
349  encoding time. */
350  int show_psnr = 0;
351  int key_frame_insert = 0;
352  uint64_t psnr_sse_total[NUM_ENCODERS] = {0};
353  uint64_t psnr_samples_total[NUM_ENCODERS] = {0};
354  double psnr_totals[NUM_ENCODERS][4] = {{0,0}};
355  int psnr_count[NUM_ENCODERS] = {0};
356 
357  double cx_time = 0;
358  struct timeval tv1, tv2, difftv;
359 
360  /* Set the required target bitrates for each resolution level.
361  * If target bitrate for highest-resolution level is set to 0,
362  * (i.e. target_bitrate[0]=0), we skip encoding at that level.
363  */
364  unsigned int target_bitrate[NUM_ENCODERS]={1000, 500, 100};
365 
366  /* Enter the frame rate of the input video */
367  int framerate = 30;
368 
369  /* Set down-sampling factor for each resolution level.
370  dsf[0] controls down sampling from level 0 to level 1;
371  dsf[1] controls down sampling from level 1 to level 2;
372  dsf[2] is not used. */
373  vpx_rational_t dsf[NUM_ENCODERS] = {{2, 1}, {2, 1}, {1, 1}};
374 
375  /* Set the number of temporal layers for each encoder/resolution level,
376  * starting from highest resoln down to lowest resoln. */
377  unsigned int num_temporal_layers[NUM_ENCODERS] = {3, 3, 3};
378 
379  if(argc!= (7 + 3 * NUM_ENCODERS))
380  die("Usage: %s <width> <height> <frame_rate> <infile> <outfile(s)> "
381  "<rate_encoder(s)> <temporal_layer(s)> <key_frame_insert> <output psnr?> \n",
382  argv[0]);
383 
384  printf("Using %s\n",vpx_codec_iface_name(interface));
385 
386  width = strtol(argv[1], NULL, 0);
387  height = strtol(argv[2], NULL, 0);
388  framerate = strtol(argv[3], NULL, 0);
389 
390  if(width < 16 || width%2 || height <16 || height%2)
391  die("Invalid resolution: %ldx%ld", width, height);
392 
393  /* Open input video file for encoding */
394  if(!(infile = fopen(argv[4], "rb")))
395  die("Failed to open %s for reading", argv[4]);
396 
397  /* Open output file for each encoder to output bitstreams */
398  for (i=0; i< NUM_ENCODERS; i++)
399  {
400  if(!target_bitrate[i])
401  {
402  outfile[i] = NULL;
403  continue;
404  }
405 
406  if(!(outfile[i] = fopen(argv[i+5], "wb")))
407  die("Failed to open %s for writing", argv[i+4]);
408  }
409 
410  // Bitrates per spatial layer: overwrite default rates above.
411  for (i=0; i< NUM_ENCODERS; i++)
412  {
413  target_bitrate[i] = strtol(argv[NUM_ENCODERS + 5 + i], NULL, 0);
414  }
415 
416  // Temporal layers per spatial layers: overwrite default settings above.
417  for (i=0; i< NUM_ENCODERS; i++)
418  {
419  num_temporal_layers[i] = strtol(argv[2 * NUM_ENCODERS + 5 + i], NULL, 0);
420  if (num_temporal_layers[i] < 1 || num_temporal_layers[i] > 3)
421  die("Invalid temporal layers: %d, Must be 1, 2, or 3. \n",
422  num_temporal_layers);
423  }
424 
425  /* Open file to write out each spatially downsampled input stream. */
426  for (i=0; i< NUM_ENCODERS - 1; i++)
427  {
428  // Highest resoln is encoder 0.
429  if (sprintf(filename,"ds%d.yuv",NUM_ENCODERS - i) < 0)
430  {
431  return EXIT_FAILURE;
432  }
433  downsampled_input[i] = fopen(filename,"wb");
434  }
435 
436  key_frame_insert = strtol(argv[3 * NUM_ENCODERS + 5], NULL, 0);
437 
438  show_psnr = strtol(argv[3 * NUM_ENCODERS + 6], NULL, 0);
439 
440 
441  /* Populate default encoder configuration */
442  for (i=0; i< NUM_ENCODERS; i++)
443  {
444  res[i] = vpx_codec_enc_config_default(interface, &cfg[i], 0);
445  if(res[i]) {
446  printf("Failed to get config: %s\n", vpx_codec_err_to_string(res[i]));
447  return EXIT_FAILURE;
448  }
449  }
450 
451  /*
452  * Update the default configuration according to needs of the application.
453  */
454  /* Highest-resolution encoder settings */
455  cfg[0].g_w = width;
456  cfg[0].g_h = height;
457  cfg[0].rc_dropframe_thresh = 0;
458  cfg[0].rc_end_usage = VPX_CBR;
459  cfg[0].rc_resize_allowed = 0;
460  cfg[0].rc_min_quantizer = 2;
461  cfg[0].rc_max_quantizer = 56;
462  cfg[0].rc_undershoot_pct = 100;
463  cfg[0].rc_overshoot_pct = 15;
464  cfg[0].rc_buf_initial_sz = 500;
465  cfg[0].rc_buf_optimal_sz = 600;
466  cfg[0].rc_buf_sz = 1000;
467  cfg[0].g_error_resilient = 1; /* Enable error resilient mode */
468  cfg[0].g_lag_in_frames = 0;
469 
470  /* Disable automatic keyframe placement */
471  /* Note: These 3 settings are copied to all levels. But, except the lowest
472  * resolution level, all other levels are set to VPX_KF_DISABLED internally.
473  */
474  cfg[0].kf_mode = VPX_KF_AUTO;
475  cfg[0].kf_min_dist = 3000;
476  cfg[0].kf_max_dist = 3000;
477 
478  cfg[0].rc_target_bitrate = target_bitrate[0]; /* Set target bitrate */
479  cfg[0].g_timebase.num = 1; /* Set fps */
480  cfg[0].g_timebase.den = framerate;
481 
482  /* Other-resolution encoder settings */
483  for (i=1; i< NUM_ENCODERS; i++)
484  {
485  memcpy(&cfg[i], &cfg[0], sizeof(vpx_codec_enc_cfg_t));
486 
487  cfg[i].rc_target_bitrate = target_bitrate[i];
488 
489  /* Note: Width & height of other-resolution encoders are calculated
490  * from the highest-resolution encoder's size and the corresponding
491  * down_sampling_factor.
492  */
493  {
494  unsigned int iw = cfg[i-1].g_w*dsf[i-1].den + dsf[i-1].num - 1;
495  unsigned int ih = cfg[i-1].g_h*dsf[i-1].den + dsf[i-1].num - 1;
496  cfg[i].g_w = iw/dsf[i-1].num;
497  cfg[i].g_h = ih/dsf[i-1].num;
498  }
499 
500  /* Make width & height to be multiplier of 2. */
501  // Should support odd size ???
502  if((cfg[i].g_w)%2)cfg[i].g_w++;
503  if((cfg[i].g_h)%2)cfg[i].g_h++;
504  }
505 
506 
507  // Set the number of threads per encode/spatial layer.
508  // (1, 1, 1) means no encoder threading.
509  cfg[0].g_threads = 2;
510  cfg[1].g_threads = 1;
511  cfg[2].g_threads = 1;
512 
513  /* Allocate image for each encoder */
514  for (i=0; i< NUM_ENCODERS; i++)
515  if(!vpx_img_alloc(&raw[i], VPX_IMG_FMT_I420, cfg[i].g_w, cfg[i].g_h, 32))
516  die("Failed to allocate image", cfg[i].g_w, cfg[i].g_h);
517 
518  if (raw[0].stride[VPX_PLANE_Y] == raw[0].d_w)
519  read_frame_p = read_frame;
520  else
521  read_frame_p = read_frame_by_row;
522 
523  for (i=0; i< NUM_ENCODERS; i++)
524  if(outfile[i])
525  write_ivf_file_header(outfile[i], &cfg[i], 0);
526 
527  /* Temporal layers settings */
528  for ( i=0; i<NUM_ENCODERS; i++)
529  {
530  set_temporal_layer_pattern(num_temporal_layers[i],
531  &cfg[i],
532  cfg[i].rc_target_bitrate,
533  &layer_flags[i * VPX_TS_MAX_PERIODICITY]);
534  }
535 
536  /* Initialize multi-encoder */
537  if(vpx_codec_enc_init_multi(&codec[0], interface, &cfg[0], NUM_ENCODERS,
538  (show_psnr ? VPX_CODEC_USE_PSNR : 0), &dsf[0]))
539  die_codec(&codec[0], "Failed to initialize encoder");
540 
541  /* The extra encoding configuration parameters can be set as follows. */
542  /* Set encoding speed */
543  for ( i=0; i<NUM_ENCODERS; i++)
544  {
545  int speed = -6;
546  /* Lower speed for the lowest resolution. */
547  if (i == NUM_ENCODERS - 1) speed = -4;
548  if(vpx_codec_control(&codec[i], VP8E_SET_CPUUSED, speed))
549  die_codec(&codec[i], "Failed to set cpu_used");
550  }
551 
552  /* Set static threshold = 1 for all encoders */
553  for ( i=0; i<NUM_ENCODERS; i++)
554  {
556  die_codec(&codec[i], "Failed to set static threshold");
557  }
558 
559  /* Set NOISE_SENSITIVITY to do TEMPORAL_DENOISING */
560  /* Enable denoising for the highest-resolution encoder. */
562  die_codec(&codec[0], "Failed to set noise_sensitivity");
563  for ( i=1; i< NUM_ENCODERS; i++)
564  {
566  die_codec(&codec[i], "Failed to set noise_sensitivity");
567  }
568 
569  /* Set the number of token partitions */
570  for ( i=0; i<NUM_ENCODERS; i++)
571  {
573  die_codec(&codec[i], "Failed to set static threshold");
574  }
575 
576  /* Set the max intra target bitrate */
577  for ( i=0; i<NUM_ENCODERS; i++)
578  {
579  unsigned int max_intra_size_pct =
580  (int)(((double)cfg[0].rc_buf_optimal_sz * 0.5) * framerate / 10);
582  max_intra_size_pct))
583  die_codec(&codec[i], "Failed to set static threshold");
584  //printf("%d %d \n",i,max_intra_size_pct);
585  }
586 
587  frame_avail = 1;
588  got_data = 0;
589 
590  while(frame_avail || got_data)
591  {
592  vpx_codec_iter_t iter[NUM_ENCODERS]={NULL};
593  const vpx_codec_cx_pkt_t *pkt[NUM_ENCODERS];
594 
595  flags = 0;
596  frame_avail = read_frame_p(infile, &raw[0]);
597 
598  if(frame_avail)
599  {
600  for ( i=1; i<NUM_ENCODERS; i++)
601  {
602  /*Scale the image down a number of times by downsampling factor*/
603  /* FilterMode 1 or 2 give better psnr than FilterMode 0. */
604  I420Scale(raw[i-1].planes[VPX_PLANE_Y], raw[i-1].stride[VPX_PLANE_Y],
605  raw[i-1].planes[VPX_PLANE_U], raw[i-1].stride[VPX_PLANE_U],
606  raw[i-1].planes[VPX_PLANE_V], raw[i-1].stride[VPX_PLANE_V],
607  raw[i-1].d_w, raw[i-1].d_h,
608  raw[i].planes[VPX_PLANE_Y], raw[i].stride[VPX_PLANE_Y],
609  raw[i].planes[VPX_PLANE_U], raw[i].stride[VPX_PLANE_U],
610  raw[i].planes[VPX_PLANE_V], raw[i].stride[VPX_PLANE_V],
611  raw[i].d_w, raw[i].d_h, 1);
612  /* Write out down-sampled input. */
613  length_frame = cfg[i].g_w * cfg[i].g_h *3/2;
614  if (fwrite(raw[i].planes[0], 1, length_frame,
615  downsampled_input[NUM_ENCODERS - i - 1]) !=
616  length_frame)
617  {
618  return EXIT_FAILURE;
619  }
620  }
621  }
622 
623  /* Set the flags (reference and update) for all the encoders.*/
624  for ( i=0; i<NUM_ENCODERS; i++)
625  {
626  layer_id = cfg[i].ts_layer_id[frame_cnt % cfg[i].ts_periodicity];
627  flags = 0;
628  flag_periodicity = periodicity_to_num_layers
629  [num_temporal_layers[i] - 1];
630  flags = layer_flags[i * VPX_TS_MAX_PERIODICITY +
631  frame_cnt % flag_periodicity];
632  // Key frame flag for first frame.
633  if (frame_cnt == 0)
634  {
635  flags |= VPX_EFLAG_FORCE_KF;
636  }
637  if (frame_cnt > 0 && frame_cnt == key_frame_insert)
638  {
639  flags = VPX_EFLAG_FORCE_KF;
640  }
641 
642  vpx_codec_control(&codec[i], VP8E_SET_FRAME_FLAGS, flags);
643  vpx_codec_control(&codec[i], VP8E_SET_TEMPORAL_LAYER_ID, layer_id);
644  }
645 
646  gettimeofday(&tv1, NULL);
647  /* Encode each frame at multi-levels */
648  /* Note the flags must be set to 0 in the encode call if they are set
649  for each frame with the vpx_codec_control(), as done above. */
650  if(vpx_codec_encode(&codec[0], frame_avail? &raw[0] : NULL,
651  frame_cnt, 1, 0, arg_deadline))
652  {
653  die_codec(&codec[0], "Failed to encode frame");
654  }
655  gettimeofday(&tv2, NULL);
656  timersub(&tv2, &tv1, &difftv);
657  cx_time += (double)(difftv.tv_sec * 1000000 + difftv.tv_usec);
658  for (i=NUM_ENCODERS-1; i>=0 ; i--)
659  {
660  got_data = 0;
661  while( (pkt[i] = vpx_codec_get_cx_data(&codec[i], &iter[i])) )
662  {
663  got_data = 1;
664  switch(pkt[i]->kind) {
666  write_ivf_frame_header(outfile[i], pkt[i]);
667  (void) fwrite(pkt[i]->data.frame.buf, 1,
668  pkt[i]->data.frame.sz, outfile[i]);
669  break;
670  case VPX_CODEC_PSNR_PKT:
671  if (show_psnr)
672  {
673  int j;
674 
675  psnr_sse_total[i] += pkt[i]->data.psnr.sse[0];
676  psnr_samples_total[i] += pkt[i]->data.psnr.samples[0];
677  for (j = 0; j < 4; j++)
678  {
679  psnr_totals[i][j] += pkt[i]->data.psnr.psnr[j];
680  }
681  psnr_count[i]++;
682  }
683 
684  break;
685  default:
686  break;
687  }
688  printf(pkt[i]->kind == VPX_CODEC_CX_FRAME_PKT
689  && (pkt[i]->data.frame.flags & VPX_FRAME_IS_KEY)? "K":"");
690  fflush(stdout);
691  }
692  }
693  frame_cnt++;
694  }
695  printf("\n");
696  printf("FPS for encoding %d %f %f \n", frame_cnt, (float)cx_time / 1000000,
697  1000000 * (double)frame_cnt / (double)cx_time);
698 
699  fclose(infile);
700 
701  printf("Processed %ld frames.\n",(long int)frame_cnt-1);
702  for (i=0; i< NUM_ENCODERS; i++)
703  {
704  /* Calculate PSNR and print it out */
705  if ( (show_psnr) && (psnr_count[i]>0) )
706  {
707  int j;
708  double ovpsnr = sse_to_psnr(psnr_samples_total[i], 255.0,
709  psnr_sse_total[i]);
710 
711  fprintf(stderr, "\n ENC%d PSNR (Overall/Avg/Y/U/V)", i);
712 
713  fprintf(stderr, " %.3lf", ovpsnr);
714  for (j = 0; j < 4; j++)
715  {
716  fprintf(stderr, " %.3lf", psnr_totals[i][j]/psnr_count[i]);
717  }
718  }
719 
720  if(vpx_codec_destroy(&codec[i]))
721  die_codec(&codec[i], "Failed to destroy codec");
722 
723  vpx_img_free(&raw[i]);
724 
725  if(!outfile[i])
726  continue;
727 
728  /* Try to rewrite the file header with the actual frame count */
729  if(!fseek(outfile[i], 0, SEEK_SET))
730  write_ivf_file_header(outfile[i], &cfg[i], frame_cnt-1);
731  fclose(outfile[i]);
732  }
733  printf("\n");
734 
735  return EXIT_SUCCESS;
736 }
Rational Number.
Definition: vpx_encoder.h:259
unsigned int rc_buf_initial_sz
Decoder Buffer Initial Size.
Definition: vpx_encoder.h:610
unsigned int ts_number_layers
Number of temporal coding layers.
Definition: vpx_encoder.h:715
Codec control function to set encoder internal speed settings.
Definition: vp8cx.h:173
#define VP8_EFLAG_NO_UPD_GF
Don&#39;t update the golden frame.
Definition: vp8cx.h:101
Image Descriptor.
Definition: vpx_image.h:88
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:397
Definition: vpx_encoder.h:276
unsigned int rc_buf_sz
Decoder Buffer Size.
Definition: vpx_encoder.h:600
#define VP8_EFLAG_NO_REF_GF
Don&#39;t reference the golden frame.
Definition: vp8cx.h:76
Codec control function to set reference and update frame flags.
Definition: vp8cx.h:275
enum vpx_kf_mode kf_mode
Keyframe placement mode.
Definition: vpx_encoder.h:665
int den
Definition: vpx_encoder.h:261
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:552
unsigned int rc_min_quantizer
Minimum (Best Quality) Quantizer.
Definition: vpx_encoder.h:541
unsigned int kf_max_dist
Keyframe maximum interval.
Definition: vpx_encoder.h:685
unsigned int g_lag_in_frames
Allow lagged encoding.
Definition: vpx_encoder.h:429
Encoder configuration structure.
Definition: vpx_encoder.h:314
Definition: vpx_encoder.h:179
Definition: vpx_encoder.h:292
Codec control function to set Max data rate for Intra frames.
Definition: vp8cx.h:269
Encoder output packet.
Definition: vpx_encoder.h:195
unsigned int rc_overshoot_pct
Rate control adaptation overshoot control.
Definition: vpx_encoder.h:583
unsigned int ts_rate_decimator[5]
Frame rate decimation factor for each temporal layer.
Definition: vpx_encoder.h:729
unsigned int rc_buf_optimal_sz
Decoder Buffer Optimal Size.
Definition: vpx_encoder.h:620
#define VPX_PLANE_V
Definition: vpx_image.h:114
unsigned int kf_min_dist
Keyframe minimum interval.
Definition: vpx_encoder.h:675
Definition: vpx_encoder.h:269
unsigned int ts_layer_id[16]
Template defining the membership of frames to temporal layers.
Definition: vpx_encoder.h:747
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:99
unsigned int g_w
Width of the frame.
Definition: vpx_encoder.h:357
unsigned int ts_target_bitrate[5]
Target bitrate for each temporal layer.
Definition: vpx_encoder.h:722
unsigned int rc_undershoot_pct
Rate control adaptation undershoot control.
Definition: vpx_encoder.h:570
unsigned int g_h
Height of the frame.
Definition: vpx_encoder.h:367
int stride[4]
Definition: vpx_image.h:117
enum vpx_codec_cx_pkt_kind kind
Definition: vpx_encoder.h:196
unsigned int rc_dropframe_thresh
Temporal resampling configuration, if supported by the codec.
Definition: vpx_encoder.h:452
Codec control function to set the temporal layer id.
Definition: vp8cx.h:316
#define VP8_EFLAG_NO_UPD_LAST
Don&#39;t update the last frame.
Definition: vp8cx.h:93
void vpx_img_free(vpx_image_t *img)
Close an image descriptor.
vpx_img_fmt_t fmt
Definition: vpx_image.h:89
unsigned char * planes[4]
Definition: vpx_image.h:116
Codec control function to set the number of token partitions.
Definition: vp8cx.h:206
unsigned int rc_target_bitrate
Target data rate.
Definition: vpx_encoder.h:525
#define VPX_DL_REALTIME
Definition: vpx_encoder.h:911
int num
Definition: vpx_encoder.h:260
control function to set noise sensitivity
Definition: vp8cx.h:188
enum vpx_enc_pass g_pass
Multi-pass Encoding Mode.
Definition: vpx_encoder.h:414
double psnr[4]
Definition: vpx_encoder.h:219
unsigned int g_threads
Maximum number of threads to use.
Definition: vpx_encoder.h:335
Provides definitions for using VP8 or VP9 encoder algorithm within the vpx Codec Interface.
#define VPX_PLANE_U
Definition: vpx_image.h:113
unsigned int rc_resize_allowed
Enable/disable spatial resampling, if supported by the codec.
Definition: vpx_encoder.h:462
unsigned int h
Definition: vpx_image.h:95
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:850
int64_t vpx_codec_pts_t
Time Stamp Type.
Definition: vpx_encoder.h:119
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:738
vpx_codec_err_t vpx_codec_destroy(vpx_codec_ctx_t *ctx)
Destroy a codec instance.
unsigned int d_h
Definition: vpx_image.h:100
unsigned int w
Definition: vpx_image.h:94
Codec control function to set the threshold for MBs treated static.
Definition: vp8cx.h:200
#define VPX_FRAME_IS_KEY
Definition: vpx_encoder.h:130
#define VPX_EFLAG_FORCE_KF
Definition: vpx_encoder.h:305
const void * vpx_codec_iter_t
Iterator.
Definition: vpx_codec.h:188
Definition: vpx_encoder.h:176
vpx_codec_er_flags_t g_error_resilient
Enable error resilient modes.
Definition: vpx_encoder.h:406
#define VP8_EFLAG_NO_UPD_ARF
Don&#39;t update the alternate reference frame.
Definition: vp8cx.h:109
#define VP8_EFLAG_NO_UPD_ENTROPY
Disable entropy update.
Definition: vp8cx.h:133
enum vpx_rc_mode rc_end_usage
Rate control algorithm to use.
Definition: vpx_encoder.h:504
Definition: vpx_encoder.h:267
Codec context structure.
Definition: vpx_codec.h:199