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 "vpx/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 #define OUTPUT_RC_STATS 1
34
35 static const arg_def_t skip_frames_arg =
36 ARG_DEF("s", "skip-frames", 1, "input frames to skip");
37 static const arg_def_t frames_arg =
38 ARG_DEF("f", "frames", 1, "number of frames to encode");
39 static const arg_def_t threads_arg =
40 ARG_DEF("th", "threads", 1, "number of threads to use");
41 #if OUTPUT_RC_STATS
42 static const arg_def_t output_rc_stats_arg =
43 ARG_DEF("rcstat", "output_rc_stats", 1, "output rc stats");
44 #endif
45 static const arg_def_t width_arg = ARG_DEF("w", "width", 1, "source width");
46 static const arg_def_t height_arg = ARG_DEF("h", "height", 1, "source height");
47 static const arg_def_t timebase_arg =
48 ARG_DEF("t", "timebase", 1, "timebase (num/den)");
49 static const arg_def_t bitrate_arg = ARG_DEF(
50 "b", "target-bitrate", 1, "encoding bitrate, in kilobits per second");
51 static const arg_def_t spatial_layers_arg =
52 ARG_DEF("sl", "spatial-layers", 1, "number of spatial SVC layers");
53 static const arg_def_t temporal_layers_arg =
54 ARG_DEF("tl", "temporal-layers", 1, "number of temporal SVC layers");
55 static const arg_def_t temporal_layering_mode_arg =
56 ARG_DEF("tlm", "temporal-layering-mode", 1,
57 "temporal layering scheme."
58 "VP9E_TEMPORAL_LAYERING_MODE");
59 static const arg_def_t kf_dist_arg =
60 ARG_DEF("k", "kf-dist", 1, "number of frames between keyframes");
61 static const arg_def_t scale_factors_arg =
62 ARG_DEF("r", "scale-factors", 1, "scale factors (lowest to highest layer)");
63 static const arg_def_t passes_arg =
64 ARG_DEF("p", "passes", 1, "Number of passes (1/2)");
65 static const arg_def_t pass_arg =
66 ARG_DEF(NULL, "pass", 1, "Pass to execute (1/2)");
67 static const arg_def_t fpf_name_arg =
68 ARG_DEF(NULL, "fpf", 1, "First pass statistics file name");
69 static const arg_def_t min_q_arg =
70 ARG_DEF(NULL, "min-q", 1, "Minimum quantizer");
71 static const arg_def_t max_q_arg =
72 ARG_DEF(NULL, "max-q", 1, "Maximum quantizer");
73 static const arg_def_t min_bitrate_arg =
74 ARG_DEF(NULL, "min-bitrate", 1, "Minimum bitrate");
75 static const arg_def_t max_bitrate_arg =
76 ARG_DEF(NULL, "max-bitrate", 1, "Maximum bitrate");
77 static const arg_def_t lag_in_frame_arg =
78 ARG_DEF(NULL, "lag-in-frames", 1,
79 "Number of frame to input before "
80 "generating any outputs");
81 static const arg_def_t rc_end_usage_arg =
82 ARG_DEF(NULL, "rc-end-usage", 1, "0 - 3: VBR, CBR, CQ, Q");
83 static const arg_def_t speed_arg =
84 ARG_DEF("sp", "speed", 1, "speed configuration");
85 static const arg_def_t aqmode_arg =
86 ARG_DEF("aq", "aqmode", 1, "aq-mode off/on");
87 static const arg_def_t bitrates_arg =
88 ARG_DEF("bl", "bitrates", 1, "bitrates[sl * num_tl + tl]");
89
90 #if CONFIG_VP9_HIGHBITDEPTH
91 static const struct arg_enum_list bitdepth_enum[] = {
92 { "8", VPX_BITS_8 }, { "10", VPX_BITS_10 }, { "12", VPX_BITS_12 }, { NULL, 0 }
93 };
94
95 static const arg_def_t bitdepth_arg = ARG_DEF_ENUM(
96 "d", "bit-depth", 1, "Bit depth for codec 8, 10 or 12. ", bitdepth_enum);
97 #endif // CONFIG_VP9_HIGHBITDEPTH
98
99 static const arg_def_t *svc_args[] = { &frames_arg,
100 &width_arg,
101 &height_arg,
102 &timebase_arg,
103 &bitrate_arg,
104 &skip_frames_arg,
105 &spatial_layers_arg,
106 &kf_dist_arg,
107 &scale_factors_arg,
108 &passes_arg,
109 &pass_arg,
110 &fpf_name_arg,
111 &min_q_arg,
112 &max_q_arg,
113 &min_bitrate_arg,
114 &max_bitrate_arg,
115 &temporal_layers_arg,
116 &temporal_layering_mode_arg,
117 &lag_in_frame_arg,
118 &threads_arg,
119 &aqmode_arg,
120 #if OUTPUT_RC_STATS
121 &output_rc_stats_arg,
122 #endif
123
124 #if CONFIG_VP9_HIGHBITDEPTH
125 &bitdepth_arg,
126 #endif
127 &speed_arg,
128 &rc_end_usage_arg,
129 &bitrates_arg,
130 NULL };
131
132 static const uint32_t default_frames_to_skip = 0;
133 static const uint32_t default_frames_to_code = 60 * 60;
134 static const uint32_t default_width = 1920;
135 static const uint32_t default_height = 1080;
136 static const uint32_t default_timebase_num = 1;
137 static const uint32_t default_timebase_den = 60;
138 static const uint32_t default_bitrate = 1000;
139 static const uint32_t default_spatial_layers = 5;
140 static const uint32_t default_temporal_layers = 1;
141 static const uint32_t default_kf_dist = 100;
142 static const uint32_t default_temporal_layering_mode = 0;
143 static const uint32_t default_output_rc_stats = 0;
144 static const int32_t default_speed = -1; // -1 means use library default.
145 static const uint32_t default_threads = 0; // zero means use library default.
146
147 typedef struct {
148 const char *input_filename;
149 const char *output_filename;
150 uint32_t frames_to_code;
151 uint32_t frames_to_skip;
152 struct VpxInputContext input_ctx;
153 stats_io_t rc_stats;
154 int passes;
155 int pass;
156 } AppInput;
157
158 static const char *exec_name;
159
usage_exit(void)160 void usage_exit(void) {
161 fprintf(stderr, "Usage: %s <options> input_filename output_filename\n",
162 exec_name);
163 fprintf(stderr, "Options:\n");
164 arg_show_usage(stderr, svc_args);
165 exit(EXIT_FAILURE);
166 }
167
parse_command_line(int argc,const char ** argv_,AppInput * app_input,SvcContext * svc_ctx,vpx_codec_enc_cfg_t * enc_cfg)168 static void parse_command_line(int argc, const char **argv_,
169 AppInput *app_input, SvcContext *svc_ctx,
170 vpx_codec_enc_cfg_t *enc_cfg) {
171 struct arg arg = { 0 };
172 char **argv = NULL;
173 char **argi = NULL;
174 char **argj = NULL;
175 vpx_codec_err_t res;
176 int passes = 0;
177 int pass = 0;
178 const char *fpf_file_name = NULL;
179 unsigned int min_bitrate = 0;
180 unsigned int max_bitrate = 0;
181 char string_options[1024] = { 0 };
182
183 // initialize SvcContext with parameters that will be passed to vpx_svc_init
184 svc_ctx->log_level = SVC_LOG_DEBUG;
185 svc_ctx->spatial_layers = default_spatial_layers;
186 svc_ctx->temporal_layers = default_temporal_layers;
187 svc_ctx->temporal_layering_mode = default_temporal_layering_mode;
188 #if OUTPUT_RC_STATS
189 svc_ctx->output_rc_stat = default_output_rc_stats;
190 #endif
191 svc_ctx->speed = default_speed;
192 svc_ctx->threads = default_threads;
193
194 // start with default encoder configuration
195 res = vpx_codec_enc_config_default(vpx_codec_vp9_cx(), enc_cfg, 0);
196 if (res) {
197 die("Failed to get config: %s\n", vpx_codec_err_to_string(res));
198 }
199 // update enc_cfg with app default values
200 enc_cfg->g_w = default_width;
201 enc_cfg->g_h = default_height;
202 enc_cfg->g_timebase.num = default_timebase_num;
203 enc_cfg->g_timebase.den = default_timebase_den;
204 enc_cfg->rc_target_bitrate = default_bitrate;
205 enc_cfg->kf_min_dist = default_kf_dist;
206 enc_cfg->kf_max_dist = default_kf_dist;
207 enc_cfg->rc_end_usage = VPX_CQ;
208
209 // initialize AppInput with default values
210 app_input->frames_to_code = default_frames_to_code;
211 app_input->frames_to_skip = default_frames_to_skip;
212
213 // process command line options
214 argv = argv_dup(argc - 1, argv_ + 1);
215 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) {
216 arg.argv_step = 1;
217
218 if (arg_match(&arg, &frames_arg, argi)) {
219 app_input->frames_to_code = arg_parse_uint(&arg);
220 } else if (arg_match(&arg, &width_arg, argi)) {
221 enc_cfg->g_w = arg_parse_uint(&arg);
222 } else if (arg_match(&arg, &height_arg, argi)) {
223 enc_cfg->g_h = arg_parse_uint(&arg);
224 } else if (arg_match(&arg, &timebase_arg, argi)) {
225 enc_cfg->g_timebase = arg_parse_rational(&arg);
226 } else if (arg_match(&arg, &bitrate_arg, argi)) {
227 enc_cfg->rc_target_bitrate = arg_parse_uint(&arg);
228 } else if (arg_match(&arg, &skip_frames_arg, argi)) {
229 app_input->frames_to_skip = arg_parse_uint(&arg);
230 } else if (arg_match(&arg, &spatial_layers_arg, argi)) {
231 svc_ctx->spatial_layers = arg_parse_uint(&arg);
232 } else if (arg_match(&arg, &temporal_layers_arg, argi)) {
233 svc_ctx->temporal_layers = arg_parse_uint(&arg);
234 #if OUTPUT_RC_STATS
235 } else if (arg_match(&arg, &output_rc_stats_arg, argi)) {
236 svc_ctx->output_rc_stat = arg_parse_uint(&arg);
237 #endif
238 } else if (arg_match(&arg, &speed_arg, argi)) {
239 svc_ctx->speed = arg_parse_uint(&arg);
240 } else if (arg_match(&arg, &aqmode_arg, argi)) {
241 svc_ctx->aqmode = arg_parse_uint(&arg);
242 } else if (arg_match(&arg, &threads_arg, argi)) {
243 svc_ctx->threads = arg_parse_uint(&arg);
244 } else if (arg_match(&arg, &temporal_layering_mode_arg, argi)) {
245 svc_ctx->temporal_layering_mode = enc_cfg->temporal_layering_mode =
246 arg_parse_int(&arg);
247 if (svc_ctx->temporal_layering_mode) {
248 enc_cfg->g_error_resilient = 1;
249 }
250 } else if (arg_match(&arg, &kf_dist_arg, argi)) {
251 enc_cfg->kf_min_dist = arg_parse_uint(&arg);
252 enc_cfg->kf_max_dist = enc_cfg->kf_min_dist;
253 } else if (arg_match(&arg, &scale_factors_arg, argi)) {
254 snprintf(string_options, sizeof(string_options), "%s scale-factors=%s",
255 string_options, arg.val);
256 } else if (arg_match(&arg, &bitrates_arg, argi)) {
257 snprintf(string_options, sizeof(string_options), "%s bitrates=%s",
258 string_options, arg.val);
259 } else if (arg_match(&arg, &passes_arg, argi)) {
260 passes = arg_parse_uint(&arg);
261 if (passes < 1 || passes > 2) {
262 die("Error: Invalid number of passes (%d)\n", passes);
263 }
264 } else if (arg_match(&arg, &pass_arg, argi)) {
265 pass = arg_parse_uint(&arg);
266 if (pass < 1 || pass > 2) {
267 die("Error: Invalid pass selected (%d)\n", pass);
268 }
269 } else if (arg_match(&arg, &fpf_name_arg, argi)) {
270 fpf_file_name = arg.val;
271 } else if (arg_match(&arg, &min_q_arg, argi)) {
272 snprintf(string_options, sizeof(string_options), "%s min-quantizers=%s",
273 string_options, arg.val);
274 } else if (arg_match(&arg, &max_q_arg, argi)) {
275 snprintf(string_options, sizeof(string_options), "%s max-quantizers=%s",
276 string_options, arg.val);
277 } else if (arg_match(&arg, &min_bitrate_arg, argi)) {
278 min_bitrate = arg_parse_uint(&arg);
279 } else if (arg_match(&arg, &max_bitrate_arg, argi)) {
280 max_bitrate = arg_parse_uint(&arg);
281 } else if (arg_match(&arg, &lag_in_frame_arg, argi)) {
282 enc_cfg->g_lag_in_frames = arg_parse_uint(&arg);
283 } else if (arg_match(&arg, &rc_end_usage_arg, argi)) {
284 enc_cfg->rc_end_usage = arg_parse_uint(&arg);
285 #if CONFIG_VP9_HIGHBITDEPTH
286 } else if (arg_match(&arg, &bitdepth_arg, argi)) {
287 enc_cfg->g_bit_depth = arg_parse_enum_or_int(&arg);
288 switch (enc_cfg->g_bit_depth) {
289 case VPX_BITS_8:
290 enc_cfg->g_input_bit_depth = 8;
291 enc_cfg->g_profile = 0;
292 break;
293 case VPX_BITS_10:
294 enc_cfg->g_input_bit_depth = 10;
295 enc_cfg->g_profile = 2;
296 break;
297 case VPX_BITS_12:
298 enc_cfg->g_input_bit_depth = 12;
299 enc_cfg->g_profile = 2;
300 break;
301 default:
302 die("Error: Invalid bit depth selected (%d)\n", enc_cfg->g_bit_depth);
303 break;
304 }
305 #endif // CONFIG_VP9_HIGHBITDEPTH
306 } else {
307 ++argj;
308 }
309 }
310
311 // There will be a space in front of the string options
312 if (strlen(string_options) > 0)
313 vpx_svc_set_options(svc_ctx, string_options + 1);
314
315 if (passes == 0 || passes == 1) {
316 if (pass) {
317 fprintf(stderr, "pass is ignored since there's only one pass\n");
318 }
319 enc_cfg->g_pass = VPX_RC_ONE_PASS;
320 } else {
321 if (pass == 0) {
322 die("pass must be specified when passes is 2\n");
323 }
324
325 if (fpf_file_name == NULL) {
326 die("fpf must be specified when passes is 2\n");
327 }
328
329 if (pass == 1) {
330 enc_cfg->g_pass = VPX_RC_FIRST_PASS;
331 if (!stats_open_file(&app_input->rc_stats, fpf_file_name, 0)) {
332 fatal("Failed to open statistics store");
333 }
334 } else {
335 enc_cfg->g_pass = VPX_RC_LAST_PASS;
336 if (!stats_open_file(&app_input->rc_stats, fpf_file_name, 1)) {
337 fatal("Failed to open statistics store");
338 }
339 enc_cfg->rc_twopass_stats_in = stats_get(&app_input->rc_stats);
340 }
341 app_input->passes = passes;
342 app_input->pass = pass;
343 }
344
345 if (enc_cfg->rc_target_bitrate > 0) {
346 if (min_bitrate > 0) {
347 enc_cfg->rc_2pass_vbr_minsection_pct =
348 min_bitrate * 100 / enc_cfg->rc_target_bitrate;
349 }
350 if (max_bitrate > 0) {
351 enc_cfg->rc_2pass_vbr_maxsection_pct =
352 max_bitrate * 100 / enc_cfg->rc_target_bitrate;
353 }
354 }
355
356 // Check for unrecognized options
357 for (argi = argv; *argi; ++argi)
358 if (argi[0][0] == '-' && strlen(argi[0]) > 1)
359 die("Error: Unrecognized option %s\n", *argi);
360
361 if (argv[0] == NULL || argv[1] == 0) {
362 usage_exit();
363 }
364 app_input->input_filename = argv[0];
365 app_input->output_filename = argv[1];
366 free(argv);
367
368 if (enc_cfg->g_w < 16 || enc_cfg->g_w % 2 || enc_cfg->g_h < 16 ||
369 enc_cfg->g_h % 2)
370 die("Invalid resolution: %d x %d\n", enc_cfg->g_w, enc_cfg->g_h);
371
372 printf(
373 "Codec %s\nframes: %d, skip: %d\n"
374 "layers: %d\n"
375 "width %d, height: %d,\n"
376 "num: %d, den: %d, bitrate: %d,\n"
377 "gop size: %d\n",
378 vpx_codec_iface_name(vpx_codec_vp9_cx()), app_input->frames_to_code,
379 app_input->frames_to_skip, svc_ctx->spatial_layers, enc_cfg->g_w,
380 enc_cfg->g_h, enc_cfg->g_timebase.num, enc_cfg->g_timebase.den,
381 enc_cfg->rc_target_bitrate, enc_cfg->kf_max_dist);
382 }
383
384 #if OUTPUT_RC_STATS
385 // For rate control encoding stats.
386 struct RateControlStats {
387 // Number of input frames per layer.
388 int layer_input_frames[VPX_MAX_LAYERS];
389 // Total (cumulative) number of encoded frames per layer.
390 int layer_tot_enc_frames[VPX_MAX_LAYERS];
391 // Number of encoded non-key frames per layer.
392 int layer_enc_frames[VPX_MAX_LAYERS];
393 // Framerate per layer (cumulative).
394 double layer_framerate[VPX_MAX_LAYERS];
395 // Target average frame size per layer (per-frame-bandwidth per layer).
396 double layer_pfb[VPX_MAX_LAYERS];
397 // Actual average frame size per layer.
398 double layer_avg_frame_size[VPX_MAX_LAYERS];
399 // Average rate mismatch per layer (|target - actual| / target).
400 double layer_avg_rate_mismatch[VPX_MAX_LAYERS];
401 // Actual encoding bitrate per layer (cumulative).
402 double layer_encoding_bitrate[VPX_MAX_LAYERS];
403 // Average of the short-time encoder actual bitrate.
404 // TODO(marpan): Should we add these short-time stats for each layer?
405 double avg_st_encoding_bitrate;
406 // Variance of the short-time encoder actual bitrate.
407 double variance_st_encoding_bitrate;
408 // Window (number of frames) for computing short-time encoding bitrate.
409 int window_size;
410 // Number of window measurements.
411 int window_count;
412 };
413
414 // Note: these rate control stats assume only 1 key frame in the
415 // sequence (i.e., first frame only).
set_rate_control_stats(struct RateControlStats * rc,vpx_codec_enc_cfg_t * cfg)416 static void set_rate_control_stats(struct RateControlStats *rc,
417 vpx_codec_enc_cfg_t *cfg) {
418 unsigned int sl, tl;
419 // Set the layer (cumulative) framerate and the target layer (non-cumulative)
420 // per-frame-bandwidth, for the rate control encoding stats below.
421 const double framerate = cfg->g_timebase.den / cfg->g_timebase.num;
422
423 for (sl = 0; sl < cfg->ss_number_layers; ++sl) {
424 for (tl = 0; tl < cfg->ts_number_layers; ++tl) {
425 const int layer = sl * cfg->ts_number_layers + tl;
426 if (cfg->ts_number_layers == 1)
427 rc->layer_framerate[layer] = framerate;
428 else
429 rc->layer_framerate[layer] = framerate / cfg->ts_rate_decimator[tl];
430 if (tl > 0) {
431 rc->layer_pfb[layer] =
432 1000.0 * (cfg->layer_target_bitrate[layer] -
433 cfg->layer_target_bitrate[layer - 1]) /
434 (rc->layer_framerate[layer] - rc->layer_framerate[layer - 1]);
435 } else {
436 rc->layer_pfb[layer] = 1000.0 * cfg->layer_target_bitrate[layer] /
437 rc->layer_framerate[layer];
438 }
439 rc->layer_input_frames[layer] = 0;
440 rc->layer_enc_frames[layer] = 0;
441 rc->layer_tot_enc_frames[layer] = 0;
442 rc->layer_encoding_bitrate[layer] = 0.0;
443 rc->layer_avg_frame_size[layer] = 0.0;
444 rc->layer_avg_rate_mismatch[layer] = 0.0;
445 }
446 }
447 rc->window_count = 0;
448 rc->window_size = 15;
449 rc->avg_st_encoding_bitrate = 0.0;
450 rc->variance_st_encoding_bitrate = 0.0;
451 }
452
printout_rate_control_summary(struct RateControlStats * rc,vpx_codec_enc_cfg_t * cfg,int frame_cnt)453 static void printout_rate_control_summary(struct RateControlStats *rc,
454 vpx_codec_enc_cfg_t *cfg,
455 int frame_cnt) {
456 unsigned int sl, tl;
457 double perc_fluctuation = 0.0;
458 int tot_num_frames = 0;
459 printf("Total number of processed frames: %d\n\n", frame_cnt - 1);
460 printf("Rate control layer stats for sl%d tl%d layer(s):\n\n",
461 cfg->ss_number_layers, cfg->ts_number_layers);
462 for (sl = 0; sl < cfg->ss_number_layers; ++sl) {
463 tot_num_frames = 0;
464 for (tl = 0; tl < cfg->ts_number_layers; ++tl) {
465 const int layer = sl * cfg->ts_number_layers + tl;
466 const int num_dropped =
467 (tl > 0)
468 ? (rc->layer_input_frames[layer] - rc->layer_enc_frames[layer])
469 : (rc->layer_input_frames[layer] - rc->layer_enc_frames[layer] -
470 1);
471 tot_num_frames += rc->layer_input_frames[layer];
472 rc->layer_encoding_bitrate[layer] = 0.001 * rc->layer_framerate[layer] *
473 rc->layer_encoding_bitrate[layer] /
474 tot_num_frames;
475 rc->layer_avg_frame_size[layer] =
476 rc->layer_avg_frame_size[layer] / rc->layer_enc_frames[layer];
477 rc->layer_avg_rate_mismatch[layer] = 100.0 *
478 rc->layer_avg_rate_mismatch[layer] /
479 rc->layer_enc_frames[layer];
480 printf("For layer#: sl%d tl%d \n", sl, tl);
481 printf("Bitrate (target vs actual): %d %f.0 kbps\n",
482 cfg->layer_target_bitrate[layer],
483 rc->layer_encoding_bitrate[layer]);
484 printf("Average frame size (target vs actual): %f %f bits\n",
485 rc->layer_pfb[layer], rc->layer_avg_frame_size[layer]);
486 printf("Average rate_mismatch: %f\n", rc->layer_avg_rate_mismatch[layer]);
487 printf(
488 "Number of input frames, encoded (non-key) frames, "
489 "and percent dropped frames: %d %d %f.0 \n",
490 rc->layer_input_frames[layer], rc->layer_enc_frames[layer],
491 100.0 * num_dropped / rc->layer_input_frames[layer]);
492 printf("\n");
493 }
494 }
495 rc->avg_st_encoding_bitrate = rc->avg_st_encoding_bitrate / rc->window_count;
496 rc->variance_st_encoding_bitrate =
497 rc->variance_st_encoding_bitrate / rc->window_count -
498 (rc->avg_st_encoding_bitrate * rc->avg_st_encoding_bitrate);
499 perc_fluctuation = 100.0 * sqrt(rc->variance_st_encoding_bitrate) /
500 rc->avg_st_encoding_bitrate;
501 printf("Short-time stats, for window of %d frames: \n", rc->window_size);
502 printf("Average, rms-variance, and percent-fluct: %f %f %f \n",
503 rc->avg_st_encoding_bitrate, sqrt(rc->variance_st_encoding_bitrate),
504 perc_fluctuation);
505 if (frame_cnt != tot_num_frames)
506 die("Error: Number of input frames not equal to output encoded frames != "
507 "%d tot_num_frames = %d\n",
508 frame_cnt, tot_num_frames);
509 }
510
parse_superframe_index(const uint8_t * data,size_t data_sz,uint32_t sizes[8],int * count)511 vpx_codec_err_t parse_superframe_index(const uint8_t *data, size_t data_sz,
512 uint32_t sizes[8], int *count) {
513 // A chunk ending with a byte matching 0xc0 is an invalid chunk unless
514 // it is a super frame index. If the last byte of real video compression
515 // data is 0xc0 the encoder must add a 0 byte. If we have the marker but
516 // not the associated matching marker byte at the front of the index we have
517 // an invalid bitstream and need to return an error.
518
519 uint8_t marker;
520
521 marker = *(data + data_sz - 1);
522 *count = 0;
523
524 if ((marker & 0xe0) == 0xc0) {
525 const uint32_t frames = (marker & 0x7) + 1;
526 const uint32_t mag = ((marker >> 3) & 0x3) + 1;
527 const size_t index_sz = 2 + mag * frames;
528
529 // This chunk is marked as having a superframe index but doesn't have
530 // enough data for it, thus it's an invalid superframe index.
531 if (data_sz < index_sz) return VPX_CODEC_CORRUPT_FRAME;
532
533 {
534 const uint8_t marker2 = *(data + data_sz - index_sz);
535
536 // This chunk is marked as having a superframe index but doesn't have
537 // the matching marker byte at the front of the index therefore it's an
538 // invalid chunk.
539 if (marker != marker2) return VPX_CODEC_CORRUPT_FRAME;
540 }
541
542 {
543 // Found a valid superframe index.
544 uint32_t i, j;
545 const uint8_t *x = &data[data_sz - index_sz + 1];
546
547 for (i = 0; i < frames; ++i) {
548 uint32_t this_sz = 0;
549
550 for (j = 0; j < mag; ++j) this_sz |= (*x++) << (j * 8);
551 sizes[i] = this_sz;
552 }
553 *count = frames;
554 }
555 }
556 return VPX_CODEC_OK;
557 }
558 #endif
559
560 // Example pattern for spatial layers and 2 temporal layers used in the
561 // bypass/flexible mode. The pattern corresponds to the pattern
562 // VP9E_TEMPORAL_LAYERING_MODE_0101 (temporal_layering_mode == 2) used in
563 // non-flexible mode.
set_frame_flags_bypass_mode(int sl,int tl,int num_spatial_layers,int is_key_frame,vpx_svc_ref_frame_config_t * ref_frame_config)564 void set_frame_flags_bypass_mode(int sl, int tl, int num_spatial_layers,
565 int is_key_frame,
566 vpx_svc_ref_frame_config_t *ref_frame_config) {
567 for (sl = 0; sl < num_spatial_layers; ++sl) {
568 if (!tl) {
569 if (!sl) {
570 ref_frame_config->frame_flags[sl] =
571 VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_GF |
572 VP8_EFLAG_NO_UPD_ARF;
573 } else {
574 if (is_key_frame) {
575 ref_frame_config->frame_flags[sl] =
576 VP8_EFLAG_NO_REF_LAST | VP8_EFLAG_NO_REF_ARF |
577 VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
578 } else {
579 ref_frame_config->frame_flags[sl] =
580 VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
581 }
582 }
583 } else if (tl == 1) {
584 if (!sl) {
585 ref_frame_config->frame_flags[sl] =
586 VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_LAST |
587 VP8_EFLAG_NO_UPD_GF;
588 } else {
589 ref_frame_config->frame_flags[sl] =
590 VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_GF;
591 }
592 }
593 if (tl == 0) {
594 ref_frame_config->lst_fb_idx[sl] = sl;
595 if (sl)
596 ref_frame_config->gld_fb_idx[sl] = sl - 1;
597 else
598 ref_frame_config->gld_fb_idx[sl] = 0;
599 ref_frame_config->alt_fb_idx[sl] = 0;
600 } else if (tl == 1) {
601 ref_frame_config->lst_fb_idx[sl] = sl;
602 ref_frame_config->gld_fb_idx[sl] = num_spatial_layers + sl - 1;
603 ref_frame_config->alt_fb_idx[sl] = num_spatial_layers + sl;
604 }
605 }
606 }
607
main(int argc,const char ** argv)608 int main(int argc, const char **argv) {
609 AppInput app_input = { 0 };
610 VpxVideoWriter *writer = NULL;
611 VpxVideoInfo info = { 0 };
612 vpx_codec_ctx_t codec;
613 vpx_codec_enc_cfg_t enc_cfg;
614 SvcContext svc_ctx;
615 uint32_t i;
616 uint32_t frame_cnt = 0;
617 vpx_image_t raw;
618 vpx_codec_err_t res;
619 int pts = 0; /* PTS starts at 0 */
620 int frame_duration = 1; /* 1 timebase tick per frame */
621 FILE *infile = NULL;
622 int end_of_stream = 0;
623 int frames_received = 0;
624 #if OUTPUT_RC_STATS
625 VpxVideoWriter *outfile[VPX_TS_MAX_LAYERS] = { NULL };
626 struct RateControlStats rc;
627 vpx_svc_layer_id_t layer_id;
628 vpx_svc_ref_frame_config_t ref_frame_config;
629 unsigned int sl, tl;
630 double sum_bitrate = 0.0;
631 double sum_bitrate2 = 0.0;
632 double framerate = 30.0;
633 #endif
634 struct vpx_usec_timer timer;
635 int64_t cx_time = 0;
636 memset(&svc_ctx, 0, sizeof(svc_ctx));
637 svc_ctx.log_print = 1;
638 exec_name = argv[0];
639 parse_command_line(argc, argv, &app_input, &svc_ctx, &enc_cfg);
640
641 // Allocate image buffer
642 #if CONFIG_VP9_HIGHBITDEPTH
643 if (!vpx_img_alloc(&raw, enc_cfg.g_input_bit_depth == 8 ? VPX_IMG_FMT_I420
644 : VPX_IMG_FMT_I42016,
645 enc_cfg.g_w, enc_cfg.g_h, 32)) {
646 die("Failed to allocate image %dx%d\n", enc_cfg.g_w, enc_cfg.g_h);
647 }
648 #else
649 if (!vpx_img_alloc(&raw, VPX_IMG_FMT_I420, enc_cfg.g_w, enc_cfg.g_h, 32)) {
650 die("Failed to allocate image %dx%d\n", enc_cfg.g_w, enc_cfg.g_h);
651 }
652 #endif // CONFIG_VP9_HIGHBITDEPTH
653
654 if (!(infile = fopen(app_input.input_filename, "rb")))
655 die("Failed to open %s for reading\n", app_input.input_filename);
656
657 // Initialize codec
658 if (vpx_svc_init(&svc_ctx, &codec, vpx_codec_vp9_cx(), &enc_cfg) !=
659 VPX_CODEC_OK)
660 die("Failed to initialize encoder\n");
661
662 #if OUTPUT_RC_STATS
663 if (svc_ctx.output_rc_stat) {
664 set_rate_control_stats(&rc, &enc_cfg);
665 framerate = enc_cfg.g_timebase.den / enc_cfg.g_timebase.num;
666 }
667 #endif
668
669 info.codec_fourcc = VP9_FOURCC;
670 info.time_base.numerator = enc_cfg.g_timebase.num;
671 info.time_base.denominator = enc_cfg.g_timebase.den;
672
673 if (!(app_input.passes == 2 && app_input.pass == 1)) {
674 // We don't save the bitstream for the 1st pass on two pass rate control
675 writer =
676 vpx_video_writer_open(app_input.output_filename, kContainerIVF, &info);
677 if (!writer)
678 die("Failed to open %s for writing\n", app_input.output_filename);
679 }
680 #if OUTPUT_RC_STATS
681 // For now, just write temporal layer streams.
682 // TODO(marpan): do spatial by re-writing superframe.
683 if (svc_ctx.output_rc_stat) {
684 for (tl = 0; tl < enc_cfg.ts_number_layers; ++tl) {
685 char file_name[PATH_MAX];
686
687 snprintf(file_name, sizeof(file_name), "%s_t%d.ivf",
688 app_input.output_filename, tl);
689 outfile[tl] = vpx_video_writer_open(file_name, kContainerIVF, &info);
690 if (!outfile[tl]) die("Failed to open %s for writing", file_name);
691 }
692 }
693 #endif
694
695 // skip initial frames
696 for (i = 0; i < app_input.frames_to_skip; ++i) vpx_img_read(&raw, infile);
697
698 if (svc_ctx.speed != -1)
699 vpx_codec_control(&codec, VP8E_SET_CPUUSED, svc_ctx.speed);
700 if (svc_ctx.threads) {
701 vpx_codec_control(&codec, VP9E_SET_TILE_COLUMNS, (svc_ctx.threads >> 1));
702 vpx_codec_control(&codec, VP9E_SET_ROW_MT, 0);
703 }
704 if (svc_ctx.speed >= 5 && svc_ctx.aqmode == 1)
705 vpx_codec_control(&codec, VP9E_SET_AQ_MODE, 3);
706 if (svc_ctx.speed >= 5)
707 vpx_codec_control(&codec, VP8E_SET_STATIC_THRESHOLD, 1);
708
709 // Encode frames
710 while (!end_of_stream) {
711 vpx_codec_iter_t iter = NULL;
712 const vpx_codec_cx_pkt_t *cx_pkt;
713 if (frame_cnt >= app_input.frames_to_code || !vpx_img_read(&raw, infile)) {
714 // We need one extra vpx_svc_encode call at end of stream to flush
715 // encoder and get remaining data
716 end_of_stream = 1;
717 }
718
719 // For BYPASS/FLEXIBLE mode, set the frame flags (reference and updates)
720 // and the buffer indices for each spatial layer of the current
721 // (super)frame to be encoded. The temporal layer_id for the current frame
722 // also needs to be set.
723 // TODO(marpan): Should rename the "VP9E_TEMPORAL_LAYERING_MODE_BYPASS"
724 // mode to "VP9E_LAYERING_MODE_BYPASS".
725 if (svc_ctx.temporal_layering_mode == VP9E_TEMPORAL_LAYERING_MODE_BYPASS) {
726 layer_id.spatial_layer_id = 0;
727 // Example for 2 temporal layers.
728 if (frame_cnt % 2 == 0)
729 layer_id.temporal_layer_id = 0;
730 else
731 layer_id.temporal_layer_id = 1;
732 // Note that we only set the temporal layer_id, since we are calling
733 // the encode for the whole superframe. The encoder will internally loop
734 // over all the spatial layers for the current superframe.
735 vpx_codec_control(&codec, VP9E_SET_SVC_LAYER_ID, &layer_id);
736 set_frame_flags_bypass_mode(sl, layer_id.temporal_layer_id,
737 svc_ctx.spatial_layers, frame_cnt == 0,
738 &ref_frame_config);
739 vpx_codec_control(&codec, VP9E_SET_SVC_REF_FRAME_CONFIG,
740 &ref_frame_config);
741 // Keep track of input frames, to account for frame drops in rate control
742 // stats/metrics.
743 for (sl = 0; sl < (unsigned int)enc_cfg.ss_number_layers; ++sl) {
744 ++rc.layer_input_frames[sl * enc_cfg.ts_number_layers +
745 layer_id.temporal_layer_id];
746 }
747 }
748
749 vpx_usec_timer_start(&timer);
750 res = vpx_svc_encode(
751 &svc_ctx, &codec, (end_of_stream ? NULL : &raw), pts, frame_duration,
752 svc_ctx.speed >= 5 ? VPX_DL_REALTIME : VPX_DL_GOOD_QUALITY);
753 vpx_usec_timer_mark(&timer);
754 cx_time += vpx_usec_timer_elapsed(&timer);
755
756 printf("%s", vpx_svc_get_message(&svc_ctx));
757 fflush(stdout);
758 if (res != VPX_CODEC_OK) {
759 die_codec(&codec, "Failed to encode frame");
760 }
761
762 while ((cx_pkt = vpx_codec_get_cx_data(&codec, &iter)) != NULL) {
763 switch (cx_pkt->kind) {
764 case VPX_CODEC_CX_FRAME_PKT: {
765 SvcInternal_t *const si = (SvcInternal_t *)svc_ctx.internal;
766 if (cx_pkt->data.frame.sz > 0) {
767 #if OUTPUT_RC_STATS
768 uint32_t sizes[8];
769 int count = 0;
770 #endif
771 vpx_video_writer_write_frame(writer, cx_pkt->data.frame.buf,
772 cx_pkt->data.frame.sz,
773 cx_pkt->data.frame.pts);
774 #if OUTPUT_RC_STATS
775 // TODO(marpan): Put this (to line728) in separate function.
776 if (svc_ctx.output_rc_stat) {
777 vpx_codec_control(&codec, VP9E_GET_SVC_LAYER_ID, &layer_id);
778 parse_superframe_index(cx_pkt->data.frame.buf,
779 cx_pkt->data.frame.sz, sizes, &count);
780 // Note computing input_layer_frames here won't account for frame
781 // drops in rate control stats.
782 // TODO(marpan): Fix this for non-bypass mode so we can get stats
783 // for dropped frames.
784 if (svc_ctx.temporal_layering_mode !=
785 VP9E_TEMPORAL_LAYERING_MODE_BYPASS) {
786 for (sl = 0; sl < enc_cfg.ss_number_layers; ++sl) {
787 ++rc.layer_input_frames[sl * enc_cfg.ts_number_layers +
788 layer_id.temporal_layer_id];
789 }
790 }
791 for (tl = layer_id.temporal_layer_id;
792 tl < enc_cfg.ts_number_layers; ++tl) {
793 vpx_video_writer_write_frame(
794 outfile[tl], cx_pkt->data.frame.buf, cx_pkt->data.frame.sz,
795 cx_pkt->data.frame.pts);
796 }
797
798 for (sl = 0; sl < enc_cfg.ss_number_layers; ++sl) {
799 for (tl = layer_id.temporal_layer_id;
800 tl < enc_cfg.ts_number_layers; ++tl) {
801 const int layer = sl * enc_cfg.ts_number_layers + tl;
802 ++rc.layer_tot_enc_frames[layer];
803 rc.layer_encoding_bitrate[layer] += 8.0 * sizes[sl];
804 // Keep count of rate control stats per layer, for non-key
805 // frames.
806 if (tl == (unsigned int)layer_id.temporal_layer_id &&
807 !(cx_pkt->data.frame.flags & VPX_FRAME_IS_KEY)) {
808 rc.layer_avg_frame_size[layer] += 8.0 * sizes[sl];
809 rc.layer_avg_rate_mismatch[layer] +=
810 fabs(8.0 * sizes[sl] - rc.layer_pfb[layer]) /
811 rc.layer_pfb[layer];
812 ++rc.layer_enc_frames[layer];
813 }
814 }
815 }
816
817 // Update for short-time encoding bitrate states, for moving
818 // window of size rc->window, shifted by rc->window / 2.
819 // Ignore first window segment, due to key frame.
820 if (frame_cnt > (unsigned int)rc.window_size) {
821 tl = layer_id.temporal_layer_id;
822 for (sl = 0; sl < enc_cfg.ss_number_layers; ++sl) {
823 sum_bitrate += 0.001 * 8.0 * sizes[sl] * framerate;
824 }
825 if (frame_cnt % rc.window_size == 0) {
826 rc.window_count += 1;
827 rc.avg_st_encoding_bitrate += sum_bitrate / rc.window_size;
828 rc.variance_st_encoding_bitrate +=
829 (sum_bitrate / rc.window_size) *
830 (sum_bitrate / rc.window_size);
831 sum_bitrate = 0.0;
832 }
833 }
834
835 // Second shifted window.
836 if (frame_cnt >
837 (unsigned int)(rc.window_size + rc.window_size / 2)) {
838 tl = layer_id.temporal_layer_id;
839 for (sl = 0; sl < enc_cfg.ss_number_layers; ++sl) {
840 sum_bitrate2 += 0.001 * 8.0 * sizes[sl] * framerate;
841 }
842
843 if (frame_cnt > (unsigned int)(2 * rc.window_size) &&
844 frame_cnt % rc.window_size == 0) {
845 rc.window_count += 1;
846 rc.avg_st_encoding_bitrate += sum_bitrate2 / rc.window_size;
847 rc.variance_st_encoding_bitrate +=
848 (sum_bitrate2 / rc.window_size) *
849 (sum_bitrate2 / rc.window_size);
850 sum_bitrate2 = 0.0;
851 }
852 }
853 }
854 #endif
855 }
856 /*
857 printf("SVC frame: %d, kf: %d, size: %d, pts: %d\n", frames_received,
858 !!(cx_pkt->data.frame.flags & VPX_FRAME_IS_KEY),
859 (int)cx_pkt->data.frame.sz, (int)cx_pkt->data.frame.pts);
860 */
861 if (enc_cfg.ss_number_layers == 1 && enc_cfg.ts_number_layers == 1)
862 si->bytes_sum[0] += (int)cx_pkt->data.frame.sz;
863 ++frames_received;
864 break;
865 }
866 case VPX_CODEC_STATS_PKT: {
867 stats_write(&app_input.rc_stats, cx_pkt->data.twopass_stats.buf,
868 cx_pkt->data.twopass_stats.sz);
869 break;
870 }
871 default: { break; }
872 }
873 }
874
875 if (!end_of_stream) {
876 ++frame_cnt;
877 pts += frame_duration;
878 }
879 }
880
881 // Compensate for the extra frame count for the bypass mode.
882 if (svc_ctx.temporal_layering_mode == VP9E_TEMPORAL_LAYERING_MODE_BYPASS) {
883 for (sl = 0; sl < enc_cfg.ss_number_layers; ++sl) {
884 const int layer =
885 sl * enc_cfg.ts_number_layers + layer_id.temporal_layer_id;
886 --rc.layer_input_frames[layer];
887 }
888 }
889
890 printf("Processed %d frames\n", frame_cnt);
891 fclose(infile);
892 #if OUTPUT_RC_STATS
893 if (svc_ctx.output_rc_stat) {
894 printout_rate_control_summary(&rc, &enc_cfg, frame_cnt);
895 printf("\n");
896 }
897 #endif
898 if (vpx_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec");
899 if (app_input.passes == 2) stats_close(&app_input.rc_stats, 1);
900 if (writer) {
901 vpx_video_writer_close(writer);
902 }
903 #if OUTPUT_RC_STATS
904 if (svc_ctx.output_rc_stat) {
905 for (tl = 0; tl < enc_cfg.ts_number_layers; ++tl) {
906 vpx_video_writer_close(outfile[tl]);
907 }
908 }
909 #endif
910 printf("Frame cnt and encoding time/FPS stats for encoding: %d %f %f \n",
911 frame_cnt, 1000 * (float)cx_time / (double)(frame_cnt * 1000000),
912 1000000 * (double)frame_cnt / (double)cx_time);
913 vpx_img_free(&raw);
914 // display average size, psnr
915 printf("%s", vpx_svc_dump_statistics(&svc_ctx));
916 vpx_svc_release(&svc_ctx);
917 return EXIT_SUCCESS;
918 }
919