AOMedia Codec SDK
svc_encoder_rtc
1/*
2 * Copyright (c) 2019, Alliance for Open Media. 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// This is an example demonstrating how to implement a multi-layer AOM
12// encoding scheme for RTC video applications.
13
14#include <assert.h>
15#include <math.h>
16#include <stdio.h>
17#include <stdlib.h>
18#include <string.h>
19
20#include "aom/aom_encoder.h"
21#include "aom/aomcx.h"
22#include "av1/common/enums.h"
23#include "common/tools_common.h"
24#include "common/video_writer.h"
25#include "aom_ports/aom_timer.h"
26
27#define zero(Dest) memset(&(Dest), 0, sizeof(Dest));
28
29static const char *exec_name;
30
31void usage_exit(void) { exit(EXIT_FAILURE); }
32
33static int mode_to_num_layers[4] = { 1, 2, 3, 3 };
34
35// For rate control encoding stats.
36struct RateControlMetrics {
37 // Number of input frames per layer.
38 int layer_input_frames[AOM_MAX_TS_LAYERS];
39 // Total (cumulative) number of encoded frames per layer.
40 int layer_tot_enc_frames[AOM_MAX_TS_LAYERS];
41 // Number of encoded non-key frames per layer.
42 int layer_enc_frames[AOM_MAX_TS_LAYERS];
43 // Framerate per layer layer (cumulative).
44 double layer_framerate[AOM_MAX_TS_LAYERS];
45 // Target average frame size per layer (per-frame-bandwidth per layer).
46 double layer_pfb[AOM_MAX_TS_LAYERS];
47 // Actual average frame size per layer.
48 double layer_avg_frame_size[AOM_MAX_TS_LAYERS];
49 // Average rate mismatch per layer (|target - actual| / target).
50 double layer_avg_rate_mismatch[AOM_MAX_TS_LAYERS];
51 // Actual encoding bitrate per layer (cumulative).
52 double layer_encoding_bitrate[AOM_MAX_TS_LAYERS];
53 // Average of the short-time encoder actual bitrate.
54 // TODO(marpan): Should we add these short-time stats for each layer?
55 double avg_st_encoding_bitrate;
56 // Variance of the short-time encoder actual bitrate.
57 double variance_st_encoding_bitrate;
58 // Window (number of frames) for computing short-timee encoding bitrate.
59 int window_size;
60 // Number of window measurements.
61 int window_count;
62 int layer_target_bitrate[AOM_MAX_TS_LAYERS];
63};
64
65static int read_frame(struct AvxInputContext *input_ctx, aom_image_t *img) {
66 FILE *f = input_ctx->file;
67 y4m_input *y4m = &input_ctx->y4m;
68 int shortread = 0;
69
70 if (input_ctx->file_type == FILE_TYPE_Y4M) {
71 if (y4m_input_fetch_frame(y4m, f, img) < 1) return 0;
72 } else {
73 shortread = read_yuv_frame(input_ctx, img);
74 }
75
76 return !shortread;
77}
78
79static int file_is_y4m(const char detect[4]) {
80 if (memcmp(detect, "YUV4", 4) == 0) {
81 return 1;
82 }
83 return 0;
84}
85
86static int fourcc_is_ivf(const char detect[4]) {
87 if (memcmp(detect, "DKIF", 4) == 0) {
88 return 1;
89 }
90 return 0;
91}
92
93static void close_input_file(struct AvxInputContext *input) {
94 fclose(input->file);
95 if (input->file_type == FILE_TYPE_Y4M) y4m_input_close(&input->y4m);
96}
97
98static void open_input_file(struct AvxInputContext *input,
100 /* Parse certain options from the input file, if possible */
101 input->file = strcmp(input->filename, "-") ? fopen(input->filename, "rb")
102 : set_binary_mode(stdin);
103
104 if (!input->file) fatal("Failed to open input file");
105
106 if (!fseeko(input->file, 0, SEEK_END)) {
107 /* Input file is seekable. Figure out how long it is, so we can get
108 * progress info.
109 */
110 input->length = ftello(input->file);
111 rewind(input->file);
112 }
113
114 /* Default to 1:1 pixel aspect ratio. */
115 input->pixel_aspect_ratio.numerator = 1;
116 input->pixel_aspect_ratio.denominator = 1;
117
118 /* For RAW input sources, these bytes will applied on the first frame
119 * in read_frame().
120 */
121 input->detect.buf_read = fread(input->detect.buf, 1, 4, input->file);
122 input->detect.position = 0;
123
124 if (input->detect.buf_read == 4 && file_is_y4m(input->detect.buf)) {
125 if (y4m_input_open(&input->y4m, input->file, input->detect.buf, 4, csp,
126 input->only_i420) >= 0) {
127 input->file_type = FILE_TYPE_Y4M;
128 input->width = input->y4m.pic_w;
129 input->height = input->y4m.pic_h;
130 input->pixel_aspect_ratio.numerator = input->y4m.par_n;
131 input->pixel_aspect_ratio.denominator = input->y4m.par_d;
132 input->framerate.numerator = input->y4m.fps_n;
133 input->framerate.denominator = input->y4m.fps_d;
134 input->fmt = input->y4m.aom_fmt;
135 input->bit_depth = input->y4m.bit_depth;
136 } else {
137 fatal("Unsupported Y4M stream.");
138 }
139 } else if (input->detect.buf_read == 4 && fourcc_is_ivf(input->detect.buf)) {
140 fatal("IVF is not supported as input.");
141 } else {
142 input->file_type = FILE_TYPE_RAW;
143 }
144}
145
146// Note: these rate control metrics assume only 1 key frame in the
147// sequence (i.e., first frame only). So for temporal pattern# 7
148// (which has key frame for every frame on base layer), the metrics
149// computation will be off/wrong.
150// TODO(marpan): Update these metrics to account for multiple key frames
151// in the stream.
152static void set_rate_control_metrics(struct RateControlMetrics *rc,
153 double framerate,
154 unsigned int ts_number_layers) {
155 int ts_rate_decimator[AOM_MAX_TS_LAYERS] = { 1 };
156 ts_rate_decimator[0] = 1;
157 if (ts_number_layers == 2) {
158 ts_rate_decimator[0] = 2;
159 ts_rate_decimator[1] = 1;
160 }
161 if (ts_number_layers == 3) {
162 ts_rate_decimator[0] = 4;
163 ts_rate_decimator[1] = 2;
164 ts_rate_decimator[2] = 1;
165 }
166 // Set the layer (cumulative) framerate and the target layer (non-cumulative)
167 // per-frame-bandwidth, for the rate control encoding stats below.
168 rc->layer_framerate[0] = framerate / ts_rate_decimator[0];
169 rc->layer_pfb[0] =
170 1000.0 * rc->layer_target_bitrate[0] / rc->layer_framerate[0];
171 for (unsigned int i = 0; i < ts_number_layers; ++i) {
172 if (i > 0) {
173 rc->layer_framerate[i] = framerate / ts_rate_decimator[i];
174 rc->layer_pfb[i] =
175 1000.0 *
176 (rc->layer_target_bitrate[i] - rc->layer_target_bitrate[i - 1]) /
177 (rc->layer_framerate[i] - rc->layer_framerate[i - 1]);
178 }
179 rc->layer_input_frames[i] = 0;
180 rc->layer_enc_frames[i] = 0;
181 rc->layer_tot_enc_frames[i] = 0;
182 rc->layer_encoding_bitrate[i] = 0.0;
183 rc->layer_avg_frame_size[i] = 0.0;
184 rc->layer_avg_rate_mismatch[i] = 0.0;
185 }
186 rc->window_count = 0;
187 rc->window_size = 15;
188 rc->avg_st_encoding_bitrate = 0.0;
189 rc->variance_st_encoding_bitrate = 0.0;
190}
191
192static void printout_rate_control_summary(struct RateControlMetrics *rc,
193 int frame_cnt,
194 unsigned int ts_number_layers) {
195 int tot_num_frames = 0;
196 double perc_fluctuation = 0.0;
197 printf("Total number of processed frames: %d\n\n", frame_cnt - 1);
198 printf("Rate control layer stats for %d layer(s):\n\n", ts_number_layers);
199 for (unsigned int i = 0; i < ts_number_layers; ++i) {
200 const int num_dropped =
201 i > 0 ? rc->layer_input_frames[i] - rc->layer_enc_frames[i]
202 : rc->layer_input_frames[i] - rc->layer_enc_frames[i] - 1;
203 tot_num_frames += rc->layer_input_frames[i];
204 rc->layer_encoding_bitrate[i] = 0.001 * rc->layer_framerate[i] *
205 rc->layer_encoding_bitrate[i] /
206 tot_num_frames;
207 rc->layer_avg_frame_size[i] =
208 rc->layer_avg_frame_size[i] / rc->layer_enc_frames[i];
209 rc->layer_avg_rate_mismatch[i] =
210 100.0 * rc->layer_avg_rate_mismatch[i] / rc->layer_enc_frames[i];
211 printf("For layer#: %d\n", i);
212 printf("Bitrate (target vs actual): %d %f\n", rc->layer_target_bitrate[i],
213 rc->layer_encoding_bitrate[i]);
214 printf("Average frame size (target vs actual): %f %f\n", rc->layer_pfb[i],
215 rc->layer_avg_frame_size[i]);
216 printf("Average rate_mismatch: %f\n", rc->layer_avg_rate_mismatch[i]);
217 printf(
218 "Number of input frames, encoded (non-key) frames, "
219 "and perc dropped frames: %d %d %f\n",
220 rc->layer_input_frames[i], rc->layer_enc_frames[i],
221 100.0 * num_dropped / rc->layer_input_frames[i]);
222 printf("\n");
223 }
224 rc->avg_st_encoding_bitrate = rc->avg_st_encoding_bitrate / rc->window_count;
225 rc->variance_st_encoding_bitrate =
226 rc->variance_st_encoding_bitrate / rc->window_count -
227 (rc->avg_st_encoding_bitrate * rc->avg_st_encoding_bitrate);
228 perc_fluctuation = 100.0 * sqrt(rc->variance_st_encoding_bitrate) /
229 rc->avg_st_encoding_bitrate;
230 printf("Short-time stats, for window of %d frames:\n", rc->window_size);
231 printf("Average, rms-variance, and percent-fluct: %f %f %f\n",
232 rc->avg_st_encoding_bitrate, sqrt(rc->variance_st_encoding_bitrate),
233 perc_fluctuation);
234 if (frame_cnt - 1 != tot_num_frames)
235 die("Error: Number of input frames not equal to output!\n");
236}
237
238// Layer pattern configuration.
239static int set_layer_pattern(int layering_mode, int frame_cnt,
240 aom_svc_layer_id_t *layer_id,
241 aom_svc_ref_frame_config_t *ref_frame_config) {
242 int i;
243 // No spatial layers in this test.
244 layer_id->spatial_layer_id = 0;
245 // Set the referende map buffer idx for the 7 references:
246 // LAST_FRAME (0), LAST2_FRAME(1), LAST3_FRAME(2), GOLDEN_FRAME(3),
247 // BWDREF_FRAME(4), ALTREF2_FRAME(5), ALTREF_FRAME(6).
248 for (i = 0; i < INTER_REFS_PER_FRAME; i++) ref_frame_config->ref_idx[i] = i;
249 for (i = 0; i < REF_FRAMES; i++) ref_frame_config->refresh[i] = 0;
250 // Note only use LAST and GF for prediction in non-rd mode (speed 8).
254 switch (layering_mode) {
255 case 0:
256 // 1-layer: update LAST on every frame, reference LAST and GF.
257 layer_id->temporal_layer_id = 0;
258 ref_frame_config->refresh[0] = 1;
259 break;
260 case 1:
261 // 2-layer.
262 // 1 3 5
263 // 0 2 4
264 if (frame_cnt % 2 == 0) {
265 layer_id->temporal_layer_id = 0;
266 // Update LAST on layer 0, reference LAST and GF.
267 ref_frame_config->refresh[0] = 1;
268 } else {
269 layer_id->temporal_layer_id = 1;
270 // No updates on layer 1, only reference LAST (TL0).
271 layer_flags |= AOM_EFLAG_NO_REF_GF;
272 }
273 break;
274 case 2:
275 // 3-layer:
276 // 1 3 5 7
277 // 2 6
278 // 0 4 8
279 if (frame_cnt % 4 == 0) {
280 // Base layer.
281 layer_id->temporal_layer_id = 0;
282 // Update LAST on layer 0, reference LAST and GF.
283 ref_frame_config->refresh[0] = 1;
284 } else if ((frame_cnt - 1) % 4 == 0) {
285 layer_id->temporal_layer_id = 2;
286 // First top layer: no updates, only reference LAST (TL0).
287 layer_flags |= AOM_EFLAG_NO_REF_GF;
288 } else if ((frame_cnt - 2) % 4 == 0) {
289 layer_id->temporal_layer_id = 1;
290 // Middle layer (TL1): update LAST2, only reference LAST (TL0).
291 ref_frame_config->refresh[1] = 1;
292 layer_flags |= AOM_EFLAG_NO_REF_GF;
293 } else if ((frame_cnt - 3) % 4 == 0) {
294 layer_id->temporal_layer_id = 2;
295 // Second top layer: no updates, only reference LAST.
296 // Set buffer idx for LAST to slot 1, since that was the slot
297 // updated in previous frame. So LAST is TL1 frame.
298 ref_frame_config->ref_idx[0] = 1;
299 ref_frame_config->ref_idx[1] = 0;
300 layer_flags |= AOM_EFLAG_NO_REF_GF;
301 }
302 break;
303 case 3:
304 // 3-layer: but middle layer updates GF, so 2nd TL2 will only
305 // reference GF (not LAST). Other frames only reference LAST.
306 // 1 3 5 7
307 // 2 6
308 // 0 4 8
309 if (frame_cnt % 4 == 0) {
310 // Base layer.
311 layer_id->temporal_layer_id = 0;
312 // Update LAST on layer 0, only reference LAST.
313 ref_frame_config->refresh[0] = 1;
314 layer_flags |= AOM_EFLAG_NO_REF_GF;
315 } else if ((frame_cnt - 1) % 4 == 0) {
316 layer_id->temporal_layer_id = 2;
317 // First top layer: no updates, only reference LAST (TL0).
318 layer_flags |= AOM_EFLAG_NO_REF_GF;
319 } else if ((frame_cnt - 2) % 4 == 0) {
320 layer_id->temporal_layer_id = 1;
321 // Middle layer (TL1): update GF, only reference LAST (TL0).
322 ref_frame_config->refresh[3] = 1;
323 layer_flags |= AOM_EFLAG_NO_REF_GF;
324 } else if ((frame_cnt - 3) % 4 == 0) {
325 layer_id->temporal_layer_id = 2;
326 // Second top layer: no updates, only reference GF.
327 layer_flags |= AOM_EFLAG_NO_REF_LAST;
328 }
329 break;
330 default: assert(0); die("Error: Unsupported temporal layering mode!\n");
331 }
332 return layer_flags;
333}
334
335int main(int argc, char **argv) {
336 AvxVideoWriter *outfile[AOM_MAX_TS_LAYERS] = { NULL };
337 aom_codec_ctx_t codec;
339 int frame_cnt = 0;
340 aom_image_t raw;
341 aom_codec_err_t res;
342 unsigned int width;
343 unsigned int height;
344 uint32_t error_resilient = 0;
345 int speed;
346 int frame_avail;
347 int got_data;
348 int flags = 0;
349 unsigned int i;
350 int pts = 0; // PTS starts at 0.
351 int frame_duration = 1; // 1 timebase tick per frame.
352 int layering_mode = 0;
353 aom_svc_layer_id_t layer_id;
354 aom_svc_params_t svc_params;
355 aom_svc_ref_frame_config_t ref_frame_config;
356 const AvxInterface *encoder = NULL;
357 struct AvxInputContext input_ctx;
358 struct RateControlMetrics rc;
359 int64_t cx_time = 0;
360 const int min_args_base = 13;
361 const int min_args = min_args_base;
362 double sum_bitrate = 0.0;
363 double sum_bitrate2 = 0.0;
364 double framerate = 30.0;
365 zero(rc.layer_target_bitrate);
366 memset(&layer_id, 0, sizeof(aom_svc_layer_id_t));
367 memset(&input_ctx, 0, sizeof(input_ctx));
368 memset(&svc_params, 0, sizeof(svc_params));
369
370 /* Setup default input stream settings */
371 input_ctx.framerate.numerator = 30;
372 input_ctx.framerate.denominator = 1;
373 input_ctx.only_i420 = 1;
374 input_ctx.bit_depth = 0;
375 unsigned int ts_number_layers = 1;
376 unsigned int ss_number_layers = 1;
377 exec_name = argv[0];
378 // Check usage and arguments.
379 if (argc < min_args) {
380 die("Usage: %s <infile> <outfile> <codec_type(av1)> <width> <height> "
381 "<rate_num> <rate_den> <speed> <frame_drop_threshold> "
382 "<error_resilient> <threads> <mode> "
383 "<Rate_0> ... <Rate_nlayers-1>\n",
384 argv[0]);
385 }
386
387 encoder = get_aom_encoder_by_name(argv[3]);
388
389 width = (unsigned int)strtoul(argv[4], NULL, 0);
390 height = (unsigned int)strtoul(argv[5], NULL, 0);
391 if (width < 16 || width % 2 || height < 16 || height % 2) {
392 die("Invalid resolution: %d x %d", width, height);
393 }
394
395 layering_mode = (int)strtol(argv[12], NULL, 0);
396 if (layering_mode < 0 || layering_mode > 13) {
397 die("Invalid layering mode (0..12) %s", argv[12]);
398 }
399
400 if (argc != min_args + mode_to_num_layers[layering_mode]) {
401 die("Invalid number of arguments");
402 }
403
404 ts_number_layers = mode_to_num_layers[layering_mode];
405
406 input_ctx.filename = argv[1];
407 open_input_file(&input_ctx, 0);
408
409 // Y4M reader has its own allocation.
410 if (input_ctx.file_type != FILE_TYPE_Y4M) {
411 if (!aom_img_alloc(&raw, AOM_IMG_FMT_I420, width, height, 32)) {
412 die("Failed to allocate image", width, height);
413 }
414 }
415
416 // Populate encoder configuration.
417 res = aom_codec_enc_config_default(encoder->codec_interface(), &cfg, 0);
418 if (res) {
419 printf("Failed to get config: %s\n", aom_codec_err_to_string(res));
420 return EXIT_FAILURE;
421 }
422
423 // Update the default configuration with our settings.
424 cfg.g_w = width;
425 cfg.g_h = height;
426
427 // Timebase format e.g. 30fps: numerator=1, demoninator = 30.
428 cfg.g_timebase.num = (int)strtol(argv[6], NULL, 0);
429 cfg.g_timebase.den = (int)strtol(argv[7], NULL, 0);
430
431 speed = (int)strtol(argv[8], NULL, 0);
432 if (speed < 0 || speed > 8) {
433 die("Invalid speed setting: must be positive");
434 }
435
436 for (i = min_args_base;
437 (int)i < min_args_base + mode_to_num_layers[layering_mode]; ++i) {
438 rc.layer_target_bitrate[i - 13] = (int)strtol(argv[i], NULL, 0);
439 svc_params.layer_target_bitrate[i - 13] = rc.layer_target_bitrate[i - 13];
440 }
441
442 cfg.rc_target_bitrate = svc_params.layer_target_bitrate[ts_number_layers - 1];
443
444 svc_params.framerate_factor[0] = 1;
445 if (ts_number_layers == 2) {
446 svc_params.framerate_factor[0] = 2;
447 svc_params.framerate_factor[1] = 1;
448 } else if (ts_number_layers == 3) {
449 svc_params.framerate_factor[0] = 4;
450 svc_params.framerate_factor[1] = 2;
451 svc_params.framerate_factor[2] = 1;
452 }
453
454 // Real time parameters.
456
457 cfg.rc_dropframe_thresh = (unsigned int)strtoul(argv[9], NULL, 0);
458 cfg.rc_end_usage = AOM_CBR;
459 cfg.rc_min_quantizer = 2;
460 cfg.rc_max_quantizer = 52;
461 cfg.rc_undershoot_pct = 50;
462 cfg.rc_overshoot_pct = 50;
463 cfg.rc_buf_initial_sz = 600;
464 cfg.rc_buf_optimal_sz = 600;
465 cfg.rc_buf_sz = 1000;
466
467 // Use 1 thread as default.
468 cfg.g_threads = (unsigned int)strtoul(argv[11], NULL, 0);
469
470 error_resilient = (uint32_t)strtoul(argv[10], NULL, 0);
471 if (error_resilient != 0 && error_resilient != 1) {
472 die("Invalid value for error resilient (0, 1): %d.", error_resilient);
473 }
474 // Enable error resilient mode.
475 cfg.g_error_resilient = error_resilient;
476 cfg.g_lag_in_frames = 0;
477 cfg.kf_mode = AOM_KF_AUTO;
478
479 // Disable automatic keyframe placement.
480 cfg.kf_min_dist = cfg.kf_max_dist = 3000;
481
482 framerate = cfg.g_timebase.den / cfg.g_timebase.num;
483 set_rate_control_metrics(&rc, framerate, ts_number_layers);
484
485 if (input_ctx.file_type == FILE_TYPE_Y4M) {
486 if (input_ctx.width != cfg.g_w || input_ctx.height != cfg.g_h) {
487 die("Incorrect width or height: %d x %d", cfg.g_w, cfg.g_h);
488 }
489 if (input_ctx.framerate.numerator != cfg.g_timebase.den ||
490 input_ctx.framerate.denominator != cfg.g_timebase.num) {
491 die("Incorrect framerate: numerator %d denominator %d",
492 cfg.g_timebase.num, cfg.g_timebase.den);
493 }
494 }
495
496 // Open an output file for each stream.
497 for (i = 0; i < ts_number_layers; ++i) {
498 char file_name[PATH_MAX];
499 AvxVideoInfo info;
500 info.codec_fourcc = encoder->fourcc;
501 info.frame_width = cfg.g_w;
502 info.frame_height = cfg.g_h;
503 info.time_base.numerator = cfg.g_timebase.num;
504 info.time_base.denominator = cfg.g_timebase.den;
505
506 snprintf(file_name, sizeof(file_name), "%s_%d.av1", argv[2], i);
507 outfile[i] = aom_video_writer_open(file_name, kContainerIVF, &info);
508 if (!outfile[i]) die("Failed to open %s for writing", file_name);
509
510 assert(outfile[i] != NULL);
511 }
512
513 // Initialize codec.
514 if (aom_codec_enc_init(&codec, encoder->codec_interface(), &cfg, 0))
515 die_codec(&codec, "Failed to initialize encoder");
516
517 aom_codec_control(&codec, AOME_SET_CPUUSED, speed);
521
522 svc_params.number_spatial_layers = ss_number_layers;
523 svc_params.number_temporal_layers = ts_number_layers;
524 for (i = 0; i < ts_number_layers; ++i) {
525 svc_params.max_quantizers[i] = cfg.rc_max_quantizer;
526 svc_params.min_quantizers[i] = cfg.rc_min_quantizer;
527 }
528 for (i = 0; i < ss_number_layers; ++i) {
529 svc_params.scaling_factor_num[i] = 1;
530 svc_params.scaling_factor_den[i] = 1;
531 }
532 aom_codec_control(&codec, AV1E_SET_SVC_PARAMS, &svc_params);
533
534 // This controls the maximum target size of the key frame.
535 // For generating smaller key frames, use a smaller max_intra_size_pct
536 // value, like 100 or 200.
537 {
538 const int max_intra_size_pct = 300;
540 max_intra_size_pct);
541 }
542
543 frame_avail = 1;
544 while (frame_avail || got_data) {
545 struct aom_usec_timer timer;
546 aom_codec_iter_t iter = NULL;
547 const aom_codec_cx_pkt_t *pkt;
548
549 // Set the reference/update flags, layer_id, and reference_map
550 // buffer index.
551 flags = set_layer_pattern(layering_mode, frame_cnt, &layer_id,
552 &ref_frame_config);
553 aom_codec_control(&codec, AV1E_SET_SVC_LAYER_ID, &layer_id);
554 aom_codec_control(&codec, AV1E_SET_SVC_REF_FRAME_CONFIG, &ref_frame_config);
555
556 frame_avail = read_frame(&input_ctx, &raw);
557 if (frame_avail) ++rc.layer_input_frames[layer_id.temporal_layer_id];
558 aom_usec_timer_start(&timer);
559 if (aom_codec_encode(&codec, frame_avail ? &raw : NULL, pts, 1, flags)) {
560 die_codec(&codec, "Failed to encode frame");
561 }
562 aom_usec_timer_mark(&timer);
563 cx_time += aom_usec_timer_elapsed(&timer);
564 got_data = 0;
565 while ((pkt = aom_codec_get_cx_data(&codec, &iter))) {
566 got_data = 1;
567 switch (pkt->kind) {
569 for (i = layer_id.temporal_layer_id; i < ts_number_layers; ++i) {
570 aom_video_writer_write_frame(outfile[i], pkt->data.frame.buf,
571 pkt->data.frame.sz, pts);
572 ++rc.layer_tot_enc_frames[i];
573 rc.layer_encoding_bitrate[i] += 8.0 * pkt->data.frame.sz;
574 // Keep count of rate control stats per layer (for non-key frames).
575 if (i == (unsigned int)layer_id.temporal_layer_id &&
576 !(pkt->data.frame.flags & AOM_FRAME_IS_KEY)) {
577 rc.layer_avg_frame_size[i] += 8.0 * pkt->data.frame.sz;
578 rc.layer_avg_rate_mismatch[i] +=
579 fabs(8.0 * pkt->data.frame.sz - rc.layer_pfb[i]) /
580 rc.layer_pfb[i];
581 ++rc.layer_enc_frames[i];
582 }
583 }
584 // Update for short-time encoding bitrate states, for moving window
585 // of size rc->window, shifted by rc->window / 2.
586 // Ignore first window segment, due to key frame.
587 if (frame_cnt > rc.window_size) {
588 sum_bitrate += 0.001 * 8.0 * pkt->data.frame.sz * framerate;
589 rc.window_size = (rc.window_size <= 0) ? 1 : rc.window_size;
590 if (frame_cnt % rc.window_size == 0) {
591 rc.window_count += 1;
592 rc.avg_st_encoding_bitrate += sum_bitrate / rc.window_size;
593 rc.variance_st_encoding_bitrate +=
594 (sum_bitrate / rc.window_size) *
595 (sum_bitrate / rc.window_size);
596 sum_bitrate = 0.0;
597 }
598 }
599 // Second shifted window.
600 if (frame_cnt > rc.window_size + rc.window_size / 2) {
601 sum_bitrate2 += 0.001 * 8.0 * pkt->data.frame.sz * framerate;
602 if (frame_cnt > 2 * rc.window_size &&
603 frame_cnt % rc.window_size == 0) {
604 rc.window_count += 1;
605 rc.avg_st_encoding_bitrate += sum_bitrate2 / rc.window_size;
606 rc.variance_st_encoding_bitrate +=
607 (sum_bitrate2 / rc.window_size) *
608 (sum_bitrate2 / rc.window_size);
609 sum_bitrate2 = 0.0;
610 }
611 }
612 break;
613 default: break;
614 }
615 }
616 ++frame_cnt;
617 pts += frame_duration;
618 }
619 close_input_file(&input_ctx);
620 printout_rate_control_summary(&rc, frame_cnt, ts_number_layers);
621 printf("\n");
622 printf("Frame cnt and encoding time/FPS stats for encoding: %d %f %f\n",
623 frame_cnt, 1000 * (float)cx_time / (double)(frame_cnt * 1000000),
624 1000000 * (double)frame_cnt / (double)cx_time);
625
626 if (aom_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec");
627
628 // Try to rewrite the output file headers with the actual frame count.
629 for (i = 0; i < ts_number_layers; ++i) aom_video_writer_close(outfile[i]);
630
631 if (input_ctx.file_type != FILE_TYPE_Y4M) {
632 aom_img_free(&raw);
633 }
634 return EXIT_SUCCESS;
635}
Describes the encoder algorithm interface to applications.
enum aom_chroma_sample_position aom_chroma_sample_position_t
List of chroma sample positions.
aom_image_t * aom_img_alloc(aom_image_t *img, aom_img_fmt_t fmt, unsigned int d_w, unsigned int d_h, unsigned int align)
Open a descriptor, allocating storage for the underlying image.
@ AOM_IMG_FMT_I420
Definition: aom_image.h:45
void aom_img_free(aom_image_t *img)
Close an image descriptor.
Provides definitions for using AOM or AV1 encoder algorithm within the aom Codec Interface.
#define AOM_EFLAG_NO_REF_LAST2
Don't reference the last2 frame.
Definition: aomcx.h:57
#define AOM_MAX_TS_LAYERS
Definition: aomcx.h:1235
#define AOM_EFLAG_NO_REF_BWD
Don't reference the bwd reference frame.
Definition: aomcx.h:86
#define AOM_EFLAG_NO_REF_ARF
Don't reference the alternate reference frame.
Definition: aomcx.h:79
#define AOM_EFLAG_NO_REF_LAST3
Don't reference the last3 frame.
Definition: aomcx.h:64
#define AOM_EFLAG_NO_REF_GF
Don't reference the golden frame.
Definition: aomcx.h:71
#define AOM_EFLAG_NO_REF_ARF2
Don't reference the alt2 reference frame.
Definition: aomcx.h:93
#define AOM_EFLAG_NO_REF_LAST
Don't reference the last frame.
Definition: aomcx.h:50
@ AV1E_SET_AQ_MODE
Codec control function to set adaptive quantization mode.
Definition: aomcx.h:372
@ AV1E_SET_SVC_LAYER_ID
Codec control function to set the layer id.
Definition: aomcx.h:1134
@ AV1E_SET_SVC_REF_FRAME_CONFIG
Codec control function to set reference frame config: the ref_idx and the refresh flags for each buff...
Definition: aomcx.h:1143
@ AV1E_SET_ENABLE_CDEF
Codec control function to encode with CDEF.
Definition: aomcx.h:563
@ AV1E_SET_SVC_PARAMS
Codec control function to set SVC paramaeters.
Definition: aomcx.h:1138
@ AOME_SET_MAX_INTRA_BITRATE_PCT
Codec control function to set Max data rate for Intra frames.
Definition: aomcx.h:240
@ AOME_SET_CPUUSED
Codec control function to set encoder internal speed settings.
Definition: aomcx.h:182
@ AV1E_SET_GF_CBR_BOOST_PCT
Boost percentage for Golden Frame in CBR mode.
Definition: aomcx.h:270
#define aom_codec_control(ctx, id, data)
aom_codec_control wrapper macro
Definition: aom_codec.h:414
aom_codec_err_t aom_codec_destroy(aom_codec_ctx_t *ctx)
Destroy a codec instance.
const char * aom_codec_err_to_string(aom_codec_err_t err)
Convert error number to printable string.
aom_codec_err_t
Algorithm return codes.
Definition: aom_codec.h:101
const void * aom_codec_iter_t
Iterator.
Definition: aom_codec.h:194
const aom_codec_cx_pkt_t * aom_codec_get_cx_data(aom_codec_ctx_t *ctx, aom_codec_iter_t *iter)
Encoded data iterator.
aom_codec_err_t aom_codec_encode(aom_codec_ctx_t *ctx, const aom_image_t *img, aom_codec_pts_t pts, unsigned long duration, aom_enc_frame_flags_t flags)
Encode a frame.
#define aom_codec_enc_init(ctx, iface, cfg, flags)
Convenience macro for aom_codec_enc_init_ver()
Definition: aom_encoder.h:764
aom_codec_err_t aom_codec_enc_config_default(aom_codec_iface_t *iface, aom_codec_enc_cfg_t *cfg, unsigned int reserved)
Get a default configuration.
#define AOM_USAGE_REALTIME
usage parameter analogous to AV1 REALTIME mode.
Definition: aom_encoder.h:867
#define AOM_FRAME_IS_KEY
Definition: aom_encoder.h:96
@ AOM_CBR
Definition: aom_encoder.h:193
@ AOM_KF_AUTO
Definition: aom_encoder.h:208
@ AOM_CODEC_CX_FRAME_PKT
Definition: aom_encoder.h:126
Codec context structure.
Definition: aom_codec.h:204
Encoder output packet.
Definition: aom_encoder.h:138
size_t sz
Definition: aom_encoder.h:143
enum aom_codec_cx_pkt_kind kind
Definition: aom_encoder.h:139
union aom_codec_cx_pkt::@1 data
struct aom_codec_cx_pkt::@1::@2 frame
aom_codec_frame_flags_t flags
Definition: aom_encoder.h:148
void * buf
Definition: aom_encoder.h:142
Encoder configuration structure.
Definition: aom_encoder.h:228
unsigned int rc_dropframe_thresh
Temporal resampling configuration, if supported by the codec.
Definition: aom_encoder.h:376
struct aom_rational g_timebase
Stream timebase units.
Definition: aom_encoder.h:325
unsigned int g_usage
Algorithm specific "usage" value.
Definition: aom_encoder.h:240
unsigned int rc_buf_sz
Decoder Buffer Size.
Definition: aom_encoder.h:548
unsigned int g_h
Height of the frame.
Definition: aom_encoder.h:276
enum aom_kf_mode kf_mode
Keyframe placement mode.
Definition: aom_encoder.h:611
enum aom_rc_mode rc_end_usage
Rate control algorithm to use.
Definition: aom_encoder.h:461
unsigned int g_threads
Maximum number of threads to use.
Definition: aom_encoder.h:248
unsigned int kf_min_dist
Keyframe minimum interval.
Definition: aom_encoder.h:620
unsigned int g_lag_in_frames
Allow lagged encoding.
Definition: aom_encoder.h:354
unsigned int rc_buf_initial_sz
Decoder Buffer Initial Size.
Definition: aom_encoder.h:557
unsigned int g_w
Width of the frame.
Definition: aom_encoder.h:267
unsigned int rc_undershoot_pct
Rate control adaptation undershoot control.
Definition: aom_encoder.h:521
unsigned int kf_max_dist
Keyframe maximum interval.
Definition: aom_encoder.h:629
aom_codec_er_flags_t g_error_resilient
Enable error resilient modes.
Definition: aom_encoder.h:333
unsigned int rc_max_quantizer
Maximum (Worst Quality) Quantizer.
Definition: aom_encoder.h:505
unsigned int rc_buf_optimal_sz
Decoder Buffer Optimal Size.
Definition: aom_encoder.h:566
unsigned int rc_min_quantizer
Minimum (Best Quality) Quantizer.
Definition: aom_encoder.h:495
unsigned int rc_target_bitrate
Target data rate.
Definition: aom_encoder.h:481
unsigned int rc_overshoot_pct
Rate control adaptation overshoot control.
Definition: aom_encoder.h:533
Image Descriptor.
Definition: aom_image.h:141
int num
Definition: aom_encoder.h:179
int den
Definition: aom_encoder.h:180
Definition: aomcx.h:1238
int temporal_layer_id
Definition: aomcx.h:1240
int spatial_layer_id
Definition: aomcx.h:1239
Definition: aomcx.h:1244
int max_quantizers[32]
Definition: aomcx.h:1247
int number_spatial_layers
Definition: aomcx.h:1245
int layer_target_bitrate[32]
Definition: aomcx.h:1252
int framerate_factor[8]
Definition: aomcx.h:1254
int min_quantizers[32]
Definition: aomcx.h:1248
int scaling_factor_den[4]
Definition: aomcx.h:1250
int number_temporal_layers
Definition: aomcx.h:1246
int scaling_factor_num[4]
Definition: aomcx.h:1249
Definition: aomcx.h:1258
int refresh[8]
Definition: aomcx.h:1263
int ref_idx[7]
Definition: aomcx.h:1262