2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
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.
12 #include "./vpx_config.h"
22 #include "vpx/vpx_encoder.h"
24 #include "vpx/vpx_decoder.h"
27 #include "third_party/libyuv/include/libyuv/scale.h"
31 #if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER
32 #include "vpx/vp8cx.h"
34 #if CONFIG_VP8_DECODER || CONFIG_VP9_DECODER
35 #include "vpx/vp8dx.h"
38 #include "./tools_common.h"
39 #include "vpx_ports/mem_ops.h"
40 #include "vpx_ports/vpx_timer.h"
41 #include "./vpxstats.h"
42 #include "./warnings.h"
43 #include "./webmenc.h"
44 #include "./y4minput.h"
46 /* Swallow warnings about unused results of fread/fwrite */
47 static size_t wrap_fread(void *ptr, size_t size, size_t nmemb,
49 return fread(ptr, size, nmemb, stream);
51 #define fread wrap_fread
53 static size_t wrap_fwrite(const void *ptr, size_t size, size_t nmemb,
55 return fwrite(ptr, size, nmemb, stream);
57 #define fwrite wrap_fwrite
60 static const char *exec_name;
62 static const struct codec_item {
64 const vpx_codec_iface_t *(*iface)(void);
65 const vpx_codec_iface_t *(*dx_iface)(void);
68 #if CONFIG_VP8_ENCODER && CONFIG_VP8_DECODER
69 {"vp8", &vpx_codec_vp8_cx, &vpx_codec_vp8_dx, VP8_FOURCC},
70 #elif CONFIG_VP8_ENCODER && !CONFIG_VP8_DECODER
71 {"vp8", &vpx_codec_vp8_cx, NULL, VP8_FOURCC},
73 #if CONFIG_VP9_ENCODER && CONFIG_VP9_DECODER
74 {"vp9", &vpx_codec_vp9_cx, &vpx_codec_vp9_dx, VP9_FOURCC},
75 #elif CONFIG_VP9_ENCODER && !CONFIG_VP9_DECODER
76 {"vp9", &vpx_codec_vp9_cx, NULL, VP9_FOURCC},
80 static void warn_or_exit_on_errorv(vpx_codec_ctx_t *ctx, int fatal,
81 const char *s, va_list ap) {
83 const char *detail = vpx_codec_error_detail(ctx);
85 vfprintf(stderr, s, ap);
86 fprintf(stderr, ": %s\n", vpx_codec_error(ctx));
89 fprintf(stderr, " %s\n", detail);
96 static void ctx_exit_on_error(vpx_codec_ctx_t *ctx, const char *s, ...) {
100 warn_or_exit_on_errorv(ctx, 1, s, ap);
104 static void warn_or_exit_on_error(vpx_codec_ctx_t *ctx, int fatal,
105 const char *s, ...) {
109 warn_or_exit_on_errorv(ctx, fatal, s, ap);
113 int read_frame(struct VpxInputContext *input_ctx, vpx_image_t *img) {
114 FILE *f = input_ctx->file;
115 y4m_input *y4m = &input_ctx->y4m;
118 if (input_ctx->file_type == FILE_TYPE_Y4M) {
119 if (y4m_input_fetch_frame(y4m, f, img) < 1)
122 shortread = read_yuv_frame(input_ctx, img);
128 int file_is_y4m(const char detect[4]) {
129 if (memcmp(detect, "YUV4", 4) == 0) {
135 int fourcc_is_ivf(const char detect[4]) {
136 if (memcmp(detect, "DKIF", 4) == 0) {
142 /* Murmur hash derived from public domain reference implementation at
143 * http:// sites.google.com/site/murmurhash/
145 static unsigned int murmur(const void *key, int len, unsigned int seed) {
146 const unsigned int m = 0x5bd1e995;
149 unsigned int h = seed ^ len;
151 const unsigned char *data = (const unsigned char *)key;
156 k = (unsigned int)data[0];
157 k |= (unsigned int)data[1] << 8;
158 k |= (unsigned int)data[2] << 16;
159 k |= (unsigned int)data[3] << 24;
190 static const arg_def_t debugmode = ARG_DEF("D", "debug", 0,
191 "Debug mode (makes output deterministic)");
192 static const arg_def_t outputfile = ARG_DEF("o", "output", 1,
194 static const arg_def_t use_yv12 = ARG_DEF(NULL, "yv12", 0,
195 "Input file is YV12 ");
196 static const arg_def_t use_i420 = ARG_DEF(NULL, "i420", 0,
197 "Input file is I420 (default)");
198 static const arg_def_t codecarg = ARG_DEF(NULL, "codec", 1,
200 static const arg_def_t passes = ARG_DEF("p", "passes", 1,
201 "Number of passes (1/2)");
202 static const arg_def_t pass_arg = ARG_DEF(NULL, "pass", 1,
203 "Pass to execute (1/2)");
204 static const arg_def_t fpf_name = ARG_DEF(NULL, "fpf", 1,
205 "First pass statistics file name");
206 static const arg_def_t limit = ARG_DEF(NULL, "limit", 1,
207 "Stop encoding after n input frames");
208 static const arg_def_t skip = ARG_DEF(NULL, "skip", 1,
209 "Skip the first n input frames");
210 static const arg_def_t deadline = ARG_DEF("d", "deadline", 1,
211 "Deadline per frame (usec)");
212 static const arg_def_t best_dl = ARG_DEF(NULL, "best", 0,
213 "Use Best Quality Deadline");
214 static const arg_def_t good_dl = ARG_DEF(NULL, "good", 0,
215 "Use Good Quality Deadline");
216 static const arg_def_t rt_dl = ARG_DEF(NULL, "rt", 0,
217 "Use Realtime Quality Deadline");
218 static const arg_def_t quietarg = ARG_DEF("q", "quiet", 0,
219 "Do not print encode progress");
220 static const arg_def_t verbosearg = ARG_DEF("v", "verbose", 0,
221 "Show encoder parameters");
222 static const arg_def_t psnrarg = ARG_DEF(NULL, "psnr", 0,
223 "Show PSNR in status line");
225 static const struct arg_enum_list test_decode_enum[] = {
226 {"off", TEST_DECODE_OFF},
227 {"fatal", TEST_DECODE_FATAL},
228 {"warn", TEST_DECODE_WARN},
231 static const arg_def_t recontest = ARG_DEF_ENUM(NULL, "test-decode", 1,
232 "Test encode/decode mismatch",
234 static const arg_def_t framerate = ARG_DEF(NULL, "fps", 1,
235 "Stream frame rate (rate/scale)");
236 static const arg_def_t use_ivf = ARG_DEF(NULL, "ivf", 0,
237 "Output IVF (default is WebM)");
238 static const arg_def_t out_part = ARG_DEF("P", "output-partitions", 0,
239 "Makes encoder output partitions. Requires IVF output!");
240 static const arg_def_t q_hist_n = ARG_DEF(NULL, "q-hist", 1,
241 "Show quantizer histogram (n-buckets)");
242 static const arg_def_t rate_hist_n = ARG_DEF(NULL, "rate-hist", 1,
243 "Show rate histogram (n-buckets)");
244 static const arg_def_t disable_warnings =
245 ARG_DEF(NULL, "disable-warnings", 0,
246 "Disable warnings about potentially incorrect encode settings.");
247 static const arg_def_t disable_warning_prompt =
248 ARG_DEF("y", "disable-warning-prompt", 0,
249 "Display warnings, but do not prompt user to continue.");
251 static const arg_def_t *main_args[] = {
253 &outputfile, &codecarg, &passes, &pass_arg, &fpf_name, &limit, &skip,
254 &deadline, &best_dl, &good_dl, &rt_dl,
255 &quietarg, &verbosearg, &psnrarg, &use_ivf, &out_part, &q_hist_n,
256 &rate_hist_n, &disable_warnings, &disable_warning_prompt,
260 static const arg_def_t usage = ARG_DEF("u", "usage", 1,
261 "Usage profile number to use");
262 static const arg_def_t threads = ARG_DEF("t", "threads", 1,
263 "Max number of threads to use");
264 static const arg_def_t profile = ARG_DEF(NULL, "profile", 1,
265 "Bitstream profile number to use");
266 static const arg_def_t width = ARG_DEF("w", "width", 1,
268 static const arg_def_t height = ARG_DEF("h", "height", 1,
270 static const struct arg_enum_list stereo_mode_enum[] = {
271 {"mono", STEREO_FORMAT_MONO},
272 {"left-right", STEREO_FORMAT_LEFT_RIGHT},
273 {"bottom-top", STEREO_FORMAT_BOTTOM_TOP},
274 {"top-bottom", STEREO_FORMAT_TOP_BOTTOM},
275 {"right-left", STEREO_FORMAT_RIGHT_LEFT},
278 static const arg_def_t stereo_mode = ARG_DEF_ENUM(NULL, "stereo-mode", 1,
279 "Stereo 3D video format", stereo_mode_enum);
280 static const arg_def_t timebase = ARG_DEF(NULL, "timebase", 1,
281 "Output timestamp precision (fractional seconds)");
282 static const arg_def_t error_resilient = ARG_DEF(NULL, "error-resilient", 1,
283 "Enable error resiliency features");
284 static const arg_def_t lag_in_frames = ARG_DEF(NULL, "lag-in-frames", 1,
285 "Max number of frames to lag");
287 static const arg_def_t *global_args[] = {
288 &use_yv12, &use_i420, &usage, &threads, &profile,
289 &width, &height, &stereo_mode, &timebase, &framerate,
294 static const arg_def_t dropframe_thresh = ARG_DEF(NULL, "drop-frame", 1,
295 "Temporal resampling threshold (buf %)");
296 static const arg_def_t resize_allowed = ARG_DEF(NULL, "resize-allowed", 1,
297 "Spatial resampling enabled (bool)");
298 static const arg_def_t resize_up_thresh = ARG_DEF(NULL, "resize-up", 1,
299 "Upscale threshold (buf %)");
300 static const arg_def_t resize_down_thresh = ARG_DEF(NULL, "resize-down", 1,
301 "Downscale threshold (buf %)");
302 static const struct arg_enum_list end_usage_enum[] = {
309 static const arg_def_t end_usage = ARG_DEF_ENUM(NULL, "end-usage", 1,
310 "Rate control mode", end_usage_enum);
311 static const arg_def_t target_bitrate = ARG_DEF(NULL, "target-bitrate", 1,
313 static const arg_def_t min_quantizer = ARG_DEF(NULL, "min-q", 1,
314 "Minimum (best) quantizer");
315 static const arg_def_t max_quantizer = ARG_DEF(NULL, "max-q", 1,
316 "Maximum (worst) quantizer");
317 static const arg_def_t undershoot_pct = ARG_DEF(NULL, "undershoot-pct", 1,
318 "Datarate undershoot (min) target (%)");
319 static const arg_def_t overshoot_pct = ARG_DEF(NULL, "overshoot-pct", 1,
320 "Datarate overshoot (max) target (%)");
321 static const arg_def_t buf_sz = ARG_DEF(NULL, "buf-sz", 1,
322 "Client buffer size (ms)");
323 static const arg_def_t buf_initial_sz = ARG_DEF(NULL, "buf-initial-sz", 1,
324 "Client initial buffer size (ms)");
325 static const arg_def_t buf_optimal_sz = ARG_DEF(NULL, "buf-optimal-sz", 1,
326 "Client optimal buffer size (ms)");
327 static const arg_def_t *rc_args[] = {
328 &dropframe_thresh, &resize_allowed, &resize_up_thresh, &resize_down_thresh,
329 &end_usage, &target_bitrate, &min_quantizer, &max_quantizer,
330 &undershoot_pct, &overshoot_pct, &buf_sz, &buf_initial_sz, &buf_optimal_sz,
335 static const arg_def_t bias_pct = ARG_DEF(NULL, "bias-pct", 1,
336 "CBR/VBR bias (0=CBR, 100=VBR)");
337 static const arg_def_t minsection_pct = ARG_DEF(NULL, "minsection-pct", 1,
338 "GOP min bitrate (% of target)");
339 static const arg_def_t maxsection_pct = ARG_DEF(NULL, "maxsection-pct", 1,
340 "GOP max bitrate (% of target)");
341 static const arg_def_t *rc_twopass_args[] = {
342 &bias_pct, &minsection_pct, &maxsection_pct, NULL
346 static const arg_def_t kf_min_dist = ARG_DEF(NULL, "kf-min-dist", 1,
347 "Minimum keyframe interval (frames)");
348 static const arg_def_t kf_max_dist = ARG_DEF(NULL, "kf-max-dist", 1,
349 "Maximum keyframe interval (frames)");
350 static const arg_def_t kf_disabled = ARG_DEF(NULL, "disable-kf", 0,
351 "Disable keyframe placement");
352 static const arg_def_t *kf_args[] = {
353 &kf_min_dist, &kf_max_dist, &kf_disabled, NULL
357 static const arg_def_t noise_sens = ARG_DEF(NULL, "noise-sensitivity", 1,
358 "Noise sensitivity (frames to blur)");
359 static const arg_def_t sharpness = ARG_DEF(NULL, "sharpness", 1,
360 "Filter sharpness (0-7)");
361 static const arg_def_t static_thresh = ARG_DEF(NULL, "static-thresh", 1,
362 "Motion detection threshold");
363 static const arg_def_t cpu_used = ARG_DEF(NULL, "cpu-used", 1,
364 "CPU Used (-16..16)");
365 static const arg_def_t auto_altref = ARG_DEF(NULL, "auto-alt-ref", 1,
366 "Enable automatic alt reference frames");
367 static const arg_def_t arnr_maxframes = ARG_DEF(NULL, "arnr-maxframes", 1,
368 "AltRef Max Frames");
369 static const arg_def_t arnr_strength = ARG_DEF(NULL, "arnr-strength", 1,
371 static const arg_def_t arnr_type = ARG_DEF(NULL, "arnr-type", 1,
373 static const struct arg_enum_list tuning_enum[] = {
374 {"psnr", VP8_TUNE_PSNR},
375 {"ssim", VP8_TUNE_SSIM},
378 static const arg_def_t tune_ssim = ARG_DEF_ENUM(NULL, "tune", 1,
379 "Material to favor", tuning_enum);
380 static const arg_def_t cq_level = ARG_DEF(NULL, "cq-level", 1,
381 "Constant/Constrained Quality level");
382 static const arg_def_t max_intra_rate_pct = ARG_DEF(NULL, "max-intra-rate", 1,
383 "Max I-frame bitrate (pct)");
385 #if CONFIG_VP8_ENCODER
386 static const arg_def_t token_parts =
387 ARG_DEF(NULL, "token-parts", 1, "Number of token partitions to use, log2");
388 static const arg_def_t *vp8_args[] = {
389 &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh,
390 &token_parts, &arnr_maxframes, &arnr_strength, &arnr_type,
391 &tune_ssim, &cq_level, &max_intra_rate_pct,
394 static const int vp8_arg_ctrl_map[] = {
395 VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF,
396 VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD,
397 VP8E_SET_TOKEN_PARTITIONS,
398 VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH, VP8E_SET_ARNR_TYPE,
399 VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, VP8E_SET_MAX_INTRA_BITRATE_PCT,
404 #if CONFIG_VP9_ENCODER
405 static const arg_def_t tile_cols =
406 ARG_DEF(NULL, "tile-columns", 1, "Number of tile columns to use, log2");
407 static const arg_def_t tile_rows =
408 ARG_DEF(NULL, "tile-rows", 1, "Number of tile rows to use, log2");
409 static const arg_def_t lossless = ARG_DEF(NULL, "lossless", 1, "Lossless mode");
410 static const arg_def_t frame_parallel_decoding = ARG_DEF(
411 NULL, "frame-parallel", 1, "Enable frame parallel decodability features");
412 static const arg_def_t aq_mode = ARG_DEF(
414 "Adaptive q mode (0: off (by default), 1: variance 2: complexity)");
416 static const arg_def_t *vp9_args[] = {
417 &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh,
418 &tile_cols, &tile_rows, &arnr_maxframes, &arnr_strength, &arnr_type,
419 &tune_ssim, &cq_level, &max_intra_rate_pct, &lossless,
420 &frame_parallel_decoding, &aq_mode,
423 static const int vp9_arg_ctrl_map[] = {
424 VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF,
425 VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD,
426 VP9E_SET_TILE_COLUMNS, VP9E_SET_TILE_ROWS,
427 VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH, VP8E_SET_ARNR_TYPE,
428 VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, VP8E_SET_MAX_INTRA_BITRATE_PCT,
429 VP9E_SET_LOSSLESS, VP9E_SET_FRAME_PARALLEL_DECODING, VP9E_SET_AQ_MODE,
434 static const arg_def_t *no_args[] = { NULL };
439 fprintf(stderr, "Usage: %s <options> -o dst_filename src_filename \n",
442 fprintf(stderr, "\nOptions:\n");
443 arg_show_usage(stderr, main_args);
444 fprintf(stderr, "\nEncoder Global Options:\n");
445 arg_show_usage(stderr, global_args);
446 fprintf(stderr, "\nRate Control Options:\n");
447 arg_show_usage(stderr, rc_args);
448 fprintf(stderr, "\nTwopass Rate Control Options:\n");
449 arg_show_usage(stderr, rc_twopass_args);
450 fprintf(stderr, "\nKeyframe Placement Options:\n");
451 arg_show_usage(stderr, kf_args);
452 #if CONFIG_VP8_ENCODER
453 fprintf(stderr, "\nVP8 Specific Options:\n");
454 arg_show_usage(stderr, vp8_args);
456 #if CONFIG_VP9_ENCODER
457 fprintf(stderr, "\nVP9 Specific Options:\n");
458 arg_show_usage(stderr, vp9_args);
460 fprintf(stderr, "\nStream timebase (--timebase):\n"
461 " The desired precision of timestamps in the output, expressed\n"
462 " in fractional seconds. Default is 1/1000.\n");
464 "Included encoders:\n"
467 for (i = 0; i < sizeof(codecs) / sizeof(codecs[0]); i++)
468 fprintf(stderr, " %-6s - %s\n",
470 vpx_codec_iface_name(codecs[i].iface()));
476 #define HIST_BAR_MAX 40
478 int low, high, count;
482 static int merge_hist_buckets(struct hist_bucket *bucket,
485 int small_bucket = 0, merge_bucket = INT_MAX, big_bucket = 0;
486 int buckets = *buckets_;
489 /* Find the extrema for this list of buckets */
490 big_bucket = small_bucket = 0;
491 for (i = 0; i < buckets; i++) {
492 if (bucket[i].count < bucket[small_bucket].count)
494 if (bucket[i].count > bucket[big_bucket].count)
498 /* If we have too many buckets, merge the smallest with an adjacent
501 while (buckets > max_buckets) {
502 int last_bucket = buckets - 1;
504 /* merge the small bucket with an adjacent one. */
505 if (small_bucket == 0)
507 else if (small_bucket == last_bucket)
508 merge_bucket = last_bucket - 1;
509 else if (bucket[small_bucket - 1].count < bucket[small_bucket + 1].count)
510 merge_bucket = small_bucket - 1;
512 merge_bucket = small_bucket + 1;
514 assert(abs(merge_bucket - small_bucket) <= 1);
515 assert(small_bucket < buckets);
516 assert(big_bucket < buckets);
517 assert(merge_bucket < buckets);
519 if (merge_bucket < small_bucket) {
520 bucket[merge_bucket].high = bucket[small_bucket].high;
521 bucket[merge_bucket].count += bucket[small_bucket].count;
523 bucket[small_bucket].high = bucket[merge_bucket].high;
524 bucket[small_bucket].count += bucket[merge_bucket].count;
525 merge_bucket = small_bucket;
528 assert(bucket[merge_bucket].low != bucket[merge_bucket].high);
532 /* Remove the merge_bucket from the list, and find the new small
533 * and big buckets while we're at it
535 big_bucket = small_bucket = 0;
536 for (i = 0; i < buckets; i++) {
537 if (i > merge_bucket)
538 bucket[i] = bucket[i + 1];
540 if (bucket[i].count < bucket[small_bucket].count)
542 if (bucket[i].count > bucket[big_bucket].count)
549 return bucket[big_bucket].count;
553 static void show_histogram(const struct hist_bucket *bucket,
557 const char *pat1, *pat2;
560 switch ((int)(log(bucket[buckets - 1].high) / log(10)) + 1) {
587 pat1 = "%12d %10s: ";
588 pat2 = "%12d-%10d: ";
592 for (i = 0; i < buckets; i++) {
597 pct = (float)(100.0 * bucket[i].count / total);
598 len = HIST_BAR_MAX * bucket[i].count / scale;
601 assert(len <= HIST_BAR_MAX);
603 if (bucket[i].low == bucket[i].high)
604 fprintf(stderr, pat1, bucket[i].low, "");
606 fprintf(stderr, pat2, bucket[i].low, bucket[i].high);
608 for (j = 0; j < HIST_BAR_MAX; j++)
609 fprintf(stderr, j < len ? "=" : " ");
610 fprintf(stderr, "\t%5d (%6.2f%%)\n", bucket[i].count, pct);
615 static void show_q_histogram(const int counts[64], int max_buckets) {
616 struct hist_bucket bucket[64];
623 for (i = 0; i < 64; i++) {
625 bucket[buckets].low = bucket[buckets].high = i;
626 bucket[buckets].count = counts[i];
632 fprintf(stderr, "\nQuantizer Selection:\n");
633 scale = merge_hist_buckets(bucket, &buckets, max_buckets);
634 show_histogram(bucket, buckets, total, scale);
638 #define RATE_BINS (100)
644 struct hist_bucket bucket[RATE_BINS];
649 static void init_rate_histogram(struct rate_hist *hist,
650 const vpx_codec_enc_cfg_t *cfg,
651 const vpx_rational_t *fps) {
654 /* Determine the number of samples in the buffer. Use the file's framerate
655 * to determine the number of frames in rc_buf_sz milliseconds, with an
656 * adjustment (5/4) to account for alt-refs
658 hist->samples = cfg->rc_buf_sz * 5 / 4 * fps->num / fps->den / 1000;
660 /* prevent division by zero */
661 if (hist->samples == 0)
664 hist->pts = calloc(hist->samples, sizeof(*hist->pts));
665 hist->sz = calloc(hist->samples, sizeof(*hist->sz));
666 for (i = 0; i < RATE_BINS; i++) {
667 hist->bucket[i].low = INT_MAX;
668 hist->bucket[i].high = 0;
669 hist->bucket[i].count = 0;
674 static void destroy_rate_histogram(struct rate_hist *hist) {
680 static void update_rate_histogram(struct rate_hist *hist,
681 const vpx_codec_enc_cfg_t *cfg,
682 const vpx_codec_cx_pkt_t *pkt) {
684 int64_t now, then, sum_sz = 0, avg_bitrate;
686 now = pkt->data.frame.pts * 1000
687 * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
689 idx = hist->frames++ % hist->samples;
690 hist->pts[idx] = now;
691 hist->sz[idx] = (int)pkt->data.frame.sz;
693 if (now < cfg->rc_buf_initial_sz)
698 /* Sum the size over the past rc_buf_sz ms */
699 for (i = hist->frames; i > 0 && hist->frames - i < hist->samples; i--) {
700 int i_idx = (i - 1) % hist->samples;
702 then = hist->pts[i_idx];
703 if (now - then > cfg->rc_buf_sz)
705 sum_sz += hist->sz[i_idx];
711 avg_bitrate = sum_sz * 8 * 1000 / (now - then);
712 idx = (int)(avg_bitrate * (RATE_BINS / 2) / (cfg->rc_target_bitrate * 1000));
715 if (idx > RATE_BINS - 1)
717 if (hist->bucket[idx].low > avg_bitrate)
718 hist->bucket[idx].low = (int)avg_bitrate;
719 if (hist->bucket[idx].high < avg_bitrate)
720 hist->bucket[idx].high = (int)avg_bitrate;
721 hist->bucket[idx].count++;
726 static void show_rate_histogram(struct rate_hist *hist,
727 const vpx_codec_enc_cfg_t *cfg,
732 for (i = 0; i < RATE_BINS; i++) {
733 if (hist->bucket[i].low == INT_MAX)
735 hist->bucket[buckets++] = hist->bucket[i];
738 fprintf(stderr, "\nRate (over %dms window):\n", cfg->rc_buf_sz);
739 scale = merge_hist_buckets(hist->bucket, &buckets, max_buckets);
740 show_histogram(hist->bucket, buckets, hist->total, scale);
743 #define mmin(a, b) ((a) < (b) ? (a) : (b))
744 static void find_mismatch(vpx_image_t *img1, vpx_image_t *img2,
745 int yloc[4], int uloc[4], int vloc[4]) {
746 const unsigned int bsize = 64;
747 const unsigned int bsizey = bsize >> img1->y_chroma_shift;
748 const unsigned int bsizex = bsize >> img1->x_chroma_shift;
749 const int c_w = (img1->d_w + img1->x_chroma_shift) >> img1->x_chroma_shift;
750 const int c_h = (img1->d_h + img1->y_chroma_shift) >> img1->y_chroma_shift;
751 unsigned int match = 1;
753 yloc[0] = yloc[1] = yloc[2] = yloc[3] = -1;
754 for (i = 0, match = 1; match && i < img1->d_h; i += bsize) {
755 for (j = 0; match && j < img1->d_w; j += bsize) {
757 int si = mmin(i + bsize, img1->d_h) - i;
758 int sj = mmin(j + bsize, img1->d_w) - j;
759 for (k = 0; match && k < si; k++)
760 for (l = 0; match && l < sj; l++) {
761 if (*(img1->planes[VPX_PLANE_Y] +
762 (i + k) * img1->stride[VPX_PLANE_Y] + j + l) !=
763 *(img2->planes[VPX_PLANE_Y] +
764 (i + k) * img2->stride[VPX_PLANE_Y] + j + l)) {
767 yloc[2] = *(img1->planes[VPX_PLANE_Y] +
768 (i + k) * img1->stride[VPX_PLANE_Y] + j + l);
769 yloc[3] = *(img2->planes[VPX_PLANE_Y] +
770 (i + k) * img2->stride[VPX_PLANE_Y] + j + l);
778 uloc[0] = uloc[1] = uloc[2] = uloc[3] = -1;
779 for (i = 0, match = 1; match && i < c_h; i += bsizey) {
780 for (j = 0; match && j < c_w; j += bsizex) {
782 int si = mmin(i + bsizey, c_h - i);
783 int sj = mmin(j + bsizex, c_w - j);
784 for (k = 0; match && k < si; k++)
785 for (l = 0; match && l < sj; l++) {
786 if (*(img1->planes[VPX_PLANE_U] +
787 (i + k) * img1->stride[VPX_PLANE_U] + j + l) !=
788 *(img2->planes[VPX_PLANE_U] +
789 (i + k) * img2->stride[VPX_PLANE_U] + j + l)) {
792 uloc[2] = *(img1->planes[VPX_PLANE_U] +
793 (i + k) * img1->stride[VPX_PLANE_U] + j + l);
794 uloc[3] = *(img2->planes[VPX_PLANE_U] +
795 (i + k) * img2->stride[VPX_PLANE_V] + j + l);
802 vloc[0] = vloc[1] = vloc[2] = vloc[3] = -1;
803 for (i = 0, match = 1; match && i < c_h; i += bsizey) {
804 for (j = 0; match && j < c_w; j += bsizex) {
806 int si = mmin(i + bsizey, c_h - i);
807 int sj = mmin(j + bsizex, c_w - j);
808 for (k = 0; match && k < si; k++)
809 for (l = 0; match && l < sj; l++) {
810 if (*(img1->planes[VPX_PLANE_V] +
811 (i + k) * img1->stride[VPX_PLANE_V] + j + l) !=
812 *(img2->planes[VPX_PLANE_V] +
813 (i + k) * img2->stride[VPX_PLANE_V] + j + l)) {
816 vloc[2] = *(img1->planes[VPX_PLANE_V] +
817 (i + k) * img1->stride[VPX_PLANE_V] + j + l);
818 vloc[3] = *(img2->planes[VPX_PLANE_V] +
819 (i + k) * img2->stride[VPX_PLANE_V] + j + l);
828 static int compare_img(vpx_image_t *img1, vpx_image_t *img2)
830 const int c_w = (img1->d_w + img1->x_chroma_shift) >> img1->x_chroma_shift;
831 const int c_h = (img1->d_h + img1->y_chroma_shift) >> img1->y_chroma_shift;
835 match &= (img1->fmt == img2->fmt);
836 match &= (img1->d_w == img2->d_w);
837 match &= (img1->d_h == img2->d_h);
839 for (i = 0; i < img1->d_h; i++)
840 match &= (memcmp(img1->planes[VPX_PLANE_Y]+i*img1->stride[VPX_PLANE_Y],
841 img2->planes[VPX_PLANE_Y]+i*img2->stride[VPX_PLANE_Y],
844 for (i = 0; i < c_h; i++)
845 match &= (memcmp(img1->planes[VPX_PLANE_U]+i*img1->stride[VPX_PLANE_U],
846 img2->planes[VPX_PLANE_U]+i*img2->stride[VPX_PLANE_U],
849 for (i = 0; i < c_h; i++)
850 match &= (memcmp(img1->planes[VPX_PLANE_V]+i*img1->stride[VPX_PLANE_U],
851 img2->planes[VPX_PLANE_V]+i*img2->stride[VPX_PLANE_U],
858 #define NELEMENTS(x) (sizeof(x)/sizeof(x[0]))
859 #define MAX(x,y) ((x)>(y)?(x):(y))
860 #if CONFIG_VP8_ENCODER && !CONFIG_VP9_ENCODER
861 #define ARG_CTRL_CNT_MAX NELEMENTS(vp8_arg_ctrl_map)
862 #elif !CONFIG_VP8_ENCODER && CONFIG_VP9_ENCODER
863 #define ARG_CTRL_CNT_MAX NELEMENTS(vp9_arg_ctrl_map)
865 #define ARG_CTRL_CNT_MAX MAX(NELEMENTS(vp8_arg_ctrl_map), \
866 NELEMENTS(vp9_arg_ctrl_map))
869 /* Per-stream configuration */
870 struct stream_config {
871 struct vpx_codec_enc_cfg cfg;
873 const char *stats_fn;
874 stereo_format_t stereo_fmt;
875 int arg_ctrls[ARG_CTRL_CNT_MAX][2];
878 int have_kf_max_dist;
882 struct stream_state {
884 struct stream_state *next;
885 struct stream_config config;
887 struct rate_hist rate_hist;
888 struct EbmlGlobal ebml;
890 uint64_t psnr_sse_total;
891 uint64_t psnr_samples_total;
892 double psnr_totals[4];
895 vpx_codec_ctx_t encoder;
896 unsigned int frames_out;
900 struct vpx_image *img;
901 vpx_codec_ctx_t decoder;
906 void validate_positive_rational(const char *msg,
907 struct vpx_rational *rat) {
914 die("Error: %s must be positive\n", msg);
917 die("Error: %s has zero denominator\n", msg);
921 static void parse_global_config(struct VpxEncoderConfig *global, char **argv) {
925 /* Initialize default parameters */
926 memset(global, 0, sizeof(*global));
927 global->codec = codecs;
929 global->use_i420 = 1;
930 /* Assign default deadline to good quality */
931 global->deadline = VPX_DL_GOOD_QUALITY;
933 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) {
936 if (arg_match(&arg, &codecarg, argi)) {
939 for (j = 0; j < sizeof(codecs) / sizeof(codecs[0]); j++)
940 if (!strcmp(codecs[j].name, arg.val))
944 global->codec = codecs + k;
946 die("Error: Unrecognized argument (%s) to --codec\n",
949 } else if (arg_match(&arg, &passes, argi)) {
950 global->passes = arg_parse_uint(&arg);
952 if (global->passes < 1 || global->passes > 2)
953 die("Error: Invalid number of passes (%d)\n", global->passes);
954 } else if (arg_match(&arg, &pass_arg, argi)) {
955 global->pass = arg_parse_uint(&arg);
957 if (global->pass < 1 || global->pass > 2)
958 die("Error: Invalid pass selected (%d)\n",
960 } else if (arg_match(&arg, &usage, argi))
961 global->usage = arg_parse_uint(&arg);
962 else if (arg_match(&arg, &deadline, argi))
963 global->deadline = arg_parse_uint(&arg);
964 else if (arg_match(&arg, &best_dl, argi))
965 global->deadline = VPX_DL_BEST_QUALITY;
966 else if (arg_match(&arg, &good_dl, argi))
967 global->deadline = VPX_DL_GOOD_QUALITY;
968 else if (arg_match(&arg, &rt_dl, argi))
969 global->deadline = VPX_DL_REALTIME;
970 else if (arg_match(&arg, &use_yv12, argi))
971 global->use_i420 = 0;
972 else if (arg_match(&arg, &use_i420, argi))
973 global->use_i420 = 1;
974 else if (arg_match(&arg, &quietarg, argi))
976 else if (arg_match(&arg, &verbosearg, argi))
978 else if (arg_match(&arg, &limit, argi))
979 global->limit = arg_parse_uint(&arg);
980 else if (arg_match(&arg, &skip, argi))
981 global->skip_frames = arg_parse_uint(&arg);
982 else if (arg_match(&arg, &psnrarg, argi))
983 global->show_psnr = 1;
984 else if (arg_match(&arg, &recontest, argi))
985 global->test_decode = arg_parse_enum_or_int(&arg);
986 else if (arg_match(&arg, &framerate, argi)) {
987 global->framerate = arg_parse_rational(&arg);
988 validate_positive_rational(arg.name, &global->framerate);
989 global->have_framerate = 1;
990 } else if (arg_match(&arg, &out_part, argi))
991 global->out_part = 1;
992 else if (arg_match(&arg, &debugmode, argi))
994 else if (arg_match(&arg, &q_hist_n, argi))
995 global->show_q_hist_buckets = arg_parse_uint(&arg);
996 else if (arg_match(&arg, &rate_hist_n, argi))
997 global->show_rate_hist_buckets = arg_parse_uint(&arg);
998 else if (arg_match(&arg, &disable_warnings, argi))
999 global->disable_warnings = 1;
1000 else if (arg_match(&arg, &disable_warning_prompt, argi))
1001 global->disable_warning_prompt = 1;
1006 /* Validate global config */
1007 if (global->passes == 0) {
1008 #if CONFIG_VP9_ENCODER
1009 // Make default VP9 passes = 2 until there is a better quality 1-pass
1011 global->passes = (global->codec->iface == vpx_codec_vp9_cx ? 2 : 1);
1018 /* DWIM: Assume the user meant passes=2 if pass=2 is specified */
1019 if (global->pass > global->passes) {
1020 warn("Assuming --pass=%d implies --passes=%d\n",
1021 global->pass, global->pass);
1022 global->passes = global->pass;
1028 void open_input_file(struct VpxInputContext *input) {
1029 /* Parse certain options from the input file, if possible */
1030 input->file = strcmp(input->filename, "-")
1031 ? fopen(input->filename, "rb") : set_binary_mode(stdin);
1034 fatal("Failed to open input file");
1036 if (!fseeko(input->file, 0, SEEK_END)) {
1037 /* Input file is seekable. Figure out how long it is, so we can get
1040 input->length = ftello(input->file);
1041 rewind(input->file);
1044 /* For RAW input sources, these bytes will applied on the first frame
1047 input->detect.buf_read = fread(input->detect.buf, 1, 4, input->file);
1048 input->detect.position = 0;
1050 if (input->detect.buf_read == 4
1051 && file_is_y4m(input->detect.buf)) {
1052 if (y4m_input_open(&input->y4m, input->file, input->detect.buf, 4,
1053 input->only_i420) >= 0) {
1054 input->file_type = FILE_TYPE_Y4M;
1055 input->width = input->y4m.pic_w;
1056 input->height = input->y4m.pic_h;
1057 input->framerate.numerator = input->y4m.fps_n;
1058 input->framerate.denominator = input->y4m.fps_d;
1059 input->use_i420 = 0;
1061 fatal("Unsupported Y4M stream.");
1062 } else if (input->detect.buf_read == 4 && fourcc_is_ivf(input->detect.buf)) {
1063 fatal("IVF is not supported as input.");
1065 input->file_type = FILE_TYPE_RAW;
1070 static void close_input_file(struct VpxInputContext *input) {
1071 fclose(input->file);
1072 if (input->file_type == FILE_TYPE_Y4M)
1073 y4m_input_close(&input->y4m);
1076 static struct stream_state *new_stream(struct VpxEncoderConfig *global,
1077 struct stream_state *prev) {
1078 struct stream_state *stream;
1080 stream = calloc(1, sizeof(*stream));
1082 fatal("Failed to allocate new stream.");
1084 memcpy(stream, prev, sizeof(*stream));
1086 prev->next = stream;
1088 vpx_codec_err_t res;
1090 /* Populate encoder configuration */
1091 res = vpx_codec_enc_config_default(global->codec->iface(),
1092 &stream->config.cfg,
1095 fatal("Failed to get config: %s\n", vpx_codec_err_to_string(res));
1097 /* Change the default timebase to a high enough value so that the
1098 * encoder will always create strictly increasing timestamps.
1100 stream->config.cfg.g_timebase.den = 1000;
1102 /* Never use the library's default resolution, require it be parsed
1103 * from the file or set on the command line.
1105 stream->config.cfg.g_w = 0;
1106 stream->config.cfg.g_h = 0;
1108 /* Initialize remaining stream parameters */
1109 stream->config.stereo_fmt = STEREO_FORMAT_MONO;
1110 stream->config.write_webm = 1;
1111 stream->ebml.last_pts_ms = -1;
1113 /* Allows removal of the application version from the EBML tags */
1114 stream->ebml.debug = global->debug;
1117 /* Output files must be specified for each stream */
1118 stream->config.out_fn = NULL;
1120 stream->next = NULL;
1125 static int parse_stream_params(struct VpxEncoderConfig *global,
1126 struct stream_state *stream,
1128 char **argi, **argj;
1130 static const arg_def_t **ctrl_args = no_args;
1131 static const int *ctrl_args_map = NULL;
1132 struct stream_config *config = &stream->config;
1133 int eos_mark_found = 0;
1135 /* Handle codec specific options */
1137 #if CONFIG_VP8_ENCODER
1138 } else if (global->codec->iface == vpx_codec_vp8_cx) {
1139 ctrl_args = vp8_args;
1140 ctrl_args_map = vp8_arg_ctrl_map;
1142 #if CONFIG_VP9_ENCODER
1143 } else if (global->codec->iface == vpx_codec_vp9_cx) {
1144 ctrl_args = vp9_args;
1145 ctrl_args_map = vp9_arg_ctrl_map;
1149 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) {
1152 /* Once we've found an end-of-stream marker (--) we want to continue
1153 * shifting arguments but not consuming them.
1155 if (eos_mark_found) {
1158 } else if (!strcmp(*argj, "--")) {
1164 else if (arg_match(&arg, &outputfile, argi))
1165 config->out_fn = arg.val;
1166 else if (arg_match(&arg, &fpf_name, argi))
1167 config->stats_fn = arg.val;
1168 else if (arg_match(&arg, &use_ivf, argi))
1169 config->write_webm = 0;
1170 else if (arg_match(&arg, &threads, argi))
1171 config->cfg.g_threads = arg_parse_uint(&arg);
1172 else if (arg_match(&arg, &profile, argi))
1173 config->cfg.g_profile = arg_parse_uint(&arg);
1174 else if (arg_match(&arg, &width, argi))
1175 config->cfg.g_w = arg_parse_uint(&arg);
1176 else if (arg_match(&arg, &height, argi))
1177 config->cfg.g_h = arg_parse_uint(&arg);
1178 else if (arg_match(&arg, &stereo_mode, argi))
1179 config->stereo_fmt = arg_parse_enum_or_int(&arg);
1180 else if (arg_match(&arg, &timebase, argi)) {
1181 config->cfg.g_timebase = arg_parse_rational(&arg);
1182 validate_positive_rational(arg.name, &config->cfg.g_timebase);
1183 } else if (arg_match(&arg, &error_resilient, argi))
1184 config->cfg.g_error_resilient = arg_parse_uint(&arg);
1185 else if (arg_match(&arg, &lag_in_frames, argi))
1186 config->cfg.g_lag_in_frames = arg_parse_uint(&arg);
1187 else if (arg_match(&arg, &dropframe_thresh, argi))
1188 config->cfg.rc_dropframe_thresh = arg_parse_uint(&arg);
1189 else if (arg_match(&arg, &resize_allowed, argi))
1190 config->cfg.rc_resize_allowed = arg_parse_uint(&arg);
1191 else if (arg_match(&arg, &resize_up_thresh, argi))
1192 config->cfg.rc_resize_up_thresh = arg_parse_uint(&arg);
1193 else if (arg_match(&arg, &resize_down_thresh, argi))
1194 config->cfg.rc_resize_down_thresh = arg_parse_uint(&arg);
1195 else if (arg_match(&arg, &end_usage, argi))
1196 config->cfg.rc_end_usage = arg_parse_enum_or_int(&arg);
1197 else if (arg_match(&arg, &target_bitrate, argi))
1198 config->cfg.rc_target_bitrate = arg_parse_uint(&arg);
1199 else if (arg_match(&arg, &min_quantizer, argi))
1200 config->cfg.rc_min_quantizer = arg_parse_uint(&arg);
1201 else if (arg_match(&arg, &max_quantizer, argi))
1202 config->cfg.rc_max_quantizer = arg_parse_uint(&arg);
1203 else if (arg_match(&arg, &undershoot_pct, argi))
1204 config->cfg.rc_undershoot_pct = arg_parse_uint(&arg);
1205 else if (arg_match(&arg, &overshoot_pct, argi))
1206 config->cfg.rc_overshoot_pct = arg_parse_uint(&arg);
1207 else if (arg_match(&arg, &buf_sz, argi))
1208 config->cfg.rc_buf_sz = arg_parse_uint(&arg);
1209 else if (arg_match(&arg, &buf_initial_sz, argi))
1210 config->cfg.rc_buf_initial_sz = arg_parse_uint(&arg);
1211 else if (arg_match(&arg, &buf_optimal_sz, argi))
1212 config->cfg.rc_buf_optimal_sz = arg_parse_uint(&arg);
1213 else if (arg_match(&arg, &bias_pct, argi)) {
1214 config->cfg.rc_2pass_vbr_bias_pct = arg_parse_uint(&arg);
1216 if (global->passes < 2)
1217 warn("option %s ignored in one-pass mode.\n", arg.name);
1218 } else if (arg_match(&arg, &minsection_pct, argi)) {
1219 config->cfg.rc_2pass_vbr_minsection_pct = arg_parse_uint(&arg);
1221 if (global->passes < 2)
1222 warn("option %s ignored in one-pass mode.\n", arg.name);
1223 } else if (arg_match(&arg, &maxsection_pct, argi)) {
1224 config->cfg.rc_2pass_vbr_maxsection_pct = arg_parse_uint(&arg);
1226 if (global->passes < 2)
1227 warn("option %s ignored in one-pass mode.\n", arg.name);
1228 } else if (arg_match(&arg, &kf_min_dist, argi))
1229 config->cfg.kf_min_dist = arg_parse_uint(&arg);
1230 else if (arg_match(&arg, &kf_max_dist, argi)) {
1231 config->cfg.kf_max_dist = arg_parse_uint(&arg);
1232 config->have_kf_max_dist = 1;
1233 } else if (arg_match(&arg, &kf_disabled, argi))
1234 config->cfg.kf_mode = VPX_KF_DISABLED;
1238 for (i = 0; ctrl_args[i]; i++) {
1239 if (arg_match(&arg, ctrl_args[i], argi)) {
1243 /* Point either to the next free element or the first
1244 * instance of this control.
1246 for (j = 0; j < config->arg_ctrl_cnt; j++)
1247 if (config->arg_ctrls[j][0] == ctrl_args_map[i])
1251 assert(j < ARG_CTRL_CNT_MAX);
1252 if (j < ARG_CTRL_CNT_MAX) {
1253 config->arg_ctrls[j][0] = ctrl_args_map[i];
1254 config->arg_ctrls[j][1] = arg_parse_enum_or_int(&arg);
1255 if (j == config->arg_ctrl_cnt)
1256 config->arg_ctrl_cnt++;
1267 return eos_mark_found;
1271 #define FOREACH_STREAM(func) \
1273 struct stream_state *stream; \
1274 for (stream = streams; stream; stream = stream->next) { \
1280 static void validate_stream_config(struct stream_state *stream) {
1281 struct stream_state *streami;
1283 if (!stream->config.cfg.g_w || !stream->config.cfg.g_h)
1284 fatal("Stream %d: Specify stream dimensions with --width (-w) "
1285 " and --height (-h)", stream->index);
1287 for (streami = stream; streami; streami = streami->next) {
1288 /* All streams require output files */
1289 if (!streami->config.out_fn)
1290 fatal("Stream %d: Output file is required (specify with -o)",
1293 /* Check for two streams outputting to the same file */
1294 if (streami != stream) {
1295 const char *a = stream->config.out_fn;
1296 const char *b = streami->config.out_fn;
1297 if (!strcmp(a, b) && strcmp(a, "/dev/null") && strcmp(a, ":nul"))
1298 fatal("Stream %d: duplicate output file (from stream %d)",
1299 streami->index, stream->index);
1302 /* Check for two streams sharing a stats file. */
1303 if (streami != stream) {
1304 const char *a = stream->config.stats_fn;
1305 const char *b = streami->config.stats_fn;
1306 if (a && b && !strcmp(a, b))
1307 fatal("Stream %d: duplicate stats file (from stream %d)",
1308 streami->index, stream->index);
1314 static void set_stream_dimensions(struct stream_state *stream,
1317 if (!stream->config.cfg.g_w) {
1318 if (!stream->config.cfg.g_h)
1319 stream->config.cfg.g_w = w;
1321 stream->config.cfg.g_w = w * stream->config.cfg.g_h / h;
1323 if (!stream->config.cfg.g_h) {
1324 stream->config.cfg.g_h = h * stream->config.cfg.g_w / w;
1329 static void set_default_kf_interval(struct stream_state *stream,
1330 struct VpxEncoderConfig *global) {
1331 /* Use a max keyframe interval of 5 seconds, if none was
1332 * specified on the command line.
1334 if (!stream->config.have_kf_max_dist) {
1335 double framerate = (double)global->framerate.num / global->framerate.den;
1336 if (framerate > 0.0)
1337 stream->config.cfg.kf_max_dist = (unsigned int)(5.0 * framerate);
1342 static void show_stream_config(struct stream_state *stream,
1343 struct VpxEncoderConfig *global,
1344 struct VpxInputContext *input) {
1346 #define SHOW(field) \
1347 fprintf(stderr, " %-28s = %d\n", #field, stream->config.cfg.field)
1349 if (stream->index == 0) {
1350 fprintf(stderr, "Codec: %s\n",
1351 vpx_codec_iface_name(global->codec->iface()));
1352 fprintf(stderr, "Source file: %s Format: %s\n", input->filename,
1353 input->use_i420 ? "I420" : "YV12");
1355 if (stream->next || stream->index)
1356 fprintf(stderr, "\nStream Index: %d\n", stream->index);
1357 fprintf(stderr, "Destination file: %s\n", stream->config.out_fn);
1358 fprintf(stderr, "Encoder parameters:\n");
1365 SHOW(g_timebase.num);
1366 SHOW(g_timebase.den);
1367 SHOW(g_error_resilient);
1369 SHOW(g_lag_in_frames);
1370 SHOW(rc_dropframe_thresh);
1371 SHOW(rc_resize_allowed);
1372 SHOW(rc_resize_up_thresh);
1373 SHOW(rc_resize_down_thresh);
1375 SHOW(rc_target_bitrate);
1376 SHOW(rc_min_quantizer);
1377 SHOW(rc_max_quantizer);
1378 SHOW(rc_undershoot_pct);
1379 SHOW(rc_overshoot_pct);
1381 SHOW(rc_buf_initial_sz);
1382 SHOW(rc_buf_optimal_sz);
1383 SHOW(rc_2pass_vbr_bias_pct);
1384 SHOW(rc_2pass_vbr_minsection_pct);
1385 SHOW(rc_2pass_vbr_maxsection_pct);
1392 static void open_output_file(struct stream_state *stream,
1393 struct VpxEncoderConfig *global) {
1394 const char *fn = stream->config.out_fn;
1395 const struct vpx_codec_enc_cfg *const cfg = &stream->config.cfg;
1397 if (cfg->g_pass == VPX_RC_FIRST_PASS)
1400 stream->file = strcmp(fn, "-") ? fopen(fn, "wb") : set_binary_mode(stdout);
1403 fatal("Failed to open output file");
1405 if (stream->config.write_webm && fseek(stream->file, 0, SEEK_CUR))
1406 fatal("WebM output to pipes not supported.");
1408 if (stream->config.write_webm) {
1409 stream->ebml.stream = stream->file;
1410 write_webm_file_header(&stream->ebml, cfg,
1412 stream->config.stereo_fmt,
1413 global->codec->fourcc);
1415 ivf_write_file_header(stream->file, cfg, global->codec->fourcc, 0);
1420 static void close_output_file(struct stream_state *stream,
1421 unsigned int fourcc) {
1422 const struct vpx_codec_enc_cfg *const cfg = &stream->config.cfg;
1424 if (cfg->g_pass == VPX_RC_FIRST_PASS)
1427 if (stream->config.write_webm) {
1428 write_webm_file_footer(&stream->ebml, stream->hash);
1429 free(stream->ebml.cue_list);
1430 stream->ebml.cue_list = NULL;
1432 if (!fseek(stream->file, 0, SEEK_SET))
1433 ivf_write_file_header(stream->file, &stream->config.cfg,
1435 stream->frames_out);
1438 fclose(stream->file);
1442 static void setup_pass(struct stream_state *stream,
1443 struct VpxEncoderConfig *global,
1445 if (stream->config.stats_fn) {
1446 if (!stats_open_file(&stream->stats, stream->config.stats_fn,
1448 fatal("Failed to open statistics store");
1450 if (!stats_open_mem(&stream->stats, pass))
1451 fatal("Failed to open statistics store");
1454 stream->config.cfg.g_pass = global->passes == 2
1455 ? pass ? VPX_RC_LAST_PASS : VPX_RC_FIRST_PASS
1458 stream->config.cfg.rc_twopass_stats_in = stats_get(&stream->stats);
1460 stream->cx_time = 0;
1462 stream->frames_out = 0;
1466 static void initialize_encoder(struct stream_state *stream,
1467 struct VpxEncoderConfig *global) {
1471 flags |= global->show_psnr ? VPX_CODEC_USE_PSNR : 0;
1472 flags |= global->out_part ? VPX_CODEC_USE_OUTPUT_PARTITION : 0;
1474 /* Construct Encoder Context */
1475 vpx_codec_enc_init(&stream->encoder, global->codec->iface(),
1476 &stream->config.cfg, flags);
1477 ctx_exit_on_error(&stream->encoder, "Failed to initialize encoder");
1479 /* Note that we bypass the vpx_codec_control wrapper macro because
1480 * we're being clever to store the control IDs in an array. Real
1481 * applications will want to make use of the enumerations directly
1483 for (i = 0; i < stream->config.arg_ctrl_cnt; i++) {
1484 int ctrl = stream->config.arg_ctrls[i][0];
1485 int value = stream->config.arg_ctrls[i][1];
1486 if (vpx_codec_control_(&stream->encoder, ctrl, value))
1487 fprintf(stderr, "Error: Tried to set control %d = %d\n",
1490 ctx_exit_on_error(&stream->encoder, "Failed to control codec");
1494 if (global->test_decode != TEST_DECODE_OFF) {
1495 vpx_codec_dec_init(&stream->decoder, global->codec->dx_iface(), NULL, 0);
1501 static void encode_frame(struct stream_state *stream,
1502 struct VpxEncoderConfig *global,
1503 struct vpx_image *img,
1504 unsigned int frames_in) {
1505 vpx_codec_pts_t frame_start, next_frame_start;
1506 struct vpx_codec_enc_cfg *cfg = &stream->config.cfg;
1507 struct vpx_usec_timer timer;
1509 frame_start = (cfg->g_timebase.den * (int64_t)(frames_in - 1)
1510 * global->framerate.den)
1511 / cfg->g_timebase.num / global->framerate.num;
1512 next_frame_start = (cfg->g_timebase.den * (int64_t)(frames_in)
1513 * global->framerate.den)
1514 / cfg->g_timebase.num / global->framerate.num;
1516 /* Scale if necessary */
1517 if (img && (img->d_w != cfg->g_w || img->d_h != cfg->g_h)) {
1519 stream->img = vpx_img_alloc(NULL, VPX_IMG_FMT_I420,
1520 cfg->g_w, cfg->g_h, 16);
1521 I420Scale(img->planes[VPX_PLANE_Y], img->stride[VPX_PLANE_Y],
1522 img->planes[VPX_PLANE_U], img->stride[VPX_PLANE_U],
1523 img->planes[VPX_PLANE_V], img->stride[VPX_PLANE_V],
1525 stream->img->planes[VPX_PLANE_Y],
1526 stream->img->stride[VPX_PLANE_Y],
1527 stream->img->planes[VPX_PLANE_U],
1528 stream->img->stride[VPX_PLANE_U],
1529 stream->img->planes[VPX_PLANE_V],
1530 stream->img->stride[VPX_PLANE_V],
1531 stream->img->d_w, stream->img->d_h,
1537 vpx_usec_timer_start(&timer);
1538 vpx_codec_encode(&stream->encoder, img, frame_start,
1539 (unsigned long)(next_frame_start - frame_start),
1540 0, global->deadline);
1541 vpx_usec_timer_mark(&timer);
1542 stream->cx_time += vpx_usec_timer_elapsed(&timer);
1543 ctx_exit_on_error(&stream->encoder, "Stream %d: Failed to encode frame",
1548 static void update_quantizer_histogram(struct stream_state *stream) {
1549 if (stream->config.cfg.g_pass != VPX_RC_FIRST_PASS) {
1552 vpx_codec_control(&stream->encoder, VP8E_GET_LAST_QUANTIZER_64, &q);
1553 ctx_exit_on_error(&stream->encoder, "Failed to read quantizer");
1554 stream->counts[q]++;
1559 static void get_cx_data(struct stream_state *stream,
1560 struct VpxEncoderConfig *global,
1562 const vpx_codec_cx_pkt_t *pkt;
1563 const struct vpx_codec_enc_cfg *cfg = &stream->config.cfg;
1564 vpx_codec_iter_t iter = NULL;
1567 while ((pkt = vpx_codec_get_cx_data(&stream->encoder, &iter))) {
1568 static size_t fsize = 0;
1569 static off_t ivf_header_pos = 0;
1571 switch (pkt->kind) {
1572 case VPX_CODEC_CX_FRAME_PKT:
1573 if (!(pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT)) {
1574 stream->frames_out++;
1577 fprintf(stderr, " %6luF", (unsigned long)pkt->data.frame.sz);
1579 update_rate_histogram(&stream->rate_hist, cfg, pkt);
1580 if (stream->config.write_webm) {
1581 /* Update the hash */
1582 if (!stream->ebml.debug)
1583 stream->hash = murmur(pkt->data.frame.buf,
1584 (int)pkt->data.frame.sz,
1587 write_webm_block(&stream->ebml, cfg, pkt);
1589 if (pkt->data.frame.partition_id <= 0) {
1590 ivf_header_pos = ftello(stream->file);
1591 fsize = pkt->data.frame.sz;
1593 ivf_write_frame_header(stream->file, pkt);
1595 fsize += pkt->data.frame.sz;
1597 if (!(pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT)) {
1598 off_t currpos = ftello(stream->file);
1599 fseeko(stream->file, ivf_header_pos, SEEK_SET);
1600 ivf_write_frame_size(stream->file, fsize);
1601 fseeko(stream->file, currpos, SEEK_SET);
1605 (void) fwrite(pkt->data.frame.buf, 1, pkt->data.frame.sz,
1608 stream->nbytes += pkt->data.raw.sz;
1612 if (global->test_decode != TEST_DECODE_OFF && !stream->mismatch_seen) {
1613 vpx_codec_decode(&stream->decoder, pkt->data.frame.buf,
1614 pkt->data.frame.sz, NULL, 0);
1615 if (stream->decoder.err) {
1616 warn_or_exit_on_error(&stream->decoder,
1617 global->test_decode == TEST_DECODE_FATAL,
1618 "Failed to decode frame %d in stream %d",
1619 stream->frames_out + 1, stream->index);
1620 stream->mismatch_seen = stream->frames_out + 1;
1625 case VPX_CODEC_STATS_PKT:
1626 stream->frames_out++;
1627 stats_write(&stream->stats,
1628 pkt->data.twopass_stats.buf,
1629 pkt->data.twopass_stats.sz);
1630 stream->nbytes += pkt->data.raw.sz;
1632 case VPX_CODEC_PSNR_PKT:
1634 if (global->show_psnr) {
1637 stream->psnr_sse_total += pkt->data.psnr.sse[0];
1638 stream->psnr_samples_total += pkt->data.psnr.samples[0];
1639 for (i = 0; i < 4; i++) {
1641 fprintf(stderr, "%.3f ", pkt->data.psnr.psnr[i]);
1642 stream->psnr_totals[i] += pkt->data.psnr.psnr[i];
1644 stream->psnr_count++;
1655 static void show_psnr(struct stream_state *stream) {
1659 if (!stream->psnr_count)
1662 fprintf(stderr, "Stream %d PSNR (Overall/Avg/Y/U/V)", stream->index);
1663 ovpsnr = vp8_mse2psnr((double)stream->psnr_samples_total, 255.0,
1664 (double)stream->psnr_sse_total);
1665 fprintf(stderr, " %.3f", ovpsnr);
1667 for (i = 0; i < 4; i++) {
1668 fprintf(stderr, " %.3f", stream->psnr_totals[i] / stream->psnr_count);
1670 fprintf(stderr, "\n");
1674 static float usec_to_fps(uint64_t usec, unsigned int frames) {
1675 return (float)(usec > 0 ? frames * 1000000.0 / (float)usec : 0);
1679 static void test_decode(struct stream_state *stream,
1680 enum TestDecodeFatality fatal,
1681 const struct codec_item *codec) {
1682 vpx_image_t enc_img, dec_img;
1684 if (stream->mismatch_seen)
1687 /* Get the internal reference frame */
1688 if (codec->fourcc == VP8_FOURCC) {
1689 struct vpx_ref_frame ref_enc, ref_dec;
1692 width = (stream->config.cfg.g_w + 15) & ~15;
1693 height = (stream->config.cfg.g_h + 15) & ~15;
1694 vpx_img_alloc(&ref_enc.img, VPX_IMG_FMT_I420, width, height, 1);
1695 enc_img = ref_enc.img;
1696 vpx_img_alloc(&ref_dec.img, VPX_IMG_FMT_I420, width, height, 1);
1697 dec_img = ref_dec.img;
1699 ref_enc.frame_type = VP8_LAST_FRAME;
1700 ref_dec.frame_type = VP8_LAST_FRAME;
1701 vpx_codec_control(&stream->encoder, VP8_COPY_REFERENCE, &ref_enc);
1702 vpx_codec_control(&stream->decoder, VP8_COPY_REFERENCE, &ref_dec);
1704 struct vp9_ref_frame ref;
1707 vpx_codec_control(&stream->encoder, VP9_GET_REFERENCE, &ref);
1709 vpx_codec_control(&stream->decoder, VP9_GET_REFERENCE, &ref);
1712 ctx_exit_on_error(&stream->encoder, "Failed to get encoder reference frame");
1713 ctx_exit_on_error(&stream->decoder, "Failed to get decoder reference frame");
1715 if (!compare_img(&enc_img, &dec_img)) {
1716 int y[4], u[4], v[4];
1717 find_mismatch(&enc_img, &dec_img, y, u, v);
1718 stream->decoder.err = 1;
1719 warn_or_exit_on_error(&stream->decoder, fatal == TEST_DECODE_FATAL,
1720 "Stream %d: Encode/decode mismatch on frame %d at"
1721 " Y[%d, %d] {%d/%d},"
1722 " U[%d, %d] {%d/%d},"
1723 " V[%d, %d] {%d/%d}",
1724 stream->index, stream->frames_out,
1725 y[0], y[1], y[2], y[3],
1726 u[0], u[1], u[2], u[3],
1727 v[0], v[1], v[2], v[3]);
1728 stream->mismatch_seen = stream->frames_out;
1731 vpx_img_free(&enc_img);
1732 vpx_img_free(&dec_img);
1736 static void print_time(const char *label, int64_t etl) {
1737 int hours, mins, secs;
1741 etl -= hours * 3600;
1746 fprintf(stderr, "[%3s %2d:%02d:%02d] ",
1747 label, hours, mins, secs);
1749 fprintf(stderr, "[%3s unknown] ", label);
1754 int main(int argc, const char **argv_) {
1757 int frame_avail, got_data;
1759 struct VpxInputContext input = {0};
1760 struct VpxEncoderConfig global;
1761 struct stream_state *streams = NULL;
1762 char **argv, **argi;
1763 uint64_t cx_time = 0;
1767 exec_name = argv_[0];
1772 /* Setup default input stream settings */
1773 input.framerate.numerator = 30;
1774 input.framerate.denominator = 1;
1776 input.only_i420 = 1;
1778 /* First parse the global configuration values, because we want to apply
1779 * other parameters on top of the default configuration provided by the
1782 argv = argv_dup(argc - 1, argv_ + 1);
1783 parse_global_config(&global, argv);
1787 /* Now parse each stream's parameters. Using a local scope here
1788 * due to the use of 'stream' as loop variable in FOREACH_STREAM
1791 struct stream_state *stream = NULL;
1794 stream = new_stream(&global, stream);
1798 } while (parse_stream_params(&global, stream, argv));
1801 /* Check for unrecognized options */
1802 for (argi = argv; *argi; argi++)
1803 if (argi[0][0] == '-' && argi[0][1])
1804 die("Error: Unrecognized option %s\n", *argi);
1806 FOREACH_STREAM(check_encoder_config(global.disable_warning_prompt,
1807 &global, &stream->config.cfg););
1809 /* Handle non-option arguments */
1810 input.filename = argv[0];
1812 if (!input.filename)
1816 /* Decide if other chroma subsamplings than 4:2:0 are supported */
1817 if (global.codec->fourcc == VP9_FOURCC)
1818 input.only_i420 = 0;
1821 for (pass = global.pass ? global.pass - 1 : 0; pass < global.passes; pass++) {
1822 int frames_in = 0, seen_frames = 0;
1823 int64_t estimated_time_left = -1;
1824 int64_t average_rate = -1;
1825 off_t lagged_count = 0;
1827 open_input_file(&input);
1829 /* If the input file doesn't specify its w/h (raw files), try to get
1830 * the data from the first stream's configuration.
1832 if (!input.width || !input.height)
1834 if (stream->config.cfg.g_w && stream->config.cfg.g_h) {
1835 input.width = stream->config.cfg.g_w;
1836 input.height = stream->config.cfg.g_h;
1841 /* Update stream configurations from the input file's parameters */
1842 if (!input.width || !input.height)
1843 fatal("Specify stream dimensions with --width (-w) "
1844 " and --height (-h)");
1845 FOREACH_STREAM(set_stream_dimensions(stream, input.width, input.height));
1846 FOREACH_STREAM(validate_stream_config(stream));
1848 /* Ensure that --passes and --pass are consistent. If --pass is set and
1849 * --passes=2, ensure --fpf was set.
1851 if (global.pass && global.passes == 2)
1853 if (!stream->config.stats_fn)
1854 die("Stream %d: Must specify --fpf when --pass=%d"
1855 " and --passes=2\n", stream->index, global.pass);
1858 /* Use the frame rate from the file only if none was specified
1859 * on the command-line.
1861 if (!global.have_framerate) {
1862 global.framerate.num = input.framerate.numerator;
1863 global.framerate.den = input.framerate.denominator;
1866 FOREACH_STREAM(set_default_kf_interval(stream, &global));
1868 /* Show configuration */
1869 if (global.verbose && pass == 0)
1870 FOREACH_STREAM(show_stream_config(stream, &global, &input));
1872 if (pass == (global.pass ? global.pass - 1 : 0)) {
1873 if (input.file_type == FILE_TYPE_Y4M)
1874 /*The Y4M reader does its own allocation.
1875 Just initialize this here to avoid problems if we never read any
1877 memset(&raw, 0, sizeof(raw));
1880 input.use_i420 ? VPX_IMG_FMT_I420
1882 input.width, input.height, 32);
1884 FOREACH_STREAM(init_rate_histogram(&stream->rate_hist,
1885 &stream->config.cfg,
1886 &global.framerate));
1889 FOREACH_STREAM(setup_pass(stream, &global, pass));
1890 FOREACH_STREAM(open_output_file(stream, &global));
1891 FOREACH_STREAM(initialize_encoder(stream, &global));
1896 while (frame_avail || got_data) {
1897 struct vpx_usec_timer timer;
1899 if (!global.limit || frames_in < global.limit) {
1900 frame_avail = read_frame(&input, &raw);
1904 seen_frames = frames_in > global.skip_frames ?
1905 frames_in - global.skip_frames : 0;
1907 if (!global.quiet) {
1908 float fps = usec_to_fps(cx_time, seen_frames);
1909 fprintf(stderr, "\rPass %d/%d ", pass + 1, global.passes);
1911 if (stream_cnt == 1)
1913 "frame %4d/%-4d %7"PRId64"B ",
1914 frames_in, streams->frames_out, (int64_t)streams->nbytes);
1916 fprintf(stderr, "frame %4d ", frames_in);
1918 fprintf(stderr, "%7"PRId64" %s %.2f %s ",
1919 cx_time > 9999999 ? cx_time / 1000 : cx_time,
1920 cx_time > 9999999 ? "ms" : "us",
1921 fps >= 1.0 ? fps : fps * 60,
1922 fps >= 1.0 ? "fps" : "fpm");
1923 print_time("ETA", estimated_time_left);
1924 fprintf(stderr, "\033[K");
1930 if (frames_in > global.skip_frames) {
1931 vpx_usec_timer_start(&timer);
1932 FOREACH_STREAM(encode_frame(stream, &global,
1933 frame_avail ? &raw : NULL,
1935 vpx_usec_timer_mark(&timer);
1936 cx_time += vpx_usec_timer_elapsed(&timer);
1938 FOREACH_STREAM(update_quantizer_histogram(stream));
1941 FOREACH_STREAM(get_cx_data(stream, &global, &got_data));
1943 if (!got_data && input.length && !streams->frames_out) {
1944 lagged_count = global.limit ? seen_frames : ftello(input.file);
1945 } else if (input.length) {
1950 int frame_in_lagged = (seen_frames - lagged_count) * 1000;
1952 rate = cx_time ? frame_in_lagged * (int64_t)1000000 / cx_time : 0;
1953 remaining = 1000 * (global.limit - global.skip_frames
1954 - seen_frames + lagged_count);
1956 off_t input_pos = ftello(input.file);
1957 off_t input_pos_lagged = input_pos - lagged_count;
1958 int64_t limit = input.length;
1960 rate = cx_time ? input_pos_lagged * (int64_t)1000000 / cx_time : 0;
1961 remaining = limit - input_pos + lagged_count;
1964 average_rate = (average_rate <= 0)
1966 : (average_rate * 7 + rate) / 8;
1967 estimated_time_left = average_rate ? remaining / average_rate : -1;
1970 if (got_data && global.test_decode != TEST_DECODE_OFF)
1971 FOREACH_STREAM(test_decode(stream, global.test_decode, global.codec));
1978 fprintf(stderr, "\n");
1981 FOREACH_STREAM(fprintf(
1983 "\rPass %d/%d frame %4d/%-4d %7"PRId64"B %7lub/f %7"PRId64"b/s"
1984 " %7"PRId64" %s (%.2f fps)\033[K\n", pass + 1,
1985 global.passes, frames_in, stream->frames_out, (int64_t)stream->nbytes,
1986 seen_frames ? (unsigned long)(stream->nbytes * 8 / seen_frames) : 0,
1987 seen_frames ? (int64_t)stream->nbytes * 8
1988 * (int64_t)global.framerate.num / global.framerate.den
1991 stream->cx_time > 9999999 ? stream->cx_time / 1000 : stream->cx_time,
1992 stream->cx_time > 9999999 ? "ms" : "us",
1993 usec_to_fps(stream->cx_time, seen_frames));
1996 if (global.show_psnr)
1997 FOREACH_STREAM(show_psnr(stream));
1999 FOREACH_STREAM(vpx_codec_destroy(&stream->encoder));
2001 if (global.test_decode != TEST_DECODE_OFF) {
2002 FOREACH_STREAM(vpx_codec_destroy(&stream->decoder));
2005 close_input_file(&input);
2007 if (global.test_decode == TEST_DECODE_FATAL) {
2008 FOREACH_STREAM(res |= stream->mismatch_seen);
2010 FOREACH_STREAM(close_output_file(stream, global.codec->fourcc));
2012 FOREACH_STREAM(stats_close(&stream->stats, global.passes - 1));
2018 if (global.show_q_hist_buckets)
2019 FOREACH_STREAM(show_q_histogram(stream->counts,
2020 global.show_q_hist_buckets));
2022 if (global.show_rate_hist_buckets)
2023 FOREACH_STREAM(show_rate_histogram(&stream->rate_hist,
2024 &stream->config.cfg,
2025 global.show_rate_hist_buckets));
2026 FOREACH_STREAM(destroy_rate_histogram(&stream->rate_hist));
2028 #if CONFIG_INTERNAL_STATS
2029 /* TODO(jkoleszar): This doesn't belong in this executable. Do it for now,
2030 * to match some existing utilities.
2032 if (!(global.pass == 1 && global.passes == 2))
2034 FILE *f = fopen("opsnr.stt", "a");
2035 if (stream->mismatch_seen) {
2036 fprintf(f, "First mismatch occurred in frame %d\n",
2037 stream->mismatch_seen);
2039 fprintf(f, "No mismatch detected in recon buffers\n");
2048 return res ? EXIT_FAILURE : EXIT_SUCCESS;