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.
11 #include "./vpx_config.h"
21 #include "vpx/vpx_encoder.h"
23 #include "vpx/vpx_decoder.h"
26 #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 "./webmenc.h"
43 #include "./y4minput.h"
45 /* Swallow warnings about unused results of fread/fwrite */
46 static size_t wrap_fread(void *ptr, size_t size, size_t nmemb,
48 return fread(ptr, size, nmemb, stream);
50 #define fread wrap_fread
52 static size_t wrap_fwrite(const void *ptr, size_t size, size_t nmemb,
54 return fwrite(ptr, size, nmemb, stream);
56 #define fwrite wrap_fwrite
59 static const char *exec_name;
61 static const struct codec_item {
63 const vpx_codec_iface_t *(*iface)(void);
64 const vpx_codec_iface_t *(*dx_iface)(void);
67 #if CONFIG_VP8_ENCODER && CONFIG_VP8_DECODER
68 {"vp8", &vpx_codec_vp8_cx, &vpx_codec_vp8_dx, VP8_FOURCC},
69 #elif CONFIG_VP8_ENCODER && !CONFIG_VP8_DECODER
70 {"vp8", &vpx_codec_vp8_cx, NULL, VP8_FOURCC},
72 #if CONFIG_VP9_ENCODER && CONFIG_VP9_DECODER
73 {"vp9", &vpx_codec_vp9_cx, &vpx_codec_vp9_dx, VP9_FOURCC},
74 #elif CONFIG_VP9_ENCODER && !CONFIG_VP9_DECODER
75 {"vp9", &vpx_codec_vp9_cx, NULL, VP9_FOURCC},
79 static void warn_or_exit_on_errorv(vpx_codec_ctx_t *ctx, int fatal,
80 const char *s, va_list ap) {
82 const char *detail = vpx_codec_error_detail(ctx);
84 vfprintf(stderr, s, ap);
85 fprintf(stderr, ": %s\n", vpx_codec_error(ctx));
88 fprintf(stderr, " %s\n", detail);
95 static void ctx_exit_on_error(vpx_codec_ctx_t *ctx, const char *s, ...) {
99 warn_or_exit_on_errorv(ctx, 1, s, ap);
103 static void warn_or_exit_on_error(vpx_codec_ctx_t *ctx, int fatal,
104 const char *s, ...) {
108 warn_or_exit_on_errorv(ctx, fatal, s, ap);
112 int read_frame(struct VpxInputContext *input_ctx, vpx_image_t *img) {
113 FILE *f = input_ctx->file;
114 y4m_input *y4m = &input_ctx->y4m;
117 if (input_ctx->file_type == FILE_TYPE_Y4M) {
118 if (y4m_input_fetch_frame(y4m, f, img) < 1)
121 shortread = read_yuv_frame(input_ctx, img);
127 int file_is_y4m(FILE *infile, y4m_input *y4m, const char detect[4]) {
128 if (memcmp(detect, "YUV4", 4) == 0) {
135 /* Murmur hash derived from public domain reference implementation at
136 * http:// sites.google.com/site/murmurhash/
138 static unsigned int murmur(const void *key, int len, unsigned int seed) {
139 const unsigned int m = 0x5bd1e995;
142 unsigned int h = seed ^ len;
144 const unsigned char *data = (const unsigned char *)key;
149 k = (unsigned int)data[0];
150 k |= (unsigned int)data[1] << 8;
151 k |= (unsigned int)data[2] << 16;
152 k |= (unsigned int)data[3] << 24;
183 static const arg_def_t debugmode = ARG_DEF("D", "debug", 0,
184 "Debug mode (makes output deterministic)");
185 static const arg_def_t outputfile = ARG_DEF("o", "output", 1,
187 static const arg_def_t use_yv12 = ARG_DEF(NULL, "yv12", 0,
188 "Input file is YV12 ");
189 static const arg_def_t use_i420 = ARG_DEF(NULL, "i420", 0,
190 "Input file is I420 (default)");
191 static const arg_def_t codecarg = ARG_DEF(NULL, "codec", 1,
193 static const arg_def_t passes = ARG_DEF("p", "passes", 1,
194 "Number of passes (1/2)");
195 static const arg_def_t pass_arg = ARG_DEF(NULL, "pass", 1,
196 "Pass to execute (1/2)");
197 static const arg_def_t fpf_name = ARG_DEF(NULL, "fpf", 1,
198 "First pass statistics file name");
199 static const arg_def_t limit = ARG_DEF(NULL, "limit", 1,
200 "Stop encoding after n input frames");
201 static const arg_def_t skip = ARG_DEF(NULL, "skip", 1,
202 "Skip the first n input frames");
203 static const arg_def_t deadline = ARG_DEF("d", "deadline", 1,
204 "Deadline per frame (usec)");
205 static const arg_def_t best_dl = ARG_DEF(NULL, "best", 0,
206 "Use Best Quality Deadline");
207 static const arg_def_t good_dl = ARG_DEF(NULL, "good", 0,
208 "Use Good Quality Deadline");
209 static const arg_def_t rt_dl = ARG_DEF(NULL, "rt", 0,
210 "Use Realtime Quality Deadline");
211 static const arg_def_t quietarg = ARG_DEF("q", "quiet", 0,
212 "Do not print encode progress");
213 static const arg_def_t verbosearg = ARG_DEF("v", "verbose", 0,
214 "Show encoder parameters");
215 static const arg_def_t psnrarg = ARG_DEF(NULL, "psnr", 0,
216 "Show PSNR in status line");
217 enum TestDecodeFatality {
222 static const struct arg_enum_list test_decode_enum[] = {
223 {"off", TEST_DECODE_OFF},
224 {"fatal", TEST_DECODE_FATAL},
225 {"warn", TEST_DECODE_WARN},
228 static const arg_def_t recontest = ARG_DEF_ENUM(NULL, "test-decode", 1,
229 "Test encode/decode mismatch",
231 static const arg_def_t framerate = ARG_DEF(NULL, "fps", 1,
232 "Stream frame rate (rate/scale)");
233 static const arg_def_t use_ivf = ARG_DEF(NULL, "ivf", 0,
234 "Output IVF (default is WebM)");
235 static const arg_def_t out_part = ARG_DEF("P", "output-partitions", 0,
236 "Makes encoder output partitions. Requires IVF output!");
237 static const arg_def_t q_hist_n = ARG_DEF(NULL, "q-hist", 1,
238 "Show quantizer histogram (n-buckets)");
239 static const arg_def_t rate_hist_n = ARG_DEF(NULL, "rate-hist", 1,
240 "Show rate histogram (n-buckets)");
241 static const arg_def_t *main_args[] = {
243 &outputfile, &codecarg, &passes, &pass_arg, &fpf_name, &limit, &skip,
244 &deadline, &best_dl, &good_dl, &rt_dl,
245 &quietarg, &verbosearg, &psnrarg, &use_ivf, &out_part, &q_hist_n, &rate_hist_n,
249 static const arg_def_t usage = ARG_DEF("u", "usage", 1,
250 "Usage profile number to use");
251 static const arg_def_t threads = ARG_DEF("t", "threads", 1,
252 "Max number of threads to use");
253 static const arg_def_t profile = ARG_DEF(NULL, "profile", 1,
254 "Bitstream profile number to use");
255 static const arg_def_t width = ARG_DEF("w", "width", 1,
257 static const arg_def_t height = ARG_DEF("h", "height", 1,
259 static const struct arg_enum_list stereo_mode_enum[] = {
260 {"mono", STEREO_FORMAT_MONO},
261 {"left-right", STEREO_FORMAT_LEFT_RIGHT},
262 {"bottom-top", STEREO_FORMAT_BOTTOM_TOP},
263 {"top-bottom", STEREO_FORMAT_TOP_BOTTOM},
264 {"right-left", STEREO_FORMAT_RIGHT_LEFT},
267 static const arg_def_t stereo_mode = ARG_DEF_ENUM(NULL, "stereo-mode", 1,
268 "Stereo 3D video format", stereo_mode_enum);
269 static const arg_def_t timebase = ARG_DEF(NULL, "timebase", 1,
270 "Output timestamp precision (fractional seconds)");
271 static const arg_def_t error_resilient = ARG_DEF(NULL, "error-resilient", 1,
272 "Enable error resiliency features");
273 static const arg_def_t lag_in_frames = ARG_DEF(NULL, "lag-in-frames", 1,
274 "Max number of frames to lag");
276 static const arg_def_t *global_args[] = {
277 &use_yv12, &use_i420, &usage, &threads, &profile,
278 &width, &height, &stereo_mode, &timebase, &framerate,
283 static const arg_def_t dropframe_thresh = ARG_DEF(NULL, "drop-frame", 1,
284 "Temporal resampling threshold (buf %)");
285 static const arg_def_t resize_allowed = ARG_DEF(NULL, "resize-allowed", 1,
286 "Spatial resampling enabled (bool)");
287 static const arg_def_t resize_up_thresh = ARG_DEF(NULL, "resize-up", 1,
288 "Upscale threshold (buf %)");
289 static const arg_def_t resize_down_thresh = ARG_DEF(NULL, "resize-down", 1,
290 "Downscale threshold (buf %)");
291 static const struct arg_enum_list end_usage_enum[] = {
298 static const arg_def_t end_usage = ARG_DEF_ENUM(NULL, "end-usage", 1,
299 "Rate control mode", end_usage_enum);
300 static const arg_def_t target_bitrate = ARG_DEF(NULL, "target-bitrate", 1,
302 static const arg_def_t min_quantizer = ARG_DEF(NULL, "min-q", 1,
303 "Minimum (best) quantizer");
304 static const arg_def_t max_quantizer = ARG_DEF(NULL, "max-q", 1,
305 "Maximum (worst) quantizer");
306 static const arg_def_t undershoot_pct = ARG_DEF(NULL, "undershoot-pct", 1,
307 "Datarate undershoot (min) target (%)");
308 static const arg_def_t overshoot_pct = ARG_DEF(NULL, "overshoot-pct", 1,
309 "Datarate overshoot (max) target (%)");
310 static const arg_def_t buf_sz = ARG_DEF(NULL, "buf-sz", 1,
311 "Client buffer size (ms)");
312 static const arg_def_t buf_initial_sz = ARG_DEF(NULL, "buf-initial-sz", 1,
313 "Client initial buffer size (ms)");
314 static const arg_def_t buf_optimal_sz = ARG_DEF(NULL, "buf-optimal-sz", 1,
315 "Client optimal buffer size (ms)");
316 static const arg_def_t *rc_args[] = {
317 &dropframe_thresh, &resize_allowed, &resize_up_thresh, &resize_down_thresh,
318 &end_usage, &target_bitrate, &min_quantizer, &max_quantizer,
319 &undershoot_pct, &overshoot_pct, &buf_sz, &buf_initial_sz, &buf_optimal_sz,
324 static const arg_def_t bias_pct = ARG_DEF(NULL, "bias-pct", 1,
325 "CBR/VBR bias (0=CBR, 100=VBR)");
326 static const arg_def_t minsection_pct = ARG_DEF(NULL, "minsection-pct", 1,
327 "GOP min bitrate (% of target)");
328 static const arg_def_t maxsection_pct = ARG_DEF(NULL, "maxsection-pct", 1,
329 "GOP max bitrate (% of target)");
330 static const arg_def_t *rc_twopass_args[] = {
331 &bias_pct, &minsection_pct, &maxsection_pct, NULL
335 static const arg_def_t kf_min_dist = ARG_DEF(NULL, "kf-min-dist", 1,
336 "Minimum keyframe interval (frames)");
337 static const arg_def_t kf_max_dist = ARG_DEF(NULL, "kf-max-dist", 1,
338 "Maximum keyframe interval (frames)");
339 static const arg_def_t kf_disabled = ARG_DEF(NULL, "disable-kf", 0,
340 "Disable keyframe placement");
341 static const arg_def_t *kf_args[] = {
342 &kf_min_dist, &kf_max_dist, &kf_disabled, NULL
346 static const arg_def_t noise_sens = ARG_DEF(NULL, "noise-sensitivity", 1,
347 "Noise sensitivity (frames to blur)");
348 static const arg_def_t sharpness = ARG_DEF(NULL, "sharpness", 1,
349 "Filter sharpness (0-7)");
350 static const arg_def_t static_thresh = ARG_DEF(NULL, "static-thresh", 1,
351 "Motion detection threshold");
352 static const arg_def_t cpu_used = ARG_DEF(NULL, "cpu-used", 1,
353 "CPU Used (-16..16)");
354 static const arg_def_t token_parts = ARG_DEF(NULL, "token-parts", 1,
355 "Number of token partitions to use, log2");
356 static const arg_def_t tile_cols = ARG_DEF(NULL, "tile-columns", 1,
357 "Number of tile columns to use, log2");
358 static const arg_def_t tile_rows = ARG_DEF(NULL, "tile-rows", 1,
359 "Number of tile rows to use, log2");
360 static const arg_def_t auto_altref = ARG_DEF(NULL, "auto-alt-ref", 1,
361 "Enable automatic alt reference frames");
362 static const arg_def_t arnr_maxframes = ARG_DEF(NULL, "arnr-maxframes", 1,
363 "AltRef Max Frames");
364 static const arg_def_t arnr_strength = ARG_DEF(NULL, "arnr-strength", 1,
366 static const arg_def_t arnr_type = ARG_DEF(NULL, "arnr-type", 1,
368 static const struct arg_enum_list tuning_enum[] = {
369 {"psnr", VP8_TUNE_PSNR},
370 {"ssim", VP8_TUNE_SSIM},
373 static const arg_def_t tune_ssim = ARG_DEF_ENUM(NULL, "tune", 1,
374 "Material to favor", tuning_enum);
375 static const arg_def_t cq_level = ARG_DEF(NULL, "cq-level", 1,
376 "Constant/Constrained Quality level");
377 static const arg_def_t max_intra_rate_pct = ARG_DEF(NULL, "max-intra-rate", 1,
378 "Max I-frame bitrate (pct)");
379 static const arg_def_t lossless = ARG_DEF(NULL, "lossless", 1, "Lossless mode");
380 #if CONFIG_VP9_ENCODER
381 static const arg_def_t frame_parallel_decoding = ARG_DEF(
382 NULL, "frame-parallel", 1, "Enable frame parallel decodability features");
383 static const arg_def_t aq_mode = ARG_DEF(
385 "Adaptive quantization mode (0: disabled (by default), 1: variance based)");
388 #if CONFIG_VP8_ENCODER
389 static const arg_def_t *vp8_args[] = {
390 &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh,
391 &token_parts, &arnr_maxframes, &arnr_strength, &arnr_type,
392 &tune_ssim, &cq_level, &max_intra_rate_pct,
395 static const int vp8_arg_ctrl_map[] = {
396 VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF,
397 VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD,
398 VP8E_SET_TOKEN_PARTITIONS,
399 VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH, VP8E_SET_ARNR_TYPE,
400 VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, VP8E_SET_MAX_INTRA_BITRATE_PCT,
405 #if CONFIG_VP9_ENCODER
406 static const arg_def_t *vp9_args[] = {
407 &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh,
408 &tile_cols, &tile_rows, &arnr_maxframes, &arnr_strength, &arnr_type,
409 &tune_ssim, &cq_level, &max_intra_rate_pct, &lossless,
410 &frame_parallel_decoding, &aq_mode,
413 static const int vp9_arg_ctrl_map[] = {
414 VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF,
415 VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD,
416 VP9E_SET_TILE_COLUMNS, VP9E_SET_TILE_ROWS,
417 VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH, VP8E_SET_ARNR_TYPE,
418 VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, VP8E_SET_MAX_INTRA_BITRATE_PCT,
419 VP9E_SET_LOSSLESS, VP9E_SET_FRAME_PARALLEL_DECODING, VP9E_SET_AQ_MODE,
424 static const arg_def_t *no_args[] = { NULL };
429 fprintf(stderr, "Usage: %s <options> -o dst_filename src_filename \n",
432 fprintf(stderr, "\nOptions:\n");
433 arg_show_usage(stderr, main_args);
434 fprintf(stderr, "\nEncoder Global Options:\n");
435 arg_show_usage(stderr, global_args);
436 fprintf(stderr, "\nRate Control Options:\n");
437 arg_show_usage(stderr, rc_args);
438 fprintf(stderr, "\nTwopass Rate Control Options:\n");
439 arg_show_usage(stderr, rc_twopass_args);
440 fprintf(stderr, "\nKeyframe Placement Options:\n");
441 arg_show_usage(stderr, kf_args);
442 #if CONFIG_VP8_ENCODER
443 fprintf(stderr, "\nVP8 Specific Options:\n");
444 arg_show_usage(stderr, vp8_args);
446 #if CONFIG_VP9_ENCODER
447 fprintf(stderr, "\nVP9 Specific Options:\n");
448 arg_show_usage(stderr, vp9_args);
450 fprintf(stderr, "\nStream timebase (--timebase):\n"
451 " The desired precision of timestamps in the output, expressed\n"
452 " in fractional seconds. Default is 1/1000.\n");
454 "Included encoders:\n"
457 for (i = 0; i < sizeof(codecs) / sizeof(codecs[0]); i++)
458 fprintf(stderr, " %-6s - %s\n",
460 vpx_codec_iface_name(codecs[i].iface()));
466 #define HIST_BAR_MAX 40
468 int low, high, count;
472 static int merge_hist_buckets(struct hist_bucket *bucket,
475 int small_bucket = 0, merge_bucket = INT_MAX, big_bucket = 0;
476 int buckets = *buckets_;
479 /* Find the extrema for this list of buckets */
480 big_bucket = small_bucket = 0;
481 for (i = 0; i < buckets; i++) {
482 if (bucket[i].count < bucket[small_bucket].count)
484 if (bucket[i].count > bucket[big_bucket].count)
488 /* If we have too many buckets, merge the smallest with an adjacent
491 while (buckets > max_buckets) {
492 int last_bucket = buckets - 1;
494 /* merge the small bucket with an adjacent one. */
495 if (small_bucket == 0)
497 else if (small_bucket == last_bucket)
498 merge_bucket = last_bucket - 1;
499 else if (bucket[small_bucket - 1].count < bucket[small_bucket + 1].count)
500 merge_bucket = small_bucket - 1;
502 merge_bucket = small_bucket + 1;
504 assert(abs(merge_bucket - small_bucket) <= 1);
505 assert(small_bucket < buckets);
506 assert(big_bucket < buckets);
507 assert(merge_bucket < buckets);
509 if (merge_bucket < small_bucket) {
510 bucket[merge_bucket].high = bucket[small_bucket].high;
511 bucket[merge_bucket].count += bucket[small_bucket].count;
513 bucket[small_bucket].high = bucket[merge_bucket].high;
514 bucket[small_bucket].count += bucket[merge_bucket].count;
515 merge_bucket = small_bucket;
518 assert(bucket[merge_bucket].low != bucket[merge_bucket].high);
522 /* Remove the merge_bucket from the list, and find the new small
523 * and big buckets while we're at it
525 big_bucket = small_bucket = 0;
526 for (i = 0; i < buckets; i++) {
527 if (i > merge_bucket)
528 bucket[i] = bucket[i + 1];
530 if (bucket[i].count < bucket[small_bucket].count)
532 if (bucket[i].count > bucket[big_bucket].count)
539 return bucket[big_bucket].count;
543 static void show_histogram(const struct hist_bucket *bucket,
547 const char *pat1, *pat2;
550 switch ((int)(log(bucket[buckets - 1].high) / log(10)) + 1) {
577 pat1 = "%12d %10s: ";
578 pat2 = "%12d-%10d: ";
582 for (i = 0; i < buckets; i++) {
587 pct = (float)(100.0 * bucket[i].count / total);
588 len = HIST_BAR_MAX * bucket[i].count / scale;
591 assert(len <= HIST_BAR_MAX);
593 if (bucket[i].low == bucket[i].high)
594 fprintf(stderr, pat1, bucket[i].low, "");
596 fprintf(stderr, pat2, bucket[i].low, bucket[i].high);
598 for (j = 0; j < HIST_BAR_MAX; j++)
599 fprintf(stderr, j < len ? "=" : " ");
600 fprintf(stderr, "\t%5d (%6.2f%%)\n", bucket[i].count, pct);
605 static void show_q_histogram(const int counts[64], int max_buckets) {
606 struct hist_bucket bucket[64];
613 for (i = 0; i < 64; i++) {
615 bucket[buckets].low = bucket[buckets].high = i;
616 bucket[buckets].count = counts[i];
622 fprintf(stderr, "\nQuantizer Selection:\n");
623 scale = merge_hist_buckets(bucket, &buckets, max_buckets);
624 show_histogram(bucket, buckets, total, scale);
628 #define RATE_BINS (100)
634 struct hist_bucket bucket[RATE_BINS];
639 static void init_rate_histogram(struct rate_hist *hist,
640 const vpx_codec_enc_cfg_t *cfg,
641 const vpx_rational_t *fps) {
644 /* Determine the number of samples in the buffer. Use the file's framerate
645 * to determine the number of frames in rc_buf_sz milliseconds, with an
646 * adjustment (5/4) to account for alt-refs
648 hist->samples = cfg->rc_buf_sz * 5 / 4 * fps->num / fps->den / 1000;
650 /* prevent division by zero */
651 if (hist->samples == 0)
654 hist->pts = calloc(hist->samples, sizeof(*hist->pts));
655 hist->sz = calloc(hist->samples, sizeof(*hist->sz));
656 for (i = 0; i < RATE_BINS; i++) {
657 hist->bucket[i].low = INT_MAX;
658 hist->bucket[i].high = 0;
659 hist->bucket[i].count = 0;
664 static void destroy_rate_histogram(struct rate_hist *hist) {
670 static void update_rate_histogram(struct rate_hist *hist,
671 const vpx_codec_enc_cfg_t *cfg,
672 const vpx_codec_cx_pkt_t *pkt) {
674 int64_t now, then, sum_sz = 0, avg_bitrate;
676 now = pkt->data.frame.pts * 1000
677 * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
679 idx = hist->frames++ % hist->samples;
680 hist->pts[idx] = now;
681 hist->sz[idx] = (int)pkt->data.frame.sz;
683 if (now < cfg->rc_buf_initial_sz)
688 /* Sum the size over the past rc_buf_sz ms */
689 for (i = hist->frames; i > 0 && hist->frames - i < hist->samples; i--) {
690 int i_idx = (i - 1) % hist->samples;
692 then = hist->pts[i_idx];
693 if (now - then > cfg->rc_buf_sz)
695 sum_sz += hist->sz[i_idx];
701 avg_bitrate = sum_sz * 8 * 1000 / (now - then);
702 idx = (int)(avg_bitrate * (RATE_BINS / 2) / (cfg->rc_target_bitrate * 1000));
705 if (idx > RATE_BINS - 1)
707 if (hist->bucket[idx].low > avg_bitrate)
708 hist->bucket[idx].low = (int)avg_bitrate;
709 if (hist->bucket[idx].high < avg_bitrate)
710 hist->bucket[idx].high = (int)avg_bitrate;
711 hist->bucket[idx].count++;
716 static void show_rate_histogram(struct rate_hist *hist,
717 const vpx_codec_enc_cfg_t *cfg,
722 for (i = 0; i < RATE_BINS; i++) {
723 if (hist->bucket[i].low == INT_MAX)
725 hist->bucket[buckets++] = hist->bucket[i];
728 fprintf(stderr, "\nRate (over %dms window):\n", cfg->rc_buf_sz);
729 scale = merge_hist_buckets(hist->bucket, &buckets, max_buckets);
730 show_histogram(hist->bucket, buckets, hist->total, scale);
733 #define mmin(a, b) ((a) < (b) ? (a) : (b))
734 static void find_mismatch(vpx_image_t *img1, vpx_image_t *img2,
735 int yloc[4], int uloc[4], int vloc[4]) {
736 const unsigned int bsize = 64;
737 const unsigned int bsizey = bsize >> img1->y_chroma_shift;
738 const unsigned int bsizex = bsize >> img1->x_chroma_shift;
739 const int c_w = (img1->d_w + img1->x_chroma_shift) >> img1->x_chroma_shift;
740 const int c_h = (img1->d_h + img1->y_chroma_shift) >> img1->y_chroma_shift;
741 unsigned int match = 1;
743 yloc[0] = yloc[1] = yloc[2] = yloc[3] = -1;
744 for (i = 0, match = 1; match && i < img1->d_h; i += bsize) {
745 for (j = 0; match && j < img1->d_w; j += bsize) {
747 int si = mmin(i + bsize, img1->d_h) - i;
748 int sj = mmin(j + bsize, img1->d_w) - j;
749 for (k = 0; match && k < si; k++)
750 for (l = 0; match && l < sj; l++) {
751 if (*(img1->planes[VPX_PLANE_Y] +
752 (i + k) * img1->stride[VPX_PLANE_Y] + j + l) !=
753 *(img2->planes[VPX_PLANE_Y] +
754 (i + k) * img2->stride[VPX_PLANE_Y] + j + l)) {
757 yloc[2] = *(img1->planes[VPX_PLANE_Y] +
758 (i + k) * img1->stride[VPX_PLANE_Y] + j + l);
759 yloc[3] = *(img2->planes[VPX_PLANE_Y] +
760 (i + k) * img2->stride[VPX_PLANE_Y] + j + l);
768 uloc[0] = uloc[1] = uloc[2] = uloc[3] = -1;
769 for (i = 0, match = 1; match && i < c_h; i += bsizey) {
770 for (j = 0; match && j < c_w; j += bsizex) {
772 int si = mmin(i + bsizey, c_h - i);
773 int sj = mmin(j + bsizex, c_w - j);
774 for (k = 0; match && k < si; k++)
775 for (l = 0; match && l < sj; l++) {
776 if (*(img1->planes[VPX_PLANE_U] +
777 (i + k) * img1->stride[VPX_PLANE_U] + j + l) !=
778 *(img2->planes[VPX_PLANE_U] +
779 (i + k) * img2->stride[VPX_PLANE_U] + j + l)) {
782 uloc[2] = *(img1->planes[VPX_PLANE_U] +
783 (i + k) * img1->stride[VPX_PLANE_U] + j + l);
784 uloc[3] = *(img2->planes[VPX_PLANE_U] +
785 (i + k) * img2->stride[VPX_PLANE_V] + j + l);
792 vloc[0] = vloc[1] = vloc[2] = vloc[3] = -1;
793 for (i = 0, match = 1; match && i < c_h; i += bsizey) {
794 for (j = 0; match && j < c_w; j += bsizex) {
796 int si = mmin(i + bsizey, c_h - i);
797 int sj = mmin(j + bsizex, c_w - j);
798 for (k = 0; match && k < si; k++)
799 for (l = 0; match && l < sj; l++) {
800 if (*(img1->planes[VPX_PLANE_V] +
801 (i + k) * img1->stride[VPX_PLANE_V] + j + l) !=
802 *(img2->planes[VPX_PLANE_V] +
803 (i + k) * img2->stride[VPX_PLANE_V] + j + l)) {
806 vloc[2] = *(img1->planes[VPX_PLANE_V] +
807 (i + k) * img1->stride[VPX_PLANE_V] + j + l);
808 vloc[3] = *(img2->planes[VPX_PLANE_V] +
809 (i + k) * img2->stride[VPX_PLANE_V] + j + l);
818 static int compare_img(vpx_image_t *img1, vpx_image_t *img2)
820 const int c_w = (img1->d_w + img1->x_chroma_shift) >> img1->x_chroma_shift;
821 const int c_h = (img1->d_h + img1->y_chroma_shift) >> img1->y_chroma_shift;
825 match &= (img1->fmt == img2->fmt);
826 match &= (img1->w == img2->w);
827 match &= (img1->h == img2->h);
829 for (i = 0; i < img1->d_h; i++)
830 match &= (memcmp(img1->planes[VPX_PLANE_Y]+i*img1->stride[VPX_PLANE_Y],
831 img2->planes[VPX_PLANE_Y]+i*img2->stride[VPX_PLANE_Y],
834 for (i = 0; i < c_h; i++)
835 match &= (memcmp(img1->planes[VPX_PLANE_U]+i*img1->stride[VPX_PLANE_U],
836 img2->planes[VPX_PLANE_U]+i*img2->stride[VPX_PLANE_U],
839 for (i = 0; i < c_h; i++)
840 match &= (memcmp(img1->planes[VPX_PLANE_V]+i*img1->stride[VPX_PLANE_U],
841 img2->planes[VPX_PLANE_V]+i*img2->stride[VPX_PLANE_U],
848 #define NELEMENTS(x) (sizeof(x)/sizeof(x[0]))
849 #define MAX(x,y) ((x)>(y)?(x):(y))
850 #if CONFIG_VP8_ENCODER && !CONFIG_VP9_ENCODER
851 #define ARG_CTRL_CNT_MAX NELEMENTS(vp8_arg_ctrl_map)
852 #elif !CONFIG_VP8_ENCODER && CONFIG_VP9_ENCODER
853 #define ARG_CTRL_CNT_MAX NELEMENTS(vp9_arg_ctrl_map)
855 #define ARG_CTRL_CNT_MAX MAX(NELEMENTS(vp8_arg_ctrl_map), \
856 NELEMENTS(vp9_arg_ctrl_map))
859 /* Configuration elements common to all streams */
860 struct global_config {
861 const struct codec_item *codec;
872 enum TestDecodeFatality test_decode;
874 struct vpx_rational framerate;
877 int show_q_hist_buckets;
878 int show_rate_hist_buckets;
882 /* Per-stream configuration */
883 struct stream_config {
884 struct vpx_codec_enc_cfg cfg;
886 const char *stats_fn;
887 stereo_format_t stereo_fmt;
888 int arg_ctrls[ARG_CTRL_CNT_MAX][2];
891 int have_kf_max_dist;
895 struct stream_state {
897 struct stream_state *next;
898 struct stream_config config;
900 struct rate_hist rate_hist;
901 struct EbmlGlobal ebml;
903 uint64_t psnr_sse_total;
904 uint64_t psnr_samples_total;
905 double psnr_totals[4];
908 vpx_codec_ctx_t encoder;
909 unsigned int frames_out;
913 struct vpx_image *img;
914 vpx_codec_ctx_t decoder;
919 void validate_positive_rational(const char *msg,
920 struct vpx_rational *rat) {
927 die("Error: %s must be positive\n", msg);
930 die("Error: %s has zero denominator\n", msg);
934 static void parse_global_config(struct global_config *global, char **argv) {
938 /* Initialize default parameters */
939 memset(global, 0, sizeof(*global));
940 global->codec = codecs;
942 global->use_i420 = 1;
943 /* Assign default deadline to good quality */
944 global->deadline = VPX_DL_GOOD_QUALITY;
946 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) {
949 if (arg_match(&arg, &codecarg, argi)) {
952 for (j = 0; j < sizeof(codecs) / sizeof(codecs[0]); j++)
953 if (!strcmp(codecs[j].name, arg.val))
957 global->codec = codecs + k;
959 die("Error: Unrecognized argument (%s) to --codec\n",
962 } else if (arg_match(&arg, &passes, argi)) {
963 global->passes = arg_parse_uint(&arg);
965 if (global->passes < 1 || global->passes > 2)
966 die("Error: Invalid number of passes (%d)\n", global->passes);
967 } else if (arg_match(&arg, &pass_arg, argi)) {
968 global->pass = arg_parse_uint(&arg);
970 if (global->pass < 1 || global->pass > 2)
971 die("Error: Invalid pass selected (%d)\n",
973 } else if (arg_match(&arg, &usage, argi))
974 global->usage = arg_parse_uint(&arg);
975 else if (arg_match(&arg, &deadline, argi))
976 global->deadline = arg_parse_uint(&arg);
977 else if (arg_match(&arg, &best_dl, argi))
978 global->deadline = VPX_DL_BEST_QUALITY;
979 else if (arg_match(&arg, &good_dl, argi))
980 global->deadline = VPX_DL_GOOD_QUALITY;
981 else if (arg_match(&arg, &rt_dl, argi))
982 global->deadline = VPX_DL_REALTIME;
983 else if (arg_match(&arg, &use_yv12, argi))
984 global->use_i420 = 0;
985 else if (arg_match(&arg, &use_i420, argi))
986 global->use_i420 = 1;
987 else if (arg_match(&arg, &quietarg, argi))
989 else if (arg_match(&arg, &verbosearg, argi))
991 else if (arg_match(&arg, &limit, argi))
992 global->limit = arg_parse_uint(&arg);
993 else if (arg_match(&arg, &skip, argi))
994 global->skip_frames = arg_parse_uint(&arg);
995 else if (arg_match(&arg, &psnrarg, argi))
996 global->show_psnr = 1;
997 else if (arg_match(&arg, &recontest, argi))
998 global->test_decode = arg_parse_enum_or_int(&arg);
999 else if (arg_match(&arg, &framerate, argi)) {
1000 global->framerate = arg_parse_rational(&arg);
1001 validate_positive_rational(arg.name, &global->framerate);
1002 global->have_framerate = 1;
1003 } else if (arg_match(&arg, &out_part, argi))
1004 global->out_part = 1;
1005 else if (arg_match(&arg, &debugmode, argi))
1007 else if (arg_match(&arg, &q_hist_n, argi))
1008 global->show_q_hist_buckets = arg_parse_uint(&arg);
1009 else if (arg_match(&arg, &rate_hist_n, argi))
1010 global->show_rate_hist_buckets = arg_parse_uint(&arg);
1015 /* Validate global config */
1016 if (global->passes == 0) {
1017 #if CONFIG_VP9_ENCODER
1018 // Make default VP9 passes = 2 until there is a better quality 1-pass
1020 global->passes = (global->codec->iface == vpx_codec_vp9_cx ? 2 : 1);
1027 /* DWIM: Assume the user meant passes=2 if pass=2 is specified */
1028 if (global->pass > global->passes) {
1029 warn("Assuming --pass=%d implies --passes=%d\n",
1030 global->pass, global->pass);
1031 global->passes = global->pass;
1037 void open_input_file(struct VpxInputContext *input) {
1038 /* Parse certain options from the input file, if possible */
1039 input->file = strcmp(input->filename, "-")
1040 ? fopen(input->filename, "rb") : set_binary_mode(stdin);
1043 fatal("Failed to open input file");
1045 if (!fseeko(input->file, 0, SEEK_END)) {
1046 /* Input file is seekable. Figure out how long it is, so we can get
1049 input->length = ftello(input->file);
1050 rewind(input->file);
1053 /* For RAW input sources, these bytes will applied on the first frame
1056 input->detect.buf_read = fread(input->detect.buf, 1, 4, input->file);
1057 input->detect.position = 0;
1059 if (input->detect.buf_read == 4
1060 && file_is_y4m(input->file, &input->y4m, input->detect.buf)) {
1061 if (y4m_input_open(&input->y4m, input->file, input->detect.buf, 4,
1062 input->only_i420) >= 0) {
1063 input->file_type = FILE_TYPE_Y4M;
1064 input->width = input->y4m.pic_w;
1065 input->height = input->y4m.pic_h;
1066 input->framerate.numerator = input->y4m.fps_n;
1067 input->framerate.denominator = input->y4m.fps_d;
1068 input->use_i420 = 0;
1070 fatal("Unsupported Y4M stream.");
1071 } else if (input->detect.buf_read == 4 && file_is_ivf(input)) {
1072 fatal("IVF is not supported as input.");
1074 input->file_type = FILE_TYPE_RAW;
1079 static void close_input_file(struct VpxInputContext *input) {
1080 fclose(input->file);
1081 if (input->file_type == FILE_TYPE_Y4M)
1082 y4m_input_close(&input->y4m);
1085 static struct stream_state *new_stream(struct global_config *global,
1086 struct stream_state *prev) {
1087 struct stream_state *stream;
1089 stream = calloc(1, sizeof(*stream));
1091 fatal("Failed to allocate new stream.");
1093 memcpy(stream, prev, sizeof(*stream));
1095 prev->next = stream;
1097 vpx_codec_err_t res;
1099 /* Populate encoder configuration */
1100 res = vpx_codec_enc_config_default(global->codec->iface(),
1101 &stream->config.cfg,
1104 fatal("Failed to get config: %s\n", vpx_codec_err_to_string(res));
1106 /* Change the default timebase to a high enough value so that the
1107 * encoder will always create strictly increasing timestamps.
1109 stream->config.cfg.g_timebase.den = 1000;
1111 /* Never use the library's default resolution, require it be parsed
1112 * from the file or set on the command line.
1114 stream->config.cfg.g_w = 0;
1115 stream->config.cfg.g_h = 0;
1117 /* Initialize remaining stream parameters */
1118 stream->config.stereo_fmt = STEREO_FORMAT_MONO;
1119 stream->config.write_webm = 1;
1120 stream->ebml.last_pts_ms = -1;
1122 /* Allows removal of the application version from the EBML tags */
1123 stream->ebml.debug = global->debug;
1126 /* Output files must be specified for each stream */
1127 stream->config.out_fn = NULL;
1129 stream->next = NULL;
1134 static int parse_stream_params(struct global_config *global,
1135 struct stream_state *stream,
1137 char **argi, **argj;
1139 static const arg_def_t **ctrl_args = no_args;
1140 static const int *ctrl_args_map = NULL;
1141 struct stream_config *config = &stream->config;
1142 int eos_mark_found = 0;
1144 /* Handle codec specific options */
1146 #if CONFIG_VP8_ENCODER
1147 } else if (global->codec->iface == vpx_codec_vp8_cx) {
1148 ctrl_args = vp8_args;
1149 ctrl_args_map = vp8_arg_ctrl_map;
1151 #if CONFIG_VP9_ENCODER
1152 } else if (global->codec->iface == vpx_codec_vp9_cx) {
1153 ctrl_args = vp9_args;
1154 ctrl_args_map = vp9_arg_ctrl_map;
1158 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) {
1161 /* Once we've found an end-of-stream marker (--) we want to continue
1162 * shifting arguments but not consuming them.
1164 if (eos_mark_found) {
1167 } else if (!strcmp(*argj, "--")) {
1173 else if (arg_match(&arg, &outputfile, argi))
1174 config->out_fn = arg.val;
1175 else if (arg_match(&arg, &fpf_name, argi))
1176 config->stats_fn = arg.val;
1177 else if (arg_match(&arg, &use_ivf, argi))
1178 config->write_webm = 0;
1179 else if (arg_match(&arg, &threads, argi))
1180 config->cfg.g_threads = arg_parse_uint(&arg);
1181 else if (arg_match(&arg, &profile, argi))
1182 config->cfg.g_profile = arg_parse_uint(&arg);
1183 else if (arg_match(&arg, &width, argi))
1184 config->cfg.g_w = arg_parse_uint(&arg);
1185 else if (arg_match(&arg, &height, argi))
1186 config->cfg.g_h = arg_parse_uint(&arg);
1187 else if (arg_match(&arg, &stereo_mode, argi))
1188 config->stereo_fmt = arg_parse_enum_or_int(&arg);
1189 else if (arg_match(&arg, &timebase, argi)) {
1190 config->cfg.g_timebase = arg_parse_rational(&arg);
1191 validate_positive_rational(arg.name, &config->cfg.g_timebase);
1192 } else if (arg_match(&arg, &error_resilient, argi))
1193 config->cfg.g_error_resilient = arg_parse_uint(&arg);
1194 else if (arg_match(&arg, &lag_in_frames, argi))
1195 config->cfg.g_lag_in_frames = arg_parse_uint(&arg);
1196 else if (arg_match(&arg, &dropframe_thresh, argi))
1197 config->cfg.rc_dropframe_thresh = arg_parse_uint(&arg);
1198 else if (arg_match(&arg, &resize_allowed, argi))
1199 config->cfg.rc_resize_allowed = arg_parse_uint(&arg);
1200 else if (arg_match(&arg, &resize_up_thresh, argi))
1201 config->cfg.rc_resize_up_thresh = arg_parse_uint(&arg);
1202 else if (arg_match(&arg, &resize_down_thresh, argi))
1203 config->cfg.rc_resize_down_thresh = arg_parse_uint(&arg);
1204 else if (arg_match(&arg, &end_usage, argi))
1205 config->cfg.rc_end_usage = arg_parse_enum_or_int(&arg);
1206 else if (arg_match(&arg, &target_bitrate, argi))
1207 config->cfg.rc_target_bitrate = arg_parse_uint(&arg);
1208 else if (arg_match(&arg, &min_quantizer, argi))
1209 config->cfg.rc_min_quantizer = arg_parse_uint(&arg);
1210 else if (arg_match(&arg, &max_quantizer, argi))
1211 config->cfg.rc_max_quantizer = arg_parse_uint(&arg);
1212 else if (arg_match(&arg, &undershoot_pct, argi))
1213 config->cfg.rc_undershoot_pct = arg_parse_uint(&arg);
1214 else if (arg_match(&arg, &overshoot_pct, argi))
1215 config->cfg.rc_overshoot_pct = arg_parse_uint(&arg);
1216 else if (arg_match(&arg, &buf_sz, argi))
1217 config->cfg.rc_buf_sz = arg_parse_uint(&arg);
1218 else if (arg_match(&arg, &buf_initial_sz, argi))
1219 config->cfg.rc_buf_initial_sz = arg_parse_uint(&arg);
1220 else if (arg_match(&arg, &buf_optimal_sz, argi))
1221 config->cfg.rc_buf_optimal_sz = arg_parse_uint(&arg);
1222 else if (arg_match(&arg, &bias_pct, argi)) {
1223 config->cfg.rc_2pass_vbr_bias_pct = arg_parse_uint(&arg);
1225 if (global->passes < 2)
1226 warn("option %s ignored in one-pass mode.\n", arg.name);
1227 } else if (arg_match(&arg, &minsection_pct, argi)) {
1228 config->cfg.rc_2pass_vbr_minsection_pct = arg_parse_uint(&arg);
1230 if (global->passes < 2)
1231 warn("option %s ignored in one-pass mode.\n", arg.name);
1232 } else if (arg_match(&arg, &maxsection_pct, argi)) {
1233 config->cfg.rc_2pass_vbr_maxsection_pct = arg_parse_uint(&arg);
1235 if (global->passes < 2)
1236 warn("option %s ignored in one-pass mode.\n", arg.name);
1237 } else if (arg_match(&arg, &kf_min_dist, argi))
1238 config->cfg.kf_min_dist = arg_parse_uint(&arg);
1239 else if (arg_match(&arg, &kf_max_dist, argi)) {
1240 config->cfg.kf_max_dist = arg_parse_uint(&arg);
1241 config->have_kf_max_dist = 1;
1242 } else if (arg_match(&arg, &kf_disabled, argi))
1243 config->cfg.kf_mode = VPX_KF_DISABLED;
1247 for (i = 0; ctrl_args[i]; i++) {
1248 if (arg_match(&arg, ctrl_args[i], argi)) {
1252 /* Point either to the next free element or the first
1253 * instance of this control.
1255 for (j = 0; j < config->arg_ctrl_cnt; j++)
1256 if (config->arg_ctrls[j][0] == ctrl_args_map[i])
1260 assert(j < ARG_CTRL_CNT_MAX);
1261 if (j < ARG_CTRL_CNT_MAX) {
1262 config->arg_ctrls[j][0] = ctrl_args_map[i];
1263 config->arg_ctrls[j][1] = arg_parse_enum_or_int(&arg);
1264 if (j == config->arg_ctrl_cnt)
1265 config->arg_ctrl_cnt++;
1276 return eos_mark_found;
1280 #define FOREACH_STREAM(func)\
1283 struct stream_state *stream;\
1285 for(stream = streams; stream; stream = stream->next)\
1290 static void validate_stream_config(struct stream_state *stream) {
1291 struct stream_state *streami;
1293 if (!stream->config.cfg.g_w || !stream->config.cfg.g_h)
1294 fatal("Stream %d: Specify stream dimensions with --width (-w) "
1295 " and --height (-h)", stream->index);
1297 for (streami = stream; streami; streami = streami->next) {
1298 /* All streams require output files */
1299 if (!streami->config.out_fn)
1300 fatal("Stream %d: Output file is required (specify with -o)",
1303 /* Check for two streams outputting to the same file */
1304 if (streami != stream) {
1305 const char *a = stream->config.out_fn;
1306 const char *b = streami->config.out_fn;
1307 if (!strcmp(a, b) && strcmp(a, "/dev/null") && strcmp(a, ":nul"))
1308 fatal("Stream %d: duplicate output file (from stream %d)",
1309 streami->index, stream->index);
1312 /* Check for two streams sharing a stats file. */
1313 if (streami != stream) {
1314 const char *a = stream->config.stats_fn;
1315 const char *b = streami->config.stats_fn;
1316 if (a && b && !strcmp(a, b))
1317 fatal("Stream %d: duplicate stats file (from stream %d)",
1318 streami->index, stream->index);
1324 static void set_stream_dimensions(struct stream_state *stream,
1327 if (!stream->config.cfg.g_w) {
1328 if (!stream->config.cfg.g_h)
1329 stream->config.cfg.g_w = w;
1331 stream->config.cfg.g_w = w * stream->config.cfg.g_h / h;
1333 if (!stream->config.cfg.g_h) {
1334 stream->config.cfg.g_h = h * stream->config.cfg.g_w / w;
1339 static void set_default_kf_interval(struct stream_state *stream,
1340 struct global_config *global) {
1341 /* Use a max keyframe interval of 5 seconds, if none was
1342 * specified on the command line.
1344 if (!stream->config.have_kf_max_dist) {
1345 double framerate = (double)global->framerate.num / global->framerate.den;
1346 if (framerate > 0.0)
1347 stream->config.cfg.kf_max_dist = (unsigned int)(5.0 * framerate);
1352 static void show_stream_config(struct stream_state *stream,
1353 struct global_config *global,
1354 struct VpxInputContext *input) {
1356 #define SHOW(field) \
1357 fprintf(stderr, " %-28s = %d\n", #field, stream->config.cfg.field)
1359 if (stream->index == 0) {
1360 fprintf(stderr, "Codec: %s\n",
1361 vpx_codec_iface_name(global->codec->iface()));
1362 fprintf(stderr, "Source file: %s Format: %s\n", input->filename,
1363 input->use_i420 ? "I420" : "YV12");
1365 if (stream->next || stream->index)
1366 fprintf(stderr, "\nStream Index: %d\n", stream->index);
1367 fprintf(stderr, "Destination file: %s\n", stream->config.out_fn);
1368 fprintf(stderr, "Encoder parameters:\n");
1375 SHOW(g_timebase.num);
1376 SHOW(g_timebase.den);
1377 SHOW(g_error_resilient);
1379 SHOW(g_lag_in_frames);
1380 SHOW(rc_dropframe_thresh);
1381 SHOW(rc_resize_allowed);
1382 SHOW(rc_resize_up_thresh);
1383 SHOW(rc_resize_down_thresh);
1385 SHOW(rc_target_bitrate);
1386 SHOW(rc_min_quantizer);
1387 SHOW(rc_max_quantizer);
1388 SHOW(rc_undershoot_pct);
1389 SHOW(rc_overshoot_pct);
1391 SHOW(rc_buf_initial_sz);
1392 SHOW(rc_buf_optimal_sz);
1393 SHOW(rc_2pass_vbr_bias_pct);
1394 SHOW(rc_2pass_vbr_minsection_pct);
1395 SHOW(rc_2pass_vbr_maxsection_pct);
1402 static void open_output_file(struct stream_state *stream,
1403 struct global_config *global) {
1404 const char *fn = stream->config.out_fn;
1406 stream->file = strcmp(fn, "-") ? fopen(fn, "wb") : set_binary_mode(stdout);
1409 fatal("Failed to open output file");
1411 if (stream->config.write_webm && fseek(stream->file, 0, SEEK_CUR))
1412 fatal("WebM output to pipes not supported.");
1414 if (stream->config.write_webm) {
1415 stream->ebml.stream = stream->file;
1416 write_webm_file_header(&stream->ebml, &stream->config.cfg,
1418 stream->config.stereo_fmt,
1419 global->codec->fourcc);
1421 ivf_write_file_header(stream->file, &stream->config.cfg,
1422 global->codec->fourcc, 0);
1426 static void close_output_file(struct stream_state *stream,
1427 unsigned int fourcc) {
1428 if (stream->config.write_webm) {
1429 write_webm_file_footer(&stream->ebml, stream->hash);
1430 free(stream->ebml.cue_list);
1431 stream->ebml.cue_list = NULL;
1433 if (!fseek(stream->file, 0, SEEK_SET))
1434 ivf_write_file_header(stream->file, &stream->config.cfg,
1436 stream->frames_out);
1439 fclose(stream->file);
1443 static void setup_pass(struct stream_state *stream,
1444 struct global_config *global,
1446 if (stream->config.stats_fn) {
1447 if (!stats_open_file(&stream->stats, stream->config.stats_fn,
1449 fatal("Failed to open statistics store");
1451 if (!stats_open_mem(&stream->stats, pass))
1452 fatal("Failed to open statistics store");
1455 stream->config.cfg.g_pass = global->passes == 2
1456 ? pass ? VPX_RC_LAST_PASS : VPX_RC_FIRST_PASS
1459 stream->config.cfg.rc_twopass_stats_in = stats_get(&stream->stats);
1461 stream->cx_time = 0;
1463 stream->frames_out = 0;
1467 static void initialize_encoder(struct stream_state *stream,
1468 struct global_config *global) {
1472 flags |= global->show_psnr ? VPX_CODEC_USE_PSNR : 0;
1473 flags |= global->out_part ? VPX_CODEC_USE_OUTPUT_PARTITION : 0;
1475 /* Construct Encoder Context */
1476 vpx_codec_enc_init(&stream->encoder, global->codec->iface(),
1477 &stream->config.cfg, flags);
1478 ctx_exit_on_error(&stream->encoder, "Failed to initialize encoder");
1480 /* Note that we bypass the vpx_codec_control wrapper macro because
1481 * we're being clever to store the control IDs in an array. Real
1482 * applications will want to make use of the enumerations directly
1484 for (i = 0; i < stream->config.arg_ctrl_cnt; i++) {
1485 int ctrl = stream->config.arg_ctrls[i][0];
1486 int value = stream->config.arg_ctrls[i][1];
1487 if (vpx_codec_control_(&stream->encoder, ctrl, value))
1488 fprintf(stderr, "Error: Tried to set control %d = %d\n",
1491 ctx_exit_on_error(&stream->encoder, "Failed to control codec");
1495 if (global->test_decode != TEST_DECODE_OFF) {
1496 vpx_codec_dec_init(&stream->decoder, global->codec->dx_iface(), NULL, 0);
1502 static void encode_frame(struct stream_state *stream,
1503 struct global_config *global,
1504 struct vpx_image *img,
1505 unsigned int frames_in) {
1506 vpx_codec_pts_t frame_start, next_frame_start;
1507 struct vpx_codec_enc_cfg *cfg = &stream->config.cfg;
1508 struct vpx_usec_timer timer;
1510 frame_start = (cfg->g_timebase.den * (int64_t)(frames_in - 1)
1511 * global->framerate.den)
1512 / cfg->g_timebase.num / global->framerate.num;
1513 next_frame_start = (cfg->g_timebase.den * (int64_t)(frames_in)
1514 * global->framerate.den)
1515 / cfg->g_timebase.num / global->framerate.num;
1517 /* Scale if necessary */
1518 if (img && (img->d_w != cfg->g_w || img->d_h != cfg->g_h)) {
1520 stream->img = vpx_img_alloc(NULL, VPX_IMG_FMT_I420,
1521 cfg->g_w, cfg->g_h, 16);
1522 I420Scale(img->planes[VPX_PLANE_Y], img->stride[VPX_PLANE_Y],
1523 img->planes[VPX_PLANE_U], img->stride[VPX_PLANE_U],
1524 img->planes[VPX_PLANE_V], img->stride[VPX_PLANE_V],
1526 stream->img->planes[VPX_PLANE_Y],
1527 stream->img->stride[VPX_PLANE_Y],
1528 stream->img->planes[VPX_PLANE_U],
1529 stream->img->stride[VPX_PLANE_U],
1530 stream->img->planes[VPX_PLANE_V],
1531 stream->img->stride[VPX_PLANE_V],
1532 stream->img->d_w, stream->img->d_h,
1538 vpx_usec_timer_start(&timer);
1539 vpx_codec_encode(&stream->encoder, img, frame_start,
1540 (unsigned long)(next_frame_start - frame_start),
1541 0, global->deadline);
1542 vpx_usec_timer_mark(&timer);
1543 stream->cx_time += vpx_usec_timer_elapsed(&timer);
1544 ctx_exit_on_error(&stream->encoder, "Stream %d: Failed to encode frame",
1549 static void update_quantizer_histogram(struct stream_state *stream) {
1550 if (stream->config.cfg.g_pass != VPX_RC_FIRST_PASS) {
1553 vpx_codec_control(&stream->encoder, VP8E_GET_LAST_QUANTIZER_64, &q);
1554 ctx_exit_on_error(&stream->encoder, "Failed to read quantizer");
1555 stream->counts[q]++;
1560 static void get_cx_data(struct stream_state *stream,
1561 struct global_config *global,
1563 const vpx_codec_cx_pkt_t *pkt;
1564 const struct vpx_codec_enc_cfg *cfg = &stream->config.cfg;
1565 vpx_codec_iter_t iter = NULL;
1568 while ((pkt = vpx_codec_get_cx_data(&stream->encoder, &iter))) {
1569 static size_t fsize = 0;
1570 static off_t ivf_header_pos = 0;
1572 switch (pkt->kind) {
1573 case VPX_CODEC_CX_FRAME_PKT:
1574 if (!(pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT)) {
1575 stream->frames_out++;
1578 fprintf(stderr, " %6luF", (unsigned long)pkt->data.frame.sz);
1580 update_rate_histogram(&stream->rate_hist, cfg, pkt);
1581 if (stream->config.write_webm) {
1582 /* Update the hash */
1583 if (!stream->ebml.debug)
1584 stream->hash = murmur(pkt->data.frame.buf,
1585 (int)pkt->data.frame.sz,
1588 write_webm_block(&stream->ebml, cfg, pkt);
1590 if (pkt->data.frame.partition_id <= 0) {
1591 ivf_header_pos = ftello(stream->file);
1592 fsize = pkt->data.frame.sz;
1594 ivf_write_frame_header(stream->file, pkt);
1596 fsize += pkt->data.frame.sz;
1598 if (!(pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT)) {
1599 off_t currpos = ftello(stream->file);
1600 fseeko(stream->file, ivf_header_pos, SEEK_SET);
1601 ivf_write_frame_size(stream->file, fsize);
1602 fseeko(stream->file, currpos, SEEK_SET);
1606 (void) fwrite(pkt->data.frame.buf, 1, pkt->data.frame.sz,
1609 stream->nbytes += pkt->data.raw.sz;
1613 if (global->test_decode != TEST_DECODE_OFF && !stream->mismatch_seen) {
1614 vpx_codec_decode(&stream->decoder, pkt->data.frame.buf,
1615 pkt->data.frame.sz, NULL, 0);
1616 if (stream->decoder.err) {
1617 warn_or_exit_on_error(&stream->decoder,
1618 global->test_decode == TEST_DECODE_FATAL,
1619 "Failed to decode frame %d in stream %d",
1620 stream->frames_out + 1, stream->index);
1621 stream->mismatch_seen = stream->frames_out + 1;
1626 case VPX_CODEC_STATS_PKT:
1627 stream->frames_out++;
1628 stats_write(&stream->stats,
1629 pkt->data.twopass_stats.buf,
1630 pkt->data.twopass_stats.sz);
1631 stream->nbytes += pkt->data.raw.sz;
1633 case VPX_CODEC_PSNR_PKT:
1635 if (global->show_psnr) {
1638 stream->psnr_sse_total += pkt->data.psnr.sse[0];
1639 stream->psnr_samples_total += pkt->data.psnr.samples[0];
1640 for (i = 0; i < 4; i++) {
1642 fprintf(stderr, "%.3f ", pkt->data.psnr.psnr[i]);
1643 stream->psnr_totals[i] += pkt->data.psnr.psnr[i];
1645 stream->psnr_count++;
1656 static void show_psnr(struct stream_state *stream) {
1660 if (!stream->psnr_count)
1663 fprintf(stderr, "Stream %d PSNR (Overall/Avg/Y/U/V)", stream->index);
1664 ovpsnr = vp8_mse2psnr((double)stream->psnr_samples_total, 255.0,
1665 (double)stream->psnr_sse_total);
1666 fprintf(stderr, " %.3f", ovpsnr);
1668 for (i = 0; i < 4; i++) {
1669 fprintf(stderr, " %.3f", stream->psnr_totals[i] / stream->psnr_count);
1671 fprintf(stderr, "\n");
1675 static float usec_to_fps(uint64_t usec, unsigned int frames) {
1676 return (float)(usec > 0 ? frames * 1000000.0 / (float)usec : 0);
1680 static void test_decode(struct stream_state *stream,
1681 enum TestDecodeFatality fatal,
1682 const struct codec_item *codec) {
1683 vpx_image_t enc_img, dec_img;
1685 if (stream->mismatch_seen)
1688 /* Get the internal reference frame */
1689 if (codec->fourcc == VP8_FOURCC) {
1690 struct vpx_ref_frame ref_enc, ref_dec;
1693 width = (stream->config.cfg.g_w + 15) & ~15;
1694 height = (stream->config.cfg.g_h + 15) & ~15;
1695 vpx_img_alloc(&ref_enc.img, VPX_IMG_FMT_I420, width, height, 1);
1696 enc_img = ref_enc.img;
1697 vpx_img_alloc(&ref_dec.img, VPX_IMG_FMT_I420, width, height, 1);
1698 dec_img = ref_dec.img;
1700 ref_enc.frame_type = VP8_LAST_FRAME;
1701 ref_dec.frame_type = VP8_LAST_FRAME;
1702 vpx_codec_control(&stream->encoder, VP8_COPY_REFERENCE, &ref_enc);
1703 vpx_codec_control(&stream->decoder, VP8_COPY_REFERENCE, &ref_dec);
1705 struct vp9_ref_frame ref;
1708 vpx_codec_control(&stream->encoder, VP9_GET_REFERENCE, &ref);
1710 vpx_codec_control(&stream->decoder, VP9_GET_REFERENCE, &ref);
1713 ctx_exit_on_error(&stream->encoder, "Failed to get encoder reference frame");
1714 ctx_exit_on_error(&stream->decoder, "Failed to get decoder reference frame");
1716 if (!compare_img(&enc_img, &dec_img)) {
1717 int y[4], u[4], v[4];
1718 find_mismatch(&enc_img, &dec_img, y, u, v);
1719 stream->decoder.err = 1;
1720 warn_or_exit_on_error(&stream->decoder, fatal == TEST_DECODE_FATAL,
1721 "Stream %d: Encode/decode mismatch on frame %d at"
1722 " Y[%d, %d] {%d/%d},"
1723 " U[%d, %d] {%d/%d},"
1724 " V[%d, %d] {%d/%d}",
1725 stream->index, stream->frames_out,
1726 y[0], y[1], y[2], y[3],
1727 u[0], u[1], u[2], u[3],
1728 v[0], v[1], v[2], v[3]);
1729 stream->mismatch_seen = stream->frames_out;
1732 vpx_img_free(&enc_img);
1733 vpx_img_free(&dec_img);
1737 static void print_time(const char *label, int64_t etl) {
1738 int hours, mins, secs;
1742 etl -= hours * 3600;
1747 fprintf(stderr, "[%3s %2d:%02d:%02d] ",
1748 label, hours, mins, secs);
1750 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 global_config 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);
1786 /* Now parse each stream's parameters. Using a local scope here
1787 * due to the use of 'stream' as loop variable in FOREACH_STREAM
1790 struct stream_state *stream = NULL;
1793 stream = new_stream(&global, stream);
1797 } while (parse_stream_params(&global, stream, argv));
1800 /* Check for unrecognized options */
1801 for (argi = argv; *argi; argi++)
1802 if (argi[0][0] == '-' && argi[0][1])
1803 die("Error: Unrecognized option %s\n", *argi);
1805 /* Handle non-option arguments */
1806 input.filename = argv[0];
1808 if (!input.filename)
1812 /* Decide if other chroma subsamplings than 4:2:0 are supported */
1813 if (global.codec->fourcc == VP9_FOURCC)
1814 input.only_i420 = 0;
1817 for (pass = global.pass ? global.pass - 1 : 0; pass < global.passes; pass++) {
1818 int frames_in = 0, seen_frames = 0;
1819 int64_t estimated_time_left = -1;
1820 int64_t average_rate = -1;
1821 off_t lagged_count = 0;
1823 open_input_file(&input);
1825 /* If the input file doesn't specify its w/h (raw files), try to get
1826 * the data from the first stream's configuration.
1828 if (!input.width || !input.height)
1830 if (stream->config.cfg.g_w && stream->config.cfg.g_h) {
1831 input.width = stream->config.cfg.g_w;
1832 input.height = stream->config.cfg.g_h;
1837 /* Update stream configurations from the input file's parameters */
1838 if (!input.width || !input.height)
1839 fatal("Specify stream dimensions with --width (-w) "
1840 " and --height (-h)");
1841 FOREACH_STREAM(set_stream_dimensions(stream, input.width, input.height));
1842 FOREACH_STREAM(validate_stream_config(stream));
1844 /* Ensure that --passes and --pass are consistent. If --pass is set and
1845 * --passes=2, ensure --fpf was set.
1847 if (global.pass && global.passes == 2)
1849 if (!stream->config.stats_fn)
1850 die("Stream %d: Must specify --fpf when --pass=%d"
1851 " and --passes=2\n", stream->index, global.pass);
1854 /* Use the frame rate from the file only if none was specified
1855 * on the command-line.
1857 if (!global.have_framerate) {
1858 global.framerate.num = input.framerate.numerator;
1859 global.framerate.den = input.framerate.denominator;
1862 FOREACH_STREAM(set_default_kf_interval(stream, &global));
1864 /* Show configuration */
1865 if (global.verbose && pass == 0)
1866 FOREACH_STREAM(show_stream_config(stream, &global, &input));
1868 if (pass == (global.pass ? global.pass - 1 : 0)) {
1869 if (input.file_type == FILE_TYPE_Y4M)
1870 /*The Y4M reader does its own allocation.
1871 Just initialize this here to avoid problems if we never read any
1873 memset(&raw, 0, sizeof(raw));
1876 input.use_i420 ? VPX_IMG_FMT_I420
1878 input.width, input.height, 32);
1880 FOREACH_STREAM(init_rate_histogram(&stream->rate_hist,
1881 &stream->config.cfg,
1882 &global.framerate));
1885 FOREACH_STREAM(setup_pass(stream, &global, pass));
1886 FOREACH_STREAM(open_output_file(stream, &global));
1887 FOREACH_STREAM(initialize_encoder(stream, &global));
1892 while (frame_avail || got_data) {
1893 struct vpx_usec_timer timer;
1895 if (!global.limit || frames_in < global.limit) {
1896 frame_avail = read_frame(&input, &raw);
1900 seen_frames = frames_in > global.skip_frames ?
1901 frames_in - global.skip_frames : 0;
1903 if (!global.quiet) {
1904 float fps = usec_to_fps(cx_time, seen_frames);
1905 fprintf(stderr, "\rPass %d/%d ", pass + 1, global.passes);
1907 if (stream_cnt == 1)
1909 "frame %4d/%-4d %7"PRId64"B ",
1910 frames_in, streams->frames_out, (int64_t)streams->nbytes);
1912 fprintf(stderr, "frame %4d ", frames_in);
1914 fprintf(stderr, "%7"PRId64" %s %.2f %s ",
1915 cx_time > 9999999 ? cx_time / 1000 : cx_time,
1916 cx_time > 9999999 ? "ms" : "us",
1917 fps >= 1.0 ? fps : fps * 60,
1918 fps >= 1.0 ? "fps" : "fpm");
1919 print_time("ETA", estimated_time_left);
1920 fprintf(stderr, "\033[K");
1926 if (frames_in > global.skip_frames) {
1927 vpx_usec_timer_start(&timer);
1928 FOREACH_STREAM(encode_frame(stream, &global,
1929 frame_avail ? &raw : NULL,
1931 vpx_usec_timer_mark(&timer);
1932 cx_time += vpx_usec_timer_elapsed(&timer);
1934 FOREACH_STREAM(update_quantizer_histogram(stream));
1937 FOREACH_STREAM(get_cx_data(stream, &global, &got_data));
1939 if (!got_data && input.length && !streams->frames_out) {
1940 lagged_count = global.limit ? seen_frames : ftello(input.file);
1941 } else if (input.length) {
1946 int frame_in_lagged = (seen_frames - lagged_count) * 1000;
1948 rate = cx_time ? frame_in_lagged * (int64_t)1000000 / cx_time : 0;
1949 remaining = 1000 * (global.limit - global.skip_frames
1950 - seen_frames + lagged_count);
1952 off_t input_pos = ftello(input.file);
1953 off_t input_pos_lagged = input_pos - lagged_count;
1954 int64_t limit = input.length;
1956 rate = cx_time ? input_pos_lagged * (int64_t)1000000 / cx_time : 0;
1957 remaining = limit - input_pos + lagged_count;
1960 average_rate = (average_rate <= 0)
1962 : (average_rate * 7 + rate) / 8;
1963 estimated_time_left = average_rate ? remaining / average_rate : -1;
1966 if (got_data && global.test_decode != TEST_DECODE_OFF)
1967 FOREACH_STREAM(test_decode(stream, global.test_decode, global.codec));
1974 fprintf(stderr, "\n");
1977 FOREACH_STREAM(fprintf(
1979 "\rPass %d/%d frame %4d/%-4d %7"PRId64"B %7lub/f %7"PRId64"b/s"
1980 " %7"PRId64" %s (%.2f fps)\033[K\n", pass + 1,
1981 global.passes, frames_in, stream->frames_out, (int64_t)stream->nbytes,
1982 seen_frames ? (unsigned long)(stream->nbytes * 8 / seen_frames) : 0,
1983 seen_frames ? (int64_t)stream->nbytes * 8
1984 * (int64_t)global.framerate.num / global.framerate.den
1987 stream->cx_time > 9999999 ? stream->cx_time / 1000 : stream->cx_time,
1988 stream->cx_time > 9999999 ? "ms" : "us",
1989 usec_to_fps(stream->cx_time, seen_frames));
1992 if (global.show_psnr)
1993 FOREACH_STREAM(show_psnr(stream));
1995 FOREACH_STREAM(vpx_codec_destroy(&stream->encoder));
1997 if (global.test_decode != TEST_DECODE_OFF) {
1998 FOREACH_STREAM(vpx_codec_destroy(&stream->decoder));
2001 close_input_file(&input);
2003 if (global.test_decode == TEST_DECODE_FATAL) {
2004 FOREACH_STREAM(res |= stream->mismatch_seen);
2006 FOREACH_STREAM(close_output_file(stream, global.codec->fourcc));
2008 FOREACH_STREAM(stats_close(&stream->stats, global.passes - 1));
2014 if (global.show_q_hist_buckets)
2015 FOREACH_STREAM(show_q_histogram(stream->counts,
2016 global.show_q_hist_buckets));
2018 if (global.show_rate_hist_buckets)
2019 FOREACH_STREAM(show_rate_histogram(&stream->rate_hist,
2020 &stream->config.cfg,
2021 global.show_rate_hist_buckets));
2022 FOREACH_STREAM(destroy_rate_histogram(&stream->rate_hist));
2024 #if CONFIG_INTERNAL_STATS
2025 /* TODO(jkoleszar): This doesn't belong in this executable. Do it for now,
2026 * to match some existing utilities.
2028 if (!(global.pass == 1 && global.passes == 2))
2030 FILE *f = fopen("opsnr.stt", "a");
2031 if (stream->mismatch_seen) {
2032 fprintf(f, "First mismatch occurred in frame %d\n",
2033 stream->mismatch_seen);
2035 fprintf(f, "No mismatch detected in recon buffers\n");
2044 return res ? EXIT_FAILURE : EXIT_SUCCESS;