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 /* This is a simple program that encodes YV12 files and generates ivf
13 * files using the new interface.
16 #define USE_POSIX_MMAP 0
18 #define USE_POSIX_MMAP 1
26 #include "vpx/vpx_encoder.h"
28 #include <sys/types.h>
34 #include "vpx_version.h"
35 #include "vpx/vp8cx.h"
36 #include "vpx_ports/mem_ops.h"
37 #include "vpx_ports/vpx_timer.h"
39 #include "libmkv/EbmlWriter.h"
40 #include "libmkv/EbmlIDs.h"
42 /* Need special handling of these functions on Windows */
44 /* MSVS doesn't define off_t, and uses _f{seek,tell}i64 */
45 typedef __int64 off_t;
46 #define fseeko _fseeki64
47 #define ftello _ftelli64
49 /* MinGW defines off_t, and uses f{seek,tell}o64 */
50 #define fseeko fseeko64
51 #define ftello ftello64
55 #define LITERALU64(n) n
57 #define LITERALU64(n) n##LLU
60 static const char *exec_name;
62 static const struct codec_item
65 const vpx_codec_iface_t *iface;
69 #if CONFIG_VP8_ENCODER
70 {"vp8", &vpx_codec_vp8_cx_algo, 0x30385056},
74 static void usage_exit();
76 void die(const char *fmt, ...)
80 vfprintf(stderr, fmt, ap);
81 fprintf(stderr, "\n");
85 static void ctx_exit_on_error(vpx_codec_ctx_t *ctx, const char *s)
89 const char *detail = vpx_codec_error_detail(ctx);
91 fprintf(stderr, "%s: %s\n", s, vpx_codec_error(ctx));
94 fprintf(stderr, " %s\n", detail);
100 /* This structure is used to abstract the different ways of handling
101 * first pass statistics.
112 int stats_open_file(stats_io_t *stats, const char *fpf, int pass)
120 stats->file = fopen(fpf, "wb");
122 stats->buf.buf = NULL,
123 res = (stats->file != NULL);
129 struct stat stat_buf;
132 fd = open(fpf, O_RDONLY);
133 stats->file = fdopen(fd, "rb");
134 fstat(fd, &stat_buf);
135 stats->buf.sz = stat_buf.st_size;
136 stats->buf.buf = mmap(NULL, stats->buf.sz, PROT_READ, MAP_PRIVATE,
138 res = (stats->buf.buf != NULL);
142 stats->file = fopen(fpf, "rb");
144 if (fseek(stats->file, 0, SEEK_END))
146 fprintf(stderr, "First-pass stats file must be seekable!\n");
150 stats->buf.sz = stats->buf_alloc_sz = ftell(stats->file);
153 stats->buf.buf = malloc(stats->buf_alloc_sz);
157 fprintf(stderr, "Failed to allocate first-pass stats buffer (%d bytes)\n",
158 stats->buf_alloc_sz);
162 nbytes = fread(stats->buf.buf, 1, stats->buf.sz, stats->file);
163 res = (nbytes == stats->buf.sz);
170 int stats_open_mem(stats_io_t *stats, int pass)
178 stats->buf_alloc_sz = 64 * 1024;
179 stats->buf.buf = malloc(stats->buf_alloc_sz);
182 stats->buf_ptr = stats->buf.buf;
183 res = (stats->buf.buf != NULL);
188 void stats_close(stats_io_t *stats)
192 if (stats->pass == 1)
196 munmap(stats->buf.buf, stats->buf.sz);
198 free(stats->buf.buf);
207 if (stats->pass == 1)
208 free(stats->buf.buf);
212 void stats_write(stats_io_t *stats, const void *pkt, size_t len)
216 if(fwrite(pkt, 1, len, stats->file));
220 if (stats->buf.sz + len > stats->buf_alloc_sz)
222 size_t new_sz = stats->buf_alloc_sz + 64 * 1024;
223 char *new_ptr = realloc(stats->buf.buf, new_sz);
227 stats->buf_ptr = new_ptr + (stats->buf_ptr - (char *)stats->buf.buf);
228 stats->buf.buf = new_ptr;
229 stats->buf_alloc_sz = new_sz;
233 memcpy(stats->buf_ptr, pkt, len);
234 stats->buf.sz += len;
235 stats->buf_ptr += len;
239 vpx_fixed_buf_t stats_get(stats_io_t *stats)
251 struct detect_buffer {
257 #define IVF_FRAME_HDR_SZ (4+8) /* 4 byte size + 8 byte timestamp */
258 static int read_frame(FILE *f, vpx_image_t *img, unsigned int file_type,
259 y4m_input *y4m, struct detect_buffer *detect)
264 if (file_type == FILE_TYPE_Y4M)
266 if (y4m_input_fetch_frame(y4m, f, img) < 1)
271 if (file_type == FILE_TYPE_IVF)
273 char junk[IVF_FRAME_HDR_SZ];
275 /* Skip the frame header. We know how big the frame should be. See
276 * write_ivf_frame_header() for documentation on the frame header
279 if(fread(junk, 1, IVF_FRAME_HDR_SZ, f));
282 for (plane = 0; plane < 3; plane++)
285 int w = (plane ? (1 + img->d_w) / 2 : img->d_w);
286 int h = (plane ? (1 + img->d_h) / 2 : img->d_h);
289 /* Determine the correct plane based on the image format. The for-loop
290 * always counts in Y,U,V order, but this may not match the order of
296 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U];
299 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V];
302 ptr = img->planes[plane];
305 for (r = 0; r < h; r++)
309 memcpy(ptr, detect->buf, 4);
310 shortread |= fread(ptr+4, 1, w-4, f) < w-4;
314 shortread |= fread(ptr, 1, w, f) < w;
316 ptr += img->stride[plane];
325 unsigned int file_is_y4m(FILE *infile,
329 if(memcmp(detect, "YUV4", 4) == 0)
336 #define IVF_FILE_HDR_SZ (32)
337 unsigned int file_is_ivf(FILE *infile,
338 unsigned int *fourcc,
340 unsigned int *height,
343 char raw_hdr[IVF_FILE_HDR_SZ];
346 if(memcmp(detect, "DKIF", 4) != 0)
349 /* See write_ivf_file_header() for more documentation on the file header
352 if (fread(raw_hdr + 4, 1, IVF_FILE_HDR_SZ - 4, infile)
353 == IVF_FILE_HDR_SZ - 4)
358 if (mem_get_le16(raw_hdr + 4) != 0)
359 fprintf(stderr, "Error: Unrecognized IVF version! This file may not"
360 " decode properly.");
362 *fourcc = mem_get_le32(raw_hdr + 8);
368 *width = mem_get_le16(raw_hdr + 12);
369 *height = mem_get_le16(raw_hdr + 14);
376 static void write_ivf_file_header(FILE *outfile,
377 const vpx_codec_enc_cfg_t *cfg,
383 if (cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS)
390 mem_put_le16(header + 4, 0); /* version */
391 mem_put_le16(header + 6, 32); /* headersize */
392 mem_put_le32(header + 8, fourcc); /* headersize */
393 mem_put_le16(header + 12, cfg->g_w); /* width */
394 mem_put_le16(header + 14, cfg->g_h); /* height */
395 mem_put_le32(header + 16, cfg->g_timebase.den); /* rate */
396 mem_put_le32(header + 20, cfg->g_timebase.num); /* scale */
397 mem_put_le32(header + 24, frame_cnt); /* length */
398 mem_put_le32(header + 28, 0); /* unused */
400 if(fwrite(header, 1, 32, outfile));
404 static void write_ivf_frame_header(FILE *outfile,
405 const vpx_codec_cx_pkt_t *pkt)
410 if (pkt->kind != VPX_CODEC_CX_FRAME_PKT)
413 pts = pkt->data.frame.pts;
414 mem_put_le32(header, pkt->data.frame.sz);
415 mem_put_le32(header + 4, pts & 0xFFFFFFFF);
416 mem_put_le32(header + 8, pts >> 32);
418 if(fwrite(header, 1, 12, outfile));
422 typedef off_t EbmlLoc;
437 uint64_t last_pts_ms;
438 vpx_rational_t framerate;
440 /* These pointers are to the start of an element */
441 off_t position_reference;
443 off_t segment_info_pos;
448 /* This pointer is to a specific element to be serialized */
451 /* These pointers are to the size field of the element */
452 EbmlLoc startSegment;
453 EbmlLoc startCluster;
455 uint32_t cluster_timecode;
458 struct cue_entry *cue_list;
464 void Ebml_Write(EbmlGlobal *glob, const void *buffer_in, unsigned long len)
466 if(fwrite(buffer_in, 1, len, glob->stream));
470 void Ebml_Serialize(EbmlGlobal *glob, const void *buffer_in, unsigned long len)
472 const unsigned char *q = (const unsigned char *)buffer_in + len - 1;
475 Ebml_Write(glob, q--, 1);
479 /* Need a fixed size serializer for the track ID. libmkv provdes a 64 bit
480 * one, but not a 32 bit one.
482 static void Ebml_SerializeUnsigned32(EbmlGlobal *glob, unsigned long class_id, uint64_t ui)
484 unsigned char sizeSerialized = 4 | 0x80;
485 Ebml_WriteID(glob, class_id);
486 Ebml_Serialize(glob, &sizeSerialized, 1);
487 Ebml_Serialize(glob, &ui, 4);
492 Ebml_StartSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc,
493 unsigned long class_id)
495 //todo this is always taking 8 bytes, this may need later optimization
496 //this is a key that says lenght unknown
497 unsigned long long unknownLen = LITERALU64(0x01FFFFFFFFFFFFFF);
499 Ebml_WriteID(glob, class_id);
500 *ebmlLoc = ftello(glob->stream);
501 Ebml_Serialize(glob, &unknownLen, 8);
505 Ebml_EndSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc)
510 /* Save the current stream pointer */
511 pos = ftello(glob->stream);
513 /* Calculate the size of this element */
514 size = pos - *ebmlLoc - 8;
515 size |= LITERALU64(0x0100000000000000);
517 /* Seek back to the beginning of the element and write the new size */
518 fseeko(glob->stream, *ebmlLoc, SEEK_SET);
519 Ebml_Serialize(glob, &size, 8);
521 /* Reset the stream pointer */
522 fseeko(glob->stream, pos, SEEK_SET);
527 write_webm_seek_element(EbmlGlobal *ebml, unsigned long id, off_t pos)
529 uint64_t offset = pos - ebml->position_reference;
531 Ebml_StartSubElement(ebml, &start, Seek);
532 Ebml_SerializeBinary(ebml, SeekID, id);
533 Ebml_SerializeUnsigned64(ebml, SeekPosition, offset);
534 Ebml_EndSubElement(ebml, &start);
539 write_webm_seek_info(EbmlGlobal *ebml)
544 /* Save the current stream pointer */
545 pos = ftello(ebml->stream);
547 if(ebml->seek_info_pos)
548 fseeko(ebml->stream, ebml->seek_info_pos, SEEK_SET);
550 ebml->seek_info_pos = pos;
555 Ebml_StartSubElement(ebml, &start, SeekHead);
556 write_webm_seek_element(ebml, Tracks, ebml->track_pos);
557 write_webm_seek_element(ebml, Cues, ebml->cue_pos);
558 write_webm_seek_element(ebml, Info, ebml->segment_info_pos);
559 Ebml_EndSubElement(ebml, &start);
566 frame_time = (uint64_t)1000 * ebml->framerate.den
567 / ebml->framerate.num;
568 ebml->segment_info_pos = ftello(ebml->stream);
569 Ebml_StartSubElement(ebml, &startInfo, Info);
570 Ebml_SerializeUnsigned(ebml, TimecodeScale, 1000000);
571 Ebml_SerializeFloat(ebml, Segment_Duration,
572 ebml->last_pts_ms + frame_time);
573 Ebml_SerializeString(ebml, 0x4D80,
574 ebml->debug ? "vpxenc" : "vpxenc" VERSION_STRING);
575 Ebml_SerializeString(ebml, 0x5741,
576 ebml->debug ? "vpxenc" : "vpxenc" VERSION_STRING);
577 Ebml_EndSubElement(ebml, &startInfo);
583 write_webm_file_header(EbmlGlobal *glob,
584 const vpx_codec_enc_cfg_t *cfg,
585 const struct vpx_rational *fps)
589 Ebml_StartSubElement(glob, &start, EBML);
590 Ebml_SerializeUnsigned(glob, EBMLVersion, 1);
591 Ebml_SerializeUnsigned(glob, EBMLReadVersion, 1); //EBML Read Version
592 Ebml_SerializeUnsigned(glob, EBMLMaxIDLength, 4); //EBML Max ID Length
593 Ebml_SerializeUnsigned(glob, EBMLMaxSizeLength, 8); //EBML Max Size Length
594 Ebml_SerializeString(glob, DocType, "webm"); //Doc Type
595 Ebml_SerializeUnsigned(glob, DocTypeVersion, 2); //Doc Type Version
596 Ebml_SerializeUnsigned(glob, DocTypeReadVersion, 2); //Doc Type Read Version
597 Ebml_EndSubElement(glob, &start);
600 Ebml_StartSubElement(glob, &glob->startSegment, Segment); //segment
601 glob->position_reference = ftello(glob->stream);
602 glob->framerate = *fps;
603 write_webm_seek_info(glob);
607 glob->track_pos = ftello(glob->stream);
608 Ebml_StartSubElement(glob, &trackStart, Tracks);
610 unsigned int trackNumber = 1;
611 uint64_t trackID = 0;
614 Ebml_StartSubElement(glob, &start, TrackEntry);
615 Ebml_SerializeUnsigned(glob, TrackNumber, trackNumber);
616 glob->track_id_pos = ftello(glob->stream);
617 Ebml_SerializeUnsigned32(glob, TrackUID, trackID);
618 Ebml_SerializeUnsigned(glob, TrackType, 1); //video is always 1
619 Ebml_SerializeString(glob, CodecID, "V_VP8");
621 unsigned int pixelWidth = cfg->g_w;
622 unsigned int pixelHeight = cfg->g_h;
623 float frameRate = (float)fps->num/(float)fps->den;
626 Ebml_StartSubElement(glob, &videoStart, Video);
627 Ebml_SerializeUnsigned(glob, PixelWidth, pixelWidth);
628 Ebml_SerializeUnsigned(glob, PixelHeight, pixelHeight);
629 Ebml_SerializeFloat(glob, FrameRate, frameRate);
630 Ebml_EndSubElement(glob, &videoStart); //Video
632 Ebml_EndSubElement(glob, &start); //Track Entry
634 Ebml_EndSubElement(glob, &trackStart);
636 // segment element is open
642 write_webm_block(EbmlGlobal *glob,
643 const vpx_codec_enc_cfg_t *cfg,
644 const vpx_codec_cx_pkt_t *pkt)
646 unsigned long block_length;
647 unsigned char track_number;
648 unsigned short block_timecode = 0;
651 int start_cluster = 0, is_keyframe;
653 /* Calculate the PTS of this frame in milliseconds */
654 pts_ms = pkt->data.frame.pts * 1000
655 * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
656 if(pts_ms <= glob->last_pts_ms)
657 pts_ms = glob->last_pts_ms + 1;
658 glob->last_pts_ms = pts_ms;
660 /* Calculate the relative time of this block */
661 if(pts_ms - glob->cluster_timecode > SHRT_MAX)
664 block_timecode = pts_ms - glob->cluster_timecode;
666 is_keyframe = (pkt->data.frame.flags & VPX_FRAME_IS_KEY);
667 if(start_cluster || is_keyframe)
669 if(glob->cluster_open)
670 Ebml_EndSubElement(glob, &glob->startCluster);
672 /* Open the new cluster */
674 glob->cluster_open = 1;
675 glob->cluster_timecode = pts_ms;
676 glob->cluster_pos = ftello(glob->stream);
677 Ebml_StartSubElement(glob, &glob->startCluster, Cluster); //cluster
678 Ebml_SerializeUnsigned(glob, Timecode, glob->cluster_timecode);
680 /* Save a cue point if this is a keyframe. */
683 struct cue_entry *cue;
685 glob->cue_list = realloc(glob->cue_list,
686 (glob->cues+1) * sizeof(struct cue_entry));
687 cue = &glob->cue_list[glob->cues];
688 cue->time = glob->cluster_timecode;
689 cue->loc = glob->cluster_pos;
694 /* Write the Simple Block */
695 Ebml_WriteID(glob, SimpleBlock);
697 block_length = pkt->data.frame.sz + 4;
698 block_length |= 0x10000000;
699 Ebml_Serialize(glob, &block_length, 4);
702 track_number |= 0x80;
703 Ebml_Write(glob, &track_number, 1);
705 Ebml_Serialize(glob, &block_timecode, 2);
710 if(pkt->data.frame.flags & VPX_FRAME_IS_INVISIBLE)
712 Ebml_Write(glob, &flags, 1);
714 Ebml_Write(glob, pkt->data.frame.buf, pkt->data.frame.sz);
719 write_webm_file_footer(EbmlGlobal *glob, long hash)
722 if(glob->cluster_open)
723 Ebml_EndSubElement(glob, &glob->startCluster);
729 glob->cue_pos = ftello(glob->stream);
730 Ebml_StartSubElement(glob, &start, Cues);
731 for(i=0; i<glob->cues; i++)
733 struct cue_entry *cue = &glob->cue_list[i];
736 Ebml_StartSubElement(glob, &start, CuePoint);
740 Ebml_SerializeUnsigned(glob, CueTime, cue->time);
742 Ebml_StartSubElement(glob, &start, CueTrackPositions);
743 Ebml_SerializeUnsigned(glob, CueTrack, 1);
744 Ebml_SerializeUnsigned64(glob, CueClusterPosition,
745 cue->loc - glob->position_reference);
746 //Ebml_SerializeUnsigned(glob, CueBlockNumber, cue->blockNumber);
747 Ebml_EndSubElement(glob, &start);
749 Ebml_EndSubElement(glob, &start);
751 Ebml_EndSubElement(glob, &start);
754 Ebml_EndSubElement(glob, &glob->startSegment);
756 /* Patch up the seek info block */
757 write_webm_seek_info(glob);
759 /* Patch up the track id */
760 fseeko(glob->stream, glob->track_id_pos, SEEK_SET);
761 Ebml_SerializeUnsigned32(glob, TrackUID, glob->debug ? 0xDEADBEEF : hash);
763 fseeko(glob->stream, 0, SEEK_END);
767 /* Murmur hash derived from public domain reference implementation at
768 * http://sites.google.com/site/murmurhash/
770 static unsigned int murmur ( const void * key, int len, unsigned int seed )
772 const unsigned int m = 0x5bd1e995;
775 unsigned int h = seed ^ len;
777 const unsigned char * data = (const unsigned char *)key;
801 case 3: h ^= data[2] << 16;
802 case 2: h ^= data[1] << 8;
803 case 1: h ^= data[0];
816 static double vp8_mse2psnr(double Samples, double Peak, double Mse)
820 if ((double)Mse > 0.0)
821 psnr = 10.0 * log10(Peak * Peak * Samples / Mse);
823 psnr = 60; // Limit to prevent / 0
834 static const arg_def_t debugmode = ARG_DEF("D", "debug", 0,
835 "Debug mode (makes output deterministic)");
836 static const arg_def_t outputfile = ARG_DEF("o", "output", 1,
838 static const arg_def_t use_yv12 = ARG_DEF(NULL, "yv12", 0,
839 "Input file is YV12 ");
840 static const arg_def_t use_i420 = ARG_DEF(NULL, "i420", 0,
841 "Input file is I420 (default)");
842 static const arg_def_t codecarg = ARG_DEF(NULL, "codec", 1,
844 static const arg_def_t passes = ARG_DEF("p", "passes", 1,
845 "Number of passes (1/2)");
846 static const arg_def_t pass_arg = ARG_DEF(NULL, "pass", 1,
847 "Pass to execute (1/2)");
848 static const arg_def_t fpf_name = ARG_DEF(NULL, "fpf", 1,
849 "First pass statistics file name");
850 static const arg_def_t limit = ARG_DEF(NULL, "limit", 1,
851 "Stop encoding after n input frames");
852 static const arg_def_t deadline = ARG_DEF("d", "deadline", 1,
853 "Deadline per frame (usec)");
854 static const arg_def_t best_dl = ARG_DEF(NULL, "best", 0,
855 "Use Best Quality Deadline");
856 static const arg_def_t good_dl = ARG_DEF(NULL, "good", 0,
857 "Use Good Quality Deadline");
858 static const arg_def_t rt_dl = ARG_DEF(NULL, "rt", 0,
859 "Use Realtime Quality Deadline");
860 static const arg_def_t verbosearg = ARG_DEF("v", "verbose", 0,
861 "Show encoder parameters");
862 static const arg_def_t psnrarg = ARG_DEF(NULL, "psnr", 0,
863 "Show PSNR in status line");
864 static const arg_def_t framerate = ARG_DEF(NULL, "fps", 1,
865 "Stream frame rate (rate/scale)");
866 static const arg_def_t use_ivf = ARG_DEF(NULL, "ivf", 0,
867 "Output IVF (default is WebM)");
868 static const arg_def_t *main_args[] =
871 &outputfile, &codecarg, &passes, &pass_arg, &fpf_name, &limit, &deadline,
872 &best_dl, &good_dl, &rt_dl,
873 &verbosearg, &psnrarg, &use_ivf, &framerate,
877 static const arg_def_t usage = ARG_DEF("u", "usage", 1,
878 "Usage profile number to use");
879 static const arg_def_t threads = ARG_DEF("t", "threads", 1,
880 "Max number of threads to use");
881 static const arg_def_t profile = ARG_DEF(NULL, "profile", 1,
882 "Bitstream profile number to use");
883 static const arg_def_t width = ARG_DEF("w", "width", 1,
885 static const arg_def_t height = ARG_DEF("h", "height", 1,
887 static const arg_def_t timebase = ARG_DEF(NULL, "timebase", 1,
888 "Stream timebase (frame duration)");
889 static const arg_def_t error_resilient = ARG_DEF(NULL, "error-resilient", 1,
890 "Enable error resiliency features");
891 static const arg_def_t lag_in_frames = ARG_DEF(NULL, "lag-in-frames", 1,
892 "Max number of frames to lag");
894 static const arg_def_t *global_args[] =
896 &use_yv12, &use_i420, &usage, &threads, &profile,
897 &width, &height, &timebase, &framerate, &error_resilient,
901 static const arg_def_t dropframe_thresh = ARG_DEF(NULL, "drop-frame", 1,
902 "Temporal resampling threshold (buf %)");
903 static const arg_def_t resize_allowed = ARG_DEF(NULL, "resize-allowed", 1,
904 "Spatial resampling enabled (bool)");
905 static const arg_def_t resize_up_thresh = ARG_DEF(NULL, "resize-up", 1,
906 "Upscale threshold (buf %)");
907 static const arg_def_t resize_down_thresh = ARG_DEF(NULL, "resize-down", 1,
908 "Downscale threshold (buf %)");
909 static const arg_def_t end_usage = ARG_DEF(NULL, "end-usage", 1,
911 static const arg_def_t target_bitrate = ARG_DEF(NULL, "target-bitrate", 1,
913 static const arg_def_t min_quantizer = ARG_DEF(NULL, "min-q", 1,
914 "Minimum (best) quantizer");
915 static const arg_def_t max_quantizer = ARG_DEF(NULL, "max-q", 1,
916 "Maximum (worst) quantizer");
917 static const arg_def_t undershoot_pct = ARG_DEF(NULL, "undershoot-pct", 1,
918 "Datarate undershoot (min) target (%)");
919 static const arg_def_t overshoot_pct = ARG_DEF(NULL, "overshoot-pct", 1,
920 "Datarate overshoot (max) target (%)");
921 static const arg_def_t buf_sz = ARG_DEF(NULL, "buf-sz", 1,
922 "Client buffer size (ms)");
923 static const arg_def_t buf_initial_sz = ARG_DEF(NULL, "buf-initial-sz", 1,
924 "Client initial buffer size (ms)");
925 static const arg_def_t buf_optimal_sz = ARG_DEF(NULL, "buf-optimal-sz", 1,
926 "Client optimal buffer size (ms)");
927 static const arg_def_t *rc_args[] =
929 &dropframe_thresh, &resize_allowed, &resize_up_thresh, &resize_down_thresh,
930 &end_usage, &target_bitrate, &min_quantizer, &max_quantizer,
931 &undershoot_pct, &overshoot_pct, &buf_sz, &buf_initial_sz, &buf_optimal_sz,
936 static const arg_def_t bias_pct = ARG_DEF(NULL, "bias-pct", 1,
937 "CBR/VBR bias (0=CBR, 100=VBR)");
938 static const arg_def_t minsection_pct = ARG_DEF(NULL, "minsection-pct", 1,
939 "GOP min bitrate (% of target)");
940 static const arg_def_t maxsection_pct = ARG_DEF(NULL, "maxsection-pct", 1,
941 "GOP max bitrate (% of target)");
942 static const arg_def_t *rc_twopass_args[] =
944 &bias_pct, &minsection_pct, &maxsection_pct, NULL
948 static const arg_def_t kf_min_dist = ARG_DEF(NULL, "kf-min-dist", 1,
949 "Minimum keyframe interval (frames)");
950 static const arg_def_t kf_max_dist = ARG_DEF(NULL, "kf-max-dist", 1,
951 "Maximum keyframe interval (frames)");
952 static const arg_def_t kf_disabled = ARG_DEF(NULL, "disable-kf", 0,
953 "Disable keyframe placement");
954 static const arg_def_t *kf_args[] =
956 &kf_min_dist, &kf_max_dist, &kf_disabled, NULL
960 #if CONFIG_VP8_ENCODER
961 static const arg_def_t noise_sens = ARG_DEF(NULL, "noise-sensitivity", 1,
962 "Noise sensitivity (frames to blur)");
963 static const arg_def_t sharpness = ARG_DEF(NULL, "sharpness", 1,
964 "Filter sharpness (0-7)");
965 static const arg_def_t static_thresh = ARG_DEF(NULL, "static-thresh", 1,
966 "Motion detection threshold");
969 #if CONFIG_VP8_ENCODER
970 static const arg_def_t cpu_used = ARG_DEF(NULL, "cpu-used", 1,
971 "CPU Used (-16..16)");
975 #if CONFIG_VP8_ENCODER
976 static const arg_def_t token_parts = ARG_DEF(NULL, "token-parts", 1,
977 "Number of token partitions to use, log2");
978 static const arg_def_t auto_altref = ARG_DEF(NULL, "auto-alt-ref", 1,
979 "Enable automatic alt reference frames");
980 static const arg_def_t arnr_maxframes = ARG_DEF(NULL, "arnr-maxframes", 1,
981 "alt_ref Max Frames");
982 static const arg_def_t arnr_strength = ARG_DEF(NULL, "arnr-strength", 1,
984 static const arg_def_t arnr_type = ARG_DEF(NULL, "arnr-type", 1,
987 static const arg_def_t *vp8_args[] =
989 &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh,
990 &token_parts, &arnr_maxframes, &arnr_strength, &arnr_type, NULL
992 static const int vp8_arg_ctrl_map[] =
994 VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF,
995 VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD,
996 VP8E_SET_TOKEN_PARTITIONS,
997 VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH , VP8E_SET_ARNR_TYPE, 0
1001 static const arg_def_t *no_args[] = { NULL };
1003 static void usage_exit()
1007 fprintf(stderr, "Usage: %s <options> -o dst_filename src_filename \n",
1010 fprintf(stderr, "\nOptions:\n");
1011 arg_show_usage(stdout, main_args);
1012 fprintf(stderr, "\nEncoder Global Options:\n");
1013 arg_show_usage(stdout, global_args);
1014 fprintf(stderr, "\nRate Control Options:\n");
1015 arg_show_usage(stdout, rc_args);
1016 fprintf(stderr, "\nTwopass Rate Control Options:\n");
1017 arg_show_usage(stdout, rc_twopass_args);
1018 fprintf(stderr, "\nKeyframe Placement Options:\n");
1019 arg_show_usage(stdout, kf_args);
1020 #if CONFIG_VP8_ENCODER
1021 fprintf(stderr, "\nVP8 Specific Options:\n");
1022 arg_show_usage(stdout, vp8_args);
1024 fprintf(stderr, "\n"
1025 "Included encoders:\n"
1028 for (i = 0; i < sizeof(codecs) / sizeof(codecs[0]); i++)
1029 fprintf(stderr, " %-6s - %s\n",
1031 vpx_codec_iface_name(codecs[i].iface));
1036 #define ARG_CTRL_CNT_MAX 10
1039 int main(int argc, const char **argv_)
1041 vpx_codec_ctx_t encoder;
1042 const char *in_fn = NULL, *out_fn = NULL, *stats_fn = NULL;
1044 FILE *infile, *outfile;
1045 vpx_codec_enc_cfg_t cfg;
1046 vpx_codec_err_t res;
1047 int pass, one_pass_only = 0;
1050 const struct codec_item *codec = codecs;
1051 int frame_avail, got_data;
1054 char **argv, **argi, **argj;
1055 int arg_usage = 0, arg_passes = 1, arg_deadline = 0;
1056 int arg_ctrls[ARG_CTRL_CNT_MAX][2], arg_ctrl_cnt = 0;
1058 static const arg_def_t **ctrl_args = no_args;
1059 static const int *ctrl_args_map = NULL;
1060 int verbose = 0, show_psnr = 0;
1061 int arg_use_i420 = 1;
1062 unsigned long cx_time = 0;
1063 unsigned int file_type, fourcc;
1065 struct vpx_rational arg_framerate = {30, 1};
1066 int arg_have_framerate = 0;
1068 EbmlGlobal ebml = {0};
1070 uint64_t psnr_sse_total = 0;
1071 uint64_t psnr_samples_total = 0;
1072 double psnr_totals[4] = {0, 0, 0, 0};
1075 exec_name = argv_[0];
1081 /* First parse the codec and usage values, because we want to apply other
1082 * parameters on top of the default configuration provided by the codec.
1084 argv = argv_dup(argc - 1, argv_ + 1);
1086 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
1090 if (arg_match(&arg, &codecarg, argi))
1094 for (j = 0; j < sizeof(codecs) / sizeof(codecs[0]); j++)
1095 if (!strcmp(codecs[j].name, arg.val))
1101 die("Error: Unrecognized argument (%s) to --codec\n",
1105 else if (arg_match(&arg, &passes, argi))
1107 arg_passes = arg_parse_uint(&arg);
1109 if (arg_passes < 1 || arg_passes > 2)
1110 die("Error: Invalid number of passes (%d)\n", arg_passes);
1112 else if (arg_match(&arg, &pass_arg, argi))
1114 one_pass_only = arg_parse_uint(&arg);
1116 if (one_pass_only < 1 || one_pass_only > 2)
1117 die("Error: Invalid pass selected (%d)\n", one_pass_only);
1119 else if (arg_match(&arg, &fpf_name, argi))
1121 else if (arg_match(&arg, &usage, argi))
1122 arg_usage = arg_parse_uint(&arg);
1123 else if (arg_match(&arg, &deadline, argi))
1124 arg_deadline = arg_parse_uint(&arg);
1125 else if (arg_match(&arg, &best_dl, argi))
1126 arg_deadline = VPX_DL_BEST_QUALITY;
1127 else if (arg_match(&arg, &good_dl, argi))
1128 arg_deadline = VPX_DL_GOOD_QUALITY;
1129 else if (arg_match(&arg, &rt_dl, argi))
1130 arg_deadline = VPX_DL_REALTIME;
1131 else if (arg_match(&arg, &use_yv12, argi))
1135 else if (arg_match(&arg, &use_i420, argi))
1139 else if (arg_match(&arg, &verbosearg, argi))
1141 else if (arg_match(&arg, &limit, argi))
1142 arg_limit = arg_parse_uint(&arg);
1143 else if (arg_match(&arg, &psnrarg, argi))
1145 else if (arg_match(&arg, &framerate, argi))
1147 arg_framerate = arg_parse_rational(&arg);
1148 arg_have_framerate = 1;
1150 else if (arg_match(&arg, &use_ivf, argi))
1152 else if (arg_match(&arg, &outputfile, argi))
1154 else if (arg_match(&arg, &debugmode, argi))
1160 /* Ensure that --passes and --pass are consistent. If --pass is set and --passes=2,
1161 * ensure --fpf was set.
1165 /* DWIM: Assume the user meant passes=2 if pass=2 is specified */
1166 if (one_pass_only > arg_passes)
1168 fprintf(stderr, "Warning: Assuming --pass=%d implies --passes=%d\n",
1169 one_pass_only, one_pass_only);
1170 arg_passes = one_pass_only;
1173 if (arg_passes == 2 && !stats_fn)
1174 die("Must specify --fpf when --pass=%d and --passes=2\n", one_pass_only);
1177 /* Populate encoder configuration */
1178 res = vpx_codec_enc_config_default(codec->iface, &cfg, arg_usage);
1182 fprintf(stderr, "Failed to get config: %s\n",
1183 vpx_codec_err_to_string(res));
1184 return EXIT_FAILURE;
1187 /* Change the default timebase to a high enough value so that the encoder
1188 * will always create strictly increasing timestamps.
1190 cfg.g_timebase.den = 1000;
1192 /* Now parse the remainder of the parameters. */
1193 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
1198 else if (arg_match(&arg, &threads, argi))
1199 cfg.g_threads = arg_parse_uint(&arg);
1200 else if (arg_match(&arg, &profile, argi))
1201 cfg.g_profile = arg_parse_uint(&arg);
1202 else if (arg_match(&arg, &width, argi))
1203 cfg.g_w = arg_parse_uint(&arg);
1204 else if (arg_match(&arg, &height, argi))
1205 cfg.g_h = arg_parse_uint(&arg);
1206 else if (arg_match(&arg, &timebase, argi))
1207 cfg.g_timebase = arg_parse_rational(&arg);
1208 else if (arg_match(&arg, &error_resilient, argi))
1209 cfg.g_error_resilient = arg_parse_uint(&arg);
1210 else if (arg_match(&arg, &lag_in_frames, argi))
1211 cfg.g_lag_in_frames = arg_parse_uint(&arg);
1212 else if (arg_match(&arg, &dropframe_thresh, argi))
1213 cfg.rc_dropframe_thresh = arg_parse_uint(&arg);
1214 else if (arg_match(&arg, &resize_allowed, argi))
1215 cfg.rc_resize_allowed = arg_parse_uint(&arg);
1216 else if (arg_match(&arg, &resize_up_thresh, argi))
1217 cfg.rc_resize_up_thresh = arg_parse_uint(&arg);
1218 else if (arg_match(&arg, &resize_down_thresh, argi))
1219 cfg.rc_resize_down_thresh = arg_parse_uint(&arg);
1220 else if (arg_match(&arg, &resize_down_thresh, argi))
1221 cfg.rc_resize_down_thresh = arg_parse_uint(&arg);
1222 else if (arg_match(&arg, &end_usage, argi))
1223 cfg.rc_end_usage = arg_parse_uint(&arg);
1224 else if (arg_match(&arg, &target_bitrate, argi))
1225 cfg.rc_target_bitrate = arg_parse_uint(&arg);
1226 else if (arg_match(&arg, &min_quantizer, argi))
1227 cfg.rc_min_quantizer = arg_parse_uint(&arg);
1228 else if (arg_match(&arg, &max_quantizer, argi))
1229 cfg.rc_max_quantizer = arg_parse_uint(&arg);
1230 else if (arg_match(&arg, &undershoot_pct, argi))
1231 cfg.rc_undershoot_pct = arg_parse_uint(&arg);
1232 else if (arg_match(&arg, &overshoot_pct, argi))
1233 cfg.rc_overshoot_pct = arg_parse_uint(&arg);
1234 else if (arg_match(&arg, &buf_sz, argi))
1235 cfg.rc_buf_sz = arg_parse_uint(&arg);
1236 else if (arg_match(&arg, &buf_initial_sz, argi))
1237 cfg.rc_buf_initial_sz = arg_parse_uint(&arg);
1238 else if (arg_match(&arg, &buf_optimal_sz, argi))
1239 cfg.rc_buf_optimal_sz = arg_parse_uint(&arg);
1240 else if (arg_match(&arg, &bias_pct, argi))
1242 cfg.rc_2pass_vbr_bias_pct = arg_parse_uint(&arg);
1246 "Warning: option %s ignored in one-pass mode.\n",
1249 else if (arg_match(&arg, &minsection_pct, argi))
1251 cfg.rc_2pass_vbr_minsection_pct = arg_parse_uint(&arg);
1255 "Warning: option %s ignored in one-pass mode.\n",
1258 else if (arg_match(&arg, &maxsection_pct, argi))
1260 cfg.rc_2pass_vbr_maxsection_pct = arg_parse_uint(&arg);
1264 "Warning: option %s ignored in one-pass mode.\n",
1267 else if (arg_match(&arg, &kf_min_dist, argi))
1268 cfg.kf_min_dist = arg_parse_uint(&arg);
1269 else if (arg_match(&arg, &kf_max_dist, argi))
1270 cfg.kf_max_dist = arg_parse_uint(&arg);
1271 else if (arg_match(&arg, &kf_disabled, argi))
1272 cfg.kf_mode = VPX_KF_DISABLED;
1277 /* Handle codec specific options */
1278 #if CONFIG_VP8_ENCODER
1280 if (codec->iface == &vpx_codec_vp8_cx_algo)
1282 ctrl_args = vp8_args;
1283 ctrl_args_map = vp8_arg_ctrl_map;
1288 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
1294 for (i = 0; ctrl_args[i]; i++)
1296 if (arg_match(&arg, ctrl_args[i], argi))
1300 if (arg_ctrl_cnt < ARG_CTRL_CNT_MAX)
1302 arg_ctrls[arg_ctrl_cnt][0] = ctrl_args_map[i];
1303 arg_ctrls[arg_ctrl_cnt][1] = arg_parse_int(&arg);
1313 /* Check for unrecognized options */
1314 for (argi = argv; *argi; argi++)
1315 if (argi[0][0] == '-' && argi[0][1])
1316 die("Error: Unrecognized option %s\n", *argi);
1318 /* Handle non-option arguments */
1325 die("Error: Output file is required (specify with -o)\n");
1327 memset(&stats, 0, sizeof(stats));
1329 for (pass = one_pass_only ? one_pass_only - 1 : 0; pass < arg_passes; pass++)
1331 int frames_in = 0, frames_out = 0;
1332 unsigned long nbytes = 0;
1333 size_t detect_bytes;
1334 struct detect_buffer detect;
1336 /* Parse certain options from the input file, if possible */
1337 infile = strcmp(in_fn, "-") ? fopen(in_fn, "rb") : stdin;
1341 fprintf(stderr, "Failed to open input file\n");
1342 return EXIT_FAILURE;
1345 /* For RAW input sources, these bytes will applied on the first frame
1347 * We can always read 4 bytes because the minimum supported frame size
1350 detect_bytes = fread(detect.buf, 1, 4, infile);
1353 if (detect_bytes == 4 && file_is_y4m(infile, &y4m, detect.buf))
1355 if (y4m_input_open(&y4m, infile, detect.buf, 4) >= 0)
1357 file_type = FILE_TYPE_Y4M;
1358 cfg.g_w = y4m.pic_w;
1359 cfg.g_h = y4m.pic_h;
1361 /* Use the frame rate from the file only if none was specified
1362 * on the command-line.
1364 if (!arg_have_framerate)
1366 arg_framerate.num = y4m.fps_n;
1367 arg_framerate.den = y4m.fps_d;
1374 fprintf(stderr, "Unsupported Y4M stream.\n");
1375 return EXIT_FAILURE;
1378 else if (detect_bytes == 4 &&
1379 file_is_ivf(infile, &fourcc, &cfg.g_w, &cfg.g_h, detect.buf))
1381 file_type = FILE_TYPE_IVF;
1391 fprintf(stderr, "Unsupported fourcc (%08x) in IVF\n", fourcc);
1392 return EXIT_FAILURE;
1397 file_type = FILE_TYPE_RAW;
1400 #define SHOW(field) fprintf(stderr, " %-28s = %d\n", #field, cfg.field)
1402 if (verbose && pass == 0)
1404 fprintf(stderr, "Codec: %s\n", vpx_codec_iface_name(codec->iface));
1405 fprintf(stderr, "Source file: %s Format: %s\n", in_fn,
1406 arg_use_i420 ? "I420" : "YV12");
1407 fprintf(stderr, "Destination file: %s\n", out_fn);
1408 fprintf(stderr, "Encoder parameters:\n");
1415 SHOW(g_timebase.num);
1416 SHOW(g_timebase.den);
1417 SHOW(g_error_resilient);
1419 SHOW(g_lag_in_frames);
1420 SHOW(rc_dropframe_thresh);
1421 SHOW(rc_resize_allowed);
1422 SHOW(rc_resize_up_thresh);
1423 SHOW(rc_resize_down_thresh);
1425 SHOW(rc_target_bitrate);
1426 SHOW(rc_min_quantizer);
1427 SHOW(rc_max_quantizer);
1428 SHOW(rc_undershoot_pct);
1429 SHOW(rc_overshoot_pct);
1431 SHOW(rc_buf_initial_sz);
1432 SHOW(rc_buf_optimal_sz);
1433 SHOW(rc_2pass_vbr_bias_pct);
1434 SHOW(rc_2pass_vbr_minsection_pct);
1435 SHOW(rc_2pass_vbr_maxsection_pct);
1441 if(pass == (one_pass_only ? one_pass_only - 1 : 0)) {
1442 if (file_type == FILE_TYPE_Y4M)
1443 /*The Y4M reader does its own allocation.
1444 Just initialize this here to avoid problems if we never read any
1446 memset(&raw, 0, sizeof(raw));
1448 vpx_img_alloc(&raw, arg_use_i420 ? VPX_IMG_FMT_I420 : VPX_IMG_FMT_YV12,
1449 cfg.g_w, cfg.g_h, 1);
1452 outfile = strcmp(out_fn, "-") ? fopen(out_fn, "wb") : stdout;
1456 fprintf(stderr, "Failed to open output file\n");
1457 return EXIT_FAILURE;
1460 if(write_webm && fseek(outfile, 0, SEEK_CUR))
1462 fprintf(stderr, "WebM output to pipes not supported.\n");
1463 return EXIT_FAILURE;
1468 if (!stats_open_file(&stats, stats_fn, pass))
1470 fprintf(stderr, "Failed to open statistics store\n");
1471 return EXIT_FAILURE;
1476 if (!stats_open_mem(&stats, pass))
1478 fprintf(stderr, "Failed to open statistics store\n");
1479 return EXIT_FAILURE;
1483 cfg.g_pass = arg_passes == 2
1484 ? pass ? VPX_RC_LAST_PASS : VPX_RC_FIRST_PASS
1486 #if VPX_ENCODER_ABI_VERSION > (1 + VPX_CODEC_ABI_VERSION)
1490 cfg.rc_twopass_stats_in = stats_get(&stats);
1497 ebml.stream = outfile;
1498 write_webm_file_header(&ebml, &cfg, &arg_framerate);
1501 write_ivf_file_header(outfile, &cfg, codec->fourcc, 0);
1504 /* Construct Encoder Context */
1505 vpx_codec_enc_init(&encoder, codec->iface, &cfg,
1506 show_psnr ? VPX_CODEC_USE_PSNR : 0);
1507 ctx_exit_on_error(&encoder, "Failed to initialize encoder");
1509 /* Note that we bypass the vpx_codec_control wrapper macro because
1510 * we're being clever to store the control IDs in an array. Real
1511 * applications will want to make use of the enumerations directly
1513 for (i = 0; i < arg_ctrl_cnt; i++)
1515 if (vpx_codec_control_(&encoder, arg_ctrls[i][0], arg_ctrls[i][1]))
1516 fprintf(stderr, "Error: Tried to set control %d = %d\n",
1517 arg_ctrls[i][0], arg_ctrls[i][1]);
1519 ctx_exit_on_error(&encoder, "Failed to control codec");
1525 while (frame_avail || got_data)
1527 vpx_codec_iter_t iter = NULL;
1528 const vpx_codec_cx_pkt_t *pkt;
1529 struct vpx_usec_timer timer;
1530 int64_t frame_start;
1532 if (!arg_limit || frames_in < arg_limit)
1534 frame_avail = read_frame(infile, &raw, file_type, &y4m,
1541 "\rPass %d/%d frame %4d/%-4d %7ldB \033[K", pass + 1,
1542 arg_passes, frames_in, frames_out, nbytes);
1547 vpx_usec_timer_start(&timer);
1549 frame_start = (cfg.g_timebase.den * (int64_t)(frames_in - 1)
1550 * arg_framerate.den) / cfg.g_timebase.num / arg_framerate.num;
1551 vpx_codec_encode(&encoder, frame_avail ? &raw : NULL, frame_start,
1552 cfg.g_timebase.den * arg_framerate.den
1553 / cfg.g_timebase.num / arg_framerate.num,
1555 vpx_usec_timer_mark(&timer);
1556 cx_time += vpx_usec_timer_elapsed(&timer);
1557 ctx_exit_on_error(&encoder, "Failed to encode frame");
1560 while ((pkt = vpx_codec_get_cx_data(&encoder, &iter)))
1566 case VPX_CODEC_CX_FRAME_PKT:
1568 fprintf(stderr, " %6luF",
1569 (unsigned long)pkt->data.frame.sz);
1573 /* Update the hash */
1575 hash = murmur(pkt->data.frame.buf,
1576 pkt->data.frame.sz, hash);
1578 write_webm_block(&ebml, &cfg, pkt);
1582 write_ivf_frame_header(outfile, pkt);
1583 if(fwrite(pkt->data.frame.buf, 1,
1584 pkt->data.frame.sz, outfile));
1586 nbytes += pkt->data.raw.sz;
1588 case VPX_CODEC_STATS_PKT:
1590 fprintf(stderr, " %6luS",
1591 (unsigned long)pkt->data.twopass_stats.sz);
1593 pkt->data.twopass_stats.buf,
1594 pkt->data.twopass_stats.sz);
1595 nbytes += pkt->data.raw.sz;
1597 case VPX_CODEC_PSNR_PKT:
1603 psnr_sse_total += pkt->data.psnr.sse[0];
1604 psnr_samples_total += pkt->data.psnr.samples[0];
1605 for (i = 0; i < 4; i++)
1607 fprintf(stderr, "%.3lf ", pkt->data.psnr.psnr[i]);
1608 psnr_totals[i] += pkt->data.psnr.psnr[i];
1623 "\rPass %d/%d frame %4d/%-4d %7ldB %7ldb/f %7"PRId64"b/s"
1624 " %7lu %s (%.2f fps)\033[K", pass + 1,
1625 arg_passes, frames_in, frames_out, nbytes, nbytes * 8 / frames_in,
1626 nbytes * 8 *(int64_t)arg_framerate.num / arg_framerate.den / frames_in,
1627 cx_time > 9999999 ? cx_time / 1000 : cx_time,
1628 cx_time > 9999999 ? "ms" : "us",
1629 (float)frames_in * 1000000.0 / (float)cx_time);
1631 if ( (show_psnr) && (psnr_count>0) )
1634 double ovpsnr = vp8_mse2psnr(psnr_samples_total, 255.0,
1637 fprintf(stderr, "\nPSNR (Overall/Avg/Y/U/V)");
1639 fprintf(stderr, " %.3lf", ovpsnr);
1640 for (i = 0; i < 4; i++)
1642 fprintf(stderr, " %.3lf", psnr_totals[i]/psnr_count);
1646 vpx_codec_destroy(&encoder);
1652 write_webm_file_footer(&ebml, hash);
1656 if (!fseek(outfile, 0, SEEK_SET))
1657 write_ivf_file_header(outfile, &cfg, codec->fourcc, frames_out);
1661 stats_close(&stats);
1662 fprintf(stderr, "\n");
1670 return EXIT_SUCCESS;