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.
15 #if defined(_WIN32) || !CONFIG_OS_SUPPORT
16 #define USE_POSIX_MMAP 0
18 #define USE_POSIX_MMAP 1
27 #include "vpx/vpx_encoder.h"
29 #include <sys/types.h>
35 #include "vpx/vp8cx.h"
36 #include "vpx_ports/mem_ops.h"
37 #include "vpx_ports/vpx_timer.h"
38 #include "tools_common.h"
40 #include "libmkv/EbmlWriter.h"
41 #include "libmkv/EbmlIDs.h"
43 /* Need special handling of these functions on Windows */
45 /* MSVS doesn't define off_t, and uses _f{seek,tell}i64 */
46 typedef __int64 off_t;
47 #define fseeko _fseeki64
48 #define ftello _ftelli64
50 /* MinGW defines off_t as long
51 and uses f{seek,tell}o64/off64_t for large files */
52 #define fseeko fseeko64
53 #define ftello ftello64
57 #define LITERALU64(hi,lo) ((((uint64_t)hi)<<32)|lo)
59 /* We should use 32-bit file operations in WebM file format
60 * when building ARM executable file (.axf) with RVCT */
61 #if !CONFIG_OS_SUPPORT
67 /* Swallow warnings about unused results of fread/fwrite */
68 static size_t wrap_fread(void *ptr, size_t size, size_t nmemb,
71 return fread(ptr, size, nmemb, stream);
73 #define fread wrap_fread
75 static size_t wrap_fwrite(const void *ptr, size_t size, size_t nmemb,
78 return fwrite(ptr, size, nmemb, stream);
80 #define fwrite wrap_fwrite
83 static const char *exec_name;
85 static const struct codec_item
88 vpx_codec_iface_t *iface;
92 #if CONFIG_VP8_ENCODER
93 {"vp8", &vpx_codec_vp8_cx_algo, 0x30385056},
97 static void usage_exit();
99 #define LOG_ERROR(label) do \
101 const char *l=label;\
105 fprintf(stderr, "%s: ", l);\
106 vfprintf(stderr, fmt, ap);\
107 fprintf(stderr, "\n");\
111 void die(const char *fmt, ...)
118 void fatal(const char *fmt, ...)
125 void warn(const char *fmt, ...)
127 LOG_ERROR("Warning");
131 static void ctx_exit_on_error(vpx_codec_ctx_t *ctx, const char *s, ...)
138 const char *detail = vpx_codec_error_detail(ctx);
140 vfprintf(stderr, s, ap);
141 fprintf(stderr, ": %s\n", vpx_codec_error(ctx));
144 fprintf(stderr, " %s\n", detail);
150 /* This structure is used to abstract the different ways of handling
151 * first pass statistics.
162 int stats_open_file(stats_io_t *stats, const char *fpf, int pass)
170 stats->file = fopen(fpf, "wb");
172 stats->buf.buf = NULL,
173 res = (stats->file != NULL);
179 struct stat stat_buf;
182 fd = open(fpf, O_RDONLY);
183 stats->file = fdopen(fd, "rb");
184 fstat(fd, &stat_buf);
185 stats->buf.sz = stat_buf.st_size;
186 stats->buf.buf = mmap(NULL, stats->buf.sz, PROT_READ, MAP_PRIVATE,
188 res = (stats->buf.buf != NULL);
192 stats->file = fopen(fpf, "rb");
194 if (fseek(stats->file, 0, SEEK_END))
195 fatal("First-pass stats file must be seekable!");
197 stats->buf.sz = stats->buf_alloc_sz = ftell(stats->file);
200 stats->buf.buf = malloc(stats->buf_alloc_sz);
203 fatal("Failed to allocate first-pass stats buffer (%lu bytes)",
204 (unsigned long)stats->buf_alloc_sz);
206 nbytes = fread(stats->buf.buf, 1, stats->buf.sz, stats->file);
207 res = (nbytes == stats->buf.sz);
214 int stats_open_mem(stats_io_t *stats, int pass)
222 stats->buf_alloc_sz = 64 * 1024;
223 stats->buf.buf = malloc(stats->buf_alloc_sz);
226 stats->buf_ptr = stats->buf.buf;
227 res = (stats->buf.buf != NULL);
232 void stats_close(stats_io_t *stats, int last_pass)
236 if (stats->pass == last_pass)
240 munmap(stats->buf.buf, stats->buf.sz);
242 free(stats->buf.buf);
251 if (stats->pass == last_pass)
252 free(stats->buf.buf);
256 void stats_write(stats_io_t *stats, const void *pkt, size_t len)
260 (void) fwrite(pkt, 1, len, stats->file);
264 if (stats->buf.sz + len > stats->buf_alloc_sz)
266 size_t new_sz = stats->buf_alloc_sz + 64 * 1024;
267 char *new_ptr = realloc(stats->buf.buf, new_sz);
271 stats->buf_ptr = new_ptr + (stats->buf_ptr - (char *)stats->buf.buf);
272 stats->buf.buf = new_ptr;
273 stats->buf_alloc_sz = new_sz;
276 fatal("Failed to realloc firstpass stats buffer.");
279 memcpy(stats->buf_ptr, pkt, len);
280 stats->buf.sz += len;
281 stats->buf_ptr += len;
285 vpx_fixed_buf_t stats_get(stats_io_t *stats)
290 /* Stereo 3D packed frame format */
291 typedef enum stereo_format
293 STEREO_FORMAT_MONO = 0,
294 STEREO_FORMAT_LEFT_RIGHT = 1,
295 STEREO_FORMAT_BOTTOM_TOP = 2,
296 STEREO_FORMAT_TOP_BOTTOM = 3,
297 STEREO_FORMAT_RIGHT_LEFT = 11
307 struct detect_buffer {
319 struct detect_buffer detect;
320 enum video_file_type file_type;
323 struct vpx_rational framerate;
328 #define IVF_FRAME_HDR_SZ (4+8) /* 4 byte size + 8 byte timestamp */
329 static int read_frame(struct input_state *input, vpx_image_t *img)
331 FILE *f = input->file;
332 enum video_file_type file_type = input->file_type;
333 y4m_input *y4m = &input->y4m;
334 struct detect_buffer *detect = &input->detect;
338 if (file_type == FILE_TYPE_Y4M)
340 if (y4m_input_fetch_frame(y4m, f, img) < 1)
345 if (file_type == FILE_TYPE_IVF)
347 char junk[IVF_FRAME_HDR_SZ];
349 /* Skip the frame header. We know how big the frame should be. See
350 * write_ivf_frame_header() for documentation on the frame header
353 (void) fread(junk, 1, IVF_FRAME_HDR_SZ, f);
356 for (plane = 0; plane < 3; plane++)
359 int w = (plane ? (1 + img->d_w) / 2 : img->d_w);
360 int h = (plane ? (1 + img->d_h) / 2 : img->d_h);
363 /* Determine the correct plane based on the image format. The for-loop
364 * always counts in Y,U,V order, but this may not match the order of
370 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U];
373 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V];
376 ptr = img->planes[plane];
379 for (r = 0; r < h; r++)
382 size_t buf_position = 0;
383 const size_t left = detect->buf_read - detect->position;
386 const size_t more = (left < needed) ? left : needed;
387 memcpy(ptr, detect->buf + detect->position, more);
390 detect->position += more;
394 shortread |= (fread(ptr + buf_position, 1, needed, f) < needed);
397 ptr += img->stride[plane];
406 unsigned int file_is_y4m(FILE *infile,
410 if(memcmp(detect, "YUV4", 4) == 0)
417 #define IVF_FILE_HDR_SZ (32)
418 unsigned int file_is_ivf(struct input_state *input,
419 unsigned int *fourcc)
421 char raw_hdr[IVF_FILE_HDR_SZ];
423 FILE *infile = input->file;
424 unsigned int *width = &input->w;
425 unsigned int *height = &input->h;
426 struct detect_buffer *detect = &input->detect;
428 if(memcmp(detect->buf, "DKIF", 4) != 0)
431 /* See write_ivf_file_header() for more documentation on the file header
434 if (fread(raw_hdr + 4, 1, IVF_FILE_HDR_SZ - 4, infile)
435 == IVF_FILE_HDR_SZ - 4)
440 if (mem_get_le16(raw_hdr + 4) != 0)
441 warn("Unrecognized IVF version! This file may not decode "
444 *fourcc = mem_get_le32(raw_hdr + 8);
450 *width = mem_get_le16(raw_hdr + 12);
451 *height = mem_get_le16(raw_hdr + 14);
452 detect->position = 4;
459 static void write_ivf_file_header(FILE *outfile,
460 const vpx_codec_enc_cfg_t *cfg,
466 if (cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS)
473 mem_put_le16(header + 4, 0); /* version */
474 mem_put_le16(header + 6, 32); /* headersize */
475 mem_put_le32(header + 8, fourcc); /* headersize */
476 mem_put_le16(header + 12, cfg->g_w); /* width */
477 mem_put_le16(header + 14, cfg->g_h); /* height */
478 mem_put_le32(header + 16, cfg->g_timebase.den); /* rate */
479 mem_put_le32(header + 20, cfg->g_timebase.num); /* scale */
480 mem_put_le32(header + 24, frame_cnt); /* length */
481 mem_put_le32(header + 28, 0); /* unused */
483 (void) fwrite(header, 1, 32, outfile);
487 static void write_ivf_frame_header(FILE *outfile,
488 const vpx_codec_cx_pkt_t *pkt)
493 if (pkt->kind != VPX_CODEC_CX_FRAME_PKT)
496 pts = pkt->data.frame.pts;
497 mem_put_le32(header, pkt->data.frame.sz);
498 mem_put_le32(header + 4, pts & 0xFFFFFFFF);
499 mem_put_le32(header + 8, pts >> 32);
501 (void) fwrite(header, 1, 12, outfile);
504 static void write_ivf_frame_size(FILE *outfile, size_t size)
507 mem_put_le32(header, size);
508 (void) fwrite(header, 1, 4, outfile);
512 typedef off_t EbmlLoc;
528 vpx_rational_t framerate;
530 /* These pointers are to the start of an element */
531 off_t position_reference;
533 off_t segment_info_pos;
538 /* This pointer is to a specific element to be serialized */
541 /* These pointers are to the size field of the element */
542 EbmlLoc startSegment;
543 EbmlLoc startCluster;
545 uint32_t cluster_timecode;
548 struct cue_entry *cue_list;
554 void Ebml_Write(EbmlGlobal *glob, const void *buffer_in, unsigned long len)
556 (void) fwrite(buffer_in, 1, len, glob->stream);
559 #define WRITE_BUFFER(s) \
560 for(i = len-1; i>=0; i--)\
562 x = *(const s *)buffer_in >> (i * CHAR_BIT); \
563 Ebml_Write(glob, &x, 1); \
565 void Ebml_Serialize(EbmlGlobal *glob, const void *buffer_in, int buffer_size, unsigned long len)
582 WRITE_BUFFER(int16_t)
585 WRITE_BUFFER(int32_t)
588 WRITE_BUFFER(int64_t)
596 /* Need a fixed size serializer for the track ID. libmkv provides a 64 bit
597 * one, but not a 32 bit one.
599 static void Ebml_SerializeUnsigned32(EbmlGlobal *glob, unsigned long class_id, uint64_t ui)
601 unsigned char sizeSerialized = 4 | 0x80;
602 Ebml_WriteID(glob, class_id);
603 Ebml_Serialize(glob, &sizeSerialized, sizeof(sizeSerialized), 1);
604 Ebml_Serialize(glob, &ui, sizeof(ui), 4);
609 Ebml_StartSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc,
610 unsigned long class_id)
612 /* todo this is always taking 8 bytes, this may need later optimization */
613 /* this is a key that says length unknown */
614 uint64_t unknownLen = LITERALU64(0x01FFFFFF, 0xFFFFFFFF);
616 Ebml_WriteID(glob, class_id);
617 *ebmlLoc = ftello(glob->stream);
618 Ebml_Serialize(glob, &unknownLen, sizeof(unknownLen), 8);
622 Ebml_EndSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc)
627 /* Save the current stream pointer */
628 pos = ftello(glob->stream);
630 /* Calculate the size of this element */
631 size = pos - *ebmlLoc - 8;
632 size |= LITERALU64(0x01000000,0x00000000);
634 /* Seek back to the beginning of the element and write the new size */
635 fseeko(glob->stream, *ebmlLoc, SEEK_SET);
636 Ebml_Serialize(glob, &size, sizeof(size), 8);
638 /* Reset the stream pointer */
639 fseeko(glob->stream, pos, SEEK_SET);
644 write_webm_seek_element(EbmlGlobal *ebml, unsigned long id, off_t pos)
646 uint64_t offset = pos - ebml->position_reference;
648 Ebml_StartSubElement(ebml, &start, Seek);
649 Ebml_SerializeBinary(ebml, SeekID, id);
650 Ebml_SerializeUnsigned64(ebml, SeekPosition, offset);
651 Ebml_EndSubElement(ebml, &start);
656 write_webm_seek_info(EbmlGlobal *ebml)
661 /* Save the current stream pointer */
662 pos = ftello(ebml->stream);
664 if(ebml->seek_info_pos)
665 fseeko(ebml->stream, ebml->seek_info_pos, SEEK_SET);
667 ebml->seek_info_pos = pos;
672 Ebml_StartSubElement(ebml, &start, SeekHead);
673 write_webm_seek_element(ebml, Tracks, ebml->track_pos);
674 write_webm_seek_element(ebml, Cues, ebml->cue_pos);
675 write_webm_seek_element(ebml, Info, ebml->segment_info_pos);
676 Ebml_EndSubElement(ebml, &start);
682 char version_string[64];
684 /* Assemble version string */
686 strcpy(version_string, "vpxenc");
689 strcpy(version_string, "vpxenc ");
690 strncat(version_string,
691 vpx_codec_version_str(),
692 sizeof(version_string) - 1 - strlen(version_string));
695 frame_time = (uint64_t)1000 * ebml->framerate.den
696 / ebml->framerate.num;
697 ebml->segment_info_pos = ftello(ebml->stream);
698 Ebml_StartSubElement(ebml, &startInfo, Info);
699 Ebml_SerializeUnsigned(ebml, TimecodeScale, 1000000);
700 Ebml_SerializeFloat(ebml, Segment_Duration,
701 ebml->last_pts_ms + frame_time);
702 Ebml_SerializeString(ebml, 0x4D80, version_string);
703 Ebml_SerializeString(ebml, 0x5741, version_string);
704 Ebml_EndSubElement(ebml, &startInfo);
710 write_webm_file_header(EbmlGlobal *glob,
711 const vpx_codec_enc_cfg_t *cfg,
712 const struct vpx_rational *fps,
713 stereo_format_t stereo_fmt)
717 Ebml_StartSubElement(glob, &start, EBML);
718 Ebml_SerializeUnsigned(glob, EBMLVersion, 1);
719 Ebml_SerializeUnsigned(glob, EBMLReadVersion, 1);
720 Ebml_SerializeUnsigned(glob, EBMLMaxIDLength, 4);
721 Ebml_SerializeUnsigned(glob, EBMLMaxSizeLength, 8);
722 Ebml_SerializeString(glob, DocType, "webm");
723 Ebml_SerializeUnsigned(glob, DocTypeVersion, 2);
724 Ebml_SerializeUnsigned(glob, DocTypeReadVersion, 2);
725 Ebml_EndSubElement(glob, &start);
728 Ebml_StartSubElement(glob, &glob->startSegment, Segment);
729 glob->position_reference = ftello(glob->stream);
730 glob->framerate = *fps;
731 write_webm_seek_info(glob);
735 glob->track_pos = ftello(glob->stream);
736 Ebml_StartSubElement(glob, &trackStart, Tracks);
738 unsigned int trackNumber = 1;
739 uint64_t trackID = 0;
742 Ebml_StartSubElement(glob, &start, TrackEntry);
743 Ebml_SerializeUnsigned(glob, TrackNumber, trackNumber);
744 glob->track_id_pos = ftello(glob->stream);
745 Ebml_SerializeUnsigned32(glob, TrackUID, trackID);
746 Ebml_SerializeUnsigned(glob, TrackType, 1);
747 Ebml_SerializeString(glob, CodecID, "V_VP8");
749 unsigned int pixelWidth = cfg->g_w;
750 unsigned int pixelHeight = cfg->g_h;
751 float frameRate = (float)fps->num/(float)fps->den;
754 Ebml_StartSubElement(glob, &videoStart, Video);
755 Ebml_SerializeUnsigned(glob, PixelWidth, pixelWidth);
756 Ebml_SerializeUnsigned(glob, PixelHeight, pixelHeight);
757 Ebml_SerializeUnsigned(glob, StereoMode, stereo_fmt);
758 Ebml_SerializeFloat(glob, FrameRate, frameRate);
759 Ebml_EndSubElement(glob, &videoStart);
761 Ebml_EndSubElement(glob, &start); /* Track Entry */
763 Ebml_EndSubElement(glob, &trackStart);
765 /* segment element is open */
771 write_webm_block(EbmlGlobal *glob,
772 const vpx_codec_enc_cfg_t *cfg,
773 const vpx_codec_cx_pkt_t *pkt)
775 unsigned long block_length;
776 unsigned char track_number;
777 unsigned short block_timecode = 0;
780 int start_cluster = 0, is_keyframe;
782 /* Calculate the PTS of this frame in milliseconds */
783 pts_ms = pkt->data.frame.pts * 1000
784 * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
785 if(pts_ms <= glob->last_pts_ms)
786 pts_ms = glob->last_pts_ms + 1;
787 glob->last_pts_ms = pts_ms;
789 /* Calculate the relative time of this block */
790 if(pts_ms - glob->cluster_timecode > SHRT_MAX)
793 block_timecode = pts_ms - glob->cluster_timecode;
795 is_keyframe = (pkt->data.frame.flags & VPX_FRAME_IS_KEY);
796 if(start_cluster || is_keyframe)
798 if(glob->cluster_open)
799 Ebml_EndSubElement(glob, &glob->startCluster);
801 /* Open the new cluster */
803 glob->cluster_open = 1;
804 glob->cluster_timecode = pts_ms;
805 glob->cluster_pos = ftello(glob->stream);
806 Ebml_StartSubElement(glob, &glob->startCluster, Cluster); /* cluster */
807 Ebml_SerializeUnsigned(glob, Timecode, glob->cluster_timecode);
809 /* Save a cue point if this is a keyframe. */
812 struct cue_entry *cue, *new_cue_list;
814 new_cue_list = realloc(glob->cue_list,
815 (glob->cues+1) * sizeof(struct cue_entry));
817 glob->cue_list = new_cue_list;
819 fatal("Failed to realloc cue list.");
821 cue = &glob->cue_list[glob->cues];
822 cue->time = glob->cluster_timecode;
823 cue->loc = glob->cluster_pos;
828 /* Write the Simple Block */
829 Ebml_WriteID(glob, SimpleBlock);
831 block_length = pkt->data.frame.sz + 4;
832 block_length |= 0x10000000;
833 Ebml_Serialize(glob, &block_length, sizeof(block_length), 4);
836 track_number |= 0x80;
837 Ebml_Write(glob, &track_number, 1);
839 Ebml_Serialize(glob, &block_timecode, sizeof(block_timecode), 2);
844 if(pkt->data.frame.flags & VPX_FRAME_IS_INVISIBLE)
846 Ebml_Write(glob, &flags, 1);
848 Ebml_Write(glob, pkt->data.frame.buf, pkt->data.frame.sz);
853 write_webm_file_footer(EbmlGlobal *glob, long hash)
856 if(glob->cluster_open)
857 Ebml_EndSubElement(glob, &glob->startCluster);
863 glob->cue_pos = ftello(glob->stream);
864 Ebml_StartSubElement(glob, &start, Cues);
865 for(i=0; i<glob->cues; i++)
867 struct cue_entry *cue = &glob->cue_list[i];
870 Ebml_StartSubElement(glob, &start, CuePoint);
874 Ebml_SerializeUnsigned(glob, CueTime, cue->time);
876 Ebml_StartSubElement(glob, &start, CueTrackPositions);
877 Ebml_SerializeUnsigned(glob, CueTrack, 1);
878 Ebml_SerializeUnsigned64(glob, CueClusterPosition,
879 cue->loc - glob->position_reference);
880 Ebml_EndSubElement(glob, &start);
882 Ebml_EndSubElement(glob, &start);
884 Ebml_EndSubElement(glob, &start);
887 Ebml_EndSubElement(glob, &glob->startSegment);
889 /* Patch up the seek info block */
890 write_webm_seek_info(glob);
892 /* Patch up the track id */
893 fseeko(glob->stream, glob->track_id_pos, SEEK_SET);
894 Ebml_SerializeUnsigned32(glob, TrackUID, glob->debug ? 0xDEADBEEF : hash);
896 fseeko(glob->stream, 0, SEEK_END);
900 /* Murmur hash derived from public domain reference implementation at
901 * http://sites.google.com/site/murmurhash/
903 static unsigned int murmur ( const void * key, int len, unsigned int seed )
905 const unsigned int m = 0x5bd1e995;
908 unsigned int h = seed ^ len;
910 const unsigned char * data = (const unsigned char *)key;
934 case 3: h ^= data[2] << 16;
935 case 2: h ^= data[1] << 8;
936 case 1: h ^= data[0];
949 static double vp8_mse2psnr(double Samples, double Peak, double Mse)
953 if ((double)Mse > 0.0)
954 psnr = 10.0 * log10(Peak * Peak * Samples / Mse);
956 psnr = 60; /* Limit to prevent / 0 */
966 static const arg_def_t debugmode = ARG_DEF("D", "debug", 0,
967 "Debug mode (makes output deterministic)");
968 static const arg_def_t outputfile = ARG_DEF("o", "output", 1,
970 static const arg_def_t use_yv12 = ARG_DEF(NULL, "yv12", 0,
971 "Input file is YV12 ");
972 static const arg_def_t use_i420 = ARG_DEF(NULL, "i420", 0,
973 "Input file is I420 (default)");
974 static const arg_def_t codecarg = ARG_DEF(NULL, "codec", 1,
976 static const arg_def_t passes = ARG_DEF("p", "passes", 1,
977 "Number of passes (1/2)");
978 static const arg_def_t pass_arg = ARG_DEF(NULL, "pass", 1,
979 "Pass to execute (1/2)");
980 static const arg_def_t fpf_name = ARG_DEF(NULL, "fpf", 1,
981 "First pass statistics file name");
982 static const arg_def_t limit = ARG_DEF(NULL, "limit", 1,
983 "Stop encoding after n input frames");
984 static const arg_def_t deadline = ARG_DEF("d", "deadline", 1,
985 "Deadline per frame (usec)");
986 static const arg_def_t best_dl = ARG_DEF(NULL, "best", 0,
987 "Use Best Quality Deadline");
988 static const arg_def_t good_dl = ARG_DEF(NULL, "good", 0,
989 "Use Good Quality Deadline");
990 static const arg_def_t rt_dl = ARG_DEF(NULL, "rt", 0,
991 "Use Realtime Quality Deadline");
992 static const arg_def_t verbosearg = ARG_DEF("v", "verbose", 0,
993 "Show encoder parameters");
994 static const arg_def_t psnrarg = ARG_DEF(NULL, "psnr", 0,
995 "Show PSNR in status line");
996 static const arg_def_t framerate = ARG_DEF(NULL, "fps", 1,
997 "Stream frame rate (rate/scale)");
998 static const arg_def_t use_ivf = ARG_DEF(NULL, "ivf", 0,
999 "Output IVF (default is WebM)");
1000 static const arg_def_t out_part = ARG_DEF("P", "output-partitions", 0,
1001 "Makes encoder output partitions. Requires IVF output!");
1002 static const arg_def_t q_hist_n = ARG_DEF(NULL, "q-hist", 1,
1003 "Show quantizer histogram (n-buckets)");
1004 static const arg_def_t rate_hist_n = ARG_DEF(NULL, "rate-hist", 1,
1005 "Show rate histogram (n-buckets)");
1006 static const arg_def_t *main_args[] =
1009 &outputfile, &codecarg, &passes, &pass_arg, &fpf_name, &limit, &deadline,
1010 &best_dl, &good_dl, &rt_dl,
1011 &verbosearg, &psnrarg, &use_ivf, &out_part, &q_hist_n, &rate_hist_n,
1015 static const arg_def_t usage = ARG_DEF("u", "usage", 1,
1016 "Usage profile number to use");
1017 static const arg_def_t threads = ARG_DEF("t", "threads", 1,
1018 "Max number of threads to use");
1019 static const arg_def_t profile = ARG_DEF(NULL, "profile", 1,
1020 "Bitstream profile number to use");
1021 static const arg_def_t width = ARG_DEF("w", "width", 1,
1023 static const arg_def_t height = ARG_DEF("h", "height", 1,
1025 static const struct arg_enum_list stereo_mode_enum[] = {
1026 {"mono" , STEREO_FORMAT_MONO},
1027 {"left-right", STEREO_FORMAT_LEFT_RIGHT},
1028 {"bottom-top", STEREO_FORMAT_BOTTOM_TOP},
1029 {"top-bottom", STEREO_FORMAT_TOP_BOTTOM},
1030 {"right-left", STEREO_FORMAT_RIGHT_LEFT},
1033 static const arg_def_t stereo_mode = ARG_DEF_ENUM(NULL, "stereo-mode", 1,
1034 "Stereo 3D video format", stereo_mode_enum);
1035 static const arg_def_t timebase = ARG_DEF(NULL, "timebase", 1,
1036 "Output timestamp precision (fractional seconds)");
1037 static const arg_def_t error_resilient = ARG_DEF(NULL, "error-resilient", 1,
1038 "Enable error resiliency features");
1039 static const arg_def_t lag_in_frames = ARG_DEF(NULL, "lag-in-frames", 1,
1040 "Max number of frames to lag");
1042 static const arg_def_t *global_args[] =
1044 &use_yv12, &use_i420, &usage, &threads, &profile,
1045 &width, &height, &stereo_mode, &timebase, &framerate, &error_resilient,
1046 &lag_in_frames, NULL
1049 static const arg_def_t dropframe_thresh = ARG_DEF(NULL, "drop-frame", 1,
1050 "Temporal resampling threshold (buf %)");
1051 static const arg_def_t resize_allowed = ARG_DEF(NULL, "resize-allowed", 1,
1052 "Spatial resampling enabled (bool)");
1053 static const arg_def_t resize_up_thresh = ARG_DEF(NULL, "resize-up", 1,
1054 "Upscale threshold (buf %)");
1055 static const arg_def_t resize_down_thresh = ARG_DEF(NULL, "resize-down", 1,
1056 "Downscale threshold (buf %)");
1057 static const struct arg_enum_list end_usage_enum[] = {
1063 static const arg_def_t end_usage = ARG_DEF_ENUM(NULL, "end-usage", 1,
1064 "Rate control mode", end_usage_enum);
1065 static const arg_def_t target_bitrate = ARG_DEF(NULL, "target-bitrate", 1,
1067 static const arg_def_t min_quantizer = ARG_DEF(NULL, "min-q", 1,
1068 "Minimum (best) quantizer");
1069 static const arg_def_t max_quantizer = ARG_DEF(NULL, "max-q", 1,
1070 "Maximum (worst) quantizer");
1071 static const arg_def_t undershoot_pct = ARG_DEF(NULL, "undershoot-pct", 1,
1072 "Datarate undershoot (min) target (%)");
1073 static const arg_def_t overshoot_pct = ARG_DEF(NULL, "overshoot-pct", 1,
1074 "Datarate overshoot (max) target (%)");
1075 static const arg_def_t buf_sz = ARG_DEF(NULL, "buf-sz", 1,
1076 "Client buffer size (ms)");
1077 static const arg_def_t buf_initial_sz = ARG_DEF(NULL, "buf-initial-sz", 1,
1078 "Client initial buffer size (ms)");
1079 static const arg_def_t buf_optimal_sz = ARG_DEF(NULL, "buf-optimal-sz", 1,
1080 "Client optimal buffer size (ms)");
1081 static const arg_def_t *rc_args[] =
1083 &dropframe_thresh, &resize_allowed, &resize_up_thresh, &resize_down_thresh,
1084 &end_usage, &target_bitrate, &min_quantizer, &max_quantizer,
1085 &undershoot_pct, &overshoot_pct, &buf_sz, &buf_initial_sz, &buf_optimal_sz,
1090 static const arg_def_t bias_pct = ARG_DEF(NULL, "bias-pct", 1,
1091 "CBR/VBR bias (0=CBR, 100=VBR)");
1092 static const arg_def_t minsection_pct = ARG_DEF(NULL, "minsection-pct", 1,
1093 "GOP min bitrate (% of target)");
1094 static const arg_def_t maxsection_pct = ARG_DEF(NULL, "maxsection-pct", 1,
1095 "GOP max bitrate (% of target)");
1096 static const arg_def_t *rc_twopass_args[] =
1098 &bias_pct, &minsection_pct, &maxsection_pct, NULL
1102 static const arg_def_t kf_min_dist = ARG_DEF(NULL, "kf-min-dist", 1,
1103 "Minimum keyframe interval (frames)");
1104 static const arg_def_t kf_max_dist = ARG_DEF(NULL, "kf-max-dist", 1,
1105 "Maximum keyframe interval (frames)");
1106 static const arg_def_t kf_disabled = ARG_DEF(NULL, "disable-kf", 0,
1107 "Disable keyframe placement");
1108 static const arg_def_t *kf_args[] =
1110 &kf_min_dist, &kf_max_dist, &kf_disabled, NULL
1114 #if CONFIG_VP8_ENCODER
1115 static const arg_def_t noise_sens = ARG_DEF(NULL, "noise-sensitivity", 1,
1116 "Noise sensitivity (frames to blur)");
1117 static const arg_def_t sharpness = ARG_DEF(NULL, "sharpness", 1,
1118 "Filter sharpness (0-7)");
1119 static const arg_def_t static_thresh = ARG_DEF(NULL, "static-thresh", 1,
1120 "Motion detection threshold");
1123 #if CONFIG_VP8_ENCODER
1124 static const arg_def_t cpu_used = ARG_DEF(NULL, "cpu-used", 1,
1125 "CPU Used (-16..16)");
1129 #if CONFIG_VP8_ENCODER
1130 static const arg_def_t token_parts = ARG_DEF(NULL, "token-parts", 1,
1131 "Number of token partitions to use, log2");
1132 static const arg_def_t auto_altref = ARG_DEF(NULL, "auto-alt-ref", 1,
1133 "Enable automatic alt reference frames");
1134 static const arg_def_t arnr_maxframes = ARG_DEF(NULL, "arnr-maxframes", 1,
1135 "AltRef Max Frames");
1136 static const arg_def_t arnr_strength = ARG_DEF(NULL, "arnr-strength", 1,
1138 static const arg_def_t arnr_type = ARG_DEF(NULL, "arnr-type", 1,
1140 static const struct arg_enum_list tuning_enum[] = {
1141 {"psnr", VP8_TUNE_PSNR},
1142 {"ssim", VP8_TUNE_SSIM},
1145 static const arg_def_t tune_ssim = ARG_DEF_ENUM(NULL, "tune", 1,
1146 "Material to favor", tuning_enum);
1147 static const arg_def_t cq_level = ARG_DEF(NULL, "cq-level", 1,
1148 "Constrained Quality Level");
1149 static const arg_def_t max_intra_rate_pct = ARG_DEF(NULL, "max-intra-rate", 1,
1150 "Max I-frame bitrate (pct)");
1152 static const arg_def_t *vp8_args[] =
1154 &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh,
1155 &token_parts, &arnr_maxframes, &arnr_strength, &arnr_type,
1156 &tune_ssim, &cq_level, &max_intra_rate_pct, NULL
1158 static const int vp8_arg_ctrl_map[] =
1160 VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF,
1161 VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD,
1162 VP8E_SET_TOKEN_PARTITIONS,
1163 VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH , VP8E_SET_ARNR_TYPE,
1164 VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, VP8E_SET_MAX_INTRA_BITRATE_PCT, 0
1168 static const arg_def_t *no_args[] = { NULL };
1170 static void usage_exit()
1174 fprintf(stderr, "Usage: %s <options> -o dst_filename src_filename \n",
1177 fprintf(stderr, "\nOptions:\n");
1178 arg_show_usage(stdout, main_args);
1179 fprintf(stderr, "\nEncoder Global Options:\n");
1180 arg_show_usage(stdout, global_args);
1181 fprintf(stderr, "\nRate Control Options:\n");
1182 arg_show_usage(stdout, rc_args);
1183 fprintf(stderr, "\nTwopass Rate Control Options:\n");
1184 arg_show_usage(stdout, rc_twopass_args);
1185 fprintf(stderr, "\nKeyframe Placement Options:\n");
1186 arg_show_usage(stdout, kf_args);
1187 #if CONFIG_VP8_ENCODER
1188 fprintf(stderr, "\nVP8 Specific Options:\n");
1189 arg_show_usage(stdout, vp8_args);
1191 fprintf(stderr, "\nStream timebase (--timebase):\n"
1192 " The desired precision of timestamps in the output, expressed\n"
1193 " in fractional seconds. Default is 1/1000.\n");
1194 fprintf(stderr, "\n"
1195 "Included encoders:\n"
1198 for (i = 0; i < sizeof(codecs) / sizeof(codecs[0]); i++)
1199 fprintf(stderr, " %-6s - %s\n",
1201 vpx_codec_iface_name(codecs[i].iface));
1207 #define HIST_BAR_MAX 40
1210 int low, high, count;
1214 static int merge_hist_buckets(struct hist_bucket *bucket,
1218 int small_bucket = 0, merge_bucket = INT_MAX, big_bucket=0;
1219 int buckets = *buckets_;
1222 /* Find the extrema for this list of buckets */
1223 big_bucket = small_bucket = 0;
1224 for(i=0; i < buckets; i++)
1226 if(bucket[i].count < bucket[small_bucket].count)
1228 if(bucket[i].count > bucket[big_bucket].count)
1232 /* If we have too many buckets, merge the smallest with an adjacent
1235 while(buckets > max_buckets)
1237 int last_bucket = buckets - 1;
1239 /* merge the small bucket with an adjacent one. */
1240 if(small_bucket == 0)
1242 else if(small_bucket == last_bucket)
1243 merge_bucket = last_bucket - 1;
1244 else if(bucket[small_bucket - 1].count < bucket[small_bucket + 1].count)
1245 merge_bucket = small_bucket - 1;
1247 merge_bucket = small_bucket + 1;
1249 assert(abs(merge_bucket - small_bucket) <= 1);
1250 assert(small_bucket < buckets);
1251 assert(big_bucket < buckets);
1252 assert(merge_bucket < buckets);
1254 if(merge_bucket < small_bucket)
1256 bucket[merge_bucket].high = bucket[small_bucket].high;
1257 bucket[merge_bucket].count += bucket[small_bucket].count;
1261 bucket[small_bucket].high = bucket[merge_bucket].high;
1262 bucket[small_bucket].count += bucket[merge_bucket].count;
1263 merge_bucket = small_bucket;
1266 assert(bucket[merge_bucket].low != bucket[merge_bucket].high);
1270 /* Remove the merge_bucket from the list, and find the new small
1271 * and big buckets while we're at it
1273 big_bucket = small_bucket = 0;
1274 for(i=0; i < buckets; i++)
1276 if(i > merge_bucket)
1277 bucket[i] = bucket[i+1];
1279 if(bucket[i].count < bucket[small_bucket].count)
1281 if(bucket[i].count > bucket[big_bucket].count)
1287 *buckets_ = buckets;
1288 return bucket[big_bucket].count;
1292 static void show_histogram(const struct hist_bucket *bucket,
1297 const char *pat1, *pat2;
1300 switch((int)(log(bucket[buckets-1].high)/log(10))+1)
1328 pat1 = "%12d %10s: ";
1329 pat2 = "%12d-%10d: ";
1333 for(i=0; i<buckets; i++)
1339 pct = 100.0 * (float)bucket[i].count / (float)total;
1340 len = HIST_BAR_MAX * bucket[i].count / scale;
1343 assert(len <= HIST_BAR_MAX);
1345 if(bucket[i].low == bucket[i].high)
1346 fprintf(stderr, pat1, bucket[i].low, "");
1348 fprintf(stderr, pat2, bucket[i].low, bucket[i].high);
1350 for(j=0; j<HIST_BAR_MAX; j++)
1351 fprintf(stderr, j<len?"=":" ");
1352 fprintf(stderr, "\t%5d (%6.2f%%)\n",bucket[i].count,pct);
1357 static void show_q_histogram(const int counts[64], int max_buckets)
1359 struct hist_bucket bucket[64];
1370 bucket[buckets].low = bucket[buckets].high = i;
1371 bucket[buckets].count = counts[i];
1377 fprintf(stderr, "\nQuantizer Selection:\n");
1378 scale = merge_hist_buckets(bucket, &buckets, max_buckets);
1379 show_histogram(bucket, buckets, total, scale);
1383 #define RATE_BINS (100)
1390 struct hist_bucket bucket[RATE_BINS];
1395 static void init_rate_histogram(struct rate_hist *hist,
1396 const vpx_codec_enc_cfg_t *cfg,
1397 const vpx_rational_t *fps)
1401 /* Determine the number of samples in the buffer. Use the file's framerate
1402 * to determine the number of frames in rc_buf_sz milliseconds, with an
1403 * adjustment (5/4) to account for alt-refs
1405 hist->samples = cfg->rc_buf_sz * 5 / 4 * fps->num / fps->den / 1000;
1407 /* prevent division by zero */
1408 if (hist->samples == 0)
1411 hist->pts = calloc(hist->samples, sizeof(*hist->pts));
1412 hist->sz = calloc(hist->samples, sizeof(*hist->sz));
1413 for(i=0; i<RATE_BINS; i++)
1415 hist->bucket[i].low = INT_MAX;
1416 hist->bucket[i].high = 0;
1417 hist->bucket[i].count = 0;
1422 static void destroy_rate_histogram(struct rate_hist *hist)
1429 static void update_rate_histogram(struct rate_hist *hist,
1430 const vpx_codec_enc_cfg_t *cfg,
1431 const vpx_codec_cx_pkt_t *pkt)
1434 int64_t now, then, sum_sz = 0, avg_bitrate;
1436 now = pkt->data.frame.pts * 1000
1437 * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
1439 idx = hist->frames++ % hist->samples;
1440 hist->pts[idx] = now;
1441 hist->sz[idx] = pkt->data.frame.sz;
1443 if(now < cfg->rc_buf_initial_sz)
1448 /* Sum the size over the past rc_buf_sz ms */
1449 for(i = hist->frames; i > 0 && hist->frames - i < hist->samples; i--)
1451 int i_idx = (i-1) % hist->samples;
1453 then = hist->pts[i_idx];
1454 if(now - then > cfg->rc_buf_sz)
1456 sum_sz += hist->sz[i_idx];
1462 avg_bitrate = sum_sz * 8 * 1000 / (now - then);
1463 idx = avg_bitrate * (RATE_BINS/2) / (cfg->rc_target_bitrate * 1000);
1466 if(idx > RATE_BINS-1)
1468 if(hist->bucket[idx].low > avg_bitrate)
1469 hist->bucket[idx].low = avg_bitrate;
1470 if(hist->bucket[idx].high < avg_bitrate)
1471 hist->bucket[idx].high = avg_bitrate;
1472 hist->bucket[idx].count++;
1477 static void show_rate_histogram(struct rate_hist *hist,
1478 const vpx_codec_enc_cfg_t *cfg,
1484 for(i = 0; i < RATE_BINS; i++)
1486 if(hist->bucket[i].low == INT_MAX)
1488 hist->bucket[buckets++] = hist->bucket[i];
1491 fprintf(stderr, "\nRate (over %dms window):\n", cfg->rc_buf_sz);
1492 scale = merge_hist_buckets(hist->bucket, &buckets, max_buckets);
1493 show_histogram(hist->bucket, buckets, hist->total, scale);
1496 #define NELEMENTS(x) (sizeof(x)/sizeof(x[0]))
1497 #define ARG_CTRL_CNT_MAX NELEMENTS(vp8_arg_ctrl_map)
1500 /* Configuration elements common to all streams */
1501 struct global_config
1503 const struct codec_item *codec;
1513 struct vpx_rational framerate;
1516 int show_q_hist_buckets;
1517 int show_rate_hist_buckets;
1521 /* Per-stream configuration */
1522 struct stream_config
1524 struct vpx_codec_enc_cfg cfg;
1526 const char *stats_fn;
1527 stereo_format_t stereo_fmt;
1528 int arg_ctrls[ARG_CTRL_CNT_MAX][2];
1531 int have_kf_max_dist;
1538 struct stream_state *next;
1539 struct stream_config config;
1541 struct rate_hist rate_hist;
1544 uint64_t psnr_sse_total;
1545 uint64_t psnr_samples_total;
1546 double psnr_totals[4];
1549 vpx_codec_ctx_t encoder;
1550 unsigned int frames_out;
1557 void validate_positive_rational(const char *msg,
1558 struct vpx_rational *rat)
1567 die("Error: %s must be positive\n", msg);
1570 die("Error: %s has zero denominator\n", msg);
1574 static void parse_global_config(struct global_config *global, char **argv)
1576 char **argi, **argj;
1579 /* Initialize default parameters */
1580 memset(global, 0, sizeof(*global));
1581 global->codec = codecs;
1583 global->use_i420 = 1;
1585 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
1589 if (arg_match(&arg, &codecarg, argi))
1593 for (j = 0; j < sizeof(codecs) / sizeof(codecs[0]); j++)
1594 if (!strcmp(codecs[j].name, arg.val))
1598 global->codec = codecs + k;
1600 die("Error: Unrecognized argument (%s) to --codec\n",
1604 else if (arg_match(&arg, &passes, argi))
1606 global->passes = arg_parse_uint(&arg);
1608 if (global->passes < 1 || global->passes > 2)
1609 die("Error: Invalid number of passes (%d)\n", global->passes);
1611 else if (arg_match(&arg, &pass_arg, argi))
1613 global->pass = arg_parse_uint(&arg);
1615 if (global->pass < 1 || global->pass > 2)
1616 die("Error: Invalid pass selected (%d)\n",
1619 else if (arg_match(&arg, &usage, argi))
1620 global->usage = arg_parse_uint(&arg);
1621 else if (arg_match(&arg, &deadline, argi))
1622 global->deadline = arg_parse_uint(&arg);
1623 else if (arg_match(&arg, &best_dl, argi))
1624 global->deadline = VPX_DL_BEST_QUALITY;
1625 else if (arg_match(&arg, &good_dl, argi))
1626 global->deadline = VPX_DL_GOOD_QUALITY;
1627 else if (arg_match(&arg, &rt_dl, argi))
1628 global->deadline = VPX_DL_REALTIME;
1629 else if (arg_match(&arg, &use_yv12, argi))
1630 global->use_i420 = 0;
1631 else if (arg_match(&arg, &use_i420, argi))
1632 global->use_i420 = 1;
1633 else if (arg_match(&arg, &verbosearg, argi))
1634 global->verbose = 1;
1635 else if (arg_match(&arg, &limit, argi))
1636 global->limit = arg_parse_uint(&arg);
1637 else if (arg_match(&arg, &psnrarg, argi))
1638 global->show_psnr = 1;
1639 else if (arg_match(&arg, &framerate, argi))
1641 global->framerate = arg_parse_rational(&arg);
1642 validate_positive_rational(arg.name, &global->framerate);
1643 global->have_framerate = 1;
1645 else if (arg_match(&arg,&out_part, argi))
1646 global->out_part = 1;
1647 else if (arg_match(&arg, &debugmode, argi))
1649 else if (arg_match(&arg, &q_hist_n, argi))
1650 global->show_q_hist_buckets = arg_parse_uint(&arg);
1651 else if (arg_match(&arg, &rate_hist_n, argi))
1652 global->show_rate_hist_buckets = arg_parse_uint(&arg);
1657 /* Validate global config */
1661 /* DWIM: Assume the user meant passes=2 if pass=2 is specified */
1662 if (global->pass > global->passes)
1664 warn("Assuming --pass=%d implies --passes=%d\n",
1665 global->pass, global->pass);
1666 global->passes = global->pass;
1672 void open_input_file(struct input_state *input)
1674 unsigned int fourcc;
1676 /* Parse certain options from the input file, if possible */
1677 input->file = strcmp(input->fn, "-") ? fopen(input->fn, "rb")
1678 : set_binary_mode(stdin);
1681 fatal("Failed to open input file");
1683 /* For RAW input sources, these bytes will applied on the first frame
1686 input->detect.buf_read = fread(input->detect.buf, 1, 4, input->file);
1687 input->detect.position = 0;
1689 if (input->detect.buf_read == 4
1690 && file_is_y4m(input->file, &input->y4m, input->detect.buf))
1692 if (y4m_input_open(&input->y4m, input->file, input->detect.buf, 4) >= 0)
1694 input->file_type = FILE_TYPE_Y4M;
1695 input->w = input->y4m.pic_w;
1696 input->h = input->y4m.pic_h;
1697 input->framerate.num = input->y4m.fps_n;
1698 input->framerate.den = input->y4m.fps_d;
1699 input->use_i420 = 0;
1702 fatal("Unsupported Y4M stream.");
1704 else if (input->detect.buf_read == 4 && file_is_ivf(input, &fourcc))
1706 input->file_type = FILE_TYPE_IVF;
1710 input->use_i420 = 0;
1713 input->use_i420 = 1;
1716 fatal("Unsupported fourcc (%08x) in IVF", fourcc);
1721 input->file_type = FILE_TYPE_RAW;
1726 static void close_input_file(struct input_state *input)
1728 fclose(input->file);
1729 if (input->file_type == FILE_TYPE_Y4M)
1730 y4m_input_close(&input->y4m);
1733 static struct stream_state *new_stream(struct global_config *global,
1734 struct stream_state *prev)
1736 struct stream_state *stream;
1738 stream = calloc(1, sizeof(*stream));
1740 fatal("Failed to allocate new stream.");
1743 memcpy(stream, prev, sizeof(*stream));
1745 prev->next = stream;
1749 vpx_codec_err_t res;
1751 /* Populate encoder configuration */
1752 res = vpx_codec_enc_config_default(global->codec->iface,
1753 &stream->config.cfg,
1756 fatal("Failed to get config: %s\n", vpx_codec_err_to_string(res));
1758 /* Change the default timebase to a high enough value so that the
1759 * encoder will always create strictly increasing timestamps.
1761 stream->config.cfg.g_timebase.den = 1000;
1763 /* Never use the library's default resolution, require it be parsed
1764 * from the file or set on the command line.
1766 stream->config.cfg.g_w = 0;
1767 stream->config.cfg.g_h = 0;
1769 /* Initialize remaining stream parameters */
1770 stream->config.stereo_fmt = STEREO_FORMAT_MONO;
1771 stream->config.write_webm = 1;
1772 stream->ebml.last_pts_ms = -1;
1774 /* Allows removal of the application version from the EBML tags */
1775 stream->ebml.debug = global->debug;
1778 /* Output files must be specified for each stream */
1779 stream->config.out_fn = NULL;
1781 stream->next = NULL;
1786 static int parse_stream_params(struct global_config *global,
1787 struct stream_state *stream,
1790 char **argi, **argj;
1792 static const arg_def_t **ctrl_args = no_args;
1793 static const int *ctrl_args_map = NULL;
1794 struct stream_config *config = &stream->config;
1795 int eos_mark_found = 0;
1797 /* Handle codec specific options */
1798 if (global->codec->iface == &vpx_codec_vp8_cx_algo)
1800 ctrl_args = vp8_args;
1801 ctrl_args_map = vp8_arg_ctrl_map;
1804 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
1808 /* Once we've found an end-of-stream marker (--) we want to continue
1809 * shifting arguments but not consuming them.
1816 else if (!strcmp(*argj, "--"))
1823 else if (arg_match(&arg, &outputfile, argi))
1824 config->out_fn = arg.val;
1825 else if (arg_match(&arg, &fpf_name, argi))
1826 config->stats_fn = arg.val;
1827 else if (arg_match(&arg, &use_ivf, argi))
1828 config->write_webm = 0;
1829 else if (arg_match(&arg, &threads, argi))
1830 config->cfg.g_threads = arg_parse_uint(&arg);
1831 else if (arg_match(&arg, &profile, argi))
1832 config->cfg.g_profile = arg_parse_uint(&arg);
1833 else if (arg_match(&arg, &width, argi))
1834 config->cfg.g_w = arg_parse_uint(&arg);
1835 else if (arg_match(&arg, &height, argi))
1836 config->cfg.g_h = arg_parse_uint(&arg);
1837 else if (arg_match(&arg, &stereo_mode, argi))
1838 config->stereo_fmt = arg_parse_enum_or_int(&arg);
1839 else if (arg_match(&arg, &timebase, argi))
1841 config->cfg.g_timebase = arg_parse_rational(&arg);
1842 validate_positive_rational(arg.name, &config->cfg.g_timebase);
1844 else if (arg_match(&arg, &error_resilient, argi))
1845 config->cfg.g_error_resilient = arg_parse_uint(&arg);
1846 else if (arg_match(&arg, &lag_in_frames, argi))
1847 config->cfg.g_lag_in_frames = arg_parse_uint(&arg);
1848 else if (arg_match(&arg, &dropframe_thresh, argi))
1849 config->cfg.rc_dropframe_thresh = arg_parse_uint(&arg);
1850 else if (arg_match(&arg, &resize_allowed, argi))
1851 config->cfg.rc_resize_allowed = arg_parse_uint(&arg);
1852 else if (arg_match(&arg, &resize_up_thresh, argi))
1853 config->cfg.rc_resize_up_thresh = arg_parse_uint(&arg);
1854 else if (arg_match(&arg, &resize_down_thresh, argi))
1855 config->cfg.rc_resize_down_thresh = arg_parse_uint(&arg);
1856 else if (arg_match(&arg, &end_usage, argi))
1857 config->cfg.rc_end_usage = arg_parse_enum_or_int(&arg);
1858 else if (arg_match(&arg, &target_bitrate, argi))
1859 config->cfg.rc_target_bitrate = arg_parse_uint(&arg);
1860 else if (arg_match(&arg, &min_quantizer, argi))
1861 config->cfg.rc_min_quantizer = arg_parse_uint(&arg);
1862 else if (arg_match(&arg, &max_quantizer, argi))
1863 config->cfg.rc_max_quantizer = arg_parse_uint(&arg);
1864 else if (arg_match(&arg, &undershoot_pct, argi))
1865 config->cfg.rc_undershoot_pct = arg_parse_uint(&arg);
1866 else if (arg_match(&arg, &overshoot_pct, argi))
1867 config->cfg.rc_overshoot_pct = arg_parse_uint(&arg);
1868 else if (arg_match(&arg, &buf_sz, argi))
1869 config->cfg.rc_buf_sz = arg_parse_uint(&arg);
1870 else if (arg_match(&arg, &buf_initial_sz, argi))
1871 config->cfg.rc_buf_initial_sz = arg_parse_uint(&arg);
1872 else if (arg_match(&arg, &buf_optimal_sz, argi))
1873 config->cfg.rc_buf_optimal_sz = arg_parse_uint(&arg);
1874 else if (arg_match(&arg, &bias_pct, argi))
1876 config->cfg.rc_2pass_vbr_bias_pct = arg_parse_uint(&arg);
1878 if (global->passes < 2)
1879 warn("option %s ignored in one-pass mode.\n", arg.name);
1881 else if (arg_match(&arg, &minsection_pct, argi))
1883 config->cfg.rc_2pass_vbr_minsection_pct = arg_parse_uint(&arg);
1885 if (global->passes < 2)
1886 warn("option %s ignored in one-pass mode.\n", arg.name);
1888 else if (arg_match(&arg, &maxsection_pct, argi))
1890 config->cfg.rc_2pass_vbr_maxsection_pct = arg_parse_uint(&arg);
1892 if (global->passes < 2)
1893 warn("option %s ignored in one-pass mode.\n", arg.name);
1895 else if (arg_match(&arg, &kf_min_dist, argi))
1896 config->cfg.kf_min_dist = arg_parse_uint(&arg);
1897 else if (arg_match(&arg, &kf_max_dist, argi))
1899 config->cfg.kf_max_dist = arg_parse_uint(&arg);
1900 config->have_kf_max_dist = 1;
1902 else if (arg_match(&arg, &kf_disabled, argi))
1903 config->cfg.kf_mode = VPX_KF_DISABLED;
1908 for (i = 0; ctrl_args[i]; i++)
1910 if (arg_match(&arg, ctrl_args[i], argi))
1915 /* Point either to the next free element or the first
1916 * instance of this control.
1918 for(j=0; j<config->arg_ctrl_cnt; j++)
1919 if(config->arg_ctrls[j][0] == ctrl_args_map[i])
1923 assert(j < ARG_CTRL_CNT_MAX);
1924 if (j < ARG_CTRL_CNT_MAX)
1926 config->arg_ctrls[j][0] = ctrl_args_map[i];
1927 config->arg_ctrls[j][1] = arg_parse_enum_or_int(&arg);
1928 if(j == config->arg_ctrl_cnt)
1929 config->arg_ctrl_cnt++;
1940 return eos_mark_found;
1944 #define FOREACH_STREAM(func)\
1947 struct stream_state *stream;\
1949 for(stream = streams; stream; stream = stream->next)\
1954 static void validate_stream_config(struct stream_state *stream)
1956 struct stream_state *streami;
1958 if(!stream->config.cfg.g_w || !stream->config.cfg.g_h)
1959 fatal("Stream %d: Specify stream dimensions with --width (-w) "
1960 " and --height (-h)", stream->index);
1962 for(streami = stream; streami; streami = streami->next)
1964 /* All streams require output files */
1965 if(!streami->config.out_fn)
1966 fatal("Stream %d: Output file is required (specify with -o)",
1969 /* Check for two streams outputting to the same file */
1970 if(streami != stream)
1972 const char *a = stream->config.out_fn;
1973 const char *b = streami->config.out_fn;
1974 if(!strcmp(a,b) && strcmp(a, "/dev/null") && strcmp(a, ":nul"))
1975 fatal("Stream %d: duplicate output file (from stream %d)",
1976 streami->index, stream->index);
1979 /* Check for two streams sharing a stats file. */
1980 if(streami != stream)
1982 const char *a = stream->config.stats_fn;
1983 const char *b = streami->config.stats_fn;
1984 if(a && b && !strcmp(a,b))
1985 fatal("Stream %d: duplicate stats file (from stream %d)",
1986 streami->index, stream->index);
1992 static void set_stream_dimensions(struct stream_state *stream,
1996 if ((stream->config.cfg.g_w && stream->config.cfg.g_w != w)
1997 ||(stream->config.cfg.g_h && stream->config.cfg.g_h != h))
1998 fatal("Stream %d: Resizing not yet supported", stream->index);
1999 stream->config.cfg.g_w = w;
2000 stream->config.cfg.g_h = h;
2004 static void set_default_kf_interval(struct stream_state *stream,
2005 struct global_config *global)
2007 /* Use a max keyframe interval of 5 seconds, if none was
2008 * specified on the command line.
2010 if (!stream->config.have_kf_max_dist)
2012 double framerate = (double)global->framerate.num/global->framerate.den;
2013 if (framerate > 0.0)
2014 stream->config.cfg.kf_max_dist = 5.0*framerate;
2019 static void show_stream_config(struct stream_state *stream,
2020 struct global_config *global,
2021 struct input_state *input)
2024 #define SHOW(field) \
2025 fprintf(stderr, " %-28s = %d\n", #field, stream->config.cfg.field)
2027 if(stream->index == 0)
2029 fprintf(stderr, "Codec: %s\n",
2030 vpx_codec_iface_name(global->codec->iface));
2031 fprintf(stderr, "Source file: %s Format: %s\n", input->fn,
2032 input->use_i420 ? "I420" : "YV12");
2034 if(stream->next || stream->index)
2035 fprintf(stderr, "\nStream Index: %d\n", stream->index);
2036 fprintf(stderr, "Destination file: %s\n", stream->config.out_fn);
2037 fprintf(stderr, "Encoder parameters:\n");
2044 SHOW(g_timebase.num);
2045 SHOW(g_timebase.den);
2046 SHOW(g_error_resilient);
2048 SHOW(g_lag_in_frames);
2049 SHOW(rc_dropframe_thresh);
2050 SHOW(rc_resize_allowed);
2051 SHOW(rc_resize_up_thresh);
2052 SHOW(rc_resize_down_thresh);
2054 SHOW(rc_target_bitrate);
2055 SHOW(rc_min_quantizer);
2056 SHOW(rc_max_quantizer);
2057 SHOW(rc_undershoot_pct);
2058 SHOW(rc_overshoot_pct);
2060 SHOW(rc_buf_initial_sz);
2061 SHOW(rc_buf_optimal_sz);
2062 SHOW(rc_2pass_vbr_bias_pct);
2063 SHOW(rc_2pass_vbr_minsection_pct);
2064 SHOW(rc_2pass_vbr_maxsection_pct);
2071 static void open_output_file(struct stream_state *stream,
2072 struct global_config *global)
2074 const char *fn = stream->config.out_fn;
2076 stream->file = strcmp(fn, "-") ? fopen(fn, "wb") : set_binary_mode(stdout);
2079 fatal("Failed to open output file");
2081 if(stream->config.write_webm && fseek(stream->file, 0, SEEK_CUR))
2082 fatal("WebM output to pipes not supported.");
2084 if(stream->config.write_webm)
2086 stream->ebml.stream = stream->file;
2087 write_webm_file_header(&stream->ebml, &stream->config.cfg,
2089 stream->config.stereo_fmt);
2092 write_ivf_file_header(stream->file, &stream->config.cfg,
2093 global->codec->fourcc, 0);
2097 static void close_output_file(struct stream_state *stream,
2098 unsigned int fourcc)
2100 if(stream->config.write_webm)
2102 write_webm_file_footer(&stream->ebml, stream->hash);
2103 free(stream->ebml.cue_list);
2104 stream->ebml.cue_list = NULL;
2108 if (!fseek(stream->file, 0, SEEK_SET))
2109 write_ivf_file_header(stream->file, &stream->config.cfg,
2111 stream->frames_out);
2114 fclose(stream->file);
2118 static void setup_pass(struct stream_state *stream,
2119 struct global_config *global,
2122 if (stream->config.stats_fn)
2124 if (!stats_open_file(&stream->stats, stream->config.stats_fn,
2126 fatal("Failed to open statistics store");
2130 if (!stats_open_mem(&stream->stats, pass))
2131 fatal("Failed to open statistics store");
2134 stream->config.cfg.g_pass = global->passes == 2
2135 ? pass ? VPX_RC_LAST_PASS : VPX_RC_FIRST_PASS
2138 stream->config.cfg.rc_twopass_stats_in = stats_get(&stream->stats);
2140 stream->cx_time = 0;
2142 stream->frames_out = 0;
2146 static void initialize_encoder(struct stream_state *stream,
2147 struct global_config *global)
2152 flags |= global->show_psnr ? VPX_CODEC_USE_PSNR : 0;
2153 flags |= global->out_part ? VPX_CODEC_USE_OUTPUT_PARTITION : 0;
2155 /* Construct Encoder Context */
2156 vpx_codec_enc_init(&stream->encoder, global->codec->iface,
2157 &stream->config.cfg, flags);
2158 ctx_exit_on_error(&stream->encoder, "Failed to initialize encoder");
2160 /* Note that we bypass the vpx_codec_control wrapper macro because
2161 * we're being clever to store the control IDs in an array. Real
2162 * applications will want to make use of the enumerations directly
2164 for (i = 0; i < stream->config.arg_ctrl_cnt; i++)
2166 int ctrl = stream->config.arg_ctrls[i][0];
2167 int value = stream->config.arg_ctrls[i][1];
2168 if (vpx_codec_control_(&stream->encoder, ctrl, value))
2169 fprintf(stderr, "Error: Tried to set control %d = %d\n",
2172 ctx_exit_on_error(&stream->encoder, "Failed to control codec");
2177 static void encode_frame(struct stream_state *stream,
2178 struct global_config *global,
2179 struct vpx_image *img,
2180 unsigned int frames_in)
2182 vpx_codec_pts_t frame_start, next_frame_start;
2183 struct vpx_codec_enc_cfg *cfg = &stream->config.cfg;
2184 struct vpx_usec_timer timer;
2186 frame_start = (cfg->g_timebase.den * (int64_t)(frames_in - 1)
2187 * global->framerate.den)
2188 / cfg->g_timebase.num / global->framerate.num;
2189 next_frame_start = (cfg->g_timebase.den * (int64_t)(frames_in)
2190 * global->framerate.den)
2191 / cfg->g_timebase.num / global->framerate.num;
2192 vpx_usec_timer_start(&timer);
2193 vpx_codec_encode(&stream->encoder, img, frame_start,
2194 next_frame_start - frame_start,
2195 0, global->deadline);
2196 vpx_usec_timer_mark(&timer);
2197 stream->cx_time += vpx_usec_timer_elapsed(&timer);
2198 ctx_exit_on_error(&stream->encoder, "Stream %d: Failed to encode frame",
2203 static void update_quantizer_histogram(struct stream_state *stream)
2205 if(stream->config.cfg.g_pass != VPX_RC_FIRST_PASS)
2209 vpx_codec_control(&stream->encoder, VP8E_GET_LAST_QUANTIZER_64, &q);
2210 ctx_exit_on_error(&stream->encoder, "Failed to read quantizer");
2211 stream->counts[q]++;
2216 static void get_cx_data(struct stream_state *stream,
2217 struct global_config *global,
2220 const vpx_codec_cx_pkt_t *pkt;
2221 const struct vpx_codec_enc_cfg *cfg = &stream->config.cfg;
2222 vpx_codec_iter_t iter = NULL;
2224 while ((pkt = vpx_codec_get_cx_data(&stream->encoder, &iter)))
2226 static size_t fsize = 0;
2227 static off_t ivf_header_pos = 0;
2233 case VPX_CODEC_CX_FRAME_PKT:
2234 if (!(pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT))
2236 stream->frames_out++;
2238 fprintf(stderr, " %6luF",
2239 (unsigned long)pkt->data.frame.sz);
2241 update_rate_histogram(&stream->rate_hist, cfg, pkt);
2242 if(stream->config.write_webm)
2244 /* Update the hash */
2245 if(!stream->ebml.debug)
2246 stream->hash = murmur(pkt->data.frame.buf,
2247 pkt->data.frame.sz, stream->hash);
2249 write_webm_block(&stream->ebml, cfg, pkt);
2253 if (pkt->data.frame.partition_id <= 0)
2255 ivf_header_pos = ftello(stream->file);
2256 fsize = pkt->data.frame.sz;
2258 write_ivf_frame_header(stream->file, pkt);
2262 fsize += pkt->data.frame.sz;
2264 if (!(pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT))
2266 off_t currpos = ftello(stream->file);
2267 fseeko(stream->file, ivf_header_pos, SEEK_SET);
2268 write_ivf_frame_size(stream->file, fsize);
2269 fseeko(stream->file, currpos, SEEK_SET);
2273 (void) fwrite(pkt->data.frame.buf, 1, pkt->data.frame.sz,
2276 stream->nbytes += pkt->data.raw.sz;
2278 case VPX_CODEC_STATS_PKT:
2279 stream->frames_out++;
2280 fprintf(stderr, " %6luS",
2281 (unsigned long)pkt->data.twopass_stats.sz);
2282 stats_write(&stream->stats,
2283 pkt->data.twopass_stats.buf,
2284 pkt->data.twopass_stats.sz);
2285 stream->nbytes += pkt->data.raw.sz;
2287 case VPX_CODEC_PSNR_PKT:
2289 if (global->show_psnr)
2293 stream->psnr_sse_total += pkt->data.psnr.sse[0];
2294 stream->psnr_samples_total += pkt->data.psnr.samples[0];
2295 for (i = 0; i < 4; i++)
2297 fprintf(stderr, "%.3f ", pkt->data.psnr.psnr[i]);
2298 stream->psnr_totals[i] += pkt->data.psnr.psnr[i];
2300 stream->psnr_count++;
2311 static void show_psnr(struct stream_state *stream)
2316 if (!stream->psnr_count)
2319 fprintf(stderr, "Stream %d PSNR (Overall/Avg/Y/U/V)", stream->index);
2320 ovpsnr = vp8_mse2psnr(stream->psnr_samples_total, 255.0,
2321 stream->psnr_sse_total);
2322 fprintf(stderr, " %.3f", ovpsnr);
2324 for (i = 0; i < 4; i++)
2326 fprintf(stderr, " %.3f", stream->psnr_totals[i]/stream->psnr_count);
2328 fprintf(stderr, "\n");
2332 float usec_to_fps(uint64_t usec, unsigned int frames)
2334 return usec > 0 ? (float)frames * 1000000.0 / (float)usec : 0;
2338 int main(int argc, const char **argv_)
2342 int frame_avail, got_data;
2344 struct input_state input = {0};
2345 struct global_config global;
2346 struct stream_state *streams = NULL;
2347 char **argv, **argi;
2348 unsigned long cx_time = 0;
2351 exec_name = argv_[0];
2356 /* Setup default input stream settings */
2357 input.framerate.num = 30;
2358 input.framerate.den = 1;
2361 /* First parse the global configuration values, because we want to apply
2362 * other parameters on top of the default configuration provided by the
2365 argv = argv_dup(argc - 1, argv_ + 1);
2366 parse_global_config(&global, argv);
2369 /* Now parse each stream's parameters. Using a local scope here
2370 * due to the use of 'stream' as loop variable in FOREACH_STREAM
2373 struct stream_state *stream = NULL;
2377 stream = new_stream(&global, stream);
2381 } while(parse_stream_params(&global, stream, argv));
2384 /* Check for unrecognized options */
2385 for (argi = argv; *argi; argi++)
2386 if (argi[0][0] == '-' && argi[0][1])
2387 die("Error: Unrecognized option %s\n", *argi);
2389 /* Handle non-option arguments */
2395 for (pass = global.pass ? global.pass - 1 : 0; pass < global.passes; pass++)
2399 open_input_file(&input);
2401 /* If the input file doesn't specify its w/h (raw files), try to get
2402 * the data from the first stream's configuration.
2404 if(!input.w || !input.h)
2406 if(stream->config.cfg.g_w && stream->config.cfg.g_h)
2408 input.w = stream->config.cfg.g_w;
2409 input.h = stream->config.cfg.g_h;
2414 /* Update stream configurations from the input file's parameters */
2415 FOREACH_STREAM(set_stream_dimensions(stream, input.w, input.h));
2416 FOREACH_STREAM(validate_stream_config(stream));
2418 /* Ensure that --passes and --pass are consistent. If --pass is set and
2419 * --passes=2, ensure --fpf was set.
2421 if (global.pass && global.passes == 2)
2423 if(!stream->config.stats_fn)
2424 die("Stream %d: Must specify --fpf when --pass=%d"
2425 " and --passes=2\n", stream->index, global.pass);
2429 /* Use the frame rate from the file only if none was specified
2430 * on the command-line.
2432 if (!global.have_framerate)
2433 global.framerate = input.framerate;
2435 FOREACH_STREAM(set_default_kf_interval(stream, &global));
2437 /* Show configuration */
2438 if (global.verbose && pass == 0)
2439 FOREACH_STREAM(show_stream_config(stream, &global, &input));
2441 if(pass == (global.pass ? global.pass - 1 : 0)) {
2442 if (input.file_type == FILE_TYPE_Y4M)
2443 /*The Y4M reader does its own allocation.
2444 Just initialize this here to avoid problems if we never read any
2446 memset(&raw, 0, sizeof(raw));
2449 input.use_i420 ? VPX_IMG_FMT_I420
2451 input.w, input.h, 32);
2453 FOREACH_STREAM(init_rate_histogram(&stream->rate_hist,
2454 &stream->config.cfg,
2455 &global.framerate));
2458 FOREACH_STREAM(open_output_file(stream, &global));
2459 FOREACH_STREAM(setup_pass(stream, &global, pass));
2460 FOREACH_STREAM(initialize_encoder(stream, &global));
2465 while (frame_avail || got_data)
2467 struct vpx_usec_timer timer;
2469 if (!global.limit || frames_in < global.limit)
2471 frame_avail = read_frame(&input, &raw);
2478 "\rPass %d/%d frame %4d/%-4d %7"PRId64"B \033[K",
2479 pass + 1, global.passes, frames_in,
2480 streams->frames_out, (int64_t)streams->nbytes);
2483 "\rPass %d/%d frame %4d %7lu %s (%.2f fps)\033[K",
2484 pass + 1, global.passes, frames_in,
2485 cx_time > 9999999 ? cx_time / 1000 : cx_time,
2486 cx_time > 9999999 ? "ms" : "us",
2487 usec_to_fps(cx_time, frames_in));
2493 vpx_usec_timer_start(&timer);
2494 FOREACH_STREAM(encode_frame(stream, &global,
2495 frame_avail ? &raw : NULL,
2497 vpx_usec_timer_mark(&timer);
2498 cx_time += vpx_usec_timer_elapsed(&timer);
2500 FOREACH_STREAM(update_quantizer_histogram(stream));
2503 FOREACH_STREAM(get_cx_data(stream, &global, &got_data));
2509 fprintf(stderr, "\n");
2511 FOREACH_STREAM(fprintf(
2513 "\rPass %d/%d frame %4d/%-4d %7"PRId64"B %7lub/f %7"PRId64"b/s"
2514 " %7"PRId64" %s (%.2f fps)\033[K\n", pass + 1,
2515 global.passes, frames_in, stream->frames_out, (int64_t)stream->nbytes,
2516 frames_in ? (unsigned long)(stream->nbytes * 8 / frames_in) : 0,
2517 frames_in ? (int64_t)stream->nbytes * 8
2518 * (int64_t)global.framerate.num / global.framerate.den
2521 stream->cx_time > 9999999 ? stream->cx_time / 1000 : stream->cx_time,
2522 stream->cx_time > 9999999 ? "ms" : "us",
2523 usec_to_fps(stream->cx_time, frames_in));
2526 if (global.show_psnr)
2527 FOREACH_STREAM(show_psnr(stream));
2529 FOREACH_STREAM(vpx_codec_destroy(&stream->encoder));
2531 close_input_file(&input);
2533 FOREACH_STREAM(close_output_file(stream, global.codec->fourcc));
2535 FOREACH_STREAM(stats_close(&stream->stats, global.passes-1));
2541 if (global.show_q_hist_buckets)
2542 FOREACH_STREAM(show_q_histogram(stream->counts,
2543 global.show_q_hist_buckets));
2545 if (global.show_rate_hist_buckets)
2546 FOREACH_STREAM(show_rate_histogram(&stream->rate_hist,
2547 &stream->config.cfg,
2548 global.show_rate_hist_buckets));
2549 FOREACH_STREAM(destroy_rate_histogram(&stream->rate_hist));
2554 return EXIT_SUCCESS;