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, and uses f{seek,tell}o64 */
51 #define fseeko fseeko64
52 #define ftello ftello64
56 #define LITERALU64(n) n
58 #define LITERALU64(n) n##LLU
61 /* We should use 32-bit file operations in WebM file format
62 * when building ARM executable file (.axf) with RVCT */
63 #if !CONFIG_OS_SUPPORT
69 static const char *exec_name;
71 static const struct codec_item
74 const vpx_codec_iface_t *iface;
78 #if CONFIG_VP8_ENCODER
79 {"vp8", &vpx_codec_vp8_cx_algo, 0x30385056},
83 static void usage_exit();
85 void die(const char *fmt, ...)
89 vfprintf(stderr, fmt, ap);
90 fprintf(stderr, "\n");
94 static void ctx_exit_on_error(vpx_codec_ctx_t *ctx, const char *s)
98 const char *detail = vpx_codec_error_detail(ctx);
100 fprintf(stderr, "%s: %s\n", s, vpx_codec_error(ctx));
103 fprintf(stderr, " %s\n", detail);
109 /* This structure is used to abstract the different ways of handling
110 * first pass statistics.
121 int stats_open_file(stats_io_t *stats, const char *fpf, int pass)
129 stats->file = fopen(fpf, "wb");
131 stats->buf.buf = NULL,
132 res = (stats->file != NULL);
138 struct stat stat_buf;
141 fd = open(fpf, O_RDONLY);
142 stats->file = fdopen(fd, "rb");
143 fstat(fd, &stat_buf);
144 stats->buf.sz = stat_buf.st_size;
145 stats->buf.buf = mmap(NULL, stats->buf.sz, PROT_READ, MAP_PRIVATE,
147 res = (stats->buf.buf != NULL);
151 stats->file = fopen(fpf, "rb");
153 if (fseek(stats->file, 0, SEEK_END))
155 fprintf(stderr, "First-pass stats file must be seekable!\n");
159 stats->buf.sz = stats->buf_alloc_sz = ftell(stats->file);
162 stats->buf.buf = malloc(stats->buf_alloc_sz);
166 fprintf(stderr, "Failed to allocate first-pass stats buffer (%lu bytes)\n",
167 (unsigned long)stats->buf_alloc_sz);
171 nbytes = fread(stats->buf.buf, 1, stats->buf.sz, stats->file);
172 res = (nbytes == stats->buf.sz);
179 int stats_open_mem(stats_io_t *stats, int pass)
187 stats->buf_alloc_sz = 64 * 1024;
188 stats->buf.buf = malloc(stats->buf_alloc_sz);
191 stats->buf_ptr = stats->buf.buf;
192 res = (stats->buf.buf != NULL);
197 void stats_close(stats_io_t *stats, int last_pass)
201 if (stats->pass == last_pass)
205 munmap(stats->buf.buf, stats->buf.sz);
207 free(stats->buf.buf);
216 if (stats->pass == last_pass)
217 free(stats->buf.buf);
221 void stats_write(stats_io_t *stats, const void *pkt, size_t len)
225 if(fwrite(pkt, 1, len, stats->file));
229 if (stats->buf.sz + len > stats->buf_alloc_sz)
231 size_t new_sz = stats->buf_alloc_sz + 64 * 1024;
232 char *new_ptr = realloc(stats->buf.buf, new_sz);
236 stats->buf_ptr = new_ptr + (stats->buf_ptr - (char *)stats->buf.buf);
237 stats->buf.buf = new_ptr;
238 stats->buf_alloc_sz = new_sz;
243 "\nFailed to realloc firstpass stats buffer.\n");
248 memcpy(stats->buf_ptr, pkt, len);
249 stats->buf.sz += len;
250 stats->buf_ptr += len;
254 vpx_fixed_buf_t stats_get(stats_io_t *stats)
259 /* Stereo 3D packed frame format */
260 typedef enum stereo_format
262 STEREO_FORMAT_MONO = 0,
263 STEREO_FORMAT_LEFT_RIGHT = 1,
264 STEREO_FORMAT_BOTTOM_TOP = 2,
265 STEREO_FORMAT_TOP_BOTTOM = 3,
266 STEREO_FORMAT_RIGHT_LEFT = 11
276 struct detect_buffer {
283 #define IVF_FRAME_HDR_SZ (4+8) /* 4 byte size + 8 byte timestamp */
284 static int read_frame(FILE *f, vpx_image_t *img, unsigned int file_type,
285 y4m_input *y4m, struct detect_buffer *detect)
290 if (file_type == FILE_TYPE_Y4M)
292 if (y4m_input_fetch_frame(y4m, f, img) < 1)
297 if (file_type == FILE_TYPE_IVF)
299 char junk[IVF_FRAME_HDR_SZ];
301 /* Skip the frame header. We know how big the frame should be. See
302 * write_ivf_frame_header() for documentation on the frame header
305 if(fread(junk, 1, IVF_FRAME_HDR_SZ, f));
308 for (plane = 0; plane < 3; plane++)
311 int w = (plane ? (1 + img->d_w) / 2 : img->d_w);
312 int h = (plane ? (1 + img->d_h) / 2 : img->d_h);
315 /* Determine the correct plane based on the image format. The for-loop
316 * always counts in Y,U,V order, but this may not match the order of
322 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U];
325 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V];
328 ptr = img->planes[plane];
331 for (r = 0; r < h; r++)
334 size_t buf_position = 0;
335 const size_t left = detect->buf_read - detect->position;
338 const size_t more = (left < needed) ? left : needed;
339 memcpy(ptr, detect->buf + detect->position, more);
342 detect->position += more;
346 shortread |= (fread(ptr + buf_position, 1, needed, f) < needed);
349 ptr += img->stride[plane];
358 unsigned int file_is_y4m(FILE *infile,
362 if(memcmp(detect, "YUV4", 4) == 0)
369 #define IVF_FILE_HDR_SZ (32)
370 unsigned int file_is_ivf(FILE *infile,
371 unsigned int *fourcc,
373 unsigned int *height,
374 struct detect_buffer *detect)
376 char raw_hdr[IVF_FILE_HDR_SZ];
379 if(memcmp(detect->buf, "DKIF", 4) != 0)
382 /* See write_ivf_file_header() for more documentation on the file header
385 if (fread(raw_hdr + 4, 1, IVF_FILE_HDR_SZ - 4, infile)
386 == IVF_FILE_HDR_SZ - 4)
391 if (mem_get_le16(raw_hdr + 4) != 0)
392 fprintf(stderr, "Error: Unrecognized IVF version! This file may not"
393 " decode properly.");
395 *fourcc = mem_get_le32(raw_hdr + 8);
401 *width = mem_get_le16(raw_hdr + 12);
402 *height = mem_get_le16(raw_hdr + 14);
403 detect->position = 4;
410 static void write_ivf_file_header(FILE *outfile,
411 const vpx_codec_enc_cfg_t *cfg,
417 if (cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS)
424 mem_put_le16(header + 4, 0); /* version */
425 mem_put_le16(header + 6, 32); /* headersize */
426 mem_put_le32(header + 8, fourcc); /* headersize */
427 mem_put_le16(header + 12, cfg->g_w); /* width */
428 mem_put_le16(header + 14, cfg->g_h); /* height */
429 mem_put_le32(header + 16, cfg->g_timebase.den); /* rate */
430 mem_put_le32(header + 20, cfg->g_timebase.num); /* scale */
431 mem_put_le32(header + 24, frame_cnt); /* length */
432 mem_put_le32(header + 28, 0); /* unused */
434 if(fwrite(header, 1, 32, outfile));
438 static void write_ivf_frame_header(FILE *outfile,
439 const vpx_codec_cx_pkt_t *pkt)
444 if (pkt->kind != VPX_CODEC_CX_FRAME_PKT)
447 pts = pkt->data.frame.pts;
448 mem_put_le32(header, pkt->data.frame.sz);
449 mem_put_le32(header + 4, pts & 0xFFFFFFFF);
450 mem_put_le32(header + 8, pts >> 32);
452 if(fwrite(header, 1, 12, outfile));
456 typedef off_t EbmlLoc;
472 vpx_rational_t framerate;
474 /* These pointers are to the start of an element */
475 off_t position_reference;
477 off_t segment_info_pos;
482 /* This pointer is to a specific element to be serialized */
485 /* These pointers are to the size field of the element */
486 EbmlLoc startSegment;
487 EbmlLoc startCluster;
489 uint32_t cluster_timecode;
492 struct cue_entry *cue_list;
498 void Ebml_Write(EbmlGlobal *glob, const void *buffer_in, unsigned long len)
500 if(fwrite(buffer_in, 1, len, glob->stream));
503 #define WRITE_BUFFER(s) \
504 for(i = len-1; i>=0; i--)\
506 x = *(const s *)buffer_in >> (i * CHAR_BIT); \
507 Ebml_Write(glob, &x, 1); \
509 void Ebml_Serialize(EbmlGlobal *glob, const void *buffer_in, int buffer_size, unsigned long len)
526 WRITE_BUFFER(int16_t)
529 WRITE_BUFFER(int32_t)
532 WRITE_BUFFER(int64_t)
540 /* Need a fixed size serializer for the track ID. libmkv provides a 64 bit
541 * one, but not a 32 bit one.
543 static void Ebml_SerializeUnsigned32(EbmlGlobal *glob, unsigned long class_id, uint64_t ui)
545 unsigned char sizeSerialized = 4 | 0x80;
546 Ebml_WriteID(glob, class_id);
547 Ebml_Serialize(glob, &sizeSerialized, sizeof(sizeSerialized), 1);
548 Ebml_Serialize(glob, &ui, sizeof(ui), 4);
553 Ebml_StartSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc,
554 unsigned long class_id)
556 //todo this is always taking 8 bytes, this may need later optimization
557 //this is a key that says length unknown
558 uint64_t unknownLen = LITERALU64(0x01FFFFFFFFFFFFFF);
560 Ebml_WriteID(glob, class_id);
561 *ebmlLoc = ftello(glob->stream);
562 Ebml_Serialize(glob, &unknownLen, sizeof(unknownLen), 8);
566 Ebml_EndSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc)
571 /* Save the current stream pointer */
572 pos = ftello(glob->stream);
574 /* Calculate the size of this element */
575 size = pos - *ebmlLoc - 8;
576 size |= LITERALU64(0x0100000000000000);
578 /* Seek back to the beginning of the element and write the new size */
579 fseeko(glob->stream, *ebmlLoc, SEEK_SET);
580 Ebml_Serialize(glob, &size, sizeof(size), 8);
582 /* Reset the stream pointer */
583 fseeko(glob->stream, pos, SEEK_SET);
588 write_webm_seek_element(EbmlGlobal *ebml, unsigned long id, off_t pos)
590 uint64_t offset = pos - ebml->position_reference;
592 Ebml_StartSubElement(ebml, &start, Seek);
593 Ebml_SerializeBinary(ebml, SeekID, id);
594 Ebml_SerializeUnsigned64(ebml, SeekPosition, offset);
595 Ebml_EndSubElement(ebml, &start);
600 write_webm_seek_info(EbmlGlobal *ebml)
605 /* Save the current stream pointer */
606 pos = ftello(ebml->stream);
608 if(ebml->seek_info_pos)
609 fseeko(ebml->stream, ebml->seek_info_pos, SEEK_SET);
611 ebml->seek_info_pos = pos;
616 Ebml_StartSubElement(ebml, &start, SeekHead);
617 write_webm_seek_element(ebml, Tracks, ebml->track_pos);
618 write_webm_seek_element(ebml, Cues, ebml->cue_pos);
619 write_webm_seek_element(ebml, Info, ebml->segment_info_pos);
620 Ebml_EndSubElement(ebml, &start);
626 char version_string[64];
628 /* Assemble version string */
630 strcpy(version_string, "vpxenc");
633 strcpy(version_string, "vpxenc ");
634 strncat(version_string,
635 vpx_codec_version_str(),
636 sizeof(version_string) - 1 - strlen(version_string));
639 frame_time = (uint64_t)1000 * ebml->framerate.den
640 / ebml->framerate.num;
641 ebml->segment_info_pos = ftello(ebml->stream);
642 Ebml_StartSubElement(ebml, &startInfo, Info);
643 Ebml_SerializeUnsigned(ebml, TimecodeScale, 1000000);
644 Ebml_SerializeFloat(ebml, Segment_Duration,
645 ebml->last_pts_ms + frame_time);
646 Ebml_SerializeString(ebml, 0x4D80, version_string);
647 Ebml_SerializeString(ebml, 0x5741, version_string);
648 Ebml_EndSubElement(ebml, &startInfo);
654 write_webm_file_header(EbmlGlobal *glob,
655 const vpx_codec_enc_cfg_t *cfg,
656 const struct vpx_rational *fps,
657 stereo_format_t stereo_fmt)
661 Ebml_StartSubElement(glob, &start, EBML);
662 Ebml_SerializeUnsigned(glob, EBMLVersion, 1);
663 Ebml_SerializeUnsigned(glob, EBMLReadVersion, 1); //EBML Read Version
664 Ebml_SerializeUnsigned(glob, EBMLMaxIDLength, 4); //EBML Max ID Length
665 Ebml_SerializeUnsigned(glob, EBMLMaxSizeLength, 8); //EBML Max Size Length
666 Ebml_SerializeString(glob, DocType, "webm"); //Doc Type
667 Ebml_SerializeUnsigned(glob, DocTypeVersion, 2); //Doc Type Version
668 Ebml_SerializeUnsigned(glob, DocTypeReadVersion, 2); //Doc Type Read Version
669 Ebml_EndSubElement(glob, &start);
672 Ebml_StartSubElement(glob, &glob->startSegment, Segment); //segment
673 glob->position_reference = ftello(glob->stream);
674 glob->framerate = *fps;
675 write_webm_seek_info(glob);
679 glob->track_pos = ftello(glob->stream);
680 Ebml_StartSubElement(glob, &trackStart, Tracks);
682 unsigned int trackNumber = 1;
683 uint64_t trackID = 0;
686 Ebml_StartSubElement(glob, &start, TrackEntry);
687 Ebml_SerializeUnsigned(glob, TrackNumber, trackNumber);
688 glob->track_id_pos = ftello(glob->stream);
689 Ebml_SerializeUnsigned32(glob, TrackUID, trackID);
690 Ebml_SerializeUnsigned(glob, TrackType, 1); //video is always 1
691 Ebml_SerializeString(glob, CodecID, "V_VP8");
693 unsigned int pixelWidth = cfg->g_w;
694 unsigned int pixelHeight = cfg->g_h;
695 float frameRate = (float)fps->num/(float)fps->den;
698 Ebml_StartSubElement(glob, &videoStart, Video);
699 Ebml_SerializeUnsigned(glob, PixelWidth, pixelWidth);
700 Ebml_SerializeUnsigned(glob, PixelHeight, pixelHeight);
701 Ebml_SerializeUnsigned(glob, StereoMode, stereo_fmt);
702 Ebml_SerializeFloat(glob, FrameRate, frameRate);
703 Ebml_EndSubElement(glob, &videoStart); //Video
705 Ebml_EndSubElement(glob, &start); //Track Entry
707 Ebml_EndSubElement(glob, &trackStart);
709 // segment element is open
715 write_webm_block(EbmlGlobal *glob,
716 const vpx_codec_enc_cfg_t *cfg,
717 const vpx_codec_cx_pkt_t *pkt)
719 unsigned long block_length;
720 unsigned char track_number;
721 unsigned short block_timecode = 0;
724 int start_cluster = 0, is_keyframe;
726 /* Calculate the PTS of this frame in milliseconds */
727 pts_ms = pkt->data.frame.pts * 1000
728 * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
729 if(pts_ms <= glob->last_pts_ms)
730 pts_ms = glob->last_pts_ms + 1;
731 glob->last_pts_ms = pts_ms;
733 /* Calculate the relative time of this block */
734 if(pts_ms - glob->cluster_timecode > SHRT_MAX)
737 block_timecode = pts_ms - glob->cluster_timecode;
739 is_keyframe = (pkt->data.frame.flags & VPX_FRAME_IS_KEY);
740 if(start_cluster || is_keyframe)
742 if(glob->cluster_open)
743 Ebml_EndSubElement(glob, &glob->startCluster);
745 /* Open the new cluster */
747 glob->cluster_open = 1;
748 glob->cluster_timecode = pts_ms;
749 glob->cluster_pos = ftello(glob->stream);
750 Ebml_StartSubElement(glob, &glob->startCluster, Cluster); //cluster
751 Ebml_SerializeUnsigned(glob, Timecode, glob->cluster_timecode);
753 /* Save a cue point if this is a keyframe. */
756 struct cue_entry *cue, *new_cue_list;
758 new_cue_list = realloc(glob->cue_list,
759 (glob->cues+1) * sizeof(struct cue_entry));
761 glob->cue_list = new_cue_list;
764 fprintf(stderr, "\nFailed to realloc cue list.\n");
768 cue = &glob->cue_list[glob->cues];
769 cue->time = glob->cluster_timecode;
770 cue->loc = glob->cluster_pos;
775 /* Write the Simple Block */
776 Ebml_WriteID(glob, SimpleBlock);
778 block_length = pkt->data.frame.sz + 4;
779 block_length |= 0x10000000;
780 Ebml_Serialize(glob, &block_length, sizeof(block_length), 4);
783 track_number |= 0x80;
784 Ebml_Write(glob, &track_number, 1);
786 Ebml_Serialize(glob, &block_timecode, sizeof(block_timecode), 2);
791 if(pkt->data.frame.flags & VPX_FRAME_IS_INVISIBLE)
793 Ebml_Write(glob, &flags, 1);
795 Ebml_Write(glob, pkt->data.frame.buf, pkt->data.frame.sz);
800 write_webm_file_footer(EbmlGlobal *glob, long hash)
803 if(glob->cluster_open)
804 Ebml_EndSubElement(glob, &glob->startCluster);
810 glob->cue_pos = ftello(glob->stream);
811 Ebml_StartSubElement(glob, &start, Cues);
812 for(i=0; i<glob->cues; i++)
814 struct cue_entry *cue = &glob->cue_list[i];
817 Ebml_StartSubElement(glob, &start, CuePoint);
821 Ebml_SerializeUnsigned(glob, CueTime, cue->time);
823 Ebml_StartSubElement(glob, &start, CueTrackPositions);
824 Ebml_SerializeUnsigned(glob, CueTrack, 1);
825 Ebml_SerializeUnsigned64(glob, CueClusterPosition,
826 cue->loc - glob->position_reference);
827 //Ebml_SerializeUnsigned(glob, CueBlockNumber, cue->blockNumber);
828 Ebml_EndSubElement(glob, &start);
830 Ebml_EndSubElement(glob, &start);
832 Ebml_EndSubElement(glob, &start);
835 Ebml_EndSubElement(glob, &glob->startSegment);
837 /* Patch up the seek info block */
838 write_webm_seek_info(glob);
840 /* Patch up the track id */
841 fseeko(glob->stream, glob->track_id_pos, SEEK_SET);
842 Ebml_SerializeUnsigned32(glob, TrackUID, glob->debug ? 0xDEADBEEF : hash);
844 fseeko(glob->stream, 0, SEEK_END);
848 /* Murmur hash derived from public domain reference implementation at
849 * http://sites.google.com/site/murmurhash/
851 static unsigned int murmur ( const void * key, int len, unsigned int seed )
853 const unsigned int m = 0x5bd1e995;
856 unsigned int h = seed ^ len;
858 const unsigned char * data = (const unsigned char *)key;
882 case 3: h ^= data[2] << 16;
883 case 2: h ^= data[1] << 8;
884 case 1: h ^= data[0];
897 static double vp8_mse2psnr(double Samples, double Peak, double Mse)
901 if ((double)Mse > 0.0)
902 psnr = 10.0 * log10(Peak * Peak * Samples / Mse);
904 psnr = 60; // Limit to prevent / 0
915 static const arg_def_t debugmode = ARG_DEF("D", "debug", 0,
916 "Debug mode (makes output deterministic)");
917 static const arg_def_t outputfile = ARG_DEF("o", "output", 1,
919 static const arg_def_t use_yv12 = ARG_DEF(NULL, "yv12", 0,
920 "Input file is YV12 ");
921 static const arg_def_t use_i420 = ARG_DEF(NULL, "i420", 0,
922 "Input file is I420 (default)");
923 static const arg_def_t codecarg = ARG_DEF(NULL, "codec", 1,
925 static const arg_def_t passes = ARG_DEF("p", "passes", 1,
926 "Number of passes (1/2)");
927 static const arg_def_t pass_arg = ARG_DEF(NULL, "pass", 1,
928 "Pass to execute (1/2)");
929 static const arg_def_t fpf_name = ARG_DEF(NULL, "fpf", 1,
930 "First pass statistics file name");
931 static const arg_def_t limit = ARG_DEF(NULL, "limit", 1,
932 "Stop encoding after n input frames");
933 static const arg_def_t deadline = ARG_DEF("d", "deadline", 1,
934 "Deadline per frame (usec)");
935 static const arg_def_t best_dl = ARG_DEF(NULL, "best", 0,
936 "Use Best Quality Deadline");
937 static const arg_def_t good_dl = ARG_DEF(NULL, "good", 0,
938 "Use Good Quality Deadline");
939 static const arg_def_t rt_dl = ARG_DEF(NULL, "rt", 0,
940 "Use Realtime Quality Deadline");
941 static const arg_def_t verbosearg = ARG_DEF("v", "verbose", 0,
942 "Show encoder parameters");
943 static const arg_def_t psnrarg = ARG_DEF(NULL, "psnr", 0,
944 "Show PSNR in status line");
945 static const arg_def_t framerate = ARG_DEF(NULL, "fps", 1,
946 "Stream frame rate (rate/scale)");
947 static const arg_def_t use_ivf = ARG_DEF(NULL, "ivf", 0,
948 "Output IVF (default is WebM)");
949 static const arg_def_t q_hist_n = ARG_DEF(NULL, "q-hist", 1,
950 "Show quantizer histogram (n-buckets)");
951 static const arg_def_t rate_hist_n = ARG_DEF(NULL, "rate-hist", 1,
952 "Show rate histogram (n-buckets)");
953 static const arg_def_t *main_args[] =
956 &outputfile, &codecarg, &passes, &pass_arg, &fpf_name, &limit, &deadline,
957 &best_dl, &good_dl, &rt_dl,
958 &verbosearg, &psnrarg, &use_ivf, &q_hist_n, &rate_hist_n,
962 static const arg_def_t usage = ARG_DEF("u", "usage", 1,
963 "Usage profile number to use");
964 static const arg_def_t threads = ARG_DEF("t", "threads", 1,
965 "Max number of threads to use");
966 static const arg_def_t profile = ARG_DEF(NULL, "profile", 1,
967 "Bitstream profile number to use");
968 static const arg_def_t width = ARG_DEF("w", "width", 1,
970 static const arg_def_t height = ARG_DEF("h", "height", 1,
972 static const struct arg_enum_list stereo_mode_enum[] = {
973 {"mono" , STEREO_FORMAT_MONO},
974 {"left-right", STEREO_FORMAT_LEFT_RIGHT},
975 {"bottom-top", STEREO_FORMAT_BOTTOM_TOP},
976 {"top-bottom", STEREO_FORMAT_TOP_BOTTOM},
977 {"right-left", STEREO_FORMAT_RIGHT_LEFT},
980 static const arg_def_t stereo_mode = ARG_DEF_ENUM(NULL, "stereo-mode", 1,
981 "Stereo 3D video format", stereo_mode_enum);
982 static const arg_def_t timebase = ARG_DEF(NULL, "timebase", 1,
983 "Output timestamp precision (fractional seconds)");
984 static const arg_def_t error_resilient = ARG_DEF(NULL, "error-resilient", 1,
985 "Enable error resiliency features");
986 static const arg_def_t lag_in_frames = ARG_DEF(NULL, "lag-in-frames", 1,
987 "Max number of frames to lag");
989 static const arg_def_t *global_args[] =
991 &use_yv12, &use_i420, &usage, &threads, &profile,
992 &width, &height, &stereo_mode, &timebase, &framerate, &error_resilient,
996 static const arg_def_t dropframe_thresh = ARG_DEF(NULL, "drop-frame", 1,
997 "Temporal resampling threshold (buf %)");
998 static const arg_def_t resize_allowed = ARG_DEF(NULL, "resize-allowed", 1,
999 "Spatial resampling enabled (bool)");
1000 static const arg_def_t resize_up_thresh = ARG_DEF(NULL, "resize-up", 1,
1001 "Upscale threshold (buf %)");
1002 static const arg_def_t resize_down_thresh = ARG_DEF(NULL, "resize-down", 1,
1003 "Downscale threshold (buf %)");
1004 static const struct arg_enum_list end_usage_enum[] = {
1010 static const arg_def_t end_usage = ARG_DEF_ENUM(NULL, "end-usage", 1,
1011 "Rate control mode", end_usage_enum);
1012 static const arg_def_t target_bitrate = ARG_DEF(NULL, "target-bitrate", 1,
1014 static const arg_def_t min_quantizer = ARG_DEF(NULL, "min-q", 1,
1015 "Minimum (best) quantizer");
1016 static const arg_def_t max_quantizer = ARG_DEF(NULL, "max-q", 1,
1017 "Maximum (worst) quantizer");
1018 static const arg_def_t undershoot_pct = ARG_DEF(NULL, "undershoot-pct", 1,
1019 "Datarate undershoot (min) target (%)");
1020 static const arg_def_t overshoot_pct = ARG_DEF(NULL, "overshoot-pct", 1,
1021 "Datarate overshoot (max) target (%)");
1022 static const arg_def_t buf_sz = ARG_DEF(NULL, "buf-sz", 1,
1023 "Client buffer size (ms)");
1024 static const arg_def_t buf_initial_sz = ARG_DEF(NULL, "buf-initial-sz", 1,
1025 "Client initial buffer size (ms)");
1026 static const arg_def_t buf_optimal_sz = ARG_DEF(NULL, "buf-optimal-sz", 1,
1027 "Client optimal buffer size (ms)");
1028 static const arg_def_t *rc_args[] =
1030 &dropframe_thresh, &resize_allowed, &resize_up_thresh, &resize_down_thresh,
1031 &end_usage, &target_bitrate, &min_quantizer, &max_quantizer,
1032 &undershoot_pct, &overshoot_pct, &buf_sz, &buf_initial_sz, &buf_optimal_sz,
1037 static const arg_def_t bias_pct = ARG_DEF(NULL, "bias-pct", 1,
1038 "CBR/VBR bias (0=CBR, 100=VBR)");
1039 static const arg_def_t minsection_pct = ARG_DEF(NULL, "minsection-pct", 1,
1040 "GOP min bitrate (% of target)");
1041 static const arg_def_t maxsection_pct = ARG_DEF(NULL, "maxsection-pct", 1,
1042 "GOP max bitrate (% of target)");
1043 static const arg_def_t *rc_twopass_args[] =
1045 &bias_pct, &minsection_pct, &maxsection_pct, NULL
1049 static const arg_def_t kf_min_dist = ARG_DEF(NULL, "kf-min-dist", 1,
1050 "Minimum keyframe interval (frames)");
1051 static const arg_def_t kf_max_dist = ARG_DEF(NULL, "kf-max-dist", 1,
1052 "Maximum keyframe interval (frames)");
1053 static const arg_def_t kf_disabled = ARG_DEF(NULL, "disable-kf", 0,
1054 "Disable keyframe placement");
1055 static const arg_def_t *kf_args[] =
1057 &kf_min_dist, &kf_max_dist, &kf_disabled, NULL
1061 #if CONFIG_VP8_ENCODER
1062 static const arg_def_t noise_sens = ARG_DEF(NULL, "noise-sensitivity", 1,
1063 "Noise sensitivity (frames to blur)");
1064 static const arg_def_t sharpness = ARG_DEF(NULL, "sharpness", 1,
1065 "Filter sharpness (0-7)");
1066 static const arg_def_t static_thresh = ARG_DEF(NULL, "static-thresh", 1,
1067 "Motion detection threshold");
1070 #if CONFIG_VP8_ENCODER
1071 static const arg_def_t cpu_used = ARG_DEF(NULL, "cpu-used", 1,
1072 "CPU Used (-16..16)");
1076 #if CONFIG_VP8_ENCODER
1077 static const arg_def_t token_parts = ARG_DEF(NULL, "token-parts", 1,
1078 "Number of token partitions to use, log2");
1079 static const arg_def_t auto_altref = ARG_DEF(NULL, "auto-alt-ref", 1,
1080 "Enable automatic alt reference frames");
1081 static const arg_def_t arnr_maxframes = ARG_DEF(NULL, "arnr-maxframes", 1,
1082 "AltRef Max Frames");
1083 static const arg_def_t arnr_strength = ARG_DEF(NULL, "arnr-strength", 1,
1085 static const arg_def_t arnr_type = ARG_DEF(NULL, "arnr-type", 1,
1087 static const struct arg_enum_list tuning_enum[] = {
1088 {"psnr", VP8_TUNE_PSNR},
1089 {"ssim", VP8_TUNE_SSIM},
1092 static const arg_def_t tune_ssim = ARG_DEF_ENUM(NULL, "tune", 1,
1093 "Material to favor", tuning_enum);
1094 static const arg_def_t cq_level = ARG_DEF(NULL, "cq-level", 1,
1095 "Constrained Quality Level");
1096 static const arg_def_t max_intra_rate_pct = ARG_DEF(NULL, "max-intra-rate", 1,
1097 "Max I-frame bitrate (pct)");
1099 static const arg_def_t *vp8_args[] =
1101 &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh,
1102 &token_parts, &arnr_maxframes, &arnr_strength, &arnr_type,
1103 &tune_ssim, &cq_level, &max_intra_rate_pct, NULL
1105 static const int vp8_arg_ctrl_map[] =
1107 VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF,
1108 VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD,
1109 VP8E_SET_TOKEN_PARTITIONS,
1110 VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH , VP8E_SET_ARNR_TYPE,
1111 VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, VP8E_SET_MAX_INTRA_BITRATE_PCT, 0
1115 static const arg_def_t *no_args[] = { NULL };
1117 static void usage_exit()
1121 fprintf(stderr, "Usage: %s <options> -o dst_filename src_filename \n",
1124 fprintf(stderr, "\nOptions:\n");
1125 arg_show_usage(stdout, main_args);
1126 fprintf(stderr, "\nEncoder Global Options:\n");
1127 arg_show_usage(stdout, global_args);
1128 fprintf(stderr, "\nRate Control Options:\n");
1129 arg_show_usage(stdout, rc_args);
1130 fprintf(stderr, "\nTwopass Rate Control Options:\n");
1131 arg_show_usage(stdout, rc_twopass_args);
1132 fprintf(stderr, "\nKeyframe Placement Options:\n");
1133 arg_show_usage(stdout, kf_args);
1134 #if CONFIG_VP8_ENCODER
1135 fprintf(stderr, "\nVP8 Specific Options:\n");
1136 arg_show_usage(stdout, vp8_args);
1138 fprintf(stderr, "\nStream timebase (--timebase):\n"
1139 " The desired precision of timestamps in the output, expressed\n"
1140 " in fractional seconds. Default is 1/1000.\n");
1141 fprintf(stderr, "\n"
1142 "Included encoders:\n"
1145 for (i = 0; i < sizeof(codecs) / sizeof(codecs[0]); i++)
1146 fprintf(stderr, " %-6s - %s\n",
1148 vpx_codec_iface_name(codecs[i].iface));
1154 #define HIST_BAR_MAX 40
1157 int low, high, count;
1161 static int merge_hist_buckets(struct hist_bucket *bucket,
1165 int small_bucket = 0, merge_bucket = INT_MAX, big_bucket=0;
1166 int buckets = *buckets_;
1169 /* Find the extrema for this list of buckets */
1170 big_bucket = small_bucket = 0;
1171 for(i=0; i < buckets; i++)
1173 if(bucket[i].count < bucket[small_bucket].count)
1175 if(bucket[i].count > bucket[big_bucket].count)
1179 /* If we have too many buckets, merge the smallest with an adjacent
1182 while(buckets > max_buckets)
1184 int last_bucket = buckets - 1;
1186 // merge the small bucket with an adjacent one.
1187 if(small_bucket == 0)
1189 else if(small_bucket == last_bucket)
1190 merge_bucket = last_bucket - 1;
1191 else if(bucket[small_bucket - 1].count < bucket[small_bucket + 1].count)
1192 merge_bucket = small_bucket - 1;
1194 merge_bucket = small_bucket + 1;
1196 assert(abs(merge_bucket - small_bucket) <= 1);
1197 assert(small_bucket < buckets);
1198 assert(big_bucket < buckets);
1199 assert(merge_bucket < buckets);
1201 if(merge_bucket < small_bucket)
1203 bucket[merge_bucket].high = bucket[small_bucket].high;
1204 bucket[merge_bucket].count += bucket[small_bucket].count;
1208 bucket[small_bucket].high = bucket[merge_bucket].high;
1209 bucket[small_bucket].count += bucket[merge_bucket].count;
1210 merge_bucket = small_bucket;
1213 assert(bucket[merge_bucket].low != bucket[merge_bucket].high);
1217 /* Remove the merge_bucket from the list, and find the new small
1218 * and big buckets while we're at it
1220 big_bucket = small_bucket = 0;
1221 for(i=0; i < buckets; i++)
1223 if(i > merge_bucket)
1224 bucket[i] = bucket[i+1];
1226 if(bucket[i].count < bucket[small_bucket].count)
1228 if(bucket[i].count > bucket[big_bucket].count)
1234 *buckets_ = buckets;
1235 return bucket[big_bucket].count;
1239 static void show_histogram(const struct hist_bucket *bucket,
1244 const char *pat1, *pat2;
1247 switch((int)(log(bucket[buckets-1].high)/log(10))+1)
1275 pat1 = "%12d %10s: ";
1276 pat2 = "%12d-%10d: ";
1280 for(i=0; i<buckets; i++)
1286 pct = 100.0 * (float)bucket[i].count / (float)total;
1287 len = HIST_BAR_MAX * bucket[i].count / scale;
1290 assert(len <= HIST_BAR_MAX);
1292 if(bucket[i].low == bucket[i].high)
1293 fprintf(stderr, pat1, bucket[i].low, "");
1295 fprintf(stderr, pat2, bucket[i].low, bucket[i].high);
1297 for(j=0; j<HIST_BAR_MAX; j++)
1298 fprintf(stderr, j<len?"=":" ");
1299 fprintf(stderr, "\t%5d (%6.2f%%)\n",bucket[i].count,pct);
1304 static void show_q_histogram(const int counts[64], int max_buckets)
1306 struct hist_bucket bucket[64];
1317 bucket[buckets].low = bucket[buckets].high = i;
1318 bucket[buckets].count = counts[i];
1324 fprintf(stderr, "\nQuantizer Selection:\n");
1325 scale = merge_hist_buckets(bucket, &buckets, max_buckets);
1326 show_histogram(bucket, buckets, total, scale);
1330 #define RATE_BINS (100)
1337 struct hist_bucket bucket[RATE_BINS];
1342 static void init_rate_histogram(struct rate_hist *hist,
1343 const vpx_codec_enc_cfg_t *cfg,
1344 const vpx_rational_t *fps)
1348 /* Determine the number of samples in the buffer. Use the file's framerate
1349 * to determine the number of frames in rc_buf_sz milliseconds, with an
1350 * adjustment (5/4) to account for alt-refs
1352 hist->samples = cfg->rc_buf_sz * 5 / 4 * fps->num / fps->den / 1000;
1354 // prevent division by zero
1355 if (hist->samples == 0)
1358 hist->pts = calloc(hist->samples, sizeof(*hist->pts));
1359 hist->sz = calloc(hist->samples, sizeof(*hist->sz));
1360 for(i=0; i<RATE_BINS; i++)
1362 hist->bucket[i].low = INT_MAX;
1363 hist->bucket[i].high = 0;
1364 hist->bucket[i].count = 0;
1369 static void destroy_rate_histogram(struct rate_hist *hist)
1376 static void update_rate_histogram(struct rate_hist *hist,
1377 const vpx_codec_enc_cfg_t *cfg,
1378 const vpx_codec_cx_pkt_t *pkt)
1381 int64_t now, then, sum_sz = 0, avg_bitrate;
1383 now = pkt->data.frame.pts * 1000
1384 * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
1386 idx = hist->frames++ % hist->samples;
1387 hist->pts[idx] = now;
1388 hist->sz[idx] = pkt->data.frame.sz;
1390 if(now < cfg->rc_buf_initial_sz)
1395 /* Sum the size over the past rc_buf_sz ms */
1396 for(i = hist->frames; i > 0 && hist->frames - i < hist->samples; i--)
1398 int i_idx = (i-1) % hist->samples;
1400 then = hist->pts[i_idx];
1401 if(now - then > cfg->rc_buf_sz)
1403 sum_sz += hist->sz[i_idx];
1409 avg_bitrate = sum_sz * 8 * 1000 / (now - then);
1410 idx = avg_bitrate * (RATE_BINS/2) / (cfg->rc_target_bitrate * 1000);
1413 if(idx > RATE_BINS-1)
1415 if(hist->bucket[idx].low > avg_bitrate)
1416 hist->bucket[idx].low = avg_bitrate;
1417 if(hist->bucket[idx].high < avg_bitrate)
1418 hist->bucket[idx].high = avg_bitrate;
1419 hist->bucket[idx].count++;
1424 static void show_rate_histogram(struct rate_hist *hist,
1425 const vpx_codec_enc_cfg_t *cfg,
1431 for(i = 0; i < RATE_BINS; i++)
1433 if(hist->bucket[i].low == INT_MAX)
1435 hist->bucket[buckets++] = hist->bucket[i];
1438 fprintf(stderr, "\nRate (over %dms window):\n", cfg->rc_buf_sz);
1439 scale = merge_hist_buckets(hist->bucket, &buckets, max_buckets);
1440 show_histogram(hist->bucket, buckets, hist->total, scale);
1443 #define ARG_CTRL_CNT_MAX 10
1445 int main(int argc, const char **argv_)
1447 vpx_codec_ctx_t encoder;
1448 const char *in_fn = NULL, *out_fn = NULL, *stats_fn = NULL;
1450 FILE *infile, *outfile;
1451 vpx_codec_enc_cfg_t cfg;
1452 vpx_codec_err_t res;
1453 int pass, one_pass_only = 0;
1456 const struct codec_item *codec = codecs;
1457 int frame_avail, got_data;
1460 char **argv, **argi, **argj;
1461 int arg_usage = 0, arg_passes = 1, arg_deadline = 0;
1462 int arg_ctrls[ARG_CTRL_CNT_MAX][2], arg_ctrl_cnt = 0;
1464 static const arg_def_t **ctrl_args = no_args;
1465 static const int *ctrl_args_map = NULL;
1466 int verbose = 0, show_psnr = 0;
1467 int arg_use_i420 = 1;
1468 unsigned long cx_time = 0;
1469 unsigned int file_type, fourcc;
1471 struct vpx_rational arg_framerate = {30, 1};
1472 int arg_have_framerate = 0;
1474 EbmlGlobal ebml = {0};
1476 uint64_t psnr_sse_total = 0;
1477 uint64_t psnr_samples_total = 0;
1478 double psnr_totals[4] = {0, 0, 0, 0};
1480 stereo_format_t stereo_fmt = STEREO_FORMAT_MONO;
1482 int show_q_hist_buckets=0;
1483 int show_rate_hist_buckets=0;
1484 struct rate_hist rate_hist={0};
1486 exec_name = argv_[0];
1487 ebml.last_pts_ms = -1;
1493 /* First parse the codec and usage values, because we want to apply other
1494 * parameters on top of the default configuration provided by the codec.
1496 argv = argv_dup(argc - 1, argv_ + 1);
1498 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
1502 if (arg_match(&arg, &codecarg, argi))
1506 for (j = 0; j < sizeof(codecs) / sizeof(codecs[0]); j++)
1507 if (!strcmp(codecs[j].name, arg.val))
1513 die("Error: Unrecognized argument (%s) to --codec\n",
1517 else if (arg_match(&arg, &passes, argi))
1519 arg_passes = arg_parse_uint(&arg);
1521 if (arg_passes < 1 || arg_passes > 2)
1522 die("Error: Invalid number of passes (%d)\n", arg_passes);
1524 else if (arg_match(&arg, &pass_arg, argi))
1526 one_pass_only = arg_parse_uint(&arg);
1528 if (one_pass_only < 1 || one_pass_only > 2)
1529 die("Error: Invalid pass selected (%d)\n", one_pass_only);
1531 else if (arg_match(&arg, &fpf_name, argi))
1533 else if (arg_match(&arg, &usage, argi))
1534 arg_usage = arg_parse_uint(&arg);
1535 else if (arg_match(&arg, &deadline, argi))
1536 arg_deadline = arg_parse_uint(&arg);
1537 else if (arg_match(&arg, &best_dl, argi))
1538 arg_deadline = VPX_DL_BEST_QUALITY;
1539 else if (arg_match(&arg, &good_dl, argi))
1540 arg_deadline = VPX_DL_GOOD_QUALITY;
1541 else if (arg_match(&arg, &rt_dl, argi))
1542 arg_deadline = VPX_DL_REALTIME;
1543 else if (arg_match(&arg, &use_yv12, argi))
1547 else if (arg_match(&arg, &use_i420, argi))
1551 else if (arg_match(&arg, &verbosearg, argi))
1553 else if (arg_match(&arg, &limit, argi))
1554 arg_limit = arg_parse_uint(&arg);
1555 else if (arg_match(&arg, &psnrarg, argi))
1557 else if (arg_match(&arg, &framerate, argi))
1559 arg_framerate = arg_parse_rational(&arg);
1560 arg_have_framerate = 1;
1562 else if (arg_match(&arg, &use_ivf, argi))
1564 else if (arg_match(&arg, &outputfile, argi))
1566 else if (arg_match(&arg, &debugmode, argi))
1568 else if (arg_match(&arg, &q_hist_n, argi))
1569 show_q_hist_buckets = arg_parse_uint(&arg);
1570 else if (arg_match(&arg, &rate_hist_n, argi))
1571 show_rate_hist_buckets = arg_parse_uint(&arg);
1576 /* Ensure that --passes and --pass are consistent. If --pass is set and --passes=2,
1577 * ensure --fpf was set.
1581 /* DWIM: Assume the user meant passes=2 if pass=2 is specified */
1582 if (one_pass_only > arg_passes)
1584 fprintf(stderr, "Warning: Assuming --pass=%d implies --passes=%d\n",
1585 one_pass_only, one_pass_only);
1586 arg_passes = one_pass_only;
1589 if (arg_passes == 2 && !stats_fn)
1590 die("Must specify --fpf when --pass=%d and --passes=2\n", one_pass_only);
1593 /* Populate encoder configuration */
1594 res = vpx_codec_enc_config_default(codec->iface, &cfg, arg_usage);
1598 fprintf(stderr, "Failed to get config: %s\n",
1599 vpx_codec_err_to_string(res));
1600 return EXIT_FAILURE;
1603 /* Change the default timebase to a high enough value so that the encoder
1604 * will always create strictly increasing timestamps.
1606 cfg.g_timebase.den = 1000;
1608 /* Never use the library's default resolution, require it be parsed
1609 * from the file or set on the command line.
1614 /* Now parse the remainder of the parameters. */
1615 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
1620 else if (arg_match(&arg, &threads, argi))
1621 cfg.g_threads = arg_parse_uint(&arg);
1622 else if (arg_match(&arg, &profile, argi))
1623 cfg.g_profile = arg_parse_uint(&arg);
1624 else if (arg_match(&arg, &width, argi))
1625 cfg.g_w = arg_parse_uint(&arg);
1626 else if (arg_match(&arg, &height, argi))
1627 cfg.g_h = arg_parse_uint(&arg);
1628 else if (arg_match(&arg, &stereo_mode, argi))
1629 stereo_fmt = arg_parse_enum_or_int(&arg);
1630 else if (arg_match(&arg, &timebase, argi))
1631 cfg.g_timebase = arg_parse_rational(&arg);
1632 else if (arg_match(&arg, &error_resilient, argi))
1633 cfg.g_error_resilient = arg_parse_uint(&arg);
1634 else if (arg_match(&arg, &lag_in_frames, argi))
1635 cfg.g_lag_in_frames = arg_parse_uint(&arg);
1636 else if (arg_match(&arg, &dropframe_thresh, argi))
1637 cfg.rc_dropframe_thresh = arg_parse_uint(&arg);
1638 else if (arg_match(&arg, &resize_allowed, argi))
1639 cfg.rc_resize_allowed = arg_parse_uint(&arg);
1640 else if (arg_match(&arg, &resize_up_thresh, argi))
1641 cfg.rc_resize_up_thresh = arg_parse_uint(&arg);
1642 else if (arg_match(&arg, &resize_down_thresh, argi))
1643 cfg.rc_resize_down_thresh = arg_parse_uint(&arg);
1644 else if (arg_match(&arg, &resize_down_thresh, argi))
1645 cfg.rc_resize_down_thresh = arg_parse_uint(&arg);
1646 else if (arg_match(&arg, &end_usage, argi))
1647 cfg.rc_end_usage = arg_parse_enum_or_int(&arg);
1648 else if (arg_match(&arg, &target_bitrate, argi))
1649 cfg.rc_target_bitrate = arg_parse_uint(&arg);
1650 else if (arg_match(&arg, &min_quantizer, argi))
1651 cfg.rc_min_quantizer = arg_parse_uint(&arg);
1652 else if (arg_match(&arg, &max_quantizer, argi))
1653 cfg.rc_max_quantizer = arg_parse_uint(&arg);
1654 else if (arg_match(&arg, &undershoot_pct, argi))
1655 cfg.rc_undershoot_pct = arg_parse_uint(&arg);
1656 else if (arg_match(&arg, &overshoot_pct, argi))
1657 cfg.rc_overshoot_pct = arg_parse_uint(&arg);
1658 else if (arg_match(&arg, &buf_sz, argi))
1659 cfg.rc_buf_sz = arg_parse_uint(&arg);
1660 else if (arg_match(&arg, &buf_initial_sz, argi))
1661 cfg.rc_buf_initial_sz = arg_parse_uint(&arg);
1662 else if (arg_match(&arg, &buf_optimal_sz, argi))
1663 cfg.rc_buf_optimal_sz = arg_parse_uint(&arg);
1664 else if (arg_match(&arg, &bias_pct, argi))
1666 cfg.rc_2pass_vbr_bias_pct = arg_parse_uint(&arg);
1670 "Warning: option %s ignored in one-pass mode.\n",
1673 else if (arg_match(&arg, &minsection_pct, argi))
1675 cfg.rc_2pass_vbr_minsection_pct = arg_parse_uint(&arg);
1679 "Warning: option %s ignored in one-pass mode.\n",
1682 else if (arg_match(&arg, &maxsection_pct, argi))
1684 cfg.rc_2pass_vbr_maxsection_pct = arg_parse_uint(&arg);
1688 "Warning: option %s ignored in one-pass mode.\n",
1691 else if (arg_match(&arg, &kf_min_dist, argi))
1692 cfg.kf_min_dist = arg_parse_uint(&arg);
1693 else if (arg_match(&arg, &kf_max_dist, argi))
1694 cfg.kf_max_dist = arg_parse_uint(&arg);
1695 else if (arg_match(&arg, &kf_disabled, argi))
1696 cfg.kf_mode = VPX_KF_DISABLED;
1701 /* Handle codec specific options */
1702 #if CONFIG_VP8_ENCODER
1704 if (codec->iface == &vpx_codec_vp8_cx_algo)
1706 ctrl_args = vp8_args;
1707 ctrl_args_map = vp8_arg_ctrl_map;
1712 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
1718 for (i = 0; ctrl_args[i]; i++)
1720 if (arg_match(&arg, ctrl_args[i], argi))
1724 if (arg_ctrl_cnt < ARG_CTRL_CNT_MAX)
1726 arg_ctrls[arg_ctrl_cnt][0] = ctrl_args_map[i];
1727 arg_ctrls[arg_ctrl_cnt][1] = arg_parse_enum_or_int(&arg);
1737 /* Check for unrecognized options */
1738 for (argi = argv; *argi; argi++)
1739 if (argi[0][0] == '-' && argi[0][1])
1740 die("Error: Unrecognized option %s\n", *argi);
1742 /* Handle non-option arguments */
1749 die("Error: Output file is required (specify with -o)\n");
1751 memset(&stats, 0, sizeof(stats));
1753 for (pass = one_pass_only ? one_pass_only - 1 : 0; pass < arg_passes; pass++)
1755 int frames_in = 0, frames_out = 0;
1757 struct detect_buffer detect;
1759 /* Parse certain options from the input file, if possible */
1760 infile = strcmp(in_fn, "-") ? fopen(in_fn, "rb")
1761 : set_binary_mode(stdin);
1765 fprintf(stderr, "Failed to open input file\n");
1766 return EXIT_FAILURE;
1769 /* For RAW input sources, these bytes will applied on the first frame
1772 detect.buf_read = fread(detect.buf, 1, 4, infile);
1773 detect.position = 0;
1775 if (detect.buf_read == 4 && file_is_y4m(infile, &y4m, detect.buf))
1777 if (y4m_input_open(&y4m, infile, detect.buf, 4) >= 0)
1779 file_type = FILE_TYPE_Y4M;
1780 cfg.g_w = y4m.pic_w;
1781 cfg.g_h = y4m.pic_h;
1783 /* Use the frame rate from the file only if none was specified
1784 * on the command-line.
1786 if (!arg_have_framerate)
1788 arg_framerate.num = y4m.fps_n;
1789 arg_framerate.den = y4m.fps_d;
1796 fprintf(stderr, "Unsupported Y4M stream.\n");
1797 return EXIT_FAILURE;
1800 else if (detect.buf_read == 4 &&
1801 file_is_ivf(infile, &fourcc, &cfg.g_w, &cfg.g_h, &detect))
1803 file_type = FILE_TYPE_IVF;
1813 fprintf(stderr, "Unsupported fourcc (%08x) in IVF\n", fourcc);
1814 return EXIT_FAILURE;
1819 file_type = FILE_TYPE_RAW;
1822 if(!cfg.g_w || !cfg.g_h)
1824 fprintf(stderr, "Specify stream dimensions with --width (-w) "
1825 " and --height (-h).\n");
1826 return EXIT_FAILURE;
1829 #define SHOW(field) fprintf(stderr, " %-28s = %d\n", #field, cfg.field)
1831 if (verbose && pass == 0)
1833 fprintf(stderr, "Codec: %s\n", vpx_codec_iface_name(codec->iface));
1834 fprintf(stderr, "Source file: %s Format: %s\n", in_fn,
1835 arg_use_i420 ? "I420" : "YV12");
1836 fprintf(stderr, "Destination file: %s\n", out_fn);
1837 fprintf(stderr, "Encoder parameters:\n");
1844 SHOW(g_timebase.num);
1845 SHOW(g_timebase.den);
1846 SHOW(g_error_resilient);
1848 SHOW(g_lag_in_frames);
1849 SHOW(rc_dropframe_thresh);
1850 SHOW(rc_resize_allowed);
1851 SHOW(rc_resize_up_thresh);
1852 SHOW(rc_resize_down_thresh);
1854 SHOW(rc_target_bitrate);
1855 SHOW(rc_min_quantizer);
1856 SHOW(rc_max_quantizer);
1857 SHOW(rc_undershoot_pct);
1858 SHOW(rc_overshoot_pct);
1860 SHOW(rc_buf_initial_sz);
1861 SHOW(rc_buf_optimal_sz);
1862 SHOW(rc_2pass_vbr_bias_pct);
1863 SHOW(rc_2pass_vbr_minsection_pct);
1864 SHOW(rc_2pass_vbr_maxsection_pct);
1870 if(pass == (one_pass_only ? one_pass_only - 1 : 0)) {
1871 if (file_type == FILE_TYPE_Y4M)
1872 /*The Y4M reader does its own allocation.
1873 Just initialize this here to avoid problems if we never read any
1875 memset(&raw, 0, sizeof(raw));
1877 vpx_img_alloc(&raw, arg_use_i420 ? VPX_IMG_FMT_I420 : VPX_IMG_FMT_YV12,
1878 cfg.g_w, cfg.g_h, 1);
1880 init_rate_histogram(&rate_hist, &cfg, &arg_framerate);
1883 outfile = strcmp(out_fn, "-") ? fopen(out_fn, "wb")
1884 : set_binary_mode(stdout);
1888 fprintf(stderr, "Failed to open output file\n");
1889 return EXIT_FAILURE;
1892 if(write_webm && fseek(outfile, 0, SEEK_CUR))
1894 fprintf(stderr, "WebM output to pipes not supported.\n");
1895 return EXIT_FAILURE;
1900 if (!stats_open_file(&stats, stats_fn, pass))
1902 fprintf(stderr, "Failed to open statistics store\n");
1903 return EXIT_FAILURE;
1908 if (!stats_open_mem(&stats, pass))
1910 fprintf(stderr, "Failed to open statistics store\n");
1911 return EXIT_FAILURE;
1915 cfg.g_pass = arg_passes == 2
1916 ? pass ? VPX_RC_LAST_PASS : VPX_RC_FIRST_PASS
1918 #if VPX_ENCODER_ABI_VERSION > (1 + VPX_CODEC_ABI_VERSION)
1922 cfg.rc_twopass_stats_in = stats_get(&stats);
1929 ebml.stream = outfile;
1930 write_webm_file_header(&ebml, &cfg, &arg_framerate, stereo_fmt);
1933 write_ivf_file_header(outfile, &cfg, codec->fourcc, 0);
1936 /* Construct Encoder Context */
1937 vpx_codec_enc_init(&encoder, codec->iface, &cfg,
1938 show_psnr ? VPX_CODEC_USE_PSNR : 0);
1939 ctx_exit_on_error(&encoder, "Failed to initialize encoder");
1941 /* Note that we bypass the vpx_codec_control wrapper macro because
1942 * we're being clever to store the control IDs in an array. Real
1943 * applications will want to make use of the enumerations directly
1945 for (i = 0; i < arg_ctrl_cnt; i++)
1947 if (vpx_codec_control_(&encoder, arg_ctrls[i][0], arg_ctrls[i][1]))
1948 fprintf(stderr, "Error: Tried to set control %d = %d\n",
1949 arg_ctrls[i][0], arg_ctrls[i][1]);
1951 ctx_exit_on_error(&encoder, "Failed to control codec");
1957 while (frame_avail || got_data)
1959 vpx_codec_iter_t iter = NULL;
1960 const vpx_codec_cx_pkt_t *pkt;
1961 struct vpx_usec_timer timer;
1962 int64_t frame_start, next_frame_start;
1964 if (!arg_limit || frames_in < arg_limit)
1966 frame_avail = read_frame(infile, &raw, file_type, &y4m,
1973 "\rPass %d/%d frame %4d/%-4d %7"PRId64"B \033[K",
1974 pass + 1, arg_passes, frames_in, frames_out, nbytes);
1979 vpx_usec_timer_start(&timer);
1981 frame_start = (cfg.g_timebase.den * (int64_t)(frames_in - 1)
1982 * arg_framerate.den) / cfg.g_timebase.num / arg_framerate.num;
1983 next_frame_start = (cfg.g_timebase.den * (int64_t)(frames_in)
1984 * arg_framerate.den)
1985 / cfg.g_timebase.num / arg_framerate.num;
1986 vpx_codec_encode(&encoder, frame_avail ? &raw : NULL, frame_start,
1987 next_frame_start - frame_start,
1989 vpx_usec_timer_mark(&timer);
1990 cx_time += vpx_usec_timer_elapsed(&timer);
1991 ctx_exit_on_error(&encoder, "Failed to encode frame");
1993 if(cfg.g_pass != VPX_RC_FIRST_PASS)
1997 vpx_codec_control(&encoder, VP8E_GET_LAST_QUANTIZER_64, &q);
1998 ctx_exit_on_error(&encoder, "Failed to read quantizer");
2004 while ((pkt = vpx_codec_get_cx_data(&encoder, &iter)))
2010 case VPX_CODEC_CX_FRAME_PKT:
2012 fprintf(stderr, " %6luF",
2013 (unsigned long)pkt->data.frame.sz);
2015 update_rate_histogram(&rate_hist, &cfg, pkt);
2018 /* Update the hash */
2020 hash = murmur(pkt->data.frame.buf,
2021 pkt->data.frame.sz, hash);
2023 write_webm_block(&ebml, &cfg, pkt);
2027 write_ivf_frame_header(outfile, pkt);
2028 if(fwrite(pkt->data.frame.buf, 1,
2029 pkt->data.frame.sz, outfile));
2031 nbytes += pkt->data.raw.sz;
2033 case VPX_CODEC_STATS_PKT:
2035 fprintf(stderr, " %6luS",
2036 (unsigned long)pkt->data.twopass_stats.sz);
2038 pkt->data.twopass_stats.buf,
2039 pkt->data.twopass_stats.sz);
2040 nbytes += pkt->data.raw.sz;
2042 case VPX_CODEC_PSNR_PKT:
2048 psnr_sse_total += pkt->data.psnr.sse[0];
2049 psnr_samples_total += pkt->data.psnr.samples[0];
2050 for (i = 0; i < 4; i++)
2052 fprintf(stderr, "%.3lf ", pkt->data.psnr.psnr[i]);
2053 psnr_totals[i] += pkt->data.psnr.psnr[i];
2068 "\rPass %d/%d frame %4d/%-4d %7"PRId64"B %7lub/f %7"PRId64"b/s"
2069 " %7lu %s (%.2f fps)\033[K", pass + 1,
2070 arg_passes, frames_in, frames_out, nbytes,
2071 frames_in ? (unsigned long)(nbytes * 8 / frames_in) : 0,
2072 frames_in ? nbytes * 8 *(int64_t)arg_framerate.num / arg_framerate.den / frames_in : 0,
2073 cx_time > 9999999 ? cx_time / 1000 : cx_time,
2074 cx_time > 9999999 ? "ms" : "us",
2075 cx_time > 0 ? (float)frames_in * 1000000.0 / (float)cx_time : 0);
2077 if ( (show_psnr) && (psnr_count>0) )
2080 double ovpsnr = vp8_mse2psnr(psnr_samples_total, 255.0,
2083 fprintf(stderr, "\nPSNR (Overall/Avg/Y/U/V)");
2085 fprintf(stderr, " %.3lf", ovpsnr);
2086 for (i = 0; i < 4; i++)
2088 fprintf(stderr, " %.3lf", psnr_totals[i]/psnr_count);
2092 vpx_codec_destroy(&encoder);
2095 if (file_type == FILE_TYPE_Y4M)
2096 y4m_input_close(&y4m);
2100 write_webm_file_footer(&ebml, hash);
2101 free(ebml.cue_list);
2102 ebml.cue_list = NULL;
2106 if (!fseek(outfile, 0, SEEK_SET))
2107 write_ivf_file_header(outfile, &cfg, codec->fourcc, frames_out);
2111 stats_close(&stats, arg_passes-1);
2112 fprintf(stderr, "\n");
2118 if (show_q_hist_buckets)
2119 show_q_histogram(counts, show_q_hist_buckets);
2121 if (show_rate_hist_buckets)
2122 show_rate_histogram(&rate_hist, &cfg, show_rate_hist_buckets);
2123 destroy_rate_histogram(&rate_hist);
2127 return EXIT_SUCCESS;