Merge "Update armv6 loopfilter to new interface"
[libvpx.git] / vpxenc.c
blob042f07b81344993867598ace09dcfa0ba2c3c427
1 /*
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.
9 */
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
17 #else
18 #define USE_POSIX_MMAP 1
19 #endif
21 #include <stdio.h>
22 #include <stdlib.h>
23 #include <stdarg.h>
24 #include <string.h>
25 #include <limits.h>
26 #include <assert.h>
27 #include "vpx/vpx_encoder.h"
28 #if USE_POSIX_MMAP
29 #include <sys/types.h>
30 #include <sys/stat.h>
31 #include <sys/mman.h>
32 #include <fcntl.h>
33 #include <unistd.h>
34 #endif
35 #include "vpx_version.h"
36 #include "vpx/vp8cx.h"
37 #include "vpx_ports/mem_ops.h"
38 #include "vpx_ports/vpx_timer.h"
39 #include "tools_common.h"
40 #include "y4minput.h"
41 #include "libmkv/EbmlWriter.h"
42 #include "libmkv/EbmlIDs.h"
44 /* Need special handling of these functions on Windows */
45 #if defined(_MSC_VER)
46 /* MSVS doesn't define off_t, and uses _f{seek,tell}i64 */
47 typedef __int64 off_t;
48 #define fseeko _fseeki64
49 #define ftello _ftelli64
50 #elif defined(_WIN32)
51 /* MinGW defines off_t, and uses f{seek,tell}o64 */
52 #define fseeko fseeko64
53 #define ftello ftello64
54 #endif
56 #if defined(_MSC_VER)
57 #define LITERALU64(n) n
58 #else
59 #define LITERALU64(n) n##LLU
60 #endif
62 /* We should use 32-bit file operations in WebM file format
63 * when building ARM executable file (.axf) with RVCT */
64 #if !CONFIG_OS_SUPPORT
65 typedef long off_t;
66 #define fseeko fseek
67 #define ftello ftell
68 #endif
70 static const char *exec_name;
72 static const struct codec_item
74 char const *name;
75 const vpx_codec_iface_t *iface;
76 unsigned int fourcc;
77 } codecs[] =
79 #if CONFIG_VP8_ENCODER
80 {"vp8", &vpx_codec_vp8_cx_algo, 0x30385056},
81 #endif
84 static void usage_exit();
86 void die(const char *fmt, ...)
88 va_list ap;
89 va_start(ap, fmt);
90 vfprintf(stderr, fmt, ap);
91 fprintf(stderr, "\n");
92 usage_exit();
95 static void ctx_exit_on_error(vpx_codec_ctx_t *ctx, const char *s)
97 if (ctx->err)
99 const char *detail = vpx_codec_error_detail(ctx);
101 fprintf(stderr, "%s: %s\n", s, vpx_codec_error(ctx));
103 if (detail)
104 fprintf(stderr, " %s\n", detail);
106 exit(EXIT_FAILURE);
110 /* This structure is used to abstract the different ways of handling
111 * first pass statistics.
113 typedef struct
115 vpx_fixed_buf_t buf;
116 int pass;
117 FILE *file;
118 char *buf_ptr;
119 size_t buf_alloc_sz;
120 } stats_io_t;
122 int stats_open_file(stats_io_t *stats, const char *fpf, int pass)
124 int res;
126 stats->pass = pass;
128 if (pass == 0)
130 stats->file = fopen(fpf, "wb");
131 stats->buf.sz = 0;
132 stats->buf.buf = NULL,
133 res = (stats->file != NULL);
135 else
137 #if 0
138 #elif USE_POSIX_MMAP
139 struct stat stat_buf;
140 int fd;
142 fd = open(fpf, O_RDONLY);
143 stats->file = fdopen(fd, "rb");
144 fstat(fd, &stat_buf);
145 stats->buf.sz = stat_buf.st_size;
146 stats->buf.buf = mmap(NULL, stats->buf.sz, PROT_READ, MAP_PRIVATE,
147 fd, 0);
148 res = (stats->buf.buf != NULL);
149 #else
150 size_t nbytes;
152 stats->file = fopen(fpf, "rb");
154 if (fseek(stats->file, 0, SEEK_END))
156 fprintf(stderr, "First-pass stats file must be seekable!\n");
157 exit(EXIT_FAILURE);
160 stats->buf.sz = stats->buf_alloc_sz = ftell(stats->file);
161 rewind(stats->file);
163 stats->buf.buf = malloc(stats->buf_alloc_sz);
165 if (!stats->buf.buf)
167 fprintf(stderr, "Failed to allocate first-pass stats buffer (%lu bytes)\n",
168 (unsigned long)stats->buf_alloc_sz);
169 exit(EXIT_FAILURE);
172 nbytes = fread(stats->buf.buf, 1, stats->buf.sz, stats->file);
173 res = (nbytes == stats->buf.sz);
174 #endif
177 return res;
180 int stats_open_mem(stats_io_t *stats, int pass)
182 int res;
183 stats->pass = pass;
185 if (!pass)
187 stats->buf.sz = 0;
188 stats->buf_alloc_sz = 64 * 1024;
189 stats->buf.buf = malloc(stats->buf_alloc_sz);
192 stats->buf_ptr = stats->buf.buf;
193 res = (stats->buf.buf != NULL);
194 return res;
198 void stats_close(stats_io_t *stats, int last_pass)
200 if (stats->file)
202 if (stats->pass == last_pass)
204 #if 0
205 #elif USE_POSIX_MMAP
206 munmap(stats->buf.buf, stats->buf.sz);
207 #else
208 free(stats->buf.buf);
209 #endif
212 fclose(stats->file);
213 stats->file = NULL;
215 else
217 if (stats->pass == last_pass)
218 free(stats->buf.buf);
222 void stats_write(stats_io_t *stats, const void *pkt, size_t len)
224 if (stats->file)
226 if(fwrite(pkt, 1, len, stats->file));
228 else
230 if (stats->buf.sz + len > stats->buf_alloc_sz)
232 size_t new_sz = stats->buf_alloc_sz + 64 * 1024;
233 char *new_ptr = realloc(stats->buf.buf, new_sz);
235 if (new_ptr)
237 stats->buf_ptr = new_ptr + (stats->buf_ptr - (char *)stats->buf.buf);
238 stats->buf.buf = new_ptr;
239 stats->buf_alloc_sz = new_sz;
241 else
243 fprintf(stderr,
244 "\nFailed to realloc firstpass stats buffer.\n");
245 exit(EXIT_FAILURE);
249 memcpy(stats->buf_ptr, pkt, len);
250 stats->buf.sz += len;
251 stats->buf_ptr += len;
255 vpx_fixed_buf_t stats_get(stats_io_t *stats)
257 return stats->buf;
260 /* Stereo 3D packed frame format */
261 typedef enum stereo_format
263 STEREO_FORMAT_MONO = 0,
264 STEREO_FORMAT_LEFT_RIGHT = 1,
265 STEREO_FORMAT_BOTTOM_TOP = 2,
266 STEREO_FORMAT_TOP_BOTTOM = 3,
267 STEREO_FORMAT_RIGHT_LEFT = 11
268 } stereo_format_t;
270 enum video_file_type
272 FILE_TYPE_RAW,
273 FILE_TYPE_IVF,
274 FILE_TYPE_Y4M
277 struct detect_buffer {
278 char buf[4];
279 size_t buf_read;
280 size_t position;
284 #define IVF_FRAME_HDR_SZ (4+8) /* 4 byte size + 8 byte timestamp */
285 static int read_frame(FILE *f, vpx_image_t *img, unsigned int file_type,
286 y4m_input *y4m, struct detect_buffer *detect)
288 int plane = 0;
289 int shortread = 0;
291 if (file_type == FILE_TYPE_Y4M)
293 if (y4m_input_fetch_frame(y4m, f, img) < 1)
294 return 0;
296 else
298 if (file_type == FILE_TYPE_IVF)
300 char junk[IVF_FRAME_HDR_SZ];
302 /* Skip the frame header. We know how big the frame should be. See
303 * write_ivf_frame_header() for documentation on the frame header
304 * layout.
306 if(fread(junk, 1, IVF_FRAME_HDR_SZ, f));
309 for (plane = 0; plane < 3; plane++)
311 unsigned char *ptr;
312 int w = (plane ? (1 + img->d_w) / 2 : img->d_w);
313 int h = (plane ? (1 + img->d_h) / 2 : img->d_h);
314 int r;
316 /* Determine the correct plane based on the image format. The for-loop
317 * always counts in Y,U,V order, but this may not match the order of
318 * the data on disk.
320 switch (plane)
322 case 1:
323 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U];
324 break;
325 case 2:
326 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V];
327 break;
328 default:
329 ptr = img->planes[plane];
332 for (r = 0; r < h; r++)
334 size_t needed = w;
335 size_t buf_position = 0;
336 const size_t left = detect->buf_read - detect->position;
337 if (left > 0)
339 const size_t more = (left < needed) ? left : needed;
340 memcpy(ptr, detect->buf + detect->position, more);
341 buf_position = more;
342 needed -= more;
343 detect->position += more;
345 if (needed > 0)
347 shortread |= (fread(ptr + buf_position, 1, needed, f) < needed);
350 ptr += img->stride[plane];
355 return !shortread;
359 unsigned int file_is_y4m(FILE *infile,
360 y4m_input *y4m,
361 char detect[4])
363 if(memcmp(detect, "YUV4", 4) == 0)
365 return 1;
367 return 0;
370 #define IVF_FILE_HDR_SZ (32)
371 unsigned int file_is_ivf(FILE *infile,
372 unsigned int *fourcc,
373 unsigned int *width,
374 unsigned int *height,
375 struct detect_buffer *detect)
377 char raw_hdr[IVF_FILE_HDR_SZ];
378 int is_ivf = 0;
380 if(memcmp(detect->buf, "DKIF", 4) != 0)
381 return 0;
383 /* See write_ivf_file_header() for more documentation on the file header
384 * layout.
386 if (fread(raw_hdr + 4, 1, IVF_FILE_HDR_SZ - 4, infile)
387 == IVF_FILE_HDR_SZ - 4)
390 is_ivf = 1;
392 if (mem_get_le16(raw_hdr + 4) != 0)
393 fprintf(stderr, "Error: Unrecognized IVF version! This file may not"
394 " decode properly.");
396 *fourcc = mem_get_le32(raw_hdr + 8);
400 if (is_ivf)
402 *width = mem_get_le16(raw_hdr + 12);
403 *height = mem_get_le16(raw_hdr + 14);
404 detect->position = 4;
407 return is_ivf;
411 static void write_ivf_file_header(FILE *outfile,
412 const vpx_codec_enc_cfg_t *cfg,
413 unsigned int fourcc,
414 int frame_cnt)
416 char header[32];
418 if (cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS)
419 return;
421 header[0] = 'D';
422 header[1] = 'K';
423 header[2] = 'I';
424 header[3] = 'F';
425 mem_put_le16(header + 4, 0); /* version */
426 mem_put_le16(header + 6, 32); /* headersize */
427 mem_put_le32(header + 8, fourcc); /* headersize */
428 mem_put_le16(header + 12, cfg->g_w); /* width */
429 mem_put_le16(header + 14, cfg->g_h); /* height */
430 mem_put_le32(header + 16, cfg->g_timebase.den); /* rate */
431 mem_put_le32(header + 20, cfg->g_timebase.num); /* scale */
432 mem_put_le32(header + 24, frame_cnt); /* length */
433 mem_put_le32(header + 28, 0); /* unused */
435 if(fwrite(header, 1, 32, outfile));
439 static void write_ivf_frame_header(FILE *outfile,
440 const vpx_codec_cx_pkt_t *pkt)
442 char header[12];
443 vpx_codec_pts_t pts;
445 if (pkt->kind != VPX_CODEC_CX_FRAME_PKT)
446 return;
448 pts = pkt->data.frame.pts;
449 mem_put_le32(header, pkt->data.frame.sz);
450 mem_put_le32(header + 4, pts & 0xFFFFFFFF);
451 mem_put_le32(header + 8, pts >> 32);
453 if(fwrite(header, 1, 12, outfile));
457 typedef off_t EbmlLoc;
460 struct cue_entry
462 unsigned int time;
463 uint64_t loc;
467 struct EbmlGlobal
469 int debug;
471 FILE *stream;
472 int64_t last_pts_ms;
473 vpx_rational_t framerate;
475 /* These pointers are to the start of an element */
476 off_t position_reference;
477 off_t seek_info_pos;
478 off_t segment_info_pos;
479 off_t track_pos;
480 off_t cue_pos;
481 off_t cluster_pos;
483 /* This pointer is to a specific element to be serialized */
484 off_t track_id_pos;
486 /* These pointers are to the size field of the element */
487 EbmlLoc startSegment;
488 EbmlLoc startCluster;
490 uint32_t cluster_timecode;
491 int cluster_open;
493 struct cue_entry *cue_list;
494 unsigned int cues;
499 void Ebml_Write(EbmlGlobal *glob, const void *buffer_in, unsigned long len)
501 if(fwrite(buffer_in, 1, len, glob->stream));
505 void Ebml_Serialize(EbmlGlobal *glob, const void *buffer_in, unsigned long len)
507 const unsigned char *q = (const unsigned char *)buffer_in + len - 1;
509 for(; len; len--)
510 Ebml_Write(glob, q--, 1);
514 /* Need a fixed size serializer for the track ID. libmkv provdes a 64 bit
515 * one, but not a 32 bit one.
517 static void Ebml_SerializeUnsigned32(EbmlGlobal *glob, unsigned long class_id, uint64_t ui)
519 unsigned char sizeSerialized = 4 | 0x80;
520 Ebml_WriteID(glob, class_id);
521 Ebml_Serialize(glob, &sizeSerialized, 1);
522 Ebml_Serialize(glob, &ui, 4);
526 static void
527 Ebml_StartSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc,
528 unsigned long class_id)
530 //todo this is always taking 8 bytes, this may need later optimization
531 //this is a key that says lenght unknown
532 unsigned long long unknownLen = LITERALU64(0x01FFFFFFFFFFFFFF);
534 Ebml_WriteID(glob, class_id);
535 *ebmlLoc = ftello(glob->stream);
536 Ebml_Serialize(glob, &unknownLen, 8);
539 static void
540 Ebml_EndSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc)
542 off_t pos;
543 uint64_t size;
545 /* Save the current stream pointer */
546 pos = ftello(glob->stream);
548 /* Calculate the size of this element */
549 size = pos - *ebmlLoc - 8;
550 size |= LITERALU64(0x0100000000000000);
552 /* Seek back to the beginning of the element and write the new size */
553 fseeko(glob->stream, *ebmlLoc, SEEK_SET);
554 Ebml_Serialize(glob, &size, 8);
556 /* Reset the stream pointer */
557 fseeko(glob->stream, pos, SEEK_SET);
561 static void
562 write_webm_seek_element(EbmlGlobal *ebml, unsigned long id, off_t pos)
564 uint64_t offset = pos - ebml->position_reference;
565 EbmlLoc start;
566 Ebml_StartSubElement(ebml, &start, Seek);
567 Ebml_SerializeBinary(ebml, SeekID, id);
568 Ebml_SerializeUnsigned64(ebml, SeekPosition, offset);
569 Ebml_EndSubElement(ebml, &start);
573 static void
574 write_webm_seek_info(EbmlGlobal *ebml)
577 off_t pos;
579 /* Save the current stream pointer */
580 pos = ftello(ebml->stream);
582 if(ebml->seek_info_pos)
583 fseeko(ebml->stream, ebml->seek_info_pos, SEEK_SET);
584 else
585 ebml->seek_info_pos = pos;
588 EbmlLoc start;
590 Ebml_StartSubElement(ebml, &start, SeekHead);
591 write_webm_seek_element(ebml, Tracks, ebml->track_pos);
592 write_webm_seek_element(ebml, Cues, ebml->cue_pos);
593 write_webm_seek_element(ebml, Info, ebml->segment_info_pos);
594 Ebml_EndSubElement(ebml, &start);
597 //segment info
598 EbmlLoc startInfo;
599 uint64_t frame_time;
601 frame_time = (uint64_t)1000 * ebml->framerate.den
602 / ebml->framerate.num;
603 ebml->segment_info_pos = ftello(ebml->stream);
604 Ebml_StartSubElement(ebml, &startInfo, Info);
605 Ebml_SerializeUnsigned(ebml, TimecodeScale, 1000000);
606 Ebml_SerializeFloat(ebml, Segment_Duration,
607 ebml->last_pts_ms + frame_time);
608 Ebml_SerializeString(ebml, 0x4D80,
609 ebml->debug ? "vpxenc" : "vpxenc" VERSION_STRING);
610 Ebml_SerializeString(ebml, 0x5741,
611 ebml->debug ? "vpxenc" : "vpxenc" VERSION_STRING);
612 Ebml_EndSubElement(ebml, &startInfo);
617 static void
618 write_webm_file_header(EbmlGlobal *glob,
619 const vpx_codec_enc_cfg_t *cfg,
620 const struct vpx_rational *fps,
621 stereo_format_t stereo_fmt)
624 EbmlLoc start;
625 Ebml_StartSubElement(glob, &start, EBML);
626 Ebml_SerializeUnsigned(glob, EBMLVersion, 1);
627 Ebml_SerializeUnsigned(glob, EBMLReadVersion, 1); //EBML Read Version
628 Ebml_SerializeUnsigned(glob, EBMLMaxIDLength, 4); //EBML Max ID Length
629 Ebml_SerializeUnsigned(glob, EBMLMaxSizeLength, 8); //EBML Max Size Length
630 Ebml_SerializeString(glob, DocType, "webm"); //Doc Type
631 Ebml_SerializeUnsigned(glob, DocTypeVersion, 2); //Doc Type Version
632 Ebml_SerializeUnsigned(glob, DocTypeReadVersion, 2); //Doc Type Read Version
633 Ebml_EndSubElement(glob, &start);
636 Ebml_StartSubElement(glob, &glob->startSegment, Segment); //segment
637 glob->position_reference = ftello(glob->stream);
638 glob->framerate = *fps;
639 write_webm_seek_info(glob);
642 EbmlLoc trackStart;
643 glob->track_pos = ftello(glob->stream);
644 Ebml_StartSubElement(glob, &trackStart, Tracks);
646 unsigned int trackNumber = 1;
647 uint64_t trackID = 0;
649 EbmlLoc start;
650 Ebml_StartSubElement(glob, &start, TrackEntry);
651 Ebml_SerializeUnsigned(glob, TrackNumber, trackNumber);
652 glob->track_id_pos = ftello(glob->stream);
653 Ebml_SerializeUnsigned32(glob, TrackUID, trackID);
654 Ebml_SerializeUnsigned(glob, TrackType, 1); //video is always 1
655 Ebml_SerializeString(glob, CodecID, "V_VP8");
657 unsigned int pixelWidth = cfg->g_w;
658 unsigned int pixelHeight = cfg->g_h;
659 float frameRate = (float)fps->num/(float)fps->den;
661 EbmlLoc videoStart;
662 Ebml_StartSubElement(glob, &videoStart, Video);
663 Ebml_SerializeUnsigned(glob, PixelWidth, pixelWidth);
664 Ebml_SerializeUnsigned(glob, PixelHeight, pixelHeight);
665 Ebml_SerializeUnsigned(glob, StereoMode, stereo_fmt);
666 Ebml_SerializeFloat(glob, FrameRate, frameRate);
667 Ebml_EndSubElement(glob, &videoStart); //Video
669 Ebml_EndSubElement(glob, &start); //Track Entry
671 Ebml_EndSubElement(glob, &trackStart);
673 // segment element is open
678 static void
679 write_webm_block(EbmlGlobal *glob,
680 const vpx_codec_enc_cfg_t *cfg,
681 const vpx_codec_cx_pkt_t *pkt)
683 unsigned long block_length;
684 unsigned char track_number;
685 unsigned short block_timecode = 0;
686 unsigned char flags;
687 int64_t pts_ms;
688 int start_cluster = 0, is_keyframe;
690 /* Calculate the PTS of this frame in milliseconds */
691 pts_ms = pkt->data.frame.pts * 1000
692 * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
693 if(pts_ms <= glob->last_pts_ms)
694 pts_ms = glob->last_pts_ms + 1;
695 glob->last_pts_ms = pts_ms;
697 /* Calculate the relative time of this block */
698 if(pts_ms - glob->cluster_timecode > SHRT_MAX)
699 start_cluster = 1;
700 else
701 block_timecode = pts_ms - glob->cluster_timecode;
703 is_keyframe = (pkt->data.frame.flags & VPX_FRAME_IS_KEY);
704 if(start_cluster || is_keyframe)
706 if(glob->cluster_open)
707 Ebml_EndSubElement(glob, &glob->startCluster);
709 /* Open the new cluster */
710 block_timecode = 0;
711 glob->cluster_open = 1;
712 glob->cluster_timecode = pts_ms;
713 glob->cluster_pos = ftello(glob->stream);
714 Ebml_StartSubElement(glob, &glob->startCluster, Cluster); //cluster
715 Ebml_SerializeUnsigned(glob, Timecode, glob->cluster_timecode);
717 /* Save a cue point if this is a keyframe. */
718 if(is_keyframe)
720 struct cue_entry *cue, *new_cue_list;
722 new_cue_list = realloc(glob->cue_list,
723 (glob->cues+1) * sizeof(struct cue_entry));
724 if(new_cue_list)
725 glob->cue_list = new_cue_list;
726 else
728 fprintf(stderr, "\nFailed to realloc cue list.\n");
729 exit(EXIT_FAILURE);
732 cue = &glob->cue_list[glob->cues];
733 cue->time = glob->cluster_timecode;
734 cue->loc = glob->cluster_pos;
735 glob->cues++;
739 /* Write the Simple Block */
740 Ebml_WriteID(glob, SimpleBlock);
742 block_length = pkt->data.frame.sz + 4;
743 block_length |= 0x10000000;
744 Ebml_Serialize(glob, &block_length, 4);
746 track_number = 1;
747 track_number |= 0x80;
748 Ebml_Write(glob, &track_number, 1);
750 Ebml_Serialize(glob, &block_timecode, 2);
752 flags = 0;
753 if(is_keyframe)
754 flags |= 0x80;
755 if(pkt->data.frame.flags & VPX_FRAME_IS_INVISIBLE)
756 flags |= 0x08;
757 Ebml_Write(glob, &flags, 1);
759 Ebml_Write(glob, pkt->data.frame.buf, pkt->data.frame.sz);
763 static void
764 write_webm_file_footer(EbmlGlobal *glob, long hash)
767 if(glob->cluster_open)
768 Ebml_EndSubElement(glob, &glob->startCluster);
771 EbmlLoc start;
772 int i;
774 glob->cue_pos = ftello(glob->stream);
775 Ebml_StartSubElement(glob, &start, Cues);
776 for(i=0; i<glob->cues; i++)
778 struct cue_entry *cue = &glob->cue_list[i];
779 EbmlLoc start;
781 Ebml_StartSubElement(glob, &start, CuePoint);
783 EbmlLoc start;
785 Ebml_SerializeUnsigned(glob, CueTime, cue->time);
787 Ebml_StartSubElement(glob, &start, CueTrackPositions);
788 Ebml_SerializeUnsigned(glob, CueTrack, 1);
789 Ebml_SerializeUnsigned64(glob, CueClusterPosition,
790 cue->loc - glob->position_reference);
791 //Ebml_SerializeUnsigned(glob, CueBlockNumber, cue->blockNumber);
792 Ebml_EndSubElement(glob, &start);
794 Ebml_EndSubElement(glob, &start);
796 Ebml_EndSubElement(glob, &start);
799 Ebml_EndSubElement(glob, &glob->startSegment);
801 /* Patch up the seek info block */
802 write_webm_seek_info(glob);
804 /* Patch up the track id */
805 fseeko(glob->stream, glob->track_id_pos, SEEK_SET);
806 Ebml_SerializeUnsigned32(glob, TrackUID, glob->debug ? 0xDEADBEEF : hash);
808 fseeko(glob->stream, 0, SEEK_END);
812 /* Murmur hash derived from public domain reference implementation at
813 * http://sites.google.com/site/murmurhash/
815 static unsigned int murmur ( const void * key, int len, unsigned int seed )
817 const unsigned int m = 0x5bd1e995;
818 const int r = 24;
820 unsigned int h = seed ^ len;
822 const unsigned char * data = (const unsigned char *)key;
824 while(len >= 4)
826 unsigned int k;
828 k = data[0];
829 k |= data[1] << 8;
830 k |= data[2] << 16;
831 k |= data[3] << 24;
833 k *= m;
834 k ^= k >> r;
835 k *= m;
837 h *= m;
838 h ^= k;
840 data += 4;
841 len -= 4;
844 switch(len)
846 case 3: h ^= data[2] << 16;
847 case 2: h ^= data[1] << 8;
848 case 1: h ^= data[0];
849 h *= m;
852 h ^= h >> 13;
853 h *= m;
854 h ^= h >> 15;
856 return h;
859 #include "math.h"
861 static double vp8_mse2psnr(double Samples, double Peak, double Mse)
863 double psnr;
865 if ((double)Mse > 0.0)
866 psnr = 10.0 * log10(Peak * Peak * Samples / Mse);
867 else
868 psnr = 60; // Limit to prevent / 0
870 if (psnr > 60)
871 psnr = 60;
873 return psnr;
877 #include "args.h"
879 static const arg_def_t debugmode = ARG_DEF("D", "debug", 0,
880 "Debug mode (makes output deterministic)");
881 static const arg_def_t outputfile = ARG_DEF("o", "output", 1,
882 "Output filename");
883 static const arg_def_t use_yv12 = ARG_DEF(NULL, "yv12", 0,
884 "Input file is YV12 ");
885 static const arg_def_t use_i420 = ARG_DEF(NULL, "i420", 0,
886 "Input file is I420 (default)");
887 static const arg_def_t codecarg = ARG_DEF(NULL, "codec", 1,
888 "Codec to use");
889 static const arg_def_t passes = ARG_DEF("p", "passes", 1,
890 "Number of passes (1/2)");
891 static const arg_def_t pass_arg = ARG_DEF(NULL, "pass", 1,
892 "Pass to execute (1/2)");
893 static const arg_def_t fpf_name = ARG_DEF(NULL, "fpf", 1,
894 "First pass statistics file name");
895 static const arg_def_t limit = ARG_DEF(NULL, "limit", 1,
896 "Stop encoding after n input frames");
897 static const arg_def_t deadline = ARG_DEF("d", "deadline", 1,
898 "Deadline per frame (usec)");
899 static const arg_def_t best_dl = ARG_DEF(NULL, "best", 0,
900 "Use Best Quality Deadline");
901 static const arg_def_t good_dl = ARG_DEF(NULL, "good", 0,
902 "Use Good Quality Deadline");
903 static const arg_def_t rt_dl = ARG_DEF(NULL, "rt", 0,
904 "Use Realtime Quality Deadline");
905 static const arg_def_t verbosearg = ARG_DEF("v", "verbose", 0,
906 "Show encoder parameters");
907 static const arg_def_t psnrarg = ARG_DEF(NULL, "psnr", 0,
908 "Show PSNR in status line");
909 static const arg_def_t framerate = ARG_DEF(NULL, "fps", 1,
910 "Stream frame rate (rate/scale)");
911 static const arg_def_t use_ivf = ARG_DEF(NULL, "ivf", 0,
912 "Output IVF (default is WebM)");
913 static const arg_def_t q_hist_n = ARG_DEF(NULL, "q-hist", 1,
914 "Show quantizer histogram (n-buckets)");
915 static const arg_def_t rate_hist_n = ARG_DEF(NULL, "rate-hist", 1,
916 "Show rate histogram (n-buckets)");
917 static const arg_def_t *main_args[] =
919 &debugmode,
920 &outputfile, &codecarg, &passes, &pass_arg, &fpf_name, &limit, &deadline,
921 &best_dl, &good_dl, &rt_dl,
922 &verbosearg, &psnrarg, &use_ivf, &q_hist_n, &rate_hist_n,
923 NULL
926 static const arg_def_t usage = ARG_DEF("u", "usage", 1,
927 "Usage profile number to use");
928 static const arg_def_t threads = ARG_DEF("t", "threads", 1,
929 "Max number of threads to use");
930 static const arg_def_t profile = ARG_DEF(NULL, "profile", 1,
931 "Bitstream profile number to use");
932 static const arg_def_t width = ARG_DEF("w", "width", 1,
933 "Frame width");
934 static const arg_def_t height = ARG_DEF("h", "height", 1,
935 "Frame height");
936 static const struct arg_enum_list stereo_mode_enum[] = {
937 {"mono" , STEREO_FORMAT_MONO},
938 {"left-right", STEREO_FORMAT_LEFT_RIGHT},
939 {"bottom-top", STEREO_FORMAT_BOTTOM_TOP},
940 {"top-bottom", STEREO_FORMAT_TOP_BOTTOM},
941 {"right-left", STEREO_FORMAT_RIGHT_LEFT},
942 {NULL, 0}
944 static const arg_def_t stereo_mode = ARG_DEF_ENUM(NULL, "stereo-mode", 1,
945 "Stereo 3D video format", stereo_mode_enum);
946 static const arg_def_t timebase = ARG_DEF(NULL, "timebase", 1,
947 "Stream timebase (frame duration)");
948 static const arg_def_t error_resilient = ARG_DEF(NULL, "error-resilient", 1,
949 "Enable error resiliency features");
950 static const arg_def_t lag_in_frames = ARG_DEF(NULL, "lag-in-frames", 1,
951 "Max number of frames to lag");
953 static const arg_def_t *global_args[] =
955 &use_yv12, &use_i420, &usage, &threads, &profile,
956 &width, &height, &stereo_mode, &timebase, &framerate, &error_resilient,
957 &lag_in_frames, NULL
960 static const arg_def_t dropframe_thresh = ARG_DEF(NULL, "drop-frame", 1,
961 "Temporal resampling threshold (buf %)");
962 static const arg_def_t resize_allowed = ARG_DEF(NULL, "resize-allowed", 1,
963 "Spatial resampling enabled (bool)");
964 static const arg_def_t resize_up_thresh = ARG_DEF(NULL, "resize-up", 1,
965 "Upscale threshold (buf %)");
966 static const arg_def_t resize_down_thresh = ARG_DEF(NULL, "resize-down", 1,
967 "Downscale threshold (buf %)");
968 static const struct arg_enum_list end_usage_enum[] = {
969 {"vbr", VPX_VBR},
970 {"cbr", VPX_CBR},
971 {"cq", VPX_CQ},
972 {NULL, 0}
974 static const arg_def_t end_usage = ARG_DEF_ENUM(NULL, "end-usage", 1,
975 "Rate control mode", end_usage_enum);
976 static const arg_def_t target_bitrate = ARG_DEF(NULL, "target-bitrate", 1,
977 "Bitrate (kbps)");
978 static const arg_def_t min_quantizer = ARG_DEF(NULL, "min-q", 1,
979 "Minimum (best) quantizer");
980 static const arg_def_t max_quantizer = ARG_DEF(NULL, "max-q", 1,
981 "Maximum (worst) quantizer");
982 static const arg_def_t undershoot_pct = ARG_DEF(NULL, "undershoot-pct", 1,
983 "Datarate undershoot (min) target (%)");
984 static const arg_def_t overshoot_pct = ARG_DEF(NULL, "overshoot-pct", 1,
985 "Datarate overshoot (max) target (%)");
986 static const arg_def_t buf_sz = ARG_DEF(NULL, "buf-sz", 1,
987 "Client buffer size (ms)");
988 static const arg_def_t buf_initial_sz = ARG_DEF(NULL, "buf-initial-sz", 1,
989 "Client initial buffer size (ms)");
990 static const arg_def_t buf_optimal_sz = ARG_DEF(NULL, "buf-optimal-sz", 1,
991 "Client optimal buffer size (ms)");
992 static const arg_def_t max_intra_rate_pct = ARG_DEF(NULL, "max-intra-rate", 1,
993 "Max I-frame bitrate (pct)");
994 static const arg_def_t *rc_args[] =
996 &dropframe_thresh, &resize_allowed, &resize_up_thresh, &resize_down_thresh,
997 &end_usage, &target_bitrate, &min_quantizer, &max_quantizer,
998 &undershoot_pct, &overshoot_pct, &buf_sz, &buf_initial_sz, &buf_optimal_sz,
999 &max_intra_rate_pct,
1000 NULL
1004 static const arg_def_t bias_pct = ARG_DEF(NULL, "bias-pct", 1,
1005 "CBR/VBR bias (0=CBR, 100=VBR)");
1006 static const arg_def_t minsection_pct = ARG_DEF(NULL, "minsection-pct", 1,
1007 "GOP min bitrate (% of target)");
1008 static const arg_def_t maxsection_pct = ARG_DEF(NULL, "maxsection-pct", 1,
1009 "GOP max bitrate (% of target)");
1010 static const arg_def_t *rc_twopass_args[] =
1012 &bias_pct, &minsection_pct, &maxsection_pct, NULL
1016 static const arg_def_t kf_min_dist = ARG_DEF(NULL, "kf-min-dist", 1,
1017 "Minimum keyframe interval (frames)");
1018 static const arg_def_t kf_max_dist = ARG_DEF(NULL, "kf-max-dist", 1,
1019 "Maximum keyframe interval (frames)");
1020 static const arg_def_t kf_disabled = ARG_DEF(NULL, "disable-kf", 0,
1021 "Disable keyframe placement");
1022 static const arg_def_t *kf_args[] =
1024 &kf_min_dist, &kf_max_dist, &kf_disabled, NULL
1028 #if CONFIG_VP8_ENCODER
1029 static const arg_def_t noise_sens = ARG_DEF(NULL, "noise-sensitivity", 1,
1030 "Noise sensitivity (frames to blur)");
1031 static const arg_def_t sharpness = ARG_DEF(NULL, "sharpness", 1,
1032 "Filter sharpness (0-7)");
1033 static const arg_def_t static_thresh = ARG_DEF(NULL, "static-thresh", 1,
1034 "Motion detection threshold");
1035 #endif
1037 #if CONFIG_VP8_ENCODER
1038 static const arg_def_t cpu_used = ARG_DEF(NULL, "cpu-used", 1,
1039 "CPU Used (-16..16)");
1040 #endif
1043 #if CONFIG_VP8_ENCODER
1044 static const arg_def_t token_parts = ARG_DEF(NULL, "token-parts", 1,
1045 "Number of token partitions to use, log2");
1046 static const arg_def_t auto_altref = ARG_DEF(NULL, "auto-alt-ref", 1,
1047 "Enable automatic alt reference frames");
1048 static const arg_def_t arnr_maxframes = ARG_DEF(NULL, "arnr-maxframes", 1,
1049 "AltRef Max Frames");
1050 static const arg_def_t arnr_strength = ARG_DEF(NULL, "arnr-strength", 1,
1051 "AltRef Strength");
1052 static const arg_def_t arnr_type = ARG_DEF(NULL, "arnr-type", 1,
1053 "AltRef Type");
1054 static const struct arg_enum_list tuning_enum[] = {
1055 {"psnr", VP8_TUNE_PSNR},
1056 {"ssim", VP8_TUNE_SSIM},
1057 {NULL, 0}
1059 static const arg_def_t tune_ssim = ARG_DEF_ENUM(NULL, "tune", 1,
1060 "Material to favor", tuning_enum);
1061 static const arg_def_t cq_level = ARG_DEF(NULL, "cq-level", 1,
1062 "Constrained Quality Level");
1064 static const arg_def_t *vp8_args[] =
1066 &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh,
1067 &token_parts, &arnr_maxframes, &arnr_strength, &arnr_type,
1068 &tune_ssim, &cq_level, NULL
1070 static const int vp8_arg_ctrl_map[] =
1072 VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF,
1073 VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD,
1074 VP8E_SET_TOKEN_PARTITIONS,
1075 VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH , VP8E_SET_ARNR_TYPE,
1076 VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, 0
1078 #endif
1080 static const arg_def_t *no_args[] = { NULL };
1082 static void usage_exit()
1084 int i;
1086 fprintf(stderr, "Usage: %s <options> -o dst_filename src_filename \n",
1087 exec_name);
1089 fprintf(stderr, "\nOptions:\n");
1090 arg_show_usage(stdout, main_args);
1091 fprintf(stderr, "\nEncoder Global Options:\n");
1092 arg_show_usage(stdout, global_args);
1093 fprintf(stderr, "\nRate Control Options:\n");
1094 arg_show_usage(stdout, rc_args);
1095 fprintf(stderr, "\nTwopass Rate Control Options:\n");
1096 arg_show_usage(stdout, rc_twopass_args);
1097 fprintf(stderr, "\nKeyframe Placement Options:\n");
1098 arg_show_usage(stdout, kf_args);
1099 #if CONFIG_VP8_ENCODER
1100 fprintf(stderr, "\nVP8 Specific Options:\n");
1101 arg_show_usage(stdout, vp8_args);
1102 #endif
1103 fprintf(stderr, "\n"
1104 "Included encoders:\n"
1105 "\n");
1107 for (i = 0; i < sizeof(codecs) / sizeof(codecs[0]); i++)
1108 fprintf(stderr, " %-6s - %s\n",
1109 codecs[i].name,
1110 vpx_codec_iface_name(codecs[i].iface));
1112 exit(EXIT_FAILURE);
1116 #define HIST_BAR_MAX 40
1117 struct hist_bucket
1119 int low, high, count;
1123 static int merge_hist_buckets(struct hist_bucket *bucket,
1124 int *buckets_,
1125 int max_buckets)
1127 int small_bucket = 0, merge_bucket = INT_MAX, big_bucket=0;
1128 int buckets = *buckets_;
1129 int i;
1131 /* Find the extrema for this list of buckets */
1132 big_bucket = small_bucket = 0;
1133 for(i=0; i < buckets; i++)
1135 if(bucket[i].count < bucket[small_bucket].count)
1136 small_bucket = i;
1137 if(bucket[i].count > bucket[big_bucket].count)
1138 big_bucket = i;
1141 /* If we have too many buckets, merge the smallest with an ajacent
1142 * bucket.
1144 while(buckets > max_buckets)
1146 int last_bucket = buckets - 1;
1148 // merge the small bucket with an adjacent one.
1149 if(small_bucket == 0)
1150 merge_bucket = 1;
1151 else if(small_bucket == last_bucket)
1152 merge_bucket = last_bucket - 1;
1153 else if(bucket[small_bucket - 1].count < bucket[small_bucket + 1].count)
1154 merge_bucket = small_bucket - 1;
1155 else
1156 merge_bucket = small_bucket + 1;
1158 assert(abs(merge_bucket - small_bucket) <= 1);
1159 assert(small_bucket < buckets);
1160 assert(big_bucket < buckets);
1161 assert(merge_bucket < buckets);
1163 if(merge_bucket < small_bucket)
1165 bucket[merge_bucket].high = bucket[small_bucket].high;
1166 bucket[merge_bucket].count += bucket[small_bucket].count;
1168 else
1170 bucket[small_bucket].high = bucket[merge_bucket].high;
1171 bucket[small_bucket].count += bucket[merge_bucket].count;
1172 merge_bucket = small_bucket;
1175 assert(bucket[merge_bucket].low != bucket[merge_bucket].high);
1177 buckets--;
1179 /* Remove the merge_bucket from the list, and find the new small
1180 * and big buckets while we're at it
1182 big_bucket = small_bucket = 0;
1183 for(i=0; i < buckets; i++)
1185 if(i > merge_bucket)
1186 bucket[i] = bucket[i+1];
1188 if(bucket[i].count < bucket[small_bucket].count)
1189 small_bucket = i;
1190 if(bucket[i].count > bucket[big_bucket].count)
1191 big_bucket = i;
1196 *buckets_ = buckets;
1197 return bucket[big_bucket].count;
1201 static void show_histogram(const struct hist_bucket *bucket,
1202 int buckets,
1203 int total,
1204 int scale)
1206 const char *pat1, *pat2;
1207 int i;
1209 switch((int)(log(bucket[buckets-1].high)/log(10))+1)
1211 case 1:
1212 case 2:
1213 pat1 = "%4d %2s: ";
1214 pat2 = "%4d-%2d: ";
1215 break;
1216 case 3:
1217 pat1 = "%5d %3s: ";
1218 pat2 = "%5d-%3d: ";
1219 break;
1220 case 4:
1221 pat1 = "%6d %4s: ";
1222 pat2 = "%6d-%4d: ";
1223 break;
1224 case 5:
1225 pat1 = "%7d %5s: ";
1226 pat2 = "%7d-%5d: ";
1227 break;
1228 case 6:
1229 pat1 = "%8d %6s: ";
1230 pat2 = "%8d-%6d: ";
1231 break;
1232 case 7:
1233 pat1 = "%9d %7s: ";
1234 pat2 = "%9d-%7d: ";
1235 break;
1236 default:
1237 pat1 = "%12d %10s: ";
1238 pat2 = "%12d-%10d: ";
1239 break;
1242 for(i=0; i<buckets; i++)
1244 int len;
1245 int j;
1246 float pct;
1248 pct = 100.0 * (float)bucket[i].count / (float)total;
1249 len = HIST_BAR_MAX * bucket[i].count / scale;
1250 if(len < 1)
1251 len = 1;
1252 assert(len <= HIST_BAR_MAX);
1254 if(bucket[i].low == bucket[i].high)
1255 fprintf(stderr, pat1, bucket[i].low, "");
1256 else
1257 fprintf(stderr, pat2, bucket[i].low, bucket[i].high);
1259 for(j=0; j<HIST_BAR_MAX; j++)
1260 fprintf(stderr, j<len?"=":" ");
1261 fprintf(stderr, "\t%5d (%6.2f%%)\n",bucket[i].count,pct);
1266 static void show_q_histogram(const int counts[64], int max_buckets)
1268 struct hist_bucket bucket[64];
1269 int buckets = 0;
1270 int total = 0;
1271 int scale;
1272 int i;
1275 for(i=0; i<64; i++)
1277 if(counts[i])
1279 bucket[buckets].low = bucket[buckets].high = i;
1280 bucket[buckets].count = counts[i];
1281 buckets++;
1282 total += counts[i];
1286 fprintf(stderr, "\nQuantizer Selection:\n");
1287 scale = merge_hist_buckets(bucket, &buckets, max_buckets);
1288 show_histogram(bucket, buckets, total, scale);
1292 #define RATE_BINS (100)
1293 struct rate_hist
1295 int64_t *pts;
1296 int *sz;
1297 int samples;
1298 int frames;
1299 struct hist_bucket bucket[RATE_BINS];
1300 int total;
1304 static void init_rate_histogram(struct rate_hist *hist,
1305 const vpx_codec_enc_cfg_t *cfg,
1306 const vpx_rational_t *fps)
1308 int i;
1310 /* Determine the number of samples in the buffer. Use the file's framerate
1311 * to determine the number of frames in rc_buf_sz milliseconds, with an
1312 * adjustment (5/4) to account for alt-refs
1314 hist->samples = cfg->rc_buf_sz * 5 / 4 * fps->num / fps->den / 1000;
1315 hist->pts = calloc(hist->samples, sizeof(*hist->pts));
1316 hist->sz = calloc(hist->samples, sizeof(*hist->sz));
1317 for(i=0; i<RATE_BINS; i++)
1319 hist->bucket[i].low = INT_MAX;
1320 hist->bucket[i].high = 0;
1321 hist->bucket[i].count = 0;
1326 static void destroy_rate_histogram(struct rate_hist *hist)
1328 free(hist->pts);
1329 free(hist->sz);
1333 static void update_rate_histogram(struct rate_hist *hist,
1334 const vpx_codec_enc_cfg_t *cfg,
1335 const vpx_codec_cx_pkt_t *pkt)
1337 int i, idx;
1338 int64_t now, then, sum_sz = 0, avg_bitrate;
1340 now = pkt->data.frame.pts * 1000
1341 * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
1343 idx = hist->frames++ % hist->samples;
1344 hist->pts[idx] = now;
1345 hist->sz[idx] = pkt->data.frame.sz;
1347 if(now < cfg->rc_buf_initial_sz)
1348 return;
1350 then = now;
1352 /* Sum the size over the past rc_buf_sz ms */
1353 for(i = hist->frames; i > 0 && hist->frames - i < hist->samples; i--)
1355 int i_idx = (i-1) % hist->samples;
1357 then = hist->pts[i_idx];
1358 if(now - then > cfg->rc_buf_sz)
1359 break;
1360 sum_sz += hist->sz[i_idx];
1363 if (now == then)
1364 return;
1366 avg_bitrate = sum_sz * 8 * 1000 / (now - then);
1367 idx = avg_bitrate * (RATE_BINS/2) / (cfg->rc_target_bitrate * 1000);
1368 if(idx < 0)
1369 idx = 0;
1370 if(idx > RATE_BINS-1)
1371 idx = RATE_BINS-1;
1372 if(hist->bucket[idx].low > avg_bitrate)
1373 hist->bucket[idx].low = avg_bitrate;
1374 if(hist->bucket[idx].high < avg_bitrate)
1375 hist->bucket[idx].high = avg_bitrate;
1376 hist->bucket[idx].count++;
1377 hist->total++;
1381 static void show_rate_histogram(struct rate_hist *hist,
1382 const vpx_codec_enc_cfg_t *cfg,
1383 int max_buckets)
1385 int i, scale;
1386 int buckets = 0;
1388 for(i = 0; i < RATE_BINS; i++)
1390 if(hist->bucket[i].low == INT_MAX)
1391 continue;
1392 hist->bucket[buckets++] = hist->bucket[i];
1395 fprintf(stderr, "\nRate (over %dms window):\n", cfg->rc_buf_sz);
1396 scale = merge_hist_buckets(hist->bucket, &buckets, max_buckets);
1397 show_histogram(hist->bucket, buckets, hist->total, scale);
1400 #define ARG_CTRL_CNT_MAX 10
1402 int main(int argc, const char **argv_)
1404 vpx_codec_ctx_t encoder;
1405 const char *in_fn = NULL, *out_fn = NULL, *stats_fn = NULL;
1406 int i;
1407 FILE *infile, *outfile;
1408 vpx_codec_enc_cfg_t cfg;
1409 vpx_codec_err_t res;
1410 int pass, one_pass_only = 0;
1411 stats_io_t stats;
1412 vpx_image_t raw;
1413 const struct codec_item *codec = codecs;
1414 int frame_avail, got_data;
1416 struct arg arg;
1417 char **argv, **argi, **argj;
1418 int arg_usage = 0, arg_passes = 1, arg_deadline = 0;
1419 int arg_ctrls[ARG_CTRL_CNT_MAX][2], arg_ctrl_cnt = 0;
1420 int arg_limit = 0;
1421 static const arg_def_t **ctrl_args = no_args;
1422 static const int *ctrl_args_map = NULL;
1423 int verbose = 0, show_psnr = 0;
1424 int arg_use_i420 = 1;
1425 unsigned long cx_time = 0;
1426 unsigned int file_type, fourcc;
1427 y4m_input y4m;
1428 struct vpx_rational arg_framerate = {30, 1};
1429 int arg_have_framerate = 0;
1430 int write_webm = 1;
1431 EbmlGlobal ebml = {0};
1432 uint32_t hash = 0;
1433 uint64_t psnr_sse_total = 0;
1434 uint64_t psnr_samples_total = 0;
1435 double psnr_totals[4] = {0, 0, 0, 0};
1436 int psnr_count = 0;
1437 stereo_format_t stereo_fmt = STEREO_FORMAT_MONO;
1438 int counts[64]={0};
1439 int show_q_hist_buckets=0;
1440 int show_rate_hist_buckets=0;
1441 struct rate_hist rate_hist={0};
1443 exec_name = argv_[0];
1444 ebml.last_pts_ms = -1;
1446 if (argc < 3)
1447 usage_exit();
1450 /* First parse the codec and usage values, because we want to apply other
1451 * parameters on top of the default configuration provided by the codec.
1453 argv = argv_dup(argc - 1, argv_ + 1);
1455 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
1457 arg.argv_step = 1;
1459 if (arg_match(&arg, &codecarg, argi))
1461 int j, k = -1;
1463 for (j = 0; j < sizeof(codecs) / sizeof(codecs[0]); j++)
1464 if (!strcmp(codecs[j].name, arg.val))
1465 k = j;
1467 if (k >= 0)
1468 codec = codecs + k;
1469 else
1470 die("Error: Unrecognized argument (%s) to --codec\n",
1471 arg.val);
1474 else if (arg_match(&arg, &passes, argi))
1476 arg_passes = arg_parse_uint(&arg);
1478 if (arg_passes < 1 || arg_passes > 2)
1479 die("Error: Invalid number of passes (%d)\n", arg_passes);
1481 else if (arg_match(&arg, &pass_arg, argi))
1483 one_pass_only = arg_parse_uint(&arg);
1485 if (one_pass_only < 1 || one_pass_only > 2)
1486 die("Error: Invalid pass selected (%d)\n", one_pass_only);
1488 else if (arg_match(&arg, &fpf_name, argi))
1489 stats_fn = arg.val;
1490 else if (arg_match(&arg, &usage, argi))
1491 arg_usage = arg_parse_uint(&arg);
1492 else if (arg_match(&arg, &deadline, argi))
1493 arg_deadline = arg_parse_uint(&arg);
1494 else if (arg_match(&arg, &best_dl, argi))
1495 arg_deadline = VPX_DL_BEST_QUALITY;
1496 else if (arg_match(&arg, &good_dl, argi))
1497 arg_deadline = VPX_DL_GOOD_QUALITY;
1498 else if (arg_match(&arg, &rt_dl, argi))
1499 arg_deadline = VPX_DL_REALTIME;
1500 else if (arg_match(&arg, &use_yv12, argi))
1502 arg_use_i420 = 0;
1504 else if (arg_match(&arg, &use_i420, argi))
1506 arg_use_i420 = 1;
1508 else if (arg_match(&arg, &verbosearg, argi))
1509 verbose = 1;
1510 else if (arg_match(&arg, &limit, argi))
1511 arg_limit = arg_parse_uint(&arg);
1512 else if (arg_match(&arg, &psnrarg, argi))
1513 show_psnr = 1;
1514 else if (arg_match(&arg, &framerate, argi))
1516 arg_framerate = arg_parse_rational(&arg);
1517 arg_have_framerate = 1;
1519 else if (arg_match(&arg, &use_ivf, argi))
1520 write_webm = 0;
1521 else if (arg_match(&arg, &outputfile, argi))
1522 out_fn = arg.val;
1523 else if (arg_match(&arg, &debugmode, argi))
1524 ebml.debug = 1;
1525 else if (arg_match(&arg, &q_hist_n, argi))
1526 show_q_hist_buckets = arg_parse_uint(&arg);
1527 else if (arg_match(&arg, &rate_hist_n, argi))
1528 show_rate_hist_buckets = arg_parse_uint(&arg);
1529 else
1530 argj++;
1533 /* Ensure that --passes and --pass are consistent. If --pass is set and --passes=2,
1534 * ensure --fpf was set.
1536 if (one_pass_only)
1538 /* DWIM: Assume the user meant passes=2 if pass=2 is specified */
1539 if (one_pass_only > arg_passes)
1541 fprintf(stderr, "Warning: Assuming --pass=%d implies --passes=%d\n",
1542 one_pass_only, one_pass_only);
1543 arg_passes = one_pass_only;
1546 if (arg_passes == 2 && !stats_fn)
1547 die("Must specify --fpf when --pass=%d and --passes=2\n", one_pass_only);
1550 /* Populate encoder configuration */
1551 res = vpx_codec_enc_config_default(codec->iface, &cfg, arg_usage);
1553 if (res)
1555 fprintf(stderr, "Failed to get config: %s\n",
1556 vpx_codec_err_to_string(res));
1557 return EXIT_FAILURE;
1560 /* Change the default timebase to a high enough value so that the encoder
1561 * will always create strictly increasing timestamps.
1563 cfg.g_timebase.den = 1000;
1565 /* Never use the library's default resolution, require it be parsed
1566 * from the file or set on the command line.
1568 cfg.g_w = 0;
1569 cfg.g_h = 0;
1571 /* Now parse the remainder of the parameters. */
1572 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
1574 arg.argv_step = 1;
1576 if (0);
1577 else if (arg_match(&arg, &threads, argi))
1578 cfg.g_threads = arg_parse_uint(&arg);
1579 else if (arg_match(&arg, &profile, argi))
1580 cfg.g_profile = arg_parse_uint(&arg);
1581 else if (arg_match(&arg, &width, argi))
1582 cfg.g_w = arg_parse_uint(&arg);
1583 else if (arg_match(&arg, &height, argi))
1584 cfg.g_h = arg_parse_uint(&arg);
1585 else if (arg_match(&arg, &stereo_mode, argi))
1586 stereo_fmt = arg_parse_enum_or_int(&arg);
1587 else if (arg_match(&arg, &timebase, argi))
1588 cfg.g_timebase = arg_parse_rational(&arg);
1589 else if (arg_match(&arg, &error_resilient, argi))
1590 cfg.g_error_resilient = arg_parse_uint(&arg);
1591 else if (arg_match(&arg, &lag_in_frames, argi))
1592 cfg.g_lag_in_frames = arg_parse_uint(&arg);
1593 else if (arg_match(&arg, &dropframe_thresh, argi))
1594 cfg.rc_dropframe_thresh = arg_parse_uint(&arg);
1595 else if (arg_match(&arg, &resize_allowed, argi))
1596 cfg.rc_resize_allowed = arg_parse_uint(&arg);
1597 else if (arg_match(&arg, &resize_up_thresh, argi))
1598 cfg.rc_resize_up_thresh = arg_parse_uint(&arg);
1599 else if (arg_match(&arg, &resize_down_thresh, argi))
1600 cfg.rc_resize_down_thresh = arg_parse_uint(&arg);
1601 else if (arg_match(&arg, &resize_down_thresh, argi))
1602 cfg.rc_resize_down_thresh = arg_parse_uint(&arg);
1603 else if (arg_match(&arg, &end_usage, argi))
1604 cfg.rc_end_usage = arg_parse_enum_or_int(&arg);
1605 else if (arg_match(&arg, &target_bitrate, argi))
1606 cfg.rc_target_bitrate = arg_parse_uint(&arg);
1607 else if (arg_match(&arg, &max_intra_rate_pct, argi))
1608 cfg.rc_max_intra_bitrate_pct = arg_parse_uint(&arg);
1609 else if (arg_match(&arg, &min_quantizer, argi))
1610 cfg.rc_min_quantizer = arg_parse_uint(&arg);
1611 else if (arg_match(&arg, &max_quantizer, argi))
1612 cfg.rc_max_quantizer = arg_parse_uint(&arg);
1613 else if (arg_match(&arg, &undershoot_pct, argi))
1614 cfg.rc_undershoot_pct = arg_parse_uint(&arg);
1615 else if (arg_match(&arg, &overshoot_pct, argi))
1616 cfg.rc_overshoot_pct = arg_parse_uint(&arg);
1617 else if (arg_match(&arg, &buf_sz, argi))
1618 cfg.rc_buf_sz = arg_parse_uint(&arg);
1619 else if (arg_match(&arg, &buf_initial_sz, argi))
1620 cfg.rc_buf_initial_sz = arg_parse_uint(&arg);
1621 else if (arg_match(&arg, &buf_optimal_sz, argi))
1622 cfg.rc_buf_optimal_sz = arg_parse_uint(&arg);
1623 else if (arg_match(&arg, &bias_pct, argi))
1625 cfg.rc_2pass_vbr_bias_pct = arg_parse_uint(&arg);
1627 if (arg_passes < 2)
1628 fprintf(stderr,
1629 "Warning: option %s ignored in one-pass mode.\n",
1630 arg.name);
1632 else if (arg_match(&arg, &minsection_pct, argi))
1634 cfg.rc_2pass_vbr_minsection_pct = arg_parse_uint(&arg);
1636 if (arg_passes < 2)
1637 fprintf(stderr,
1638 "Warning: option %s ignored in one-pass mode.\n",
1639 arg.name);
1641 else if (arg_match(&arg, &maxsection_pct, argi))
1643 cfg.rc_2pass_vbr_maxsection_pct = arg_parse_uint(&arg);
1645 if (arg_passes < 2)
1646 fprintf(stderr,
1647 "Warning: option %s ignored in one-pass mode.\n",
1648 arg.name);
1650 else if (arg_match(&arg, &kf_min_dist, argi))
1651 cfg.kf_min_dist = arg_parse_uint(&arg);
1652 else if (arg_match(&arg, &kf_max_dist, argi))
1653 cfg.kf_max_dist = arg_parse_uint(&arg);
1654 else if (arg_match(&arg, &kf_disabled, argi))
1655 cfg.kf_mode = VPX_KF_DISABLED;
1656 else
1657 argj++;
1660 /* Handle codec specific options */
1661 #if CONFIG_VP8_ENCODER
1663 if (codec->iface == &vpx_codec_vp8_cx_algo)
1665 ctrl_args = vp8_args;
1666 ctrl_args_map = vp8_arg_ctrl_map;
1669 #endif
1671 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
1673 int match = 0;
1675 arg.argv_step = 1;
1677 for (i = 0; ctrl_args[i]; i++)
1679 if (arg_match(&arg, ctrl_args[i], argi))
1681 match = 1;
1683 if (arg_ctrl_cnt < ARG_CTRL_CNT_MAX)
1685 arg_ctrls[arg_ctrl_cnt][0] = ctrl_args_map[i];
1686 arg_ctrls[arg_ctrl_cnt][1] = arg_parse_enum_or_int(&arg);
1687 arg_ctrl_cnt++;
1692 if (!match)
1693 argj++;
1696 /* Check for unrecognized options */
1697 for (argi = argv; *argi; argi++)
1698 if (argi[0][0] == '-' && argi[0][1])
1699 die("Error: Unrecognized option %s\n", *argi);
1701 /* Handle non-option arguments */
1702 in_fn = argv[0];
1704 if (!in_fn)
1705 usage_exit();
1707 if(!out_fn)
1708 die("Error: Output file is required (specify with -o)\n");
1710 memset(&stats, 0, sizeof(stats));
1712 for (pass = one_pass_only ? one_pass_only - 1 : 0; pass < arg_passes; pass++)
1714 int frames_in = 0, frames_out = 0;
1715 unsigned long nbytes = 0;
1716 struct detect_buffer detect;
1718 /* Parse certain options from the input file, if possible */
1719 infile = strcmp(in_fn, "-") ? fopen(in_fn, "rb")
1720 : set_binary_mode(stdin);
1722 if (!infile)
1724 fprintf(stderr, "Failed to open input file\n");
1725 return EXIT_FAILURE;
1728 /* For RAW input sources, these bytes will applied on the first frame
1729 * in read_frame().
1731 detect.buf_read = fread(detect.buf, 1, 4, infile);
1732 detect.position = 0;
1734 if (detect.buf_read == 4 && file_is_y4m(infile, &y4m, detect.buf))
1736 if (y4m_input_open(&y4m, infile, detect.buf, 4) >= 0)
1738 file_type = FILE_TYPE_Y4M;
1739 cfg.g_w = y4m.pic_w;
1740 cfg.g_h = y4m.pic_h;
1742 /* Use the frame rate from the file only if none was specified
1743 * on the command-line.
1745 if (!arg_have_framerate)
1747 arg_framerate.num = y4m.fps_n;
1748 arg_framerate.den = y4m.fps_d;
1751 arg_use_i420 = 0;
1753 else
1755 fprintf(stderr, "Unsupported Y4M stream.\n");
1756 return EXIT_FAILURE;
1759 else if (detect.buf_read == 4 &&
1760 file_is_ivf(infile, &fourcc, &cfg.g_w, &cfg.g_h, &detect))
1762 file_type = FILE_TYPE_IVF;
1763 switch (fourcc)
1765 case 0x32315659:
1766 arg_use_i420 = 0;
1767 break;
1768 case 0x30323449:
1769 arg_use_i420 = 1;
1770 break;
1771 default:
1772 fprintf(stderr, "Unsupported fourcc (%08x) in IVF\n", fourcc);
1773 return EXIT_FAILURE;
1776 else
1778 file_type = FILE_TYPE_RAW;
1781 if(!cfg.g_w || !cfg.g_h)
1783 fprintf(stderr, "Specify stream dimensions with --width (-w) "
1784 " and --height (-h).\n");
1785 return EXIT_FAILURE;
1788 #define SHOW(field) fprintf(stderr, " %-28s = %d\n", #field, cfg.field)
1790 if (verbose && pass == 0)
1792 fprintf(stderr, "Codec: %s\n", vpx_codec_iface_name(codec->iface));
1793 fprintf(stderr, "Source file: %s Format: %s\n", in_fn,
1794 arg_use_i420 ? "I420" : "YV12");
1795 fprintf(stderr, "Destination file: %s\n", out_fn);
1796 fprintf(stderr, "Encoder parameters:\n");
1798 SHOW(g_usage);
1799 SHOW(g_threads);
1800 SHOW(g_profile);
1801 SHOW(g_w);
1802 SHOW(g_h);
1803 SHOW(g_timebase.num);
1804 SHOW(g_timebase.den);
1805 SHOW(g_error_resilient);
1806 SHOW(g_pass);
1807 SHOW(g_lag_in_frames);
1808 SHOW(rc_dropframe_thresh);
1809 SHOW(rc_resize_allowed);
1810 SHOW(rc_resize_up_thresh);
1811 SHOW(rc_resize_down_thresh);
1812 SHOW(rc_end_usage);
1813 SHOW(rc_target_bitrate);
1814 SHOW(rc_min_quantizer);
1815 SHOW(rc_max_quantizer);
1816 SHOW(rc_undershoot_pct);
1817 SHOW(rc_overshoot_pct);
1818 SHOW(rc_buf_sz);
1819 SHOW(rc_buf_initial_sz);
1820 SHOW(rc_buf_optimal_sz);
1821 SHOW(rc_2pass_vbr_bias_pct);
1822 SHOW(rc_2pass_vbr_minsection_pct);
1823 SHOW(rc_2pass_vbr_maxsection_pct);
1824 SHOW(kf_mode);
1825 SHOW(kf_min_dist);
1826 SHOW(kf_max_dist);
1829 if(pass == (one_pass_only ? one_pass_only - 1 : 0)) {
1830 if (file_type == FILE_TYPE_Y4M)
1831 /*The Y4M reader does its own allocation.
1832 Just initialize this here to avoid problems if we never read any
1833 frames.*/
1834 memset(&raw, 0, sizeof(raw));
1835 else
1836 vpx_img_alloc(&raw, arg_use_i420 ? VPX_IMG_FMT_I420 : VPX_IMG_FMT_YV12,
1837 cfg.g_w, cfg.g_h, 1);
1839 init_rate_histogram(&rate_hist, &cfg, &arg_framerate);
1842 outfile = strcmp(out_fn, "-") ? fopen(out_fn, "wb")
1843 : set_binary_mode(stdout);
1845 if (!outfile)
1847 fprintf(stderr, "Failed to open output file\n");
1848 return EXIT_FAILURE;
1851 if(write_webm && fseek(outfile, 0, SEEK_CUR))
1853 fprintf(stderr, "WebM output to pipes not supported.\n");
1854 return EXIT_FAILURE;
1857 if (stats_fn)
1859 if (!stats_open_file(&stats, stats_fn, pass))
1861 fprintf(stderr, "Failed to open statistics store\n");
1862 return EXIT_FAILURE;
1865 else
1867 if (!stats_open_mem(&stats, pass))
1869 fprintf(stderr, "Failed to open statistics store\n");
1870 return EXIT_FAILURE;
1874 cfg.g_pass = arg_passes == 2
1875 ? pass ? VPX_RC_LAST_PASS : VPX_RC_FIRST_PASS
1876 : VPX_RC_ONE_PASS;
1877 #if VPX_ENCODER_ABI_VERSION > (1 + VPX_CODEC_ABI_VERSION)
1879 if (pass)
1881 cfg.rc_twopass_stats_in = stats_get(&stats);
1884 #endif
1886 if(write_webm)
1888 ebml.stream = outfile;
1889 write_webm_file_header(&ebml, &cfg, &arg_framerate, stereo_fmt);
1891 else
1892 write_ivf_file_header(outfile, &cfg, codec->fourcc, 0);
1895 /* Construct Encoder Context */
1896 vpx_codec_enc_init(&encoder, codec->iface, &cfg,
1897 show_psnr ? VPX_CODEC_USE_PSNR : 0);
1898 ctx_exit_on_error(&encoder, "Failed to initialize encoder");
1900 /* Note that we bypass the vpx_codec_control wrapper macro because
1901 * we're being clever to store the control IDs in an array. Real
1902 * applications will want to make use of the enumerations directly
1904 for (i = 0; i < arg_ctrl_cnt; i++)
1906 if (vpx_codec_control_(&encoder, arg_ctrls[i][0], arg_ctrls[i][1]))
1907 fprintf(stderr, "Error: Tried to set control %d = %d\n",
1908 arg_ctrls[i][0], arg_ctrls[i][1]);
1910 ctx_exit_on_error(&encoder, "Failed to control codec");
1913 frame_avail = 1;
1914 got_data = 0;
1916 while (frame_avail || got_data)
1918 vpx_codec_iter_t iter = NULL;
1919 const vpx_codec_cx_pkt_t *pkt;
1920 struct vpx_usec_timer timer;
1921 int64_t frame_start, next_frame_start;
1923 if (!arg_limit || frames_in < arg_limit)
1925 frame_avail = read_frame(infile, &raw, file_type, &y4m,
1926 &detect);
1928 if (frame_avail)
1929 frames_in++;
1931 fprintf(stderr,
1932 "\rPass %d/%d frame %4d/%-4d %7ldB \033[K", pass + 1,
1933 arg_passes, frames_in, frames_out, nbytes);
1935 else
1936 frame_avail = 0;
1938 vpx_usec_timer_start(&timer);
1940 frame_start = (cfg.g_timebase.den * (int64_t)(frames_in - 1)
1941 * arg_framerate.den) / cfg.g_timebase.num / arg_framerate.num;
1942 next_frame_start = (cfg.g_timebase.den * (int64_t)(frames_in)
1943 * arg_framerate.den)
1944 / cfg.g_timebase.num / arg_framerate.num;
1945 vpx_codec_encode(&encoder, frame_avail ? &raw : NULL, frame_start,
1946 next_frame_start - frame_start,
1947 0, arg_deadline);
1948 vpx_usec_timer_mark(&timer);
1949 cx_time += vpx_usec_timer_elapsed(&timer);
1950 ctx_exit_on_error(&encoder, "Failed to encode frame");
1952 if(cfg.g_pass != VPX_RC_FIRST_PASS)
1954 int q;
1956 vpx_codec_control(&encoder, VP8E_GET_LAST_QUANTIZER_64, &q);
1957 ctx_exit_on_error(&encoder, "Failed to read quantizer");
1958 counts[q]++;
1961 got_data = 0;
1963 while ((pkt = vpx_codec_get_cx_data(&encoder, &iter)))
1965 got_data = 1;
1967 switch (pkt->kind)
1969 case VPX_CODEC_CX_FRAME_PKT:
1970 frames_out++;
1971 fprintf(stderr, " %6luF",
1972 (unsigned long)pkt->data.frame.sz);
1974 update_rate_histogram(&rate_hist, &cfg, pkt);
1975 if(write_webm)
1977 /* Update the hash */
1978 if(!ebml.debug)
1979 hash = murmur(pkt->data.frame.buf,
1980 pkt->data.frame.sz, hash);
1982 write_webm_block(&ebml, &cfg, pkt);
1984 else
1986 write_ivf_frame_header(outfile, pkt);
1987 if(fwrite(pkt->data.frame.buf, 1,
1988 pkt->data.frame.sz, outfile));
1990 nbytes += pkt->data.raw.sz;
1991 break;
1992 case VPX_CODEC_STATS_PKT:
1993 frames_out++;
1994 fprintf(stderr, " %6luS",
1995 (unsigned long)pkt->data.twopass_stats.sz);
1996 stats_write(&stats,
1997 pkt->data.twopass_stats.buf,
1998 pkt->data.twopass_stats.sz);
1999 nbytes += pkt->data.raw.sz;
2000 break;
2001 case VPX_CODEC_PSNR_PKT:
2003 if (show_psnr)
2005 int i;
2007 psnr_sse_total += pkt->data.psnr.sse[0];
2008 psnr_samples_total += pkt->data.psnr.samples[0];
2009 for (i = 0; i < 4; i++)
2011 fprintf(stderr, "%.3lf ", pkt->data.psnr.psnr[i]);
2012 psnr_totals[i] += pkt->data.psnr.psnr[i];
2014 psnr_count++;
2017 break;
2018 default:
2019 break;
2023 fflush(stdout);
2026 fprintf(stderr,
2027 "\rPass %d/%d frame %4d/%-4d %7ldB %7ldb/f %7"PRId64"b/s"
2028 " %7lu %s (%.2f fps)\033[K", pass + 1,
2029 arg_passes, frames_in, frames_out, nbytes, nbytes * 8 / frames_in,
2030 nbytes * 8 *(int64_t)arg_framerate.num / arg_framerate.den / frames_in,
2031 cx_time > 9999999 ? cx_time / 1000 : cx_time,
2032 cx_time > 9999999 ? "ms" : "us",
2033 (float)frames_in * 1000000.0 / (float)cx_time);
2035 if ( (show_psnr) && (psnr_count>0) )
2037 int i;
2038 double ovpsnr = vp8_mse2psnr(psnr_samples_total, 255.0,
2039 psnr_sse_total);
2041 fprintf(stderr, "\nPSNR (Overall/Avg/Y/U/V)");
2043 fprintf(stderr, " %.3lf", ovpsnr);
2044 for (i = 0; i < 4; i++)
2046 fprintf(stderr, " %.3lf", psnr_totals[i]/psnr_count);
2050 vpx_codec_destroy(&encoder);
2052 fclose(infile);
2053 if (file_type == FILE_TYPE_Y4M)
2054 y4m_input_close(&y4m);
2056 if(write_webm)
2058 write_webm_file_footer(&ebml, hash);
2059 free(ebml.cue_list);
2060 ebml.cue_list = NULL;
2062 else
2064 if (!fseek(outfile, 0, SEEK_SET))
2065 write_ivf_file_header(outfile, &cfg, codec->fourcc, frames_out);
2068 fclose(outfile);
2069 stats_close(&stats, arg_passes-1);
2070 fprintf(stderr, "\n");
2072 if (one_pass_only)
2073 break;
2076 if (show_q_hist_buckets)
2077 show_q_histogram(counts, show_q_hist_buckets);
2079 if (show_rate_hist_buckets)
2080 show_rate_histogram(&rate_hist, &cfg, show_rate_hist_buckets);
2081 destroy_rate_histogram(&rate_hist);
2083 vpx_img_free(&raw);
2084 free(argv);
2085 return EXIT_SUCCESS;