Merge "ARMv6 optimized fdct4x4"
[libvpx.git] / vpxenc.c
blob6c13cd1bdaf7f6fae6facb2823a3537399f004d1
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 "vpx/vpx_encoder.h"
27 #if USE_POSIX_MMAP
28 #include <sys/types.h>
29 #include <sys/stat.h>
30 #include <sys/mman.h>
31 #include <fcntl.h>
32 #include <unistd.h>
33 #endif
34 #include "vpx_version.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"
39 #include "y4minput.h"
40 #include "libmkv/EbmlWriter.h"
41 #include "libmkv/EbmlIDs.h"
43 /* Need special handling of these functions on Windows */
44 #if defined(_MSC_VER)
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
49 #elif defined(_WIN32)
50 /* MinGW defines off_t, and uses f{seek,tell}o64 */
51 #define fseeko fseeko64
52 #define ftello ftello64
53 #endif
55 #if defined(_MSC_VER)
56 #define LITERALU64(n) n
57 #else
58 #define LITERALU64(n) n##LLU
59 #endif
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
64 typedef long off_t;
65 #define fseeko fseek
66 #define ftello ftell
67 #endif
69 static const char *exec_name;
71 static const struct codec_item
73 char const *name;
74 const vpx_codec_iface_t *iface;
75 unsigned int fourcc;
76 } codecs[] =
78 #if CONFIG_VP8_ENCODER
79 {"vp8", &vpx_codec_vp8_cx_algo, 0x30385056},
80 #endif
83 static void usage_exit();
85 void die(const char *fmt, ...)
87 va_list ap;
88 va_start(ap, fmt);
89 vfprintf(stderr, fmt, ap);
90 fprintf(stderr, "\n");
91 usage_exit();
94 static void ctx_exit_on_error(vpx_codec_ctx_t *ctx, const char *s)
96 if (ctx->err)
98 const char *detail = vpx_codec_error_detail(ctx);
100 fprintf(stderr, "%s: %s\n", s, vpx_codec_error(ctx));
102 if (detail)
103 fprintf(stderr, " %s\n", detail);
105 exit(EXIT_FAILURE);
109 /* This structure is used to abstract the different ways of handling
110 * first pass statistics.
112 typedef struct
114 vpx_fixed_buf_t buf;
115 int pass;
116 FILE *file;
117 char *buf_ptr;
118 size_t buf_alloc_sz;
119 } stats_io_t;
121 int stats_open_file(stats_io_t *stats, const char *fpf, int pass)
123 int res;
125 stats->pass = pass;
127 if (pass == 0)
129 stats->file = fopen(fpf, "wb");
130 stats->buf.sz = 0;
131 stats->buf.buf = NULL,
132 res = (stats->file != NULL);
134 else
136 #if 0
137 #elif USE_POSIX_MMAP
138 struct stat stat_buf;
139 int fd;
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,
146 fd, 0);
147 res = (stats->buf.buf != NULL);
148 #else
149 size_t nbytes;
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");
156 exit(EXIT_FAILURE);
159 stats->buf.sz = stats->buf_alloc_sz = ftell(stats->file);
160 rewind(stats->file);
162 stats->buf.buf = malloc(stats->buf_alloc_sz);
164 if (!stats->buf.buf)
166 fprintf(stderr, "Failed to allocate first-pass stats buffer (%lu bytes)\n",
167 (unsigned long)stats->buf_alloc_sz);
168 exit(EXIT_FAILURE);
171 nbytes = fread(stats->buf.buf, 1, stats->buf.sz, stats->file);
172 res = (nbytes == stats->buf.sz);
173 #endif
176 return res;
179 int stats_open_mem(stats_io_t *stats, int pass)
181 int res;
182 stats->pass = pass;
184 if (!pass)
186 stats->buf.sz = 0;
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);
193 return res;
197 void stats_close(stats_io_t *stats, int last_pass)
199 if (stats->file)
201 if (stats->pass == last_pass)
203 #if 0
204 #elif USE_POSIX_MMAP
205 munmap(stats->buf.buf, stats->buf.sz);
206 #else
207 free(stats->buf.buf);
208 #endif
211 fclose(stats->file);
212 stats->file = NULL;
214 else
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)
223 if (stats->file)
225 if(fwrite(pkt, 1, len, stats->file));
227 else
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);
234 if (new_ptr)
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;
239 } /* else ... */
242 memcpy(stats->buf_ptr, pkt, len);
243 stats->buf.sz += len;
244 stats->buf_ptr += len;
248 vpx_fixed_buf_t stats_get(stats_io_t *stats)
250 return stats->buf;
253 enum video_file_type
255 FILE_TYPE_RAW,
256 FILE_TYPE_IVF,
257 FILE_TYPE_Y4M
260 struct detect_buffer {
261 char buf[4];
262 size_t buf_read;
263 size_t position;
267 #define IVF_FRAME_HDR_SZ (4+8) /* 4 byte size + 8 byte timestamp */
268 static int read_frame(FILE *f, vpx_image_t *img, unsigned int file_type,
269 y4m_input *y4m, struct detect_buffer *detect)
271 int plane = 0;
272 int shortread = 0;
274 if (file_type == FILE_TYPE_Y4M)
276 if (y4m_input_fetch_frame(y4m, f, img) < 1)
277 return 0;
279 else
281 if (file_type == FILE_TYPE_IVF)
283 char junk[IVF_FRAME_HDR_SZ];
285 /* Skip the frame header. We know how big the frame should be. See
286 * write_ivf_frame_header() for documentation on the frame header
287 * layout.
289 if(fread(junk, 1, IVF_FRAME_HDR_SZ, f));
292 for (plane = 0; plane < 3; plane++)
294 unsigned char *ptr;
295 int w = (plane ? (1 + img->d_w) / 2 : img->d_w);
296 int h = (plane ? (1 + img->d_h) / 2 : img->d_h);
297 int r;
299 /* Determine the correct plane based on the image format. The for-loop
300 * always counts in Y,U,V order, but this may not match the order of
301 * the data on disk.
303 switch (plane)
305 case 1:
306 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U];
307 break;
308 case 2:
309 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V];
310 break;
311 default:
312 ptr = img->planes[plane];
315 for (r = 0; r < h; r++)
317 size_t needed = w;
318 size_t buf_position = 0;
319 const size_t left = detect->buf_read - detect->position;
320 if (left > 0)
322 const size_t more = (left < needed) ? left : needed;
323 memcpy(ptr, detect->buf + detect->position, more);
324 buf_position = more;
325 needed -= more;
326 detect->position += more;
328 if (needed > 0)
330 shortread |= (fread(ptr + buf_position, 1, needed, f) < needed);
333 ptr += img->stride[plane];
338 return !shortread;
342 unsigned int file_is_y4m(FILE *infile,
343 y4m_input *y4m,
344 char detect[4])
346 if(memcmp(detect, "YUV4", 4) == 0)
348 return 1;
350 return 0;
353 #define IVF_FILE_HDR_SZ (32)
354 unsigned int file_is_ivf(FILE *infile,
355 unsigned int *fourcc,
356 unsigned int *width,
357 unsigned int *height,
358 struct detect_buffer *detect)
360 char raw_hdr[IVF_FILE_HDR_SZ];
361 int is_ivf = 0;
363 if(memcmp(detect->buf, "DKIF", 4) != 0)
364 return 0;
366 /* See write_ivf_file_header() for more documentation on the file header
367 * layout.
369 if (fread(raw_hdr + 4, 1, IVF_FILE_HDR_SZ - 4, infile)
370 == IVF_FILE_HDR_SZ - 4)
373 is_ivf = 1;
375 if (mem_get_le16(raw_hdr + 4) != 0)
376 fprintf(stderr, "Error: Unrecognized IVF version! This file may not"
377 " decode properly.");
379 *fourcc = mem_get_le32(raw_hdr + 8);
383 if (is_ivf)
385 *width = mem_get_le16(raw_hdr + 12);
386 *height = mem_get_le16(raw_hdr + 14);
387 detect->position = 4;
390 return is_ivf;
394 static void write_ivf_file_header(FILE *outfile,
395 const vpx_codec_enc_cfg_t *cfg,
396 unsigned int fourcc,
397 int frame_cnt)
399 char header[32];
401 if (cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS)
402 return;
404 header[0] = 'D';
405 header[1] = 'K';
406 header[2] = 'I';
407 header[3] = 'F';
408 mem_put_le16(header + 4, 0); /* version */
409 mem_put_le16(header + 6, 32); /* headersize */
410 mem_put_le32(header + 8, fourcc); /* headersize */
411 mem_put_le16(header + 12, cfg->g_w); /* width */
412 mem_put_le16(header + 14, cfg->g_h); /* height */
413 mem_put_le32(header + 16, cfg->g_timebase.den); /* rate */
414 mem_put_le32(header + 20, cfg->g_timebase.num); /* scale */
415 mem_put_le32(header + 24, frame_cnt); /* length */
416 mem_put_le32(header + 28, 0); /* unused */
418 if(fwrite(header, 1, 32, outfile));
422 static void write_ivf_frame_header(FILE *outfile,
423 const vpx_codec_cx_pkt_t *pkt)
425 char header[12];
426 vpx_codec_pts_t pts;
428 if (pkt->kind != VPX_CODEC_CX_FRAME_PKT)
429 return;
431 pts = pkt->data.frame.pts;
432 mem_put_le32(header, pkt->data.frame.sz);
433 mem_put_le32(header + 4, pts & 0xFFFFFFFF);
434 mem_put_le32(header + 8, pts >> 32);
436 if(fwrite(header, 1, 12, outfile));
440 typedef off_t EbmlLoc;
443 struct cue_entry
445 unsigned int time;
446 uint64_t loc;
450 struct EbmlGlobal
452 int debug;
454 FILE *stream;
455 int64_t last_pts_ms;
456 vpx_rational_t framerate;
458 /* These pointers are to the start of an element */
459 off_t position_reference;
460 off_t seek_info_pos;
461 off_t segment_info_pos;
462 off_t track_pos;
463 off_t cue_pos;
464 off_t cluster_pos;
466 /* This pointer is to a specific element to be serialized */
467 off_t track_id_pos;
469 /* These pointers are to the size field of the element */
470 EbmlLoc startSegment;
471 EbmlLoc startCluster;
473 uint32_t cluster_timecode;
474 int cluster_open;
476 struct cue_entry *cue_list;
477 unsigned int cues;
482 void Ebml_Write(EbmlGlobal *glob, const void *buffer_in, unsigned long len)
484 if(fwrite(buffer_in, 1, len, glob->stream));
488 void Ebml_Serialize(EbmlGlobal *glob, const void *buffer_in, unsigned long len)
490 const unsigned char *q = (const unsigned char *)buffer_in + len - 1;
492 for(; len; len--)
493 Ebml_Write(glob, q--, 1);
497 /* Need a fixed size serializer for the track ID. libmkv provdes a 64 bit
498 * one, but not a 32 bit one.
500 static void Ebml_SerializeUnsigned32(EbmlGlobal *glob, unsigned long class_id, uint64_t ui)
502 unsigned char sizeSerialized = 4 | 0x80;
503 Ebml_WriteID(glob, class_id);
504 Ebml_Serialize(glob, &sizeSerialized, 1);
505 Ebml_Serialize(glob, &ui, 4);
509 static void
510 Ebml_StartSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc,
511 unsigned long class_id)
513 //todo this is always taking 8 bytes, this may need later optimization
514 //this is a key that says lenght unknown
515 unsigned long long unknownLen = LITERALU64(0x01FFFFFFFFFFFFFF);
517 Ebml_WriteID(glob, class_id);
518 *ebmlLoc = ftello(glob->stream);
519 Ebml_Serialize(glob, &unknownLen, 8);
522 static void
523 Ebml_EndSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc)
525 off_t pos;
526 uint64_t size;
528 /* Save the current stream pointer */
529 pos = ftello(glob->stream);
531 /* Calculate the size of this element */
532 size = pos - *ebmlLoc - 8;
533 size |= LITERALU64(0x0100000000000000);
535 /* Seek back to the beginning of the element and write the new size */
536 fseeko(glob->stream, *ebmlLoc, SEEK_SET);
537 Ebml_Serialize(glob, &size, 8);
539 /* Reset the stream pointer */
540 fseeko(glob->stream, pos, SEEK_SET);
544 static void
545 write_webm_seek_element(EbmlGlobal *ebml, unsigned long id, off_t pos)
547 uint64_t offset = pos - ebml->position_reference;
548 EbmlLoc start;
549 Ebml_StartSubElement(ebml, &start, Seek);
550 Ebml_SerializeBinary(ebml, SeekID, id);
551 Ebml_SerializeUnsigned64(ebml, SeekPosition, offset);
552 Ebml_EndSubElement(ebml, &start);
556 static void
557 write_webm_seek_info(EbmlGlobal *ebml)
560 off_t pos;
562 /* Save the current stream pointer */
563 pos = ftello(ebml->stream);
565 if(ebml->seek_info_pos)
566 fseeko(ebml->stream, ebml->seek_info_pos, SEEK_SET);
567 else
568 ebml->seek_info_pos = pos;
571 EbmlLoc start;
573 Ebml_StartSubElement(ebml, &start, SeekHead);
574 write_webm_seek_element(ebml, Tracks, ebml->track_pos);
575 write_webm_seek_element(ebml, Cues, ebml->cue_pos);
576 write_webm_seek_element(ebml, Info, ebml->segment_info_pos);
577 Ebml_EndSubElement(ebml, &start);
580 //segment info
581 EbmlLoc startInfo;
582 uint64_t frame_time;
584 frame_time = (uint64_t)1000 * ebml->framerate.den
585 / ebml->framerate.num;
586 ebml->segment_info_pos = ftello(ebml->stream);
587 Ebml_StartSubElement(ebml, &startInfo, Info);
588 Ebml_SerializeUnsigned(ebml, TimecodeScale, 1000000);
589 Ebml_SerializeFloat(ebml, Segment_Duration,
590 ebml->last_pts_ms + frame_time);
591 Ebml_SerializeString(ebml, 0x4D80,
592 ebml->debug ? "vpxenc" : "vpxenc" VERSION_STRING);
593 Ebml_SerializeString(ebml, 0x5741,
594 ebml->debug ? "vpxenc" : "vpxenc" VERSION_STRING);
595 Ebml_EndSubElement(ebml, &startInfo);
600 static void
601 write_webm_file_header(EbmlGlobal *glob,
602 const vpx_codec_enc_cfg_t *cfg,
603 const struct vpx_rational *fps)
606 EbmlLoc start;
607 Ebml_StartSubElement(glob, &start, EBML);
608 Ebml_SerializeUnsigned(glob, EBMLVersion, 1);
609 Ebml_SerializeUnsigned(glob, EBMLReadVersion, 1); //EBML Read Version
610 Ebml_SerializeUnsigned(glob, EBMLMaxIDLength, 4); //EBML Max ID Length
611 Ebml_SerializeUnsigned(glob, EBMLMaxSizeLength, 8); //EBML Max Size Length
612 Ebml_SerializeString(glob, DocType, "webm"); //Doc Type
613 Ebml_SerializeUnsigned(glob, DocTypeVersion, 2); //Doc Type Version
614 Ebml_SerializeUnsigned(glob, DocTypeReadVersion, 2); //Doc Type Read Version
615 Ebml_EndSubElement(glob, &start);
618 Ebml_StartSubElement(glob, &glob->startSegment, Segment); //segment
619 glob->position_reference = ftello(glob->stream);
620 glob->framerate = *fps;
621 write_webm_seek_info(glob);
624 EbmlLoc trackStart;
625 glob->track_pos = ftello(glob->stream);
626 Ebml_StartSubElement(glob, &trackStart, Tracks);
628 unsigned int trackNumber = 1;
629 uint64_t trackID = 0;
631 EbmlLoc start;
632 Ebml_StartSubElement(glob, &start, TrackEntry);
633 Ebml_SerializeUnsigned(glob, TrackNumber, trackNumber);
634 glob->track_id_pos = ftello(glob->stream);
635 Ebml_SerializeUnsigned32(glob, TrackUID, trackID);
636 Ebml_SerializeUnsigned(glob, TrackType, 1); //video is always 1
637 Ebml_SerializeString(glob, CodecID, "V_VP8");
639 unsigned int pixelWidth = cfg->g_w;
640 unsigned int pixelHeight = cfg->g_h;
641 float frameRate = (float)fps->num/(float)fps->den;
643 EbmlLoc videoStart;
644 Ebml_StartSubElement(glob, &videoStart, Video);
645 Ebml_SerializeUnsigned(glob, PixelWidth, pixelWidth);
646 Ebml_SerializeUnsigned(glob, PixelHeight, pixelHeight);
647 Ebml_SerializeFloat(glob, FrameRate, frameRate);
648 Ebml_EndSubElement(glob, &videoStart); //Video
650 Ebml_EndSubElement(glob, &start); //Track Entry
652 Ebml_EndSubElement(glob, &trackStart);
654 // segment element is open
659 static void
660 write_webm_block(EbmlGlobal *glob,
661 const vpx_codec_enc_cfg_t *cfg,
662 const vpx_codec_cx_pkt_t *pkt)
664 unsigned long block_length;
665 unsigned char track_number;
666 unsigned short block_timecode = 0;
667 unsigned char flags;
668 int64_t pts_ms;
669 int start_cluster = 0, is_keyframe;
671 /* Calculate the PTS of this frame in milliseconds */
672 pts_ms = pkt->data.frame.pts * 1000
673 * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
674 if(pts_ms <= glob->last_pts_ms)
675 pts_ms = glob->last_pts_ms + 1;
676 glob->last_pts_ms = pts_ms;
678 /* Calculate the relative time of this block */
679 if(pts_ms - glob->cluster_timecode > SHRT_MAX)
680 start_cluster = 1;
681 else
682 block_timecode = pts_ms - glob->cluster_timecode;
684 is_keyframe = (pkt->data.frame.flags & VPX_FRAME_IS_KEY);
685 if(start_cluster || is_keyframe)
687 if(glob->cluster_open)
688 Ebml_EndSubElement(glob, &glob->startCluster);
690 /* Open the new cluster */
691 block_timecode = 0;
692 glob->cluster_open = 1;
693 glob->cluster_timecode = pts_ms;
694 glob->cluster_pos = ftello(glob->stream);
695 Ebml_StartSubElement(glob, &glob->startCluster, Cluster); //cluster
696 Ebml_SerializeUnsigned(glob, Timecode, glob->cluster_timecode);
698 /* Save a cue point if this is a keyframe. */
699 if(is_keyframe)
701 struct cue_entry *cue;
703 glob->cue_list = realloc(glob->cue_list,
704 (glob->cues+1) * sizeof(struct cue_entry));
705 cue = &glob->cue_list[glob->cues];
706 cue->time = glob->cluster_timecode;
707 cue->loc = glob->cluster_pos;
708 glob->cues++;
712 /* Write the Simple Block */
713 Ebml_WriteID(glob, SimpleBlock);
715 block_length = pkt->data.frame.sz + 4;
716 block_length |= 0x10000000;
717 Ebml_Serialize(glob, &block_length, 4);
719 track_number = 1;
720 track_number |= 0x80;
721 Ebml_Write(glob, &track_number, 1);
723 Ebml_Serialize(glob, &block_timecode, 2);
725 flags = 0;
726 if(is_keyframe)
727 flags |= 0x80;
728 if(pkt->data.frame.flags & VPX_FRAME_IS_INVISIBLE)
729 flags |= 0x08;
730 Ebml_Write(glob, &flags, 1);
732 Ebml_Write(glob, pkt->data.frame.buf, pkt->data.frame.sz);
736 static void
737 write_webm_file_footer(EbmlGlobal *glob, long hash)
740 if(glob->cluster_open)
741 Ebml_EndSubElement(glob, &glob->startCluster);
744 EbmlLoc start;
745 int i;
747 glob->cue_pos = ftello(glob->stream);
748 Ebml_StartSubElement(glob, &start, Cues);
749 for(i=0; i<glob->cues; i++)
751 struct cue_entry *cue = &glob->cue_list[i];
752 EbmlLoc start;
754 Ebml_StartSubElement(glob, &start, CuePoint);
756 EbmlLoc start;
758 Ebml_SerializeUnsigned(glob, CueTime, cue->time);
760 Ebml_StartSubElement(glob, &start, CueTrackPositions);
761 Ebml_SerializeUnsigned(glob, CueTrack, 1);
762 Ebml_SerializeUnsigned64(glob, CueClusterPosition,
763 cue->loc - glob->position_reference);
764 //Ebml_SerializeUnsigned(glob, CueBlockNumber, cue->blockNumber);
765 Ebml_EndSubElement(glob, &start);
767 Ebml_EndSubElement(glob, &start);
769 Ebml_EndSubElement(glob, &start);
772 Ebml_EndSubElement(glob, &glob->startSegment);
774 /* Patch up the seek info block */
775 write_webm_seek_info(glob);
777 /* Patch up the track id */
778 fseeko(glob->stream, glob->track_id_pos, SEEK_SET);
779 Ebml_SerializeUnsigned32(glob, TrackUID, glob->debug ? 0xDEADBEEF : hash);
781 fseeko(glob->stream, 0, SEEK_END);
785 /* Murmur hash derived from public domain reference implementation at
786 * http://sites.google.com/site/murmurhash/
788 static unsigned int murmur ( const void * key, int len, unsigned int seed )
790 const unsigned int m = 0x5bd1e995;
791 const int r = 24;
793 unsigned int h = seed ^ len;
795 const unsigned char * data = (const unsigned char *)key;
797 while(len >= 4)
799 unsigned int k;
801 k = data[0];
802 k |= data[1] << 8;
803 k |= data[2] << 16;
804 k |= data[3] << 24;
806 k *= m;
807 k ^= k >> r;
808 k *= m;
810 h *= m;
811 h ^= k;
813 data += 4;
814 len -= 4;
817 switch(len)
819 case 3: h ^= data[2] << 16;
820 case 2: h ^= data[1] << 8;
821 case 1: h ^= data[0];
822 h *= m;
825 h ^= h >> 13;
826 h *= m;
827 h ^= h >> 15;
829 return h;
832 #include "math.h"
834 static double vp8_mse2psnr(double Samples, double Peak, double Mse)
836 double psnr;
838 if ((double)Mse > 0.0)
839 psnr = 10.0 * log10(Peak * Peak * Samples / Mse);
840 else
841 psnr = 60; // Limit to prevent / 0
843 if (psnr > 60)
844 psnr = 60;
846 return psnr;
850 #include "args.h"
852 static const arg_def_t debugmode = ARG_DEF("D", "debug", 0,
853 "Debug mode (makes output deterministic)");
854 static const arg_def_t outputfile = ARG_DEF("o", "output", 1,
855 "Output filename");
856 static const arg_def_t use_yv12 = ARG_DEF(NULL, "yv12", 0,
857 "Input file is YV12 ");
858 static const arg_def_t use_i420 = ARG_DEF(NULL, "i420", 0,
859 "Input file is I420 (default)");
860 static const arg_def_t codecarg = ARG_DEF(NULL, "codec", 1,
861 "Codec to use");
862 static const arg_def_t passes = ARG_DEF("p", "passes", 1,
863 "Number of passes (1/2)");
864 static const arg_def_t pass_arg = ARG_DEF(NULL, "pass", 1,
865 "Pass to execute (1/2)");
866 static const arg_def_t fpf_name = ARG_DEF(NULL, "fpf", 1,
867 "First pass statistics file name");
868 static const arg_def_t limit = ARG_DEF(NULL, "limit", 1,
869 "Stop encoding after n input frames");
870 static const arg_def_t deadline = ARG_DEF("d", "deadline", 1,
871 "Deadline per frame (usec)");
872 static const arg_def_t best_dl = ARG_DEF(NULL, "best", 0,
873 "Use Best Quality Deadline");
874 static const arg_def_t good_dl = ARG_DEF(NULL, "good", 0,
875 "Use Good Quality Deadline");
876 static const arg_def_t rt_dl = ARG_DEF(NULL, "rt", 0,
877 "Use Realtime Quality Deadline");
878 static const arg_def_t verbosearg = ARG_DEF("v", "verbose", 0,
879 "Show encoder parameters");
880 static const arg_def_t psnrarg = ARG_DEF(NULL, "psnr", 0,
881 "Show PSNR in status line");
882 static const arg_def_t framerate = ARG_DEF(NULL, "fps", 1,
883 "Stream frame rate (rate/scale)");
884 static const arg_def_t use_ivf = ARG_DEF(NULL, "ivf", 0,
885 "Output IVF (default is WebM)");
886 static const arg_def_t *main_args[] =
888 &debugmode,
889 &outputfile, &codecarg, &passes, &pass_arg, &fpf_name, &limit, &deadline,
890 &best_dl, &good_dl, &rt_dl,
891 &verbosearg, &psnrarg, &use_ivf, &framerate,
892 NULL
895 static const arg_def_t usage = ARG_DEF("u", "usage", 1,
896 "Usage profile number to use");
897 static const arg_def_t threads = ARG_DEF("t", "threads", 1,
898 "Max number of threads to use");
899 static const arg_def_t profile = ARG_DEF(NULL, "profile", 1,
900 "Bitstream profile number to use");
901 static const arg_def_t width = ARG_DEF("w", "width", 1,
902 "Frame width");
903 static const arg_def_t height = ARG_DEF("h", "height", 1,
904 "Frame height");
905 static const arg_def_t timebase = ARG_DEF(NULL, "timebase", 1,
906 "Stream timebase (frame duration)");
907 static const arg_def_t error_resilient = ARG_DEF(NULL, "error-resilient", 1,
908 "Enable error resiliency features");
909 static const arg_def_t lag_in_frames = ARG_DEF(NULL, "lag-in-frames", 1,
910 "Max number of frames to lag");
912 static const arg_def_t *global_args[] =
914 &use_yv12, &use_i420, &usage, &threads, &profile,
915 &width, &height, &timebase, &framerate, &error_resilient,
916 &lag_in_frames, NULL
919 static const arg_def_t dropframe_thresh = ARG_DEF(NULL, "drop-frame", 1,
920 "Temporal resampling threshold (buf %)");
921 static const arg_def_t resize_allowed = ARG_DEF(NULL, "resize-allowed", 1,
922 "Spatial resampling enabled (bool)");
923 static const arg_def_t resize_up_thresh = ARG_DEF(NULL, "resize-up", 1,
924 "Upscale threshold (buf %)");
925 static const arg_def_t resize_down_thresh = ARG_DEF(NULL, "resize-down", 1,
926 "Downscale threshold (buf %)");
927 static const struct arg_enum_list end_usage_enum[] = {
928 {"vbr", VPX_VBR},
929 {"cbr", VPX_CBR},
930 {"cq", VPX_CQ},
931 {NULL, 0}
933 static const arg_def_t end_usage = ARG_DEF_ENUM(NULL, "end-usage", 1,
934 "Rate control mode", end_usage_enum);
935 static const arg_def_t target_bitrate = ARG_DEF(NULL, "target-bitrate", 1,
936 "Bitrate (kbps)");
937 static const arg_def_t min_quantizer = ARG_DEF(NULL, "min-q", 1,
938 "Minimum (best) quantizer");
939 static const arg_def_t max_quantizer = ARG_DEF(NULL, "max-q", 1,
940 "Maximum (worst) quantizer");
941 static const arg_def_t undershoot_pct = ARG_DEF(NULL, "undershoot-pct", 1,
942 "Datarate undershoot (min) target (%)");
943 static const arg_def_t overshoot_pct = ARG_DEF(NULL, "overshoot-pct", 1,
944 "Datarate overshoot (max) target (%)");
945 static const arg_def_t buf_sz = ARG_DEF(NULL, "buf-sz", 1,
946 "Client buffer size (ms)");
947 static const arg_def_t buf_initial_sz = ARG_DEF(NULL, "buf-initial-sz", 1,
948 "Client initial buffer size (ms)");
949 static const arg_def_t buf_optimal_sz = ARG_DEF(NULL, "buf-optimal-sz", 1,
950 "Client optimal buffer size (ms)");
951 static const arg_def_t *rc_args[] =
953 &dropframe_thresh, &resize_allowed, &resize_up_thresh, &resize_down_thresh,
954 &end_usage, &target_bitrate, &min_quantizer, &max_quantizer,
955 &undershoot_pct, &overshoot_pct, &buf_sz, &buf_initial_sz, &buf_optimal_sz,
956 NULL
960 static const arg_def_t bias_pct = ARG_DEF(NULL, "bias-pct", 1,
961 "CBR/VBR bias (0=CBR, 100=VBR)");
962 static const arg_def_t minsection_pct = ARG_DEF(NULL, "minsection-pct", 1,
963 "GOP min bitrate (% of target)");
964 static const arg_def_t maxsection_pct = ARG_DEF(NULL, "maxsection-pct", 1,
965 "GOP max bitrate (% of target)");
966 static const arg_def_t *rc_twopass_args[] =
968 &bias_pct, &minsection_pct, &maxsection_pct, NULL
972 static const arg_def_t kf_min_dist = ARG_DEF(NULL, "kf-min-dist", 1,
973 "Minimum keyframe interval (frames)");
974 static const arg_def_t kf_max_dist = ARG_DEF(NULL, "kf-max-dist", 1,
975 "Maximum keyframe interval (frames)");
976 static const arg_def_t kf_disabled = ARG_DEF(NULL, "disable-kf", 0,
977 "Disable keyframe placement");
978 static const arg_def_t *kf_args[] =
980 &kf_min_dist, &kf_max_dist, &kf_disabled, NULL
984 #if CONFIG_VP8_ENCODER
985 static const arg_def_t noise_sens = ARG_DEF(NULL, "noise-sensitivity", 1,
986 "Noise sensitivity (frames to blur)");
987 static const arg_def_t sharpness = ARG_DEF(NULL, "sharpness", 1,
988 "Filter sharpness (0-7)");
989 static const arg_def_t static_thresh = ARG_DEF(NULL, "static-thresh", 1,
990 "Motion detection threshold");
991 #endif
993 #if CONFIG_VP8_ENCODER
994 static const arg_def_t cpu_used = ARG_DEF(NULL, "cpu-used", 1,
995 "CPU Used (-16..16)");
996 #endif
999 #if CONFIG_VP8_ENCODER
1000 static const arg_def_t token_parts = ARG_DEF(NULL, "token-parts", 1,
1001 "Number of token partitions to use, log2");
1002 static const arg_def_t auto_altref = ARG_DEF(NULL, "auto-alt-ref", 1,
1003 "Enable automatic alt reference frames");
1004 static const arg_def_t arnr_maxframes = ARG_DEF(NULL, "arnr-maxframes", 1,
1005 "AltRef Max Frames");
1006 static const arg_def_t arnr_strength = ARG_DEF(NULL, "arnr-strength", 1,
1007 "AltRef Strength");
1008 static const arg_def_t arnr_type = ARG_DEF(NULL, "arnr-type", 1,
1009 "AltRef Type");
1010 static const struct arg_enum_list tuning_enum[] = {
1011 {"psnr", VP8_TUNE_PSNR},
1012 {"ssim", VP8_TUNE_SSIM},
1013 {NULL, 0}
1015 static const arg_def_t tune_ssim = ARG_DEF_ENUM(NULL, "tune", 1,
1016 "Material to favor", tuning_enum);
1017 static const arg_def_t cq_level = ARG_DEF(NULL, "cq-level", 1,
1018 "Constrained Quality Level");
1020 static const arg_def_t *vp8_args[] =
1022 &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh,
1023 &token_parts, &arnr_maxframes, &arnr_strength, &arnr_type,
1024 &tune_ssim, &cq_level, NULL
1026 static const int vp8_arg_ctrl_map[] =
1028 VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF,
1029 VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD,
1030 VP8E_SET_TOKEN_PARTITIONS,
1031 VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH , VP8E_SET_ARNR_TYPE,
1032 VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, 0
1034 #endif
1036 static const arg_def_t *no_args[] = { NULL };
1038 static void usage_exit()
1040 int i;
1042 fprintf(stderr, "Usage: %s <options> -o dst_filename src_filename \n",
1043 exec_name);
1045 fprintf(stderr, "\nOptions:\n");
1046 arg_show_usage(stdout, main_args);
1047 fprintf(stderr, "\nEncoder Global Options:\n");
1048 arg_show_usage(stdout, global_args);
1049 fprintf(stderr, "\nRate Control Options:\n");
1050 arg_show_usage(stdout, rc_args);
1051 fprintf(stderr, "\nTwopass Rate Control Options:\n");
1052 arg_show_usage(stdout, rc_twopass_args);
1053 fprintf(stderr, "\nKeyframe Placement Options:\n");
1054 arg_show_usage(stdout, kf_args);
1055 #if CONFIG_VP8_ENCODER
1056 fprintf(stderr, "\nVP8 Specific Options:\n");
1057 arg_show_usage(stdout, vp8_args);
1058 #endif
1059 fprintf(stderr, "\n"
1060 "Included encoders:\n"
1061 "\n");
1063 for (i = 0; i < sizeof(codecs) / sizeof(codecs[0]); i++)
1064 fprintf(stderr, " %-6s - %s\n",
1065 codecs[i].name,
1066 vpx_codec_iface_name(codecs[i].iface));
1068 exit(EXIT_FAILURE);
1071 #define ARG_CTRL_CNT_MAX 10
1074 int main(int argc, const char **argv_)
1076 vpx_codec_ctx_t encoder;
1077 const char *in_fn = NULL, *out_fn = NULL, *stats_fn = NULL;
1078 int i;
1079 FILE *infile, *outfile;
1080 vpx_codec_enc_cfg_t cfg;
1081 vpx_codec_err_t res;
1082 int pass, one_pass_only = 0;
1083 stats_io_t stats;
1084 vpx_image_t raw;
1085 const struct codec_item *codec = codecs;
1086 int frame_avail, got_data;
1088 struct arg arg;
1089 char **argv, **argi, **argj;
1090 int arg_usage = 0, arg_passes = 1, arg_deadline = 0;
1091 int arg_ctrls[ARG_CTRL_CNT_MAX][2], arg_ctrl_cnt = 0;
1092 int arg_limit = 0;
1093 static const arg_def_t **ctrl_args = no_args;
1094 static const int *ctrl_args_map = NULL;
1095 int verbose = 0, show_psnr = 0;
1096 int arg_use_i420 = 1;
1097 unsigned long cx_time = 0;
1098 unsigned int file_type, fourcc;
1099 y4m_input y4m;
1100 struct vpx_rational arg_framerate = {30, 1};
1101 int arg_have_framerate = 0;
1102 int write_webm = 1;
1103 EbmlGlobal ebml = {0};
1104 uint32_t hash = 0;
1105 uint64_t psnr_sse_total = 0;
1106 uint64_t psnr_samples_total = 0;
1107 double psnr_totals[4] = {0, 0, 0, 0};
1108 int psnr_count = 0;
1110 exec_name = argv_[0];
1111 ebml.last_pts_ms = -1;
1113 if (argc < 3)
1114 usage_exit();
1117 /* First parse the codec and usage values, because we want to apply other
1118 * parameters on top of the default configuration provided by the codec.
1120 argv = argv_dup(argc - 1, argv_ + 1);
1122 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
1124 arg.argv_step = 1;
1126 if (arg_match(&arg, &codecarg, argi))
1128 int j, k = -1;
1130 for (j = 0; j < sizeof(codecs) / sizeof(codecs[0]); j++)
1131 if (!strcmp(codecs[j].name, arg.val))
1132 k = j;
1134 if (k >= 0)
1135 codec = codecs + k;
1136 else
1137 die("Error: Unrecognized argument (%s) to --codec\n",
1138 arg.val);
1141 else if (arg_match(&arg, &passes, argi))
1143 arg_passes = arg_parse_uint(&arg);
1145 if (arg_passes < 1 || arg_passes > 2)
1146 die("Error: Invalid number of passes (%d)\n", arg_passes);
1148 else if (arg_match(&arg, &pass_arg, argi))
1150 one_pass_only = arg_parse_uint(&arg);
1152 if (one_pass_only < 1 || one_pass_only > 2)
1153 die("Error: Invalid pass selected (%d)\n", one_pass_only);
1155 else if (arg_match(&arg, &fpf_name, argi))
1156 stats_fn = arg.val;
1157 else if (arg_match(&arg, &usage, argi))
1158 arg_usage = arg_parse_uint(&arg);
1159 else if (arg_match(&arg, &deadline, argi))
1160 arg_deadline = arg_parse_uint(&arg);
1161 else if (arg_match(&arg, &best_dl, argi))
1162 arg_deadline = VPX_DL_BEST_QUALITY;
1163 else if (arg_match(&arg, &good_dl, argi))
1164 arg_deadline = VPX_DL_GOOD_QUALITY;
1165 else if (arg_match(&arg, &rt_dl, argi))
1166 arg_deadline = VPX_DL_REALTIME;
1167 else if (arg_match(&arg, &use_yv12, argi))
1169 arg_use_i420 = 0;
1171 else if (arg_match(&arg, &use_i420, argi))
1173 arg_use_i420 = 1;
1175 else if (arg_match(&arg, &verbosearg, argi))
1176 verbose = 1;
1177 else if (arg_match(&arg, &limit, argi))
1178 arg_limit = arg_parse_uint(&arg);
1179 else if (arg_match(&arg, &psnrarg, argi))
1180 show_psnr = 1;
1181 else if (arg_match(&arg, &framerate, argi))
1183 arg_framerate = arg_parse_rational(&arg);
1184 arg_have_framerate = 1;
1186 else if (arg_match(&arg, &use_ivf, argi))
1187 write_webm = 0;
1188 else if (arg_match(&arg, &outputfile, argi))
1189 out_fn = arg.val;
1190 else if (arg_match(&arg, &debugmode, argi))
1191 ebml.debug = 1;
1192 else
1193 argj++;
1196 /* Ensure that --passes and --pass are consistent. If --pass is set and --passes=2,
1197 * ensure --fpf was set.
1199 if (one_pass_only)
1201 /* DWIM: Assume the user meant passes=2 if pass=2 is specified */
1202 if (one_pass_only > arg_passes)
1204 fprintf(stderr, "Warning: Assuming --pass=%d implies --passes=%d\n",
1205 one_pass_only, one_pass_only);
1206 arg_passes = one_pass_only;
1209 if (arg_passes == 2 && !stats_fn)
1210 die("Must specify --fpf when --pass=%d and --passes=2\n", one_pass_only);
1213 /* Populate encoder configuration */
1214 res = vpx_codec_enc_config_default(codec->iface, &cfg, arg_usage);
1216 if (res)
1218 fprintf(stderr, "Failed to get config: %s\n",
1219 vpx_codec_err_to_string(res));
1220 return EXIT_FAILURE;
1223 /* Change the default timebase to a high enough value so that the encoder
1224 * will always create strictly increasing timestamps.
1226 cfg.g_timebase.den = 1000;
1228 /* Never use the library's default resolution, require it be parsed
1229 * from the file or set on the command line.
1231 cfg.g_w = 0;
1232 cfg.g_h = 0;
1234 /* Now parse the remainder of the parameters. */
1235 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
1237 arg.argv_step = 1;
1239 if (0);
1240 else if (arg_match(&arg, &threads, argi))
1241 cfg.g_threads = arg_parse_uint(&arg);
1242 else if (arg_match(&arg, &profile, argi))
1243 cfg.g_profile = arg_parse_uint(&arg);
1244 else if (arg_match(&arg, &width, argi))
1245 cfg.g_w = arg_parse_uint(&arg);
1246 else if (arg_match(&arg, &height, argi))
1247 cfg.g_h = arg_parse_uint(&arg);
1248 else if (arg_match(&arg, &timebase, argi))
1249 cfg.g_timebase = arg_parse_rational(&arg);
1250 else if (arg_match(&arg, &error_resilient, argi))
1251 cfg.g_error_resilient = arg_parse_uint(&arg);
1252 else if (arg_match(&arg, &lag_in_frames, argi))
1253 cfg.g_lag_in_frames = arg_parse_uint(&arg);
1254 else if (arg_match(&arg, &dropframe_thresh, argi))
1255 cfg.rc_dropframe_thresh = arg_parse_uint(&arg);
1256 else if (arg_match(&arg, &resize_allowed, argi))
1257 cfg.rc_resize_allowed = arg_parse_uint(&arg);
1258 else if (arg_match(&arg, &resize_up_thresh, argi))
1259 cfg.rc_resize_up_thresh = arg_parse_uint(&arg);
1260 else if (arg_match(&arg, &resize_down_thresh, argi))
1261 cfg.rc_resize_down_thresh = arg_parse_uint(&arg);
1262 else if (arg_match(&arg, &resize_down_thresh, argi))
1263 cfg.rc_resize_down_thresh = arg_parse_uint(&arg);
1264 else if (arg_match(&arg, &end_usage, argi))
1265 cfg.rc_end_usage = arg_parse_enum_or_int(&arg);
1266 else if (arg_match(&arg, &target_bitrate, argi))
1267 cfg.rc_target_bitrate = arg_parse_uint(&arg);
1268 else if (arg_match(&arg, &min_quantizer, argi))
1269 cfg.rc_min_quantizer = arg_parse_uint(&arg);
1270 else if (arg_match(&arg, &max_quantizer, argi))
1271 cfg.rc_max_quantizer = arg_parse_uint(&arg);
1272 else if (arg_match(&arg, &undershoot_pct, argi))
1273 cfg.rc_undershoot_pct = arg_parse_uint(&arg);
1274 else if (arg_match(&arg, &overshoot_pct, argi))
1275 cfg.rc_overshoot_pct = arg_parse_uint(&arg);
1276 else if (arg_match(&arg, &buf_sz, argi))
1277 cfg.rc_buf_sz = arg_parse_uint(&arg);
1278 else if (arg_match(&arg, &buf_initial_sz, argi))
1279 cfg.rc_buf_initial_sz = arg_parse_uint(&arg);
1280 else if (arg_match(&arg, &buf_optimal_sz, argi))
1281 cfg.rc_buf_optimal_sz = arg_parse_uint(&arg);
1282 else if (arg_match(&arg, &bias_pct, argi))
1284 cfg.rc_2pass_vbr_bias_pct = arg_parse_uint(&arg);
1286 if (arg_passes < 2)
1287 fprintf(stderr,
1288 "Warning: option %s ignored in one-pass mode.\n",
1289 arg.name);
1291 else if (arg_match(&arg, &minsection_pct, argi))
1293 cfg.rc_2pass_vbr_minsection_pct = arg_parse_uint(&arg);
1295 if (arg_passes < 2)
1296 fprintf(stderr,
1297 "Warning: option %s ignored in one-pass mode.\n",
1298 arg.name);
1300 else if (arg_match(&arg, &maxsection_pct, argi))
1302 cfg.rc_2pass_vbr_maxsection_pct = arg_parse_uint(&arg);
1304 if (arg_passes < 2)
1305 fprintf(stderr,
1306 "Warning: option %s ignored in one-pass mode.\n",
1307 arg.name);
1309 else if (arg_match(&arg, &kf_min_dist, argi))
1310 cfg.kf_min_dist = arg_parse_uint(&arg);
1311 else if (arg_match(&arg, &kf_max_dist, argi))
1312 cfg.kf_max_dist = arg_parse_uint(&arg);
1313 else if (arg_match(&arg, &kf_disabled, argi))
1314 cfg.kf_mode = VPX_KF_DISABLED;
1315 else
1316 argj++;
1319 /* Handle codec specific options */
1320 #if CONFIG_VP8_ENCODER
1322 if (codec->iface == &vpx_codec_vp8_cx_algo)
1324 ctrl_args = vp8_args;
1325 ctrl_args_map = vp8_arg_ctrl_map;
1328 #endif
1330 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
1332 int match = 0;
1334 arg.argv_step = 1;
1336 for (i = 0; ctrl_args[i]; i++)
1338 if (arg_match(&arg, ctrl_args[i], argi))
1340 match = 1;
1342 if (arg_ctrl_cnt < ARG_CTRL_CNT_MAX)
1344 arg_ctrls[arg_ctrl_cnt][0] = ctrl_args_map[i];
1345 arg_ctrls[arg_ctrl_cnt][1] = arg_parse_enum_or_int(&arg);
1346 arg_ctrl_cnt++;
1351 if (!match)
1352 argj++;
1355 /* Check for unrecognized options */
1356 for (argi = argv; *argi; argi++)
1357 if (argi[0][0] == '-' && argi[0][1])
1358 die("Error: Unrecognized option %s\n", *argi);
1360 /* Handle non-option arguments */
1361 in_fn = argv[0];
1363 if (!in_fn)
1364 usage_exit();
1366 if(!out_fn)
1367 die("Error: Output file is required (specify with -o)\n");
1369 memset(&stats, 0, sizeof(stats));
1371 for (pass = one_pass_only ? one_pass_only - 1 : 0; pass < arg_passes; pass++)
1373 int frames_in = 0, frames_out = 0;
1374 unsigned long nbytes = 0;
1375 struct detect_buffer detect;
1377 /* Parse certain options from the input file, if possible */
1378 infile = strcmp(in_fn, "-") ? fopen(in_fn, "rb")
1379 : set_binary_mode(stdin);
1381 if (!infile)
1383 fprintf(stderr, "Failed to open input file\n");
1384 return EXIT_FAILURE;
1387 /* For RAW input sources, these bytes will applied on the first frame
1388 * in read_frame().
1390 detect.buf_read = fread(detect.buf, 1, 4, infile);
1391 detect.position = 0;
1393 if (detect.buf_read == 4 && file_is_y4m(infile, &y4m, detect.buf))
1395 if (y4m_input_open(&y4m, infile, detect.buf, 4) >= 0)
1397 file_type = FILE_TYPE_Y4M;
1398 cfg.g_w = y4m.pic_w;
1399 cfg.g_h = y4m.pic_h;
1401 /* Use the frame rate from the file only if none was specified
1402 * on the command-line.
1404 if (!arg_have_framerate)
1406 arg_framerate.num = y4m.fps_n;
1407 arg_framerate.den = y4m.fps_d;
1410 arg_use_i420 = 0;
1412 else
1414 fprintf(stderr, "Unsupported Y4M stream.\n");
1415 return EXIT_FAILURE;
1418 else if (detect.buf_read == 4 &&
1419 file_is_ivf(infile, &fourcc, &cfg.g_w, &cfg.g_h, &detect))
1421 file_type = FILE_TYPE_IVF;
1422 switch (fourcc)
1424 case 0x32315659:
1425 arg_use_i420 = 0;
1426 break;
1427 case 0x30323449:
1428 arg_use_i420 = 1;
1429 break;
1430 default:
1431 fprintf(stderr, "Unsupported fourcc (%08x) in IVF\n", fourcc);
1432 return EXIT_FAILURE;
1435 else
1437 file_type = FILE_TYPE_RAW;
1440 if(!cfg.g_w || !cfg.g_h)
1442 fprintf(stderr, "Specify stream dimensions with --width (-w) "
1443 " and --height (-h).\n");
1444 return EXIT_FAILURE;
1447 #define SHOW(field) fprintf(stderr, " %-28s = %d\n", #field, cfg.field)
1449 if (verbose && pass == 0)
1451 fprintf(stderr, "Codec: %s\n", vpx_codec_iface_name(codec->iface));
1452 fprintf(stderr, "Source file: %s Format: %s\n", in_fn,
1453 arg_use_i420 ? "I420" : "YV12");
1454 fprintf(stderr, "Destination file: %s\n", out_fn);
1455 fprintf(stderr, "Encoder parameters:\n");
1457 SHOW(g_usage);
1458 SHOW(g_threads);
1459 SHOW(g_profile);
1460 SHOW(g_w);
1461 SHOW(g_h);
1462 SHOW(g_timebase.num);
1463 SHOW(g_timebase.den);
1464 SHOW(g_error_resilient);
1465 SHOW(g_pass);
1466 SHOW(g_lag_in_frames);
1467 SHOW(rc_dropframe_thresh);
1468 SHOW(rc_resize_allowed);
1469 SHOW(rc_resize_up_thresh);
1470 SHOW(rc_resize_down_thresh);
1471 SHOW(rc_end_usage);
1472 SHOW(rc_target_bitrate);
1473 SHOW(rc_min_quantizer);
1474 SHOW(rc_max_quantizer);
1475 SHOW(rc_undershoot_pct);
1476 SHOW(rc_overshoot_pct);
1477 SHOW(rc_buf_sz);
1478 SHOW(rc_buf_initial_sz);
1479 SHOW(rc_buf_optimal_sz);
1480 SHOW(rc_2pass_vbr_bias_pct);
1481 SHOW(rc_2pass_vbr_minsection_pct);
1482 SHOW(rc_2pass_vbr_maxsection_pct);
1483 SHOW(kf_mode);
1484 SHOW(kf_min_dist);
1485 SHOW(kf_max_dist);
1488 if(pass == (one_pass_only ? one_pass_only - 1 : 0)) {
1489 if (file_type == FILE_TYPE_Y4M)
1490 /*The Y4M reader does its own allocation.
1491 Just initialize this here to avoid problems if we never read any
1492 frames.*/
1493 memset(&raw, 0, sizeof(raw));
1494 else
1495 vpx_img_alloc(&raw, arg_use_i420 ? VPX_IMG_FMT_I420 : VPX_IMG_FMT_YV12,
1496 cfg.g_w, cfg.g_h, 1);
1499 outfile = strcmp(out_fn, "-") ? fopen(out_fn, "wb")
1500 : set_binary_mode(stdout);
1502 if (!outfile)
1504 fprintf(stderr, "Failed to open output file\n");
1505 return EXIT_FAILURE;
1508 if(write_webm && fseek(outfile, 0, SEEK_CUR))
1510 fprintf(stderr, "WebM output to pipes not supported.\n");
1511 return EXIT_FAILURE;
1514 if (stats_fn)
1516 if (!stats_open_file(&stats, stats_fn, pass))
1518 fprintf(stderr, "Failed to open statistics store\n");
1519 return EXIT_FAILURE;
1522 else
1524 if (!stats_open_mem(&stats, pass))
1526 fprintf(stderr, "Failed to open statistics store\n");
1527 return EXIT_FAILURE;
1531 cfg.g_pass = arg_passes == 2
1532 ? pass ? VPX_RC_LAST_PASS : VPX_RC_FIRST_PASS
1533 : VPX_RC_ONE_PASS;
1534 #if VPX_ENCODER_ABI_VERSION > (1 + VPX_CODEC_ABI_VERSION)
1536 if (pass)
1538 cfg.rc_twopass_stats_in = stats_get(&stats);
1541 #endif
1543 if(write_webm)
1545 ebml.stream = outfile;
1546 write_webm_file_header(&ebml, &cfg, &arg_framerate);
1548 else
1549 write_ivf_file_header(outfile, &cfg, codec->fourcc, 0);
1552 /* Construct Encoder Context */
1553 vpx_codec_enc_init(&encoder, codec->iface, &cfg,
1554 show_psnr ? VPX_CODEC_USE_PSNR : 0);
1555 ctx_exit_on_error(&encoder, "Failed to initialize encoder");
1557 /* Note that we bypass the vpx_codec_control wrapper macro because
1558 * we're being clever to store the control IDs in an array. Real
1559 * applications will want to make use of the enumerations directly
1561 for (i = 0; i < arg_ctrl_cnt; i++)
1563 if (vpx_codec_control_(&encoder, arg_ctrls[i][0], arg_ctrls[i][1]))
1564 fprintf(stderr, "Error: Tried to set control %d = %d\n",
1565 arg_ctrls[i][0], arg_ctrls[i][1]);
1567 ctx_exit_on_error(&encoder, "Failed to control codec");
1570 frame_avail = 1;
1571 got_data = 0;
1573 while (frame_avail || got_data)
1575 vpx_codec_iter_t iter = NULL;
1576 const vpx_codec_cx_pkt_t *pkt;
1577 struct vpx_usec_timer timer;
1578 int64_t frame_start, next_frame_start;
1580 if (!arg_limit || frames_in < arg_limit)
1582 frame_avail = read_frame(infile, &raw, file_type, &y4m,
1583 &detect);
1585 if (frame_avail)
1586 frames_in++;
1588 fprintf(stderr,
1589 "\rPass %d/%d frame %4d/%-4d %7ldB \033[K", pass + 1,
1590 arg_passes, frames_in, frames_out, nbytes);
1592 else
1593 frame_avail = 0;
1595 vpx_usec_timer_start(&timer);
1597 frame_start = (cfg.g_timebase.den * (int64_t)(frames_in - 1)
1598 * arg_framerate.den) / cfg.g_timebase.num / arg_framerate.num;
1599 next_frame_start = (cfg.g_timebase.den * (int64_t)(frames_in)
1600 * arg_framerate.den)
1601 / cfg.g_timebase.num / arg_framerate.num;
1602 vpx_codec_encode(&encoder, frame_avail ? &raw : NULL, frame_start,
1603 next_frame_start - frame_start,
1604 0, arg_deadline);
1605 vpx_usec_timer_mark(&timer);
1606 cx_time += vpx_usec_timer_elapsed(&timer);
1607 ctx_exit_on_error(&encoder, "Failed to encode frame");
1608 got_data = 0;
1610 while ((pkt = vpx_codec_get_cx_data(&encoder, &iter)))
1612 got_data = 1;
1614 switch (pkt->kind)
1616 case VPX_CODEC_CX_FRAME_PKT:
1617 frames_out++;
1618 fprintf(stderr, " %6luF",
1619 (unsigned long)pkt->data.frame.sz);
1621 if(write_webm)
1623 /* Update the hash */
1624 if(!ebml.debug)
1625 hash = murmur(pkt->data.frame.buf,
1626 pkt->data.frame.sz, hash);
1628 write_webm_block(&ebml, &cfg, pkt);
1630 else
1632 write_ivf_frame_header(outfile, pkt);
1633 if(fwrite(pkt->data.frame.buf, 1,
1634 pkt->data.frame.sz, outfile));
1636 nbytes += pkt->data.raw.sz;
1637 break;
1638 case VPX_CODEC_STATS_PKT:
1639 frames_out++;
1640 fprintf(stderr, " %6luS",
1641 (unsigned long)pkt->data.twopass_stats.sz);
1642 stats_write(&stats,
1643 pkt->data.twopass_stats.buf,
1644 pkt->data.twopass_stats.sz);
1645 nbytes += pkt->data.raw.sz;
1646 break;
1647 case VPX_CODEC_PSNR_PKT:
1649 if (show_psnr)
1651 int i;
1653 psnr_sse_total += pkt->data.psnr.sse[0];
1654 psnr_samples_total += pkt->data.psnr.samples[0];
1655 for (i = 0; i < 4; i++)
1657 fprintf(stderr, "%.3lf ", pkt->data.psnr.psnr[i]);
1658 psnr_totals[i] += pkt->data.psnr.psnr[i];
1660 psnr_count++;
1663 break;
1664 default:
1665 break;
1669 fflush(stdout);
1672 fprintf(stderr,
1673 "\rPass %d/%d frame %4d/%-4d %7ldB %7ldb/f %7"PRId64"b/s"
1674 " %7lu %s (%.2f fps)\033[K", pass + 1,
1675 arg_passes, frames_in, frames_out, nbytes, nbytes * 8 / frames_in,
1676 nbytes * 8 *(int64_t)arg_framerate.num / arg_framerate.den / frames_in,
1677 cx_time > 9999999 ? cx_time / 1000 : cx_time,
1678 cx_time > 9999999 ? "ms" : "us",
1679 (float)frames_in * 1000000.0 / (float)cx_time);
1681 if ( (show_psnr) && (psnr_count>0) )
1683 int i;
1684 double ovpsnr = vp8_mse2psnr(psnr_samples_total, 255.0,
1685 psnr_sse_total);
1687 fprintf(stderr, "\nPSNR (Overall/Avg/Y/U/V)");
1689 fprintf(stderr, " %.3lf", ovpsnr);
1690 for (i = 0; i < 4; i++)
1692 fprintf(stderr, " %.3lf", psnr_totals[i]/psnr_count);
1696 vpx_codec_destroy(&encoder);
1698 fclose(infile);
1700 if(write_webm)
1702 write_webm_file_footer(&ebml, hash);
1704 else
1706 if (!fseek(outfile, 0, SEEK_SET))
1707 write_ivf_file_header(outfile, &cfg, codec->fourcc, frames_out);
1710 fclose(outfile);
1711 stats_close(&stats, arg_passes-1);
1712 fprintf(stderr, "\n");
1714 if (one_pass_only)
1715 break;
1718 vpx_img_free(&raw);
1719 free(argv);
1720 return EXIT_SUCCESS;