Moved zbin_mode_boost to macroblock struct
[aom.git] / vpxenc.c
blobc9547ea3c8b3da688ea16006556cf7196562b9f5
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/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 as long
51 and uses f{seek,tell}o64/off64_t for large files */
52 #define fseeko fseeko64
53 #define ftello ftello64
54 #define off_t off64_t
55 #endif
57 #define LITERALU64(hi,lo) ((((uint64_t)hi)<<32)|lo)
59 /* We should use 32-bit file operations in WebM file format
60 * when building ARM executable file (.axf) with RVCT */
61 #if !CONFIG_OS_SUPPORT
62 typedef long off_t;
63 #define fseeko fseek
64 #define ftello ftell
65 #endif
67 /* Swallow warnings about unused results of fread/fwrite */
68 static size_t wrap_fread(void *ptr, size_t size, size_t nmemb,
69 FILE *stream)
71 return fread(ptr, size, nmemb, stream);
73 #define fread wrap_fread
75 static size_t wrap_fwrite(const void *ptr, size_t size, size_t nmemb,
76 FILE *stream)
78 return fwrite(ptr, size, nmemb, stream);
80 #define fwrite wrap_fwrite
83 static const char *exec_name;
85 static const struct codec_item
87 char const *name;
88 vpx_codec_iface_t *iface;
89 unsigned int fourcc;
90 } codecs[] =
92 #if CONFIG_VP8_ENCODER
93 {"vp8", &vpx_codec_vp8_cx_algo, 0x30385056},
94 #endif
97 static void usage_exit();
99 #define LOG_ERROR(label) do \
101 const char *l=label;\
102 va_list ap;\
103 va_start(ap, fmt);\
104 if(l)\
105 fprintf(stderr, "%s: ", l);\
106 vfprintf(stderr, fmt, ap);\
107 fprintf(stderr, "\n");\
108 va_end(ap);\
109 } while(0)
111 void die(const char *fmt, ...)
113 LOG_ERROR(NULL);
114 usage_exit();
118 void fatal(const char *fmt, ...)
120 LOG_ERROR("Fatal");
121 exit(EXIT_FAILURE);
125 void warn(const char *fmt, ...)
127 LOG_ERROR("Warning");
131 static void ctx_exit_on_error(vpx_codec_ctx_t *ctx, const char *s, ...)
133 va_list ap;
135 va_start(ap, s);
136 if (ctx->err)
138 const char *detail = vpx_codec_error_detail(ctx);
140 vfprintf(stderr, s, ap);
141 fprintf(stderr, ": %s\n", vpx_codec_error(ctx));
143 if (detail)
144 fprintf(stderr, " %s\n", detail);
146 exit(EXIT_FAILURE);
150 /* This structure is used to abstract the different ways of handling
151 * first pass statistics.
153 typedef struct
155 vpx_fixed_buf_t buf;
156 int pass;
157 FILE *file;
158 char *buf_ptr;
159 size_t buf_alloc_sz;
160 } stats_io_t;
162 int stats_open_file(stats_io_t *stats, const char *fpf, int pass)
164 int res;
166 stats->pass = pass;
168 if (pass == 0)
170 stats->file = fopen(fpf, "wb");
171 stats->buf.sz = 0;
172 stats->buf.buf = NULL,
173 res = (stats->file != NULL);
175 else
177 #if 0
178 #elif USE_POSIX_MMAP
179 struct stat stat_buf;
180 int fd;
182 fd = open(fpf, O_RDONLY);
183 stats->file = fdopen(fd, "rb");
184 fstat(fd, &stat_buf);
185 stats->buf.sz = stat_buf.st_size;
186 stats->buf.buf = mmap(NULL, stats->buf.sz, PROT_READ, MAP_PRIVATE,
187 fd, 0);
188 res = (stats->buf.buf != NULL);
189 #else
190 size_t nbytes;
192 stats->file = fopen(fpf, "rb");
194 if (fseek(stats->file, 0, SEEK_END))
195 fatal("First-pass stats file must be seekable!");
197 stats->buf.sz = stats->buf_alloc_sz = ftell(stats->file);
198 rewind(stats->file);
200 stats->buf.buf = malloc(stats->buf_alloc_sz);
202 if (!stats->buf.buf)
203 fatal("Failed to allocate first-pass stats buffer (%lu bytes)",
204 (unsigned long)stats->buf_alloc_sz);
206 nbytes = fread(stats->buf.buf, 1, stats->buf.sz, stats->file);
207 res = (nbytes == stats->buf.sz);
208 #endif
211 return res;
214 int stats_open_mem(stats_io_t *stats, int pass)
216 int res;
217 stats->pass = pass;
219 if (!pass)
221 stats->buf.sz = 0;
222 stats->buf_alloc_sz = 64 * 1024;
223 stats->buf.buf = malloc(stats->buf_alloc_sz);
226 stats->buf_ptr = stats->buf.buf;
227 res = (stats->buf.buf != NULL);
228 return res;
232 void stats_close(stats_io_t *stats, int last_pass)
234 if (stats->file)
236 if (stats->pass == last_pass)
238 #if 0
239 #elif USE_POSIX_MMAP
240 munmap(stats->buf.buf, stats->buf.sz);
241 #else
242 free(stats->buf.buf);
243 #endif
246 fclose(stats->file);
247 stats->file = NULL;
249 else
251 if (stats->pass == last_pass)
252 free(stats->buf.buf);
256 void stats_write(stats_io_t *stats, const void *pkt, size_t len)
258 if (stats->file)
260 (void) fwrite(pkt, 1, len, stats->file);
262 else
264 if (stats->buf.sz + len > stats->buf_alloc_sz)
266 size_t new_sz = stats->buf_alloc_sz + 64 * 1024;
267 char *new_ptr = realloc(stats->buf.buf, new_sz);
269 if (new_ptr)
271 stats->buf_ptr = new_ptr + (stats->buf_ptr - (char *)stats->buf.buf);
272 stats->buf.buf = new_ptr;
273 stats->buf_alloc_sz = new_sz;
275 else
276 fatal("Failed to realloc firstpass stats buffer.");
279 memcpy(stats->buf_ptr, pkt, len);
280 stats->buf.sz += len;
281 stats->buf_ptr += len;
285 vpx_fixed_buf_t stats_get(stats_io_t *stats)
287 return stats->buf;
290 /* Stereo 3D packed frame format */
291 typedef enum stereo_format
293 STEREO_FORMAT_MONO = 0,
294 STEREO_FORMAT_LEFT_RIGHT = 1,
295 STEREO_FORMAT_BOTTOM_TOP = 2,
296 STEREO_FORMAT_TOP_BOTTOM = 3,
297 STEREO_FORMAT_RIGHT_LEFT = 11
298 } stereo_format_t;
300 enum video_file_type
302 FILE_TYPE_RAW,
303 FILE_TYPE_IVF,
304 FILE_TYPE_Y4M
307 struct detect_buffer {
308 char buf[4];
309 size_t buf_read;
310 size_t position;
314 struct input_state
316 char *fn;
317 FILE *file;
318 y4m_input y4m;
319 struct detect_buffer detect;
320 enum video_file_type file_type;
321 unsigned int w;
322 unsigned int h;
323 struct vpx_rational framerate;
324 int use_i420;
328 #define IVF_FRAME_HDR_SZ (4+8) /* 4 byte size + 8 byte timestamp */
329 static int read_frame(struct input_state *input, vpx_image_t *img)
331 FILE *f = input->file;
332 enum video_file_type file_type = input->file_type;
333 y4m_input *y4m = &input->y4m;
334 struct detect_buffer *detect = &input->detect;
335 int plane = 0;
336 int shortread = 0;
338 if (file_type == FILE_TYPE_Y4M)
340 if (y4m_input_fetch_frame(y4m, f, img) < 1)
341 return 0;
343 else
345 if (file_type == FILE_TYPE_IVF)
347 char junk[IVF_FRAME_HDR_SZ];
349 /* Skip the frame header. We know how big the frame should be. See
350 * write_ivf_frame_header() for documentation on the frame header
351 * layout.
353 (void) fread(junk, 1, IVF_FRAME_HDR_SZ, f);
356 for (plane = 0; plane < 3; plane++)
358 unsigned char *ptr;
359 int w = (plane ? (1 + img->d_w) / 2 : img->d_w);
360 int h = (plane ? (1 + img->d_h) / 2 : img->d_h);
361 int r;
363 /* Determine the correct plane based on the image format. The for-loop
364 * always counts in Y,U,V order, but this may not match the order of
365 * the data on disk.
367 switch (plane)
369 case 1:
370 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U];
371 break;
372 case 2:
373 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V];
374 break;
375 default:
376 ptr = img->planes[plane];
379 for (r = 0; r < h; r++)
381 size_t needed = w;
382 size_t buf_position = 0;
383 const size_t left = detect->buf_read - detect->position;
384 if (left > 0)
386 const size_t more = (left < needed) ? left : needed;
387 memcpy(ptr, detect->buf + detect->position, more);
388 buf_position = more;
389 needed -= more;
390 detect->position += more;
392 if (needed > 0)
394 shortread |= (fread(ptr + buf_position, 1, needed, f) < needed);
397 ptr += img->stride[plane];
402 return !shortread;
406 unsigned int file_is_y4m(FILE *infile,
407 y4m_input *y4m,
408 char detect[4])
410 if(memcmp(detect, "YUV4", 4) == 0)
412 return 1;
414 return 0;
417 #define IVF_FILE_HDR_SZ (32)
418 unsigned int file_is_ivf(struct input_state *input,
419 unsigned int *fourcc)
421 char raw_hdr[IVF_FILE_HDR_SZ];
422 int is_ivf = 0;
423 FILE *infile = input->file;
424 unsigned int *width = &input->w;
425 unsigned int *height = &input->h;
426 struct detect_buffer *detect = &input->detect;
428 if(memcmp(detect->buf, "DKIF", 4) != 0)
429 return 0;
431 /* See write_ivf_file_header() for more documentation on the file header
432 * layout.
434 if (fread(raw_hdr + 4, 1, IVF_FILE_HDR_SZ - 4, infile)
435 == IVF_FILE_HDR_SZ - 4)
438 is_ivf = 1;
440 if (mem_get_le16(raw_hdr + 4) != 0)
441 warn("Unrecognized IVF version! This file may not decode "
442 "properly.");
444 *fourcc = mem_get_le32(raw_hdr + 8);
448 if (is_ivf)
450 *width = mem_get_le16(raw_hdr + 12);
451 *height = mem_get_le16(raw_hdr + 14);
452 detect->position = 4;
455 return is_ivf;
459 static void write_ivf_file_header(FILE *outfile,
460 const vpx_codec_enc_cfg_t *cfg,
461 unsigned int fourcc,
462 int frame_cnt)
464 char header[32];
466 if (cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS)
467 return;
469 header[0] = 'D';
470 header[1] = 'K';
471 header[2] = 'I';
472 header[3] = 'F';
473 mem_put_le16(header + 4, 0); /* version */
474 mem_put_le16(header + 6, 32); /* headersize */
475 mem_put_le32(header + 8, fourcc); /* headersize */
476 mem_put_le16(header + 12, cfg->g_w); /* width */
477 mem_put_le16(header + 14, cfg->g_h); /* height */
478 mem_put_le32(header + 16, cfg->g_timebase.den); /* rate */
479 mem_put_le32(header + 20, cfg->g_timebase.num); /* scale */
480 mem_put_le32(header + 24, frame_cnt); /* length */
481 mem_put_le32(header + 28, 0); /* unused */
483 (void) fwrite(header, 1, 32, outfile);
487 static void write_ivf_frame_header(FILE *outfile,
488 const vpx_codec_cx_pkt_t *pkt)
490 char header[12];
491 vpx_codec_pts_t pts;
493 if (pkt->kind != VPX_CODEC_CX_FRAME_PKT)
494 return;
496 pts = pkt->data.frame.pts;
497 mem_put_le32(header, (int)pkt->data.frame.sz);
498 mem_put_le32(header + 4, pts & 0xFFFFFFFF);
499 mem_put_le32(header + 8, pts >> 32);
501 (void) fwrite(header, 1, 12, outfile);
504 static void write_ivf_frame_size(FILE *outfile, size_t size)
506 char header[4];
507 mem_put_le32(header, (int)size);
508 (void) fwrite(header, 1, 4, outfile);
512 typedef off_t EbmlLoc;
515 struct cue_entry
517 unsigned int time;
518 uint64_t loc;
522 struct EbmlGlobal
524 int debug;
526 FILE *stream;
527 int64_t last_pts_ms;
528 vpx_rational_t framerate;
530 /* These pointers are to the start of an element */
531 off_t position_reference;
532 off_t seek_info_pos;
533 off_t segment_info_pos;
534 off_t track_pos;
535 off_t cue_pos;
536 off_t cluster_pos;
538 /* This pointer is to a specific element to be serialized */
539 off_t track_id_pos;
541 /* These pointers are to the size field of the element */
542 EbmlLoc startSegment;
543 EbmlLoc startCluster;
545 uint32_t cluster_timecode;
546 int cluster_open;
548 struct cue_entry *cue_list;
549 unsigned int cues;
554 void Ebml_Write(EbmlGlobal *glob, const void *buffer_in, unsigned long len)
556 (void) fwrite(buffer_in, 1, len, glob->stream);
559 #define WRITE_BUFFER(s) \
560 for(i = len-1; i>=0; i--)\
562 x = (char)(*(const s *)buffer_in >> (i * CHAR_BIT)); \
563 Ebml_Write(glob, &x, 1); \
565 void Ebml_Serialize(EbmlGlobal *glob, const void *buffer_in, int buffer_size, unsigned long len)
567 char x;
568 int i;
570 /* buffer_size:
571 * 1 - int8_t;
572 * 2 - int16_t;
573 * 3 - int32_t;
574 * 4 - int64_t;
576 switch (buffer_size)
578 case 1:
579 WRITE_BUFFER(int8_t)
580 break;
581 case 2:
582 WRITE_BUFFER(int16_t)
583 break;
584 case 4:
585 WRITE_BUFFER(int32_t)
586 break;
587 case 8:
588 WRITE_BUFFER(int64_t)
589 break;
590 default:
591 break;
594 #undef WRITE_BUFFER
596 /* Need a fixed size serializer for the track ID. libmkv provides a 64 bit
597 * one, but not a 32 bit one.
599 static void Ebml_SerializeUnsigned32(EbmlGlobal *glob, unsigned long class_id, uint64_t ui)
601 unsigned char sizeSerialized = 4 | 0x80;
602 Ebml_WriteID(glob, class_id);
603 Ebml_Serialize(glob, &sizeSerialized, sizeof(sizeSerialized), 1);
604 Ebml_Serialize(glob, &ui, sizeof(ui), 4);
608 static void
609 Ebml_StartSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc,
610 unsigned long class_id)
612 /* todo this is always taking 8 bytes, this may need later optimization */
613 /* this is a key that says length unknown */
614 uint64_t unknownLen = LITERALU64(0x01FFFFFF, 0xFFFFFFFF);
616 Ebml_WriteID(glob, class_id);
617 *ebmlLoc = ftello(glob->stream);
618 Ebml_Serialize(glob, &unknownLen, sizeof(unknownLen), 8);
621 static void
622 Ebml_EndSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc)
624 off_t pos;
625 uint64_t size;
627 /* Save the current stream pointer */
628 pos = ftello(glob->stream);
630 /* Calculate the size of this element */
631 size = pos - *ebmlLoc - 8;
632 size |= LITERALU64(0x01000000,0x00000000);
634 /* Seek back to the beginning of the element and write the new size */
635 fseeko(glob->stream, *ebmlLoc, SEEK_SET);
636 Ebml_Serialize(glob, &size, sizeof(size), 8);
638 /* Reset the stream pointer */
639 fseeko(glob->stream, pos, SEEK_SET);
643 static void
644 write_webm_seek_element(EbmlGlobal *ebml, unsigned long id, off_t pos)
646 uint64_t offset = pos - ebml->position_reference;
647 EbmlLoc start;
648 Ebml_StartSubElement(ebml, &start, Seek);
649 Ebml_SerializeBinary(ebml, SeekID, id);
650 Ebml_SerializeUnsigned64(ebml, SeekPosition, offset);
651 Ebml_EndSubElement(ebml, &start);
655 static void
656 write_webm_seek_info(EbmlGlobal *ebml)
659 off_t pos;
661 /* Save the current stream pointer */
662 pos = ftello(ebml->stream);
664 if(ebml->seek_info_pos)
665 fseeko(ebml->stream, ebml->seek_info_pos, SEEK_SET);
666 else
667 ebml->seek_info_pos = pos;
670 EbmlLoc start;
672 Ebml_StartSubElement(ebml, &start, SeekHead);
673 write_webm_seek_element(ebml, Tracks, ebml->track_pos);
674 write_webm_seek_element(ebml, Cues, ebml->cue_pos);
675 write_webm_seek_element(ebml, Info, ebml->segment_info_pos);
676 Ebml_EndSubElement(ebml, &start);
679 /* segment info */
680 EbmlLoc startInfo;
681 uint64_t frame_time;
682 char version_string[64];
684 /* Assemble version string */
685 if(ebml->debug)
686 strcpy(version_string, "vpxenc");
687 else
689 strcpy(version_string, "vpxenc ");
690 strncat(version_string,
691 vpx_codec_version_str(),
692 sizeof(version_string) - 1 - strlen(version_string));
695 frame_time = (uint64_t)1000 * ebml->framerate.den
696 / ebml->framerate.num;
697 ebml->segment_info_pos = ftello(ebml->stream);
698 Ebml_StartSubElement(ebml, &startInfo, Info);
699 Ebml_SerializeUnsigned(ebml, TimecodeScale, 1000000);
700 Ebml_SerializeFloat(ebml, Segment_Duration,
701 (double)(ebml->last_pts_ms + frame_time));
702 Ebml_SerializeString(ebml, 0x4D80, version_string);
703 Ebml_SerializeString(ebml, 0x5741, version_string);
704 Ebml_EndSubElement(ebml, &startInfo);
709 static void
710 write_webm_file_header(EbmlGlobal *glob,
711 const vpx_codec_enc_cfg_t *cfg,
712 const struct vpx_rational *fps,
713 stereo_format_t stereo_fmt)
716 EbmlLoc start;
717 Ebml_StartSubElement(glob, &start, EBML);
718 Ebml_SerializeUnsigned(glob, EBMLVersion, 1);
719 Ebml_SerializeUnsigned(glob, EBMLReadVersion, 1);
720 Ebml_SerializeUnsigned(glob, EBMLMaxIDLength, 4);
721 Ebml_SerializeUnsigned(glob, EBMLMaxSizeLength, 8);
722 Ebml_SerializeString(glob, DocType, "webm");
723 Ebml_SerializeUnsigned(glob, DocTypeVersion, 2);
724 Ebml_SerializeUnsigned(glob, DocTypeReadVersion, 2);
725 Ebml_EndSubElement(glob, &start);
728 Ebml_StartSubElement(glob, &glob->startSegment, Segment);
729 glob->position_reference = ftello(glob->stream);
730 glob->framerate = *fps;
731 write_webm_seek_info(glob);
734 EbmlLoc trackStart;
735 glob->track_pos = ftello(glob->stream);
736 Ebml_StartSubElement(glob, &trackStart, Tracks);
738 unsigned int trackNumber = 1;
739 uint64_t trackID = 0;
741 EbmlLoc start;
742 Ebml_StartSubElement(glob, &start, TrackEntry);
743 Ebml_SerializeUnsigned(glob, TrackNumber, trackNumber);
744 glob->track_id_pos = ftello(glob->stream);
745 Ebml_SerializeUnsigned32(glob, TrackUID, trackID);
746 Ebml_SerializeUnsigned(glob, TrackType, 1);
747 Ebml_SerializeString(glob, CodecID, "V_VP8");
749 unsigned int pixelWidth = cfg->g_w;
750 unsigned int pixelHeight = cfg->g_h;
751 float frameRate = (float)fps->num/(float)fps->den;
753 EbmlLoc videoStart;
754 Ebml_StartSubElement(glob, &videoStart, Video);
755 Ebml_SerializeUnsigned(glob, PixelWidth, pixelWidth);
756 Ebml_SerializeUnsigned(glob, PixelHeight, pixelHeight);
757 Ebml_SerializeUnsigned(glob, StereoMode, stereo_fmt);
758 Ebml_SerializeFloat(glob, FrameRate, frameRate);
759 Ebml_EndSubElement(glob, &videoStart);
761 Ebml_EndSubElement(glob, &start); /* Track Entry */
763 Ebml_EndSubElement(glob, &trackStart);
765 /* segment element is open */
770 static void
771 write_webm_block(EbmlGlobal *glob,
772 const vpx_codec_enc_cfg_t *cfg,
773 const vpx_codec_cx_pkt_t *pkt)
775 unsigned long block_length;
776 unsigned char track_number;
777 unsigned short block_timecode = 0;
778 unsigned char flags;
779 int64_t pts_ms;
780 int start_cluster = 0, is_keyframe;
782 /* Calculate the PTS of this frame in milliseconds */
783 pts_ms = pkt->data.frame.pts * 1000
784 * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
785 if(pts_ms <= glob->last_pts_ms)
786 pts_ms = glob->last_pts_ms + 1;
787 glob->last_pts_ms = pts_ms;
789 /* Calculate the relative time of this block */
790 if(pts_ms - glob->cluster_timecode > SHRT_MAX)
791 start_cluster = 1;
792 else
793 block_timecode = (unsigned short)pts_ms - glob->cluster_timecode;
795 is_keyframe = (pkt->data.frame.flags & VPX_FRAME_IS_KEY);
796 if(start_cluster || is_keyframe)
798 if(glob->cluster_open)
799 Ebml_EndSubElement(glob, &glob->startCluster);
801 /* Open the new cluster */
802 block_timecode = 0;
803 glob->cluster_open = 1;
804 glob->cluster_timecode = (uint32_t)pts_ms;
805 glob->cluster_pos = ftello(glob->stream);
806 Ebml_StartSubElement(glob, &glob->startCluster, Cluster); /* cluster */
807 Ebml_SerializeUnsigned(glob, Timecode, glob->cluster_timecode);
809 /* Save a cue point if this is a keyframe. */
810 if(is_keyframe)
812 struct cue_entry *cue, *new_cue_list;
814 new_cue_list = realloc(glob->cue_list,
815 (glob->cues+1) * sizeof(struct cue_entry));
816 if(new_cue_list)
817 glob->cue_list = new_cue_list;
818 else
819 fatal("Failed to realloc cue list.");
821 cue = &glob->cue_list[glob->cues];
822 cue->time = glob->cluster_timecode;
823 cue->loc = glob->cluster_pos;
824 glob->cues++;
828 /* Write the Simple Block */
829 Ebml_WriteID(glob, SimpleBlock);
831 block_length = (unsigned long)pkt->data.frame.sz + 4;
832 block_length |= 0x10000000;
833 Ebml_Serialize(glob, &block_length, sizeof(block_length), 4);
835 track_number = 1;
836 track_number |= 0x80;
837 Ebml_Write(glob, &track_number, 1);
839 Ebml_Serialize(glob, &block_timecode, sizeof(block_timecode), 2);
841 flags = 0;
842 if(is_keyframe)
843 flags |= 0x80;
844 if(pkt->data.frame.flags & VPX_FRAME_IS_INVISIBLE)
845 flags |= 0x08;
846 Ebml_Write(glob, &flags, 1);
848 Ebml_Write(glob, pkt->data.frame.buf, (unsigned long)pkt->data.frame.sz);
852 static void
853 write_webm_file_footer(EbmlGlobal *glob, long hash)
856 if(glob->cluster_open)
857 Ebml_EndSubElement(glob, &glob->startCluster);
860 EbmlLoc start;
861 unsigned int i;
863 glob->cue_pos = ftello(glob->stream);
864 Ebml_StartSubElement(glob, &start, Cues);
865 for(i=0; i<glob->cues; i++)
867 struct cue_entry *cue = &glob->cue_list[i];
868 EbmlLoc start;
870 Ebml_StartSubElement(glob, &start, CuePoint);
872 EbmlLoc start;
874 Ebml_SerializeUnsigned(glob, CueTime, cue->time);
876 Ebml_StartSubElement(glob, &start, CueTrackPositions);
877 Ebml_SerializeUnsigned(glob, CueTrack, 1);
878 Ebml_SerializeUnsigned64(glob, CueClusterPosition,
879 cue->loc - glob->position_reference);
880 Ebml_EndSubElement(glob, &start);
882 Ebml_EndSubElement(glob, &start);
884 Ebml_EndSubElement(glob, &start);
887 Ebml_EndSubElement(glob, &glob->startSegment);
889 /* Patch up the seek info block */
890 write_webm_seek_info(glob);
892 /* Patch up the track id */
893 fseeko(glob->stream, glob->track_id_pos, SEEK_SET);
894 Ebml_SerializeUnsigned32(glob, TrackUID, glob->debug ? 0xDEADBEEF : hash);
896 fseeko(glob->stream, 0, SEEK_END);
900 /* Murmur hash derived from public domain reference implementation at
901 * http://sites.google.com/site/murmurhash/
903 static unsigned int murmur ( const void * key, int len, unsigned int seed )
905 const unsigned int m = 0x5bd1e995;
906 const int r = 24;
908 unsigned int h = seed ^ len;
910 const unsigned char * data = (const unsigned char *)key;
912 while(len >= 4)
914 unsigned int k;
916 k = data[0];
917 k |= data[1] << 8;
918 k |= data[2] << 16;
919 k |= data[3] << 24;
921 k *= m;
922 k ^= k >> r;
923 k *= m;
925 h *= m;
926 h ^= k;
928 data += 4;
929 len -= 4;
932 switch(len)
934 case 3: h ^= data[2] << 16;
935 case 2: h ^= data[1] << 8;
936 case 1: h ^= data[0];
937 h *= m;
940 h ^= h >> 13;
941 h *= m;
942 h ^= h >> 15;
944 return h;
947 #include "math.h"
949 static double vp8_mse2psnr(double Samples, double Peak, double Mse)
951 double psnr;
953 if ((double)Mse > 0.0)
954 psnr = 10.0 * log10(Peak * Peak * Samples / Mse);
955 else
956 psnr = 60; /* Limit to prevent / 0 */
958 if (psnr > 60)
959 psnr = 60;
961 return psnr;
965 #include "args.h"
966 static const arg_def_t debugmode = ARG_DEF("D", "debug", 0,
967 "Debug mode (makes output deterministic)");
968 static const arg_def_t outputfile = ARG_DEF("o", "output", 1,
969 "Output filename");
970 static const arg_def_t use_yv12 = ARG_DEF(NULL, "yv12", 0,
971 "Input file is YV12 ");
972 static const arg_def_t use_i420 = ARG_DEF(NULL, "i420", 0,
973 "Input file is I420 (default)");
974 static const arg_def_t codecarg = ARG_DEF(NULL, "codec", 1,
975 "Codec to use");
976 static const arg_def_t passes = ARG_DEF("p", "passes", 1,
977 "Number of passes (1/2)");
978 static const arg_def_t pass_arg = ARG_DEF(NULL, "pass", 1,
979 "Pass to execute (1/2)");
980 static const arg_def_t fpf_name = ARG_DEF(NULL, "fpf", 1,
981 "First pass statistics file name");
982 static const arg_def_t limit = ARG_DEF(NULL, "limit", 1,
983 "Stop encoding after n input frames");
984 static const arg_def_t deadline = ARG_DEF("d", "deadline", 1,
985 "Deadline per frame (usec)");
986 static const arg_def_t best_dl = ARG_DEF(NULL, "best", 0,
987 "Use Best Quality Deadline");
988 static const arg_def_t good_dl = ARG_DEF(NULL, "good", 0,
989 "Use Good Quality Deadline");
990 static const arg_def_t rt_dl = ARG_DEF(NULL, "rt", 0,
991 "Use Realtime Quality Deadline");
992 static const arg_def_t quietarg = ARG_DEF("q", "quiet", 0,
993 "Do not print encode progress");
994 static const arg_def_t verbosearg = ARG_DEF("v", "verbose", 0,
995 "Show encoder parameters");
996 static const arg_def_t psnrarg = ARG_DEF(NULL, "psnr", 0,
997 "Show PSNR in status line");
998 static const arg_def_t framerate = ARG_DEF(NULL, "fps", 1,
999 "Stream frame rate (rate/scale)");
1000 static const arg_def_t use_ivf = ARG_DEF(NULL, "ivf", 0,
1001 "Output IVF (default is WebM)");
1002 static const arg_def_t out_part = ARG_DEF("P", "output-partitions", 0,
1003 "Makes encoder output partitions. Requires IVF output!");
1004 static const arg_def_t q_hist_n = ARG_DEF(NULL, "q-hist", 1,
1005 "Show quantizer histogram (n-buckets)");
1006 static const arg_def_t rate_hist_n = ARG_DEF(NULL, "rate-hist", 1,
1007 "Show rate histogram (n-buckets)");
1008 static const arg_def_t *main_args[] =
1010 &debugmode,
1011 &outputfile, &codecarg, &passes, &pass_arg, &fpf_name, &limit, &deadline,
1012 &best_dl, &good_dl, &rt_dl,
1013 &quietarg, &verbosearg, &psnrarg, &use_ivf, &out_part, &q_hist_n, &rate_hist_n,
1014 NULL
1017 static const arg_def_t usage = ARG_DEF("u", "usage", 1,
1018 "Usage profile number to use");
1019 static const arg_def_t threads = ARG_DEF("t", "threads", 1,
1020 "Max number of threads to use");
1021 static const arg_def_t profile = ARG_DEF(NULL, "profile", 1,
1022 "Bitstream profile number to use");
1023 static const arg_def_t width = ARG_DEF("w", "width", 1,
1024 "Frame width");
1025 static const arg_def_t height = ARG_DEF("h", "height", 1,
1026 "Frame height");
1027 static const struct arg_enum_list stereo_mode_enum[] = {
1028 {"mono" , STEREO_FORMAT_MONO},
1029 {"left-right", STEREO_FORMAT_LEFT_RIGHT},
1030 {"bottom-top", STEREO_FORMAT_BOTTOM_TOP},
1031 {"top-bottom", STEREO_FORMAT_TOP_BOTTOM},
1032 {"right-left", STEREO_FORMAT_RIGHT_LEFT},
1033 {NULL, 0}
1035 static const arg_def_t stereo_mode = ARG_DEF_ENUM(NULL, "stereo-mode", 1,
1036 "Stereo 3D video format", stereo_mode_enum);
1037 static const arg_def_t timebase = ARG_DEF(NULL, "timebase", 1,
1038 "Output timestamp precision (fractional seconds)");
1039 static const arg_def_t error_resilient = ARG_DEF(NULL, "error-resilient", 1,
1040 "Enable error resiliency features");
1041 static const arg_def_t lag_in_frames = ARG_DEF(NULL, "lag-in-frames", 1,
1042 "Max number of frames to lag");
1044 static const arg_def_t *global_args[] =
1046 &use_yv12, &use_i420, &usage, &threads, &profile,
1047 &width, &height, &stereo_mode, &timebase, &framerate, &error_resilient,
1048 &lag_in_frames, NULL
1051 static const arg_def_t dropframe_thresh = ARG_DEF(NULL, "drop-frame", 1,
1052 "Temporal resampling threshold (buf %)");
1053 static const arg_def_t resize_allowed = ARG_DEF(NULL, "resize-allowed", 1,
1054 "Spatial resampling enabled (bool)");
1055 static const arg_def_t resize_up_thresh = ARG_DEF(NULL, "resize-up", 1,
1056 "Upscale threshold (buf %)");
1057 static const arg_def_t resize_down_thresh = ARG_DEF(NULL, "resize-down", 1,
1058 "Downscale threshold (buf %)");
1059 static const struct arg_enum_list end_usage_enum[] = {
1060 {"vbr", VPX_VBR},
1061 {"cbr", VPX_CBR},
1062 {"cq", VPX_CQ},
1063 {NULL, 0}
1065 static const arg_def_t end_usage = ARG_DEF_ENUM(NULL, "end-usage", 1,
1066 "Rate control mode", end_usage_enum);
1067 static const arg_def_t target_bitrate = ARG_DEF(NULL, "target-bitrate", 1,
1068 "Bitrate (kbps)");
1069 static const arg_def_t min_quantizer = ARG_DEF(NULL, "min-q", 1,
1070 "Minimum (best) quantizer");
1071 static const arg_def_t max_quantizer = ARG_DEF(NULL, "max-q", 1,
1072 "Maximum (worst) quantizer");
1073 static const arg_def_t undershoot_pct = ARG_DEF(NULL, "undershoot-pct", 1,
1074 "Datarate undershoot (min) target (%)");
1075 static const arg_def_t overshoot_pct = ARG_DEF(NULL, "overshoot-pct", 1,
1076 "Datarate overshoot (max) target (%)");
1077 static const arg_def_t buf_sz = ARG_DEF(NULL, "buf-sz", 1,
1078 "Client buffer size (ms)");
1079 static const arg_def_t buf_initial_sz = ARG_DEF(NULL, "buf-initial-sz", 1,
1080 "Client initial buffer size (ms)");
1081 static const arg_def_t buf_optimal_sz = ARG_DEF(NULL, "buf-optimal-sz", 1,
1082 "Client optimal buffer size (ms)");
1083 static const arg_def_t *rc_args[] =
1085 &dropframe_thresh, &resize_allowed, &resize_up_thresh, &resize_down_thresh,
1086 &end_usage, &target_bitrate, &min_quantizer, &max_quantizer,
1087 &undershoot_pct, &overshoot_pct, &buf_sz, &buf_initial_sz, &buf_optimal_sz,
1088 NULL
1092 static const arg_def_t bias_pct = ARG_DEF(NULL, "bias-pct", 1,
1093 "CBR/VBR bias (0=CBR, 100=VBR)");
1094 static const arg_def_t minsection_pct = ARG_DEF(NULL, "minsection-pct", 1,
1095 "GOP min bitrate (% of target)");
1096 static const arg_def_t maxsection_pct = ARG_DEF(NULL, "maxsection-pct", 1,
1097 "GOP max bitrate (% of target)");
1098 static const arg_def_t *rc_twopass_args[] =
1100 &bias_pct, &minsection_pct, &maxsection_pct, NULL
1104 static const arg_def_t kf_min_dist = ARG_DEF(NULL, "kf-min-dist", 1,
1105 "Minimum keyframe interval (frames)");
1106 static const arg_def_t kf_max_dist = ARG_DEF(NULL, "kf-max-dist", 1,
1107 "Maximum keyframe interval (frames)");
1108 static const arg_def_t kf_disabled = ARG_DEF(NULL, "disable-kf", 0,
1109 "Disable keyframe placement");
1110 static const arg_def_t *kf_args[] =
1112 &kf_min_dist, &kf_max_dist, &kf_disabled, NULL
1116 #if CONFIG_VP8_ENCODER
1117 static const arg_def_t noise_sens = ARG_DEF(NULL, "noise-sensitivity", 1,
1118 "Noise sensitivity (frames to blur)");
1119 static const arg_def_t sharpness = ARG_DEF(NULL, "sharpness", 1,
1120 "Filter sharpness (0-7)");
1121 static const arg_def_t static_thresh = ARG_DEF(NULL, "static-thresh", 1,
1122 "Motion detection threshold");
1123 #endif
1125 #if CONFIG_VP8_ENCODER
1126 static const arg_def_t cpu_used = ARG_DEF(NULL, "cpu-used", 1,
1127 "CPU Used (-16..16)");
1128 #endif
1131 #if CONFIG_VP8_ENCODER
1132 static const arg_def_t token_parts = ARG_DEF(NULL, "token-parts", 1,
1133 "Number of token partitions to use, log2");
1134 static const arg_def_t auto_altref = ARG_DEF(NULL, "auto-alt-ref", 1,
1135 "Enable automatic alt reference frames");
1136 static const arg_def_t arnr_maxframes = ARG_DEF(NULL, "arnr-maxframes", 1,
1137 "AltRef Max Frames");
1138 static const arg_def_t arnr_strength = ARG_DEF(NULL, "arnr-strength", 1,
1139 "AltRef Strength");
1140 static const arg_def_t arnr_type = ARG_DEF(NULL, "arnr-type", 1,
1141 "AltRef Type");
1142 static const struct arg_enum_list tuning_enum[] = {
1143 {"psnr", VP8_TUNE_PSNR},
1144 {"ssim", VP8_TUNE_SSIM},
1145 {NULL, 0}
1147 static const arg_def_t tune_ssim = ARG_DEF_ENUM(NULL, "tune", 1,
1148 "Material to favor", tuning_enum);
1149 static const arg_def_t cq_level = ARG_DEF(NULL, "cq-level", 1,
1150 "Constrained Quality Level");
1151 static const arg_def_t max_intra_rate_pct = ARG_DEF(NULL, "max-intra-rate", 1,
1152 "Max I-frame bitrate (pct)");
1154 static const arg_def_t *vp8_args[] =
1156 &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh,
1157 &token_parts, &arnr_maxframes, &arnr_strength, &arnr_type,
1158 &tune_ssim, &cq_level, &max_intra_rate_pct, NULL
1160 static const int vp8_arg_ctrl_map[] =
1162 VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF,
1163 VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD,
1164 VP8E_SET_TOKEN_PARTITIONS,
1165 VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH , VP8E_SET_ARNR_TYPE,
1166 VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, VP8E_SET_MAX_INTRA_BITRATE_PCT, 0
1168 #endif
1170 static const arg_def_t *no_args[] = { NULL };
1172 static void usage_exit()
1174 int i;
1176 fprintf(stderr, "Usage: %s <options> -o dst_filename src_filename \n",
1177 exec_name);
1179 fprintf(stderr, "\nOptions:\n");
1180 arg_show_usage(stdout, main_args);
1181 fprintf(stderr, "\nEncoder Global Options:\n");
1182 arg_show_usage(stdout, global_args);
1183 fprintf(stderr, "\nRate Control Options:\n");
1184 arg_show_usage(stdout, rc_args);
1185 fprintf(stderr, "\nTwopass Rate Control Options:\n");
1186 arg_show_usage(stdout, rc_twopass_args);
1187 fprintf(stderr, "\nKeyframe Placement Options:\n");
1188 arg_show_usage(stdout, kf_args);
1189 #if CONFIG_VP8_ENCODER
1190 fprintf(stderr, "\nVP8 Specific Options:\n");
1191 arg_show_usage(stdout, vp8_args);
1192 #endif
1193 fprintf(stderr, "\nStream timebase (--timebase):\n"
1194 " The desired precision of timestamps in the output, expressed\n"
1195 " in fractional seconds. Default is 1/1000.\n");
1196 fprintf(stderr, "\n"
1197 "Included encoders:\n"
1198 "\n");
1200 for (i = 0; i < sizeof(codecs) / sizeof(codecs[0]); i++)
1201 fprintf(stderr, " %-6s - %s\n",
1202 codecs[i].name,
1203 vpx_codec_iface_name(codecs[i].iface));
1205 exit(EXIT_FAILURE);
1209 #define HIST_BAR_MAX 40
1210 struct hist_bucket
1212 int low, high, count;
1216 static int merge_hist_buckets(struct hist_bucket *bucket,
1217 int *buckets_,
1218 int max_buckets)
1220 int small_bucket = 0, merge_bucket = INT_MAX, big_bucket=0;
1221 int buckets = *buckets_;
1222 int i;
1224 /* Find the extrema for this list of buckets */
1225 big_bucket = small_bucket = 0;
1226 for(i=0; i < buckets; i++)
1228 if(bucket[i].count < bucket[small_bucket].count)
1229 small_bucket = i;
1230 if(bucket[i].count > bucket[big_bucket].count)
1231 big_bucket = i;
1234 /* If we have too many buckets, merge the smallest with an adjacent
1235 * bucket.
1237 while(buckets > max_buckets)
1239 int last_bucket = buckets - 1;
1241 /* merge the small bucket with an adjacent one. */
1242 if(small_bucket == 0)
1243 merge_bucket = 1;
1244 else if(small_bucket == last_bucket)
1245 merge_bucket = last_bucket - 1;
1246 else if(bucket[small_bucket - 1].count < bucket[small_bucket + 1].count)
1247 merge_bucket = small_bucket - 1;
1248 else
1249 merge_bucket = small_bucket + 1;
1251 assert(abs(merge_bucket - small_bucket) <= 1);
1252 assert(small_bucket < buckets);
1253 assert(big_bucket < buckets);
1254 assert(merge_bucket < buckets);
1256 if(merge_bucket < small_bucket)
1258 bucket[merge_bucket].high = bucket[small_bucket].high;
1259 bucket[merge_bucket].count += bucket[small_bucket].count;
1261 else
1263 bucket[small_bucket].high = bucket[merge_bucket].high;
1264 bucket[small_bucket].count += bucket[merge_bucket].count;
1265 merge_bucket = small_bucket;
1268 assert(bucket[merge_bucket].low != bucket[merge_bucket].high);
1270 buckets--;
1272 /* Remove the merge_bucket from the list, and find the new small
1273 * and big buckets while we're at it
1275 big_bucket = small_bucket = 0;
1276 for(i=0; i < buckets; i++)
1278 if(i > merge_bucket)
1279 bucket[i] = bucket[i+1];
1281 if(bucket[i].count < bucket[small_bucket].count)
1282 small_bucket = i;
1283 if(bucket[i].count > bucket[big_bucket].count)
1284 big_bucket = i;
1289 *buckets_ = buckets;
1290 return bucket[big_bucket].count;
1294 static void show_histogram(const struct hist_bucket *bucket,
1295 int buckets,
1296 int total,
1297 int scale)
1299 const char *pat1, *pat2;
1300 int i;
1302 switch((int)(log(bucket[buckets-1].high)/log(10))+1)
1304 case 1:
1305 case 2:
1306 pat1 = "%4d %2s: ";
1307 pat2 = "%4d-%2d: ";
1308 break;
1309 case 3:
1310 pat1 = "%5d %3s: ";
1311 pat2 = "%5d-%3d: ";
1312 break;
1313 case 4:
1314 pat1 = "%6d %4s: ";
1315 pat2 = "%6d-%4d: ";
1316 break;
1317 case 5:
1318 pat1 = "%7d %5s: ";
1319 pat2 = "%7d-%5d: ";
1320 break;
1321 case 6:
1322 pat1 = "%8d %6s: ";
1323 pat2 = "%8d-%6d: ";
1324 break;
1325 case 7:
1326 pat1 = "%9d %7s: ";
1327 pat2 = "%9d-%7d: ";
1328 break;
1329 default:
1330 pat1 = "%12d %10s: ";
1331 pat2 = "%12d-%10d: ";
1332 break;
1335 for(i=0; i<buckets; i++)
1337 int len;
1338 int j;
1339 float pct;
1341 pct = (float)(100.0 * bucket[i].count / total);
1342 len = HIST_BAR_MAX * bucket[i].count / scale;
1343 if(len < 1)
1344 len = 1;
1345 assert(len <= HIST_BAR_MAX);
1347 if(bucket[i].low == bucket[i].high)
1348 fprintf(stderr, pat1, bucket[i].low, "");
1349 else
1350 fprintf(stderr, pat2, bucket[i].low, bucket[i].high);
1352 for(j=0; j<HIST_BAR_MAX; j++)
1353 fprintf(stderr, j<len?"=":" ");
1354 fprintf(stderr, "\t%5d (%6.2f%%)\n",bucket[i].count,pct);
1359 static void show_q_histogram(const int counts[64], int max_buckets)
1361 struct hist_bucket bucket[64];
1362 int buckets = 0;
1363 int total = 0;
1364 int scale;
1365 int i;
1368 for(i=0; i<64; i++)
1370 if(counts[i])
1372 bucket[buckets].low = bucket[buckets].high = i;
1373 bucket[buckets].count = counts[i];
1374 buckets++;
1375 total += counts[i];
1379 fprintf(stderr, "\nQuantizer Selection:\n");
1380 scale = merge_hist_buckets(bucket, &buckets, max_buckets);
1381 show_histogram(bucket, buckets, total, scale);
1385 #define RATE_BINS (100)
1386 struct rate_hist
1388 int64_t *pts;
1389 int *sz;
1390 int samples;
1391 int frames;
1392 struct hist_bucket bucket[RATE_BINS];
1393 int total;
1397 static void init_rate_histogram(struct rate_hist *hist,
1398 const vpx_codec_enc_cfg_t *cfg,
1399 const vpx_rational_t *fps)
1401 int i;
1403 /* Determine the number of samples in the buffer. Use the file's framerate
1404 * to determine the number of frames in rc_buf_sz milliseconds, with an
1405 * adjustment (5/4) to account for alt-refs
1407 hist->samples = cfg->rc_buf_sz * 5 / 4 * fps->num / fps->den / 1000;
1409 /* prevent division by zero */
1410 if (hist->samples == 0)
1411 hist->samples=1;
1413 hist->pts = calloc(hist->samples, sizeof(*hist->pts));
1414 hist->sz = calloc(hist->samples, sizeof(*hist->sz));
1415 for(i=0; i<RATE_BINS; i++)
1417 hist->bucket[i].low = INT_MAX;
1418 hist->bucket[i].high = 0;
1419 hist->bucket[i].count = 0;
1424 static void destroy_rate_histogram(struct rate_hist *hist)
1426 free(hist->pts);
1427 free(hist->sz);
1431 static void update_rate_histogram(struct rate_hist *hist,
1432 const vpx_codec_enc_cfg_t *cfg,
1433 const vpx_codec_cx_pkt_t *pkt)
1435 int i, idx;
1436 int64_t now, then, sum_sz = 0, avg_bitrate;
1438 now = pkt->data.frame.pts * 1000
1439 * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
1441 idx = hist->frames++ % hist->samples;
1442 hist->pts[idx] = now;
1443 hist->sz[idx] = (int)pkt->data.frame.sz;
1445 if(now < cfg->rc_buf_initial_sz)
1446 return;
1448 then = now;
1450 /* Sum the size over the past rc_buf_sz ms */
1451 for(i = hist->frames; i > 0 && hist->frames - i < hist->samples; i--)
1453 int i_idx = (i-1) % hist->samples;
1455 then = hist->pts[i_idx];
1456 if(now - then > cfg->rc_buf_sz)
1457 break;
1458 sum_sz += hist->sz[i_idx];
1461 if (now == then)
1462 return;
1464 avg_bitrate = sum_sz * 8 * 1000 / (now - then);
1465 idx = (int)(avg_bitrate * (RATE_BINS/2) / (cfg->rc_target_bitrate * 1000));
1466 if(idx < 0)
1467 idx = 0;
1468 if(idx > RATE_BINS-1)
1469 idx = RATE_BINS-1;
1470 if(hist->bucket[idx].low > avg_bitrate)
1471 hist->bucket[idx].low = (int)avg_bitrate;
1472 if(hist->bucket[idx].high < avg_bitrate)
1473 hist->bucket[idx].high = (int)avg_bitrate;
1474 hist->bucket[idx].count++;
1475 hist->total++;
1479 static void show_rate_histogram(struct rate_hist *hist,
1480 const vpx_codec_enc_cfg_t *cfg,
1481 int max_buckets)
1483 int i, scale;
1484 int buckets = 0;
1486 for(i = 0; i < RATE_BINS; i++)
1488 if(hist->bucket[i].low == INT_MAX)
1489 continue;
1490 hist->bucket[buckets++] = hist->bucket[i];
1493 fprintf(stderr, "\nRate (over %dms window):\n", cfg->rc_buf_sz);
1494 scale = merge_hist_buckets(hist->bucket, &buckets, max_buckets);
1495 show_histogram(hist->bucket, buckets, hist->total, scale);
1498 #define NELEMENTS(x) (sizeof(x)/sizeof(x[0]))
1499 #define ARG_CTRL_CNT_MAX NELEMENTS(vp8_arg_ctrl_map)
1502 /* Configuration elements common to all streams */
1503 struct global_config
1505 const struct codec_item *codec;
1506 int passes;
1507 int pass;
1508 int usage;
1509 int deadline;
1510 int use_i420;
1511 int quiet;
1512 int verbose;
1513 int limit;
1514 int show_psnr;
1515 int have_framerate;
1516 struct vpx_rational framerate;
1517 int out_part;
1518 int debug;
1519 int show_q_hist_buckets;
1520 int show_rate_hist_buckets;
1524 /* Per-stream configuration */
1525 struct stream_config
1527 struct vpx_codec_enc_cfg cfg;
1528 const char *out_fn;
1529 const char *stats_fn;
1530 stereo_format_t stereo_fmt;
1531 int arg_ctrls[ARG_CTRL_CNT_MAX][2];
1532 int arg_ctrl_cnt;
1533 int write_webm;
1534 int have_kf_max_dist;
1538 struct stream_state
1540 int index;
1541 struct stream_state *next;
1542 struct stream_config config;
1543 FILE *file;
1544 struct rate_hist rate_hist;
1545 EbmlGlobal ebml;
1546 uint32_t hash;
1547 uint64_t psnr_sse_total;
1548 uint64_t psnr_samples_total;
1549 double psnr_totals[4];
1550 int psnr_count;
1551 int counts[64];
1552 vpx_codec_ctx_t encoder;
1553 unsigned int frames_out;
1554 uint64_t cx_time;
1555 size_t nbytes;
1556 stats_io_t stats;
1560 void validate_positive_rational(const char *msg,
1561 struct vpx_rational *rat)
1563 if (rat->den < 0)
1565 rat->num *= -1;
1566 rat->den *= -1;
1569 if (rat->num < 0)
1570 die("Error: %s must be positive\n", msg);
1572 if (!rat->den)
1573 die("Error: %s has zero denominator\n", msg);
1577 static void parse_global_config(struct global_config *global, char **argv)
1579 char **argi, **argj;
1580 struct arg arg;
1582 /* Initialize default parameters */
1583 memset(global, 0, sizeof(*global));
1584 global->codec = codecs;
1585 global->passes = 1;
1586 global->use_i420 = 1;
1588 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
1590 arg.argv_step = 1;
1592 if (arg_match(&arg, &codecarg, argi))
1594 int j, k = -1;
1596 for (j = 0; j < sizeof(codecs) / sizeof(codecs[0]); j++)
1597 if (!strcmp(codecs[j].name, arg.val))
1598 k = j;
1600 if (k >= 0)
1601 global->codec = codecs + k;
1602 else
1603 die("Error: Unrecognized argument (%s) to --codec\n",
1604 arg.val);
1607 else if (arg_match(&arg, &passes, argi))
1609 global->passes = arg_parse_uint(&arg);
1611 if (global->passes < 1 || global->passes > 2)
1612 die("Error: Invalid number of passes (%d)\n", global->passes);
1614 else if (arg_match(&arg, &pass_arg, argi))
1616 global->pass = arg_parse_uint(&arg);
1618 if (global->pass < 1 || global->pass > 2)
1619 die("Error: Invalid pass selected (%d)\n",
1620 global->pass);
1622 else if (arg_match(&arg, &usage, argi))
1623 global->usage = arg_parse_uint(&arg);
1624 else if (arg_match(&arg, &deadline, argi))
1625 global->deadline = arg_parse_uint(&arg);
1626 else if (arg_match(&arg, &best_dl, argi))
1627 global->deadline = VPX_DL_BEST_QUALITY;
1628 else if (arg_match(&arg, &good_dl, argi))
1629 global->deadline = VPX_DL_GOOD_QUALITY;
1630 else if (arg_match(&arg, &rt_dl, argi))
1631 global->deadline = VPX_DL_REALTIME;
1632 else if (arg_match(&arg, &use_yv12, argi))
1633 global->use_i420 = 0;
1634 else if (arg_match(&arg, &use_i420, argi))
1635 global->use_i420 = 1;
1636 else if (arg_match(&arg, &quietarg, argi))
1637 global->quiet = 1;
1638 else if (arg_match(&arg, &verbosearg, argi))
1639 global->verbose = 1;
1640 else if (arg_match(&arg, &limit, argi))
1641 global->limit = arg_parse_uint(&arg);
1642 else if (arg_match(&arg, &psnrarg, argi))
1643 global->show_psnr = 1;
1644 else if (arg_match(&arg, &framerate, argi))
1646 global->framerate = arg_parse_rational(&arg);
1647 validate_positive_rational(arg.name, &global->framerate);
1648 global->have_framerate = 1;
1650 else if (arg_match(&arg,&out_part, argi))
1651 global->out_part = 1;
1652 else if (arg_match(&arg, &debugmode, argi))
1653 global->debug = 1;
1654 else if (arg_match(&arg, &q_hist_n, argi))
1655 global->show_q_hist_buckets = arg_parse_uint(&arg);
1656 else if (arg_match(&arg, &rate_hist_n, argi))
1657 global->show_rate_hist_buckets = arg_parse_uint(&arg);
1658 else
1659 argj++;
1662 /* Validate global config */
1664 if (global->pass)
1666 /* DWIM: Assume the user meant passes=2 if pass=2 is specified */
1667 if (global->pass > global->passes)
1669 warn("Assuming --pass=%d implies --passes=%d\n",
1670 global->pass, global->pass);
1671 global->passes = global->pass;
1677 void open_input_file(struct input_state *input)
1679 unsigned int fourcc;
1681 /* Parse certain options from the input file, if possible */
1682 input->file = strcmp(input->fn, "-") ? fopen(input->fn, "rb")
1683 : set_binary_mode(stdin);
1685 if (!input->file)
1686 fatal("Failed to open input file");
1688 /* For RAW input sources, these bytes will applied on the first frame
1689 * in read_frame().
1691 input->detect.buf_read = fread(input->detect.buf, 1, 4, input->file);
1692 input->detect.position = 0;
1694 if (input->detect.buf_read == 4
1695 && file_is_y4m(input->file, &input->y4m, input->detect.buf))
1697 if (y4m_input_open(&input->y4m, input->file, input->detect.buf, 4) >= 0)
1699 input->file_type = FILE_TYPE_Y4M;
1700 input->w = input->y4m.pic_w;
1701 input->h = input->y4m.pic_h;
1702 input->framerate.num = input->y4m.fps_n;
1703 input->framerate.den = input->y4m.fps_d;
1704 input->use_i420 = 0;
1706 else
1707 fatal("Unsupported Y4M stream.");
1709 else if (input->detect.buf_read == 4 && file_is_ivf(input, &fourcc))
1711 input->file_type = FILE_TYPE_IVF;
1712 switch (fourcc)
1714 case 0x32315659:
1715 input->use_i420 = 0;
1716 break;
1717 case 0x30323449:
1718 input->use_i420 = 1;
1719 break;
1720 default:
1721 fatal("Unsupported fourcc (%08x) in IVF", fourcc);
1724 else
1726 input->file_type = FILE_TYPE_RAW;
1731 static void close_input_file(struct input_state *input)
1733 fclose(input->file);
1734 if (input->file_type == FILE_TYPE_Y4M)
1735 y4m_input_close(&input->y4m);
1738 static struct stream_state *new_stream(struct global_config *global,
1739 struct stream_state *prev)
1741 struct stream_state *stream;
1743 stream = calloc(1, sizeof(*stream));
1744 if(!stream)
1745 fatal("Failed to allocate new stream.");
1746 if(prev)
1748 memcpy(stream, prev, sizeof(*stream));
1749 stream->index++;
1750 prev->next = stream;
1752 else
1754 vpx_codec_err_t res;
1756 /* Populate encoder configuration */
1757 res = vpx_codec_enc_config_default(global->codec->iface,
1758 &stream->config.cfg,
1759 global->usage);
1760 if (res)
1761 fatal("Failed to get config: %s\n", vpx_codec_err_to_string(res));
1763 /* Change the default timebase to a high enough value so that the
1764 * encoder will always create strictly increasing timestamps.
1766 stream->config.cfg.g_timebase.den = 1000;
1768 /* Never use the library's default resolution, require it be parsed
1769 * from the file or set on the command line.
1771 stream->config.cfg.g_w = 0;
1772 stream->config.cfg.g_h = 0;
1774 /* Initialize remaining stream parameters */
1775 stream->config.stereo_fmt = STEREO_FORMAT_MONO;
1776 stream->config.write_webm = 1;
1777 stream->ebml.last_pts_ms = -1;
1779 /* Allows removal of the application version from the EBML tags */
1780 stream->ebml.debug = global->debug;
1783 /* Output files must be specified for each stream */
1784 stream->config.out_fn = NULL;
1786 stream->next = NULL;
1787 return stream;
1791 static int parse_stream_params(struct global_config *global,
1792 struct stream_state *stream,
1793 char **argv)
1795 char **argi, **argj;
1796 struct arg arg;
1797 static const arg_def_t **ctrl_args = no_args;
1798 static const int *ctrl_args_map = NULL;
1799 struct stream_config *config = &stream->config;
1800 int eos_mark_found = 0;
1802 /* Handle codec specific options */
1803 if (global->codec->iface == &vpx_codec_vp8_cx_algo)
1805 ctrl_args = vp8_args;
1806 ctrl_args_map = vp8_arg_ctrl_map;
1809 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
1811 arg.argv_step = 1;
1813 /* Once we've found an end-of-stream marker (--) we want to continue
1814 * shifting arguments but not consuming them.
1816 if (eos_mark_found)
1818 argj++;
1819 continue;
1821 else if (!strcmp(*argj, "--"))
1823 eos_mark_found = 1;
1824 continue;
1827 if (0);
1828 else if (arg_match(&arg, &outputfile, argi))
1829 config->out_fn = arg.val;
1830 else if (arg_match(&arg, &fpf_name, argi))
1831 config->stats_fn = arg.val;
1832 else if (arg_match(&arg, &use_ivf, argi))
1833 config->write_webm = 0;
1834 else if (arg_match(&arg, &threads, argi))
1835 config->cfg.g_threads = arg_parse_uint(&arg);
1836 else if (arg_match(&arg, &profile, argi))
1837 config->cfg.g_profile = arg_parse_uint(&arg);
1838 else if (arg_match(&arg, &width, argi))
1839 config->cfg.g_w = arg_parse_uint(&arg);
1840 else if (arg_match(&arg, &height, argi))
1841 config->cfg.g_h = arg_parse_uint(&arg);
1842 else if (arg_match(&arg, &stereo_mode, argi))
1843 config->stereo_fmt = arg_parse_enum_or_int(&arg);
1844 else if (arg_match(&arg, &timebase, argi))
1846 config->cfg.g_timebase = arg_parse_rational(&arg);
1847 validate_positive_rational(arg.name, &config->cfg.g_timebase);
1849 else if (arg_match(&arg, &error_resilient, argi))
1850 config->cfg.g_error_resilient = arg_parse_uint(&arg);
1851 else if (arg_match(&arg, &lag_in_frames, argi))
1852 config->cfg.g_lag_in_frames = arg_parse_uint(&arg);
1853 else if (arg_match(&arg, &dropframe_thresh, argi))
1854 config->cfg.rc_dropframe_thresh = arg_parse_uint(&arg);
1855 else if (arg_match(&arg, &resize_allowed, argi))
1856 config->cfg.rc_resize_allowed = arg_parse_uint(&arg);
1857 else if (arg_match(&arg, &resize_up_thresh, argi))
1858 config->cfg.rc_resize_up_thresh = arg_parse_uint(&arg);
1859 else if (arg_match(&arg, &resize_down_thresh, argi))
1860 config->cfg.rc_resize_down_thresh = arg_parse_uint(&arg);
1861 else if (arg_match(&arg, &end_usage, argi))
1862 config->cfg.rc_end_usage = arg_parse_enum_or_int(&arg);
1863 else if (arg_match(&arg, &target_bitrate, argi))
1864 config->cfg.rc_target_bitrate = arg_parse_uint(&arg);
1865 else if (arg_match(&arg, &min_quantizer, argi))
1866 config->cfg.rc_min_quantizer = arg_parse_uint(&arg);
1867 else if (arg_match(&arg, &max_quantizer, argi))
1868 config->cfg.rc_max_quantizer = arg_parse_uint(&arg);
1869 else if (arg_match(&arg, &undershoot_pct, argi))
1870 config->cfg.rc_undershoot_pct = arg_parse_uint(&arg);
1871 else if (arg_match(&arg, &overshoot_pct, argi))
1872 config->cfg.rc_overshoot_pct = arg_parse_uint(&arg);
1873 else if (arg_match(&arg, &buf_sz, argi))
1874 config->cfg.rc_buf_sz = arg_parse_uint(&arg);
1875 else if (arg_match(&arg, &buf_initial_sz, argi))
1876 config->cfg.rc_buf_initial_sz = arg_parse_uint(&arg);
1877 else if (arg_match(&arg, &buf_optimal_sz, argi))
1878 config->cfg.rc_buf_optimal_sz = arg_parse_uint(&arg);
1879 else if (arg_match(&arg, &bias_pct, argi))
1881 config->cfg.rc_2pass_vbr_bias_pct = arg_parse_uint(&arg);
1883 if (global->passes < 2)
1884 warn("option %s ignored in one-pass mode.\n", arg.name);
1886 else if (arg_match(&arg, &minsection_pct, argi))
1888 config->cfg.rc_2pass_vbr_minsection_pct = arg_parse_uint(&arg);
1890 if (global->passes < 2)
1891 warn("option %s ignored in one-pass mode.\n", arg.name);
1893 else if (arg_match(&arg, &maxsection_pct, argi))
1895 config->cfg.rc_2pass_vbr_maxsection_pct = arg_parse_uint(&arg);
1897 if (global->passes < 2)
1898 warn("option %s ignored in one-pass mode.\n", arg.name);
1900 else if (arg_match(&arg, &kf_min_dist, argi))
1901 config->cfg.kf_min_dist = arg_parse_uint(&arg);
1902 else if (arg_match(&arg, &kf_max_dist, argi))
1904 config->cfg.kf_max_dist = arg_parse_uint(&arg);
1905 config->have_kf_max_dist = 1;
1907 else if (arg_match(&arg, &kf_disabled, argi))
1908 config->cfg.kf_mode = VPX_KF_DISABLED;
1909 else
1911 int i, match = 0;
1913 for (i = 0; ctrl_args[i]; i++)
1915 if (arg_match(&arg, ctrl_args[i], argi))
1917 int j;
1918 match = 1;
1920 /* Point either to the next free element or the first
1921 * instance of this control.
1923 for(j=0; j<config->arg_ctrl_cnt; j++)
1924 if(config->arg_ctrls[j][0] == ctrl_args_map[i])
1925 break;
1927 /* Update/insert */
1928 assert(j < ARG_CTRL_CNT_MAX);
1929 if (j < ARG_CTRL_CNT_MAX)
1931 config->arg_ctrls[j][0] = ctrl_args_map[i];
1932 config->arg_ctrls[j][1] = arg_parse_enum_or_int(&arg);
1933 if(j == config->arg_ctrl_cnt)
1934 config->arg_ctrl_cnt++;
1940 if (!match)
1941 argj++;
1945 return eos_mark_found;
1949 #define FOREACH_STREAM(func)\
1952 struct stream_state *stream;\
1954 for(stream = streams; stream; stream = stream->next)\
1955 func;\
1956 }while(0)
1959 static void validate_stream_config(struct stream_state *stream)
1961 struct stream_state *streami;
1963 if(!stream->config.cfg.g_w || !stream->config.cfg.g_h)
1964 fatal("Stream %d: Specify stream dimensions with --width (-w) "
1965 " and --height (-h)", stream->index);
1967 for(streami = stream; streami; streami = streami->next)
1969 /* All streams require output files */
1970 if(!streami->config.out_fn)
1971 fatal("Stream %d: Output file is required (specify with -o)",
1972 streami->index);
1974 /* Check for two streams outputting to the same file */
1975 if(streami != stream)
1977 const char *a = stream->config.out_fn;
1978 const char *b = streami->config.out_fn;
1979 if(!strcmp(a,b) && strcmp(a, "/dev/null") && strcmp(a, ":nul"))
1980 fatal("Stream %d: duplicate output file (from stream %d)",
1981 streami->index, stream->index);
1984 /* Check for two streams sharing a stats file. */
1985 if(streami != stream)
1987 const char *a = stream->config.stats_fn;
1988 const char *b = streami->config.stats_fn;
1989 if(a && b && !strcmp(a,b))
1990 fatal("Stream %d: duplicate stats file (from stream %d)",
1991 streami->index, stream->index);
1997 static void set_stream_dimensions(struct stream_state *stream,
1998 unsigned int w,
1999 unsigned int h)
2001 if ((stream->config.cfg.g_w && stream->config.cfg.g_w != w)
2002 ||(stream->config.cfg.g_h && stream->config.cfg.g_h != h))
2003 fatal("Stream %d: Resizing not yet supported", stream->index);
2004 stream->config.cfg.g_w = w;
2005 stream->config.cfg.g_h = h;
2009 static void set_default_kf_interval(struct stream_state *stream,
2010 struct global_config *global)
2012 /* Use a max keyframe interval of 5 seconds, if none was
2013 * specified on the command line.
2015 if (!stream->config.have_kf_max_dist)
2017 double framerate = (double)global->framerate.num/global->framerate.den;
2018 if (framerate > 0.0)
2019 stream->config.cfg.kf_max_dist = (unsigned int)(5.0*framerate);
2024 static void show_stream_config(struct stream_state *stream,
2025 struct global_config *global,
2026 struct input_state *input)
2029 #define SHOW(field) \
2030 fprintf(stderr, " %-28s = %d\n", #field, stream->config.cfg.field)
2032 if(stream->index == 0)
2034 fprintf(stderr, "Codec: %s\n",
2035 vpx_codec_iface_name(global->codec->iface));
2036 fprintf(stderr, "Source file: %s Format: %s\n", input->fn,
2037 input->use_i420 ? "I420" : "YV12");
2039 if(stream->next || stream->index)
2040 fprintf(stderr, "\nStream Index: %d\n", stream->index);
2041 fprintf(stderr, "Destination file: %s\n", stream->config.out_fn);
2042 fprintf(stderr, "Encoder parameters:\n");
2044 SHOW(g_usage);
2045 SHOW(g_threads);
2046 SHOW(g_profile);
2047 SHOW(g_w);
2048 SHOW(g_h);
2049 SHOW(g_timebase.num);
2050 SHOW(g_timebase.den);
2051 SHOW(g_error_resilient);
2052 SHOW(g_pass);
2053 SHOW(g_lag_in_frames);
2054 SHOW(rc_dropframe_thresh);
2055 SHOW(rc_resize_allowed);
2056 SHOW(rc_resize_up_thresh);
2057 SHOW(rc_resize_down_thresh);
2058 SHOW(rc_end_usage);
2059 SHOW(rc_target_bitrate);
2060 SHOW(rc_min_quantizer);
2061 SHOW(rc_max_quantizer);
2062 SHOW(rc_undershoot_pct);
2063 SHOW(rc_overshoot_pct);
2064 SHOW(rc_buf_sz);
2065 SHOW(rc_buf_initial_sz);
2066 SHOW(rc_buf_optimal_sz);
2067 SHOW(rc_2pass_vbr_bias_pct);
2068 SHOW(rc_2pass_vbr_minsection_pct);
2069 SHOW(rc_2pass_vbr_maxsection_pct);
2070 SHOW(kf_mode);
2071 SHOW(kf_min_dist);
2072 SHOW(kf_max_dist);
2076 static void open_output_file(struct stream_state *stream,
2077 struct global_config *global)
2079 const char *fn = stream->config.out_fn;
2081 stream->file = strcmp(fn, "-") ? fopen(fn, "wb") : set_binary_mode(stdout);
2083 if (!stream->file)
2084 fatal("Failed to open output file");
2086 if(stream->config.write_webm && fseek(stream->file, 0, SEEK_CUR))
2087 fatal("WebM output to pipes not supported.");
2089 if(stream->config.write_webm)
2091 stream->ebml.stream = stream->file;
2092 write_webm_file_header(&stream->ebml, &stream->config.cfg,
2093 &global->framerate,
2094 stream->config.stereo_fmt);
2096 else
2097 write_ivf_file_header(stream->file, &stream->config.cfg,
2098 global->codec->fourcc, 0);
2102 static void close_output_file(struct stream_state *stream,
2103 unsigned int fourcc)
2105 if(stream->config.write_webm)
2107 write_webm_file_footer(&stream->ebml, stream->hash);
2108 free(stream->ebml.cue_list);
2109 stream->ebml.cue_list = NULL;
2111 else
2113 if (!fseek(stream->file, 0, SEEK_SET))
2114 write_ivf_file_header(stream->file, &stream->config.cfg,
2115 fourcc,
2116 stream->frames_out);
2119 fclose(stream->file);
2123 static void setup_pass(struct stream_state *stream,
2124 struct global_config *global,
2125 int pass)
2127 if (stream->config.stats_fn)
2129 if (!stats_open_file(&stream->stats, stream->config.stats_fn,
2130 pass))
2131 fatal("Failed to open statistics store");
2133 else
2135 if (!stats_open_mem(&stream->stats, pass))
2136 fatal("Failed to open statistics store");
2139 stream->config.cfg.g_pass = global->passes == 2
2140 ? pass ? VPX_RC_LAST_PASS : VPX_RC_FIRST_PASS
2141 : VPX_RC_ONE_PASS;
2142 if (pass)
2143 stream->config.cfg.rc_twopass_stats_in = stats_get(&stream->stats);
2145 stream->cx_time = 0;
2146 stream->nbytes = 0;
2147 stream->frames_out = 0;
2151 static void initialize_encoder(struct stream_state *stream,
2152 struct global_config *global)
2154 int i;
2155 int flags = 0;
2157 flags |= global->show_psnr ? VPX_CODEC_USE_PSNR : 0;
2158 flags |= global->out_part ? VPX_CODEC_USE_OUTPUT_PARTITION : 0;
2160 /* Construct Encoder Context */
2161 vpx_codec_enc_init(&stream->encoder, global->codec->iface,
2162 &stream->config.cfg, flags);
2163 ctx_exit_on_error(&stream->encoder, "Failed to initialize encoder");
2165 /* Note that we bypass the vpx_codec_control wrapper macro because
2166 * we're being clever to store the control IDs in an array. Real
2167 * applications will want to make use of the enumerations directly
2169 for (i = 0; i < stream->config.arg_ctrl_cnt; i++)
2171 int ctrl = stream->config.arg_ctrls[i][0];
2172 int value = stream->config.arg_ctrls[i][1];
2173 if (vpx_codec_control_(&stream->encoder, ctrl, value))
2174 fprintf(stderr, "Error: Tried to set control %d = %d\n",
2175 ctrl, value);
2177 ctx_exit_on_error(&stream->encoder, "Failed to control codec");
2182 static void encode_frame(struct stream_state *stream,
2183 struct global_config *global,
2184 struct vpx_image *img,
2185 unsigned int frames_in)
2187 vpx_codec_pts_t frame_start, next_frame_start;
2188 struct vpx_codec_enc_cfg *cfg = &stream->config.cfg;
2189 struct vpx_usec_timer timer;
2191 frame_start = (cfg->g_timebase.den * (int64_t)(frames_in - 1)
2192 * global->framerate.den)
2193 / cfg->g_timebase.num / global->framerate.num;
2194 next_frame_start = (cfg->g_timebase.den * (int64_t)(frames_in)
2195 * global->framerate.den)
2196 / cfg->g_timebase.num / global->framerate.num;
2197 vpx_usec_timer_start(&timer);
2198 vpx_codec_encode(&stream->encoder, img, frame_start,
2199 (unsigned long)(next_frame_start - frame_start),
2200 0, global->deadline);
2201 vpx_usec_timer_mark(&timer);
2202 stream->cx_time += vpx_usec_timer_elapsed(&timer);
2203 ctx_exit_on_error(&stream->encoder, "Stream %d: Failed to encode frame",
2204 stream->index);
2208 static void update_quantizer_histogram(struct stream_state *stream)
2210 if(stream->config.cfg.g_pass != VPX_RC_FIRST_PASS)
2212 int q;
2214 vpx_codec_control(&stream->encoder, VP8E_GET_LAST_QUANTIZER_64, &q);
2215 ctx_exit_on_error(&stream->encoder, "Failed to read quantizer");
2216 stream->counts[q]++;
2221 static void get_cx_data(struct stream_state *stream,
2222 struct global_config *global,
2223 int *got_data)
2225 const vpx_codec_cx_pkt_t *pkt;
2226 const struct vpx_codec_enc_cfg *cfg = &stream->config.cfg;
2227 vpx_codec_iter_t iter = NULL;
2229 while ((pkt = vpx_codec_get_cx_data(&stream->encoder, &iter)))
2231 static size_t fsize = 0;
2232 static off_t ivf_header_pos = 0;
2234 *got_data = 1;
2236 switch (pkt->kind)
2238 case VPX_CODEC_CX_FRAME_PKT:
2239 if (!(pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT))
2241 stream->frames_out++;
2243 if (!global->quiet)
2244 fprintf(stderr, " %6luF",
2245 (unsigned long)pkt->data.frame.sz);
2247 update_rate_histogram(&stream->rate_hist, cfg, pkt);
2248 if(stream->config.write_webm)
2250 /* Update the hash */
2251 if(!stream->ebml.debug)
2252 stream->hash = murmur(pkt->data.frame.buf,
2253 (int)pkt->data.frame.sz,
2254 stream->hash);
2256 write_webm_block(&stream->ebml, cfg, pkt);
2258 else
2260 if (pkt->data.frame.partition_id <= 0)
2262 ivf_header_pos = ftello(stream->file);
2263 fsize = pkt->data.frame.sz;
2265 write_ivf_frame_header(stream->file, pkt);
2267 else
2269 fsize += pkt->data.frame.sz;
2271 if (!(pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT))
2273 off_t currpos = ftello(stream->file);
2274 fseeko(stream->file, ivf_header_pos, SEEK_SET);
2275 write_ivf_frame_size(stream->file, fsize);
2276 fseeko(stream->file, currpos, SEEK_SET);
2280 (void) fwrite(pkt->data.frame.buf, 1, pkt->data.frame.sz,
2281 stream->file);
2283 stream->nbytes += pkt->data.raw.sz;
2284 break;
2285 case VPX_CODEC_STATS_PKT:
2286 stream->frames_out++;
2287 if (!global->quiet)
2288 fprintf(stderr, " %6luS",
2289 (unsigned long)pkt->data.twopass_stats.sz);
2290 stats_write(&stream->stats,
2291 pkt->data.twopass_stats.buf,
2292 pkt->data.twopass_stats.sz);
2293 stream->nbytes += pkt->data.raw.sz;
2294 break;
2295 case VPX_CODEC_PSNR_PKT:
2297 if (global->show_psnr)
2299 int i;
2301 stream->psnr_sse_total += pkt->data.psnr.sse[0];
2302 stream->psnr_samples_total += pkt->data.psnr.samples[0];
2303 for (i = 0; i < 4; i++)
2305 if (!global->quiet)
2306 fprintf(stderr, "%.3f ", pkt->data.psnr.psnr[i]);
2307 stream->psnr_totals[i] += pkt->data.psnr.psnr[i];
2309 stream->psnr_count++;
2312 break;
2313 default:
2314 break;
2320 static void show_psnr(struct stream_state *stream)
2322 int i;
2323 double ovpsnr;
2325 if (!stream->psnr_count)
2326 return;
2328 fprintf(stderr, "Stream %d PSNR (Overall/Avg/Y/U/V)", stream->index);
2329 ovpsnr = vp8_mse2psnr((double)stream->psnr_samples_total, 255.0,
2330 (double)stream->psnr_sse_total);
2331 fprintf(stderr, " %.3f", ovpsnr);
2333 for (i = 0; i < 4; i++)
2335 fprintf(stderr, " %.3f", stream->psnr_totals[i]/stream->psnr_count);
2337 fprintf(stderr, "\n");
2341 float usec_to_fps(uint64_t usec, unsigned int frames)
2343 return (float)(usec > 0 ? frames * 1000000.0 / (float)usec : 0);
2347 int main(int argc, const char **argv_)
2349 int pass;
2350 vpx_image_t raw;
2351 int frame_avail, got_data;
2353 struct input_state input = {0};
2354 struct global_config global;
2355 struct stream_state *streams = NULL;
2356 char **argv, **argi;
2357 unsigned long cx_time = 0;
2358 int stream_cnt = 0;
2360 exec_name = argv_[0];
2362 if (argc < 3)
2363 usage_exit();
2365 /* Setup default input stream settings */
2366 input.framerate.num = 30;
2367 input.framerate.den = 1;
2368 input.use_i420 = 1;
2370 /* First parse the global configuration values, because we want to apply
2371 * other parameters on top of the default configuration provided by the
2372 * codec.
2374 argv = argv_dup(argc - 1, argv_ + 1);
2375 parse_global_config(&global, argv);
2378 /* Now parse each stream's parameters. Using a local scope here
2379 * due to the use of 'stream' as loop variable in FOREACH_STREAM
2380 * loops
2382 struct stream_state *stream = NULL;
2386 stream = new_stream(&global, stream);
2387 stream_cnt++;
2388 if(!streams)
2389 streams = stream;
2390 } while(parse_stream_params(&global, stream, argv));
2393 /* Check for unrecognized options */
2394 for (argi = argv; *argi; argi++)
2395 if (argi[0][0] == '-' && argi[0][1])
2396 die("Error: Unrecognized option %s\n", *argi);
2398 /* Handle non-option arguments */
2399 input.fn = argv[0];
2401 if (!input.fn)
2402 usage_exit();
2404 for (pass = global.pass ? global.pass - 1 : 0; pass < global.passes; pass++)
2406 int frames_in = 0;
2408 open_input_file(&input);
2410 /* If the input file doesn't specify its w/h (raw files), try to get
2411 * the data from the first stream's configuration.
2413 if(!input.w || !input.h)
2414 FOREACH_STREAM({
2415 if(stream->config.cfg.g_w && stream->config.cfg.g_h)
2417 input.w = stream->config.cfg.g_w;
2418 input.h = stream->config.cfg.g_h;
2419 break;
2423 /* Update stream configurations from the input file's parameters */
2424 FOREACH_STREAM(set_stream_dimensions(stream, input.w, input.h));
2425 FOREACH_STREAM(validate_stream_config(stream));
2427 /* Ensure that --passes and --pass are consistent. If --pass is set and
2428 * --passes=2, ensure --fpf was set.
2430 if (global.pass && global.passes == 2)
2431 FOREACH_STREAM({
2432 if(!stream->config.stats_fn)
2433 die("Stream %d: Must specify --fpf when --pass=%d"
2434 " and --passes=2\n", stream->index, global.pass);
2438 /* Use the frame rate from the file only if none was specified
2439 * on the command-line.
2441 if (!global.have_framerate)
2442 global.framerate = input.framerate;
2444 FOREACH_STREAM(set_default_kf_interval(stream, &global));
2446 /* Show configuration */
2447 if (global.verbose && pass == 0)
2448 FOREACH_STREAM(show_stream_config(stream, &global, &input));
2450 if(pass == (global.pass ? global.pass - 1 : 0)) {
2451 if (input.file_type == FILE_TYPE_Y4M)
2452 /*The Y4M reader does its own allocation.
2453 Just initialize this here to avoid problems if we never read any
2454 frames.*/
2455 memset(&raw, 0, sizeof(raw));
2456 else
2457 vpx_img_alloc(&raw,
2458 input.use_i420 ? VPX_IMG_FMT_I420
2459 : VPX_IMG_FMT_YV12,
2460 input.w, input.h, 32);
2462 FOREACH_STREAM(init_rate_histogram(&stream->rate_hist,
2463 &stream->config.cfg,
2464 &global.framerate));
2467 FOREACH_STREAM(open_output_file(stream, &global));
2468 FOREACH_STREAM(setup_pass(stream, &global, pass));
2469 FOREACH_STREAM(initialize_encoder(stream, &global));
2471 frame_avail = 1;
2472 got_data = 0;
2474 while (frame_avail || got_data)
2476 struct vpx_usec_timer timer;
2478 if (!global.limit || frames_in < global.limit)
2480 frame_avail = read_frame(&input, &raw);
2482 if (frame_avail)
2483 frames_in++;
2485 if (!global.quiet)
2487 if(stream_cnt == 1)
2488 fprintf(stderr,
2489 "\rPass %d/%d frame %4d/%-4d %7"PRId64"B \033[K",
2490 pass + 1, global.passes, frames_in,
2491 streams->frames_out, (int64_t)streams->nbytes);
2492 else
2493 fprintf(stderr,
2494 "\rPass %d/%d frame %4d %7lu %s (%.2f fps)\033[K",
2495 pass + 1, global.passes, frames_in,
2496 cx_time > 9999999 ? cx_time / 1000 : cx_time,
2497 cx_time > 9999999 ? "ms" : "us",
2498 usec_to_fps(cx_time, frames_in));
2502 else
2503 frame_avail = 0;
2505 vpx_usec_timer_start(&timer);
2506 FOREACH_STREAM(encode_frame(stream, &global,
2507 frame_avail ? &raw : NULL,
2508 frames_in));
2509 vpx_usec_timer_mark(&timer);
2510 cx_time += (unsigned long)vpx_usec_timer_elapsed(&timer);
2512 FOREACH_STREAM(update_quantizer_histogram(stream));
2514 got_data = 0;
2515 FOREACH_STREAM(get_cx_data(stream, &global, &got_data));
2517 fflush(stdout);
2520 if(stream_cnt > 1)
2521 fprintf(stderr, "\n");
2523 if (!global.quiet)
2524 FOREACH_STREAM(fprintf(
2525 stderr,
2526 "\rPass %d/%d frame %4d/%-4d %7"PRId64"B %7lub/f %7"PRId64"b/s"
2527 " %7"PRId64" %s (%.2f fps)\033[K\n", pass + 1,
2528 global.passes, frames_in, stream->frames_out, (int64_t)stream->nbytes,
2529 frames_in ? (unsigned long)(stream->nbytes * 8 / frames_in) : 0,
2530 frames_in ? (int64_t)stream->nbytes * 8
2531 * (int64_t)global.framerate.num / global.framerate.den
2532 / frames_in
2533 : 0,
2534 stream->cx_time > 9999999 ? stream->cx_time / 1000 : stream->cx_time,
2535 stream->cx_time > 9999999 ? "ms" : "us",
2536 usec_to_fps(stream->cx_time, frames_in));
2539 if (global.show_psnr)
2540 FOREACH_STREAM(show_psnr(stream));
2542 FOREACH_STREAM(vpx_codec_destroy(&stream->encoder));
2544 close_input_file(&input);
2546 FOREACH_STREAM(close_output_file(stream, global.codec->fourcc));
2548 FOREACH_STREAM(stats_close(&stream->stats, global.passes-1));
2550 if (global.pass)
2551 break;
2554 if (global.show_q_hist_buckets)
2555 FOREACH_STREAM(show_q_histogram(stream->counts,
2556 global.show_q_hist_buckets));
2558 if (global.show_rate_hist_buckets)
2559 FOREACH_STREAM(show_rate_histogram(&stream->rate_hist,
2560 &stream->config.cfg,
2561 global.show_rate_hist_buckets));
2562 FOREACH_STREAM(destroy_rate_histogram(&stream->rate_hist));
2564 vpx_img_free(&raw);
2565 free(argv);
2566 free(streams);
2567 return EXIT_SUCCESS;