vpxenc

00001 /*
00002  *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
00003  *
00004  *  Use of this source code is governed by a BSD-style license
00005  *  that can be found in the LICENSE file in the root of the source
00006  *  tree. An additional intellectual property rights grant can be found
00007  *  in the file PATENTS.  All contributing project authors may
00008  *  be found in the AUTHORS file in the root of the source tree.
00009  */
00010 
00011 
00012 /* This is a simple program that encodes YV12 files and generates ivf
00013  * files using the new interface.
00014  */
00015 #if defined(_WIN32) || !CONFIG_OS_SUPPORT
00016 #define USE_POSIX_MMAP 0
00017 #else
00018 #define USE_POSIX_MMAP 1
00019 #endif
00020 
00021 #include <stdio.h>
00022 #include <stdlib.h>
00023 #include <stdarg.h>
00024 #include <string.h>
00025 #include <limits.h>
00026 #include <assert.h>
00027 #include "vpx/vpx_encoder.h"
00028 #if USE_POSIX_MMAP
00029 #include <sys/types.h>
00030 #include <sys/stat.h>
00031 #include <sys/mman.h>
00032 #include <fcntl.h>
00033 #include <unistd.h>
00034 #endif
00035 #include "vpx/vp8cx.h"
00036 #include "vpx_ports/mem_ops.h"
00037 #include "vpx_ports/vpx_timer.h"
00038 #include "tools_common.h"
00039 #include "y4minput.h"
00040 #include "libmkv/EbmlWriter.h"
00041 #include "libmkv/EbmlIDs.h"
00042 
00043 /* Need special handling of these functions on Windows */
00044 #if defined(_MSC_VER)
00045 /* MSVS doesn't define off_t, and uses _f{seek,tell}i64 */
00046 typedef __int64 off_t;
00047 #define fseeko _fseeki64
00048 #define ftello _ftelli64
00049 #elif defined(_WIN32)
00050 /* MinGW defines off_t as long
00051    and uses f{seek,tell}o64/off64_t for large files */
00052 #define fseeko fseeko64
00053 #define ftello ftello64
00054 #define off_t off64_t
00055 #endif
00056 
00057 #if defined(_MSC_VER)
00058 #define LITERALU64(n) n
00059 #else
00060 #define LITERALU64(n) n##LLU
00061 #endif
00062 
00063 /* We should use 32-bit file operations in WebM file format
00064  * when building ARM executable file (.axf) with RVCT */
00065 #if !CONFIG_OS_SUPPORT
00066 typedef long off_t;
00067 #define fseeko fseek
00068 #define ftello ftell
00069 #endif
00070 
00071 static const char *exec_name;
00072 
00073 static const struct codec_item
00074 {
00075     char const              *name;
00076     const vpx_codec_iface_t *iface;
00077     unsigned int             fourcc;
00078 } codecs[] =
00079 {
00080 #if CONFIG_VP8_ENCODER
00081     {"vp8",  &vpx_codec_vp8_cx_algo, 0x30385056},
00082 #endif
00083 };
00084 
00085 static void usage_exit();
00086 
00087 void die(const char *fmt, ...)
00088 {
00089     va_list ap;
00090     va_start(ap, fmt);
00091     vfprintf(stderr, fmt, ap);
00092     fprintf(stderr, "\n");
00093     usage_exit();
00094 }
00095 
00096 static void ctx_exit_on_error(vpx_codec_ctx_t *ctx, const char *s)
00097 {
00098     if (ctx->err)
00099     {
00100         const char *detail = vpx_codec_error_detail(ctx);
00101 
00102         fprintf(stderr, "%s: %s\n", s, vpx_codec_error(ctx));
00103 
00104         if (detail)
00105             fprintf(stderr, "    %s\n", detail);
00106 
00107         exit(EXIT_FAILURE);
00108     }
00109 }
00110 
00111 /* This structure is used to abstract the different ways of handling
00112  * first pass statistics.
00113  */
00114 typedef struct
00115 {
00116     vpx_fixed_buf_t buf;
00117     int             pass;
00118     FILE           *file;
00119     char           *buf_ptr;
00120     size_t          buf_alloc_sz;
00121 } stats_io_t;
00122 
00123 int stats_open_file(stats_io_t *stats, const char *fpf, int pass)
00124 {
00125     int res;
00126 
00127     stats->pass = pass;
00128 
00129     if (pass == 0)
00130     {
00131         stats->file = fopen(fpf, "wb");
00132         stats->buf.sz = 0;
00133         stats->buf.buf = NULL,
00134                    res = (stats->file != NULL);
00135     }
00136     else
00137     {
00138 #if 0
00139 #elif USE_POSIX_MMAP
00140         struct stat stat_buf;
00141         int fd;
00142 
00143         fd = open(fpf, O_RDONLY);
00144         stats->file = fdopen(fd, "rb");
00145         fstat(fd, &stat_buf);
00146         stats->buf.sz = stat_buf.st_size;
00147         stats->buf.buf = mmap(NULL, stats->buf.sz, PROT_READ, MAP_PRIVATE,
00148                               fd, 0);
00149         res = (stats->buf.buf != NULL);
00150 #else
00151         size_t nbytes;
00152 
00153         stats->file = fopen(fpf, "rb");
00154 
00155         if (fseek(stats->file, 0, SEEK_END))
00156         {
00157             fprintf(stderr, "First-pass stats file must be seekable!\n");
00158             exit(EXIT_FAILURE);
00159         }
00160 
00161         stats->buf.sz = stats->buf_alloc_sz = ftell(stats->file);
00162         rewind(stats->file);
00163 
00164         stats->buf.buf = malloc(stats->buf_alloc_sz);
00165 
00166         if (!stats->buf.buf)
00167         {
00168             fprintf(stderr, "Failed to allocate first-pass stats buffer (%lu bytes)\n",
00169                     (unsigned long)stats->buf_alloc_sz);
00170             exit(EXIT_FAILURE);
00171         }
00172 
00173         nbytes = fread(stats->buf.buf, 1, stats->buf.sz, stats->file);
00174         res = (nbytes == stats->buf.sz);
00175 #endif
00176     }
00177 
00178     return res;
00179 }
00180 
00181 int stats_open_mem(stats_io_t *stats, int pass)
00182 {
00183     int res;
00184     stats->pass = pass;
00185 
00186     if (!pass)
00187     {
00188         stats->buf.sz = 0;
00189         stats->buf_alloc_sz = 64 * 1024;
00190         stats->buf.buf = malloc(stats->buf_alloc_sz);
00191     }
00192 
00193     stats->buf_ptr = stats->buf.buf;
00194     res = (stats->buf.buf != NULL);
00195     return res;
00196 }
00197 
00198 
00199 void stats_close(stats_io_t *stats, int last_pass)
00200 {
00201     if (stats->file)
00202     {
00203         if (stats->pass == last_pass)
00204         {
00205 #if 0
00206 #elif USE_POSIX_MMAP
00207             munmap(stats->buf.buf, stats->buf.sz);
00208 #else
00209             free(stats->buf.buf);
00210 #endif
00211         }
00212 
00213         fclose(stats->file);
00214         stats->file = NULL;
00215     }
00216     else
00217     {
00218         if (stats->pass == last_pass)
00219             free(stats->buf.buf);
00220     }
00221 }
00222 
00223 void stats_write(stats_io_t *stats, const void *pkt, size_t len)
00224 {
00225     if (stats->file)
00226     {
00227         if(fwrite(pkt, 1, len, stats->file));
00228     }
00229     else
00230     {
00231         if (stats->buf.sz + len > stats->buf_alloc_sz)
00232         {
00233             size_t  new_sz = stats->buf_alloc_sz + 64 * 1024;
00234             char   *new_ptr = realloc(stats->buf.buf, new_sz);
00235 
00236             if (new_ptr)
00237             {
00238                 stats->buf_ptr = new_ptr + (stats->buf_ptr - (char *)stats->buf.buf);
00239                 stats->buf.buf = new_ptr;
00240                 stats->buf_alloc_sz = new_sz;
00241             }
00242             else
00243             {
00244                 fprintf(stderr,
00245                         "\nFailed to realloc firstpass stats buffer.\n");
00246                 exit(EXIT_FAILURE);
00247             }
00248         }
00249 
00250         memcpy(stats->buf_ptr, pkt, len);
00251         stats->buf.sz += len;
00252         stats->buf_ptr += len;
00253     }
00254 }
00255 
00256 vpx_fixed_buf_t stats_get(stats_io_t *stats)
00257 {
00258     return stats->buf;
00259 }
00260 
00261 /* Stereo 3D packed frame format */
00262 typedef enum stereo_format
00263 {
00264     STEREO_FORMAT_MONO       = 0,
00265     STEREO_FORMAT_LEFT_RIGHT = 1,
00266     STEREO_FORMAT_BOTTOM_TOP = 2,
00267     STEREO_FORMAT_TOP_BOTTOM = 3,
00268     STEREO_FORMAT_RIGHT_LEFT = 11
00269 } stereo_format_t;
00270 
00271 enum video_file_type
00272 {
00273     FILE_TYPE_RAW,
00274     FILE_TYPE_IVF,
00275     FILE_TYPE_Y4M
00276 };
00277 
00278 struct detect_buffer {
00279     char buf[4];
00280     size_t buf_read;
00281     size_t position;
00282 };
00283 
00284 
00285 #define IVF_FRAME_HDR_SZ (4+8) /* 4 byte size + 8 byte timestamp */
00286 static int read_frame(FILE *f, vpx_image_t *img, unsigned int file_type,
00287                       y4m_input *y4m, struct detect_buffer *detect)
00288 {
00289     int plane = 0;
00290     int shortread = 0;
00291 
00292     if (file_type == FILE_TYPE_Y4M)
00293     {
00294         if (y4m_input_fetch_frame(y4m, f, img) < 1)
00295            return 0;
00296     }
00297     else
00298     {
00299         if (file_type == FILE_TYPE_IVF)
00300         {
00301             char junk[IVF_FRAME_HDR_SZ];
00302 
00303             /* Skip the frame header. We know how big the frame should be. See
00304              * write_ivf_frame_header() for documentation on the frame header
00305              * layout.
00306              */
00307             if(fread(junk, 1, IVF_FRAME_HDR_SZ, f));
00308         }
00309 
00310         for (plane = 0; plane < 3; plane++)
00311         {
00312             unsigned char *ptr;
00313             int w = (plane ? (1 + img->d_w) / 2 : img->d_w);
00314             int h = (plane ? (1 + img->d_h) / 2 : img->d_h);
00315             int r;
00316 
00317             /* Determine the correct plane based on the image format. The for-loop
00318              * always counts in Y,U,V order, but this may not match the order of
00319              * the data on disk.
00320              */
00321             switch (plane)
00322             {
00323             case 1:
00324                 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U];
00325                 break;
00326             case 2:
00327                 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V];
00328                 break;
00329             default:
00330                 ptr = img->planes[plane];
00331             }
00332 
00333             for (r = 0; r < h; r++)
00334             {
00335                 size_t needed = w;
00336                 size_t buf_position = 0;
00337                 const size_t left = detect->buf_read - detect->position;
00338                 if (left > 0)
00339                 {
00340                     const size_t more = (left < needed) ? left : needed;
00341                     memcpy(ptr, detect->buf + detect->position, more);
00342                     buf_position = more;
00343                     needed -= more;
00344                     detect->position += more;
00345                 }
00346                 if (needed > 0)
00347                 {
00348                     shortread |= (fread(ptr + buf_position, 1, needed, f) < needed);
00349                 }
00350 
00351                 ptr += img->stride[plane];
00352             }
00353         }
00354     }
00355 
00356     return !shortread;
00357 }
00358 
00359 
00360 unsigned int file_is_y4m(FILE      *infile,
00361                          y4m_input *y4m,
00362                          char       detect[4])
00363 {
00364     if(memcmp(detect, "YUV4", 4) == 0)
00365     {
00366         return 1;
00367     }
00368     return 0;
00369 }
00370 
00371 #define IVF_FILE_HDR_SZ (32)
00372 unsigned int file_is_ivf(FILE *infile,
00373                          unsigned int *fourcc,
00374                          unsigned int *width,
00375                          unsigned int *height,
00376                          struct detect_buffer *detect)
00377 {
00378     char raw_hdr[IVF_FILE_HDR_SZ];
00379     int is_ivf = 0;
00380 
00381     if(memcmp(detect->buf, "DKIF", 4) != 0)
00382         return 0;
00383 
00384     /* See write_ivf_file_header() for more documentation on the file header
00385      * layout.
00386      */
00387     if (fread(raw_hdr + 4, 1, IVF_FILE_HDR_SZ - 4, infile)
00388         == IVF_FILE_HDR_SZ - 4)
00389     {
00390         {
00391             is_ivf = 1;
00392 
00393             if (mem_get_le16(raw_hdr + 4) != 0)
00394                 fprintf(stderr, "Error: Unrecognized IVF version! This file may not"
00395                         " decode properly.");
00396 
00397             *fourcc = mem_get_le32(raw_hdr + 8);
00398         }
00399     }
00400 
00401     if (is_ivf)
00402     {
00403         *width = mem_get_le16(raw_hdr + 12);
00404         *height = mem_get_le16(raw_hdr + 14);
00405         detect->position = 4;
00406     }
00407 
00408     return is_ivf;
00409 }
00410 
00411 
00412 static void write_ivf_file_header(FILE *outfile,
00413                                   const vpx_codec_enc_cfg_t *cfg,
00414                                   unsigned int fourcc,
00415                                   int frame_cnt)
00416 {
00417     char header[32];
00418 
00419     if (cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS)
00420         return;
00421 
00422     header[0] = 'D';
00423     header[1] = 'K';
00424     header[2] = 'I';
00425     header[3] = 'F';
00426     mem_put_le16(header + 4,  0);                 /* version */
00427     mem_put_le16(header + 6,  32);                /* headersize */
00428     mem_put_le32(header + 8,  fourcc);            /* headersize */
00429     mem_put_le16(header + 12, cfg->g_w);          /* width */
00430     mem_put_le16(header + 14, cfg->g_h);          /* height */
00431     mem_put_le32(header + 16, cfg->g_timebase.den); /* rate */
00432     mem_put_le32(header + 20, cfg->g_timebase.num); /* scale */
00433     mem_put_le32(header + 24, frame_cnt);         /* length */
00434     mem_put_le32(header + 28, 0);                 /* unused */
00435 
00436     if(fwrite(header, 1, 32, outfile));
00437 }
00438 
00439 
00440 static void write_ivf_frame_header(FILE *outfile,
00441                                    const vpx_codec_cx_pkt_t *pkt)
00442 {
00443     char             header[12];
00444     vpx_codec_pts_t  pts;
00445 
00446     if (pkt->kind != VPX_CODEC_CX_FRAME_PKT)
00447         return;
00448 
00449     pts = pkt->data.frame.pts;
00450     mem_put_le32(header, pkt->data.frame.sz);
00451     mem_put_le32(header + 4, pts & 0xFFFFFFFF);
00452     mem_put_le32(header + 8, pts >> 32);
00453 
00454     if(fwrite(header, 1, 12, outfile));
00455 }
00456 
00457 
00458 typedef off_t EbmlLoc;
00459 
00460 
00461 struct cue_entry
00462 {
00463     unsigned int time;
00464     uint64_t     loc;
00465 };
00466 
00467 
00468 struct EbmlGlobal
00469 {
00470     int debug;
00471 
00472     FILE    *stream;
00473     int64_t last_pts_ms;
00474     vpx_rational_t  framerate;
00475 
00476     /* These pointers are to the start of an element */
00477     off_t    position_reference;
00478     off_t    seek_info_pos;
00479     off_t    segment_info_pos;
00480     off_t    track_pos;
00481     off_t    cue_pos;
00482     off_t    cluster_pos;
00483 
00484     /* This pointer is to a specific element to be serialized */
00485     off_t    track_id_pos;
00486 
00487     /* These pointers are to the size field of the element */
00488     EbmlLoc  startSegment;
00489     EbmlLoc  startCluster;
00490 
00491     uint32_t cluster_timecode;
00492     int      cluster_open;
00493 
00494     struct cue_entry *cue_list;
00495     unsigned int      cues;
00496 
00497 };
00498 
00499 
00500 void Ebml_Write(EbmlGlobal *glob, const void *buffer_in, unsigned long len)
00501 {
00502     if(fwrite(buffer_in, 1, len, glob->stream));
00503 }
00504 
00505 #define WRITE_BUFFER(s) \
00506 for(i = len-1; i>=0; i--)\
00507 { \
00508     x = *(const s *)buffer_in >> (i * CHAR_BIT); \
00509     Ebml_Write(glob, &x, 1); \
00510 }
00511 void Ebml_Serialize(EbmlGlobal *glob, const void *buffer_in, int buffer_size, unsigned long len)
00512 {
00513     char x;
00514     int i;
00515 
00516     /* buffer_size:
00517      * 1 - int8_t;
00518      * 2 - int16_t;
00519      * 3 - int32_t;
00520      * 4 - int64_t;
00521      */
00522     switch (buffer_size)
00523     {
00524         case 1:
00525             WRITE_BUFFER(int8_t)
00526             break;
00527         case 2:
00528             WRITE_BUFFER(int16_t)
00529             break;
00530         case 4:
00531             WRITE_BUFFER(int32_t)
00532             break;
00533         case 8:
00534             WRITE_BUFFER(int64_t)
00535             break;
00536         default:
00537             break;
00538     }
00539 }
00540 #undef WRITE_BUFFER
00541 
00542 /* Need a fixed size serializer for the track ID. libmkv provides a 64 bit
00543  * one, but not a 32 bit one.
00544  */
00545 static void Ebml_SerializeUnsigned32(EbmlGlobal *glob, unsigned long class_id, uint64_t ui)
00546 {
00547     unsigned char sizeSerialized = 4 | 0x80;
00548     Ebml_WriteID(glob, class_id);
00549     Ebml_Serialize(glob, &sizeSerialized, sizeof(sizeSerialized), 1);
00550     Ebml_Serialize(glob, &ui, sizeof(ui), 4);
00551 }
00552 
00553 
00554 static void
00555 Ebml_StartSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc,
00556                           unsigned long class_id)
00557 {
00558     //todo this is always taking 8 bytes, this may need later optimization
00559     //this is a key that says length unknown
00560     uint64_t unknownLen =  LITERALU64(0x01FFFFFFFFFFFFFF);
00561 
00562     Ebml_WriteID(glob, class_id);
00563     *ebmlLoc = ftello(glob->stream);
00564     Ebml_Serialize(glob, &unknownLen, sizeof(unknownLen), 8);
00565 }
00566 
00567 static void
00568 Ebml_EndSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc)
00569 {
00570     off_t pos;
00571     uint64_t size;
00572 
00573     /* Save the current stream pointer */
00574     pos = ftello(glob->stream);
00575 
00576     /* Calculate the size of this element */
00577     size = pos - *ebmlLoc - 8;
00578     size |=  LITERALU64(0x0100000000000000);
00579 
00580     /* Seek back to the beginning of the element and write the new size */
00581     fseeko(glob->stream, *ebmlLoc, SEEK_SET);
00582     Ebml_Serialize(glob, &size, sizeof(size), 8);
00583 
00584     /* Reset the stream pointer */
00585     fseeko(glob->stream, pos, SEEK_SET);
00586 }
00587 
00588 
00589 static void
00590 write_webm_seek_element(EbmlGlobal *ebml, unsigned long id, off_t pos)
00591 {
00592     uint64_t offset = pos - ebml->position_reference;
00593     EbmlLoc start;
00594     Ebml_StartSubElement(ebml, &start, Seek);
00595     Ebml_SerializeBinary(ebml, SeekID, id);
00596     Ebml_SerializeUnsigned64(ebml, SeekPosition, offset);
00597     Ebml_EndSubElement(ebml, &start);
00598 }
00599 
00600 
00601 static void
00602 write_webm_seek_info(EbmlGlobal *ebml)
00603 {
00604 
00605     off_t pos;
00606 
00607     /* Save the current stream pointer */
00608     pos = ftello(ebml->stream);
00609 
00610     if(ebml->seek_info_pos)
00611         fseeko(ebml->stream, ebml->seek_info_pos, SEEK_SET);
00612     else
00613         ebml->seek_info_pos = pos;
00614 
00615     {
00616         EbmlLoc start;
00617 
00618         Ebml_StartSubElement(ebml, &start, SeekHead);
00619         write_webm_seek_element(ebml, Tracks, ebml->track_pos);
00620         write_webm_seek_element(ebml, Cues,   ebml->cue_pos);
00621         write_webm_seek_element(ebml, Info,   ebml->segment_info_pos);
00622         Ebml_EndSubElement(ebml, &start);
00623     }
00624     {
00625         //segment info
00626         EbmlLoc startInfo;
00627         uint64_t frame_time;
00628         char version_string[64];
00629 
00630         /* Assemble version string */
00631         if(ebml->debug)
00632             strcpy(version_string, "vpxenc");
00633         else
00634         {
00635             strcpy(version_string, "vpxenc ");
00636             strncat(version_string,
00637                     vpx_codec_version_str(),
00638                     sizeof(version_string) - 1 - strlen(version_string));
00639         }
00640 
00641         frame_time = (uint64_t)1000 * ebml->framerate.den
00642                      / ebml->framerate.num;
00643         ebml->segment_info_pos = ftello(ebml->stream);
00644         Ebml_StartSubElement(ebml, &startInfo, Info);
00645         Ebml_SerializeUnsigned(ebml, TimecodeScale, 1000000);
00646         Ebml_SerializeFloat(ebml, Segment_Duration,
00647                             ebml->last_pts_ms + frame_time);
00648         Ebml_SerializeString(ebml, 0x4D80, version_string);
00649         Ebml_SerializeString(ebml, 0x5741, version_string);
00650         Ebml_EndSubElement(ebml, &startInfo);
00651     }
00652 }
00653 
00654 
00655 static void
00656 write_webm_file_header(EbmlGlobal                *glob,
00657                        const vpx_codec_enc_cfg_t *cfg,
00658                        const struct vpx_rational *fps,
00659                        stereo_format_t            stereo_fmt)
00660 {
00661     {
00662         EbmlLoc start;
00663         Ebml_StartSubElement(glob, &start, EBML);
00664         Ebml_SerializeUnsigned(glob, EBMLVersion, 1);
00665         Ebml_SerializeUnsigned(glob, EBMLReadVersion, 1); //EBML Read Version
00666         Ebml_SerializeUnsigned(glob, EBMLMaxIDLength, 4); //EBML Max ID Length
00667         Ebml_SerializeUnsigned(glob, EBMLMaxSizeLength, 8); //EBML Max Size Length
00668         Ebml_SerializeString(glob, DocType, "webm"); //Doc Type
00669         Ebml_SerializeUnsigned(glob, DocTypeVersion, 2); //Doc Type Version
00670         Ebml_SerializeUnsigned(glob, DocTypeReadVersion, 2); //Doc Type Read Version
00671         Ebml_EndSubElement(glob, &start);
00672     }
00673     {
00674         Ebml_StartSubElement(glob, &glob->startSegment, Segment); //segment
00675         glob->position_reference = ftello(glob->stream);
00676         glob->framerate = *fps;
00677         write_webm_seek_info(glob);
00678 
00679         {
00680             EbmlLoc trackStart;
00681             glob->track_pos = ftello(glob->stream);
00682             Ebml_StartSubElement(glob, &trackStart, Tracks);
00683             {
00684                 unsigned int trackNumber = 1;
00685                 uint64_t     trackID = 0;
00686 
00687                 EbmlLoc start;
00688                 Ebml_StartSubElement(glob, &start, TrackEntry);
00689                 Ebml_SerializeUnsigned(glob, TrackNumber, trackNumber);
00690                 glob->track_id_pos = ftello(glob->stream);
00691                 Ebml_SerializeUnsigned32(glob, TrackUID, trackID);
00692                 Ebml_SerializeUnsigned(glob, TrackType, 1); //video is always 1
00693                 Ebml_SerializeString(glob, CodecID, "V_VP8");
00694                 {
00695                     unsigned int pixelWidth = cfg->g_w;
00696                     unsigned int pixelHeight = cfg->g_h;
00697                     float        frameRate   = (float)fps->num/(float)fps->den;
00698 
00699                     EbmlLoc videoStart;
00700                     Ebml_StartSubElement(glob, &videoStart, Video);
00701                     Ebml_SerializeUnsigned(glob, PixelWidth, pixelWidth);
00702                     Ebml_SerializeUnsigned(glob, PixelHeight, pixelHeight);
00703                     Ebml_SerializeUnsigned(glob, StereoMode, stereo_fmt);
00704                     Ebml_SerializeFloat(glob, FrameRate, frameRate);
00705                     Ebml_EndSubElement(glob, &videoStart); //Video
00706                 }
00707                 Ebml_EndSubElement(glob, &start); //Track Entry
00708             }
00709             Ebml_EndSubElement(glob, &trackStart);
00710         }
00711         // segment element is open
00712     }
00713 }
00714 
00715 
00716 static void
00717 write_webm_block(EbmlGlobal                *glob,
00718                  const vpx_codec_enc_cfg_t *cfg,
00719                  const vpx_codec_cx_pkt_t  *pkt)
00720 {
00721     unsigned long  block_length;
00722     unsigned char  track_number;
00723     unsigned short block_timecode = 0;
00724     unsigned char  flags;
00725     int64_t        pts_ms;
00726     int            start_cluster = 0, is_keyframe;
00727 
00728     /* Calculate the PTS of this frame in milliseconds */
00729     pts_ms = pkt->data.frame.pts * 1000
00730              * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
00731     if(pts_ms <= glob->last_pts_ms)
00732         pts_ms = glob->last_pts_ms + 1;
00733     glob->last_pts_ms = pts_ms;
00734 
00735     /* Calculate the relative time of this block */
00736     if(pts_ms - glob->cluster_timecode > SHRT_MAX)
00737         start_cluster = 1;
00738     else
00739         block_timecode = pts_ms - glob->cluster_timecode;
00740 
00741     is_keyframe = (pkt->data.frame.flags & VPX_FRAME_IS_KEY);
00742     if(start_cluster || is_keyframe)
00743     {
00744         if(glob->cluster_open)
00745             Ebml_EndSubElement(glob, &glob->startCluster);
00746 
00747         /* Open the new cluster */
00748         block_timecode = 0;
00749         glob->cluster_open = 1;
00750         glob->cluster_timecode = pts_ms;
00751         glob->cluster_pos = ftello(glob->stream);
00752         Ebml_StartSubElement(glob, &glob->startCluster, Cluster); //cluster
00753         Ebml_SerializeUnsigned(glob, Timecode, glob->cluster_timecode);
00754 
00755         /* Save a cue point if this is a keyframe. */
00756         if(is_keyframe)
00757         {
00758             struct cue_entry *cue, *new_cue_list;
00759 
00760             new_cue_list = realloc(glob->cue_list,
00761                                    (glob->cues+1) * sizeof(struct cue_entry));
00762             if(new_cue_list)
00763                 glob->cue_list = new_cue_list;
00764             else
00765             {
00766                 fprintf(stderr, "\nFailed to realloc cue list.\n");
00767                 exit(EXIT_FAILURE);
00768             }
00769 
00770             cue = &glob->cue_list[glob->cues];
00771             cue->time = glob->cluster_timecode;
00772             cue->loc = glob->cluster_pos;
00773             glob->cues++;
00774         }
00775     }
00776 
00777     /* Write the Simple Block */
00778     Ebml_WriteID(glob, SimpleBlock);
00779 
00780     block_length = pkt->data.frame.sz + 4;
00781     block_length |= 0x10000000;
00782     Ebml_Serialize(glob, &block_length, sizeof(block_length), 4);
00783 
00784     track_number = 1;
00785     track_number |= 0x80;
00786     Ebml_Write(glob, &track_number, 1);
00787 
00788     Ebml_Serialize(glob, &block_timecode, sizeof(block_timecode), 2);
00789 
00790     flags = 0;
00791     if(is_keyframe)
00792         flags |= 0x80;
00793     if(pkt->data.frame.flags & VPX_FRAME_IS_INVISIBLE)
00794         flags |= 0x08;
00795     Ebml_Write(glob, &flags, 1);
00796 
00797     Ebml_Write(glob, pkt->data.frame.buf, pkt->data.frame.sz);
00798 }
00799 
00800 
00801 static void
00802 write_webm_file_footer(EbmlGlobal *glob, long hash)
00803 {
00804 
00805     if(glob->cluster_open)
00806         Ebml_EndSubElement(glob, &glob->startCluster);
00807 
00808     {
00809         EbmlLoc start;
00810         unsigned int i;
00811 
00812         glob->cue_pos = ftello(glob->stream);
00813         Ebml_StartSubElement(glob, &start, Cues);
00814         for(i=0; i<glob->cues; i++)
00815         {
00816             struct cue_entry *cue = &glob->cue_list[i];
00817             EbmlLoc start;
00818 
00819             Ebml_StartSubElement(glob, &start, CuePoint);
00820             {
00821                 EbmlLoc start;
00822 
00823                 Ebml_SerializeUnsigned(glob, CueTime, cue->time);
00824 
00825                 Ebml_StartSubElement(glob, &start, CueTrackPositions);
00826                 Ebml_SerializeUnsigned(glob, CueTrack, 1);
00827                 Ebml_SerializeUnsigned64(glob, CueClusterPosition,
00828                                          cue->loc - glob->position_reference);
00829                 //Ebml_SerializeUnsigned(glob, CueBlockNumber, cue->blockNumber);
00830                 Ebml_EndSubElement(glob, &start);
00831             }
00832             Ebml_EndSubElement(glob, &start);
00833         }
00834         Ebml_EndSubElement(glob, &start);
00835     }
00836 
00837     Ebml_EndSubElement(glob, &glob->startSegment);
00838 
00839     /* Patch up the seek info block */
00840     write_webm_seek_info(glob);
00841 
00842     /* Patch up the track id */
00843     fseeko(glob->stream, glob->track_id_pos, SEEK_SET);
00844     Ebml_SerializeUnsigned32(glob, TrackUID, glob->debug ? 0xDEADBEEF : hash);
00845 
00846     fseeko(glob->stream, 0, SEEK_END);
00847 }
00848 
00849 
00850 /* Murmur hash derived from public domain reference implementation at
00851  *   http://sites.google.com/site/murmurhash/
00852  */
00853 static unsigned int murmur ( const void * key, int len, unsigned int seed )
00854 {
00855     const unsigned int m = 0x5bd1e995;
00856     const int r = 24;
00857 
00858     unsigned int h = seed ^ len;
00859 
00860     const unsigned char * data = (const unsigned char *)key;
00861 
00862     while(len >= 4)
00863     {
00864         unsigned int k;
00865 
00866         k  = data[0];
00867         k |= data[1] << 8;
00868         k |= data[2] << 16;
00869         k |= data[3] << 24;
00870 
00871         k *= m;
00872         k ^= k >> r;
00873         k *= m;
00874 
00875         h *= m;
00876         h ^= k;
00877 
00878         data += 4;
00879         len -= 4;
00880     }
00881 
00882     switch(len)
00883     {
00884     case 3: h ^= data[2] << 16;
00885     case 2: h ^= data[1] << 8;
00886     case 1: h ^= data[0];
00887             h *= m;
00888     };
00889 
00890     h ^= h >> 13;
00891     h *= m;
00892     h ^= h >> 15;
00893 
00894     return h;
00895 }
00896 
00897 #include "math.h"
00898 
00899 static double vp8_mse2psnr(double Samples, double Peak, double Mse)
00900 {
00901     double psnr;
00902 
00903     if ((double)Mse > 0.0)
00904         psnr = 10.0 * log10(Peak * Peak * Samples / Mse);
00905     else
00906         psnr = 60;      // Limit to prevent / 0
00907 
00908     if (psnr > 60)
00909         psnr = 60;
00910 
00911     return psnr;
00912 }
00913 
00914 
00915 #include "args.h"
00916 
00917 static const arg_def_t debugmode = ARG_DEF("D", "debug", 0,
00918         "Debug mode (makes output deterministic)");
00919 static const arg_def_t outputfile = ARG_DEF("o", "output", 1,
00920         "Output filename");
00921 static const arg_def_t use_yv12 = ARG_DEF(NULL, "yv12", 0,
00922                                   "Input file is YV12 ");
00923 static const arg_def_t use_i420 = ARG_DEF(NULL, "i420", 0,
00924                                   "Input file is I420 (default)");
00925 static const arg_def_t codecarg = ARG_DEF(NULL, "codec", 1,
00926                                   "Codec to use");
00927 static const arg_def_t passes           = ARG_DEF("p", "passes", 1,
00928         "Number of passes (1/2)");
00929 static const arg_def_t pass_arg         = ARG_DEF(NULL, "pass", 1,
00930         "Pass to execute (1/2)");
00931 static const arg_def_t fpf_name         = ARG_DEF(NULL, "fpf", 1,
00932         "First pass statistics file name");
00933 static const arg_def_t limit = ARG_DEF(NULL, "limit", 1,
00934                                        "Stop encoding after n input frames");
00935 static const arg_def_t deadline         = ARG_DEF("d", "deadline", 1,
00936         "Deadline per frame (usec)");
00937 static const arg_def_t best_dl          = ARG_DEF(NULL, "best", 0,
00938         "Use Best Quality Deadline");
00939 static const arg_def_t good_dl          = ARG_DEF(NULL, "good", 0,
00940         "Use Good Quality Deadline");
00941 static const arg_def_t rt_dl            = ARG_DEF(NULL, "rt", 0,
00942         "Use Realtime Quality Deadline");
00943 static const arg_def_t verbosearg       = ARG_DEF("v", "verbose", 0,
00944         "Show encoder parameters");
00945 static const arg_def_t psnrarg          = ARG_DEF(NULL, "psnr", 0,
00946         "Show PSNR in status line");
00947 static const arg_def_t framerate        = ARG_DEF(NULL, "fps", 1,
00948         "Stream frame rate (rate/scale)");
00949 static const arg_def_t use_ivf          = ARG_DEF(NULL, "ivf", 0,
00950         "Output IVF (default is WebM)");
00951 static const arg_def_t q_hist_n         = ARG_DEF(NULL, "q-hist", 1,
00952         "Show quantizer histogram (n-buckets)");
00953 static const arg_def_t rate_hist_n         = ARG_DEF(NULL, "rate-hist", 1,
00954         "Show rate histogram (n-buckets)");
00955 static const arg_def_t *main_args[] =
00956 {
00957     &debugmode,
00958     &outputfile, &codecarg, &passes, &pass_arg, &fpf_name, &limit, &deadline,
00959     &best_dl, &good_dl, &rt_dl,
00960     &verbosearg, &psnrarg, &use_ivf, &q_hist_n, &rate_hist_n,
00961     NULL
00962 };
00963 
00964 static const arg_def_t usage            = ARG_DEF("u", "usage", 1,
00965         "Usage profile number to use");
00966 static const arg_def_t threads          = ARG_DEF("t", "threads", 1,
00967         "Max number of threads to use");
00968 static const arg_def_t profile          = ARG_DEF(NULL, "profile", 1,
00969         "Bitstream profile number to use");
00970 static const arg_def_t width            = ARG_DEF("w", "width", 1,
00971         "Frame width");
00972 static const arg_def_t height           = ARG_DEF("h", "height", 1,
00973         "Frame height");
00974 static const struct arg_enum_list stereo_mode_enum[] = {
00975     {"mono"      , STEREO_FORMAT_MONO},
00976     {"left-right", STEREO_FORMAT_LEFT_RIGHT},
00977     {"bottom-top", STEREO_FORMAT_BOTTOM_TOP},
00978     {"top-bottom", STEREO_FORMAT_TOP_BOTTOM},
00979     {"right-left", STEREO_FORMAT_RIGHT_LEFT},
00980     {NULL, 0}
00981 };
00982 static const arg_def_t stereo_mode      = ARG_DEF_ENUM(NULL, "stereo-mode", 1,
00983         "Stereo 3D video format", stereo_mode_enum);
00984 static const arg_def_t timebase         = ARG_DEF(NULL, "timebase", 1,
00985         "Output timestamp precision (fractional seconds)");
00986 static const arg_def_t error_resilient  = ARG_DEF(NULL, "error-resilient", 1,
00987         "Enable error resiliency features");
00988 static const arg_def_t lag_in_frames    = ARG_DEF(NULL, "lag-in-frames", 1,
00989         "Max number of frames to lag");
00990 
00991 static const arg_def_t *global_args[] =
00992 {
00993     &use_yv12, &use_i420, &usage, &threads, &profile,
00994     &width, &height, &stereo_mode, &timebase, &framerate, &error_resilient,
00995     &lag_in_frames, NULL
00996 };
00997 
00998 static const arg_def_t dropframe_thresh   = ARG_DEF(NULL, "drop-frame", 1,
00999         "Temporal resampling threshold (buf %)");
01000 static const arg_def_t resize_allowed     = ARG_DEF(NULL, "resize-allowed", 1,
01001         "Spatial resampling enabled (bool)");
01002 static const arg_def_t resize_up_thresh   = ARG_DEF(NULL, "resize-up", 1,
01003         "Upscale threshold (buf %)");
01004 static const arg_def_t resize_down_thresh = ARG_DEF(NULL, "resize-down", 1,
01005         "Downscale threshold (buf %)");
01006 static const struct arg_enum_list end_usage_enum[] = {
01007     {"vbr", VPX_VBR},
01008     {"cbr", VPX_CBR},
01009     {"cq",  VPX_CQ},
01010     {NULL, 0}
01011 };
01012 static const arg_def_t end_usage          = ARG_DEF_ENUM(NULL, "end-usage", 1,
01013         "Rate control mode", end_usage_enum);
01014 static const arg_def_t target_bitrate     = ARG_DEF(NULL, "target-bitrate", 1,
01015         "Bitrate (kbps)");
01016 static const arg_def_t min_quantizer      = ARG_DEF(NULL, "min-q", 1,
01017         "Minimum (best) quantizer");
01018 static const arg_def_t max_quantizer      = ARG_DEF(NULL, "max-q", 1,
01019         "Maximum (worst) quantizer");
01020 static const arg_def_t undershoot_pct     = ARG_DEF(NULL, "undershoot-pct", 1,
01021         "Datarate undershoot (min) target (%)");
01022 static const arg_def_t overshoot_pct      = ARG_DEF(NULL, "overshoot-pct", 1,
01023         "Datarate overshoot (max) target (%)");
01024 static const arg_def_t buf_sz             = ARG_DEF(NULL, "buf-sz", 1,
01025         "Client buffer size (ms)");
01026 static const arg_def_t buf_initial_sz     = ARG_DEF(NULL, "buf-initial-sz", 1,
01027         "Client initial buffer size (ms)");
01028 static const arg_def_t buf_optimal_sz     = ARG_DEF(NULL, "buf-optimal-sz", 1,
01029         "Client optimal buffer size (ms)");
01030 static const arg_def_t *rc_args[] =
01031 {
01032     &dropframe_thresh, &resize_allowed, &resize_up_thresh, &resize_down_thresh,
01033     &end_usage, &target_bitrate, &min_quantizer, &max_quantizer,
01034     &undershoot_pct, &overshoot_pct, &buf_sz, &buf_initial_sz, &buf_optimal_sz,
01035     NULL
01036 };
01037 
01038 
01039 static const arg_def_t bias_pct = ARG_DEF(NULL, "bias-pct", 1,
01040                                   "CBR/VBR bias (0=CBR, 100=VBR)");
01041 static const arg_def_t minsection_pct = ARG_DEF(NULL, "minsection-pct", 1,
01042                                         "GOP min bitrate (% of target)");
01043 static const arg_def_t maxsection_pct = ARG_DEF(NULL, "maxsection-pct", 1,
01044                                         "GOP max bitrate (% of target)");
01045 static const arg_def_t *rc_twopass_args[] =
01046 {
01047     &bias_pct, &minsection_pct, &maxsection_pct, NULL
01048 };
01049 
01050 
01051 static const arg_def_t kf_min_dist = ARG_DEF(NULL, "kf-min-dist", 1,
01052                                      "Minimum keyframe interval (frames)");
01053 static const arg_def_t kf_max_dist = ARG_DEF(NULL, "kf-max-dist", 1,
01054                                      "Maximum keyframe interval (frames)");
01055 static const arg_def_t kf_disabled = ARG_DEF(NULL, "disable-kf", 0,
01056                                      "Disable keyframe placement");
01057 static const arg_def_t *kf_args[] =
01058 {
01059     &kf_min_dist, &kf_max_dist, &kf_disabled, NULL
01060 };
01061 
01062 
01063 #if CONFIG_VP8_ENCODER
01064 static const arg_def_t noise_sens = ARG_DEF(NULL, "noise-sensitivity", 1,
01065                                     "Noise sensitivity (frames to blur)");
01066 static const arg_def_t sharpness = ARG_DEF(NULL, "sharpness", 1,
01067                                    "Filter sharpness (0-7)");
01068 static const arg_def_t static_thresh = ARG_DEF(NULL, "static-thresh", 1,
01069                                        "Motion detection threshold");
01070 #endif
01071 
01072 #if CONFIG_VP8_ENCODER
01073 static const arg_def_t cpu_used = ARG_DEF(NULL, "cpu-used", 1,
01074                                   "CPU Used (-16..16)");
01075 #endif
01076 
01077 
01078 #if CONFIG_VP8_ENCODER
01079 static const arg_def_t token_parts = ARG_DEF(NULL, "token-parts", 1,
01080                                      "Number of token partitions to use, log2");
01081 static const arg_def_t auto_altref = ARG_DEF(NULL, "auto-alt-ref", 1,
01082                                      "Enable automatic alt reference frames");
01083 static const arg_def_t arnr_maxframes = ARG_DEF(NULL, "arnr-maxframes", 1,
01084                                         "AltRef Max Frames");
01085 static const arg_def_t arnr_strength = ARG_DEF(NULL, "arnr-strength", 1,
01086                                        "AltRef Strength");
01087 static const arg_def_t arnr_type = ARG_DEF(NULL, "arnr-type", 1,
01088                                    "AltRef Type");
01089 static const struct arg_enum_list tuning_enum[] = {
01090     {"psnr", VP8_TUNE_PSNR},
01091     {"ssim", VP8_TUNE_SSIM},
01092     {NULL, 0}
01093 };
01094 static const arg_def_t tune_ssim = ARG_DEF_ENUM(NULL, "tune", 1,
01095                                    "Material to favor", tuning_enum);
01096 static const arg_def_t cq_level = ARG_DEF(NULL, "cq-level", 1,
01097                                    "Constrained Quality Level");
01098 static const arg_def_t max_intra_rate_pct = ARG_DEF(NULL, "max-intra-rate", 1,
01099         "Max I-frame bitrate (pct)");
01100 
01101 static const arg_def_t *vp8_args[] =
01102 {
01103     &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh,
01104     &token_parts, &arnr_maxframes, &arnr_strength, &arnr_type,
01105     &tune_ssim, &cq_level, &max_intra_rate_pct, NULL
01106 };
01107 static const int vp8_arg_ctrl_map[] =
01108 {
01109     VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF,
01110     VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD,
01111     VP8E_SET_TOKEN_PARTITIONS,
01112     VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH , VP8E_SET_ARNR_TYPE,
01113     VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, VP8E_SET_MAX_INTRA_BITRATE_PCT, 0
01114 };
01115 #endif
01116 
01117 static const arg_def_t *no_args[] = { NULL };
01118 
01119 static void usage_exit()
01120 {
01121     int i;
01122 
01123     fprintf(stderr, "Usage: %s <options> -o dst_filename src_filename \n",
01124             exec_name);
01125 
01126     fprintf(stderr, "\nOptions:\n");
01127     arg_show_usage(stdout, main_args);
01128     fprintf(stderr, "\nEncoder Global Options:\n");
01129     arg_show_usage(stdout, global_args);
01130     fprintf(stderr, "\nRate Control Options:\n");
01131     arg_show_usage(stdout, rc_args);
01132     fprintf(stderr, "\nTwopass Rate Control Options:\n");
01133     arg_show_usage(stdout, rc_twopass_args);
01134     fprintf(stderr, "\nKeyframe Placement Options:\n");
01135     arg_show_usage(stdout, kf_args);
01136 #if CONFIG_VP8_ENCODER
01137     fprintf(stderr, "\nVP8 Specific Options:\n");
01138     arg_show_usage(stdout, vp8_args);
01139 #endif
01140     fprintf(stderr, "\nStream timebase (--timebase):\n"
01141             "  The desired precision of timestamps in the output, expressed\n"
01142             "  in fractional seconds. Default is 1/1000.\n");
01143     fprintf(stderr, "\n"
01144            "Included encoders:\n"
01145            "\n");
01146 
01147     for (i = 0; i < sizeof(codecs) / sizeof(codecs[0]); i++)
01148         fprintf(stderr, "    %-6s - %s\n",
01149                codecs[i].name,
01150                vpx_codec_iface_name(codecs[i].iface));
01151 
01152     exit(EXIT_FAILURE);
01153 }
01154 
01155 
01156 #define HIST_BAR_MAX 40
01157 struct hist_bucket
01158 {
01159     int low, high, count;
01160 };
01161 
01162 
01163 static int merge_hist_buckets(struct hist_bucket *bucket,
01164                               int *buckets_,
01165                               int max_buckets)
01166 {
01167     int small_bucket = 0, merge_bucket = INT_MAX, big_bucket=0;
01168     int buckets = *buckets_;
01169     int i;
01170 
01171     /* Find the extrema for this list of buckets */
01172     big_bucket = small_bucket = 0;
01173     for(i=0; i < buckets; i++)
01174     {
01175         if(bucket[i].count < bucket[small_bucket].count)
01176             small_bucket = i;
01177         if(bucket[i].count > bucket[big_bucket].count)
01178             big_bucket = i;
01179     }
01180 
01181     /* If we have too many buckets, merge the smallest with an adjacent
01182      * bucket.
01183      */
01184     while(buckets > max_buckets)
01185     {
01186         int last_bucket = buckets - 1;
01187 
01188         // merge the small bucket with an adjacent one.
01189         if(small_bucket == 0)
01190             merge_bucket = 1;
01191         else if(small_bucket == last_bucket)
01192             merge_bucket = last_bucket - 1;
01193         else if(bucket[small_bucket - 1].count < bucket[small_bucket + 1].count)
01194             merge_bucket = small_bucket - 1;
01195         else
01196             merge_bucket = small_bucket + 1;
01197 
01198         assert(abs(merge_bucket - small_bucket) <= 1);
01199         assert(small_bucket < buckets);
01200         assert(big_bucket < buckets);
01201         assert(merge_bucket < buckets);
01202 
01203         if(merge_bucket < small_bucket)
01204         {
01205             bucket[merge_bucket].high = bucket[small_bucket].high;
01206             bucket[merge_bucket].count += bucket[small_bucket].count;
01207         }
01208         else
01209         {
01210             bucket[small_bucket].high = bucket[merge_bucket].high;
01211             bucket[small_bucket].count += bucket[merge_bucket].count;
01212             merge_bucket = small_bucket;
01213         }
01214 
01215         assert(bucket[merge_bucket].low != bucket[merge_bucket].high);
01216 
01217         buckets--;
01218 
01219         /* Remove the merge_bucket from the list, and find the new small
01220          * and big buckets while we're at it
01221          */
01222         big_bucket = small_bucket = 0;
01223         for(i=0; i < buckets; i++)
01224         {
01225             if(i > merge_bucket)
01226                 bucket[i] = bucket[i+1];
01227 
01228             if(bucket[i].count < bucket[small_bucket].count)
01229                 small_bucket = i;
01230             if(bucket[i].count > bucket[big_bucket].count)
01231                 big_bucket = i;
01232         }
01233 
01234     }
01235 
01236     *buckets_ = buckets;
01237     return bucket[big_bucket].count;
01238 }
01239 
01240 
01241 static void show_histogram(const struct hist_bucket *bucket,
01242                            int                       buckets,
01243                            int                       total,
01244                            int                       scale)
01245 {
01246     const char *pat1, *pat2;
01247     int i;
01248 
01249     switch((int)(log(bucket[buckets-1].high)/log(10))+1)
01250     {
01251         case 1:
01252         case 2:
01253             pat1 = "%4d %2s: ";
01254             pat2 = "%4d-%2d: ";
01255             break;
01256         case 3:
01257             pat1 = "%5d %3s: ";
01258             pat2 = "%5d-%3d: ";
01259             break;
01260         case 4:
01261             pat1 = "%6d %4s: ";
01262             pat2 = "%6d-%4d: ";
01263             break;
01264         case 5:
01265             pat1 = "%7d %5s: ";
01266             pat2 = "%7d-%5d: ";
01267             break;
01268         case 6:
01269             pat1 = "%8d %6s: ";
01270             pat2 = "%8d-%6d: ";
01271             break;
01272         case 7:
01273             pat1 = "%9d %7s: ";
01274             pat2 = "%9d-%7d: ";
01275             break;
01276         default:
01277             pat1 = "%12d %10s: ";
01278             pat2 = "%12d-%10d: ";
01279             break;
01280     }
01281 
01282     for(i=0; i<buckets; i++)
01283     {
01284         int len;
01285         int j;
01286         float pct;
01287 
01288         pct = 100.0 * (float)bucket[i].count / (float)total;
01289         len = HIST_BAR_MAX * bucket[i].count / scale;
01290         if(len < 1)
01291             len = 1;
01292         assert(len <= HIST_BAR_MAX);
01293 
01294         if(bucket[i].low == bucket[i].high)
01295             fprintf(stderr, pat1, bucket[i].low, "");
01296         else
01297             fprintf(stderr, pat2, bucket[i].low, bucket[i].high);
01298 
01299         for(j=0; j<HIST_BAR_MAX; j++)
01300             fprintf(stderr, j<len?"=":" ");
01301         fprintf(stderr, "\t%5d (%6.2f%%)\n",bucket[i].count,pct);
01302     }
01303 }
01304 
01305 
01306 static void show_q_histogram(const int counts[64], int max_buckets)
01307 {
01308     struct hist_bucket bucket[64];
01309     int buckets = 0;
01310     int total = 0;
01311     int scale;
01312     int i;
01313 
01314 
01315     for(i=0; i<64; i++)
01316     {
01317         if(counts[i])
01318         {
01319             bucket[buckets].low = bucket[buckets].high = i;
01320             bucket[buckets].count = counts[i];
01321             buckets++;
01322             total += counts[i];
01323         }
01324     }
01325 
01326     fprintf(stderr, "\nQuantizer Selection:\n");
01327     scale = merge_hist_buckets(bucket, &buckets, max_buckets);
01328     show_histogram(bucket, buckets, total, scale);
01329 }
01330 
01331 
01332 #define RATE_BINS (100)
01333 struct rate_hist
01334 {
01335     int64_t            *pts;
01336     int                *sz;
01337     int                 samples;
01338     int                 frames;
01339     struct hist_bucket  bucket[RATE_BINS];
01340     int                 total;
01341 };
01342 
01343 
01344 static void init_rate_histogram(struct rate_hist          *hist,
01345                                 const vpx_codec_enc_cfg_t *cfg,
01346                                 const vpx_rational_t      *fps)
01347 {
01348     int i;
01349 
01350     /* Determine the number of samples in the buffer. Use the file's framerate
01351      * to determine the number of frames in rc_buf_sz milliseconds, with an
01352      * adjustment (5/4) to account for alt-refs
01353      */
01354     hist->samples = cfg->rc_buf_sz * 5 / 4 * fps->num / fps->den / 1000;
01355 
01356     // prevent division by zero
01357     if (hist->samples == 0)
01358       hist->samples=1;
01359 
01360     hist->pts = calloc(hist->samples, sizeof(*hist->pts));
01361     hist->sz = calloc(hist->samples, sizeof(*hist->sz));
01362     for(i=0; i<RATE_BINS; i++)
01363     {
01364         hist->bucket[i].low = INT_MAX;
01365         hist->bucket[i].high = 0;
01366         hist->bucket[i].count = 0;
01367     }
01368 }
01369 
01370 
01371 static void destroy_rate_histogram(struct rate_hist *hist)
01372 {
01373     free(hist->pts);
01374     free(hist->sz);
01375 }
01376 
01377 
01378 static void update_rate_histogram(struct rate_hist          *hist,
01379                                   const vpx_codec_enc_cfg_t *cfg,
01380                                   const vpx_codec_cx_pkt_t  *pkt)
01381 {
01382     int i, idx;
01383     int64_t now, then, sum_sz = 0, avg_bitrate;
01384 
01385     now = pkt->data.frame.pts * 1000
01386           * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den;
01387 
01388     idx = hist->frames++ % hist->samples;
01389     hist->pts[idx] = now;
01390     hist->sz[idx] = pkt->data.frame.sz;
01391 
01392     if(now < cfg->rc_buf_initial_sz)
01393         return;
01394 
01395     then = now;
01396 
01397     /* Sum the size over the past rc_buf_sz ms */
01398     for(i = hist->frames; i > 0 && hist->frames - i < hist->samples; i--)
01399     {
01400         int i_idx = (i-1) % hist->samples;
01401 
01402         then = hist->pts[i_idx];
01403         if(now - then > cfg->rc_buf_sz)
01404             break;
01405         sum_sz += hist->sz[i_idx];
01406     }
01407 
01408     if (now == then)
01409         return;
01410 
01411     avg_bitrate = sum_sz * 8 * 1000 / (now - then);
01412     idx = avg_bitrate * (RATE_BINS/2) / (cfg->rc_target_bitrate * 1000);
01413     if(idx < 0)
01414         idx = 0;
01415     if(idx > RATE_BINS-1)
01416         idx = RATE_BINS-1;
01417     if(hist->bucket[idx].low > avg_bitrate)
01418         hist->bucket[idx].low = avg_bitrate;
01419     if(hist->bucket[idx].high < avg_bitrate)
01420         hist->bucket[idx].high = avg_bitrate;
01421     hist->bucket[idx].count++;
01422     hist->total++;
01423 }
01424 
01425 
01426 static void show_rate_histogram(struct rate_hist          *hist,
01427                                 const vpx_codec_enc_cfg_t *cfg,
01428                                 int                        max_buckets)
01429 {
01430     int i, scale;
01431     int buckets = 0;
01432 
01433     for(i = 0; i < RATE_BINS; i++)
01434     {
01435         if(hist->bucket[i].low == INT_MAX)
01436             continue;
01437         hist->bucket[buckets++] = hist->bucket[i];
01438     }
01439 
01440     fprintf(stderr, "\nRate (over %dms window):\n", cfg->rc_buf_sz);
01441     scale = merge_hist_buckets(hist->bucket, &buckets, max_buckets);
01442     show_histogram(hist->bucket, buckets, hist->total, scale);
01443 }
01444 
01445 #define NELEMENTS(x) (sizeof(x)/sizeof(x[0]))
01446 #define ARG_CTRL_CNT_MAX NELEMENTS(vp8_arg_ctrl_map)
01447 
01448 int main(int argc, const char **argv_)
01449 {
01450     vpx_codec_ctx_t        encoder;
01451     const char                  *in_fn = NULL, *out_fn = NULL, *stats_fn = NULL;
01452     int                    i;
01453     FILE                  *infile, *outfile;
01454     vpx_codec_enc_cfg_t    cfg;
01455     vpx_codec_err_t        res;
01456     int                    pass, one_pass_only = 0;
01457     stats_io_t             stats;
01458     vpx_image_t            raw;
01459     const struct codec_item  *codec = codecs;
01460     int                    frame_avail, got_data;
01461 
01462     struct arg               arg;
01463     char                   **argv, **argi, **argj;
01464     int                      arg_usage = 0, arg_passes = 1, arg_deadline = 0;
01465     int                      arg_ctrls[ARG_CTRL_CNT_MAX][2], arg_ctrl_cnt = 0;
01466     int                      arg_limit = 0;
01467     static const arg_def_t **ctrl_args = no_args;
01468     static const int        *ctrl_args_map = NULL;
01469     int                      verbose = 0, show_psnr = 0;
01470     int                      arg_use_i420 = 1;
01471     unsigned long            cx_time = 0;
01472     unsigned int             file_type, fourcc;
01473     y4m_input                y4m;
01474     struct vpx_rational      arg_framerate = {30, 1};
01475     int                      arg_have_framerate = 0;
01476     int                      write_webm = 1;
01477     EbmlGlobal               ebml = {0};
01478     uint32_t                 hash = 0;
01479     uint64_t                 psnr_sse_total = 0;
01480     uint64_t                 psnr_samples_total = 0;
01481     double                   psnr_totals[4] = {0, 0, 0, 0};
01482     int                      psnr_count = 0;
01483     stereo_format_t          stereo_fmt = STEREO_FORMAT_MONO;
01484     int                      counts[64]={0};
01485     int                      show_q_hist_buckets=0;
01486     int                      show_rate_hist_buckets=0;
01487     struct rate_hist         rate_hist={0};
01488 
01489     exec_name = argv_[0];
01490     ebml.last_pts_ms = -1;
01491 
01492     if (argc < 3)
01493         usage_exit();
01494 
01495 
01496     /* First parse the codec and usage values, because we want to apply other
01497      * parameters on top of the default configuration provided by the codec.
01498      */
01499     argv = argv_dup(argc - 1, argv_ + 1);
01500 
01501     for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
01502     {
01503         arg.argv_step = 1;
01504 
01505         if (arg_match(&arg, &codecarg, argi))
01506         {
01507             int j, k = -1;
01508 
01509             for (j = 0; j < sizeof(codecs) / sizeof(codecs[0]); j++)
01510                 if (!strcmp(codecs[j].name, arg.val))
01511                     k = j;
01512 
01513             if (k >= 0)
01514                 codec = codecs + k;
01515             else
01516                 die("Error: Unrecognized argument (%s) to --codec\n",
01517                     arg.val);
01518 
01519         }
01520         else if (arg_match(&arg, &passes, argi))
01521         {
01522             arg_passes = arg_parse_uint(&arg);
01523 
01524             if (arg_passes < 1 || arg_passes > 2)
01525                 die("Error: Invalid number of passes (%d)\n", arg_passes);
01526         }
01527         else if (arg_match(&arg, &pass_arg, argi))
01528         {
01529             one_pass_only = arg_parse_uint(&arg);
01530 
01531             if (one_pass_only < 1 || one_pass_only > 2)
01532                 die("Error: Invalid pass selected (%d)\n", one_pass_only);
01533         }
01534         else if (arg_match(&arg, &fpf_name, argi))
01535             stats_fn = arg.val;
01536         else if (arg_match(&arg, &usage, argi))
01537             arg_usage = arg_parse_uint(&arg);
01538         else if (arg_match(&arg, &deadline, argi))
01539             arg_deadline = arg_parse_uint(&arg);
01540         else if (arg_match(&arg, &best_dl, argi))
01541             arg_deadline = VPX_DL_BEST_QUALITY;
01542         else if (arg_match(&arg, &good_dl, argi))
01543             arg_deadline = VPX_DL_GOOD_QUALITY;
01544         else if (arg_match(&arg, &rt_dl, argi))
01545             arg_deadline = VPX_DL_REALTIME;
01546         else if (arg_match(&arg, &use_yv12, argi))
01547         {
01548             arg_use_i420 = 0;
01549         }
01550         else if (arg_match(&arg, &use_i420, argi))
01551         {
01552             arg_use_i420 = 1;
01553         }
01554         else if (arg_match(&arg, &verbosearg, argi))
01555             verbose = 1;
01556         else if (arg_match(&arg, &limit, argi))
01557             arg_limit = arg_parse_uint(&arg);
01558         else if (arg_match(&arg, &psnrarg, argi))
01559             show_psnr = 1;
01560         else if (arg_match(&arg, &framerate, argi))
01561         {
01562             arg_framerate = arg_parse_rational(&arg);
01563             arg_have_framerate = 1;
01564         }
01565         else if (arg_match(&arg, &use_ivf, argi))
01566             write_webm = 0;
01567         else if (arg_match(&arg, &outputfile, argi))
01568             out_fn = arg.val;
01569         else if (arg_match(&arg, &debugmode, argi))
01570             ebml.debug = 1;
01571         else if (arg_match(&arg, &q_hist_n, argi))
01572             show_q_hist_buckets = arg_parse_uint(&arg);
01573         else if (arg_match(&arg, &rate_hist_n, argi))
01574             show_rate_hist_buckets = arg_parse_uint(&arg);
01575         else
01576             argj++;
01577     }
01578 
01579     /* Ensure that --passes and --pass are consistent. If --pass is set and --passes=2,
01580      * ensure --fpf was set.
01581      */
01582     if (one_pass_only)
01583     {
01584         /* DWIM: Assume the user meant passes=2 if pass=2 is specified */
01585         if (one_pass_only > arg_passes)
01586         {
01587             fprintf(stderr, "Warning: Assuming --pass=%d implies --passes=%d\n",
01588                    one_pass_only, one_pass_only);
01589             arg_passes = one_pass_only;
01590         }
01591 
01592         if (arg_passes == 2 && !stats_fn)
01593             die("Must specify --fpf when --pass=%d and --passes=2\n", one_pass_only);
01594     }
01595 
01596     /* Populate encoder configuration */
01597     res = vpx_codec_enc_config_default(codec->iface, &cfg, arg_usage);
01598 
01599     if (res)
01600     {
01601         fprintf(stderr, "Failed to get config: %s\n",
01602                 vpx_codec_err_to_string(res));
01603         return EXIT_FAILURE;
01604     }
01605 
01606     /* Change the default timebase to a high enough value so that the encoder
01607      * will always create strictly increasing timestamps.
01608      */
01609     cfg.g_timebase.den = 1000;
01610 
01611     /* Never use the library's default resolution, require it be parsed
01612      * from the file or set on the command line.
01613      */
01614     cfg.g_w = 0;
01615     cfg.g_h = 0;
01616 
01617     /* Now parse the remainder of the parameters. */
01618     for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
01619     {
01620         arg.argv_step = 1;
01621 
01622         if (0);
01623         else if (arg_match(&arg, &threads, argi))
01624             cfg.g_threads = arg_parse_uint(&arg);
01625         else if (arg_match(&arg, &profile, argi))
01626             cfg.g_profile = arg_parse_uint(&arg);
01627         else if (arg_match(&arg, &width, argi))
01628             cfg.g_w = arg_parse_uint(&arg);
01629         else if (arg_match(&arg, &height, argi))
01630             cfg.g_h = arg_parse_uint(&arg);
01631         else if (arg_match(&arg, &stereo_mode, argi))
01632             stereo_fmt = arg_parse_enum_or_int(&arg);
01633         else if (arg_match(&arg, &timebase, argi))
01634             cfg.g_timebase = arg_parse_rational(&arg);
01635         else if (arg_match(&arg, &error_resilient, argi))
01636             cfg.g_error_resilient = arg_parse_uint(&arg);
01637         else if (arg_match(&arg, &lag_in_frames, argi))
01638             cfg.g_lag_in_frames = arg_parse_uint(&arg);
01639         else if (arg_match(&arg, &dropframe_thresh, argi))
01640             cfg.rc_dropframe_thresh = arg_parse_uint(&arg);
01641         else if (arg_match(&arg, &resize_allowed, argi))
01642             cfg.rc_resize_allowed = arg_parse_uint(&arg);
01643         else if (arg_match(&arg, &resize_up_thresh, argi))
01644             cfg.rc_resize_up_thresh = arg_parse_uint(&arg);
01645         else if (arg_match(&arg, &resize_down_thresh, argi))
01646             cfg.rc_resize_down_thresh = arg_parse_uint(&arg);
01647         else if (arg_match(&arg, &end_usage, argi))
01648             cfg.rc_end_usage = arg_parse_enum_or_int(&arg);
01649         else if (arg_match(&arg, &target_bitrate, argi))
01650             cfg.rc_target_bitrate = arg_parse_uint(&arg);
01651         else if (arg_match(&arg, &min_quantizer, argi))
01652             cfg.rc_min_quantizer = arg_parse_uint(&arg);
01653         else if (arg_match(&arg, &max_quantizer, argi))
01654             cfg.rc_max_quantizer = arg_parse_uint(&arg);
01655         else if (arg_match(&arg, &undershoot_pct, argi))
01656             cfg.rc_undershoot_pct = arg_parse_uint(&arg);
01657         else if (arg_match(&arg, &overshoot_pct, argi))
01658             cfg.rc_overshoot_pct = arg_parse_uint(&arg);
01659         else if (arg_match(&arg, &buf_sz, argi))
01660             cfg.rc_buf_sz = arg_parse_uint(&arg);
01661         else if (arg_match(&arg, &buf_initial_sz, argi))
01662             cfg.rc_buf_initial_sz = arg_parse_uint(&arg);
01663         else if (arg_match(&arg, &buf_optimal_sz, argi))
01664             cfg.rc_buf_optimal_sz = arg_parse_uint(&arg);
01665         else if (arg_match(&arg, &bias_pct, argi))
01666         {
01667             cfg.rc_2pass_vbr_bias_pct = arg_parse_uint(&arg);
01668 
01669             if (arg_passes < 2)
01670                 fprintf(stderr,
01671                         "Warning: option %s ignored in one-pass mode.\n",
01672                         arg.name);
01673         }
01674         else if (arg_match(&arg, &minsection_pct, argi))
01675         {
01676             cfg.rc_2pass_vbr_minsection_pct = arg_parse_uint(&arg);
01677 
01678             if (arg_passes < 2)
01679                 fprintf(stderr,
01680                         "Warning: option %s ignored in one-pass mode.\n",
01681                         arg.name);
01682         }
01683         else if (arg_match(&arg, &maxsection_pct, argi))
01684         {
01685             cfg.rc_2pass_vbr_maxsection_pct = arg_parse_uint(&arg);
01686 
01687             if (arg_passes < 2)
01688                 fprintf(stderr,
01689                         "Warning: option %s ignored in one-pass mode.\n",
01690                         arg.name);
01691         }
01692         else if (arg_match(&arg, &kf_min_dist, argi))
01693             cfg.kf_min_dist = arg_parse_uint(&arg);
01694         else if (arg_match(&arg, &kf_max_dist, argi))
01695             cfg.kf_max_dist = arg_parse_uint(&arg);
01696         else if (arg_match(&arg, &kf_disabled, argi))
01697             cfg.kf_mode = VPX_KF_DISABLED;
01698         else
01699             argj++;
01700     }
01701 
01702     /* Handle codec specific options */
01703 #if CONFIG_VP8_ENCODER
01704 
01705     if (codec->iface == &vpx_codec_vp8_cx_algo)
01706     {
01707         ctrl_args = vp8_args;
01708         ctrl_args_map = vp8_arg_ctrl_map;
01709     }
01710 
01711 #endif
01712 
01713     for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step)
01714     {
01715         int match = 0;
01716 
01717         arg.argv_step = 1;
01718 
01719         for (i = 0; ctrl_args[i]; i++)
01720         {
01721             if (arg_match(&arg, ctrl_args[i], argi))
01722             {
01723                 int j;
01724                 match = 1;
01725 
01726                 /* Point either to the next free element or the first
01727                 * instance of this control.
01728                 */
01729                 for(j=0; j<arg_ctrl_cnt; j++)
01730                     if(arg_ctrls[j][0] == ctrl_args_map[i])
01731                         break;
01732 
01733                 /* Update/insert */
01734                 assert(j < ARG_CTRL_CNT_MAX);
01735                 if (j < ARG_CTRL_CNT_MAX)
01736                 {
01737                     arg_ctrls[j][0] = ctrl_args_map[i];
01738                     arg_ctrls[j][1] = arg_parse_enum_or_int(&arg);
01739                     if(j == arg_ctrl_cnt)
01740                         arg_ctrl_cnt++;
01741                 }
01742 
01743             }
01744         }
01745 
01746         if (!match)
01747             argj++;
01748     }
01749 
01750     /* Check for unrecognized options */
01751     for (argi = argv; *argi; argi++)
01752         if (argi[0][0] == '-' && argi[0][1])
01753             die("Error: Unrecognized option %s\n", *argi);
01754 
01755     /* Handle non-option arguments */
01756     in_fn = argv[0];
01757 
01758     if (!in_fn)
01759         usage_exit();
01760 
01761     if(!out_fn)
01762         die("Error: Output file is required (specify with -o)\n");
01763 
01764     memset(&stats, 0, sizeof(stats));
01765 
01766     for (pass = one_pass_only ? one_pass_only - 1 : 0; pass < arg_passes; pass++)
01767     {
01768         int frames_in = 0, frames_out = 0;
01769         int64_t nbytes = 0;
01770         struct detect_buffer detect;
01771 
01772         /* Parse certain options from the input file, if possible */
01773         infile = strcmp(in_fn, "-") ? fopen(in_fn, "rb")
01774                                     : set_binary_mode(stdin);
01775 
01776         if (!infile)
01777         {
01778             fprintf(stderr, "Failed to open input file\n");
01779             return EXIT_FAILURE;
01780         }
01781 
01782         /* For RAW input sources, these bytes will applied on the first frame
01783          *  in read_frame().
01784          */
01785         detect.buf_read = fread(detect.buf, 1, 4, infile);
01786         detect.position = 0;
01787 
01788         if (detect.buf_read == 4 && file_is_y4m(infile, &y4m, detect.buf))
01789         {
01790             if (y4m_input_open(&y4m, infile, detect.buf, 4) >= 0)
01791             {
01792                 file_type = FILE_TYPE_Y4M;
01793                 cfg.g_w = y4m.pic_w;
01794                 cfg.g_h = y4m.pic_h;
01795 
01796                 /* Use the frame rate from the file only if none was specified
01797                  * on the command-line.
01798                  */
01799                 if (!arg_have_framerate)
01800                 {
01801                     arg_framerate.num = y4m.fps_n;
01802                     arg_framerate.den = y4m.fps_d;
01803                 }
01804 
01805                 arg_use_i420 = 0;
01806             }
01807             else
01808             {
01809                 fprintf(stderr, "Unsupported Y4M stream.\n");
01810                 return EXIT_FAILURE;
01811             }
01812         }
01813         else if (detect.buf_read == 4 &&
01814                  file_is_ivf(infile, &fourcc, &cfg.g_w, &cfg.g_h, &detect))
01815         {
01816             file_type = FILE_TYPE_IVF;
01817             switch (fourcc)
01818             {
01819             case 0x32315659:
01820                 arg_use_i420 = 0;
01821                 break;
01822             case 0x30323449:
01823                 arg_use_i420 = 1;
01824                 break;
01825             default:
01826                 fprintf(stderr, "Unsupported fourcc (%08x) in IVF\n", fourcc);
01827                 return EXIT_FAILURE;
01828             }
01829         }
01830         else
01831         {
01832             file_type = FILE_TYPE_RAW;
01833         }
01834 
01835         if(!cfg.g_w || !cfg.g_h)
01836         {
01837             fprintf(stderr, "Specify stream dimensions with --width (-w) "
01838                             " and --height (-h).\n");
01839             return EXIT_FAILURE;
01840         }
01841 
01842 #define SHOW(field) fprintf(stderr, "    %-28s = %d\n", #field, cfg.field)
01843 
01844         if (verbose && pass == 0)
01845         {
01846             fprintf(stderr, "Codec: %s\n", vpx_codec_iface_name(codec->iface));
01847             fprintf(stderr, "Source file: %s Format: %s\n", in_fn,
01848                     arg_use_i420 ? "I420" : "YV12");
01849             fprintf(stderr, "Destination file: %s\n", out_fn);
01850             fprintf(stderr, "Encoder parameters:\n");
01851 
01852             SHOW(g_usage);
01853             SHOW(g_threads);
01854             SHOW(g_profile);
01855             SHOW(g_w);
01856             SHOW(g_h);
01857             SHOW(g_timebase.num);
01858             SHOW(g_timebase.den);
01859             SHOW(g_error_resilient);
01860             SHOW(g_pass);
01861             SHOW(g_lag_in_frames);
01862             SHOW(rc_dropframe_thresh);
01863             SHOW(rc_resize_allowed);
01864             SHOW(rc_resize_up_thresh);
01865             SHOW(rc_resize_down_thresh);
01866             SHOW(rc_end_usage);
01867             SHOW(rc_target_bitrate);
01868             SHOW(rc_min_quantizer);
01869             SHOW(rc_max_quantizer);
01870             SHOW(rc_undershoot_pct);
01871             SHOW(rc_overshoot_pct);
01872             SHOW(rc_buf_sz);
01873             SHOW(rc_buf_initial_sz);
01874             SHOW(rc_buf_optimal_sz);
01875             SHOW(rc_2pass_vbr_bias_pct);
01876             SHOW(rc_2pass_vbr_minsection_pct);
01877             SHOW(rc_2pass_vbr_maxsection_pct);
01878             SHOW(kf_mode);
01879             SHOW(kf_min_dist);
01880             SHOW(kf_max_dist);
01881         }
01882 
01883         if(pass == (one_pass_only ? one_pass_only - 1 : 0)) {
01884             if (file_type == FILE_TYPE_Y4M)
01885                 /*The Y4M reader does its own allocation.
01886                   Just initialize this here to avoid problems if we never read any
01887                    frames.*/
01888                 memset(&raw, 0, sizeof(raw));
01889             else
01890                 vpx_img_alloc(&raw, arg_use_i420 ? VPX_IMG_FMT_I420 : VPX_IMG_FMT_YV12,
01891                               cfg.g_w, cfg.g_h, 1);
01892 
01893             init_rate_histogram(&rate_hist, &cfg, &arg_framerate);
01894         }
01895 
01896         outfile = strcmp(out_fn, "-") ? fopen(out_fn, "wb")
01897                                       : set_binary_mode(stdout);
01898 
01899         if (!outfile)
01900         {
01901             fprintf(stderr, "Failed to open output file\n");
01902             return EXIT_FAILURE;
01903         }
01904 
01905         if(write_webm && fseek(outfile, 0, SEEK_CUR))
01906         {
01907             fprintf(stderr, "WebM output to pipes not supported.\n");
01908             return EXIT_FAILURE;
01909         }
01910 
01911         if (stats_fn)
01912         {
01913             if (!stats_open_file(&stats, stats_fn, pass))
01914             {
01915                 fprintf(stderr, "Failed to open statistics store\n");
01916                 return EXIT_FAILURE;
01917             }
01918         }
01919         else
01920         {
01921             if (!stats_open_mem(&stats, pass))
01922             {
01923                 fprintf(stderr, "Failed to open statistics store\n");
01924                 return EXIT_FAILURE;
01925             }
01926         }
01927 
01928         cfg.g_pass = arg_passes == 2
01929                      ? pass ? VPX_RC_LAST_PASS : VPX_RC_FIRST_PASS
01930                  : VPX_RC_ONE_PASS;
01931 #if VPX_ENCODER_ABI_VERSION > (1 + VPX_CODEC_ABI_VERSION)
01932 
01933         if (pass)
01934         {
01935             cfg.rc_twopass_stats_in = stats_get(&stats);
01936         }
01937 
01938 #endif
01939 
01940         if(write_webm)
01941         {
01942             ebml.stream = outfile;
01943             write_webm_file_header(&ebml, &cfg, &arg_framerate, stereo_fmt);
01944         }
01945         else
01946             write_ivf_file_header(outfile, &cfg, codec->fourcc, 0);
01947 
01948 
01949         /* Construct Encoder Context */
01950         vpx_codec_enc_init(&encoder, codec->iface, &cfg,
01951                            show_psnr ? VPX_CODEC_USE_PSNR : 0);
01952         ctx_exit_on_error(&encoder, "Failed to initialize encoder");
01953 
01954         /* Note that we bypass the vpx_codec_control wrapper macro because
01955          * we're being clever to store the control IDs in an array. Real
01956          * applications will want to make use of the enumerations directly
01957          */
01958         for (i = 0; i < arg_ctrl_cnt; i++)
01959         {
01960             if (vpx_codec_control_(&encoder, arg_ctrls[i][0], arg_ctrls[i][1]))
01961                 fprintf(stderr, "Error: Tried to set control %d = %d\n",
01962                         arg_ctrls[i][0], arg_ctrls[i][1]);
01963 
01964             ctx_exit_on_error(&encoder, "Failed to control codec");
01965         }
01966 
01967         frame_avail = 1;
01968         got_data = 0;
01969 
01970         while (frame_avail || got_data)
01971         {
01972             vpx_codec_iter_t iter = NULL;
01973             const vpx_codec_cx_pkt_t *pkt;
01974             struct vpx_usec_timer timer;
01975             int64_t frame_start, next_frame_start;
01976 
01977             if (!arg_limit || frames_in < arg_limit)
01978             {
01979                 frame_avail = read_frame(infile, &raw, file_type, &y4m,
01980                                          &detect);
01981 
01982                 if (frame_avail)
01983                     frames_in++;
01984 
01985                 fprintf(stderr,
01986                         "\rPass %d/%d frame %4d/%-4d %7"PRId64"B \033[K",
01987                         pass + 1, arg_passes, frames_in, frames_out, nbytes);
01988             }
01989             else
01990                 frame_avail = 0;
01991 
01992             vpx_usec_timer_start(&timer);
01993 
01994             frame_start = (cfg.g_timebase.den * (int64_t)(frames_in - 1)
01995                           * arg_framerate.den) / cfg.g_timebase.num / arg_framerate.num;
01996             next_frame_start = (cfg.g_timebase.den * (int64_t)(frames_in)
01997                                 * arg_framerate.den)
01998                                 / cfg.g_timebase.num / arg_framerate.num;
01999             vpx_codec_encode(&encoder, frame_avail ? &raw : NULL, frame_start,
02000                              next_frame_start - frame_start,
02001                              0, arg_deadline);
02002             vpx_usec_timer_mark(&timer);
02003             cx_time += vpx_usec_timer_elapsed(&timer);
02004             ctx_exit_on_error(&encoder, "Failed to encode frame");
02005 
02006             if(cfg.g_pass != VPX_RC_FIRST_PASS)
02007             {
02008                 int q;
02009 
02010                 vpx_codec_control(&encoder, VP8E_GET_LAST_QUANTIZER_64, &q);
02011                 ctx_exit_on_error(&encoder, "Failed to read quantizer");
02012                 counts[q]++;
02013             }
02014 
02015             got_data = 0;
02016 
02017             while ((pkt = vpx_codec_get_cx_data(&encoder, &iter)))
02018             {
02019                 got_data = 1;
02020 
02021                 switch (pkt->kind)
02022                 {
02023                 case VPX_CODEC_CX_FRAME_PKT:
02024                     frames_out++;
02025                     fprintf(stderr, " %6luF",
02026                             (unsigned long)pkt->data.frame.sz);
02027 
02028                     update_rate_histogram(&rate_hist, &cfg, pkt);
02029                     if(write_webm)
02030                     {
02031                         /* Update the hash */
02032                         if(!ebml.debug)
02033                             hash = murmur(pkt->data.frame.buf,
02034                                           pkt->data.frame.sz, hash);
02035 
02036                         write_webm_block(&ebml, &cfg, pkt);
02037                     }
02038                     else
02039                     {
02040                         write_ivf_frame_header(outfile, pkt);
02041                         if(fwrite(pkt->data.frame.buf, 1,
02042                                   pkt->data.frame.sz, outfile));
02043                     }
02044                     nbytes += pkt->data.raw.sz;
02045                     break;
02046                 case VPX_CODEC_STATS_PKT:
02047                     frames_out++;
02048                     fprintf(stderr, " %6luS",
02049                            (unsigned long)pkt->data.twopass_stats.sz);
02050                     stats_write(&stats,
02051                                 pkt->data.twopass_stats.buf,
02052                                 pkt->data.twopass_stats.sz);
02053                     nbytes += pkt->data.raw.sz;
02054                     break;
02055                 case VPX_CODEC_PSNR_PKT:
02056 
02057                     if (show_psnr)
02058                     {
02059                         int i;
02060 
02061                         psnr_sse_total += pkt->data.psnr.sse[0];
02062                         psnr_samples_total += pkt->data.psnr.samples[0];
02063                         for (i = 0; i < 4; i++)
02064                         {
02065                             fprintf(stderr, "%.3lf ", pkt->data.psnr.psnr[i]);
02066                             psnr_totals[i] += pkt->data.psnr.psnr[i];
02067                         }
02068                         psnr_count++;
02069                     }
02070 
02071                     break;
02072                 default:
02073                     break;
02074                 }
02075             }
02076 
02077             fflush(stdout);
02078         }
02079 
02080         fprintf(stderr,
02081                "\rPass %d/%d frame %4d/%-4d %7"PRId64"B %7lub/f %7"PRId64"b/s"
02082                " %7lu %s (%.2f fps)\033[K", pass + 1,
02083                arg_passes, frames_in, frames_out, nbytes,
02084                frames_in ? (unsigned long)(nbytes * 8 / frames_in) : 0,
02085                frames_in ? nbytes * 8 *(int64_t)arg_framerate.num / arg_framerate.den / frames_in : 0,
02086                cx_time > 9999999 ? cx_time / 1000 : cx_time,
02087                cx_time > 9999999 ? "ms" : "us",
02088                cx_time > 0 ? (float)frames_in * 1000000.0 / (float)cx_time : 0);
02089 
02090         if ( (show_psnr) && (psnr_count>0) )
02091         {
02092             int i;
02093             double ovpsnr = vp8_mse2psnr(psnr_samples_total, 255.0,
02094                                          psnr_sse_total);
02095 
02096             fprintf(stderr, "\nPSNR (Overall/Avg/Y/U/V)");
02097 
02098             fprintf(stderr, " %.3lf", ovpsnr);
02099             for (i = 0; i < 4; i++)
02100             {
02101                 fprintf(stderr, " %.3lf", psnr_totals[i]/psnr_count);
02102             }
02103         }
02104 
02105         vpx_codec_destroy(&encoder);
02106 
02107         fclose(infile);
02108         if (file_type == FILE_TYPE_Y4M)
02109             y4m_input_close(&y4m);
02110 
02111         if(write_webm)
02112         {
02113             write_webm_file_footer(&ebml, hash);
02114             free(ebml.cue_list);
02115             ebml.cue_list = NULL;
02116         }
02117         else
02118         {
02119             if (!fseek(outfile, 0, SEEK_SET))
02120                 write_ivf_file_header(outfile, &cfg, codec->fourcc, frames_out);
02121         }
02122 
02123         fclose(outfile);
02124         stats_close(&stats, arg_passes-1);
02125         fprintf(stderr, "\n");
02126 
02127         if (one_pass_only)
02128             break;
02129     }
02130 
02131     if (show_q_hist_buckets)
02132         show_q_histogram(counts, show_q_hist_buckets);
02133 
02134     if (show_rate_hist_buckets)
02135         show_rate_histogram(&rate_hist, &cfg, show_rate_hist_buckets);
02136     destroy_rate_histogram(&rate_hist);
02137 
02138     vpx_img_free(&raw);
02139     free(argv);
02140     return EXIT_SUCCESS;
02141 }

Generated on 13 May 2012 for WebM VP8 Codec SDK by  doxygen 1.6.1