WebM VP8 Codec SDK
|
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_version.h" 00036 #include "vpx/vp8cx.h" 00037 #include "vpx_ports/mem_ops.h" 00038 #include "vpx_ports/vpx_timer.h" 00039 #include "tools_common.h" 00040 #include "y4minput.h" 00041 #include "libmkv/EbmlWriter.h" 00042 #include "libmkv/EbmlIDs.h" 00043 00044 /* Need special handling of these functions on Windows */ 00045 #if defined(_MSC_VER) 00046 /* MSVS doesn't define off_t, and uses _f{seek,tell}i64 */ 00047 typedef __int64 off_t; 00048 #define fseeko _fseeki64 00049 #define ftello _ftelli64 00050 #elif defined(_WIN32) 00051 /* MinGW defines off_t, and uses f{seek,tell}o64 */ 00052 #define fseeko fseeko64 00053 #define ftello ftello64 00054 #endif 00055 00056 #if defined(_MSC_VER) 00057 #define LITERALU64(n) n 00058 #else 00059 #define LITERALU64(n) n##LLU 00060 #endif 00061 00062 /* We should use 32-bit file operations in WebM file format 00063 * when building ARM executable file (.axf) with RVCT */ 00064 #if !CONFIG_OS_SUPPORT 00065 typedef long off_t; 00066 #define fseeko fseek 00067 #define ftello ftell 00068 #endif 00069 00070 static const char *exec_name; 00071 00072 static const struct codec_item 00073 { 00074 char const *name; 00075 const vpx_codec_iface_t *iface; 00076 unsigned int fourcc; 00077 } codecs[] = 00078 { 00079 #if CONFIG_VP8_ENCODER 00080 {"vp8", &vpx_codec_vp8_cx_algo, 0x30385056}, 00081 #endif 00082 }; 00083 00084 static void usage_exit(); 00085 00086 void die(const char *fmt, ...) 00087 { 00088 va_list ap; 00089 va_start(ap, fmt); 00090 vfprintf(stderr, fmt, ap); 00091 fprintf(stderr, "\n"); 00092 usage_exit(); 00093 } 00094 00095 static void ctx_exit_on_error(vpx_codec_ctx_t *ctx, const char *s) 00096 { 00097 if (ctx->err) 00098 { 00099 const char *detail = vpx_codec_error_detail(ctx); 00100 00101 fprintf(stderr, "%s: %s\n", s, vpx_codec_error(ctx)); 00102 00103 if (detail) 00104 fprintf(stderr, " %s\n", detail); 00105 00106 exit(EXIT_FAILURE); 00107 } 00108 } 00109 00110 /* This structure is used to abstract the different ways of handling 00111 * first pass statistics. 00112 */ 00113 typedef struct 00114 { 00115 vpx_fixed_buf_t buf; 00116 int pass; 00117 FILE *file; 00118 char *buf_ptr; 00119 size_t buf_alloc_sz; 00120 } stats_io_t; 00121 00122 int stats_open_file(stats_io_t *stats, const char *fpf, int pass) 00123 { 00124 int res; 00125 00126 stats->pass = pass; 00127 00128 if (pass == 0) 00129 { 00130 stats->file = fopen(fpf, "wb"); 00131 stats->buf.sz = 0; 00132 stats->buf.buf = NULL, 00133 res = (stats->file != NULL); 00134 } 00135 else 00136 { 00137 #if 0 00138 #elif USE_POSIX_MMAP 00139 struct stat stat_buf; 00140 int fd; 00141 00142 fd = open(fpf, O_RDONLY); 00143 stats->file = fdopen(fd, "rb"); 00144 fstat(fd, &stat_buf); 00145 stats->buf.sz = stat_buf.st_size; 00146 stats->buf.buf = mmap(NULL, stats->buf.sz, PROT_READ, MAP_PRIVATE, 00147 fd, 0); 00148 res = (stats->buf.buf != NULL); 00149 #else 00150 size_t nbytes; 00151 00152 stats->file = fopen(fpf, "rb"); 00153 00154 if (fseek(stats->file, 0, SEEK_END)) 00155 { 00156 fprintf(stderr, "First-pass stats file must be seekable!\n"); 00157 exit(EXIT_FAILURE); 00158 } 00159 00160 stats->buf.sz = stats->buf_alloc_sz = ftell(stats->file); 00161 rewind(stats->file); 00162 00163 stats->buf.buf = malloc(stats->buf_alloc_sz); 00164 00165 if (!stats->buf.buf) 00166 { 00167 fprintf(stderr, "Failed to allocate first-pass stats buffer (%lu bytes)\n", 00168 (unsigned long)stats->buf_alloc_sz); 00169 exit(EXIT_FAILURE); 00170 } 00171 00172 nbytes = fread(stats->buf.buf, 1, stats->buf.sz, stats->file); 00173 res = (nbytes == stats->buf.sz); 00174 #endif 00175 } 00176 00177 return res; 00178 } 00179 00180 int stats_open_mem(stats_io_t *stats, int pass) 00181 { 00182 int res; 00183 stats->pass = pass; 00184 00185 if (!pass) 00186 { 00187 stats->buf.sz = 0; 00188 stats->buf_alloc_sz = 64 * 1024; 00189 stats->buf.buf = malloc(stats->buf_alloc_sz); 00190 } 00191 00192 stats->buf_ptr = stats->buf.buf; 00193 res = (stats->buf.buf != NULL); 00194 return res; 00195 } 00196 00197 00198 void stats_close(stats_io_t *stats, int last_pass) 00199 { 00200 if (stats->file) 00201 { 00202 if (stats->pass == last_pass) 00203 { 00204 #if 0 00205 #elif USE_POSIX_MMAP 00206 munmap(stats->buf.buf, stats->buf.sz); 00207 #else 00208 free(stats->buf.buf); 00209 #endif 00210 } 00211 00212 fclose(stats->file); 00213 stats->file = NULL; 00214 } 00215 else 00216 { 00217 if (stats->pass == last_pass) 00218 free(stats->buf.buf); 00219 } 00220 } 00221 00222 void stats_write(stats_io_t *stats, const void *pkt, size_t len) 00223 { 00224 if (stats->file) 00225 { 00226 if(fwrite(pkt, 1, len, stats->file)); 00227 } 00228 else 00229 { 00230 if (stats->buf.sz + len > stats->buf_alloc_sz) 00231 { 00232 size_t new_sz = stats->buf_alloc_sz + 64 * 1024; 00233 char *new_ptr = realloc(stats->buf.buf, new_sz); 00234 00235 if (new_ptr) 00236 { 00237 stats->buf_ptr = new_ptr + (stats->buf_ptr - (char *)stats->buf.buf); 00238 stats->buf.buf = new_ptr; 00239 stats->buf_alloc_sz = new_sz; 00240 } 00241 else 00242 { 00243 fprintf(stderr, 00244 "\nFailed to realloc firstpass stats buffer.\n"); 00245 exit(EXIT_FAILURE); 00246 } 00247 } 00248 00249 memcpy(stats->buf_ptr, pkt, len); 00250 stats->buf.sz += len; 00251 stats->buf_ptr += len; 00252 } 00253 } 00254 00255 vpx_fixed_buf_t stats_get(stats_io_t *stats) 00256 { 00257 return stats->buf; 00258 } 00259 00260 /* Stereo 3D packed frame format */ 00261 typedef enum stereo_format 00262 { 00263 STEREO_FORMAT_MONO = 0, 00264 STEREO_FORMAT_LEFT_RIGHT = 1, 00265 STEREO_FORMAT_BOTTOM_TOP = 2, 00266 STEREO_FORMAT_TOP_BOTTOM = 3, 00267 STEREO_FORMAT_RIGHT_LEFT = 11 00268 } stereo_format_t; 00269 00270 enum video_file_type 00271 { 00272 FILE_TYPE_RAW, 00273 FILE_TYPE_IVF, 00274 FILE_TYPE_Y4M 00275 }; 00276 00277 struct detect_buffer { 00278 char buf[4]; 00279 size_t buf_read; 00280 size_t position; 00281 }; 00282 00283 00284 #define IVF_FRAME_HDR_SZ (4+8) /* 4 byte size + 8 byte timestamp */ 00285 static int read_frame(FILE *f, vpx_image_t *img, unsigned int file_type, 00286 y4m_input *y4m, struct detect_buffer *detect) 00287 { 00288 int plane = 0; 00289 int shortread = 0; 00290 00291 if (file_type == FILE_TYPE_Y4M) 00292 { 00293 if (y4m_input_fetch_frame(y4m, f, img) < 1) 00294 return 0; 00295 } 00296 else 00297 { 00298 if (file_type == FILE_TYPE_IVF) 00299 { 00300 char junk[IVF_FRAME_HDR_SZ]; 00301 00302 /* Skip the frame header. We know how big the frame should be. See 00303 * write_ivf_frame_header() for documentation on the frame header 00304 * layout. 00305 */ 00306 if(fread(junk, 1, IVF_FRAME_HDR_SZ, f)); 00307 } 00308 00309 for (plane = 0; plane < 3; plane++) 00310 { 00311 unsigned char *ptr; 00312 int w = (plane ? (1 + img->d_w) / 2 : img->d_w); 00313 int h = (plane ? (1 + img->d_h) / 2 : img->d_h); 00314 int r; 00315 00316 /* Determine the correct plane based on the image format. The for-loop 00317 * always counts in Y,U,V order, but this may not match the order of 00318 * the data on disk. 00319 */ 00320 switch (plane) 00321 { 00322 case 1: 00323 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U]; 00324 break; 00325 case 2: 00326 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V]; 00327 break; 00328 default: 00329 ptr = img->planes[plane]; 00330 } 00331 00332 for (r = 0; r < h; r++) 00333 { 00334 size_t needed = w; 00335 size_t buf_position = 0; 00336 const size_t left = detect->buf_read - detect->position; 00337 if (left > 0) 00338 { 00339 const size_t more = (left < needed) ? left : needed; 00340 memcpy(ptr, detect->buf + detect->position, more); 00341 buf_position = more; 00342 needed -= more; 00343 detect->position += more; 00344 } 00345 if (needed > 0) 00346 { 00347 shortread |= (fread(ptr + buf_position, 1, needed, f) < needed); 00348 } 00349 00350 ptr += img->stride[plane]; 00351 } 00352 } 00353 } 00354 00355 return !shortread; 00356 } 00357 00358 00359 unsigned int file_is_y4m(FILE *infile, 00360 y4m_input *y4m, 00361 char detect[4]) 00362 { 00363 if(memcmp(detect, "YUV4", 4) == 0) 00364 { 00365 return 1; 00366 } 00367 return 0; 00368 } 00369 00370 #define IVF_FILE_HDR_SZ (32) 00371 unsigned int file_is_ivf(FILE *infile, 00372 unsigned int *fourcc, 00373 unsigned int *width, 00374 unsigned int *height, 00375 struct detect_buffer *detect) 00376 { 00377 char raw_hdr[IVF_FILE_HDR_SZ]; 00378 int is_ivf = 0; 00379 00380 if(memcmp(detect->buf, "DKIF", 4) != 0) 00381 return 0; 00382 00383 /* See write_ivf_file_header() for more documentation on the file header 00384 * layout. 00385 */ 00386 if (fread(raw_hdr + 4, 1, IVF_FILE_HDR_SZ - 4, infile) 00387 == IVF_FILE_HDR_SZ - 4) 00388 { 00389 { 00390 is_ivf = 1; 00391 00392 if (mem_get_le16(raw_hdr + 4) != 0) 00393 fprintf(stderr, "Error: Unrecognized IVF version! This file may not" 00394 " decode properly."); 00395 00396 *fourcc = mem_get_le32(raw_hdr + 8); 00397 } 00398 } 00399 00400 if (is_ivf) 00401 { 00402 *width = mem_get_le16(raw_hdr + 12); 00403 *height = mem_get_le16(raw_hdr + 14); 00404 detect->position = 4; 00405 } 00406 00407 return is_ivf; 00408 } 00409 00410 00411 static void write_ivf_file_header(FILE *outfile, 00412 const vpx_codec_enc_cfg_t *cfg, 00413 unsigned int fourcc, 00414 int frame_cnt) 00415 { 00416 char header[32]; 00417 00418 if (cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS) 00419 return; 00420 00421 header[0] = 'D'; 00422 header[1] = 'K'; 00423 header[2] = 'I'; 00424 header[3] = 'F'; 00425 mem_put_le16(header + 4, 0); /* version */ 00426 mem_put_le16(header + 6, 32); /* headersize */ 00427 mem_put_le32(header + 8, fourcc); /* headersize */ 00428 mem_put_le16(header + 12, cfg->g_w); /* width */ 00429 mem_put_le16(header + 14, cfg->g_h); /* height */ 00430 mem_put_le32(header + 16, cfg->g_timebase.den); /* rate */ 00431 mem_put_le32(header + 20, cfg->g_timebase.num); /* scale */ 00432 mem_put_le32(header + 24, frame_cnt); /* length */ 00433 mem_put_le32(header + 28, 0); /* unused */ 00434 00435 if(fwrite(header, 1, 32, outfile)); 00436 } 00437 00438 00439 static void write_ivf_frame_header(FILE *outfile, 00440 const vpx_codec_cx_pkt_t *pkt) 00441 { 00442 char header[12]; 00443 vpx_codec_pts_t pts; 00444 00445 if (pkt->kind != VPX_CODEC_CX_FRAME_PKT) 00446 return; 00447 00448 pts = pkt->data.frame.pts; 00449 mem_put_le32(header, pkt->data.frame.sz); 00450 mem_put_le32(header + 4, pts & 0xFFFFFFFF); 00451 mem_put_le32(header + 8, pts >> 32); 00452 00453 if(fwrite(header, 1, 12, outfile)); 00454 } 00455 00456 00457 typedef off_t EbmlLoc; 00458 00459 00460 struct cue_entry 00461 { 00462 unsigned int time; 00463 uint64_t loc; 00464 }; 00465 00466 00467 struct EbmlGlobal 00468 { 00469 int debug; 00470 00471 FILE *stream; 00472 int64_t last_pts_ms; 00473 vpx_rational_t framerate; 00474 00475 /* These pointers are to the start of an element */ 00476 off_t position_reference; 00477 off_t seek_info_pos; 00478 off_t segment_info_pos; 00479 off_t track_pos; 00480 off_t cue_pos; 00481 off_t cluster_pos; 00482 00483 /* This pointer is to a specific element to be serialized */ 00484 off_t track_id_pos; 00485 00486 /* These pointers are to the size field of the element */ 00487 EbmlLoc startSegment; 00488 EbmlLoc startCluster; 00489 00490 uint32_t cluster_timecode; 00491 int cluster_open; 00492 00493 struct cue_entry *cue_list; 00494 unsigned int cues; 00495 00496 }; 00497 00498 00499 void Ebml_Write(EbmlGlobal *glob, const void *buffer_in, unsigned long len) 00500 { 00501 if(fwrite(buffer_in, 1, len, glob->stream)); 00502 } 00503 00504 #define WRITE_BUFFER(s) \ 00505 for(i = len-1; i>=0; i--)\ 00506 { \ 00507 x = *(const s *)buffer_in >> (i * CHAR_BIT); \ 00508 Ebml_Write(glob, &x, 1); \ 00509 } 00510 void Ebml_Serialize(EbmlGlobal *glob, const void *buffer_in, int buffer_size, unsigned long len) 00511 { 00512 char x; 00513 int i; 00514 00515 /* buffer_size: 00516 * 1 - int8_t; 00517 * 2 - int16_t; 00518 * 3 - int32_t; 00519 * 4 - int64_t; 00520 */ 00521 switch (buffer_size) 00522 { 00523 case 1: 00524 WRITE_BUFFER(int8_t) 00525 break; 00526 case 2: 00527 WRITE_BUFFER(int16_t) 00528 break; 00529 case 4: 00530 WRITE_BUFFER(int32_t) 00531 break; 00532 case 8: 00533 WRITE_BUFFER(int64_t) 00534 break; 00535 default: 00536 break; 00537 } 00538 } 00539 #undef WRITE_BUFFER 00540 00541 /* Need a fixed size serializer for the track ID. libmkv provides a 64 bit 00542 * one, but not a 32 bit one. 00543 */ 00544 static void Ebml_SerializeUnsigned32(EbmlGlobal *glob, unsigned long class_id, uint64_t ui) 00545 { 00546 unsigned char sizeSerialized = 4 | 0x80; 00547 Ebml_WriteID(glob, class_id); 00548 Ebml_Serialize(glob, &sizeSerialized, sizeof(sizeSerialized), 1); 00549 Ebml_Serialize(glob, &ui, sizeof(ui), 4); 00550 } 00551 00552 00553 static void 00554 Ebml_StartSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc, 00555 unsigned long class_id) 00556 { 00557 //todo this is always taking 8 bytes, this may need later optimization 00558 //this is a key that says length unknown 00559 uint64_t unknownLen = LITERALU64(0x01FFFFFFFFFFFFFF); 00560 00561 Ebml_WriteID(glob, class_id); 00562 *ebmlLoc = ftello(glob->stream); 00563 Ebml_Serialize(glob, &unknownLen, sizeof(unknownLen), 8); 00564 } 00565 00566 static void 00567 Ebml_EndSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc) 00568 { 00569 off_t pos; 00570 uint64_t size; 00571 00572 /* Save the current stream pointer */ 00573 pos = ftello(glob->stream); 00574 00575 /* Calculate the size of this element */ 00576 size = pos - *ebmlLoc - 8; 00577 size |= LITERALU64(0x0100000000000000); 00578 00579 /* Seek back to the beginning of the element and write the new size */ 00580 fseeko(glob->stream, *ebmlLoc, SEEK_SET); 00581 Ebml_Serialize(glob, &size, sizeof(size), 8); 00582 00583 /* Reset the stream pointer */ 00584 fseeko(glob->stream, pos, SEEK_SET); 00585 } 00586 00587 00588 static void 00589 write_webm_seek_element(EbmlGlobal *ebml, unsigned long id, off_t pos) 00590 { 00591 uint64_t offset = pos - ebml->position_reference; 00592 EbmlLoc start; 00593 Ebml_StartSubElement(ebml, &start, Seek); 00594 Ebml_SerializeBinary(ebml, SeekID, id); 00595 Ebml_SerializeUnsigned64(ebml, SeekPosition, offset); 00596 Ebml_EndSubElement(ebml, &start); 00597 } 00598 00599 00600 static void 00601 write_webm_seek_info(EbmlGlobal *ebml) 00602 { 00603 00604 off_t pos; 00605 00606 /* Save the current stream pointer */ 00607 pos = ftello(ebml->stream); 00608 00609 if(ebml->seek_info_pos) 00610 fseeko(ebml->stream, ebml->seek_info_pos, SEEK_SET); 00611 else 00612 ebml->seek_info_pos = pos; 00613 00614 { 00615 EbmlLoc start; 00616 00617 Ebml_StartSubElement(ebml, &start, SeekHead); 00618 write_webm_seek_element(ebml, Tracks, ebml->track_pos); 00619 write_webm_seek_element(ebml, Cues, ebml->cue_pos); 00620 write_webm_seek_element(ebml, Info, ebml->segment_info_pos); 00621 Ebml_EndSubElement(ebml, &start); 00622 } 00623 { 00624 //segment info 00625 EbmlLoc startInfo; 00626 uint64_t frame_time; 00627 00628 frame_time = (uint64_t)1000 * ebml->framerate.den 00629 / ebml->framerate.num; 00630 ebml->segment_info_pos = ftello(ebml->stream); 00631 Ebml_StartSubElement(ebml, &startInfo, Info); 00632 Ebml_SerializeUnsigned(ebml, TimecodeScale, 1000000); 00633 Ebml_SerializeFloat(ebml, Segment_Duration, 00634 ebml->last_pts_ms + frame_time); 00635 Ebml_SerializeString(ebml, 0x4D80, 00636 ebml->debug ? "vpxenc" : "vpxenc" VERSION_STRING); 00637 Ebml_SerializeString(ebml, 0x5741, 00638 ebml->debug ? "vpxenc" : "vpxenc" VERSION_STRING); 00639 Ebml_EndSubElement(ebml, &startInfo); 00640 } 00641 } 00642 00643 00644 static void 00645 write_webm_file_header(EbmlGlobal *glob, 00646 const vpx_codec_enc_cfg_t *cfg, 00647 const struct vpx_rational *fps, 00648 stereo_format_t stereo_fmt) 00649 { 00650 { 00651 EbmlLoc start; 00652 Ebml_StartSubElement(glob, &start, EBML); 00653 Ebml_SerializeUnsigned(glob, EBMLVersion, 1); 00654 Ebml_SerializeUnsigned(glob, EBMLReadVersion, 1); //EBML Read Version 00655 Ebml_SerializeUnsigned(glob, EBMLMaxIDLength, 4); //EBML Max ID Length 00656 Ebml_SerializeUnsigned(glob, EBMLMaxSizeLength, 8); //EBML Max Size Length 00657 Ebml_SerializeString(glob, DocType, "webm"); //Doc Type 00658 Ebml_SerializeUnsigned(glob, DocTypeVersion, 2); //Doc Type Version 00659 Ebml_SerializeUnsigned(glob, DocTypeReadVersion, 2); //Doc Type Read Version 00660 Ebml_EndSubElement(glob, &start); 00661 } 00662 { 00663 Ebml_StartSubElement(glob, &glob->startSegment, Segment); //segment 00664 glob->position_reference = ftello(glob->stream); 00665 glob->framerate = *fps; 00666 write_webm_seek_info(glob); 00667 00668 { 00669 EbmlLoc trackStart; 00670 glob->track_pos = ftello(glob->stream); 00671 Ebml_StartSubElement(glob, &trackStart, Tracks); 00672 { 00673 unsigned int trackNumber = 1; 00674 uint64_t trackID = 0; 00675 00676 EbmlLoc start; 00677 Ebml_StartSubElement(glob, &start, TrackEntry); 00678 Ebml_SerializeUnsigned(glob, TrackNumber, trackNumber); 00679 glob->track_id_pos = ftello(glob->stream); 00680 Ebml_SerializeUnsigned32(glob, TrackUID, trackID); 00681 Ebml_SerializeUnsigned(glob, TrackType, 1); //video is always 1 00682 Ebml_SerializeString(glob, CodecID, "V_VP8"); 00683 { 00684 unsigned int pixelWidth = cfg->g_w; 00685 unsigned int pixelHeight = cfg->g_h; 00686 float frameRate = (float)fps->num/(float)fps->den; 00687 00688 EbmlLoc videoStart; 00689 Ebml_StartSubElement(glob, &videoStart, Video); 00690 Ebml_SerializeUnsigned(glob, PixelWidth, pixelWidth); 00691 Ebml_SerializeUnsigned(glob, PixelHeight, pixelHeight); 00692 Ebml_SerializeUnsigned(glob, StereoMode, stereo_fmt); 00693 Ebml_SerializeFloat(glob, FrameRate, frameRate); 00694 Ebml_EndSubElement(glob, &videoStart); //Video 00695 } 00696 Ebml_EndSubElement(glob, &start); //Track Entry 00697 } 00698 Ebml_EndSubElement(glob, &trackStart); 00699 } 00700 // segment element is open 00701 } 00702 } 00703 00704 00705 static void 00706 write_webm_block(EbmlGlobal *glob, 00707 const vpx_codec_enc_cfg_t *cfg, 00708 const vpx_codec_cx_pkt_t *pkt) 00709 { 00710 unsigned long block_length; 00711 unsigned char track_number; 00712 unsigned short block_timecode = 0; 00713 unsigned char flags; 00714 int64_t pts_ms; 00715 int start_cluster = 0, is_keyframe; 00716 00717 /* Calculate the PTS of this frame in milliseconds */ 00718 pts_ms = pkt->data.frame.pts * 1000 00719 * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den; 00720 if(pts_ms <= glob->last_pts_ms) 00721 pts_ms = glob->last_pts_ms + 1; 00722 glob->last_pts_ms = pts_ms; 00723 00724 /* Calculate the relative time of this block */ 00725 if(pts_ms - glob->cluster_timecode > SHRT_MAX) 00726 start_cluster = 1; 00727 else 00728 block_timecode = pts_ms - glob->cluster_timecode; 00729 00730 is_keyframe = (pkt->data.frame.flags & VPX_FRAME_IS_KEY); 00731 if(start_cluster || is_keyframe) 00732 { 00733 if(glob->cluster_open) 00734 Ebml_EndSubElement(glob, &glob->startCluster); 00735 00736 /* Open the new cluster */ 00737 block_timecode = 0; 00738 glob->cluster_open = 1; 00739 glob->cluster_timecode = pts_ms; 00740 glob->cluster_pos = ftello(glob->stream); 00741 Ebml_StartSubElement(glob, &glob->startCluster, Cluster); //cluster 00742 Ebml_SerializeUnsigned(glob, Timecode, glob->cluster_timecode); 00743 00744 /* Save a cue point if this is a keyframe. */ 00745 if(is_keyframe) 00746 { 00747 struct cue_entry *cue, *new_cue_list; 00748 00749 new_cue_list = realloc(glob->cue_list, 00750 (glob->cues+1) * sizeof(struct cue_entry)); 00751 if(new_cue_list) 00752 glob->cue_list = new_cue_list; 00753 else 00754 { 00755 fprintf(stderr, "\nFailed to realloc cue list.\n"); 00756 exit(EXIT_FAILURE); 00757 } 00758 00759 cue = &glob->cue_list[glob->cues]; 00760 cue->time = glob->cluster_timecode; 00761 cue->loc = glob->cluster_pos; 00762 glob->cues++; 00763 } 00764 } 00765 00766 /* Write the Simple Block */ 00767 Ebml_WriteID(glob, SimpleBlock); 00768 00769 block_length = pkt->data.frame.sz + 4; 00770 block_length |= 0x10000000; 00771 Ebml_Serialize(glob, &block_length, sizeof(block_length), 4); 00772 00773 track_number = 1; 00774 track_number |= 0x80; 00775 Ebml_Write(glob, &track_number, 1); 00776 00777 Ebml_Serialize(glob, &block_timecode, sizeof(block_timecode), 2); 00778 00779 flags = 0; 00780 if(is_keyframe) 00781 flags |= 0x80; 00782 if(pkt->data.frame.flags & VPX_FRAME_IS_INVISIBLE) 00783 flags |= 0x08; 00784 Ebml_Write(glob, &flags, 1); 00785 00786 Ebml_Write(glob, pkt->data.frame.buf, pkt->data.frame.sz); 00787 } 00788 00789 00790 static void 00791 write_webm_file_footer(EbmlGlobal *glob, long hash) 00792 { 00793 00794 if(glob->cluster_open) 00795 Ebml_EndSubElement(glob, &glob->startCluster); 00796 00797 { 00798 EbmlLoc start; 00799 int i; 00800 00801 glob->cue_pos = ftello(glob->stream); 00802 Ebml_StartSubElement(glob, &start, Cues); 00803 for(i=0; i<glob->cues; i++) 00804 { 00805 struct cue_entry *cue = &glob->cue_list[i]; 00806 EbmlLoc start; 00807 00808 Ebml_StartSubElement(glob, &start, CuePoint); 00809 { 00810 EbmlLoc start; 00811 00812 Ebml_SerializeUnsigned(glob, CueTime, cue->time); 00813 00814 Ebml_StartSubElement(glob, &start, CueTrackPositions); 00815 Ebml_SerializeUnsigned(glob, CueTrack, 1); 00816 Ebml_SerializeUnsigned64(glob, CueClusterPosition, 00817 cue->loc - glob->position_reference); 00818 //Ebml_SerializeUnsigned(glob, CueBlockNumber, cue->blockNumber); 00819 Ebml_EndSubElement(glob, &start); 00820 } 00821 Ebml_EndSubElement(glob, &start); 00822 } 00823 Ebml_EndSubElement(glob, &start); 00824 } 00825 00826 Ebml_EndSubElement(glob, &glob->startSegment); 00827 00828 /* Patch up the seek info block */ 00829 write_webm_seek_info(glob); 00830 00831 /* Patch up the track id */ 00832 fseeko(glob->stream, glob->track_id_pos, SEEK_SET); 00833 Ebml_SerializeUnsigned32(glob, TrackUID, glob->debug ? 0xDEADBEEF : hash); 00834 00835 fseeko(glob->stream, 0, SEEK_END); 00836 } 00837 00838 00839 /* Murmur hash derived from public domain reference implementation at 00840 * http://sites.google.com/site/murmurhash/ 00841 */ 00842 static unsigned int murmur ( const void * key, int len, unsigned int seed ) 00843 { 00844 const unsigned int m = 0x5bd1e995; 00845 const int r = 24; 00846 00847 unsigned int h = seed ^ len; 00848 00849 const unsigned char * data = (const unsigned char *)key; 00850 00851 while(len >= 4) 00852 { 00853 unsigned int k; 00854 00855 k = data[0]; 00856 k |= data[1] << 8; 00857 k |= data[2] << 16; 00858 k |= data[3] << 24; 00859 00860 k *= m; 00861 k ^= k >> r; 00862 k *= m; 00863 00864 h *= m; 00865 h ^= k; 00866 00867 data += 4; 00868 len -= 4; 00869 } 00870 00871 switch(len) 00872 { 00873 case 3: h ^= data[2] << 16; 00874 case 2: h ^= data[1] << 8; 00875 case 1: h ^= data[0]; 00876 h *= m; 00877 }; 00878 00879 h ^= h >> 13; 00880 h *= m; 00881 h ^= h >> 15; 00882 00883 return h; 00884 } 00885 00886 #include "math.h" 00887 00888 static double vp8_mse2psnr(double Samples, double Peak, double Mse) 00889 { 00890 double psnr; 00891 00892 if ((double)Mse > 0.0) 00893 psnr = 10.0 * log10(Peak * Peak * Samples / Mse); 00894 else 00895 psnr = 60; // Limit to prevent / 0 00896 00897 if (psnr > 60) 00898 psnr = 60; 00899 00900 return psnr; 00901 } 00902 00903 00904 #include "args.h" 00905 00906 static const arg_def_t debugmode = ARG_DEF("D", "debug", 0, 00907 "Debug mode (makes output deterministic)"); 00908 static const arg_def_t outputfile = ARG_DEF("o", "output", 1, 00909 "Output filename"); 00910 static const arg_def_t use_yv12 = ARG_DEF(NULL, "yv12", 0, 00911 "Input file is YV12 "); 00912 static const arg_def_t use_i420 = ARG_DEF(NULL, "i420", 0, 00913 "Input file is I420 (default)"); 00914 static const arg_def_t codecarg = ARG_DEF(NULL, "codec", 1, 00915 "Codec to use"); 00916 static const arg_def_t passes = ARG_DEF("p", "passes", 1, 00917 "Number of passes (1/2)"); 00918 static const arg_def_t pass_arg = ARG_DEF(NULL, "pass", 1, 00919 "Pass to execute (1/2)"); 00920 static const arg_def_t fpf_name = ARG_DEF(NULL, "fpf", 1, 00921 "First pass statistics file name"); 00922 static const arg_def_t limit = ARG_DEF(NULL, "limit", 1, 00923 "Stop encoding after n input frames"); 00924 static const arg_def_t deadline = ARG_DEF("d", "deadline", 1, 00925 "Deadline per frame (usec)"); 00926 static const arg_def_t best_dl = ARG_DEF(NULL, "best", 0, 00927 "Use Best Quality Deadline"); 00928 static const arg_def_t good_dl = ARG_DEF(NULL, "good", 0, 00929 "Use Good Quality Deadline"); 00930 static const arg_def_t rt_dl = ARG_DEF(NULL, "rt", 0, 00931 "Use Realtime Quality Deadline"); 00932 static const arg_def_t verbosearg = ARG_DEF("v", "verbose", 0, 00933 "Show encoder parameters"); 00934 static const arg_def_t psnrarg = ARG_DEF(NULL, "psnr", 0, 00935 "Show PSNR in status line"); 00936 static const arg_def_t framerate = ARG_DEF(NULL, "fps", 1, 00937 "Stream frame rate (rate/scale)"); 00938 static const arg_def_t use_ivf = ARG_DEF(NULL, "ivf", 0, 00939 "Output IVF (default is WebM)"); 00940 static const arg_def_t q_hist_n = ARG_DEF(NULL, "q-hist", 1, 00941 "Show quantizer histogram (n-buckets)"); 00942 static const arg_def_t rate_hist_n = ARG_DEF(NULL, "rate-hist", 1, 00943 "Show rate histogram (n-buckets)"); 00944 static const arg_def_t *main_args[] = 00945 { 00946 &debugmode, 00947 &outputfile, &codecarg, &passes, &pass_arg, &fpf_name, &limit, &deadline, 00948 &best_dl, &good_dl, &rt_dl, 00949 &verbosearg, &psnrarg, &use_ivf, &q_hist_n, &rate_hist_n, 00950 NULL 00951 }; 00952 00953 static const arg_def_t usage = ARG_DEF("u", "usage", 1, 00954 "Usage profile number to use"); 00955 static const arg_def_t threads = ARG_DEF("t", "threads", 1, 00956 "Max number of threads to use"); 00957 static const arg_def_t profile = ARG_DEF(NULL, "profile", 1, 00958 "Bitstream profile number to use"); 00959 static const arg_def_t width = ARG_DEF("w", "width", 1, 00960 "Frame width"); 00961 static const arg_def_t height = ARG_DEF("h", "height", 1, 00962 "Frame height"); 00963 static const struct arg_enum_list stereo_mode_enum[] = { 00964 {"mono" , STEREO_FORMAT_MONO}, 00965 {"left-right", STEREO_FORMAT_LEFT_RIGHT}, 00966 {"bottom-top", STEREO_FORMAT_BOTTOM_TOP}, 00967 {"top-bottom", STEREO_FORMAT_TOP_BOTTOM}, 00968 {"right-left", STEREO_FORMAT_RIGHT_LEFT}, 00969 {NULL, 0} 00970 }; 00971 static const arg_def_t stereo_mode = ARG_DEF_ENUM(NULL, "stereo-mode", 1, 00972 "Stereo 3D video format", stereo_mode_enum); 00973 static const arg_def_t timebase = ARG_DEF(NULL, "timebase", 1, 00974 "Output timestamp precision (fractional seconds)"); 00975 static const arg_def_t error_resilient = ARG_DEF(NULL, "error-resilient", 1, 00976 "Enable error resiliency features"); 00977 static const arg_def_t lag_in_frames = ARG_DEF(NULL, "lag-in-frames", 1, 00978 "Max number of frames to lag"); 00979 00980 static const arg_def_t *global_args[] = 00981 { 00982 &use_yv12, &use_i420, &usage, &threads, &profile, 00983 &width, &height, &stereo_mode, &timebase, &framerate, &error_resilient, 00984 &lag_in_frames, NULL 00985 }; 00986 00987 static const arg_def_t dropframe_thresh = ARG_DEF(NULL, "drop-frame", 1, 00988 "Temporal resampling threshold (buf %)"); 00989 static const arg_def_t resize_allowed = ARG_DEF(NULL, "resize-allowed", 1, 00990 "Spatial resampling enabled (bool)"); 00991 static const arg_def_t resize_up_thresh = ARG_DEF(NULL, "resize-up", 1, 00992 "Upscale threshold (buf %)"); 00993 static const arg_def_t resize_down_thresh = ARG_DEF(NULL, "resize-down", 1, 00994 "Downscale threshold (buf %)"); 00995 static const struct arg_enum_list end_usage_enum[] = { 00996 {"vbr", VPX_VBR}, 00997 {"cbr", VPX_CBR}, 00998 {"cq", VPX_CQ}, 00999 {NULL, 0} 01000 }; 01001 static const arg_def_t end_usage = ARG_DEF_ENUM(NULL, "end-usage", 1, 01002 "Rate control mode", end_usage_enum); 01003 static const arg_def_t target_bitrate = ARG_DEF(NULL, "target-bitrate", 1, 01004 "Bitrate (kbps)"); 01005 static const arg_def_t min_quantizer = ARG_DEF(NULL, "min-q", 1, 01006 "Minimum (best) quantizer"); 01007 static const arg_def_t max_quantizer = ARG_DEF(NULL, "max-q", 1, 01008 "Maximum (worst) quantizer"); 01009 static const arg_def_t undershoot_pct = ARG_DEF(NULL, "undershoot-pct", 1, 01010 "Datarate undershoot (min) target (%)"); 01011 static const arg_def_t overshoot_pct = ARG_DEF(NULL, "overshoot-pct", 1, 01012 "Datarate overshoot (max) target (%)"); 01013 static const arg_def_t buf_sz = ARG_DEF(NULL, "buf-sz", 1, 01014 "Client buffer size (ms)"); 01015 static const arg_def_t buf_initial_sz = ARG_DEF(NULL, "buf-initial-sz", 1, 01016 "Client initial buffer size (ms)"); 01017 static const arg_def_t buf_optimal_sz = ARG_DEF(NULL, "buf-optimal-sz", 1, 01018 "Client optimal buffer size (ms)"); 01019 static const arg_def_t *rc_args[] = 01020 { 01021 &dropframe_thresh, &resize_allowed, &resize_up_thresh, &resize_down_thresh, 01022 &end_usage, &target_bitrate, &min_quantizer, &max_quantizer, 01023 &undershoot_pct, &overshoot_pct, &buf_sz, &buf_initial_sz, &buf_optimal_sz, 01024 NULL 01025 }; 01026 01027 01028 static const arg_def_t bias_pct = ARG_DEF(NULL, "bias-pct", 1, 01029 "CBR/VBR bias (0=CBR, 100=VBR)"); 01030 static const arg_def_t minsection_pct = ARG_DEF(NULL, "minsection-pct", 1, 01031 "GOP min bitrate (% of target)"); 01032 static const arg_def_t maxsection_pct = ARG_DEF(NULL, "maxsection-pct", 1, 01033 "GOP max bitrate (% of target)"); 01034 static const arg_def_t *rc_twopass_args[] = 01035 { 01036 &bias_pct, &minsection_pct, &maxsection_pct, NULL 01037 }; 01038 01039 01040 static const arg_def_t kf_min_dist = ARG_DEF(NULL, "kf-min-dist", 1, 01041 "Minimum keyframe interval (frames)"); 01042 static const arg_def_t kf_max_dist = ARG_DEF(NULL, "kf-max-dist", 1, 01043 "Maximum keyframe interval (frames)"); 01044 static const arg_def_t kf_disabled = ARG_DEF(NULL, "disable-kf", 0, 01045 "Disable keyframe placement"); 01046 static const arg_def_t *kf_args[] = 01047 { 01048 &kf_min_dist, &kf_max_dist, &kf_disabled, NULL 01049 }; 01050 01051 01052 #if CONFIG_VP8_ENCODER 01053 static const arg_def_t noise_sens = ARG_DEF(NULL, "noise-sensitivity", 1, 01054 "Noise sensitivity (frames to blur)"); 01055 static const arg_def_t sharpness = ARG_DEF(NULL, "sharpness", 1, 01056 "Filter sharpness (0-7)"); 01057 static const arg_def_t static_thresh = ARG_DEF(NULL, "static-thresh", 1, 01058 "Motion detection threshold"); 01059 #endif 01060 01061 #if CONFIG_VP8_ENCODER 01062 static const arg_def_t cpu_used = ARG_DEF(NULL, "cpu-used", 1, 01063 "CPU Used (-16..16)"); 01064 #endif 01065 01066 01067 #if CONFIG_VP8_ENCODER 01068 static const arg_def_t token_parts = ARG_DEF(NULL, "token-parts", 1, 01069 "Number of token partitions to use, log2"); 01070 static const arg_def_t auto_altref = ARG_DEF(NULL, "auto-alt-ref", 1, 01071 "Enable automatic alt reference frames"); 01072 static const arg_def_t arnr_maxframes = ARG_DEF(NULL, "arnr-maxframes", 1, 01073 "AltRef Max Frames"); 01074 static const arg_def_t arnr_strength = ARG_DEF(NULL, "arnr-strength", 1, 01075 "AltRef Strength"); 01076 static const arg_def_t arnr_type = ARG_DEF(NULL, "arnr-type", 1, 01077 "AltRef Type"); 01078 static const struct arg_enum_list tuning_enum[] = { 01079 {"psnr", VP8_TUNE_PSNR}, 01080 {"ssim", VP8_TUNE_SSIM}, 01081 {NULL, 0} 01082 }; 01083 static const arg_def_t tune_ssim = ARG_DEF_ENUM(NULL, "tune", 1, 01084 "Material to favor", tuning_enum); 01085 static const arg_def_t cq_level = ARG_DEF(NULL, "cq-level", 1, 01086 "Constrained Quality Level"); 01087 static const arg_def_t max_intra_rate_pct = ARG_DEF(NULL, "max-intra-rate", 1, 01088 "Max I-frame bitrate (pct)"); 01089 01090 static const arg_def_t *vp8_args[] = 01091 { 01092 &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh, 01093 &token_parts, &arnr_maxframes, &arnr_strength, &arnr_type, 01094 &tune_ssim, &cq_level, &max_intra_rate_pct, NULL 01095 }; 01096 static const int vp8_arg_ctrl_map[] = 01097 { 01098 VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF, 01099 VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD, 01100 VP8E_SET_TOKEN_PARTITIONS, 01101 VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH , VP8E_SET_ARNR_TYPE, 01102 VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, VP8E_SET_MAX_INTRA_BITRATE_PCT, 0 01103 }; 01104 #endif 01105 01106 static const arg_def_t *no_args[] = { NULL }; 01107 01108 static void usage_exit() 01109 { 01110 int i; 01111 01112 fprintf(stderr, "Usage: %s <options> -o dst_filename src_filename \n", 01113 exec_name); 01114 01115 fprintf(stderr, "\nOptions:\n"); 01116 arg_show_usage(stdout, main_args); 01117 fprintf(stderr, "\nEncoder Global Options:\n"); 01118 arg_show_usage(stdout, global_args); 01119 fprintf(stderr, "\nRate Control Options:\n"); 01120 arg_show_usage(stdout, rc_args); 01121 fprintf(stderr, "\nTwopass Rate Control Options:\n"); 01122 arg_show_usage(stdout, rc_twopass_args); 01123 fprintf(stderr, "\nKeyframe Placement Options:\n"); 01124 arg_show_usage(stdout, kf_args); 01125 #if CONFIG_VP8_ENCODER 01126 fprintf(stderr, "\nVP8 Specific Options:\n"); 01127 arg_show_usage(stdout, vp8_args); 01128 #endif 01129 fprintf(stderr, "\nStream timebase (--timebase):\n" 01130 " The desired precision of timestamps in the output, expressed\n" 01131 " in fractional seconds. Default is 1/1000.\n"); 01132 fprintf(stderr, "\n" 01133 "Included encoders:\n" 01134 "\n"); 01135 01136 for (i = 0; i < sizeof(codecs) / sizeof(codecs[0]); i++) 01137 fprintf(stderr, " %-6s - %s\n", 01138 codecs[i].name, 01139 vpx_codec_iface_name(codecs[i].iface)); 01140 01141 exit(EXIT_FAILURE); 01142 } 01143 01144 01145 #define HIST_BAR_MAX 40 01146 struct hist_bucket 01147 { 01148 int low, high, count; 01149 }; 01150 01151 01152 static int merge_hist_buckets(struct hist_bucket *bucket, 01153 int *buckets_, 01154 int max_buckets) 01155 { 01156 int small_bucket = 0, merge_bucket = INT_MAX, big_bucket=0; 01157 int buckets = *buckets_; 01158 int i; 01159 01160 /* Find the extrema for this list of buckets */ 01161 big_bucket = small_bucket = 0; 01162 for(i=0; i < buckets; i++) 01163 { 01164 if(bucket[i].count < bucket[small_bucket].count) 01165 small_bucket = i; 01166 if(bucket[i].count > bucket[big_bucket].count) 01167 big_bucket = i; 01168 } 01169 01170 /* If we have too many buckets, merge the smallest with an adjacent 01171 * bucket. 01172 */ 01173 while(buckets > max_buckets) 01174 { 01175 int last_bucket = buckets - 1; 01176 01177 // merge the small bucket with an adjacent one. 01178 if(small_bucket == 0) 01179 merge_bucket = 1; 01180 else if(small_bucket == last_bucket) 01181 merge_bucket = last_bucket - 1; 01182 else if(bucket[small_bucket - 1].count < bucket[small_bucket + 1].count) 01183 merge_bucket = small_bucket - 1; 01184 else 01185 merge_bucket = small_bucket + 1; 01186 01187 assert(abs(merge_bucket - small_bucket) <= 1); 01188 assert(small_bucket < buckets); 01189 assert(big_bucket < buckets); 01190 assert(merge_bucket < buckets); 01191 01192 if(merge_bucket < small_bucket) 01193 { 01194 bucket[merge_bucket].high = bucket[small_bucket].high; 01195 bucket[merge_bucket].count += bucket[small_bucket].count; 01196 } 01197 else 01198 { 01199 bucket[small_bucket].high = bucket[merge_bucket].high; 01200 bucket[small_bucket].count += bucket[merge_bucket].count; 01201 merge_bucket = small_bucket; 01202 } 01203 01204 assert(bucket[merge_bucket].low != bucket[merge_bucket].high); 01205 01206 buckets--; 01207 01208 /* Remove the merge_bucket from the list, and find the new small 01209 * and big buckets while we're at it 01210 */ 01211 big_bucket = small_bucket = 0; 01212 for(i=0; i < buckets; i++) 01213 { 01214 if(i > merge_bucket) 01215 bucket[i] = bucket[i+1]; 01216 01217 if(bucket[i].count < bucket[small_bucket].count) 01218 small_bucket = i; 01219 if(bucket[i].count > bucket[big_bucket].count) 01220 big_bucket = i; 01221 } 01222 01223 } 01224 01225 *buckets_ = buckets; 01226 return bucket[big_bucket].count; 01227 } 01228 01229 01230 static void show_histogram(const struct hist_bucket *bucket, 01231 int buckets, 01232 int total, 01233 int scale) 01234 { 01235 const char *pat1, *pat2; 01236 int i; 01237 01238 switch((int)(log(bucket[buckets-1].high)/log(10))+1) 01239 { 01240 case 1: 01241 case 2: 01242 pat1 = "%4d %2s: "; 01243 pat2 = "%4d-%2d: "; 01244 break; 01245 case 3: 01246 pat1 = "%5d %3s: "; 01247 pat2 = "%5d-%3d: "; 01248 break; 01249 case 4: 01250 pat1 = "%6d %4s: "; 01251 pat2 = "%6d-%4d: "; 01252 break; 01253 case 5: 01254 pat1 = "%7d %5s: "; 01255 pat2 = "%7d-%5d: "; 01256 break; 01257 case 6: 01258 pat1 = "%8d %6s: "; 01259 pat2 = "%8d-%6d: "; 01260 break; 01261 case 7: 01262 pat1 = "%9d %7s: "; 01263 pat2 = "%9d-%7d: "; 01264 break; 01265 default: 01266 pat1 = "%12d %10s: "; 01267 pat2 = "%12d-%10d: "; 01268 break; 01269 } 01270 01271 for(i=0; i<buckets; i++) 01272 { 01273 int len; 01274 int j; 01275 float pct; 01276 01277 pct = 100.0 * (float)bucket[i].count / (float)total; 01278 len = HIST_BAR_MAX * bucket[i].count / scale; 01279 if(len < 1) 01280 len = 1; 01281 assert(len <= HIST_BAR_MAX); 01282 01283 if(bucket[i].low == bucket[i].high) 01284 fprintf(stderr, pat1, bucket[i].low, ""); 01285 else 01286 fprintf(stderr, pat2, bucket[i].low, bucket[i].high); 01287 01288 for(j=0; j<HIST_BAR_MAX; j++) 01289 fprintf(stderr, j<len?"=":" "); 01290 fprintf(stderr, "\t%5d (%6.2f%%)\n",bucket[i].count,pct); 01291 } 01292 } 01293 01294 01295 static void show_q_histogram(const int counts[64], int max_buckets) 01296 { 01297 struct hist_bucket bucket[64]; 01298 int buckets = 0; 01299 int total = 0; 01300 int scale; 01301 int i; 01302 01303 01304 for(i=0; i<64; i++) 01305 { 01306 if(counts[i]) 01307 { 01308 bucket[buckets].low = bucket[buckets].high = i; 01309 bucket[buckets].count = counts[i]; 01310 buckets++; 01311 total += counts[i]; 01312 } 01313 } 01314 01315 fprintf(stderr, "\nQuantizer Selection:\n"); 01316 scale = merge_hist_buckets(bucket, &buckets, max_buckets); 01317 show_histogram(bucket, buckets, total, scale); 01318 } 01319 01320 01321 #define RATE_BINS (100) 01322 struct rate_hist 01323 { 01324 int64_t *pts; 01325 int *sz; 01326 int samples; 01327 int frames; 01328 struct hist_bucket bucket[RATE_BINS]; 01329 int total; 01330 }; 01331 01332 01333 static void init_rate_histogram(struct rate_hist *hist, 01334 const vpx_codec_enc_cfg_t *cfg, 01335 const vpx_rational_t *fps) 01336 { 01337 int i; 01338 01339 /* Determine the number of samples in the buffer. Use the file's framerate 01340 * to determine the number of frames in rc_buf_sz milliseconds, with an 01341 * adjustment (5/4) to account for alt-refs 01342 */ 01343 hist->samples = cfg->rc_buf_sz * 5 / 4 * fps->num / fps->den / 1000; 01344 01345 // prevent division by zero 01346 if (hist->samples == 0) 01347 hist->samples=1; 01348 01349 hist->pts = calloc(hist->samples, sizeof(*hist->pts)); 01350 hist->sz = calloc(hist->samples, sizeof(*hist->sz)); 01351 for(i=0; i<RATE_BINS; i++) 01352 { 01353 hist->bucket[i].low = INT_MAX; 01354 hist->bucket[i].high = 0; 01355 hist->bucket[i].count = 0; 01356 } 01357 } 01358 01359 01360 static void destroy_rate_histogram(struct rate_hist *hist) 01361 { 01362 free(hist->pts); 01363 free(hist->sz); 01364 } 01365 01366 01367 static void update_rate_histogram(struct rate_hist *hist, 01368 const vpx_codec_enc_cfg_t *cfg, 01369 const vpx_codec_cx_pkt_t *pkt) 01370 { 01371 int i, idx; 01372 int64_t now, then, sum_sz = 0, avg_bitrate; 01373 01374 now = pkt->data.frame.pts * 1000 01375 * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den; 01376 01377 idx = hist->frames++ % hist->samples; 01378 hist->pts[idx] = now; 01379 hist->sz[idx] = pkt->data.frame.sz; 01380 01381 if(now < cfg->rc_buf_initial_sz) 01382 return; 01383 01384 then = now; 01385 01386 /* Sum the size over the past rc_buf_sz ms */ 01387 for(i = hist->frames; i > 0 && hist->frames - i < hist->samples; i--) 01388 { 01389 int i_idx = (i-1) % hist->samples; 01390 01391 then = hist->pts[i_idx]; 01392 if(now - then > cfg->rc_buf_sz) 01393 break; 01394 sum_sz += hist->sz[i_idx]; 01395 } 01396 01397 if (now == then) 01398 return; 01399 01400 avg_bitrate = sum_sz * 8 * 1000 / (now - then); 01401 idx = avg_bitrate * (RATE_BINS/2) / (cfg->rc_target_bitrate * 1000); 01402 if(idx < 0) 01403 idx = 0; 01404 if(idx > RATE_BINS-1) 01405 idx = RATE_BINS-1; 01406 if(hist->bucket[idx].low > avg_bitrate) 01407 hist->bucket[idx].low = avg_bitrate; 01408 if(hist->bucket[idx].high < avg_bitrate) 01409 hist->bucket[idx].high = avg_bitrate; 01410 hist->bucket[idx].count++; 01411 hist->total++; 01412 } 01413 01414 01415 static void show_rate_histogram(struct rate_hist *hist, 01416 const vpx_codec_enc_cfg_t *cfg, 01417 int max_buckets) 01418 { 01419 int i, scale; 01420 int buckets = 0; 01421 01422 for(i = 0; i < RATE_BINS; i++) 01423 { 01424 if(hist->bucket[i].low == INT_MAX) 01425 continue; 01426 hist->bucket[buckets++] = hist->bucket[i]; 01427 } 01428 01429 fprintf(stderr, "\nRate (over %dms window):\n", cfg->rc_buf_sz); 01430 scale = merge_hist_buckets(hist->bucket, &buckets, max_buckets); 01431 show_histogram(hist->bucket, buckets, hist->total, scale); 01432 } 01433 01434 #define ARG_CTRL_CNT_MAX 10 01435 01436 int main(int argc, const char **argv_) 01437 { 01438 vpx_codec_ctx_t encoder; 01439 const char *in_fn = NULL, *out_fn = NULL, *stats_fn = NULL; 01440 int i; 01441 FILE *infile, *outfile; 01442 vpx_codec_enc_cfg_t cfg; 01443 vpx_codec_err_t res; 01444 int pass, one_pass_only = 0; 01445 stats_io_t stats; 01446 vpx_image_t raw; 01447 const struct codec_item *codec = codecs; 01448 int frame_avail, got_data; 01449 01450 struct arg arg; 01451 char **argv, **argi, **argj; 01452 int arg_usage = 0, arg_passes = 1, arg_deadline = 0; 01453 int arg_ctrls[ARG_CTRL_CNT_MAX][2], arg_ctrl_cnt = 0; 01454 int arg_limit = 0; 01455 static const arg_def_t **ctrl_args = no_args; 01456 static const int *ctrl_args_map = NULL; 01457 int verbose = 0, show_psnr = 0; 01458 int arg_use_i420 = 1; 01459 unsigned long cx_time = 0; 01460 unsigned int file_type, fourcc; 01461 y4m_input y4m; 01462 struct vpx_rational arg_framerate = {30, 1}; 01463 int arg_have_framerate = 0; 01464 int write_webm = 1; 01465 EbmlGlobal ebml = {0}; 01466 uint32_t hash = 0; 01467 uint64_t psnr_sse_total = 0; 01468 uint64_t psnr_samples_total = 0; 01469 double psnr_totals[4] = {0, 0, 0, 0}; 01470 int psnr_count = 0; 01471 stereo_format_t stereo_fmt = STEREO_FORMAT_MONO; 01472 int counts[64]={0}; 01473 int show_q_hist_buckets=0; 01474 int show_rate_hist_buckets=0; 01475 struct rate_hist rate_hist={0}; 01476 01477 exec_name = argv_[0]; 01478 ebml.last_pts_ms = -1; 01479 01480 if (argc < 3) 01481 usage_exit(); 01482 01483 01484 /* First parse the codec and usage values, because we want to apply other 01485 * parameters on top of the default configuration provided by the codec. 01486 */ 01487 argv = argv_dup(argc - 1, argv_ + 1); 01488 01489 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) 01490 { 01491 arg.argv_step = 1; 01492 01493 if (arg_match(&arg, &codecarg, argi)) 01494 { 01495 int j, k = -1; 01496 01497 for (j = 0; j < sizeof(codecs) / sizeof(codecs[0]); j++) 01498 if (!strcmp(codecs[j].name, arg.val)) 01499 k = j; 01500 01501 if (k >= 0) 01502 codec = codecs + k; 01503 else 01504 die("Error: Unrecognized argument (%s) to --codec\n", 01505 arg.val); 01506 01507 } 01508 else if (arg_match(&arg, &passes, argi)) 01509 { 01510 arg_passes = arg_parse_uint(&arg); 01511 01512 if (arg_passes < 1 || arg_passes > 2) 01513 die("Error: Invalid number of passes (%d)\n", arg_passes); 01514 } 01515 else if (arg_match(&arg, &pass_arg, argi)) 01516 { 01517 one_pass_only = arg_parse_uint(&arg); 01518 01519 if (one_pass_only < 1 || one_pass_only > 2) 01520 die("Error: Invalid pass selected (%d)\n", one_pass_only); 01521 } 01522 else if (arg_match(&arg, &fpf_name, argi)) 01523 stats_fn = arg.val; 01524 else if (arg_match(&arg, &usage, argi)) 01525 arg_usage = arg_parse_uint(&arg); 01526 else if (arg_match(&arg, &deadline, argi)) 01527 arg_deadline = arg_parse_uint(&arg); 01528 else if (arg_match(&arg, &best_dl, argi)) 01529 arg_deadline = VPX_DL_BEST_QUALITY; 01530 else if (arg_match(&arg, &good_dl, argi)) 01531 arg_deadline = VPX_DL_GOOD_QUALITY; 01532 else if (arg_match(&arg, &rt_dl, argi)) 01533 arg_deadline = VPX_DL_REALTIME; 01534 else if (arg_match(&arg, &use_yv12, argi)) 01535 { 01536 arg_use_i420 = 0; 01537 } 01538 else if (arg_match(&arg, &use_i420, argi)) 01539 { 01540 arg_use_i420 = 1; 01541 } 01542 else if (arg_match(&arg, &verbosearg, argi)) 01543 verbose = 1; 01544 else if (arg_match(&arg, &limit, argi)) 01545 arg_limit = arg_parse_uint(&arg); 01546 else if (arg_match(&arg, &psnrarg, argi)) 01547 show_psnr = 1; 01548 else if (arg_match(&arg, &framerate, argi)) 01549 { 01550 arg_framerate = arg_parse_rational(&arg); 01551 arg_have_framerate = 1; 01552 } 01553 else if (arg_match(&arg, &use_ivf, argi)) 01554 write_webm = 0; 01555 else if (arg_match(&arg, &outputfile, argi)) 01556 out_fn = arg.val; 01557 else if (arg_match(&arg, &debugmode, argi)) 01558 ebml.debug = 1; 01559 else if (arg_match(&arg, &q_hist_n, argi)) 01560 show_q_hist_buckets = arg_parse_uint(&arg); 01561 else if (arg_match(&arg, &rate_hist_n, argi)) 01562 show_rate_hist_buckets = arg_parse_uint(&arg); 01563 else 01564 argj++; 01565 } 01566 01567 /* Ensure that --passes and --pass are consistent. If --pass is set and --passes=2, 01568 * ensure --fpf was set. 01569 */ 01570 if (one_pass_only) 01571 { 01572 /* DWIM: Assume the user meant passes=2 if pass=2 is specified */ 01573 if (one_pass_only > arg_passes) 01574 { 01575 fprintf(stderr, "Warning: Assuming --pass=%d implies --passes=%d\n", 01576 one_pass_only, one_pass_only); 01577 arg_passes = one_pass_only; 01578 } 01579 01580 if (arg_passes == 2 && !stats_fn) 01581 die("Must specify --fpf when --pass=%d and --passes=2\n", one_pass_only); 01582 } 01583 01584 /* Populate encoder configuration */ 01585 res = vpx_codec_enc_config_default(codec->iface, &cfg, arg_usage); 01586 01587 if (res) 01588 { 01589 fprintf(stderr, "Failed to get config: %s\n", 01590 vpx_codec_err_to_string(res)); 01591 return EXIT_FAILURE; 01592 } 01593 01594 /* Change the default timebase to a high enough value so that the encoder 01595 * will always create strictly increasing timestamps. 01596 */ 01597 cfg.g_timebase.den = 1000; 01598 01599 /* Never use the library's default resolution, require it be parsed 01600 * from the file or set on the command line. 01601 */ 01602 cfg.g_w = 0; 01603 cfg.g_h = 0; 01604 01605 /* Now parse the remainder of the parameters. */ 01606 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) 01607 { 01608 arg.argv_step = 1; 01609 01610 if (0); 01611 else if (arg_match(&arg, &threads, argi)) 01612 cfg.g_threads = arg_parse_uint(&arg); 01613 else if (arg_match(&arg, &profile, argi)) 01614 cfg.g_profile = arg_parse_uint(&arg); 01615 else if (arg_match(&arg, &width, argi)) 01616 cfg.g_w = arg_parse_uint(&arg); 01617 else if (arg_match(&arg, &height, argi)) 01618 cfg.g_h = arg_parse_uint(&arg); 01619 else if (arg_match(&arg, &stereo_mode, argi)) 01620 stereo_fmt = arg_parse_enum_or_int(&arg); 01621 else if (arg_match(&arg, &timebase, argi)) 01622 cfg.g_timebase = arg_parse_rational(&arg); 01623 else if (arg_match(&arg, &error_resilient, argi)) 01624 cfg.g_error_resilient = arg_parse_uint(&arg); 01625 else if (arg_match(&arg, &lag_in_frames, argi)) 01626 cfg.g_lag_in_frames = arg_parse_uint(&arg); 01627 else if (arg_match(&arg, &dropframe_thresh, argi)) 01628 cfg.rc_dropframe_thresh = arg_parse_uint(&arg); 01629 else if (arg_match(&arg, &resize_allowed, argi)) 01630 cfg.rc_resize_allowed = arg_parse_uint(&arg); 01631 else if (arg_match(&arg, &resize_up_thresh, argi)) 01632 cfg.rc_resize_up_thresh = arg_parse_uint(&arg); 01633 else if (arg_match(&arg, &resize_down_thresh, argi)) 01634 cfg.rc_resize_down_thresh = arg_parse_uint(&arg); 01635 else if (arg_match(&arg, &resize_down_thresh, argi)) 01636 cfg.rc_resize_down_thresh = arg_parse_uint(&arg); 01637 else if (arg_match(&arg, &end_usage, argi)) 01638 cfg.rc_end_usage = arg_parse_enum_or_int(&arg); 01639 else if (arg_match(&arg, &target_bitrate, argi)) 01640 cfg.rc_target_bitrate = arg_parse_uint(&arg); 01641 else if (arg_match(&arg, &min_quantizer, argi)) 01642 cfg.rc_min_quantizer = arg_parse_uint(&arg); 01643 else if (arg_match(&arg, &max_quantizer, argi)) 01644 cfg.rc_max_quantizer = arg_parse_uint(&arg); 01645 else if (arg_match(&arg, &undershoot_pct, argi)) 01646 cfg.rc_undershoot_pct = arg_parse_uint(&arg); 01647 else if (arg_match(&arg, &overshoot_pct, argi)) 01648 cfg.rc_overshoot_pct = arg_parse_uint(&arg); 01649 else if (arg_match(&arg, &buf_sz, argi)) 01650 cfg.rc_buf_sz = arg_parse_uint(&arg); 01651 else if (arg_match(&arg, &buf_initial_sz, argi)) 01652 cfg.rc_buf_initial_sz = arg_parse_uint(&arg); 01653 else if (arg_match(&arg, &buf_optimal_sz, argi)) 01654 cfg.rc_buf_optimal_sz = arg_parse_uint(&arg); 01655 else if (arg_match(&arg, &bias_pct, argi)) 01656 { 01657 cfg.rc_2pass_vbr_bias_pct = arg_parse_uint(&arg); 01658 01659 if (arg_passes < 2) 01660 fprintf(stderr, 01661 "Warning: option %s ignored in one-pass mode.\n", 01662 arg.name); 01663 } 01664 else if (arg_match(&arg, &minsection_pct, argi)) 01665 { 01666 cfg.rc_2pass_vbr_minsection_pct = arg_parse_uint(&arg); 01667 01668 if (arg_passes < 2) 01669 fprintf(stderr, 01670 "Warning: option %s ignored in one-pass mode.\n", 01671 arg.name); 01672 } 01673 else if (arg_match(&arg, &maxsection_pct, argi)) 01674 { 01675 cfg.rc_2pass_vbr_maxsection_pct = arg_parse_uint(&arg); 01676 01677 if (arg_passes < 2) 01678 fprintf(stderr, 01679 "Warning: option %s ignored in one-pass mode.\n", 01680 arg.name); 01681 } 01682 else if (arg_match(&arg, &kf_min_dist, argi)) 01683 cfg.kf_min_dist = arg_parse_uint(&arg); 01684 else if (arg_match(&arg, &kf_max_dist, argi)) 01685 cfg.kf_max_dist = arg_parse_uint(&arg); 01686 else if (arg_match(&arg, &kf_disabled, argi)) 01687 cfg.kf_mode = VPX_KF_DISABLED; 01688 else 01689 argj++; 01690 } 01691 01692 /* Handle codec specific options */ 01693 #if CONFIG_VP8_ENCODER 01694 01695 if (codec->iface == &vpx_codec_vp8_cx_algo) 01696 { 01697 ctrl_args = vp8_args; 01698 ctrl_args_map = vp8_arg_ctrl_map; 01699 } 01700 01701 #endif 01702 01703 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) 01704 { 01705 int match = 0; 01706 01707 arg.argv_step = 1; 01708 01709 for (i = 0; ctrl_args[i]; i++) 01710 { 01711 if (arg_match(&arg, ctrl_args[i], argi)) 01712 { 01713 match = 1; 01714 01715 if (arg_ctrl_cnt < ARG_CTRL_CNT_MAX) 01716 { 01717 arg_ctrls[arg_ctrl_cnt][0] = ctrl_args_map[i]; 01718 arg_ctrls[arg_ctrl_cnt][1] = arg_parse_enum_or_int(&arg); 01719 arg_ctrl_cnt++; 01720 } 01721 } 01722 } 01723 01724 if (!match) 01725 argj++; 01726 } 01727 01728 /* Check for unrecognized options */ 01729 for (argi = argv; *argi; argi++) 01730 if (argi[0][0] == '-' && argi[0][1]) 01731 die("Error: Unrecognized option %s\n", *argi); 01732 01733 /* Handle non-option arguments */ 01734 in_fn = argv[0]; 01735 01736 if (!in_fn) 01737 usage_exit(); 01738 01739 if(!out_fn) 01740 die("Error: Output file is required (specify with -o)\n"); 01741 01742 memset(&stats, 0, sizeof(stats)); 01743 01744 for (pass = one_pass_only ? one_pass_only - 1 : 0; pass < arg_passes; pass++) 01745 { 01746 int frames_in = 0, frames_out = 0; 01747 unsigned long nbytes = 0; 01748 struct detect_buffer detect; 01749 01750 /* Parse certain options from the input file, if possible */ 01751 infile = strcmp(in_fn, "-") ? fopen(in_fn, "rb") 01752 : set_binary_mode(stdin); 01753 01754 if (!infile) 01755 { 01756 fprintf(stderr, "Failed to open input file\n"); 01757 return EXIT_FAILURE; 01758 } 01759 01760 /* For RAW input sources, these bytes will applied on the first frame 01761 * in read_frame(). 01762 */ 01763 detect.buf_read = fread(detect.buf, 1, 4, infile); 01764 detect.position = 0; 01765 01766 if (detect.buf_read == 4 && file_is_y4m(infile, &y4m, detect.buf)) 01767 { 01768 if (y4m_input_open(&y4m, infile, detect.buf, 4) >= 0) 01769 { 01770 file_type = FILE_TYPE_Y4M; 01771 cfg.g_w = y4m.pic_w; 01772 cfg.g_h = y4m.pic_h; 01773 01774 /* Use the frame rate from the file only if none was specified 01775 * on the command-line. 01776 */ 01777 if (!arg_have_framerate) 01778 { 01779 arg_framerate.num = y4m.fps_n; 01780 arg_framerate.den = y4m.fps_d; 01781 } 01782 01783 arg_use_i420 = 0; 01784 } 01785 else 01786 { 01787 fprintf(stderr, "Unsupported Y4M stream.\n"); 01788 return EXIT_FAILURE; 01789 } 01790 } 01791 else if (detect.buf_read == 4 && 01792 file_is_ivf(infile, &fourcc, &cfg.g_w, &cfg.g_h, &detect)) 01793 { 01794 file_type = FILE_TYPE_IVF; 01795 switch (fourcc) 01796 { 01797 case 0x32315659: 01798 arg_use_i420 = 0; 01799 break; 01800 case 0x30323449: 01801 arg_use_i420 = 1; 01802 break; 01803 default: 01804 fprintf(stderr, "Unsupported fourcc (%08x) in IVF\n", fourcc); 01805 return EXIT_FAILURE; 01806 } 01807 } 01808 else 01809 { 01810 file_type = FILE_TYPE_RAW; 01811 } 01812 01813 if(!cfg.g_w || !cfg.g_h) 01814 { 01815 fprintf(stderr, "Specify stream dimensions with --width (-w) " 01816 " and --height (-h).\n"); 01817 return EXIT_FAILURE; 01818 } 01819 01820 #define SHOW(field) fprintf(stderr, " %-28s = %d\n", #field, cfg.field) 01821 01822 if (verbose && pass == 0) 01823 { 01824 fprintf(stderr, "Codec: %s\n", vpx_codec_iface_name(codec->iface)); 01825 fprintf(stderr, "Source file: %s Format: %s\n", in_fn, 01826 arg_use_i420 ? "I420" : "YV12"); 01827 fprintf(stderr, "Destination file: %s\n", out_fn); 01828 fprintf(stderr, "Encoder parameters:\n"); 01829 01830 SHOW(g_usage); 01831 SHOW(g_threads); 01832 SHOW(g_profile); 01833 SHOW(g_w); 01834 SHOW(g_h); 01835 SHOW(g_timebase.num); 01836 SHOW(g_timebase.den); 01837 SHOW(g_error_resilient); 01838 SHOW(g_pass); 01839 SHOW(g_lag_in_frames); 01840 SHOW(rc_dropframe_thresh); 01841 SHOW(rc_resize_allowed); 01842 SHOW(rc_resize_up_thresh); 01843 SHOW(rc_resize_down_thresh); 01844 SHOW(rc_end_usage); 01845 SHOW(rc_target_bitrate); 01846 SHOW(rc_min_quantizer); 01847 SHOW(rc_max_quantizer); 01848 SHOW(rc_undershoot_pct); 01849 SHOW(rc_overshoot_pct); 01850 SHOW(rc_buf_sz); 01851 SHOW(rc_buf_initial_sz); 01852 SHOW(rc_buf_optimal_sz); 01853 SHOW(rc_2pass_vbr_bias_pct); 01854 SHOW(rc_2pass_vbr_minsection_pct); 01855 SHOW(rc_2pass_vbr_maxsection_pct); 01856 SHOW(kf_mode); 01857 SHOW(kf_min_dist); 01858 SHOW(kf_max_dist); 01859 } 01860 01861 if(pass == (one_pass_only ? one_pass_only - 1 : 0)) { 01862 if (file_type == FILE_TYPE_Y4M) 01863 /*The Y4M reader does its own allocation. 01864 Just initialize this here to avoid problems if we never read any 01865 frames.*/ 01866 memset(&raw, 0, sizeof(raw)); 01867 else 01868 vpx_img_alloc(&raw, arg_use_i420 ? VPX_IMG_FMT_I420 : VPX_IMG_FMT_YV12, 01869 cfg.g_w, cfg.g_h, 1); 01870 01871 init_rate_histogram(&rate_hist, &cfg, &arg_framerate); 01872 } 01873 01874 outfile = strcmp(out_fn, "-") ? fopen(out_fn, "wb") 01875 : set_binary_mode(stdout); 01876 01877 if (!outfile) 01878 { 01879 fprintf(stderr, "Failed to open output file\n"); 01880 return EXIT_FAILURE; 01881 } 01882 01883 if(write_webm && fseek(outfile, 0, SEEK_CUR)) 01884 { 01885 fprintf(stderr, "WebM output to pipes not supported.\n"); 01886 return EXIT_FAILURE; 01887 } 01888 01889 if (stats_fn) 01890 { 01891 if (!stats_open_file(&stats, stats_fn, pass)) 01892 { 01893 fprintf(stderr, "Failed to open statistics store\n"); 01894 return EXIT_FAILURE; 01895 } 01896 } 01897 else 01898 { 01899 if (!stats_open_mem(&stats, pass)) 01900 { 01901 fprintf(stderr, "Failed to open statistics store\n"); 01902 return EXIT_FAILURE; 01903 } 01904 } 01905 01906 cfg.g_pass = arg_passes == 2 01907 ? pass ? VPX_RC_LAST_PASS : VPX_RC_FIRST_PASS 01908 : VPX_RC_ONE_PASS; 01909 #if VPX_ENCODER_ABI_VERSION > (1 + VPX_CODEC_ABI_VERSION) 01910 01911 if (pass) 01912 { 01913 cfg.rc_twopass_stats_in = stats_get(&stats); 01914 } 01915 01916 #endif 01917 01918 if(write_webm) 01919 { 01920 ebml.stream = outfile; 01921 write_webm_file_header(&ebml, &cfg, &arg_framerate, stereo_fmt); 01922 } 01923 else 01924 write_ivf_file_header(outfile, &cfg, codec->fourcc, 0); 01925 01926 01927 /* Construct Encoder Context */ 01928 vpx_codec_enc_init(&encoder, codec->iface, &cfg, 01929 show_psnr ? VPX_CODEC_USE_PSNR : 0); 01930 ctx_exit_on_error(&encoder, "Failed to initialize encoder"); 01931 01932 /* Note that we bypass the vpx_codec_control wrapper macro because 01933 * we're being clever to store the control IDs in an array. Real 01934 * applications will want to make use of the enumerations directly 01935 */ 01936 for (i = 0; i < arg_ctrl_cnt; i++) 01937 { 01938 if (vpx_codec_control_(&encoder, arg_ctrls[i][0], arg_ctrls[i][1])) 01939 fprintf(stderr, "Error: Tried to set control %d = %d\n", 01940 arg_ctrls[i][0], arg_ctrls[i][1]); 01941 01942 ctx_exit_on_error(&encoder, "Failed to control codec"); 01943 } 01944 01945 frame_avail = 1; 01946 got_data = 0; 01947 01948 while (frame_avail || got_data) 01949 { 01950 vpx_codec_iter_t iter = NULL; 01951 const vpx_codec_cx_pkt_t *pkt; 01952 struct vpx_usec_timer timer; 01953 int64_t frame_start, next_frame_start; 01954 01955 if (!arg_limit || frames_in < arg_limit) 01956 { 01957 frame_avail = read_frame(infile, &raw, file_type, &y4m, 01958 &detect); 01959 01960 if (frame_avail) 01961 frames_in++; 01962 01963 fprintf(stderr, 01964 "\rPass %d/%d frame %4d/%-4d %7ldB \033[K", pass + 1, 01965 arg_passes, frames_in, frames_out, nbytes); 01966 } 01967 else 01968 frame_avail = 0; 01969 01970 vpx_usec_timer_start(&timer); 01971 01972 frame_start = (cfg.g_timebase.den * (int64_t)(frames_in - 1) 01973 * arg_framerate.den) / cfg.g_timebase.num / arg_framerate.num; 01974 next_frame_start = (cfg.g_timebase.den * (int64_t)(frames_in) 01975 * arg_framerate.den) 01976 / cfg.g_timebase.num / arg_framerate.num; 01977 vpx_codec_encode(&encoder, frame_avail ? &raw : NULL, frame_start, 01978 next_frame_start - frame_start, 01979 0, arg_deadline); 01980 vpx_usec_timer_mark(&timer); 01981 cx_time += vpx_usec_timer_elapsed(&timer); 01982 ctx_exit_on_error(&encoder, "Failed to encode frame"); 01983 01984 if(cfg.g_pass != VPX_RC_FIRST_PASS) 01985 { 01986 int q; 01987 01988 vpx_codec_control(&encoder, VP8E_GET_LAST_QUANTIZER_64, &q); 01989 ctx_exit_on_error(&encoder, "Failed to read quantizer"); 01990 counts[q]++; 01991 } 01992 01993 got_data = 0; 01994 01995 while ((pkt = vpx_codec_get_cx_data(&encoder, &iter))) 01996 { 01997 got_data = 1; 01998 01999 switch (pkt->kind) 02000 { 02001 case VPX_CODEC_CX_FRAME_PKT: 02002 frames_out++; 02003 fprintf(stderr, " %6luF", 02004 (unsigned long)pkt->data.frame.sz); 02005 02006 update_rate_histogram(&rate_hist, &cfg, pkt); 02007 if(write_webm) 02008 { 02009 /* Update the hash */ 02010 if(!ebml.debug) 02011 hash = murmur(pkt->data.frame.buf, 02012 pkt->data.frame.sz, hash); 02013 02014 write_webm_block(&ebml, &cfg, pkt); 02015 } 02016 else 02017 { 02018 write_ivf_frame_header(outfile, pkt); 02019 if(fwrite(pkt->data.frame.buf, 1, 02020 pkt->data.frame.sz, outfile)); 02021 } 02022 nbytes += pkt->data.raw.sz; 02023 break; 02024 case VPX_CODEC_STATS_PKT: 02025 frames_out++; 02026 fprintf(stderr, " %6luS", 02027 (unsigned long)pkt->data.twopass_stats.sz); 02028 stats_write(&stats, 02029 pkt->data.twopass_stats.buf, 02030 pkt->data.twopass_stats.sz); 02031 nbytes += pkt->data.raw.sz; 02032 break; 02033 case VPX_CODEC_PSNR_PKT: 02034 02035 if (show_psnr) 02036 { 02037 int i; 02038 02039 psnr_sse_total += pkt->data.psnr.sse[0]; 02040 psnr_samples_total += pkt->data.psnr.samples[0]; 02041 for (i = 0; i < 4; i++) 02042 { 02043 fprintf(stderr, "%.3lf ", pkt->data.psnr.psnr[i]); 02044 psnr_totals[i] += pkt->data.psnr.psnr[i]; 02045 } 02046 psnr_count++; 02047 } 02048 02049 break; 02050 default: 02051 break; 02052 } 02053 } 02054 02055 fflush(stdout); 02056 } 02057 02058 fprintf(stderr, 02059 "\rPass %d/%d frame %4d/%-4d %7ldB %7ldb/f %7"PRId64"b/s" 02060 " %7lu %s (%.2f fps)\033[K", pass + 1, 02061 arg_passes, frames_in, frames_out, nbytes, nbytes * 8 / frames_in, 02062 nbytes * 8 *(int64_t)arg_framerate.num / arg_framerate.den / frames_in, 02063 cx_time > 9999999 ? cx_time / 1000 : cx_time, 02064 cx_time > 9999999 ? "ms" : "us", 02065 (float)frames_in * 1000000.0 / (float)cx_time); 02066 02067 if ( (show_psnr) && (psnr_count>0) ) 02068 { 02069 int i; 02070 double ovpsnr = vp8_mse2psnr(psnr_samples_total, 255.0, 02071 psnr_sse_total); 02072 02073 fprintf(stderr, "\nPSNR (Overall/Avg/Y/U/V)"); 02074 02075 fprintf(stderr, " %.3lf", ovpsnr); 02076 for (i = 0; i < 4; i++) 02077 { 02078 fprintf(stderr, " %.3lf", psnr_totals[i]/psnr_count); 02079 } 02080 } 02081 02082 vpx_codec_destroy(&encoder); 02083 02084 fclose(infile); 02085 if (file_type == FILE_TYPE_Y4M) 02086 y4m_input_close(&y4m); 02087 02088 if(write_webm) 02089 { 02090 write_webm_file_footer(&ebml, hash); 02091 free(ebml.cue_list); 02092 ebml.cue_list = NULL; 02093 } 02094 else 02095 { 02096 if (!fseek(outfile, 0, SEEK_SET)) 02097 write_ivf_file_header(outfile, &cfg, codec->fourcc, frames_out); 02098 } 02099 02100 fclose(outfile); 02101 stats_close(&stats, arg_passes-1); 02102 fprintf(stderr, "\n"); 02103 02104 if (one_pass_only) 02105 break; 02106 } 02107 02108 if (show_q_hist_buckets) 02109 show_q_histogram(counts, show_q_hist_buckets); 02110 02111 if (show_rate_hist_buckets) 02112 show_rate_histogram(&rate_hist, &cfg, show_rate_hist_buckets); 02113 destroy_rate_histogram(&rate_hist); 02114 02115 vpx_img_free(&raw); 02116 free(argv); 02117 return EXIT_SUCCESS; 02118 }