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/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 #define LOG_ERROR(label) do \ 00088 {\ 00089 const char *l=label;\ 00090 va_list ap;\ 00091 va_start(ap, fmt);\ 00092 if(l)\ 00093 fprintf(stderr, "%s: ", l);\ 00094 vfprintf(stderr, fmt, ap);\ 00095 fprintf(stderr, "\n");\ 00096 va_end(ap);\ 00097 } while(0) 00098 00099 void die(const char *fmt, ...) 00100 { 00101 LOG_ERROR(NULL); 00102 usage_exit(); 00103 } 00104 00105 00106 void fatal(const char *fmt, ...) 00107 { 00108 LOG_ERROR("Fatal"); 00109 exit(EXIT_FAILURE); 00110 } 00111 00112 00113 void warn(const char *fmt, ...) 00114 { 00115 LOG_ERROR("Warning"); 00116 } 00117 00118 00119 static void ctx_exit_on_error(vpx_codec_ctx_t *ctx, const char *s, ...) 00120 { 00121 va_list ap; 00122 00123 va_start(ap, s); 00124 if (ctx->err) 00125 { 00126 const char *detail = vpx_codec_error_detail(ctx); 00127 00128 vfprintf(stderr, s, ap); 00129 fprintf(stderr, ": %s\n", vpx_codec_error(ctx)); 00130 00131 if (detail) 00132 fprintf(stderr, " %s\n", detail); 00133 00134 exit(EXIT_FAILURE); 00135 } 00136 } 00137 00138 /* This structure is used to abstract the different ways of handling 00139 * first pass statistics. 00140 */ 00141 typedef struct 00142 { 00143 vpx_fixed_buf_t buf; 00144 int pass; 00145 FILE *file; 00146 char *buf_ptr; 00147 size_t buf_alloc_sz; 00148 } stats_io_t; 00149 00150 int stats_open_file(stats_io_t *stats, const char *fpf, int pass) 00151 { 00152 int res; 00153 00154 stats->pass = pass; 00155 00156 if (pass == 0) 00157 { 00158 stats->file = fopen(fpf, "wb"); 00159 stats->buf.sz = 0; 00160 stats->buf.buf = NULL, 00161 res = (stats->file != NULL); 00162 } 00163 else 00164 { 00165 #if 0 00166 #elif USE_POSIX_MMAP 00167 struct stat stat_buf; 00168 int fd; 00169 00170 fd = open(fpf, O_RDONLY); 00171 stats->file = fdopen(fd, "rb"); 00172 fstat(fd, &stat_buf); 00173 stats->buf.sz = stat_buf.st_size; 00174 stats->buf.buf = mmap(NULL, stats->buf.sz, PROT_READ, MAP_PRIVATE, 00175 fd, 0); 00176 res = (stats->buf.buf != NULL); 00177 #else 00178 size_t nbytes; 00179 00180 stats->file = fopen(fpf, "rb"); 00181 00182 if (fseek(stats->file, 0, SEEK_END)) 00183 fatal("First-pass stats file must be seekable!"); 00184 00185 stats->buf.sz = stats->buf_alloc_sz = ftell(stats->file); 00186 rewind(stats->file); 00187 00188 stats->buf.buf = malloc(stats->buf_alloc_sz); 00189 00190 if (!stats->buf.buf) 00191 fatal("Failed to allocate first-pass stats buffer (%lu bytes)", 00192 (unsigned long)stats->buf_alloc_sz); 00193 00194 nbytes = fread(stats->buf.buf, 1, stats->buf.sz, stats->file); 00195 res = (nbytes == stats->buf.sz); 00196 #endif 00197 } 00198 00199 return res; 00200 } 00201 00202 int stats_open_mem(stats_io_t *stats, int pass) 00203 { 00204 int res; 00205 stats->pass = pass; 00206 00207 if (!pass) 00208 { 00209 stats->buf.sz = 0; 00210 stats->buf_alloc_sz = 64 * 1024; 00211 stats->buf.buf = malloc(stats->buf_alloc_sz); 00212 } 00213 00214 stats->buf_ptr = stats->buf.buf; 00215 res = (stats->buf.buf != NULL); 00216 return res; 00217 } 00218 00219 00220 void stats_close(stats_io_t *stats, int last_pass) 00221 { 00222 if (stats->file) 00223 { 00224 if (stats->pass == last_pass) 00225 { 00226 #if 0 00227 #elif USE_POSIX_MMAP 00228 munmap(stats->buf.buf, stats->buf.sz); 00229 #else 00230 free(stats->buf.buf); 00231 #endif 00232 } 00233 00234 fclose(stats->file); 00235 stats->file = NULL; 00236 } 00237 else 00238 { 00239 if (stats->pass == last_pass) 00240 free(stats->buf.buf); 00241 } 00242 } 00243 00244 void stats_write(stats_io_t *stats, const void *pkt, size_t len) 00245 { 00246 if (stats->file) 00247 { 00248 if(fwrite(pkt, 1, len, stats->file)); 00249 } 00250 else 00251 { 00252 if (stats->buf.sz + len > stats->buf_alloc_sz) 00253 { 00254 size_t new_sz = stats->buf_alloc_sz + 64 * 1024; 00255 char *new_ptr = realloc(stats->buf.buf, new_sz); 00256 00257 if (new_ptr) 00258 { 00259 stats->buf_ptr = new_ptr + (stats->buf_ptr - (char *)stats->buf.buf); 00260 stats->buf.buf = new_ptr; 00261 stats->buf_alloc_sz = new_sz; 00262 } 00263 else 00264 fatal("Failed to realloc firstpass stats buffer."); 00265 } 00266 00267 memcpy(stats->buf_ptr, pkt, len); 00268 stats->buf.sz += len; 00269 stats->buf_ptr += len; 00270 } 00271 } 00272 00273 vpx_fixed_buf_t stats_get(stats_io_t *stats) 00274 { 00275 return stats->buf; 00276 } 00277 00278 /* Stereo 3D packed frame format */ 00279 typedef enum stereo_format 00280 { 00281 STEREO_FORMAT_MONO = 0, 00282 STEREO_FORMAT_LEFT_RIGHT = 1, 00283 STEREO_FORMAT_BOTTOM_TOP = 2, 00284 STEREO_FORMAT_TOP_BOTTOM = 3, 00285 STEREO_FORMAT_RIGHT_LEFT = 11 00286 } stereo_format_t; 00287 00288 enum video_file_type 00289 { 00290 FILE_TYPE_RAW, 00291 FILE_TYPE_IVF, 00292 FILE_TYPE_Y4M 00293 }; 00294 00295 struct detect_buffer { 00296 char buf[4]; 00297 size_t buf_read; 00298 size_t position; 00299 }; 00300 00301 00302 struct input_state 00303 { 00304 char *fn; 00305 FILE *file; 00306 y4m_input y4m; 00307 struct detect_buffer detect; 00308 enum video_file_type file_type; 00309 unsigned int w; 00310 unsigned int h; 00311 struct vpx_rational framerate; 00312 int use_i420; 00313 }; 00314 00315 00316 #define IVF_FRAME_HDR_SZ (4+8) /* 4 byte size + 8 byte timestamp */ 00317 static int read_frame(struct input_state *input, vpx_image_t *img) 00318 { 00319 FILE *f = input->file; 00320 enum video_file_type file_type = input->file_type; 00321 y4m_input *y4m = &input->y4m; 00322 struct detect_buffer *detect = &input->detect; 00323 int plane = 0; 00324 int shortread = 0; 00325 00326 if (file_type == FILE_TYPE_Y4M) 00327 { 00328 if (y4m_input_fetch_frame(y4m, f, img) < 1) 00329 return 0; 00330 } 00331 else 00332 { 00333 if (file_type == FILE_TYPE_IVF) 00334 { 00335 char junk[IVF_FRAME_HDR_SZ]; 00336 00337 /* Skip the frame header. We know how big the frame should be. See 00338 * write_ivf_frame_header() for documentation on the frame header 00339 * layout. 00340 */ 00341 if(fread(junk, 1, IVF_FRAME_HDR_SZ, f)); 00342 } 00343 00344 for (plane = 0; plane < 3; plane++) 00345 { 00346 unsigned char *ptr; 00347 int w = (plane ? (1 + img->d_w) / 2 : img->d_w); 00348 int h = (plane ? (1 + img->d_h) / 2 : img->d_h); 00349 int r; 00350 00351 /* Determine the correct plane based on the image format. The for-loop 00352 * always counts in Y,U,V order, but this may not match the order of 00353 * the data on disk. 00354 */ 00355 switch (plane) 00356 { 00357 case 1: 00358 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U]; 00359 break; 00360 case 2: 00361 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V]; 00362 break; 00363 default: 00364 ptr = img->planes[plane]; 00365 } 00366 00367 for (r = 0; r < h; r++) 00368 { 00369 size_t needed = w; 00370 size_t buf_position = 0; 00371 const size_t left = detect->buf_read - detect->position; 00372 if (left > 0) 00373 { 00374 const size_t more = (left < needed) ? left : needed; 00375 memcpy(ptr, detect->buf + detect->position, more); 00376 buf_position = more; 00377 needed -= more; 00378 detect->position += more; 00379 } 00380 if (needed > 0) 00381 { 00382 shortread |= (fread(ptr + buf_position, 1, needed, f) < needed); 00383 } 00384 00385 ptr += img->stride[plane]; 00386 } 00387 } 00388 } 00389 00390 return !shortread; 00391 } 00392 00393 00394 unsigned int file_is_y4m(FILE *infile, 00395 y4m_input *y4m, 00396 char detect[4]) 00397 { 00398 if(memcmp(detect, "YUV4", 4) == 0) 00399 { 00400 return 1; 00401 } 00402 return 0; 00403 } 00404 00405 #define IVF_FILE_HDR_SZ (32) 00406 unsigned int file_is_ivf(struct input_state *input, 00407 unsigned int *fourcc) 00408 { 00409 char raw_hdr[IVF_FILE_HDR_SZ]; 00410 int is_ivf = 0; 00411 FILE *infile = input->file; 00412 unsigned int *width = &input->w; 00413 unsigned int *height = &input->h; 00414 struct detect_buffer *detect = &input->detect; 00415 00416 if(memcmp(detect->buf, "DKIF", 4) != 0) 00417 return 0; 00418 00419 /* See write_ivf_file_header() for more documentation on the file header 00420 * layout. 00421 */ 00422 if (fread(raw_hdr + 4, 1, IVF_FILE_HDR_SZ - 4, infile) 00423 == IVF_FILE_HDR_SZ - 4) 00424 { 00425 { 00426 is_ivf = 1; 00427 00428 if (mem_get_le16(raw_hdr + 4) != 0) 00429 warn("Unrecognized IVF version! This file may not decode " 00430 "properly."); 00431 00432 *fourcc = mem_get_le32(raw_hdr + 8); 00433 } 00434 } 00435 00436 if (is_ivf) 00437 { 00438 *width = mem_get_le16(raw_hdr + 12); 00439 *height = mem_get_le16(raw_hdr + 14); 00440 detect->position = 4; 00441 } 00442 00443 return is_ivf; 00444 } 00445 00446 00447 static void write_ivf_file_header(FILE *outfile, 00448 const vpx_codec_enc_cfg_t *cfg, 00449 unsigned int fourcc, 00450 int frame_cnt) 00451 { 00452 char header[32]; 00453 00454 if (cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS) 00455 return; 00456 00457 header[0] = 'D'; 00458 header[1] = 'K'; 00459 header[2] = 'I'; 00460 header[3] = 'F'; 00461 mem_put_le16(header + 4, 0); /* version */ 00462 mem_put_le16(header + 6, 32); /* headersize */ 00463 mem_put_le32(header + 8, fourcc); /* headersize */ 00464 mem_put_le16(header + 12, cfg->g_w); /* width */ 00465 mem_put_le16(header + 14, cfg->g_h); /* height */ 00466 mem_put_le32(header + 16, cfg->g_timebase.den); /* rate */ 00467 mem_put_le32(header + 20, cfg->g_timebase.num); /* scale */ 00468 mem_put_le32(header + 24, frame_cnt); /* length */ 00469 mem_put_le32(header + 28, 0); /* unused */ 00470 00471 if(fwrite(header, 1, 32, outfile)); 00472 } 00473 00474 00475 static void write_ivf_frame_header(FILE *outfile, 00476 const vpx_codec_cx_pkt_t *pkt) 00477 { 00478 char header[12]; 00479 vpx_codec_pts_t pts; 00480 00481 if (pkt->kind != VPX_CODEC_CX_FRAME_PKT) 00482 return; 00483 00484 pts = pkt->data.frame.pts; 00485 mem_put_le32(header, pkt->data.frame.sz); 00486 mem_put_le32(header + 4, pts & 0xFFFFFFFF); 00487 mem_put_le32(header + 8, pts >> 32); 00488 00489 if(fwrite(header, 1, 12, outfile)); 00490 } 00491 00492 static void write_ivf_frame_size(FILE *outfile, size_t size) 00493 { 00494 char header[4]; 00495 mem_put_le32(header, size); 00496 fwrite(header, 1, 4, outfile); 00497 } 00498 00499 00500 typedef off_t EbmlLoc; 00501 00502 00503 struct cue_entry 00504 { 00505 unsigned int time; 00506 uint64_t loc; 00507 }; 00508 00509 00510 struct EbmlGlobal 00511 { 00512 int debug; 00513 00514 FILE *stream; 00515 int64_t last_pts_ms; 00516 vpx_rational_t framerate; 00517 00518 /* These pointers are to the start of an element */ 00519 off_t position_reference; 00520 off_t seek_info_pos; 00521 off_t segment_info_pos; 00522 off_t track_pos; 00523 off_t cue_pos; 00524 off_t cluster_pos; 00525 00526 /* This pointer is to a specific element to be serialized */ 00527 off_t track_id_pos; 00528 00529 /* These pointers are to the size field of the element */ 00530 EbmlLoc startSegment; 00531 EbmlLoc startCluster; 00532 00533 uint32_t cluster_timecode; 00534 int cluster_open; 00535 00536 struct cue_entry *cue_list; 00537 unsigned int cues; 00538 00539 }; 00540 00541 00542 void Ebml_Write(EbmlGlobal *glob, const void *buffer_in, unsigned long len) 00543 { 00544 if(fwrite(buffer_in, 1, len, glob->stream)); 00545 } 00546 00547 #define WRITE_BUFFER(s) \ 00548 for(i = len-1; i>=0; i--)\ 00549 { \ 00550 x = *(const s *)buffer_in >> (i * CHAR_BIT); \ 00551 Ebml_Write(glob, &x, 1); \ 00552 } 00553 void Ebml_Serialize(EbmlGlobal *glob, const void *buffer_in, int buffer_size, unsigned long len) 00554 { 00555 char x; 00556 int i; 00557 00558 /* buffer_size: 00559 * 1 - int8_t; 00560 * 2 - int16_t; 00561 * 3 - int32_t; 00562 * 4 - int64_t; 00563 */ 00564 switch (buffer_size) 00565 { 00566 case 1: 00567 WRITE_BUFFER(int8_t) 00568 break; 00569 case 2: 00570 WRITE_BUFFER(int16_t) 00571 break; 00572 case 4: 00573 WRITE_BUFFER(int32_t) 00574 break; 00575 case 8: 00576 WRITE_BUFFER(int64_t) 00577 break; 00578 default: 00579 break; 00580 } 00581 } 00582 #undef WRITE_BUFFER 00583 00584 /* Need a fixed size serializer for the track ID. libmkv provides a 64 bit 00585 * one, but not a 32 bit one. 00586 */ 00587 static void Ebml_SerializeUnsigned32(EbmlGlobal *glob, unsigned long class_id, uint64_t ui) 00588 { 00589 unsigned char sizeSerialized = 4 | 0x80; 00590 Ebml_WriteID(glob, class_id); 00591 Ebml_Serialize(glob, &sizeSerialized, sizeof(sizeSerialized), 1); 00592 Ebml_Serialize(glob, &ui, sizeof(ui), 4); 00593 } 00594 00595 00596 static void 00597 Ebml_StartSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc, 00598 unsigned long class_id) 00599 { 00600 //todo this is always taking 8 bytes, this may need later optimization 00601 //this is a key that says length unknown 00602 uint64_t unknownLen = LITERALU64(0x01FFFFFFFFFFFFFF); 00603 00604 Ebml_WriteID(glob, class_id); 00605 *ebmlLoc = ftello(glob->stream); 00606 Ebml_Serialize(glob, &unknownLen, sizeof(unknownLen), 8); 00607 } 00608 00609 static void 00610 Ebml_EndSubElement(EbmlGlobal *glob, EbmlLoc *ebmlLoc) 00611 { 00612 off_t pos; 00613 uint64_t size; 00614 00615 /* Save the current stream pointer */ 00616 pos = ftello(glob->stream); 00617 00618 /* Calculate the size of this element */ 00619 size = pos - *ebmlLoc - 8; 00620 size |= LITERALU64(0x0100000000000000); 00621 00622 /* Seek back to the beginning of the element and write the new size */ 00623 fseeko(glob->stream, *ebmlLoc, SEEK_SET); 00624 Ebml_Serialize(glob, &size, sizeof(size), 8); 00625 00626 /* Reset the stream pointer */ 00627 fseeko(glob->stream, pos, SEEK_SET); 00628 } 00629 00630 00631 static void 00632 write_webm_seek_element(EbmlGlobal *ebml, unsigned long id, off_t pos) 00633 { 00634 uint64_t offset = pos - ebml->position_reference; 00635 EbmlLoc start; 00636 Ebml_StartSubElement(ebml, &start, Seek); 00637 Ebml_SerializeBinary(ebml, SeekID, id); 00638 Ebml_SerializeUnsigned64(ebml, SeekPosition, offset); 00639 Ebml_EndSubElement(ebml, &start); 00640 } 00641 00642 00643 static void 00644 write_webm_seek_info(EbmlGlobal *ebml) 00645 { 00646 00647 off_t pos; 00648 00649 /* Save the current stream pointer */ 00650 pos = ftello(ebml->stream); 00651 00652 if(ebml->seek_info_pos) 00653 fseeko(ebml->stream, ebml->seek_info_pos, SEEK_SET); 00654 else 00655 ebml->seek_info_pos = pos; 00656 00657 { 00658 EbmlLoc start; 00659 00660 Ebml_StartSubElement(ebml, &start, SeekHead); 00661 write_webm_seek_element(ebml, Tracks, ebml->track_pos); 00662 write_webm_seek_element(ebml, Cues, ebml->cue_pos); 00663 write_webm_seek_element(ebml, Info, ebml->segment_info_pos); 00664 Ebml_EndSubElement(ebml, &start); 00665 } 00666 { 00667 //segment info 00668 EbmlLoc startInfo; 00669 uint64_t frame_time; 00670 char version_string[64]; 00671 00672 /* Assemble version string */ 00673 if(ebml->debug) 00674 strcpy(version_string, "vpxenc"); 00675 else 00676 { 00677 strcpy(version_string, "vpxenc "); 00678 strncat(version_string, 00679 vpx_codec_version_str(), 00680 sizeof(version_string) - 1 - strlen(version_string)); 00681 } 00682 00683 frame_time = (uint64_t)1000 * ebml->framerate.den 00684 / ebml->framerate.num; 00685 ebml->segment_info_pos = ftello(ebml->stream); 00686 Ebml_StartSubElement(ebml, &startInfo, Info); 00687 Ebml_SerializeUnsigned(ebml, TimecodeScale, 1000000); 00688 Ebml_SerializeFloat(ebml, Segment_Duration, 00689 ebml->last_pts_ms + frame_time); 00690 Ebml_SerializeString(ebml, 0x4D80, version_string); 00691 Ebml_SerializeString(ebml, 0x5741, version_string); 00692 Ebml_EndSubElement(ebml, &startInfo); 00693 } 00694 } 00695 00696 00697 static void 00698 write_webm_file_header(EbmlGlobal *glob, 00699 const vpx_codec_enc_cfg_t *cfg, 00700 const struct vpx_rational *fps, 00701 stereo_format_t stereo_fmt) 00702 { 00703 { 00704 EbmlLoc start; 00705 Ebml_StartSubElement(glob, &start, EBML); 00706 Ebml_SerializeUnsigned(glob, EBMLVersion, 1); 00707 Ebml_SerializeUnsigned(glob, EBMLReadVersion, 1); //EBML Read Version 00708 Ebml_SerializeUnsigned(glob, EBMLMaxIDLength, 4); //EBML Max ID Length 00709 Ebml_SerializeUnsigned(glob, EBMLMaxSizeLength, 8); //EBML Max Size Length 00710 Ebml_SerializeString(glob, DocType, "webm"); //Doc Type 00711 Ebml_SerializeUnsigned(glob, DocTypeVersion, 2); //Doc Type Version 00712 Ebml_SerializeUnsigned(glob, DocTypeReadVersion, 2); //Doc Type Read Version 00713 Ebml_EndSubElement(glob, &start); 00714 } 00715 { 00716 Ebml_StartSubElement(glob, &glob->startSegment, Segment); //segment 00717 glob->position_reference = ftello(glob->stream); 00718 glob->framerate = *fps; 00719 write_webm_seek_info(glob); 00720 00721 { 00722 EbmlLoc trackStart; 00723 glob->track_pos = ftello(glob->stream); 00724 Ebml_StartSubElement(glob, &trackStart, Tracks); 00725 { 00726 unsigned int trackNumber = 1; 00727 uint64_t trackID = 0; 00728 00729 EbmlLoc start; 00730 Ebml_StartSubElement(glob, &start, TrackEntry); 00731 Ebml_SerializeUnsigned(glob, TrackNumber, trackNumber); 00732 glob->track_id_pos = ftello(glob->stream); 00733 Ebml_SerializeUnsigned32(glob, TrackUID, trackID); 00734 Ebml_SerializeUnsigned(glob, TrackType, 1); //video is always 1 00735 Ebml_SerializeString(glob, CodecID, "V_VP8"); 00736 { 00737 unsigned int pixelWidth = cfg->g_w; 00738 unsigned int pixelHeight = cfg->g_h; 00739 float frameRate = (float)fps->num/(float)fps->den; 00740 00741 EbmlLoc videoStart; 00742 Ebml_StartSubElement(glob, &videoStart, Video); 00743 Ebml_SerializeUnsigned(glob, PixelWidth, pixelWidth); 00744 Ebml_SerializeUnsigned(glob, PixelHeight, pixelHeight); 00745 Ebml_SerializeUnsigned(glob, StereoMode, stereo_fmt); 00746 Ebml_SerializeFloat(glob, FrameRate, frameRate); 00747 Ebml_EndSubElement(glob, &videoStart); //Video 00748 } 00749 Ebml_EndSubElement(glob, &start); //Track Entry 00750 } 00751 Ebml_EndSubElement(glob, &trackStart); 00752 } 00753 // segment element is open 00754 } 00755 } 00756 00757 00758 static void 00759 write_webm_block(EbmlGlobal *glob, 00760 const vpx_codec_enc_cfg_t *cfg, 00761 const vpx_codec_cx_pkt_t *pkt) 00762 { 00763 unsigned long block_length; 00764 unsigned char track_number; 00765 unsigned short block_timecode = 0; 00766 unsigned char flags; 00767 int64_t pts_ms; 00768 int start_cluster = 0, is_keyframe; 00769 00770 /* Calculate the PTS of this frame in milliseconds */ 00771 pts_ms = pkt->data.frame.pts * 1000 00772 * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den; 00773 if(pts_ms <= glob->last_pts_ms) 00774 pts_ms = glob->last_pts_ms + 1; 00775 glob->last_pts_ms = pts_ms; 00776 00777 /* Calculate the relative time of this block */ 00778 if(pts_ms - glob->cluster_timecode > SHRT_MAX) 00779 start_cluster = 1; 00780 else 00781 block_timecode = pts_ms - glob->cluster_timecode; 00782 00783 is_keyframe = (pkt->data.frame.flags & VPX_FRAME_IS_KEY); 00784 if(start_cluster || is_keyframe) 00785 { 00786 if(glob->cluster_open) 00787 Ebml_EndSubElement(glob, &glob->startCluster); 00788 00789 /* Open the new cluster */ 00790 block_timecode = 0; 00791 glob->cluster_open = 1; 00792 glob->cluster_timecode = pts_ms; 00793 glob->cluster_pos = ftello(glob->stream); 00794 Ebml_StartSubElement(glob, &glob->startCluster, Cluster); //cluster 00795 Ebml_SerializeUnsigned(glob, Timecode, glob->cluster_timecode); 00796 00797 /* Save a cue point if this is a keyframe. */ 00798 if(is_keyframe) 00799 { 00800 struct cue_entry *cue, *new_cue_list; 00801 00802 new_cue_list = realloc(glob->cue_list, 00803 (glob->cues+1) * sizeof(struct cue_entry)); 00804 if(new_cue_list) 00805 glob->cue_list = new_cue_list; 00806 else 00807 fatal("Failed to realloc cue list."); 00808 00809 cue = &glob->cue_list[glob->cues]; 00810 cue->time = glob->cluster_timecode; 00811 cue->loc = glob->cluster_pos; 00812 glob->cues++; 00813 } 00814 } 00815 00816 /* Write the Simple Block */ 00817 Ebml_WriteID(glob, SimpleBlock); 00818 00819 block_length = pkt->data.frame.sz + 4; 00820 block_length |= 0x10000000; 00821 Ebml_Serialize(glob, &block_length, sizeof(block_length), 4); 00822 00823 track_number = 1; 00824 track_number |= 0x80; 00825 Ebml_Write(glob, &track_number, 1); 00826 00827 Ebml_Serialize(glob, &block_timecode, sizeof(block_timecode), 2); 00828 00829 flags = 0; 00830 if(is_keyframe) 00831 flags |= 0x80; 00832 if(pkt->data.frame.flags & VPX_FRAME_IS_INVISIBLE) 00833 flags |= 0x08; 00834 Ebml_Write(glob, &flags, 1); 00835 00836 Ebml_Write(glob, pkt->data.frame.buf, pkt->data.frame.sz); 00837 } 00838 00839 00840 static void 00841 write_webm_file_footer(EbmlGlobal *glob, long hash) 00842 { 00843 00844 if(glob->cluster_open) 00845 Ebml_EndSubElement(glob, &glob->startCluster); 00846 00847 { 00848 EbmlLoc start; 00849 unsigned int i; 00850 00851 glob->cue_pos = ftello(glob->stream); 00852 Ebml_StartSubElement(glob, &start, Cues); 00853 for(i=0; i<glob->cues; i++) 00854 { 00855 struct cue_entry *cue = &glob->cue_list[i]; 00856 EbmlLoc start; 00857 00858 Ebml_StartSubElement(glob, &start, CuePoint); 00859 { 00860 EbmlLoc start; 00861 00862 Ebml_SerializeUnsigned(glob, CueTime, cue->time); 00863 00864 Ebml_StartSubElement(glob, &start, CueTrackPositions); 00865 Ebml_SerializeUnsigned(glob, CueTrack, 1); 00866 Ebml_SerializeUnsigned64(glob, CueClusterPosition, 00867 cue->loc - glob->position_reference); 00868 //Ebml_SerializeUnsigned(glob, CueBlockNumber, cue->blockNumber); 00869 Ebml_EndSubElement(glob, &start); 00870 } 00871 Ebml_EndSubElement(glob, &start); 00872 } 00873 Ebml_EndSubElement(glob, &start); 00874 } 00875 00876 Ebml_EndSubElement(glob, &glob->startSegment); 00877 00878 /* Patch up the seek info block */ 00879 write_webm_seek_info(glob); 00880 00881 /* Patch up the track id */ 00882 fseeko(glob->stream, glob->track_id_pos, SEEK_SET); 00883 Ebml_SerializeUnsigned32(glob, TrackUID, glob->debug ? 0xDEADBEEF : hash); 00884 00885 fseeko(glob->stream, 0, SEEK_END); 00886 } 00887 00888 00889 /* Murmur hash derived from public domain reference implementation at 00890 * http://sites.google.com/site/murmurhash/ 00891 */ 00892 static unsigned int murmur ( const void * key, int len, unsigned int seed ) 00893 { 00894 const unsigned int m = 0x5bd1e995; 00895 const int r = 24; 00896 00897 unsigned int h = seed ^ len; 00898 00899 const unsigned char * data = (const unsigned char *)key; 00900 00901 while(len >= 4) 00902 { 00903 unsigned int k; 00904 00905 k = data[0]; 00906 k |= data[1] << 8; 00907 k |= data[2] << 16; 00908 k |= data[3] << 24; 00909 00910 k *= m; 00911 k ^= k >> r; 00912 k *= m; 00913 00914 h *= m; 00915 h ^= k; 00916 00917 data += 4; 00918 len -= 4; 00919 } 00920 00921 switch(len) 00922 { 00923 case 3: h ^= data[2] << 16; 00924 case 2: h ^= data[1] << 8; 00925 case 1: h ^= data[0]; 00926 h *= m; 00927 }; 00928 00929 h ^= h >> 13; 00930 h *= m; 00931 h ^= h >> 15; 00932 00933 return h; 00934 } 00935 00936 #include "math.h" 00937 00938 static double vp8_mse2psnr(double Samples, double Peak, double Mse) 00939 { 00940 double psnr; 00941 00942 if ((double)Mse > 0.0) 00943 psnr = 10.0 * log10(Peak * Peak * Samples / Mse); 00944 else 00945 psnr = 60; // Limit to prevent / 0 00946 00947 if (psnr > 60) 00948 psnr = 60; 00949 00950 return psnr; 00951 } 00952 00953 00954 #include "args.h" 00955 static const arg_def_t debugmode = ARG_DEF("D", "debug", 0, 00956 "Debug mode (makes output deterministic)"); 00957 static const arg_def_t outputfile = ARG_DEF("o", "output", 1, 00958 "Output filename"); 00959 static const arg_def_t use_yv12 = ARG_DEF(NULL, "yv12", 0, 00960 "Input file is YV12 "); 00961 static const arg_def_t use_i420 = ARG_DEF(NULL, "i420", 0, 00962 "Input file is I420 (default)"); 00963 static const arg_def_t codecarg = ARG_DEF(NULL, "codec", 1, 00964 "Codec to use"); 00965 static const arg_def_t passes = ARG_DEF("p", "passes", 1, 00966 "Number of passes (1/2)"); 00967 static const arg_def_t pass_arg = ARG_DEF(NULL, "pass", 1, 00968 "Pass to execute (1/2)"); 00969 static const arg_def_t fpf_name = ARG_DEF(NULL, "fpf", 1, 00970 "First pass statistics file name"); 00971 static const arg_def_t limit = ARG_DEF(NULL, "limit", 1, 00972 "Stop encoding after n input frames"); 00973 static const arg_def_t deadline = ARG_DEF("d", "deadline", 1, 00974 "Deadline per frame (usec)"); 00975 static const arg_def_t best_dl = ARG_DEF(NULL, "best", 0, 00976 "Use Best Quality Deadline"); 00977 static const arg_def_t good_dl = ARG_DEF(NULL, "good", 0, 00978 "Use Good Quality Deadline"); 00979 static const arg_def_t rt_dl = ARG_DEF(NULL, "rt", 0, 00980 "Use Realtime Quality Deadline"); 00981 static const arg_def_t verbosearg = ARG_DEF("v", "verbose", 0, 00982 "Show encoder parameters"); 00983 static const arg_def_t psnrarg = ARG_DEF(NULL, "psnr", 0, 00984 "Show PSNR in status line"); 00985 static const arg_def_t framerate = ARG_DEF(NULL, "fps", 1, 00986 "Stream frame rate (rate/scale)"); 00987 static const arg_def_t use_ivf = ARG_DEF(NULL, "ivf", 0, 00988 "Output IVF (default is WebM)"); 00989 static const arg_def_t out_part = ARG_DEF("P", "output-partitions", 0, 00990 "Makes encoder output partitions. Requires IVF output!"); 00991 static const arg_def_t q_hist_n = ARG_DEF(NULL, "q-hist", 1, 00992 "Show quantizer histogram (n-buckets)"); 00993 static const arg_def_t rate_hist_n = ARG_DEF(NULL, "rate-hist", 1, 00994 "Show rate histogram (n-buckets)"); 00995 static const arg_def_t *main_args[] = 00996 { 00997 &debugmode, 00998 &outputfile, &codecarg, &passes, &pass_arg, &fpf_name, &limit, &deadline, 00999 &best_dl, &good_dl, &rt_dl, 01000 &verbosearg, &psnrarg, &use_ivf, &out_part, &q_hist_n, &rate_hist_n, 01001 NULL 01002 }; 01003 01004 static const arg_def_t usage = ARG_DEF("u", "usage", 1, 01005 "Usage profile number to use"); 01006 static const arg_def_t threads = ARG_DEF("t", "threads", 1, 01007 "Max number of threads to use"); 01008 static const arg_def_t profile = ARG_DEF(NULL, "profile", 1, 01009 "Bitstream profile number to use"); 01010 static const arg_def_t width = ARG_DEF("w", "width", 1, 01011 "Frame width"); 01012 static const arg_def_t height = ARG_DEF("h", "height", 1, 01013 "Frame height"); 01014 static const struct arg_enum_list stereo_mode_enum[] = { 01015 {"mono" , STEREO_FORMAT_MONO}, 01016 {"left-right", STEREO_FORMAT_LEFT_RIGHT}, 01017 {"bottom-top", STEREO_FORMAT_BOTTOM_TOP}, 01018 {"top-bottom", STEREO_FORMAT_TOP_BOTTOM}, 01019 {"right-left", STEREO_FORMAT_RIGHT_LEFT}, 01020 {NULL, 0} 01021 }; 01022 static const arg_def_t stereo_mode = ARG_DEF_ENUM(NULL, "stereo-mode", 1, 01023 "Stereo 3D video format", stereo_mode_enum); 01024 static const arg_def_t timebase = ARG_DEF(NULL, "timebase", 1, 01025 "Output timestamp precision (fractional seconds)"); 01026 static const arg_def_t error_resilient = ARG_DEF(NULL, "error-resilient", 1, 01027 "Enable error resiliency features"); 01028 static const arg_def_t lag_in_frames = ARG_DEF(NULL, "lag-in-frames", 1, 01029 "Max number of frames to lag"); 01030 01031 static const arg_def_t *global_args[] = 01032 { 01033 &use_yv12, &use_i420, &usage, &threads, &profile, 01034 &width, &height, &stereo_mode, &timebase, &framerate, &error_resilient, 01035 &lag_in_frames, NULL 01036 }; 01037 01038 static const arg_def_t dropframe_thresh = ARG_DEF(NULL, "drop-frame", 1, 01039 "Temporal resampling threshold (buf %)"); 01040 static const arg_def_t resize_allowed = ARG_DEF(NULL, "resize-allowed", 1, 01041 "Spatial resampling enabled (bool)"); 01042 static const arg_def_t resize_up_thresh = ARG_DEF(NULL, "resize-up", 1, 01043 "Upscale threshold (buf %)"); 01044 static const arg_def_t resize_down_thresh = ARG_DEF(NULL, "resize-down", 1, 01045 "Downscale threshold (buf %)"); 01046 static const struct arg_enum_list end_usage_enum[] = { 01047 {"vbr", VPX_VBR}, 01048 {"cbr", VPX_CBR}, 01049 {"cq", VPX_CQ}, 01050 {NULL, 0} 01051 }; 01052 static const arg_def_t end_usage = ARG_DEF_ENUM(NULL, "end-usage", 1, 01053 "Rate control mode", end_usage_enum); 01054 static const arg_def_t target_bitrate = ARG_DEF(NULL, "target-bitrate", 1, 01055 "Bitrate (kbps)"); 01056 static const arg_def_t min_quantizer = ARG_DEF(NULL, "min-q", 1, 01057 "Minimum (best) quantizer"); 01058 static const arg_def_t max_quantizer = ARG_DEF(NULL, "max-q", 1, 01059 "Maximum (worst) quantizer"); 01060 static const arg_def_t undershoot_pct = ARG_DEF(NULL, "undershoot-pct", 1, 01061 "Datarate undershoot (min) target (%)"); 01062 static const arg_def_t overshoot_pct = ARG_DEF(NULL, "overshoot-pct", 1, 01063 "Datarate overshoot (max) target (%)"); 01064 static const arg_def_t buf_sz = ARG_DEF(NULL, "buf-sz", 1, 01065 "Client buffer size (ms)"); 01066 static const arg_def_t buf_initial_sz = ARG_DEF(NULL, "buf-initial-sz", 1, 01067 "Client initial buffer size (ms)"); 01068 static const arg_def_t buf_optimal_sz = ARG_DEF(NULL, "buf-optimal-sz", 1, 01069 "Client optimal buffer size (ms)"); 01070 static const arg_def_t *rc_args[] = 01071 { 01072 &dropframe_thresh, &resize_allowed, &resize_up_thresh, &resize_down_thresh, 01073 &end_usage, &target_bitrate, &min_quantizer, &max_quantizer, 01074 &undershoot_pct, &overshoot_pct, &buf_sz, &buf_initial_sz, &buf_optimal_sz, 01075 NULL 01076 }; 01077 01078 01079 static const arg_def_t bias_pct = ARG_DEF(NULL, "bias-pct", 1, 01080 "CBR/VBR bias (0=CBR, 100=VBR)"); 01081 static const arg_def_t minsection_pct = ARG_DEF(NULL, "minsection-pct", 1, 01082 "GOP min bitrate (% of target)"); 01083 static const arg_def_t maxsection_pct = ARG_DEF(NULL, "maxsection-pct", 1, 01084 "GOP max bitrate (% of target)"); 01085 static const arg_def_t *rc_twopass_args[] = 01086 { 01087 &bias_pct, &minsection_pct, &maxsection_pct, NULL 01088 }; 01089 01090 01091 static const arg_def_t kf_min_dist = ARG_DEF(NULL, "kf-min-dist", 1, 01092 "Minimum keyframe interval (frames)"); 01093 static const arg_def_t kf_max_dist = ARG_DEF(NULL, "kf-max-dist", 1, 01094 "Maximum keyframe interval (frames)"); 01095 static const arg_def_t kf_disabled = ARG_DEF(NULL, "disable-kf", 0, 01096 "Disable keyframe placement"); 01097 static const arg_def_t *kf_args[] = 01098 { 01099 &kf_min_dist, &kf_max_dist, &kf_disabled, NULL 01100 }; 01101 01102 01103 #if CONFIG_VP8_ENCODER 01104 static const arg_def_t noise_sens = ARG_DEF(NULL, "noise-sensitivity", 1, 01105 "Noise sensitivity (frames to blur)"); 01106 static const arg_def_t sharpness = ARG_DEF(NULL, "sharpness", 1, 01107 "Filter sharpness (0-7)"); 01108 static const arg_def_t static_thresh = ARG_DEF(NULL, "static-thresh", 1, 01109 "Motion detection threshold"); 01110 #endif 01111 01112 #if CONFIG_VP8_ENCODER 01113 static const arg_def_t cpu_used = ARG_DEF(NULL, "cpu-used", 1, 01114 "CPU Used (-16..16)"); 01115 #endif 01116 01117 01118 #if CONFIG_VP8_ENCODER 01119 static const arg_def_t token_parts = ARG_DEF(NULL, "token-parts", 1, 01120 "Number of token partitions to use, log2"); 01121 static const arg_def_t auto_altref = ARG_DEF(NULL, "auto-alt-ref", 1, 01122 "Enable automatic alt reference frames"); 01123 static const arg_def_t arnr_maxframes = ARG_DEF(NULL, "arnr-maxframes", 1, 01124 "AltRef Max Frames"); 01125 static const arg_def_t arnr_strength = ARG_DEF(NULL, "arnr-strength", 1, 01126 "AltRef Strength"); 01127 static const arg_def_t arnr_type = ARG_DEF(NULL, "arnr-type", 1, 01128 "AltRef Type"); 01129 static const struct arg_enum_list tuning_enum[] = { 01130 {"psnr", VP8_TUNE_PSNR}, 01131 {"ssim", VP8_TUNE_SSIM}, 01132 {NULL, 0} 01133 }; 01134 static const arg_def_t tune_ssim = ARG_DEF_ENUM(NULL, "tune", 1, 01135 "Material to favor", tuning_enum); 01136 static const arg_def_t cq_level = ARG_DEF(NULL, "cq-level", 1, 01137 "Constrained Quality Level"); 01138 static const arg_def_t max_intra_rate_pct = ARG_DEF(NULL, "max-intra-rate", 1, 01139 "Max I-frame bitrate (pct)"); 01140 01141 static const arg_def_t *vp8_args[] = 01142 { 01143 &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh, 01144 &token_parts, &arnr_maxframes, &arnr_strength, &arnr_type, 01145 &tune_ssim, &cq_level, &max_intra_rate_pct, NULL 01146 }; 01147 static const int vp8_arg_ctrl_map[] = 01148 { 01149 VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF, 01150 VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD, 01151 VP8E_SET_TOKEN_PARTITIONS, 01152 VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH , VP8E_SET_ARNR_TYPE, 01153 VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, VP8E_SET_MAX_INTRA_BITRATE_PCT, 0 01154 }; 01155 #endif 01156 01157 static const arg_def_t *no_args[] = { NULL }; 01158 01159 static void usage_exit() 01160 { 01161 int i; 01162 01163 fprintf(stderr, "Usage: %s <options> -o dst_filename src_filename \n", 01164 exec_name); 01165 01166 fprintf(stderr, "\nOptions:\n"); 01167 arg_show_usage(stdout, main_args); 01168 fprintf(stderr, "\nEncoder Global Options:\n"); 01169 arg_show_usage(stdout, global_args); 01170 fprintf(stderr, "\nRate Control Options:\n"); 01171 arg_show_usage(stdout, rc_args); 01172 fprintf(stderr, "\nTwopass Rate Control Options:\n"); 01173 arg_show_usage(stdout, rc_twopass_args); 01174 fprintf(stderr, "\nKeyframe Placement Options:\n"); 01175 arg_show_usage(stdout, kf_args); 01176 #if CONFIG_VP8_ENCODER 01177 fprintf(stderr, "\nVP8 Specific Options:\n"); 01178 arg_show_usage(stdout, vp8_args); 01179 #endif 01180 fprintf(stderr, "\nStream timebase (--timebase):\n" 01181 " The desired precision of timestamps in the output, expressed\n" 01182 " in fractional seconds. Default is 1/1000.\n"); 01183 fprintf(stderr, "\n" 01184 "Included encoders:\n" 01185 "\n"); 01186 01187 for (i = 0; i < sizeof(codecs) / sizeof(codecs[0]); i++) 01188 fprintf(stderr, " %-6s - %s\n", 01189 codecs[i].name, 01190 vpx_codec_iface_name(codecs[i].iface)); 01191 01192 exit(EXIT_FAILURE); 01193 } 01194 01195 01196 #define HIST_BAR_MAX 40 01197 struct hist_bucket 01198 { 01199 int low, high, count; 01200 }; 01201 01202 01203 static int merge_hist_buckets(struct hist_bucket *bucket, 01204 int *buckets_, 01205 int max_buckets) 01206 { 01207 int small_bucket = 0, merge_bucket = INT_MAX, big_bucket=0; 01208 int buckets = *buckets_; 01209 int i; 01210 01211 /* Find the extrema for this list of buckets */ 01212 big_bucket = small_bucket = 0; 01213 for(i=0; i < buckets; i++) 01214 { 01215 if(bucket[i].count < bucket[small_bucket].count) 01216 small_bucket = i; 01217 if(bucket[i].count > bucket[big_bucket].count) 01218 big_bucket = i; 01219 } 01220 01221 /* If we have too many buckets, merge the smallest with an adjacent 01222 * bucket. 01223 */ 01224 while(buckets > max_buckets) 01225 { 01226 int last_bucket = buckets - 1; 01227 01228 // merge the small bucket with an adjacent one. 01229 if(small_bucket == 0) 01230 merge_bucket = 1; 01231 else if(small_bucket == last_bucket) 01232 merge_bucket = last_bucket - 1; 01233 else if(bucket[small_bucket - 1].count < bucket[small_bucket + 1].count) 01234 merge_bucket = small_bucket - 1; 01235 else 01236 merge_bucket = small_bucket + 1; 01237 01238 assert(abs(merge_bucket - small_bucket) <= 1); 01239 assert(small_bucket < buckets); 01240 assert(big_bucket < buckets); 01241 assert(merge_bucket < buckets); 01242 01243 if(merge_bucket < small_bucket) 01244 { 01245 bucket[merge_bucket].high = bucket[small_bucket].high; 01246 bucket[merge_bucket].count += bucket[small_bucket].count; 01247 } 01248 else 01249 { 01250 bucket[small_bucket].high = bucket[merge_bucket].high; 01251 bucket[small_bucket].count += bucket[merge_bucket].count; 01252 merge_bucket = small_bucket; 01253 } 01254 01255 assert(bucket[merge_bucket].low != bucket[merge_bucket].high); 01256 01257 buckets--; 01258 01259 /* Remove the merge_bucket from the list, and find the new small 01260 * and big buckets while we're at it 01261 */ 01262 big_bucket = small_bucket = 0; 01263 for(i=0; i < buckets; i++) 01264 { 01265 if(i > merge_bucket) 01266 bucket[i] = bucket[i+1]; 01267 01268 if(bucket[i].count < bucket[small_bucket].count) 01269 small_bucket = i; 01270 if(bucket[i].count > bucket[big_bucket].count) 01271 big_bucket = i; 01272 } 01273 01274 } 01275 01276 *buckets_ = buckets; 01277 return bucket[big_bucket].count; 01278 } 01279 01280 01281 static void show_histogram(const struct hist_bucket *bucket, 01282 int buckets, 01283 int total, 01284 int scale) 01285 { 01286 const char *pat1, *pat2; 01287 int i; 01288 01289 switch((int)(log(bucket[buckets-1].high)/log(10))+1) 01290 { 01291 case 1: 01292 case 2: 01293 pat1 = "%4d %2s: "; 01294 pat2 = "%4d-%2d: "; 01295 break; 01296 case 3: 01297 pat1 = "%5d %3s: "; 01298 pat2 = "%5d-%3d: "; 01299 break; 01300 case 4: 01301 pat1 = "%6d %4s: "; 01302 pat2 = "%6d-%4d: "; 01303 break; 01304 case 5: 01305 pat1 = "%7d %5s: "; 01306 pat2 = "%7d-%5d: "; 01307 break; 01308 case 6: 01309 pat1 = "%8d %6s: "; 01310 pat2 = "%8d-%6d: "; 01311 break; 01312 case 7: 01313 pat1 = "%9d %7s: "; 01314 pat2 = "%9d-%7d: "; 01315 break; 01316 default: 01317 pat1 = "%12d %10s: "; 01318 pat2 = "%12d-%10d: "; 01319 break; 01320 } 01321 01322 for(i=0; i<buckets; i++) 01323 { 01324 int len; 01325 int j; 01326 float pct; 01327 01328 pct = 100.0 * (float)bucket[i].count / (float)total; 01329 len = HIST_BAR_MAX * bucket[i].count / scale; 01330 if(len < 1) 01331 len = 1; 01332 assert(len <= HIST_BAR_MAX); 01333 01334 if(bucket[i].low == bucket[i].high) 01335 fprintf(stderr, pat1, bucket[i].low, ""); 01336 else 01337 fprintf(stderr, pat2, bucket[i].low, bucket[i].high); 01338 01339 for(j=0; j<HIST_BAR_MAX; j++) 01340 fprintf(stderr, j<len?"=":" "); 01341 fprintf(stderr, "\t%5d (%6.2f%%)\n",bucket[i].count,pct); 01342 } 01343 } 01344 01345 01346 static void show_q_histogram(const int counts[64], int max_buckets) 01347 { 01348 struct hist_bucket bucket[64]; 01349 int buckets = 0; 01350 int total = 0; 01351 int scale; 01352 int i; 01353 01354 01355 for(i=0; i<64; i++) 01356 { 01357 if(counts[i]) 01358 { 01359 bucket[buckets].low = bucket[buckets].high = i; 01360 bucket[buckets].count = counts[i]; 01361 buckets++; 01362 total += counts[i]; 01363 } 01364 } 01365 01366 fprintf(stderr, "\nQuantizer Selection:\n"); 01367 scale = merge_hist_buckets(bucket, &buckets, max_buckets); 01368 show_histogram(bucket, buckets, total, scale); 01369 } 01370 01371 01372 #define RATE_BINS (100) 01373 struct rate_hist 01374 { 01375 int64_t *pts; 01376 int *sz; 01377 int samples; 01378 int frames; 01379 struct hist_bucket bucket[RATE_BINS]; 01380 int total; 01381 }; 01382 01383 01384 static void init_rate_histogram(struct rate_hist *hist, 01385 const vpx_codec_enc_cfg_t *cfg, 01386 const vpx_rational_t *fps) 01387 { 01388 int i; 01389 01390 /* Determine the number of samples in the buffer. Use the file's framerate 01391 * to determine the number of frames in rc_buf_sz milliseconds, with an 01392 * adjustment (5/4) to account for alt-refs 01393 */ 01394 hist->samples = cfg->rc_buf_sz * 5 / 4 * fps->num / fps->den / 1000; 01395 01396 // prevent division by zero 01397 if (hist->samples == 0) 01398 hist->samples=1; 01399 01400 hist->pts = calloc(hist->samples, sizeof(*hist->pts)); 01401 hist->sz = calloc(hist->samples, sizeof(*hist->sz)); 01402 for(i=0; i<RATE_BINS; i++) 01403 { 01404 hist->bucket[i].low = INT_MAX; 01405 hist->bucket[i].high = 0; 01406 hist->bucket[i].count = 0; 01407 } 01408 } 01409 01410 01411 static void destroy_rate_histogram(struct rate_hist *hist) 01412 { 01413 free(hist->pts); 01414 free(hist->sz); 01415 } 01416 01417 01418 static void update_rate_histogram(struct rate_hist *hist, 01419 const vpx_codec_enc_cfg_t *cfg, 01420 const vpx_codec_cx_pkt_t *pkt) 01421 { 01422 int i, idx; 01423 int64_t now, then, sum_sz = 0, avg_bitrate; 01424 01425 now = pkt->data.frame.pts * 1000 01426 * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den; 01427 01428 idx = hist->frames++ % hist->samples; 01429 hist->pts[idx] = now; 01430 hist->sz[idx] = pkt->data.frame.sz; 01431 01432 if(now < cfg->rc_buf_initial_sz) 01433 return; 01434 01435 then = now; 01436 01437 /* Sum the size over the past rc_buf_sz ms */ 01438 for(i = hist->frames; i > 0 && hist->frames - i < hist->samples; i--) 01439 { 01440 int i_idx = (i-1) % hist->samples; 01441 01442 then = hist->pts[i_idx]; 01443 if(now - then > cfg->rc_buf_sz) 01444 break; 01445 sum_sz += hist->sz[i_idx]; 01446 } 01447 01448 if (now == then) 01449 return; 01450 01451 avg_bitrate = sum_sz * 8 * 1000 / (now - then); 01452 idx = avg_bitrate * (RATE_BINS/2) / (cfg->rc_target_bitrate * 1000); 01453 if(idx < 0) 01454 idx = 0; 01455 if(idx > RATE_BINS-1) 01456 idx = RATE_BINS-1; 01457 if(hist->bucket[idx].low > avg_bitrate) 01458 hist->bucket[idx].low = avg_bitrate; 01459 if(hist->bucket[idx].high < avg_bitrate) 01460 hist->bucket[idx].high = avg_bitrate; 01461 hist->bucket[idx].count++; 01462 hist->total++; 01463 } 01464 01465 01466 static void show_rate_histogram(struct rate_hist *hist, 01467 const vpx_codec_enc_cfg_t *cfg, 01468 int max_buckets) 01469 { 01470 int i, scale; 01471 int buckets = 0; 01472 01473 for(i = 0; i < RATE_BINS; i++) 01474 { 01475 if(hist->bucket[i].low == INT_MAX) 01476 continue; 01477 hist->bucket[buckets++] = hist->bucket[i]; 01478 } 01479 01480 fprintf(stderr, "\nRate (over %dms window):\n", cfg->rc_buf_sz); 01481 scale = merge_hist_buckets(hist->bucket, &buckets, max_buckets); 01482 show_histogram(hist->bucket, buckets, hist->total, scale); 01483 } 01484 01485 #define NELEMENTS(x) (sizeof(x)/sizeof(x[0])) 01486 #define ARG_CTRL_CNT_MAX NELEMENTS(vp8_arg_ctrl_map) 01487 01488 01489 /* Configuration elements common to all streams */ 01490 struct global_config 01491 { 01492 const struct codec_item *codec; 01493 int passes; 01494 int pass; 01495 int usage; 01496 int deadline; 01497 int use_i420; 01498 int verbose; 01499 int limit; 01500 int show_psnr; 01501 int have_framerate; 01502 struct vpx_rational framerate; 01503 int out_part; 01504 int debug; 01505 int show_q_hist_buckets; 01506 int show_rate_hist_buckets; 01507 }; 01508 01509 01510 /* Per-stream configuration */ 01511 struct stream_config 01512 { 01513 struct vpx_codec_enc_cfg cfg; 01514 const char *out_fn; 01515 const char *stats_fn; 01516 stereo_format_t stereo_fmt; 01517 int arg_ctrls[ARG_CTRL_CNT_MAX][2]; 01518 int arg_ctrl_cnt; 01519 int write_webm; 01520 int have_kf_max_dist; 01521 }; 01522 01523 01524 struct stream_state 01525 { 01526 int index; 01527 struct stream_state *next; 01528 struct stream_config config; 01529 FILE *file; 01530 struct rate_hist rate_hist; 01531 EbmlGlobal ebml; 01532 uint32_t hash; 01533 uint64_t psnr_sse_total; 01534 uint64_t psnr_samples_total; 01535 double psnr_totals[4]; 01536 int psnr_count; 01537 int counts[64]; 01538 vpx_codec_ctx_t encoder; 01539 unsigned int frames_out; 01540 uint64_t cx_time; 01541 size_t nbytes; 01542 stats_io_t stats; 01543 }; 01544 01545 01546 void validate_positive_rational(const char *msg, 01547 struct vpx_rational *rat) 01548 { 01549 if (rat->den < 0) 01550 { 01551 rat->num *= -1; 01552 rat->den *= -1; 01553 } 01554 01555 if (rat->num < 0) 01556 die("Error: %s must be positive\n", msg); 01557 01558 if (!rat->den) 01559 die("Error: %s has zero denominator\n", msg); 01560 } 01561 01562 01563 static void parse_global_config(struct global_config *global, char **argv) 01564 { 01565 char **argi, **argj; 01566 struct arg arg; 01567 01568 /* Initialize default parameters */ 01569 memset(global, 0, sizeof(*global)); 01570 global->codec = codecs; 01571 global->passes = 1; 01572 global->use_i420 = 1; 01573 01574 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) 01575 { 01576 arg.argv_step = 1; 01577 01578 if (arg_match(&arg, &codecarg, argi)) 01579 { 01580 int j, k = -1; 01581 01582 for (j = 0; j < sizeof(codecs) / sizeof(codecs[0]); j++) 01583 if (!strcmp(codecs[j].name, arg.val)) 01584 k = j; 01585 01586 if (k >= 0) 01587 global->codec = codecs + k; 01588 else 01589 die("Error: Unrecognized argument (%s) to --codec\n", 01590 arg.val); 01591 01592 } 01593 else if (arg_match(&arg, &passes, argi)) 01594 { 01595 global->passes = arg_parse_uint(&arg); 01596 01597 if (global->passes < 1 || global->passes > 2) 01598 die("Error: Invalid number of passes (%d)\n", global->passes); 01599 } 01600 else if (arg_match(&arg, &pass_arg, argi)) 01601 { 01602 global->pass = arg_parse_uint(&arg); 01603 01604 if (global->pass < 1 || global->pass > 2) 01605 die("Error: Invalid pass selected (%d)\n", 01606 global->pass); 01607 } 01608 else if (arg_match(&arg, &usage, argi)) 01609 global->usage = arg_parse_uint(&arg); 01610 else if (arg_match(&arg, &deadline, argi)) 01611 global->deadline = arg_parse_uint(&arg); 01612 else if (arg_match(&arg, &best_dl, argi)) 01613 global->deadline = VPX_DL_BEST_QUALITY; 01614 else if (arg_match(&arg, &good_dl, argi)) 01615 global->deadline = VPX_DL_GOOD_QUALITY; 01616 else if (arg_match(&arg, &rt_dl, argi)) 01617 global->deadline = VPX_DL_REALTIME; 01618 else if (arg_match(&arg, &use_yv12, argi)) 01619 global->use_i420 = 0; 01620 else if (arg_match(&arg, &use_i420, argi)) 01621 global->use_i420 = 1; 01622 else if (arg_match(&arg, &verbosearg, argi)) 01623 global->verbose = 1; 01624 else if (arg_match(&arg, &limit, argi)) 01625 global->limit = arg_parse_uint(&arg); 01626 else if (arg_match(&arg, &psnrarg, argi)) 01627 global->show_psnr = 1; 01628 else if (arg_match(&arg, &framerate, argi)) 01629 { 01630 global->framerate = arg_parse_rational(&arg); 01631 validate_positive_rational(arg.name, &global->framerate); 01632 global->have_framerate = 1; 01633 } 01634 else if (arg_match(&arg,&out_part, argi)) 01635 global->out_part = 1; 01636 else if (arg_match(&arg, &debugmode, argi)) 01637 global->debug = 1; 01638 else if (arg_match(&arg, &q_hist_n, argi)) 01639 global->show_q_hist_buckets = arg_parse_uint(&arg); 01640 else if (arg_match(&arg, &rate_hist_n, argi)) 01641 global->show_rate_hist_buckets = arg_parse_uint(&arg); 01642 else 01643 argj++; 01644 } 01645 01646 /* Validate global config */ 01647 01648 if (global->pass) 01649 { 01650 /* DWIM: Assume the user meant passes=2 if pass=2 is specified */ 01651 if (global->pass > global->passes) 01652 { 01653 warn("Assuming --pass=%d implies --passes=%d\n", 01654 global->pass, global->pass); 01655 global->passes = global->pass; 01656 } 01657 } 01658 } 01659 01660 01661 void open_input_file(struct input_state *input) 01662 { 01663 unsigned int fourcc; 01664 01665 /* Parse certain options from the input file, if possible */ 01666 input->file = strcmp(input->fn, "-") ? fopen(input->fn, "rb") 01667 : set_binary_mode(stdin); 01668 01669 if (!input->file) 01670 fatal("Failed to open input file"); 01671 01672 /* For RAW input sources, these bytes will applied on the first frame 01673 * in read_frame(). 01674 */ 01675 input->detect.buf_read = fread(input->detect.buf, 1, 4, input->file); 01676 input->detect.position = 0; 01677 01678 if (input->detect.buf_read == 4 01679 && file_is_y4m(input->file, &input->y4m, input->detect.buf)) 01680 { 01681 if (y4m_input_open(&input->y4m, input->file, input->detect.buf, 4) >= 0) 01682 { 01683 input->file_type = FILE_TYPE_Y4M; 01684 input->w = input->y4m.pic_w; 01685 input->h = input->y4m.pic_h; 01686 input->framerate.num = input->y4m.fps_n; 01687 input->framerate.den = input->y4m.fps_d; 01688 input->use_i420 = 0; 01689 } 01690 else 01691 fatal("Unsupported Y4M stream."); 01692 } 01693 else if (input->detect.buf_read == 4 && file_is_ivf(input, &fourcc)) 01694 { 01695 input->file_type = FILE_TYPE_IVF; 01696 switch (fourcc) 01697 { 01698 case 0x32315659: 01699 input->use_i420 = 0; 01700 break; 01701 case 0x30323449: 01702 input->use_i420 = 1; 01703 break; 01704 default: 01705 fatal("Unsupported fourcc (%08x) in IVF", fourcc); 01706 } 01707 } 01708 else 01709 { 01710 input->file_type = FILE_TYPE_RAW; 01711 } 01712 } 01713 01714 01715 static void close_input_file(struct input_state *input) 01716 { 01717 fclose(input->file); 01718 if (input->file_type == FILE_TYPE_Y4M) 01719 y4m_input_close(&input->y4m); 01720 } 01721 01722 static struct stream_state *new_stream(struct global_config *global, 01723 struct stream_state *prev) 01724 { 01725 struct stream_state *stream; 01726 01727 stream = calloc(1, sizeof(*stream)); 01728 if(!stream) 01729 fatal("Failed to allocate new stream."); 01730 if(prev) 01731 { 01732 memcpy(stream, prev, sizeof(*stream)); 01733 stream->index++; 01734 prev->next = stream; 01735 } 01736 else 01737 { 01738 vpx_codec_err_t res; 01739 01740 /* Populate encoder configuration */ 01741 res = vpx_codec_enc_config_default(global->codec->iface, 01742 &stream->config.cfg, 01743 global->usage); 01744 if (res) 01745 fatal("Failed to get config: %s\n", vpx_codec_err_to_string(res)); 01746 01747 /* Change the default timebase to a high enough value so that the 01748 * encoder will always create strictly increasing timestamps. 01749 */ 01750 stream->config.cfg.g_timebase.den = 1000; 01751 01752 /* Never use the library's default resolution, require it be parsed 01753 * from the file or set on the command line. 01754 */ 01755 stream->config.cfg.g_w = 0; 01756 stream->config.cfg.g_h = 0; 01757 01758 /* Initialize remaining stream parameters */ 01759 stream->config.stereo_fmt = STEREO_FORMAT_MONO; 01760 stream->config.write_webm = 1; 01761 stream->ebml.last_pts_ms = -1; 01762 01763 /* Allows removal of the application version from the EBML tags */ 01764 stream->ebml.debug = global->debug; 01765 } 01766 01767 /* Output files must be specified for each stream */ 01768 stream->config.out_fn = NULL; 01769 01770 stream->next = NULL; 01771 return stream; 01772 } 01773 01774 01775 static int parse_stream_params(struct global_config *global, 01776 struct stream_state *stream, 01777 char **argv) 01778 { 01779 char **argi, **argj; 01780 struct arg arg; 01781 static const arg_def_t **ctrl_args = no_args; 01782 static const int *ctrl_args_map = NULL; 01783 struct stream_config *config = &stream->config; 01784 int eos_mark_found = 0; 01785 01786 /* Handle codec specific options */ 01787 if (global->codec->iface == &vpx_codec_vp8_cx_algo) 01788 { 01789 ctrl_args = vp8_args; 01790 ctrl_args_map = vp8_arg_ctrl_map; 01791 } 01792 01793 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) 01794 { 01795 arg.argv_step = 1; 01796 01797 /* Once we've found an end-of-stream marker (--) we want to continue 01798 * shifting arguments but not consuming them. 01799 */ 01800 if (eos_mark_found) 01801 { 01802 argj++; 01803 continue; 01804 } 01805 else if (!strcmp(*argj, "--")) 01806 { 01807 eos_mark_found = 1; 01808 continue; 01809 } 01810 01811 if (0); 01812 else if (arg_match(&arg, &outputfile, argi)) 01813 config->out_fn = arg.val; 01814 else if (arg_match(&arg, &fpf_name, argi)) 01815 config->stats_fn = arg.val; 01816 else if (arg_match(&arg, &use_ivf, argi)) 01817 config->write_webm = 0; 01818 else if (arg_match(&arg, &threads, argi)) 01819 config->cfg.g_threads = arg_parse_uint(&arg); 01820 else if (arg_match(&arg, &profile, argi)) 01821 config->cfg.g_profile = arg_parse_uint(&arg); 01822 else if (arg_match(&arg, &width, argi)) 01823 config->cfg.g_w = arg_parse_uint(&arg); 01824 else if (arg_match(&arg, &height, argi)) 01825 config->cfg.g_h = arg_parse_uint(&arg); 01826 else if (arg_match(&arg, &stereo_mode, argi)) 01827 config->stereo_fmt = arg_parse_enum_or_int(&arg); 01828 else if (arg_match(&arg, &timebase, argi)) 01829 { 01830 config->cfg.g_timebase = arg_parse_rational(&arg); 01831 validate_positive_rational(arg.name, &config->cfg.g_timebase); 01832 } 01833 else if (arg_match(&arg, &error_resilient, argi)) 01834 config->cfg.g_error_resilient = arg_parse_uint(&arg); 01835 else if (arg_match(&arg, &lag_in_frames, argi)) 01836 config->cfg.g_lag_in_frames = arg_parse_uint(&arg); 01837 else if (arg_match(&arg, &dropframe_thresh, argi)) 01838 config->cfg.rc_dropframe_thresh = arg_parse_uint(&arg); 01839 else if (arg_match(&arg, &resize_allowed, argi)) 01840 config->cfg.rc_resize_allowed = arg_parse_uint(&arg); 01841 else if (arg_match(&arg, &resize_up_thresh, argi)) 01842 config->cfg.rc_resize_up_thresh = arg_parse_uint(&arg); 01843 else if (arg_match(&arg, &resize_down_thresh, argi)) 01844 config->cfg.rc_resize_down_thresh = arg_parse_uint(&arg); 01845 else if (arg_match(&arg, &end_usage, argi)) 01846 config->cfg.rc_end_usage = arg_parse_enum_or_int(&arg); 01847 else if (arg_match(&arg, &target_bitrate, argi)) 01848 config->cfg.rc_target_bitrate = arg_parse_uint(&arg); 01849 else if (arg_match(&arg, &min_quantizer, argi)) 01850 config->cfg.rc_min_quantizer = arg_parse_uint(&arg); 01851 else if (arg_match(&arg, &max_quantizer, argi)) 01852 config->cfg.rc_max_quantizer = arg_parse_uint(&arg); 01853 else if (arg_match(&arg, &undershoot_pct, argi)) 01854 config->cfg.rc_undershoot_pct = arg_parse_uint(&arg); 01855 else if (arg_match(&arg, &overshoot_pct, argi)) 01856 config->cfg.rc_overshoot_pct = arg_parse_uint(&arg); 01857 else if (arg_match(&arg, &buf_sz, argi)) 01858 config->cfg.rc_buf_sz = arg_parse_uint(&arg); 01859 else if (arg_match(&arg, &buf_initial_sz, argi)) 01860 config->cfg.rc_buf_initial_sz = arg_parse_uint(&arg); 01861 else if (arg_match(&arg, &buf_optimal_sz, argi)) 01862 config->cfg.rc_buf_optimal_sz = arg_parse_uint(&arg); 01863 else if (arg_match(&arg, &bias_pct, argi)) 01864 { 01865 config->cfg.rc_2pass_vbr_bias_pct = arg_parse_uint(&arg); 01866 01867 if (global->passes < 2) 01868 warn("option %s ignored in one-pass mode.\n", arg.name); 01869 } 01870 else if (arg_match(&arg, &minsection_pct, argi)) 01871 { 01872 config->cfg.rc_2pass_vbr_minsection_pct = arg_parse_uint(&arg); 01873 01874 if (global->passes < 2) 01875 warn("option %s ignored in one-pass mode.\n", arg.name); 01876 } 01877 else if (arg_match(&arg, &maxsection_pct, argi)) 01878 { 01879 config->cfg.rc_2pass_vbr_maxsection_pct = arg_parse_uint(&arg); 01880 01881 if (global->passes < 2) 01882 warn("option %s ignored in one-pass mode.\n", arg.name); 01883 } 01884 else if (arg_match(&arg, &kf_min_dist, argi)) 01885 config->cfg.kf_min_dist = arg_parse_uint(&arg); 01886 else if (arg_match(&arg, &kf_max_dist, argi)) 01887 { 01888 config->cfg.kf_max_dist = arg_parse_uint(&arg); 01889 config->have_kf_max_dist = 1; 01890 } 01891 else if (arg_match(&arg, &kf_disabled, argi)) 01892 config->cfg.kf_mode = VPX_KF_DISABLED; 01893 else 01894 { 01895 int i, match = 0; 01896 01897 for (i = 0; ctrl_args[i]; i++) 01898 { 01899 if (arg_match(&arg, ctrl_args[i], argi)) 01900 { 01901 int j; 01902 match = 1; 01903 01904 /* Point either to the next free element or the first 01905 * instance of this control. 01906 */ 01907 for(j=0; j<config->arg_ctrl_cnt; j++) 01908 if(config->arg_ctrls[j][0] == ctrl_args_map[i]) 01909 break; 01910 01911 /* Update/insert */ 01912 assert(j < ARG_CTRL_CNT_MAX); 01913 if (j < ARG_CTRL_CNT_MAX) 01914 { 01915 config->arg_ctrls[j][0] = ctrl_args_map[i]; 01916 config->arg_ctrls[j][1] = arg_parse_enum_or_int(&arg); 01917 if(j == config->arg_ctrl_cnt) 01918 config->arg_ctrl_cnt++; 01919 } 01920 01921 } 01922 } 01923 01924 if (!match) 01925 argj++; 01926 } 01927 } 01928 01929 return eos_mark_found; 01930 } 01931 01932 01933 #define FOREACH_STREAM(func)\ 01934 do\ 01935 {\ 01936 struct stream_state *stream;\ 01937 \ 01938 for(stream = streams; stream; stream = stream->next)\ 01939 func;\ 01940 }while(0) 01941 01942 01943 static void validate_stream_config(struct stream_state *stream) 01944 { 01945 struct stream_state *streami; 01946 01947 if(!stream->config.cfg.g_w || !stream->config.cfg.g_h) 01948 fatal("Stream %d: Specify stream dimensions with --width (-w) " 01949 " and --height (-h)", stream->index); 01950 01951 for(streami = stream; streami; streami = streami->next) 01952 { 01953 /* All streams require output files */ 01954 if(!streami->config.out_fn) 01955 fatal("Stream %d: Output file is required (specify with -o)", 01956 streami->index); 01957 01958 /* Check for two streams outputting to the same file */ 01959 if(streami != stream) 01960 { 01961 const char *a = stream->config.out_fn; 01962 const char *b = streami->config.out_fn; 01963 if(!strcmp(a,b) && strcmp(a, "/dev/null") && strcmp(a, ":nul")) 01964 fatal("Stream %d: duplicate output file (from stream %d)", 01965 streami->index, stream->index); 01966 } 01967 01968 /* Check for two streams sharing a stats file. */ 01969 if(streami != stream) 01970 { 01971 const char *a = stream->config.stats_fn; 01972 const char *b = streami->config.stats_fn; 01973 if(a && b && !strcmp(a,b)) 01974 fatal("Stream %d: duplicate stats file (from stream %d)", 01975 streami->index, stream->index); 01976 } 01977 } 01978 } 01979 01980 01981 static void set_stream_dimensions(struct stream_state *stream, 01982 unsigned int w, 01983 unsigned int h) 01984 { 01985 if ((stream->config.cfg.g_w && stream->config.cfg.g_w != w) 01986 ||(stream->config.cfg.g_h && stream->config.cfg.g_h != h)) 01987 fatal("Stream %d: Resizing not yet supported", stream->index); 01988 stream->config.cfg.g_w = w; 01989 stream->config.cfg.g_h = h; 01990 } 01991 01992 01993 static void set_default_kf_interval(struct stream_state *stream, 01994 struct global_config *global) 01995 { 01996 /* Use a max keyframe interval of 5 seconds, if none was 01997 * specified on the command line. 01998 */ 01999 if (!stream->config.have_kf_max_dist) 02000 { 02001 double framerate = (double)global->framerate.num/global->framerate.den; 02002 if (framerate > 0.0) 02003 stream->config.cfg.kf_max_dist = 5.0*framerate; 02004 } 02005 } 02006 02007 02008 static void show_stream_config(struct stream_state *stream, 02009 struct global_config *global, 02010 struct input_state *input) 02011 { 02012 02013 #define SHOW(field) \ 02014 fprintf(stderr, " %-28s = %d\n", #field, stream->config.cfg.field) 02015 02016 if(stream->index == 0) 02017 { 02018 fprintf(stderr, "Codec: %s\n", 02019 vpx_codec_iface_name(global->codec->iface)); 02020 fprintf(stderr, "Source file: %s Format: %s\n", input->fn, 02021 input->use_i420 ? "I420" : "YV12"); 02022 } 02023 if(stream->next || stream->index) 02024 fprintf(stderr, "\nStream Index: %d\n", stream->index); 02025 fprintf(stderr, "Destination file: %s\n", stream->config.out_fn); 02026 fprintf(stderr, "Encoder parameters:\n"); 02027 02028 SHOW(g_usage); 02029 SHOW(g_threads); 02030 SHOW(g_profile); 02031 SHOW(g_w); 02032 SHOW(g_h); 02033 SHOW(g_timebase.num); 02034 SHOW(g_timebase.den); 02035 SHOW(g_error_resilient); 02036 SHOW(g_pass); 02037 SHOW(g_lag_in_frames); 02038 SHOW(rc_dropframe_thresh); 02039 SHOW(rc_resize_allowed); 02040 SHOW(rc_resize_up_thresh); 02041 SHOW(rc_resize_down_thresh); 02042 SHOW(rc_end_usage); 02043 SHOW(rc_target_bitrate); 02044 SHOW(rc_min_quantizer); 02045 SHOW(rc_max_quantizer); 02046 SHOW(rc_undershoot_pct); 02047 SHOW(rc_overshoot_pct); 02048 SHOW(rc_buf_sz); 02049 SHOW(rc_buf_initial_sz); 02050 SHOW(rc_buf_optimal_sz); 02051 SHOW(rc_2pass_vbr_bias_pct); 02052 SHOW(rc_2pass_vbr_minsection_pct); 02053 SHOW(rc_2pass_vbr_maxsection_pct); 02054 SHOW(kf_mode); 02055 SHOW(kf_min_dist); 02056 SHOW(kf_max_dist); 02057 } 02058 02059 02060 static void open_output_file(struct stream_state *stream, 02061 struct global_config *global) 02062 { 02063 const char *fn = stream->config.out_fn; 02064 02065 stream->file = strcmp(fn, "-") ? fopen(fn, "wb") : set_binary_mode(stdout); 02066 02067 if (!stream->file) 02068 fatal("Failed to open output file"); 02069 02070 if(stream->config.write_webm && fseek(stream->file, 0, SEEK_CUR)) 02071 fatal("WebM output to pipes not supported."); 02072 02073 if(stream->config.write_webm) 02074 { 02075 stream->ebml.stream = stream->file; 02076 write_webm_file_header(&stream->ebml, &stream->config.cfg, 02077 &global->framerate, 02078 stream->config.stereo_fmt); 02079 } 02080 else 02081 write_ivf_file_header(stream->file, &stream->config.cfg, 02082 global->codec->fourcc, 0); 02083 } 02084 02085 02086 static void close_output_file(struct stream_state *stream, 02087 unsigned int fourcc) 02088 { 02089 if(stream->config.write_webm) 02090 { 02091 write_webm_file_footer(&stream->ebml, stream->hash); 02092 free(stream->ebml.cue_list); 02093 stream->ebml.cue_list = NULL; 02094 } 02095 else 02096 { 02097 if (!fseek(stream->file, 0, SEEK_SET)) 02098 write_ivf_file_header(stream->file, &stream->config.cfg, 02099 fourcc, 02100 stream->frames_out); 02101 } 02102 02103 fclose(stream->file); 02104 } 02105 02106 02107 static void setup_pass(struct stream_state *stream, 02108 struct global_config *global, 02109 int pass) 02110 { 02111 if (stream->config.stats_fn) 02112 { 02113 if (!stats_open_file(&stream->stats, stream->config.stats_fn, 02114 pass)) 02115 fatal("Failed to open statistics store"); 02116 } 02117 else 02118 { 02119 if (!stats_open_mem(&stream->stats, pass)) 02120 fatal("Failed to open statistics store"); 02121 } 02122 02123 stream->config.cfg.g_pass = global->passes == 2 02124 ? pass ? VPX_RC_LAST_PASS : VPX_RC_FIRST_PASS 02125 : VPX_RC_ONE_PASS; 02126 if (pass) 02127 stream->config.cfg.rc_twopass_stats_in = stats_get(&stream->stats); 02128 02129 stream->cx_time = 0; 02130 stream->nbytes = 0; 02131 stream->frames_out = 0; 02132 } 02133 02134 02135 static void initialize_encoder(struct stream_state *stream, 02136 struct global_config *global) 02137 { 02138 int i; 02139 int flags = 0; 02140 02141 flags |= global->show_psnr ? VPX_CODEC_USE_PSNR : 0; 02142 flags |= global->out_part ? VPX_CODEC_USE_OUTPUT_PARTITION : 0; 02143 02144 /* Construct Encoder Context */ 02145 vpx_codec_enc_init(&stream->encoder, global->codec->iface, 02146 &stream->config.cfg, flags); 02147 ctx_exit_on_error(&stream->encoder, "Failed to initialize encoder"); 02148 02149 /* Note that we bypass the vpx_codec_control wrapper macro because 02150 * we're being clever to store the control IDs in an array. Real 02151 * applications will want to make use of the enumerations directly 02152 */ 02153 for (i = 0; i < stream->config.arg_ctrl_cnt; i++) 02154 { 02155 int ctrl = stream->config.arg_ctrls[i][0]; 02156 int value = stream->config.arg_ctrls[i][1]; 02157 if (vpx_codec_control_(&stream->encoder, ctrl, value)) 02158 fprintf(stderr, "Error: Tried to set control %d = %d\n", 02159 ctrl, value); 02160 02161 ctx_exit_on_error(&stream->encoder, "Failed to control codec"); 02162 } 02163 } 02164 02165 02166 static void encode_frame(struct stream_state *stream, 02167 struct global_config *global, 02168 struct vpx_image *img, 02169 unsigned int frames_in) 02170 { 02171 vpx_codec_pts_t frame_start, next_frame_start; 02172 struct vpx_codec_enc_cfg *cfg = &stream->config.cfg; 02173 struct vpx_usec_timer timer; 02174 02175 frame_start = (cfg->g_timebase.den * (int64_t)(frames_in - 1) 02176 * global->framerate.den) 02177 / cfg->g_timebase.num / global->framerate.num; 02178 next_frame_start = (cfg->g_timebase.den * (int64_t)(frames_in) 02179 * global->framerate.den) 02180 / cfg->g_timebase.num / global->framerate.num; 02181 vpx_usec_timer_start(&timer); 02182 vpx_codec_encode(&stream->encoder, img, frame_start, 02183 next_frame_start - frame_start, 02184 0, global->deadline); 02185 vpx_usec_timer_mark(&timer); 02186 stream->cx_time += vpx_usec_timer_elapsed(&timer); 02187 ctx_exit_on_error(&stream->encoder, "Stream %d: Failed to encode frame", 02188 stream->index); 02189 } 02190 02191 02192 static void update_quantizer_histogram(struct stream_state *stream) 02193 { 02194 if(stream->config.cfg.g_pass != VPX_RC_FIRST_PASS) 02195 { 02196 int q; 02197 02198 vpx_codec_control(&stream->encoder, VP8E_GET_LAST_QUANTIZER_64, &q); 02199 ctx_exit_on_error(&stream->encoder, "Failed to read quantizer"); 02200 stream->counts[q]++; 02201 } 02202 } 02203 02204 02205 static void get_cx_data(struct stream_state *stream, 02206 struct global_config *global, 02207 int *got_data) 02208 { 02209 const vpx_codec_cx_pkt_t *pkt; 02210 const struct vpx_codec_enc_cfg *cfg = &stream->config.cfg; 02211 vpx_codec_iter_t iter = NULL; 02212 02213 while ((pkt = vpx_codec_get_cx_data(&stream->encoder, &iter))) 02214 { 02215 static size_t fsize = 0; 02216 static off_t ivf_header_pos = 0; 02217 02218 *got_data = 1; 02219 02220 switch (pkt->kind) 02221 { 02222 case VPX_CODEC_CX_FRAME_PKT: 02223 if (!(pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT)) 02224 { 02225 stream->frames_out++; 02226 } 02227 fprintf(stderr, " %6luF", 02228 (unsigned long)pkt->data.frame.sz); 02229 02230 update_rate_histogram(&stream->rate_hist, cfg, pkt); 02231 if(stream->config.write_webm) 02232 { 02233 /* Update the hash */ 02234 if(!stream->ebml.debug) 02235 stream->hash = murmur(pkt->data.frame.buf, 02236 pkt->data.frame.sz, stream->hash); 02237 02238 write_webm_block(&stream->ebml, cfg, pkt); 02239 } 02240 else 02241 { 02242 if (pkt->data.frame.partition_id <= 0) 02243 { 02244 ivf_header_pos = ftello(stream->file); 02245 fsize = pkt->data.frame.sz; 02246 02247 write_ivf_frame_header(stream->file, pkt); 02248 } 02249 else 02250 { 02251 fsize += pkt->data.frame.sz; 02252 02253 if (!(pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT)) 02254 { 02255 off_t currpos = ftello(stream->file); 02256 fseeko(stream->file, ivf_header_pos, SEEK_SET); 02257 write_ivf_frame_size(stream->file, fsize); 02258 fseeko(stream->file, currpos, SEEK_SET); 02259 } 02260 } 02261 02262 fwrite(pkt->data.frame.buf, 1, 02263 pkt->data.frame.sz, stream->file); 02264 } 02265 stream->nbytes += pkt->data.raw.sz; 02266 break; 02267 case VPX_CODEC_STATS_PKT: 02268 stream->frames_out++; 02269 fprintf(stderr, " %6luS", 02270 (unsigned long)pkt->data.twopass_stats.sz); 02271 stats_write(&stream->stats, 02272 pkt->data.twopass_stats.buf, 02273 pkt->data.twopass_stats.sz); 02274 stream->nbytes += pkt->data.raw.sz; 02275 break; 02276 case VPX_CODEC_PSNR_PKT: 02277 02278 if (global->show_psnr) 02279 { 02280 int i; 02281 02282 stream->psnr_sse_total += pkt->data.psnr.sse[0]; 02283 stream->psnr_samples_total += pkt->data.psnr.samples[0]; 02284 for (i = 0; i < 4; i++) 02285 { 02286 fprintf(stderr, "%.3lf ", pkt->data.psnr.psnr[i]); 02287 stream->psnr_totals[i] += pkt->data.psnr.psnr[i]; 02288 } 02289 stream->psnr_count++; 02290 } 02291 02292 break; 02293 default: 02294 break; 02295 } 02296 } 02297 } 02298 02299 02300 static void show_psnr(struct stream_state *stream) 02301 { 02302 int i; 02303 double ovpsnr; 02304 02305 if (!stream->psnr_count) 02306 return; 02307 02308 fprintf(stderr, "Stream %d PSNR (Overall/Avg/Y/U/V)", stream->index); 02309 ovpsnr = vp8_mse2psnr(stream->psnr_samples_total, 255.0, 02310 stream->psnr_sse_total); 02311 fprintf(stderr, " %.3lf", ovpsnr); 02312 02313 for (i = 0; i < 4; i++) 02314 { 02315 fprintf(stderr, " %.3lf", stream->psnr_totals[i]/stream->psnr_count); 02316 } 02317 fprintf(stderr, "\n"); 02318 } 02319 02320 02321 float usec_to_fps(uint64_t usec, unsigned int frames) 02322 { 02323 return usec > 0 ? (float)frames * 1000000.0 / (float)usec : 0; 02324 } 02325 02326 02327 int main(int argc, const char **argv_) 02328 { 02329 int pass; 02330 vpx_image_t raw; 02331 int frame_avail, got_data; 02332 02333 struct input_state input = {0}; 02334 struct global_config global; 02335 struct stream_state *streams = NULL; 02336 char **argv, **argi; 02337 unsigned long cx_time = 0; 02338 int stream_cnt = 0; 02339 02340 exec_name = argv_[0]; 02341 02342 if (argc < 3) 02343 usage_exit(); 02344 02345 /* Setup default input stream settings */ 02346 input.framerate.num = 30; 02347 input.framerate.den = 1; 02348 input.use_i420 = 1; 02349 02350 /* First parse the global configuration values, because we want to apply 02351 * other parameters on top of the default configuration provided by the 02352 * codec. 02353 */ 02354 argv = argv_dup(argc - 1, argv_ + 1); 02355 parse_global_config(&global, argv); 02356 02357 { 02358 /* Now parse each stream's parameters. Using a local scope here 02359 * due to the use of 'stream' as loop variable in FOREACH_STREAM 02360 * loops 02361 */ 02362 struct stream_state *stream = NULL; 02363 02364 do 02365 { 02366 stream = new_stream(&global, stream); 02367 stream_cnt++; 02368 if(!streams) 02369 streams = stream; 02370 } while(parse_stream_params(&global, stream, argv)); 02371 } 02372 02373 /* Check for unrecognized options */ 02374 for (argi = argv; *argi; argi++) 02375 if (argi[0][0] == '-' && argi[0][1]) 02376 die("Error: Unrecognized option %s\n", *argi); 02377 02378 /* Handle non-option arguments */ 02379 input.fn = argv[0]; 02380 02381 if (!input.fn) 02382 usage_exit(); 02383 02384 for (pass = global.pass ? global.pass - 1 : 0; pass < global.passes; pass++) 02385 { 02386 int frames_in = 0; 02387 02388 open_input_file(&input); 02389 02390 /* If the input file doesn't specify its w/h (raw files), try to get 02391 * the data from the first stream's configuration. 02392 */ 02393 if(!input.w || !input.h) 02394 FOREACH_STREAM({ 02395 if(stream->config.cfg.g_w && stream->config.cfg.g_h) 02396 { 02397 input.w = stream->config.cfg.g_w; 02398 input.h = stream->config.cfg.g_h; 02399 break; 02400 } 02401 }); 02402 02403 /* Update stream configurations from the input file's parameters */ 02404 FOREACH_STREAM(set_stream_dimensions(stream, input.w, input.h)); 02405 FOREACH_STREAM(validate_stream_config(stream)); 02406 02407 /* Ensure that --passes and --pass are consistent. If --pass is set and 02408 * --passes=2, ensure --fpf was set. 02409 */ 02410 if (global.pass && global.passes == 2) 02411 FOREACH_STREAM({ 02412 if(!stream->config.stats_fn) 02413 die("Stream %d: Must specify --fpf when --pass=%d" 02414 " and --passes=2\n", stream->index, global.pass); 02415 }); 02416 02417 02418 /* Use the frame rate from the file only if none was specified 02419 * on the command-line. 02420 */ 02421 if (!global.have_framerate) 02422 global.framerate = input.framerate; 02423 02424 FOREACH_STREAM(set_default_kf_interval(stream, &global)); 02425 02426 /* Show configuration */ 02427 if (global.verbose && pass == 0) 02428 FOREACH_STREAM(show_stream_config(stream, &global, &input)); 02429 02430 if(pass == (global.pass ? global.pass - 1 : 0)) { 02431 if (input.file_type == FILE_TYPE_Y4M) 02432 /*The Y4M reader does its own allocation. 02433 Just initialize this here to avoid problems if we never read any 02434 frames.*/ 02435 memset(&raw, 0, sizeof(raw)); 02436 else 02437 vpx_img_alloc(&raw, 02438 input.use_i420 ? VPX_IMG_FMT_I420 02439 : VPX_IMG_FMT_YV12, 02440 input.w, input.h, 1); 02441 02442 FOREACH_STREAM(init_rate_histogram(&stream->rate_hist, 02443 &stream->config.cfg, 02444 &global.framerate)); 02445 } 02446 02447 FOREACH_STREAM(open_output_file(stream, &global)); 02448 FOREACH_STREAM(setup_pass(stream, &global, pass)); 02449 FOREACH_STREAM(initialize_encoder(stream, &global)); 02450 02451 frame_avail = 1; 02452 got_data = 0; 02453 02454 while (frame_avail || got_data) 02455 { 02456 struct vpx_usec_timer timer; 02457 02458 if (!global.limit || frames_in < global.limit) 02459 { 02460 frame_avail = read_frame(&input, &raw); 02461 02462 if (frame_avail) 02463 frames_in++; 02464 02465 if(stream_cnt == 1) 02466 fprintf(stderr, 02467 "\rPass %d/%d frame %4d/%-4d %7"PRId64"B \033[K", 02468 pass + 1, global.passes, frames_in, 02469 streams->frames_out, (int64_t)streams->nbytes); 02470 else 02471 fprintf(stderr, 02472 "\rPass %d/%d frame %4d %7lu %s (%.2f fps)\033[K", 02473 pass + 1, global.passes, frames_in, 02474 cx_time > 9999999 ? cx_time / 1000 : cx_time, 02475 cx_time > 9999999 ? "ms" : "us", 02476 usec_to_fps(cx_time, frames_in)); 02477 02478 } 02479 else 02480 frame_avail = 0; 02481 02482 vpx_usec_timer_start(&timer); 02483 FOREACH_STREAM(encode_frame(stream, &global, 02484 frame_avail ? &raw : NULL, 02485 frames_in)); 02486 vpx_usec_timer_mark(&timer); 02487 cx_time += vpx_usec_timer_elapsed(&timer); 02488 02489 FOREACH_STREAM(update_quantizer_histogram(stream)); 02490 02491 got_data = 0; 02492 FOREACH_STREAM(get_cx_data(stream, &global, &got_data)); 02493 02494 fflush(stdout); 02495 } 02496 02497 if(stream_cnt > 1) 02498 fprintf(stderr, "\n"); 02499 02500 FOREACH_STREAM(fprintf( 02501 stderr, 02502 "\rPass %d/%d frame %4d/%-4d %7"PRId64"B %7lub/f %7"PRId64"b/s" 02503 " %7"PRId64" %s (%.2f fps)\033[K\n", pass + 1, 02504 global.passes, frames_in, stream->frames_out, (int64_t)stream->nbytes, 02505 frames_in ? (unsigned long)(stream->nbytes * 8 / frames_in) : 0, 02506 frames_in ? (int64_t)stream->nbytes * 8 02507 * (int64_t)global.framerate.num / global.framerate.den 02508 / frames_in 02509 : 0, 02510 stream->cx_time > 9999999 ? stream->cx_time / 1000 : stream->cx_time, 02511 stream->cx_time > 9999999 ? "ms" : "us", 02512 usec_to_fps(stream->cx_time, frames_in)); 02513 ); 02514 02515 if (global.show_psnr) 02516 FOREACH_STREAM(show_psnr(stream)); 02517 02518 FOREACH_STREAM(vpx_codec_destroy(&stream->encoder)); 02519 02520 close_input_file(&input); 02521 02522 FOREACH_STREAM(close_output_file(stream, global.codec->fourcc)); 02523 02524 FOREACH_STREAM(stats_close(&stream->stats, global.passes-1)); 02525 02526 if (global.pass) 02527 break; 02528 } 02529 02530 if (global.show_q_hist_buckets) 02531 FOREACH_STREAM(show_q_histogram(stream->counts, 02532 global.show_q_hist_buckets)); 02533 02534 if (global.show_rate_hist_buckets) 02535 FOREACH_STREAM(show_rate_histogram(&stream->rate_hist, 02536 &stream->config.cfg, 02537 global.show_rate_hist_buckets)); 02538 FOREACH_STREAM(destroy_rate_histogram(&stream->rate_hist)); 02539 02540 vpx_img_free(&raw); 02541 free(argv); 02542 free(streams); 02543 return EXIT_SUCCESS; 02544 }