1 /* 2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved. 3 * 4 * Use of this source code is governed by a BSD-style license 5 * that can be found in the LICENSE file in the root of the source 6 * tree. An additional intellectual property rights grant can be found 7 * in the file PATENTS. All contributing project authors may 8 * be found in the AUTHORS file in the root of the source tree. 9 */ 10 11 #include "./vpx_config.h" 12 13 #if defined(_WIN32) || defined(__OS2__) || !CONFIG_OS_SUPPORT 14 #define USE_POSIX_MMAP 0 15 #else 16 #define USE_POSIX_MMAP 1 17 #endif 18 19 #include <math.h> 20 #include <stdio.h> 21 #include <stdlib.h> 22 #include <stdarg.h> 23 #include <string.h> 24 #include <limits.h> 25 #include <assert.h> 26 #include "vpx/vpx_encoder.h" 27 #if CONFIG_DECODERS 28 #include "vpx/vpx_decoder.h" 29 #endif 30 #if USE_POSIX_MMAP 31 #include <sys/types.h> 32 #include <sys/stat.h> 33 #include <sys/mman.h> 34 #include <fcntl.h> 35 #include <unistd.h> 36 #endif 37 38 #include "third_party/libyuv/include/libyuv/scale.h" 39 40 #if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER 41 #include "vpx/vp8cx.h" 42 #endif 43 #if CONFIG_VP8_DECODER || CONFIG_VP9_DECODER 44 #include "vpx/vp8dx.h" 45 #endif 46 47 #include "./tools_common.h" 48 #include "vpx_ports/mem_ops.h" 49 #include "vpx_ports/vpx_timer.h" 50 #include "./vpxstats.h" 51 #include "./webmenc.h" 52 #include "./y4minput.h" 53 54 /* Swallow warnings about unused results of fread/fwrite */ 55 static size_t wrap_fread(void *ptr, size_t size, size_t nmemb, 56 FILE *stream) { 57 return fread(ptr, size, nmemb, stream); 58 } 59 #define fread wrap_fread 60 61 static size_t wrap_fwrite(const void *ptr, size_t size, size_t nmemb, 62 FILE *stream) { 63 return fwrite(ptr, size, nmemb, stream); 64 } 65 #define fwrite wrap_fwrite 66 67 68 static const char *exec_name; 69 70 static const struct codec_item { 71 char const *name; 72 const vpx_codec_iface_t *(*iface)(void); 73 const vpx_codec_iface_t *(*dx_iface)(void); 74 unsigned int fourcc; 75 } codecs[] = { 76 #if CONFIG_VP8_ENCODER && CONFIG_VP8_DECODER 77 {"vp8", &vpx_codec_vp8_cx, &vpx_codec_vp8_dx, VP8_FOURCC}, 78 #elif CONFIG_VP8_ENCODER && !CONFIG_VP8_DECODER 79 {"vp8", &vpx_codec_vp8_cx, NULL, VP8_FOURCC}, 80 #endif 81 #if CONFIG_VP9_ENCODER && CONFIG_VP9_DECODER 82 {"vp9", &vpx_codec_vp9_cx, &vpx_codec_vp9_dx, VP9_FOURCC}, 83 #elif CONFIG_VP9_ENCODER && !CONFIG_VP9_DECODER 84 {"vp9", &vpx_codec_vp9_cx, NULL, VP9_FOURCC}, 85 #endif 86 }; 87 88 static void warn_or_exit_on_errorv(vpx_codec_ctx_t *ctx, int fatal, 89 const char *s, va_list ap) { 90 if (ctx->err) { 91 const char *detail = vpx_codec_error_detail(ctx); 92 93 vfprintf(stderr, s, ap); 94 fprintf(stderr, ": %s\n", vpx_codec_error(ctx)); 95 96 if (detail) 97 fprintf(stderr, " %s\n", detail); 98 99 if (fatal) 100 exit(EXIT_FAILURE); 101 } 102 } 103 104 static void ctx_exit_on_error(vpx_codec_ctx_t *ctx, const char *s, ...) { 105 va_list ap; 106 107 va_start(ap, s); 108 warn_or_exit_on_errorv(ctx, 1, s, ap); 109 va_end(ap); 110 } 111 112 static void warn_or_exit_on_error(vpx_codec_ctx_t *ctx, int fatal, 113 const char *s, ...) { 114 va_list ap; 115 116 va_start(ap, s); 117 warn_or_exit_on_errorv(ctx, fatal, s, ap); 118 va_end(ap); 119 } 120 121 enum video_file_type { 122 FILE_TYPE_RAW, 123 FILE_TYPE_IVF, 124 FILE_TYPE_Y4M 125 }; 126 127 struct detect_buffer { 128 char buf[4]; 129 size_t buf_read; 130 size_t position; 131 }; 132 133 134 struct input_state { 135 char *fn; 136 FILE *file; 137 off_t length; 138 y4m_input y4m; 139 struct detect_buffer detect; 140 enum video_file_type file_type; 141 unsigned int w; 142 unsigned int h; 143 struct vpx_rational framerate; 144 int use_i420; 145 int only_i420; 146 }; 147 148 #define IVF_FRAME_HDR_SZ (4+8) /* 4 byte size + 8 byte timestamp */ 149 static int read_frame(struct input_state *input, vpx_image_t *img) { 150 FILE *f = input->file; 151 enum video_file_type file_type = input->file_type; 152 y4m_input *y4m = &input->y4m; 153 struct detect_buffer *detect = &input->detect; 154 int plane = 0; 155 int shortread = 0; 156 157 if (file_type == FILE_TYPE_Y4M) { 158 if (y4m_input_fetch_frame(y4m, f, img) < 1) 159 return 0; 160 } else { 161 if (file_type == FILE_TYPE_IVF) { 162 char junk[IVF_FRAME_HDR_SZ]; 163 164 /* Skip the frame header. We know how big the frame should be. See 165 * write_ivf_frame_header() for documentation on the frame header 166 * layout. 167 */ 168 (void) fread(junk, 1, IVF_FRAME_HDR_SZ, f); 169 } 170 171 for (plane = 0; plane < 3; plane++) { 172 unsigned char *ptr; 173 int w = (plane ? (1 + img->d_w) / 2 : img->d_w); 174 int h = (plane ? (1 + img->d_h) / 2 : img->d_h); 175 int r; 176 177 /* Determine the correct plane based on the image format. The for-loop 178 * always counts in Y,U,V order, but this may not match the order of 179 * the data on disk. 180 */ 181 switch (plane) { 182 case 1: 183 ptr = img->planes[img->fmt == VPX_IMG_FMT_YV12 ? VPX_PLANE_V : VPX_PLANE_U]; 184 break; 185 case 2: 186 ptr = img->planes[img->fmt == VPX_IMG_FMT_YV12 ? VPX_PLANE_U : VPX_PLANE_V]; 187 break; 188 default: 189 ptr = img->planes[plane]; 190 } 191 192 for (r = 0; r < h; r++) { 193 size_t needed = w; 194 size_t buf_position = 0; 195 const size_t left = detect->buf_read - detect->position; 196 if (left > 0) { 197 const size_t more = (left < needed) ? left : needed; 198 memcpy(ptr, detect->buf + detect->position, more); 199 buf_position = more; 200 needed -= more; 201 detect->position += more; 202 } 203 if (needed > 0) { 204 shortread |= (fread(ptr + buf_position, 1, needed, f) < needed); 205 } 206 207 ptr += img->stride[plane]; 208 } 209 } 210 } 211 212 return !shortread; 213 } 214 215 216 unsigned int file_is_y4m(FILE *infile, 217 y4m_input *y4m, 218 char detect[4]) { 219 if (memcmp(detect, "YUV4", 4) == 0) { 220 return 1; 221 } 222 return 0; 223 } 224 225 #define IVF_FILE_HDR_SZ (32) 226 unsigned int file_is_ivf(struct input_state *input, 227 unsigned int *fourcc) { 228 char raw_hdr[IVF_FILE_HDR_SZ]; 229 int is_ivf = 0; 230 FILE *infile = input->file; 231 unsigned int *width = &input->w; 232 unsigned int *height = &input->h; 233 struct detect_buffer *detect = &input->detect; 234 235 if (memcmp(detect->buf, "DKIF", 4) != 0) 236 return 0; 237 238 /* See write_ivf_file_header() for more documentation on the file header 239 * layout. 240 */ 241 if (fread(raw_hdr + 4, 1, IVF_FILE_HDR_SZ - 4, infile) 242 == IVF_FILE_HDR_SZ - 4) { 243 { 244 is_ivf = 1; 245 246 if (mem_get_le16(raw_hdr + 4) != 0) 247 warn("Unrecognized IVF version! This file may not decode " 248 "properly."); 249 250 *fourcc = mem_get_le32(raw_hdr + 8); 251 } 252 } 253 254 if (is_ivf) { 255 *width = mem_get_le16(raw_hdr + 12); 256 *height = mem_get_le16(raw_hdr + 14); 257 detect->position = 4; 258 } 259 260 return is_ivf; 261 } 262 263 264 static void write_ivf_file_header(FILE *outfile, 265 const vpx_codec_enc_cfg_t *cfg, 266 unsigned int fourcc, 267 int frame_cnt) { 268 char header[32]; 269 270 if (cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS) 271 return; 272 273 header[0] = 'D'; 274 header[1] = 'K'; 275 header[2] = 'I'; 276 header[3] = 'F'; 277 mem_put_le16(header + 4, 0); /* version */ 278 mem_put_le16(header + 6, 32); /* headersize */ 279 mem_put_le32(header + 8, fourcc); /* headersize */ 280 mem_put_le16(header + 12, cfg->g_w); /* width */ 281 mem_put_le16(header + 14, cfg->g_h); /* height */ 282 mem_put_le32(header + 16, cfg->g_timebase.den); /* rate */ 283 mem_put_le32(header + 20, cfg->g_timebase.num); /* scale */ 284 mem_put_le32(header + 24, frame_cnt); /* length */ 285 mem_put_le32(header + 28, 0); /* unused */ 286 287 (void) fwrite(header, 1, 32, outfile); 288 } 289 290 291 static void write_ivf_frame_header(FILE *outfile, 292 const vpx_codec_cx_pkt_t *pkt) { 293 char header[12]; 294 vpx_codec_pts_t pts; 295 296 if (pkt->kind != VPX_CODEC_CX_FRAME_PKT) 297 return; 298 299 pts = pkt->data.frame.pts; 300 mem_put_le32(header, (int)pkt->data.frame.sz); 301 mem_put_le32(header + 4, pts & 0xFFFFFFFF); 302 mem_put_le32(header + 8, pts >> 32); 303 304 (void) fwrite(header, 1, 12, outfile); 305 } 306 307 static void write_ivf_frame_size(FILE *outfile, size_t size) { 308 char header[4]; 309 mem_put_le32(header, (int)size); 310 (void) fwrite(header, 1, 4, outfile); 311 } 312 313 314 315 /* Murmur hash derived from public domain reference implementation at 316 * http:// sites.google.com/site/murmurhash/ 317 */ 318 static unsigned int murmur(const void *key, int len, unsigned int seed) { 319 const unsigned int m = 0x5bd1e995; 320 const int r = 24; 321 322 unsigned int h = seed ^ len; 323 324 const unsigned char *data = (const unsigned char *)key; 325 326 while (len >= 4) { 327 unsigned int k; 328 329 k = (unsigned int)data[0]; 330 k |= (unsigned int)data[1] << 8; 331 k |= (unsigned int)data[2] << 16; 332 k |= (unsigned int)data[3] << 24; 333 334 k *= m; 335 k ^= k >> r; 336 k *= m; 337 338 h *= m; 339 h ^= k; 340 341 data += 4; 342 len -= 4; 343 } 344 345 switch (len) { 346 case 3: 347 h ^= data[2] << 16; 348 case 2: 349 h ^= data[1] << 8; 350 case 1: 351 h ^= data[0]; 352 h *= m; 353 }; 354 355 h ^= h >> 13; 356 h *= m; 357 h ^= h >> 15; 358 359 return h; 360 } 361 362 363 #include "args.h" 364 static const arg_def_t debugmode = ARG_DEF("D", "debug", 0, 365 "Debug mode (makes output deterministic)"); 366 static const arg_def_t outputfile = ARG_DEF("o", "output", 1, 367 "Output filename"); 368 static const arg_def_t use_yv12 = ARG_DEF(NULL, "yv12", 0, 369 "Input file is YV12 "); 370 static const arg_def_t use_i420 = ARG_DEF(NULL, "i420", 0, 371 "Input file is I420 (default)"); 372 static const arg_def_t codecarg = ARG_DEF(NULL, "codec", 1, 373 "Codec to use"); 374 static const arg_def_t passes = ARG_DEF("p", "passes", 1, 375 "Number of passes (1/2)"); 376 static const arg_def_t pass_arg = ARG_DEF(NULL, "pass", 1, 377 "Pass to execute (1/2)"); 378 static const arg_def_t fpf_name = ARG_DEF(NULL, "fpf", 1, 379 "First pass statistics file name"); 380 static const arg_def_t limit = ARG_DEF(NULL, "limit", 1, 381 "Stop encoding after n input frames"); 382 static const arg_def_t skip = ARG_DEF(NULL, "skip", 1, 383 "Skip the first n input frames"); 384 static const arg_def_t deadline = ARG_DEF("d", "deadline", 1, 385 "Deadline per frame (usec)"); 386 static const arg_def_t best_dl = ARG_DEF(NULL, "best", 0, 387 "Use Best Quality Deadline"); 388 static const arg_def_t good_dl = ARG_DEF(NULL, "good", 0, 389 "Use Good Quality Deadline"); 390 static const arg_def_t rt_dl = ARG_DEF(NULL, "rt", 0, 391 "Use Realtime Quality Deadline"); 392 static const arg_def_t quietarg = ARG_DEF("q", "quiet", 0, 393 "Do not print encode progress"); 394 static const arg_def_t verbosearg = ARG_DEF("v", "verbose", 0, 395 "Show encoder parameters"); 396 static const arg_def_t psnrarg = ARG_DEF(NULL, "psnr", 0, 397 "Show PSNR in status line"); 398 enum TestDecodeFatality { 399 TEST_DECODE_OFF, 400 TEST_DECODE_FATAL, 401 TEST_DECODE_WARN, 402 }; 403 static const struct arg_enum_list test_decode_enum[] = { 404 {"off", TEST_DECODE_OFF}, 405 {"fatal", TEST_DECODE_FATAL}, 406 {"warn", TEST_DECODE_WARN}, 407 {NULL, 0} 408 }; 409 static const arg_def_t recontest = ARG_DEF_ENUM(NULL, "test-decode", 1, 410 "Test encode/decode mismatch", 411 test_decode_enum); 412 static const arg_def_t framerate = ARG_DEF(NULL, "fps", 1, 413 "Stream frame rate (rate/scale)"); 414 static const arg_def_t use_ivf = ARG_DEF(NULL, "ivf", 0, 415 "Output IVF (default is WebM)"); 416 static const arg_def_t out_part = ARG_DEF("P", "output-partitions", 0, 417 "Makes encoder output partitions. Requires IVF output!"); 418 static const arg_def_t q_hist_n = ARG_DEF(NULL, "q-hist", 1, 419 "Show quantizer histogram (n-buckets)"); 420 static const arg_def_t rate_hist_n = ARG_DEF(NULL, "rate-hist", 1, 421 "Show rate histogram (n-buckets)"); 422 static const arg_def_t *main_args[] = { 423 &debugmode, 424 &outputfile, &codecarg, &passes, &pass_arg, &fpf_name, &limit, &skip, 425 &deadline, &best_dl, &good_dl, &rt_dl, 426 &quietarg, &verbosearg, &psnrarg, &use_ivf, &out_part, &q_hist_n, &rate_hist_n, 427 NULL 428 }; 429 430 static const arg_def_t usage = ARG_DEF("u", "usage", 1, 431 "Usage profile number to use"); 432 static const arg_def_t threads = ARG_DEF("t", "threads", 1, 433 "Max number of threads to use"); 434 static const arg_def_t profile = ARG_DEF(NULL, "profile", 1, 435 "Bitstream profile number to use"); 436 static const arg_def_t width = ARG_DEF("w", "width", 1, 437 "Frame width"); 438 static const arg_def_t height = ARG_DEF("h", "height", 1, 439 "Frame height"); 440 static const struct arg_enum_list stereo_mode_enum[] = { 441 {"mono", STEREO_FORMAT_MONO}, 442 {"left-right", STEREO_FORMAT_LEFT_RIGHT}, 443 {"bottom-top", STEREO_FORMAT_BOTTOM_TOP}, 444 {"top-bottom", STEREO_FORMAT_TOP_BOTTOM}, 445 {"right-left", STEREO_FORMAT_RIGHT_LEFT}, 446 {NULL, 0} 447 }; 448 static const arg_def_t stereo_mode = ARG_DEF_ENUM(NULL, "stereo-mode", 1, 449 "Stereo 3D video format", stereo_mode_enum); 450 static const arg_def_t timebase = ARG_DEF(NULL, "timebase", 1, 451 "Output timestamp precision (fractional seconds)"); 452 static const arg_def_t error_resilient = ARG_DEF(NULL, "error-resilient", 1, 453 "Enable error resiliency features"); 454 static const arg_def_t lag_in_frames = ARG_DEF(NULL, "lag-in-frames", 1, 455 "Max number of frames to lag"); 456 457 static const arg_def_t *global_args[] = { 458 &use_yv12, &use_i420, &usage, &threads, &profile, 459 &width, &height, &stereo_mode, &timebase, &framerate, 460 &error_resilient, 461 &lag_in_frames, NULL 462 }; 463 464 static const arg_def_t dropframe_thresh = ARG_DEF(NULL, "drop-frame", 1, 465 "Temporal resampling threshold (buf %)"); 466 static const arg_def_t resize_allowed = ARG_DEF(NULL, "resize-allowed", 1, 467 "Spatial resampling enabled (bool)"); 468 static const arg_def_t resize_up_thresh = ARG_DEF(NULL, "resize-up", 1, 469 "Upscale threshold (buf %)"); 470 static const arg_def_t resize_down_thresh = ARG_DEF(NULL, "resize-down", 1, 471 "Downscale threshold (buf %)"); 472 static const struct arg_enum_list end_usage_enum[] = { 473 {"vbr", VPX_VBR}, 474 {"cbr", VPX_CBR}, 475 {"cq", VPX_CQ}, 476 {"q", VPX_Q}, 477 {NULL, 0} 478 }; 479 static const arg_def_t end_usage = ARG_DEF_ENUM(NULL, "end-usage", 1, 480 "Rate control mode", end_usage_enum); 481 static const arg_def_t target_bitrate = ARG_DEF(NULL, "target-bitrate", 1, 482 "Bitrate (kbps)"); 483 static const arg_def_t min_quantizer = ARG_DEF(NULL, "min-q", 1, 484 "Minimum (best) quantizer"); 485 static const arg_def_t max_quantizer = ARG_DEF(NULL, "max-q", 1, 486 "Maximum (worst) quantizer"); 487 static const arg_def_t undershoot_pct = ARG_DEF(NULL, "undershoot-pct", 1, 488 "Datarate undershoot (min) target (%)"); 489 static const arg_def_t overshoot_pct = ARG_DEF(NULL, "overshoot-pct", 1, 490 "Datarate overshoot (max) target (%)"); 491 static const arg_def_t buf_sz = ARG_DEF(NULL, "buf-sz", 1, 492 "Client buffer size (ms)"); 493 static const arg_def_t buf_initial_sz = ARG_DEF(NULL, "buf-initial-sz", 1, 494 "Client initial buffer size (ms)"); 495 static const arg_def_t buf_optimal_sz = ARG_DEF(NULL, "buf-optimal-sz", 1, 496 "Client optimal buffer size (ms)"); 497 static const arg_def_t *rc_args[] = { 498 &dropframe_thresh, &resize_allowed, &resize_up_thresh, &resize_down_thresh, 499 &end_usage, &target_bitrate, &min_quantizer, &max_quantizer, 500 &undershoot_pct, &overshoot_pct, &buf_sz, &buf_initial_sz, &buf_optimal_sz, 501 NULL 502 }; 503 504 505 static const arg_def_t bias_pct = ARG_DEF(NULL, "bias-pct", 1, 506 "CBR/VBR bias (0=CBR, 100=VBR)"); 507 static const arg_def_t minsection_pct = ARG_DEF(NULL, "minsection-pct", 1, 508 "GOP min bitrate (% of target)"); 509 static const arg_def_t maxsection_pct = ARG_DEF(NULL, "maxsection-pct", 1, 510 "GOP max bitrate (% of target)"); 511 static const arg_def_t *rc_twopass_args[] = { 512 &bias_pct, &minsection_pct, &maxsection_pct, NULL 513 }; 514 515 516 static const arg_def_t kf_min_dist = ARG_DEF(NULL, "kf-min-dist", 1, 517 "Minimum keyframe interval (frames)"); 518 static const arg_def_t kf_max_dist = ARG_DEF(NULL, "kf-max-dist", 1, 519 "Maximum keyframe interval (frames)"); 520 static const arg_def_t kf_disabled = ARG_DEF(NULL, "disable-kf", 0, 521 "Disable keyframe placement"); 522 static const arg_def_t *kf_args[] = { 523 &kf_min_dist, &kf_max_dist, &kf_disabled, NULL 524 }; 525 526 527 static const arg_def_t noise_sens = ARG_DEF(NULL, "noise-sensitivity", 1, 528 "Noise sensitivity (frames to blur)"); 529 static const arg_def_t sharpness = ARG_DEF(NULL, "sharpness", 1, 530 "Filter sharpness (0-7)"); 531 static const arg_def_t static_thresh = ARG_DEF(NULL, "static-thresh", 1, 532 "Motion detection threshold"); 533 static const arg_def_t cpu_used = ARG_DEF(NULL, "cpu-used", 1, 534 "CPU Used (-16..16)"); 535 static const arg_def_t token_parts = ARG_DEF(NULL, "token-parts", 1, 536 "Number of token partitions to use, log2"); 537 static const arg_def_t tile_cols = ARG_DEF(NULL, "tile-columns", 1, 538 "Number of tile columns to use, log2"); 539 static const arg_def_t tile_rows = ARG_DEF(NULL, "tile-rows", 1, 540 "Number of tile rows to use, log2"); 541 static const arg_def_t auto_altref = ARG_DEF(NULL, "auto-alt-ref", 1, 542 "Enable automatic alt reference frames"); 543 static const arg_def_t arnr_maxframes = ARG_DEF(NULL, "arnr-maxframes", 1, 544 "AltRef Max Frames"); 545 static const arg_def_t arnr_strength = ARG_DEF(NULL, "arnr-strength", 1, 546 "AltRef Strength"); 547 static const arg_def_t arnr_type = ARG_DEF(NULL, "arnr-type", 1, 548 "AltRef Type"); 549 static const struct arg_enum_list tuning_enum[] = { 550 {"psnr", VP8_TUNE_PSNR}, 551 {"ssim", VP8_TUNE_SSIM}, 552 {NULL, 0} 553 }; 554 static const arg_def_t tune_ssim = ARG_DEF_ENUM(NULL, "tune", 1, 555 "Material to favor", tuning_enum); 556 static const arg_def_t cq_level = ARG_DEF(NULL, "cq-level", 1, 557 "Constant/Constrained Quality level"); 558 static const arg_def_t max_intra_rate_pct = ARG_DEF(NULL, "max-intra-rate", 1, 559 "Max I-frame bitrate (pct)"); 560 static const arg_def_t lossless = ARG_DEF(NULL, "lossless", 1, "Lossless mode"); 561 #if CONFIG_VP9_ENCODER 562 static const arg_def_t frame_parallel_decoding = ARG_DEF( 563 NULL, "frame-parallel", 1, "Enable frame parallel decodability features"); 564 #endif 565 566 #if CONFIG_VP8_ENCODER 567 static const arg_def_t *vp8_args[] = { 568 &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh, 569 &token_parts, &arnr_maxframes, &arnr_strength, &arnr_type, 570 &tune_ssim, &cq_level, &max_intra_rate_pct, 571 NULL 572 }; 573 static const int vp8_arg_ctrl_map[] = { 574 VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF, 575 VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD, 576 VP8E_SET_TOKEN_PARTITIONS, 577 VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH, VP8E_SET_ARNR_TYPE, 578 VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, VP8E_SET_MAX_INTRA_BITRATE_PCT, 579 0 580 }; 581 #endif 582 583 #if CONFIG_VP9_ENCODER 584 static const arg_def_t *vp9_args[] = { 585 &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh, 586 &tile_cols, &tile_rows, &arnr_maxframes, &arnr_strength, &arnr_type, 587 &tune_ssim, &cq_level, &max_intra_rate_pct, &lossless, 588 &frame_parallel_decoding, 589 NULL 590 }; 591 static const int vp9_arg_ctrl_map[] = { 592 VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF, 593 VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD, 594 VP9E_SET_TILE_COLUMNS, VP9E_SET_TILE_ROWS, 595 VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH, VP8E_SET_ARNR_TYPE, 596 VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, VP8E_SET_MAX_INTRA_BITRATE_PCT, 597 VP9E_SET_LOSSLESS, VP9E_SET_FRAME_PARALLEL_DECODING, 598 0 599 }; 600 #endif 601 602 static const arg_def_t *no_args[] = { NULL }; 603 604 void usage_exit() { 605 int i; 606 607 fprintf(stderr, "Usage: %s <options> -o dst_filename src_filename \n", 608 exec_name); 609 610 fprintf(stderr, "\nOptions:\n"); 611 arg_show_usage(stderr, main_args); 612 fprintf(stderr, "\nEncoder Global Options:\n"); 613 arg_show_usage(stderr, global_args); 614 fprintf(stderr, "\nRate Control Options:\n"); 615 arg_show_usage(stderr, rc_args); 616 fprintf(stderr, "\nTwopass Rate Control Options:\n"); 617 arg_show_usage(stderr, rc_twopass_args); 618 fprintf(stderr, "\nKeyframe Placement Options:\n"); 619 arg_show_usage(stderr, kf_args); 620 #if CONFIG_VP8_ENCODER 621 fprintf(stderr, "\nVP8 Specific Options:\n"); 622 arg_show_usage(stderr, vp8_args); 623 #endif 624 #if CONFIG_VP9_ENCODER 625 fprintf(stderr, "\nVP9 Specific Options:\n"); 626 arg_show_usage(stderr, vp9_args); 627 #endif 628 fprintf(stderr, "\nStream timebase (--timebase):\n" 629 " The desired precision of timestamps in the output, expressed\n" 630 " in fractional seconds. Default is 1/1000.\n"); 631 fprintf(stderr, "\n" 632 "Included encoders:\n" 633 "\n"); 634 635 for (i = 0; i < sizeof(codecs) / sizeof(codecs[0]); i++) 636 fprintf(stderr, " %-6s - %s\n", 637 codecs[i].name, 638 vpx_codec_iface_name(codecs[i].iface())); 639 640 exit(EXIT_FAILURE); 641 } 642 643 644 #define HIST_BAR_MAX 40 645 struct hist_bucket { 646 int low, high, count; 647 }; 648 649 650 static int merge_hist_buckets(struct hist_bucket *bucket, 651 int *buckets_, 652 int max_buckets) { 653 int small_bucket = 0, merge_bucket = INT_MAX, big_bucket = 0; 654 int buckets = *buckets_; 655 int i; 656 657 /* Find the extrema for this list of buckets */ 658 big_bucket = small_bucket = 0; 659 for (i = 0; i < buckets; i++) { 660 if (bucket[i].count < bucket[small_bucket].count) 661 small_bucket = i; 662 if (bucket[i].count > bucket[big_bucket].count) 663 big_bucket = i; 664 } 665 666 /* If we have too many buckets, merge the smallest with an adjacent 667 * bucket. 668 */ 669 while (buckets > max_buckets) { 670 int last_bucket = buckets - 1; 671 672 /* merge the small bucket with an adjacent one. */ 673 if (small_bucket == 0) 674 merge_bucket = 1; 675 else if (small_bucket == last_bucket) 676 merge_bucket = last_bucket - 1; 677 else if (bucket[small_bucket - 1].count < bucket[small_bucket + 1].count) 678 merge_bucket = small_bucket - 1; 679 else 680 merge_bucket = small_bucket + 1; 681 682 assert(abs(merge_bucket - small_bucket) <= 1); 683 assert(small_bucket < buckets); 684 assert(big_bucket < buckets); 685 assert(merge_bucket < buckets); 686 687 if (merge_bucket < small_bucket) { 688 bucket[merge_bucket].high = bucket[small_bucket].high; 689 bucket[merge_bucket].count += bucket[small_bucket].count; 690 } else { 691 bucket[small_bucket].high = bucket[merge_bucket].high; 692 bucket[small_bucket].count += bucket[merge_bucket].count; 693 merge_bucket = small_bucket; 694 } 695 696 assert(bucket[merge_bucket].low != bucket[merge_bucket].high); 697 698 buckets--; 699 700 /* Remove the merge_bucket from the list, and find the new small 701 * and big buckets while we're at it 702 */ 703 big_bucket = small_bucket = 0; 704 for (i = 0; i < buckets; i++) { 705 if (i > merge_bucket) 706 bucket[i] = bucket[i + 1]; 707 708 if (bucket[i].count < bucket[small_bucket].count) 709 small_bucket = i; 710 if (bucket[i].count > bucket[big_bucket].count) 711 big_bucket = i; 712 } 713 714 } 715 716 *buckets_ = buckets; 717 return bucket[big_bucket].count; 718 } 719 720 721 static void show_histogram(const struct hist_bucket *bucket, 722 int buckets, 723 int total, 724 int scale) { 725 const char *pat1, *pat2; 726 int i; 727 728 switch ((int)(log(bucket[buckets - 1].high) / log(10)) + 1) { 729 case 1: 730 case 2: 731 pat1 = "%4d %2s: "; 732 pat2 = "%4d-%2d: "; 733 break; 734 case 3: 735 pat1 = "%5d %3s: "; 736 pat2 = "%5d-%3d: "; 737 break; 738 case 4: 739 pat1 = "%6d %4s: "; 740 pat2 = "%6d-%4d: "; 741 break; 742 case 5: 743 pat1 = "%7d %5s: "; 744 pat2 = "%7d-%5d: "; 745 break; 746 case 6: 747 pat1 = "%8d %6s: "; 748 pat2 = "%8d-%6d: "; 749 break; 750 case 7: 751 pat1 = "%9d %7s: "; 752 pat2 = "%9d-%7d: "; 753 break; 754 default: 755 pat1 = "%12d %10s: "; 756 pat2 = "%12d-%10d: "; 757 break; 758 } 759 760 for (i = 0; i < buckets; i++) { 761 int len; 762 int j; 763 float pct; 764 765 pct = (float)(100.0 * bucket[i].count / total); 766 len = HIST_BAR_MAX * bucket[i].count / scale; 767 if (len < 1) 768 len = 1; 769 assert(len <= HIST_BAR_MAX); 770 771 if (bucket[i].low == bucket[i].high) 772 fprintf(stderr, pat1, bucket[i].low, ""); 773 else 774 fprintf(stderr, pat2, bucket[i].low, bucket[i].high); 775 776 for (j = 0; j < HIST_BAR_MAX; j++) 777 fprintf(stderr, j < len ? "=" : " "); 778 fprintf(stderr, "\t%5d (%6.2f%%)\n", bucket[i].count, pct); 779 } 780 } 781 782 783 static void show_q_histogram(const int counts[64], int max_buckets) { 784 struct hist_bucket bucket[64]; 785 int buckets = 0; 786 int total = 0; 787 int scale; 788 int i; 789 790 791 for (i = 0; i < 64; i++) { 792 if (counts[i]) { 793 bucket[buckets].low = bucket[buckets].high = i; 794 bucket[buckets].count = counts[i]; 795 buckets++; 796 total += counts[i]; 797 } 798 } 799 800 fprintf(stderr, "\nQuantizer Selection:\n"); 801 scale = merge_hist_buckets(bucket, &buckets, max_buckets); 802 show_histogram(bucket, buckets, total, scale); 803 } 804 805 806 #define RATE_BINS (100) 807 struct rate_hist { 808 int64_t *pts; 809 int *sz; 810 int samples; 811 int frames; 812 struct hist_bucket bucket[RATE_BINS]; 813 int total; 814 }; 815 816 817 static void init_rate_histogram(struct rate_hist *hist, 818 const vpx_codec_enc_cfg_t *cfg, 819 const vpx_rational_t *fps) { 820 int i; 821 822 /* Determine the number of samples in the buffer. Use the file's framerate 823 * to determine the number of frames in rc_buf_sz milliseconds, with an 824 * adjustment (5/4) to account for alt-refs 825 */ 826 hist->samples = cfg->rc_buf_sz * 5 / 4 * fps->num / fps->den / 1000; 827 828 /* prevent division by zero */ 829 if (hist->samples == 0) 830 hist->samples = 1; 831 832 hist->pts = calloc(hist->samples, sizeof(*hist->pts)); 833 hist->sz = calloc(hist->samples, sizeof(*hist->sz)); 834 for (i = 0; i < RATE_BINS; i++) { 835 hist->bucket[i].low = INT_MAX; 836 hist->bucket[i].high = 0; 837 hist->bucket[i].count = 0; 838 } 839 } 840 841 842 static void destroy_rate_histogram(struct rate_hist *hist) { 843 free(hist->pts); 844 free(hist->sz); 845 } 846 847 848 static void update_rate_histogram(struct rate_hist *hist, 849 const vpx_codec_enc_cfg_t *cfg, 850 const vpx_codec_cx_pkt_t *pkt) { 851 int i, idx; 852 int64_t now, then, sum_sz = 0, avg_bitrate; 853 854 now = pkt->data.frame.pts * 1000 855 * (uint64_t)cfg->g_timebase.num / (uint64_t)cfg->g_timebase.den; 856 857 idx = hist->frames++ % hist->samples; 858 hist->pts[idx] = now; 859 hist->sz[idx] = (int)pkt->data.frame.sz; 860 861 if (now < cfg->rc_buf_initial_sz) 862 return; 863 864 then = now; 865 866 /* Sum the size over the past rc_buf_sz ms */ 867 for (i = hist->frames; i > 0 && hist->frames - i < hist->samples; i--) { 868 int i_idx = (i - 1) % hist->samples; 869 870 then = hist->pts[i_idx]; 871 if (now - then > cfg->rc_buf_sz) 872 break; 873 sum_sz += hist->sz[i_idx]; 874 } 875 876 if (now == then) 877 return; 878 879 avg_bitrate = sum_sz * 8 * 1000 / (now - then); 880 idx = (int)(avg_bitrate * (RATE_BINS / 2) / (cfg->rc_target_bitrate * 1000)); 881 if (idx < 0) 882 idx = 0; 883 if (idx > RATE_BINS - 1) 884 idx = RATE_BINS - 1; 885 if (hist->bucket[idx].low > avg_bitrate) 886 hist->bucket[idx].low = (int)avg_bitrate; 887 if (hist->bucket[idx].high < avg_bitrate) 888 hist->bucket[idx].high = (int)avg_bitrate; 889 hist->bucket[idx].count++; 890 hist->total++; 891 } 892 893 894 static void show_rate_histogram(struct rate_hist *hist, 895 const vpx_codec_enc_cfg_t *cfg, 896 int max_buckets) { 897 int i, scale; 898 int buckets = 0; 899 900 for (i = 0; i < RATE_BINS; i++) { 901 if (hist->bucket[i].low == INT_MAX) 902 continue; 903 hist->bucket[buckets++] = hist->bucket[i]; 904 } 905 906 fprintf(stderr, "\nRate (over %dms window):\n", cfg->rc_buf_sz); 907 scale = merge_hist_buckets(hist->bucket, &buckets, max_buckets); 908 show_histogram(hist->bucket, buckets, hist->total, scale); 909 } 910 911 #define mmin(a, b) ((a) < (b) ? (a) : (b)) 912 static void find_mismatch(vpx_image_t *img1, vpx_image_t *img2, 913 int yloc[4], int uloc[4], int vloc[4]) { 914 const unsigned int bsize = 64; 915 const unsigned int bsizey = bsize >> img1->y_chroma_shift; 916 const unsigned int bsizex = bsize >> img1->x_chroma_shift; 917 const int c_w = (img1->d_w + img1->x_chroma_shift) >> img1->x_chroma_shift; 918 const int c_h = (img1->d_h + img1->y_chroma_shift) >> img1->y_chroma_shift; 919 unsigned int match = 1; 920 unsigned int i, j; 921 yloc[0] = yloc[1] = yloc[2] = yloc[3] = -1; 922 for (i = 0, match = 1; match && i < img1->d_h; i += bsize) { 923 for (j = 0; match && j < img1->d_w; j += bsize) { 924 int k, l; 925 int si = mmin(i + bsize, img1->d_h) - i; 926 int sj = mmin(j + bsize, img1->d_w) - j; 927 for (k = 0; match && k < si; k++) 928 for (l = 0; match && l < sj; l++) { 929 if (*(img1->planes[VPX_PLANE_Y] + 930 (i + k) * img1->stride[VPX_PLANE_Y] + j + l) != 931 *(img2->planes[VPX_PLANE_Y] + 932 (i + k) * img2->stride[VPX_PLANE_Y] + j + l)) { 933 yloc[0] = i + k; 934 yloc[1] = j + l; 935 yloc[2] = *(img1->planes[VPX_PLANE_Y] + 936 (i + k) * img1->stride[VPX_PLANE_Y] + j + l); 937 yloc[3] = *(img2->planes[VPX_PLANE_Y] + 938 (i + k) * img2->stride[VPX_PLANE_Y] + j + l); 939 match = 0; 940 break; 941 } 942 } 943 } 944 } 945 946 uloc[0] = uloc[1] = uloc[2] = uloc[3] = -1; 947 for (i = 0, match = 1; match && i < c_h; i += bsizey) { 948 for (j = 0; match && j < c_w; j += bsizex) { 949 int k, l; 950 int si = mmin(i + bsizey, c_h - i); 951 int sj = mmin(j + bsizex, c_w - j); 952 for (k = 0; match && k < si; k++) 953 for (l = 0; match && l < sj; l++) { 954 if (*(img1->planes[VPX_PLANE_U] + 955 (i + k) * img1->stride[VPX_PLANE_U] + j + l) != 956 *(img2->planes[VPX_PLANE_U] + 957 (i + k) * img2->stride[VPX_PLANE_U] + j + l)) { 958 uloc[0] = i + k; 959 uloc[1] = j + l; 960 uloc[2] = *(img1->planes[VPX_PLANE_U] + 961 (i + k) * img1->stride[VPX_PLANE_U] + j + l); 962 uloc[3] = *(img2->planes[VPX_PLANE_U] + 963 (i + k) * img2->stride[VPX_PLANE_V] + j + l); 964 match = 0; 965 break; 966 } 967 } 968 } 969 } 970 vloc[0] = vloc[1] = vloc[2] = vloc[3] = -1; 971 for (i = 0, match = 1; match && i < c_h; i += bsizey) { 972 for (j = 0; match && j < c_w; j += bsizex) { 973 int k, l; 974 int si = mmin(i + bsizey, c_h - i); 975 int sj = mmin(j + bsizex, c_w - j); 976 for (k = 0; match && k < si; k++) 977 for (l = 0; match && l < sj; l++) { 978 if (*(img1->planes[VPX_PLANE_V] + 979 (i + k) * img1->stride[VPX_PLANE_V] + j + l) != 980 *(img2->planes[VPX_PLANE_V] + 981 (i + k) * img2->stride[VPX_PLANE_V] + j + l)) { 982 vloc[0] = i + k; 983 vloc[1] = j + l; 984 vloc[2] = *(img1->planes[VPX_PLANE_V] + 985 (i + k) * img1->stride[VPX_PLANE_V] + j + l); 986 vloc[3] = *(img2->planes[VPX_PLANE_V] + 987 (i + k) * img2->stride[VPX_PLANE_V] + j + l); 988 match = 0; 989 break; 990 } 991 } 992 } 993 } 994 } 995 996 static int compare_img(vpx_image_t *img1, vpx_image_t *img2) 997 { 998 const int c_w = (img1->d_w + img1->x_chroma_shift) >> img1->x_chroma_shift; 999 const int c_h = (img1->d_h + img1->y_chroma_shift) >> img1->y_chroma_shift; 1000 int match = 1; 1001 unsigned int i; 1002 1003 match &= (img1->fmt == img2->fmt); 1004 match &= (img1->w == img2->w); 1005 match &= (img1->h == img2->h); 1006 1007 for (i = 0; i < img1->d_h; i++) 1008 match &= (memcmp(img1->planes[VPX_PLANE_Y]+i*img1->stride[VPX_PLANE_Y], 1009 img2->planes[VPX_PLANE_Y]+i*img2->stride[VPX_PLANE_Y], 1010 img1->d_w) == 0); 1011 1012 for (i = 0; i < c_h; i++) 1013 match &= (memcmp(img1->planes[VPX_PLANE_U]+i*img1->stride[VPX_PLANE_U], 1014 img2->planes[VPX_PLANE_U]+i*img2->stride[VPX_PLANE_U], 1015 c_w) == 0); 1016 1017 for (i = 0; i < c_h; i++) 1018 match &= (memcmp(img1->planes[VPX_PLANE_V]+i*img1->stride[VPX_PLANE_U], 1019 img2->planes[VPX_PLANE_V]+i*img2->stride[VPX_PLANE_U], 1020 c_w) == 0); 1021 1022 return match; 1023 } 1024 1025 1026 #define NELEMENTS(x) (sizeof(x)/sizeof(x[0])) 1027 #define MAX(x,y) ((x)>(y)?(x):(y)) 1028 #if CONFIG_VP8_ENCODER && !CONFIG_VP9_ENCODER 1029 #define ARG_CTRL_CNT_MAX NELEMENTS(vp8_arg_ctrl_map) 1030 #elif !CONFIG_VP8_ENCODER && CONFIG_VP9_ENCODER 1031 #define ARG_CTRL_CNT_MAX NELEMENTS(vp9_arg_ctrl_map) 1032 #else 1033 #define ARG_CTRL_CNT_MAX MAX(NELEMENTS(vp8_arg_ctrl_map), \ 1034 NELEMENTS(vp9_arg_ctrl_map)) 1035 #endif 1036 1037 /* Configuration elements common to all streams */ 1038 struct global_config { 1039 const struct codec_item *codec; 1040 int passes; 1041 int pass; 1042 int usage; 1043 int deadline; 1044 int use_i420; 1045 int quiet; 1046 int verbose; 1047 int limit; 1048 int skip_frames; 1049 int show_psnr; 1050 enum TestDecodeFatality test_decode; 1051 int have_framerate; 1052 struct vpx_rational framerate; 1053 int out_part; 1054 int debug; 1055 int show_q_hist_buckets; 1056 int show_rate_hist_buckets; 1057 }; 1058 1059 1060 /* Per-stream configuration */ 1061 struct stream_config { 1062 struct vpx_codec_enc_cfg cfg; 1063 const char *out_fn; 1064 const char *stats_fn; 1065 stereo_format_t stereo_fmt; 1066 int arg_ctrls[ARG_CTRL_CNT_MAX][2]; 1067 int arg_ctrl_cnt; 1068 int write_webm; 1069 int have_kf_max_dist; 1070 }; 1071 1072 1073 struct stream_state { 1074 int index; 1075 struct stream_state *next; 1076 struct stream_config config; 1077 FILE *file; 1078 struct rate_hist rate_hist; 1079 struct EbmlGlobal ebml; 1080 uint32_t hash; 1081 uint64_t psnr_sse_total; 1082 uint64_t psnr_samples_total; 1083 double psnr_totals[4]; 1084 int psnr_count; 1085 int counts[64]; 1086 vpx_codec_ctx_t encoder; 1087 unsigned int frames_out; 1088 uint64_t cx_time; 1089 size_t nbytes; 1090 stats_io_t stats; 1091 struct vpx_image *img; 1092 vpx_codec_ctx_t decoder; 1093 int mismatch_seen; 1094 }; 1095 1096 1097 void validate_positive_rational(const char *msg, 1098 struct vpx_rational *rat) { 1099 if (rat->den < 0) { 1100 rat->num *= -1; 1101 rat->den *= -1; 1102 } 1103 1104 if (rat->num < 0) 1105 die("Error: %s must be positive\n", msg); 1106 1107 if (!rat->den) 1108 die("Error: %s has zero denominator\n", msg); 1109 } 1110 1111 1112 static void parse_global_config(struct global_config *global, char **argv) { 1113 char **argi, **argj; 1114 struct arg arg; 1115 1116 /* Initialize default parameters */ 1117 memset(global, 0, sizeof(*global)); 1118 global->codec = codecs; 1119 global->passes = 0; 1120 global->use_i420 = 1; 1121 /* Assign default deadline to good quality */ 1122 global->deadline = VPX_DL_GOOD_QUALITY; 1123 1124 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) { 1125 arg.argv_step = 1; 1126 1127 if (arg_match(&arg, &codecarg, argi)) { 1128 int j, k = -1; 1129 1130 for (j = 0; j < sizeof(codecs) / sizeof(codecs[0]); j++) 1131 if (!strcmp(codecs[j].name, arg.val)) 1132 k = j; 1133 1134 if (k >= 0) 1135 global->codec = codecs + k; 1136 else 1137 die("Error: Unrecognized argument (%s) to --codec\n", 1138 arg.val); 1139 1140 } else if (arg_match(&arg, &passes, argi)) { 1141 global->passes = arg_parse_uint(&arg); 1142 1143 if (global->passes < 1 || global->passes > 2) 1144 die("Error: Invalid number of passes (%d)\n", global->passes); 1145 } else if (arg_match(&arg, &pass_arg, argi)) { 1146 global->pass = arg_parse_uint(&arg); 1147 1148 if (global->pass < 1 || global->pass > 2) 1149 die("Error: Invalid pass selected (%d)\n", 1150 global->pass); 1151 } else if (arg_match(&arg, &usage, argi)) 1152 global->usage = arg_parse_uint(&arg); 1153 else if (arg_match(&arg, &deadline, argi)) 1154 global->deadline = arg_parse_uint(&arg); 1155 else if (arg_match(&arg, &best_dl, argi)) 1156 global->deadline = VPX_DL_BEST_QUALITY; 1157 else if (arg_match(&arg, &good_dl, argi)) 1158 global->deadline = VPX_DL_GOOD_QUALITY; 1159 else if (arg_match(&arg, &rt_dl, argi)) 1160 global->deadline = VPX_DL_REALTIME; 1161 else if (arg_match(&arg, &use_yv12, argi)) 1162 global->use_i420 = 0; 1163 else if (arg_match(&arg, &use_i420, argi)) 1164 global->use_i420 = 1; 1165 else if (arg_match(&arg, &quietarg, argi)) 1166 global->quiet = 1; 1167 else if (arg_match(&arg, &verbosearg, argi)) 1168 global->verbose = 1; 1169 else if (arg_match(&arg, &limit, argi)) 1170 global->limit = arg_parse_uint(&arg); 1171 else if (arg_match(&arg, &skip, argi)) 1172 global->skip_frames = arg_parse_uint(&arg); 1173 else if (arg_match(&arg, &psnrarg, argi)) 1174 global->show_psnr = 1; 1175 else if (arg_match(&arg, &recontest, argi)) 1176 global->test_decode = arg_parse_enum_or_int(&arg); 1177 else if (arg_match(&arg, &framerate, argi)) { 1178 global->framerate = arg_parse_rational(&arg); 1179 validate_positive_rational(arg.name, &global->framerate); 1180 global->have_framerate = 1; 1181 } else if (arg_match(&arg, &out_part, argi)) 1182 global->out_part = 1; 1183 else if (arg_match(&arg, &debugmode, argi)) 1184 global->debug = 1; 1185 else if (arg_match(&arg, &q_hist_n, argi)) 1186 global->show_q_hist_buckets = arg_parse_uint(&arg); 1187 else if (arg_match(&arg, &rate_hist_n, argi)) 1188 global->show_rate_hist_buckets = arg_parse_uint(&arg); 1189 else 1190 argj++; 1191 } 1192 1193 /* Validate global config */ 1194 if (global->passes == 0) { 1195 #if CONFIG_VP9_ENCODER 1196 // Make default VP9 passes = 2 until there is a better quality 1-pass 1197 // encoder 1198 global->passes = (global->codec->iface == vpx_codec_vp9_cx ? 2 : 1); 1199 #else 1200 global->passes = 1; 1201 #endif 1202 } 1203 1204 if (global->pass) { 1205 /* DWIM: Assume the user meant passes=2 if pass=2 is specified */ 1206 if (global->pass > global->passes) { 1207 warn("Assuming --pass=%d implies --passes=%d\n", 1208 global->pass, global->pass); 1209 global->passes = global->pass; 1210 } 1211 } 1212 } 1213 1214 1215 void open_input_file(struct input_state *input) { 1216 unsigned int fourcc; 1217 1218 /* Parse certain options from the input file, if possible */ 1219 input->file = strcmp(input->fn, "-") ? fopen(input->fn, "rb") 1220 : set_binary_mode(stdin); 1221 1222 if (!input->file) 1223 fatal("Failed to open input file"); 1224 1225 if (!fseeko(input->file, 0, SEEK_END)) { 1226 /* Input file is seekable. Figure out how long it is, so we can get 1227 * progress info. 1228 */ 1229 input->length = ftello(input->file); 1230 rewind(input->file); 1231 } 1232 1233 /* For RAW input sources, these bytes will applied on the first frame 1234 * in read_frame(). 1235 */ 1236 input->detect.buf_read = fread(input->detect.buf, 1, 4, input->file); 1237 input->detect.position = 0; 1238 1239 if (input->detect.buf_read == 4 1240 && file_is_y4m(input->file, &input->y4m, input->detect.buf)) { 1241 if (y4m_input_open(&input->y4m, input->file, input->detect.buf, 4, 1242 input->only_i420) >= 0) { 1243 input->file_type = FILE_TYPE_Y4M; 1244 input->w = input->y4m.pic_w; 1245 input->h = input->y4m.pic_h; 1246 input->framerate.num = input->y4m.fps_n; 1247 input->framerate.den = input->y4m.fps_d; 1248 input->use_i420 = 0; 1249 } else 1250 fatal("Unsupported Y4M stream."); 1251 } else if (input->detect.buf_read == 4 && file_is_ivf(input, &fourcc)) { 1252 fatal("IVF is not supported as input."); 1253 } else { 1254 input->file_type = FILE_TYPE_RAW; 1255 } 1256 } 1257 1258 1259 static void close_input_file(struct input_state *input) { 1260 fclose(input->file); 1261 if (input->file_type == FILE_TYPE_Y4M) 1262 y4m_input_close(&input->y4m); 1263 } 1264 1265 static struct stream_state *new_stream(struct global_config *global, 1266 struct stream_state *prev) { 1267 struct stream_state *stream; 1268 1269 stream = calloc(1, sizeof(*stream)); 1270 if (!stream) 1271 fatal("Failed to allocate new stream."); 1272 if (prev) { 1273 memcpy(stream, prev, sizeof(*stream)); 1274 stream->index++; 1275 prev->next = stream; 1276 } else { 1277 vpx_codec_err_t res; 1278 1279 /* Populate encoder configuration */ 1280 res = vpx_codec_enc_config_default(global->codec->iface(), 1281 &stream->config.cfg, 1282 global->usage); 1283 if (res) 1284 fatal("Failed to get config: %s\n", vpx_codec_err_to_string(res)); 1285 1286 /* Change the default timebase to a high enough value so that the 1287 * encoder will always create strictly increasing timestamps. 1288 */ 1289 stream->config.cfg.g_timebase.den = 1000; 1290 1291 /* Never use the library's default resolution, require it be parsed 1292 * from the file or set on the command line. 1293 */ 1294 stream->config.cfg.g_w = 0; 1295 stream->config.cfg.g_h = 0; 1296 1297 /* Initialize remaining stream parameters */ 1298 stream->config.stereo_fmt = STEREO_FORMAT_MONO; 1299 stream->config.write_webm = 1; 1300 stream->ebml.last_pts_ms = -1; 1301 1302 /* Allows removal of the application version from the EBML tags */ 1303 stream->ebml.debug = global->debug; 1304 } 1305 1306 /* Output files must be specified for each stream */ 1307 stream->config.out_fn = NULL; 1308 1309 stream->next = NULL; 1310 return stream; 1311 } 1312 1313 1314 static int parse_stream_params(struct global_config *global, 1315 struct stream_state *stream, 1316 char **argv) { 1317 char **argi, **argj; 1318 struct arg arg; 1319 static const arg_def_t **ctrl_args = no_args; 1320 static const int *ctrl_args_map = NULL; 1321 struct stream_config *config = &stream->config; 1322 int eos_mark_found = 0; 1323 1324 /* Handle codec specific options */ 1325 if (0) { 1326 #if CONFIG_VP8_ENCODER 1327 } else if (global->codec->iface == vpx_codec_vp8_cx) { 1328 ctrl_args = vp8_args; 1329 ctrl_args_map = vp8_arg_ctrl_map; 1330 #endif 1331 #if CONFIG_VP9_ENCODER 1332 } else if (global->codec->iface == vpx_codec_vp9_cx) { 1333 ctrl_args = vp9_args; 1334 ctrl_args_map = vp9_arg_ctrl_map; 1335 #endif 1336 } 1337 1338 for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) { 1339 arg.argv_step = 1; 1340 1341 /* Once we've found an end-of-stream marker (--) we want to continue 1342 * shifting arguments but not consuming them. 1343 */ 1344 if (eos_mark_found) { 1345 argj++; 1346 continue; 1347 } else if (!strcmp(*argj, "--")) { 1348 eos_mark_found = 1; 1349 continue; 1350 } 1351 1352 if (0); 1353 else if (arg_match(&arg, &outputfile, argi)) 1354 config->out_fn = arg.val; 1355 else if (arg_match(&arg, &fpf_name, argi)) 1356 config->stats_fn = arg.val; 1357 else if (arg_match(&arg, &use_ivf, argi)) 1358 config->write_webm = 0; 1359 else if (arg_match(&arg, &threads, argi)) 1360 config->cfg.g_threads = arg_parse_uint(&arg); 1361 else if (arg_match(&arg, &profile, argi)) 1362 config->cfg.g_profile = arg_parse_uint(&arg); 1363 else if (arg_match(&arg, &width, argi)) 1364 config->cfg.g_w = arg_parse_uint(&arg); 1365 else if (arg_match(&arg, &height, argi)) 1366 config->cfg.g_h = arg_parse_uint(&arg); 1367 else if (arg_match(&arg, &stereo_mode, argi)) 1368 config->stereo_fmt = arg_parse_enum_or_int(&arg); 1369 else if (arg_match(&arg, &timebase, argi)) { 1370 config->cfg.g_timebase = arg_parse_rational(&arg); 1371 validate_positive_rational(arg.name, &config->cfg.g_timebase); 1372 } else if (arg_match(&arg, &error_resilient, argi)) 1373 config->cfg.g_error_resilient = arg_parse_uint(&arg); 1374 else if (arg_match(&arg, &lag_in_frames, argi)) 1375 config->cfg.g_lag_in_frames = arg_parse_uint(&arg); 1376 else if (arg_match(&arg, &dropframe_thresh, argi)) 1377 config->cfg.rc_dropframe_thresh = arg_parse_uint(&arg); 1378 else if (arg_match(&arg, &resize_allowed, argi)) 1379 config->cfg.rc_resize_allowed = arg_parse_uint(&arg); 1380 else if (arg_match(&arg, &resize_up_thresh, argi)) 1381 config->cfg.rc_resize_up_thresh = arg_parse_uint(&arg); 1382 else if (arg_match(&arg, &resize_down_thresh, argi)) 1383 config->cfg.rc_resize_down_thresh = arg_parse_uint(&arg); 1384 else if (arg_match(&arg, &end_usage, argi)) 1385 config->cfg.rc_end_usage = arg_parse_enum_or_int(&arg); 1386 else if (arg_match(&arg, &target_bitrate, argi)) 1387 config->cfg.rc_target_bitrate = arg_parse_uint(&arg); 1388 else if (arg_match(&arg, &min_quantizer, argi)) 1389 config->cfg.rc_min_quantizer = arg_parse_uint(&arg); 1390 else if (arg_match(&arg, &max_quantizer, argi)) 1391 config->cfg.rc_max_quantizer = arg_parse_uint(&arg); 1392 else if (arg_match(&arg, &undershoot_pct, argi)) 1393 config->cfg.rc_undershoot_pct = arg_parse_uint(&arg); 1394 else if (arg_match(&arg, &overshoot_pct, argi)) 1395 config->cfg.rc_overshoot_pct = arg_parse_uint(&arg); 1396 else if (arg_match(&arg, &buf_sz, argi)) 1397 config->cfg.rc_buf_sz = arg_parse_uint(&arg); 1398 else if (arg_match(&arg, &buf_initial_sz, argi)) 1399 config->cfg.rc_buf_initial_sz = arg_parse_uint(&arg); 1400 else if (arg_match(&arg, &buf_optimal_sz, argi)) 1401 config->cfg.rc_buf_optimal_sz = arg_parse_uint(&arg); 1402 else if (arg_match(&arg, &bias_pct, argi)) { 1403 config->cfg.rc_2pass_vbr_bias_pct = arg_parse_uint(&arg); 1404 1405 if (global->passes < 2) 1406 warn("option %s ignored in one-pass mode.\n", arg.name); 1407 } else if (arg_match(&arg, &minsection_pct, argi)) { 1408 config->cfg.rc_2pass_vbr_minsection_pct = arg_parse_uint(&arg); 1409 1410 if (global->passes < 2) 1411 warn("option %s ignored in one-pass mode.\n", arg.name); 1412 } else if (arg_match(&arg, &maxsection_pct, argi)) { 1413 config->cfg.rc_2pass_vbr_maxsection_pct = arg_parse_uint(&arg); 1414 1415 if (global->passes < 2) 1416 warn("option %s ignored in one-pass mode.\n", arg.name); 1417 } else if (arg_match(&arg, &kf_min_dist, argi)) 1418 config->cfg.kf_min_dist = arg_parse_uint(&arg); 1419 else if (arg_match(&arg, &kf_max_dist, argi)) { 1420 config->cfg.kf_max_dist = arg_parse_uint(&arg); 1421 config->have_kf_max_dist = 1; 1422 } else if (arg_match(&arg, &kf_disabled, argi)) 1423 config->cfg.kf_mode = VPX_KF_DISABLED; 1424 else { 1425 int i, match = 0; 1426 1427 for (i = 0; ctrl_args[i]; i++) { 1428 if (arg_match(&arg, ctrl_args[i], argi)) { 1429 int j; 1430 match = 1; 1431 1432 /* Point either to the next free element or the first 1433 * instance of this control. 1434 */ 1435 for (j = 0; j < config->arg_ctrl_cnt; j++) 1436 if (config->arg_ctrls[j][0] == ctrl_args_map[i]) 1437 break; 1438 1439 /* Update/insert */ 1440 assert(j < ARG_CTRL_CNT_MAX); 1441 if (j < ARG_CTRL_CNT_MAX) { 1442 config->arg_ctrls[j][0] = ctrl_args_map[i]; 1443 config->arg_ctrls[j][1] = arg_parse_enum_or_int(&arg); 1444 if (j == config->arg_ctrl_cnt) 1445 config->arg_ctrl_cnt++; 1446 } 1447 1448 } 1449 } 1450 1451 if (!match) 1452 argj++; 1453 } 1454 } 1455 1456 return eos_mark_found; 1457 } 1458 1459 1460 #define FOREACH_STREAM(func)\ 1461 do\ 1462 {\ 1463 struct stream_state *stream;\ 1464 \ 1465 for(stream = streams; stream; stream = stream->next)\ 1466 func;\ 1467 }while(0) 1468 1469 1470 static void validate_stream_config(struct stream_state *stream) { 1471 struct stream_state *streami; 1472 1473 if (!stream->config.cfg.g_w || !stream->config.cfg.g_h) 1474 fatal("Stream %d: Specify stream dimensions with --width (-w) " 1475 " and --height (-h)", stream->index); 1476 1477 for (streami = stream; streami; streami = streami->next) { 1478 /* All streams require output files */ 1479 if (!streami->config.out_fn) 1480 fatal("Stream %d: Output file is required (specify with -o)", 1481 streami->index); 1482 1483 /* Check for two streams outputting to the same file */ 1484 if (streami != stream) { 1485 const char *a = stream->config.out_fn; 1486 const char *b = streami->config.out_fn; 1487 if (!strcmp(a, b) && strcmp(a, "/dev/null") && strcmp(a, ":nul")) 1488 fatal("Stream %d: duplicate output file (from stream %d)", 1489 streami->index, stream->index); 1490 } 1491 1492 /* Check for two streams sharing a stats file. */ 1493 if (streami != stream) { 1494 const char *a = stream->config.stats_fn; 1495 const char *b = streami->config.stats_fn; 1496 if (a && b && !strcmp(a, b)) 1497 fatal("Stream %d: duplicate stats file (from stream %d)", 1498 streami->index, stream->index); 1499 } 1500 } 1501 } 1502 1503 1504 static void set_stream_dimensions(struct stream_state *stream, 1505 unsigned int w, 1506 unsigned int h) { 1507 if (!stream->config.cfg.g_w) { 1508 if (!stream->config.cfg.g_h) 1509 stream->config.cfg.g_w = w; 1510 else 1511 stream->config.cfg.g_w = w * stream->config.cfg.g_h / h; 1512 } 1513 if (!stream->config.cfg.g_h) { 1514 stream->config.cfg.g_h = h * stream->config.cfg.g_w / w; 1515 } 1516 } 1517 1518 1519 static void set_default_kf_interval(struct stream_state *stream, 1520 struct global_config *global) { 1521 /* Use a max keyframe interval of 5 seconds, if none was 1522 * specified on the command line. 1523 */ 1524 if (!stream->config.have_kf_max_dist) { 1525 double framerate = (double)global->framerate.num / global->framerate.den; 1526 if (framerate > 0.0) 1527 stream->config.cfg.kf_max_dist = (unsigned int)(5.0 * framerate); 1528 } 1529 } 1530 1531 1532 static void show_stream_config(struct stream_state *stream, 1533 struct global_config *global, 1534 struct input_state *input) { 1535 1536 #define SHOW(field) \ 1537 fprintf(stderr, " %-28s = %d\n", #field, stream->config.cfg.field) 1538 1539 if (stream->index == 0) { 1540 fprintf(stderr, "Codec: %s\n", 1541 vpx_codec_iface_name(global->codec->iface())); 1542 fprintf(stderr, "Source file: %s Format: %s\n", input->fn, 1543 input->use_i420 ? "I420" : "YV12"); 1544 } 1545 if (stream->next || stream->index) 1546 fprintf(stderr, "\nStream Index: %d\n", stream->index); 1547 fprintf(stderr, "Destination file: %s\n", stream->config.out_fn); 1548 fprintf(stderr, "Encoder parameters:\n"); 1549 1550 SHOW(g_usage); 1551 SHOW(g_threads); 1552 SHOW(g_profile); 1553 SHOW(g_w); 1554 SHOW(g_h); 1555 SHOW(g_timebase.num); 1556 SHOW(g_timebase.den); 1557 SHOW(g_error_resilient); 1558 SHOW(g_pass); 1559 SHOW(g_lag_in_frames); 1560 SHOW(rc_dropframe_thresh); 1561 SHOW(rc_resize_allowed); 1562 SHOW(rc_resize_up_thresh); 1563 SHOW(rc_resize_down_thresh); 1564 SHOW(rc_end_usage); 1565 SHOW(rc_target_bitrate); 1566 SHOW(rc_min_quantizer); 1567 SHOW(rc_max_quantizer); 1568 SHOW(rc_undershoot_pct); 1569 SHOW(rc_overshoot_pct); 1570 SHOW(rc_buf_sz); 1571 SHOW(rc_buf_initial_sz); 1572 SHOW(rc_buf_optimal_sz); 1573 SHOW(rc_2pass_vbr_bias_pct); 1574 SHOW(rc_2pass_vbr_minsection_pct); 1575 SHOW(rc_2pass_vbr_maxsection_pct); 1576 SHOW(kf_mode); 1577 SHOW(kf_min_dist); 1578 SHOW(kf_max_dist); 1579 } 1580 1581 1582 static void open_output_file(struct stream_state *stream, 1583 struct global_config *global) { 1584 const char *fn = stream->config.out_fn; 1585 1586 stream->file = strcmp(fn, "-") ? fopen(fn, "wb") : set_binary_mode(stdout); 1587 1588 if (!stream->file) 1589 fatal("Failed to open output file"); 1590 1591 if (stream->config.write_webm && fseek(stream->file, 0, SEEK_CUR)) 1592 fatal("WebM output to pipes not supported."); 1593 1594 if (stream->config.write_webm) { 1595 stream->ebml.stream = stream->file; 1596 write_webm_file_header(&stream->ebml, &stream->config.cfg, 1597 &global->framerate, 1598 stream->config.stereo_fmt, 1599 global->codec->fourcc); 1600 } else 1601 write_ivf_file_header(stream->file, &stream->config.cfg, 1602 global->codec->fourcc, 0); 1603 } 1604 1605 1606 static void close_output_file(struct stream_state *stream, 1607 unsigned int fourcc) { 1608 if (stream->config.write_webm) { 1609 write_webm_file_footer(&stream->ebml, stream->hash); 1610 free(stream->ebml.cue_list); 1611 stream->ebml.cue_list = NULL; 1612 } else { 1613 if (!fseek(stream->file, 0, SEEK_SET)) 1614 write_ivf_file_header(stream->file, &stream->config.cfg, 1615 fourcc, 1616 stream->frames_out); 1617 } 1618 1619 fclose(stream->file); 1620 } 1621 1622 1623 static void setup_pass(struct stream_state *stream, 1624 struct global_config *global, 1625 int pass) { 1626 if (stream->config.stats_fn) { 1627 if (!stats_open_file(&stream->stats, stream->config.stats_fn, 1628 pass)) 1629 fatal("Failed to open statistics store"); 1630 } else { 1631 if (!stats_open_mem(&stream->stats, pass)) 1632 fatal("Failed to open statistics store"); 1633 } 1634 1635 stream->config.cfg.g_pass = global->passes == 2 1636 ? pass ? VPX_RC_LAST_PASS : VPX_RC_FIRST_PASS 1637 : VPX_RC_ONE_PASS; 1638 if (pass) 1639 stream->config.cfg.rc_twopass_stats_in = stats_get(&stream->stats); 1640 1641 stream->cx_time = 0; 1642 stream->nbytes = 0; 1643 stream->frames_out = 0; 1644 } 1645 1646 1647 static void initialize_encoder(struct stream_state *stream, 1648 struct global_config *global) { 1649 int i; 1650 int flags = 0; 1651 1652 flags |= global->show_psnr ? VPX_CODEC_USE_PSNR : 0; 1653 flags |= global->out_part ? VPX_CODEC_USE_OUTPUT_PARTITION : 0; 1654 1655 /* Construct Encoder Context */ 1656 vpx_codec_enc_init(&stream->encoder, global->codec->iface(), 1657 &stream->config.cfg, flags); 1658 ctx_exit_on_error(&stream->encoder, "Failed to initialize encoder"); 1659 1660 /* Note that we bypass the vpx_codec_control wrapper macro because 1661 * we're being clever to store the control IDs in an array. Real 1662 * applications will want to make use of the enumerations directly 1663 */ 1664 for (i = 0; i < stream->config.arg_ctrl_cnt; i++) { 1665 int ctrl = stream->config.arg_ctrls[i][0]; 1666 int value = stream->config.arg_ctrls[i][1]; 1667 if (vpx_codec_control_(&stream->encoder, ctrl, value)) 1668 fprintf(stderr, "Error: Tried to set control %d = %d\n", 1669 ctrl, value); 1670 1671 ctx_exit_on_error(&stream->encoder, "Failed to control codec"); 1672 } 1673 1674 #if CONFIG_DECODERS 1675 if (global->test_decode != TEST_DECODE_OFF) { 1676 vpx_codec_dec_init(&stream->decoder, global->codec->dx_iface(), NULL, 0); 1677 } 1678 #endif 1679 } 1680 1681 1682 static void encode_frame(struct stream_state *stream, 1683 struct global_config *global, 1684 struct vpx_image *img, 1685 unsigned int frames_in) { 1686 vpx_codec_pts_t frame_start, next_frame_start; 1687 struct vpx_codec_enc_cfg *cfg = &stream->config.cfg; 1688 struct vpx_usec_timer timer; 1689 1690 frame_start = (cfg->g_timebase.den * (int64_t)(frames_in - 1) 1691 * global->framerate.den) 1692 / cfg->g_timebase.num / global->framerate.num; 1693 next_frame_start = (cfg->g_timebase.den * (int64_t)(frames_in) 1694 * global->framerate.den) 1695 / cfg->g_timebase.num / global->framerate.num; 1696 1697 /* Scale if necessary */ 1698 if (img && (img->d_w != cfg->g_w || img->d_h != cfg->g_h)) { 1699 if (!stream->img) 1700 stream->img = vpx_img_alloc(NULL, VPX_IMG_FMT_I420, 1701 cfg->g_w, cfg->g_h, 16); 1702 I420Scale(img->planes[VPX_PLANE_Y], img->stride[VPX_PLANE_Y], 1703 img->planes[VPX_PLANE_U], img->stride[VPX_PLANE_U], 1704 img->planes[VPX_PLANE_V], img->stride[VPX_PLANE_V], 1705 img->d_w, img->d_h, 1706 stream->img->planes[VPX_PLANE_Y], 1707 stream->img->stride[VPX_PLANE_Y], 1708 stream->img->planes[VPX_PLANE_U], 1709 stream->img->stride[VPX_PLANE_U], 1710 stream->img->planes[VPX_PLANE_V], 1711 stream->img->stride[VPX_PLANE_V], 1712 stream->img->d_w, stream->img->d_h, 1713 kFilterBox); 1714 1715 img = stream->img; 1716 } 1717 1718 vpx_usec_timer_start(&timer); 1719 vpx_codec_encode(&stream->encoder, img, frame_start, 1720 (unsigned long)(next_frame_start - frame_start), 1721 0, global->deadline); 1722 vpx_usec_timer_mark(&timer); 1723 stream->cx_time += vpx_usec_timer_elapsed(&timer); 1724 ctx_exit_on_error(&stream->encoder, "Stream %d: Failed to encode frame", 1725 stream->index); 1726 } 1727 1728 1729 static void update_quantizer_histogram(struct stream_state *stream) { 1730 if (stream->config.cfg.g_pass != VPX_RC_FIRST_PASS) { 1731 int q; 1732 1733 vpx_codec_control(&stream->encoder, VP8E_GET_LAST_QUANTIZER_64, &q); 1734 ctx_exit_on_error(&stream->encoder, "Failed to read quantizer"); 1735 stream->counts[q]++; 1736 } 1737 } 1738 1739 1740 static void get_cx_data(struct stream_state *stream, 1741 struct global_config *global, 1742 int *got_data) { 1743 const vpx_codec_cx_pkt_t *pkt; 1744 const struct vpx_codec_enc_cfg *cfg = &stream->config.cfg; 1745 vpx_codec_iter_t iter = NULL; 1746 1747 *got_data = 0; 1748 while ((pkt = vpx_codec_get_cx_data(&stream->encoder, &iter))) { 1749 static size_t fsize = 0; 1750 static off_t ivf_header_pos = 0; 1751 1752 switch (pkt->kind) { 1753 case VPX_CODEC_CX_FRAME_PKT: 1754 if (!(pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT)) { 1755 stream->frames_out++; 1756 } 1757 if (!global->quiet) 1758 fprintf(stderr, " %6luF", (unsigned long)pkt->data.frame.sz); 1759 1760 update_rate_histogram(&stream->rate_hist, cfg, pkt); 1761 if (stream->config.write_webm) { 1762 /* Update the hash */ 1763 if (!stream->ebml.debug) 1764 stream->hash = murmur(pkt->data.frame.buf, 1765 (int)pkt->data.frame.sz, 1766 stream->hash); 1767 1768 write_webm_block(&stream->ebml, cfg, pkt); 1769 } else { 1770 if (pkt->data.frame.partition_id <= 0) { 1771 ivf_header_pos = ftello(stream->file); 1772 fsize = pkt->data.frame.sz; 1773 1774 write_ivf_frame_header(stream->file, pkt); 1775 } else { 1776 fsize += pkt->data.frame.sz; 1777 1778 if (!(pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT)) { 1779 off_t currpos = ftello(stream->file); 1780 fseeko(stream->file, ivf_header_pos, SEEK_SET); 1781 write_ivf_frame_size(stream->file, fsize); 1782 fseeko(stream->file, currpos, SEEK_SET); 1783 } 1784 } 1785 1786 (void) fwrite(pkt->data.frame.buf, 1, pkt->data.frame.sz, 1787 stream->file); 1788 } 1789 stream->nbytes += pkt->data.raw.sz; 1790 1791 *got_data = 1; 1792 #if CONFIG_DECODERS 1793 if (global->test_decode != TEST_DECODE_OFF && !stream->mismatch_seen) { 1794 vpx_codec_decode(&stream->decoder, pkt->data.frame.buf, 1795 pkt->data.frame.sz, NULL, 0); 1796 if (stream->decoder.err) { 1797 warn_or_exit_on_error(&stream->decoder, 1798 global->test_decode == TEST_DECODE_FATAL, 1799 "Failed to decode frame %d in stream %d", 1800 stream->frames_out + 1, stream->index); 1801 stream->mismatch_seen = stream->frames_out + 1; 1802 } 1803 } 1804 #endif 1805 break; 1806 case VPX_CODEC_STATS_PKT: 1807 stream->frames_out++; 1808 stats_write(&stream->stats, 1809 pkt->data.twopass_stats.buf, 1810 pkt->data.twopass_stats.sz); 1811 stream->nbytes += pkt->data.raw.sz; 1812 break; 1813 case VPX_CODEC_PSNR_PKT: 1814 1815 if (global->show_psnr) { 1816 int i; 1817 1818 stream->psnr_sse_total += pkt->data.psnr.sse[0]; 1819 stream->psnr_samples_total += pkt->data.psnr.samples[0]; 1820 for (i = 0; i < 4; i++) { 1821 if (!global->quiet) 1822 fprintf(stderr, "%.3f ", pkt->data.psnr.psnr[i]); 1823 stream->psnr_totals[i] += pkt->data.psnr.psnr[i]; 1824 } 1825 stream->psnr_count++; 1826 } 1827 1828 break; 1829 default: 1830 break; 1831 } 1832 } 1833 } 1834 1835 1836 static void show_psnr(struct stream_state *stream) { 1837 int i; 1838 double ovpsnr; 1839 1840 if (!stream->psnr_count) 1841 return; 1842 1843 fprintf(stderr, "Stream %d PSNR (Overall/Avg/Y/U/V)", stream->index); 1844 ovpsnr = vp8_mse2psnr((double)stream->psnr_samples_total, 255.0, 1845 (double)stream->psnr_sse_total); 1846 fprintf(stderr, " %.3f", ovpsnr); 1847 1848 for (i = 0; i < 4; i++) { 1849 fprintf(stderr, " %.3f", stream->psnr_totals[i] / stream->psnr_count); 1850 } 1851 fprintf(stderr, "\n"); 1852 } 1853 1854 1855 static float usec_to_fps(uint64_t usec, unsigned int frames) { 1856 return (float)(usec > 0 ? frames * 1000000.0 / (float)usec : 0); 1857 } 1858 1859 1860 static void test_decode(struct stream_state *stream, 1861 enum TestDecodeFatality fatal, 1862 const struct codec_item *codec) { 1863 vpx_image_t enc_img, dec_img; 1864 1865 if (stream->mismatch_seen) 1866 return; 1867 1868 /* Get the internal reference frame */ 1869 if (codec->fourcc == VP8_FOURCC) { 1870 struct vpx_ref_frame ref_enc, ref_dec; 1871 int width, height; 1872 1873 width = (stream->config.cfg.g_w + 15) & ~15; 1874 height = (stream->config.cfg.g_h + 15) & ~15; 1875 vpx_img_alloc(&ref_enc.img, VPX_IMG_FMT_I420, width, height, 1); 1876 enc_img = ref_enc.img; 1877 vpx_img_alloc(&ref_dec.img, VPX_IMG_FMT_I420, width, height, 1); 1878 dec_img = ref_dec.img; 1879 1880 ref_enc.frame_type = VP8_LAST_FRAME; 1881 ref_dec.frame_type = VP8_LAST_FRAME; 1882 vpx_codec_control(&stream->encoder, VP8_COPY_REFERENCE, &ref_enc); 1883 vpx_codec_control(&stream->decoder, VP8_COPY_REFERENCE, &ref_dec); 1884 } else { 1885 struct vp9_ref_frame ref; 1886 1887 ref.idx = 0; 1888 vpx_codec_control(&stream->encoder, VP9_GET_REFERENCE, &ref); 1889 enc_img = ref.img; 1890 vpx_codec_control(&stream->decoder, VP9_GET_REFERENCE, &ref); 1891 dec_img = ref.img; 1892 } 1893 ctx_exit_on_error(&stream->encoder, "Failed to get encoder reference frame"); 1894 ctx_exit_on_error(&stream->decoder, "Failed to get decoder reference frame"); 1895 1896 if (!compare_img(&enc_img, &dec_img)) { 1897 int y[4], u[4], v[4]; 1898 find_mismatch(&enc_img, &dec_img, y, u, v); 1899 stream->decoder.err = 1; 1900 warn_or_exit_on_error(&stream->decoder, fatal == TEST_DECODE_FATAL, 1901 "Stream %d: Encode/decode mismatch on frame %d at" 1902 " Y[%d, %d] {%d/%d}," 1903 " U[%d, %d] {%d/%d}," 1904 " V[%d, %d] {%d/%d}", 1905 stream->index, stream->frames_out, 1906 y[0], y[1], y[2], y[3], 1907 u[0], u[1], u[2], u[3], 1908 v[0], v[1], v[2], v[3]); 1909 stream->mismatch_seen = stream->frames_out; 1910 } 1911 1912 vpx_img_free(&enc_img); 1913 vpx_img_free(&dec_img); 1914 } 1915 1916 1917 static void print_time(const char *label, int64_t etl) { 1918 int hours, mins, secs; 1919 1920 if (etl >= 0) { 1921 hours = etl / 3600; 1922 etl -= hours * 3600; 1923 mins = etl / 60; 1924 etl -= mins * 60; 1925 secs = etl; 1926 1927 fprintf(stderr, "[%3s %2d:%02d:%02d] ", 1928 label, hours, mins, secs); 1929 } else { 1930 fprintf(stderr, "[%3s unknown] ", label); 1931 } 1932 } 1933 1934 int main(int argc, const char **argv_) { 1935 int pass; 1936 vpx_image_t raw; 1937 int frame_avail, got_data; 1938 1939 struct input_state input = {0}; 1940 struct global_config global; 1941 struct stream_state *streams = NULL; 1942 char **argv, **argi; 1943 uint64_t cx_time = 0; 1944 int stream_cnt = 0; 1945 int res = 0; 1946 1947 exec_name = argv_[0]; 1948 1949 if (argc < 3) 1950 usage_exit(); 1951 1952 /* Setup default input stream settings */ 1953 input.framerate.num = 30; 1954 input.framerate.den = 1; 1955 input.use_i420 = 1; 1956 input.only_i420 = 1; 1957 1958 /* First parse the global configuration values, because we want to apply 1959 * other parameters on top of the default configuration provided by the 1960 * codec. 1961 */ 1962 argv = argv_dup(argc - 1, argv_ + 1); 1963 parse_global_config(&global, argv); 1964 1965 { 1966 /* Now parse each stream's parameters. Using a local scope here 1967 * due to the use of 'stream' as loop variable in FOREACH_STREAM 1968 * loops 1969 */ 1970 struct stream_state *stream = NULL; 1971 1972 do { 1973 stream = new_stream(&global, stream); 1974 stream_cnt++; 1975 if (!streams) 1976 streams = stream; 1977 } while (parse_stream_params(&global, stream, argv)); 1978 } 1979 1980 /* Check for unrecognized options */ 1981 for (argi = argv; *argi; argi++) 1982 if (argi[0][0] == '-' && argi[0][1]) 1983 die("Error: Unrecognized option %s\n", *argi); 1984 1985 /* Handle non-option arguments */ 1986 input.fn = argv[0]; 1987 1988 if (!input.fn) 1989 usage_exit(); 1990 1991 #if CONFIG_NON420 1992 /* Decide if other chroma subsamplings than 4:2:0 are supported */ 1993 if (global.codec->fourcc == VP9_FOURCC) 1994 input.only_i420 = 0; 1995 #endif 1996 1997 for (pass = global.pass ? global.pass - 1 : 0; pass < global.passes; pass++) { 1998 int frames_in = 0, seen_frames = 0; 1999 int64_t estimated_time_left = -1; 2000 int64_t average_rate = -1; 2001 off_t lagged_count = 0; 2002 2003 open_input_file(&input); 2004 2005 /* If the input file doesn't specify its w/h (raw files), try to get 2006 * the data from the first stream's configuration. 2007 */ 2008 if (!input.w || !input.h) 2009 FOREACH_STREAM( { 2010 if (stream->config.cfg.g_w && stream->config.cfg.g_h) { 2011 input.w = stream->config.cfg.g_w; 2012 input.h = stream->config.cfg.g_h; 2013 break; 2014 } 2015 }); 2016 2017 /* Update stream configurations from the input file's parameters */ 2018 if (!input.w || !input.h) 2019 fatal("Specify stream dimensions with --width (-w) " 2020 " and --height (-h)"); 2021 FOREACH_STREAM(set_stream_dimensions(stream, input.w, input.h)); 2022 FOREACH_STREAM(validate_stream_config(stream)); 2023 2024 /* Ensure that --passes and --pass are consistent. If --pass is set and 2025 * --passes=2, ensure --fpf was set. 2026 */ 2027 if (global.pass && global.passes == 2) 2028 FOREACH_STREAM( { 2029 if (!stream->config.stats_fn) 2030 die("Stream %d: Must specify --fpf when --pass=%d" 2031 " and --passes=2\n", stream->index, global.pass); 2032 }); 2033 2034 /* Use the frame rate from the file only if none was specified 2035 * on the command-line. 2036 */ 2037 if (!global.have_framerate) 2038 global.framerate = input.framerate; 2039 2040 FOREACH_STREAM(set_default_kf_interval(stream, &global)); 2041 2042 /* Show configuration */ 2043 if (global.verbose && pass == 0) 2044 FOREACH_STREAM(show_stream_config(stream, &global, &input)); 2045 2046 if (pass == (global.pass ? global.pass - 1 : 0)) { 2047 if (input.file_type == FILE_TYPE_Y4M) 2048 /*The Y4M reader does its own allocation. 2049 Just initialize this here to avoid problems if we never read any 2050 frames.*/ 2051 memset(&raw, 0, sizeof(raw)); 2052 else 2053 vpx_img_alloc(&raw, 2054 input.use_i420 ? VPX_IMG_FMT_I420 2055 : VPX_IMG_FMT_YV12, 2056 input.w, input.h, 32); 2057 2058 FOREACH_STREAM(init_rate_histogram(&stream->rate_hist, 2059 &stream->config.cfg, 2060 &global.framerate)); 2061 } 2062 2063 FOREACH_STREAM(setup_pass(stream, &global, pass)); 2064 FOREACH_STREAM(open_output_file(stream, &global)); 2065 FOREACH_STREAM(initialize_encoder(stream, &global)); 2066 2067 frame_avail = 1; 2068 got_data = 0; 2069 2070 while (frame_avail || got_data) { 2071 struct vpx_usec_timer timer; 2072 2073 if (!global.limit || frames_in < global.limit) { 2074 frame_avail = read_frame(&input, &raw); 2075 2076 if (frame_avail) 2077 frames_in++; 2078 seen_frames = frames_in > global.skip_frames ? 2079 frames_in - global.skip_frames : 0; 2080 2081 if (!global.quiet) { 2082 float fps = usec_to_fps(cx_time, seen_frames); 2083 fprintf(stderr, "\rPass %d/%d ", pass + 1, global.passes); 2084 2085 if (stream_cnt == 1) 2086 fprintf(stderr, 2087 "frame %4d/%-4d %7"PRId64"B ", 2088 frames_in, streams->frames_out, (int64_t)streams->nbytes); 2089 else 2090 fprintf(stderr, "frame %4d ", frames_in); 2091 2092 fprintf(stderr, "%7"PRId64" %s %.2f %s ", 2093 cx_time > 9999999 ? cx_time / 1000 : cx_time, 2094 cx_time > 9999999 ? "ms" : "us", 2095 fps >= 1.0 ? fps : fps * 60, 2096 fps >= 1.0 ? "fps" : "fpm"); 2097 print_time("ETA", estimated_time_left); 2098 fprintf(stderr, "\033[K"); 2099 } 2100 2101 } else 2102 frame_avail = 0; 2103 2104 if (frames_in > global.skip_frames) { 2105 vpx_usec_timer_start(&timer); 2106 FOREACH_STREAM(encode_frame(stream, &global, 2107 frame_avail ? &raw : NULL, 2108 frames_in)); 2109 vpx_usec_timer_mark(&timer); 2110 cx_time += vpx_usec_timer_elapsed(&timer); 2111 2112 FOREACH_STREAM(update_quantizer_histogram(stream)); 2113 2114 got_data = 0; 2115 FOREACH_STREAM(get_cx_data(stream, &global, &got_data)); 2116 2117 if (!got_data && input.length && !streams->frames_out) { 2118 lagged_count = global.limit ? seen_frames : ftello(input.file); 2119 } else if (input.length) { 2120 int64_t remaining; 2121 int64_t rate; 2122 2123 if (global.limit) { 2124 int frame_in_lagged = (seen_frames - lagged_count) * 1000; 2125 2126 rate = cx_time ? frame_in_lagged * (int64_t)1000000 / cx_time : 0; 2127 remaining = 1000 * (global.limit - global.skip_frames 2128 - seen_frames + lagged_count); 2129 } else { 2130 off_t input_pos = ftello(input.file); 2131 off_t input_pos_lagged = input_pos - lagged_count; 2132 int64_t limit = input.length; 2133 2134 rate = cx_time ? input_pos_lagged * (int64_t)1000000 / cx_time : 0; 2135 remaining = limit - input_pos + lagged_count; 2136 } 2137 2138 average_rate = (average_rate <= 0) 2139 ? rate 2140 : (average_rate * 7 + rate) / 8; 2141 estimated_time_left = average_rate ? remaining / average_rate : -1; 2142 } 2143 2144 if (got_data && global.test_decode != TEST_DECODE_OFF) 2145 FOREACH_STREAM(test_decode(stream, global.test_decode, global.codec)); 2146 } 2147 2148 fflush(stdout); 2149 } 2150 2151 if (stream_cnt > 1) 2152 fprintf(stderr, "\n"); 2153 2154 if (!global.quiet) 2155 FOREACH_STREAM(fprintf( 2156 stderr, 2157 "\rPass %d/%d frame %4d/%-4d %7"PRId64"B %7lub/f %7"PRId64"b/s" 2158 " %7"PRId64" %s (%.2f fps)\033[K\n", pass + 1, 2159 global.passes, frames_in, stream->frames_out, (int64_t)stream->nbytes, 2160 seen_frames ? (unsigned long)(stream->nbytes * 8 / seen_frames) : 0, 2161 seen_frames ? (int64_t)stream->nbytes * 8 2162 * (int64_t)global.framerate.num / global.framerate.den 2163 / seen_frames 2164 : 0, 2165 stream->cx_time > 9999999 ? stream->cx_time / 1000 : stream->cx_time, 2166 stream->cx_time > 9999999 ? "ms" : "us", 2167 usec_to_fps(stream->cx_time, seen_frames)); 2168 ); 2169 2170 if (global.show_psnr) 2171 FOREACH_STREAM(show_psnr(stream)); 2172 2173 FOREACH_STREAM(vpx_codec_destroy(&stream->encoder)); 2174 2175 if (global.test_decode != TEST_DECODE_OFF) { 2176 FOREACH_STREAM(vpx_codec_destroy(&stream->decoder)); 2177 } 2178 2179 close_input_file(&input); 2180 2181 if (global.test_decode == TEST_DECODE_FATAL) { 2182 FOREACH_STREAM(res |= stream->mismatch_seen); 2183 } 2184 FOREACH_STREAM(close_output_file(stream, global.codec->fourcc)); 2185 2186 FOREACH_STREAM(stats_close(&stream->stats, global.passes - 1)); 2187 2188 if (global.pass) 2189 break; 2190 } 2191 2192 if (global.show_q_hist_buckets) 2193 FOREACH_STREAM(show_q_histogram(stream->counts, 2194 global.show_q_hist_buckets)); 2195 2196 if (global.show_rate_hist_buckets) 2197 FOREACH_STREAM(show_rate_histogram(&stream->rate_hist, 2198 &stream->config.cfg, 2199 global.show_rate_hist_buckets)); 2200 FOREACH_STREAM(destroy_rate_histogram(&stream->rate_hist)); 2201 2202 #if CONFIG_INTERNAL_STATS 2203 /* TODO(jkoleszar): This doesn't belong in this executable. Do it for now, 2204 * to match some existing utilities. 2205 */ 2206 if (!(global.pass == 1 && global.passes == 2)) 2207 FOREACH_STREAM({ 2208 FILE *f = fopen("opsnr.stt", "a"); 2209 if (stream->mismatch_seen) { 2210 fprintf(f, "First mismatch occurred in frame %d\n", 2211 stream->mismatch_seen); 2212 } else { 2213 fprintf(f, "No mismatch detected in recon buffers\n"); 2214 } 2215 fclose(f); 2216 }); 2217 #endif 2218 2219 vpx_img_free(&raw); 2220 free(argv); 2221 free(streams); 2222 return res ? EXIT_FAILURE : EXIT_SUCCESS; 2223 } 2224
