// Initialize encoder, given setup in the encoder parameters bool LumaEncoder::initialize(const char *outputFile, const unsigned int w, const unsigned int h, bool verbose) { // Initialize base. Creates a Matroska file for writing LumaEncoderBase::initialize(outputFile, w, h); // Adjust profile for the specified bit depth (0-1 for 8 bits, and 2-3 for higher bit depths) if (m_params.profile > 1 && m_params.bitDepth == 8) m_params.profile -=2; if (m_params.profile < 2 && m_params.bitDepth > 8) m_params.profile +=2; // Initialize quantizer m_quant.setQuantizer(m_params.ptf, m_params.ptfBitDepth, m_params.colorSpace, m_params.colorBitDepth); // Add attachments with meta data to Matroska file unsigned int *buffer1 = new unsigned int; *buffer1 = m_params.ptfBitDepth; m_writer.addAttachment(430, (const binary*)buffer1, sizeof(unsigned int), "PTF bit depth"); unsigned int *buffer2 = new unsigned int; *buffer2 = m_params.colorBitDepth; m_writer.addAttachment(431, (const binary*)buffer2, sizeof(unsigned int), "Color bit depth"); LumaQuantizer::ptf_t *buffer3 = new LumaQuantizer::ptf_t; *buffer3 = m_params.ptf; m_writer.addAttachment(432, (const binary*)buffer3, sizeof(LumaQuantizer::ptf_t), "PTF description"); LumaQuantizer::colorSpace_t *buffer4 = new LumaQuantizer::colorSpace_t; *buffer4 = m_params.colorSpace; m_writer.addAttachment(433, (const binary*)buffer4, sizeof(LumaQuantizer::colorSpace_t), "Color space"); float *buffer5 = new float[m_quant.getSize()]; memcpy((void*)buffer5, (void*)m_quant.getMapping(), (m_quant.getSize())*sizeof(float)); m_writer.addAttachment(434, (const binary*)buffer5, (m_quant.getSize())*sizeof(float), "PTF"); float *buffer6 = new float; *buffer6 = m_params.preScaling; m_writer.addAttachment(435, (const binary*)buffer6, sizeof(float), "Scaling"); m_writer.writeAttachments(); // Framerate for timecodes m_writer.setFramerate(m_params.fps); m_writer.setVerbose(verbose); // Initialize VPX codec vpx_codec_err_t res; vpx_codec_enc_cfg_t cfg; const vpx_codec_iface_t *(*const vpx_encoder)() = &vpx_codec_vp9_cx; if (w <= 0 || h <= 0 || (w % 2) != 0 || (h % 2) != 0) throw LumaException("Invalid frame size"); if (m_params.profile == 0 && !vpx_img_alloc(&m_rawFrame, VPX_IMG_FMT_I420, w, h, 32)) throw LumaException("Failed to allocate 8 bit 420 image"); else if (m_params.profile == 1 && !vpx_img_alloc(&m_rawFrame, VPX_IMG_FMT_I444, w, h, 32)) throw LumaException("Failed to allocate 8 bit 444 image"); else if (m_params.profile == 2 && !vpx_img_alloc(&m_rawFrame, VPX_IMG_FMT_I42016, w, h, 32)) throw LumaException("Failed to allocate 16 bit 420 image"); else if (m_params.profile == 3 && !vpx_img_alloc(&m_rawFrame, VPX_IMG_FMT_I44416, w, h, 32)) throw LumaException("Failed to allocate 16 bit 444 image"); res = vpx_codec_enc_config_default(vpx_encoder(), &cfg, 0); if (res) throw LumaException("Failed to get default codec config"); cfg.g_threads = 6; cfg.rc_min_quantizer = m_params.quantizerScale; cfg.rc_max_quantizer = m_params.quantizerScale; cfg.g_w = w; cfg.g_h = h; cfg.g_timebase.num = 1; cfg.g_timebase.den = 25; cfg.rc_target_bitrate = m_params.bitrate; cfg.g_error_resilient = 0; cfg.g_pass = VPX_RC_ONE_PASS; cfg.rc_end_usage = VPX_VBR; cfg.g_lag_in_frames = 0; cfg.rc_end_usage = VPX_Q; cfg.kf_max_dist = 25; cfg.kf_mode = VPX_KF_AUTO; cfg.g_profile = m_params.profile; fprintf(stderr, "Encoding options:\n"); fprintf(stderr, "--------------------------------------------------------\n"); fprintf(stderr, "Transfer function (PTF): %s\n", m_quant.name(m_params.ptf).c_str()); fprintf(stderr, "Color space: %s\n", m_quant.name(m_params.colorSpace).c_str()); fprintf(stderr, "PTF bit depth: %d\n", m_params.ptfBitDepth); fprintf(stderr, "Color bit depth: %d\n", m_params.colorBitDepth); fprintf(stderr, "Encoding profile: %d (4%d%d)\n", m_params.profile, (m_params.profile%2==0) ? 2 : 4, (m_params.profile%2==0) ? 2 : 4); fprintf(stderr, "Encoding bit depth: "); if (m_params.bitDepth == 8 || m_params.profile < 2) { cfg.g_bit_depth = VPX_BITS_8; fprintf(stderr, "8\n"); } else if (m_params.bitDepth == 10) { cfg.g_bit_depth = VPX_BITS_10; fprintf(stderr, "10\n"); } else { cfg.g_bit_depth = VPX_BITS_12; fprintf(stderr, "12\n"); } fprintf(stderr, "Codec: %s\n", vpx_codec_iface_name(vpx_encoder())); fprintf(stderr, "Output: %s\n", outputFile); fprintf(stderr, "--------------------------------------------------------\n\n"); int flags = m_params.profile < 2 ? 0 : VPX_CODEC_USE_HIGHBITDEPTH; if (vpx_codec_enc_init(&m_codec, vpx_encoder(), &cfg, flags)) throw LumaException("Failed to initialize vpxEncoder\n"); if (m_params.lossLess && vpx_codec_control_(&m_codec, VP9E_SET_LOSSLESS, 1)) throw LumaException("Failed to use lossless mode\n"); m_initialized = true; return true; }
int main(int argc, const char **argv_) { vpx_codec_ctx_t encoder; const char *in_fn = NULL, *out_fn = NULL, *stats_fn = NULL; int i; FILE *infile, *outfile; vpx_codec_enc_cfg_t cfg; vpx_codec_err_t res; int pass, one_pass_only = 0; stats_io_t stats; vpx_image_t raw; const struct codec_item *codec = codecs; int frame_avail, got_data; struct arg arg; char **argv, **argi, **argj; int arg_usage = 0, arg_passes = 1, arg_deadline = 0; int arg_ctrls[ARG_CTRL_CNT_MAX][2], arg_ctrl_cnt = 0; int arg_limit = 0; static const arg_def_t **ctrl_args = no_args; static const int *ctrl_args_map = NULL; int verbose = 0, show_psnr = 0; int arg_use_i420 = 1; int arg_have_timebase = 0; unsigned long cx_time = 0; unsigned int file_type, fourcc; y4m_input y4m; exec_name = argv_[0]; if (argc < 3) usage_exit(); /* First parse the codec and usage values, because we want to apply other * parameters on top of the default configuration provided by the codec. */ argv = argv_dup(argc - 1, argv_ + 1); for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) { arg.argv_step = 1; if (arg_match(&arg, &codecarg, argi)) { int j, k = -1; for (j = 0; j < sizeof(codecs) / sizeof(codecs[0]); j++) if (!strcmp(codecs[j].name, arg.val)) k = j; if (k >= 0) codec = codecs + k; else die("Error: Unrecognized argument (%s) to --codec\n", arg.val); } else if (arg_match(&arg, &passes, argi)) { arg_passes = arg_parse_uint(&arg); if (arg_passes < 1 || arg_passes > 2) die("Error: Invalid number of passes (%d)\n", arg_passes); } else if (arg_match(&arg, &pass_arg, argi)) { one_pass_only = arg_parse_uint(&arg); if (one_pass_only < 1 || one_pass_only > 2) die("Error: Invalid pass selected (%d)\n", one_pass_only); } else if (arg_match(&arg, &fpf_name, argi)) stats_fn = arg.val; else if (arg_match(&arg, &usage, argi)) arg_usage = arg_parse_uint(&arg); else if (arg_match(&arg, &deadline, argi)) arg_deadline = arg_parse_uint(&arg); else if (arg_match(&arg, &best_dl, argi)) arg_deadline = VPX_DL_BEST_QUALITY; else if (arg_match(&arg, &good_dl, argi)) arg_deadline = VPX_DL_GOOD_QUALITY; else if (arg_match(&arg, &rt_dl, argi)) arg_deadline = VPX_DL_REALTIME; else if (arg_match(&arg, &use_yv12, argi)) { arg_use_i420 = 0; } else if (arg_match(&arg, &use_i420, argi)) { arg_use_i420 = 1; } else if (arg_match(&arg, &verbosearg, argi)) verbose = 1; else if (arg_match(&arg, &limit, argi)) arg_limit = arg_parse_uint(&arg); else if (arg_match(&arg, &psnrarg, argi)) show_psnr = 1; else argj++; } /* Ensure that --passes and --pass are consistent. If --pass is set and --passes=2, * ensure --fpf was set. */ if (one_pass_only) { /* DWIM: Assume the user meant passes=2 if pass=2 is specified */ if (one_pass_only > arg_passes) { fprintf(stderr, "Warning: Assuming --pass=%d implies --passes=%d\n", one_pass_only, one_pass_only); arg_passes = one_pass_only; } if (arg_passes == 2 && !stats_fn) die("Must specify --fpf when --pass=%d and --passes=2\n", one_pass_only); } /* Populate encoder configuration */ res = vpx_codec_enc_config_default(codec->iface, &cfg, arg_usage); if (res) { fprintf(stderr, "Failed to get config: %s\n", vpx_codec_err_to_string(res)); return EXIT_FAILURE; } /* Now parse the remainder of the parameters. */ for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) { arg.argv_step = 1; if (0); else if (arg_match(&arg, &threads, argi)) cfg.g_threads = arg_parse_uint(&arg); else if (arg_match(&arg, &profile, argi)) cfg.g_profile = arg_parse_uint(&arg); else if (arg_match(&arg, &width, argi)) cfg.g_w = arg_parse_uint(&arg); else if (arg_match(&arg, &height, argi)) cfg.g_h = arg_parse_uint(&arg); else if (arg_match(&arg, &timebase, argi)) { cfg.g_timebase = arg_parse_rational(&arg); arg_have_timebase = 1; } else if (arg_match(&arg, &error_resilient, argi)) cfg.g_error_resilient = arg_parse_uint(&arg); else if (arg_match(&arg, &lag_in_frames, argi)) cfg.g_lag_in_frames = arg_parse_uint(&arg); else if (arg_match(&arg, &dropframe_thresh, argi)) cfg.rc_dropframe_thresh = arg_parse_uint(&arg); else if (arg_match(&arg, &resize_allowed, argi)) cfg.rc_resize_allowed = arg_parse_uint(&arg); else if (arg_match(&arg, &resize_up_thresh, argi)) cfg.rc_resize_up_thresh = arg_parse_uint(&arg); else if (arg_match(&arg, &resize_down_thresh, argi)) cfg.rc_resize_down_thresh = arg_parse_uint(&arg); else if (arg_match(&arg, &resize_down_thresh, argi)) cfg.rc_resize_down_thresh = arg_parse_uint(&arg); else if (arg_match(&arg, &end_usage, argi)) cfg.rc_end_usage = arg_parse_uint(&arg); else if (arg_match(&arg, &target_bitrate, argi)) cfg.rc_target_bitrate = arg_parse_uint(&arg); else if (arg_match(&arg, &min_quantizer, argi)) cfg.rc_min_quantizer = arg_parse_uint(&arg); else if (arg_match(&arg, &max_quantizer, argi)) cfg.rc_max_quantizer = arg_parse_uint(&arg); else if (arg_match(&arg, &undershoot_pct, argi)) cfg.rc_undershoot_pct = arg_parse_uint(&arg); else if (arg_match(&arg, &overshoot_pct, argi)) cfg.rc_overshoot_pct = arg_parse_uint(&arg); else if (arg_match(&arg, &buf_sz, argi)) cfg.rc_buf_sz = arg_parse_uint(&arg); else if (arg_match(&arg, &buf_initial_sz, argi)) cfg.rc_buf_initial_sz = arg_parse_uint(&arg); else if (arg_match(&arg, &buf_optimal_sz, argi)) cfg.rc_buf_optimal_sz = arg_parse_uint(&arg); else if (arg_match(&arg, &bias_pct, argi)) { cfg.rc_2pass_vbr_bias_pct = arg_parse_uint(&arg); if (arg_passes < 2) fprintf(stderr, "Warning: option %s ignored in one-pass mode.\n", arg.name); } else if (arg_match(&arg, &minsection_pct, argi)) { cfg.rc_2pass_vbr_minsection_pct = arg_parse_uint(&arg); if (arg_passes < 2) fprintf(stderr, "Warning: option %s ignored in one-pass mode.\n", arg.name); } else if (arg_match(&arg, &maxsection_pct, argi)) { cfg.rc_2pass_vbr_maxsection_pct = arg_parse_uint(&arg); if (arg_passes < 2) fprintf(stderr, "Warning: option %s ignored in one-pass mode.\n", arg.name); } else if (arg_match(&arg, &kf_min_dist, argi)) cfg.kf_min_dist = arg_parse_uint(&arg); else if (arg_match(&arg, &kf_max_dist, argi)) cfg.kf_max_dist = arg_parse_uint(&arg); else if (arg_match(&arg, &kf_disabled, argi)) cfg.kf_mode = VPX_KF_DISABLED; else argj++; } /* Handle codec specific options */ #if CONFIG_VP8_ENCODER if (codec->iface == &vpx_codec_vp8_cx_algo) { ctrl_args = vp8_args; ctrl_args_map = vp8_arg_ctrl_map; } #endif for (argi = argj = argv; (*argj = *argi); argi += arg.argv_step) { int match = 0; arg.argv_step = 1; for (i = 0; ctrl_args[i]; i++) { if (arg_match(&arg, ctrl_args[i], argi)) { match = 1; if (arg_ctrl_cnt < ARG_CTRL_CNT_MAX) { arg_ctrls[arg_ctrl_cnt][0] = ctrl_args_map[i]; arg_ctrls[arg_ctrl_cnt][1] = arg_parse_int(&arg); arg_ctrl_cnt++; } } } if (!match) argj++; } /* Check for unrecognized options */ for (argi = argv; *argi; argi++) if (argi[0][0] == '-' && argi[0][1]) die("Error: Unrecognized option %s\n", *argi); /* Handle non-option arguments */ in_fn = argv[0]; out_fn = argv[1]; if (!in_fn || !out_fn) usage_exit(); memset(&stats, 0, sizeof(stats)); for (pass = one_pass_only ? one_pass_only - 1 : 0; pass < arg_passes; pass++) { int frames_in = 0, frames_out = 0; unsigned long nbytes = 0; struct detect_buffer detect; /* Parse certain options from the input file, if possible */ infile = strcmp(in_fn, "-") ? fopen(in_fn, "rb") : stdin; if (!infile) { fprintf(stderr, "Failed to open input file\n"); return EXIT_FAILURE; } fread(detect.buf, 1, 4, infile); detect.valid = 0; if (file_is_y4m(infile, &y4m, detect.buf)) { if (y4m_input_open(&y4m, infile, detect.buf, 4) >= 0) { file_type = FILE_TYPE_Y4M; cfg.g_w = y4m.pic_w; cfg.g_h = y4m.pic_h; /* Use the frame rate from the file only if none was specified * on the command-line. */ if (!arg_have_timebase) { cfg.g_timebase.num = y4m.fps_d; cfg.g_timebase.den = y4m.fps_n; } arg_use_i420 = 0; } else { fprintf(stderr, "Unsupported Y4M stream.\n"); return EXIT_FAILURE; } } else if (file_is_ivf(infile, &fourcc, &cfg.g_w, &cfg.g_h, detect.buf)) { file_type = FILE_TYPE_IVF; switch (fourcc) { case 0x32315659: arg_use_i420 = 0; break; case 0x30323449: arg_use_i420 = 1; break; default: fprintf(stderr, "Unsupported fourcc (%08x) in IVF\n", fourcc); return EXIT_FAILURE; } } else { file_type = FILE_TYPE_RAW; detect.valid = 1; } #define SHOW(field) fprintf(stderr, " %-28s = %d\n", #field, cfg.field) if (verbose && pass == 0) { fprintf(stderr, "Codec: %s\n", vpx_codec_iface_name(codec->iface)); fprintf(stderr, "Source file: %s Format: %s\n", in_fn, arg_use_i420 ? "I420" : "YV12"); fprintf(stderr, "Destination file: %s\n", out_fn); fprintf(stderr, "Encoder parameters:\n"); SHOW(g_usage); SHOW(g_threads); SHOW(g_profile); SHOW(g_w); SHOW(g_h); SHOW(g_timebase.num); SHOW(g_timebase.den); SHOW(g_error_resilient); SHOW(g_pass); SHOW(g_lag_in_frames); SHOW(rc_dropframe_thresh); SHOW(rc_resize_allowed); SHOW(rc_resize_up_thresh); SHOW(rc_resize_down_thresh); SHOW(rc_end_usage); SHOW(rc_target_bitrate); SHOW(rc_min_quantizer); SHOW(rc_max_quantizer); SHOW(rc_undershoot_pct); SHOW(rc_overshoot_pct); SHOW(rc_buf_sz); SHOW(rc_buf_initial_sz); SHOW(rc_buf_optimal_sz); SHOW(rc_2pass_vbr_bias_pct); SHOW(rc_2pass_vbr_minsection_pct); SHOW(rc_2pass_vbr_maxsection_pct); SHOW(kf_mode); SHOW(kf_min_dist); SHOW(kf_max_dist); } if(pass == (one_pass_only ? one_pass_only - 1 : 0)) { if (file_type == FILE_TYPE_Y4M) /*The Y4M reader does its own allocation. Just initialize this here to avoid problems if we never read any frames.*/ memset(&raw, 0, sizeof(raw)); else vpx_img_alloc(&raw, arg_use_i420 ? VPX_IMG_FMT_I420 : VPX_IMG_FMT_YV12, cfg.g_w, cfg.g_h, 1); // This was added so that ivfenc will create monotically increasing // timestamps. Since we create new timestamps for alt-reference frames // we need to make room in the series of timestamps. Since there can // only be 1 alt-ref frame ( current bitstream) multiplying by 2 // gives us enough room. cfg.g_timebase.den *= 2; } outfile = strcmp(out_fn, "-") ? fopen(out_fn, "wb") : stdout; if (!outfile) { fprintf(stderr, "Failed to open output file\n"); return EXIT_FAILURE; } if (stats_fn) { if (!stats_open_file(&stats, stats_fn, pass)) { fprintf(stderr, "Failed to open statistics store\n"); return EXIT_FAILURE; } } else { if (!stats_open_mem(&stats, pass)) { fprintf(stderr, "Failed to open statistics store\n"); return EXIT_FAILURE; } } cfg.g_pass = arg_passes == 2 ? pass ? VPX_RC_LAST_PASS : VPX_RC_FIRST_PASS : VPX_RC_ONE_PASS; #if VPX_ENCODER_ABI_VERSION > (1 + VPX_CODEC_ABI_VERSION) if (pass) { cfg.rc_twopass_stats_in = stats_get(&stats); } #endif write_ivf_file_header(outfile, &cfg, codec->fourcc, 0); /* Construct Encoder Context */ vpx_codec_enc_init(&encoder, codec->iface, &cfg, show_psnr ? VPX_CODEC_USE_PSNR : 0); ctx_exit_on_error(&encoder, "Failed to initialize encoder"); /* Note that we bypass the vpx_codec_control wrapper macro because * we're being clever to store the control IDs in an array. Real * applications will want to make use of the enumerations directly */ for (i = 0; i < arg_ctrl_cnt; i++) { if (vpx_codec_control_(&encoder, arg_ctrls[i][0], arg_ctrls[i][1])) fprintf(stderr, "Error: Tried to set control %d = %d\n", arg_ctrls[i][0], arg_ctrls[i][1]); ctx_exit_on_error(&encoder, "Failed to control codec"); } frame_avail = 1; got_data = 0; while (frame_avail || got_data) { vpx_codec_iter_t iter = NULL; const vpx_codec_cx_pkt_t *pkt; struct vpx_usec_timer timer; if (!arg_limit || frames_in < arg_limit) { frame_avail = read_frame(infile, &raw, file_type, &y4m, &detect); if (frame_avail) frames_in++; fprintf(stderr, "\rPass %d/%d frame %4d/%-4d %7ldB \033[K", pass + 1, arg_passes, frames_in, frames_out, nbytes); } else frame_avail = 0; vpx_usec_timer_start(&timer); // since we halved our timebase we need to double the timestamps // and duration we pass in. vpx_codec_encode(&encoder, frame_avail ? &raw : NULL, (frames_in - 1) * 2, 2, 0, arg_deadline); vpx_usec_timer_mark(&timer); cx_time += vpx_usec_timer_elapsed(&timer); ctx_exit_on_error(&encoder, "Failed to encode frame"); got_data = 0; while ((pkt = vpx_codec_get_cx_data(&encoder, &iter))) { got_data = 1; switch (pkt->kind) { case VPX_CODEC_CX_FRAME_PKT: frames_out++; fprintf(stderr, " %6luF", (unsigned long)pkt->data.frame.sz); write_ivf_frame_header(outfile, pkt); fwrite(pkt->data.frame.buf, 1, pkt->data.frame.sz, outfile); nbytes += pkt->data.raw.sz; break; case VPX_CODEC_STATS_PKT: frames_out++; fprintf(stderr, " %6luS", (unsigned long)pkt->data.twopass_stats.sz); stats_write(&stats, pkt->data.twopass_stats.buf, pkt->data.twopass_stats.sz); nbytes += pkt->data.raw.sz; break; case VPX_CODEC_PSNR_PKT: if (show_psnr) { int i; for (i = 0; i < 4; i++) fprintf(stderr, "%.3lf ", pkt->data.psnr.psnr[i]); } break; default: break; } } fflush(stdout); } /* this bitrate calc is simplified and relies on the fact that this * application uses 1/timebase for framerate. */ fprintf(stderr, "\rPass %d/%d frame %4d/%-4d %7ldB %7ldb/f %7"PRId64"b/s" " %7lu %s (%.2f fps)\033[K", pass + 1, arg_passes, frames_in, frames_out, nbytes, nbytes * 8 / frames_in, nbytes * 8 *(int64_t)cfg.g_timebase.den/2/ cfg.g_timebase.num / frames_in, cx_time > 9999999 ? cx_time / 1000 : cx_time, cx_time > 9999999 ? "ms" : "us", (float)frames_in * 1000000.0 / (float)cx_time); vpx_codec_destroy(&encoder); fclose(infile); if (!fseek(outfile, 0, SEEK_SET)) write_ivf_file_header(outfile, &cfg, codec->fourcc, frames_out); fclose(outfile); stats_close(&stats); fprintf(stderr, "\n"); if (one_pass_only) break; } vpx_img_free(&raw); free(argv); return EXIT_SUCCESS; }