/* work.c Copyright (c) 2003-2017 HandBrake Team This file is part of the HandBrake source code Homepage: . It may be used under the terms of the GNU General Public License v2. For full terms see the file COPYING file or visit http://www.gnu.org/licenses/gpl-2.0.html */ #include "hb.h" #include "libavformat/avformat.h" #include "decomb.h" #ifdef USE_QSV #include "qsv_common.h" #include "qsv_filter_pp.h" #endif typedef struct { hb_list_t * jobs; hb_job_t ** current_job; hb_error_code * error; volatile int * die; } hb_work_t; static void work_func(); static void do_job( hb_job_t *); static void filter_loop( void * ); #define FIFO_UNBOUNDED 65536 #define FIFO_UNBOUNDED_WAKE 65535 #define FIFO_LARGE 32 #define FIFO_LARGE_WAKE 16 #define FIFO_SMALL 16 #define FIFO_SMALL_WAKE 15 #define FIFO_MINI 4 #define FIFO_MINI_WAKE 3 /** * Allocates work object and launches work thread with work_func. * @param jobs Handle to hb_list_t. * @param die Handle to user inititated exit indicator. * @param error Handle to error indicator. */ hb_thread_t * hb_work_init( hb_list_t * jobs, volatile int * die, hb_error_code * error, hb_job_t ** job ) { hb_work_t * work = calloc( sizeof( hb_work_t ), 1 ); work->jobs = jobs; work->current_job = job; work->die = die; work->error = error; return hb_thread_init( "work", work_func, work, HB_LOW_PRIORITY ); } static void InitWorkState(hb_handle_t *h, int pass_id, int pass, int pass_count) { hb_state_t state; state.state = HB_STATE_WORKING; #define p state.param.working p.pass_id = pass_id; p.pass = pass; p.pass_count = pass_count; p.progress = 0.0; p.rate_cur = 0.0; p.rate_avg = 0.0; p.hours = -1; p.minutes = -1; p.seconds = -1; #undef p hb_set_state( h, &state ); } /** * Iterates through job list and calls do_job for each job. * @param _work Handle work object. */ static void work_func( void * _work ) { hb_work_t * work = _work; hb_job_t * job; hb_log( "%d job(s) to process", hb_list_count( work->jobs ) ); while( !*work->die && ( job = hb_list_item( work->jobs, 0 ) ) ) { hb_handle_t * h = job->h; hb_list_rem( work->jobs, job ); hb_list_t * passes = hb_list_init(); // JSON jobs get special treatment. We want to perform the title // scan for the JSON job automatically. This requires that we delay // filling the job struct till we have performed the title scan // because the default values for the job come from the title. if (job->json != NULL) { hb_deep_log(1, "json job:\n%s", job->json); // Perform title scan for json job hb_json_job_scan(job->h, job->json); // Expand json string to full job struct hb_job_t *new_job = hb_json_to_job(job->h, job->json); if (new_job == NULL) { hb_job_close(&job); hb_list_close(&passes); *work->error = HB_ERROR_INIT; *work->die = 1; break; } new_job->h = job->h; new_job->sequence_id = job->sequence_id; hb_job_close(&job); job = new_job; } hb_job_setup_passes(job->h, job, passes); hb_job_close(&job); int pass_count, pass; pass_count = hb_list_count(passes); for (pass = 0; pass < pass_count && !*work->die; pass++) { job = hb_list_item(passes, pass); job->die = work->die; job->done_error = work->error; *(work->current_job) = job; InitWorkState(job->h, job->pass_id, pass + 1, pass_count); do_job( job ); *(work->current_job) = NULL; } // Clean up any incomplete jobs for (; pass < pass_count; pass++) { job = hb_list_item(passes, pass); hb_job_close(&job); } hb_list_close(&passes); // Force rescan of next source processed by this hb_handle_t // TODO: Fix this ugly hack! hb_force_rescan(h); } free( work ); } hb_work_object_t * hb_get_work( hb_handle_t *h, int id ) { hb_work_object_t * w; for( w = hb_objects; w; w = w->next ) { if( w->id == id ) { hb_work_object_t *wc = malloc( sizeof(*w) ); *wc = *w; wc->h = h; return wc; } } return NULL; } hb_work_object_t* hb_audio_decoder(hb_handle_t *h, int codec) { hb_work_object_t * w = NULL; if (codec & HB_ACODEC_FF_MASK) { w = hb_get_work(h, WORK_DECAVCODEC); } switch (codec) { case HB_ACODEC_LPCM: w = hb_get_work(h, WORK_DECLPCM); break; default: break; } return w; } hb_work_object_t* hb_video_decoder(hb_handle_t *h, int vcodec, int param) { hb_work_object_t * w; w = hb_get_work(h, vcodec); if (w == NULL) { hb_error("Invalid video decoder: codec %d, param %d", vcodec, param); return NULL; } w->codec_param = param; return w; } hb_work_object_t* hb_video_encoder(hb_handle_t *h, int vcodec) { hb_work_object_t * w = NULL; switch (vcodec) { case HB_VCODEC_FFMPEG_MPEG4: w = hb_get_work(h, WORK_ENCAVCODEC); w->codec_param = AV_CODEC_ID_MPEG4; break; case HB_VCODEC_FFMPEG_MPEG2: w = hb_get_work(h, WORK_ENCAVCODEC); w->codec_param = AV_CODEC_ID_MPEG2VIDEO; break; case HB_VCODEC_FFMPEG_VP8: w = hb_get_work(h, WORK_ENCAVCODEC); w->codec_param = AV_CODEC_ID_VP8; break; case HB_VCODEC_FFMPEG_VP9: w = hb_get_work(h, WORK_ENCAVCODEC); w->codec_param = AV_CODEC_ID_VP9; break; case HB_VCODEC_X264_8BIT: case HB_VCODEC_X264_10BIT: w = hb_get_work(h, WORK_ENCX264); break; case HB_VCODEC_QSV_H264: case HB_VCODEC_QSV_H265: case HB_VCODEC_QSV_H265_10BIT: w = hb_get_work(h, WORK_ENCQSV); break; case HB_VCODEC_THEORA: w = hb_get_work(h, WORK_ENCTHEORA); break; #ifdef USE_X265 case HB_VCODEC_X265_8BIT: case HB_VCODEC_X265_10BIT: case HB_VCODEC_X265_12BIT: case HB_VCODEC_X265_16BIT: w = hb_get_work(h, WORK_ENCX265); break; #endif default: hb_error("Unknown video codec (0x%x)", vcodec ); } return w; } hb_work_object_t* hb_audio_encoder(hb_handle_t *h, int codec) { if (codec & HB_ACODEC_FF_MASK) { return hb_get_work(h, WORK_ENCAVCODEC_AUDIO); } switch (codec) { case HB_ACODEC_AC3: case HB_ACODEC_LAME: return hb_get_work(h, WORK_ENCAVCODEC_AUDIO); case HB_ACODEC_VORBIS: return hb_get_work(h, WORK_ENCVORBIS); case HB_ACODEC_CA_AAC: return hb_get_work(h, WORK_ENC_CA_AAC); case HB_ACODEC_CA_HAAC: return hb_get_work(h, WORK_ENC_CA_HAAC); default: break; } return NULL; } /** * Displays job parameters in the debug log. * @param job Handle work hb_job_t. */ void hb_display_job_info(hb_job_t *job) { int i; hb_title_t *title = job->title; hb_audio_t *audio; hb_subtitle_t *subtitle; hb_log("job configuration:"); hb_log( " * source"); hb_log( " + %s", title->path ); if( job->pts_to_start || job->pts_to_stop ) { int64_t stop; int hr_start, min_start, hr_stop, min_stop; float sec_start, sec_stop; stop = job->pts_to_start + job->pts_to_stop; hr_start = job->pts_to_start / (90000 * 60 * 60); min_start = job->pts_to_start / (90000 * 60); sec_start = (float)job->pts_to_start / 90000.0 - min_start * 60; min_start %= 60; hr_stop = stop / (90000 * 60 * 60); min_stop = stop / (90000 * 60); sec_stop = (float)stop / 90000.0 - min_stop * 60; min_stop %= 60; if (job->pts_to_stop) { hb_log(" + title %d, start %02d:%02d:%02.2f stop %02d:%02d:%02.2f", title->index, hr_start, min_start, sec_start, hr_stop, min_stop, sec_stop); } else { hb_log(" + title %d, start %02d:%02d:%02.2f", title->index, hr_start, min_start, sec_start); } } else if( job->frame_to_start || job->frame_to_stop ) { hb_log(" + title %d, frames %d to %d", title->index, job->frame_to_start, job->frame_to_start + job->frame_to_stop - 1); } else { hb_log( " + title %d, chapter(s) %d to %d", title->index, job->chapter_start, job->chapter_end ); } if( title->container_name != NULL ) hb_log( " + container: %s", title->container_name); if( title->data_rate ) { hb_log( " + data rate: %d kbps", title->data_rate / 1000 ); } hb_log( " * destination"); hb_log( " + %s", job->file ); hb_log(" + container: %s", hb_container_get_long_name(job->mux)); switch (job->mux) { case HB_MUX_AV_MP4: if (job->mp4_optimize) hb_log(" + optimized for HTTP streaming (fast start)"); if (job->ipod_atom) hb_log(" + compatibility atom for iPod 5G"); break; default: break; } if (job->align_av_start) { hb_log(" + align initial A/V stream timestamps"); } if (job->inline_parameter_sets) { hb_log(" + optimized for adaptive streaming (inline parameter sets)"); } if( job->chapter_markers ) { hb_log( " + chapter markers" ); } hb_log(" * video track"); #ifdef USE_QSV if (hb_qsv_decode_is_enabled(job)) { hb_log(" + decoder: %s", hb_qsv_decode_get_codec_name(title->video_codec_param)); } else #endif { hb_log(" + decoder: %s", title->video_codec_name); } if( title->video_bitrate ) { hb_log( " + bitrate %d kbps", title->video_bitrate / 1000 ); } // Filters can modify dimensions. So show them first. if( hb_list_count( job->list_filter ) ) { hb_log(" + %s", hb_list_count( job->list_filter) > 1 ? "filters" : "filter" ); for( i = 0; i < hb_list_count( job->list_filter ); i++ ) { hb_filter_object_t * filter = hb_list_item( job->list_filter, i ); char * settings = hb_filter_settings_string(filter->id, filter->settings); if (settings != NULL) hb_log(" + %s (%s)", filter->name, settings); else hb_log(" + %s (default settings)", filter->name); free(settings); if (filter->info) { hb_filter_info_t * info; info = filter->info(filter); if (info != NULL && info->human_readable_desc != NULL && info->human_readable_desc[0] != 0) { char * line, * pos = NULL; char * tmp = strdup(info->human_readable_desc); for (line = strtok_r(tmp, "\n", &pos); line != NULL; line = strtok_r(NULL, "\n", &pos)) { hb_log(" + %s", line); } free(tmp); } hb_filter_info_close(&info); } } } hb_log( " + Output geometry" ); hb_log( " + storage dimensions: %d x %d", job->width, job->height ); hb_log( " + pixel aspect ratio: %d : %d", job->par.num, job->par.den ); hb_log( " + display dimensions: %d x %d", job->width * job->par.num / job->par.den, job->height ); if( !job->indepth_scan ) { /* Video encoder */ hb_log(" + encoder: %s", hb_video_encoder_get_long_name(job->vcodec)); if (job->encoder_preset && *job->encoder_preset && hb_video_encoder_get_presets(job->vcodec) != NULL) { hb_log(" + preset: %s", job->encoder_preset); } if (job->encoder_tune && *job->encoder_tune) { switch (job->vcodec) { case HB_VCODEC_X264_8BIT: case HB_VCODEC_X264_10BIT: case HB_VCODEC_X265_8BIT: case HB_VCODEC_X265_10BIT: case HB_VCODEC_X265_12BIT: case HB_VCODEC_X265_16BIT: hb_log(" + tune: %s", job->encoder_tune); default: break; } } if (job->encoder_options != NULL && *job->encoder_options && job->vcodec != HB_VCODEC_THEORA) { hb_log(" + options: %s", job->encoder_options); } if (job->encoder_profile && *job->encoder_profile) { switch (job->vcodec) { case HB_VCODEC_X264_8BIT: case HB_VCODEC_X264_10BIT: case HB_VCODEC_X265_8BIT: case HB_VCODEC_X265_10BIT: case HB_VCODEC_X265_12BIT: case HB_VCODEC_X265_16BIT: case HB_VCODEC_QSV_H264: case HB_VCODEC_QSV_H265: case HB_VCODEC_QSV_H265_10BIT: hb_log(" + profile: %s", job->encoder_profile); default: break; } } if (job->encoder_level && *job->encoder_level) { switch (job->vcodec) { case HB_VCODEC_X264_8BIT: case HB_VCODEC_X264_10BIT: case HB_VCODEC_QSV_H264: case HB_VCODEC_QSV_H265: case HB_VCODEC_QSV_H265_10BIT: hb_log(" + level: %s", job->encoder_level); default: break; } } if (job->vquality > HB_INVALID_VIDEO_QUALITY) { hb_log(" + quality: %.2f (%s)", job->vquality, hb_video_quality_get_name(job->vcodec)); } else { hb_log( " + bitrate: %d kbps, pass: %d", job->vbitrate, job->pass_id ); if(job->pass_id == HB_PASS_ENCODE_1ST && job->fastfirstpass == 1 && ((job->vcodec & HB_VCODEC_X264_MASK) || (job->vcodec & HB_VCODEC_X265_MASK))) { hb_log( " + fast first pass" ); if (job->vcodec & HB_VCODEC_X264_MASK) { hb_log( " + options: ref=1:8x8dct=0:me=dia:trellis=0" ); hb_log( " analyse=i4x4 (if originally enabled, else analyse=none)" ); hb_log( " subq=2 (if originally greater than 2, else subq unchanged)" ); } } } if (job->color_matrix_code && ((job->vcodec & HB_VCODEC_X264_MASK) || (job->vcodec & HB_VCODEC_X265_MASK))) { // color matrix is set: // 1) at the stream level (x264, x265, qsv only), hb_log(" + custom color matrix: %s", job->color_matrix_code == 1 ? "ITU Bt.601 (NTSC)" : job->color_matrix_code == 2 ? "ITU Bt.601 (PAL)" : job->color_matrix_code == 3 ? "ITU Bt.709 (HD)" : job->color_matrix_code == 4 ? "ITU Bt.2020 (UHD)" : "Custom"); } } if (job->indepth_scan) { hb_log( " * Foreign Audio Search: %s%s%s", job->select_subtitle_config.dest == RENDERSUB ? "Render/Burn-in" : "Passthrough", job->select_subtitle_config.force ? ", Forced Only" : "", job->select_subtitle_config.default_track ? ", Default" : "" ); } for( i = 0; i < hb_list_count( job->list_subtitle ); i++ ) { subtitle = hb_list_item( job->list_subtitle, i ); if( subtitle ) { if( job->indepth_scan ) { hb_log( " + subtitle, %s (track %d, id 0x%x, %s)", subtitle->lang, subtitle->track, subtitle->id, subtitle->format == PICTURESUB ? "Picture" : "Text"); } else if( subtitle->source == SRTSUB ) { /* For SRT, print offset and charset too */ hb_log(" * subtitle track %d, %s (track %d, id 0x%x, Text) -> " "%s%s, offset: %"PRId64", charset: %s", subtitle->out_track, subtitle->lang, subtitle->track, subtitle->id, subtitle->config.dest == RENDERSUB ? "Render/Burn-in" : "Passthrough", subtitle->config.default_track ? ", Default" : "", subtitle->config.offset, subtitle->config.src_codeset); } else { hb_log(" * subtitle track %d, %s (track %d, id 0x%x, %s) -> " "%s%s%s", subtitle->out_track, subtitle->lang, subtitle->track, subtitle->id, subtitle->format == PICTURESUB ? "Picture" : "Text", subtitle->config.dest == RENDERSUB ? "Render/Burn-in" : "Passthrough", subtitle->config.force ? ", Forced Only" : "", subtitle->config.default_track ? ", Default" : "" ); } } } if( !job->indepth_scan ) { for( i = 0; i < hb_list_count( job->list_audio ); i++ ) { audio = hb_list_item( job->list_audio, i ); hb_log( " * audio track %d", audio->config.out.track ); if( audio->config.out.name ) hb_log( " + name: %s", audio->config.out.name ); hb_log( " + decoder: %s (track %d, id 0x%x)", audio->config.lang.description, audio->config.in.track + 1, audio->id ); if (audio->config.in.bitrate >= 1000) hb_log(" + bitrate: %d kbps, samplerate: %d Hz", audio->config.in.bitrate / 1000, audio->config.in.samplerate); else hb_log(" + samplerate: %d Hz", audio->config.in.samplerate); if( audio->config.out.codec & HB_ACODEC_PASS_FLAG ) { hb_log(" + %s", hb_audio_encoder_get_name(audio->config.out.codec)); } else { hb_log(" + mixdown: %s", hb_mixdown_get_name(audio->config.out.mixdown)); if( audio->config.out.normalize_mix_level != 0 ) { hb_log( " + normalized mixing levels" ); } if( audio->config.out.gain != 0.0 ) { hb_log( " + gain: %.fdB", audio->config.out.gain ); } if (audio->config.out.dynamic_range_compression > 0.0f && hb_audio_can_apply_drc(audio->config.in.codec, audio->config.in.codec_param, audio->config.out.codec)) { hb_log( " + dynamic range compression: %f", audio->config.out.dynamic_range_compression ); } if (hb_audio_dither_is_supported(audio->config.out.codec)) { hb_log(" + dither: %s", hb_audio_dither_get_description(audio->config.out.dither_method)); } hb_log(" + encoder: %s", hb_audio_encoder_get_long_name(audio->config.out.codec)); if (audio->config.out.bitrate > 0) { hb_log(" + bitrate: %d kbps, samplerate: %d Hz", audio->config.out.bitrate, audio->config.out.samplerate); } else if (audio->config.out.quality != HB_INVALID_AUDIO_QUALITY) { hb_log(" + quality: %.2f, samplerate: %d Hz", audio->config.out.quality, audio->config.out.samplerate); } else if (audio->config.out.samplerate > 0) { hb_log(" + samplerate: %d Hz", audio->config.out.samplerate); } if (audio->config.out.compression_level >= 0) { hb_log(" + compression level: %.2f", audio->config.out.compression_level); } } } } } /* Corrects framerates when actual duration and frame count numbers are known. */ void correct_framerate( hb_interjob_t * interjob, hb_job_t * job ) { if (interjob->total_time <= 0 || interjob->out_frame_count <= 0 || job->cfr == 1) { // Invalid or uninitialized frame statistics // Or CFR output return; } // compute actual output vrate from first pass int64_t num, den; num = interjob->out_frame_count * 90000LL; den = interjob->total_time; hb_limit_rational64(&num, &den, num, den, INT_MAX); job->vrate.num = num; job->vrate.den = den; den = hb_video_framerate_get_close(&job->vrate, 2.); if (den > 0) { int low, high, clock; hb_video_framerate_get_limits(&low, &high, &clock); job->vrate.num = clock; job->vrate.den = den; } if (ABS(((double)job->orig_vrate.num / job->orig_vrate.den) - ((double) job->vrate.num / job->vrate.den)) > 0.05) { hb_log("work: correcting framerate, %d/%d -> %d/%d", job->orig_vrate.num, job->orig_vrate.den, job->vrate.num, job->vrate.den); } } static void analyze_subtitle_scan( hb_job_t * job ) { hb_subtitle_t *subtitle; int subtitle_highest = 0; int subtitle_lowest = 0; int subtitle_lowest_id = 0; int subtitle_forced_id = 0; int subtitle_forced_hits = 0; int subtitle_hit = 0; int i; // Before closing the title print out our subtitle stats if we need to // find the highest and lowest. for (i = 0; i < hb_list_count(job->list_subtitle); i++) { subtitle = hb_list_item(job->list_subtitle, i); hb_log("Subtitle track %d (id 0x%x) '%s': %d hits (%d forced)", subtitle->track, subtitle->id, subtitle->lang, subtitle->hits, subtitle->forced_hits); if (subtitle->hits == 0) continue; if (subtitle_highest < subtitle->hits) { subtitle_highest = subtitle->hits; } if (subtitle_lowest == 0 || subtitle_lowest > subtitle->hits) { subtitle_lowest = subtitle->hits; subtitle_lowest_id = subtitle->id; } // pick the track with fewest forced hits if (subtitle->forced_hits > 0 && (subtitle_forced_hits == 0 || subtitle_forced_hits > subtitle->forced_hits)) { subtitle_forced_id = subtitle->id; subtitle_forced_hits = subtitle->forced_hits; } } if (subtitle_forced_id && job->select_subtitle_config.force) { // If there is a subtitle stream with forced subtitles and forced-only // is set, then select it in preference to the lowest. subtitle_hit = subtitle_forced_id; hb_log("Found a subtitle candidate with id 0x%x (contains forced subs)", subtitle_hit ); } else if (subtitle_lowest > 0 && subtitle_lowest < subtitle_highest * 0.1) { // OK we have more than one, and the lowest is lower, // but how much lower to qualify for turning it on by // default? // // Let's say 10% as a default. subtitle_hit = subtitle_lowest_id; hb_log( "Found a subtitle candidate with id 0x%x", subtitle_hit ); } else { hb_log( "No candidate detected during subtitle scan" ); } for (i = 0; i < hb_list_count( job->list_subtitle ); i++) { subtitle = hb_list_item( job->list_subtitle, i ); if (subtitle->id == subtitle_hit) { hb_interjob_t *interjob = hb_interjob_get(job->h); subtitle->config = job->select_subtitle_config; // Remove from list since we are taking ownership // of the subtitle. hb_list_rem(job->list_subtitle, subtitle); interjob->select_subtitle = subtitle; break; } } } static int sanitize_subtitles( hb_job_t * job ) { int i; uint8_t one_burned = 0; hb_interjob_t * interjob = hb_interjob_get(job->h); hb_subtitle_t * subtitle; if (job->indepth_scan) { // Subtitles are set by hb_add() during subtitle scan return 0; } /* Look for the scanned subtitle in the existing subtitle list * select_subtitle implies that we did a scan. */ if (interjob->select_subtitle != NULL) { /* Disable forced subtitles if we didn't find any in the scan, so that * we display normal subtitles instead. */ if( interjob->select_subtitle->config.force && interjob->select_subtitle->forced_hits == 0 ) { interjob->select_subtitle->config.force = 0; } for (i = 0; i < hb_list_count( job->list_subtitle );) { subtitle = hb_list_item(job->list_subtitle, i); /* Remove the scanned subtitle from the list if * it would result in: * - an emty track (forced and no forced hits) * - an identical, duplicate subtitle track: * -> both (or neither) are forced * -> subtitle is not forced but all its hits are forced */ if( ( interjob->select_subtitle->id == subtitle->id ) && ( ( subtitle->config.force && interjob->select_subtitle->forced_hits == 0 ) || ( subtitle->config.force == interjob->select_subtitle->config.force ) || ( !subtitle->config.force && interjob->select_subtitle->hits == interjob->select_subtitle->forced_hits ) ) ) { hb_list_rem( job->list_subtitle, subtitle ); free( subtitle ); continue; } /* Adjust output track number, in case we removed one. * Output tracks sadly still need to be in sequential order. * Note: out.track starts at 1, i starts at 0, and track 1 is * interjob->select_subtitle */ subtitle->out_track = ++i + 1; } /* Add the subtitle that we found on the subtitle scan pass. * * Make sure it's the first subtitle in the list so that it becomes the * first burned subtitle (explicitly or after sanitizing) - which should * ensure that it doesn't get dropped. */ interjob->select_subtitle->out_track = 1; if (job->pass_id == HB_PASS_ENCODE || job->pass_id == HB_PASS_ENCODE_2ND) { // final pass, interjob->select_subtitle is no longer needed hb_list_insert(job->list_subtitle, 0, interjob->select_subtitle); interjob->select_subtitle = NULL; } else { // this is not the final pass, so we need to copy it instead hb_list_insert(job->list_subtitle, 0, hb_subtitle_copy(interjob->select_subtitle)); } } for (i = 0; i < hb_list_count(job->list_subtitle);) { subtitle = hb_list_item(job->list_subtitle, i); if (subtitle->config.dest == RENDERSUB) { if (one_burned) { if (!hb_subtitle_can_pass(subtitle->source, job->mux)) { hb_log( "More than one subtitle burn-in requested, dropping track %d.", i ); hb_list_rem(job->list_subtitle, subtitle); free(subtitle); continue; } else { hb_log("More than one subtitle burn-in requested. Changing track %d to soft subtitle.", i); subtitle->config.dest = PASSTHRUSUB; } } else if (!hb_subtitle_can_burn(subtitle->source)) { hb_log("Subtitle burn-in requested and input track can not be rendered. Changing track %d to soft subtitle.", i); subtitle->config.dest = PASSTHRUSUB; } else { one_burned = 1; } } if (subtitle->config.dest == PASSTHRUSUB && !hb_subtitle_can_pass(subtitle->source, job->mux)) { if (!one_burned) { hb_log("Subtitle pass-thru requested and input track is not compatible with container. Changing track %d to burned-in subtitle.", i); subtitle->config.dest = RENDERSUB; subtitle->config.default_track = 0; one_burned = 1; } else { hb_log("Subtitle pass-thru requested and input track is not compatible with container. One track already burned, dropping track %d.", i); hb_list_rem(job->list_subtitle, subtitle); free(subtitle); continue; } } /* Adjust output track number, in case we removed one. * Output tracks sadly still need to be in sequential order. * Note: out.track starts at 1, i starts at 0 */ subtitle->out_track = ++i; } if (one_burned) { // Add subtitle rendering filter // Note that if the filter is already in the filter chain, this // has no effect. Note also that this means the front-end is // not required to add the subtitle rendering filter since // we will always try to do it here. hb_filter_object_t *filter = hb_filter_init(HB_FILTER_RENDER_SUB); hb_add_filter_dict(job, filter, NULL); } return 0; } static int sanitize_audio(hb_job_t *job) { int i; hb_audio_t * audio; if (job->indepth_scan) { // Audio is not processed during subtitle scan return 0; } // apply Auto Passthru settings hb_autopassthru_apply_settings(job); for (i = 0; i < hb_list_count(job->list_audio);) { audio = hb_list_item(job->list_audio, i); if (audio->config.out.codec == HB_ACODEC_AUTO_PASS) { // Auto Passthru should have been handled above // remove track to avoid a crash hb_log("Auto Passthru error, dropping track %d", audio->config.out.track); hb_list_rem(job->list_audio, audio); free(audio); continue; } if ((audio->config.out.codec & HB_ACODEC_PASS_FLAG) && !(audio->config.in.codec & audio->config.out.codec & HB_ACODEC_PASS_MASK)) { hb_log("Passthru requested and input codec is not the same as output codec for track %d, dropping track", audio->config.out.track); hb_list_rem(job->list_audio, audio); free(audio); continue; } /* Adjust output track number, in case we removed one. * Output tracks sadly still need to be in sequential order. * Note: out.track starts at 1, i starts at 0 */ audio->config.out.track = ++i; } int best_mixdown = 0; int best_bitrate = 0; int best_samplerate = 0; for (i = 0; i < hb_list_count(job->list_audio); i++) { audio = hb_list_item(job->list_audio, i); /* Passthru audio */ if (audio->config.out.codec & HB_ACODEC_PASS_FLAG) { // Muxer needs these to be set correctly in order to // set audio track MP4 time base. audio->config.out.samples_per_frame = audio->config.in.samples_per_frame; audio->config.out.samplerate = audio->config.in.samplerate; continue; } /* Vorbis language information */ if (audio->config.out.codec == HB_ACODEC_VORBIS) audio->priv.config.vorbis.language = audio->config.lang.simple; /* sense-check the requested samplerate */ if (audio->config.out.samplerate <= 0) { // if not specified, set to same as input audio->config.out.samplerate = audio->config.in.samplerate; } best_samplerate = hb_audio_samplerate_find_closest(audio->config.out.samplerate, audio->config.out.codec); if (best_samplerate != audio->config.out.samplerate) { hb_log("work: sanitizing track %d unsupported samplerate %d Hz to %s kHz", audio->config.out.track, audio->config.out.samplerate, hb_audio_samplerate_get_name(best_samplerate)); audio->config.out.samplerate = best_samplerate; } /* sense-check the requested mixdown */ if (audio->config.out.mixdown <= HB_AMIXDOWN_NONE) { /* Mixdown not specified, set the default mixdown */ audio->config.out.mixdown = hb_mixdown_get_default(audio->config.out.codec, audio->config.in.channel_layout); hb_log("work: mixdown not specified, track %d setting mixdown %s", audio->config.out.track, hb_mixdown_get_name(audio->config.out.mixdown)); } else { best_mixdown = hb_mixdown_get_best(audio->config.out.codec, audio->config.in.channel_layout, audio->config.out.mixdown); if (audio->config.out.mixdown != best_mixdown) { /* log the output mixdown */ hb_log("work: sanitizing track %d mixdown %s to %s", audio->config.out.track, hb_mixdown_get_name(audio->config.out.mixdown), hb_mixdown_get_name(best_mixdown)); audio->config.out.mixdown = best_mixdown; } } /* sense-check the requested compression level */ if (audio->config.out.compression_level < 0) { audio->config.out.compression_level = hb_audio_compression_get_default(audio->config.out.codec); if (audio->config.out.compression_level >= 0) { hb_log("work: compression level not specified, track %d setting compression level %.2f", audio->config.out.track, audio->config.out.compression_level); } } else { float best_compression = hb_audio_compression_get_best(audio->config.out.codec, audio->config.out.compression_level); if (best_compression != audio->config.out.compression_level) { if (best_compression == -1) { hb_log("work: track %d, compression level not supported by codec", audio->config.out.track); } else { hb_log("work: sanitizing track %d compression level %.2f to %.2f", audio->config.out.track, audio->config.out.compression_level, best_compression); } audio->config.out.compression_level = best_compression; } } /* sense-check the requested quality */ if (audio->config.out.quality != HB_INVALID_AUDIO_QUALITY) { float best_quality = hb_audio_quality_get_best(audio->config.out.codec, audio->config.out.quality); if (best_quality != audio->config.out.quality) { if (best_quality == HB_INVALID_AUDIO_QUALITY) { hb_log("work: track %d, quality mode not supported by codec", audio->config.out.track); } else { hb_log("work: sanitizing track %d quality %.2f to %.2f", audio->config.out.track, audio->config.out.quality, best_quality); } audio->config.out.quality = best_quality; } } /* sense-check the requested bitrate */ if (audio->config.out.quality == HB_INVALID_AUDIO_QUALITY) { if (audio->config.out.bitrate <= 0) { /* Bitrate not specified, set the default bitrate */ audio->config.out.bitrate = hb_audio_bitrate_get_default(audio->config.out.codec, audio->config.out.samplerate, audio->config.out.mixdown); if (audio->config.out.bitrate > 0) { hb_log("work: bitrate not specified, track %d setting bitrate %d Kbps", audio->config.out.track, audio->config.out.bitrate); } } else { best_bitrate = hb_audio_bitrate_get_best(audio->config.out.codec, audio->config.out.bitrate, audio->config.out.samplerate, audio->config.out.mixdown); if (best_bitrate > 0 && best_bitrate != audio->config.out.bitrate) { /* log the output bitrate */ hb_log("work: sanitizing track %d bitrate %d to %d Kbps", audio->config.out.track, audio->config.out.bitrate, best_bitrate); } audio->config.out.bitrate = best_bitrate; } } /* sense-check the requested dither */ if (hb_audio_dither_is_supported(audio->config.out.codec)) { if (audio->config.out.dither_method == hb_audio_dither_get_default()) { /* "auto", enable with default settings */ audio->config.out.dither_method = hb_audio_dither_get_default_method(); } } else if (audio->config.out.dither_method != hb_audio_dither_get_default()) { /* specific dither requested but dithering not supported */ hb_log("work: track %d, dithering not supported by codec", audio->config.out.track); } } return 0; } static int sanitize_qsv( hb_job_t * job ) { #ifdef USE_QSV #if 0 // TODO: re-implement QSV VPP filtering and QSV zerocopy path int i; /* * XXX: mfxCoreInterface's CopyFrame doesn't work in old drivers, and our * workaround is really slow. If we have validated CPU-based filters in * the list and we can't use CopyFrame, disable QSV decoding until a * better solution is implemented. */ if (hb_qsv_copyframe_is_slow(job->vcodec)) { if (job->list_filter != NULL) { int encode_only = 0; for (i = 0; i < hb_list_count(job->list_filter) && !encode_only; i++) { hb_filter_object_t *filter = hb_list_item(job->list_filter, i); switch (filter->id) { // validated, CPU-based filters case HB_FILTER_ROTATE: case HB_FILTER_RENDER_SUB: case HB_FILTER_AVFILTER: encode_only = 1; break; // CPU-based deinterlace (validated) case HB_FILTER_DEINTERLACE: { int mode = hb_dict_get_int(filter->settings, "mode"); if (!(mode & MODE_DEINTERLACE_QSV)) { encode_only = 1; } } break; // other filters will be removed default: break; } } if (encode_only) { hb_log("do_job: QSV: possible CopyFrame bug, using encode-only path"); if (hb_get_cpu_platform() >= HB_CPU_PLATFORM_INTEL_IVB) { hb_log("do_job: QSV: please update your Intel graphics driver to version 9.18.10.3257 or later"); } job->qsv.decode = 0; } } } /* * When QSV's VPP is used for filtering, not all CPU filters * are supported, so we need to do a little extra setup here. */ if (job->vcodec & HB_VCODEC_QSV_MASK) { int vpp_settings[7]; int num_cpu_filters = 0; hb_filter_object_t *filter; // default values for VPP filter vpp_settings[0] = job->title->geometry.width; vpp_settings[1] = job->title->geometry.height; vpp_settings[2] = job->title->crop[0]; vpp_settings[3] = job->title->crop[1]; vpp_settings[4] = job->title->crop[2]; vpp_settings[5] = job->title->crop[3]; vpp_settings[6] = 0; // deinterlace: off if (job->list_filter != NULL && hb_list_count(job->list_filter) > 0) { while (hb_list_count(job->list_filter) > num_cpu_filters) { filter = hb_list_item(job->list_filter, num_cpu_filters); switch (filter->id) { // cropping and scaling always done via VPP filter case HB_FILTER_CROP_SCALE: hb_dict_extract_int(&vpp_settings[0], filter->settings, "width"); hb_dict_extract_int(&vpp_settings[1], filter->settings, "height"); hb_dict_extract_int(&vpp_settings[2], filter->settings, "crop-top"); hb_dict_extract_int(&vpp_settings[3], filter->settings, "crop-bottom"); hb_dict_extract_int(&vpp_settings[4], filter->settings, "crop-left"); hb_dict_extract_int(&vpp_settings[5], filter->settings, "crop-right"); hb_list_rem(job->list_filter, filter); hb_filter_close(&filter); break; // pick VPP or CPU deinterlace depending on settings case HB_FILTER_DEINTERLACE: { int mode = hb_dict_get_int(filter->settings, "mode"); if (mode & MODE_DEINTERLACE_QSV) { // deinterlacing via VPP filter vpp_settings[6] = 1; hb_list_rem(job->list_filter, filter); hb_filter_close(&filter); } else { // validated num_cpu_filters++; } } break; // then, validated filters case HB_FILTER_ROTATE: // TODO: use Media SDK for this case HB_FILTER_RENDER_SUB: case HB_FILTER_AVFILTER: num_cpu_filters++; break; // finally, drop all unsupported filters default: hb_log("do_job: QSV: full path, removing unsupported filter '%s'", filter->name); hb_list_rem(job->list_filter, filter); hb_filter_close(&filter); break; } } if (num_cpu_filters > 0) { // we need filters to copy to system memory and back filter = hb_filter_init(HB_FILTER_QSV_PRE); hb_add_filter_dict(job, filter, NULL); filter = hb_filter_init(HB_FILTER_QSV_POST); hb_add_filter_dict(job, filter, NULL); } if (vpp_settings[0] != job->title->geometry.width || vpp_settings[1] != job->title->geometry.height || vpp_settings[2] >= 1 /* crop */ || vpp_settings[3] >= 1 /* crop */ || vpp_settings[4] >= 1 /* crop */ || vpp_settings[5] >= 1 /* crop */ || vpp_settings[6] >= 1 /* deinterlace */) { // we need the VPP filter hb_dict_t * dict = hb_dict_init(); hb_dict_set(dict, "width", hb_value_int(vpp_settings[0])); hb_dict_set(dict, "height", hb_value_int(vpp_settings[1])); hb_dict_set(dict, "crop-top", hb_value_int(vpp_settings[2])); hb_dict_set(dict, "crop-bottom", hb_value_int(vpp_settings[3])); hb_dict_set(dict, "crop-left", hb_value_int(vpp_settings[4])); hb_dict_set(dict, "crop-right", hb_value_int(vpp_settings[5])); hb_dict_set(dict, "deinterlace", hb_value_int(vpp_settings[6])); filter = hb_filter_init(HB_FILTER_QSV); hb_add_filter_dict(job, filter, dict); hb_value_free(&dict); } } } #endif // QSV VPP filtering and QSV zerocopy path #endif // USE_QSV return 0; } static void sanitize_filter_list(hb_list_t *list) { // Add selective deinterlacing mode if comb detection is enabled if (hb_filter_find(list, HB_FILTER_COMB_DETECT) != NULL) { int selective[] = {HB_FILTER_DECOMB, HB_FILTER_DEINTERLACE}; int ii, count = sizeof(selective) / sizeof(int); for (ii = 0; ii < count; ii++) { hb_filter_object_t * filter = hb_filter_find(list, selective[ii]); if (filter != NULL) { int mode = hb_dict_get_int(filter->settings, "mode"); mode |= MODE_DECOMB_SELECTIVE; hb_dict_set(filter->settings, "mode", hb_value_int(mode)); break; } } } // Combine HB_FILTER_AVFILTERs that are sequential hb_avfilter_combine(list); } /** * Job initialization rountine. * * Initializes fifos. * Creates work objects for synchronizer, video decoder, video renderer, * video decoder, audio decoder, audio encoder, reader, muxer. * Launches thread for each work object with work_loop. * Waits for completion of last work object. * Closes threads and frees fifos. * @param job Handle work hb_job_t. */ static void do_job(hb_job_t *job) { int i, result; hb_title_t * title; hb_interjob_t * interjob; hb_work_object_t * w; hb_audio_t * audio; hb_subtitle_t * subtitle; title = job->title; interjob = hb_interjob_get(job->h); if (job->sequence_id != interjob->sequence_id) { // New job sequence, clear interjob hb_subtitle_close(&interjob->select_subtitle); memset(interjob, 0, sizeof(*interjob)); interjob->sequence_id = job->sequence_id; } job->list_work = hb_list_init(); w = hb_get_work(job->h, WORK_READER); hb_list_add(job->list_work, w); hb_log( "starting job" ); // This must be performed before initializing filters because // it can add the subtitle render filter. result = sanitize_subtitles(job); if (result) { *job->done_error = HB_ERROR_WRONG_INPUT; *job->die = 1; goto cleanup; } // sanitize_qsv looks for subtitle render filter, so must happen after // sanitize_subtitle result = sanitize_qsv(job); if (result) { *job->done_error = HB_ERROR_WRONG_INPUT; *job->die = 1; goto cleanup; } // Filters have an effect on settings. // So initialize the filters and update the job. if (job->list_filter && hb_list_count(job->list_filter)) { hb_filter_init_t init; sanitize_filter_list(job->list_filter); memset(&init, 0, sizeof(init)); init.job = job; init.pix_fmt = AV_PIX_FMT_YUV420P; init.geometry.width = title->geometry.width; init.geometry.height = title->geometry.height; init.geometry.par = job->par; memcpy(init.crop, title->crop, sizeof(int[4])); init.vrate = job->vrate; init.cfr = 0; init.grayscale = 0; for( i = 0; i < hb_list_count( job->list_filter ); ) { hb_filter_object_t * filter = hb_list_item( job->list_filter, i ); filter->done = &job->done; if (filter->init(filter, &init)) { hb_log( "Failure to initialise filter '%s', disabling", filter->name ); hb_list_rem( job->list_filter, filter ); hb_filter_close( &filter ); continue; } i++; } job->width = init.geometry.width; job->height = init.geometry.height; job->par = init.geometry.par; memcpy(job->crop, init.crop, sizeof(int[4])); job->vrate = init.vrate; job->cfr = init.cfr; job->grayscale = init.grayscale; // Perform filter post_init which informs filters of final // job configuration. e.g. rendersub filter needs to know the // final crop dimensions. for( i = 0; i < hb_list_count( job->list_filter ); ) { hb_filter_object_t * filter = hb_list_item( job->list_filter, i ); if (filter->post_init != NULL && filter->post_init(filter, job)) { hb_log( "Failure to initialise filter '%s', disabling", filter->name ); hb_list_rem( job->list_filter, filter ); hb_filter_close( &filter ); continue; } i++; } } else { job->width = title->geometry.width; job->height = title->geometry.height; job->par = title->geometry.par; memset(job->crop, 0, sizeof(int[4])); job->vrate = title->vrate; job->cfr = 0; } job->orig_vrate = job->vrate; if (job->pass_id == HB_PASS_ENCODE_2ND) { correct_framerate(interjob, job); } /* * The frame rate may affect the bitstream's time base, lose superfluous * factors for consistency (some encoders reduce fractions, some don't). */ hb_reduce(&job->orig_vrate.num, &job->orig_vrate.den, job->orig_vrate.num, job->orig_vrate.den); hb_reduce(&job->vrate.num, &job->vrate.den, job->vrate.num, job->vrate.den); #ifdef USE_QSV #if 0 // TODO: re-implement QSV zerocopy path if (hb_qsv_decode_is_enabled(job) && (job->vcodec & HB_VCODEC_QSV_MASK)) { job->fifo_mpeg2 = hb_fifo_init( FIFO_MINI, FIFO_MINI_WAKE ); job->fifo_raw = hb_fifo_init( FIFO_MINI, FIFO_MINI_WAKE ); if (!job->indepth_scan) { // When doing subtitle indepth scan, the pipeline ends at sync job->fifo_sync = hb_fifo_init( FIFO_MINI, FIFO_MINI_WAKE ); job->fifo_render = hb_fifo_init( FIFO_MINI, FIFO_MINI_WAKE ); job->fifo_mpeg4 = hb_fifo_init( FIFO_MINI, FIFO_MINI_WAKE ); } } else #endif // QSV zerocopy path #endif { job->fifo_mpeg2 = hb_fifo_init( FIFO_SMALL, FIFO_SMALL_WAKE ); job->fifo_raw = hb_fifo_init( FIFO_SMALL, FIFO_SMALL_WAKE ); if (!job->indepth_scan) { // When doing subtitle indepth scan, the pipeline ends at sync job->fifo_sync = hb_fifo_init( FIFO_SMALL, FIFO_SMALL_WAKE ); job->fifo_render = NULL; // Attached to filter chain job->fifo_mpeg4 = hb_fifo_init( FIFO_LARGE, FIFO_LARGE_WAKE ); } } result = sanitize_audio(job); if (result) { *job->done_error = HB_ERROR_WRONG_INPUT; *job->die = 1; goto cleanup; } if (!job->indepth_scan) { // Set up audio decoder work objects // Audio fifos must be initialized before sync for (i = 0; i < hb_list_count(job->list_audio); i++) { audio = hb_list_item(job->list_audio, i); /* set up the audio work fifos */ audio->priv.fifo_in = hb_fifo_init(FIFO_LARGE, FIFO_LARGE_WAKE); audio->priv.fifo_raw = hb_fifo_init(FIFO_SMALL, FIFO_SMALL_WAKE); audio->priv.fifo_sync = hb_fifo_init(FIFO_SMALL, FIFO_SMALL_WAKE); audio->priv.fifo_out = hb_fifo_init(FIFO_LARGE, FIFO_LARGE_WAKE); // Add audio decoder work object w = hb_audio_decoder(job->h, audio->config.in.codec); if (w == NULL) { hb_error("Invalid input codec: %d", audio->config.in.codec); *job->done_error = HB_ERROR_WRONG_INPUT; *job->die = 1; goto cleanup; } w->fifo_in = audio->priv.fifo_in; w->fifo_out = audio->priv.fifo_raw; w->config = &audio->priv.config; w->audio = audio; w->codec_param = audio->config.in.codec_param; hb_list_add( job->list_work, w ); } } // Subtitle fifos must be initialized before sync for (i = 0; i < hb_list_count( job->list_subtitle ); i++) { subtitle = hb_list_item( job->list_subtitle, i ); w = hb_get_work( job->h, subtitle->codec ); // Must set capacity of the raw-FIFO to be set >= the maximum // number of subtitle lines that could be decoded prior to a // video frame in order to prevent the following deadlock // condition: // 1. Subtitle decoder blocks trying to generate more subtitle // lines than will fit in the FIFO. // 2. Blocks the processing of further subtitle packets read // from the input stream. // 3. And that blocks the processing of any further video // packets read from the input stream. // 4. And that blocks the sync work-object from running, which // is needed to consume the subtitle lines in the raw-FIFO. // Since that number is unbounded, the FIFO must be made // (effectively) unbounded in capacity. subtitle->fifo_raw = hb_fifo_init( FIFO_UNBOUNDED, FIFO_UNBOUNDED_WAKE ); if (w->id != WORK_DECSRTSUB) { // decsrtsub is a buffer source like reader. It's input comes // from a file. subtitle->fifo_in = hb_fifo_init( FIFO_SMALL, FIFO_SMALL_WAKE ); } if (!job->indepth_scan) { // When doing subtitle indepth scan, the pipeline ends at sync subtitle->fifo_out = hb_fifo_init( FIFO_SMALL, FIFO_SMALL_WAKE ); } w->fifo_in = subtitle->fifo_in; w->fifo_out = subtitle->fifo_raw; w->subtitle = subtitle; hb_list_add( job->list_work, w ); } // Video decoder w = hb_video_decoder(job->h, title->video_codec, title->video_codec_param); if (w == NULL) { *job->done_error = HB_ERROR_WRONG_INPUT; *job->die = 1; goto cleanup; } w->fifo_in = job->fifo_mpeg2; w->fifo_out = job->fifo_raw; hb_list_add(job->list_work, w); // Synchronization w = hb_get_work(job->h, WORK_SYNC_VIDEO); hb_list_add(job->list_work, w); if (!job->indepth_scan) { for( i = 0; i < hb_list_count( job->list_audio ); i++ ) { audio = hb_list_item( job->list_audio, i ); /* * Audio Encoder Thread */ if ( !(audio->config.out.codec & HB_ACODEC_PASS_FLAG ) ) { /* * Add the encoder thread if not doing pass through */ w = hb_audio_encoder( job->h, audio->config.out.codec); if (w == NULL) { hb_error("Invalid audio codec: %#x", audio->config.out.codec); w = NULL; *job->done_error = HB_ERROR_WRONG_INPUT; *job->die = 1; goto cleanup; } w->fifo_in = audio->priv.fifo_sync; w->fifo_out = audio->priv.fifo_out; w->config = &audio->priv.config; w->audio = audio; hb_list_add( job->list_work, w ); } } /* Set up the video filter fifo pipeline */ if ( job->list_filter ) { hb_fifo_t * fifo_in = job->fifo_sync; for (i = 0; i < hb_list_count(job->list_filter); i++) { hb_filter_object_t * filter = hb_list_item(job->list_filter, i); filter->fifo_in = fifo_in; filter->fifo_out = hb_fifo_init( FIFO_MINI, FIFO_MINI_WAKE ); fifo_in = filter->fifo_out; } job->fifo_render = fifo_in; } else if ( !job->list_filter ) { hb_log("work: Internal Error: no filters"); job->fifo_render = NULL; } // Video encoder w = hb_video_encoder(job->h, job->vcodec); if (w == NULL) { *job->done_error = HB_ERROR_INIT; *job->die = 1; goto cleanup; } // Handle case where there are no filters. // This really should never happen. if ( job->fifo_render ) w->fifo_in = job->fifo_render; else w->fifo_in = job->fifo_sync; w->fifo_out = job->fifo_mpeg4; w->config = &job->config; hb_list_add( job->list_work, w ); } // Add Muxer work object // Muxer work object should be the last object added to the list // during regular encoding pass. For subtitle scan, sync is last. if (!job->indepth_scan) { w = hb_get_work(job->h, WORK_MUX); hb_list_add(job->list_work, w); } if( job->chapter_markers && job->chapter_start == job->chapter_end ) { job->chapter_markers = 0; hb_log("work: only 1 chapter, disabling chapter markers"); } /* Display settings */ hb_display_job_info( job ); // Initialize all work objects job->done = 0; for (i = 0; i < hb_list_count( job->list_work ); i++) { w = hb_list_item( job->list_work, i ); w->done = &job->done; if (w->init( w, job )) { hb_error( "Failure to initialise thread '%s'", w->name ); *job->done_error = HB_ERROR_INIT; *job->die = 1; goto cleanup; } } /* Launch processing threads */ for (i = 0; i < hb_list_count( job->list_work ); i++) { w = hb_list_item(job->list_work, i); w->thread = hb_thread_init(w->name, hb_work_loop, w, HB_LOW_PRIORITY); } if (job->list_filter && !job->indepth_scan) { for (i = 0; i < hb_list_count(job->list_filter); i++) { hb_filter_object_t * filter = hb_list_item(job->list_filter, i); // Filters were initialized earlier, so we just need // to start the filter's thread filter->thread = hb_thread_init( filter->name, filter_loop, filter, HB_LOW_PRIORITY ); } } // Wait for the thread of the last work object to complete // Note that other threads may still be running even though the // last thread has exited. So we must be careful with the sequence // of closing threads below. w = hb_list_item(job->list_work, hb_list_count(job->list_work) - 1); w->die = job->die; hb_thread_close(&w->thread); hb_handle_t * h = job->h; hb_state_t state; hb_get_state2( h, &state ); hb_log("work: average encoding speed for job is %f fps", state.param.working.rate_avg); cleanup: job->done = 1; // Close render filter pipeline if (job->list_filter) { for (i = 0; i < hb_list_count(job->list_filter); i++) { hb_filter_object_t * filter = hb_list_item(job->list_filter, i); if( filter->thread != NULL ) { hb_thread_close(&filter->thread); } filter->close(filter); } } // Close work objects // A work thread can use data created by another work thread's init. // So close all work threads before closing thread data. for (i = 0; i < hb_list_count(job->list_work); i++) { w = hb_list_item(job->list_work, i); if (w->thread != NULL) { hb_thread_close(&w->thread); } } while ((w = hb_list_item(job->list_work, 0))) { hb_list_rem(job->list_work, w); w->close(w); free(w); } hb_list_close( &job->list_work ); /* Close fifos */ hb_fifo_close( &job->fifo_mpeg2 ); hb_fifo_close( &job->fifo_raw ); hb_fifo_close( &job->fifo_sync ); hb_fifo_close( &job->fifo_mpeg4 ); for (i = 0; i < hb_list_count( job->list_subtitle ); i++) { subtitle = hb_list_item( job->list_subtitle, i ); if( subtitle ) { hb_fifo_close( &subtitle->fifo_in ); hb_fifo_close( &subtitle->fifo_raw ); hb_fifo_close( &subtitle->fifo_out ); } } for (i = 0; i < hb_list_count( job->list_audio ); i++) { audio = hb_list_item( job->list_audio, i ); if( audio->priv.fifo_in != NULL ) hb_fifo_close( &audio->priv.fifo_in ); if( audio->priv.fifo_raw != NULL ) hb_fifo_close( &audio->priv.fifo_raw ); if( audio->priv.fifo_sync != NULL ) hb_fifo_close( &audio->priv.fifo_sync ); if( audio->priv.fifo_out != NULL ) hb_fifo_close( &audio->priv.fifo_out ); } if (job->list_filter) { for (i = 0; i < hb_list_count( job->list_filter ); i++) { hb_filter_object_t * filter = hb_list_item( job->list_filter, i ); hb_fifo_close( &filter->fifo_out ); } } if (job->indepth_scan) { analyze_subtitle_scan(job); } hb_buffer_pool_free(); hb_job_close(&job); } static inline void copy_chapter( hb_buffer_t * dst, hb_buffer_t * src ) { // Propagate any chapter breaks for the worker if and only if the // output frame has the same time stamp as the input frame (any // worker that delays frames has to propagate the chapter marks itself // and workers that move chapter marks to a different time should set // 'src' to NULL so that this code won't generate spurious duplicates.) if( src && dst && src->s.start == dst->s.start && src->s.new_chap != 0) { // restore log below to debug chapter mark propagation problems dst->s.new_chap = src->s.new_chap; } } /** * Performs the work object's specific work function. * Loops calling work function for associated work object. Sleeps when fifo is full. * Monitors work done indicator. * Exits loop when work indiactor is set. * @param _w Handle to work object. */ void hb_work_loop( void * _w ) { hb_work_object_t * w = _w; hb_buffer_t * buf_in = NULL, * buf_out = NULL; while ((w->die == NULL || !*w->die) && !*w->done && w->status != HB_WORK_DONE) { // fifo_in == NULL means this is a data source (e.g. reader) if (w->fifo_in != NULL) { buf_in = hb_fifo_get_wait( w->fifo_in ); if ( buf_in == NULL ) continue; if ( *w->done ) { if( buf_in ) { hb_buffer_close( &buf_in ); } break; } } // Invalidate buf_out so that if there is no output // we don't try to pass along junk. buf_out = NULL; w->status = w->work( w, &buf_in, &buf_out ); copy_chapter( buf_out, buf_in ); if( buf_in ) { hb_buffer_close( &buf_in ); } if ( buf_out && w->fifo_out == NULL ) { hb_buffer_close( &buf_out ); } if( buf_out ) { while ( !*w->done ) { if ( hb_fifo_full_wait( w->fifo_out ) ) { hb_fifo_push( w->fifo_out, buf_out ); buf_out = NULL; break; } } } else if (w->fifo_in == NULL) { // If this work object is a generator (no input fifo) and it // generated no output, it may be waiting for status from // another thread. Yield so that we don't spin doing nothing. hb_yield(); } } if ( buf_out ) { hb_buffer_close( &buf_out ); } // Consume data in incoming fifo till job completes so that // residual data does not stall the pipeline. There can be // residual data during point-to-point encoding. hb_deep_log(3, "worker %s waiting to die", w->name); while ((w->die == NULL || !*w->die) && !*w->done && w->fifo_in != NULL) { buf_in = hb_fifo_get_wait( w->fifo_in ); if ( buf_in != NULL ) hb_buffer_close( &buf_in ); } } /** * Performs the filter object's specific work function. * Loops calling work function for associated filter object. * Sleeps when fifo is full. * Monitors work done indicator. * Exits loop when work indiactor is set. * @param _w Handle to work object. */ static void filter_loop( void * _f ) { hb_filter_object_t * f = _f; hb_buffer_t * buf_in, * buf_out = NULL; while( !*f->done && f->status != HB_FILTER_DONE ) { buf_in = hb_fifo_get_wait( f->fifo_in ); if ( buf_in == NULL ) continue; // Filters can drop buffers. Remember chapter information // so that it can be propagated to the next buffer if ( buf_in->s.new_chap ) { f->chapter_time = buf_in->s.start; f->chapter_val = buf_in->s.new_chap; // don't let 'filter_loop' put a chapter mark on the wrong buffer buf_in->s.new_chap = 0; } if ( *f->done ) { if( buf_in ) { hb_buffer_close( &buf_in ); } break; } buf_out = NULL; #ifdef USE_QSV hb_buffer_t *last_buf_in = buf_in; #endif f->status = f->work( f, &buf_in, &buf_out ); #ifdef USE_QSV if (f->status == HB_FILTER_DELAY && last_buf_in->qsv_details.filter_details != NULL && buf_out == NULL) { hb_filter_private_t_qsv *qsv_user = buf_in ? buf_in->qsv_details.filter_details : last_buf_in->qsv_details.filter_details ; qsv_user->post.status = f->status; hb_lock(qsv_user->post.frame_completed_lock); qsv_user->post.frame_go = 1; hb_cond_broadcast(qsv_user->post.frame_completed); hb_unlock(qsv_user->post.frame_completed_lock); } #endif if ( buf_out && f->chapter_val && f->chapter_time <= buf_out->s.start ) { buf_out->s.new_chap = f->chapter_val; f->chapter_val = 0; } if( buf_in ) { hb_buffer_close( &buf_in ); } if ( buf_out && f->fifo_out == NULL ) { hb_buffer_close( &buf_out ); } if( buf_out ) { while ( !*f->done ) { if ( hb_fifo_full_wait( f->fifo_out ) ) { hb_fifo_push( f->fifo_out, buf_out ); buf_out = NULL; break; } } } } if ( buf_out ) { hb_buffer_close( &buf_out ); } // Consume data in incoming fifo till job complete so that // residual data does not stall the pipeline while( !*f->done ) { buf_in = hb_fifo_get_wait( f->fifo_in ); if ( buf_in != NULL ) hb_buffer_close( &buf_in ); } }