X-Git-Url: http://git.osdn.jp/view?a=blobdiff_plain;f=libhb%2Fdecavcodec.c;h=f8a06992f80d946ca64de7e1c3a10ce03304b5aa;hb=b06cc6f04d0fcd061593536a03fd6e26d99c8cf8;hp=5320597062408b7a16c17afff04885956741e9ea;hpb=347f9ab6f9e978ea727c5a20afc88ff8147b2129;p=handbrake-jp%2Fhandbrake-jp-git.git diff --git a/libhb/decavcodec.c b/libhb/decavcodec.c index 53205970..f8a06992 100644 --- a/libhb/decavcodec.c +++ b/libhb/decavcodec.c @@ -110,6 +110,10 @@ struct hb_work_private_s struct SwsContext *sws_context; // if we have to rescale or convert color space }; +static void decodeAudio( hb_work_private_t *pv, uint8_t *data, int size ); +static hb_buffer_t *link_buf_list( hb_work_private_t *pv ); + + static int64_t heap_pop( pts_heap_t *heap ) { int64_t result; @@ -183,6 +187,7 @@ static int decavcodecInit( hb_work_object_t * w, hb_job_t * job ) w->private_data = pv; pv->job = job; + pv->list = hb_list_init(); int codec_id = w->codec_param; /*XXX*/ @@ -233,7 +238,7 @@ static void decavcodecClose( hb_work_object_t * w ) } if ( pv->buffer ) { - free( pv->buffer ); + av_free( pv->buffer ); pv->buffer = NULL; } free( pv ); @@ -250,17 +255,12 @@ static int decavcodecWork( hb_work_object_t * w, hb_buffer_t ** buf_in, hb_buffer_t ** buf_out ) { hb_work_private_t * pv = w->private_data; - hb_buffer_t * in = *buf_in, * buf, * last = NULL; - int pos, len, out_size, i, uncompressed_len; - short buffer[AVCODEC_MAX_AUDIO_FRAME_SIZE]; - uint64_t cur; - unsigned char *parser_output_buffer; - int parser_output_buffer_len; + hb_buffer_t * in = *buf_in; - if ( (*buf_in)->size <= 0 ) + if ( in->size <= 0 ) { /* EOF on input stream - send it downstream & say that we're done */ - *buf_out = *buf_in; + *buf_out = in; *buf_in = NULL; return HB_WORK_DONE; } @@ -273,71 +273,40 @@ static int decavcodecWork( hb_work_object_t * w, hb_buffer_t ** buf_in, return HB_WORK_OK; } - cur = ( in->start < 0 )? pv->pts_next : in->start; + // if the packet has a timestamp use it + if ( in->start != -1 ) + { + pv->pts_next = in->start; + } - pos = 0; - while( pos < in->size ) + int pos, len; + for ( pos = 0; pos < in->size; pos += len ) { - len = av_parser_parse( pv->parser, pv->context, - &parser_output_buffer, &parser_output_buffer_len, - in->data + pos, in->size - pos, cur, cur ); - out_size = 0; - uncompressed_len = 0; + uint8_t *parser_output_buffer; + int parser_output_buffer_len; + int64_t cur = pv->pts_next; + + len = av_parser_parse2( pv->parser, pv->context, + &parser_output_buffer, &parser_output_buffer_len, + in->data + pos, in->size - pos, cur, cur, AV_NOPTS_VALUE ); if (parser_output_buffer_len) { - out_size = sizeof(buffer); - uncompressed_len = avcodec_decode_audio2( pv->context, buffer, - &out_size, - parser_output_buffer, - parser_output_buffer_len ); - } - if( out_size ) - { - short * s16; - float * fl32; - - buf = hb_buffer_init( 2 * out_size ); - - int sample_size_in_bytes = 2; // Default to 2 bytes - switch (pv->context->sample_fmt) - { - case SAMPLE_FMT_S16: - sample_size_in_bytes = 2; - break; - /* We should handle other formats here - but that needs additional format conversion work below */ - /* For now we'll just report the error and try to carry on */ - default: - hb_log("decavcodecWork - Unknown Sample Format from avcodec_decode_audio (%d) !", pv->context->sample_fmt); - break; - } - - buf->start = cur; - buf->stop = cur + 90000 * ( out_size / (sample_size_in_bytes * pv->context->channels) ) / - pv->context->sample_rate; - cur = buf->stop; - - s16 = buffer; - fl32 = (float *) buf->data; - for( i = 0; i < out_size / 2; i++ ) - { - fl32[i] = s16[i]; - } - - if( last ) + // set the duration on every frame since the stream format can + // change (it shouldn't but there's no way to guarantee it). + // duration is a scaling factor to go from #bytes in the decoded + // frame to frame time (in 90KHz mpeg ticks). 'channels' converts + // total samples to per-channel samples. 'sample_rate' converts + // per-channel samples to seconds per sample and the 90000 + // is mpeg ticks per second. + if ( pv->context->sample_rate && pv->context->channels ) { - last = last->next = buf; - } - else - { - *buf_out = last = buf; + pv->duration = 90000. / + (double)( pv->context->sample_rate * pv->context->channels ); } + decodeAudio( pv, parser_output_buffer, parser_output_buffer_len ); } - - pos += len; } - - pv->pts_next = cur; - + *buf_out = link_buf_list( pv ); return HB_WORK_OK; } @@ -401,25 +370,25 @@ static int decavcodecBSInfo( hb_work_object_t *w, const hb_buffer_t *buf, AVCodecParserContext *parser = av_parser_init( codec->id ); AVCodecContext *context = avcodec_alloc_context(); hb_avcodec_open( context, codec ); -#ifdef SYS_CYGWIN - uint8_t *buffer = memalign(16, AVCODEC_MAX_AUDIO_FRAME_SIZE); -#else - uint8_t *buffer = malloc( AVCODEC_MAX_AUDIO_FRAME_SIZE ); -#endif + uint8_t *buffer = av_malloc( AVCODEC_MAX_AUDIO_FRAME_SIZE ); int out_size = AVCODEC_MAX_AUDIO_FRAME_SIZE; unsigned char *pbuffer; int pos = 0, pbuffer_size; while ( pos < buf->size ) { - int len = av_parser_parse( parser, context, &pbuffer, &pbuffer_size, - buf->data + pos, buf->size - pos, - buf->start, buf->start ); + int len = av_parser_parse2( parser, context, &pbuffer, &pbuffer_size, + buf->data + pos, buf->size - pos, + buf->start, buf->start, AV_NOPTS_VALUE ); pos += len; if ( pbuffer_size > 0 ) { - len = avcodec_decode_audio2( context, (int16_t*)buffer, &out_size, - pbuffer, pbuffer_size ); + AVPacket avp; + av_init_packet( &avp ); + avp.data = pbuffer; + avp.size = pbuffer_size; + + len = avcodec_decode_audio3( context, (int16_t*)buffer, &out_size, &avp ); if ( len > 0 && context->sample_rate > 0 ) { info->bitrate = context->bit_rate; @@ -431,7 +400,7 @@ static int decavcodecBSInfo( hb_work_object_t *w, const hb_buffer_t *buf, } } } - free( buffer ); + av_free( buffer ); av_parser_close( parser ); hb_avcodec_close( context ); return ret; @@ -521,12 +490,12 @@ static void log_chapter( hb_work_private_t *pv, int chap_num, int64_t pts ) hb_chapter_t *c = hb_list_item( pv->job->title->list_chapter, chap_num - 1 ); if ( c && c->title ) { - hb_log( "%s: \"%s\" (%d) at frame %u time %lld", + hb_log( "%s: \"%s\" (%d) at frame %u time %"PRId64, pv->context->codec->name, c->title, chap_num, pv->nframes, pts ); } else { - hb_log( "%s: Chapter %d at frame %u time %lld", + hb_log( "%s: Chapter %d at frame %u time %"PRId64, pv->context->codec->name, chap_num, pv->nframes, pts ); } } @@ -550,13 +519,18 @@ static int decodeFrame( hb_work_private_t *pv, uint8_t *data, int size ) { int got_picture, oldlevel = 0; AVFrame frame; + AVPacket avp; if ( global_verbosity_level <= 1 ) { oldlevel = av_log_get_level(); av_log_set_level( AV_LOG_QUIET ); } - if ( avcodec_decode_video( pv->context, &frame, &got_picture, data, size ) < 0 ) + + av_init_packet( &avp ); + avp.data = data; + avp.size = size; + if ( avcodec_decode_video2( pv->context, &frame, &got_picture, &avp ) < 0 ) { ++pv->decode_errors; } @@ -589,6 +563,14 @@ static int decodeFrame( hb_work_private_t *pv, uint8_t *data, int size ) { frame_dur += frame.repeat_pict * frame_dur * 0.5; } + // XXX Unlike every other video decoder, the Raw decoder doesn't + // use the standard buffer allocation routines so we never + // get to put a PTS in the frame. Do it now. + if ( pv->context->codec_id == CODEC_ID_RAWVIDEO ) + { + frame.pts = pv->pts; + pv->pts = -1; + } // If there was no pts for this frame, assume constant frame rate // video & estimate the next frame time from the last & duration. double pts = frame.pts; @@ -675,8 +657,8 @@ static void decodeVideo( hb_work_private_t *pv, uint8_t *data, int size, do { uint8_t *pout; int pout_len; - int len = av_parser_parse( pv->parser, pv->context, &pout, &pout_len, - data + pos, size - pos, pts, dts ); + int len = av_parser_parse2( pv->parser, pv->context, &pout, &pout_len, + data + pos, size - pos, pts, dts, AV_NOPTS_VALUE ); pos += len; if ( pout_len > 0 ) @@ -902,13 +884,22 @@ static int decavcodecvInfo( hb_work_object_t *w, hb_work_info_t *info ) info->rate_base *= context->ticks_per_frame; } - /* Sometimes there's no pixel aspect set in the source. In that case, - assume a 1:1 PAR. Otherwise, preserve the source PAR. */ - info->pixel_aspect_width = context->sample_aspect_ratio.num ? - context->sample_aspect_ratio.num : 1; - info->pixel_aspect_height = context->sample_aspect_ratio.den ? - context->sample_aspect_ratio.den : 1; - + info->pixel_aspect_width = context->sample_aspect_ratio.num; + info->pixel_aspect_height = context->sample_aspect_ratio.den; + + /* Sometimes there's no pixel aspect set in the source ffmpeg context + * which appears to come from the video stream. In that case, + * try the pixel aspect in AVStream (which appears to come from + * the container). Else assume a 1:1 PAR. */ + if ( info->pixel_aspect_width == 0 || + info->pixel_aspect_height == 0 ) + { + AVStream *st = hb_ffmpeg_avstream( w->codec_param ); + info->pixel_aspect_width = st->sample_aspect_ratio.num ? + st->sample_aspect_ratio.num : 1; + info->pixel_aspect_height = st->sample_aspect_ratio.den ? + st->sample_aspect_ratio.den : 1; + } /* ffmpeg returns the Pixel Aspect Ratio (PAR). Handbrake wants the * Display Aspect Ratio so we convert by scaling by the Storage * Aspect Ratio (w/h). We do the calc in floating point to get the @@ -1124,26 +1115,17 @@ static void decodeAudio( hb_work_private_t *pv, uint8_t *data, int size ) int16_t *buffer = pv->buffer; if ( buffer == NULL ) { - // XXX ffmpeg bug workaround - // malloc a buffer for the audio decode. On an x86, ffmpeg - // uses mmx/sse instructions on this buffer without checking - // that it's 16 byte aligned and this will cause an abort if - // the buffer is allocated on our stack. Rather than doing - // complicated, machine dependent alignment here we use the - // fact that malloc returns an aligned pointer on most architectures. - - #ifdef SYS_CYGWIN - // Cygwin's malloc doesn't appear to return 16-byte aligned memory so use memalign instead. - pv->buffer = memalign(16, AVCODEC_MAX_AUDIO_FRAME_SIZE); - #else - pv->buffer = malloc( AVCODEC_MAX_AUDIO_FRAME_SIZE ); - #endif - + pv->buffer = av_malloc( AVCODEC_MAX_AUDIO_FRAME_SIZE ); buffer = pv->buffer; } + + AVPacket avp; + av_init_packet( &avp ); + avp.data = data + pos; + avp.size = size - pos; + int out_size = AVCODEC_MAX_AUDIO_FRAME_SIZE; - int len = avcodec_decode_audio2( context, buffer, &out_size, - data + pos, size - pos ); + int len = avcodec_decode_audio3( context, buffer, &out_size, &avp ); if ( len <= 0 ) { return; @@ -1168,7 +1150,7 @@ static void decodeAudio( hb_work_private_t *pv, uint8_t *data, int size ) NULL, 0 ); // get output buffer size (in 2-byte samples) then malloc a buffer out_size = ( out_size * 2 ) / isamp; - buffer = malloc( out_size ); + buffer = av_malloc( out_size ); // we're doing straight sample format conversion which behaves as if // there were only one channel. @@ -1202,7 +1184,7 @@ static void decodeAudio( hb_work_private_t *pv, uint8_t *data, int size ) // if we allocated a buffer for sample format conversion, free it if ( buffer != pv->buffer ) { - free( buffer ); + av_free( buffer ); } } }