diff options
Diffstat (limited to 'libavcodec/h264.c')
| -rw-r--r-- | libavcodec/h264.c | 152 |
1 files changed, 142 insertions, 10 deletions
diff --git a/libavcodec/h264.c b/libavcodec/h264.c index d40ce0d049..dfe9d4a6a5 100644 --- a/libavcodec/h264.c +++ b/libavcodec/h264.c @@ -246,6 +246,141 @@ static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){ return 0; } +static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n, int height, + int y_offset, int list){ + int raw_my= h->mv_cache[list][ scan8[n] ][1]; + int filter_height= (raw_my&3) ? 2 : 0; + int full_my= (raw_my>>2) + y_offset; + int top = full_my - filter_height, bottom = full_my + height + filter_height; + + return FFMAX(abs(top), bottom); +} + +static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n, int height, + int y_offset, int list0, int list1, int *nrefs){ + MpegEncContext * const s = &h->s; + int my; + + y_offset += 16*(s->mb_y >> MB_FIELD); + + if(list0){ + int ref_n = h->ref_cache[0][ scan8[n] ]; + Picture *ref= &h->ref_list[0][ref_n]; + + // Error resilience puts the current picture in the ref list. + // Don't try to wait on these as it will cause a deadlock. + // Fields can wait on each other, though. + if(ref->thread_opaque != s->current_picture.thread_opaque || + (ref->reference&3) != s->picture_structure) { + my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0); + if (refs[0][ref_n] < 0) nrefs[0] += 1; + refs[0][ref_n] = FFMAX(refs[0][ref_n], my); + } + } + + if(list1){ + int ref_n = h->ref_cache[1][ scan8[n] ]; + Picture *ref= &h->ref_list[1][ref_n]; + + if(ref->thread_opaque != s->current_picture.thread_opaque || + (ref->reference&3) != s->picture_structure) { + my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1); + if (refs[1][ref_n] < 0) nrefs[1] += 1; + refs[1][ref_n] = FFMAX(refs[1][ref_n], my); + } + } +} + +/** + * Wait until all reference frames are available for MC operations. + * + * @param h the H264 context + */ +static void await_references(H264Context *h){ + MpegEncContext * const s = &h->s; + const int mb_xy= h->mb_xy; + const int mb_type= s->current_picture.mb_type[mb_xy]; + int refs[2][48]; + int nrefs[2] = {0}; + int ref, list; + + memset(refs, -1, sizeof(refs)); + + if(IS_16X16(mb_type)){ + get_lowest_part_y(h, refs, 0, 16, 0, + IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs); + }else if(IS_16X8(mb_type)){ + get_lowest_part_y(h, refs, 0, 8, 0, + IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs); + get_lowest_part_y(h, refs, 8, 8, 8, + IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs); + }else if(IS_8X16(mb_type)){ + get_lowest_part_y(h, refs, 0, 16, 0, + IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs); + get_lowest_part_y(h, refs, 4, 16, 0, + IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs); + }else{ + int i; + + assert(IS_8X8(mb_type)); + + for(i=0; i<4; i++){ + const int sub_mb_type= h->sub_mb_type[i]; + const int n= 4*i; + int y_offset= (i&2)<<2; + + if(IS_SUB_8X8(sub_mb_type)){ + get_lowest_part_y(h, refs, n , 8, y_offset, + IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs); + }else if(IS_SUB_8X4(sub_mb_type)){ + get_lowest_part_y(h, refs, n , 4, y_offset, + IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs); + get_lowest_part_y(h, refs, n+2, 4, y_offset+4, + IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs); + }else if(IS_SUB_4X8(sub_mb_type)){ + get_lowest_part_y(h, refs, n , 8, y_offset, + IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs); + get_lowest_part_y(h, refs, n+1, 8, y_offset, + IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs); + }else{ + int j; + assert(IS_SUB_4X4(sub_mb_type)); + for(j=0; j<4; j++){ + int sub_y_offset= y_offset + 2*(j&2); + get_lowest_part_y(h, refs, n+j, 4, sub_y_offset, + IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs); + } + } + } + } + + for(list=h->list_count-1; list>=0; list--){ + for(ref=0; ref<48 && nrefs[list]; ref++){ + int row = refs[list][ref]; + if(row >= 0){ + Picture *ref_pic = &h->ref_list[list][ref]; + int ref_field = ref_pic->reference - 1; + int ref_field_picture = ref_pic->field_picture; + int pic_height = 16*s->mb_height >> ref_field_picture; + + row <<= MB_MBAFF; + nrefs[list]--; + + if(!FIELD_PICTURE && ref_field_picture){ // frame referencing two fields + ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) - !(row&1), pic_height-1), 1); + ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) , pic_height-1), 0); + }else if(FIELD_PICTURE && !ref_field_picture){ // field referencing one field of a frame + ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row*2 + ref_field , pic_height-1), 0); + }else if(FIELD_PICTURE){ + ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), ref_field); + }else{ + ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), 0); + } + } + } + } +} + #if 0 /** * DCT transforms the 16 dc values. @@ -315,6 +450,7 @@ static void chroma_dc_dct_c(DCTELEM *block){ static void free_tables(H264Context *h, int free_rbsp){ int i; H264Context *hx; + av_freep(&h->intra4x4_pred_mode); av_freep(&h->chroma_pred_mode_table); av_freep(&h->cbp_table); @@ -611,6 +747,7 @@ av_cold int ff_h264_decode_init(AVCodecContext *avctx){ return 0; } + #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b)+(size)))) static void copy_picture_range(Picture **to, Picture **from, int count, MpegEncContext *new_base, MpegEncContext *old_base) { @@ -711,7 +848,8 @@ static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContex copy_fields(h, h1, poc_lsb, redundant_pic_count); //reference lists - copy_fields(h, h1, ref_count, intra_gb); + copy_fields(h, h1, ref_count, list_count); + copy_fields(h, h1, ref_list, intra_gb); copy_fields(h, h1, short_ref, cabac_init_idc); copy_picture_range(h->short_ref, h1->short_ref, 32, s, s1); @@ -738,6 +876,7 @@ int ff_h264_frame_start(H264Context *h){ MpegEncContext * const s = &h->s; int i; const int pixel_shift = h->pixel_shift; + int thread_count = (s->avctx->active_thread_type & FF_THREAD_SLICE) ? s->avctx->thread_count : 1; if(MPV_frame_start(s, s->avctx) < 0) return -1; @@ -766,7 +905,7 @@ int ff_h264_frame_start(H264Context *h){ /* can't be in alloc_tables because linesize isn't known there. * FIXME: redo bipred weight to not require extra buffer? */ - for(i = 0; i < s->avctx->thread_count; i++) + for(i = 0; i < thread_count; i++) if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad) h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize); @@ -2910,12 +3049,6 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){ int nal_index; h->max_contexts = (HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_SLICE)) ? avctx->thread_count : 1; -#if 0 - int i; - for(i=0; i<50; i++){ - av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]); - } -#endif if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){ h->current_slice = 0; if (!s->first_field) @@ -3491,8 +3624,7 @@ AVCodec ff_h264_decoder = { ff_h264_decode_end, decode_frame, /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY | - CODEC_CAP_FRAME_THREADS | - CODEC_CAP_SLICE_THREADS, + CODEC_CAP_SLICE_THREADS | CODEC_CAP_FRAME_THREADS, .flush= flush_dpb, .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"), .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy), |