diff --git a/libavcodec/h264.h b/libavcodec/h264.h index 203a04bd1d..007ed3cc57 100644 --- a/libavcodec/h264.h +++ b/libavcodec/h264.h @@ -320,6 +320,19 @@ typedef struct H264SliceContext { int chroma_pred_mode; int intra16x16_pred_mode; + + int topleft_mb_xy; + int top_mb_xy; + int topright_mb_xy; + int left_mb_xy[LEFT_MBS]; + + int topleft_type; + int top_type; + int topright_type; + int left_type[LEFT_MBS]; + + const uint8_t *left_block; + int topleft_partition; } H264SliceContext; /** @@ -357,19 +370,6 @@ typedef struct H264Context { int workaround_bugs; // prediction stuff - int topleft_mb_xy; - int top_mb_xy; - int topright_mb_xy; - int left_mb_xy[LEFT_MBS]; - - int topleft_type; - int top_type; - int topright_type; - int left_type[LEFT_MBS]; - - const uint8_t *left_block; - int topleft_partition; - int8_t intra4x4_pred_mode_cache[5 * 8]; int8_t(*intra4x4_pred_mode); H264PredContext hpc; diff --git a/libavcodec/h264_cabac.c b/libavcodec/h264_cabac.c index 30b97836f0..505d1c4e20 100644 --- a/libavcodec/h264_cabac.c +++ b/libavcodec/h264_cabac.c @@ -1293,15 +1293,17 @@ static int decode_cabac_field_decoding_flag(H264Context *h) { return get_cabac_noinline( &h->cabac, &(h->cabac_state+70)[ctx] ); } -static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) { +static int decode_cabac_intra_mb_type(H264Context *h, H264SliceContext *sl, + int ctx_base, int intra_slice) +{ uint8_t *state= &h->cabac_state[ctx_base]; int mb_type; if(intra_slice){ int ctx=0; - if( h->left_type[LTOP] & (MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)) + if (sl->left_type[LTOP] & (MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)) ctx++; - if( h->top_type & (MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)) + if (sl->top_type & (MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)) ctx++; if( get_cabac_noinline( &h->cabac, &state[ctx] ) == 0 ) return 0; /* I4x4 */ @@ -1371,17 +1373,18 @@ static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) { return mode + ( mode >= pred_mode ); } -static int decode_cabac_mb_chroma_pre_mode( H264Context *h) { - const int mba_xy = h->left_mb_xy[0]; - const int mbb_xy = h->top_mb_xy; +static int decode_cabac_mb_chroma_pre_mode(H264Context *h, H264SliceContext *sl) +{ + const int mba_xy = sl->left_mb_xy[0]; + const int mbb_xy = sl->top_mb_xy; int ctx = 0; /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */ - if( h->left_type[LTOP] && h->chroma_pred_mode_table[mba_xy] != 0 ) + if (sl->left_type[LTOP] && h->chroma_pred_mode_table[mba_xy] != 0) ctx++; - if( h->top_type && h->chroma_pred_mode_table[mbb_xy] != 0 ) + if (sl->top_type && h->chroma_pred_mode_table[mbb_xy] != 0) ctx++; if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+ctx] ) == 0 ) @@ -1928,15 +1931,15 @@ int ff_h264_decode_mb_cabac(H264Context *h, H264SliceContext *sl) sl->prev_mb_skipped = 0; - fill_decode_neighbors(h, -(MB_FIELD(h))); + fill_decode_neighbors(h, sl, -(MB_FIELD(h))); if( h->slice_type_nos == AV_PICTURE_TYPE_B ) { int ctx = 0; assert(h->slice_type_nos == AV_PICTURE_TYPE_B); - if( !IS_DIRECT( h->left_type[LTOP]-1 ) ) + if (!IS_DIRECT(sl->left_type[LTOP] - 1)) ctx++; - if( !IS_DIRECT( h->top_type-1 ) ) + if (!IS_DIRECT(sl->top_type - 1)) ctx++; if( !get_cabac_noinline( &h->cabac, &h->cabac_state[27+ctx] ) ){ @@ -1952,7 +1955,7 @@ int ff_h264_decode_mb_cabac(H264Context *h, H264SliceContext *sl) if( bits < 8 ){ mb_type= bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */ }else if( bits == 13 ){ - mb_type= decode_cabac_intra_mb_type(h, 32, 0); + mb_type = decode_cabac_intra_mb_type(h, sl, 32, 0); goto decode_intra_mb; }else if( bits == 14 ){ mb_type= 11; /* B_L1_L0_8x16 */ @@ -1978,11 +1981,11 @@ int ff_h264_decode_mb_cabac(H264Context *h, H264SliceContext *sl) partition_count= p_mb_type_info[mb_type].partition_count; mb_type= p_mb_type_info[mb_type].type; } else { - mb_type= decode_cabac_intra_mb_type(h, 17, 0); + mb_type = decode_cabac_intra_mb_type(h, sl, 17, 0); goto decode_intra_mb; } } else { - mb_type= decode_cabac_intra_mb_type(h, 3, 1); + mb_type = decode_cabac_intra_mb_type(h, sl, 3, 1); if(h->slice_type == AV_PICTURE_TYPE_SI && mb_type) mb_type--; assert(h->slice_type_nos == AV_PICTURE_TYPE_I); @@ -2031,7 +2034,7 @@ decode_intra_mb: return 0; } - fill_decode_caches(h, mb_type); + fill_decode_caches(h, sl, mb_type); if( IS_INTRA( mb_type ) ) { int i, pred_mode; @@ -2060,7 +2063,7 @@ decode_intra_mb: } if(decode_chroma){ h->chroma_pred_mode_table[mb_xy] = - pred_mode = decode_cabac_mb_chroma_pre_mode( h ); + pred_mode = decode_cabac_mb_chroma_pre_mode(h, sl); pred_mode= ff_h264_check_intra_pred_mode( h, pred_mode, 1 ); if( pred_mode < 0 ) return -1; @@ -2136,7 +2139,7 @@ decode_intra_mb: const int index= 4*i + block_width*j; int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ]; uint8_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ]; - pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my); + pred_motion(h, sl, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my); DECODE_CABAC_MB_MVD( h, list, index) tprintf(h->avctx, "final mv:%d %d\n", mx, my); @@ -2200,7 +2203,7 @@ decode_intra_mb: for(list=0; listlist_count; list++){ if(IS_DIR(mb_type, 0, list)){ int mx,my,mpx,mpy; - pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my); + pred_motion(h, sl, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my); DECODE_CABAC_MB_MVD( h, list, 0) tprintf(h->avctx, "final mv:%d %d\n", mx, my); @@ -2231,7 +2234,7 @@ decode_intra_mb: for(i=0; i<2; i++){ if(IS_DIR(mb_type, i, list)){ int mx,my,mpx,mpy; - pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my); + pred_16x8_motion(h, sl, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my); DECODE_CABAC_MB_MVD( h, list, 8*i) tprintf(h->avctx, "final mv:%d %d\n", mx, my); @@ -2266,7 +2269,7 @@ decode_intra_mb: for(i=0; i<2; i++){ if(IS_DIR(mb_type, i, list)){ int mx,my,mpx,mpy; - pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my); + pred_8x16_motion(h, sl, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my); DECODE_CABAC_MB_MVD( h, list, 4*i) tprintf(h->avctx, "final mv:%d %d\n", mx, my); @@ -2304,7 +2307,7 @@ decode_intra_mb: int i; uint8_t *nnz_cache = h->non_zero_count_cache; for (i = 0; i < 2; i++){ - if (h->left_type[LEFT(i)] && !IS_8x8DCT(h->left_type[LEFT(i)])){ + if (sl->left_type[LEFT(i)] && !IS_8x8DCT(sl->left_type[LEFT(i)])) { nnz_cache[3+8* 1 + 2*8*i]= nnz_cache[3+8* 2 + 2*8*i]= nnz_cache[3+8* 6 + 2*8*i]= @@ -2313,7 +2316,7 @@ decode_intra_mb: nnz_cache[3+8*12 + 2*8*i]= IS_INTRA(mb_type) ? 64 : 0; } } - if (h->top_type && !IS_8x8DCT(h->top_type)){ + if (sl->top_type && !IS_8x8DCT(sl->top_type)){ uint32_t top_empty = CABAC(h) && !IS_INTRA(mb_type) ? 0 : 0x40404040; AV_WN32A(&nnz_cache[4+8* 0], top_empty); AV_WN32A(&nnz_cache[4+8* 5], top_empty); diff --git a/libavcodec/h264_cavlc.c b/libavcodec/h264_cavlc.c index 7244654a30..b6fe77b6ba 100644 --- a/libavcodec/h264_cavlc.c +++ b/libavcodec/h264_cavlc.c @@ -781,8 +781,8 @@ decode_intra_mb: return 0; } - fill_decode_neighbors(h, mb_type); - fill_decode_caches(h, mb_type); + fill_decode_neighbors(h, sl, mb_type); + fill_decode_caches(h, sl, mb_type); //mb_pred if(IS_INTRA(mb_type)){ @@ -903,7 +903,7 @@ decode_intra_mb: int mx, my; const int index= 4*i + block_width*j; int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ]; - pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my); + pred_motion(h, sl, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my); mx += get_se_golomb(&h->gb); my += get_se_golomb(&h->gb); tprintf(h->avctx, "final mv:%d %d\n", mx, my); @@ -957,7 +957,7 @@ decode_intra_mb: } for(list=0; listlist_count; list++){ if(IS_DIR(mb_type, 0, list)){ - pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my); + pred_motion(h, sl, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my); mx += get_se_golomb(&h->gb); my += get_se_golomb(&h->gb); tprintf(h->avctx, "final mv:%d %d\n", mx, my); @@ -992,7 +992,7 @@ decode_intra_mb: for(i=0; i<2; i++){ unsigned int val; if(IS_DIR(mb_type, i, list)){ - pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my); + pred_16x8_motion(h, sl, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my); mx += get_se_golomb(&h->gb); my += get_se_golomb(&h->gb); tprintf(h->avctx, "final mv:%d %d\n", mx, my); @@ -1030,7 +1030,7 @@ decode_intra_mb: for(i=0; i<2; i++){ unsigned int val; if(IS_DIR(mb_type, i, list)){ - pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my); + pred_8x16_motion(h, sl, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my); mx += get_se_golomb(&h->gb); my += get_se_golomb(&h->gb); tprintf(h->avctx, "final mv:%d %d\n", mx, my); diff --git a/libavcodec/h264_loopfilter.c b/libavcodec/h264_loopfilter.c index f6457286a1..a6a0b927f0 100644 --- a/libavcodec/h264_loopfilter.c +++ b/libavcodec/h264_loopfilter.c @@ -235,6 +235,7 @@ static av_always_inline void filter_mb_edgech(uint8_t *pix, int stride, } static av_always_inline void h264_filter_mb_fast_internal(H264Context *h, + H264SliceContext *sl, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, @@ -248,8 +249,8 @@ static av_always_inline void h264_filter_mb_fast_internal(H264Context *h, int chroma422 = CHROMA422(h); int mb_xy = h->mb_xy; - int left_type= h->left_type[LTOP]; - int top_type= h->top_type; + int left_type = sl->left_type[LTOP]; + int top_type = sl->top_type; int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8); int a = 52 + h->slice_alpha_c0_offset - qp_bd_offset; @@ -258,7 +259,7 @@ static av_always_inline void h264_filter_mb_fast_internal(H264Context *h, int mb_type = h->cur_pic.mb_type[mb_xy]; int qp = h->cur_pic.qscale_table[mb_xy]; int qp0 = h->cur_pic.qscale_table[mb_xy - 1]; - int qp1 = h->cur_pic.qscale_table[h->top_mb_xy]; + int qp1 = h->cur_pic.qscale_table[sl->top_mb_xy]; int qpc = get_chroma_qp( h, 0, qp ); int qpc0 = get_chroma_qp( h, 0, qp0 ); int qpc1 = get_chroma_qp( h, 0, qp1 ); @@ -427,12 +428,12 @@ void ff_h264_filter_mb_fast(H264Context *h, H264SliceContext *sl, } #if CONFIG_SMALL - h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, h->pixel_shift); + h264_filter_mb_fast_internal(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, h->pixel_shift); #else if(h->pixel_shift){ - h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 1); + h264_filter_mb_fast_internal(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 1); }else{ - h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 0); + h264_filter_mb_fast_internal(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 0); } #endif } @@ -478,8 +479,8 @@ static av_always_inline void filter_mb_dir(H264Context *h, H264SliceContext *sl, int chroma_qp_avg[2]; int chroma444 = CHROMA444(h); int chroma422 = CHROMA422(h); - const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy; - const int mbm_type = dir == 0 ? h->left_type[LTOP] : h->top_type; + const int mbm_xy = dir == 0 ? mb_xy -1 : sl->top_mb_xy; + const int mbm_type = dir == 0 ? sl->left_type[LTOP] : sl->top_type; // how often to recheck mv-based bS when iterating between edges static const uint8_t mask_edge_tab[2][8]={{0,3,3,3,1,1,1,1}, @@ -726,9 +727,9 @@ void ff_h264_filter_mb(H264Context *h, H264SliceContext *sl, if (FRAME_MBAFF(h) // and current and left pair do not have the same interlaced type - && IS_INTERLACED(mb_type^h->left_type[LTOP]) + && IS_INTERLACED(mb_type ^ sl->left_type[LTOP]) // and left mb is in available to us - && h->left_type[LTOP]) { + && sl->left_type[LTOP]) { /* First vertical edge is different in MBAFF frames * There are 8 different bS to compute and 2 different Qp */ @@ -756,8 +757,8 @@ void ff_h264_filter_mb(H264Context *h, H264SliceContext *sl, const uint8_t *off= offset[MB_FIELD(h)][mb_y&1]; for( i = 0; i < 8; i++ ) { int j= MB_FIELD(h) ? i>>2 : i&1; - int mbn_xy = h->left_mb_xy[LEFT(j)]; - int mbn_type= h->left_type[LEFT(j)]; + int mbn_xy = sl->left_mb_xy[LEFT(j)]; + int mbn_type = sl->left_type[LEFT(j)]; if( IS_INTRA( mbn_type ) ) bS[i] = 4; @@ -772,8 +773,8 @@ void ff_h264_filter_mb(H264Context *h, H264SliceContext *sl, } mb_qp = h->cur_pic.qscale_table[mb_xy]; - mbn0_qp = h->cur_pic.qscale_table[h->left_mb_xy[0]]; - mbn1_qp = h->cur_pic.qscale_table[h->left_mb_xy[1]]; + mbn0_qp = h->cur_pic.qscale_table[sl->left_mb_xy[0]]; + mbn1_qp = h->cur_pic.qscale_table[sl->left_mb_xy[1]]; qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1; bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) + get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1; diff --git a/libavcodec/h264_mb.c b/libavcodec/h264_mb.c index e26f5c9f70..1515bc800d 100644 --- a/libavcodec/h264_mb.c +++ b/libavcodec/h264_mb.c @@ -496,7 +496,8 @@ static av_always_inline void prefetch_motion(H264Context *h, int list, } } -static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y, +static av_always_inline void xchg_mb_border(H264Context *h, H264SliceContext *sl, + uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg, int chroma444, @@ -519,7 +520,7 @@ static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y, if (h->deblocking_filter == 2) { deblock_topleft = h->slice_table[h->mb_xy - 1 - h->mb_stride] == h->slice_num; - deblock_top = h->top_type; + deblock_top = sl->top_type; } else { deblock_topleft = (h->mb_x > 0); deblock_top = (h->mb_y > !!MB_FIELD(h)); diff --git a/libavcodec/h264_mb_template.c b/libavcodec/h264_mb_template.c index eb34533d15..4dd543a5a2 100644 --- a/libavcodec/h264_mb_template.c +++ b/libavcodec/h264_mb_template.c @@ -159,7 +159,7 @@ static av_noinline void FUNC(hl_decode_mb)(H264Context *h, H264SliceContext *sl) } else { if (IS_INTRA(mb_type)) { if (h->deblocking_filter) - xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, + xchg_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, 0, SIMPLE, PIXEL_SHIFT); if (SIMPLE || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) { @@ -172,7 +172,7 @@ static av_noinline void FUNC(hl_decode_mb)(H264Context *h, H264SliceContext *sl) block_offset, linesize, dest_y, 0); if (h->deblocking_filter) - xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, + xchg_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, SIMPLE, PIXEL_SHIFT); } else if (is_h264) { if (chroma422) { @@ -343,7 +343,7 @@ static av_noinline void FUNC(hl_decode_mb_444)(H264Context *h, H264SliceContext } else { if (IS_INTRA(mb_type)) { if (h->deblocking_filter) - xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, + xchg_mb_border(h, sl, dest[0], dest[1], dest[2], linesize, linesize, 1, 1, SIMPLE, PIXEL_SHIFT); for (p = 0; p < plane_count; p++) @@ -352,7 +352,7 @@ static av_noinline void FUNC(hl_decode_mb_444)(H264Context *h, H264SliceContext block_offset, linesize, dest[p], p); if (h->deblocking_filter) - xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, + xchg_mb_border(h, sl, dest[0], dest[1], dest[2], linesize, linesize, 0, 1, SIMPLE, PIXEL_SHIFT); } else { FUNC(hl_motion_444)(h, sl, dest[0], dest[1], dest[2], diff --git a/libavcodec/h264_mvpred.h b/libavcodec/h264_mvpred.h index 8f9932d974..42d5e3d076 100644 --- a/libavcodec/h264_mvpred.h +++ b/libavcodec/h264_mvpred.h @@ -35,7 +35,8 @@ #include -static av_always_inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, +static av_always_inline int fetch_diagonal_mv(H264Context *h, H264SliceContext *sl, + const int16_t **C, int i, int list, int part_width) { const int topright_ref = h->ref_cache[list][i - 8 + part_width]; @@ -61,13 +62,13 @@ static av_always_inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, AV_ZERO32(h->mv_cache[list][scan8[0] - 2]); *C = h->mv_cache[list][scan8[0] - 2]; - if (!MB_FIELD(h) && IS_INTERLACED(h->left_type[0])) { - SET_DIAG_MV(* 2, >> 1, h->left_mb_xy[0] + h->mb_stride, + if (!MB_FIELD(h) && IS_INTERLACED(sl->left_type[0])) { + SET_DIAG_MV(* 2, >> 1, sl->left_mb_xy[0] + h->mb_stride, (h->mb_y & 1) * 2 + (i >> 5)); } - if (MB_FIELD(h) && !IS_INTERLACED(h->left_type[0])) { + if (MB_FIELD(h) && !IS_INTERLACED(sl->left_type[0])) { // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK. - SET_DIAG_MV(/ 2, << 1, h->left_mb_xy[i >= 36], ((i >> 2)) & 3); + SET_DIAG_MV(/ 2, << 1, sl->left_mb_xy[i >= 36], ((i >> 2)) & 3); } } #undef SET_DIAG_MV @@ -91,7 +92,9 @@ static av_always_inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, * @param mx the x component of the predicted motion vector * @param my the y component of the predicted motion vector */ -static av_always_inline void pred_motion(H264Context *const h, int n, +static av_always_inline void pred_motion(H264Context *const h, + H264SliceContext *sl, + int n, int part_width, int list, int ref, int *const mx, int *const my) { @@ -113,7 +116,7 @@ static av_always_inline void pred_motion(H264Context *const h, int n, * . . . L . . . . */ - diagonal_ref = fetch_diagonal_mv(h, &C, index8, list, part_width); + diagonal_ref = fetch_diagonal_mv(h, sl, &C, index8, list, part_width); match_count = (diagonal_ref == ref) + (top_ref == ref) + (left_ref == ref); tprintf(h->avctx, "pred_motion match_count=%d\n", match_count); if (match_count > 1) { //most common @@ -155,6 +158,7 @@ static av_always_inline void pred_motion(H264Context *const h, int n, * @param my the y component of the predicted motion vector */ static av_always_inline void pred_16x8_motion(H264Context *const h, + H264SliceContext *sl, int n, int list, int ref, int *const mx, int *const my) { @@ -185,7 +189,7 @@ static av_always_inline void pred_16x8_motion(H264Context *const h, } //RARE - pred_motion(h, n, 4, list, ref, mx, my); + pred_motion(h, sl, n, 4, list, ref, mx, my); } /** @@ -195,6 +199,7 @@ static av_always_inline void pred_16x8_motion(H264Context *const h, * @param my the y component of the predicted motion vector */ static av_always_inline void pred_8x16_motion(H264Context *const h, + H264SliceContext *sl, int n, int list, int ref, int *const mx, int *const my) { @@ -214,7 +219,7 @@ static av_always_inline void pred_8x16_motion(H264Context *const h, const int16_t *C; int diagonal_ref; - diagonal_ref = fetch_diagonal_mv(h, &C, scan8[4], list, 2); + diagonal_ref = fetch_diagonal_mv(h, sl, &C, scan8[4], list, 2); tprintf(h->avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->mb_x, h->mb_y, n, list); @@ -227,7 +232,7 @@ static av_always_inline void pred_8x16_motion(H264Context *const h, } //RARE - pred_motion(h, n, 2, list, ref, mx, my); + pred_motion(h, sl, n, 2, list, ref, mx, my); } #define FIX_MV_MBAFF(type, refn, mvn, idx) \ @@ -249,7 +254,8 @@ static av_always_inline void pred_8x16_motion(H264Context *const h, } \ } -static av_always_inline void pred_pskip_motion(H264Context *const h) +static av_always_inline void pred_pskip_motion(H264Context *const h, + H264SliceContext *sl) { DECLARE_ALIGNED(4, static const int16_t, zeromv)[2] = { 0 }; DECLARE_ALIGNED(4, int16_t, mvbuf)[3][2]; @@ -266,26 +272,26 @@ static av_always_inline void pred_pskip_motion(H264Context *const h) * FIXME: this is a partial duplicate of the logic in fill_decode_caches, * but it's faster this way. Is there a way to avoid this duplication? */ - if (USES_LIST(h->left_type[LTOP], 0)) { - left_ref = ref[4 * h->left_mb_xy[LTOP] + 1 + (h->left_block[0] & ~1)]; - A = mv[h->mb2b_xy[h->left_mb_xy[LTOP]] + 3 + b_stride * h->left_block[0]]; - FIX_MV_MBAFF(h->left_type[LTOP], left_ref, A, 0); + if (USES_LIST(sl->left_type[LTOP], 0)) { + left_ref = ref[4 * sl->left_mb_xy[LTOP] + 1 + (sl->left_block[0] & ~1)]; + A = mv[h->mb2b_xy[sl->left_mb_xy[LTOP]] + 3 + b_stride * sl->left_block[0]]; + FIX_MV_MBAFF(sl->left_type[LTOP], left_ref, A, 0); if (!(left_ref | AV_RN32A(A))) goto zeromv; - } else if (h->left_type[LTOP]) { + } else if (sl->left_type[LTOP]) { left_ref = LIST_NOT_USED; A = zeromv; } else { goto zeromv; } - if (USES_LIST(h->top_type, 0)) { - top_ref = ref[4 * h->top_mb_xy + 2]; - B = mv[h->mb2b_xy[h->top_mb_xy] + 3 * b_stride]; - FIX_MV_MBAFF(h->top_type, top_ref, B, 1); + if (USES_LIST(sl->top_type, 0)) { + top_ref = ref[4 * sl->top_mb_xy + 2]; + B = mv[h->mb2b_xy[sl->top_mb_xy] + 3 * b_stride]; + FIX_MV_MBAFF(sl->top_type, top_ref, B, 1); if (!(top_ref | AV_RN32A(B))) goto zeromv; - } else if (h->top_type) { + } else if (sl->top_type) { top_ref = LIST_NOT_USED; B = zeromv; } else { @@ -295,21 +301,21 @@ static av_always_inline void pred_pskip_motion(H264Context *const h) tprintf(h->avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->mb_x, h->mb_y); - if (USES_LIST(h->topright_type, 0)) { - diagonal_ref = ref[4 * h->topright_mb_xy + 2]; - C = mv[h->mb2b_xy[h->topright_mb_xy] + 3 * b_stride]; - FIX_MV_MBAFF(h->topright_type, diagonal_ref, C, 2); - } else if (h->topright_type) { + if (USES_LIST(sl->topright_type, 0)) { + diagonal_ref = ref[4 * sl->topright_mb_xy + 2]; + C = mv[h->mb2b_xy[sl->topright_mb_xy] + 3 * b_stride]; + FIX_MV_MBAFF(sl->topright_type, diagonal_ref, C, 2); + } else if (sl->topright_type) { diagonal_ref = LIST_NOT_USED; C = zeromv; } else { - if (USES_LIST(h->topleft_type, 0)) { - diagonal_ref = ref[4 * h->topleft_mb_xy + 1 + - (h->topleft_partition & 2)]; - C = mv[h->mb2b_xy[h->topleft_mb_xy] + 3 + b_stride + - (h->topleft_partition & 2 * b_stride)]; - FIX_MV_MBAFF(h->topleft_type, diagonal_ref, C, 2); - } else if (h->topleft_type) { + if (USES_LIST(sl->topleft_type, 0)) { + diagonal_ref = ref[4 * sl->topleft_mb_xy + 1 + + (sl->topleft_partition & 2)]; + C = mv[h->mb2b_xy[sl->topleft_mb_xy] + 3 + b_stride + + (sl->topleft_partition & 2 * b_stride)]; + FIX_MV_MBAFF(sl->topleft_type, diagonal_ref, C, 2); + } else if (sl->topleft_type) { diagonal_ref = LIST_NOT_USED; C = zeromv; } else { @@ -347,7 +353,7 @@ zeromv: return; } -static void fill_decode_neighbors(H264Context *h, int mb_type) +static void fill_decode_neighbors(H264Context *h, H264SliceContext *sl, int mb_type) { const int mb_xy = h->mb_xy; int topleft_xy, top_xy, topright_xy, left_xy[LEFT_MBS]; @@ -358,7 +364,7 @@ static void fill_decode_neighbors(H264Context *h, int mb_type) { 0, 2, 0, 2, 7, 10, 7, 10, 3 + 0 * 4, 3 + 2 * 4, 3 + 0 * 4, 3 + 2 * 4, 1 + 4 * 4, 1 + 8 * 4, 1 + 4 * 4, 1 + 8 * 4 } }; - h->topleft_partition = -1; + sl->topleft_partition = -1; top_xy = mb_xy - (h->mb_stride << MB_FIELD(h)); @@ -368,7 +374,7 @@ static void fill_decode_neighbors(H264Context *h, int mb_type) topleft_xy = top_xy - 1; topright_xy = top_xy + 1; left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1; - h->left_block = left_block_options[0]; + sl->left_block = left_block_options[0]; if (FRAME_MBAFF(h)) { const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]); const int curr_mb_field_flag = IS_INTERLACED(mb_type); @@ -377,13 +383,13 @@ static void fill_decode_neighbors(H264Context *h, int mb_type) left_xy[LBOT] = left_xy[LTOP] = mb_xy - h->mb_stride - 1; if (curr_mb_field_flag) { left_xy[LBOT] += h->mb_stride; - h->left_block = left_block_options[3]; + sl->left_block = left_block_options[3]; } else { topleft_xy += h->mb_stride; /* take top left mv from the middle of the mb, as opposed * to all other modes which use the bottom right partition */ - h->topleft_partition = 0; - h->left_block = left_block_options[1]; + sl->topleft_partition = 0; + sl->left_block = left_block_options[1]; } } } else { @@ -395,66 +401,66 @@ static void fill_decode_neighbors(H264Context *h, int mb_type) if (left_mb_field_flag != curr_mb_field_flag) { if (curr_mb_field_flag) { left_xy[LBOT] += h->mb_stride; - h->left_block = left_block_options[3]; + sl->left_block = left_block_options[3]; } else { - h->left_block = left_block_options[2]; + sl->left_block = left_block_options[2]; } } } } - h->topleft_mb_xy = topleft_xy; - h->top_mb_xy = top_xy; - h->topright_mb_xy = topright_xy; - h->left_mb_xy[LTOP] = left_xy[LTOP]; - h->left_mb_xy[LBOT] = left_xy[LBOT]; + sl->topleft_mb_xy = topleft_xy; + sl->top_mb_xy = top_xy; + sl->topright_mb_xy = topright_xy; + sl->left_mb_xy[LTOP] = left_xy[LTOP]; + sl->left_mb_xy[LBOT] = left_xy[LBOT]; //FIXME do we need all in the context? - h->topleft_type = h->cur_pic.mb_type[topleft_xy]; - h->top_type = h->cur_pic.mb_type[top_xy]; - h->topright_type = h->cur_pic.mb_type[topright_xy]; - h->left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]]; - h->left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]]; + sl->topleft_type = h->cur_pic.mb_type[topleft_xy]; + sl->top_type = h->cur_pic.mb_type[top_xy]; + sl->topright_type = h->cur_pic.mb_type[topright_xy]; + sl->left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]]; + sl->left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]]; if (FMO) { if (h->slice_table[topleft_xy] != h->slice_num) - h->topleft_type = 0; + sl->topleft_type = 0; if (h->slice_table[top_xy] != h->slice_num) - h->top_type = 0; + sl->top_type = 0; if (h->slice_table[left_xy[LTOP]] != h->slice_num) - h->left_type[LTOP] = h->left_type[LBOT] = 0; + sl->left_type[LTOP] = sl->left_type[LBOT] = 0; } else { if (h->slice_table[topleft_xy] != h->slice_num) { - h->topleft_type = 0; + sl->topleft_type = 0; if (h->slice_table[top_xy] != h->slice_num) - h->top_type = 0; + sl->top_type = 0; if (h->slice_table[left_xy[LTOP]] != h->slice_num) - h->left_type[LTOP] = h->left_type[LBOT] = 0; + sl->left_type[LTOP] = sl->left_type[LBOT] = 0; } } if (h->slice_table[topright_xy] != h->slice_num) - h->topright_type = 0; + sl->topright_type = 0; } -static void fill_decode_caches(H264Context *h, int mb_type) +static void fill_decode_caches(H264Context *h, H264SliceContext *sl, int mb_type) { int topleft_xy, top_xy, topright_xy, left_xy[LEFT_MBS]; int topleft_type, top_type, topright_type, left_type[LEFT_MBS]; - const uint8_t *left_block = h->left_block; + const uint8_t *left_block = sl->left_block; int i; uint8_t *nnz; uint8_t *nnz_cache; - topleft_xy = h->topleft_mb_xy; - top_xy = h->top_mb_xy; - topright_xy = h->topright_mb_xy; - left_xy[LTOP] = h->left_mb_xy[LTOP]; - left_xy[LBOT] = h->left_mb_xy[LBOT]; - topleft_type = h->topleft_type; - top_type = h->top_type; - topright_type = h->topright_type; - left_type[LTOP] = h->left_type[LTOP]; - left_type[LBOT] = h->left_type[LBOT]; + topleft_xy = sl->topleft_mb_xy; + top_xy = sl->top_mb_xy; + topright_xy = sl->topright_mb_xy; + left_xy[LTOP] = sl->left_mb_xy[LTOP]; + left_xy[LBOT] = sl->left_mb_xy[LBOT]; + topleft_type = sl->topleft_type; + top_type = sl->top_type; + topright_type = sl->topright_type; + left_type[LTOP] = sl->left_type[LTOP]; + left_type[LBOT] = sl->left_type[LBOT]; if (!IS_SKIP(mb_type)) { if (IS_INTRA(mb_type)) { @@ -667,8 +673,8 @@ static void fill_decode_caches(H264Context *h, int mb_type) if (ref_cache[4 - 1 * 8] < 0) { if (USES_LIST(topleft_type, list)) { const int b_xy = h->mb2b_xy[topleft_xy] + 3 + b_stride + - (h->topleft_partition & 2 * b_stride); - const int b8_xy = 4 * topleft_xy + 1 + (h->topleft_partition & 2); + (sl->topleft_partition & 2 * b_stride); + const int b8_xy = 4 * topleft_xy + 1 + (sl->topleft_partition & 2); AV_COPY32(mv_cache[-1 - 1 * 8], mv[b_xy]); ref_cache[-1 - 1 * 8] = ref[b8_xy]; } else { @@ -808,16 +814,16 @@ static void av_unused decode_mb_skip(H264Context *h, H264SliceContext *sl) // just for fill_caches. pred_direct_motion will set the real mb_type mb_type |= MB_TYPE_L0L1 | MB_TYPE_DIRECT2 | MB_TYPE_SKIP; if (h->direct_spatial_mv_pred) { - fill_decode_neighbors(h, mb_type); - fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ... + fill_decode_neighbors(h, sl, mb_type); + fill_decode_caches(h, sl, mb_type); //FIXME check what is needed and what not ... } ff_h264_pred_direct_motion(h, &mb_type); mb_type |= MB_TYPE_SKIP; } else { mb_type |= MB_TYPE_16x16 | MB_TYPE_P0L0 | MB_TYPE_P1L0 | MB_TYPE_SKIP; - fill_decode_neighbors(h, mb_type); - pred_pskip_motion(h); + fill_decode_neighbors(h, sl, mb_type); + pred_pskip_motion(h, sl); } write_back_motion(h, mb_type); diff --git a/libavcodec/h264_slice.c b/libavcodec/h264_slice.c index c69dac9068..b40757b140 100644 --- a/libavcodec/h264_slice.c +++ b/libavcodec/h264_slice.c @@ -1932,9 +1932,9 @@ static int fill_filter_caches(H264Context *h, H264SliceContext *sl, int mb_type) } } - h->top_mb_xy = top_xy; - h->left_mb_xy[LTOP] = left_xy[LTOP]; - h->left_mb_xy[LBOT] = left_xy[LBOT]; + sl->top_mb_xy = top_xy; + sl->left_mb_xy[LTOP] = left_xy[LTOP]; + sl->left_mb_xy[LBOT] = left_xy[LBOT]; { /* For sufficiently low qp, filtering wouldn't do anything. * This is a conservative estimate: could also check beta_offset @@ -1970,9 +1970,9 @@ static int fill_filter_caches(H264Context *h, H264SliceContext *sl, int mb_type) if (h->slice_table[left_xy[LBOT]] == 0xFFFF) left_type[LTOP] = left_type[LBOT] = 0; } - h->top_type = top_type; - h->left_type[LTOP] = left_type[LTOP]; - h->left_type[LBOT] = left_type[LBOT]; + sl->top_type = top_type; + sl->left_type[LTOP] = left_type[LTOP]; + sl->left_type[LBOT] = left_type[LBOT]; if (IS_INTRA(mb_type)) return 0; diff --git a/libavcodec/svq3.c b/libavcodec/svq3.c index cad546bc61..87ddc1a580 100644 --- a/libavcodec/svq3.c +++ b/libavcodec/svq3.c @@ -369,6 +369,7 @@ static inline int svq3_mc_dir(SVQ3Context *s, int size, int mode, { int i, j, k, mx, my, dx, dy, x, y; H264Context *h = &s->h; + H264SliceContext *sl = &h->slice_ctx[0]; const int part_width = ((size & 5) == 4) ? 4 : 16 >> (size & 1); const int part_height = 16 >> ((unsigned)(size + 1) / 3); const int extra_width = (mode == PREDICT_MODE) ? -16 * 6 : 0; @@ -386,7 +387,7 @@ static inline int svq3_mc_dir(SVQ3Context *s, int size, int mode, (j >> 1 & 4) + (i & 8); if (mode != PREDICT_MODE) { - pred_motion(h, k, part_width >> 2, dir, 1, &mx, &my); + pred_motion(h, sl, k, part_width >> 2, dir, 1, &mx, &my); } else { mx = s->next_pic->motion_val[0][b_xy][0] << 1; my = s->next_pic->motion_val[0][b_xy][1] << 1;