diff --git a/libavcodec/h264_mvpred.h b/libavcodec/h264_mvpred.h index 2bd4458f0b..5244c290f2 100644 --- a/libavcodec/h264_mvpred.h +++ b/libavcodec/h264_mvpred.h @@ -35,53 +35,53 @@ //#undef NDEBUG #include -static av_always_inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){ - const int topright_ref= h->ref_cache[list][ i - 8 + part_width ]; - MpegEncContext *s = &h->s; +static av_always_inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, + int i, int list, int part_width) +{ + const int topright_ref = h->ref_cache[list][i - 8 + part_width]; + MpegEncContext *s = &h->s; /* there is no consistent mapping of mvs to neighboring locations that will * make mbaff happy, so we can't move all this logic to fill_caches */ - if(FRAME_MBAFF){ + if (FRAME_MBAFF) { +#define SET_DIAG_MV(MV_OP, REF_OP, XY, Y4) \ + const int xy = XY, y4 = Y4; \ + const int mb_type = mb_types[xy + (y4 >> 2) * s->mb_stride]; \ + if (!USES_LIST(mb_type, list)) \ + return LIST_NOT_USED; \ + mv = s->current_picture_ptr->f.motion_val[list][h->mb2b_xy[xy] + 3 + y4 * h->b_stride]; \ + h->mv_cache[list][scan8[0] - 2][0] = mv[0]; \ + h->mv_cache[list][scan8[0] - 2][1] = mv[1] MV_OP; \ + return s->current_picture_ptr->f.ref_index[list][4 * xy + 1 + (y4 & ~1)] REF_OP; -#define SET_DIAG_MV(MV_OP, REF_OP, XY, Y4)\ - const int xy = XY, y4 = Y4;\ - const int mb_type = mb_types[xy+(y4>>2)*s->mb_stride];\ - if(!USES_LIST(mb_type,list))\ - return LIST_NOT_USED;\ - mv = s->current_picture_ptr->f.motion_val[list][h->mb2b_xy[xy] + 3 + y4*h->b_stride];\ - h->mv_cache[list][scan8[0]-2][0] = mv[0];\ - h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\ - return s->current_picture_ptr->f.ref_index[list][4*xy + 1 + (y4 & ~1)] REF_OP; - - if(topright_ref == PART_NOT_AVAILABLE - && i >= scan8[0]+8 && (i&7)==4 - && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){ + if (topright_ref == PART_NOT_AVAILABLE + && i >= scan8[0] + 8 && (i & 7) == 4 + && h->ref_cache[list][scan8[0] - 1] != PART_NOT_AVAILABLE) { const uint32_t *mb_types = s->current_picture_ptr->f.mb_type; const int16_t *mv; - AV_ZERO32(h->mv_cache[list][scan8[0]-2]); - *C = h->mv_cache[list][scan8[0]-2]; + AV_ZERO32(h->mv_cache[list][scan8[0] - 2]); + *C = h->mv_cache[list][scan8[0] - 2]; - if(!MB_FIELD - && IS_INTERLACED(h->left_type[0])){ - SET_DIAG_MV(*2, >>1, h->left_mb_xy[0]+s->mb_stride, (s->mb_y&1)*2+(i>>5)); + if (!MB_FIELD && IS_INTERLACED(h->left_type[0])) { + SET_DIAG_MV(* 2, >> 1, h->left_mb_xy[0] + s->mb_stride, + (s->mb_y & 1) * 2 + (i >> 5)); } - if(MB_FIELD - && !IS_INTERLACED(h->left_type[0])){ + if (MB_FIELD && !IS_INTERLACED(h->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, h->left_mb_xy[i >= 36], ((i >> 2)) & 3); } } #undef SET_DIAG_MV } - if(topright_ref != PART_NOT_AVAILABLE){ - *C= h->mv_cache[list][ i - 8 + part_width ]; + if (topright_ref != PART_NOT_AVAILABLE) { + *C = h->mv_cache[list][i - 8 + part_width]; return topright_ref; - }else{ + } else { tprintf(s->avctx, "topright MV not available\n"); - *C= h->mv_cache[list][ i - 8 - 1 ]; - return h->ref_cache[list][ i - 8 - 1 ]; + *C = h->mv_cache[list][i - 8 - 1]; + return h->ref_cache[list][i - 8 - 1]; } } @@ -92,53 +92,61 @@ 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, int part_width, int list, int ref, int * const mx, int * const my){ - const int index8= scan8[n]; - const int top_ref= h->ref_cache[list][ index8 - 8 ]; - const int left_ref= h->ref_cache[list][ index8 - 1 ]; - const int16_t * const A= h->mv_cache[list][ index8 - 1 ]; - const int16_t * const B= h->mv_cache[list][ index8 - 8 ]; - const int16_t * C; +static av_always_inline void pred_motion(H264Context *const h, int n, + int part_width, int list, int ref, + int *const mx, int *const my) +{ + const int index8 = scan8[n]; + const int top_ref = h->ref_cache[list][index8 - 8]; + const int left_ref = h->ref_cache[list][index8 - 1]; + const int16_t *const A = h->mv_cache[list][index8 - 1]; + const int16_t *const B = h->mv_cache[list][index8 - 8]; + const int16_t *C; int diagonal_ref, match_count; - assert(part_width==1 || part_width==2 || part_width==4); + assert(part_width == 1 || part_width == 2 || part_width == 4); /* mv_cache - B . . A T T T T - U . . L . . , . - U . . L . . . . - U . . L . . , . - . . . L . . . . -*/ + * B . . A T T T T + * U . . L . . , . + * U . . L . . . . + * U . . L . . , . + * . . . L . . . . + */ - diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width); - match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref); + diagonal_ref = fetch_diagonal_mv(h, &C, index8, list, part_width); + match_count = (diagonal_ref == ref) + (top_ref == ref) + (left_ref == ref); tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count); - if(match_count > 1){ //most common - *mx= mid_pred(A[0], B[0], C[0]); - *my= mid_pred(A[1], B[1], C[1]); - }else if(match_count==1){ - if(left_ref==ref){ - *mx= A[0]; - *my= A[1]; - }else if(top_ref==ref){ - *mx= B[0]; - *my= B[1]; - }else{ - *mx= C[0]; - *my= C[1]; + if (match_count > 1) { //most common + *mx = mid_pred(A[0], B[0], C[0]); + *my = mid_pred(A[1], B[1], C[1]); + } else if (match_count == 1) { + if (left_ref == ref) { + *mx = A[0]; + *my = A[1]; + } else if (top_ref == ref) { + *mx = B[0]; + *my = B[1]; + } else { + *mx = C[0]; + *my = C[1]; } - }else{ - if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){ - *mx= A[0]; - *my= A[1]; - }else{ - *mx= mid_pred(A[0], B[0], C[0]); - *my= mid_pred(A[1], B[1], C[1]); + } else { + if (top_ref == PART_NOT_AVAILABLE && + diagonal_ref == PART_NOT_AVAILABLE && + left_ref != PART_NOT_AVAILABLE) { + *mx = A[0]; + *my = A[1]; + } else { + *mx = mid_pred(A[0], B[0], C[0]); + *my = mid_pred(A[1], B[1], C[1]); } } - tprintf(h->s.avctx, "pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list); + tprintf(h->s.avctx, + "pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", + top_ref, B[0], B[1], diagonal_ref, C[0], C[1], left_ref, + A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list); } /** @@ -147,27 +155,32 @@ static av_always_inline void pred_motion(H264Context * const h, int n, int part_ * @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_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){ - if(n==0){ - const int top_ref= h->ref_cache[list][ scan8[0] - 8 ]; - const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ]; +static av_always_inline void pred_16x8_motion(H264Context *const h, + int n, int list, int ref, + int *const mx, int *const my) +{ + if (n == 0) { + const int top_ref = h->ref_cache[list][scan8[0] - 8]; + const int16_t *const B = h->mv_cache[list][scan8[0] - 8]; - tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list); + tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", + top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list); - if(top_ref == ref){ - *mx= B[0]; - *my= B[1]; + if (top_ref == ref) { + *mx = B[0]; + *my = B[1]; return; } - }else{ - const int left_ref= h->ref_cache[list][ scan8[8] - 1 ]; - const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ]; + } else { + const int left_ref = h->ref_cache[list][scan8[8] - 1]; + const int16_t *const A = h->mv_cache[list][scan8[8] - 1]; - tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list); + tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", + left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list); - if(left_ref == ref){ - *mx= A[0]; - *my= A[1]; + if (left_ref == ref) { + *mx = A[0]; + *my = A[1]; return; } } @@ -182,29 +195,34 @@ static av_always_inline void pred_16x8_motion(H264Context * const h, int n, int * @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_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){ - if(n==0){ - const int left_ref= h->ref_cache[list][ scan8[0] - 1 ]; - const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ]; +static av_always_inline void pred_8x16_motion(H264Context *const h, + int n, int list, int ref, + int *const mx, int *const my) +{ + if (n == 0) { + const int left_ref = h->ref_cache[list][scan8[0] - 1]; + const int16_t *const A = h->mv_cache[list][scan8[0] - 1]; - tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list); + tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", + left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list); - if(left_ref == ref){ - *mx= A[0]; - *my= A[1]; + if (left_ref == ref) { + *mx = A[0]; + *my = A[1]; return; } - }else{ - const int16_t * C; + } else { + const int16_t *C; int diagonal_ref; - diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2); + diagonal_ref = fetch_diagonal_mv(h, &C, scan8[4], list, 2); - tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list); + tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", + diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list); - if(diagonal_ref == ref){ - *mx= C[0]; - *my= C[1]; + if (diagonal_ref == ref) { + *mx = C[0]; + *my = C[1]; return; } } @@ -213,168 +231,174 @@ static av_always_inline void pred_8x16_motion(H264Context * const h, int n, int pred_motion(h, n, 2, list, ref, mx, my); } -#define FIX_MV_MBAFF(type, refn, mvn, idx)\ - if(FRAME_MBAFF){\ - if(MB_FIELD){\ - if(!IS_INTERLACED(type)){\ - refn <<= 1;\ - AV_COPY32(mvbuf[idx], mvn);\ - mvbuf[idx][1] /= 2;\ - mvn = mvbuf[idx];\ - }\ - }else{\ - if(IS_INTERLACED(type)){\ - refn >>= 1;\ - AV_COPY32(mvbuf[idx], mvn);\ - mvbuf[idx][1] <<= 1;\ - mvn = mvbuf[idx];\ - }\ - }\ +#define FIX_MV_MBAFF(type, refn, mvn, idx) \ + if (FRAME_MBAFF) { \ + if (MB_FIELD) { \ + if (!IS_INTERLACED(type)) { \ + refn <<= 1; \ + AV_COPY32(mvbuf[idx], mvn); \ + mvbuf[idx][1] /= 2; \ + mvn = mvbuf[idx]; \ + } \ + } else { \ + if (IS_INTERLACED(type)) { \ + refn >>= 1; \ + AV_COPY32(mvbuf[idx], mvn); \ + mvbuf[idx][1] <<= 1; \ + mvn = mvbuf[idx]; \ + } \ + } \ } -static av_always_inline void pred_pskip_motion(H264Context * const h){ - DECLARE_ALIGNED(4, static const int16_t, zeromv)[2] = {0}; +static av_always_inline void pred_pskip_motion(H264Context *const h) +{ + DECLARE_ALIGNED(4, static const int16_t, zeromv)[2] = { 0 }; DECLARE_ALIGNED(4, int16_t, mvbuf)[3][2]; - MpegEncContext * const s = &h->s; - int8_t *ref = s->current_picture.f.ref_index[0]; - int16_t (*mv)[2] = s->current_picture.f.motion_val[0]; + MpegEncContext *const s = &h->s; + int8_t *ref = s->current_picture.f.ref_index[0]; + int16_t(*mv)[2] = s->current_picture.f.motion_val[0]; int top_ref, left_ref, diagonal_ref, match_count, mx, my; const int16_t *A, *B, *C; int b_stride = h->b_stride; fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1); - /* To avoid doing an entire fill_decode_caches, we inline the relevant parts here. - * 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? + /* To avoid doing an entire fill_decode_caches, we inline the relevant + * parts here. + * 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]]; + 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(!(left_ref | AV_RN32A(A))){ + if (!(left_ref | AV_RN32A(A))) goto zeromv; - } - }else if(h->left_type[LTOP]){ + } else if (h->left_type[LTOP]) { left_ref = LIST_NOT_USED; - A = zeromv; - }else{ + 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]; + 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(!(top_ref | AV_RN32A(B))){ + if (!(top_ref | AV_RN32A(B))) goto zeromv; - } - }else if(h->top_type){ + } else if (h->top_type) { top_ref = LIST_NOT_USED; - B = zeromv; - }else{ + B = zeromv; + } else { goto zeromv; } - tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y); + tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", + top_ref, left_ref, h->s.mb_x, h->s.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]; + 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){ + } else if (h->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)]; + } 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){ + } else if (h->topleft_type) { diagonal_ref = LIST_NOT_USED; - C = zeromv; - }else{ + C = zeromv; + } else { diagonal_ref = PART_NOT_AVAILABLE; - C = zeromv; + C = zeromv; } } - match_count= !diagonal_ref + !top_ref + !left_ref; + match_count = !diagonal_ref + !top_ref + !left_ref; tprintf(h->s.avctx, "pred_pskip_motion match_count=%d\n", match_count); - if(match_count > 1){ + if (match_count > 1) { mx = mid_pred(A[0], B[0], C[0]); my = mid_pred(A[1], B[1], C[1]); - }else if(match_count==1){ - if(!left_ref){ + } else if (match_count == 1) { + if (!left_ref) { mx = A[0]; my = A[1]; - }else if(!top_ref){ + } else if (!top_ref) { mx = B[0]; my = B[1]; - }else{ + } else { mx = C[0]; my = C[1]; } - }else{ + } else { mx = mid_pred(A[0], B[0], C[0]); my = mid_pred(A[1], B[1], C[1]); } - fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4); + fill_rectangle(h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx, my), 4); return; + zeromv: - fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4); + fill_rectangle(h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4); return; } -static void fill_decode_neighbors(H264Context *h, int mb_type){ - MpegEncContext * const s = &h->s; - const int mb_xy= h->mb_xy; +static void fill_decode_neighbors(H264Context *h, int mb_type) +{ + MpegEncContext *const s = &h->s; + const int mb_xy = h->mb_xy; int topleft_xy, top_xy, topright_xy, left_xy[LEFT_MBS]; - static const uint8_t left_block_options[4][32]={ - {0,1,2,3,7,10,8,11,3+0*4, 3+1*4, 3+2*4, 3+3*4, 1+4*4, 1+8*4, 1+5*4, 1+9*4}, - {2,2,3,3,8,11,8,11,3+2*4, 3+2*4, 3+3*4, 3+3*4, 1+5*4, 1+9*4, 1+5*4, 1+9*4}, - {0,0,1,1,7,10,7,10,3+0*4, 3+0*4, 3+1*4, 3+1*4, 1+4*4, 1+8*4, 1+4*4, 1+8*4}, - {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} + static const uint8_t left_block_options[4][32] = { + { 0, 1, 2, 3, 7, 10, 8, 11, 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4, 1 + 4 * 4, 1 + 8 * 4, 1 + 5 * 4, 1 + 9 * 4 }, + { 2, 2, 3, 3, 8, 11, 8, 11, 3 + 2 * 4, 3 + 2 * 4, 3 + 3 * 4, 3 + 3 * 4, 1 + 5 * 4, 1 + 9 * 4, 1 + 5 * 4, 1 + 9 * 4 }, + { 0, 0, 1, 1, 7, 10, 7, 10, 3 + 0 * 4, 3 + 0 * 4, 3 + 1 * 4, 3 + 1 * 4, 1 + 4 * 4, 1 + 8 * 4, 1 + 4 * 4, 1 + 8 * 4 }, + { 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; + h->topleft_partition = -1; - top_xy = mb_xy - (s->mb_stride << MB_FIELD); + top_xy = mb_xy - (s->mb_stride << MB_FIELD); /* Wow, what a mess, why didn't they simplify the interlacing & intra * stuff, I can't imagine that these complex rules are worth it. */ - topleft_xy = top_xy - 1; - topright_xy= top_xy + 1; - left_xy[LBOT] = left_xy[LTOP] = mb_xy-1; + 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]; - if(FRAME_MBAFF){ + if (FRAME_MBAFF) { const int left_mb_field_flag = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]); const int curr_mb_field_flag = IS_INTERLACED(mb_type); - if(s->mb_y&1){ + if (s->mb_y & 1) { if (left_mb_field_flag != curr_mb_field_flag) { left_xy[LBOT] = left_xy[LTOP] = mb_xy - s->mb_stride - 1; if (curr_mb_field_flag) { left_xy[LBOT] += s->mb_stride; - h->left_block = left_block_options[3]; + h->left_block = left_block_options[3]; } else { topleft_xy += s->mb_stride; - // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition + /* 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]; + h->left_block = left_block_options[1]; } } - }else{ - if(curr_mb_field_flag){ + } else { + if (curr_mb_field_flag) { topleft_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy - 1] >> 7) & 1) - 1); topright_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy + 1] >> 7) & 1) - 1); - top_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy ] >> 7) & 1) - 1); + top_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy] >> 7) & 1) - 1); } if (left_mb_field_flag != curr_mb_field_flag) { if (curr_mb_field_flag) { left_xy[LBOT] += s->mb_stride; - h->left_block = left_block_options[3]; + h->left_block = left_block_options[3]; } else { h->left_block = left_block_options[2]; } @@ -382,9 +406,9 @@ static void fill_decode_neighbors(H264Context *h, int mb_type){ } } - h->topleft_mb_xy = topleft_xy; - h->top_mb_xy = top_xy; - h->topright_mb_xy= topright_xy; + 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]; //FIXME do we need all in the context? @@ -395,351 +419,372 @@ static void fill_decode_neighbors(H264Context *h, int mb_type){ h->left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]]; h->left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]]; - if(FMO){ - if(h->slice_table[topleft_xy ] != h->slice_num) h->topleft_type = 0; - if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0; - if(h->slice_table[left_xy[LTOP] ] != h->slice_num) h->left_type[LTOP] = h->left_type[LBOT] = 0; - }else{ - if(h->slice_table[topleft_xy ] != h->slice_num){ + if (FMO) { + if (h->slice_table[topleft_xy] != h->slice_num) h->topleft_type = 0; - if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0; - if(h->slice_table[left_xy[LTOP] ] != h->slice_num) h->left_type[LTOP] = h->left_type[LBOT] = 0; + if (h->slice_table[top_xy] != h->slice_num) + h->top_type = 0; + if (h->slice_table[left_xy[LTOP]] != h->slice_num) + h->left_type[LTOP] = h->left_type[LBOT] = 0; + } else { + if (h->slice_table[topleft_xy] != h->slice_num) { + h->topleft_type = 0; + if (h->slice_table[top_xy] != h->slice_num) + h->top_type = 0; + if (h->slice_table[left_xy[LTOP]] != h->slice_num) + h->left_type[LTOP] = h->left_type[LBOT] = 0; } } - if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0; + if (h->slice_table[topright_xy] != h->slice_num) + h->topright_type = 0; } -static void fill_decode_caches(H264Context *h, int mb_type){ - MpegEncContext * const s = &h->s; +static void fill_decode_caches(H264Context *h, int mb_type) +{ + MpegEncContext *const s = &h->s; 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 = h->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 = 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]; - if(!IS_SKIP(mb_type)){ - if(IS_INTRA(mb_type)){ - int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1; - h->topleft_samples_available= - h->top_samples_available= - h->left_samples_available= 0xFFFF; - h->topright_samples_available= 0xEEEA; + if (!IS_SKIP(mb_type)) { + if (IS_INTRA(mb_type)) { + int type_mask = h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1; + h->topleft_samples_available = + h->top_samples_available = + h->left_samples_available = 0xFFFF; + h->topright_samples_available = 0xEEEA; - if(!(top_type & type_mask)){ - h->topleft_samples_available= 0xB3FF; - h->top_samples_available= 0x33FF; - h->topright_samples_available= 0x26EA; + if (!(top_type & type_mask)) { + h->topleft_samples_available = 0xB3FF; + h->top_samples_available = 0x33FF; + h->topright_samples_available = 0x26EA; } - if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[LTOP])){ - if(IS_INTERLACED(mb_type)){ - if(!(left_type[LTOP] & type_mask)){ - h->topleft_samples_available&= 0xDFFF; - h->left_samples_available&= 0x5FFF; + if (IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[LTOP])) { + if (IS_INTERLACED(mb_type)) { + if (!(left_type[LTOP] & type_mask)) { + h->topleft_samples_available &= 0xDFFF; + h->left_samples_available &= 0x5FFF; } - if(!(left_type[LBOT] & type_mask)){ - h->topleft_samples_available&= 0xFF5F; - h->left_samples_available&= 0xFF5F; + if (!(left_type[LBOT] & type_mask)) { + h->topleft_samples_available &= 0xFF5F; + h->left_samples_available &= 0xFF5F; } - }else{ + } else { int left_typei = s->current_picture.f.mb_type[left_xy[LTOP] + s->mb_stride]; assert(left_xy[LTOP] == left_xy[LBOT]); - if(!((left_typei & type_mask) && (left_type[LTOP] & type_mask))){ - h->topleft_samples_available&= 0xDF5F; - h->left_samples_available&= 0x5F5F; + if (!((left_typei & type_mask) && (left_type[LTOP] & type_mask))) { + h->topleft_samples_available &= 0xDF5F; + h->left_samples_available &= 0x5F5F; } } - }else{ - if(!(left_type[LTOP] & type_mask)){ - h->topleft_samples_available&= 0xDF5F; - h->left_samples_available&= 0x5F5F; + } else { + if (!(left_type[LTOP] & type_mask)) { + h->topleft_samples_available &= 0xDF5F; + h->left_samples_available &= 0x5F5F; } } - if(!(topleft_type & type_mask)) - h->topleft_samples_available&= 0x7FFF; + if (!(topleft_type & type_mask)) + h->topleft_samples_available &= 0x7FFF; - if(!(topright_type & type_mask)) - h->topright_samples_available&= 0xFBFF; + if (!(topright_type & type_mask)) + h->topright_samples_available &= 0xFBFF; - if(IS_INTRA4x4(mb_type)){ - if(IS_INTRA4x4(top_type)){ - AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]); - }else{ - h->intra4x4_pred_mode_cache[4+8*0]= - h->intra4x4_pred_mode_cache[5+8*0]= - h->intra4x4_pred_mode_cache[6+8*0]= - h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask); + if (IS_INTRA4x4(mb_type)) { + if (IS_INTRA4x4(top_type)) { + AV_COPY32(h->intra4x4_pred_mode_cache + 4 + 8 * 0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]); + } else { + h->intra4x4_pred_mode_cache[4 + 8 * 0] = + h->intra4x4_pred_mode_cache[5 + 8 * 0] = + h->intra4x4_pred_mode_cache[6 + 8 * 0] = + h->intra4x4_pred_mode_cache[7 + 8 * 0] = 2 - 3 * !(top_type & type_mask); } - for(i=0; i<2; i++){ - if(IS_INTRA4x4(left_type[LEFT(i)])){ - int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[LEFT(i)]]; - h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]]; - h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]]; - }else{ - h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= - h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[LEFT(i)] & type_mask); + for (i = 0; i < 2; i++) { + if (IS_INTRA4x4(left_type[LEFT(i)])) { + int8_t *mode = h->intra4x4_pred_mode + h->mb2br_xy[left_xy[LEFT(i)]]; + h->intra4x4_pred_mode_cache[3 + 8 * 1 + 2 * 8 * i] = mode[6 - left_block[0 + 2 * i]]; + h->intra4x4_pred_mode_cache[3 + 8 * 2 + 2 * 8 * i] = mode[6 - left_block[1 + 2 * i]]; + } else { + h->intra4x4_pred_mode_cache[3 + 8 * 1 + 2 * 8 * i] = + h->intra4x4_pred_mode_cache[3 + 8 * 2 + 2 * 8 * i] = 2 - 3 * !(left_type[LEFT(i)] & type_mask); } } } } - -/* -0 . T T. T T T T -1 L . .L . . . . -2 L . .L . . . . -3 . T TL . . . . -4 L . .L . . . . -5 L . .. . . . . -*/ -//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec) - nnz_cache = h->non_zero_count_cache; - if(top_type){ - nnz = h->non_zero_count[top_xy]; - AV_COPY32(&nnz_cache[4+8* 0], &nnz[4*3]); - if(!s->chroma_y_shift){ - AV_COPY32(&nnz_cache[4+8* 5], &nnz[4* 7]); - AV_COPY32(&nnz_cache[4+8*10], &nnz[4*11]); - }else{ - AV_COPY32(&nnz_cache[4+8* 5], &nnz[4* 5]); - AV_COPY32(&nnz_cache[4+8*10], &nnz[4* 9]); - } - }else{ - uint32_t top_empty = CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040; - AV_WN32A(&nnz_cache[4+8* 0], top_empty); - AV_WN32A(&nnz_cache[4+8* 5], top_empty); - AV_WN32A(&nnz_cache[4+8*10], top_empty); - } - - for (i=0; i<2; i++) { - if(left_type[LEFT(i)]){ - nnz = h->non_zero_count[left_xy[LEFT(i)]]; - nnz_cache[3+8* 1 + 2*8*i]= nnz[left_block[8+0+2*i]]; - nnz_cache[3+8* 2 + 2*8*i]= nnz[left_block[8+1+2*i]]; - if(CHROMA444){ - nnz_cache[3+8* 6 + 2*8*i]= nnz[left_block[8+0+2*i]+4*4]; - nnz_cache[3+8* 7 + 2*8*i]= nnz[left_block[8+1+2*i]+4*4]; - nnz_cache[3+8*11 + 2*8*i]= nnz[left_block[8+0+2*i]+8*4]; - nnz_cache[3+8*12 + 2*8*i]= nnz[left_block[8+1+2*i]+8*4]; - }else if(CHROMA422) { - nnz_cache[3+8* 6 + 2*8*i]= nnz[left_block[8+0+2*i]-2+4*4]; - nnz_cache[3+8* 7 + 2*8*i]= nnz[left_block[8+1+2*i]-2+4*4]; - nnz_cache[3+8*11 + 2*8*i]= nnz[left_block[8+0+2*i]-2+8*4]; - nnz_cache[3+8*12 + 2*8*i]= nnz[left_block[8+1+2*i]-2+8*4]; - }else{ - nnz_cache[3+8* 6 + 8*i]= nnz[left_block[8+4+2*i]]; - nnz_cache[3+8*11 + 8*i]= nnz[left_block[8+5+2*i]]; + /* + * 0 . T T. T T T T + * 1 L . .L . . . . + * 2 L . .L . . . . + * 3 . T TL . . . . + * 4 L . .L . . . . + * 5 L . .. . . . . + */ + /* FIXME: constraint_intra_pred & partitioning & nnz + * (let us hope this is just a typo in the spec) */ + nnz_cache = h->non_zero_count_cache; + if (top_type) { + nnz = h->non_zero_count[top_xy]; + AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[4 * 3]); + if (!s->chroma_y_shift) { + AV_COPY32(&nnz_cache[4 + 8 * 5], &nnz[4 * 7]); + AV_COPY32(&nnz_cache[4 + 8 * 10], &nnz[4 * 11]); + } else { + AV_COPY32(&nnz_cache[4 + 8 * 5], &nnz[4 * 5]); + AV_COPY32(&nnz_cache[4 + 8 * 10], &nnz[4 * 9]); + } + } else { + uint32_t top_empty = CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040; + AV_WN32A(&nnz_cache[4 + 8 * 0], top_empty); + AV_WN32A(&nnz_cache[4 + 8 * 5], top_empty); + AV_WN32A(&nnz_cache[4 + 8 * 10], top_empty); + } + + for (i = 0; i < 2; i++) { + if (left_type[LEFT(i)]) { + nnz = h->non_zero_count[left_xy[LEFT(i)]]; + nnz_cache[3 + 8 * 1 + 2 * 8 * i] = nnz[left_block[8 + 0 + 2 * i]]; + nnz_cache[3 + 8 * 2 + 2 * 8 * i] = nnz[left_block[8 + 1 + 2 * i]]; + if (CHROMA444) { + nnz_cache[3 + 8 * 6 + 2 * 8 * i] = nnz[left_block[8 + 0 + 2 * i] + 4 * 4]; + nnz_cache[3 + 8 * 7 + 2 * 8 * i] = nnz[left_block[8 + 1 + 2 * i] + 4 * 4]; + nnz_cache[3 + 8 * 11 + 2 * 8 * i] = nnz[left_block[8 + 0 + 2 * i] + 8 * 4]; + nnz_cache[3 + 8 * 12 + 2 * 8 * i] = nnz[left_block[8 + 1 + 2 * i] + 8 * 4]; + } else if (CHROMA422) { + nnz_cache[3 + 8 * 6 + 2 * 8 * i] = nnz[left_block[8 + 0 + 2 * i] - 2 + 4 * 4]; + nnz_cache[3 + 8 * 7 + 2 * 8 * i] = nnz[left_block[8 + 1 + 2 * i] - 2 + 4 * 4]; + nnz_cache[3 + 8 * 11 + 2 * 8 * i] = nnz[left_block[8 + 0 + 2 * i] - 2 + 8 * 4]; + nnz_cache[3 + 8 * 12 + 2 * 8 * i] = nnz[left_block[8 + 1 + 2 * i] - 2 + 8 * 4]; + } else { + nnz_cache[3 + 8 * 6 + 8 * i] = nnz[left_block[8 + 4 + 2 * i]]; + nnz_cache[3 + 8 * 11 + 8 * i] = nnz[left_block[8 + 5 + 2 * i]]; + } + } else { + nnz_cache[3 + 8 * 1 + 2 * 8 * i] = + nnz_cache[3 + 8 * 2 + 2 * 8 * i] = + nnz_cache[3 + 8 * 6 + 2 * 8 * i] = + nnz_cache[3 + 8 * 7 + 2 * 8 * i] = + nnz_cache[3 + 8 * 11 + 2 * 8 * i] = + nnz_cache[3 + 8 * 12 + 2 * 8 * i] = CABAC && !IS_INTRA(mb_type) ? 0 : 64; + } + } + + if (CABAC) { + // top_cbp + if (top_type) + h->top_cbp = h->cbp_table[top_xy]; + else + h->top_cbp = IS_INTRA(mb_type) ? 0x7CF : 0x00F; + // left_cbp + if (left_type[LTOP]) { + h->left_cbp = (h->cbp_table[left_xy[LTOP]] & 0x7F0) | + ((h->cbp_table[left_xy[LTOP]] >> (left_block[0] & (~1))) & 2) | + (((h->cbp_table[left_xy[LBOT]] >> (left_block[2] & (~1))) & 2) << 2); + } else { + h->left_cbp = IS_INTRA(mb_type) ? 0x7CF : 0x00F; } - }else{ - nnz_cache[3+8* 1 + 2*8*i]= - nnz_cache[3+8* 2 + 2*8*i]= - nnz_cache[3+8* 6 + 2*8*i]= - nnz_cache[3+8* 7 + 2*8*i]= - nnz_cache[3+8*11 + 2*8*i]= - nnz_cache[3+8*12 + 2*8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64; } } - if( CABAC ) { - // top_cbp - if(top_type) { - h->top_cbp = h->cbp_table[top_xy]; - } else { - h->top_cbp = IS_INTRA(mb_type) ? 0x7CF : 0x00F; - } - // left_cbp - if (left_type[LTOP]) { - h->left_cbp = (h->cbp_table[left_xy[LTOP]] & 0x7F0) - | ((h->cbp_table[left_xy[LTOP]]>>(left_block[0]&(~1)))&2) - | (((h->cbp_table[left_xy[LBOT]]>>(left_block[2]&(~1)))&2) << 2); - } else { - h->left_cbp = IS_INTRA(mb_type) ? 0x7CF : 0x00F; - } - } - } - - if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){ + if (IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)) { int list; int b_stride = h->b_stride; - for(list=0; listlist_count; list++){ + for (list = 0; list < h->list_count; list++) { int8_t *ref_cache = &h->ref_cache[list][scan8[0]]; int8_t *ref = s->current_picture.f.ref_index[list]; - int16_t (*mv_cache)[2] = &h->mv_cache[list][scan8[0]]; - int16_t (*mv)[2] = s->current_picture.f.motion_val[list]; - if(!USES_LIST(mb_type, list)){ + int16_t(*mv_cache)[2] = &h->mv_cache[list][scan8[0]]; + int16_t(*mv)[2] = s->current_picture.f.motion_val[list]; + if (!USES_LIST(mb_type, list)) continue; - } assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)); - if(USES_LIST(top_type, list)){ - const int b_xy= h->mb2b_xy[top_xy] + 3*b_stride; - AV_COPY128(mv_cache[0 - 1*8], mv[b_xy + 0]); - ref_cache[0 - 1*8]= - ref_cache[1 - 1*8]= ref[4*top_xy + 2]; - ref_cache[2 - 1*8]= - ref_cache[3 - 1*8]= ref[4*top_xy + 3]; - }else{ - AV_ZERO128(mv_cache[0 - 1*8]); - AV_WN32A(&ref_cache[0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101u); + if (USES_LIST(top_type, list)) { + const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride; + AV_COPY128(mv_cache[0 - 1 * 8], mv[b_xy + 0]); + ref_cache[0 - 1 * 8] = + ref_cache[1 - 1 * 8] = ref[4 * top_xy + 2]; + ref_cache[2 - 1 * 8] = + ref_cache[3 - 1 * 8] = ref[4 * top_xy + 3]; + } else { + AV_ZERO128(mv_cache[0 - 1 * 8]); + AV_WN32A(&ref_cache[0 - 1 * 8], + ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE) & 0xFF) * 0x01010101u); } - if(mb_type & (MB_TYPE_16x8|MB_TYPE_8x8)){ - for(i=0; i<2; i++){ - int cache_idx = -1 + i*2*8; - if(USES_LIST(left_type[LEFT(i)], list)){ - const int b_xy= h->mb2b_xy[left_xy[LEFT(i)]] + 3; - const int b8_xy= 4*left_xy[LEFT(i)] + 1; - AV_COPY32(mv_cache[cache_idx ], mv[b_xy + b_stride*left_block[0+i*2]]); - AV_COPY32(mv_cache[cache_idx+8], mv[b_xy + b_stride*left_block[1+i*2]]); - ref_cache[cache_idx ]= ref[b8_xy + (left_block[0+i*2]&~1)]; - ref_cache[cache_idx+8]= ref[b8_xy + (left_block[1+i*2]&~1)]; - }else{ - AV_ZERO32(mv_cache[cache_idx ]); - AV_ZERO32(mv_cache[cache_idx+8]); - ref_cache[cache_idx ]= - ref_cache[cache_idx+8]= (left_type[LEFT(i)]) ? LIST_NOT_USED : PART_NOT_AVAILABLE; + if (mb_type & (MB_TYPE_16x8 | MB_TYPE_8x8)) { + for (i = 0; i < 2; i++) { + int cache_idx = -1 + i * 2 * 8; + if (USES_LIST(left_type[LEFT(i)], list)) { + const int b_xy = h->mb2b_xy[left_xy[LEFT(i)]] + 3; + const int b8_xy = 4 * left_xy[LEFT(i)] + 1; + AV_COPY32(mv_cache[cache_idx], + mv[b_xy + b_stride * left_block[0 + i * 2]]); + AV_COPY32(mv_cache[cache_idx + 8], + mv[b_xy + b_stride * left_block[1 + i * 2]]); + ref_cache[cache_idx] = ref[b8_xy + (left_block[0 + i * 2] & ~1)]; + ref_cache[cache_idx + 8] = ref[b8_xy + (left_block[1 + i * 2] & ~1)]; + } else { + AV_ZERO32(mv_cache[cache_idx]); + AV_ZERO32(mv_cache[cache_idx + 8]); + ref_cache[cache_idx] = + ref_cache[cache_idx + 8] = (left_type[LEFT(i)]) ? LIST_NOT_USED + : PART_NOT_AVAILABLE; + } } - } - }else{ - if(USES_LIST(left_type[LTOP], list)){ - const int b_xy= h->mb2b_xy[left_xy[LTOP]] + 3; - const int b8_xy= 4*left_xy[LTOP] + 1; - AV_COPY32(mv_cache[-1], mv[b_xy + b_stride*left_block[0]]); - ref_cache[-1]= ref[b8_xy + (left_block[0]&~1)]; - }else{ + } else { + if (USES_LIST(left_type[LTOP], list)) { + const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3; + const int b8_xy = 4 * left_xy[LTOP] + 1; + AV_COPY32(mv_cache[-1], mv[b_xy + b_stride * left_block[0]]); + ref_cache[-1] = ref[b8_xy + (left_block[0] & ~1)]; + } else { AV_ZERO32(mv_cache[-1]); - ref_cache[-1]= left_type[LTOP] ? LIST_NOT_USED : PART_NOT_AVAILABLE; + ref_cache[-1] = left_type[LTOP] ? LIST_NOT_USED + : PART_NOT_AVAILABLE; } } - if(USES_LIST(topright_type, list)){ - const int b_xy= h->mb2b_xy[topright_xy] + 3*b_stride; - AV_COPY32(mv_cache[4 - 1*8], mv[b_xy]); - ref_cache[4 - 1*8]= ref[4*topright_xy + 2]; - }else{ - AV_ZERO32(mv_cache[4 - 1*8]); - ref_cache[4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; + if (USES_LIST(topright_type, list)) { + const int b_xy = h->mb2b_xy[topright_xy] + 3 * b_stride; + AV_COPY32(mv_cache[4 - 1 * 8], mv[b_xy]); + ref_cache[4 - 1 * 8] = ref[4 * topright_xy + 2]; + } else { + AV_ZERO32(mv_cache[4 - 1 * 8]); + ref_cache[4 - 1 * 8] = topright_type ? LIST_NOT_USED + : PART_NOT_AVAILABLE; } - 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); - AV_COPY32(mv_cache[-1 - 1*8], mv[b_xy]); - ref_cache[-1 - 1*8]= ref[b8_xy]; - }else{ - AV_ZERO32(mv_cache[-1 - 1*8]); - ref_cache[-1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; + 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); + AV_COPY32(mv_cache[-1 - 1 * 8], mv[b_xy]); + ref_cache[-1 - 1 * 8] = ref[b8_xy]; + } else { + AV_ZERO32(mv_cache[-1 - 1 * 8]); + ref_cache[-1 - 1 * 8] = topleft_type ? LIST_NOT_USED + : PART_NOT_AVAILABLE; } } - if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF) + if ((mb_type & (MB_TYPE_SKIP | MB_TYPE_DIRECT2)) && !FRAME_MBAFF) continue; - if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))){ - uint8_t (*mvd_cache)[2] = &h->mvd_cache[list][scan8[0]]; - uint8_t (*mvd)[2] = h->mvd_table[list]; - ref_cache[2+8*0] = - ref_cache[2+8*2] = PART_NOT_AVAILABLE; - AV_ZERO32(mv_cache[2+8*0]); - AV_ZERO32(mv_cache[2+8*2]); + if (!(mb_type & (MB_TYPE_SKIP | MB_TYPE_DIRECT2))) { + uint8_t(*mvd_cache)[2] = &h->mvd_cache[list][scan8[0]]; + uint8_t(*mvd)[2] = h->mvd_table[list]; + ref_cache[2 + 8 * 0] = + ref_cache[2 + 8 * 2] = PART_NOT_AVAILABLE; + AV_ZERO32(mv_cache[2 + 8 * 0]); + AV_ZERO32(mv_cache[2 + 8 * 2]); - if( CABAC ) { - if(USES_LIST(top_type, list)){ - const int b_xy= h->mb2br_xy[top_xy]; - AV_COPY64(mvd_cache[0 - 1*8], mvd[b_xy + 0]); - }else{ - AV_ZERO64(mvd_cache[0 - 1*8]); + if (CABAC) { + if (USES_LIST(top_type, list)) { + const int b_xy = h->mb2br_xy[top_xy]; + AV_COPY64(mvd_cache[0 - 1 * 8], mvd[b_xy + 0]); + } else { + AV_ZERO64(mvd_cache[0 - 1 * 8]); } - if(USES_LIST(left_type[LTOP], list)){ - const int b_xy= h->mb2br_xy[left_xy[LTOP]] + 6; - AV_COPY16(mvd_cache[-1 + 0*8], mvd[b_xy - left_block[0]]); - AV_COPY16(mvd_cache[-1 + 1*8], mvd[b_xy - left_block[1]]); - }else{ - AV_ZERO16(mvd_cache[-1 + 0*8]); - AV_ZERO16(mvd_cache[-1 + 1*8]); + if (USES_LIST(left_type[LTOP], list)) { + const int b_xy = h->mb2br_xy[left_xy[LTOP]] + 6; + AV_COPY16(mvd_cache[-1 + 0 * 8], mvd[b_xy - left_block[0]]); + AV_COPY16(mvd_cache[-1 + 1 * 8], mvd[b_xy - left_block[1]]); + } else { + AV_ZERO16(mvd_cache[-1 + 0 * 8]); + AV_ZERO16(mvd_cache[-1 + 1 * 8]); } - if(USES_LIST(left_type[LBOT], list)){ - const int b_xy= h->mb2br_xy[left_xy[LBOT]] + 6; - AV_COPY16(mvd_cache[-1 + 2*8], mvd[b_xy - left_block[2]]); - AV_COPY16(mvd_cache[-1 + 3*8], mvd[b_xy - left_block[3]]); - }else{ - AV_ZERO16(mvd_cache[-1 + 2*8]); - AV_ZERO16(mvd_cache[-1 + 3*8]); + if (USES_LIST(left_type[LBOT], list)) { + const int b_xy = h->mb2br_xy[left_xy[LBOT]] + 6; + AV_COPY16(mvd_cache[-1 + 2 * 8], mvd[b_xy - left_block[2]]); + AV_COPY16(mvd_cache[-1 + 3 * 8], mvd[b_xy - left_block[3]]); + } else { + AV_ZERO16(mvd_cache[-1 + 2 * 8]); + AV_ZERO16(mvd_cache[-1 + 3 * 8]); } - AV_ZERO16(mvd_cache[2+8*0]); - AV_ZERO16(mvd_cache[2+8*2]); - if(h->slice_type_nos == AV_PICTURE_TYPE_B){ + AV_ZERO16(mvd_cache[2 + 8 * 0]); + AV_ZERO16(mvd_cache[2 + 8 * 2]); + if (h->slice_type_nos == AV_PICTURE_TYPE_B) { uint8_t *direct_cache = &h->direct_cache[scan8[0]]; uint8_t *direct_table = h->direct_table; - fill_rectangle(direct_cache, 4, 4, 8, MB_TYPE_16x16>>1, 1); + fill_rectangle(direct_cache, 4, 4, 8, MB_TYPE_16x16 >> 1, 1); - if(IS_DIRECT(top_type)){ - AV_WN32A(&direct_cache[-1*8], 0x01010101u*(MB_TYPE_DIRECT2>>1)); - }else if(IS_8X8(top_type)){ - int b8_xy = 4*top_xy; - direct_cache[0 - 1*8]= direct_table[b8_xy + 2]; - direct_cache[2 - 1*8]= direct_table[b8_xy + 3]; - }else{ - AV_WN32A(&direct_cache[-1*8], 0x01010101*(MB_TYPE_16x16>>1)); + if (IS_DIRECT(top_type)) { + AV_WN32A(&direct_cache[-1 * 8], + 0x01010101u * (MB_TYPE_DIRECT2 >> 1)); + } else if (IS_8X8(top_type)) { + int b8_xy = 4 * top_xy; + direct_cache[0 - 1 * 8] = direct_table[b8_xy + 2]; + direct_cache[2 - 1 * 8] = direct_table[b8_xy + 3]; + } else { + AV_WN32A(&direct_cache[-1 * 8], + 0x01010101 * (MB_TYPE_16x16 >> 1)); } - if(IS_DIRECT(left_type[LTOP])) - direct_cache[-1 + 0*8]= MB_TYPE_DIRECT2>>1; - else if(IS_8X8(left_type[LTOP])) - direct_cache[-1 + 0*8]= direct_table[4*left_xy[LTOP] + 1 + (left_block[0]&~1)]; + if (IS_DIRECT(left_type[LTOP])) + direct_cache[-1 + 0 * 8] = MB_TYPE_DIRECT2 >> 1; + else if (IS_8X8(left_type[LTOP])) + direct_cache[-1 + 0 * 8] = direct_table[4 * left_xy[LTOP] + 1 + (left_block[0] & ~1)]; else - direct_cache[-1 + 0*8]= MB_TYPE_16x16>>1; + direct_cache[-1 + 0 * 8] = MB_TYPE_16x16 >> 1; - if(IS_DIRECT(left_type[LBOT])) - direct_cache[-1 + 2*8]= MB_TYPE_DIRECT2>>1; - else if(IS_8X8(left_type[LBOT])) - direct_cache[-1 + 2*8]= direct_table[4*left_xy[LBOT] + 1 + (left_block[2]&~1)]; + if (IS_DIRECT(left_type[LBOT])) + direct_cache[-1 + 2 * 8] = MB_TYPE_DIRECT2 >> 1; + else if (IS_8X8(left_type[LBOT])) + direct_cache[-1 + 2 * 8] = direct_table[4 * left_xy[LBOT] + 1 + (left_block[2] & ~1)]; else - direct_cache[-1 + 2*8]= MB_TYPE_16x16>>1; + direct_cache[-1 + 2 * 8] = MB_TYPE_16x16 >> 1; } } } - if(FRAME_MBAFF){ -#define MAP_MVS\ - MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\ - MAP_F2F(scan8[0] + 0 - 1*8, top_type)\ - MAP_F2F(scan8[0] + 1 - 1*8, top_type)\ - MAP_F2F(scan8[0] + 2 - 1*8, top_type)\ - MAP_F2F(scan8[0] + 3 - 1*8, top_type)\ - MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\ - MAP_F2F(scan8[0] - 1 + 0*8, left_type[LTOP])\ - MAP_F2F(scan8[0] - 1 + 1*8, left_type[LTOP])\ - MAP_F2F(scan8[0] - 1 + 2*8, left_type[LBOT])\ - MAP_F2F(scan8[0] - 1 + 3*8, left_type[LBOT]) - if(MB_FIELD){ -#define MAP_F2F(idx, mb_type)\ - if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\ - h->ref_cache[list][idx] <<= 1;\ - h->mv_cache[list][idx][1] /= 2;\ - h->mvd_cache[list][idx][1] >>=1;\ - } + +#define MAP_MVS \ + MAP_F2F(scan8[0] - 1 - 1 * 8, topleft_type) \ + MAP_F2F(scan8[0] + 0 - 1 * 8, top_type) \ + MAP_F2F(scan8[0] + 1 - 1 * 8, top_type) \ + MAP_F2F(scan8[0] + 2 - 1 * 8, top_type) \ + MAP_F2F(scan8[0] + 3 - 1 * 8, top_type) \ + MAP_F2F(scan8[0] + 4 - 1 * 8, topright_type) \ + MAP_F2F(scan8[0] - 1 + 0 * 8, left_type[LTOP]) \ + MAP_F2F(scan8[0] - 1 + 1 * 8, left_type[LTOP]) \ + MAP_F2F(scan8[0] - 1 + 2 * 8, left_type[LBOT]) \ + MAP_F2F(scan8[0] - 1 + 3 * 8, left_type[LBOT]) + + if (FRAME_MBAFF) { + if (MB_FIELD) { + +#define MAP_F2F(idx, mb_type) \ + if (!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0) { \ + h->ref_cache[list][idx] <<= 1; \ + h->mv_cache[list][idx][1] /= 2; \ + h->mvd_cache[list][idx][1] >>= 1; \ + } + MAP_MVS + } else { + #undef MAP_F2F - }else{ -#define MAP_F2F(idx, mb_type)\ - if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\ - h->ref_cache[list][idx] >>= 1;\ - h->mv_cache[list][idx][1] <<= 1;\ - h->mvd_cache[list][idx][1] <<= 1;\ - } +#define MAP_F2F(idx, mb_type) \ + if (IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0) { \ + h->ref_cache[list][idx] >>= 1; \ + h->mv_cache[list][idx][1] <<= 1; \ + h->mvd_cache[list][idx][1] <<= 1; \ + } + MAP_MVS #undef MAP_F2F } @@ -747,36 +792,34 @@ static void fill_decode_caches(H264Context *h, int mb_type){ } } - h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[LTOP]); + h->neighbor_transform_size = !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[LTOP]); } /** * decodes a P_SKIP or B_SKIP macroblock */ -static void av_unused decode_mb_skip(H264Context *h){ - MpegEncContext * const s = &h->s; - const int mb_xy= h->mb_xy; - int mb_type=0; +static void av_unused decode_mb_skip(H264Context *h) +{ + MpegEncContext *const s = &h->s; + const int mb_xy = h->mb_xy; + int mb_type = 0; memset(h->non_zero_count[mb_xy], 0, 48); - if(MB_FIELD) - mb_type|= MB_TYPE_INTERLACED; + if (MB_FIELD) + mb_type |= MB_TYPE_INTERLACED; - if( h->slice_type_nos == AV_PICTURE_TYPE_B ) - { + if (h->slice_type_nos == AV_PICTURE_TYPE_B) { // 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){ + 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_caches(h, 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; + 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); @@ -785,8 +828,8 @@ static void av_unused decode_mb_skip(H264Context *h){ write_back_motion(h, mb_type); s->current_picture.f.mb_type[mb_xy] = mb_type; s->current_picture.f.qscale_table[mb_xy] = s->qscale; - h->slice_table[ mb_xy ]= h->slice_num; - h->prev_mb_skipped= 1; + h->slice_table[mb_xy] = h->slice_num; + h->prev_mb_skipped = 1; } #endif /* AVCODEC_H264_MVPRED_H */ diff --git a/libavcodec/h264dsp.h b/libavcodec/h264dsp.h index 08a25a54c9..248c7d0e08 100644 --- a/libavcodec/h264dsp.h +++ b/libavcodec/h264dsp.h @@ -28,56 +28,90 @@ #define AVCODEC_H264DSP_H #include + #include "dsputil.h" typedef void (*h264_weight_func)(uint8_t *block, int stride, int height, int log2_denom, int weight, int offset); -typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int height, - int log2_denom, int weightd, int weights, int offset); +typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, + int stride, int height, int log2_denom, + int weightd, int weights, int offset); /** * Context for storing H.264 DSP functions */ -typedef struct H264DSPContext{ +typedef struct H264DSPContext { /* weighted MC */ h264_weight_func weight_h264_pixels_tab[4]; h264_biweight_func biweight_h264_pixels_tab[4]; /* loop filter */ - void (*h264_v_loop_filter_luma)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0); - void (*h264_h_loop_filter_luma)(uint8_t *pix/*align 4 */, int stride, int alpha, int beta, int8_t *tc0); - void (*h264_h_loop_filter_luma_mbaff)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0); + void (*h264_v_loop_filter_luma)(uint8_t *pix /*align 16*/, int stride, + int alpha, int beta, int8_t *tc0); + void (*h264_h_loop_filter_luma)(uint8_t *pix /*align 4 */, int stride, + int alpha, int beta, int8_t *tc0); + void (*h264_h_loop_filter_luma_mbaff)(uint8_t *pix /*align 16*/, int stride, + int alpha, int beta, int8_t *tc0); /* v/h_loop_filter_luma_intra: align 16 */ - void (*h264_v_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta); - void (*h264_h_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta); - void (*h264_h_loop_filter_luma_mbaff_intra)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta); - void (*h264_v_loop_filter_chroma)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0); - void (*h264_h_loop_filter_chroma)(uint8_t *pix/*align 4*/, int stride, int alpha, int beta, int8_t *tc0); - void (*h264_h_loop_filter_chroma_mbaff)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0); - void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta); - void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta); - void (*h264_h_loop_filter_chroma_mbaff_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta); + void (*h264_v_loop_filter_luma_intra)(uint8_t *pix, int stride, + int alpha, int beta); + void (*h264_h_loop_filter_luma_intra)(uint8_t *pix, int stride, + int alpha, int beta); + void (*h264_h_loop_filter_luma_mbaff_intra)(uint8_t *pix /*align 16*/, + int stride, int alpha, int beta); + void (*h264_v_loop_filter_chroma)(uint8_t *pix /*align 8*/, int stride, + int alpha, int beta, int8_t *tc0); + void (*h264_h_loop_filter_chroma)(uint8_t *pix /*align 4*/, int stride, + int alpha, int beta, int8_t *tc0); + void (*h264_h_loop_filter_chroma_mbaff)(uint8_t *pix /*align 8*/, + int stride, int alpha, int beta, + int8_t *tc0); + void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix /*align 8*/, + int stride, int alpha, int beta); + void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix /*align 8*/, + int stride, int alpha, int beta); + void (*h264_h_loop_filter_chroma_mbaff_intra)(uint8_t *pix /*align 8*/, + int stride, int alpha, int beta); // h264_loop_filter_strength: simd only. the C version is inlined in h264.c - void (*h264_loop_filter_strength)(int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2], - int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field); + void (*h264_loop_filter_strength)(int16_t bS[2][4][4], uint8_t nnz[40], + int8_t ref[2][40], int16_t mv[2][40][2], + int bidir, int edges, int step, + int mask_mv0, int mask_mv1, int field); /* IDCT */ - void (*h264_idct_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride); - void (*h264_idct8_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride); - void (*h264_idct_dc_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride); - void (*h264_idct8_dc_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride); + void (*h264_idct_add)(uint8_t *dst /*align 4*/, + DCTELEM *block /*align 16*/, int stride); + void (*h264_idct8_add)(uint8_t *dst /*align 8*/, + DCTELEM *block /*align 16*/, int stride); + void (*h264_idct_dc_add)(uint8_t *dst /*align 4*/, + DCTELEM *block /*align 16*/, int stride); + void (*h264_idct8_dc_add)(uint8_t *dst /*align 8*/, + DCTELEM *block /*align 16*/, int stride); - void (*h264_idct_add16)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[15*8]); - void (*h264_idct8_add4)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[15*8]); - void (*h264_idct_add8)(uint8_t **dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[15*8]); - void (*h264_idct_add16intra)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[15*8]); - void (*h264_luma_dc_dequant_idct)(DCTELEM *output, DCTELEM *input/*align 16*/, int qmul); + void (*h264_idct_add16)(uint8_t *dst /*align 16*/, const int *blockoffset, + DCTELEM *block /*align 16*/, int stride, + const uint8_t nnzc[15 * 8]); + void (*h264_idct8_add4)(uint8_t *dst /*align 16*/, const int *blockoffset, + DCTELEM *block /*align 16*/, int stride, + const uint8_t nnzc[15 * 8]); + void (*h264_idct_add8)(uint8_t **dst /*align 16*/, const int *blockoffset, + DCTELEM *block /*align 16*/, int stride, + const uint8_t nnzc[15 * 8]); + void (*h264_idct_add16intra)(uint8_t *dst /*align 16*/, const int *blockoffset, + DCTELEM *block /*align 16*/, + int stride, const uint8_t nnzc[15 * 8]); + void (*h264_luma_dc_dequant_idct)(DCTELEM *output, + DCTELEM *input /*align 16*/, int qmul); void (*h264_chroma_dc_dequant_idct)(DCTELEM *block, int qmul); -}H264DSPContext; +} H264DSPContext; -void ff_h264dsp_init(H264DSPContext *c, const int bit_depth, const int chroma_format_idc); -void ff_h264dsp_init_arm(H264DSPContext *c, const int bit_depth, const int chroma_format_idc); -void ff_h264dsp_init_ppc(H264DSPContext *c, const int bit_depth, const int chroma_format_idc); -void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth, const int chroma_format_idc); +void ff_h264dsp_init(H264DSPContext *c, const int bit_depth, + const int chroma_format_idc); +void ff_h264dsp_init_arm(H264DSPContext *c, const int bit_depth, + const int chroma_format_idc); +void ff_h264dsp_init_ppc(H264DSPContext *c, const int bit_depth, + const int chroma_format_idc); +void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth, + const int chroma_format_idc); #endif /* AVCODEC_H264DSP_H */