yading@10: /* yading@10: * H.26L/H.264/AVC/JVT/14496-10/... motion vector predicion yading@10: * Copyright (c) 2003 Michael Niedermayer yading@10: * yading@10: * This file is part of FFmpeg. yading@10: * yading@10: * FFmpeg is free software; you can redistribute it and/or yading@10: * modify it under the terms of the GNU Lesser General Public yading@10: * License as published by the Free Software Foundation; either yading@10: * version 2.1 of the License, or (at your option) any later version. yading@10: * yading@10: * FFmpeg is distributed in the hope that it will be useful, yading@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of yading@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU yading@10: * Lesser General Public License for more details. yading@10: * yading@10: * You should have received a copy of the GNU Lesser General Public yading@10: * License along with FFmpeg; if not, write to the Free Software yading@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA yading@10: */ yading@10: yading@10: /** yading@10: * @file yading@10: * H.264 / AVC / MPEG4 part10 motion vector predicion. yading@10: * @author Michael Niedermayer yading@10: */ yading@10: yading@10: #ifndef AVCODEC_H264_MVPRED_H yading@10: #define AVCODEC_H264_MVPRED_H yading@10: yading@10: #include "internal.h" yading@10: #include "avcodec.h" yading@10: #include "h264.h" yading@10: #include "libavutil/avassert.h" yading@10: yading@10: yading@10: static av_always_inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, yading@10: int i, int list, int part_width) yading@10: { yading@10: const int topright_ref = h->ref_cache[list][i - 8 + part_width]; yading@10: yading@10: /* there is no consistent mapping of mvs to neighboring locations that will yading@10: * make mbaff happy, so we can't move all this logic to fill_caches */ yading@10: if (FRAME_MBAFF(h)) { yading@10: #define SET_DIAG_MV(MV_OP, REF_OP, XY, Y4) \ yading@10: const int xy = XY, y4 = Y4; \ yading@10: const int mb_type = mb_types[xy + (y4 >> 2) * h->mb_stride]; \ yading@10: if (!USES_LIST(mb_type, list)) \ yading@10: return LIST_NOT_USED; \ yading@10: mv = h->cur_pic_ptr->motion_val[list][h->mb2b_xy[xy] + 3 + y4 * h->b_stride]; \ yading@10: h->mv_cache[list][scan8[0] - 2][0] = mv[0]; \ yading@10: h->mv_cache[list][scan8[0] - 2][1] = mv[1] MV_OP; \ yading@10: return h->cur_pic_ptr->ref_index[list][4 * xy + 1 + (y4 & ~1)] REF_OP; yading@10: yading@10: if (topright_ref == PART_NOT_AVAILABLE yading@10: && i >= scan8[0] + 8 && (i & 7) == 4 yading@10: && h->ref_cache[list][scan8[0] - 1] != PART_NOT_AVAILABLE) { yading@10: const uint32_t *mb_types = h->cur_pic_ptr->mb_type; yading@10: const int16_t *mv; yading@10: AV_ZERO32(h->mv_cache[list][scan8[0] - 2]); yading@10: *C = h->mv_cache[list][scan8[0] - 2]; yading@10: yading@10: if (!MB_FIELD(h) && IS_INTERLACED(h->left_type[0])) { yading@10: SET_DIAG_MV(* 2, >> 1, h->left_mb_xy[0] + h->mb_stride, yading@10: (h->mb_y & 1) * 2 + (i >> 5)); yading@10: } yading@10: if (MB_FIELD(h) && !IS_INTERLACED(h->left_type[0])) { yading@10: // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK. yading@10: SET_DIAG_MV(/ 2, << 1, h->left_mb_xy[i >= 36], ((i >> 2)) & 3); yading@10: } yading@10: } yading@10: #undef SET_DIAG_MV yading@10: } yading@10: yading@10: if (topright_ref != PART_NOT_AVAILABLE) { yading@10: *C = h->mv_cache[list][i - 8 + part_width]; yading@10: return topright_ref; yading@10: } else { yading@10: tprintf(h->avctx, "topright MV not available\n"); yading@10: yading@10: *C = h->mv_cache[list][i - 8 - 1]; yading@10: return h->ref_cache[list][i - 8 - 1]; yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * Get the predicted MV. yading@10: * @param n the block index yading@10: * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4) yading@10: * @param mx the x component of the predicted motion vector yading@10: * @param my the y component of the predicted motion vector yading@10: */ yading@10: static av_always_inline void pred_motion(H264Context *const h, int n, yading@10: int part_width, int list, int ref, yading@10: int *const mx, int *const my) yading@10: { yading@10: const int index8 = scan8[n]; yading@10: const int top_ref = h->ref_cache[list][index8 - 8]; yading@10: const int left_ref = h->ref_cache[list][index8 - 1]; yading@10: const int16_t *const A = h->mv_cache[list][index8 - 1]; yading@10: const int16_t *const B = h->mv_cache[list][index8 - 8]; yading@10: const int16_t *C; yading@10: int diagonal_ref, match_count; yading@10: yading@10: av_assert2(part_width == 1 || part_width == 2 || part_width == 4); yading@10: yading@10: /* mv_cache yading@10: * B . . A T T T T yading@10: * U . . L . . , . yading@10: * U . . L . . . . yading@10: * U . . L . . , . yading@10: * . . . L . . . . yading@10: */ yading@10: yading@10: diagonal_ref = fetch_diagonal_mv(h, &C, index8, list, part_width); yading@10: match_count = (diagonal_ref == ref) + (top_ref == ref) + (left_ref == ref); yading@10: tprintf(h->avctx, "pred_motion match_count=%d\n", match_count); yading@10: if (match_count > 1) { //most common yading@10: *mx = mid_pred(A[0], B[0], C[0]); yading@10: *my = mid_pred(A[1], B[1], C[1]); yading@10: } else if (match_count == 1) { yading@10: if (left_ref == ref) { yading@10: *mx = A[0]; yading@10: *my = A[1]; yading@10: } else if (top_ref == ref) { yading@10: *mx = B[0]; yading@10: *my = B[1]; yading@10: } else { yading@10: *mx = C[0]; yading@10: *my = C[1]; yading@10: } yading@10: } else { yading@10: if (top_ref == PART_NOT_AVAILABLE && yading@10: diagonal_ref == PART_NOT_AVAILABLE && yading@10: left_ref != PART_NOT_AVAILABLE) { yading@10: *mx = A[0]; yading@10: *my = A[1]; yading@10: } else { yading@10: *mx = mid_pred(A[0], B[0], C[0]); yading@10: *my = mid_pred(A[1], B[1], C[1]); yading@10: } yading@10: } yading@10: yading@10: tprintf(h->avctx, yading@10: "pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", yading@10: top_ref, B[0], B[1], diagonal_ref, C[0], C[1], left_ref, yading@10: A[0], A[1], ref, *mx, *my, h->mb_x, h->mb_y, n, list); yading@10: } yading@10: yading@10: /** yading@10: * Get the directionally predicted 16x8 MV. yading@10: * @param n the block index yading@10: * @param mx the x component of the predicted motion vector yading@10: * @param my the y component of the predicted motion vector yading@10: */ yading@10: static av_always_inline void pred_16x8_motion(H264Context *const h, yading@10: int n, int list, int ref, yading@10: int *const mx, int *const my) yading@10: { yading@10: if (n == 0) { yading@10: const int top_ref = h->ref_cache[list][scan8[0] - 8]; yading@10: const int16_t *const B = h->mv_cache[list][scan8[0] - 8]; yading@10: yading@10: tprintf(h->avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", yading@10: top_ref, B[0], B[1], h->mb_x, h->mb_y, n, list); yading@10: yading@10: if (top_ref == ref) { yading@10: *mx = B[0]; yading@10: *my = B[1]; yading@10: return; yading@10: } yading@10: } else { yading@10: const int left_ref = h->ref_cache[list][scan8[8] - 1]; yading@10: const int16_t *const A = h->mv_cache[list][scan8[8] - 1]; yading@10: yading@10: tprintf(h->avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", yading@10: left_ref, A[0], A[1], h->mb_x, h->mb_y, n, list); yading@10: yading@10: if (left_ref == ref) { yading@10: *mx = A[0]; yading@10: *my = A[1]; yading@10: return; yading@10: } yading@10: } yading@10: yading@10: //RARE yading@10: pred_motion(h, n, 4, list, ref, mx, my); yading@10: } yading@10: yading@10: /** yading@10: * Get the directionally predicted 8x16 MV. yading@10: * @param n the block index yading@10: * @param mx the x component of the predicted motion vector yading@10: * @param my the y component of the predicted motion vector yading@10: */ yading@10: static av_always_inline void pred_8x16_motion(H264Context *const h, yading@10: int n, int list, int ref, yading@10: int *const mx, int *const my) yading@10: { yading@10: if (n == 0) { yading@10: const int left_ref = h->ref_cache[list][scan8[0] - 1]; yading@10: const int16_t *const A = h->mv_cache[list][scan8[0] - 1]; yading@10: yading@10: tprintf(h->avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", yading@10: left_ref, A[0], A[1], h->mb_x, h->mb_y, n, list); yading@10: yading@10: if (left_ref == ref) { yading@10: *mx = A[0]; yading@10: *my = A[1]; yading@10: return; yading@10: } yading@10: } else { yading@10: const int16_t *C; yading@10: int diagonal_ref; yading@10: yading@10: diagonal_ref = fetch_diagonal_mv(h, &C, scan8[4], list, 2); yading@10: yading@10: tprintf(h->avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", yading@10: diagonal_ref, C[0], C[1], h->mb_x, h->mb_y, n, list); yading@10: yading@10: if (diagonal_ref == ref) { yading@10: *mx = C[0]; yading@10: *my = C[1]; yading@10: return; yading@10: } yading@10: } yading@10: yading@10: //RARE yading@10: pred_motion(h, n, 2, list, ref, mx, my); yading@10: } yading@10: yading@10: #define FIX_MV_MBAFF(type, refn, mvn, idx) \ yading@10: if (FRAME_MBAFF(h)) { \ yading@10: if (MB_FIELD(h)) { \ yading@10: if (!IS_INTERLACED(type)) { \ yading@10: refn <<= 1; \ yading@10: AV_COPY32(mvbuf[idx], mvn); \ yading@10: mvbuf[idx][1] /= 2; \ yading@10: mvn = mvbuf[idx]; \ yading@10: } \ yading@10: } else { \ yading@10: if (IS_INTERLACED(type)) { \ yading@10: refn >>= 1; \ yading@10: AV_COPY32(mvbuf[idx], mvn); \ yading@10: mvbuf[idx][1] <<= 1; \ yading@10: mvn = mvbuf[idx]; \ yading@10: } \ yading@10: } \ yading@10: } yading@10: yading@10: static av_always_inline void pred_pskip_motion(H264Context *const h) yading@10: { yading@10: DECLARE_ALIGNED(4, static const int16_t, zeromv)[2] = { 0 }; yading@10: DECLARE_ALIGNED(4, int16_t, mvbuf)[3][2]; yading@10: int8_t *ref = h->cur_pic.ref_index[0]; yading@10: int16_t(*mv)[2] = h->cur_pic.motion_val[0]; yading@10: int top_ref, left_ref, diagonal_ref, match_count, mx, my; yading@10: const int16_t *A, *B, *C; yading@10: int b_stride = h->b_stride; yading@10: yading@10: fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1); yading@10: yading@10: /* To avoid doing an entire fill_decode_caches, we inline the relevant yading@10: * parts here. yading@10: * FIXME: this is a partial duplicate of the logic in fill_decode_caches, yading@10: * but it's faster this way. Is there a way to avoid this duplication? yading@10: */ yading@10: if (USES_LIST(h->left_type[LTOP], 0)) { yading@10: left_ref = ref[4 * h->left_mb_xy[LTOP] + 1 + (h->left_block[0] & ~1)]; yading@10: A = mv[h->mb2b_xy[h->left_mb_xy[LTOP]] + 3 + b_stride * h->left_block[0]]; yading@10: FIX_MV_MBAFF(h->left_type[LTOP], left_ref, A, 0); yading@10: if (!(left_ref | AV_RN32A(A))) yading@10: goto zeromv; yading@10: } else if (h->left_type[LTOP]) { yading@10: left_ref = LIST_NOT_USED; yading@10: A = zeromv; yading@10: } else { yading@10: goto zeromv; yading@10: } yading@10: yading@10: if (USES_LIST(h->top_type, 0)) { yading@10: top_ref = ref[4 * h->top_mb_xy + 2]; yading@10: B = mv[h->mb2b_xy[h->top_mb_xy] + 3 * b_stride]; yading@10: FIX_MV_MBAFF(h->top_type, top_ref, B, 1); yading@10: if (!(top_ref | AV_RN32A(B))) yading@10: goto zeromv; yading@10: } else if (h->top_type) { yading@10: top_ref = LIST_NOT_USED; yading@10: B = zeromv; yading@10: } else { yading@10: goto zeromv; yading@10: } yading@10: yading@10: tprintf(h->avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", yading@10: top_ref, left_ref, h->mb_x, h->mb_y); yading@10: yading@10: if (USES_LIST(h->topright_type, 0)) { yading@10: diagonal_ref = ref[4 * h->topright_mb_xy + 2]; yading@10: C = mv[h->mb2b_xy[h->topright_mb_xy] + 3 * b_stride]; yading@10: FIX_MV_MBAFF(h->topright_type, diagonal_ref, C, 2); yading@10: } else if (h->topright_type) { yading@10: diagonal_ref = LIST_NOT_USED; yading@10: C = zeromv; yading@10: } else { yading@10: if (USES_LIST(h->topleft_type, 0)) { yading@10: diagonal_ref = ref[4 * h->topleft_mb_xy + 1 + yading@10: (h->topleft_partition & 2)]; yading@10: C = mv[h->mb2b_xy[h->topleft_mb_xy] + 3 + b_stride + yading@10: (h->topleft_partition & 2 * b_stride)]; yading@10: FIX_MV_MBAFF(h->topleft_type, diagonal_ref, C, 2); yading@10: } else if (h->topleft_type) { yading@10: diagonal_ref = LIST_NOT_USED; yading@10: C = zeromv; yading@10: } else { yading@10: diagonal_ref = PART_NOT_AVAILABLE; yading@10: C = zeromv; yading@10: } yading@10: } yading@10: yading@10: match_count = !diagonal_ref + !top_ref + !left_ref; yading@10: tprintf(h->avctx, "pred_pskip_motion match_count=%d\n", match_count); yading@10: if (match_count > 1) { yading@10: mx = mid_pred(A[0], B[0], C[0]); yading@10: my = mid_pred(A[1], B[1], C[1]); yading@10: } else if (match_count == 1) { yading@10: if (!left_ref) { yading@10: mx = A[0]; yading@10: my = A[1]; yading@10: } else if (!top_ref) { yading@10: mx = B[0]; yading@10: my = B[1]; yading@10: } else { yading@10: mx = C[0]; yading@10: my = C[1]; yading@10: } yading@10: } else { yading@10: mx = mid_pred(A[0], B[0], C[0]); yading@10: my = mid_pred(A[1], B[1], C[1]); yading@10: } yading@10: yading@10: fill_rectangle(h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx, my), 4); yading@10: return; yading@10: yading@10: zeromv: yading@10: fill_rectangle(h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4); yading@10: return; yading@10: } yading@10: yading@10: static void fill_decode_neighbors(H264Context *h, int mb_type) yading@10: { yading@10: const int mb_xy = h->mb_xy; yading@10: int topleft_xy, top_xy, topright_xy, left_xy[LEFT_MBS]; yading@10: static const uint8_t left_block_options[4][32] = { yading@10: { 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 }, yading@10: { 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 }, yading@10: { 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 }, yading@10: { 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 } yading@10: }; yading@10: yading@10: h->topleft_partition = -1; yading@10: yading@10: top_xy = mb_xy - (h->mb_stride << MB_FIELD(h)); yading@10: yading@10: /* Wow, what a mess, why didn't they simplify the interlacing & intra yading@10: * stuff, I can't imagine that these complex rules are worth it. */ yading@10: yading@10: topleft_xy = top_xy - 1; yading@10: topright_xy = top_xy + 1; yading@10: left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1; yading@10: h->left_block = left_block_options[0]; yading@10: if (FRAME_MBAFF(h)) { yading@10: const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]); yading@10: const int curr_mb_field_flag = IS_INTERLACED(mb_type); yading@10: if (h->mb_y & 1) { yading@10: if (left_mb_field_flag != curr_mb_field_flag) { yading@10: left_xy[LBOT] = left_xy[LTOP] = mb_xy - h->mb_stride - 1; yading@10: if (curr_mb_field_flag) { yading@10: left_xy[LBOT] += h->mb_stride; yading@10: h->left_block = left_block_options[3]; yading@10: } else { yading@10: topleft_xy += h->mb_stride; yading@10: /* take top left mv from the middle of the mb, as opposed yading@10: * to all other modes which use the bottom right partition */ yading@10: h->topleft_partition = 0; yading@10: h->left_block = left_block_options[1]; yading@10: } yading@10: } yading@10: } else { yading@10: if (curr_mb_field_flag) { yading@10: topleft_xy += h->mb_stride & (((h->cur_pic.mb_type[top_xy - 1] >> 7) & 1) - 1); yading@10: topright_xy += h->mb_stride & (((h->cur_pic.mb_type[top_xy + 1] >> 7) & 1) - 1); yading@10: top_xy += h->mb_stride & (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1); yading@10: } yading@10: if (left_mb_field_flag != curr_mb_field_flag) { yading@10: if (curr_mb_field_flag) { yading@10: left_xy[LBOT] += h->mb_stride; yading@10: h->left_block = left_block_options[3]; yading@10: } else { yading@10: h->left_block = left_block_options[2]; yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: h->topleft_mb_xy = topleft_xy; yading@10: h->top_mb_xy = top_xy; yading@10: h->topright_mb_xy = topright_xy; yading@10: h->left_mb_xy[LTOP] = left_xy[LTOP]; yading@10: h->left_mb_xy[LBOT] = left_xy[LBOT]; yading@10: //FIXME do we need all in the context? yading@10: yading@10: h->topleft_type = h->cur_pic.mb_type[topleft_xy]; yading@10: h->top_type = h->cur_pic.mb_type[top_xy]; yading@10: h->topright_type = h->cur_pic.mb_type[topright_xy]; yading@10: h->left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]]; yading@10: h->left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]]; yading@10: yading@10: if (FMO) { yading@10: if (h->slice_table[topleft_xy] != h->slice_num) yading@10: h->topleft_type = 0; yading@10: if (h->slice_table[top_xy] != h->slice_num) yading@10: h->top_type = 0; yading@10: if (h->slice_table[left_xy[LTOP]] != h->slice_num) yading@10: h->left_type[LTOP] = h->left_type[LBOT] = 0; yading@10: } else { yading@10: if (h->slice_table[topleft_xy] != h->slice_num) { yading@10: h->topleft_type = 0; yading@10: if (h->slice_table[top_xy] != h->slice_num) yading@10: h->top_type = 0; yading@10: if (h->slice_table[left_xy[LTOP]] != h->slice_num) yading@10: h->left_type[LTOP] = h->left_type[LBOT] = 0; yading@10: } yading@10: } yading@10: if (h->slice_table[topright_xy] != h->slice_num) yading@10: h->topright_type = 0; yading@10: } yading@10: yading@10: static void fill_decode_caches(H264Context *h, int mb_type) yading@10: { yading@10: int topleft_xy, top_xy, topright_xy, left_xy[LEFT_MBS]; yading@10: int topleft_type, top_type, topright_type, left_type[LEFT_MBS]; yading@10: const uint8_t *left_block = h->left_block; yading@10: int i; yading@10: uint8_t *nnz; yading@10: uint8_t *nnz_cache; yading@10: yading@10: topleft_xy = h->topleft_mb_xy; yading@10: top_xy = h->top_mb_xy; yading@10: topright_xy = h->topright_mb_xy; yading@10: left_xy[LTOP] = h->left_mb_xy[LTOP]; yading@10: left_xy[LBOT] = h->left_mb_xy[LBOT]; yading@10: topleft_type = h->topleft_type; yading@10: top_type = h->top_type; yading@10: topright_type = h->topright_type; yading@10: left_type[LTOP] = h->left_type[LTOP]; yading@10: left_type[LBOT] = h->left_type[LBOT]; yading@10: yading@10: if (!IS_SKIP(mb_type)) { yading@10: if (IS_INTRA(mb_type)) { yading@10: int type_mask = h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1; yading@10: h->topleft_samples_available = yading@10: h->top_samples_available = yading@10: h->left_samples_available = 0xFFFF; yading@10: h->topright_samples_available = 0xEEEA; yading@10: yading@10: if (!(top_type & type_mask)) { yading@10: h->topleft_samples_available = 0xB3FF; yading@10: h->top_samples_available = 0x33FF; yading@10: h->topright_samples_available = 0x26EA; yading@10: } yading@10: if (IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[LTOP])) { yading@10: if (IS_INTERLACED(mb_type)) { yading@10: if (!(left_type[LTOP] & type_mask)) { yading@10: h->topleft_samples_available &= 0xDFFF; yading@10: h->left_samples_available &= 0x5FFF; yading@10: } yading@10: if (!(left_type[LBOT] & type_mask)) { yading@10: h->topleft_samples_available &= 0xFF5F; yading@10: h->left_samples_available &= 0xFF5F; yading@10: } yading@10: } else { yading@10: int left_typei = h->cur_pic.mb_type[left_xy[LTOP] + h->mb_stride]; yading@10: yading@10: av_assert2(left_xy[LTOP] == left_xy[LBOT]); yading@10: if (!((left_typei & type_mask) && (left_type[LTOP] & type_mask))) { yading@10: h->topleft_samples_available &= 0xDF5F; yading@10: h->left_samples_available &= 0x5F5F; yading@10: } yading@10: } yading@10: } else { yading@10: if (!(left_type[LTOP] & type_mask)) { yading@10: h->topleft_samples_available &= 0xDF5F; yading@10: h->left_samples_available &= 0x5F5F; yading@10: } yading@10: } yading@10: yading@10: if (!(topleft_type & type_mask)) yading@10: h->topleft_samples_available &= 0x7FFF; yading@10: yading@10: if (!(topright_type & type_mask)) yading@10: h->topright_samples_available &= 0xFBFF; yading@10: yading@10: if (IS_INTRA4x4(mb_type)) { yading@10: if (IS_INTRA4x4(top_type)) { yading@10: AV_COPY32(h->intra4x4_pred_mode_cache + 4 + 8 * 0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]); yading@10: } else { yading@10: h->intra4x4_pred_mode_cache[4 + 8 * 0] = yading@10: h->intra4x4_pred_mode_cache[5 + 8 * 0] = yading@10: h->intra4x4_pred_mode_cache[6 + 8 * 0] = yading@10: h->intra4x4_pred_mode_cache[7 + 8 * 0] = 2 - 3 * !(top_type & type_mask); yading@10: } yading@10: for (i = 0; i < 2; i++) { yading@10: if (IS_INTRA4x4(left_type[LEFT(i)])) { yading@10: int8_t *mode = h->intra4x4_pred_mode + h->mb2br_xy[left_xy[LEFT(i)]]; yading@10: h->intra4x4_pred_mode_cache[3 + 8 * 1 + 2 * 8 * i] = mode[6 - left_block[0 + 2 * i]]; yading@10: h->intra4x4_pred_mode_cache[3 + 8 * 2 + 2 * 8 * i] = mode[6 - left_block[1 + 2 * i]]; yading@10: } else { yading@10: h->intra4x4_pred_mode_cache[3 + 8 * 1 + 2 * 8 * i] = yading@10: h->intra4x4_pred_mode_cache[3 + 8 * 2 + 2 * 8 * i] = 2 - 3 * !(left_type[LEFT(i)] & type_mask); yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: /* yading@10: * 0 . T T. T T T T yading@10: * 1 L . .L . . . . yading@10: * 2 L . .L . . . . yading@10: * 3 . T TL . . . . yading@10: * 4 L . .L . . . . yading@10: * 5 L . .. . . . . yading@10: */ yading@10: /* FIXME: constraint_intra_pred & partitioning & nnz yading@10: * (let us hope this is just a typo in the spec) */ yading@10: nnz_cache = h->non_zero_count_cache; yading@10: if (top_type) { yading@10: nnz = h->non_zero_count[top_xy]; yading@10: AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[4 * 3]); yading@10: if (!h->chroma_y_shift) { yading@10: AV_COPY32(&nnz_cache[4 + 8 * 5], &nnz[4 * 7]); yading@10: AV_COPY32(&nnz_cache[4 + 8 * 10], &nnz[4 * 11]); yading@10: } else { yading@10: AV_COPY32(&nnz_cache[4 + 8 * 5], &nnz[4 * 5]); yading@10: AV_COPY32(&nnz_cache[4 + 8 * 10], &nnz[4 * 9]); yading@10: } yading@10: } else { yading@10: uint32_t top_empty = CABAC(h) && !IS_INTRA(mb_type) ? 0 : 0x40404040; yading@10: AV_WN32A(&nnz_cache[4 + 8 * 0], top_empty); yading@10: AV_WN32A(&nnz_cache[4 + 8 * 5], top_empty); yading@10: AV_WN32A(&nnz_cache[4 + 8 * 10], top_empty); yading@10: } yading@10: yading@10: for (i = 0; i < 2; i++) { yading@10: if (left_type[LEFT(i)]) { yading@10: nnz = h->non_zero_count[left_xy[LEFT(i)]]; yading@10: nnz_cache[3 + 8 * 1 + 2 * 8 * i] = nnz[left_block[8 + 0 + 2 * i]]; yading@10: nnz_cache[3 + 8 * 2 + 2 * 8 * i] = nnz[left_block[8 + 1 + 2 * i]]; yading@10: if (CHROMA444(h)) { yading@10: nnz_cache[3 + 8 * 6 + 2 * 8 * i] = nnz[left_block[8 + 0 + 2 * i] + 4 * 4]; yading@10: nnz_cache[3 + 8 * 7 + 2 * 8 * i] = nnz[left_block[8 + 1 + 2 * i] + 4 * 4]; yading@10: nnz_cache[3 + 8 * 11 + 2 * 8 * i] = nnz[left_block[8 + 0 + 2 * i] + 8 * 4]; yading@10: nnz_cache[3 + 8 * 12 + 2 * 8 * i] = nnz[left_block[8 + 1 + 2 * i] + 8 * 4]; yading@10: } else if (CHROMA422(h)) { yading@10: nnz_cache[3 + 8 * 6 + 2 * 8 * i] = nnz[left_block[8 + 0 + 2 * i] - 2 + 4 * 4]; yading@10: nnz_cache[3 + 8 * 7 + 2 * 8 * i] = nnz[left_block[8 + 1 + 2 * i] - 2 + 4 * 4]; yading@10: nnz_cache[3 + 8 * 11 + 2 * 8 * i] = nnz[left_block[8 + 0 + 2 * i] - 2 + 8 * 4]; yading@10: nnz_cache[3 + 8 * 12 + 2 * 8 * i] = nnz[left_block[8 + 1 + 2 * i] - 2 + 8 * 4]; yading@10: } else { yading@10: nnz_cache[3 + 8 * 6 + 8 * i] = nnz[left_block[8 + 4 + 2 * i]]; yading@10: nnz_cache[3 + 8 * 11 + 8 * i] = nnz[left_block[8 + 5 + 2 * i]]; yading@10: } yading@10: } else { yading@10: nnz_cache[3 + 8 * 1 + 2 * 8 * i] = yading@10: nnz_cache[3 + 8 * 2 + 2 * 8 * i] = yading@10: nnz_cache[3 + 8 * 6 + 2 * 8 * i] = yading@10: nnz_cache[3 + 8 * 7 + 2 * 8 * i] = yading@10: nnz_cache[3 + 8 * 11 + 2 * 8 * i] = yading@10: nnz_cache[3 + 8 * 12 + 2 * 8 * i] = CABAC(h) && !IS_INTRA(mb_type) ? 0 : 64; yading@10: } yading@10: } yading@10: yading@10: if (CABAC(h)) { yading@10: // top_cbp yading@10: if (top_type) yading@10: h->top_cbp = h->cbp_table[top_xy]; yading@10: else yading@10: h->top_cbp = IS_INTRA(mb_type) ? 0x7CF : 0x00F; yading@10: // left_cbp yading@10: if (left_type[LTOP]) { yading@10: h->left_cbp = (h->cbp_table[left_xy[LTOP]] & 0x7F0) | yading@10: ((h->cbp_table[left_xy[LTOP]] >> (left_block[0] & (~1))) & 2) | yading@10: (((h->cbp_table[left_xy[LBOT]] >> (left_block[2] & (~1))) & 2) << 2); yading@10: } else { yading@10: h->left_cbp = IS_INTRA(mb_type) ? 0x7CF : 0x00F; yading@10: } yading@10: } yading@10: } yading@10: yading@10: if (IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)) { yading@10: int list; yading@10: int b_stride = h->b_stride; yading@10: for (list = 0; list < h->list_count; list++) { yading@10: int8_t *ref_cache = &h->ref_cache[list][scan8[0]]; yading@10: int8_t *ref = h->cur_pic.ref_index[list]; yading@10: int16_t(*mv_cache)[2] = &h->mv_cache[list][scan8[0]]; yading@10: int16_t(*mv)[2] = h->cur_pic.motion_val[list]; yading@10: if (!USES_LIST(mb_type, list)) yading@10: continue; yading@10: av_assert2(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)); yading@10: yading@10: if (USES_LIST(top_type, list)) { yading@10: const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride; yading@10: AV_COPY128(mv_cache[0 - 1 * 8], mv[b_xy + 0]); yading@10: ref_cache[0 - 1 * 8] = yading@10: ref_cache[1 - 1 * 8] = ref[4 * top_xy + 2]; yading@10: ref_cache[2 - 1 * 8] = yading@10: ref_cache[3 - 1 * 8] = ref[4 * top_xy + 3]; yading@10: } else { yading@10: AV_ZERO128(mv_cache[0 - 1 * 8]); yading@10: AV_WN32A(&ref_cache[0 - 1 * 8], yading@10: ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE) & 0xFF) * 0x01010101u); yading@10: } yading@10: yading@10: if (mb_type & (MB_TYPE_16x8 | MB_TYPE_8x8)) { yading@10: for (i = 0; i < 2; i++) { yading@10: int cache_idx = -1 + i * 2 * 8; yading@10: if (USES_LIST(left_type[LEFT(i)], list)) { yading@10: const int b_xy = h->mb2b_xy[left_xy[LEFT(i)]] + 3; yading@10: const int b8_xy = 4 * left_xy[LEFT(i)] + 1; yading@10: AV_COPY32(mv_cache[cache_idx], yading@10: mv[b_xy + b_stride * left_block[0 + i * 2]]); yading@10: AV_COPY32(mv_cache[cache_idx + 8], yading@10: mv[b_xy + b_stride * left_block[1 + i * 2]]); yading@10: ref_cache[cache_idx] = ref[b8_xy + (left_block[0 + i * 2] & ~1)]; yading@10: ref_cache[cache_idx + 8] = ref[b8_xy + (left_block[1 + i * 2] & ~1)]; yading@10: } else { yading@10: AV_ZERO32(mv_cache[cache_idx]); yading@10: AV_ZERO32(mv_cache[cache_idx + 8]); yading@10: ref_cache[cache_idx] = yading@10: ref_cache[cache_idx + 8] = (left_type[LEFT(i)]) ? LIST_NOT_USED yading@10: : PART_NOT_AVAILABLE; yading@10: } yading@10: } yading@10: } else { yading@10: if (USES_LIST(left_type[LTOP], list)) { yading@10: const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3; yading@10: const int b8_xy = 4 * left_xy[LTOP] + 1; yading@10: AV_COPY32(mv_cache[-1], mv[b_xy + b_stride * left_block[0]]); yading@10: ref_cache[-1] = ref[b8_xy + (left_block[0] & ~1)]; yading@10: } else { yading@10: AV_ZERO32(mv_cache[-1]); yading@10: ref_cache[-1] = left_type[LTOP] ? LIST_NOT_USED yading@10: : PART_NOT_AVAILABLE; yading@10: } yading@10: } yading@10: yading@10: if (USES_LIST(topright_type, list)) { yading@10: const int b_xy = h->mb2b_xy[topright_xy] + 3 * b_stride; yading@10: AV_COPY32(mv_cache[4 - 1 * 8], mv[b_xy]); yading@10: ref_cache[4 - 1 * 8] = ref[4 * topright_xy + 2]; yading@10: } else { yading@10: AV_ZERO32(mv_cache[4 - 1 * 8]); yading@10: ref_cache[4 - 1 * 8] = topright_type ? LIST_NOT_USED yading@10: : PART_NOT_AVAILABLE; yading@10: } yading@10: if(ref_cache[2 - 1*8] < 0 || ref_cache[4 - 1*8] < 0){ yading@10: if (USES_LIST(topleft_type, list)) { yading@10: const int b_xy = h->mb2b_xy[topleft_xy] + 3 + b_stride + yading@10: (h->topleft_partition & 2 * b_stride); yading@10: const int b8_xy = 4 * topleft_xy + 1 + (h->topleft_partition & 2); yading@10: AV_COPY32(mv_cache[-1 - 1 * 8], mv[b_xy]); yading@10: ref_cache[-1 - 1 * 8] = ref[b8_xy]; yading@10: } else { yading@10: AV_ZERO32(mv_cache[-1 - 1 * 8]); yading@10: ref_cache[-1 - 1 * 8] = topleft_type ? LIST_NOT_USED yading@10: : PART_NOT_AVAILABLE; yading@10: } yading@10: } yading@10: yading@10: if ((mb_type & (MB_TYPE_SKIP | MB_TYPE_DIRECT2)) && !FRAME_MBAFF(h)) yading@10: continue; yading@10: yading@10: if (!(mb_type & (MB_TYPE_SKIP | MB_TYPE_DIRECT2))) { yading@10: uint8_t(*mvd_cache)[2] = &h->mvd_cache[list][scan8[0]]; yading@10: uint8_t(*mvd)[2] = h->mvd_table[list]; yading@10: ref_cache[2 + 8 * 0] = yading@10: ref_cache[2 + 8 * 2] = PART_NOT_AVAILABLE; yading@10: AV_ZERO32(mv_cache[2 + 8 * 0]); yading@10: AV_ZERO32(mv_cache[2 + 8 * 2]); yading@10: yading@10: if (CABAC(h)) { yading@10: if (USES_LIST(top_type, list)) { yading@10: const int b_xy = h->mb2br_xy[top_xy]; yading@10: AV_COPY64(mvd_cache[0 - 1 * 8], mvd[b_xy + 0]); yading@10: } else { yading@10: AV_ZERO64(mvd_cache[0 - 1 * 8]); yading@10: } yading@10: if (USES_LIST(left_type[LTOP], list)) { yading@10: const int b_xy = h->mb2br_xy[left_xy[LTOP]] + 6; yading@10: AV_COPY16(mvd_cache[-1 + 0 * 8], mvd[b_xy - left_block[0]]); yading@10: AV_COPY16(mvd_cache[-1 + 1 * 8], mvd[b_xy - left_block[1]]); yading@10: } else { yading@10: AV_ZERO16(mvd_cache[-1 + 0 * 8]); yading@10: AV_ZERO16(mvd_cache[-1 + 1 * 8]); yading@10: } yading@10: if (USES_LIST(left_type[LBOT], list)) { yading@10: const int b_xy = h->mb2br_xy[left_xy[LBOT]] + 6; yading@10: AV_COPY16(mvd_cache[-1 + 2 * 8], mvd[b_xy - left_block[2]]); yading@10: AV_COPY16(mvd_cache[-1 + 3 * 8], mvd[b_xy - left_block[3]]); yading@10: } else { yading@10: AV_ZERO16(mvd_cache[-1 + 2 * 8]); yading@10: AV_ZERO16(mvd_cache[-1 + 3 * 8]); yading@10: } yading@10: AV_ZERO16(mvd_cache[2 + 8 * 0]); yading@10: AV_ZERO16(mvd_cache[2 + 8 * 2]); yading@10: if (h->slice_type_nos == AV_PICTURE_TYPE_B) { yading@10: uint8_t *direct_cache = &h->direct_cache[scan8[0]]; yading@10: uint8_t *direct_table = h->direct_table; yading@10: fill_rectangle(direct_cache, 4, 4, 8, MB_TYPE_16x16 >> 1, 1); yading@10: yading@10: if (IS_DIRECT(top_type)) { yading@10: AV_WN32A(&direct_cache[-1 * 8], yading@10: 0x01010101u * (MB_TYPE_DIRECT2 >> 1)); yading@10: } else if (IS_8X8(top_type)) { yading@10: int b8_xy = 4 * top_xy; yading@10: direct_cache[0 - 1 * 8] = direct_table[b8_xy + 2]; yading@10: direct_cache[2 - 1 * 8] = direct_table[b8_xy + 3]; yading@10: } else { yading@10: AV_WN32A(&direct_cache[-1 * 8], yading@10: 0x01010101 * (MB_TYPE_16x16 >> 1)); yading@10: } yading@10: yading@10: if (IS_DIRECT(left_type[LTOP])) yading@10: direct_cache[-1 + 0 * 8] = MB_TYPE_DIRECT2 >> 1; yading@10: else if (IS_8X8(left_type[LTOP])) yading@10: direct_cache[-1 + 0 * 8] = direct_table[4 * left_xy[LTOP] + 1 + (left_block[0] & ~1)]; yading@10: else yading@10: direct_cache[-1 + 0 * 8] = MB_TYPE_16x16 >> 1; yading@10: yading@10: if (IS_DIRECT(left_type[LBOT])) yading@10: direct_cache[-1 + 2 * 8] = MB_TYPE_DIRECT2 >> 1; yading@10: else if (IS_8X8(left_type[LBOT])) yading@10: direct_cache[-1 + 2 * 8] = direct_table[4 * left_xy[LBOT] + 1 + (left_block[2] & ~1)]; yading@10: else yading@10: direct_cache[-1 + 2 * 8] = MB_TYPE_16x16 >> 1; yading@10: } yading@10: } yading@10: } yading@10: yading@10: #define MAP_MVS \ yading@10: MAP_F2F(scan8[0] - 1 - 1 * 8, topleft_type) \ yading@10: MAP_F2F(scan8[0] + 0 - 1 * 8, top_type) \ yading@10: MAP_F2F(scan8[0] + 1 - 1 * 8, top_type) \ yading@10: MAP_F2F(scan8[0] + 2 - 1 * 8, top_type) \ yading@10: MAP_F2F(scan8[0] + 3 - 1 * 8, top_type) \ yading@10: MAP_F2F(scan8[0] + 4 - 1 * 8, topright_type) \ yading@10: MAP_F2F(scan8[0] - 1 + 0 * 8, left_type[LTOP]) \ yading@10: MAP_F2F(scan8[0] - 1 + 1 * 8, left_type[LTOP]) \ yading@10: MAP_F2F(scan8[0] - 1 + 2 * 8, left_type[LBOT]) \ yading@10: MAP_F2F(scan8[0] - 1 + 3 * 8, left_type[LBOT]) yading@10: yading@10: if (FRAME_MBAFF(h)) { yading@10: if (MB_FIELD(h)) { yading@10: yading@10: #define MAP_F2F(idx, mb_type) \ yading@10: if (!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0) { \ yading@10: h->ref_cache[list][idx] <<= 1; \ yading@10: h->mv_cache[list][idx][1] /= 2; \ yading@10: h->mvd_cache[list][idx][1] >>= 1; \ yading@10: } yading@10: yading@10: MAP_MVS yading@10: } else { yading@10: yading@10: #undef MAP_F2F yading@10: #define MAP_F2F(idx, mb_type) \ yading@10: if (IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0) { \ yading@10: h->ref_cache[list][idx] >>= 1; \ yading@10: h->mv_cache[list][idx][1] <<= 1; \ yading@10: h->mvd_cache[list][idx][1] <<= 1; \ yading@10: } yading@10: yading@10: MAP_MVS yading@10: #undef MAP_F2F yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: h->neighbor_transform_size = !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[LTOP]); yading@10: } yading@10: yading@10: /** yading@10: * decodes a P_SKIP or B_SKIP macroblock yading@10: */ yading@10: static void av_unused decode_mb_skip(H264Context *h) yading@10: { yading@10: const int mb_xy = h->mb_xy; yading@10: int mb_type = 0; yading@10: yading@10: memset(h->non_zero_count[mb_xy], 0, 48); yading@10: yading@10: if (MB_FIELD(h)) yading@10: mb_type |= MB_TYPE_INTERLACED; yading@10: yading@10: if (h->slice_type_nos == AV_PICTURE_TYPE_B) { yading@10: // just for fill_caches. pred_direct_motion will set the real mb_type yading@10: mb_type |= MB_TYPE_L0L1 | MB_TYPE_DIRECT2 | MB_TYPE_SKIP; yading@10: if (h->direct_spatial_mv_pred) { yading@10: fill_decode_neighbors(h, mb_type); yading@10: fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ... yading@10: } yading@10: ff_h264_pred_direct_motion(h, &mb_type); yading@10: mb_type |= MB_TYPE_SKIP; yading@10: } else { yading@10: mb_type |= MB_TYPE_16x16 | MB_TYPE_P0L0 | MB_TYPE_P1L0 | MB_TYPE_SKIP; yading@10: yading@10: fill_decode_neighbors(h, mb_type); yading@10: pred_pskip_motion(h); yading@10: } yading@10: yading@10: write_back_motion(h, mb_type); yading@10: h->cur_pic.mb_type[mb_xy] = mb_type; yading@10: h->cur_pic.qscale_table[mb_xy] = h->qscale; yading@10: h->slice_table[mb_xy] = h->slice_num; yading@10: h->prev_mb_skipped = 1; yading@10: } yading@10: yading@10: #endif /* AVCODEC_H264_MVPRED_H */