From: Jingning Han Date: Fri, 22 Nov 2013 19:56:29 +0000 (-0800) Subject: Use separate inter predictors for enc/dec X-Git-Tag: v1.4.0~2956^2 X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=ba8b5e8d6d5db6972d1ce662626c839457b63f28;p=libvpx Use separate inter predictors for enc/dec The decoder will construct inter predictor using lazy border extension, while the encoder, going with multiple runs of motion search in the rate- distortion optimization loop for each block, does border extension at frame level. This commit makes separate the inter predictors for encoder and decoder, respectively. Change-Id: Ieca2fecba3a7201a6d64ef9f219e5d91e50559c3 --- diff --git a/vp9/common/vp9_reconinter.c b/vp9/common/vp9_reconinter.c index afec0faf4..09a4fc826 100644 --- a/vp9/common/vp9_reconinter.c +++ b/vp9/common/vp9_reconinter.c @@ -98,7 +98,6 @@ MV clamp_mv_to_umv_border_sb(const MACROBLOCKD *xd, const MV *src_mv, return clamped_mv; } - // TODO(jkoleszar): In principle, pred_w, pred_h are unnecessary, as we could // calculate the subsampled BLOCK_SIZE, but that type isn't defined for // sizes smaller than 16x16 yet. @@ -206,6 +205,96 @@ void vp9_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col, MAX_MB_PLANE - 1); } +// TODO(jingning): This function serves as a placeholder for decoder prediction +// using on demand border extension. It should be moved to /decoder/ directory. +static void dec_build_inter_predictors(MACROBLOCKD *xd, int plane, int block, + BLOCK_SIZE bsize, int pred_w, int pred_h, + int mi_x, int mi_y) { + struct macroblockd_plane *const pd = &xd->plane[plane]; + const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd); + const int bwl = b_width_log2(plane_bsize); + const int bw = 4 << bwl; + const int bh = 4 * num_4x4_blocks_high_lookup[plane_bsize]; + const int x = 4 * (block & ((1 << bwl) - 1)); + const int y = 4 * (block >> bwl); + const MODE_INFO *mi = xd->mi_8x8[0]; + const int is_compound = has_second_ref(&mi->mbmi); + int ref; + + assert(x < bw); + assert(y < bh); + assert(mi->mbmi.sb_type < BLOCK_8X8 || 4 << pred_w == bw); + assert(mi->mbmi.sb_type < BLOCK_8X8 || 4 << pred_h == bh); + + for (ref = 0; ref < 1 + is_compound; ++ref) { + struct scale_factors *const scale = &xd->scale_factor[ref]; + struct buf_2d *const pre_buf = &pd->pre[ref]; + struct buf_2d *const dst_buf = &pd->dst; + uint8_t *const dst = dst_buf->buf + dst_buf->stride * y + x; + + // TODO(jkoleszar): All chroma MVs in SPLITMV mode are taken as the + // same MV (the average of the 4 luma MVs) but we could do something + // smarter for non-4:2:0. Just punt for now, pending the changes to get + // rid of SPLITMV mode entirely. + const MV mv = mi->mbmi.sb_type < BLOCK_8X8 + ? (plane == 0 ? mi->bmi[block].as_mv[ref].as_mv + : mi_mv_pred_q4(mi, ref)) + : mi->mbmi.mv[ref].as_mv; + + // TODO(jkoleszar): This clamping is done in the incorrect place for the + // scaling case. It needs to be done on the scaled MV, not the pre-scaling + // MV. Note however that it performs the subsampling aware scaling so + // that the result is always q4. + // mv_precision precision is MV_PRECISION_Q4. + const MV mv_q4 = clamp_mv_to_umv_border_sb(xd, &mv, bw, bh, + pd->subsampling_x, + pd->subsampling_y); + + uint8_t *pre; + MV32 scaled_mv; + int xs, ys; + + if (vp9_is_scaled(scale->sfc)) { + pre = pre_buf->buf + scaled_buffer_offset(x, y, pre_buf->stride, scale); + scale->sfc->set_scaled_offsets(scale, mi_y + y, mi_x + x); + scaled_mv = scale->sfc->scale_mv(&mv_q4, scale); + xs = scale->sfc->x_step_q4; + ys = scale->sfc->y_step_q4; + } else { + pre = pre_buf->buf + (y * pre_buf->stride + x); + scaled_mv.row = mv_q4.row; + scaled_mv.col = mv_q4.col; + xs = ys = 16; + } + + inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride, + &scaled_mv, scale, + 4 << pred_w, 4 << pred_h, ref, + &xd->subpix, xs, ys); + } +} + +void vp9_dec_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col, + BLOCK_SIZE bsize) { + int plane; + for (plane = 0; plane < MAX_MB_PLANE; ++plane) { + const int mi_x = mi_col * MI_SIZE; + const int mi_y = mi_row * MI_SIZE; + const int bwl = b_width_log2(bsize) - xd->plane[plane].subsampling_x; + const int bhl = b_height_log2(bsize) - xd->plane[plane].subsampling_y; + + if (xd->mi_8x8[0]->mbmi.sb_type < BLOCK_8X8) { + int i = 0, x, y; + assert(bsize == BLOCK_8X8); + for (y = 0; y < 1 << bhl; ++y) + for (x = 0; x < 1 << bwl; ++x) + dec_build_inter_predictors(xd, plane, i++, bsize, 0, 0, mi_x, mi_y); + } else { + dec_build_inter_predictors(xd, plane, 0, bsize, bwl, bhl, mi_x, mi_y); + } + } +} + // TODO(dkovalev: find better place for this function) void vp9_setup_scale_factors(VP9_COMMON *cm, int i) { const int ref = cm->active_ref_idx[i]; diff --git a/vp9/common/vp9_reconinter.h b/vp9/common/vp9_reconinter.h index b328754e7..4a302f988 100644 --- a/vp9/common/vp9_reconinter.h +++ b/vp9/common/vp9_reconinter.h @@ -24,6 +24,9 @@ void vp9_build_inter_predictors_sbuv(MACROBLOCKD *xd, int mi_row, int mi_col, void vp9_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col, BLOCK_SIZE bsize); +void vp9_dec_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col, + BLOCK_SIZE bsize); + void vp9_build_inter_predictor(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, const MV *mv_q3, diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c index f30079e54..ea185703c 100644 --- a/vp9/decoder/vp9_decodeframe.c +++ b/vp9/decoder/vp9_decodeframe.c @@ -444,7 +444,7 @@ static void decode_modes_b(VP9_COMMON *const cm, MACROBLOCKD *const xd, vp9_get_filter_kernel(mbmi->interp_filter); // Prediction - vp9_build_inter_predictors_sb(xd, mi_row, mi_col, bsize); + vp9_dec_build_inter_predictors_sb(xd, mi_row, mi_col, bsize); // Reconstruction if (!mbmi->skip_coeff) {