diff options
Diffstat (limited to 'vp9')
-rw-r--r-- | vp9/common/vp9_blockd.h | 11 | ||||
-rw-r--r-- | vp9/common/vp9_findnearmv.h | 15 | ||||
-rw-r--r-- | vp9/common/vp9_mbpitch.c | 8 | ||||
-rw-r--r-- | vp9/decoder/vp9_decodemv.c | 15 | ||||
-rw-r--r-- | vp9/decoder/vp9_decodframe.c | 205 | ||||
-rw-r--r-- | vp9/decoder/vp9_detokenize.c | 283 | ||||
-rw-r--r-- | vp9/encoder/vp9_rdopt.c | 6 |
7 files changed, 191 insertions, 352 deletions
diff --git a/vp9/common/vp9_blockd.h b/vp9/common/vp9_blockd.h index d0c77077e..d372325a6 100644 --- a/vp9/common/vp9_blockd.h +++ b/vp9/common/vp9_blockd.h @@ -237,6 +237,14 @@ static INLINE int mb_height_log2(BLOCK_SIZE_TYPE sb_type) { } } +typedef enum { + BLOCK_4X4_LG2 = 0, + BLOCK_8X8_LG2 = 2, + BLOCK_16X16_LG2 = 4, + BLOCK_32X32_LG2 = 6, + BLOCK_64X64_LG2 = 8 +} BLOCK_SIZE_LG2; + typedef struct { MB_PREDICTION_MODE mode, uv_mode; #if CONFIG_COMP_INTERINTRA_PRED @@ -321,6 +329,9 @@ struct mb_plane { DECLARE_ALIGNED(16, int16_t, qcoeff[64 * 64]); DECLARE_ALIGNED(16, int16_t, dqcoeff[64 * 64]); DECLARE_ALIGNED(16, uint16_t, eobs[256]); + PLANE_TYPE plane_type; + int subsampling_x; + int subsampling_y; }; #define BLOCK_OFFSET(x, i, n) ((x) + (i) * (n)) diff --git a/vp9/common/vp9_findnearmv.h b/vp9/common/vp9_findnearmv.h index 6887b044f..1c96f0ad0 100644 --- a/vp9/common/vp9_findnearmv.h +++ b/vp9/common/vp9_findnearmv.h @@ -17,8 +17,8 @@ #include "vp9/common/vp9_treecoder.h" #include "vp9/common/vp9_onyxc_int.h" -#define LEFT_TOP_MARGIN (16 << 3) -#define RIGHT_BOTTOM_MARGIN (16 << 3) +#define LEFT_TOP_MARGIN ((VP9BORDERINPIXELS - VP9_INTERP_EXTEND) << 3) +#define RIGHT_BOTTOM_MARGIN ((VP9BORDERINPIXELS - VP9_INTERP_EXTEND) << 3) /* check a list of motion vectors by sad score using a number rows of pixels * above and a number cols of pixels in the left to select the one with best @@ -43,7 +43,7 @@ static void mv_bias(int refmb_ref_frame_sign_bias, int refframe, mvp->as_mv = xmv; } - +// TODO(jingning): this mv clamping function should be block size dependent. static void clamp_mv(int_mv *mv, int mb_to_left_edge, int mb_to_right_edge, @@ -59,12 +59,19 @@ static void clamp_mv(int_mv *mv, mb_to_bottom_edge : mv->as_mv.row; } -static void clamp_mv2(int_mv *mv, const MACROBLOCKD *xd) { +static int clamp_mv2(int_mv *mv, const MACROBLOCKD *xd) { + int_mv tmp_mv; + int mv_clampped = 0; + tmp_mv.as_int = mv->as_int; clamp_mv(mv, xd->mb_to_left_edge - LEFT_TOP_MARGIN, xd->mb_to_right_edge + RIGHT_BOTTOM_MARGIN, xd->mb_to_top_edge - LEFT_TOP_MARGIN, xd->mb_to_bottom_edge + RIGHT_BOTTOM_MARGIN); + if (tmp_mv.as_int != mv->as_int) + mv_clampped = 1; + + return mv_clampped; } static unsigned int check_mv_bounds(int_mv *mv, diff --git a/vp9/common/vp9_mbpitch.c b/vp9/common/vp9_mbpitch.c index b357c9ac9..aba950e9a 100644 --- a/vp9/common/vp9_mbpitch.c +++ b/vp9/common/vp9_mbpitch.c @@ -70,7 +70,7 @@ static void setup_macroblock(MACROBLOCKD *mb, BLOCKSET bs) { } void vp9_setup_block_dptrs(MACROBLOCKD *mb) { - int r, c; + int r, c, i; BLOCKD *blockd = mb->block; for (r = 0; r < 4; r++) { @@ -99,6 +99,12 @@ void vp9_setup_block_dptrs(MACROBLOCKD *mb) { blockd[to].predictor = &mb->predictor[from]; } } + + for (i = 0; i < MAX_MB_PLANE; i++) { + mb->plane[i].plane_type = i ? PLANE_TYPE_UV : PLANE_TYPE_Y_WITH_DC; + mb->plane[i].subsampling_x = !!i; + mb->plane[i].subsampling_y = !!i; + } } void vp9_build_block_doffsets(MACROBLOCKD *mb) { diff --git a/vp9/decoder/vp9_decodemv.c b/vp9/decoder/vp9_decodemv.c index e033a2435..f081e7174 100644 --- a/vp9/decoder/vp9_decodemv.c +++ b/vp9/decoder/vp9_decodemv.c @@ -1026,11 +1026,6 @@ static void read_mb_modes_mv(VP9D_COMP *pbi, MODE_INFO *mi, MB_MODE_INFO *mbmi, case NEWMV: process_mv(bc, &mv->as_mv, &best_mv.as_mv, nmvc, &cm->fc.NMVcount, xd->allow_high_precision_mv); - - // Don't need to check this on NEARMV and NEARESTMV modes - // since those modes clamp the MV. The NEWMV mode does not, - // so signal to the prediction stage whether special - // handling may be required. mbmi->need_to_clamp_mvs = check_mv_bounds(mv, mb_to_left_edge, mb_to_right_edge, @@ -1040,11 +1035,11 @@ static void read_mb_modes_mv(VP9D_COMP *pbi, MODE_INFO *mi, MB_MODE_INFO *mbmi, if (mbmi->second_ref_frame > 0) { process_mv(bc, &mbmi->mv[1].as_mv, &best_mv_second.as_mv, nmvc, &cm->fc.NMVcount, xd->allow_high_precision_mv); - mbmi->need_to_clamp_secondmv |= check_mv_bounds(&mbmi->mv[1], - mb_to_left_edge, - mb_to_right_edge, - mb_to_top_edge, - mb_to_bottom_edge); + mbmi->need_to_clamp_secondmv = check_mv_bounds(&mbmi->mv[1], + mb_to_left_edge, + mb_to_right_edge, + mb_to_top_edge, + mb_to_bottom_edge); } break; default: diff --git a/vp9/decoder/vp9_decodframe.c b/vp9/decoder/vp9_decodframe.c index 9ad060fe5..965310e7d 100644 --- a/vp9/decoder/vp9_decodframe.c +++ b/vp9/decoder/vp9_decodframe.c @@ -53,8 +53,8 @@ static int read_le32(const uint8_t *p) { } // len == 0 is not allowed -static int read_is_valid(const unsigned char *start, size_t len, - const unsigned char *end) { +static int read_is_valid(const uint8_t *start, size_t len, + const uint8_t *end) { return start + len > start && start + len <= end; } @@ -227,23 +227,6 @@ static void skip_recon_mb(VP9D_COMP *pbi, MACROBLOCKD *xd, static void decode_16x16(VP9D_COMP *pbi, MACROBLOCKD *xd, BOOL_DECODER* const bc) { const TX_TYPE tx_type = get_tx_type_16x16(xd, 0); -#if 0 // def DEC_DEBUG - if (dec_debug) { - int i; - printf("\n"); - printf("qcoeff 16x16\n"); - for (i = 0; i < 400; i++) { - printf("%3d ", xd->qcoeff[i]); - if (i % 16 == 15) printf("\n"); - } - printf("\n"); - printf("predictor\n"); - for (i = 0; i < 400; i++) { - printf("%3d ", xd->predictor[i]); - if (i % 16 == 15) printf("\n"); - } - } -#endif if (tx_type != DCT_DCT) { vp9_ht_dequant_idct_add_16x16_c(tx_type, xd->plane[0].qcoeff, xd->block[0].dequant, xd->predictor, @@ -266,21 +249,11 @@ static void decode_16x16(VP9D_COMP *pbi, MACROBLOCKD *xd, static void decode_8x8(VP9D_COMP *pbi, MACROBLOCKD *xd, BOOL_DECODER* const bc) { - // First do Y + const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode; + // luma // if the first one is DCT_DCT assume all the rest are as well TX_TYPE tx_type = get_tx_type_8x8(xd, 0); -#if 0 // def DEC_DEBUG - if (dec_debug) { - int i; - printf("\n"); - printf("qcoeff 8x8\n"); - for (i = 0; i < 384; i++) { - printf("%3d ", xd->qcoeff[i]); - if (i % 16 == 15) printf("\n"); - } - } -#endif - if (tx_type != DCT_DCT || xd->mode_info_context->mbmi.mode == I8X8_PRED) { + if (tx_type != DCT_DCT || mode == I8X8_PRED) { int i; for (i = 0; i < 4; i++) { int ib = vp9_i8x8_block[i]; @@ -291,7 +264,7 @@ static void decode_8x8(VP9D_COMP *pbi, MACROBLOCKD *xd, uint8_t *dst = *(xd->block[ib].base_dst) + xd->block[ib].dst; int stride = xd->dst.y_stride; BLOCKD *b = &xd->block[ib]; - if (xd->mode_info_context->mbmi.mode == I8X8_PRED) { + if (mode == I8X8_PRED) { int i8x8mode = b->bmi.as_mode.first; vp9_intra8x8_predict(xd, b, i8x8mode, b->predictor); } @@ -313,8 +286,8 @@ static void decode_8x8(VP9D_COMP *pbi, MACROBLOCKD *xd, xd); } - // Now do UV - if (xd->mode_info_context->mbmi.mode == I8X8_PRED) { + // chroma + if (mode == I8X8_PRED) { int i; for (i = 0; i < 4; i++) { int ib = vp9_i8x8_block[i]; @@ -335,7 +308,7 @@ static void decode_8x8(VP9D_COMP *pbi, MACROBLOCKD *xd, *(b->base_dst) + b->dst, 8, b->dst_stride, xd->plane[2].eobs[i]); } - } else if (xd->mode_info_context->mbmi.mode == SPLITMV) { + } else if (mode == SPLITMV) { xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant, xd->predictor + 16 * 16, xd->dst.u_buffer, xd->dst.uv_stride, xd->plane[1].eobs); @@ -351,35 +324,13 @@ static void decode_8x8(VP9D_COMP *pbi, MACROBLOCKD *xd, xd->predictor + 16 * 16 + 64, xd->dst.v_buffer, 8, xd->dst.uv_stride, xd->plane[2].eobs[0]); } -#if 0 // def DEC_DEBUG - if (dec_debug) { - int i; - printf("\n"); - printf("predictor\n"); - for (i = 0; i < 384; i++) { - printf("%3d ", xd->predictor[i]); - if (i % 16 == 15) printf("\n"); - } - } -#endif } static void decode_4x4(VP9D_COMP *pbi, MACROBLOCKD *xd, BOOL_DECODER* const bc) { TX_TYPE tx_type; int i = 0; - MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode; -#if 0 // def DEC_DEBUG - if (dec_debug) { - int i; - printf("\n"); - printf("predictor\n"); - for (i = 0; i < 384; i++) { - printf("%3d ", xd->predictor[i]); - if (i % 16 == 15) printf("\n"); - } - } -#endif + const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode; if (mode == I8X8_PRED) { for (i = 0; i < 4; i++) { int ib = vp9_i8x8_block[i]; @@ -468,23 +419,6 @@ static void decode_4x4(VP9D_COMP *pbi, MACROBLOCKD *xd, xd->predictor + 16 * 16 + 64, xd->dst.v_buffer, xd->dst.uv_stride, xd->plane[2].eobs); } else { -#if 0 // def DEC_DEBUG - if (dec_debug) { - int i; - printf("\n"); - printf("qcoeff 4x4\n"); - for (i = 0; i < 400; i++) { - printf("%3d ", xd->qcoeff[i]); - if (i % 16 == 15) printf("\n"); - } - printf("\n"); - printf("predictor\n"); - for (i = 0; i < 400; i++) { - printf("%3d ", xd->predictor[i]); - if (i % 16 == 15) printf("\n"); - } - } -#endif for (i = 0; i < 16; i++) { BLOCKD *b = &xd->block[i]; tx_type = get_tx_type_4x4(xd, i); @@ -990,6 +924,8 @@ static void set_refs(VP9D_COMP *pbi, int block_size, int mb_row, int mb_col) { // Select the appropriate reference frame for this MB const int second_fb_idx = cm->active_ref_idx[mbmi->second_ref_frame - 1]; const YV12_BUFFER_CONFIG *second_cfg = &cm->yv12_fb[second_fb_idx]; + xd->scale_factor[1] = cm->active_ref_scale[mbmi->second_ref_frame - 1]; + xd->scale_factor_uv[1] = cm->active_ref_scale[mbmi->second_ref_frame - 1]; setup_pred_block(&xd->second_pre, second_cfg, mb_row, mb_col, &xd->scale_factor[1], &xd->scale_factor_uv[1]); xd->corrupted |= second_cfg->corrupted; @@ -998,9 +934,9 @@ static void set_refs(VP9D_COMP *pbi, int block_size, int mb_row, int mb_col) { } /* Decode a row of Superblocks (2x2 region of MBs) */ -static void decode_sb_row(VP9D_COMP *pbi, VP9_COMMON *pc, - int mb_row, MACROBLOCKD *xd, - BOOL_DECODER* const bc) { +static void decode_sb_row(VP9D_COMP *pbi, int mb_row, vp9_reader* r) { + VP9_COMMON *const pc = &pbi->common; + MACROBLOCKD *const xd = &pbi->mb; int mb_col; // For a SB there are 2 left contexts, each pertaining to a MB row within @@ -1008,72 +944,52 @@ static void decode_sb_row(VP9D_COMP *pbi, VP9_COMMON *pc, for (mb_col = pc->cur_tile_mb_col_start; mb_col < pc->cur_tile_mb_col_end; mb_col += 4) { - if (vp9_read(bc, pc->prob_sb64_coded)) { -#ifdef DEC_DEBUG - dec_debug = (pc->current_video_frame == 11 && pc->show_frame && - mb_row == 8 && mb_col == 0); - if (dec_debug) - printf("Debug Decode SB64\n"); -#endif + if (vp9_read(r, pc->prob_sb64_coded)) { + // SB64 decoding set_offsets(pbi, 64, mb_row, mb_col); - vp9_decode_mb_mode_mv(pbi, xd, mb_row, mb_col, bc); + vp9_decode_mb_mode_mv(pbi, xd, mb_row, mb_col, r); set_refs(pbi, 64, mb_row, mb_col); - decode_sb64(pbi, xd, mb_row, mb_col, bc); - xd->corrupted |= bool_error(bc); + decode_sb64(pbi, xd, mb_row, mb_col, r); + xd->corrupted |= bool_error(r); } else { + // not SB64 int j; - for (j = 0; j < 4; j++) { - const int x_idx_sb = (j & 1) << 1, y_idx_sb = j & 2; + const int x_idx_sb = mb_col + 2 * (j % 2); + const int y_idx_sb = mb_row + 2 * (j / 2); - if (mb_row + y_idx_sb >= pc->mb_rows || - mb_col + x_idx_sb >= pc->mb_cols) { - // MB lies outside frame, skip on to next - continue; - } + if (y_idx_sb >= pc->mb_rows || x_idx_sb >= pc->mb_cols) + continue; // MB lies outside frame, skip on to next xd->sb_index = j; - if (vp9_read(bc, pc->prob_sb32_coded)) { -#ifdef DEC_DEBUG - dec_debug = (pc->current_video_frame == 11 && pc->show_frame && - mb_row + y_idx_sb == 8 && mb_col + x_idx_sb == 0); - if (dec_debug) - printf("Debug Decode SB32\n"); -#endif - set_offsets(pbi, 32, mb_row + y_idx_sb, mb_col + x_idx_sb); - vp9_decode_mb_mode_mv(pbi, - xd, mb_row + y_idx_sb, mb_col + x_idx_sb, bc); - set_refs(pbi, 32, mb_row + y_idx_sb, mb_col + x_idx_sb); - decode_sb32(pbi, xd, mb_row + y_idx_sb, mb_col + x_idx_sb, bc); - xd->corrupted |= bool_error(bc); + if (vp9_read(r, pc->prob_sb32_coded)) { + // SB32 decoding + set_offsets(pbi, 32, y_idx_sb, x_idx_sb); + vp9_decode_mb_mode_mv(pbi, xd, y_idx_sb, x_idx_sb, r); + set_refs(pbi, 32, y_idx_sb, x_idx_sb); + decode_sb32(pbi, xd, y_idx_sb, x_idx_sb, r); + xd->corrupted |= bool_error(r); } else { - int i; - + // not SB32 // Process the 4 MBs within the SB in the order: // top-left, top-right, bottom-left, bottom-right + int i; for (i = 0; i < 4; i++) { - const int x_idx = x_idx_sb + (i & 1), y_idx = y_idx_sb + (i >> 1); + const int x_idx_mb = x_idx_sb + (i % 2); + const int y_idx_mb = y_idx_sb + (i / 2); - if (mb_row + y_idx >= pc->mb_rows || - mb_col + x_idx >= pc->mb_cols) { - // MB lies outside frame, skip on to next - continue; - } -#ifdef DEC_DEBUG - dec_debug = (pc->current_video_frame == 11 && pc->show_frame && - mb_row + y_idx == 8 && mb_col + x_idx == 0); - if (dec_debug) - printf("Debug Decode MB\n"); -#endif + if (y_idx_mb >= pc->mb_rows || x_idx_mb >= pc->mb_cols) + continue; // MB lies outside frame, skip on to next - set_offsets(pbi, 16, mb_row + y_idx, mb_col + x_idx); xd->mb_index = i; - vp9_decode_mb_mode_mv(pbi, xd, mb_row + y_idx, mb_col + x_idx, bc); - set_refs(pbi, 16, mb_row + y_idx, mb_col + x_idx); - decode_mb(pbi, xd, mb_row + y_idx, mb_col + x_idx, bc); - xd->corrupted |= bool_error(bc); + // MB decoding + set_offsets(pbi, 16, y_idx_mb, x_idx_mb); + vp9_decode_mb_mode_mv(pbi, xd, y_idx_mb, x_idx_mb, r); + set_refs(pbi, 16, y_idx_mb, x_idx_mb); + decode_mb(pbi, xd, y_idx_mb, x_idx_mb, r); + xd->corrupted |= bool_error(r); } } } @@ -1083,25 +999,21 @@ static void decode_sb_row(VP9D_COMP *pbi, VP9_COMMON *pc, static void setup_token_decoder(VP9D_COMP *pbi, - const unsigned char *cx_data, - BOOL_DECODER* const bool_decoder) { + const uint8_t *data, + vp9_reader *r) { VP9_COMMON *pc = &pbi->common; - const uint8_t *user_data_end = pbi->source + pbi->source_sz; - const uint8_t *partition = cx_data; - ptrdiff_t bytes_left = user_data_end - partition; - ptrdiff_t partition_size = bytes_left; + const uint8_t *data_end = pbi->source + pbi->source_sz; + const size_t partition_size = data_end - data; // Validate the calculated partition length. If the buffer // described by the partition can't be fully read, then restrict // it to the portion that can be (for EC mode) or throw an error. - if (!read_is_valid(partition, partition_size, user_data_end)) { + if (!read_is_valid(data, partition_size, data_end)) vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME, "Truncated packet or corrupt partition " "%d length", 1); - } - if (vp9_start_decode(bool_decoder, - partition, (unsigned int)partition_size)) + if (vp9_start_decode(r, data, partition_size)) vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR, "Failed to allocate bool decoder %d", 1); } @@ -1533,7 +1445,6 @@ static void decode_tiles(VP9D_COMP *pbi, const uint8_t *data, int first_partition_size, BOOL_DECODER *header_bc, BOOL_DECODER *residual_bc) { VP9_COMMON *const pc = &pbi->common; - MACROBLOCKD *const xd = &pbi->mb; const uint8_t *data_ptr = data + first_partition_size; int tile_row, tile_col, delta_log2_tiles; @@ -1587,7 +1498,7 @@ static void decode_tiles(VP9D_COMP *pbi, // Decode a row of superblocks for (mb_row = pc->cur_tile_mb_row_start; mb_row < pc->cur_tile_mb_row_end; mb_row += 4) { - decode_sb_row(pbi, pc, mb_row, xd, residual_bc); + decode_sb_row(pbi, mb_row, residual_bc); } if (tile_row == pc->tile_rows - 1 && tile_col == n_cols - 1) @@ -1612,7 +1523,7 @@ static void decode_tiles(VP9D_COMP *pbi, // Decode a row of superblocks for (mb_row = pc->cur_tile_mb_row_start; mb_row < pc->cur_tile_mb_row_end; mb_row += 4) { - decode_sb_row(pbi, pc, mb_row, xd, residual_bc); + decode_sb_row(pbi, mb_row, residual_bc); } if (has_more) { @@ -1630,7 +1541,7 @@ int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) { MACROBLOCKD *const xd = &pbi->mb; const uint8_t *data = pbi->source; const uint8_t *data_end = data + pbi->source_sz; - ptrdiff_t first_partition_length_in_bytes = 0; + size_t first_partition_size = 0; int i, corrupt_tokens = 0; // printf("Decoding frame %d\n", pc->current_video_frame); @@ -1647,9 +1558,9 @@ int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) { pc->version = (data[0] >> 1) & 7; pc->show_frame = (data[0] >> 4) & 1; scaling_active = (data[0] >> 5) & 1; - first_partition_length_in_bytes = read_le16(data + 1); + first_partition_size = read_le16(data + 1); - if (!read_is_valid(data, first_partition_length_in_bytes, data_end)) + if (!read_is_valid(data, first_partition_size, data_end)) vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME, "Truncated packet or corrupt partition 0 length"); @@ -1683,8 +1594,7 @@ int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) { pc->width, pc->height, VP9BORDERINPIXELS); - if (vp9_start_decode(&header_bc, data, - (unsigned int)first_partition_length_in_bytes)) + if (vp9_start_decode(&header_bc, data, first_partition_size)) vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR, "Failed to allocate bool decoder 0"); @@ -1858,8 +1768,7 @@ int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) { vp9_decode_mode_mvs_init(pbi, &header_bc); - decode_tiles(pbi, data, first_partition_length_in_bytes, - &header_bc, &residual_bc); + decode_tiles(pbi, data, first_partition_size, &header_bc, &residual_bc); corrupt_tokens |= xd->corrupted; // keep track of the last coded dimensions diff --git a/vp9/decoder/vp9_detokenize.c b/vp9/decoder/vp9_detokenize.c index a4ada2b7e..9077fcde1 100644 --- a/vp9/decoder/vp9_detokenize.c +++ b/vp9/decoder/vp9_detokenize.c @@ -380,182 +380,122 @@ static int get_eob(MACROBLOCKD* const xd, int segment_id, int eob_max) { return vp9_get_segdata(xd, segment_id, SEG_LVL_SKIP) ? 0 : eob_max; } -static INLINE int decode_sb(VP9D_COMP* const pbi, - MACROBLOCKD* const xd, - BOOL_DECODER* const bc, - int offset, int count, int inc, - int eob_max, TX_SIZE tx_size) { - const int segment_id = xd->mode_info_context->mbmi.segment_id; - const int seg_eob = get_eob(xd, segment_id, eob_max); +/* TODO(jkoleszar): Probably best to remove instances that require this, + * as the data likely becomes per-plane and stored in the per-plane structures. + * This is a stub to work with the existing code. + */ +static INLINE int block_idx_4x4(MACROBLOCKD* const xd, int block_size_b, + int plane, int i) { + const int luma_blocks = 1 << block_size_b; + assert(xd->plane[0].subsampling_x == 0); + assert(xd->plane[0].subsampling_y == 0); + assert(xd->plane[1].subsampling_x == 1); + assert(xd->plane[1].subsampling_y == 1); + assert(xd->plane[2].subsampling_x == 1); + assert(xd->plane[2].subsampling_y == 1); + return plane == 0 ? i : + plane == 1 ? luma_blocks + i : + luma_blocks * 5 / 4 + i; +} + +static INLINE int decode_block_plane(VP9D_COMP* const pbi, + MACROBLOCKD* const xd, + BOOL_DECODER* const bc, + BLOCK_SIZE_LG2 block_size, + int segment_id, + int plane, + int is_split) { + // block and transform sizes, in number of 4x4 blocks log 2 ("*_b") + // 4x4=0, 8x8=2, 16x16=4, 32x32=6, 64x64=8 + const TX_SIZE tx_size = xd->mode_info_context->mbmi.txfm_size; + const BLOCK_SIZE_LG2 block_size_b = block_size; + const BLOCK_SIZE_LG2 txfrm_size_b = tx_size * 2; + + // subsampled size of the block + const int ss_sum = xd->plane[plane].subsampling_x + + xd->plane[plane].subsampling_y; + const BLOCK_SIZE_LG2 ss_block_size = block_size_b - ss_sum; + + // size of the transform to use. scale the transform down if it's larger + // than the size of the subsampled data, or forced externally by the mb mode. + const int ss_max = MAX(xd->plane[plane].subsampling_x, + xd->plane[plane].subsampling_y); + const BLOCK_SIZE_LG2 ss_txfrm_size = txfrm_size_b > ss_block_size || is_split + ? txfrm_size_b - ss_max * 2 + : txfrm_size_b; + const TX_SIZE ss_tx_size = ss_txfrm_size / 2; + + // TODO(jkoleszar): 1 may not be correct here with larger chroma planes. + const int inc = is_split ? 1 : (1 << ss_txfrm_size); + + // find the maximum eob for this transform size, adjusted by segment + const int seg_eob = get_eob(xd, segment_id, 16 << ss_txfrm_size); + int i, eobtotal = 0; - assert(count == offset * 3 / 2); + assert(txfrm_size_b <= block_size_b); + assert(ss_txfrm_size <= ss_block_size); - // luma blocks - for (i = 0; i < offset; i += inc) { - const int c = decode_coefs(pbi, xd, bc, i, PLANE_TYPE_Y_WITH_DC, seg_eob, - BLOCK_OFFSET(xd->plane[0].qcoeff, i, 16), - tx_size); - xd->plane[0].eobs[i] = c; - eobtotal += c; - } + // step through the block by the size of the transform in use. + for (i = 0; i < (1 << ss_block_size); i += inc) { + const int block_idx = block_idx_4x4(xd, block_size_b, plane, i); - // chroma blocks - for (i = offset; i < offset * 5 / 4; i += inc) { - const int b = i - offset; - const int c = decode_coefs(pbi, xd, bc, i, PLANE_TYPE_UV, seg_eob, - BLOCK_OFFSET(xd->plane[1].qcoeff, b, 16), - tx_size); - xd->plane[1].eobs[b] = c; + const int c = decode_coefs(pbi, xd, bc, block_idx, + xd->plane[plane].plane_type, seg_eob, + BLOCK_OFFSET(xd->plane[plane].qcoeff, i, 16), + ss_tx_size); + xd->plane[plane].eobs[i] = c; eobtotal += c; } - for (i = offset * 5 / 4; i < count; i += inc) { - const int b = i - offset * 5 / 4; - const int c = decode_coefs(pbi, xd, bc, i, PLANE_TYPE_UV, seg_eob, - BLOCK_OFFSET(xd->plane[2].qcoeff, b, 16), - tx_size); - xd->plane[2].eobs[b] = c; - eobtotal += c; - } - return eobtotal; } -int vp9_decode_sb_tokens(VP9D_COMP* const pbi, - MACROBLOCKD* const xd, - BOOL_DECODER* const bc) { - switch (xd->mode_info_context->mbmi.txfm_size) { - case TX_32X32: { - // 32x32 luma block - const int segment_id = xd->mode_info_context->mbmi.segment_id; - int eobtotal = 0, seg_eob; - int c = decode_coefs(pbi, xd, bc, 0, PLANE_TYPE_Y_WITH_DC, - get_eob(xd, segment_id, 1024), - xd->plane[0].qcoeff, TX_32X32); - xd->plane[0].eobs[0] = c; - eobtotal += c; - - // 16x16 chroma blocks - seg_eob = get_eob(xd, segment_id, 256); - - c = decode_coefs(pbi, xd, bc, 64, PLANE_TYPE_UV, seg_eob, - xd->plane[1].qcoeff, TX_16X16); - xd->plane[1].eobs[0] = c; - eobtotal += c; - c = decode_coefs(pbi, xd, bc, 80, PLANE_TYPE_UV, seg_eob, - xd->plane[2].qcoeff, TX_16X16); - xd->plane[2].eobs[0] = c; - eobtotal += c; - return eobtotal; - } - case TX_16X16: - return decode_sb(pbi, xd, bc, 64, 96, 16, 16 * 16, TX_16X16); - case TX_8X8: - return decode_sb(pbi, xd, bc, 64, 96, 4, 8 * 8, TX_8X8); - case TX_4X4: - return decode_sb(pbi, xd, bc, 64, 96, 1, 4 * 4, TX_4X4); - default: - assert(0); - return 0; +static INLINE int decode_blocks_helper(VP9D_COMP* const pbi, + MACROBLOCKD* const xd, + BOOL_DECODER* const bc, + int block_size, + int is_split_chroma) { + const int segment_id = xd->mode_info_context->mbmi.segment_id; + int plane, eobtotal = 0; + + for (plane = 0; plane < MAX_MB_PLANE; plane++) { + const int is_split = is_split_chroma && + xd->plane[plane].plane_type == PLANE_TYPE_UV; + eobtotal += decode_block_plane(pbi, xd, bc, block_size, segment_id, + plane, is_split); } + return eobtotal; +} + +static INLINE int decode_blocks(VP9D_COMP* const pbi, + MACROBLOCKD* const xd, + BOOL_DECODER* const bc, + int block_size) { + const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode; + const TX_SIZE tx_size = xd->mode_info_context->mbmi.txfm_size; + return decode_blocks_helper(pbi, xd, bc, block_size, + tx_size == TX_8X8 && (mode == I8X8_PRED || mode == SPLITMV)); } int vp9_decode_sb64_tokens(VP9D_COMP* const pbi, MACROBLOCKD* const xd, BOOL_DECODER* const bc) { - switch (xd->mode_info_context->mbmi.txfm_size) { - case TX_32X32: - return decode_sb(pbi, xd, bc, 256, 384, 64, 32 * 32, TX_32X32); - case TX_16X16: - return decode_sb(pbi, xd, bc, 256, 384, 16, 16 * 16, TX_16X16); - case TX_8X8: - return decode_sb(pbi, xd, bc, 256, 384, 4, 8 * 8, TX_8X8); - case TX_4X4: - return decode_sb(pbi, xd, bc, 256, 384, 1, 4 * 4, TX_4X4); - default: - assert(0); - return 0; - } + return decode_blocks(pbi, xd, bc, BLOCK_64X64_LG2); } -static int vp9_decode_mb_tokens_16x16(VP9D_COMP* const pbi, - MACROBLOCKD* const xd, - BOOL_DECODER* const bc) { - const int segment_id = xd->mode_info_context->mbmi.segment_id; - int eobtotal = 0, seg_eob; - - // Luma block - int c = decode_coefs(pbi, xd, bc, 0, PLANE_TYPE_Y_WITH_DC, - get_eob(xd, segment_id, 256), - xd->plane[0].qcoeff, TX_16X16); - xd->plane[0].eobs[0] = c; - eobtotal += c; - - // 8x8 chroma blocks - seg_eob = get_eob(xd, segment_id, 64); - - c = decode_coefs(pbi, xd, bc, 16, PLANE_TYPE_UV, - seg_eob, xd->plane[1].qcoeff, TX_8X8); - xd->plane[1].eobs[0] = c; - eobtotal += c; - c = decode_coefs(pbi, xd, bc, 20, PLANE_TYPE_UV, - seg_eob, xd->plane[2].qcoeff, TX_8X8); - xd->plane[2].eobs[0] = c; - eobtotal += c; - return eobtotal; +int vp9_decode_sb_tokens(VP9D_COMP* const pbi, + MACROBLOCKD* const xd, + BOOL_DECODER* const bc) { + return decode_blocks(pbi, xd, bc, BLOCK_32X32_LG2); } -static int vp9_decode_mb_tokens_8x8(VP9D_COMP* const pbi, - MACROBLOCKD* const xd, - BOOL_DECODER* const bc) { - int i, eobtotal = 0; - const int segment_id = xd->mode_info_context->mbmi.segment_id; - - // luma blocks - int seg_eob = get_eob(xd, segment_id, 64); - for (i = 0; i < 16; i += 4) { - const int c = decode_coefs(pbi, xd, bc, i, PLANE_TYPE_Y_WITH_DC, seg_eob, - BLOCK_OFFSET(xd->plane[0].qcoeff, i, 16), - TX_8X8); - xd->plane[0].eobs[i] = c; - eobtotal += c; - } - - // chroma blocks - if (xd->mode_info_context->mbmi.mode == I8X8_PRED || - xd->mode_info_context->mbmi.mode == SPLITMV) { - // use 4x4 transform for U, V components in I8X8/splitmv prediction mode - seg_eob = get_eob(xd, segment_id, 16); - for (i = 16; i < 20; i++) { - const int c = decode_coefs(pbi, xd, bc, i, PLANE_TYPE_UV, seg_eob, - BLOCK_OFFSET(xd->plane[1].qcoeff, i - 16, 16), - TX_4X4); - xd->plane[1].eobs[i - 16] = c; - eobtotal += c; - } - for (i = 20; i < 24; i++) { - const int c = decode_coefs(pbi, xd, bc, i, PLANE_TYPE_UV, seg_eob, - BLOCK_OFFSET(xd->plane[2].qcoeff, i - 20, 16), - TX_4X4); - xd->plane[2].eobs[i - 20] = c; - eobtotal += c; - } - } else { - int c; - - c = decode_coefs(pbi, xd, bc, 16, PLANE_TYPE_UV, seg_eob, - xd->plane[1].qcoeff, TX_8X8); - xd->plane[1].eobs[0] = c; - eobtotal += c; - c = decode_coefs(pbi, xd, bc, 20, PLANE_TYPE_UV, seg_eob, - xd->plane[2].qcoeff, TX_8X8); - xd->plane[2].eobs[0] = c; - eobtotal += c; - } - - return eobtotal; +int vp9_decode_mb_tokens(VP9D_COMP* const pbi, + MACROBLOCKD* const xd, + BOOL_DECODER* const bc) { + return decode_blocks(pbi, xd, bc, BLOCK_16X16_LG2); } +#if CONFIG_NEWBINTRAMODES static int decode_coefs_4x4(VP9D_COMP *dx, MACROBLOCKD *xd, BOOL_DECODER* const bc, PLANE_TYPE type, int i, int seg_eob) { @@ -588,39 +528,6 @@ int vp9_decode_mb_tokens_4x4_uv(VP9D_COMP* const dx, return decode_mb_tokens_4x4_uv(dx, xd, bc, seg_eob); } -static int vp9_decode_mb_tokens_4x4(VP9D_COMP* const dx, - MACROBLOCKD* const xd, - BOOL_DECODER* const bc) { - int i, eobtotal = 0; - const int segment_id = xd->mode_info_context->mbmi.segment_id; - const int seg_eob = get_eob(xd, segment_id, 16); - - // luma blocks - for (i = 0; i < 16; ++i) - eobtotal += decode_coefs_4x4(dx, xd, bc, PLANE_TYPE_Y_WITH_DC, i, seg_eob); - - // chroma blocks - eobtotal += decode_mb_tokens_4x4_uv(dx, xd, bc, seg_eob); - - return eobtotal; -} - -int vp9_decode_mb_tokens(VP9D_COMP* const dx, - MACROBLOCKD* const xd, - BOOL_DECODER* const bc) { - const TX_SIZE tx_size = xd->mode_info_context->mbmi.txfm_size; - switch (tx_size) { - case TX_16X16: - return vp9_decode_mb_tokens_16x16(dx, xd, bc); - case TX_8X8: - return vp9_decode_mb_tokens_8x8(dx, xd, bc); - default: - assert(tx_size == TX_4X4); - return vp9_decode_mb_tokens_4x4(dx, xd, bc); - } -} - -#if CONFIG_NEWBINTRAMODES int vp9_decode_coefs_4x4(VP9D_COMP *dx, MACROBLOCKD *xd, BOOL_DECODER* const bc, PLANE_TYPE type, int i) { diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c index 566fb60ef..4df117088 100644 --- a/vp9/encoder/vp9_rdopt.c +++ b/vp9/encoder/vp9_rdopt.c @@ -3711,7 +3711,11 @@ static int64_t handle_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, for (i = 0; i < num_refs; ++i) { cur_mv[i] = frame_mv[this_mode][refs[i]]; // Clip "next_nearest" so that it does not extend to far out of image - clamp_mv2(&cur_mv[i], xd); + if (this_mode == NEWMV) + assert(!clamp_mv2(&cur_mv[i], xd)); + else + clamp_mv2(&cur_mv[i], xd); + if (mv_check_bounds(x, &cur_mv[i])) return INT64_MAX; mbmi->mv[i].as_int = cur_mv[i].as_int; |