1 /* 2 * Copyright (c) 2016, Alliance for Open Media. All rights reserved 3 * 4 * This source code is subject to the terms of the BSD 2 Clause License and 5 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License 6 * was not distributed with this source code in the LICENSE file, you can 7 * obtain it at www.aomedia.org/license/software. If the Alliance for Open 8 * Media Patent License 1.0 was not distributed with this source code in the 9 * PATENTS file, you can obtain it at www.aomedia.org/license/patent. 10 */ 11 12 #ifndef AV1_ENCODER_BLOCK_H_ 13 #define AV1_ENCODER_BLOCK_H_ 14 15 #include "av1/common/entropymv.h" 16 #include "av1/common/entropy.h" 17 #if CONFIG_PVQ 18 #include "av1/encoder/encint.h" 19 #endif 20 #include "av1/common/mvref_common.h" 21 #include "av1/encoder/hash.h" 22 #if CONFIG_DIST_8X8 23 #include "aom/aomcx.h" 24 #endif 25 26 #ifdef __cplusplus 27 extern "C" { 28 #endif 29 30 #if CONFIG_PVQ 31 // Maximum possible # of tx blocks in luma plane, which is currently 256, 32 // since there can be 16x16 of 4x4 tx. 33 #define MAX_PVQ_BLOCKS_IN_SB (MAX_SB_SQUARE >> 2 * OD_LOG_BSIZE0) 34 #endif 35 36 typedef struct { 37 unsigned int sse; 38 int sum; 39 unsigned int var; 40 } DIFF; 41 42 typedef struct macroblock_plane { 43 DECLARE_ALIGNED(16, int16_t, src_diff[MAX_SB_SQUARE]); 44 #if CONFIG_PVQ 45 DECLARE_ALIGNED(16, int16_t, src_int16[MAX_SB_SQUARE]); 46 #endif 47 tran_low_t *qcoeff; 48 tran_low_t *coeff; 49 uint16_t *eobs; 50 #if CONFIG_LV_MAP 51 uint8_t *txb_entropy_ctx; 52 #endif 53 struct buf_2d src; 54 55 // Quantizer setings 56 const int16_t *quant_fp; 57 const int16_t *round_fp; 58 const int16_t *quant; 59 const int16_t *quant_shift; 60 const int16_t *zbin; 61 const int16_t *round; 62 #if CONFIG_NEW_QUANT 63 const cuml_bins_type_nuq *cuml_bins_nuq[QUANT_PROFILES]; 64 #endif // CONFIG_NEW_QUANT 65 } MACROBLOCK_PLANE; 66 67 typedef int av1_coeff_cost[PLANE_TYPES][REF_TYPES][COEF_BANDS][COEFF_CONTEXTS] 68 [TAIL_TOKENS]; 69 70 #if CONFIG_LV_MAP 71 typedef struct { 72 int txb_skip_cost[TXB_SKIP_CONTEXTS][2]; 73 int nz_map_cost[SIG_COEF_CONTEXTS][2]; 74 int eob_cost[EOB_COEF_CONTEXTS][2]; 75 int dc_sign_cost[DC_SIGN_CONTEXTS][2]; 76 int base_cost[NUM_BASE_LEVELS][COEFF_BASE_CONTEXTS][2]; 77 #if BR_NODE 78 int lps_cost[LEVEL_CONTEXTS][COEFF_BASE_RANGE + 1]; 79 int br_cost[BASE_RANGE_SETS][LEVEL_CONTEXTS][2]; 80 #else // BR_NODE 81 int lps_cost[LEVEL_CONTEXTS][2]; 82 #endif // BR_NODE 83 #if CONFIG_CTX1D 84 int eob_mode_cost[TX_CLASSES][2]; 85 int empty_line_cost[TX_CLASSES][EMPTY_LINE_CONTEXTS][2]; 86 int hv_eob_cost[TX_CLASSES][HV_EOB_CONTEXTS][2]; 87 #endif 88 } LV_MAP_COEFF_COST; 89 90 typedef struct { 91 tran_low_t tcoeff[MAX_MB_PLANE][MAX_SB_SQUARE]; 92 uint16_t eobs[MAX_MB_PLANE][MAX_SB_SQUARE / (TX_SIZE_W_MIN * TX_SIZE_H_MIN)]; 93 uint8_t txb_skip_ctx[MAX_MB_PLANE] 94 [MAX_SB_SQUARE / (TX_SIZE_W_MIN * TX_SIZE_H_MIN)]; 95 int dc_sign_ctx[MAX_MB_PLANE] 96 [MAX_SB_SQUARE / (TX_SIZE_W_MIN * TX_SIZE_H_MIN)]; 97 } CB_COEFF_BUFFER; 98 #endif 99 100 typedef struct { 101 int_mv ref_mvs[MODE_CTX_REF_FRAMES][MAX_MV_REF_CANDIDATES]; 102 int16_t mode_context[MODE_CTX_REF_FRAMES]; 103 #if CONFIG_LV_MAP 104 // TODO(angiebird): Reduce the buffer size according to sb_type 105 tran_low_t *tcoeff[MAX_MB_PLANE]; 106 uint16_t *eobs[MAX_MB_PLANE]; 107 uint8_t *txb_skip_ctx[MAX_MB_PLANE]; 108 int *dc_sign_ctx[MAX_MB_PLANE]; 109 #endif 110 uint8_t ref_mv_count[MODE_CTX_REF_FRAMES]; 111 CANDIDATE_MV ref_mv_stack[MODE_CTX_REF_FRAMES][MAX_REF_MV_STACK_SIZE]; 112 int16_t compound_mode_context[MODE_CTX_REF_FRAMES]; 113 } MB_MODE_INFO_EXT; 114 115 typedef struct { 116 int col_min; 117 int col_max; 118 int row_min; 119 int row_max; 120 } MvLimits; 121 122 typedef struct { 123 uint8_t best_palette_color_map[MAX_SB_SQUARE]; 124 float kmeans_data_buf[2 * MAX_SB_SQUARE]; 125 } PALETTE_BUFFER; 126 127 typedef struct { 128 TX_TYPE tx_type; 129 TX_SIZE tx_size; 130 #if CONFIG_VAR_TX 131 TX_SIZE min_tx_size; 132 TX_SIZE inter_tx_size[MAX_MIB_SIZE][MAX_MIB_SIZE]; 133 uint8_t blk_skip[MAX_MIB_SIZE * MAX_MIB_SIZE * 8]; 134 #endif // CONFIG_VAR_TX 135 #if CONFIG_TXK_SEL 136 TX_TYPE txk_type[MAX_SB_SQUARE / (TX_SIZE_W_MIN * TX_SIZE_H_MIN)]; 137 #endif // CONFIG_TXK_SEL 138 RD_STATS rd_stats; 139 uint32_t hash_value; 140 } TX_RD_INFO; 141 142 #define RD_RECORD_BUFFER_LEN 8 143 typedef struct { 144 TX_RD_INFO tx_rd_info[RD_RECORD_BUFFER_LEN]; // Circular buffer. 145 int index_start; 146 int num; 147 CRC_CALCULATOR crc_calculator; // Hash function. 148 } TX_RD_RECORD; 149 150 typedef struct macroblock MACROBLOCK; 151 struct macroblock { 152 struct macroblock_plane plane[MAX_MB_PLANE]; 153 154 // Save the transform RD search info. 155 TX_RD_RECORD tx_rd_record; 156 157 MACROBLOCKD e_mbd; 158 MB_MODE_INFO_EXT *mbmi_ext; 159 int skip_block; 160 int qindex; 161 162 // The equivalent error at the current rdmult of one whole bit (not one 163 // bitcost unit). 164 int errorperbit; 165 // The equivalend SAD error of one (whole) bit at the current quantizer 166 // for large blocks. 167 int sadperbit16; 168 // The equivalend SAD error of one (whole) bit at the current quantizer 169 // for sub-8x8 blocks. 170 int sadperbit4; 171 int rdmult; 172 int mb_energy; 173 int *m_search_count_ptr; 174 int *ex_search_count_ptr; 175 176 #if CONFIG_VAR_TX 177 unsigned int txb_split_count; 178 #endif 179 180 // These are set to their default values at the beginning, and then adjusted 181 // further in the encoding process. 182 BLOCK_SIZE min_partition_size; 183 BLOCK_SIZE max_partition_size; 184 185 int mv_best_ref_index[TOTAL_REFS_PER_FRAME]; 186 unsigned int max_mv_context[TOTAL_REFS_PER_FRAME]; 187 unsigned int source_variance; 188 unsigned int pred_sse[TOTAL_REFS_PER_FRAME]; 189 int pred_mv_sad[TOTAL_REFS_PER_FRAME]; 190 191 int *nmvjointcost; 192 int nmv_vec_cost[NMV_CONTEXTS][MV_JOINTS]; 193 int *nmvcost[NMV_CONTEXTS][2]; 194 int *nmvcost_hp[NMV_CONTEXTS][2]; 195 int **mv_cost_stack[NMV_CONTEXTS]; 196 int **mvcost; 197 198 #if CONFIG_MOTION_VAR 199 int32_t *wsrc_buf; 200 int32_t *mask_buf; 201 uint8_t *above_pred_buf; 202 uint8_t *left_pred_buf; 203 #endif // CONFIG_MOTION_VAR 204 205 PALETTE_BUFFER *palette_buffer; 206 207 // These define limits to motion vector components to prevent them 208 // from extending outside the UMV borders 209 MvLimits mv_limits; 210 211 #if CONFIG_VAR_TX 212 uint8_t blk_skip[MAX_MB_PLANE][MAX_MIB_SIZE * MAX_MIB_SIZE * 8]; 213 uint8_t blk_skip_drl[MAX_MB_PLANE][MAX_MIB_SIZE * MAX_MIB_SIZE * 8]; 214 #endif 215 216 int skip; 217 218 #if CONFIG_CB4X4 219 int skip_chroma_rd; 220 #endif 221 222 #if CONFIG_LV_MAP 223 LV_MAP_COEFF_COST coeff_costs[TX_SIZES][PLANE_TYPES]; 224 uint16_t cb_offset; 225 #endif 226 227 av1_coeff_cost token_head_costs[TX_SIZES]; 228 av1_coeff_cost token_tail_costs[TX_SIZES]; 229 230 // mode costs 231 int mbmode_cost[BLOCK_SIZE_GROUPS][INTRA_MODES]; 232 int newmv_mode_cost[NEWMV_MODE_CONTEXTS][2]; 233 int zeromv_mode_cost[ZEROMV_MODE_CONTEXTS][2]; 234 int refmv_mode_cost[REFMV_MODE_CONTEXTS][2]; 235 int drl_mode_cost0[DRL_MODE_CONTEXTS][2]; 236 237 int inter_compound_mode_cost[INTER_MODE_CONTEXTS][INTER_COMPOUND_MODES]; 238 int compound_type_cost[BLOCK_SIZES_ALL][COMPOUND_TYPES]; 239 #if CONFIG_COMPOUND_SINGLEREF 240 int inter_singleref_comp_mode_cost[INTER_MODE_CONTEXTS] 241 [INTER_SINGLEREF_COMP_MODES]; 242 #endif // CONFIG_COMPOUND_SINGLEREF 243 #if CONFIG_INTERINTRA 244 int interintra_mode_cost[BLOCK_SIZE_GROUPS][INTERINTRA_MODES]; 245 #endif // CONFIG_INTERINTRA 246 #if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION 247 int motion_mode_cost[BLOCK_SIZES_ALL][MOTION_MODES]; 248 #if CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION 249 int motion_mode_cost1[BLOCK_SIZES_ALL][2]; 250 #if CONFIG_NCOBMC_ADAPT_WEIGHT 251 int motion_mode_cost2[BLOCK_SIZES_ALL][OBMC_FAMILY_MODES]; 252 #endif 253 #endif // CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION 254 #if CONFIG_MOTION_VAR && CONFIG_NCOBMC_ADAPT_WEIGHT 255 int ncobmc_mode_cost[ADAPT_OVERLAP_BLOCKS][MAX_NCOBMC_MODES]; 256 #endif // CONFIG_MOTION_VAR && CONFIG_NCOBMC_ADAPT_WEIGHT 257 #endif // CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION 258 int intra_uv_mode_cost[INTRA_MODES][UV_INTRA_MODES]; 259 int y_mode_costs[INTRA_MODES][INTRA_MODES][INTRA_MODES]; 260 int switchable_interp_costs[SWITCHABLE_FILTER_CONTEXTS][SWITCHABLE_FILTERS]; 261 #if CONFIG_EXT_PARTITION_TYPES 262 int partition_cost[PARTITION_CONTEXTS + CONFIG_UNPOISON_PARTITION_CTX] 263 [EXT_PARTITION_TYPES]; 264 #else 265 int partition_cost[PARTITION_CONTEXTS + CONFIG_UNPOISON_PARTITION_CTX] 266 [PARTITION_TYPES]; 267 #endif // CONFIG_EXT_PARTITION_TYPES 268 #if CONFIG_MRC_TX 269 int mrc_mask_inter_cost[PALETTE_SIZES][PALETTE_COLOR_INDEX_CONTEXTS] 270 [PALETTE_COLORS]; 271 int mrc_mask_intra_cost[PALETTE_SIZES][PALETTE_COLOR_INDEX_CONTEXTS] 272 [PALETTE_COLORS]; 273 #endif // CONFIG_MRC_TX 274 int palette_y_size_cost[PALETTE_BLOCK_SIZES][PALETTE_SIZES]; 275 int palette_uv_size_cost[PALETTE_BLOCK_SIZES][PALETTE_SIZES]; 276 int palette_y_color_cost[PALETTE_SIZES][PALETTE_COLOR_INDEX_CONTEXTS] 277 [PALETTE_COLORS]; 278 int palette_uv_color_cost[PALETTE_SIZES][PALETTE_COLOR_INDEX_CONTEXTS] 279 [PALETTE_COLORS]; 280 #if CONFIG_CFL 281 // The rate associated with each alpha codeword 282 int cfl_cost[CFL_JOINT_SIGNS][CFL_PRED_PLANES][CFL_ALPHABET_SIZE]; 283 #endif // CONFIG_CFL 284 int tx_size_cost[TX_SIZES - 1][TX_SIZE_CONTEXTS][TX_SIZES]; 285 #if CONFIG_EXT_TX 286 #if CONFIG_LGT_FROM_PRED 287 int intra_lgt_cost[LGT_SIZES][INTRA_MODES][2]; 288 int inter_lgt_cost[LGT_SIZES][2]; 289 #endif 290 int inter_tx_type_costs[EXT_TX_SETS_INTER][EXT_TX_SIZES][TX_TYPES]; 291 int intra_tx_type_costs[EXT_TX_SETS_INTRA][EXT_TX_SIZES][INTRA_MODES] 292 [TX_TYPES]; 293 #else 294 int intra_tx_type_costs[EXT_TX_SIZES][TX_TYPES][TX_TYPES]; 295 int inter_tx_type_costs[EXT_TX_SIZES][TX_TYPES]; 296 #endif // CONFIG_EXT_TX 297 #if CONFIG_EXT_INTRA 298 #if CONFIG_INTRA_INTERP 299 int intra_filter_cost[INTRA_FILTERS + 1][INTRA_FILTERS]; 300 #endif // CONFIG_INTRA_INTERP 301 #endif // CONFIG_EXT_INTRA 302 #if CONFIG_LOOP_RESTORATION 303 int switchable_restore_cost[RESTORE_SWITCHABLE_TYPES]; 304 #endif // CONFIG_LOOP_RESTORATION 305 #if CONFIG_INTRABC 306 int intrabc_cost[2]; 307 #endif // CONFIG_INTRABC 308 309 int optimize; 310 311 // Used to store sub partition's choices. 312 MV pred_mv[TOTAL_REFS_PER_FRAME]; 313 314 // Store the best motion vector during motion search 315 int_mv best_mv; 316 // Store the second best motion vector during full-pixel motion search 317 int_mv second_best_mv; 318 319 // use default transform and skip transform type search for intra modes 320 int use_default_intra_tx_type; 321 // use default transform and skip transform type search for inter modes 322 int use_default_inter_tx_type; 323 #if CONFIG_PVQ 324 int rate; 325 // 1 if neither AC nor DC is coded. Only used during RDO. 326 int pvq_skip[MAX_MB_PLANE]; 327 PVQ_QUEUE *pvq_q; 328 329 // Storage for PVQ tx block encodings in a superblock. 330 // There can be max 16x16 of 4x4 blocks (and YUV) encode by PVQ 331 // 256 is the max # of 4x4 blocks in a SB (64x64), which comes from: 332 // 1) Since PVQ is applied to each trasnform-ed block 333 // 2) 4x4 is the smallest tx size in AV1 334 // 3) AV1 allows using smaller tx size than block (i.e. partition) size 335 // TODO(yushin) : The memory usage could be improved a lot, since this has 336 // storage for 10 bands and 128 coefficients for every 4x4 block, 337 PVQ_INFO pvq[MAX_PVQ_BLOCKS_IN_SB][MAX_MB_PLANE]; 338 daala_enc_ctx daala_enc; 339 int pvq_speed; 340 int pvq_coded; // Indicates whether pvq_info needs be stored to tokenize 341 #endif 342 #if CONFIG_DIST_8X8 343 int using_dist_8x8; 344 aom_tune_metric tune_metric; 345 #if CONFIG_CB4X4 346 #if CONFIG_HIGHBITDEPTH 347 DECLARE_ALIGNED(16, uint16_t, decoded_8x8[8 * 8]); 348 #else 349 DECLARE_ALIGNED(16, uint8_t, decoded_8x8[8 * 8]); 350 #endif 351 #endif // CONFIG_CB4X4 352 #endif // CONFIG_DIST_8X8 353 }; 354 355 #ifdef __cplusplus 356 } // extern "C" 357 #endif 358 359 #endif // AV1_ENCODER_BLOCK_H_ 360