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 AOM_AV1_COMMON_ONYXC_INT_H_ 13 #define AOM_AV1_COMMON_ONYXC_INT_H_ 14 15 #include "config/aom_config.h" 16 #include "config/av1_rtcd.h" 17 18 #include "aom/internal/aom_codec_internal.h" 19 #include "aom_util/aom_thread.h" 20 #include "av1/common/alloccommon.h" 21 #include "av1/common/av1_loopfilter.h" 22 #include "av1/common/entropy.h" 23 #include "av1/common/entropymode.h" 24 #include "av1/common/entropymv.h" 25 #include "av1/common/enums.h" 26 #include "av1/common/frame_buffers.h" 27 #include "av1/common/mv.h" 28 #include "av1/common/quant_common.h" 29 #include "av1/common/restoration.h" 30 #include "av1/common/tile_common.h" 31 #include "av1/common/timing.h" 32 #include "av1/common/odintrin.h" 33 #include "av1/encoder/hash_motion.h" 34 #include "aom_dsp/grain_synthesis.h" 35 #include "aom_dsp/grain_table.h" 36 #ifdef __cplusplus 37 extern "C" { 38 #endif 39 40 #if defined(__clang__) && defined(__has_warning) 41 #if __has_feature(cxx_attributes) && __has_warning("-Wimplicit-fallthrough") 42 #define AOM_FALLTHROUGH_INTENDED [[clang::fallthrough]] // NOLINT 43 #endif 44 #elif defined(__GNUC__) && __GNUC__ >= 7 45 #define AOM_FALLTHROUGH_INTENDED __attribute__((fallthrough)) // NOLINT 46 #endif 47 48 #ifndef AOM_FALLTHROUGH_INTENDED 49 #define AOM_FALLTHROUGH_INTENDED \ 50 do { \ 51 } while (0) 52 #endif 53 54 #define CDEF_MAX_STRENGTHS 16 55 56 /* Constant values while waiting for the sequence header */ 57 #define FRAME_ID_LENGTH 15 58 #define DELTA_FRAME_ID_LENGTH 14 59 60 #define FRAME_CONTEXTS (FRAME_BUFFERS + 1) 61 // Extra frame context which is always kept at default values 62 #define FRAME_CONTEXT_DEFAULTS (FRAME_CONTEXTS - 1) 63 #define PRIMARY_REF_BITS 3 64 #define PRIMARY_REF_NONE 7 65 66 #define NUM_PING_PONG_BUFFERS 2 67 68 #define MAX_NUM_TEMPORAL_LAYERS 8 69 #define MAX_NUM_SPATIAL_LAYERS 4 70 /* clang-format off */ 71 // clang-format seems to think this is a pointer dereference and not a 72 // multiplication. 73 #define MAX_NUM_OPERATING_POINTS \ 74 MAX_NUM_TEMPORAL_LAYERS * MAX_NUM_SPATIAL_LAYERS 75 /* clang-format on*/ 76 77 // TODO(jingning): Turning this on to set up transform coefficient 78 // processing timer. 79 #define TXCOEFF_TIMER 0 80 #define TXCOEFF_COST_TIMER 0 81 82 enum { 83 SINGLE_REFERENCE = 0, 84 COMPOUND_REFERENCE = 1, 85 REFERENCE_MODE_SELECT = 2, 86 REFERENCE_MODES = 3, 87 } UENUM1BYTE(REFERENCE_MODE); 88 89 enum { 90 /** 91 * Frame context updates are disabled 92 */ 93 REFRESH_FRAME_CONTEXT_DISABLED, 94 /** 95 * Update frame context to values resulting from backward probability 96 * updates based on entropy/counts in the decoded frame 97 */ 98 REFRESH_FRAME_CONTEXT_BACKWARD, 99 } UENUM1BYTE(REFRESH_FRAME_CONTEXT_MODE); 100 101 #define MFMV_STACK_SIZE 3 102 typedef struct { 103 int_mv mfmv0; 104 uint8_t ref_frame_offset; 105 } TPL_MV_REF; 106 107 typedef struct { 108 int_mv mv; 109 MV_REFERENCE_FRAME ref_frame; 110 } MV_REF; 111 112 113 typedef struct RefCntBuffer { 114 // For a RefCntBuffer, the following are reference-holding variables: 115 // - cm->ref_frame_map[] 116 // - cm->cur_frame 117 // - cm->scaled_ref_buf[] (encoder only) 118 // - cm->next_ref_frame_map[] (decoder only) 119 // - pbi->output_frame_index[] (decoder only) 120 // With that definition, 'ref_count' is the number of reference-holding 121 // variables that are currently referencing this buffer. 122 // For example: 123 // - suppose this buffer is at index 'k' in the buffer pool, and 124 // - Total 'n' of the variables / array elements above have value 'k' (that 125 // is, they are pointing to buffer at index 'k'). 126 // Then, pool->frame_bufs[k].ref_count = n. 127 int ref_count; 128 129 unsigned int order_hint; 130 unsigned int ref_order_hints[INTER_REFS_PER_FRAME]; 131 132 MV_REF *mvs; 133 uint8_t *seg_map; 134 struct segmentation seg; 135 int mi_rows; 136 int mi_cols; 137 // Width and height give the size of the buffer (before any upscaling, unlike 138 // the sizes that can be derived from the buf structure) 139 int width; 140 int height; 141 WarpedMotionParams global_motion[REF_FRAMES]; 142 int showable_frame; // frame can be used as show existing frame in future 143 uint8_t film_grain_params_present; 144 aom_film_grain_t film_grain_params; 145 aom_codec_frame_buffer_t raw_frame_buffer; 146 YV12_BUFFER_CONFIG buf; 147 hash_table hash_table; 148 FRAME_TYPE frame_type; 149 150 // This is only used in the encoder but needs to be indexed per ref frame 151 // so it's extremely convenient to keep it here. 152 int interp_filter_selected[SWITCHABLE]; 153 154 // Inter frame reference frame delta for loop filter 155 int8_t ref_deltas[REF_FRAMES]; 156 157 // 0 = ZERO_MV, MV 158 int8_t mode_deltas[MAX_MODE_LF_DELTAS]; 159 160 FRAME_CONTEXT frame_context; 161 } RefCntBuffer; 162 163 typedef struct BufferPool { 164 // Protect BufferPool from being accessed by several FrameWorkers at 165 // the same time during frame parallel decode. 166 // TODO(hkuang): Try to use atomic variable instead of locking the whole pool. 167 // TODO(wtc): Remove this. See 168 // https://chromium-review.googlesource.com/c/webm/libvpx/+/560630. 169 #if CONFIG_MULTITHREAD 170 pthread_mutex_t pool_mutex; 171 #endif 172 173 // Private data associated with the frame buffer callbacks. 174 void *cb_priv; 175 176 aom_get_frame_buffer_cb_fn_t get_fb_cb; 177 aom_release_frame_buffer_cb_fn_t release_fb_cb; 178 179 RefCntBuffer frame_bufs[FRAME_BUFFERS]; 180 181 // Frame buffers allocated internally by the codec. 182 InternalFrameBufferList int_frame_buffers; 183 } BufferPool; 184 185 typedef struct { 186 int cdef_pri_damping; 187 int cdef_sec_damping; 188 int nb_cdef_strengths; 189 int cdef_strengths[CDEF_MAX_STRENGTHS]; 190 int cdef_uv_strengths[CDEF_MAX_STRENGTHS]; 191 int cdef_bits; 192 } CdefInfo; 193 194 typedef struct { 195 int delta_q_present_flag; 196 // Resolution of delta quant 197 int delta_q_res; 198 int delta_lf_present_flag; 199 // Resolution of delta lf level 200 int delta_lf_res; 201 // This is a flag for number of deltas of loop filter level 202 // 0: use 1 delta, for y_vertical, y_horizontal, u, and v 203 // 1: use separate deltas for each filter level 204 int delta_lf_multi; 205 } DeltaQInfo; 206 207 typedef struct { 208 int enable_order_hint; // 0 - disable order hint, and related tools 209 int order_hint_bits_minus_1; // dist_wtd_comp, ref_frame_mvs, 210 // frame_sign_bias 211 // if 0, enable_dist_wtd_comp and 212 // enable_ref_frame_mvs must be set as 0. 213 int enable_dist_wtd_comp; // 0 - disable dist-wtd compound modes 214 // 1 - enable it 215 int enable_ref_frame_mvs; // 0 - disable ref frame mvs 216 // 1 - enable it 217 } OrderHintInfo; 218 219 // Sequence header structure. 220 // Note: All syntax elements of sequence_header_obu that need to be 221 // bit-identical across multiple sequence headers must be part of this struct, 222 // so that consistency is checked by are_seq_headers_consistent() function. 223 typedef struct SequenceHeader { 224 int num_bits_width; 225 int num_bits_height; 226 int max_frame_width; 227 int max_frame_height; 228 uint8_t frame_id_numbers_present_flag; 229 int frame_id_length; 230 int delta_frame_id_length; 231 BLOCK_SIZE sb_size; // Size of the superblock used for this frame 232 int mib_size; // Size of the superblock in units of MI blocks 233 int mib_size_log2; // Log 2 of above. 234 235 OrderHintInfo order_hint_info; 236 237 uint8_t force_screen_content_tools; // 0 - force off 238 // 1 - force on 239 // 2 - adaptive 240 uint8_t still_picture; // Video is a single frame still picture 241 uint8_t reduced_still_picture_hdr; // Use reduced header for still picture 242 uint8_t force_integer_mv; // 0 - Don't force. MV can use subpel 243 // 1 - force to integer 244 // 2 - adaptive 245 uint8_t enable_filter_intra; // enables/disables filterintra 246 uint8_t enable_intra_edge_filter; // enables/disables edge upsampling 247 uint8_t enable_interintra_compound; // enables/disables interintra_compound 248 uint8_t enable_masked_compound; // enables/disables masked compound 249 uint8_t enable_dual_filter; // 0 - disable dual interpolation filter 250 // 1 - enable vert/horz filter selection 251 uint8_t enable_warped_motion; // 0 - disable warp for the sequence 252 // 1 - enable warp for the sequence 253 uint8_t enable_superres; // 0 - Disable superres for the sequence 254 // and no frame level superres flag 255 // 1 - Enable superres for the sequence 256 // enable per-frame superres flag 257 uint8_t enable_cdef; // To turn on/off CDEF 258 uint8_t enable_restoration; // To turn on/off loop restoration 259 BITSTREAM_PROFILE profile; 260 261 // Operating point info. 262 int operating_points_cnt_minus_1; 263 int operating_point_idc[MAX_NUM_OPERATING_POINTS]; 264 uint8_t display_model_info_present_flag; 265 uint8_t decoder_model_info_present_flag; 266 AV1_LEVEL seq_level_idx[MAX_NUM_OPERATING_POINTS]; 267 uint8_t tier[MAX_NUM_OPERATING_POINTS]; // seq_tier in the spec. One bit: 0 268 // or 1. 269 270 // Color config. 271 aom_bit_depth_t bit_depth; // AOM_BITS_8 in profile 0 or 1, 272 // AOM_BITS_10 or AOM_BITS_12 in profile 2 or 3. 273 uint8_t use_highbitdepth; // If true, we need to use 16bit frame buffers. 274 uint8_t monochrome; // Monochorme video 275 aom_color_primaries_t color_primaries; 276 aom_transfer_characteristics_t transfer_characteristics; 277 aom_matrix_coefficients_t matrix_coefficients; 278 int color_range; 279 int subsampling_x; // Chroma subsampling for x 280 int subsampling_y; // Chroma subsampling for y 281 aom_chroma_sample_position_t chroma_sample_position; 282 uint8_t separate_uv_delta_q; 283 uint8_t film_grain_params_present; 284 } SequenceHeader; 285 286 typedef struct { 287 int skip_mode_allowed; 288 int skip_mode_flag; 289 int ref_frame_idx_0; 290 int ref_frame_idx_1; 291 } SkipModeInfo; 292 293 typedef struct { 294 FRAME_TYPE frame_type; 295 REFERENCE_MODE reference_mode; 296 297 unsigned int order_hint; 298 unsigned int frame_number; 299 SkipModeInfo skip_mode_info; 300 int refresh_frame_flags; // Which ref frames are overwritten by this frame 301 int frame_refs_short_signaling; 302 } CurrentFrame; 303 304 typedef struct AV1Common { 305 CurrentFrame current_frame; 306 struct aom_internal_error_info error; 307 int width; 308 int height; 309 int render_width; 310 int render_height; 311 int timing_info_present; 312 aom_timing_info_t timing_info; 313 int buffer_removal_time_present; 314 aom_dec_model_info_t buffer_model; 315 aom_dec_model_op_parameters_t op_params[MAX_NUM_OPERATING_POINTS + 1]; 316 aom_op_timing_info_t op_frame_timing[MAX_NUM_OPERATING_POINTS + 1]; 317 uint32_t frame_presentation_time; 318 319 int context_update_tile_id; 320 321 // Scale of the current frame with respect to itself. 322 struct scale_factors sf_identity; 323 324 RefCntBuffer *prev_frame; 325 326 // TODO(hkuang): Combine this with cur_buf in macroblockd. 327 RefCntBuffer *cur_frame; 328 329 // For encoder, we have a two-level mapping from reference frame type to the 330 // corresponding buffer in the buffer pool: 331 // * 'remapped_ref_idx[i - 1]' maps reference type 'i' (range: LAST_FRAME ... 332 // EXTREF_FRAME) to a remapped index 'j' (in range: 0 ... REF_FRAMES - 1) 333 // * Later, 'cm->ref_frame_map[j]' maps the remapped index 'j' to a pointer to 334 // the reference counted buffer structure RefCntBuffer, taken from the buffer 335 // pool cm->buffer_pool->frame_bufs. 336 // 337 // LAST_FRAME, ..., EXTREF_FRAME 338 // | | 339 // v v 340 // remapped_ref_idx[LAST_FRAME - 1], ..., remapped_ref_idx[EXTREF_FRAME - 1] 341 // | | 342 // v v 343 // ref_frame_map[], ..., ref_frame_map[] 344 // 345 // Note: INTRA_FRAME always refers to the current frame, so there's no need to 346 // have a remapped index for the same. 347 int remapped_ref_idx[REF_FRAMES]; 348 349 struct scale_factors ref_scale_factors[REF_FRAMES]; 350 351 // For decoder, ref_frame_map[i] maps reference type 'i' to a pointer to 352 // the buffer in the buffer pool 'cm->buffer_pool.frame_bufs'. 353 // For encoder, ref_frame_map[j] (where j = remapped_ref_idx[i]) maps 354 // remapped reference index 'j' (that is, original reference type 'i') to 355 // a pointer to the buffer in the buffer pool 'cm->buffer_pool.frame_bufs'. 356 RefCntBuffer *ref_frame_map[REF_FRAMES]; 357 358 // Prepare ref_frame_map for the next frame. 359 // Only used in frame parallel decode. 360 RefCntBuffer *next_ref_frame_map[REF_FRAMES]; 361 FRAME_TYPE last_frame_type; /* last frame's frame type for motion search.*/ 362 363 int show_frame; 364 int showable_frame; // frame can be used as show existing frame in future 365 int show_existing_frame; 366 367 uint8_t disable_cdf_update; 368 int allow_high_precision_mv; 369 uint8_t cur_frame_force_integer_mv; // 0 the default in AOM, 1 only integer 370 371 uint8_t allow_screen_content_tools; 372 int allow_intrabc; 373 int allow_warped_motion; 374 375 // MBs, mb_rows/cols is in 16-pixel units; mi_rows/cols is in 376 // MB_MODE_INFO (8-pixel) units. 377 int MBs; 378 int mb_rows, mi_rows; 379 int mb_cols, mi_cols; 380 int mi_stride; 381 382 /* profile settings */ 383 TX_MODE tx_mode; 384 385 #if CONFIG_ENTROPY_STATS 386 int coef_cdf_category; 387 #endif 388 389 int base_qindex; 390 int y_dc_delta_q; 391 int u_dc_delta_q; 392 int v_dc_delta_q; 393 int u_ac_delta_q; 394 int v_ac_delta_q; 395 396 // The dequantizers below are true dequantizers used only in the 397 // dequantization process. They have the same coefficient 398 // shift/scale as TX. 399 int16_t y_dequant_QTX[MAX_SEGMENTS][2]; 400 int16_t u_dequant_QTX[MAX_SEGMENTS][2]; 401 int16_t v_dequant_QTX[MAX_SEGMENTS][2]; 402 403 // Global quant matrix tables 404 const qm_val_t *giqmatrix[NUM_QM_LEVELS][3][TX_SIZES_ALL]; 405 const qm_val_t *gqmatrix[NUM_QM_LEVELS][3][TX_SIZES_ALL]; 406 407 // Local quant matrix tables for each frame 408 const qm_val_t *y_iqmatrix[MAX_SEGMENTS][TX_SIZES_ALL]; 409 const qm_val_t *u_iqmatrix[MAX_SEGMENTS][TX_SIZES_ALL]; 410 const qm_val_t *v_iqmatrix[MAX_SEGMENTS][TX_SIZES_ALL]; 411 412 // Encoder 413 int using_qmatrix; 414 int qm_y; 415 int qm_u; 416 int qm_v; 417 int min_qmlevel; 418 int max_qmlevel; 419 int use_quant_b_adapt; 420 421 /* We allocate a MB_MODE_INFO struct for each macroblock, together with 422 an extra row on top and column on the left to simplify prediction. */ 423 int mi_alloc_size; 424 MB_MODE_INFO *mip; /* Base of allocated array */ 425 MB_MODE_INFO *mi; /* Corresponds to upper left visible macroblock */ 426 427 // TODO(agrange): Move prev_mi into encoder structure. 428 // prev_mip and prev_mi will only be allocated in encoder. 429 MB_MODE_INFO *prev_mip; /* MB_MODE_INFO array 'mip' from last decoded frame */ 430 MB_MODE_INFO *prev_mi; /* 'mi' from last frame (points into prev_mip) */ 431 432 // Separate mi functions between encoder and decoder. 433 int (*alloc_mi)(struct AV1Common *cm, int mi_size); 434 void (*free_mi)(struct AV1Common *cm); 435 void (*setup_mi)(struct AV1Common *cm); 436 437 // Grid of pointers to 8x8 MB_MODE_INFO structs. Any 8x8 not in the visible 438 // area will be NULL. 439 MB_MODE_INFO **mi_grid_base; 440 MB_MODE_INFO **mi_grid_visible; 441 MB_MODE_INFO **prev_mi_grid_base; 442 MB_MODE_INFO **prev_mi_grid_visible; 443 444 // Whether to use previous frames' motion vectors for prediction. 445 int allow_ref_frame_mvs; 446 447 uint8_t *last_frame_seg_map; 448 449 InterpFilter interp_filter; 450 451 int switchable_motion_mode; 452 453 loop_filter_info_n lf_info; 454 // The denominator of the superres scale; the numerator is fixed. 455 uint8_t superres_scale_denominator; 456 int superres_upscaled_width; 457 int superres_upscaled_height; 458 RestorationInfo rst_info[MAX_MB_PLANE]; 459 460 // Pointer to a scratch buffer used by self-guided restoration 461 int32_t *rst_tmpbuf; 462 RestorationLineBuffers *rlbs; 463 464 // Output of loop restoration 465 YV12_BUFFER_CONFIG rst_frame; 466 467 // Flag signaling how frame contexts should be updated at the end of 468 // a frame decode 469 REFRESH_FRAME_CONTEXT_MODE refresh_frame_context; 470 471 int ref_frame_sign_bias[REF_FRAMES]; /* Two state 0, 1 */ 472 473 struct loopfilter lf; 474 struct segmentation seg; 475 int coded_lossless; // frame is fully lossless at the coded resolution. 476 int all_lossless; // frame is fully lossless at the upscaled resolution. 477 478 int reduced_tx_set_used; 479 480 // Context probabilities for reference frame prediction 481 MV_REFERENCE_FRAME comp_fwd_ref[FWD_REFS]; 482 MV_REFERENCE_FRAME comp_bwd_ref[BWD_REFS]; 483 484 FRAME_CONTEXT *fc; /* this frame entropy */ 485 FRAME_CONTEXT *default_frame_context; 486 int primary_ref_frame; 487 488 int error_resilient_mode; 489 490 int tile_cols, tile_rows; 491 492 int max_tile_width_sb; 493 int min_log2_tile_cols; 494 int max_log2_tile_cols; 495 int max_log2_tile_rows; 496 int min_log2_tile_rows; 497 int min_log2_tiles; 498 int max_tile_height_sb; 499 int uniform_tile_spacing_flag; 500 int log2_tile_cols; // only valid for uniform tiles 501 int log2_tile_rows; // only valid for uniform tiles 502 int tile_col_start_sb[MAX_TILE_COLS + 1]; // valid for 0 <= i <= tile_cols 503 int tile_row_start_sb[MAX_TILE_ROWS + 1]; // valid for 0 <= i <= tile_rows 504 int tile_width, tile_height; // In MI units 505 int min_inner_tile_width; // min width of non-rightmost tile 506 507 unsigned int large_scale_tile; 508 unsigned int single_tile_decoding; 509 510 int byte_alignment; 511 int skip_loop_filter; 512 int skip_film_grain; 513 514 // External BufferPool passed from outside. 515 BufferPool *buffer_pool; 516 517 PARTITION_CONTEXT **above_seg_context; 518 ENTROPY_CONTEXT **above_context[MAX_MB_PLANE]; 519 TXFM_CONTEXT **above_txfm_context; 520 WarpedMotionParams global_motion[REF_FRAMES]; 521 aom_film_grain_t film_grain_params; 522 523 CdefInfo cdef_info; 524 DeltaQInfo delta_q_info; // Delta Q and Delta LF parameters 525 526 int num_tg; 527 SequenceHeader seq_params; 528 int current_frame_id; 529 int ref_frame_id[REF_FRAMES]; 530 int valid_for_referencing[REF_FRAMES]; 531 TPL_MV_REF *tpl_mvs; 532 int tpl_mvs_mem_size; 533 // TODO(jingning): This can be combined with sign_bias later. 534 int8_t ref_frame_side[REF_FRAMES]; 535 536 int is_annexb; 537 538 int temporal_layer_id; 539 int spatial_layer_id; 540 unsigned int number_temporal_layers; 541 unsigned int number_spatial_layers; 542 int num_allocated_above_context_mi_col; 543 int num_allocated_above_contexts; 544 int num_allocated_above_context_planes; 545 546 #if TXCOEFF_TIMER 547 int64_t cum_txcoeff_timer; 548 int64_t txcoeff_timer; 549 int txb_count; 550 #endif 551 552 #if TXCOEFF_COST_TIMER 553 int64_t cum_txcoeff_cost_timer; 554 int64_t txcoeff_cost_timer; 555 int64_t txcoeff_cost_count; 556 #endif 557 const cfg_options_t *options; 558 int is_decoding; 559 } AV1_COMMON; 560 561 // TODO(hkuang): Don't need to lock the whole pool after implementing atomic 562 // frame reference count. 563 static void lock_buffer_pool(BufferPool *const pool) { 564 #if CONFIG_MULTITHREAD 565 pthread_mutex_lock(&pool->pool_mutex); 566 #else 567 (void)pool; 568 #endif 569 } 570 571 static void unlock_buffer_pool(BufferPool *const pool) { 572 #if CONFIG_MULTITHREAD 573 pthread_mutex_unlock(&pool->pool_mutex); 574 #else 575 (void)pool; 576 #endif 577 } 578 579 static INLINE YV12_BUFFER_CONFIG *get_ref_frame(AV1_COMMON *cm, int index) { 580 if (index < 0 || index >= REF_FRAMES) return NULL; 581 if (cm->ref_frame_map[index] == NULL) return NULL; 582 return &cm->ref_frame_map[index]->buf; 583 } 584 585 static INLINE int get_free_fb(AV1_COMMON *cm) { 586 RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs; 587 int i; 588 589 lock_buffer_pool(cm->buffer_pool); 590 for (i = 0; i < FRAME_BUFFERS; ++i) 591 if (frame_bufs[i].ref_count == 0) break; 592 593 if (i != FRAME_BUFFERS) { 594 if (frame_bufs[i].buf.use_external_reference_buffers) { 595 // If this frame buffer's y_buffer, u_buffer, and v_buffer point to the 596 // external reference buffers. Restore the buffer pointers to point to the 597 // internally allocated memory. 598 YV12_BUFFER_CONFIG *ybf = &frame_bufs[i].buf; 599 ybf->y_buffer = ybf->store_buf_adr[0]; 600 ybf->u_buffer = ybf->store_buf_adr[1]; 601 ybf->v_buffer = ybf->store_buf_adr[2]; 602 ybf->use_external_reference_buffers = 0; 603 } 604 605 frame_bufs[i].ref_count = 1; 606 } else { 607 // We should never run out of free buffers. If this assertion fails, there 608 // is a reference leak. 609 assert(0 && "Ran out of free frame buffers. Likely a reference leak."); 610 // Reset i to be INVALID_IDX to indicate no free buffer found. 611 i = INVALID_IDX; 612 } 613 614 unlock_buffer_pool(cm->buffer_pool); 615 return i; 616 } 617 618 static INLINE RefCntBuffer *assign_cur_frame_new_fb(AV1_COMMON *const cm) { 619 // Release the previously-used frame-buffer 620 if (cm->cur_frame != NULL) { 621 --cm->cur_frame->ref_count; 622 cm->cur_frame = NULL; 623 } 624 625 // Assign a new framebuffer 626 const int new_fb_idx = get_free_fb(cm); 627 if (new_fb_idx == INVALID_IDX) return NULL; 628 629 cm->cur_frame = &cm->buffer_pool->frame_bufs[new_fb_idx]; 630 cm->cur_frame->buf.buf_8bit_valid = 0; 631 av1_zero(cm->cur_frame->interp_filter_selected); 632 return cm->cur_frame; 633 } 634 635 // Modify 'lhs_ptr' to reference the buffer at 'rhs_ptr', and update the ref 636 // counts accordingly. 637 static INLINE void assign_frame_buffer_p(RefCntBuffer **lhs_ptr, 638 RefCntBuffer *rhs_ptr) { 639 RefCntBuffer *const old_ptr = *lhs_ptr; 640 if (old_ptr != NULL) { 641 assert(old_ptr->ref_count > 0); 642 // One less reference to the buffer at 'old_ptr', so decrease ref count. 643 --old_ptr->ref_count; 644 } 645 646 *lhs_ptr = rhs_ptr; 647 // One more reference to the buffer at 'rhs_ptr', so increase ref count. 648 ++rhs_ptr->ref_count; 649 } 650 651 static INLINE int frame_is_intra_only(const AV1_COMMON *const cm) { 652 return cm->current_frame.frame_type == KEY_FRAME || 653 cm->current_frame.frame_type == INTRA_ONLY_FRAME; 654 } 655 656 static INLINE int frame_is_sframe(const AV1_COMMON *cm) { 657 return cm->current_frame.frame_type == S_FRAME; 658 } 659 660 // These functions take a reference frame label between LAST_FRAME and 661 // EXTREF_FRAME inclusive. Note that this is different to the indexing 662 // previously used by the frame_refs[] array. 663 static INLINE int get_ref_frame_map_idx(const AV1_COMMON *const cm, 664 const MV_REFERENCE_FRAME ref_frame) { 665 return (ref_frame >= LAST_FRAME && ref_frame <= EXTREF_FRAME) 666 ? cm->remapped_ref_idx[ref_frame - LAST_FRAME] 667 : INVALID_IDX; 668 } 669 670 static INLINE RefCntBuffer *get_ref_frame_buf( 671 const AV1_COMMON *const cm, const MV_REFERENCE_FRAME ref_frame) { 672 const int map_idx = get_ref_frame_map_idx(cm, ref_frame); 673 return (map_idx != INVALID_IDX) ? cm->ref_frame_map[map_idx] : NULL; 674 } 675 676 // Both const and non-const versions of this function are provided so that it 677 // can be used with a const AV1_COMMON if needed. 678 static INLINE const struct scale_factors *get_ref_scale_factors_const( 679 const AV1_COMMON *const cm, const MV_REFERENCE_FRAME ref_frame) { 680 const int map_idx = get_ref_frame_map_idx(cm, ref_frame); 681 return (map_idx != INVALID_IDX) ? &cm->ref_scale_factors[map_idx] : NULL; 682 } 683 684 static INLINE struct scale_factors *get_ref_scale_factors( 685 AV1_COMMON *const cm, const MV_REFERENCE_FRAME ref_frame) { 686 const int map_idx = get_ref_frame_map_idx(cm, ref_frame); 687 return (map_idx != INVALID_IDX) ? &cm->ref_scale_factors[map_idx] : NULL; 688 } 689 690 static INLINE RefCntBuffer *get_primary_ref_frame_buf( 691 const AV1_COMMON *const cm) { 692 if (cm->primary_ref_frame == PRIMARY_REF_NONE) return NULL; 693 const int map_idx = get_ref_frame_map_idx(cm, cm->primary_ref_frame + 1); 694 return (map_idx != INVALID_IDX) ? cm->ref_frame_map[map_idx] : NULL; 695 } 696 697 // Returns 1 if this frame might allow mvs from some reference frame. 698 static INLINE int frame_might_allow_ref_frame_mvs(const AV1_COMMON *cm) { 699 return !cm->error_resilient_mode && 700 cm->seq_params.order_hint_info.enable_ref_frame_mvs && 701 cm->seq_params.order_hint_info.enable_order_hint && 702 !frame_is_intra_only(cm); 703 } 704 705 // Returns 1 if this frame might use warped_motion 706 static INLINE int frame_might_allow_warped_motion(const AV1_COMMON *cm) { 707 return !cm->error_resilient_mode && !frame_is_intra_only(cm) && 708 cm->seq_params.enable_warped_motion; 709 } 710 711 static INLINE void ensure_mv_buffer(RefCntBuffer *buf, AV1_COMMON *cm) { 712 const int buf_rows = buf->mi_rows; 713 const int buf_cols = buf->mi_cols; 714 715 if (buf->mvs == NULL || buf_rows != cm->mi_rows || buf_cols != cm->mi_cols) { 716 aom_free(buf->mvs); 717 buf->mi_rows = cm->mi_rows; 718 buf->mi_cols = cm->mi_cols; 719 CHECK_MEM_ERROR(cm, buf->mvs, 720 (MV_REF *)aom_calloc( 721 ((cm->mi_rows + 1) >> 1) * ((cm->mi_cols + 1) >> 1), 722 sizeof(*buf->mvs))); 723 aom_free(buf->seg_map); 724 CHECK_MEM_ERROR(cm, buf->seg_map, 725 (uint8_t *)aom_calloc(cm->mi_rows * cm->mi_cols, 726 sizeof(*buf->seg_map))); 727 } 728 729 const int mem_size = 730 ((cm->mi_rows + MAX_MIB_SIZE) >> 1) * (cm->mi_stride >> 1); 731 int realloc = cm->tpl_mvs == NULL; 732 if (cm->tpl_mvs) realloc |= cm->tpl_mvs_mem_size < mem_size; 733 734 if (realloc) { 735 aom_free(cm->tpl_mvs); 736 CHECK_MEM_ERROR(cm, cm->tpl_mvs, 737 (TPL_MV_REF *)aom_calloc(mem_size, sizeof(*cm->tpl_mvs))); 738 cm->tpl_mvs_mem_size = mem_size; 739 } 740 } 741 742 void cfl_init(CFL_CTX *cfl, const SequenceHeader *seq_params); 743 744 static INLINE int av1_num_planes(const AV1_COMMON *cm) { 745 return cm->seq_params.monochrome ? 1 : MAX_MB_PLANE; 746 } 747 748 static INLINE void av1_init_above_context(AV1_COMMON *cm, MACROBLOCKD *xd, 749 const int tile_row) { 750 const int num_planes = av1_num_planes(cm); 751 for (int i = 0; i < num_planes; ++i) { 752 xd->above_context[i] = cm->above_context[i][tile_row]; 753 } 754 xd->above_seg_context = cm->above_seg_context[tile_row]; 755 xd->above_txfm_context = cm->above_txfm_context[tile_row]; 756 } 757 758 static INLINE void av1_init_macroblockd(AV1_COMMON *cm, MACROBLOCKD *xd, 759 tran_low_t *dqcoeff) { 760 const int num_planes = av1_num_planes(cm); 761 for (int i = 0; i < num_planes; ++i) { 762 xd->plane[i].dqcoeff = dqcoeff; 763 764 if (xd->plane[i].plane_type == PLANE_TYPE_Y) { 765 memcpy(xd->plane[i].seg_dequant_QTX, cm->y_dequant_QTX, 766 sizeof(cm->y_dequant_QTX)); 767 memcpy(xd->plane[i].seg_iqmatrix, cm->y_iqmatrix, sizeof(cm->y_iqmatrix)); 768 769 } else { 770 if (i == AOM_PLANE_U) { 771 memcpy(xd->plane[i].seg_dequant_QTX, cm->u_dequant_QTX, 772 sizeof(cm->u_dequant_QTX)); 773 memcpy(xd->plane[i].seg_iqmatrix, cm->u_iqmatrix, 774 sizeof(cm->u_iqmatrix)); 775 } else { 776 memcpy(xd->plane[i].seg_dequant_QTX, cm->v_dequant_QTX, 777 sizeof(cm->v_dequant_QTX)); 778 memcpy(xd->plane[i].seg_iqmatrix, cm->v_iqmatrix, 779 sizeof(cm->v_iqmatrix)); 780 } 781 } 782 } 783 xd->mi_stride = cm->mi_stride; 784 xd->error_info = &cm->error; 785 cfl_init(&xd->cfl, &cm->seq_params); 786 } 787 788 static INLINE void set_skip_context(MACROBLOCKD *xd, int mi_row, int mi_col, 789 const int num_planes) { 790 int i; 791 int row_offset = mi_row; 792 int col_offset = mi_col; 793 for (i = 0; i < num_planes; ++i) { 794 struct macroblockd_plane *const pd = &xd->plane[i]; 795 // Offset the buffer pointer 796 const BLOCK_SIZE bsize = xd->mi[0]->sb_type; 797 if (pd->subsampling_y && (mi_row & 0x01) && (mi_size_high[bsize] == 1)) 798 row_offset = mi_row - 1; 799 if (pd->subsampling_x && (mi_col & 0x01) && (mi_size_wide[bsize] == 1)) 800 col_offset = mi_col - 1; 801 int above_idx = col_offset; 802 int left_idx = row_offset & MAX_MIB_MASK; 803 pd->above_context = &xd->above_context[i][above_idx >> pd->subsampling_x]; 804 pd->left_context = &xd->left_context[i][left_idx >> pd->subsampling_y]; 805 } 806 } 807 808 static INLINE int calc_mi_size(int len) { 809 // len is in mi units. Align to a multiple of SBs. 810 return ALIGN_POWER_OF_TWO(len, MAX_MIB_SIZE_LOG2); 811 } 812 813 static INLINE void set_plane_n4(MACROBLOCKD *const xd, int bw, int bh, 814 const int num_planes) { 815 int i; 816 for (i = 0; i < num_planes; i++) { 817 xd->plane[i].width = (bw * MI_SIZE) >> xd->plane[i].subsampling_x; 818 xd->plane[i].height = (bh * MI_SIZE) >> xd->plane[i].subsampling_y; 819 820 xd->plane[i].width = AOMMAX(xd->plane[i].width, 4); 821 xd->plane[i].height = AOMMAX(xd->plane[i].height, 4); 822 } 823 } 824 825 static INLINE void set_mi_row_col(MACROBLOCKD *xd, const TileInfo *const tile, 826 int mi_row, int bh, int mi_col, int bw, 827 int mi_rows, int mi_cols) { 828 xd->mb_to_top_edge = -((mi_row * MI_SIZE) * 8); 829 xd->mb_to_bottom_edge = ((mi_rows - bh - mi_row) * MI_SIZE) * 8; 830 xd->mb_to_left_edge = -((mi_col * MI_SIZE) * 8); 831 xd->mb_to_right_edge = ((mi_cols - bw - mi_col) * MI_SIZE) * 8; 832 833 // Are edges available for intra prediction? 834 xd->up_available = (mi_row > tile->mi_row_start); 835 836 const int ss_x = xd->plane[1].subsampling_x; 837 const int ss_y = xd->plane[1].subsampling_y; 838 839 xd->left_available = (mi_col > tile->mi_col_start); 840 xd->chroma_up_available = xd->up_available; 841 xd->chroma_left_available = xd->left_available; 842 if (ss_x && bw < mi_size_wide[BLOCK_8X8]) 843 xd->chroma_left_available = (mi_col - 1) > tile->mi_col_start; 844 if (ss_y && bh < mi_size_high[BLOCK_8X8]) 845 xd->chroma_up_available = (mi_row - 1) > tile->mi_row_start; 846 if (xd->up_available) { 847 xd->above_mbmi = xd->mi[-xd->mi_stride]; 848 } else { 849 xd->above_mbmi = NULL; 850 } 851 852 if (xd->left_available) { 853 xd->left_mbmi = xd->mi[-1]; 854 } else { 855 xd->left_mbmi = NULL; 856 } 857 858 const int chroma_ref = ((mi_row & 0x01) || !(bh & 0x01) || !ss_y) && 859 ((mi_col & 0x01) || !(bw & 0x01) || !ss_x); 860 if (chroma_ref) { 861 // To help calculate the "above" and "left" chroma blocks, note that the 862 // current block may cover multiple luma blocks (eg, if partitioned into 863 // 4x4 luma blocks). 864 // First, find the top-left-most luma block covered by this chroma block 865 MB_MODE_INFO **base_mi = 866 &xd->mi[-(mi_row & ss_y) * xd->mi_stride - (mi_col & ss_x)]; 867 868 // Then, we consider the luma region covered by the left or above 4x4 chroma 869 // prediction. We want to point to the chroma reference block in that 870 // region, which is the bottom-right-most mi unit. 871 // This leads to the following offsets: 872 MB_MODE_INFO *chroma_above_mi = 873 xd->chroma_up_available ? base_mi[-xd->mi_stride + ss_x] : NULL; 874 xd->chroma_above_mbmi = chroma_above_mi; 875 876 MB_MODE_INFO *chroma_left_mi = 877 xd->chroma_left_available ? base_mi[ss_y * xd->mi_stride - 1] : NULL; 878 xd->chroma_left_mbmi = chroma_left_mi; 879 } 880 881 xd->n4_h = bh; 882 xd->n4_w = bw; 883 xd->is_sec_rect = 0; 884 if (xd->n4_w < xd->n4_h) { 885 // Only mark is_sec_rect as 1 for the last block. 886 // For PARTITION_VERT_4, it would be (0, 0, 0, 1); 887 // For other partitions, it would be (0, 1). 888 if (!((mi_col + xd->n4_w) & (xd->n4_h - 1))) xd->is_sec_rect = 1; 889 } 890 891 if (xd->n4_w > xd->n4_h) 892 if (mi_row & (xd->n4_w - 1)) xd->is_sec_rect = 1; 893 } 894 895 static INLINE aom_cdf_prob *get_y_mode_cdf(FRAME_CONTEXT *tile_ctx, 896 const MB_MODE_INFO *above_mi, 897 const MB_MODE_INFO *left_mi) { 898 const PREDICTION_MODE above = av1_above_block_mode(above_mi); 899 const PREDICTION_MODE left = av1_left_block_mode(left_mi); 900 const int above_ctx = intra_mode_context[above]; 901 const int left_ctx = intra_mode_context[left]; 902 return tile_ctx->kf_y_cdf[above_ctx][left_ctx]; 903 } 904 905 static INLINE void update_partition_context(MACROBLOCKD *xd, int mi_row, 906 int mi_col, BLOCK_SIZE subsize, 907 BLOCK_SIZE bsize) { 908 PARTITION_CONTEXT *const above_ctx = xd->above_seg_context + mi_col; 909 PARTITION_CONTEXT *const left_ctx = 910 xd->left_seg_context + (mi_row & MAX_MIB_MASK); 911 912 const int bw = mi_size_wide[bsize]; 913 const int bh = mi_size_high[bsize]; 914 memset(above_ctx, partition_context_lookup[subsize].above, bw); 915 memset(left_ctx, partition_context_lookup[subsize].left, bh); 916 } 917 918 static INLINE int is_chroma_reference(int mi_row, int mi_col, BLOCK_SIZE bsize, 919 int subsampling_x, int subsampling_y) { 920 const int bw = mi_size_wide[bsize]; 921 const int bh = mi_size_high[bsize]; 922 int ref_pos = ((mi_row & 0x01) || !(bh & 0x01) || !subsampling_y) && 923 ((mi_col & 0x01) || !(bw & 0x01) || !subsampling_x); 924 return ref_pos; 925 } 926 927 static INLINE BLOCK_SIZE scale_chroma_bsize(BLOCK_SIZE bsize, int subsampling_x, 928 int subsampling_y) { 929 BLOCK_SIZE bs = bsize; 930 switch (bsize) { 931 case BLOCK_4X4: 932 if (subsampling_x == 1 && subsampling_y == 1) 933 bs = BLOCK_8X8; 934 else if (subsampling_x == 1) 935 bs = BLOCK_8X4; 936 else if (subsampling_y == 1) 937 bs = BLOCK_4X8; 938 break; 939 case BLOCK_4X8: 940 if (subsampling_x == 1 && subsampling_y == 1) 941 bs = BLOCK_8X8; 942 else if (subsampling_x == 1) 943 bs = BLOCK_8X8; 944 else if (subsampling_y == 1) 945 bs = BLOCK_4X8; 946 break; 947 case BLOCK_8X4: 948 if (subsampling_x == 1 && subsampling_y == 1) 949 bs = BLOCK_8X8; 950 else if (subsampling_x == 1) 951 bs = BLOCK_8X4; 952 else if (subsampling_y == 1) 953 bs = BLOCK_8X8; 954 break; 955 case BLOCK_4X16: 956 if (subsampling_x == 1 && subsampling_y == 1) 957 bs = BLOCK_8X16; 958 else if (subsampling_x == 1) 959 bs = BLOCK_8X16; 960 else if (subsampling_y == 1) 961 bs = BLOCK_4X16; 962 break; 963 case BLOCK_16X4: 964 if (subsampling_x == 1 && subsampling_y == 1) 965 bs = BLOCK_16X8; 966 else if (subsampling_x == 1) 967 bs = BLOCK_16X4; 968 else if (subsampling_y == 1) 969 bs = BLOCK_16X8; 970 break; 971 default: break; 972 } 973 return bs; 974 } 975 976 static INLINE aom_cdf_prob cdf_element_prob(const aom_cdf_prob *cdf, 977 size_t element) { 978 assert(cdf != NULL); 979 return (element > 0 ? cdf[element - 1] : CDF_PROB_TOP) - cdf[element]; 980 } 981 982 static INLINE void partition_gather_horz_alike(aom_cdf_prob *out, 983 const aom_cdf_prob *const in, 984 BLOCK_SIZE bsize) { 985 (void)bsize; 986 out[0] = CDF_PROB_TOP; 987 out[0] -= cdf_element_prob(in, PARTITION_HORZ); 988 out[0] -= cdf_element_prob(in, PARTITION_SPLIT); 989 out[0] -= cdf_element_prob(in, PARTITION_HORZ_A); 990 out[0] -= cdf_element_prob(in, PARTITION_HORZ_B); 991 out[0] -= cdf_element_prob(in, PARTITION_VERT_A); 992 if (bsize != BLOCK_128X128) out[0] -= cdf_element_prob(in, PARTITION_HORZ_4); 993 out[0] = AOM_ICDF(out[0]); 994 out[1] = AOM_ICDF(CDF_PROB_TOP); 995 } 996 997 static INLINE void partition_gather_vert_alike(aom_cdf_prob *out, 998 const aom_cdf_prob *const in, 999 BLOCK_SIZE bsize) { 1000 (void)bsize; 1001 out[0] = CDF_PROB_TOP; 1002 out[0] -= cdf_element_prob(in, PARTITION_VERT); 1003 out[0] -= cdf_element_prob(in, PARTITION_SPLIT); 1004 out[0] -= cdf_element_prob(in, PARTITION_HORZ_A); 1005 out[0] -= cdf_element_prob(in, PARTITION_VERT_A); 1006 out[0] -= cdf_element_prob(in, PARTITION_VERT_B); 1007 if (bsize != BLOCK_128X128) out[0] -= cdf_element_prob(in, PARTITION_VERT_4); 1008 out[0] = AOM_ICDF(out[0]); 1009 out[1] = AOM_ICDF(CDF_PROB_TOP); 1010 } 1011 1012 static INLINE void update_ext_partition_context(MACROBLOCKD *xd, int mi_row, 1013 int mi_col, BLOCK_SIZE subsize, 1014 BLOCK_SIZE bsize, 1015 PARTITION_TYPE partition) { 1016 if (bsize >= BLOCK_8X8) { 1017 const int hbs = mi_size_wide[bsize] / 2; 1018 BLOCK_SIZE bsize2 = get_partition_subsize(bsize, PARTITION_SPLIT); 1019 switch (partition) { 1020 case PARTITION_SPLIT: 1021 if (bsize != BLOCK_8X8) break; 1022 AOM_FALLTHROUGH_INTENDED; 1023 case PARTITION_NONE: 1024 case PARTITION_HORZ: 1025 case PARTITION_VERT: 1026 case PARTITION_HORZ_4: 1027 case PARTITION_VERT_4: 1028 update_partition_context(xd, mi_row, mi_col, subsize, bsize); 1029 break; 1030 case PARTITION_HORZ_A: 1031 update_partition_context(xd, mi_row, mi_col, bsize2, subsize); 1032 update_partition_context(xd, mi_row + hbs, mi_col, subsize, subsize); 1033 break; 1034 case PARTITION_HORZ_B: 1035 update_partition_context(xd, mi_row, mi_col, subsize, subsize); 1036 update_partition_context(xd, mi_row + hbs, mi_col, bsize2, subsize); 1037 break; 1038 case PARTITION_VERT_A: 1039 update_partition_context(xd, mi_row, mi_col, bsize2, subsize); 1040 update_partition_context(xd, mi_row, mi_col + hbs, subsize, subsize); 1041 break; 1042 case PARTITION_VERT_B: 1043 update_partition_context(xd, mi_row, mi_col, subsize, subsize); 1044 update_partition_context(xd, mi_row, mi_col + hbs, bsize2, subsize); 1045 break; 1046 default: assert(0 && "Invalid partition type"); 1047 } 1048 } 1049 } 1050 1051 static INLINE int partition_plane_context(const MACROBLOCKD *xd, int mi_row, 1052 int mi_col, BLOCK_SIZE bsize) { 1053 const PARTITION_CONTEXT *above_ctx = xd->above_seg_context + mi_col; 1054 const PARTITION_CONTEXT *left_ctx = 1055 xd->left_seg_context + (mi_row & MAX_MIB_MASK); 1056 // Minimum partition point is 8x8. Offset the bsl accordingly. 1057 const int bsl = mi_size_wide_log2[bsize] - mi_size_wide_log2[BLOCK_8X8]; 1058 int above = (*above_ctx >> bsl) & 1, left = (*left_ctx >> bsl) & 1; 1059 1060 assert(mi_size_wide_log2[bsize] == mi_size_high_log2[bsize]); 1061 assert(bsl >= 0); 1062 1063 return (left * 2 + above) + bsl * PARTITION_PLOFFSET; 1064 } 1065 1066 // Return the number of elements in the partition CDF when 1067 // partitioning the (square) block with luma block size of bsize. 1068 static INLINE int partition_cdf_length(BLOCK_SIZE bsize) { 1069 if (bsize <= BLOCK_8X8) 1070 return PARTITION_TYPES; 1071 else if (bsize == BLOCK_128X128) 1072 return EXT_PARTITION_TYPES - 2; 1073 else 1074 return EXT_PARTITION_TYPES; 1075 } 1076 1077 static INLINE int max_block_wide(const MACROBLOCKD *xd, BLOCK_SIZE bsize, 1078 int plane) { 1079 int max_blocks_wide = block_size_wide[bsize]; 1080 const struct macroblockd_plane *const pd = &xd->plane[plane]; 1081 1082 if (xd->mb_to_right_edge < 0) 1083 max_blocks_wide += xd->mb_to_right_edge >> (3 + pd->subsampling_x); 1084 1085 // Scale the width in the transform block unit. 1086 return max_blocks_wide >> tx_size_wide_log2[0]; 1087 } 1088 1089 static INLINE int max_block_high(const MACROBLOCKD *xd, BLOCK_SIZE bsize, 1090 int plane) { 1091 int max_blocks_high = block_size_high[bsize]; 1092 const struct macroblockd_plane *const pd = &xd->plane[plane]; 1093 1094 if (xd->mb_to_bottom_edge < 0) 1095 max_blocks_high += xd->mb_to_bottom_edge >> (3 + pd->subsampling_y); 1096 1097 // Scale the height in the transform block unit. 1098 return max_blocks_high >> tx_size_high_log2[0]; 1099 } 1100 1101 static INLINE int max_intra_block_width(const MACROBLOCKD *xd, 1102 BLOCK_SIZE plane_bsize, int plane, 1103 TX_SIZE tx_size) { 1104 const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane) 1105 << tx_size_wide_log2[0]; 1106 return ALIGN_POWER_OF_TWO(max_blocks_wide, tx_size_wide_log2[tx_size]); 1107 } 1108 1109 static INLINE int max_intra_block_height(const MACROBLOCKD *xd, 1110 BLOCK_SIZE plane_bsize, int plane, 1111 TX_SIZE tx_size) { 1112 const int max_blocks_high = max_block_high(xd, plane_bsize, plane) 1113 << tx_size_high_log2[0]; 1114 return ALIGN_POWER_OF_TWO(max_blocks_high, tx_size_high_log2[tx_size]); 1115 } 1116 1117 static INLINE void av1_zero_above_context(AV1_COMMON *const cm, const MACROBLOCKD *xd, 1118 int mi_col_start, int mi_col_end, const int tile_row) { 1119 const SequenceHeader *const seq_params = &cm->seq_params; 1120 const int num_planes = av1_num_planes(cm); 1121 const int width = mi_col_end - mi_col_start; 1122 const int aligned_width = 1123 ALIGN_POWER_OF_TWO(width, seq_params->mib_size_log2); 1124 1125 const int offset_y = mi_col_start; 1126 const int width_y = aligned_width; 1127 const int offset_uv = offset_y >> seq_params->subsampling_x; 1128 const int width_uv = width_y >> seq_params->subsampling_x; 1129 1130 av1_zero_array(cm->above_context[0][tile_row] + offset_y, width_y); 1131 if (num_planes > 1) { 1132 if (cm->above_context[1][tile_row] && cm->above_context[2][tile_row]) { 1133 av1_zero_array(cm->above_context[1][tile_row] + offset_uv, width_uv); 1134 av1_zero_array(cm->above_context[2][tile_row] + offset_uv, width_uv); 1135 } else { 1136 aom_internal_error(xd->error_info, AOM_CODEC_CORRUPT_FRAME, 1137 "Invalid value of planes"); 1138 } 1139 } 1140 1141 av1_zero_array(cm->above_seg_context[tile_row] + mi_col_start, aligned_width); 1142 1143 memset(cm->above_txfm_context[tile_row] + mi_col_start, 1144 tx_size_wide[TX_SIZES_LARGEST], 1145 aligned_width * sizeof(TXFM_CONTEXT)); 1146 } 1147 1148 static INLINE void av1_zero_left_context(MACROBLOCKD *const xd) { 1149 av1_zero(xd->left_context); 1150 av1_zero(xd->left_seg_context); 1151 1152 memset(xd->left_txfm_context_buffer, tx_size_high[TX_SIZES_LARGEST], 1153 sizeof(xd->left_txfm_context_buffer)); 1154 } 1155 1156 // Disable array-bounds checks as the TX_SIZE enum contains values larger than 1157 // TX_SIZES_ALL (TX_INVALID) which make extending the array as a workaround 1158 // infeasible. The assert is enough for static analysis and this or other tools 1159 // asan, valgrind would catch oob access at runtime. 1160 #if defined(__GNUC__) && __GNUC__ >= 4 1161 #pragma GCC diagnostic ignored "-Warray-bounds" 1162 #endif 1163 1164 #if defined(__GNUC__) && __GNUC__ >= 4 1165 #pragma GCC diagnostic warning "-Warray-bounds" 1166 #endif 1167 1168 static INLINE void set_txfm_ctx(TXFM_CONTEXT *txfm_ctx, uint8_t txs, int len) { 1169 int i; 1170 for (i = 0; i < len; ++i) txfm_ctx[i] = txs; 1171 } 1172 1173 static INLINE void set_txfm_ctxs(TX_SIZE tx_size, int n4_w, int n4_h, int skip, 1174 const MACROBLOCKD *xd) { 1175 uint8_t bw = tx_size_wide[tx_size]; 1176 uint8_t bh = tx_size_high[tx_size]; 1177 1178 if (skip) { 1179 bw = n4_w * MI_SIZE; 1180 bh = n4_h * MI_SIZE; 1181 } 1182 1183 set_txfm_ctx(xd->above_txfm_context, bw, n4_w); 1184 set_txfm_ctx(xd->left_txfm_context, bh, n4_h); 1185 } 1186 1187 static INLINE void txfm_partition_update(TXFM_CONTEXT *above_ctx, 1188 TXFM_CONTEXT *left_ctx, 1189 TX_SIZE tx_size, TX_SIZE txb_size) { 1190 BLOCK_SIZE bsize = txsize_to_bsize[txb_size]; 1191 int bh = mi_size_high[bsize]; 1192 int bw = mi_size_wide[bsize]; 1193 uint8_t txw = tx_size_wide[tx_size]; 1194 uint8_t txh = tx_size_high[tx_size]; 1195 int i; 1196 for (i = 0; i < bh; ++i) left_ctx[i] = txh; 1197 for (i = 0; i < bw; ++i) above_ctx[i] = txw; 1198 } 1199 1200 static INLINE TX_SIZE get_sqr_tx_size(int tx_dim) { 1201 switch (tx_dim) { 1202 case 128: 1203 case 64: return TX_64X64; break; 1204 case 32: return TX_32X32; break; 1205 case 16: return TX_16X16; break; 1206 case 8: return TX_8X8; break; 1207 default: return TX_4X4; 1208 } 1209 } 1210 1211 static INLINE TX_SIZE get_tx_size(int width, int height) { 1212 if (width == height) { 1213 return get_sqr_tx_size(width); 1214 } 1215 if (width < height) { 1216 if (width + width == height) { 1217 switch (width) { 1218 case 4: return TX_4X8; break; 1219 case 8: return TX_8X16; break; 1220 case 16: return TX_16X32; break; 1221 case 32: return TX_32X64; break; 1222 } 1223 } else { 1224 switch (width) { 1225 case 4: return TX_4X16; break; 1226 case 8: return TX_8X32; break; 1227 case 16: return TX_16X64; break; 1228 } 1229 } 1230 } else { 1231 if (height + height == width) { 1232 switch (height) { 1233 case 4: return TX_8X4; break; 1234 case 8: return TX_16X8; break; 1235 case 16: return TX_32X16; break; 1236 case 32: return TX_64X32; break; 1237 } 1238 } else { 1239 switch (height) { 1240 case 4: return TX_16X4; break; 1241 case 8: return TX_32X8; break; 1242 case 16: return TX_64X16; break; 1243 } 1244 } 1245 } 1246 assert(0); 1247 return TX_4X4; 1248 } 1249 1250 static INLINE int txfm_partition_context(const TXFM_CONTEXT *const above_ctx, 1251 const TXFM_CONTEXT *const left_ctx, 1252 BLOCK_SIZE bsize, TX_SIZE tx_size) { 1253 const uint8_t txw = tx_size_wide[tx_size]; 1254 const uint8_t txh = tx_size_high[tx_size]; 1255 const int above = *above_ctx < txw; 1256 const int left = *left_ctx < txh; 1257 int category = TXFM_PARTITION_CONTEXTS; 1258 1259 // dummy return, not used by others. 1260 if (tx_size <= TX_4X4) return 0; 1261 1262 TX_SIZE max_tx_size = 1263 get_sqr_tx_size(AOMMAX(block_size_wide[bsize], block_size_high[bsize])); 1264 1265 if (max_tx_size >= TX_8X8) { 1266 category = 1267 (txsize_sqr_up_map[tx_size] != max_tx_size && max_tx_size > TX_8X8) + 1268 (TX_SIZES - 1 - max_tx_size) * 2; 1269 } 1270 assert(category != TXFM_PARTITION_CONTEXTS); 1271 return category * 3 + above + left; 1272 } 1273 1274 // Compute the next partition in the direction of the sb_type stored in the mi 1275 // array, starting with bsize. 1276 static INLINE PARTITION_TYPE get_partition(const AV1_COMMON *const cm, 1277 int mi_row, int mi_col, 1278 BLOCK_SIZE bsize) { 1279 if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return PARTITION_INVALID; 1280 1281 const int offset = mi_row * cm->mi_stride + mi_col; 1282 MB_MODE_INFO **mi = cm->mi_grid_visible + offset; 1283 const BLOCK_SIZE subsize = mi[0]->sb_type; 1284 1285 if (subsize == bsize) return PARTITION_NONE; 1286 1287 const int bhigh = mi_size_high[bsize]; 1288 const int bwide = mi_size_wide[bsize]; 1289 const int sshigh = mi_size_high[subsize]; 1290 const int sswide = mi_size_wide[subsize]; 1291 1292 if (bsize > BLOCK_8X8 && mi_row + bwide / 2 < cm->mi_rows && 1293 mi_col + bhigh / 2 < cm->mi_cols) { 1294 // In this case, the block might be using an extended partition 1295 // type. 1296 const MB_MODE_INFO *const mbmi_right = mi[bwide / 2]; 1297 const MB_MODE_INFO *const mbmi_below = mi[bhigh / 2 * cm->mi_stride]; 1298 1299 if (sswide == bwide) { 1300 // Smaller height but same width. Is PARTITION_HORZ_4, PARTITION_HORZ or 1301 // PARTITION_HORZ_B. To distinguish the latter two, check if the lower 1302 // half was split. 1303 if (sshigh * 4 == bhigh) return PARTITION_HORZ_4; 1304 assert(sshigh * 2 == bhigh); 1305 1306 if (mbmi_below->sb_type == subsize) 1307 return PARTITION_HORZ; 1308 else 1309 return PARTITION_HORZ_B; 1310 } else if (sshigh == bhigh) { 1311 // Smaller width but same height. Is PARTITION_VERT_4, PARTITION_VERT or 1312 // PARTITION_VERT_B. To distinguish the latter two, check if the right 1313 // half was split. 1314 if (sswide * 4 == bwide) return PARTITION_VERT_4; 1315 assert(sswide * 2 == bhigh); 1316 1317 if (mbmi_right->sb_type == subsize) 1318 return PARTITION_VERT; 1319 else 1320 return PARTITION_VERT_B; 1321 } else { 1322 // Smaller width and smaller height. Might be PARTITION_SPLIT or could be 1323 // PARTITION_HORZ_A or PARTITION_VERT_A. If subsize isn't halved in both 1324 // dimensions, we immediately know this is a split (which will recurse to 1325 // get to subsize). Otherwise look down and to the right. With 1326 // PARTITION_VERT_A, the right block will have height bhigh; with 1327 // PARTITION_HORZ_A, the lower block with have width bwide. Otherwise 1328 // it's PARTITION_SPLIT. 1329 if (sswide * 2 != bwide || sshigh * 2 != bhigh) return PARTITION_SPLIT; 1330 1331 if (mi_size_wide[mbmi_below->sb_type] == bwide) return PARTITION_HORZ_A; 1332 if (mi_size_high[mbmi_right->sb_type] == bhigh) return PARTITION_VERT_A; 1333 1334 return PARTITION_SPLIT; 1335 } 1336 } 1337 const int vert_split = sswide < bwide; 1338 const int horz_split = sshigh < bhigh; 1339 const int split_idx = (vert_split << 1) | horz_split; 1340 assert(split_idx != 0); 1341 1342 static const PARTITION_TYPE base_partitions[4] = { 1343 PARTITION_INVALID, PARTITION_HORZ, PARTITION_VERT, PARTITION_SPLIT 1344 }; 1345 1346 return base_partitions[split_idx]; 1347 } 1348 1349 static INLINE void set_sb_size(SequenceHeader *const seq_params, 1350 BLOCK_SIZE sb_size) { 1351 seq_params->sb_size = sb_size; 1352 seq_params->mib_size = mi_size_wide[seq_params->sb_size]; 1353 seq_params->mib_size_log2 = mi_size_wide_log2[seq_params->sb_size]; 1354 } 1355 1356 // Returns true if the frame is fully lossless at the coded resolution. 1357 // Note: If super-resolution is used, such a frame will still NOT be lossless at 1358 // the upscaled resolution. 1359 static INLINE int is_coded_lossless(const AV1_COMMON *cm, 1360 const MACROBLOCKD *xd) { 1361 int coded_lossless = 1; 1362 if (cm->seg.enabled) { 1363 for (int i = 0; i < MAX_SEGMENTS; ++i) { 1364 if (!xd->lossless[i]) { 1365 coded_lossless = 0; 1366 break; 1367 } 1368 } 1369 } else { 1370 coded_lossless = xd->lossless[0]; 1371 } 1372 return coded_lossless; 1373 } 1374 1375 static INLINE int is_valid_seq_level_idx(AV1_LEVEL seq_level_idx) { 1376 return seq_level_idx < SEQ_LEVELS || seq_level_idx == SEQ_LEVEL_MAX; 1377 } 1378 1379 #ifdef __cplusplus 1380 } // extern "C" 1381 #endif 1382 1383 #endif // AOM_AV1_COMMON_ONYXC_INT_H_ 1384
