ref_mvs.c 79.3 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
/*
 * Copyright (c) 2001-2016, Alliance for Open Media. All rights reserved
 *
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
 */

/*
 * Changes made compared to libaom version:
 * - we disable TMV and enable MV_COMPRESS so that the
 *   input array for prev_frames can be at 4x4 instead of
 *   8x8 resolution, and therefore shared between cur_frame
 *   and prev_frame. To make enc/dec behave consistent, we
 *   also make this change around line 2580:
#if 0
                AOMMIN(((mi_row >> 1) << 1) + 1 + (((xd->n8_h - 1) >> 1) << 1),
                       mi_row_end - 1) *
                    prev_frame_mvs_stride +
                AOMMIN(((mi_col >> 1) << 1) + 1 + (((xd->n8_w - 1) >> 1) << 1),
                       mi_col_end - 1)
#else
                (((mi_row >> 1) << 1) + 1) * prev_frame_mvs_stride +
                (((mi_col >> 1) << 1) + 1)
#endif
 *   and the same change (swap mi_cols from prev_frame.mv_stride) on line 2407
 * - we disable rect-block overhanging edge inclusion (see
 *   line 2642):
  if (num_8x8_blocks_wide == num_8x8_blocks_high || 1) {
    mv_ref_search[5].row = -1;
    mv_ref_search[5].col = 0;
    mv_ref_search[6].row = 0;
    mv_ref_search[6].col = -1;
  } else {
    mv_ref_search[5].row = -1;
    mv_ref_search[5].col = num_8x8_blocks_wide;
    mv_ref_search[6].row = num_8x8_blocks_high;
    mv_ref_search[6].col = -1;
  }
 *   Note that this is a bitstream change and needs the same
 *   change on the decoder side also.
 * - we change xd->mi to be a pointer instead of a double ptr.
 */

#include "config.h"

#include <assert.h>
51
#include <errno.h>
52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768
#include <limits.h>
#include <stddef.h>
#include <stdint.h>
#include <stdlib.h>
#include <string.h>

#include "common/intops.h"

#define av1_zero(a) memset(a, 0, sizeof(a))

#define ATTRIBUTE_PACKED
#define INLINE inline
#define IMPLIES(a, b) (!(a) || (b))  //  Logical 'a implies b' (or 'a -> b')

#define ROUND_POWER_OF_TWO(value, n) (((value) + (((1 << (n)) >> 1))) >> (n))
#define ROUND_POWER_OF_TWO_SIGNED(value, n)           \
  (((value) < 0) ? -ROUND_POWER_OF_TWO(-(value), (n)) \
                 : ROUND_POWER_OF_TWO((value), (n)))
#define NELEMENTS(x) (int)(sizeof(x) / sizeof(x[0]))

#define MAX_MV_REF_CANDIDATES 2

#define MAX_REF_MV_STACK_SIZE 8
#define REF_CAT_LEVEL 640

#define FRAME_OFFSET_BITS 5
#define MAX_FRAME_DISTANCE ((1 << FRAME_OFFSET_BITS) - 1)
#define INVALID_MV 0x80008000

#define COMP_NEWMV_CTXS 5
#define REFMV_OFFSET 4
#define REFMV_CTX_MASK ((1 << (8 - REFMV_OFFSET)) - 1)

#define MV_IN_USE_BITS 14
#define MV_UPP (1 << MV_IN_USE_BITS)
#define MV_LOW (-(1 << MV_IN_USE_BITS))

typedef struct MV {
    int16_t row;
    int16_t col;
} MV;
typedef union int_mv {
    uint32_t as_int;
    MV as_mv;
} int_mv;
typedef int8_t MV_REFERENCE_FRAME;
#define MFMV_STACK_SIZE 3
typedef struct {
  int_mv mfmv0;
  uint8_t ref_frame_offset;
} TPL_MV_REF;
typedef struct {
    int_mv mv[2];
    MV_REFERENCE_FRAME ref_frame[2];
    int8_t mode, sb_type;
} MV_REF;
#define MB_MODE_INFO MV_REF

#define AOMMAX(a,b) ((a)>(b)?(a):(b))
#define AOMMIN(a,b) ((a)<(b)?(a):(b))

typedef struct candidate_mv {
    int_mv this_mv;
    int_mv comp_mv;
    int weight;
} CANDIDATE_MV;
#define NONE_FRAME -1
#define INTRA_FRAME 0
#define LAST_FRAME 1

#define LAST2_FRAME 2
#define LAST3_FRAME 3
#define GOLDEN_FRAME 4
#define BWDREF_FRAME 5
#define ALTREF2_FRAME 6
#define ALTREF_FRAME 7
#define LAST_REF_FRAMES (LAST3_FRAME - LAST_FRAME + 1)

#define INTER_REFS_PER_FRAME (ALTREF_FRAME - LAST_FRAME + 1)
#define TOTAL_REFS_PER_FRAME (ALTREF_FRAME - INTRA_FRAME + 1)

#define FWD_REFS (GOLDEN_FRAME - LAST_FRAME + 1)
#define FWD_RF_OFFSET(ref) (ref - LAST_FRAME)
#define BWD_REFS (ALTREF_FRAME - BWDREF_FRAME + 1)
#define BWD_RF_OFFSET(ref) (ref - BWDREF_FRAME)
#define FWD_REFS (GOLDEN_FRAME - LAST_FRAME + 1)
#define SINGLE_REFS (FWD_REFS + BWD_REFS)
typedef enum ATTRIBUTE_PACKED {
  LAST_LAST2_FRAMES,      // { LAST_FRAME, LAST2_FRAME }
  LAST_LAST3_FRAMES,      // { LAST_FRAME, LAST3_FRAME }
  LAST_GOLDEN_FRAMES,     // { LAST_FRAME, GOLDEN_FRAME }
  BWDREF_ALTREF_FRAMES,   // { BWDREF_FRAME, ALTREF_FRAME }
  LAST2_LAST3_FRAMES,     // { LAST2_FRAME, LAST3_FRAME }
  LAST2_GOLDEN_FRAMES,    // { LAST2_FRAME, GOLDEN_FRAME }
  LAST3_GOLDEN_FRAMES,    // { LAST3_FRAME, GOLDEN_FRAME }
  BWDREF_ALTREF2_FRAMES,  // { BWDREF_FRAME, ALTREF2_FRAME }
  ALTREF2_ALTREF_FRAMES,  // { ALTREF2_FRAME, ALTREF_FRAME }
  TOTAL_UNIDIR_COMP_REFS,
  // NOTE: UNIDIR_COMP_REFS is the number of uni-directional reference pairs
  //       that are explicitly signaled.
  UNIDIR_COMP_REFS = BWDREF_ALTREF_FRAMES + 1,
} UNIDIR_COMP_REF;
#define TOTAL_COMP_REFS (FWD_REFS * BWD_REFS + TOTAL_UNIDIR_COMP_REFS)
#define MODE_CTX_REF_FRAMES (TOTAL_REFS_PER_FRAME + TOTAL_COMP_REFS)

#define GLOBALMV_OFFSET 3
#define NEWMV_CTX_MASK ((1 << GLOBALMV_OFFSET) - 1)
#define GLOBALMV_CTX_MASK ((1 << (REFMV_OFFSET - GLOBALMV_OFFSET)) - 1)
#define MI_SIZE_LOG2 2
#define MI_SIZE (1 << MI_SIZE_LOG2)
#define MAX_SB_SIZE_LOG2 7
#define MAX_MIB_SIZE_LOG2 (MAX_SB_SIZE_LOG2 - MI_SIZE_LOG2)
#define MIN_MIB_SIZE_LOG2 (MIN_SB_SIZE_LOG2 - MI_SIZE_LOG2)
#define MAX_MIB_SIZE (1 << MAX_MIB_SIZE_LOG2)
#define MI_SIZE_64X64 (64 >> MI_SIZE_LOG2)
#define MI_SIZE_128X128 (128 >> MI_SIZE_LOG2)
#define REFMV_OFFSET 4

typedef enum ATTRIBUTE_PACKED {
  BLOCK_4X4,
  BLOCK_4X8,
  BLOCK_8X4,
  BLOCK_8X8,
  BLOCK_8X16,
  BLOCK_16X8,
  BLOCK_16X16,
  BLOCK_16X32,
  BLOCK_32X16,
  BLOCK_32X32,
  BLOCK_32X64,
  BLOCK_64X32,
  BLOCK_64X64,
  BLOCK_64X128,
  BLOCK_128X64,
  BLOCK_128X128,
  BLOCK_4X16,
  BLOCK_16X4,
  BLOCK_8X32,
  BLOCK_32X8,
  BLOCK_16X64,
  BLOCK_64X16,
  BLOCK_32X128,
  BLOCK_128X32,
  BLOCK_SIZES_ALL,
  BLOCK_SIZES = BLOCK_4X16,
  BLOCK_INVALID = 255,
  BLOCK_LARGEST = (BLOCK_SIZES - 1)
} BLOCK_SIZE;

typedef enum ATTRIBUTE_PACKED {
  PARTITION_NONE,
  PARTITION_HORZ,
  PARTITION_VERT,
  PARTITION_SPLIT,
  PARTITION_HORZ_A,  // HORZ split and the top partition is split again
  PARTITION_HORZ_B,  // HORZ split and the bottom partition is split again
  PARTITION_VERT_A,  // VERT split and the left partition is split again
  PARTITION_VERT_B,  // VERT split and the right partition is split again
  PARTITION_HORZ_4,  // 4:1 horizontal partition
  PARTITION_VERT_4,  // 4:1 vertical partition
  EXT_PARTITION_TYPES,
  PARTITION_TYPES = PARTITION_SPLIT + 1,
  PARTITION_INVALID = 255
} PARTITION_TYPE;
typedef struct CUR_MODE_INFO {
  PARTITION_TYPE partition;
} CUR_MODE_INFO ;

typedef enum ATTRIBUTE_PACKED {
  DC_PRED,        // Average of above and left pixels
  V_PRED,         // Vertical
  H_PRED,         // Horizontal
  D45_PRED,       // Directional 45  deg = round(arctan(1/1) * 180/pi)
  D135_PRED,      // Directional 135 deg = 180 - 45
  D117_PRED,      // Directional 117 deg = 180 - 63
  D153_PRED,      // Directional 153 deg = 180 - 27
  D207_PRED,      // Directional 207 deg = 180 + 27
  D63_PRED,       // Directional 63  deg = round(arctan(2/1) * 180/pi)
  SMOOTH_PRED,    // Combination of horizontal and vertical interpolation
  SMOOTH_V_PRED,  // Vertical interpolation
  SMOOTH_H_PRED,  // Horizontal interpolation
  PAETH_PRED,     // Predict from the direction of smallest gradient
  NEARESTMV,
  NEARMV,
  GLOBALMV,
  NEWMV,
  // Compound ref compound modes
  NEAREST_NEARESTMV,
  NEAR_NEARMV,
  NEAREST_NEWMV,
  NEW_NEARESTMV,
  NEAR_NEWMV,
  NEW_NEARMV,
  GLOBAL_GLOBALMV,
  NEW_NEWMV,
  MB_MODE_COUNT,
  INTRA_MODES = PAETH_PRED + 1,  // PAETH_PRED has to be the last intra mode.
  INTRA_INVALID = MB_MODE_COUNT  // For uv_mode in inter blocks
} PREDICTION_MODE;
typedef enum {
  IDENTITY = 0,      // identity transformation, 0-parameter
  TRANSLATION = 1,   // translational motion 2-parameter
  ROTZOOM = 2,       // simplified affine with rotation + zoom only, 4-parameter
  AFFINE = 3,        // affine, 6-parameter
  TRANS_TYPES,
} TransformationType;

#define LEAST_SQUARES_SAMPLES_MAX_BITS 3
#define LEAST_SQUARES_SAMPLES_MAX (1 << LEAST_SQUARES_SAMPLES_MAX_BITS)
#define SAMPLES_ARRAY_SIZE (LEAST_SQUARES_SAMPLES_MAX * 2)

static const uint8_t mi_size_wide[BLOCK_SIZES_ALL] = {
  1, 1, 2, 2, 2, 4, 4, 4, 8, 8, 8, 16, 16,
  16, 32, 32,  1, 4, 2, 8, 4, 16, 8, 32
};
static const uint8_t mi_size_high[BLOCK_SIZES_ALL] = {
  1, 2, 1, 2, 4, 2, 4, 8, 4, 8, 16, 8, 16,
  32, 16, 32,  4, 1, 8, 2, 16, 4, 32, 8
};

static const uint8_t block_size_wide[BLOCK_SIZES_ALL] = {
  4,  4,
  8,  8,
  8,  16,
  16, 16,
  32, 32,
  32, 64,
  64, 64, 128, 128, 4,
  16, 8,
  32, 16,
  64, 32, 128
};

static const uint8_t block_size_high[BLOCK_SIZES_ALL] = {
  4,  8,
  4,  8,
  16, 8,
  16, 32,
  16, 32,
  64, 32,
  64, 128, 64, 128, 16,
  4,  32,
  8,  64,
  16, 128, 32
};

static INLINE int is_global_mv_block(const MB_MODE_INFO *const mbmi,
                                     TransformationType type) {
  const PREDICTION_MODE mode = mbmi->mode;
  const BLOCK_SIZE bsize = mbmi->sb_type;
  const int block_size_allowed =
      AOMMIN(block_size_wide[bsize], block_size_high[bsize]) >= 8;
  return (mode == GLOBALMV || mode == GLOBAL_GLOBALMV) && type > TRANSLATION &&
         block_size_allowed;
}

typedef struct {
  TransformationType wmtype;
  int32_t wmmat[6];
  int16_t alpha, beta, gamma, delta;
} WarpedMotionParams;

#define REF_FRAMES_LOG2 3
#define REF_FRAMES (1 << REF_FRAMES_LOG2)
#define FRAME_BUFFERS (REF_FRAMES + 7)
typedef struct {

  unsigned int cur_frame_offset;
  unsigned int ref_frame_offset[INTER_REFS_PER_FRAME];

  MV_REF *mvs;
  ptrdiff_t mv_stride;
  int mi_rows;
  int mi_cols;
  uint8_t intra_only;
} RefCntBuffer;

#define INVALID_IDX -1  // Invalid buffer index.
typedef struct TileInfo {
  int mi_row_start, mi_row_end;
  int mi_col_start, mi_col_end;
  int tg_horz_boundary;
} TileInfo;
typedef struct macroblockd {
  TileInfo tile;
  int mi_stride;

  CUR_MODE_INFO cur_mi;
  MB_MODE_INFO *mi;
  int up_available;
  int left_available;
  /* Distance of MB away from frame edges in subpixels (1/8th pixel)  */
  int mb_to_left_edge;
  int mb_to_right_edge;
  int mb_to_top_edge;
  int mb_to_bottom_edge;
  // block dimension in the unit of mode_info.
  uint8_t n8_w, n8_h;
  uint8_t is_sec_rect;

} MACROBLOCKD;
typedef struct RefBuffer {
  int idx;  // frame buf idx
} RefBuffer;
typedef struct BufferPool {
  RefCntBuffer frame_bufs[FRAME_BUFFERS];
} BufferPool;
typedef struct AV1Common {

  // TODO(hkuang): Combine this with cur_buf in macroblockd.
  RefCntBuffer cur_frame;

  // Each Inter frame can reference INTER_REFS_PER_FRAME buffers
  RefBuffer frame_refs[INTER_REFS_PER_FRAME];

  int allow_high_precision_mv;
  int cur_frame_force_integer_mv;  // 0 the default in AOM, 1 only integer
  int mi_rows;
  int mi_cols;
  int mi_stride;

  // Whether to use previous frame's motion vectors for prediction.
  int allow_ref_frame_mvs;

  int ref_frame_sign_bias[TOTAL_REFS_PER_FRAME]; /* Two state 0, 1 */
  int frame_parallel_decode;  // frame-based threading.

  unsigned int frame_offset;

  // External BufferPool passed from outside.
  BufferPool buffer_pool;

  WarpedMotionParams global_motion[TOTAL_REFS_PER_FRAME];
  struct {
    BLOCK_SIZE sb_size;
    int enable_order_hint;
    int order_hint_bits_minus1;
  } seq_params;
  TPL_MV_REF *tpl_mvs;
  // TODO(jingning): This can be combined with sign_bias later.
  int8_t ref_frame_side[TOTAL_REFS_PER_FRAME];

    int ref_buf_idx[INTER_REFS_PER_FRAME];
    int ref_order_hint[INTER_REFS_PER_FRAME];
} AV1_COMMON;

static INLINE void integer_mv_precision(MV *mv) {
  int mod = (mv->row % 8);
  if (mod != 0) {
    mv->row -= mod;
    if (abs(mod) > 4) {
      if (mod > 0) {
        mv->row += 8;
      } else {
        mv->row -= 8;
      }
    }
  }

  mod = (mv->col % 8);
  if (mod != 0) {
    mv->col -= mod;
    if (abs(mod) > 4) {
      if (mod > 0) {
        mv->col += 8;
      } else {
        mv->col -= 8;
      }
    }
  }
}

static INLINE int clamp(int value, int low, int high) {
  return value < low ? low : (value > high ? high : value);
}

static INLINE void clamp_mv(MV *mv, int min_col, int max_col, int min_row,
                            int max_row) {
  mv->col = clamp(mv->col, min_col, max_col);
  mv->row = clamp(mv->row, min_row, max_row);
}

static INLINE int is_intrabc_block(const MB_MODE_INFO *mbmi) {
  return mbmi->ref_frame[0] == INTRA_FRAME && mbmi->mv[0].as_mv.row != -0x8000;
  //return mbmi->use_intrabc;
}

static INLINE int is_inter_block(const MB_MODE_INFO *mbmi) {
  if (is_intrabc_block(mbmi)) return 1;
  return mbmi->ref_frame[0] > INTRA_FRAME;
}

static INLINE MV_REFERENCE_FRAME comp_ref0(int ref_idx) {
  static const MV_REFERENCE_FRAME lut[] = {
    LAST_FRAME,     // LAST_LAST2_FRAMES,
    LAST_FRAME,     // LAST_LAST3_FRAMES,
    LAST_FRAME,     // LAST_GOLDEN_FRAMES,
    BWDREF_FRAME,   // BWDREF_ALTREF_FRAMES,
    LAST2_FRAME,    // LAST2_LAST3_FRAMES
    LAST2_FRAME,    // LAST2_GOLDEN_FRAMES,
    LAST3_FRAME,    // LAST3_GOLDEN_FRAMES,
    BWDREF_FRAME,   // BWDREF_ALTREF2_FRAMES,
    ALTREF2_FRAME,  // ALTREF2_ALTREF_FRAMES,
  };
  assert(NELEMENTS(lut) == TOTAL_UNIDIR_COMP_REFS);
  return lut[ref_idx];
}

static INLINE MV_REFERENCE_FRAME comp_ref1(int ref_idx) {
  static const MV_REFERENCE_FRAME lut[] = {
    LAST2_FRAME,    // LAST_LAST2_FRAMES,
    LAST3_FRAME,    // LAST_LAST3_FRAMES,
    GOLDEN_FRAME,   // LAST_GOLDEN_FRAMES,
    ALTREF_FRAME,   // BWDREF_ALTREF_FRAMES,
    LAST3_FRAME,    // LAST2_LAST3_FRAMES
    GOLDEN_FRAME,   // LAST2_GOLDEN_FRAMES,
    GOLDEN_FRAME,   // LAST3_GOLDEN_FRAMES,
    ALTREF2_FRAME,  // BWDREF_ALTREF2_FRAMES,
    ALTREF_FRAME,   // ALTREF2_ALTREF_FRAMES,
  };
  assert(NELEMENTS(lut) == TOTAL_UNIDIR_COMP_REFS);
  return lut[ref_idx];
}

#define WARPEDMODEL_PREC_BITS 16
#define GM_TRANS_ONLY_PREC_DIFF (WARPEDMODEL_PREC_BITS - 3)
#define WARPEDMODEL_ROW3HOMO_PREC_BITS 16

static INLINE int convert_to_trans_prec(int allow_hp, int coor) {
  if (allow_hp)
    return ROUND_POWER_OF_TWO_SIGNED(coor, WARPEDMODEL_PREC_BITS - 3);
  else
    return ROUND_POWER_OF_TWO_SIGNED(coor, WARPEDMODEL_PREC_BITS - 2) * 2;
}

static INLINE int block_center_x(int mi_col, BLOCK_SIZE bs) {
  const int bw = block_size_wide[bs];
  return mi_col * MI_SIZE + bw / 2 - 1;
}

static INLINE int block_center_y(int mi_row, BLOCK_SIZE bs) {
  const int bh = block_size_high[bs];
  return mi_row * MI_SIZE + bh / 2 - 1;
}

// Convert a global motion vector into a motion vector at the centre of the
// given block.
//
// The resulting motion vector will have three fractional bits of precision. If
// allow_hp is zero, the bottom bit will always be zero. If CONFIG_AMVR and
// is_integer is true, the bottom three bits will be zero (so the motion vector
// represents an integer)
static INLINE int_mv gm_get_motion_vector(const WarpedMotionParams *gm,
                                          int allow_hp, BLOCK_SIZE bsize,
                                          int mi_col, int mi_row,
                                          int is_integer) {
  int_mv res;
  const int32_t *mat = gm->wmmat;
  int x, y, tx, ty;

  if (gm->wmtype == TRANSLATION) {
    // All global motion vectors are stored with WARPEDMODEL_PREC_BITS (16)
    // bits of fractional precision. The offset for a translation is stored in
    // entries 0 and 1. For translations, all but the top three (two if
    // cm->allow_high_precision_mv is false) fractional bits are always zero.
    //
    // After the right shifts, there are 3 fractional bits of precision. If
    // allow_hp is false, the bottom bit is always zero (so we don't need a
    // call to convert_to_trans_prec here)
    res.as_mv.row = gm->wmmat[0] >> GM_TRANS_ONLY_PREC_DIFF;
    res.as_mv.col = gm->wmmat[1] >> GM_TRANS_ONLY_PREC_DIFF;
    assert(IMPLIES(1 & (res.as_mv.row | res.as_mv.col), allow_hp));
    if (is_integer) {
      integer_mv_precision(&res.as_mv);
    }
    return res;
  }

  x = block_center_x(mi_col, bsize);
  y = block_center_y(mi_row, bsize);

  if (gm->wmtype == ROTZOOM) {
    assert(gm->wmmat[5] == gm->wmmat[2]);
    assert(gm->wmmat[4] == -gm->wmmat[3]);
  }
  if (gm->wmtype > AFFINE) {
    int xc = (int)((int64_t)mat[2] * x + (int64_t)mat[3] * y + mat[0]);
    int yc = (int)((int64_t)mat[4] * x + (int64_t)mat[5] * y + mat[1]);
    const int Z = (int)((int64_t)mat[6] * x + (int64_t)mat[7] * y +
                        (1 << WARPEDMODEL_ROW3HOMO_PREC_BITS));
    xc *= 1 << (WARPEDMODEL_ROW3HOMO_PREC_BITS - WARPEDMODEL_PREC_BITS);
    yc *= 1 << (WARPEDMODEL_ROW3HOMO_PREC_BITS - WARPEDMODEL_PREC_BITS);
    xc = (int)(xc > 0 ? ((int64_t)xc + Z / 2) / Z : ((int64_t)xc - Z / 2) / Z);
    yc = (int)(yc > 0 ? ((int64_t)yc + Z / 2) / Z : ((int64_t)yc - Z / 2) / Z);
    tx = convert_to_trans_prec(allow_hp, xc) - (x << 3);
    ty = convert_to_trans_prec(allow_hp, yc) - (y << 3);
  } else {
    const int xc =
        (mat[2] - (1 << WARPEDMODEL_PREC_BITS)) * x + mat[3] * y + mat[0];
    const int yc =
        mat[4] * x + (mat[5] - (1 << WARPEDMODEL_PREC_BITS)) * y + mat[1];
    tx = convert_to_trans_prec(allow_hp, xc);
    ty = convert_to_trans_prec(allow_hp, yc);
  }

  res.as_mv.row = ty;
  res.as_mv.col = tx;

  if (is_integer) {
    integer_mv_precision(&res.as_mv);
  }
  return res;
}

static INLINE int have_newmv_in_inter_mode(PREDICTION_MODE mode) {
  return (mode == NEWMV || mode == NEW_NEWMV || mode == NEAREST_NEWMV ||
          mode == NEW_NEARESTMV || mode == NEAR_NEWMV || mode == NEW_NEARMV);
}

/*
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
 *
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
 */
#ifndef AV1_COMMON_MVREF_COMMON_H_
#define AV1_COMMON_MVREF_COMMON_H_

//#include "av1/common/onyxc_int.h"
//#include "av1/common/blockd.h"

#ifdef __cplusplus
extern "C" {
#endif

#define MVREF_ROW_COLS 3

// Set the upper limit of the motion vector component magnitude.
// This would make a motion vector fit in 26 bits. Plus 3 bits for the
// reference frame index. A tuple of motion vector can hence be stored within
// 32 bit range for efficient load/store operations.
#define REFMVS_LIMIT ((1 << 12) - 1)

typedef struct position {
  int row;
  int col;
} POSITION;

// clamp_mv_ref
#define MV_BORDER (16 << 3)  // Allow 16 pels in 1/8th pel units

static INLINE int get_relative_dist(const AV1_COMMON *cm, int a, int b) {
  if (!cm->seq_params.enable_order_hint) return 0;

  const int bits = cm->seq_params.order_hint_bits_minus1 + 1;

  assert(bits >= 1);
  assert(a >= 0 && a < (1 << bits));
  assert(b >= 0 && b < (1 << bits));

  int diff = a - b;
  int m = 1 << (bits - 1);
  diff = (diff & (m - 1)) - (diff & m);
  return diff;
}

static INLINE void clamp_mv_ref(MV *mv, int bw, int bh, const MACROBLOCKD *xd) {
  clamp_mv(mv, xd->mb_to_left_edge - bw * 8 - MV_BORDER,
           xd->mb_to_right_edge + bw * 8 + MV_BORDER,
           xd->mb_to_top_edge - bh * 8 - MV_BORDER,
           xd->mb_to_bottom_edge + bh * 8 + MV_BORDER);
}

// This function returns either the appropriate sub block or block's mv
// on whether the block_size < 8x8 and we have check_sub_blocks set.
static INLINE int_mv get_sub_block_mv(const MB_MODE_INFO *candidate,
                                      int which_mv, int search_col) {
  (void)search_col;
  return candidate->mv[which_mv];
}

// Checks that the given mi_row, mi_col and search point
// are inside the borders of the tile.
static INLINE int is_inside(const TileInfo *const tile, int mi_col, int mi_row,
                            int mi_rows, const POSITION *mi_pos) {
  const int dependent_horz_tile_flag = 0;
  if (dependent_horz_tile_flag && !tile->tg_horz_boundary) {
    return !(mi_row + mi_pos->row < 0 ||
             mi_col + mi_pos->col < tile->mi_col_start ||
             mi_row + mi_pos->row >= mi_rows ||
             mi_col + mi_pos->col >= tile->mi_col_end);
  } else {
    return !(mi_row + mi_pos->row < tile->mi_row_start ||
             mi_col + mi_pos->col < tile->mi_col_start ||
             mi_row + mi_pos->row >= tile->mi_row_end ||
             mi_col + mi_pos->col >= tile->mi_col_end);
  }
}

static INLINE int find_valid_row_offset(const TileInfo *const tile, int mi_row,
                                        int mi_rows, int row_offset) {
  const int dependent_horz_tile_flag = 0;
  if (dependent_horz_tile_flag && !tile->tg_horz_boundary)
    return clamp(row_offset, -mi_row, mi_rows - mi_row - 1);
  else
    return clamp(row_offset, tile->mi_row_start - mi_row,
                 tile->mi_row_end - mi_row - 1);
}

static INLINE int find_valid_col_offset(const TileInfo *const tile, int mi_col,
                                        int col_offset) {
  return clamp(col_offset, tile->mi_col_start - mi_col,
               tile->mi_col_end - mi_col - 1);
}

static INLINE void lower_mv_precision(MV *mv, int allow_hp,
                                      int is_integer) {
  if (is_integer) {
    integer_mv_precision(mv);
  } else {
    if (!allow_hp) {
      if (mv->row & 1) mv->row += (mv->row > 0 ? -1 : 1);
      if (mv->col & 1) mv->col += (mv->col > 0 ? -1 : 1);
    }
  }
}

static INLINE int8_t get_uni_comp_ref_idx(const MV_REFERENCE_FRAME *const rf) {
  // Single ref pred
  if (rf[1] <= INTRA_FRAME) return -1;

  // Bi-directional comp ref pred
  if ((rf[0] < BWDREF_FRAME) && (rf[1] >= BWDREF_FRAME)) return -1;

  for (int8_t ref_idx = 0; ref_idx < TOTAL_UNIDIR_COMP_REFS; ++ref_idx) {
    if (rf[0] == comp_ref0(ref_idx) && rf[1] == comp_ref1(ref_idx))
      return ref_idx;
  }
  return -1;
}

static INLINE int8_t av1_ref_frame_type(const MV_REFERENCE_FRAME *const rf) {
  if (rf[1] > INTRA_FRAME) {
    const int8_t uni_comp_ref_idx = get_uni_comp_ref_idx(rf);
    if (uni_comp_ref_idx >= 0) {
      assert((REF_FRAMES + FWD_REFS * BWD_REFS + uni_comp_ref_idx) <
             MODE_CTX_REF_FRAMES);
      return REF_FRAMES + FWD_REFS * BWD_REFS + uni_comp_ref_idx;
    } else {
      return REF_FRAMES + FWD_RF_OFFSET(rf[0]) +
             BWD_RF_OFFSET(rf[1]) * FWD_REFS;
    }
  }

  return rf[0];
}

// clang-format off
static MV_REFERENCE_FRAME ref_frame_map[TOTAL_COMP_REFS][2] = {
  { LAST_FRAME, BWDREF_FRAME },  { LAST2_FRAME, BWDREF_FRAME },
  { LAST3_FRAME, BWDREF_FRAME }, { GOLDEN_FRAME, BWDREF_FRAME },

  { LAST_FRAME, ALTREF2_FRAME },  { LAST2_FRAME, ALTREF2_FRAME },
  { LAST3_FRAME, ALTREF2_FRAME }, { GOLDEN_FRAME, ALTREF2_FRAME },

  { LAST_FRAME, ALTREF_FRAME },  { LAST2_FRAME, ALTREF_FRAME },
  { LAST3_FRAME, ALTREF_FRAME }, { GOLDEN_FRAME, ALTREF_FRAME },

  { LAST_FRAME, LAST2_FRAME }, { LAST_FRAME, LAST3_FRAME },
  { LAST_FRAME, GOLDEN_FRAME }, { BWDREF_FRAME, ALTREF_FRAME },

  // NOTE: Following reference frame pairs are not supported to be explicitly
  //       signalled, but they are possibly chosen by the use of skip_mode,
  //       which may use the most recent one-sided reference frame pair.
  { LAST2_FRAME, LAST3_FRAME }, { LAST2_FRAME, GOLDEN_FRAME },
  { LAST3_FRAME, GOLDEN_FRAME }, {BWDREF_FRAME, ALTREF2_FRAME},
  { ALTREF2_FRAME, ALTREF_FRAME }
};
// clang-format on

static INLINE void av1_set_ref_frame(MV_REFERENCE_FRAME *rf,
                                     int8_t ref_frame_type) {
  if (ref_frame_type >= REF_FRAMES) {
    rf[0] = ref_frame_map[ref_frame_type - REF_FRAMES][0];
    rf[1] = ref_frame_map[ref_frame_type - REF_FRAMES][1];
  } else {
    rf[0] = ref_frame_type;
    rf[1] = NONE_FRAME;
    assert(ref_frame_type > NONE_FRAME);
  }
}

static uint16_t compound_mode_ctx_map[3][COMP_NEWMV_CTXS] = {
  { 0, 1, 1, 1, 1 },
  { 1, 2, 3, 4, 4 },
  { 4, 4, 5, 6, 7 },
};

static INLINE int16_t av1_mode_context_analyzer(
    const int16_t *const mode_context, const MV_REFERENCE_FRAME *const rf) {
  const int8_t ref_frame = av1_ref_frame_type(rf);

  if (rf[1] <= INTRA_FRAME) return mode_context[ref_frame];

  const int16_t newmv_ctx = mode_context[ref_frame] & NEWMV_CTX_MASK;
  const int16_t refmv_ctx =
      (mode_context[ref_frame] >> REFMV_OFFSET) & REFMV_CTX_MASK;

  const int16_t comp_ctx = compound_mode_ctx_map[refmv_ctx >> 1][AOMMIN(
      newmv_ctx, COMP_NEWMV_CTXS - 1)];
  return comp_ctx;
}

769
static void av1_setup_frame_buf_refs(AV1_COMMON *cm);
770 771 772 773 774 775
void av1_setup_frame_sign_bias(AV1_COMMON *cm);
void av1_setup_skip_mode_allowed(AV1_COMMON *cm);

void av1_copy_frame_mvs(const AV1_COMMON *const cm, MB_MODE_INFO *mi,
                        int mi_row, int mi_col, int x_mis, int y_mis);

776
static void av1_find_mv_refs(const AV1_COMMON *cm, const MACROBLOCKD *xd,
777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606
                      MB_MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
                      uint8_t ref_mv_count[MODE_CTX_REF_FRAMES],
                      CANDIDATE_MV ref_mv_stack[][MAX_REF_MV_STACK_SIZE],
                      int_mv mv_ref_list[][MAX_MV_REF_CANDIDATES],
                      int_mv *global_mvs, int mi_row, int mi_col,
                      int16_t *mode_context);

int selectSamples(MV *mv, int *pts, int *pts_inref, int len, BLOCK_SIZE bsize);
int findSamples(const AV1_COMMON *cm, MACROBLOCKD *xd, int mi_row, int mi_col,
                int *pts, int *pts_inref);

#define INTRABC_DELAY_PIXELS 256  //  Delay of 256 pixels
#define INTRABC_DELAY_SB64 (INTRABC_DELAY_PIXELS / 64)
#define USE_WAVE_FRONT 1  // Use only top left area of frame for reference.

#ifdef __cplusplus
}  // extern "C"
#endif

#endif  // AV1_COMMON_MVREF_COMMON_H_

/*
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
 *
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
 */

#include <stdlib.h>

//#include "av1/common/mvref_common.h"
//#include "av1/common/warped_motion.h"

// Although we assign 32 bit integers, all the values are strictly under 14
// bits.
static int div_mult[32] = { 0,    16384, 8192, 5461, 4096, 3276, 2730, 2340,
                            2048, 1820,  1638, 1489, 1365, 1260, 1170, 1092,
                            1024, 963,   910,  862,  819,  780,  744,  712,
                            682,  655,   630,  606,  585,  564,  546,  528 };

// TODO(jingning): Consider the use of lookup table for (num / den)
// altogether.
static void get_mv_projection(MV *output, MV ref, int num, int den) {
  den = AOMMIN(den, MAX_FRAME_DISTANCE);
  num = num > 0 ? AOMMIN(num, MAX_FRAME_DISTANCE)
                : AOMMAX(num, -MAX_FRAME_DISTANCE);
  int mv_row = ROUND_POWER_OF_TWO_SIGNED(ref.row * num * div_mult[den], 14);
  int mv_col = ROUND_POWER_OF_TWO_SIGNED(ref.col * num * div_mult[den], 14);
  const int clamp_max = MV_UPP - 1;
  const int clamp_min = MV_LOW + 1;
  output->row = (int16_t)clamp(mv_row, clamp_min, clamp_max);
  output->col = (int16_t)clamp(mv_col, clamp_min, clamp_max);
}

static void add_ref_mv_candidate(
    const MB_MODE_INFO *const candidate, const MV_REFERENCE_FRAME rf[2],
    uint8_t *refmv_count, uint8_t *ref_match_count, uint8_t *newmv_count,
    CANDIDATE_MV *ref_mv_stack, int_mv *gm_mv_candidates,
    const WarpedMotionParams *gm_params, int col, int weight) {
  if (!is_inter_block(candidate)) return;  // for intrabc
  int index = 0, ref;
  assert(weight % 2 == 0);

  if (rf[1] == NONE_FRAME) {
    // single reference frame
    for (ref = 0; ref < 2; ++ref) {
      if (candidate->ref_frame[ref] == rf[0]) {
        int_mv this_refmv;
        if (is_global_mv_block(candidate, gm_params[rf[0]].wmtype))
          this_refmv = gm_mv_candidates[0];
        else
          this_refmv = get_sub_block_mv(candidate, ref, col);

        for (index = 0; index < *refmv_count; ++index)
          if (ref_mv_stack[index].this_mv.as_int == this_refmv.as_int) break;

        if (index < *refmv_count) ref_mv_stack[index].weight += weight;

        // Add a new item to the list.
        if (index == *refmv_count && *refmv_count < MAX_REF_MV_STACK_SIZE) {
          ref_mv_stack[index].this_mv = this_refmv;
          ref_mv_stack[index].weight = weight;
          ++(*refmv_count);
        }
        if (have_newmv_in_inter_mode(candidate->mode)) ++*newmv_count;
        ++*ref_match_count;
      }
    }
  } else {
    // compound reference frame
    if (candidate->ref_frame[0] == rf[0] && candidate->ref_frame[1] == rf[1]) {
      int_mv this_refmv[2];

      for (ref = 0; ref < 2; ++ref) {
        if (is_global_mv_block(candidate, gm_params[rf[ref]].wmtype))
          this_refmv[ref] = gm_mv_candidates[ref];
        else
          this_refmv[ref] = get_sub_block_mv(candidate, ref, col);
      }

      for (index = 0; index < *refmv_count; ++index)
        if ((ref_mv_stack[index].this_mv.as_int == this_refmv[0].as_int) &&
            (ref_mv_stack[index].comp_mv.as_int == this_refmv[1].as_int))
          break;

      if (index < *refmv_count) ref_mv_stack[index].weight += weight;

      // Add a new item to the list.
      if (index == *refmv_count && *refmv_count < MAX_REF_MV_STACK_SIZE) {
        ref_mv_stack[index].this_mv = this_refmv[0];
        ref_mv_stack[index].comp_mv = this_refmv[1];
        ref_mv_stack[index].weight = weight;
        ++(*refmv_count);
      }
      if (have_newmv_in_inter_mode(candidate->mode)) ++*newmv_count;
      ++*ref_match_count;
    }
  }
}

static void scan_row_mbmi(const AV1_COMMON *cm, const MACROBLOCKD *xd,
                          int mi_row, int mi_col,
                          const MV_REFERENCE_FRAME rf[2], int row_offset,
                          CANDIDATE_MV *ref_mv_stack, uint8_t *refmv_count,
                          uint8_t *ref_match_count, uint8_t *newmv_count,
                          int_mv *gm_mv_candidates, int max_row_offset,
                          int *processed_rows) {
  int end_mi = AOMMIN(xd->n8_w, cm->mi_cols - mi_col);
  end_mi = AOMMIN(end_mi, mi_size_wide[BLOCK_64X64]);
  const int n8_w_8 = mi_size_wide[BLOCK_8X8];
  const int n8_w_16 = mi_size_wide[BLOCK_16X16];
  int i;
  int col_offset = 0;
  const int shift = 0;
  // TODO(jingning): Revisit this part after cb4x4 is stable.
  if (abs(row_offset) > 1) {
    col_offset = 1;
    if ((mi_col & 0x01) && xd->n8_w < n8_w_8) --col_offset;
  }
  const int use_step_16 = (xd->n8_w >= 16);
  MB_MODE_INFO *const candidate_mi0 = xd->mi + row_offset * xd->mi_stride;
  (void)mi_row;

  for (i = 0; i < end_mi;) {
    const MB_MODE_INFO *const candidate = &candidate_mi0[col_offset + i];
    const int candidate_bsize = candidate->sb_type;
    const int n8_w = mi_size_wide[candidate_bsize];
    int len = AOMMIN(xd->n8_w, n8_w);
    if (use_step_16)
      len = AOMMAX(n8_w_16, len);
    else if (abs(row_offset) > 1)
      len = AOMMAX(len, n8_w_8);

    int weight = 2;
    if (xd->n8_w >= n8_w_8 && xd->n8_w <= n8_w) {
      int inc = AOMMIN(-max_row_offset + row_offset + 1,
                       mi_size_high[candidate_bsize]);
      // Obtain range used in weight calculation.
      weight = AOMMAX(weight, (inc << shift));
      // Update processed rows.
      *processed_rows = inc - row_offset - 1;
    }

    add_ref_mv_candidate(candidate, rf, refmv_count, ref_match_count,
                         newmv_count, ref_mv_stack, gm_mv_candidates,
                         cm->global_motion, col_offset + i, len * weight);

    i += len;
  }
}

static void scan_col_mbmi(const AV1_COMMON *cm, const MACROBLOCKD *xd,
                          int mi_row, int mi_col,
                          const MV_REFERENCE_FRAME rf[2], int col_offset,
                          CANDIDATE_MV *ref_mv_stack, uint8_t *refmv_count,
                          uint8_t *ref_match_count, uint8_t *newmv_count,
                          int_mv *gm_mv_candidates, int max_col_offset,
                          int *processed_cols) {
  int end_mi = AOMMIN(xd->n8_h, cm->mi_rows - mi_row);
  end_mi = AOMMIN(end_mi, mi_size_high[BLOCK_64X64]);
  const int n8_h_8 = mi_size_high[BLOCK_8X8];
  const int n8_h_16 = mi_size_high[BLOCK_16X16];
  int i;
  int row_offset = 0;
  const int shift = 0;
  if (abs(col_offset) > 1) {
    row_offset = 1;
    if ((mi_row & 0x01) && xd->n8_h < n8_h_8) --row_offset;
  }
  const int use_step_16 = (xd->n8_h >= 16);
  (void)mi_col;

  for (i = 0; i < end_mi;) {
    const MB_MODE_INFO *const candidate =
        &xd->mi[(row_offset + i) * xd->mi_stride + col_offset];
    const int candidate_bsize = candidate->sb_type;
    const int n8_h = mi_size_high[candidate_bsize];
    int len = AOMMIN(xd->n8_h, n8_h);
    if (use_step_16)
      len = AOMMAX(n8_h_16, len);
    else if (abs(col_offset) > 1)
      len = AOMMAX(len, n8_h_8);

    int weight = 2;
    if (xd->n8_h >= n8_h_8 && xd->n8_h <= n8_h) {
      int inc = AOMMIN(-max_col_offset + col_offset + 1,
                       mi_size_wide[candidate_bsize]);
      // Obtain range used in weight calculation.
      weight = AOMMAX(weight, (inc << shift));
      // Update processed cols.
      *processed_cols = inc - col_offset - 1;
    }

    add_ref_mv_candidate(candidate, rf, refmv_count, ref_match_count,
                         newmv_count, ref_mv_stack, gm_mv_candidates,
                         cm->global_motion, col_offset, len * weight);

    i += len;
  }
}

static void scan_blk_mbmi(const AV1_COMMON *cm, const MACROBLOCKD *xd,
                          const int mi_row, const int mi_col,
                          const MV_REFERENCE_FRAME rf[2], int row_offset,
                          int col_offset, CANDIDATE_MV *ref_mv_stack,
                          uint8_t *ref_match_count, uint8_t *newmv_count,
                          int_mv *gm_mv_candidates,
                          uint8_t refmv_count[MODE_CTX_REF_FRAMES]) {
  const TileInfo *const tile = &xd->tile;
  POSITION mi_pos;

  mi_pos.row = row_offset;
  mi_pos.col = col_offset;

  if (is_inside(tile, mi_col, mi_row, cm->mi_rows, &mi_pos)) {
    const MB_MODE_INFO *const candidate =
        &xd->mi[mi_pos.row * xd->mi_stride + mi_pos.col];
    const int len = mi_size_wide[BLOCK_8X8];

    add_ref_mv_candidate(candidate, rf, refmv_count, ref_match_count,
                         newmv_count, ref_mv_stack, gm_mv_candidates,
                         cm->global_motion, mi_pos.col, 2 * len);
  }  // Analyze a single 8x8 block motion information.
}

static int has_top_right(const AV1_COMMON *cm, const MACROBLOCKD *xd,
                         int mi_row, int mi_col, int bs) {
  const int sb_mi_size = mi_size_wide[cm->seq_params.sb_size];
  const int mask_row = mi_row & (sb_mi_size - 1);
  const int mask_col = mi_col & (sb_mi_size - 1);

  if (bs > mi_size_wide[BLOCK_64X64]) return 0;

  // In a split partition all apart from the bottom right has a top right
  int has_tr = !((mask_row & bs) && (mask_col & bs));

  // bs > 0 and bs is a power of 2
  assert(bs > 0 && !(bs & (bs - 1)));

  // For each 4x4 group of blocks, when the bottom right is decoded the blocks
  // to the right have not been decoded therefore the bottom right does
  // not have a top right
  while (bs < sb_mi_size) {
    if (mask_col & bs) {
      if ((mask_col & (2 * bs)) && (mask_row & (2 * bs))) {
        has_tr = 0;
        break;
      }
    } else {
      break;
    }
    bs <<= 1;
  }

  // The left hand of two vertical rectangles always has a top right (as the
  // block above will have been decoded)
  if (xd->n8_w < xd->n8_h)
    if (!xd->is_sec_rect) has_tr = 1;

  // The bottom of two horizontal rectangles never has a top right (as the block
  // to the right won't have been decoded)
  if (xd->n8_w > xd->n8_h)
    if (xd->is_sec_rect) has_tr = 0;

  // The bottom left square of a Vertical A (in the old format) does
  // not have a top right as it is decoded before the right hand
  // rectangle of the partition
  if (xd->cur_mi.partition == PARTITION_VERT_A) {
    if (xd->n8_w == xd->n8_h)
      if (mask_row & bs) has_tr = 0;
  }

  return has_tr;
}

static int check_sb_border(const int mi_row, const int mi_col,
                           const int row_offset, const int col_offset) {
  const int sb_mi_size = mi_size_wide[BLOCK_64X64];
  const int row = mi_row & (sb_mi_size - 1);
  const int col = mi_col & (sb_mi_size - 1);

  if (row + row_offset < 0 || row + row_offset >= sb_mi_size ||
      col + col_offset < 0 || col + col_offset >= sb_mi_size)
    return 0;

  return 1;
}

static int add_tpl_ref_mv(const AV1_COMMON *cm, const MACROBLOCKD *xd,
                          int mi_row, int mi_col, MV_REFERENCE_FRAME ref_frame,
                          int blk_row, int blk_col, int_mv *gm_mv_candidates,
                          uint8_t refmv_count[MODE_CTX_REF_FRAMES],
                          CANDIDATE_MV ref_mv_stacks[][MAX_REF_MV_STACK_SIZE],
                          int16_t *mode_context) {
  POSITION mi_pos;
  int idx;
  const int weight_unit = 1;  // mi_size_wide[BLOCK_8X8];

  mi_pos.row = (mi_row & 0x01) ? blk_row : blk_row + 1;
  mi_pos.col = (mi_col & 0x01) ? blk_col : blk_col + 1;

  if (!is_inside(&xd->tile, mi_col, mi_row, cm->mi_rows, &mi_pos)) return 0;

  const TPL_MV_REF *prev_frame_mvs =
      cm->tpl_mvs + ((mi_row + mi_pos.row) >> 1) * (cm->mi_stride >> 1) +
      ((mi_col + mi_pos.col) >> 1);

  MV_REFERENCE_FRAME rf[2];
  av1_set_ref_frame(rf, ref_frame);

  if (rf[1] == NONE_FRAME) {
    int cur_frame_index = cm->cur_frame.cur_frame_offset;
    int buf_idx_0 = cm->frame_refs[FWD_RF_OFFSET(rf[0])].idx;
    int frame0_index = cm->buffer_pool.frame_bufs[buf_idx_0].cur_frame_offset;
    int cur_offset_0 = get_relative_dist(cm, cur_frame_index, frame0_index);
    CANDIDATE_MV *ref_mv_stack = ref_mv_stacks[rf[0]];

    if (prev_frame_mvs->mfmv0.as_int != INVALID_MV) {
      int_mv this_refmv;

      get_mv_projection(&this_refmv.as_mv, prev_frame_mvs->mfmv0.as_mv,
                        cur_offset_0, prev_frame_mvs->ref_frame_offset);
      lower_mv_precision(&this_refmv.as_mv, cm->allow_high_precision_mv,
                         cm->cur_frame_force_integer_mv);

      if (blk_row == 0 && blk_col == 0)
        if (abs(this_refmv.as_mv.row - gm_mv_candidates[0].as_mv.row) >= 16 ||
            abs(this_refmv.as_mv.col - gm_mv_candidates[0].as_mv.col) >= 16)
          mode_context[ref_frame] |= (1 << GLOBALMV_OFFSET);

      for (idx = 0; idx < refmv_count[rf[0]]; ++idx)
        if (this_refmv.as_int == ref_mv_stack[idx].this_mv.as_int) break;

      if (idx < refmv_count[rf[0]]) ref_mv_stack[idx].weight += 2 * weight_unit;

      if (idx == refmv_count[rf[0]] &&
          refmv_count[rf[0]] < MAX_REF_MV_STACK_SIZE) {
        ref_mv_stack[idx].this_mv.as_int = this_refmv.as_int;
        ref_mv_stack[idx].weight = 2 * weight_unit;
        ++(refmv_count[rf[0]]);
      }

      return 1;
    }
  } else {
    // Process compound inter mode
    int cur_frame_index = cm->cur_frame.cur_frame_offset;
    int buf_idx_0 = cm->frame_refs[FWD_RF_OFFSET(rf[0])].idx;
    int frame0_index = cm->buffer_pool.frame_bufs[buf_idx_0].cur_frame_offset;

    int cur_offset_0 = get_relative_dist(cm, cur_frame_index, frame0_index);
    int buf_idx_1 = cm->frame_refs[FWD_RF_OFFSET(rf[1])].idx;
    int frame1_index = cm->buffer_pool.frame_bufs[buf_idx_1].cur_frame_offset;
    int cur_offset_1 = get_relative_dist(cm, cur_frame_index, frame1_index);
    CANDIDATE_MV *ref_mv_stack = ref_mv_stacks[ref_frame];

    if (prev_frame_mvs->mfmv0.as_int != INVALID_MV) {
      int_mv this_refmv;
      int_mv comp_refmv;
      get_mv_projection(&this_refmv.as_mv, prev_frame_mvs->mfmv0.as_mv,
                        cur_offset_0, prev_frame_mvs->ref_frame_offset);
      get_mv_projection(&comp_refmv.as_mv, prev_frame_mvs->mfmv0.as_mv,
                        cur_offset_1, prev_frame_mvs->ref_frame_offset);

      lower_mv_precision(&this_refmv.as_mv, cm->allow_high_precision_mv,
                         cm->cur_frame_force_integer_mv);
      lower_mv_precision(&comp_refmv.as_mv, cm->allow_high_precision_mv,
                         cm->cur_frame_force_integer_mv);

      if (blk_row == 0 && blk_col == 0)
        if (abs(this_refmv.as_mv.row - gm_mv_candidates[0].as_mv.row) >= 16 ||
            abs(this_refmv.as_mv.col - gm_mv_candidates[0].as_mv.col) >= 16 ||
            abs(comp_refmv.as_mv.row - gm_mv_candidates[1].as_mv.row) >= 16 ||
            abs(comp_refmv.as_mv.col - gm_mv_candidates[1].as_mv.col) >= 16)
          mode_context[ref_frame] |= (1 << GLOBALMV_OFFSET);

      for (idx = 0; idx < refmv_count[ref_frame]; ++idx)
        if (this_refmv.as_int == ref_mv_stack[idx].this_mv.as_int &&
            comp_refmv.as_int == ref_mv_stack[idx].comp_mv.as_int)
          break;

      if (idx < refmv_count[ref_frame])
        ref_mv_stack[idx].weight += 2 * weight_unit;

      if (idx == refmv_count[ref_frame] &&
          refmv_count[ref_frame] < MAX_REF_MV_STACK_SIZE) {
        ref_mv_stack[idx].this_mv.as_int = this_refmv.as_int;
        ref_mv_stack[idx].comp_mv.as_int = comp_refmv.as_int;
        ref_mv_stack[idx].weight = 2 * weight_unit;
        ++(refmv_count[ref_frame]);
      }
      return 1;
    }
  }
  return 0;
}

static void setup_ref_mv_list(
    const AV1_COMMON *cm, const MACROBLOCKD *xd, MV_REFERENCE_FRAME ref_frame,
    uint8_t refmv_count[MODE_CTX_REF_FRAMES],
    CANDIDATE_MV ref_mv_stack[][MAX_REF_MV_STACK_SIZE],
    int_mv mv_ref_list[][MAX_MV_REF_CANDIDATES], int_mv *gm_mv_candidates,
    int mi_row, int mi_col, int16_t *mode_context) {
  const int bs = AOMMAX(xd->n8_w, xd->n8_h);
  const int has_tr = has_top_right(cm, xd, mi_row, mi_col, bs);
  MV_REFERENCE_FRAME rf[2];

  const TileInfo *const tile = &xd->tile;
  int max_row_offset = 0, max_col_offset = 0;
  const int row_adj = (xd->n8_h < mi_size_high[BLOCK_8X8]) && (mi_row & 0x01);
  const int col_adj = (xd->n8_w < mi_size_wide[BLOCK_8X8]) && (mi_col & 0x01);
  int processed_rows = 0;
  int processed_cols = 0;

  av1_set_ref_frame(rf, ref_frame);
  mode_context[ref_frame] = 0;
  refmv_count[ref_frame] = 0;

  // Find valid maximum row/col offset.
  if (xd->up_available) {
    max_row_offset = -(MVREF_ROW_COLS << 1) + row_adj;

    if (xd->n8_h < mi_size_high[BLOCK_8X8])
      max_row_offset = -(2 << 1) + row_adj;

    max_row_offset =
        find_valid_row_offset(tile, mi_row, cm->mi_rows, max_row_offset);
  }

  if (xd->left_available) {
    max_col_offset = -(MVREF_ROW_COLS << 1) + col_adj;

    if (xd->n8_w < mi_size_wide[BLOCK_8X8])
      max_col_offset = -(2 << 1) + col_adj;

    max_col_offset = find_valid_col_offset(tile, mi_col, max_col_offset);
  }

  uint8_t col_match_count = 0;
  uint8_t row_match_count = 0;
  uint8_t newmv_count = 0;

  // Scan the first above row mode info. row_offset = -1;
  if (abs(max_row_offset) >= 1)
    scan_row_mbmi(cm, xd, mi_row, mi_col, rf, -1, ref_mv_stack[ref_frame],
                  &refmv_count[ref_frame], &row_match_count, &newmv_count,
                  gm_mv_candidates, max_row_offset, &processed_rows);
  // Scan the first left column mode info. col_offset = -1;
  if (abs(max_col_offset) >= 1)
    scan_col_mbmi(cm, xd, mi_row, mi_col, rf, -1, ref_mv_stack[ref_frame],
                  &refmv_count[ref_frame], &col_match_count, &newmv_count,
                  gm_mv_candidates, max_col_offset, &processed_cols);
  // Check top-right boundary
  if (has_tr)
    scan_blk_mbmi(cm, xd, mi_row, mi_col, rf, -1, xd->n8_w,
                  ref_mv_stack[ref_frame], &row_match_count, &newmv_count,
                  gm_mv_candidates, &refmv_count[ref_frame]);

  uint8_t nearest_match = (row_match_count > 0) + (col_match_count > 0);
  uint8_t nearest_refmv_count = refmv_count[ref_frame];

  // TODO(yunqing): for comp_search, do it for all 3 cases.
  for (int idx = 0; idx < nearest_refmv_count; ++idx)
    ref_mv_stack[ref_frame][idx].weight += REF_CAT_LEVEL;

  if (cm->allow_ref_frame_mvs) {
    int is_available = 0;
    const int voffset = AOMMAX(mi_size_high[BLOCK_8X8], xd->n8_h);
    const int hoffset = AOMMAX(mi_size_wide[BLOCK_8X8], xd->n8_w);
    const int blk_row_end = AOMMIN(xd->n8_h, mi_size_high[BLOCK_64X64]);
    const int blk_col_end = AOMMIN(xd->n8_w, mi_size_wide[BLOCK_64X64]);

    const int tpl_sample_pos[3][2] = {
      { voffset, -2 },
      { voffset, hoffset },
      { voffset - 2, hoffset },
    };
    const int allow_extension = (xd->n8_h >= mi_size_high[BLOCK_8X8]) &&
                                (xd->n8_h < mi_size_high[BLOCK_64X64]) &&
                                (xd->n8_w >= mi_size_wide[BLOCK_8X8]) &&
                                (xd->n8_w < mi_size_wide[BLOCK_64X64]);

    int step_h = (xd->n8_h >= mi_size_high[BLOCK_64X64])
                     ? mi_size_high[BLOCK_16X16]
                     : mi_size_high[BLOCK_8X8];
    int step_w = (xd->n8_w >= mi_size_wide[BLOCK_64X64])
                     ? mi_size_wide[BLOCK_16X16]
                     : mi_size_wide[BLOCK_8X8];

    for (int blk_row = 0; blk_row < blk_row_end; blk_row += step_h) {
      for (int blk_col = 0; blk_col < blk_col_end; blk_col += step_w) {
        int ret = add_tpl_ref_mv(cm, xd, mi_row, mi_col, ref_frame, blk_row,
                                 blk_col, gm_mv_candidates, refmv_count,
                                 ref_mv_stack, mode_context);
        if (blk_row == 0 && blk_col == 0) is_available = ret;
      }
    }

    if (is_available == 0) mode_context[ref_frame] |= (1 << GLOBALMV_OFFSET);

    for (int i = 0; i < 3 && allow_extension; ++i) {
      const int blk_row = tpl_sample_pos[i][0];
      const int blk_col = tpl_sample_pos[i][1];

      if (!check_sb_border(mi_row, mi_col, blk_row, blk_col)) continue;
      add_tpl_ref_mv(cm, xd, mi_row, mi_col, ref_frame, blk_row, blk_col,
                     gm_mv_candidates, refmv_count, ref_mv_stack, mode_context);
    }
  }

  uint8_t dummy_newmv_count = 0;

  // Scan the second outer area.
  scan_blk_mbmi(cm, xd, mi_row, mi_col, rf, -1, -1, ref_mv_stack[ref_frame],
                &row_match_count, &dummy_newmv_count, gm_mv_candidates,
                &refmv_count[ref_frame]);

  for (int idx = 2; idx <= MVREF_ROW_COLS; ++idx) {
    const int row_offset = -(idx << 1) + 1 + row_adj;
    const int col_offset = -(idx << 1) + 1 + col_adj;

    if (abs(row_offset) <= abs(max_row_offset) &&
        abs(row_offset) > processed_rows)
      scan_row_mbmi(cm, xd, mi_row, mi_col, rf, row_offset,
                    ref_mv_stack[ref_frame], &refmv_count[ref_frame],
                    &row_match_count, &dummy_newmv_count, gm_mv_candidates,
                    max_row_offset, &processed_rows);

    if (abs(col_offset) <= abs(max_col_offset) &&
        abs(col_offset) > processed_cols)
      scan_col_mbmi(cm, xd, mi_row, mi_col, rf, col_offset,
                    ref_mv_stack[ref_frame], &refmv_count[ref_frame],
                    &col_match_count, &dummy_newmv_count, gm_mv_candidates,
                    max_col_offset, &processed_cols);
  }

  uint8_t ref_match_count = (row_match_count > 0) + (col_match_count > 0);

  switch (nearest_match) {
    case 0:
      mode_context[ref_frame] |= 0;
      if (ref_match_count >= 1) mode_context[ref_frame] |= 1;
      if (ref_match_count == 1)
        mode_context[ref_frame] |= (1 << REFMV_OFFSET);
      else if (ref_match_count >= 2)
        mode_context[ref_frame] |= (2 << REFMV_OFFSET);
      break;
    case 1:
      mode_context[ref_frame] |= (newmv_count > 0) ? 2 : 3;
      if (ref_match_count == 1)
        mode_context[ref_frame] |= (3 << REFMV_OFFSET);
      else if (ref_match_count >= 2)
        mode_context[ref_frame] |= (4 << REFMV_OFFSET);
      break;
    case 2:
    default:
      if (newmv_count >= 1)
        mode_context[ref_frame] |= 4;
      else
        mode_context[ref_frame] |= 5;

      mode_context[ref_frame] |= (5 << REFMV_OFFSET);
      break;
  }

  // Rank the likelihood and assign nearest and near mvs.
  int len = nearest_refmv_count;
  while (len > 0) {
    int nr_len = 0;
    for (int idx = 1; idx < len; ++idx) {
      if (ref_mv_stack[ref_frame][idx - 1].weight <
          ref_mv_stack[ref_frame][idx].weight) {
        CANDIDATE_MV tmp_mv = ref_mv_stack[ref_frame][idx - 1];
        ref_mv_stack[ref_frame][idx - 1] = ref_mv_stack[ref_frame][idx];
        ref_mv_stack[ref_frame][idx] = tmp_mv;
        nr_len = idx;
      }
    }
    len = nr_len;
  }

  len = refmv_count[ref_frame];
  while (len > nearest_refmv_count) {
    int nr_len = nearest_refmv_count;
    for (int idx = nearest_refmv_count + 1; idx < len; ++idx) {
      if (ref_mv_stack[ref_frame][idx - 1].weight <
          ref_mv_stack[ref_frame][idx].weight) {
        CANDIDATE_MV tmp_mv = ref_mv_stack[ref_frame][idx - 1];
        ref_mv_stack[ref_frame][idx - 1] = ref_mv_stack[ref_frame][idx];
        ref_mv_stack[ref_frame][idx] = tmp_mv;
        nr_len = idx;
      }
    }
    len = nr_len;
  }

  if (rf[1] > NONE_FRAME) {
    // TODO(jingning, yunqing): Refactor and consolidate the compound and
    // single reference frame modes. Reduce unnecessary redundancy.
    if (refmv_count[ref_frame] < MAX_MV_REF_CANDIDATES) {
      int_mv ref_id[2][2], ref_diff[2][2];
      int ref_id_count[2] = { 0 }, ref_diff_count[2] = { 0 };

      int mi_width = AOMMIN(mi_size_wide[BLOCK_64X64], xd->n8_w);
      mi_width = AOMMIN(mi_width, cm->mi_cols - mi_col);
      int mi_height = AOMMIN(mi_size_high[BLOCK_64X64], xd->n8_h);
      mi_height = AOMMIN(mi_height, cm->mi_rows - mi_row);
      int mi_size = AOMMIN(mi_width, mi_height);

      for (int idx = 0; abs(max_row_offset) >= 1 && idx < mi_size;) {
        const MB_MODE_INFO *const candidate = &xd->mi[-xd->mi_stride + idx];
        const int candidate_bsize = candidate->sb_type;

        for (int rf_idx = 0; rf_idx < 2; ++rf_idx) {
          MV_REFERENCE_FRAME can_rf = candidate->ref_frame[rf_idx];

          for (int cmp_idx = 0; cmp_idx < 2; ++cmp_idx) {
            if (can_rf == rf[cmp_idx] && ref_id_count[cmp_idx] < 2) {
              ref_id[cmp_idx][ref_id_count[cmp_idx]] = candidate->mv[rf_idx];
              ++ref_id_count[cmp_idx];
            } else if (can_rf > INTRA_FRAME && ref_diff_count[cmp_idx] < 2) {
              int_mv this_mv = candidate->mv[rf_idx];
              if (cm->ref_frame_sign_bias[can_rf] !=
                  cm->ref_frame_sign_bias[rf[cmp_idx]]) {
                this_mv.as_mv.row = -this_mv.as_mv.row;
                this_mv.as_mv.col = -this_mv.as_mv.col;
              }
              ref_diff[cmp_idx][ref_diff_count[cmp_idx]] = this_mv;
              ++ref_diff_count[cmp_idx];
            }
          }
        }
        idx += mi_size_wide[candidate_bsize];
      }

      for (int idx = 0; abs(max_col_offset) >= 1 && idx < mi_size;) {
        const MB_MODE_INFO *const candidate = &xd->mi[idx * xd->mi_stride - 1];
        const int candidate_bsize = candidate->sb_type;

        for (int rf_idx = 0; rf_idx < 2; ++rf_idx) {
          MV_REFERENCE_FRAME can_rf = candidate->ref_frame[rf_idx];

          for (int cmp_idx = 0; cmp_idx < 2; ++cmp_idx) {
            if (can_rf == rf[cmp_idx] && ref_id_count[cmp_idx] < 2) {
              ref_id[cmp_idx][ref_id_count[cmp_idx]] = candidate->mv[rf_idx];
              ++ref_id_count[cmp_idx];
            } else if (can_rf > INTRA_FRAME && ref_diff_count[cmp_idx] < 2) {
              int_mv this_mv = candidate->mv[rf_idx];
              if (cm->ref_frame_sign_bias[can_rf] !=
                  cm->ref_frame_sign_bias[rf[cmp_idx]]) {
                this_mv.as_mv.row = -this_mv.as_mv.row;
                this_mv.as_mv.col = -this_mv.as_mv.col;
              }
              ref_diff[cmp_idx][ref_diff_count[cmp_idx]] = this_mv;
              ++ref_diff_count[cmp_idx];
            }
          }
        }
        idx += mi_size_high[candidate_bsize];
      }

      // Build up the compound mv predictor
      int_mv comp_list[3][2];

      for (int idx = 0; idx < 2; ++idx) {
        int comp_idx = 0;
        for (int list_idx = 0; list_idx < ref_id_count[idx] && comp_idx < 2;
             ++list_idx, ++comp_idx)
          comp_list[comp_idx][idx] = ref_id[idx][list_idx];
        for (int list_idx = 0; list_idx < ref_diff_count[idx] && comp_idx < 2;
             ++list_idx, ++comp_idx)
          comp_list[comp_idx][idx] = ref_diff[idx][list_idx];
        for (; comp_idx < 3; ++comp_idx)
          comp_list[comp_idx][idx] = gm_mv_candidates[idx];
      }

      if (refmv_count[ref_frame]) {
        assert(refmv_count[ref_frame] == 1);
        if (comp_list[0][0].as_int ==
                ref_mv_stack[ref_frame][0].this_mv.as_int &&
            comp_list[0][1].as_int ==
                ref_mv_stack[ref_frame][0].comp_mv.as_int) {
          ref_mv_stack[ref_frame][refmv_count[ref_frame]].this_mv =
              comp_list[1][0];
          ref_mv_stack[ref_frame][refmv_count[ref_frame]].comp_mv =
              comp_list[1][1];
        } else {
          ref_mv_stack[ref_frame][refmv_count[ref_frame]].this_mv =
              comp_list[0][0];
          ref_mv_stack[ref_frame][refmv_count[ref_frame]].comp_mv =
              comp_list[0][1];
        }
        ref_mv_stack[ref_frame][refmv_count[ref_frame]].weight = 2;
        ++refmv_count[ref_frame];
      } else {
        for (int idx = 0; idx < MAX_MV_REF_CANDIDATES; ++idx) {
          ref_mv_stack[ref_frame][refmv_count[ref_frame]].this_mv =
              comp_list[idx][0];
          ref_mv_stack[ref_frame][refmv_count[ref_frame]].comp_mv =
              comp_list[idx][1];
          ref_mv_stack[ref_frame][refmv_count[ref_frame]].weight = 2;
          ++refmv_count[ref_frame];
        }
      }
    }

    assert(refmv_count[ref_frame] >= 2);

    for (int idx = 0; idx < refmv_count[ref_frame]; ++idx) {
      clamp_mv_ref(&ref_mv_stack[ref_frame][idx].this_mv.as_mv,
                   xd->n8_w << MI_SIZE_LOG2, xd->n8_h << MI_SIZE_LOG2, xd);
      clamp_mv_ref(&ref_mv_stack[ref_frame][idx].comp_mv.as_mv,
                   xd->n8_w << MI_SIZE_LOG2, xd->n8_h << MI_SIZE_LOG2, xd);
    }
  } else {
    // Handle single reference frame extension
    int mi_width = AOMMIN(mi_size_wide[BLOCK_64X64], xd->n8_w);
    mi_width = AOMMIN(mi_width, cm->mi_cols - mi_col);
    int mi_height = AOMMIN(mi_size_high[BLOCK_64X64], xd->n8_h);
    mi_height = AOMMIN(mi_height, cm->mi_rows - mi_row);
    int mi_size = AOMMIN(mi_width, mi_height);

    for (int idx = 0; abs(max_row_offset) >= 1 && idx < mi_size &&
                      refmv_count[ref_frame] < MAX_MV_REF_CANDIDATES;) {
      const MB_MODE_INFO *const candidate = &xd->mi[-xd->mi_stride + idx];
      const int candidate_bsize = candidate->sb_type;

      // TODO(jingning): Refactor the following code.
      for (int rf_idx = 0; rf_idx < 2; ++rf_idx) {
        if (candidate->ref_frame[rf_idx] > INTRA_FRAME) {
          int_mv this_mv = candidate->mv[rf_idx];
          if (cm->ref_frame_sign_bias[candidate->ref_frame[rf_idx]] !=
              cm->ref_frame_sign_bias[ref_frame]) {
            this_mv.as_mv.row = -this_mv.as_mv.row;
            this_mv.as_mv.col = -this_mv.as_mv.col;
          }
          int stack_idx;
          for (stack_idx = 0; stack_idx < refmv_count[ref_frame]; ++stack_idx) {
            int_mv stack_mv = ref_mv_stack[ref_frame][stack_idx].this_mv;
            if (this_mv.as_int == stack_mv.as_int) break;
          }

          if (stack_idx == refmv_count[ref_frame]) {
            ref_mv_stack[ref_frame][stack_idx].this_mv = this_mv;

            // TODO(jingning): Set an arbitrary small number here. The weight
            // doesn't matter as long as it is properly initialized.
            ref_mv_stack[ref_frame][stack_idx].weight = 2;
            ++refmv_count[ref_frame];
          }
        }
      }
      idx += mi_size_wide[candidate_bsize];
    }

    for (int idx = 0; abs(max_col_offset) >= 1 && idx < mi_size &&
                      refmv_count[ref_frame] < MAX_MV_REF_CANDIDATES;) {
      const MB_MODE_INFO *const candidate = &xd->mi[idx * xd->mi_stride - 1];
      const int candidate_bsize = candidate->sb_type;

      // TODO(jingning): Refactor the following code.
      for (int rf_idx = 0; rf_idx < 2; ++rf_idx) {
        if (candidate->ref_frame[rf_idx] > INTRA_FRAME) {
          int_mv this_mv = candidate->mv[rf_idx];
          if (cm->ref_frame_sign_bias[candidate->ref_frame[rf_idx]] !=
              cm->ref_frame_sign_bias[ref_frame]) {
            this_mv.as_mv.row = -this_mv.as_mv.row;
            this_mv.as_mv.col = -this_mv.as_mv.col;
          }
          int stack_idx;
          for (stack_idx = 0; stack_idx < refmv_count[ref_frame]; ++stack_idx) {
            int_mv stack_mv = ref_mv_stack[ref_frame][stack_idx].this_mv;
            if (this_mv.as_int == stack_mv.as_int) break;
          }

          if (stack_idx == refmv_count[ref_frame]) {
            ref_mv_stack[ref_frame][stack_idx].this_mv = this_mv;

            // TODO(jingning): Set an arbitrary small number here. The weight
            // doesn't matter as long as it is properly initialized.
            ref_mv_stack[ref_frame][stack_idx].weight = 2;
            ++refmv_count[ref_frame];
          }
        }
      }
      idx += mi_size_high[candidate_bsize];
    }

    for (int idx = 0; idx < refmv_count[ref_frame]; ++idx) {
      clamp_mv_ref(&ref_mv_stack[ref_frame][idx].this_mv.as_mv,
                   xd->n8_w << MI_SIZE_LOG2, xd->n8_h << MI_SIZE_LOG2, xd);
    }

    if (mv_ref_list != NULL) {
      for (int idx = refmv_count[ref_frame]; idx < MAX_MV_REF_CANDIDATES; ++idx)
        mv_ref_list[rf[0]][idx].as_int = gm_mv_candidates[0].as_int;

      for (int idx = 0;
           idx < AOMMIN(MAX_MV_REF_CANDIDATES, refmv_count[ref_frame]); ++idx) {
        mv_ref_list[rf[0]][idx].as_int =
            ref_mv_stack[ref_frame][idx].this_mv.as_int;
      }
    }
  }
}

1607
static void av1_find_mv_refs(const AV1_COMMON *cm, const MACROBLOCKD *xd,
1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649
                      MB_MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
                      uint8_t ref_mv_count[MODE_CTX_REF_FRAMES],
                      CANDIDATE_MV ref_mv_stack[][MAX_REF_MV_STACK_SIZE],
                      int_mv mv_ref_list[][MAX_MV_REF_CANDIDATES],
                      int_mv *global_mvs, int mi_row, int mi_col,
                      int16_t *mode_context) {
  int_mv zeromv[2];
  BLOCK_SIZE bsize = mi->sb_type;
  MV_REFERENCE_FRAME rf[2];
  av1_set_ref_frame(rf, ref_frame);

  if (ref_frame < REF_FRAMES) {
    if (ref_frame != INTRA_FRAME) {
      global_mvs[ref_frame] = gm_get_motion_vector(
          &cm->global_motion[ref_frame], cm->allow_high_precision_mv, bsize,
          mi_col, mi_row, cm->cur_frame_force_integer_mv);
    } else {
      global_mvs[ref_frame].as_int = INVALID_MV;
    }
  }

  if (ref_frame != INTRA_FRAME) {
    zeromv[0].as_int =
        gm_get_motion_vector(&cm->global_motion[rf[0]],
                             cm->allow_high_precision_mv, bsize, mi_col, mi_row,
                             cm->cur_frame_force_integer_mv)
            .as_int;
    zeromv[1].as_int =
        (rf[1] != NONE_FRAME)
            ? gm_get_motion_vector(&cm->global_motion[rf[1]],
                                   cm->allow_high_precision_mv, bsize, mi_col,
                                   mi_row, cm->cur_frame_force_integer_mv)
                  .as_int
            : 0;
  } else {
    zeromv[0].as_int = zeromv[1].as_int = 0;
  }

  setup_ref_mv_list(cm, xd, ref_frame, ref_mv_count, ref_mv_stack, mv_ref_list,
                    zeromv, mi_row, mi_col, mode_context);
}

1650
static void av1_setup_frame_buf_refs(AV1_COMMON *cm) {
1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792
  cm->cur_frame.cur_frame_offset = cm->frame_offset;

  MV_REFERENCE_FRAME ref_frame;
  for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
    const int buf_idx = cm->frame_refs[ref_frame - LAST_FRAME].idx;
    if (buf_idx >= 0)
      cm->cur_frame.ref_frame_offset[ref_frame - LAST_FRAME] =
          cm->buffer_pool.frame_bufs[buf_idx].cur_frame_offset;
  }
}

#define MAX_OFFSET_WIDTH 64
#define MAX_OFFSET_HEIGHT 0

static int get_block_position(AV1_COMMON *cm, int *mi_r, int *mi_c, int blk_row,
                              int blk_col, MV mv, int sign_bias) {
  const int base_blk_row = (blk_row >> 3) << 3;
  const int base_blk_col = (blk_col >> 3) << 3;

  const int row_offset = (mv.row >= 0) ? (mv.row >> (4 + MI_SIZE_LOG2))
                                       : -((-mv.row) >> (4 + MI_SIZE_LOG2));

  const int col_offset = (mv.col >= 0) ? (mv.col >> (4 + MI_SIZE_LOG2))
                                       : -((-mv.col) >> (4 + MI_SIZE_LOG2));

  int row = (sign_bias == 1) ? blk_row - row_offset : blk_row + row_offset;
  int col = (sign_bias == 1) ? blk_col - col_offset : blk_col + col_offset;

  if (row < 0 || row >= (cm->mi_rows >> 1) || col < 0 ||
      col >= (cm->mi_cols >> 1))
    return 0;

  if (row < base_blk_row - (MAX_OFFSET_HEIGHT >> 3) ||
      row >= base_blk_row + 8 + (MAX_OFFSET_HEIGHT >> 3) ||
      col < base_blk_col - (MAX_OFFSET_WIDTH >> 3) ||
      col >= base_blk_col + 8 + (MAX_OFFSET_WIDTH >> 3))
    return 0;

  *mi_r = row;
  *mi_c = col;

  return 1;
}

static int motion_field_projection(AV1_COMMON *cm, MV_REFERENCE_FRAME ref_frame,
                                   int dir,
                                   const int from_x4, const int to_x4,
                                   const int from_y4, const int to_y4) {
  TPL_MV_REF *tpl_mvs_base = cm->tpl_mvs;
  int ref_offset[TOTAL_REFS_PER_FRAME] = { 0 };
  int ref_sign[TOTAL_REFS_PER_FRAME] = { 0 };

  (void)dir;

  int ref_frame_idx = cm->frame_refs[FWD_RF_OFFSET(ref_frame)].idx;
  if (ref_frame_idx < 0) return 0;

  if (cm->buffer_pool.frame_bufs[ref_frame_idx].intra_only) return 0;

  if (cm->buffer_pool.frame_bufs[ref_frame_idx].mi_rows != cm->mi_rows ||
      cm->buffer_pool.frame_bufs[ref_frame_idx].mi_cols != cm->mi_cols)
    return 0;

  int ref_frame_index =
      cm->buffer_pool.frame_bufs[ref_frame_idx].cur_frame_offset;
  unsigned int *ref_rf_idx =
      &cm->buffer_pool.frame_bufs[ref_frame_idx].ref_frame_offset[0];
   int cur_frame_index = cm->cur_frame.cur_frame_offset;
  int ref_to_cur = get_relative_dist(cm, ref_frame_index, cur_frame_index);

  for (MV_REFERENCE_FRAME rf = LAST_FRAME; rf <= INTER_REFS_PER_FRAME; ++rf) {
    ref_offset[rf] =
        get_relative_dist(cm, ref_frame_index, ref_rf_idx[rf - LAST_FRAME]);
    // note the inverted sign
    ref_sign[rf] =
        get_relative_dist(cm, ref_rf_idx[rf - LAST_FRAME], ref_frame_index) < 0;
  }

  if (dir == 2) ref_to_cur = -ref_to_cur;

  MV_REF *mv_ref_base = cm->buffer_pool.frame_bufs[ref_frame_idx].mvs;
  const ptrdiff_t mv_stride =
    cm->buffer_pool.frame_bufs[ref_frame_idx].mv_stride;
  const int mvs_rows = (cm->mi_rows + 1) >> 1;
  const int mvs_cols = (cm->mi_cols + 1) >> 1;

  assert(from_y4 >= 0);
  const int row_start8 = from_y4 >> 1;
  const int row_end8 = imin(to_y4 >> 1, mvs_rows);
  const int col_start8 = imax((from_x4 - (MAX_OFFSET_WIDTH >> 2)) >> 1, 0);
  const int col_end8 = imin((to_x4 + (MAX_OFFSET_WIDTH >> 2)) >> 1, mvs_cols);
  for (int blk_row = row_start8; blk_row < row_end8; ++blk_row) {
    for (int blk_col = col_start8; blk_col < col_end8; ++blk_col) {
      MV_REF *mv_ref = &mv_ref_base[((blk_row << 1) + 1) * mv_stride +
                                     (blk_col << 1) + 1];
      int diridx;
      const int ref0 = mv_ref->ref_frame[0], ref1 = mv_ref->ref_frame[1];
      if (ref1 > 0 && ref_sign[ref1] &&
          abs(mv_ref->mv[1].as_mv.row) < (1 << 12) &&
          abs(mv_ref->mv[1].as_mv.col) < (1 << 12))
      {
        diridx = 1;
      } else if (ref0 > 0 && ref_sign[ref0] &&
                 abs(mv_ref->mv[0].as_mv.row) < (1 << 12) &&
                 abs(mv_ref->mv[0].as_mv.col) < (1 << 12))
      {
        diridx = 0;
      } else {
        continue;
      }
      MV fwd_mv = mv_ref->mv[diridx].as_mv;

      if (mv_ref->ref_frame[diridx] > INTRA_FRAME) {
        int_mv this_mv;
        int mi_r, mi_c;
        const int ref_frame_offset = ref_offset[mv_ref->ref_frame[diridx]];

        int pos_valid = abs(ref_frame_offset) <= MAX_FRAME_DISTANCE &&
                        ref_frame_offset > 0 &&
                        abs(ref_to_cur) <= MAX_FRAME_DISTANCE;

        if (pos_valid) {
          get_mv_projection(&this_mv.as_mv, fwd_mv, ref_to_cur,
                            ref_frame_offset);
          pos_valid = get_block_position(cm, &mi_r, &mi_c, blk_row, blk_col,
                                         this_mv.as_mv, dir >> 1);
        }

        if (pos_valid && mi_c >= (from_x4 >> 1) && mi_c < (to_x4 >> 1)) {
          int mi_offset = mi_r * (cm->mi_stride >> 1) + mi_c;

          tpl_mvs_base[mi_offset].mfmv0.as_mv.row = fwd_mv.row;
          tpl_mvs_base[mi_offset].mfmv0.as_mv.col = fwd_mv.col;
          tpl_mvs_base[mi_offset].ref_frame_offset = ref_frame_offset;
        }
      }
    }
  }

  return 1;
}

1793
static void av1_setup_motion_field(AV1_COMMON *cm) {
1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851
  if (!cm->seq_params.enable_order_hint) return;

  TPL_MV_REF *tpl_mvs_base = cm->tpl_mvs;
  int size = (((cm->mi_rows + 31) & ~31) >> 1) * (cm->mi_stride >> 1);
  for (int idx = 0; idx < size; ++idx) {
    tpl_mvs_base[idx].mfmv0.as_int = INVALID_MV;
    tpl_mvs_base[idx].ref_frame_offset = 0;
  }

  memset(cm->ref_frame_side, 0, sizeof(cm->ref_frame_side));
  RefCntBuffer *const frame_bufs = cm->buffer_pool.frame_bufs;

  const int cur_order_hint = cm->cur_frame.cur_frame_offset;
  int *const ref_buf_idx = cm->ref_buf_idx;
  int *const ref_order_hint = cm->ref_order_hint;

  for (int ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ref_frame++) {
    const int ref_idx = ref_frame - LAST_FRAME;
    const int buf_idx = cm->frame_refs[ref_idx].idx;
    int order_hint = 0;

    if (buf_idx >= 0) order_hint = frame_bufs[buf_idx].cur_frame_offset;

    ref_buf_idx[ref_idx] = buf_idx;
    ref_order_hint[ref_idx] = order_hint;

    if (get_relative_dist(cm, order_hint, cur_order_hint) > 0)
      cm->ref_frame_side[ref_frame] = 1;
    else if (order_hint == cur_order_hint)
      cm->ref_frame_side[ref_frame] = -1;
  }
}

enum BlockSize {
    BS_128x128,
    BS_128x64,
    BS_64x128,
    BS_64x64,
    BS_64x32,
    BS_64x16,
    BS_32x64,
    BS_32x32,
    BS_32x16,
    BS_32x8,
    BS_16x64,
    BS_16x32,
    BS_16x16,
    BS_16x8,
    BS_16x4,
    BS_8x32,
    BS_8x16,
    BS_8x8,
    BS_8x4,
    BS_4x16,
    BS_4x8,
    BS_4x4,
    N_BS_SIZES,
};
1852
extern const uint8_t dav1d_block_dimensions[N_BS_SIZES][4];
1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903
const uint8_t bs_to_sbtype[N_BS_SIZES] = {
    [BS_128x128] = BLOCK_128X128,
    [BS_128x64] = BLOCK_128X64,
    [BS_64x128] = BLOCK_64X128,
    [BS_64x64] = BLOCK_64X64,
    [BS_64x32] = BLOCK_64X32,
    [BS_64x16] = BLOCK_64X16,
    [BS_32x64] = BLOCK_32X64,
    [BS_32x32] = BLOCK_32X32,
    [BS_32x16] = BLOCK_32X16,
    [BS_32x8] = BLOCK_32X8,
    [BS_16x64] = BLOCK_16X64,
    [BS_16x32] = BLOCK_16X32,
    [BS_16x16] = BLOCK_16X16,
    [BS_16x8] = BLOCK_16X8,
    [BS_16x4] = BLOCK_16X4,
    [BS_8x32] = BLOCK_8X32,
    [BS_8x16] = BLOCK_8X16,
    [BS_8x8] = BLOCK_8X8,
    [BS_8x4] = BLOCK_8X4,
    [BS_4x16] = BLOCK_4X16,
    [BS_4x8] = BLOCK_4X8,
    [BS_4x4] = BLOCK_4X4,
};
const uint8_t sbtype_to_bs[BLOCK_SIZES_ALL] = {
    [BLOCK_128X128] = BS_128x128,
    [BLOCK_128X64] = BS_128x64,
    [BLOCK_64X128] = BS_64x128,
    [BLOCK_64X64] = BS_64x64,
    [BLOCK_64X32] = BS_64x32,
    [BLOCK_64X16] = BS_64x16,
    [BLOCK_32X64] = BS_32x64,
    [BLOCK_32X32] = BS_32x32,
    [BLOCK_32X16] = BS_32x16,
    [BLOCK_32X8] = BS_32x8,
    [BLOCK_16X64] = BS_16x64,
    [BLOCK_16X32] = BS_16x32,
    [BLOCK_16X16] = BS_16x16,
    [BLOCK_16X8] = BS_16x8,
    [BLOCK_16X4] = BS_16x4,
    [BLOCK_8X32] = BS_8x32,
    [BLOCK_8X16] = BS_8x16,
    [BLOCK_8X8] = BS_8x8,
    [BLOCK_8X4] = BS_8x4,
    [BLOCK_4X16] = BS_4x16,
    [BLOCK_4X8] = BS_4x8,
    [BLOCK_4X4] = BS_4x4,
};

#include <stdio.h>

1904 1905 1906 1907 1908 1909
void av1_find_ref_mvs(CANDIDATE_MV *mvstack, int *cnt, int_mv (*mvlist)[2],
                      int *ctx, int refidx_dav1d[2],
                      int w4, int h4, int bs, int bp, int by4, int bx4,
                      int tile_col_start4, int tile_col_end4,
                      int tile_row_start4, int tile_row_end4,
                      AV1_COMMON *cm);
1910 1911 1912 1913 1914 1915 1916
void av1_find_ref_mvs(CANDIDATE_MV *mvstack, int *cnt, int_mv (*mvlist)[2],
                      int *ctx, int refidx_dav1d[2],
                      int w4, int h4, int bs, int bp, int by4, int bx4,
                      int tile_col_start4, int tile_col_end4,
                      int tile_row_start4, int tile_row_end4,
                      AV1_COMMON *cm)
{
1917 1918
    const int bw4 = dav1d_block_dimensions[bs][0];
    const int bh4 = dav1d_block_dimensions[bs][1];
1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048
    int stride = cm->cur_frame.mv_stride;
    MACROBLOCKD xd = (MACROBLOCKD) {
        .n8_w = bw4,
        .n8_h = bh4,
        .mi_stride = stride,
        .up_available = by4 > tile_row_start4,
        .left_available = bx4 > tile_col_start4,
        .tile = {
            .mi_col_end = AOMMIN(w4, tile_col_end4),
            .mi_row_end = AOMMIN(h4, tile_row_end4),
            .tg_horz_boundary = 0,
            .mi_row_start = tile_row_start4,
            .mi_col_start = tile_col_start4,
        },
        .mi = (MB_MODE_INFO *) &cm->cur_frame.mvs[by4 * stride + bx4],
        .mb_to_bottom_edge = (h4 - bh4 - by4) * 32,
        .mb_to_left_edge = -bx4 * 32,
        .mb_to_right_edge = (w4 - bw4 - bx4) * 32,
        .mb_to_top_edge = -by4 * 32,
        .is_sec_rect = 0,
        .cur_mi = {
            .partition = bp,
        },
    };
    xd.mi->sb_type = bs_to_sbtype[bs];
    if (xd.n8_w < xd.n8_h) {
        // Only mark is_sec_rect as 1 for the last block.
        // For PARTITION_VERT_4, it would be (0, 0, 0, 1);
        // For other partitions, it would be (0, 1).
        if (!((bx4 + xd.n8_w) & (xd.n8_h - 1))) xd.is_sec_rect = 1;
    }

    if (xd.n8_w > xd.n8_h)
        if (by4 & (xd.n8_w - 1)) xd.is_sec_rect = 1;

    MV_REFERENCE_FRAME rf[2] = { refidx_dav1d[0] + 1, refidx_dav1d[1] + 1 };
    const int refidx = av1_ref_frame_type(rf);
    int16_t single_context[MODE_CTX_REF_FRAMES];
    uint8_t mv_cnt[MODE_CTX_REF_FRAMES] = { 0 };
    CANDIDATE_MV mv_stack[MODE_CTX_REF_FRAMES][MAX_REF_MV_STACK_SIZE];
    memset(mv_stack, 0, sizeof(mv_stack));
    int_mv mv_list[MODE_CTX_REF_FRAMES][MAX_MV_REF_CANDIDATES] = { { { 0 } } };
    int_mv gmvs[MODE_CTX_REF_FRAMES];
    av1_find_mv_refs(cm, &xd, xd.mi, refidx, mv_cnt,
                     mv_stack, mv_list, gmvs, by4, bx4,
                     single_context);
    for (int i = 0; i < mv_cnt[refidx]; i++)
        mvstack[i] = mv_stack[refidx][i];
    *cnt = mv_cnt[refidx];

    mvlist[0][0] = mv_list[refidx_dav1d[0] + 1][0];
    mvlist[0][1] = mv_list[refidx_dav1d[0] + 1][1];
    if (refidx_dav1d[1] != -1) {
        mvlist[1][0] = mv_list[refidx_dav1d[1] + 1][0];
        mvlist[1][1] = mv_list[refidx_dav1d[1] + 1][1];
    }

    if (ctx) {
        if (refidx_dav1d[1] == -1)
            *ctx = single_context[refidx_dav1d[0] + 1];
        else
            *ctx = av1_mode_context_analyzer(single_context, rf);
    }

    if (0 && bx4 == 38 && by4 == 15 && cm->frame_offset == 3 &&
        refidx_dav1d[1] == -1 && refidx_dav1d[0] == 4 &&
        bw4 == 1 && bh4 == 1 && bp == 3)
    {
        MV_REF *l = bx4 ? &cm->cur_frame.mvs[by4*stride+bx4-1] : NULL;
        MV_REF *a = by4 ? &cm->cur_frame.mvs[by4*stride+bx4-stride] : NULL;
        printf("Input: left=[0]y:%d,x:%d,r:%d,[1]y:%d,x:%d,r:%d,mode=%d, "
               "above=[0]y:%d,x:%d,r:%d,[1]y:%d,x:%d,r:%d,mode=%d, "
               "temp=y:%d,x:%d,r:%d [use_ref=%d]\n",
               l ? l->mv[0].as_mv.row : -1,
               l ? l->mv[0].as_mv.col : -1,
               l ? l->ref_frame[0]: -1,
               l ? l->mv[1].as_mv.row : -1,
               l ? l->mv[1].as_mv.col : -1,
               l ? l->ref_frame[1]: -1,
               l ? l->mode : -1,
               a ? a->mv[0].as_mv.row: -1,
               a ? a->mv[0].as_mv.col : -1,
               a ? a->ref_frame[0] : -1,
               a ? a->mv[1].as_mv.row: -1,
               a ? a->mv[1].as_mv.col : -1,
               a ? a->ref_frame[1] : -1,
               a ? a->mode : -1,
               cm->tpl_mvs[(by4 >> 1) * (cm->mi_stride >> 1) + (bx4 >> 1)].mfmv0.as_mv.row,
               cm->tpl_mvs[(by4 >> 1) * (cm->mi_stride >> 1) + (bx4 >> 1)].mfmv0.as_mv.col,
               cm->tpl_mvs[(by4 >> 1) * (cm->mi_stride >> 1) +
                           (bx4 >> 1)].ref_frame_offset,
               cm->allow_ref_frame_mvs);
        printf("Edges: l=%d,t=%d,r=%d,b=%d,w=%d,h=%d,border=%d\n",
               xd.mb_to_left_edge,
               xd.mb_to_top_edge,
               xd.mb_to_right_edge,
               xd.mb_to_bottom_edge,
               xd.n8_w << MI_SIZE_LOG2,
               xd.n8_h << MI_SIZE_LOG2,
               MV_BORDER);
        printf("bp=%d, x=%d, y=%d, refs=%d/%d, n_mvs: %d, "
               "first mv: y=%d,x=%d | y=%d,x=%d, "
               "first comp mv: y=%d,x=%d | y=%d,x=%d, "
               "second mv: y=%d, x=%d | y=%d, x=%d, "
               "second comp mv: y=%d, x=%d | y=%d, x=%d, "
               "third mv: y=%d, x=%d, "
               "ctx=%d\n",
               bp, bx4, by4, refidx_dav1d[0], refidx_dav1d[1], mv_cnt[refidx],
               mv_stack[refidx][0].this_mv.as_mv.row,
               mv_stack[refidx][0].this_mv.as_mv.col,
               mv_list[refidx_dav1d[0] + 1][0].as_mv.row,
               mv_list[refidx_dav1d[0] + 1][0].as_mv.col,
               mv_stack[refidx][0].comp_mv.as_mv.row,
               mv_stack[refidx][0].comp_mv.as_mv.col,
               mv_list[refidx_dav1d[1] + 1][0].as_mv.row,
               mv_list[refidx_dav1d[1] + 1][0].as_mv.col,
               mv_stack[refidx][1].this_mv.as_mv.row,
               mv_stack[refidx][1].this_mv.as_mv.col,
               mv_list[refidx_dav1d[0] + 1][1].as_mv.row,
               mv_list[refidx_dav1d[0] + 1][1].as_mv.col,
               mv_stack[refidx][1].comp_mv.as_mv.row,
               mv_stack[refidx][1].comp_mv.as_mv.col,
               mv_list[refidx_dav1d[1] + 1][1].as_mv.row,
               mv_list[refidx_dav1d[1] + 1][1].as_mv.col,
               mv_stack[refidx][2].this_mv.as_mv.row,
               mv_stack[refidx][2].this_mv.as_mv.col,
               *ctx);
    }
}

2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062
int av1_init_ref_mv_common(AV1_COMMON *cm,
                           const int w8, const int h8,
                           const ptrdiff_t stride,
                           const int allow_sb128,
                           MV_REF *cur,
                           MV_REF *ref_mvs[7],
                           const unsigned cur_poc,
                           const unsigned ref_poc[7],
                           const unsigned ref_ref_poc[7][7],
                           const WarpedMotionParams gmv[7],
                           const int allow_hp,
                           const int force_int_mv,
                           const int allow_ref_frame_mvs,
                           const int order_hint);
2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076
int av1_init_ref_mv_common(AV1_COMMON *cm,
                           const int w8, const int h8,
                           const ptrdiff_t stride,
                           const int allow_sb128,
                           MV_REF *cur,
                           MV_REF *ref_mvs[7],
                           const unsigned cur_poc,
                           const unsigned ref_poc[7],
                           const unsigned ref_ref_poc[7][7],
                           const WarpedMotionParams gmv[7],
                           const int allow_hp,
                           const int force_int_mv,
                           const int allow_ref_frame_mvs,
                           const int order_hint)
2077 2078 2079 2080 2081
{
    if (cm->mi_cols != (w8 << 1) || cm->mi_rows != (h8 << 1)) {
        const int align_h = (h8 + 15) & ~15;
        if (cm->tpl_mvs) free(cm->tpl_mvs);
        cm->tpl_mvs = malloc(sizeof(*cm->tpl_mvs) * (stride >> 1) * align_h);
2082
        if (!cm->tpl_mvs) return -ENOMEM;
2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123
        for (int i = 0; i < 7; i++)
            cm->frame_refs[i].idx = i;
        cm->mi_cols = w8 << 1;
        cm->mi_rows = h8 << 1;
        cm->mi_stride = stride;
        for (int i = 0; i < 7; i++) {
            cm->buffer_pool.frame_bufs[i].mi_rows = cm->mi_rows;
            cm->buffer_pool.frame_bufs[i].mi_cols = cm->mi_cols;
            cm->buffer_pool.frame_bufs[i].mv_stride = stride;
        }
        cm->cur_frame.mv_stride = stride;
    }

    cm->allow_high_precision_mv = allow_hp;
    cm->seq_params.sb_size = allow_sb128 ? BLOCK_128X128 : BLOCK_64X64;

    cm->seq_params.enable_order_hint = !!order_hint;
    cm->seq_params.order_hint_bits_minus1 = order_hint - 1;
    // FIXME get these from the sequence/frame headers instead of hardcoding
    cm->frame_parallel_decode = 0;
    cm->cur_frame_force_integer_mv = force_int_mv;

    memcpy(&cm->global_motion[1], gmv, sizeof(*gmv) * 7);

    cm->frame_offset = cur_poc;
    cm->allow_ref_frame_mvs = allow_ref_frame_mvs;
    cm->cur_frame.mvs = cur;
    for (int i = 0; i < 7; i++) {
        cm->buffer_pool.frame_bufs[i].mvs = ref_mvs[i];
        cm->buffer_pool.frame_bufs[i].intra_only = ref_mvs[i] == NULL;
        cm->buffer_pool.frame_bufs[i].cur_frame_offset = ref_poc[i];
        for (int j = 0; j < 7; j++)
            cm->buffer_pool.frame_bufs[i].ref_frame_offset[j] =
                ref_ref_poc[i][j];
    }
    av1_setup_frame_buf_refs(cm);
    for (int i = 0; i < 7; i++) {
        const int ref_poc = cm->buffer_pool.frame_bufs[i].cur_frame_offset;
        cm->ref_frame_sign_bias[1 + i] = get_relative_dist(cm, ref_poc, cur_poc) > 0;
    }
    av1_setup_motion_field(cm);
2124 2125

    return 0;
2126 2127
}

2128 2129 2130
void av1_init_ref_mv_tile_row(AV1_COMMON *cm,
                              int tile_col_start4, int tile_col_end4,
                              int row_start4, int row_end4);
2131 2132 2133 2134
void av1_init_ref_mv_tile_row(AV1_COMMON *cm,
                              int tile_col_start4, int tile_col_end4,
                              int row_start4, int row_end4)
{
2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179
  RefCntBuffer *const frame_bufs = cm->buffer_pool.frame_bufs;
  const int cur_order_hint = cm->cur_frame.cur_frame_offset;
  int *const ref_buf_idx = cm->ref_buf_idx;
  int *const ref_order_hint = cm->ref_order_hint;

  int ref_stamp = MFMV_STACK_SIZE - 1;

  if (ref_buf_idx[LAST_FRAME - LAST_FRAME] >= 0) {
    const int alt_of_lst_order_hint =
        frame_bufs[ref_buf_idx[LAST_FRAME - LAST_FRAME]]
            .ref_frame_offset[ALTREF_FRAME - LAST_FRAME];

    const int is_lst_overlay =
        (alt_of_lst_order_hint == ref_order_hint[GOLDEN_FRAME - LAST_FRAME]);
      if (!is_lst_overlay) motion_field_projection(cm, LAST_FRAME, 2,
                                                   tile_col_start4, tile_col_end4,
                                                   row_start4, row_end4);
    --ref_stamp;
  }

  if (get_relative_dist(cm, ref_order_hint[BWDREF_FRAME - LAST_FRAME],
                        cur_order_hint) > 0) {
      if (motion_field_projection(cm, BWDREF_FRAME, 0,
                                  tile_col_start4, tile_col_end4,
                                  row_start4, row_end4)) --ref_stamp;
  }

  if (get_relative_dist(cm, ref_order_hint[ALTREF2_FRAME - LAST_FRAME],
                        cur_order_hint) > 0) {
      if (motion_field_projection(cm, ALTREF2_FRAME, 0,
                                  tile_col_start4, tile_col_end4,
                                  row_start4, row_end4)) --ref_stamp;
  }

  if (get_relative_dist(cm, ref_order_hint[ALTREF_FRAME - LAST_FRAME],
                        cur_order_hint) > 0 &&
      ref_stamp >= 0)
      if (motion_field_projection(cm, ALTREF_FRAME, 0,
                                  tile_col_start4, tile_col_end4,
                                  row_start4, row_end4)) --ref_stamp;

  if (ref_stamp >= 0 && ref_buf_idx[LAST2_FRAME - LAST_FRAME] >= 0)
      if (motion_field_projection(cm, LAST2_FRAME, 2,
                                  tile_col_start4, tile_col_end4,
                                  row_start4, row_end4)) --ref_stamp;
2180 2181
}

2182
AV1_COMMON *av1_alloc_ref_mv_common(void);
2183 2184
AV1_COMMON *av1_alloc_ref_mv_common(void) {
    AV1_COMMON *cm = malloc(sizeof(*cm));
2185
    if (!cm) return NULL;
2186 2187 2188 2189
    memset(cm, 0, sizeof(*cm));
    return cm;
}

2190
void av1_free_ref_mv_common(AV1_COMMON *cm);
2191 2192 2193 2194
void av1_free_ref_mv_common(AV1_COMMON *cm) {
    if (cm->tpl_mvs) free(cm->tpl_mvs);
    free(cm);
}