av1_loopfilter.c 137 KB
Newer Older
Jingning Han's avatar
Jingning Han committed
1
/*
Yaowu Xu's avatar
Yaowu Xu committed
2
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
Jingning Han's avatar
Jingning Han committed
3
 *
Yaowu Xu's avatar
Yaowu Xu committed
4 5 6 7 8 9
 * 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.
Jingning Han's avatar
Jingning Han committed
10 11
 */

Debargha Mukherjee's avatar
Debargha Mukherjee committed
12 13
#include <math.h>

Yaowu Xu's avatar
Yaowu Xu committed
14 15 16 17
#include "./aom_config.h"
#include "./aom_dsp_rtcd.h"
#include "aom_dsp/aom_dsp_common.h"
#include "aom_mem/aom_mem.h"
18
#include "aom_ports/mem.h"
19 20 21
#include "av1/common/av1_loopfilter.h"
#include "av1/common/onyxc_int.h"
#include "av1/common/reconinter.h"
22
#include "av1/common/seg_common.h"
Jingning Han's avatar
Jingning Han committed
23

24 25 26 27 28 29
#if CONFIG_LOOPFILTER_LEVEL
static const SEG_LVL_FEATURES seg_lvl_lf_lut[MAX_MB_PLANE][2] = {
  { SEG_LVL_ALT_LF_Y_V, SEG_LVL_ALT_LF_Y_H },
  { SEG_LVL_ALT_LF_U, SEG_LVL_ALT_LF_U },
  { SEG_LVL_ALT_LF_V, SEG_LVL_ALT_LF_V }
};
30 31 32 33 34 35

#if CONFIG_EXT_DELTA_Q
static const int delta_lf_id_lut[MAX_MB_PLANE][2] = {
  { 0, 1 }, { 2, 2 }, { 3, 3 }
};
#endif  // CONFIG_EXT_DELTA_Q
36 37
#endif  // CONFIG_LOOPFILTER_LEVEL

Cheng Chen's avatar
Cheng Chen committed
38 39 40 41 42 43 44 45 46 47 48 49 50 51 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
#if CONFIG_LPF_DIRECT
static void pick_filter_pixel_left(uint8_t *const src, uint8_t *const line,
                                   int *const orig_pos, int length, int row,
                                   int col, int width, int height, int pitch,
                                   int pivot, int direct) {
  int i;
  int pos = row * pitch + col;

  for (i = 0; i < length; ++i) {
    int dy = 0;
    switch (direct) {
      case VERT_HORZ: dy = 0; break;
      case DEGREE_45: dy = 1; break;
      case DEGREE_135: dy = -1; break;
    }
    col -= 1;
    row += dy;
    if (col >= 0 && col < width && row >= 0 && row < height) {
      pos = row * pitch + col;
      line[pivot - 1 - i] = src[pos];
      orig_pos[pivot - 1 - i] = pos;
    }
  }
}

static void pick_filter_pixel_right(uint8_t *const src, uint8_t *const line,
                                    int *const orig_pos, int length, int row,
                                    int col, int width, int height, int pitch,
                                    int pivot, int direct) {
  int i;
  int pos = row * pitch + col;

  line[pivot] = src[pos];
  orig_pos[pivot] = pos;

  for (i = 1; i < length; ++i) {
    int dy = 0;
    switch (direct) {
      case VERT_HORZ: dy = 0; break;
      case DEGREE_45: dy = -1; break;
      case DEGREE_135: dy = 1; break;
    }
    col += 1;
    row += dy;
    if (col >= 0 && col < width && row >= 0 && row < height) {
      pos = row * pitch + col;
      line[pivot + i] = src[pos];
      orig_pos[pivot + i] = pos;
    }
  }
}

static void pick_filter_pixel_above(uint8_t *const src, uint8_t *const line,
                                    int *const orig_pos, int length, int row,
                                    int col, int width, int height, int pitch,
                                    int pivot, int direct) {
  int i;
  int pos = row * pitch + col;

  for (i = 0; i < length; ++i) {
    int dx = 0;
    switch (direct) {
      case VERT_HORZ: dx = 0; break;
      case DEGREE_45: dx = 1; break;
      case DEGREE_135: dx = -1; break;
    }
    col += dx;
    row -= 1;
    if (col >= 0 && col < width && row >= 0 && row < height) {
      pos = row * pitch + col;
      line[pivot - 1 - i] = src[pos];
      orig_pos[pivot - 1 - i] = pos;
    }
  }
}

static void pick_filter_pixel_bot(uint8_t *const src, uint8_t *const line,
                                  int *const orig_pos, int length, int row,
                                  int col, int width, int height, int pitch,
                                  int pivot, int direct) {
  int i;
  int pos = row * pitch + col;

  line[pivot] = src[pos];
  orig_pos[pivot] = pos;

  for (i = 1; i < length; ++i) {
    int dx = 0;
    switch (direct) {
      case VERT_HORZ: dx = 0; break;
      case DEGREE_45: dx = -1; break;
      case DEGREE_135: dx = 1; break;
    }
    col += dx;
    row += 1;
    if (col >= 0 && col < width && row >= 0 && row < height) {
      pos = row * pitch + col;
      line[pivot + i] = src[pos];
      orig_pos[pivot + i] = pos;
    }
  }
}

static void pick_filter_block_vert(uint8_t *const src, uint8_t *const block,
                                   int *const orig_pos, int length, int row,
                                   int col, int width, int height, int pitch,
                                   int pivot, int line_length, int unit,
                                   int direct) {
  int i;
  for (i = 0; i < 8 * unit; ++i) {
    pick_filter_pixel_left(src, block + i * line_length,
                           orig_pos + i * line_length, length, row + i, col,
                           width, height, pitch, pivot, direct);
    pick_filter_pixel_right(src, block + i * line_length,
                            orig_pos + i * line_length, length, row + i, col,
                            width, height, pitch, pivot, direct);
  }
}

static void pick_filter_block_horz(uint8_t *const src, uint8_t *const block,
                                   int *const orig_pos, int length, int row,
                                   int col, int width, int height, int pitch,
                                   int pivot, int line_length, int unit,
                                   int direct) {
  int i, j;
  int num = 8 * unit;
  for (i = 0; i < num; ++i) {
    pick_filter_pixel_above(src, block + i * line_length,
                            orig_pos + i * line_length, length, row, col + i,
                            width, height, pitch, pivot, direct);
    pick_filter_pixel_bot(src, block + i * line_length,
                          orig_pos + i * line_length, length, row, col + i,
                          width, height, pitch, pivot, direct);
  }

  // rearrange block
  // TODO(chengchen): make it in-place or a stand alone function
  uint8_t tmp_block[256];
  int tmp_pos[256];
  for (i = 0; i < 256; ++i) {
    tmp_block[i] = 0;
    tmp_pos[i] = -1;
  }
  for (i = 0; i < num; ++i) {
    for (j = 0; j < line_length; ++j) {
      tmp_block[j * line_length + i] = block[i * line_length + j];
      tmp_pos[j * line_length + i] = orig_pos[i * line_length + j];
    }
  }
  for (i = 0; i < 256; ++i) {
    block[i] = tmp_block[i];
    orig_pos[i] = tmp_pos[i];
  }
}

static int compute_block_grad(uint8_t *const src, int length, int row, int col,
                              int width, int height, int pitch, int unit,
                              int vert_or_horz, int direct) {
  int i, j;
  int r0, c0, pos0, r1 = 0, c1 = 0, pos1;
  int sum_grad = 0;
  for (i = 0; i < 8 * unit; ++i) {
    // vert_or_horz: 0 vertical edge, 1 horizontal edge
    r0 = vert_or_horz ? row : row + i;
    c0 = vert_or_horz ? col + i : col;
    pos0 = r0 * pitch + c0;

    for (j = 0; j < length; ++j) {
      if (vert_or_horz == 0) {
        switch (direct) {
          case VERT_HORZ: r1 = r0; break;
          case DEGREE_45: r1 = r0 + 1; break;
          case DEGREE_135: r1 = r0 - 1; break;
        }
        c1 = c0 - 1;
      } else {
        r1 = r0 - 1;
        switch (direct) {
          case VERT_HORZ: c1 = c0; break;
          case DEGREE_45: c1 = c0 + 1; break;
          case DEGREE_135: c1 = c0 - 1; break;
        }
      }
      pos1 = r1 * pitch + c1;

      if (r0 >= 0 && r0 < height && c0 >= 0 && c0 < width && r1 >= 0 &&
          r1 < height && c1 >= 0 && c1 < width) {
        sum_grad += abs(src[pos1] - src[pos0]);
      } else {
        sum_grad += 255;  // penalize unreachable boundary
      }
      r0 = r1;
      c0 = c1;
      pos0 = pos1;
    }

    r0 = vert_or_horz ? row : row + i;
    c0 = vert_or_horz ? col + i : col;
    pos0 = r0 * pitch + c0;

    for (j = 0; j < length - 1; ++j) {
      if (vert_or_horz == 0) {
        switch (direct) {
          case VERT_HORZ: r1 = r0; break;
          case DEGREE_45: r1 = r0 - 1; break;
          case DEGREE_135: r1 = r0 + 1; break;
        }
        c1 = c0 + 1;
      } else {
        r1 = r0 + 1;
        switch (direct) {
          case VERT_HORZ: c1 = c0; break;
          case DEGREE_45: c1 = c0 - 1; break;
          case DEGREE_135: c1 = c0 + 1; break;
        }
      }
      pos1 = r1 * pitch + c1;

      if (r0 >= 0 && r0 < height && c0 >= 0 && c0 < width && r1 >= 0 &&
          r1 < height && c1 >= 0 && c1 < width) {
        sum_grad += abs(src[pos1] - src[pos0]);
      } else {
        sum_grad += 255;  // penalize unreachable boundary
      }
      r0 = r1;
      c0 = c1;
      pos0 = pos1;
    }
  }

  return sum_grad;
}

static int pick_min_grad_direct(uint8_t *const src, int length, int row,
                                int col, int width, int height, int pitch,
                                int unit, int vert_or_horz) {
  int direct = VERT_HORZ;
  int min_grad = INT_MAX, sum_grad = 0;

  int degree;
  for (degree = 0; degree < FILTER_DEGREES; ++degree) {
    // compute abs gradient along each line for the filter block
    sum_grad = compute_block_grad(src, length, row, col, width, height, pitch,
                                  unit, vert_or_horz, degree);
    if (sum_grad < min_grad) {
      min_grad = sum_grad;
      direct = degree;
    }
  }

  return direct;
}
#endif  // CONFIG_LPF_DIRECT

292
#define PARALLEL_DEBLOCKING_15TAPLUMAONLY 1
293
#define PARALLEL_DEBLOCKING_DISABLE_15TAP 0
Ola Hugosson's avatar
Ola Hugosson committed
294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316
#if CONFIG_DEBLOCK_13TAP
#define PARALLEL_DEBLOCKING_5_TAP_CHROMA 1
#else
#define PARALLEL_DEBLOCKING_5_TAP_CHROMA 0
#endif

#if PARALLEL_DEBLOCKING_5_TAP_CHROMA
extern void aom_lpf_vertical_6_c(uint8_t *s, int pitch, const uint8_t *blimit,
                                 const uint8_t *limit, const uint8_t *thresh);

extern void aom_lpf_horizontal_6_c(uint8_t *s, int p, const uint8_t *blimit,
                                   const uint8_t *limit, const uint8_t *thresh);

extern void aom_highbd_lpf_horizontal_6_c(uint16_t *s, int p,
                                          const uint8_t *blimit,
                                          const uint8_t *limit,
                                          const uint8_t *thresh, int bd);

extern void aom_highbd_lpf_vertical_6_c(uint16_t *s, int pitch,
                                        const uint8_t *blimit,
                                        const uint8_t *limit,
                                        const uint8_t *thresh, int bd);
#endif
317

Jingning Han's avatar
Jingning Han committed
318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334
// 64 bit masks for left transform size. Each 1 represents a position where
// we should apply a loop filter across the left border of an 8x8 block
// boundary.
//
// In the case of TX_16X16->  ( in low order byte first we end up with
// a mask that looks like this
//
//    10101010
//    10101010
//    10101010
//    10101010
//    10101010
//    10101010
//    10101010
//    10101010
//
// A loopfilter should be applied to every other 8x8 horizontally.
clang-format's avatar
clang-format committed
335
static const uint64_t left_64x64_txform_mask[TX_SIZES] = {
336
#if CONFIG_CHROMA_2X2
337 338
  0xffffffffffffffffULL,  // TX_2X2
#endif
Jingning Han's avatar
Jingning Han committed
339 340 341 342
  0xffffffffffffffffULL,  // TX_4X4
  0xffffffffffffffffULL,  // TX_8x8
  0x5555555555555555ULL,  // TX_16x16
  0x1111111111111111ULL,  // TX_32x32
343 344 345
#if CONFIG_TX64X64
  0x0101010101010101ULL,  // TX_64x64
#endif                    // CONFIG_TX64X64
Jingning Han's avatar
Jingning Han committed
346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364
};

// 64 bit masks for above transform size. Each 1 represents a position where
// we should apply a loop filter across the top border of an 8x8 block
// boundary.
//
// In the case of TX_32x32 ->  ( in low order byte first we end up with
// a mask that looks like this
//
//    11111111
//    00000000
//    00000000
//    00000000
//    11111111
//    00000000
//    00000000
//    00000000
//
// A loopfilter should be applied to every other 4 the row vertically.
clang-format's avatar
clang-format committed
365
static const uint64_t above_64x64_txform_mask[TX_SIZES] = {
366
#if CONFIG_CHROMA_2X2
367 368
  0xffffffffffffffffULL,  // TX_4X4
#endif
Jingning Han's avatar
Jingning Han committed
369 370 371 372
  0xffffffffffffffffULL,  // TX_4X4
  0xffffffffffffffffULL,  // TX_8x8
  0x00ff00ff00ff00ffULL,  // TX_16x16
  0x000000ff000000ffULL,  // TX_32x32
373 374 375
#if CONFIG_TX64X64
  0x00000000000000ffULL,  // TX_64x64
#endif                    // CONFIG_TX64X64
Jingning Han's avatar
Jingning Han committed
376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392
};

// 64 bit masks for prediction sizes (left). Each 1 represents a position
// where left border of an 8x8 block. These are aligned to the right most
// appropriate bit, and then shifted into place.
//
// In the case of TX_16x32 ->  ( low order byte first ) we end up with
// a mask that looks like this :
//
//  10000000
//  10000000
//  10000000
//  10000000
//  00000000
//  00000000
//  00000000
//  00000000
393
static const uint64_t left_prediction_mask[BLOCK_SIZES_ALL] = {
394
#if CONFIG_CHROMA_2X2 || CONFIG_CHROMA_SUB8X8
Jingning Han's avatar
Jingning Han committed
395 396 397 398
  0x0000000000000001ULL,  // BLOCK_2X2,
  0x0000000000000001ULL,  // BLOCK_2X4,
  0x0000000000000001ULL,  // BLOCK_4X2,
#endif
Jingning Han's avatar
Jingning Han committed
399 400 401 402 403 404 405 406 407 408 409 410
  0x0000000000000001ULL,  // BLOCK_4X4,
  0x0000000000000001ULL,  // BLOCK_4X8,
  0x0000000000000001ULL,  // BLOCK_8X4,
  0x0000000000000001ULL,  // BLOCK_8X8,
  0x0000000000000101ULL,  // BLOCK_8X16,
  0x0000000000000001ULL,  // BLOCK_16X8,
  0x0000000000000101ULL,  // BLOCK_16X16,
  0x0000000001010101ULL,  // BLOCK_16X32,
  0x0000000000000101ULL,  // BLOCK_32X16,
  0x0000000001010101ULL,  // BLOCK_32X32,
  0x0101010101010101ULL,  // BLOCK_32X64,
  0x0000000001010101ULL,  // BLOCK_64X32,
411 412 413 414
  0x0101010101010101ULL,  // BLOCK_64X64,
  0x0000000000000101ULL,  // BLOCK_4X16,
  0x0000000000000001ULL,  // BLOCK_16X4,
  0x0000000001010101ULL,  // BLOCK_8X32,
415 416 417
  0x0000000000000001ULL,  // BLOCK_32X8,
  0x0101010101010101ULL,  // BLOCK_16X64,
  0x0000000000000101ULL,  // BLOCK_64X16
Jingning Han's avatar
Jingning Han committed
418 419 420
};

// 64 bit mask to shift and set for each prediction size.
421
static const uint64_t above_prediction_mask[BLOCK_SIZES_ALL] = {
422
#if CONFIG_CHROMA_2X2 || CONFIG_CHROMA_SUB8X8
Jingning Han's avatar
Jingning Han committed
423 424 425 426
  0x0000000000000001ULL,  // BLOCK_2X2
  0x0000000000000001ULL,  // BLOCK_2X4
  0x0000000000000001ULL,  // BLOCK_4X2
#endif
Jingning Han's avatar
Jingning Han committed
427 428 429 430 431 432 433 434 435 436 437 438
  0x0000000000000001ULL,  // BLOCK_4X4
  0x0000000000000001ULL,  // BLOCK_4X8
  0x0000000000000001ULL,  // BLOCK_8X4
  0x0000000000000001ULL,  // BLOCK_8X8
  0x0000000000000001ULL,  // BLOCK_8X16,
  0x0000000000000003ULL,  // BLOCK_16X8
  0x0000000000000003ULL,  // BLOCK_16X16
  0x0000000000000003ULL,  // BLOCK_16X32,
  0x000000000000000fULL,  // BLOCK_32X16,
  0x000000000000000fULL,  // BLOCK_32X32,
  0x000000000000000fULL,  // BLOCK_32X64,
  0x00000000000000ffULL,  // BLOCK_64X32,
439 440 441 442
  0x00000000000000ffULL,  // BLOCK_64X64,
  0x0000000000000001ULL,  // BLOCK_4X16,
  0x0000000000000003ULL,  // BLOCK_16X4,
  0x0000000000000001ULL,  // BLOCK_8X32,
443 444 445
  0x000000000000000fULL,  // BLOCK_32X8,
  0x0000000000000003ULL,  // BLOCK_16X64,
  0x00000000000000ffULL,  // BLOCK_64X16
Jingning Han's avatar
Jingning Han committed
446 447
};
// 64 bit mask to shift and set for each prediction size. A bit is set for
448
// each 8x8 block that would be in the top left most block of the given block
Jingning Han's avatar
Jingning Han committed
449
// size in the 64x64 block.
450
static const uint64_t size_mask[BLOCK_SIZES_ALL] = {
451
#if CONFIG_CHROMA_2X2 || CONFIG_CHROMA_SUB8X8
Jingning Han's avatar
Jingning Han committed
452 453 454 455
  0x0000000000000001ULL,  // BLOCK_2X2
  0x0000000000000001ULL,  // BLOCK_2X4
  0x0000000000000001ULL,  // BLOCK_4X2
#endif
Jingning Han's avatar
Jingning Han committed
456 457 458 459 460 461 462 463 464 465 466 467
  0x0000000000000001ULL,  // BLOCK_4X4
  0x0000000000000001ULL,  // BLOCK_4X8
  0x0000000000000001ULL,  // BLOCK_8X4
  0x0000000000000001ULL,  // BLOCK_8X8
  0x0000000000000101ULL,  // BLOCK_8X16,
  0x0000000000000003ULL,  // BLOCK_16X8
  0x0000000000000303ULL,  // BLOCK_16X16
  0x0000000003030303ULL,  // BLOCK_16X32,
  0x0000000000000f0fULL,  // BLOCK_32X16,
  0x000000000f0f0f0fULL,  // BLOCK_32X32,
  0x0f0f0f0f0f0f0f0fULL,  // BLOCK_32X64,
  0x00000000ffffffffULL,  // BLOCK_64X32,
468 469 470 471
  0xffffffffffffffffULL,  // BLOCK_64X64,
  0x0000000000000101ULL,  // BLOCK_4X16,
  0x0000000000000003ULL,  // BLOCK_16X4,
  0x0000000001010101ULL,  // BLOCK_8X32,
472 473 474
  0x000000000000000fULL,  // BLOCK_32X8,
  0x0303030303030303ULL,  // BLOCK_16X64,
  0x000000000000ffffULL,  // BLOCK_64X16
Jingning Han's avatar
Jingning Han committed
475 476
};

477
// These are used for masking the left and above 32x32 borders.
clang-format's avatar
clang-format committed
478
static const uint64_t left_border = 0x1111111111111111ULL;
Jingning Han's avatar
Jingning Han committed
479 480 481
static const uint64_t above_border = 0x000000ff000000ffULL;

// 16 bit masks for uv transform sizes.
clang-format's avatar
clang-format committed
482
static const uint16_t left_64x64_txform_mask_uv[TX_SIZES] = {
483
#if CONFIG_CHROMA_2X2
484 485
  0xffff,  // TX_2X2
#endif
Jingning Han's avatar
Jingning Han committed
486 487 488 489
  0xffff,  // TX_4X4
  0xffff,  // TX_8x8
  0x5555,  // TX_16x16
  0x1111,  // TX_32x32
490 491 492
#if CONFIG_TX64X64
  0x0101,  // TX_64x64, never used
#endif     // CONFIG_TX64X64
Jingning Han's avatar
Jingning Han committed
493 494
};

clang-format's avatar
clang-format committed
495
static const uint16_t above_64x64_txform_mask_uv[TX_SIZES] = {
496
#if CONFIG_CHROMA_2X2
497 498
  0xffff,  // TX_2X2
#endif
Jingning Han's avatar
Jingning Han committed
499 500 501 502
  0xffff,  // TX_4X4
  0xffff,  // TX_8x8
  0x0f0f,  // TX_16x16
  0x000f,  // TX_32x32
503 504 505
#if CONFIG_TX64X64
  0x0003,  // TX_64x64, never used
#endif     // CONFIG_TX64X64
Jingning Han's avatar
Jingning Han committed
506 507 508
};

// 16 bit left mask to shift and set for each uv prediction size.
509
static const uint16_t left_prediction_mask_uv[BLOCK_SIZES_ALL] = {
510
#if CONFIG_CHROMA_2X2 || CONFIG_CHROMA_SUB8X8
Jingning Han's avatar
Jingning Han committed
511 512 513 514
  0x0001,  // BLOCK_2X2,
  0x0001,  // BLOCK_2X4,
  0x0001,  // BLOCK_4X2,
#endif
Jingning Han's avatar
Jingning Han committed
515 516 517 518 519 520 521 522 523 524 525 526
  0x0001,  // BLOCK_4X4,
  0x0001,  // BLOCK_4X8,
  0x0001,  // BLOCK_8X4,
  0x0001,  // BLOCK_8X8,
  0x0001,  // BLOCK_8X16,
  0x0001,  // BLOCK_16X8,
  0x0001,  // BLOCK_16X16,
  0x0011,  // BLOCK_16X32,
  0x0001,  // BLOCK_32X16,
  0x0011,  // BLOCK_32X32,
  0x1111,  // BLOCK_32X64
  0x0011,  // BLOCK_64X32,
527 528 529 530
  0x1111,  // BLOCK_64X64,
  0x0001,  // BLOCK_4X16,
  0x0001,  // BLOCK_16X4,
  0x0011,  // BLOCK_8X32,
531 532 533
  0x0001,  // BLOCK_32X8,
  0x1111,  // BLOCK_16X64,
  0x0001,  // BLOCK_64X16,
Jingning Han's avatar
Jingning Han committed
534
};
535

Jingning Han's avatar
Jingning Han committed
536
// 16 bit above mask to shift and set for uv each prediction size.
537
static const uint16_t above_prediction_mask_uv[BLOCK_SIZES_ALL] = {
538
#if CONFIG_CHROMA_2X2 || CONFIG_CHROMA_SUB8X8
Jingning Han's avatar
Jingning Han committed
539 540 541 542
  0x0001,  // BLOCK_2X2
  0x0001,  // BLOCK_2X4
  0x0001,  // BLOCK_4X2
#endif
Jingning Han's avatar
Jingning Han committed
543 544 545 546 547 548 549 550 551 552 553 554
  0x0001,  // BLOCK_4X4
  0x0001,  // BLOCK_4X8
  0x0001,  // BLOCK_8X4
  0x0001,  // BLOCK_8X8
  0x0001,  // BLOCK_8X16,
  0x0001,  // BLOCK_16X8
  0x0001,  // BLOCK_16X16
  0x0001,  // BLOCK_16X32,
  0x0003,  // BLOCK_32X16,
  0x0003,  // BLOCK_32X32,
  0x0003,  // BLOCK_32X64,
  0x000f,  // BLOCK_64X32,
555 556 557 558
  0x000f,  // BLOCK_64X64,
  0x0001,  // BLOCK_4X16,
  0x0001,  // BLOCK_16X4,
  0x0001,  // BLOCK_8X32,
559 560 561
  0x0003,  // BLOCK_32X8,
  0x0001,  // BLOCK_16X64,
  0x000f,  // BLOCK_64X16
Jingning Han's avatar
Jingning Han committed
562 563 564
};

// 64 bit mask to shift and set for each uv prediction size
565
static const uint16_t size_mask_uv[BLOCK_SIZES_ALL] = {
566
#if CONFIG_CHROMA_2X2 || CONFIG_CHROMA_SUB8X8
Jingning Han's avatar
Jingning Han committed
567 568 569 570
  0x0001,  // BLOCK_2X2
  0x0001,  // BLOCK_2X4
  0x0001,  // BLOCK_4X2
#endif
Jingning Han's avatar
Jingning Han committed
571 572 573 574 575 576 577 578 579 580 581 582
  0x0001,  // BLOCK_4X4
  0x0001,  // BLOCK_4X8
  0x0001,  // BLOCK_8X4
  0x0001,  // BLOCK_8X8
  0x0001,  // BLOCK_8X16,
  0x0001,  // BLOCK_16X8
  0x0001,  // BLOCK_16X16
  0x0011,  // BLOCK_16X32,
  0x0003,  // BLOCK_32X16,
  0x0033,  // BLOCK_32X32,
  0x3333,  // BLOCK_32X64,
  0x00ff,  // BLOCK_64X32,
583 584 585 586
  0xffff,  // BLOCK_64X64,
  0x0001,  // BLOCK_4X16,
  0x0001,  // BLOCK_16X4,
  0x0011,  // BLOCK_8X32,
587 588 589
  0x0003,  // BLOCK_32X8,
  0x1111,  // BLOCK_16X64,
  0x000f,  // BLOCK_64X16
Jingning Han's avatar
Jingning Han committed
590
};
clang-format's avatar
clang-format committed
591
static const uint16_t left_border_uv = 0x1111;
Jingning Han's avatar
Jingning Han committed
592 593
static const uint16_t above_border_uv = 0x000f;

594
static const int mode_lf_lut[] = {
Jingning Han's avatar
Jingning Han committed
595
  0, 0, 0, 0, 0, 0, 0, 0, 0, 0,  // INTRA_MODES
596
  0,
Urvang Joshi's avatar
Urvang Joshi committed
597 598 599
#if CONFIG_SMOOTH_HV
  0, 0,
#endif         // CONFIG_SMOOTH_HV
600
  1, 1, 0, 1,  // INTER_MODES (ZEROMV == 0)
601 602 603 604 605
#if CONFIG_COMPOUND_SINGLEREF
  // 1, 1, 1, 1, 1,       // INTER_SINGLEREF_COMP_MODES
  // NOTE(zoeliu): Remove SR_NEAREST_NEWMV
  1, 1, 1, 1,             // INTER_SINGLEREF_COMP_MODES
#endif                    // CONFIG_COMPOUND_SINGLEREF
606
  1, 1, 1, 1, 1, 1, 0, 1  // INTER_COMPOUND_MODES (ZERO_ZEROMV == 0)
Jingning Han's avatar
Jingning Han committed
607 608 609 610 611 612 613 614 615 616 617 618 619 620 621
};

static void update_sharpness(loop_filter_info_n *lfi, int sharpness_lvl) {
  int lvl;

  // For each possible value for the loop filter fill out limits
  for (lvl = 0; lvl <= MAX_LOOP_FILTER; lvl++) {
    // Set loop filter parameters that control sharpness.
    int block_inside_limit = lvl >> ((sharpness_lvl > 0) + (sharpness_lvl > 4));

    if (sharpness_lvl > 0) {
      if (block_inside_limit > (9 - sharpness_lvl))
        block_inside_limit = (9 - sharpness_lvl);
    }

clang-format's avatar
clang-format committed
622
    if (block_inside_limit < 1) block_inside_limit = 1;
Jingning Han's avatar
Jingning Han committed
623 624 625 626 627 628

    memset(lfi->lfthr[lvl].lim, block_inside_limit, SIMD_WIDTH);
    memset(lfi->lfthr[lvl].mblim, (2 * (lvl + 2) + block_inside_limit),
           SIMD_WIDTH);
  }
}
Fangwen Fu's avatar
Fangwen Fu committed
629 630 631
#if CONFIG_EXT_DELTA_Q
static uint8_t get_filter_level(const AV1_COMMON *cm,
                                const loop_filter_info_n *lfi_n,
Cheng Chen's avatar
Cheng Chen committed
632
#if CONFIG_LOOPFILTER_LEVEL
633
                                const int dir_idx, int plane,
Cheng Chen's avatar
Cheng Chen committed
634 635 636
#endif
#if CONFIG_LPF_SB
                                int mi_row, int mi_col,
637
#endif
Fangwen Fu's avatar
Fangwen Fu committed
638
                                const MB_MODE_INFO *mbmi) {
639
#if CONFIG_LPF_SB
Cheng Chen's avatar
Cheng Chen committed
640
  return cm->mi[mi_row * cm->mi_stride + mi_col].mbmi.filt_lvl;
641 642
#endif

Fangwen Fu's avatar
Fangwen Fu committed
643 644
  const int segment_id = mbmi->segment_id;
  if (cm->delta_lf_present_flag) {
Cheng Chen's avatar
Cheng Chen committed
645
#if CONFIG_LOOPFILTER_LEVEL
646 647 648 649 650 651 652
    int delta_lf;
    if (cm->delta_lf_multi) {
      const int delta_lf_idx = delta_lf_id_lut[plane][dir_idx];
      delta_lf = mbmi->curr_delta_lf[delta_lf_idx];
    } else {
      delta_lf = mbmi->current_delta_lf_from_base;
    }
653
    int lvl_seg =
654
        clamp(delta_lf + cm->lf.filter_level[dir_idx], 0, MAX_LOOP_FILTER);
655
#else
Fangwen Fu's avatar
Fangwen Fu committed
656 657
    int lvl_seg = clamp(mbmi->current_delta_lf_from_base + cm->lf.filter_level,
                        0, MAX_LOOP_FILTER);
658
#endif
Fangwen Fu's avatar
Fangwen Fu committed
659
    const int scale = 1 << (lvl_seg >> 5);
660 661 662 663 664 665 666 667 668 669
#if CONFIG_LOOPFILTER_LEVEL
    assert(plane >= 0 && plane <= 2);
    const int seg_lf_feature_id = seg_lvl_lf_lut[plane][dir_idx];
    if (segfeature_active(&cm->seg, segment_id, seg_lf_feature_id)) {
      const int data = get_segdata(&cm->seg, segment_id, seg_lf_feature_id);
      lvl_seg =
          clamp(cm->seg.abs_delta == SEGMENT_ABSDATA ? data : lvl_seg + data, 0,
                MAX_LOOP_FILTER);
    }
#else
Fangwen Fu's avatar
Fangwen Fu committed
670 671 672 673 674 675
    if (segfeature_active(&cm->seg, segment_id, SEG_LVL_ALT_LF)) {
      const int data = get_segdata(&cm->seg, segment_id, SEG_LVL_ALT_LF);
      lvl_seg =
          clamp(cm->seg.abs_delta == SEGMENT_ABSDATA ? data : lvl_seg + data, 0,
                MAX_LOOP_FILTER);
    }
676
#endif  // CONFIG_LOOPFILTER_LEVEL
Jingning Han's avatar
Jingning Han committed
677

Fangwen Fu's avatar
Fangwen Fu committed
678 679 680 681 682 683 684 685
    if (cm->lf.mode_ref_delta_enabled) {
      lvl_seg += cm->lf.ref_deltas[mbmi->ref_frame[0]] * scale;
      if (mbmi->ref_frame[0] > INTRA_FRAME)
        lvl_seg += cm->lf.mode_deltas[mode_lf_lut[mbmi->mode]] * scale;
      lvl_seg = clamp(lvl_seg, 0, MAX_LOOP_FILTER);
    }
    return lvl_seg;
  } else {
Cheng Chen's avatar
Cheng Chen committed
686
#if CONFIG_LOOPFILTER_LEVEL
687 688 689
    return lfi_n
        ->lvl[segment_id][dir_idx][mbmi->ref_frame[0]][mode_lf_lut[mbmi->mode]];
#else
Fangwen Fu's avatar
Fangwen Fu committed
690
    return lfi_n->lvl[segment_id][mbmi->ref_frame[0]][mode_lf_lut[mbmi->mode]];
691
#endif
Fangwen Fu's avatar
Fangwen Fu committed
692 693 694
  }
}
#else
Jingning Han's avatar
Jingning Han committed
695 696
static uint8_t get_filter_level(const loop_filter_info_n *lfi_n,
                                const MB_MODE_INFO *mbmi) {
Cheng Chen's avatar
Cheng Chen committed
697 698 699 700
#if CONFIG_LPF_SB
  return mbmi->filt_lvl;
#endif

701 702
  const int segment_id = mbmi->segment_id;
  return lfi_n->lvl[segment_id][mbmi->ref_frame[0]][mode_lf_lut[mbmi->mode]];
Jingning Han's avatar
Jingning Han committed
703
}
Fangwen Fu's avatar
Fangwen Fu committed
704
#endif
Jingning Han's avatar
Jingning Han committed
705

Yaowu Xu's avatar
Yaowu Xu committed
706
void av1_loop_filter_init(AV1_COMMON *cm) {
707
  assert(MB_MODE_COUNT == NELEMENTS(mode_lf_lut));
Jingning Han's avatar
Jingning Han committed
708 709 710 711 712 713 714 715 716 717 718 719 720
  loop_filter_info_n *lfi = &cm->lf_info;
  struct loopfilter *lf = &cm->lf;
  int lvl;

  // init limits for given sharpness
  update_sharpness(lfi, lf->sharpness_level);
  lf->last_sharpness_level = lf->sharpness_level;

  // init hev threshold const vectors
  for (lvl = 0; lvl <= MAX_LOOP_FILTER; lvl++)
    memset(lfi->lfthr[lvl].hev_thr, (lvl >> 4), SIMD_WIDTH);
}

721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741
#if CONFIG_LPF_SB
void av1_loop_filter_sb_level_init(AV1_COMMON *cm, int mi_row, int mi_col,
                                   int lvl) {
  const int mi_row_start = AOMMAX(0, mi_row - FILT_BOUNDARY_MI_OFFSET);
  const int mi_col_start = AOMMAX(0, mi_col - FILT_BOUNDARY_MI_OFFSET);
  const int mi_row_range = mi_row - FILT_BOUNDARY_MI_OFFSET + MAX_MIB_SIZE;
  const int mi_col_range = mi_col - FILT_BOUNDARY_MI_OFFSET + MAX_MIB_SIZE;
  const int mi_row_end = AOMMIN(mi_row_range, cm->mi_rows);
  const int mi_col_end = AOMMIN(mi_col_range, cm->mi_cols);

  int row, col;
  for (row = mi_row_start; row < mi_row_end; ++row) {
    for (col = mi_col_start; col < mi_col_end; ++col) {
      // Note: can't use cm->mi_grid_visible. Because for each partition,
      // all visible pointers will point to the first of the partition.
      cm->mi[row * cm->mi_stride + col].mbmi.filt_lvl = lvl;
    }
  }
}
#endif  // CONFIG_LPF_SB

742
void av1_loop_filter_frame_init(AV1_COMMON *cm, int default_filt_lvl,
743 744 745 746 747 748
                                int default_filt_lvl_r
#if CONFIG_LOOPFILTER_LEVEL
                                ,
                                int plane
#endif
                                ) {
Jingning Han's avatar
Jingning Han committed
749 750 751 752
  int seg_id;
  // n_shift is the multiplier for lf_deltas
  // the multiplier is 1 for when filter_lvl is between 0 and 31;
  // 2 when filter_lvl is between 32 and 63
753
  int scale = 1 << (default_filt_lvl >> 5);
Jingning Han's avatar
Jingning Han committed
754 755 756 757 758 759 760 761 762 763 764
  loop_filter_info_n *const lfi = &cm->lf_info;
  struct loopfilter *const lf = &cm->lf;
  const struct segmentation *const seg = &cm->seg;

  // update limits if sharpness has changed
  if (lf->last_sharpness_level != lf->sharpness_level) {
    update_sharpness(lfi, lf->sharpness_level);
    lf->last_sharpness_level = lf->sharpness_level;
  }

  for (seg_id = 0; seg_id < MAX_SEGMENTS; seg_id++) {
765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783
    for (int dir = 0; dir < 2; ++dir) {
      int lvl_seg = (dir == 0) ? default_filt_lvl : default_filt_lvl_r;
#if CONFIG_LOOPFILTER_LEVEL
      assert(plane >= 0 && plane <= 2);
      const int seg_lf_feature_id = seg_lvl_lf_lut[plane][dir];
      if (segfeature_active(seg, seg_id, seg_lf_feature_id)) {
        const int data = get_segdata(&cm->seg, seg_id, seg_lf_feature_id);
        lvl_seg = clamp(
            seg->abs_delta == SEGMENT_ABSDATA ? data : default_filt_lvl + data,
            0, MAX_LOOP_FILTER);
      }
#else
      if (segfeature_active(seg, seg_id, SEG_LVL_ALT_LF)) {
        const int data = get_segdata(seg, seg_id, SEG_LVL_ALT_LF);
        lvl_seg = clamp(
            seg->abs_delta == SEGMENT_ABSDATA ? data : default_filt_lvl + data,
            0, MAX_LOOP_FILTER);
      }
#endif  // CONFIG_LOOPFILTER_LEVEL
Jingning Han's avatar
Jingning Han committed
784

785 786 787 788 789 790 791 792 793 794
      if (!lf->mode_ref_delta_enabled) {
// we could get rid of this if we assume that deltas are set to
// zero when not in use; encoder always uses deltas
#if CONFIG_LOOPFILTER_LEVEL
        memset(lfi->lvl[seg_id][dir], lvl_seg, sizeof(lfi->lvl[seg_id][dir]));
#else
        memset(lfi->lvl[seg_id], lvl_seg, sizeof(lfi->lvl[seg_id]));
#endif  // CONFIG_LOOPFILTER_LEVEL
      } else {
        int ref, mode;
Cheng Chen's avatar
Cheng Chen committed
795
#if CONFIG_LOOPFILTER_LEVEL
796 797 798 799 800 801 802 803 804 805 806 807 808 809 810
        scale = 1 << (lvl_seg >> 5);

        const int intra_lvl = lvl_seg + lf->ref_deltas[INTRA_FRAME] * scale;
        lfi->lvl[seg_id][dir][INTRA_FRAME][0] =
            clamp(intra_lvl, 0, MAX_LOOP_FILTER);

        for (ref = LAST_FRAME; ref < TOTAL_REFS_PER_FRAME; ++ref) {
          for (mode = 0; mode < MAX_MODE_LF_DELTAS; ++mode) {
            const int inter_lvl = lvl_seg + lf->ref_deltas[ref] * scale +
                                  lf->mode_deltas[mode] * scale;
            lfi->lvl[seg_id][dir][ref][mode] =
                clamp(inter_lvl, 0, MAX_LOOP_FILTER);
          }
        }
#else
811 812 813 814 815 816 817 818 819 820
        (void)default_filt_lvl_r;
        const int intra_lvl = lvl_seg + lf->ref_deltas[INTRA_FRAME] * scale;
        lfi->lvl[seg_id][INTRA_FRAME][0] = clamp(intra_lvl, 0, MAX_LOOP_FILTER);

        for (ref = LAST_FRAME; ref < TOTAL_REFS_PER_FRAME; ++ref) {
          for (mode = 0; mode < MAX_MODE_LF_DELTAS; ++mode) {
            const int inter_lvl = lvl_seg + lf->ref_deltas[ref] * scale +
                                  lf->mode_deltas[mode] * scale;
            lfi->lvl[seg_id][ref][mode] = clamp(inter_lvl, 0, MAX_LOOP_FILTER);
          }
Jingning Han's avatar
Jingning Han committed
821
        }
822
#endif
823
      }
Jingning Han's avatar
Jingning Han committed
824 825 826 827
    }
  }
}

clang-format's avatar
clang-format committed
828 829
static void filter_selectively_vert_row2(int subsampling_factor, uint8_t *s,
                                         int pitch, unsigned int mask_16x16_l,
Jingning Han's avatar
Jingning Han committed
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
                                         unsigned int mask_8x8_l,
                                         unsigned int mask_4x4_l,
                                         unsigned int mask_4x4_int_l,
                                         const loop_filter_info_n *lfi_n,
                                         const uint8_t *lfl) {
  const int mask_shift = subsampling_factor ? 4 : 8;
  const int mask_cutoff = subsampling_factor ? 0xf : 0xff;
  const int lfl_forward = subsampling_factor ? 4 : 8;

  unsigned int mask_16x16_0 = mask_16x16_l & mask_cutoff;
  unsigned int mask_8x8_0 = mask_8x8_l & mask_cutoff;
  unsigned int mask_4x4_0 = mask_4x4_l & mask_cutoff;
  unsigned int mask_4x4_int_0 = mask_4x4_int_l & mask_cutoff;
  unsigned int mask_16x16_1 = (mask_16x16_l >> mask_shift) & mask_cutoff;
  unsigned int mask_8x8_1 = (mask_8x8_l >> mask_shift) & mask_cutoff;
  unsigned int mask_4x4_1 = (mask_4x4_l >> mask_shift) & mask_cutoff;
  unsigned int mask_4x4_int_1 = (mask_4x4_int_l >> mask_shift) & mask_cutoff;
  unsigned int mask;

  for (mask = mask_16x16_0 | mask_8x8_0 | mask_4x4_0 | mask_4x4_int_0 |
              mask_16x16_1 | mask_8x8_1 | mask_4x4_1 | mask_4x4_int_1;
       mask; mask >>= 1) {
    const loop_filter_thresh *lfi0 = lfi_n->lfthr + *lfl;
    const loop_filter_thresh *lfi1 = lfi_n->lfthr + *(lfl + lfl_forward);

    if (mask & 1) {
      if ((mask_16x16_0 | mask_16x16_1) & 1) {
        if ((mask_16x16_0 & mask_16x16_1) & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
858
          aom_lpf_vertical_16_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
859 860
                                   lfi0->hev_thr);
        } else if (mask_16x16_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
861
          aom_lpf_vertical_16(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr);
Jingning Han's avatar
Jingning Han committed
862
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
863
          aom_lpf_vertical_16(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim,
clang-format's avatar
clang-format committed
864
                              lfi1->hev_thr);
Jingning Han's avatar
Jingning Han committed
865 866 867 868 869
        }
      }

      if ((mask_8x8_0 | mask_8x8_1) & 1) {
        if ((mask_8x8_0 & mask_8x8_1) & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
870
          aom_lpf_vertical_8_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
871 872 873
                                  lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                  lfi1->hev_thr);
        } else if (mask_8x8_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
874
          aom_lpf_vertical_8(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr);
Jingning Han's avatar
Jingning Han committed
875
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
876
          aom_lpf_vertical_8(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim,
877
                             lfi1->hev_thr);
Jingning Han's avatar
Jingning Han committed
878 879 880 881 882
        }
      }

      if ((mask_4x4_0 | mask_4x4_1) & 1) {
        if ((mask_4x4_0 & mask_4x4_1) & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
883
          aom_lpf_vertical_4_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
884 885 886
                                  lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                  lfi1->hev_thr);
        } else if (mask_4x4_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
887
          aom_lpf_vertical_4(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr);
Jingning Han's avatar
Jingning Han committed
888
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
889
          aom_lpf_vertical_4(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim,
890
                             lfi1->hev_thr);
Jingning Han's avatar
Jingning Han committed
891 892 893 894 895
        }
      }

      if ((mask_4x4_int_0 | mask_4x4_int_1) & 1) {
        if ((mask_4x4_int_0 & mask_4x4_int_1) & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
896
          aom_lpf_vertical_4_dual(s + 4, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
897 898 899
                                  lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                  lfi1->hev_thr);
        } else if (mask_4x4_int_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
900
          aom_lpf_vertical_4(s + 4, pitch, lfi0->mblim, lfi0->lim,
901
                             lfi0->hev_thr);
Jingning Han's avatar
Jingning Han committed
902
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
903
          aom_lpf_vertical_4(s + 8 * pitch + 4, pitch, lfi1->mblim, lfi1->lim,
904
                             lfi1->hev_thr);
Jingning Han's avatar
Jingning Han committed
905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921
        }
      }
    }

    s += 8;
    lfl += 1;
    mask_16x16_0 >>= 1;
    mask_8x8_0 >>= 1;
    mask_4x4_0 >>= 1;
    mask_4x4_int_0 >>= 1;
    mask_16x16_1 >>= 1;
    mask_8x8_1 >>= 1;
    mask_4x4_1 >>= 1;
    mask_4x4_int_1 >>= 1;
  }
}

922
#if CONFIG_HIGHBITDEPTH
clang-format's avatar
clang-format committed
923 924 925 926 927
static void highbd_filter_selectively_vert_row2(
    int subsampling_factor, uint16_t *s, int pitch, unsigned int mask_16x16_l,
    unsigned int mask_8x8_l, unsigned int mask_4x4_l,
    unsigned int mask_4x4_int_l, const loop_filter_info_n *lfi_n,
    const uint8_t *lfl, int bd) {
Jingning Han's avatar
Jingning Han committed
928 929 930 931 932 933 934 935 936 937 938 939 940 941 942
  const int mask_shift = subsampling_factor ? 4 : 8;
  const int mask_cutoff = subsampling_factor ? 0xf : 0xff;
  const int lfl_forward = subsampling_factor ? 4 : 8;

  unsigned int mask_16x16_0 = mask_16x16_l & mask_cutoff;
  unsigned int mask_8x8_0 = mask_8x8_l & mask_cutoff;
  unsigned int mask_4x4_0 = mask_4x4_l & mask_cutoff;
  unsigned int mask_4x4_int_0 = mask_4x4_int_l & mask_cutoff;
  unsigned int mask_16x16_1 = (mask_16x16_l >> mask_shift) & mask_cutoff;
  unsigned int mask_8x8_1 = (mask_8x8_l >> mask_shift) & mask_cutoff;
  unsigned int mask_4x4_1 = (mask_4x4_l >> mask_shift) & mask_cutoff;
  unsigned int mask_4x4_int_1 = (mask_4x4_int_l >> mask_shift) & mask_cutoff;
  unsigned int mask;

  for (mask = mask_16x16_0 | mask_8x8_0 | mask_4x4_0 | mask_4x4_int_0 |
clang-format's avatar
clang-format committed
943
              mask_16x16_1 | mask_8x8_1 | mask_4x4_1 | mask_4x4_int_1;
Jingning Han's avatar
Jingning Han committed
944 945 946 947 948 949 950
       mask; mask >>= 1) {
    const loop_filter_thresh *lfi0 = lfi_n->lfthr + *lfl;
    const loop_filter_thresh *lfi1 = lfi_n->lfthr + *(lfl + lfl_forward);

    if (mask & 1) {
      if ((mask_16x16_0 | mask_16x16_1) & 1) {
        if ((mask_16x16_0 & mask_16x16_1) & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
951
          aom_highbd_lpf_vertical_16_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
952 953
                                          lfi0->hev_thr, bd);
        } else if (mask_16x16_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
954
          aom_highbd_lpf_vertical_16(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
955 956
                                     lfi0->hev_thr, bd);
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
957
          aom_highbd_lpf_vertical_16(s + 8 * pitch, pitch, lfi1->mblim,
Jingning Han's avatar
Jingning Han committed
958 959 960 961 962 963
                                     lfi1->lim, lfi1->hev_thr, bd);
        }
      }

      if ((mask_8x8_0 | mask_8x8_1) & 1) {
        if ((mask_8x8_0 & mask_8x8_1) & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
964
          aom_highbd_lpf_vertical_8_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
965 966 967
                                         lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                         lfi1->hev_thr, bd);
        } else if (mask_8x8_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
968
          aom_highbd_lpf_vertical_8(s, pitch, lfi0->mblim, lfi0->lim,
969
                                    lfi0->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
970
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
971
          aom_highbd_lpf_vertical_8(s + 8 * pitch, pitch, lfi1->mblim,
972
                                    lfi1->lim, lfi1->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
973 974 975 976 977
        }
      }

      if ((mask_4x4_0 | mask_4x4_1) & 1) {
        if ((mask_4x4_0 & mask_4x4_1) & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
978
          aom_highbd_lpf_vertical_4_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
979 980 981
                                         lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                         lfi1->hev_thr, bd);
        } else if (mask_4x4_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
982
          aom_highbd_lpf_vertical_4(s, pitch, lfi0->mblim, lfi0->lim,
983
                                    lfi0->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
984
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
985
          aom_highbd_lpf_vertical_4(s + 8 * pitch, pitch, lfi1->mblim,
986
                                    lfi1->lim, lfi1->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
987 988 989 990 991
        }
      }

      if ((mask_4x4_int_0 | mask_4x4_int_1) & 1) {
        if ((mask_4x4_int_0 & mask_4x4_int_1) & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
992
          aom_highbd_lpf_vertical_4_dual(s + 4, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
993 994 995
                                         lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                         lfi1->hev_thr, bd);
        } else if (mask_4x4_int_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
996
          aom_highbd_lpf_vertical_4(s + 4, pitch, lfi0->mblim, lfi0->lim,
997
                                    lfi0->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
998
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
999
          aom_highbd_lpf_vertical_4(s + 8 * pitch + 4, pitch, lfi1->mblim,
1000
                                    lfi1->lim, lfi1->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016
        }
      }
    }

    s += 8;
    lfl += 1;
    mask_16x16_0 >>= 1;
    mask_8x8_0 >>= 1;
    mask_4x4_0 >>= 1;
    mask_4x4_int_0 >>= 1;
    mask_16x16_1 >>= 1;
    mask_8x8_1 >>= 1;
    mask_4x4_1 >>= 1;
    mask_4x4_int_1 >>= 1;
  }
}
1017
#endif  // CONFIG_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
1018

clang-format's avatar
clang-format committed
1019 1020 1021
static void filter_selectively_horiz(
    uint8_t *s, int pitch, unsigned int mask_16x16, unsigned int mask_8x8,
    unsigned int mask_4x4, unsigned int mask_4x4_int,
Cheng Chen's avatar
Cheng Chen committed
1022 1023 1024 1025 1026 1027 1028
    const loop_filter_info_n *lfi_n, const uint8_t *lfl
#if CONFIG_LPF_DIRECT
    ,
    uint8_t *const src, int mi_row, int mi_col, int idx_r, int col_step,
    int width, int height, int ss_x, int ss_y
#endif
    ) {
Jingning Han's avatar
Jingning Han committed
1029 1030
  unsigned int mask;
  int count;
Cheng Chen's avatar
Cheng Chen committed
1031
#if CONFIG_LPF_DIRECT
1032 1033 1034
  // scale for u, v plane
  width >>= ss_x;
  height >>= ss_y;
Cheng Chen's avatar
Cheng Chen committed
1035 1036
  int idx_c = 0;
#endif
Jingning Han's avatar
Jingning Han committed
1037

clang-format's avatar
clang-format committed
1038 1039
  for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; mask;
       mask >>= count) {
Jingning Han's avatar
Jingning Han committed
1040 1041 1042 1043
    const loop_filter_thresh *lfi = lfi_n->lfthr + *lfl;

    count = 1;
    if (mask & 1) {
Cheng Chen's avatar
Cheng Chen committed
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
#if CONFIG_LPF_DIRECT
      int i;
      const int line_length = 16;
      const int pivot = 8;
      const int above_filt_len = mask_16x16 & 1 ? 8 : 4;
      const int bot_filt_len = mask_16x16 & 1 ? 8 : 4;
      uint8_t block[256];  // line_length * size_of(BLOCK_8X8) * two_blocks
      int orig_pos[256];
      int direct;

      assert(above_filt_len == bot_filt_len);
      (void)bot_filt_len;
      for (i = 0; i < 256; ++i) {
        block[i] = 0;
        orig_pos[i] = -1;
      }

      // actual position for current pixel
      const int row = (mi_row + idx_r) * MI_SIZE >> ss_y;
      const int col = (mi_col + idx_c) * MI_SIZE >> ss_x;

      // Next block's thresholds.
      const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + 1);

      if (mask_16x16 & 1) {
        if ((mask_16x16 & 3) == 3) {
          // Could use asymmetric length in the future
          direct = pick_min_grad_direct(src, above_filt_len, row, col, width,
                                        height, pitch, 2, 1);

          pick_filter_block_horz(src, block, orig_pos, above_filt_len, row, col,
                                 width, height, pitch, pivot, line_length, 2,
                                 direct);

          aom_lpf_horizontal_edge_16(block + pivot * line_length, line_length,
                                     lfi->mblim, lfi->lim, lfi->hev_thr);
          count = 2;
        } else {
          direct = pick_min_grad_direct(src, above_filt_len, row, col, width,
                                        height, pitch, 1, 1);

          pick_filter_block_horz(src, block, orig_pos, above_filt_len, row, col,
                                 width, height, pitch, pivot, line_length, 1,
                                 direct);

          aom_lpf_horizontal_edge_8(block + pivot * line_length, line_length,
                                    lfi->mblim, lfi->lim, lfi->hev_thr);
        }

        for (i = 0; i < 256; ++i)
          if (orig_pos[i] >= 0) src[orig_pos[i]] = block[i];
      } else if (mask_8x8 & 1) {
        if ((mask_8x8 & 3) == 3) {
          count = 2;
          direct = pick_min_grad_direct(src, above_filt_len, row, col, width,
                                        height, pitch, 2, 1);

          pick_filter_block_horz(src, block, orig_pos, above_filt_len, row, col,
                                 width, height, pitch, pivot, line_length, 2,
                                 direct);

          aom_lpf_horizontal_8_dual(block + pivot * line_length, line_length,
                                    lfi->mblim, lfi->lim, lfi->hev_thr,
                                    lfin->mblim, lfin->lim, lfin->hev_thr);

          for (i = 0; i < 256; ++i)
            if (orig_pos[i] >= 0) src[orig_pos[i]] = block[i];

          if ((mask_4x4_int & 3) == 3) {
            for (i = 0; i < 256; ++i) {
              block[i] = 0;
              orig_pos[i] = -1;
            }

            direct = pick_min_grad_direct(src, 4, row, col, width, height,
                                          pitch, 2, 1);

            pick_filter_block_horz(src, block, orig_pos, 4, row + 4, col, width,
                                   height, pitch, pivot, line_length, 2,
                                   direct);

            aom_lpf_horizontal_4_dual(block + pivot * line_length, line_length,
                                      lfi->mblim, lfi->lim, lfi->hev_thr,
                                      lfin->mblim, lfin->lim, lfin->hev_thr);

            for (i = 0; i < 256; ++i)
              if (orig_pos[i] >= 0) src[orig_pos[i]] = block[i];
          } else {
            for (i = 0; i < 256; ++i) {
              block[i] = 0;
              orig_pos[i] = -1;
            }

            if (mask_4x4_int & 1) {
              direct = pick_min_grad_direct(src, 4, row, col, width, height,
                                            pitch, 1, 1);

              pick_filter_block_horz(src, block, orig_pos, 4, row + 4, col,
                                     width, height, pitch, pivot, line_length,
                                     1, direct);

              aom_lpf_horizontal_4(block + pivot * line_length, line_length,
                                   lfi->mblim, lfi->lim, lfi->hev_thr);
            } else if (mask_4x4_int & 2) {
              direct = pick_min_grad_direct(src, 4, row, col, width, height,
                                            pitch, 1, 1);

              pick_filter_block_horz(src, block, orig_pos, 4, row + 4, col + 8,
                                     width, height, pitch, pivot, line_length,
                                     1, direct);

              aom_lpf_horizontal_4(block + pivot * line_length, line_length,
                                   lfin->mblim, lfin->lim, lfin->hev_thr);
            }

            for (i = 0; i < 256; ++i)
              if (orig_pos[i] >= 0) src[orig_pos[i]] = block[i];
          }
        } else {
          direct = pick_min_grad_direct(src, above_filt_len, row, col, width,
                                        height, pitch, 1, 1);

          pick_filter_block_horz(src, block, orig_pos, above_filt_len, row, col,
                                 width, height, pitch, pivot, line_length, 1,
                                 direct);

          aom_lpf_horizontal_8(block + pivot * line_length, line_length,
                               lfi->mblim, lfi->lim, lfi->hev_thr);

          for (i = 0; i < 256; ++i)
            if (orig_pos[i] >= 0) src[orig_pos[i]] = block[i];

          if (mask_4x4_int & 1) {
            for (i = 0; i < 256; ++i) {
              block[i] = 0;
              orig_pos[i] = -1;
            }
            direct = pick_min_grad_direct(src, 4, row, col, width, height,
                                          pitch, 1, 1);

            pick_filter_block_horz(src, block, orig_pos, 4, row + 4, col, width,
                                   height, pitch, pivot, line_length, 1,
                                   direct);

            aom_lpf_horizontal_4(block + pivot * line_length, line_length,
                                 lfi->mblim, lfi->lim, lfi->hev_thr);

            for (i = 0; i < 256; ++i)
              if (orig_pos[i] >= 0) src[orig_pos[i]] = block[i];
          }
        }
      } else if (mask_4x4 & 1) {
        if ((mask_4x4 & 3) == 3) {
          count = 2;
          direct = pick_min_grad_direct(src, 4, row, col, width, height, pitch,
                                        2, 1);

          pick_filter_block_horz(src, block, orig_pos, 4, row, col, width,
                                 height, pitch, pivot, line_length, 2, direct);

          aom_lpf_horizontal_4_dual(block + pivot * line_length, line_length,
                                    lfi->mblim, lfi->lim, lfi->hev_thr,
                                    lfin->mblim, lfin->lim, lfin->hev_thr);

          for (i = 0; i < 256; ++i)
            if (