loopfilter.c 63.7 KB
Newer Older
Jingning Han's avatar
Jingning Han committed
1
/*
2
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
Jingning Han's avatar
Jingning Han committed
3
 *
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
 */

Adrian Grange's avatar
Adrian Grange committed
12 13
#include "./aom_config.h"
#include "./aom_dsp_rtcd.h"
Yaowu Xu's avatar
Yaowu Xu committed
14 15 16
#include "av1/common/loopfilter.h"
#include "av1/common/onyxc_int.h"
#include "av1/common/reconinter.h"
Adrian Grange's avatar
Adrian Grange committed
17 18
#include "aom_dsp/aom_dsp_common.h"
#include "aom_mem/aom_mem.h"
Yaowu Xu's avatar
Yaowu Xu committed
19
#include "aom_ports/mem.h"
Jingning Han's avatar
Jingning Han committed
20

Yaowu Xu's avatar
Yaowu Xu committed
21
#include "av1/common/seg_common.h"
Jingning Han's avatar
Jingning Han committed
22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39

// 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
40
static const uint64_t left_64x64_txform_mask[TX_SIZES] = {
Jingning Han's avatar
Jingning Han committed
41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63
  0xffffffffffffffffULL,  // TX_4X4
  0xffffffffffffffffULL,  // TX_8x8
  0x5555555555555555ULL,  // TX_16x16
  0x1111111111111111ULL,  // TX_32x32
};

// 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
64
static const uint64_t above_64x64_txform_mask[TX_SIZES] = {
Jingning Han's avatar
Jingning Han committed
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
  0xffffffffffffffffULL,  // TX_4X4
  0xffffffffffffffffULL,  // TX_8x8
  0x00ff00ff00ff00ffULL,  // TX_16x16
  0x000000ff000000ffULL,  // TX_32x32
};

// 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
static const uint64_t left_prediction_mask[BLOCK_SIZES] = {
  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,
  0x0101010101010101ULL,  // BLOCK_64X64
};

// 64 bit mask to shift and set for each prediction size.
static const uint64_t above_prediction_mask[BLOCK_SIZES] = {
  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,
  0x00000000000000ffULL,  // BLOCK_64X64
};
// 64 bit mask to shift and set for each prediction size. A bit is set for
// each 8x8 block that would be in the left most block of the given block
// size in the 64x64 block.
static const uint64_t size_mask[BLOCK_SIZES] = {
  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,
  0xffffffffffffffffULL,  // BLOCK_64X64
};

// These are used for masking the left and above borders.
clang-format's avatar
clang-format committed
138
static const uint64_t left_border = 0x1111111111111111ULL;
Jingning Han's avatar
Jingning Han committed
139 140 141
static const uint64_t above_border = 0x000000ff000000ffULL;

// 16 bit masks for uv transform sizes.
clang-format's avatar
clang-format committed
142
static const uint16_t left_64x64_txform_mask_uv[TX_SIZES] = {
Jingning Han's avatar
Jingning Han committed
143 144 145 146 147 148
  0xffff,  // TX_4X4
  0xffff,  // TX_8x8
  0x5555,  // TX_16x16
  0x1111,  // TX_32x32
};

clang-format's avatar
clang-format committed
149
static const uint16_t above_64x64_txform_mask_uv[TX_SIZES] = {
Jingning Han's avatar
Jingning Han committed
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
  0xffff,  // TX_4X4
  0xffff,  // TX_8x8
  0x0f0f,  // TX_16x16
  0x000f,  // TX_32x32
};

// 16 bit left mask to shift and set for each uv prediction size.
static const uint16_t left_prediction_mask_uv[BLOCK_SIZES] = {
  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,
  0x1111,  // BLOCK_64X64
};
// 16 bit above mask to shift and set for uv each prediction size.
static const uint16_t above_prediction_mask_uv[BLOCK_SIZES] = {
  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,
  0x000f,  // BLOCK_64X64
};

// 64 bit mask to shift and set for each uv prediction size
static const uint16_t size_mask_uv[BLOCK_SIZES] = {
  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,
  0xffff,  // BLOCK_64X64
};
clang-format's avatar
clang-format committed
205
static const uint16_t left_border_uv = 0x1111;
Jingning Han's avatar
Jingning Han committed
206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225
static const uint16_t above_border_uv = 0x000f;

static const int mode_lf_lut[MB_MODE_COUNT] = {
  0, 0, 0, 0, 0, 0, 0, 0, 0, 0,  // INTRA_MODES
  1, 1, 0, 1                     // INTER_MODES (ZEROMV == 0)
};

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
226
    if (block_inside_limit < 1) block_inside_limit = 1;
Jingning Han's avatar
Jingning Han committed
227 228 229 230 231 232 233 234 235

    memset(lfi->lfthr[lvl].lim, block_inside_limit, SIMD_WIDTH);
    memset(lfi->lfthr[lvl].mblim, (2 * (lvl + 2) + block_inside_limit),
           SIMD_WIDTH);
  }
}

static uint8_t get_filter_level(const loop_filter_info_n *lfi_n,
                                const MB_MODE_INFO *mbmi) {
clang-format's avatar
clang-format committed
236 237
  return lfi_n
      ->lvl[mbmi->segment_id][mbmi->ref_frame[0]][mode_lf_lut[mbmi->mode]];
Jingning Han's avatar
Jingning Han committed
238 239
}

240
void av1_loop_filter_init(AV1_COMMON *cm) {
Jingning Han's avatar
Jingning Han committed
241 242 243 244 245 246 247 248 249 250 251 252 253
  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);
}

254
void av1_loop_filter_frame_init(AV1_COMMON *cm, int default_filt_lvl) {
Jingning Han's avatar
Jingning Han committed
255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273
  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
  const int scale = 1 << (default_filt_lvl >> 5);
  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++) {
    int lvl_seg = default_filt_lvl;
    if (segfeature_active(seg, seg_id, SEG_LVL_ALT_LF)) {
      const int data = get_segdata(seg, seg_id, SEG_LVL_ALT_LF);
clang-format's avatar
clang-format committed
274 275 276
      lvl_seg = clamp(
          seg->abs_delta == SEGMENT_ABSDATA ? data : default_filt_lvl + data, 0,
          MAX_LOOP_FILTER);
Jingning Han's avatar
Jingning Han committed
277 278 279 280 281 282 283 284 285 286 287 288 289
    }

    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
      memset(lfi->lvl[seg_id], lvl_seg, sizeof(lfi->lvl[seg_id]));
    } else {
      int ref, mode;
      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 < MAX_REF_FRAMES; ++ref) {
        for (mode = 0; mode < MAX_MODE_LF_DELTAS; ++mode) {
clang-format's avatar
clang-format committed
290 291
          const int inter_lvl = lvl_seg + lf->ref_deltas[ref] * scale +
                                lf->mode_deltas[mode] * scale;
Jingning Han's avatar
Jingning Han committed
292 293 294 295 296 297 298
          lfi->lvl[seg_id][ref][mode] = clamp(inter_lvl, 0, MAX_LOOP_FILTER);
        }
      }
    }
  }
}

clang-format's avatar
clang-format committed
299 300
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
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
                                         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);

    // TODO(yunqingwang): count in loopfilter functions should be removed.
    if (mask & 1) {
      if ((mask_16x16_0 | mask_16x16_1) & 1) {
        if ((mask_16x16_0 & mask_16x16_1) & 1) {
Adrian Grange's avatar
Adrian Grange committed
330
          aom_lpf_vertical_16_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
331 332
                                   lfi0->hev_thr);
        } else if (mask_16x16_0 & 1) {
Adrian Grange's avatar
Adrian Grange committed
333
          aom_lpf_vertical_16(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr);
Jingning Han's avatar
Jingning Han committed
334
        } else {
Adrian Grange's avatar
Adrian Grange committed
335
          aom_lpf_vertical_16(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim,
clang-format's avatar
clang-format committed
336
                              lfi1->hev_thr);
Jingning Han's avatar
Jingning Han committed
337 338 339 340 341
        }
      }

      if ((mask_8x8_0 | mask_8x8_1) & 1) {
        if ((mask_8x8_0 & mask_8x8_1) & 1) {
Adrian Grange's avatar
Adrian Grange committed
342
          aom_lpf_vertical_8_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
343 344 345
                                  lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                  lfi1->hev_thr);
        } else if (mask_8x8_0 & 1) {
Adrian Grange's avatar
Adrian Grange committed
346
          aom_lpf_vertical_8(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr,
Jingning Han's avatar
Jingning Han committed
347 348
                             1);
        } else {
Adrian Grange's avatar
Adrian Grange committed
349
          aom_lpf_vertical_8(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim,
Jingning Han's avatar
Jingning Han committed
350 351 352 353 354 355
                             lfi1->hev_thr, 1);
        }
      }

      if ((mask_4x4_0 | mask_4x4_1) & 1) {
        if ((mask_4x4_0 & mask_4x4_1) & 1) {
Adrian Grange's avatar
Adrian Grange committed
356
          aom_lpf_vertical_4_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
357 358 359
                                  lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                  lfi1->hev_thr);
        } else if (mask_4x4_0 & 1) {
Adrian Grange's avatar
Adrian Grange committed
360
          aom_lpf_vertical_4(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr,
Jingning Han's avatar
Jingning Han committed
361 362
                             1);
        } else {
Adrian Grange's avatar
Adrian Grange committed
363
          aom_lpf_vertical_4(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim,
Jingning Han's avatar
Jingning Han committed
364 365 366 367 368 369
                             lfi1->hev_thr, 1);
        }
      }

      if ((mask_4x4_int_0 | mask_4x4_int_1) & 1) {
        if ((mask_4x4_int_0 & mask_4x4_int_1) & 1) {
Adrian Grange's avatar
Adrian Grange committed
370
          aom_lpf_vertical_4_dual(s + 4, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
371 372 373
                                  lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                  lfi1->hev_thr);
        } else if (mask_4x4_int_0 & 1) {
Adrian Grange's avatar
Adrian Grange committed
374
          aom_lpf_vertical_4(s + 4, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
375 376
                             lfi0->hev_thr, 1);
        } else {
Adrian Grange's avatar
Adrian Grange committed
377
          aom_lpf_vertical_4(s + 8 * pitch + 4, pitch, lfi1->mblim, lfi1->lim,
Jingning Han's avatar
Jingning Han committed
378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395
                             lfi1->hev_thr, 1);
        }
      }
    }

    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;
  }
}

396
#if CONFIG_AOM_HIGHBITDEPTH
clang-format's avatar
clang-format committed
397 398 399 400 401
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
402 403 404 405 406 407 408 409 410 411 412 413 414 415 416
  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
417
              mask_16x16_1 | mask_8x8_1 | mask_4x4_1 | mask_4x4_int_1;
Jingning Han's avatar
Jingning Han committed
418 419 420 421 422 423 424 425
       mask; mask >>= 1) {
    const loop_filter_thresh *lfi0 = lfi_n->lfthr + *lfl;
    const loop_filter_thresh *lfi1 = lfi_n->lfthr + *(lfl + lfl_forward);

    // TODO(yunqingwang): count in loopfilter functions should be removed.
    if (mask & 1) {
      if ((mask_16x16_0 | mask_16x16_1) & 1) {
        if ((mask_16x16_0 & mask_16x16_1) & 1) {
Adrian Grange's avatar
Adrian Grange committed
426
          aom_highbd_lpf_vertical_16_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
427 428
                                          lfi0->hev_thr, bd);
        } else if (mask_16x16_0 & 1) {
Adrian Grange's avatar
Adrian Grange committed
429
          aom_highbd_lpf_vertical_16(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
430 431
                                     lfi0->hev_thr, bd);
        } else {
Adrian Grange's avatar
Adrian Grange committed
432
          aom_highbd_lpf_vertical_16(s + 8 * pitch, pitch, lfi1->mblim,
Jingning Han's avatar
Jingning Han committed
433 434 435 436 437 438
                                     lfi1->lim, lfi1->hev_thr, bd);
        }
      }

      if ((mask_8x8_0 | mask_8x8_1) & 1) {
        if ((mask_8x8_0 & mask_8x8_1) & 1) {
Adrian Grange's avatar
Adrian Grange committed
439
          aom_highbd_lpf_vertical_8_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
440 441 442
                                         lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                         lfi1->hev_thr, bd);
        } else if (mask_8x8_0 & 1) {
Adrian Grange's avatar
Adrian Grange committed
443
          aom_highbd_lpf_vertical_8(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
444 445
                                    lfi0->hev_thr, 1, bd);
        } else {
Adrian Grange's avatar
Adrian Grange committed
446
          aom_highbd_lpf_vertical_8(s + 8 * pitch, pitch, lfi1->mblim,
Jingning Han's avatar
Jingning Han committed
447 448 449 450 451 452
                                    lfi1->lim, lfi1->hev_thr, 1, bd);
        }
      }

      if ((mask_4x4_0 | mask_4x4_1) & 1) {
        if ((mask_4x4_0 & mask_4x4_1) & 1) {
Adrian Grange's avatar
Adrian Grange committed
453
          aom_highbd_lpf_vertical_4_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
454 455 456
                                         lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                         lfi1->hev_thr, bd);
        } else if (mask_4x4_0 & 1) {
Adrian Grange's avatar
Adrian Grange committed
457
          aom_highbd_lpf_vertical_4(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
458 459
                                    lfi0->hev_thr, 1, bd);
        } else {
Adrian Grange's avatar
Adrian Grange committed
460
          aom_highbd_lpf_vertical_4(s + 8 * pitch, pitch, lfi1->mblim,
Jingning Han's avatar
Jingning Han committed
461 462 463 464 465 466
                                    lfi1->lim, lfi1->hev_thr, 1, bd);
        }
      }

      if ((mask_4x4_int_0 | mask_4x4_int_1) & 1) {
        if ((mask_4x4_int_0 & mask_4x4_int_1) & 1) {
Adrian Grange's avatar
Adrian Grange committed
467
          aom_highbd_lpf_vertical_4_dual(s + 4, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
468 469 470
                                         lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                         lfi1->hev_thr, bd);
        } else if (mask_4x4_int_0 & 1) {
Adrian Grange's avatar
Adrian Grange committed
471
          aom_highbd_lpf_vertical_4(s + 4, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
472 473
                                    lfi0->hev_thr, 1, bd);
        } else {
Adrian Grange's avatar
Adrian Grange committed
474
          aom_highbd_lpf_vertical_4(s + 8 * pitch + 4, pitch, lfi1->mblim,
Jingning Han's avatar
Jingning Han committed
475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491
                                    lfi1->lim, lfi1->hev_thr, 1, bd);
        }
      }
    }

    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;
  }
}
492
#endif  // CONFIG_AOM_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
493

clang-format's avatar
clang-format committed
494 495 496 497
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,
    const loop_filter_info_n *lfi_n, const uint8_t *lfl) {
Jingning Han's avatar
Jingning Han committed
498 499 500
  unsigned int mask;
  int count;

clang-format's avatar
clang-format committed
501 502
  for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; mask;
       mask >>= count) {
Jingning Han's avatar
Jingning Han committed
503 504 505 506 507 508
    const loop_filter_thresh *lfi = lfi_n->lfthr + *lfl;

    count = 1;
    if (mask & 1) {
      if (mask_16x16 & 1) {
        if ((mask_16x16 & 3) == 3) {
Adrian Grange's avatar
Adrian Grange committed
509
          aom_lpf_horizontal_16(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr,
clang-format's avatar
clang-format committed
510
                                2);
Jingning Han's avatar
Jingning Han committed
511 512
          count = 2;
        } else {
Adrian Grange's avatar
Adrian Grange committed
513
          aom_lpf_horizontal_16(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr,
clang-format's avatar
clang-format committed
514
                                1);
Jingning Han's avatar
Jingning Han committed
515 516 517 518 519 520
        }
      } else if (mask_8x8 & 1) {
        if ((mask_8x8 & 3) == 3) {
          // Next block's thresholds.
          const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + 1);

Adrian Grange's avatar
Adrian Grange committed
521
          aom_lpf_horizontal_8_dual(s, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
522 523 524 525
                                    lfi->hev_thr, lfin->mblim, lfin->lim,
                                    lfin->hev_thr);

          if ((mask_4x4_int & 3) == 3) {
Adrian Grange's avatar
Adrian Grange committed
526
            aom_lpf_horizontal_4_dual(s + 4 * pitch, pitch, lfi->mblim,
Jingning Han's avatar
Jingning Han committed
527 528 529 530
                                      lfi->lim, lfi->hev_thr, lfin->mblim,
                                      lfin->lim, lfin->hev_thr);
          } else {
            if (mask_4x4_int & 1)
Adrian Grange's avatar
Adrian Grange committed
531
              aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
532 533
                                   lfi->hev_thr, 1);
            else if (mask_4x4_int & 2)
Adrian Grange's avatar
Adrian Grange committed
534
              aom_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
Jingning Han's avatar
Jingning Han committed
535 536 537 538
                                   lfin->lim, lfin->hev_thr, 1);
          }
          count = 2;
        } else {
Adrian Grange's avatar
Adrian Grange committed
539
          aom_lpf_horizontal_8(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
Jingning Han's avatar
Jingning Han committed
540 541

          if (mask_4x4_int & 1)
Adrian Grange's avatar
Adrian Grange committed
542
            aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
543 544 545 546 547 548 549
                                 lfi->hev_thr, 1);
        }
      } else if (mask_4x4 & 1) {
        if ((mask_4x4 & 3) == 3) {
          // Next block's thresholds.
          const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + 1);

Adrian Grange's avatar
Adrian Grange committed
550
          aom_lpf_horizontal_4_dual(s, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
551 552 553
                                    lfi->hev_thr, lfin->mblim, lfin->lim,
                                    lfin->hev_thr);
          if ((mask_4x4_int & 3) == 3) {
Adrian Grange's avatar
Adrian Grange committed
554
            aom_lpf_horizontal_4_dual(s + 4 * pitch, pitch, lfi->mblim,
Jingning Han's avatar
Jingning Han committed
555 556 557 558
                                      lfi->lim, lfi->hev_thr, lfin->mblim,
                                      lfin->lim, lfin->hev_thr);
          } else {
            if (mask_4x4_int & 1)
Adrian Grange's avatar
Adrian Grange committed
559
              aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
560 561
                                   lfi->hev_thr, 1);
            else if (mask_4x4_int & 2)
Adrian Grange's avatar
Adrian Grange committed
562
              aom_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
Jingning Han's avatar
Jingning Han committed
563 564 565 566
                                   lfin->lim, lfin->hev_thr, 1);
          }
          count = 2;
        } else {
Adrian Grange's avatar
Adrian Grange committed
567
          aom_lpf_horizontal_4(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
Jingning Han's avatar
Jingning Han committed
568 569

          if (mask_4x4_int & 1)
Adrian Grange's avatar
Adrian Grange committed
570
            aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
571 572 573
                                 lfi->hev_thr, 1);
        }
      } else if (mask_4x4_int & 1) {
Adrian Grange's avatar
Adrian Grange committed
574
        aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
575 576 577 578 579 580 581 582 583 584 585 586
                             lfi->hev_thr, 1);
      }
    }
    s += 8 * count;
    lfl += count;
    mask_16x16 >>= count;
    mask_8x8 >>= count;
    mask_4x4 >>= count;
    mask_4x4_int >>= count;
  }
}

587
#if CONFIG_AOM_HIGHBITDEPTH
clang-format's avatar
clang-format committed
588 589 590 591
static void highbd_filter_selectively_horiz(
    uint16_t *s, int pitch, unsigned int mask_16x16, unsigned int mask_8x8,
    unsigned int mask_4x4, unsigned int mask_4x4_int,
    const loop_filter_info_n *lfi_n, const uint8_t *lfl, int bd) {
Jingning Han's avatar
Jingning Han committed
592 593 594
  unsigned int mask;
  int count;

clang-format's avatar
clang-format committed
595 596
  for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; mask;
       mask >>= count) {
Jingning Han's avatar
Jingning Han committed
597 598 599 600 601 602
    const loop_filter_thresh *lfi = lfi_n->lfthr + *lfl;

    count = 1;
    if (mask & 1) {
      if (mask_16x16 & 1) {
        if ((mask_16x16 & 3) == 3) {
Adrian Grange's avatar
Adrian Grange committed
603
          aom_highbd_lpf_horizontal_16(s, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
604 605 606
                                       lfi->hev_thr, 2, bd);
          count = 2;
        } else {
Adrian Grange's avatar
Adrian Grange committed
607
          aom_highbd_lpf_horizontal_16(s, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
608 609 610 611 612 613 614
                                       lfi->hev_thr, 1, bd);
        }
      } else if (mask_8x8 & 1) {
        if ((mask_8x8 & 3) == 3) {
          // Next block's thresholds.
          const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + 1);

Adrian Grange's avatar
Adrian Grange committed
615
          aom_highbd_lpf_horizontal_8_dual(s, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
616 617 618 619
                                           lfi->hev_thr, lfin->mblim, lfin->lim,
                                           lfin->hev_thr, bd);

          if ((mask_4x4_int & 3) == 3) {
Adrian Grange's avatar
Adrian Grange committed
620
            aom_highbd_lpf_horizontal_4_dual(
clang-format's avatar
clang-format committed
621 622
                s + 4 * pitch, pitch, lfi->mblim, lfi->lim, lfi->hev_thr,
                lfin->mblim, lfin->lim, lfin->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
623 624
          } else {
            if (mask_4x4_int & 1) {
Adrian Grange's avatar
Adrian Grange committed
625
              aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
Jingning Han's avatar
Jingning Han committed
626 627
                                          lfi->lim, lfi->hev_thr, 1, bd);
            } else if (mask_4x4_int & 2) {
Adrian Grange's avatar
Adrian Grange committed
628
              aom_highbd_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
Jingning Han's avatar
Jingning Han committed
629 630 631 632 633
                                          lfin->lim, lfin->hev_thr, 1, bd);
            }
          }
          count = 2;
        } else {
Adrian Grange's avatar
Adrian Grange committed
634
          aom_highbd_lpf_horizontal_8(s, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
635 636 637
                                      lfi->hev_thr, 1, bd);

          if (mask_4x4_int & 1) {
Adrian Grange's avatar
Adrian Grange committed
638
            aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
Jingning Han's avatar
Jingning Han committed
639 640 641 642 643 644 645 646
                                        lfi->lim, lfi->hev_thr, 1, bd);
          }
        }
      } else if (mask_4x4 & 1) {
        if ((mask_4x4 & 3) == 3) {
          // Next block's thresholds.
          const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + 1);

Adrian Grange's avatar
Adrian Grange committed
647
          aom_highbd_lpf_horizontal_4_dual(s, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
648 649 650
                                           lfi->hev_thr, lfin->mblim, lfin->lim,
                                           lfin->hev_thr, bd);
          if ((mask_4x4_int & 3) == 3) {
Adrian Grange's avatar
Adrian Grange committed
651
            aom_highbd_lpf_horizontal_4_dual(
clang-format's avatar
clang-format committed
652 653
                s + 4 * pitch, pitch, lfi->mblim, lfi->lim, lfi->hev_thr,
                lfin->mblim, lfin->lim, lfin->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
654 655
          } else {
            if (mask_4x4_int & 1) {
Adrian Grange's avatar
Adrian Grange committed
656
              aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
Jingning Han's avatar
Jingning Han committed
657 658
                                          lfi->lim, lfi->hev_thr, 1, bd);
            } else if (mask_4x4_int & 2) {
Adrian Grange's avatar
Adrian Grange committed
659
              aom_highbd_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
Jingning Han's avatar
Jingning Han committed
660 661 662 663 664
                                          lfin->lim, lfin->hev_thr, 1, bd);
            }
          }
          count = 2;
        } else {
Adrian Grange's avatar
Adrian Grange committed
665
          aom_highbd_lpf_horizontal_4(s, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
666 667 668
                                      lfi->hev_thr, 1, bd);

          if (mask_4x4_int & 1) {
Adrian Grange's avatar
Adrian Grange committed
669
            aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
Jingning Han's avatar
Jingning Han committed
670 671 672 673
                                        lfi->lim, lfi->hev_thr, 1, bd);
          }
        }
      } else if (mask_4x4_int & 1) {
Adrian Grange's avatar
Adrian Grange committed
674
        aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
675 676 677 678 679 680 681 682 683 684 685
                                    lfi->hev_thr, 1, bd);
      }
    }
    s += 8 * count;
    lfl += count;
    mask_16x16 >>= count;
    mask_8x8 >>= count;
    mask_4x4 >>= count;
    mask_4x4_int >>= count;
  }
}
686
#endif  // CONFIG_AOM_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
687 688 689 690 691 692 693 694 695 696

// This function ors into the current lfm structure, where to do loop
// filters for the specific mi we are looking at. It uses information
// including the block_size_type (32x16, 32x32, etc.), the transform size,
// whether there were any coefficients encoded, and the loop filter strength
// block we are currently looking at. Shift is used to position the
// 1's we produce.
// TODO(JBB) Need another function for different resolution color..
static void build_masks(const loop_filter_info_n *const lfi_n,
                        const MODE_INFO *mi, const int shift_y,
clang-format's avatar
clang-format committed
697
                        const int shift_uv, LOOP_FILTER_MASK *lfm) {
Jingning Han's avatar
Jingning Han committed
698 699 700 701 702 703 704 705 706 707
  const MB_MODE_INFO *mbmi = &mi->mbmi;
  const BLOCK_SIZE block_size = mbmi->sb_type;
  const TX_SIZE tx_size_y = mbmi->tx_size;
  const TX_SIZE tx_size_uv = get_uv_tx_size_impl(tx_size_y, block_size, 1, 1);
  const int filter_level = get_filter_level(lfi_n, mbmi);
  uint64_t *const left_y = &lfm->left_y[tx_size_y];
  uint64_t *const above_y = &lfm->above_y[tx_size_y];
  uint64_t *const int_4x4_y = &lfm->int_4x4_y;
  uint16_t *const left_uv = &lfm->left_uv[tx_size_uv];
  uint16_t *const above_uv = &lfm->above_uv[tx_size_uv];
708 709 710
#if CONFIG_MISC_FIXES
  uint16_t *const int_4x4_uv = &lfm->left_int_4x4_uv;
#else
Jingning Han's avatar
Jingning Han committed
711
  uint16_t *const int_4x4_uv = &lfm->int_4x4_uv;
712
#endif
Jingning Han's avatar
Jingning Han committed
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
  int i;

  // If filter level is 0 we don't loop filter.
  if (!filter_level) {
    return;
  } else {
    const int w = num_8x8_blocks_wide_lookup[block_size];
    const int h = num_8x8_blocks_high_lookup[block_size];
    int index = shift_y;
    for (i = 0; i < h; i++) {
      memset(&lfm->lfl_y[index], filter_level, w);
      index += 8;
    }
  }

  // These set 1 in the current block size for the block size edges.
  // For instance if the block size is 32x16, we'll set:
  //    above =   1111
  //              0000
  //    and
  //    left  =   1000
  //          =   1000
  // NOTE : In this example the low bit is left most ( 1000 ) is stored as
  //        1,  not 8...
  //
  // U and V set things on a 16 bit scale.
  //
  *above_y |= above_prediction_mask[block_size] << shift_y;
  *above_uv |= above_prediction_mask_uv[block_size] << shift_uv;
  *left_y |= left_prediction_mask[block_size] << shift_y;
  *left_uv |= left_prediction_mask_uv[block_size] << shift_uv;

clang-format's avatar
clang-format committed
745 746
// If the block has no coefficients and is not intra we skip applying
// the loop filter on block edges.
747
#if CONFIG_MISC_FIXES
clang-format's avatar
clang-format committed
748
  if ((mbmi->skip || mbmi->has_no_coeffs) && is_inter_block(mbmi)) return;
749
#else
clang-format's avatar
clang-format committed
750
  if (mbmi->skip && is_inter_block(mbmi)) return;
751
#endif
Jingning Han's avatar
Jingning Han committed
752 753 754 755 756

  // Here we are adding a mask for the transform size. The transform
  // size mask is set to be correct for a 64x64 prediction block size. We
  // mask to match the size of the block we are working on and then shift it
  // into place..
clang-format's avatar
clang-format committed
757 758 759 760 761
  *above_y |= (size_mask[block_size] & above_64x64_txform_mask[tx_size_y])
              << shift_y;
  *above_uv |=
      (size_mask_uv[block_size] & above_64x64_txform_mask_uv[tx_size_uv])
      << shift_uv;
Jingning Han's avatar
Jingning Han committed
762

clang-format's avatar
clang-format committed
763 764 765 766
  *left_y |= (size_mask[block_size] & left_64x64_txform_mask[tx_size_y])
             << shift_y;
  *left_uv |= (size_mask_uv[block_size] & left_64x64_txform_mask_uv[tx_size_uv])
              << shift_uv;
Jingning Han's avatar
Jingning Han committed
767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808

  // Here we are trying to determine what to do with the internal 4x4 block
  // boundaries.  These differ from the 4x4 boundaries on the outside edge of
  // an 8x8 in that the internal ones can be skipped and don't depend on
  // the prediction block size.
  if (tx_size_y == TX_4X4)
    *int_4x4_y |= (size_mask[block_size] & 0xffffffffffffffffULL) << shift_y;

  if (tx_size_uv == TX_4X4)
    *int_4x4_uv |= (size_mask_uv[block_size] & 0xffff) << shift_uv;
}

// This function does the same thing as the one above with the exception that
// it only affects the y masks. It exists because for blocks < 16x16 in size,
// we only update u and v masks on the first block.
static void build_y_mask(const loop_filter_info_n *const lfi_n,
                         const MODE_INFO *mi, const int shift_y,
                         LOOP_FILTER_MASK *lfm) {
  const MB_MODE_INFO *mbmi = &mi->mbmi;
  const BLOCK_SIZE block_size = mbmi->sb_type;
  const TX_SIZE tx_size_y = mbmi->tx_size;
  const int filter_level = get_filter_level(lfi_n, mbmi);
  uint64_t *const left_y = &lfm->left_y[tx_size_y];
  uint64_t *const above_y = &lfm->above_y[tx_size_y];
  uint64_t *const int_4x4_y = &lfm->int_4x4_y;
  int i;

  if (!filter_level) {
    return;
  } else {
    const int w = num_8x8_blocks_wide_lookup[block_size];
    const int h = num_8x8_blocks_high_lookup[block_size];
    int index = shift_y;
    for (i = 0; i < h; i++) {
      memset(&lfm->lfl_y[index], filter_level, w);
      index += 8;
    }
  }

  *above_y |= above_prediction_mask[block_size] << shift_y;
  *left_y |= left_prediction_mask[block_size] << shift_y;

809
#if CONFIG_MISC_FIXES
clang-format's avatar
clang-format committed
810
  if ((mbmi->skip || mbmi->has_no_coeffs) && is_inter_block(mbmi)) return;
811
#else
clang-format's avatar
clang-format committed
812
  if (mbmi->skip && is_inter_block(mbmi)) return;
813
#endif
Jingning Han's avatar
Jingning Han committed
814

clang-format's avatar
clang-format committed
815 816
  *above_y |= (size_mask[block_size] & above_64x64_txform_mask[tx_size_y])
              << shift_y;
Jingning Han's avatar
Jingning Han committed
817

clang-format's avatar
clang-format committed
818 819
  *left_y |= (size_mask[block_size] & left_64x64_txform_mask[tx_size_y])
             << shift_y;
Jingning Han's avatar
Jingning Han committed
820 821 822 823 824 825 826 827

  if (tx_size_y == TX_4X4)
    *int_4x4_y |= (size_mask[block_size] & 0xffffffffffffffffULL) << shift_y;
}

// This function sets up the bit masks for the entire 64x64 region represented
// by mi_row, mi_col.
// TODO(JBB): This function only works for yv12.
828
void av1_setup_mask(AV1_COMMON *const cm, const int mi_row, const int mi_col,
clang-format's avatar
clang-format committed
829 830
                    MODE_INFO **mi, const int mode_info_stride,
                    LOOP_FILTER_MASK *lfm) {
Jingning Han's avatar
Jingning Han committed
831 832 833 834 835 836 837 838 839
  int idx_32, idx_16, idx_8;
  const loop_filter_info_n *const lfi_n = &cm->lf_info;
  MODE_INFO **mip = mi;
  MODE_INFO **mip2 = mi;

  // These are offsets to the next mi in the 64x64 block. It is what gets
  // added to the mi ptr as we go through each loop. It helps us to avoid
  // setting up special row and column counters for each index. The last step
  // brings us out back to the starting position.
clang-format's avatar
clang-format committed
840 841 842 843 844
  const int offset_32[] = { 4, (mode_info_stride << 2) - 4, 4,
                            -(mode_info_stride << 2) - 4 };
  const int offset_16[] = { 2, (mode_info_stride << 1) - 2, 2,
                            -(mode_info_stride << 1) - 2 };
  const int offset[] = { 1, mode_info_stride - 1, 1, -mode_info_stride - 1 };
Jingning Han's avatar
Jingning Han committed
845 846 847 848 849

  // Following variables represent shifts to position the current block
  // mask over the appropriate block. A shift of 36 to the left will move
  // the bits for the final 32 by 32 block in the 64x64 up 4 rows and left
  // 4 rows to the appropriate spot.
clang-format's avatar
clang-format committed
850 851 852 853 854
  const int shift_32_y[] = { 0, 4, 32, 36 };
  const int shift_16_y[] = { 0, 2, 16, 18 };
  const int shift_8_y[] = { 0, 1, 8, 9 };
  const int shift_32_uv[] = { 0, 2, 8, 10 };
  const int shift_16_uv[] = { 0, 1, 4, 5 };
Jingning Han's avatar
Jingning Han committed
855
  int i;
clang-format's avatar
clang-format committed
856
  const int max_rows =
857 858
      (mi_row + MAX_MIB_SIZE > cm->mi_rows ? cm->mi_rows - mi_row
                                           : MAX_MIB_SIZE);
clang-format's avatar
clang-format committed
859
  const int max_cols =
860 861
      (mi_col + MAX_MIB_SIZE > cm->mi_cols ? cm->mi_cols - mi_col
                                           : MAX_MIB_SIZE);
Jingning Han's avatar
Jingning Han committed
862

863
  av1_zero(*lfm);
Jingning Han's avatar
Jingning Han committed
864 865 866 867 868 869
  assert(mip[0] != NULL);

  // TODO(jimbankoski): Try moving most of the following code into decode
  // loop and storing lfm in the mbmi structure so that we don't have to go
  // through the recursive loop structure multiple times.
  switch (mip[0]->mbmi.sb_type) {
clang-format's avatar
clang-format committed
870
    case BLOCK_64X64: build_masks(lfi_n, mip[0], 0, 0, lfm); break;
Jingning Han's avatar
Jingning Han committed
871 872 873
    case BLOCK_64X32:
      build_masks(lfi_n, mip[0], 0, 0, lfm);
      mip2 = mip + mode_info_stride * 4;
clang-format's avatar
clang-format committed
874
      if (4 >= max_rows) break;
Jingning Han's avatar
Jingning Han committed
875 876 877 878 879
      build_masks(lfi_n, mip2[0], 32, 8, lfm);
      break;
    case BLOCK_32X64:
      build_masks(lfi_n, mip[0], 0, 0, lfm);
      mip2 = mip + 4;
clang-format's avatar
clang-format committed
880
      if (4 >= max_cols) break;
Jingning Han's avatar
Jingning Han committed
881 882 883 884
      build_masks(lfi_n, mip2[0], 4, 2, lfm);
      break;
    default:
      for (idx_32 = 0; idx_32 < 4; mip += offset_32[idx_32], ++idx_32) {
885 886
        const int shift_y_32 = shift_32_y[idx_32];
        const int shift_uv_32 = shift_32_uv[idx_32];
Jingning Han's avatar
Jingning Han committed
887 888 889 890 891 892
        const int mi_32_col_offset = ((idx_32 & 1) << 2);
        const int mi_32_row_offset = ((idx_32 >> 1) << 2);
        if (mi_32_col_offset >= max_cols || mi_32_row_offset >= max_rows)
          continue;
        switch (mip[0]->mbmi.sb_type) {
          case BLOCK_32X32:
893
            build_masks(lfi_n, mip[0], shift_y_32, shift_uv_32, lfm);
Jingning Han's avatar
Jingning Han committed
894 895
            break;
          case BLOCK_32X16:
896
            build_masks(lfi_n, mip[0], shift_y_32, shift_uv_32, lfm);
clang-format's avatar
clang-format committed
897
            if (mi_32_row_offset + 2 >= max_rows) continue;
Jingning Han's avatar
Jingning Han committed
898
            mip2 = mip + mode_info_stride * 2;
899
            build_masks(lfi_n, mip2[0], shift_y_32 + 16, shift_uv_32 + 4, lfm);
Jingning Han's avatar
Jingning Han committed
900 901
            break;
          case BLOCK_16X32:
902
            build_masks(lfi_n, mip[0], shift_y_32, shift_uv_32, lfm);
clang-format's avatar
clang-format committed
903
            if (mi_32_col_offset + 2 >= max_cols) continue;
Jingning Han's avatar
Jingning Han committed
904
            mip2 = mip + 2;
905
            build_masks(lfi_n, mip2[0], shift_y_32 + 2, shift_uv_32 + 1, lfm);
Jingning Han's avatar
Jingning Han committed
906 907 908
            break;
          default:
            for (idx_16 = 0; idx_16 < 4; mip += offset_16[idx_16], ++idx_16) {
909 910
              const int shift_y_32_16 = shift_y_32 + shift_16_y[idx_16];
              const int shift_uv_32_16 = shift_uv_32 + shift_16_uv[idx_16];
clang-format's avatar
clang-format committed
911 912 913 914
              const int mi_16_col_offset =
                  mi_32_col_offset + ((idx_16 & 1) << 1);
              const int mi_16_row_offset =
                  mi_32_row_offset + ((idx_16 >> 1) << 1);
Jingning Han's avatar
Jingning Han committed
915 916 917 918 919 920

              if (mi_16_col_offset >= max_cols || mi_16_row_offset >= max_rows)
                continue;

              switch (mip[0]->mbmi.sb_type) {
                case BLOCK_16X16:
921 922
                  build_masks(lfi_n, mip[0], shift_y_32_16, shift_uv_32_16,
                              lfm);
Jingning Han's avatar
Jingning Han committed
923 924
                  break;
                case BLOCK_16X8:
925 926
                  build_masks(lfi_n, mip[0], shift_y_32_16, shift_uv_32_16,
                              lfm);
clang-format's avatar
clang-format committed
927
                  if (mi_16_row_offset + 1 >= max_rows) continue;
Jingning Han's avatar
Jingning Han committed
928
                  mip2 = mip + mode_info_stride;
929
                  build_y_mask(lfi_n, mip2[0], shift_y_32_16 + 8, lfm);
Jingning Han's avatar
Jingning Han committed
930 931
                  break;
                case BLOCK_8X16:
932 933
                  build_masks(lfi_n, mip[0], shift_y_32_16, shift_uv_32_16,
                              lfm);
clang-format's avatar
clang-format committed
934
                  if (mi_16_col_offset + 1 >= max_cols) continue;
Jingning Han's avatar
Jingning Han committed
935
                  mip2 = mip + 1;
936
                  build_y_mask(lfi_n, mip2[0], shift_y_32_16 + 1, lfm);
Jingning Han's avatar
Jingning Han committed
937 938
                  break;
                default: {
939 940 941
                  const int shift_y_32_16_8_zero = shift_y_32_16 + shift_8_y[0];
                  build_masks(lfi_n, mip[0], shift_y_32_16_8_zero,
                              shift_uv_32_16, lfm);
Jingning Han's avatar
Jingning Han committed
942 943
                  mip += offset[0];
                  for (idx_8 = 1; idx_8 < 4; mip += offset[idx_8], ++idx_8) {
944 945
                    const int shift_y_32_16_8 =
                        shift_y_32_16 + shift_8_y[idx_8];
clang-format's avatar
clang-format committed
946 947 948 949
                    const int mi_8_col_offset =
                        mi_16_col_offset + ((idx_8 & 1));
                    const int mi_8_row_offset =
                        mi_16_row_offset + ((idx_8 >> 1));
Jingning Han's avatar
Jingning Han committed
950 951 952 953

                    if (mi_8_col_offset >= max_cols ||
                        mi_8_row_offset >= max_rows)
                      continue;
954
                    build_y_mask(lfi_n, mip[0], shift_y_32_16_8, lfm);
Jingning Han's avatar
Jingning Han committed
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
                  }
                  break;
                }
              }
            }
            break;
        }
      }
      break;
  }
  // The largest loopfilter we have is 16x16 so we use the 16x16 mask
  // for 32x32 transforms also.
  lfm->left_y[TX_16X16] |= lfm->left_y[TX_32X32];
  lfm->above_y[TX_16X16] |= lfm->above_y[TX_32X32];
  lfm->left_uv[TX_16X16] |= lfm->left_uv[TX_32X32];
  lfm->above_uv[TX_16X16] |= lfm->above_uv[TX_32X32];

  // We do at least 8 tap filter on every 32x32 even if the transform size
  // is 4x4. So if the 4x4 is set on a border pixel add it to the 8x8 and
  // remove it from the 4x4.
  lfm->left_y[TX_8X8] |= lfm->left_y[TX_4X4] & left_border;
  lfm->left_y[TX_4X4] &= ~left_border;
  lfm->above_y[TX_8X8] |= lfm->above_y[TX_4X4] & above_border;
  lfm->above_y[TX_4X4] &= ~above_border;
  lfm->left_uv[TX_8X8] |= lfm->left_uv[TX_4X4] & left_border_uv;
  lfm->left_uv[TX_4X4] &= ~left_border_uv;
  lfm->above_uv[TX_8X8] |= lfm->above_uv[TX_4X4] & above_border_uv;
  lfm->above_uv[TX_4X4] &= ~above_border_uv;

  // We do some special edge handling.
985
  if (mi_row + MAX_MIB_SIZE > cm->mi_rows) {
Jingning Han's avatar
Jingning Han committed
986 987 988
    const uint64_t rows = cm->mi_rows - mi_row;

    // Each pixel inside the border gets a 1,
clang-format's avatar
clang-format committed
989 990
    const uint64_t mask_y = (((uint64_t)1 << (rows << 3)) - 1);
    const uint16_t mask_uv = (((uint16_t)1 << (((rows + 1) >> 1) << 2)) - 1);
Jingning Han's avatar
Jingning Han committed
991 992 993 994 995 996 997 998 999

    // Remove values completely outside our border.
    for (i = 0; i < TX_32X32; i++) {
      lfm->left_y[i] &= mask_y;
      lfm->above_y[i] &= mask_y;
      lfm->left_uv[i] &= mask_uv;
      lfm->above_uv[i] &= mask_uv;
    }
    lfm->int_4x4_y &= mask_y;
1000 1001 1002
#if CONFIG_MISC_FIXES
    lfm->above_int_4x4_uv = lfm->left_int_4x4_uv & mask_uv;
#else
Jingning Han's avatar
Jingning Han committed
1003
    lfm->int_4x4_uv &= mask_uv;
1004
#endif
Jingning Han's avatar
Jingning Han committed
1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017

    // We don't apply a wide loop filter on the last uv block row. If set
    // apply the shorter one instead.
    if (rows == 1) {
      lfm->above_uv[TX_8X8] |= lfm->above_uv[TX_16X16];
      lfm->above_uv[TX_16X16] = 0;
    }
    if (rows == 5) {
      lfm->above_uv[TX_8X8] |= lfm->above_uv[TX_16X16] & 0xff00;
      lfm->above_uv[TX_16X16] &= ~(lfm->above_uv[TX_16X16] & 0xff00);
    }
  }

1018
  if (mi_col + MAX_MIB_SIZE > cm->mi_cols) {
Jingning Han's avatar
Jingning Han committed
1019 1020 1021 1022
    const uint64_t columns = cm->mi_cols - mi_col;

    // Each pixel inside the border gets a 1, the multiply copies the border
    // to where we need it.
clang-format's avatar
clang-format committed
1023
    const uint64_t mask_y = (((1 << columns) - 1)) * 0x0101010101010101ULL;
Jingning Han's avatar
Jingning Han committed
1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037
    const uint16_t mask_uv = ((1 << ((columns + 1) >> 1)) - 1) * 0x1111;

    // Internal edges are not applied on the last column of the image so
    // we mask 1 more for the internal edges
    const uint16_t mask_uv_int = ((1 << (columns >> 1)) - 1) * 0x1111;

    // Remove the bits outside the image edge.
    for (i = 0; i < TX_32X32; i++) {
      lfm->left_y[i] &= mask_y;
      lfm->above_y[i] &= mask_y;
      lfm->left_uv[i] &= mask_uv;
      lfm->above_uv[i] &= mask_uv;
    }
    lfm->int_4x4_y &= mask_y;
1038 1039 1040
#if CONFIG_MISC_FIXES
    lfm->left_int_4x4_uv &= mask_uv_int;
#else
Jingning Han's avatar
Jingning Han committed
1041
    lfm->int_4x4_uv &= mask_uv_int;
1042
#endif
Jingning Han's avatar
Jingning Han committed
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

    // We don't apply a wide loop filter on the last uv column. If set
    // apply the shorter one instead.
    if (columns == 1) {
      lfm->left_uv[TX_8X8] |= lfm->left_uv[TX_16X16];
      lfm->left_uv[TX_16X16] = 0;
    }
    if (columns == 5) {
      lfm->left_uv[TX_8X8] |= (lfm->left_uv[TX_16X16] & 0xcccc);
      lfm->left_uv[TX_16X16] &= ~(lfm->left_uv[TX_16X16] & 0xcccc);
    }
  }
  // We don't apply a loop filter on the first column in the image, mask that
  // out.
  if (mi_col == 0) {
    for (i = 0; i < TX_32X32; i++) {
      lfm->left_y[i] &= 0xfefefefefefefefeULL;
      lfm->left_uv[i] &= 0xeeee;
    }
  }

  // Assert if we try to apply 2 different loop filters at the same position.
  assert(!(lfm->left_y[TX_16X16] & lfm->left_y[TX_8X8]));
  assert(!(lfm->left_y[TX_16X16] & lfm->left_y[TX_4X4]));
  assert(!(lfm->left_y[TX_8X8] & lfm->left_y[TX_4X4]));
  assert(!(lfm->int_4x4_y & lfm->left_y[TX_16X16]));
clang-format's avatar
clang-format committed
1069
  assert(!(lfm->left_uv[TX_16X16] & lfm->left_uv[TX_8X8]));
Jingning Han's avatar
Jingning Han committed
1070 1071
  assert(!(lfm->left_uv[TX_16X16] & lfm->left_uv[TX_4X4]));
  assert(!(lfm->left_uv[TX_8X8] & lfm->left_uv[TX_4X4]));
1072 1073 1074
#if CONFIG_MISC_FIXES
  assert(!(lfm->left_int_4x4_uv & lfm->left_uv[TX_16X16]));
#else
Jingning Han's avatar
Jingning Han committed
1075
  assert(!(lfm->int_4x4_uv & lfm->left_uv[TX_16X16]));
1076
#endif
Jingning Han's avatar
Jingning Han committed
1077 1078 1079 1080 1081 1082 1083
  assert(!(lfm->above_y[TX_16X16] & lfm->above_y[TX_8X8]));
  assert(!(lfm->above_y[TX_16X16] & lfm->above_y[TX_4X4]));
  assert(!(lfm->above_y[TX_8X8] & lfm->above_y[TX_4X4]));
  assert(!(lfm->int_4x4_y & lfm->above_y[TX_16X16]));
  assert(!(lfm->above_uv[TX_16X16] & lfm->above_uv[TX_8X8]));
  assert(!(lfm->above_uv[TX_16X16] & lfm->above_uv[TX_4X4]));
  assert(!(lfm->above_uv[TX_8X8] & lfm->above_uv[TX_4X4]));
1084 1085 1086
#if CONFIG_MISC_FIXES
  assert(!(lfm->above_int_4x4_uv & lfm->above_uv[TX_16X16]));
#else
Jingning Han's avatar
Jingning Han committed
1087
  assert(!(lfm->int_4x4_uv & lfm->above_uv[TX_16X16]));
1088
#endif
Jingning Han's avatar
Jingning Han committed
1089 1090
}

clang-format's avatar
clang-format committed
1091 1092 1093 1094
static void filter_selectively_vert(
    uint8_t *s, int pitch, unsigned int mask_16x16, unsigned int mask_8x8,
    unsigned int mask_4x4, unsigned int mask_4x4_int,
    const loop_filter_info_n *lfi_n, const uint8_t *lfl) {
Jingning Han's avatar
Jingning Han committed
1095 1096
  unsigned int mask;

clang-format's avatar
clang-format committed
1097 1098
  for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; mask;
       mask >>= 1) {
Jingning Han's avatar
Jingning Han committed
1099 1100 1101 1102
    const loop_filter_thresh *lfi = lfi_n->lfthr + *lfl;

    if (mask & 1) {
      if (mask_16x16 & 1) {
Adrian Grange's avatar
Adrian Grange committed
1103
        aom_lpf_vertical_16(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
1104
      } else if (mask_8x8 & 1) {
Adrian Grange's avatar
Adrian Grange committed
1105
        aom_lpf_vertical_8(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
Jingning Han's avatar
Jingning Han committed
1106
      } else if (mask_4x4 & 1) {
Adrian Grange's avatar
Adrian Grange committed
1107
        aom_lpf_vertical_4(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
Jingning Han's avatar
Jingning Han committed
1108 1109 1110
      }
    }
    if (mask_4x4_int & 1)
Adrian Grange's avatar
Adrian Grange committed
1111
      aom_lpf_vertical_4(s + 4, pitch, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
Jingning Han's avatar
Jingning Han committed
1112 1113 1114 1115 1116 1117 1118 1119 1120
    s += 8;
    lfl += 1;
    mask_16x16 >>= 1;
    mask_8x8 >>= 1;
    mask_4x4 >>= 1;
    mask_4x4_int >>= 1;
  }
}

1121
#if CONFIG_AOM_HIGHBITDEPTH
clang-format's avatar
clang-format committed
1122 1123 1124 1125
static void highbd_filter_selectively_vert(
    uint16_t *s, int pitch, unsigned int mask_16x16, unsigned int mask_8x8,
    unsigned int mask_4x4, unsigned int mask_4x4_int,
    const loop_filter_info_n *lfi_n, const uint8_t *lfl, int bd) {
Jingning Han's avatar
Jingning Han committed
1126 1127
  unsigned int mask;

clang-format's avatar
clang-format committed
1128 1129
  for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; mask;
       mask >>= 1) {
Jingning Han's avatar
Jingning Han committed
1130 1131 1132 1133
    const loop_filter_thresh *lfi = lfi_n->lfthr + *lfl;

    if (mask & 1) {
      if (mask_16x16 & 1) {
Adrian Grange's avatar
Adrian Grange committed
1134
        aom_highbd_lpf_vertical_16(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr,
clang-format's avatar
clang-format committed
1135
                                   bd);
Jingning Han's avatar
Jingning Han committed
1136
      } else if (mask_8x8 & 1) {
Adrian Grange's avatar
Adrian Grange committed
1137
        aom_highbd_lpf_vertical_8(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr,
clang-format's avatar
clang-format committed
1138
                                  1, bd);
Jingning Han's avatar
Jingning Han committed
1139
      } else if (mask_4x4 & 1) {
Adrian Grange's avatar
Adrian Grange committed
1140
        aom_highbd_lpf_vertical_4(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr,
clang-format's avatar
clang-format committed
1141
                                  1, bd);
Jingning Han's avatar
Jingning Han committed
1142 1143 1144
      }
    }
    if (mask_4x4_int & 1)
Adrian Grange's avatar
Adrian Grange committed
1145
      aom_highbd_lpf_vertical_4(s + 4, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
1146 1147 1148 1149 1150 1151 1152 1153 1154
                                lfi->hev_thr, 1, bd);
    s += 8;
    lfl += 1;
    mask_16x16 >>= 1;
    mask_8x8 >>= 1;
    mask_4x4 >>= 1;
    mask_4x4_int >>= 1;
  }
}
1155
#endif  // CONFIG_AOM_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
1156

1157 1158 1159 1160
void av1_filter_block_plane_non420_ver(AV1_COMMON *cm,
                                       struct macroblockd_plane *plane,
                                       MODE_INFO **mi_8x8, int mi_row,
                                       int mi_col) {
Jingning Han's avatar
Jingning Han committed
1161 1162 1163 1164 1165 1166
  const int ss_x = plane->subsampling_x;
  const int ss_y = plane->subsampling_y;
  const int row_step = 1 << ss_y;
  const int col_step = 1 << ss_x;
  const int row_step_stride = cm->mi_stride * row_step;
  struct buf_2d *const dst = &plane->dst;
clang-format's avatar
clang-format committed
1167
  uint8_t *const dst0 = dst->buf;
1168 1169 1170 1171 1172
  unsigned int mask_16x16[MAX_MIB_SIZE] = { 0 };
  unsigned int mask_8x8[MAX_MIB_SIZE] = { 0 };
  unsigned int mask_4x4[MAX_MIB_SIZE] = { 0 };
  unsigned int mask_4x4_int[MAX_MIB_SIZE] = { 0 };
  uint8_t lfl[MAX_MIB_SIZE * MAX_MIB_SIZE];
Jingning Han's avatar
Jingning Han committed
1173 1174
  int r, c;

1175
  for (r = 0; r < MAX_MIB_SIZE && mi_row + r < cm->mi_rows; r += row_step) {