loopfilter.c 63.5 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
#include "./aom_config.h"
#include "./aom_dsp_rtcd.h"
16 17 18 19
#include "av1/common/loopfilter.h"
#include "av1/common/onyxc_int.h"
#include "av1/common/reconinter.h"
#include "av1/common/restoration.h"
Yaowu Xu's avatar
Yaowu Xu committed
20 21
#include "aom_dsp/aom_dsp_common.h"
#include "aom_mem/aom_mem.h"
22
#include "aom_ports/mem.h"
Jingning Han's avatar
Jingning Han committed
23

24
#include "av1/common/seg_common.h"
Jingning Han's avatar
Jingning Han committed
25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42

// 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
43
static const uint64_t left_64x64_txform_mask[TX_SIZES] = {
Jingning Han's avatar
Jingning Han committed
44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66
  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
67
static const uint64_t above_64x64_txform_mask[TX_SIZES] = {
Jingning Han's avatar
Jingning Han committed
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
  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
141
static const uint64_t left_border = 0x1111111111111111ULL;
Jingning Han's avatar
Jingning Han committed
142 143 144
static const uint64_t above_border = 0x000000ff000000ffULL;

// 16 bit masks for uv transform sizes.
clang-format's avatar
clang-format committed
145
static const uint16_t left_64x64_txform_mask_uv[TX_SIZES] = {
Jingning Han's avatar
Jingning Han committed
146 147 148 149 150 151
  0xffff,  // TX_4X4
  0xffff,  // TX_8x8
  0x5555,  // TX_16x16
  0x1111,  // TX_32x32
};

clang-format's avatar
clang-format committed
152
static const uint16_t above_64x64_txform_mask_uv[TX_SIZES] = {
Jingning Han's avatar
Jingning Han committed
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
  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
208
static const uint16_t left_border_uv = 0x1111;
Jingning Han's avatar
Jingning Han committed
209 210 211 212 213
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)
Yue Chen's avatar
Yue Chen committed
214
#if CONFIG_EXT_INTER
clang-format's avatar
clang-format committed
215 216 217 218
  ,
  1,                            // NEWFROMNEARMV mode
  1, 1, 1, 1, 1, 1, 1, 1, 0, 1  // INTER_COMPOUND_MODES (ZERO_ZEROMV == 0)
#endif                          // CONFIG_EXT_INTER
Jingning Han's avatar
Jingning Han committed
219 220 221 222 223 224 225 226 227 228 229 230 231 232 233
};

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
234
    if (block_inside_limit < 1) block_inside_limit = 1;
Jingning Han's avatar
Jingning Han committed
235 236 237 238 239 240 241 242 243

    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) {
244
#if CONFIG_SUPERTX
Yaowu Xu's avatar
Yaowu Xu committed
245
  const int segment_id = AOMMIN(mbmi->segment_id, mbmi->segment_id_supertx);
clang-format's avatar
clang-format committed
246 247
  assert(
      IMPLIES(supertx_enabled(mbmi), mbmi->segment_id_supertx != MAX_SEGMENTS));
248 249 250 251 252 253
  assert(IMPLIES(supertx_enabled(mbmi),
                 mbmi->segment_id_supertx <= mbmi->segment_id));
#else
  const int segment_id = mbmi->segment_id;
#endif  // CONFIG_SUPERTX
  return lfi_n->lvl[segment_id][mbmi->ref_frame[0]][mode_lf_lut[mbmi->mode]];
Jingning Han's avatar
Jingning Han committed
254 255
}

Yaowu Xu's avatar
Yaowu Xu committed
256
void av1_loop_filter_init(AV1_COMMON *cm) {
Jingning Han's avatar
Jingning Han committed
257 258 259 260 261 262 263 264 265 266 267 268 269
  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);
}

Yaowu Xu's avatar
Yaowu Xu committed
270
void av1_loop_filter_frame_init(AV1_COMMON *cm, int default_filt_lvl) {
Jingning Han's avatar
Jingning Han committed
271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289
  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
290 291 292
      lvl_seg = clamp(
          seg->abs_delta == SEGMENT_ABSDATA ? data : default_filt_lvl + data, 0,
          MAX_LOOP_FILTER);
Jingning Han's avatar
Jingning Han committed
293 294 295 296 297 298 299 300 301 302 303
    }

    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);

304
      for (ref = LAST_FRAME; ref < TOTAL_REFS_PER_FRAME; ++ref) {
Jingning Han's avatar
Jingning Han committed
305
        for (mode = 0; mode < MAX_MODE_LF_DELTAS; ++mode) {
clang-format's avatar
clang-format committed
306 307
          const int inter_lvl = lvl_seg + lf->ref_deltas[ref] * scale +
                                lf->mode_deltas[mode] * scale;
Jingning Han's avatar
Jingning Han committed
308 309 310 311 312 313 314
          lfi->lvl[seg_id][ref][mode] = clamp(inter_lvl, 0, MAX_LOOP_FILTER);
        }
      }
    }
  }
}

clang-format's avatar
clang-format committed
315 316
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
317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344
                                         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
345
          aom_lpf_vertical_16_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
346 347
                                   lfi0->hev_thr);
        } else if (mask_16x16_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
348
          aom_lpf_vertical_16(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr);
Jingning Han's avatar
Jingning Han committed
349
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
350
          aom_lpf_vertical_16(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim,
clang-format's avatar
clang-format committed
351
                              lfi1->hev_thr);
Jingning Han's avatar
Jingning Han committed
352 353 354 355 356
        }
      }

      if ((mask_8x8_0 | mask_8x8_1) & 1) {
        if ((mask_8x8_0 & mask_8x8_1) & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
357
          aom_lpf_vertical_8_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
358 359 360
                                  lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                  lfi1->hev_thr);
        } else if (mask_8x8_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
361
          aom_lpf_vertical_8(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr);
Jingning Han's avatar
Jingning Han committed
362
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
363
          aom_lpf_vertical_8(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim,
364
                             lfi1->hev_thr);
Jingning Han's avatar
Jingning Han committed
365 366 367 368 369
        }
      }

      if ((mask_4x4_0 | mask_4x4_1) & 1) {
        if ((mask_4x4_0 & mask_4x4_1) & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
370
          aom_lpf_vertical_4_dual(s, 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_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
374
          aom_lpf_vertical_4(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr);
Jingning Han's avatar
Jingning Han committed
375
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
376
          aom_lpf_vertical_4(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim,
377
                             lfi1->hev_thr);
Jingning Han's avatar
Jingning Han committed
378 379 380 381 382
        }
      }

      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
383
          aom_lpf_vertical_4_dual(s + 4, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
384 385 386
                                  lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                  lfi1->hev_thr);
        } else if (mask_4x4_int_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
387
          aom_lpf_vertical_4(s + 4, pitch, lfi0->mblim, lfi0->lim,
388
                             lfi0->hev_thr);
Jingning Han's avatar
Jingning Han committed
389
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
390
          aom_lpf_vertical_4(s + 8 * pitch + 4, pitch, lfi1->mblim, lfi1->lim,
391
                             lfi1->hev_thr);
Jingning Han's avatar
Jingning Han committed
392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408
        }
      }
    }

    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;
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
409
#if CONFIG_AOM_HIGHBITDEPTH
clang-format's avatar
clang-format committed
410 411 412 413 414
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
415 416 417 418 419 420 421 422 423 424 425 426 427 428 429
  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
430
              mask_16x16_1 | mask_8x8_1 | mask_4x4_1 | mask_4x4_int_1;
Jingning Han's avatar
Jingning Han committed
431 432 433 434 435 436 437
       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
438
          aom_highbd_lpf_vertical_16_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
439 440
                                          lfi0->hev_thr, bd);
        } else if (mask_16x16_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
441
          aom_highbd_lpf_vertical_16(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
442 443
                                     lfi0->hev_thr, bd);
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
444
          aom_highbd_lpf_vertical_16(s + 8 * pitch, pitch, lfi1->mblim,
Jingning Han's avatar
Jingning Han committed
445 446 447 448 449 450
                                     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
451
          aom_highbd_lpf_vertical_8_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
452 453 454
                                         lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                         lfi1->hev_thr, bd);
        } else if (mask_8x8_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
455
          aom_highbd_lpf_vertical_8(s, pitch, lfi0->mblim, lfi0->lim,
456
                                    lfi0->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
457
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
458
          aom_highbd_lpf_vertical_8(s + 8 * pitch, pitch, lfi1->mblim,
459
                                    lfi1->lim, lfi1->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
460 461 462 463 464
        }
      }

      if ((mask_4x4_0 | mask_4x4_1) & 1) {
        if ((mask_4x4_0 & mask_4x4_1) & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
465
          aom_highbd_lpf_vertical_4_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
466 467 468
                                         lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                         lfi1->hev_thr, bd);
        } else if (mask_4x4_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
469
          aom_highbd_lpf_vertical_4(s, pitch, lfi0->mblim, lfi0->lim,
470
                                    lfi0->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
471
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
472
          aom_highbd_lpf_vertical_4(s + 8 * pitch, pitch, lfi1->mblim,
473
                                    lfi1->lim, lfi1->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
474 475 476 477 478
        }
      }

      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
479
          aom_highbd_lpf_vertical_4_dual(s + 4, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
480 481 482
                                         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
483
          aom_highbd_lpf_vertical_4(s + 4, pitch, lfi0->mblim, lfi0->lim,
484
                                    lfi0->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
485
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
486
          aom_highbd_lpf_vertical_4(s + 8 * pitch + 4, pitch, lfi1->mblim,
487
                                    lfi1->lim, lfi1->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503
        }
      }
    }

    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;
  }
}
Yaowu Xu's avatar
Yaowu Xu committed
504
#endif  // CONFIG_AOM_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
505

clang-format's avatar
clang-format committed
506 507 508 509
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
510 511 512
  unsigned int mask;
  int count;

clang-format's avatar
clang-format committed
513 514
  for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; mask;
       mask >>= count) {
Jingning Han's avatar
Jingning Han committed
515 516 517 518 519 520
    const loop_filter_thresh *lfi = lfi_n->lfthr + *lfl;

    count = 1;
    if (mask & 1) {
      if (mask_16x16 & 1) {
        if ((mask_16x16 & 3) == 3) {
Yaowu Xu's avatar
Yaowu Xu committed
521
          aom_lpf_horizontal_edge_16(s, pitch, lfi->mblim, lfi->lim,
522
                                     lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
523 524
          count = 2;
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
525
          aom_lpf_horizontal_edge_8(s, pitch, lfi->mblim, lfi->lim,
526
                                    lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
527 528 529 530 531 532
        }
      } else if (mask_8x8 & 1) {
        if ((mask_8x8 & 3) == 3) {
          // Next block's thresholds.
          const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + 1);

Yaowu Xu's avatar
Yaowu Xu committed
533
          aom_lpf_horizontal_8_dual(s, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
534 535 536 537
                                    lfi->hev_thr, lfin->mblim, lfin->lim,
                                    lfin->hev_thr);

          if ((mask_4x4_int & 3) == 3) {
Yaowu Xu's avatar
Yaowu Xu committed
538
            aom_lpf_horizontal_4_dual(s + 4 * pitch, pitch, lfi->mblim,
Jingning Han's avatar
Jingning Han committed
539 540 541 542
                                      lfi->lim, lfi->hev_thr, lfin->mblim,
                                      lfin->lim, lfin->hev_thr);
          } else {
            if (mask_4x4_int & 1)
Yaowu Xu's avatar
Yaowu Xu committed
543
              aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
544
                                   lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
545
            else if (mask_4x4_int & 2)
Yaowu Xu's avatar
Yaowu Xu committed
546
              aom_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
547
                                   lfin->lim, lfin->hev_thr);
Jingning Han's avatar
Jingning Han committed
548 549 550
          }
          count = 2;
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
551
          aom_lpf_horizontal_8(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
552 553

          if (mask_4x4_int & 1)
Yaowu Xu's avatar
Yaowu Xu committed
554
            aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
555
                                 lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
556 557 558 559 560 561
        }
      } else if (mask_4x4 & 1) {
        if ((mask_4x4 & 3) == 3) {
          // Next block's thresholds.
          const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + 1);

Yaowu Xu's avatar
Yaowu Xu committed
562
          aom_lpf_horizontal_4_dual(s, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
563 564 565
                                    lfi->hev_thr, lfin->mblim, lfin->lim,
                                    lfin->hev_thr);
          if ((mask_4x4_int & 3) == 3) {
Yaowu Xu's avatar
Yaowu Xu committed
566
            aom_lpf_horizontal_4_dual(s + 4 * pitch, pitch, lfi->mblim,
Jingning Han's avatar
Jingning Han committed
567 568 569 570
                                      lfi->lim, lfi->hev_thr, lfin->mblim,
                                      lfin->lim, lfin->hev_thr);
          } else {
            if (mask_4x4_int & 1)
Yaowu Xu's avatar
Yaowu Xu committed
571
              aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
572
                                   lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
573
            else if (mask_4x4_int & 2)
Yaowu Xu's avatar
Yaowu Xu committed
574
              aom_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
575
                                   lfin->lim, lfin->hev_thr);
Jingning Han's avatar
Jingning Han committed
576 577 578
          }
          count = 2;
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
579
          aom_lpf_horizontal_4(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
580 581

          if (mask_4x4_int & 1)
Yaowu Xu's avatar
Yaowu Xu committed
582
            aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
583
                                 lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
584 585
        }
      } else if (mask_4x4_int & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
586
        aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
587
                             lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
588 589 590 591 592 593 594 595 596 597 598
      }
    }
    s += 8 * count;
    lfl += count;
    mask_16x16 >>= count;
    mask_8x8 >>= count;
    mask_4x4 >>= count;
    mask_4x4_int >>= count;
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
599
#if CONFIG_AOM_HIGHBITDEPTH
clang-format's avatar
clang-format committed
600 601 602 603
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
604 605 606
  unsigned int mask;
  int count;

clang-format's avatar
clang-format committed
607 608
  for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; mask;
       mask >>= count) {
Jingning Han's avatar
Jingning Han committed
609 610 611 612 613 614
    const loop_filter_thresh *lfi = lfi_n->lfthr + *lfl;

    count = 1;
    if (mask & 1) {
      if (mask_16x16 & 1) {
        if ((mask_16x16 & 3) == 3) {
Yaowu Xu's avatar
Yaowu Xu committed
615
          aom_highbd_lpf_horizontal_edge_16(s, pitch, lfi->mblim, lfi->lim,
616
                                            lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
617 618
          count = 2;
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
619
          aom_highbd_lpf_horizontal_edge_8(s, pitch, lfi->mblim, lfi->lim,
620
                                           lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
621 622 623 624 625 626
        }
      } else if (mask_8x8 & 1) {
        if ((mask_8x8 & 3) == 3) {
          // Next block's thresholds.
          const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + 1);

Yaowu Xu's avatar
Yaowu Xu committed
627
          aom_highbd_lpf_horizontal_8_dual(s, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
628 629 630 631
                                           lfi->hev_thr, lfin->mblim, lfin->lim,
                                           lfin->hev_thr, bd);

          if ((mask_4x4_int & 3) == 3) {
Yaowu Xu's avatar
Yaowu Xu committed
632
            aom_highbd_lpf_horizontal_4_dual(
clang-format's avatar
clang-format committed
633 634
                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
635 636
          } else {
            if (mask_4x4_int & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
637
              aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
638
                                          lfi->lim, lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
639
            } else if (mask_4x4_int & 2) {
Yaowu Xu's avatar
Yaowu Xu committed
640
              aom_highbd_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
641
                                          lfin->lim, lfin->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
642 643 644 645
            }
          }
          count = 2;
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
646
          aom_highbd_lpf_horizontal_8(s, pitch, lfi->mblim, lfi->lim,
647
                                      lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
648 649

          if (mask_4x4_int & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
650
            aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
651
                                        lfi->lim, lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
652 653 654 655 656 657 658
          }
        }
      } else if (mask_4x4 & 1) {
        if ((mask_4x4 & 3) == 3) {
          // Next block's thresholds.
          const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + 1);

Yaowu Xu's avatar
Yaowu Xu committed
659
          aom_highbd_lpf_horizontal_4_dual(s, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
660 661 662
                                           lfi->hev_thr, lfin->mblim, lfin->lim,
                                           lfin->hev_thr, bd);
          if ((mask_4x4_int & 3) == 3) {
Yaowu Xu's avatar
Yaowu Xu committed
663
            aom_highbd_lpf_horizontal_4_dual(
clang-format's avatar
clang-format committed
664 665
                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
666 667
          } else {
            if (mask_4x4_int & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
668
              aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
669
                                          lfi->lim, lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
670
            } else if (mask_4x4_int & 2) {
Yaowu Xu's avatar
Yaowu Xu committed
671
              aom_highbd_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
672
                                          lfin->lim, lfin->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
673 674 675 676
            }
          }
          count = 2;
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
677
          aom_highbd_lpf_horizontal_4(s, pitch, lfi->mblim, lfi->lim,
678
                                      lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
679 680

          if (mask_4x4_int & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
681
            aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
682
                                        lfi->lim, lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
683 684 685
          }
        }
      } else if (mask_4x4_int & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
686
        aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
687
                                    lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
688 689 690 691 692 693 694 695 696 697
      }
    }
    s += 8 * count;
    lfl += count;
    mask_16x16 >>= count;
    mask_8x8 >>= count;
    mask_4x4 >>= count;
    mask_4x4_int >>= count;
  }
}
Yaowu Xu's avatar
Yaowu Xu committed
698
#endif  // CONFIG_AOM_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
699 700 701 702 703 704 705 706 707 708

// 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
709
                        const int shift_uv, LOOP_FILTER_MASK *lfm) {
Jingning Han's avatar
Jingning Han committed
710 711
  const MB_MODE_INFO *mbmi = &mi->mbmi;
  const BLOCK_SIZE block_size = mbmi->sb_type;
712 713
  // TODO(debargha): Check if masks can be setup correctly when
  // rectangular transfroms are used with the EXT_TX expt.
Debargha Mukherjee's avatar
Debargha Mukherjee committed
714
  const TX_SIZE tx_size_y = txsize_sqr_up_map[mbmi->tx_size];
715
  const TX_SIZE tx_size_uv =
716
      txsize_sqr_up_map[uv_txsize_lookup[block_size][mbmi->tx_size][1][1]];
Jingning Han's avatar
Jingning Han committed
717 718 719 720 721 722
  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];
723
  uint16_t *const int_4x4_uv = &lfm->left_int_4x4_uv;
Jingning Han's avatar
Jingning Han committed
724 725 726 727 728 729 730 731
  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];
732 733 734
    const int row = (shift_y >> MAX_MIB_SIZE_LOG2);
    const int col = shift_y - (row << MAX_MIB_SIZE_LOG2);

clang-format's avatar
clang-format committed
735
    for (i = 0; i < h; i++) memset(&lfm->lfl_y[row + i][col], filter_level, w);
Jingning Han's avatar
Jingning Han committed
736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756
  }

  // 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;

  // If the block has no coefficients and is not intra we skip applying
  // the loop filter on block edges.
clang-format's avatar
clang-format committed
757
  if ((mbmi->skip || mbmi->has_no_coeffs) && is_inter_block(mbmi)) return;
Jingning Han's avatar
Jingning Han committed
758 759 760 761 762

  // 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
763 764 765 766 767
  *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
768

clang-format's avatar
clang-format committed
769 770 771 772
  *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
773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789

  // 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,
790 791 792
#if CONFIG_SUPERTX
                         int supertx_enabled,
#endif  // CONFIG_SUPERTX
Jingning Han's avatar
Jingning Han committed
793 794
                         LOOP_FILTER_MASK *lfm) {
  const MB_MODE_INFO *mbmi = &mi->mbmi;
Debargha Mukherjee's avatar
Debargha Mukherjee committed
795
  const TX_SIZE tx_size_y = txsize_sqr_up_map[mbmi->tx_size];
796 797 798 799 800 801
#if CONFIG_SUPERTX
  const BLOCK_SIZE block_size =
      supertx_enabled ? (BLOCK_SIZE)(3 * tx_size_y) : mbmi->sb_type;
#else
  const BLOCK_SIZE block_size = mbmi->sb_type;
#endif
Jingning Han's avatar
Jingning Han committed
802 803 804 805 806 807 808 809 810 811 812
  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];
813 814 815
    const int row = (shift_y >> MAX_MIB_SIZE_LOG2);
    const int col = shift_y - (row << MAX_MIB_SIZE_LOG2);

clang-format's avatar
clang-format committed
816
    for (i = 0; i < h; i++) memset(&lfm->lfl_y[row + i][col], filter_level, w);
Jingning Han's avatar
Jingning Han committed
817 818 819 820 821
  }

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

clang-format's avatar
clang-format committed
822
  if ((mbmi->skip || mbmi->has_no_coeffs) && is_inter_block(mbmi)) return;
Jingning Han's avatar
Jingning Han committed
823

clang-format's avatar
clang-format committed
824 825
  *above_y |= (size_mask[block_size] & above_64x64_txform_mask[tx_size_y])
              << shift_y;
Jingning Han's avatar
Jingning Han committed
826

clang-format's avatar
clang-format committed
827 828
  *left_y |= (size_mask[block_size] & left_64x64_txform_mask[tx_size_y])
             << shift_y;
Jingning Han's avatar
Jingning Han committed
829 830 831 832 833 834 835 836

  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.
Yaowu Xu's avatar
Yaowu Xu committed
837 838 839
void av1_setup_mask(AV1_COMMON *const cm, const int mi_row, const int mi_col,
                    MODE_INFO **mi, const int mode_info_stride,
                    LOOP_FILTER_MASK *lfm) {
Jingning Han's avatar
Jingning Han committed
840 841 842 843 844 845 846 847 848
  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
849 850 851 852 853
  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
854 855 856 857 858

  // 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
859 860 861 862 863
  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
864
  int i;
Yaowu Xu's avatar
Yaowu Xu committed
865 866
  const int max_rows = AOMMIN(cm->mi_rows - mi_row, MAX_MIB_SIZE);
  const int max_cols = AOMMIN(cm->mi_cols - mi_col, MAX_MIB_SIZE);
867 868 869
#if CONFIG_EXT_PARTITION
  assert(0 && "Not yet updated");
#endif  // CONFIG_EXT_PARTITION
Jingning Han's avatar
Jingning Han committed
870

Yaowu Xu's avatar
Yaowu Xu committed
871
  av1_zero(*lfm);
Jingning Han's avatar
Jingning Han committed
872 873 874 875 876 877
  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
878
    case BLOCK_64X64: build_masks(lfi_n, mip[0], 0, 0, lfm); break;
Jingning Han's avatar
Jingning Han committed
879 880 881
    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
882
      if (4 >= max_rows) break;
Jingning Han's avatar
Jingning Han committed
883 884 885 886 887
      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
888
      if (4 >= max_cols) break;
Jingning Han's avatar
Jingning Han committed
889 890 891 892 893 894 895 896 897 898 899 900 901 902
      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) {
        const int shift_y = shift_32_y[idx_32];
        const int shift_uv = shift_32_uv[idx_32];
        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:
            build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
            break;
clang-format's avatar
clang-format committed
903
          case BLOCK_32X16: build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
904
#if CONFIG_SUPERTX
clang-format's avatar
clang-format committed
905
            if (supertx_enabled(&mip[0]->mbmi)) break;
906
#endif
clang-format's avatar
clang-format committed
907
            if (mi_32_row_offset + 2 >= max_rows) continue;
Jingning Han's avatar
Jingning Han committed
908 909 910
            mip2 = mip + mode_info_stride * 2;
            build_masks(lfi_n, mip2[0], shift_y + 16, shift_uv + 4, lfm);
            break;
clang-format's avatar
clang-format committed
911
          case BLOCK_16X32: build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
912
#if CONFIG_SUPERTX
clang-format's avatar
clang-format committed
913
            if (supertx_enabled(&mip[0]->mbmi)) break;
914
#endif
clang-format's avatar
clang-format committed
915
            if (mi_32_col_offset + 2 >= max_cols) continue;
Jingning Han's avatar
Jingning Han committed
916 917 918 919
            mip2 = mip + 2;
            build_masks(lfi_n, mip2[0], shift_y + 2, shift_uv + 1, lfm);
            break;
          default:
920 921 922 923 924 925
#if CONFIG_SUPERTX
            if (mip[0]->mbmi.tx_size == TX_32X32) {
              build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
              break;
            }
#endif
Jingning Han's avatar
Jingning Han committed
926 927 928
            for (idx_16 = 0; idx_16 < 4; mip += offset_16[idx_16], ++idx_16) {
              const int shift_y = shift_32_y[idx_32] + shift_16_y[idx_16];
              const int shift_uv = shift_32_uv[idx_32] + shift_16_uv[idx_16];
clang-format's avatar
clang-format committed
929 930 931 932
              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
933 934 935 936 937 938 939 940 941

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

              switch (mip[0]->mbmi.sb_type) {
                case BLOCK_16X16:
                  build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
                  break;
                case BLOCK_16X8:
942
#if CONFIG_SUPERTX
clang-format's avatar
clang-format committed
943
                  if (supertx_enabled(&mip[0]->mbmi)) break;
944
#endif
Jingning Han's avatar
Jingning Han committed
945
                  build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
clang-format's avatar
clang-format committed
946
                  if (mi_16_row_offset + 1 >= max_rows) continue;
Jingning Han's avatar
Jingning Han committed
947
                  mip2 = mip + mode_info_stride;
clang-format's avatar
clang-format committed
948
                  build_y_mask(lfi_n, mip2[0], shift_y + 8,
949 950 951 952
#if CONFIG_SUPERTX
                               0,
#endif
                               lfm);
Jingning Han's avatar
Jingning Han committed
953 954
                  break;
                case BLOCK_8X16:
955
#if CONFIG_SUPERTX
clang-format's avatar
clang-format committed
956
                  if (supertx_enabled(&mip[0]->mbmi)) break;
957
#endif
Jingning Han's avatar
Jingning Han committed
958
                  build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
clang-format's avatar
clang-format committed
959
                  if (mi_16_col_offset + 1 >= max_cols) continue;
Jingning Han's avatar
Jingning Han committed
960
                  mip2 = mip + 1;
clang-format's avatar
clang-format committed
961
                  build_y_mask(lfi_n, mip2[0], shift_y + 1,
962 963 964 965
#if CONFIG_SUPERTX
                               0,
#endif
                               lfm);
Jingning Han's avatar
Jingning Han committed
966 967
                  break;
                default: {
clang-format's avatar
clang-format committed
968 969
                  const int shift_y =
                      shift_32_y[idx_32] + shift_16_y[idx_16] + shift_8_y[0];
970 971 972 973 974 975
#if CONFIG_SUPERTX
                  if (mip[0]->mbmi.tx_size == TX_16X16) {
                    build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
                    break;
                  }
#endif
Jingning Han's avatar
Jingning Han committed
976 977 978 979
                  build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
                  mip += offset[0];
                  for (idx_8 = 1; idx_8 < 4; mip += offset[idx_8], ++idx_8) {
                    const int shift_y = shift_32_y[idx_32] +
clang-format's avatar
clang-format committed
980 981 982 983 984
                                        shift_16_y[idx_16] + shift_8_y[idx_8];
                    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
985 986 987 988

                    if (mi_8_col_offset >= max_cols ||
                        mi_8_row_offset >= max_rows)
                      continue;
989 990 991 992 993
                    build_y_mask(lfi_n, mip[0], shift_y,
#if CONFIG_SUPERTX
                                 supertx_enabled(&mip[0]->mbmi),
#endif
                                 lfm);
Jingning Han's avatar
Jingning Han committed
994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023
                  }
                  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.
1024
  if (mi_row + MAX_MIB_SIZE > cm->mi_rows) {
Jingning Han's avatar
Jingning Han committed
1025 1026 1027
    const uint64_t rows = cm->mi_rows - mi_row;

    // Each pixel inside the border gets a 1,
clang-format's avatar
clang-format committed
1028
    const uint64_t mask_y = (((uint64_t)1 << (rows << MAX_MIB_SIZE_LOG2)) - 1);
1029
    const uint16_t mask_uv =
clang-format's avatar
clang-format committed
1030
        (((uint16_t)1 << (((rows + 1) >> 1) << (MAX_MIB_SIZE_LOG2 - 1))) - 1);
Jingning Han's avatar
Jingning Han committed
1031 1032 1033 1034 1035 1036 1037 1038 1039

    // 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;
1040
    lfm->above_int_4x4_uv = lfm->left_int_4x4_uv & mask_uv;
Jingning Han's avatar
Jingning Han committed
1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053

    // 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);
    }
  }