loopfilter.c 78.2 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
#include "av1/common/loopfilter.h"
#include "av1/common/onyxc_int.h"
#include "av1/common/reconinter.h"
Yaowu Xu's avatar
Yaowu Xu committed
19 20
#include "aom_dsp/aom_dsp_common.h"
#include "aom_mem/aom_mem.h"
21
#include "aom_ports/mem.h"
Jingning Han's avatar
Jingning Han committed
22

23
#include "av1/common/seg_common.h"
Jingning Han's avatar
Jingning Han committed
24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41

// 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
42
static const uint64_t left_64x64_txform_mask[TX_SIZES] = {
43 44 45
#if CONFIG_CB4X4
  0xffffffffffffffffULL,  // TX_2X2
#endif
Jingning Han's avatar
Jingning Han committed
46 47 48 49
  0xffffffffffffffffULL,  // TX_4X4
  0xffffffffffffffffULL,  // TX_8x8
  0x5555555555555555ULL,  // TX_16x16
  0x1111111111111111ULL,  // TX_32x32
50 51 52
#if CONFIG_TX64X64
  0x0101010101010101ULL,  // TX_64x64
#endif                    // CONFIG_TX64X64
Jingning Han's avatar
Jingning Han committed
53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71
};

// 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
72
static const uint64_t above_64x64_txform_mask[TX_SIZES] = {
73 74 75
#if CONFIG_CB4X4
  0xffffffffffffffffULL,  // TX_4X4
#endif
Jingning Han's avatar
Jingning Han committed
76 77 78 79
  0xffffffffffffffffULL,  // TX_4X4
  0xffffffffffffffffULL,  // TX_8x8
  0x00ff00ff00ff00ffULL,  // TX_16x16
  0x000000ff000000ffULL,  // TX_32x32
80 81 82
#if CONFIG_TX64X64
  0x00000000000000ffULL,  // TX_64x64
#endif                    // CONFIG_TX64X64
Jingning Han's avatar
Jingning Han committed
83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100
};

// 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] = {
Jingning Han's avatar
Jingning Han committed
101 102 103 104 105
#if CONFIG_CB4X4
  0x0000000000000001ULL,  // BLOCK_2X2,
  0x0000000000000001ULL,  // BLOCK_2X4,
  0x0000000000000001ULL,  // BLOCK_4X2,
#endif
Jingning Han's avatar
Jingning Han committed
106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122
  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] = {
Jingning Han's avatar
Jingning Han committed
123 124 125 126 127
#if CONFIG_CB4X4
  0x0000000000000001ULL,  // BLOCK_2X2
  0x0000000000000001ULL,  // BLOCK_2X4
  0x0000000000000001ULL,  // BLOCK_4X2
#endif
Jingning Han's avatar
Jingning Han committed
128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145
  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] = {
Jingning Han's avatar
Jingning Han committed
146 147 148 149 150
#if CONFIG_CB4X4
  0x0000000000000001ULL,  // BLOCK_2X2
  0x0000000000000001ULL,  // BLOCK_2X4
  0x0000000000000001ULL,  // BLOCK_4X2
#endif
Jingning Han's avatar
Jingning Han committed
151 152 153 154 155 156 157 158 159 160 161 162 163 164 165
  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
};

166
// These are used for masking the left and above 32x32 borders.
clang-format's avatar
clang-format committed
167
static const uint64_t left_border = 0x1111111111111111ULL;
Jingning Han's avatar
Jingning Han committed
168 169 170
static const uint64_t above_border = 0x000000ff000000ffULL;

// 16 bit masks for uv transform sizes.
clang-format's avatar
clang-format committed
171
static const uint16_t left_64x64_txform_mask_uv[TX_SIZES] = {
172 173 174
#if CONFIG_CB4X4
  0xffff,  // TX_2X2
#endif
Jingning Han's avatar
Jingning Han committed
175 176 177 178
  0xffff,  // TX_4X4
  0xffff,  // TX_8x8
  0x5555,  // TX_16x16
  0x1111,  // TX_32x32
179 180 181
#if CONFIG_TX64X64
  0x0101,  // TX_64x64, never used
#endif     // CONFIG_TX64X64
Jingning Han's avatar
Jingning Han committed
182 183
};

clang-format's avatar
clang-format committed
184
static const uint16_t above_64x64_txform_mask_uv[TX_SIZES] = {
185 186 187
#if CONFIG_CB4X4
  0xffff,  // TX_2X2
#endif
Jingning Han's avatar
Jingning Han committed
188 189 190 191
  0xffff,  // TX_4X4
  0xffff,  // TX_8x8
  0x0f0f,  // TX_16x16
  0x000f,  // TX_32x32
192 193 194
#if CONFIG_TX64X64
  0x0003,  // TX_64x64, never used
#endif     // CONFIG_TX64X64
Jingning Han's avatar
Jingning Han committed
195 196 197 198
};

// 16 bit left mask to shift and set for each uv prediction size.
static const uint16_t left_prediction_mask_uv[BLOCK_SIZES] = {
Jingning Han's avatar
Jingning Han committed
199 200 201 202 203
#if CONFIG_CB4X4
  0x0001,  // BLOCK_2X2,
  0x0001,  // BLOCK_2X4,
  0x0001,  // BLOCK_4X2,
#endif
Jingning Han's avatar
Jingning Han committed
204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219
  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] = {
Jingning Han's avatar
Jingning Han committed
220 221 222 223 224
#if CONFIG_CB4X4
  0x0001,  // BLOCK_2X2
  0x0001,  // BLOCK_2X4
  0x0001,  // BLOCK_4X2
#endif
Jingning Han's avatar
Jingning Han committed
225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241
  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] = {
Jingning Han's avatar
Jingning Han committed
242 243 244 245 246
#if CONFIG_CB4X4
  0x0001,  // BLOCK_2X2
  0x0001,  // BLOCK_2X4
  0x0001,  // BLOCK_4X2
#endif
Jingning Han's avatar
Jingning Han committed
247 248 249 250 251 252 253 254 255 256 257 258 259 260
  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
261
static const uint16_t left_border_uv = 0x1111;
Jingning Han's avatar
Jingning Han committed
262 263 264 265 266
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
267
#if CONFIG_EXT_INTER
clang-format's avatar
clang-format committed
268 269 270 271
  ,
  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
272 273 274 275 276 277 278 279 280 281 282 283 284 285 286
};

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
287
    if (block_inside_limit < 1) block_inside_limit = 1;
Jingning Han's avatar
Jingning Han committed
288 289 290 291 292 293 294 295 296

    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) {
297
#if CONFIG_SUPERTX
Yaowu Xu's avatar
Yaowu Xu committed
298
  const int segment_id = AOMMIN(mbmi->segment_id, mbmi->segment_id_supertx);
clang-format's avatar
clang-format committed
299 300
  assert(
      IMPLIES(supertx_enabled(mbmi), mbmi->segment_id_supertx != MAX_SEGMENTS));
301 302 303 304 305 306
  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
307 308
}

Yaowu Xu's avatar
Yaowu Xu committed
309
void av1_loop_filter_init(AV1_COMMON *cm) {
Jingning Han's avatar
Jingning Han committed
310 311 312 313 314 315 316 317 318 319 320 321 322
  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
323
void av1_loop_filter_frame_init(AV1_COMMON *cm, int default_filt_lvl) {
Jingning Han's avatar
Jingning Han committed
324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342
  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
343 344 345
      lvl_seg = clamp(
          seg->abs_delta == SEGMENT_ABSDATA ? data : default_filt_lvl + data, 0,
          MAX_LOOP_FILTER);
Jingning Han's avatar
Jingning Han committed
346 347 348 349 350 351 352 353 354 355 356
    }

    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);

357
      for (ref = LAST_FRAME; ref < TOTAL_REFS_PER_FRAME; ++ref) {
Jingning Han's avatar
Jingning Han committed
358
        for (mode = 0; mode < MAX_MODE_LF_DELTAS; ++mode) {
clang-format's avatar
clang-format committed
359 360
          const int inter_lvl = lvl_seg + lf->ref_deltas[ref] * scale +
                                lf->mode_deltas[mode] * scale;
Jingning Han's avatar
Jingning Han committed
361 362 363 364 365 366 367
          lfi->lvl[seg_id][ref][mode] = clamp(inter_lvl, 0, MAX_LOOP_FILTER);
        }
      }
    }
  }
}

clang-format's avatar
clang-format committed
368 369
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
370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397
                                         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
398
          aom_lpf_vertical_16_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
399 400
                                   lfi0->hev_thr);
        } else if (mask_16x16_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
401
          aom_lpf_vertical_16(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr);
Jingning Han's avatar
Jingning Han committed
402
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
403
          aom_lpf_vertical_16(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim,
clang-format's avatar
clang-format committed
404
                              lfi1->hev_thr);
Jingning Han's avatar
Jingning Han committed
405 406 407 408 409
        }
      }

      if ((mask_8x8_0 | mask_8x8_1) & 1) {
        if ((mask_8x8_0 & mask_8x8_1) & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
410
          aom_lpf_vertical_8_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
411 412 413
                                  lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                  lfi1->hev_thr);
        } else if (mask_8x8_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
414
          aom_lpf_vertical_8(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr);
Jingning Han's avatar
Jingning Han committed
415
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
416
          aom_lpf_vertical_8(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim,
417
                             lfi1->hev_thr);
Jingning Han's avatar
Jingning Han committed
418 419 420 421 422
        }
      }

      if ((mask_4x4_0 | mask_4x4_1) & 1) {
        if ((mask_4x4_0 & mask_4x4_1) & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
423
          aom_lpf_vertical_4_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
424 425 426
                                  lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                  lfi1->hev_thr);
        } else if (mask_4x4_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
427
          aom_lpf_vertical_4(s, pitch, lfi0->mblim, lfi0->lim, lfi0->hev_thr);
Jingning Han's avatar
Jingning Han committed
428
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
429
          aom_lpf_vertical_4(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim,
430
                             lfi1->hev_thr);
Jingning Han's avatar
Jingning Han committed
431 432 433 434 435
        }
      }

      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
436
          aom_lpf_vertical_4_dual(s + 4, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
437 438 439
                                  lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                  lfi1->hev_thr);
        } else if (mask_4x4_int_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
440
          aom_lpf_vertical_4(s + 4, pitch, lfi0->mblim, lfi0->lim,
441
                             lfi0->hev_thr);
Jingning Han's avatar
Jingning Han committed
442
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
443
          aom_lpf_vertical_4(s + 8 * pitch + 4, pitch, lfi1->mblim, lfi1->lim,
444
                             lfi1->hev_thr);
Jingning Han's avatar
Jingning Han committed
445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461
        }
      }
    }

    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
462
#if CONFIG_AOM_HIGHBITDEPTH
clang-format's avatar
clang-format committed
463 464 465 466 467
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
468 469 470 471 472 473 474 475 476 477 478 479 480 481 482
  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
483
              mask_16x16_1 | mask_8x8_1 | mask_4x4_1 | mask_4x4_int_1;
Jingning Han's avatar
Jingning Han committed
484 485 486 487 488 489 490
       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
491
          aom_highbd_lpf_vertical_16_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
492 493
                                          lfi0->hev_thr, bd);
        } else if (mask_16x16_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
494
          aom_highbd_lpf_vertical_16(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
495 496
                                     lfi0->hev_thr, bd);
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
497
          aom_highbd_lpf_vertical_16(s + 8 * pitch, pitch, lfi1->mblim,
Jingning Han's avatar
Jingning Han committed
498 499 500 501 502 503
                                     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
504
          aom_highbd_lpf_vertical_8_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
505 506 507
                                         lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                         lfi1->hev_thr, bd);
        } else if (mask_8x8_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
508
          aom_highbd_lpf_vertical_8(s, pitch, lfi0->mblim, lfi0->lim,
509
                                    lfi0->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
510
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
511
          aom_highbd_lpf_vertical_8(s + 8 * pitch, pitch, lfi1->mblim,
512
                                    lfi1->lim, lfi1->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
513 514 515 516 517
        }
      }

      if ((mask_4x4_0 | mask_4x4_1) & 1) {
        if ((mask_4x4_0 & mask_4x4_1) & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
518
          aom_highbd_lpf_vertical_4_dual(s, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
519 520 521
                                         lfi0->hev_thr, lfi1->mblim, lfi1->lim,
                                         lfi1->hev_thr, bd);
        } else if (mask_4x4_0 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
522
          aom_highbd_lpf_vertical_4(s, pitch, lfi0->mblim, lfi0->lim,
523
                                    lfi0->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
524
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
525
          aom_highbd_lpf_vertical_4(s + 8 * pitch, pitch, lfi1->mblim,
526
                                    lfi1->lim, lfi1->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
527 528 529 530 531
        }
      }

      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
532
          aom_highbd_lpf_vertical_4_dual(s + 4, pitch, lfi0->mblim, lfi0->lim,
Jingning Han's avatar
Jingning Han committed
533 534 535
                                         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
536
          aom_highbd_lpf_vertical_4(s + 4, pitch, lfi0->mblim, lfi0->lim,
537
                                    lfi0->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
538
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
539
          aom_highbd_lpf_vertical_4(s + 8 * pitch + 4, pitch, lfi1->mblim,
540
                                    lfi1->lim, lfi1->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556
        }
      }
    }

    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
557
#endif  // CONFIG_AOM_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
558

clang-format's avatar
clang-format committed
559 560 561 562
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
563 564 565
  unsigned int mask;
  int count;

clang-format's avatar
clang-format committed
566 567
  for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; mask;
       mask >>= count) {
Jingning Han's avatar
Jingning Han committed
568 569 570 571 572 573
    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
574
          aom_lpf_horizontal_edge_16(s, pitch, lfi->mblim, lfi->lim,
575
                                     lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
576 577
          count = 2;
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
578
          aom_lpf_horizontal_edge_8(s, pitch, lfi->mblim, lfi->lim,
579
                                    lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
580 581 582 583 584 585
        }
      } 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
586
          aom_lpf_horizontal_8_dual(s, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
587 588 589 590
                                    lfi->hev_thr, lfin->mblim, lfin->lim,
                                    lfin->hev_thr);

          if ((mask_4x4_int & 3) == 3) {
Yaowu Xu's avatar
Yaowu Xu committed
591
            aom_lpf_horizontal_4_dual(s + 4 * pitch, pitch, lfi->mblim,
Jingning Han's avatar
Jingning Han committed
592 593 594 595
                                      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
596
              aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
597
                                   lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
598
            else if (mask_4x4_int & 2)
Yaowu Xu's avatar
Yaowu Xu committed
599
              aom_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
600
                                   lfin->lim, lfin->hev_thr);
Jingning Han's avatar
Jingning Han committed
601 602 603
          }
          count = 2;
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
604
          aom_lpf_horizontal_8(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
605 606

          if (mask_4x4_int & 1)
Yaowu Xu's avatar
Yaowu Xu committed
607
            aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
608
                                 lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
609 610 611 612 613 614
        }
      } 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
615
          aom_lpf_horizontal_4_dual(s, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
616 617 618
                                    lfi->hev_thr, lfin->mblim, lfin->lim,
                                    lfin->hev_thr);
          if ((mask_4x4_int & 3) == 3) {
Yaowu Xu's avatar
Yaowu Xu committed
619
            aom_lpf_horizontal_4_dual(s + 4 * pitch, pitch, lfi->mblim,
Jingning Han's avatar
Jingning Han committed
620 621 622 623
                                      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
624
              aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
625
                                   lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
626
            else if (mask_4x4_int & 2)
Yaowu Xu's avatar
Yaowu Xu committed
627
              aom_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
628
                                   lfin->lim, lfin->hev_thr);
Jingning Han's avatar
Jingning Han committed
629 630 631
          }
          count = 2;
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
632
          aom_lpf_horizontal_4(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
633 634

          if (mask_4x4_int & 1)
Yaowu Xu's avatar
Yaowu Xu committed
635
            aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
636
                                 lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
637 638
        }
      } else if (mask_4x4_int & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
639
        aom_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
640
                             lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
641 642 643 644 645 646 647 648 649 650 651
      }
    }
    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
652
#if CONFIG_AOM_HIGHBITDEPTH
clang-format's avatar
clang-format committed
653 654 655 656
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
657 658 659
  unsigned int mask;
  int count;

clang-format's avatar
clang-format committed
660 661
  for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; mask;
       mask >>= count) {
Jingning Han's avatar
Jingning Han committed
662 663 664 665 666 667
    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
668
          aom_highbd_lpf_horizontal_edge_16(s, pitch, lfi->mblim, lfi->lim,
669
                                            lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
670 671
          count = 2;
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
672
          aom_highbd_lpf_horizontal_edge_8(s, pitch, lfi->mblim, lfi->lim,
673
                                           lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
674 675 676 677 678 679
        }
      } 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
680
          aom_highbd_lpf_horizontal_8_dual(s, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
681 682 683 684
                                           lfi->hev_thr, lfin->mblim, lfin->lim,
                                           lfin->hev_thr, bd);

          if ((mask_4x4_int & 3) == 3) {
Yaowu Xu's avatar
Yaowu Xu committed
685
            aom_highbd_lpf_horizontal_4_dual(
clang-format's avatar
clang-format committed
686 687
                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
688 689
          } else {
            if (mask_4x4_int & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
690
              aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
691
                                          lfi->lim, lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
692
            } else if (mask_4x4_int & 2) {
Yaowu Xu's avatar
Yaowu Xu committed
693
              aom_highbd_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
694
                                          lfin->lim, lfin->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
695 696 697 698
            }
          }
          count = 2;
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
699
          aom_highbd_lpf_horizontal_8(s, pitch, lfi->mblim, lfi->lim,
700
                                      lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
701 702

          if (mask_4x4_int & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
703
            aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
704
                                        lfi->lim, lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
705 706 707 708 709 710 711
          }
        }
      } 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
712
          aom_highbd_lpf_horizontal_4_dual(s, pitch, lfi->mblim, lfi->lim,
Jingning Han's avatar
Jingning Han committed
713 714 715
                                           lfi->hev_thr, lfin->mblim, lfin->lim,
                                           lfin->hev_thr, bd);
          if ((mask_4x4_int & 3) == 3) {
Yaowu Xu's avatar
Yaowu Xu committed
716
            aom_highbd_lpf_horizontal_4_dual(
clang-format's avatar
clang-format committed
717 718
                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
719 720
          } else {
            if (mask_4x4_int & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
721
              aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
722
                                          lfi->lim, lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
723
            } else if (mask_4x4_int & 2) {
Yaowu Xu's avatar
Yaowu Xu committed
724
              aom_highbd_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
725
                                          lfin->lim, lfin->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
726 727 728 729
            }
          }
          count = 2;
        } else {
Yaowu Xu's avatar
Yaowu Xu committed
730
          aom_highbd_lpf_horizontal_4(s, pitch, lfi->mblim, lfi->lim,
731
                                      lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
732 733

          if (mask_4x4_int & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
734
            aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
735
                                        lfi->lim, lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
736 737 738
          }
        }
      } else if (mask_4x4_int & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
739
        aom_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
740
                                    lfi->hev_thr, bd);
Jingning Han's avatar
Jingning Han committed
741 742 743 744 745 746 747 748 749 750
      }
    }
    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
751
#endif  // CONFIG_AOM_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
752 753 754 755 756 757 758 759 760 761

// 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
762
                        const int shift_uv, LOOP_FILTER_MASK *lfm) {
Jingning Han's avatar
Jingning Han committed
763 764
  const MB_MODE_INFO *mbmi = &mi->mbmi;
  const BLOCK_SIZE block_size = mbmi->sb_type;
765 766
  // TODO(debargha): Check if masks can be setup correctly when
  // rectangular transfroms are used with the EXT_TX expt.
767 768 769
  const TX_SIZE tx_size_y = txsize_sqr_map[mbmi->tx_size];
  const TX_SIZE tx_size_y_left = txsize_horz_map[mbmi->tx_size];
  const TX_SIZE tx_size_y_above = txsize_vert_map[mbmi->tx_size];
770
  const TX_SIZE tx_size_uv =
771 772 773 774 775
      txsize_sqr_map[uv_txsize_lookup[block_size][mbmi->tx_size][1][1]];
  const TX_SIZE tx_size_uv_left =
      txsize_horz_map[uv_txsize_lookup[block_size][mbmi->tx_size][1][1]];
  const TX_SIZE tx_size_uv_above =
      txsize_vert_map[uv_txsize_lookup[block_size][mbmi->tx_size][1][1]];
Jingning Han's avatar
Jingning Han committed
776
  const int filter_level = get_filter_level(lfi_n, mbmi);
777 778
  uint64_t *const left_y = &lfm->left_y[tx_size_y_left];
  uint64_t *const above_y = &lfm->above_y[tx_size_y_above];
Jingning Han's avatar
Jingning Han committed
779
  uint64_t *const int_4x4_y = &lfm->int_4x4_y;
780 781
  uint16_t *const left_uv = &lfm->left_uv[tx_size_uv_left];
  uint16_t *const above_uv = &lfm->above_uv[tx_size_uv_above];
782
  uint16_t *const int_4x4_uv = &lfm->left_int_4x4_uv;
Jingning Han's avatar
Jingning Han committed
783 784 785 786 787 788 789 790
  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];
791 792 793
    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
794
    for (i = 0; i < h; i++) memset(&lfm->lfl_y[row + i][col], filter_level, w);
Jingning Han's avatar
Jingning Han committed
795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815
  }

  // 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.
816
  if (mbmi->skip && is_inter_block(mbmi)) return;
Jingning Han's avatar
Jingning Han committed
817 818 819 820 821

  // 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..
822
  *above_y |= (size_mask[block_size] & above_64x64_txform_mask[tx_size_y_above])
clang-format's avatar
clang-format committed
823 824
              << shift_y;
  *above_uv |=
825
      (size_mask_uv[block_size] & above_64x64_txform_mask_uv[tx_size_uv_above])
clang-format's avatar
clang-format committed
826
      << shift_uv;
Jingning Han's avatar
Jingning Han committed
827

828
  *left_y |= (size_mask[block_size] & left_64x64_txform_mask[tx_size_y_left])
clang-format's avatar
clang-format committed
829
             << shift_y;
830 831 832
  *left_uv |=
      (size_mask_uv[block_size] & left_64x64_txform_mask_uv[tx_size_uv_left])
      << shift_uv;
Jingning Han's avatar
Jingning Han committed
833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849

  // 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,
850 851 852
#if CONFIG_SUPERTX
                         int supertx_enabled,
#endif  // CONFIG_SUPERTX
Jingning Han's avatar
Jingning Han committed
853 854
                         LOOP_FILTER_MASK *lfm) {
  const MB_MODE_INFO *mbmi = &mi->mbmi;
855 856 857
  const TX_SIZE tx_size_y = txsize_sqr_map[mbmi->tx_size];
  const TX_SIZE tx_size_y_left = txsize_horz_map[mbmi->tx_size];
  const TX_SIZE tx_size_y_above = txsize_vert_map[mbmi->tx_size];
858 859 860 861 862 863
#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
864
  const int filter_level = get_filter_level(lfi_n, mbmi);
865 866
  uint64_t *const left_y = &lfm->left_y[tx_size_y_left];
  uint64_t *const above_y = &lfm->above_y[tx_size_y_above];
Jingning Han's avatar
Jingning Han committed
867 868 869 870 871 872 873 874
  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];
875 876 877
    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
878
    for (i = 0; i < h; i++) memset(&lfm->lfl_y[row + i][col], filter_level, w);
Jingning Han's avatar
Jingning Han committed
879 880 881 882 883
  }

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

884
  if (mbmi->skip && is_inter_block(mbmi)) return;
Jingning Han's avatar
Jingning Han committed
885

886
  *above_y |= (size_mask[block_size] & above_64x64_txform_mask[tx_size_y_above])
clang-format's avatar
clang-format committed
887
              << shift_y;
Jingning Han's avatar
Jingning Han committed
888

889
  *left_y |= (size_mask[block_size] & left_64x64_txform_mask[tx_size_y_left])
clang-format's avatar
clang-format committed
890
             << shift_y;
Jingning Han's avatar
Jingning Han committed
891 892 893 894 895

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

896
#if CONFIG_LOOPFILTERING_ACROSS_TILES
897 898 899 900 901 902 903 904 905 906 907
// This function update the bit masks for the entire 64x64 region represented
// by mi_row, mi_col. In case one of the edge is a tile boundary, loop filtering
// for that edge is disabled. This function only check the tile boundary info
// for the top left corner mi to determine the boundary information for the
// top and left edge of the whole super block
static void update_tile_boundary_filter_mask(AV1_COMMON *const cm,
                                             const int mi_row, const int mi_col,
                                             LOOP_FILTER_MASK *lfm) {
  int i;
  MODE_INFO *const mi = cm->mi + mi_row * cm->mi_stride + mi_col;

908
  if (mi->mbmi.boundary_info & TILE_LEFT_BOUNDARY) {
909 910 911 912 913 914
    for (i = 0; i <= TX_32X32; i++) {
      lfm->left_y[i] &= 0xfefefefefefefefeULL;
      lfm->left_uv[i] &= 0xeeee;
    }
  }

915
  if (mi->mbmi.boundary_info & TILE_ABOVE_BOUNDARY) {
916 917 918 919 920 921
    for (i = 0; i <= TX_32X32; i++) {
      lfm->above_y[i] &= 0xffffffffffffff00ULL;
      lfm->above_uv[i] &= 0xfff0;
    }
  }
}
922
#endif  // CONFIG_LOOPFILTERING_ACROSS_TILES
923

Jingning Han's avatar
Jingning Han committed
924 925 926
// 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
927 928 929
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
930 931 932 933 934 935 936 937 938
  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
939 940 941 942 943
  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
944 945 946 947 948

  // 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
949 950 951 952 953
  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
954
  int i;
Yaowu Xu's avatar
Yaowu Xu committed
955 956
  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);
957 958 959
#if CONFIG_EXT_PARTITION
  assert(0 && "Not yet updated");
#endif  // CONFIG_EXT_PARTITION
Jingning Han's avatar
Jingning Han committed
960

Yaowu Xu's avatar
Yaowu Xu committed
961
  av1_zero(*lfm);
Jingning Han's avatar
Jingning Han committed
962 963 964 965 966 967
  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
968
    case BLOCK_64X64: build_masks(lfi_n, mip[0], 0, 0, lfm); break;
969 970
    case BLOCK_64X32: build_masks(lfi_n, mip[0], 0, 0, lfm);
#if CONFIG_SUPERTX && CONFIG_TX64X64
971
      if (supertx_enabled(&mip[0]->mbmi)) break;
972
#endif  // CONFIG_SUPERTX && CONFIG_TX64X64
Jingning Han's avatar
Jingning Han committed
973
      mip2 = mip + mode_info_stride * 4;
clang-format's avatar
clang-format committed
974
      if (4 >= max_rows) break;
Jingning Han's avatar
Jingning Han committed
975 976
      build_masks(lfi_n, mip2[0], 32, 8, lfm);
      break;
977 978
    case BLOCK_32X64: build_masks(lfi_n, mip[0], 0, 0, lfm);
#if CONFIG_SUPERTX && CONFIG_TX64X64
979
      if (supertx_enabled(&mip[0]->mbmi)) break;
980
#endif  // CONFIG_SUPERTX && CONFIG_TX64X64
Jingning Han's avatar
Jingning Han committed
981
      mip2 = mip + 4;
clang-format's avatar
clang-format committed
982
      if (4 >= max_cols) break;
Jingning Han's avatar
Jingning Han committed
983 984 985
      build_masks(lfi_n, mip2[0], 4, 2, lfm);
      break;
    default:
986
#if CONFIG_SUPERTX && CONFIG_TX64X64
987 988
      if (mip[0]->mbmi.tx_size == TX_64X64) {
        build_masks(lfi_n, mip[0], 0, 0, lfm);
989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003
      } else {
#endif  // CONFIG_SUPERTX && CONFIG_TX64X64
        for (idx_32 = 0; idx_32 < 4; mip += offset_32[idx_32], ++idx_32) {
          const int shift_y_32 = shift_32_y[idx_32];
          const int shift_uv_32 = 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_32, shift_uv_32, lfm);
              break;
            case BLOCK_32X16:
              build_masks(lfi_n, mip[0], shift_y_32, shift_uv_32, lfm);
1004
#if CONFIG_SUPERTX
1005
              if (supertx_enabled(&mip[0]->mbmi)) break;
1006
#endif
1007 1008 1009 1010 1011 1012 1013
              if (mi_32_row_offset + 2 >= max_rows) continue;
              mip2 = mip + mode_info_stride * 2;
              build_masks(lfi_n, mip2[0], shift_y_32 + 16, shift_uv_32 + 4,
                          lfm);
              break;
            case BLOCK_16X32:
              build_masks(lfi_n, mip[0], shift_y_32, shift_uv_32, lfm);
1014
#if CONFIG_SUPERTX
1015
              if (supertx_enabled(&mip[0]->mbmi)) break;
1016
#endif
1017 1018 1019
              if (mi_32_col_offset + 2 >= max_cols) continue;
              mip2 = mip + 2;
              build_masks(lfi_n, mip2[0], shift_y_32 + 2, shift_uv_32 + 1, lfm);
1020
              break;
1021 1022 1023 1024 1025 1026
            default:
#if CONFIG_SUPERTX
              if (mip[0]->mbmi.tx_size == TX_32X32) {
                build_masks(lfi_n, mip[0], shift_y_32, shift_uv_32, lfm);
                break;
              }
1027
#endif
1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045
              for (idx_16 = 0; idx_16 < 4; mip += offset_16[idx_16], ++idx_16) {
                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];
                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);

                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_32_16, shift_uv_32_16,
                                lfm);
                    break;
                  case BLOCK_16X8:
1046
#if CONFIG_SUPERTX
1047
                    if (supertx_enabled(&mip[0]->mbmi)) break;
1048
#endif
1049 1050 1051 1052 1053
                    build_masks(lfi_n, mip[0], shift_y_32_16, shift_uv_32_16,
                                lfm);
                    if (mi_16_row_offset + 1 >= max_rows) continue;
                    mip2 = mip + mode_info_stride;
                    build_y_mask(lfi_n, mip2[0], shift_y_32_16 + 8,
1054
#if CONFIG_SUPERTX
1055
                                 0,
1056
#endif
1057 1058 1059
                                 lfm);
                    break;
                  case BLOCK_8X16:
1060
#if CONFIG_SUPERTX
1061
                    if (supertx_enabled(&mip[0]->mbmi)) break;
1062
#endif
1063 1064 1065 1066 1067
                    build_masks(lfi_n, mip[0], shift_y_32_16, shift_uv_32_16,
                                lfm);
                    if (mi_16_col_offset + 1 >= max_cols) continue;
                    mip2 = mip + 1;
                    build_y_mask(lfi_n, mip2[0], shift_y_32_16 + 1,
1068
#if CONFIG_SUPERTX
1069
                                 0,
1070
#endif
1071 1072 1073 1074 1075
                                 lfm);
                    break;
                  default: {
                    const int shift_y_32_16_8_zero =
                        shift_y_32_16 + shift_8_y[0];
1076
#if CONFIG_SUPERTX
1077 1078 1079 1080 1081 1082
                    if (mip[0]->mbmi.tx_size == TX_16X16) {
                      build_masks(lfi_n, mip[0], shift_y_32_16_8_zero,
                                  shift_uv_32_16, lfm);
                      break;
                    }
#endif
1083 1084
                    build_masks(lfi_n, mip[0], shift_y_32_16_8_zero,
                                shift_uv_32_16, lfm);
1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097
                    mip += offset[0];
                    for (idx_8 = 1; idx_8 < 4; mip += offset[idx_8], ++idx_8) {
                      const int shift_y_32_16_8 =
                          shift_y_32_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));

                      if (mi_8_col_offset >= max_cols ||
                          mi_8_row_offset >= max_rows)
                        continue;
                      build_y_mask(lfi_n, mip[0], shift_y_32_16_8,
1098
#if CONFIG_SUPERTX
1099
                                   supertx_enabled(&mip[0]->mbmi),
1100
#endif
1101 1102 1103
                                   lfm);
                    }
                    break;
Jingning Han's avatar
Jingning Han committed
1104 1105 1106
                  }
                }
              }
1107 1108
              break;
          }
Jingning Han's avatar
Jingning Han committed
1109
        }
1110
#if CONFIG_SUPERTX && CONFIG_TX64X64
Jingning Han's avatar
Jingning Han committed
1111
      }
1112
#endif  // CONFIG_SUPERTX && CONFIG_TX64X64
Jingning Han's avatar
Jingning Han committed
1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134
      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.
1135
  if (mi_row + MAX_MIB_SIZE > cm->mi_rows) {
Jingning Han's avatar
Jingning Han committed
1136 1137 1138
    const uint64_t rows = cm->mi_rows - mi_row;

    // Each pixel inside the border gets a 1,
clang-format's avatar
clang-format committed
1139
    const uint64_t mask_y = (((uint64_t)1 << (rows << MAX_MIB_SIZE_LOG2)) - 1);
1140
    const uint16_t mask_uv =
clang-format's avatar
clang-format committed
1141
        (((uint16_t)1 << (((rows + 1) >> 1) << (MAX_MIB_SIZE_LOG2 - 1))) - 1);
Jingning Han's avatar
Jingning Han committed
1142 1143 1144 1145 1146 1147 1148 1149 1150

    // 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;
1151
    lfm->above_int_4x4_uv = lfm->left_int_4x4_uv & mask_uv;
Jingning Han's avatar
Jingning Han committed
1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162

    // 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);
    }
1163 1164
  } else {
    lfm->above_int_4x4_uv = lfm->left_int_4x4_uv;
Jingning Han's avatar
Jingning Han committed
1165 1166
  }

1167
  if (mi_col + MAX_MIB_SIZE > cm->mi_cols) {
Jingning Han's avatar
Jingning Han committed
1168 1169 1170 1171
    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
1172
    const uint64_t mask_y = (((1 << columns) - 1)) * 0x0101010101010101ULL;
Jingning Han's avatar
Jingning Han committed
1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186
    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;
1187
    lfm->left_int_4x4_uv &= mask_uv_int;
Jingning Han's avatar
Jingning Han committed
1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208

    // 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;
    }
  }

1209
#if CONFIG_LOOPFILTERING_ACROSS_TILES
1210 1211 1212
  if (av1_disable_loopfilter_on_tile_boundary(cm)) {
    update_tile_boundary_filter_mask(cm, mi_row, mi_col, lfm);
  }
1213
#endif  // CONFIG_LOOPFILTERING_ACROSS_TILES
1214

Jingning Han's avatar
Jingning Han committed
1215 1216 1217 1218 1219
  // 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
1220
  assert(!(lfm->left_uv[TX_16X16] & lfm->left_uv[TX_8X8]));
Jingning Han's avatar
Jingning Han committed
1221 1222
  assert(!(lfm->left_uv[TX_16X16] & lfm->left_uv[TX_4X4]));
  assert(!(lfm->left_uv[TX_8X8] & lfm->left_uv[TX_4X4]));
1223
  assert(!(lfm->left_int_4x4_uv & lfm->left_uv[TX_16X16]));
Jingning Han's avatar
Jingning Han committed
1224 1225 1226 1227 1228 1229 1230
  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]));
1231
  assert(!(lfm->above_int_4x4_uv & lfm->above_uv[TX_16X16]));
Jingning Han's avatar
Jingning Han committed
1232 1233
}

clang-format's avatar
clang-format committed
1234 1235 1236 1237
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
1238 1239
  unsigned int mask;

clang-format's avatar
clang-format committed
1240 1241
  for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; mask;
       mask >>= 1) {
Jingning Han's avatar
Jingning Han committed
1242 1243 1244 1245
    const loop_filter_thresh *lfi = lfi_n->lfthr + *lfl;

    if (mask & 1) {
      if (mask_16x16 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
1246
        aom_lpf_vertical_16(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed
1247
      } else if (mask_8x8 & 1) {
Yaowu Xu's avatar
Yaowu Xu committed
1248
        aom_lpf_vertical_8(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr);
Jingning Han's avatar
Jingning Han committed