loopfilter.c 29.7 KB
Newer Older
John Koleszar's avatar
John Koleszar committed
1
/*
2
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
John Koleszar's avatar
John Koleszar committed
3
 *
4 5 6 7 8 9
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
John Koleszar's avatar
John Koleszar committed
10
 */
11

Zoe Liu's avatar
Zoe Liu committed
12 13
#include <stdlib.h>

Adrian Grange's avatar
Adrian Grange committed
14 15
#include "./aom_config.h"
#include "aom_dsp/aom_dsp_common.h"
Yaowu Xu's avatar
Yaowu Xu committed
16
#include "aom_ports/mem.h"
John Koleszar's avatar
John Koleszar committed
17

18
static INLINE int8_t signed_char_clamp(int t) {
19
  return (int8_t)clamp(t, -128, 127);
John Koleszar's avatar
John Koleszar committed
20 21
}

22
#if CONFIG_AOM_HIGHBITDEPTH
23 24
static INLINE int16_t signed_char_clamp_high(int t, int bd) {
  switch (bd) {
clang-format's avatar
clang-format committed
25 26
    case 10: return (int16_t)clamp(t, -128 * 4, 128 * 4 - 1);
    case 12: return (int16_t)clamp(t, -128 * 16, 128 * 16 - 1);
27
    case 8:
clang-format's avatar
clang-format committed
28
    default: return (int16_t)clamp(t, -128, 128 - 1);
29 30 31 32
  }
}
#endif

Dmitry Kovalev's avatar
Dmitry Kovalev committed
33
// should we apply any filter at all: 11111111 yes, 00000000 no
clang-format's avatar
clang-format committed
34 35 36
static INLINE int8_t filter_mask(uint8_t limit, uint8_t blimit, uint8_t p3,
                                 uint8_t p2, uint8_t p1, uint8_t p0, uint8_t q0,
                                 uint8_t q1, uint8_t q2, uint8_t q3) {
37
  int8_t mask = 0;
John Koleszar's avatar
John Koleszar committed
38 39 40 41 42 43
  mask |= (abs(p3 - p2) > limit) * -1;
  mask |= (abs(p2 - p1) > limit) * -1;
  mask |= (abs(p1 - p0) > limit) * -1;
  mask |= (abs(q1 - q0) > limit) * -1;
  mask |= (abs(q2 - q1) > limit) * -1;
  mask |= (abs(q3 - q2) > limit) * -1;
clang-format's avatar
clang-format committed
44
  mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit) * -1;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
45
  return ~mask;
John Koleszar's avatar
John Koleszar committed
46 47
}

clang-format's avatar
clang-format committed
48 49
static INLINE int8_t flat_mask4(uint8_t thresh, uint8_t p3, uint8_t p2,
                                uint8_t p1, uint8_t p0, uint8_t q0, uint8_t q1,
50
                                uint8_t q2, uint8_t q3) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
51 52 53 54 55 56 57 58
  int8_t mask = 0;
  mask |= (abs(p1 - p0) > thresh) * -1;
  mask |= (abs(q1 - q0) > thresh) * -1;
  mask |= (abs(p2 - p0) > thresh) * -1;
  mask |= (abs(q2 - q0) > thresh) * -1;
  mask |= (abs(p3 - p0) > thresh) * -1;
  mask |= (abs(q3 - q0) > thresh) * -1;
  return ~mask;
59 60
}

clang-format's avatar
clang-format committed
61 62 63 64
static INLINE int8_t flat_mask5(uint8_t thresh, uint8_t p4, uint8_t p3,
                                uint8_t p2, uint8_t p1, uint8_t p0, uint8_t q0,
                                uint8_t q1, uint8_t q2, uint8_t q3,
                                uint8_t q4) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
65 66 67 68
  int8_t mask = ~flat_mask4(thresh, p3, p2, p1, p0, q0, q1, q2, q3);
  mask |= (abs(p4 - p0) > thresh) * -1;
  mask |= (abs(q4 - q0) > thresh) * -1;
  return ~mask;
69 70
}

71
// is there high edge variance internal edge: 11111111 yes, 00000000 no
72 73
static INLINE int8_t hev_mask(uint8_t thresh, uint8_t p1, uint8_t p0,
                              uint8_t q0, uint8_t q1) {
74
  int8_t hev = 0;
clang-format's avatar
clang-format committed
75 76
  hev |= (abs(p1 - p0) > thresh) * -1;
  hev |= (abs(q1 - q0) > thresh) * -1;
John Koleszar's avatar
John Koleszar committed
77
  return hev;
John Koleszar's avatar
John Koleszar committed
78 79
}

80
static INLINE void filter4(int8_t mask, uint8_t thresh, uint8_t *op1,
81
                           uint8_t *op0, uint8_t *oq0, uint8_t *oq1) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
82
  int8_t filter1, filter2;
John Koleszar's avatar
John Koleszar committed
83

clang-format's avatar
clang-format committed
84 85 86 87
  const int8_t ps1 = (int8_t)*op1 ^ 0x80;
  const int8_t ps0 = (int8_t)*op0 ^ 0x80;
  const int8_t qs0 = (int8_t)*oq0 ^ 0x80;
  const int8_t qs1 = (int8_t)*oq1 ^ 0x80;
88
  const uint8_t hev = hev_mask(thresh, *op1, *op0, *oq0, *oq1);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105

  // add outer taps if we have high edge variance
  int8_t filter = signed_char_clamp(ps1 - qs1) & hev;

  // inner taps
  filter = signed_char_clamp(filter + 3 * (qs0 - ps0)) & mask;

  // save bottom 3 bits so that we round one side +4 and the other +3
  // if it equals 4 we'll set to adjust by -1 to account for the fact
  // we'd round 3 the other way
  filter1 = signed_char_clamp(filter + 4) >> 3;
  filter2 = signed_char_clamp(filter + 3) >> 3;

  *oq0 = signed_char_clamp(qs0 - filter1) ^ 0x80;
  *op0 = signed_char_clamp(ps0 + filter2) ^ 0x80;

  // outer tap adjustments
Dmitry Kovalev's avatar
Dmitry Kovalev committed
106
  filter = ROUND_POWER_OF_TWO(filter1, 1) & ~hev;
John Koleszar's avatar
John Koleszar committed
107

Dmitry Kovalev's avatar
Dmitry Kovalev committed
108 109
  *oq1 = signed_char_clamp(qs1 - filter) ^ 0x80;
  *op1 = signed_char_clamp(ps1 + filter) ^ 0x80;
John Koleszar's avatar
John Koleszar committed
110
}
111

Adrian Grange's avatar
Adrian Grange committed
112
void aom_lpf_horizontal_4_c(uint8_t *s, int p /* pitch */,
Jim Bankoski's avatar
Jim Bankoski committed
113 114
                            const uint8_t *blimit, const uint8_t *limit,
                            const uint8_t *thresh, int count) {
115
  int i;
John Koleszar's avatar
John Koleszar committed
116

Dmitry Kovalev's avatar
Dmitry Kovalev committed
117 118
  // loop filter designed to work using chars so that we can make maximum use
  // of 8 bit simd instructions.
119 120
  for (i = 0; i < 8 * count; ++i) {
    const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
clang-format's avatar
clang-format committed
121 122 123
    const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
    const int8_t mask =
        filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3);
124
    filter4(mask, *thresh, s - 2 * p, s - 1 * p, s, s + 1 * p);
John Koleszar's avatar
John Koleszar committed
125
    ++s;
126
  }
John Koleszar's avatar
John Koleszar committed
127 128
}

Adrian Grange's avatar
Adrian Grange committed
129
void aom_lpf_horizontal_4_dual_c(uint8_t *s, int p, const uint8_t *blimit0,
Jim Bankoski's avatar
Jim Bankoski committed
130 131 132
                                 const uint8_t *limit0, const uint8_t *thresh0,
                                 const uint8_t *blimit1, const uint8_t *limit1,
                                 const uint8_t *thresh1) {
Adrian Grange's avatar
Adrian Grange committed
133 134
  aom_lpf_horizontal_4_c(s, p, blimit0, limit0, thresh0, 1);
  aom_lpf_horizontal_4_c(s + 8, p, blimit1, limit1, thresh1, 1);
135 136
}

Adrian Grange's avatar
Adrian Grange committed
137
void aom_lpf_vertical_4_c(uint8_t *s, int pitch, const uint8_t *blimit,
Jim Bankoski's avatar
Jim Bankoski committed
138 139
                          const uint8_t *limit, const uint8_t *thresh,
                          int count) {
140
  int i;
John Koleszar's avatar
John Koleszar committed
141

Dmitry Kovalev's avatar
Dmitry Kovalev committed
142 143
  // loop filter designed to work using chars so that we can make maximum use
  // of 8 bit simd instructions.
144 145
  for (i = 0; i < 8 * count; ++i) {
    const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
clang-format's avatar
clang-format committed
146 147 148
    const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
    const int8_t mask =
        filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3);
149
    filter4(mask, *thresh, s - 2, s - 1, s, s + 1);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
150
    s += pitch;
151
  }
John Koleszar's avatar
John Koleszar committed
152
}
153

Adrian Grange's avatar
Adrian Grange committed
154
void aom_lpf_vertical_4_dual_c(uint8_t *s, int pitch, const uint8_t *blimit0,
Jim Bankoski's avatar
Jim Bankoski committed
155 156 157
                               const uint8_t *limit0, const uint8_t *thresh0,
                               const uint8_t *blimit1, const uint8_t *limit1,
                               const uint8_t *thresh1) {
Adrian Grange's avatar
Adrian Grange committed
158 159
  aom_lpf_vertical_4_c(s, pitch, blimit0, limit0, thresh0, 1);
  aom_lpf_vertical_4_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1, 1);
160 161
}

162
static INLINE void filter8(int8_t mask, uint8_t thresh, uint8_t flat,
clang-format's avatar
clang-format committed
163 164
                           uint8_t *op3, uint8_t *op2, uint8_t *op1,
                           uint8_t *op0, uint8_t *oq0, uint8_t *oq1,
165
                           uint8_t *oq2, uint8_t *oq3) {
John Koleszar's avatar
John Koleszar committed
166
  if (flat && mask) {
167 168
    const uint8_t p3 = *op3, p2 = *op2, p1 = *op1, p0 = *op0;
    const uint8_t q0 = *oq0, q1 = *oq1, q2 = *oq2, q3 = *oq3;
John Koleszar's avatar
John Koleszar committed
169

Dmitry Kovalev's avatar
Dmitry Kovalev committed
170 171 172 173 174 175 176
    // 7-tap filter [1, 1, 1, 2, 1, 1, 1]
    *op2 = ROUND_POWER_OF_TWO(p3 + p3 + p3 + 2 * p2 + p1 + p0 + q0, 3);
    *op1 = ROUND_POWER_OF_TWO(p3 + p3 + p2 + 2 * p1 + p0 + q0 + q1, 3);
    *op0 = ROUND_POWER_OF_TWO(p3 + p2 + p1 + 2 * p0 + q0 + q1 + q2, 3);
    *oq0 = ROUND_POWER_OF_TWO(p2 + p1 + p0 + 2 * q0 + q1 + q2 + q3, 3);
    *oq1 = ROUND_POWER_OF_TWO(p1 + p0 + q0 + 2 * q1 + q2 + q3 + q3, 3);
    *oq2 = ROUND_POWER_OF_TWO(p0 + q0 + q1 + 2 * q2 + q3 + q3 + q3, 3);
John Koleszar's avatar
John Koleszar committed
177
  } else {
clang-format's avatar
clang-format committed
178
    filter4(mask, thresh, op1, op0, oq0, oq1);
John Koleszar's avatar
John Koleszar committed
179
  }
180
}
181

Adrian Grange's avatar
Adrian Grange committed
182
void aom_lpf_horizontal_8_c(uint8_t *s, int p, const uint8_t *blimit,
Jim Bankoski's avatar
Jim Bankoski committed
183 184
                            const uint8_t *limit, const uint8_t *thresh,
                            int count) {
185
  int i;
John Koleszar's avatar
John Koleszar committed
186

Dmitry Kovalev's avatar
Dmitry Kovalev committed
187 188
  // loop filter designed to work using chars so that we can make maximum use
  // of 8 bit simd instructions.
189 190 191 192
  for (i = 0; i < 8 * count; ++i) {
    const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
    const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];

clang-format's avatar
clang-format committed
193 194
    const int8_t mask =
        filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3);
195
    const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3);
clang-format's avatar
clang-format committed
196 197
    filter8(mask, *thresh, flat, s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p, s,
            s + 1 * p, s + 2 * p, s + 3 * p);
John Koleszar's avatar
John Koleszar committed
198
    ++s;
199
  }
John Koleszar's avatar
John Koleszar committed
200
}
201

Adrian Grange's avatar
Adrian Grange committed
202
void aom_lpf_horizontal_8_dual_c(uint8_t *s, int p, const uint8_t *blimit0,
Jim Bankoski's avatar
Jim Bankoski committed
203 204 205
                                 const uint8_t *limit0, const uint8_t *thresh0,
                                 const uint8_t *blimit1, const uint8_t *limit1,
                                 const uint8_t *thresh1) {
Adrian Grange's avatar
Adrian Grange committed
206 207
  aom_lpf_horizontal_8_c(s, p, blimit0, limit0, thresh0, 1);
  aom_lpf_horizontal_8_c(s + 8, p, blimit1, limit1, thresh1, 1);
208 209
}

Adrian Grange's avatar
Adrian Grange committed
210
void aom_lpf_vertical_8_c(uint8_t *s, int pitch, const uint8_t *blimit,
Jim Bankoski's avatar
Jim Bankoski committed
211 212
                          const uint8_t *limit, const uint8_t *thresh,
                          int count) {
213 214 215 216 217
  int i;

  for (i = 0; i < 8 * count; ++i) {
    const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
    const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
clang-format's avatar
clang-format committed
218 219
    const int8_t mask =
        filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3);
220
    const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3);
clang-format's avatar
clang-format committed
221 222
    filter8(mask, *thresh, flat, s - 4, s - 3, s - 2, s - 1, s, s + 1, s + 2,
            s + 3);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
223
    s += pitch;
224
  }
John Koleszar's avatar
John Koleszar committed
225 226
}

Adrian Grange's avatar
Adrian Grange committed
227
void aom_lpf_vertical_8_dual_c(uint8_t *s, int pitch, const uint8_t *blimit0,
Jim Bankoski's avatar
Jim Bankoski committed
228 229 230
                               const uint8_t *limit0, const uint8_t *thresh0,
                               const uint8_t *blimit1, const uint8_t *limit1,
                               const uint8_t *thresh1) {
Adrian Grange's avatar
Adrian Grange committed
231 232
  aom_lpf_vertical_8_c(s, pitch, blimit0, limit0, thresh0, 1);
  aom_lpf_vertical_8_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1, 1);
clang-format's avatar
clang-format committed
233 234 235 236 237 238 239 240
}

static INLINE void filter16(int8_t mask, uint8_t thresh, uint8_t flat,
                            uint8_t flat2, uint8_t *op7, uint8_t *op6,
                            uint8_t *op5, uint8_t *op4, uint8_t *op3,
                            uint8_t *op2, uint8_t *op1, uint8_t *op0,
                            uint8_t *oq0, uint8_t *oq1, uint8_t *oq2,
                            uint8_t *oq3, uint8_t *oq4, uint8_t *oq5,
241
                            uint8_t *oq6, uint8_t *oq7) {
242
  if (flat2 && flat && mask) {
clang-format's avatar
clang-format committed
243 244
    const uint8_t p7 = *op7, p6 = *op6, p5 = *op5, p4 = *op4, p3 = *op3,
                  p2 = *op2, p1 = *op1, p0 = *op0;
245

clang-format's avatar
clang-format committed
246 247
    const uint8_t q0 = *oq0, q1 = *oq1, q2 = *oq2, q3 = *oq3, q4 = *oq4,
                  q5 = *oq5, q6 = *oq6, q7 = *oq7;
248

Dmitry Kovalev's avatar
Dmitry Kovalev committed
249
    // 15-tap filter [1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1]
clang-format's avatar
clang-format committed
250 251 252 253 254 255 256 257 258 259 260
    *op6 = ROUND_POWER_OF_TWO(
        p7 * 7 + p6 * 2 + p5 + p4 + p3 + p2 + p1 + p0 + q0, 4);
    *op5 = ROUND_POWER_OF_TWO(
        p7 * 6 + p6 + p5 * 2 + p4 + p3 + p2 + p1 + p0 + q0 + q1, 4);
    *op4 = ROUND_POWER_OF_TWO(
        p7 * 5 + p6 + p5 + p4 * 2 + p3 + p2 + p1 + p0 + q0 + q1 + q2, 4);
    *op3 = ROUND_POWER_OF_TWO(
        p7 * 4 + p6 + p5 + p4 + p3 * 2 + p2 + p1 + p0 + q0 + q1 + q2 + q3, 4);
    *op2 = ROUND_POWER_OF_TWO(
        p7 * 3 + p6 + p5 + p4 + p3 + p2 * 2 + p1 + p0 + q0 + q1 + q2 + q3 + q4,
        4);
261
    *op1 = ROUND_POWER_OF_TWO(p7 * 2 + p6 + p5 + p4 + p3 + p2 + p1 * 2 + p0 +
clang-format's avatar
clang-format committed
262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283
                                  q0 + q1 + q2 + q3 + q4 + q5,
                              4);
    *op0 = ROUND_POWER_OF_TWO(p7 + p6 + p5 + p4 + p3 + p2 + p1 + p0 * 2 + q0 +
                                  q1 + q2 + q3 + q4 + q5 + q6,
                              4);
    *oq0 = ROUND_POWER_OF_TWO(p6 + p5 + p4 + p3 + p2 + p1 + p0 + q0 * 2 + q1 +
                                  q2 + q3 + q4 + q5 + q6 + q7,
                              4);
    *oq1 = ROUND_POWER_OF_TWO(p5 + p4 + p3 + p2 + p1 + p0 + q0 + q1 * 2 + q2 +
                                  q3 + q4 + q5 + q6 + q7 * 2,
                              4);
    *oq2 = ROUND_POWER_OF_TWO(
        p4 + p3 + p2 + p1 + p0 + q0 + q1 + q2 * 2 + q3 + q4 + q5 + q6 + q7 * 3,
        4);
    *oq3 = ROUND_POWER_OF_TWO(
        p3 + p2 + p1 + p0 + q0 + q1 + q2 + q3 * 2 + q4 + q5 + q6 + q7 * 4, 4);
    *oq4 = ROUND_POWER_OF_TWO(
        p2 + p1 + p0 + q0 + q1 + q2 + q3 + q4 * 2 + q5 + q6 + q7 * 5, 4);
    *oq5 = ROUND_POWER_OF_TWO(
        p1 + p0 + q0 + q1 + q2 + q3 + q4 + q5 * 2 + q6 + q7 * 6, 4);
    *oq6 = ROUND_POWER_OF_TWO(
        p0 + q0 + q1 + q2 + q3 + q4 + q5 + q6 * 2 + q7 * 7, 4);
284
  } else {
285
    filter8(mask, thresh, flat, op3, op2, op1, op0, oq0, oq1, oq2, oq3);
286 287 288
  }
}

Adrian Grange's avatar
Adrian Grange committed
289
void aom_lpf_horizontal_16_c(uint8_t *s, int p, const uint8_t *blimit,
Jim Bankoski's avatar
Jim Bankoski committed
290 291
                             const uint8_t *limit, const uint8_t *thresh,
                             int count) {
292
  int i;
293

Dmitry Kovalev's avatar
Dmitry Kovalev committed
294 295
  // loop filter designed to work using chars so that we can make maximum use
  // of 8 bit simd instructions.
296
  for (i = 0; i < 8 * count; ++i) {
297 298
    const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
    const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
clang-format's avatar
clang-format committed
299 300
    const int8_t mask =
        filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3);
301
    const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3);
clang-format's avatar
clang-format committed
302 303 304 305 306 307 308 309
    const int8_t flat2 =
        flat_mask5(1, s[-8 * p], s[-7 * p], s[-6 * p], s[-5 * p], p0, q0,
                   s[4 * p], s[5 * p], s[6 * p], s[7 * p]);

    filter16(mask, *thresh, flat, flat2, s - 8 * p, s - 7 * p, s - 6 * p,
             s - 5 * p, s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p, s,
             s + 1 * p, s + 2 * p, s + 3 * p, s + 4 * p, s + 5 * p, s + 6 * p,
             s + 7 * p);
310
    ++s;
311
  }
312
}
313

clang-format's avatar
clang-format committed
314 315
static void mb_lpf_vertical_edge_w(uint8_t *s, int p, const uint8_t *blimit,
                                   const uint8_t *limit, const uint8_t *thresh,
316
                                   int count) {
317 318
  int i;

319
  for (i = 0; i < count; ++i) {
320
    const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
clang-format's avatar
clang-format committed
321 322 323
    const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
    const int8_t mask =
        filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3);
324
    const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3);
clang-format's avatar
clang-format committed
325 326
    const int8_t flat2 = flat_mask5(1, s[-8], s[-7], s[-6], s[-5], p0, q0, s[4],
                                    s[5], s[6], s[7]);
327

clang-format's avatar
clang-format committed
328 329 330
    filter16(mask, *thresh, flat, flat2, s - 8, s - 7, s - 6, s - 5, s - 4,
             s - 3, s - 2, s - 1, s, s + 1, s + 2, s + 3, s + 4, s + 5, s + 6,
             s + 7);
331
    s += p;
332
  }
333
}
334

Adrian Grange's avatar
Adrian Grange committed
335
void aom_lpf_vertical_16_c(uint8_t *s, int p, const uint8_t *blimit,
Jim Bankoski's avatar
Jim Bankoski committed
336
                           const uint8_t *limit, const uint8_t *thresh) {
337 338 339
  mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 8);
}

Adrian Grange's avatar
Adrian Grange committed
340
void aom_lpf_vertical_16_dual_c(uint8_t *s, int p, const uint8_t *blimit,
Jim Bankoski's avatar
Jim Bankoski committed
341
                                const uint8_t *limit, const uint8_t *thresh) {
342
  mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 16);
343
}
344

345
#if CONFIG_AOM_HIGHBITDEPTH
346
// Should we apply any filter at all: 11111111 yes, 00000000 no ?
347
static INLINE int8_t highbd_filter_mask(uint8_t limit, uint8_t blimit,
clang-format's avatar
clang-format committed
348 349
                                        uint16_t p3, uint16_t p2, uint16_t p1,
                                        uint16_t p0, uint16_t q0, uint16_t q1,
350
                                        uint16_t q2, uint16_t q3, int bd) {
351 352 353 354 355 356 357 358 359
  int8_t mask = 0;
  int16_t limit16 = (uint16_t)limit << (bd - 8);
  int16_t blimit16 = (uint16_t)blimit << (bd - 8);
  mask |= (abs(p3 - p2) > limit16) * -1;
  mask |= (abs(p2 - p1) > limit16) * -1;
  mask |= (abs(p1 - p0) > limit16) * -1;
  mask |= (abs(q1 - q0) > limit16) * -1;
  mask |= (abs(q2 - q1) > limit16) * -1;
  mask |= (abs(q3 - q2) > limit16) * -1;
clang-format's avatar
clang-format committed
360
  mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit16) * -1;
361 362 363
  return ~mask;
}

clang-format's avatar
clang-format committed
364 365 366 367
static INLINE int8_t highbd_flat_mask4(uint8_t thresh, uint16_t p3, uint16_t p2,
                                       uint16_t p1, uint16_t p0, uint16_t q0,
                                       uint16_t q1, uint16_t q2, uint16_t q3,
                                       int bd) {
368 369 370 371 372 373 374 375 376 377 378
  int8_t mask = 0;
  int16_t thresh16 = (uint16_t)thresh << (bd - 8);
  mask |= (abs(p1 - p0) > thresh16) * -1;
  mask |= (abs(q1 - q0) > thresh16) * -1;
  mask |= (abs(p2 - p0) > thresh16) * -1;
  mask |= (abs(q2 - q0) > thresh16) * -1;
  mask |= (abs(p3 - p0) > thresh16) * -1;
  mask |= (abs(q3 - q0) > thresh16) * -1;
  return ~mask;
}

clang-format's avatar
clang-format committed
379 380 381
static INLINE int8_t highbd_flat_mask5(uint8_t thresh, uint16_t p4, uint16_t p3,
                                       uint16_t p2, uint16_t p1, uint16_t p0,
                                       uint16_t q0, uint16_t q1, uint16_t q2,
382 383
                                       uint16_t q3, uint16_t q4, int bd) {
  int8_t mask = ~highbd_flat_mask4(thresh, p3, p2, p1, p0, q0, q1, q2, q3, bd);
384 385 386 387 388 389 390 391
  int16_t thresh16 = (uint16_t)thresh << (bd - 8);
  mask |= (abs(p4 - p0) > thresh16) * -1;
  mask |= (abs(q4 - q0) > thresh16) * -1;
  return ~mask;
}

// Is there high edge variance internal edge:
// 11111111_11111111 yes, 00000000_00000000 no ?
392 393
static INLINE int16_t highbd_hev_mask(uint8_t thresh, uint16_t p1, uint16_t p0,
                                      uint16_t q0, uint16_t q1, int bd) {
394 395 396 397 398 399 400
  int16_t hev = 0;
  int16_t thresh16 = (uint16_t)thresh << (bd - 8);
  hev |= (abs(p1 - p0) > thresh16) * -1;
  hev |= (abs(q1 - q0) > thresh16) * -1;
  return hev;
}

401 402 403
static INLINE void highbd_filter4(int8_t mask, uint8_t thresh, uint16_t *op1,
                                  uint16_t *op0, uint16_t *oq0, uint16_t *oq1,
                                  int bd) {
404 405 406 407 408 409 410 411
  int16_t filter1, filter2;
  // ^0x80 equivalent to subtracting 0x80 from the values to turn them
  // into -128 to +127 instead of 0 to 255.
  int shift = bd - 8;
  const int16_t ps1 = (int16_t)*op1 - (0x80 << shift);
  const int16_t ps0 = (int16_t)*op0 - (0x80 << shift);
  const int16_t qs0 = (int16_t)*oq0 - (0x80 << shift);
  const int16_t qs1 = (int16_t)*oq1 - (0x80 << shift);
412
  const uint16_t hev = highbd_hev_mask(thresh, *op1, *op0, *oq0, *oq1, bd);
413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435

  // Add outer taps if we have high edge variance.
  int16_t filter = signed_char_clamp_high(ps1 - qs1, bd) & hev;

  // Inner taps.
  filter = signed_char_clamp_high(filter + 3 * (qs0 - ps0), bd) & mask;

  // Save bottom 3 bits so that we round one side +4 and the other +3
  // if it equals 4 we'll set to adjust by -1 to account for the fact
  // we'd round 3 the other way.
  filter1 = signed_char_clamp_high(filter + 4, bd) >> 3;
  filter2 = signed_char_clamp_high(filter + 3, bd) >> 3;

  *oq0 = signed_char_clamp_high(qs0 - filter1, bd) + (0x80 << shift);
  *op0 = signed_char_clamp_high(ps0 + filter2, bd) + (0x80 << shift);

  // Outer tap adjustments.
  filter = ROUND_POWER_OF_TWO(filter1, 1) & ~hev;

  *oq1 = signed_char_clamp_high(qs1 - filter, bd) + (0x80 << shift);
  *op1 = signed_char_clamp_high(ps1 + filter, bd) + (0x80 << shift);
}

Adrian Grange's avatar
Adrian Grange committed
436
void aom_highbd_lpf_horizontal_4_c(uint16_t *s, int p /* pitch */,
437 438 439 440 441 442 443 444 445 446 447 448 449 450 451
                                   const uint8_t *blimit, const uint8_t *limit,
                                   const uint8_t *thresh, int count, int bd) {
  int i;

  // loop filter designed to work using chars so that we can make maximum use
  // of 8 bit simd instructions.
  for (i = 0; i < 8 * count; ++i) {
    const uint16_t p3 = s[-4 * p];
    const uint16_t p2 = s[-3 * p];
    const uint16_t p1 = s[-2 * p];
    const uint16_t p0 = s[-p];
    const uint16_t q0 = s[0 * p];
    const uint16_t q1 = s[1 * p];
    const uint16_t q2 = s[2 * p];
    const uint16_t q3 = s[3 * p];
clang-format's avatar
clang-format committed
452 453
    const int8_t mask =
        highbd_filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3, bd);
454
    highbd_filter4(mask, *thresh, s - 2 * p, s - 1 * p, s, s + 1 * p, bd);
455 456 457 458
    ++s;
  }
}

Adrian Grange's avatar
Adrian Grange committed
459
void aom_highbd_lpf_horizontal_4_dual_c(
clang-format's avatar
clang-format committed
460 461 462
    uint16_t *s, int p, const uint8_t *blimit0, const uint8_t *limit0,
    const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1,
    const uint8_t *thresh1, int bd) {
Adrian Grange's avatar
Adrian Grange committed
463 464
  aom_highbd_lpf_horizontal_4_c(s, p, blimit0, limit0, thresh0, 1, bd);
  aom_highbd_lpf_horizontal_4_c(s + 8, p, blimit1, limit1, thresh1, 1, bd);
465 466
}

Adrian Grange's avatar
Adrian Grange committed
467
void aom_highbd_lpf_vertical_4_c(uint16_t *s, int pitch, const uint8_t *blimit,
468 469 470 471 472 473 474 475
                                 const uint8_t *limit, const uint8_t *thresh,
                                 int count, int bd) {
  int i;

  // loop filter designed to work using chars so that we can make maximum use
  // of 8 bit simd instructions.
  for (i = 0; i < 8 * count; ++i) {
    const uint16_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
clang-format's avatar
clang-format committed
476 477 478
    const uint16_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
    const int8_t mask =
        highbd_filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3, bd);
479
    highbd_filter4(mask, *thresh, s - 2, s - 1, s, s + 1, bd);
480 481 482 483
    s += pitch;
  }
}

Adrian Grange's avatar
Adrian Grange committed
484
void aom_highbd_lpf_vertical_4_dual_c(
clang-format's avatar
clang-format committed
485 486 487
    uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0,
    const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1,
    const uint8_t *thresh1, int bd) {
Adrian Grange's avatar
Adrian Grange committed
488 489
  aom_highbd_lpf_vertical_4_c(s, pitch, blimit0, limit0, thresh0, 1, bd);
  aom_highbd_lpf_vertical_4_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1, 1,
clang-format's avatar
clang-format committed
490
                              bd);
491 492
}

493
static INLINE void highbd_filter8(int8_t mask, uint8_t thresh, uint8_t flat,
clang-format's avatar
clang-format committed
494 495
                                  uint16_t *op3, uint16_t *op2, uint16_t *op1,
                                  uint16_t *op0, uint16_t *oq0, uint16_t *oq1,
496
                                  uint16_t *oq2, uint16_t *oq3, int bd) {
497 498 499 500 501 502 503 504 505 506 507 508
  if (flat && mask) {
    const uint16_t p3 = *op3, p2 = *op2, p1 = *op1, p0 = *op0;
    const uint16_t q0 = *oq0, q1 = *oq1, q2 = *oq2, q3 = *oq3;

    // 7-tap filter [1, 1, 1, 2, 1, 1, 1]
    *op2 = ROUND_POWER_OF_TWO(p3 + p3 + p3 + 2 * p2 + p1 + p0 + q0, 3);
    *op1 = ROUND_POWER_OF_TWO(p3 + p3 + p2 + 2 * p1 + p0 + q0 + q1, 3);
    *op0 = ROUND_POWER_OF_TWO(p3 + p2 + p1 + 2 * p0 + q0 + q1 + q2, 3);
    *oq0 = ROUND_POWER_OF_TWO(p2 + p1 + p0 + 2 * q0 + q1 + q2 + q3, 3);
    *oq1 = ROUND_POWER_OF_TWO(p1 + p0 + q0 + 2 * q1 + q2 + q3 + q3, 3);
    *oq2 = ROUND_POWER_OF_TWO(p0 + q0 + q1 + 2 * q2 + q3 + q3 + q3, 3);
  } else {
clang-format's avatar
clang-format committed
509
    highbd_filter4(mask, thresh, op1, op0, oq0, oq1, bd);
510 511 512
  }
}

Adrian Grange's avatar
Adrian Grange committed
513
void aom_highbd_lpf_horizontal_8_c(uint16_t *s, int p, const uint8_t *blimit,
514 515 516 517 518 519 520 521 522 523
                                   const uint8_t *limit, const uint8_t *thresh,
                                   int count, int bd) {
  int i;

  // loop filter designed to work using chars so that we can make maximum use
  // of 8 bit simd instructions.
  for (i = 0; i < 8 * count; ++i) {
    const uint16_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
    const uint16_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];

clang-format's avatar
clang-format committed
524 525 526 527 528 529
    const int8_t mask =
        highbd_filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3, bd);
    const int8_t flat =
        highbd_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, bd);
    highbd_filter8(mask, *thresh, flat, s - 4 * p, s - 3 * p, s - 2 * p,
                   s - 1 * p, s, s + 1 * p, s + 2 * p, s + 3 * p, bd);
530 531 532 533
    ++s;
  }
}

Adrian Grange's avatar
Adrian Grange committed
534
void aom_highbd_lpf_horizontal_8_dual_c(
clang-format's avatar
clang-format committed
535 536 537
    uint16_t *s, int p, const uint8_t *blimit0, const uint8_t *limit0,
    const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1,
    const uint8_t *thresh1, int bd) {
Adrian Grange's avatar
Adrian Grange committed
538 539
  aom_highbd_lpf_horizontal_8_c(s, p, blimit0, limit0, thresh0, 1, bd);
  aom_highbd_lpf_horizontal_8_c(s + 8, p, blimit1, limit1, thresh1, 1, bd);
540 541
}

Adrian Grange's avatar
Adrian Grange committed
542
void aom_highbd_lpf_vertical_8_c(uint16_t *s, int pitch, const uint8_t *blimit,
543 544 545 546 547 548 549
                                 const uint8_t *limit, const uint8_t *thresh,
                                 int count, int bd) {
  int i;

  for (i = 0; i < 8 * count; ++i) {
    const uint16_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
    const uint16_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
clang-format's avatar
clang-format committed
550 551 552 553 554 555
    const int8_t mask =
        highbd_filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3, bd);
    const int8_t flat =
        highbd_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, bd);
    highbd_filter8(mask, *thresh, flat, s - 4, s - 3, s - 2, s - 1, s, s + 1,
                   s + 2, s + 3, bd);
556 557 558 559
    s += pitch;
  }
}

Adrian Grange's avatar
Adrian Grange committed
560
void aom_highbd_lpf_vertical_8_dual_c(
clang-format's avatar
clang-format committed
561 562 563
    uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0,
    const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1,
    const uint8_t *thresh1, int bd) {
Adrian Grange's avatar
Adrian Grange committed
564 565
  aom_highbd_lpf_vertical_8_c(s, pitch, blimit0, limit0, thresh0, 1, bd);
  aom_highbd_lpf_vertical_8_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1, 1,
clang-format's avatar
clang-format committed
566 567 568 569 570 571 572 573 574
                              bd);
}

static INLINE void highbd_filter16(int8_t mask, uint8_t thresh, uint8_t flat,
                                   uint8_t flat2, uint16_t *op7, uint16_t *op6,
                                   uint16_t *op5, uint16_t *op4, uint16_t *op3,
                                   uint16_t *op2, uint16_t *op1, uint16_t *op0,
                                   uint16_t *oq0, uint16_t *oq1, uint16_t *oq2,
                                   uint16_t *oq3, uint16_t *oq4, uint16_t *oq5,
575
                                   uint16_t *oq6, uint16_t *oq7, int bd) {
576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594
  if (flat2 && flat && mask) {
    const uint16_t p7 = *op7;
    const uint16_t p6 = *op6;
    const uint16_t p5 = *op5;
    const uint16_t p4 = *op4;
    const uint16_t p3 = *op3;
    const uint16_t p2 = *op2;
    const uint16_t p1 = *op1;
    const uint16_t p0 = *op0;
    const uint16_t q0 = *oq0;
    const uint16_t q1 = *oq1;
    const uint16_t q2 = *oq2;
    const uint16_t q3 = *oq3;
    const uint16_t q4 = *oq4;
    const uint16_t q5 = *oq5;
    const uint16_t q6 = *oq6;
    const uint16_t q7 = *oq7;

    // 15-tap filter [1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1]
clang-format's avatar
clang-format committed
595 596 597 598 599 600 601 602 603 604 605
    *op6 = ROUND_POWER_OF_TWO(
        p7 * 7 + p6 * 2 + p5 + p4 + p3 + p2 + p1 + p0 + q0, 4);
    *op5 = ROUND_POWER_OF_TWO(
        p7 * 6 + p6 + p5 * 2 + p4 + p3 + p2 + p1 + p0 + q0 + q1, 4);
    *op4 = ROUND_POWER_OF_TWO(
        p7 * 5 + p6 + p5 + p4 * 2 + p3 + p2 + p1 + p0 + q0 + q1 + q2, 4);
    *op3 = ROUND_POWER_OF_TWO(
        p7 * 4 + p6 + p5 + p4 + p3 * 2 + p2 + p1 + p0 + q0 + q1 + q2 + q3, 4);
    *op2 = ROUND_POWER_OF_TWO(
        p7 * 3 + p6 + p5 + p4 + p3 + p2 * 2 + p1 + p0 + q0 + q1 + q2 + q3 + q4,
        4);
606
    *op1 = ROUND_POWER_OF_TWO(p7 * 2 + p6 + p5 + p4 + p3 + p2 + p1 * 2 + p0 +
clang-format's avatar
clang-format committed
607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628
                                  q0 + q1 + q2 + q3 + q4 + q5,
                              4);
    *op0 = ROUND_POWER_OF_TWO(p7 + p6 + p5 + p4 + p3 + p2 + p1 + p0 * 2 + q0 +
                                  q1 + q2 + q3 + q4 + q5 + q6,
                              4);
    *oq0 = ROUND_POWER_OF_TWO(p6 + p5 + p4 + p3 + p2 + p1 + p0 + q0 * 2 + q1 +
                                  q2 + q3 + q4 + q5 + q6 + q7,
                              4);
    *oq1 = ROUND_POWER_OF_TWO(p5 + p4 + p3 + p2 + p1 + p0 + q0 + q1 * 2 + q2 +
                                  q3 + q4 + q5 + q6 + q7 * 2,
                              4);
    *oq2 = ROUND_POWER_OF_TWO(
        p4 + p3 + p2 + p1 + p0 + q0 + q1 + q2 * 2 + q3 + q4 + q5 + q6 + q7 * 3,
        4);
    *oq3 = ROUND_POWER_OF_TWO(
        p3 + p2 + p1 + p0 + q0 + q1 + q2 + q3 * 2 + q4 + q5 + q6 + q7 * 4, 4);
    *oq4 = ROUND_POWER_OF_TWO(
        p2 + p1 + p0 + q0 + q1 + q2 + q3 + q4 * 2 + q5 + q6 + q7 * 5, 4);
    *oq5 = ROUND_POWER_OF_TWO(
        p1 + p0 + q0 + q1 + q2 + q3 + q4 + q5 * 2 + q6 + q7 * 6, 4);
    *oq6 = ROUND_POWER_OF_TWO(
        p0 + q0 + q1 + q2 + q3 + q4 + q5 + q6 * 2 + q7 * 7, 4);
629
  } else {
630 631
    highbd_filter8(mask, thresh, flat, op3, op2, op1, op0, oq0, oq1, oq2, oq3,
                   bd);
632 633 634
  }
}

Adrian Grange's avatar
Adrian Grange committed
635
void aom_highbd_lpf_horizontal_16_c(uint16_t *s, int p, const uint8_t *blimit,
636 637 638 639 640 641 642 643 644 645 646 647 648 649 650
                                    const uint8_t *limit, const uint8_t *thresh,
                                    int count, int bd) {
  int i;

  // loop filter designed to work using chars so that we can make maximum use
  // of 8 bit simd instructions.
  for (i = 0; i < 8 * count; ++i) {
    const uint16_t p3 = s[-4 * p];
    const uint16_t p2 = s[-3 * p];
    const uint16_t p1 = s[-2 * p];
    const uint16_t p0 = s[-p];
    const uint16_t q0 = s[0 * p];
    const uint16_t q1 = s[1 * p];
    const uint16_t q2 = s[2 * p];
    const uint16_t q3 = s[3 * p];
clang-format's avatar
clang-format committed
651 652 653 654 655 656 657 658 659 660 661 662
    const int8_t mask =
        highbd_filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3, bd);
    const int8_t flat =
        highbd_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, bd);
    const int8_t flat2 =
        highbd_flat_mask5(1, s[-8 * p], s[-7 * p], s[-6 * p], s[-5 * p], p0, q0,
                          s[4 * p], s[5 * p], s[6 * p], s[7 * p], bd);

    highbd_filter16(mask, *thresh, flat, flat2, s - 8 * p, s - 7 * p, s - 6 * p,
                    s - 5 * p, s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p, s,
                    s + 1 * p, s + 2 * p, s + 3 * p, s + 4 * p, s + 5 * p,
                    s + 6 * p, s + 7 * p, bd);
663 664 665 666
    ++s;
  }
}

667 668 669
static void highbd_mb_lpf_vertical_edge_w(uint16_t *s, int p,
                                          const uint8_t *blimit,
                                          const uint8_t *limit,
clang-format's avatar
clang-format committed
670 671
                                          const uint8_t *thresh, int count,
                                          int bd) {
672 673 674 675 676 677 678 679 680 681 682
  int i;

  for (i = 0; i < count; ++i) {
    const uint16_t p3 = s[-4];
    const uint16_t p2 = s[-3];
    const uint16_t p1 = s[-2];
    const uint16_t p0 = s[-1];
    const uint16_t q0 = s[0];
    const uint16_t q1 = s[1];
    const uint16_t q2 = s[2];
    const uint16_t q3 = s[3];
clang-format's avatar
clang-format committed
683 684 685 686
    const int8_t mask =
        highbd_filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3, bd);
    const int8_t flat =
        highbd_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, bd);
687 688 689
    const int8_t flat2 = highbd_flat_mask5(1, s[-8], s[-7], s[-6], s[-5], p0,
                                           q0, s[4], s[5], s[6], s[7], bd);

clang-format's avatar
clang-format committed
690 691 692
    highbd_filter16(mask, *thresh, flat, flat2, s - 8, s - 7, s - 6, s - 5,
                    s - 4, s - 3, s - 2, s - 1, s, s + 1, s + 2, s + 3, s + 4,
                    s + 5, s + 6, s + 7, bd);
693 694 695 696
    s += p;
  }
}

Adrian Grange's avatar
Adrian Grange committed
697
void aom_highbd_lpf_vertical_16_c(uint16_t *s, int p, const uint8_t *blimit,
698 699
                                  const uint8_t *limit, const uint8_t *thresh,
                                  int bd) {
700
  highbd_mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 8, bd);
701 702
}

Adrian Grange's avatar
Adrian Grange committed
703
void aom_highbd_lpf_vertical_16_dual_c(uint16_t *s, int p,
704 705
                                       const uint8_t *blimit,
                                       const uint8_t *limit,
clang-format's avatar
clang-format committed
706
                                       const uint8_t *thresh, int bd) {
707
  highbd_mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 16, bd);
708
}
709
#endif  // CONFIG_AOM_HIGHBITDEPTH