reconinter.c 116 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
12
13
 */

#include <assert.h>

Yaowu Xu's avatar
Yaowu Xu committed
14
15
16
#include "./aom_scale_rtcd.h"
#include "./aom_dsp_rtcd.h"
#include "./aom_config.h"
Jingning Han's avatar
Jingning Han committed
17

Yaowu Xu's avatar
Yaowu Xu committed
18
#include "aom/aom_integer.h"
19
#include "aom_dsp/blend.h"
Jingning Han's avatar
Jingning Han committed
20

21
22
23
#include "av1/common/blockd.h"
#include "av1/common/reconinter.h"
#include "av1/common/reconintra.h"
Yue Chen's avatar
Yue Chen committed
24
#if CONFIG_MOTION_VAR
25
#include "av1/common/onyxc_int.h"
Yue Chen's avatar
Yue Chen committed
26
#endif  // CONFIG_MOTION_VAR
Jingning Han's avatar
Jingning Han committed
27

28
#if CONFIG_EXT_INTER
29

clang-format's avatar
clang-format committed
30
#define NSMOOTHERS 1
31
static int get_masked_weight(int m, int smoothness) {
clang-format's avatar
clang-format committed
32
33
#define SMOOTHER_LEN 32
  static const uint8_t smoothfn[NSMOOTHERS][2 * SMOOTHER_LEN + 1] = { {
clang-format's avatar
clang-format committed
34
35
36
37
      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
      0,  0,  0,  0,  0,  0,  0,  0,  0,  1,  2,  4,  7,  13, 21, 32, 43,
      51, 57, 60, 62, 63, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64,
      64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64,
clang-format's avatar
clang-format committed
38
  } };
39
40
41
42
43
  if (m < -SMOOTHER_LEN)
    return 0;
  else if (m > SMOOTHER_LEN)
    return (1 << WEDGE_WEIGHT_BITS);
  else
44
    return smoothfn[smoothness][m + SMOOTHER_LEN];
45
46
}

47
// [smoother][negative][direction]
clang-format's avatar
clang-format committed
48
49
50
DECLARE_ALIGNED(16, static uint8_t,
                wedge_mask_obl[NSMOOTHERS][2][WEDGE_DIRECTIONS]
                              [MASK_MASTER_SIZE * MASK_MASTER_SIZE]);
51

clang-format's avatar
clang-format committed
52
53
DECLARE_ALIGNED(16, static uint8_t,
                wedge_signflip_lookup[BLOCK_SIZES][MAX_WEDGE_TYPES]);
54

55
56
// 3 * MAX_WEDGE_SQUARE is an easy to compute and fairly tight upper bound
// on the sum of all mask sizes up to an including MAX_WEDGE_SQUARE.
clang-format's avatar
clang-format committed
57
58
DECLARE_ALIGNED(16, static uint8_t,
                wedge_mask_buf[2 * MAX_WEDGE_TYPES * 3 * MAX_WEDGE_SQUARE]);
59
60
61

static wedge_masks_type wedge_masks[BLOCK_SIZES][2];

62
63
64
// Some unused wedge codebooks left temporarily to facilitate experiments.
// To be removed when setteld.
static wedge_code_type wedge_codebook_8_hgtw[8] = {
clang-format's avatar
clang-format committed
65
66
67
68
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_OBLIQUE27, 4, 2 },  { WEDGE_OBLIQUE27, 4, 6 },
  { WEDGE_OBLIQUE153, 4, 2 }, { WEDGE_OBLIQUE153, 4, 6 },
69
70
};

71
static wedge_code_type wedge_codebook_8_hltw[8] = {
clang-format's avatar
clang-format committed
72
73
74
75
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_OBLIQUE63, 2, 4 },  { WEDGE_OBLIQUE63, 6, 4 },
  { WEDGE_OBLIQUE117, 2, 4 }, { WEDGE_OBLIQUE117, 6, 4 },
76
77
};

78
static wedge_code_type wedge_codebook_8_heqw[8] = {
clang-format's avatar
clang-format committed
79
80
81
82
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_HORIZONTAL, 4, 2 }, { WEDGE_HORIZONTAL, 4, 6 },
  { WEDGE_VERTICAL, 2, 4 },   { WEDGE_VERTICAL, 6, 4 },
83
84
};

85
86
#if !USE_LARGE_WEDGE_CODEBOOK
static const wedge_code_type wedge_codebook_16_hgtw[16] = {
clang-format's avatar
clang-format committed
87
88
89
90
91
92
93
94
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_HORIZONTAL, 4, 2 }, { WEDGE_HORIZONTAL, 4, 4 },
  { WEDGE_HORIZONTAL, 4, 6 }, { WEDGE_VERTICAL, 4, 4 },
  { WEDGE_OBLIQUE27, 4, 2 },  { WEDGE_OBLIQUE27, 4, 6 },
  { WEDGE_OBLIQUE153, 4, 2 }, { WEDGE_OBLIQUE153, 4, 6 },
  { WEDGE_OBLIQUE63, 2, 4 },  { WEDGE_OBLIQUE63, 6, 4 },
  { WEDGE_OBLIQUE117, 2, 4 }, { WEDGE_OBLIQUE117, 6, 4 },
Debargha Mukherjee's avatar
Debargha Mukherjee committed
95
};
96

97
static const wedge_code_type wedge_codebook_16_hltw[16] = {
clang-format's avatar
clang-format committed
98
99
100
101
102
103
104
105
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_VERTICAL, 2, 4 },   { WEDGE_VERTICAL, 4, 4 },
  { WEDGE_VERTICAL, 6, 4 },   { WEDGE_HORIZONTAL, 4, 4 },
  { WEDGE_OBLIQUE27, 4, 2 },  { WEDGE_OBLIQUE27, 4, 6 },
  { WEDGE_OBLIQUE153, 4, 2 }, { WEDGE_OBLIQUE153, 4, 6 },
  { WEDGE_OBLIQUE63, 2, 4 },  { WEDGE_OBLIQUE63, 6, 4 },
  { WEDGE_OBLIQUE117, 2, 4 }, { WEDGE_OBLIQUE117, 6, 4 },
106
107
};

108
static const wedge_code_type wedge_codebook_16_heqw[16] = {
clang-format's avatar
clang-format committed
109
110
111
112
113
114
115
116
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_HORIZONTAL, 4, 2 }, { WEDGE_HORIZONTAL, 4, 6 },
  { WEDGE_VERTICAL, 2, 4 },   { WEDGE_VERTICAL, 6, 4 },
  { WEDGE_OBLIQUE27, 4, 2 },  { WEDGE_OBLIQUE27, 4, 6 },
  { WEDGE_OBLIQUE153, 4, 2 }, { WEDGE_OBLIQUE153, 4, 6 },
  { WEDGE_OBLIQUE63, 2, 4 },  { WEDGE_OBLIQUE63, 6, 4 },
  { WEDGE_OBLIQUE117, 2, 4 }, { WEDGE_OBLIQUE117, 6, 4 },
117
118
};

119
const wedge_params_type wedge_params_lookup[BLOCK_SIZES] = {
120
121
122
123
124
#if CONFIG_CB4X4
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
#endif
clang-format's avatar
clang-format committed
125
126
127
128
129
130
131
132
133
134
135
136
137
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
  { 4, wedge_codebook_16_heqw, wedge_signflip_lookup[3], 0, wedge_masks[3] },
  { 4, wedge_codebook_16_hgtw, wedge_signflip_lookup[4], 0, wedge_masks[4] },
  { 4, wedge_codebook_16_hltw, wedge_signflip_lookup[5], 0, wedge_masks[5] },
  { 4, wedge_codebook_16_heqw, wedge_signflip_lookup[6], 0, wedge_masks[6] },
  { 4, wedge_codebook_16_hgtw, wedge_signflip_lookup[7], 0, wedge_masks[7] },
  { 4, wedge_codebook_16_hltw, wedge_signflip_lookup[8], 0, wedge_masks[8] },
  { 4, wedge_codebook_16_heqw, wedge_signflip_lookup[9], 0, wedge_masks[9] },
  { 0, wedge_codebook_8_hgtw, wedge_signflip_lookup[10], 0, wedge_masks[10] },
  { 0, wedge_codebook_8_hltw, wedge_signflip_lookup[11], 0, wedge_masks[11] },
  { 0, wedge_codebook_8_heqw, wedge_signflip_lookup[12], 0, wedge_masks[12] },
138
#if CONFIG_EXT_PARTITION
clang-format's avatar
clang-format committed
139
140
141
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
142
#endif  // CONFIG_EXT_PARTITION
143
144
};

145
146
147
#else

static const wedge_code_type wedge_codebook_32_hgtw[32] = {
clang-format's avatar
clang-format committed
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_HORIZONTAL, 4, 2 }, { WEDGE_HORIZONTAL, 4, 4 },
  { WEDGE_HORIZONTAL, 4, 6 }, { WEDGE_VERTICAL, 4, 4 },
  { WEDGE_OBLIQUE27, 4, 1 },  { WEDGE_OBLIQUE27, 4, 2 },
  { WEDGE_OBLIQUE27, 4, 3 },  { WEDGE_OBLIQUE27, 4, 5 },
  { WEDGE_OBLIQUE27, 4, 6 },  { WEDGE_OBLIQUE27, 4, 7 },
  { WEDGE_OBLIQUE153, 4, 1 }, { WEDGE_OBLIQUE153, 4, 2 },
  { WEDGE_OBLIQUE153, 4, 3 }, { WEDGE_OBLIQUE153, 4, 5 },
  { WEDGE_OBLIQUE153, 4, 6 }, { WEDGE_OBLIQUE153, 4, 7 },
  { WEDGE_OBLIQUE63, 1, 4 },  { WEDGE_OBLIQUE63, 2, 4 },
  { WEDGE_OBLIQUE63, 3, 4 },  { WEDGE_OBLIQUE63, 5, 4 },
  { WEDGE_OBLIQUE63, 6, 4 },  { WEDGE_OBLIQUE63, 7, 4 },
  { WEDGE_OBLIQUE117, 1, 4 }, { WEDGE_OBLIQUE117, 2, 4 },
  { WEDGE_OBLIQUE117, 3, 4 }, { WEDGE_OBLIQUE117, 5, 4 },
  { WEDGE_OBLIQUE117, 6, 4 }, { WEDGE_OBLIQUE117, 7, 4 },
164
165
};

166
static const wedge_code_type wedge_codebook_32_hltw[32] = {
clang-format's avatar
clang-format committed
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_VERTICAL, 2, 4 },   { WEDGE_VERTICAL, 4, 4 },
  { WEDGE_VERTICAL, 6, 4 },   { WEDGE_HORIZONTAL, 4, 4 },
  { WEDGE_OBLIQUE27, 4, 1 },  { WEDGE_OBLIQUE27, 4, 2 },
  { WEDGE_OBLIQUE27, 4, 3 },  { WEDGE_OBLIQUE27, 4, 5 },
  { WEDGE_OBLIQUE27, 4, 6 },  { WEDGE_OBLIQUE27, 4, 7 },
  { WEDGE_OBLIQUE153, 4, 1 }, { WEDGE_OBLIQUE153, 4, 2 },
  { WEDGE_OBLIQUE153, 4, 3 }, { WEDGE_OBLIQUE153, 4, 5 },
  { WEDGE_OBLIQUE153, 4, 6 }, { WEDGE_OBLIQUE153, 4, 7 },
  { WEDGE_OBLIQUE63, 1, 4 },  { WEDGE_OBLIQUE63, 2, 4 },
  { WEDGE_OBLIQUE63, 3, 4 },  { WEDGE_OBLIQUE63, 5, 4 },
  { WEDGE_OBLIQUE63, 6, 4 },  { WEDGE_OBLIQUE63, 7, 4 },
  { WEDGE_OBLIQUE117, 1, 4 }, { WEDGE_OBLIQUE117, 2, 4 },
  { WEDGE_OBLIQUE117, 3, 4 }, { WEDGE_OBLIQUE117, 5, 4 },
  { WEDGE_OBLIQUE117, 6, 4 }, { WEDGE_OBLIQUE117, 7, 4 },
183
184
};

185
static const wedge_code_type wedge_codebook_32_heqw[32] = {
clang-format's avatar
clang-format committed
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
  { WEDGE_OBLIQUE27, 4, 4 },  { WEDGE_OBLIQUE63, 4, 4 },
  { WEDGE_OBLIQUE117, 4, 4 }, { WEDGE_OBLIQUE153, 4, 4 },
  { WEDGE_HORIZONTAL, 4, 2 }, { WEDGE_HORIZONTAL, 4, 6 },
  { WEDGE_VERTICAL, 2, 4 },   { WEDGE_VERTICAL, 6, 4 },
  { WEDGE_OBLIQUE27, 4, 1 },  { WEDGE_OBLIQUE27, 4, 2 },
  { WEDGE_OBLIQUE27, 4, 3 },  { WEDGE_OBLIQUE27, 4, 5 },
  { WEDGE_OBLIQUE27, 4, 6 },  { WEDGE_OBLIQUE27, 4, 7 },
  { WEDGE_OBLIQUE153, 4, 1 }, { WEDGE_OBLIQUE153, 4, 2 },
  { WEDGE_OBLIQUE153, 4, 3 }, { WEDGE_OBLIQUE153, 4, 5 },
  { WEDGE_OBLIQUE153, 4, 6 }, { WEDGE_OBLIQUE153, 4, 7 },
  { WEDGE_OBLIQUE63, 1, 4 },  { WEDGE_OBLIQUE63, 2, 4 },
  { WEDGE_OBLIQUE63, 3, 4 },  { WEDGE_OBLIQUE63, 5, 4 },
  { WEDGE_OBLIQUE63, 6, 4 },  { WEDGE_OBLIQUE63, 7, 4 },
  { WEDGE_OBLIQUE117, 1, 4 }, { WEDGE_OBLIQUE117, 2, 4 },
  { WEDGE_OBLIQUE117, 3, 4 }, { WEDGE_OBLIQUE117, 5, 4 },
  { WEDGE_OBLIQUE117, 6, 4 }, { WEDGE_OBLIQUE117, 7, 4 },
202
203
};

204
const wedge_params_type wedge_params_lookup[BLOCK_SIZES] = {
205
206
207
208
209
#if CONFIG_CB4X4
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
#endif
clang-format's avatar
clang-format committed
210
211
212
213
214
215
216
217
218
219
220
221
222
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
  { 5, wedge_codebook_32_heqw, wedge_signflip_lookup[3], 0, wedge_masks[3] },
  { 5, wedge_codebook_32_hgtw, wedge_signflip_lookup[4], 0, wedge_masks[4] },
  { 5, wedge_codebook_32_hltw, wedge_signflip_lookup[5], 0, wedge_masks[5] },
  { 5, wedge_codebook_32_heqw, wedge_signflip_lookup[6], 0, wedge_masks[6] },
  { 5, wedge_codebook_32_hgtw, wedge_signflip_lookup[7], 0, wedge_masks[7] },
  { 5, wedge_codebook_32_hltw, wedge_signflip_lookup[8], 0, wedge_masks[8] },
  { 5, wedge_codebook_32_heqw, wedge_signflip_lookup[9], 0, wedge_masks[9] },
  { 0, wedge_codebook_8_hgtw, wedge_signflip_lookup[10], 0, wedge_masks[10] },
  { 0, wedge_codebook_8_hltw, wedge_signflip_lookup[11], 0, wedge_masks[11] },
  { 0, wedge_codebook_8_heqw, wedge_signflip_lookup[12], 0, wedge_masks[12] },
Debargha Mukherjee's avatar
Debargha Mukherjee committed
223
#if CONFIG_EXT_PARTITION
clang-format's avatar
clang-format committed
224
225
226
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
Debargha Mukherjee's avatar
Debargha Mukherjee committed
227
#endif  // CONFIG_EXT_PARTITION
228
};
229
#endif  // USE_LARGE_WEDGE_CODEBOOK
230

clang-format's avatar
clang-format committed
231
static const uint8_t *get_wedge_mask_inplace(int wedge_index, int neg,
232
                                             BLOCK_SIZE sb_type) {
Debargha Mukherjee's avatar
Debargha Mukherjee committed
233
  const uint8_t *master;
234
235
  const int bh = block_size_high[sb_type];
  const int bw = block_size_wide[sb_type];
236
237
238
  const wedge_code_type *a =
      wedge_params_lookup[sb_type].codebook + wedge_index;
  const int smoother = wedge_params_lookup[sb_type].smoother;
239
  int woff, hoff;
240
241
242
243
244
245
246
  const uint8_t wsignflip = wedge_params_lookup[sb_type].signflip[wedge_index];

  assert(wedge_index >= 0 &&
         wedge_index < (1 << get_wedge_bits_lookup(sb_type)));
  woff = (a->x_offset * bw) >> 3;
  hoff = (a->y_offset * bh) >> 3;
  master = wedge_mask_obl[smoother][neg ^ wsignflip][a->direction] +
clang-format's avatar
clang-format committed
247
248
           MASK_MASTER_STRIDE * (MASK_MASTER_SIZE / 2 - hoff) +
           MASK_MASTER_SIZE / 2 - woff;
Debargha Mukherjee's avatar
Debargha Mukherjee committed
249
250
251
  return master;
}

Yaowu Xu's avatar
Yaowu Xu committed
252
253
254
const uint8_t *av1_get_soft_mask(int wedge_index, int wedge_sign,
                                 BLOCK_SIZE sb_type, int offset_x,
                                 int offset_y) {
255
  const uint8_t *mask =
256
      get_wedge_mask_inplace(wedge_index, wedge_sign, sb_type);
clang-format's avatar
clang-format committed
257
  if (mask) mask -= (offset_x + offset_y * MASK_MASTER_STRIDE);
258
  return mask;
259
260
}

261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
#if CONFIG_COMPOUND_SEGMENT
static uint8_t *invert_mask(uint8_t *mask_inv_buffer, const uint8_t *const mask,
                            int h, int w, int stride) {
  int i, j;

  for (i = 0; i < h; ++i)
    for (j = 0; j < w; ++j) {
      mask_inv_buffer[i * stride + j] =
          AOM_BLEND_A64_MAX_ALPHA - mask[i * stride + j];
    }
  return mask_inv_buffer;
}
#endif  // CONFIG_COMPOUND_SEGMENT

const uint8_t *av1_get_compound_type_mask_inverse(
    const INTERINTER_COMPOUND_DATA *const comp_data,
#if CONFIG_COMPOUND_SEGMENT
    uint8_t *mask_buffer, int h, int w, int stride,
#endif
    BLOCK_SIZE sb_type) {
281
282
283
  assert(is_masked_compound_type(comp_data->type));
  switch (comp_data->type) {
    case COMPOUND_WEDGE:
284
285
      return av1_get_contiguous_soft_mask(comp_data->wedge_index,
                                          !comp_data->wedge_sign, sb_type);
286
287
#if CONFIG_COMPOUND_SEGMENT
    case COMPOUND_SEG:
288
      return invert_mask(mask_buffer, comp_data->seg_mask, h, w, stride);
289
#endif  // CONFIG_COMPOUND_SEGMENT
290
291
292
    default: assert(0); return NULL;
  }
}
293

294
295
296
297
298
299
300
const uint8_t *av1_get_compound_type_mask(
    const INTERINTER_COMPOUND_DATA *const comp_data, BLOCK_SIZE sb_type) {
  assert(is_masked_compound_type(comp_data->type));
  switch (comp_data->type) {
    case COMPOUND_WEDGE:
      return av1_get_contiguous_soft_mask(comp_data->wedge_index,
                                          comp_data->wedge_sign, sb_type);
301
#if CONFIG_COMPOUND_SEGMENT
302
303
304
305
306
307
308
    case COMPOUND_SEG: return comp_data->seg_mask;
#endif  // CONFIG_COMPOUND_SEGMENT
    default: assert(0); return NULL;
  }
}

#if CONFIG_COMPOUND_SEGMENT
309
310
311
#if COMPOUND_SEGMENT_TYPE == 0
static void uniform_mask(uint8_t *mask, int which_inverse, BLOCK_SIZE sb_type,
                         int h, int w, int mask_val) {
312
313
314
315
316
317
318
319
320
321
  int i, j;
  int block_stride = block_size_wide[sb_type];
  for (i = 0; i < h; ++i)
    for (j = 0; j < w; ++j) {
      mask[i * block_stride + j] =
          which_inverse ? AOM_BLEND_A64_MAX_ALPHA - mask_val : mask_val;
    }
}

void build_compound_seg_mask(uint8_t *mask, SEG_MASK_TYPE mask_type,
322
323
324
325
326
327
328
                             const uint8_t *src0, int src0_stride,
                             const uint8_t *src1, int src1_stride,
                             BLOCK_SIZE sb_type, int h, int w) {
  (void)src0;
  (void)src1;
  (void)src0_stride;
  (void)src1_stride;
329
330
331
332
333
  switch (mask_type) {
    case UNIFORM_45: uniform_mask(mask, 0, sb_type, h, w, 45); break;
    case UNIFORM_45_INV: uniform_mask(mask, 1, sb_type, h, w, 45); break;
    default: assert(0);
  }
334
}
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377

#if CONFIG_AOM_HIGHBITDEPTH
void build_compound_seg_mask_highbd(uint8_t *mask, SEG_MASK_TYPE mask_type,
                                    const uint8_t *src0, int src0_stride,
                                    const uint8_t *src1, int src1_stride,
                                    BLOCK_SIZE sb_type, int h, int w, int bd) {
  (void)src0;
  (void)src1;
  (void)src0_stride;
  (void)src1_stride;
  (void)bd;
  switch (mask_type) {
    case UNIFORM_45: uniform_mask(mask, 0, sb_type, h, w, 45); break;
    case UNIFORM_45_INV: uniform_mask(mask, 1, sb_type, h, w, 45); break;
    default: assert(0);
  }
}
#endif  // CONFIG_AOM_HIGHBITDEPTH

#elif COMPOUND_SEGMENT_TYPE == 1
#define DIFF_FACTOR 16
static void diffwtd_mask(uint8_t *mask, int which_inverse, int mask_base,
                         const uint8_t *src0, int src0_stride,
                         const uint8_t *src1, int src1_stride,
                         BLOCK_SIZE sb_type, int h, int w) {
  int i, j, m, diff;
  int block_stride = block_size_wide[sb_type];
  for (i = 0; i < h; ++i) {
    for (j = 0; j < w; ++j) {
      diff =
          abs((int)src0[i * src0_stride + j] - (int)src1[i * src1_stride + j]);
      m = clamp(mask_base + (diff / DIFF_FACTOR), 0, AOM_BLEND_A64_MAX_ALPHA);
      mask[i * block_stride + j] =
          which_inverse ? AOM_BLEND_A64_MAX_ALPHA - m : m;
    }
  }
}

void build_compound_seg_mask(uint8_t *mask, SEG_MASK_TYPE mask_type,
                             const uint8_t *src0, int src0_stride,
                             const uint8_t *src1, int src1_stride,
                             BLOCK_SIZE sb_type, int h, int w) {
  switch (mask_type) {
Yaowu Xu's avatar
Yaowu Xu committed
378
379
    case DIFFWTD_42:
      diffwtd_mask(mask, 0, 42, src0, src0_stride, src1, src1_stride, sb_type,
380
381
                   h, w);
      break;
Yaowu Xu's avatar
Yaowu Xu committed
382
383
    case DIFFWTD_42_INV:
      diffwtd_mask(mask, 1, 42, src0, src0_stride, src1, src1_stride, sb_type,
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
                   h, w);
      break;
    default: assert(0);
  }
}

#if CONFIG_AOM_HIGHBITDEPTH
static void diffwtd_mask_highbd(uint8_t *mask, int which_inverse, int mask_base,
                                const uint16_t *src0, int src0_stride,
                                const uint16_t *src1, int src1_stride,
                                BLOCK_SIZE sb_type, int h, int w, int bd) {
  int i, j, m, diff;
  int block_stride = block_size_wide[sb_type];
  for (i = 0; i < h; ++i) {
    for (j = 0; j < w; ++j) {
      diff = abs((int)src0[i * src0_stride + j] -
                 (int)src1[i * src1_stride + j]) >>
             (bd - 8);
      m = clamp(mask_base + (diff / DIFF_FACTOR), 0, AOM_BLEND_A64_MAX_ALPHA);
      mask[i * block_stride + j] =
          which_inverse ? AOM_BLEND_A64_MAX_ALPHA - m : m;
    }
  }
}

void build_compound_seg_mask_highbd(uint8_t *mask, SEG_MASK_TYPE mask_type,
                                    const uint8_t *src0, int src0_stride,
                                    const uint8_t *src1, int src1_stride,
                                    BLOCK_SIZE sb_type, int h, int w, int bd) {
  switch (mask_type) {
    case DIFFWTD_42:
      diffwtd_mask_highbd(mask, 0, 42, CONVERT_TO_SHORTPTR(src0), src0_stride,
                          CONVERT_TO_SHORTPTR(src1), src1_stride, sb_type, h, w,
                          bd);
      break;
    case DIFFWTD_42_INV:
      diffwtd_mask_highbd(mask, 1, 42, CONVERT_TO_SHORTPTR(src0), src0_stride,
                          CONVERT_TO_SHORTPTR(src1), src1_stride, sb_type, h, w,
                          bd);
      break;
    default: assert(0);
  }
}
#endif  // CONFIG_AOM_HIGHBITDEPTH
#endif  // COMPOUND_SEGMENT_TYPE
429
#endif  // CONFIG_COMPOUND_SEGMENT
430

431
static void init_wedge_master_masks() {
432
433
434
435
  int i, j, s;
  const int w = MASK_MASTER_SIZE;
  const int h = MASK_MASTER_SIZE;
  const int stride = MASK_MASTER_STRIDE;
clang-format's avatar
clang-format committed
436
  const int a[2] = { 2, 1 };
437
438
439
440
441
442
443
444
  const double asqrt = sqrt(a[0] * a[0] + a[1] * a[1]);
  for (s = 0; s < NSMOOTHERS; s++) {
    for (i = 0; i < h; ++i)
      for (j = 0; j < w; ++j) {
        int x = (2 * j + 1 - w);
        int y = (2 * i + 1 - h);
        int m = (int)rint((a[0] * x + a[1] * y) / asqrt);
        wedge_mask_obl[s][1][WEDGE_OBLIQUE63][i * stride + j] =
clang-format's avatar
clang-format committed
445
446
            wedge_mask_obl[s][1][WEDGE_OBLIQUE27][j * stride + i] =
                get_masked_weight(m, s);
447
        wedge_mask_obl[s][1][WEDGE_OBLIQUE117][i * stride + w - 1 - j] =
clang-format's avatar
clang-format committed
448
449
            wedge_mask_obl[s][1][WEDGE_OBLIQUE153][(w - 1 - j) * stride + i] =
                (1 << WEDGE_WEIGHT_BITS) - get_masked_weight(m, s);
450
        wedge_mask_obl[s][0][WEDGE_OBLIQUE63][i * stride + j] =
clang-format's avatar
clang-format committed
451
452
            wedge_mask_obl[s][0][WEDGE_OBLIQUE27][j * stride + i] =
                (1 << WEDGE_WEIGHT_BITS) - get_masked_weight(m, s);
453
        wedge_mask_obl[s][0][WEDGE_OBLIQUE117][i * stride + w - 1 - j] =
clang-format's avatar
clang-format committed
454
455
            wedge_mask_obl[s][0][WEDGE_OBLIQUE153][(w - 1 - j) * stride + i] =
                get_masked_weight(m, s);
456
        wedge_mask_obl[s][1][WEDGE_VERTICAL][i * stride + j] =
clang-format's avatar
clang-format committed
457
458
            wedge_mask_obl[s][1][WEDGE_HORIZONTAL][j * stride + i] =
                get_masked_weight(x, s);
459
        wedge_mask_obl[s][0][WEDGE_VERTICAL][i * stride + j] =
clang-format's avatar
clang-format committed
460
461
            wedge_mask_obl[s][0][WEDGE_HORIZONTAL][j * stride + i] =
                (1 << WEDGE_WEIGHT_BITS) - get_masked_weight(x, s);
462
463
      }
  }
464
465
466
467
468
469
470
471
472
}

// If the signs for the wedges for various blocksizes are
// inconsistent flip the sign flag. Do it only once for every
// wedge codebook.
static void init_wedge_signs() {
  BLOCK_SIZE sb_type;
  memset(wedge_signflip_lookup, 0, sizeof(wedge_signflip_lookup));
  for (sb_type = BLOCK_4X4; sb_type < BLOCK_SIZES; ++sb_type) {
473
474
    const int bw = block_size_wide[sb_type];
    const int bh = block_size_high[sb_type];
475
476
477
478
479
480
481
482
    const wedge_params_type wedge_params = wedge_params_lookup[sb_type];
    const int wbits = wedge_params.bits;
    const int wtypes = 1 << wbits;
    int i, w;
    if (wbits == 0) continue;
    for (w = 0; w < wtypes; ++w) {
      const uint8_t *mask = get_wedge_mask_inplace(w, 0, sb_type);
      int sum = 0;
clang-format's avatar
clang-format committed
483
484
      for (i = 0; i < bw; ++i) sum += mask[i];
      for (i = 0; i < bh; ++i) sum += mask[i * MASK_MASTER_STRIDE];
485
486
487
488
489
490
491
492
493
494
495
496
      sum = (sum + (bw + bh) / 2) / (bw + bh);
      wedge_params.signflip[w] = (sum < 32);
    }
  }
}

static void init_wedge_masks() {
  uint8_t *dst = wedge_mask_buf;
  BLOCK_SIZE bsize;
  memset(wedge_masks, 0, sizeof(wedge_masks));
  for (bsize = BLOCK_4X4; bsize < BLOCK_SIZES; ++bsize) {
    const uint8_t *mask;
497
498
    const int bw = block_size_wide[bsize];
    const int bh = block_size_high[bsize];
499
500
501
502
503
504
505
    const wedge_params_type *wedge_params = &wedge_params_lookup[bsize];
    const int wbits = wedge_params->bits;
    const int wtypes = 1 << wbits;
    int w;
    if (wbits == 0) continue;
    for (w = 0; w < wtypes; ++w) {
      mask = get_wedge_mask_inplace(w, 0, bsize);
Yaowu Xu's avatar
Yaowu Xu committed
506
      aom_convolve_copy(mask, MASK_MASTER_STRIDE, dst, bw, NULL, 0, NULL, 0, bw,
clang-format's avatar
clang-format committed
507
                        bh);
508
509
510
511
      wedge_params->masks[0][w] = dst;
      dst += bw * bh;

      mask = get_wedge_mask_inplace(w, 1, bsize);
Yaowu Xu's avatar
Yaowu Xu committed
512
      aom_convolve_copy(mask, MASK_MASTER_STRIDE, dst, bw, NULL, 0, NULL, 0, bw,
clang-format's avatar
clang-format committed
513
                        bh);
514
515
516
517
518
519
520
521
      wedge_params->masks[1][w] = dst;
      dst += bw * bh;
    }
    assert(sizeof(wedge_mask_buf) >= (size_t)(dst - wedge_mask_buf));
  }
}

// Equation of line: f(x, y) = a[0]*(x - a[2]*w/8) + a[1]*(y - a[3]*h/8) = 0
Yaowu Xu's avatar
Yaowu Xu committed
522
void av1_init_wedge_masks() {
523
  init_wedge_master_masks();
524
  init_wedge_signs();
525
  init_wedge_masks();
526
527
}

528
529
#if CONFIG_SUPERTX
static void build_masked_compound_wedge_extend(
clang-format's avatar
clang-format committed
530
    uint8_t *dst, int dst_stride, const uint8_t *src0, int src0_stride,
531
532
533
    const uint8_t *src1, int src1_stride,
    const INTERINTER_COMPOUND_DATA *const comp_data, BLOCK_SIZE sb_type,
    int wedge_offset_x, int wedge_offset_y, int h, int w) {
534
535
  const int subh = (2 << b_height_log2_lookup[sb_type]) == h;
  const int subw = (2 << b_width_log2_lookup[sb_type]) == w;
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
  const uint8_t *mask;
  size_t mask_stride;
  switch (comp_data->type) {
    case COMPOUND_WEDGE:
      mask = av1_get_soft_mask(comp_data->wedge_index, comp_data->wedge_sign,
                               sb_type, wedge_offset_x, wedge_offset_y);
      mask_stride = MASK_MASTER_STRIDE;
      break;
#if CONFIG_COMPOUND_SEGMENT
    case COMPOUND_SEG:
      mask = comp_data->seg_mask;
      mask_stride = block_size_wide[sb_type];
      break;
#endif
    default: assert(0); return;
  }
Yaowu Xu's avatar
Yaowu Xu committed
552
  aom_blend_a64_mask(dst, dst_stride, src0, src0_stride, src1, src1_stride,
553
                     mask, mask_stride, h, w, subh, subw);
554
555
}

Yaowu Xu's avatar
Yaowu Xu committed
556
#if CONFIG_AOM_HIGHBITDEPTH
557
static void build_masked_compound_wedge_extend_highbd(
clang-format's avatar
clang-format committed
558
    uint8_t *dst_8, int dst_stride, const uint8_t *src0_8, int src0_stride,
559
560
561
    const uint8_t *src1_8, int src1_stride,
    const INTERINTER_COMPOUND_DATA *const comp_data, BLOCK_SIZE sb_type,
    int wedge_offset_x, int wedge_offset_y, int h, int w, int bd) {
562
563
  const int subh = (2 << b_height_log2_lookup[sb_type]) == h;
  const int subw = (2 << b_width_log2_lookup[sb_type]) == w;
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
  const uint8_t *mask;
  size_t mask_stride;
  switch (comp_data->type) {
    case COMPOUND_WEDGE:
      mask = av1_get_soft_mask(comp_data->wedge_index, comp_data->wedge_sign,
                               sb_type, wedge_offset_x, wedge_offset_y);
      mask_stride = MASK_MASTER_STRIDE;
      break;
#if CONFIG_COMPOUND_SEGMENT
    case COMPOUND_SEG:
      mask = comp_data->seg_mask;
      mask_stride = block_size_wide[sb_type];
      break;
#endif
    default: assert(0); return;
  }
Yaowu Xu's avatar
Yaowu Xu committed
580
  aom_highbd_blend_a64_mask(dst_8, dst_stride, src0_8, src0_stride, src1_8,
581
582
                            src1_stride, mask, mask_stride, h, w, subh, subw,
                            bd);
583
}
Yaowu Xu's avatar
Yaowu Xu committed
584
#endif  // CONFIG_AOM_HIGHBITDEPTH
585
#else
586
587
588
589
590
static void build_masked_compound(
    uint8_t *dst, int dst_stride, const uint8_t *src0, int src0_stride,
    const uint8_t *src1, int src1_stride,
    const INTERINTER_COMPOUND_DATA *const comp_data, BLOCK_SIZE sb_type, int h,
    int w) {
591
592
593
594
  // Derive subsampling from h and w passed in. May be refactored to
  // pass in subsampling factors directly.
  const int subh = (2 << b_height_log2_lookup[sb_type]) == h;
  const int subw = (2 << b_width_log2_lookup[sb_type]) == w;
595
  const uint8_t *mask = av1_get_compound_type_mask(comp_data, sb_type);
Yaowu Xu's avatar
Yaowu Xu committed
596
  aom_blend_a64_mask(dst, dst_stride, src0, src0_stride, src1, src1_stride,
597
                     mask, block_size_wide[sb_type], h, w, subh, subw);
598
599
}

Yaowu Xu's avatar
Yaowu Xu committed
600
#if CONFIG_AOM_HIGHBITDEPTH
601
static void build_masked_compound_highbd(
clang-format's avatar
clang-format committed
602
    uint8_t *dst_8, int dst_stride, const uint8_t *src0_8, int src0_stride,
603
604
605
    const uint8_t *src1_8, int src1_stride,
    const INTERINTER_COMPOUND_DATA *const comp_data, BLOCK_SIZE sb_type, int h,
    int w, int bd) {
606
607
608
609
  // Derive subsampling from h and w passed in. May be refactored to
  // pass in subsampling factors directly.
  const int subh = (2 << b_height_log2_lookup[sb_type]) == h;
  const int subw = (2 << b_width_log2_lookup[sb_type]) == w;
610
611
612
  const uint8_t *mask = av1_get_compound_type_mask(comp_data, sb_type);
  // const uint8_t *mask =
  //     av1_get_contiguous_soft_mask(wedge_index, wedge_sign, sb_type);
613
614
615
  aom_highbd_blend_a64_mask(dst_8, dst_stride, src0_8, src0_stride, src1_8,
                            src1_stride, mask, block_size_wide[sb_type], h, w,
                            subh, subw, bd);
616
}
Yaowu Xu's avatar
Yaowu Xu committed
617
#endif  // CONFIG_AOM_HIGHBITDEPTH
618
#endif  // CONFIG_SUPERTX
619

Yaowu Xu's avatar
Yaowu Xu committed
620
621
622
623
624
void av1_make_masked_inter_predictor(const uint8_t *pre, int pre_stride,
                                     uint8_t *dst, int dst_stride,
                                     const int subpel_x, const int subpel_y,
                                     const struct scale_factors *sf, int w,
                                     int h,
625
#if CONFIG_DUAL_FILTER
James Zern's avatar
James Zern committed
626
                                     const InterpFilter *interp_filter,
627
#else
James Zern's avatar
James Zern committed
628
                                     const InterpFilter interp_filter,
629
#endif
Yaowu Xu's avatar
Yaowu Xu committed
630
                                     int xs, int ys,
631
#if CONFIG_SUPERTX
Yaowu Xu's avatar
Yaowu Xu committed
632
                                     int wedge_offset_x, int wedge_offset_y,
633
#endif  // CONFIG_SUPERTX
634
                                     int plane,
635
636
637
638
#if CONFIG_GLOBAL_MOTION
                                     int is_global, int p_col, int p_row,
                                     int ref,
#endif  // CONFIG_GLOBAL_MOTION
639
640
641
                                     MACROBLOCKD *xd) {
  MODE_INFO *mi = xd->mi[0];
  INTERINTER_COMPOUND_DATA *comp_data = &mi->mbmi.interinter_compound_data;
clang-format's avatar
clang-format committed
642
643
// The prediction filter types used here should be those for
// the second reference block.
Geza Lore's avatar
Geza Lore committed
644
#if CONFIG_DUAL_FILTER
James Zern's avatar
James Zern committed
645
  InterpFilter tmp_ipf[4] = {
Geza Lore's avatar
Geza Lore committed
646
    interp_filter[2], interp_filter[3], interp_filter[2], interp_filter[3],
647
  };
Geza Lore's avatar
Geza Lore committed
648
#else
James Zern's avatar
James Zern committed
649
  InterpFilter tmp_ipf = interp_filter;
Geza Lore's avatar
Geza Lore committed
650
#endif  // CONFIG_DUAL_FILTER
651
  ConvolveParams conv_params = get_conv_params(0, plane);
652

Yaowu Xu's avatar
Yaowu Xu committed
653
#if CONFIG_AOM_HIGHBITDEPTH
654
  DECLARE_ALIGNED(16, uint8_t, tmp_dst_[2 * MAX_SB_SQUARE]);
clang-format's avatar
clang-format committed
655
656
657
  uint8_t *tmp_dst = (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
                         ? CONVERT_TO_BYTEPTR(tmp_dst_)
                         : tmp_dst_;
Yaowu Xu's avatar
Yaowu Xu committed
658
  av1_make_inter_predictor(pre, pre_stride, tmp_dst, MAX_SB_SIZE, subpel_x,
659
660
661
662
663
                           subpel_y, sf, w, h, &conv_params, tmp_ipf,
#if CONFIG_GLOBAL_MOTION
                           is_global, p_col, p_row, plane, ref,
#endif  // CONFIG_GLOBAL_MOTION
                           xs, ys, xd);
664
665
666
667
668
669
670
671
672
673
674
675
#if CONFIG_COMPOUND_SEGMENT
  if (!plane && comp_data->type == COMPOUND_SEG) {
    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
      build_compound_seg_mask_highbd(comp_data->seg_mask, comp_data->mask_type,
                                     dst, dst_stride, tmp_dst, MAX_SB_SIZE,
                                     mi->mbmi.sb_type, h, w, xd->bd);
    else
      build_compound_seg_mask(comp_data->seg_mask, comp_data->mask_type, dst,
                              dst_stride, tmp_dst, MAX_SB_SIZE,
                              mi->mbmi.sb_type, h, w);
  }
#endif  // CONFIG_COMPOUND_SEGMENT
676
677
678
679
680
681
682
683
684
685
686

#if CONFIG_SUPERTX
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
    build_masked_compound_wedge_extend_highbd(
        dst, dst_stride, dst, dst_stride, tmp_dst, MAX_SB_SIZE, comp_data,
        mi->mbmi.sb_type, wedge_offset_x, wedge_offset_y, h, w, xd->bd);
  else
    build_masked_compound_wedge_extend(
        dst, dst_stride, dst, dst_stride, tmp_dst, MAX_SB_SIZE, comp_data,
        mi->mbmi.sb_type, wedge_offset_x, wedge_offset_y, h, w);
#else
687
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
688
689
690
    build_masked_compound_highbd(dst, dst_stride, dst, dst_stride, tmp_dst,
                                 MAX_SB_SIZE, comp_data, mi->mbmi.sb_type, h, w,
                                 xd->bd);
691
  else
692
693
    build_masked_compound(dst, dst_stride, dst, dst_stride, tmp_dst,
                          MAX_SB_SIZE, comp_data, mi->mbmi.sb_type, h, w);
694
#endif  // CONFIG_SUPERTX
695
696

#else  // CONFIG_AOM_HIGHBITDEPTH
697
  DECLARE_ALIGNED(16, uint8_t, tmp_dst[MAX_SB_SQUARE]);
Yaowu Xu's avatar
Yaowu Xu committed
698
  av1_make_inter_predictor(pre, pre_stride, tmp_dst, MAX_SB_SIZE, subpel_x,
699
700
701
702
703
                           subpel_y, sf, w, h, &conv_params, tmp_ipf,
#if CONFIG_GLOBAL_MOTION
                           is_global, p_col, p_row, plane, ref,
#endif  // CONFIG_GLOBAL_MOTION
                           xs, ys, xd);
704
705
#if CONFIG_COMPOUND_SEGMENT
  if (!plane && comp_data->type == COMPOUND_SEG)
706
707
708
    build_compound_seg_mask(comp_data->seg_mask, comp_data->mask_type, dst,
                            dst_stride, tmp_dst, MAX_SB_SIZE, mi->mbmi.sb_type,
                            h, w);
709
#endif  // CONFIG_COMPOUND_SEGMENT
710
711
712
713
714
#if CONFIG_SUPERTX
  build_masked_compound_wedge_extend(dst, dst_stride, dst, dst_stride, tmp_dst,
                                     MAX_SB_SIZE, comp_data, mi->mbmi.sb_type,
                                     wedge_offset_x, wedge_offset_y, h, w);
#else
715
716
  build_masked_compound(dst, dst_stride, dst, dst_stride, tmp_dst, MAX_SB_SIZE,
                        comp_data, mi->mbmi.sb_type, h, w);
717
#endif  // CONFIG_SUPERTX
Yaowu Xu's avatar
Yaowu Xu committed
718
#endif  // CONFIG_AOM_HIGHBITDEPTH
719
720
721
#if CONFIG_COMPOUND_SEGMENT
  (void)plane;
#endif  // CONFIG_COMPOUND_SEGMENT
722
723
}
#endif  // CONFIG_EXT_INTER
724

Yaowu Xu's avatar
Yaowu Xu committed
725
726
#if CONFIG_AOM_HIGHBITDEPTH
void av1_highbd_build_inter_predictor(
clang-format's avatar
clang-format committed
727
728
    const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride,
    const MV *src_mv, const struct scale_factors *sf, int w, int h, int ref,
729
#if CONFIG_DUAL_FILTER
James Zern's avatar
James Zern committed
730
    const InterpFilter *interp_filter,
731
#else
James Zern's avatar
James Zern committed
732
    const InterpFilter interp_filter,
733
#endif
clang-format's avatar
clang-format committed
734
    enum mv_precision precision, int x, int y, int bd) {
Jingning Han's avatar
Jingning Han committed
735
736
737
  const int is_q4 = precision == MV_PRECISION_Q4;
  const MV mv_q4 = { is_q4 ? src_mv->row : src_mv->row * 2,
                     is_q4 ? src_mv->col : src_mv->col * 2 };
Yaowu Xu's avatar
Yaowu Xu committed
738
  MV32 mv = av1_scale_mv(&mv_q4, x, y, sf);
Jingning Han's avatar
Jingning Han committed
739
740
741
742
743
  const int subpel_x = mv.col & SUBPEL_MASK;
  const int subpel_y = mv.row & SUBPEL_MASK;

  src += (mv.row >> SUBPEL_BITS) * src_stride + (mv.col >> SUBPEL_BITS);

744
  highbd_inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y,
745
746
                         sf, w, h, ref, interp_filter, sf->x_step_q4,
                         sf->y_step_q4, bd);
Jingning Han's avatar
Jingning Han committed
747
}
Yaowu Xu's avatar
Yaowu Xu committed
748
#endif  // CONFIG_AOM_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
749

Yaowu Xu's avatar
Yaowu Xu committed
750
751
752
void av1_build_inter_predictor(const uint8_t *src, int src_stride, uint8_t *dst,
                               int dst_stride, const MV *src_mv,
                               const struct scale_factors *sf, int w, int h,
753
                               ConvolveParams *conv_params,
754
#if CONFIG_DUAL_FILTER
James Zern's avatar
James Zern committed
755
                               const InterpFilter *interp_filter,
756
#else
James Zern's avatar
James Zern committed
757
                               const InterpFilter interp_filter,
758
#endif
Yaowu Xu's avatar
Yaowu Xu committed
759
                               enum mv_precision precision, int x, int y) {
Jingning Han's avatar
Jingning Han committed
760
761
762
  const int is_q4 = precision == MV_PRECISION_Q4;
  const MV mv_q4 = { is_q4 ? src_mv->row : src_mv->row * 2,
                     is_q4 ? src_mv->col : src_mv->col * 2 };
Yaowu Xu's avatar
Yaowu Xu committed
763
  MV32 mv = av1_scale_mv(&mv_q4, x, y, sf);
Jingning Han's avatar
Jingning Han committed
764
765
766
767
768
  const int subpel_x = mv.col & SUBPEL_MASK;
  const int subpel_y = mv.row & SUBPEL_MASK;

  src += (mv.row >> SUBPEL_BITS) * src_stride + (mv.col >> SUBPEL_BITS);

clang-format's avatar
clang-format committed
769
  inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y, sf, w,
770
                  h, conv_params, interp_filter, sf->x_step_q4, sf->y_step_q4);
Jingning Han's avatar
Jingning Han committed
771
772
}

Angie Chiang's avatar
Angie Chiang committed
773
774
775
776
777
778
779
typedef struct SubpelParams {
  int xs;
  int ys;
  int subpel_x;
  int subpel_y;
} SubpelParams;

780
void build_inter_predictors(MACROBLOCKD *xd, int plane,
Yue Chen's avatar
Yue Chen committed
781
#if CONFIG_MOTION_VAR
782
                            int mi_col_offset, int mi_row_offset,
Yue Chen's avatar
Yue Chen committed
783
#endif  // CONFIG_MOTION_VAR
clang-format's avatar
clang-format committed
784
785
                            int block, int bw, int bh, int x, int y, int w,
                            int h,
786
787
788
#if CONFIG_SUPERTX && CONFIG_EXT_INTER
                            int wedge_offset_x, int wedge_offset_y,
#endif  // CONFIG_SUPERTX && CONFIG_EXT_INTER
789
                            int mi_x, int mi_y) {
Jingning Han's avatar
Jingning Han committed
790
  struct macroblockd_plane *const pd = &xd->plane[plane];
Yue Chen's avatar
Yue Chen committed
791
#if CONFIG_MOTION_VAR
792
  const MODE_INFO *mi = xd->mi[mi_col_offset + xd->mi_stride * mi_row_offset];
Yue Chen's avatar
Yue Chen committed
793
#if !CONFIG_CB4X4
794
  const int build_for_obmc = !(mi_col_offset == 0 && mi_row_offset == 0);
Yue Chen's avatar
Yue Chen committed
795
#endif
796
#else
Jingning Han's avatar
Jingning Han committed
797
  const MODE_INFO *mi = xd->mi[0];
Yue Chen's avatar
Yue Chen committed
798
#endif  // CONFIG_MOTION_VAR
Jingning Han's avatar
Jingning Han committed
799
800
  const int is_compound = has_second_ref(&mi->mbmi);
  int ref;
801
802
803
#if CONFIG_GLOBAL_MOTION
  int is_global[2];
  for (ref = 0; ref < 1 + is_compound; ++ref) {
804
    WarpedMotionParams *const wm = &xd->global_motion[mi->mbmi.ref_frame[ref]];
clang-format's avatar
clang-format committed
805
    is_global[ref] =
806
        (get_y_mode(mi, block) == ZEROMV && wm->wmtype > TRANSLATION);
807
808
  }
#endif  // CONFIG_GLOBAL_MOTION
Jingning Han's avatar
Jingning Han committed
809

810
811
812
813
#if CONFIG_CB4X4
  (void)block;
#endif

814
#if CONFIG_SUB8X8_MC
815
816
817
#if CONFIG_MOTION_VAR
  if (mi->mbmi.sb_type < BLOCK_8X8 && plane > 0 && !build_for_obmc) {
#else
818
  if (mi->mbmi.sb_type < BLOCK_8X8 && plane > 0) {
819
#endif  // CONFIG_MOTION_VAR
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
    // block size in log2
    const int b4_wl = b_width_log2_lookup[mi->mbmi.sb_type];
    const int b4_hl = b_height_log2_lookup[mi->mbmi.sb_type];
    const int b8_sl = b_width_log2_lookup[BLOCK_8X8];

    // block size
    const int b4_w = 1 << b4_wl;
    const int b4_h = 1 << b4_hl;
    const int b8_s = 1 << b8_sl;
    int idx, idy;

    const int x_base = x;
    const int y_base = y;

    // processing unit size
    const int x_step = w >> (b8_sl - b4_wl);
    const int y_step = h >> (b8_sl - b4_hl);

    for (idy = 0; idy < b8_s; idy += b4_h) {
      for (idx = 0; idx < b8_s; idx += b4_w) {
        const int chr_idx = (idy * 2) + idx;
        for (ref = 0; ref < 1 + is_compound; ++ref) {
          const struct scale_factors *const sf = &xd->block_refs[ref]->sf;
          struct buf_2d *const pre_buf = &pd->pre[ref];
          struct buf_2d *const dst_buf = &pd->dst;
          uint8_t *dst = dst_buf->buf;
          const MV mv = mi->bmi[chr_idx].as_mv[ref].as_mv;
          const MV mv_q4 = clamp_mv_to_umv_border_sb(
              xd, &mv, bw, bh, pd->subsampling_x, pd->subsampling_y);
          uint8_t *pre;
          MV32 scaled_mv;
          int xs, ys, subpel_x, subpel_y;
Yaowu Xu's avatar
Yaowu Xu committed
852
          const int is_scaled = av1_is_scaled(sf);
853
          ConvolveParams conv_params = get_conv_params(ref, plane);
854
855
856
857
858
859
860
861
862

          x = x_base + idx * x_step;
          y = y_base + idy * y_step;

          dst += dst_buf->stride * y + x;

          if (is_scaled) {
            pre =
                pre_buf->buf + scaled_buffer_offset(x, y, pre_buf->stride, sf);
Yaowu Xu's avatar
Yaowu Xu committed
863
            scaled_mv = av1_scale_mv(&mv_q4, mi_x + x, mi_y + y, sf);
864
865
866
867
868
869
870
871
872
873
874
875
876
            xs = sf->x_step_q4;
            ys = sf->y_step_q4;
          } else {
            pre = pre_buf->buf + y * pre_buf->stride + x;
            scaled_mv.row = mv_q4.row;
            scaled_mv.col = mv_q4.col;
            xs = ys = 16;
          }

          subpel_x = scaled_mv.col & SUBPEL_MASK;
          subpel_y = scaled_mv.row & SUBPEL_MASK;
          pre += (scaled_mv.row >> SUBPEL_BITS) * pre_buf->stride +
                 (scaled_mv.col >> SUBPEL_BITS);
877

878
#if CONFIG_EXT_INTER
879
880
          if (ref &&
              is_masked_compound_type(mi->mbmi.interinter_compound_data.type))
Yaowu Xu's avatar
Yaowu Xu committed
881
            av1_make_masked_inter_predictor(
882
883
884
885
886
                pre, pre_buf->stride, dst, dst_buf->stride, subpel_x, subpel_y,
                sf, w, h, mi->mbmi.interp_filter, xs, ys,
#if CONFIG_SUPERTX
                wedge_offset_x, wedge_offset_y,
#endif  // CONFIG_SUPERTX
887
                plane,
888
889
890
891
892
#if CONFIG_GLOBAL_MOTION
                is_global[ref], (mi_x >> pd->subsampling_x) + x,
                (mi_y >> pd->subsampling_y) + y, ref
#endif  // CONFIG_GLOBAL_MOTION
                                                     xd);
893
894
          else
#endif  // CONFIG_EXT_INTER
895
896
897
898
899
900
901
902
            av1_make_inter_predictor(
                pre, pre_buf->stride, dst, dst_buf->stride, subpel_x, subpel_y,
                sf, x_step, y_step, &conv_params, mi->mbmi.interp_filter,
#if CONFIG_GLOBAL_MOTION
                is_global[ref], (mi_x >> pd->subsampling_x) + x,
                (mi_y >> pd->subsampling_y) + y, plane, ref,
#endif  // CONFIG_GLOBAL_MOTION
                xs, ys, xd);
903
        }
904
905
906
907
908
909
      }
    }
    return;
  }
#endif

Angie Chiang's avatar
Angie Chiang committed
910
  {
Jingning Han's avatar
Jingning Han committed
911
912
    struct buf_2d *const dst_buf = &pd->dst;
    uint8_t *const dst = dst_buf->buf + dst_buf->stride * y + x;
Angie Chiang's avatar
Angie Chiang committed
913
914
915
    uint8_t *pre[2];
    MV32 scaled_mv[2];
    SubpelParams subpel_params[2];
Angie Chiang's avatar
Angie Chiang committed
916
#if CONFIG_CONVOLVE_ROUND
Angie Chiang's avatar
Angie Chiang committed
917
918
    DECLARE_ALIGNED(16, int32_t, tmp_dst[MAX_SB_SIZE * MAX_SB_SIZE]);
    av1_zero(tmp_dst);
Angie Chiang's avatar
Angie Chiang committed
919
#endif  // CONFIG_CONVOLVE_ROUND
Angie Chiang's avatar
Angie Chiang committed
920
921
922
923

    for (ref = 0; ref < 1 + is_compound; ++ref) {
      const struct scale_factors *const sf = &xd->block_refs[ref]->sf;
      struct buf_2d *const pre_buf = &pd->pre[ref];
924
#if CONFIG_CB4X4
Angie Chiang's avatar
Angie Chiang committed
925
      const MV mv = mi->mbmi.mv[ref].as_mv;
926
#else
Angie Chiang's avatar
Angie Chiang committed
927
928
      const MV mv =
          mi->mbmi.sb_type < BLOCK_8X8
929
#if CONFIG_MOTION_VAR
Angie Chiang's avatar
Angie Chiang committed
930
931
              ? (build_for_obmc ? mi->bmi[block].as_mv[ref].as_mv
                                : average_split_mvs(pd, mi, ref, block))
932
#else
Angie Chiang's avatar
Angie Chiang committed
933
              ? average_split_mvs(pd, mi, ref, block)
934
#endif  // CONFIG_MOTION_VAR
Angie Chiang's avatar
Angie Chiang committed
935
              : mi->mbmi.mv[ref].as_mv;
936
#endif
Jingning Han's avatar
Jingning Han committed
937

Angie Chiang's avatar
Angie Chiang committed
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
      // TODO(jkoleszar): This clamping is done in the incorrect place for the
      // scaling case. It needs to be done on the scaled MV, not the pre-scaling
      // MV. Note however that it performs the subsampling aware scaling so
      // that the result is always q4.
      // mv_precision precision is MV_PRECISION_Q4.
      const MV mv_q4 = clamp_mv_to_umv_border_sb(
          xd, &mv, bw, bh, pd->subsampling_x, pd->subsampling_y);

      const int is_scaled = av1_is_scaled(sf);

      if (is_scaled) {
        pre[ref] =
            pre_buf->buf + scaled_buffer_offset(x, y, pre_buf->stride, sf);
        scaled_mv[ref] = av1_scale_mv(&mv_q4, mi_x + x, mi_y + y, sf);
        subpel_params[ref].xs = sf->x_step_q4;
        subpel_params[ref].ys = sf->y_step_q4;
      } else {
        pre[ref] = pre_buf->buf + (y * pre_buf->stride + x);
        scaled_mv[ref].row = mv_q4.row;
        scaled_mv[ref].col = mv_q4.col;
        subpel_params[ref].xs = 16;
        subpel_params[ref].ys = 16;
      }
961

Angie Chiang's avatar
Angie Chiang committed
962
963
964
965
966
      subpel_params[ref].subpel_x = scaled_mv[ref].col & SUBPEL_MASK;
      subpel_params[ref].subpel_y = scaled_mv[ref].row & SUBPEL_MASK;
      pre[ref] += (scaled_mv[ref].row >> SUBPEL_BITS) * pre_buf->stride +
                  (scaled_mv[ref].col >> SUBPEL_BITS);
    }
Jingning Han's avatar
Jingning Han committed
967

Angie Chiang's avatar
Angie Chiang committed
968
#if CONFIG_CONVOLVE_ROUND
Angie Chiang's avatar
Angie Chiang committed
969
    ConvolveParams conv_params =
970
        get_conv_params_no_round(ref, plane, tmp_dst, MAX_SB_SIZE);
Angie Chiang's avatar
Angie Chiang committed
971
#else
972
    ConvolveParams conv_params = get_conv_params(ref, plane);
Angie Chiang's avatar
Angie Chiang committed
973
#endif  // CONFIG_CONVOLVE_ROUND
Angie Chiang's avatar
Angie Chiang committed
974
975
976
977
    for (ref = 0; ref < 1 + is_compound; ++ref) {
      const struct scale_factors *const sf = &xd->block_refs[ref]->sf;
      struct buf_2d *const pre_buf = &pd->pre[ref];
      conv_params.ref = ref;
978
#if CONFIG_EXT_INTER
Angie Chiang's avatar
Angie Chiang committed
979
980
981
982
983
984
985
      if (ref &&
          is_masked_compound_type(mi->mbmi.interinter_compound_data.type))
        av1_make_masked_inter_predictor(
            pre[ref], pre_buf->stride, dst, dst_buf->stride,
            subpel_params[ref].subpel_x, subpel_params[ref].subpel_y, sf, w, h,
            mi->mbmi.interp_filter, subpel_params[ref].xs,
            subpel_params[ref].ys,
986
#if CONFIG_SUPERTX
Angie Chiang's avatar
Angie Chiang committed
987
            wedge_offset_x, wedge_offset_y,
988
#endif  // CONFIG_SUPERTX
Angie Chiang's avatar
Angie Chiang committed
989
            plane,
Zoe Liu's avatar
Zoe Liu committed
990
#if CONFIG_GLOBAL_MOTION
991
992
993
            is_global[ref], (mi_x >> pd->subsampling_x) + x,
            (mi_y >> pd->subsampling_y) + y, ref,
#endif  // CONFIG_GLOBAL_MOTION
Angie Chiang's avatar
Angie Chiang committed
994
995
            xd);
      else
996
#endif  // CONFIG_EXT_INTER
Angie Chiang's avatar
Angie Chiang committed
997
998
999
        av1_make_inter_predictor(
            pre[ref], pre_buf->stride, dst, dst_buf->stride,
            subpel_params[ref].subpel_x, subpel_params[ref].subpel_y, sf, w, h,
1000
1001
1002
1003
1004
1005
            &conv_params, mi->mbmi.interp_filter,
#if CONFIG_GLOBAL_MOTION
            is_global[ref], (mi_x >> pd->subsampling_x) + x,
            (mi_y >> pd->subsampling_y) + y, plane, ref,
#endif  // CONFIG_GLOBAL_MOTION
            subpel_params[ref].xs, subpel_params[ref].ys, xd);
Angie Chiang's avatar
Angie Chiang committed
1006
    }
Angie Chiang's avatar
Angie Chiang committed
1007

Angie Chiang's avatar
Angie Chiang committed
1008
#if CONFIG_CONVOLVE_ROUND
1009
1010
1011
// TODO(angiebird): This part needs optimization
#if CONFIG_AOM_HIGHBITDEPTH
    if (!(xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH))
Zoe Liu's avatar
Zoe Liu committed
1012
#endif  // CONFIG_AOM_HIGHBITDEPTH
Angie Chiang's avatar
Angie Chiang committed
1013
1014
1015
      av1_convolve_rounding(tmp_dst, MAX_SB_SIZE, dst, dst_buf->stride, w, h,
                            FILTER_BITS * 2 + is_compound -
                                conv_params.round_0 - conv_params.round_1);
Angie Chiang's avatar
Angie Chiang committed
1016
#endif  // CONFIG_CONVOLVE_ROUND
Jingning Han's avatar
Jingning Han committed
1017
1018
1019
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
1020
1021
void av1_build_inter_predictor_sub8x8(MACROBLOCKD *xd, int plane, int i, int ir,
                                      int ic, int mi_row, int mi_col) {
1022
1023
1024
  struct macroblockd_plane *const pd = &xd->plane[plane];
  MODE_INFO *const mi = xd->mi[0];
  const BLOCK_SIZE plane_bsize = get_plane_block_size(mi->mbmi.sb_type, pd);
1025
1026
  const int width = block_size_wide[plane_bsize];
  const int height = block_size_high[plane_bsize];
1027
1028
1029
1030
1031
  uint8_t *const dst = &pd->dst.buf[(ir * pd->dst.stride + ic) << 2];
  int ref;
  const int is_compound = has_second_ref(&mi->mbmi);

  for (ref = 0; ref < 1 + is_compound; ++ref) {
1032
    ConvolveParams conv_params = get_conv_params(ref, plane);
1033
1034
    const uint8_t *pre =
        &pd->pre[ref].buf[(ir * pd->pre[ref].stride + ic) << 2];
Yaowu Xu's avatar
Yaowu Xu committed
1035
#if CONFIG_AOM_HIGHBITDEPTH
Zoe Liu's avatar
Zoe Liu committed
1036
    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
Yaowu Xu's avatar
Yaowu Xu committed
1037
      av1_highbd_build_inter_predictor(
clang-format's avatar
clang-format committed