reconinter.c 118 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
Yue Chen's avatar
Yue Chen committed
27
#if CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
28
#include "av1/common/warped_motion.h"
Yue Chen's avatar
Yue Chen committed
29
#endif  // CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
Jingning Han's avatar
Jingning Han committed
30

31
#if CONFIG_EXT_INTER
32

clang-format's avatar
clang-format committed
33
#define NSMOOTHERS 1
34
static int get_masked_weight(int m, int smoothness) {
clang-format's avatar
clang-format committed
35
36
#define SMOOTHER_LEN 32
  static const uint8_t smoothfn[NSMOOTHERS][2 * SMOOTHER_LEN + 1] = { {
clang-format's avatar
clang-format committed
37
38
39
40
      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
41
  } };
42
43
44
45
46
  if (m < -SMOOTHER_LEN)
    return 0;
  else if (m > SMOOTHER_LEN)
    return (1 << WEDGE_WEIGHT_BITS);
  else
47
    return smoothfn[smoothness][m + SMOOTHER_LEN];
48
49
}

50
// [smoother][negative][direction]
clang-format's avatar
clang-format committed
51
52
53
DECLARE_ALIGNED(16, static uint8_t,
                wedge_mask_obl[NSMOOTHERS][2][WEDGE_DIRECTIONS]
                              [MASK_MASTER_SIZE * MASK_MASTER_SIZE]);
54

clang-format's avatar
clang-format committed
55
56
DECLARE_ALIGNED(16, static uint8_t,
                wedge_signflip_lookup[BLOCK_SIZES][MAX_WEDGE_TYPES]);
57

58
59
// 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
60
61
DECLARE_ALIGNED(16, static uint8_t,
                wedge_mask_buf[2 * MAX_WEDGE_TYPES * 3 * MAX_WEDGE_SQUARE]);
62
63
64

static wedge_masks_type wedge_masks[BLOCK_SIZES][2];

65
66
67
// 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
68
69
70
71
  { 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 },
72
73
};

74
static wedge_code_type wedge_codebook_8_hltw[8] = {
clang-format's avatar
clang-format committed
75
76
77
78
  { 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 },
79
80
};

81
static wedge_code_type wedge_codebook_8_heqw[8] = {
clang-format's avatar
clang-format committed
82
83
84
85
  { 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 },
86
87
};

88
89
#if !USE_LARGE_WEDGE_CODEBOOK
static const wedge_code_type wedge_codebook_16_hgtw[16] = {
clang-format's avatar
clang-format committed
90
91
92
93
94
95
96
97
  { 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
98
};
99

100
static const wedge_code_type wedge_codebook_16_hltw[16] = {
clang-format's avatar
clang-format committed
101
102
103
104
105
106
107
108
  { 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 },
109
110
};

111
static const wedge_code_type wedge_codebook_16_heqw[16] = {
clang-format's avatar
clang-format committed
112
113
114
115
116
117
118
119
  { 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 },
120
121
};

122
const wedge_params_type wedge_params_lookup[BLOCK_SIZES] = {
clang-format's avatar
clang-format committed
123
124
125
126
127
128
129
130
131
132
133
134
135
  { 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] },
136
#if CONFIG_EXT_PARTITION
clang-format's avatar
clang-format committed
137
138
139
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
140
#endif  // CONFIG_EXT_PARTITION
141
142
};

143
144
145
#else

static const wedge_code_type wedge_codebook_32_hgtw[32] = {
clang-format's avatar
clang-format committed
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
  { 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 },
162
163
};

164
static const wedge_code_type wedge_codebook_32_hltw[32] = {
clang-format's avatar
clang-format committed
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
  { 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 },
181
182
};

183
static const wedge_code_type wedge_codebook_32_heqw[32] = {
clang-format's avatar
clang-format committed
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
  { 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 },
200
201
};

202
const wedge_params_type wedge_params_lookup[BLOCK_SIZES] = {
clang-format's avatar
clang-format committed
203
204
205
206
207
208
209
210
211
212
213
214
215
  { 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
216
#if CONFIG_EXT_PARTITION
clang-format's avatar
clang-format committed
217
218
219
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
  { 0, NULL, NULL, 0, NULL },
Debargha Mukherjee's avatar
Debargha Mukherjee committed
220
#endif  // CONFIG_EXT_PARTITION
221
};
222
#endif  // USE_LARGE_WEDGE_CODEBOOK
223

clang-format's avatar
clang-format committed
224
static const uint8_t *get_wedge_mask_inplace(int wedge_index, int neg,
225
                                             BLOCK_SIZE sb_type) {
Debargha Mukherjee's avatar
Debargha Mukherjee committed
226
  const uint8_t *master;
227
228
  const int bh = 4 << b_height_log2_lookup[sb_type];
  const int bw = 4 << b_width_log2_lookup[sb_type];
229
230
231
  const wedge_code_type *a =
      wedge_params_lookup[sb_type].codebook + wedge_index;
  const int smoother = wedge_params_lookup[sb_type].smoother;
232
  int woff, hoff;
233
234
235
236
237
238
239
  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
240
241
           MASK_MASTER_STRIDE * (MASK_MASTER_SIZE / 2 - hoff) +
           MASK_MASTER_SIZE / 2 - woff;
Debargha Mukherjee's avatar
Debargha Mukherjee committed
242
243
244
  return master;
}

Yaowu Xu's avatar
Yaowu Xu committed
245
246
247
const uint8_t *av1_get_soft_mask(int wedge_index, int wedge_sign,
                                 BLOCK_SIZE sb_type, int offset_x,
                                 int offset_y) {
248
  const uint8_t *mask =
249
      get_wedge_mask_inplace(wedge_index, wedge_sign, sb_type);
clang-format's avatar
clang-format committed
250
  if (mask) mask -= (offset_x + offset_y * MASK_MASTER_STRIDE);
251
  return mask;
252
253
}

254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
#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) {
274
275
276
  assert(is_masked_compound_type(comp_data->type));
  switch (comp_data->type) {
    case COMPOUND_WEDGE:
277
278
      return av1_get_contiguous_soft_mask(comp_data->wedge_index,
                                          !comp_data->wedge_sign, sb_type);
279
280
#if CONFIG_COMPOUND_SEGMENT
    case COMPOUND_SEG:
281
      return invert_mask(mask_buffer, comp_data->seg_mask, h, w, stride);
282
#endif  // CONFIG_COMPOUND_SEGMENT
283
284
285
    default: assert(0); return NULL;
  }
}
286

287
288
289
290
291
292
293
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);
294
#if CONFIG_COMPOUND_SEGMENT
295
296
297
298
299
300
301
    case COMPOUND_SEG: return comp_data->seg_mask;
#endif  // CONFIG_COMPOUND_SEGMENT
    default: assert(0); return NULL;
  }
}

#if CONFIG_COMPOUND_SEGMENT
302
303
304
#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) {
305
306
307
308
309
310
311
312
313
314
  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,
315
316
317
318
319
320
321
                             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;
322
323
324
325
326
  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);
  }
327
}
328
329
330
331
332
333
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
378
379
380
381
382
383
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

#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) {
    case DIFFWTD_45:
      diffwtd_mask(mask, 0, 47, src0, src0_stride, src1, src1_stride, sb_type,
                   h, w);
      break;
    case DIFFWTD_45_INV:
      diffwtd_mask(mask, 1, 47, src0, src0_stride, src1, src1_stride, sb_type,
                   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
422
#endif  // CONFIG_COMPOUND_SEGMENT
423

424
static void init_wedge_master_masks() {
425
426
427
428
  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
429
  const int a[2] = { 2, 1 };
430
431
432
433
434
435
436
437
  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
438
439
            wedge_mask_obl[s][1][WEDGE_OBLIQUE27][j * stride + i] =
                get_masked_weight(m, s);
440
        wedge_mask_obl[s][1][WEDGE_OBLIQUE117][i * stride + w - 1 - j] =
clang-format's avatar
clang-format committed
441
442
            wedge_mask_obl[s][1][WEDGE_OBLIQUE153][(w - 1 - j) * stride + i] =
                (1 << WEDGE_WEIGHT_BITS) - get_masked_weight(m, s);
443
        wedge_mask_obl[s][0][WEDGE_OBLIQUE63][i * stride + j] =
clang-format's avatar
clang-format committed
444
445
            wedge_mask_obl[s][0][WEDGE_OBLIQUE27][j * stride + i] =
                (1 << WEDGE_WEIGHT_BITS) - get_masked_weight(m, s);
446
        wedge_mask_obl[s][0][WEDGE_OBLIQUE117][i * stride + w - 1 - j] =
clang-format's avatar
clang-format committed
447
448
            wedge_mask_obl[s][0][WEDGE_OBLIQUE153][(w - 1 - j) * stride + i] =
                get_masked_weight(m, s);
449
        wedge_mask_obl[s][1][WEDGE_VERTICAL][i * stride + j] =
clang-format's avatar
clang-format committed
450
451
            wedge_mask_obl[s][1][WEDGE_HORIZONTAL][j * stride + i] =
                get_masked_weight(x, s);
452
        wedge_mask_obl[s][0][WEDGE_VERTICAL][i * stride + j] =
clang-format's avatar
clang-format committed
453
454
            wedge_mask_obl[s][0][WEDGE_HORIZONTAL][j * stride + i] =
                (1 << WEDGE_WEIGHT_BITS) - get_masked_weight(x, s);
455
456
      }
  }
457
458
459
460
461
462
463
464
465
}

// 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) {
466
467
    const int bw = block_size_wide[sb_type];
    const int bh = block_size_high[sb_type];
468
469
470
471
472
473
474
475
    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
476
477
      for (i = 0; i < bw; ++i) sum += mask[i];
      for (i = 0; i < bh; ++i) sum += mask[i * MASK_MASTER_STRIDE];
478
479
480
481
482
483
484
485
486
487
488
489
      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;
490
491
    const int bw = block_size_wide[bsize];
    const int bh = block_size_high[bsize];
492
493
494
495
496
497
498
    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
499
      aom_convolve_copy(mask, MASK_MASTER_STRIDE, dst, bw, NULL, 0, NULL, 0, bw,
clang-format's avatar
clang-format committed
500
                        bh);
501
502
503
504
      wedge_params->masks[0][w] = dst;
      dst += bw * bh;

      mask = get_wedge_mask_inplace(w, 1, bsize);
Yaowu Xu's avatar
Yaowu Xu committed
505
      aom_convolve_copy(mask, MASK_MASTER_STRIDE, dst, bw, NULL, 0, NULL, 0, bw,
clang-format's avatar
clang-format committed
506
                        bh);
507
508
509
510
511
512
513
514
      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
515
void av1_init_wedge_masks() {
516
  init_wedge_master_masks();
517
  init_wedge_signs();
518
  init_wedge_masks();
519
520
}

521
522
#if CONFIG_SUPERTX
static void build_masked_compound_wedge_extend(
clang-format's avatar
clang-format committed
523
524
525
    uint8_t *dst, int dst_stride, const uint8_t *src0, int src0_stride,
    const uint8_t *src1, int src1_stride, int wedge_index, int wedge_sign,
    BLOCK_SIZE sb_type, int wedge_offset_x, int wedge_offset_y, int h, int w) {
526
527
  const int subh = (2 << b_height_log2_lookup[sb_type]) == h;
  const int subw = (2 << b_width_log2_lookup[sb_type]) == w;
Yaowu Xu's avatar
Yaowu Xu committed
528
529
530
  const uint8_t *mask = av1_get_soft_mask(wedge_index, wedge_sign, sb_type,
                                          wedge_offset_x, wedge_offset_y);
  aom_blend_a64_mask(dst, dst_stride, src0, src0_stride, src1, src1_stride,
clang-format's avatar
clang-format committed
531
                     mask, MASK_MASTER_STRIDE, h, w, subh, subw);
532
533
}

Yaowu Xu's avatar
Yaowu Xu committed
534
#if CONFIG_AOM_HIGHBITDEPTH
535
static void build_masked_compound_wedge_extend_highbd(
clang-format's avatar
clang-format committed
536
537
538
539
    uint8_t *dst_8, int dst_stride, const uint8_t *src0_8, int src0_stride,
    const uint8_t *src1_8, int src1_stride, int wedge_index, int wedge_sign,
    BLOCK_SIZE sb_type, int wedge_offset_x, int wedge_offset_y, int h, int w,
    int bd) {
540
541
  const int subh = (2 << b_height_log2_lookup[sb_type]) == h;
  const int subw = (2 << b_width_log2_lookup[sb_type]) == w;
Yaowu Xu's avatar
Yaowu Xu committed
542
543
544
  const uint8_t *mask = av1_get_soft_mask(wedge_index, wedge_sign, sb_type,
                                          wedge_offset_x, wedge_offset_y);
  aom_highbd_blend_a64_mask(dst_8, dst_stride, src0_8, src0_stride, src1_8,
clang-format's avatar
clang-format committed
545
546
                            src1_stride, mask, MASK_MASTER_STRIDE, h, w, subh,
                            subw, bd);
547
}
Yaowu Xu's avatar
Yaowu Xu committed
548
#endif  // CONFIG_AOM_HIGHBITDEPTH
549
#endif  // CONFIG_SUPERTX
550

551
552
553
554
555
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) {
556
557
558
559
  // 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;
560
  const uint8_t *mask = av1_get_compound_type_mask(comp_data, sb_type);
Yaowu Xu's avatar
Yaowu Xu committed
561
  aom_blend_a64_mask(dst, dst_stride, src0, src0_stride, src1, src1_stride,
562
                     mask, block_size_wide[sb_type], h, w, subh, subw);
563
564
}

Yaowu Xu's avatar
Yaowu Xu committed
565
#if CONFIG_AOM_HIGHBITDEPTH
566
static void build_masked_compound_highbd(
clang-format's avatar
clang-format committed
567
    uint8_t *dst_8, int dst_stride, const uint8_t *src0_8, int src0_stride,
568
569
570
    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) {
571
572
573
574
  // 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;
575
576
577
  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);
578
579
580
  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);
581
}
Yaowu Xu's avatar
Yaowu Xu committed
582
#endif  // CONFIG_AOM_HIGHBITDEPTH
583

Yaowu Xu's avatar
Yaowu Xu committed
584
585
586
587
588
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,
589
#if CONFIG_DUAL_FILTER
James Zern's avatar
James Zern committed
590
                                     const InterpFilter *interp_filter,
591
#else
James Zern's avatar
James Zern committed
592
                                     const InterpFilter interp_filter,
593
#endif
Yaowu Xu's avatar
Yaowu Xu committed
594
                                     int xs, int ys,
595
#if CONFIG_SUPERTX
Yaowu Xu's avatar
Yaowu Xu committed
596
                                     int wedge_offset_x, int wedge_offset_y,
597
#endif  // CONFIG_SUPERTX
598
599
600
601
602
603
#if CONFIG_COMPOUND_SEGMENT
                                     int plane,
#endif  // CONFIG_COMPOUND_SEGMENT
                                     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
604
605
// The prediction filter types used here should be those for
// the second reference block.
Geza Lore's avatar
Geza Lore committed
606
#if CONFIG_DUAL_FILTER
James Zern's avatar
James Zern committed
607
  InterpFilter tmp_ipf[4] = {
Geza Lore's avatar
Geza Lore committed
608
    interp_filter[2], interp_filter[3], interp_filter[2], interp_filter[3],
609
  };
Geza Lore's avatar
Geza Lore committed
610
#else
James Zern's avatar
James Zern committed
611
  InterpFilter tmp_ipf = interp_filter;
Geza Lore's avatar
Geza Lore committed
612
#endif  // CONFIG_DUAL_FILTER
613
  ConvolveParams conv_params = get_conv_params(0);
Yaowu Xu's avatar
Yaowu Xu committed
614
#if CONFIG_AOM_HIGHBITDEPTH
615
  DECLARE_ALIGNED(16, uint8_t, tmp_dst_[2 * MAX_SB_SQUARE]);
clang-format's avatar
clang-format committed
616
617
618
  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
619
  av1_make_inter_predictor(pre, pre_stride, tmp_dst, MAX_SB_SIZE, subpel_x,
620
621
                           subpel_y, sf, w, h, &conv_params, tmp_ipf, xs, ys,
                           xd);
622
623
#if CONFIG_SUPERTX
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
624
    build_masked_compound_wedge_extend_highbd(
clang-format's avatar
clang-format committed
625
        dst, dst_stride, dst, dst_stride, tmp_dst, MAX_SB_SIZE,
626
627
        comp_data->wedge_index, comp_data->wedge_sign, mi->mbmi.sb_type,
        wedge_offset_x, wedge_offset_y, h, w, xd->bd);
628
  else
629
    build_masked_compound_wedge_extend(
clang-format's avatar
clang-format committed
630
        dst, dst_stride, dst, dst_stride, tmp_dst, MAX_SB_SIZE,
631
632
        comp_data->wedge_index, comp_data->wedge_sign, mi->mbmi.sb_type,
        wedge_offset_x, wedge_offset_y, h, w);
633
#else
634
635
636
637
638
639
640
641
642
643
644
645
#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
646
  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
647
648
649
    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);
650
  else
651
652
    build_masked_compound(dst, dst_stride, dst, dst_stride, tmp_dst,
                          MAX_SB_SIZE, comp_data, mi->mbmi.sb_type, h, w);
653
#endif  // CONFIG_SUPERTX
Yaowu Xu's avatar
Yaowu Xu committed
654
#else   // CONFIG_AOM_HIGHBITDEPTH
655
  DECLARE_ALIGNED(16, uint8_t, tmp_dst[MAX_SB_SQUARE]);
Yaowu Xu's avatar
Yaowu Xu committed
656
  av1_make_inter_predictor(pre, pre_stride, tmp_dst, MAX_SB_SIZE, subpel_x,
657
658
                           subpel_y, sf, w, h, &conv_params, tmp_ipf, xs, ys,
                           xd);
659
#if CONFIG_SUPERTX
660
661
662
663
  build_masked_compound_wedge_extend(dst, dst_stride, dst, dst_stride, tmp_dst,
                                     MAX_SB_SIZE, comp_data->wedge_index,
                                     comp_data->wedge_sign, mi->mbmi.sb_type,
                                     wedge_offset_x, wedge_offset_y, h, w);
664
#else
665
666
#if CONFIG_COMPOUND_SEGMENT
  if (!plane && comp_data->type == COMPOUND_SEG)
667
668
669
    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);
670
#endif  // CONFIG_COMPOUND_SEGMENT
671
672
  build_masked_compound(dst, dst_stride, dst, dst_stride, tmp_dst, MAX_SB_SIZE,
                        comp_data, mi->mbmi.sb_type, h, w);
673
#endif  // CONFIG_SUPERTX
Yaowu Xu's avatar
Yaowu Xu committed
674
#endif  // CONFIG_AOM_HIGHBITDEPTH
675
676
}
#endif  // CONFIG_EXT_INTER
677

Yaowu Xu's avatar
Yaowu Xu committed
678
679
#if CONFIG_AOM_HIGHBITDEPTH
void av1_highbd_build_inter_predictor(
clang-format's avatar
clang-format committed
680
681
    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,
682
#if CONFIG_DUAL_FILTER
James Zern's avatar
James Zern committed
683
    const InterpFilter *interp_filter,
684
#else
James Zern's avatar
James Zern committed
685
    const InterpFilter interp_filter,
686
#endif
clang-format's avatar
clang-format committed
687
    enum mv_precision precision, int x, int y, int bd) {
Jingning Han's avatar
Jingning Han committed
688
689
690
  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
691
  MV32 mv = av1_scale_mv(&mv_q4, x, y, sf);
Jingning Han's avatar
Jingning Han committed
692
693
694
695
696
  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);

697
  highbd_inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y,
698
699
                         sf, w, h, ref, interp_filter, sf->x_step_q4,
                         sf->y_step_q4, bd);
Jingning Han's avatar
Jingning Han committed
700
}
Yaowu Xu's avatar
Yaowu Xu committed
701
#endif  // CONFIG_AOM_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
702

Yaowu Xu's avatar
Yaowu Xu committed
703
704
705
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,
706
                               ConvolveParams *conv_params,
707
#if CONFIG_DUAL_FILTER
James Zern's avatar
James Zern committed
708
                               const InterpFilter *interp_filter,
709
#else
James Zern's avatar
James Zern committed
710
                               const InterpFilter interp_filter,
711
#endif
Yaowu Xu's avatar
Yaowu Xu committed
712
                               enum mv_precision precision, int x, int y) {
Jingning Han's avatar
Jingning Han committed
713
714
715
  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
716
  MV32 mv = av1_scale_mv(&mv_q4, x, y, sf);
Jingning Han's avatar
Jingning Han committed
717
718
719
720
721
  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
722
  inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y, sf, w,
723
                  h, conv_params, interp_filter, sf->x_step_q4, sf->y_step_q4);
Jingning Han's avatar
Jingning Han committed
724
725
}

Angie Chiang's avatar
Angie Chiang committed
726
727
728
729
730
731
732
typedef struct SubpelParams {
  int xs;
  int ys;
  int subpel_x;
  int subpel_y;
} SubpelParams;

733
void build_inter_predictors(MACROBLOCKD *xd, int plane,
Yue Chen's avatar
Yue Chen committed
734
#if CONFIG_MOTION_VAR
735
                            int mi_col_offset, int mi_row_offset,
Yue Chen's avatar
Yue Chen committed
736
#endif  // CONFIG_MOTION_VAR
clang-format's avatar
clang-format committed
737
738
                            int block, int bw, int bh, int x, int y, int w,
                            int h,
739
740
741
#if CONFIG_SUPERTX && CONFIG_EXT_INTER
                            int wedge_offset_x, int wedge_offset_y,
#endif  // CONFIG_SUPERTX && CONFIG_EXT_INTER
742
                            int mi_x, int mi_y) {
Jingning Han's avatar
Jingning Han committed
743
  struct macroblockd_plane *const pd = &xd->plane[plane];
Yue Chen's avatar
Yue Chen committed
744
#if CONFIG_MOTION_VAR
745
  const MODE_INFO *mi = xd->mi[mi_col_offset + xd->mi_stride * mi_row_offset];
Yue Chen's avatar
Yue Chen committed
746
#if !CONFIG_CB4X4
747
  const int build_for_obmc = !(mi_col_offset == 0 && mi_row_offset == 0);
Yue Chen's avatar
Yue Chen committed
748
#endif
749
#else
Jingning Han's avatar
Jingning Han committed
750
  const MODE_INFO *mi = xd->mi[0];
Yue Chen's avatar
Yue Chen committed
751
#endif  // CONFIG_MOTION_VAR
Jingning Han's avatar
Jingning Han committed
752
753
  const int is_compound = has_second_ref(&mi->mbmi);
  int ref;
754
#if CONFIG_GLOBAL_MOTION
755
  WarpedMotionParams *gm[2];
756
757
758
  int is_global[2];
  for (ref = 0; ref < 1 + is_compound; ++ref) {
    gm[ref] = &xd->global_motion[mi->mbmi.ref_frame[ref]];
clang-format's avatar
clang-format committed
759
    is_global[ref] =
760
        (get_y_mode(mi, block) == ZEROMV && gm[ref]->wmtype > TRANSLATION);
761
762
  }
  // TODO(sarahparker) remove these once gm works with all experiments
clang-format's avatar
clang-format committed
763
764
  (void)gm;
  (void)is_global;
765
#endif  // CONFIG_GLOBAL_MOTION
Jingning Han's avatar
Jingning Han committed
766

767
768
769
770
#if CONFIG_CB4X4
  (void)block;
#endif

771
772
// TODO(sarahparker) enable the use of DUAL_FILTER in warped motion functions
// in order to allow GLOBAL_MOTION and DUAL_FILTER to work together
773
#if CONFIG_SUB8X8_MC
774
775
776
#if CONFIG_MOTION_VAR
  if (mi->mbmi.sb_type < BLOCK_8X8 && plane > 0 && !build_for_obmc) {
#else
777
  if (mi->mbmi.sb_type < BLOCK_8X8 && plane > 0) {
778
#endif  // CONFIG_MOTION_VAR
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
    // 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
811
          const int is_scaled = av1_is_scaled(sf);
812
          ConvolveParams conv_params = get_conv_params(ref);
813
814
815
816
817
818
819
820
821

          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
822
            scaled_mv = av1_scale_mv(&mv_q4, mi_x + x, mi_y + y, sf);
823
824
825
826
827
828
829
830
831
832
833
834
835
            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);
836

837
#if CONFIG_EXT_INTER
838
839
          if (ref &&
              is_masked_compound_type(mi->mbmi.interinter_compound_data.type))
Yaowu Xu's avatar
Yaowu Xu committed
840
            av1_make_masked_inter_predictor(
841
842
843
844
845
                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
846
847
848
#if CONFIG_COMPOUND_SEGMENT
                plane,
#endif  // CONFIG_COMPOUND_SEGMENT
849
850
851
                xd);
          else
#endif  // CONFIG_EXT_INTER
Yaowu Xu's avatar
Yaowu Xu committed
852
853
            av1_make_inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
                                     subpel_x, subpel_y, sf, x_step, y_step,
854
855
                                     &conv_params, mi->mbmi.interp_filter, xs,
                                     ys, xd);
856
        }
857
858
859
860
861
862
      }
    }
    return;
  }
#endif

Angie Chiang's avatar
Angie Chiang committed
863
  {
Jingning Han's avatar
Jingning Han committed
864
865
    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
866
867
868
    uint8_t *pre[2];
    MV32 scaled_mv[2];
    SubpelParams subpel_params[2];
Angie Chiang's avatar
Angie Chiang committed
869
870
871
#if CONVOLVE_POST_ROUNDING
    int32_t tmp_dst[MAX_SB_SIZE * MAX_SB_SIZE];
#endif  // CONVOLVE_POST_ROUNDING
Angie Chiang's avatar
Angie Chiang committed
872
873
874
875

    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];
876
#if CONFIG_CB4X4
Angie Chiang's avatar
Angie Chiang committed
877
      const MV mv = mi->mbmi.mv[ref].as_mv;
878
#else
Angie Chiang's avatar
Angie Chiang committed
879
880
      const MV mv =
          mi->mbmi.sb_type < BLOCK_8X8
881
#if CONFIG_MOTION_VAR
Angie Chiang's avatar
Angie Chiang committed
882
883
              ? (build_for_obmc ? mi->bmi[block].as_mv[ref].as_mv
                                : average_split_mvs(pd, mi, ref, block))
884
#else
Angie Chiang's avatar
Angie Chiang committed
885
              ? average_split_mvs(pd, mi, ref, block)
886
#endif  // CONFIG_MOTION_VAR
Angie Chiang's avatar
Angie Chiang committed
887
              : mi->mbmi.mv[ref].as_mv;
888
#endif
Jingning Han's avatar
Jingning Han committed
889

Angie Chiang's avatar
Angie Chiang committed
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
      // 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;
      }
913

Angie Chiang's avatar
Angie Chiang committed
914
915
916
917
918
      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
919

Angie Chiang's avatar
Angie Chiang committed
920
921
922
    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];
Angie Chiang's avatar
Angie Chiang committed
923
924
925
926
927
928
#if CONVOLVE_POST_ROUNDING
      ConvolveParams conv_params =
          get_conv_params_no_round(ref, tmp_dst, MAX_SB_SIZE);
#else
      ConvolveParams conv_params = get_conv_params(ref);
#endif  // CONVOLVE_POST_ROUNDING
929
#if CONFIG_EXT_INTER
Angie Chiang's avatar
Angie Chiang committed
930
931
932
933
934
935
936
      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,
937
#if CONFIG_SUPERTX
Angie Chiang's avatar
Angie Chiang committed
938
            wedge_offset_x, wedge_offset_y,
939
#endif  // CONFIG_SUPERTX
940
#if CONFIG_COMPOUND_SEGMENT
Angie Chiang's avatar
Angie Chiang committed
941
            plane,
942
#endif  // CONFIG_COMPOUND_SEGMENT
Angie Chiang's avatar
Angie Chiang committed
943
944
            xd);
      else
945
946
#else  // CONFIG_EXT_INTER
#if CONFIG_GLOBAL_MOTION
Angie Chiang's avatar
Angie Chiang committed
947
948
      if (is_global[ref])
        av1_warp_plane(gm[ref],
Yaowu Xu's avatar
Yaowu Xu committed
949
#if CONFIG_AOM_HIGHBITDEPTH
Angie Chiang's avatar
Angie Chiang committed
950
                       xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH, xd->bd,
Yaowu Xu's avatar
Yaowu Xu committed
951
#endif  // CONFIG_AOM_HIGHBITDEPTH
Angie Chiang's avatar
Angie Chiang committed
952
953
954
955
956
957
                       pre_buf->buf0, pre_buf->width, pre_buf->height,
                       pre_buf->stride, dst, (mi_x >> pd->subsampling_x) + x,
                       (mi_y >> pd->subsampling_y) + y, w, h, dst_buf->stride,
                       pd->subsampling_x, pd->subsampling_y,
                       subpel_params[ref].xs, subpel_params[ref].ys, ref);
      else
958
#endif  // CONFIG_GLOBAL_MOTION
959
#endif  // CONFIG_EXT_INTER
Angie Chiang's avatar
Angie Chiang committed
960
961
962
963
964
965
        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,
            &conv_params, mi->mbmi.interp_filter, subpel_params[ref].xs,
            subpel_params[ref].ys, xd);
    }
Angie Chiang's avatar
Angie Chiang committed
966
967

#if CONVOLVE_POST_ROUNDING
968
969
970
971
972
// TODO(angiebird): This part needs optimization
#if CONFIG_AOM_HIGHBITDEPTH
    if (!(xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH))
      av1_convolve_rounding(tmp_dst, MAX_SB_SIZE, dst, dst_buf->stride, w, h);
#else
Angie Chiang's avatar
Angie Chiang committed
973
    av1_convolve_rounding(tmp_dst, MAX_SB_SIZE, dst, dst_buf->stride, w, h);
974
#endif
Angie Chiang's avatar
Angie Chiang committed
975
#endif  // CONVOLVE_POST_ROUNDING
Jingning Han's avatar
Jingning Han committed
976
977
978
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
979
980
void av1_build_inter_predictor_sub8x8(MACROBLOCKD *xd, int plane, int i, int ir,
                                      int ic, int mi_row, int mi_col) {
981
982
983
  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);
984
985
  const int width = block_size_wide[plane_bsize];
  const int height = block_size_high[plane_bsize];
986
987
988
989
990
  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) {
991
    ConvolveParams conv_params = get_conv_params(ref);
992
993
    const uint8_t *pre =
        &pd->pre[ref].buf[(ir * pd->pre[ref].stride + ic) << 2];
Yaowu Xu's avatar
Yaowu Xu committed
994
#if CONFIG_AOM_HIGHBITDEPTH
clang-format's avatar
clang-format committed
995
    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
Yaowu Xu's avatar
Yaowu Xu committed
996
      av1_highbd_build_inter_predictor(
clang-format's avatar
clang-format committed
997
998
999
1000
1001
          pre, pd->pre[ref].stride, dst, pd->dst.stride,
          &mi->bmi[i].as_mv[ref].as_mv, &xd->block_refs[ref]->sf, width, height,
          ref, mi->mbmi.interp_filter, MV_PRECISION_Q3,
          mi_col * MI_SIZE + 4 * ic, mi_row * MI_SIZE + 4 * ir, xd->bd);
    } else {
Yaowu Xu's avatar
Yaowu Xu committed
1002
      av1_build_inter_predictor(
clang-format's avatar
clang-format committed
1003
1004
          pre, pd->pre[ref].stride, dst, pd->dst.stride,
          &mi->bmi[i].as_mv[ref].as_mv, &xd->block_refs[ref]->sf, width, height,
1005
          &conv_params, mi->mbmi.interp_filter, MV_PRECISION_Q3,
clang-format's avatar
clang-format committed
1006
1007
          mi_col * MI_SIZE + 4 * ic, mi_row * MI_SIZE + 4 * ir);
    }
1008
#else
Yaowu Xu's avatar
Yaowu Xu committed
1009
    av1_build_inter_predictor(
clang-format's avatar
clang-format committed
1010
1011
        pre, pd->pre[ref].stride, dst, pd->dst.stride,