vp9_reconinter.c 67.4 KB
Newer Older
John Koleszar's avatar
John Koleszar committed
1
/*
2
 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
John Koleszar's avatar
John Koleszar committed
3
 *
4
 *  Use of this source code is governed by a BSD-style license
5
6
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
7
 *  in the file PATENTS.  All contributing project authors may
8
 *  be found in the AUTHORS file in the root of the source tree.
John Koleszar's avatar
John Koleszar committed
9
10
 */

11
#include <assert.h>
John Koleszar's avatar
John Koleszar committed
12

13
#include "./vpx_config.h"
14
#include "vpx/vpx_integer.h"
15
#include "vp9/common/vp9_blockd.h"
16
#include "vp9/common/vp9_filter.h"
17
#include "vp9/common/vp9_reconinter.h"
18
#include "vp9/common/vp9_reconintra.h"
John Koleszar's avatar
John Koleszar committed
19

20
21
22
void vp9_setup_scale_factors_for_frame(struct scale_factors *scale,
                                       YV12_BUFFER_CONFIG *other,
                                       int this_w, int this_h) {
John Koleszar's avatar
John Koleszar committed
23
24
  int other_h = other->y_crop_height;
  int other_w = other->y_crop_width;
25
26
27
28
29

  scale->x_num = other_w;
  scale->x_den = this_w;
  scale->x_offset_q4 = 0;  // calculated per-mb
  scale->x_step_q4 = 16 * other_w / this_w;
30

31
32
33
34
35
  scale->y_num = other_h;
  scale->y_den = this_h;
  scale->y_offset_q4 = 0;  // calculated per-mb
  scale->y_step_q4 = 16 * other_h / this_h;

36
37
38
39
40
41
  // TODO(agrange): Investigate the best choice of functions to use here
  // for EIGHTTAP_SMOOTH. Since it is not interpolating, need to choose what
  // to do at full-pel offsets. The current selection, where the filter is
  // applied in one direction only, and not at all for 0,0, seems to give the
  // best quality, but it may be worth trying an additional mode that does
  // do the filtering on full-pel.
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
#if CONFIG_IMPLICIT_COMPOUNDINTER_WEIGHT
  if (scale->x_step_q4 == 16) {
    if (scale->y_step_q4 == 16) {
      // No scaling in either direction.
      scale->predict[0][0][0] = vp9_convolve_copy;
      scale->predict[0][0][1] = vp9_convolve_1by8;
      scale->predict[0][0][2] = vp9_convolve_qtr;
      scale->predict[0][0][3] = vp9_convolve_3by8;
      scale->predict[0][0][4] = vp9_convolve_avg;
      scale->predict[0][0][5] = vp9_convolve_5by8;
      scale->predict[0][0][6] = vp9_convolve_3qtr;
      scale->predict[0][0][7] = vp9_convolve_7by8;
      scale->predict[0][1][0] = vp9_convolve8_vert;
      scale->predict[0][1][1] = vp9_convolve8_1by8_vert;
      scale->predict[0][1][2] = vp9_convolve8_qtr_vert;
      scale->predict[0][1][3] = vp9_convolve8_3by8_vert;
      scale->predict[0][1][4] = vp9_convolve8_avg_vert;
      scale->predict[0][1][5] = vp9_convolve8_5by8_vert;
      scale->predict[0][1][6] = vp9_convolve8_3qtr_vert;
      scale->predict[0][1][7] = vp9_convolve8_7by8_vert;
      scale->predict[1][0][0] = vp9_convolve8_horiz;
      scale->predict[1][0][1] = vp9_convolve8_1by8_horiz;
      scale->predict[1][0][2] = vp9_convolve8_qtr_horiz;
      scale->predict[1][0][3] = vp9_convolve8_3by8_horiz;
      scale->predict[1][0][4] = vp9_convolve8_avg_horiz;
      scale->predict[1][0][5] = vp9_convolve8_5by8_horiz;
      scale->predict[1][0][6] = vp9_convolve8_3qtr_horiz;
      scale->predict[1][0][7] = vp9_convolve8_7by8_horiz;
    } else {
      // No scaling in x direction. Must always scale in the y direction.
      scale->predict[0][0][0] = vp9_convolve8_vert;
      scale->predict[0][0][1] = vp9_convolve8_1by8_vert;
      scale->predict[0][0][2] = vp9_convolve8_qtr_vert;
      scale->predict[0][0][3] = vp9_convolve8_3by8_vert;
      scale->predict[0][0][4] = vp9_convolve8_avg_vert;
      scale->predict[0][0][5] = vp9_convolve8_5by8_vert;
      scale->predict[0][0][6] = vp9_convolve8_3qtr_vert;
      scale->predict[0][0][7] = vp9_convolve8_7by8_vert;
      scale->predict[0][1][0] = vp9_convolve8_vert;
      scale->predict[0][1][1] = vp9_convolve8_1by8_vert;
      scale->predict[0][1][2] = vp9_convolve8_qtr_vert;
      scale->predict[0][1][3] = vp9_convolve8_3by8_vert;
      scale->predict[0][1][4] = vp9_convolve8_avg_vert;
      scale->predict[0][1][5] = vp9_convolve8_5by8_vert;
      scale->predict[0][1][6] = vp9_convolve8_3qtr_vert;
      scale->predict[0][1][7] = vp9_convolve8_7by8_vert;
      scale->predict[1][0][0] = vp9_convolve8;
      scale->predict[1][0][1] = vp9_convolve8_1by8;
      scale->predict[1][0][2] = vp9_convolve8_qtr;
      scale->predict[1][0][3] = vp9_convolve8_3by8;
      scale->predict[1][0][4] = vp9_convolve8_avg;
      scale->predict[1][0][5] = vp9_convolve8_5by8;
      scale->predict[1][0][6] = vp9_convolve8_3qtr;
      scale->predict[1][0][7] = vp9_convolve8_7by8;
    }
  } else {
    if (scale->y_step_q4 == 16) {
      // No scaling in the y direction. Must always scale in the x direction.
      scale->predict[0][0][0] = vp9_convolve8_horiz;
      scale->predict[0][0][1] = vp9_convolve8_1by8_horiz;
      scale->predict[0][0][2] = vp9_convolve8_qtr_horiz;
      scale->predict[0][0][3] = vp9_convolve8_3by8_horiz;
      scale->predict[0][0][4] = vp9_convolve8_avg_horiz;
      scale->predict[0][0][5] = vp9_convolve8_5by8_horiz;
      scale->predict[0][0][6] = vp9_convolve8_3qtr_horiz;
      scale->predict[0][0][7] = vp9_convolve8_7by8_horiz;
      scale->predict[0][1][0] = vp9_convolve8;
      scale->predict[0][1][1] = vp9_convolve8_1by8;
      scale->predict[0][1][2] = vp9_convolve8_qtr;
      scale->predict[0][1][3] = vp9_convolve8_3by8;
      scale->predict[0][1][4] = vp9_convolve8_avg;
      scale->predict[0][1][5] = vp9_convolve8_5by8;
      scale->predict[0][1][6] = vp9_convolve8_3qtr;
      scale->predict[0][1][7] = vp9_convolve8_7by8;
      scale->predict[1][0][0] = vp9_convolve8_horiz;
      scale->predict[1][0][1] = vp9_convolve8_1by8_horiz;
      scale->predict[1][0][2] = vp9_convolve8_qtr_horiz;
      scale->predict[1][0][3] = vp9_convolve8_3by8_horiz;
      scale->predict[1][0][4] = vp9_convolve8_avg_horiz;
      scale->predict[1][0][5] = vp9_convolve8_5by8_horiz;
      scale->predict[1][0][6] = vp9_convolve8_3qtr_horiz;
      scale->predict[1][0][7] = vp9_convolve8_7by8_horiz;
    } else {
      // Must always scale in both directions.
      scale->predict[0][0][0] = vp9_convolve8;
      scale->predict[0][0][1] = vp9_convolve8_1by8;
      scale->predict[0][0][2] = vp9_convolve8_qtr;
      scale->predict[0][0][3] = vp9_convolve8_3by8;
      scale->predict[0][0][4] = vp9_convolve8_avg;
      scale->predict[0][0][5] = vp9_convolve8_5by8;
      scale->predict[0][0][6] = vp9_convolve8_3qtr;
      scale->predict[0][0][7] = vp9_convolve8_7by8;
      scale->predict[0][1][0] = vp9_convolve8;
      scale->predict[0][1][1] = vp9_convolve8_1by8;
      scale->predict[0][1][2] = vp9_convolve8_qtr;
      scale->predict[0][1][3] = vp9_convolve8_3by8;
      scale->predict[0][1][4] = vp9_convolve8_avg;
      scale->predict[0][1][5] = vp9_convolve8_5by8;
      scale->predict[0][1][6] = vp9_convolve8_3qtr;
      scale->predict[0][1][7] = vp9_convolve8_7by8;
      scale->predict[1][0][0] = vp9_convolve8;
      scale->predict[1][0][1] = vp9_convolve8_1by8;
      scale->predict[1][0][2] = vp9_convolve8_qtr;
      scale->predict[1][0][3] = vp9_convolve8_3by8;
      scale->predict[1][0][4] = vp9_convolve8_avg;
      scale->predict[1][0][5] = vp9_convolve8_5by8;
      scale->predict[1][0][6] = vp9_convolve8_3qtr;
      scale->predict[1][0][7] = vp9_convolve8_7by8;
    }
  }
  // 2D subpel motion always gets filtered in both directions
  scale->predict[1][1][0] = vp9_convolve8;
  scale->predict[1][1][1] = vp9_convolve8_1by8;
  scale->predict[1][1][2] = vp9_convolve8_qtr;
  scale->predict[1][1][3] = vp9_convolve8_3by8;
  scale->predict[1][1][4] = vp9_convolve8_avg;
  scale->predict[1][1][5] = vp9_convolve8_5by8;
  scale->predict[1][1][6] = vp9_convolve8_3qtr;
  scale->predict[1][1][7] = vp9_convolve8_7by8;
}
#else
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
  if (scale->x_step_q4 == 16) {
    if (scale->y_step_q4 == 16) {
      // No scaling in either direction.
      scale->predict[0][0][0] = vp9_convolve_copy;
      scale->predict[0][0][1] = vp9_convolve_avg;
      scale->predict[0][1][0] = vp9_convolve8_vert;
      scale->predict[0][1][1] = vp9_convolve8_avg_vert;
      scale->predict[1][0][0] = vp9_convolve8_horiz;
      scale->predict[1][0][1] = vp9_convolve8_avg_horiz;
    } else {
      // No scaling in x direction. Must always scale in the y direction.
      scale->predict[0][0][0] = vp9_convolve8_vert;
      scale->predict[0][0][1] = vp9_convolve8_avg_vert;
      scale->predict[0][1][0] = vp9_convolve8_vert;
      scale->predict[0][1][1] = vp9_convolve8_avg_vert;
      scale->predict[1][0][0] = vp9_convolve8;
      scale->predict[1][0][1] = vp9_convolve8_avg;
    }
  } else {
    if (scale->y_step_q4 == 16) {
      // No scaling in the y direction. Must always scale in the x direction.
      scale->predict[0][0][0] = vp9_convolve8_horiz;
      scale->predict[0][0][1] = vp9_convolve8_avg_horiz;
      scale->predict[0][1][0] = vp9_convolve8;
      scale->predict[0][1][1] = vp9_convolve8_avg;
      scale->predict[1][0][0] = vp9_convolve8_horiz;
      scale->predict[1][0][1] = vp9_convolve8_avg_horiz;
    } else {
      // Must always scale in both directions.
      scale->predict[0][0][0] = vp9_convolve8;
      scale->predict[0][0][1] = vp9_convolve8_avg;
      scale->predict[0][1][0] = vp9_convolve8;
      scale->predict[0][1][1] = vp9_convolve8_avg;
      scale->predict[1][0][0] = vp9_convolve8;
      scale->predict[1][0][1] = vp9_convolve8_avg;
    }
  }
  // 2D subpel motion always gets filtered in both directions
  scale->predict[1][1][0] = vp9_convolve8;
  scale->predict[1][1][1] = vp9_convolve8_avg;
}
204
#endif
205
206
207
208
209
210
211
212
213
214
215
216
217

void vp9_setup_interp_filters(MACROBLOCKD *xd,
                              INTERPOLATIONFILTERTYPE mcomp_filter_type,
                              VP9_COMMON *cm) {
  if (xd->mode_info_context) {
    MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;

    set_scale_factors(xd,
                      mbmi->ref_frame - 1,
                      mbmi->second_ref_frame - 1,
                      cm->active_ref_scale);
  }

218
219
220
221
222
223
224
225
226
227
228
229
230
231
  switch (mcomp_filter_type) {
    case EIGHTTAP:
    case SWITCHABLE:
      xd->subpix.filter_x = xd->subpix.filter_y = vp9_sub_pel_filters_8;
      break;
    case EIGHTTAP_SMOOTH:
      xd->subpix.filter_x = xd->subpix.filter_y = vp9_sub_pel_filters_8lp;
      break;
    case EIGHTTAP_SHARP:
      xd->subpix.filter_x = xd->subpix.filter_y = vp9_sub_pel_filters_8s;
      break;
    case BILINEAR:
      xd->subpix.filter_x = xd->subpix.filter_y = vp9_bilinear_filters;
      break;
Adrian Grange's avatar
Adrian Grange committed
232
#if CONFIG_ENABLE_6TAP
233
234
235
    case SIXTAP:
      xd->subpix.filter_x = xd->subpix.filter_y = vp9_sub_pel_filters_6;
      break;
Adrian Grange's avatar
Adrian Grange committed
236
237
#endif
  }
238
  assert(((intptr_t)xd->subpix.filter_x & 0xff) == 0);
239
240
}

241
void vp9_copy_mem16x16_c(const uint8_t *src,
242
                         int src_stride,
243
                         uint8_t *dst,
244
                         int dst_stride) {
John Koleszar's avatar
John Koleszar committed
245
  int r;
John Koleszar's avatar
John Koleszar committed
246

John Koleszar's avatar
John Koleszar committed
247
  for (r = 0; r < 16; r++) {
248
#if !(CONFIG_FAST_UNALIGNED)
John Koleszar's avatar
John Koleszar committed
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
    dst[0] = src[0];
    dst[1] = src[1];
    dst[2] = src[2];
    dst[3] = src[3];
    dst[4] = src[4];
    dst[5] = src[5];
    dst[6] = src[6];
    dst[7] = src[7];
    dst[8] = src[8];
    dst[9] = src[9];
    dst[10] = src[10];
    dst[11] = src[11];
    dst[12] = src[12];
    dst[13] = src[13];
    dst[14] = src[14];
    dst[15] = src[15];
John Koleszar's avatar
John Koleszar committed
265
266

#else
267
268
269
270
    ((uint32_t *)dst)[0] = ((const uint32_t *)src)[0];
    ((uint32_t *)dst)[1] = ((const uint32_t *)src)[1];
    ((uint32_t *)dst)[2] = ((const uint32_t *)src)[2];
    ((uint32_t *)dst)[3] = ((const uint32_t *)src)[3];
John Koleszar's avatar
John Koleszar committed
271
272

#endif
John Koleszar's avatar
John Koleszar committed
273
274
275
    src += src_stride;
    dst += dst_stride;
  }
John Koleszar's avatar
John Koleszar committed
276
277
}

278
void vp9_copy_mem8x8_c(const uint8_t *src,
279
                       int src_stride,
280
                       uint8_t *dst,
281
                       int dst_stride) {
John Koleszar's avatar
John Koleszar committed
282
283
284
  int r;

  for (r = 0; r < 8; r++) {
285
#if !(CONFIG_FAST_UNALIGNED)
John Koleszar's avatar
John Koleszar committed
286
287
288
289
290
291
292
293
    dst[0] = src[0];
    dst[1] = src[1];
    dst[2] = src[2];
    dst[3] = src[3];
    dst[4] = src[4];
    dst[5] = src[5];
    dst[6] = src[6];
    dst[7] = src[7];
John Koleszar's avatar
John Koleszar committed
294
#else
295
296
    ((uint32_t *)dst)[0] = ((const uint32_t *)src)[0];
    ((uint32_t *)dst)[1] = ((const uint32_t *)src)[1];
John Koleszar's avatar
John Koleszar committed
297
#endif
John Koleszar's avatar
John Koleszar committed
298
299
300
    src += src_stride;
    dst += dst_stride;
  }
John Koleszar's avatar
John Koleszar committed
301
302
}

303
void vp9_copy_mem8x4_c(const uint8_t *src,
304
                       int src_stride,
305
                       uint8_t *dst,
306
                       int dst_stride) {
John Koleszar's avatar
John Koleszar committed
307
308
309
  int r;

  for (r = 0; r < 4; r++) {
310
#if !(CONFIG_FAST_UNALIGNED)
John Koleszar's avatar
John Koleszar committed
311
312
313
314
315
316
317
318
    dst[0] = src[0];
    dst[1] = src[1];
    dst[2] = src[2];
    dst[3] = src[3];
    dst[4] = src[4];
    dst[5] = src[5];
    dst[6] = src[6];
    dst[7] = src[7];
John Koleszar's avatar
John Koleszar committed
319
#else
320
321
    ((uint32_t *)dst)[0] = ((const uint32_t *)src)[0];
    ((uint32_t *)dst)[1] = ((const uint32_t *)src)[1];
John Koleszar's avatar
John Koleszar committed
322
#endif
John Koleszar's avatar
John Koleszar committed
323
324
325
    src += src_stride;
    dst += dst_stride;
  }
John Koleszar's avatar
John Koleszar committed
326
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
static void set_scaled_offsets(struct scale_factors *scale,
                               int row, int col) {
  const int x_q4 = 16 * col;
  const int y_q4 = 16 * row;

  scale->x_offset_q4 = (x_q4 * scale->x_num / scale->x_den) & 0xf;
  scale->y_offset_q4 = (y_q4 * scale->y_num / scale->y_den) & 0xf;
}

static int32_t scale_motion_vector_component_q3(int mv_q3,
                                                int num,
                                                int den,
                                                int offset_q4) {
  // returns the scaled and offset value of the mv component.
  const int32_t mv_q4 = mv_q3 << 1;

  /* TODO(jkoleszar): make fixed point, or as a second multiply? */
  return mv_q4 * num / den + offset_q4;
}

static int32_t scale_motion_vector_component_q4(int mv_q4,
                                                int num,
                                                int den,
                                                int offset_q4) {
352
353
354
  // returns the scaled and offset value of the mv component.

  /* TODO(jkoleszar): make fixed point, or as a second multiply? */
355
  return mv_q4 * num / den + offset_q4;
John Koleszar's avatar
John Koleszar committed
356
357
}

358
359
360
static int_mv32 scale_motion_vector_q3_to_q4(
    const int_mv *src_mv,
    const struct scale_factors *scale) {
361
362
363
  // returns mv * scale + offset
  int_mv32 result;

364
365
366
367
368
369
370
371
  result.as_mv.row = scale_motion_vector_component_q3(src_mv->as_mv.row,
                                                      scale->y_num,
                                                      scale->y_den,
                                                      scale->y_offset_q4);
  result.as_mv.col = scale_motion_vector_component_q3(src_mv->as_mv.col,
                                                      scale->x_num,
                                                      scale->x_den,
                                                      scale->x_offset_q4);
372
  return result;
373
374
}

375
376
377
378
void vp9_build_inter_predictor(const uint8_t *src, int src_stride,
                               uint8_t *dst, int dst_stride,
                               const int_mv *mv_q3,
                               const struct scale_factors *scale,
379
                               int w, int h, int weight,
380
                               const struct subpix_fn_table *subpix) {
381
382
  int_mv32 mv = scale_motion_vector_q3_to_q4(mv_q3, scale);
  src += (mv.as_mv.row >> 4) * src_stride + (mv.as_mv.col >> 4);
383
  scale->predict[!!(mv.as_mv.col & 15)][!!(mv.as_mv.row & 15)][weight](
384
      src, src_stride, dst, dst_stride,
385
386
      subpix->filter_x[mv.as_mv.col & 15], scale->x_step_q4,
      subpix->filter_y[mv.as_mv.row & 15], scale->y_step_q4,
387
      w, h);
John Koleszar's avatar
John Koleszar committed
388
389
}

390
391
/* Like vp9_build_inter_predictor, but takes the full-pel part of the
 * mv separately, and the fractional part as a q4.
392
 */
393
394
395
396
397
void vp9_build_inter_predictor_q4(const uint8_t *src, int src_stride,
                                  uint8_t *dst, int dst_stride,
                                  const int_mv *fullpel_mv_q3,
                                  const int_mv *frac_mv_q4,
                                  const struct scale_factors *scale,
398
                                  int w, int h, int weight,
399
400
401
402
403
404
                                  const struct subpix_fn_table *subpix) {
  const int mv_row_q4 = ((fullpel_mv_q3->as_mv.row >> 3) << 4)
                        + (frac_mv_q4->as_mv.row & 0xf);
  const int mv_col_q4 = ((fullpel_mv_q3->as_mv.col >> 3) << 4)
                        + (frac_mv_q4->as_mv.col & 0xf);
  const int scaled_mv_row_q4 =
405
406
      scale_motion_vector_component_q4(mv_row_q4, scale->y_num, scale->y_den,
                                       scale->y_offset_q4);
407
  const int scaled_mv_col_q4 =
408
409
      scale_motion_vector_component_q4(mv_col_q4, scale->x_num, scale->x_den,
                                       scale->x_offset_q4);
410
411
412
  const int subpel_x = scaled_mv_col_q4 & 15;
  const int subpel_y = scaled_mv_row_q4 & 15;

413
  src += (scaled_mv_row_q4 >> 4) * src_stride + (scaled_mv_col_q4 >> 4);
414
  scale->predict[!!subpel_x][!!subpel_y][weight](
415
      src, src_stride, dst, dst_stride,
416
417
      subpix->filter_x[subpel_x], scale->x_step_q4,
      subpix->filter_y[subpel_y], scale->y_step_q4,
418
      w, h);
419
420
}

421
422
423
424
425
426
427
428
static void build_2x1_inter_predictor_wh(const BLOCKD *d0, const BLOCKD *d1,
                                         struct scale_factors *scale,
                                         uint8_t *predictor,
                                         int block_size, int stride,
                                         int which_mv, int weight,
                                         int width, int height,
                                         const struct subpix_fn_table *subpix,
                                         int row, int col) {
429
430
431
  assert(d1->predictor - d0->predictor == block_size);
  assert(d1->pre == d0->pre + block_size);

432
433
  set_scaled_offsets(&scale[which_mv], row, col);

434
435
436
437
438
  if (d0->bmi.as_mv[which_mv].as_int == d1->bmi.as_mv[which_mv].as_int) {
    uint8_t **base_pre = which_mv ? d0->base_second_pre : d0->base_pre;

    vp9_build_inter_predictor(*base_pre + d0->pre,
                              d0->pre_stride,
439
                              predictor, stride,
440
441
                              &d0->bmi.as_mv[which_mv],
                              &scale[which_mv],
442
443
                              width, height,
                              weight, subpix);
444
445
446
447
448
449
450

  } else {
    uint8_t **base_pre0 = which_mv ? d0->base_second_pre : d0->base_pre;
    uint8_t **base_pre1 = which_mv ? d1->base_second_pre : d1->base_pre;

    vp9_build_inter_predictor(*base_pre0 + d0->pre,
                              d0->pre_stride,
451
                              predictor, stride,
452
453
                              &d0->bmi.as_mv[which_mv],
                              &scale[which_mv],
454
455
456
457
                              width > block_size ? block_size : width, height,
                              weight, subpix);

    if (width <= block_size) return;
458
459
460

    set_scaled_offsets(&scale[which_mv], row, col + block_size);

461
462
    vp9_build_inter_predictor(*base_pre1 + d1->pre,
                              d1->pre_stride,
463
                              predictor + block_size, stride,
464
465
                              &d1->bmi.as_mv[which_mv],
                              &scale[which_mv],
466
467
                              width - block_size, height,
                              weight, subpix);
468
  }
John Koleszar's avatar
John Koleszar committed
469
470
}

471
472
473
474
475
476
477
478
static void build_2x1_inter_predictor(const BLOCKD *d0, const BLOCKD *d1,
                                      struct scale_factors *scale,
                                      int block_size, int stride,
                                      int which_mv, int weight,
                                      const struct subpix_fn_table *subpix,
                                      int row, int col) {
  assert(d1->predictor - d0->predictor == block_size);
  assert(d1->pre == d0->pre + block_size);
479

480
  set_scaled_offsets(&scale[which_mv], row, col);
John Koleszar's avatar
John Koleszar committed
481

482
483
  if (d0->bmi.as_mv[which_mv].as_int == d1->bmi.as_mv[which_mv].as_int) {
    uint8_t **base_pre = which_mv ? d0->base_second_pre : d0->base_pre;
John Koleszar's avatar
John Koleszar committed
484

485
486
487
488
489
490
491
    vp9_build_inter_predictor(*base_pre + d0->pre,
                              d0->pre_stride,
                              d0->predictor, stride,
                              &d0->bmi.as_mv[which_mv],
                              &scale[which_mv],
                              2 * block_size, block_size,
                              weight, subpix);
John Koleszar's avatar
John Koleszar committed
492

493
494
495
  } else {
    uint8_t **base_pre0 = which_mv ? d0->base_second_pre : d0->base_pre;
    uint8_t **base_pre1 = which_mv ? d1->base_second_pre : d1->base_pre;
John Koleszar's avatar
John Koleszar committed
496

497
498
499
500
501
502
503
    vp9_build_inter_predictor(*base_pre0 + d0->pre,
                              d0->pre_stride,
                              d0->predictor, stride,
                              &d0->bmi.as_mv[which_mv],
                              &scale[which_mv],
                              block_size, block_size,
                              weight, subpix);
504

505
    set_scaled_offsets(&scale[which_mv], row, col + block_size);
John Koleszar's avatar
John Koleszar committed
506

507
508
509
510
511
512
513
    vp9_build_inter_predictor(*base_pre1 + d1->pre,
                              d1->pre_stride,
                              d1->predictor, stride,
                              &d1->bmi.as_mv[which_mv],
                              &scale[which_mv],
                              block_size, block_size,
                              weight, subpix);
John Koleszar's avatar
John Koleszar committed
514
  }
John Koleszar's avatar
John Koleszar committed
515
516
}

John Koleszar's avatar
John Koleszar committed
517
518
519
520
521
522
523
524
525
526
static void clamp_mv_to_umv_border(MV *mv, const MACROBLOCKD *xd) {
  /* If the MV points so far into the UMV border that no visible pixels
   * are used for reconstruction, the subpel part of the MV can be
   * discarded and the MV limited to 16 pixels with equivalent results.
   *
   * This limit kicks in at 19 pixels for the top and left edges, for
   * the 16 pixels plus 3 taps right of the central pixel when subpel
   * filtering. The bottom and right edges use 16 pixels plus 2 pixels
   * left of the central pixel when filtering.
   */
John Koleszar's avatar
John Koleszar committed
527
  if (mv->col < (xd->mb_to_left_edge - ((16 + VP9_INTERP_EXTEND) << 3)))
John Koleszar's avatar
John Koleszar committed
528
    mv->col = xd->mb_to_left_edge - (16 << 3);
John Koleszar's avatar
John Koleszar committed
529
  else if (mv->col > xd->mb_to_right_edge + ((15 + VP9_INTERP_EXTEND) << 3))
John Koleszar's avatar
John Koleszar committed
530
531
    mv->col = xd->mb_to_right_edge + (16 << 3);

John Koleszar's avatar
John Koleszar committed
532
  if (mv->row < (xd->mb_to_top_edge - ((16 + VP9_INTERP_EXTEND) << 3)))
John Koleszar's avatar
John Koleszar committed
533
    mv->row = xd->mb_to_top_edge - (16 << 3);
John Koleszar's avatar
John Koleszar committed
534
  else if (mv->row > xd->mb_to_bottom_edge + ((15 + VP9_INTERP_EXTEND) << 3))
John Koleszar's avatar
John Koleszar committed
535
    mv->row = xd->mb_to_bottom_edge + (16 << 3);
536
537
538
}

/* A version of the above function for chroma block MVs.*/
John Koleszar's avatar
John Koleszar committed
539
static void clamp_uvmv_to_umv_border(MV *mv, const MACROBLOCKD *xd) {
John Koleszar's avatar
John Koleszar committed
540
541
542
  const int extend = VP9_INTERP_EXTEND;

  mv->col = (2 * mv->col < (xd->mb_to_left_edge - ((16 + extend) << 3))) ?
John Koleszar's avatar
John Koleszar committed
543
            (xd->mb_to_left_edge - (16 << 3)) >> 1 : mv->col;
John Koleszar's avatar
John Koleszar committed
544
  mv->col = (2 * mv->col > xd->mb_to_right_edge + ((15 + extend) << 3)) ?
John Koleszar's avatar
John Koleszar committed
545
546
            (xd->mb_to_right_edge + (16 << 3)) >> 1 : mv->col;

John Koleszar's avatar
John Koleszar committed
547
  mv->row = (2 * mv->row < (xd->mb_to_top_edge - ((16 + extend) << 3))) ?
John Koleszar's avatar
John Koleszar committed
548
            (xd->mb_to_top_edge - (16 << 3)) >> 1 : mv->row;
John Koleszar's avatar
John Koleszar committed
549
  mv->row = (2 * mv->row > xd->mb_to_bottom_edge + ((15 + extend) << 3)) ?
John Koleszar's avatar
John Koleszar committed
550
            (xd->mb_to_bottom_edge + (16 << 3)) >> 1 : mv->row;
551
552
}

553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
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
811
812
813
814
815
816
817
818
819
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
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
#define AVERAGE_WEIGHT  (1 << (2 * CONFIG_IMPLICIT_COMPOUNDINTER_WEIGHT))

#if CONFIG_IMPLICIT_COMPOUNDINTER_WEIGHT

// Whether to use implicit weighting for UV
#define USE_IMPLICIT_WEIGHT_UV

// Whether to use implicit weighting for SplitMV
// #define USE_IMPLICIT_WEIGHT_SPLITMV

// #define SEARCH_MIN3
static int64_t get_consistency_metric(MACROBLOCKD *xd,
                                      uint8_t *tmp_y, int tmp_ystride) {
  int block_size = 16 <<  xd->mode_info_context->mbmi.sb_type;
  uint8_t *rec_y = xd->dst.y_buffer;
  int rec_ystride = xd->dst.y_stride;
  int64_t metric = 0;
  int i;
  if (xd->up_available) {
    for (i = 0; i < block_size; ++i) {
      int diff = abs(*(rec_y - rec_ystride + i) -
                     *(tmp_y + i));
#ifdef SEARCH_MIN3
      // Searches for the min abs diff among 3 pixel neighbors in the border
      int diff1 = xd->left_available ?
          abs(*(rec_y - rec_ystride + i - 1) - *(tmp_y + i)) : diff;
      int diff2 = i < block_size - 1 ?
          abs(*(rec_y - rec_ystride + i + 1) - *(tmp_y + i)) : diff;
      diff = diff <= diff1 ? diff : diff1;
      diff = diff <= diff2 ? diff : diff2;
#endif
      metric += diff;
    }
  }
  if (xd->left_available) {
    for (i = 0; i < block_size; ++i) {
      int diff = abs(*(rec_y - 1 + i * rec_ystride) -
                     *(tmp_y + i * tmp_ystride));
#ifdef SEARCH_MIN3
      // Searches for the min abs diff among 3 pixel neighbors in the border
      int diff1 = xd->up_available ?
          abs(*(rec_y - 1 + (i - 1) * rec_ystride) -
                      *(tmp_y + i * tmp_ystride)) : diff;
      int diff2 = i < block_size - 1 ?
          abs(*(rec_y - 1 + (i + 1) * rec_ystride) -
              *(tmp_y + i * tmp_ystride)) : diff;
      diff = diff <= diff1 ? diff : diff1;
      diff = diff <= diff2 ? diff : diff2;
#endif
      metric += diff;
    }
  }
  return metric;
}

static int get_weight(MACROBLOCKD *xd, int64_t metric_1, int64_t metric_2) {
  int weight = AVERAGE_WEIGHT;
  if (2 * metric_1 < metric_2)
    weight = 6;
  else if (4 * metric_1 < 3 * metric_2)
    weight = 5;
  else if (2 * metric_2 < metric_1)
    weight = 2;
  else if (4 * metric_2 < 3 * metric_1)
    weight = 3;
  return weight;
}

#ifdef USE_IMPLICIT_WEIGHT_SPLITMV
static int get_implicit_compoundinter_weight_splitmv(
    MACROBLOCKD *xd, int mb_row, int mb_col) {
  MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
  BLOCKD *blockd = xd->block;
  const int use_second_ref = mbmi->second_ref_frame > 0;
  int64_t metric_2 = 0, metric_1 = 0;
  int i, which_mv, weight;
  uint8_t tmp_y[256];
  const int tmp_ystride = 16;

  if (!use_second_ref) return 0;
  if (!(xd->up_available || xd->left_available))
    return AVERAGE_WEIGHT;

  assert(xd->mode_info_context->mbmi.mode == SPLITMV);

  which_mv = 1;  // second predictor
  if (xd->mode_info_context->mbmi.partitioning != PARTITIONING_4X4) {
    for (i = 0; i < 16; i += 8) {
      BLOCKD *d0 = &blockd[i];
      BLOCKD *d1 = &blockd[i + 2];
      const int y = i & 8;

      blockd[i + 0].bmi = xd->mode_info_context->bmi[i + 0];
      blockd[i + 2].bmi = xd->mode_info_context->bmi[i + 2];

      if (mbmi->need_to_clamp_mvs) {
        clamp_mv_to_umv_border(&blockd[i + 0].bmi.as_mv[which_mv].as_mv, xd);
        clamp_mv_to_umv_border(&blockd[i + 2].bmi.as_mv[which_mv].as_mv, xd);
      }
      if (i == 0) {
        build_2x1_inter_predictor_wh(d0, d1, xd->scale_factor, tmp_y, 8, 16,
                                     which_mv, 0, 16, 1,
                                     &xd->subpix, mb_row * 16 + y, mb_col * 16);
        build_2x1_inter_predictor_wh(d0, d1, xd->scale_factor, tmp_y, 8, 16,
                                     which_mv, 0, 1, 8,
                                     &xd->subpix, mb_row * 16 + y, mb_col * 16);
      } else {
        build_2x1_inter_predictor_wh(d0, d1, xd->scale_factor, tmp_y + 8 * 16,
                                     8, 16, which_mv, 0, 1, 8,
                                     &xd->subpix, mb_row * 16 + y, mb_col * 16);
      }
    }
  } else {
    for (i = 0; i < 16; i += 2) {
      BLOCKD *d0 = &blockd[i];
      BLOCKD *d1 = &blockd[i + 1];
      const int x = (i & 3) * 4;
      const int y = (i >> 2) * 4;

      blockd[i + 0].bmi = xd->mode_info_context->bmi[i + 0];
      blockd[i + 1].bmi = xd->mode_info_context->bmi[i + 1];

      if (i >= 4 && (i & 3) != 0) continue;

      if (i == 0) {
        build_2x1_inter_predictor_wh(d0, d1, xd->scale_factor, tmp_y, 4, 16,
                                     which_mv, 0, 8, 1, &xd->subpix,
                                     mb_row * 16 + y, mb_col * 16 + x);
        build_2x1_inter_predictor_wh(d0, d1, xd->scale_factor, tmp_y, 4, 16,
                                     which_mv, 0, 1, 4, &xd->subpix,
                                     mb_row * 16 + y, mb_col * 16 + x);
      } else if (i < 4) {
        build_2x1_inter_predictor_wh(d0, d1, xd->scale_factor, tmp_y + x, 4, 16,
                                     which_mv, 0, 8, 1, &xd->subpix,
                                     mb_row * 16 + y, mb_col * 16 + x);
      } else {
        build_2x1_inter_predictor_wh(d0, d1, xd->scale_factor, tmp_y + y * 16,
                                     4, 16, which_mv, 0, 1, 4, &xd->subpix,
                                     mb_row * 16 + y, mb_col * 16 + x);
      }
    }
  }
  metric_2 = get_consistency_metric(xd, tmp_y, tmp_ystride);

  which_mv = 0;  // first predictor
  if (xd->mode_info_context->mbmi.partitioning != PARTITIONING_4X4) {
    for (i = 0; i < 16; i += 8) {
      BLOCKD *d0 = &blockd[i];
      BLOCKD *d1 = &blockd[i + 2];
      const int y = i & 8;

      blockd[i + 0].bmi = xd->mode_info_context->bmi[i + 0];
      blockd[i + 2].bmi = xd->mode_info_context->bmi[i + 2];

      if (mbmi->need_to_clamp_mvs) {
        clamp_mv_to_umv_border(&blockd[i + 0].bmi.as_mv[which_mv].as_mv, xd);
        clamp_mv_to_umv_border(&blockd[i + 2].bmi.as_mv[which_mv].as_mv, xd);
      }
      if (i == 0) {
        build_2x1_inter_predictor_wh(d0, d1, xd->scale_factor, tmp_y, 8, 16,
                                     which_mv, 0, 16, 1,
                                     &xd->subpix, mb_row * 16 + y, mb_col * 16);
        build_2x1_inter_predictor_wh(d0, d1, xd->scale_factor, tmp_y, 8, 16,
                                     which_mv, 0, 1, 8,
                                     &xd->subpix, mb_row * 16 + y, mb_col * 16);
      } else {
        build_2x1_inter_predictor_wh(d0, d1, xd->scale_factor, tmp_y + 8 * 16,
                                     8, 16, which_mv, 0, 1, 8,
                                     &xd->subpix, mb_row * 16 + y, mb_col * 16);
      }
    }
  } else {
    for (i = 0; i < 16; i += 2) {
      BLOCKD *d0 = &blockd[i];
      BLOCKD *d1 = &blockd[i + 1];
      const int x = (i & 3) * 4;
      const int y = (i >> 2) * 4;

      blockd[i + 0].bmi = xd->mode_info_context->bmi[i + 0];
      blockd[i + 1].bmi = xd->mode_info_context->bmi[i + 1];

      if (i >= 4 && (i & 3) != 0) continue;

      if (i == 0) {
        build_2x1_inter_predictor_wh(d0, d1, xd->scale_factor, tmp_y, 4, 16,
                                     which_mv, 0, 8, 1, &xd->subpix,
                                     mb_row * 16 + y, mb_col * 16 + x);
        build_2x1_inter_predictor_wh(d0, d1, xd->scale_factor, tmp_y, 4, 16,
                                     which_mv, 0, 1, 4, &xd->subpix,
                                     mb_row * 16 + y, mb_col * 16 + x);
      } else if (i < 4) {
        build_2x1_inter_predictor_wh(d0, d1, xd->scale_factor, tmp_y + x, 4, 16,
                                     which_mv, 0, 8, 1, &xd->subpix,
                                     mb_row * 16 + y, mb_col * 16 + x);
      } else {
        build_2x1_inter_predictor_wh(d0, d1, xd->scale_factor, tmp_y + y * 16,
                                     4, 16, which_mv, 0, 1, 4, &xd->subpix,
                                     mb_row * 16 + y, mb_col * 16 + x);
      }
    }
  }
  metric_1 = get_consistency_metric(xd, tmp_y, tmp_ystride);

  // Choose final weight for averaging
  weight = get_weight(xd, metric_1, metric_2);
  return weight;
}
#endif

static int get_implicit_compoundinter_weight(MACROBLOCKD *xd,
                                             int mb_row,
                                             int mb_col) {
  const int use_second_ref = xd->mode_info_context->mbmi.second_ref_frame > 0;
  int64_t metric_2 = 0, metric_1 = 0;
  int n, clamp_mvs, pre_stride;
  uint8_t *base_pre;
  int_mv ymv;
  uint8_t tmp_y[4096];
  const int tmp_ystride = 64;
  int weight;
  int edge[4];
  int block_size = 16 <<  xd->mode_info_context->mbmi.sb_type;

  if (!use_second_ref) return 0;
  if (!(xd->up_available || xd->left_available))
    return AVERAGE_WEIGHT;

  edge[0] = xd->mb_to_top_edge;
  edge[1] = xd->mb_to_bottom_edge;
  edge[2] = xd->mb_to_left_edge;
  edge[3] = xd->mb_to_right_edge;

  clamp_mvs = xd->mode_info_context->mbmi.need_to_clamp_secondmv;
  base_pre = xd->second_pre.y_buffer;
  pre_stride = xd->second_pre.y_stride;
  ymv.as_int = xd->mode_info_context->mbmi.mv[1].as_int;
  // First generate the second predictor
  for (n = 0; n < block_size; n += 16) {
    xd->mb_to_left_edge   = edge[2] - (n << 3);
    xd->mb_to_right_edge  = edge[3] + ((16 - n) << 3);
    if (clamp_mvs)
      clamp_mv_to_umv_border(&ymv.as_mv, xd);
    set_scaled_offsets(&xd->scale_factor[1], mb_row * 16, mb_col * 16 + n);
    // predict a single row of pixels
    vp9_build_inter_predictor(
        base_pre + scaled_buffer_offset(n, 0, pre_stride, &xd->scale_factor[1]),
        pre_stride, tmp_y + n, tmp_ystride, &ymv, &xd->scale_factor[1],
        16, 1, 0, &xd->subpix);
  }
  xd->mb_to_left_edge = edge[2];
  xd->mb_to_right_edge = edge[3];
  for (n = 0; n < block_size; n += 16) {
    xd->mb_to_top_edge    = edge[0] - (n << 3);
    xd->mb_to_bottom_edge = edge[1] + ((16 - n) << 3);
    if (clamp_mvs)
      clamp_mv_to_umv_border(&ymv.as_mv, xd);
    set_scaled_offsets(&xd->scale_factor[1], mb_row * 16 + n, mb_col * 16);
    // predict a single col of pixels
    vp9_build_inter_predictor(
        base_pre + scaled_buffer_offset(0, n, pre_stride, &xd->scale_factor[1]),
        pre_stride, tmp_y + n * tmp_ystride, tmp_ystride, &ymv,
        &xd->scale_factor[1], 1, 16, 0, &xd->subpix);
  }
  xd->mb_to_top_edge = edge[0];
  xd->mb_to_bottom_edge = edge[1];
  // Compute consistency metric
  metric_2 = get_consistency_metric(xd, tmp_y, tmp_ystride);

  clamp_mvs = xd->mode_info_context->mbmi.need_to_clamp_mvs;
  base_pre = xd->pre.y_buffer;
  pre_stride = xd->pre.y_stride;
  ymv.as_int = xd->mode_info_context->mbmi.mv[0].as_int;
  // Now generate the first predictor
  for (n = 0; n < block_size; n += 16) {
    xd->mb_to_left_edge   = edge[2] - (n << 3);
    xd->mb_to_right_edge  = edge[3] + ((16 - n) << 3);
    if (clamp_mvs)
      clamp_mv_to_umv_border(&ymv.as_mv, xd);
    set_scaled_offsets(&xd->scale_factor[0], mb_row * 16, mb_col * 16 + n);
    // predict a single row of pixels
    vp9_build_inter_predictor(
        base_pre + scaled_buffer_offset(n, 0, pre_stride, &xd->scale_factor[0]),
        pre_stride, tmp_y + n, tmp_ystride, &ymv, &xd->scale_factor[0],
        16, 1, 0, &xd->subpix);
  }
  xd->mb_to_left_edge = edge[2];
  xd->mb_to_right_edge = edge[3];
  for (n = 0; n < block_size; n += 16) {
    xd->mb_to_top_edge    = edge[0] - (n << 3);
    xd->mb_to_bottom_edge = edge[1] + ((16 - n) << 3);
    if (clamp_mvs)
      clamp_mv_to_umv_border(&ymv.as_mv, xd);
    set_scaled_offsets(&xd->scale_factor[0], mb_row * 16 + n, mb_col * 16);
    // predict a single col of pixels
    vp9_build_inter_predictor(
        base_pre + scaled_buffer_offset(0, n, pre_stride, &xd->scale_factor[0]),
        pre_stride, tmp_y + n * tmp_ystride, tmp_ystride, &ymv,
        &xd->scale_factor[0], 1, 16, 0, &xd->subpix);
  }
  xd->mb_to_top_edge = edge[0];
  xd->mb_to_bottom_edge = edge[1];
  metric_1 = get_consistency_metric(xd, tmp_y, tmp_ystride);

  // Choose final weight for averaging
  weight = get_weight(xd, metric_1, metric_2);
  return weight;
}

static void build_inter16x16_predictors_mby_w(MACROBLOCKD *xd,
                                              uint8_t *dst_y,
                                              int dst_ystride,
                                              int weight,
                                              int mb_row,
                                              int mb_col) {
867
868
869
870
  const int use_second_ref = xd->mode_info_context->mbmi.second_ref_frame > 0;
  int which_mv;

  for (which_mv = 0; which_mv < 1 + use_second_ref; ++which_mv) {
871
    const int clamp_mvs = which_mv ?
872
        xd->mode_info_context->mbmi.need_to_clamp_secondmv :
873
         xd->mode_info_context->mbmi.need_to_clamp_mvs;
874

875
876
877
    uint8_t *base_pre = which_mv ? xd->second_pre.y_buffer : xd->pre.y_buffer;
    int pre_stride = which_mv ? xd->second_pre.y_stride : xd->pre.y_stride;
    int_mv ymv;
878
    ymv.as_int = xd->mode_info_context->mbmi.mv[which_mv].as_int;
879

880
881
882
    if (clamp_mvs)
      clamp_mv_to_umv_border(&ymv.as_mv, xd);

883
884
885
    set_scaled_offsets(&xd->scale_factor[which_mv], mb_row * 16, mb_col * 16);

    vp9_build_inter_predictor(base_pre, pre_stride,
886
887
                              dst_y, dst_ystride,
                              &ymv, &xd->scale_factor[which_mv],
888
                              16, 16, which_mv ? weight : 0, &xd->subpix);
889
  }
890
891
}

892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
void vp9_build_inter16x16_predictors_mby(MACROBLOCKD *xd,
                                         uint8_t *dst_y,
                                         int dst_ystride,
                                         int mb_row,
                                         int mb_col) {
  int weight = get_implicit_compoundinter_weight(xd, mb_row, mb_col);

  build_inter16x16_predictors_mby_w(xd, dst_y, dst_ystride, weight,
                                    mb_row, mb_col);
}

#else

void vp9_build_inter16x16_predictors_mby(MACROBLOCKD *xd,
                                         uint8_t *dst_y,
                                         int dst_ystride,
                                         int mb_row,
                                         int mb_col) {
  const int use_second_ref = xd->mode_info_context->mbmi.second_ref_frame > 0;
  int which_mv;

  for (which_mv = 0; which_mv < 1 + use_second_ref; ++which_mv) {
    const int clamp_mvs = which_mv ?
         xd->mode_info_context->mbmi.need_to_clamp_secondmv :
         xd->mode_info_context->mbmi.need_to_clamp_mvs;

    uint8_t *base_pre = which_mv ? xd->second_pre.y_buffer : xd->pre.y_buffer;
    int pre_stride = which_mv ? xd->second_pre.y_stride : xd->pre.y_stride;
    int_mv ymv;
    ymv.as_int = xd->mode_info_context->mbmi.mv[which_mv].as_int;

    if (clamp_mvs)
      clamp_mv_to_umv_border(&ymv.as_mv, xd);

    set_scaled_offsets(&xd->scale_factor[which_mv], mb_row * 16, mb_col * 16);

    vp9_build_inter_predictor(base_pre, pre_stride,
                              dst_y, dst_ystride,
                              &ymv, &xd->scale_factor[which_mv],
                              16, 16, which_mv, &xd->subpix);
  }
}
#endif

#if CONFIG_IMPLICIT_COMPOUNDINTER_WEIGHT
static void build_inter16x16_predictors_mbuv_w(MACROBLOCKD *xd,
                                               uint8_t *dst_u,
                                               uint8_t *dst_v,
                                               int dst_uvstride,
                                               int weight,
                                               int mb_row,
                                               int mb_col) {
  const int use_second_ref = xd->mode_info_context->mbmi.second_ref_frame > 0;
  int which_mv;

947
948
949
950
951
  for (which_mv = 0; which_mv < 1 + use_second_ref; ++which_mv) {
    const int clamp_mvs =
        which_mv ? xd->mode_info_context->mbmi.need_to_clamp_secondmv
                 : xd->mode_info_context->mbmi.need_to_clamp_mvs;
    uint8_t *uptr, *vptr;
952
953
    int pre_stride = which_mv ? xd->second_pre.uv_stride
                              : xd->pre.uv_stride;
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
    int_mv _o16x16mv;
    int_mv _16x16mv;

    _16x16mv.as_int = xd->mode_info_context->mbmi.mv[which_mv].as_int;

    if (clamp_mvs)
      clamp_mv_to_umv_border(&_16x16mv.as_mv, xd);

    _o16x16mv = _16x16mv;
    /* calc uv motion vectors */
    if (_16x16mv.as_mv.row < 0)
      _16x16mv.as_mv.row -= 1;
    else
      _16x16mv.as_mv.row += 1;

    if (_16x16mv.as_mv.col < 0)
      _16x16mv.as_mv.col -= 1;
    else
      _16x16mv.as_mv.col += 1;

    _16x16mv.as_mv.row /= 2;
    _16x16mv.as_mv.col /= 2;

    _16x16mv.as_mv.row &= xd->fullpixel_mask;
    _16x16mv.as_mv.col &= xd->fullpixel_mask;

    uptr = (which_mv ? xd->second_pre.u_buffer : xd->pre.u_buffer);
    vptr = (which_mv ? xd->second_pre.v_buffer : xd->pre.v_buffer);

983
984
985
    set_scaled_offsets(&xd->scale_factor_uv[which_mv],
                       mb_row * 16, mb_col * 16);

986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
    vp9_build_inter_predictor_q4(
        uptr, pre_stride, dst_u, dst_uvstride, &_16x16mv, &_o16x16mv,
        &xd->scale_factor_uv[which_mv], 8, 8,
        which_mv ? weight : 0, &xd->subpix);

    vp9_build_inter_predictor_q4(
        vptr, pre_stride, dst_v, dst_uvstride, &_16x16mv, &_o16x16mv,
        &xd->scale_factor_uv[which_mv], 8, 8,
        which_mv ? weight : 0, &xd->subpix);
  }
}

void vp9_build_inter16x16_predictors_mbuv(MACROBLOCKD *xd,
                                          uint8_t *dst_u,
                                          uint8_t *dst_v,
For faster browsing, not all history is shown. View entire blame