vp9_reconinter.c 58 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
42
43
44
45
46
47
48
49
50
51
52
53
  if (scale->x_num == scale->x_den && scale->y_num == scale->y_den) {
    scale->scale_value_x = unscaled_value;
    scale->scale_value_y = unscaled_value;
    scale->set_scaled_offsets = set_offsets_without_scaling;
    scale->scale_motion_vector_q3_to_q4 =
        motion_vector_q3_to_q4_without_scaling;
    scale->scale_motion_vector_component_q4 =
        motion_vector_component_q4_without_scaling;
  } else {
    scale->scale_value_x = scale_value_x_with_scaling;
    scale->scale_value_y = scale_value_y_with_scaling;
    scale->set_scaled_offsets = set_offsets_with_scaling;
    scale->scale_motion_vector_q3_to_q4 =
        motion_vector_q3_to_q4_with_scaling;
    scale->scale_motion_vector_component_q4 =
        motion_vector_component_q4_with_scaling;
  }

54
55
56
57
58
59
  // 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.
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
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
#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
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
  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;
}
222
#endif
223
224
225
226
227
228
229
230
231
232
233
234
235

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);
  }

236
237
238
239
240
241
242
243
244
245
246
247
248
249
  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
250
#if CONFIG_ENABLE_6TAP
251
252
253
    case SIXTAP:
      xd->subpix.filter_x = xd->subpix.filter_y = vp9_sub_pel_filters_6;
      break;
Adrian Grange's avatar
Adrian Grange committed
254
255
#endif
  }
256
  assert(((intptr_t)xd->subpix.filter_x & 0xff) == 0);
257
258
}

259
void vp9_copy_mem16x16_c(const uint8_t *src,
260
                         int src_stride,
261
                         uint8_t *dst,
262
                         int dst_stride) {
John Koleszar's avatar
John Koleszar committed
263
  int r;
John Koleszar's avatar
John Koleszar committed
264

John Koleszar's avatar
John Koleszar committed
265
  for (r = 0; r < 16; r++) {
266
#if !(CONFIG_FAST_UNALIGNED)
John Koleszar's avatar
John Koleszar committed
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
    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
283
284

#else
285
286
287
288
    ((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
289
290

#endif
John Koleszar's avatar
John Koleszar committed
291
292
293
    src += src_stride;
    dst += dst_stride;
  }
John Koleszar's avatar
John Koleszar committed
294
295
}

296
void vp9_copy_mem8x8_c(const uint8_t *src,
297
                       int src_stride,
298
                       uint8_t *dst,
299
                       int dst_stride) {
John Koleszar's avatar
John Koleszar committed
300
301
302
  int r;

  for (r = 0; r < 8; r++) {
303
#if !(CONFIG_FAST_UNALIGNED)
John Koleszar's avatar
John Koleszar committed
304
305
306
307
308
309
310
311
    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
312
#else
313
314
    ((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
315
#endif
John Koleszar's avatar
John Koleszar committed
316
317
318
    src += src_stride;
    dst += dst_stride;
  }
John Koleszar's avatar
John Koleszar committed
319
320
}

321
void vp9_copy_mem8x4_c(const uint8_t *src,
322
                       int src_stride,
323
                       uint8_t *dst,
324
                       int dst_stride) {
John Koleszar's avatar
John Koleszar committed
325
326
327
  int r;

  for (r = 0; r < 4; r++) {
328
#if !(CONFIG_FAST_UNALIGNED)
John Koleszar's avatar
John Koleszar committed
329
330
331
332
333
334
335
336
    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
337
#else
338
339
    ((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
340
#endif
John Koleszar's avatar
John Koleszar committed
341
342
343
    src += src_stride;
    dst += dst_stride;
  }
John Koleszar's avatar
John Koleszar committed
344
345
}

346
347
348
349
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,
350
                               int w, int h, int weight,
351
                               const struct subpix_fn_table *subpix) {
352
  int_mv32 mv = scale->scale_motion_vector_q3_to_q4(mv_q3, scale);
353
  src += (mv.as_mv.row >> 4) * src_stride + (mv.as_mv.col >> 4);
354
  scale->predict[!!(mv.as_mv.col & 15)][!!(mv.as_mv.row & 15)][weight](
355
      src, src_stride, dst, dst_stride,
356
357
      subpix->filter_x[mv.as_mv.col & 15], scale->x_step_q4,
      subpix->filter_y[mv.as_mv.row & 15], scale->y_step_q4,
358
      w, h);
John Koleszar's avatar
John Koleszar committed
359
360
}

361
362
/* Like vp9_build_inter_predictor, but takes the full-pel part of the
 * mv separately, and the fractional part as a q4.
363
 */
364
365
void vp9_build_inter_predictor_q4(const uint8_t *src, int src_stride,
                                  uint8_t *dst, int dst_stride,
366
                                  const int_mv *mv_q4,
367
                                  const struct scale_factors *scale,
368
                                  int w, int h, int weight,
369
370
                                  const struct subpix_fn_table *subpix) {
  const int scaled_mv_row_q4 =
371
372
373
      scale->scale_motion_vector_component_q4(mv_q4->as_mv.row,
                                              scale->y_num, scale->y_den,
                                              scale->y_offset_q4);
374
  const int scaled_mv_col_q4 =
375
376
377
      scale->scale_motion_vector_component_q4(mv_q4->as_mv.col,
                                              scale->x_num, scale->x_den,
                                              scale->x_offset_q4);
378
379
380
  const int subpel_x = scaled_mv_col_q4 & 15;
  const int subpel_y = scaled_mv_row_q4 & 15;

381
  src += (scaled_mv_row_q4 >> 4) * src_stride + (scaled_mv_col_q4 >> 4);
382
  scale->predict[!!subpel_x][!!subpel_y][weight](
383
      src, src_stride, dst, dst_stride,
384
385
      subpix->filter_x[subpel_x], scale->x_step_q4,
      subpix->filter_y[subpel_y], scale->y_step_q4,
386
      w, h);
387
388
}

389
static void build_2x1_inter_predictor_wh(const BLOCKD *d0, const BLOCKD *d1,
390
                                         struct scale_factors *s,
391
392
393
394
395
396
                                         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) {
397
398
  struct scale_factors * scale = &s[which_mv];

399
  assert(d1->dst - d0->dst == block_size);
400
401
  assert(d1->pre == d0->pre + block_size);

402
  scale->set_scaled_offsets(scale, row, col);
403

404
405
406
407
408
  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,
409
                              predictor, stride,
410
                              &d0->bmi.as_mv[which_mv],
411
                              scale,
412
413
                              width, height,
                              weight, subpix);
414
415
416
417
418
419
420

  } 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,
421
                              predictor, stride,
422
                              &d0->bmi.as_mv[which_mv],
423
                              scale,
424
425
426
427
                              width > block_size ? block_size : width, height,
                              weight, subpix);

    if (width <= block_size) return;
428

429
    scale->set_scaled_offsets(scale, row, col + block_size);
430

431
432
    vp9_build_inter_predictor(*base_pre1 + d1->pre,
                              d1->pre_stride,
433
                              predictor + block_size, stride,
434
                              &d1->bmi.as_mv[which_mv],
435
                              scale,
436
437
                              width - block_size, height,
                              weight, subpix);
438
  }
John Koleszar's avatar
John Koleszar committed
439
440
}

441
static void build_2x1_inter_predictor(const BLOCKD *d0, const BLOCKD *d1,
442
                                      struct scale_factors *s,
443
444
445
                                      int block_size, int stride,
                                      int which_mv, int weight,
                                      const struct subpix_fn_table *subpix,
446
447
448
                                      int row, int col) {
  uint8_t *d0_predictor = *(d0->base_dst) + d0->dst;
  uint8_t *d1_predictor = *(d1->base_dst) + d1->dst;
449
  struct scale_factors * scale = &s[which_mv];
450
  stride = d0->dst_stride;
451

452
  assert(d1_predictor - d0_predictor == block_size);
453
  assert(d1->pre == d0->pre + block_size);
454

455
  scale->set_scaled_offsets(scale, row, col);
John Koleszar's avatar
John Koleszar committed
456

457
458
  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
459

460
461
    vp9_build_inter_predictor(*base_pre + d0->pre,
                              d0->pre_stride,
462
                              d0_predictor, stride,
463
                              &d0->bmi.as_mv[which_mv],
464
                              scale,
465
466
467
468
469
                              2 * block_size, block_size,
                              weight, subpix);
  } 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
470

471
472
    vp9_build_inter_predictor(*base_pre0 + d0->pre,
                              d0->pre_stride,
473
                              d0_predictor, stride,
474
                              &d0->bmi.as_mv[which_mv],
475
                              scale,
476
477
                              block_size, block_size,
                              weight, subpix);
478

479
    scale->set_scaled_offsets(scale, row, col + block_size);
John Koleszar's avatar
John Koleszar committed
480

481
482
    vp9_build_inter_predictor(*base_pre1 + d1->pre,
                              d1->pre_stride,
483
                              d1_predictor, stride,
484
                              &d1->bmi.as_mv[which_mv],
485
                              scale,
486
487
                              block_size, block_size,
                              weight, subpix);
John Koleszar's avatar
John Koleszar committed
488
  }
John Koleszar's avatar
John Koleszar committed
489
490
}

John Koleszar's avatar
John Koleszar committed
491
492
493
494
495
496
497
498
499
500
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
501
  if (mv->col < (xd->mb_to_left_edge - ((16 + VP9_INTERP_EXTEND) << 3)))
John Koleszar's avatar
John Koleszar committed
502
    mv->col = xd->mb_to_left_edge - (16 << 3);
John Koleszar's avatar
John Koleszar committed
503
  else if (mv->col > xd->mb_to_right_edge + ((15 + VP9_INTERP_EXTEND) << 3))
John Koleszar's avatar
John Koleszar committed
504
505
    mv->col = xd->mb_to_right_edge + (16 << 3);

John Koleszar's avatar
John Koleszar committed
506
  if (mv->row < (xd->mb_to_top_edge - ((16 + VP9_INTERP_EXTEND) << 3)))
John Koleszar's avatar
John Koleszar committed
507
    mv->row = xd->mb_to_top_edge - (16 << 3);
John Koleszar's avatar
John Koleszar committed
508
  else if (mv->row > xd->mb_to_bottom_edge + ((15 + VP9_INTERP_EXTEND) << 3))
John Koleszar's avatar
John Koleszar committed
509
    mv->row = xd->mb_to_bottom_edge + (16 << 3);
510
511
512
}

/* A version of the above function for chroma block MVs.*/
John Koleszar's avatar
John Koleszar committed
513
static void clamp_uvmv_to_umv_border(MV *mv, const MACROBLOCKD *xd) {
John Koleszar's avatar
John Koleszar committed
514
515
516
  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
517
            (xd->mb_to_left_edge - (16 << 3)) >> 1 : mv->col;
John Koleszar's avatar
John Koleszar committed
518
  mv->col = (2 * mv->col > xd->mb_to_right_edge + ((15 + extend) << 3)) ?
John Koleszar's avatar
John Koleszar committed
519
520
            (xd->mb_to_right_edge + (16 << 3)) >> 1 : mv->col;

John Koleszar's avatar
John Koleszar committed
521
  mv->row = (2 * mv->row < (xd->mb_to_top_edge - ((16 + extend) << 3))) ?
John Koleszar's avatar
John Koleszar committed
522
            (xd->mb_to_top_edge - (16 << 3)) >> 1 : mv->row;
John Koleszar's avatar
John Koleszar committed
523
  mv->row = (2 * mv->row > xd->mb_to_bottom_edge + ((15 + extend) << 3)) ?
John Koleszar's avatar
John Koleszar committed
524
            (xd->mb_to_bottom_edge + (16 << 3)) >> 1 : mv->row;
525
526
}

527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
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
#if !CONFIG_IMPLICIT_COMPOUNDINTER_WEIGHT
// TODO(jkoleszar): yet another mv clamping function :-(
MV clamp_mv_to_umv_border_sb(const MV *src_mv,
    int bwl, int bhl,
    int mb_to_left_edge, int mb_to_top_edge,
    int mb_to_right_edge, int mb_to_bottom_edge) {
  /* 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.
   */
  const int epel_left = (VP9_INTERP_EXTEND + (4 << bwl)) << 3;
  const int epel_right = epel_left - (1 << 3);
  const int epel_top = (VP9_INTERP_EXTEND + (4 << bhl)) << 3;
  const int epel_bottom = epel_top - (1 << 3);
  MV clamped_mv;
  clamped_mv.col = clamp(src_mv->col,
                         mb_to_left_edge - epel_left,
                         mb_to_right_edge + epel_right);
  clamped_mv.row = clamp(src_mv->row,
                         mb_to_top_edge - epel_top,
                         mb_to_bottom_edge + epel_bottom);
  return clamped_mv;
}

struct build_inter_predictors_args {
  MACROBLOCKD *xd;
  uint8_t* dst[MAX_MB_PLANE];
  int dst_stride[MAX_MB_PLANE];
  int x;
  int y;
};
static void build_inter_predictors(int plane, int block,
                                   BLOCK_SIZE_TYPE bsize,
                                   int pred_w, int pred_h,
                                   void *argv) {
  const struct build_inter_predictors_args* const arg = argv;
  const int bwl = pred_w, bw = 4 << bwl;
  const int bhl = pred_h, bh = 4 << bhl;
  const int x_idx = block & ((1 << bwl) - 1), y_idx = block >> bwl;
  const int x = x_idx * 4, y = y_idx * 4;
  MACROBLOCKD * const xd = arg->xd;
  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 MV* const mv = (xd->mode_info_context->mbmi.mode == SPLITMV)
         ? &xd->block[block].bmi.as_mv[which_mv].as_mv
         : &xd->mode_info_context->mbmi.mv[which_mv].as_mv;

    const uint8_t * const base_pre = which_mv ? xd->second_pre.y_buffer
                                             : xd->pre.y_buffer;
    const int pre_stride = which_mv ? xd->second_pre.y_stride
                                    : xd->pre.y_stride;
    const uint8_t *const pre = base_pre +
        scaled_buffer_offset(x, y, pre_stride, &xd->scale_factor[which_mv]);
    struct scale_factors * const scale =
      plane == 0 ? &xd->scale_factor[which_mv] : &xd->scale_factor_uv[which_mv];

    int_mv clamped_mv;
    clamped_mv.as_mv = clamp_mv_to_umv_border_sb(mv, bwl, bhl,
                                                 xd->mb_to_left_edge,
                                                 xd->mb_to_top_edge,
                                                 xd->mb_to_right_edge,
                                                 xd->mb_to_bottom_edge);

    scale->set_scaled_offsets(scale, arg->y + y, arg->x + x);

    vp9_build_inter_predictor(pre, pre_stride,
                              arg->dst[plane], arg->dst_stride[plane],
                              &clamped_mv, &xd->scale_factor[which_mv],
                              bw, bh, which_mv, &xd->subpix);
  }
}
void vp9_build_inter_predictors_sby(MACROBLOCKD *xd,
                                    uint8_t *dst_y,
                                    int dst_ystride,
                                    int mb_row,
                                    int mb_col,
                                    BLOCK_SIZE_TYPE bsize) {
  struct build_inter_predictors_args args = {
    xd, {dst_y, NULL, NULL}, {dst_ystride, 0, 0}, mb_col * 16, mb_row * 16
  };
  foreach_predicted_block_in_plane(xd, bsize, 0, build_inter_predictors, &args);
}
#endif

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
#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;
835
  struct scale_factors *scale;
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850

  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
851
  scale = &xd->scale_factor[1];
852
853
854
855
856
  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);
857
    scale->set_scaled_offsets(scale, mb_row * 16, mb_col * 16 + n);
858
    // predict a single row of pixels
859
860
861
    vp9_build_inter_predictor(base_pre +
        scaled_buffer_offset(n, 0, pre_stride, scale),
        pre_stride, tmp_y + n, tmp_ystride, &ymv, scale, 16, 1, 0, &xd->subpix);
862
863
864
865
866
867
868
869
  }
  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);
870
    scale->set_scaled_offsets(scale, mb_row * 16 + n, mb_col * 16);
871
    // predict a single col of pixels
872
873
    vp9_build_inter_predictor(base_pre +
        scaled_buffer_offset(0, n, pre_stride, scale),
874
        pre_stride, tmp_y + n * tmp_ystride, tmp_ystride, &ymv,
875
        scale, 1, 16, 0, &xd->subpix);
876
877
878
879
880
881
882
883
884
885
886
  }
  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
887
  scale = &xd->scale_factor[0];
888
889
890
891
892
  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);
893
    scale->set_scaled_offsets(scale, mb_row * 16, mb_col * 16 + n);
894
    // predict a single row of pixels
895
896
897
    vp9_build_inter_predictor(base_pre +
        scaled_buffer_offset(n, 0, pre_stride, scale),
        pre_stride, tmp_y + n, tmp_ystride, &ymv, scale, 16, 1, 0, &xd->subpix);
898
899
900
901
902
903
904
905
  }
  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);
906
    scale->set_scaled_offsets(scale, mb_row * 16 + n, mb_col * 16);
907
    // predict a single col of pixels
908
909
    vp9_build_inter_predictor(base_pre +
        scaled_buffer_offset(0, n, pre_stride, scale),
910
        pre_stride, tmp_y + n * tmp_ystride, tmp_ystride, &ymv,
911
        scale, 1, 16, 0, &xd->subpix);
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
  }
  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) {
928
929
930
931
  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) {
932
    const int clamp_mvs = which_mv ?
933
        xd->mode_info_context->mbmi.need_to_clamp_secondmv :
934
         xd->mode_info_context->mbmi.need_to_clamp_mvs;
935

936
937
938
    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;
939
940
    struct scale_factors *scale = &xd->scale_factor[which_mv];

941
    ymv.as_int = xd->mode_info_context->mbmi.mv[which_mv].as_int;
942

943
944
945
    if (clamp_mvs)
      clamp_mv_to_umv_border(&ymv.as_mv, xd);

946
    scale->set_scaled_offsets(scale, mb_row * 16, mb_col * 16);
947

948
949
950
    vp9_build_inter_predictor(base_pre, pre_stride, dst_y, dst_ystride,
                              &ymv, scale, 16, 16,
                              which_mv ? weight : 0, &xd->subpix);
951
  }
952
}
953
954
955
956
957
958
959
960
961
962
963
964
965
#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;

966
967
968
969
970
  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;
971
972
    int pre_stride = which_mv ? xd->second_pre.uv_stride
                              : xd->pre.uv_stride;
973
    int_mv mv;
974

975
    struct scale_factors *scale = &xd->scale_factor_uv[which_mv];
976
    mv.as_int = xd->mode_info_context->mbmi.mv[which_mv].as_int;
977

978
979

    if (clamp_mvs)
980
      clamp_mv_to_umv_border(&mv.as_mv, xd);
981
982
983
984

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

985
    scale->set_scaled_offsets(scale, mb_row * 16, mb_col * 16);
986

987
    vp9_build_inter_predictor_q4(
988
        uptr, pre_stride, dst_u, dst_uvstride, &mv,
989
        scale, 8, 8, which_mv ? weight : 0, &xd->subpix);
990
991

    vp9_build_inter_predictor_q4(
992
        vptr, pre_stride, dst_v, dst_uvstride, &mv,
993
        scale, 8, 8, which_mv ? weight : 0, &xd->subpix);
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
  }
}

void vp9_build_inter16x16_predictors_mbuv(MACROBLOCKD *xd,
                                          uint8_t *dst_u,
                                          uint8_t *dst_v,
                                          int dst_uvstride,
                                          int mb_row,
                                          int mb_col) {
#ifdef USE_IMPLICIT_WEIGHT_UV
  int weight = get_implicit_compoundinter_weight(xd, mb_row, mb_col);
#else
  int weight = AVERAGE_WEIGHT;
#endif
  build_inter16x16_predictors_mbuv_w(xd, dst_u, dst_v, dst_uvstride,
                                     weight, mb_row, mb_col);
}

#else

void vp9_build_inter16x16_predictors_mbuv(MACROBLOCKD *xd,
                                          uint8_t *dst_u,
                                          uint8_t *dst_v,
                                          int dst_uvstride,
                                          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 *uptr, *vptr;
    int pre_stride = which_mv ? xd->second_pre.uv_stride
                              : xd->pre.uv_stride;
1030
    int_mv mv;
1031

1032
    struct scale_factors *scale = &xd->scale_factor_uv[which_mv];
1033
    mv.as_int = xd->mode_info_context->mbmi.mv[which_mv].as_int;
1034

1035
1036

    if (clamp_mvs)
1037
      clamp_mv_to_umv_border(&mv.as_mv, xd);
1038
1039
1040
1041

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

1042
    scale->set_scaled_offsets(scale, mb_row * 16, mb_col * 16);
1043
1044

    vp9_build_inter_predictor_q4(
1045
        uptr, pre_stride, dst_u, dst_uvstride, &mv,
1046
        scale, 8, 8,
1047
1048
1049
        which_mv << (2 * CONFIG_IMPLICIT_COMPOUNDINTER_WEIGHT), &xd->subpix);

    vp9_build_inter_predictor_q4(
1050
        vptr, pre_stride, dst_v, dst_uvstride, &mv,
1051
        scale, 8, 8,
1052
1053
1054
1055
1056
1057
        which_mv << (2 * CONFIG_IMPLICIT_COMPOUNDINTER_WEIGHT), &xd->subpix);
  }
}
#endif

#if CONFIG_IMPLICIT_COMPOUNDINTER_WEIGHT
1058
1059
1060
1061
1062
1063
1064
1065
1066
static void build_inter_predictors_sby_w(MACROBLOCKD *x,
                                         uint8_t *dst_y,
                                         int dst_ystride,
                                         int weight,
                                         int mb_row,
                                         int mb_col,
                                         BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize),  bw = 1 << bwl;
  const int bhl = mb_height_log2(bsize), bh = 1 << bhl;
1067
1068
1069
1070
1071
1072
1073
1074
1075
  uint8_t *y1 = x->pre.y_buffer;
  uint8_t *y2 = x->second_pre.y_buffer;
  int edge[4], n;

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

1076
1077
  for (n = 0; n < bw * bh; n++) {
    const int x_idx = n & (bw - 1), y_idx = n >> bwl;
1078

1079
1080
1081
1082
    x->mb_to_top_edge    = edge[0] -           ((y_idx  * 16) << 3);
    x->mb_to_bottom_edge = edge[1] + (((bh - 1 - y_idx) * 16) << 3);
    x->mb_to_left_edge   = edge[2] -           ((x_idx  * 16) << 3);
    x->mb_to_right_edge  = edge[3] + (((bw - 1 - x_idx) * 16) << 3);
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109

    x->pre.y_buffer = y1 + scaled_buffer_offset(x_idx * 16,
                                                y_idx * 16,
                                                x->pre.y_stride,
                                                &x->scale_factor[0]);
    if (x->mode_info_context->mbmi.second_ref_frame > 0) {
      x->second_pre.y_buffer = y2 +
          scaled_buffer_offset(x_idx * 16,
                               y_idx * 16,
                               x->second_pre.y_stride,
                               &x->scale_factor[1]);
    }
    build_inter16x16_predictors_mby_w(x,
        dst_y + y_idx * 16 * dst_ystride  + x_idx * 16,
        dst_ystride, weight, mb_row + y_idx, mb_col + x_idx);
  }
  x->mb_to_top_edge    = edge[0];
  x->mb_to_bottom_edge = edge[1];
  x->mb_to_left_edge   = edge[2];
  x->mb_to_right_edge  = edge[3];

  x->pre.y_buffer = y1;
  if (x->mode_info_context->mbmi.second_ref_frame > 0) {
    x->second_pre.y_buffer = y2;
  }
}

1110
void vp9_build_inter_predictors_sby(MACROBLOCKD *x,
1111
1112
1113
                                         uint8_t *dst_y,
                                         int dst_ystride,
                                         int mb_row,
1114
1115
                                         int mb_col,
                                         BLOCK_SIZE_TYPE bsize) {
1116
  int weight = get_implicit_compoundinter_weight(x, mb_row, mb_col);
1117
1118
  build_inter_predictors_sby_w(x, dst_y, dst_ystride, weight,
                                    mb_row, mb_col, bsize);
1119
1120
1121
1122
}
#endif

#if CONFIG_IMPLICIT_COMPOUNDINTER_WEIGHT
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
static void build_inter_predictors_sbuv_w(MACROBLOCKD *x,
                                          uint8_t *dst_u,
                                          uint8_t *dst_v,
                                          int dst_uvstride,
                                          int weight,
                                          int mb_row,
                                          int mb_col,
                                          BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize),  bw = 1 << bwl;
  const int bhl = mb_height_log2(bsize), bh = 1 << bhl;
1133
1134
1135
1136
1137
1138
1139
1140
1141
  uint8_t *u1 = x->pre.u_buffer, *v1 = x->pre.v_buffer;
  uint8_t *u2 = x->second_pre.u_buffer, *v2 = x->second_pre.v_buffer;
  int edge[4], n;

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

1142
  for (n = 0; n < bw * bh; n++) {
1143
    int scaled_uv_offset;
1144
    const int x_idx = n & (bw - 1), y_idx = n >> bwl;
1145

1146
1147
1148
1149
    x->mb_to_top_edge    = edge[0] -           ((y_idx  * 16) << 3);
    x->mb_to_bottom_edge = edge[1] + (((bh - 1 - y_idx) * 16) << 3);
    x->mb_to_left_edge   = edge[2] -           ((x_idx  * 16) << 3);
    x->mb_to_right_edge  = edge[3] + (((bw - 1 - x_idx) * 16) << 3);
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185

    scaled_uv_offset = scaled_buffer_offset(x_idx * 8,
                                            y_idx * 8,
                                            x->pre.uv_stride,
                                            &x->scale_factor_uv[0]);
    x->pre.u_buffer = u1 + scaled_uv_offset;
    x->pre.v_buffer = v1 + scaled_uv_offset;

    if (x->mode_info_context->mbmi.second_ref_frame > 0) {
      scaled_uv_offset = scaled_buffer_offset(x_idx * 8,
                                              y_idx * 8,
                                              x->second_pre.uv_stride,
                                              &x->scale_factor_uv[1]);
      x->second_pre.u_buffer = u2 + scaled_uv_offset;
      x->second_pre.v_buffer = v2 + scaled_uv_offset;
    }

    build_inter16x16_predictors_mbuv_w(x,
        dst_u + y_idx *  8 * dst_uvstride + x_idx *  8,
        dst_v + y_idx *  8 * dst_uvstride + x_idx *  8,
        dst_uvstride, weight, mb_row + y_idx, mb_col + x_idx);
  }
  x->mb_to_top_edge    = edge[0];
  x->mb_to_bottom_edge = edge[1];
  x->mb_to_left_edge   = edge[2];
  x->mb_to_right_edge  = edge[3];

  x->pre.u_buffer = u1;
  x->pre.v_buffer = v1;

  if (x->mode_info_context->mbmi.second_ref_frame > 0) {
    x->second_pre.u_buffer = u2;
    x->second_pre.v_buffer = v2;
  }
}

1186
1187
1188
1189
1190
1191
1192
void vp9_build_inter_predictors_sbuv(MACROBLOCKD *xd,
                                     uint8_t *dst_u,
                                     uint8_t *dst_v,
                                     int dst_uvstride,
                                     int mb_row,
                                     int mb_col,
                                     BLOCK_SIZE_TYPE bsize) {
1193
1194
1195
1196
1197
#ifdef USE_IMPLICIT_WEIGHT_UV
  int weight = get_implicit_compoundinter_weight(xd, mb_row, mb_col);
#else
  int weight = AVERAGE_WEIGHT;
#endif
1198
1199
  build_inter_predictors_sbuv_w(xd, dst_u, dst_v, dst_uvstride,
                                weight, mb_row, mb_col, bsize);
1200
1201
1202
1203
}

#else

1204
1205
1206
1207
1208
1209
1210
1211
1212
void vp9_build_inter_predictors_sbuv(MACROBLOCKD *x,
                                     uint8_t *dst_u,
                                     uint8_t *dst_v,
                                     int dst_uvstride,
                                     int mb_row,
                                     int mb_col,
                                     BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize),  bw = 1 << bwl;
  const int bhl = mb_height_log2(bsize), bh = 1 << bhl;
1213
1214
1215
1216
1217
1218
1219
1220
1221
  uint8_t *u1 = x->pre.u_buffer, *v1 = x->pre.v_buffer;
  uint8_t *u2 = x->second_pre.u_buffer, *v2 = x->second_pre.v_buffer;
  int edge[4], n;

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

1222
  for (n = 0; n < bw * bh; n++) {
1223
    int scaled_uv_offset;
1224
    const int x_idx = n & (bw - 1), y_idx = n >> bwl;
1225

1226
1227
1228
1229
    x->mb_to_top_edge    = edge[0] -           ((y_idx  * 16) << 3);
    x->mb_to_bottom_edge = edge[1] + (((bh - 1 - y_idx) * 16) << 3);
    x->mb_to_left_edge   = edge[2] -           ((x_idx  * 16) << 3);
    x->mb_to_right_edge  = edge[3] + (((bw - 1 - x_idx) * 16) << 3);
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258

    scaled_uv_offset = scaled_buffer_offset(x_idx * 8,
                                            y_idx * 8,
                                            x->pre.uv_stride,
                                            &x->scale_factor_uv[0]);
    x->pre.u_buffer = u1 + scaled_uv_offset;
    x->pre.v_buffer = v1 + scaled_uv_offset;

    if (x->mode_info_context->mbmi.second_ref_frame > 0) {
      scaled_uv_offset = scaled_buffer_offset(x_idx * 8,
                                              y_idx * 8,
                                              x->second_pre.uv_stride,
                                              &x->scale_factor_uv[1]);
      x->second_pre.u_buffer = u2 + scaled_uv_offset;
      x->second_pre.v_buffer = v2 + scaled_uv_offset;
    }

    vp9_build_inter16x16_predictors_mbuv(x,
        dst_u + y_idx *  8 * dst_uvstride + x_idx *  8,
        dst_v + y_idx *  8 * dst_uvstride + x_idx *  8,
        dst_uvstride, mb_row + y_idx, mb_col + x_idx);
  }
  x->mb_to_top_edge    = edge[0];
  x->mb_to_bottom_edge = edge[1];
  x->mb_to_left_edge   = edge[2];
  x->mb_to_right_edge  = edge[3];

  x->pre.u_buffer = u1;
  x->pre.v_buffer = v1;
1259

1260
1261
1262
  if (x->mode_info_context->mbmi.second_ref_frame > 0) {
    x->second_pre.u_buffer = u2;
    x->second_pre.v_buffer = v2;
1263
  }
1264
}
1265
#endif
John Koleszar's avatar
John Koleszar committed
1266

1267
1268
1269
void vp9_build_inter_predictors_sb(MACROBLOCKD *mb,
                                   int mb_row, int mb_col,
                                   BLOCK_SIZE_TYPE bsize) {
1270
1271
1272
1273
1274
1275
  uint8_t *const y = mb->dst.y_buffer;
  uint8_t *const u = mb->dst.u_buffer;
  uint8_t *const v = mb->dst.v_buffer;
  const int y_stride = mb->dst.y_stride;
  const int uv_stride = mb->dst.uv_stride;

1276
1277
  vp9_build_inter_predictors_sby(mb, y, y_stride, mb_row, mb_col, bsize);
  vp9_build_inter_predictors_sbuv(mb, u, v, uv_stride, mb_row, mb_col, bsize);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
1278
#if CONFIG_COMP_INTERINTRA_PRED
1279
  if (mb->mode_info_context->mbmi.second_ref_frame == INTRA_FRAME) {
1280
1281
1282
1283
1284
1285
    if (bsize == BLOCK_SIZE_SB32X32)
      vp9_build_interintra_32x32_predictors_sb(mb, y, u, v,
                                               y_stride, uv_stride);
    else
      vp9_build_interintra_64x64_predictors_sb(mb, y, u, v,
                                               y_stride, uv_stride);
1286
1287
  }
#endif
Ronald S. Bultje's avatar
Ronald S. Bultje committed
1288
1289
}

1290
static void build_inter4x4_predictors_mb(MACROBLOCKD *xd,
1291
                                         int mb_row, int mb_col) {
John Koleszar's avatar
John Koleszar committed
12