rd.c 53.9 KB
Newer Older
Jingning Han's avatar
Jingning Han committed
1
/*
Yaowu Xu's avatar
Yaowu Xu committed
2
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
Jingning Han's avatar
Jingning Han committed
3
 *
Yaowu Xu's avatar
Yaowu Xu committed
4
5
6
7
8
9
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
Jingning Han's avatar
Jingning Han committed
10
11
12
13
14
15
 */

#include <assert.h>
#include <math.h>
#include <stdio.h>

Yaowu Xu's avatar
Yaowu Xu committed
16
#include "./av1_rtcd.h"
Jingning Han's avatar
Jingning Han committed
17

Yaowu Xu's avatar
Yaowu Xu committed
18
19
#include "aom_dsp/aom_dsp_common.h"
#include "aom_mem/aom_mem.h"
20
21
22
#include "aom_ports/bitops.h"
#include "aom_ports/mem.h"
#include "aom_ports/system_state.h"
Jingning Han's avatar
Jingning Han committed
23

24
25
26
27
28
29
30
31
32
#include "av1/common/common.h"
#include "av1/common/entropy.h"
#include "av1/common/entropymode.h"
#include "av1/common/mvref_common.h"
#include "av1/common/pred_common.h"
#include "av1/common/quant_common.h"
#include "av1/common/reconinter.h"
#include "av1/common/reconintra.h"
#include "av1/common/seg_common.h"
33

34
#include "av1/encoder/av1_quantize.h"
35
36
37
38
39
40
41
42
#include "av1/encoder/cost.h"
#include "av1/encoder/encodemb.h"
#include "av1/encoder/encodemv.h"
#include "av1/encoder/encoder.h"
#include "av1/encoder/mcomp.h"
#include "av1/encoder/ratectrl.h"
#include "av1/encoder/rd.h"
#include "av1/encoder/tokenize.h"
Jingning Han's avatar
Jingning Han committed
43

44
#define RD_THRESH_POW 1.25
Jingning Han's avatar
Jingning Han committed
45
46
47
48
49
50
51
52

// Factor to weigh the rate for switchable interp filters.
#define SWITCHABLE_INTERP_RATE_FACTOR 1

// The baseline rd thresholds for breaking out of the rd loop for
// certain modes are assumed to be based on 8x8 blocks.
// This table is used to correct for block size.
// The factors here are << 2 (2 = x0.5, 32 = x8 etc).
53
static const uint8_t rd_thresh_block_size_factor[BLOCK_SIZES_ALL] = {
54
#if CONFIG_CHROMA_2X2 || CONFIG_CHROMA_SUB8X8
Jingning Han's avatar
Jingning Han committed
55
56
  2,  2,  2,
#endif
57
  2,  3,  3,  4, 6, 6, 8, 12, 12, 16, 24, 24, 32,
58
#if CONFIG_EXT_PARTITION
59
  48, 48, 64,
60
#endif  // CONFIG_EXT_PARTITION
61
  4,  4,  8,  8
Jingning Han's avatar
Jingning Han committed
62
63
};

64
65
void av1_fill_mode_rates(AV1_COMMON *const cm, MACROBLOCK *x,
                         FRAME_CONTEXT *fc) {
Jingning Han's avatar
Jingning Han committed
66
67
  int i, j;

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
  if (cm->frame_type == KEY_FRAME) {
#if CONFIG_EXT_PARTITION_TYPES
    for (i = 0; i < PARTITION_PLOFFSET; ++i)
      av1_cost_tokens(x->partition_cost[i], fc->partition_prob[i],
                      av1_partition_tree);
    for (; i < PARTITION_CONTEXTS_PRIMARY; ++i)
      av1_cost_tokens(x->partition_cost[i], fc->partition_prob[i],
                      av1_ext_partition_tree);
#else
    for (i = 0; i < PARTITION_CONTEXTS_PRIMARY; ++i)
      av1_cost_tokens(x->partition_cost[i], fc->partition_prob[i],
                      av1_partition_tree);
#endif  // CONFIG_EXT_PARTITION_TYPES
#if CONFIG_UNPOISON_PARTITION_CTX
    for (; i < PARTITION_CONTEXTS_PRIMARY + PARTITION_BLOCK_SIZES; ++i) {
      aom_prob p = fc->partition_prob[i][PARTITION_VERT];
      assert(p > 0);
      x->partition_cost[i][PARTITION_NONE] = INT_MAX;
      x->partition_cost[i][PARTITION_HORZ] = INT_MAX;
      x->partition_cost[i][PARTITION_VERT] = av1_cost_bit(p, 0);
      x->partition_cost[i][PARTITION_SPLIT] = av1_cost_bit(p, 1);
    }
    for (; i < PARTITION_CONTEXTS_PRIMARY + 2 * PARTITION_BLOCK_SIZES; ++i) {
      aom_prob p = fc->partition_prob[i][PARTITION_HORZ];
      assert(p > 0);
      x->partition_cost[i][PARTITION_NONE] = INT_MAX;
      x->partition_cost[i][PARTITION_HORZ] = av1_cost_bit(p, 0);
      x->partition_cost[i][PARTITION_VERT] = INT_MAX;
      x->partition_cost[i][PARTITION_SPLIT] = av1_cost_bit(p, 1);
    }
    x->partition_cost[PARTITION_CONTEXTS][PARTITION_NONE] = INT_MAX;
    x->partition_cost[PARTITION_CONTEXTS][PARTITION_HORZ] = INT_MAX;
    x->partition_cost[PARTITION_CONTEXTS][PARTITION_VERT] = INT_MAX;
    x->partition_cost[PARTITION_CONTEXTS][PARTITION_SPLIT] = 0;
#endif  // CONFIG_UNPOISON_PARTITION_CTX
  }

Jingning Han's avatar
Jingning Han committed
105
106
  for (i = 0; i < INTRA_MODES; ++i)
    for (j = 0; j < INTRA_MODES; ++j)
107
      av1_cost_tokens_from_cdf(x->y_mode_costs[i][j], av1_kf_y_mode_cdf[i][j],
108
                               av1_intra_mode_inv);
Jingning Han's avatar
Jingning Han committed
109

110
  for (i = 0; i < BLOCK_SIZE_GROUPS; ++i)
111
    av1_cost_tokens_from_cdf(x->mbmode_cost[i], fc->y_mode_cdf[i],
112
                             av1_intra_mode_inv);
113
114
115
116
117
118
119
  const int *uv_mode_inv_map =
#if CONFIG_CFL
      // CfL codes the uv_mode without reordering it
      NULL;
#else
      av1_intra_mode_inv;
#endif
120
  for (i = 0; i < INTRA_MODES; ++i)
121
    av1_cost_tokens_from_cdf(x->intra_uv_mode_cost[i], fc->uv_mode_cdf[i],
122
                             uv_mode_inv_map);
Jingning Han's avatar
Jingning Han committed
123
124

  for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; ++i)
125
    av1_cost_tokens(x->switchable_interp_costs[i],
Yaowu Xu's avatar
Yaowu Xu committed
126
                    fc->switchable_interp_prob[i], av1_switchable_interp_tree);
hui su's avatar
hui su committed
127

128
#if CONFIG_PALETTE
hui su's avatar
hui su committed
129
  for (i = 0; i < PALETTE_BLOCK_SIZES; ++i) {
130
    av1_cost_tokens_from_cdf(x->palette_y_size_cost[i],
131
                             fc->palette_y_size_cdf[i], NULL);
132
    av1_cost_tokens_from_cdf(x->palette_uv_size_cost[i],
133
                             fc->palette_uv_size_cdf[i], NULL);
hui su's avatar
hui su committed
134
135
  }

136
  for (i = 0; i < PALETTE_SIZES; ++i) {
137
    for (j = 0; j < PALETTE_COLOR_INDEX_CONTEXTS; ++j) {
138
      av1_cost_tokens_from_cdf(x->palette_y_color_cost[i][j],
139
                               fc->palette_y_color_index_cdf[i][j], NULL);
140
      av1_cost_tokens_from_cdf(x->palette_uv_color_cost[i][j],
141
                               fc->palette_uv_color_index_cdf[i][j], NULL);
hui su's avatar
hui su committed
142
    }
143
144
  }
#endif  // CONFIG_PALETTE
145

Jingning Han's avatar
Jingning Han committed
146
  for (i = 0; i < MAX_TX_DEPTH; ++i)
147
    for (j = 0; j < TX_SIZE_CONTEXTS; ++j)
148
      av1_cost_tokens(x->tx_size_cost[i][j], fc->tx_size_probs[i][j],
Yaowu Xu's avatar
Yaowu Xu committed
149
                      av1_tx_size_tree[i]);
150

151
#if CONFIG_EXT_TX
152
153
154
  for (i = TX_4X4; i < EXT_TX_SIZES; ++i) {
    int s;
    for (s = 1; s < EXT_TX_SETS_INTER; ++s) {
155
      if (use_inter_ext_tx_for_txsize[s][i]) {
156
        av1_cost_tokens(x->inter_tx_type_costs[s][i],
Yaowu Xu's avatar
Yaowu Xu committed
157
                        fc->inter_ext_tx_prob[s][i], av1_ext_tx_inter_tree[s]);
158
159
160
      }
    }
    for (s = 1; s < EXT_TX_SETS_INTRA; ++s) {
161
      if (use_intra_ext_tx_for_txsize[s][i]) {
162
        for (j = 0; j < INTRA_MODES; ++j)
163
          av1_cost_tokens(x->intra_tx_type_costs[s][i][j],
Yaowu Xu's avatar
Yaowu Xu committed
164
165
                          fc->intra_ext_tx_prob[s][i][j],
                          av1_ext_tx_intra_tree[s]);
166
167
      }
    }
168
  }
169
#else
170
171
  for (i = TX_4X4; i < EXT_TX_SIZES; ++i) {
    for (j = 0; j < TX_TYPES; ++j)
172
173
      av1_cost_tokens(x->intra_tx_type_costs[i][j], fc->intra_ext_tx_prob[i][j],
                      av1_ext_tx_tree);
174
175
  }
  for (i = TX_4X4; i < EXT_TX_SIZES; ++i) {
176
    av1_cost_tokens(x->inter_tx_type_costs[i], fc->inter_ext_tx_prob[i],
Yaowu Xu's avatar
Yaowu Xu committed
177
                    av1_ext_tx_tree);
178
  }
179
#endif  // CONFIG_EXT_TX
180
#if CONFIG_EXT_INTRA
hui su's avatar
hui su committed
181
#if CONFIG_INTRA_INTERP
182
  for (i = 0; i < INTRA_FILTERS + 1; ++i)
183
    av1_cost_tokens(x->intra_filter_cost[i], fc->intra_filter_probs[i],
Yaowu Xu's avatar
Yaowu Xu committed
184
                    av1_intra_filter_tree);
hui su's avatar
hui su committed
185
#endif  // CONFIG_INTRA_INTERP
186
#endif  // CONFIG_EXT_INTRA
187
#if CONFIG_LOOP_RESTORATION
188
  av1_cost_tokens(x->switchable_restore_cost, fc->switchable_restore_prob,
189
190
                  av1_switchable_restore_tree);
#endif  // CONFIG_LOOP_RESTORATION
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
222
223
224
225
226
227
228
229
230

  if (!frame_is_intra_only(cm)) {
    for (i = 0; i < NEWMV_MODE_CONTEXTS; ++i) {
      x->newmv_mode_cost[i][0] = av1_cost_bit(fc->newmv_prob[i], 0);
      x->newmv_mode_cost[i][1] = av1_cost_bit(fc->newmv_prob[i], 1);
    }

    for (i = 0; i < ZEROMV_MODE_CONTEXTS; ++i) {
      x->zeromv_mode_cost[i][0] = av1_cost_bit(fc->zeromv_prob[i], 0);
      x->zeromv_mode_cost[i][1] = av1_cost_bit(fc->zeromv_prob[i], 1);
    }

    for (i = 0; i < REFMV_MODE_CONTEXTS; ++i) {
      x->refmv_mode_cost[i][0] = av1_cost_bit(fc->refmv_prob[i], 0);
      x->refmv_mode_cost[i][1] = av1_cost_bit(fc->refmv_prob[i], 1);
    }

    for (i = 0; i < DRL_MODE_CONTEXTS; ++i) {
      x->drl_mode_cost0[i][0] = av1_cost_bit(fc->drl_prob[i], 0);
      x->drl_mode_cost0[i][1] = av1_cost_bit(fc->drl_prob[i], 1);
    }
#if CONFIG_EXT_INTER
    for (i = 0; i < INTER_MODE_CONTEXTS; ++i)
      av1_cost_tokens(x->inter_compound_mode_cost[i],
                      fc->inter_compound_mode_probs[i],
                      av1_inter_compound_mode_tree);
#if CONFIG_COMPOUND_SINGLEREF
    for (i = 0; i < INTER_MODE_CONTEXTS; ++i)
      av1_cost_tokens(x->inter_singleref_comp_mode_cost[i],
                      fc->inter_singleref_comp_mode_probs[i],
                      av1_inter_singleref_comp_mode_tree);
#endif  // CONFIG_COMPOUND_SINGLEREF
#if CONFIG_INTERINTRA
    for (i = 0; i < BLOCK_SIZE_GROUPS; ++i)
      av1_cost_tokens(x->interintra_mode_cost[i], fc->interintra_mode_prob[i],
                      av1_interintra_mode_tree);
#endif  // CONFIG_INTERINTRA
#endif  // CONFIG_EXT_INTER
#if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
    for (i = BLOCK_8X8; i < BLOCK_SIZES_ALL; i++) {
231
232
      av1_cost_tokens_from_cdf(x->motion_mode_cost[i], fc->motion_mode_cdf[i],
                               NULL);
233
234
235
236
237
238
239
240
241
    }
#if CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
    for (i = BLOCK_8X8; i < BLOCK_SIZES_ALL; i++) {
      x->motion_mode_cost1[i][0] = av1_cost_bit(fc->obmc_prob[i], 0);
      x->motion_mode_cost1[i][1] = av1_cost_bit(fc->obmc_prob[i], 1);
    }
#endif  // CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
#if CONFIG_MOTION_VAR && CONFIG_NCOBMC_ADAPT_WEIGHT
    for (i = ADAPT_OVERLAP_BLOCK_8X8; i < ADAPT_OVERLAP_BLOCKS; ++i) {
242
243
      av1_cost_tokens_from_cdf(x->ncobmc_mode_cost[i], fc->ncobmc_mode_cdf[i],
                               NULL);
244
245
246
247
    }
#endif  // CONFIG_MOTION_VAR && CONFIG_NCOBMC_ADAPT_WEIGHT
#endif  // CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
  }
Jingning Han's avatar
Jingning Han committed
248
249
}

Yaowu Xu's avatar
Yaowu Xu committed
250
251
void av1_fill_token_costs(av1_coeff_cost *c,
                          av1_coeff_probs_model (*p)[PLANE_TYPES]) {
Jingning Han's avatar
Jingning Han committed
252
253
  int i, j, k, l;
  TX_SIZE t;
254
  for (t = 0; t < TX_SIZES; ++t)
Jingning Han's avatar
Jingning Han committed
255
256
257
258
    for (i = 0; i < PLANE_TYPES; ++i)
      for (j = 0; j < REF_TYPES; ++j)
        for (k = 0; k < COEF_BANDS; ++k)
          for (l = 0; l < BAND_COEFF_CONTEXTS(k); ++l) {
Yaowu Xu's avatar
Yaowu Xu committed
259
260
261
262
263
            aom_prob probs[ENTROPY_NODES];
            av1_model_to_full_probs(p[t][i][j][k][l], probs);
            av1_cost_tokens((int *)c[t][i][j][k][0][l], probs, av1_coef_tree);
            av1_cost_tokens_skip((int *)c[t][i][j][k][1][l], probs,
                                 av1_coef_tree);
Jingning Han's avatar
Jingning Han committed
264
265
266
267
268
269
270
271
272
            assert(c[t][i][j][k][0][l][EOB_TOKEN] ==
                   c[t][i][j][k][1][l][EOB_TOKEN]);
          }
}

// Values are now correlated to quantizer.
static int sad_per_bit16lut_8[QINDEX_RANGE];
static int sad_per_bit4lut_8[QINDEX_RANGE];

273
#if CONFIG_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
274
275
276
277
278
279
280
static int sad_per_bit16lut_10[QINDEX_RANGE];
static int sad_per_bit4lut_10[QINDEX_RANGE];
static int sad_per_bit16lut_12[QINDEX_RANGE];
static int sad_per_bit4lut_12[QINDEX_RANGE];
#endif

static void init_me_luts_bd(int *bit16lut, int *bit4lut, int range,
Yaowu Xu's avatar
Yaowu Xu committed
281
                            aom_bit_depth_t bit_depth) {
Jingning Han's avatar
Jingning Han committed
282
283
284
285
286
  int i;
  // Initialize the sad lut tables using a formulaic calculation for now.
  // This is to make it easier to resolve the impact of experimental changes
  // to the quantizer tables.
  for (i = 0; i < range; i++) {
Yaowu Xu's avatar
Yaowu Xu committed
287
    const double q = av1_convert_qindex_to_q(i, bit_depth);
Jingning Han's avatar
Jingning Han committed
288
289
290
291
292
    bit16lut[i] = (int)(0.0418 * q + 2.4107);
    bit4lut[i] = (int)(0.063 * q + 2.742);
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
293
void av1_init_me_luts(void) {
Jingning Han's avatar
Jingning Han committed
294
  init_me_luts_bd(sad_per_bit16lut_8, sad_per_bit4lut_8, QINDEX_RANGE,
Yaowu Xu's avatar
Yaowu Xu committed
295
                  AOM_BITS_8);
296
#if CONFIG_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
297
  init_me_luts_bd(sad_per_bit16lut_10, sad_per_bit4lut_10, QINDEX_RANGE,
Yaowu Xu's avatar
Yaowu Xu committed
298
                  AOM_BITS_10);
Jingning Han's avatar
Jingning Han committed
299
  init_me_luts_bd(sad_per_bit16lut_12, sad_per_bit4lut_12, QINDEX_RANGE,
Yaowu Xu's avatar
Yaowu Xu committed
300
                  AOM_BITS_12);
Jingning Han's avatar
Jingning Han committed
301
302
303
#endif
}

304
305
static const int rd_boost_factor[16] = { 64, 32, 32, 32, 24, 16, 12, 12,
                                         8,  8,  4,  4,  2,  2,  1,  0 };
Jingning Han's avatar
Jingning Han committed
306
static const int rd_frame_type_factor[FRAME_UPDATE_TYPES] = {
307
  128, 144, 128, 128, 144,
308
#if CONFIG_EXT_REFS
309
310
  // TODO(zoeliu): To adjust further following factor values.
  128, 128, 128
311
312
  // TODO(weitinglin): We should investigate if the values should be the same
  //                   as the value used by OVERLAY frame
Yaowu Xu's avatar
Yaowu Xu committed
313
  ,
Zoe Liu's avatar
Zoe Liu committed
314
315
316
317
318
  144  // INTNL_OVERLAY_UPDATE
#if CONFIG_ALTREF2
  ,
  128   // INTNL_ARF_UPDATE
#endif  // CONFIG_ALTREF2
319
#endif  // CONFIG_EXT_REFS
Jingning Han's avatar
Jingning Han committed
320
321
};

Yaowu Xu's avatar
Yaowu Xu committed
322
323
int av1_compute_rd_mult(const AV1_COMP *cpi, int qindex) {
  const int64_t q = av1_dc_quant(qindex, 0, cpi->common.bit_depth);
324
#if CONFIG_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
325
326
  int64_t rdmult = 0;
  switch (cpi->common.bit_depth) {
Yaowu Xu's avatar
Yaowu Xu committed
327
328
329
    case AOM_BITS_8: rdmult = 88 * q * q / 24; break;
    case AOM_BITS_10: rdmult = ROUND_POWER_OF_TWO(88 * q * q / 24, 4); break;
    case AOM_BITS_12: rdmult = ROUND_POWER_OF_TWO(88 * q * q / 24, 8); break;
Jingning Han's avatar
Jingning Han committed
330
    default:
Yaowu Xu's avatar
Yaowu Xu committed
331
      assert(0 && "bit_depth should be AOM_BITS_8, AOM_BITS_10 or AOM_BITS_12");
Jingning Han's avatar
Jingning Han committed
332
333
334
335
      return -1;
  }
#else
  int64_t rdmult = 88 * q * q / 24;
336
#endif  // CONFIG_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
337
338
339
  if (cpi->oxcf.pass == 2 && (cpi->common.frame_type != KEY_FRAME)) {
    const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
    const FRAME_UPDATE_TYPE frame_type = gf_group->update_type[gf_group->index];
Yaowu Xu's avatar
Yaowu Xu committed
340
    const int boost_index = AOMMIN(15, (cpi->rc.gfu_boost / 100));
Jingning Han's avatar
Jingning Han committed
341
342
343
344

    rdmult = (rdmult * rd_frame_type_factor[frame_type]) >> 7;
    rdmult += ((rdmult * rd_boost_factor[boost_index]) >> 7);
  }
345
  if (rdmult < 1) rdmult = 1;
Jingning Han's avatar
Jingning Han committed
346
347
348
  return (int)rdmult;
}

Yaowu Xu's avatar
Yaowu Xu committed
349
static int compute_rd_thresh_factor(int qindex, aom_bit_depth_t bit_depth) {
Jingning Han's avatar
Jingning Han committed
350
  double q;
351
#if CONFIG_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
352
  switch (bit_depth) {
Yaowu Xu's avatar
Yaowu Xu committed
353
354
355
    case AOM_BITS_8: q = av1_dc_quant(qindex, 0, AOM_BITS_8) / 4.0; break;
    case AOM_BITS_10: q = av1_dc_quant(qindex, 0, AOM_BITS_10) / 16.0; break;
    case AOM_BITS_12: q = av1_dc_quant(qindex, 0, AOM_BITS_12) / 64.0; break;
Jingning Han's avatar
Jingning Han committed
356
    default:
Yaowu Xu's avatar
Yaowu Xu committed
357
      assert(0 && "bit_depth should be AOM_BITS_8, AOM_BITS_10 or AOM_BITS_12");
Jingning Han's avatar
Jingning Han committed
358
359
360
      return -1;
  }
#else
361
  (void)bit_depth;
Yaowu Xu's avatar
Yaowu Xu committed
362
  q = av1_dc_quant(qindex, 0, AOM_BITS_8) / 4.0;
363
#endif  // CONFIG_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
364
  // TODO(debargha): Adjust the function below.
Yaowu Xu's avatar
Yaowu Xu committed
365
  return AOMMAX((int)(pow(q, RD_THRESH_POW) * 5.12), 8);
Jingning Han's avatar
Jingning Han committed
366
367
}

Yaowu Xu's avatar
Yaowu Xu committed
368
void av1_initialize_me_consts(const AV1_COMP *cpi, MACROBLOCK *x, int qindex) {
369
#if CONFIG_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
370
  switch (cpi->common.bit_depth) {
Yaowu Xu's avatar
Yaowu Xu committed
371
    case AOM_BITS_8:
Jingning Han's avatar
Jingning Han committed
372
373
374
      x->sadperbit16 = sad_per_bit16lut_8[qindex];
      x->sadperbit4 = sad_per_bit4lut_8[qindex];
      break;
Yaowu Xu's avatar
Yaowu Xu committed
375
    case AOM_BITS_10:
Jingning Han's avatar
Jingning Han committed
376
377
378
      x->sadperbit16 = sad_per_bit16lut_10[qindex];
      x->sadperbit4 = sad_per_bit4lut_10[qindex];
      break;
Yaowu Xu's avatar
Yaowu Xu committed
379
    case AOM_BITS_12:
Jingning Han's avatar
Jingning Han committed
380
381
382
383
      x->sadperbit16 = sad_per_bit16lut_12[qindex];
      x->sadperbit4 = sad_per_bit4lut_12[qindex];
      break;
    default:
Yaowu Xu's avatar
Yaowu Xu committed
384
      assert(0 && "bit_depth should be AOM_BITS_8, AOM_BITS_10 or AOM_BITS_12");
Jingning Han's avatar
Jingning Han committed
385
386
387
388
389
  }
#else
  (void)cpi;
  x->sadperbit16 = sad_per_bit16lut_8[qindex];
  x->sadperbit4 = sad_per_bit4lut_8[qindex];
390
#endif  // CONFIG_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
391
392
}

Yaowu Xu's avatar
Yaowu Xu committed
393
static void set_block_thresholds(const AV1_COMMON *cm, RD_OPT *rd) {
Jingning Han's avatar
Jingning Han committed
394
395
396
397
  int i, bsize, segment_id;

  for (segment_id = 0; segment_id < MAX_SEGMENTS; ++segment_id) {
    const int qindex =
Yaowu Xu's avatar
Yaowu Xu committed
398
        clamp(av1_get_qindex(&cm->seg, segment_id, cm->base_qindex) +
399
400
                  cm->y_dc_delta_q,
              0, MAXQ);
Jingning Han's avatar
Jingning Han committed
401
402
    const int q = compute_rd_thresh_factor(qindex, cm->bit_depth);

403
    for (bsize = 0; bsize < BLOCK_SIZES_ALL; ++bsize) {
Jingning Han's avatar
Jingning Han committed
404
405
406
407
408
      // Threshold here seems unnecessarily harsh but fine given actual
      // range of values used for cpi->sf.thresh_mult[].
      const int t = q * rd_thresh_block_size_factor[bsize];
      const int thresh_max = INT_MAX / t;

409
410
411
412
413
414
#if CONFIG_CB4X4
      for (i = 0; i < MAX_MODES; ++i)
        rd->threshes[segment_id][bsize][i] = rd->thresh_mult[i] < thresh_max
                                                 ? rd->thresh_mult[i] * t / 4
                                                 : INT_MAX;
#else
Jingning Han's avatar
Jingning Han committed
415
416
      if (bsize >= BLOCK_8X8) {
        for (i = 0; i < MAX_MODES; ++i)
417
418
419
          rd->threshes[segment_id][bsize][i] = rd->thresh_mult[i] < thresh_max
                                                   ? rd->thresh_mult[i] * t / 4
                                                   : INT_MAX;
Jingning Han's avatar
Jingning Han committed
420
421
422
423
424
425
426
      } else {
        for (i = 0; i < MAX_REFS; ++i)
          rd->threshes[segment_id][bsize][i] =
              rd->thresh_mult_sub8x8[i] < thresh_max
                  ? rd->thresh_mult_sub8x8[i] * t / 4
                  : INT_MAX;
      }
427
#endif
Jingning Han's avatar
Jingning Han committed
428
429
430
431
    }
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
432
433
void av1_set_mvcost(MACROBLOCK *x, MV_REFERENCE_FRAME ref_frame, int ref,
                    int ref_mv_idx) {
434
  MB_MODE_INFO_EXT *mbmi_ext = x->mbmi_ext;
Yaowu Xu's avatar
Yaowu Xu committed
435
436
437
438
  int8_t rf_type = av1_ref_frame_type(x->e_mbd.mi[0]->mbmi.ref_frame);
  int nmv_ctx = av1_nmv_ctx(mbmi_ext->ref_mv_count[rf_type],
                            mbmi_ext->ref_mv_stack[rf_type], ref, ref_mv_idx);
  (void)ref_frame;
439
440
441
442
  x->mvcost = x->mv_cost_stack[nmv_ctx];
  x->nmvjointcost = x->nmv_vec_cost[nmv_ctx];
}

Yaowu Xu's avatar
Yaowu Xu committed
443
444
void av1_initialize_rd_consts(AV1_COMP *cpi) {
  AV1_COMMON *const cm = &cpi->common;
Jingning Han's avatar
Jingning Han committed
445
446
  MACROBLOCK *const x = &cpi->td.mb;
  RD_OPT *const rd = &cpi->rd;
Jingning Han's avatar
Jingning Han committed
447
  int nmv_ctx;
Jingning Han's avatar
Jingning Han committed
448

Yaowu Xu's avatar
Yaowu Xu committed
449
  aom_clear_system_state();
Jingning Han's avatar
Jingning Han committed
450

Yaowu Xu's avatar
Yaowu Xu committed
451
  rd->RDMULT = av1_compute_rd_mult(cpi, cm->base_qindex + cm->y_dc_delta_q);
Jingning Han's avatar
Jingning Han committed
452

453
  set_error_per_bit(x, rd->RDMULT);
Jingning Han's avatar
Jingning Han committed
454
455
456

  set_block_thresholds(cm, rd);

Jingning Han's avatar
Jingning Han committed
457
  for (nmv_ctx = 0; nmv_ctx < NMV_CONTEXTS; ++nmv_ctx) {
Yaowu Xu's avatar
Yaowu Xu committed
458
    av1_build_nmv_cost_table(
Jingning Han's avatar
Jingning Han committed
459
460
461
462
        x->nmv_vec_cost[nmv_ctx],
        cm->allow_high_precision_mv ? x->nmvcost_hp[nmv_ctx]
                                    : x->nmvcost[nmv_ctx],
        &cm->fc->nmvc[nmv_ctx], cm->allow_high_precision_mv);
463
  }
Jingning Han's avatar
Jingning Han committed
464
465
466
  x->mvcost = x->mv_cost_stack[0];
  x->nmvjointcost = x->nmv_vec_cost[0];

Alex Converse's avatar
Alex Converse committed
467
468
469
470
471
472
473
474
475
476
#if CONFIG_INTRABC
  if (frame_is_intra_only(cm) && cm->allow_screen_content_tools &&
      cpi->oxcf.pass != 1) {
    av1_build_nmv_cost_table(
        x->nmv_vec_cost[0],
        cm->allow_high_precision_mv ? x->nmvcost_hp[0] : x->nmvcost[0],
        &cm->fc->ndvc, MV_SUBPEL_NONE);
  }
#endif

477
  if (cpi->oxcf.pass != 1) {
478
    av1_fill_token_costs(x->token_costs, cm->fc->coef_probs);
479
#if CONFIG_GLOBAL_MOTION
480
    for (int i = 0; i < TRANS_TYPES; ++i)
481
482
483
      cpi->gmtype_cost[i] = (1 + (i > 0 ? GLOBAL_TYPE_BITS : 0))
                            << AV1_PROB_COST_SHIFT;
#endif  // CONFIG_GLOBAL_MOTION
Jingning Han's avatar
Jingning Han committed
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
  }
}

static void model_rd_norm(int xsq_q10, int *r_q10, int *d_q10) {
  // NOTE: The tables below must be of the same size.

  // The functions described below are sampled at the four most significant
  // bits of x^2 + 8 / 256.

  // Normalized rate:
  // This table models the rate for a Laplacian source with given variance
  // when quantized with a uniform quantizer with given stepsize. The
  // closed form expression is:
  // Rn(x) = H(sqrt(r)) + sqrt(r)*[1 + H(r)/(1 - r)],
  // where r = exp(-sqrt(2) * x) and x = qpstep / sqrt(variance),
  // and H(x) is the binary entropy function.
  static const int rate_tab_q10[] = {
501
502
503
504
505
506
507
508
509
    65536, 6086, 5574, 5275, 5063, 4899, 4764, 4651, 4553, 4389, 4255, 4142,
    4044,  3958, 3881, 3811, 3748, 3635, 3538, 3453, 3376, 3307, 3244, 3186,
    3133,  3037, 2952, 2877, 2809, 2747, 2690, 2638, 2589, 2501, 2423, 2353,
    2290,  2232, 2179, 2130, 2084, 2001, 1928, 1862, 1802, 1748, 1698, 1651,
    1608,  1530, 1460, 1398, 1342, 1290, 1243, 1199, 1159, 1086, 1021, 963,
    911,   864,  821,  781,  745,  680,  623,  574,  530,  490,  455,  424,
    395,   345,  304,  269,  239,  213,  190,  171,  154,  126,  104,  87,
    73,    61,   52,   44,   38,   28,   21,   16,   12,   10,   8,    6,
    5,     3,    2,    1,    1,    1,    0,    0,
Jingning Han's avatar
Jingning Han committed
510
511
512
513
514
515
516
517
518
  };
  // Normalized distortion:
  // This table models the normalized distortion for a Laplacian source
  // with given variance when quantized with a uniform quantizer
  // with given stepsize. The closed form expression is:
  // Dn(x) = 1 - 1/sqrt(2) * x / sinh(x/sqrt(2))
  // where x = qpstep / sqrt(variance).
  // Note the actual distortion is Dn * variance.
  static const int dist_tab_q10[] = {
519
520
521
522
523
524
525
526
527
    0,    0,    1,    1,    1,    2,    2,    2,    3,    3,    4,    5,
    5,    6,    7,    7,    8,    9,    11,   12,   13,   15,   16,   17,
    18,   21,   24,   26,   29,   31,   34,   36,   39,   44,   49,   54,
    59,   64,   69,   73,   78,   88,   97,   106,  115,  124,  133,  142,
    151,  167,  184,  200,  215,  231,  245,  260,  274,  301,  327,  351,
    375,  397,  418,  439,  458,  495,  528,  559,  587,  613,  637,  659,
    680,  717,  749,  777,  801,  823,  842,  859,  874,  899,  919,  936,
    949,  960,  969,  977,  983,  994,  1001, 1006, 1010, 1013, 1015, 1017,
    1018, 1020, 1022, 1022, 1023, 1023, 1023, 1024,
Jingning Han's avatar
Jingning Han committed
528
529
  };
  static const int xsq_iq_q10[] = {
530
531
532
533
534
535
536
537
538
539
540
541
    0,      4,      8,      12,     16,     20,     24,     28,     32,
    40,     48,     56,     64,     72,     80,     88,     96,     112,
    128,    144,    160,    176,    192,    208,    224,    256,    288,
    320,    352,    384,    416,    448,    480,    544,    608,    672,
    736,    800,    864,    928,    992,    1120,   1248,   1376,   1504,
    1632,   1760,   1888,   2016,   2272,   2528,   2784,   3040,   3296,
    3552,   3808,   4064,   4576,   5088,   5600,   6112,   6624,   7136,
    7648,   8160,   9184,   10208,  11232,  12256,  13280,  14304,  15328,
    16352,  18400,  20448,  22496,  24544,  26592,  28640,  30688,  32736,
    36832,  40928,  45024,  49120,  53216,  57312,  61408,  65504,  73696,
    81888,  90080,  98272,  106464, 114656, 122848, 131040, 147424, 163808,
    180192, 196576, 212960, 229344, 245728,
Jingning Han's avatar
Jingning Han committed
542
543
544
545
546
547
548
549
550
551
552
  };
  const int tmp = (xsq_q10 >> 2) + 8;
  const int k = get_msb(tmp) - 3;
  const int xq = (k << 3) + ((tmp >> k) & 0x7);
  const int one_q10 = 1 << 10;
  const int a_q10 = ((xsq_q10 - xsq_iq_q10[xq]) << 10) >> (2 + k);
  const int b_q10 = one_q10 - a_q10;
  *r_q10 = (rate_tab_q10[xq] * b_q10 + rate_tab_q10[xq + 1] * a_q10) >> 10;
  *d_q10 = (dist_tab_q10[xq] * b_q10 + dist_tab_q10[xq + 1] * a_q10) >> 10;
}

Yaowu Xu's avatar
Yaowu Xu committed
553
554
555
void av1_model_rd_from_var_lapndz(int64_t var, unsigned int n_log2,
                                  unsigned int qstep, int *rate,
                                  int64_t *dist) {
Jingning Han's avatar
Jingning Han committed
556
557
558
559
560
561
562
563
564
565
566
567
568
569
  // This function models the rate and distortion for a Laplacian
  // source with given variance when quantized with a uniform quantizer
  // with given stepsize. The closed form expressions are in:
  // Hang and Chen, "Source Model for transform video coder and its
  // application - Part I: Fundamental Theory", IEEE Trans. Circ.
  // Sys. for Video Tech., April 1997.
  if (var == 0) {
    *rate = 0;
    *dist = 0;
  } else {
    int d_q10, r_q10;
    static const uint32_t MAX_XSQ_Q10 = 245727;
    const uint64_t xsq_q10_64 =
        (((uint64_t)qstep * qstep << (n_log2 + 10)) + (var >> 1)) / var;
Yaowu Xu's avatar
Yaowu Xu committed
570
    const int xsq_q10 = (int)AOMMIN(xsq_q10_64, MAX_XSQ_Q10);
Jingning Han's avatar
Jingning Han committed
571
    model_rd_norm(xsq_q10, &r_q10, &d_q10);
Yaowu Xu's avatar
Yaowu Xu committed
572
    *rate = ROUND_POWER_OF_TWO(r_q10 << n_log2, 10 - AV1_PROB_COST_SHIFT);
Jingning Han's avatar
Jingning Han committed
573
574
575
576
    *dist = (var * (int64_t)d_q10 + 512) >> 10;
  }
}

577
578
579
580
static void get_entropy_contexts_plane(
    BLOCK_SIZE plane_bsize, TX_SIZE tx_size, const struct macroblockd_plane *pd,
    ENTROPY_CONTEXT t_above[2 * MAX_MIB_SIZE],
    ENTROPY_CONTEXT t_left[2 * MAX_MIB_SIZE]) {
581
582
  const int num_4x4_w = block_size_wide[plane_bsize] >> tx_size_wide_log2[0];
  const int num_4x4_h = block_size_high[plane_bsize] >> tx_size_high_log2[0];
Jingning Han's avatar
Jingning Han committed
583
584
585
  const ENTROPY_CONTEXT *const above = pd->above_context;
  const ENTROPY_CONTEXT *const left = pd->left_context;

586
587
588
589
590
591
#if CONFIG_LV_MAP
  memcpy(t_above, above, sizeof(ENTROPY_CONTEXT) * num_4x4_w);
  memcpy(t_left, left, sizeof(ENTROPY_CONTEXT) * num_4x4_h);
  return;
#endif  // CONFIG_LV_MAP

Jingning Han's avatar
Jingning Han committed
592
  int i;
593

594
#if CONFIG_CHROMA_2X2
595
  switch (tx_size) {
596
    case TX_2X2:
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
      memcpy(t_above, above, sizeof(ENTROPY_CONTEXT) * num_4x4_w);
      memcpy(t_left, left, sizeof(ENTROPY_CONTEXT) * num_4x4_h);
      break;
    case TX_4X4:
      for (i = 0; i < num_4x4_w; i += 2)
        t_above[i] = !!*(const uint16_t *)&above[i];
      for (i = 0; i < num_4x4_h; i += 2)
        t_left[i] = !!*(const uint16_t *)&left[i];
      break;
    case TX_8X8:
      for (i = 0; i < num_4x4_w; i += 4)
        t_above[i] = !!*(const uint32_t *)&above[i];
      for (i = 0; i < num_4x4_h; i += 4)
        t_left[i] = !!*(const uint32_t *)&left[i];
      break;
    case TX_16X16:
      for (i = 0; i < num_4x4_w; i += 8)
        t_above[i] = !!*(const uint64_t *)&above[i];
      for (i = 0; i < num_4x4_h; i += 8)
        t_left[i] = !!*(const uint64_t *)&left[i];
      break;
    case TX_32X32:
      for (i = 0; i < num_4x4_w; i += 16)
        t_above[i] =
            !!(*(const uint64_t *)&above[i] | *(const uint64_t *)&above[i + 8]);
      for (i = 0; i < num_4x4_h; i += 16)
        t_left[i] =
            !!(*(const uint64_t *)&left[i] | *(const uint64_t *)&left[i + 8]);
      break;
626
627
628
629
630
631
632
633
634
635
636
637
638
639
#if CONFIG_TX64X64
    case TX_64X64:
      for (i = 0; i < num_4x4_w; i += 32)
        t_above[i] =
            !!(*(const uint64_t *)&above[i] | *(const uint64_t *)&above[i + 8] |
               *(const uint64_t *)&above[i + 16] |
               *(const uint64_t *)&above[i + 24]);
      for (i = 0; i < num_4x4_h; i += 32)
        t_left[i] =
            !!(*(const uint64_t *)&left[i] | *(const uint64_t *)&left[i + 8] |
               *(const uint64_t *)&left[i + 16] |
               *(const uint64_t *)&left[i + 24]);
      break;
#endif  // CONFIG_TX64X64
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
    case TX_4X8:
      for (i = 0; i < num_4x4_w; i += 2)
        t_above[i] = !!*(const uint16_t *)&above[i];
      for (i = 0; i < num_4x4_h; i += 4)
        t_left[i] = !!*(const uint32_t *)&left[i];
      break;
    case TX_8X4:
      for (i = 0; i < num_4x4_w; i += 4)
        t_above[i] = !!*(const uint32_t *)&above[i];
      for (i = 0; i < num_4x4_h; i += 2)
        t_left[i] = !!*(const uint16_t *)&left[i];
      break;
    case TX_8X16:
      for (i = 0; i < num_4x4_w; i += 4)
        t_above[i] = !!*(const uint32_t *)&above[i];
      for (i = 0; i < num_4x4_h; i += 8)
        t_left[i] = !!*(const uint64_t *)&left[i];
      break;
    case TX_16X8:
      for (i = 0; i < num_4x4_w; i += 8)
        t_above[i] = !!*(const uint64_t *)&above[i];
      for (i = 0; i < num_4x4_h; i += 4)
        t_left[i] = !!*(const uint32_t *)&left[i];
      break;
    case TX_16X32:
      for (i = 0; i < num_4x4_w; i += 8)
        t_above[i] = !!*(const uint64_t *)&above[i];
      for (i = 0; i < num_4x4_h; i += 16)
        t_left[i] =
            !!(*(const uint64_t *)&left[i] | *(const uint64_t *)&left[i + 8]);
      break;
    case TX_32X16:
      for (i = 0; i < num_4x4_w; i += 16)
        t_above[i] =
            !!(*(const uint64_t *)&above[i] | *(const uint64_t *)&above[i + 8]);
      for (i = 0; i < num_4x4_h; i += 8)
        t_left[i] = !!*(const uint64_t *)&left[i];
677
      break;
Yue Chen's avatar
Yue Chen committed
678
#if CONFIG_RECT_TX_EXT && (CONFIG_EXT_TX || CONFIG_VAR_TX)
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
    case TX_4X16:
      for (i = 0; i < num_4x4_w; i += 2)
        t_above[i] = !!*(const uint16_t *)&above[i];
      for (i = 0; i < num_4x4_h; i += 8)
        t_left[i] = !!*(const uint64_t *)&left[i];
      break;
    case TX_16X4:
      for (i = 0; i < num_4x4_w; i += 8)
        t_above[i] = !!*(const uint64_t *)&above[i];
      for (i = 0; i < num_4x4_h; i += 2)
        t_left[i] = !!*(const uint16_t *)&left[i];
      break;
    case TX_8X32:
      for (i = 0; i < num_4x4_w; i += 4)
        t_above[i] = !!*(const uint32_t *)&above[i];
      for (i = 0; i < num_4x4_h; i += 16)
        t_left[i] =
            !!(*(const uint64_t *)&left[i] | *(const uint64_t *)&left[i + 8]);
      break;
    case TX_32X8:
      for (i = 0; i < num_4x4_w; i += 16)
        t_above[i] =
            !!(*(const uint64_t *)&above[i] | *(const uint64_t *)&above[i + 8]);
      for (i = 0; i < num_4x4_h; i += 4)
        t_left[i] = !!*(const uint32_t *)&left[i];
      break;
Yue Chen's avatar
Yue Chen committed
705
#endif
706
707
708
709

    default: assert(0 && "Invalid transform size."); break;
  }
  return;
710
#endif  // CONFIG_CHROMA_2X2
711
712

  switch (tx_size) {
Jingning Han's avatar
Jingning Han committed
713
714
715
716
    case TX_4X4:
      memcpy(t_above, above, sizeof(ENTROPY_CONTEXT) * num_4x4_w);
      memcpy(t_left, left, sizeof(ENTROPY_CONTEXT) * num_4x4_h);
      break;
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
    case TX_8X8:
      for (i = 0; i < num_4x4_w; i += 2)
        t_above[i] = !!*(const uint16_t *)&above[i];
      for (i = 0; i < num_4x4_h; i += 2)
        t_left[i] = !!*(const uint16_t *)&left[i];
      break;
    case TX_16X16:
      for (i = 0; i < num_4x4_w; i += 4)
        t_above[i] = !!*(const uint32_t *)&above[i];
      for (i = 0; i < num_4x4_h; i += 4)
        t_left[i] = !!*(const uint32_t *)&left[i];
      break;
    case TX_32X32:
      for (i = 0; i < num_4x4_w; i += 8)
        t_above[i] = !!*(const uint64_t *)&above[i];
      for (i = 0; i < num_4x4_h; i += 8)
        t_left[i] = !!*(const uint64_t *)&left[i];
      break;
735
736
737
738
739
740
#if CONFIG_TX64X64
    case TX_64X64:
      for (i = 0; i < num_4x4_w; i += 16)
        t_above[i] =
            !!(*(const uint64_t *)&above[i] | *(const uint64_t *)&above[i + 8]);
      for (i = 0; i < num_4x4_h; i += 16)
741
742
        t_left[i] =
            !!(*(const uint64_t *)&left[i] | *(const uint64_t *)&left[i + 8]);
743
744
      break;
#endif  // CONFIG_TX64X64
745
746
747
748
749
750
751
752
753
754
    case TX_4X8:
      memcpy(t_above, above, sizeof(ENTROPY_CONTEXT) * num_4x4_w);
      for (i = 0; i < num_4x4_h; i += 2)
        t_left[i] = !!*(const uint16_t *)&left[i];
      break;
    case TX_8X4:
      for (i = 0; i < num_4x4_w; i += 2)
        t_above[i] = !!*(const uint16_t *)&above[i];
      memcpy(t_left, left, sizeof(ENTROPY_CONTEXT) * num_4x4_h);
      break;
755
    case TX_8X16:
Jingning Han's avatar
Jingning Han committed
756
757
      for (i = 0; i < num_4x4_w; i += 2)
        t_above[i] = !!*(const uint16_t *)&above[i];
758
759
760
761
762
763
      for (i = 0; i < num_4x4_h; i += 4)
        t_left[i] = !!*(const uint32_t *)&left[i];
      break;
    case TX_16X8:
      for (i = 0; i < num_4x4_w; i += 4)
        t_above[i] = !!*(const uint32_t *)&above[i];
Jingning Han's avatar
Jingning Han committed
764
765
766
      for (i = 0; i < num_4x4_h; i += 2)
        t_left[i] = !!*(const uint16_t *)&left[i];
      break;
767
    case TX_16X32:
Jingning Han's avatar
Jingning Han committed
768
769
      for (i = 0; i < num_4x4_w; i += 4)
        t_above[i] = !!*(const uint32_t *)&above[i];
770
771
      for (i = 0; i < num_4x4_h; i += 8)
        t_left[i] = !!*(const uint64_t *)&left[i];
Jingning Han's avatar
Jingning Han committed
772
      break;
773
    case TX_32X16:
Jingning Han's avatar
Jingning Han committed
774
775
      for (i = 0; i < num_4x4_w; i += 8)
        t_above[i] = !!*(const uint64_t *)&above[i];
776
777
778
      for (i = 0; i < num_4x4_h; i += 4)
        t_left[i] = !!*(const uint32_t *)&left[i];
      break;
Yue Chen's avatar
Yue Chen committed
779
#if CONFIG_RECT_TX_EXT && (CONFIG_EXT_TX || CONFIG_VAR_TX)
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
    case TX_4X16:
      memcpy(t_above, above, sizeof(ENTROPY_CONTEXT) * num_4x4_w);
      for (i = 0; i < num_4x4_h; i += 4)
        t_left[i] = !!*(const uint32_t *)&left[i];
      break;
    case TX_16X4:
      for (i = 0; i < num_4x4_w; i += 4)
        t_above[i] = !!*(const uint32_t *)&above[i];
      memcpy(t_left, left, sizeof(ENTROPY_CONTEXT) * num_4x4_h);
      break;
    case TX_8X32:
      for (i = 0; i < num_4x4_w; i += 2)
        t_above[i] = !!*(const uint16_t *)&above[i];
      for (i = 0; i < num_4x4_h; i += 8)
        t_left[i] = !!*(const uint64_t *)&left[i];
      break;
    case TX_32X8:
      for (i = 0; i < num_4x4_w; i += 8)
        t_above[i] = !!*(const uint64_t *)&above[i];
      for (i = 0; i < num_4x4_h; i += 2)
        t_left[i] = !!*(const uint16_t *)&left[i];
      break;
Yue Chen's avatar
Yue Chen committed
802
#endif
clang-format's avatar
clang-format committed
803
    default: assert(0 && "Invalid transform size."); break;
Jingning Han's avatar
Jingning Han committed
804
805
806
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
807
808
809
810
void av1_get_entropy_contexts(BLOCK_SIZE bsize, TX_SIZE tx_size,
                              const struct macroblockd_plane *pd,
                              ENTROPY_CONTEXT t_above[2 * MAX_MIB_SIZE],
                              ENTROPY_CONTEXT t_left[2 * MAX_MIB_SIZE]) {
811
#if CONFIG_CHROMA_SUB8X8
812
813
814
  const BLOCK_SIZE plane_bsize =
      AOMMAX(BLOCK_4X4, get_plane_block_size(bsize, pd));
#else
815
  const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
816
#endif
817
  get_entropy_contexts_plane(plane_bsize, tx_size, pd, t_above, t_left);
818
819
}

820
void av1_mv_pred(const AV1_COMP *cpi, MACROBLOCK *x, uint8_t *ref_y_buffer,
Yaowu Xu's avatar
Yaowu Xu committed
821
                 int ref_y_stride, int ref_frame, BLOCK_SIZE block_size) {
Jingning Han's avatar
Jingning Han committed
822
823
824
825
826
827
828
829
  int i;
  int zero_seen = 0;
  int best_index = 0;
  int best_sad = INT_MAX;
  int this_sad = INT_MAX;
  int max_mv = 0;
  uint8_t *src_y_ptr = x->plane[0].src.buf;
  uint8_t *ref_y_ptr;
830
  MV pred_mv[MAX_MV_REF_CANDIDATES + 1];
831
832
833
834
835
836
  int num_mv_refs = 0;

  pred_mv[num_mv_refs++] = x->mbmi_ext->ref_mvs[ref_frame][0].as_mv;
  if (x->mbmi_ext->ref_mvs[ref_frame][0].as_int !=
      x->mbmi_ext->ref_mvs[ref_frame][1].as_int) {
    pred_mv[num_mv_refs++] = x->mbmi_ext->ref_mvs[ref_frame][1].as_mv;
837
  }
838
839
840
  if (cpi->sf.adaptive_motion_search && block_size < x->max_partition_size)
    pred_mv[num_mv_refs++] = x->pred_mv[ref_frame];

Jingning Han's avatar
Jingning Han committed
841
  assert(num_mv_refs <= (int)(sizeof(pred_mv) / sizeof(pred_mv[0])));
842

Jingning Han's avatar
Jingning Han committed
843
844
845
846
847
848
  // Get the sad for each candidate reference mv.
  for (i = 0; i < num_mv_refs; ++i) {
    const MV *this_mv = &pred_mv[i];
    int fp_row, fp_col;
    fp_row = (this_mv->row + 3 + (this_mv->row >= 0)) >> 3;
    fp_col = (this_mv->col + 3 + (this_mv->col >= 0)) >> 3;
Yaowu Xu's avatar
Yaowu Xu committed
849
    max_mv = AOMMAX(max_mv, AOMMAX(abs(this_mv->row), abs(this_mv->col)) >> 3);
Jingning Han's avatar
Jingning Han committed
850

851
852
    if (fp_row == 0 && fp_col == 0 && zero_seen) continue;
    zero_seen |= (fp_row == 0 && fp_col == 0);
Jingning Han's avatar
Jingning Han committed
853

854
    ref_y_ptr = &ref_y_buffer[ref_y_stride * fp_row + fp_col];
Jingning Han's avatar
Jingning Han committed
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
    // Find sad for current vector.
    this_sad = cpi->fn_ptr[block_size].sdf(src_y_ptr, x->plane[0].src.stride,
                                           ref_y_ptr, ref_y_stride);
    // Note if it is the best so far.
    if (this_sad < best_sad) {
      best_sad = this_sad;
      best_index = i;
    }
  }

  // Note the index of the mv that worked best in the reference list.
  x->mv_best_ref_index[ref_frame] = best_index;
  x->max_mv_context[ref_frame] = max_mv;
  x->pred_mv_sad[ref_frame] = best_sad;
}

Yaowu Xu's avatar
Yaowu Xu committed
871
872
873
874
875
void av1_setup_pred_block(const MACROBLOCKD *xd,
                          struct buf_2d dst[MAX_MB_PLANE],
                          const YV12_BUFFER_CONFIG *src, int mi_row, int mi_col,
                          const struct scale_factors *scale,
                          const struct scale_factors *scale_uv) {
Jingning Han's avatar
Jingning Han committed
876
877
878
879
880
881
882
883
884
  int i;

  dst[0].buf = src->y_buffer;
  dst[0].stride = src->y_stride;
  dst[1].buf = src->u_buffer;
  dst[2].buf = src->v_buffer;
  dst[1].stride = dst[2].stride = src->uv_stride;

  for (i = 0; i < MAX_MB_PLANE; ++i) {
885
    setup_pred_plane(dst + i, xd->mi[0]->mbmi.sb_type, dst[i].buf,
886
887
                     i ? src->uv_crop_width : src->y_crop_width,
                     i ? src->uv_crop_height : src->y_crop_height,
888
                     dst[i].stride, mi_row, mi_col, i ? scale_uv : scale,
Jingning Han's avatar
Jingning Han committed
889
890
891
892
                     xd->plane[i].subsampling_x, xd->plane[i].subsampling_y);
  }
}

Yaowu Xu's avatar
Yaowu Xu committed
893
894
int av1_raster_block_offset(BLOCK_SIZE plane_bsize, int raster_block,
                            int stride) {
Jingning Han's avatar
Jingning Han committed
895
896
897
898
899
900
  const int bw = b_width_log2_lookup[plane_bsize];
  const int y = 4 * (raster_block >> bw);
  const int x = 4 * (raster_block & ((1 << bw) - 1));
  return y * stride + x;
}

Yaowu Xu's avatar
Yaowu Xu committed
901
902
int16_t *av1_raster_block_offset_int16(BLOCK_SIZE plane_bsize, int raster_block,
                                       int16_t *base) {
903
  const int stride = block_size_wide[plane_bsize];
Yaowu Xu's avatar
Yaowu Xu committed
904
  return base + av1_raster_block_offset(plane_bsize, raster_block, stride);
Jingning Han's avatar
Jingning Han committed
905
906
}

Yaowu Xu's avatar
Yaowu Xu committed
907
908
909
YV12_BUFFER_CONFIG *av1_get_scaled_ref_frame(const AV1_COMP *cpi,
                                             int ref_frame) {
  const AV1_COMMON *const cm = &cpi->common;
Jingning Han's avatar
Jingning Han committed
910
911
  const int scaled_idx = cpi->scaled_ref_idx[ref_frame - 1];
  const int ref_idx = get_ref_frame_buf_idx(cpi, ref_frame);
912
913
914
  return (scaled_idx != ref_idx && scaled_idx != INVALID_IDX)
             ? &cm->buffer_pool->frame_bufs[scaled_idx].buf
             : NULL;
Jingning Han's avatar
Jingning Han committed
915
916
}

917
#if CONFIG_DUAL_FILTER
918
919
int av1_get_switchable_rate(const AV1_COMMON *const cm, MACROBLOCK *x,
                            const MACROBLOCKD *xd) {
920
921
922
923
924
925
926
927
928
929
930
  if (cm->interp_filter == SWITCHABLE) {
    const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
    int inter_filter_cost = 0;
    int dir;

    for (dir = 0; dir < 2; ++dir) {
      if (has_subpel_mv_component(xd->mi[0], xd, dir) ||
          (mbmi->ref_frame[1] > INTRA_FRAME &&
           has_subpel_mv_component(xd->mi[0], xd, dir + 2))) {
        const int ctx = av1_get_pred_context_switchable_interp(xd, dir);
        inter_filter_cost +=
931
            x->switchable_interp_costs[ctx][mbmi->interp_filter[dir]];
932
      }
933
    }
934
935
936
    return SWITCHABLE_INTERP_RATE_FACTOR * inter_filter_cost;
  } else {
    return 0;
937
938
939
  }
}
#else
940
941
int av1_get_switchable_rate(const AV1_COMMON *const cm, MACROBLOCK *x,
                            const MACROBLOCKD *xd) {
942
  if (cm->interp_filter == SWITCHABLE) {
943
944
945
    const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
    const int ctx = av1_get_pred_context_switchable_interp(xd);
    return SWITCHABLE_INTERP_RATE_FACTOR *
946
           x->switchable_interp_costs[ctx][mbmi->interp_filter];
947
948
  }
  return 0;
Jingning Han's avatar
Jingning Han committed
949
}
950
#endif
Jingning Han's avatar
Jingning Han committed
951

Yaowu Xu's avatar
Yaowu Xu committed
952
void av1_set_rd_speed_thresholds(AV1_COMP *cpi) {
Jingning Han's avatar
Jingning Han committed
953
954
955
956
957
  int i;
  RD_OPT *const rd = &cpi->rd;
  SPEED_FEATURES *const sf = &cpi->sf;

  // Set baseline threshold values.
958
  for (i = 0; i < MAX_MODES; ++i) rd->thresh_mult[i] = cpi->oxcf.mode == 0;
Jingning Han's avatar
Jingning Han committed
959
960
961

  if (sf->adaptive_rd_thresh) {
    rd->thresh_mult[THR_NEARESTMV] = 300;
962
963
964
#if CONFIG_EXT_REFS
    rd->thresh_mult[THR_NEARESTL2] = 300;
    rd->thresh_mult[THR_NEARESTL3] = 300;
965
    rd->thresh_mult[THR_NEARESTB] = 300;
Zoe Liu's avatar
Zoe Liu committed
966
967
968
#if CONFIG_ALTREF2
    rd->thresh_mult[THR_NEARESTA2] = 300;
#endif  // CONFIG_ALTREF2
969
#endif  // CONFIG_EXT_REFS
Jingning Han's avatar
Jingning Han committed
970
    rd->thresh_mult[THR_NEARESTA] = 300;
971
    rd->thresh_mult[THR_NEARESTG] = 300;
Jingning Han's avatar
Jingning Han committed
972
973
  } else {
    rd->thresh_mult[THR_NEARESTMV] = 0;
974
975
976
#if CONFIG_EXT_REFS
    rd->thresh_mult[THR_NEARESTL2] = 0;
    rd->thresh_mult[THR_NEARESTL3] = 0;
977
    rd->thresh_mult[THR_NEARESTB] = 0;
Zoe Liu's avatar
Zoe Liu committed
978
979
980
#if CONFIG_ALTREF2
    rd->thresh_mult[THR_NEARESTA2] = 0;
#endif  // CONFIG_ALTREF2
981
#endif  // CONFIG_EXT_REFS
Jingning Han's avatar
Jingning Han committed
982
    rd->thresh_mult[THR_NEARESTA] = 0;
983
    rd->thresh_mult[THR_NEARESTG] = 0;
Jingning Han's avatar
Jingning Han committed
984
985
986
987
988
  }

  rd->thresh_mult[THR_DC] += 1000;

  rd->thresh_mult[THR_NEWMV] += 1000;
989
990
991
#if CONFIG_EXT_REFS
  rd->thresh_mult[THR_NEWL2] += 1000;
  rd->thresh_mult[THR_NEWL3] += 1000;
992
  rd->thresh_mult[THR_NEWB] += 1000;
Zoe Liu's avatar
Zoe Liu committed
993
994
995
#if CONFIG_ALTREF2
  rd->thresh_mult[THR_NEWA2] = 1000;
#endif  // CONFIG_ALTREF2
996
#endif  // CONFIG_EXT_REFS
Jingning Han's avatar
Jingning Han committed
997
998
999
1000
  rd->thresh_mult[THR_NEWA] += 1000;
  rd->thresh_mult[THR_NEWG] += 1000;

  rd->thresh_mult[THR_NEARMV] += 1000;
1001
1002
1003
#if CONFIG_EXT_REFS
  rd->thresh_mult[THR_NEARL2] += 1000;
  rd->thresh_mult[THR_NEARL3] += 1000;
1004
  rd->thresh_mult[THR_NEARB] += 1000;
Zoe Liu's avatar
Zoe Liu committed
1005
1006
1007
#if CONFIG_ALTREF2
  rd->thresh_mult[THR_NEARA2] = 1000;
#endif  // CONFIG_ALTREF2
1008
#endif  // CONFIG_EXT_REFS
Jingning Han's avatar
Jingning Han committed
1009
  rd->thresh_mult[THR_NEARA] += 1000;
1010
1011
1012
1013
1014
1015
  rd->thresh_mult[THR_NEARG] += 1000;

  rd->thresh_mult[THR_ZEROMV] += 2000;
#if CONFIG_EXT_REFS
  rd->thresh_mult[THR_ZEROL2] += 2000;
  rd->thresh_mult[THR_ZEROL3] += 2000;
1016
  rd->thresh_mult[THR_ZEROB] += 2000;
Zoe Liu's avatar
Zoe Liu committed
1017
1018
1019
#if CONFIG_ALTREF2
  rd->thresh_mult[THR_ZEROA2] = 2000;
#endif  // CONFIG_ALTREF2
1020
1021
1022
#endif  // CONFIG_EXT_REFS
  rd->thresh_mult[THR_ZEROG] += 2000;
  rd->thresh_mult[THR_ZEROA] += 2000;
Jingning Han's avatar
Jingning Han committed
1023
1024
1025

  rd->thresh_mult[THR_TM] += 1000;

1026
#if CONFIG_EXT_INTER
1027

1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
#if CONFIG_COMPOUND_SINGLEREF
  rd->thresh_mult[THR_SR_NEAREST_NEARMV] += 1200;
#if CONFIG_EXT_REFS
  rd->thresh_mult[THR_SR_NEAREST_NEARL2] += 1200;
  rd->thresh_mult[THR_SR_NEAREST_NEARL3] += 1200;
  rd->thresh_mult[THR_SR_NEAREST_NEARB] += 1200;
#endif  // CONFIG_EXT_REFS
  rd->thresh_mult[THR_SR_NEAREST_NEARA] += 1200;
  rd->thresh_mult[THR_SR_NEAREST_NEARG] += 1200;

  /*
  rd->thresh_mult[THR_SR_NEAREST_NEWMV] += 1200;
#if CONFIG_EXT_REFS
  rd->thresh_mult[THR_SR_NEAREST_NEWL2] += 1200;
  rd->thresh_mult[THR_SR_NEAREST_NEWL3] += 1200;
  rd->thresh_mult[THR_SR_NEAREST_NEWB] += 1200;
#endif  // CONFIG_EXT_REFS
  rd->thresh_mult[THR_SR_NEAREST_NEWA] += 1200;
  rd->thresh_mult[THR_SR_NEAREST_NEWG] += 1200;*/

  rd->thresh_mult[THR_SR_NEAR_NEWMV] += 1500;
#if CONFIG_EXT_REFS
  rd->thresh_mult[THR_SR_NEAR_NEWL2] += 1500;
  rd->thresh_mult[THR_SR_NEAR_NEWL3] += 1500;
  rd->thresh_mult[THR_SR_NEAR_NEWB] += 1500;
#endif  // CONFIG_EXT_REFS
  rd->thresh_mult[THR_SR_NEAR_NEWA] += 1500;
  rd->thresh_mult[THR_SR_NEAR_NEWG] += 1500;

  rd->thresh_mult[THR_SR_ZERO_NEWMV] += 2000;
#if CONFIG_EXT_REFS
  rd->thresh_mult[THR_SR_ZERO_NEWL2] += 2000;
  rd->thresh_mult[THR_SR_ZERO_NEWL3] += 2000;
  rd->thresh_mult[THR_SR_ZERO_NEWB] += 2000;
#endif  // CONFIG_EXT_REFS
  rd->thresh_mult[THR_SR_ZERO_NEWA] += 2000;
  rd->thresh_mult[THR_SR_ZERO_NEWG] += 2000;

  rd->thresh_mult[THR_SR_NEW_NEWMV] += 1700;
#if CONFIG_EXT_REFS
  rd->thresh_mult[THR_SR_NEW_NEWL2] += 1700;
  rd->thresh_mult[THR_SR_NEW_NEWL3] += 1700;
  rd->thresh_mult[THR_SR_NEW_NEWB] += 1700;
#endif  // CONFIG_EXT_REFS
  rd->thresh_mult[THR_SR_NEW_NEWA] += 1700;
  rd->thresh_mult[THR_SR_NEW_NEWG] += 1700;
#endif  // CONFIG_COMPOUND_SINGLEREF

1076
  rd->thresh_mult[THR_COMP_NEAREST_NEARESTLA] += 1000;
1077
1078
1079
1080
#if CONFIG_EXT_REFS
  rd->thresh_mult[THR_COMP_NEAREST_NEARESTL2A] += 1000;
  rd->thresh_mult[THR_COMP_NEAREST_NEARESTL3A] += 1000;
#endif  // CONFIG_EXT_REFS
1081
  rd->thresh_mult[THR_COMP_NEAREST_NEARESTGA] += 1000;
1082
1083
1084
1085
1086
#if CONFIG_EXT_REFS
  rd->thresh_mult[THR_COMP_NEAREST_NEARESTLB] += 1000;
  rd->thresh_mult[THR_COMP_NEAREST_NEARESTL2B] += 1000;
  rd->thresh_mult[THR_COMP_NEAREST_NEARESTL3B] += 1000;
  rd->thresh_mult[THR_COMP_NEAREST_NEARESTGB] += 1000;
Zoe Liu's avatar
Zoe Liu committed
1087
1088
1089
1090
1091
1092
#if CONFIG_ALTREF2
  rd->thresh_mult[THR_COMP_NEAREST_NEARESTLA2] += 1000;
  rd->thresh_mult[THR_COMP_NEAREST_NEARESTL2A2] += 1000;
  rd->thresh_mult[THR_COMP_NEAREST_NEARESTL3A2] += 1000;
  rd->thresh_mult[THR_COMP_NEAREST_NEARESTGA2] += 1000;
#endif  // CONFIG_ALTREF2
Zoe Liu's avatar
Zoe Liu committed
1093
1094
1095

#if CONFIG_EXT_COMP_REFS
  rd->thresh_mult[THR_COMP_NEAREST_NEARESTLL2] += 1000;
1096
  rd->thresh_mult[THR_COMP_NEAREST_NEARESTLL3] += 1000;
Zoe Liu's avatar
Zoe Liu committed
1097
1098
1099
  rd->thresh_mult[THR_COMP_NEAREST_NEARESTLG] += 1000;
  rd->thresh_mult[THR_COMP_NEAREST_NEARESTBA] += 1000;
#endif  // CONFIG_EXT_COMP_REFS
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
#endif  // CONFIG_EXT_REFS

#else  // CONFIG_EXT_INTER

  rd->thresh_mult[THR_COMP_NEARESTLA] += 1000;
#if CONFIG_EXT_REFS
  rd->thresh_mult[THR_COMP_NEARESTL2A] += 1000;
  rd->thresh_mult[THR_COMP_NEARESTL3A] += 1000;
#endif  // CONFIG_EXT_REFS
  rd->thresh_mult[THR_COMP_NEARESTGA] += 1000;
#if CONFIG_EXT_REFS
  rd->thresh_mult[THR_COMP_NEARESTLB] += 1000;
  rd->thresh_mult[THR_COMP_NEARESTL2B] += 1000;
  rd->thresh_mult[THR_COMP_NEARESTL3B] += 1000;
  rd->thresh_mult[THR_COMP_NEARESTGB] += 1000;
Zoe Liu's avatar
Zoe Liu committed
1115
1116
1117
1118
1119
1120
1121
#if CONFIG_ALTREF2
  rd->thresh_mult[THR_COMP_NEARESTLA2] += 1000;
  rd->thresh_mult[THR_COMP_NEARESTL2A2] += 1000;
  rd->thresh_mult[THR_COMP_NEARESTL3A2] += 1000;
  rd->thresh_mult[THR_COMP_NEARESTGA2] += 1000;
#endif  // CONFIG_ALTREF2

Zoe Liu's avatar
Zoe Liu committed
1122
1123
#if CONFIG_EXT_COMP_REFS
  rd->thresh_mult[THR_COMP_NEARESTLL2] += 1000;