rdopt.c 477 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
 */

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

Yaowu Xu's avatar
Yaowu Xu committed
15
#include "./aom_dsp_rtcd.h"
Jingning Han's avatar
Jingning Han committed
16
#include "./av1_rtcd.h"
Jingning Han's avatar
Jingning Han committed
17

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

24
25
26
#if CONFIG_CFL
#include "av1/common/cfl.h"
#endif
27
28
29
30
31
32
#include "av1/common/common.h"
#include "av1/common/common_data.h"
#include "av1/common/entropy.h"
#include "av1/common/entropymode.h"
#include "av1/common/idct.h"
#include "av1/common/mvref_common.h"
33
#if CONFIG_MOTION_VAR
34
#include "av1/common/obmc.h"
35
#endif
36
37
38
39
40
41
#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/scan.h"
#include "av1/common/seg_common.h"
42
43
44
#if CONFIG_LV_MAP
#include "av1/common/txb_common.h"
#endif
Yue Chen's avatar
Yue Chen committed
45
46
47
#if CONFIG_WARPED_MOTION
#include "av1/common/warped_motion.h"
#endif  // CONFIG_WARPED_MOTION
Jingning Han's avatar
Jingning Han committed
48

Jingning Han's avatar
Jingning Han committed
49
#include "av1/encoder/aq_variance.h"
50
#include "av1/encoder/av1_quantize.h"
51
52
53
54
#include "av1/encoder/cost.h"
#include "av1/encoder/encodemb.h"
#include "av1/encoder/encodemv.h"
#include "av1/encoder/encoder.h"
55
56
57
#if CONFIG_LV_MAP
#include "av1/encoder/encodetxb.h"
#endif
58
59
60
61
62
63
#include "av1/encoder/hybrid_fwd_txfm.h"
#include "av1/encoder/mcomp.h"
#include "av1/encoder/palette.h"
#include "av1/encoder/ratectrl.h"
#include "av1/encoder/rd.h"
#include "av1/encoder/rdopt.h"
64
#include "av1/encoder/tokenize.h"
65
66
67
#if CONFIG_EXT_TX
#include "av1/encoder/tx_prune_model_weights.h"
#endif  // CONFIG_EXT_TX
68
69
#if CONFIG_PVQ
#include "av1/encoder/pvq_encoder.h"
Yushin Cho's avatar
Yushin Cho committed
70
#include "av1/common/pvq.h"
71
#endif  // CONFIG_PVQ
72
#if CONFIG_DUAL_FILTER
Angie Chiang's avatar
Angie Chiang committed
73
#define DUAL_FILTER_SET_SIZE (SWITCHABLE_FILTERS * SWITCHABLE_FILTERS)
74
#if USE_EXTRA_FILTER
Angie Chiang's avatar
Angie Chiang committed
75
static const int filter_sets[DUAL_FILTER_SET_SIZE][2] = {
76
77
78
  { 0, 0 }, { 0, 1 }, { 0, 2 }, { 0, 3 }, { 1, 0 }, { 1, 1 },
  { 1, 2 }, { 1, 3 }, { 2, 0 }, { 2, 1 }, { 2, 2 }, { 2, 3 },
  { 3, 0 }, { 3, 1 }, { 3, 2 }, { 3, 3 },
79
};
80
81
82
83
84
85
#else   // USE_EXTRA_FILTER
static const int filter_sets[DUAL_FILTER_SET_SIZE][2] = {
  { 0, 0 }, { 0, 1 }, { 0, 2 }, { 1, 0 }, { 1, 1 },
  { 1, 2 }, { 2, 0 }, { 2, 1 }, { 2, 2 },
};
#endif  // USE_EXTRA_FILTER
Angie Chiang's avatar
Angie Chiang committed
86
#endif  // CONFIG_DUAL_FILTER
87

Zoe Liu's avatar
Zoe Liu committed
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
#define LAST_FRAME_MODE_MASK                                          \
  ((1 << INTRA_FRAME) | (1 << LAST2_FRAME) | (1 << LAST3_FRAME) |     \
   (1 << GOLDEN_FRAME) | (1 << BWDREF_FRAME) | (1 << ALTREF2_FRAME) | \
   (1 << ALTREF_FRAME))
#define LAST2_FRAME_MODE_MASK                                         \
  ((1 << INTRA_FRAME) | (1 << LAST_FRAME) | (1 << LAST3_FRAME) |      \
   (1 << GOLDEN_FRAME) | (1 << BWDREF_FRAME) | (1 << ALTREF2_FRAME) | \
   (1 << ALTREF_FRAME))
#define LAST3_FRAME_MODE_MASK                                         \
  ((1 << INTRA_FRAME) | (1 << LAST_FRAME) | (1 << LAST2_FRAME) |      \
   (1 << GOLDEN_FRAME) | (1 << BWDREF_FRAME) | (1 << ALTREF2_FRAME) | \
   (1 << ALTREF_FRAME))
#define GOLDEN_FRAME_MODE_MASK                                       \
  ((1 << INTRA_FRAME) | (1 << LAST_FRAME) | (1 << LAST2_FRAME) |     \
   (1 << LAST3_FRAME) | (1 << BWDREF_FRAME) | (1 << ALTREF2_FRAME) | \
   (1 << ALTREF_FRAME))
#define BWDREF_FRAME_MODE_MASK                                       \
  ((1 << INTRA_FRAME) | (1 << LAST_FRAME) | (1 << LAST2_FRAME) |     \
   (1 << LAST3_FRAME) | (1 << GOLDEN_FRAME) | (1 << ALTREF2_FRAME) | \
   (1 << ALTREF_FRAME))
#define ALTREF2_FRAME_MODE_MASK                                     \
  ((1 << INTRA_FRAME) | (1 << LAST_FRAME) | (1 << LAST2_FRAME) |    \
   (1 << LAST3_FRAME) | (1 << GOLDEN_FRAME) | (1 << BWDREF_FRAME) | \
   (1 << ALTREF_FRAME))
#define ALTREF_FRAME_MODE_MASK                                      \
  ((1 << INTRA_FRAME) | (1 << LAST_FRAME) | (1 << LAST2_FRAME) |    \
   (1 << LAST3_FRAME) | (1 << GOLDEN_FRAME) | (1 << BWDREF_FRAME) | \
   (1 << ALTREF2_FRAME))

Zoe Liu's avatar
Zoe Liu committed
117
#if CONFIG_EXT_COMP_REFS
118
119
120
#define SECOND_REF_FRAME_MASK                                         \
  ((1 << ALTREF_FRAME) | (1 << ALTREF2_FRAME) | (1 << BWDREF_FRAME) | \
   (1 << GOLDEN_FRAME) | (1 << LAST2_FRAME) | 0x01)
121
#else  // !CONFIG_EXT_COMP_REFS
Zoe Liu's avatar
Zoe Liu committed
122
123
#define SECOND_REF_FRAME_MASK \
  ((1 << ALTREF_FRAME) | (1 << ALTREF2_FRAME) | (1 << BWDREF_FRAME) | 0x01)
Zoe Liu's avatar
Zoe Liu committed
124
#endif  // CONFIG_EXT_COMP_REFS
Jingning Han's avatar
Jingning Han committed
125

126
127
#define MIN_EARLY_TERM_INDEX 3
#define NEW_MV_DISCOUNT_FACTOR 8
Jingning Han's avatar
Jingning Han committed
128

129
130
131
132
133
#if CONFIG_EXT_INTRA
#define ANGLE_SKIP_THRESH 10
#define FILTER_FAST_SEARCH 1
#endif  // CONFIG_EXT_INTRA

134
135
136
// Setting this to 1 will disable trellis optimization within the
// transform search. Trellis optimization will still be applied
// in the final encode.
137
#ifndef DISABLE_TRELLISQ_SEARCH
138
#define DISABLE_TRELLISQ_SEARCH 0
139
#endif
140

141
142
143
144
145
146
static const double ADST_FLIP_SVM[8] = {
  /* vertical */
  -6.6623, -2.8062, -3.2531, 3.1671,
  /* horizontal */
  -7.7051, -3.2234, -3.6193, 3.4533
};
147

Jingning Han's avatar
Jingning Han committed
148
149
150
151
152
typedef struct {
  PREDICTION_MODE mode;
  MV_REFERENCE_FRAME ref_frame[2];
} MODE_DEFINITION;

153
typedef struct { MV_REFERENCE_FRAME ref_frame[2]; } REF_DEFINITION;
Jingning Han's avatar
Jingning Han committed
154
155

struct rdcost_block_args {
Yaowu Xu's avatar
Yaowu Xu committed
156
  const AV1_COMP *cpi;
Jingning Han's avatar
Jingning Han committed
157
  MACROBLOCK *x;
158
159
  ENTROPY_CONTEXT t_above[2 * MAX_MIB_SIZE];
  ENTROPY_CONTEXT t_left[2 * MAX_MIB_SIZE];
160
  RD_STATS rd_stats;
Jingning Han's avatar
Jingning Han committed
161
162
163
164
165
166
167
  int64_t this_rd;
  int64_t best_rd;
  int exit_early;
  int use_fast_coef_costing;
};

#define LAST_NEW_MV_INDEX 6
Yaowu Xu's avatar
Yaowu Xu committed
168
static const MODE_DEFINITION av1_mode_order[MAX_MODES] = {
Emil Keyder's avatar
Emil Keyder committed
169
170
171
172
  { NEARESTMV, { LAST_FRAME, NONE_FRAME } },
  { NEARESTMV, { LAST2_FRAME, NONE_FRAME } },
  { NEARESTMV, { LAST3_FRAME, NONE_FRAME } },
  { NEARESTMV, { BWDREF_FRAME, NONE_FRAME } },
Zoe Liu's avatar
Zoe Liu committed
173
  { NEARESTMV, { ALTREF2_FRAME, NONE_FRAME } },
Emil Keyder's avatar
Emil Keyder committed
174
175
  { NEARESTMV, { ALTREF_FRAME, NONE_FRAME } },
  { NEARESTMV, { GOLDEN_FRAME, NONE_FRAME } },
Jingning Han's avatar
Jingning Han committed
176

Emil Keyder's avatar
Emil Keyder committed
177
  { DC_PRED, { INTRA_FRAME, NONE_FRAME } },
Jingning Han's avatar
Jingning Han committed
178

Emil Keyder's avatar
Emil Keyder committed
179
180
181
182
  { NEWMV, { LAST_FRAME, NONE_FRAME } },
  { NEWMV, { LAST2_FRAME, NONE_FRAME } },
  { NEWMV, { LAST3_FRAME, NONE_FRAME } },
  { NEWMV, { BWDREF_FRAME, NONE_FRAME } },
Zoe Liu's avatar
Zoe Liu committed
183
  { NEWMV, { ALTREF2_FRAME, NONE_FRAME } },
Emil Keyder's avatar
Emil Keyder committed
184
185
  { NEWMV, { ALTREF_FRAME, NONE_FRAME } },
  { NEWMV, { GOLDEN_FRAME, NONE_FRAME } },
Jingning Han's avatar
Jingning Han committed
186

Emil Keyder's avatar
Emil Keyder committed
187
188
189
190
  { NEARMV, { LAST_FRAME, NONE_FRAME } },
  { NEARMV, { LAST2_FRAME, NONE_FRAME } },
  { NEARMV, { LAST3_FRAME, NONE_FRAME } },
  { NEARMV, { BWDREF_FRAME, NONE_FRAME } },
Zoe Liu's avatar
Zoe Liu committed
191
  { NEARMV, { ALTREF2_FRAME, NONE_FRAME } },
Emil Keyder's avatar
Emil Keyder committed
192
193
  { NEARMV, { ALTREF_FRAME, NONE_FRAME } },
  { NEARMV, { GOLDEN_FRAME, NONE_FRAME } },
Jingning Han's avatar
Jingning Han committed
194

Emil Keyder's avatar
Emil Keyder committed
195
196
197
198
  { ZEROMV, { LAST_FRAME, NONE_FRAME } },
  { ZEROMV, { LAST2_FRAME, NONE_FRAME } },
  { ZEROMV, { LAST3_FRAME, NONE_FRAME } },
  { ZEROMV, { BWDREF_FRAME, NONE_FRAME } },
Zoe Liu's avatar
Zoe Liu committed
199
  { ZEROMV, { ALTREF2_FRAME, NONE_FRAME } },
Emil Keyder's avatar
Emil Keyder committed
200
201
  { ZEROMV, { GOLDEN_FRAME, NONE_FRAME } },
  { ZEROMV, { ALTREF_FRAME, NONE_FRAME } },
Jingning Han's avatar
Jingning Han committed
202

203
// TODO(zoeliu): May need to reconsider the order on the modes to check
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
231
232
233
234
235
236
237
238
239
240
241
242
243
#if CONFIG_COMPOUND_SINGLEREF
  // Single ref comp mode
  { SR_NEAREST_NEARMV, { LAST_FRAME, NONE_FRAME } },
  { SR_NEAREST_NEARMV, { LAST2_FRAME, NONE_FRAME } },
  { SR_NEAREST_NEARMV, { LAST3_FRAME, NONE_FRAME } },
  { SR_NEAREST_NEARMV, { BWDREF_FRAME, NONE_FRAME } },
  { SR_NEAREST_NEARMV, { GOLDEN_FRAME, NONE_FRAME } },
  { SR_NEAREST_NEARMV, { ALTREF_FRAME, NONE_FRAME } },

  /*
  { SR_NEAREST_NEWMV, { LAST_FRAME, NONE_FRAME } },
  { SR_NEAREST_NEWMV, { LAST2_FRAME, NONE_FRAME } },
  { SR_NEAREST_NEWMV, { LAST3_FRAME, NONE_FRAME } },
  { SR_NEAREST_NEWMV, { BWDREF_FRAME, NONE_FRAME } },
  { SR_NEAREST_NEWMV, { GOLDEN_FRAME, NONE_FRAME } },
  { SR_NEAREST_NEWMV, { ALTREF_FRAME, NONE_FRAME } },*/

  { SR_NEAR_NEWMV, { LAST_FRAME, NONE_FRAME } },
  { SR_NEAR_NEWMV, { LAST2_FRAME, NONE_FRAME } },
  { SR_NEAR_NEWMV, { LAST3_FRAME, NONE_FRAME } },
  { SR_NEAR_NEWMV, { BWDREF_FRAME, NONE_FRAME } },
  { SR_NEAR_NEWMV, { GOLDEN_FRAME, NONE_FRAME } },
  { SR_NEAR_NEWMV, { ALTREF_FRAME, NONE_FRAME } },

  { SR_ZERO_NEWMV, { LAST_FRAME, NONE_FRAME } },
  { SR_ZERO_NEWMV, { LAST2_FRAME, NONE_FRAME } },
  { SR_ZERO_NEWMV, { LAST3_FRAME, NONE_FRAME } },
  { SR_ZERO_NEWMV, { BWDREF_FRAME, NONE_FRAME } },
  { SR_ZERO_NEWMV, { GOLDEN_FRAME, NONE_FRAME } },
  { SR_ZERO_NEWMV, { ALTREF_FRAME, NONE_FRAME } },

  { SR_NEW_NEWMV, { LAST_FRAME, NONE_FRAME } },
  { SR_NEW_NEWMV, { LAST2_FRAME, NONE_FRAME } },
  { SR_NEW_NEWMV, { LAST3_FRAME, NONE_FRAME } },
  { SR_NEW_NEWMV, { BWDREF_FRAME, NONE_FRAME } },
  { SR_NEW_NEWMV, { GOLDEN_FRAME, NONE_FRAME } },
  { SR_NEW_NEWMV, { ALTREF_FRAME, NONE_FRAME } },
#endif  // CONFIG_COMPOUND_SINGLEREF

244
245
246
247
248
249
250
251
  { NEAREST_NEARESTMV, { LAST_FRAME, ALTREF_FRAME } },
  { NEAREST_NEARESTMV, { LAST2_FRAME, ALTREF_FRAME } },
  { NEAREST_NEARESTMV, { LAST3_FRAME, ALTREF_FRAME } },
  { NEAREST_NEARESTMV, { GOLDEN_FRAME, ALTREF_FRAME } },
  { NEAREST_NEARESTMV, { LAST_FRAME, BWDREF_FRAME } },
  { NEAREST_NEARESTMV, { LAST2_FRAME, BWDREF_FRAME } },
  { NEAREST_NEARESTMV, { LAST3_FRAME, BWDREF_FRAME } },
  { NEAREST_NEARESTMV, { GOLDEN_FRAME, BWDREF_FRAME } },
Zoe Liu's avatar
Zoe Liu committed
252
253
254
255
  { NEAREST_NEARESTMV, { LAST_FRAME, ALTREF2_FRAME } },
  { NEAREST_NEARESTMV, { LAST2_FRAME, ALTREF2_FRAME } },
  { NEAREST_NEARESTMV, { LAST3_FRAME, ALTREF2_FRAME } },
  { NEAREST_NEARESTMV, { GOLDEN_FRAME, ALTREF2_FRAME } },
Zoe Liu's avatar
Zoe Liu committed
256
257
258

#if CONFIG_EXT_COMP_REFS
  { NEAREST_NEARESTMV, { LAST_FRAME, LAST2_FRAME } },
259
  { NEAREST_NEARESTMV, { LAST_FRAME, LAST3_FRAME } },
Zoe Liu's avatar
Zoe Liu committed
260
261
262
  { NEAREST_NEARESTMV, { LAST_FRAME, GOLDEN_FRAME } },
  { NEAREST_NEARESTMV, { BWDREF_FRAME, ALTREF_FRAME } },
#endif  // CONFIG_EXT_COMP_REFS
263

Urvang Joshi's avatar
Urvang Joshi committed
264
  { PAETH_PRED, { INTRA_FRAME, NONE_FRAME } },
Jingning Han's avatar
Jingning Han committed
265

Emil Keyder's avatar
Emil Keyder committed
266
  { SMOOTH_PRED, { INTRA_FRAME, NONE_FRAME } },
Urvang Joshi's avatar
Urvang Joshi committed
267
268
269
270
#if CONFIG_SMOOTH_HV
  { SMOOTH_V_PRED, { INTRA_FRAME, NONE_FRAME } },
  { SMOOTH_H_PRED, { INTRA_FRAME, NONE_FRAME } },
#endif  // CONFIG_SMOOTH_HV
271

272
273
274
275
276
277
278
  { NEAR_NEARMV, { LAST_FRAME, ALTREF_FRAME } },
  { NEW_NEARESTMV, { LAST_FRAME, ALTREF_FRAME } },
  { NEAREST_NEWMV, { LAST_FRAME, ALTREF_FRAME } },
  { NEW_NEARMV, { LAST_FRAME, ALTREF_FRAME } },
  { NEAR_NEWMV, { LAST_FRAME, ALTREF_FRAME } },
  { NEW_NEWMV, { LAST_FRAME, ALTREF_FRAME } },
  { ZERO_ZEROMV, { LAST_FRAME, ALTREF_FRAME } },
279

280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
  { NEAR_NEARMV, { LAST2_FRAME, ALTREF_FRAME } },
  { NEW_NEARESTMV, { LAST2_FRAME, ALTREF_FRAME } },
  { NEAREST_NEWMV, { LAST2_FRAME, ALTREF_FRAME } },
  { NEW_NEARMV, { LAST2_FRAME, ALTREF_FRAME } },
  { NEAR_NEWMV, { LAST2_FRAME, ALTREF_FRAME } },
  { NEW_NEWMV, { LAST2_FRAME, ALTREF_FRAME } },
  { ZERO_ZEROMV, { LAST2_FRAME, ALTREF_FRAME } },

  { NEAR_NEARMV, { LAST3_FRAME, ALTREF_FRAME } },
  { NEW_NEARESTMV, { LAST3_FRAME, ALTREF_FRAME } },
  { NEAREST_NEWMV, { LAST3_FRAME, ALTREF_FRAME } },
  { NEW_NEARMV, { LAST3_FRAME, ALTREF_FRAME } },
  { NEAR_NEWMV, { LAST3_FRAME, ALTREF_FRAME } },
  { NEW_NEWMV, { LAST3_FRAME, ALTREF_FRAME } },
  { ZERO_ZEROMV, { LAST3_FRAME, ALTREF_FRAME } },
295

296
297
298
299
300
301
302
  { NEAR_NEARMV, { GOLDEN_FRAME, ALTREF_FRAME } },
  { NEW_NEARESTMV, { GOLDEN_FRAME, ALTREF_FRAME } },
  { NEAREST_NEWMV, { GOLDEN_FRAME, ALTREF_FRAME } },
  { NEW_NEARMV, { GOLDEN_FRAME, ALTREF_FRAME } },
  { NEAR_NEWMV, { GOLDEN_FRAME, ALTREF_FRAME } },
  { NEW_NEWMV, { GOLDEN_FRAME, ALTREF_FRAME } },
  { ZERO_ZEROMV, { GOLDEN_FRAME, ALTREF_FRAME } },
303

304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
  { NEAR_NEARMV, { LAST_FRAME, BWDREF_FRAME } },
  { NEW_NEARESTMV, { LAST_FRAME, BWDREF_FRAME } },
  { NEAREST_NEWMV, { LAST_FRAME, BWDREF_FRAME } },
  { NEW_NEARMV, { LAST_FRAME, BWDREF_FRAME } },
  { NEAR_NEWMV, { LAST_FRAME, BWDREF_FRAME } },
  { NEW_NEWMV, { LAST_FRAME, BWDREF_FRAME } },
  { ZERO_ZEROMV, { LAST_FRAME, BWDREF_FRAME } },

  { NEAR_NEARMV, { LAST2_FRAME, BWDREF_FRAME } },
  { NEW_NEARESTMV, { LAST2_FRAME, BWDREF_FRAME } },
  { NEAREST_NEWMV, { LAST2_FRAME, BWDREF_FRAME } },
  { NEW_NEARMV, { LAST2_FRAME, BWDREF_FRAME } },
  { NEAR_NEWMV, { LAST2_FRAME, BWDREF_FRAME } },
  { NEW_NEWMV, { LAST2_FRAME, BWDREF_FRAME } },
  { ZERO_ZEROMV, { LAST2_FRAME, BWDREF_FRAME } },

  { NEAR_NEARMV, { LAST3_FRAME, BWDREF_FRAME } },
  { NEW_NEARESTMV, { LAST3_FRAME, BWDREF_FRAME } },
  { NEAREST_NEWMV, { LAST3_FRAME, BWDREF_FRAME } },
  { NEW_NEARMV, { LAST3_FRAME, BWDREF_FRAME } },
  { NEAR_NEWMV, { LAST3_FRAME, BWDREF_FRAME } },
  { NEW_NEWMV, { LAST3_FRAME, BWDREF_FRAME } },
  { ZERO_ZEROMV, { LAST3_FRAME, BWDREF_FRAME } },

  { NEAR_NEARMV, { GOLDEN_FRAME, BWDREF_FRAME } },
  { NEW_NEARESTMV, { GOLDEN_FRAME, BWDREF_FRAME } },
  { NEAREST_NEWMV, { GOLDEN_FRAME, BWDREF_FRAME } },
  { NEW_NEARMV, { GOLDEN_FRAME, BWDREF_FRAME } },
  { NEAR_NEWMV, { GOLDEN_FRAME, BWDREF_FRAME } },
  { NEW_NEWMV, { GOLDEN_FRAME, BWDREF_FRAME } },
  { ZERO_ZEROMV, { GOLDEN_FRAME, BWDREF_FRAME } },
Zoe Liu's avatar
Zoe Liu committed
335

Zoe Liu's avatar
Zoe Liu committed
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
  { NEAR_NEARMV, { LAST_FRAME, ALTREF2_FRAME } },
  { NEW_NEARESTMV, { LAST_FRAME, ALTREF2_FRAME } },
  { NEAREST_NEWMV, { LAST_FRAME, ALTREF2_FRAME } },
  { NEW_NEARMV, { LAST_FRAME, ALTREF2_FRAME } },
  { NEAR_NEWMV, { LAST_FRAME, ALTREF2_FRAME } },
  { NEW_NEWMV, { LAST_FRAME, ALTREF2_FRAME } },
  { ZERO_ZEROMV, { LAST_FRAME, ALTREF2_FRAME } },

  { NEAR_NEARMV, { LAST2_FRAME, ALTREF2_FRAME } },
  { NEW_NEARESTMV, { LAST2_FRAME, ALTREF2_FRAME } },
  { NEAREST_NEWMV, { LAST2_FRAME, ALTREF2_FRAME } },
  { NEW_NEARMV, { LAST2_FRAME, ALTREF2_FRAME } },
  { NEAR_NEWMV, { LAST2_FRAME, ALTREF2_FRAME } },
  { NEW_NEWMV, { LAST2_FRAME, ALTREF2_FRAME } },
  { ZERO_ZEROMV, { LAST2_FRAME, ALTREF2_FRAME } },

  { NEAR_NEARMV, { LAST3_FRAME, ALTREF2_FRAME } },
  { NEW_NEARESTMV, { LAST3_FRAME, ALTREF2_FRAME } },
  { NEAREST_NEWMV, { LAST3_FRAME, ALTREF2_FRAME } },
  { NEW_NEARMV, { LAST3_FRAME, ALTREF2_FRAME } },
  { NEAR_NEWMV, { LAST3_FRAME, ALTREF2_FRAME } },
  { NEW_NEWMV, { LAST3_FRAME, ALTREF2_FRAME } },
  { ZERO_ZEROMV, { LAST3_FRAME, ALTREF2_FRAME } },

  { NEAR_NEARMV, { GOLDEN_FRAME, ALTREF2_FRAME } },
  { NEW_NEARESTMV, { GOLDEN_FRAME, ALTREF2_FRAME } },
  { NEAREST_NEWMV, { GOLDEN_FRAME, ALTREF2_FRAME } },
  { NEW_NEARMV, { GOLDEN_FRAME, ALTREF2_FRAME } },
  { NEAR_NEWMV, { GOLDEN_FRAME, ALTREF2_FRAME } },
  { NEW_NEWMV, { GOLDEN_FRAME, ALTREF2_FRAME } },
  { ZERO_ZEROMV, { GOLDEN_FRAME, ALTREF2_FRAME } },

Zoe Liu's avatar
Zoe Liu committed
368
369
370
371
372
373
374
375
376
#if CONFIG_EXT_COMP_REFS
  { NEAR_NEARMV, { LAST_FRAME, LAST2_FRAME } },
  { NEW_NEARESTMV, { LAST_FRAME, LAST2_FRAME } },
  { NEAREST_NEWMV, { LAST_FRAME, LAST2_FRAME } },
  { NEW_NEARMV, { LAST_FRAME, LAST2_FRAME } },
  { NEAR_NEWMV, { LAST_FRAME, LAST2_FRAME } },
  { NEW_NEWMV, { LAST_FRAME, LAST2_FRAME } },
  { ZERO_ZEROMV, { LAST_FRAME, LAST2_FRAME } },

377
378
379
380
381
382
383
384
  { NEAR_NEARMV, { LAST_FRAME, LAST3_FRAME } },
  { NEW_NEARESTMV, { LAST_FRAME, LAST3_FRAME } },
  { NEAREST_NEWMV, { LAST_FRAME, LAST3_FRAME } },
  { NEW_NEARMV, { LAST_FRAME, LAST3_FRAME } },
  { NEAR_NEWMV, { LAST_FRAME, LAST3_FRAME } },
  { NEW_NEWMV, { LAST_FRAME, LAST3_FRAME } },
  { ZERO_ZEROMV, { LAST_FRAME, LAST3_FRAME } },

Zoe Liu's avatar
Zoe Liu committed
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
  { NEAR_NEARMV, { LAST_FRAME, GOLDEN_FRAME } },
  { NEW_NEARESTMV, { LAST_FRAME, GOLDEN_FRAME } },
  { NEAREST_NEWMV, { LAST_FRAME, GOLDEN_FRAME } },
  { NEW_NEARMV, { LAST_FRAME, GOLDEN_FRAME } },
  { NEAR_NEWMV, { LAST_FRAME, GOLDEN_FRAME } },
  { NEW_NEWMV, { LAST_FRAME, GOLDEN_FRAME } },
  { ZERO_ZEROMV, { LAST_FRAME, GOLDEN_FRAME } },

  { NEAR_NEARMV, { BWDREF_FRAME, ALTREF_FRAME } },
  { NEW_NEARESTMV, { BWDREF_FRAME, ALTREF_FRAME } },
  { NEAREST_NEWMV, { BWDREF_FRAME, ALTREF_FRAME } },
  { NEW_NEARMV, { BWDREF_FRAME, ALTREF_FRAME } },
  { NEAR_NEWMV, { BWDREF_FRAME, ALTREF_FRAME } },
  { NEW_NEWMV, { BWDREF_FRAME, ALTREF_FRAME } },
  { ZERO_ZEROMV, { BWDREF_FRAME, ALTREF_FRAME } },
#endif  // CONFIG_EXT_COMP_REFS
401

Emil Keyder's avatar
Emil Keyder committed
402
403
404
405
406
407
408
409
  { H_PRED, { INTRA_FRAME, NONE_FRAME } },
  { V_PRED, { INTRA_FRAME, NONE_FRAME } },
  { D135_PRED, { INTRA_FRAME, NONE_FRAME } },
  { D207_PRED, { INTRA_FRAME, NONE_FRAME } },
  { D153_PRED, { INTRA_FRAME, NONE_FRAME } },
  { D63_PRED, { INTRA_FRAME, NONE_FRAME } },
  { D117_PRED, { INTRA_FRAME, NONE_FRAME } },
  { D45_PRED, { INTRA_FRAME, NONE_FRAME } },
410

411
412
413
414
  { ZEROMV, { LAST_FRAME, INTRA_FRAME } },
  { NEARESTMV, { LAST_FRAME, INTRA_FRAME } },
  { NEARMV, { LAST_FRAME, INTRA_FRAME } },
  { NEWMV, { LAST_FRAME, INTRA_FRAME } },
415

416
417
418
419
420
421
422
423
424
  { ZEROMV, { LAST2_FRAME, INTRA_FRAME } },
  { NEARESTMV, { LAST2_FRAME, INTRA_FRAME } },
  { NEARMV, { LAST2_FRAME, INTRA_FRAME } },
  { NEWMV, { LAST2_FRAME, INTRA_FRAME } },

  { ZEROMV, { LAST3_FRAME, INTRA_FRAME } },
  { NEARESTMV, { LAST3_FRAME, INTRA_FRAME } },
  { NEARMV, { LAST3_FRAME, INTRA_FRAME } },
  { NEWMV, { LAST3_FRAME, INTRA_FRAME } },
425

426
427
428
429
  { ZEROMV, { GOLDEN_FRAME, INTRA_FRAME } },
  { NEARESTMV, { GOLDEN_FRAME, INTRA_FRAME } },
  { NEARMV, { GOLDEN_FRAME, INTRA_FRAME } },
  { NEWMV, { GOLDEN_FRAME, INTRA_FRAME } },
430

431
432
433
434
  { ZEROMV, { BWDREF_FRAME, INTRA_FRAME } },
  { NEARESTMV, { BWDREF_FRAME, INTRA_FRAME } },
  { NEARMV, { BWDREF_FRAME, INTRA_FRAME } },
  { NEWMV, { BWDREF_FRAME, INTRA_FRAME } },
Zoe Liu's avatar
Zoe Liu committed
435
436
437
438
439

  { ZEROMV, { ALTREF2_FRAME, INTRA_FRAME } },
  { NEARESTMV, { ALTREF2_FRAME, INTRA_FRAME } },
  { NEARMV, { ALTREF2_FRAME, INTRA_FRAME } },
  { NEWMV, { ALTREF2_FRAME, INTRA_FRAME } },
440

441
442
443
444
  { ZEROMV, { ALTREF_FRAME, INTRA_FRAME } },
  { NEARESTMV, { ALTREF_FRAME, INTRA_FRAME } },
  { NEARMV, { ALTREF_FRAME, INTRA_FRAME } },
  { NEWMV, { ALTREF_FRAME, INTRA_FRAME } },
Jingning Han's avatar
Jingning Han committed
445
446
};

hui su's avatar
hui su committed
447
static const PREDICTION_MODE intra_rd_search_mode_order[INTRA_MODES] = {
Urvang Joshi's avatar
Urvang Joshi committed
448
  DC_PRED,       H_PRED,        V_PRED,    SMOOTH_PRED, PAETH_PRED,
Urvang Joshi's avatar
Urvang Joshi committed
449
#if CONFIG_SMOOTH_HV
hui su's avatar
hui su committed
450
  SMOOTH_V_PRED, SMOOTH_H_PRED,
Urvang Joshi's avatar
Urvang Joshi committed
451
#endif  // CONFIG_SMOOTH_HV
Urvang Joshi's avatar
Urvang Joshi committed
452
  D135_PRED,     D207_PRED,     D153_PRED, D63_PRED,    D117_PRED,  D45_PRED,
hui su's avatar
hui su committed
453
454
};

Luc Trudeau's avatar
Luc Trudeau committed
455
456
#if CONFIG_CFL
static const UV_PREDICTION_MODE uv_rd_search_mode_order[UV_INTRA_MODES] = {
Urvang Joshi's avatar
Urvang Joshi committed
457
  UV_DC_PRED,       UV_CFL_PRED,      UV_H_PRED,
Urvang Joshi's avatar
Urvang Joshi committed
458
  UV_V_PRED,        UV_SMOOTH_PRED,   UV_PAETH_PRED,
Urvang Joshi's avatar
Urvang Joshi committed
459
#if CONFIG_SMOOTH_HV
Luc Trudeau's avatar
Luc Trudeau committed
460
  UV_SMOOTH_V_PRED, UV_SMOOTH_H_PRED,
Urvang Joshi's avatar
Urvang Joshi committed
461
462
463
#endif  // CONFIG_SMOOTH_HV
  UV_D135_PRED,     UV_D207_PRED,     UV_D153_PRED,
  UV_D63_PRED,      UV_D117_PRED,     UV_D45_PRED,
Luc Trudeau's avatar
Luc Trudeau committed
464
465
466
467
468
};
#else
#define uv_rd_search_mode_order intra_rd_search_mode_order
#endif  // CONFIG_CFL

hui su's avatar
hui su committed
469
static INLINE int write_uniform_cost(int n, int v) {
470
471
  const int l = get_unsigned_bits(n);
  const int m = (1 << l) - n;
472
  if (l == 0) return 0;
hui su's avatar
hui su committed
473
  if (v < m)
Yaowu Xu's avatar
Yaowu Xu committed
474
    return (l - 1) * av1_cost_bit(128, 0);
hui su's avatar
hui su committed
475
  else
Yaowu Xu's avatar
Yaowu Xu committed
476
    return l * av1_cost_bit(128, 0);
hui su's avatar
hui su committed
477
478
}

479
480
481
// constants for prune 1 and prune 2 decision boundaries
#define FAST_EXT_TX_CORR_MID 0.0
#define FAST_EXT_TX_EDST_MID 0.1
482
483
484
#define FAST_EXT_TX_CORR_MARGIN 0.5
#define FAST_EXT_TX_EDST_MARGIN 0.3

485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
static unsigned pixel_dist_visible_only(
    const AV1_COMP *const cpi, const MACROBLOCK *x, const uint8_t *src,
    const int src_stride, const uint8_t *dst, const int dst_stride,
    const BLOCK_SIZE tx_bsize, int txb_rows, int txb_cols, int visible_rows,
    int visible_cols) {
  unsigned sse;

  if (txb_rows == visible_rows && txb_cols == visible_cols
#if CONFIG_RECT_TX_EXT && (CONFIG_EXT_TX || CONFIG_VAR_TX)
      && tx_bsize < BLOCK_SIZES
#endif
      ) {
    cpi->fn_ptr[tx_bsize].vf(src, src_stride, dst, dst_stride, &sse);
    return sse;
  }
#if CONFIG_HIGHBITDEPTH
  const MACROBLOCKD *xd = &x->e_mbd;

  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
    uint64_t sse64 = aom_highbd_sse_odd_size(src, src_stride, dst, dst_stride,
                                             visible_cols, visible_rows);
    return (unsigned int)ROUND_POWER_OF_TWO(sse64, (xd->bd - 8) * 2);
  }
#else
  (void)x;
#endif  // CONFIG_HIGHBITDEPTH
  sse = aom_sse_odd_size(src, src_stride, dst, dst_stride, visible_cols,
                         visible_rows);
  return sse;
}

516
#if CONFIG_DIST_8X8
Yushin Cho's avatar
Yushin Cho committed
517
518
519
520
521
522
523
524
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
static uint64_t cdef_dist_8x8_16bit(uint16_t *dst, int dstride, uint16_t *src,
                                    int sstride, int coeff_shift) {
  uint64_t svar = 0;
  uint64_t dvar = 0;
  uint64_t sum_s = 0;
  uint64_t sum_d = 0;
  uint64_t sum_s2 = 0;
  uint64_t sum_d2 = 0;
  uint64_t sum_sd = 0;
  uint64_t dist = 0;

  int i, j;
  for (i = 0; i < 8; i++) {
    for (j = 0; j < 8; j++) {
      sum_s += src[i * sstride + j];
      sum_d += dst[i * dstride + j];
      sum_s2 += src[i * sstride + j] * src[i * sstride + j];
      sum_d2 += dst[i * dstride + j] * dst[i * dstride + j];
      sum_sd += src[i * sstride + j] * dst[i * dstride + j];
    }
  }
  /* Compute the variance -- the calculation cannot go negative. */
  svar = sum_s2 - ((sum_s * sum_s + 32) >> 6);
  dvar = sum_d2 - ((sum_d * sum_d + 32) >> 6);

  // Tuning of jm's original dering distortion metric used in CDEF tool,
  // suggested by jm
  const uint64_t a = 4;
  const uint64_t b = 2;
  const uint64_t c1 = (400 * a << 2 * coeff_shift);
  const uint64_t c2 = (b * 20000 * a * a << 4 * coeff_shift);

  dist =
      (uint64_t)floor(.5 +
                      (sum_d2 + sum_s2 - 2 * sum_sd) * .5 * (svar + dvar + c1) /
                          (sqrt(svar * (double)dvar + c2)));

  // Calibrate dist to have similar rate for the same QP with MSE only
  // distortion (as in master branch)
  dist = (uint64_t)((float)dist * 0.75);

  return dist;
}

Yushin Cho's avatar
Yushin Cho committed
561
static int od_compute_var_4x4(uint16_t *x, int stride) {
Yushin Cho's avatar
Yushin Cho committed
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
  int sum;
  int s2;
  int i;
  sum = 0;
  s2 = 0;
  for (i = 0; i < 4; i++) {
    int j;
    for (j = 0; j < 4; j++) {
      int t;

      t = x[i * stride + j];
      sum += t;
      s2 += t * t;
    }
  }
Yushin Cho's avatar
Yushin Cho committed
577

Yushin Cho's avatar
Yushin Cho committed
578
579
580
  return (s2 - (sum * sum >> 4)) >> 4;
}

581
582
583
584
585
586
587
/* OD_DIST_LP_MID controls the frequency weighting filter used for computing
   the distortion. For a value X, the filter is [1 X 1]/(X + 2) and
   is applied both horizontally and vertically. For X=5, the filter is
   a good approximation for the OD_QM8_Q4_HVS quantization matrix. */
#define OD_DIST_LP_MID (5)
#define OD_DIST_LP_NORM (OD_DIST_LP_MID + 2)

Yushin Cho's avatar
Yushin Cho committed
588
589
static double od_compute_dist_8x8(int use_activity_masking, uint16_t *x,
                                  uint16_t *y, od_coeff *e_lp, int stride) {
Yushin Cho's avatar
Yushin Cho committed
590
591
592
593
594
595
596
597
598
599
600
  double sum;
  int min_var;
  double mean_var;
  double var_stat;
  double activity;
  double calibration;
  int i;
  int j;
  double vardist;

  vardist = 0;
Yushin Cho's avatar
Yushin Cho committed
601

Yushin Cho's avatar
Yushin Cho committed
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
#if 1
  min_var = INT_MAX;
  mean_var = 0;
  for (i = 0; i < 3; i++) {
    for (j = 0; j < 3; j++) {
      int varx;
      int vary;
      varx = od_compute_var_4x4(x + 2 * i * stride + 2 * j, stride);
      vary = od_compute_var_4x4(y + 2 * i * stride + 2 * j, stride);
      min_var = OD_MINI(min_var, varx);
      mean_var += 1. / (1 + varx);
      /* The cast to (double) is to avoid an overflow before the sqrt.*/
      vardist += varx - 2 * sqrt(varx * (double)vary) + vary;
    }
  }
  /* We use a different variance statistic depending on whether activity
James Zern's avatar
James Zern committed
618
     masking is used, since the harmonic mean appeared slightly worse with
Yushin Cho's avatar
Yushin Cho committed
619
620
621
622
623
624
625
626
627
628
629
630
631
632
     masking off. The calibration constant just ensures that we preserve the
     rate compared to activity=1. */
  if (use_activity_masking) {
    calibration = 1.95;
    var_stat = 9. / mean_var;
  } else {
    calibration = 1.62;
    var_stat = min_var;
  }
  /* 1.62 is a calibration constant, 0.25 is a noise floor and 1/6 is the
     activity masking constant. */
  activity = calibration * pow(.25 + var_stat, -1. / 6);
#else
  activity = 1;
633
#endif  // 1
Yushin Cho's avatar
Yushin Cho committed
634
635
636
  sum = 0;
  for (i = 0; i < 8; i++) {
    for (j = 0; j < 8; j++)
637
      sum += e_lp[i * stride + j] * (double)e_lp[i * stride + j];
Yushin Cho's avatar
Yushin Cho committed
638
  }
639
640
641
  /* Normalize the filter to unit DC response. */
  sum *= 1. / (OD_DIST_LP_NORM * OD_DIST_LP_NORM * OD_DIST_LP_NORM *
               OD_DIST_LP_NORM);
Yushin Cho's avatar
Yushin Cho committed
642
643
644
645
  return activity * activity * (sum + vardist);
}

// Note : Inputs x and y are in a pixel domain
Yushin Cho's avatar
Yushin Cho committed
646
647
static double od_compute_dist_common(int activity_masking, uint16_t *x,
                                     uint16_t *y, int bsize_w, int bsize_h,
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
                                     int qindex, od_coeff *tmp,
                                     od_coeff *e_lp) {
  int i, j;
  double sum = 0;
  const int mid = OD_DIST_LP_MID;

  for (j = 0; j < bsize_w; j++) {
    e_lp[j] = mid * tmp[j] + 2 * tmp[bsize_w + j];
    e_lp[(bsize_h - 1) * bsize_w + j] = mid * tmp[(bsize_h - 1) * bsize_w + j] +
                                        2 * tmp[(bsize_h - 2) * bsize_w + j];
  }
  for (i = 1; i < bsize_h - 1; i++) {
    for (j = 0; j < bsize_w; j++) {
      e_lp[i * bsize_w + j] = mid * tmp[i * bsize_w + j] +
                              tmp[(i - 1) * bsize_w + j] +
                              tmp[(i + 1) * bsize_w + j];
    }
  }
  for (i = 0; i < bsize_h; i += 8) {
    for (j = 0; j < bsize_w; j += 8) {
Yushin Cho's avatar
Yushin Cho committed
668
      sum += od_compute_dist_8x8(activity_masking, &x[i * bsize_w + j],
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
                                 &y[i * bsize_w + j], &e_lp[i * bsize_w + j],
                                 bsize_w);
    }
  }
  /* Scale according to linear regression against SSE, for 8x8 blocks. */
  if (activity_masking) {
    sum *= 2.2 + (1.7 - 2.2) * (qindex - 99) / (210 - 99) +
           (qindex < 99 ? 2.5 * (qindex - 99) / 99 * (qindex - 99) / 99 : 0);
  } else {
    sum *= qindex >= 128
               ? 1.4 + (0.9 - 1.4) * (qindex - 128) / (209 - 128)
               : qindex <= 43 ? 1.5 + (2.0 - 1.5) * (qindex - 43) / (16 - 43)
                              : 1.5 + (1.4 - 1.5) * (qindex - 43) / (128 - 43);
  }

  return sum;
}

Yushin Cho's avatar
Yushin Cho committed
687
688
static double od_compute_dist(uint16_t *x, uint16_t *y, int bsize_w,
                              int bsize_h, int qindex) {
Yushin Cho's avatar
Yushin Cho committed
689
  assert(bsize_w >= 8 && bsize_h >= 8);
Yushin Cho's avatar
Yushin Cho committed
690
691
692
693
694
#if CONFIG_PVQ
  int activity_masking = 1;
#else
  int activity_masking = 0;
#endif
Luc Trudeau's avatar
Luc Trudeau committed
695
696
697
698
699
700
701
  int i, j;
  DECLARE_ALIGNED(16, od_coeff, e[MAX_TX_SQUARE]);
  DECLARE_ALIGNED(16, od_coeff, tmp[MAX_TX_SQUARE]);
  DECLARE_ALIGNED(16, od_coeff, e_lp[MAX_TX_SQUARE]);
  for (i = 0; i < bsize_h; i++) {
    for (j = 0; j < bsize_w; j++) {
      e[i * bsize_w + j] = x[i * bsize_w + j] - y[i * bsize_w + j];
702
    }
Luc Trudeau's avatar
Luc Trudeau committed
703
704
705
706
707
708
709
710
711
  }
  int mid = OD_DIST_LP_MID;
  for (i = 0; i < bsize_h; i++) {
    tmp[i * bsize_w] = mid * e[i * bsize_w] + 2 * e[i * bsize_w + 1];
    tmp[i * bsize_w + bsize_w - 1] =
        mid * e[i * bsize_w + bsize_w - 1] + 2 * e[i * bsize_w + bsize_w - 2];
    for (j = 1; j < bsize_w - 1; j++) {
      tmp[i * bsize_w + j] = mid * e[i * bsize_w + j] + e[i * bsize_w + j - 1] +
                             e[i * bsize_w + j + 1];
712
    }
713
  }
Luc Trudeau's avatar
Luc Trudeau committed
714
715
  return od_compute_dist_common(activity_masking, x, y, bsize_w, bsize_h,
                                qindex, tmp, e_lp);
716
717
}

Yushin Cho's avatar
Yushin Cho committed
718
719
static double od_compute_dist_diff(uint16_t *x, int16_t *e, int bsize_w,
                                   int bsize_h, int qindex) {
720
  assert(bsize_w >= 8 && bsize_h >= 8);
Yushin Cho's avatar
Yushin Cho committed
721
722
723
724
725
#if CONFIG_PVQ
  int activity_masking = 1;
#else
  int activity_masking = 0;
#endif
Luc Trudeau's avatar
Luc Trudeau committed
726
727
728
729
730
731
732
  DECLARE_ALIGNED(16, uint16_t, y[MAX_TX_SQUARE]);
  DECLARE_ALIGNED(16, od_coeff, tmp[MAX_TX_SQUARE]);
  DECLARE_ALIGNED(16, od_coeff, e_lp[MAX_TX_SQUARE]);
  int i, j;
  for (i = 0; i < bsize_h; i++) {
    for (j = 0; j < bsize_w; j++) {
      y[i * bsize_w + j] = x[i * bsize_w + j] - e[i * bsize_w + j];
733
    }
Luc Trudeau's avatar
Luc Trudeau committed
734
735
736
737
738
739
740
741
742
  }
  int mid = OD_DIST_LP_MID;
  for (i = 0; i < bsize_h; i++) {
    tmp[i * bsize_w] = mid * e[i * bsize_w] + 2 * e[i * bsize_w + 1];
    tmp[i * bsize_w + bsize_w - 1] =
        mid * e[i * bsize_w + bsize_w - 1] + 2 * e[i * bsize_w + bsize_w - 2];
    for (j = 1; j < bsize_w - 1; j++) {
      tmp[i * bsize_w + j] = mid * e[i * bsize_w + j] + e[i * bsize_w + j - 1] +
                             e[i * bsize_w + j + 1];
Yushin Cho's avatar
Yushin Cho committed
743
744
    }
  }
Luc Trudeau's avatar
Luc Trudeau committed
745
746
  return od_compute_dist_common(activity_masking, x, y, bsize_w, bsize_h,
                                qindex, tmp, e_lp);
Yushin Cho's avatar
Yushin Cho committed
747
748
}

749
int64_t av1_dist_8x8(const AV1_COMP *const cpi, const MACROBLOCK *x,
Yushin Cho's avatar
Yushin Cho committed
750
751
752
753
                     const uint8_t *src, int src_stride, const uint8_t *dst,
                     int dst_stride, const BLOCK_SIZE tx_bsize, int bsw,
                     int bsh, int visible_w, int visible_h, int qindex) {
  int64_t d = 0;
Yushin Cho's avatar
Yushin Cho committed
754
  int i, j;
755
  const MACROBLOCKD *xd = &x->e_mbd;
Yushin Cho's avatar
Yushin Cho committed
756
757
758
759

  DECLARE_ALIGNED(16, uint16_t, orig[MAX_TX_SQUARE]);
  DECLARE_ALIGNED(16, uint16_t, rec[MAX_TX_SQUARE]);

Yushin Cho's avatar
Yushin Cho committed
760
761
762
763
764
  assert(bsw >= 8);
  assert(bsh >= 8);
  assert((bsw & 0x07) == 0);
  assert((bsh & 0x07) == 0);

765
766
  if (x->tune_metric == AOM_TUNE_CDEF_DIST ||
      x->tune_metric == AOM_TUNE_DAALA_DIST) {
767
#if CONFIG_HIGHBITDEPTH
768
    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
769
      for (j = 0; j < bsh; j++)
770
        for (i = 0; i < bsw; i++)
771
          orig[j * bsw + i] = CONVERT_TO_SHORTPTR(src)[j * src_stride + i];
772

773
      if ((bsw == visible_w) && (bsh == visible_h)) {
774
775
        for (j = 0; j < bsh; j++)
          for (i = 0; i < bsw; i++)
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
            rec[j * bsw + i] = CONVERT_TO_SHORTPTR(dst)[j * dst_stride + i];
      } else {
        for (j = 0; j < visible_h; j++)
          for (i = 0; i < visible_w; i++)
            rec[j * bsw + i] = CONVERT_TO_SHORTPTR(dst)[j * dst_stride + i];

        if (visible_w < bsw) {
          for (j = 0; j < bsh; j++)
            for (i = visible_w; i < bsw; i++)
              rec[j * bsw + i] = CONVERT_TO_SHORTPTR(src)[j * src_stride + i];
        }

        if (visible_h < bsh) {
          for (j = visible_h; j < bsh; j++)
            for (i = 0; i < bsw; i++)
              rec[j * bsw + i] = CONVERT_TO_SHORTPTR(src)[j * src_stride + i];
        }
793
      }
794
    } else {
795
796
#endif
      for (j = 0; j < bsh; j++)
797
        for (i = 0; i < bsw; i++) orig[j * bsw + i] = src[j * src_stride + i];
798

799
      if ((bsw == visible_w) && (bsh == visible_h)) {
800
        for (j = 0; j < bsh; j++)
801
802
803
804
805
806
807
808
809
810
811
          for (i = 0; i < bsw; i++) rec[j * bsw + i] = dst[j * dst_stride + i];
      } else {
        for (j = 0; j < visible_h; j++)
          for (i = 0; i < visible_w; i++)
            rec[j * bsw + i] = dst[j * dst_stride + i];

        if (visible_w < bsw) {
          for (j = 0; j < bsh; j++)
            for (i = visible_w; i < bsw; i++)
              rec[j * bsw + i] = src[j * src_stride + i];
        }
812

813
814
815
816
817
        if (visible_h < bsh) {
          for (j = visible_h; j < bsh; j++)
            for (i = 0; i < bsw; i++)
              rec[j * bsw + i] = src[j * src_stride + i];
        }
818
819
      }
#if CONFIG_HIGHBITDEPTH
820
    }
821
#endif  // CONFIG_HIGHBITDEPTH
822
  }
Yushin Cho's avatar
Yushin Cho committed
823

824
825
826
  if (x->tune_metric == AOM_TUNE_DAALA_DIST) {
    d = (int64_t)od_compute_dist(orig, rec, bsw, bsh, qindex);
  } else if (x->tune_metric == AOM_TUNE_CDEF_DIST) {
Yushin Cho's avatar
Yushin Cho committed
827
828
829
830
831
832
833
834
835
836
837
838
    int coeff_shift = AOMMAX(xd->bd - 8, 0);

    for (i = 0; i < bsh; i += 8) {
      for (j = 0; j < bsw; j += 8) {
        d += cdef_dist_8x8_16bit(&rec[i * bsw + j], bsw, &orig[i * bsw + j],
                                 bsw, coeff_shift);
      }
    }
#if CONFIG_HIGHBITDEPTH
    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
      d = ((uint64_t)d) >> 2 * coeff_shift;
#endif
839
840
  } else {
    // Otherwise, MSE by default
841
842
    d = pixel_dist_visible_only(cpi, x, src, src_stride, dst, dst_stride,
                                tx_bsize, bsh, bsw, visible_h, visible_w);
Yushin Cho's avatar
Yushin Cho committed
843
  }
844

Yushin Cho's avatar
Yushin Cho committed
845
846
  return d;
}
847

848
static int64_t av1_dist_8x8_diff(const MACROBLOCK *x, const uint8_t *src,
Yushin Cho's avatar
Yushin Cho committed
849
850
851
852
                                 int src_stride, const int16_t *diff,
                                 int diff_stride, int bsw, int bsh,
                                 int visible_w, int visible_h, int qindex) {
  int64_t d = 0;
853
  int i, j;
854
  const MACROBLOCKD *xd = &x->e_mbd;
Yushin Cho's avatar
Yushin Cho committed
855
856
857
858

  DECLARE_ALIGNED(16, uint16_t, orig[MAX_TX_SQUARE]);
  DECLARE_ALIGNED(16, int16_t, diff16[MAX_TX_SQUARE]);

Yushin Cho's avatar
Yushin Cho committed
859
860
861
862
863
  assert(bsw >= 8);
  assert(bsh >= 8);
  assert((bsw & 0x07) == 0);
  assert((bsh & 0x07) == 0);

864
865
  if (x->tune_metric == AOM_TUNE_CDEF_DIST ||
      x->tune_metric == AOM_TUNE_DAALA_DIST) {
866
#if CONFIG_HIGHBITDEPTH
867
868
869
870
871
    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
      for (j = 0; j < bsh; j++)
        for (i = 0; i < bsw; i++)
          orig[j * bsw + i] = CONVERT_TO_SHORTPTR(src)[j * src_stride + i];
    } else {
872
#endif
873
874
      for (j = 0; j < bsh; j++)
        for (i = 0; i < bsw; i++) orig[j * bsw + i] = src[j * src_stride + i];
875
#if CONFIG_HIGHBITDEPTH
876
    }
877
#endif  // CONFIG_HIGHBITDEPTH
878

879
    if ((bsw == visible_w) && (bsh == visible_h)) {
880
      for (j = 0; j < bsh; j++)
881
882
883
884
885
886
887
888
889
890
891
        for (i = 0; i < bsw; i++)
          diff16[j * bsw + i] = diff[j * diff_stride + i];
    } else {
      for (j = 0; j < visible_h; j++)
        for (i = 0; i < visible_w; i++)
          diff16[j * bsw + i] = diff[j * diff_stride + i];

      if (visible_w < bsw) {
        for (j = 0; j < bsh; j++)
          for (i = visible_w; i < bsw; i++) diff16[j * bsw + i] = 0;
      }
892

893
894
895
896
      if (visible_h < bsh) {
        for (j = visible_h; j < bsh; j++)
          for (i = 0; i < bsw; i++) diff16[j * bsw + i] = 0;
      }
897
898
    }
  }
899

900
901
902
  if (x->tune_metric == AOM_TUNE_DAALA_DIST) {
    d = (int64_t)od_compute_dist_diff(orig, diff16, bsw, bsh, qindex);
  } else if (x->tune_metric == AOM_TUNE_CDEF_DIST) {
Yushin Cho's avatar
Yushin Cho committed
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
    int coeff_shift = AOMMAX(xd->bd - 8, 0);
    DECLARE_ALIGNED(16, uint16_t, dst16[MAX_TX_SQUARE]);

    for (i = 0; i < bsh; i++) {
      for (j = 0; j < bsw; j++) {
        dst16[i * bsw + j] = orig[i * bsw + j] - diff16[i * bsw + j];
      }
    }

    for (i = 0; i < bsh; i += 8) {
      for (j = 0; j < bsw; j += 8) {
        d += cdef_dist_8x8_16bit(&dst16[i * bsw + j], bsw, &orig[i * bsw + j],
                                 bsw, coeff_shift);
      }
    }
    // Don't scale 'd' for HBD since it will be done by caller side for diff
    // input
920
921
  } else {
    // Otherwise, MSE by default
922
    d = aom_sum_squares_2d_i16(diff, diff_stride, visible_w, visible_h);
Yushin Cho's avatar
Yushin Cho committed
923
  }
924
925
926

  return d;
}
Yushin Cho's avatar
Yushin Cho committed
927
#endif  // CONFIG_DIST_8X8
Yushin Cho's avatar
Yushin Cho committed
928

Yaowu Xu's avatar
Yaowu Xu committed
929
static void get_energy_distribution_fine(const AV1_COMP *cpi, BLOCK_SIZE bsize,
930
931
                                         const uint8_t *src, int src_stride,
                                         const uint8_t *dst, int dst_stride,
932
                                         double *hordist, double *verdist) {
933
934
  const int bw = block_size_wide[bsize];
  const int bh = block_size_high[bsize];
935
  unsigned int esq[16] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
936

937
  const int f_index = bsize - BLOCK_16X16;
938
  if (f_index < 0) {
939
940
    const int w_shift = bw == 8 ? 1 : 2;
    const int h_shift = bh == 8 ? 1 : 2;
941
#if CONFIG_HIGHBITDEPTH
942
    if (cpi->common.use_highbitdepth) {
943
944
945
946
947
      const uint16_t *src16 = CONVERT_TO_SHORTPTR(src);
      const uint16_t *dst16 = CONVERT_TO_SHORTPTR(dst);
      for (int i = 0; i < bh; ++i)
        for (int j = 0; j < bw; ++j) {
          const int index = (j >> w_shift) + ((i >> h_shift) << 2);
948
949
950
          esq[index] +=
              (src16[j + i * src_stride] - dst16[j + i * dst_stride]) *
              (src16[j + i * src_stride] - dst16[j + i * dst_stride]);
951
952
        }
    } else {
953
#endif  // CONFIG_HIGHBITDEPTH
954

955
956
957
      for (int i = 0; i < bh; ++i)
        for (int j = 0; j < bw; ++j) {
          const int index = (j >> w_shift) + ((i >> h_shift) << 2);
958
959
960
          esq[index] += (src[j + i * src_stride] - dst[j + i * dst_stride]) *
                        (src[j + i * src_stride] - dst[j + i * dst_stride]);
        }
961
#if CONFIG_HIGHBITDEPTH
962
    }
963
#endif  // CONFIG_HIGHBITDEPTH
964
  } else {
965
966
967
968
969
970
971
    cpi->fn_ptr[f_index].vf(src, src_stride, dst, dst_stride, &esq[0]);
    cpi->fn_ptr[f_index].vf(src + bw / 4, src_stride, dst + bw / 4, dst_stride,
                            &esq[1]);
    cpi->fn_ptr[f_index].vf(src + bw / 2, src_stride, dst + bw / 2, dst_stride,
                            &esq[2]);
    cpi->fn_ptr[f_index].vf(src + 3 * bw / 4, src_stride, dst + 3 * bw / 4,
                            dst_stride, &esq[3]);
972
973
974
    src += bh / 4 * src_stride;
    dst += bh / 4 * dst_stride;

975
976
977
978
979
980
981
    cpi->fn_ptr[f_index].vf(src, src_stride, dst, dst_stride, &esq[4]);
    cpi->fn_ptr[f_index].vf(src + bw / 4, src_stride, dst + bw / 4, dst_stride,
                            &esq[5]);
    cpi->fn_ptr[f_index].vf(src + bw / 2, src_stride, dst + bw / 2, dst_stride,
                            &esq[6]);
    cpi->fn_ptr[f_index].vf(src + 3 * bw / 4, src_stride, dst + 3 * bw / 4,
                            dst_stride, &esq[7]);
982
983
984
    src += bh / 4 * src_stride;
    dst += bh / 4 * dst_stride;

985
986
987
988
989
990
991
    cpi->fn_ptr[f_index].vf(src, src_stride, dst, dst_stride, &esq[8]);
    cpi->fn_ptr[f_index].vf(src + bw / 4, src_stride, dst + bw / 4, dst_stride,
                            &esq[9]);
    cpi->fn_ptr[f_index].vf(src + bw / 2, src_stride, dst + bw / 2, dst_stride,
                            &esq[10]);
    cpi->fn_ptr[f_index].vf(src + 3 * bw / 4, src_stride, dst + 3 * bw / 4,
                            dst_stride, &esq[11]);
992
993
994
    src += bh / 4 * src_stride;
    dst += bh / 4 * dst_stride;

995
996
997
998
999
1000
1001
    cpi->fn_ptr[f_index].vf(src, src_stride, dst, dst_stride, &esq[12]);
    cpi->fn_ptr[f_index].vf(src + bw / 4, src_stride, dst + bw / 4, dst_stride,
                            &esq[13]);
    cpi->fn_ptr[f_index].vf(src + bw / 2, src_stride, dst + bw / 2, dst_stride,
                            &esq[14]);
    cpi->fn_ptr[f_index].vf(src + 3 * bw / 4, src_stride, dst + 3 * bw / 4,
                            dst_stride, &esq[15]);
1002
1003
  }

1004
1005
1006
  double total = (double)esq[0] + esq[1] + esq[2] + esq[3] + esq[4] + esq[5] +
                 esq[6] + esq[7] + esq[8] + esq[9] + esq[10] + esq[11] +
                 esq[12] + esq[13] + esq[14] + esq[15];
1007
1008
  if (total > 0) {
    const double e_recip = 1.0 / total;
1009
1010
1011
1012
1013
1014
    hordist[0] = ((double)esq[0] + esq[4] + esq[8] + esq[12]) * e_recip;
    hordist[1] = ((double)esq[1] + esq[5] + esq[9] + esq[13]) * e_recip;
    hordist[2] = ((double)esq[2] + esq[6] + esq[10] + esq[14]) * e_recip;
    verdist[0] = ((double)esq[0] + esq[1] + esq[2] + esq[3]) * e_recip;
    verdist[1] = ((double)esq[4] + esq[5] + esq[6] + esq[7]) * e_recip;
    verdist[2] = ((double)esq[8] + esq[9] + esq[10] + esq[11]) * e_recip;
1015
1016
1017
1018
1019
1020
1021
  } else {
    hordist[0] = verdist[0] = 0.25;
    hordist[1] = verdist[1] = 0.25;
    hordist[2] = verdist[2] = 0.25;
  }
}

Urvang Joshi's avatar
Urvang Joshi committed
1022
1023
1024
static int adst_vs_flipadst(const AV1_COMP *cpi, BLOCK_SIZE bsize,
                            const uint8_t *src, int src_stride,
                            const uint8_t *dst, int dst_stride) {
1025
1026
  int prune_bitmask = 0;
  double svm_proj_h = 0, svm_proj_v = 0;
Alex Converse's avatar
Alex Converse committed
1027
  double hdist[3] = { 0, 0, 0 }, vdist[3] = { 0, 0, 0 };
1028
1029
  get_energy_distribution_fine(cpi, bsize, src, src_stride, dst, dst_stride,
                               hdist, vdist);
1030

1031
  svm_proj_v = vdist[0] * ADST_FLIP_SVM[0] + vdist[1] * ADST_FLIP_SVM[1] +
1032
               vdist[2] * ADST_FLIP_SVM[2] + ADST_FLIP_SVM[3];
1033
  svm_proj_h = hdist[0] * ADST_FLIP_SVM[4] + hdist[1] * ADST_FLIP_SVM[5] +
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
               hdist[2] * ADST_FLIP_SVM[6] + ADST_FLIP_SVM[7];
  if (svm_proj_v > FAST_EXT_TX_EDST_MID + FAST_EXT_TX_EDST_MARGIN)
    prune_bitmask |= 1 << FLIPADST_1D;
  else if (svm_proj_v < FAST_EXT_TX_EDST_MID - FAST_EXT_TX_EDST_MARGIN)
    prune_bitmask |= 1 << ADST_1D;

  if (svm_proj_h > FAST_EXT_TX_EDST_MID + FAST_EXT_TX_EDST_MARGIN)
    prune_bitmask |= 1 << (FLIPADST_1D + 8);
  else if (svm_proj_h < FAST_EXT_TX_EDST_MID - FAST_EXT_TX_EDST_MARGIN)
    prune_bitmask |= 1 << (ADST_1D + 8);

  return prune_bitmask;
}

#if CONFIG_EXT_TX
Alex Converse's avatar
Alex Converse committed
1049
1050
static void get_horver_correlation(const int16_t *diff, int stride, int w,
                                   int h, double *hcorr, double *vcorr) {
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
1076
1077
  // Returns hor/ver correlation coefficient
  const int num = (h - 1) * (w - 1);
  double num_r;
  int i, j;
  int64_t xy_sum = 0, xz_sum = 0;
  int64_t x_sum = 0, y_sum = 0, z_sum = 0;
  int64_t x2_sum = 0, y2_sum = 0, z2_sum = 0;
  double x_var_n, y_var_n, z_var_n, xy_var_n, xz_var_n;
  *hcorr = *vcorr = 1;

  assert(num > 0);
  num_r = 1.0 / num;
  for (i = 1; i < h; ++i) {
    for (j = 1; j < w; ++j) {
      const int16_t x = diff[i * stride + j];
      const int16_t y = diff[i * stride + j - 1];
      const int16_t z = diff[(i - 1) * stride + j];
      xy_sum += x * y;
      xz_sum += x * z;
      x_sum += x;
      y_sum += y;
      z_sum += z;
      x2_sum += x * x;
      y2_sum += y * y;
      z2_sum += z * z;
    }
  }
1078
1079
1080
  x_var_n = x2_sum - (x_sum * x_sum) * num_r;
  y_var_n = y2_sum - (y_sum * y_sum) * num_r;
  z_var_n = z2_sum - (z_sum * z_sum) * num_r;
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
  xy_var_n = xy_sum - (x_sum * y_sum) * num_r;
  xz_var_n = xz_sum - (x_sum * z_sum) * num_r;
  if (x_var_n > 0 && y_var_n > 0) {
    *hcorr = xy_var_n / sqrt(x_var_n * y_var_n);
    *hcorr = *hcorr < 0 ? 0 : *hcorr;
  }
  if (x_var_n > 0 && z_var_n > 0) {
    *vcorr = xz_var_n / sqrt(x_var_n * z_var_n);
    *vcorr = *vcorr < 0 ? 0 : *vcorr;
  }
}

Alex Converse's avatar
Alex Converse committed
1093
1094
int dct_vs_idtx(const int16_t *diff, int stride, int w, int h) {
  double hcorr, vcorr;
1095
  int prune_bitmask = 0;
Alex Converse's avatar
Alex Converse committed
1096
  get_horver_correlation(diff, stride, w, h, &hcorr, &vcorr);
1097

Alex Converse's avatar
Alex Converse committed
1098
  if (vcorr > FAST_EXT_TX_CORR_MID + FAST_EXT_TX_CORR_MARGIN)
1099
    prune_bitmask |= 1 << IDTX_1D;
Alex Converse's avatar
Alex Converse committed
1100
  else if (vcorr < FAST_EXT_TX_CORR_MID - FAST_EXT_TX_CORR_MARGIN)
1101
1102
    prune_bitmask |= 1 << DCT_1D;

Alex Converse's avatar
Alex Converse committed
1103
  if (hcorr > FAST_EXT_TX_CORR_MID + FAST_EXT_TX_CORR_MARGIN)
1104
    prune_bitmask |= 1 << (IDTX_1D + 8);
Alex Converse's avatar
Alex Converse committed
1105
  else if (hcorr < FAST_EXT_TX_CORR_MID - FAST_EXT_TX_CORR_MARGIN)
1106
1107
1108
1109
1110
    prune_bitmask |= 1 << (DCT_1D + 8);
  return prune_bitmask;
}

// Performance drop: 0.5%, Speed improvement: 24%
Yaowu Xu's avatar
Yaowu Xu committed
1111
static int prune_two_for_sby(const AV1_COMP *cpi, BLOCK_SIZE bsize,
Alex Converse's avatar
Alex Converse committed
1112
1113
                             MACROBLOCK *x, const MACROBLOCKD *xd,
                             int adst_flipadst, int dct_idtx) {
1114
  int prune = 0;
1115

Alex Converse's avatar
Alex Converse committed
1116
1117
1118
  if (adst_flipadst) {
    const struct macroblock_plane *const p = &x->plane[0];
    const struct macroblockd_plane *const pd = &xd->plane[0];
1119
    prune |= adst_vs_flipadst(cpi, bsize, p->src.buf, p->src.stride,
Alex Converse's avatar
Alex Converse committed
1120
1121
1122
1123
1124
1125
1126
1127
1128
                              pd->dst.buf, pd->dst.stride);
  }
  if (dct_idtx) {
    av1_subtract_plane(x, bsize, 0);
    const struct macroblock_plane *const p = &x->plane[0];
    const int bw = 4 << (b_width_log2_lookup[bsize]);
    const int bh = 4 << (b_height_log2_lookup[bsize]);
    prune |= dct_vs_idtx(p->src_diff, bw, bw, bh);
  }
1129
1130
1131

  return prune;
}
1132
1133
#endif  // CONFIG_EXT_TX

1134
// Performance drop: 0.3%, Speed improvement: 5%
Yaowu Xu's avatar
Yaowu Xu committed
1135
static int prune_one_for_sby(const AV1_COMP *cpi, BLOCK_SIZE bsize,
Alex Converse's avatar
Alex Converse committed
1136
1137
1138
                             const MACROBLOCK *x, const MACROBLOCKD *xd) {
  const struct macroblock_plane *const p = &x->plane[0];
  const struct macroblockd_plane *const pd = &xd->plane[0];
1139
  return adst_vs_flipadst(cpi, bsize, p->src.buf, p->src.stride, pd->dst.buf,
Alex Converse's avatar
Alex Converse committed
1140
                          pd->dst.stride);
1141
1142
}

Hui Su's avatar
Hui Su committed
1143
1144
1145
1146
1147
1148
1149
1150
1151
#if CONFIG_EXT_TX
// 1D Transforms used in inter set, this needs to be changed if
// ext_tx_used_inter is changed
static const int ext_tx_used_inter_1D[EXT_TX_SETS_INTER][TX_TYPES_1D] = {
  { 1, 0, 0, 0 }, { 1, 1, 1, 1 }, { 1, 1, 1, 1 }, { 1, 0, 0, 1 },
#if CONFIG_MRC_TX
  { 1, 0, 0, 1 },
#endif  // CONFIG_MRC_TX
};
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
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
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
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392

static void get_energy_distribution_finer(const int16_t *diff, int stride,
                                          int bw, int bh, float *hordist,
                                          float *verdist) {
  // First compute downscaled block energy values (esq); downscale factors
  // are defined by w_shift and h_shift.
  unsigned int esq[256];
  const int w_shift = bw <= 8 ? 0 : 1;
  const int h_shift = bh <= 8 ? 0 : 1;
  const int esq_w = bw <= 8 ? bw : bw / 2;
  const int esq_h = bh <= 8 ? bh : bh / 2;
  const int esq_sz = esq_w * esq_h;
  int i, j;
  memset(esq, 0, esq_sz * sizeof(esq[0]));
  for (i = 0; i < bh; i++) {
    unsigned int *cur_esq_row = esq + (i >> h_shift) * esq_w;
    const int16_t *cur_diff_row = diff + i * stride;
    for (j = 0; j < bw; j++) {
      cur_esq_row[j >> w_shift] += cur_diff_row[j] * cur_diff_row[j];
    }
  }

  uint64_t total = 0;
  for (i = 0; i < esq_sz; i++) total += esq[i];

  // Output hordist and verdist arrays are normalized 1D projections of esq
  if (total == 0) {
    float hor_val = 1.0f / esq_w;
    for (j = 0; j < esq_w - 1; j++) hordist[j] = hor_val;
    float ver_val = 1.0f / esq_h;
    for (i = 0; i < esq_h - 1; i++) verdist[i] = ver_val;
    return;
  }

  const float e_recip = 1.0f / (float)total;
  memset(hordist, 0, (esq_w - 1) * sizeof(hordist[0]));
  memset(verdist, 0, (esq_h - 1) * sizeof(verdist[0]));
  const unsigned int *cur_esq_row;
  for (i = 0; i < esq_h - 1; i++) {
    cur_esq_row = esq + i * esq_w;
    for (j = 0; j < esq_w - 1; j++) {
      hordist[j] += (float)cur_esq_row[j];
      verdist[i] += (float)cur_esq_row[j];
    }
    verdist[i] += (float)cur_esq_row[j];
  }
  cur_esq_row = esq + i * esq_w;
  for (j = 0; j < esq_w - 1; j++) hordist[j] += (float)cur_esq_row[j];

  for (j = 0; j < esq_w - 1; j++) hordist[j] *= e_recip;
  for (i = 0; i < esq_h - 1; i++) verdist[i] *= e_recip;
}

// Similar to get_horver_correlation, but also takes into account first
// row/column, when computing horizontal/vertical correlation.
static void get_horver_correlation_full(const int16_t *diff, int stride, int w,
                                        int h, float *hcorr, float *vcorr) {
  const float num_hor = h * (w - 1);
  const float num_ver = (h - 1) * w;
  int i, j;

  // The following notation is used:
  // x - current pixel
  // y - left neighbor pixel
  // z - top neighbor pixel
  int64_t xy_sum = 0, xz_sum = 0;
  int64_t xhor_sum = 0, xver_sum = 0, y_sum = 0, z_sum = 0;
  int64_t x2hor_sum = 0, x2ver_sum = 0, y2_sum = 0, z2_sum = 0;

  int16_t x, y, z;
  for (j = 1; j < w; ++j) {
    x = diff[j];
    y = diff[j - 1];
    xy_sum += x * y;
    xhor_sum += x;
    y_sum += y;
    x2hor_sum += x * x;
    y2_sum += y * y;
  }
  for (i = 1; i < h; ++i) {
    x = diff[i * stride];
    z = diff[(i - 1) * stride];
    xz_sum += x * z;
    xver_sum += x;
    z_sum += z;
    x2ver_sum += x * x;
    z2_sum += z * z;
    for (j = 1; j < w; ++j) {
      x = diff[i * stride + j];
      y = diff[i * stride + j - 1];
      z = diff[(i - 1) * stride + j];
      xy_sum += x * y;
      xz_sum += x * z;
      xhor_sum += x;
      xver_sum += x;
      y_sum += y;
      z_sum += z;
      x2hor_sum += x * x;
      x2ver_sum += x * x;
      y2_sum += y * y;
      z2_sum += z * z;
    }
  }
  const float xhor_var_n = x2hor_sum - (xhor_sum * xhor_sum) / num_hor;
  const float y_var_n = y2_sum - (y_sum * y_sum) / num_hor;
  const float xy_var_n = xy_sum - (xhor_sum * y_sum) / num_hor;
  const float xver_var_n = x2ver_sum - (xver_sum * xver_sum) / num_ver;
  const float z_var_n = z2_sum - (z_sum * z_sum) / num_ver;
  const float xz_var_n = xz_sum - (xver_sum * z_sum) / num_ver;

  *hcorr = *vcorr = 1;
  if (xhor_var_n > 0 && y_var_n > 0) {
    *hcorr = xy_var_n / sqrtf(xhor_var_n * y_var_n);
    *hcorr = *hcorr < 0 ? 0 : *hcorr;
  }
  if (xver_var_n > 0 && z_var_n > 0) {
    *vcorr = xz_var_n / sqrtf(xver_var_n * z_var_n);
    *vcorr = *vcorr < 0 ? 0 : *vcorr;
  }
}

// Performs a forward pass through a neural network with 2 fully-connected
// layers, assuming ReLU as activation function. Number of output neurons
// is always equal to 4.
// fc1, fc2 - weight matrices of the respective layers.
// b1, b2 - bias vectors of the respective layers.
static void compute_1D_scores(float *features, int num_features,
                              const float *fc1, const float *b1,
                              const float *fc2, const float *b2,
                              int num_hidden_units, float *dst_scores) {
  assert(num_hidden_units <= 32);
  float hidden_layer[32];
  for (int i = 0; i < num_hidden_units; i++) {
    const float *cur_coef = fc1 + i * num_features;
    hidden_layer[i] = 0.0f;
    for (int j = 0; j < num_features; j++)
      hidden_layer[i] += cur_coef[j] * features[j];
    hidden_layer[i] = AOMMAX(hidden_layer[i] + b1[i], 0.0f);
  }
  for (int i = 0; i < 4; i++) {
    const float *cur_coef = fc2 + i * num_hidden_units;
    dst_scores[i] = 0.0f;
    for (int j = 0; j < num_hidden_units; j++)
      dst_scores[i] += cur_coef[j] * hidden_layer[j];
    dst_scores[i] += b2[i];
  }
}

// Transforms raw scores into a probability distribution across 16 TX types
static void score_2D_transform_pow8(float *scores_2D, float shift) {
  float sum = 0.0f;
  int i;

  for (i = 0; i < 16; i++) {
    float v, v2, v4;
    v = AOMMAX(scores_2D[i] + shift, 0.0f);
    v2 = v * v;
    v4 = v2 * v2;
    scores_2D[i] = v4 * v4;
    sum += scores_2D[i];
  }
  for (i = 0; i < 16; i++) scores_2D[i] /= sum;
}

static int prune_tx_types_2D(BLOCK_SIZE bsize, const MACROBLOCK *x,
                             int tx_set_type, int pruning_aggressiveness) {
  if (bsize >= BLOCK_32X32) return 0;
  const struct macroblock_plane *const p = &x->plane[0];
  const int bidx = AOMMAX(bsize - BLOCK_4X4, 0);
  const float score_thresh =
      av1_prune_2D_adaptive_thresholds[bidx][pruning_aggressiveness - 1];

  float hfeatures[16], vfeatures[16];
  float hscores[4], vscores[4];
  float scores_2D[16];
  int tx_type_table_2D[16] = {
    DCT_DCT,      DCT_ADST,      DCT_FLIPADST,      V_DCT,
    ADST_DCT,     ADST_ADST,     ADST_FLIPADST,     V_ADST,
    FLIPADST_DCT, FLIPADST_ADST, FLIPADST_FLIPADST, V_FLIPADST,
    H_DCT,        H_ADST,        H_FLIPADST,        IDTX
  };
  const int bw = block_size_wide[bsize], bh = block_size_high[bsize];
  const int hfeatures_num = bw <= 8 ? bw : bw / 2;
  const int vfeatures_num = bh <= 8 ? bh : bh / 2;
  assert(hfeatures_num <= 16);
  assert(vfeatures_num <= 16);

  get_energy_distribution_finer(p->src_diff, bw, bw, bh, hfeatures, vfeatures);
  get_horver_correlation_full(p->src_diff, bw, bw, bh,
                              &hfeatures[hfeatures_num - 1],
                              &vfeatures[vfeatures_num - 1]);

  const float *fc1_hor = av1_prune_2D_learned_weights_hor[bidx];
  const float *b1_hor =
      fc1_hor + av1_prune_2D_num_hidden_units_hor[bidx] * hfeatures_num;
  const float *fc2_hor = b1_hor + av1_prune_2D_num_hidden_units_hor[bidx];
  const float *b2_hor = fc2_hor + av1_prune_2D_num_hidden_units_hor[bidx] * 4;
  compute_1D_scores(hfeatures, hfeatures_num, fc1_hor, b1_hor, fc2_hor, b2_hor,
                    av1_prune_2D_num_hidden_units_hor[bidx], hscores);

  const float *fc1_ver = av1_prune_2D_learned_weights_ver[bidx];
  const float *b1_ver =
      fc1_ver + av1_prune_2D_num_hidden_units_ver[bidx] * vfeatures_num;
  const float *fc2_ver = b1_ver + av1_prune_2D_num_hidden_units_ver[bidx];
  const float *b2_ver = fc2_ver + av1_prune_2D_num_hidden_units_ver[bidx] * 4;
  compute_1D_scores(vfeatures, vfeatures_num, fc1_ver, b1_ver, fc2_ver, b2_ver,
                    av1_prune_2D_num_hidden_units_ver[bidx], vscores);

  float score_2D_average = 0.0f;
  for (int i = 0; i < 4; i++) {
    float *cur_scores_2D = scores_2D + i * 4;
    cur_scores_2D[0] = vscores[i] * hscores[0];
    cur_scores_2D[1] = vscores[i] * hscores[1];
    cur_scores_2D[2] = vscores[i] * hscores[2];
    cur_scores_2D[3] = vscores[i] * hscores[3];
    score_2D_average += cur_scores_2D[0] + cur_scores_2D[1] + cur_scores_2D[2] +
                        cur_scores_2D[3];
  }
  score_2D_average /= 16;
  score_2D_transform_pow8(scores_2D, (20 - score_2D_average));

  // Always keep the TX type with the highest score, prune all others with
  // score below score_thresh.
  int max_score_i = 0;
  float max_score = 0.0f;
  for (int i = 0; i < 16; i++) {
    if (scores_2D[i] > max_score &&
        av1_ext_tx_used[tx_set_type][tx_type_table_2D[i]]) {
      max_score = scores_2D[i];
      max_score_i = i;
    }
  }

  int prune_bitmask = 0;
  for (int i = 0; i < 16; i++) {
    if (scores_2D[i] < score_thresh && i != max_score_i)
      prune_bitmask |= (1 << tx_type_table_2D[i]);
  }

  return prune_bitmask;
}
Hui Su's avatar
Hui Su committed
1393
1394
#endif  // CONFIG_EXT_TX

Yaowu Xu's avatar
Yaowu Xu committed
1395
static int prune_tx_types(const AV1_COMP *cpi, BLOCK_SIZE bsize, MACROBLOCK *x,
1396
                          const MACROBLOCKD *const xd, int tx_set_type) {
1397
#if CONFIG_EXT_TX
1398
1399
1400
  int tx_set = ext_tx_set_index[1][tx_set_type];
  assert(tx_set >= 0);
  const int *tx_set_1D = ext_tx_used_inter_1D[tx_set];
1401
#else
1402
  const int tx_set_1D[TX_TYPES_1D] = { 0 };
1403
  (void)tx_set_type;
1404
#endif  // CONFIG_EXT_TX
1405

1406
  switch (cpi->sf.tx_type_search.prune_mode) {
1407
1408
    case NO_PRUNE: return 0; break;
    case PRUNE_ONE:
1409
      if (!(tx_set_1D[FLIPADST_1D] & tx_set_1D[ADST_1D])) return 0;
1410
1411
      return prune_one_for_sby(cpi, bsize, x, xd);
      break;
1412
1413
#if CONFIG_EXT_TX
    case PRUNE_TWO:
1414
      if (!(tx_set_1D[FLIPADST_1D] & tx_set_1D[ADST_1D])) {
1415
        if (!(tx_set_1D[DCT_1D] & tx_set_1D[IDTX_1D])) return 0;
1416
1417
        return prune_two_for_sby(cpi, bsize, x, xd, 0, 1);
      }
1418
      if (!(tx_set_1D[DCT_1D] & tx_set_1D[IDTX_1D]))
1419
1420
        return prune_two_for_sby(cpi, bsize, x, xd, 1, 0);
      return prune_two_for_sby(cpi, bsize, x, xd, 1, 1);
1421
      break;
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
    case PRUNE_2D_ACCURATE:
      if (tx_set_type == EXT_TX_SET_ALL16)
        return prune_tx_types_2D(bsize, x, tx_set_type, 6);
      else if (tx_set_type == EXT_TX_SET_DTT9_IDTX_1DDCT)
        return prune_tx_types_2D(bsize, x, tx_set_type, 4);
      else
        return 0;
      break;
    case PRUNE_2D_FAST:
      if (tx_set_type == EXT_TX_SET_ALL16)
        return prune_tx_types_2D(bsize, x, tx_set_type, 10);
      else if (tx_set_type == EXT_TX_SET_DTT9_IDTX_1DDCT)
        return prune_tx_types_2D(bsize, x, tx_set_type, 7);
      else
        return 0;
      break;
1438
#endif  // CONFIG_EXT_TX
1439
1440
1441
1442
1443
  }
  assert(0);
  return 0;
}

1444
1445
static int do_tx_type_search(TX_TYPE tx_type, int prune,
                             TX_TYPE_PRUNE_MODE mode) {
1446
1447
// TODO(sarahparker) implement for non ext tx
#if CONFIG_EXT_TX
1448
1449
1450
1451
1452
1453
  if (mode >= PRUNE_2D_ACCURATE) {
    return !((prune >> tx_type) & 1);
  } else {
    return !(((prune >> vtx_tab[tx_type]) & 1) |
             ((prune >> (htx_tab[tx_type] + 8)) & 1));
  }
1454
1455
#else
  // temporary to avoid compiler warnings
1456
1457
1458
1459
  (void)vtx_tab;
  (void)htx_tab;
  (void)tx_type;
  (void)prune;
1460
  (void)mode;
1461
  return 1;
1462
#endif  // CONFIG_EXT_TX
1463
1464
}

Yaowu Xu's avatar
Yaowu Xu committed
1465
static void model_rd_from_sse(const AV1_COMP *const cpi,
1466
1467
                              const MACROBLOCKD *const xd, BLOCK_SIZE bsize,
                              int plane, int64_t sse, int *rate,
Geza Lore's avatar
Geza Lore committed
1468
1469
1470
                              int64_t *dist) {
  const struct macroblockd_plane *const pd = &xd->plane[plane];
  const int dequant_shift =
1471
#if CONFIG_HIGHBITDEPTH
1472
      (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) ? xd->bd - 5 :
1473
#endif  // CONFIG_HIGHBITDEPTH
1474
                                                    3;
Geza Lore's avatar
Geza Lore committed
1475
1476
1477
1478
1479
1480
1481

  // Fast approximate the modelling function.
  if (cpi->sf.simple_model_rd_from_var) {
    const int64_t square_error = sse;
    int quantizer = (pd->dequant[1] >> dequant_shift);

    if (quantizer < 120)
Yaowu Xu's avatar
Yaowu Xu committed
1482
      *rate = (int)((square_error * (280 - quantizer)) >>
Yaowu Xu's avatar
Yaowu Xu committed
1483
                    (16 - AV1_PROB_COST_SHIFT));
Geza Lore's avatar
Geza Lore committed
1484
1485
1486
1487
    else
      *rate = 0;
    *dist = (square_error * quantizer) >> 8;
  } else {
Yaowu Xu's avatar
Yaowu Xu committed
1488
1489
    av1_model_rd_from_var_lapndz(sse, num_pels_log2_lookup[bsize],
                                 pd->dequant[1] >> dequant_shift, rate, dist);
Geza Lore's avatar
Geza Lore committed
1490
1491
1492
1493
1494
  }

  *dist <<= 4;
}

Yaowu Xu's avatar
Yaowu Xu committed
1495
static void model_rd_for_sb(const AV1_COMP *const cpi, BLOCK_SIZE bsize,
1496
1497
1498
1499
                            MACROBLOCK *x, MACROBLOCKD *xd, int plane_from,
                            int plane_to, int *out_rate_sum,
                            int64_t *out_dist_sum, int *skip_txfm_sb,
                            int64_t *skip_sse_sb) {
Jingning Han's avatar
Jingning Han committed
1500
1501
1502
  // Note our transform coeffs are 8 times an orthogonal transform.
  // Hence quantizer step is also 8 times. To get effective quantizer
  // we need to divide by 8 before sending to modeling function.
Geza Lore's avatar
Geza Lore committed
1503
1504
1505
  int plane;
  const int ref = xd->mi[0]->mbmi.ref_frame[0];

Jingning Han's avatar
Jingning Han committed
1506
1507
1508
1509
1510
1511
  int64_t rate_sum = 0;
  int64_t dist_sum = 0;
  int64_t total_sse = 0;

  x->pred_sse[ref] = 0;

Geza Lore's avatar
Geza Lore committed
1512
1513
1514
  for (plane = plane_from; plane <= plane_to; ++plane) {
    struct macroblock_plane *const p = &x->plane[plane];
    struct macroblockd_plane *const pd = &xd->plane[plane];
1515
#if CONFIG_CHROMA_SUB8X8
1516
1517
    const BLOCK_SIZE bs = AOMMAX(BLOCK_4X4, get_plane_block_size(bsize, pd));
#else
Jingning Han's avatar
Jingning Han committed
1518
    const BLOCK_SIZE bs = get_plane_block_size(bsize, pd);
1519
#endif  // CONFIG_CHROMA_SUB8X8
Jingning Han's avatar
Jingning Han committed
1520

Geza Lore's avatar
Geza Lore committed
1521
1522
1523
    unsigned int sse;
    int rate;
    int64_t dist;
Jingning Han's avatar
Jingning Han committed
1524

1525
1526
#if CONFIG_CB4X4
    if (x->skip_chroma_rd && plane) continue;
1527
#endif  // CONFIG_CB4X4
1528

Geza Lore's avatar
Geza Lore committed
1529
1530
    // TODO(geza): Write direct sse functions that do not compute
    // variance as well.
1531
1532
    cpi->fn_ptr[bs].vf(p->src.buf, p->src.stride, pd->dst.buf, pd->dst.stride,
                       &sse);
Jingning Han's avatar
Jingning Han committed
1533

1534
    if (plane == 0) x->pred_sse[ref] = sse;
Jingning Han's avatar
Jingning Han committed
1535

Geza Lore's avatar
Geza Lore committed
1536
    total_sse += sse;
Jingning Han's avatar
Jingning Han committed
1537

Geza Lore's avatar
Geza Lore committed
1538
    model_rd_from_sse(cpi, xd, bs, plane, sse, &rate, &dist);
Geza Lore's avatar
Geza Lore committed
1539
1540
1541

    rate_sum += rate;
    dist_sum += dist;
Jingning Han's avatar
Jingning Han committed
1542
1543
  }

Geza Lore's avatar
Geza Lore committed
1544
  *skip_txfm_sb = total_sse == 0;
Jingning Han's avatar
Jingning Han committed
1545
1546
  *skip_sse_sb = total_sse << 4;
  *out_rate_sum = (int)rate_sum;
Geza Lore's avatar
Geza Lore committed
1547
  *out_dist_sum = dist_sum;
Jingning Han's avatar
Jingning Han committed
1548
1549
}

Yaowu Xu's avatar
Yaowu Xu committed
1550
1551
int64_t av1_block_error_c(const tran_low_t *coeff, const tran_low_t *dqcoeff,
                          intptr_t block_size, int64_t *ssz) {
Jingning Han's avatar
Jingning Han committed
1552
1553
1554
1555
1556
  int i;
  int64_t error = 0, sqcoeff = 0;

  for (i = 0; i < block_size; i++) {
    const int diff = coeff[i] - dqcoeff[i];
1557
    error += diff * diff;
Jingning Han's avatar
Jingning Han committed
1558
1559
1560
1561
1562
1563
1564
    sqcoeff += coeff[i] * coeff[i];
  }

  *ssz = sqcoeff;
  return error;
}

Yaowu Xu's avatar
Yaowu Xu committed
1565
1566
int64_t av1_block_error_fp_c(const int16_t *coeff, const int16_t *dqcoeff,
                             int block_size) {
Jingning Han's avatar
Jingning Han committed
1567
1568
1569
1570
1571
  int i;
  int64_t error = 0;

  for (i = 0; i < block_size; i++) {
    const int diff = coeff[i] - dqcoeff[i];
1572
    error += diff * diff;
Jingning Han's avatar
Jingning Han committed
1573
1574
1575
1576
1577
  }

  return error;
}

1578
#if CONFIG_HIGHBITDEPTH
Yaowu Xu's avatar
Yaowu Xu committed
1579
1580
1581
int64_t av1_highbd_block_error_c(const tran_low_t *coeff,
                                 const tran_low_t *dqcoeff, intptr_t block_size,
                                 int64_t *ssz, int bd) {
Jingning Han's avatar
Jingning Han committed
1582
1583
1584
1585
1586
1587
1588
  int i;
  int64_t error = 0, sqcoeff = 0;
  int shift = 2 * (bd - 8);
  int rounding = shift > 0 ? 1 << (shift - 1) : 0;

  for (i = 0; i < block_size; i++) {
    const int64_t diff = coeff[i] - dqcoeff[i];
1589
    error += diff * diff;
Jingning Han's avatar
Jingning Han committed
1590
1591