vp9_onyx_int.h 28.5 KB
Newer Older
John Koleszar's avatar
John Koleszar committed
1
/*
2
 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
John Koleszar's avatar
John Koleszar committed
3
 *
4
 *  Use of this source code is governed by a BSD-style license
5
6
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
7
 *  in the file PATENTS.  All contributing project authors may
8
 *  be found in the AUTHORS file in the root of the source tree.
John Koleszar's avatar
John Koleszar committed
9
10
 */

11
12
#ifndef VP9_ENCODER_VP9_ONYX_INT_H_
#define VP9_ENCODER_VP9_ONYX_INT_H_
John Koleszar's avatar
John Koleszar committed
13
14

#include <stdio.h>
15

16
#include "./vpx_config.h"
17
18
#include "vpx_ports/mem.h"
#include "vpx/internal/vpx_codec_internal.h"
19
#include "vpx/vp8cx.h"
20

21
#include "vp9/common/vp9_ppflags.h"
22
23
#include "vp9/common/vp9_entropy.h"
#include "vp9/common/vp9_entropymode.h"
24
#include "vp9/common/vp9_onyxc_int.h"
25

26
#include "vp9/encoder/vp9_encodemb.h"
27
#include "vp9/encoder/vp9_firstpass.h"
28
#include "vp9/encoder/vp9_lookahead.h"
29
#include "vp9/encoder/vp9_mbgraph.h"
30
31
32
#include "vp9/encoder/vp9_mcomp.h"
#include "vp9/encoder/vp9_quantize.h"
#include "vp9/encoder/vp9_ratectrl.h"
33
#include "vp9/encoder/vp9_svc_layercontext.h"
34
35
#include "vp9/encoder/vp9_tokenize.h"
#include "vp9/encoder/vp9_variance.h"
John Koleszar's avatar
John Koleszar committed
36

37
38
39
40
#ifdef __cplusplus
extern "C" {
#endif

Paul Wilkins's avatar
Paul Wilkins committed
41
// #define MODE_TEST_HIT_STATS
42

43
44
45
46
#if CONFIG_MULTIPLE_ARF
// Set MIN_GF_INTERVAL to 1 for the full decomposition.
#define MIN_GF_INTERVAL             2
#else
John Koleszar's avatar
John Koleszar committed
47
#define MIN_GF_INTERVAL             4
48
#endif
49
#define DEFAULT_GF_INTERVAL         10
John Koleszar's avatar
John Koleszar committed
50
51
52

#define KEY_FRAME_CONTEXT 5

53
54
#define MAX_MODES 30
#define MAX_REFS  6
55

John Koleszar's avatar
John Koleszar committed
56
57
58
#define MIN_THRESHMULT  32
#define MAX_THRESHMULT  512

John Koleszar's avatar
John Koleszar committed
59
typedef struct {
60
61
62
  int nmvjointcost[MV_JOINTS];
  int nmvcosts[2][MV_VALS];
  int nmvcosts_hp[2][MV_VALS];
John Koleszar's avatar
John Koleszar committed
63

64
  vp9_prob segment_pred_probs[PREDICTION_PROBS];
John Koleszar's avatar
John Koleszar committed
65

John Koleszar's avatar
John Koleszar committed
66
  unsigned char *last_frame_seg_map_copy;
67

John Koleszar's avatar
John Koleszar committed
68
69
  // 0 = Intra, Last, GF, ARF
  signed char last_ref_lf_deltas[MAX_REF_LF_DELTAS];
70
  // 0 = ZERO_MV, MV
John Koleszar's avatar
John Koleszar committed
71
  signed char last_mode_lf_deltas[MAX_MODE_LF_DELTAS];
John Koleszar's avatar
John Koleszar committed
72

73
  FRAME_CONTEXT fc;
John Koleszar's avatar
John Koleszar committed
74
75
} CODING_CONTEXT;

76
77
// This enumerator type needs to be kept aligned with the mode order in
// const MODE_DEFINITION vp9_mode_order[MAX_MODES] used in the rd code.
John Koleszar's avatar
John Koleszar committed
78
79
typedef enum {
  THR_NEARESTMV,
Paul Wilkins's avatar
Paul Wilkins committed
80
81
  THR_NEARESTA,
  THR_NEARESTG,
John Koleszar's avatar
John Koleszar committed
82

83
84
85
  THR_DC,

  THR_NEWMV,
Paul Wilkins's avatar
Paul Wilkins committed
86
  THR_NEWA,
87
88
89
  THR_NEWG,

  THR_NEARMV,
John Koleszar's avatar
John Koleszar committed
90
  THR_NEARA,
91
92
  THR_COMP_NEARESTLA,
  THR_COMP_NEARESTGA,
John Koleszar's avatar
John Koleszar committed
93
94

  THR_TM,
95

Paul Wilkins's avatar
Paul Wilkins committed
96
97
98
99
100
  THR_COMP_NEARLA,
  THR_COMP_NEWLA,
  THR_NEARG,
  THR_COMP_NEARGA,
  THR_COMP_NEWGA,
101

Paul Wilkins's avatar
Paul Wilkins committed
102
103
104
  THR_ZEROMV,
  THR_ZEROG,
  THR_ZEROA,
John Koleszar's avatar
John Koleszar committed
105
106
  THR_COMP_ZEROLA,
  THR_COMP_ZEROGA,
107

Paul Wilkins's avatar
Paul Wilkins committed
108
109
110
  THR_H_PRED,
  THR_V_PRED,
  THR_D135_PRED,
Dmitry Kovalev's avatar
Dmitry Kovalev committed
111
  THR_D207_PRED,
Paul Wilkins's avatar
Paul Wilkins committed
112
113
114
115
  THR_D153_PRED,
  THR_D63_PRED,
  THR_D117_PRED,
  THR_D45_PRED,
Ronald S. Bultje's avatar
Ronald S. Bultje committed
116
} THR_MODES;
John Koleszar's avatar
John Koleszar committed
117

118
119
120
121
122
123
124
125
126
typedef enum {
  THR_LAST,
  THR_GOLD,
  THR_ALTR,
  THR_COMP_LA,
  THR_COMP_GA,
  THR_INTRA,
} THR_MODES_SUB8X8;

John Koleszar's avatar
John Koleszar committed
127
128
129
typedef enum {
  DIAMOND = 0,
  NSTEP = 1,
130
131
  HEX = 2,
  BIGDIA = 3,
132
  SQUARE = 4,
133
134
  FAST_HEX = 5,
  FAST_DIAMOND = 6
John Koleszar's avatar
John Koleszar committed
135
136
} SEARCH_METHODS;

Deb Mukherjee's avatar
Deb Mukherjee committed
137
138
139
140
141
142
143
typedef enum {
  USE_FULL_RD = 0,
  USE_LARGESTINTRA,
  USE_LARGESTINTRA_MODELINTER,
  USE_LARGESTALL
} TX_SIZE_SEARCH_METHOD;

144
145
146
147
148
149
typedef enum {
  NOT_IN_USE = 0,
  RELAXED_NEIGHBORING_MIN_MAX = 1,
  STRICT_NEIGHBORING_MIN_MAX = 2
} AUTO_MIN_MAX_MODE;

150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
typedef enum {
  // Values should be powers of 2 so that they can be selected as bits of
  // an integer flags field

  // terminate search early based on distortion so far compared to
  // qp step, distortion in the neighborhood of the frame, etc.
  FLAG_EARLY_TERMINATE = 1,

  // skips comp inter modes if the best so far is an intra mode
  FLAG_SKIP_COMP_BESTINTRA = 2,

  // skips comp inter modes if the best single intermode so far does
  // not have the same reference as one of the two references being
  // tested
  FLAG_SKIP_COMP_REFMISMATCH = 4,

  // skips oblique intra modes if the best so far is an inter mode
  FLAG_SKIP_INTRA_BESTINTER = 8,

  // skips oblique intra modes  at angles 27, 63, 117, 153 if the best
  // intra so far is not one of the neighboring directions
  FLAG_SKIP_INTRA_DIRMISMATCH = 16,
172
173
174
175

  // skips intra modes other than DC_PRED if the source variance
  // is small
  FLAG_SKIP_INTRA_LOWVAR = 32,
176
177
} MODE_SEARCH_SKIP_LOGIC;

178
typedef enum {
179
  SUBPEL_TREE = 0,
180
181
182
  // Other methods to come
} SUBPEL_SEARCH_METHODS;

Paul Wilkins's avatar
Paul Wilkins committed
183
184
185
186
187
188
typedef enum {
  LAST_FRAME_PARTITION_OFF = 0,
  LAST_FRAME_PARTITION_LOW_MOTION = 1,
  LAST_FRAME_PARTITION_ALL = 2
} LAST_FRAME_PARTITION_METHOD;

189
190
191
192
193
194
195
196
197
198
199
typedef enum {
  // No recode.
  DISALLOW_RECODE = 0,
  // Allow recode for KF and exceeding maximum frame bandwidth.
  ALLOW_RECODE_KFMAXBW = 1,
  // Allow recode only for KF/ARF/GF frames.
  ALLOW_RECODE_KFARFGF = 2,
  // Allow recode for all frames based on bitrate constraints.
  ALLOW_RECODE = 3,
} RECODE_LOOP_TYPE;

200
201
202
203
204
205
206
207
208
typedef enum {
  // encode_breakout is disabled.
  ENCODE_BREAKOUT_DISABLED = 0,
  // encode_breakout is enabled.
  ENCODE_BREAKOUT_ENABLED = 1,
  // encode_breakout is enabled with small max_thresh limit.
  ENCODE_BREAKOUT_LIMITED = 2
} ENCODE_BREAKOUT_TYPE;

209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
typedef enum {
  // Search partitions using RD/NONRD criterion
  SEARCH_PARTITION = 0,

  // Always use a fixed size partition
  FIXED_PARTITION = 1,

  // Use a fixed size partition in every 64X64 SB, where the size is
  // determined based on source variance
  VAR_BASED_FIXED_PARTITION = 2,

  // Use an arbitrary partitioning scheme based on source variance within
  // a 64X64 SB
  VAR_BASED_PARTITION
} PARTITION_SEARCH_TYPE;

John Koleszar's avatar
John Koleszar committed
225
typedef struct {
226
227
  // Frame level coding parameter update
  int frame_parameter_update;
Jim Bankoski's avatar
Jim Bankoski committed
228
229

  // Motion search method (Diamond, NSTEP, Hex, Big Diamond, Square, etc).
John Koleszar's avatar
John Koleszar committed
230
  SEARCH_METHODS search_method;
Jim Bankoski's avatar
Jim Bankoski committed
231

232
  RECODE_LOOP_TYPE recode_loop;
Jim Bankoski's avatar
Jim Bankoski committed
233
234
235
236
237

  // Subpel_search_method can only be subpel_tree which does a subpixel
  // logarithmic search that keeps stepping at 1/2 pixel units until
  // you stop getting a gain, and then goes on to 1/4 and repeats
  // the same process. Along the way it skips many diagonals.
238
  SUBPEL_SEARCH_METHODS subpel_search_method;
Jim Bankoski's avatar
Jim Bankoski committed
239
240

  // Maximum number of steps in logarithmic subpel search before giving up.
241
  int subpel_iters_per_step;
Jim Bankoski's avatar
Jim Bankoski committed
242

Yaowu Xu's avatar
Yaowu Xu committed
243
244
245
  // Control when to stop subpel search
  int subpel_force_stop;

Jim Bankoski's avatar
Jim Bankoski committed
246
247
248
249
  // Thresh_mult is used to set a threshold for the rd score. A higher value
  // means that we will accept the best mode so far more often. This number
  // is used in combination with the current block size, and thresh_freq_fact
  // to pick a threshold.
John Koleszar's avatar
John Koleszar committed
250
  int thresh_mult[MAX_MODES];
251
  int thresh_mult_sub8x8[MAX_REFS];
Jim Bankoski's avatar
Jim Bankoski committed
252
253
254

  // This parameter controls the number of steps we'll do in a diamond
  // search.
John Koleszar's avatar
John Koleszar committed
255
  int max_step_search_steps;
Jim Bankoski's avatar
Jim Bankoski committed
256
257
258

  // This parameter controls which step in the n-step process we start at.
  // It's changed adaptively based on circumstances.
259
  int reduce_first_step_size;
Jim Bankoski's avatar
Jim Bankoski committed
260
261
262

  // If this is set to 1, we limit the motion search range to 2 times the
  // largest motion vector found in the last frame.
Paul Wilkins's avatar
Paul Wilkins committed
263
  int auto_mv_step_size;
Jim Bankoski's avatar
Jim Bankoski committed
264
265

  // Trellis (dynamic programming) optimization of quantized values (+1, 0).
John Koleszar's avatar
John Koleszar committed
266
  int optimize_coefficients;
Jim Bankoski's avatar
Jim Bankoski committed
267
268
269
270
271
272

  // Always set to 0. If on it enables 0 cost background transmission
  // (except for the initial transmission of the segmentation). The feature is
  // disabled because the addition of very large block sizes make the
  // backgrounds very to cheap to encode, and the segmentation we have
  // adds overhead.
273
  int static_segmentation;
Jim Bankoski's avatar
Jim Bankoski committed
274
275
276
277
278

  // If 1 we iterate finding a best reference for 2 ref frames together - via
  // a log search that iterates 4 times (check around mv for last for best
  // error of combined predictor then check around mv for alt). If 0 we
  // we just use the best motion vector found for each frame by itself.
279
  int comp_inter_joint_search_thresh;
Jim Bankoski's avatar
Jim Bankoski committed
280
281
282

  // This variable is used to cap the maximum number of times we skip testing a
  // mode to be evaluated. A high value means we will be faster.
283
  int adaptive_rd_thresh;
Jim Bankoski's avatar
Jim Bankoski committed
284
285
286
287

  // Enables skipping the reconstruction step (idct, recon) in the
  // intermediate steps assuming the last frame didn't have too many intra
  // blocks and the q is less than a threshold.
288
  int skip_encode_sb;
289
  int skip_encode_frame;
Jim Bankoski's avatar
Jim Bankoski committed
290
291
292
293
294
295
296
297

  // This variable allows us to reuse the last frames partition choices
  // (64x64 v 32x32 etc) for this frame. It can be set to only use the last
  // frame as a starting point in low motion scenes or always use it. If set
  // we use last partitioning_redo frequency to determine how often to redo
  // the partitioning from scratch. Adjust_partitioning_from_last_frame
  // enables us to adjust up or down one partitioning from the last frames
  // partitioning.
Paul Wilkins's avatar
Paul Wilkins committed
298
  LAST_FRAME_PARTITION_METHOD use_lastframe_partitioning;
Jim Bankoski's avatar
Jim Bankoski committed
299
300
301
302

  // Determine which method we use to determine transform size. We can choose
  // between options like full rd, largest for prediction size, largest
  // for intra and model coefs for the rest.
Deb Mukherjee's avatar
Deb Mukherjee committed
303
  TX_SIZE_SEARCH_METHOD tx_size_search_method;
Jim Bankoski's avatar
Jim Bankoski committed
304
305
306

  // Low precision 32x32 fdct keeps everything in 16 bits and thus is less
  // precise but significantly faster than the non lp version.
307
  int use_lp32x32fdct;
Jim Bankoski's avatar
Jim Bankoski committed
308
309
310
311
312
313

  // TODO(JBB): remove this as its no longer used.

  // After looking at the first set of modes (set by index here), skip
  // checking modes for reference frames that don't match the reference frame
  // of the best so far.
314
  int mode_skip_start;
Jim Bankoski's avatar
Jim Bankoski committed
315
316

  // TODO(JBB): Remove this.
Paul Wilkins's avatar
Paul Wilkins committed
317
  int reference_masking;
Jim Bankoski's avatar
Jim Bankoski committed
318

319
320
321
  PARTITION_SEARCH_TYPE partition_search_type;

  // Used if partition_search_type = FIXED_SIZE_PARTITION
322
  BLOCK_SIZE always_this_block_size;
Jim Bankoski's avatar
Jim Bankoski committed
323

324
325
326
327
328
329
330
  // Skip rectangular partition test when partition type none gives better
  // rd than partition type split.
  int less_rectangular_check;

  // Disable testing non square partitions. (eg 16x32)
  int use_square_partition_only;

Jim Bankoski's avatar
Jim Bankoski committed
331
  // Sets min and max partition sizes for this 64x64 region based on the
332
333
  // same 64x64 in last encoded frame, and the left and above neighbor.
  AUTO_MIN_MAX_MODE auto_min_max_partition_size;
Jim Bankoski's avatar
Jim Bankoski committed
334
335
336

  // Min and max partition size we enable (block_size) as per auto
  // min max, but also used by adjust partitioning, and pick_partitioning.
337
338
  BLOCK_SIZE min_partition_size;
  BLOCK_SIZE max_partition_size;
Jim Bankoski's avatar
Jim Bankoski committed
339
340
341

  // Whether or not we allow partitions one smaller or one greater than the last
  // frame's partitioning. Only used if use_lastframe_partitioning is set.
342
  int adjust_partitioning_from_last_frame;
Jim Bankoski's avatar
Jim Bankoski committed
343
344
345

  // How frequently we re do the partitioning from scratch. Only used if
  // use_lastframe_partitioning is set.
346
  int last_partitioning_redo_frequency;
Jim Bankoski's avatar
Jim Bankoski committed
347
348
349
350

  // Disables sub 8x8 blocksizes in different scenarios: Choices are to disable
  // it always, to allow it for only Last frame and Intra, disable it for all
  // inter modes or to enable it always.
351
  int disable_split_mask;
Jim Bankoski's avatar
Jim Bankoski committed
352

Jingning Han's avatar
Jingning Han committed
353
  // TODO(jingning): combine the related motion search speed features
Jim Bankoski's avatar
Jim Bankoski committed
354
355
  // This allows us to use motion search at other sizes as a starting
  // point for this motion search and limits the search range around it.
Jingning Han's avatar
Jingning Han committed
356
  int adaptive_motion_search;
Jim Bankoski's avatar
Jim Bankoski committed
357
358
359
360
361

  // Allows sub 8x8 modes to use the prediction filter that was determined
  // best for 8x8 mode. If set to 0 we always re check all the filters for
  // sizes less than 8x8, 1 means we check all filter modes if no 8x8 filter
  // was selected, and 2 means we use 8 tap if no 8x8 filter mode was selected.
362
  int adaptive_pred_interp_filter;
363
364
365
366
367

  // Implements various heuristics to skip searching modes
  // The heuristics selected are based on  flags
  // defined in the MODE_SEARCH_SKIP_HEURISTICS enum
  unsigned int mode_search_skip_flags;
Jim Bankoski's avatar
Jim Bankoski committed
368

369
370
  // A source variance threshold below which the split mode is disabled
  unsigned int disable_split_var_thresh;
Jim Bankoski's avatar
Jim Bankoski committed
371

372
373
374
  // A source variance threshold below which filter search is disabled
  // Choose a very large value (UINT_MAX) to use 8-tap always
  unsigned int disable_filter_search_var_thresh;
Jim Bankoski's avatar
Jim Bankoski committed
375
376
377

  // These bit masks allow you to enable or disable intra modes for each
  // transform size separately.
378
379
  int intra_y_mode_mask[TX_SIZES];
  int intra_uv_mode_mask[TX_SIZES];
Jim Bankoski's avatar
Jim Bankoski committed
380
381
382
383

  // This variable enables an early break out of mode testing if the model for
  // rd built from the prediction signal indicates a value that's much
  // higher than the best rd we've seen so far.
384
  int use_rd_breakout;
Jim Bankoski's avatar
Jim Bankoski committed
385
386
387
388

  // This enables us to use an estimate for intra rd based on dc mode rather
  // than choosing an actual uv mode in the stage of encoding before the actual
  // final encode.
389
  int use_uv_intra_rd_estimate;
Jim Bankoski's avatar
Jim Bankoski committed
390

391
392
393
394
  // This feature controls how the loop filter level is determined:
  // 0: Try the full image with different values.
  // 1: Try a small portion of the image with different values.
  // 2: Estimate the level based on quantizer and frame type
395
  int use_fast_lpf_pick;
Jim Bankoski's avatar
Jim Bankoski committed
396
397
398

  // This feature limits the number of coefficients updates we actually do
  // by only looking at counts from 1/2 the bands.
399
  int use_fast_coef_updates;  // 0: 2-loop, 1: 1-loop, 2: 1-loop reduced
400

401
  // This flag controls the use of non-RD mode decision.
402
  int use_nonrd_pick_mode;
403
404
405
406

  // This variable sets the encode_breakout threshold. Currently, it is only
  // enabled in real time mode.
  int encode_breakout_thresh;
407
408
409
410

  // A binary mask indicating if NEARESTMV, NEARMV, ZEROMV, NEWMV
  // modes are disabled in order from LSB to MSB for each BLOCK_SIZE.
  int disable_inter_mode_mask[BLOCK_SIZES];
411
412
413
414

  // This feature controls whether we do the expensive context update and
  // calculation in the rd coefficient costing loop.
  int use_fast_coef_costing;
415
416
417
418
419

  // This variable controls the maximum block size where intra blocks can be
  // used in inter frames.
  // TODO(aconverse): Fold this into one of the other many mode skips
  BLOCK_SIZE max_intra_bsize;
John Koleszar's avatar
John Koleszar committed
420
421
} SPEED_FEATURES;

422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
typedef enum {
  NORMAL      = 0,
  FOURFIVE    = 1,
  THREEFIVE   = 2,
  ONETWO      = 3
} VPX_SCALING;

typedef enum {
  VP9_LAST_FLAG = 1,
  VP9_GOLD_FLAG = 2,
  VP9_ALT_FLAG = 4
} VP9_REFFRAME;

typedef enum {
  USAGE_LOCAL_FILE_PLAYBACK   = 0x0,
  USAGE_STREAM_FROM_SERVER    = 0x1,
  USAGE_CONSTRAINED_QUALITY   = 0x2,
  USAGE_CONSTANT_QUALITY      = 0x3,
} END_USAGE;


typedef enum {
  MODE_GOODQUALITY    = 0x1,
  MODE_BESTQUALITY    = 0x2,
  MODE_FIRSTPASS      = 0x3,
  MODE_SECONDPASS     = 0x4,
  MODE_SECONDPASS_BEST = 0x5,
  MODE_REALTIME       = 0x6,
} MODE;

typedef enum {
  FRAMEFLAGS_KEY    = 1,
  FRAMEFLAGS_GOLDEN = 2,
  FRAMEFLAGS_ALTREF = 4,
} FRAMETYPE_FLAGS;

typedef enum {
  NO_AQ = 0,
  VARIANCE_AQ = 1,
  COMPLEXITY_AQ = 2,
462
463
  AQ_MODE_COUNT  // This should always be the last member of the enum
} AQ_MODE;
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499

typedef struct {
  int version;  // 4 versions of bitstream defined:
                //   0 - best quality/slowest decode,
                //   3 - lowest quality/fastest decode
  int width;  // width of data passed to the compressor
  int height;  // height of data passed to the compressor
  double framerate;  // set to passed in framerate
  int64_t target_bandwidth;  // bandwidth to be used in kilobits per second

  int noise_sensitivity;  // pre processing blur: recommendation 0
  int sharpness;  // sharpening output: recommendation 0:
  int cpu_used;
  unsigned int rc_max_intra_bitrate_pct;

  // mode ->
  // (0)=Realtime/Live Encoding. This mode is optimized for realtime
  //     encoding (for example, capturing a television signal or feed from
  //     a live camera). ( speed setting controls how fast )
  // (1)=Good Quality Fast Encoding. The encoder balances quality with the
  //     amount of time it takes to encode the output. ( speed setting
  //     controls how fast )
  // (2)=One Pass - Best Quality. The encoder places priority on the
  //     quality of the output over encoding speed. The output is compressed
  //     at the highest possible quality. This option takes the longest
  //     amount of time to encode. ( speed setting ignored )
  // (3)=Two Pass - First Pass. The encoder generates a file of statistics
  //     for use in the second encoding pass. ( speed setting controls how
  //     fast )
  // (4)=Two Pass - Second Pass. The encoder uses the statistics that were
  //     generated in the first encoding pass to create the compressed
  //     output. ( speed setting controls how fast )
  // (5)=Two Pass - Second Pass Best.  The encoder uses the statistics that
  //     were generated in the first encoding pass to create the compressed
  //     output using the highest possible quality, and taking a
  //    longer amount of time to encode.. ( speed setting ignored )
500
  MODE mode;
501
502
503
504
505
506
507
508
509
510

  // Key Framing Operations
  int auto_key;  // autodetect cut scenes and set the keyframes
  int key_freq;  // maximum distance to key frame.

  int lag_in_frames;  // how many frames lag before we start encoding

  // ----------------------------------------------------------------
  // DATARATE CONTROL OPTIONS

511
  END_USAGE end_usage;  // vbr or cbr
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530

  // buffer targeting aggressiveness
  int under_shoot_pct;
  int over_shoot_pct;

  // buffering parameters
  int64_t starting_buffer_level;  // in seconds
  int64_t optimal_buffer_level;
  int64_t maximum_buffer_size;

  // Frame drop threshold.
  int drop_frames_water_mark;

  // controlling quality
  int fixed_q;
  int worst_allowed_q;
  int best_allowed_q;
  int cq_level;
  int lossless;
531
  AQ_MODE aq_mode;  // Adaptive Quantization mode
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579

  // two pass datarate control
  int two_pass_vbrbias;        // two pass datarate control tweaks
  int two_pass_vbrmin_section;
  int two_pass_vbrmax_section;
  // END DATARATE CONTROL OPTIONS
  // ----------------------------------------------------------------

  // Spatial and temporal scalability.
  int ss_number_layers;  // Number of spatial layers.
  int ts_number_layers;  // Number of temporal layers.
  // Bitrate allocation for spatial layers.
  int ss_target_bitrate[VPX_SS_MAX_LAYERS];
  // Bitrate allocation (CBR mode) and framerate factor, for temporal layers.
  int ts_target_bitrate[VPX_TS_MAX_LAYERS];
  int ts_rate_decimator[VPX_TS_MAX_LAYERS];

  // these parameters aren't to be used in final build don't use!!!
  int play_alternate;
  int alt_freq;

  int encode_breakout;  // early breakout : for video conf recommend 800

  /* Bitfield defining the error resiliency features to enable.
   * Can provide decodable frames after losses in previous
   * frames and decodable partitions after losses in the same frame.
   */
  unsigned int error_resilient_mode;

  /* Bitfield defining the parallel decoding mode where the
   * decoding in successive frames may be conducted in parallel
   * just by decoding the frame headers.
   */
  unsigned int frame_parallel_decoding_mode;

  int arnr_max_frames;
  int arnr_strength;
  int arnr_type;

  int tile_columns;
  int tile_rows;

  struct vpx_fixed_buf         two_pass_stats_in;
  struct vpx_codec_pkt_list  *output_pkt_list;

  vp8e_tuning tuning;
} VP9_CONFIG;

580
typedef struct VP9_COMP {
581
582
583
584
  DECLARE_ALIGNED(16, int16_t, y_quant[QINDEX_RANGE][8]);
  DECLARE_ALIGNED(16, int16_t, y_quant_shift[QINDEX_RANGE][8]);
  DECLARE_ALIGNED(16, int16_t, y_zbin[QINDEX_RANGE][8]);
  DECLARE_ALIGNED(16, int16_t, y_round[QINDEX_RANGE][8]);
John Koleszar's avatar
John Koleszar committed
585

586
587
588
589
  DECLARE_ALIGNED(16, int16_t, uv_quant[QINDEX_RANGE][8]);
  DECLARE_ALIGNED(16, int16_t, uv_quant_shift[QINDEX_RANGE][8]);
  DECLARE_ALIGNED(16, int16_t, uv_zbin[QINDEX_RANGE][8]);
  DECLARE_ALIGNED(16, int16_t, uv_round[QINDEX_RANGE][8]);
John Koleszar's avatar
John Koleszar committed
590

591
#if CONFIG_ALPHA
592
593
594
595
  DECLARE_ALIGNED(16, int16_t, a_quant[QINDEX_RANGE][8]);
  DECLARE_ALIGNED(16, int16_t, a_quant_shift[QINDEX_RANGE][8]);
  DECLARE_ALIGNED(16, int16_t, a_zbin[QINDEX_RANGE][8]);
  DECLARE_ALIGNED(16, int16_t, a_round[QINDEX_RANGE][8]);
596
#endif
John Koleszar's avatar
John Koleszar committed
597

John Koleszar's avatar
John Koleszar committed
598
  MACROBLOCK mb;
599
  VP9_COMMON common;
600
  VP9_CONFIG oxcf;
John Koleszar's avatar
John Koleszar committed
601
602
  struct lookahead_ctx    *lookahead;
  struct lookahead_entry  *source;
603
#if CONFIG_MULTIPLE_ARF
Dmitry Kovalev's avatar
Dmitry Kovalev committed
604
  struct lookahead_entry  *alt_ref_source[REF_FRAMES];
605
#else
John Koleszar's avatar
John Koleszar committed
606
  struct lookahead_entry  *alt_ref_source;
607
#endif
John Koleszar's avatar
John Koleszar committed
608

John Koleszar's avatar
John Koleszar committed
609
610
611
  YV12_BUFFER_CONFIG *Source;
  YV12_BUFFER_CONFIG *un_scaled_source;
  YV12_BUFFER_CONFIG scaled_source;
John Koleszar's avatar
John Koleszar committed
612

613
  int key_frame_frequency;
John Koleszar's avatar
John Koleszar committed
614

615
616
  int gold_is_last;  // gold same as last frame ( short circuit gold searches)
  int alt_is_last;  // Alt same as last ( short circuit altref search)
John Koleszar's avatar
John Koleszar committed
617
  int gold_is_alt;  // don't do both alt and gold search ( just do gold).
John Koleszar's avatar
John Koleszar committed
618

619
  int scaled_ref_idx[3];
620
621
622
  int lst_fb_idx;
  int gld_fb_idx;
  int alt_fb_idx;
623

624
#if CONFIG_MULTIPLE_ARF
Dmitry Kovalev's avatar
Dmitry Kovalev committed
625
  int alt_ref_fb_idx[REF_FRAMES - 3];
626
#endif
627
628
629
  int refresh_last_frame;
  int refresh_golden_frame;
  int refresh_alt_ref_frame;
Deb Mukherjee's avatar
Deb Mukherjee committed
630
631
632
633
634
635
636
637
638

  int ext_refresh_frame_flags_pending;
  int ext_refresh_last_frame;
  int ext_refresh_golden_frame;
  int ext_refresh_alt_ref_frame;

  int ext_refresh_frame_context_pending;
  int ext_refresh_frame_context;

John Koleszar's avatar
John Koleszar committed
639
  YV12_BUFFER_CONFIG last_frame_uf;
John Koleszar's avatar
John Koleszar committed
640

John Koleszar's avatar
John Koleszar committed
641
  TOKENEXTRA *tok;
Ronald S. Bultje's avatar
Ronald S. Bultje committed
642
  unsigned int tok_count[4][1 << 6];
John Koleszar's avatar
John Koleszar committed
643

644
645
646
647
648
649
#if CONFIG_MULTIPLE_ARF
  // Position within a frame coding order (including any additional ARF frames).
  unsigned int sequence_number;
  // Next frame in naturally occurring order that has not yet been coded.
  int next_frame_in_order;
#endif
John Koleszar's avatar
John Koleszar committed
650

John Koleszar's avatar
John Koleszar committed
651
652
  // Ambient reconstruction err target for force key frames
  int ambient_err;
653

654
  int rd_threshes[MAX_SEGMENTS][BLOCK_SIZES][MAX_MODES];
655
  int rd_thresh_freq_fact[BLOCK_SIZES][MAX_MODES];
656
  int rd_thresh_sub8x8[MAX_SEGMENTS][BLOCK_SIZES][MAX_REFS];
657
  int rd_thresh_freq_sub8x8[BLOCK_SIZES][MAX_REFS];
658

659
  int64_t rd_comp_pred_diff[REFERENCE_MODES];
660
  int64_t rd_prediction_type_threshes[MAX_REF_FRAMES][REFERENCE_MODES];
661
  int64_t rd_tx_select_diff[TX_MODES];
662
  // FIXME(rbultje) can this overflow?
663
  int rd_tx_select_threshes[MAX_REF_FRAMES][TX_MODES];
John Koleszar's avatar
John Koleszar committed
664

665
  int64_t rd_filter_diff[SWITCHABLE_FILTER_CONTEXTS];
666
  int64_t rd_filter_threshes[MAX_REF_FRAMES][SWITCHABLE_FILTER_CONTEXTS];
667
  int64_t rd_filter_cache[SWITCHABLE_FILTER_CONTEXTS];
668
  int64_t mask_filter_rd;
669

John Koleszar's avatar
John Koleszar committed
670
671
  int RDMULT;
  int RDDIV;
John Koleszar's avatar
John Koleszar committed
672

John Koleszar's avatar
John Koleszar committed
673
  CODING_CONTEXT coding_context;
John Koleszar's avatar
John Koleszar committed
674

675
676
  int zbin_mode_boost;
  int zbin_mode_boost_enabled;
John Koleszar's avatar
John Koleszar committed
677
  int active_arnr_frames;           // <= cpi->oxcf.arnr_max_frames
678
  int active_arnr_strength;         // <= cpi->oxcf.arnr_max_strength
John Koleszar's avatar
John Koleszar committed
679

James Zern's avatar
James Zern committed
680
  double output_framerate;
John Koleszar's avatar
John Koleszar committed
681
682
683
  int64_t last_time_stamp_seen;
  int64_t last_end_time_stamp_seen;
  int64_t first_time_stamp_ever;
John Koleszar's avatar
John Koleszar committed
684

685
  RATE_CONTROL rc;
John Koleszar's avatar
John Koleszar committed
686

John Koleszar's avatar
John Koleszar committed
687
  int cq_target_quality;
688

689
690
691
  vp9_coeff_count coef_counts[TX_SIZES][PLANE_TYPES];
  vp9_coeff_probs_model frame_coef_probs[TX_SIZES][PLANE_TYPES];
  vp9_coeff_stats frame_branch_ct[TX_SIZES][PLANE_TYPES];
692

John Koleszar's avatar
John Koleszar committed
693
  struct vpx_codec_pkt_list  *output_pkt_list;
694

John Koleszar's avatar
John Koleszar committed
695
696
697
698
699
700
  MBGRAPH_FRAME_STATS mbgraph_stats[MAX_LAG_BUFFERS];
  int mbgraph_n_frames;             // number of frames filled in the above
  int static_mb_pct;                // % forced skip mbs by segmentation
  int seg0_progress, seg0_idx, seg0_cnt;

  // for real time encoding
701
  int speed;
John Koleszar's avatar
John Koleszar committed
702
703
704

  int cpu_used;
  int pass;
John Koleszar's avatar
John Koleszar committed
705

John Koleszar's avatar
John Koleszar committed
706
  int ref_frame_flags;
John Koleszar's avatar
John Koleszar committed
707

John Koleszar's avatar
John Koleszar committed
708
  SPEED_FEATURES sf;
John Koleszar's avatar
John Koleszar committed
709

Paul Wilkins's avatar
Paul Wilkins committed
710
  unsigned int max_mv_magnitude;
711
  int mv_step_param;
Paul Wilkins's avatar
Paul Wilkins committed
712

713
714
715
716
717
718
719
  // Default value is 1. From first pass stats, encode_breakout may be disabled.
  ENCODE_BREAKOUT_TYPE allow_encode_breakout;

  // Get threshold from external input. In real time mode, it can be
  // overwritten according to encoding speed.
  int encode_breakout;

John Koleszar's avatar
John Koleszar committed
720
  unsigned char *segmentation_map;
Paul Wilkins's avatar
Paul Wilkins committed
721

John Koleszar's avatar
John Koleszar committed
722
  // segment threashold for encode breakout
Paul Wilkins's avatar
Paul Wilkins committed
723
  int  segment_encode_breakout[MAX_SEGMENTS];
John Koleszar's avatar
John Koleszar committed
724

725
726
  unsigned char *complexity_map;

John Koleszar's avatar
John Koleszar committed
727
728
  unsigned char *active_map;
  unsigned int active_map_enabled;
John Koleszar's avatar
John Koleszar committed
729

John Koleszar's avatar
John Koleszar committed
730
  fractional_mv_step_fp *find_fractional_mv_step;
731
  fractional_mv_step_comp_fp *find_fractional_mv_step_comp;
732
733
734
  vp9_full_search_fn_t full_search_sad;
  vp9_refining_search_fn_t refining_search_sad;
  vp9_diamond_search_fn_t diamond_search_sad;
735
  vp9_variance_fn_ptr_t fn_ptr[BLOCK_SIZES];
Yaowu Xu's avatar
Yaowu Xu committed
736
737
738
  uint64_t time_receive_data;
  uint64_t time_compress_data;
  uint64_t time_pick_lpf;
739
  uint64_t time_encode_sb_row;
John Koleszar's avatar
John Koleszar committed
740

741
  struct twopass_rc twopass;
John Koleszar's avatar
John Koleszar committed
742

John Koleszar's avatar
John Koleszar committed
743
744
745
  YV12_BUFFER_CONFIG alt_ref_buffer;
  YV12_BUFFER_CONFIG *frames[MAX_LAG_BUFFERS];
  int fixed_divide[512];
John Koleszar's avatar
John Koleszar committed
746

747
#if CONFIG_INTERNAL_STATS
748
749
  unsigned int mode_chosen_counts[MAX_MODES];

John Koleszar's avatar
John Koleszar committed
750
751
752
753
754
  int    count;
  double total_y;
  double total_u;
  double total_v;
  double total;
755
756
757
  uint64_t total_sq_error;
  uint64_t total_samples;

John Koleszar's avatar
John Koleszar committed
758
759
760
761
  double totalp_y;
  double totalp_u;
  double totalp_v;
  double totalp;
762
763
764
  uint64_t totalp_sq_error;
  uint64_t totalp_samples;

John Koleszar's avatar
John Koleszar committed
765
766
767
  int    bytes;
  double summed_quality;
  double summed_weights;
768
769
  double summedp_quality;
  double summedp_weights;
John Koleszar's avatar
John Koleszar committed
770
771
772
773
774
775
776
777
778
  unsigned int tot_recode_hits;


  double total_ssimg_y;
  double total_ssimg_u;
  double total_ssimg_v;
  double total_ssimg_all;

  int b_calculate_ssimg;
John Koleszar's avatar
John Koleszar committed
779
#endif
John Koleszar's avatar
John Koleszar committed
780
  int b_calculate_psnr;
781

John Koleszar's avatar
John Koleszar committed
782
783
784
785
786
  // Per MB activity measurement
  unsigned int activity_avg;
  unsigned int *mb_activity_map;
  int *mb_norm_activity_map;
  int output_partition;
787

Jim Bankoski's avatar
Jim Bankoski committed
788
  // Force next frame to intra when kf_auto says so.
John Koleszar's avatar
John Koleszar committed
789
  int force_next_frame_intra;
790

John Koleszar's avatar
John Koleszar committed
791
  int droppable;
Adrian Grange's avatar
Adrian Grange committed
792

John Koleszar's avatar
John Koleszar committed
793
  int dummy_packing;    /* flag to indicate if packing is dummy */
794

Dmitry Kovalev's avatar
Dmitry Kovalev committed
795
  unsigned int tx_stepdown_count[TX_SIZES];
Deb Mukherjee's avatar
Deb Mukherjee committed
796

797
798
  int initial_width;
  int initial_height;
799

800
801
  int use_svc;

802
  SVC svc;
803

804
805
806
807
808
809
810
811
812
813
814
815
#if CONFIG_MULTIPLE_ARF
  // ARF tracking variables.
  int multi_arf_enabled;
  unsigned int frame_coding_order_period;
  unsigned int new_frame_coding_order_period;
  int frame_coding_order[MAX_LAG_BUFFERS * 2];
  int arf_buffer_idx[MAX_LAG_BUFFERS * 3 / 2];
  int arf_weight[MAX_LAG_BUFFERS];
  int arf_buffered;
  int this_frame_weight;
  int max_arf_level;
#endif
Paul Wilkins's avatar
Paul Wilkins committed
816

817
818
819
820
#ifdef MODE_TEST_HIT_STATS
  // Debug / test stats
  int64_t mode_test_hits[BLOCK_SIZES];
#endif
821

Jim Bankoski's avatar
Jim Bankoski committed
822
  // Y,U,V,(A)
823
824
825
  ENTROPY_CONTEXT *above_context[MAX_MB_PLANE];
  ENTROPY_CONTEXT left_context[MAX_MB_PLANE][16];

826
827
  PARTITION_CONTEXT *above_seg_context;
  PARTITION_CONTEXT left_seg_context[8];
828
} VP9_COMP;
John Koleszar's avatar
John Koleszar committed
829

830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
void vp9_initialize_enc();

struct VP9_COMP *vp9_create_compressor(VP9_CONFIG *oxcf);
void vp9_remove_compressor(VP9_COMP *cpi);

void vp9_change_config(VP9_COMP *cpi, VP9_CONFIG *oxcf);

  // receive a frames worth of data. caller can assume that a copy of this
  // frame is made and not just a copy of the pointer..
int vp9_receive_raw_frame(VP9_COMP *cpi, unsigned int frame_flags,
                          YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
                          int64_t end_time_stamp);

int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
                            size_t *size, uint8_t *dest,
                            int64_t *time_stamp, int64_t *time_end, int flush);

int vp9_get_preview_raw_frame(VP9_COMP *cpi, YV12_BUFFER_CONFIG *dest,
                              vp9_ppflags_t *flags);

int vp9_use_as_reference(VP9_COMP *cpi, int ref_frame_flags);

int vp9_update_reference(VP9_COMP *cpi, int ref_frame_flags);

int vp9_copy_reference_enc(VP9_COMP *cpi, VP9_REFFRAME ref_frame_flag,
                           YV12_BUFFER_CONFIG *sd);

int vp9_get_reference_enc(VP9_COMP *cpi, int index,
                          YV12_BUFFER_CONFIG **fb);

int vp9_set_reference_enc(VP9_COMP *cpi, VP9_REFFRAME ref_frame_flag,
                          YV12_BUFFER_CONFIG *sd);

int vp9_update_entropy(VP9_COMP *cpi, int update);

int vp9_set_roimap(VP9_COMP *cpi, unsigned char *map,
                   unsigned int rows, unsigned int cols,
                   int delta_q[MAX_SEGMENTS],
                   int delta_lf[MAX_SEGMENTS],
                   unsigned int threshold[MAX_SEGMENTS]);

int vp9_set_active_map(VP9_COMP *cpi, unsigned char *map,
                       unsigned int rows, unsigned int cols);

int vp9_set_internal_size(VP9_COMP *cpi,
                          VPX_SCALING horiz_mode, VPX_SCALING vert_mode);

int vp9_set_size_literal(VP9_COMP *cpi, unsigned int width,
                         unsigned int height);

void vp9_set_svc(VP9_COMP *cpi, int use_svc);

int vp9_get_quantizer(struct VP9_COMP *cpi);

884
885
static int get_ref_frame_idx(const VP9_COMP *cpi,
                             MV_REFERENCE_FRAME ref_frame) {
886
887
888
889
890
891
892
893
894
  if (ref_frame == LAST_FRAME) {
    return cpi->lst_fb_idx;
  } else if (ref_frame == GOLDEN_FRAME) {
    return cpi->gld_fb_idx;
  } else {
    return cpi->alt_fb_idx;
  }
}

895
896
897
static YV12_BUFFER_CONFIG *get_ref_frame_buffer(VP9_COMP *cpi,
                                                MV_REFERENCE_FRAME ref_frame) {
  VP9_COMMON *const cm = &cpi->common;
898
899
  return &cm->frame_bufs[cm->ref_frame_map[get_ref_frame_idx(cpi,
                                                             ref_frame)]].buf;
900
901
}

902
void vp9_set_speed_features(VP9_COMP *cpi);
Yaowu Xu's avatar
Yaowu Xu committed
903

904
905
int vp9_calc_ss_err(const YV12_BUFFER_CONFIG *source,
                    const YV12_BUFFER_CONFIG *reference);
906

907
void vp9_alloc_compressor_data(VP9_COMP *cpi);
908

Deb Mukherjee's avatar
Deb Mukherjee committed
909
int vp9_compute_qdelta(const VP9_COMP *cpi, double qstart, double qtarget);
910

911
912
913
914
static int get_token_alloc(int mb_rows, int mb_cols) {
  return mb_rows * mb_cols * (48 * 16 + 4);
}

915
916
917
918
extern const int q_trans[];

int64_t vp9_rescale(int64_t val, int64_t num, int denom);

919
920
static void set_ref_ptrs(VP9_COMMON *cm, MACROBLOCKD *xd,
                         MV_REFERENCE_FRAME ref0, MV_REFERENCE_FRAME ref1) {
921
922
923
924
  xd->block_refs[0] = &cm->frame_refs[ref0 >= LAST_FRAME ? ref0 - LAST_FRAME
                                                         : 0];
  xd->block_refs[1] = &cm->frame_refs[ref1 >= LAST_FRAME ? ref1 - LAST_FRAME
                                                         : 0];
925
926
}

927
928
929
930
#ifdef __cplusplus
}  // extern "C"
#endif

931
#endif  // VP9_ENCODER_VP9_ONYX_INT_H_