rdopt.c 181 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
13
14
15
 */


#include <stdio.h>
#include <math.h>
#include <limits.h>
#include <assert.h>
John Koleszar's avatar
John Koleszar committed
16
#include "vp8/common/pragmas.h"
John Koleszar's avatar
John Koleszar committed
17
18
19
20
21
22

#include "tokenize.h"
#include "treewriter.h"
#include "onyx_int.h"
#include "modecosts.h"
#include "encodeintra.h"
John Koleszar's avatar
John Koleszar committed
23
24
25
26
27
#include "vp8/common/entropymode.h"
#include "vp8/common/reconinter.h"
#include "vp8/common/reconintra.h"
#include "vp8/common/reconintra4x4.h"
#include "vp8/common/findnearmv.h"
Christian Duvivier's avatar
Christian Duvivier committed
28
#include "vp8/common/quant_common.h"
John Koleszar's avatar
John Koleszar committed
29
30
#include "encodemb.h"
#include "quantize.h"
John Koleszar's avatar
John Koleszar committed
31
32
#include "vp8/common/idct.h"
#include "vp8/common/g_common.h"
John Koleszar's avatar
John Koleszar committed
33
34
#include "variance.h"
#include "mcomp.h"
Yunqing Wang's avatar
Yunqing Wang committed
35
#include "rdopt.h"
Paul Wilkins's avatar
Paul Wilkins committed
36
#include "ratectrl.h"
John Koleszar's avatar
John Koleszar committed
37
38
#include "vpx_mem/vpx_mem.h"
#include "dct.h"
John Koleszar's avatar
John Koleszar committed
39
#include "vp8/common/systemdependent.h"
40
#include "vp8/encoder/encodemv.h"
John Koleszar's avatar
John Koleszar committed
41

42
#include "vp8/common/seg_common.h"
43
#include "vp8/common/pred_common.h"
44
#include "vp8/common/entropy.h"
45

46
#if CONFIG_NEWBESTREFMV
Paul Wilkins's avatar
Paul Wilkins committed
47
48
49
#include "vp8/common/mvref_common.h"
#endif

John Koleszar's avatar
John Koleszar committed
50
51
52
53
54
55
#if CONFIG_RUNTIME_CPU_DETECT
#define IF_RTCD(x)  (x)
#else
#define IF_RTCD(x)  NULL
#endif

Scott LaVarnway's avatar
Scott LaVarnway committed
56
57
58
extern void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x);
extern void vp8_update_zbin_extra(VP8_COMP *cpi, MACROBLOCK *x);

John Koleszar's avatar
John Koleszar committed
59
60
#define MAXF(a,b)            (((a) > (b)) ? (a) : (b))

61
62
#define INVALID_MV 0x80008000

63
64
65
/* Factor to weigh the rate for switchable interp filters */
#define SWITCHABLE_INTERP_RATE_FACTOR 1

John Koleszar's avatar
John Koleszar committed
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
static const int auto_speed_thresh[17] = {
  1000,
  200,
  150,
  130,
  150,
  125,
  120,
  115,
  115,
  115,
  115,
  115,
  115,
  115,
  115,
  115,
  105
John Koleszar's avatar
John Koleszar committed
84
85
};

86
#if CONFIG_PRED_FILTER
John Koleszar's avatar
John Koleszar committed
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
const MODE_DEFINITION vp8_mode_order[MAX_MODES] = {
  {ZEROMV,    LAST_FRAME,   0,  0},
  {ZEROMV,    LAST_FRAME,   0,  1},
  {DC_PRED,   INTRA_FRAME,  0,  0},

  {NEARESTMV, LAST_FRAME,   0,  0},
  {NEARESTMV, LAST_FRAME,   0,  1},
  {NEARMV,    LAST_FRAME,   0,  0},
  {NEARMV,    LAST_FRAME,   0,  1},

  {ZEROMV,    GOLDEN_FRAME, 0,  0},
  {ZEROMV,    GOLDEN_FRAME, 0,  1},
  {NEARESTMV, GOLDEN_FRAME, 0,  0},
  {NEARESTMV, GOLDEN_FRAME, 0,  1},

  {ZEROMV,    ALTREF_FRAME, 0,  0},
  {ZEROMV,    ALTREF_FRAME, 0,  1},
  {NEARESTMV, ALTREF_FRAME, 0,  0},
  {NEARESTMV, ALTREF_FRAME, 0,  1},

  {NEARMV,    GOLDEN_FRAME, 0,  0},
  {NEARMV,    GOLDEN_FRAME, 0,  1},
  {NEARMV,    ALTREF_FRAME, 0,  0},
  {NEARMV,    ALTREF_FRAME, 0,  1},

  {V_PRED,    INTRA_FRAME,  0,  0},
  {H_PRED,    INTRA_FRAME,  0,  0},
  {D45_PRED,  INTRA_FRAME,  0,  0},
  {D135_PRED, INTRA_FRAME,  0,  0},
  {D117_PRED, INTRA_FRAME,  0,  0},
  {D153_PRED, INTRA_FRAME,  0,  0},
  {D27_PRED,  INTRA_FRAME,  0,  0},
  {D63_PRED,  INTRA_FRAME,  0,  0},

  {TM_PRED,   INTRA_FRAME,  0,  0},

  {NEWMV,     LAST_FRAME,   0,  0},
  {NEWMV,     LAST_FRAME,   0,  1},
  {NEWMV,     GOLDEN_FRAME, 0,  0},
  {NEWMV,     GOLDEN_FRAME, 0,  1},
  {NEWMV,     ALTREF_FRAME, 0,  0},
  {NEWMV,     ALTREF_FRAME, 0,  1},

  {SPLITMV,   LAST_FRAME,   0,  0},
  {SPLITMV,   GOLDEN_FRAME, 0,  0},
  {SPLITMV,   ALTREF_FRAME, 0,  0},

  {B_PRED,    INTRA_FRAME,  0,  0},
  {I8X8_PRED, INTRA_FRAME,  0,  0},

  /* compound prediction modes */
  {ZEROMV,    LAST_FRAME,   GOLDEN_FRAME, 0},
  {NEARESTMV, LAST_FRAME,   GOLDEN_FRAME, 0},
  {NEARMV,    LAST_FRAME,   GOLDEN_FRAME, 0},

  {ZEROMV,    ALTREF_FRAME, LAST_FRAME,   0},
  {NEARESTMV, ALTREF_FRAME, LAST_FRAME,   0},
  {NEARMV,    ALTREF_FRAME, LAST_FRAME,   0},

  {ZEROMV,    GOLDEN_FRAME, ALTREF_FRAME, 0},
  {NEARESTMV, GOLDEN_FRAME, ALTREF_FRAME, 0},
  {NEARMV,    GOLDEN_FRAME, ALTREF_FRAME, 0},

  {NEWMV,     LAST_FRAME,   GOLDEN_FRAME, 0},
  {NEWMV,     ALTREF_FRAME, LAST_FRAME,   0},
  {NEWMV,     GOLDEN_FRAME, ALTREF_FRAME, 0},

  {SPLITMV,   LAST_FRAME,   GOLDEN_FRAME, 0},
  {SPLITMV,   ALTREF_FRAME, LAST_FRAME,   0},
  {SPLITMV,   GOLDEN_FRAME, ALTREF_FRAME, 0}
John Koleszar's avatar
John Koleszar committed
157
};
158
#else
John Koleszar's avatar
John Koleszar committed
159
160
161
const MODE_DEFINITION vp8_mode_order[MAX_MODES] = {
  {ZEROMV,    LAST_FRAME,   0},
  {DC_PRED,   INTRA_FRAME,  0},
John Koleszar's avatar
John Koleszar committed
162

John Koleszar's avatar
John Koleszar committed
163
164
  {NEARESTMV, LAST_FRAME,   0},
  {NEARMV,    LAST_FRAME,   0},
John Koleszar's avatar
John Koleszar committed
165

John Koleszar's avatar
John Koleszar committed
166
167
  {ZEROMV,    GOLDEN_FRAME, 0},
  {NEARESTMV, GOLDEN_FRAME, 0},
John Koleszar's avatar
John Koleszar committed
168

John Koleszar's avatar
John Koleszar committed
169
170
  {ZEROMV,    ALTREF_FRAME, 0},
  {NEARESTMV, ALTREF_FRAME, 0},
John Koleszar's avatar
John Koleszar committed
171

John Koleszar's avatar
John Koleszar committed
172
173
  {NEARMV,    GOLDEN_FRAME, 0},
  {NEARMV,    ALTREF_FRAME, 0},
John Koleszar's avatar
John Koleszar committed
174

John Koleszar's avatar
John Koleszar committed
175
176
177
178
179
180
181
182
  {V_PRED,    INTRA_FRAME,  0},
  {H_PRED,    INTRA_FRAME,  0},
  {D45_PRED,  INTRA_FRAME,  0},
  {D135_PRED, INTRA_FRAME,  0},
  {D117_PRED, INTRA_FRAME,  0},
  {D153_PRED, INTRA_FRAME,  0},
  {D27_PRED,  INTRA_FRAME,  0},
  {D63_PRED,  INTRA_FRAME,  0},
John Koleszar's avatar
John Koleszar committed
183

John Koleszar's avatar
John Koleszar committed
184
  {TM_PRED,   INTRA_FRAME,  0},
John Koleszar's avatar
John Koleszar committed
185

John Koleszar's avatar
John Koleszar committed
186
187
188
  {NEWMV,     LAST_FRAME,   0},
  {NEWMV,     GOLDEN_FRAME, 0},
  {NEWMV,     ALTREF_FRAME, 0},
John Koleszar's avatar
John Koleszar committed
189

John Koleszar's avatar
John Koleszar committed
190
191
192
  {SPLITMV,   LAST_FRAME,   0},
  {SPLITMV,   GOLDEN_FRAME, 0},
  {SPLITMV,   ALTREF_FRAME, 0},
193

John Koleszar's avatar
John Koleszar committed
194
195
  {B_PRED,    INTRA_FRAME,  0},
  {I8X8_PRED, INTRA_FRAME,  0},
196

John Koleszar's avatar
John Koleszar committed
197
198
199
200
  /* compound prediction modes */
  {ZEROMV,    LAST_FRAME,   GOLDEN_FRAME},
  {NEARESTMV, LAST_FRAME,   GOLDEN_FRAME},
  {NEARMV,    LAST_FRAME,   GOLDEN_FRAME},
201

John Koleszar's avatar
John Koleszar committed
202
203
204
  {ZEROMV,    ALTREF_FRAME, LAST_FRAME},
  {NEARESTMV, ALTREF_FRAME, LAST_FRAME},
  {NEARMV,    ALTREF_FRAME, LAST_FRAME},
205

John Koleszar's avatar
John Koleszar committed
206
207
208
  {ZEROMV,    GOLDEN_FRAME, ALTREF_FRAME},
  {NEARESTMV, GOLDEN_FRAME, ALTREF_FRAME},
  {NEARMV,    GOLDEN_FRAME, ALTREF_FRAME},
209

John Koleszar's avatar
John Koleszar committed
210
211
212
  {NEWMV,     LAST_FRAME,   GOLDEN_FRAME},
  {NEWMV,     ALTREF_FRAME, LAST_FRAME  },
  {NEWMV,     GOLDEN_FRAME, ALTREF_FRAME},
213

John Koleszar's avatar
John Koleszar committed
214
215
216
  {SPLITMV,   LAST_FRAME,   GOLDEN_FRAME},
  {SPLITMV,   ALTREF_FRAME, LAST_FRAME  },
  {SPLITMV,   GOLDEN_FRAME, ALTREF_FRAME}
John Koleszar's avatar
John Koleszar committed
217
};
218
#endif
John Koleszar's avatar
John Koleszar committed
219
220

static void fill_token_costs(
John Koleszar's avatar
John Koleszar committed
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
  unsigned int (*c)[COEF_BANDS] [PREV_COEF_CONTEXTS] [MAX_ENTROPY_TOKENS],
  const vp8_prob(*p)[COEF_BANDS] [PREV_COEF_CONTEXTS] [ENTROPY_NODES],
  int block_type_counts) {
  int i, j, k;

  for (i = 0; i < block_type_counts; i++)
    for (j = 0; j < COEF_BANDS; j++)
      for (k = 0; k < PREV_COEF_CONTEXTS; k++) {
        if (k == 0 && ((j > 0 && i > 0) || (j > 1 && i == 0)))
          vp8_cost_tokens_skip((int *)(c [i][j][k]),
                               p [i][j][k],
                               vp8_coef_tree);
        else
          vp8_cost_tokens((int *)(c [i][j][k]),
                          p [i][j][k],
                          vp8_coef_tree);
      }
John Koleszar's avatar
John Koleszar committed
238
239
}

240

Paul Wilkins's avatar
Paul Wilkins committed
241
static int rd_iifactor [ 32 ] =  {    4,   4,   3,   2,   1,   0,   0,   0,
John Koleszar's avatar
John Koleszar committed
242
243
244
245
246
                                      0,   0,   0,   0,   0,   0,   0,   0,
                                      0,   0,   0,   0,   0,   0,   0,   0,
                                      0,   0,   0,   0,   0,   0,   0,   0,
                                 };

247
// 3* dc_qlookup[Q]*dc_qlookup[Q];
248

249
/* values are now correlated to quantizer */
Paul Wilkins's avatar
Paul Wilkins committed
250
251
252
static int sad_per_bit16lut[QINDEX_RANGE];
static int sad_per_bit4lut[QINDEX_RANGE];

John Koleszar's avatar
John Koleszar committed
253
254
255
256
257
258
259
260
261
262
263
void vp8_init_me_luts() {
  int i;

  // Initialize the sad lut tables using a formulaic calculation for now
  // This is to make it easier to resolve the impact of experimental changes
  // to the quantizer tables.
  for (i = 0; i < QINDEX_RANGE; i++) {
    sad_per_bit16lut[i] =
      (int)((0.0418 * vp8_convert_qindex_to_q(i)) + 2.4107);
    sad_per_bit4lut[i] = (int)((0.063 * vp8_convert_qindex_to_q(i)) + 2.742);
  }
Paul Wilkins's avatar
Paul Wilkins committed
264
}
John Koleszar's avatar
John Koleszar committed
265

John Koleszar's avatar
John Koleszar committed
266
267
int compute_rd_mult(int qindex) {
  int q;
268

John Koleszar's avatar
John Koleszar committed
269
270
  q = vp8_dc_quant(qindex, 0);
  return (11 * q * q) >> 6;
271
272
}

John Koleszar's avatar
John Koleszar committed
273
274
275
void vp8cx_initialize_me_consts(VP8_COMP *cpi, int QIndex) {
  cpi->mb.sadperbit16 =  sad_per_bit16lut[QIndex];
  cpi->mb.sadperbit4  =  sad_per_bit4lut[QIndex];
John Koleszar's avatar
John Koleszar committed
276
277
}

278

John Koleszar's avatar
John Koleszar committed
279
void vp8_initialize_rd_consts(VP8_COMP *cpi, int QIndex) {
280
  int q, i;
John Koleszar's avatar
John Koleszar committed
281

John Koleszar's avatar
John Koleszar committed
282
  vp8_clear_system_state();  // __asm emms;
John Koleszar's avatar
John Koleszar committed
283

John Koleszar's avatar
John Koleszar committed
284
285
286
287
288
  // Further tests required to see if optimum is different
  // for key frames, golden frames and arf frames.
  // if (cpi->common.refresh_golden_frame ||
  //     cpi->common.refresh_alt_ref_frame)
  QIndex = (QIndex < 0) ? 0 : ((QIndex > MAXQ) ? MAXQ : QIndex);
289

John Koleszar's avatar
John Koleszar committed
290
  cpi->RDMULT = compute_rd_mult(QIndex);
John Koleszar's avatar
John Koleszar committed
291

John Koleszar's avatar
John Koleszar committed
292
293
294
  // Extend rate multiplier along side quantizer zbin increases
  if (cpi->zbin_over_quant  > 0) {
    double oq_factor;
295

John Koleszar's avatar
John Koleszar committed
296
297
298
299
300
    // Experimental code using the same basic equation as used for Q above
    // The units of cpi->zbin_over_quant are 1/128 of Q bin size
    oq_factor = 1.0 + ((double)0.0015625 * cpi->zbin_over_quant);
    cpi->RDMULT = (int)((double)cpi->RDMULT * oq_factor * oq_factor);
  }
John Koleszar's avatar
John Koleszar committed
301

John Koleszar's avatar
John Koleszar committed
302
303
304
305
306
307
308
  if (cpi->pass == 2 && (cpi->common.frame_type != KEY_FRAME)) {
    if (cpi->twopass.next_iiratio > 31)
      cpi->RDMULT += (cpi->RDMULT * rd_iifactor[31]) >> 4;
    else
      cpi->RDMULT +=
        (cpi->RDMULT * rd_iifactor[cpi->twopass.next_iiratio]) >> 4;
  }
John Koleszar's avatar
John Koleszar committed
309

John Koleszar's avatar
John Koleszar committed
310
311
  if (cpi->RDMULT < 7)
    cpi->RDMULT = 7;
312

John Koleszar's avatar
John Koleszar committed
313
314
  cpi->mb.errorperbit = (cpi->RDMULT / 110);
  cpi->mb.errorperbit += (cpi->mb.errorperbit == 0);
315

John Koleszar's avatar
John Koleszar committed
316
  vp8_set_speed_features(cpi);
John Koleszar's avatar
John Koleszar committed
317

John Koleszar's avatar
John Koleszar committed
318
319
320
  q = (int)pow(vp8_dc_quant(QIndex, 0) >> 2, 1.25);
  q = q << 2;
  cpi->RDMULT = cpi->RDMULT << 4;
Paul Wilkins's avatar
Paul Wilkins committed
321

John Koleszar's avatar
John Koleszar committed
322
323
  if (q < 8)
    q = 8;
324

John Koleszar's avatar
John Koleszar committed
325
326
327
  if (cpi->RDMULT > 1000) {
    cpi->RDDIV = 1;
    cpi->RDMULT /= 100;
John Koleszar's avatar
John Koleszar committed
328

John Koleszar's avatar
John Koleszar committed
329
330
331
332
333
334
    for (i = 0; i < MAX_MODES; i++) {
      if (cpi->sf.thresh_mult[i] < INT_MAX) {
        cpi->rd_threshes[i] = cpi->sf.thresh_mult[i] * q / 100;
      } else {
        cpi->rd_threshes[i] = INT_MAX;
      }
John Koleszar's avatar
John Koleszar committed
335

John Koleszar's avatar
John Koleszar committed
336
      cpi->rd_baseline_thresh[i] = cpi->rd_threshes[i];
John Koleszar's avatar
John Koleszar committed
337
    }
John Koleszar's avatar
John Koleszar committed
338
339
  } else {
    cpi->RDDIV = 100;
John Koleszar's avatar
John Koleszar committed
340

John Koleszar's avatar
John Koleszar committed
341
342
343
344
345
346
    for (i = 0; i < MAX_MODES; i++) {
      if (cpi->sf.thresh_mult[i] < (INT_MAX / q)) {
        cpi->rd_threshes[i] = cpi->sf.thresh_mult[i] * q;
      } else {
        cpi->rd_threshes[i] = INT_MAX;
      }
John Koleszar's avatar
John Koleszar committed
347

John Koleszar's avatar
John Koleszar committed
348
      cpi->rd_baseline_thresh[i] = cpi->rd_threshes[i];
John Koleszar's avatar
John Koleszar committed
349
    }
John Koleszar's avatar
John Koleszar committed
350
  }
John Koleszar's avatar
John Koleszar committed
351

John Koleszar's avatar
John Koleszar committed
352
  fill_token_costs(
353
    cpi->mb.token_costs[TX_4X4],
John Koleszar's avatar
John Koleszar committed
354
355
    (const vp8_prob( *)[8][PREV_COEF_CONTEXTS][11]) cpi->common.fc.coef_probs,
    BLOCK_TYPES);
356
357
358
359
360
  fill_token_costs(
    cpi->mb.hybrid_token_costs[TX_4X4],
    (const vp8_prob( *)[8][PREV_COEF_CONTEXTS][11])
    cpi->common.fc.hybrid_coef_probs,
    BLOCK_TYPES);
John Koleszar's avatar
John Koleszar committed
361

John Koleszar's avatar
John Koleszar committed
362
  fill_token_costs(
363
    cpi->mb.token_costs[TX_8X8],
John Koleszar's avatar
John Koleszar committed
364
365
    (const vp8_prob( *)[8][PREV_COEF_CONTEXTS][11]) cpi->common.fc.coef_probs_8x8,
    BLOCK_TYPES_8X8);
366
367
368
369
370
  fill_token_costs(
    cpi->mb.hybrid_token_costs[TX_8X8],
    (const vp8_prob( *)[8][PREV_COEF_CONTEXTS][11])
    cpi->common.fc.hybrid_coef_probs_8x8,
    BLOCK_TYPES_8X8);
371

Daniel Kang's avatar
Daniel Kang committed
372
  fill_token_costs(
373
    cpi->mb.token_costs[TX_16X16],
Daniel Kang's avatar
Daniel Kang committed
374
375
    (const vp8_prob(*)[8][PREV_COEF_CONTEXTS][11]) cpi->common.fc.coef_probs_16x16,
    BLOCK_TYPES_16X16);
376
377
378
379
380
  fill_token_costs(
    cpi->mb.hybrid_token_costs[TX_16X16],
    (const vp8_prob(*)[8][PREV_COEF_CONTEXTS][11])
    cpi->common.fc.hybrid_coef_probs_16x16,
    BLOCK_TYPES_16X16);
Daniel Kang's avatar
Daniel Kang committed
381

John Koleszar's avatar
John Koleszar committed
382
383
384
  /*rough estimate for costing*/
  cpi->common.kf_ymode_probs_index = cpi->common.base_qindex >> 4;
  vp8_init_mode_costs(cpi);
John Koleszar's avatar
John Koleszar committed
385

386
387
388
389
390
391
392
393
394
  if (cpi->common.frame_type != KEY_FRAME)
  {
    vp8_build_nmv_cost_table(
        cpi->mb.nmvjointcost,
        cpi->mb.e_mbd.allow_high_precision_mv ?
        cpi->mb.nmvcost_hp : cpi->mb.nmvcost,
        &cpi->common.fc.nmvc,
        cpi->mb.e_mbd.allow_high_precision_mv, 1, 1);
  }
John Koleszar's avatar
John Koleszar committed
395
396
}

John Koleszar's avatar
John Koleszar committed
397
398
void vp8_auto_select_speed(VP8_COMP *cpi) {
  int milliseconds_for_compress = (int)(1000000 / cpi->oxcf.frame_rate);
John Koleszar's avatar
John Koleszar committed
399

John Koleszar's avatar
John Koleszar committed
400
  milliseconds_for_compress = milliseconds_for_compress * (16 - cpi->oxcf.cpu_used) / 16;
John Koleszar's avatar
John Koleszar committed
401

John Koleszar's avatar
John Koleszar committed
402
403
404
405
406
407
408
  /*
  // this is done during parameter valid check
  if( cpi->oxcf.cpu_used > 16)
      cpi->oxcf.cpu_used = 16;
  if( cpi->oxcf.cpu_used < -16)
      cpi->oxcf.cpu_used = -16;
  */
John Koleszar's avatar
John Koleszar committed
409

410
411
412
  if (cpi->avg_pick_mode_time < milliseconds_for_compress &&
      (cpi->avg_encode_time - cpi->avg_pick_mode_time) <
      milliseconds_for_compress) {
John Koleszar's avatar
John Koleszar committed
413
414
415
416
417
418
419
420
421
422
    if (cpi->avg_pick_mode_time == 0) {
      cpi->Speed = 4;
    } else {
      if (milliseconds_for_compress * 100 < cpi->avg_encode_time * 95) {
        cpi->Speed          += 2;
        cpi->avg_pick_mode_time = 0;
        cpi->avg_encode_time = 0;

        if (cpi->Speed > 16) {
          cpi->Speed = 16;
John Koleszar's avatar
John Koleszar committed
423
        }
John Koleszar's avatar
John Koleszar committed
424
      }
John Koleszar's avatar
John Koleszar committed
425

426
427
      if (milliseconds_for_compress * 100 >
          cpi->avg_encode_time * auto_speed_thresh[cpi->Speed]) {
John Koleszar's avatar
John Koleszar committed
428
429
430
        cpi->Speed          -= 1;
        cpi->avg_pick_mode_time = 0;
        cpi->avg_encode_time = 0;
John Koleszar's avatar
John Koleszar committed
431

John Koleszar's avatar
John Koleszar committed
432
433
434
        // In real-time mode, cpi->speed is in [4, 16].
        if (cpi->Speed < 4) {      // if ( cpi->Speed < 0 )
          cpi->Speed = 4;        // cpi->Speed = 0;
John Koleszar's avatar
John Koleszar committed
435
        }
John Koleszar's avatar
John Koleszar committed
436
      }
John Koleszar's avatar
John Koleszar committed
437
    }
John Koleszar's avatar
John Koleszar committed
438
439
  } else {
    cpi->Speed += 4;
John Koleszar's avatar
John Koleszar committed
440

John Koleszar's avatar
John Koleszar committed
441
442
    if (cpi->Speed > 16)
      cpi->Speed = 16;
John Koleszar's avatar
John Koleszar committed
443
444


John Koleszar's avatar
John Koleszar committed
445
446
447
    cpi->avg_pick_mode_time = 0;
    cpi->avg_encode_time = 0;
  }
John Koleszar's avatar
John Koleszar committed
448
449
}

450
451
int vp8_block_error_c(short *coeff, short *dqcoeff, int block_size) {
  int i, error = 0;
John Koleszar's avatar
John Koleszar committed
452

453
  for (i = 0; i < block_size; i++) {
John Koleszar's avatar
John Koleszar committed
454
455
456
    int this_diff = coeff[i] - dqcoeff[i];
    error += this_diff * this_diff;
  }
John Koleszar's avatar
John Koleszar committed
457

John Koleszar's avatar
John Koleszar committed
458
  return error;
John Koleszar's avatar
John Koleszar committed
459
460
}

John Koleszar's avatar
John Koleszar committed
461
462
463
464
465
int vp8_mbblock_error_c(MACROBLOCK *mb, int dc) {
  BLOCK  *be;
  BLOCKD *bd;
  int i, j;
  int berror, error = 0;
John Koleszar's avatar
John Koleszar committed
466

John Koleszar's avatar
John Koleszar committed
467
468
469
  for (i = 0; i < 16; i++) {
    be = &mb->block[i];
    bd = &mb->e_mbd.block[i];
John Koleszar's avatar
John Koleszar committed
470

John Koleszar's avatar
John Koleszar committed
471
    berror = 0;
John Koleszar's avatar
John Koleszar committed
472

John Koleszar's avatar
John Koleszar committed
473
474
475
    for (j = dc; j < 16; j++) {
      int this_diff = be->coeff[j] - bd->dqcoeff[j];
      berror += this_diff * this_diff;
John Koleszar's avatar
John Koleszar committed
476
477
    }

John Koleszar's avatar
John Koleszar committed
478
479
480
481
    error += berror;
  }

  return error;
John Koleszar's avatar
John Koleszar committed
482
483
}

John Koleszar's avatar
John Koleszar committed
484
485
486
int vp8_mbuverror_c(MACROBLOCK *mb) {
  BLOCK  *be;
  BLOCKD *bd;
John Koleszar's avatar
John Koleszar committed
487

488
  int i, error = 0;
John Koleszar's avatar
John Koleszar committed
489

John Koleszar's avatar
John Koleszar committed
490
491
492
  for (i = 16; i < 24; i++) {
    be = &mb->block[i];
    bd = &mb->e_mbd.block[i];
John Koleszar's avatar
John Koleszar committed
493

494
    error += vp8_block_error_c(be->coeff, bd->dqcoeff, 16);
John Koleszar's avatar
John Koleszar committed
495
  }
John Koleszar's avatar
John Koleszar committed
496

John Koleszar's avatar
John Koleszar committed
497
  return error;
John Koleszar's avatar
John Koleszar committed
498
499
}

500
int vp8_uvsse(MACROBLOCK *x) {
John Koleszar's avatar
John Koleszar committed
501
502
503
504
505
506
507
  unsigned char *uptr, *vptr;
  unsigned char *upred_ptr = (*(x->block[16].base_src) + x->block[16].src);
  unsigned char *vpred_ptr = (*(x->block[20].base_src) + x->block[20].src);
  int uv_stride = x->block[16].src_stride;

  unsigned int sse1 = 0;
  unsigned int sse2 = 0;
508
509
  int mv_row = x->e_mbd.mode_info_context->mbmi.mv[0].as_mv.row;
  int mv_col = x->e_mbd.mode_info_context->mbmi.mv[0].as_mv.col;
John Koleszar's avatar
John Koleszar committed
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
  int offset;
  int pre_stride = x->e_mbd.block[16].pre_stride;

  if (mv_row < 0)
    mv_row -= 1;
  else
    mv_row += 1;

  if (mv_col < 0)
    mv_col -= 1;
  else
    mv_col += 1;

  mv_row /= 2;
  mv_col /= 2;

  offset = (mv_row >> 3) * pre_stride + (mv_col >> 3);
  uptr = x->e_mbd.pre.u_buffer + offset;
  vptr = x->e_mbd.pre.v_buffer + offset;

  if ((mv_row | mv_col) & 7) {
531
532
533
534
    vp8_sub_pixel_variance8x8(uptr, pre_stride, (mv_col & 7) << 1,
                              (mv_row & 7) << 1, upred_ptr, uv_stride, &sse2);
    vp8_sub_pixel_variance8x8(vptr, pre_stride, (mv_col & 7) << 1,
                              (mv_row & 7) << 1, vpred_ptr, uv_stride, &sse1);
John Koleszar's avatar
John Koleszar committed
535
536
    sse2 += sse1;
  } else {
537
538
    vp8_variance8x8(uptr, pre_stride, upred_ptr, uv_stride, &sse2);
    vp8_variance8x8(vptr, pre_stride, vpred_ptr, uv_stride, &sse1);
John Koleszar's avatar
John Koleszar committed
539
540
541
    sse2 += sse1;
  }
  return sse2;
John Koleszar's avatar
John Koleszar committed
542
543
544

}

545
static int cost_coeffs_2x2(MACROBLOCK *mb,
546
                           BLOCKD *b, PLANE_TYPE type,
547
                           ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l) {
548
  int c = (type == PLANE_TYPE_Y_NO_DC); /* start at coef 0, unless Y with Y2 */
549
550
551
552
553
554
555
556
557
558
559
  int eob = b->eob;
  int pt;    /* surrounding block/prev coef predictor */
  int cost = 0;
  short *qcoeff_ptr = b->qcoeff;

  VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l);
  assert(eob <= 4);

  for (; c < eob; c++) {
    int v = qcoeff_ptr[vp8_default_zig_zag1d[c]];
    int t = vp8_dct_value_tokens_ptr[v].Token;
560
    cost += mb->token_costs[TX_8X8][type][vp8_coef_bands[c]][pt][t];
561
562
563
564
565
    cost += vp8_dct_value_cost_ptr[v];
    pt = vp8_prev_token_class[t];
  }

  if (c < 4)
566
    cost += mb->token_costs[TX_8X8][type][vp8_coef_bands[c]]
567
568
569
570
571
572
573
            [pt] [DCT_EOB_TOKEN];

  pt = (c != !type); // is eob first coefficient;
  *a = *l = pt;
  return cost;
}

574
static int cost_coeffs(MACROBLOCK *mb, BLOCKD *b, PLANE_TYPE type,
Daniel Kang's avatar
Daniel Kang committed
575
                       ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l,
576
                       int tx_size) {
Daniel Kang's avatar
Daniel Kang committed
577
  const int eob = b->eob;
578
  int c = (type == PLANE_TYPE_Y_NO_DC); /* start at coef 0, unless Y with Y2 */
579
  int cost = 0, default_eob, seg_eob;
Daniel Kang's avatar
Daniel Kang committed
580
581
  int pt;                     /* surrounding block/prev coef predictor */
  int const *scan, *band;
John Koleszar's avatar
John Koleszar committed
582
  short *qcoeff_ptr = b->qcoeff;
583
584
585
  MACROBLOCKD *xd = &mb->e_mbd;
  MB_MODE_INFO *mbmi = &mb->e_mbd.mode_info_context->mbmi;
  TX_TYPE tx_type = DCT_DCT;
Paul Wilkins's avatar
Paul Wilkins committed
586
  int segment_id = mbmi->segment_id;
587

588
  switch (tx_size) {
Daniel Kang's avatar
Daniel Kang committed
589
590
591
592
    case TX_4X4:
      scan = vp8_default_zig_zag1d;
      band = vp8_coef_bands;
      default_eob = 16;
Deb Mukherjee's avatar
Deb Mukherjee committed
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
      if (type == PLANE_TYPE_Y_WITH_DC) {
        tx_type = get_tx_type_4x4(xd, b);
        if (tx_type != DCT_DCT) {
          switch (tx_type) {
            case ADST_DCT:
              scan = vp8_row_scan;
              break;

            case DCT_ADST:
              scan = vp8_col_scan;
              break;

            default:
              scan = vp8_default_zig_zag1d;
              break;
          }
609
        }
Daniel Kang's avatar
Daniel Kang committed
610
      }
Deb Mukherjee's avatar
Deb Mukherjee committed
611

Daniel Kang's avatar
Daniel Kang committed
612
613
614
615
616
      break;
    case TX_8X8:
      scan = vp8_default_zig_zag1d_8x8;
      band = vp8_coef_bands_8x8;
      default_eob = 64;
Deb Mukherjee's avatar
Deb Mukherjee committed
617
      if (type == PLANE_TYPE_Y_WITH_DC) {
618
619
        BLOCKD *bb;
        int ib = (b - xd->block);
620
621
622
        if (ib < 16) {
          ib = (ib & 8) + ((ib & 4) >> 1);
          bb = xd->block + ib;
Deb Mukherjee's avatar
Deb Mukherjee committed
623
          tx_type = get_tx_type_8x8(xd, bb);
624
        }
625
      }
Daniel Kang's avatar
Daniel Kang committed
626
627
628
629
630
      break;
    case TX_16X16:
      scan = vp8_default_zig_zag1d_16x16;
      band = vp8_coef_bands_16x16;
      default_eob = 256;
Deb Mukherjee's avatar
Deb Mukherjee committed
631
632
633
      if (type == PLANE_TYPE_Y_WITH_DC) {
        tx_type = get_tx_type_16x16(xd, b);
      }
Daniel Kang's avatar
Daniel Kang committed
634
635
636
637
      break;
    default:
      break;
  }
638
639
640
641
642
  if (segfeature_active(&mb->e_mbd, segment_id, SEG_LVL_EOB))
    seg_eob = get_segdata(&mb->e_mbd, segment_id, SEG_LVL_EOB);
  else
    seg_eob = default_eob;

643
644
  VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l);

645
646
647
648
649
650
651
652
653
654
655
  if (tx_type != DCT_DCT) {
    for (; c < eob; c++) {
      int v = qcoeff_ptr[scan[c]];
      int t = vp8_dct_value_tokens_ptr[v].Token;
      cost += mb->hybrid_token_costs[tx_size][type][band[c]][pt][t];
      cost += vp8_dct_value_cost_ptr[v];
      pt = vp8_prev_token_class[t];
    }
    if (c < seg_eob)
      cost += mb->hybrid_token_costs[tx_size][type][band[c]]
          [pt][DCT_EOB_TOKEN];
656
  } else {
657
658
659
660
661
662
663
664
665
666
    for (; c < eob; c++) {
      int v = qcoeff_ptr[scan[c]];
      int t = vp8_dct_value_tokens_ptr[v].Token;
      cost += mb->token_costs[tx_size][type][band[c]][pt][t];
      cost += vp8_dct_value_cost_ptr[v];
      pt = vp8_prev_token_class[t];
    }
    if (c < seg_eob)
      cost += mb->token_costs[tx_size][type][band[c]]
          [pt][DCT_EOB_TOKEN];
667
668
669
670
671
672
673
  }

  pt = (c != !type); // is eob first coefficient;
  *a = *l = pt;
  return cost;
}

John Koleszar's avatar
John Koleszar committed
674
675
676
static int vp8_rdcost_mby(MACROBLOCK *mb) {
  int cost = 0;
  int b;
Paul Wilkins's avatar
Paul Wilkins committed
677
  MACROBLOCKD *xd = &mb->e_mbd;
John Koleszar's avatar
John Koleszar committed
678
679
680
  ENTROPY_CONTEXT_PLANES t_above, t_left;
  ENTROPY_CONTEXT *ta;
  ENTROPY_CONTEXT *tl;
681

Paul Wilkins's avatar
Paul Wilkins committed
682
683
  vpx_memcpy(&t_above, xd->above_context, sizeof(ENTROPY_CONTEXT_PLANES));
  vpx_memcpy(&t_left, xd->left_context, sizeof(ENTROPY_CONTEXT_PLANES));
John Koleszar's avatar
John Koleszar committed
684

John Koleszar's avatar
John Koleszar committed
685
686
  ta = (ENTROPY_CONTEXT *)&t_above;
  tl = (ENTROPY_CONTEXT *)&t_left;
John Koleszar's avatar
John Koleszar committed
687

John Koleszar's avatar
John Koleszar committed
688
  for (b = 0; b < 16; b++)
Paul Wilkins's avatar
Paul Wilkins committed
689
    cost += cost_coeffs(mb, xd->block + b, PLANE_TYPE_Y_NO_DC,
Daniel Kang's avatar
Daniel Kang committed
690
691
                        ta + vp8_block2above[b], tl + vp8_block2left[b],
                        TX_4X4);
John Koleszar's avatar
John Koleszar committed
692

Paul Wilkins's avatar
Paul Wilkins committed
693
  cost += cost_coeffs(mb, xd->block + 24, PLANE_TYPE_Y2,
Daniel Kang's avatar
Daniel Kang committed
694
695
                      ta + vp8_block2above[24], tl + vp8_block2left[24],
                      TX_4X4);
John Koleszar's avatar
John Koleszar committed
696

John Koleszar's avatar
John Koleszar committed
697
  return cost;
John Koleszar's avatar
John Koleszar committed
698
699
}

700
701
702
703
704
static void macro_block_yrd_4x4(MACROBLOCK *mb,
                                int *Rate,
                                int *Distortion,
                                const VP8_ENCODER_RTCD *rtcd,
                                int *skippable) {
John Koleszar's avatar
John Koleszar committed
705
  int b;
Paul Wilkins's avatar
Paul Wilkins committed
706
  MACROBLOCKD *const xd = &mb->e_mbd;
John Koleszar's avatar
John Koleszar committed
707
  BLOCK   *const mb_y2 = mb->block + 24;
Paul Wilkins's avatar
Paul Wilkins committed
708
  BLOCKD *const x_y2  = xd->block + 24;
John Koleszar's avatar
John Koleszar committed
709
710
711
  short *Y2DCPtr = mb_y2->src_diff;
  BLOCK *beptr;
  int d;
712

John Koleszar's avatar
John Koleszar committed
713
714
715
  ENCODEMB_INVOKE(&rtcd->encodemb, submby)(
    mb->src_diff,
    *(mb->block[0].base_src),
Paul Wilkins's avatar
Paul Wilkins committed
716
    xd->predictor,
John Koleszar's avatar
John Koleszar committed
717
    mb->block[0].src_stride);
718

John Koleszar's avatar
John Koleszar committed
719
720
721
722
723
724
725
726
727
728
729
730
  // Fdct and building the 2nd order block
  for (beptr = mb->block; beptr < mb->block + 16; beptr += 2) {
    mb->vp8_short_fdct8x4(beptr->src_diff, beptr->coeff, 32);
    *Y2DCPtr++ = beptr->coeff[0];
    *Y2DCPtr++ = beptr->coeff[16];
  }

  // 2nd order fdct
  mb->short_walsh4x4(mb_y2->src_diff, mb_y2->coeff, 8);

  // Quantization
  for (b = 0; b < 16; b++) {
731
    mb->quantize_b_4x4(&mb->block[b], &xd->block[b]);
John Koleszar's avatar
John Koleszar committed
732
  }
733

John Koleszar's avatar
John Koleszar committed
734
  // DC predication and Quantization of 2nd Order block
735
  mb->quantize_b_4x4(mb_y2, x_y2);
736

John Koleszar's avatar
John Koleszar committed
737
738
  // Distortion
  d = ENCODEMB_INVOKE(&rtcd->encodemb, mberr)(mb, 1);
739

740
  d += ENCODEMB_INVOKE(&rtcd->encodemb, berr)(mb_y2->coeff, x_y2->dqcoeff, 16);
741

John Koleszar's avatar
John Koleszar committed
742
743
744
  *Distortion = (d >> 2);
  // rate
  *Rate = vp8_rdcost_mby(mb);
745
  *skippable = mby_is_skippable_4x4(&mb->e_mbd, 1);
746
}
John Koleszar's avatar
John Koleszar committed
747

Ronald S. Bultje's avatar
Ronald S. Bultje committed
748
static int vp8_rdcost_mby_8x8(MACROBLOCK *mb, int backup) {
John Koleszar's avatar
John Koleszar committed
749
750
  int cost = 0;
  int b;
Paul Wilkins's avatar
Paul Wilkins committed
751
  MACROBLOCKD *xd = &mb->e_mbd;
John Koleszar's avatar
John Koleszar committed
752
753
754
755
  ENTROPY_CONTEXT_PLANES t_above, t_left;
  ENTROPY_CONTEXT *ta;
  ENTROPY_CONTEXT *tl;

Ronald S. Bultje's avatar
Ronald S. Bultje committed
756
757
758
  if (backup) {
    vpx_memcpy(&t_above,xd->above_context, sizeof(ENTROPY_CONTEXT_PLANES));
    vpx_memcpy(&t_left, xd->left_context, sizeof(ENTROPY_CONTEXT_PLANES));
John Koleszar's avatar
John Koleszar committed
759

Ronald S. Bultje's avatar
Ronald S. Bultje committed
760
761
762
763
764
765
    ta = (ENTROPY_CONTEXT *)&t_above;
    tl = (ENTROPY_CONTEXT *)&t_left;
  } else {
    ta = (ENTROPY_CONTEXT *)mb->e_mbd.above_context;
    tl = (ENTROPY_CONTEXT *)mb->e_mbd.left_context;
  }
John Koleszar's avatar
John Koleszar committed
766
767

  for (b = 0; b < 16; b += 4)
Paul Wilkins's avatar
Paul Wilkins committed
768
    cost += cost_coeffs(mb, xd->block + b, PLANE_TYPE_Y_NO_DC,
Daniel Kang's avatar
Daniel Kang committed
769
770
                        ta + vp8_block2above_8x8[b], tl + vp8_block2left_8x8[b],
                        TX_8X8);
John Koleszar's avatar
John Koleszar committed
771

Paul Wilkins's avatar
Paul Wilkins committed
772
  cost += cost_coeffs_2x2(mb, xd->block + 24, PLANE_TYPE_Y2,
John Koleszar's avatar
John Koleszar committed
773
774
                          ta + vp8_block2above[24], tl + vp8_block2left[24]);
  return cost;
775
776
}

John Koleszar's avatar
John Koleszar committed
777
778
779
static void macro_block_yrd_8x8(MACROBLOCK *mb,
                                int *Rate,
                                int *Distortion,
780
781
                                const VP8_ENCODER_RTCD *rtcd,
                                int *skippable) {
Paul Wilkins's avatar
Paul Wilkins committed
782
  MACROBLOCKD *const xd = &mb->e_mbd;
John Koleszar's avatar
John Koleszar committed
783
  BLOCK   *const mb_y2 = mb->block + 24;
Paul Wilkins's avatar
Paul Wilkins committed
784
  BLOCKD *const x_y2  = xd->block + 24;
John Koleszar's avatar
John Koleszar committed
785
786
787
788
789
  int d;

  ENCODEMB_INVOKE(&rtcd->encodemb, submby)(
    mb->src_diff,
    *(mb->block[0].base_src),
Paul Wilkins's avatar
Paul Wilkins committed
790
    xd->predictor,
John Koleszar's avatar
John Koleszar committed
791
792
793
794
795
796
797
798
799
800
    mb->block[0].src_stride);

  vp8_transform_mby_8x8(mb);
  vp8_quantize_mby_8x8(mb);

  /* remove 1st order dc to properly combine 1st/2nd order distortion */
  mb->coeff[0] = 0;
  mb->coeff[64] = 0;
  mb->coeff[128] = 0;
  mb->coeff[192] = 0;
Paul Wilkins's avatar
Paul Wilkins committed
801
802
803
804
  xd->dqcoeff[0] = 0;
  xd->dqcoeff[64] = 0;
  xd->dqcoeff[128] = 0;
  xd->dqcoeff[192] = 0;
John Koleszar's avatar
John Koleszar committed
805
806

  d = ENCODEMB_INVOKE(&rtcd->encodemb, mberr)(mb, 0);
807
  d += ENCODEMB_INVOKE(&rtcd->encodemb, berr)(mb_y2->coeff, x_y2->dqcoeff, 16);
John Koleszar's avatar
John Koleszar committed
808
809
810

  *Distortion = (d >> 2);
  // rate
Ronald S. Bultje's avatar
Ronald S. Bultje committed
811
  *Rate = vp8_rdcost_mby_8x8(mb, 1);
812
  *skippable = mby_is_skippable_8x8(&mb->e_mbd, 1);
813
}
814

Daniel Kang's avatar
Daniel Kang committed
815
816
static int vp8_rdcost_mby_16x16(MACROBLOCK *mb) {
  int cost;
Paul Wilkins's avatar
Paul Wilkins committed
817
  MACROBLOCKD *xd = &mb->e_mbd;
Daniel Kang's avatar
Daniel Kang committed
818
819
820
  ENTROPY_CONTEXT_PLANES t_above, t_left;
  ENTROPY_CONTEXT *ta, *tl;

Paul Wilkins's avatar
Paul Wilkins committed
821
822
  vpx_memcpy(&t_above, xd->above_context, sizeof(ENTROPY_CONTEXT_PLANES));
  vpx_memcpy(&t_left, xd->left_context, sizeof(ENTROPY_CONTEXT_PLANES));
Daniel Kang's avatar
Daniel Kang committed
823
824
825
826

  ta = (ENTROPY_CONTEXT *)&t_above;
  tl = (ENTROPY_CONTEXT *)&t_left;

Paul Wilkins's avatar
Paul Wilkins committed
827
  cost = cost_coeffs(mb, xd->block, PLANE_TYPE_Y_WITH_DC, ta, tl, TX_16X16);
Daniel Kang's avatar
Daniel Kang committed
828
829
  return cost;
}
830

Daniel Kang's avatar
Daniel Kang committed
831
static void macro_block_yrd_16x16(MACROBLOCK *mb, int *Rate, int *Distortion,
832
                                  const VP8_ENCODER_RTCD *rtcd, int *skippable) {
Daniel Kang's avatar
Daniel Kang committed
833
  int d;
Deb Mukherjee's avatar
Deb Mukherjee committed
834
835
836
837
  MACROBLOCKD *xd = &mb->e_mbd;
  BLOCKD *b  = &mb->e_mbd.block[0];
  BLOCK  *be = &mb->block[0];
  TX_TYPE tx_type;
Daniel Kang's avatar
Daniel Kang committed
838
839
840
841
842
843
844

  ENCODEMB_INVOKE(&rtcd->encodemb, submby)(
    mb->src_diff,
    *(mb->block[0].base_src),
    mb->e_mbd.predictor,
    mb->block[0].src_stride);

Deb Mukherjee's avatar
Deb Mukherjee committed
845
846
847
  tx_type = get_tx_type_16x16(xd, b);
  if (tx_type != DCT_DCT) {
    vp8_fht_c(be->src_diff, be->coeff, 32, tx_type, 16);
848
849
850
  } else
    vp8_transform_mby_16x16(mb);

Daniel Kang's avatar
Daniel Kang committed
851
  vp8_quantize_mby_16x16(mb);
852
853
854
  // TODO(jingning) is it possible to quickly determine whether to force
  //                trailing coefficients to be zero, instead of running trellis
  //                optimization in the rate-distortion optimization loop?
855
  if (mb->e_mbd.mode_info_context->mbmi.mode < I8X8_PRED)
856
857
    vp8_optimize_mby_16x16(mb, rtcd);

Daniel Kang's avatar
Daniel Kang committed
858
859
860
861
862
  d = ENCODEMB_INVOKE(&rtcd->encodemb, mberr)(mb, 0);

  *Distortion = (d >> 2);
  // rate
  *Rate = vp8_rdcost_mby_16x16(mb);
863
  *skippable = mby_is_skippable_16x16(&mb->e_mbd);
Daniel Kang's avatar
Daniel Kang committed
864
865
}

866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
static void macro_block_yrd(VP8_COMP *cpi, MACROBLOCK *x, int *rate,
                            int *distortion, int *skippable,
                            int64_t txfm_cache[NB_TXFM_MODES]) {
  VP8_COMMON *cm = &cpi->common;
  MB_MODE_INFO *mbmi = &x->e_mbd.mode_info_context->mbmi;

  MACROBLOCKD *xd = &x->e_mbd;
  int can_skip = cm->mb_no_coeff_skip;
  vp8_prob skip_prob = can_skip ? get_pred_prob(cm, xd, PRED_MBSKIP) : 128;
  int s0, s1;
  int r4x4, r4x4s, r8x8, r8x8s, d4x4, d8x8, s4x4, s8x8;
  int64_t rd4x4, rd8x8, rd4x4s, rd8x8s;
  int d16x16, r16x16, r16x16s, s16x16;
  int64_t rd16x16, rd16x16s;

  // FIXME don't do sub x3
  if (skip_prob == 0)
    skip_prob = 1;
  s0 = vp8_cost_bit(skip_prob, 0);
  s1 = vp8_cost_bit(skip_prob, 1);
  macro_block_yrd_16x16(x, &r16x16, &d16x16, IF_RTCD(&cpi->rtcd), &s16x16);
  if (can_skip) {
    if (s16x16) {
      rd16x16 = RDCOST(x->rdmult, x->rddiv, s1, d16x16);
    } else {
      rd16x16 = RDCOST(x->rdmult, x->rddiv, r16x16 + s0, d16x16);
    }
  } else {
    rd16x16 = RDCOST(x->rdmult, x->rddiv, r16x16, d16x16);
  }
  r16x16s = r16x16 + vp8_cost_one(cm->prob_tx[0]) + vp8_cost_one(cm->prob_tx[1]);
  if (can_skip) {
    if (s16x16) {
      rd16x16s = RDCOST(x->rdmult, x->rddiv, s1, d16x16);
    } else {
      rd16x16s = RDCOST(x->rdmult, x->rddiv, r16x16s + s0, d16x16);
    }
  } else {
    rd16x16s = RDCOST(x->rdmult, x->rddiv, r16x16s, d16x16);
  }
  macro_block_yrd_8x8(x, &r8x8, &d8x8, IF_RTCD(&cpi->rtcd), &s8x8);
  if (can_skip) {
    if (s8x8) {
      rd8x8 = RDCOST(x->rdmult, x->rddiv, s1, d8x8);
    } else {
      rd8x8 = RDCOST(x->rdmult, x->rddiv, r8x8 + s0, d8x8);
    }
  } else {
    rd8x8 = RDCOST(x->rdmult, x->rddiv, r8x8, d8x8);
  }
  r8x8s = r8x8 + vp8_cost_one(cm->prob_tx[0]);
  r8x8s += vp8_cost_zero(cm->prob_tx[1]);
  if (can_skip) {
    if (s8x8) {
      rd8x8s = RDCOST(x->rdmult, x->rddiv, s1, d8x8);
    } else {
      rd8x8s = RDCOST(x->rdmult, x->rddiv, r8x8s + s0, d8x8);
    }
  } else {
    rd8x8s = RDCOST(x->rdmult, x->rddiv, r8x8s, d8x8);
  }
  macro_block_yrd_4x4(x, &r4x4, &d4x4, IF_RTCD(&cpi->rtcd), &s4x4);
  if (can_skip) {
    if (s4x4) {
      rd4x4 = RDCOST(x->rdmult, x->rddiv, s1, d4x4);
    } else {
      rd4x4 = RDCOST(x->rdmult, x->rddiv, r4x4 + s0, d4x4);
    }
  } else {
    rd4x4 = RDCOST(x->rdmult, x->rddiv, r4x4, d4x4);
  }
  r4x4s = r4x4 + vp8_cost_zero(cm->prob_tx[0]);
  if (can_skip) {
    if (s4x4) {
      rd4x4s = RDCOST(x->rdmult, x->rddiv, s1, d4x4);
    } else {
      rd4x4s = RDCOST(x->rdmult, x->rddiv, r4x4s + s0, d4x4);
    }
  } else {
    rd4x4s = RDCOST(x->rdmult, x->rddiv, r4x4s, d4x4);
  }

  if ( cpi->common.txfm_mode == ALLOW_16X16 ||
      (cpi->common.txfm_mode == TX_MODE_SELECT &&
       rd16x16s < rd8x8s && rd16x16s < rd4x4s)) {
    mbmi->txfm_size = TX_16X16;
    *skippable = s16x16;
    *distortion = d16x16;
    *rate = (cpi->common.txfm_mode == ALLOW_16X16) ? r16x16 : r16x16s;
  } else
  if ( cpi->common.txfm_mode == ALLOW_8X8 ||
      (cpi->common.txfm_mode == TX_MODE_SELECT && rd8x8s < rd4x4s)) {
    mbmi->txfm_size = TX_8X8;
    *skippable = s8x8;
    *distortion = d8x8;
    *rate = (cpi->common.txfm_mode == ALLOW_8X8) ? r8x8 : r8x8s;
  } else {
    assert(cpi->common.txfm_mode == ONLY_4X4 ||
           (cpi->common.txfm_mode == TX_MODE_SELECT && rd4x4s <= rd8x8s));
    mbmi->txfm_size = TX_4X4;
    *skippable = s4x4;
    *distortion = d4x4;
    *rate = (cpi->common.txfm_mode == ONLY_4X4) ? r4x4 : r4x4s;
  }

  txfm_cache[ONLY_4X4] = rd4x4;
  txfm_cache[ALLOW_8X8] = rd8x8;
  txfm_cache[ALLOW_16X16] = rd16x16;
  if (rd16x16s < rd8x8s && rd16x16s < rd4x4s)
    txfm_cache[TX_MODE_SELECT] = rd16x16s;
  else
    txfm_cache[TX_MODE_SELECT] = rd4x4s < rd8x8s ? rd4x4s : rd8x8s;

}

John Koleszar's avatar
John Koleszar committed
981
982
983
984
985
986
987
static void copy_predictor(unsigned char *dst, const unsigned char *predictor) {
  const unsigned int *p = (const unsigned int *)predictor;
  unsigned int *d = (unsigned int *)dst;
  d[0] = p[0];
  d[4] = p[4];
  d[8] = p[8];
  d[12] = p[12];
988
}
Yaowu Xu's avatar
Yaowu Xu committed
989

Ronald S. Bultje's avatar
Ronald S. Bultje committed
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
#if CONFIG_SUPERBLOCKS
static void super_block_yrd_8x8(MACROBLOCK *x,
                                int *rate,
                                int *distortion,
                                const VP8_ENCODER_RTCD *rtcd, int *skip)
{
  MACROBLOCKD *const xd = &x->e_mbd;
  BLOCK *const by2 = x->block + 24;
  BLOCKD *const bdy2  = xd->block + 24;
  int d = 0, r = 0, n;
  const uint8_t *src = x->src.y_buffer, *dst = xd->dst.y_buffer;
  int src_y_stride = x->src.y_stride, dst_y_stride = xd->dst.y_stride;
  ENTROPY_CONTEXT_PLANES *ta = xd->above_context;
  ENTROPY_CONTEXT_PLANES *tl = xd->left_context;
  ENTROPY_CONTEXT_PLANES t_above[2];
  ENTROPY_CONTEXT_PLANES t_left[2];
  int skippable = 1;

  vpx_memcpy(t_above, xd->above_context, sizeof(t_above));
  vpx_memcpy(t_left, xd->left_context, sizeof(t_left));

  for (n = 0; n < 4; n++) {
    int x_idx = n & 1, y_idx = n >> 1;

    vp8_subtract_mby_s_c(x->src_diff,
                         src + x_idx * 16 + y_idx * 16 * src_y_stride,
                         src_y_stride,
                         dst + x_idx * 16 + y_idx * 16 * dst_y_stride,
                         dst_y_stride);
    vp8_transform_mby_8x8(x);
    vp8_quantize_mby_8x8(x);

    /* remove 1st order dc to properly combine 1st/2nd order distortion */
    x->coeff[  0] = 0;
    x->coeff[ 64] = 0;
    x->coeff[128] = 0;
    x->coeff[192] = 0;
    xd->dqcoeff[  0] = 0;
    xd->dqcoeff[ 64] = 0;
    xd->dqcoeff[128] = 0;
    xd->dqcoeff[192] = 0;

    d += ENCODEMB_INVOKE(&rtcd->encodemb, mberr)(x, 0);
    d += ENCODEMB_INVOKE(&rtcd->encodemb, berr)(by2->coeff, bdy2->dqcoeff, 16);
    xd->above_context = ta + x_idx;
    xd->left_context = tl + y_idx;
    r += vp8_rdcost_mby_8x8(x, 0);
Deb Mukherjee's avatar
Deb Mukherjee committed
1037
    skippable = skippable && mby_is_skippable_8x8(xd, 1);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
  }

  *distortion = (d >> 2);
  *rate       = r;
  if (skip) *skip = skippable;
  xd->above_context = ta;
  xd->left_context = tl;
  vpx_memcpy(xd->above_context, &t_above, sizeof(t_above));
  vpx_memcpy(xd->left_context, &t_left, sizeof(t_left));
}
#endif

John Koleszar's avatar
John Koleszar committed
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
static void copy_predictor_8x8(unsigned char *dst, const unsigned char *predictor) {
  const unsigned int *p = (const unsigned int *)predictor;
  unsigned int *d = (unsigned int *)dst;
  d[0] = p[0];
  d[1] = p[1];
  d[4] = p[4];
  d[5] = p[5];
  d[8] = p[8];
  d[9] = p[9];
  d[12] = p[12];
  d[13] = p[13];
  d[16] = p[16];
  d[17] = p[17];
  d[20] = p[20];
  d[21] = p[21];
  d[24] = p[24];
  d[25] = p[25];
  d[28] = p[28];
  d[29] = p[29];
Yaowu Xu's avatar
Yaowu Xu committed
1069
1070
}

1071
1072
static int64_t rd_pick_intra4x4block(VP8_COMP *cpi, MACROBLOCK *x, BLOCK *be,
                                     BLOCKD *b, B_PREDICTION_MODE *best_mode,
1073
#if CONFIG_COMP_INTRA_PRED
1074
1075
1076
1077
1078
1079
1080
                                     B_PREDICTION_MODE *best_second_mode,
                                     int allow_comp,
#endif
                                     int *bmode_costs,
                                     ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l,
                                     int *bestrate, int *bestratey,
                                     int *bestdistortion) {
John Koleszar's avatar
John Koleszar committed
1081
  B_PREDICTION_MODE mode;
Deb Mukherjee's avatar
Deb Mukherjee committed
1082
  MACROBLOCKD *xd = &x->e_mbd;
Jingning Han's avatar
Jingning Han committed
1083

1084
#if CONFIG_COMP_INTRA_PRED
John Koleszar's avatar
John Koleszar committed
1085
1086
  B_PREDICTION_MODE mode2;
#endif
1087
  int64_t best_rd = INT64_MAX;
John Koleszar's avatar
John Koleszar committed
1088
1089
1090
1091
1092
  int rate = 0;
  int distortion;

  ENTROPY_CONTEXT ta = *a, tempa = *a;
  ENTROPY_CONTEXT tl = *l, templ = *l;
Deb Mukherjee's avatar
Deb Mukherjee committed
1093
1094
  TX_TYPE tx_type = DCT_DCT;
  TX_TYPE best_tx_type = DCT_DCT;
John Koleszar's avatar
John Koleszar committed
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
  /*
   * The predictor buffer is a 2d buffer with a stride of 16.  Create
   * a temp buffer that meets the stride requirements, but we are only
   * interested in the left 4x4 block
   * */
  DECLARE_ALIGNED_ARRAY(16, unsigned char,  best_predictor, 16 * 4);
  DECLARE_ALIGNED_ARRAY(16, short, best_dqcoeff, 16);

  for (mode = B_DC_PRED; mode <= B_HU_PRED; mode++) {
#if CONFIG_COMP_INTRA_PRED
Jingning Han's avatar
Jingning Han committed
1105
1106
    for (mode2 = (allow_comp ? 0 : (B_DC_PRED - 1));
                   mode2 != (allow_comp ? (mode + 1) : 0); mode2++) {
1107
#endif
1108
      int64_t this_rd;
John Koleszar's avatar
John Koleszar committed
1109
      int ratey;
John Koleszar's avatar
John Koleszar committed
1110

1111
      b->bmi.as_mode.first = mode;
John Koleszar's avatar
John Koleszar committed
1112
      rate = bmode_costs[mode];
1113

1114
#if CONFIG_COMP_INTRA_PRED
John Koleszar's avatar
John Koleszar committed
1115
      if (mode2 == (B_PREDICTION_MODE)(B_DC_PRED - 1)) {
1116
#endif
1117
        vp8_intra4x4_predict(b, mode, b->predictor);
1118
#if CONFIG_COMP_INTRA_PRED
John Koleszar's avatar
John Koleszar committed
1119
      } else {
1120
        vp8_comp_intra4x4_predict(b, mode, mode2, b->predictor);
John Koleszar's avatar
John Koleszar committed
1121
1122
1123
1124
1125
        rate += bmode_costs[mode2];
      }
#endif
      ENCODEMB_INVOKE(IF_RTCD(&cpi->rtcd.encodemb), subb)(be, b, 16);

Deb Mukherjee's avatar
Deb Mukherjee committed
1126
1127
1128
1129
1130
      b->bmi.as_mode.first = mode;
      tx_type = get_tx_type_4x4(xd, b);
      if (tx_type != DCT_DCT) {
        vp8_fht_c(be->src_diff, be->coeff, 32, tx_type, 4);
        vp8_ht_quantize_b_4x4(be, b, tx_type);
Jingning Han's avatar
Jingning Han committed
1131
1132
      } else {
        x->vp8_short_fdct4x4(be->src_diff, be->coeff, 32);
1133
        x->quantize_b_4x4(be, b);
Jingning Han's avatar
Jingning Han committed
1134
      }
John Koleszar's avatar
John Koleszar committed
1135

Deb Mukherjee's avatar
Deb Mukherjee committed
1136
1137
      tempa = ta;
      templ = tl;
John Koleszar's avatar
John Koleszar committed
1138

Deb Mukherjee's avatar
Deb Mukherjee committed
1139
1140
1141
1142
      ratey = cost_coeffs(x, b, PLANE_TYPE_Y_WITH_DC, &tempa, &templ, TX_4X4);
      rate += ratey;
      distortion = ENCODEMB_INVOKE(IF_RTCD(&cpi->rtcd.encodemb), berr)(
          be->coeff, b->dqcoeff, 16) >> 2;
Jingning Han's avatar
Jingning Han committed
1143

Deb Mukherjee's avatar
Deb Mukherjee committed
1144
      this_rd = RDCOST(x->rdmult, x->rddiv, rate, distortion);
Jingning Han's avatar
Jingning Han committed
1145

Deb Mukherjee's avatar
Deb Mukherjee committed
1146
1147
1148
1149
1150
1151
1152
      if (this_rd < best_rd) {
        *bestrate = rate;
        *bestratey = ratey;
        *bestdistortion = distortion;
        best_rd = this_rd;
        *best_mode = mode;
        best_tx_type = tx_type;
Jingning Han's avatar
Jingning Han committed
1153

1154
#if CONFIG_COMP_INTRA_PRED
Deb Mukherjee's avatar
Deb Mukherjee committed
1155
        *best_second_mode = mode2;
1156
#endif
Deb Mukherjee's avatar
Deb Mukherjee committed
1157
1158
1159
1160
1161
        *a = tempa;
        *l = templ;
        copy_predictor(best_predictor, b->predictor);
        vpx_memcpy(best_dqcoeff, b->dqcoeff, 32);
      }
Jingning Han's avatar
Jingning Han committed
1162
#if CONFIG_COMP_INTRA_PRED
John Koleszar's avatar
John Koleszar committed
1163
    }
Jingning Han's avatar
Jingning Han committed
1164
#endif
John Koleszar's avatar
John Koleszar committed
1165
1166
  }
  b->bmi.as_mode.first = (B_PREDICTION_MODE)(*best_mode);
1167
#if CONFIG_COMP_INTRA_PRED
John Koleszar's avatar
John Koleszar committed
1168
  b->bmi.as_mode.second = (B_PREDICTION_MODE)(*best_second_mode);
1169
#endif
1170