vp9_rdopt.c 183 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>
16
17
#include "vp9/common/vp9_pragmas.h"

18
19
20
21
22
#include "vp9/encoder/vp9_tokenize.h"
#include "vp9/encoder/vp9_treewriter.h"
#include "vp9/encoder/vp9_onyx_int.h"
#include "vp9/encoder/vp9_modecosts.h"
#include "vp9/encoder/vp9_encodeintra.h"
23
24
25
26
27
28
#include "vp9/common/vp9_entropymode.h"
#include "vp9/common/vp9_reconinter.h"
#include "vp9/common/vp9_reconintra.h"
#include "vp9/common/vp9_reconintra4x4.h"
#include "vp9/common/vp9_findnearmv.h"
#include "vp9/common/vp9_quant_common.h"
29
30
31
32
33
34
#include "vp9/encoder/vp9_encodemb.h"
#include "vp9/encoder/vp9_quantize.h"
#include "vp9/encoder/vp9_variance.h"
#include "vp9/encoder/vp9_mcomp.h"
#include "vp9/encoder/vp9_rdopt.h"
#include "vp9/encoder/vp9_ratectrl.h"
John Koleszar's avatar
John Koleszar committed
35
#include "vpx_mem/vpx_mem.h"
36
37
#include "vp9/common/vp9_systemdependent.h"
#include "vp9/encoder/vp9_encodemv.h"
John Koleszar's avatar
John Koleszar committed
38

39
40
41
#include "vp9/common/vp9_seg_common.h"
#include "vp9/common/vp9_pred_common.h"
#include "vp9/common/vp9_entropy.h"
42
#include "vp9_rtcd.h"
43
#include "vp9/common/vp9_mvref_common.h"
Ronald S. Bultje's avatar
Ronald S. Bultje committed
44
#include "vp9/common/vp9_common.h"
Paul Wilkins's avatar
Paul Wilkins committed
45

John Koleszar's avatar
John Koleszar committed
46
47
#define MAXF(a,b)            (((a) > (b)) ? (a) : (b))

48
49
#define INVALID_MV 0x80008000

50
51
52
/* Factor to weigh the rate for switchable interp filters */
#define SWITCHABLE_INTERP_RATE_FACTOR 1

John Koleszar's avatar
John Koleszar committed
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
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
71
72
};

73
const MODE_DEFINITION vp9_mode_order[MAX_MODES] = {
74
75
  {ZEROMV,    LAST_FRAME,   NONE},
  {DC_PRED,   INTRA_FRAME,  NONE},
John Koleszar's avatar
John Koleszar committed
76

77
78
  {NEARESTMV, LAST_FRAME,   NONE},
  {NEARMV,    LAST_FRAME,   NONE},
John Koleszar's avatar
John Koleszar committed
79

80
81
  {ZEROMV,    GOLDEN_FRAME, NONE},
  {NEARESTMV, GOLDEN_FRAME, NONE},
John Koleszar's avatar
John Koleszar committed
82

83
84
  {ZEROMV,    ALTREF_FRAME, NONE},
  {NEARESTMV, ALTREF_FRAME, NONE},
John Koleszar's avatar
John Koleszar committed
85

86
87
  {NEARMV,    GOLDEN_FRAME, NONE},
  {NEARMV,    ALTREF_FRAME, NONE},
John Koleszar's avatar
John Koleszar committed
88

89
90
91
92
93
94
95
96
  {V_PRED,    INTRA_FRAME,  NONE},
  {H_PRED,    INTRA_FRAME,  NONE},
  {D45_PRED,  INTRA_FRAME,  NONE},
  {D135_PRED, INTRA_FRAME,  NONE},
  {D117_PRED, INTRA_FRAME,  NONE},
  {D153_PRED, INTRA_FRAME,  NONE},
  {D27_PRED,  INTRA_FRAME,  NONE},
  {D63_PRED,  INTRA_FRAME,  NONE},
John Koleszar's avatar
John Koleszar committed
97

98
  {TM_PRED,   INTRA_FRAME,  NONE},
John Koleszar's avatar
John Koleszar committed
99

100
101
102
  {NEWMV,     LAST_FRAME,   NONE},
  {NEWMV,     GOLDEN_FRAME, NONE},
  {NEWMV,     ALTREF_FRAME, NONE},
John Koleszar's avatar
John Koleszar committed
103

104
105
106
  {SPLITMV,   LAST_FRAME,   NONE},
  {SPLITMV,   GOLDEN_FRAME, NONE},
  {SPLITMV,   ALTREF_FRAME, NONE},
107

108
109
  {B_PRED,    INTRA_FRAME,  NONE},
  {I8X8_PRED, INTRA_FRAME,  NONE},
110

John Koleszar's avatar
John Koleszar committed
111
112
113
114
  /* compound prediction modes */
  {ZEROMV,    LAST_FRAME,   GOLDEN_FRAME},
  {NEARESTMV, LAST_FRAME,   GOLDEN_FRAME},
  {NEARMV,    LAST_FRAME,   GOLDEN_FRAME},
115

John Koleszar's avatar
John Koleszar committed
116
117
118
  {ZEROMV,    ALTREF_FRAME, LAST_FRAME},
  {NEARESTMV, ALTREF_FRAME, LAST_FRAME},
  {NEARMV,    ALTREF_FRAME, LAST_FRAME},
119

John Koleszar's avatar
John Koleszar committed
120
121
122
  {ZEROMV,    GOLDEN_FRAME, ALTREF_FRAME},
  {NEARESTMV, GOLDEN_FRAME, ALTREF_FRAME},
  {NEARMV,    GOLDEN_FRAME, ALTREF_FRAME},
123

John Koleszar's avatar
John Koleszar committed
124
125
126
  {NEWMV,     LAST_FRAME,   GOLDEN_FRAME},
  {NEWMV,     ALTREF_FRAME, LAST_FRAME  },
  {NEWMV,     GOLDEN_FRAME, ALTREF_FRAME},
127

John Koleszar's avatar
John Koleszar committed
128
129
  {SPLITMV,   LAST_FRAME,   GOLDEN_FRAME},
  {SPLITMV,   ALTREF_FRAME, LAST_FRAME  },
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
  {SPLITMV,   GOLDEN_FRAME, ALTREF_FRAME},

#if CONFIG_COMP_INTERINTRA_PRED
  /* compound inter-intra prediction */
  {ZEROMV,    LAST_FRAME,   INTRA_FRAME},
  {NEARESTMV, LAST_FRAME,   INTRA_FRAME},
  {NEARMV,    LAST_FRAME,   INTRA_FRAME},
  {NEWMV,     LAST_FRAME,   INTRA_FRAME},

  {ZEROMV,    GOLDEN_FRAME,   INTRA_FRAME},
  {NEARESTMV, GOLDEN_FRAME,   INTRA_FRAME},
  {NEARMV,    GOLDEN_FRAME,   INTRA_FRAME},
  {NEWMV,     GOLDEN_FRAME,   INTRA_FRAME},

  {ZEROMV,    ALTREF_FRAME,   INTRA_FRAME},
  {NEARESTMV, ALTREF_FRAME,   INTRA_FRAME},
  {NEARMV,    ALTREF_FRAME,   INTRA_FRAME},
  {NEWMV,     ALTREF_FRAME,   INTRA_FRAME},
#endif
John Koleszar's avatar
John Koleszar committed
149
150
};

151
152
153
static void fill_token_costs(vp9_coeff_count *c,
                             vp9_coeff_probs *p,
                             int block_type_counts) {
John Koleszar's avatar
John Koleszar committed
154
155
156
157
158
159
  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)))
160
161
          vp9_cost_tokens_skip((int *)(c[i][j][k]),
                               p[i][j][k],
162
                               vp9_coef_tree);
John Koleszar's avatar
John Koleszar committed
163
        else
164
165
          vp9_cost_tokens((int *)(c[i][j][k]),
                          p[i][j][k],
166
                          vp9_coef_tree);
John Koleszar's avatar
John Koleszar committed
167
      }
John Koleszar's avatar
John Koleszar committed
168
169
}

170

171
172
173
174
static int rd_iifactor[32] =  { 4, 4, 3, 2, 1, 0, 0, 0,
                                0, 0, 0, 0, 0, 0, 0, 0,
                                0, 0, 0, 0, 0, 0, 0, 0,
                                0, 0, 0, 0, 0, 0, 0, 0, };
John Koleszar's avatar
John Koleszar committed
175

176
// 3* dc_qlookup[Q]*dc_qlookup[Q];
177

178
/* values are now correlated to quantizer */
Paul Wilkins's avatar
Paul Wilkins committed
179
180
181
static int sad_per_bit16lut[QINDEX_RANGE];
static int sad_per_bit4lut[QINDEX_RANGE];

182
void vp9_init_me_luts() {
John Koleszar's avatar
John Koleszar committed
183
184
185
186
187
188
189
  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] =
190
191
      (int)((0.0418 * vp9_convert_qindex_to_q(i)) + 2.4107);
    sad_per_bit4lut[i] = (int)((0.063 * vp9_convert_qindex_to_q(i)) + 2.742);
John Koleszar's avatar
John Koleszar committed
192
  }
Paul Wilkins's avatar
Paul Wilkins committed
193
}
John Koleszar's avatar
John Koleszar committed
194

195
static int compute_rd_mult(int qindex) {
John Koleszar's avatar
John Koleszar committed
196
  int q;
197

198
  q = vp9_dc_quant(qindex, 0);
John Koleszar's avatar
John Koleszar committed
199
  return (11 * q * q) >> 6;
200
201
}

202
void vp9_initialize_me_consts(VP9_COMP *cpi, int QIndex) {
John Koleszar's avatar
John Koleszar committed
203
204
  cpi->mb.sadperbit16 =  sad_per_bit16lut[QIndex];
  cpi->mb.sadperbit4  =  sad_per_bit4lut[QIndex];
John Koleszar's avatar
John Koleszar committed
205
206
}

207

208
void vp9_initialize_rd_consts(VP9_COMP *cpi, int QIndex) {
209
  int q, i;
John Koleszar's avatar
John Koleszar committed
210

211
  vp9_clear_system_state();  // __asm emms;
John Koleszar's avatar
John Koleszar committed
212

John Koleszar's avatar
John Koleszar committed
213
214
215
216
217
  // 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);
218

John Koleszar's avatar
John Koleszar committed
219
  cpi->RDMULT = compute_rd_mult(QIndex);
John Koleszar's avatar
John Koleszar committed
220

John Koleszar's avatar
John Koleszar committed
221
222
223
  // Extend rate multiplier along side quantizer zbin increases
  if (cpi->zbin_over_quant  > 0) {
    double oq_factor;
224

John Koleszar's avatar
John Koleszar committed
225
226
227
228
229
    // 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
230

John Koleszar's avatar
John Koleszar committed
231
232
233
234
235
236
237
  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
238

John Koleszar's avatar
John Koleszar committed
239
240
  if (cpi->RDMULT < 7)
    cpi->RDMULT = 7;
241

John Koleszar's avatar
John Koleszar committed
242
243
  cpi->mb.errorperbit = (cpi->RDMULT / 110);
  cpi->mb.errorperbit += (cpi->mb.errorperbit == 0);
244

245
  vp9_set_speed_features(cpi);
John Koleszar's avatar
John Koleszar committed
246

247
  q = (int)pow(vp9_dc_quant(QIndex, 0) >> 2, 1.25);
John Koleszar's avatar
John Koleszar committed
248
249
  q = q << 2;
  cpi->RDMULT = cpi->RDMULT << 4;
Paul Wilkins's avatar
Paul Wilkins committed
250

John Koleszar's avatar
John Koleszar committed
251
252
  if (q < 8)
    q = 8;
253

John Koleszar's avatar
John Koleszar committed
254
255
256
  if (cpi->RDMULT > 1000) {
    cpi->RDDIV = 1;
    cpi->RDMULT /= 100;
John Koleszar's avatar
John Koleszar committed
257

John Koleszar's avatar
John Koleszar committed
258
259
260
261
262
263
    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
264

John Koleszar's avatar
John Koleszar committed
265
      cpi->rd_baseline_thresh[i] = cpi->rd_threshes[i];
John Koleszar's avatar
John Koleszar committed
266
    }
John Koleszar's avatar
John Koleszar committed
267
268
  } else {
    cpi->RDDIV = 100;
John Koleszar's avatar
John Koleszar committed
269

John Koleszar's avatar
John Koleszar committed
270
271
272
273
274
275
    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
276

John Koleszar's avatar
John Koleszar committed
277
      cpi->rd_baseline_thresh[i] = cpi->rd_threshes[i];
John Koleszar's avatar
John Koleszar committed
278
    }
John Koleszar's avatar
John Koleszar committed
279
  }
John Koleszar's avatar
John Koleszar committed
280

281
282
283
284
285
286
287
288
289
290
291
292
293
294
  fill_token_costs(cpi->mb.token_costs[TX_4X4],
                   cpi->common.fc.coef_probs_4x4, BLOCK_TYPES_4X4);
  fill_token_costs(cpi->mb.hybrid_token_costs[TX_4X4],
                   cpi->common.fc.hybrid_coef_probs_4x4, BLOCK_TYPES_4X4);

  fill_token_costs(cpi->mb.token_costs[TX_8X8],
                   cpi->common.fc.coef_probs_8x8, BLOCK_TYPES_8X8);
  fill_token_costs(cpi->mb.hybrid_token_costs[TX_8X8],
                   cpi->common.fc.hybrid_coef_probs_8x8, BLOCK_TYPES_8X8);

  fill_token_costs(cpi->mb.token_costs[TX_16X16],
                   cpi->common.fc.coef_probs_16x16, BLOCK_TYPES_16X16);
  fill_token_costs(cpi->mb.hybrid_token_costs[TX_16X16],
                   cpi->common.fc.hybrid_coef_probs_16x16, BLOCK_TYPES_16X16);
Daniel Kang's avatar
Daniel Kang committed
295

296
#if CONFIG_TX32X32
297
298
  fill_token_costs(cpi->mb.token_costs[TX_32X32],
                   cpi->common.fc.coef_probs_32x32, BLOCK_TYPES_32X32);
299
300
#endif

John Koleszar's avatar
John Koleszar committed
301
302
  /*rough estimate for costing*/
  cpi->common.kf_ymode_probs_index = cpi->common.base_qindex >> 4;
303
  vp9_init_mode_costs(cpi);
John Koleszar's avatar
John Koleszar committed
304

305
  if (cpi->common.frame_type != KEY_FRAME) {
306
    vp9_build_nmv_cost_table(
307
308
309
310
311
312
        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
313
314
}

315
int vp9_block_error_c(int16_t *coeff, int16_t *dqcoeff, int block_size) {
316
  int i, error = 0;
John Koleszar's avatar
John Koleszar committed
317

318
  for (i = 0; i < block_size; i++) {
John Koleszar's avatar
John Koleszar committed
319
320
321
    int this_diff = coeff[i] - dqcoeff[i];
    error += this_diff * this_diff;
  }
John Koleszar's avatar
John Koleszar committed
322

John Koleszar's avatar
John Koleszar committed
323
  return error;
John Koleszar's avatar
John Koleszar committed
324
325
}

326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
int vp9_mbblock_error_8x8_c(MACROBLOCK *mb, int dc) {
  BLOCK  *be;
  BLOCKD *bd;
  int i, j;
  int berror, error = 0;

  for (i = 0; i < 16; i+=4) {
    be = &mb->block[i];
    bd = &mb->e_mbd.block[i];
    berror = 0;
    for (j = dc; j < 64; j++) {
      int this_diff = be->coeff[j] - bd->dqcoeff[j];
      berror += this_diff * this_diff;
    }
    error += berror;
  }
  return error;
}

345
int vp9_mbblock_error_c(MACROBLOCK *mb, int dc) {
John Koleszar's avatar
John Koleszar committed
346
347
348
349
  BLOCK  *be;
  BLOCKD *bd;
  int i, j;
  int berror, error = 0;
John Koleszar's avatar
John Koleszar committed
350

John Koleszar's avatar
John Koleszar committed
351
352
353
354
355
356
357
  for (i = 0; i < 16; i++) {
    be = &mb->block[i];
    bd = &mb->e_mbd.block[i];
    berror = 0;
    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
358
    }
John Koleszar's avatar
John Koleszar committed
359
360
361
    error += berror;
  }
  return error;
John Koleszar's avatar
John Koleszar committed
362
363
}

364
int vp9_mbuverror_c(MACROBLOCK *mb) {
John Koleszar's avatar
John Koleszar committed
365
366
  BLOCK  *be;
  BLOCKD *bd;
John Koleszar's avatar
John Koleszar committed
367

368
  int i, error = 0;
John Koleszar's avatar
John Koleszar committed
369

John Koleszar's avatar
John Koleszar committed
370
371
372
  for (i = 16; i < 24; i++) {
    be = &mb->block[i];
    bd = &mb->e_mbd.block[i];
John Koleszar's avatar
John Koleszar committed
373

374
    error += vp9_block_error_c(be->coeff, bd->dqcoeff, 16);
John Koleszar's avatar
John Koleszar committed
375
  }
John Koleszar's avatar
John Koleszar committed
376

John Koleszar's avatar
John Koleszar committed
377
  return error;
John Koleszar's avatar
John Koleszar committed
378
379
}

380
int vp9_uvsse(MACROBLOCK *x) {
381
382
383
  uint8_t *uptr, *vptr;
  uint8_t *upred_ptr = (*(x->block[16].base_src) + x->block[16].src);
  uint8_t *vpred_ptr = (*(x->block[20].base_src) + x->block[20].src);
John Koleszar's avatar
John Koleszar committed
384
385
386
387
  int uv_stride = x->block[16].src_stride;

  unsigned int sse1 = 0;
  unsigned int sse2 = 0;
388
389
  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
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
  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) {
411
    vp9_sub_pixel_variance8x8(uptr, pre_stride, (mv_col & 7) << 1,
412
                              (mv_row & 7) << 1, upred_ptr, uv_stride, &sse2);
413
    vp9_sub_pixel_variance8x8(vptr, pre_stride, (mv_col & 7) << 1,
414
                              (mv_row & 7) << 1, vpred_ptr, uv_stride, &sse1);
John Koleszar's avatar
John Koleszar committed
415
416
    sse2 += sse1;
  } else {
417
418
    vp9_variance8x8(uptr, pre_stride, upred_ptr, uv_stride, &sse2);
    vp9_variance8x8(vptr, pre_stride, vpred_ptr, uv_stride, &sse1);
John Koleszar's avatar
John Koleszar committed
419
420
421
    sse2 += sse1;
  }
  return sse2;
John Koleszar's avatar
John Koleszar committed
422
423
424

}

425
426
427
428
429
#if CONFIG_NEWCOEFCONTEXT
#define PT pn
#else
#define PT pt
#endif
430
431
432
433
static int cost_coeffs(MACROBLOCK *mb,
                       BLOCKD *b, PLANE_TYPE type,
                       ENTROPY_CONTEXT *a,
                       ENTROPY_CONTEXT *l,
434
                       TX_SIZE tx_size) {
435
  int pt;
Daniel Kang's avatar
Daniel Kang committed
436
  const int eob = b->eob;
437
  MACROBLOCKD *xd = &mb->e_mbd;
438
#if CONFIG_TX32X32
439
440
441
442
443
444
445
446
447
  const int ib = (int)(b - xd->block);
#endif
  int c = (type == PLANE_TYPE_Y_NO_DC) ? 1 : 0;
  int cost = 0, seg_eob;
  const int segment_id = xd->mode_info_context->mbmi.segment_id;
  const int *scan, *band;
  int16_t *qcoeff_ptr = b->qcoeff;
  const TX_TYPE tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
                          get_tx_type(xd, b) : DCT_DCT;
448
449
450
451
#if CONFIG_NEWCOEFCONTEXT
  const int *neighbors;
  int pn;
#endif
452
453

  ENTROPY_CONTEXT a_ec = *a, l_ec = *l;
454

455
  switch (tx_size) {
Daniel Kang's avatar
Daniel Kang committed
456
    case TX_4X4:
457
458
459
      scan = vp9_default_zig_zag1d_4x4;
      band = vp9_coef_bands_4x4;
      seg_eob = 16;
Deb Mukherjee's avatar
Deb Mukherjee committed
460
      if (type == PLANE_TYPE_Y_WITH_DC) {
461
462
463
464
        if (tx_type == ADST_DCT) {
          scan = vp9_row_scan_4x4;
        } else if (tx_type == DCT_ADST) {
          scan = vp9_col_scan_4x4;
465
        }
Daniel Kang's avatar
Daniel Kang committed
466
467
468
      }
      break;
    case TX_8X8:
469
      if (type == PLANE_TYPE_Y2) {
470
471
        scan = vp9_default_zig_zag1d_4x4;
        band = vp9_coef_bands_4x4;
472
473
474
475
476
        seg_eob = 4;
      } else {
        scan = vp9_default_zig_zag1d_8x8;
        band = vp9_coef_bands_8x8;
        seg_eob = 64;
477
      }
Daniel Kang's avatar
Daniel Kang committed
478
479
      break;
    case TX_16X16:
480
481
      scan = vp9_default_zig_zag1d_16x16;
      band = vp9_coef_bands_16x16;
482
      seg_eob = 256;
483
#if CONFIG_TX32X32
484
485
486
      if (type == PLANE_TYPE_UV) {
        const int uv_idx = ib - 16;
        qcoeff_ptr = xd->sb_coeff_data.qcoeff + 1024 + 64 * uv_idx;
Deb Mukherjee's avatar
Deb Mukherjee committed
487
      }
488
#endif
Daniel Kang's avatar
Daniel Kang committed
489
      break;
490
#if CONFIG_TX32X32
491
492
493
    case TX_32X32:
      scan = vp9_default_zig_zag1d_32x32;
      band = vp9_coef_bands_32x32;
494
      seg_eob = 1024;
495
496
497
      qcoeff_ptr = xd->sb_coeff_data.qcoeff;
      break;
#endif
Daniel Kang's avatar
Daniel Kang committed
498
    default:
499
      abort();
Daniel Kang's avatar
Daniel Kang committed
500
501
      break;
  }
502

503
  VP9_COMBINEENTROPYCONTEXTS(pt, a_ec, l_ec);
504
505
506
507
#if CONFIG_NEWCOEFCONTEXT
  neighbors = vp9_get_coef_neighbors_handle(scan);
  pn = pt;
#endif
508
509
510
511

  if (vp9_segfeature_active(xd, segment_id, SEG_LVL_EOB))
    seg_eob = vp9_get_segdata(xd, segment_id, SEG_LVL_EOB);

512
513
514
  if (tx_type != DCT_DCT) {
    for (; c < eob; c++) {
      int v = qcoeff_ptr[scan[c]];
515
      int t = vp9_dct_value_tokens_ptr[v].Token;
516
      cost += mb->hybrid_token_costs[tx_size][type][band[c]][PT][t];
517
518
      cost += vp9_dct_value_cost_ptr[v];
      pt = vp9_prev_token_class[t];
519
520
521
#if CONFIG_NEWCOEFCONTEXT
      if (c < seg_eob - 1 && NEWCOEFCONTEXT_BAND_COND(band[c + 1]))
        pn = vp9_get_coef_neighbor_context(
522
            qcoeff_ptr, (type == PLANE_TYPE_Y_NO_DC), neighbors, scan[c + 1]);
523
524
525
      else
        pn = pt;
#endif
526
527
528
    }
    if (c < seg_eob)
      cost += mb->hybrid_token_costs[tx_size][type][band[c]]
529
          [PT][DCT_EOB_TOKEN];
530
  } else {
531
532
    for (; c < eob; c++) {
      int v = qcoeff_ptr[scan[c]];
533
      int t = vp9_dct_value_tokens_ptr[v].Token;
534
      cost += mb->token_costs[tx_size][type][band[c]][pt][t];
535
536
      cost += vp9_dct_value_cost_ptr[v];
      pt = vp9_prev_token_class[t];
537
538
539
#if CONFIG_NEWCOEFCONTEXT
      if (c < seg_eob - 1 && NEWCOEFCONTEXT_BAND_COND(band[c + 1]))
        pn = vp9_get_coef_neighbor_context(
540
            qcoeff_ptr, (type == PLANE_TYPE_Y_NO_DC), neighbors, scan[c + 1]);
541
542
543
      else
        pn = pt;
#endif
544
545
546
    }
    if (c < seg_eob)
      cost += mb->token_costs[tx_size][type][band[c]]
547
          [PT][DCT_EOB_TOKEN];
548
549
  }

550
551
  // is eob first coefficient;
  pt = (c > !type);
552
553
554
555
  *a = *l = pt;
  return cost;
}

556
static int rdcost_mby_4x4(MACROBLOCK *mb, int has_2nd_order, int backup) {
John Koleszar's avatar
John Koleszar committed
557
558
  int cost = 0;
  int b;
Paul Wilkins's avatar
Paul Wilkins committed
559
  MACROBLOCKD *xd = &mb->e_mbd;
John Koleszar's avatar
John Koleszar committed
560
561
562
  ENTROPY_CONTEXT_PLANES t_above, t_left;
  ENTROPY_CONTEXT *ta;
  ENTROPY_CONTEXT *tl;
563

564
565
566
  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
567

568
569
570
571
572
573
    ta = (ENTROPY_CONTEXT *)&t_above;
    tl = (ENTROPY_CONTEXT *)&t_left;
  } else {
    ta = (ENTROPY_CONTEXT *)xd->above_context;
    tl = (ENTROPY_CONTEXT *)xd->left_context;
  }
John Koleszar's avatar
John Koleszar committed
574

John Koleszar's avatar
John Koleszar committed
575
  for (b = 0; b < 16; b++)
576
577
578
    cost += cost_coeffs(mb, xd->block + b,
                        (has_2nd_order ?
                         PLANE_TYPE_Y_NO_DC : PLANE_TYPE_Y_WITH_DC),
Yaowu Xu's avatar
Yaowu Xu committed
579
580
                        ta + vp9_block2above[TX_4X4][b],
                        tl + vp9_block2left[TX_4X4][b],
Daniel Kang's avatar
Daniel Kang committed
581
                        TX_4X4);
John Koleszar's avatar
John Koleszar committed
582

583
584
  if (has_2nd_order)
    cost += cost_coeffs(mb, xd->block + 24, PLANE_TYPE_Y2,
Yaowu Xu's avatar
Yaowu Xu committed
585
586
                        ta + vp9_block2above[TX_4X4][24],
                        tl + vp9_block2left[TX_4X4][24],
587
                        TX_4X4);
John Koleszar's avatar
John Koleszar committed
588

John Koleszar's avatar
John Koleszar committed
589
  return cost;
John Koleszar's avatar
John Koleszar committed
590
591
}

592
593
594
static void macro_block_yrd_4x4(MACROBLOCK *mb,
                                int *Rate,
                                int *Distortion,
595
                                int *skippable, int backup) {
Paul Wilkins's avatar
Paul Wilkins committed
596
  MACROBLOCKD *const xd = &mb->e_mbd;
John Koleszar's avatar
John Koleszar committed
597
  BLOCK   *const mb_y2 = mb->block + 24;
Paul Wilkins's avatar
Paul Wilkins committed
598
  BLOCKD *const x_y2  = xd->block + 24;
Jim Bankoski's avatar
Jim Bankoski committed
599
  int d, has_2nd_order;
600

601
602
  xd->mode_info_context->mbmi.txfm_size = TX_4X4;
  has_2nd_order = get_2nd_order_usage(xd);
John Koleszar's avatar
John Koleszar committed
603
  // Fdct and building the 2nd order block
604
605
606
607
608
  vp9_transform_mby_4x4(mb);
  vp9_quantize_mby_4x4(mb);
  d = vp9_mbblock_error(mb, has_2nd_order);
  if (has_2nd_order)
    d += vp9_block_error(mb_y2->coeff, x_y2->dqcoeff, 16);
609

John Koleszar's avatar
John Koleszar committed
610
611
  *Distortion = (d >> 2);
  // rate
612
613
  *Rate = rdcost_mby_4x4(mb, has_2nd_order, backup);
  *skippable = vp9_mby_is_skippable_4x4(&mb->e_mbd, has_2nd_order);
614
}
John Koleszar's avatar
John Koleszar committed
615

616
static int rdcost_mby_8x8(MACROBLOCK *mb, int has_2nd_order, int backup) {
John Koleszar's avatar
John Koleszar committed
617
618
  int cost = 0;
  int b;
Paul Wilkins's avatar
Paul Wilkins committed
619
  MACROBLOCKD *xd = &mb->e_mbd;
John Koleszar's avatar
John Koleszar committed
620
621
622
623
  ENTROPY_CONTEXT_PLANES t_above, t_left;
  ENTROPY_CONTEXT *ta;
  ENTROPY_CONTEXT *tl;

Ronald S. Bultje's avatar
Ronald S. Bultje committed
624
625
626
  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
627

Ronald S. Bultje's avatar
Ronald S. Bultje committed
628
629
630
631
632
633
    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
634
635

  for (b = 0; b < 16; b += 4)
636
637
638
    cost += cost_coeffs(mb, xd->block + b,
                        (has_2nd_order ?
                         PLANE_TYPE_Y_NO_DC : PLANE_TYPE_Y_WITH_DC),
Yaowu Xu's avatar
Yaowu Xu committed
639
640
                        ta + vp9_block2above[TX_8X8][b],
                        tl + vp9_block2left[TX_8X8][b],
Daniel Kang's avatar
Daniel Kang committed
641
                        TX_8X8);
John Koleszar's avatar
John Koleszar committed
642

643
  if (has_2nd_order)
644
    cost += cost_coeffs(mb, xd->block + 24, PLANE_TYPE_Y2,
Yaowu Xu's avatar
Yaowu Xu committed
645
                            ta + vp9_block2above[TX_8X8][24],
646
647
                            tl + vp9_block2left[TX_8X8][24],
                            TX_8X8);
John Koleszar's avatar
John Koleszar committed
648
  return cost;
649
650
}

John Koleszar's avatar
John Koleszar committed
651
652
653
static void macro_block_yrd_8x8(MACROBLOCK *mb,
                                int *Rate,
                                int *Distortion,
654
                                int *skippable, int backup) {
Paul Wilkins's avatar
Paul Wilkins committed
655
  MACROBLOCKD *const xd = &mb->e_mbd;
John Koleszar's avatar
John Koleszar committed
656
  BLOCK   *const mb_y2 = mb->block + 24;
Paul Wilkins's avatar
Paul Wilkins committed
657
  BLOCKD *const x_y2  = xd->block + 24;
658
659
660
  int d, has_2nd_order;

  xd->mode_info_context->mbmi.txfm_size = TX_8X8;
John Koleszar's avatar
John Koleszar committed
661

662
663
  vp9_transform_mby_8x8(mb);
  vp9_quantize_mby_8x8(mb);
664
665
666
667
  has_2nd_order = get_2nd_order_usage(xd);
  d = vp9_mbblock_error_8x8_c(mb, has_2nd_order);
  if (has_2nd_order)
    d += vp9_block_error(mb_y2->coeff, x_y2->dqcoeff, 16);
John Koleszar's avatar
John Koleszar committed
668
669
670

  *Distortion = (d >> 2);
  // rate
671
672
  *Rate = rdcost_mby_8x8(mb, has_2nd_order, backup);
  *skippable = vp9_mby_is_skippable_8x8(&mb->e_mbd, has_2nd_order);
673
}
674

675
static int rdcost_mby_16x16(MACROBLOCK *mb, int backup) {
Daniel Kang's avatar
Daniel Kang committed
676
  int cost;
Paul Wilkins's avatar
Paul Wilkins committed
677
  MACROBLOCKD *xd = &mb->e_mbd;
Daniel Kang's avatar
Daniel Kang committed
678
679
680
  ENTROPY_CONTEXT_PLANES t_above, t_left;
  ENTROPY_CONTEXT *ta, *tl;

681
682
683
  if (backup) {
    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
684

685
686
687
688
689
690
    ta = (ENTROPY_CONTEXT *)&t_above;
    tl = (ENTROPY_CONTEXT *)&t_left;
  } else {
    ta = (ENTROPY_CONTEXT *)xd->above_context;
    tl = (ENTROPY_CONTEXT *)xd->left_context;
  }
Daniel Kang's avatar
Daniel Kang committed
691

Paul Wilkins's avatar
Paul Wilkins committed
692
  cost = cost_coeffs(mb, xd->block, PLANE_TYPE_Y_WITH_DC, ta, tl, TX_16X16);
Daniel Kang's avatar
Daniel Kang committed
693
694
  return cost;
}
695

Daniel Kang's avatar
Daniel Kang committed
696
static void macro_block_yrd_16x16(MACROBLOCK *mb, int *Rate, int *Distortion,
697
                                  int *skippable, int backup) {
Daniel Kang's avatar
Daniel Kang committed
698
  int d;
Deb Mukherjee's avatar
Deb Mukherjee committed
699
  MACROBLOCKD *xd = &mb->e_mbd;
Daniel Kang's avatar
Daniel Kang committed
700

701
702
  xd->mode_info_context->mbmi.txfm_size = TX_16X16;
  vp9_transform_mby_16x16(mb);
703
  vp9_quantize_mby_16x16(mb);
704
705
706
  // 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?
707
  if (mb->e_mbd.mode_info_context->mbmi.mode < I8X8_PRED)
708
    vp9_optimize_mby_16x16(mb);
709

710
  d = vp9_mbblock_error(mb, 0);
Daniel Kang's avatar
Daniel Kang committed
711
712
713

  *Distortion = (d >> 2);
  // rate
714
  *Rate = rdcost_mby_16x16(mb, backup);
715
  *skippable = vp9_mby_is_skippable_16x16(&mb->e_mbd);
Daniel Kang's avatar
Daniel Kang committed
716
717
}

718
static void choose_txfm_size_from_rd(VP9_COMP *cpi, MACROBLOCK *x,
719
720
721
722
723
                                     int (*r)[2], int *rate,
                                     int *d, int *distortion,
                                     int *s, int *skip,
                                     int64_t txfm_cache[NB_TXFM_MODES],
                                     TX_SIZE max_txfm_size) {
724
725
726
727
728
  VP9_COMMON *const cm = &cpi->common;
  MACROBLOCKD *const xd = &x->e_mbd;
  MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
  vp9_prob skip_prob = cm->mb_no_coeff_skip ?
                       vp9_get_pred_prob(cm, xd, PRED_MBSKIP) : 128;
729
730
731
732
733
734
735
736
737
738
739
740
  int64_t rd[TX_SIZE_MAX_SB][2];
  int n, m;

  for (n = TX_4X4; n <= max_txfm_size; n++) {
    r[n][1] = r[n][0];
    for (m = 0; m <= n - (n == max_txfm_size); m++) {
      if (m == n)
        r[n][1] += vp9_cost_zero(cm->prob_tx[m]);
      else
        r[n][1] += vp9_cost_one(cm->prob_tx[m]);
    }
  }
741
742
743
744
745
746
747
748

  if (cm->mb_no_coeff_skip) {
    int s0, s1;

    assert(skip_prob > 0);
    s0 = vp9_cost_bit(skip_prob, 0);
    s1 = vp9_cost_bit(skip_prob, 1);

749
    for (n = TX_4X4; n <= max_txfm_size; n++) {
750
      if (s[n]) {
751
        rd[n][0] = rd[n][1] = RDCOST(x->rdmult, x->rddiv, s1, d[n]);
752
      } else {
753
754
        rd[n][0] = RDCOST(x->rdmult, x->rddiv, r[n][0] + s0, d[n]);
        rd[n][1] = RDCOST(x->rdmult, x->rddiv, r[n][1] + s0, d[n]);
755
      }
756
757
    }
  } else {
758
759
760
    for (n = TX_4X4; n <= max_txfm_size; n++) {
      rd[n][0] = RDCOST(x->rdmult, x->rddiv, r[n][0], d[n]);
      rd[n][1] = RDCOST(x->rdmult, x->rddiv, r[n][1], d[n]);
761
762
763
    }
  }

764
#if CONFIG_TX32X32
765
766
767
768
769
770
771
772
  if (max_txfm_size == TX_32X32 &&
      (cm->txfm_mode == ALLOW_32X32 ||
       (cm->txfm_mode == TX_MODE_SELECT &&
        rd[TX_32X32][1] < rd[TX_16X16][1] && rd[TX_32X32][1] < rd[TX_8X8][1] &&
        rd[TX_32X32][1] < rd[TX_4X4][1]))) {
    mbmi->txfm_size = TX_32X32;
  } else
#endif
773
  if ( cm->txfm_mode == ALLOW_16X16 ||
774
#if CONFIG_TX32X32
775
776
      (max_txfm_size == TX_16X16 && cm->txfm_mode == ALLOW_32X32) ||
#endif
777
      (cm->txfm_mode == TX_MODE_SELECT &&
778
       rd[TX_16X16][1] < rd[TX_8X8][1] && rd[TX_16X16][1] < rd[TX_4X4][1])) {
779
    mbmi->txfm_size = TX_16X16;
780
  } else if (cm->txfm_mode == ALLOW_8X8 ||
781
           (cm->txfm_mode == TX_MODE_SELECT && rd[TX_8X8][1] < rd[TX_4X4][1])) {
782
783
    mbmi->txfm_size = TX_8X8;
  } else {
784
    assert(cm->txfm_mode == ONLY_4X4 || cm->txfm_mode == TX_MODE_SELECT);
785
786
787
    mbmi->txfm_size = TX_4X4;
  }

788
  *distortion = d[mbmi->txfm_size];
789
  *rate       = r[mbmi->txfm_size][cm->txfm_mode == TX_MODE_SELECT];
790
791
  *skip       = s[mbmi->txfm_size];

792
793
794
  txfm_cache[ONLY_4X4] = rd[TX_4X4][0];
  txfm_cache[ALLOW_8X8] = rd[TX_8X8][0];
  txfm_cache[ALLOW_16X16] = rd[TX_16X16][0];
795
#if CONFIG_TX32X32
796
797
798
799
800
801
802
803
804
  txfm_cache[ALLOW_32X32] = rd[max_txfm_size][0];
  if (max_txfm_size == TX_32X32 &&
      rd[TX_32X32][1] < rd[TX_16X16][1] && rd[TX_32X32][1] < rd[TX_8X8][1] &&
      rd[TX_32X32][1] < rd[TX_4X4][1])
    txfm_cache[TX_MODE_SELECT] = rd[TX_32X32][1];
  else
#endif
  if (rd[TX_16X16][1] < rd[TX_8X8][1] && rd[TX_16X16][1] < rd[TX_4X4][1])
    txfm_cache[TX_MODE_SELECT] = rd[TX_16X16][1];
805
  else
806
807
    txfm_cache[TX_MODE_SELECT] = rd[TX_4X4][1] < rd[TX_8X8][1] ?
                                 rd[TX_4X4][1] : rd[TX_8X8][1];
808
809
810
811
812
813
}

static void macro_block_yrd(VP9_COMP *cpi, MACROBLOCK *x, int *rate,
                            int *distortion, int *skippable,
                            int64_t txfm_cache[NB_TXFM_MODES]) {
  MACROBLOCKD *const xd = &x->e_mbd;
814
  int r[TX_SIZE_MAX_MB][2], d[TX_SIZE_MAX_MB], s[TX_SIZE_MAX_MB];
815
816
817

  vp9_subtract_mby(x->src_diff, *(x->block[0].base_src), xd->predictor,
                   x->block[0].src_stride);
818

819
820
821
  macro_block_yrd_16x16(x, &r[TX_16X16][0], &d[TX_16X16], &s[TX_16X16], 1);
  macro_block_yrd_8x8(x, &r[TX_8X8][0], &d[TX_8X8], &s[TX_8X8], 1);
  macro_block_yrd_4x4(x, &r[TX_4X4][0], &d[TX_4X4], &s[TX_4X4], 1);
822
823

  choose_txfm_size_from_rd(cpi, x, r, rate, d, distortion, s, skippable,
824
                           txfm_cache, TX_16X16);
825
826
}

827
static void copy_predictor(uint8_t *dst, const uint8_t *predictor) {
John Koleszar's avatar
John Koleszar committed
828
829
830
831
832
833
  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];
834
}
Yaowu Xu's avatar
Yaowu Xu committed
835

836
#if CONFIG_TX32X32
Ronald S. Bultje's avatar
Ronald S. Bultje committed
837
static int rdcost_sby_32x32(MACROBLOCK *x, int backup) {
838
839
  MACROBLOCKD * const xd = &x->e_mbd;
  ENTROPY_CONTEXT_PLANES t_above, t_left;
Ronald S. Bultje's avatar
Ronald S. Bultje committed
840
  ENTROPY_CONTEXT *ta, *tl;
841

Ronald S. Bultje's avatar
Ronald S. Bultje committed
842
843
844
845
846
847
848
849
850
851
  if (backup) {
    ta = (ENTROPY_CONTEXT *) &t_above,
    tl = (ENTROPY_CONTEXT *) &t_left;

    vpx_memcpy(&t_above, xd->above_context, sizeof(ENTROPY_CONTEXT_PLANES));
    vpx_memcpy(&t_left,  xd->left_context,  sizeof(ENTROPY_CONTEXT_PLANES));
  } else {
    ta = (ENTROPY_CONTEXT *) xd->above_context;
    tl = (ENTROPY_CONTEXT *) xd->left_context;
  }
852
853
854
855

  return cost_coeffs(x, xd->block, PLANE_TYPE_Y_WITH_DC, ta, tl, TX_32X32);
}

856
857
static int vp9_sb_block_error_c(int16_t *coeff, int16_t *dqcoeff,
                                int block_size) {
858
859
860
861
862
863
864
865
866
867
868
869
870
  int i;
  int64_t error = 0;

  for (i = 0; i < block_size; i++) {
    unsigned int this_diff = coeff[i] - dqcoeff[i];
    error += this_diff * this_diff;
  }

  return error > INT_MAX ? INT_MAX : error;
}

#define DEBUG_ERROR 0
static void super_block_yrd_32x32(MACROBLOCK *x,
Ronald S. Bultje's avatar
Ronald S. Bultje committed
871
872
                                  int *rate, int *distortion, int *skippable,
                                  int backup) {
873
874
875
  SUPERBLOCK  * const x_sb = &x->sb_coeff_data;
  MACROBLOCKD * const xd = &x->e_mbd;
  SUPERBLOCKD * const xd_sb = &xd->sb_coeff_data;
876
#if DEBUG_ERROR || CONFIG_DWTDCTHYBRID
877
  int16_t out[1024];
878
879
880
881
#endif

  vp9_transform_sby_32x32(x);
  vp9_quantize_sby_32x32(x);
882
#if DEBUG_ERROR || CONFIG_DWTDCTHYBRID
883
884
885
  vp9_short_idct32x32(xd_sb->dqcoeff, out, 64);
#endif

886
#if !CONFIG_DWTDCTHYBRID
887
888
889
890
891
892
893
894
  *distortion = vp9_sb_block_error_c(x_sb->coeff, xd_sb->dqcoeff, 1024);
#else
  *distortion = vp9_block_error_c(x_sb->src_diff, out, 1024) << 4;
#endif
#if DEBUG_ERROR
  printf("IDCT/FDCT error 32x32: %d (d: %d)\n",
         vp9_block_error_c(x_sb->src_diff, out, 1024), *distortion);
#endif
Ronald S. Bultje's avatar
Ronald S. Bultje committed
895
  *rate       = rdcost_sby_32x32(x, backup);
896
897
898
899
  *skippable  = vp9_sby_is_skippable_32x32(&x->e_mbd);
}
#endif

900
901
static void super_block_yrd(VP9_COMP *cpi,
                            MACROBLOCK *x, int *rate, int *distortion,
902
                            int *skip,
903
                            int64_t txfm_cache[NB_TXFM_MODES]) {
Ronald S. Bultje's avatar
Ronald S. Bultje committed
904
  MACROBLOCKD *const xd = &x->e_mbd;
905
  int r[TX_SIZE_MAX_SB][2], d[TX_SIZE_MAX_SB], s[TX_SIZE_MAX_SB], n;
Ronald S. Bultje's avatar
Ronald S. Bultje committed
906
907
  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;
908
909
910
911
  ENTROPY_CONTEXT_PLANES t_above[TX_SIZE_MAX_MB][2],
                        *orig_above = xd->above_context;
  ENTROPY_CONTEXT_PLANES t_left[TX_SIZE_MAX_MB][2],
                        *orig_left = xd->left_context;
912

913
  for (n = TX_4X4; n < TX_SIZE_MAX_MB; n++) {
914
915
    vpx_memcpy(t_above[n], xd->above_context, sizeof(t_above[n]));
    vpx_memcpy(t_left[n], xd->left_context, sizeof(t_left[n]));
916
    r[n][0] = 0;
917
918
919
    d[n] = 0;
    s[n] = 1;
  }
Ronald S. Bultje's avatar
Ronald S. Bultje committed
920

921
922
923
#if CONFIG_TX32X32
  vp9_subtract_sby_s_c(x->sb_coeff_data.src_diff, src, src_y_stride,
                       dst, dst_y_stride);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
924
  super_block_yrd_32x32(x, &r[TX_32X32][0], &d[TX_32X32], &s[TX_32X32], 1);
925
926
927
928
929
#endif

#if DEBUG_ERROR
  int err[3] = { 0, 0, 0 };
#endif
Ronald S. Bultje's avatar
Ronald S. Bultje committed
930
931
  for (n = 0; n < 4; n++) {
    int x_idx = n & 1, y_idx = n >> 1;
932
    int r_tmp, d_tmp, s_tmp;
Ronald S. Bultje's avatar
Ronald S. Bultje committed
933

934
    vp9_subtract_mby_s_c(x->src_diff,
Ronald S. Bultje's avatar
Ronald S. Bultje committed
935
936
937
938
                         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);
939
940
941

    xd->above_context = &t_above[TX_16X16][x_idx];
    xd->left_context = &t_left[TX_16X16][y_idx];
942
    macro_block_yrd_16x16(x, &r_tmp, &d_tmp, &s_tmp, 0);
943
    d[TX_16X16] += d_tmp;
944
    r[TX_16X16][0] += r_tmp;
945
    s[TX_16X16] = s[TX_16X16] && s_tmp;
946
947
948
949
#if DEBUG_ERROR
    vp9_inverse_transform_mby_16x16(xd);
    err[2] += vp9_block_error_c(xd->diff, x->src_diff, 256);
#endif
950
951
952

    xd->above_context = &t_above[TX_4X4][x_idx];
    xd->left_context = &t_left[TX_4X4][y_idx];