vp9_rdopt.c 206 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
#include "vp9/common/vp9_entropymode.h"
#include "vp9/common/vp9_reconinter.h"
#include "vp9/common/vp9_reconintra.h"
#include "vp9/common/vp9_findnearmv.h"
#include "vp9/common/vp9_quant_common.h"
28 29 30 31 32 33
#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
34
#include "vpx_mem/vpx_mem.h"
35 36
#include "vp9/common/vp9_systemdependent.h"
#include "vp9/encoder/vp9_encodemv.h"
John Koleszar's avatar
John Koleszar committed
37

38 39 40
#include "vp9/common/vp9_seg_common.h"
#include "vp9/common/vp9_pred_common.h"
#include "vp9/common/vp9_entropy.h"
41
#include "vp9_rtcd.h"
42
#include "vp9/common/vp9_mvref_common.h"
Ronald S. Bultje's avatar
Ronald S. Bultje committed
43
#include "vp9/common/vp9_common.h"
Paul Wilkins's avatar
Paul Wilkins committed
44

John Koleszar's avatar
John Koleszar committed
45 46
#define MAXF(a,b)            (((a) > (b)) ? (a) : (b))

47 48
#define INVALID_MV 0x80008000

49 50 51
/* Factor to weigh the rate for switchable interp filters */
#define SWITCHABLE_INTERP_RATE_FACTOR 1

John Koleszar's avatar
John Koleszar committed
52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69
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
70 71
};

72
const MODE_DEFINITION vp9_mode_order[MAX_MODES] = {
73 74
  {ZEROMV,    LAST_FRAME,   NONE},
  {DC_PRED,   INTRA_FRAME,  NONE},
John Koleszar's avatar
John Koleszar committed
75

76 77
  {NEARESTMV, LAST_FRAME,   NONE},
  {NEARMV,    LAST_FRAME,   NONE},
John Koleszar's avatar
John Koleszar committed
78

79 80
  {ZEROMV,    GOLDEN_FRAME, NONE},
  {NEARESTMV, GOLDEN_FRAME, NONE},
John Koleszar's avatar
John Koleszar committed
81

82 83
  {ZEROMV,    ALTREF_FRAME, NONE},
  {NEARESTMV, ALTREF_FRAME, NONE},
John Koleszar's avatar
John Koleszar committed
84

85 86
  {NEARMV,    GOLDEN_FRAME, NONE},
  {NEARMV,    ALTREF_FRAME, NONE},
John Koleszar's avatar
John Koleszar committed
87

88 89 90 91 92 93 94 95
  {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
96

97
  {TM_PRED,   INTRA_FRAME,  NONE},
John Koleszar's avatar
John Koleszar committed
98

99 100 101
  {NEWMV,     LAST_FRAME,   NONE},
  {NEWMV,     GOLDEN_FRAME, NONE},
  {NEWMV,     ALTREF_FRAME, NONE},
John Koleszar's avatar
John Koleszar committed
102

103 104 105
  {SPLITMV,   LAST_FRAME,   NONE},
  {SPLITMV,   GOLDEN_FRAME, NONE},
  {SPLITMV,   ALTREF_FRAME, NONE},
106

107 108
  {B_PRED,    INTRA_FRAME,  NONE},
  {I8X8_PRED, INTRA_FRAME,  NONE},
109

John Koleszar's avatar
John Koleszar committed
110 111 112 113
  /* compound prediction modes */
  {ZEROMV,    LAST_FRAME,   GOLDEN_FRAME},
  {NEARESTMV, LAST_FRAME,   GOLDEN_FRAME},
  {NEARMV,    LAST_FRAME,   GOLDEN_FRAME},
114

John Koleszar's avatar
John Koleszar committed
115 116 117
  {ZEROMV,    ALTREF_FRAME, LAST_FRAME},
  {NEARESTMV, ALTREF_FRAME, LAST_FRAME},
  {NEARMV,    ALTREF_FRAME, LAST_FRAME},
118

John Koleszar's avatar
John Koleszar committed
119 120 121
  {ZEROMV,    GOLDEN_FRAME, ALTREF_FRAME},
  {NEARESTMV, GOLDEN_FRAME, ALTREF_FRAME},
  {NEARMV,    GOLDEN_FRAME, ALTREF_FRAME},
122

John Koleszar's avatar
John Koleszar committed
123 124 125
  {NEWMV,     LAST_FRAME,   GOLDEN_FRAME},
  {NEWMV,     ALTREF_FRAME, LAST_FRAME  },
  {NEWMV,     GOLDEN_FRAME, ALTREF_FRAME},
126

John Koleszar's avatar
John Koleszar committed
127 128
  {SPLITMV,   LAST_FRAME,   GOLDEN_FRAME},
  {SPLITMV,   ALTREF_FRAME, LAST_FRAME  },
129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147
  {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
148 149
};

150 151 152
static void fill_token_costs(vp9_coeff_count *c,
                             vp9_coeff_probs *p,
                             int block_type_counts) {
153
  int i, j, k, l;
John Koleszar's avatar
John Koleszar committed
154 155

  for (i = 0; i < block_type_counts; i++)
156 157 158
    for (j = 0; j < REF_TYPES; j++)
      for (k = 0; k < COEF_BANDS; k++)
        for (l = 0; l < PREV_COEF_CONTEXTS; l++) {
159 160 161 162 163 164
#if CONFIG_CODE_NONZEROCOUNT
          // All costs are without the EOB node
          vp9_cost_tokens_skip((int *)(c[i][j][k][l]),
                               p[i][j][k][l],
                               vp9_coef_tree);
#else
165 166 167 168 169 170 171 172
          if (l == 0 && k > 0)
            vp9_cost_tokens_skip((int *)(c[i][j][k][l]),
                                 p[i][j][k][l],
                                 vp9_coef_tree);
          else
            vp9_cost_tokens((int *)(c[i][j][k][l]),
                            p[i][j][k][l],
                            vp9_coef_tree);
173 174 175 176 177 178 179 180 181 182 183 184 185
#endif
        }
}

#if CONFIG_CODE_NONZEROCOUNT
static void fill_nzc_costs(VP9_COMP *cpi, int block_size) {
  int nzc_context, r, b, nzc, values;
  int cost[16];
  values = block_size * block_size + 1;

  for (nzc_context = 0; nzc_context < MAX_NZC_CONTEXTS; ++nzc_context) {
    for (r = 0; r < REF_TYPES; ++r) {
      for (b = 0; b < BLOCK_TYPES; ++b) {
186 187
        unsigned int *nzc_costs;
        if (block_size == 4) {
188 189 190
          vp9_cost_tokens(cost,
                          cpi->common.fc.nzc_probs_4x4[nzc_context][r][b],
                          vp9_nzc4x4_tree);
191 192
          nzc_costs = cpi->mb.nzc_costs_4x4[nzc_context][r][b];
        } else if (block_size == 8) {
193 194 195
          vp9_cost_tokens(cost,
                          cpi->common.fc.nzc_probs_8x8[nzc_context][r][b],
                          vp9_nzc8x8_tree);
196 197
          nzc_costs = cpi->mb.nzc_costs_8x8[nzc_context][r][b];
        } else if (block_size == 16) {
198 199 200
          vp9_cost_tokens(cost,
                          cpi->common.fc.nzc_probs_16x16[nzc_context][r][b],
                          vp9_nzc16x16_tree);
201 202
          nzc_costs = cpi->mb.nzc_costs_16x16[nzc_context][r][b];
        } else {
203 204 205
          vp9_cost_tokens(cost,
                          cpi->common.fc.nzc_probs_32x32[nzc_context][r][b],
                          vp9_nzc32x32_tree);
206 207
          nzc_costs = cpi->mb.nzc_costs_32x32[nzc_context][r][b];
        }
208 209 210 211 212

        for (nzc = 0; nzc < values; ++nzc) {
          int e, c, totalcost = 0;
          c = codenzc(nzc);
          totalcost = cost[c];
213 214
          if ((e = vp9_extranzcbits[c])) {
            int x = nzc - vp9_basenzcvalue[c];
215
            while (e--) {
216 217 218 219
              totalcost += vp9_cost_bit(
                  cpi->common.fc.nzc_pcat_probs[nzc_context]
                                               [c - NZC_TOKENS_NOEXTRA][e],
                  ((x >> e) & 1));
220 221
            }
          }
222
          nzc_costs[nzc] = totalcost;
223
        }
224 225 226
      }
    }
  }
John Koleszar's avatar
John Koleszar committed
227
}
228
#endif
John Koleszar's avatar
John Koleszar committed
229

230

231 232 233 234
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
235

236
// 3* dc_qlookup[Q]*dc_qlookup[Q];
237

238
/* values are now correlated to quantizer */
Paul Wilkins's avatar
Paul Wilkins committed
239 240 241
static int sad_per_bit16lut[QINDEX_RANGE];
static int sad_per_bit4lut[QINDEX_RANGE];

242
void vp9_init_me_luts() {
John Koleszar's avatar
John Koleszar committed
243 244 245 246 247 248 249
  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] =
250 251
      (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
252
  }
Paul Wilkins's avatar
Paul Wilkins committed
253
}
John Koleszar's avatar
John Koleszar committed
254

255
static int compute_rd_mult(int qindex) {
Dmitry Kovalev's avatar
Dmitry Kovalev committed
256
  int q = vp9_dc_quant(qindex, 0);
257
  return (11 * q * q) >> 2;
258 259
}

Dmitry Kovalev's avatar
Dmitry Kovalev committed
260 261 262
void vp9_initialize_me_consts(VP9_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
263 264
}

265

Dmitry Kovalev's avatar
Dmitry Kovalev committed
266
void vp9_initialize_rd_consts(VP9_COMP *cpi, int qindex) {
267
  int q, i;
John Koleszar's avatar
John Koleszar committed
268

269
  vp9_clear_system_state();  // __asm emms;
John Koleszar's avatar
John Koleszar committed
270

John Koleszar's avatar
John Koleszar committed
271 272 273 274
  // 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)
Dmitry Kovalev's avatar
Dmitry Kovalev committed
275
  qindex = (qindex < 0) ? 0 : ((qindex > MAXQ) ? MAXQ : qindex);
276

Dmitry Kovalev's avatar
Dmitry Kovalev committed
277
  cpi->RDMULT = compute_rd_mult(qindex);
John Koleszar's avatar
John Koleszar committed
278 279 280 281 282
  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 +=
Dmitry Kovalev's avatar
Dmitry Kovalev committed
283
          (cpi->RDMULT * rd_iifactor[cpi->twopass.next_iiratio]) >> 4;
John Koleszar's avatar
John Koleszar committed
284
  }
285
  cpi->mb.errorperbit = cpi->RDMULT >> 6;
John Koleszar's avatar
John Koleszar committed
286
  cpi->mb.errorperbit += (cpi->mb.errorperbit == 0);
287

288
  vp9_set_speed_features(cpi);
John Koleszar's avatar
John Koleszar committed
289

Dmitry Kovalev's avatar
Dmitry Kovalev committed
290 291
  q = (int)pow(vp9_dc_quant(qindex, 0) >> 2, 1.25);
  q <<= 2;
John Koleszar's avatar
John Koleszar committed
292 293
  if (q < 8)
    q = 8;
294

John Koleszar's avatar
John Koleszar committed
295 296 297
  if (cpi->RDMULT > 1000) {
    cpi->RDDIV = 1;
    cpi->RDMULT /= 100;
John Koleszar's avatar
John Koleszar committed
298

John Koleszar's avatar
John Koleszar committed
299 300 301 302 303 304
    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
305

John Koleszar's avatar
John Koleszar committed
306
      cpi->rd_baseline_thresh[i] = cpi->rd_threshes[i];
John Koleszar's avatar
John Koleszar committed
307
    }
John Koleszar's avatar
John Koleszar committed
308 309
  } else {
    cpi->RDDIV = 100;
John Koleszar's avatar
John Koleszar committed
310

John Koleszar's avatar
John Koleszar committed
311 312 313 314 315 316
    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
317

John Koleszar's avatar
John Koleszar committed
318
      cpi->rd_baseline_thresh[i] = cpi->rd_threshes[i];
John Koleszar's avatar
John Koleszar committed
319
    }
John Koleszar's avatar
John Koleszar committed
320
  }
John Koleszar's avatar
John Koleszar committed
321

322
  fill_token_costs(cpi->mb.token_costs[TX_4X4],
323
                   cpi->common.fc.coef_probs_4x4, BLOCK_TYPES);
324
  fill_token_costs(cpi->mb.token_costs[TX_8X8],
325
                   cpi->common.fc.coef_probs_8x8, BLOCK_TYPES);
326
  fill_token_costs(cpi->mb.token_costs[TX_16X16],
327
                   cpi->common.fc.coef_probs_16x16, BLOCK_TYPES);
328
  fill_token_costs(cpi->mb.token_costs[TX_32X32],
329
                   cpi->common.fc.coef_probs_32x32, BLOCK_TYPES);
330 331 332 333 334 335
#if CONFIG_CODE_NONZEROCOUNT
  fill_nzc_costs(cpi, 4);
  fill_nzc_costs(cpi, 8);
  fill_nzc_costs(cpi, 16);
  fill_nzc_costs(cpi, 32);
#endif
336

John Koleszar's avatar
John Koleszar committed
337 338
  /*rough estimate for costing*/
  cpi->common.kf_ymode_probs_index = cpi->common.base_qindex >> 4;
339
  vp9_init_mode_costs(cpi);
John Koleszar's avatar
John Koleszar committed
340

341
  if (cpi->common.frame_type != KEY_FRAME) {
342
    vp9_build_nmv_cost_table(
343 344 345 346 347 348
        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
349 350
}

351
int vp9_block_error_c(int16_t *coeff, int16_t *dqcoeff, int block_size) {
352
  int i, error = 0;
John Koleszar's avatar
John Koleszar committed
353

354
  for (i = 0; i < block_size; i++) {
John Koleszar's avatar
John Koleszar committed
355 356 357
    int this_diff = coeff[i] - dqcoeff[i];
    error += this_diff * this_diff;
  }
John Koleszar's avatar
John Koleszar committed
358

John Koleszar's avatar
John Koleszar committed
359
  return error;
John Koleszar's avatar
John Koleszar committed
360 361
}

362
int vp9_mbblock_error_c(MACROBLOCK *mb) {
John Koleszar's avatar
John Koleszar committed
363 364 365 366
  BLOCK  *be;
  BLOCKD *bd;
  int i, j;
  int berror, error = 0;
John Koleszar's avatar
John Koleszar committed
367

John Koleszar's avatar
John Koleszar committed
368 369 370 371
  for (i = 0; i < 16; i++) {
    be = &mb->block[i];
    bd = &mb->e_mbd.block[i];
    berror = 0;
372
    for (j = 0; j < 16; j++) {
John Koleszar's avatar
John Koleszar committed
373 374
      int this_diff = be->coeff[j] - bd->dqcoeff[j];
      berror += this_diff * this_diff;
John Koleszar's avatar
John Koleszar committed
375
    }
John Koleszar's avatar
John Koleszar committed
376 377 378
    error += berror;
  }
  return error;
John Koleszar's avatar
John Koleszar committed
379 380
}

381
int vp9_mbuverror_c(MACROBLOCK *mb) {
John Koleszar's avatar
John Koleszar committed
382 383
  BLOCK  *be;
  BLOCKD *bd;
John Koleszar's avatar
John Koleszar committed
384

385
  int i, error = 0;
John Koleszar's avatar
John Koleszar committed
386

John Koleszar's avatar
John Koleszar committed
387 388 389
  for (i = 16; i < 24; i++) {
    be = &mb->block[i];
    bd = &mb->e_mbd.block[i];
John Koleszar's avatar
John Koleszar committed
390

391
    error += vp9_block_error_c(be->coeff, bd->dqcoeff, 16);
John Koleszar's avatar
John Koleszar committed
392
  }
John Koleszar's avatar
John Koleszar committed
393

John Koleszar's avatar
John Koleszar committed
394
  return error;
John Koleszar's avatar
John Koleszar committed
395 396
}

397
int vp9_uvsse(MACROBLOCK *x) {
398 399 400
  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
401 402 403 404
  int uv_stride = x->block[16].src_stride;

  unsigned int sse1 = 0;
  unsigned int sse2 = 0;
405 406
  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
407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427
  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) {
428
    vp9_sub_pixel_variance8x8(uptr, pre_stride, (mv_col & 7) << 1,
429
                              (mv_row & 7) << 1, upred_ptr, uv_stride, &sse2);
430
    vp9_sub_pixel_variance8x8(vptr, pre_stride, (mv_col & 7) << 1,
431
                              (mv_row & 7) << 1, vpred_ptr, uv_stride, &sse1);
John Koleszar's avatar
John Koleszar committed
432 433
    sse2 += sse1;
  } else {
434 435
    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
436 437 438
    sse2 += sse1;
  }
  return sse2;
John Koleszar's avatar
John Koleszar committed
439 440
}

441
static INLINE int cost_coeffs(VP9_COMMON *const cm, MACROBLOCK *mb,
442
                              int ib, PLANE_TYPE type,
443 444 445
                              ENTROPY_CONTEXT *a,
                              ENTROPY_CONTEXT *l,
                              TX_SIZE tx_size) {
446
  MACROBLOCKD *const xd = &mb->e_mbd;
447 448
  MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
  int pt;
449
  const int eob = xd->eobs[ib];
450
  int c = 0;
451
  int cost = 0;
452
  const int *scan;
453 454
  const int16_t *qcoeff_ptr = xd->qcoeff + ib * 16;
  const int ref = mbmi->ref_frame != INTRA_FRAME;
455
  unsigned int (*token_costs)[PREV_COEF_CONTEXTS][MAX_ENTROPY_TOKENS] =
456
      mb->token_costs[tx_size][type][ref];
457
  ENTROPY_CONTEXT a_ec, l_ec;
458 459 460 461
  ENTROPY_CONTEXT *const a1 = a +
      sizeof(ENTROPY_CONTEXT_PLANES)/sizeof(ENTROPY_CONTEXT);
  ENTROPY_CONTEXT *const l1 = l +
      sizeof(ENTROPY_CONTEXT_PLANES)/sizeof(ENTROPY_CONTEXT);
462

463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478
#if CONFIG_CODE_NONZEROCOUNT
  int nzc_context = vp9_get_nzc_context(cm, xd, ib);
  unsigned int *nzc_cost;
#else
  int seg_eob;
  const int segment_id = xd->mode_info_context->mbmi.segment_id;
#endif

  // Check for consistency of tx_size with mode info
  if (type == PLANE_TYPE_Y_WITH_DC) {
    assert(xd->mode_info_context->mbmi.txfm_size == tx_size);
  } else {
    TX_SIZE tx_size_uv = get_uv_tx_size(xd);
    assert(tx_size == tx_size_uv);
  }

479
  switch (tx_size) {
480 481 482
    case TX_4X4: {
      const TX_TYPE tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
                              get_tx_type_4x4(xd, ib) : DCT_DCT;
483 484
      a_ec = *a;
      l_ec = *l;
485 486 487
#if CONFIG_CODE_NONZEROCOUNT
      nzc_cost = mb->nzc_costs_4x4[nzc_context][ref][type];
#else
488
      seg_eob = 16;
489
#endif
490 491 492 493 494 495
      if (tx_type == ADST_DCT) {
        scan = vp9_row_scan_4x4;
      } else if (tx_type == DCT_ADST) {
        scan = vp9_col_scan_4x4;
      } else {
        scan = vp9_default_zig_zag1d_4x4;
Daniel Kang's avatar
Daniel Kang committed
496 497
      }
      break;
498
    }
Daniel Kang's avatar
Daniel Kang committed
499
    case TX_8X8:
500 501
      a_ec = (a[0] + a[1]) != 0;
      l_ec = (l[0] + l[1]) != 0;
502
      scan = vp9_default_zig_zag1d_8x8;
503 504 505
#if CONFIG_CODE_NONZEROCOUNT
      nzc_cost = mb->nzc_costs_8x8[nzc_context][ref][type];
#else
506
      seg_eob = 64;
507
#endif
Daniel Kang's avatar
Daniel Kang committed
508 509
      break;
    case TX_16X16:
510
      scan = vp9_default_zig_zag1d_16x16;
511 512 513
#if CONFIG_CODE_NONZEROCOUNT
      nzc_cost = mb->nzc_costs_16x16[nzc_context][ref][type];
#else
514
      seg_eob = 256;
515
#endif
516
      if (type == PLANE_TYPE_UV) {
517 518 519 520 521
        a_ec = (a[0] + a[1] + a1[0] + a1[1]) != 0;
        l_ec = (l[0] + l[1] + l1[0] + l1[1]) != 0;
      } else {
        a_ec = (a[0] + a[1] + a[2] + a[3]) != 0;
        l_ec = (l[0] + l[1] + l[2] + l[3]) != 0;
Deb Mukherjee's avatar
Deb Mukherjee committed
522
      }
Daniel Kang's avatar
Daniel Kang committed
523
      break;
524 525
    case TX_32X32:
      scan = vp9_default_zig_zag1d_32x32;
526 527 528
#if CONFIG_CODE_NONZEROCOUNT
      nzc_cost = mb->nzc_costs_32x32[nzc_context][ref][type];
#else
529
      seg_eob = 1024;
530
#endif
531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546
      if (type == PLANE_TYPE_UV) {
        ENTROPY_CONTEXT *a2, *a3, *l2, *l3;
        a2 = a1 + sizeof(ENTROPY_CONTEXT_PLANES) / sizeof(ENTROPY_CONTEXT);
        a3 = a2 + sizeof(ENTROPY_CONTEXT_PLANES) / sizeof(ENTROPY_CONTEXT);
        l2 = l1 + sizeof(ENTROPY_CONTEXT_PLANES) / sizeof(ENTROPY_CONTEXT);
        l3 = l2 + sizeof(ENTROPY_CONTEXT_PLANES) / sizeof(ENTROPY_CONTEXT);
        a_ec = (a[0] + a[1] + a1[0] + a1[1] +
                a2[0] + a2[1] + a3[0] + a3[1]) != 0;
        l_ec = (l[0] + l[1] + l1[0] + l1[1] +
                l2[0] + l2[1] + l3[0] + l3[1]) != 0;
      } else {
        a_ec = (a[0] + a[1] + a[2] + a[3] +
                a1[0] + a1[1] + a1[2] + a1[3]) != 0;
        l_ec = (l[0] + l[1] + l[2] + l[3] +
                l1[0] + l1[1] + l1[2] + l1[3]) != 0;
      }
547
      break;
Daniel Kang's avatar
Daniel Kang committed
548
    default:
549
      abort();
Daniel Kang's avatar
Daniel Kang committed
550 551
      break;
  }
552

553 554
  VP9_COMBINEENTROPYCONTEXTS(pt, a_ec, l_ec);

555
#if CONFIG_CODE_NONZEROCOUNT == 0
Paul Wilkins's avatar
Paul Wilkins committed
556 557
  if (vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP))
    seg_eob = 0;
558
#endif
559

560
  {
561
    int recent_energy = 0;
562 563 564
#if CONFIG_CODE_NONZEROCOUNT
    int nzc = 0;
#endif
565 566
    for (; c < eob; c++) {
      int v = qcoeff_ptr[scan[c]];
567
      int t = vp9_dct_value_tokens_ptr[v].Token;
568 569 570
#if CONFIG_CODE_NONZEROCOUNT
      nzc += (v != 0);
#endif
571
      cost += token_costs[get_coef_band(tx_size, c)][pt][t];
572
      cost += vp9_dct_value_cost_ptr[v];
573
      pt = vp9_get_coef_context(&recent_energy, t);
574
    }
575 576 577
#if CONFIG_CODE_NONZEROCOUNT
    cost += nzc_cost[nzc];
#else
578
    if (c < seg_eob)
579
      cost += mb->token_costs[tx_size][type][ref][get_coef_band(tx_size, c)]
580
          [pt][DCT_EOB_TOKEN];
581
#endif
582 583
  }

584
  // is eob first coefficient;
585
  pt = (c > 0);
586
  *a = *l = pt;
587 588 589 590 591 592 593 594 595 596 597 598 599 600
  if (tx_size >= TX_8X8) {
    a[1] = l[1] = pt;
    if (tx_size >= TX_16X16) {
      if (type == PLANE_TYPE_UV) {
        a1[0] = a1[1] = l1[0] = l1[1] = pt;
      } else {
        a[2] = a[3] = l[2] = l[3] = pt;
        if (tx_size >= TX_32X32) {
          a1[0] = a1[1] = a1[2] = a1[3] = pt;
          l1[0] = l1[1] = l1[2] = l1[3] = pt;
        }
      }
    }
  }
601 602 603
  return cost;
}

604
static int rdcost_mby_4x4(VP9_COMMON *const cm, MACROBLOCK *mb) {
John Koleszar's avatar
John Koleszar committed
605 606
  int cost = 0;
  int b;
Paul Wilkins's avatar
Paul Wilkins committed
607
  MACROBLOCKD *xd = &mb->e_mbd;
John Koleszar's avatar
John Koleszar committed
608
  ENTROPY_CONTEXT_PLANES t_above, t_left;
609 610
  ENTROPY_CONTEXT *ta = (ENTROPY_CONTEXT *)&t_above;
  ENTROPY_CONTEXT *tl = (ENTROPY_CONTEXT *)&t_left;
John Koleszar's avatar
John Koleszar committed
611

612 613
  vpx_memcpy(&t_above, xd->above_context, sizeof(t_above));
  vpx_memcpy(&t_left, xd->left_context, sizeof(t_left));
John Koleszar's avatar
John Koleszar committed
614

John Koleszar's avatar
John Koleszar committed
615
  for (b = 0; b < 16; b++)
616
    cost += cost_coeffs(cm, mb, b, PLANE_TYPE_Y_WITH_DC,
Yaowu Xu's avatar
Yaowu Xu committed
617 618
                        ta + vp9_block2above[TX_4X4][b],
                        tl + vp9_block2left[TX_4X4][b],
Daniel Kang's avatar
Daniel Kang committed
619
                        TX_4X4);
John Koleszar's avatar
John Koleszar committed
620

John Koleszar's avatar
John Koleszar committed
621
  return cost;
John Koleszar's avatar
John Koleszar committed
622 623
}

624 625
static void macro_block_yrd_4x4(VP9_COMMON *const cm,
                                MACROBLOCK *mb,
626 627 628
                                int *rate,
                                int *distortion,
                                int *skippable) {
Paul Wilkins's avatar
Paul Wilkins committed
629
  MACROBLOCKD *const xd = &mb->e_mbd;
630

631 632 633
  xd->mode_info_context->mbmi.txfm_size = TX_4X4;
  vp9_transform_mby_4x4(mb);
  vp9_quantize_mby_4x4(mb);
634

635 636
  *distortion = vp9_mbblock_error(mb) >> 2;
  *rate = rdcost_mby_4x4(cm, mb);
637
  *skippable = vp9_mby_is_skippable_4x4(xd);
638
}
John Koleszar's avatar
John Koleszar committed
639

640
static int rdcost_mby_8x8(VP9_COMMON *const cm, MACROBLOCK *mb) {
John Koleszar's avatar
John Koleszar committed
641 642
  int cost = 0;
  int b;
Paul Wilkins's avatar
Paul Wilkins committed
643
  MACROBLOCKD *xd = &mb->e_mbd;
John Koleszar's avatar
John Koleszar committed
644
  ENTROPY_CONTEXT_PLANES t_above, t_left;
645 646
  ENTROPY_CONTEXT *ta = (ENTROPY_CONTEXT *)&t_above;
  ENTROPY_CONTEXT *tl = (ENTROPY_CONTEXT *)&t_left;
John Koleszar's avatar
John Koleszar committed
647

648 649
  vpx_memcpy(&t_above, xd->above_context, sizeof(t_above));
  vpx_memcpy(&t_left,  xd->left_context, sizeof(t_left));
John Koleszar's avatar
John Koleszar committed
650 651

  for (b = 0; b < 16; b += 4)
652
    cost += cost_coeffs(cm, mb, b, PLANE_TYPE_Y_WITH_DC,
Yaowu Xu's avatar
Yaowu Xu committed
653 654
                        ta + vp9_block2above[TX_8X8][b],
                        tl + vp9_block2left[TX_8X8][b],
Daniel Kang's avatar
Daniel Kang committed
655
                        TX_8X8);
John Koleszar's avatar
John Koleszar committed
656 657

  return cost;
658 659
}

660 661
static void macro_block_yrd_8x8(VP9_COMMON *const cm,
                                MACROBLOCK *mb,
662 663 664
                                int *rate,
                                int *distortion,
                                int *skippable) {
Paul Wilkins's avatar
Paul Wilkins committed
665
  MACROBLOCKD *const xd = &mb->e_mbd;
666 667

  xd->mode_info_context->mbmi.txfm_size = TX_8X8;
668 669
  vp9_transform_mby_8x8(mb);
  vp9_quantize_mby_8x8(mb);
John Koleszar's avatar
John Koleszar committed
670

671 672
  *distortion = vp9_mbblock_error(mb) >> 2;
  *rate = rdcost_mby_8x8(cm, mb);
673
  *skippable = vp9_mby_is_skippable_8x8(xd);
674
}
675

676 677
static int rdcost_mby_16x16(VP9_COMMON *const cm, MACROBLOCK *mb) {
  MACROBLOCKD *const xd = &mb->e_mbd;
Daniel Kang's avatar
Daniel Kang committed
678
  ENTROPY_CONTEXT_PLANES t_above, t_left;
679 680
  ENTROPY_CONTEXT *ta = (ENTROPY_CONTEXT *)&t_above;
  ENTROPY_CONTEXT *tl = (ENTROPY_CONTEXT *)&t_left;
Daniel Kang's avatar
Daniel Kang committed
681

682 683
  vpx_memcpy(&t_above, xd->above_context, sizeof(t_above));
  vpx_memcpy(&t_left, xd->left_context, sizeof(t_left));
Daniel Kang's avatar
Daniel Kang committed
684

685
  return cost_coeffs(cm, mb, 0, PLANE_TYPE_Y_WITH_DC, ta, tl, TX_16X16);
Daniel Kang's avatar
Daniel Kang committed
686
}
687

688
static void macro_block_yrd_16x16(VP9_COMMON *const cm, MACROBLOCK *mb,
689 690
                                  int *rate, int *distortion, int *skippable) {
  MACROBLOCKD *const xd = &mb->e_mbd;
Daniel Kang's avatar
Daniel Kang committed
691

692 693
  xd->mode_info_context->mbmi.txfm_size = TX_16X16;
  vp9_transform_mby_16x16(mb);
694
  vp9_quantize_mby_16x16(mb);
695 696 697
  // 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?
698 699
  if (mb->optimize &&
      xd->mode_info_context->mbmi.mode < I8X8_PRED)
700
    vp9_optimize_mby_16x16(cm, mb);
701

702 703
  *distortion = vp9_mbblock_error(mb) >> 2;
  *rate = rdcost_mby_16x16(cm, mb);
704
  *skippable = vp9_mby_is_skippable_16x16(xd);
Daniel Kang's avatar
Daniel Kang committed
705 706
}

707
static void choose_txfm_size_from_rd(VP9_COMP *cpi, MACROBLOCK *x,
708 709 710 711 712
                                     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) {
713 714 715 716 717
  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;
718 719 720 721 722 723 724 725 726 727 728 729
  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]);
    }
  }
730 731 732 733 734 735 736 737

  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);

738
    for (n = TX_4X4; n <= max_txfm_size; n++) {
739
      if (s[n]) {
740
        rd[n][0] = rd[n][1] = RDCOST(x->rdmult, x->rddiv, s1, d[n]);
741
      } else {
742 743
        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]);
744
      }
745 746
    }
  } else {
747 748 749
    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]);
750 751 752
    }
  }

753 754 755 756 757 758
  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;
Ronald S. Bultje's avatar
Ronald S. Bultje committed
759 760 761 762 763
  } else if ( cm->txfm_mode == ALLOW_16X16 ||
             (max_txfm_size == TX_16X16 && cm->txfm_mode == ALLOW_32X32) ||
             (cm->txfm_mode == TX_MODE_SELECT &&
              rd[TX_16X16][1] < rd[TX_8X8][1] &&
              rd[TX_16X16][1] < rd[TX_4X4][1])) {
764
    mbmi->txfm_size = TX_16X16;
765
  } else if (cm->txfm_mode == ALLOW_8X8 ||
766
           (cm->txfm_mode == TX_MODE_SELECT && rd[TX_8X8][1] < rd[TX_4X4][1])) {
767 768
    mbmi->txfm_size = TX_8X8;
  } else {
769
    assert(cm->txfm_mode == ONLY_4X4 || cm->txfm_mode == TX_MODE_SELECT);
770 771 772
    mbmi->txfm_size = TX_4X4;
  }

773
  *distortion = d[mbmi->txfm_size];
774
  *rate       = r[mbmi->txfm_size][cm->txfm_mode == TX_MODE_SELECT];
775 776
  *skip       = s[mbmi->txfm_size];

777 778 779 780 781 782 783 784
  txfm_cache[ONLY_4X4] = rd[TX_4X4][0];
  txfm_cache[ALLOW_8X8] = rd[TX_8X8][0];
  txfm_cache[ALLOW_16X16] = rd[TX_16X16][0];
  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];
Ronald S. Bultje's avatar
Ronald S. Bultje committed
785
  else if (rd[TX_16X16][1] < rd[TX_8X8][1] && rd[TX_16X16][1] < rd[TX_4X4][1])
786
    txfm_cache[TX_MODE_SELECT] = rd[TX_16X16][1];
787
  else
788 789
    txfm_cache[TX_MODE_SELECT] = rd[TX_4X4][1] < rd[TX_8X8][1] ?
                                 rd[TX_4X4][1] : rd[TX_8X8][1];
790 791 792 793 794
}

static void macro_block_yrd(VP9_COMP *cpi, MACROBLOCK *x, int *rate,
                            int *distortion, int *skippable,
                            int64_t txfm_cache[NB_TXFM_MODES]) {
795
  VP9_COMMON *const cm = &cpi->common;
796
  MACROBLOCKD *const xd = &x->e_mbd;
797
  int r[TX_SIZE_MAX_MB][2], d[TX_SIZE_MAX_MB], s[TX_SIZE_MAX_MB];
798 799 800

  vp9_subtract_mby(x->src_diff, *(x->block[0].base_src), xd->predictor,
                   x->block[0].src_stride);
801

802 803 804
  macro_block_yrd_16x16(cm, x, &r[TX_16X16][0], &d[TX_16X16], &s[TX_16X16]);
  macro_block_yrd_8x8(cm, x, &r[TX_8X8][0], &d[TX_8X8], &s[TX_8X8]);
  macro_block_yrd_4x4(cm, x, &r[TX_4X4][0], &d[TX_4X4], &s[TX_4X4]);
805 806

  choose_txfm_size_from_rd(cpi, x, r, rate, d, distortion, s, skippable,
807
                           txfm_cache, TX_16X16);
808 809
}

810
static void copy_predictor(uint8_t *dst, const uint8_t *predictor) {
John Koleszar's avatar
John Koleszar committed
811 812 813 814 815 816
  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];
817
}
Yaowu Xu's avatar
Yaowu Xu committed
818

819
static int vp9_sb_block_error_c(int16_t *coeff, int16_t *dqcoeff,
820
                                int block_size, int shift) {
821 822 823 824 825 826 827
  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;
  }
828
  error >>= shift;
829

Frank Galligan's avatar
Frank Galligan committed
830
  return error > INT_MAX ? INT_MAX : (int)error;