vp9_rdopt.c

/*
 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */


#include <stdio.h>
#include <math.h>
#include <limits.h>
#include <assert.h>
#include "vp9/common/vp9_pragmas.h"

#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"
#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"
#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"
#include "vpx_mem/vpx_mem.h"
#include "vp9/common/vp9_systemdependent.h"
#include "vp9/encoder/vp9_encodemv.h"

#include "vp9/common/vp9_seg_common.h"
#include "vp9/common/vp9_pred_common.h"
#include "vp9/common/vp9_entropy.h"
#include "vp9_rtcd.h"
#include "vp9/common/vp9_mvref_common.h"
#include "vp9/common/vp9_common.h"

#define MAXF(a,b)            (((a) > (b)) ? (a) : (b))

#define INVALID_MV 0x80008000

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

static const int auto_speed_thresh[17] = {
  1000,
  200,
  150,
  130,
  150,
  125,
  120,
  115,
  115,
  115,
  115,
  115,
  115,
  115,
  115,
  115,
  105
};

const MODE_DEFINITION vp9_mode_order[MAX_MODES] = {
  {ZEROMV,    LAST_FRAME,   NONE},
  {DC_PRED,   INTRA_FRAME,  NONE},

  {NEARESTMV, LAST_FRAME,   NONE},
  {NEARMV,    LAST_FRAME,   NONE},

  {ZEROMV,    GOLDEN_FRAME, NONE},
  {NEARESTMV, GOLDEN_FRAME, NONE},

  {ZEROMV,    ALTREF_FRAME, NONE},
  {NEARESTMV, ALTREF_FRAME, NONE},

  {NEARMV,    GOLDEN_FRAME, NONE},
  {NEARMV,    ALTREF_FRAME, NONE},

  {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},

  {TM_PRED,   INTRA_FRAME,  NONE},

  {NEWMV,     LAST_FRAME,   NONE},
  {NEWMV,     GOLDEN_FRAME, NONE},
  {NEWMV,     ALTREF_FRAME, NONE},

  {SPLITMV,   LAST_FRAME,   NONE},
  {SPLITMV,   GOLDEN_FRAME, NONE},
  {SPLITMV,   ALTREF_FRAME, NONE},

  {B_PRED,    INTRA_FRAME,  NONE},
  {I8X8_PRED, INTRA_FRAME,  NONE},

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

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

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

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

  {SPLITMV,   LAST_FRAME,   GOLDEN_FRAME},
  {SPLITMV,   ALTREF_FRAME, LAST_FRAME  },
  {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
};

static void fill_token_costs(vp9_coeff_count *c,
                             vp9_coeff_probs *p,
                             int block_type_counts) {
  int i, j, k, l;

  for (i = 0; i < block_type_counts; i++)
    for (j = 0; j < REF_TYPES; j++)
      for (k = 0; k < COEF_BANDS; k++)
        for (l = 0; l < PREV_COEF_CONTEXTS; l++) {
          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);
        }
}


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, };

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

/* values are now correlated to quantizer */
static int sad_per_bit16lut[QINDEX_RANGE];
static int sad_per_bit4lut[QINDEX_RANGE];

void vp9_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 * vp9_convert_qindex_to_q(i)) + 2.4107);
    sad_per_bit4lut[i] = (int)((0.063 * vp9_convert_qindex_to_q(i)) + 2.742);
  }
}

static int compute_rd_mult(int qindex) {
  int q;

  q = vp9_dc_quant(qindex, 0);
  return (11 * q * q) >> 6;
}

void vp9_initialize_me_consts(VP9_COMP *cpi, int QIndex) {
  cpi->mb.sadperbit16 =  sad_per_bit16lut[QIndex];
  cpi->mb.sadperbit4  =  sad_per_bit4lut[QIndex];
}


void vp9_initialize_rd_consts(VP9_COMP *cpi, int QIndex) {
  int q, i;

  vp9_clear_system_state();  // __asm emms;

  // 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);

  cpi->RDMULT = compute_rd_mult(QIndex);

  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;
  }

  if (cpi->RDMULT < 7)
    cpi->RDMULT = 7;

  cpi->mb.errorperbit = (cpi->RDMULT / 110);
  cpi->mb.errorperbit += (cpi->mb.errorperbit == 0);

  vp9_set_speed_features(cpi);

  q = (int)pow(vp9_dc_quant(QIndex, 0) >> 2, 1.25);
  q = q << 2;
  cpi->RDMULT = cpi->RDMULT << 4;

  if (q < 8)
    q = 8;

  if (cpi->RDMULT > 1000) {
    cpi->RDDIV = 1;
    cpi->RDMULT /= 100;

    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;
      }

      cpi->rd_baseline_thresh[i] = cpi->rd_threshes[i];
    }
  } else {
    cpi->RDDIV = 100;

    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;
      }

      cpi->rd_baseline_thresh[i] = cpi->rd_threshes[i];
    }
  }

  fill_token_costs(cpi->mb.token_costs[TX_4X4],
                   cpi->common.fc.coef_probs_4x4, BLOCK_TYPES);
  fill_token_costs(cpi->mb.token_costs[TX_8X8],
                   cpi->common.fc.coef_probs_8x8, BLOCK_TYPES);
  fill_token_costs(cpi->mb.token_costs[TX_16X16],
                   cpi->common.fc.coef_probs_16x16, BLOCK_TYPES);
  fill_token_costs(cpi->mb.token_costs[TX_32X32],
                   cpi->common.fc.coef_probs_32x32, BLOCK_TYPES_32X32);

  /*rough estimate for costing*/
  cpi->common.kf_ymode_probs_index = cpi->common.base_qindex >> 4;
  vp9_init_mode_costs(cpi);

  if (cpi->common.frame_type != KEY_FRAME) {
    vp9_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);
  }
}

int vp9_block_error_c(int16_t *coeff, int16_t *dqcoeff, int block_size) {
  int i, error = 0;

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

  return error;
}

int vp9_mbblock_error_c(MACROBLOCK *mb) {
  BLOCK  *be;
  BLOCKD *bd;
  int i, j;
  int berror, error = 0;

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

int vp9_mbuverror_c(MACROBLOCK *mb) {
  BLOCK  *be;
  BLOCKD *bd;

  int i, error = 0;

  for (i = 16; i < 24; i++) {
    be = &mb->block[i];
    bd = &mb->e_mbd.block[i];

    error += vp9_block_error_c(be->coeff, bd->dqcoeff, 16);
  }

  return error;
}

int vp9_uvsse(MACROBLOCK *x) {
  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);
  int uv_stride = x->block[16].src_stride;

  unsigned int sse1 = 0;
  unsigned int sse2 = 0;
  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;
  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) {
    vp9_sub_pixel_variance8x8(uptr, pre_stride, (mv_col & 7) << 1,
                              (mv_row & 7) << 1, upred_ptr, uv_stride, &sse2);
    vp9_sub_pixel_variance8x8(vptr, pre_stride, (mv_col & 7) << 1,
                              (mv_row & 7) << 1, vpred_ptr, uv_stride, &sse1);
    sse2 += sse1;
  } else {
    vp9_variance8x8(uptr, pre_stride, upred_ptr, uv_stride, &sse2);
    vp9_variance8x8(vptr, pre_stride, vpred_ptr, uv_stride, &sse1);
    sse2 += sse1;
  }
  return sse2;

}

static INLINE int cost_coeffs(MACROBLOCK *mb,
                              BLOCKD *b, PLANE_TYPE type,
                              ENTROPY_CONTEXT *a,
                              ENTROPY_CONTEXT *l,
                              TX_SIZE tx_size) {
  int pt;
  const int eob = b->eob;
  MACROBLOCKD *xd = &mb->e_mbd;
  const int ib = (int)(b - xd->block);
  int c = 0;
  int cost = 0, seg_eob;
  const int segment_id = xd->mode_info_context->mbmi.segment_id;
  const int *scan;
  int16_t *qcoeff_ptr = b->qcoeff;
  const int ref = xd->mode_info_context->mbmi.ref_frame != INTRA_FRAME;
  const TX_TYPE tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
                          get_tx_type(xd, b) : DCT_DCT;
  unsigned int (*token_costs)[PREV_COEF_CONTEXTS][MAX_ENTROPY_TOKENS] =
      mb->token_costs[tx_size][type][ref];
  ENTROPY_CONTEXT a_ec = *a, l_ec = *l;

  switch (tx_size) {
    case TX_4X4:
      scan = vp9_default_zig_zag1d_4x4;
      seg_eob = 16;
      if (type == PLANE_TYPE_Y_WITH_DC) {
        if (tx_type == ADST_DCT) {
          scan = vp9_row_scan_4x4;
        } else if (tx_type == DCT_ADST) {
          scan = vp9_col_scan_4x4;
        }
      }
      break;
    case TX_8X8:
      scan = vp9_default_zig_zag1d_8x8;
      seg_eob = 64;
      break;
    case TX_16X16:
      scan = vp9_default_zig_zag1d_16x16;
      seg_eob = 256;
      if (type == PLANE_TYPE_UV) {
        const int uv_idx = ib - 16;
        qcoeff_ptr = xd->sb_coeff_data.qcoeff + 1024 + 64 * uv_idx;
      }
      break;
    case TX_32X32:
      scan = vp9_default_zig_zag1d_32x32;
      seg_eob = 1024;
      qcoeff_ptr = xd->sb_coeff_data.qcoeff;
      break;
    default:
      abort();
      break;
  }

  VP9_COMBINEENTROPYCONTEXTS(pt, a_ec, l_ec);

  if (vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP))
    seg_eob = 0;

  {
    int recent_energy = 0;
    for (; c < eob; c++) {
      int v = qcoeff_ptr[scan[c]];
      int t = vp9_dct_value_tokens_ptr[v].Token;
      cost += token_costs[get_coef_band(tx_size, c)][pt][t];
      cost += vp9_dct_value_cost_ptr[v];
      pt = vp9_get_coef_context(&recent_energy, t);
    }
    if (c < seg_eob)
      cost += mb->token_costs[tx_size][type][ref][get_coef_band(tx_size, c)]
          [pt][DCT_EOB_TOKEN];
  }

  // is eob first coefficient;
  pt = (c > 0);
  *a = *l = pt;
  return cost;
}

static int rdcost_mby_4x4(MACROBLOCK *mb, int backup) {
  int cost = 0;
  int b;
  MACROBLOCKD *xd = &mb->e_mbd;
  ENTROPY_CONTEXT_PLANES t_above, t_left;
  ENTROPY_CONTEXT *ta;
  ENTROPY_CONTEXT *tl;

  if (backup) {
    vpx_memcpy(&t_above, xd->above_context, sizeof(ENTROPY_CONTEXT_PLANES));
    vpx_memcpy(&t_left, xd->left_context, sizeof(ENTROPY_CONTEXT_PLANES));

    ta = (ENTROPY_CONTEXT *)&t_above;
    tl = (ENTROPY_CONTEXT *)&t_left;
  } else {
    ta = (ENTROPY_CONTEXT *)xd->above_context;
    tl = (ENTROPY_CONTEXT *)xd->left_context;
  }

  for (b = 0; b < 16; b++)
    cost += cost_coeffs(mb, xd->block + b, PLANE_TYPE_Y_WITH_DC,
                        ta + vp9_block2above[TX_4X4][b],
                        tl + vp9_block2left[TX_4X4][b],
                        TX_4X4);

  return cost;
}

static void macro_block_yrd_4x4(MACROBLOCK *mb,
                                int *Rate,
                                int *Distortion,
                                int *skippable, int backup) {
  MACROBLOCKD *const xd = &mb->e_mbd;

  xd->mode_info_context->mbmi.txfm_size = TX_4X4;
  vp9_transform_mby_4x4(mb);
  vp9_quantize_mby_4x4(mb);

  *Distortion = vp9_mbblock_error(mb) >> 2;
  *Rate = rdcost_mby_4x4(mb, backup);
  *skippable = vp9_mby_is_skippable_4x4(xd);
}

static int rdcost_mby_8x8(MACROBLOCK *mb, int backup) {
  int cost = 0;
  int b;
  MACROBLOCKD *xd = &mb->e_mbd;
  ENTROPY_CONTEXT_PLANES t_above, t_left;
  ENTROPY_CONTEXT *ta;
  ENTROPY_CONTEXT *tl;

  if (backup) {
    vpx_memcpy(&t_above,xd->above_context, sizeof(ENTROPY_CONTEXT_PLANES));
    vpx_memcpy(&t_left, xd->left_context, sizeof(ENTROPY_CONTEXT_PLANES));

    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;
  }

  for (b = 0; b < 16; b += 4)
    cost += cost_coeffs(mb, xd->block + b, PLANE_TYPE_Y_WITH_DC,
                        ta + vp9_block2above[TX_8X8][b],
                        tl + vp9_block2left[TX_8X8][b],
                        TX_8X8);

  return cost;
}

static void macro_block_yrd_8x8(MACROBLOCK *mb,
                                int *Rate,
                                int *Distortion,
                                int *skippable, int backup) {
  MACROBLOCKD *const xd = &mb->e_mbd;

  xd->mode_info_context->mbmi.txfm_size = TX_8X8;
  vp9_transform_mby_8x8(mb);
  vp9_quantize_mby_8x8(mb);

  *Distortion = vp9_mbblock_error(mb) >> 2;
  *Rate = rdcost_mby_8x8(mb, backup);
  *skippable = vp9_mby_is_skippable_8x8(xd);
}

static int rdcost_mby_16x16(MACROBLOCK *mb, int backup) {
  int cost;
  MACROBLOCKD *xd = &mb->e_mbd;
  ENTROPY_CONTEXT_PLANES t_above, t_left;
  ENTROPY_CONTEXT *ta, *tl;

  if (backup) {
    vpx_memcpy(&t_above, xd->above_context, sizeof(ENTROPY_CONTEXT_PLANES));
    vpx_memcpy(&t_left, xd->left_context, sizeof(ENTROPY_CONTEXT_PLANES));

    ta = (ENTROPY_CONTEXT *)&t_above;
    tl = (ENTROPY_CONTEXT *)&t_left;
  } else {
    ta = (ENTROPY_CONTEXT *)xd->above_context;
    tl = (ENTROPY_CONTEXT *)xd->left_context;
  }

  cost = cost_coeffs(mb, xd->block, PLANE_TYPE_Y_WITH_DC, ta, tl, TX_16X16);
  return cost;
}

static void macro_block_yrd_16x16(MACROBLOCK *mb, int *Rate, int *Distortion,
                                  int *skippable, int backup) {
  MACROBLOCKD *xd = &mb->e_mbd;

  xd->mode_info_context->mbmi.txfm_size = TX_16X16;
  vp9_transform_mby_16x16(mb);
  vp9_quantize_mby_16x16(mb);
  // 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?
  if (mb->optimize &&
      xd->mode_info_context->mbmi.mode < I8X8_PRED)
    vp9_optimize_mby_16x16(mb);

  *Distortion = vp9_mbblock_error(mb) >> 2;
  *Rate = rdcost_mby_16x16(mb, backup);
  *skippable = vp9_mby_is_skippable_16x16(xd);
}

static void choose_txfm_size_from_rd(VP9_COMP *cpi, MACROBLOCK *x,
                                     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) {
  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;
  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]);
    }
  }

  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);

    for (n = TX_4X4; n <= max_txfm_size; n++) {
      if (s[n]) {
        rd[n][0] = rd[n][1] = RDCOST(x->rdmult, x->rddiv, s1, d[n]);
      } else {
        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]);
      }
    }
  } else {
    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]);
    }
  }

  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 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])) {
    mbmi->txfm_size = TX_16X16;
  } else if (cm->txfm_mode == ALLOW_8X8 ||
           (cm->txfm_mode == TX_MODE_SELECT && rd[TX_8X8][1] < rd[TX_4X4][1])) {
    mbmi->txfm_size = TX_8X8;
  } else {
    assert(cm->txfm_mode == ONLY_4X4 || cm->txfm_mode == TX_MODE_SELECT);
    mbmi->txfm_size = TX_4X4;
  }

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

  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];
  else 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];
  else
    txfm_cache[TX_MODE_SELECT] = rd[TX_4X4][1] < rd[TX_8X8][1] ?
                                 rd[TX_4X4][1] : rd[TX_8X8][1];
}

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;
  int r[TX_SIZE_MAX_MB][2], d[TX_SIZE_MAX_MB], s[TX_SIZE_MAX_MB];

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

  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);

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

static void copy_predictor(uint8_t *dst, const uint8_t *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];
}

static int rdcost_sby_32x32(MACROBLOCK *x, int backup) {
  MACROBLOCKD * const xd = &x->e_mbd;
  ENTROPY_CONTEXT_PLANES t_above, t_left;
  ENTROPY_CONTEXT *ta, *tl;

  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;
  }

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

static int vp9_sb_block_error_c(int16_t *coeff, int16_t *dqcoeff,
                                int block_size) {
  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,
                                  int *rate, int *distortion, int *skippable,
                                  int backup) {
  SUPERBLOCK  * const x_sb = &x->sb_coeff_data;
  MACROBLOCKD * const xd = &x->e_mbd;
  SUPERBLOCKD * const xd_sb = &xd->sb_coeff_data;
#if DEBUG_ERROR
  int16_t out[1024];
#endif

  vp9_transform_sby_32x32(x);
  vp9_quantize_sby_32x32(x);
#if DEBUG_ERROR
  vp9_short_idct32x32(xd_sb->dqcoeff, out, 64);
#endif

  *distortion = vp9_sb_block_error_c(x_sb->coeff, xd_sb->dqcoeff, 1024);

#if DEBUG_ERROR
  printf("IDCT/FDCT error 32x32: %d (d: %d)\n",
         vp9_block_error_c(x_sb->src_diff, out, 1024), *distortion);
#endif
  *rate       = rdcost_sby_32x32(x, backup);
  *skippable  = vp9_sby_is_skippable_32x32(&x->e_mbd);
}

static void super_block_yrd(VP9_COMP *cpi,
                            MACROBLOCK *x, int *rate, int *distortion,
                            int *skip,
                            int64_t txfm_cache[NB_TXFM_MODES]) {
  MACROBLOCKD *const xd = &x->e_mbd;
  int r[TX_SIZE_MAX_SB][2], d[TX_SIZE_MAX_SB], s[TX_SIZE_MAX_SB], 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 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;

  for (n = TX_4X4; n < TX_SIZE_MAX_MB; n++) {
    vpx_memcpy(t_above[n], xd->above_context, sizeof(t_above[n]));
    vpx_memcpy(t_left[n], xd->left_context, sizeof(t_left[n]));
    r[n][0] = 0;
    d[n] = 0;
    s[n] = 1;
  }

  vp9_subtract_sby_s_c(x->sb_coeff_data.src_diff, src, src_y_stride,
                       dst, dst_y_stride);
  super_block_yrd_32x32(x, &r[TX_32X32][0], &d[TX_32X32], &s[TX_32X32], 1);

#if DEBUG_ERROR
  int err[3] = { 0, 0, 0 };
#endif
  for (n = 0; n < 4; n++) {
    int x_idx = n & 1, y_idx = n >> 1;
    int r_tmp, d_tmp, s_tmp;

    vp9_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);

    xd->above_context = &t_above[TX_16X16][x_idx];
    xd->left_context = &t_left[TX_16X16][y_idx];
    macro_block_yrd_16x16(x, &r_tmp, &d_tmp, &s_tmp, 0);
    d[TX_16X16] += d_tmp;
    r[TX_16X16][0] += r_tmp;
    s[TX_16X16] = s[TX_16X16] && s_tmp;
#if DEBUG_ERROR
    vp9_inverse_transform_mby_16x16(xd);
    err[2] += vp9_block_error_c(xd->diff, x->src_diff, 256);
#endif

    xd->above_context = &t_above[TX_4X4][x_idx];
    xd->left_context = &t_left[TX_4X4][y_idx];
    macro_block_yrd_4x4(x, &r_tmp, &d_tmp, &s_tmp, 0);
    d[TX_4X4] += d_tmp;
    r[TX_4X4][0] += r_tmp;
    s[TX_4X4] = s[TX_4X4] && s_tmp;
#if DEBUG_ERROR
    vp9_inverse_transform_mby_4x4(xd);
    err[0] += vp9_block_error_c(xd->diff, x->src_diff, 256);
#endif

    xd->above_context = &t_above[TX_8X8][x_idx];
    xd->left_context = &t_left[TX_8X8][y_idx];
    macro_block_yrd_8x8(x, &r_tmp, &d_tmp, &s_tmp, 0);
    d[TX_8X8] += d_tmp;
    r[TX_8X8][0] += r_tmp;
    s[TX_8X8] = s[TX_8X8] && s_tmp;
#if DEBUG_ERROR
    vp9_inverse_transform_mby_8x8(xd);
    err[1] += vp9_block_error_c(xd->diff, x->src_diff, 256);
#endif
  }
#if DEBUG_ERROR
  printf("IDCT/FDCT error 16x16: %d (d: %d)\n", err[2], d[2]);
  printf("IDCT/FDCT error 8x8: %d (d: %d)\n", err[1], d[1]);
  printf("IDCT/FDCT error 4x4: %d (d: %d)\n", err[0], d[0]);
#endif
  choose_txfm_size_from_rd(cpi, x, r, rate, d, distortion, s, skip, txfm_cache,
                           TX_SIZE_MAX_SB - 1);

  xd->above_context = orig_above;
  xd->left_context = orig_left;
}

static void super_block_64_yrd(VP9_COMP *cpi,
                               MACROBLOCK *x, int *rate, int *distortion,
                               int *skip,
                               int64_t txfm_cache[NB_TXFM_MODES]) {
  MACROBLOCKD *const xd = &x->e_mbd;
  int r[TX_SIZE_MAX_SB][2], d[TX_SIZE_MAX_SB], s[TX_SIZE_MAX_SB], 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 t_above[TX_SIZE_MAX_SB][4],
                        *orig_above = xd->above_context;
  ENTROPY_CONTEXT_PLANES t_left[TX_SIZE_MAX_SB][4],
                        *orig_left = xd->left_context;

  for (n = TX_4X4; n < TX_SIZE_MAX_SB; n++) {
    vpx_memcpy(t_above[n], xd->above_context, sizeof(t_above[n]));
    vpx_memcpy(t_left[n], xd->left_context, sizeof(t_left[n]));
    r[n][0] = 0;
    d[n] = 0;
    s[n] = 1;
  }

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

    xd->above_context = &t_above[TX_32X32][x_idx << 1];
    xd->left_context = &t_left[TX_32X32][y_idx << 1];
    vp9_subtract_sby_s_c(x->sb_coeff_data.src_diff,
                         src + 32 * x_idx + 32 * y_idx * src_y_stride,
                         src_y_stride,
                         dst + 32 * x_idx + 32 * y_idx * dst_y_stride,
                         dst_y_stride);
    super_block_yrd_32x32(x, &r_tmp, &d_tmp, &s_tmp, 0);
    r[TX_32X32][0] += r_tmp;
    d[TX_32X32] += d_tmp;
    s[TX_32X32] = s[TX_32X32] && s_tmp;
  }

#if DEBUG_ERROR
  int err[3] = { 0, 0, 0 };
#endif
  for (n = 0; n < 16; n++) {
    int x_idx = n & 3, y_idx = n >> 2;
    int r_tmp, d_tmp, s_tmp;

    vp9_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);

    xd->above_context = &t_above[TX_16X16][x_idx];
    xd->left_context = &t_left[TX_16X16][y_idx];
    macro_block_yrd_16x16(x, &r_tmp, &d_tmp, &s_tmp, 0);
    d[TX_16X16] += d_tmp;
    r[TX_16X16][0] += r_tmp;
    s[TX_16X16] = s[TX_16X16] && s_tmp;
#if DEBUG_ERROR
    vp9_inverse_transform_mby_16x16(xd);
    err[2] += vp9_block_error_c(xd->diff, x->src_diff, 256);
#endif

    xd->above_context = &t_above[TX_4X4][x_idx];
    xd->left_context = &t_left[TX_4X4][y_idx];
    macro_block_yrd_4x4(x, &r_tmp, &d_tmp, &s_tmp, 0);
    d[TX_4X4] += d_tmp;
    r[TX_4X4][0] += r_tmp;
    s[TX_4X4] = s[TX_4X4] && s_tmp;
#if DEBUG_ERROR
    vp9_inverse_transform_mby_4x4(xd);
    err[0] += vp9_block_error_c(xd->diff, x->src_diff, 256);
#endif

    xd->above_context = &t_above[TX_8X8][x_idx];
    xd->left_context = &t_left[TX_8X8][y_idx];
    macro_block_yrd_8x8(x, &r_tmp, &d_tmp, &s_tmp, 0);
    d[TX_8X8] += d_tmp;
    r[TX_8X8][0] += r_tmp;
    s[TX_8X8] = s[TX_8X8] && s_tmp;
#if DEBUG_ERROR
    vp9_inverse_transform_mby_8x8(xd);
    err[1] += vp9_block_error_c(xd->diff, x->src_diff, 256);
#endif
  }
#if DEBUG_ERROR
  printf("IDCT/FDCT error 16x16: %d (d: %d)\n", err[2], d[2]);
  printf("IDCT/FDCT error 8x8: %d (d: %d)\n", err[1], d[1]);
  printf("IDCT/FDCT error 4x4: %d (d: %d)\n", err[0], d[0]);
#endif
  choose_txfm_size_from_rd(cpi, x, r, rate, d, distortion, s, skip, txfm_cache,
                           TX_SIZE_MAX_SB - 1);

  xd->above_context = orig_above;
  xd->left_context = orig_left;
}

static void copy_predictor_8x8(uint8_t *dst, const uint8_t *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];
}

static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, BLOCK *be,
                                     BLOCKD *b, B_PREDICTION_MODE *best_mode,
                                     int *bmode_costs,
                                     ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l,
                                     int *bestrate, int *bestratey,
                                     int *bestdistortion) {
  B_PREDICTION_MODE mode;
  MACROBLOCKD *xd = &x->e_mbd;
  int64_t best_rd = INT64_MAX;
  int rate = 0;
  int distortion;

  ENTROPY_CONTEXT ta = *a, tempa = *a;
  ENTROPY_CONTEXT tl = *l, templ = *l;
  TX_TYPE tx_type = DCT