Roll latest libvpx into Android.

Make the VP9 decoding 2X faster than the old one.
Checkout is from master branch(hash:242157c756314827ad9244952c7253e8900b9626).

Change-Id: Ibe67b3ee19f82b87df2416826b63a67f7f79b63a

1 files changed, 284 insertions, 584 deletions

diff --git a/libvpx/vp9/encoder/vp9_bitstream.c b/libvpx/vp9/encoder/vp9_bitstream.c
index 09ab2db..ad0f6c5 100644
--- a/libvpx/vp9/encoder/vp9_bitstream.c
+++ b/libvpx/vp9/encoder/vp9_bitstream.c
@@ -32,6 +32,7 @@
 #include "vp9/encoder/vp9_encodemv.h"
 #include "vp9/encoder/vp9_bitstream.h"
 #include "vp9/encoder/vp9_segmentation.h"
+#include "vp9/encoder/vp9_subexp.h"
 #include "vp9/encoder/vp9_write_bit_buffer.h"
 
 
@@ -48,8 +49,6 @@ vp9_coeff_stats tree_update_hist[TX_SIZE_MAX_SB][BLOCK_TYPES];
 extern unsigned int active_section;
 #endif
 
-#define vp9_cost_upd  ((int)(vp9_cost_one(upd) - vp9_cost_zero(upd)) >> 8)
-#define vp9_cost_upd256  ((int)(vp9_cost_one(upd) - vp9_cost_zero(upd)))
 
 #ifdef MODE_STATS
 int64_t tx_count_32x32p_stats[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB];
@@ -155,8 +154,6 @@ void write_switchable_interp_stats() {
 }
 #endif
 
-static int update_bits[255];
-
 static INLINE void write_be32(uint8_t *p, int value) {
   p[0] = value >> 24;
   p[1] = value >> 16;
@@ -164,248 +161,11 @@ static INLINE void write_be32(uint8_t *p, int value) {
   p[3] = value;
 }
 
-
-
-int recenter_nonneg(int v, int m) {
-  if (v > (m << 1))
-    return v;
-  else if (v >= m)
-    return ((v - m) << 1);
-  else
-    return ((m - v) << 1) - 1;
-}
-
-static int get_unsigned_bits(unsigned num_values) {
-  int cat = 0;
-  if ((num_values--) <= 1) return 0;
-  while (num_values > 0) {
-    cat++;
-    num_values >>= 1;
-  }
-  return cat;
-}
-
 void vp9_encode_unsigned_max(struct vp9_write_bit_buffer *wb,
                              int data, int max) {
   vp9_wb_write_literal(wb, data, get_unsigned_bits(max));
 }
 
-void encode_uniform(vp9_writer *w, int v, int n) {
-  int l = get_unsigned_bits(n);
-  int m;
-  if (l == 0)
-    return;
-  m = (1 << l) - n;
-  if (v < m) {
-    vp9_write_literal(w, v, l - 1);
-  } else {
-    vp9_write_literal(w, m + ((v - m) >> 1), l - 1);
-    vp9_write_literal(w, (v - m) & 1, 1);
-  }
-}
-
-int count_uniform(int v, int n) {
-  int l = get_unsigned_bits(n);
-  int m;
-  if (l == 0) return 0;
-  m = (1 << l) - n;
-  if (v < m)
-    return l - 1;
-  else
-    return l;
-}
-
-void encode_term_subexp(vp9_writer *w, int word, int k, int num_syms) {
-  int i = 0;
-  int mk = 0;
-  while (1) {
-    int b = (i ? k + i - 1 : k);
-    int a = (1 << b);
-    if (num_syms <= mk + 3 * a) {
-      encode_uniform(w, word - mk, num_syms - mk);
-      break;
-    } else {
-      int t = (word >= mk + a);
-      vp9_write_literal(w, t, 1);
-      if (t) {
-        i = i + 1;
-        mk += a;
-      } else {
-        vp9_write_literal(w, word - mk, b);
-        break;
-      }
-    }
-  }
-}
-
-int count_term_subexp(int word, int k, int num_syms) {
-  int count = 0;
-  int i = 0;
-  int mk = 0;
-  while (1) {
-    int b = (i ? k + i - 1 : k);
-    int a = (1 << b);
-    if (num_syms <= mk + 3 * a) {
-      count += count_uniform(word - mk, num_syms - mk);
-      break;
-    } else {
-      int t = (word >= mk + a);
-      count++;
-      if (t) {
-        i = i + 1;
-        mk += a;
-      } else {
-        count += b;
-        break;
-      }
-    }
-  }
-  return count;
-}
-
-static void compute_update_table() {
-  int i;
-  for (i = 0; i < 254; i++)
-    update_bits[i] = count_term_subexp(i, SUBEXP_PARAM, 255);
-}
-
-static int split_index(int i, int n, int modulus) {
-  int max1 = (n - 1 - modulus / 2) / modulus + 1;
-  if (i % modulus == modulus / 2) i = i / modulus;
-  else i = max1 + i - (i + modulus - modulus / 2) / modulus;
-  return i;
-}
-
-static int remap_prob(int v, int m) {
-  const int n = 255;
-  const int modulus = MODULUS_PARAM;
-  int i;
-  v--;
-  m--;
-  if ((m << 1) <= n)
-    i = recenter_nonneg(v, m) - 1;
-  else
-    i = recenter_nonneg(n - 1 - v, n - 1 - m) - 1;
-
-  i = split_index(i, n - 1, modulus);
-  return i;
-}
-
-static void write_prob_diff_update(vp9_writer *w,
-                                   vp9_prob newp, vp9_prob oldp) {
-  int delp = remap_prob(newp, oldp);
-  encode_term_subexp(w, delp, SUBEXP_PARAM, 255);
-}
-
-static int prob_diff_update_cost(vp9_prob newp, vp9_prob oldp) {
-  int delp = remap_prob(newp, oldp);
-  return update_bits[delp] * 256;
-}
-
-static int prob_update_savings(const unsigned int *ct,
-                               const vp9_prob oldp, const vp9_prob newp,
-                               const vp9_prob upd) {
-  const int old_b = cost_branch256(ct, oldp);
-  const int new_b = cost_branch256(ct, newp);
-  const int update_b = 2048 + vp9_cost_upd256;
-  return old_b - new_b - update_b;
-}
-
-static int prob_diff_update_savings_search(const unsigned int *ct,
-                                           const vp9_prob oldp, vp9_prob *bestp,
-                                           const vp9_prob upd) {
-  const int old_b = cost_branch256(ct, oldp);
-  int new_b, update_b, savings, bestsavings, step;
-  vp9_prob newp, bestnewp;
-
-  bestsavings = 0;
-  bestnewp = oldp;
-
-  step = (*bestp > oldp ? -1 : 1);
-  for (newp = *bestp; newp != oldp; newp += step) {
-    new_b = cost_branch256(ct, newp);
-    update_b = prob_diff_update_cost(newp, oldp) + vp9_cost_upd256;
-    savings = old_b - new_b - update_b;
-    if (savings > bestsavings) {
-      bestsavings = savings;
-      bestnewp = newp;
-    }
-  }
-  *bestp = bestnewp;
-  return bestsavings;
-}
-
-static int prob_diff_update_savings_search_model(const unsigned int *ct,
-                                                 const vp9_prob *oldp,
-                                                 vp9_prob *bestp,
-                                                 const vp9_prob upd,
-                                                 int b, int r) {
-  int i, old_b, new_b, update_b, savings, bestsavings, step;
-  int newp;
-  vp9_prob bestnewp, newplist[ENTROPY_NODES], oldplist[ENTROPY_NODES];
-  vp9_model_to_full_probs(oldp, oldplist);
-  vpx_memcpy(newplist, oldp, sizeof(vp9_prob) * UNCONSTRAINED_NODES);
-  for (i = UNCONSTRAINED_NODES, old_b = 0; i < ENTROPY_NODES; ++i)
-    old_b += cost_branch256(ct + 2 * i, oldplist[i]);
-  old_b += cost_branch256(ct + 2 * PIVOT_NODE, oldplist[PIVOT_NODE]);
-
-  bestsavings = 0;
-  bestnewp = oldp[PIVOT_NODE];
-
-  step = (*bestp > oldp[PIVOT_NODE] ? -1 : 1);
-  newp = *bestp;
-  for (; newp != oldp[PIVOT_NODE]; newp += step) {
-    if (newp < 1 || newp > 255) continue;
-    newplist[PIVOT_NODE] = newp;
-    vp9_model_to_full_probs(newplist, newplist);
-    for (i = UNCONSTRAINED_NODES, new_b = 0; i < ENTROPY_NODES; ++i)
-      new_b += cost_branch256(ct + 2 * i, newplist[i]);
-    new_b += cost_branch256(ct + 2 * PIVOT_NODE, newplist[PIVOT_NODE]);
-    update_b = prob_diff_update_cost(newp, oldp[PIVOT_NODE]) +
-        vp9_cost_upd256;
-    savings = old_b - new_b - update_b;
-    if (savings > bestsavings) {
-      bestsavings = savings;
-      bestnewp = newp;
-    }
-  }
-  *bestp = bestnewp;
-  return bestsavings;
-}
-
-static void vp9_cond_prob_update(vp9_writer *bc, vp9_prob *oldp, vp9_prob upd,
-                                 unsigned int *ct) {
-  vp9_prob newp;
-  int savings;
-  newp = get_binary_prob(ct[0], ct[1]);
-  assert(newp >= 1);
-  savings = prob_update_savings(ct, *oldp, newp, upd);
-  if (savings > 0) {
-    vp9_write(bc, 1, upd);
-    vp9_write_prob(bc, newp);
-    *oldp = newp;
-  } else {
-    vp9_write(bc, 0, upd);
-  }
-}
-
-static void vp9_cond_prob_diff_update(vp9_writer *bc, vp9_prob *oldp,
-                                      vp9_prob upd,
-                                      unsigned int *ct) {
-  vp9_prob newp;
-  int savings;
-  newp = get_binary_prob(ct[0], ct[1]);
-  assert(newp >= 1);
-  savings = prob_diff_update_savings_search(ct, *oldp, &newp, upd);
-  if (savings > 0) {
-    vp9_write(bc, 1, upd);
-    write_prob_diff_update(bc, newp, *oldp);
-    *oldp = newp;
-  } else {
-    vp9_write(bc, 0, upd);
-  }
-}
-
 static void update_mode(
   vp9_writer *w,
   int n,
@@ -440,16 +200,39 @@ static void update_mbintra_mode_probs(VP9_COMP* const cpi,
                 (unsigned int *)cpi->y_mode_count[j]);
 }
 
-void vp9_update_skip_probs(VP9_COMP *cpi, vp9_writer *bc) {
-  VP9_COMMON *const pc = &cpi->common;
-  int k;
+static void write_selected_txfm_size(const VP9_COMP *cpi, TX_SIZE tx_size,
+                                     BLOCK_SIZE_TYPE bsize, vp9_writer *w) {
+  const MACROBLOCKD *const xd = &cpi->mb.e_mbd;
+  const vp9_prob *tx_probs = get_tx_probs2(xd, &cpi->common.fc.tx_probs);
+  vp9_write(w, tx_size != TX_4X4, tx_probs[0]);
+  if (bsize >= BLOCK_SIZE_MB16X16 && tx_size != TX_4X4) {
+    vp9_write(w, tx_size != TX_8X8, tx_probs[1]);
+    if (bsize >= BLOCK_SIZE_SB32X32 && tx_size != TX_8X8)
+      vp9_write(w, tx_size != TX_16X16, tx_probs[2]);
+  }
+}
 
-  for (k = 0; k < MBSKIP_CONTEXTS; ++k) {
-    vp9_cond_prob_diff_update(bc, &pc->fc.mbskip_probs[k],
-                              VP9_MODE_UPDATE_PROB, pc->fc.mbskip_count[k]);
+static int write_skip_coeff(const VP9_COMP *cpi, int segment_id, MODE_INFO *m,
+                            vp9_writer *w) {
+  const MACROBLOCKD *const xd = &cpi->mb.e_mbd;
+  if (vp9_segfeature_active(&xd->seg, segment_id, SEG_LVL_SKIP)) {
+    return 1;
+  } else {
+    const int skip_coeff = m->mbmi.mb_skip_coeff;
+    vp9_write(w, skip_coeff, vp9_get_pred_prob_mbskip(&cpi->common, xd));
+    return skip_coeff;
   }
 }
 
+void vp9_update_skip_probs(VP9_COMP *cpi, vp9_writer *w) {
+  VP9_COMMON *cm = &cpi->common;
+  int k;
+
+  for (k = 0; k < MBSKIP_CONTEXTS; ++k)
+    vp9_cond_prob_diff_update(w, &cm->fc.mbskip_probs[k],
+                              VP9_MODE_UPDATE_PROB, cm->counts.mbskip[k]);
+}
+
 static void write_intra_mode(vp9_writer *bc, int m, const vp9_prob *p) {
   write_token(bc, vp9_intra_mode_tree, p, vp9_intra_mode_encodings + m);
 }
@@ -465,7 +248,7 @@ static void update_switchable_interp_probs(VP9_COMP *const cpi,
     vp9_tree_probs_from_distribution(
         vp9_switchable_interp_tree,
         new_prob[j], branch_ct[j],
-        pc->fc.switchable_interp_count[j], 0);
+        pc->counts.switchable_interp[j], 0);
   }
   for (j = 0; j <= VP9_SWITCHABLE_FILTERS; ++j) {
     for (i = 0; i < VP9_SWITCHABLE_FILTERS - 1; ++i) {
@@ -486,7 +269,7 @@ static void update_inter_mode_probs(VP9_COMMON *pc, vp9_writer* const bc) {
     for (j = 0; j < VP9_INTER_MODES - 1; j++) {
       vp9_cond_prob_diff_update(bc, &pc->fc.inter_mode_probs[i][j],
                                 VP9_MODE_UPDATE_PROB,
-                                pc->fc.inter_mode_counts[i][j]);
+                                pc->counts.inter_mode[i][j]);
     }
   }
 }
@@ -519,22 +302,13 @@ static void pack_mb_tokens(vp9_writer* const bc,
     assert(pp != 0);
 
     /* skip one or two nodes */
-#if !CONFIG_BALANCED_COEFTREE
     if (p->skip_eob_node) {
       n -= p->skip_eob_node;
       i = 2 * p->skip_eob_node;
     }
-#endif
 
     do {
       const int bb = (v >> --n) & 1;
-#if CONFIG_BALANCED_COEFTREE
-      if (i == 2 && p->skip_eob_node) {
-        i += 2;
-        assert(bb == 1);
-        continue;
-      }
-#endif
       vp9_write(bc, bb, pp[i >> 1]);
       i = vp9_coef_tree[i + bb];
     } while (n);
@@ -563,22 +337,18 @@ static void pack_mb_tokens(vp9_writer* const bc,
   *tp = p;
 }
 
-static void write_sb_mv_ref(vp9_writer *bc, MB_PREDICTION_MODE m,
+static void write_sb_mv_ref(vp9_writer *w, MB_PREDICTION_MODE mode,
                             const vp9_prob *p) {
-#if CONFIG_DEBUG
-  assert(NEARESTMV <= m && m <= NEWMV);
-#endif
-  write_token(bc, vp9_sb_mv_ref_tree, p,
-              vp9_sb_mv_ref_encoding_array - NEARESTMV + m);
+  assert(is_inter_mode(mode));
+  write_token(w, vp9_inter_mode_tree, p,
+              &vp9_inter_mode_encodings[mode - NEARESTMV]);
 }
 
-// This function writes the current macro block's segnment id to the bitstream
-// It should only be called if a segment map update is indicated.
-static void write_mb_segid(vp9_writer *bc,
-                           const MB_MODE_INFO *mi, const MACROBLOCKD *xd) {
-  if (xd->segmentation_enabled && xd->update_mb_segmentation_map)
-    treed_write(bc, vp9_segment_tree, xd->mb_segment_tree_probs,
-                mi->segment_id, 3);
+
+static void write_segment_id(vp9_writer *w, const struct segmentation *seg,
+                             int segment_id) {
+  if (seg->enabled && seg->update_map)
+    treed_write(w, vp9_segment_tree, seg->tree_probs, segment_id, 3);
 }
 
 // This function encodes the reference frame
@@ -588,7 +358,7 @@ static void encode_ref_frame(VP9_COMP *cpi, vp9_writer *bc) {
   MACROBLOCKD *const xd = &x->e_mbd;
   MB_MODE_INFO *mi = &xd->mode_info_context->mbmi;
   const int segment_id = mi->segment_id;
-  int seg_ref_active = vp9_segfeature_active(xd, segment_id,
+  int seg_ref_active = vp9_segfeature_active(&xd->seg, segment_id,
                                              SEG_LVL_REF_FRAME);
   // If segment level coding of this signal is disabled...
   // or the segment allows multiple reference frame options
@@ -597,7 +367,7 @@ static void encode_ref_frame(VP9_COMP *cpi, vp9_writer *bc) {
     // (if not specified at the frame/segment level)
     if (pc->comp_pred_mode == HYBRID_PREDICTION) {
       vp9_write(bc, mi->ref_frame[1] > INTRA_FRAME,
-                vp9_get_pred_prob(pc, xd, PRED_COMP_INTER_INTER));
+                vp9_get_pred_prob_comp_inter_inter(pc, xd));
     } else {
       assert((mi->ref_frame[1] <= INTRA_FRAME) ==
                  (pc->comp_pred_mode == SINGLE_PREDICTION_ONLY));
@@ -605,17 +375,17 @@ static void encode_ref_frame(VP9_COMP *cpi, vp9_writer *bc) {
 
     if (mi->ref_frame[1] > INTRA_FRAME) {
       vp9_write(bc, mi->ref_frame[0] == GOLDEN_FRAME,
-                vp9_get_pred_prob(pc, xd, PRED_COMP_REF_P));
+                vp9_get_pred_prob_comp_ref_p(pc, xd));
     } else {
       vp9_write(bc, mi->ref_frame[0] != LAST_FRAME,
-                vp9_get_pred_prob(pc, xd, PRED_SINGLE_REF_P1));
+                vp9_get_pred_prob_single_ref_p1(pc, xd));
       if (mi->ref_frame[0] != LAST_FRAME)
         vp9_write(bc, mi->ref_frame[0] != GOLDEN_FRAME,
-                  vp9_get_pred_prob(pc, xd, PRED_SINGLE_REF_P2));
+                  vp9_get_pred_prob_single_ref_p2(pc, xd));
     }
   } else {
     assert(mi->ref_frame[1] <= INTRA_FRAME);
-    assert(vp9_get_segdata(xd, segment_id, SEG_LVL_REF_FRAME) ==
+    assert(vp9_get_segdata(&xd->seg, segment_id, SEG_LVL_REF_FRAME) ==
            mi->ref_frame[0]);
   }
 
@@ -629,60 +399,42 @@ static void pack_inter_mode_mvs(VP9_COMP *cpi, MODE_INFO *m,
   const nmv_context *nmvc = &pc->fc.nmvc;
   MACROBLOCK *const x = &cpi->mb;
   MACROBLOCKD *const xd = &x->e_mbd;
+  struct segmentation *seg = &xd->seg;
   MB_MODE_INFO *const mi = &m->mbmi;
   const MV_REFERENCE_FRAME rf = mi->ref_frame[0];
   const MB_PREDICTION_MODE mode = mi->mode;
   const int segment_id = mi->segment_id;
   int skip_coeff;
+  const BLOCK_SIZE_TYPE bsize = mi->sb_type;
 
-  xd->prev_mode_info_context = pc->prev_mi + (m - pc->mi);
   x->partition_info = x->pi + (m - pc->mi);
 
 #ifdef ENTROPY_STATS
   active_section = 9;
 #endif
 
-  if (cpi->mb.e_mbd.update_mb_segmentation_map) {
-    // Is temporal coding of the segment map enabled
-    if (pc->temporal_update) {
-      unsigned char prediction_flag = vp9_get_pred_flag(xd, PRED_SEG_ID);
-      vp9_prob pred_prob = vp9_get_pred_prob(pc, xd, PRED_SEG_ID);
-
-      // Code the segment id prediction flag for this mb
-      vp9_write(bc, prediction_flag, pred_prob);
-
-      // If the mb segment id wasn't predicted code explicitly
-      if (!prediction_flag)
-        write_mb_segid(bc, mi, &cpi->mb.e_mbd);
+  if (seg->update_map) {
+    if (seg->temporal_update) {
+      const int pred_flag = mi->seg_id_predicted;
+      vp9_prob pred_prob = vp9_get_pred_prob_seg_id(xd);
+      vp9_write(bc, pred_flag, pred_prob);
+      if (!pred_flag)
+        write_segment_id(bc, seg, segment_id);
     } else {
-      // Normal unpredicted coding
-      write_mb_segid(bc, mi, &cpi->mb.e_mbd);
+      write_segment_id(bc, seg, segment_id);
     }
   }
 
-  if (vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP)) {
-    skip_coeff = 1;
-  } else {
-    skip_coeff = m->mbmi.mb_skip_coeff;
-    vp9_write(bc, skip_coeff,
-              vp9_get_pred_prob(pc, xd, PRED_MBSKIP));
-  }
+  skip_coeff = write_skip_coeff(cpi, segment_id, m, bc);
 
-  if (!vp9_segfeature_active(xd, segment_id, SEG_LVL_REF_FRAME))
+  if (!vp9_segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME))
     vp9_write(bc, rf != INTRA_FRAME,
-              vp9_get_pred_prob(pc, xd, PRED_INTRA_INTER));
+              vp9_get_pred_prob_intra_inter(pc, xd));
 
-  if (mi->sb_type >= BLOCK_SIZE_SB8X8 && pc->txfm_mode == TX_MODE_SELECT &&
+  if (bsize >= BLOCK_SIZE_SB8X8 && pc->tx_mode == TX_MODE_SELECT &&
       !(rf != INTRA_FRAME &&
-        (skip_coeff || vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP)))) {
-    TX_SIZE sz = mi->txfm_size;
-    const vp9_prob *tx_probs = vp9_get_pred_probs(pc, xd, PRED_TX_SIZE);
-    vp9_write(bc, sz != TX_4X4, tx_probs[0]);
-    if (mi->sb_type >= BLOCK_SIZE_MB16X16 && sz != TX_4X4) {
-      vp9_write(bc, sz != TX_8X8, tx_probs[1]);
-      if (mi->sb_type >= BLOCK_SIZE_SB32X32 && sz != TX_8X8)
-        vp9_write(bc, sz != TX_16X16, tx_probs[2]);
-    }
+        (skip_coeff || vp9_segfeature_active(seg, segment_id, SEG_LVL_SKIP)))) {
+    write_selected_txfm_size(cpi, mi->txfm_size, bsize, bc);
   }
 
   if (rf == INTRA_FRAME) {
@@ -690,28 +442,24 @@ static void pack_inter_mode_mvs(VP9_COMP *cpi, MODE_INFO *m,
     active_section = 6;
 #endif
 
-    if (m->mbmi.sb_type >= BLOCK_SIZE_SB8X8) {
-      const BLOCK_SIZE_TYPE bsize = xd->mode_info_context->mbmi.sb_type;
+    if (bsize >= BLOCK_SIZE_SB8X8) {
       const int bwl = b_width_log2(bsize), bhl = b_height_log2(bsize);
       const int bsl = MIN(bwl, bhl);
       write_intra_mode(bc, mode, pc->fc.y_mode_prob[MIN(3, bsl)]);
     } else {
       int idx, idy;
-      int bw = 1 << b_width_log2(mi->sb_type);
-      int bh = 1 << b_height_log2(mi->sb_type);
-      for (idy = 0; idy < 2; idy += bh)
-        for (idx = 0; idx < 2; idx += bw) {
-          MB_PREDICTION_MODE bm = m->bmi[idy * 2 + idx].as_mode.first;
+      int num_4x4_blocks_wide = num_4x4_blocks_wide_lookup[bsize];
+      int num_4x4_blocks_high = num_4x4_blocks_high_lookup[bsize];
+      for (idy = 0; idy < 2; idy += num_4x4_blocks_high)
+        for (idx = 0; idx < 2; idx += num_4x4_blocks_wide) {
+          const MB_PREDICTION_MODE bm = m->bmi[idy * 2 + idx].as_mode;
           write_intra_mode(bc, bm, pc->fc.y_mode_prob[0]);
         }
     }
-    write_intra_mode(bc, mi->uv_mode,
-                     pc->fc.uv_mode_prob[mode]);
+    write_intra_mode(bc, mi->uv_mode, pc->fc.uv_mode_prob[mode]);
   } else {
     vp9_prob *mv_ref_p;
-
     encode_ref_frame(cpi, bc);
-
     mv_ref_p = cpi->common.fc.inter_mode_probs[mi->mb_mode_context[rf]];
 
 #ifdef ENTROPY_STATS
@@ -719,8 +467,8 @@ static void pack_inter_mode_mvs(VP9_COMP *cpi, MODE_INFO *m,
 #endif
 
     // If segment skip is not enabled code the mode.
-    if (!vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP)) {
-      if (mi->sb_type >= BLOCK_SIZE_SB8X8) {
+    if (!vp9_segfeature_active(seg, segment_id, SEG_LVL_SKIP)) {
+      if (bsize >= BLOCK_SIZE_SB8X8) {
         write_sb_mv_ref(bc, mode, mv_ref_p);
         vp9_accum_mv_refs(&cpi->common, mode, mi->mb_mode_context[rf]);
       }
@@ -728,38 +476,37 @@ static void pack_inter_mode_mvs(VP9_COMP *cpi, MODE_INFO *m,
 
     if (cpi->common.mcomp_filter_type == SWITCHABLE) {
       write_token(bc, vp9_switchable_interp_tree,
-                  vp9_get_pred_probs(&cpi->common, xd,
-                                     PRED_SWITCHABLE_INTERP),
+                  vp9_get_pred_probs_switchable_interp(&cpi->common, xd),
                   vp9_switchable_interp_encodings +
                   vp9_switchable_interp_map[mi->interp_filter]);
     } else {
       assert(mi->interp_filter == cpi->common.mcomp_filter_type);
     }
 
-    if (xd->mode_info_context->mbmi.sb_type < BLOCK_SIZE_SB8X8) {
+    if (bsize < BLOCK_SIZE_SB8X8) {
       int j;
       MB_PREDICTION_MODE blockmode;
       int_mv blockmv;
-      int bwl = b_width_log2(mi->sb_type), bw = 1 << bwl;
-      int bhl = b_height_log2(mi->sb_type), bh = 1 << bhl;
+      int num_4x4_blocks_wide = num_4x4_blocks_wide_lookup[bsize];
+      int num_4x4_blocks_high = num_4x4_blocks_high_lookup[bsize];
       int idx, idy;
-      for (idy = 0; idy < 2; idy += bh) {
-        for (idx = 0; idx < 2; idx += bw) {
+      for (idy = 0; idy < 2; idy += num_4x4_blocks_high) {
+        for (idx = 0; idx < 2; idx += num_4x4_blocks_wide) {
           j = idy * 2 + idx;
-          blockmode = cpi->mb.partition_info->bmi[j].mode;
-          blockmv = cpi->mb.partition_info->bmi[j].mv;
+          blockmode = x->partition_info->bmi[j].mode;
+          blockmv = m->bmi[j].as_mv[0];
           write_sb_mv_ref(bc, blockmode, mv_ref_p);
           vp9_accum_mv_refs(&cpi->common, blockmode, mi->mb_mode_context[rf]);
           if (blockmode == NEWMV) {
 #ifdef ENTROPY_STATS
             active_section = 11;
 #endif
-            vp9_encode_mv(bc, &blockmv.as_mv, &mi->best_mv.as_mv,
+            vp9_encode_mv(cpi, bc, &blockmv.as_mv, &mi->best_mv.as_mv,
                           nmvc, xd->allow_high_precision_mv);
 
             if (mi->ref_frame[1] > INTRA_FRAME)
-              vp9_encode_mv(bc,
-                            &cpi->mb.partition_info->bmi[j].second_mv.as_mv,
+              vp9_encode_mv(cpi, bc,
+                            &m->bmi[j].as_mv[1].as_mv,
                             &mi->best_second_mv.as_mv,
                             nmvc, xd->allow_high_precision_mv);
           }
@@ -769,12 +516,12 @@ static void pack_inter_mode_mvs(VP9_COMP *cpi, MODE_INFO *m,
 #ifdef ENTROPY_STATS
       active_section = 5;
 #endif
-      vp9_encode_mv(bc,
+      vp9_encode_mv(cpi, bc,
                     &mi->mv[0].as_mv, &mi->best_mv.as_mv,
                     nmvc, xd->allow_high_precision_mv);
 
       if (mi->ref_frame[1] > INTRA_FRAME)
-        vp9_encode_mv(bc,
+        vp9_encode_mv(cpi, bc,
                       &mi->mv[1].as_mv, &mi->best_second_mv.as_mv,
                       nmvc, xd->allow_high_precision_mv);
     }
@@ -789,54 +536,40 @@ static void write_mb_modes_kf(const VP9_COMP *cpi,
   const int ym = m->mbmi.mode;
   const int mis = c->mode_info_stride;
   const int segment_id = m->mbmi.segment_id;
-  int skip_coeff;
 
-  if (xd->update_mb_segmentation_map)
-    write_mb_segid(bc, &m->mbmi, xd);
+  if (xd->seg.update_map)
+    write_segment_id(bc, &xd->seg, m->mbmi.segment_id);
 
-  if (vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP)) {
-    skip_coeff = 1;
-  } else {
-    skip_coeff = m->mbmi.mb_skip_coeff;
-    vp9_write(bc, skip_coeff, vp9_get_pred_prob(c, xd, PRED_MBSKIP));
-  }
+  write_skip_coeff(cpi, segment_id, m, bc);
 
-  if (m->mbmi.sb_type >= BLOCK_SIZE_SB8X8 && c->txfm_mode == TX_MODE_SELECT) {
-    TX_SIZE sz = m->mbmi.txfm_size;
-    const vp9_prob *tx_probs = vp9_get_pred_probs(c, xd, PRED_TX_SIZE);
-    vp9_write(bc, sz != TX_4X4, tx_probs[0]);
-    if (m->mbmi.sb_type >= BLOCK_SIZE_MB16X16 && sz != TX_4X4) {
-      vp9_write(bc, sz != TX_8X8, tx_probs[1]);
-      if (m->mbmi.sb_type >= BLOCK_SIZE_SB32X32 && sz != TX_8X8)
-        vp9_write(bc, sz != TX_16X16, tx_probs[2]);
-    }
-  }
+  if (m->mbmi.sb_type >= BLOCK_SIZE_SB8X8 && c->tx_mode == TX_MODE_SELECT)
+    write_selected_txfm_size(cpi, m->mbmi.txfm_size, m->mbmi.sb_type, bc);
 
   if (m->mbmi.sb_type >= BLOCK_SIZE_SB8X8) {
     const MB_PREDICTION_MODE A = above_block_mode(m, 0, mis);
     const MB_PREDICTION_MODE L = xd->left_available ?
                                  left_block_mode(m, 0) : DC_PRED;
-    write_intra_mode(bc, ym, c->kf_y_mode_prob[A][L]);
+    write_intra_mode(bc, ym, vp9_kf_y_mode_prob[A][L]);
   } else {
     int idx, idy;
-    int bw = 1 << b_width_log2(m->mbmi.sb_type);
-    int bh = 1 << b_height_log2(m->mbmi.sb_type);
-    for (idy = 0; idy < 2; idy += bh) {
-      for (idx = 0; idx < 2; idx += bw) {
+    int num_4x4_blocks_wide = num_4x4_blocks_wide_lookup[m->mbmi.sb_type];
+    int num_4x4_blocks_high = num_4x4_blocks_high_lookup[m->mbmi.sb_type];
+    for (idy = 0; idy < 2; idy += num_4x4_blocks_high) {
+      for (idx = 0; idx < 2; idx += num_4x4_blocks_wide) {
         int i = idy * 2 + idx;
         const MB_PREDICTION_MODE A = above_block_mode(m, i, mis);
         const MB_PREDICTION_MODE L = (xd->left_available || idx) ?
                                      left_block_mode(m, i) : DC_PRED;
-        const int bm = m->bmi[i].as_mode.first;
+        const int bm = m->bmi[i].as_mode;
 #ifdef ENTROPY_STATS
         ++intra_mode_stats[A][L][bm];
 #endif
-        write_intra_mode(bc, bm, c->kf_y_mode_prob[A][L]);
+        write_intra_mode(bc, bm, vp9_kf_y_mode_prob[A][L]);
       }
     }
   }
 
-  write_intra_mode(bc, m->mbmi.uv_mode, c->kf_uv_mode_prob[ym]);
+  write_intra_mode(bc, m->mbmi.uv_mode, vp9_kf_uv_mode_prob[ym]);
 }
 
 static void write_modes_b(VP9_COMP *cpi, MODE_INFO *m, vp9_writer *bc,
@@ -875,30 +608,16 @@ static void write_modes_sb(VP9_COMP *cpi, MODE_INFO *m, vp9_writer *bc,
   VP9_COMMON *const cm = &cpi->common;
   MACROBLOCKD *xd = &cpi->mb.e_mbd;
   const int mis = cm->mode_info_stride;
-  int bwl, bhl;
   int bsl = b_width_log2(bsize);
   int bs = (1 << bsl) / 4;  // mode_info step for subsize
   int n;
-  PARTITION_TYPE partition;
+  PARTITION_TYPE partition = PARTITION_NONE;
   BLOCK_SIZE_TYPE subsize;
 
   if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
     return;
 
-  bwl = b_width_log2(m->mbmi.sb_type);
-  bhl = b_height_log2(m->mbmi.sb_type);
-
-  // parse the partition type
-  if ((bwl == bsl) && (bhl == bsl))
-    partition = PARTITION_NONE;
-  else if ((bwl == bsl) && (bhl < bsl))
-    partition = PARTITION_HORZ;
-  else if ((bwl < bsl) && (bhl == bsl))
-    partition = PARTITION_VERT;
-  else if ((bwl < bsl) && (bhl < bsl))
-    partition = PARTITION_SPLIT;
-  else
-    assert(0);
+  partition = partition_lookup[bsl][m->mbmi.sb_type];
 
   if (bsize < BLOCK_SIZE_SB8X8)
     if (xd->ab_index > 0)
@@ -906,9 +625,8 @@ static void write_modes_sb(VP9_COMP *cpi, MODE_INFO *m, vp9_writer *bc,
 
   if (bsize >= BLOCK_SIZE_SB8X8) {
     int pl;
-    int idx = check_bsize_coverage(cm, xd, mi_row, mi_col, bsize);
-    xd->left_seg_context = cm->left_seg_context + (mi_row & MI_MASK);
-    xd->above_seg_context = cm->above_seg_context + mi_col;
+    const int idx = check_bsize_coverage(cm, xd, mi_row, mi_col, bsize);
+    set_partition_seg_context(cm, xd, mi_row, mi_col);
     pl = partition_plane_context(xd, bsize);
     // encode the partition information
     if (idx == 0)
@@ -968,14 +686,12 @@ static void write_modes(VP9_COMP *cpi, vp9_writer* const bc,
 
   m_ptr += c->cur_tile_mi_col_start + c->cur_tile_mi_row_start * mis;
 
-  for (mi_row = c->cur_tile_mi_row_start;
-       mi_row < c->cur_tile_mi_row_end;
+  for (mi_row = c->cur_tile_mi_row_start; mi_row < c->cur_tile_mi_row_end;
        mi_row += 8, m_ptr += 8 * mis) {
     m = m_ptr;
-    vpx_memset(c->left_seg_context, 0, sizeof(c->left_seg_context));
-    for (mi_col = c->cur_tile_mi_col_start;
-         mi_col < c->cur_tile_mi_col_end;
-         mi_col += 64 / MI_SIZE, m += 64 / MI_SIZE)
+    vp9_zero(c->left_seg_context);
+    for (mi_col = c->cur_tile_mi_col_start; mi_col < c->cur_tile_mi_col_end;
+         mi_col += MI_BLOCK_SIZE, m += MI_BLOCK_SIZE)
       write_modes_sb(cpi, m, bc, tok, tok_end, mi_row, mi_col,
                      BLOCK_SIZE_SB64X64);
   }
@@ -1014,7 +730,7 @@ static void build_tree_distribution(VP9_COMP *cpi, TX_SIZE txfm_size) {
   vp9_coeff_probs_model *coef_probs = cpi->frame_coef_probs[txfm_size];
   vp9_coeff_count *coef_counts = cpi->coef_counts[txfm_size];
   unsigned int (*eob_branch_ct)[REF_TYPES][COEF_BANDS][PREV_COEF_CONTEXTS] =
-      cpi->common.fc.eob_branch_counts[txfm_size];
+      cpi->common.counts.eob_branch[txfm_size];
   vp9_coeff_stats *coef_branch_ct = cpi->frame_branch_ct[txfm_size];
   vp9_prob full_probs[ENTROPY_NODES];
   int i, j, k, l;
@@ -1031,19 +747,11 @@ static void build_tree_distribution(VP9_COMP *cpi, TX_SIZE txfm_size) {
                                            coef_counts[i][j][k][l], 0);
           vpx_memcpy(coef_probs[i][j][k][l], full_probs,
                      sizeof(vp9_prob) * UNCONSTRAINED_NODES);
-#if CONFIG_BALANCED_COEFTREE
-          coef_branch_ct[i][j][k][l][1][1] = eob_branch_ct[i][j][k][l] -
-                                             coef_branch_ct[i][j][k][l][1][0];
-          coef_probs[i][j][k][l][1] =
-              get_binary_prob(coef_branch_ct[i][j][k][l][1][0],
-                              coef_branch_ct[i][j][k][l][1][1]);
-#else
           coef_branch_ct[i][j][k][l][0][1] = eob_branch_ct[i][j][k][l] -
                                              coef_branch_ct[i][j][k][l][0][0];
           coef_probs[i][j][k][l][0] =
               get_binary_prob(coef_branch_ct[i][j][k][l][0][0],
                               coef_branch_ct[i][j][k][l][0][1]);
-#endif
 #ifdef ENTROPY_STATS
           if (!cpi->dummy_packing) {
             int t;
@@ -1096,11 +804,11 @@ static void update_coef_probs_common(vp9_writer* const bc, VP9_COMP *cpi,
             if (l >= 3 && k == 0)
               continue;
             if (t == PIVOT_NODE)
-              s = prob_diff_update_savings_search_model(
+              s = vp9_prob_diff_update_savings_search_model(
                   frame_branch_ct[i][j][k][l][0],
                   old_frame_coef_probs[i][j][k][l], &newp, upd, i, j);
             else
-              s = prob_diff_update_savings_search(
+              s = vp9_prob_diff_update_savings_search(
                   frame_branch_ct[i][j][k][l][t], oldp, &newp, upd);
             if (s > 0 && newp != oldp)
               u = 1;
@@ -1137,11 +845,11 @@ static void update_coef_probs_common(vp9_writer* const bc, VP9_COMP *cpi,
             if (l >= 3 && k == 0)
               continue;
             if (t == PIVOT_NODE)
-              s = prob_diff_update_savings_search_model(
+              s = vp9_prob_diff_update_savings_search_model(
                   frame_branch_ct[i][j][k][l][0],
                   old_frame_coef_probs[i][j][k][l], &newp, upd, i, j);
             else
-              s = prob_diff_update_savings_search(
+              s = vp9_prob_diff_update_savings_search(
                   frame_branch_ct[i][j][k][l][t],
                   *oldp, &newp, upd);
             if (s > 0 && newp != *oldp)
@@ -1153,7 +861,7 @@ static void update_coef_probs_common(vp9_writer* const bc, VP9_COMP *cpi,
 #endif
             if (u) {
               /* send/use new probability */
-              write_prob_diff_update(bc, newp, *oldp);
+              vp9_write_prob_diff_update(bc, newp, *oldp);
               *oldp = newp;
             }
           }
@@ -1164,7 +872,7 @@ static void update_coef_probs_common(vp9_writer* const bc, VP9_COMP *cpi,
 }
 
 static void update_coef_probs(VP9_COMP* const cpi, vp9_writer* const bc) {
-  const TXFM_MODE txfm_mode = cpi->common.txfm_mode;
+  const TX_MODE tx_mode = cpi->common.tx_mode;
 
   vp9_clear_system_state();
 
@@ -1174,39 +882,39 @@ static void update_coef_probs(VP9_COMP* const cpi, vp9_writer* const bc) {
   update_coef_probs_common(bc, cpi, TX_4X4);
 
   // do not do this if not even allowed
-  if (txfm_mode > ONLY_4X4)
+  if (tx_mode > ONLY_4X4)
     update_coef_probs_common(bc, cpi, TX_8X8);
 
-  if (txfm_mode > ALLOW_8X8)
+  if (tx_mode > ALLOW_8X8)
     update_coef_probs_common(bc, cpi, TX_16X16);
 
-  if (txfm_mode > ALLOW_16X16)
+  if (tx_mode > ALLOW_16X16)
     update_coef_probs_common(bc, cpi, TX_32X32);
 }
 
-static void encode_loopfilter(VP9_COMMON *pc, MACROBLOCKD *xd,
+static void encode_loopfilter(struct loopfilter *lf,
                               struct vp9_write_bit_buffer *wb) {
   int i;
 
   // Encode the loop filter level and type
-  vp9_wb_write_literal(wb, pc->filter_level, 6);
-  vp9_wb_write_literal(wb, pc->sharpness_level, 3);
+  vp9_wb_write_literal(wb, lf->filter_level, 6);
+  vp9_wb_write_literal(wb, lf->sharpness_level, 3);
 
   // Write out loop filter deltas applied at the MB level based on mode or
   // ref frame (if they are enabled).
-  vp9_wb_write_bit(wb, xd->mode_ref_lf_delta_enabled);
+  vp9_wb_write_bit(wb, lf->mode_ref_delta_enabled);
 
-  if (xd->mode_ref_lf_delta_enabled) {
+  if (lf->mode_ref_delta_enabled) {
     // Do the deltas need to be updated
-    vp9_wb_write_bit(wb, xd->mode_ref_lf_delta_update);
-    if (xd->mode_ref_lf_delta_update) {
+    vp9_wb_write_bit(wb, lf->mode_ref_delta_update);
+    if (lf->mode_ref_delta_update) {
       // Send update
       for (i = 0; i < MAX_REF_LF_DELTAS; i++) {
-        const int delta = xd->ref_lf_deltas[i];
+        const int delta = lf->ref_deltas[i];
 
         // Frame level data
-        if (delta != xd->last_ref_lf_deltas[i]) {
-          xd->last_ref_lf_deltas[i] = delta;
+        if (delta != lf->last_ref_deltas[i]) {
+          lf->last_ref_deltas[i] = delta;
           vp9_wb_write_bit(wb, 1);
 
           assert(delta != 0);
@@ -1219,9 +927,9 @@ static void encode_loopfilter(VP9_COMMON *pc, MACROBLOCKD *xd,
 
       // Send update
       for (i = 0; i < MAX_MODE_LF_DELTAS; i++) {
-        const int delta = xd->mode_lf_deltas[i];
-        if (delta != xd->last_mode_lf_deltas[i]) {
-          xd->last_mode_lf_deltas[i] = delta;
+        const int delta = lf->mode_deltas[i];
+        if (delta != lf->last_mode_deltas[i]) {
+          lf->last_mode_deltas[i] = delta;
           vp9_wb_write_bit(wb, 1);
 
           assert(delta != 0);
@@ -1255,23 +963,23 @@ static void encode_quantization(VP9_COMMON *cm,
 
 
 static void encode_segmentation(VP9_COMP *cpi,
-                               struct vp9_write_bit_buffer *wb) {
+                                struct vp9_write_bit_buffer *wb) {
   int i, j;
-  VP9_COMMON *const cm = &cpi->common;
-  MACROBLOCKD *const xd = &cpi->mb.e_mbd;
 
-  vp9_wb_write_bit(wb, xd->segmentation_enabled);
-  if (!xd->segmentation_enabled)
+  struct segmentation *seg = &cpi->mb.e_mbd.seg;
+
+  vp9_wb_write_bit(wb, seg->enabled);
+  if (!seg->enabled)
     return;
 
   // Segmentation map
-  vp9_wb_write_bit(wb, xd->update_mb_segmentation_map);
-  if (xd->update_mb_segmentation_map) {
+  vp9_wb_write_bit(wb, seg->update_map);
+  if (seg->update_map) {
     // Select the coding strategy (temporal or spatial)
     vp9_choose_segmap_coding_method(cpi);
     // Write out probabilities used to decode unpredicted  macro-block segments
-    for (i = 0; i < MB_SEG_TREE_PROBS; i++) {
-      const int prob = xd->mb_segment_tree_probs[i];
+    for (i = 0; i < SEG_TREE_PROBS; i++) {
+      const int prob = seg->tree_probs[i];
       const int update = prob != MAX_PROB;
       vp9_wb_write_bit(wb, update);
       if (update)
@@ -1279,10 +987,10 @@ static void encode_segmentation(VP9_COMP *cpi,
     }
 
     // Write out the chosen coding method.
-    vp9_wb_write_bit(wb, cm->temporal_update);
-    if (cm->temporal_update) {
+    vp9_wb_write_bit(wb, seg->temporal_update);
+    if (seg->temporal_update) {
       for (i = 0; i < PREDICTION_PROBS; i++) {
-        const int prob = cm->segment_pred_probs[i];
+        const int prob = seg->pred_probs[i];
         const int update = prob != MAX_PROB;
         vp9_wb_write_bit(wb, update);
         if (update)
@@ -1292,16 +1000,16 @@ static void encode_segmentation(VP9_COMP *cpi,
   }
 
   // Segmentation data
-  vp9_wb_write_bit(wb, xd->update_mb_segmentation_data);
-  if (xd->update_mb_segmentation_data) {
-    vp9_wb_write_bit(wb, xd->mb_segment_abs_delta);
+  vp9_wb_write_bit(wb, seg->update_data);
+  if (seg->update_data) {
+    vp9_wb_write_bit(wb, seg->abs_delta);
 
-    for (i = 0; i < MAX_MB_SEGMENTS; i++) {
+    for (i = 0; i < MAX_SEGMENTS; i++) {
       for (j = 0; j < SEG_LVL_MAX; j++) {
-        const int active = vp9_segfeature_active(xd, i, j);
+        const int active = vp9_segfeature_active(seg, i, j);
         vp9_wb_write_bit(wb, active);
         if (active) {
-          const int data = vp9_get_segdata(xd, i, j);
+          const int data = vp9_get_segdata(seg, i, j);
           const int data_max = vp9_seg_feature_data_max(j);
 
           if (vp9_is_segfeature_signed(j)) {
@@ -1321,12 +1029,12 @@ static void encode_txfm_probs(VP9_COMP *cpi, vp9_writer *w) {
   VP9_COMMON *const cm = &cpi->common;
 
   // Mode
-  vp9_write_literal(w, MIN(cm->txfm_mode, ALLOW_32X32), 2);
-  if (cm->txfm_mode >= ALLOW_32X32)
-    vp9_write_bit(w, cm->txfm_mode == TX_MODE_SELECT);
+  vp9_write_literal(w, MIN(cm->tx_mode, ALLOW_32X32), 2);
+  if (cm->tx_mode >= ALLOW_32X32)
+    vp9_write_bit(w, cm->tx_mode == TX_MODE_SELECT);
 
   // Probabilities
-  if (cm->txfm_mode == TX_MODE_SELECT) {
+  if (cm->tx_mode == TX_MODE_SELECT) {
     int i, j;
     unsigned int ct_8x8p[TX_SIZE_MAX_SB - 3][2];
     unsigned int ct_16x16p[TX_SIZE_MAX_SB - 2][2];
@@ -1334,28 +1042,26 @@ static void encode_txfm_probs(VP9_COMP *cpi, vp9_writer *w) {
 
 
     for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
-      tx_counts_to_branch_counts_8x8(cm->fc.tx_count_8x8p[i],
+      tx_counts_to_branch_counts_8x8(cm->counts.tx.p8x8[i],
                                      ct_8x8p);
-      for (j = 0; j < TX_SIZE_MAX_SB - 3; j++) {
-        vp9_cond_prob_diff_update(w, &cm->fc.tx_probs_8x8p[i][j],
+      for (j = 0; j < TX_SIZE_MAX_SB - 3; j++)
+        vp9_cond_prob_diff_update(w, &cm->fc.tx_probs.p8x8[i][j],
                                   VP9_MODE_UPDATE_PROB, ct_8x8p[j]);
-      }
     }
+
     for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
-      tx_counts_to_branch_counts_16x16(cm->fc.tx_count_16x16p[i],
+      tx_counts_to_branch_counts_16x16(cm->counts.tx.p16x16[i],
                                        ct_16x16p);
-      for (j = 0; j < TX_SIZE_MAX_SB - 2; j++) {
-        vp9_cond_prob_diff_update(w, &cm->fc.tx_probs_16x16p[i][j],
+      for (j = 0; j < TX_SIZE_MAX_SB - 2; j++)
+        vp9_cond_prob_diff_update(w, &cm->fc.tx_probs.p16x16[i][j],
                                   VP9_MODE_UPDATE_PROB, ct_16x16p[j]);
-      }
     }
+
     for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
-      tx_counts_to_branch_counts_32x32(cm->fc.tx_count_32x32p[i],
-                                       ct_32x32p);
-      for (j = 0; j < TX_SIZE_MAX_SB - 1; j++) {
-        vp9_cond_prob_diff_update(w, &cm->fc.tx_probs_32x32p[i][j],
+      tx_counts_to_branch_counts_32x32(cm->counts.tx.p32x32[i], ct_32x32p);
+      for (j = 0; j < TX_SIZE_MAX_SB - 1; j++)
+        vp9_cond_prob_diff_update(w, &cm->fc.tx_probs.p32x32[i][j],
                                   VP9_MODE_UPDATE_PROB, ct_32x32p[j]);
-      }
     }
 #ifdef MODE_STATS
     if (!cpi->dummy_packing)
@@ -1381,7 +1087,7 @@ static void fix_mcomp_filter_type(VP9_COMP *cpi) {
     for (i = 0; i < VP9_SWITCHABLE_FILTERS; ++i) {
       count[i] = 0;
       for (j = 0; j <= VP9_SWITCHABLE_FILTERS; ++j)
-        count[i] += cm->fc.switchable_interp_count[j][i];
+        count[i] += cm->counts.switchable_interp[j][i];
       c += (count[i] > 0);
     }
     if (c == 1) {
@@ -1397,18 +1103,18 @@ static void fix_mcomp_filter_type(VP9_COMP *cpi) {
 }
 
 static void write_tile_info(VP9_COMMON *cm, struct vp9_write_bit_buffer *wb) {
-  int min_log2_tiles, delta_log2_tiles, n_tile_bits, n;
-  vp9_get_tile_n_bits(cm, &min_log2_tiles, &delta_log2_tiles);
-  n_tile_bits = cm->log2_tile_columns - min_log2_tiles;
-  for (n = 0; n < delta_log2_tiles; n++) {
-    if (n_tile_bits--) {
-      vp9_wb_write_bit(wb, 1);
-    } else {
-      vp9_wb_write_bit(wb, 0);
-      break;
-    }
-  }
+  int min_log2_tile_cols, max_log2_tile_cols, ones;
+  vp9_get_tile_n_bits(cm->mi_cols, &min_log2_tile_cols, &max_log2_tile_cols);
 
+  // columns
+  ones = cm->log2_tile_cols - min_log2_tile_cols;
+  while (ones--)
+    vp9_wb_write_bit(wb, 1);
+
+  if (cm->log2_tile_cols < max_log2_tile_cols)
+    vp9_wb_write_bit(wb, 0);
+
+  // rows
   vp9_wb_write_bit(wb, cm->log2_tile_rows != 0);
   if (cm->log2_tile_rows != 0)
     vp9_wb_write_bit(wb, cm->log2_tile_rows != 1);
@@ -1449,6 +1155,57 @@ static int get_refresh_mask(VP9_COMP *cpi) {
     }
 }
 
+static size_t encode_tiles(VP9_COMP *cpi, uint8_t *data_ptr) {
+  VP9_COMMON *const cm = &cpi->common;
+  vp9_writer residual_bc;
+
+  int tile_row, tile_col;
+  TOKENEXTRA *tok[4][1 << 6], *tok_end;
+  size_t total_size = 0;
+  const int tile_cols = 1 << cm->log2_tile_cols;
+  const int tile_rows = 1 << cm->log2_tile_rows;
+
+  vpx_memset(cm->above_seg_context, 0, sizeof(PARTITION_CONTEXT) *
+             mi_cols_aligned_to_sb(cm->mi_cols));
+
+  tok[0][0] = cpi->tok;
+  for (tile_row = 0; tile_row < tile_rows; tile_row++) {
+    if (tile_row)
+      tok[tile_row][0] = tok[tile_row - 1][tile_cols - 1] +
+                         cpi->tok_count[tile_row - 1][tile_cols - 1];
+
+    for (tile_col = 1; tile_col < tile_cols; tile_col++)
+      tok[tile_row][tile_col] = tok[tile_row][tile_col - 1] +
+                                cpi->tok_count[tile_row][tile_col - 1];
+  }
+
+  for (tile_row = 0; tile_row < tile_rows; tile_row++) {
+    vp9_get_tile_row_offsets(cm, tile_row);
+    for (tile_col = 0; tile_col < tile_cols; tile_col++) {
+      vp9_get_tile_col_offsets(cm, tile_col);
+      tok_end = tok[tile_row][tile_col] + cpi->tok_count[tile_row][tile_col];
+
+      if (tile_col < tile_cols - 1 || tile_row < tile_rows - 1)
+        vp9_start_encode(&residual_bc, data_ptr + total_size + 4);
+      else
+        vp9_start_encode(&residual_bc, data_ptr + total_size);
+
+      write_modes(cpi, &residual_bc, &tok[tile_row][tile_col], tok_end);
+      assert(tok[tile_row][tile_col] == tok_end);
+      vp9_stop_encode(&residual_bc);
+      if (tile_col < tile_cols - 1 || tile_row < tile_rows - 1) {
+        // size of this tile
+        write_be32(data_ptr + total_size, residual_bc.pos);
+        total_size += 4;
+      }
+
+      total_size += residual_bc.pos;
+    }
+  }
+
+  return total_size;
+}
+
 static void write_display_size(VP9_COMP *cpi, struct vp9_write_bit_buffer *wb) {
   VP9_COMMON *const cm = &cpi->common;
 
@@ -1562,7 +1319,7 @@ static void write_uncompressed_header(VP9_COMP *cpi,
       int i;
       vp9_wb_write_literal(wb, get_refresh_mask(cpi), NUM_REF_FRAMES);
       for (i = 0; i < ALLOWED_REFS_PER_FRAME; ++i) {
-        vp9_wb_write_literal(wb, refs[i], NUM_REF_FRAMES_LG2);
+        vp9_wb_write_literal(wb, refs[i], NUM_REF_FRAMES_LOG2);
         vp9_wb_write_bit(wb, cm->ref_frame_sign_bias[LAST_FRAME + i]);
       }
 
@@ -1580,66 +1337,27 @@ static void write_uncompressed_header(VP9_COMP *cpi,
     vp9_wb_write_bit(wb, cm->frame_parallel_decoding_mode);
   }
 
-  vp9_wb_write_literal(wb, cm->frame_context_idx, NUM_FRAME_CONTEXTS_LG2);
+  vp9_wb_write_literal(wb, cm->frame_context_idx, NUM_FRAME_CONTEXTS_LOG2);
 
-  encode_loopfilter(cm, xd, wb);
+  encode_loopfilter(&xd->lf, wb);
   encode_quantization(cm, wb);
   encode_segmentation(cpi, wb);
 
   write_tile_info(cm, wb);
 }
 
-void vp9_pack_bitstream(VP9_COMP *cpi, uint8_t *dest, unsigned long *size) {
-  int i, bytes_packed;
-  VP9_COMMON *const pc = &cpi->common;
-  vp9_writer header_bc, residual_bc;
+static size_t write_compressed_header(VP9_COMP *cpi, uint8_t *data) {
+  VP9_COMMON *const cm = &cpi->common;
   MACROBLOCKD *const xd = &cpi->mb.e_mbd;
+  FRAME_CONTEXT *const fc = &cm->fc;
+  vp9_writer header_bc;
 
-  uint8_t *cx_data = dest;
-  struct vp9_write_bit_buffer wb = {dest, 0};
-  struct vp9_write_bit_buffer first_partition_size_wb;
-
-  write_uncompressed_header(cpi, &wb);
-  first_partition_size_wb = wb;
-  vp9_wb_write_literal(&wb, 0, 16);  // don't know in advance first part. size
-
-  bytes_packed = vp9_rb_bytes_written(&wb);
-  cx_data += bytes_packed;
+  vp9_start_encode(&header_bc, data);
 
-  compute_update_table();
-
-  vp9_start_encode(&header_bc, cx_data);
-
-#ifdef ENTROPY_STATS
-  if (pc->frame_type == INTER_FRAME)
-    active_section = 0;
+  if (xd->lossless)
+    cm->tx_mode = ONLY_4X4;
   else
-    active_section = 7;
-#endif
-
-  vp9_clear_system_state();  // __asm emms;
-
-  vp9_copy(pc->fc.pre_coef_probs, pc->fc.coef_probs);
-  vp9_copy(pc->fc.pre_y_mode_prob, pc->fc.y_mode_prob);
-  vp9_copy(pc->fc.pre_uv_mode_prob, pc->fc.uv_mode_prob);
-  vp9_copy(pc->fc.pre_partition_prob, pc->fc.partition_prob[INTER_FRAME]);
-  pc->fc.pre_nmvc = pc->fc.nmvc;
-  vp9_copy(pc->fc.pre_switchable_interp_prob, pc->fc.switchable_interp_prob);
-  vp9_copy(pc->fc.pre_inter_mode_probs, pc->fc.inter_mode_probs);
-  vp9_copy(pc->fc.pre_intra_inter_prob, pc->fc.intra_inter_prob);
-  vp9_copy(pc->fc.pre_comp_inter_prob, pc->fc.comp_inter_prob);
-  vp9_copy(pc->fc.pre_comp_ref_prob, pc->fc.comp_ref_prob);
-  vp9_copy(pc->fc.pre_single_ref_prob, pc->fc.single_ref_prob);
-  vp9_copy(pc->fc.pre_tx_probs_8x8p, pc->fc.tx_probs_8x8p);
-  vp9_copy(pc->fc.pre_tx_probs_16x16p, pc->fc.tx_probs_16x16p);
-  vp9_copy(pc->fc.pre_tx_probs_32x32p, pc->fc.tx_probs_32x32p);
-  vp9_copy(pc->fc.pre_mbskip_probs, pc->fc.mbskip_probs);
-
-  if (xd->lossless) {
-    pc->txfm_mode = ONLY_4X4;
-  } else {
     encode_txfm_probs(cpi, &header_bc);
-  }
 
   update_coef_probs(cpi, &header_bc);
 
@@ -1649,124 +1367,106 @@ void vp9_pack_bitstream(VP9_COMP *cpi, uint8_t *dest, unsigned long *size) {
 
   vp9_update_skip_probs(cpi, &header_bc);
 
-  if (pc->frame_type != KEY_FRAME) {
+  if (cm->frame_type != KEY_FRAME) {
+    int i;
 #ifdef ENTROPY_STATS
     active_section = 1;
 #endif
 
-    update_inter_mode_probs(pc, &header_bc);
-    vp9_zero(cpi->common.fc.inter_mode_counts);
+    update_inter_mode_probs(cm, &header_bc);
+    vp9_zero(cm->counts.inter_mode);
 
-    if (pc->mcomp_filter_type == SWITCHABLE)
+    if (cm->mcomp_filter_type == SWITCHABLE)
       update_switchable_interp_probs(cpi, &header_bc);
 
     for (i = 0; i < INTRA_INTER_CONTEXTS; i++)
-      vp9_cond_prob_diff_update(&header_bc, &pc->fc.intra_inter_prob[i],
+      vp9_cond_prob_diff_update(&header_bc, &fc->intra_inter_prob[i],
                                 VP9_MODE_UPDATE_PROB,
                                 cpi->intra_inter_count[i]);
 
-    if (pc->allow_comp_inter_inter) {
+    if (cm->allow_comp_inter_inter) {
       const int comp_pred_mode = cpi->common.comp_pred_mode;
-      const int use_compound_pred = (comp_pred_mode != SINGLE_PREDICTION_ONLY);
-      const int use_hybrid_pred = (comp_pred_mode == HYBRID_PREDICTION);
+      const int use_compound_pred = comp_pred_mode != SINGLE_PREDICTION_ONLY;
+      const int use_hybrid_pred = comp_pred_mode == HYBRID_PREDICTION;
 
       vp9_write_bit(&header_bc, use_compound_pred);
       if (use_compound_pred) {
         vp9_write_bit(&header_bc, use_hybrid_pred);
-        if (use_hybrid_pred) {
+        if (use_hybrid_pred)
           for (i = 0; i < COMP_INTER_CONTEXTS; i++)
-            vp9_cond_prob_diff_update(&header_bc, &pc->fc.comp_inter_prob[i],
+            vp9_cond_prob_diff_update(&header_bc, &fc->comp_inter_prob[i],
                                       VP9_MODE_UPDATE_PROB,
                                       cpi->comp_inter_count[i]);
-        }
       }
     }
 
-    if (pc->comp_pred_mode != COMP_PREDICTION_ONLY) {
+    if (cm->comp_pred_mode != COMP_PREDICTION_ONLY) {
       for (i = 0; i < REF_CONTEXTS; i++) {
-        vp9_cond_prob_diff_update(&header_bc, &pc->fc.single_ref_prob[i][0],
+        vp9_cond_prob_diff_update(&header_bc, &fc->single_ref_prob[i][0],
                                   VP9_MODE_UPDATE_PROB,
                                   cpi->single_ref_count[i][0]);
-        vp9_cond_prob_diff_update(&header_bc, &pc->fc.single_ref_prob[i][1],
+        vp9_cond_prob_diff_update(&header_bc, &fc->single_ref_prob[i][1],
                                   VP9_MODE_UPDATE_PROB,
                                   cpi->single_ref_count[i][1]);
       }
     }
 
-    if (pc->comp_pred_mode != SINGLE_PREDICTION_ONLY) {
+    if (cm->comp_pred_mode != SINGLE_PREDICTION_ONLY)
       for (i = 0; i < REF_CONTEXTS; i++)
-        vp9_cond_prob_diff_update(&header_bc, &pc->fc.comp_ref_prob[i],
+        vp9_cond_prob_diff_update(&header_bc, &fc->comp_ref_prob[i],
                                   VP9_MODE_UPDATE_PROB,
                                   cpi->comp_ref_count[i]);
-    }
 
     update_mbintra_mode_probs(cpi, &header_bc);
 
     for (i = 0; i < NUM_PARTITION_CONTEXTS; ++i) {
-      vp9_prob Pnew[PARTITION_TYPES - 1];
+      vp9_prob pnew[PARTITION_TYPES - 1];
       unsigned int bct[PARTITION_TYPES - 1][2];
       update_mode(&header_bc, PARTITION_TYPES, vp9_partition_encodings,
-                  vp9_partition_tree, Pnew,
-                  pc->fc.partition_prob[pc->frame_type][i], bct,
+                  vp9_partition_tree, pnew,
+                  fc->partition_prob[cm->frame_type][i], bct,
                   (unsigned int *)cpi->partition_count[i]);
     }
 
     vp9_write_nmv_probs(cpi, xd->allow_high_precision_mv, &header_bc);
   }
 
-
   vp9_stop_encode(&header_bc);
+  assert(header_bc.pos <= 0xffff);
 
+  return header_bc.pos;
+}
 
-  // first partition size
-  assert(header_bc.pos <= 0xffff);
-  vp9_wb_write_literal(&first_partition_size_wb, header_bc.pos, 16);
-  *size = bytes_packed + header_bc.pos;
-
-  {
-    int tile_row, tile_col, total_size = 0;
-    unsigned char *data_ptr = cx_data + header_bc.pos;
-    TOKENEXTRA *tok[4][1 << 6], *tok_end;
-
-    vpx_memset(cpi->common.above_seg_context, 0, sizeof(PARTITION_CONTEXT) *
-               mi_cols_aligned_to_sb(&cpi->common));
-    tok[0][0] = cpi->tok;
-    for (tile_row = 0; tile_row < pc->tile_rows; tile_row++) {
-      if (tile_row) {
-        tok[tile_row][0] = tok[tile_row - 1][pc->tile_columns - 1] +
-                           cpi->tok_count[tile_row - 1][pc->tile_columns - 1];
-      }
-      for (tile_col = 1; tile_col < pc->tile_columns; tile_col++) {
-        tok[tile_row][tile_col] = tok[tile_row][tile_col - 1] +
-                                  cpi->tok_count[tile_row][tile_col - 1];
-      }
-    }
+void vp9_pack_bitstream(VP9_COMP *cpi, uint8_t *dest, unsigned long *size) {
+  uint8_t *data = dest;
+  size_t first_part_size;
+  struct vp9_write_bit_buffer wb = {data, 0};
+  struct vp9_write_bit_buffer saved_wb;
 
-    for (tile_row = 0; tile_row < pc->tile_rows; tile_row++) {
-      vp9_get_tile_row_offsets(pc, tile_row);
-      for (tile_col = 0; tile_col < pc->tile_columns; tile_col++) {
-        vp9_get_tile_col_offsets(pc, tile_col);
-        tok_end = tok[tile_row][tile_col] + cpi->tok_count[tile_row][tile_col];
-
-        if (tile_col < pc->tile_columns - 1 || tile_row < pc->tile_rows - 1)
-          vp9_start_encode(&residual_bc, data_ptr + total_size + 4);
-        else
-          vp9_start_encode(&residual_bc, data_ptr + total_size);
-        write_modes(cpi, &residual_bc, &tok[tile_row][tile_col], tok_end);
-        assert(tok[tile_row][tile_col] == tok_end);
-        vp9_stop_encode(&residual_bc);
-        if (tile_col < pc->tile_columns - 1 || tile_row < pc->tile_rows - 1) {
-          // size of this tile
-          write_be32(data_ptr + total_size, residual_bc.pos);
-          total_size += 4;
-        }
+  write_uncompressed_header(cpi, &wb);
+  saved_wb = wb;
+  vp9_wb_write_literal(&wb, 0, 16);  // don't know in advance first part. size
 
-        total_size += residual_bc.pos;
-      }
-    }
+  data += vp9_rb_bytes_written(&wb);
 
-    *size += total_size;
-  }
+  vp9_compute_update_table();
+
+#ifdef ENTROPY_STATS
+  if (pc->frame_type == INTER_FRAME)
+    active_section = 0;
+  else
+    active_section = 7;
+#endif
+
+  vp9_clear_system_state();  // __asm emms;
+
+  first_part_size = write_compressed_header(cpi, data);
+  data += first_part_size;
+  vp9_wb_write_literal(&saved_wb, first_part_size, 16);
+
+  data += encode_tiles(cpi, data);
+
+  *size = data - dest;
 }
 
 #ifdef ENTROPY_STATS