Roll latest libvpx into Android.

The latest libvpx just added initial multithread vp9 decoding support and more neon optimization.
Checkout is from master branch(hash:33afddadb9af6569bd8296ef1d48d0511b651e9d).

Change-Id: I54be2f48bc033c00876b6b1d0a3ff1eccb92a2fa

11 files changed, 661 insertions, 303 deletions

diff --git a/libvpx/vp9/decoder/vp9_decodemv.c b/libvpx/vp9/decoder/vp9_decodemv.c
index 6f0044a..a3e2ad3 100644
--- a/libvpx/vp9/decoder/vp9_decodemv.c
+++ b/libvpx/vp9/decoder/vp9_decodemv.c
@@ -30,8 +30,12 @@ static MB_PREDICTION_MODE read_intra_mode(vp9_reader *r, const vp9_prob *p) {
   return (MB_PREDICTION_MODE)treed_read(r, vp9_intra_mode_tree, p);
 }
 
-static MB_PREDICTION_MODE read_inter_mode(vp9_reader *r, const vp9_prob *p) {
-  return (MB_PREDICTION_MODE)treed_read(r, vp9_inter_mode_tree, p);
+static MB_PREDICTION_MODE read_inter_mode(VP9_COMMON *cm, vp9_reader *r,
+                                          uint8_t context) {
+  MB_PREDICTION_MODE mode = treed_read(r, vp9_inter_mode_tree,
+                            cm->fc.inter_mode_probs[context]);
+  ++cm->counts.inter_mode[context][inter_mode_offset(mode)];
+  return mode;
 }
 
 static int read_segment_id(vp9_reader *r, const struct segmentation *seg) {
@@ -43,9 +47,9 @@ static TX_SIZE read_selected_tx_size(VP9_COMMON *cm, MACROBLOCKD *xd,
   const uint8_t context = vp9_get_pred_context_tx_size(xd);
   const vp9_prob *tx_probs = get_tx_probs(bsize, context, &cm->fc.tx_probs);
   TX_SIZE tx_size = vp9_read(r, tx_probs[0]);
-  if (tx_size != TX_4X4 && bsize >= BLOCK_SIZE_MB16X16) {
+  if (tx_size != TX_4X4 && bsize >= BLOCK_16X16) {
     tx_size += vp9_read(r, tx_probs[1]);
-    if (tx_size != TX_8X8 && bsize >= BLOCK_SIZE_SB32X32)
+    if (tx_size != TX_8X8 && bsize >= BLOCK_32X32)
       tx_size += vp9_read(r, tx_probs[2]);
   }
 
@@ -54,18 +58,18 @@ static TX_SIZE read_selected_tx_size(VP9_COMMON *cm, MACROBLOCKD *xd,
 }
 
 static TX_SIZE read_tx_size(VP9D_COMP *pbi, TX_MODE tx_mode,
-                            BLOCK_SIZE_TYPE bsize, int select_cond,
+                            BLOCK_SIZE_TYPE bsize, int allow_select,
                             vp9_reader *r) {
   VP9_COMMON *const cm = &pbi->common;
   MACROBLOCKD *const xd = &pbi->mb;
 
-  if (tx_mode == TX_MODE_SELECT && bsize >= BLOCK_SIZE_SB8X8 && select_cond)
+  if (allow_select && tx_mode == TX_MODE_SELECT && bsize >= BLOCK_8X8)
     return read_selected_tx_size(cm, xd, bsize, r);
-  else if (tx_mode >= ALLOW_32X32 && bsize >= BLOCK_SIZE_SB32X32)
+  else if (tx_mode >= ALLOW_32X32 && bsize >= BLOCK_32X32)
     return TX_32X32;
-  else if (tx_mode >= ALLOW_16X16 && bsize >= BLOCK_SIZE_MB16X16)
+  else if (tx_mode >= ALLOW_16X16 && bsize >= BLOCK_16X16)
     return TX_16X16;
-  else if (tx_mode >= ALLOW_8X8 && bsize >= BLOCK_SIZE_SB8X8)
+  else if (tx_mode >= ALLOW_8X8 && bsize >= BLOCK_8X8)
     return TX_8X8;
   else
     return TX_4X4;
@@ -146,8 +150,8 @@ static uint8_t read_skip_coeff(VP9D_COMP *pbi, int segment_id, vp9_reader *r) {
   return skip_coeff;
 }
 
-static void read_intra_mode_info(VP9D_COMP *pbi, MODE_INFO *m,
-                                 int mi_row, int mi_col, vp9_reader *r) {
+static void read_intra_frame_mode_info(VP9D_COMP *pbi, MODE_INFO *m,
+                                       int mi_row, int mi_col, vp9_reader *r) {
   VP9_COMMON *const cm = &pbi->common;
   MACROBLOCKD *const xd = &pbi->mb;
   MB_MODE_INFO *const mbmi = &m->mbmi;
@@ -158,6 +162,7 @@ static void read_intra_mode_info(VP9D_COMP *pbi, MODE_INFO *m,
   mbmi->mb_skip_coeff = read_skip_coeff(pbi, mbmi->segment_id, r);
   mbmi->txfm_size = read_tx_size(pbi, cm->tx_mode, bsize, 1, r);
   mbmi->ref_frame[0] = INTRA_FRAME;
+  mbmi->ref_frame[1] = NONE;
 
   if (bsize >= BLOCK_SIZE_SB8X8) {
     const MB_PREDICTION_MODE A = above_block_mode(m, 0, mis);
@@ -166,12 +171,12 @@ static void read_intra_mode_info(VP9D_COMP *pbi, MODE_INFO *m,
     mbmi->mode = read_intra_mode(r, vp9_kf_y_mode_prob[A][L]);
   } else {
     // Only 4x4, 4x8, 8x4 blocks
-    const int bw = 1 << b_width_log2(bsize);
-    const int bh = 1 << b_height_log2(bsize);
+    const int num_4x4_w = num_4x4_blocks_wide_lookup[bsize];  // 1 or 2
+    const int num_4x4_h = num_4x4_blocks_high_lookup[bsize];  // 1 or 2
     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_h) {
+      for (idx = 0; idx < 2; idx += num_4x4_w) {
         const int ib = idy * 2 + idx;
         const MB_PREDICTION_MODE A = above_block_mode(m, ib, mis);
         const MB_PREDICTION_MODE L = (xd->left_available || idx) ?
@@ -179,9 +184,9 @@ static void read_intra_mode_info(VP9D_COMP *pbi, MODE_INFO *m,
         const MB_PREDICTION_MODE b_mode = read_intra_mode(r,
                                               vp9_kf_y_mode_prob[A][L]);
         m->bmi[ib].as_mode = b_mode;
-        if (bh == 2)
+        if (num_4x4_h == 2)
           m->bmi[ib + 2].as_mode = b_mode;
-        if (bw == 2)
+        if (num_4x4_w == 2)
           m->bmi[ib + 1].as_mode = b_mode;
       }
     }
@@ -228,16 +233,16 @@ static int read_mv_component(vp9_reader *r,
 
 static INLINE void read_mv(vp9_reader *r, MV *mv, const MV *ref,
                            const nmv_context *ctx,
-                           nmv_context_counts *counts, int usehp) {
+                           nmv_context_counts *counts, int allow_hp) {
   const MV_JOINT_TYPE j = treed_read(r, vp9_mv_joint_tree, ctx->joints);
+  const int use_hp = allow_hp && vp9_use_mv_hp(ref);
   MV diff = {0, 0};
 
-  usehp = usehp && vp9_use_mv_hp(ref);
   if (mv_joint_vertical(j))
-    diff.row = read_mv_component(r, &ctx->comps[0], usehp);
+    diff.row = read_mv_component(r, &ctx->comps[0], use_hp);
 
   if (mv_joint_horizontal(j))
-    diff.col = read_mv_component(r, &ctx->comps[1], usehp);
+    diff.col = read_mv_component(r, &ctx->comps[1], use_hp);
 
   vp9_inc_mv(&diff, counts);
 
@@ -245,29 +250,30 @@ static INLINE void read_mv(vp9_reader *r, MV *mv, const MV *ref,
   mv->col = ref->col + diff.col;
 }
 
-static void update_mv(vp9_reader *r, vp9_prob *p, vp9_prob upd_p) {
-  if (vp9_read(r, upd_p))
+static void update_mv(vp9_reader *r, vp9_prob *p) {
+  if (vp9_read(r, VP9_NMV_UPDATE_PROB))
     *p = (vp9_read_literal(r, 7) << 1) | 1;
 }
 
-static void read_mv_probs(vp9_reader *r, nmv_context *mvc, int usehp) {
+static void read_mv_probs(vp9_reader *r, nmv_context *mvc, int allow_hp) {
   int i, j, k;
 
   for (j = 0; j < MV_JOINTS - 1; ++j)
-    update_mv(r, &mvc->joints[j], VP9_NMV_UPDATE_PROB);
+    update_mv(r, &mvc->joints[j]);
 
   for (i = 0; i < 2; ++i) {
     nmv_component *const comp = &mvc->comps[i];
 
-    update_mv(r, &comp->sign, VP9_NMV_UPDATE_PROB);
+    update_mv(r, &comp->sign);
+
     for (j = 0; j < MV_CLASSES - 1; ++j)
-      update_mv(r, &comp->classes[j], VP9_NMV_UPDATE_PROB);
+      update_mv(r, &comp->classes[j]);
 
     for (j = 0; j < CLASS0_SIZE - 1; ++j)
-      update_mv(r, &comp->class0[j], VP9_NMV_UPDATE_PROB);
+      update_mv(r, &comp->class0[j]);
 
     for (j = 0; j < MV_OFFSET_BITS; ++j)
-      update_mv(r, &comp->bits[j], VP9_NMV_UPDATE_PROB);
+      update_mv(r, &comp->bits[j]);
   }
 
   for (i = 0; i < 2; ++i) {
@@ -275,23 +281,23 @@ static void read_mv_probs(vp9_reader *r, nmv_context *mvc, int usehp) {
 
     for (j = 0; j < CLASS0_SIZE; ++j)
       for (k = 0; k < 3; ++k)
-        update_mv(r, &comp->class0_fp[j][k], VP9_NMV_UPDATE_PROB);
+        update_mv(r, &comp->class0_fp[j][k]);
 
     for (j = 0; j < 3; ++j)
-      update_mv(r, &comp->fp[j], VP9_NMV_UPDATE_PROB);
+      update_mv(r, &comp->fp[j]);
   }
 
-  if (usehp) {
+  if (allow_hp) {
     for (i = 0; i < 2; ++i) {
-      update_mv(r, &mvc->comps[i].class0_hp, VP9_NMV_UPDATE_PROB);
-      update_mv(r, &mvc->comps[i].hp, VP9_NMV_UPDATE_PROB);
+      update_mv(r, &mvc->comps[i].class0_hp);
+      update_mv(r, &mvc->comps[i].hp);
     }
   }
 }
 
 // Read the referncence frame
-static void read_ref_frame(VP9D_COMP *pbi, vp9_reader *r,
-                           int segment_id, MV_REFERENCE_FRAME ref_frame[2]) {
+static void read_ref_frames(VP9D_COMP *pbi, vp9_reader *r,
+                            int segment_id, MV_REFERENCE_FRAME ref_frame[2]) {
   VP9_COMMON *const cm = &pbi->common;
   MACROBLOCKD *const xd = &pbi->mb;
   FRAME_CONTEXT *const fc = &cm->fc;
@@ -320,18 +326,19 @@ static void read_ref_frame(VP9D_COMP *pbi, vp9_reader *r,
       ref_frame[fix_ref_idx] = cm->comp_fixed_ref;
       ref_frame[!fix_ref_idx] = cm->comp_var_ref[b];
     } else {
-      const int ref1_ctx = vp9_get_pred_context_single_ref_p1(xd);
-      ref_frame[1] = NONE;
-      if (vp9_read(r, fc->single_ref_prob[ref1_ctx][0])) {
-        const int ref2_ctx = vp9_get_pred_context_single_ref_p2(xd);
-        const int b = vp9_read(r, fc->single_ref_prob[ref2_ctx][1]);
-        ref_frame[0] = b ? ALTREF_FRAME : GOLDEN_FRAME;
-        counts->single_ref[ref1_ctx][0][1]++;
-        counts->single_ref[ref2_ctx][1][b]++;
+      const int ctx0 = vp9_get_pred_context_single_ref_p1(xd);
+      const int bit0 = vp9_read(r, fc->single_ref_prob[ctx0][0]);
+      ++counts->single_ref[ctx0][0][bit0];
+      if (bit0) {
+        const int ctx1 = vp9_get_pred_context_single_ref_p2(xd);
+        const int bit1 = vp9_read(r, fc->single_ref_prob[ctx1][1]);
+        ref_frame[0] = bit1 ? ALTREF_FRAME : GOLDEN_FRAME;
+        ++counts->single_ref[ctx1][1][bit1];
       } else {
         ref_frame[0] = LAST_FRAME;
-        counts->single_ref[ref1_ctx][0][0]++;
       }
+
+      ref_frame[1] = NONE;
     }
   }
 }
@@ -359,16 +366,6 @@ static INLINE COMPPREDMODE_TYPE read_comp_pred_mode(vp9_reader *r) {
   return mode;
 }
 
-static INLINE void assign_and_clamp_mv(int_mv *dst, const int_mv *src,
-                                       int mb_to_left_edge,
-                                       int mb_to_right_edge,
-                                       int mb_to_top_edge,
-                                       int mb_to_bottom_edge) {
-  dst->as_int = src->as_int;
-  clamp_mv(dst, mb_to_left_edge, mb_to_right_edge, mb_to_top_edge,
-           mb_to_bottom_edge);
-}
-
 static INLINE INTERPOLATIONFILTERTYPE read_switchable_filter_type(
     VP9D_COMP *pbi, vp9_reader *r) {
   VP9_COMMON *const cm = &pbi->common;
@@ -380,32 +377,35 @@ static INLINE INTERPOLATIONFILTERTYPE read_switchable_filter_type(
   return vp9_switchable_interp[index];
 }
 
-static void read_intra_block_modes(VP9D_COMP *pbi, MODE_INFO *mi,
-                                   vp9_reader *r) {
+static void read_intra_block_mode_info(VP9D_COMP *pbi, MODE_INFO *mi,
+                                  vp9_reader *r) {
   VP9_COMMON *const cm = &pbi->common;
   MB_MODE_INFO *const mbmi = &mi->mbmi;
   const BLOCK_SIZE_TYPE bsize = mi->mbmi.sb_type;
-  const int bwl = b_width_log2(bsize), bhl = b_height_log2(bsize);
+
+  mbmi->ref_frame[0] = INTRA_FRAME;
+  mbmi->ref_frame[1] = NONE;
 
   if (bsize >= BLOCK_SIZE_SB8X8) {
-    const int size_group = MIN(3, MIN(bwl, bhl));
+    const int size_group = size_group_lookup[bsize];
     mbmi->mode = read_intra_mode(r, cm->fc.y_mode_prob[size_group]);
     cm->counts.y_mode[size_group][mbmi->mode]++;
   } else {
      // Only 4x4, 4x8, 8x4 blocks
-     const int bw = 1 << bwl, bh = 1 << bhl;
+     const int num_4x4_w = num_4x4_blocks_wide_lookup[bsize];  // 1 or 2
+     const int num_4x4_h = num_4x4_blocks_high_lookup[bsize];  // 1 or 2
      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_h) {
+       for (idx = 0; idx < 2; idx += num_4x4_w) {
          const int ib = idy * 2 + idx;
          const int b_mode = read_intra_mode(r, cm->fc.y_mode_prob[0]);
          mi->bmi[ib].as_mode = b_mode;
          cm->counts.y_mode[0][b_mode]++;
 
-         if (bh == 2)
+         if (num_4x4_h == 2)
            mi->bmi[ib + 2].as_mode = b_mode;
-         if (bw == 2)
+         if (num_4x4_w == 2)
            mi->bmi[ib + 1].as_mode = b_mode;
       }
     }
@@ -416,203 +416,197 @@ static void read_intra_block_modes(VP9D_COMP *pbi, MODE_INFO *mi,
   cm->counts.uv_mode[mbmi->mode][mbmi->uv_mode]++;
 }
 
-static MV_REFERENCE_FRAME read_reference_frame(VP9D_COMP *pbi, int segment_id,
-                                               vp9_reader *r) {
+static int read_is_inter_block(VP9D_COMP *pbi, int segment_id, vp9_reader *r) {
   VP9_COMMON *const cm = &pbi->common;
   MACROBLOCKD *const xd = &pbi->mb;
 
-  MV_REFERENCE_FRAME ref;
-  if (!vp9_segfeature_active(&xd->seg, segment_id, SEG_LVL_REF_FRAME)) {
-    const int ctx = vp9_get_pred_context_intra_inter(xd);
-    ref = (MV_REFERENCE_FRAME)
-              vp9_read(r, vp9_get_pred_prob_intra_inter(cm, xd));
-    cm->counts.intra_inter[ctx][ref != INTRA_FRAME]++;
+  if (vp9_segfeature_active(&xd->seg, segment_id, SEG_LVL_REF_FRAME)) {
+    return vp9_get_segdata(&xd->seg, segment_id, SEG_LVL_REF_FRAME) !=
+           INTRA_FRAME;
   } else {
-    ref = (MV_REFERENCE_FRAME) vp9_get_segdata(&xd->seg, segment_id,
-                                   SEG_LVL_REF_FRAME) != INTRA_FRAME;
+    const int ctx = vp9_get_pred_context_intra_inter(xd);
+    const int is_inter = vp9_read(r, vp9_get_pred_prob_intra_inter(cm, xd));
+    ++cm->counts.intra_inter[ctx][is_inter];
+    return is_inter;
   }
-  return ref;
 }
 
-static void read_inter_mode_info(VP9D_COMP *pbi, MODE_INFO *mi,
-                                 int mi_row, int mi_col, vp9_reader *r) {
+static void read_inter_block_mode_info(VP9D_COMP *pbi, MODE_INFO *mi,
+                                       int mi_row, int mi_col, vp9_reader *r) {
   VP9_COMMON *const cm = &pbi->common;
   MACROBLOCKD *const xd = &pbi->mb;
   nmv_context *const nmvc = &cm->fc.nmvc;
   MB_MODE_INFO *const mbmi = &mi->mbmi;
-
   int_mv *const mv0 = &mbmi->mv[0];
   int_mv *const mv1 = &mbmi->mv[1];
-  const BLOCK_SIZE_TYPE bsize = mi->mbmi.sb_type;
-  const int bw = 1 << b_width_log2(bsize);
-  const int bh = 1 << b_height_log2(bsize);
-
-  int idx, idy;
+  const BLOCK_SIZE_TYPE bsize = mbmi->sb_type;
+  const int allow_hp = xd->allow_high_precision_mv;
 
-  mbmi->segment_id = read_inter_segment_id(pbi, mi_row, mi_col, r);
-  mbmi->mb_skip_coeff = read_skip_coeff(pbi, mbmi->segment_id, r);
-  mbmi->ref_frame[0] = read_reference_frame(pbi, mbmi->segment_id, r);
-  mbmi->ref_frame[1] = NONE;
-  mbmi->txfm_size = read_tx_size(pbi, cm->tx_mode, bsize,
-     (!mbmi->mb_skip_coeff || mbmi->ref_frame[0] == INTRA_FRAME), r);
+  int_mv nearest, nearby, best_mv;
+  int_mv nearest_second, nearby_second, best_mv_second;
+  uint8_t inter_mode_ctx;
+  MV_REFERENCE_FRAME ref0, ref1;
 
-  if (mbmi->ref_frame[0] != INTRA_FRAME) {
-    int_mv nearest, nearby, best_mv;
-    int_mv nearest_second, nearby_second, best_mv_second;
-    vp9_prob *mv_ref_p;
-    MV_REFERENCE_FRAME ref0, ref1;
+  read_ref_frames(pbi, r, mbmi->segment_id, mbmi->ref_frame);
+  ref0 = mbmi->ref_frame[0];
+  ref1 = mbmi->ref_frame[1];
 
-    read_ref_frame(pbi, r, mbmi->segment_id, mbmi->ref_frame);
-    ref0 = mbmi->ref_frame[0];
-    ref1 = mbmi->ref_frame[1];
+  vp9_find_mv_refs(cm, xd, mi, xd->prev_mode_info_context,
+                   ref0, mbmi->ref_mvs[ref0], cm->ref_frame_sign_bias,
+                   mi_row, mi_col);
 
-    vp9_find_mv_refs(cm, xd, mi, xd->prev_mode_info_context,
-                     ref0, mbmi->ref_mvs[ref0], cm->ref_frame_sign_bias);
+  inter_mode_ctx = mbmi->mb_mode_context[ref0];
 
-    mv_ref_p = cm->fc.inter_mode_probs[mbmi->mb_mode_context[ref0]];
+  if (vp9_segfeature_active(&xd->seg, mbmi->segment_id, SEG_LVL_SKIP))
+    mbmi->mode = ZEROMV;
+  else if (bsize >= BLOCK_SIZE_SB8X8)
+    mbmi->mode = read_inter_mode(cm, r, inter_mode_ctx);
 
-    if (vp9_segfeature_active(&xd->seg, mbmi->segment_id, SEG_LVL_SKIP)) {
-      mbmi->mode = ZEROMV;
-    } else if (bsize >= BLOCK_SIZE_SB8X8) {
-      mbmi->mode = read_inter_mode(r, mv_ref_p);
-      vp9_accum_mv_refs(cm, mbmi->mode, mbmi->mb_mode_context[ref0]);
-    }
-    mbmi->uv_mode = DC_PRED;
+  mbmi->uv_mode = DC_PRED;
 
-    // nearest, nearby
-    if (bsize < BLOCK_SIZE_SB8X8 || mbmi->mode != ZEROMV) {
-      vp9_find_best_ref_mvs(xd, mbmi->ref_mvs[ref0], &nearest, &nearby);
-      best_mv.as_int = mbmi->ref_mvs[ref0][0].as_int;
-    }
+  // nearest, nearby
+  if (bsize < BLOCK_SIZE_SB8X8 || mbmi->mode != ZEROMV) {
+    vp9_find_best_ref_mvs(xd, mbmi->ref_mvs[ref0], &nearest, &nearby);
+    best_mv.as_int = mbmi->ref_mvs[ref0][0].as_int;
+  }
 
-    mbmi->interp_filter = cm->mcomp_filter_type == SWITCHABLE
-                              ? read_switchable_filter_type(pbi, r)
-                              : cm->mcomp_filter_type;
+  mbmi->interp_filter = cm->mcomp_filter_type == SWITCHABLE
+                            ? read_switchable_filter_type(pbi, r)
+                            : cm->mcomp_filter_type;
 
-    if (ref1 > INTRA_FRAME) {
-      vp9_find_mv_refs(cm, xd, mi, xd->prev_mode_info_context,
-                       ref1, mbmi->ref_mvs[ref1], cm->ref_frame_sign_bias);
+  if (ref1 > INTRA_FRAME) {
+    vp9_find_mv_refs(cm, xd, mi, xd->prev_mode_info_context,
+                     ref1, mbmi->ref_mvs[ref1], cm->ref_frame_sign_bias,
+                     mi_row, mi_col);
 
-      if (bsize < BLOCK_SIZE_SB8X8 || mbmi->mode != ZEROMV) {
-        vp9_find_best_ref_mvs(xd, mbmi->ref_mvs[ref1],
-                              &nearest_second, &nearby_second);
-        best_mv_second.as_int = mbmi->ref_mvs[ref1][0].as_int;
-      }
+    if (bsize < BLOCK_SIZE_SB8X8 || mbmi->mode != ZEROMV) {
+      vp9_find_best_ref_mvs(xd, mbmi->ref_mvs[ref1],
+                            &nearest_second, &nearby_second);
+      best_mv_second.as_int = mbmi->ref_mvs[ref1][0].as_int;
     }
+  }
 
+  if (bsize < BLOCK_SIZE_SB8X8) {
+    const int num_4x4_w = num_4x4_blocks_wide_lookup[bsize];  // 1 or 2
+    const int num_4x4_h = num_4x4_blocks_high_lookup[bsize];  // 1 or 2
+    int idx, idy;
+    for (idy = 0; idy < 2; idy += num_4x4_h) {
+      for (idx = 0; idx < 2; idx += num_4x4_w) {
+        int_mv blockmv, secondmv;
+        const int j = idy * 2 + idx;
+        const int b_mode = read_inter_mode(cm, r, inter_mode_ctx);
 
-    if (mbmi->sb_type < BLOCK_SIZE_SB8X8) {
-      for (idy = 0; idy < 2; idy += bh) {
-        for (idx = 0; idx < 2; idx += bw) {
-          int_mv blockmv, secondmv;
-          const int j = idy * 2 + idx;
-          const int blockmode = read_inter_mode(r, mv_ref_p);
+        if (b_mode == NEARESTMV || b_mode == NEARMV) {
+          vp9_append_sub8x8_mvs_for_idx(cm, xd, &nearest, &nearby, j, 0,
+                                        mi_row, mi_col);
 
-          vp9_accum_mv_refs(cm, blockmode, mbmi->mb_mode_context[ref0]);
-          if (blockmode == NEARESTMV || blockmode == NEARMV) {
-            vp9_append_sub8x8_mvs_for_idx(cm, xd, &nearest, &nearby, j, 0);
-            if (ref1 > 0)
-              vp9_append_sub8x8_mvs_for_idx(cm, xd,  &nearest_second,
-                                            &nearby_second, j, 1);
-          }
-
-          switch (blockmode) {
-            case NEWMV:
-              read_mv(r, &blockmv.as_mv, &best_mv.as_mv, nmvc,
-                      &cm->counts.mv, xd->allow_high_precision_mv);
-
-              if (ref1 > 0)
-                read_mv(r, &secondmv.as_mv, &best_mv_second.as_mv, nmvc,
-                        &cm->counts.mv, xd->allow_high_precision_mv);
-              break;
-            case NEARESTMV:
-              blockmv.as_int = nearest.as_int;
-              if (ref1 > 0)
-                secondmv.as_int = nearest_second.as_int;
-              break;
-            case NEARMV:
-              blockmv.as_int = nearby.as_int;
-              if (ref1 > 0)
-                secondmv.as_int = nearby_second.as_int;
-              break;
-            case ZEROMV:
-              blockmv.as_int = 0;
-              if (ref1 > 0)
-                secondmv.as_int = 0;
-              break;
-            default:
-              assert(!"Invalid inter mode value");
-          }
-          mi->bmi[j].as_mv[0].as_int = blockmv.as_int;
           if (ref1 > 0)
-            mi->bmi[j].as_mv[1].as_int = secondmv.as_int;
-
-          if (bh == 2)
-            mi->bmi[j + 2] = mi->bmi[j];
-          if (bw == 2)
-            mi->bmi[j + 1] = mi->bmi[j];
-          mi->mbmi.mode = blockmode;
+            vp9_append_sub8x8_mvs_for_idx(cm, xd,  &nearest_second,
+                                         &nearby_second, j, 1,
+                                         mi_row, mi_col);
         }
-      }
 
-      mv0->as_int = mi->bmi[3].as_mv[0].as_int;
-      mv1->as_int = mi->bmi[3].as_mv[1].as_int;
-    } else {
-      const int mb_to_top_edge = xd->mb_to_top_edge - LEFT_TOP_MARGIN;
-      const int mb_to_bottom_edge = xd->mb_to_bottom_edge + RIGHT_BOTTOM_MARGIN;
-      const int mb_to_left_edge = xd->mb_to_left_edge - LEFT_TOP_MARGIN;
-      const int mb_to_right_edge = xd->mb_to_right_edge + RIGHT_BOTTOM_MARGIN;
-
-      switch (mbmi->mode) {
-        case NEARMV:
-          // Clip "next_nearest" so that it does not extend to far out of image
-          assign_and_clamp_mv(mv0, &nearby, mb_to_left_edge,
-                                            mb_to_right_edge,
-                                            mb_to_top_edge,
-                                            mb_to_bottom_edge);
-          if (ref1 > 0)
-            assign_and_clamp_mv(mv1, &nearby_second, mb_to_left_edge,
-                                                     mb_to_right_edge,
-                                                     mb_to_top_edge,
-                                                     mb_to_bottom_edge);
-          break;
-
-        case NEARESTMV:
-          // Clip "next_nearest" so that it does not extend to far out of image
-          assign_and_clamp_mv(mv0, &nearest, mb_to_left_edge,
-                                             mb_to_right_edge,
-                                             mb_to_top_edge,
-                                             mb_to_bottom_edge);
-          if (ref1 > 0)
-            assign_and_clamp_mv(mv1, &nearest_second, mb_to_left_edge,
-                                                      mb_to_right_edge,
-                                                      mb_to_top_edge,
-                                                      mb_to_bottom_edge);
-          break;
-
-        case ZEROMV:
-          mv0->as_int = 0;
-          if (ref1 > 0)
-            mv1->as_int = 0;
-          break;
+        switch (b_mode) {
+          case NEWMV:
+            read_mv(r, &blockmv.as_mv, &best_mv.as_mv, nmvc,
+                    &cm->counts.mv, allow_hp);
 
-        case NEWMV:
-          read_mv(r, &mv0->as_mv, &best_mv.as_mv, nmvc, &cm->counts.mv,
-                  xd->allow_high_precision_mv);
-          if (ref1 > 0)
-            read_mv(r, &mv1->as_mv, &best_mv_second.as_mv, nmvc,
-                    &cm->counts.mv, xd->allow_high_precision_mv);
-          break;
-        default:
-          assert(!"Invalid inter mode value");
+            if (ref1 > 0)
+              read_mv(r, &secondmv.as_mv, &best_mv_second.as_mv, nmvc,
+                      &cm->counts.mv, allow_hp);
+            break;
+          case NEARESTMV:
+            blockmv.as_int = nearest.as_int;
+            if (ref1 > 0)
+              secondmv.as_int = nearest_second.as_int;
+            break;
+          case NEARMV:
+            blockmv.as_int = nearby.as_int;
+            if (ref1 > 0)
+              secondmv.as_int = nearby_second.as_int;
+            break;
+          case ZEROMV:
+            blockmv.as_int = 0;
+            if (ref1 > 0)
+              secondmv.as_int = 0;
+            break;
+          default:
+            assert(!"Invalid inter mode value");
+        }
+        mi->bmi[j].as_mv[0].as_int = blockmv.as_int;
+        if (ref1 > 0)
+          mi->bmi[j].as_mv[1].as_int = secondmv.as_int;
+
+        if (num_4x4_h == 2)
+          mi->bmi[j + 2] = mi->bmi[j];
+        if (num_4x4_w == 2)
+          mi->bmi[j + 1] = mi->bmi[j];
+        mi->mbmi.mode = b_mode;
       }
     }
+
+    mv0->as_int = mi->bmi[3].as_mv[0].as_int;
+    mv1->as_int = mi->bmi[3].as_mv[1].as_int;
   } else {
-    mv0->as_int = 0;  // required for left and above block mv
-    read_intra_block_modes(pbi, mi, r);
+    switch (mbmi->mode) {
+      case NEARMV:
+        mv0->as_int = nearby.as_int;
+        clamp_mv2(&mv0->as_mv, xd);
+
+        if (ref1 > 0) {
+          mv1->as_int = nearby_second.as_int;
+          clamp_mv2(&mv1->as_mv, xd);
+        }
+        break;
+
+      case NEARESTMV:
+        mv0->as_int = nearest.as_int;
+        clamp_mv2(&mv0->as_mv, xd);
+
+        if (ref1 > 0) {
+          mv1->as_int = nearest_second.as_int;
+          clamp_mv2(&mv1->as_mv, xd);
+        }
+        break;
+
+      case ZEROMV:
+        mv0->as_int = 0;
+        if (ref1 > 0)
+          mv1->as_int = 0;
+        break;
+
+      case NEWMV:
+        read_mv(r, &mv0->as_mv, &best_mv.as_mv, nmvc, &cm->counts.mv, allow_hp);
+        if (ref1 > 0)
+          read_mv(r, &mv1->as_mv, &best_mv_second.as_mv, nmvc, &cm->counts.mv,
+                  allow_hp);
+        break;
+      default:
+        assert(!"Invalid inter mode value");
+    }
   }
 }
 
+static void read_inter_frame_mode_info(VP9D_COMP *pbi, MODE_INFO *mi,
+                                       int mi_row, int mi_col, vp9_reader *r) {
+  VP9_COMMON *const cm = &pbi->common;
+  MB_MODE_INFO *const mbmi = &mi->mbmi;
+  int inter_block;
+
+  mbmi->mv[0].as_int = 0;
+  mbmi->mv[1].as_int = 0;
+  mbmi->segment_id = read_inter_segment_id(pbi, mi_row, mi_col, r);
+  mbmi->mb_skip_coeff = read_skip_coeff(pbi, mbmi->segment_id, r);
+  inter_block = read_is_inter_block(pbi, mbmi->segment_id, r);
+  mbmi->txfm_size = read_tx_size(pbi, cm->tx_mode, mbmi->sb_type,
+                                 !mbmi->mb_skip_coeff || !inter_block, r);
+
+  if (inter_block)
+    read_inter_block_mode_info(pbi, mi, mi_row, mi_col, r);
+  else
+    read_intra_block_mode_info(pbi, mi, r);
+}
+
 static void read_comp_pred(VP9_COMMON *cm, vp9_reader *r) {
   int i;
 
@@ -690,9 +684,9 @@ void vp9_read_mode_info(VP9D_COMP* pbi, int mi_row, int mi_col, vp9_reader *r) {
   int x, y;
 
   if (cm->frame_type == KEY_FRAME || cm->intra_only)
-    read_intra_mode_info(pbi, mi, mi_row, mi_col, r);
+    read_intra_frame_mode_info(pbi, mi, mi_row, mi_col, r);
   else
-    read_inter_mode_info(pbi, mi, mi_row, mi_col, r);
+    read_inter_frame_mode_info(pbi, mi, mi_row, mi_col, r);
 
   for (y = 0; y < y_mis; y++)
     for (x = !y; x < x_mis; x++)
diff --git a/libvpx/vp9/decoder/vp9_decodemv.h b/libvpx/vp9/decoder/vp9_decodemv.h
index 4073d9e..462d2e3 100644
--- a/libvpx/vp9/decoder/vp9_decodemv.h
+++ b/libvpx/vp9/decoder/vp9_decodemv.h
@@ -12,6 +12,7 @@
 #define VP9_DECODER_VP9_DECODEMV_H_
 
 #include "vp9/decoder/vp9_onyxd_int.h"
+#include "vp9/decoder/vp9_dboolhuff.h"
 
 void vp9_prepare_read_mode_info(VP9D_COMP* pbi, vp9_reader *r);
 
diff --git a/libvpx/vp9/decoder/vp9_decodframe.c b/libvpx/vp9/decoder/vp9_decodframe.c
index ffec8ea..feb6024 100644
--- a/libvpx/vp9/decoder/vp9_decodframe.c
+++ b/libvpx/vp9/decoder/vp9_decodframe.c
@@ -31,8 +31,11 @@
 #include "vp9/decoder/vp9_detokenize.h"
 #include "vp9/decoder/vp9_decodemv.h"
 #include "vp9/decoder/vp9_dsubexp.h"
+#include "vp9/decoder/vp9_idct_blk.h"
 #include "vp9/decoder/vp9_onyxd_int.h"
 #include "vp9/decoder/vp9_read_bit_buffer.h"
+#include "vp9/decoder/vp9_thread.h"
+#include "vp9/decoder/vp9_treereader.h"
 
 static int read_be32(const uint8_t *p) {
   return (p[0] << 24) | (p[1] << 16) | (p[2] << 8) | p[3];
@@ -59,17 +62,17 @@ static void read_tx_probs(struct tx_probs *tx_probs, vp9_reader *r) {
   int i, j;
 
   for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
-    for (j = 0; j < TX_SIZE_MAX_SB - 3; ++j)
+    for (j = 0; j < TX_SIZES - 3; ++j)
       if (vp9_read(r, VP9_MODE_UPDATE_PROB))
         vp9_diff_update_prob(r, &tx_probs->p8x8[i][j]);
 
   for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
-    for (j = 0; j < TX_SIZE_MAX_SB - 2; ++j)
+    for (j = 0; j < TX_SIZES - 2; ++j)
       if (vp9_read(r, VP9_MODE_UPDATE_PROB))
         vp9_diff_update_prob(r, &tx_probs->p16x16[i][j]);
 
   for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
-    for (j = 0; j < TX_SIZE_MAX_SB - 1; ++j)
+    for (j = 0; j < TX_SIZES - 1; ++j)
       if (vp9_read(r, VP9_MODE_UPDATE_PROB))
         vp9_diff_update_prob(r, &tx_probs->p32x32[i][j]);
 }
@@ -138,8 +141,8 @@ static void decode_block_intra(int plane, int block, BLOCK_SIZE_TYPE bsize,
   const int mode = plane == 0 ? mi->mbmi.mode
                               : mi->mbmi.uv_mode;
 
-  if (plane == 0 && mi->mbmi.sb_type < BLOCK_SIZE_SB8X8) {
-    assert(bsize == BLOCK_SIZE_SB8X8);
+  if (plane == 0 && mi->mbmi.sb_type < BLOCK_8X8) {
+    assert(bsize == BLOCK_8X8);
     b_mode = mi->bmi[raster_block].as_mode;
   } else {
     b_mode = mode;
@@ -223,7 +226,7 @@ static void decode_modes_b(VP9D_COMP *pbi, int mi_row, int mi_col,
                            vp9_reader *r, BLOCK_SIZE_TYPE bsize) {
   VP9_COMMON *const cm = &pbi->common;
   MACROBLOCKD *const xd = &pbi->mb;
-  const int less8x8 = bsize < BLOCK_SIZE_SB8X8;
+  const int less8x8 = bsize < BLOCK_8X8;
   MB_MODE_INFO *mbmi;
 
   if (less8x8)
@@ -234,12 +237,12 @@ static void decode_modes_b(VP9D_COMP *pbi, int mi_row, int mi_col,
   vp9_read_mode_info(pbi, mi_row, mi_col, r);
 
   if (less8x8)
-    bsize = BLOCK_SIZE_SB8X8;
+    bsize = BLOCK_8X8;
 
   // Has to be called after set_offsets
   mbmi = &xd->mode_info_context->mbmi;
 
-  if (mbmi->ref_frame[0] == INTRA_FRAME) {
+  if (!is_inter_block(mbmi)) {
     // Intra reconstruction
     decode_tokens(pbi, bsize, r);
     foreach_transformed_block(xd, bsize, decode_block_intra, xd);
@@ -280,12 +283,12 @@ static void decode_modes_sb(VP9D_COMP *pbi, int mi_row, int mi_col,
   if (mi_row >= pc->mi_rows || mi_col >= pc->mi_cols)
     return;
 
-  if (bsize < BLOCK_SIZE_SB8X8) {
+  if (bsize < BLOCK_8X8) {
     if (xd->ab_index != 0)
       return;
   } else {
     int pl;
-    const int idx = check_bsize_coverage(pc, xd, mi_row, mi_col, bsize);
+    const int idx = check_bsize_coverage(pc, mi_row, mi_col, bsize);
     set_partition_seg_context(pc, xd, mi_row, mi_col);
     pl = partition_plane_context(xd, bsize);
 
@@ -332,8 +335,8 @@ static void decode_modes_sb(VP9D_COMP *pbi, int mi_row, int mi_col,
   }
 
   // update partition context
-  if (bsize >= BLOCK_SIZE_SB8X8 &&
-      (bsize == BLOCK_SIZE_SB8X8 || partition != PARTITION_SPLIT)) {
+  if (bsize >= BLOCK_8X8 &&
+      (bsize == BLOCK_8X8 || partition != PARTITION_SPLIT)) {
     set_partition_seg_context(pc, xd, mi_row, mi_col);
     update_partition_context(xd, subsize, bsize);
   }
@@ -499,7 +502,7 @@ static INTERPOLATIONFILTERTYPE read_interp_filter_type(
                              : vp9_rb_read_literal(rb, 2);
 }
 
-static void read_frame_size(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb,
+static void read_frame_size(struct vp9_read_bit_buffer *rb,
                             int *width, int *height) {
   const int w = vp9_rb_read_literal(rb, 16) + 1;
   const int h = vp9_rb_read_literal(rb, 16) + 1;
@@ -507,12 +510,11 @@ static void read_frame_size(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb,
   *height = h;
 }
 
-static void setup_display_size(VP9D_COMP *pbi, struct vp9_read_bit_buffer *rb) {
-  VP9_COMMON *const cm = &pbi->common;
+static void setup_display_size(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
   cm->display_width = cm->width;
   cm->display_height = cm->height;
   if (vp9_rb_read_bit(rb))
-    read_frame_size(cm, rb, &cm->display_width, &cm->display_height);
+    read_frame_size(rb, &cm->display_width, &cm->display_height);
 }
 
 static void apply_frame_size(VP9D_COMP *pbi, int width, int height) {
@@ -548,10 +550,9 @@ static void apply_frame_size(VP9D_COMP *pbi, int width, int height) {
 
 static void setup_frame_size(VP9D_COMP *pbi,
                              struct vp9_read_bit_buffer *rb) {
-  VP9_COMMON *const cm = &pbi->common;
   int width, height;
-  read_frame_size(cm, rb, &width, &height);
-  setup_display_size(pbi, rb);
+  read_frame_size(rb, &width, &height);
+  setup_display_size(&pbi->common, rb);
   apply_frame_size(pbi, width, height);
 }
 
@@ -572,21 +573,29 @@ static void setup_frame_size_with_refs(VP9D_COMP *pbi,
   }
 
   if (!found)
-    read_frame_size(cm, rb, &width, &height);
+    read_frame_size(rb, &width, &height);
 
   if (!width || !height)
     vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
                        "Referenced frame with invalid size");
 
-  setup_display_size(pbi, rb);
+  setup_display_size(cm, rb);
   apply_frame_size(pbi, width, height);
 }
 
 static void decode_tile(VP9D_COMP *pbi, vp9_reader *r) {
+  const int num_threads = pbi->oxcf.max_threads;
   VP9_COMMON *const pc = &pbi->common;
   int mi_row, mi_col;
 
   if (pbi->do_loopfilter_inline) {
+    if (num_threads > 1) {
+      LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1;
+      lf_data->frame_buffer = &pbi->common.yv12_fb[pbi->common.new_fb_idx];
+      lf_data->cm = pc;
+      lf_data->xd = pbi->mb;
+      lf_data->y_only = 0;
+    }
     vp9_loop_filter_frame_init(pc, &pbi->mb, pbi->mb.lf.filter_level);
   }
 
@@ -597,21 +606,37 @@ static void decode_tile(VP9D_COMP *pbi, vp9_reader *r) {
     vpx_memset(pc->left_seg_context, 0, sizeof(pc->left_seg_context));
     for (mi_col = pc->cur_tile_mi_col_start; mi_col < pc->cur_tile_mi_col_end;
          mi_col += MI_BLOCK_SIZE) {
-      decode_modes_sb(pbi, mi_row, mi_col, r, BLOCK_SIZE_SB64X64);
+      decode_modes_sb(pbi, mi_row, mi_col, r, BLOCK_64X64);
     }
 
     if (pbi->do_loopfilter_inline) {
-      YV12_BUFFER_CONFIG *const fb =
-          &pbi->common.yv12_fb[pbi->common.new_fb_idx];
       // delay the loopfilter by 1 macroblock row.
       const int lf_start = mi_row - MI_BLOCK_SIZE;
       if (lf_start < 0) continue;
-      vp9_loop_filter_rows(fb, pc, &pbi->mb, lf_start, mi_row, 0);
+
+      if (num_threads > 1) {
+        LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1;
+
+        vp9_worker_sync(&pbi->lf_worker);
+        lf_data->start = lf_start;
+        lf_data->stop = mi_row;
+        pbi->lf_worker.hook = vp9_loop_filter_worker;
+        vp9_worker_launch(&pbi->lf_worker);
+      } else {
+        YV12_BUFFER_CONFIG *const fb =
+            &pbi->common.yv12_fb[pbi->common.new_fb_idx];
+        vp9_loop_filter_rows(fb, pc, &pbi->mb, lf_start, mi_row, 0);
+      }
     }
   }
 
   if (pbi->do_loopfilter_inline) {
     YV12_BUFFER_CONFIG *const fb = &pbi->common.yv12_fb[pbi->common.new_fb_idx];
+    if (num_threads > 1) {
+      // TODO(jzern): since the loop filter is delayed one mb row, this will be
+      // forced to wait for the last row scheduled in the for loop.
+      vp9_worker_sync(&pbi->lf_worker);
+    }
     vp9_loop_filter_rows(fb, pc, &pbi->mb,
                          mi_row - MI_BLOCK_SIZE, pc->mi_rows, 0);
   }
@@ -994,7 +1019,6 @@ int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) {
 
     if (!keyframe && !pc->intra_only) {
       vp9_adapt_mode_probs(pc);
-      vp9_adapt_mode_context(pc);
       vp9_adapt_mv_probs(pc, xd->allow_high_precision_mv);
     }
   }
diff --git a/libvpx/vp9/decoder/vp9_detokenize.c b/libvpx/vp9/decoder/vp9_detokenize.c
index 01c1db0..0021643 100644
--- a/libvpx/vp9/decoder/vp9_detokenize.c
+++ b/libvpx/vp9/decoder/vp9_detokenize.c
@@ -15,8 +15,10 @@
 #include "vp9/common/vp9_common.h"
 #include "vp9/common/vp9_seg_common.h"
 
+#include "vp9/decoder/vp9_dboolhuff.h"
 #include "vp9/decoder/vp9_detokenize.h"
 #include "vp9/decoder/vp9_onyxd_int.h"
+#include "vp9/decoder/vp9_treereader.h"
 
 #define EOB_CONTEXT_NODE            0
 #define ZERO_CONTEXT_NODE           1
@@ -73,7 +75,7 @@ DECLARE_ALIGNED(16, extern const uint8_t,
 #define WRITE_COEF_CONTINUE(val, token)                  \
   {                                                      \
     qcoeff_ptr[scan[c]] = vp9_read_and_apply_sign(r, val) * \
-                            dq[c > 0] / (1 + (txfm_size == TX_32X32)); \
+                            dq[c > 0] / (1 + (tx_size == TX_32X32)); \
     INCREMENT_COUNT(token);                              \
     c++;                                                 \
     continue;                                            \
@@ -88,33 +90,24 @@ DECLARE_ALIGNED(16, extern const uint8_t,
 static int decode_coefs(VP9_COMMON *cm, const MACROBLOCKD *xd,
                         vp9_reader *r, int block_idx,
                         PLANE_TYPE type, int seg_eob, int16_t *qcoeff_ptr,
-                        TX_SIZE txfm_size, const int16_t *dq,
+                        TX_SIZE tx_size, const int16_t *dq,
                         ENTROPY_CONTEXT *A, ENTROPY_CONTEXT *L) {
   FRAME_CONTEXT *const fc = &cm->fc;
   FRAME_COUNTS *const counts = &cm->counts;
   ENTROPY_CONTEXT above_ec, left_ec;
-  int pt, c = 0;
-  int band;
-  vp9_prob (*coef_probs)[PREV_COEF_CONTEXTS][UNCONSTRAINED_NODES];
+  const int ref = is_inter_block(&xd->mode_info_context->mbmi);
+  int band, pt, c = 0;
+  vp9_prob (*coef_probs)[PREV_COEF_CONTEXTS][UNCONSTRAINED_NODES] =
+      fc->coef_probs[tx_size][type][ref];
   vp9_prob coef_probs_full[COEF_BANDS][PREV_COEF_CONTEXTS][ENTROPY_NODES];
-  uint8_t load_map[COEF_BANDS][PREV_COEF_CONTEXTS] = {
-    {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, 0, 0, 0},
-    {0, 0, 0, 0, 0, 0},
-  };
-
+  uint8_t load_map[COEF_BANDS][PREV_COEF_CONTEXTS] = { { 0 } };
   vp9_prob *prob;
-  vp9_coeff_count_model *coef_counts;
-  const int ref = xd->mode_info_context->mbmi.ref_frame[0] != INTRA_FRAME;
+  vp9_coeff_count_model *coef_counts = counts->coef[tx_size];
   const int16_t *scan, *nb;
   uint8_t token_cache[1024];
   const uint8_t * band_translate;
-  coef_probs  = fc->coef_probs[txfm_size][type][ref];
-  coef_counts = counts->coef[txfm_size];
-  switch (txfm_size) {
+
+  switch (tx_size) {
     default:
     case TX_4X4: {
       scan = get_scan_4x4(get_tx_type_4x4(type, xd, block_idx));
@@ -125,22 +118,22 @@ static int decode_coefs(VP9_COMMON *cm, const MACROBLOCKD *xd,
     }
     case TX_8X8: {
       scan = get_scan_8x8(get_tx_type_8x8(type, xd));
-      above_ec = (A[0] + A[1]) != 0;
-      left_ec = (L[0] + L[1]) != 0;
+      above_ec = !!*(uint16_t *)A;
+      left_ec  = !!*(uint16_t *)L;
       band_translate = vp9_coefband_trans_8x8plus;
       break;
     }
     case TX_16X16: {
       scan = get_scan_16x16(get_tx_type_16x16(type, xd));
-      above_ec = (A[0] + A[1] + A[2] + A[3]) != 0;
-      left_ec = (L[0] + L[1] + L[2] + L[3]) != 0;
+      above_ec = !!*(uint32_t *)A;
+      left_ec  = !!*(uint32_t *)L;
       band_translate = vp9_coefband_trans_8x8plus;
       break;
     }
     case TX_32X32:
       scan = vp9_default_scan_32x32;
-      above_ec = (A[0] + A[1] + A[2] + A[3] + A[4] + A[5] + A[6] + A[7]) != 0;
-      left_ec = (L[0] + L[1] + L[2] + L[3] + L[4] + L[5] + L[6] + L[7]) != 0;
+      above_ec = !!*(uint64_t *)A;
+      left_ec  = !!*(uint64_t *)L;
       band_translate = vp9_coefband_trans_8x8plus;
       break;
   }
@@ -157,7 +150,7 @@ static int decode_coefs(VP9_COMMON *cm, const MACROBLOCKD *xd,
       pt = get_coef_context(nb, token_cache, c);
     band = get_coef_band(band_translate, c);
     prob = coef_probs[band][pt];
-    counts->eob_branch[txfm_size][type][ref][band][pt]++;
+    counts->eob_branch[tx_size][type][ref][band][pt]++;
     if (!vp9_read(r, prob[EOB_CONTEXT_NODE]))
       break;
 
@@ -276,7 +269,7 @@ static void decode_block(int plane, int block,
   const int mod = bw - ss_tx_size - pd->subsampling_x;
   const int aoff = (off & ((1 << mod) - 1)) << ss_tx_size;
   const int loff = (off >> mod) << ss_tx_size;
-
+  const int tx_size_in_blocks = 1 << ss_tx_size;
   ENTROPY_CONTEXT *A = pd->above_context + aoff;
   ENTROPY_CONTEXT *L = pd->left_context + loff;
   const int eob = decode_coefs(&arg->pbi->common, xd, arg->r, block,
@@ -285,10 +278,11 @@ static void decode_block(int plane, int block,
                                ss_tx_size, pd->dequant, A, L);
 
   if (xd->mb_to_right_edge < 0 || xd->mb_to_bottom_edge < 0) {
-    set_contexts_on_border(xd, bsize, plane, ss_tx_size, eob, aoff, loff, A, L);
+    set_contexts_on_border(xd, bsize, plane, tx_size_in_blocks, eob, aoff, loff,
+                           A, L);
   } else {
     int pt;
-    for (pt = 0; pt < (1 << ss_tx_size); pt++)
+    for (pt = 0; pt < tx_size_in_blocks; pt++)
       A[pt] = L[pt] = eob > 0;
   }
   pd->eobs[block] = eob;
diff --git a/libvpx/vp9/decoder/vp9_detokenize.h b/libvpx/vp9/decoder/vp9_detokenize.h
index d46b596..f98fe8d 100644
--- a/libvpx/vp9/decoder/vp9_detokenize.h
+++ b/libvpx/vp9/decoder/vp9_detokenize.h
@@ -13,6 +13,7 @@
 #define VP9_DECODER_VP9_DETOKENIZE_H_
 
 #include "vp9/decoder/vp9_onyxd_int.h"
+#include "vp9/decoder/vp9_dboolhuff.h"
 
 int vp9_decode_tokens(VP9D_COMP* pbi, vp9_reader *r, BLOCK_SIZE_TYPE bsize);
 
diff --git a/libvpx/vp9/decoder/vp9_idct_blk.c b/libvpx/vp9/decoder/vp9_idct_blk.c
index 0217919..395e636 100644
--- a/libvpx/vp9/decoder/vp9_idct_blk.c
+++ b/libvpx/vp9/decoder/vp9_idct_blk.c
@@ -93,15 +93,11 @@ void vp9_idct_add_8x8_c(int16_t *input, uint8_t *dest, int stride, int eob) {
   if (eob) {
     if (eob == 1) {
       // DC only DCT coefficient
-      int16_t in = input[0];
-      int16_t out;
-
-      // Note: the idct1 will need to be modified accordingly whenever
-      // vp9_short_idct8x8_c() is modified.
-      vp9_short_idct1_8x8_c(&in, &out);
+      vp9_short_idct8x8_1_add(input, dest, stride);
       input[0] = 0;
-
-      vp9_add_constant_residual_8x8(out, dest, stride);
+    } else if (eob <= 10) {
+      vp9_short_idct10_8x8_add(input, dest, stride);
+      vpx_memset(input, 0, 128);
     } else {
       vp9_short_idct8x8_add(input, dest, stride);
       vpx_memset(input, 0, 128);
@@ -127,14 +123,11 @@ void vp9_idct_add_16x16_c(int16_t *input, uint8_t *dest, int stride, int eob) {
   if (eob) {
     if (eob == 1) {
       /* DC only DCT coefficient. */
-      int16_t in = input[0];
-      int16_t out;
-      /* Note: the idct1 will need to be modified accordingly whenever
-       * vp9_short_idct16x16() is modified. */
-      vp9_short_idct1_16x16_c(&in, &out);
+      vp9_short_idct16x16_1_add(input, dest, stride);
       input[0] = 0;
-
-      vp9_add_constant_residual_16x16(out, dest, stride);
+    } else if (eob <= 10) {
+      vp9_short_idct10_16x16_add(input, dest, stride);
+      vpx_memset(input, 0, 512);
     } else {
       vp9_short_idct16x16_add(input, dest, stride);
       vpx_memset(input, 0, 512);
diff --git a/libvpx/vp9/decoder/vp9_onyxd_if.c b/libvpx/vp9/decoder/vp9_onyxd_if.c
index cb72920..5a01dd7 100644
--- a/libvpx/vp9/decoder/vp9_onyxd_if.c
+++ b/libvpx/vp9/decoder/vp9_onyxd_if.c
@@ -8,9 +8,9 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-
-#include <stdio.h>
 #include <assert.h>
+#include <limits.h>
+#include <stdio.h>
 
 #include "vp9/common/vp9_onyxc_int.h"
 #if CONFIG_POSTPROC
@@ -114,7 +114,7 @@ VP9D_PTR vp9_create_decompressor(VP9D_CONFIG *oxcf) {
   if (!pbi)
     return NULL;
 
-  vpx_memset(pbi, 0, sizeof(VP9D_COMP));
+  vp9_zero(*pbi);
 
   if (setjmp(pbi->common.error.jmp)) {
     pbi->common.error.setjmp = 0;
@@ -141,6 +141,16 @@ VP9D_PTR vp9_create_decompressor(VP9D_CONFIG *oxcf) {
   pbi->common.error.setjmp = 0;
   pbi->decoded_key_frame = 0;
 
+  if (pbi->oxcf.max_threads > 1) {
+    vp9_worker_init(&pbi->lf_worker);
+    pbi->lf_worker.data1 = vpx_malloc(sizeof(LFWorkerData));
+    pbi->lf_worker.hook = (VP9WorkerHook)vp9_loop_filter_worker;
+    if (pbi->lf_worker.data1 == NULL || !vp9_worker_reset(&pbi->lf_worker)) {
+      vp9_remove_decompressor(pbi);
+      return NULL;
+    }
+  }
+
   return pbi;
 }
 
@@ -154,6 +164,8 @@ void vp9_remove_decompressor(VP9D_PTR ptr) {
     vpx_free(pbi->common.last_frame_seg_map);
 
   vp9_remove_common(&pbi->common);
+  vp9_worker_end(&pbi->lf_worker);
+  vpx_free(pbi->lf_worker.data1);
   vpx_free(pbi);
 }
 
diff --git a/libvpx/vp9/decoder/vp9_onyxd_int.h b/libvpx/vp9/decoder/vp9_onyxd_int.h
index 4760066..a051971 100644
--- a/libvpx/vp9/decoder/vp9_onyxd_int.h
+++ b/libvpx/vp9/decoder/vp9_onyxd_int.h
@@ -14,10 +14,8 @@
 #include "./vpx_config.h"
 
 #include "vp9/common/vp9_onyxc_int.h"
-
-#include "vp9/decoder/vp9_idct_blk.h"
 #include "vp9/decoder/vp9_onyxd.h"
-#include "vp9/decoder/vp9_treereader.h"
+#include "vp9/decoder/vp9_thread.h"
 
 typedef struct VP9Decompressor {
   DECLARE_ALIGNED(16, MACROBLOCKD, mb);
@@ -40,6 +38,7 @@ typedef struct VP9Decompressor {
   int initial_height;
 
   int do_loopfilter_inline;  // apply loopfilter to available rows immediately
+  VP9Worker lf_worker;
 } VP9D_COMP;
 
 #endif  // VP9_DECODER_VP9_TREEREADER_H_
diff --git a/libvpx/vp9/decoder/vp9_thread.c b/libvpx/vp9/decoder/vp9_thread.c
new file mode 100644
index 0000000..dc3b681
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_thread.c
@@ -0,0 +1,248 @@
+// Copyright 2013 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING 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.
+// -----------------------------------------------------------------------------
+//
+// Multi-threaded worker
+//
+// Original source:
+//  http://git.chromium.org/webm/libwebp.git
+//  100644 blob eff8f2a8c20095aade3c292b0e9292dac6cb3587  src/utils/thread.c
+
+
+#include <assert.h>
+#include <string.h>   // for memset()
+#include "./vp9_thread.h"
+
+#if defined(__cplusplus) || defined(c_plusplus)
+extern "C" {
+#endif
+
+#if CONFIG_MULTITHREAD
+
+#if defined(_WIN32)
+
+//------------------------------------------------------------------------------
+// simplistic pthread emulation layer
+
+#include <process.h>
+
+// _beginthreadex requires __stdcall
+#define THREADFN unsigned int __stdcall
+#define THREAD_RETURN(val) (unsigned int)((DWORD_PTR)val)
+
+static int pthread_create(pthread_t* const thread, const void* attr,
+                          unsigned int (__stdcall *start)(void*), void* arg) {
+  (void)attr;
+  *thread = (pthread_t)_beginthreadex(NULL,   /* void *security */
+                                      0,      /* unsigned stack_size */
+                                      start,
+                                      arg,
+                                      0,      /* unsigned initflag */
+                                      NULL);  /* unsigned *thrdaddr */
+  if (*thread == NULL) return 1;
+  SetThreadPriority(*thread, THREAD_PRIORITY_ABOVE_NORMAL);
+  return 0;
+}
+
+static int pthread_join(pthread_t thread, void** value_ptr) {
+  (void)value_ptr;
+  return (WaitForSingleObject(thread, INFINITE) != WAIT_OBJECT_0 ||
+          CloseHandle(thread) == 0);
+}
+
+// Mutex
+static int pthread_mutex_init(pthread_mutex_t* const mutex, void* mutexattr) {
+  (void)mutexattr;
+  InitializeCriticalSection(mutex);
+  return 0;
+}
+
+static int pthread_mutex_lock(pthread_mutex_t* const mutex) {
+  EnterCriticalSection(mutex);
+  return 0;
+}
+
+static int pthread_mutex_unlock(pthread_mutex_t* const mutex) {
+  LeaveCriticalSection(mutex);
+  return 0;
+}
+
+static int pthread_mutex_destroy(pthread_mutex_t* const mutex) {
+  DeleteCriticalSection(mutex);
+  return 0;
+}
+
+// Condition
+static int pthread_cond_destroy(pthread_cond_t* const condition) {
+  int ok = 1;
+  ok &= (CloseHandle(condition->waiting_sem_) != 0);
+  ok &= (CloseHandle(condition->received_sem_) != 0);
+  ok &= (CloseHandle(condition->signal_event_) != 0);
+  return !ok;
+}
+
+static int pthread_cond_init(pthread_cond_t* const condition, void* cond_attr) {
+  (void)cond_attr;
+  condition->waiting_sem_ = CreateSemaphore(NULL, 0, 1, NULL);
+  condition->received_sem_ = CreateSemaphore(NULL, 0, 1, NULL);
+  condition->signal_event_ = CreateEvent(NULL, FALSE, FALSE, NULL);
+  if (condition->waiting_sem_ == NULL ||
+      condition->received_sem_ == NULL ||
+      condition->signal_event_ == NULL) {
+    pthread_cond_destroy(condition);
+    return 1;
+  }
+  return 0;
+}
+
+static int pthread_cond_signal(pthread_cond_t* const condition) {
+  int ok = 1;
+  if (WaitForSingleObject(condition->waiting_sem_, 0) == WAIT_OBJECT_0) {
+    // a thread is waiting in pthread_cond_wait: allow it to be notified
+    ok = SetEvent(condition->signal_event_);
+    // wait until the event is consumed so the signaler cannot consume
+    // the event via its own pthread_cond_wait.
+    ok &= (WaitForSingleObject(condition->received_sem_, INFINITE) !=
+           WAIT_OBJECT_0);
+  }
+  return !ok;
+}
+
+static int pthread_cond_wait(pthread_cond_t* const condition,
+                             pthread_mutex_t* const mutex) {
+  int ok;
+  // note that there is a consumer available so the signal isn't dropped in
+  // pthread_cond_signal
+  if (!ReleaseSemaphore(condition->waiting_sem_, 1, NULL))
+    return 1;
+  // now unlock the mutex so pthread_cond_signal may be issued
+  pthread_mutex_unlock(mutex);
+  ok = (WaitForSingleObject(condition->signal_event_, INFINITE) ==
+        WAIT_OBJECT_0);
+  ok &= ReleaseSemaphore(condition->received_sem_, 1, NULL);
+  pthread_mutex_lock(mutex);
+  return !ok;
+}
+
+#else  // _WIN32
+# define THREADFN void*
+# define THREAD_RETURN(val) val
+#endif
+
+//------------------------------------------------------------------------------
+
+static THREADFN thread_loop(void *ptr) {    // thread loop
+  VP9Worker* const worker = (VP9Worker*)ptr;
+  int done = 0;
+  while (!done) {
+    pthread_mutex_lock(&worker->mutex_);
+    while (worker->status_ == OK) {   // wait in idling mode
+      pthread_cond_wait(&worker->condition_, &worker->mutex_);
+    }
+    if (worker->status_ == WORK) {
+      if (worker->hook) {
+        worker->had_error |= !worker->hook(worker->data1, worker->data2);
+      }
+      worker->status_ = OK;
+    } else if (worker->status_ == NOT_OK) {   // finish the worker
+      done = 1;
+    }
+    // signal to the main thread that we're done (for Sync())
+    pthread_cond_signal(&worker->condition_);
+    pthread_mutex_unlock(&worker->mutex_);
+  }
+  return THREAD_RETURN(NULL);    // Thread is finished
+}
+
+// main thread state control
+static void change_state(VP9Worker* const worker,
+                         VP9WorkerStatus new_status) {
+  // no-op when attempting to change state on a thread that didn't come up
+  if (worker->status_ < OK) return;
+
+  pthread_mutex_lock(&worker->mutex_);
+  // wait for the worker to finish
+  while (worker->status_ != OK) {
+    pthread_cond_wait(&worker->condition_, &worker->mutex_);
+  }
+  // assign new status and release the working thread if needed
+  if (new_status != OK) {
+    worker->status_ = new_status;
+    pthread_cond_signal(&worker->condition_);
+  }
+  pthread_mutex_unlock(&worker->mutex_);
+}
+
+#endif
+
+//------------------------------------------------------------------------------
+
+void vp9_worker_init(VP9Worker* const worker) {
+  memset(worker, 0, sizeof(*worker));
+  worker->status_ = NOT_OK;
+}
+
+int vp9_worker_sync(VP9Worker* const worker) {
+#if CONFIG_MULTITHREAD
+  change_state(worker, OK);
+#endif
+  assert(worker->status_ <= OK);
+  return !worker->had_error;
+}
+
+int vp9_worker_reset(VP9Worker* const worker) {
+  int ok = 1;
+  worker->had_error = 0;
+  if (worker->status_ < OK) {
+#if CONFIG_MULTITHREAD
+    if (pthread_mutex_init(&worker->mutex_, NULL) ||
+        pthread_cond_init(&worker->condition_, NULL)) {
+      return 0;
+    }
+    pthread_mutex_lock(&worker->mutex_);
+    ok = !pthread_create(&worker->thread_, NULL, thread_loop, worker);
+    if (ok) worker->status_ = OK;
+    pthread_mutex_unlock(&worker->mutex_);
+#else
+    worker->status_ = OK;
+#endif
+  } else if (worker->status_ > OK) {
+    ok = vp9_worker_sync(worker);
+  }
+  assert(!ok || (worker->status_ == OK));
+  return ok;
+}
+
+void vp9_worker_launch(VP9Worker* const worker) {
+#if CONFIG_MULTITHREAD
+  change_state(worker, WORK);
+#else
+  if (worker->hook)
+    worker->had_error |= !worker->hook(worker->data1, worker->data2);
+#endif
+}
+
+void vp9_worker_end(VP9Worker* const worker) {
+  if (worker->status_ >= OK) {
+#if CONFIG_MULTITHREAD
+    change_state(worker, NOT_OK);
+    pthread_join(worker->thread_, NULL);
+    pthread_mutex_destroy(&worker->mutex_);
+    pthread_cond_destroy(&worker->condition_);
+#else
+    worker->status_ = NOT_OK;
+#endif
+  }
+  assert(worker->status_ == NOT_OK);
+}
+
+//------------------------------------------------------------------------------
+
+#if defined(__cplusplus) || defined(c_plusplus)
+}    // extern "C"
+#endif
diff --git a/libvpx/vp9/decoder/vp9_thread.h b/libvpx/vp9/decoder/vp9_thread.h
new file mode 100644
index 0000000..a8f7e04
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_thread.h
@@ -0,0 +1,93 @@
+// Copyright 2013 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING 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.
+// -----------------------------------------------------------------------------
+//
+// Multi-threaded worker
+//
+// Original source:
+//  http://git.chromium.org/webm/libwebp.git
+//  100644 blob 13a61a4c84194c3374080cbf03d881d3cd6af40d  src/utils/thread.h
+
+
+#ifndef VP9_DECODER_VP9_THREAD_H_
+#define VP9_DECODER_VP9_THREAD_H_
+
+#include "vpx_config.h"
+
+#if defined(__cplusplus) || defined(c_plusplus)
+extern "C" {
+#endif
+
+#if CONFIG_MULTITHREAD
+
+#if defined(_WIN32)
+
+#include <windows.h>
+typedef HANDLE pthread_t;
+typedef CRITICAL_SECTION pthread_mutex_t;
+typedef struct {
+  HANDLE waiting_sem_;
+  HANDLE received_sem_;
+  HANDLE signal_event_;
+} pthread_cond_t;
+
+#else
+
+#include <pthread.h>
+
+#endif    /* _WIN32 */
+#endif    /* CONFIG_MULTITHREAD */
+
+// State of the worker thread object
+typedef enum {
+  NOT_OK = 0,   // object is unusable
+  OK,           // ready to work
+  WORK          // busy finishing the current task
+} VP9WorkerStatus;
+
+// Function to be called by the worker thread. Takes two opaque pointers as
+// arguments (data1 and data2), and should return false in case of error.
+typedef int (*VP9WorkerHook)(void*, void*);
+
+// Synchronize object used to launch job in the worker thread
+typedef struct {
+#if CONFIG_MULTITHREAD
+  pthread_mutex_t mutex_;
+  pthread_cond_t  condition_;
+  pthread_t       thread_;
+#endif
+  VP9WorkerStatus status_;
+  VP9WorkerHook hook;     // hook to call
+  void* data1;            // first argument passed to 'hook'
+  void* data2;            // second argument passed to 'hook'
+  int had_error;          // return value of the last call to 'hook'
+} VP9Worker;
+
+// Must be called first, before any other method.
+void vp9_worker_init(VP9Worker* const worker);
+// Must be called to initialize the object and spawn the thread. Re-entrant.
+// Will potentially launch the thread. Returns false in case of error.
+int vp9_worker_reset(VP9Worker* const worker);
+// Makes sure the previous work is finished. Returns true if worker->had_error
+// was not set and no error condition was triggered by the working thread.
+int vp9_worker_sync(VP9Worker* const worker);
+// Triggers the thread to call hook() with data1 and data2 argument. These
+// hook/data1/data2 can be changed at any time before calling this function,
+// but not be changed afterward until the next call to vp9_worker_sync().
+void vp9_worker_launch(VP9Worker* const worker);
+// Kill the thread and terminate the object. To use the object again, one
+// must call vp9_worker_reset() again.
+void vp9_worker_end(VP9Worker* const worker);
+
+//------------------------------------------------------------------------------
+
+#if defined(__cplusplus) || defined(c_plusplus)
+}    // extern "C"
+#endif
+
+#endif  /* VP9_DECODER_VP9_THREAD_H_ */
diff --git a/libvpx/vp9/decoder/vp9_treereader.h b/libvpx/vp9/decoder/vp9_treereader.h
index 4535688..710cc4c 100644
--- a/libvpx/vp9/decoder/vp9_treereader.h
+++ b/libvpx/vp9/decoder/vp9_treereader.h
@@ -15,7 +15,6 @@
 #include "vp9/common/vp9_treecoder.h"
 #include "vp9/decoder/vp9_dboolhuff.h"
 
-#define vp9_read_prob(r) ((vp9_prob)vp9_read_literal(r, 8))
 #define vp9_read_and_apply_sign(r, value) (vp9_read_bit(r) ? -(value) : (value))
 
 // Intent of tree data structure is to make decoding trivial.