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

14 files changed, 1111 insertions, 1150 deletions

diff --git a/libvpx/vp9/decoder/arm/neon/vp9_add_constant_residual_neon.asm b/libvpx/vp9/decoder/arm/neon/vp9_add_constant_residual_neon.asm
new file mode 100644
index 0000000..174e747
--- /dev/null
+++ b/libvpx/vp9/decoder/arm/neon/vp9_add_constant_residual_neon.asm
@@ -0,0 +1,230 @@
+;
+;   Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+;
+;   Use of this source code is governed by a BSD-style license
+;   that can be found in the LICENSE file in the root of the source
+;   tree. An additional intellectual property rights grant can be found
+;   in the file PATENTS.  All contributing project authors may
+;   be found in the AUTHORS file in the root of the source tree.
+;
+
+    EXPORT |vp9_add_constant_residual_8x8_neon|
+    EXPORT |vp9_add_constant_residual_16x16_neon|
+    EXPORT |vp9_add_constant_residual_32x32_neon|
+    ARM
+
+    AREA ||.text||, CODE, READONLY, ALIGN=2
+
+    MACRO
+    LD_16x8 $src, $stride
+    vld1.8              {q8},       [$src],     $stride
+    vld1.8              {q9},       [$src],     $stride
+    vld1.8              {q10},      [$src],     $stride
+    vld1.8              {q11},      [$src],     $stride
+    vld1.8              {q12},      [$src],     $stride
+    vld1.8              {q13},      [$src],     $stride
+    vld1.8              {q14},      [$src],     $stride
+    vld1.8              {q15},      [$src],     $stride
+    MEND
+
+    MACRO
+    ADD_DIFF_16x8 $diff
+    vqadd.u8            q8,         q8,         $diff
+    vqadd.u8            q9,         q9,         $diff
+    vqadd.u8            q10,        q10,        $diff
+    vqadd.u8            q11,        q11,        $diff
+    vqadd.u8            q12,        q12,        $diff
+    vqadd.u8            q13,        q13,        $diff
+    vqadd.u8            q14,        q14,        $diff
+    vqadd.u8            q15,        q15,        $diff
+    MEND
+
+    MACRO
+    SUB_DIFF_16x8 $diff
+    vqsub.u8            q8,         q8,         $diff
+    vqsub.u8            q9,         q9,         $diff
+    vqsub.u8            q10,        q10,        $diff
+    vqsub.u8            q11,        q11,        $diff
+    vqsub.u8            q12,        q12,        $diff
+    vqsub.u8            q13,        q13,        $diff
+    vqsub.u8            q14,        q14,        $diff
+    vqsub.u8            q15,        q15,        $diff
+    MEND
+
+    MACRO
+    ST_16x8 $dst, $stride
+    vst1.8              {q8},       [$dst],     $stride
+    vst1.8              {q9},       [$dst],     $stride
+    vst1.8              {q10},      [$dst],     $stride
+    vst1.8              {q11},      [$dst],     $stride
+    vst1.8              {q12},      [$dst],     $stride
+    vst1.8              {q13},      [$dst],     $stride
+    vst1.8              {q14},      [$dst],     $stride
+    vst1.8              {q15},      [$dst],     $stride
+    MEND
+
+; void add_constant_residual(const int16_t diff, uint8_t *dest, int stride,
+;                             int width, int height) {
+;  int r, c;
+;
+;  for (r = 0; r < height; r++) {
+;    for (c = 0; c < width; c++)
+;      dest[c] = clip_pixel(diff + dest[c]);
+;
+;    dest += stride;
+;  }
+;}
+;void vp9_add_constant_residual_8x8_c(const int16_t diff, uint8_t *dest,
+;                                     int stride) {
+;  add_constant_residual(diff, dest, stride, 8, 8);
+;}
+;       r0      : const int16_t diff
+;       r1      : const uint8_t *dest
+;       r2      : int stride
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+|vp9_add_constant_residual_8x8_neon| PROC
+    mov                 r3,         r1                      ; r3: save dest to r3
+    vld1.8              {d0},       [r1],       r2
+    vld1.8              {d1},       [r1],       r2
+    vld1.8              {d2},       [r1],       r2
+    vld1.8              {d3},       [r1],       r2
+    vld1.8              {d4},       [r1],       r2
+    vld1.8              {d5},       [r1],       r2
+    vld1.8              {d6},       [r1],       r2
+    vld1.8              {d7},       [r1],       r2
+    cmp                 r0,         #0
+    bge                 DIFF_POSITIVE_8x8
+
+DIFF_NEGATIVE_8x8                                           ; diff < 0
+    neg                 r0,         r0
+    usat                r0,         #8,         r0
+    vdup.u8             q8,         r0
+
+    vqsub.u8            q0,         q0,         q8
+    vqsub.u8            q1,         q1,         q8
+    vqsub.u8            q2,         q2,         q8
+    vqsub.u8            q3,         q3,         q8
+    b                   DIFF_SAVE_8x8
+
+DIFF_POSITIVE_8x8                                           ; diff >= 0
+    usat                r0,         #8,         r0
+    vdup.u8             q8,         r0
+
+    vqadd.u8            q0,         q0,         q8
+    vqadd.u8            q1,         q1,         q8
+    vqadd.u8            q2,         q2,         q8
+    vqadd.u8            q3,         q3,         q8
+
+DIFF_SAVE_8x8
+    vst1.8              {d0},       [r3],       r2
+    vst1.8              {d1},       [r3],       r2
+    vst1.8              {d2},       [r3],       r2
+    vst1.8              {d3},       [r3],       r2
+    vst1.8              {d4},       [r3],       r2
+    vst1.8              {d5},       [r3],       r2
+    vst1.8              {d6},       [r3],       r2
+    vst1.8              {d7},       [r3],       r2
+
+    bx                  lr
+    ENDP
+
+;void vp9_add_constant_residual_16x16_c(const int16_t diff, uint8_t *dest,
+;                                       int stride) {
+;  add_constant_residual(diff, dest, stride, 16, 16);
+;}
+;       r0      : const int16_t diff
+;       r1      : const uint8_t *dest
+;       r2      : int stride
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+|vp9_add_constant_residual_16x16_neon| PROC
+    mov                 r3,         r1
+    LD_16x8             r1,         r2
+    cmp                 r0,         #0
+    bge                 DIFF_POSITIVE_16x16
+
+|DIFF_NEGATIVE_16x16|
+    neg                 r0,         r0
+    usat                r0,         #8,         r0
+    vdup.u8             q0,         r0
+
+    SUB_DIFF_16x8       q0
+    ST_16x8             r3,         r2
+    LD_16x8             r1,         r2
+    SUB_DIFF_16x8       q0
+    b                   DIFF_SAVE_16x16
+
+|DIFF_POSITIVE_16x16|
+    usat                r0,         #8,         r0
+    vdup.u8             q0,         r0
+
+    ADD_DIFF_16x8       q0
+    ST_16x8             r3,         r2
+    LD_16x8             r1,         r2
+    ADD_DIFF_16x8       q0
+
+|DIFF_SAVE_16x16|
+    ST_16x8             r3,         r2
+    bx                  lr
+    ENDP
+
+;void vp9_add_constant_residual_32x32_c(const int16_t diff, uint8_t *dest,
+;                                       int stride) {
+;  add_constant_residual(diff, dest, stride, 32, 32);
+;}
+;       r0      : const int16_t diff
+;       r1      : const uint8_t *dest
+;       r2      : int stride
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+|vp9_add_constant_residual_32x32_neon| PROC
+    push                {r4,lr}
+    pld                 [r1]
+    mov                 r3,         r1
+    add                 r4,         r1,         #16         ; r4 dest + 16 for second loop
+    cmp                 r0,         #0
+    bge                 DIFF_POSITIVE_32x32
+
+|DIFF_NEGATIVE_32x32|
+    neg                 r0,         r0
+    usat                r0,         #8,         r0
+    vdup.u8             q0,         r0
+    mov                 r0,         #4
+
+|DIFF_NEGATIVE_32x32_LOOP|
+    sub                 r0,         #1
+    LD_16x8             r1,         r2
+    SUB_DIFF_16x8       q0
+    ST_16x8             r3,         r2
+
+    LD_16x8             r1,         r2
+    SUB_DIFF_16x8       q0
+    ST_16x8             r3,         r2
+    cmp                 r0,         #2
+    moveq               r1,         r4
+    moveq               r3,         r4
+    cmp                 r0,         #0
+    bne                 DIFF_NEGATIVE_32x32_LOOP
+    pop                 {r4,pc}
+
+|DIFF_POSITIVE_32x32|
+    usat                r0,         #8,         r0
+    vdup.u8             q0,         r0
+    mov                 r0,         #4
+
+|DIFF_POSITIVE_32x32_LOOP|
+    sub                 r0,         #1
+    LD_16x8             r1,         r2
+    ADD_DIFF_16x8       q0
+    ST_16x8             r3,         r2
+
+    LD_16x8             r1,         r2
+    ADD_DIFF_16x8       q0
+    ST_16x8             r3,         r2
+    cmp                 r0,         #2
+    moveq               r1,         r4
+    moveq               r3,         r4
+    cmp                 r0,         #0
+    bne                 DIFF_POSITIVE_32x32_LOOP
+    pop                 {r4,pc}
+    ENDP
+
+    END
diff --git a/libvpx/vp9/decoder/vp9_dboolhuff.c b/libvpx/vp9/decoder/vp9_dboolhuff.c
index df77d65..31b1ae2 100644
--- a/libvpx/vp9/decoder/vp9_dboolhuff.c
+++ b/libvpx/vp9/decoder/vp9_dboolhuff.c
@@ -13,6 +13,12 @@
 
 #include "vp9/decoder/vp9_dboolhuff.h"
 
+// This is meant to be a large, positive constant that can still be efficiently
+// loaded as an immediate (on platforms like ARM, for example).
+// Even relatively modest values like 100 would work fine.
+#define VP9_LOTS_OF_BITS 0x40000000
+
+
 int vp9_reader_init(vp9_reader *r, const uint8_t *buffer, size_t size) {
   int marker_bit;
 
@@ -67,3 +73,20 @@ const uint8_t *vp9_reader_find_end(vp9_reader *r) {
   return r->buffer;
 }
 
+int vp9_reader_has_error(vp9_reader *r) {
+  // Check if we have reached the end of the buffer.
+  //
+  // Variable 'count' stores the number of bits in the 'value' buffer, minus
+  // 8. The top byte is part of the algorithm, and the remainder is buffered
+  // to be shifted into it. So if count == 8, the top 16 bits of 'value' are
+  // occupied, 8 for the algorithm and 8 in the buffer.
+  //
+  // When reading a byte from the user's buffer, count is filled with 8 and
+  // one byte is filled into the value buffer. When we reach the end of the
+  // data, count is additionally filled with VP9_LOTS_OF_BITS. So when
+  // count == VP9_LOTS_OF_BITS - 1, the user's data has been exhausted.
+  //
+  // 1 if we have tried to decode bits after the end of stream was encountered.
+  // 0 No error.
+  return r->count > VP9_BD_VALUE_SIZE && r->count < VP9_LOTS_OF_BITS;
+}
diff --git a/libvpx/vp9/decoder/vp9_dboolhuff.h b/libvpx/vp9/decoder/vp9_dboolhuff.h
index b50aa35..c46dd73 100644
--- a/libvpx/vp9/decoder/vp9_dboolhuff.h
+++ b/libvpx/vp9/decoder/vp9_dboolhuff.h
@@ -22,11 +22,6 @@ typedef size_t VP9_BD_VALUE;
 
 #define VP9_BD_VALUE_SIZE ((int)sizeof(VP9_BD_VALUE)*CHAR_BIT)
 
-// This is meant to be a large, positive constant that can still be efficiently
-// loaded as an immediate (on platforms like ARM, for example).
-// Even relatively modest values like 100 would work fine.
-#define VP9_LOTS_OF_BITS 0x40000000
-
 typedef struct {
   const uint8_t *buffer_end;
   const uint8_t *buffer;
@@ -93,22 +88,6 @@ static int vp9_read_literal(vp9_reader *br, int bits) {
   return z;
 }
 
-static int vp9_reader_has_error(vp9_reader *r) {
-  // Check if we have reached the end of the buffer.
-  //
-  // Variable 'count' stores the number of bits in the 'value' buffer, minus
-  // 8. The top byte is part of the algorithm, and the remainder is buffered
-  // to be shifted into it. So if count == 8, the top 16 bits of 'value' are
-  // occupied, 8 for the algorithm and 8 in the buffer.
-  //
-  // When reading a byte from the user's buffer, count is filled with 8 and
-  // one byte is filled into the value buffer. When we reach the end of the
-  // data, count is additionally filled with VP9_LOTS_OF_BITS. So when
-  // count == VP9_LOTS_OF_BITS - 1, the user's data has been exhausted.
-  //
-  // 1 if we have tried to decode bits after the end of stream was encountered.
-  // 0 No error.
-  return r->count > VP9_BD_VALUE_SIZE && r->count < VP9_LOTS_OF_BITS;
-}
+int vp9_reader_has_error(vp9_reader *r);
 
 #endif  // VP9_DECODER_VP9_DBOOLHUFF_H_
diff --git a/libvpx/vp9/decoder/vp9_decodemv.c b/libvpx/vp9/decoder/vp9_decodemv.c
index b3d41be..6f0044a 100644
--- a/libvpx/vp9/decoder/vp9_decodemv.c
+++ b/libvpx/vp9/decoder/vp9_decodemv.c
@@ -8,151 +8,188 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
+#include <assert.h>
 
-#include "vp9/decoder/vp9_treereader.h"
-#include "vp9/common/vp9_entropymv.h"
+#include "vp9/common/vp9_common.h"
+#include "vp9/common/vp9_entropy.h"
 #include "vp9/common/vp9_entropymode.h"
-#include "vp9/common/vp9_reconinter.h"
-#include "vp9/decoder/vp9_onyxd_int.h"
+#include "vp9/common/vp9_entropymv.h"
 #include "vp9/common/vp9_findnearmv.h"
-#include "vp9/common/vp9_common.h"
-#include "vp9/common/vp9_seg_common.h"
+#include "vp9/common/vp9_mvref_common.h"
 #include "vp9/common/vp9_pred_common.h"
-#include "vp9/common/vp9_entropy.h"
+#include "vp9/common/vp9_reconinter.h"
+#include "vp9/common/vp9_seg_common.h"
+
 #include "vp9/decoder/vp9_decodemv.h"
 #include "vp9/decoder/vp9_decodframe.h"
-#include "vp9/common/vp9_mvref_common.h"
-#if CONFIG_DEBUG
-#include <assert.h>
-#endif
+#include "vp9/decoder/vp9_onyxd_int.h"
+#include "vp9/decoder/vp9_dsubexp.h"
+#include "vp9/decoder/vp9_treereader.h"
+
+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);
+}
 
-// #define DEBUG_DEC_MV
-#ifdef DEBUG_DEC_MV
-int dec_mvcount = 0;
-#endif
+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);
+}
 
-// #define DEC_DEBUG
-#ifdef DEC_DEBUG
-extern int dec_debug;
-#endif
+static int read_segment_id(vp9_reader *r, const struct segmentation *seg) {
+  return treed_read(r, vp9_segment_tree, seg->tree_probs);
+}
 
-static MB_PREDICTION_MODE read_intra_mode(vp9_reader *r, const vp9_prob *p) {
-  MB_PREDICTION_MODE m = treed_read(r, vp9_intra_mode_tree, p);
-  return m;
+static TX_SIZE read_selected_tx_size(VP9_COMMON *cm, MACROBLOCKD *xd,
+                                     BLOCK_SIZE_TYPE bsize, vp9_reader *r) {
+  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) {
+    tx_size += vp9_read(r, tx_probs[1]);
+    if (tx_size != TX_8X8 && bsize >= BLOCK_SIZE_SB32X32)
+      tx_size += vp9_read(r, tx_probs[2]);
+  }
+
+  update_tx_counts(bsize, context, tx_size, &cm->counts.tx);
+  return tx_size;
 }
 
-static int read_mb_segid(vp9_reader *r, MACROBLOCKD *xd) {
-  return treed_read(r, vp9_segment_tree, xd->mb_segment_tree_probs);
+static TX_SIZE read_tx_size(VP9D_COMP *pbi, TX_MODE tx_mode,
+                            BLOCK_SIZE_TYPE bsize, int select_cond,
+                            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)
+    return read_selected_tx_size(cm, xd, bsize, r);
+  else if (tx_mode >= ALLOW_32X32 && bsize >= BLOCK_SIZE_SB32X32)
+    return TX_32X32;
+  else if (tx_mode >= ALLOW_16X16 && bsize >= BLOCK_SIZE_MB16X16)
+    return TX_16X16;
+  else if (tx_mode >= ALLOW_8X8 && bsize >= BLOCK_SIZE_SB8X8)
+    return TX_8X8;
+  else
+    return TX_4X4;
 }
 
-static void set_segment_id(VP9_COMMON *cm, MB_MODE_INFO *mbmi,
+static void set_segment_id(VP9_COMMON *cm, BLOCK_SIZE_TYPE bsize,
                            int mi_row, int mi_col, int segment_id) {
-  const int mi_index = mi_row * cm->mi_cols + mi_col;
-  const BLOCK_SIZE_TYPE sb_type = mbmi->sb_type;
-  const int bw = 1 << mi_width_log2(sb_type);
-  const int bh = 1 << mi_height_log2(sb_type);
-  const int ymis = MIN(cm->mi_rows - mi_row, bh);
+  const int mi_offset = mi_row * cm->mi_cols + mi_col;
+  const int bw = 1 << mi_width_log2(bsize);
+  const int bh = 1 << mi_height_log2(bsize);
   const int xmis = MIN(cm->mi_cols - mi_col, bw);
+  const int ymis = MIN(cm->mi_rows - mi_row, bh);
   int x, y;
 
-  for (y = 0; y < ymis; y++) {
-    for (x = 0; x < xmis; x++) {
-      const int index = mi_index + (y * cm->mi_cols + x);
-      cm->last_frame_seg_map[index] = segment_id;
-    }
-  }
-}
+  assert(segment_id >= 0 && segment_id < MAX_SEGMENTS);
 
-static TX_SIZE select_txfm_size(VP9_COMMON *cm, MACROBLOCKD *xd,
-                                vp9_reader *r, BLOCK_SIZE_TYPE bsize) {
-  const int context = vp9_get_pred_context(cm, xd, PRED_TX_SIZE);
-  const vp9_prob *tx_probs = vp9_get_pred_probs(cm, xd, PRED_TX_SIZE);
-  TX_SIZE txfm_size = vp9_read(r, tx_probs[0]);
-  if (txfm_size != TX_4X4 && bsize >= BLOCK_SIZE_MB16X16) {
-    txfm_size += vp9_read(r, tx_probs[1]);
-    if (txfm_size != TX_8X8 && bsize >= BLOCK_SIZE_SB32X32)
-      txfm_size += vp9_read(r, tx_probs[2]);
-  }
-  if (bsize >= BLOCK_SIZE_SB32X32) {
-    cm->fc.tx_count_32x32p[context][txfm_size]++;
-  } else if (bsize >= BLOCK_SIZE_MB16X16) {
-    cm->fc.tx_count_16x16p[context][txfm_size]++;
-  } else {
-    cm->fc.tx_count_8x8p[context][txfm_size]++;
-  }
-  return txfm_size;
+  for (y = 0; y < ymis; y++)
+    for (x = 0; x < xmis; x++)
+      cm->last_frame_seg_map[mi_offset + y * cm->mi_cols + x] = segment_id;
 }
 
+static int read_intra_segment_id(VP9D_COMP *pbi, int mi_row, int mi_col,
+                                 vp9_reader *r) {
+  MACROBLOCKD *const xd = &pbi->mb;
+  struct segmentation *const seg = &xd->seg;
+  const BLOCK_SIZE_TYPE bsize = xd->mode_info_context->mbmi.sb_type;
+  int segment_id;
+
+  if (!seg->enabled)
+    return 0;  // Default for disabled segmentation
+
+  if (!seg->update_map)
+    return 0;
 
-static void kfread_modes(VP9D_COMP *pbi, MODE_INFO *m,
-                         int mi_row, int mi_col,
-                         vp9_reader *r) {
+  segment_id = read_segment_id(r, seg);
+  set_segment_id(&pbi->common, bsize, mi_row, mi_col, segment_id);
+  return segment_id;
+}
+
+static int read_inter_segment_id(VP9D_COMP *pbi, int mi_row, int mi_col,
+                                 vp9_reader *r) {
   VP9_COMMON *const cm = &pbi->common;
   MACROBLOCKD *const xd = &pbi->mb;
-  const int mis = cm->mode_info_stride;
+  struct segmentation *const seg = &xd->seg;
+  const BLOCK_SIZE_TYPE bsize = xd->mode_info_context->mbmi.sb_type;
+  int pred_segment_id, segment_id;
 
-  // Read segmentation map if it is being updated explicitly this frame
-  m->mbmi.segment_id = 0;
-  if (xd->segmentation_enabled && xd->update_mb_segmentation_map) {
-    m->mbmi.segment_id = read_mb_segid(r, xd);
-    set_segment_id(cm, &m->mbmi, mi_row, mi_col, m->mbmi.segment_id);
-  }
+  if (!seg->enabled)
+    return 0;  // Default for disabled segmentation
 
-  m->mbmi.mb_skip_coeff = vp9_segfeature_active(xd, m->mbmi.segment_id,
-                                                SEG_LVL_SKIP);
-  if (!m->mbmi.mb_skip_coeff) {
-    m->mbmi.mb_skip_coeff = vp9_read(r, vp9_get_pred_prob(cm, xd, PRED_MBSKIP));
-    cm->fc.mbskip_count[vp9_get_pred_context(cm, xd, PRED_MBSKIP)]
-                       [m->mbmi.mb_skip_coeff]++;
+  pred_segment_id = vp9_get_segment_id(cm, cm->last_frame_seg_map,
+                                       bsize, mi_row, mi_col);
+  if (!seg->update_map)
+    return pred_segment_id;
+
+  if (seg->temporal_update) {
+    const vp9_prob pred_prob = vp9_get_pred_prob_seg_id(xd);
+    const int pred_flag = vp9_read(r, pred_prob);
+    vp9_set_pred_flag_seg_id(cm, bsize, mi_row, mi_col, pred_flag);
+    segment_id = pred_flag ? pred_segment_id
+                           : read_segment_id(r, seg);
+  } else {
+    segment_id = read_segment_id(r, seg);
   }
+  set_segment_id(cm, bsize, mi_row, mi_col, segment_id);
+  return segment_id;
+}
 
-  if (cm->txfm_mode == TX_MODE_SELECT &&
-      m->mbmi.sb_type >= BLOCK_SIZE_SB8X8) {
-    m->mbmi.txfm_size = select_txfm_size(cm, xd, r, m->mbmi.sb_type);
-  } else if (cm->txfm_mode >= ALLOW_32X32 &&
-             m->mbmi.sb_type >= BLOCK_SIZE_SB32X32) {
-    m->mbmi.txfm_size = TX_32X32;
-  } else if (cm->txfm_mode >= ALLOW_16X16 &&
-             m->mbmi.sb_type >= BLOCK_SIZE_MB16X16) {
-    m->mbmi.txfm_size = TX_16X16;
-  } else if (cm->txfm_mode >= ALLOW_8X8 &&
-             m->mbmi.sb_type >= BLOCK_SIZE_SB8X8) {
-    m->mbmi.txfm_size = TX_8X8;
-  } else {
-    m->mbmi.txfm_size = TX_4X4;
+static uint8_t read_skip_coeff(VP9D_COMP *pbi, int segment_id, vp9_reader *r) {
+  VP9_COMMON *const cm = &pbi->common;
+  MACROBLOCKD *const xd = &pbi->mb;
+  int skip_coeff = vp9_segfeature_active(&xd->seg, segment_id, SEG_LVL_SKIP);
+  if (!skip_coeff) {
+    const int ctx = vp9_get_pred_context_mbskip(xd);
+    skip_coeff = vp9_read(r, vp9_get_pred_prob_mbskip(cm, xd));
+    cm->counts.mbskip[ctx][skip_coeff]++;
   }
+  return skip_coeff;
+}
 
-  // luma mode
-  m->mbmi.ref_frame[0] = INTRA_FRAME;
-  if (m->mbmi.sb_type >= BLOCK_SIZE_SB8X8) {
+static void read_intra_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;
+  const BLOCK_SIZE_TYPE bsize = mbmi->sb_type;
+  const int mis = cm->mode_info_stride;
+
+  mbmi->segment_id = read_intra_segment_id(pbi, mi_row, mi_col, r);
+  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;
+
+  if (bsize >= 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;
-    m->mbmi.mode = read_intra_mode(r, cm->kf_y_mode_prob[A][L]);
+    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);
     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 ib = idy * 2 + idx;
-        int k;
+        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) ?
                                       left_block_mode(m, ib) : DC_PRED;
-        m->bmi[ib].as_mode.first =
-            read_intra_mode(r, cm->kf_y_mode_prob[A][L]);
-        for (k = 1; k < bh; ++k)
-          m->bmi[ib + k * 2].as_mode.first = m->bmi[ib].as_mode.first;
-        for (k = 1; k < bw; ++k)
-          m->bmi[ib + k].as_mode.first = m->bmi[ib].as_mode.first;
+        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)
+          m->bmi[ib + 2].as_mode = b_mode;
+        if (bw == 2)
+          m->bmi[ib + 1].as_mode = b_mode;
       }
     }
-    m->mbmi.mode = m->bmi[3].as_mode.first;
+
+    mbmi->mode = m->bmi[3].as_mode;
   }
 
-  m->mbmi.uv_mode = read_intra_mode(r, cm->kf_uv_mode_prob[m->mbmi.mode]);
+  mbmi->uv_mode = read_intra_mode(r, vp9_kf_uv_mode_prob[mbmi->mode]);
 }
 
 static int read_mv_component(vp9_reader *r,
@@ -161,9 +198,10 @@ static int read_mv_component(vp9_reader *r,
   int mag, d, fr, hp;
   const int sign = vp9_read(r, mvcomp->sign);
   const int mv_class = treed_read(r, vp9_mv_class_tree, mvcomp->classes);
+  const int class0 = mv_class == MV_CLASS_0;
 
   // Integer part
-  if (mv_class == MV_CLASS_0) {
+  if (class0) {
     d = treed_read(r, vp9_mv_class0_tree, mvcomp->class0);
   } else {
     int i;
@@ -176,66 +214,77 @@ static int read_mv_component(vp9_reader *r,
 
   // Fractional part
   fr = treed_read(r, vp9_mv_fp_tree,
-                  mv_class == MV_CLASS_0 ? mvcomp->class0_fp[d] : mvcomp->fp);
+                  class0 ? mvcomp->class0_fp[d] : mvcomp->fp);
 
 
   // High precision part (if hp is not used, the default value of the hp is 1)
-  hp = usehp ? vp9_read(r,
-                        mv_class == MV_CLASS_0 ? mvcomp->class0_hp : mvcomp->hp)
+  hp = usehp ? vp9_read(r, class0 ? mvcomp->class0_hp : mvcomp->hp)
              : 1;
 
-  // result
+  // Result
   mag = vp9_get_mv_mag(mv_class, (d << 3) | (fr << 1) | hp) + 1;
   return sign ? -mag : mag;
 }
 
-static void update_nmv(vp9_reader *r, vp9_prob *const p,
-                       const vp9_prob upd_p) {
-  if (vp9_read(r, upd_p)) {
-#ifdef LOW_PRECISION_MV_UPDATE
+static INLINE void read_mv(vp9_reader *r, MV *mv, const MV *ref,
+                           const nmv_context *ctx,
+                           nmv_context_counts *counts, int usehp) {
+  const MV_JOINT_TYPE j = treed_read(r, vp9_mv_joint_tree, ctx->joints);
+  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);
+
+  if (mv_joint_horizontal(j))
+    diff.col = read_mv_component(r, &ctx->comps[1], usehp);
+
+  vp9_inc_mv(&diff, counts);
+
+  mv->row = ref->row + diff.row;
+  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))
     *p = (vp9_read_literal(r, 7) << 1) | 1;
-#else
-    *p = (vp9_read_literal(r, 8));
-#endif
-  }
 }
 
-static void read_nmvprobs(vp9_reader *r, nmv_context *mvctx,
-                          int usehp) {
+static void read_mv_probs(vp9_reader *r, nmv_context *mvc, int usehp) {
   int i, j, k;
 
-#ifdef MV_GROUP_UPDATE
-  if (!vp9_read_bit(r))
-    return;
-#endif
   for (j = 0; j < MV_JOINTS - 1; ++j)
-    update_nmv(r, &mvctx->joints[j], VP9_NMV_UPDATE_PROB);
+    update_mv(r, &mvc->joints[j], VP9_NMV_UPDATE_PROB);
 
   for (i = 0; i < 2; ++i) {
-    update_nmv(r, &mvctx->comps[i].sign, VP9_NMV_UPDATE_PROB);
+    nmv_component *const comp = &mvc->comps[i];
+
+    update_mv(r, &comp->sign, VP9_NMV_UPDATE_PROB);
     for (j = 0; j < MV_CLASSES - 1; ++j)
-      update_nmv(r, &mvctx->comps[i].classes[j], VP9_NMV_UPDATE_PROB);
+      update_mv(r, &comp->classes[j], VP9_NMV_UPDATE_PROB);
 
     for (j = 0; j < CLASS0_SIZE - 1; ++j)
-      update_nmv(r, &mvctx->comps[i].class0[j], VP9_NMV_UPDATE_PROB);
+      update_mv(r, &comp->class0[j], VP9_NMV_UPDATE_PROB);
 
     for (j = 0; j < MV_OFFSET_BITS; ++j)
-      update_nmv(r, &mvctx->comps[i].bits[j], VP9_NMV_UPDATE_PROB);
+      update_mv(r, &comp->bits[j], VP9_NMV_UPDATE_PROB);
   }
 
   for (i = 0; i < 2; ++i) {
+    nmv_component *const comp = &mvc->comps[i];
+
     for (j = 0; j < CLASS0_SIZE; ++j)
       for (k = 0; k < 3; ++k)
-        update_nmv(r, &mvctx->comps[i].class0_fp[j][k], VP9_NMV_UPDATE_PROB);
+        update_mv(r, &comp->class0_fp[j][k], VP9_NMV_UPDATE_PROB);
 
     for (j = 0; j < 3; ++j)
-      update_nmv(r, &mvctx->comps[i].fp[j], VP9_NMV_UPDATE_PROB);
+      update_mv(r, &comp->fp[j], VP9_NMV_UPDATE_PROB);
   }
 
   if (usehp) {
     for (i = 0; i < 2; ++i) {
-      update_nmv(r, &mvctx->comps[i].class0_hp, VP9_NMV_UPDATE_PROB);
-      update_nmv(r, &mvctx->comps[i].hp, VP9_NMV_UPDATE_PROB);
+      update_mv(r, &mvc->comps[i].class0_hp, VP9_NMV_UPDATE_PROB);
+      update_mv(r, &mvc->comps[i].hp, VP9_NMV_UPDATE_PROB);
     }
   }
 }
@@ -245,205 +294,71 @@ static void read_ref_frame(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;
-  const int seg_ref_active = vp9_segfeature_active(xd, segment_id,
-                                                   SEG_LVL_REF_FRAME);
+  FRAME_CONTEXT *const fc = &cm->fc;
+  FRAME_COUNTS *const counts = &cm->counts;
 
-  // Segment reference frame features not available.
-  if (!seg_ref_active) {
+  if (vp9_segfeature_active(&xd->seg, segment_id, SEG_LVL_REF_FRAME)) {
+    ref_frame[0] = vp9_get_segdata(&xd->seg, segment_id, SEG_LVL_REF_FRAME);
+    ref_frame[1] = NONE;
+  } else {
+    const int comp_ctx = vp9_get_pred_context_comp_inter_inter(cm, xd);
     int is_comp;
-    int comp_ctx = vp9_get_pred_context(cm, xd, PRED_COMP_INTER_INTER);
 
     if (cm->comp_pred_mode == HYBRID_PREDICTION) {
-      is_comp = vp9_read(r, cm->fc.comp_inter_prob[comp_ctx]);
-      cm->fc.comp_inter_count[comp_ctx][is_comp]++;
+      is_comp = vp9_read(r, fc->comp_inter_prob[comp_ctx]);
+      counts->comp_inter[comp_ctx][is_comp]++;
     } else {
       is_comp = cm->comp_pred_mode == COMP_PREDICTION_ONLY;
     }
 
     // FIXME(rbultje) I'm pretty sure this breaks segmentation ref frame coding
     if (is_comp) {
-      int b, fix_ref_idx = cm->ref_frame_sign_bias[cm->comp_fixed_ref];
-      int ref_ctx = vp9_get_pred_context(cm, xd, PRED_COMP_REF_P);
-
-      ref_frame[fix_ref_idx]  = cm->comp_fixed_ref;
-      b = vp9_read(r, cm->fc.comp_ref_prob[ref_ctx]);
-      cm->fc.comp_ref_count[ref_ctx][b]++;
+      const int fix_ref_idx = cm->ref_frame_sign_bias[cm->comp_fixed_ref];
+      const int ref_ctx = vp9_get_pred_context_comp_ref_p(cm, xd);
+      const int b = vp9_read(r, fc->comp_ref_prob[ref_ctx]);
+      counts->comp_ref[ref_ctx][b]++;
+      ref_frame[fix_ref_idx] = cm->comp_fixed_ref;
       ref_frame[!fix_ref_idx] = cm->comp_var_ref[b];
     } else {
-      int ref1_ctx = vp9_get_pred_context(cm, xd, PRED_SINGLE_REF_P1);
+      const int ref1_ctx = vp9_get_pred_context_single_ref_p1(xd);
       ref_frame[1] = NONE;
-      if (vp9_read(r, cm->fc.single_ref_prob[ref1_ctx][0])) {
-        int ref2_ctx = vp9_get_pred_context(cm, xd, PRED_SINGLE_REF_P2);
-        int b2 = vp9_read(r, cm->fc.single_ref_prob[ref2_ctx][1]);
-        ref_frame[0] = b2 ? ALTREF_FRAME : GOLDEN_FRAME;
-        cm->fc.single_ref_count[ref1_ctx][0][1]++;
-        cm->fc.single_ref_count[ref2_ctx][1][b2]++;
+      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]++;
       } else {
         ref_frame[0] = LAST_FRAME;
-        cm->fc.single_ref_count[ref1_ctx][0][0]++;
+        counts->single_ref[ref1_ctx][0][0]++;
       }
     }
-  } else {
-    ref_frame[0] = vp9_get_segdata(xd, segment_id, SEG_LVL_REF_FRAME);
-    ref_frame[1] = NONE;
   }
 }
 
-static MB_PREDICTION_MODE read_sb_mv_ref(vp9_reader *r, const vp9_prob *p) {
-  return (MB_PREDICTION_MODE) treed_read(r, vp9_sb_mv_ref_tree, p);
-}
-
-#ifdef VPX_MODE_COUNT
-unsigned int vp9_mv_cont_count[5][4] = {
-  { 0, 0, 0, 0 },
-  { 0, 0, 0, 0 },
-  { 0, 0, 0, 0 },
-  { 0, 0, 0, 0 },
-  { 0, 0, 0, 0 }
-};
-#endif
-
-static void read_switchable_interp_probs(VP9_COMMON* const cm, vp9_reader *r) {
+static void read_switchable_interp_probs(FRAME_CONTEXT *fc, vp9_reader *r) {
   int i, j;
-  for (j = 0; j <= VP9_SWITCHABLE_FILTERS; ++j)
-    for (i = 0; i < VP9_SWITCHABLE_FILTERS - 1; ++i) {
-      if (vp9_read(r, VP9_MODE_UPDATE_PROB)) {
-        cm->fc.switchable_interp_prob[j][i] =
-            // vp9_read_prob(r);
-            vp9_read_prob_diff_update(r, cm->fc.switchable_interp_prob[j][i]);
-      }
-    }
+  for (j = 0; j < VP9_SWITCHABLE_FILTERS + 1; ++j)
+    for (i = 0; i < VP9_SWITCHABLE_FILTERS - 1; ++i)
+      if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+        vp9_diff_update_prob(r, &fc->switchable_interp_prob[j][i]);
 }
 
-static void read_inter_mode_probs(VP9_COMMON *const cm, vp9_reader *r) {
+static void read_inter_mode_probs(FRAME_CONTEXT *fc, vp9_reader *r) {
   int i, j;
   for (i = 0; i < INTER_MODE_CONTEXTS; ++i)
-    for (j = 0; j < VP9_INTER_MODES - 1; ++j) {
-      if (vp9_read(r, VP9_MODE_UPDATE_PROB)) {
-        // cm->fc.inter_mode_probs[i][j] = vp9_read_prob(r);
-        cm->fc.inter_mode_probs[i][j] =
-            vp9_read_prob_diff_update(r, cm->fc.inter_mode_probs[i][j]);
-      }
-    }
+    for (j = 0; j < VP9_INTER_MODES - 1; ++j)
+      if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+        vp9_diff_update_prob(r, &fc->inter_mode_probs[i][j]);
 }
 
 static INLINE COMPPREDMODE_TYPE read_comp_pred_mode(vp9_reader *r) {
   COMPPREDMODE_TYPE mode = vp9_read_bit(r);
   if (mode)
-     mode += vp9_read_bit(r);
+    mode += vp9_read_bit(r);
   return mode;
 }
 
-static void mb_mode_mv_init(VP9D_COMP *pbi, vp9_reader *r) {
-  VP9_COMMON *const cm = &pbi->common;
-
-  if ((cm->frame_type != KEY_FRAME) && (!cm->intra_only)) {
-    nmv_context *const nmvc = &pbi->common.fc.nmvc;
-    MACROBLOCKD *const xd = &pbi->mb;
-    int i, j;
-
-    read_inter_mode_probs(cm, r);
-
-    if (cm->mcomp_filter_type == SWITCHABLE)
-      read_switchable_interp_probs(cm, r);
-
-    for (i = 0; i < INTRA_INTER_CONTEXTS; i++) {
-      if (vp9_read(r, VP9_MODE_UPDATE_PROB))
-        cm->fc.intra_inter_prob[i] =
-            vp9_read_prob_diff_update(r, cm->fc.intra_inter_prob[i]);
-    }
-
-    if (cm->allow_comp_inter_inter) {
-      cm->comp_pred_mode = read_comp_pred_mode(r);
-      if (cm->comp_pred_mode == HYBRID_PREDICTION)
-        for (i = 0; i < COMP_INTER_CONTEXTS; i++)
-          if (vp9_read(r, VP9_MODE_UPDATE_PROB))
-            cm->fc.comp_inter_prob[i] =
-                vp9_read_prob_diff_update(r, cm->fc.comp_inter_prob[i]);
-    } else {
-      cm->comp_pred_mode = SINGLE_PREDICTION_ONLY;
-    }
-
-    if (cm->comp_pred_mode != COMP_PREDICTION_ONLY)
-      for (i = 0; i < REF_CONTEXTS; i++) {
-        if (vp9_read(r, VP9_MODE_UPDATE_PROB))
-          cm->fc.single_ref_prob[i][0] =
-              vp9_read_prob_diff_update(r, cm->fc.single_ref_prob[i][0]);
-        if (vp9_read(r, VP9_MODE_UPDATE_PROB))
-          cm->fc.single_ref_prob[i][1] =
-              vp9_read_prob_diff_update(r, cm->fc.single_ref_prob[i][1]);
-      }
-
-    if (cm->comp_pred_mode != SINGLE_PREDICTION_ONLY)
-      for (i = 0; i < REF_CONTEXTS; i++)
-        if (vp9_read(r, VP9_MODE_UPDATE_PROB))
-          cm->fc.comp_ref_prob[i] =
-              vp9_read_prob_diff_update(r, cm->fc.comp_ref_prob[i]);
-
-    // VP9_INTRA_MODES
-    for (j = 0; j < BLOCK_SIZE_GROUPS; j++) {
-      for (i = 0; i < VP9_INTRA_MODES - 1; ++i) {
-        if (vp9_read(r, VP9_MODE_UPDATE_PROB)) {
-          cm->fc.y_mode_prob[j][i] =
-              vp9_read_prob_diff_update(r, cm->fc.y_mode_prob[j][i]);
-        }
-      }
-    }
-    for (j = 0; j < NUM_PARTITION_CONTEXTS; ++j) {
-      for (i = 0; i < PARTITION_TYPES - 1; ++i) {
-        if (vp9_read(r, VP9_MODE_UPDATE_PROB)) {
-          cm->fc.partition_prob[INTER_FRAME][j][i] =
-              vp9_read_prob_diff_update(r,
-                  cm->fc.partition_prob[INTER_FRAME][j][i]);
-        }
-      }
-    }
-
-    read_nmvprobs(r, nmvc, xd->allow_high_precision_mv);
-  }
-}
-
-// This function either reads the segment id for the current macroblock from
-// the bitstream or if the value is temporally predicted asserts the predicted
-// value
-static int read_mb_segment_id(VP9D_COMP *pbi, int mi_row, int mi_col,
-                              vp9_reader *r) {
-  VP9_COMMON *const cm = &pbi->common;
-  MACROBLOCKD *const xd = &pbi->mb;
-  MODE_INFO *const mi = xd->mode_info_context;
-  MB_MODE_INFO *const mbmi = &mi->mbmi;
-
-  if (!xd->segmentation_enabled)
-    return 0;  // Default for disabled segmentation
-
-  if (xd->update_mb_segmentation_map) {
-    int segment_id;
-
-    if (cm->temporal_update) {
-      // Temporal coding of the segment id for this mb is enabled.
-      // Get the context based probability for reading the
-      // prediction status flag
-      const vp9_prob pred_prob = vp9_get_pred_prob(cm, xd, PRED_SEG_ID);
-      const int pred_flag = vp9_read(r, pred_prob);
-      vp9_set_pred_flag(xd, PRED_SEG_ID, pred_flag);
-
-      // If the value is flagged as correctly predicted
-      // then use the predicted value, otherwise decode it explicitly
-      segment_id = pred_flag ? vp9_get_pred_mi_segid(cm, mbmi->sb_type,
-                                                     mi_row, mi_col)
-                             : read_mb_segid(r, xd);
-    } else {
-      segment_id = read_mb_segid(r, xd);  // Normal unpredicted coding mode
-    }
-
-    set_segment_id(cm, mbmi, mi_row, mi_col, segment_id);  // Side effect
-    return segment_id;
-  } else {
-    return vp9_get_pred_mi_segid(cm, mbmi->sb_type, mi_row, mi_col);
-  }
-}
-
-
 static INLINE void assign_and_clamp_mv(int_mv *dst, const int_mv *src,
                                        int mb_to_left_edge,
                                        int mb_to_right_edge,
@@ -454,242 +369,188 @@ static INLINE void assign_and_clamp_mv(int_mv *dst, const int_mv *src,
            mb_to_bottom_edge);
 }
 
-static INLINE void decode_mv(vp9_reader *r, MV *mv, const MV *ref,
-                             const nmv_context *ctx,
-                             nmv_context_counts *counts,
-                             int usehp) {
-  const MV_JOINT_TYPE j = treed_read(r, vp9_mv_joint_tree, ctx->joints);
-  MV diff = {0, 0};
-
-  usehp = usehp && vp9_use_nmv_hp(ref);
-  if (mv_joint_vertical(j))
-    diff.row = read_mv_component(r, &ctx->comps[0], usehp);
+static INLINE INTERPOLATIONFILTERTYPE read_switchable_filter_type(
+    VP9D_COMP *pbi, vp9_reader *r) {
+  VP9_COMMON *const cm = &pbi->common;
+  MACROBLOCKD *const xd = &pbi->mb;
+  const vp9_prob *probs = vp9_get_pred_probs_switchable_interp(cm, xd);
+  const int index = treed_read(r, vp9_switchable_interp_tree, probs);
+  const int ctx = vp9_get_pred_context_switchable_interp(xd);
+  ++cm->counts.switchable_interp[ctx][index];
+  return vp9_switchable_interp[index];
+}
 
-  if (mv_joint_horizontal(j))
-    diff.col = read_mv_component(r, &ctx->comps[1], usehp);
+static void read_intra_block_modes(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);
 
-  vp9_increment_nmv(&diff, ref, counts, usehp);
+  if (bsize >= BLOCK_SIZE_SB8X8) {
+    const int size_group = MIN(3, MIN(bwl, bhl));
+    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;
+     int idx, idy;
+
+     for (idy = 0; idy < 2; idy += bh) {
+       for (idx = 0; idx < 2; idx += bw) {
+         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)
+           mi->bmi[ib + 2].as_mode = b_mode;
+         if (bw == 2)
+           mi->bmi[ib + 1].as_mode = b_mode;
+      }
+    }
+    mbmi->mode = mi->bmi[3].as_mode;
+  }
 
-  mv->row = diff.row + ref->row;
-  mv->col = diff.col + ref->col;
+  mbmi->uv_mode = read_intra_mode(r, cm->fc.uv_mode_prob[mbmi->mode]);
+  cm->counts.uv_mode[mbmi->mode][mbmi->uv_mode]++;
 }
 
-static INLINE INTERPOLATIONFILTERTYPE read_switchable_filter_type(
-    VP9D_COMP *pbi, vp9_reader *r) {
-  const int index = treed_read(r, vp9_switchable_interp_tree,
-                               vp9_get_pred_probs(&pbi->common, &pbi->mb,
-                                                  PRED_SWITCHABLE_INTERP));
-  ++pbi->common.fc.switchable_interp_count
-                [vp9_get_pred_context(
-                    &pbi->common, &pbi->mb, PRED_SWITCHABLE_INTERP)][index];
-  return vp9_switchable_interp[index];
+static MV_REFERENCE_FRAME read_reference_frame(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]++;
+  } else {
+    ref = (MV_REFERENCE_FRAME) vp9_get_segdata(&xd->seg, segment_id,
+                                   SEG_LVL_REF_FRAME) != INTRA_FRAME;
+  }
+  return ref;
 }
 
-static void read_mb_modes_mv(VP9D_COMP *pbi, MODE_INFO *mi, MB_MODE_INFO *mbmi,
-                             int mi_row, int mi_col,
-                             vp9_reader *r) {
+static void read_inter_mode_info(VP9D_COMP *pbi, MODE_INFO *mi,
+                                 int mi_row, int mi_col, vp9_reader *r) {
   VP9_COMMON *const cm = &pbi->common;
-  nmv_context *const nmvc = &cm->fc.nmvc;
   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];
-  BLOCK_SIZE_TYPE bsize = mi->mbmi.sb_type;
-  int bw = 1 << b_width_log2(bsize);
-  int bh = 1 << b_height_log2(bsize);
+  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 mb_to_left_edge, mb_to_right_edge, mb_to_top_edge, mb_to_bottom_edge;
-  int j, idx, idy;
+  int idx, idy;
 
+  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);
 
-  // Make sure the MACROBLOCKD mode info pointer is pointed at the
-  // correct entry for the current macroblock.
-  xd->mode_info_context = mi;
-
-  // Distance of Mb to the various image edges.
-  // These specified to 8th pel as they are always compared to MV values
-  // that are in 1/8th pel units
-  set_mi_row_col(cm, xd, mi_row, 1 << mi_height_log2(bsize),
-                         mi_col, 1 << mi_width_log2(bsize));
-
-  mb_to_top_edge = xd->mb_to_top_edge - LEFT_TOP_MARGIN;
-  mb_to_bottom_edge = xd->mb_to_bottom_edge + RIGHT_BOTTOM_MARGIN;
-  mb_to_left_edge = xd->mb_to_left_edge - LEFT_TOP_MARGIN;
-  mb_to_right_edge = xd->mb_to_right_edge + RIGHT_BOTTOM_MARGIN;
-
-  // Read the macroblock segment id.
-  mbmi->segment_id = read_mb_segment_id(pbi, mi_row, mi_col, r);
-
-  mbmi->mb_skip_coeff = vp9_segfeature_active(xd, mbmi->segment_id,
-                                              SEG_LVL_SKIP);
-  if (!mbmi->mb_skip_coeff) {
-    mbmi->mb_skip_coeff = vp9_read(r, vp9_get_pred_prob(cm, xd, PRED_MBSKIP));
-    cm->fc.mbskip_count[vp9_get_pred_context(cm, xd, PRED_MBSKIP)]
-                       [mbmi->mb_skip_coeff]++;
-  }
-
-  // Read the reference frame
-  if (!vp9_segfeature_active(xd, mbmi->segment_id, SEG_LVL_REF_FRAME)) {
-    mbmi->ref_frame[0] =
-        vp9_read(r, vp9_get_pred_prob(cm, xd, PRED_INTRA_INTER));
-    cm->fc.intra_inter_count[vp9_get_pred_context(cm, xd, PRED_INTRA_INTER)]
-                            [mbmi->ref_frame[0] != INTRA_FRAME]++;
-  } else {
-    mbmi->ref_frame[0] =
-        vp9_get_segdata(xd, mbmi->segment_id, SEG_LVL_REF_FRAME) != INTRA_FRAME;
-  }
-
-  if (cm->txfm_mode == TX_MODE_SELECT &&
-      (mbmi->mb_skip_coeff == 0 || mbmi->ref_frame[0] == INTRA_FRAME) &&
-      bsize >= BLOCK_SIZE_SB8X8) {
-    mbmi->txfm_size = select_txfm_size(cm, xd, r, bsize);
-  } else if (bsize >= BLOCK_SIZE_SB32X32 &&
-             cm->txfm_mode >= ALLOW_32X32) {
-    mbmi->txfm_size = TX_32X32;
-  } else if (cm->txfm_mode >= ALLOW_16X16 &&
-             bsize >= BLOCK_SIZE_MB16X16) {
-    mbmi->txfm_size = TX_16X16;
-  } else if (cm->txfm_mode >= ALLOW_8X8 && (bsize >= BLOCK_SIZE_SB8X8)) {
-    mbmi->txfm_size = TX_8X8;
-  } else {
-    mbmi->txfm_size = TX_4X4;
-  }
-
-  // If reference frame is an Inter frame
   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_frame(pbi, r, mbmi->segment_id, mbmi->ref_frame);
+    ref0 = mbmi->ref_frame[0];
+    ref1 = mbmi->ref_frame[1];
 
-    {
-#ifdef DEC_DEBUG
-      if (dec_debug)
-        printf("%d %d\n", xd->mode_info_context->mbmi.mv[0].as_mv.row,
-               xd->mode_info_context->mbmi.mv[0].as_mv.col);
-#endif
-      vp9_find_mv_refs(cm, xd, mi, xd->prev_mode_info_context,
-                       mbmi->ref_frame[0], mbmi->ref_mvs[mbmi->ref_frame[0]],
-                       cm->ref_frame_sign_bias);
-
-      mv_ref_p = cm->fc.inter_mode_probs[
-        mbmi->mb_mode_context[mbmi->ref_frame[0]]];
-
-      // If the segment level skip mode enabled
-      if (vp9_segfeature_active(xd, mbmi->segment_id, SEG_LVL_SKIP)) {
-        mbmi->mode = ZEROMV;
-      } else if (bsize >= BLOCK_SIZE_SB8X8) {
-        mbmi->mode = read_sb_mv_ref(r, mv_ref_p);
-        vp9_accum_mv_refs(cm, mbmi->mode,
-                          mbmi->mb_mode_context[mbmi->ref_frame[0]]);
-      }
+    vp9_find_mv_refs(cm, xd, mi, xd->prev_mode_info_context,
+                     ref0, mbmi->ref_mvs[ref0], cm->ref_frame_sign_bias);
 
-      if (bsize < BLOCK_SIZE_SB8X8 || mbmi->mode != ZEROMV) {
-        vp9_find_best_ref_mvs(xd,
-                              mbmi->ref_mvs[mbmi->ref_frame[0]],
-                              &nearest, &nearby);
+    mv_ref_p = cm->fc.inter_mode_probs[mbmi->mb_mode_context[ref0]];
 
-        best_mv.as_int = mbmi->ref_mvs[mbmi->ref_frame[0]][0].as_int;
-      }
+    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;
 
-#ifdef DEC_DEBUG
-      if (dec_debug)
-        printf("[D %d %d] %d %d %d %d\n", ref_frame,
-               mbmi->mb_mode_context[ref_frame],
-               mv_ref_p[0], mv_ref_p[1], mv_ref_p[2], mv_ref_p[3]);
-#endif
+    // 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;
 
-    if (mbmi->ref_frame[1] > INTRA_FRAME) {
+    if (ref1 > INTRA_FRAME) {
       vp9_find_mv_refs(cm, xd, mi, xd->prev_mode_info_context,
-                       mbmi->ref_frame[1],
-                       mbmi->ref_mvs[mbmi->ref_frame[1]],
-                       cm->ref_frame_sign_bias);
+                       ref1, mbmi->ref_mvs[ref1], cm->ref_frame_sign_bias);
 
       if (bsize < BLOCK_SIZE_SB8X8 || mbmi->mode != ZEROMV) {
-        vp9_find_best_ref_mvs(xd,
-                              mbmi->ref_mvs[mbmi->ref_frame[1]],
-                              &nearest_second,
-                              &nearby_second);
-        best_mv_second.as_int = mbmi->ref_mvs[mbmi->ref_frame[1]][0].as_int;
+        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;
       }
     }
 
-    mbmi->uv_mode = DC_PRED;
+
     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;
-          int blockmode;
-          int i;
-          j = idy * 2 + idx;
+          const int j = idy * 2 + idx;
+          const int blockmode = read_inter_mode(r, mv_ref_p);
 
-          blockmode = read_sb_mv_ref(r, mv_ref_p);
-          vp9_accum_mv_refs(cm, blockmode,
-                            mbmi->mb_mode_context[mbmi->ref_frame[0]]);
+          vp9_accum_mv_refs(cm, blockmode, mbmi->mb_mode_context[ref0]);
           if (blockmode == NEARESTMV || blockmode == NEARMV) {
-            MV_REFERENCE_FRAME rf2 = mbmi->ref_frame[1];
             vp9_append_sub8x8_mvs_for_idx(cm, xd, &nearest, &nearby, j, 0);
-            if (rf2 > 0) {
+            if (ref1 > 0)
               vp9_append_sub8x8_mvs_for_idx(cm, xd,  &nearest_second,
                                             &nearby_second, j, 1);
-            }
           }
 
           switch (blockmode) {
             case NEWMV:
-              decode_mv(r, &blockmv.as_mv, &best_mv.as_mv, nmvc,
-                         &cm->fc.NMVcount, xd->allow_high_precision_mv);
+              read_mv(r, &blockmv.as_mv, &best_mv.as_mv, nmvc,
+                      &cm->counts.mv, xd->allow_high_precision_mv);
 
-              if (mbmi->ref_frame[1] > 0)
-                decode_mv(r, &secondmv.as_mv, &best_mv_second.as_mv, nmvc,
-                          &cm->fc.NMVcount, xd->allow_high_precision_mv);
-
-#ifdef VPX_MODE_COUNT
-              vp9_mv_cont_count[mv_contz][3]++;
-#endif
+              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 (mbmi->ref_frame[1] > 0)
+              if (ref1 > 0)
                 secondmv.as_int = nearest_second.as_int;
-#ifdef VPX_MODE_COUNT
-              vp9_mv_cont_count[mv_contz][0]++;
-#endif
               break;
             case NEARMV:
               blockmv.as_int = nearby.as_int;
-              if (mbmi->ref_frame[1] > 0)
+              if (ref1 > 0)
                 secondmv.as_int = nearby_second.as_int;
-#ifdef VPX_MODE_COUNT
-              vp9_mv_cont_count[mv_contz][1]++;
-#endif
               break;
             case ZEROMV:
               blockmv.as_int = 0;
-              if (mbmi->ref_frame[1] > 0)
+              if (ref1 > 0)
                 secondmv.as_int = 0;
-#ifdef VPX_MODE_COUNT
-              vp9_mv_cont_count[mv_contz][2]++;
-#endif
               break;
             default:
-              break;
+              assert(!"Invalid inter mode value");
           }
           mi->bmi[j].as_mv[0].as_int = blockmv.as_int;
-          if (mbmi->ref_frame[1] > 0)
+          if (ref1 > 0)
             mi->bmi[j].as_mv[1].as_int = secondmv.as_int;
 
-          for (i = 1; i < bh; ++i)
-            vpx_memcpy(&mi->bmi[j + i * 2], &mi->bmi[j], sizeof(mi->bmi[j]));
-          for (i = 1; i < bw; ++i)
-            vpx_memcpy(&mi->bmi[j + i], &mi->bmi[j], sizeof(mi->bmi[j]));
+          if (bh == 2)
+            mi->bmi[j + 2] = mi->bmi[j];
+          if (bw == 2)
+            mi->bmi[j + 1] = mi->bmi[j];
           mi->mbmi.mode = blockmode;
         }
       }
@@ -697,6 +558,11 @@ static void read_mb_modes_mv(VP9D_COMP *pbi, MODE_INFO *mi, MB_MODE_INFO *mbmi,
       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
@@ -704,7 +570,7 @@ static void read_mb_modes_mv(VP9D_COMP *pbi, MODE_INFO *mi, MB_MODE_INFO *mbmi,
                                             mb_to_right_edge,
                                             mb_to_top_edge,
                                             mb_to_bottom_edge);
-          if (mbmi->ref_frame[1] > 0)
+          if (ref1 > 0)
             assign_and_clamp_mv(mv1, &nearby_second, mb_to_left_edge,
                                                      mb_to_right_edge,
                                                      mb_to_top_edge,
@@ -717,7 +583,7 @@ static void read_mb_modes_mv(VP9D_COMP *pbi, MODE_INFO *mi, MB_MODE_INFO *mbmi,
                                              mb_to_right_edge,
                                              mb_to_top_edge,
                                              mb_to_bottom_edge);
-          if (mbmi->ref_frame[1] > 0)
+          if (ref1 > 0)
             assign_and_clamp_mv(mv1, &nearest_second, mb_to_left_edge,
                                                       mb_to_right_edge,
                                                       mb_to_top_edge,
@@ -726,98 +592,109 @@ static void read_mb_modes_mv(VP9D_COMP *pbi, MODE_INFO *mi, MB_MODE_INFO *mbmi,
 
         case ZEROMV:
           mv0->as_int = 0;
-          if (mbmi->ref_frame[1] > 0)
+          if (ref1 > 0)
             mv1->as_int = 0;
           break;
 
         case NEWMV:
-          decode_mv(r, &mv0->as_mv, &best_mv.as_mv, nmvc, &cm->fc.NMVcount,
-                    xd->allow_high_precision_mv);
-          if (mbmi->ref_frame[1] > 0)
-            decode_mv(r, &mv1->as_mv, &best_mv_second.as_mv, nmvc,
-                      &cm->fc.NMVcount, xd->allow_high_precision_mv);
+          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:
-#if CONFIG_DEBUG
-          assert(0);
-#endif
-          break;
+          assert(!"Invalid inter mode value");
       }
     }
   } else {
-    // required for left and above block mv
-    mv0->as_int = 0;
-
-    if (bsize >= BLOCK_SIZE_SB8X8) {
-      const BLOCK_SIZE_TYPE bsize = xd->mode_info_context->mbmi.sb_type;
-      const int bwl = b_width_log2(bsize), bhl = b_height_log2(bsize);
-      const int bsl = MIN(bwl, bhl);
-      mbmi->mode = read_intra_mode(r, cm->fc.y_mode_prob[MIN(3, bsl)]);
-      cm->fc.y_mode_counts[MIN(3, bsl)][mbmi->mode]++;
-    } else {
-      int idx, idy;
-      for (idy = 0; idy < 2; idy += bh) {
-        for (idx = 0; idx < 2; idx += bw) {
-          int ib = idy * 2 + idx, k;
-          int m = read_intra_mode(r, cm->fc.y_mode_prob[0]);
-          mi->bmi[ib].as_mode.first = m;
-          cm->fc.y_mode_counts[0][m]++;
-          for (k = 1; k < bh; ++k)
-            mi->bmi[ib + k * 2].as_mode.first = m;
-          for (k = 1; k < bw; ++k)
-            mi->bmi[ib + k].as_mode.first = m;
-        }
-      }
-      mbmi->mode = mi->bmi[3].as_mode.first;
+    mv0->as_int = 0;  // required for left and above block mv
+    read_intra_block_modes(pbi, mi, r);
+  }
+}
+
+static void read_comp_pred(VP9_COMMON *cm, vp9_reader *r) {
+  int i;
+
+  cm->comp_pred_mode = cm->allow_comp_inter_inter ? read_comp_pred_mode(r)
+                                                  : SINGLE_PREDICTION_ONLY;
+
+  if (cm->comp_pred_mode == HYBRID_PREDICTION)
+    for (i = 0; i < COMP_INTER_CONTEXTS; i++)
+      if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+        vp9_diff_update_prob(r, &cm->fc.comp_inter_prob[i]);
+
+  if (cm->comp_pred_mode != COMP_PREDICTION_ONLY)
+    for (i = 0; i < REF_CONTEXTS; i++) {
+      if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+        vp9_diff_update_prob(r, &cm->fc.single_ref_prob[i][0]);
+      if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+        vp9_diff_update_prob(r, &cm->fc.single_ref_prob[i][1]);
     }
 
-    mbmi->uv_mode = read_intra_mode(r, cm->fc.uv_mode_prob[mbmi->mode]);
-    cm->fc.uv_mode_counts[mbmi->mode][mbmi->uv_mode]++;
-  }
+  if (cm->comp_pred_mode != SINGLE_PREDICTION_ONLY)
+    for (i = 0; i < REF_CONTEXTS; i++)
+      if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+        vp9_diff_update_prob(r, &cm->fc.comp_ref_prob[i]);
 }
 
-void vp9_decode_mode_mvs_init(VP9D_COMP* const pbi, vp9_reader *r) {
-  VP9_COMMON *cm = &pbi->common;
+void vp9_prepare_read_mode_info(VP9D_COMP* pbi, vp9_reader *r) {
+  VP9_COMMON *const cm = &pbi->common;
   int k;
 
   // TODO(jkoleszar): does this clear more than MBSKIP_CONTEXTS? Maybe remove.
   // vpx_memset(cm->fc.mbskip_probs, 0, sizeof(cm->fc.mbskip_probs));
-  for (k = 0; k < MBSKIP_CONTEXTS; ++k) {
-    if (vp9_read(r, VP9_MODE_UPDATE_PROB)) {
-      cm->fc.mbskip_probs[k] =
-          vp9_read_prob_diff_update(r, cm->fc.mbskip_probs[k]);
-    }
-    // cm->fc.mbskip_probs[k] = vp9_read_prob(r);
-  }
+  for (k = 0; k < MBSKIP_CONTEXTS; ++k)
+    if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+      vp9_diff_update_prob(r, &cm->fc.mbskip_probs[k]);
+
+  if (cm->frame_type != KEY_FRAME && !cm->intra_only) {
+    nmv_context *const nmvc = &pbi->common.fc.nmvc;
+    MACROBLOCKD *const xd = &pbi->mb;
+    int i, j;
+
+    read_inter_mode_probs(&cm->fc, r);
 
-  mb_mode_mv_init(pbi, r);
+    if (cm->mcomp_filter_type == SWITCHABLE)
+      read_switchable_interp_probs(&cm->fc, r);
+
+    for (i = 0; i < INTRA_INTER_CONTEXTS; i++)
+      if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+        vp9_diff_update_prob(r, &cm->fc.intra_inter_prob[i]);
+
+    read_comp_pred(cm, r);
+
+    for (j = 0; j < BLOCK_SIZE_GROUPS; j++)
+      for (i = 0; i < VP9_INTRA_MODES - 1; ++i)
+        if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+          vp9_diff_update_prob(r, &cm->fc.y_mode_prob[j][i]);
+
+    for (j = 0; j < NUM_PARTITION_CONTEXTS; ++j)
+      for (i = 0; i < PARTITION_TYPES - 1; ++i)
+        if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+          vp9_diff_update_prob(r, &cm->fc.partition_prob[INTER_FRAME][j][i]);
+
+    read_mv_probs(r, nmvc, xd->allow_high_precision_mv);
+  }
 }
 
-void vp9_decode_mb_mode_mv(VP9D_COMP* const pbi,
-                           MACROBLOCKD* const xd,
-                           int mi_row,
-                           int mi_col,
-                           vp9_reader *r) {
+void vp9_read_mode_info(VP9D_COMP* pbi, int mi_row, int mi_col, vp9_reader *r) {
   VP9_COMMON *const cm = &pbi->common;
+  MACROBLOCKD *const xd = &pbi->mb;
   MODE_INFO *mi = xd->mode_info_context;
-  MB_MODE_INFO *const mbmi = &mi->mbmi;
+  const BLOCK_SIZE_TYPE bsize = mi->mbmi.sb_type;
+  const int bw = 1 << mi_width_log2(bsize);
+  const int bh = 1 << mi_height_log2(bsize);
+  const int y_mis = MIN(bh, cm->mi_rows - mi_row);
+  const int x_mis = MIN(bw, cm->mi_cols - mi_col);
+  int x, y;
 
-  if ((cm->frame_type == KEY_FRAME) || cm->intra_only) {
-    kfread_modes(pbi, mi, mi_row, mi_col, r);
-  } else {
-    read_mb_modes_mv(pbi, mi, &mi->mbmi, mi_row, mi_col, r);
-  }
+  if (cm->frame_type == KEY_FRAME || cm->intra_only)
+    read_intra_mode_info(pbi, mi, mi_row, mi_col, r);
+  else
+    read_inter_mode_info(pbi, mi, mi_row, mi_col, r);
 
-  if (1) {
-    const int bw = 1 << mi_width_log2(mbmi->sb_type);
-    const int bh = 1 << mi_height_log2(mbmi->sb_type);
-    const int y_mis = MIN(bh, cm->mi_rows - mi_row);
-    const int x_mis = MIN(bw, cm->mi_cols - mi_col);
-    const int mis = cm->mode_info_stride;
-    int x, y;
-
-    for (y = 0; y < y_mis; y++)
-      for (x = !y; x < x_mis; x++)
-        mi[y * mis + x] = *mi;
-  }
+  for (y = 0; y < y_mis; y++)
+    for (x = !y; x < x_mis; x++)
+      mi[y * cm->mode_info_stride + x] = *mi;
 }
diff --git a/libvpx/vp9/decoder/vp9_decodemv.h b/libvpx/vp9/decoder/vp9_decodemv.h
index bf5e83c..4073d9e 100644
--- a/libvpx/vp9/decoder/vp9_decodemv.h
+++ b/libvpx/vp9/decoder/vp9_decodemv.h
@@ -13,11 +13,8 @@
 
 #include "vp9/decoder/vp9_onyxd_int.h"
 
-void vp9_decode_mb_mode_mv(VP9D_COMP* const pbi,
-                           MACROBLOCKD* const xd,
-                           int mb_row,
-                           int mb_col,
-                           vp9_reader *r);
-void vp9_decode_mode_mvs_init(VP9D_COMP* const pbi, vp9_reader *r);
+void vp9_prepare_read_mode_info(VP9D_COMP* pbi, vp9_reader *r);
+
+void vp9_read_mode_info(VP9D_COMP* pbi, int mi_row, int mi_col, vp9_reader *r);
 
 #endif  // VP9_DECODER_VP9_DECODEMV_H_
diff --git a/libvpx/vp9/decoder/vp9_decodframe.c b/libvpx/vp9/decoder/vp9_decodframe.c
index 49b181d..ffec8ea 100644
--- a/libvpx/vp9/decoder/vp9_decodframe.c
+++ b/libvpx/vp9/decoder/vp9_decodframe.c
@@ -14,15 +14,15 @@
 #include "vpx_mem/vpx_mem.h"
 #include "vpx_scale/vpx_scale.h"
 
-#include "vp9/common/vp9_extend.h"
-#include "vp9/common/vp9_modecont.h"
+#include "vp9/common/vp9_alloccommon.h"
 #include "vp9/common/vp9_common.h"
-#include "vp9/common/vp9_reconintra.h"
-#include "vp9/common/vp9_reconinter.h"
 #include "vp9/common/vp9_entropy.h"
-#include "vp9/common/vp9_alloccommon.h"
 #include "vp9/common/vp9_entropymode.h"
+#include "vp9/common/vp9_extend.h"
+#include "vp9/common/vp9_pred_common.h"
 #include "vp9/common/vp9_quant_common.h"
+#include "vp9/common/vp9_reconintra.h"
+#include "vp9/common/vp9_reconinter.h"
 #include "vp9/common/vp9_seg_common.h"
 #include "vp9/common/vp9_tile_common.h"
 
@@ -30,169 +30,58 @@
 #include "vp9/decoder/vp9_decodframe.h"
 #include "vp9/decoder/vp9_detokenize.h"
 #include "vp9/decoder/vp9_decodemv.h"
+#include "vp9/decoder/vp9_dsubexp.h"
 #include "vp9/decoder/vp9_onyxd_int.h"
 #include "vp9/decoder/vp9_read_bit_buffer.h"
 
-
-// #define DEC_DEBUG
-#ifdef DEC_DEBUG
-int dec_debug = 0;
-#endif
-
 static int read_be32(const uint8_t *p) {
   return (p[0] << 24) | (p[1] << 16) | (p[2] << 8) | p[3];
 }
 
 // len == 0 is not allowed
-static int read_is_valid(const uint8_t *start, size_t len,
-                         const uint8_t *end) {
+static int read_is_valid(const uint8_t *start, size_t len, const uint8_t *end) {
   return start + len > start && start + len <= end;
 }
 
-static void setup_txfm_mode(VP9_COMMON *pc, int lossless, vp9_reader *r) {
-  if (lossless) {
-    pc->txfm_mode = ONLY_4X4;
-  } else {
-    pc->txfm_mode = vp9_read_literal(r, 2);
-    if (pc->txfm_mode == ALLOW_32X32)
-      pc->txfm_mode += vp9_read_bit(r);
-    if (pc->txfm_mode == TX_MODE_SELECT) {
-      int i, j;
-      for (i = 0; i < TX_SIZE_CONTEXTS; ++i) {
-        for (j = 0; j < TX_SIZE_MAX_SB - 3; ++j) {
-          if (vp9_read(r, VP9_MODE_UPDATE_PROB))
-            pc->fc.tx_probs_8x8p[i][j] =
-                vp9_read_prob_diff_update(r, pc->fc.tx_probs_8x8p[i][j]);
-        }
-      }
-      for (i = 0; i < TX_SIZE_CONTEXTS; ++i) {
-        for (j = 0; j < TX_SIZE_MAX_SB - 2; ++j) {
-          if (vp9_read(r, VP9_MODE_UPDATE_PROB))
-            pc->fc.tx_probs_16x16p[i][j] =
-                vp9_read_prob_diff_update(r, pc->fc.tx_probs_16x16p[i][j]);
-        }
-      }
-      for (i = 0; i < TX_SIZE_CONTEXTS; ++i) {
-        for (j = 0; j < TX_SIZE_MAX_SB - 1; ++j) {
-          if (vp9_read(r, VP9_MODE_UPDATE_PROB))
-            pc->fc.tx_probs_32x32p[i][j] =
-                vp9_read_prob_diff_update(r, pc->fc.tx_probs_32x32p[i][j]);
-        }
-      }
-    }
-  }
-}
-
-static int get_unsigned_bits(unsigned int num_values) {
-  int cat = 0;
-  if (num_values <= 1)
-    return 0;
-  num_values--;
-  while (num_values > 0) {
-    cat++;
-    num_values >>= 1;
-  }
-  return cat;
-}
-
-static int inv_recenter_nonneg(int v, int m) {
-  if (v > 2 * m)
-    return v;
-
-  return v % 2 ? m - (v + 1) / 2 : m + v / 2;
-}
-
-static int decode_uniform(vp9_reader *r, int n) {
-  int v;
-  const int l = get_unsigned_bits(n);
-  const int m = (1 << l) - n;
-  if (!l)
-    return 0;
-
-  v = vp9_read_literal(r, l - 1);
-  return v < m ?  v : (v << 1) - m + vp9_read_bit(r);
-}
-
-static int decode_term_subexp(vp9_reader *r, int k, int num_syms) {
-  int i = 0, mk = 0, word;
-  while (1) {
-    const int b = i ? k + i - 1 : k;
-    const int a = 1 << b;
-    if (num_syms <= mk + 3 * a) {
-      word = decode_uniform(r, num_syms - mk) + mk;
-      break;
-    } else {
-      if (vp9_read_bit(r)) {
-        i++;
-        mk += a;
-      } else {
-        word = vp9_read_literal(r, b) + mk;
-        break;
-      }
-    }
-  }
-  return word;
-}
-
 static int decode_unsigned_max(struct vp9_read_bit_buffer *rb, int max) {
   const int data = vp9_rb_read_literal(rb, get_unsigned_bits(max));
   return data > max ? max : data;
 }
 
-static int merge_index(int v, int n, int modulus) {
-  int max1 = (n - 1 - modulus / 2) / modulus + 1;
-  if (v < max1) {
-    v = v * modulus + modulus / 2;
-  } else {
-    int w;
-    v -= max1;
-    w = v;
-    v += (v + modulus - modulus / 2) / modulus;
-    while (v % modulus == modulus / 2 ||
-           w != v - (v + modulus - modulus / 2) / modulus) v++;
-  }
-  return v;
-}
-
-static int inv_remap_prob(int v, int m) {
-  const int n = 255;
-
-  v = merge_index(v, n - 1, MODULUS_PARAM);
-  m--;
-  if ((m << 1) <= n) {
-    return 1 + inv_recenter_nonneg(v + 1, m);
-  } else {
-    return n - inv_recenter_nonneg(v + 1, n - 1 - m);
-  }
+static TX_MODE read_tx_mode(vp9_reader *r) {
+  TX_MODE tx_mode = vp9_read_literal(r, 2);
+  if (tx_mode == ALLOW_32X32)
+    tx_mode += vp9_read_bit(r);
+  return tx_mode;
 }
 
-vp9_prob vp9_read_prob_diff_update(vp9_reader *r, int oldp) {
-  int delp = decode_term_subexp(r, SUBEXP_PARAM, 255);
-  return (vp9_prob)inv_remap_prob(delp, oldp);
-}
+static void read_tx_probs(struct tx_probs *tx_probs, vp9_reader *r) {
+  int i, j;
 
-void vp9_init_dequantizer(VP9_COMMON *pc) {
-  int q;
+  for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
+    for (j = 0; j < TX_SIZE_MAX_SB - 3; ++j)
+      if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+        vp9_diff_update_prob(r, &tx_probs->p8x8[i][j]);
 
-  for (q = 0; q < QINDEX_RANGE; q++) {
-    // DC value
-    pc->y_dequant[q][0] = vp9_dc_quant(q, pc->y_dc_delta_q);
-    pc->uv_dequant[q][0] = vp9_dc_quant(q, pc->uv_dc_delta_q);
+  for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
+    for (j = 0; j < TX_SIZE_MAX_SB - 2; ++j)
+      if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+        vp9_diff_update_prob(r, &tx_probs->p16x16[i][j]);
 
-    // AC values
-    pc->y_dequant[q][1] = vp9_ac_quant(q, 0);
-    pc->uv_dequant[q][1] = vp9_ac_quant(q, pc->uv_ac_delta_q);
-  }
+  for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
+    for (j = 0; j < TX_SIZE_MAX_SB - 1; ++j)
+      if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+        vp9_diff_update_prob(r, &tx_probs->p32x32[i][j]);
 }
 
-static void mb_init_dequantizer(VP9_COMMON *pc, MACROBLOCKD *xd) {
+static void init_dequantizer(VP9_COMMON *cm, MACROBLOCKD *xd) {
   int i;
   const int segment_id = xd->mode_info_context->mbmi.segment_id;
-  xd->q_index = vp9_get_qindex(xd, segment_id, pc->base_qindex);
+  xd->q_index = vp9_get_qindex(xd, segment_id, cm->base_qindex);
 
-  xd->plane[0].dequant = pc->y_dequant[xd->q_index];
+  xd->plane[0].dequant = cm->y_dequant[xd->q_index];
   for (i = 1; i < MAX_MB_PLANE; i++)
-    xd->plane[i].dequant = pc->uv_dequant[xd->q_index];
+    xd->plane[i].dequant = cm->uv_dequant[xd->q_index];
 }
 
 static void decode_block(int plane, int block, BLOCK_SIZE_TYPE bsize,
@@ -201,32 +90,32 @@ static void decode_block(int plane, int block, BLOCK_SIZE_TYPE bsize,
   struct macroblockd_plane *pd = &xd->plane[plane];
   int16_t* const qcoeff = BLOCK_OFFSET(pd->qcoeff, block, 16);
   const int stride = pd->dst.stride;
+  const int eob = pd->eobs[block];
   const int raster_block = txfrm_block_to_raster_block(xd, bsize, plane,
                                                        block, ss_txfrm_size);
   uint8_t* const dst = raster_block_offset_uint8(xd, bsize, plane,
                                                  raster_block,
                                                  pd->dst.buf, stride);
 
-  TX_TYPE tx_type;
-
   switch (ss_txfrm_size / 2) {
-    case TX_4X4:
-      tx_type = plane == 0 ? get_tx_type_4x4(xd, raster_block) : DCT_DCT;
+    case TX_4X4: {
+      const TX_TYPE tx_type = get_tx_type_4x4(pd->plane_type, xd, raster_block);
       if (tx_type == DCT_DCT)
-        xd->itxm_add(qcoeff, dst, stride, pd->eobs[block]);
+        xd->itxm_add(qcoeff, dst, stride, eob);
       else
-        vp9_iht_add_c(tx_type, qcoeff, dst, stride, pd->eobs[block]);
+        vp9_iht_add_c(tx_type, qcoeff, dst, stride, eob);
       break;
+    }
     case TX_8X8:
-      tx_type = plane == 0 ? get_tx_type_8x8(xd, raster_block) : DCT_DCT;
-      vp9_iht_add_8x8_c(tx_type, qcoeff, dst, stride, pd->eobs[block]);
+      vp9_iht_add_8x8_c(get_tx_type_8x8(pd->plane_type, xd), qcoeff, dst,
+                        stride, eob);
       break;
     case TX_16X16:
-      tx_type = plane == 0 ? get_tx_type_16x16(xd, raster_block) : DCT_DCT;
-      vp9_iht_add_16x16_c(tx_type, qcoeff, dst, stride, pd->eobs[block]);
+      vp9_iht_add_16x16_c(get_tx_type_16x16(pd->plane_type, xd), qcoeff, dst,
+                          stride, eob);
       break;
     case TX_32X32:
-      vp9_idct_add_32x32(qcoeff, dst, stride, pd->eobs[block]);
+      vp9_idct_add_32x32(qcoeff, dst, stride, eob);
       break;
   }
 }
@@ -235,6 +124,7 @@ static void decode_block_intra(int plane, int block, BLOCK_SIZE_TYPE bsize,
                                int ss_txfrm_size, void *arg) {
   MACROBLOCKD* const xd = arg;
   struct macroblockd_plane *pd = &xd->plane[plane];
+  MODE_INFO *const mi = xd->mode_info_context;
 
   const int raster_block = txfrm_block_to_raster_block(xd, bsize, plane,
                                                        block, ss_txfrm_size);
@@ -245,13 +135,12 @@ static void decode_block_intra(int plane, int block, BLOCK_SIZE_TYPE bsize,
   int b_mode;
   int plane_b_size;
   const int tx_ib = raster_block >> tx_size;
-  const int mode = plane == 0 ? xd->mode_info_context->mbmi.mode
-                              : xd->mode_info_context->mbmi.uv_mode;
+  const int mode = plane == 0 ? mi->mbmi.mode
+                              : mi->mbmi.uv_mode;
 
-
-  if (plane == 0 && xd->mode_info_context->mbmi.sb_type < BLOCK_SIZE_SB8X8) {
+  if (plane == 0 && mi->mbmi.sb_type < BLOCK_SIZE_SB8X8) {
     assert(bsize == BLOCK_SIZE_SB8X8);
-    b_mode = xd->mode_info_context->bmi[raster_block].as_mode.first;
+    b_mode = mi->bmi[raster_block].as_mode;
   } else {
     b_mode = mode;
   }
@@ -261,97 +150,28 @@ static void decode_block_intra(int plane, int block, BLOCK_SIZE_TYPE bsize,
 
   plane_b_size = b_width_log2(bsize) - pd->subsampling_x;
   vp9_predict_intra_block(xd, tx_ib, plane_b_size, tx_size, b_mode,
+                          dst, pd->dst.stride,
                           dst, pd->dst.stride);
 
   // Early exit if there are no coefficients
-  if (xd->mode_info_context->mbmi.mb_skip_coeff)
+  if (mi->mbmi.mb_skip_coeff)
     return;
 
   decode_block(plane, block, bsize, ss_txfrm_size, arg);
 }
 
-static void decode_atom(VP9D_COMP *pbi, MACROBLOCKD *xd,
-                        int mi_row, int mi_col,
-                        vp9_reader *r, BLOCK_SIZE_TYPE bsize) {
-  MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
-
-  assert(mbmi->ref_frame[0] != INTRA_FRAME);
-
-  if ((pbi->common.frame_type != KEY_FRAME) && (!pbi->common.intra_only))
-    vp9_setup_interp_filters(xd, mbmi->interp_filter, &pbi->common);
-
-  // prediction
-  vp9_build_inter_predictors_sb(xd, mi_row, mi_col, bsize);
-
-  if (mbmi->mb_skip_coeff) {
-    vp9_reset_sb_tokens_context(xd, bsize);
-  } else {
-    if (xd->segmentation_enabled)
-      mb_init_dequantizer(&pbi->common, xd);
-
-    if (!vp9_reader_has_error(r))
-      vp9_decode_tokens(pbi, r, bsize);
-
-    foreach_transformed_block(xd, bsize, decode_block, xd);
-  }
-}
+static int decode_tokens(VP9D_COMP *pbi, BLOCK_SIZE_TYPE bsize, vp9_reader *r) {
+  MACROBLOCKD *const xd = &pbi->mb;
 
-static void decode_sb_intra(VP9D_COMP *pbi, MACROBLOCKD *xd,
-                          int mi_row, int mi_col,
-                          vp9_reader *r, BLOCK_SIZE_TYPE bsize) {
-  MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
-  if (mbmi->mb_skip_coeff) {
+  if (xd->mode_info_context->mbmi.mb_skip_coeff) {
     vp9_reset_sb_tokens_context(xd, bsize);
+    return -1;
   } else {
-    if (xd->segmentation_enabled)
-      mb_init_dequantizer(&pbi->common, xd);
-
-    if (!vp9_reader_has_error(r))
-      vp9_decode_tokens(pbi, r, bsize);
-  }
+    if (xd->seg.enabled)
+      init_dequantizer(&pbi->common, xd);
 
-  foreach_transformed_block(xd, bsize, decode_block_intra, xd);
-}
-
-
-static void decode_sb(VP9D_COMP *pbi, MACROBLOCKD *xd, int mi_row, int mi_col,
-                      vp9_reader *r, BLOCK_SIZE_TYPE bsize) {
-  const int bwl = mi_width_log2(bsize), bhl = mi_height_log2(bsize);
-  const int bw = 1 << bwl, bh = 1 << bhl;
-  int n, eobtotal;
-  VP9_COMMON *const pc = &pbi->common;
-  MODE_INFO *const mi = xd->mode_info_context;
-  MB_MODE_INFO *const mbmi = &mi->mbmi;
-  const int mis = pc->mode_info_stride;
-
-  assert(mbmi->sb_type == bsize);
-  assert(mbmi->ref_frame[0] != INTRA_FRAME);
-
-  if (pbi->common.frame_type != KEY_FRAME)
-    vp9_setup_interp_filters(xd, mbmi->interp_filter, pc);
-
-  // generate prediction
-  vp9_build_inter_predictors_sb(xd, mi_row, mi_col, bsize);
-
-  if (mbmi->mb_skip_coeff) {
-    vp9_reset_sb_tokens_context(xd, bsize);
-  } else {
-    // re-initialize macroblock dequantizer before detokenization
-    if (xd->segmentation_enabled)
-      mb_init_dequantizer(pc, xd);
-
-    // dequantization and idct
-    eobtotal = vp9_decode_tokens(pbi, r, bsize);
-    if (eobtotal == 0) {  // skip loopfilter
-      for (n = 0; n < bw * bh; n++) {
-        const int x_idx = n & (bw - 1), y_idx = n >> bwl;
-
-        if (mi_col + x_idx < pc->mi_cols && mi_row + y_idx < pc->mi_rows)
-          mi[y_idx * mis + x_idx].mbmi.mb_skip_coeff = 1;
-      }
-    } else {
-      foreach_transformed_block(xd, bsize, decode_block, xd);
-    }
+    // TODO(dkovalev) if (!vp9_reader_has_error(r))
+    return vp9_decode_tokens(pbi, r, bsize);
   }
 }
 
@@ -377,8 +197,8 @@ static void set_offsets(VP9D_COMP *pbi, BLOCK_SIZE_TYPE bsize,
     pd->left_context = cm->left_context[i] +
                            (((mi_row * 2) & 15) >> pd->subsampling_y);
   }
-  xd->above_seg_context = cm->above_seg_context + mi_col;
-  xd->left_seg_context  = cm->left_seg_context + (mi_row & MI_MASK);
+
+  set_partition_seg_context(cm, xd, mi_row, mi_col);
 
   // Distance of Mb to the various image edges. These are specified to 8th pel
   // as they are always compared to values that are in 1/8th pel units
@@ -387,53 +207,65 @@ static void set_offsets(VP9D_COMP *pbi, BLOCK_SIZE_TYPE bsize,
   setup_dst_planes(xd, &cm->yv12_fb[cm->new_fb_idx], mi_row, mi_col);
 }
 
-static void set_refs(VP9D_COMP *pbi, int mi_row, int mi_col) {
+static void set_ref(VP9D_COMP *pbi, int i, int mi_row, int mi_col) {
   VP9_COMMON *const cm = &pbi->common;
   MACROBLOCKD *const xd = &pbi->mb;
   MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
+  const int ref = mbmi->ref_frame[i] - 1;
 
-  if (mbmi->ref_frame[0] > INTRA_FRAME) {
-    // Select the appropriate reference frame for this MB
-    const int fb_idx = cm->active_ref_idx[mbmi->ref_frame[0] - 1];
-    const YV12_BUFFER_CONFIG *cfg = &cm->yv12_fb[fb_idx];
-    xd->scale_factor[0]    = cm->active_ref_scale[mbmi->ref_frame[0] - 1];
-    xd->scale_factor_uv[0] = cm->active_ref_scale[mbmi->ref_frame[0] - 1];
-    setup_pre_planes(xd, cfg, NULL, mi_row, mi_col,
-                     xd->scale_factor, xd->scale_factor_uv);
-    xd->corrupted |= cfg->corrupted;
-
-    if (mbmi->ref_frame[1] > INTRA_FRAME) {
-      // Select the appropriate reference frame for this MB
-      const int second_fb_idx = cm->active_ref_idx[mbmi->ref_frame[1] - 1];
-      const YV12_BUFFER_CONFIG *second_cfg = &cm->yv12_fb[second_fb_idx];
-      xd->scale_factor[1]    = cm->active_ref_scale[mbmi->ref_frame[1] - 1];
-      xd->scale_factor_uv[1] = cm->active_ref_scale[mbmi->ref_frame[1] - 1];
-      setup_pre_planes(xd, NULL, second_cfg, mi_row, mi_col,
-                       xd->scale_factor, xd->scale_factor_uv);
-      xd->corrupted |= second_cfg->corrupted;
-    }
-  }
+  const YV12_BUFFER_CONFIG *cfg = &cm->yv12_fb[cm->active_ref_idx[ref]];
+  xd->scale_factor[i] = cm->active_ref_scale[ref];
+  setup_pre_planes(xd, i, cfg, mi_row, mi_col, &xd->scale_factor[i]);
+  xd->corrupted |= cfg->corrupted;
 }
 
 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;
+  MB_MODE_INFO *mbmi;
 
-  if (bsize < BLOCK_SIZE_SB8X8)
+  if (less8x8)
     if (xd->ab_index > 0)
       return;
+
   set_offsets(pbi, bsize, mi_row, mi_col);
-  vp9_decode_mb_mode_mv(pbi, xd, mi_row, mi_col, r);
-  set_refs(pbi, mi_row, mi_col);
+  vp9_read_mode_info(pbi, mi_row, mi_col, r);
 
-  if (xd->mode_info_context->mbmi.ref_frame[0] == INTRA_FRAME)
-    decode_sb_intra(pbi, xd, mi_row, mi_col, r, (bsize < BLOCK_SIZE_SB8X8) ?
-                                     BLOCK_SIZE_SB8X8 : bsize);
-  else if (bsize < BLOCK_SIZE_SB8X8)
-    decode_atom(pbi, xd, mi_row, mi_col, r, BLOCK_SIZE_SB8X8);
-  else
-    decode_sb(pbi, xd, mi_row, mi_col, r, bsize);
+  if (less8x8)
+    bsize = BLOCK_SIZE_SB8X8;
 
+  // Has to be called after set_offsets
+  mbmi = &xd->mode_info_context->mbmi;
+
+  if (mbmi->ref_frame[0] == INTRA_FRAME) {
+    // Intra reconstruction
+    decode_tokens(pbi, bsize, r);
+    foreach_transformed_block(xd, bsize, decode_block_intra, xd);
+  } else {
+    // Inter reconstruction
+    int eobtotal;
+
+    set_ref(pbi, 0, mi_row, mi_col);
+    if (mbmi->ref_frame[1] > INTRA_FRAME)
+      set_ref(pbi, 1, mi_row, mi_col);
+
+    vp9_setup_interp_filters(xd, mbmi->interp_filter, cm);
+    vp9_build_inter_predictors_sb(xd, mi_row, mi_col, bsize);
+    eobtotal = decode_tokens(pbi, bsize, r);
+    if (less8x8) {
+      if (eobtotal >= 0)
+        foreach_transformed_block(xd, bsize, decode_block, xd);
+    } else {
+      assert(mbmi->sb_type == bsize);
+      if (eobtotal == 0)
+        // skip loopfilter
+        vp9_set_pred_flag_mbskip(cm, bsize, mi_row, mi_col, 1);
+      else if (eobtotal > 0)
+        foreach_transformed_block(xd, bsize, decode_block, xd);
+    }
+  }
   xd->corrupted |= vp9_reader_has_error(r);
 }
 
@@ -448,16 +280,13 @@ 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_SIZE_SB8X8) {
     if (xd->ab_index != 0)
       return;
-
-  if (bsize >= BLOCK_SIZE_SB8X8) {
+  } else {
     int pl;
-    int idx = check_bsize_coverage(pc, xd, mi_row, mi_col, bsize);
-    // read the partition information
-    xd->left_seg_context = pc->left_seg_context + (mi_row & MI_MASK);
-    xd->above_seg_context = pc->above_seg_context + mi_col;
+    const int idx = check_bsize_coverage(pc, xd, mi_row, mi_col, bsize);
+    set_partition_seg_context(pc, xd, mi_row, mi_col);
     pl = partition_plane_context(xd, bsize);
 
     if (idx == 0)
@@ -469,7 +298,7 @@ static void decode_modes_sb(VP9D_COMP *pbi, int mi_row, int mi_col,
     else
       partition = PARTITION_SPLIT;
 
-    pc->fc.partition_counts[pl][partition]++;
+    pc->counts.partition[pl][partition]++;
   }
 
   subsize = get_subsize(bsize, partition);
@@ -499,8 +328,9 @@ static void decode_modes_sb(VP9D_COMP *pbi, int mi_row, int mi_col,
       }
       break;
     default:
-      assert(0);
+      assert(!"Invalid partition type");
   }
+
   // update partition context
   if (bsize >= BLOCK_SIZE_SB8X8 &&
       (bsize == BLOCK_SIZE_SB8X8 || partition != PARTITION_SPLIT)) {
@@ -527,142 +357,118 @@ static void setup_token_decoder(VP9D_COMP *pbi,
                        "Failed to allocate bool decoder %d", 1);
 }
 
-static void read_coef_probs_common(FRAME_CONTEXT *fc, TX_SIZE tx_size,
+static void read_coef_probs_common(vp9_coeff_probs_model *coef_probs,
                                    vp9_reader *r) {
-  vp9_coeff_probs_model *coef_probs = fc->coef_probs[tx_size];
-
-  if (vp9_read_bit(r)) {
-    int i, j, k, l, m;
-    for (i = 0; i < BLOCK_TYPES; i++) {
-      for (j = 0; j < REF_TYPES; j++) {
-        for (k = 0; k < COEF_BANDS; k++) {
-          for (l = 0; l < PREV_COEF_CONTEXTS; l++) {
-            if (l >= 3 && k == 0)
-              continue;
-
-            for (m = 0; m < UNCONSTRAINED_NODES; m++) {
-              vp9_prob *const p = coef_probs[i][j][k][l] + m;
-
-              if (vp9_read(r, VP9_COEF_UPDATE_PROB))
-                *p = vp9_read_prob_diff_update(r, *p);
-            }
-          }
-        }
-      }
-    }
-  }
-}
+  int i, j, k, l, m;
 
-static void read_coef_probs(VP9D_COMP *pbi, vp9_reader *r) {
-  const TXFM_MODE txfm_mode = pbi->common.txfm_mode;
-  FRAME_CONTEXT *const fc = &pbi->common.fc;
+  if (vp9_read_bit(r))
+    for (i = 0; i < BLOCK_TYPES; i++)
+      for (j = 0; j < REF_TYPES; j++)
+        for (k = 0; k < COEF_BANDS; k++)
+          for (l = 0; l < PREV_COEF_CONTEXTS; l++)
+            if (k > 0 || l < 3)
+              for (m = 0; m < UNCONSTRAINED_NODES; m++)
+                if (vp9_read(r, VP9_COEF_UPDATE_PROB))
+                  vp9_diff_update_prob(r, &coef_probs[i][j][k][l][m]);
+}
 
-  read_coef_probs_common(fc, TX_4X4, r);
+static void read_coef_probs(FRAME_CONTEXT *fc, TX_MODE tx_mode,
+                            vp9_reader *r) {
+  read_coef_probs_common(fc->coef_probs[TX_4X4], r);
 
-  if (txfm_mode > ONLY_4X4)
-    read_coef_probs_common(fc, TX_8X8, r);
+  if (tx_mode > ONLY_4X4)
+    read_coef_probs_common(fc->coef_probs[TX_8X8], r);
 
-  if (txfm_mode > ALLOW_8X8)
-    read_coef_probs_common(fc, TX_16X16, r);
+  if (tx_mode > ALLOW_8X8)
+    read_coef_probs_common(fc->coef_probs[TX_16X16], r);
 
-  if (txfm_mode > ALLOW_16X16)
-    read_coef_probs_common(fc, TX_32X32, r);
+  if (tx_mode > ALLOW_16X16)
+    read_coef_probs_common(fc->coef_probs[TX_32X32], r);
 }
 
-static void setup_segmentation(VP9D_COMP *pbi, struct vp9_read_bit_buffer *rb) {
+static void setup_segmentation(struct segmentation *seg,
+                               struct vp9_read_bit_buffer *rb) {
   int i, j;
 
-  VP9_COMMON *const cm = &pbi->common;
-  MACROBLOCKD *const xd = &pbi->mb;
-
-  xd->update_mb_segmentation_map = 0;
-  xd->update_mb_segmentation_data = 0;
+  seg->update_map = 0;
+  seg->update_data = 0;
 
-  xd->segmentation_enabled = vp9_rb_read_bit(rb);
-  if (!xd->segmentation_enabled)
+  seg->enabled = vp9_rb_read_bit(rb);
+  if (!seg->enabled)
     return;
 
   // Segmentation map update
-  xd->update_mb_segmentation_map = vp9_rb_read_bit(rb);
-  if (xd->update_mb_segmentation_map) {
-    for (i = 0; i < MB_SEG_TREE_PROBS; i++)
-      xd->mb_segment_tree_probs[i] = vp9_rb_read_bit(rb) ?
-                                         vp9_rb_read_literal(rb, 8) : MAX_PROB;
-
-    cm->temporal_update = vp9_rb_read_bit(rb);
-    if (cm->temporal_update) {
+  seg->update_map = vp9_rb_read_bit(rb);
+  if (seg->update_map) {
+    for (i = 0; i < SEG_TREE_PROBS; i++)
+      seg->tree_probs[i] = vp9_rb_read_bit(rb) ? vp9_rb_read_literal(rb, 8)
+                                               : MAX_PROB;
+
+    seg->temporal_update = vp9_rb_read_bit(rb);
+    if (seg->temporal_update) {
       for (i = 0; i < PREDICTION_PROBS; i++)
-        cm->segment_pred_probs[i] = vp9_rb_read_bit(rb) ?
-                                        vp9_rb_read_literal(rb, 8) : MAX_PROB;
+        seg->pred_probs[i] = vp9_rb_read_bit(rb) ? vp9_rb_read_literal(rb, 8)
+                                                 : MAX_PROB;
     } else {
       for (i = 0; i < PREDICTION_PROBS; i++)
-        cm->segment_pred_probs[i] = MAX_PROB;
+        seg->pred_probs[i] = MAX_PROB;
     }
   }
 
   // Segmentation data update
-  xd->update_mb_segmentation_data = vp9_rb_read_bit(rb);
-  if (xd->update_mb_segmentation_data) {
-    xd->mb_segment_abs_delta = vp9_rb_read_bit(rb);
+  seg->update_data = vp9_rb_read_bit(rb);
+  if (seg->update_data) {
+    seg->abs_delta = vp9_rb_read_bit(rb);
 
-    vp9_clearall_segfeatures(xd);
+    vp9_clearall_segfeatures(seg);
 
-    for (i = 0; i < MAX_MB_SEGMENTS; i++) {
+    for (i = 0; i < MAX_SEGMENTS; i++) {
       for (j = 0; j < SEG_LVL_MAX; j++) {
         int data = 0;
         const int feature_enabled = vp9_rb_read_bit(rb);
         if (feature_enabled) {
-          vp9_enable_segfeature(xd, i, j);
+          vp9_enable_segfeature(seg, i, j);
           data = decode_unsigned_max(rb, vp9_seg_feature_data_max(j));
           if (vp9_is_segfeature_signed(j))
             data = vp9_rb_read_bit(rb) ? -data : data;
         }
-        vp9_set_segdata(xd, i, j, data);
+        vp9_set_segdata(seg, i, j, data);
       }
     }
   }
 }
 
-static void setup_loopfilter(VP9D_COMP *pbi, struct vp9_read_bit_buffer *rb) {
-  VP9_COMMON *const cm = &pbi->common;
-  MACROBLOCKD *const xd = &pbi->mb;
+static void setup_loopfilter(struct loopfilter *lf,
+                             struct vp9_read_bit_buffer *rb) {
 
-  cm->filter_level = vp9_rb_read_literal(rb, 6);
-  cm->sharpness_level = vp9_rb_read_literal(rb, 3);
+  lf->filter_level = vp9_rb_read_literal(rb, 6);
+  lf->sharpness_level = vp9_rb_read_literal(rb, 3);
 
   // Read in loop filter deltas applied at the MB level based on mode or ref
   // frame.
-  xd->mode_ref_lf_delta_update = 0;
+  lf->mode_ref_delta_update = 0;
 
-  xd->mode_ref_lf_delta_enabled = vp9_rb_read_bit(rb);
-  if (xd->mode_ref_lf_delta_enabled) {
-    xd->mode_ref_lf_delta_update = vp9_rb_read_bit(rb);
-    if (xd->mode_ref_lf_delta_update) {
+  lf->mode_ref_delta_enabled = vp9_rb_read_bit(rb);
+  if (lf->mode_ref_delta_enabled) {
+    lf->mode_ref_delta_update = vp9_rb_read_bit(rb);
+    if (lf->mode_ref_delta_update) {
       int i;
 
-      for (i = 0; i < MAX_REF_LF_DELTAS; i++) {
-        if (vp9_rb_read_bit(rb)) {
-          const int value = vp9_rb_read_literal(rb, 6);
-          xd->ref_lf_deltas[i] = vp9_rb_read_bit(rb) ? -value : value;
-        }
-      }
+      for (i = 0; i < MAX_REF_LF_DELTAS; i++)
+        if (vp9_rb_read_bit(rb))
+          lf->ref_deltas[i] = vp9_rb_read_signed_literal(rb, 6);
 
-      for (i = 0; i < MAX_MODE_LF_DELTAS; i++) {
-        if (vp9_rb_read_bit(rb)) {
-          const int value = vp9_rb_read_literal(rb, 6);
-          xd->mode_lf_deltas[i] = vp9_rb_read_bit(rb) ? -value : value;
-        }
-      }
+      for (i = 0; i < MAX_MODE_LF_DELTAS; i++)
+        if (vp9_rb_read_bit(rb))
+          lf->mode_deltas[i] = vp9_rb_read_signed_literal(rb, 6);
     }
   }
 }
 
 static int read_delta_q(struct vp9_read_bit_buffer *rb, int *delta_q) {
   const int old = *delta_q;
-  if (vp9_rb_read_bit(rb)) {
-    const int value = vp9_rb_read_literal(rb, 4);
-    *delta_q = vp9_rb_read_bit(rb) ? -value : value;
-  }
+  if (vp9_rb_read_bit(rb))
+    *delta_q = vp9_rb_read_signed_literal(rb, 4);
   return old != *delta_q;
 }
 
@@ -682,11 +488,9 @@ static void setup_quantization(VP9D_COMP *pbi, struct vp9_read_bit_buffer *rb) {
                  cm->y_dc_delta_q == 0 &&
                  cm->uv_dc_delta_q == 0 &&
                  cm->uv_ac_delta_q == 0;
-  if (xd->lossless) {
-    xd->itxm_add          = vp9_idct_add_lossless_c;
-  } else {
-    xd->itxm_add          = vp9_idct_add;
-  }
+
+  xd->itxm_add = xd->lossless ? vp9_idct_add_lossless_c
+                              : vp9_idct_add;
 }
 
 static INTERPOLATIONFILTERTYPE read_interp_filter_type(
@@ -778,108 +582,90 @@ static void setup_frame_size_with_refs(VP9D_COMP *pbi,
   apply_frame_size(pbi, width, height);
 }
 
-static void update_frame_context(FRAME_CONTEXT *fc) {
-  vp9_copy(fc->pre_coef_probs, fc->coef_probs);
-  vp9_copy(fc->pre_y_mode_prob, fc->y_mode_prob);
-  vp9_copy(fc->pre_uv_mode_prob, fc->uv_mode_prob);
-  vp9_copy(fc->pre_partition_prob, fc->partition_prob[1]);
-  vp9_copy(fc->pre_intra_inter_prob, fc->intra_inter_prob);
-  vp9_copy(fc->pre_comp_inter_prob, fc->comp_inter_prob);
-  vp9_copy(fc->pre_single_ref_prob, fc->single_ref_prob);
-  vp9_copy(fc->pre_comp_ref_prob, fc->comp_ref_prob);
-  fc->pre_nmvc = fc->nmvc;
-  vp9_copy(fc->pre_switchable_interp_prob, fc->switchable_interp_prob);
-  vp9_copy(fc->pre_inter_mode_probs, fc->inter_mode_probs);
-  vp9_copy(fc->pre_tx_probs_8x8p, fc->tx_probs_8x8p);
-  vp9_copy(fc->pre_tx_probs_16x16p, fc->tx_probs_16x16p);
-  vp9_copy(fc->pre_tx_probs_32x32p, fc->tx_probs_32x32p);
-  vp9_copy(fc->pre_mbskip_probs, fc->mbskip_probs);
-
-  vp9_zero(fc->coef_counts);
-  vp9_zero(fc->eob_branch_counts);
-  vp9_zero(fc->y_mode_counts);
-  vp9_zero(fc->uv_mode_counts);
-  vp9_zero(fc->NMVcount);
-  vp9_zero(fc->inter_mode_counts);
-  vp9_zero(fc->partition_counts);
-  vp9_zero(fc->switchable_interp_count);
-  vp9_zero(fc->intra_inter_count);
-  vp9_zero(fc->comp_inter_count);
-  vp9_zero(fc->single_ref_count);
-  vp9_zero(fc->comp_ref_count);
-  vp9_zero(fc->tx_count_8x8p);
-  vp9_zero(fc->tx_count_16x16p);
-  vp9_zero(fc->tx_count_32x32p);
-  vp9_zero(fc->mbskip_count);
-}
-
 static void decode_tile(VP9D_COMP *pbi, vp9_reader *r) {
   VP9_COMMON *const pc = &pbi->common;
   int mi_row, mi_col;
 
-  for (mi_row = pc->cur_tile_mi_row_start;
-       mi_row < pc->cur_tile_mi_row_end; mi_row += 64 / MI_SIZE) {
+  if (pbi->do_loopfilter_inline) {
+    vp9_loop_filter_frame_init(pc, &pbi->mb, pbi->mb.lf.filter_level);
+  }
+
+  for (mi_row = pc->cur_tile_mi_row_start; mi_row < pc->cur_tile_mi_row_end;
+       mi_row += MI_BLOCK_SIZE) {
     // For a SB there are 2 left contexts, each pertaining to a MB row within
     vpx_memset(&pc->left_context, 0, sizeof(pc->left_context));
     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 += 64 / MI_SIZE)
+    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);
+    }
+
+    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 (pbi->do_loopfilter_inline) {
+    YV12_BUFFER_CONFIG *const fb = &pbi->common.yv12_fb[pbi->common.new_fb_idx];
+    vp9_loop_filter_rows(fb, pc, &pbi->mb,
+                         mi_row - MI_BLOCK_SIZE, pc->mi_rows, 0);
   }
 }
 
 static void setup_tile_info(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
-  int delta_log2_tiles;
+  int min_log2_tile_cols, max_log2_tile_cols, max_ones;
+  vp9_get_tile_n_bits(cm->mi_cols, &min_log2_tile_cols, &max_log2_tile_cols);
 
-  vp9_get_tile_n_bits(cm, &cm->log2_tile_columns, &delta_log2_tiles);
-  while (delta_log2_tiles--) {
-    if (vp9_rb_read_bit(rb)) {
-      cm->log2_tile_columns++;
-    } else {
-      break;
-    }
-  }
+  // columns
+  max_ones = max_log2_tile_cols - min_log2_tile_cols;
+  cm->log2_tile_cols = min_log2_tile_cols;
+  while (max_ones-- && vp9_rb_read_bit(rb))
+    cm->log2_tile_cols++;
 
+  // rows
   cm->log2_tile_rows = vp9_rb_read_bit(rb);
   if (cm->log2_tile_rows)
     cm->log2_tile_rows += vp9_rb_read_bit(rb);
-
-  cm->tile_columns = 1 << cm->log2_tile_columns;
-  cm->tile_rows    = 1 << cm->log2_tile_rows;
 }
 
-static void decode_tiles(VP9D_COMP *pbi,
-                         const uint8_t *data, size_t first_partition_size,
-                         vp9_reader *residual_bc) {
+static const uint8_t *decode_tiles(VP9D_COMP *pbi, const uint8_t *data) {
+  vp9_reader residual_bc;
+
   VP9_COMMON *const pc = &pbi->common;
 
-  const uint8_t *data_ptr = data + first_partition_size;
-  const uint8_t* const data_end = pbi->source + pbi->source_sz;
+  const uint8_t *const data_end = pbi->source + pbi->source_sz;
+  const int aligned_mi_cols = mi_cols_aligned_to_sb(pc->mi_cols);
+  const int tile_cols = 1 << pc->log2_tile_cols;
+  const int tile_rows = 1 << pc->log2_tile_rows;
   int tile_row, tile_col;
 
   // Note: this memset assumes above_context[0], [1] and [2]
   // are allocated as part of the same buffer.
-  vpx_memset(pc->above_context[0], 0, sizeof(ENTROPY_CONTEXT) * 2 *
-                                      MAX_MB_PLANE * mi_cols_aligned_to_sb(pc));
+  vpx_memset(pc->above_context[0], 0,
+             sizeof(ENTROPY_CONTEXT) * 2 * MAX_MB_PLANE * aligned_mi_cols);
 
-  vpx_memset(pc->above_seg_context, 0, sizeof(PARTITION_CONTEXT) *
-                                       mi_cols_aligned_to_sb(pc));
+  vpx_memset(pc->above_seg_context, 0,
+             sizeof(PARTITION_CONTEXT) * aligned_mi_cols);
 
   if (pbi->oxcf.inv_tile_order) {
-    const int n_cols = pc->tile_columns;
     const uint8_t *data_ptr2[4][1 << 6];
     vp9_reader bc_bak = {0};
 
     // pre-initialize the offsets, we're going to read in inverse order
-    data_ptr2[0][0] = data_ptr;
-    for (tile_row = 0; tile_row < pc->tile_rows; tile_row++) {
+    data_ptr2[0][0] = data;
+    for (tile_row = 0; tile_row < tile_rows; tile_row++) {
       if (tile_row) {
-        const int size = read_be32(data_ptr2[tile_row - 1][n_cols - 1]);
-        data_ptr2[tile_row - 1][n_cols - 1] += 4;
-        data_ptr2[tile_row][0] = data_ptr2[tile_row - 1][n_cols - 1] + size;
+        const int size = read_be32(data_ptr2[tile_row - 1][tile_cols - 1]);
+        data_ptr2[tile_row - 1][tile_cols - 1] += 4;
+        data_ptr2[tile_row][0] = data_ptr2[tile_row - 1][tile_cols - 1] + size;
       }
 
-      for (tile_col = 1; tile_col < n_cols; tile_col++) {
+      for (tile_col = 1; tile_col < tile_cols; tile_col++) {
         const int size = read_be32(data_ptr2[tile_row][tile_col - 1]);
         data_ptr2[tile_row][tile_col - 1] += 4;
         data_ptr2[tile_row][tile_col] =
@@ -887,48 +673,49 @@ static void decode_tiles(VP9D_COMP *pbi,
       }
     }
 
-    for (tile_row = 0; tile_row < pc->tile_rows; tile_row++) {
+    for (tile_row = 0; tile_row < tile_rows; tile_row++) {
       vp9_get_tile_row_offsets(pc, tile_row);
-      for (tile_col = n_cols - 1; tile_col >= 0; tile_col--) {
+      for (tile_col = tile_cols - 1; tile_col >= 0; tile_col--) {
         vp9_get_tile_col_offsets(pc, tile_col);
         setup_token_decoder(pbi, data_ptr2[tile_row][tile_col],
                             data_end - data_ptr2[tile_row][tile_col],
-                            residual_bc);
-        decode_tile(pbi, residual_bc);
-        if (tile_row == pc->tile_rows - 1 && tile_col == n_cols - 1)
-          bc_bak = *residual_bc;
+                            &residual_bc);
+        decode_tile(pbi, &residual_bc);
+        if (tile_row == tile_rows - 1 && tile_col == tile_cols - 1)
+          bc_bak = residual_bc;
       }
     }
-    *residual_bc = bc_bak;
+    residual_bc = bc_bak;
   } else {
     int has_more;
 
-    for (tile_row = 0; tile_row < pc->tile_rows; tile_row++) {
+    for (tile_row = 0; tile_row < tile_rows; tile_row++) {
       vp9_get_tile_row_offsets(pc, tile_row);
-      for (tile_col = 0; tile_col < pc->tile_columns; tile_col++) {
+      for (tile_col = 0; tile_col < tile_cols; tile_col++) {
         size_t size;
 
         vp9_get_tile_col_offsets(pc, tile_col);
 
-        has_more = tile_col < pc->tile_columns - 1 ||
-                   tile_row < pc->tile_rows - 1;
+        has_more = tile_col < tile_cols - 1 || tile_row < tile_rows - 1;
         if (has_more) {
-          if (!read_is_valid(data_ptr, 4, data_end))
+          if (!read_is_valid(data, 4, data_end))
             vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
                          "Truncated packet or corrupt tile length");
 
-          size = read_be32(data_ptr);
-          data_ptr += 4;
+          size = read_be32(data);
+          data += 4;
         } else {
-          size = data_end - data_ptr;
+          size = data_end - data;
         }
 
-        setup_token_decoder(pbi, data_ptr, size, residual_bc);
-        decode_tile(pbi, residual_bc);
-        data_ptr += size;
+        setup_token_decoder(pbi, data, size, &residual_bc);
+        decode_tile(pbi, &residual_bc);
+        data += size;
       }
     }
   }
+
+  return vp9_reader_find_end(&residual_bc);
 }
 
 static void check_sync_code(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
@@ -949,10 +736,9 @@ static void setup_inter_inter(VP9_COMMON *cm) {
   int i;
 
   cm->allow_comp_inter_inter = 0;
-  for (i = 0; i < ALLOWED_REFS_PER_FRAME; ++i) {
-    cm->allow_comp_inter_inter |= i > 0 &&
+  for (i = 1; i < ALLOWED_REFS_PER_FRAME; ++i)
+    cm->allow_comp_inter_inter |=
         cm->ref_frame_sign_bias[i + 1] != cm->ref_frame_sign_bias[1];
-  }
 
   if (cm->allow_comp_inter_inter) {
     // which one is always-on in comp inter-inter?
@@ -999,7 +785,7 @@ static size_t read_uncompressed_header(VP9D_COMP *pbi,
     int frame_to_show = cm->ref_frame_map[vp9_rb_read_literal(rb, 3)];
     ref_cnt_fb(cm->fb_idx_ref_cnt, &cm->new_fb_idx, frame_to_show);
     pbi->refresh_frame_flags = 0;
-    cm->filter_level = 0;
+    xd->lf.filter_level = 0;
     return 0;
   }
 
@@ -1053,7 +839,7 @@ static size_t read_uncompressed_header(VP9D_COMP *pbi,
        pbi->refresh_frame_flags = vp9_rb_read_literal(rb, NUM_REF_FRAMES);
 
       for (i = 0; i < ALLOWED_REFS_PER_FRAME; ++i) {
-        const int ref = vp9_rb_read_literal(rb, NUM_REF_FRAMES_LG2);
+        const int ref = vp9_rb_read_literal(rb, NUM_REF_FRAMES_LOG2);
         cm->active_ref_idx[i] = cm->ref_frame_map[ref];
         cm->ref_frame_sign_bias[LAST_FRAME + i] = vp9_rb_read_bit(rb);
       }
@@ -1078,23 +864,54 @@ static size_t read_uncompressed_header(VP9D_COMP *pbi,
     cm->frame_parallel_decoding_mode = 1;
   }
 
-  cm->frame_context_idx = vp9_rb_read_literal(rb, NUM_FRAME_CONTEXTS_LG2);
+  cm->frame_context_idx = vp9_rb_read_literal(rb, NUM_FRAME_CONTEXTS_LOG2);
 
   if (cm->frame_type == KEY_FRAME || cm->error_resilient_mode || cm->intra_only)
     vp9_setup_past_independence(cm, xd);
 
-  setup_loopfilter(pbi, rb);
+  setup_loopfilter(&xd->lf, rb);
   setup_quantization(pbi, rb);
-  setup_segmentation(pbi, rb);
+  setup_segmentation(&xd->seg, rb);
 
   setup_tile_info(cm, rb);
 
   return vp9_rb_read_literal(rb, 16);
 }
 
+static int read_compressed_header(VP9D_COMP *pbi, const uint8_t *data,
+                                  size_t partition_size) {
+  VP9_COMMON *const cm = &pbi->common;
+  MACROBLOCKD *const xd = &pbi->mb;
+  vp9_reader r;
+
+  if (vp9_reader_init(&r, data, partition_size))
+    vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
+                       "Failed to allocate bool decoder 0");
+
+  cm->tx_mode = xd->lossless ? ONLY_4X4 : read_tx_mode(&r);
+  if (cm->tx_mode == TX_MODE_SELECT)
+    read_tx_probs(&cm->fc.tx_probs, &r);
+  read_coef_probs(&cm->fc, cm->tx_mode, &r);
+
+  vp9_prepare_read_mode_info(pbi, &r);
+
+  return vp9_reader_has_error(&r);
+}
+
+void vp9_init_dequantizer(VP9_COMMON *cm) {
+  int q;
+
+  for (q = 0; q < QINDEX_RANGE; q++) {
+    cm->y_dequant[q][0] = vp9_dc_quant(q, cm->y_dc_delta_q);
+    cm->y_dequant[q][1] = vp9_ac_quant(q, 0);
+
+    cm->uv_dequant[q][0] = vp9_dc_quant(q, cm->uv_dc_delta_q);
+    cm->uv_dequant[q][1] = vp9_ac_quant(q, cm->uv_ac_delta_q);
+  }
+}
+
 int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) {
   int i;
-  vp9_reader header_bc, residual_bc;
   VP9_COMMON *const pc = &pbi->common;
   MACROBLOCKD *const xd = &pbi->mb;
 
@@ -1115,6 +932,8 @@ int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) {
   data += vp9_rb_bytes_read(&rb);
   xd->corrupted = 0;
   new_fb->corrupted = 0;
+  pbi->do_loopfilter_inline =
+      (pc->log2_tile_rows | pc->log2_tile_cols) == 0 && pbi->mb.lf.filter_level;
 
   if (!pbi->decoded_key_frame && !keyframe)
     return -1;
@@ -1125,37 +944,29 @@ int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) {
 
   xd->mode_info_context = pc->mi;
   xd->prev_mode_info_context = pc->prev_mi;
-  xd->frame_type = pc->frame_type;
   xd->mode_info_stride = pc->mode_info_stride;
 
-  if (vp9_reader_init(&header_bc, data, first_partition_size))
-    vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR,
-                       "Failed to allocate bool decoder 0");
-
-  mb_init_dequantizer(pc, &pbi->mb);  // MB level dequantizer setup
+  init_dequantizer(pc, &pbi->mb);
 
   if (!keyframe)
     vp9_setup_interp_filters(xd, pc->mcomp_filter_type, pc);
 
   pc->fc = pc->frame_contexts[pc->frame_context_idx];
 
-  update_frame_context(&pc->fc);
-
-  setup_txfm_mode(pc, xd->lossless, &header_bc);
-
-  read_coef_probs(pbi, &header_bc);
+  vp9_zero(pc->counts);
 
   // Initialize xd pointers. Any reference should do for xd->pre, so use 0.
-  setup_pre_planes(xd, &pc->yv12_fb[pc->active_ref_idx[0]], NULL,
-                   0, 0, NULL, NULL);
+  setup_pre_planes(xd, 0, &pc->yv12_fb[pc->active_ref_idx[0]], 0, 0, NULL);
   setup_dst_planes(xd, new_fb, 0, 0);
 
+  new_fb->corrupted |= read_compressed_header(pbi, data, first_partition_size);
+
   // Create the segmentation map structure and set to 0
   if (!pc->last_frame_seg_map)
-    CHECK_MEM_ERROR(pc->last_frame_seg_map,
+    CHECK_MEM_ERROR(pc, pc->last_frame_seg_map,
                     vpx_calloc((pc->mi_rows * pc->mi_cols), 1));
 
-  vp9_setup_block_dptrs(xd, pc->subsampling_x, pc->subsampling_y);
+  setup_block_dptrs(xd, pc->subsampling_x, pc->subsampling_y);
 
   // clear out the coeff buffer
   for (i = 0; i < MAX_MB_PLANE; ++i)
@@ -1163,14 +974,12 @@ int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) {
 
   set_prev_mi(pc);
 
-  vp9_decode_mode_mvs_init(pbi, &header_bc);
-
-  decode_tiles(pbi, data, first_partition_size, &residual_bc);
+  *p_data_end = decode_tiles(pbi, data + first_partition_size);
 
   pc->last_width = pc->width;
   pc->last_height = pc->height;
 
-  new_fb->corrupted = vp9_reader_has_error(&header_bc) | xd->corrupted;
+  new_fb->corrupted |= xd->corrupted;
 
   if (!pbi->decoded_key_frame) {
     if (keyframe && !new_fb->corrupted)
@@ -1180,20 +989,18 @@ int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) {
                          "A stream must start with a complete key frame");
   }
 
-  // Adaptation
   if (!pc->error_resilient_mode && !pc->frame_parallel_decoding_mode) {
     vp9_adapt_coef_probs(pc);
 
-    if ((!keyframe) && (!pc->intra_only)) {
+    if (!keyframe && !pc->intra_only) {
       vp9_adapt_mode_probs(pc);
       vp9_adapt_mode_context(pc);
-      vp9_adapt_nmv_probs(pc, xd->allow_high_precision_mv);
+      vp9_adapt_mv_probs(pc, xd->allow_high_precision_mv);
     }
   }
 
   if (pc->refresh_frame_context)
     pc->frame_contexts[pc->frame_context_idx] = pc->fc;
 
-  *p_data_end = vp9_reader_find_end(&residual_bc);
   return 0;
 }
diff --git a/libvpx/vp9/decoder/vp9_decodframe.h b/libvpx/vp9/decoder/vp9_decodframe.h
index 66e951d..00b6d67 100644
--- a/libvpx/vp9/decoder/vp9_decodframe.h
+++ b/libvpx/vp9/decoder/vp9_decodframe.h
@@ -17,6 +17,5 @@ struct VP9Decompressor;
 
 void vp9_init_dequantizer(struct VP9Common *pc);
 int vp9_decode_frame(struct VP9Decompressor *cpi, const uint8_t **p_data_end);
-vp9_prob vp9_read_prob_diff_update(vp9_reader *r, int oldp);
 
 #endif  // VP9_DECODER_VP9_DECODFRAME_H_
diff --git a/libvpx/vp9/decoder/vp9_detokenize.c b/libvpx/vp9/decoder/vp9_detokenize.c
index 3bbb212..01c1db0 100644
--- a/libvpx/vp9/decoder/vp9_detokenize.c
+++ b/libvpx/vp9/decoder/vp9_detokenize.c
@@ -18,14 +18,8 @@
 #include "vp9/decoder/vp9_detokenize.h"
 #include "vp9/decoder/vp9_onyxd_int.h"
 
-#if CONFIG_BALANCED_COEFTREE
-#define ZERO_CONTEXT_NODE           0
-#define EOB_CONTEXT_NODE            1
-#else
 #define EOB_CONTEXT_NODE            0
 #define ZERO_CONTEXT_NODE           1
-#endif
-
 #define ONE_CONTEXT_NODE            2
 #define LOW_VAL_CONTEXT_NODE        3
 #define TWO_CONTEXT_NODE            4
@@ -91,13 +85,15 @@ DECLARE_ALIGNED(16, extern const uint8_t,
       val += 1 << bits_count;          \
   } while (0);
 
-static int decode_coefs(FRAME_CONTEXT *fc, const MACROBLOCKD *xd,
+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,
                         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, pad, default_eob;
+  int pt, c = 0;
   int band;
   vp9_prob (*coef_probs)[PREV_COEF_CONTEXTS][UNCONSTRAINED_NODES];
   vp9_prob coef_probs_full[COEF_BANDS][PREV_COEF_CONTEXTS][ENTROPY_NODES];
@@ -113,53 +109,31 @@ static int decode_coefs(FRAME_CONTEXT *fc, const MACROBLOCKD *xd,
   vp9_prob *prob;
   vp9_coeff_count_model *coef_counts;
   const int ref = xd->mode_info_context->mbmi.ref_frame[0] != INTRA_FRAME;
-  TX_TYPE tx_type = DCT_DCT;
-  const int *scan, *nb;
+  const int16_t *scan, *nb;
   uint8_t token_cache[1024];
   const uint8_t * band_translate;
-#if CONFIG_BALANCED_COEFTREE
-  int skip_eob_node = 0;
-#endif
-
   coef_probs  = fc->coef_probs[txfm_size][type][ref];
-  coef_counts = fc->coef_counts[txfm_size];
+  coef_counts = counts->coef[txfm_size];
   switch (txfm_size) {
     default:
     case TX_4X4: {
-      tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
-          get_tx_type_4x4(xd, block_idx) : DCT_DCT;
-      scan = get_scan_4x4(tx_type);
+      scan = get_scan_4x4(get_tx_type_4x4(type, xd, block_idx));
       above_ec = A[0] != 0;
       left_ec = L[0] != 0;
-      default_eob = 16;
       band_translate = vp9_coefband_trans_4x4;
       break;
     }
     case TX_8X8: {
-      const BLOCK_SIZE_TYPE sb_type = xd->mode_info_context->mbmi.sb_type;
-      const int sz = 1 + b_width_log2(sb_type);
-      const int x = block_idx & ((1 << sz) - 1);
-      const int y = block_idx - x;
-      tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
-          get_tx_type_8x8(xd, y + (x >> 1)) : DCT_DCT;
-      scan = get_scan_8x8(tx_type);
+      scan = get_scan_8x8(get_tx_type_8x8(type, xd));
       above_ec = (A[0] + A[1]) != 0;
       left_ec = (L[0] + L[1]) != 0;
-      default_eob = 64;
       band_translate = vp9_coefband_trans_8x8plus;
       break;
     }
     case TX_16X16: {
-      const BLOCK_SIZE_TYPE sb_type = xd->mode_info_context->mbmi.sb_type;
-      const int sz = 2 + b_width_log2(sb_type);
-      const int x = block_idx & ((1 << sz) - 1);
-      const int y = block_idx - x;
-      tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
-          get_tx_type_16x16(xd, y + (x >> 2)) : DCT_DCT;
-      scan = get_scan_16x16(tx_type);
+      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;
-      default_eob = 256;
       band_translate = vp9_coefband_trans_8x8plus;
       break;
     }
@@ -167,13 +141,12 @@ static int decode_coefs(FRAME_CONTEXT *fc, const MACROBLOCKD *xd,
       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;
-      default_eob = 1024;
       band_translate = vp9_coefband_trans_8x8plus;
       break;
   }
 
   pt = combine_entropy_contexts(above_ec, left_ec);
-  nb = vp9_get_coef_neighbors_handle(scan, &pad);
+  nb = vp9_get_coef_neighbors_handle(scan);
 
   while (1) {
     int val;
@@ -181,43 +154,26 @@ static int decode_coefs(FRAME_CONTEXT *fc, const MACROBLOCKD *xd,
     if (c >= seg_eob)
       break;
     if (c)
-      pt = vp9_get_coef_context(scan, nb, pad, token_cache,
-                                c, default_eob);
+      pt = get_coef_context(nb, token_cache, c);
     band = get_coef_band(band_translate, c);
     prob = coef_probs[band][pt];
-#if !CONFIG_BALANCED_COEFTREE
-    fc->eob_branch_counts[txfm_size][type][ref][band][pt]++;
+    counts->eob_branch[txfm_size][type][ref][band][pt]++;
     if (!vp9_read(r, prob[EOB_CONTEXT_NODE]))
       break;
 
 SKIP_START:
-#endif
     if (c >= seg_eob)
       break;
     if (c)
-      pt = vp9_get_coef_context(scan, nb, pad, token_cache,
-                                c, default_eob);
+      pt = get_coef_context(nb, token_cache, c);
     band = get_coef_band(band_translate, c);
     prob = coef_probs[band][pt];
 
     if (!vp9_read(r, prob[ZERO_CONTEXT_NODE])) {
       INCREMENT_COUNT(ZERO_TOKEN);
       ++c;
-#if CONFIG_BALANCED_COEFTREE
-      skip_eob_node = 1;
-      continue;
-#else
       goto SKIP_START;
-#endif
-    }
-#if CONFIG_BALANCED_COEFTREE
-    if (!skip_eob_node) {
-      fc->eob_branch_counts[txfm_size][type][ref][band][pt]++;
-      if (!vp9_read(r, prob[EOB_CONTEXT_NODE]))
-        break;
     }
-    skip_eob_node = 0;
-#endif
 
     // ONE_CONTEXT_NODE_0_
     if (!vp9_read(r, prob[ONE_CONTEXT_NODE])) {
@@ -293,8 +249,8 @@ SKIP_START:
   return c;
 }
 
-static int get_eob(MACROBLOCKD* const xd, int segment_id, int eob_max) {
-  return vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP) ? 0 : eob_max;
+static int get_eob(struct segmentation *seg, int segment_id, int eob_max) {
+  return vp9_segfeature_active(seg, segment_id, SEG_LVL_SKIP) ? 0 : eob_max;
 }
 
 struct decode_block_args {
@@ -315,7 +271,7 @@ static void decode_block(int plane, int block,
   struct macroblockd_plane* pd = &xd->plane[plane];
   const int segment_id = xd->mode_info_context->mbmi.segment_id;
   const TX_SIZE ss_tx_size = ss_txfrm_size / 2;
-  const int seg_eob = get_eob(xd, segment_id, 16 << ss_txfrm_size);
+  const int seg_eob = get_eob(&xd->seg, segment_id, 16 << ss_txfrm_size);
   const int off = block >> ss_txfrm_size;
   const int mod = bw - ss_tx_size - pd->subsampling_x;
   const int aoff = (off & ((1 << mod) - 1)) << ss_tx_size;
@@ -323,7 +279,7 @@ static void decode_block(int plane, int block,
 
   ENTROPY_CONTEXT *A = pd->above_context + aoff;
   ENTROPY_CONTEXT *L = pd->left_context + loff;
-  const int eob = decode_coefs(&arg->pbi->common.fc, xd, arg->r, block,
+  const int eob = decode_coefs(&arg->pbi->common, xd, arg->r, block,
                                pd->plane_type, seg_eob,
                                BLOCK_OFFSET(pd->qcoeff, block, 16),
                                ss_tx_size, pd->dequant, A, L);
diff --git a/libvpx/vp9/decoder/vp9_dsubexp.c b/libvpx/vp9/decoder/vp9_dsubexp.c
new file mode 100644
index 0000000..8cc64f7
--- /dev/null
+++ b/libvpx/vp9/decoder/vp9_dsubexp.c
@@ -0,0 +1,106 @@
+/*
+  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vp9/common/vp9_entropy.h"
+
+#include "vp9/decoder/vp9_dsubexp.h"
+
+static int inv_recenter_nonneg(int v, int m) {
+  if (v > 2 * m)
+    return v;
+
+  return v % 2 ? m - (v + 1) / 2 : m + v / 2;
+}
+
+static int decode_uniform(vp9_reader *r, int n) {
+  int v;
+  const int l = get_unsigned_bits(n);
+  const int m = (1 << l) - n;
+  if (!l)
+    return 0;
+
+  v = vp9_read_literal(r, l - 1);
+  return v < m ?  v : (v << 1) - m + vp9_read_bit(r);
+}
+
+
+static int merge_index(int v, int n, int modulus) {
+  int max1 = (n - 1 - modulus / 2) / modulus + 1;
+  if (v < max1) {
+    v = v * modulus + modulus / 2;
+  } else {
+    int w;
+    v -= max1;
+    w = v;
+    v += (v + modulus - modulus / 2) / modulus;
+    while (v % modulus == modulus / 2 ||
+           w != v - (v + modulus - modulus / 2) / modulus) v++;
+  }
+  return v;
+}
+
+static int inv_remap_prob(int v, int m) {
+  static int inv_map_table[MAX_PROB - 1] = {
+    // generated by:
+    //   inv_map_table[j] = merge_index(j, MAX_PROB - 1, MODULUS_PARAM);
+      6,  19,  32,  45,  58,  71,  84,  97, 110, 123, 136, 149, 162, 175, 188,
+    201, 214, 227, 240, 253,   0,   1,   2,   3,   4,   5,   7,   8,   9,  10,
+     11,  12,  13,  14,  15,  16,  17,  18,  20,  21,  22,  23,  24,  25,  26,
+     27,  28,  29,  30,  31,  33,  34,  35,  36,  37,  38,  39,  40,  41,  42,
+     43,  44,  46,  47,  48,  49,  50,  51,  52,  53,  54,  55,  56,  57,  59,
+     60,  61,  62,  63,  64,  65,  66,  67,  68,  69,  70,  72,  73,  74,  75,
+     76,  77,  78,  79,  80,  81,  82,  83,  85,  86,  87,  88,  89,  90,  91,
+     92,  93,  94,  95,  96,  98,  99, 100, 101, 102, 103, 104, 105, 106, 107,
+    108, 109, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 124,
+    125, 126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 137, 138, 139, 140,
+    141, 142, 143, 144, 145, 146, 147, 148, 150, 151, 152, 153, 154, 155, 156,
+    157, 158, 159, 160, 161, 163, 164, 165, 166, 167, 168, 169, 170, 171, 172,
+    173, 174, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 189,
+    190, 191, 192, 193, 194, 195, 196, 197, 198, 199, 200, 202, 203, 204, 205,
+    206, 207, 208, 209, 210, 211, 212, 213, 215, 216, 217, 218, 219, 220, 221,
+    222, 223, 224, 225, 226, 228, 229, 230, 231, 232, 233, 234, 235, 236, 237,
+    238, 239, 241, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252,
+
+  };
+  // v = merge_index(v, MAX_PROBS - 1, MODULUS_PARAM);
+  v = inv_map_table[v];
+  m--;
+  if ((m << 1) <= MAX_PROB) {
+    return 1 + inv_recenter_nonneg(v + 1, m);
+  } else {
+    return MAX_PROB - inv_recenter_nonneg(v + 1, MAX_PROB - 1 - m);
+  }
+}
+
+static int decode_term_subexp(vp9_reader *r, int k, int num_syms) {
+  int i = 0, mk = 0, word;
+  while (1) {
+    const int b = i ? k + i - 1 : k;
+    const int a = 1 << b;
+    if (num_syms <= mk + 3 * a) {
+      word = decode_uniform(r, num_syms - mk) + mk;
+      break;
+    } else {
+      if (vp9_read_bit(r)) {
+        i++;
+        mk += a;
+      } else {
+        word = vp9_read_literal(r, b) + mk;
+        break;
+      }
+    }
+  }
+  return word;
+}
+
+void vp9_diff_update_prob(vp9_reader *r, vp9_prob* p) {
+  int delp = decode_term_subexp(r, SUBEXP_PARAM, 255);
+  *p = (vp9_prob)inv_remap_prob(delp, *p);
+}
diff --git a/libvpx/vp9/decoder/vp9_asm_dec_offsets.c b/libvpx/vp9/decoder/vp9_dsubexp.h
index e4b9c97..aeb9399 100644
--- a/libvpx/vp9/decoder/vp9_asm_dec_offsets.c
+++ b/libvpx/vp9/decoder/vp9_dsubexp.h
@@ -1,5 +1,5 @@
 /*
- *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *  Copyright (c) 2013 The WebM project authors. All Rights Reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
@@ -9,12 +9,11 @@
  */
 
 
-#include "vpx_ports/asm_offsets.h"
-#include "vp9/decoder/vp9_onyxd_int.h"
+#ifndef VP9_DECODER_VP9_DSUBEXP_H_
+#define VP9_DECODER_VP9_DSUBEXP_H_
 
-BEGIN
+#include "vp9/decoder/vp9_dboolhuff.h"
 
-END
+void vp9_diff_update_prob(vp9_reader *r, vp9_prob* p);
 
-/* add asserts for any offset that is not supported by assembly code */
-/* add asserts for any size that is not supported by assembly code */
+#endif  // VP9_DECODER_VP9_DSUBEXP_H_
diff --git a/libvpx/vp9/decoder/vp9_idct_blk.c b/libvpx/vp9/decoder/vp9_idct_blk.c
index c52963c..0217919 100644
--- a/libvpx/vp9/decoder/vp9_idct_blk.c
+++ b/libvpx/vp9/decoder/vp9_idct_blk.c
@@ -66,7 +66,7 @@ void vp9_idct_add_c(int16_t *input, uint8_t *dest, int stride, int eob) {
     vp9_short_idct4x4_add(input, dest, stride);
     vpx_memset(input, 0, 32);
   } else {
-    vp9_dc_only_idct_add(input[0], dest, dest, stride, stride);
+    vp9_short_idct4x4_1_add(input, dest, stride);
     ((int *)input)[0] = 0;
   }
 }
diff --git a/libvpx/vp9/decoder/vp9_onyxd_if.c b/libvpx/vp9/decoder/vp9_onyxd_if.c
index 3cef88b..cb72920 100644
--- a/libvpx/vp9/decoder/vp9_onyxd_if.c
+++ b/libvpx/vp9/decoder/vp9_onyxd_if.c
@@ -136,7 +136,7 @@ VP9D_PTR vp9_create_decompressor(VP9D_CONFIG *oxcf) {
   // vp9_init_dequantizer() for every frame.
   vp9_init_dequantizer(&pbi->common);
 
-  vp9_loop_filter_init(&pbi->common);
+  vp9_loop_filter_init(&pbi->common, &pbi->mb.lf);
 
   pbi->common.error.setjmp = 0;
   pbi->decoded_key_frame = 0;
@@ -154,7 +154,6 @@ void vp9_remove_decompressor(VP9D_PTR ptr) {
     vpx_free(pbi->common.last_frame_seg_map);
 
   vp9_remove_common(&pbi->common);
-  vpx_free(pbi->mbc);
   vpx_free(pbi);
 }
 
@@ -347,9 +346,9 @@ int vp9_receive_compressed_data(VP9D_PTR ptr,
                              cm->current_video_frame + 1000);
 #endif
 
-    if (cm->filter_level) {
+    if (!pbi->do_loopfilter_inline) {
       /* Apply the loop filter if appropriate. */
-      vp9_loop_filter_frame(cm, &pbi->mb, cm->filter_level, 0);
+      vp9_loop_filter_frame(cm, &pbi->mb, pbi->mb.lf.filter_level, 0);
     }
 
 #if WRITE_RECON_BUFFER == 2
@@ -361,8 +360,9 @@ int vp9_receive_compressed_data(VP9D_PTR ptr,
                              cm->current_video_frame + 3000);
 #endif
 
-    vp9_extend_frame_borders(cm->frame_to_show,
-                             cm->subsampling_x, cm->subsampling_y);
+    vp9_extend_frame_inner_borders(cm->frame_to_show,
+                                   cm->subsampling_x,
+                                   cm->subsampling_y);
   }
 
 #if WRITE_RECON_BUFFER == 1
@@ -412,9 +412,8 @@ int vp9_get_raw_frame(VP9D_PTR ptr, YV12_BUFFER_CONFIG *sd,
   *time_stamp = pbi->last_time_stamp;
   *time_end_stamp = 0;
 
-  sd->clrtype = pbi->common.clr_type;
 #if CONFIG_POSTPROC
-  ret = vp9_post_proc_frame(&pbi->common, sd, flags);
+  ret = vp9_post_proc_frame(&pbi->common, &pbi->mb.lf, sd, flags);
 #else
 
   if (pbi->common.frame_to_show) {
diff --git a/libvpx/vp9/decoder/vp9_onyxd_int.h b/libvpx/vp9/decoder/vp9_onyxd_int.h
index 8698570..4760066 100644
--- a/libvpx/vp9/decoder/vp9_onyxd_int.h
+++ b/libvpx/vp9/decoder/vp9_onyxd_int.h
@@ -10,13 +10,14 @@
 
 #ifndef VP9_DECODER_VP9_ONYXD_INT_H_
 #define VP9_DECODER_VP9_ONYXD_INT_H_
+
 #include "./vpx_config.h"
-#include "vp9/decoder/vp9_onyxd.h"
-#include "vp9/decoder/vp9_treereader.h"
+
 #include "vp9/common/vp9_onyxc_int.h"
-#include "vp9/decoder/vp9_idct_blk.h"
 
-// #define DEC_DEBUG
+#include "vp9/decoder/vp9_idct_blk.h"
+#include "vp9/decoder/vp9_onyxd.h"
+#include "vp9/decoder/vp9_treereader.h"
 
 typedef struct VP9Decompressor {
   DECLARE_ALIGNED(16, MACROBLOCKD, mb);
@@ -28,35 +29,17 @@ typedef struct VP9Decompressor {
   const uint8_t *source;
   uint32_t source_sz;
 
-  vp9_reader *mbc;
   int64_t last_time_stamp;
   int ready_for_new_data;
 
   int refresh_frame_flags;
-  vp9_prob prob_skip_false;
 
   int decoded_key_frame;
 
   int initial_width;
   int initial_height;
-} VP9D_COMP;
 
-
-#if CONFIG_DEBUG
-#define CHECK_MEM_ERROR(lval,expr) do {\
-    lval = (expr); \
-    if(!lval) \
-      vpx_internal_error(&pbi->common.error, VPX_CODEC_MEM_ERROR,\
-                         "Failed to allocate "#lval" at %s:%d", \
-                         __FILE__,__LINE__);\
-  } while(0)
-#else
-#define CHECK_MEM_ERROR(lval,expr) do {\
-    lval = (expr); \
-    if(!lval) \
-      vpx_internal_error(&pbi->common.error, VPX_CODEC_MEM_ERROR,\
-                         "Failed to allocate "#lval);\
-  } while(0)
-#endif
+  int do_loopfilter_inline;  // apply loopfilter to available rows immediately
+} VP9D_COMP;
 
 #endif  // VP9_DECODER_VP9_TREEREADER_H_
diff --git a/libvpx/vp9/decoder/vp9_read_bit_buffer.h b/libvpx/vp9/decoder/vp9_read_bit_buffer.h
index f243cb4..c7fa3aa 100644
--- a/libvpx/vp9/decoder/vp9_read_bit_buffer.h
+++ b/libvpx/vp9/decoder/vp9_read_bit_buffer.h
@@ -51,4 +51,10 @@ static int vp9_rb_read_literal(struct vp9_read_bit_buffer *rb, int bits) {
   return value;
 }
 
+static int vp9_rb_read_signed_literal(struct vp9_read_bit_buffer *rb,
+                                      int bits) {
+  const int value = vp9_rb_read_literal(rb, bits);
+  return vp9_rb_read_bit(rb) ? -value : value;
+}
+
 #endif  // VP9_READ_BIT_BUFFER_