Update aom to v1.0.0

Update aom to commit id d14c5bb4f336ef1842046089849dee4a301fbbf0.

26 files changed, 5371 insertions, 8853 deletions

diff --git a/third_party/aom/av1/decoder/accounting.c b/third_party/aom/av1/decoder/accounting.c
index ba243c9e1..8d8f3dfdb 100644
--- a/third_party/aom/av1/decoder/accounting.c
+++ b/third_party/aom/av1/decoder/accounting.c
@@ -15,7 +15,7 @@
 #include <string.h>
 
 #include "aom/aom_integer.h"
-#include "./accounting.h"
+#include "av1/decoder/accounting.h"
 
 static int aom_accounting_hash(const char *str) {
   uint32_t val;
@@ -31,7 +31,7 @@ static int aom_accounting_hash(const char *str) {
 /* Dictionary lookup based on an open-addressing hash table. */
 int aom_accounting_dictionary_lookup(Accounting *accounting, const char *str) {
   int hash;
-  int len;
+  size_t len;
   AccountingDictionary *dictionary;
   dictionary = &accounting->syms.dictionary;
   hash = aom_accounting_hash(str);
diff --git a/third_party/aom/av1/decoder/accounting.h b/third_party/aom/av1/decoder/accounting.h
index 889865b2e..9099d081b 100644
--- a/third_party/aom/av1/decoder/accounting.h
+++ b/third_party/aom/av1/decoder/accounting.h
@@ -11,6 +11,7 @@
 #ifndef AOM_ACCOUNTING_H_
 #define AOM_ACCOUNTING_H_
 #include <stdlib.h>
+#include "aom/aomdx.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -58,8 +59,6 @@ typedef struct {
   AccountingDictionary dictionary;
 } AccountingSymbols;
 
-typedef struct Accounting Accounting;
-
 struct Accounting {
   AccountingSymbols syms;
   /** Size allocated for symbols (not all may be used). */
diff --git a/third_party/aom/av1/decoder/decint.h b/third_party/aom/av1/decoder/decint.h
deleted file mode 100644
index e887ad5e0..000000000
--- a/third_party/aom/av1/decoder/decint.h
+++ /dev/null
@@ -1,35 +0,0 @@
-/*
- * Copyright (c) 2001-2016, Alliance for Open Media. All rights reserved
- *
- * This source code is subject to the terms of the BSD 2 Clause License and
- * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
- * was not distributed with this source code in the LICENSE file, you can
- * obtain it at www.aomedia.org/license/software. If the Alliance for Open
- * Media Patent License 1.0 was not distributed with this source code in the
- * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
- */
-
-/* clang-format off */
-
-#if !defined(_decint_H)
-# define _decint_H (1)
-# include "av1/common/pvq_state.h"
-# include "aom_dsp/bitreader.h"
-# include "aom_dsp/entdec.h"
-
-typedef struct daala_dec_ctx daala_dec_ctx;
-
-typedef struct daala_dec_ctx od_dec_ctx;
-
-
-struct daala_dec_ctx {
-  /* Stores context-adaptive CDFs for PVQ. */
-  od_state state;
-  /* AOM entropy decoder. */
-  aom_reader *r;
-  int use_activity_masking;
-  /* Mode of quantization matrice : FLAT (0) or HVS (1) */
-  int qm;
-};
-
-#endif
diff --git a/third_party/aom/av1/decoder/decodeframe.c b/third_party/aom/av1/decoder/decodeframe.c
index 9ec3b60eb..e92c6b28c 100644
--- a/third_party/aom/av1/decoder/decodeframe.c
+++ b/third_party/aom/av1/decoder/decodeframe.c
@@ -10,12 +10,12 @@
  */
 
 #include <assert.h>
-#include <stdlib.h>  // qsort()
+#include <stddef.h>
 
-#include "./aom_config.h"
-#include "./aom_dsp_rtcd.h"
-#include "./aom_scale_rtcd.h"
-#include "./av1_rtcd.h"
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+#include "config/aom_scale_rtcd.h"
+#include "config/av1_rtcd.h"
 
 #include "aom/aom_codec.h"
 #include "aom_dsp/aom_dsp_common.h"
@@ -23,19 +23,19 @@
 #include "aom_dsp/bitreader.h"
 #include "aom_dsp/bitreader_buffer.h"
 #include "aom_mem/aom_mem.h"
+#include "aom_ports/aom_timer.h"
 #include "aom_ports/mem.h"
 #include "aom_ports/mem_ops.h"
 #include "aom_scale/aom_scale.h"
 #include "aom_util/aom_thread.h"
 
-#if CONFIG_BITSTREAM_DEBUG
+#if CONFIG_BITSTREAM_DEBUG || CONFIG_MISMATCH_DEBUG
 #include "aom_util/debug_util.h"
-#endif  // CONFIG_BITSTREAM_DEBUG
+#endif  // CONFIG_BITSTREAM_DEBUG || CONFIG_MISMATCH_DEBUG
 
 #include "av1/common/alloccommon.h"
-#if CONFIG_CDEF
 #include "av1/common/cdef.h"
-#endif
+#include "av1/common/cfl.h"
 #if CONFIG_INSPECTION
 #include "av1/decoder/inspection.h"
 #endif
@@ -49,78 +49,69 @@
 #include "av1/common/quant_common.h"
 #include "av1/common/reconinter.h"
 #include "av1/common/reconintra.h"
-#if CONFIG_FRAME_SUPERRES
 #include "av1/common/resize.h"
-#endif  // CONFIG_FRAME_SUPERRES
 #include "av1/common/seg_common.h"
 #include "av1/common/thread_common.h"
 #include "av1/common/tile_common.h"
-
+#include "av1/common/warped_motion.h"
+#include "av1/common/obmc.h"
 #include "av1/decoder/decodeframe.h"
 #include "av1/decoder/decodemv.h"
 #include "av1/decoder/decoder.h"
-#if CONFIG_LV_MAP
 #include "av1/decoder/decodetxb.h"
-#endif
 #include "av1/decoder/detokenize.h"
-#include "av1/decoder/dsubexp.h"
-#include "av1/decoder/symbolrate.h"
 
-#if CONFIG_WARPED_MOTION || CONFIG_GLOBAL_MOTION
-#include "av1/common/warped_motion.h"
-#endif  // CONFIG_WARPED_MOTION || CONFIG_GLOBAL_MOTION
-
-#define MAX_AV1_HEADER_SIZE 80
 #define ACCT_STR __func__
 
-#if CONFIG_PVQ
-#include "av1/common/partition.h"
-#include "av1/common/pvq.h"
-#include "av1/common/scan.h"
-#include "av1/decoder/decint.h"
-#include "av1/decoder/pvq_decoder.h"
-#include "av1/encoder/encodemb.h"
-#include "av1/encoder/hybrid_fwd_txfm.h"
-#endif
+// This is needed by ext_tile related unit tests.
+#define EXT_TILE_DEBUG 1
+#define MC_TEMP_BUF_PELS                       \
+  (((MAX_SB_SIZE)*2 + (AOM_INTERP_EXTEND)*2) * \
+   ((MAX_SB_SIZE)*2 + (AOM_INTERP_EXTEND)*2))
 
-#if CONFIG_CFL
-#include "av1/common/cfl.h"
-#endif
+// Checks that the remaining bits start with a 1 and ends with 0s.
+// It consumes an additional byte, if already byte aligned before the check.
+int av1_check_trailing_bits(AV1Decoder *pbi, struct aom_read_bit_buffer *rb) {
+  AV1_COMMON *const cm = &pbi->common;
+  // bit_offset is set to 0 (mod 8) when the reader is already byte aligned
+  int bits_before_alignment = 8 - rb->bit_offset % 8;
+  int trailing = aom_rb_read_literal(rb, bits_before_alignment);
+  if (trailing != (1 << (bits_before_alignment - 1))) {
+    cm->error.error_code = AOM_CODEC_CORRUPT_FRAME;
+    return -1;
+  }
+  return 0;
+}
 
-#if CONFIG_STRIPED_LOOP_RESTORATION && !CONFIG_LOOP_RESTORATION
-#error "striped_loop_restoration requires loop_restoration"
-#endif
+// Use only_chroma = 1 to only set the chroma planes
+static void set_planes_to_neutral_grey(AV1_COMMON *const cm,
+                                       const YV12_BUFFER_CONFIG *const buf,
+                                       int only_chroma) {
+  const int val = 1 << (cm->bit_depth - 1);
+
+  for (int plane = only_chroma; plane < MAX_MB_PLANE; plane++) {
+    const int is_uv = plane > 0;
+    for (int row_idx = 0; row_idx < buf->crop_heights[is_uv]; row_idx++) {
+      if (cm->use_highbitdepth) {
+        // TODO(yaowu): replace this with aom_memset16() for speed
+        for (int col_idx = 0; col_idx < buf->crop_widths[is_uv]; col_idx++) {
+          uint16_t *base = CONVERT_TO_SHORTPTR(buf->buffers[plane]);
+          base[row_idx * buf->strides[is_uv] + col_idx] = val;
+        }
+      } else {
+        memset(&buf->buffers[plane][row_idx * buf->uv_stride], 1 << 7,
+               buf->crop_widths[is_uv]);
+      }
+    }
+  }
+}
 
-#if CONFIG_LOOP_RESTORATION
 static void loop_restoration_read_sb_coeffs(const AV1_COMMON *const cm,
                                             MACROBLOCKD *xd,
                                             aom_reader *const r, int plane,
-                                            int rtile_idx);
-#endif
-
-static struct aom_read_bit_buffer *init_read_bit_buffer(
-    AV1Decoder *pbi, struct aom_read_bit_buffer *rb, const uint8_t *data,
-    const uint8_t *data_end, uint8_t clear_data[MAX_AV1_HEADER_SIZE]);
-static int read_compressed_header(AV1Decoder *pbi, const uint8_t *data,
-                                  size_t partition_size);
-static size_t read_uncompressed_header(AV1Decoder *pbi,
-                                       struct aom_read_bit_buffer *rb);
-
-static int is_compound_reference_allowed(const AV1_COMMON *cm) {
-#if CONFIG_ONE_SIDED_COMPOUND  // Normative in decoder
-  return !frame_is_intra_only(cm);
-#else
-  int i;
-  if (frame_is_intra_only(cm)) return 0;
-  for (i = 1; i < INTER_REFS_PER_FRAME; ++i)
-    if (cm->ref_frame_sign_bias[i + 1] != cm->ref_frame_sign_bias[1]) return 1;
-
-  return 0;
-#endif  // CONFIG_ONE_SIDED_COMPOUND
-}
+                                            int runit_idx);
 
 static void setup_compound_reference_mode(AV1_COMMON *cm) {
-#if CONFIG_EXT_REFS
   cm->comp_fwd_ref[0] = LAST_FRAME;
   cm->comp_fwd_ref[1] = LAST2_FRAME;
   cm->comp_fwd_ref[2] = LAST3_FRAME;
@@ -129,1952 +120,1099 @@ static void setup_compound_reference_mode(AV1_COMMON *cm) {
   cm->comp_bwd_ref[0] = BWDREF_FRAME;
   cm->comp_bwd_ref[1] = ALTREF2_FRAME;
   cm->comp_bwd_ref[2] = ALTREF_FRAME;
-#else   // !CONFIG_EXT_REFS
-  if (cm->ref_frame_sign_bias[LAST_FRAME] ==
-      cm->ref_frame_sign_bias[GOLDEN_FRAME]) {
-    cm->comp_fixed_ref = ALTREF_FRAME;
-    cm->comp_var_ref[0] = LAST_FRAME;
-    cm->comp_var_ref[1] = GOLDEN_FRAME;
-  } else if (cm->ref_frame_sign_bias[LAST_FRAME] ==
-             cm->ref_frame_sign_bias[ALTREF_FRAME]) {
-    cm->comp_fixed_ref = GOLDEN_FRAME;
-    cm->comp_var_ref[0] = LAST_FRAME;
-    cm->comp_var_ref[1] = ALTREF_FRAME;
-  } else {
-    cm->comp_fixed_ref = LAST_FRAME;
-    cm->comp_var_ref[0] = GOLDEN_FRAME;
-    cm->comp_var_ref[1] = ALTREF_FRAME;
-  }
-#endif  // CONFIG_EXT_REFS
 }
 
 static int read_is_valid(const uint8_t *start, size_t len, const uint8_t *end) {
   return len != 0 && len <= (size_t)(end - start);
 }
 
-static int decode_unsigned_max(struct aom_read_bit_buffer *rb, int max) {
-  const int data = aom_rb_read_literal(rb, get_unsigned_bits(max));
-  return data > max ? max : data;
-}
-
 static TX_MODE read_tx_mode(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
-#if CONFIG_TX64X64
-  TX_MODE tx_mode;
-#endif
-  if (cm->all_lossless) return ONLY_4X4;
-#if CONFIG_VAR_TX_NO_TX_MODE
-  (void)rb;
-  return TX_MODE_SELECT;
-#else
-#if CONFIG_TX64X64
-  tx_mode = aom_rb_read_bit(rb) ? TX_MODE_SELECT : aom_rb_read_literal(rb, 2);
-  if (tx_mode == ALLOW_32X32) tx_mode += aom_rb_read_bit(rb);
-  return tx_mode;
-#else
-  return aom_rb_read_bit(rb) ? TX_MODE_SELECT : aom_rb_read_literal(rb, 2);
-#endif  // CONFIG_TX64X64
-#endif  // CONFIG_VAR_TX_NO_TX_MODE
+  if (cm->coded_lossless) return ONLY_4X4;
+  return aom_rb_read_bit(rb) ? TX_MODE_SELECT : TX_MODE_LARGEST;
 }
 
-#if !CONFIG_RESTRICT_COMPRESSED_HDR
-static void read_inter_mode_probs(FRAME_CONTEXT *fc, aom_reader *r) {
-  int i;
-  for (i = 0; i < NEWMV_MODE_CONTEXTS; ++i)
-    av1_diff_update_prob(r, &fc->newmv_prob[i], ACCT_STR);
-  for (i = 0; i < ZEROMV_MODE_CONTEXTS; ++i)
-    av1_diff_update_prob(r, &fc->zeromv_prob[i], ACCT_STR);
-  for (i = 0; i < REFMV_MODE_CONTEXTS; ++i)
-    av1_diff_update_prob(r, &fc->refmv_prob[i], ACCT_STR);
-  for (i = 0; i < DRL_MODE_CONTEXTS; ++i)
-    av1_diff_update_prob(r, &fc->drl_prob[i], ACCT_STR);
-}
-#endif
-
 static REFERENCE_MODE read_frame_reference_mode(
     const AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
-  if (is_compound_reference_allowed(cm)) {
-#if CONFIG_REF_ADAPT
-    return aom_rb_read_bit(rb) ? REFERENCE_MODE_SELECT : SINGLE_REFERENCE;
-#else
-    return aom_rb_read_bit(rb)
-               ? REFERENCE_MODE_SELECT
-               : (aom_rb_read_bit(rb) ? COMPOUND_REFERENCE : SINGLE_REFERENCE);
-#endif  // CONFIG_REF_ADAPT
-  } else {
+  if (frame_is_intra_only(cm)) {
     return SINGLE_REFERENCE;
+  } else {
+    return aom_rb_read_bit(rb) ? REFERENCE_MODE_SELECT : SINGLE_REFERENCE;
   }
 }
 
-#if !CONFIG_RESTRICT_COMPRESSED_HDR
-static void read_frame_reference_mode_probs(AV1_COMMON *cm, aom_reader *r) {
-  FRAME_CONTEXT *const fc = cm->fc;
-  int i;
-
-  if (cm->reference_mode == REFERENCE_MODE_SELECT)
-    for (i = 0; i < COMP_INTER_CONTEXTS; ++i)
-      av1_diff_update_prob(r, &fc->comp_inter_prob[i], ACCT_STR);
-
-  if (cm->reference_mode != COMPOUND_REFERENCE) {
-    for (i = 0; i < REF_CONTEXTS; ++i) {
-      int j;
-      for (j = 0; j < (SINGLE_REFS - 1); ++j) {
-        av1_diff_update_prob(r, &fc->single_ref_prob[i][j], ACCT_STR);
-      }
-    }
-  }
-
-  if (cm->reference_mode != SINGLE_REFERENCE) {
-#if CONFIG_EXT_COMP_REFS
-    for (i = 0; i < COMP_REF_TYPE_CONTEXTS; ++i)
-      av1_diff_update_prob(r, &fc->comp_ref_type_prob[i], ACCT_STR);
-
-    for (i = 0; i < UNI_COMP_REF_CONTEXTS; ++i) {
-      int j;
-      for (j = 0; j < (UNIDIR_COMP_REFS - 1); ++j)
-        av1_diff_update_prob(r, &fc->uni_comp_ref_prob[i][j], ACCT_STR);
-    }
-#endif  // CONFIG_EXT_COMP_REFS
-
-    for (i = 0; i < REF_CONTEXTS; ++i) {
-      int j;
-#if CONFIG_EXT_REFS
-      for (j = 0; j < (FWD_REFS - 1); ++j)
-        av1_diff_update_prob(r, &fc->comp_ref_prob[i][j], ACCT_STR);
-      for (j = 0; j < (BWD_REFS - 1); ++j)
-        av1_diff_update_prob(r, &fc->comp_bwdref_prob[i][j], ACCT_STR);
-#else
-      for (j = 0; j < (COMP_REFS - 1); ++j)
-        av1_diff_update_prob(r, &fc->comp_ref_prob[i][j], ACCT_STR);
-#endif  // CONFIG_EXT_REFS
-    }
-  }
-}
-
-static void update_mv_probs(aom_prob *p, int n, aom_reader *r) {
-  int i;
-  for (i = 0; i < n; ++i) av1_diff_update_prob(r, &p[i], ACCT_STR);
-}
-
-static void read_mv_probs(nmv_context *ctx, int allow_hp, aom_reader *r) {
-  int i;
-  if (allow_hp) {
-    for (i = 0; i < 2; ++i) {
-      nmv_component *const comp_ctx = &ctx->comps[i];
-      update_mv_probs(&comp_ctx->class0_hp, 1, r);
-      update_mv_probs(&comp_ctx->hp, 1, r);
-    }
-  }
-}
-#endif
-
 static void inverse_transform_block(MACROBLOCKD *xd, int plane,
-#if CONFIG_LGT_FROM_PRED
-                                    PREDICTION_MODE mode,
-#endif
                                     const TX_TYPE tx_type,
                                     const TX_SIZE tx_size, uint8_t *dst,
-                                    int stride, int16_t scan_line, int eob) {
+                                    int stride, int reduced_tx_set) {
   struct macroblockd_plane *const pd = &xd->plane[plane];
   tran_low_t *const dqcoeff = pd->dqcoeff;
-  av1_inverse_transform_block(xd, dqcoeff,
-#if CONFIG_LGT_FROM_PRED
-                              mode,
-#endif
-#if CONFIG_MRC_TX && SIGNAL_ANY_MRC_MASK
-                              xd->mrc_mask,
-#endif  // CONFIG_MRC_TX && SIGNAL_ANY_MRC_MASK
-                              tx_type, tx_size, dst, stride, eob);
+  eob_info *eob_data = pd->eob_data + xd->txb_offset[plane];
+  uint16_t scan_line = eob_data->max_scan_line;
+  uint16_t eob = eob_data->eob;
+
+  memcpy(dqcoeff, pd->dqcoeff_block + xd->cb_offset[plane],
+         (scan_line + 1) * sizeof(dqcoeff[0]));
+  av1_inverse_transform_block(xd, dqcoeff, plane, tx_type, tx_size, dst, stride,
+                              eob, reduced_tx_set);
   memset(dqcoeff, 0, (scan_line + 1) * sizeof(dqcoeff[0]));
 }
 
-static int get_block_idx(const MACROBLOCKD *xd, int plane, int row, int col) {
-  const int bsize = xd->mi[0]->mbmi.sb_type;
-  const struct macroblockd_plane *pd = &xd->plane[plane];
-#if CONFIG_CHROMA_SUB8X8
-  const BLOCK_SIZE plane_bsize =
-      AOMMAX(BLOCK_4X4, get_plane_block_size(bsize, pd));
-#elif CONFIG_CB4X4
-  const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
-#else
-  const BLOCK_SIZE plane_bsize =
-      get_plane_block_size(AOMMAX(BLOCK_8X8, bsize), pd);
-#endif
-  const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
-  const TX_SIZE tx_size = av1_get_tx_size(plane, xd);
-  const uint8_t txh_unit = tx_size_high_unit[tx_size];
-  return row * max_blocks_wide + col * txh_unit;
-}
-
-#if CONFIG_PVQ
-static int av1_pvq_decode_helper(MACROBLOCKD *xd, tran_low_t *ref_coeff,
-                                 tran_low_t *dqcoeff, int16_t *quant, int pli,
-                                 int bs, TX_TYPE tx_type, int xdec,
-                                 PVQ_SKIP_TYPE ac_dc_coded) {
-  unsigned int flags;  // used for daala's stream analyzer.
-  int off;
-  const int is_keyframe = 0;
-  const int has_dc_skip = 1;
-  int coeff_shift = 3 - av1_get_tx_scale(bs);
-  int hbd_downshift = 0;
-  int rounding_mask;
-  // DC quantizer for PVQ
-  int pvq_dc_quant;
-  int lossless = (quant[0] == 0);
-  const int blk_size = tx_size_wide[bs];
-  int eob = 0;
-  int i;
-  od_dec_ctx *dec = &xd->daala_dec;
-  int use_activity_masking = dec->use_activity_masking;
-  DECLARE_ALIGNED(16, tran_low_t, dqcoeff_pvq[OD_TXSIZE_MAX * OD_TXSIZE_MAX]);
-  DECLARE_ALIGNED(16, tran_low_t, ref_coeff_pvq[OD_TXSIZE_MAX * OD_TXSIZE_MAX]);
-
-  od_coeff ref_int32[OD_TXSIZE_MAX * OD_TXSIZE_MAX];
-  od_coeff out_int32[OD_TXSIZE_MAX * OD_TXSIZE_MAX];
-
-  hbd_downshift = xd->bd - 8;
-
-  od_raster_to_coding_order(ref_coeff_pvq, blk_size, tx_type, ref_coeff,
-                            blk_size);
-
-  assert(OD_COEFF_SHIFT >= 4);
-  if (lossless)
-    pvq_dc_quant = 1;
-  else {
-    if (use_activity_masking)
-      pvq_dc_quant =
-          OD_MAXI(1,
-                  (quant[0] << (OD_COEFF_SHIFT - 3) >> hbd_downshift) *
-                          dec->state.pvq_qm_q4[pli][od_qm_get_index(bs, 0)] >>
-                      4);
-    else
-      pvq_dc_quant =
-          OD_MAXI(1, quant[0] << (OD_COEFF_SHIFT - 3) >> hbd_downshift);
-  }
-
-  off = od_qm_offset(bs, xdec);
-
-  // copy int16 inputs to int32
-  for (i = 0; i < blk_size * blk_size; i++) {
-    ref_int32[i] =
-        AOM_SIGNED_SHL(ref_coeff_pvq[i], OD_COEFF_SHIFT - coeff_shift) >>
-        hbd_downshift;
-  }
-
-  od_pvq_decode(dec, ref_int32, out_int32,
-                OD_MAXI(1, quant[1] << (OD_COEFF_SHIFT - 3) >> hbd_downshift),
-                pli, bs, OD_PVQ_BETA[use_activity_masking][pli][bs],
-                is_keyframe, &flags, ac_dc_coded, dec->state.qm + off,
-                dec->state.qm_inv + off);
-
-  if (!has_dc_skip || out_int32[0]) {
-    out_int32[0] =
-        has_dc_skip + generic_decode(dec->r, &dec->state.adapt->model_dc[pli],
-                                     &dec->state.adapt->ex_dc[pli][bs][0], 2,
-                                     "dc:mag");
-    if (out_int32[0]) out_int32[0] *= aom_read_bit(dec->r, "dc:sign") ? -1 : 1;
-  }
-  out_int32[0] = out_int32[0] * pvq_dc_quant + ref_int32[0];
-
-  // copy int32 result back to int16
-  assert(OD_COEFF_SHIFT > coeff_shift);
-  rounding_mask = (1 << (OD_COEFF_SHIFT - coeff_shift - 1)) - 1;
-  for (i = 0; i < blk_size * blk_size; i++) {
-    out_int32[i] = AOM_SIGNED_SHL(out_int32[i], hbd_downshift);
-    dqcoeff_pvq[i] = (out_int32[i] + (out_int32[i] < 0) + rounding_mask) >>
-                     (OD_COEFF_SHIFT - coeff_shift);
-  }
-
-  od_coding_order_to_raster(dqcoeff, blk_size, tx_type, dqcoeff_pvq, blk_size);
-
-  eob = blk_size * blk_size;
-
-  return eob;
-}
-
-static PVQ_SKIP_TYPE read_pvq_skip(AV1_COMMON *cm, MACROBLOCKD *const xd,
-                                   int plane, TX_SIZE tx_size) {
-  // decode ac/dc coded flag. bit0: DC coded, bit1 : AC coded
-  // NOTE : we don't use 5 symbols for luma here in aom codebase,
-  // since block partition is taken care of by aom.
-  // So, only AC/DC skip info is coded
-  const int ac_dc_coded = aom_read_symbol(
-      xd->daala_dec.r,
-      xd->daala_dec.state.adapt->skip_cdf[2 * tx_size + (plane != 0)], 4,
-      "skip");
-  if (ac_dc_coded < 0 || ac_dc_coded > 3) {
-    aom_internal_error(&cm->error, AOM_CODEC_INVALID_PARAM,
-                       "Invalid PVQ Skip Type");
-  }
-  return ac_dc_coded;
-}
-
-static int av1_pvq_decode_helper2(AV1_COMMON *cm, MACROBLOCKD *const xd,
-                                  MB_MODE_INFO *const mbmi, int plane, int row,
-                                  int col, TX_SIZE tx_size, TX_TYPE tx_type) {
-  struct macroblockd_plane *const pd = &xd->plane[plane];
-  // transform block size in pixels
-  int tx_blk_size = tx_size_wide[tx_size];
-  int i, j;
-  tran_low_t *pvq_ref_coeff = pd->pvq_ref_coeff;
-  const int diff_stride = tx_blk_size;
-  int16_t *pred = pd->pred;
-  tran_low_t *const dqcoeff = pd->dqcoeff;
-  uint8_t *dst;
-  int eob;
-  const PVQ_SKIP_TYPE ac_dc_coded = read_pvq_skip(cm, xd, plane, tx_size);
-
-  eob = 0;
-  dst = &pd->dst.buf[4 * row * pd->dst.stride + 4 * col];
-
-  if (ac_dc_coded) {
-    int xdec = pd->subsampling_x;
-    int seg_id = mbmi->segment_id;
-    int16_t *quant;
-    TxfmParam txfm_param;
-    // ToDo(yaowu): correct this with optimal number from decoding process.
-    const int max_scan_line = tx_size_2d[tx_size];
-#if CONFIG_HIGHBITDEPTH
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-      for (j = 0; j < tx_blk_size; j++)
-        for (i = 0; i < tx_blk_size; i++)
-          pred[diff_stride * j + i] =
-              CONVERT_TO_SHORTPTR(dst)[pd->dst.stride * j + i];
-    } else {
+static void read_coeffs_tx_intra_block(AV1_COMMON *cm, MACROBLOCKD *const xd,
+                                       aom_reader *const r, int plane, int row,
+                                       int col, TX_SIZE tx_size) {
+  MB_MODE_INFO *mbmi = xd->mi[0];
+  if (!mbmi->skip) {
+#if TXCOEFF_TIMER
+    struct aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
 #endif
-      for (j = 0; j < tx_blk_size; j++)
-        for (i = 0; i < tx_blk_size; i++)
-          pred[diff_stride * j + i] = dst[pd->dst.stride * j + i];
-#if CONFIG_HIGHBITDEPTH
-    }
+    av1_read_coeffs_txb_facade(cm, xd, r, row, col, plane, tx_size);
+#if TXCOEFF_TIMER
+    aom_usec_timer_mark(&timer);
+    const int64_t elapsed_time = aom_usec_timer_elapsed(&timer);
+    cm->txcoeff_timer += elapsed_time;
+    ++cm->txb_count;
 #endif
-
-    txfm_param.tx_type = tx_type;
-    txfm_param.tx_size = tx_size;
-    txfm_param.lossless = xd->lossless[seg_id];
-
-#if CONFIG_HIGHBITDEPTH
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-      txfm_param.bd = xd->bd;
-      av1_highbd_fwd_txfm(pred, pvq_ref_coeff, diff_stride, &txfm_param);
-    } else {
-#endif  // CONFIG_HIGHBITDEPTH
-      av1_fwd_txfm(pred, pvq_ref_coeff, diff_stride, &txfm_param);
-#if CONFIG_HIGHBITDEPTH
-    }
-#endif  // CONFIG_HIGHBITDEPTH
-
-    quant = &pd->seg_dequant[seg_id][0];  // aom's quantizer
-
-    eob = av1_pvq_decode_helper(xd, pvq_ref_coeff, dqcoeff, quant, plane,
-                                tx_size, tx_type, xdec, ac_dc_coded);
-
-    inverse_transform_block(xd, plane, tx_type, tx_size, dst, pd->dst.stride,
-                            max_scan_line, eob);
   }
-
-  return eob;
 }
-#endif
 
-static void predict_and_reconstruct_intra_block(
-    AV1_COMMON *cm, MACROBLOCKD *const xd, aom_reader *const r,
-    MB_MODE_INFO *const mbmi, int plane, int row, int col, TX_SIZE tx_size) {
-  PLANE_TYPE plane_type = get_plane_type(plane);
-  const int block_idx = get_block_idx(xd, plane, row, col);
-#if CONFIG_PVQ
+static void predict_and_reconstruct_intra_block(AV1_COMMON *cm,
+                                                MACROBLOCKD *const xd,
+                                                aom_reader *const r, int plane,
+                                                int row, int col,
+                                                TX_SIZE tx_size) {
   (void)r;
-#endif
-  av1_predict_intra_block_facade(cm, xd, plane, block_idx, col, row, tx_size);
+  MB_MODE_INFO *mbmi = xd->mi[0];
+  PLANE_TYPE plane_type = get_plane_type(plane);
+
+  av1_predict_intra_block_facade(cm, xd, plane, col, row, tx_size);
 
   if (!mbmi->skip) {
-#if !CONFIG_PVQ
     struct macroblockd_plane *const pd = &xd->plane[plane];
-#if CONFIG_LV_MAP
-    int16_t max_scan_line = 0;
-    int eob;
-    av1_read_coeffs_txb_facade(cm, xd, r, row, col, block_idx, plane,
-                               pd->dqcoeff, tx_size, &max_scan_line, &eob);
+
     // tx_type will be read out in av1_read_coeffs_txb_facade
-    const TX_TYPE tx_type =
-        av1_get_tx_type(plane_type, xd, row, col, block_idx, tx_size);
-#else   // CONFIG_LV_MAP
-    const TX_TYPE tx_type =
-        av1_get_tx_type(plane_type, xd, row, col, block_idx, tx_size);
-    const SCAN_ORDER *scan_order = get_scan(cm, tx_size, tx_type, mbmi);
-    int16_t max_scan_line = 0;
-    const int eob =
-        av1_decode_block_tokens(cm, xd, plane, scan_order, col, row, tx_size,
-                                tx_type, &max_scan_line, r, mbmi->segment_id);
-#endif  // CONFIG_LV_MAP
-    if (eob) {
+    const TX_TYPE tx_type = av1_get_tx_type(plane_type, xd, row, col, tx_size,
+                                            cm->reduced_tx_set_used);
+    eob_info *eob_data = pd->eob_data + xd->txb_offset[plane];
+    if (eob_data->eob) {
       uint8_t *dst =
           &pd->dst.buf[(row * pd->dst.stride + col) << tx_size_wide_log2[0]];
-      inverse_transform_block(xd, plane,
-#if CONFIG_LGT_FROM_PRED
-                              mbmi->mode,
-#endif
-                              tx_type, tx_size, dst, pd->dst.stride,
-                              max_scan_line, eob);
+      inverse_transform_block(xd, plane, tx_type, tx_size, dst, pd->dst.stride,
+                              cm->reduced_tx_set_used);
     }
-#else   // !CONFIG_PVQ
-    const TX_TYPE tx_type =
-        av1_get_tx_type(plane_type, xd, row, col, block_idx, tx_size);
-    av1_pvq_decode_helper2(cm, xd, mbmi, plane, row, col, tx_size, tx_type);
-#endif  // !CONFIG_PVQ
-  }
-#if CONFIG_CFL
-  if (plane == AOM_PLANE_Y && xd->cfl->store_y) {
+  }
+  if (plane == AOM_PLANE_Y && store_cfl_required(cm, xd)) {
     cfl_store_tx(xd, row, col, tx_size, mbmi->sb_type);
   }
-#endif  // CONFIG_CFL
 }
 
-#if CONFIG_VAR_TX && !CONFIG_COEF_INTERLEAVE
+static void inverse_transform_inter_block(const AV1_COMMON *const cm,
+                                          MACROBLOCKD *const xd,
+                                          aom_reader *const r,
+                                          const int blk_row, const int blk_col,
+                                          const int plane,
+                                          const TX_SIZE tx_size) {
+  (void)r;
+  PLANE_TYPE plane_type = get_plane_type(plane);
+  const struct macroblockd_plane *const pd = &xd->plane[plane];
+  MB_MODE_INFO *mbmi = xd->mi[0];
+
+  // tx_type will be read out in av1_read_coeffs_txb_facade
+  const TX_TYPE tx_type = av1_get_tx_type(plane_type, xd, blk_row, blk_col,
+                                          tx_size, cm->reduced_tx_set_used);
+
+  if (plane == 0)
+    update_txk_array(mbmi->txk_type, mbmi->sb_type, blk_row, blk_col, tx_size,
+                     tx_type);
+
+  uint8_t *dst =
+      &pd->dst
+           .buf[(blk_row * pd->dst.stride + blk_col) << tx_size_wide_log2[0]];
+  inverse_transform_block(xd, plane, tx_type, tx_size, dst, pd->dst.stride,
+                          cm->reduced_tx_set_used);
+}
+
+static void set_cb_buffer_offsets(MACROBLOCKD *const xd, TX_SIZE tx_size,
+                                  int plane) {
+  xd->cb_offset[plane] += tx_size_wide[tx_size] * tx_size_high[tx_size];
+  xd->txb_offset[plane] =
+      xd->cb_offset[plane] / (TX_SIZE_W_MIN * TX_SIZE_H_MIN);
+}
+
 static void decode_reconstruct_tx(AV1_COMMON *cm, MACROBLOCKD *const xd,
                                   aom_reader *r, MB_MODE_INFO *const mbmi,
                                   int plane, BLOCK_SIZE plane_bsize,
                                   int blk_row, int blk_col, int block,
                                   TX_SIZE tx_size, int *eob_total) {
   const struct macroblockd_plane *const pd = &xd->plane[plane];
-  const BLOCK_SIZE bsize = txsize_to_bsize[tx_size];
-  const int tx_row = blk_row >> (1 - pd->subsampling_y);
-  const int tx_col = blk_col >> (1 - pd->subsampling_x);
   const TX_SIZE plane_tx_size =
-      plane ? uv_txsize_lookup[bsize][mbmi->inter_tx_size[tx_row][tx_col]][0][0]
-            : mbmi->inter_tx_size[tx_row][tx_col];
+      plane ? av1_get_max_uv_txsize(mbmi->sb_type, pd->subsampling_x,
+                                    pd->subsampling_y)
+            : mbmi->inter_tx_size[av1_get_txb_size_index(plane_bsize, blk_row,
+                                                         blk_col)];
   // Scale to match transform block unit.
   const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
   const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
 
   if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
 
-  if (tx_size == plane_tx_size) {
-    PLANE_TYPE plane_type = get_plane_type(plane);
-#if CONFIG_LV_MAP
-    int16_t max_scan_line = 0;
-    int eob;
-    av1_read_coeffs_txb_facade(cm, xd, r, blk_row, blk_col, block, plane,
-                               pd->dqcoeff, tx_size, &max_scan_line, &eob);
-    // tx_type will be read out in av1_read_coeffs_txb_facade
-    const TX_TYPE tx_type =
-        av1_get_tx_type(plane_type, xd, blk_row, blk_col, block, plane_tx_size);
-#else   // CONFIG_LV_MAP
-    const TX_TYPE tx_type =
-        av1_get_tx_type(plane_type, xd, blk_row, blk_col, block, plane_tx_size);
-    const SCAN_ORDER *sc = get_scan(cm, plane_tx_size, tx_type, mbmi);
-    int16_t max_scan_line = 0;
-    const int eob = av1_decode_block_tokens(
-        cm, xd, plane, sc, blk_col, blk_row, plane_tx_size, tx_type,
-        &max_scan_line, r, mbmi->segment_id);
-#endif  // CONFIG_LV_MAP
-    inverse_transform_block(xd, plane,
-#if CONFIG_LGT_FROM_PRED
-                            mbmi->mode,
-#endif
-                            tx_type, plane_tx_size,
-                            &pd->dst.buf[(blk_row * pd->dst.stride + blk_col)
-                                         << tx_size_wide_log2[0]],
-                            pd->dst.stride, max_scan_line, eob);
-    *eob_total += eob;
+  if (tx_size == plane_tx_size || plane) {
+#if TXCOEFF_TIMER
+    struct aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+#endif
+    av1_read_coeffs_txb_facade(cm, xd, r, blk_row, blk_col, plane, tx_size);
+#if TXCOEFF_TIMER
+    aom_usec_timer_mark(&timer);
+    const int64_t elapsed_time = aom_usec_timer_elapsed(&timer);
+    cm->txcoeff_timer += elapsed_time;
+    ++cm->txb_count;
+#endif
+    inverse_transform_inter_block(cm, xd, r, blk_row, blk_col, plane, tx_size);
+
+#if CONFIG_MISMATCH_DEBUG
+    int pixel_c, pixel_r;
+    BLOCK_SIZE bsize = txsize_to_bsize[tx_size];
+    int blk_w = block_size_wide[bsize];
+    int blk_h = block_size_high[bsize];
+    mi_to_pixel_loc(&pixel_c, &pixel_r, mi_col, mi_row, blk_col, blk_row,
+                    pd->subsampling_x, pd->subsampling_y);
+    mismatch_check_block_tx(dst, pd->dst.stride, cm->frame_offset, plane,
+                            pixel_c, pixel_r, blk_w, blk_h,
+                            xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH);
+#endif
+    eob_info *eob_data = pd->eob_data + xd->txb_offset[plane];
+    *eob_total += eob_data->eob;
+    set_cb_buffer_offsets(xd, tx_size, plane);
   } else {
-#if CONFIG_RECT_TX_EXT
-    int is_qttx = plane_tx_size == quarter_txsize_lookup[plane_bsize];
-    const TX_SIZE sub_txs = is_qttx ? plane_tx_size : sub_tx_size_map[tx_size];
-    if (is_qttx) assert(blk_row == 0 && blk_col == 0 && block == 0);
-#else
     const TX_SIZE sub_txs = sub_tx_size_map[tx_size];
     assert(IMPLIES(tx_size <= TX_4X4, sub_txs == tx_size));
     assert(IMPLIES(tx_size > TX_4X4, sub_txs < tx_size));
-#endif
-    const int bsl = tx_size_wide_unit[sub_txs];
-    int sub_step = tx_size_wide_unit[sub_txs] * tx_size_high_unit[sub_txs];
-    int i;
+    const int bsw = tx_size_wide_unit[sub_txs];
+    const int bsh = tx_size_high_unit[sub_txs];
+    const int sub_step = bsw * bsh;
 
-    assert(bsl > 0);
-
-    for (i = 0; i < 4; ++i) {
-#if CONFIG_RECT_TX_EXT
-      int is_wide_tx = tx_size_wide_unit[sub_txs] > tx_size_high_unit[sub_txs];
-      const int offsetr =
-          is_qttx ? (is_wide_tx ? i * tx_size_high_unit[sub_txs] : 0)
-                  : blk_row + ((i >> 1) * bsl);
-      const int offsetc =
-          is_qttx ? (is_wide_tx ? 0 : i * tx_size_wide_unit[sub_txs])
-                  : blk_col + (i & 0x01) * bsl;
-#else
-      const int offsetr = blk_row + (i >> 1) * bsl;
-      const int offsetc = blk_col + (i & 0x01) * bsl;
-#endif
+    assert(bsw > 0 && bsh > 0);
+
+    for (int row = 0; row < tx_size_high_unit[tx_size]; row += bsh) {
+      for (int col = 0; col < tx_size_wide_unit[tx_size]; col += bsw) {
+        const int offsetr = blk_row + row;
+        const int offsetc = blk_col + col;
 
-      if (offsetr >= max_blocks_high || offsetc >= max_blocks_wide) continue;
+        if (offsetr >= max_blocks_high || offsetc >= max_blocks_wide) continue;
 
-      decode_reconstruct_tx(cm, xd, r, mbmi, plane, plane_bsize, offsetr,
-                            offsetc, block, sub_txs, eob_total);
-      block += sub_step;
+        decode_reconstruct_tx(cm, xd, r, mbmi, plane, plane_bsize, offsetr,
+                              offsetc, block, sub_txs, eob_total);
+        block += sub_step;
+      }
     }
   }
 }
-#endif  // CONFIG_VAR_TX
-
-#if !CONFIG_VAR_TX || CONFIG_SUPERTX || CONFIG_COEF_INTERLEAVE || \
-    (!CONFIG_VAR_TX && CONFIG_EXT_TX && CONFIG_RECT_TX)
-static int reconstruct_inter_block(AV1_COMMON *cm, MACROBLOCKD *const xd,
-                                   aom_reader *const r, int segment_id,
-                                   int plane, int row, int col,
-                                   TX_SIZE tx_size) {
-  PLANE_TYPE plane_type = get_plane_type(plane);
-  int block_idx = get_block_idx(xd, plane, row, col);
-#if CONFIG_PVQ
-  int eob;
-  (void)r;
-  (void)segment_id;
-#else
-  struct macroblockd_plane *const pd = &xd->plane[plane];
-#endif
-
-#if !CONFIG_PVQ
-#if CONFIG_LV_MAP
-  (void)segment_id;
-  int16_t max_scan_line = 0;
-  int eob;
-  av1_read_coeffs_txb_facade(cm, xd, r, row, col, block_idx, plane, pd->dqcoeff,
-                             tx_size, &max_scan_line, &eob);
-  // tx_type will be read out in av1_read_coeffs_txb_facade
-  const TX_TYPE tx_type =
-      av1_get_tx_type(plane_type, xd, row, col, block_idx, tx_size);
-#else   // CONFIG_LV_MAP
-  int16_t max_scan_line = 0;
-  const TX_TYPE tx_type =
-      av1_get_tx_type(plane_type, xd, row, col, block_idx, tx_size);
-  const SCAN_ORDER *scan_order =
-      get_scan(cm, tx_size, tx_type, &xd->mi[0]->mbmi);
-  const int eob =
-      av1_decode_block_tokens(cm, xd, plane, scan_order, col, row, tx_size,
-                              tx_type, &max_scan_line, r, segment_id);
-#endif  // CONFIG_LV_MAP
-  uint8_t *dst =
-      &pd->dst.buf[(row * pd->dst.stride + col) << tx_size_wide_log2[0]];
-  if (eob)
-    inverse_transform_block(xd, plane,
-#if CONFIG_LGT_FROM_PRED
-                            xd->mi[0]->mbmi.mode,
-#endif
-                            tx_type, tx_size, dst, pd->dst.stride,
-                            max_scan_line, eob);
-#else
-  const TX_TYPE tx_type =
-      av1_get_tx_type(plane_type, xd, row, col, block_idx, tx_size);
-  eob = av1_pvq_decode_helper2(cm, xd, &xd->mi[0]->mbmi, plane, row, col,
-                               tx_size, tx_type);
-#endif
-  return eob;
-}
-#endif  // !CONFIG_VAR_TX || CONFIG_SUPER_TX
 
 static void set_offsets(AV1_COMMON *const cm, MACROBLOCKD *const xd,
                         BLOCK_SIZE bsize, int mi_row, int mi_col, int bw,
                         int bh, int x_mis, int y_mis) {
+  const int num_planes = av1_num_planes(cm);
+
   const int offset = mi_row * cm->mi_stride + mi_col;
-  int x, y;
   const TileInfo *const tile = &xd->tile;
 
   xd->mi = cm->mi_grid_visible + offset;
   xd->mi[0] = &cm->mi[offset];
   // TODO(slavarnway): Generate sb_type based on bwl and bhl, instead of
   // passing bsize from decode_partition().
-  xd->mi[0]->mbmi.sb_type = bsize;
+  xd->mi[0]->sb_type = bsize;
 #if CONFIG_RD_DEBUG
-  xd->mi[0]->mbmi.mi_row = mi_row;
-  xd->mi[0]->mbmi.mi_col = mi_col;
+  xd->mi[0]->mi_row = mi_row;
+  xd->mi[0]->mi_col = mi_col;
 #endif
-#if CONFIG_CFL
-  xd->cfl->mi_row = mi_row;
-  xd->cfl->mi_col = mi_col;
-#endif
-  for (y = 0; y < y_mis; ++y)
-    for (x = !y; x < x_mis; ++x) xd->mi[y * cm->mi_stride + x] = xd->mi[0];
+  xd->cfl.mi_row = mi_row;
+  xd->cfl.mi_col = mi_col;
 
-  set_plane_n4(xd, bw, bh);
-  set_skip_context(xd, mi_row, mi_col);
+  assert(x_mis && y_mis);
+  for (int x = 1; x < x_mis; ++x) xd->mi[x] = xd->mi[0];
+  int idx = cm->mi_stride;
+  for (int y = 1; y < y_mis; ++y) {
+    memcpy(&xd->mi[idx], &xd->mi[0], x_mis * sizeof(xd->mi[0]));
+    idx += cm->mi_stride;
+  }
 
-#if CONFIG_VAR_TX
-  xd->max_tx_size = max_txsize_lookup[bsize];
-#endif
+  set_plane_n4(xd, bw, bh, num_planes);
+  set_skip_context(xd, mi_row, mi_col, num_planes);
 
   // 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
-  set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw,
-#if CONFIG_DEPENDENT_HORZTILES
-                 cm->dependent_horz_tiles,
-#endif  // CONFIG_DEPENDENT_HORZTILES
-                 cm->mi_rows, cm->mi_cols);
+  set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);
 
   av1_setup_dst_planes(xd->plane, bsize, get_frame_new_buffer(cm), mi_row,
-                       mi_col);
+                       mi_col, 0, num_planes);
 }
 
-#if CONFIG_SUPERTX
-static MB_MODE_INFO *set_offsets_extend(AV1_COMMON *const cm,
-                                        MACROBLOCKD *const xd,
-                                        const TileInfo *const tile,
-                                        BLOCK_SIZE bsize_pred, int mi_row_pred,
-                                        int mi_col_pred, int mi_row_ori,
-                                        int mi_col_ori) {
-  // Used in supertx
-  // (mi_row_ori, mi_col_ori): location for mv
-  // (mi_row_pred, mi_col_pred, bsize_pred): region to predict
-  const int bw = mi_size_wide[bsize_pred];
-  const int bh = mi_size_high[bsize_pred];
-  const int offset = mi_row_ori * cm->mi_stride + mi_col_ori;
-  xd->mi = cm->mi_grid_visible + offset;
-  xd->mi[0] = cm->mi + offset;
-  set_mi_row_col(xd, tile, mi_row_pred, bh, mi_col_pred, bw,
-#if CONFIG_DEPENDENT_HORZTILES
-                 cm->dependent_horz_tiles,
-#endif  // CONFIG_DEPENDENT_HORZTILES
-                 cm->mi_rows, cm->mi_cols);
-
-  xd->up_available = (mi_row_ori > tile->mi_row_start);
-  xd->left_available = (mi_col_ori > tile->mi_col_start);
+static void decode_mbmi_block(AV1Decoder *const pbi, MACROBLOCKD *const xd,
+                              int mi_row, int mi_col, aom_reader *r,
+                              PARTITION_TYPE partition, BLOCK_SIZE bsize) {
+  AV1_COMMON *const cm = &pbi->common;
+  const int bw = mi_size_wide[bsize];
+  const int bh = mi_size_high[bsize];
+  const int x_mis = AOMMIN(bw, cm->mi_cols - mi_col);
+  const int y_mis = AOMMIN(bh, cm->mi_rows - mi_row);
 
-  set_plane_n4(xd, bw, bh);
+#if CONFIG_ACCOUNTING
+  aom_accounting_set_context(&pbi->accounting, mi_col, mi_row);
+#endif
+  set_offsets(cm, xd, bsize, mi_row, mi_col, bw, bh, x_mis, y_mis);
+  xd->mi[0]->partition = partition;
+  av1_read_mode_info(pbi, xd, mi_row, mi_col, r, x_mis, y_mis);
+  if (bsize >= BLOCK_8X8 && (cm->subsampling_x || cm->subsampling_y)) {
+    const BLOCK_SIZE uv_subsize =
+        ss_size_lookup[bsize][cm->subsampling_x][cm->subsampling_y];
+    if (uv_subsize == BLOCK_INVALID)
+      aom_internal_error(xd->error_info, AOM_CODEC_CORRUPT_FRAME,
+                         "Invalid block size.");
+  }
 
-  return &xd->mi[0]->mbmi;
+  int reader_corrupted_flag = aom_reader_has_error(r);
+  aom_merge_corrupted_flag(&xd->corrupted, reader_corrupted_flag);
 }
 
-#if CONFIG_SUPERTX
-static MB_MODE_INFO *set_mb_offsets(AV1_COMMON *const cm, MACROBLOCKD *const xd,
-                                    BLOCK_SIZE bsize, int mi_row, int mi_col,
-                                    int bw, int bh, int x_mis, int y_mis) {
-  const int offset = mi_row * cm->mi_stride + mi_col;
-  const TileInfo *const tile = &xd->tile;
-  int x, y;
+typedef struct PadBlock {
+  int x0;
+  int x1;
+  int y0;
+  int y1;
+} PadBlock;
 
-  xd->mi = cm->mi_grid_visible + offset;
-  xd->mi[0] = cm->mi + offset;
-  xd->mi[0]->mbmi.sb_type = bsize;
-  for (y = 0; y < y_mis; ++y)
-    for (x = !y; x < x_mis; ++x) xd->mi[y * cm->mi_stride + x] = xd->mi[0];
-
-  set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw,
-#if CONFIG_DEPENDENT_HORZTILES
-                 cm->dependent_horz_tiles,
-#endif  // CONFIG_DEPENDENT_HORZTILES
-                 cm->mi_rows, cm->mi_cols);
-  return &xd->mi[0]->mbmi;
-}
-#endif
+static void highbd_build_mc_border(const uint8_t *src8, int src_stride,
+                                   uint8_t *dst8, int dst_stride, int x, int y,
+                                   int b_w, int b_h, int w, int h) {
+  // Get a pointer to the start of the real data for this row.
+  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+  const uint16_t *ref_row = src - x - y * src_stride;
 
-static void set_offsets_topblock(AV1_COMMON *const cm, MACROBLOCKD *const xd,
-                                 const TileInfo *const tile, BLOCK_SIZE bsize,
-                                 int mi_row, int mi_col) {
-  const int bw = mi_size_wide[bsize];
-  const int bh = mi_size_high[bsize];
-  const int offset = mi_row * cm->mi_stride + mi_col;
+  if (y >= h)
+    ref_row += (h - 1) * src_stride;
+  else if (y > 0)
+    ref_row += y * src_stride;
 
-  xd->mi = cm->mi_grid_visible + offset;
-  xd->mi[0] = cm->mi + offset;
+  do {
+    int right = 0, copy;
+    int left = x < 0 ? -x : 0;
 
-  set_plane_n4(xd, bw, bh);
+    if (left > b_w) left = b_w;
 
-  set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw,
-#if CONFIG_DEPENDENT_HORZTILES
-                 cm->dependent_horz_tiles,
-#endif  // CONFIG_DEPENDENT_HORZTILES
-                 cm->mi_rows, cm->mi_cols);
+    if (x + b_w > w) right = x + b_w - w;
 
-  av1_setup_dst_planes(xd->plane, bsize, get_frame_new_buffer(cm), mi_row,
-                       mi_col);
-}
+    if (right > b_w) right = b_w;
 
-static void set_param_topblock(AV1_COMMON *const cm, MACROBLOCKD *const xd,
-                               BLOCK_SIZE bsize, int mi_row, int mi_col,
-                               int txfm, int skip) {
-  const int bw = mi_size_wide[bsize];
-  const int bh = mi_size_high[bsize];
-  const int x_mis = AOMMIN(bw, cm->mi_cols - mi_col);
-  const int y_mis = AOMMIN(bh, cm->mi_rows - mi_row);
-  const int offset = mi_row * cm->mi_stride + mi_col;
-  int x, y;
+    copy = b_w - left - right;
 
-  xd->mi = cm->mi_grid_visible + offset;
-  xd->mi[0] = cm->mi + offset;
+    if (left) aom_memset16(dst, ref_row[0], left);
 
-  for (y = 0; y < y_mis; ++y)
-    for (x = 0; x < x_mis; ++x) {
-      xd->mi[y * cm->mi_stride + x]->mbmi.skip = skip;
-      xd->mi[y * cm->mi_stride + x]->mbmi.tx_type = txfm;
-    }
-#if CONFIG_VAR_TX
-  xd->above_txfm_context = cm->above_txfm_context + mi_col;
-  xd->left_txfm_context =
-      xd->left_txfm_context_buffer + (mi_row & MAX_MIB_MASK);
-  set_txfm_ctxs(xd->mi[0]->mbmi.tx_size, bw, bh, skip, xd);
-#endif
-}
+    if (copy) memcpy(dst + left, ref_row + x + left, copy * sizeof(uint16_t));
 
-static void set_ref(AV1_COMMON *const cm, MACROBLOCKD *const xd, int idx,
-                    int mi_row, int mi_col) {
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
-#if CONFIG_COMPOUND_SINGLEREF
-  RefBuffer *ref_buffer =
-      has_second_ref(mbmi) ? &cm->frame_refs[mbmi->ref_frame[idx] - LAST_FRAME]
-                           : &cm->frame_refs[mbmi->ref_frame[0] - LAST_FRAME];
-#else
-  RefBuffer *ref_buffer = &cm->frame_refs[mbmi->ref_frame[idx] - LAST_FRAME];
-#endif  // CONFIG_COMPOUND_SINGLEREF
-  xd->block_refs[idx] = ref_buffer;
-  if (!av1_is_valid_scale(&ref_buffer->sf))
-    aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
-                       "Invalid scale factors");
-  av1_setup_pre_planes(xd, idx, ref_buffer->buf, mi_row, mi_col,
-                       &ref_buffer->sf);
-  aom_merge_corrupted_flag(&xd->corrupted, ref_buffer->buf->corrupted);
+    if (right) aom_memset16(dst + left + copy, ref_row[w - 1], right);
+
+    dst += dst_stride;
+    ++y;
+
+    if (y > 0 && y < h) ref_row += src_stride;
+  } while (--b_h);
 }
 
-static void dec_predict_b_extend(
-    AV1Decoder *const pbi, MACROBLOCKD *const xd, const TileInfo *const tile,
-    int block, int mi_row_ori, int mi_col_ori, int mi_row_pred, int mi_col_pred,
-    int mi_row_top, int mi_col_top, int plane, uint8_t *dst_buf, int dst_stride,
-    BLOCK_SIZE bsize_top, BLOCK_SIZE bsize_pred, int b_sub8x8, int bextend) {
-  // Used in supertx
-  // (mi_row_ori, mi_col_ori): location for mv
-  // (mi_row_pred, mi_col_pred, bsize_pred): region to predict
-  // (mi_row_top, mi_col_top, bsize_top): region of the top partition size
-  // block: sub location of sub8x8 blocks
-  // b_sub8x8: 1: ori is sub8x8; 0: ori is not sub8x8
-  // bextend: 1: region to predict is an extension of ori; 0: not
-  int r = (mi_row_pred - mi_row_top) * MI_SIZE;
-  int c = (mi_col_pred - mi_col_top) * MI_SIZE;
-  const int mi_width_top = mi_size_wide[bsize_top];
-  const int mi_height_top = mi_size_high[bsize_top];
-  MB_MODE_INFO *mbmi;
-  AV1_COMMON *const cm = &pbi->common;
+static void build_mc_border(const uint8_t *src, int src_stride, uint8_t *dst,
+                            int dst_stride, int x, int y, int b_w, int b_h,
+                            int w, int h) {
+  // Get a pointer to the start of the real data for this row.
+  const uint8_t *ref_row = src - x - y * src_stride;
 
-  if (mi_row_pred < mi_row_top || mi_col_pred < mi_col_top ||
-      mi_row_pred >= mi_row_top + mi_height_top ||
-      mi_col_pred >= mi_col_top + mi_width_top || mi_row_pred >= cm->mi_rows ||
-      mi_col_pred >= cm->mi_cols)
-    return;
+  if (y >= h)
+    ref_row += (h - 1) * src_stride;
+  else if (y > 0)
+    ref_row += y * src_stride;
 
-  mbmi = set_offsets_extend(cm, xd, tile, bsize_pred, mi_row_pred, mi_col_pred,
-                            mi_row_ori, mi_col_ori);
-  set_ref(cm, xd, 0, mi_row_pred, mi_col_pred);
-  if (has_second_ref(&xd->mi[0]->mbmi)
-#if CONFIG_COMPOUND_SINGLEREF
-      || is_inter_singleref_comp_mode(xd->mi[0]->mbmi.mode)
-#endif  // CONFIG_COMPOUND_SINGLEREF
-          )
-    set_ref(cm, xd, 1, mi_row_pred, mi_col_pred);
-  if (!bextend) mbmi->tx_size = max_txsize_lookup[bsize_top];
-
-  xd->plane[plane].dst.stride = dst_stride;
-  xd->plane[plane].dst.buf =
-      dst_buf + (r >> xd->plane[plane].subsampling_y) * dst_stride +
-      (c >> xd->plane[plane].subsampling_x);
-
-  if (!b_sub8x8)
-    av1_build_inter_predictor_sb_extend(&pbi->common, xd, mi_row_ori,
-                                        mi_col_ori, mi_row_pred, mi_col_pred,
-                                        plane, bsize_pred);
-  else
-    av1_build_inter_predictor_sb_sub8x8_extend(
-        &pbi->common, xd, mi_row_ori, mi_col_ori, mi_row_pred, mi_col_pred,
-        plane, bsize_pred, block);
-}
+  do {
+    int right = 0, copy;
+    int left = x < 0 ? -x : 0;
 
-static void dec_extend_dir(AV1Decoder *const pbi, MACROBLOCKD *const xd,
-                           const TileInfo *const tile, int block,
-                           BLOCK_SIZE bsize, BLOCK_SIZE top_bsize,
-                           int mi_row_ori, int mi_col_ori, int mi_row,
-                           int mi_col, int mi_row_top, int mi_col_top,
-                           int plane, uint8_t *dst_buf, int dst_stride,
-                           int dir) {
-  // dir: 0-lower, 1-upper, 2-left, 3-right
-  //      4-lowerleft, 5-upperleft, 6-lowerright, 7-upperright
-  const int mi_width = mi_size_wide[bsize];
-  const int mi_height = mi_size_high[bsize];
-  int xss = xd->plane[1].subsampling_x;
-  int yss = xd->plane[1].subsampling_y;
-#if CONFIG_CB4X4
-  const int unify_bsize = 1;
-#else
-  const int unify_bsize = 0;
-#endif
-  int b_sub8x8 = (bsize < BLOCK_8X8) && !unify_bsize ? 1 : 0;
-  BLOCK_SIZE extend_bsize;
-  int mi_row_pred, mi_col_pred;
-
-  int wide_unit, high_unit;
-  int i, j;
-  int ext_offset = 0;
-
-  if (dir == 0 || dir == 1) {
-    extend_bsize =
-        (mi_width == mi_size_wide[BLOCK_8X8] || bsize < BLOCK_8X8 || xss < yss)
-            ? BLOCK_8X8
-            : BLOCK_16X8;
-#if CONFIG_CB4X4
-    if (bsize < BLOCK_8X8) {
-      extend_bsize = BLOCK_4X4;
-      ext_offset = mi_size_wide[BLOCK_8X8];
-    }
-#endif
+    if (left > b_w) left = b_w;
 
-    wide_unit = mi_size_wide[extend_bsize];
-    high_unit = mi_size_high[extend_bsize];
-
-    mi_row_pred = mi_row + ((dir == 0) ? mi_height : -(mi_height + ext_offset));
-    mi_col_pred = mi_col;
-
-    for (j = 0; j < mi_height + ext_offset; j += high_unit)
-      for (i = 0; i < mi_width + ext_offset; i += wide_unit)
-        dec_predict_b_extend(pbi, xd, tile, block, mi_row_ori, mi_col_ori,
-                             mi_row_pred + j, mi_col_pred + i, mi_row_top,
-                             mi_col_top, plane, dst_buf, dst_stride, top_bsize,
-                             extend_bsize, b_sub8x8, 1);
-  } else if (dir == 2 || dir == 3) {
-    extend_bsize =
-        (mi_height == mi_size_high[BLOCK_8X8] || bsize < BLOCK_8X8 || yss < xss)
-            ? BLOCK_8X8
-            : BLOCK_8X16;
-#if CONFIG_CB4X4
-    if (bsize < BLOCK_8X8) {
-      extend_bsize = BLOCK_4X4;
-      ext_offset = mi_size_wide[BLOCK_8X8];
-    }
-#endif
+    if (x + b_w > w) right = x + b_w - w;
 
-    wide_unit = mi_size_wide[extend_bsize];
-    high_unit = mi_size_high[extend_bsize];
+    if (right > b_w) right = b_w;
 
-    mi_row_pred = mi_row;
-    mi_col_pred = mi_col + ((dir == 3) ? mi_width : -(mi_width + ext_offset));
+    copy = b_w - left - right;
 
-    for (j = 0; j < mi_height + ext_offset; j += high_unit)
-      for (i = 0; i < mi_width + ext_offset; i += wide_unit)
-        dec_predict_b_extend(pbi, xd, tile, block, mi_row_ori, mi_col_ori,
-                             mi_row_pred + j, mi_col_pred + i, mi_row_top,
-                             mi_col_top, plane, dst_buf, dst_stride, top_bsize,
-                             extend_bsize, b_sub8x8, 1);
-  } else {
-    extend_bsize = BLOCK_8X8;
-#if CONFIG_CB4X4
-    if (bsize < BLOCK_8X8) {
-      extend_bsize = BLOCK_4X4;
-      ext_offset = mi_size_wide[BLOCK_8X8];
+    if (left) memset(dst, ref_row[0], left);
+
+    if (copy) memcpy(dst + left, ref_row + x + left, copy);
+
+    if (right) memset(dst + left + copy, ref_row[w - 1], right);
+
+    dst += dst_stride;
+    ++y;
+
+    if (y > 0 && y < h) ref_row += src_stride;
+  } while (--b_h);
+}
+
+static INLINE int update_extend_mc_border_params(
+    const struct scale_factors *const sf, struct buf_2d *const pre_buf,
+    MV32 scaled_mv, PadBlock *block, int subpel_x_mv, int subpel_y_mv,
+    int do_warp, int is_intrabc, int *x_pad, int *y_pad) {
+  const int is_scaled = av1_is_scaled(sf);
+  // Get reference width and height.
+  int frame_width = pre_buf->width;
+  int frame_height = pre_buf->height;
+
+  // Do border extension if there is motion or
+  // width/height is not a multiple of 8 pixels.
+  if ((!is_intrabc) && (!do_warp) &&
+      (is_scaled || scaled_mv.col || scaled_mv.row || (frame_width & 0x7) ||
+       (frame_height & 0x7))) {
+    if (subpel_x_mv || (sf->x_step_q4 != SUBPEL_SHIFTS)) {
+      block->x0 -= AOM_INTERP_EXTEND - 1;
+      block->x1 += AOM_INTERP_EXTEND;
+      *x_pad = 1;
     }
-#endif
-    wide_unit = mi_size_wide[extend_bsize];
-    high_unit = mi_size_high[extend_bsize];
 
-    mi_row_pred = mi_row + ((dir == 4 || dir == 6) ? mi_height
-                                                   : -(mi_height + ext_offset));
-    mi_col_pred =
-        mi_col + ((dir == 6 || dir == 7) ? mi_width : -(mi_width + ext_offset));
+    if (subpel_y_mv || (sf->y_step_q4 != SUBPEL_SHIFTS)) {
+      block->y0 -= AOM_INTERP_EXTEND - 1;
+      block->y1 += AOM_INTERP_EXTEND;
+      *y_pad = 1;
+    }
 
-    for (j = 0; j < mi_height + ext_offset; j += high_unit)
-      for (i = 0; i < mi_width + ext_offset; i += wide_unit)
-        dec_predict_b_extend(pbi, xd, tile, block, mi_row_ori, mi_col_ori,
-                             mi_row_pred + j, mi_col_pred + i, mi_row_top,
-                             mi_col_top, plane, dst_buf, dst_stride, top_bsize,
-                             extend_bsize, b_sub8x8, 1);
+    // Skip border extension if block is inside the frame.
+    if (block->x0 < 0 || block->x1 > frame_width - 1 || block->y0 < 0 ||
+        block->y1 > frame_height - 1) {
+      return 1;
+    }
   }
+  return 0;
 }
 
-static void dec_extend_all(AV1Decoder *const pbi, MACROBLOCKD *const xd,
-                           const TileInfo *const tile, int block,
-                           BLOCK_SIZE bsize, BLOCK_SIZE top_bsize,
-                           int mi_row_ori, int mi_col_ori, int mi_row,
-                           int mi_col, int mi_row_top, int mi_col_top,
-                           int plane, uint8_t *dst_buf, int dst_stride) {
-  for (int i = 0; i < 8; ++i) {
-    dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row_ori,
-                   mi_col_ori, mi_row, mi_col, mi_row_top, mi_col_top, plane,
-                   dst_buf, dst_stride, i);
+static INLINE void extend_mc_border(const struct scale_factors *const sf,
+                                    struct buf_2d *const pre_buf,
+                                    MV32 scaled_mv, PadBlock block,
+                                    int subpel_x_mv, int subpel_y_mv,
+                                    int do_warp, int is_intrabc, int highbd,
+                                    uint8_t *mc_buf, uint8_t **pre,
+                                    int *src_stride) {
+  int x_pad = 0, y_pad = 0;
+  if (update_extend_mc_border_params(sf, pre_buf, scaled_mv, &block,
+                                     subpel_x_mv, subpel_y_mv, do_warp,
+                                     is_intrabc, &x_pad, &y_pad)) {
+    // Get reference block pointer.
+    const uint8_t *const buf_ptr =
+        pre_buf->buf0 + block.y0 * pre_buf->stride + block.x0;
+    int buf_stride = pre_buf->stride;
+    const int b_w = block.x1 - block.x0;
+    const int b_h = block.y1 - block.y0;
+
+    // Extend the border.
+    if (highbd) {
+      highbd_build_mc_border(buf_ptr, buf_stride, mc_buf, b_w, block.x0,
+                             block.y0, b_w, b_h, pre_buf->width,
+                             pre_buf->height);
+    } else {
+      build_mc_border(buf_ptr, buf_stride, mc_buf, b_w, block.x0, block.y0, b_w,
+                      b_h, pre_buf->width, pre_buf->height);
+    }
+    *src_stride = b_w;
+    *pre = mc_buf + y_pad * (AOM_INTERP_EXTEND - 1) * b_w +
+           x_pad * (AOM_INTERP_EXTEND - 1);
   }
 }
 
-static void dec_predict_sb_complex(AV1Decoder *const pbi, MACROBLOCKD *const xd,
-                                   const TileInfo *const tile, int mi_row,
-                                   int mi_col, int mi_row_top, int mi_col_top,
-                                   BLOCK_SIZE bsize, BLOCK_SIZE top_bsize,
-                                   uint8_t *dst_buf[3], int dst_stride[3]) {
-  const AV1_COMMON *const cm = &pbi->common;
-  const int hbs = mi_size_wide[bsize] / 2;
-  const PARTITION_TYPE partition = get_partition(cm, mi_row, mi_col, bsize);
-  const BLOCK_SIZE subsize = get_subsize(bsize, partition);
-#if CONFIG_EXT_PARTITION_TYPES
-  const BLOCK_SIZE bsize2 = get_subsize(bsize, PARTITION_SPLIT);
-#endif
-  int i;
-  const int mi_offset = mi_row * cm->mi_stride + mi_col;
-  uint8_t *dst_buf1[3], *dst_buf2[3], *dst_buf3[3];
-#if CONFIG_CB4X4
-  const int unify_bsize = 1;
-#else
-  const int unify_bsize = 0;
-#endif
-
-  DECLARE_ALIGNED(16, uint8_t, tmp_buf1[MAX_MB_PLANE * MAX_TX_SQUARE * 2]);
-  DECLARE_ALIGNED(16, uint8_t, tmp_buf2[MAX_MB_PLANE * MAX_TX_SQUARE * 2]);
-  DECLARE_ALIGNED(16, uint8_t, tmp_buf3[MAX_MB_PLANE * MAX_TX_SQUARE * 2]);
-  int dst_stride1[3] = { MAX_TX_SIZE, MAX_TX_SIZE, MAX_TX_SIZE };
-  int dst_stride2[3] = { MAX_TX_SIZE, MAX_TX_SIZE, MAX_TX_SIZE };
-  int dst_stride3[3] = { MAX_TX_SIZE, MAX_TX_SIZE, MAX_TX_SIZE };
-
-#if CONFIG_HIGHBITDEPTH
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-    int len = sizeof(uint16_t);
-    dst_buf1[0] = CONVERT_TO_BYTEPTR(tmp_buf1);
-    dst_buf1[1] = CONVERT_TO_BYTEPTR(tmp_buf1 + MAX_TX_SQUARE * len);
-    dst_buf1[2] = CONVERT_TO_BYTEPTR(tmp_buf1 + 2 * MAX_TX_SQUARE * len);
-    dst_buf2[0] = CONVERT_TO_BYTEPTR(tmp_buf2);
-    dst_buf2[1] = CONVERT_TO_BYTEPTR(tmp_buf2 + MAX_TX_SQUARE * len);
-    dst_buf2[2] = CONVERT_TO_BYTEPTR(tmp_buf2 + 2 * MAX_TX_SQUARE * len);
-    dst_buf3[0] = CONVERT_TO_BYTEPTR(tmp_buf3);
-    dst_buf3[1] = CONVERT_TO_BYTEPTR(tmp_buf3 + MAX_TX_SQUARE * len);
-    dst_buf3[2] = CONVERT_TO_BYTEPTR(tmp_buf3 + 2 * MAX_TX_SQUARE * len);
+static INLINE void dec_calc_subpel_params(
+    MACROBLOCKD *xd, const struct scale_factors *const sf, const MV mv,
+    int plane, const int pre_x, const int pre_y, int x, int y,
+    struct buf_2d *const pre_buf, SubpelParams *subpel_params, int bw, int bh,
+    PadBlock *block, int mi_x, int mi_y, MV32 *scaled_mv, int *subpel_x_mv,
+    int *subpel_y_mv) {
+  struct macroblockd_plane *const pd = &xd->plane[plane];
+  const int is_scaled = av1_is_scaled(sf);
+  if (is_scaled) {
+    int ssx = pd->subsampling_x;
+    int ssy = pd->subsampling_y;
+    int orig_pos_y = (pre_y + y) << SUBPEL_BITS;
+    orig_pos_y += mv.row * (1 << (1 - ssy));
+    int orig_pos_x = (pre_x + x) << SUBPEL_BITS;
+    orig_pos_x += mv.col * (1 << (1 - ssx));
+    int pos_y = sf->scale_value_y(orig_pos_y, sf);
+    int pos_x = sf->scale_value_x(orig_pos_x, sf);
+    pos_x += SCALE_EXTRA_OFF;
+    pos_y += SCALE_EXTRA_OFF;
+
+    const int top = -AOM_LEFT_TOP_MARGIN_SCALED(ssy);
+    const int left = -AOM_LEFT_TOP_MARGIN_SCALED(ssx);
+    const int bottom = (pre_buf->height + AOM_INTERP_EXTEND)
+                       << SCALE_SUBPEL_BITS;
+    const int right = (pre_buf->width + AOM_INTERP_EXTEND) << SCALE_SUBPEL_BITS;
+    pos_y = clamp(pos_y, top, bottom);
+    pos_x = clamp(pos_x, left, right);
+
+    subpel_params->subpel_x = pos_x & SCALE_SUBPEL_MASK;
+    subpel_params->subpel_y = pos_y & SCALE_SUBPEL_MASK;
+    subpel_params->xs = sf->x_step_q4;
+    subpel_params->ys = sf->y_step_q4;
+
+    // Get reference block top left coordinate.
+    block->x0 = pos_x >> SCALE_SUBPEL_BITS;
+    block->y0 = pos_y >> SCALE_SUBPEL_BITS;
+
+    // Get reference block bottom right coordinate.
+    block->x1 =
+        ((pos_x + (bw - 1) * subpel_params->xs) >> SCALE_SUBPEL_BITS) + 1;
+    block->y1 =
+        ((pos_y + (bh - 1) * subpel_params->ys) >> SCALE_SUBPEL_BITS) + 1;
+
+    MV temp_mv;
+    temp_mv = clamp_mv_to_umv_border_sb(xd, &mv, bw, bh, pd->subsampling_x,
+                                        pd->subsampling_y);
+    *scaled_mv = av1_scale_mv(&temp_mv, (mi_x + x), (mi_y + y), sf);
+    scaled_mv->row += SCALE_EXTRA_OFF;
+    scaled_mv->col += SCALE_EXTRA_OFF;
+
+    *subpel_x_mv = scaled_mv->col & SCALE_SUBPEL_MASK;
+    *subpel_y_mv = scaled_mv->row & SCALE_SUBPEL_MASK;
   } else {
-#endif
-    dst_buf1[0] = tmp_buf1;
-    dst_buf1[1] = tmp_buf1 + MAX_TX_SQUARE;
-    dst_buf1[2] = tmp_buf1 + 2 * MAX_TX_SQUARE;
-    dst_buf2[0] = tmp_buf2;
-    dst_buf2[1] = tmp_buf2 + MAX_TX_SQUARE;
-    dst_buf2[2] = tmp_buf2 + 2 * MAX_TX_SQUARE;
-    dst_buf3[0] = tmp_buf3;
-    dst_buf3[1] = tmp_buf3 + MAX_TX_SQUARE;
-    dst_buf3[2] = tmp_buf3 + 2 * MAX_TX_SQUARE;
-#if CONFIG_HIGHBITDEPTH
+    // Get block position in current frame.
+    int pos_x = (pre_x + x) << SUBPEL_BITS;
+    int pos_y = (pre_y + y) << SUBPEL_BITS;
+
+    const MV mv_q4 = clamp_mv_to_umv_border_sb(
+        xd, &mv, bw, bh, pd->subsampling_x, pd->subsampling_y);
+    subpel_params->xs = subpel_params->ys = SCALE_SUBPEL_SHIFTS;
+    subpel_params->subpel_x = (mv_q4.col & SUBPEL_MASK) << SCALE_EXTRA_BITS;
+    subpel_params->subpel_y = (mv_q4.row & SUBPEL_MASK) << SCALE_EXTRA_BITS;
+
+    // Get reference block top left coordinate.
+    pos_x += mv_q4.col;
+    pos_y += mv_q4.row;
+    block->x0 = pos_x >> SUBPEL_BITS;
+    block->y0 = pos_y >> SUBPEL_BITS;
+
+    // Get reference block bottom right coordinate.
+    block->x1 = (pos_x >> SUBPEL_BITS) + (bw - 1) + 1;
+    block->y1 = (pos_y >> SUBPEL_BITS) + (bh - 1) + 1;
+
+    scaled_mv->row = mv_q4.row;
+    scaled_mv->col = mv_q4.col;
+    *subpel_x_mv = scaled_mv->col & SUBPEL_MASK;
+    *subpel_y_mv = scaled_mv->row & SUBPEL_MASK;
   }
-#endif
-
-  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return;
-
-  xd->mi = cm->mi_grid_visible + mi_offset;
-  xd->mi[0] = cm->mi + mi_offset;
+}
 
-  for (i = 0; i < MAX_MB_PLANE; i++) {
-    xd->plane[i].dst.buf = dst_buf[i];
-    xd->plane[i].dst.stride = dst_stride[i];
+static INLINE void dec_build_inter_predictors(const AV1_COMMON *cm,
+                                              MACROBLOCKD *xd, int plane,
+                                              const MB_MODE_INFO *mi,
+                                              int build_for_obmc, int bw,
+                                              int bh, int mi_x, int mi_y) {
+  struct macroblockd_plane *const pd = &xd->plane[plane];
+  int is_compound = has_second_ref(mi);
+  int ref;
+  const int is_intrabc = is_intrabc_block(mi);
+  assert(IMPLIES(is_intrabc, !is_compound));
+  int is_global[2] = { 0, 0 };
+  for (ref = 0; ref < 1 + is_compound; ++ref) {
+    const WarpedMotionParams *const wm = &xd->global_motion[mi->ref_frame[ref]];
+    is_global[ref] = is_global_mv_block(mi, wm->wmtype);
+  }
+
+  const BLOCK_SIZE bsize = mi->sb_type;
+  const int ss_x = pd->subsampling_x;
+  const int ss_y = pd->subsampling_y;
+  int sub8x8_inter = (block_size_wide[bsize] < 8 && ss_x) ||
+                     (block_size_high[bsize] < 8 && ss_y);
+
+  if (is_intrabc) sub8x8_inter = 0;
+
+  // For sub8x8 chroma blocks, we may be covering more than one luma block's
+  // worth of pixels. Thus (mi_x, mi_y) may not be the correct coordinates for
+  // the top-left corner of the prediction source - the correct top-left corner
+  // is at (pre_x, pre_y).
+  const int row_start =
+      (block_size_high[bsize] == 4) && ss_y && !build_for_obmc ? -1 : 0;
+  const int col_start =
+      (block_size_wide[bsize] == 4) && ss_x && !build_for_obmc ? -1 : 0;
+  const int pre_x = (mi_x + MI_SIZE * col_start) >> ss_x;
+  const int pre_y = (mi_y + MI_SIZE * row_start) >> ss_y;
+
+  sub8x8_inter = sub8x8_inter && !build_for_obmc;
+  if (sub8x8_inter) {
+    for (int row = row_start; row <= 0 && sub8x8_inter; ++row) {
+      for (int col = col_start; col <= 0; ++col) {
+        const MB_MODE_INFO *this_mbmi = xd->mi[row * xd->mi_stride + col];
+        if (!is_inter_block(this_mbmi)) sub8x8_inter = 0;
+        if (is_intrabc_block(this_mbmi)) sub8x8_inter = 0;
+      }
+    }
   }
 
-  switch (partition) {
-    case PARTITION_NONE:
-      assert(bsize < top_bsize);
-      for (i = 0; i < MAX_MB_PLANE; i++) {
-        dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
-                             mi_row_top, mi_col_top, i, dst_buf[i],
-                             dst_stride[i], top_bsize, bsize, 0, 0);
-        dec_extend_all(pbi, xd, tile, 0, bsize, top_bsize, mi_row, mi_col,
-                       mi_row, mi_col, mi_row_top, mi_col_top, i, dst_buf[i],
-                       dst_stride[i]);
-      }
-      break;
-    case PARTITION_HORZ:
-      if (bsize == BLOCK_8X8 && !unify_bsize) {
-        for (i = 0; i < MAX_MB_PLANE; i++) {
-          // For sub8x8, predict in 8x8 unit
-          // First half
-          dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
-                               mi_row_top, mi_col_top, i, dst_buf[i],
-                               dst_stride[i], top_bsize, BLOCK_8X8, 1, 0);
-          if (bsize < top_bsize)
-            dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
-                           mi_row, mi_col, mi_row_top, mi_col_top, i,
-                           dst_buf[i], dst_stride[i]);
-
-          // Second half
-          dec_predict_b_extend(pbi, xd, tile, 2, mi_row, mi_col, mi_row, mi_col,
-                               mi_row_top, mi_col_top, i, dst_buf1[i],
-                               dst_stride1[i], top_bsize, BLOCK_8X8, 1, 1);
-          if (bsize < top_bsize)
-            dec_extend_all(pbi, xd, tile, 2, subsize, top_bsize, mi_row, mi_col,
-                           mi_row, mi_col, mi_row_top, mi_col_top, i,
-                           dst_buf1[i], dst_stride1[i]);
+  if (sub8x8_inter) {
+    // block size
+    const int b4_w = block_size_wide[bsize] >> ss_x;
+    const int b4_h = block_size_high[bsize] >> ss_y;
+    const BLOCK_SIZE plane_bsize = scale_chroma_bsize(bsize, ss_x, ss_y);
+    const int b8_w = block_size_wide[plane_bsize] >> ss_x;
+    const int b8_h = block_size_high[plane_bsize] >> ss_y;
+    assert(!is_compound);
+
+    const struct buf_2d orig_pred_buf[2] = { pd->pre[0], pd->pre[1] };
+
+    int row = row_start;
+    int src_stride;
+    for (int y = 0; y < b8_h; y += b4_h) {
+      int col = col_start;
+      for (int x = 0; x < b8_w; x += b4_w) {
+        MB_MODE_INFO *this_mbmi = xd->mi[row * xd->mi_stride + col];
+        is_compound = has_second_ref(this_mbmi);
+        DECLARE_ALIGNED(32, CONV_BUF_TYPE, tmp_dst[8 * 8]);
+        int tmp_dst_stride = 8;
+        assert(bw < 8 || bh < 8);
+        ConvolveParams conv_params = get_conv_params_no_round(
+            0, 0, plane, tmp_dst, tmp_dst_stride, is_compound, xd->bd);
+        conv_params.use_jnt_comp_avg = 0;
+        struct buf_2d *const dst_buf = &pd->dst;
+        uint8_t *dst = dst_buf->buf + dst_buf->stride * y + x;
+
+        ref = 0;
+        const RefBuffer *ref_buf =
+            &cm->frame_refs[this_mbmi->ref_frame[ref] - LAST_FRAME];
+
+        pd->pre[ref].buf0 =
+            (plane == 1) ? ref_buf->buf->u_buffer : ref_buf->buf->v_buffer;
+        pd->pre[ref].buf =
+            pd->pre[ref].buf0 + scaled_buffer_offset(pre_x, pre_y,
+                                                     ref_buf->buf->uv_stride,
+                                                     &ref_buf->sf);
+        pd->pre[ref].width = ref_buf->buf->uv_crop_width;
+        pd->pre[ref].height = ref_buf->buf->uv_crop_height;
+        pd->pre[ref].stride = ref_buf->buf->uv_stride;
+
+        const struct scale_factors *const sf =
+            is_intrabc ? &cm->sf_identity : &ref_buf->sf;
+        struct buf_2d *const pre_buf = is_intrabc ? dst_buf : &pd->pre[ref];
+
+        const MV mv = this_mbmi->mv[ref].as_mv;
+
+        uint8_t *pre;
+        SubpelParams subpel_params;
+        PadBlock block;
+        MV32 scaled_mv;
+        int subpel_x_mv, subpel_y_mv;
+        int highbd;
+        WarpTypesAllowed warp_types;
+        warp_types.global_warp_allowed = is_global[ref];
+        warp_types.local_warp_allowed = this_mbmi->motion_mode == WARPED_CAUSAL;
+
+        dec_calc_subpel_params(xd, sf, mv, plane, pre_x, pre_y, x, y, pre_buf,
+                               &subpel_params, bw, bh, &block, mi_x, mi_y,
+                               &scaled_mv, &subpel_x_mv, &subpel_y_mv);
+        pre = pre_buf->buf0 + block.y0 * pre_buf->stride + block.x0;
+        src_stride = pre_buf->stride;
+        highbd = xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH;
+        extend_mc_border(sf, pre_buf, scaled_mv, block, subpel_x_mv,
+                         subpel_y_mv, 0, is_intrabc, highbd, xd->mc_buf[ref],
+                         &pre, &src_stride);
+        conv_params.ref = ref;
+        conv_params.do_average = ref;
+        if (is_masked_compound_type(mi->interinter_comp.type)) {
+          // masked compound type has its own average mechanism
+          conv_params.do_average = 0;
         }
 
-        // weighted average to smooth the boundary
-        xd->plane[0].dst.buf = dst_buf[0];
-        xd->plane[0].dst.stride = dst_stride[0];
-        av1_build_masked_inter_predictor_complex(
-            xd, dst_buf[0], dst_stride[0], dst_buf1[0], dst_stride1[0], mi_row,
-            mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_HORZ,
-            0);
-      } else {
-        for (i = 0; i < MAX_MB_PLANE; i++) {
-#if CONFIG_CB4X4
-          const struct macroblockd_plane *pd = &xd->plane[i];
-          int handle_chroma_sub8x8 = need_handle_chroma_sub8x8(
-              subsize, pd->subsampling_x, pd->subsampling_y);
-
-          if (handle_chroma_sub8x8) {
-            int mode_offset_row = CONFIG_CHROMA_SUB8X8 ? hbs : 0;
-
-            dec_predict_b_extend(pbi, xd, tile, 0, mi_row + mode_offset_row,
-                                 mi_col, mi_row, mi_col, mi_row_top, mi_col_top,
-                                 i, dst_buf[i], dst_stride[i], top_bsize, bsize,
-                                 0, 0);
-            if (bsize < top_bsize)
-              dec_extend_all(pbi, xd, tile, 0, bsize, top_bsize,
-                             mi_row + mode_offset_row, mi_col, mi_row, mi_col,
-                             mi_row_top, mi_col_top, i, dst_buf[i],
-                             dst_stride[i]);
-          } else {
-#endif
-            // First half
-            dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row,
-                                 mi_col, mi_row_top, mi_col_top, i, dst_buf[i],
-                                 dst_stride[i], top_bsize, subsize, 0, 0);
-            if (bsize < top_bsize)
-              dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
-                             mi_col, mi_row, mi_col, mi_row_top, mi_col_top, i,
-                             dst_buf[i], dst_stride[i]);
-            else
-              dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
-                             mi_col, mi_row, mi_col, mi_row_top, mi_col_top, i,
-                             dst_buf[i], dst_stride[i], 0);
-
-            if (mi_row + hbs < cm->mi_rows) {
-              // Second half
-              dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col,
-                                   mi_row + hbs, mi_col, mi_row_top, mi_col_top,
-                                   i, dst_buf1[i], dst_stride1[i], top_bsize,
-                                   subsize, 0, 0);
-              if (bsize < top_bsize)
-                dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize,
-                               mi_row + hbs, mi_col, mi_row + hbs, mi_col,
-                               mi_row_top, mi_col_top, i, dst_buf1[i],
-                               dst_stride1[i]);
-              else
-                dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize,
-                               mi_row + hbs, mi_col, mi_row + hbs, mi_col,
-                               mi_row_top, mi_col_top, i, dst_buf1[i],
-                               dst_stride1[i], 1);
-
-              // weighted average to smooth the boundary
-              xd->plane[i].dst.buf = dst_buf[i];
-              xd->plane[i].dst.stride = dst_stride[i];
-              av1_build_masked_inter_predictor_complex(
-                  xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i],
-                  mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
-                  PARTITION_HORZ, i);
-            }
-#if CONFIG_CB4X4
-          }
-#endif
-        }
-      }
-      break;
-    case PARTITION_VERT:
-      if (bsize == BLOCK_8X8 && !unify_bsize) {
-        for (i = 0; i < MAX_MB_PLANE; i++) {
-          // First half
-          dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
-                               mi_row_top, mi_col_top, i, dst_buf[i],
-                               dst_stride[i], top_bsize, BLOCK_8X8, 1, 0);
-          if (bsize < top_bsize)
-            dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
-                           mi_row, mi_col, mi_row_top, mi_col_top, i,
-                           dst_buf[i], dst_stride[i]);
-
-          // Second half
-          dec_predict_b_extend(pbi, xd, tile, 1, mi_row, mi_col, mi_row, mi_col,
-                               mi_row_top, mi_col_top, i, dst_buf1[i],
-                               dst_stride1[i], top_bsize, BLOCK_8X8, 1, 1);
-          if (bsize < top_bsize)
-            dec_extend_all(pbi, xd, tile, 1, subsize, top_bsize, mi_row, mi_col,
-                           mi_row, mi_col, mi_row_top, mi_col_top, i,
-                           dst_buf1[i], dst_stride1[i]);
-        }
+        av1_make_inter_predictor(
+            pre, src_stride, dst, dst_buf->stride, &subpel_params, sf, b4_w,
+            b4_h, &conv_params, this_mbmi->interp_filters, &warp_types,
+            (mi_x >> pd->subsampling_x) + x, (mi_y >> pd->subsampling_y) + y,
+            plane, ref, mi, build_for_obmc, xd, cm->allow_warped_motion);
 
-        // Smooth
-        xd->plane[0].dst.buf = dst_buf[0];
-        xd->plane[0].dst.stride = dst_stride[0];
-        av1_build_masked_inter_predictor_complex(
-            xd, dst_buf[0], dst_stride[0], dst_buf1[0], dst_stride1[0], mi_row,
-            mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_VERT,
-            0);
-      } else {
-        for (i = 0; i < MAX_MB_PLANE; i++) {
-#if CONFIG_CB4X4
-          const struct macroblockd_plane *pd = &xd->plane[i];
-          int handle_chroma_sub8x8 = need_handle_chroma_sub8x8(
-              subsize, pd->subsampling_x, pd->subsampling_y);
-
-          if (handle_chroma_sub8x8) {
-            int mode_offset_col = CONFIG_CHROMA_SUB8X8 ? hbs : 0;
-            assert(i > 0 && bsize == BLOCK_8X8);
-
-            dec_predict_b_extend(pbi, xd, tile, 0, mi_row,
-                                 mi_col + mode_offset_col, mi_row, mi_col,
-                                 mi_row_top, mi_col_top, i, dst_buf[i],
-                                 dst_stride[i], top_bsize, bsize, 0, 0);
-            if (bsize < top_bsize)
-              dec_extend_all(pbi, xd, tile, 0, bsize, top_bsize, mi_row,
-                             mi_col + mode_offset_col, mi_row, mi_col,
-                             mi_row_top, mi_col_top, i, dst_buf[i],
-                             dst_stride[i]);
-          } else {
-#endif
-            // First half
-            dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row,
-                                 mi_col, mi_row_top, mi_col_top, i, dst_buf[i],
-                                 dst_stride[i], top_bsize, subsize, 0, 0);
-            if (bsize < top_bsize)
-              dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
-                             mi_col, mi_row, mi_col, mi_row_top, mi_col_top, i,
-                             dst_buf[i], dst_stride[i]);
-            else
-              dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
-                             mi_col, mi_row, mi_col, mi_row_top, mi_col_top, i,
-                             dst_buf[i], dst_stride[i], 3);
-
-            // Second half
-            if (mi_col + hbs < cm->mi_cols) {
-              dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col + hbs,
-                                   mi_row, mi_col + hbs, mi_row_top, mi_col_top,
-                                   i, dst_buf1[i], dst_stride1[i], top_bsize,
-                                   subsize, 0, 0);
-              if (bsize < top_bsize)
-                dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
-                               mi_col + hbs, mi_row, mi_col + hbs, mi_row_top,
-                               mi_col_top, i, dst_buf1[i], dst_stride1[i]);
-              else
-                dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
-                               mi_col + hbs, mi_row, mi_col + hbs, mi_row_top,
-                               mi_col_top, i, dst_buf1[i], dst_stride1[i], 2);
-
-              // Smooth
-              xd->plane[i].dst.buf = dst_buf[i];
-              xd->plane[i].dst.stride = dst_stride[i];
-              av1_build_masked_inter_predictor_complex(
-                  xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i],
-                  mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
-                  PARTITION_VERT, i);
-            }
-#if CONFIG_CB4X4
-          }
-#endif
-        }
+        ++col;
       }
-      break;
-    case PARTITION_SPLIT:
-      if (bsize == BLOCK_8X8 && !unify_bsize) {
-        for (i = 0; i < MAX_MB_PLANE; i++) {
-          dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
-                               mi_row_top, mi_col_top, i, dst_buf[i],
-                               dst_stride[i], top_bsize, BLOCK_8X8, 1, 0);
-          dec_predict_b_extend(pbi, xd, tile, 1, mi_row, mi_col, mi_row, mi_col,
-                               mi_row_top, mi_col_top, i, dst_buf1[i],
-                               dst_stride1[i], top_bsize, BLOCK_8X8, 1, 1);
-          dec_predict_b_extend(pbi, xd, tile, 2, mi_row, mi_col, mi_row, mi_col,
-                               mi_row_top, mi_col_top, i, dst_buf2[i],
-                               dst_stride2[i], top_bsize, BLOCK_8X8, 1, 1);
-          dec_predict_b_extend(pbi, xd, tile, 3, mi_row, mi_col, mi_row, mi_col,
-                               mi_row_top, mi_col_top, i, dst_buf3[i],
-                               dst_stride3[i], top_bsize, BLOCK_8X8, 1, 1);
-          if (bsize < top_bsize) {
-            dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
-                           mi_row, mi_col, mi_row_top, mi_col_top, i,
-                           dst_buf[i], dst_stride[i]);
-            dec_extend_all(pbi, xd, tile, 1, subsize, top_bsize, mi_row, mi_col,
-                           mi_row, mi_col, mi_row_top, mi_col_top, i,
-                           dst_buf1[i], dst_stride1[i]);
-            dec_extend_all(pbi, xd, tile, 2, subsize, top_bsize, mi_row, mi_col,
-                           mi_row, mi_col, mi_row_top, mi_col_top, i,
-                           dst_buf2[i], dst_stride2[i]);
-            dec_extend_all(pbi, xd, tile, 3, subsize, top_bsize, mi_row, mi_col,
-                           mi_row, mi_col, mi_row_top, mi_col_top, i,
-                           dst_buf3[i], dst_stride3[i]);
-          }
-        }
-#if CONFIG_CB4X4
-      } else if (bsize == BLOCK_8X8) {
-        for (i = 0; i < MAX_MB_PLANE; i++) {
-          const struct macroblockd_plane *pd = &xd->plane[i];
-          int handle_chroma_sub8x8 = need_handle_chroma_sub8x8(
-              subsize, pd->subsampling_x, pd->subsampling_y);
-
-          if (handle_chroma_sub8x8) {
-            int mode_offset_row =
-                CONFIG_CHROMA_SUB8X8 && mi_row + hbs < cm->mi_rows ? hbs : 0;
-            int mode_offset_col =
-                CONFIG_CHROMA_SUB8X8 && mi_col + hbs < cm->mi_cols ? hbs : 0;
-
-            dec_predict_b_extend(pbi, xd, tile, 0, mi_row + mode_offset_row,
-                                 mi_col + mode_offset_col, mi_row, mi_col,
-                                 mi_row_top, mi_col_top, i, dst_buf[i],
-                                 dst_stride[i], top_bsize, BLOCK_8X8, 0, 0);
-            if (bsize < top_bsize)
-              dec_extend_all(pbi, xd, tile, 0, BLOCK_8X8, top_bsize,
-                             mi_row + mode_offset_row, mi_col + mode_offset_col,
-                             mi_row, mi_col, mi_row_top, mi_col_top, i,
-                             dst_buf[i], dst_stride[i]);
-          } else {
-            dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row,
-                                 mi_col, mi_row_top, mi_col_top, i, dst_buf[i],
-                                 dst_stride[i], top_bsize, subsize, 0, 0);
-            if (mi_row < cm->mi_rows && mi_col + hbs < cm->mi_cols)
-              dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col + hbs,
-                                   mi_row, mi_col + hbs, mi_row_top, mi_col_top,
-                                   i, dst_buf1[i], dst_stride1[i], top_bsize,
-                                   subsize, 0, 0);
-            if (mi_row + hbs < cm->mi_rows && mi_col < cm->mi_cols)
-              dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col,
-                                   mi_row + hbs, mi_col, mi_row_top, mi_col_top,
-                                   i, dst_buf2[i], dst_stride2[i], top_bsize,
-                                   subsize, 0, 0);
-            if (mi_row + hbs < cm->mi_rows && mi_col + hbs < cm->mi_cols)
-              dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col + hbs,
-                                   mi_row + hbs, mi_col + hbs, mi_row_top,
-                                   mi_col_top, i, dst_buf3[i], dst_stride3[i],
-                                   top_bsize, subsize, 0, 0);
-
-            if (bsize < top_bsize) {
-              dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
-                             mi_col, mi_row, mi_col, mi_row_top, mi_col_top, i,
-                             dst_buf[i], dst_stride[i]);
-              if (mi_row < cm->mi_rows && mi_col + hbs < cm->mi_cols)
-                dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
-                               mi_col + hbs, mi_row, mi_col + hbs, mi_row_top,
-                               mi_col_top, i, dst_buf1[i], dst_stride1[i]);
-              if (mi_row + hbs < cm->mi_rows && mi_col < cm->mi_cols)
-                dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize,
-                               mi_row + hbs, mi_col, mi_row + hbs, mi_col,
-                               mi_row_top, mi_col_top, i, dst_buf2[i],
-                               dst_stride2[i]);
-              if (mi_row + hbs < cm->mi_rows && mi_col + hbs < cm->mi_cols)
-                dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize,
-                               mi_row + hbs, mi_col + hbs, mi_row + hbs,
-                               mi_col + hbs, mi_row_top, mi_col_top, i,
-                               dst_buf3[i], dst_stride3[i]);
-            }
-          }
-        }
-#endif
-      } else {
-        dec_predict_sb_complex(pbi, xd, tile, mi_row, mi_col, mi_row_top,
-                               mi_col_top, subsize, top_bsize, dst_buf,
-                               dst_stride);
-        if (mi_row < cm->mi_rows && mi_col + hbs < cm->mi_cols)
-          dec_predict_sb_complex(pbi, xd, tile, mi_row, mi_col + hbs,
-                                 mi_row_top, mi_col_top, subsize, top_bsize,
-                                 dst_buf1, dst_stride1);
-        if (mi_row + hbs < cm->mi_rows && mi_col < cm->mi_cols)
-          dec_predict_sb_complex(pbi, xd, tile, mi_row + hbs, mi_col,
-                                 mi_row_top, mi_col_top, subsize, top_bsize,
-                                 dst_buf2, dst_stride2);
-        if (mi_row + hbs < cm->mi_rows && mi_col + hbs < cm->mi_cols)
-          dec_predict_sb_complex(pbi, xd, tile, mi_row + hbs, mi_col + hbs,
-                                 mi_row_top, mi_col_top, subsize, top_bsize,
-                                 dst_buf3, dst_stride3);
-      }
-      for (i = 0; i < MAX_MB_PLANE; i++) {
-#if CONFIG_CB4X4
-        const struct macroblockd_plane *pd = &xd->plane[i];
-        int handle_chroma_sub8x8 = need_handle_chroma_sub8x8(
-            subsize, pd->subsampling_x, pd->subsampling_y);
-        if (handle_chroma_sub8x8) continue;  // Skip <4x4 chroma smoothing
-#else
-        if (bsize == BLOCK_8X8 && i != 0)
-          continue;  // Skip <4x4 chroma smoothing
-#endif
-        if (mi_row < cm->mi_rows && mi_col + hbs < cm->mi_cols) {
-          av1_build_masked_inter_predictor_complex(
-              xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i],
-              mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
-              PARTITION_VERT, i);
-          if (mi_row + hbs < cm->mi_rows) {
-            av1_build_masked_inter_predictor_complex(
-                xd, dst_buf2[i], dst_stride2[i], dst_buf3[i], dst_stride3[i],
-                mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
-                PARTITION_VERT, i);
-            av1_build_masked_inter_predictor_complex(
-                xd, dst_buf[i], dst_stride[i], dst_buf2[i], dst_stride2[i],
-                mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
-                PARTITION_HORZ, i);
-          }
-        } else if (mi_row + hbs < cm->mi_rows && mi_col < cm->mi_cols) {
-          av1_build_masked_inter_predictor_complex(
-              xd, dst_buf[i], dst_stride[i], dst_buf2[i], dst_stride2[i],
-              mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
-              PARTITION_HORZ, i);
-        }
-      }
-      break;
-#if CONFIG_EXT_PARTITION_TYPES
-#if CONFIG_EXT_PARTITION_TYPES_AB
-#error HORZ/VERT_A/B partitions not yet updated in superres code
-#endif
-    case PARTITION_HORZ_A:
-      dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
-                           mi_row_top, mi_col_top, dst_buf, dst_stride,
-                           top_bsize, bsize2, 0, 0);
-      dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row, mi_col,
-                     mi_row_top, mi_col_top, dst_buf, dst_stride);
-
-      dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col + hbs, mi_row,
-                           mi_col + hbs, mi_row_top, mi_col_top, dst_buf1,
-                           dst_stride1, top_bsize, bsize2, 0, 0);
-      dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row, mi_col + hbs,
-                     mi_row_top, mi_col_top, dst_buf1, dst_stride1);
-
-      dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col, mi_row + hbs,
-                           mi_col, mi_row_top, mi_col_top, dst_buf2,
-                           dst_stride2, top_bsize, subsize, 0, 0);
-      if (bsize < top_bsize)
-        dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row + hbs,
-                       mi_col, mi_row_top, mi_col_top, dst_buf2, dst_stride2);
-      else
-        dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row + hbs,
-                       mi_col, mi_row_top, mi_col_top, dst_buf2, dst_stride2,
-                       1);
-
-      for (i = 0; i < MAX_MB_PLANE; i++) {
-        xd->plane[i].dst.buf = dst_buf[i];
-        xd->plane[i].dst.stride = dst_stride[i];
-        av1_build_masked_inter_predictor_complex(
-            xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i], mi_row,
-            mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_VERT,
-            i);
-      }
-      for (i = 0; i < MAX_MB_PLANE; i++) {
-        av1_build_masked_inter_predictor_complex(
-            xd, dst_buf[i], dst_stride[i], dst_buf2[i], dst_stride2[i], mi_row,
-            mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_HORZ,
-            i);
-      }
-      break;
-    case PARTITION_VERT_A:
+      ++row;
+    }
 
-      dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
-                           mi_row_top, mi_col_top, dst_buf, dst_stride,
-                           top_bsize, bsize2, 0, 0);
-      dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row, mi_col,
-                     mi_row_top, mi_col_top, dst_buf, dst_stride);
-
-      dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col, mi_row + hbs,
-                           mi_col, mi_row_top, mi_col_top, dst_buf1,
-                           dst_stride1, top_bsize, bsize2, 0, 0);
-      dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row + hbs, mi_col,
-                     mi_row_top, mi_col_top, dst_buf1, dst_stride1);
-
-      dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col + hbs, mi_row,
-                           mi_col + hbs, mi_row_top, mi_col_top, dst_buf2,
-                           dst_stride2, top_bsize, subsize, 0, 0);
-      if (bsize < top_bsize)
-        dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
-                       mi_col + hbs, mi_row_top, mi_col_top, dst_buf2,
-                       dst_stride2);
-      else
-        dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row,
-                       mi_col + hbs, mi_row_top, mi_col_top, dst_buf2,
-                       dst_stride2, 2);
-
-      for (i = 0; i < MAX_MB_PLANE; i++) {
-        xd->plane[i].dst.buf = dst_buf[i];
-        xd->plane[i].dst.stride = dst_stride[i];
-        av1_build_masked_inter_predictor_complex(
-            xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i], mi_row,
-            mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_HORZ,
-            i);
-      }
-      for (i = 0; i < MAX_MB_PLANE; i++) {
-        av1_build_masked_inter_predictor_complex(
-            xd, dst_buf[i], dst_stride[i], dst_buf2[i], dst_stride2[i], mi_row,
-            mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_VERT,
-            i);
-      }
-      break;
-    case PARTITION_HORZ_B:
-      dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
-                           mi_row_top, mi_col_top, dst_buf, dst_stride,
-                           top_bsize, subsize, 0, 0);
-      if (bsize < top_bsize)
-        dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
-                       mi_row_top, mi_col_top, dst_buf, dst_stride);
-      else
-        dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
-                       mi_row_top, mi_col_top, dst_buf, dst_stride, 0);
-
-      dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col, mi_row + hbs,
-                           mi_col, mi_row_top, mi_col_top, dst_buf1,
-                           dst_stride1, top_bsize, bsize2, 0, 0);
-      dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row + hbs, mi_col,
-                     mi_row_top, mi_col_top, dst_buf1, dst_stride1);
-
-      dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col + hbs,
-                           mi_row + hbs, mi_col + hbs, mi_row_top, mi_col_top,
-                           dst_buf2, dst_stride2, top_bsize, bsize2, 0, 0);
-      dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row + hbs,
-                     mi_col + hbs, mi_row_top, mi_col_top, dst_buf2,
-                     dst_stride2);
-
-      for (i = 0; i < MAX_MB_PLANE; i++) {
-        xd->plane[i].dst.buf = dst_buf1[i];
-        xd->plane[i].dst.stride = dst_stride1[i];
-        av1_build_masked_inter_predictor_complex(
-            xd, dst_buf1[i], dst_stride1[i], dst_buf2[i], dst_stride2[i],
-            mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
-            PARTITION_VERT, i);
-      }
-      for (i = 0; i < MAX_MB_PLANE; i++) {
-        xd->plane[i].dst.buf = dst_buf[i];
-        xd->plane[i].dst.stride = dst_stride[i];
-        av1_build_masked_inter_predictor_complex(
-            xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i], mi_row,
-            mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_HORZ,
-            i);
+    for (ref = 0; ref < 2; ++ref) pd->pre[ref] = orig_pred_buf[ref];
+    return;
+  }
+
+  {
+    struct buf_2d *const dst_buf = &pd->dst;
+    uint8_t *const dst = dst_buf->buf;
+    uint8_t *pre[2];
+    SubpelParams subpel_params[2];
+    DECLARE_ALIGNED(32, uint16_t, tmp_dst[MAX_SB_SIZE * MAX_SB_SIZE]);
+    int src_stride[2];
+    for (ref = 0; ref < 1 + is_compound; ++ref) {
+      const struct scale_factors *const sf =
+          is_intrabc ? &cm->sf_identity : &xd->block_refs[ref]->sf;
+      struct buf_2d *const pre_buf = is_intrabc ? dst_buf : &pd->pre[ref];
+      const MV mv = mi->mv[ref].as_mv;
+      PadBlock block;
+      MV32 scaled_mv;
+      int subpel_x_mv, subpel_y_mv;
+      int highbd;
+
+      dec_calc_subpel_params(xd, sf, mv, plane, pre_x, pre_y, 0, 0, pre_buf,
+                             &subpel_params[ref], bw, bh, &block, mi_x, mi_y,
+                             &scaled_mv, &subpel_x_mv, &subpel_y_mv);
+      pre[ref] = pre_buf->buf0 + block.y0 * pre_buf->stride + block.x0;
+      src_stride[ref] = pre_buf->stride;
+      highbd = xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH;
+
+      WarpTypesAllowed warp_types;
+      warp_types.global_warp_allowed = is_global[ref];
+      warp_types.local_warp_allowed = mi->motion_mode == WARPED_CAUSAL;
+      int do_warp = (bw >= 8 && bh >= 8 &&
+                     av1_allow_warp(mi, &warp_types,
+                                    &xd->global_motion[mi->ref_frame[ref]],
+                                    build_for_obmc, subpel_params[ref].xs,
+                                    subpel_params[ref].ys, NULL));
+      do_warp = (do_warp && xd->cur_frame_force_integer_mv == 0);
+
+      extend_mc_border(sf, pre_buf, scaled_mv, block, subpel_x_mv, subpel_y_mv,
+                       do_warp, is_intrabc, highbd, xd->mc_buf[ref], &pre[ref],
+                       &src_stride[ref]);
+    }
+
+    ConvolveParams conv_params = get_conv_params_no_round(
+        0, 0, plane, tmp_dst, MAX_SB_SIZE, is_compound, xd->bd);
+    av1_jnt_comp_weight_assign(cm, mi, 0, &conv_params.fwd_offset,
+                               &conv_params.bck_offset,
+                               &conv_params.use_jnt_comp_avg, is_compound);
+
+    for (ref = 0; ref < 1 + is_compound; ++ref) {
+      const struct scale_factors *const sf =
+          is_intrabc ? &cm->sf_identity : &xd->block_refs[ref]->sf;
+      WarpTypesAllowed warp_types;
+      warp_types.global_warp_allowed = is_global[ref];
+      warp_types.local_warp_allowed = mi->motion_mode == WARPED_CAUSAL;
+      conv_params.ref = ref;
+      conv_params.do_average = ref;
+      if (is_masked_compound_type(mi->interinter_comp.type)) {
+        // masked compound type has its own average mechanism
+        conv_params.do_average = 0;
       }
-      break;
-    case PARTITION_VERT_B:
-      dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col,
-                           mi_row_top, mi_col_top, dst_buf, dst_stride,
-                           top_bsize, subsize, 0, 0);
-      if (bsize < top_bsize)
-        dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
-                       mi_row_top, mi_col_top, dst_buf, dst_stride);
+
+      if (ref && is_masked_compound_type(mi->interinter_comp.type))
+        av1_make_masked_inter_predictor(
+            pre[ref], src_stride[ref], dst, dst_buf->stride,
+            &subpel_params[ref], sf, bw, bh, &conv_params, mi->interp_filters,
+            plane, &warp_types, mi_x >> pd->subsampling_x,
+            mi_y >> pd->subsampling_y, ref, xd, cm->allow_warped_motion);
       else
-        dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col,
-                       mi_row_top, mi_col_top, dst_buf, dst_stride, 3);
-
-      dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col + hbs, mi_row,
-                           mi_col + hbs, mi_row_top, mi_col_top, dst_buf1,
-                           dst_stride1, top_bsize, bsize2, 0, 0);
-      dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row, mi_col + hbs,
-                     mi_row_top, mi_col_top, dst_buf1, dst_stride1);
-
-      dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col + hbs,
-                           mi_row + hbs, mi_col + hbs, mi_row_top, mi_col_top,
-                           dst_buf2, dst_stride2, top_bsize, bsize2, 0, 0);
-      dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row + hbs,
-                     mi_col + hbs, mi_row_top, mi_col_top, dst_buf2,
-                     dst_stride2);
-
-      for (i = 0; i < MAX_MB_PLANE; i++) {
-        xd->plane[i].dst.buf = dst_buf1[i];
-        xd->plane[i].dst.stride = dst_stride1[i];
-        av1_build_masked_inter_predictor_complex(
-            xd, dst_buf1[i], dst_stride1[i], dst_buf2[i], dst_stride2[i],
-            mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize,
-            PARTITION_HORZ, i);
-      }
-      for (i = 0; i < MAX_MB_PLANE; i++) {
-        xd->plane[i].dst.buf = dst_buf[i];
-        xd->plane[i].dst.stride = dst_stride[i];
-        av1_build_masked_inter_predictor_complex(
-            xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i], mi_row,
-            mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_VERT,
-            i);
-      }
-      break;
-#endif  // CONFIG_EXT_PARTITION_TYPES
-    default: assert(0);
+        av1_make_inter_predictor(
+            pre[ref], src_stride[ref], dst, dst_buf->stride,
+            &subpel_params[ref], sf, bw, bh, &conv_params, mi->interp_filters,
+            &warp_types, mi_x >> pd->subsampling_x, mi_y >> pd->subsampling_y,
+            plane, ref, mi, build_for_obmc, xd, cm->allow_warped_motion);
+    }
   }
 }
 
-static void set_segment_id_supertx(const AV1_COMMON *const cm, int mi_row,
-                                   int mi_col, BLOCK_SIZE bsize) {
-  const struct segmentation *seg = &cm->seg;
-  const int miw = AOMMIN(mi_size_wide[bsize], cm->mi_cols - mi_col);
-  const int mih = AOMMIN(mi_size_high[bsize], cm->mi_rows - mi_row);
-  const int mi_offset = mi_row * cm->mi_stride + mi_col;
-  MODE_INFO **const mip = cm->mi_grid_visible + mi_offset;
-  int r, c;
-  int seg_id_supertx = MAX_SEGMENTS;
+static void dec_build_inter_predictors_for_planes(const AV1_COMMON *cm,
+                                                  MACROBLOCKD *xd,
+                                                  BLOCK_SIZE bsize, int mi_row,
+                                                  int mi_col, int plane_from,
+                                                  int plane_to) {
+  int plane;
+  const int mi_x = mi_col * MI_SIZE;
+  const int mi_y = mi_row * MI_SIZE;
+  for (plane = plane_from; plane <= plane_to; ++plane) {
+    const struct macroblockd_plane *pd = &xd->plane[plane];
+    const int bw = pd->width;
+    const int bh = pd->height;
+
+    if (!is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
+                             pd->subsampling_y))
+      continue;
 
-  if (!seg->enabled) {
-    seg_id_supertx = 0;
-  } else {
-    // Find the minimum segment_id
-    for (r = 0; r < mih; r++)
-      for (c = 0; c < miw; c++)
-        seg_id_supertx =
-            AOMMIN(mip[r * cm->mi_stride + c]->mbmi.segment_id, seg_id_supertx);
-    assert(0 <= seg_id_supertx && seg_id_supertx < MAX_SEGMENTS);
-  }
-
-  // Assign the the segment_id back to segment_id_supertx
-  for (r = 0; r < mih; r++)
-    for (c = 0; c < miw; c++)
-      mip[r * cm->mi_stride + c]->mbmi.segment_id_supertx = seg_id_supertx;
+    dec_build_inter_predictors(cm, xd, plane, xd->mi[0], 0, bw, bh, mi_x, mi_y);
+  }
 }
-#endif  // CONFIG_SUPERTX
 
-static void decode_mbmi_block(AV1Decoder *const pbi, MACROBLOCKD *const xd,
-#if CONFIG_SUPERTX
-                              int supertx_enabled,
-#endif  // CONFIG_SUPERTX
-                              int mi_row, int mi_col, aom_reader *r,
-#if CONFIG_EXT_PARTITION_TYPES
-                              PARTITION_TYPE partition,
-#endif  // CONFIG_EXT_PARTITION_TYPES
-                              BLOCK_SIZE bsize) {
-  AV1_COMMON *const cm = &pbi->common;
-  const int bw = mi_size_wide[bsize];
-  const int bh = mi_size_high[bsize];
-  const int x_mis = AOMMIN(bw, cm->mi_cols - mi_col);
-  const int y_mis = AOMMIN(bh, cm->mi_rows - mi_row);
+static void dec_build_inter_predictors_sby(const AV1_COMMON *cm,
+                                           MACROBLOCKD *xd, int mi_row,
+                                           int mi_col, BUFFER_SET *ctx,
+                                           BLOCK_SIZE bsize) {
+  dec_build_inter_predictors_for_planes(cm, xd, bsize, mi_row, mi_col, 0, 0);
 
-#if CONFIG_ACCOUNTING
-  aom_accounting_set_context(&pbi->accounting, mi_col, mi_row);
-#endif
-#if CONFIG_SUPERTX
-  if (supertx_enabled) {
-    set_mb_offsets(cm, xd, bsize, mi_row, mi_col, bw, bh, x_mis, y_mis);
-  } else {
-    set_offsets(cm, xd, bsize, mi_row, mi_col, bw, bh, x_mis, y_mis);
+  if (is_interintra_pred(xd->mi[0])) {
+    BUFFER_SET default_ctx = { { xd->plane[0].dst.buf, NULL, NULL },
+                               { xd->plane[0].dst.stride, 0, 0 } };
+    if (!ctx) ctx = &default_ctx;
+    av1_build_interintra_predictors_sby(cm, xd, xd->plane[0].dst.buf,
+                                        xd->plane[0].dst.stride, ctx, bsize);
   }
-#if CONFIG_EXT_PARTITION_TYPES
-  xd->mi[0]->mbmi.partition = partition;
-#endif
-  av1_read_mode_info(pbi, xd, supertx_enabled, mi_row, mi_col, r, x_mis, y_mis);
-#else
-  set_offsets(cm, xd, bsize, mi_row, mi_col, bw, bh, x_mis, y_mis);
-#if CONFIG_EXT_PARTITION_TYPES
-  xd->mi[0]->mbmi.partition = partition;
-#endif
-  av1_read_mode_info(pbi, xd, mi_row, mi_col, r, x_mis, y_mis);
-#endif  // CONFIG_SUPERTX
-  if (bsize >= BLOCK_8X8 && (cm->subsampling_x || cm->subsampling_y)) {
-    const BLOCK_SIZE uv_subsize =
-        ss_size_lookup[bsize][cm->subsampling_x][cm->subsampling_y];
-    if (uv_subsize == BLOCK_INVALID)
-      aom_internal_error(xd->error_info, AOM_CODEC_CORRUPT_FRAME,
-                         "Invalid block size.");
+}
+
+static void dec_build_inter_predictors_sbuv(const AV1_COMMON *cm,
+                                            MACROBLOCKD *xd, int mi_row,
+                                            int mi_col, BUFFER_SET *ctx,
+                                            BLOCK_SIZE bsize) {
+  dec_build_inter_predictors_for_planes(cm, xd, bsize, mi_row, mi_col, 1,
+                                        MAX_MB_PLANE - 1);
+
+  if (is_interintra_pred(xd->mi[0])) {
+    BUFFER_SET default_ctx = {
+      { NULL, xd->plane[1].dst.buf, xd->plane[2].dst.buf },
+      { 0, xd->plane[1].dst.stride, xd->plane[2].dst.stride }
+    };
+    if (!ctx) ctx = &default_ctx;
+    av1_build_interintra_predictors_sbuv(
+        cm, xd, xd->plane[1].dst.buf, xd->plane[2].dst.buf,
+        xd->plane[1].dst.stride, xd->plane[2].dst.stride, ctx, bsize);
   }
+}
 
-#if CONFIG_SUPERTX
-  xd->mi[0]->mbmi.segment_id_supertx = MAX_SEGMENTS;
-#endif  // CONFIG_SUPERTX
+static void dec_build_inter_predictors_sb(const AV1_COMMON *cm, MACROBLOCKD *xd,
+                                          int mi_row, int mi_col,
+                                          BUFFER_SET *ctx, BLOCK_SIZE bsize) {
+  const int num_planes = av1_num_planes(cm);
+  dec_build_inter_predictors_sby(cm, xd, mi_row, mi_col, ctx, bsize);
+  if (num_planes > 1)
+    dec_build_inter_predictors_sbuv(cm, xd, mi_row, mi_col, ctx, bsize);
+}
 
-  int reader_corrupted_flag = aom_reader_has_error(r);
-  aom_merge_corrupted_flag(&xd->corrupted, reader_corrupted_flag);
+static INLINE void dec_build_prediction_by_above_pred(
+    MACROBLOCKD *xd, int rel_mi_col, uint8_t above_mi_width,
+    MB_MODE_INFO *above_mbmi, void *fun_ctxt, const int num_planes) {
+  struct build_prediction_ctxt *ctxt = (struct build_prediction_ctxt *)fun_ctxt;
+  const int above_mi_col = ctxt->mi_col + rel_mi_col;
+  int mi_x, mi_y;
+  MB_MODE_INFO backup_mbmi = *above_mbmi;
+
+  av1_setup_build_prediction_by_above_pred(xd, rel_mi_col, above_mi_width,
+                                           &backup_mbmi, ctxt, num_planes);
+  mi_x = above_mi_col << MI_SIZE_LOG2;
+  mi_y = ctxt->mi_row << MI_SIZE_LOG2;
+
+  const BLOCK_SIZE bsize = xd->mi[0]->sb_type;
+
+  for (int j = 0; j < num_planes; ++j) {
+    const struct macroblockd_plane *pd = &xd->plane[j];
+    int bw = (above_mi_width * MI_SIZE) >> pd->subsampling_x;
+    int bh = clamp(block_size_high[bsize] >> (pd->subsampling_y + 1), 4,
+                   block_size_high[BLOCK_64X64] >> (pd->subsampling_y + 1));
+
+    if (av1_skip_u4x4_pred_in_obmc(bsize, pd, 0)) continue;
+    dec_build_inter_predictors(ctxt->cm, xd, j, &backup_mbmi, 1, bw, bh, mi_x,
+                               mi_y);
+  }
 }
 
-#if CONFIG_NCOBMC_ADAPT_WEIGHT
-static void set_mode_info_offsets(AV1_COMMON *const cm, MACROBLOCKD *const xd,
-                                  int mi_row, int mi_col) {
-  const int offset = mi_row * cm->mi_stride + mi_col;
-  xd->mi = cm->mi_grid_visible + offset;
-  xd->mi[0] = &cm->mi[offset];
+static void dec_build_prediction_by_above_preds(
+    const AV1_COMMON *cm, MACROBLOCKD *xd, int mi_row, int mi_col,
+    uint8_t *tmp_buf[MAX_MB_PLANE], int tmp_width[MAX_MB_PLANE],
+    int tmp_height[MAX_MB_PLANE], int tmp_stride[MAX_MB_PLANE]) {
+  if (!xd->up_available) return;
+
+  // Adjust mb_to_bottom_edge to have the correct value for the OBMC
+  // prediction block. This is half the height of the original block,
+  // except for 128-wide blocks, where we only use a height of 32.
+  int this_height = xd->n8_h * MI_SIZE;
+  int pred_height = AOMMIN(this_height / 2, 32);
+  xd->mb_to_bottom_edge += (this_height - pred_height) * 8;
+
+  struct build_prediction_ctxt ctxt = { cm,         mi_row,
+                                        mi_col,     tmp_buf,
+                                        tmp_width,  tmp_height,
+                                        tmp_stride, xd->mb_to_right_edge };
+  BLOCK_SIZE bsize = xd->mi[0]->sb_type;
+  foreach_overlappable_nb_above(cm, xd, mi_col,
+                                max_neighbor_obmc[mi_size_wide_log2[bsize]],
+                                dec_build_prediction_by_above_pred, &ctxt);
+
+  xd->mb_to_left_edge = -((mi_col * MI_SIZE) * 8);
+  xd->mb_to_right_edge = ctxt.mb_to_far_edge;
+  xd->mb_to_bottom_edge -= (this_height - pred_height) * 8;
 }
 
-static void get_ncobmc_recon(AV1_COMMON *const cm, MACROBLOCKD *xd, int mi_row,
-                             int mi_col, int bsize, int mode) {
-  uint8_t *pred_buf[4][MAX_MB_PLANE];
-  int pred_stride[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
-  // target block in pxl
-  int pxl_row = mi_row << MI_SIZE_LOG2;
-  int pxl_col = mi_col << MI_SIZE_LOG2;
+static INLINE void dec_build_prediction_by_left_pred(
+    MACROBLOCKD *xd, int rel_mi_row, uint8_t left_mi_height,
+    MB_MODE_INFO *left_mbmi, void *fun_ctxt, const int num_planes) {
+  struct build_prediction_ctxt *ctxt = (struct build_prediction_ctxt *)fun_ctxt;
+  const int left_mi_row = ctxt->mi_row + rel_mi_row;
+  int mi_x, mi_y;
+  MB_MODE_INFO backup_mbmi = *left_mbmi;
+
+  av1_setup_build_prediction_by_left_pred(xd, rel_mi_row, left_mi_height,
+                                          &backup_mbmi, ctxt, num_planes);
+  mi_x = ctxt->mi_col << MI_SIZE_LOG2;
+  mi_y = left_mi_row << MI_SIZE_LOG2;
+  const BLOCK_SIZE bsize = xd->mi[0]->sb_type;
+
+  for (int j = 0; j < num_planes; ++j) {
+    const struct macroblockd_plane *pd = &xd->plane[j];
+    int bw = clamp(block_size_wide[bsize] >> (pd->subsampling_x + 1), 4,
+                   block_size_wide[BLOCK_64X64] >> (pd->subsampling_x + 1));
+    int bh = (left_mi_height << MI_SIZE_LOG2) >> pd->subsampling_y;
+
+    if (av1_skip_u4x4_pred_in_obmc(bsize, pd, 1)) continue;
+    dec_build_inter_predictors(ctxt->cm, xd, j, &backup_mbmi, 1, bw, bh, mi_x,
+                               mi_y);
+  }
+}
+
+static void dec_build_prediction_by_left_preds(
+    const AV1_COMMON *cm, MACROBLOCKD *xd, int mi_row, int mi_col,
+    uint8_t *tmp_buf[MAX_MB_PLANE], int tmp_width[MAX_MB_PLANE],
+    int tmp_height[MAX_MB_PLANE], int tmp_stride[MAX_MB_PLANE]) {
+  if (!xd->left_available) return;
+
+  // Adjust mb_to_right_edge to have the correct value for the OBMC
+  // prediction block. This is half the width of the original block,
+  // except for 128-wide blocks, where we only use a width of 32.
+  int this_width = xd->n8_w * MI_SIZE;
+  int pred_width = AOMMIN(this_width / 2, 32);
+  xd->mb_to_right_edge += (this_width - pred_width) * 8;
+
+  struct build_prediction_ctxt ctxt = { cm,         mi_row,
+                                        mi_col,     tmp_buf,
+                                        tmp_width,  tmp_height,
+                                        tmp_stride, xd->mb_to_bottom_edge };
+  BLOCK_SIZE bsize = xd->mi[0]->sb_type;
+  foreach_overlappable_nb_left(cm, xd, mi_row,
+                               max_neighbor_obmc[mi_size_high_log2[bsize]],
+                               dec_build_prediction_by_left_pred, &ctxt);
+
+  xd->mb_to_top_edge = -((mi_row * MI_SIZE) * 8);
+  xd->mb_to_right_edge -= (this_width - pred_width) * 8;
+  xd->mb_to_bottom_edge = ctxt.mb_to_far_edge;
+}
+
+static void dec_build_obmc_inter_predictors_sb(const AV1_COMMON *cm,
+                                               MACROBLOCKD *xd, int mi_row,
+                                               int mi_col) {
+  const int num_planes = av1_num_planes(cm);
+  DECLARE_ALIGNED(16, uint8_t, tmp_buf1[2 * MAX_MB_PLANE * MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(16, uint8_t, tmp_buf2[2 * MAX_MB_PLANE * MAX_SB_SQUARE]);
+  uint8_t *dst_buf1[MAX_MB_PLANE], *dst_buf2[MAX_MB_PLANE];
+  int dst_stride1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
+  int dst_stride2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
+  int dst_width1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
+  int dst_width2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
+  int dst_height1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
+  int dst_height2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
 
-  int plane;
-#if CONFIG_HIGHBITDEPTH
   if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
     int len = sizeof(uint16_t);
-    ASSIGN_ALIGNED_PTRS_HBD(pred_buf[0], cm->ncobmcaw_buf[0], MAX_SB_SQUARE,
-                            len);
-    ASSIGN_ALIGNED_PTRS_HBD(pred_buf[1], cm->ncobmcaw_buf[1], MAX_SB_SQUARE,
-                            len);
-    ASSIGN_ALIGNED_PTRS_HBD(pred_buf[2], cm->ncobmcaw_buf[2], MAX_SB_SQUARE,
-                            len);
-    ASSIGN_ALIGNED_PTRS_HBD(pred_buf[3], cm->ncobmcaw_buf[3], MAX_SB_SQUARE,
-                            len);
+    dst_buf1[0] = CONVERT_TO_BYTEPTR(tmp_buf1);
+    dst_buf1[1] = CONVERT_TO_BYTEPTR(tmp_buf1 + MAX_SB_SQUARE * len);
+    dst_buf1[2] = CONVERT_TO_BYTEPTR(tmp_buf1 + MAX_SB_SQUARE * 2 * len);
+    dst_buf2[0] = CONVERT_TO_BYTEPTR(tmp_buf2);
+    dst_buf2[1] = CONVERT_TO_BYTEPTR(tmp_buf2 + MAX_SB_SQUARE * len);
+    dst_buf2[2] = CONVERT_TO_BYTEPTR(tmp_buf2 + MAX_SB_SQUARE * 2 * len);
   } else {
-#endif  // CONFIG_HIGHBITDEPTH
-    ASSIGN_ALIGNED_PTRS(pred_buf[0], cm->ncobmcaw_buf[0], MAX_SB_SQUARE);
-    ASSIGN_ALIGNED_PTRS(pred_buf[1], cm->ncobmcaw_buf[1], MAX_SB_SQUARE);
-    ASSIGN_ALIGNED_PTRS(pred_buf[2], cm->ncobmcaw_buf[2], MAX_SB_SQUARE);
-    ASSIGN_ALIGNED_PTRS(pred_buf[3], cm->ncobmcaw_buf[3], MAX_SB_SQUARE);
-#if CONFIG_HIGHBITDEPTH
-  }
-#endif
-  av1_get_ext_blk_preds(cm, xd, bsize, mi_row, mi_col, pred_buf, pred_stride);
-  av1_get_ori_blk_pred(cm, xd, bsize, mi_row, mi_col, pred_buf[3], pred_stride);
-  for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
-    build_ncobmc_intrpl_pred(cm, xd, plane, pxl_row, pxl_col, bsize, pred_buf,
-                             pred_stride, mode);
+    dst_buf1[0] = tmp_buf1;
+    dst_buf1[1] = tmp_buf1 + MAX_SB_SQUARE;
+    dst_buf1[2] = tmp_buf1 + MAX_SB_SQUARE * 2;
+    dst_buf2[0] = tmp_buf2;
+    dst_buf2[1] = tmp_buf2 + MAX_SB_SQUARE;
+    dst_buf2[2] = tmp_buf2 + MAX_SB_SQUARE * 2;
+  }
+  dec_build_prediction_by_above_preds(cm, xd, mi_row, mi_col, dst_buf1,
+                                      dst_width1, dst_height1, dst_stride1);
+  dec_build_prediction_by_left_preds(cm, xd, mi_row, mi_col, dst_buf2,
+                                     dst_width2, dst_height2, dst_stride2);
+  av1_setup_dst_planes(xd->plane, xd->mi[0]->sb_type, get_frame_new_buffer(cm),
+                       mi_row, mi_col, 0, num_planes);
+  av1_build_obmc_inter_prediction(cm, xd, mi_row, mi_col, dst_buf1, dst_stride1,
+                                  dst_buf2, dst_stride2);
+}
+
+static void cfl_store_inter_block(AV1_COMMON *const cm, MACROBLOCKD *const xd) {
+  MB_MODE_INFO *mbmi = xd->mi[0];
+  if (store_cfl_required(cm, xd)) {
+    cfl_store_block(xd, mbmi->sb_type, mbmi->tx_size);
   }
 }
 
-static void av1_get_ncobmc_recon(AV1_COMMON *const cm, MACROBLOCKD *const xd,
-                                 int bsize, const int mi_row, const int mi_col,
-                                 const NCOBMC_MODE modes) {
-  const int mi_width = mi_size_wide[bsize];
-  const int mi_height = mi_size_high[bsize];
+static void predict_inter_block(AV1_COMMON *const cm, MACROBLOCKD *const xd,
+                                int mi_row, int mi_col, BLOCK_SIZE bsize) {
+  MB_MODE_INFO *mbmi = xd->mi[0];
+  const int num_planes = av1_num_planes(cm);
+  for (int ref = 0; ref < 1 + has_second_ref(mbmi); ++ref) {
+    const MV_REFERENCE_FRAME frame = mbmi->ref_frame[ref];
+    if (frame < LAST_FRAME) {
+      assert(is_intrabc_block(mbmi));
+      assert(frame == INTRA_FRAME);
+      assert(ref == 0);
+    } else {
+      RefBuffer *ref_buf = &cm->frame_refs[frame - LAST_FRAME];
 
-  assert(bsize >= BLOCK_8X8);
+      xd->block_refs[ref] = ref_buf;
+      av1_setup_pre_planes(xd, ref, ref_buf->buf, mi_row, mi_col, &ref_buf->sf,
+                           num_planes);
+    }
+  }
 
-  reset_xd_boundary(xd, mi_row, mi_height, mi_col, mi_width, cm->mi_rows,
-                    cm->mi_cols);
-  get_ncobmc_recon(cm, xd, mi_row, mi_col, bsize, modes);
+  dec_build_inter_predictors_sb(cm, xd, mi_row, mi_col, NULL, bsize);
+  if (mbmi->motion_mode == OBMC_CAUSAL)
+    dec_build_obmc_inter_predictors_sb(cm, xd, mi_row, mi_col);
 }
 
-static void recon_ncobmc_intrpl_pred(AV1_COMMON *const cm,
-                                     MACROBLOCKD *const xd, int mi_row,
-                                     int mi_col, BLOCK_SIZE bsize) {
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
-  const int mi_width = mi_size_wide[bsize];
-  const int mi_height = mi_size_high[bsize];
-  const int hbs = AOMMAX(mi_size_wide[bsize] / 2, mi_size_high[bsize] / 2);
-  const BLOCK_SIZE sqr_blk = bsize_2_sqr_bsize[bsize];
-  if (mi_width > mi_height) {
-    // horizontal partition
-    av1_get_ncobmc_recon(cm, xd, sqr_blk, mi_row, mi_col, mbmi->ncobmc_mode[0]);
-    xd->mi += hbs;
-    av1_get_ncobmc_recon(cm, xd, sqr_blk, mi_row, mi_col + hbs,
-                         mbmi->ncobmc_mode[1]);
-  } else if (mi_height > mi_width) {
-    // vertical partition
-    av1_get_ncobmc_recon(cm, xd, sqr_blk, mi_row, mi_col, mbmi->ncobmc_mode[0]);
-    xd->mi += hbs * xd->mi_stride;
-    av1_get_ncobmc_recon(cm, xd, sqr_blk, mi_row + hbs, mi_col,
-                         mbmi->ncobmc_mode[1]);
-  } else {
-    av1_get_ncobmc_recon(cm, xd, sqr_blk, mi_row, mi_col, mbmi->ncobmc_mode[0]);
-  }
-  set_mode_info_offsets(cm, xd, mi_row, mi_col);
-  // restore dst buffer and mode info
-  av1_setup_dst_planes(xd->plane, bsize, get_frame_new_buffer(cm), mi_row,
-                       mi_col);
+static void set_color_index_map_offset(MACROBLOCKD *const xd, int plane,
+                                       aom_reader *r) {
+  (void)r;
+  Av1ColorMapParam params;
+  const MB_MODE_INFO *const mbmi = xd->mi[0];
+  av1_get_block_dimensions(mbmi->sb_type, plane, xd, &params.plane_width,
+                           &params.plane_height, NULL, NULL);
+  xd->color_index_map_offset[plane] += params.plane_width * params.plane_height;
 }
-#endif  // CONFIG_NCOBMC_ADAPT_WEIGHT
 
 static void decode_token_and_recon_block(AV1Decoder *const pbi,
                                          MACROBLOCKD *const xd, int mi_row,
                                          int mi_col, aom_reader *r,
                                          BLOCK_SIZE bsize) {
   AV1_COMMON *const cm = &pbi->common;
+  const int num_planes = av1_num_planes(cm);
   const int bw = mi_size_wide[bsize];
   const int bh = mi_size_high[bsize];
   const int x_mis = AOMMIN(bw, cm->mi_cols - mi_col);
   const int y_mis = AOMMIN(bh, cm->mi_rows - mi_row);
 
   set_offsets(cm, xd, bsize, mi_row, mi_col, bw, bh, x_mis, y_mis);
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
-#if CONFIG_CFL && CONFIG_CHROMA_SUB8X8
-  CFL_CTX *const cfl = xd->cfl;
+  MB_MODE_INFO *mbmi = xd->mi[0];
+  CFL_CTX *const cfl = &xd->cfl;
   cfl->is_chroma_reference = is_chroma_reference(
       mi_row, mi_col, bsize, cfl->subsampling_x, cfl->subsampling_y);
-#endif  // CONFIG_CFL && CONFIG_CHROMA_SUB8X8
 
   if (cm->delta_q_present_flag) {
-    int i;
-    for (i = 0; i < MAX_SEGMENTS; i++) {
-#if CONFIG_EXT_DELTA_Q
+    for (int i = 0; i < MAX_SEGMENTS; i++) {
       const int current_qindex =
           av1_get_qindex(&cm->seg, i, xd->current_qindex);
-#else
-      const int current_qindex = xd->current_qindex;
-#endif  // CONFIG_EXT_DELTA_Q
-      int j;
-      for (j = 0; j < MAX_MB_PLANE; ++j) {
-        const int dc_delta_q = j == 0 ? cm->y_dc_delta_q : cm->uv_dc_delta_q;
-        const int ac_delta_q = j == 0 ? 0 : cm->uv_ac_delta_q;
-
-        xd->plane[j].seg_dequant[i][0] =
-            av1_dc_quant(current_qindex, dc_delta_q, cm->bit_depth);
-        xd->plane[j].seg_dequant[i][1] =
-            av1_ac_quant(current_qindex, ac_delta_q, cm->bit_depth);
+      for (int j = 0; j < num_planes; ++j) {
+        const int dc_delta_q =
+            j == 0 ? cm->y_dc_delta_q
+                   : (j == 1 ? cm->u_dc_delta_q : cm->v_dc_delta_q);
+        const int ac_delta_q =
+            j == 0 ? 0 : (j == 1 ? cm->u_ac_delta_q : cm->v_ac_delta_q);
+        xd->plane[j].seg_dequant_QTX[i][0] =
+            av1_dc_quant_QTX(current_qindex, dc_delta_q, cm->bit_depth);
+        xd->plane[j].seg_dequant_QTX[i][1] =
+            av1_ac_quant_QTX(current_qindex, ac_delta_q, cm->bit_depth);
       }
     }
   }
+  if (mbmi->skip) av1_reset_skip_context(xd, mi_row, mi_col, bsize, num_planes);
 
-#if CONFIG_CB4X4
-  if (mbmi->skip) av1_reset_skip_context(xd, mi_row, mi_col, bsize);
-#else
-  if (mbmi->skip) {
-    av1_reset_skip_context(xd, mi_row, mi_col, AOMMAX(BLOCK_8X8, bsize));
-  }
-#endif
-
-#if CONFIG_COEF_INTERLEAVE
-  {
-    const struct macroblockd_plane *const pd_y = &xd->plane[0];
-    const struct macroblockd_plane *const pd_c = &xd->plane[1];
-    const TX_SIZE tx_log2_y = mbmi->tx_size;
-    const TX_SIZE tx_log2_c = av1_get_uv_tx_size(mbmi, pd_c);
-    const int tx_sz_y = (1 << tx_log2_y);
-    const int tx_sz_c = (1 << tx_log2_c);
-    const int num_4x4_w_y = pd_y->n4_w;
-    const int num_4x4_h_y = pd_y->n4_h;
-    const int num_4x4_w_c = pd_c->n4_w;
-    const int num_4x4_h_c = pd_c->n4_h;
-    const int max_4x4_w_y = get_max_4x4_size(num_4x4_w_y, xd->mb_to_right_edge,
-                                             pd_y->subsampling_x);
-    const int max_4x4_h_y = get_max_4x4_size(num_4x4_h_y, xd->mb_to_bottom_edge,
-                                             pd_y->subsampling_y);
-    const int max_4x4_w_c = get_max_4x4_size(num_4x4_w_c, xd->mb_to_right_edge,
-                                             pd_c->subsampling_x);
-    const int max_4x4_h_c = get_max_4x4_size(num_4x4_h_c, xd->mb_to_bottom_edge,
-                                             pd_c->subsampling_y);
-
-    // The max_4x4_w/h may be smaller than tx_sz under some corner cases,
-    // i.e. when the SB is splitted by tile boundaries.
-    const int tu_num_w_y = (max_4x4_w_y + tx_sz_y - 1) / tx_sz_y;
-    const int tu_num_h_y = (max_4x4_h_y + tx_sz_y - 1) / tx_sz_y;
-    const int tu_num_w_c = (max_4x4_w_c + tx_sz_c - 1) / tx_sz_c;
-    const int tu_num_h_c = (max_4x4_h_c + tx_sz_c - 1) / tx_sz_c;
-    const int tu_num_c = tu_num_w_c * tu_num_h_c;
-
-    if (!is_inter_block(mbmi)) {
-      int tu_idx_c = 0;
-      int row_y, col_y, row_c, col_c;
-      int plane;
-
-// TODO(anybody) : remove this flag when PVQ supports pallete coding tool
-#if !CONFIG_PVQ
-      for (plane = 0; plane <= 1; ++plane) {
-        if (mbmi->palette_mode_info.palette_size[plane])
-          av1_decode_palette_tokens(xd, plane, r);
-      }
-#endif  // !CONFIG_PVQ
-
-      for (row_y = 0; row_y < tu_num_h_y; row_y++) {
-        for (col_y = 0; col_y < tu_num_w_y; col_y++) {
-          // luma
-          predict_and_reconstruct_intra_block(
-              cm, xd, r, mbmi, 0, row_y * tx_sz_y, col_y * tx_sz_y, tx_log2_y);
-          // chroma
-          if (tu_idx_c < tu_num_c) {
-            row_c = (tu_idx_c / tu_num_w_c) * tx_sz_c;
-            col_c = (tu_idx_c % tu_num_w_c) * tx_sz_c;
-            predict_and_reconstruct_intra_block(cm, xd, r, mbmi, 1, row_c,
-                                                col_c, tx_log2_c);
-            predict_and_reconstruct_intra_block(cm, xd, r, mbmi, 2, row_c,
-                                                col_c, tx_log2_c);
-            tu_idx_c++;
-          }
-        }
-      }
-
-      // In 422 case, it's possilbe that Chroma has more TUs than Luma
-      while (tu_idx_c < tu_num_c) {
-        row_c = (tu_idx_c / tu_num_w_c) * tx_sz_c;
-        col_c = (tu_idx_c % tu_num_w_c) * tx_sz_c;
-        predict_and_reconstruct_intra_block(cm, xd, r, mbmi, 1, row_c, col_c,
-                                            tx_log2_c);
-        predict_and_reconstruct_intra_block(cm, xd, r, mbmi, 2, row_c, col_c,
-                                            tx_log2_c);
-        tu_idx_c++;
-      }
-    } else {
-      // Prediction
-      av1_build_inter_predictors_sb(cm, xd, mi_row, mi_col, NULL,
-                                    AOMMAX(bsize, BLOCK_8X8));
-
-      // Reconstruction
-      if (!mbmi->skip) {
-        int eobtotal = 0;
-        int tu_idx_c = 0;
-        int row_y, col_y, row_c, col_c;
-
-        for (row_y = 0; row_y < tu_num_h_y; row_y++) {
-          for (col_y = 0; col_y < tu_num_w_y; col_y++) {
-            // luma
-            eobtotal += reconstruct_inter_block(cm, xd, r, mbmi->segment_id, 0,
-                                                row_y * tx_sz_y,
-                                                col_y * tx_sz_y, tx_log2_y);
-            // chroma
-            if (tu_idx_c < tu_num_c) {
-              row_c = (tu_idx_c / tu_num_w_c) * tx_sz_c;
-              col_c = (tu_idx_c % tu_num_w_c) * tx_sz_c;
-              eobtotal += reconstruct_inter_block(cm, xd, r, mbmi->segment_id,
-                                                  1, row_c, col_c, tx_log2_c);
-              eobtotal += reconstruct_inter_block(cm, xd, r, mbmi->segment_id,
-                                                  2, row_c, col_c, tx_log2_c);
-              tu_idx_c++;
+  if (!is_inter_block(mbmi)) {
+    int row, col;
+    assert(bsize == get_plane_block_size(bsize, xd->plane[0].subsampling_x,
+                                         xd->plane[0].subsampling_y));
+    const int max_blocks_wide = max_block_wide(xd, bsize, 0);
+    const int max_blocks_high = max_block_high(xd, bsize, 0);
+    const BLOCK_SIZE max_unit_bsize = BLOCK_64X64;
+    int mu_blocks_wide =
+        block_size_wide[max_unit_bsize] >> tx_size_wide_log2[0];
+    int mu_blocks_high =
+        block_size_high[max_unit_bsize] >> tx_size_high_log2[0];
+    mu_blocks_wide = AOMMIN(max_blocks_wide, mu_blocks_wide);
+    mu_blocks_high = AOMMIN(max_blocks_high, mu_blocks_high);
+
+    for (row = 0; row < max_blocks_high; row += mu_blocks_high) {
+      for (col = 0; col < max_blocks_wide; col += mu_blocks_wide) {
+        for (int plane = 0; plane < num_planes; ++plane) {
+          const struct macroblockd_plane *const pd = &xd->plane[plane];
+          if (!is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
+                                   pd->subsampling_y))
+            continue;
+
+          const TX_SIZE tx_size = av1_get_tx_size(plane, xd);
+          const int stepr = tx_size_high_unit[tx_size];
+          const int stepc = tx_size_wide_unit[tx_size];
+
+          const int unit_height = ROUND_POWER_OF_TWO(
+              AOMMIN(mu_blocks_high + row, max_blocks_high), pd->subsampling_y);
+          const int unit_width = ROUND_POWER_OF_TWO(
+              AOMMIN(mu_blocks_wide + col, max_blocks_wide), pd->subsampling_x);
+
+          for (int blk_row = row >> pd->subsampling_y; blk_row < unit_height;
+               blk_row += stepr) {
+            for (int blk_col = col >> pd->subsampling_x; blk_col < unit_width;
+                 blk_col += stepc) {
+              read_coeffs_tx_intra_block(cm, xd, r, plane, blk_row, blk_col,
+                                         tx_size);
+              predict_and_reconstruct_intra_block(cm, xd, r, plane, blk_row,
+                                                  blk_col, tx_size);
+              set_cb_buffer_offsets(xd, tx_size, plane);
             }
           }
         }
-
-        // In 422 case, it's possilbe that Chroma has more TUs than Luma
-        while (tu_idx_c < tu_num_c) {
-          row_c = (tu_idx_c / tu_num_w_c) * tx_sz_c;
-          col_c = (tu_idx_c % tu_num_w_c) * tx_sz_c;
-          eobtotal += reconstruct_inter_block(cm, xd, r, mbmi->segment_id, 1,
-                                              row_c, col_c, tx_log2_c);
-          eobtotal += reconstruct_inter_block(cm, xd, r, mbmi->segment_id, 2,
-                                              row_c, col_c, tx_log2_c);
-          tu_idx_c++;
-        }
-
-        // TODO(CONFIG_COEF_INTERLEAVE owners): bring eob == 0 corner case
-        // into line with the defaut configuration
-        if (bsize >= BLOCK_8X8 && eobtotal == 0) mbmi->skip = 1;
       }
     }
-  }
-#else  // CONFIG_COEF_INTERLEAVE
-  if (!is_inter_block(mbmi)) {
-    int plane;
-
-// TODO(anybody) : remove this flag when PVQ supports pallete coding tool
-#if !CONFIG_PVQ
-    for (plane = 0; plane <= 1; ++plane) {
-      if (mbmi->palette_mode_info.palette_size[plane])
-        av1_decode_palette_tokens(xd, plane, r);
-    }
-#endif  // #if !CONFIG_PVQ
-
-    for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
-      const struct macroblockd_plane *const pd = &xd->plane[plane];
-      const TX_SIZE tx_size = av1_get_tx_size(plane, xd);
-      const int stepr = tx_size_high_unit[tx_size];
-      const int stepc = tx_size_wide_unit[tx_size];
-#if CONFIG_CHROMA_SUB8X8
-      const BLOCK_SIZE plane_bsize =
-          AOMMAX(BLOCK_4X4, get_plane_block_size(bsize, pd));
-#elif CONFIG_CB4X4
-      const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
-#else
-      const BLOCK_SIZE plane_bsize =
-          get_plane_block_size(AOMMAX(BLOCK_8X8, bsize), pd);
-#endif
-      int row, col;
-      const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
-      const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
-#if CONFIG_CB4X4
+  } else {
+    predict_inter_block(cm, xd, mi_row, mi_col, bsize);
+#if CONFIG_MISMATCH_DEBUG
+    for (int plane = 0; plane < num_planes; ++plane) {
+      const struct macroblockd_plane *pd = &xd->plane[plane];
+      int pixel_c, pixel_r;
+      mi_to_pixel_loc(&pixel_c, &pixel_r, mi_col, mi_row, 0, 0,
+                      pd->subsampling_x, pd->subsampling_y);
       if (!is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
                                pd->subsampling_y))
         continue;
+      mismatch_check_block_pre(pd->dst.buf, pd->dst.stride, cm->frame_offset,
+                               plane, pixel_c, pixel_r, pd->width, pd->height,
+                               xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH);
+    }
 #endif
-      int blk_row, blk_col;
-      const BLOCK_SIZE max_unit_bsize = get_plane_block_size(BLOCK_64X64, pd);
+
+    // Reconstruction
+    if (!mbmi->skip) {
+      int eobtotal = 0;
+
+      const int max_blocks_wide = max_block_wide(xd, bsize, 0);
+      const int max_blocks_high = max_block_high(xd, bsize, 0);
+      int row, col;
+
+      const BLOCK_SIZE max_unit_bsize = BLOCK_64X64;
+      assert(max_unit_bsize ==
+             get_plane_block_size(BLOCK_64X64, xd->plane[0].subsampling_x,
+                                  xd->plane[0].subsampling_y));
       int mu_blocks_wide =
           block_size_wide[max_unit_bsize] >> tx_size_wide_log2[0];
       int mu_blocks_high =
           block_size_high[max_unit_bsize] >> tx_size_high_log2[0];
+
       mu_blocks_wide = AOMMIN(max_blocks_wide, mu_blocks_wide);
       mu_blocks_high = AOMMIN(max_blocks_high, mu_blocks_high);
 
       for (row = 0; row < max_blocks_high; row += mu_blocks_high) {
-        const int unit_height = AOMMIN(mu_blocks_high + row, max_blocks_high);
         for (col = 0; col < max_blocks_wide; col += mu_blocks_wide) {
-          const int unit_width = AOMMIN(mu_blocks_wide + col, max_blocks_wide);
-
-          for (blk_row = row; blk_row < unit_height; blk_row += stepr)
-            for (blk_col = col; blk_col < unit_width; blk_col += stepc)
-              predict_and_reconstruct_intra_block(cm, xd, r, mbmi, plane,
-                                                  blk_row, blk_col, tx_size);
-        }
-      }
-    }
-  } else {
-    int ref;
-
-#if CONFIG_COMPOUND_SINGLEREF
-    for (ref = 0; ref < 1 + is_inter_anyref_comp_mode(mbmi->mode); ++ref)
-#else
-    for (ref = 0; ref < 1 + has_second_ref(mbmi); ++ref)
-#endif  // CONFIG_COMPOUND_SINGLEREF
-    {
-      const MV_REFERENCE_FRAME frame =
-#if CONFIG_COMPOUND_SINGLEREF
-          has_second_ref(mbmi) ? mbmi->ref_frame[ref] : mbmi->ref_frame[0];
-#else
-          mbmi->ref_frame[ref];
-#endif  // CONFIG_COMPOUND_SINGLEREF
-      if (frame < LAST_FRAME) {
-#if CONFIG_INTRABC
-        assert(is_intrabc_block(mbmi));
-        assert(frame == INTRA_FRAME);
-        assert(ref == 0);
-#else
-        assert(0);
-#endif  // CONFIG_INTRABC
-      } else {
-        RefBuffer *ref_buf = &cm->frame_refs[frame - LAST_FRAME];
-
-        xd->block_refs[ref] = ref_buf;
-        if ((!av1_is_valid_scale(&ref_buf->sf)))
-          aom_internal_error(xd->error_info, AOM_CODEC_UNSUP_BITSTREAM,
-                             "Reference frame has invalid dimensions");
-        av1_setup_pre_planes(xd, ref, ref_buf->buf, mi_row, mi_col,
-                             &ref_buf->sf);
-      }
-    }
-
-#if CONFIG_CB4X4
-    av1_build_inter_predictors_sb(cm, xd, mi_row, mi_col, NULL, bsize);
-#else
-    av1_build_inter_predictors_sb(cm, xd, mi_row, mi_col, NULL,
-                                  AOMMAX(bsize, BLOCK_8X8));
-#endif
-
-#if CONFIG_MOTION_VAR
-    if (mbmi->motion_mode == OBMC_CAUSAL) {
-#if CONFIG_NCOBMC
-      av1_build_ncobmc_inter_predictors_sb(cm, xd, mi_row, mi_col);
-#else
-      av1_build_obmc_inter_predictors_sb(cm, xd, mi_row, mi_col);
-#endif
-    }
-#endif  // CONFIG_MOTION_VAR
-#if CONFIG_NCOBMC_ADAPT_WEIGHT
-    if (mbmi->motion_mode == NCOBMC_ADAPT_WEIGHT) {
-      int plane;
-      recon_ncobmc_intrpl_pred(cm, xd, mi_row, mi_col, bsize);
-      for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
-        get_pred_from_intrpl_buf(xd, mi_row, mi_col, bsize, plane);
-      }
-    }
-#endif
-    // Reconstruction
-    if (!mbmi->skip) {
-      int eobtotal = 0;
-      int plane;
-
-      for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
-        const struct macroblockd_plane *const pd = &xd->plane[plane];
-#if CONFIG_CHROMA_SUB8X8
-        const BLOCK_SIZE plane_bsize =
-            AOMMAX(BLOCK_4X4, get_plane_block_size(bsize, pd));
-#elif CONFIG_CB4X4
-        const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
-#else
-        const BLOCK_SIZE plane_bsize =
-            get_plane_block_size(AOMMAX(BLOCK_8X8, bsize), pd);
-#endif
-        const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
-        const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
-        int row, col;
-
-#if CONFIG_CB4X4
-        if (!is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
-                                 pd->subsampling_y))
-          continue;
-#endif
-
-#if CONFIG_VAR_TX
-        const BLOCK_SIZE max_unit_bsize = get_plane_block_size(BLOCK_64X64, pd);
-        int mu_blocks_wide =
-            block_size_wide[max_unit_bsize] >> tx_size_wide_log2[0];
-        int mu_blocks_high =
-            block_size_high[max_unit_bsize] >> tx_size_high_log2[0];
-
-        mu_blocks_wide = AOMMIN(max_blocks_wide, mu_blocks_wide);
-        mu_blocks_high = AOMMIN(max_blocks_high, mu_blocks_high);
-
-        const TX_SIZE max_tx_size = get_vartx_max_txsize(
-            mbmi, plane_bsize, pd->subsampling_x || pd->subsampling_y);
-        const int bh_var_tx = tx_size_high_unit[max_tx_size];
-        const int bw_var_tx = tx_size_wide_unit[max_tx_size];
-        int block = 0;
-        int step =
-            tx_size_wide_unit[max_tx_size] * tx_size_high_unit[max_tx_size];
-
-        for (row = 0; row < max_blocks_high; row += mu_blocks_high) {
-          for (col = 0; col < max_blocks_wide; col += mu_blocks_wide) {
+          for (int plane = 0; plane < num_planes; ++plane) {
+            const struct macroblockd_plane *const pd = &xd->plane[plane];
+            if (!is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
+                                     pd->subsampling_y))
+              continue;
+            const BLOCK_SIZE bsizec =
+                scale_chroma_bsize(bsize, pd->subsampling_x, pd->subsampling_y);
+            const BLOCK_SIZE plane_bsize = get_plane_block_size(
+                bsizec, pd->subsampling_x, pd->subsampling_y);
+
+            const TX_SIZE max_tx_size =
+                get_vartx_max_txsize(xd, plane_bsize, plane);
+            const int bh_var_tx = tx_size_high_unit[max_tx_size];
+            const int bw_var_tx = tx_size_wide_unit[max_tx_size];
+            int block = 0;
+            int step =
+                tx_size_wide_unit[max_tx_size] * tx_size_high_unit[max_tx_size];
             int blk_row, blk_col;
-            const int unit_height =
-                AOMMIN(mu_blocks_high + row, max_blocks_high);
-            const int unit_width =
-                AOMMIN(mu_blocks_wide + col, max_blocks_wide);
-            for (blk_row = row; blk_row < unit_height; blk_row += bh_var_tx) {
-              for (blk_col = col; blk_col < unit_width; blk_col += bw_var_tx) {
+            const int unit_height = ROUND_POWER_OF_TWO(
+                AOMMIN(mu_blocks_high + row, max_blocks_high),
+                pd->subsampling_y);
+            const int unit_width = ROUND_POWER_OF_TWO(
+                AOMMIN(mu_blocks_wide + col, max_blocks_wide),
+                pd->subsampling_x);
+
+            for (blk_row = row >> pd->subsampling_y; blk_row < unit_height;
+                 blk_row += bh_var_tx) {
+              for (blk_col = col >> pd->subsampling_x; blk_col < unit_width;
+                   blk_col += bw_var_tx) {
                 decode_reconstruct_tx(cm, xd, r, mbmi, plane, plane_bsize,
                                       blk_row, blk_col, block, max_tx_size,
                                       &eobtotal);
@@ -2083,388 +1221,291 @@ static void decode_token_and_recon_block(AV1Decoder *const pbi,
             }
           }
         }
-#else
-        const TX_SIZE tx_size = av1_get_tx_size(plane, xd);
-        const int stepr = tx_size_high_unit[tx_size];
-        const int stepc = tx_size_wide_unit[tx_size];
-        for (row = 0; row < max_blocks_high; row += stepr)
-          for (col = 0; col < max_blocks_wide; col += stepc)
-            eobtotal += reconstruct_inter_block(cm, xd, r, mbmi->segment_id,
-                                                plane, row, col, tx_size);
-#endif
       }
     }
+    cfl_store_inter_block(cm, xd);
   }
-#if CONFIG_CFL && CONFIG_CHROMA_SUB8X8
-  if (mbmi->uv_mode != UV_CFL_PRED) {
-#if CONFIG_DEBUG
-    if (cfl->is_chroma_reference) {
-      cfl_clear_sub8x8_val(cfl);
-    }
-#endif
-    if (!cfl->is_chroma_reference && is_inter_block(mbmi)) {
-      cfl_store_block(xd, mbmi->sb_type, mbmi->tx_size);
-    }
-  }
-#endif  // CONFIG_CFL && CONFIG_CHROMA_SUB8X8
-#endif  // CONFIG_COEF_INTERLEAVE
+
+  av1_visit_palette(pbi, xd, mi_row, mi_col, r, bsize,
+                    set_color_index_map_offset);
 
   int reader_corrupted_flag = aom_reader_has_error(r);
   aom_merge_corrupted_flag(&xd->corrupted, reader_corrupted_flag);
 }
 
-#if NC_MODE_INFO && CONFIG_MOTION_VAR
-static void detoken_and_recon_sb(AV1Decoder *const pbi, MACROBLOCKD *const xd,
-                                 int mi_row, int mi_col, aom_reader *r,
-                                 BLOCK_SIZE bsize) {
-  AV1_COMMON *const cm = &pbi->common;
-  const int hbs = mi_size_wide[bsize] >> 1;
-#if CONFIG_CB4X4
-  const int unify_bsize = 1;
-#else
-  const int unify_bsize = 0;
-#endif
-#if CONFIG_EXT_PARTITION_TYPES
-  BLOCK_SIZE bsize2 = get_subsize(bsize, PARTITION_SPLIT);
-#endif
-  PARTITION_TYPE partition;
-  BLOCK_SIZE subsize;
-  const int has_rows = (mi_row + hbs) < cm->mi_rows;
-  const int has_cols = (mi_col + hbs) < cm->mi_cols;
+static void read_tx_size_vartx(MACROBLOCKD *xd, MB_MODE_INFO *mbmi,
+                               TX_SIZE tx_size, int depth, int blk_row,
+                               int blk_col, aom_reader *r) {
+  FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
+  int is_split = 0;
+  const BLOCK_SIZE bsize = mbmi->sb_type;
+  const int max_blocks_high = max_block_high(xd, bsize, 0);
+  const int max_blocks_wide = max_block_wide(xd, bsize, 0);
+  if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
+  assert(tx_size > TX_4X4);
+
+  if (depth == MAX_VARTX_DEPTH) {
+    for (int idy = 0; idy < tx_size_high_unit[tx_size]; ++idy) {
+      for (int idx = 0; idx < tx_size_wide_unit[tx_size]; ++idx) {
+        const int index =
+            av1_get_txb_size_index(bsize, blk_row + idy, blk_col + idx);
+        mbmi->inter_tx_size[index] = tx_size;
+      }
+    }
+    mbmi->tx_size = tx_size;
+    txfm_partition_update(xd->above_txfm_context + blk_col,
+                          xd->left_txfm_context + blk_row, tx_size, tx_size);
+    return;
+  }
 
-  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return;
+  const int ctx = txfm_partition_context(xd->above_txfm_context + blk_col,
+                                         xd->left_txfm_context + blk_row,
+                                         mbmi->sb_type, tx_size);
+  is_split = aom_read_symbol(r, ec_ctx->txfm_partition_cdf[ctx], 2, ACCT_STR);
 
-  partition = get_partition(cm, mi_row, mi_col, bsize);
-  subsize = subsize_lookup[partition][bsize];
+  if (is_split) {
+    const TX_SIZE sub_txs = sub_tx_size_map[tx_size];
+    const int bsw = tx_size_wide_unit[sub_txs];
+    const int bsh = tx_size_high_unit[sub_txs];
+
+    if (sub_txs == TX_4X4) {
+      for (int idy = 0; idy < tx_size_high_unit[tx_size]; ++idy) {
+        for (int idx = 0; idx < tx_size_wide_unit[tx_size]; ++idx) {
+          const int index =
+              av1_get_txb_size_index(bsize, blk_row + idy, blk_col + idx);
+          mbmi->inter_tx_size[index] = sub_txs;
+        }
+      }
+      mbmi->tx_size = sub_txs;
+      txfm_partition_update(xd->above_txfm_context + blk_col,
+                            xd->left_txfm_context + blk_row, sub_txs, tx_size);
+      return;
+    }
 
-  if (!hbs && !unify_bsize) {
-    xd->bmode_blocks_wl = 1 >> !!(partition & PARTITION_VERT);
-    xd->bmode_blocks_hl = 1 >> !!(partition & PARTITION_HORZ);
-    decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, subsize);
+    assert(bsw > 0 && bsh > 0);
+    for (int row = 0; row < tx_size_high_unit[tx_size]; row += bsh) {
+      for (int col = 0; col < tx_size_wide_unit[tx_size]; col += bsw) {
+        int offsetr = blk_row + row;
+        int offsetc = blk_col + col;
+        read_tx_size_vartx(xd, mbmi, sub_txs, depth + 1, offsetr, offsetc, r);
+      }
+    }
   } else {
-    switch (partition) {
-      case PARTITION_NONE:
-        decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, bsize);
-        break;
-      case PARTITION_HORZ:
-        decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, subsize);
-        if (has_rows)
-          decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col, r,
-                                       subsize);
-        break;
-      case PARTITION_VERT:
-        decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, subsize);
-        if (has_cols)
-          decode_token_and_recon_block(pbi, xd, mi_row, mi_col + hbs, r,
-                                       subsize);
-        break;
-      case PARTITION_SPLIT:
-        detoken_and_recon_sb(pbi, xd, mi_row, mi_col, r, subsize);
-        detoken_and_recon_sb(pbi, xd, mi_row, mi_col + hbs, r, subsize);
-        detoken_and_recon_sb(pbi, xd, mi_row + hbs, mi_col, r, subsize);
-        detoken_and_recon_sb(pbi, xd, mi_row + hbs, mi_col + hbs, r, subsize);
-        break;
-#if CONFIG_EXT_PARTITION_TYPES
-#if CONFIG_EXT_PARTITION_TYPES_AB
-#error NC_MODE_INFO+MOTION_VAR not yet supported for new HORZ/VERT_AB partitions
-#endif
-      case PARTITION_HORZ_A:
-        decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, bsize2);
-        decode_token_and_recon_block(pbi, xd, mi_row, mi_col + hbs, r, bsize2);
-        decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col, r, subsize);
-        break;
-      case PARTITION_HORZ_B:
-        decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, subsize);
-        decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col, r, bsize2);
-        decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col + hbs, r,
-                                     bsize2);
-        break;
-      case PARTITION_VERT_A:
-        decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, bsize2);
-        decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col, r, bsize2);
-        decode_token_and_recon_block(pbi, xd, mi_row, mi_col + hbs, r, subsize);
-        break;
-      case PARTITION_VERT_B:
-        decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, subsize);
-        decode_token_and_recon_block(pbi, xd, mi_row, mi_col + hbs, r, bsize2);
-        decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col + hbs, r,
-                                     bsize2);
-        break;
-#endif
-      default: assert(0 && "Invalid partition type");
+    for (int idy = 0; idy < tx_size_high_unit[tx_size]; ++idy) {
+      for (int idx = 0; idx < tx_size_wide_unit[tx_size]; ++idx) {
+        const int index =
+            av1_get_txb_size_index(bsize, blk_row + idy, blk_col + idx);
+        mbmi->inter_tx_size[index] = tx_size;
+      }
     }
+    mbmi->tx_size = tx_size;
+    txfm_partition_update(xd->above_txfm_context + blk_col,
+                          xd->left_txfm_context + blk_row, tx_size, tx_size);
+  }
+}
+
+static TX_SIZE read_selected_tx_size(MACROBLOCKD *xd, aom_reader *r) {
+  // TODO(debargha): Clean up the logic here. This function should only
+  // be called for intra.
+  const BLOCK_SIZE bsize = xd->mi[0]->sb_type;
+  const int32_t tx_size_cat = bsize_to_tx_size_cat(bsize);
+  const int max_depths = bsize_to_max_depth(bsize);
+  const int ctx = get_tx_size_context(xd);
+  FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
+  const int depth = aom_read_symbol(r, ec_ctx->tx_size_cdf[tx_size_cat][ctx],
+                                    max_depths + 1, ACCT_STR);
+  assert(depth >= 0 && depth <= max_depths);
+  const TX_SIZE tx_size = depth_to_tx_size(depth, bsize);
+  return tx_size;
+}
+
+static TX_SIZE read_tx_size(AV1_COMMON *cm, MACROBLOCKD *xd, int is_inter,
+                            int allow_select_inter, aom_reader *r) {
+  const TX_MODE tx_mode = cm->tx_mode;
+  const BLOCK_SIZE bsize = xd->mi[0]->sb_type;
+  if (xd->lossless[xd->mi[0]->segment_id]) return TX_4X4;
+
+  if (block_signals_txsize(bsize)) {
+    if ((!is_inter || allow_select_inter) && tx_mode == TX_MODE_SELECT) {
+      const TX_SIZE coded_tx_size = read_selected_tx_size(xd, r);
+      return coded_tx_size;
+    } else {
+      return tx_size_from_tx_mode(bsize, tx_mode);
+    }
+  } else {
+    assert(IMPLIES(tx_mode == ONLY_4X4, bsize == BLOCK_4X4));
+    return max_txsize_rect_lookup[bsize];
   }
 }
-#endif
 
 static void decode_block(AV1Decoder *const pbi, MACROBLOCKD *const xd,
-#if CONFIG_SUPERTX
-                         int supertx_enabled,
-#endif  // CONFIG_SUPERTX
                          int mi_row, int mi_col, aom_reader *r,
-#if CONFIG_EXT_PARTITION_TYPES
-                         PARTITION_TYPE partition,
-#endif  // CONFIG_EXT_PARTITION_TYPES
-                         BLOCK_SIZE bsize) {
-  decode_mbmi_block(pbi, xd,
-#if CONFIG_SUPERTX
-                    supertx_enabled,
-#endif
-                    mi_row, mi_col, r,
-#if CONFIG_EXT_PARTITION_TYPES
-                    partition,
-#endif
-                    bsize);
+                         PARTITION_TYPE partition, BLOCK_SIZE bsize) {
+  decode_mbmi_block(pbi, xd, mi_row, mi_col, r, partition, bsize);
+
+  av1_visit_palette(pbi, xd, mi_row, mi_col, r, bsize,
+                    av1_decode_palette_tokens);
+
+  AV1_COMMON *cm = &pbi->common;
+  MB_MODE_INFO *mbmi = xd->mi[0];
+  int inter_block_tx = is_inter_block(mbmi) || is_intrabc_block(mbmi);
+  if (cm->tx_mode == TX_MODE_SELECT && block_signals_txsize(bsize) &&
+      !mbmi->skip && inter_block_tx && !xd->lossless[mbmi->segment_id]) {
+    const TX_SIZE max_tx_size = max_txsize_rect_lookup[bsize];
+    const int bh = tx_size_high_unit[max_tx_size];
+    const int bw = tx_size_wide_unit[max_tx_size];
+    const int width = block_size_wide[bsize] >> tx_size_wide_log2[0];
+    const int height = block_size_high[bsize] >> tx_size_high_log2[0];
+
+    for (int idy = 0; idy < height; idy += bh)
+      for (int idx = 0; idx < width; idx += bw)
+        read_tx_size_vartx(xd, mbmi, max_tx_size, 0, idy, idx, r);
+  } else {
+    mbmi->tx_size = read_tx_size(cm, xd, inter_block_tx, !mbmi->skip, r);
+    if (inter_block_tx)
+      memset(mbmi->inter_tx_size, mbmi->tx_size, sizeof(mbmi->inter_tx_size));
+    set_txfm_ctxs(mbmi->tx_size, xd->n8_w, xd->n8_h,
+                  mbmi->skip && is_inter_block(mbmi), xd);
+  }
 
-#if !(CONFIG_MOTION_VAR && NC_MODE_INFO)
-#if CONFIG_SUPERTX
-  if (!supertx_enabled)
-#endif  // CONFIG_SUPERTX
-    decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, bsize);
-#endif
+  decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, bsize);
 }
 
-static PARTITION_TYPE read_partition(AV1_COMMON *cm, MACROBLOCKD *xd,
-                                     int mi_row, int mi_col, aom_reader *r,
-                                     int has_rows, int has_cols,
+static PARTITION_TYPE read_partition(MACROBLOCKD *xd, int mi_row, int mi_col,
+                                     aom_reader *r, int has_rows, int has_cols,
                                      BLOCK_SIZE bsize) {
-#if CONFIG_UNPOISON_PARTITION_CTX
-  const int ctx =
-      partition_plane_context(xd, mi_row, mi_col, has_rows, has_cols, bsize);
-#else
   const int ctx = partition_plane_context(xd, mi_row, mi_col, bsize);
-#endif
-  PARTITION_TYPE p;
   FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
-  (void)cm;
 
-  aom_cdf_prob *partition_cdf = (ctx >= 0) ? ec_ctx->partition_cdf[ctx] : NULL;
+  if (!has_rows && !has_cols) return PARTITION_SPLIT;
 
+  assert(ctx >= 0);
+  aom_cdf_prob *partition_cdf = ec_ctx->partition_cdf[ctx];
   if (has_rows && has_cols) {
-#if CONFIG_EXT_PARTITION_TYPES
-    const int num_partition_types =
-        (mi_width_log2_lookup[bsize] > mi_width_log2_lookup[BLOCK_8X8])
-            ? EXT_PARTITION_TYPES
-            : PARTITION_TYPES;
-#else
-    const int num_partition_types = PARTITION_TYPES;
-#endif  // CONFIG_EXT_PARTITION_TYPES
-    p = (PARTITION_TYPE)aom_read_symbol(r, partition_cdf, num_partition_types,
-                                        ACCT_STR);
+    return (PARTITION_TYPE)aom_read_symbol(
+        r, partition_cdf, partition_cdf_length(bsize), ACCT_STR);
   } else if (!has_rows && has_cols) {
     assert(bsize > BLOCK_8X8);
     aom_cdf_prob cdf[2];
-    partition_gather_vert_alike(cdf, partition_cdf);
+    partition_gather_vert_alike(cdf, partition_cdf, bsize);
     assert(cdf[1] == AOM_ICDF(CDF_PROB_TOP));
-    p = aom_read_cdf(r, cdf, 2, ACCT_STR) ? PARTITION_SPLIT : PARTITION_HORZ;
-    // gather cols
-  } else if (has_rows && !has_cols) {
+    return aom_read_cdf(r, cdf, 2, ACCT_STR) ? PARTITION_SPLIT : PARTITION_HORZ;
+  } else {
+    assert(has_rows && !has_cols);
     assert(bsize > BLOCK_8X8);
     aom_cdf_prob cdf[2];
-    partition_gather_horz_alike(cdf, partition_cdf);
+    partition_gather_horz_alike(cdf, partition_cdf, bsize);
     assert(cdf[1] == AOM_ICDF(CDF_PROB_TOP));
-    p = aom_read_cdf(r, cdf, 2, ACCT_STR) ? PARTITION_SPLIT : PARTITION_VERT;
-  } else {
-    p = PARTITION_SPLIT;
+    return aom_read_cdf(r, cdf, 2, ACCT_STR) ? PARTITION_SPLIT : PARTITION_VERT;
   }
-
-  return p;
 }
 
-#if CONFIG_SUPERTX
-static int read_skip(AV1_COMMON *cm, const MACROBLOCKD *xd, int segment_id,
-                     aom_reader *r) {
-  if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP)) {
-    return 1;
-  } else {
-    const int ctx = av1_get_skip_context(xd);
-#if CONFIG_NEW_MULTISYMBOL
-    FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
-    const int skip = aom_read_symbol(r, ec_ctx->skip_cdfs[ctx], 2, ACCT_STR);
-#else
-    const int skip = aom_read(r, cm->fc->skip_probs[ctx], ACCT_STR);
-#endif
-    FRAME_COUNTS *counts = xd->counts;
-    if (counts) ++counts->skip[ctx][skip];
-    return skip;
-  }
-}
-#endif  // CONFIG_SUPERTX
-
 // TODO(slavarnway): eliminate bsize and subsize in future commits
 static void decode_partition(AV1Decoder *const pbi, MACROBLOCKD *const xd,
-#if CONFIG_SUPERTX
-                             int supertx_enabled,
-#endif
                              int mi_row, int mi_col, aom_reader *r,
                              BLOCK_SIZE bsize) {
   AV1_COMMON *const cm = &pbi->common;
-  const int num_8x8_wh = mi_size_wide[bsize];
-  const int hbs = num_8x8_wh >> 1;
-#if CONFIG_EXT_PARTITION_TYPES && CONFIG_EXT_PARTITION_TYPES_AB
-  const int qbs = num_8x8_wh >> 2;
-#endif
-#if CONFIG_CB4X4
-  const int unify_bsize = 1;
-#else
-  const int unify_bsize = 0;
-#endif
+  const int bw = mi_size_wide[bsize];
+  const int hbs = bw >> 1;
   PARTITION_TYPE partition;
   BLOCK_SIZE subsize;
-#if CONFIG_EXT_PARTITION_TYPES
-  const int quarter_step = num_8x8_wh / 4;
-  int i;
-#if !CONFIG_EXT_PARTITION_TYPES_AB
-  BLOCK_SIZE bsize2 = get_subsize(bsize, PARTITION_SPLIT);
-#endif
-#endif
+  const int quarter_step = bw / 4;
+  BLOCK_SIZE bsize2 = get_partition_subsize(bsize, PARTITION_SPLIT);
   const int has_rows = (mi_row + hbs) < cm->mi_rows;
   const int has_cols = (mi_col + hbs) < cm->mi_cols;
-#if CONFIG_SUPERTX
-  const int read_token = !supertx_enabled;
-  int skip = 0;
-  TX_SIZE supertx_size = max_txsize_lookup[bsize];
-  const TileInfo *const tile = &xd->tile;
-  int txfm = DCT_DCT;
-#endif  // CONFIG_SUPERTX
 
   if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return;
 
+  const int num_planes = av1_num_planes(cm);
+  for (int plane = 0; plane < num_planes; ++plane) {
+    int rcol0, rcol1, rrow0, rrow1, tile_tl_idx;
+    if (av1_loop_restoration_corners_in_sb(cm, plane, mi_row, mi_col, bsize,
+                                           &rcol0, &rcol1, &rrow0, &rrow1,
+                                           &tile_tl_idx)) {
+      const int rstride = cm->rst_info[plane].horz_units_per_tile;
+      for (int rrow = rrow0; rrow < rrow1; ++rrow) {
+        for (int rcol = rcol0; rcol < rcol1; ++rcol) {
+          const int runit_idx = tile_tl_idx + rcol + rrow * rstride;
+          loop_restoration_read_sb_coeffs(cm, xd, r, plane, runit_idx);
+        }
+      }
+    }
+  }
+
   partition = (bsize < BLOCK_8X8) ? PARTITION_NONE
-                                  : read_partition(cm, xd, mi_row, mi_col, r,
+                                  : read_partition(xd, mi_row, mi_col, r,
                                                    has_rows, has_cols, bsize);
-  subsize = subsize_lookup[partition][bsize];  // get_subsize(bsize, partition);
+  subsize = get_partition_subsize(bsize, partition);
 
   // Check the bitstream is conformant: if there is subsampling on the
   // chroma planes, subsize must subsample to a valid block size.
   const struct macroblockd_plane *const pd_u = &xd->plane[1];
-  if (get_plane_block_size(subsize, pd_u) == BLOCK_INVALID) {
+  if (get_plane_block_size(subsize, pd_u->subsampling_x, pd_u->subsampling_y) ==
+      BLOCK_INVALID) {
     aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
                        "Block size %dx%d invalid with this subsampling mode",
                        block_size_wide[subsize], block_size_high[subsize]);
   }
 
-#if CONFIG_PVQ
-  assert(partition < PARTITION_TYPES);
-  assert(subsize < BLOCK_SIZES_ALL);
-#endif
-#if CONFIG_SUPERTX
-  if (!frame_is_intra_only(cm) && partition != PARTITION_NONE &&
-      bsize <= MAX_SUPERTX_BLOCK_SIZE && !supertx_enabled && !xd->lossless[0]) {
-    const int supertx_context = partition_supertx_context_lookup[partition];
-    supertx_enabled = aom_read(
-        r, cm->fc->supertx_prob[supertx_context][supertx_size], ACCT_STR);
-    if (xd->counts)
-      xd->counts->supertx[supertx_context][supertx_size][supertx_enabled]++;
-#if CONFIG_VAR_TX
-    if (supertx_enabled) xd->supertx_size = supertx_size;
-#endif
-  }
-#endif  // CONFIG_SUPERTX
-
-#if CONFIG_SUPERTX
-#define DEC_BLOCK_STX_ARG supertx_enabled,
-#else
 #define DEC_BLOCK_STX_ARG
-#endif
-#if CONFIG_EXT_PARTITION_TYPES
 #define DEC_BLOCK_EPT_ARG partition,
-#else
-#define DEC_BLOCK_EPT_ARG
-#endif
 #define DEC_BLOCK(db_r, db_c, db_subsize)                   \
   decode_block(pbi, xd, DEC_BLOCK_STX_ARG(db_r), (db_c), r, \
                DEC_BLOCK_EPT_ARG(db_subsize))
 #define DEC_PARTITION(db_r, db_c, db_subsize) \
   decode_partition(pbi, xd, DEC_BLOCK_STX_ARG(db_r), (db_c), r, (db_subsize))
 
-  if (!hbs && !unify_bsize) {
-    // calculate bmode block dimensions (log 2)
-    xd->bmode_blocks_wl = 1 >> !!(partition & PARTITION_VERT);
-    xd->bmode_blocks_hl = 1 >> !!(partition & PARTITION_HORZ);
-    DEC_BLOCK(mi_row, mi_col, subsize);
-  } else {
-    switch (partition) {
-      case PARTITION_NONE: DEC_BLOCK(mi_row, mi_col, subsize); break;
-      case PARTITION_HORZ:
-        DEC_BLOCK(mi_row, mi_col, subsize);
-        if (has_rows) DEC_BLOCK(mi_row + hbs, mi_col, subsize);
-        break;
-      case PARTITION_VERT:
-        DEC_BLOCK(mi_row, mi_col, subsize);
-        if (has_cols) DEC_BLOCK(mi_row, mi_col + hbs, subsize);
-        break;
-      case PARTITION_SPLIT:
-        DEC_PARTITION(mi_row, mi_col, subsize);
-        DEC_PARTITION(mi_row, mi_col + hbs, subsize);
-        DEC_PARTITION(mi_row + hbs, mi_col, subsize);
-        DEC_PARTITION(mi_row + hbs, mi_col + hbs, subsize);
-        break;
-#if CONFIG_EXT_PARTITION_TYPES
-#if CONFIG_EXT_PARTITION_TYPES_AB
-      case PARTITION_HORZ_A:
-        DEC_BLOCK(mi_row, mi_col, get_subsize(bsize, PARTITION_HORZ_4));
-        DEC_BLOCK(mi_row + qbs, mi_col, get_subsize(bsize, PARTITION_HORZ_4));
-        DEC_BLOCK(mi_row + hbs, mi_col, subsize);
-        break;
-      case PARTITION_HORZ_B:
-        DEC_BLOCK(mi_row, mi_col, subsize);
-        DEC_BLOCK(mi_row + hbs, mi_col, get_subsize(bsize, PARTITION_HORZ_4));
-        if (mi_row + 3 * qbs < cm->mi_rows)
-          DEC_BLOCK(mi_row + 3 * qbs, mi_col,
-                    get_subsize(bsize, PARTITION_HORZ_4));
-        break;
-      case PARTITION_VERT_A:
-        DEC_BLOCK(mi_row, mi_col, get_subsize(bsize, PARTITION_VERT_4));
-        DEC_BLOCK(mi_row, mi_col + qbs, get_subsize(bsize, PARTITION_VERT_4));
-        DEC_BLOCK(mi_row, mi_col + hbs, subsize);
-        break;
-      case PARTITION_VERT_B:
-        DEC_BLOCK(mi_row, mi_col, subsize);
-        DEC_BLOCK(mi_row, mi_col + hbs, get_subsize(bsize, PARTITION_VERT_4));
-        if (mi_col + 3 * qbs < cm->mi_cols)
-          DEC_BLOCK(mi_row, mi_col + 3 * qbs,
-                    get_subsize(bsize, PARTITION_VERT_4));
-        break;
-#else
-      case PARTITION_HORZ_A:
-        DEC_BLOCK(mi_row, mi_col, bsize2);
-        DEC_BLOCK(mi_row, mi_col + hbs, bsize2);
-        DEC_BLOCK(mi_row + hbs, mi_col, subsize);
-        break;
-      case PARTITION_HORZ_B:
-        DEC_BLOCK(mi_row, mi_col, subsize);
-        DEC_BLOCK(mi_row + hbs, mi_col, bsize2);
-        DEC_BLOCK(mi_row + hbs, mi_col + hbs, bsize2);
-        break;
-      case PARTITION_VERT_A:
-        DEC_BLOCK(mi_row, mi_col, bsize2);
-        DEC_BLOCK(mi_row + hbs, mi_col, bsize2);
-        DEC_BLOCK(mi_row, mi_col + hbs, subsize);
-        break;
-      case PARTITION_VERT_B:
-        DEC_BLOCK(mi_row, mi_col, subsize);
-        DEC_BLOCK(mi_row, mi_col + hbs, bsize2);
-        DEC_BLOCK(mi_row + hbs, mi_col + hbs, bsize2);
-        break;
-#endif
-      case PARTITION_HORZ_4:
-        for (i = 0; i < 4; ++i) {
-          int this_mi_row = mi_row + i * quarter_step;
-          if (i > 0 && this_mi_row >= cm->mi_rows) break;
-          DEC_BLOCK(this_mi_row, mi_col, subsize);
-        }
-        break;
-      case PARTITION_VERT_4:
-        for (i = 0; i < 4; ++i) {
-          int this_mi_col = mi_col + i * quarter_step;
-          if (i > 0 && this_mi_col >= cm->mi_cols) break;
-          DEC_BLOCK(mi_row, this_mi_col, subsize);
-        }
-        break;
-#endif  // CONFIG_EXT_PARTITION_TYPES
-      default: assert(0 && "Invalid partition type");
-    }
+  switch (partition) {
+    case PARTITION_NONE: DEC_BLOCK(mi_row, mi_col, subsize); break;
+    case PARTITION_HORZ:
+      DEC_BLOCK(mi_row, mi_col, subsize);
+      if (has_rows) DEC_BLOCK(mi_row + hbs, mi_col, subsize);
+      break;
+    case PARTITION_VERT:
+      DEC_BLOCK(mi_row, mi_col, subsize);
+      if (has_cols) DEC_BLOCK(mi_row, mi_col + hbs, subsize);
+      break;
+    case PARTITION_SPLIT:
+      DEC_PARTITION(mi_row, mi_col, subsize);
+      DEC_PARTITION(mi_row, mi_col + hbs, subsize);
+      DEC_PARTITION(mi_row + hbs, mi_col, subsize);
+      DEC_PARTITION(mi_row + hbs, mi_col + hbs, subsize);
+      break;
+    case PARTITION_HORZ_A:
+      DEC_BLOCK(mi_row, mi_col, bsize2);
+      DEC_BLOCK(mi_row, mi_col + hbs, bsize2);
+      DEC_BLOCK(mi_row + hbs, mi_col, subsize);
+      break;
+    case PARTITION_HORZ_B:
+      DEC_BLOCK(mi_row, mi_col, subsize);
+      DEC_BLOCK(mi_row + hbs, mi_col, bsize2);
+      DEC_BLOCK(mi_row + hbs, mi_col + hbs, bsize2);
+      break;
+    case PARTITION_VERT_A:
+      DEC_BLOCK(mi_row, mi_col, bsize2);
+      DEC_BLOCK(mi_row + hbs, mi_col, bsize2);
+      DEC_BLOCK(mi_row, mi_col + hbs, subsize);
+      break;
+    case PARTITION_VERT_B:
+      DEC_BLOCK(mi_row, mi_col, subsize);
+      DEC_BLOCK(mi_row, mi_col + hbs, bsize2);
+      DEC_BLOCK(mi_row + hbs, mi_col + hbs, bsize2);
+      break;
+    case PARTITION_HORZ_4:
+      for (int i = 0; i < 4; ++i) {
+        int this_mi_row = mi_row + i * quarter_step;
+        if (i > 0 && this_mi_row >= cm->mi_rows) break;
+        DEC_BLOCK(this_mi_row, mi_col, subsize);
+      }
+      break;
+    case PARTITION_VERT_4:
+      for (int i = 0; i < 4; ++i) {
+        int this_mi_col = mi_col + i * quarter_step;
+        if (i > 0 && this_mi_col >= cm->mi_cols) break;
+        DEC_BLOCK(mi_row, this_mi_col, subsize);
+      }
+      break;
+    default: assert(0 && "Invalid partition type");
   }
 
 #undef DEC_PARTITION
@@ -2472,219 +1513,13 @@ static void decode_partition(AV1Decoder *const pbi, MACROBLOCKD *const xd,
 #undef DEC_BLOCK_EPT_ARG
 #undef DEC_BLOCK_STX_ARG
 
-#if CONFIG_SUPERTX
-  if (supertx_enabled && read_token) {
-    uint8_t *dst_buf[3];
-    int dst_stride[3], i;
-    int offset = mi_row * cm->mi_stride + mi_col;
-
-    set_segment_id_supertx(cm, mi_row, mi_col, bsize);
-
-    if (cm->delta_q_present_flag) {
-      for (i = 0; i < MAX_SEGMENTS; i++) {
-        int j;
-        for (j = 0; j < MAX_MB_PLANE; ++j) {
-          const int dc_delta_q = j == 0 ? cm->y_dc_delta_q : cm->uv_dc_delta_q;
-          const int ac_delta_q = j == 0 ? 0 : cm->uv_ac_delta_q;
-
-          xd->plane[j].seg_dequant[i][0] =
-              av1_dc_quant(xd->current_qindex, dc_delta_q, cm->bit_depth);
-          xd->plane[j].seg_dequant[i][1] =
-              av1_ac_quant(xd->current_qindex, ac_delta_q, cm->bit_depth);
-        }
-      }
-    }
-
-    xd->mi = cm->mi_grid_visible + offset;
-    xd->mi[0] = cm->mi + offset;
-    set_mi_row_col(xd, tile, mi_row, mi_size_high[bsize], mi_col,
-                   mi_size_wide[bsize],
-#if CONFIG_DEPENDENT_HORZTILES
-                   cm->dependent_horz_tiles,
-#endif  // CONFIG_DEPENDENT_HORZTILES
-                   cm->mi_rows, cm->mi_cols);
-    set_skip_context(xd, mi_row, mi_col);
-    skip = read_skip(cm, xd, xd->mi[0]->mbmi.segment_id_supertx, r);
-    if (skip) {
-      av1_reset_skip_context(xd, mi_row, mi_col, bsize);
-    } else {
-      FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
-#if CONFIG_EXT_TX
-      if (get_ext_tx_types(supertx_size, bsize, 1, cm->reduced_tx_set_used) >
-          1) {
-        const int eset =
-            get_ext_tx_set(supertx_size, bsize, 1, cm->reduced_tx_set_used);
-        if (eset > 0) {
-          const TxSetType tx_set_type = get_ext_tx_set_type(
-              supertx_size, bsize, 1, cm->reduced_tx_set_used);
-          const int packed_sym =
-              aom_read_symbol(r, ec_ctx->inter_ext_tx_cdf[eset][supertx_size],
-                              av1_num_ext_tx_set[tx_set_type], ACCT_STR);
-          txfm = av1_ext_tx_inv[tx_set_type][packed_sym];
-#if CONFIG_ENTROPY_STATS
-          if (xd->counts) ++xd->counts->inter_ext_tx[eset][supertx_size][txfm];
-#endif  // CONFIG_ENTROPY_STATS
-        }
-      }
-#else
-      if (supertx_size < TX_32X32) {
-        txfm = aom_read_symbol(r, ec_ctx->inter_ext_tx_cdf[supertx_size],
-                               TX_TYPES, ACCT_STR);
-#if CONFIG_ENTROPY_STATS
-        if (xd->counts) ++xd->counts->inter_ext_tx[supertx_size][txfm];
-#endif  // CONFIG_ENTROPY_STATS
-      }
-#endif  // CONFIG_EXT_TX
-    }
-
-    av1_setup_dst_planes(xd->plane, bsize, get_frame_new_buffer(cm), mi_row,
-                         mi_col);
-    for (i = 0; i < MAX_MB_PLANE; i++) {
-      dst_buf[i] = xd->plane[i].dst.buf;
-      dst_stride[i] = xd->plane[i].dst.stride;
-    }
-    dec_predict_sb_complex(pbi, xd, tile, mi_row, mi_col, mi_row, mi_col, bsize,
-                           bsize, dst_buf, dst_stride);
-
-    if (!skip) {
-      int eobtotal = 0;
-      MB_MODE_INFO *mbmi;
-      set_offsets_topblock(cm, xd, tile, bsize, mi_row, mi_col);
-      mbmi = &xd->mi[0]->mbmi;
-      mbmi->tx_type = txfm;
-      assert(mbmi->segment_id_supertx != MAX_SEGMENTS);
-      for (i = 0; i < MAX_MB_PLANE; ++i) {
-        const struct macroblockd_plane *const pd = &xd->plane[i];
-        int row, col;
-        const TX_SIZE tx_size = av1_get_tx_size(i, xd);
-        const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
-        const int stepr = tx_size_high_unit[tx_size];
-        const int stepc = tx_size_wide_unit[tx_size];
-        const int max_blocks_wide = max_block_wide(xd, plane_bsize, i);
-        const int max_blocks_high = max_block_high(xd, plane_bsize, i);
-
-        for (row = 0; row < max_blocks_high; row += stepr)
-          for (col = 0; col < max_blocks_wide; col += stepc)
-            eobtotal += reconstruct_inter_block(
-                cm, xd, r, mbmi->segment_id_supertx, i, row, col, tx_size);
-      }
-      if ((unify_bsize || !(subsize < BLOCK_8X8)) && eobtotal == 0) skip = 1;
-    }
-    set_param_topblock(cm, xd, bsize, mi_row, mi_col, txfm, skip);
-  }
-#endif  // CONFIG_SUPERTX
-
-#if CONFIG_EXT_PARTITION_TYPES
   update_ext_partition_context(xd, mi_row, mi_col, subsize, bsize, partition);
-#else
-  // update partition context
-  if (bsize >= BLOCK_8X8 &&
-      (bsize == BLOCK_8X8 || partition != PARTITION_SPLIT))
-    update_partition_context(xd, mi_row, mi_col, subsize, bsize);
-#endif  // CONFIG_EXT_PARTITION_TYPES
-
-#if CONFIG_LPF_SB
-  if (bsize == cm->sb_size) {
-    int filt_lvl;
-    if (mi_row == 0 && mi_col == 0) {
-      filt_lvl = aom_read_literal(r, 6, ACCT_STR);
-      cm->mi_grid_visible[0]->mbmi.reuse_sb_lvl = 0;
-      cm->mi_grid_visible[0]->mbmi.delta = 0;
-      cm->mi_grid_visible[0]->mbmi.sign = 0;
-    } else {
-      int prev_mi_row, prev_mi_col;
-      if (mi_col - MAX_MIB_SIZE < 0) {
-        prev_mi_row = mi_row - MAX_MIB_SIZE;
-        prev_mi_col = mi_col;
-      } else {
-        prev_mi_row = mi_row;
-        prev_mi_col = mi_col - MAX_MIB_SIZE;
-      }
-
-      MB_MODE_INFO *curr_mbmi =
-          &cm->mi_grid_visible[mi_row * cm->mi_stride + mi_col]->mbmi;
-      MB_MODE_INFO *prev_mbmi =
-          &cm->mi_grid_visible[prev_mi_row * cm->mi_stride + prev_mi_col]->mbmi;
-      const uint8_t prev_lvl = prev_mbmi->filt_lvl;
-
-      const int reuse_ctx = prev_mbmi->reuse_sb_lvl;
-      const int reuse_prev_lvl = aom_read_symbol(
-          r, xd->tile_ctx->lpf_reuse_cdf[reuse_ctx], 2, ACCT_STR);
-      curr_mbmi->reuse_sb_lvl = reuse_prev_lvl;
-
-      if (reuse_prev_lvl) {
-        filt_lvl = prev_lvl;
-        curr_mbmi->delta = 0;
-        curr_mbmi->sign = 0;
-      } else {
-        const int delta_ctx = prev_mbmi->delta;
-        unsigned int delta = aom_read_symbol(
-            r, xd->tile_ctx->lpf_delta_cdf[delta_ctx], DELTA_RANGE, ACCT_STR);
-        curr_mbmi->delta = delta;
-        delta *= LPF_STEP;
-
-        if (delta) {
-          const int sign_ctx = prev_mbmi->sign;
-          const int sign = aom_read_symbol(
-              r, xd->tile_ctx->lpf_sign_cdf[reuse_ctx][sign_ctx], 2, ACCT_STR);
-          curr_mbmi->sign = sign;
-          filt_lvl = sign ? prev_lvl + delta : prev_lvl - delta;
-        } else {
-          filt_lvl = prev_lvl;
-          curr_mbmi->sign = 0;
-        }
-      }
-    }
-
-    av1_loop_filter_sb_level_init(cm, mi_row, mi_col, filt_lvl);
-  }
-#endif
-
-#if CONFIG_CDEF
-  if (bsize == cm->sb_size) {
-    int width_step = mi_size_wide[BLOCK_64X64];
-    int height_step = mi_size_wide[BLOCK_64X64];
-    int w, h;
-    for (h = 0; (h < mi_size_high[cm->sb_size]) && (mi_row + h < cm->mi_rows);
-         h += height_step) {
-      for (w = 0; (w < mi_size_wide[cm->sb_size]) && (mi_col + w < cm->mi_cols);
-           w += width_step) {
-        if (!cm->all_lossless && !sb_all_skip(cm, mi_row + h, mi_col + w))
-          cm->mi_grid_visible[(mi_row + h) * cm->mi_stride + (mi_col + w)]
-              ->mbmi.cdef_strength =
-              aom_read_literal(r, cm->cdef_bits, ACCT_STR);
-        else
-          cm->mi_grid_visible[(mi_row + h) * cm->mi_stride + (mi_col + w)]
-              ->mbmi.cdef_strength = -1;
-      }
-    }
-  }
-#endif  // CONFIG_CDEF
-#if CONFIG_LOOP_RESTORATION
-  for (int plane = 0; plane < MAX_MB_PLANE; ++plane) {
-    int rcol0, rcol1, rrow0, rrow1, nhtiles;
-    if (av1_loop_restoration_corners_in_sb(cm, plane, mi_row, mi_col, bsize,
-                                           &rcol0, &rcol1, &rrow0, &rrow1,
-                                           &nhtiles)) {
-      for (int rrow = rrow0; rrow < rrow1; ++rrow) {
-        for (int rcol = rcol0; rcol < rcol1; ++rcol) {
-          int rtile_idx = rcol + rrow * nhtiles;
-          loop_restoration_read_sb_coeffs(cm, xd, r, plane, rtile_idx);
-        }
-      }
-    }
-  }
-#endif
 }
 
 static void setup_bool_decoder(const uint8_t *data, const uint8_t *data_end,
                                const size_t read_size,
                                struct aom_internal_error_info *error_info,
-                               aom_reader *r,
-#if CONFIG_ANS && ANS_MAX_SYMBOLS
-                               int window_size,
-#endif  // CONFIG_ANS && ANS_MAX_SYMBOLS
-                               aom_decrypt_cb decrypt_cb, void *decrypt_state) {
+                               aom_reader *r, uint8_t allow_update_cdf) {
   // Validate the calculated partition length. If the buffer
   // described by the partition can't be fully read, then restrict
   // it to the portion that can be (for EC mode) or throw an error.
@@ -2692,117 +1527,147 @@ static void setup_bool_decoder(const uint8_t *data, const uint8_t *data_end,
     aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME,
                        "Truncated packet or corrupt tile length");
 
-#if CONFIG_ANS && ANS_MAX_SYMBOLS
-  r->window_size = window_size;
-#endif
-  if (aom_reader_init(r, data, read_size, decrypt_cb, decrypt_state))
+  if (aom_reader_init(r, data, read_size))
     aom_internal_error(error_info, AOM_CODEC_MEM_ERROR,
                        "Failed to allocate bool decoder %d", 1);
+
+  r->allow_update_cdf = allow_update_cdf;
 }
 
 static void setup_segmentation(AV1_COMMON *const cm,
                                struct aom_read_bit_buffer *rb) {
   struct segmentation *const seg = &cm->seg;
-  int i, j;
 
   seg->update_map = 0;
   seg->update_data = 0;
   seg->temporal_update = 0;
 
   seg->enabled = aom_rb_read_bit(rb);
-  if (!seg->enabled) return;
+  if (!seg->enabled) {
+    if (cm->cur_frame->seg_map)
+      memset(cm->cur_frame->seg_map, 0, (cm->mi_rows * cm->mi_cols));
 
-  // Segmentation map update
-  if (frame_is_intra_only(cm) || cm->error_resilient_mode) {
+    memset(seg, 0, sizeof(*seg));
+    segfeatures_copy(&cm->cur_frame->seg, seg);
+    return;
+  }
+  if (cm->seg.enabled && cm->prev_frame &&
+      (cm->mi_rows == cm->prev_frame->mi_rows) &&
+      (cm->mi_cols == cm->prev_frame->mi_cols)) {
+    cm->last_frame_seg_map = cm->prev_frame->seg_map;
+  } else {
+    cm->last_frame_seg_map = NULL;
+  }
+  // Read update flags
+  if (cm->primary_ref_frame == PRIMARY_REF_NONE) {
+    // These frames can't use previous frames, so must signal map + features
     seg->update_map = 1;
+    seg->temporal_update = 0;
+    seg->update_data = 1;
   } else {
     seg->update_map = aom_rb_read_bit(rb);
-  }
-  if (seg->update_map) {
-    if (frame_is_intra_only(cm) || cm->error_resilient_mode) {
-      seg->temporal_update = 0;
-    } else {
+    if (seg->update_map) {
       seg->temporal_update = aom_rb_read_bit(rb);
+    } else {
+      seg->temporal_update = 0;
     }
+    seg->update_data = aom_rb_read_bit(rb);
   }
 
   // Segmentation data update
-  seg->update_data = aom_rb_read_bit(rb);
   if (seg->update_data) {
-    seg->abs_delta = aom_rb_read_bit(rb);
-
     av1_clearall_segfeatures(seg);
 
-    for (i = 0; i < MAX_SEGMENTS; i++) {
-      for (j = 0; j < SEG_LVL_MAX; j++) {
+    for (int i = 0; i < MAX_SEGMENTS; i++) {
+      for (int j = 0; j < SEG_LVL_MAX; j++) {
         int data = 0;
         const int feature_enabled = aom_rb_read_bit(rb);
         if (feature_enabled) {
           av1_enable_segfeature(seg, i, j);
-          data = decode_unsigned_max(rb, av1_seg_feature_data_max(j));
-          if (av1_is_segfeature_signed(j))
-            data = aom_rb_read_bit(rb) ? -data : data;
+
+          const int data_max = av1_seg_feature_data_max(j);
+          const int data_min = -data_max;
+          const int ubits = get_unsigned_bits(data_max);
+
+          if (av1_is_segfeature_signed(j)) {
+            data = aom_rb_read_inv_signed_literal(rb, ubits);
+          } else {
+            data = aom_rb_read_literal(rb, ubits);
+          }
+
+          data = clamp(data, data_min, data_max);
         }
         av1_set_segdata(seg, i, j, data);
       }
     }
+    calculate_segdata(seg);
+  } else if (cm->prev_frame) {
+    segfeatures_copy(seg, &cm->prev_frame->seg);
   }
+  segfeatures_copy(&cm->cur_frame->seg, seg);
 }
 
-#if CONFIG_LOOP_RESTORATION
 static void decode_restoration_mode(AV1_COMMON *cm,
                                     struct aom_read_bit_buffer *rb) {
-  int p;
-  RestorationInfo *rsi = &cm->rst_info[0];
-  if (aom_rb_read_bit(rb)) {
-    rsi->frame_restoration_type =
-        aom_rb_read_bit(rb) ? RESTORE_SGRPROJ : RESTORE_WIENER;
-  } else {
-    rsi->frame_restoration_type =
-        aom_rb_read_bit(rb) ? RESTORE_SWITCHABLE : RESTORE_NONE;
-  }
-  for (p = 1; p < MAX_MB_PLANE; ++p) {
-    rsi = &cm->rst_info[p];
+  assert(!cm->all_lossless);
+  const int num_planes = av1_num_planes(cm);
+  if (cm->allow_intrabc) return;
+  int all_none = 1, chroma_none = 1;
+  for (int p = 0; p < num_planes; ++p) {
+    RestorationInfo *rsi = &cm->rst_info[p];
     if (aom_rb_read_bit(rb)) {
       rsi->frame_restoration_type =
           aom_rb_read_bit(rb) ? RESTORE_SGRPROJ : RESTORE_WIENER;
     } else {
-      rsi->frame_restoration_type = RESTORE_NONE;
+      rsi->frame_restoration_type =
+          aom_rb_read_bit(rb) ? RESTORE_SWITCHABLE : RESTORE_NONE;
+    }
+    if (rsi->frame_restoration_type != RESTORE_NONE) {
+      all_none = 0;
+      chroma_none &= p == 0;
     }
   }
+  if (!all_none) {
+    assert(cm->seq_params.sb_size == BLOCK_64X64 ||
+           cm->seq_params.sb_size == BLOCK_128X128);
+    const int sb_size = cm->seq_params.sb_size == BLOCK_128X128 ? 128 : 64;
 
-  cm->rst_info[0].restoration_tilesize = RESTORATION_TILESIZE_MAX;
-  cm->rst_info[1].restoration_tilesize = RESTORATION_TILESIZE_MAX;
-  cm->rst_info[2].restoration_tilesize = RESTORATION_TILESIZE_MAX;
-  if (cm->rst_info[0].frame_restoration_type != RESTORE_NONE ||
-      cm->rst_info[1].frame_restoration_type != RESTORE_NONE ||
-      cm->rst_info[2].frame_restoration_type != RESTORE_NONE) {
-    rsi = &cm->rst_info[0];
-    rsi->restoration_tilesize >>= aom_rb_read_bit(rb);
-    if (rsi->restoration_tilesize != RESTORATION_TILESIZE_MAX) {
-      rsi->restoration_tilesize >>= aom_rb_read_bit(rb);
+    for (int p = 0; p < num_planes; ++p)
+      cm->rst_info[p].restoration_unit_size = sb_size;
+
+    RestorationInfo *rsi = &cm->rst_info[0];
+
+    if (sb_size == 64) {
+      rsi->restoration_unit_size <<= aom_rb_read_bit(rb);
+    }
+    if (rsi->restoration_unit_size > 64) {
+      rsi->restoration_unit_size <<= aom_rb_read_bit(rb);
     }
-  }
-  int s = AOMMIN(cm->subsampling_x, cm->subsampling_y);
-  if (s && (cm->rst_info[1].frame_restoration_type != RESTORE_NONE ||
-            cm->rst_info[2].frame_restoration_type != RESTORE_NONE)) {
-    cm->rst_info[1].restoration_tilesize =
-        cm->rst_info[0].restoration_tilesize >> (aom_rb_read_bit(rb) * s);
   } else {
-    cm->rst_info[1].restoration_tilesize = cm->rst_info[0].restoration_tilesize;
+    const int size = RESTORATION_UNITSIZE_MAX;
+    for (int p = 0; p < num_planes; ++p)
+      cm->rst_info[p].restoration_unit_size = size;
   }
-  cm->rst_info[2].restoration_tilesize = cm->rst_info[1].restoration_tilesize;
 
-  cm->rst_info[0].procunit_width = cm->rst_info[0].procunit_height =
-      RESTORATION_PROC_UNIT_SIZE;
-  cm->rst_info[1].procunit_width = cm->rst_info[2].procunit_width =
-      RESTORATION_PROC_UNIT_SIZE >> cm->subsampling_x;
-  cm->rst_info[1].procunit_height = cm->rst_info[2].procunit_height =
-      RESTORATION_PROC_UNIT_SIZE >> cm->subsampling_y;
+  if (num_planes > 1) {
+    int s = AOMMIN(cm->subsampling_x, cm->subsampling_y);
+    if (s && !chroma_none) {
+      cm->rst_info[1].restoration_unit_size =
+          cm->rst_info[0].restoration_unit_size >> (aom_rb_read_bit(rb) * s);
+    } else {
+      cm->rst_info[1].restoration_unit_size =
+          cm->rst_info[0].restoration_unit_size;
+    }
+    cm->rst_info[2].restoration_unit_size =
+        cm->rst_info[1].restoration_unit_size;
+  }
 }
 
 static void read_wiener_filter(int wiener_win, WienerInfo *wiener_info,
                                WienerInfo *ref_wiener_info, aom_reader *rb) {
+  memset(wiener_info->vfilter, 0, sizeof(wiener_info->vfilter));
+  memset(wiener_info->hfilter, 0, sizeof(wiener_info->hfilter));
+
   if (wiener_win == WIENER_WIN)
     wiener_info->vfilter[0] = wiener_info->vfilter[WIENER_WIN - 1] =
         aom_read_primitive_refsubexpfin(
@@ -2860,75 +1725,109 @@ static void read_wiener_filter(int wiener_win, WienerInfo *wiener_info,
 static void read_sgrproj_filter(SgrprojInfo *sgrproj_info,
                                 SgrprojInfo *ref_sgrproj_info, aom_reader *rb) {
   sgrproj_info->ep = aom_read_literal(rb, SGRPROJ_PARAMS_BITS, ACCT_STR);
-  sgrproj_info->xqd[0] =
-      aom_read_primitive_refsubexpfin(
-          rb, SGRPROJ_PRJ_MAX0 - SGRPROJ_PRJ_MIN0 + 1, SGRPROJ_PRJ_SUBEXP_K,
-          ref_sgrproj_info->xqd[0] - SGRPROJ_PRJ_MIN0, ACCT_STR) +
-      SGRPROJ_PRJ_MIN0;
-  sgrproj_info->xqd[1] =
-      aom_read_primitive_refsubexpfin(
-          rb, SGRPROJ_PRJ_MAX1 - SGRPROJ_PRJ_MIN1 + 1, SGRPROJ_PRJ_SUBEXP_K,
-          ref_sgrproj_info->xqd[1] - SGRPROJ_PRJ_MIN1, ACCT_STR) +
-      SGRPROJ_PRJ_MIN1;
+  const sgr_params_type *params = &sgr_params[sgrproj_info->ep];
+
+  if (params->r[0] == 0) {
+    sgrproj_info->xqd[0] = 0;
+    sgrproj_info->xqd[1] =
+        aom_read_primitive_refsubexpfin(
+            rb, SGRPROJ_PRJ_MAX1 - SGRPROJ_PRJ_MIN1 + 1, SGRPROJ_PRJ_SUBEXP_K,
+            ref_sgrproj_info->xqd[1] - SGRPROJ_PRJ_MIN1, ACCT_STR) +
+        SGRPROJ_PRJ_MIN1;
+  } else if (params->r[1] == 0) {
+    sgrproj_info->xqd[0] =
+        aom_read_primitive_refsubexpfin(
+            rb, SGRPROJ_PRJ_MAX0 - SGRPROJ_PRJ_MIN0 + 1, SGRPROJ_PRJ_SUBEXP_K,
+            ref_sgrproj_info->xqd[0] - SGRPROJ_PRJ_MIN0, ACCT_STR) +
+        SGRPROJ_PRJ_MIN0;
+    sgrproj_info->xqd[1] = clamp((1 << SGRPROJ_PRJ_BITS) - sgrproj_info->xqd[0],
+                                 SGRPROJ_PRJ_MIN1, SGRPROJ_PRJ_MAX1);
+  } else {
+    sgrproj_info->xqd[0] =
+        aom_read_primitive_refsubexpfin(
+            rb, SGRPROJ_PRJ_MAX0 - SGRPROJ_PRJ_MIN0 + 1, SGRPROJ_PRJ_SUBEXP_K,
+            ref_sgrproj_info->xqd[0] - SGRPROJ_PRJ_MIN0, ACCT_STR) +
+        SGRPROJ_PRJ_MIN0;
+    sgrproj_info->xqd[1] =
+        aom_read_primitive_refsubexpfin(
+            rb, SGRPROJ_PRJ_MAX1 - SGRPROJ_PRJ_MIN1 + 1, SGRPROJ_PRJ_SUBEXP_K,
+            ref_sgrproj_info->xqd[1] - SGRPROJ_PRJ_MIN1, ACCT_STR) +
+        SGRPROJ_PRJ_MIN1;
+  }
+
   memcpy(ref_sgrproj_info, sgrproj_info, sizeof(*sgrproj_info));
 }
 
 static void loop_restoration_read_sb_coeffs(const AV1_COMMON *const cm,
                                             MACROBLOCKD *xd,
                                             aom_reader *const r, int plane,
-                                            int rtile_idx) {
-  const RestorationInfo *rsi = cm->rst_info + plane;
+                                            int runit_idx) {
+  const RestorationInfo *rsi = &cm->rst_info[plane];
+  RestorationUnitInfo *rui = &rsi->unit_info[runit_idx];
   if (rsi->frame_restoration_type == RESTORE_NONE) return;
 
+  assert(!cm->all_lossless);
+
   const int wiener_win = (plane > 0) ? WIENER_WIN_CHROMA : WIENER_WIN;
   WienerInfo *wiener_info = xd->wiener_info + plane;
   SgrprojInfo *sgrproj_info = xd->sgrproj_info + plane;
 
   if (rsi->frame_restoration_type == RESTORE_SWITCHABLE) {
-    assert(plane == 0);
-    rsi->restoration_type[rtile_idx] =
-        aom_read_tree(r, av1_switchable_restore_tree,
-                      cm->fc->switchable_restore_prob, ACCT_STR);
-
-    if (rsi->restoration_type[rtile_idx] == RESTORE_WIENER) {
-      read_wiener_filter(wiener_win, &rsi->wiener_info[rtile_idx], wiener_info,
-                         r);
-    } else if (rsi->restoration_type[rtile_idx] == RESTORE_SGRPROJ) {
-      read_sgrproj_filter(&rsi->sgrproj_info[rtile_idx], sgrproj_info, r);
+    rui->restoration_type =
+        aom_read_symbol(r, xd->tile_ctx->switchable_restore_cdf,
+                        RESTORE_SWITCHABLE_TYPES, ACCT_STR);
+    switch (rui->restoration_type) {
+      case RESTORE_WIENER:
+        read_wiener_filter(wiener_win, &rui->wiener_info, wiener_info, r);
+        break;
+      case RESTORE_SGRPROJ:
+        read_sgrproj_filter(&rui->sgrproj_info, sgrproj_info, r);
+        break;
+      default: assert(rui->restoration_type == RESTORE_NONE); break;
     }
   } else if (rsi->frame_restoration_type == RESTORE_WIENER) {
-    if (aom_read(r, RESTORE_NONE_WIENER_PROB, ACCT_STR)) {
-      rsi->restoration_type[rtile_idx] = RESTORE_WIENER;
-      read_wiener_filter(wiener_win, &rsi->wiener_info[rtile_idx], wiener_info,
-                         r);
+    if (aom_read_symbol(r, xd->tile_ctx->wiener_restore_cdf, 2, ACCT_STR)) {
+      rui->restoration_type = RESTORE_WIENER;
+      read_wiener_filter(wiener_win, &rui->wiener_info, wiener_info, r);
     } else {
-      rsi->restoration_type[rtile_idx] = RESTORE_NONE;
+      rui->restoration_type = RESTORE_NONE;
     }
   } else if (rsi->frame_restoration_type == RESTORE_SGRPROJ) {
-    if (aom_read(r, RESTORE_NONE_SGRPROJ_PROB, ACCT_STR)) {
-      rsi->restoration_type[rtile_idx] = RESTORE_SGRPROJ;
-      read_sgrproj_filter(&rsi->sgrproj_info[rtile_idx], sgrproj_info, r);
+    if (aom_read_symbol(r, xd->tile_ctx->sgrproj_restore_cdf, 2, ACCT_STR)) {
+      rui->restoration_type = RESTORE_SGRPROJ;
+      read_sgrproj_filter(&rui->sgrproj_info, sgrproj_info, r);
     } else {
-      rsi->restoration_type[rtile_idx] = RESTORE_NONE;
+      rui->restoration_type = RESTORE_NONE;
     }
   }
 }
-#endif  // CONFIG_LOOP_RESTORATION
 
 static void setup_loopfilter(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
+  const int num_planes = av1_num_planes(cm);
   struct loopfilter *lf = &cm->lf;
-#if !CONFIG_LPF_SB
-#if CONFIG_LOOPFILTER_LEVEL
+  if (cm->allow_intrabc || cm->coded_lossless) {
+    // write default deltas to frame buffer
+    av1_set_default_ref_deltas(cm->cur_frame->ref_deltas);
+    av1_set_default_mode_deltas(cm->cur_frame->mode_deltas);
+    return;
+  }
+  assert(!cm->coded_lossless);
+  if (cm->prev_frame) {
+    // write deltas to frame buffer
+    memcpy(lf->ref_deltas, cm->prev_frame->ref_deltas, REF_FRAMES);
+    memcpy(lf->mode_deltas, cm->prev_frame->mode_deltas, MAX_MODE_LF_DELTAS);
+  } else {
+    av1_set_default_ref_deltas(lf->ref_deltas);
+    av1_set_default_mode_deltas(lf->mode_deltas);
+  }
   lf->filter_level[0] = aom_rb_read_literal(rb, 6);
   lf->filter_level[1] = aom_rb_read_literal(rb, 6);
-  if (lf->filter_level[0] || lf->filter_level[1]) {
-    lf->filter_level_u = aom_rb_read_literal(rb, 6);
-    lf->filter_level_v = aom_rb_read_literal(rb, 6);
+  if (num_planes > 1) {
+    if (lf->filter_level[0] || lf->filter_level[1]) {
+      lf->filter_level_u = aom_rb_read_literal(rb, 6);
+      lf->filter_level_v = aom_rb_read_literal(rb, 6);
+    }
   }
-#else
-  lf->filter_level = aom_rb_read_literal(rb, 6);
-#endif
-#endif  // CONFIG_LPF_SB
   lf->sharpness_level = aom_rb_read_literal(rb, 3);
 
   // Read in loop filter deltas applied at the MB level based on mode or ref
@@ -2939,38 +1838,33 @@ static void setup_loopfilter(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
   if (lf->mode_ref_delta_enabled) {
     lf->mode_ref_delta_update = aom_rb_read_bit(rb);
     if (lf->mode_ref_delta_update) {
-      int i;
-
-      for (i = 0; i < TOTAL_REFS_PER_FRAME; i++)
+      for (int i = 0; i < REF_FRAMES; i++)
         if (aom_rb_read_bit(rb))
           lf->ref_deltas[i] = aom_rb_read_inv_signed_literal(rb, 6);
 
-      for (i = 0; i < MAX_MODE_LF_DELTAS; i++)
+      for (int i = 0; i < MAX_MODE_LF_DELTAS; i++)
         if (aom_rb_read_bit(rb))
           lf->mode_deltas[i] = aom_rb_read_inv_signed_literal(rb, 6);
     }
   }
+
+  // write deltas to frame buffer
+  memcpy(cm->cur_frame->ref_deltas, lf->ref_deltas, REF_FRAMES);
+  memcpy(cm->cur_frame->mode_deltas, lf->mode_deltas, MAX_MODE_LF_DELTAS);
 }
 
-#if CONFIG_CDEF
 static void setup_cdef(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
-  int i;
-#if CONFIG_CDEF_SINGLEPASS
+  const int num_planes = av1_num_planes(cm);
+  if (cm->allow_intrabc) return;
   cm->cdef_pri_damping = cm->cdef_sec_damping = aom_rb_read_literal(rb, 2) + 3;
-#else
-  cm->cdef_pri_damping = aom_rb_read_literal(rb, 1) + 5;
-  cm->cdef_sec_damping = aom_rb_read_literal(rb, 2) + 3;
-#endif
   cm->cdef_bits = aom_rb_read_literal(rb, 2);
   cm->nb_cdef_strengths = 1 << cm->cdef_bits;
-  for (i = 0; i < cm->nb_cdef_strengths; i++) {
+  for (int i = 0; i < cm->nb_cdef_strengths; i++) {
     cm->cdef_strengths[i] = aom_rb_read_literal(rb, CDEF_STRENGTH_BITS);
-    cm->cdef_uv_strengths[i] = cm->subsampling_x == cm->subsampling_y
-                                   ? aom_rb_read_literal(rb, CDEF_STRENGTH_BITS)
-                                   : 0;
+    cm->cdef_uv_strengths[i] =
+        num_planes > 1 ? aom_rb_read_literal(rb, CDEF_STRENGTH_BITS) : 0;
   }
 }
-#endif  // CONFIG_CDEF
 
 static INLINE int read_delta_q(struct aom_read_bit_buffer *rb) {
   return aom_rb_read_bit(rb) ? aom_rb_read_inv_signed_literal(rb, 6) : 0;
@@ -2978,66 +1872,74 @@ static INLINE int read_delta_q(struct aom_read_bit_buffer *rb) {
 
 static void setup_quantization(AV1_COMMON *const cm,
                                struct aom_read_bit_buffer *rb) {
+  const int num_planes = av1_num_planes(cm);
   cm->base_qindex = aom_rb_read_literal(rb, QINDEX_BITS);
   cm->y_dc_delta_q = read_delta_q(rb);
-  cm->uv_dc_delta_q = read_delta_q(rb);
-  cm->uv_ac_delta_q = read_delta_q(rb);
+  if (num_planes > 1) {
+    int diff_uv_delta = 0;
+    if (cm->separate_uv_delta_q) diff_uv_delta = aom_rb_read_bit(rb);
+    cm->u_dc_delta_q = read_delta_q(rb);
+    cm->u_ac_delta_q = read_delta_q(rb);
+    if (diff_uv_delta) {
+      cm->v_dc_delta_q = read_delta_q(rb);
+      cm->v_ac_delta_q = read_delta_q(rb);
+    } else {
+      cm->v_dc_delta_q = cm->u_dc_delta_q;
+      cm->v_ac_delta_q = cm->u_ac_delta_q;
+    }
+  }
   cm->dequant_bit_depth = cm->bit_depth;
-#if CONFIG_AOM_QM
   cm->using_qmatrix = aom_rb_read_bit(rb);
   if (cm->using_qmatrix) {
-    cm->min_qmlevel = aom_rb_read_literal(rb, QM_LEVEL_BITS);
-    cm->max_qmlevel = aom_rb_read_literal(rb, QM_LEVEL_BITS);
+    cm->qm_y = aom_rb_read_literal(rb, QM_LEVEL_BITS);
+    cm->qm_u = aom_rb_read_literal(rb, QM_LEVEL_BITS);
+    if (!cm->separate_uv_delta_q)
+      cm->qm_v = cm->qm_u;
+    else
+      cm->qm_v = aom_rb_read_literal(rb, QM_LEVEL_BITS);
   } else {
-    cm->min_qmlevel = 0;
-    cm->max_qmlevel = 0;
+    cm->qm_y = 0;
+    cm->qm_u = 0;
+    cm->qm_v = 0;
   }
-#endif
 }
 
 // Build y/uv dequant values based on segmentation.
 static void setup_segmentation_dequant(AV1_COMMON *const cm) {
-#if CONFIG_AOM_QM
   const int using_qm = cm->using_qmatrix;
-  const int minqm = cm->min_qmlevel;
-  const int maxqm = cm->max_qmlevel;
-#endif
   // When segmentation is disabled, only the first value is used.  The
   // remaining are don't cares.
   const int max_segments = cm->seg.enabled ? MAX_SEGMENTS : 1;
   for (int i = 0; i < max_segments; ++i) {
     const int qindex = av1_get_qindex(&cm->seg, i, cm->base_qindex);
-    cm->y_dequant[i][0] = av1_dc_quant(qindex, cm->y_dc_delta_q, cm->bit_depth);
-    cm->y_dequant[i][1] = av1_ac_quant(qindex, 0, cm->bit_depth);
-    cm->uv_dequant[i][0] =
-        av1_dc_quant(qindex, cm->uv_dc_delta_q, cm->bit_depth);
-    cm->uv_dequant[i][1] =
-        av1_ac_quant(qindex, cm->uv_ac_delta_q, cm->bit_depth);
-#if CONFIG_AOM_QM
+    cm->y_dequant_QTX[i][0] =
+        av1_dc_quant_QTX(qindex, cm->y_dc_delta_q, cm->bit_depth);
+    cm->y_dequant_QTX[i][1] = av1_ac_quant_QTX(qindex, 0, cm->bit_depth);
+    cm->u_dequant_QTX[i][0] =
+        av1_dc_quant_QTX(qindex, cm->u_dc_delta_q, cm->bit_depth);
+    cm->u_dequant_QTX[i][1] =
+        av1_ac_quant_QTX(qindex, cm->u_ac_delta_q, cm->bit_depth);
+    cm->v_dequant_QTX[i][0] =
+        av1_dc_quant_QTX(qindex, cm->v_dc_delta_q, cm->bit_depth);
+    cm->v_dequant_QTX[i][1] =
+        av1_ac_quant_QTX(qindex, cm->v_ac_delta_q, cm->bit_depth);
     const int lossless = qindex == 0 && cm->y_dc_delta_q == 0 &&
-                         cm->uv_dc_delta_q == 0 && cm->uv_ac_delta_q == 0;
+                         cm->u_dc_delta_q == 0 && cm->u_ac_delta_q == 0 &&
+                         cm->v_dc_delta_q == 0 && cm->v_ac_delta_q == 0;
     // NB: depends on base index so there is only 1 set per frame
     // No quant weighting when lossless or signalled not using QM
-    const int qmlevel = (lossless || using_qm == 0)
-                            ? NUM_QM_LEVELS - 1
-                            : aom_get_qmlevel(cm->base_qindex, minqm, maxqm);
+    int qmlevel = (lossless || using_qm == 0) ? NUM_QM_LEVELS - 1 : cm->qm_y;
     for (int j = 0; j < TX_SIZES_ALL; ++j) {
-      cm->y_iqmatrix[i][1][j] = aom_iqmatrix(cm, qmlevel, 0, j, 1);
-      cm->y_iqmatrix[i][0][j] = aom_iqmatrix(cm, qmlevel, 0, j, 0);
-      cm->uv_iqmatrix[i][1][j] = aom_iqmatrix(cm, qmlevel, 1, j, 1);
-      cm->uv_iqmatrix[i][0][j] = aom_iqmatrix(cm, qmlevel, 1, j, 0);
+      cm->y_iqmatrix[i][j] = av1_iqmatrix(cm, qmlevel, AOM_PLANE_Y, j);
     }
-#endif  // CONFIG_AOM_QM
-#if CONFIG_NEW_QUANT
-    for (int dq = 0; dq < QUANT_PROFILES; dq++) {
-      for (int b = 0; b < COEF_BANDS; ++b) {
-        av1_get_dequant_val_nuq(cm->y_dequant[i][b != 0], b,
-                                cm->y_dequant_nuq[i][dq][b], NULL, dq);
-        av1_get_dequant_val_nuq(cm->uv_dequant[i][b != 0], b,
-                                cm->uv_dequant_nuq[i][dq][b], NULL, dq);
-      }
+    qmlevel = (lossless || using_qm == 0) ? NUM_QM_LEVELS - 1 : cm->qm_u;
+    for (int j = 0; j < TX_SIZES_ALL; ++j) {
+      cm->u_iqmatrix[i][j] = av1_iqmatrix(cm, qmlevel, AOM_PLANE_U, j);
+    }
+    qmlevel = (lossless || using_qm == 0) ? NUM_QM_LEVELS - 1 : cm->qm_v;
+    for (int j = 0; j < TX_SIZES_ALL; ++j) {
+      cm->v_iqmatrix[i][j] = av1_iqmatrix(cm, qmlevel, AOM_PLANE_V, j);
     }
-#endif  //  CONFIG_NEW_QUANT
   }
 }
 
@@ -3047,23 +1949,21 @@ static InterpFilter read_frame_interp_filter(struct aom_read_bit_buffer *rb) {
 }
 
 static void setup_render_size(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
-#if CONFIG_FRAME_SUPERRES
   cm->render_width = cm->superres_upscaled_width;
   cm->render_height = cm->superres_upscaled_height;
-#else
-  cm->render_width = cm->width;
-  cm->render_height = cm->height;
-#endif  // CONFIG_FRAME_SUPERRES
   if (aom_rb_read_bit(rb))
-    av1_read_frame_size(rb, &cm->render_width, &cm->render_height);
+    av1_read_frame_size(rb, 16, 16, &cm->render_width, &cm->render_height);
 }
 
-#if CONFIG_FRAME_SUPERRES
 // TODO(afergs): make "struct aom_read_bit_buffer *const rb"?
 static void setup_superres(AV1_COMMON *const cm, struct aom_read_bit_buffer *rb,
                            int *width, int *height) {
   cm->superres_upscaled_width = *width;
   cm->superres_upscaled_height = *height;
+
+  const SequenceHeader *const seq_params = &cm->seq_params;
+  if (!seq_params->enable_superres) return;
+
   if (aom_rb_read_bit(rb)) {
     cm->superres_scale_denominator =
         (uint8_t)aom_rb_read_literal(rb, SUPERRES_SCALE_BITS);
@@ -3077,7 +1977,6 @@ static void setup_superres(AV1_COMMON *const cm, struct aom_read_bit_buffer *rb,
     cm->superres_scale_denominator = SCALE_NUMERATOR;
   }
 }
-#endif  // CONFIG_FRAME_SUPERRES
 
 static void resize_context_buffers(AV1_COMMON *cm, int width, int height) {
 #if CONFIG_SIZE_LIMIT
@@ -3111,24 +2010,34 @@ static void resize_context_buffers(AV1_COMMON *cm, int width, int height) {
   cm->cur_frame->height = cm->height;
 }
 
-static void setup_frame_size(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
+static void setup_frame_size(AV1_COMMON *cm, int frame_size_override_flag,
+                             struct aom_read_bit_buffer *rb) {
   int width, height;
   BufferPool *const pool = cm->buffer_pool;
-  av1_read_frame_size(rb, &width, &height);
-#if CONFIG_FRAME_SUPERRES
+
+  if (frame_size_override_flag) {
+    int num_bits_width = cm->seq_params.num_bits_width;
+    int num_bits_height = cm->seq_params.num_bits_height;
+    av1_read_frame_size(rb, num_bits_width, num_bits_height, &width, &height);
+    if (width > cm->seq_params.max_frame_width ||
+        height > cm->seq_params.max_frame_height) {
+      aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
+                         "Frame dimensions are larger than the maximum values");
+    }
+  } else {
+    width = cm->seq_params.max_frame_width;
+    height = cm->seq_params.max_frame_height;
+  }
+
   setup_superres(cm, rb, &width, &height);
-#endif  // CONFIG_FRAME_SUPERRES
-  setup_render_size(cm, rb);
   resize_context_buffers(cm, width, height);
+  setup_render_size(cm, rb);
 
   lock_buffer_pool(pool);
   if (aom_realloc_frame_buffer(
           get_frame_new_buffer(cm), cm->width, cm->height, cm->subsampling_x,
-          cm->subsampling_y,
-#if CONFIG_HIGHBITDEPTH
-          cm->use_highbitdepth,
-#endif
-          AOM_BORDER_IN_PIXELS, cm->byte_alignment,
+          cm->subsampling_y, cm->use_highbitdepth, AOM_BORDER_IN_PIXELS,
+          cm->byte_alignment,
           &pool->frame_bufs[cm->new_fb_idx].raw_frame_buffer, pool->get_fb_cb,
           pool->cb_priv)) {
     unlock_buffer_pool(pool);
@@ -3140,25 +2049,22 @@ static void setup_frame_size(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
   pool->frame_bufs[cm->new_fb_idx].buf.subsampling_x = cm->subsampling_x;
   pool->frame_bufs[cm->new_fb_idx].buf.subsampling_y = cm->subsampling_y;
   pool->frame_bufs[cm->new_fb_idx].buf.bit_depth = (unsigned int)cm->bit_depth;
-  pool->frame_bufs[cm->new_fb_idx].buf.color_space = cm->color_space;
-#if CONFIG_COLORSPACE_HEADERS
-  pool->frame_bufs[cm->new_fb_idx].buf.transfer_function =
-      cm->transfer_function;
+  pool->frame_bufs[cm->new_fb_idx].buf.color_primaries = cm->color_primaries;
+  pool->frame_bufs[cm->new_fb_idx].buf.transfer_characteristics =
+      cm->transfer_characteristics;
+  pool->frame_bufs[cm->new_fb_idx].buf.matrix_coefficients =
+      cm->matrix_coefficients;
+  pool->frame_bufs[cm->new_fb_idx].buf.monochrome = cm->seq_params.monochrome;
   pool->frame_bufs[cm->new_fb_idx].buf.chroma_sample_position =
       cm->chroma_sample_position;
-#endif
   pool->frame_bufs[cm->new_fb_idx].buf.color_range = cm->color_range;
   pool->frame_bufs[cm->new_fb_idx].buf.render_width = cm->render_width;
   pool->frame_bufs[cm->new_fb_idx].buf.render_height = cm->render_height;
 }
 
-static void setup_sb_size(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
-  (void)rb;
-#if CONFIG_EXT_PARTITION
-  set_sb_size(cm, aom_rb_read_bit(rb) ? BLOCK_128X128 : BLOCK_64X64);
-#else
-  set_sb_size(cm, BLOCK_64X64);
-#endif  // CONFIG_EXT_PARTITION
+static void setup_sb_size(SequenceHeader *seq_params,
+                          struct aom_read_bit_buffer *rb) {
+  set_sb_size(seq_params, aom_rb_read_bit(rb) ? BLOCK_128X128 : BLOCK_64X64);
 }
 
 static INLINE int valid_ref_frame_img_fmt(aom_bit_depth_t ref_bit_depth,
@@ -3172,29 +2078,30 @@ static INLINE int valid_ref_frame_img_fmt(aom_bit_depth_t ref_bit_depth,
 static void setup_frame_size_with_refs(AV1_COMMON *cm,
                                        struct aom_read_bit_buffer *rb) {
   int width, height;
-  int found = 0, i;
+  int found = 0;
   int has_valid_ref_frame = 0;
   BufferPool *const pool = cm->buffer_pool;
-  for (i = 0; i < INTER_REFS_PER_FRAME; ++i) {
+  for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
     if (aom_rb_read_bit(rb)) {
       YV12_BUFFER_CONFIG *const buf = cm->frame_refs[i].buf;
       width = buf->y_crop_width;
       height = buf->y_crop_height;
       cm->render_width = buf->render_width;
       cm->render_height = buf->render_height;
-#if CONFIG_FRAME_SUPERRES
       setup_superres(cm, rb, &width, &height);
-#endif  // CONFIG_FRAME_SUPERRES
+      resize_context_buffers(cm, width, height);
       found = 1;
       break;
     }
   }
 
   if (!found) {
-    av1_read_frame_size(rb, &width, &height);
-#if CONFIG_FRAME_SUPERRES
+    int num_bits_width = cm->seq_params.num_bits_width;
+    int num_bits_height = cm->seq_params.num_bits_height;
+
+    av1_read_frame_size(rb, num_bits_width, num_bits_height, &width, &height);
     setup_superres(cm, rb, &width, &height);
-#endif  // CONFIG_FRAME_SUPERRES
+    resize_context_buffers(cm, width, height);
     setup_render_size(cm, rb);
   }
 
@@ -3204,7 +2111,7 @@ static void setup_frame_size_with_refs(AV1_COMMON *cm,
 
   // Check to make sure at least one of frames that this frame references
   // has valid dimensions.
-  for (i = 0; i < INTER_REFS_PER_FRAME; ++i) {
+  for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
     RefBuffer *const ref_frame = &cm->frame_refs[i];
     has_valid_ref_frame |=
         valid_ref_frame_size(ref_frame->buf->y_crop_width,
@@ -3213,7 +2120,7 @@ static void setup_frame_size_with_refs(AV1_COMMON *cm,
   if (!has_valid_ref_frame)
     aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
                        "Referenced frame has invalid size");
-  for (i = 0; i < INTER_REFS_PER_FRAME; ++i) {
+  for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
     RefBuffer *const ref_frame = &cm->frame_refs[i];
     if (!valid_ref_frame_img_fmt(ref_frame->buf->bit_depth,
                                  ref_frame->buf->subsampling_x,
@@ -3223,16 +2130,11 @@ static void setup_frame_size_with_refs(AV1_COMMON *cm,
                          "Referenced frame has incompatible color format");
   }
 
-  resize_context_buffers(cm, width, height);
-
   lock_buffer_pool(pool);
   if (aom_realloc_frame_buffer(
           get_frame_new_buffer(cm), cm->width, cm->height, cm->subsampling_x,
-          cm->subsampling_y,
-#if CONFIG_HIGHBITDEPTH
-          cm->use_highbitdepth,
-#endif
-          AOM_BORDER_IN_PIXELS, cm->byte_alignment,
+          cm->subsampling_y, cm->use_highbitdepth, AOM_BORDER_IN_PIXELS,
+          cm->byte_alignment,
           &pool->frame_bufs[cm->new_fb_idx].raw_frame_buffer, pool->get_fb_cb,
           pool->cb_priv)) {
     unlock_buffer_pool(pool);
@@ -3244,33 +2146,19 @@ static void setup_frame_size_with_refs(AV1_COMMON *cm,
   pool->frame_bufs[cm->new_fb_idx].buf.subsampling_x = cm->subsampling_x;
   pool->frame_bufs[cm->new_fb_idx].buf.subsampling_y = cm->subsampling_y;
   pool->frame_bufs[cm->new_fb_idx].buf.bit_depth = (unsigned int)cm->bit_depth;
-  pool->frame_bufs[cm->new_fb_idx].buf.color_space = cm->color_space;
-#if CONFIG_COLORSPACE_HEADERS
-  pool->frame_bufs[cm->new_fb_idx].buf.transfer_function =
-      cm->transfer_function;
+  pool->frame_bufs[cm->new_fb_idx].buf.color_primaries = cm->color_primaries;
+  pool->frame_bufs[cm->new_fb_idx].buf.transfer_characteristics =
+      cm->transfer_characteristics;
+  pool->frame_bufs[cm->new_fb_idx].buf.matrix_coefficients =
+      cm->matrix_coefficients;
+  pool->frame_bufs[cm->new_fb_idx].buf.monochrome = cm->seq_params.monochrome;
   pool->frame_bufs[cm->new_fb_idx].buf.chroma_sample_position =
       cm->chroma_sample_position;
-#endif
   pool->frame_bufs[cm->new_fb_idx].buf.color_range = cm->color_range;
   pool->frame_bufs[cm->new_fb_idx].buf.render_width = cm->render_width;
   pool->frame_bufs[cm->new_fb_idx].buf.render_height = cm->render_height;
 }
 
-static void read_tile_group_range(AV1Decoder *pbi,
-                                  struct aom_read_bit_buffer *const rb) {
-  AV1_COMMON *const cm = &pbi->common;
-  const int num_bits = cm->log2_tile_rows + cm->log2_tile_cols;
-  const int num_tiles =
-      cm->tile_rows * cm->tile_cols;  // Note: May be < (1<<num_bits)
-  pbi->tg_start = aom_rb_read_literal(rb, num_bits);
-  pbi->tg_size = 1 + aom_rb_read_literal(rb, num_bits);
-  if (pbi->tg_start + pbi->tg_size > num_tiles)
-    aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
-                       "Tile group extends past last tile in frame");
-}
-
-#if CONFIG_MAX_TILE
-
 // Same function as av1_read_uniform but reading from uncompresses header wb
 static int rb_read_uniform(struct aom_read_bit_buffer *const rb, int n) {
   const int l = get_unsigned_bits(n);
@@ -3285,11 +2173,10 @@ static int rb_read_uniform(struct aom_read_bit_buffer *const rb, int n) {
 
 static void read_tile_info_max_tile(AV1_COMMON *const cm,
                                     struct aom_read_bit_buffer *const rb) {
-  int width_mi = ALIGN_POWER_OF_TWO(cm->mi_cols, MAX_MIB_SIZE_LOG2);
-  int height_mi = ALIGN_POWER_OF_TWO(cm->mi_rows, MAX_MIB_SIZE_LOG2);
-  int width_sb = width_mi >> MAX_MIB_SIZE_LOG2;
-  int height_sb = height_mi >> MAX_MIB_SIZE_LOG2;
-  int start_sb, size_sb, i;
+  int width_mi = ALIGN_POWER_OF_TWO(cm->mi_cols, cm->seq_params.mib_size_log2);
+  int height_mi = ALIGN_POWER_OF_TWO(cm->mi_rows, cm->seq_params.mib_size_log2);
+  int width_sb = width_mi >> cm->seq_params.mib_size_log2;
+  int height_sb = height_mi >> cm->seq_params.mib_size_log2;
 
   av1_get_tile_limits(cm);
   cm->uniform_tile_spacing_flag = aom_rb_read_bit(rb);
@@ -3304,8 +2191,11 @@ static void read_tile_info_max_tile(AV1_COMMON *const cm,
       cm->log2_tile_cols++;
     }
   } else {
+    int i;
+    int start_sb;
     for (i = 0, start_sb = 0; width_sb > 0 && i < MAX_TILE_COLS; i++) {
-      size_sb = 1 + rb_read_uniform(rb, AOMMIN(width_sb, MAX_TILE_WIDTH_SB));
+      const int size_sb =
+          1 + rb_read_uniform(rb, AOMMIN(width_sb, cm->max_tile_width_sb));
       cm->tile_col_start_sb[i] = start_sb;
       start_sb += size_sb;
       width_sb -= size_sb;
@@ -3325,8 +2215,10 @@ static void read_tile_info_max_tile(AV1_COMMON *const cm,
       cm->log2_tile_rows++;
     }
   } else {
+    int i;
+    int start_sb;
     for (i = 0, start_sb = 0; height_sb > 0 && i < MAX_TILE_ROWS; i++) {
-      size_sb =
+      const int size_sb =
           1 + rb_read_uniform(rb, AOMMIN(height_sb, cm->max_tile_height_sb));
       cm->tile_row_start_sb[i] = start_sb;
       start_sb += size_sb;
@@ -3337,110 +2229,61 @@ static void read_tile_info_max_tile(AV1_COMMON *const cm,
   }
   av1_calculate_tile_rows(cm);
 }
-#endif
 
-static void read_tile_info(AV1Decoder *const pbi,
-                           struct aom_read_bit_buffer *const rb) {
-  AV1_COMMON *const cm = &pbi->common;
-#if CONFIG_EXT_TILE
+void av1_set_single_tile_decoding_mode(AV1_COMMON *const cm) {
   cm->single_tile_decoding = 0;
   if (cm->large_scale_tile) {
     struct loopfilter *lf = &cm->lf;
 
     // Figure out single_tile_decoding by loopfilter_level.
-    cm->single_tile_decoding = (!lf->filter_level) ? 1 : 0;
-// Read the tile width/height
-#if CONFIG_EXT_PARTITION
-    if (cm->sb_size == BLOCK_128X128) {
-      cm->tile_width = aom_rb_read_literal(rb, 5) + 1;
-      cm->tile_height = aom_rb_read_literal(rb, 5) + 1;
-    } else {
-#endif  // CONFIG_EXT_PARTITION
-      cm->tile_width = aom_rb_read_literal(rb, 6) + 1;
-      cm->tile_height = aom_rb_read_literal(rb, 6) + 1;
-#if CONFIG_EXT_PARTITION
-    }
-#endif  // CONFIG_EXT_PARTITION
-
-#if CONFIG_LOOPFILTERING_ACROSS_TILES
-    cm->loop_filter_across_tiles_enabled = aom_rb_read_bit(rb);
-#endif  // CONFIG_LOOPFILTERING_ACROSS_TILES
-
-    cm->tile_width <<= cm->mib_size_log2;
-    cm->tile_height <<= cm->mib_size_log2;
-
-    cm->tile_width = AOMMIN(cm->tile_width, cm->mi_cols);
-    cm->tile_height = AOMMIN(cm->tile_height, cm->mi_rows);
-
-    // Get the number of tiles
-    cm->tile_cols = 1;
-    while (cm->tile_cols * cm->tile_width < cm->mi_cols) ++cm->tile_cols;
-
-    cm->tile_rows = 1;
-    while (cm->tile_rows * cm->tile_height < cm->mi_rows) ++cm->tile_rows;
-
-    if (cm->tile_cols * cm->tile_rows > 1) {
-      // Read the number of bytes used to store tile size
-      pbi->tile_col_size_bytes = aom_rb_read_literal(rb, 2) + 1;
-      pbi->tile_size_bytes = aom_rb_read_literal(rb, 2) + 1;
-    }
-
-#if CONFIG_DEPENDENT_HORZTILES
-    cm->dependent_horz_tiles = 0;
-#endif
-  } else {
-#endif  // CONFIG_EXT_TILE
-
-#if CONFIG_MAX_TILE
-    read_tile_info_max_tile(cm, rb);
-#else
-  int min_log2_tile_cols, max_log2_tile_cols, max_ones;
-  av1_get_tile_n_bits(cm->mi_cols, &min_log2_tile_cols, &max_log2_tile_cols);
+    const int no_loopfilter = !(lf->filter_level[0] || lf->filter_level[1]);
+    const int no_cdef = cm->cdef_bits == 0 && cm->cdef_strengths[0] == 0 &&
+                        cm->cdef_uv_strengths[0] == 0;
+    const int no_restoration =
+        cm->rst_info[0].frame_restoration_type == RESTORE_NONE &&
+        cm->rst_info[1].frame_restoration_type == RESTORE_NONE &&
+        cm->rst_info[2].frame_restoration_type == RESTORE_NONE;
+    assert(IMPLIES(cm->coded_lossless, no_loopfilter && no_cdef));
+    assert(IMPLIES(cm->all_lossless, no_restoration));
+    cm->single_tile_decoding = no_loopfilter && no_cdef && no_restoration;
+  }
+}
 
-  // columns
-  max_ones = max_log2_tile_cols - min_log2_tile_cols;
-  cm->log2_tile_cols = min_log2_tile_cols;
-  while (max_ones-- && aom_rb_read_bit(rb)) cm->log2_tile_cols++;
+static void read_tile_info(AV1Decoder *const pbi,
+                           struct aom_read_bit_buffer *const rb) {
+  AV1_COMMON *const cm = &pbi->common;
 
-  if (cm->log2_tile_cols > 6)
-    aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
-                       "Invalid number of tile columns");
+  read_tile_info_max_tile(cm, rb);
 
-  // rows
-  cm->log2_tile_rows = aom_rb_read_bit(rb);
-  if (cm->log2_tile_rows) cm->log2_tile_rows += aom_rb_read_bit(rb);
+  cm->context_update_tile_id = 0;
+  if (cm->tile_rows * cm->tile_cols > 1) {
+    // tile to use for cdf update
+    cm->context_update_tile_id =
+        aom_rb_read_literal(rb, cm->log2_tile_rows + cm->log2_tile_cols);
+    // tile size magnitude
+    pbi->tile_size_bytes = aom_rb_read_literal(rb, 2) + 1;
+  }
+}
 
-  cm->tile_width =
-      get_tile_size(cm->mi_cols, cm->log2_tile_cols, &cm->tile_cols);
-  cm->tile_height =
-      get_tile_size(cm->mi_rows, cm->log2_tile_rows, &cm->tile_rows);
+#if EXT_TILE_DEBUG
+static void read_ext_tile_info(AV1Decoder *const pbi,
+                               struct aom_read_bit_buffer *const rb) {
+  AV1_COMMON *const cm = &pbi->common;
 
-#endif  // CONFIG_MAX_TILE
-#if CONFIG_DEPENDENT_HORZTILES
-    if (cm->tile_rows > 1)
-      cm->dependent_horz_tiles = aom_rb_read_bit(rb);
-    else
-      cm->dependent_horz_tiles = 0;
-#endif
-#if CONFIG_LOOPFILTERING_ACROSS_TILES
-    cm->loop_filter_across_tiles_enabled = aom_rb_read_bit(rb);
-#endif  // CONFIG_LOOPFILTERING_ACROSS_TILES
+  // This information is stored as a separate byte.
+  int mod = rb->bit_offset % CHAR_BIT;
+  if (mod > 0) aom_rb_read_literal(rb, CHAR_BIT - mod);
+  assert(rb->bit_offset % CHAR_BIT == 0);
 
-    // tile size magnitude
+  if (cm->tile_cols * cm->tile_rows > 1) {
+    // Read the number of bytes used to store tile size
+    pbi->tile_col_size_bytes = aom_rb_read_literal(rb, 2) + 1;
     pbi->tile_size_bytes = aom_rb_read_literal(rb, 2) + 1;
-#if CONFIG_EXT_TILE
   }
-#endif  // CONFIG_EXT_TILE
-
-// each tile group header is in its own tile group OBU
-#if !CONFIG_OBU
-  // Store an index to the location of the tile group information
-  pbi->tg_size_bit_offset = rb->bit_offset;
-  read_tile_group_range(pbi, rb);
-#endif
 }
+#endif  // EXT_TILE_DEBUG
 
-static int mem_get_varsize(const uint8_t *src, int sz) {
+static size_t mem_get_varsize(const uint8_t *src, int sz) {
   switch (sz) {
     case 1: return src[0];
     case 2: return mem_get_le16(src);
@@ -3450,14 +2293,14 @@ static int mem_get_varsize(const uint8_t *src, int sz) {
   }
 }
 
-#if CONFIG_EXT_TILE
+#if EXT_TILE_DEBUG
 // Reads the next tile returning its size and adjusting '*data' accordingly
-// based on 'is_last'.
+// based on 'is_last'. On return, '*data' is updated to point to the end of the
+// raw tile buffer in the bit stream.
 static void get_ls_tile_buffer(
     const uint8_t *const data_end, struct aom_internal_error_info *error_info,
-    const uint8_t **data, aom_decrypt_cb decrypt_cb, void *decrypt_state,
-    TileBufferDec (*const tile_buffers)[MAX_TILE_COLS], int tile_size_bytes,
-    int col, int row, int tile_copy_mode) {
+    const uint8_t **data, TileBufferDec (*const tile_buffers)[MAX_TILE_COLS],
+    int tile_size_bytes, int col, int row, int tile_copy_mode) {
   size_t size;
 
   size_t copy_size = 0;
@@ -3466,15 +2309,7 @@ static void get_ls_tile_buffer(
   if (!read_is_valid(*data, tile_size_bytes, data_end))
     aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME,
                        "Truncated packet or corrupt tile length");
-  if (decrypt_cb) {
-    uint8_t be_data[4];
-    decrypt_cb(decrypt_state, *data, be_data, tile_size_bytes);
-
-    // Only read number of bytes in cm->tile_size_bytes.
-    size = mem_get_varsize(be_data, tile_size_bytes);
-  } else {
-    size = mem_get_varsize(*data, tile_size_bytes);
-  }
+  size = mem_get_varsize(*data, tile_size_bytes);
 
   // If tile_copy_mode = 1, then the top bit of the tile header indicates copy
   // mode.
@@ -3486,6 +2321,8 @@ static void get_ls_tile_buffer(
     copy_data = tile_buffers[row - offset][col].data;
     copy_size = tile_buffers[row - offset][col].size;
     size = 0;
+  } else {
+    size += AV1_MIN_TILE_SIZE_BYTES;
   }
 
   *data += tile_size_bytes;
@@ -3503,30 +2340,31 @@ static void get_ls_tile_buffer(
   }
 
   *data += size;
-
-  tile_buffers[row][col].raw_data_end = *data;
 }
 
-static void get_ls_tile_buffers(
+// Returns the end of the last tile buffer
+// (tile_buffers[cm->tile_rows - 1][cm->tile_cols - 1]).
+static const uint8_t *get_ls_tile_buffers(
     AV1Decoder *pbi, const uint8_t *data, const uint8_t *data_end,
     TileBufferDec (*const tile_buffers)[MAX_TILE_COLS]) {
   AV1_COMMON *const cm = &pbi->common;
   const int tile_cols = cm->tile_cols;
   const int tile_rows = cm->tile_rows;
   const int have_tiles = tile_cols * tile_rows > 1;
+  const uint8_t *raw_data_end;  // The end of the last tile buffer
 
   if (!have_tiles) {
     const size_t tile_size = data_end - data;
     tile_buffers[0][0].data = data;
     tile_buffers[0][0].size = tile_size;
-    tile_buffers[0][0].raw_data_end = NULL;
+    raw_data_end = NULL;
   } else {
     // We locate only the tile buffers that are required, which are the ones
     // specified by pbi->dec_tile_col and pbi->dec_tile_row. Also, we always
     // need the last (bottom right) tile buffer, as we need to know where the
     // end of the compressed frame buffer is for proper superframe decoding.
 
-    const uint8_t *tile_col_data_end[MAX_TILE_COLS];
+    const uint8_t *tile_col_data_end[MAX_TILE_COLS] = { NULL };
     const uint8_t *const data_start = data;
 
     const int dec_tile_row = AOMMIN(pbi->dec_tile_row, tile_rows);
@@ -3543,12 +2381,11 @@ static void get_ls_tile_buffers(
     const int tile_copy_mode =
         ((AOMMAX(cm->tile_width, cm->tile_height) << MI_SIZE_LOG2) <= 256) ? 1
                                                                            : 0;
-    size_t tile_col_size;
-    int r, c;
-
     // Read tile column sizes for all columns (we need the last tile buffer)
-    for (c = 0; c < tile_cols; ++c) {
+    for (int c = 0; c < tile_cols; ++c) {
       const int is_last = c == tile_cols - 1;
+      size_t tile_col_size;
+
       if (!is_last) {
         tile_col_size = mem_get_varsize(data, tile_col_size_bytes);
         data += tile_col_size_bytes;
@@ -3563,7 +2400,7 @@ static void get_ls_tile_buffers(
     data = data_start;
 
     // Read the required tile sizes.
-    for (c = tile_cols_start; c < tile_cols_end; ++c) {
+    for (int c = tile_cols_start; c < tile_cols_end; ++c) {
       const int is_last = c == tile_cols - 1;
 
       if (c > 0) data = tile_col_data_end[c - 1];
@@ -3571,40 +2408,45 @@ static void get_ls_tile_buffers(
       if (!is_last) data += tile_col_size_bytes;
 
       // Get the whole of the last column, otherwise stop at the required tile.
-      for (r = 0; r < (is_last ? tile_rows : tile_rows_end); ++r) {
-        tile_buffers[r][c].col = c;
-
+      for (int r = 0; r < (is_last ? tile_rows : tile_rows_end); ++r) {
         get_ls_tile_buffer(tile_col_data_end[c], &pbi->common.error, &data,
-                           pbi->decrypt_cb, pbi->decrypt_state, tile_buffers,
-                           tile_size_bytes, c, r, tile_copy_mode);
+                           tile_buffers, tile_size_bytes, c, r, tile_copy_mode);
       }
     }
 
     // If we have not read the last column, then read it to get the last tile.
     if (tile_cols_end != tile_cols) {
-      c = tile_cols - 1;
+      const int c = tile_cols - 1;
 
       data = tile_col_data_end[c - 1];
 
-      for (r = 0; r < tile_rows; ++r) {
-        tile_buffers[r][c].col = c;
-
+      for (int r = 0; r < tile_rows; ++r) {
         get_ls_tile_buffer(tile_col_data_end[c], &pbi->common.error, &data,
-                           pbi->decrypt_cb, pbi->decrypt_state, tile_buffers,
-                           tile_size_bytes, c, r, tile_copy_mode);
+                           tile_buffers, tile_size_bytes, c, r, tile_copy_mode);
       }
     }
+    raw_data_end = data;
   }
+  return raw_data_end;
+}
+#endif  // EXT_TILE_DEBUG
+
+static const uint8_t *get_ls_single_tile_buffer(
+    AV1Decoder *pbi, const uint8_t *data,
+    TileBufferDec (*const tile_buffers)[MAX_TILE_COLS]) {
+  assert(pbi->dec_tile_row >= 0 && pbi->dec_tile_col >= 0);
+  tile_buffers[pbi->dec_tile_row][pbi->dec_tile_col].data = data;
+  tile_buffers[pbi->dec_tile_row][pbi->dec_tile_col].size =
+      (size_t)pbi->coded_tile_data_size;
+  return data + pbi->coded_tile_data_size;
 }
-#endif  // CONFIG_EXT_TILE
 
 // Reads the next tile returning its size and adjusting '*data' accordingly
 // based on 'is_last'.
 static void get_tile_buffer(const uint8_t *const data_end,
                             const int tile_size_bytes, int is_last,
                             struct aom_internal_error_info *error_info,
-                            const uint8_t **data, aom_decrypt_cb decrypt_cb,
-                            void *decrypt_state, TileBufferDec *const buf) {
+                            const uint8_t **data, TileBufferDec *const buf) {
   size_t size;
 
   if (!is_last) {
@@ -3612,13 +2454,7 @@ static void get_tile_buffer(const uint8_t *const data_end,
       aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME,
                          "Truncated packet or corrupt tile length");
 
-    if (decrypt_cb) {
-      uint8_t be_data[4];
-      decrypt_cb(decrypt_state, *data, be_data, tile_size_bytes);
-      size = mem_get_varsize(be_data, tile_size_bytes);
-    } else {
-      size = mem_get_varsize(*data, tile_size_bytes);
-    }
+    size = mem_get_varsize(*data, tile_size_bytes) + AV1_MIN_TILE_SIZE_BYTES;
     *data += tile_size_bytes;
 
     if (size > (size_t)(data_end - *data))
@@ -3637,140 +2473,123 @@ static void get_tile_buffer(const uint8_t *const data_end,
 static void get_tile_buffers(AV1Decoder *pbi, const uint8_t *data,
                              const uint8_t *data_end,
                              TileBufferDec (*const tile_buffers)[MAX_TILE_COLS],
-                             int startTile, int endTile) {
+                             int start_tile, int end_tile) {
   AV1_COMMON *const cm = &pbi->common;
-  int r, c;
   const int tile_cols = cm->tile_cols;
   const int tile_rows = cm->tile_rows;
   int tc = 0;
   int first_tile_in_tg = 0;
-  struct aom_read_bit_buffer rb_tg_hdr;
-  uint8_t clear_data[MAX_AV1_HEADER_SIZE];
-#if !CONFIG_OBU
-  const size_t hdr_size = pbi->uncomp_hdr_size + pbi->first_partition_size;
-  const int tg_size_bit_offset = pbi->tg_size_bit_offset;
-#else
-  const int tg_size_bit_offset = 0;
-#endif
-
-#if CONFIG_DEPENDENT_HORZTILES
-  int tile_group_start_col = 0;
-  int tile_group_start_row = 0;
-#endif
 
-  for (r = 0; r < tile_rows; ++r) {
-    for (c = 0; c < tile_cols; ++c, ++tc) {
+  for (int r = 0; r < tile_rows; ++r) {
+    for (int c = 0; c < tile_cols; ++c, ++tc) {
       TileBufferDec *const buf = &tile_buffers[r][c];
-#if CONFIG_OBU
-      const int is_last = (tc == endTile);
+
+      const int is_last = (tc == end_tile);
       const size_t hdr_offset = 0;
-#else
-      const int is_last = (r == tile_rows - 1) && (c == tile_cols - 1);
-      const size_t hdr_offset = (tc && tc == first_tile_in_tg) ? hdr_size : 0;
-#endif
 
-      if (tc < startTile || tc > endTile) continue;
+      if (tc < start_tile || tc > end_tile) continue;
 
       if (data + hdr_offset >= data_end)
         aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
                            "Data ended before all tiles were read.");
-      buf->col = c;
-      if (hdr_offset) {
-        init_read_bit_buffer(pbi, &rb_tg_hdr, data, data_end, clear_data);
-        rb_tg_hdr.bit_offset = tg_size_bit_offset;
-        read_tile_group_range(pbi, &rb_tg_hdr);
-#if CONFIG_DEPENDENT_HORZTILES
-        tile_group_start_row = r;
-        tile_group_start_col = c;
-#endif
-      }
       first_tile_in_tg += tc == first_tile_in_tg ? pbi->tg_size : 0;
       data += hdr_offset;
       get_tile_buffer(data_end, pbi->tile_size_bytes, is_last,
-                      &pbi->common.error, &data, pbi->decrypt_cb,
-                      pbi->decrypt_state, buf);
-#if CONFIG_DEPENDENT_HORZTILES
-      cm->tile_group_start_row[r][c] = tile_group_start_row;
-      cm->tile_group_start_col[r][c] = tile_group_start_col;
-#endif
+                      &pbi->common.error, &data, buf);
     }
   }
 }
 
-#if CONFIG_PVQ
-static void daala_dec_init(AV1_COMMON *const cm, daala_dec_ctx *daala_dec,
-                           aom_reader *r) {
-  daala_dec->r = r;
+static void set_cb_buffer(MACROBLOCKD *const xd, CB_BUFFER *cb_buffer,
+                          const int num_planes) {
+  for (int plane = 0; plane < num_planes; ++plane) {
+    xd->plane[plane].dqcoeff_block = cb_buffer->dqcoeff[plane];
+    xd->plane[plane].eob_data = cb_buffer->eob_data[plane];
+    xd->cb_offset[plane] = 0;
+    xd->txb_offset[plane] = 0;
+  }
+  xd->plane[0].color_index_map = cb_buffer->color_index_map[0];
+  xd->plane[1].color_index_map = cb_buffer->color_index_map[1];
+  xd->color_index_map_offset[0] = 0;
+  xd->color_index_map_offset[1] = 0;
+}
 
-  // TODO(yushin) : activity masking info needs be signaled by a bitstream
-  daala_dec->use_activity_masking = AV1_PVQ_ENABLE_ACTIVITY_MASKING;
+static void decode_tile_sb_row(AV1Decoder *pbi, ThreadData *const td,
+                               TileInfo tile_info, const int mi_row) {
+  AV1_COMMON *const cm = &pbi->common;
+  const int num_planes = av1_num_planes(cm);
+  av1_zero_left_context(&td->xd);
 
-  if (daala_dec->use_activity_masking)
-    daala_dec->qm = OD_HVS_QM;
-  else
-    daala_dec->qm = OD_FLAT_QM;
+  for (int mi_col = tile_info.mi_col_start; mi_col < tile_info.mi_col_end;
+       mi_col += cm->seq_params.mib_size) {
+    set_cb_buffer(&td->xd, &td->cb_buffer_base, num_planes);
 
-  od_init_qm(daala_dec->state.qm, daala_dec->state.qm_inv,
-             daala_dec->qm == OD_HVS_QM ? OD_QM8_Q4_HVS : OD_QM8_Q4_FLAT);
+    decode_partition(pbi, &td->xd, mi_row, mi_col, td->bit_reader,
+                     cm->seq_params.sb_size);
+  }
+}
 
-  if (daala_dec->use_activity_masking) {
-    int pli;
-    int use_masking = daala_dec->use_activity_masking;
-    int segment_id = 0;
-    int qindex = av1_get_qindex(&cm->seg, segment_id, cm->base_qindex);
+static int check_trailing_bits_after_symbol_coder(aom_reader *r) {
+  uint32_t nb_bits = aom_reader_tell(r);
+  uint32_t nb_bytes = (nb_bits + 7) >> 3;
 
-    for (pli = 0; pli < MAX_MB_PLANE; pli++) {
-      int i;
-      int q;
+  const uint8_t *p_begin = aom_reader_find_begin(r);
+  const uint8_t *p_end = aom_reader_find_end(r);
 
-      q = qindex;
-      if (q <= OD_DEFAULT_QMS[use_masking][0][pli].interp_q << OD_COEFF_SHIFT) {
-        od_interp_qm(&daala_dec->state.pvq_qm_q4[pli][0], q,
-                     &OD_DEFAULT_QMS[use_masking][0][pli], NULL);
-      } else {
-        i = 0;
-        while (OD_DEFAULT_QMS[use_masking][i + 1][pli].qm_q4 != NULL &&
-               q > OD_DEFAULT_QMS[use_masking][i + 1][pli].interp_q
-                       << OD_COEFF_SHIFT) {
-          i++;
-        }
-        od_interp_qm(&daala_dec->state.pvq_qm_q4[pli][0], q,
-                     &OD_DEFAULT_QMS[use_masking][i][pli],
-                     &OD_DEFAULT_QMS[use_masking][i + 1][pli]);
-      }
-    }
+  // It is legal to have no padding bytes (nb_bytes == p_end - p_begin).
+  if ((ptrdiff_t)nb_bytes > p_end - p_begin) return -1;
+  const uint8_t *p = p_begin + nb_bytes;
+
+  // aom_reader_tell() returns 1 for a newly initialized decoder, and the
+  // return value only increases as values are decoded. So nb_bits > 0, and
+  // thus p > p_begin. Therefore accessing p[-1] is safe.
+  uint8_t last_byte = p[-1];
+  uint8_t pattern = 128 >> ((nb_bits - 1) & 7);
+  if ((last_byte & (2 * pattern - 1)) != pattern) return -1;
+
+  // Make sure that all padding bytes are zero as required by the spec.
+  while (p < p_end) {
+    if (*p != 0) return -1;
+    p++;
   }
+  return 0;
 }
-#endif  // #if CONFIG_PVQ
 
-#if CONFIG_LOOPFILTERING_ACROSS_TILES
-static void dec_setup_across_tile_boundary_info(
-    const AV1_COMMON *const cm, const TileInfo *const tile_info) {
-  if (tile_info->mi_row_start >= tile_info->mi_row_end ||
-      tile_info->mi_col_start >= tile_info->mi_col_end)
-    return;
+static void decode_tile(AV1Decoder *pbi, ThreadData *const td, int tile_row,
+                        int tile_col) {
+  TileInfo tile_info;
+
+  AV1_COMMON *const cm = &pbi->common;
+  const int num_planes = av1_num_planes(cm);
 
-  if (!cm->loop_filter_across_tiles_enabled) {
-    av1_setup_across_tile_boundary_info(cm, tile_info);
+  av1_tile_set_row(&tile_info, cm, tile_row);
+  av1_tile_set_col(&tile_info, cm, tile_col);
+  av1_zero_above_context(cm, tile_info.mi_col_start, tile_info.mi_col_end,
+                         tile_row);
+  av1_reset_loop_restoration(&td->xd, num_planes);
+
+  for (int mi_row = tile_info.mi_row_start; mi_row < tile_info.mi_row_end;
+       mi_row += cm->seq_params.mib_size) {
+    decode_tile_sb_row(pbi, td, tile_info, mi_row);
   }
+
+  int corrupted =
+      (check_trailing_bits_after_symbol_coder(td->bit_reader)) ? 1 : 0;
+  aom_merge_corrupted_flag(&td->xd.corrupted, corrupted);
 }
-#endif  // CONFIG_LOOPFILTERING_ACROSS_TILES
 
 static const uint8_t *decode_tiles(AV1Decoder *pbi, const uint8_t *data,
-                                   const uint8_t *data_end, int startTile,
-                                   int endTile) {
+                                   const uint8_t *data_end, int start_tile,
+                                   int end_tile) {
   AV1_COMMON *const cm = &pbi->common;
-  const AVxWorkerInterface *const winterface = aom_get_worker_interface();
   const int tile_cols = cm->tile_cols;
   const int tile_rows = cm->tile_rows;
   const int n_tiles = tile_cols * tile_rows;
   TileBufferDec(*const tile_buffers)[MAX_TILE_COLS] = pbi->tile_buffers;
-#if CONFIG_EXT_TILE
   const int dec_tile_row = AOMMIN(pbi->dec_tile_row, tile_rows);
   const int single_row = pbi->dec_tile_row >= 0;
   const int dec_tile_col = AOMMIN(pbi->dec_tile_col, tile_cols);
   const int single_col = pbi->dec_tile_col >= 0;
-#endif  // CONFIG_EXT_TILE
   int tile_rows_start;
   int tile_rows_end;
   int tile_cols_start;
@@ -3778,8 +2597,9 @@ static const uint8_t *decode_tiles(AV1Decoder *pbi, const uint8_t *data,
   int inv_col_order;
   int inv_row_order;
   int tile_row, tile_col;
+  uint8_t allow_update_cdf;
+  const uint8_t *raw_data_end = NULL;
 
-#if CONFIG_EXT_TILE
   if (cm->large_scale_tile) {
     tile_rows_start = single_row ? dec_tile_row : 0;
     tile_rows_end = single_row ? dec_tile_row + 1 : tile_rows;
@@ -3787,46 +2607,38 @@ static const uint8_t *decode_tiles(AV1Decoder *pbi, const uint8_t *data,
     tile_cols_end = single_col ? tile_cols_start + 1 : tile_cols;
     inv_col_order = pbi->inv_tile_order && !single_col;
     inv_row_order = pbi->inv_tile_order && !single_row;
+    allow_update_cdf = 0;
   } else {
-#endif  // CONFIG_EXT_TILE
     tile_rows_start = 0;
     tile_rows_end = tile_rows;
     tile_cols_start = 0;
     tile_cols_end = tile_cols;
     inv_col_order = pbi->inv_tile_order;
     inv_row_order = pbi->inv_tile_order;
-#if CONFIG_EXT_TILE
-  }
-#endif  // CONFIG_EXT_TILE
-
-  if (cm->lf.filter_level && !cm->skip_loop_filter &&
-      pbi->lf_worker.data1 == NULL) {
-    CHECK_MEM_ERROR(cm, pbi->lf_worker.data1,
-                    aom_memalign(32, sizeof(LFWorkerData)));
-    pbi->lf_worker.hook = (AVxWorkerHook)av1_loop_filter_worker;
-    if (pbi->max_threads > 1 && !winterface->reset(&pbi->lf_worker)) {
-      aom_internal_error(&cm->error, AOM_CODEC_ERROR,
-                         "Loop filter thread creation failed");
-    }
+    allow_update_cdf = 1;
   }
 
-  if (cm->lf.filter_level && !cm->skip_loop_filter) {
-    LFWorkerData *const lf_data = (LFWorkerData *)pbi->lf_worker.data1;
-    // Be sure to sync as we might be resuming after a failed frame decode.
-    winterface->sync(&pbi->lf_worker);
-    av1_loop_filter_data_reset(lf_data, get_frame_new_buffer(cm), cm,
-                               pbi->mb.plane);
-  }
+  // No tiles to decode.
+  if (tile_rows_end <= tile_rows_start || tile_cols_end <= tile_cols_start ||
+      // First tile is larger than end_tile.
+      tile_rows_start * cm->tile_cols + tile_cols_start > end_tile ||
+      // Last tile is smaller than start_tile.
+      (tile_rows_end - 1) * cm->tile_cols + tile_cols_end - 1 < start_tile)
+    return data;
+
+  allow_update_cdf = allow_update_cdf && !cm->disable_cdf_update;
 
   assert(tile_rows <= MAX_TILE_ROWS);
   assert(tile_cols <= MAX_TILE_COLS);
 
-#if CONFIG_EXT_TILE
-  if (cm->large_scale_tile)
-    get_ls_tile_buffers(pbi, data, data_end, tile_buffers);
+#if EXT_TILE_DEBUG
+  if (cm->large_scale_tile && !pbi->ext_tile_debug)
+    raw_data_end = get_ls_single_tile_buffer(pbi, data, tile_buffers);
+  else if (cm->large_scale_tile && pbi->ext_tile_debug)
+    raw_data_end = get_ls_tile_buffers(pbi, data, data_end, tile_buffers);
   else
-#endif  // CONFIG_EXT_TILE
-    get_tile_buffers(pbi, data, data_end, tile_buffers, startTile, endTile);
+#endif  // EXT_TILE_DEBUG
+    get_tile_buffers(pbi, data, data_end, tile_buffers, start_tile, end_tile);
 
   if (pbi->tile_data == NULL || n_tiles != pbi->allocated_tiles) {
     aom_free(pbi->tile_data);
@@ -3839,536 +2651,411 @@ static const uint8_t *decode_tiles(AV1Decoder *pbi, const uint8_t *data,
     aom_accounting_reset(&pbi->accounting);
   }
 #endif
-  // Load all tile information into tile_data.
+  // Load all tile information into thread_data.
   for (tile_row = tile_rows_start; tile_row < tile_rows_end; ++tile_row) {
+    const int row = inv_row_order ? tile_rows - 1 - tile_row : tile_row;
+
     for (tile_col = tile_cols_start; tile_col < tile_cols_end; ++tile_col) {
-      const TileBufferDec *const buf = &tile_buffers[tile_row][tile_col];
-      TileData *const td = pbi->tile_data + tile_cols * tile_row + tile_col;
+      const int col = inv_col_order ? tile_cols - 1 - tile_col : tile_col;
+      ThreadData *const td = &pbi->td;
+      TileDataDec *const tile_data = pbi->tile_data + row * cm->tile_cols + col;
+      const TileBufferDec *const tile_bs_buf = &tile_buffers[row][col];
 
-      if (tile_row * cm->tile_cols + tile_col < startTile ||
-          tile_row * cm->tile_cols + tile_col > endTile)
+      if (row * cm->tile_cols + col < start_tile ||
+          row * cm->tile_cols + col > end_tile)
         continue;
 
-      td->cm = cm;
       td->xd = pbi->mb;
       td->xd.corrupted = 0;
-      td->xd.counts =
-          cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD
-              ? &cm->counts
-              : NULL;
+      td->xd.mc_buf[0] = pbi->td.mc_buf[0];
+      td->xd.mc_buf[1] = pbi->td.mc_buf[1];
+      td->bit_reader = &tile_data->bit_reader;
       av1_zero(td->dqcoeff);
-#if CONFIG_PVQ
-      av1_zero(td->pvq_ref_coeff);
-#endif
-      av1_tile_init(&td->xd.tile, td->cm, tile_row, tile_col);
-      setup_bool_decoder(buf->data, data_end, buf->size, &cm->error,
-                         &td->bit_reader,
-#if CONFIG_ANS && ANS_MAX_SYMBOLS
-                         1 << cm->ans_window_size_log2,
-#endif  // CONFIG_ANS && ANS_MAX_SYMBOLS
-                         pbi->decrypt_cb, pbi->decrypt_state);
+      av1_tile_init(&td->xd.tile, cm, row, col);
+      setup_bool_decoder(tile_bs_buf->data, data_end, tile_bs_buf->size,
+                         &cm->error, td->bit_reader, allow_update_cdf);
 #if CONFIG_ACCOUNTING
       if (pbi->acct_enabled) {
-        td->bit_reader.accounting = &pbi->accounting;
+        td->bit_reader->accounting = &pbi->accounting;
+        td->bit_reader->accounting->last_tell_frac =
+            aom_reader_tell_frac(td->bit_reader);
       } else {
-        td->bit_reader.accounting = NULL;
+        td->bit_reader->accounting = NULL;
       }
 #endif
-      av1_init_macroblockd(cm, &td->xd,
-#if CONFIG_PVQ
-                           td->pvq_ref_coeff,
-#endif
-#if CONFIG_CFL
-                           &td->cfl,
-#endif
-                           td->dqcoeff);
+      av1_init_macroblockd(cm, &td->xd, td->dqcoeff);
+      av1_init_above_context(cm, &td->xd, row);
 
       // Initialise the tile context from the frame context
-      td->tctx = *cm->fc;
-      td->xd.tile_ctx = &td->tctx;
-
-#if CONFIG_PVQ
-      daala_dec_init(cm, &td->xd.daala_dec, &td->bit_reader);
-      td->xd.daala_dec.state.adapt = &td->tctx.pvq_context;
-#endif
+      tile_data->tctx = *cm->fc;
+      td->xd.tile_ctx = &tile_data->tctx;
 
-      td->xd.plane[0].color_index_map = td->color_index_map[0];
-      td->xd.plane[1].color_index_map = td->color_index_map[1];
-#if CONFIG_MRC_TX
-      td->xd.mrc_mask = td->mrc_mask;
-#endif  // CONFIG_MRC_TX
+      // decode tile
+      decode_tile(pbi, &pbi->td, row, col);
+      aom_merge_corrupted_flag(&pbi->mb.corrupted, td->xd.corrupted);
+      if (pbi->mb.corrupted)
+        aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
+                           "Failed to decode tile data");
     }
   }
 
-  for (tile_row = tile_rows_start; tile_row < tile_rows_end; ++tile_row) {
-    const int row = inv_row_order ? tile_rows - 1 - tile_row : tile_row;
-    int mi_row = 0;
-    TileInfo tile_info;
-
-    av1_tile_set_row(&tile_info, cm, row);
-
-    for (tile_col = tile_cols_start; tile_col < tile_cols_end; ++tile_col) {
-      const int col = inv_col_order ? tile_cols - 1 - tile_col : tile_col;
-      TileData *const td = pbi->tile_data + tile_cols * row + col;
+  if (cm->large_scale_tile) {
+    if (n_tiles == 1) {
+      // Find the end of the single tile buffer
+      return aom_reader_find_end(&pbi->tile_data->bit_reader);
+    }
+    // Return the end of the last tile buffer
+    return raw_data_end;
+  }
+  TileDataDec *const tile_data = pbi->tile_data + end_tile;
 
-      if (tile_row * cm->tile_cols + tile_col < startTile ||
-          tile_row * cm->tile_cols + tile_col > endTile)
-        continue;
+  return aom_reader_find_end(&tile_data->bit_reader);
+}
 
-#if CONFIG_ACCOUNTING
-      if (pbi->acct_enabled) {
-        td->bit_reader.accounting->last_tell_frac =
-            aom_reader_tell_frac(&td->bit_reader);
-      }
-#endif
+static TileJobsDec *get_dec_job_info(AV1DecTileMT *tile_mt_info) {
+  TileJobsDec *cur_job_info = NULL;
+#if CONFIG_MULTITHREAD
+  pthread_mutex_lock(tile_mt_info->job_mutex);
 
-      av1_tile_set_col(&tile_info, cm, col);
+  if (tile_mt_info->jobs_dequeued < tile_mt_info->jobs_enqueued) {
+    cur_job_info = tile_mt_info->job_queue + tile_mt_info->jobs_dequeued;
+    tile_mt_info->jobs_dequeued++;
+  }
 
-#if CONFIG_DEPENDENT_HORZTILES
-      av1_tile_set_tg_boundary(&tile_info, cm, tile_row, tile_col);
-      if (!cm->dependent_horz_tiles || tile_row == 0 ||
-          tile_info.tg_horz_boundary) {
-        av1_zero_above_context(cm, tile_info.mi_col_start,
-                               tile_info.mi_col_end);
-      }
+  pthread_mutex_unlock(tile_mt_info->job_mutex);
 #else
-      av1_zero_above_context(cm, tile_info.mi_col_start, tile_info.mi_col_end);
+  (void)tile_mt_info;
 #endif
-#if CONFIG_LOOP_RESTORATION
-      for (int p = 0; p < MAX_MB_PLANE; ++p) {
-        set_default_wiener(td->xd.wiener_info + p);
-        set_default_sgrproj(td->xd.sgrproj_info + p);
-      }
-#endif  // CONFIG_LOOP_RESTORATION
+  return cur_job_info;
+}
 
-#if CONFIG_LOOPFILTERING_ACROSS_TILES
-      dec_setup_across_tile_boundary_info(cm, &tile_info);
-#endif  // CONFIG_LOOPFILTERING_ACROSS_TILES
+static int tile_worker_hook(void *arg1, void *arg2) {
+  DecWorkerData *const thread_data = (DecWorkerData *)arg1;
+  AV1Decoder *const pbi = (AV1Decoder *)arg2;
+  AV1_COMMON *cm = &pbi->common;
+  ThreadData *const td = thread_data->td;
+  uint8_t allow_update_cdf;
 
-      for (mi_row = tile_info.mi_row_start; mi_row < tile_info.mi_row_end;
-           mi_row += cm->mib_size) {
-        int mi_col;
+  if (setjmp(thread_data->error_info.jmp)) {
+    thread_data->error_info.setjmp = 0;
+    thread_data->td->xd.corrupted = 1;
+    return 0;
+  }
+  allow_update_cdf = cm->large_scale_tile ? 0 : 1;
+  allow_update_cdf = allow_update_cdf && !cm->disable_cdf_update;
 
-        av1_zero_left_context(&td->xd);
+  assert(cm->tile_cols > 0);
+  while (1) {
+    TileJobsDec *cur_job_info = get_dec_job_info(&pbi->tile_mt_info);
 
-        for (mi_col = tile_info.mi_col_start; mi_col < tile_info.mi_col_end;
-             mi_col += cm->mib_size) {
-#if CONFIG_NCOBMC_ADAPT_WEIGHT
-          alloc_ncobmc_pred_buffer(&td->xd);
-          set_sb_mi_boundaries(cm, &td->xd, mi_row, mi_col);
-#endif
-          decode_partition(pbi, &td->xd,
-#if CONFIG_SUPERTX
-                           0,
-#endif  // CONFIG_SUPERTX
-                           mi_row, mi_col, &td->bit_reader, cm->sb_size);
-#if NC_MODE_INFO && CONFIG_MOTION_VAR
-          detoken_and_recon_sb(pbi, &td->xd, mi_row, mi_col, &td->bit_reader,
-                               cm->sb_size);
-#endif
-#if CONFIG_NCOBMC_ADAPT_WEIGHT
-          free_ncobmc_pred_buffer(&td->xd);
-#endif
-        }
-        aom_merge_corrupted_flag(&pbi->mb.corrupted, td->xd.corrupted);
-        if (pbi->mb.corrupted)
-          aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
-                             "Failed to decode tile data");
-      }
-    }
-
-#if !CONFIG_OBU
-    assert(mi_row > 0);
-#endif
+    if (cur_job_info != NULL && !td->xd.corrupted) {
+      const TileBufferDec *const tile_buffer = cur_job_info->tile_buffer;
+      TileDataDec *const tile_data = cur_job_info->tile_data;
+      volatile int tile_row = tile_data->tile_info.tile_row;
+      volatile int tile_col = tile_data->tile_info.tile_col;
 
-// when Parallel deblocking is enabled, deblocking should not
-// be interleaved with decoding. Instead, deblocking should be done
-// after the entire frame is decoded.
-#if !CONFIG_VAR_TX && !CONFIG_PARALLEL_DEBLOCKING && !CONFIG_CB4X4
-    // Loopfilter one tile row.
-    // Note: If out-of-order tile decoding is used(for example, inv_row_order
-    // = 1), the loopfiltering has be done after all tile rows are decoded.
-    if (!inv_row_order && cm->lf.filter_level && !cm->skip_loop_filter) {
-      LFWorkerData *const lf_data = (LFWorkerData *)pbi->lf_worker.data1;
-      const int lf_start = AOMMAX(0, tile_info.mi_row_start - cm->mib_size);
-      const int lf_end = tile_info.mi_row_end - cm->mib_size;
-
-      // Delay the loopfilter if the first tile row is only
-      // a single superblock high.
-      if (lf_end <= 0) continue;
-
-      // Decoding has completed. Finish up the loop filter in this thread.
-      if (tile_info.mi_row_end >= cm->mi_rows) continue;
-
-      winterface->sync(&pbi->lf_worker);
-      lf_data->start = lf_start;
-      lf_data->stop = lf_end;
-      if (pbi->max_threads > 1) {
-        winterface->launch(&pbi->lf_worker);
+      td->xd = pbi->mb;
+      td->xd.corrupted = 0;
+      td->xd.mc_buf[0] = td->mc_buf[0];
+      td->xd.mc_buf[1] = td->mc_buf[1];
+      td->bit_reader = &tile_data->bit_reader;
+      av1_zero(td->dqcoeff);
+      av1_tile_init(&td->xd.tile, cm, tile_row, tile_col);
+      setup_bool_decoder(tile_buffer->data, thread_data->data_end,
+                         tile_buffer->size, &cm->error, td->bit_reader,
+                         allow_update_cdf);
+#if CONFIG_ACCOUNTING
+      if (pbi->acct_enabled) {
+        td->bit_reader->accounting = &pbi->accounting;
+        td->bit_reader->accounting->last_tell_frac =
+            aom_reader_tell_frac(td->bit_reader);
       } else {
-        winterface->execute(&pbi->lf_worker);
+        td->bit_reader->accounting = NULL;
       }
-    }
-#endif  // !CONFIG_VAR_TX && !CONFIG_PARALLEL_DEBLOCKING
-
-    // After loopfiltering, the last 7 row pixels in each superblock row may
-    // still be changed by the longest loopfilter of the next superblock row.
-    if (cm->frame_parallel_decode)
-      av1_frameworker_broadcast(pbi->cur_buf, mi_row << cm->mib_size_log2);
-  }
+#endif
+      av1_init_macroblockd(cm, &td->xd, td->dqcoeff);
+      av1_init_above_context(cm, &td->xd, tile_row);
 
-#if CONFIG_VAR_TX || CONFIG_CB4X4
-// Loopfilter the whole frame.
-#if CONFIG_LPF_SB
-  av1_loop_filter_frame(get_frame_new_buffer(cm), cm, &pbi->mb,
-                        cm->lf.filter_level, 0, 0, 0, 0);
-#else
-#if CONFIG_LOOPFILTER_LEVEL
-  if (cm->lf.filter_level[0] || cm->lf.filter_level[1]) {
-    av1_loop_filter_frame(get_frame_new_buffer(cm), cm, &pbi->mb,
-                          cm->lf.filter_level[0], cm->lf.filter_level[1], 0, 0);
-    av1_loop_filter_frame(get_frame_new_buffer(cm), cm, &pbi->mb,
-                          cm->lf.filter_level_u, cm->lf.filter_level_u, 1, 0);
-    av1_loop_filter_frame(get_frame_new_buffer(cm), cm, &pbi->mb,
-                          cm->lf.filter_level_v, cm->lf.filter_level_v, 2, 0);
-  }
-#else
-#if CONFIG_OBU
-  if (endTile == cm->tile_rows * cm->tile_cols - 1)
+      // Initialise the tile context from the frame context
+      tile_data->tctx = *cm->fc;
+      td->xd.tile_ctx = &tile_data->tctx;
+#if CONFIG_ACCOUNTING
+      if (pbi->acct_enabled) {
+        tile_data->bit_reader.accounting->last_tell_frac =
+            aom_reader_tell_frac(&tile_data->bit_reader);
+      }
 #endif
-    av1_loop_filter_frame(get_frame_new_buffer(cm), cm, &pbi->mb,
-                          cm->lf.filter_level, 0, 0);
-#endif  // CONFIG_LOOPFILTER_LEVEL
-#endif  // CONFIG_LPF_SB
-#else
-#if CONFIG_PARALLEL_DEBLOCKING
-  // Loopfilter all rows in the frame in the frame.
-  if (cm->lf.filter_level && !cm->skip_loop_filter) {
-    LFWorkerData *const lf_data = (LFWorkerData *)pbi->lf_worker.data1;
-    winterface->sync(&pbi->lf_worker);
-    lf_data->start = 0;
-    lf_data->stop = cm->mi_rows;
-    winterface->execute(&pbi->lf_worker);
-  }
-#else
-  // Loopfilter remaining rows in the frame.
-  if (cm->lf.filter_level && !cm->skip_loop_filter) {
-    LFWorkerData *const lf_data = (LFWorkerData *)pbi->lf_worker.data1;
-    winterface->sync(&pbi->lf_worker);
-    lf_data->start = lf_data->stop;
-    lf_data->stop = cm->mi_rows;
-    winterface->execute(&pbi->lf_worker);
-  }
-#endif  // CONFIG_PARALLEL_DEBLOCKING
-#endif  // CONFIG_VAR_TX
-  if (cm->frame_parallel_decode)
-    av1_frameworker_broadcast(pbi->cur_buf, INT_MAX);
-
-#if CONFIG_EXT_TILE
-  if (cm->large_scale_tile) {
-    if (n_tiles == 1) {
-#if CONFIG_ANS
-      return data_end;
-#else
-      // Find the end of the single tile buffer
-      return aom_reader_find_end(&pbi->tile_data->bit_reader);
-#endif  // CONFIG_ANS
+      // decode tile
+      decode_tile(pbi, td, tile_row, tile_col);
     } else {
-      // Return the end of the last tile buffer
-      return tile_buffers[tile_rows - 1][tile_cols - 1].raw_data_end;
+      break;
     }
-  } else {
-#endif  // CONFIG_EXT_TILE
-#if CONFIG_ANS
-    return data_end;
-#else
-#if !CONFIG_OBU
-  {
-    // Get last tile data.
-    TileData *const td = pbi->tile_data + tile_cols * tile_rows - 1;
-    return aom_reader_find_end(&td->bit_reader);
-  }
-#else
-  TileData *const td = pbi->tile_data + endTile;
-  return aom_reader_find_end(&td->bit_reader);
-#endif
-#endif  // CONFIG_ANS
-#if CONFIG_EXT_TILE
   }
-#endif  // CONFIG_EXT_TILE
+  return !td->xd.corrupted;
 }
 
-static int tile_worker_hook(TileWorkerData *const tile_data,
-                            const TileInfo *const tile) {
-  AV1Decoder *const pbi = tile_data->pbi;
-  const AV1_COMMON *const cm = &pbi->common;
-  int mi_row, mi_col;
+// sorts in descending order
+static int compare_tile_buffers(const void *a, const void *b) {
+  const TileJobsDec *const buf1 = (const TileJobsDec *)a;
+  const TileJobsDec *const buf2 = (const TileJobsDec *)b;
+  return (((int)buf2->tile_buffer->size) - ((int)buf1->tile_buffer->size));
+}
 
-  if (setjmp(tile_data->error_info.jmp)) {
-    tile_data->error_info.setjmp = 0;
-    aom_merge_corrupted_flag(&tile_data->xd.corrupted, 1);
-    return 0;
+static void enqueue_tile_jobs(AV1Decoder *pbi, AV1_COMMON *cm,
+                              int tile_rows_start, int tile_rows_end,
+                              int tile_cols_start, int tile_cols_end,
+                              int startTile, int endTile) {
+  AV1DecTileMT *tile_mt_info = &pbi->tile_mt_info;
+  TileJobsDec *tile_job_queue = tile_mt_info->job_queue;
+  tile_mt_info->jobs_enqueued = 0;
+  tile_mt_info->jobs_dequeued = 0;
+
+  for (int row = tile_rows_start; row < tile_rows_end; row++) {
+    for (int col = tile_cols_start; col < tile_cols_end; col++) {
+      if (row * cm->tile_cols + col < startTile ||
+          row * cm->tile_cols + col > endTile)
+        continue;
+      tile_job_queue->tile_buffer = &pbi->tile_buffers[row][col];
+      tile_job_queue->tile_data = pbi->tile_data + row * cm->tile_cols + col;
+      tile_job_queue++;
+      tile_mt_info->jobs_enqueued++;
+    }
   }
+}
+
+static void alloc_dec_jobs(AV1DecTileMT *tile_mt_info, AV1_COMMON *cm,
+                           int tile_rows, int tile_cols) {
+  tile_mt_info->alloc_tile_rows = tile_rows;
+  tile_mt_info->alloc_tile_cols = tile_cols;
+  int num_tiles = tile_rows * tile_cols;
+#if CONFIG_MULTITHREAD
+  {
+    CHECK_MEM_ERROR(cm, tile_mt_info->job_mutex,
+                    aom_malloc(sizeof(*tile_mt_info->job_mutex) * num_tiles));
 
-  tile_data->error_info.setjmp = 1;
-  tile_data->xd.error_info = &tile_data->error_info;
-#if CONFIG_DEPENDENT_HORZTILES
-  if (!cm->dependent_horz_tiles || tile->tg_horz_boundary) {
-    av1_zero_above_context(&pbi->common, tile->mi_col_start, tile->mi_col_end);
+    for (int i = 0; i < num_tiles; i++) {
+      pthread_mutex_init(&tile_mt_info->job_mutex[i], NULL);
+    }
   }
-#else
-  av1_zero_above_context(&pbi->common, tile->mi_col_start, tile->mi_col_end);
 #endif
+  CHECK_MEM_ERROR(cm, tile_mt_info->job_queue,
+                  aom_malloc(sizeof(*tile_mt_info->job_queue) * num_tiles));
+}
 
-  for (mi_row = tile->mi_row_start; mi_row < tile->mi_row_end;
-       mi_row += cm->mib_size) {
-    av1_zero_left_context(&tile_data->xd);
-
-    for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end;
-         mi_col += cm->mib_size) {
-      decode_partition(pbi, &tile_data->xd,
-#if CONFIG_SUPERTX
-                       0,
-#endif
-                       mi_row, mi_col, &tile_data->bit_reader, cm->sb_size);
-#if NC_MODE_INFO && CONFIG_MOTION_VAR
-      detoken_and_recon_sb(pbi, &tile_data->xd, mi_row, mi_col,
-                           &tile_data->bit_reader, cm->sb_size);
-#endif
-    }
+void av1_free_mc_tmp_buf(void *td, int use_highbd) {
+  ThreadData *thread_data = (ThreadData *)td;
+  int ref;
+  for (ref = 0; ref < 2; ref++) {
+    if (use_highbd)
+      aom_free(CONVERT_TO_SHORTPTR(thread_data->mc_buf[ref]));
+    else
+      aom_free(thread_data->mc_buf[ref]);
+    thread_data->mc_buf[ref] = NULL;
   }
-  return !tile_data->xd.corrupted;
+  thread_data->mc_buf_size = 0;
 }
 
-// sorts in descending order
-static int compare_tile_buffers(const void *a, const void *b) {
-  const TileBufferDec *const buf1 = (const TileBufferDec *)a;
-  const TileBufferDec *const buf2 = (const TileBufferDec *)b;
-  return (int)(buf2->size - buf1->size);
+static void allocate_mc_tmp_buf(AV1_COMMON *const cm, void *td, int buf_size,
+                                int use_highbd) {
+  ThreadData *thread_data = (ThreadData *)td;
+
+  for (int ref = 0; ref < 2; ref++) {
+    if (use_highbd) {
+      uint16_t *hbd_mc_buf;
+      CHECK_MEM_ERROR(cm, hbd_mc_buf, (uint16_t *)aom_memalign(16, buf_size));
+      thread_data->mc_buf[ref] = CONVERT_TO_BYTEPTR(hbd_mc_buf);
+    } else {
+      CHECK_MEM_ERROR(cm, thread_data->mc_buf[ref],
+                      (uint8_t *)aom_memalign(16, buf_size));
+    }
+  }
+  thread_data->mc_buf_size = buf_size;
 }
 
 static const uint8_t *decode_tiles_mt(AV1Decoder *pbi, const uint8_t *data,
-                                      const uint8_t *data_end) {
+                                      const uint8_t *data_end, int start_tile,
+                                      int end_tile) {
   AV1_COMMON *const cm = &pbi->common;
   const AVxWorkerInterface *const winterface = aom_get_worker_interface();
   const int tile_cols = cm->tile_cols;
   const int tile_rows = cm->tile_rows;
-  const int num_workers = AOMMIN(pbi->max_threads & ~1, tile_cols);
+  const int n_tiles = tile_cols * tile_rows;
   TileBufferDec(*const tile_buffers)[MAX_TILE_COLS] = pbi->tile_buffers;
-#if CONFIG_EXT_TILE
   const int dec_tile_row = AOMMIN(pbi->dec_tile_row, tile_rows);
   const int single_row = pbi->dec_tile_row >= 0;
   const int dec_tile_col = AOMMIN(pbi->dec_tile_col, tile_cols);
   const int single_col = pbi->dec_tile_col >= 0;
-#endif  // CONFIG_EXT_TILE
   int tile_rows_start;
   int tile_rows_end;
   int tile_cols_start;
   int tile_cols_end;
-  int tile_row, tile_col;
-  int i;
+  int tile_count_tg;
+  int num_workers;
+  int worker_idx;
+  const uint8_t *raw_data_end = NULL;
 
-#if CONFIG_EXT_TILE
   if (cm->large_scale_tile) {
     tile_rows_start = single_row ? dec_tile_row : 0;
     tile_rows_end = single_row ? dec_tile_row + 1 : tile_rows;
     tile_cols_start = single_col ? dec_tile_col : 0;
     tile_cols_end = single_col ? tile_cols_start + 1 : tile_cols;
   } else {
-#endif  // CONFIG_EXT_TILE
     tile_rows_start = 0;
     tile_rows_end = tile_rows;
     tile_cols_start = 0;
     tile_cols_end = tile_cols;
-#if CONFIG_EXT_TILE
   }
-#endif  // CONFIG_EXT_TILE
+  tile_count_tg = end_tile - start_tile + 1;
+  num_workers = AOMMIN(pbi->max_threads, tile_count_tg);
 
-#if !CONFIG_ANS
-  int final_worker = -1;
-#endif  // !CONFIG_ANS
+  // No tiles to decode.
+  if (tile_rows_end <= tile_rows_start || tile_cols_end <= tile_cols_start ||
+      // First tile is larger than end_tile.
+      tile_rows_start * tile_cols + tile_cols_start > end_tile ||
+      // Last tile is smaller than start_tile.
+      (tile_rows_end - 1) * tile_cols + tile_cols_end - 1 < start_tile)
+    return data;
 
   assert(tile_rows <= MAX_TILE_ROWS);
   assert(tile_cols <= MAX_TILE_COLS);
-
-  assert(tile_cols * tile_rows > 1);
-
-  // TODO(jzern): See if we can remove the restriction of passing in max
-  // threads to the decoder.
-  if (pbi->num_tile_workers == 0) {
-    const int num_threads = pbi->max_threads & ~1;
+  assert(tile_count_tg > 0);
+  assert(num_workers > 0);
+  assert(start_tile <= end_tile);
+  assert(start_tile >= 0 && end_tile < n_tiles);
+
+  // Create workers and thread_data
+  if (pbi->num_workers == 0) {
+    const int num_threads = pbi->max_threads;
     CHECK_MEM_ERROR(cm, pbi->tile_workers,
                     aom_malloc(num_threads * sizeof(*pbi->tile_workers)));
-    // Ensure tile data offsets will be properly aligned. This may fail on
-    // platforms without DECLARE_ALIGNED().
-    assert((sizeof(*pbi->tile_worker_data) % 16) == 0);
-    CHECK_MEM_ERROR(
-        cm, pbi->tile_worker_data,
-        aom_memalign(32, num_threads * sizeof(*pbi->tile_worker_data)));
-    CHECK_MEM_ERROR(cm, pbi->tile_worker_info,
-                    aom_malloc(num_threads * sizeof(*pbi->tile_worker_info)));
-    for (i = 0; i < num_threads; ++i) {
-      AVxWorker *const worker = &pbi->tile_workers[i];
-      ++pbi->num_tile_workers;
+    CHECK_MEM_ERROR(cm, pbi->thread_data,
+                    aom_malloc(num_threads * sizeof(*pbi->thread_data)));
+
+    for (worker_idx = 0; worker_idx < num_threads; ++worker_idx) {
+      AVxWorker *const worker = &pbi->tile_workers[worker_idx];
+      DecWorkerData *const thread_data = pbi->thread_data + worker_idx;
+      ++pbi->num_workers;
 
       winterface->init(worker);
-      if (i < num_threads - 1 && !winterface->reset(worker)) {
+      if (worker_idx < num_threads - 1 && !winterface->reset(worker)) {
         aom_internal_error(&cm->error, AOM_CODEC_ERROR,
                            "Tile decoder thread creation failed");
       }
-    }
-  }
 
-  // Reset tile decoding hook
-  for (i = 0; i < num_workers; ++i) {
-    AVxWorker *const worker = &pbi->tile_workers[i];
-    winterface->sync(worker);
-    worker->hook = (AVxWorkerHook)tile_worker_hook;
-    worker->data1 = &pbi->tile_worker_data[i];
-    worker->data2 = &pbi->tile_worker_info[i];
+      if (worker_idx < num_threads - 1) {
+        // Allocate thread data.
+        CHECK_MEM_ERROR(cm, thread_data->td,
+                        aom_memalign(32, sizeof(*thread_data->td)));
+        av1_zero(*thread_data->td);
+      } else {
+        // Main thread acts as a worker and uses the thread data in pbi
+        thread_data->td = &pbi->td;
+      }
+    }
   }
-
-  // Initialize thread frame counts.
-  if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) {
-    for (i = 0; i < num_workers; ++i) {
-      TileWorkerData *const twd = (TileWorkerData *)pbi->tile_workers[i].data1;
-      av1_zero(twd->counts);
+  const int use_highbd = cm->use_highbitdepth ? 1 : 0;
+  const int buf_size = MC_TEMP_BUF_PELS << use_highbd;
+  for (worker_idx = 0; worker_idx < pbi->max_threads - 1; ++worker_idx) {
+    DecWorkerData *const thread_data = pbi->thread_data + worker_idx;
+    if (thread_data->td->mc_buf_size != buf_size) {
+      av1_free_mc_tmp_buf(thread_data->td, use_highbd);
+      allocate_mc_tmp_buf(cm, thread_data->td, buf_size, use_highbd);
     }
   }
 
-// Load tile data into tile_buffers
-#if CONFIG_EXT_TILE
+    // get tile size in tile group
+#if EXT_TILE_DEBUG
   if (cm->large_scale_tile)
-    get_ls_tile_buffers(pbi, data, data_end, tile_buffers);
+    raw_data_end = get_ls_tile_buffers(pbi, data, data_end, tile_buffers);
   else
-#endif  // CONFIG_EXT_TILE
-    get_tile_buffers(pbi, data, data_end, tile_buffers, 0,
-                     cm->tile_rows * cm->tile_cols - 1);
+#endif  // EXT_TILE_DEBUG
+    get_tile_buffers(pbi, data, data_end, tile_buffers, start_tile, end_tile);
 
-  for (tile_row = tile_rows_start; tile_row < tile_rows_end; ++tile_row) {
-    // Sort the buffers in this tile row based on size in descending order.
-    qsort(&tile_buffers[tile_row][tile_cols_start],
-          tile_cols_end - tile_cols_start, sizeof(tile_buffers[0][0]),
-          compare_tile_buffers);
-
-    // Rearrange the tile buffers in this tile row such that per-tile group
-    // the largest, and presumably the most difficult tile will be decoded in
-    // the main thread. This should help minimize the number of instances
-    // where the main thread is waiting for a worker to complete.
-    {
-      int group_start;
-      for (group_start = tile_cols_start; group_start < tile_cols_end;
-           group_start += num_workers) {
-        const int group_end = AOMMIN(group_start + num_workers, tile_cols);
-        const TileBufferDec largest = tile_buffers[tile_row][group_start];
-        memmove(&tile_buffers[tile_row][group_start],
-                &tile_buffers[tile_row][group_start + 1],
-                (group_end - group_start - 1) * sizeof(tile_buffers[0][0]));
-        tile_buffers[tile_row][group_end - 1] = largest;
-      }
-    }
+  if (pbi->tile_data == NULL || n_tiles != pbi->allocated_tiles) {
+    aom_free(pbi->tile_data);
+    CHECK_MEM_ERROR(cm, pbi->tile_data,
+                    aom_memalign(32, n_tiles * sizeof(*pbi->tile_data)));
+    pbi->allocated_tiles = n_tiles;
+  }
 
-    for (tile_col = tile_cols_start; tile_col < tile_cols_end;) {
-      // Launch workers for individual columns
-      for (i = 0; i < num_workers && tile_col < tile_cols_end;
-           ++i, ++tile_col) {
-        TileBufferDec *const buf = &tile_buffers[tile_row][tile_col];
-        AVxWorker *const worker = &pbi->tile_workers[i];
-        TileWorkerData *const twd = (TileWorkerData *)worker->data1;
-        TileInfo *const tile_info = (TileInfo *)worker->data2;
-
-        twd->pbi = pbi;
-        twd->xd = pbi->mb;
-        twd->xd.corrupted = 0;
-        twd->xd.counts =
-            cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD
-                ? &twd->counts
-                : NULL;
-        av1_zero(twd->dqcoeff);
-        av1_tile_init(tile_info, cm, tile_row, buf->col);
-        av1_tile_init(&twd->xd.tile, cm, tile_row, buf->col);
-
-#if CONFIG_LOOPFILTERING_ACROSS_TILES
-        dec_setup_across_tile_boundary_info(cm, tile_info);
-#endif  // CONFIG_LOOPFILTERING_ACROSS_TILES
-
-        setup_bool_decoder(buf->data, data_end, buf->size, &cm->error,
-                           &twd->bit_reader,
-#if CONFIG_ANS && ANS_MAX_SYMBOLS
-                           1 << cm->ans_window_size_log2,
-#endif  // CONFIG_ANS && ANS_MAX_SYMBOLS
-                           pbi->decrypt_cb, pbi->decrypt_state);
-        av1_init_macroblockd(cm, &twd->xd,
-#if CONFIG_PVQ
-                             twd->pvq_ref_coeff,
-#endif
-#if CONFIG_CFL
-                             &twd->cfl,
-#endif
-                             twd->dqcoeff);
-#if CONFIG_PVQ
-        daala_dec_init(cm, &twd->xd.daala_dec, &twd->bit_reader);
-        twd->xd.daala_dec.state.adapt = &twd->tctx.pvq_context;
+  // Reset tile decoding hook
+  for (worker_idx = 0; worker_idx < num_workers; ++worker_idx) {
+    AVxWorker *const worker = &pbi->tile_workers[worker_idx];
+    DecWorkerData *const thread_data = pbi->thread_data + worker_idx;
+    winterface->sync(worker);
+
+    worker->hook = tile_worker_hook;
+    worker->data1 = thread_data;
+    worker->data2 = pbi;
+  }
+#if CONFIG_ACCOUNTING
+  if (pbi->acct_enabled) {
+    aom_accounting_reset(&pbi->accounting);
+  }
 #endif
-        // Initialise the tile context from the frame context
-        twd->tctx = *cm->fc;
-        twd->xd.tile_ctx = &twd->tctx;
-        twd->xd.plane[0].color_index_map = twd->color_index_map[0];
-        twd->xd.plane[1].color_index_map = twd->color_index_map[1];
-
-        worker->had_error = 0;
-        if (i == num_workers - 1 || tile_col == tile_cols_end - 1) {
-          winterface->execute(worker);
-        } else {
-          winterface->launch(worker);
-        }
+  for (int row = 0; row < tile_rows; row++) {
+    for (int col = 0; col < tile_cols; col++) {
+      TileDataDec *tile_data = pbi->tile_data + row * cm->tile_cols + col;
+      av1_tile_init(&tile_data->tile_info, cm, row, col);
+    }
+  }
 
-#if !CONFIG_ANS
-        if (tile_row == tile_rows - 1 && buf->col == tile_cols - 1) {
-          final_worker = i;
-        }
-#endif  // !CONFIG_ANS
-      }
+  if (pbi->tile_mt_info.alloc_tile_cols != tile_cols ||
+      pbi->tile_mt_info.alloc_tile_rows != tile_rows) {
+    av1_dealloc_dec_jobs(&pbi->tile_mt_info);
+    alloc_dec_jobs(&pbi->tile_mt_info, cm, tile_rows, tile_cols);
+  }
+  enqueue_tile_jobs(pbi, cm, tile_rows_start, tile_rows_end, tile_cols_start,
+                    tile_cols_end, start_tile, end_tile);
+  qsort(pbi->tile_mt_info.job_queue, pbi->tile_mt_info.jobs_enqueued,
+        sizeof(pbi->tile_mt_info.job_queue[0]), compare_tile_buffers);
 
-      // Sync all workers
-      for (; i > 0; --i) {
-        AVxWorker *const worker = &pbi->tile_workers[i - 1];
-        // TODO(jzern): The tile may have specific error data associated with
-        // its aom_internal_error_info which could be propagated to the main
-        // info in cm. Additionally once the threads have been synced and an
-        // error is detected, there's no point in continuing to decode tiles.
-        pbi->mb.corrupted |= !winterface->sync(worker);
+  {
+    const int base = tile_count_tg / num_workers;
+    const int remain = tile_count_tg % num_workers;
+    int tile_start = start_tile;
+    int corrupted = 0;
+
+    for (worker_idx = 0; worker_idx < num_workers; ++worker_idx) {
+      // compute number of tiles assign to each worker
+      const int count = base + (remain + worker_idx) / num_workers;
+      AVxWorker *const worker = &pbi->tile_workers[worker_idx];
+      DecWorkerData *const thread_data = (DecWorkerData *)worker->data1;
+
+      thread_data->data_end = data_end;
+      tile_start += count;
+
+      worker->had_error = 0;
+      if (worker_idx == num_workers - 1) {
+        winterface->execute(worker);
+      } else {
+        winterface->launch(worker);
       }
     }
-  }
 
-  // Accumulate thread frame counts.
-  if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) {
-    for (i = 0; i < num_workers; ++i) {
-      TileWorkerData *const twd = (TileWorkerData *)pbi->tile_workers[i].data1;
-      av1_accumulate_frame_counts(&cm->counts, &twd->counts);
+    for (; worker_idx > 0; --worker_idx) {
+      AVxWorker *const worker = &pbi->tile_workers[worker_idx - 1];
+      aom_merge_corrupted_flag(&corrupted, !winterface->sync(worker));
     }
+
+    pbi->mb.corrupted = corrupted;
   }
 
-#if CONFIG_EXT_TILE
+  if (pbi->mb.corrupted)
+    aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
+                       "Failed to decode tile data");
+
   if (cm->large_scale_tile) {
+    if (n_tiles == 1) {
+      // Find the end of the single tile buffer
+      return aom_reader_find_end(&pbi->tile_data->bit_reader);
+    }
     // Return the end of the last tile buffer
-    return tile_buffers[tile_rows - 1][tile_cols - 1].raw_data_end;
-  } else {
-#endif  // CONFIG_EXT_TILE
-#if CONFIG_ANS
-    return data_end;
-#else
-  assert(final_worker != -1);
-  {
-    TileWorkerData *const twd =
-        (TileWorkerData *)pbi->tile_workers[final_worker].data1;
-    return aom_reader_find_end(&twd->bit_reader);
-  }
-#endif  // CONFIG_ANS
-#if CONFIG_EXT_TILE
+    return raw_data_end;
   }
-#endif  // CONFIG_EXT_TILE
+  TileDataDec *const tile_data = pbi->tile_data + end_tile;
+
+  return aom_reader_find_end(&tile_data->bit_reader);
 }
 
 static void error_handler(void *data) {
@@ -4376,217 +3063,462 @@ static void error_handler(void *data) {
   aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, "Truncated packet");
 }
 
-static void read_bitdepth_colorspace_sampling(AV1_COMMON *cm,
-                                              struct aom_read_bit_buffer *rb,
-                                              int allow_lowbitdepth) {
-  if (cm->profile >= PROFILE_2) {
-    cm->bit_depth = aom_rb_read_bit(rb) ? AOM_BITS_12 : AOM_BITS_10;
+// Reads the high_bitdepth and twelve_bit fields in color_config() and sets
+// cm->bit_depth based on the values of those fields and cm->profile. Reports
+// errors by calling rb->error_handler() or aom_internal_error().
+static void av1_read_bitdepth(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
+  const int high_bitdepth = aom_rb_read_bit(rb);
+  if (cm->profile == PROFILE_2 && high_bitdepth) {
+    const int twelve_bit = aom_rb_read_bit(rb);
+    cm->bit_depth = twelve_bit ? AOM_BITS_12 : AOM_BITS_10;
+  } else if (cm->profile <= PROFILE_2) {
+    cm->bit_depth = high_bitdepth ? AOM_BITS_10 : AOM_BITS_8;
   } else {
-    cm->bit_depth = AOM_BITS_8;
+    aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+                       "Unsupported profile/bit-depth combination");
   }
+}
 
-#if CONFIG_HIGHBITDEPTH
-  cm->use_highbitdepth = cm->bit_depth > AOM_BITS_8 || !allow_lowbitdepth;
-#else
-  (void)allow_lowbitdepth;
-#endif
-#if CONFIG_COLORSPACE_HEADERS
-  cm->color_space = aom_rb_read_literal(rb, 5);
-  cm->transfer_function = aom_rb_read_literal(rb, 5);
-#else
-  cm->color_space = aom_rb_read_literal(rb, 3);
-#endif
-  if (cm->color_space != AOM_CS_SRGB) {
-    // [16,235] (including xvycc) vs [0,255] range
-    cm->color_range = aom_rb_read_bit(rb);
-    if (cm->profile == PROFILE_1 || cm->profile == PROFILE_3) {
-      cm->subsampling_x = aom_rb_read_bit(rb);
-      cm->subsampling_y = aom_rb_read_bit(rb);
-      if (cm->subsampling_x == 1 && cm->subsampling_y == 1)
-        aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
-                           "4:2:0 color not supported in profile 1 or 3");
-      if (aom_rb_read_bit(rb))
-        aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
-                           "Reserved bit set");
-    } else {
-      cm->subsampling_y = cm->subsampling_x = 1;
+void av1_read_film_grain_params(AV1_COMMON *cm,
+                                struct aom_read_bit_buffer *rb) {
+  aom_film_grain_t *pars = &cm->film_grain_params;
+
+  pars->apply_grain = aom_rb_read_bit(rb);
+  if (!pars->apply_grain) {
+    memset(pars, 0, sizeof(*pars));
+    return;
+  }
+
+  pars->random_seed = aom_rb_read_literal(rb, 16);
+  if (cm->frame_type == INTER_FRAME)
+    pars->update_parameters = aom_rb_read_bit(rb);
+  else
+    pars->update_parameters = 1;
+
+  if (!pars->update_parameters) {
+    // inherit parameters from a previous reference frame
+    RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs;
+    int film_grain_params_ref_idx = aom_rb_read_literal(rb, 3);
+    int buf_idx = cm->ref_frame_map[film_grain_params_ref_idx];
+    if (buf_idx == INVALID_IDX) {
+      aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+                         "Invalid Film grain reference idx");
     }
-#if CONFIG_COLORSPACE_HEADERS
-    if (cm->subsampling_x == 1 && cm->subsampling_y == 1) {
-      cm->chroma_sample_position = aom_rb_read_literal(rb, 2);
+    if (!frame_bufs[buf_idx].film_grain_params_present) {
+      aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+                         "Film grain reference parameters not available");
     }
-#endif
+    uint16_t random_seed = pars->random_seed;
+    *pars = frame_bufs[buf_idx].film_grain_params;  // inherit paramaters
+    pars->random_seed = random_seed;                // with new random seed
+    return;
+  }
+
+  // Scaling functions parameters
+  pars->num_y_points = aom_rb_read_literal(rb, 4);  // max 14
+  if (pars->num_y_points > 14)
+    aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+                       "Number of points for film grain luma scaling function "
+                       "exceeds the maximum value.");
+  for (int i = 0; i < pars->num_y_points; i++) {
+    pars->scaling_points_y[i][0] = aom_rb_read_literal(rb, 8);
+    if (i && pars->scaling_points_y[i - 1][0] >= pars->scaling_points_y[i][0])
+      aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+                         "First coordinate of the scaling function points "
+                         "shall be increasing.");
+    pars->scaling_points_y[i][1] = aom_rb_read_literal(rb, 8);
+  }
+
+  if (!cm->seq_params.monochrome)
+    pars->chroma_scaling_from_luma = aom_rb_read_bit(rb);
+
+  if (cm->seq_params.monochrome || pars->chroma_scaling_from_luma ||
+      ((cm->subsampling_x == 1) && (cm->subsampling_y == 1) &&
+       (pars->num_y_points == 0))) {
+    pars->num_cb_points = 0;
+    pars->num_cr_points = 0;
   } else {
-    if (cm->profile == PROFILE_1 || cm->profile == PROFILE_3) {
-      // Note if colorspace is SRGB then 4:4:4 chroma sampling is assumed.
-      // 4:2:2 or 4:4:0 chroma sampling is not allowed.
-      cm->subsampling_y = cm->subsampling_x = 0;
-      if (aom_rb_read_bit(rb))
+    pars->num_cb_points = aom_rb_read_literal(rb, 4);  // max 10
+    if (pars->num_cb_points > 10)
+      aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+                         "Number of points for film grain cb scaling function "
+                         "exceeds the maximum value.");
+    for (int i = 0; i < pars->num_cb_points; i++) {
+      pars->scaling_points_cb[i][0] = aom_rb_read_literal(rb, 8);
+      if (i &&
+          pars->scaling_points_cb[i - 1][0] >= pars->scaling_points_cb[i][0])
         aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
-                           "Reserved bit set");
-    } else {
+                           "First coordinate of the scaling function points "
+                           "shall be increasing.");
+      pars->scaling_points_cb[i][1] = aom_rb_read_literal(rb, 8);
+    }
+
+    pars->num_cr_points = aom_rb_read_literal(rb, 4);  // max 10
+    if (pars->num_cr_points > 10)
       aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
-                         "4:4:4 color not supported in profile 0 or 2");
+                         "Number of points for film grain cr scaling function "
+                         "exceeds the maximum value.");
+    for (int i = 0; i < pars->num_cr_points; i++) {
+      pars->scaling_points_cr[i][0] = aom_rb_read_literal(rb, 8);
+      if (i &&
+          pars->scaling_points_cr[i - 1][0] >= pars->scaling_points_cr[i][0])
+        aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+                           "First coordinate of the scaling function points "
+                           "shall be increasing.");
+      pars->scaling_points_cr[i][1] = aom_rb_read_literal(rb, 8);
     }
+
+    if ((cm->subsampling_x == 1) && (cm->subsampling_y == 1) &&
+        (((pars->num_cb_points == 0) && (pars->num_cr_points != 0)) ||
+         ((pars->num_cb_points != 0) && (pars->num_cr_points == 0))))
+      aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+                         "In YCbCr 4:2:0, film grain shall be applied "
+                         "to both chroma components or neither.");
   }
+
+  pars->scaling_shift = aom_rb_read_literal(rb, 2) + 8;  // 8 + value
+
+  // AR coefficients
+  // Only sent if the corresponsing scaling function has
+  // more than 0 points
+
+  pars->ar_coeff_lag = aom_rb_read_literal(rb, 2);
+
+  int num_pos_luma = 2 * pars->ar_coeff_lag * (pars->ar_coeff_lag + 1);
+  int num_pos_chroma = num_pos_luma;
+  if (pars->num_y_points > 0) ++num_pos_chroma;
+
+  if (pars->num_y_points)
+    for (int i = 0; i < num_pos_luma; i++)
+      pars->ar_coeffs_y[i] = aom_rb_read_literal(rb, 8) - 128;
+
+  if (pars->num_cb_points || pars->chroma_scaling_from_luma)
+    for (int i = 0; i < num_pos_chroma; i++)
+      pars->ar_coeffs_cb[i] = aom_rb_read_literal(rb, 8) - 128;
+
+  if (pars->num_cr_points || pars->chroma_scaling_from_luma)
+    for (int i = 0; i < num_pos_chroma; i++)
+      pars->ar_coeffs_cr[i] = aom_rb_read_literal(rb, 8) - 128;
+
+  pars->ar_coeff_shift = aom_rb_read_literal(rb, 2) + 6;  // 6 + value
+
+  pars->grain_scale_shift = aom_rb_read_literal(rb, 2);
+
+  if (pars->num_cb_points) {
+    pars->cb_mult = aom_rb_read_literal(rb, 8);
+    pars->cb_luma_mult = aom_rb_read_literal(rb, 8);
+    pars->cb_offset = aom_rb_read_literal(rb, 9);
+  }
+
+  if (pars->num_cr_points) {
+    pars->cr_mult = aom_rb_read_literal(rb, 8);
+    pars->cr_luma_mult = aom_rb_read_literal(rb, 8);
+    pars->cr_offset = aom_rb_read_literal(rb, 9);
+  }
+
+  pars->overlap_flag = aom_rb_read_bit(rb);
+
+  pars->clip_to_restricted_range = aom_rb_read_bit(rb);
 }
 
-#if CONFIG_REFERENCE_BUFFER
-void read_sequence_header(SequenceHeader *seq_params,
-                          struct aom_read_bit_buffer *rb) {
-  /* Placeholder for actually reading from the bitstream */
-  seq_params->frame_id_numbers_present_flag = aom_rb_read_bit(rb);
-  if (seq_params->frame_id_numbers_present_flag) {
-    seq_params->frame_id_length_minus7 = aom_rb_read_literal(rb, 4);
-    seq_params->delta_frame_id_length_minus2 = aom_rb_read_literal(rb, 4);
+static void read_film_grain(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
+  if (cm->film_grain_params_present && (cm->show_frame || cm->showable_frame)) {
+    av1_read_film_grain_params(cm, rb);
+  } else {
+    memset(&cm->film_grain_params, 0, sizeof(cm->film_grain_params));
   }
+  cm->film_grain_params.bit_depth = cm->bit_depth;
+  memcpy(&cm->cur_frame->film_grain_params, &cm->film_grain_params,
+         sizeof(aom_film_grain_t));
 }
-#endif  // CONFIG_REFERENCE_BUFFER
 
-static void read_compound_tools(AV1_COMMON *cm,
-                                struct aom_read_bit_buffer *rb) {
-  (void)cm;
-  (void)rb;
-#if CONFIG_INTERINTRA
-  if (!frame_is_intra_only(cm) && cm->reference_mode != COMPOUND_REFERENCE) {
-    cm->allow_interintra_compound = aom_rb_read_bit(rb);
+void av1_read_color_config(AV1_COMMON *cm, struct aom_read_bit_buffer *rb,
+                           int allow_lowbitdepth) {
+  av1_read_bitdepth(cm, rb);
+
+  cm->use_highbitdepth = cm->bit_depth > AOM_BITS_8 || !allow_lowbitdepth;
+  // monochrome bit (not needed for PROFILE_1)
+  const int is_monochrome = cm->profile != PROFILE_1 ? aom_rb_read_bit(rb) : 0;
+  cm->seq_params.monochrome = is_monochrome;
+  int color_description_present_flag = aom_rb_read_bit(rb);
+  if (color_description_present_flag) {
+    cm->color_primaries = aom_rb_read_literal(rb, 8);
+    cm->transfer_characteristics = aom_rb_read_literal(rb, 8);
+    cm->matrix_coefficients = aom_rb_read_literal(rb, 8);
   } else {
-    cm->allow_interintra_compound = 0;
-  }
-#endif  // CONFIG_INTERINTRA
-#if CONFIG_WEDGE || CONFIG_COMPOUND_SEGMENT
-#if CONFIG_COMPOUND_SINGLEREF
-  if (!frame_is_intra_only(cm)) {
-#else   // !CONFIG_COMPOUND_SINGLEREF
-  if (!frame_is_intra_only(cm) && cm->reference_mode != SINGLE_REFERENCE) {
-#endif  // CONFIG_COMPOUND_SINGLEREF
-    cm->allow_masked_compound = aom_rb_read_bit(rb);
+    cm->color_primaries = AOM_CICP_CP_UNSPECIFIED;
+    cm->transfer_characteristics = AOM_CICP_TC_UNSPECIFIED;
+    cm->matrix_coefficients = AOM_CICP_MC_UNSPECIFIED;
+  }
+  if (is_monochrome) {
+    // [16,235] (including xvycc) vs [0,255] range
+    cm->color_range = aom_rb_read_bit(rb);
+    cm->subsampling_y = cm->subsampling_x = 1;
+    cm->chroma_sample_position = AOM_CSP_UNKNOWN;
+    cm->separate_uv_delta_q = 0;
+    return;
+  }
+  if (cm->color_primaries == AOM_CICP_CP_BT_709 &&
+      cm->transfer_characteristics == AOM_CICP_TC_SRGB &&
+      cm->matrix_coefficients == AOM_CICP_MC_IDENTITY) {  // it would be better
+                                                          // to remove this
+                                                          // dependency too
+    cm->subsampling_y = cm->subsampling_x = 0;
+    cm->color_range = 1;  // assume full color-range
+    if (!(cm->profile == PROFILE_1 ||
+          (cm->profile == PROFILE_2 && cm->bit_depth == AOM_BITS_12))) {
+      aom_internal_error(
+          &cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+          "sRGB colorspace not compatible with specified profile");
+    }
   } else {
-    cm->allow_masked_compound = 0;
+    // [16,235] (including xvycc) vs [0,255] range
+    cm->color_range = aom_rb_read_bit(rb);
+    if (cm->profile == PROFILE_0) {
+      // 420 only
+      cm->subsampling_x = cm->subsampling_y = 1;
+    } else if (cm->profile == PROFILE_1) {
+      // 444 only
+      cm->subsampling_x = cm->subsampling_y = 0;
+    } else {
+      assert(cm->profile == PROFILE_2);
+      if (cm->bit_depth == AOM_BITS_12) {
+        cm->subsampling_x = aom_rb_read_bit(rb);
+        if (cm->subsampling_x)
+          cm->subsampling_y = aom_rb_read_bit(rb);  // 422 or 420
+        else
+          cm->subsampling_y = 0;  // 444
+      } else {
+        // 422
+        cm->subsampling_x = 1;
+        cm->subsampling_y = 0;
+      }
+    }
+    if (cm->matrix_coefficients == AOM_CICP_MC_IDENTITY &&
+        (cm->subsampling_x || cm->subsampling_y)) {
+      aom_internal_error(
+          &cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+          "Identity CICP Matrix incompatible with non 4:4:4 color sampling");
+    }
+    if (cm->subsampling_x && cm->subsampling_y) {
+      cm->chroma_sample_position = aom_rb_read_literal(rb, 2);
+    }
   }
-#endif  // CONFIG_WEDGE || CONFIG_COMPOUND_SEGMENT
+  cm->separate_uv_delta_q = aom_rb_read_bit(rb);
 }
 
-#if CONFIG_VAR_REFS
-static void check_valid_ref_frames(AV1_COMMON *cm) {
-  MV_REFERENCE_FRAME ref_frame;
-  // TODO(zoeliu): To handle ALTREF_FRAME the same way as do with other
-  //               reference frames: Current encoder invalid ALTREF when ALTREF
-  //               is the same as LAST, but invalid all the other references
-  //               when they are the same as ALTREF.
-  for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
-    RefBuffer *const ref_buf = &cm->frame_refs[ref_frame - LAST_FRAME];
-
-    if (ref_buf->idx != INVALID_IDX) {
-      ref_buf->is_valid = 1;
-
-      MV_REFERENCE_FRAME ref;
-      for (ref = LAST_FRAME; ref < ref_frame; ++ref) {
-        RefBuffer *const buf = &cm->frame_refs[ref - LAST_FRAME];
-        if (buf->is_valid && buf->idx == ref_buf->idx) {
-          if (ref_frame != ALTREF_FRAME || ref == LAST_FRAME) {
-            ref_buf->is_valid = 0;
-            break;
-          } else {
-            buf->is_valid = 0;
-          }
-        }
+void av1_read_timing_info_header(AV1_COMMON *cm,
+                                 struct aom_read_bit_buffer *rb) {
+  cm->timing_info.num_units_in_display_tick = aom_rb_read_unsigned_literal(
+      rb, 32);  // Number of units in a display tick
+  cm->timing_info.time_scale =
+      aom_rb_read_unsigned_literal(rb, 32);  // Time scale
+  if (cm->timing_info.num_units_in_display_tick == 0 ||
+      cm->timing_info.time_scale == 0) {
+    aom_internal_error(
+        &cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+        "num_units_in_display_tick and time_scale must be greater than 0.");
+  }
+  cm->timing_info.equal_picture_interval =
+      aom_rb_read_bit(rb);  // Equal picture interval bit
+  if (cm->timing_info.equal_picture_interval) {
+    cm->timing_info.num_ticks_per_picture =
+        aom_rb_read_uvlc(rb) + 1;  // ticks per picture
+    if (cm->timing_info.num_ticks_per_picture == 0) {
+      aom_internal_error(
+          &cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+          "num_ticks_per_picture_minus_1 cannot be (1 << 32) − 1.");
+    }
+  }
+}
+
+void av1_read_decoder_model_info(AV1_COMMON *cm,
+                                 struct aom_read_bit_buffer *rb) {
+  cm->buffer_model.encoder_decoder_buffer_delay_length =
+      aom_rb_read_literal(rb, 5) + 1;
+  cm->buffer_model.num_units_in_decoding_tick = aom_rb_read_unsigned_literal(
+      rb, 32);  // Number of units in a decoding tick
+  cm->buffer_model.buffer_removal_delay_length = aom_rb_read_literal(rb, 5) + 1;
+  cm->buffer_model.frame_presentation_delay_length =
+      aom_rb_read_literal(rb, 5) + 1;
+}
+
+void av1_read_op_parameters_info(AV1_COMMON *const cm,
+                                 struct aom_read_bit_buffer *rb, int op_num) {
+  // The cm->op_params array has MAX_NUM_OPERATING_POINTS + 1 elements.
+  if (op_num > MAX_NUM_OPERATING_POINTS) {
+    aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+                       "AV1 does not support %d decoder model operating points",
+                       op_num + 1);
+  }
+
+  cm->op_params[op_num].decoder_buffer_delay = aom_rb_read_literal(
+      rb, cm->buffer_model.encoder_decoder_buffer_delay_length);
+
+  cm->op_params[op_num].encoder_buffer_delay = aom_rb_read_literal(
+      rb, cm->buffer_model.encoder_decoder_buffer_delay_length);
+
+  cm->op_params[op_num].low_delay_mode_flag = aom_rb_read_bit(rb);
+}
+
+static void av1_read_tu_pts_info(AV1_COMMON *const cm,
+                                 struct aom_read_bit_buffer *rb) {
+  cm->tu_presentation_delay =
+      aom_rb_read_literal(rb, cm->buffer_model.frame_presentation_delay_length);
+}
+
+void read_sequence_header(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
+  // rb->error_handler may be triggered during aom_rb_read_bit(), raising
+  // internal errors and immediate decoding termination. We use a local variable
+  // to store the info. as we decode. At the end, if no errors have occurred,
+  // cm->seq_params is updated.
+  SequenceHeader sh = cm->seq_params;
+  SequenceHeader *const seq_params = &sh;
+  int num_bits_width = aom_rb_read_literal(rb, 4) + 1;
+  int num_bits_height = aom_rb_read_literal(rb, 4) + 1;
+  int max_frame_width = aom_rb_read_literal(rb, num_bits_width) + 1;
+  int max_frame_height = aom_rb_read_literal(rb, num_bits_height) + 1;
+
+  seq_params->num_bits_width = num_bits_width;
+  seq_params->num_bits_height = num_bits_height;
+  seq_params->max_frame_width = max_frame_width;
+  seq_params->max_frame_height = max_frame_height;
+
+  if (seq_params->reduced_still_picture_hdr) {
+    seq_params->frame_id_numbers_present_flag = 0;
+  } else {
+    seq_params->frame_id_numbers_present_flag = aom_rb_read_bit(rb);
+  }
+  if (seq_params->frame_id_numbers_present_flag) {
+    // We must always have delta_frame_id_length < frame_id_length,
+    // in order for a frame to be referenced with a unique delta.
+    // Avoid wasting bits by using a coding that enforces this restriction.
+    seq_params->delta_frame_id_length = aom_rb_read_literal(rb, 4) + 2;
+    seq_params->frame_id_length =
+        aom_rb_read_literal(rb, 3) + seq_params->delta_frame_id_length + 1;
+    if (seq_params->frame_id_length > 16)
+      aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
+                         "Invalid frame_id_length");
+  }
+
+  setup_sb_size(seq_params, rb);
+
+  seq_params->enable_filter_intra = aom_rb_read_bit(rb);
+  seq_params->enable_intra_edge_filter = aom_rb_read_bit(rb);
+
+  if (seq_params->reduced_still_picture_hdr) {
+    seq_params->enable_interintra_compound = 0;
+    seq_params->enable_masked_compound = 0;
+    seq_params->enable_warped_motion = 0;
+    seq_params->enable_dual_filter = 0;
+    seq_params->enable_order_hint = 0;
+    seq_params->enable_jnt_comp = 0;
+    seq_params->enable_ref_frame_mvs = 0;
+    seq_params->force_screen_content_tools = 2;  // SELECT_SCREEN_CONTENT_TOOLS
+    seq_params->force_integer_mv = 2;            // SELECT_INTEGER_MV
+    seq_params->order_hint_bits_minus_1 = -1;
+  } else {
+    seq_params->enable_interintra_compound = aom_rb_read_bit(rb);
+    seq_params->enable_masked_compound = aom_rb_read_bit(rb);
+    seq_params->enable_warped_motion = aom_rb_read_bit(rb);
+    seq_params->enable_dual_filter = aom_rb_read_bit(rb);
+
+    seq_params->enable_order_hint = aom_rb_read_bit(rb);
+    seq_params->enable_jnt_comp =
+        seq_params->enable_order_hint ? aom_rb_read_bit(rb) : 0;
+    seq_params->enable_ref_frame_mvs =
+        seq_params->enable_order_hint ? aom_rb_read_bit(rb) : 0;
+
+    if (aom_rb_read_bit(rb)) {
+      seq_params->force_screen_content_tools =
+          2;  // SELECT_SCREEN_CONTENT_TOOLS
+    } else {
+      seq_params->force_screen_content_tools = aom_rb_read_bit(rb);
+    }
+
+    if (seq_params->force_screen_content_tools > 0) {
+      if (aom_rb_read_bit(rb)) {
+        seq_params->force_integer_mv = 2;  // SELECT_INTEGER_MV
+      } else {
+        seq_params->force_integer_mv = aom_rb_read_bit(rb);
       }
     } else {
-      ref_buf->is_valid = 0;
+      seq_params->force_integer_mv = 2;  // SELECT_INTEGER_MV
     }
+    seq_params->order_hint_bits_minus_1 =
+        seq_params->enable_order_hint ? aom_rb_read_literal(rb, 3) : -1;
   }
+
+  seq_params->enable_superres = aom_rb_read_bit(rb);
+  seq_params->enable_cdef = aom_rb_read_bit(rb);
+  seq_params->enable_restoration = aom_rb_read_bit(rb);
+  cm->seq_params = *seq_params;
 }
-#endif  // CONFIG_VAR_REFS
 
-#if CONFIG_GLOBAL_MOTION
 static int read_global_motion_params(WarpedMotionParams *params,
                                      const WarpedMotionParams *ref_params,
                                      struct aom_read_bit_buffer *rb,
                                      int allow_hp) {
   TransformationType type = aom_rb_read_bit(rb);
   if (type != IDENTITY) {
-#if GLOBAL_TRANS_TYPES > 4
-    type += aom_rb_read_literal(rb, GLOBAL_TYPE_BITS);
-#else
     if (aom_rb_read_bit(rb))
       type = ROTZOOM;
     else
       type = aom_rb_read_bit(rb) ? TRANSLATION : AFFINE;
-#endif  // GLOBAL_TRANS_TYPES > 4
   }
 
-  int trans_bits;
-  int trans_dec_factor;
-  int trans_prec_diff;
   *params = default_warp_params;
   params->wmtype = type;
-  switch (type) {
-    case HOMOGRAPHY:
-    case HORTRAPEZOID:
-    case VERTRAPEZOID:
-      if (type != HORTRAPEZOID)
-        params->wmmat[6] =
-            aom_rb_read_signed_primitive_refsubexpfin(
-                rb, GM_ROW3HOMO_MAX + 1, SUBEXPFIN_K,
-                (ref_params->wmmat[6] >> GM_ROW3HOMO_PREC_DIFF)) *
-            GM_ROW3HOMO_DECODE_FACTOR;
-      if (type != VERTRAPEZOID)
-        params->wmmat[7] =
-            aom_rb_read_signed_primitive_refsubexpfin(
-                rb, GM_ROW3HOMO_MAX + 1, SUBEXPFIN_K,
-                (ref_params->wmmat[7] >> GM_ROW3HOMO_PREC_DIFF)) *
-            GM_ROW3HOMO_DECODE_FACTOR;
-    case AFFINE:
-    case ROTZOOM:
-      params->wmmat[2] = aom_rb_read_signed_primitive_refsubexpfin(
-                             rb, GM_ALPHA_MAX + 1, SUBEXPFIN_K,
-                             (ref_params->wmmat[2] >> GM_ALPHA_PREC_DIFF) -
-                                 (1 << GM_ALPHA_PREC_BITS)) *
-                             GM_ALPHA_DECODE_FACTOR +
-                         (1 << WARPEDMODEL_PREC_BITS);
-      if (type != VERTRAPEZOID)
-        params->wmmat[3] = aom_rb_read_signed_primitive_refsubexpfin(
-                               rb, GM_ALPHA_MAX + 1, SUBEXPFIN_K,
-                               (ref_params->wmmat[3] >> GM_ALPHA_PREC_DIFF)) *
-                           GM_ALPHA_DECODE_FACTOR;
-      if (type >= AFFINE) {
-        if (type != HORTRAPEZOID)
-          params->wmmat[4] = aom_rb_read_signed_primitive_refsubexpfin(
-                                 rb, GM_ALPHA_MAX + 1, SUBEXPFIN_K,
-                                 (ref_params->wmmat[4] >> GM_ALPHA_PREC_DIFF)) *
-                             GM_ALPHA_DECODE_FACTOR;
-        params->wmmat[5] = aom_rb_read_signed_primitive_refsubexpfin(
-                               rb, GM_ALPHA_MAX + 1, SUBEXPFIN_K,
-                               (ref_params->wmmat[5] >> GM_ALPHA_PREC_DIFF) -
-                                   (1 << GM_ALPHA_PREC_BITS)) *
-                               GM_ALPHA_DECODE_FACTOR +
-                           (1 << WARPEDMODEL_PREC_BITS);
-      } else {
-        params->wmmat[4] = -params->wmmat[3];
-        params->wmmat[5] = params->wmmat[2];
-      }
-    // fallthrough intended
-    case TRANSLATION:
-      trans_bits = (type == TRANSLATION) ? GM_ABS_TRANS_ONLY_BITS - !allow_hp
-                                         : GM_ABS_TRANS_BITS;
-      trans_dec_factor = (type == TRANSLATION)
-                             ? GM_TRANS_ONLY_DECODE_FACTOR * (1 << !allow_hp)
-                             : GM_TRANS_DECODE_FACTOR;
-      trans_prec_diff = (type == TRANSLATION)
-                            ? GM_TRANS_ONLY_PREC_DIFF + !allow_hp
-                            : GM_TRANS_PREC_DIFF;
-      params->wmmat[0] = aom_rb_read_signed_primitive_refsubexpfin(
-                             rb, (1 << trans_bits) + 1, SUBEXPFIN_K,
-                             (ref_params->wmmat[0] >> trans_prec_diff)) *
-                         trans_dec_factor;
-      params->wmmat[1] = aom_rb_read_signed_primitive_refsubexpfin(
-                             rb, (1 << trans_bits) + 1, SUBEXPFIN_K,
-                             (ref_params->wmmat[1] >> trans_prec_diff)) *
-                         trans_dec_factor;
-    case IDENTITY: break;
-    default: assert(0);
+
+  if (type >= ROTZOOM) {
+    params->wmmat[2] = aom_rb_read_signed_primitive_refsubexpfin(
+                           rb, GM_ALPHA_MAX + 1, SUBEXPFIN_K,
+                           (ref_params->wmmat[2] >> GM_ALPHA_PREC_DIFF) -
+                               (1 << GM_ALPHA_PREC_BITS)) *
+                           GM_ALPHA_DECODE_FACTOR +
+                       (1 << WARPEDMODEL_PREC_BITS);
+    params->wmmat[3] = aom_rb_read_signed_primitive_refsubexpfin(
+                           rb, GM_ALPHA_MAX + 1, SUBEXPFIN_K,
+                           (ref_params->wmmat[3] >> GM_ALPHA_PREC_DIFF)) *
+                       GM_ALPHA_DECODE_FACTOR;
+  }
+
+  if (type >= AFFINE) {
+    params->wmmat[4] = aom_rb_read_signed_primitive_refsubexpfin(
+                           rb, GM_ALPHA_MAX + 1, SUBEXPFIN_K,
+                           (ref_params->wmmat[4] >> GM_ALPHA_PREC_DIFF)) *
+                       GM_ALPHA_DECODE_FACTOR;
+    params->wmmat[5] = aom_rb_read_signed_primitive_refsubexpfin(
+                           rb, GM_ALPHA_MAX + 1, SUBEXPFIN_K,
+                           (ref_params->wmmat[5] >> GM_ALPHA_PREC_DIFF) -
+                               (1 << GM_ALPHA_PREC_BITS)) *
+                           GM_ALPHA_DECODE_FACTOR +
+                       (1 << WARPEDMODEL_PREC_BITS);
+  } else {
+    params->wmmat[4] = -params->wmmat[3];
+    params->wmmat[5] = params->wmmat[2];
+  }
+
+  if (type >= TRANSLATION) {
+    const int trans_bits = (type == TRANSLATION)
+                               ? GM_ABS_TRANS_ONLY_BITS - !allow_hp
+                               : GM_ABS_TRANS_BITS;
+    const int trans_dec_factor =
+        (type == TRANSLATION) ? GM_TRANS_ONLY_DECODE_FACTOR * (1 << !allow_hp)
+                              : GM_TRANS_DECODE_FACTOR;
+    const int trans_prec_diff = (type == TRANSLATION)
+                                    ? GM_TRANS_ONLY_PREC_DIFF + !allow_hp
+                                    : GM_TRANS_PREC_DIFF;
+    params->wmmat[0] = aom_rb_read_signed_primitive_refsubexpfin(
+                           rb, (1 << trans_bits) + 1, SUBEXPFIN_K,
+                           (ref_params->wmmat[0] >> trans_prec_diff)) *
+                       trans_dec_factor;
+    params->wmmat[1] = aom_rb_read_signed_primitive_refsubexpfin(
+                           rb, (1 << trans_bits) + 1, SUBEXPFIN_K,
+                           (ref_params->wmmat[1] >> trans_prec_diff)) *
+                       trans_dec_factor;
   }
+
   if (params->wmtype <= AFFINE) {
     int good_shear_params = get_shear_params(params);
     if (!good_shear_params) return 0;
@@ -4596,16 +3528,18 @@ static int read_global_motion_params(WarpedMotionParams *params,
 }
 
 static void read_global_motion(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
-  int frame;
-  for (frame = LAST_FRAME; frame <= ALTREF_FRAME; ++frame) {
+  for (int frame = LAST_FRAME; frame <= ALTREF_FRAME; ++frame) {
     const WarpedMotionParams *ref_params =
-        cm->error_resilient_mode ? &default_warp_params
-                                 : &cm->prev_frame->global_motion[frame];
+        cm->prev_frame ? &cm->prev_frame->global_motion[frame]
+                       : &default_warp_params;
     int good_params = read_global_motion_params(
         &cm->global_motion[frame], ref_params, rb, cm->allow_high_precision_mv);
-    if (!good_params)
-      aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
-                         "Invalid shear parameters for global motion.");
+    if (!good_params) {
+#if WARPED_MOTION_DEBUG
+      printf("Warning: unexpected global motion shear params from aomenc\n");
+#endif
+      cm->global_motion[frame].invalid = 1;
+    }
 
     // TODO(sarahparker, debargha): The logic in the commented out code below
     // does not work currently and causes mismatches when resize is on. Fix it
@@ -4631,252 +3565,397 @@ static void read_global_motion(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
            */
   }
   memcpy(cm->cur_frame->global_motion, cm->global_motion,
-         TOTAL_REFS_PER_FRAME * sizeof(WarpedMotionParams));
+         REF_FRAMES * sizeof(WarpedMotionParams));
 }
-#endif  // CONFIG_GLOBAL_MOTION
 
-static size_t read_uncompressed_header(AV1Decoder *pbi,
-                                       struct aom_read_bit_buffer *rb) {
+static void show_existing_frame_reset(AV1Decoder *const pbi,
+                                      int existing_frame_idx) {
+  AV1_COMMON *const cm = &pbi->common;
+  BufferPool *const pool = cm->buffer_pool;
+  RefCntBuffer *const frame_bufs = pool->frame_bufs;
+
+  assert(cm->show_existing_frame);
+
+  cm->frame_type = KEY_FRAME;
+
+  pbi->refresh_frame_flags = (1 << REF_FRAMES) - 1;
+
+  for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
+    cm->frame_refs[i].idx = INVALID_IDX;
+    cm->frame_refs[i].buf = NULL;
+  }
+
+  if (pbi->need_resync) {
+    memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
+    pbi->need_resync = 0;
+  }
+
+  cm->cur_frame->intra_only = 1;
+
+  if (cm->seq_params.frame_id_numbers_present_flag) {
+    /* If bitmask is set, update reference frame id values and
+       mark frames as valid for reference.
+       Note that the displayed frame be valid for referencing
+       in order to have been selected.
+    */
+    int refresh_frame_flags = pbi->refresh_frame_flags;
+    int display_frame_id = cm->ref_frame_id[existing_frame_idx];
+    for (int i = 0; i < REF_FRAMES; i++) {
+      if ((refresh_frame_flags >> i) & 1) {
+        cm->ref_frame_id[i] = display_frame_id;
+        cm->valid_for_referencing[i] = 1;
+      }
+    }
+  }
+
+  cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_DISABLED;
+
+  // Generate next_ref_frame_map.
+  lock_buffer_pool(pool);
+  int ref_index = 0;
+  for (int mask = pbi->refresh_frame_flags; mask; mask >>= 1) {
+    if (mask & 1) {
+      cm->next_ref_frame_map[ref_index] = cm->new_fb_idx;
+      ++frame_bufs[cm->new_fb_idx].ref_count;
+    } else {
+      cm->next_ref_frame_map[ref_index] = cm->ref_frame_map[ref_index];
+    }
+    // Current thread holds the reference frame.
+    if (cm->ref_frame_map[ref_index] >= 0)
+      ++frame_bufs[cm->ref_frame_map[ref_index]].ref_count;
+    ++ref_index;
+  }
+
+  for (; ref_index < REF_FRAMES; ++ref_index) {
+    cm->next_ref_frame_map[ref_index] = cm->ref_frame_map[ref_index];
+
+    // Current thread holds the reference frame.
+    if (cm->ref_frame_map[ref_index] >= 0)
+      ++frame_bufs[cm->ref_frame_map[ref_index]].ref_count;
+  }
+  unlock_buffer_pool(pool);
+  pbi->hold_ref_buf = 1;
+
+  // Reload the adapted CDFs from when we originally coded this keyframe
+  *cm->fc = cm->frame_contexts[existing_frame_idx];
+}
+
+static int read_uncompressed_header(AV1Decoder *pbi,
+                                    struct aom_read_bit_buffer *rb) {
   AV1_COMMON *const cm = &pbi->common;
   MACROBLOCKD *const xd = &pbi->mb;
   BufferPool *const pool = cm->buffer_pool;
   RefCntBuffer *const frame_bufs = pool->frame_bufs;
-  int i, mask, ref_index = 0;
-  size_t sz;
+
+  if (!pbi->sequence_header_ready) {
+    aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
+                       "No sequence header");
+  }
 
   cm->last_frame_type = cm->frame_type;
   cm->last_intra_only = cm->intra_only;
 
-#if CONFIG_EXT_REFS
   // NOTE: By default all coded frames to be used as a reference
   cm->is_reference_frame = 1;
-#endif  // CONFIG_EXT_REFS
 
-#if !CONFIG_OBU
-  if (aom_rb_read_literal(rb, 2) != AOM_FRAME_MARKER)
-    aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
-                       "Invalid frame marker");
+  if (cm->seq_params.reduced_still_picture_hdr) {
+    cm->show_existing_frame = 0;
+    cm->show_frame = 1;
+    cm->frame_type = KEY_FRAME;
+    cm->error_resilient_mode = 1;
+  } else {
+    cm->show_existing_frame = aom_rb_read_bit(rb);
+    cm->reset_decoder_state = 0;
+
+    if (cm->show_existing_frame) {
+      // Show an existing frame directly.
+      const int existing_frame_idx = aom_rb_read_literal(rb, 3);
+      const int frame_to_show = cm->ref_frame_map[existing_frame_idx];
+      if (cm->seq_params.decoder_model_info_present_flag &&
+          cm->timing_info.equal_picture_interval == 0) {
+        av1_read_tu_pts_info(cm, rb);
+      }
+      if (cm->seq_params.frame_id_numbers_present_flag) {
+        int frame_id_length = cm->seq_params.frame_id_length;
+        int display_frame_id = aom_rb_read_literal(rb, frame_id_length);
+        /* Compare display_frame_id with ref_frame_id and check valid for
+         * referencing */
+        if (display_frame_id != cm->ref_frame_id[existing_frame_idx] ||
+            cm->valid_for_referencing[existing_frame_idx] == 0)
+          aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
+                             "Reference buffer frame ID mismatch");
+      }
+      lock_buffer_pool(pool);
+      if (frame_to_show < 0 || frame_bufs[frame_to_show].ref_count < 1) {
+        unlock_buffer_pool(pool);
+        aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+                           "Buffer %d does not contain a decoded frame",
+                           frame_to_show);
+      }
+      ref_cnt_fb(frame_bufs, &cm->new_fb_idx, frame_to_show);
+      cm->reset_decoder_state =
+          frame_bufs[frame_to_show].frame_type == KEY_FRAME;
+      unlock_buffer_pool(pool);
 
-  cm->profile = av1_read_profile(rb);
+      cm->lf.filter_level[0] = 0;
+      cm->lf.filter_level[1] = 0;
+      cm->show_frame = 1;
 
-  const BITSTREAM_PROFILE MAX_SUPPORTED_PROFILE =
-      CONFIG_HIGHBITDEPTH ? MAX_PROFILES : PROFILE_2;
+      if (!frame_bufs[frame_to_show].showable_frame) {
+        aom_merge_corrupted_flag(&xd->corrupted, 1);
+      }
+      if (cm->reset_decoder_state) frame_bufs[frame_to_show].showable_frame = 0;
 
-  if (cm->profile >= MAX_SUPPORTED_PROFILE)
-    aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
-                       "Unsupported bitstream profile");
-#endif
+      cm->film_grain_params = frame_bufs[frame_to_show].film_grain_params;
 
-#if CONFIG_EXT_TILE
-  cm->large_scale_tile = aom_rb_read_literal(rb, 1);
-#if CONFIG_REFERENCE_BUFFER
-  if (cm->large_scale_tile) cm->seq_params.frame_id_numbers_present_flag = 0;
-#endif  // CONFIG_REFERENCE_BUFFER
-#endif  // CONFIG_EXT_TILE
+      if (cm->reset_decoder_state) {
+        show_existing_frame_reset(pbi, existing_frame_idx);
+      } else {
+        pbi->refresh_frame_flags = 0;
+      }
 
-  cm->show_existing_frame = aom_rb_read_bit(rb);
+      return 0;
+    }
 
-  if (cm->show_existing_frame) {
-    // Show an existing frame directly.
-    const int existing_frame_idx = aom_rb_read_literal(rb, 3);
-    const int frame_to_show = cm->ref_frame_map[existing_frame_idx];
-#if CONFIG_REFERENCE_BUFFER
-    if (cm->seq_params.frame_id_numbers_present_flag) {
-      int frame_id_length = cm->seq_params.frame_id_length_minus7 + 7;
-      int display_frame_id = aom_rb_read_literal(rb, frame_id_length);
-      /* Compare display_frame_id with ref_frame_id and check valid for
-       * referencing */
-      if (display_frame_id != cm->ref_frame_id[existing_frame_idx] ||
-          cm->valid_for_referencing[existing_frame_idx] == 0)
-        aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
-                           "Reference buffer frame ID mismatch");
+    cm->frame_type = (FRAME_TYPE)aom_rb_read_literal(rb, 2);  // 2 bits
+    cm->show_frame = aom_rb_read_bit(rb);
+    if (cm->seq_params.still_picture &&
+        (cm->frame_type != KEY_FRAME || !cm->show_frame)) {
+      aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
+                         "Still pictures must be coded as shown keyframes");
     }
-#endif
-    lock_buffer_pool(pool);
-    if (frame_to_show < 0 || frame_bufs[frame_to_show].ref_count < 1) {
-      unlock_buffer_pool(pool);
-      aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
-                         "Buffer %d does not contain a decoded frame",
-                         frame_to_show);
+    cm->showable_frame = cm->frame_type != KEY_FRAME;
+    if (cm->show_frame) {
+      if (cm->seq_params.decoder_model_info_present_flag &&
+          cm->timing_info.equal_picture_interval == 0)
+        av1_read_tu_pts_info(cm, rb);
+    } else {
+      // See if this frame can be used as show_existing_frame in future
+      cm->showable_frame = aom_rb_read_bit(rb);
     }
-    ref_cnt_fb(frame_bufs, &cm->new_fb_idx, frame_to_show);
-    unlock_buffer_pool(pool);
+    cm->cur_frame->showable_frame = cm->showable_frame;
+    cm->intra_only = cm->frame_type == INTRA_ONLY_FRAME;
+    cm->error_resilient_mode =
+        frame_is_sframe(cm) || (cm->frame_type == KEY_FRAME && cm->show_frame)
+            ? 1
+            : aom_rb_read_bit(rb);
+  }
 
-#if CONFIG_LOOPFILTER_LEVEL
-    cm->lf.filter_level[0] = 0;
-    cm->lf.filter_level[1] = 0;
-#else
-    cm->lf.filter_level = 0;
-#endif
-    cm->show_frame = 1;
-    pbi->refresh_frame_flags = 0;
+  cm->disable_cdf_update = aom_rb_read_bit(rb);
+  if (cm->seq_params.force_screen_content_tools == 2) {
+    cm->allow_screen_content_tools = aom_rb_read_bit(rb);
+  } else {
+    cm->allow_screen_content_tools = cm->seq_params.force_screen_content_tools;
+  }
 
-    if (cm->frame_parallel_decode) {
-      for (i = 0; i < REF_FRAMES; ++i)
-        cm->next_ref_frame_map[i] = cm->ref_frame_map[i];
+  if (cm->allow_screen_content_tools) {
+    if (cm->seq_params.force_integer_mv == 2) {
+      cm->cur_frame_force_integer_mv = aom_rb_read_bit(rb);
+    } else {
+      cm->cur_frame_force_integer_mv = cm->seq_params.force_integer_mv;
     }
-
-    return 0;
+  } else {
+    cm->cur_frame_force_integer_mv = 0;
   }
 
-#if !CONFIG_OBU
-  cm->frame_type = (FRAME_TYPE)aom_rb_read_bit(rb);
-  cm->show_frame = aom_rb_read_bit(rb);
-  if (cm->frame_type != KEY_FRAME)
-    cm->intra_only = cm->show_frame ? 0 : aom_rb_read_bit(rb);
-#else
-  cm->frame_type = (FRAME_TYPE)aom_rb_read_literal(rb, 2);  // 2 bits
-  cm->show_frame = aom_rb_read_bit(rb);
-  cm->intra_only = cm->frame_type == INTRA_ONLY_FRAME;
-#endif
-  cm->error_resilient_mode = aom_rb_read_bit(rb);
-#if CONFIG_REFERENCE_BUFFER
-#if !CONFIG_OBU
-  if (frame_is_intra_only(cm)) read_sequence_header(&cm->seq_params, rb);
-#endif  // !CONFIG_OBU
-  if (cm->seq_params.frame_id_numbers_present_flag) {
-    int frame_id_length = cm->seq_params.frame_id_length_minus7 + 7;
-    int diff_len = cm->seq_params.delta_frame_id_length_minus2 + 2;
-    int prev_frame_id = 0;
-    if (cm->frame_type != KEY_FRAME) {
-      prev_frame_id = cm->current_frame_id;
-    }
-    cm->current_frame_id = aom_rb_read_literal(rb, frame_id_length);
+  cm->frame_refs_short_signaling = 0;
+  int frame_size_override_flag = 0;
+  cm->allow_intrabc = 0;
+  cm->primary_ref_frame = PRIMARY_REF_NONE;
 
-    if (cm->frame_type != KEY_FRAME) {
-      int diff_frame_id;
-      if (cm->current_frame_id > prev_frame_id) {
-        diff_frame_id = cm->current_frame_id - prev_frame_id;
-      } else {
-        diff_frame_id =
-            (1 << frame_id_length) + cm->current_frame_id - prev_frame_id;
+  if (!cm->seq_params.reduced_still_picture_hdr) {
+    if (cm->seq_params.frame_id_numbers_present_flag) {
+      int frame_id_length = cm->seq_params.frame_id_length;
+      int diff_len = cm->seq_params.delta_frame_id_length;
+      int prev_frame_id = 0;
+      int have_prev_frame_id = !pbi->decoding_first_frame &&
+                               !(cm->frame_type == KEY_FRAME && cm->show_frame);
+      if (have_prev_frame_id) {
+        prev_frame_id = cm->current_frame_id;
       }
-      /* Check current_frame_id for conformance */
-      if (prev_frame_id == cm->current_frame_id ||
-          diff_frame_id >= (1 << (frame_id_length - 1))) {
-        aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
-                           "Invalid value of current_frame_id");
+      cm->current_frame_id = aom_rb_read_literal(rb, frame_id_length);
+
+      if (have_prev_frame_id) {
+        int diff_frame_id;
+        if (cm->current_frame_id > prev_frame_id) {
+          diff_frame_id = cm->current_frame_id - prev_frame_id;
+        } else {
+          diff_frame_id =
+              (1 << frame_id_length) + cm->current_frame_id - prev_frame_id;
+        }
+        /* Check current_frame_id for conformance */
+        if (prev_frame_id == cm->current_frame_id ||
+            diff_frame_id >= (1 << (frame_id_length - 1))) {
+          aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
+                             "Invalid value of current_frame_id");
+        }
       }
-    }
-    /* Check if some frames need to be marked as not valid for referencing */
-    for (i = 0; i < REF_FRAMES; i++) {
-      if (cm->frame_type == KEY_FRAME) {
-        cm->valid_for_referencing[i] = 0;
-      } else if (cm->current_frame_id - (1 << diff_len) > 0) {
-        if (cm->ref_frame_id[i] > cm->current_frame_id ||
-            cm->ref_frame_id[i] < cm->current_frame_id - (1 << diff_len))
-          cm->valid_for_referencing[i] = 0;
-      } else {
-        if (cm->ref_frame_id[i] > cm->current_frame_id &&
-            cm->ref_frame_id[i] <
-                (1 << frame_id_length) + cm->current_frame_id - (1 << diff_len))
+      /* Check if some frames need to be marked as not valid for referencing */
+      for (int i = 0; i < REF_FRAMES; i++) {
+        if (cm->frame_type == KEY_FRAME && cm->show_frame) {
           cm->valid_for_referencing[i] = 0;
+        } else if (cm->current_frame_id - (1 << diff_len) > 0) {
+          if (cm->ref_frame_id[i] > cm->current_frame_id ||
+              cm->ref_frame_id[i] < cm->current_frame_id - (1 << diff_len))
+            cm->valid_for_referencing[i] = 0;
+        } else {
+          if (cm->ref_frame_id[i] > cm->current_frame_id &&
+              cm->ref_frame_id[i] < (1 << frame_id_length) +
+                                        cm->current_frame_id - (1 << diff_len))
+            cm->valid_for_referencing[i] = 0;
+        }
+      }
+    }
+
+    frame_size_override_flag =
+        frame_is_sframe(cm) ? 1 : aom_rb_read_literal(rb, 1);
+
+    cm->frame_offset =
+        aom_rb_read_literal(rb, cm->seq_params.order_hint_bits_minus_1 + 1);
+    cm->current_video_frame = cm->frame_offset;
+
+    if (!cm->error_resilient_mode && !frame_is_intra_only(cm)) {
+      cm->primary_ref_frame = aom_rb_read_literal(rb, PRIMARY_REF_BITS);
+    }
+  }
+
+  if (cm->seq_params.decoder_model_info_present_flag) {
+    cm->buffer_removal_delay_present = aom_rb_read_bit(rb);
+    if (cm->buffer_removal_delay_present) {
+      for (int op_num = 0;
+           op_num < cm->seq_params.operating_points_cnt_minus_1 + 1; op_num++) {
+        if (cm->op_params[op_num].decoder_model_param_present_flag) {
+          if ((((cm->seq_params.operating_point_idc[op_num] >>
+                 cm->temporal_layer_id) &
+                0x1) &&
+               ((cm->seq_params.operating_point_idc[op_num] >>
+                 (cm->spatial_layer_id + 8)) &
+                0x1)) ||
+              cm->seq_params.operating_point_idc[op_num] == 0) {
+            cm->op_frame_timing[op_num].buffer_removal_delay =
+                aom_rb_read_literal(
+                    rb, cm->buffer_model.buffer_removal_delay_length);
+          } else {
+            cm->op_frame_timing[op_num].buffer_removal_delay = 0;
+          }
+        } else {
+          cm->op_frame_timing[op_num].buffer_removal_delay = 0;
+        }
       }
     }
   }
-#endif  // CONFIG_REFERENCE_BUFFER
   if (cm->frame_type == KEY_FRAME) {
-#if !CONFIG_OBU
-    read_bitdepth_colorspace_sampling(cm, rb, pbi->allow_lowbitdepth);
-#endif
-    pbi->refresh_frame_flags = (1 << REF_FRAMES) - 1;
+    if (!cm->show_frame)  // unshown keyframe (forward keyframe)
+      pbi->refresh_frame_flags = aom_rb_read_literal(rb, REF_FRAMES);
+    else  // shown keyframe
+      pbi->refresh_frame_flags = (1 << REF_FRAMES) - 1;
 
-    for (i = 0; i < INTER_REFS_PER_FRAME; ++i) {
+    for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
       cm->frame_refs[i].idx = INVALID_IDX;
       cm->frame_refs[i].buf = NULL;
-#if CONFIG_VAR_REFS
-      cm->frame_refs[i].is_valid = 0;
-#endif  // CONFIG_VAR_REFS
     }
-
-    setup_frame_size(cm, rb);
-    setup_sb_size(cm, rb);
-
     if (pbi->need_resync) {
       memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
       pbi->need_resync = 0;
     }
-#if CONFIG_ANS && ANS_MAX_SYMBOLS
-    cm->ans_window_size_log2 = aom_rb_read_literal(rb, 4) + 8;
-#endif  // CONFIG_ANS && ANS_MAX_SYMBOLS
-    cm->allow_screen_content_tools = aom_rb_read_bit(rb);
-#if CONFIG_AMVR
-    if (cm->allow_screen_content_tools) {
-      if (aom_rb_read_bit(rb)) {
-        cm->seq_mv_precision_level = 2;
-      } else {
-        cm->seq_mv_precision_level = aom_rb_read_bit(rb) ? 0 : 1;
-      }
-    } else {
-      cm->seq_mv_precision_level = 0;
-    }
-#endif
-#if CONFIG_TEMPMV_SIGNALING
-    cm->use_prev_frame_mvs = 0;
-#endif
   } else {
-    if (cm->intra_only) cm->allow_screen_content_tools = aom_rb_read_bit(rb);
-#if CONFIG_TEMPMV_SIGNALING
-    if (cm->intra_only || cm->error_resilient_mode) cm->use_prev_frame_mvs = 0;
-#endif
-#if CONFIG_NO_FRAME_CONTEXT_SIGNALING
-// The only way to reset all frame contexts to their default values is with a
-// keyframe.
-#else
-    if (cm->error_resilient_mode) {
-      cm->reset_frame_context = RESET_FRAME_CONTEXT_ALL;
-    } else {
-      if (cm->intra_only) {
-        cm->reset_frame_context = aom_rb_read_bit(rb)
-                                      ? RESET_FRAME_CONTEXT_ALL
-                                      : RESET_FRAME_CONTEXT_CURRENT;
-      } else {
-        cm->reset_frame_context = aom_rb_read_bit(rb)
-                                      ? RESET_FRAME_CONTEXT_CURRENT
-                                      : RESET_FRAME_CONTEXT_NONE;
-        if (cm->reset_frame_context == RESET_FRAME_CONTEXT_CURRENT)
-          cm->reset_frame_context = aom_rb_read_bit(rb)
-                                        ? RESET_FRAME_CONTEXT_ALL
-                                        : RESET_FRAME_CONTEXT_CURRENT;
-      }
-    }
-#endif
-
     if (cm->intra_only) {
-#if !CONFIG_OBU
-      read_bitdepth_colorspace_sampling(cm, rb, pbi->allow_lowbitdepth);
-#endif
-
       pbi->refresh_frame_flags = aom_rb_read_literal(rb, REF_FRAMES);
-      setup_frame_size(cm, rb);
-      setup_sb_size(cm, rb);
+      if (pbi->refresh_frame_flags == 0xFF) {
+        aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+                           "Intra only frames cannot have refresh flags 0xFF");
+      }
       if (pbi->need_resync) {
         memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
         pbi->need_resync = 0;
       }
-#if CONFIG_ANS && ANS_MAX_SYMBOLS
-      cm->ans_window_size_log2 = aom_rb_read_literal(rb, 4) + 8;
-#endif
     } else if (pbi->need_resync != 1) { /* Skip if need resync */
-#if CONFIG_OBU
-      pbi->refresh_frame_flags = (cm->frame_type == S_FRAME)
-                                     ? ~(1 << REF_FRAMES)
-                                     : aom_rb_read_literal(rb, REF_FRAMES);
-#else
-      pbi->refresh_frame_flags = aom_rb_read_literal(rb, REF_FRAMES);
-#endif
-
-#if CONFIG_EXT_REFS
+      pbi->refresh_frame_flags =
+          frame_is_sframe(cm) ? 0xFF : aom_rb_read_literal(rb, REF_FRAMES);
       if (!pbi->refresh_frame_flags) {
         // NOTE: "pbi->refresh_frame_flags == 0" indicates that the coded frame
         //       will not be used as a reference
         cm->is_reference_frame = 0;
       }
-#endif  // CONFIG_EXT_REFS
+    }
+  }
 
-      for (i = 0; i < INTER_REFS_PER_FRAME; ++i) {
-        const int ref = aom_rb_read_literal(rb, REF_FRAMES_LOG2);
-        const int idx = cm->ref_frame_map[ref];
+  if (!frame_is_intra_only(cm) || pbi->refresh_frame_flags != 0xFF) {
+    // Read all ref frame order hints if error_resilient_mode == 1
+    if (cm->error_resilient_mode && cm->seq_params.enable_order_hint) {
+      for (int ref_idx = 0; ref_idx < REF_FRAMES; ref_idx++) {
+        // Read order hint from bit stream
+        unsigned int frame_offset =
+            aom_rb_read_literal(rb, cm->seq_params.order_hint_bits_minus_1 + 1);
+        // Get buffer index
+        int buf_idx = cm->ref_frame_map[ref_idx];
+        assert(buf_idx < FRAME_BUFFERS);
+        if (buf_idx == -1 ||
+            frame_offset != frame_bufs[buf_idx].cur_frame_offset) {
+          if (buf_idx >= 0) {
+            lock_buffer_pool(pool);
+            decrease_ref_count(buf_idx, frame_bufs, pool);
+            unlock_buffer_pool(pool);
+          }
+          // If no corresponding buffer exists, allocate a new buffer with all
+          // pixels set to neutral grey.
+          buf_idx = get_free_fb(cm);
+          if (buf_idx == INVALID_IDX) {
+            aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
+                               "Unable to find free frame buffer");
+          }
+          lock_buffer_pool(pool);
+          if (aom_realloc_frame_buffer(
+                  &frame_bufs[buf_idx].buf, cm->seq_params.max_frame_width,
+                  cm->seq_params.max_frame_height, cm->subsampling_x,
+                  cm->subsampling_y, cm->use_highbitdepth, AOM_BORDER_IN_PIXELS,
+                  cm->byte_alignment,
+                  &pool->frame_bufs[buf_idx].raw_frame_buffer, pool->get_fb_cb,
+                  pool->cb_priv)) {
+            unlock_buffer_pool(pool);
+            aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
+                               "Failed to allocate frame buffer");
+          }
+          unlock_buffer_pool(pool);
+          set_planes_to_neutral_grey(cm, &frame_bufs[buf_idx].buf, 0);
+
+          cm->ref_frame_map[ref_idx] = buf_idx;
+          frame_bufs[buf_idx].cur_frame_offset = frame_offset;
+        }
+      }
+    }
+  }
+
+  if (cm->frame_type == KEY_FRAME) {
+    setup_frame_size(cm, frame_size_override_flag, rb);
+
+    if (cm->allow_screen_content_tools && !av1_superres_scaled(cm))
+      cm->allow_intrabc = aom_rb_read_bit(rb);
+    cm->allow_ref_frame_mvs = 0;
+    cm->prev_frame = NULL;
+  } else {
+    cm->allow_ref_frame_mvs = 0;
+
+    if (cm->intra_only) {
+      cm->cur_frame->film_grain_params_present = cm->film_grain_params_present;
+      setup_frame_size(cm, frame_size_override_flag, rb);
+      if (cm->allow_screen_content_tools && !av1_superres_scaled(cm))
+        cm->allow_intrabc = aom_rb_read_bit(rb);
+
+    } else if (pbi->need_resync != 1) { /* Skip if need resync */
+
+      // Frame refs short signaling is off when error resilient mode is on.
+      if (cm->seq_params.enable_order_hint)
+        cm->frame_refs_short_signaling = aom_rb_read_bit(rb);
+
+      if (cm->frame_refs_short_signaling) {
+        // == LAST_FRAME ==
+        const int lst_ref = aom_rb_read_literal(rb, REF_FRAMES_LOG2);
+        const int lst_idx = cm->ref_frame_map[lst_ref];
+
+        // == GOLDEN_FRAME ==
+        const int gld_ref = aom_rb_read_literal(rb, REF_FRAMES_LOG2);
+        const int gld_idx = cm->ref_frame_map[gld_ref];
 
         // Most of the time, streams start with a keyframe. In that case,
         // ref_frame_map will have been filled in at that point and will not
@@ -4884,146 +3963,136 @@ static size_t read_uncompressed_header(AV1Decoder *pbi,
         // with an intra-only frame, so long as they don't then signal a
         // reference to a slot that hasn't been set yet. That's what we are
         // checking here.
-        if (idx == -1)
+        if (lst_idx == -1)
           aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
                              "Inter frame requests nonexistent reference");
+        if (gld_idx == -1)
+          aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
+                             "Inter frame requests nonexistent reference");
+
+        av1_set_frame_refs(cm, lst_ref, gld_ref);
+      }
+
+      for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
+        int ref = 0;
+        if (!cm->frame_refs_short_signaling) {
+          ref = aom_rb_read_literal(rb, REF_FRAMES_LOG2);
+          const int idx = cm->ref_frame_map[ref];
+
+          // Most of the time, streams start with a keyframe. In that case,
+          // ref_frame_map will have been filled in at that point and will not
+          // contain any -1's. However, streams are explicitly allowed to start
+          // with an intra-only frame, so long as they don't then signal a
+          // reference to a slot that hasn't been set yet. That's what we are
+          // checking here.
+          if (idx == -1)
+            aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
+                               "Inter frame requests nonexistent reference");
+
+          RefBuffer *const ref_frame = &cm->frame_refs[i];
+          ref_frame->idx = idx;
+          ref_frame->buf = &frame_bufs[idx].buf;
+          ref_frame->map_idx = ref;
+        } else {
+          ref = cm->frame_refs[i].map_idx;
+        }
 
-        RefBuffer *const ref_frame = &cm->frame_refs[i];
-        ref_frame->idx = idx;
-        ref_frame->buf = &frame_bufs[idx].buf;
-#if CONFIG_FRAME_SIGN_BIAS
-#if CONFIG_OBU
-        // NOTE: For the scenario of (cm->frame_type != S_FRAME),
-        // ref_frame_sign_bias will be reset based on frame offsets.
         cm->ref_frame_sign_bias[LAST_FRAME + i] = 0;
-#endif  // CONFIG_OBU
-#else   // !CONFIG_FRAME_SIGN_BIAS
-#if CONFIG_OBU
-        cm->ref_frame_sign_bias[LAST_FRAME + i] =
-            (cm->frame_type == S_FRAME) ? 0 : aom_rb_read_bit(rb);
-#else   // !CONFIG_OBU
-        cm->ref_frame_sign_bias[LAST_FRAME + i] = aom_rb_read_bit(rb);
-#endif  // CONFIG_OBU
-#endif  // CONFIG_FRAME_SIGN_BIAS
-#if CONFIG_REFERENCE_BUFFER
+
         if (cm->seq_params.frame_id_numbers_present_flag) {
-          int frame_id_length = cm->seq_params.frame_id_length_minus7 + 7;
-          int diff_len = cm->seq_params.delta_frame_id_length_minus2 + 2;
-          int delta_frame_id_minus1 = aom_rb_read_literal(rb, diff_len);
+          int frame_id_length = cm->seq_params.frame_id_length;
+          int diff_len = cm->seq_params.delta_frame_id_length;
+          int delta_frame_id_minus_1 = aom_rb_read_literal(rb, diff_len);
           int ref_frame_id =
-              ((cm->current_frame_id - (delta_frame_id_minus1 + 1) +
+              ((cm->current_frame_id - (delta_frame_id_minus_1 + 1) +
                 (1 << frame_id_length)) %
                (1 << frame_id_length));
-          /* Compare values derived from delta_frame_id_minus1 and
-           * refresh_frame_flags. Also, check valid for referencing */
+          // Compare values derived from delta_frame_id_minus_1 and
+          // refresh_frame_flags. Also, check valid for referencing
           if (ref_frame_id != cm->ref_frame_id[ref] ||
               cm->valid_for_referencing[ref] == 0)
             aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
                                "Reference buffer frame ID mismatch");
         }
-#endif  // CONFIG_REFERENCE_BUFFER
       }
 
-#if CONFIG_VAR_REFS
-      check_valid_ref_frames(cm);
-#endif  // CONFIG_VAR_REFS
-
-#if CONFIG_FRAME_SIZE
-      if (cm->error_resilient_mode == 0) {
+      if (!cm->error_resilient_mode && frame_size_override_flag) {
         setup_frame_size_with_refs(cm, rb);
       } else {
-        setup_frame_size(cm, rb);
+        setup_frame_size(cm, frame_size_override_flag, rb);
       }
-#else
-      setup_frame_size_with_refs(cm, rb);
-#endif
 
-#if CONFIG_AMVR
-      if (cm->seq_mv_precision_level == 2) {
-        cm->cur_frame_mv_precision_level = aom_rb_read_bit(rb) ? 0 : 1;
+      if (cm->cur_frame_force_integer_mv) {
+        cm->allow_high_precision_mv = 0;
       } else {
-        cm->cur_frame_mv_precision_level = cm->seq_mv_precision_level;
+        cm->allow_high_precision_mv = aom_rb_read_bit(rb);
       }
-#endif
-      cm->allow_high_precision_mv = aom_rb_read_bit(rb);
       cm->interp_filter = read_frame_interp_filter(rb);
-#if CONFIG_TEMPMV_SIGNALING
-      if (frame_might_use_prev_frame_mvs(cm))
-        cm->use_prev_frame_mvs = aom_rb_read_bit(rb);
+      cm->switchable_motion_mode = aom_rb_read_bit(rb);
+    }
+
+    cm->prev_frame = get_prev_frame(cm);
+    if (cm->primary_ref_frame != PRIMARY_REF_NONE &&
+        cm->frame_refs[cm->primary_ref_frame].idx < 0) {
+      aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
+                         "Reference frame containing this frame's initial "
+                         "frame context is unavailable.");
+    }
+
+    if (!cm->intra_only && pbi->need_resync != 1) {
+      if (frame_might_allow_ref_frame_mvs(cm))
+        cm->allow_ref_frame_mvs = aom_rb_read_bit(rb);
       else
-        cm->use_prev_frame_mvs = 0;
-#endif
-      for (i = 0; i < INTER_REFS_PER_FRAME; ++i) {
+        cm->allow_ref_frame_mvs = 0;
+
+      for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
         RefBuffer *const ref_buf = &cm->frame_refs[i];
-#if CONFIG_HIGHBITDEPTH
-        av1_setup_scale_factors_for_frame(
-            &ref_buf->sf, ref_buf->buf->y_crop_width,
-            ref_buf->buf->y_crop_height, cm->width, cm->height,
-            cm->use_highbitdepth);
-#else
         av1_setup_scale_factors_for_frame(
             &ref_buf->sf, ref_buf->buf->y_crop_width,
             ref_buf->buf->y_crop_height, cm->width, cm->height);
-#endif
+        if ((!av1_is_valid_scale(&ref_buf->sf)))
+          aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+                             "Reference frame has invalid dimensions");
       }
     }
   }
 
-#if CONFIG_FRAME_MARKER
-  if (cm->show_frame == 0) {
-    cm->frame_offset = cm->current_video_frame + aom_rb_read_literal(rb, 4);
-  } else {
-    cm->frame_offset = cm->current_video_frame;
-  }
   av1_setup_frame_buf_refs(cm);
 
-#if CONFIG_FRAME_SIGN_BIAS
-#if CONFIG_OBU
-  if (cm->frame_type != S_FRAME)
-#endif  // CONFIG_OBU
-    av1_setup_frame_sign_bias(cm);
-#define FRAME_SIGN_BIAS_DEBUG 0
-#if FRAME_SIGN_BIAS_DEBUG
-  {
-    printf("\n\nDECODER: Frame=%d, show_frame=%d:", cm->current_video_frame,
-           cm->show_frame);
-    MV_REFERENCE_FRAME ref_frame;
-    for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
-      printf(" sign_bias[%d]=%d", ref_frame,
-             cm->ref_frame_sign_bias[ref_frame]);
-    }
-    printf("\n");
-  }
-#endif  // FRAME_SIGN_BIAS_DEBUG
-#undef FRAME_SIGN_BIAS_DEBUG
-#endif  // CONFIG_FRAME_SIGN_BIAS
-#endif  // CONFIG_FRAME_MARKER
+  av1_setup_frame_sign_bias(cm);
 
-#if CONFIG_TEMPMV_SIGNALING
   cm->cur_frame->intra_only = cm->frame_type == KEY_FRAME || cm->intra_only;
-#endif
+  cm->cur_frame->frame_type = cm->frame_type;
 
-#if CONFIG_REFERENCE_BUFFER
   if (cm->seq_params.frame_id_numbers_present_flag) {
     /* If bitmask is set, update reference frame id values and
        mark frames as valid for reference */
-    int refresh_frame_flags =
-        cm->frame_type == KEY_FRAME ? 0xFF : pbi->refresh_frame_flags;
-    for (i = 0; i < REF_FRAMES; i++) {
+    int refresh_frame_flags = pbi->refresh_frame_flags;
+    for (int i = 0; i < REF_FRAMES; i++) {
       if ((refresh_frame_flags >> i) & 1) {
         cm->ref_frame_id[i] = cm->current_frame_id;
         cm->valid_for_referencing[i] = 1;
       }
     }
   }
-#endif  // CONFIG_REFERENCE_BUFFER
+
+  const int might_bwd_adapt =
+      !(cm->seq_params.reduced_still_picture_hdr) && !(cm->disable_cdf_update);
+  if (might_bwd_adapt) {
+    cm->refresh_frame_context = aom_rb_read_bit(rb)
+                                    ? REFRESH_FRAME_CONTEXT_DISABLED
+                                    : REFRESH_FRAME_CONTEXT_BACKWARD;
+  } else {
+    cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_DISABLED;
+  }
 
   get_frame_new_buffer(cm)->bit_depth = cm->bit_depth;
-  get_frame_new_buffer(cm)->color_space = cm->color_space;
-#if CONFIG_COLORSPACE_HEADERS
-  get_frame_new_buffer(cm)->transfer_function = cm->transfer_function;
+  get_frame_new_buffer(cm)->color_primaries = cm->color_primaries;
+  get_frame_new_buffer(cm)->transfer_characteristics =
+      cm->transfer_characteristics;
+  get_frame_new_buffer(cm)->matrix_coefficients = cm->matrix_coefficients;
+  get_frame_new_buffer(cm)->monochrome = cm->seq_params.monochrome;
   get_frame_new_buffer(cm)->chroma_sample_position = cm->chroma_sample_position;
-#endif
   get_frame_new_buffer(cm)->color_range = cm->color_range;
   get_frame_new_buffer(cm)->render_width = cm->render_width;
   get_frame_new_buffer(cm)->render_height = cm->render_height;
@@ -5034,22 +4103,10 @@ static size_t read_uncompressed_header(AV1Decoder *pbi,
                        " state");
   }
 
-  if (!cm->error_resilient_mode) {
-    cm->refresh_frame_context = aom_rb_read_bit(rb)
-                                    ? REFRESH_FRAME_CONTEXT_FORWARD
-                                    : REFRESH_FRAME_CONTEXT_BACKWARD;
-  } else {
-    cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_FORWARD;
-  }
-#if !CONFIG_NO_FRAME_CONTEXT_SIGNALING
-  // This flag will be overridden by the call to av1_setup_past_independence
-  // below, forcing the use of context 0 for those frame types.
-  cm->frame_context_idx = aom_rb_read_literal(rb, FRAME_CONTEXTS_LOG2);
-#endif
-
   // Generate next_ref_frame_map.
   lock_buffer_pool(pool);
-  for (mask = pbi->refresh_frame_flags; mask; mask >>= 1) {
+  int ref_index = 0;
+  for (int mask = pbi->refresh_frame_flags; mask; mask >>= 1) {
     if (mask & 1) {
       cm->next_ref_frame_map[ref_index] = cm->new_fb_idx;
       ++frame_bufs[cm->new_fb_idx].ref_count;
@@ -5072,461 +4129,185 @@ static size_t read_uncompressed_header(AV1Decoder *pbi,
   unlock_buffer_pool(pool);
   pbi->hold_ref_buf = 1;
 
-  if (frame_is_intra_only(cm) || cm->error_resilient_mode)
-    av1_setup_past_independence(cm);
+  if (cm->allow_intrabc) {
+    // Set parameters corresponding to no filtering.
+    struct loopfilter *lf = &cm->lf;
+    lf->filter_level[0] = 0;
+    lf->filter_level[1] = 0;
+    cm->cdef_bits = 0;
+    cm->cdef_strengths[0] = 0;
+    cm->nb_cdef_strengths = 1;
+    cm->cdef_uv_strengths[0] = 0;
+    cm->rst_info[0].frame_restoration_type = RESTORE_NONE;
+    cm->rst_info[1].frame_restoration_type = RESTORE_NONE;
+    cm->rst_info[2].frame_restoration_type = RESTORE_NONE;
+  }
 
-  setup_loopfilter(cm, rb);
+  read_tile_info(pbi, rb);
   setup_quantization(cm, rb);
   xd->bd = (int)cm->bit_depth;
 
-#if CONFIG_Q_ADAPT_PROBS
-  av1_default_coef_probs(cm);
-  if (cm->frame_type == KEY_FRAME || cm->error_resilient_mode ||
-      cm->reset_frame_context == RESET_FRAME_CONTEXT_ALL) {
-    for (i = 0; i < FRAME_CONTEXTS; ++i) cm->frame_contexts[i] = *cm->fc;
-  } else if (cm->reset_frame_context == RESET_FRAME_CONTEXT_CURRENT) {
-#if CONFIG_NO_FRAME_CONTEXT_SIGNALING
-    if (cm->frame_refs[0].idx <= 0) {
-      cm->frame_contexts[cm->frame_refs[0].idx] = *cm->fc;
-    }
-#else
-    cm->frame_contexts[cm->frame_context_idx] = *cm->fc;
-#endif  // CONFIG_NO_FRAME_CONTEXT_SIGNALING
+  if (cm->num_allocated_above_context_planes < av1_num_planes(cm) ||
+      cm->num_allocated_above_context_mi_col < cm->mi_cols ||
+      cm->num_allocated_above_contexts < cm->tile_rows) {
+    av1_free_above_context_buffers(cm, cm->num_allocated_above_contexts);
+    if (av1_alloc_above_context_buffers(cm, cm->tile_rows))
+      aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
+                         "Failed to allocate context buffers");
   }
-#endif  // CONFIG_Q_ADAPT_PROBS
 
-  setup_segmentation(cm, rb);
+  if (cm->primary_ref_frame == PRIMARY_REF_NONE) {
+    av1_setup_past_independence(cm);
+  }
 
-  {
-    struct segmentation *const seg = &cm->seg;
-    int segment_quantizer_active = 0;
-    for (i = 0; i < MAX_SEGMENTS; i++) {
-      if (segfeature_active(seg, i, SEG_LVL_ALT_Q)) {
-        segment_quantizer_active = 1;
-      }
-    }
+  setup_segmentation(cm, rb);
 
-    cm->delta_q_res = 1;
-#if CONFIG_EXT_DELTA_Q
-    cm->delta_lf_res = 1;
-    cm->delta_lf_present_flag = 0;
-#if CONFIG_LOOPFILTER_LEVEL
-    cm->delta_lf_multi = 0;
-#endif  // CONFIG_LOOPFILTER_LEVEL
-#endif
-    if (segment_quantizer_active == 0 && cm->base_qindex > 0) {
-      cm->delta_q_present_flag = aom_rb_read_bit(rb);
-    } else {
-      cm->delta_q_present_flag = 0;
-    }
-    if (cm->delta_q_present_flag) {
-      xd->prev_qindex = cm->base_qindex;
-      cm->delta_q_res = 1 << aom_rb_read_literal(rb, 2);
-#if CONFIG_EXT_DELTA_Q
-      assert(!segment_quantizer_active);
-      cm->delta_lf_present_flag = aom_rb_read_bit(rb);
-      if (cm->delta_lf_present_flag) {
-        xd->prev_delta_lf_from_base = 0;
-        cm->delta_lf_res = 1 << aom_rb_read_literal(rb, 2);
-#if CONFIG_LOOPFILTER_LEVEL
-        cm->delta_lf_multi = aom_rb_read_bit(rb);
-        for (int lf_id = 0; lf_id < FRAME_LF_COUNT; ++lf_id)
-          xd->prev_delta_lf[lf_id] = 0;
-#endif  // CONFIG_LOOPFILTER_LEVEL
-      }
-#endif  // CONFIG_EXT_DELTA_Q
+  cm->delta_q_res = 1;
+  cm->delta_lf_res = 1;
+  cm->delta_lf_present_flag = 0;
+  cm->delta_lf_multi = 0;
+  cm->delta_q_present_flag = cm->base_qindex > 0 ? aom_rb_read_bit(rb) : 0;
+  if (cm->delta_q_present_flag) {
+    xd->current_qindex = cm->base_qindex;
+    cm->delta_q_res = 1 << aom_rb_read_literal(rb, 2);
+    if (!cm->allow_intrabc) cm->delta_lf_present_flag = aom_rb_read_bit(rb);
+    if (cm->delta_lf_present_flag) {
+      cm->delta_lf_res = 1 << aom_rb_read_literal(rb, 2);
+      cm->delta_lf_multi = aom_rb_read_bit(rb);
+      av1_reset_loop_filter_delta(xd, av1_num_planes(cm));
     }
   }
-#if CONFIG_AMVR
-  xd->cur_frame_mv_precision_level = cm->cur_frame_mv_precision_level;
-#endif
 
-  for (i = 0; i < MAX_SEGMENTS; ++i) {
+  xd->cur_frame_force_integer_mv = cm->cur_frame_force_integer_mv;
+
+  for (int i = 0; i < MAX_SEGMENTS; ++i) {
     const int qindex = cm->seg.enabled
                            ? av1_get_qindex(&cm->seg, i, cm->base_qindex)
                            : cm->base_qindex;
     xd->lossless[i] = qindex == 0 && cm->y_dc_delta_q == 0 &&
-                      cm->uv_dc_delta_q == 0 && cm->uv_ac_delta_q == 0;
+                      cm->u_dc_delta_q == 0 && cm->u_ac_delta_q == 0 &&
+                      cm->v_dc_delta_q == 0 && cm->v_ac_delta_q == 0;
     xd->qindex[i] = qindex;
   }
-  cm->all_lossless = all_lossless(cm, xd);
+  cm->coded_lossless = is_coded_lossless(cm, xd);
+  cm->all_lossless = cm->coded_lossless && !av1_superres_scaled(cm);
   setup_segmentation_dequant(cm);
-#if CONFIG_CDEF
-  if (!cm->all_lossless) {
-    setup_cdef(cm, rb);
+  if (cm->coded_lossless) {
+    cm->lf.filter_level[0] = 0;
+    cm->lf.filter_level[1] = 0;
   }
-#endif
-#if CONFIG_LOOP_RESTORATION
-  decode_restoration_mode(cm, rb);
-#endif  // CONFIG_LOOP_RESTORATION
-  cm->tx_mode = read_tx_mode(cm, rb);
-  cm->reference_mode = read_frame_reference_mode(cm, rb);
-  if (cm->reference_mode != SINGLE_REFERENCE) setup_compound_reference_mode(cm);
-  read_compound_tools(cm, rb);
-
-#if CONFIG_EXT_TX
-  cm->reduced_tx_set_used = aom_rb_read_bit(rb);
-#endif  // CONFIG_EXT_TX
-
-#if CONFIG_ADAPT_SCAN
-  cm->use_adapt_scan = aom_rb_read_bit(rb);
-  // TODO(angiebird): call av1_init_scan_order only when use_adapt_scan
-  // switches from 1 to 0
-  if (cm->use_adapt_scan == 0) av1_init_scan_order(cm);
-#endif  // CONFIG_ADAPT_SCAN
-
-#if CONFIG_EXT_REFS || CONFIG_TEMPMV_SIGNALING
-  // NOTE(zoeliu): As cm->prev_frame can take neither a frame of
-  //               show_exisiting_frame=1, nor can it take a frame not used as
-  //               a reference, it is probable that by the time it is being
-  //               referred to, the frame buffer it originally points to may
-  //               already get expired and have been reassigned to the current
-  //               newly coded frame. Hence, we need to check whether this is
-  //               the case, and if yes, we have 2 choices:
-  //               (1) Simply disable the use of previous frame mvs; or
-  //               (2) Have cm->prev_frame point to one reference frame buffer,
-  //                   e.g. LAST_FRAME.
-  if (!dec_is_ref_frame_buf(pbi, cm->prev_frame)) {
-    // Reassign the LAST_FRAME buffer to cm->prev_frame.
-    cm->prev_frame =
-        cm->frame_refs[LAST_FRAME - LAST_FRAME].idx != INVALID_IDX
-            ? &cm->buffer_pool
-                   ->frame_bufs[cm->frame_refs[LAST_FRAME - LAST_FRAME].idx]
-            : NULL;
-  }
-#endif  // CONFIG_EXT_REFS || CONFIG_TEMPMV_SIGNALING
-
-#if CONFIG_TEMPMV_SIGNALING
-  if (cm->use_prev_frame_mvs && !frame_can_use_prev_frame_mvs(cm)) {
-    aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
-                       "Frame wrongly requests previous frame MVs");
+  if (cm->coded_lossless || !cm->seq_params.enable_cdef) {
+    cm->cdef_bits = 0;
+    cm->cdef_strengths[0] = 0;
+    cm->cdef_uv_strengths[0] = 0;
   }
-#else
-  cm->use_prev_frame_mvs = !cm->error_resilient_mode && cm->prev_frame &&
-#if CONFIG_FRAME_SUPERRES
-                           cm->width == cm->last_width &&
-                           cm->height == cm->last_height &&
-#else
-                           cm->width == cm->prev_frame->buf.y_crop_width &&
-                           cm->height == cm->prev_frame->buf.y_crop_height &&
-#endif  // CONFIG_FRAME_SUPERRES
-                           !cm->last_intra_only && cm->last_show_frame &&
-                           (cm->last_frame_type != KEY_FRAME);
-#endif  // CONFIG_TEMPMV_SIGNALING
-
-#if CONFIG_GLOBAL_MOTION
-  if (!frame_is_intra_only(cm)) read_global_motion(cm, rb);
-#endif
-
-  read_tile_info(pbi, rb);
-  if (use_compressed_header(cm)) {
-    sz = aom_rb_read_literal(rb, 16);
-    if (sz == 0)
-      aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
-                         "Invalid header size");
-  } else {
-    sz = 0;
+  if (cm->all_lossless || !cm->seq_params.enable_restoration) {
+    cm->rst_info[0].frame_restoration_type = RESTORE_NONE;
+    cm->rst_info[1].frame_restoration_type = RESTORE_NONE;
+    cm->rst_info[2].frame_restoration_type = RESTORE_NONE;
   }
-  return sz;
-}
+  setup_loopfilter(cm, rb);
 
-#if CONFIG_SUPERTX
-static void read_supertx_probs(FRAME_CONTEXT *fc, aom_reader *r) {
-  int i, j;
-  if (aom_read(r, GROUP_DIFF_UPDATE_PROB, ACCT_STR)) {
-    for (i = 0; i < PARTITION_SUPERTX_CONTEXTS; ++i) {
-      for (j = TX_8X8; j < TX_SIZES; ++j) {
-        av1_diff_update_prob(r, &fc->supertx_prob[i][j], ACCT_STR);
-      }
-    }
+  if (!cm->coded_lossless && cm->seq_params.enable_cdef) {
+    setup_cdef(cm, rb);
+  }
+  if (!cm->all_lossless && cm->seq_params.enable_restoration) {
+    decode_restoration_mode(cm, rb);
   }
-}
-#endif  // CONFIG_SUPERTX
-
-static int read_compressed_header(AV1Decoder *pbi, const uint8_t *data,
-                                  size_t partition_size) {
-#if CONFIG_RESTRICT_COMPRESSED_HDR
-  (void)pbi;
-  (void)data;
-  (void)partition_size;
-  return 0;
-#else
-  AV1_COMMON *const cm = &pbi->common;
-#if CONFIG_SUPERTX
-  MACROBLOCKD *const xd = &pbi->mb;
-#endif
-  aom_reader r;
-#if !CONFIG_NEW_MULTISYMBOL
-  FRAME_CONTEXT *const fc = cm->fc;
-  int i;
-#endif
-
-#if CONFIG_ANS && ANS_MAX_SYMBOLS
-  r.window_size = 1 << cm->ans_window_size_log2;
-#endif
-  if (aom_reader_init(&r, data, partition_size, pbi->decrypt_cb,
-                      pbi->decrypt_state))
-    aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
-                       "Failed to allocate bool decoder 0");
 
-#if CONFIG_RECT_TX_EXT && (CONFIG_EXT_TX || CONFIG_VAR_TX)
-  if (cm->tx_mode == TX_MODE_SELECT)
-    av1_diff_update_prob(&r, &fc->quarter_tx_size_prob, ACCT_STR);
-#endif
+  cm->tx_mode = read_tx_mode(cm, rb);
+  cm->reference_mode = read_frame_reference_mode(cm, rb);
+  if (cm->reference_mode != SINGLE_REFERENCE) setup_compound_reference_mode(cm);
 
-#if CONFIG_LV_MAP && !LV_MAP_PROB
-  av1_read_txb_probs(fc, cm->tx_mode, &r, &cm->counts);
-#endif  // CONFIG_LV_MAP && !LV_MAP_PROB
-
-#if !CONFIG_NEW_MULTISYMBOL
-#if CONFIG_VAR_TX
-  if (cm->tx_mode == TX_MODE_SELECT)
-    for (i = 0; i < TXFM_PARTITION_CONTEXTS; ++i)
-      av1_diff_update_prob(&r, &fc->txfm_partition_prob[i], ACCT_STR);
-#endif  // CONFIG_VAR_TX
-  for (i = 0; i < SKIP_CONTEXTS; ++i)
-    av1_diff_update_prob(&r, &fc->skip_probs[i], ACCT_STR);
-#endif
+  av1_setup_skip_mode_allowed(cm);
+  cm->skip_mode_flag = cm->is_skip_mode_allowed ? aom_rb_read_bit(rb) : 0;
 
-  if (!frame_is_intra_only(cm)) {
-#if !CONFIG_NEW_MULTISYMBOL
-    read_inter_mode_probs(fc, &r);
-#endif
+  if (frame_might_allow_warped_motion(cm))
+    cm->allow_warped_motion = aom_rb_read_bit(rb);
+  else
+    cm->allow_warped_motion = 0;
 
-#if CONFIG_INTERINTRA
-    if (cm->reference_mode != COMPOUND_REFERENCE &&
-        cm->allow_interintra_compound) {
-#if !CONFIG_NEW_MULTISYMBOL
-      for (i = 0; i < BLOCK_SIZE_GROUPS; i++) {
-        if (is_interintra_allowed_bsize_group(i)) {
-          av1_diff_update_prob(&r, &fc->interintra_prob[i], ACCT_STR);
-        }
-      }
-#endif
-#if CONFIG_WEDGE && !CONFIG_NEW_MULTISYMBOL
-#if CONFIG_EXT_PARTITION_TYPES
-      int block_sizes_to_update = BLOCK_SIZES_ALL;
-#else
-      int block_sizes_to_update = BLOCK_SIZES;
-#endif
-      for (i = 0; i < block_sizes_to_update; i++) {
-        if (is_interintra_allowed_bsize(i) && is_interintra_wedge_used(i)) {
-          av1_diff_update_prob(&r, &fc->wedge_interintra_prob[i], ACCT_STR);
-        }
-      }
-#endif  // CONFIG_WEDGE
-    }
-#endif  // CONFIG_INTERINTRA
+  cm->reduced_tx_set_used = aom_rb_read_bit(rb);
 
-#if !CONFIG_NEW_MULTISYMBOL
-    for (i = 0; i < INTRA_INTER_CONTEXTS; i++)
-      av1_diff_update_prob(&r, &fc->intra_inter_prob[i], ACCT_STR);
-#endif
+  if (cm->allow_ref_frame_mvs && !frame_might_allow_ref_frame_mvs(cm)) {
+    aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
+                       "Frame wrongly requests reference frame MVs");
+  }
 
-#if !CONFIG_NEW_MULTISYMBOL
-    read_frame_reference_mode_probs(cm, &r);
-#endif
+  if (!frame_is_intra_only(cm)) read_global_motion(cm, rb);
 
-#if CONFIG_COMPOUND_SINGLEREF
-    for (i = 0; i < COMP_INTER_MODE_CONTEXTS; i++)
-      av1_diff_update_prob(&r, &fc->comp_inter_mode_prob[i], ACCT_STR);
-#endif  // CONFIG_COMPOUND_SINGLEREF
+  cm->cur_frame->film_grain_params_present = cm->film_grain_params_present;
+  read_film_grain(cm, rb);
 
-#if !CONFIG_NEW_MULTISYMBOL
-#if CONFIG_AMVR
-    if (cm->cur_frame_mv_precision_level == 0) {
-#endif
-      for (i = 0; i < NMV_CONTEXTS; ++i)
-        read_mv_probs(&fc->nmvc[i], cm->allow_high_precision_mv, &r);
-#if CONFIG_AMVR
-    }
-#endif
-#endif
-#if CONFIG_SUPERTX
-    if (!xd->lossless[0]) read_supertx_probs(fc, &r);
-#endif
+#if EXT_TILE_DEBUG
+  if (pbi->ext_tile_debug && cm->large_scale_tile) {
+    read_ext_tile_info(pbi, rb);
+    av1_set_single_tile_decoding_mode(cm);
   }
-
-  return aom_reader_has_error(&r);
-#endif  // CONFIG_RESTRICT_COMPRESSED_HDR
-}
-
-#ifdef NDEBUG
-#define debug_check_frame_counts(cm) (void)0
-#else  // !NDEBUG
-// Counts should only be incremented when frame_parallel_decoding_mode and
-// error_resilient_mode are disabled.
-static void debug_check_frame_counts(const AV1_COMMON *const cm) {
-  FRAME_COUNTS zero_counts;
-  av1_zero(zero_counts);
-  assert(cm->refresh_frame_context != REFRESH_FRAME_CONTEXT_BACKWARD ||
-         cm->error_resilient_mode);
-  assert(!memcmp(cm->counts.partition, zero_counts.partition,
-                 sizeof(cm->counts.partition)));
-  assert(!memcmp(cm->counts.switchable_interp, zero_counts.switchable_interp,
-                 sizeof(cm->counts.switchable_interp)));
-  assert(!memcmp(cm->counts.inter_compound_mode,
-                 zero_counts.inter_compound_mode,
-                 sizeof(cm->counts.inter_compound_mode)));
-#if CONFIG_INTERINTRA
-  assert(!memcmp(cm->counts.interintra, zero_counts.interintra,
-                 sizeof(cm->counts.interintra)));
-#if CONFIG_WEDGE
-  assert(!memcmp(cm->counts.wedge_interintra, zero_counts.wedge_interintra,
-                 sizeof(cm->counts.wedge_interintra)));
-#endif  // CONFIG_WEDGE
-#endif  // CONFIG_INTERINTRA
-  assert(!memcmp(cm->counts.compound_interinter,
-                 zero_counts.compound_interinter,
-                 sizeof(cm->counts.compound_interinter)));
-#if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
-  assert(!memcmp(cm->counts.motion_mode, zero_counts.motion_mode,
-                 sizeof(cm->counts.motion_mode)));
-#endif  // CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
-#if CONFIG_NCOBMC_ADAPT_WEIGHT && CONFIG_MOTION_VAR
-  assert(!memcmp(cm->counts.ncobmc_mode, zero_counts.ncobmc_mode,
-                 sizeof(cm->counts.ncobmc_mode)));
-#endif
-  assert(!memcmp(cm->counts.intra_inter, zero_counts.intra_inter,
-                 sizeof(cm->counts.intra_inter)));
-#if CONFIG_COMPOUND_SINGLEREF
-  assert(!memcmp(cm->counts.comp_inter_mode, zero_counts.comp_inter_mode,
-                 sizeof(cm->counts.comp_inter_mode)));
-#endif  // CONFIG_COMPOUND_SINGLEREF
-  assert(!memcmp(cm->counts.comp_inter, zero_counts.comp_inter,
-                 sizeof(cm->counts.comp_inter)));
-#if CONFIG_EXT_COMP_REFS
-  assert(!memcmp(cm->counts.comp_ref_type, zero_counts.comp_ref_type,
-                 sizeof(cm->counts.comp_ref_type)));
-  assert(!memcmp(cm->counts.uni_comp_ref, zero_counts.uni_comp_ref,
-                 sizeof(cm->counts.uni_comp_ref)));
-#endif  // CONFIG_EXT_COMP_REFS
-  assert(!memcmp(cm->counts.single_ref, zero_counts.single_ref,
-                 sizeof(cm->counts.single_ref)));
-  assert(!memcmp(cm->counts.comp_ref, zero_counts.comp_ref,
-                 sizeof(cm->counts.comp_ref)));
-#if CONFIG_EXT_REFS
-  assert(!memcmp(cm->counts.comp_bwdref, zero_counts.comp_bwdref,
-                 sizeof(cm->counts.comp_bwdref)));
-#endif  // CONFIG_EXT_REFS
-  assert(!memcmp(&cm->counts.tx_size, &zero_counts.tx_size,
-                 sizeof(cm->counts.tx_size)));
-  assert(!memcmp(cm->counts.skip, zero_counts.skip, sizeof(cm->counts.skip)));
-  assert(
-      !memcmp(&cm->counts.mv[0], &zero_counts.mv[0], sizeof(cm->counts.mv[0])));
-  assert(
-      !memcmp(&cm->counts.mv[1], &zero_counts.mv[1], sizeof(cm->counts.mv[0])));
+#endif  // EXT_TILE_DEBUG
+  return 0;
 }
-#endif  // NDEBUG
 
-static struct aom_read_bit_buffer *init_read_bit_buffer(
+struct aom_read_bit_buffer *av1_init_read_bit_buffer(
     AV1Decoder *pbi, struct aom_read_bit_buffer *rb, const uint8_t *data,
-    const uint8_t *data_end, uint8_t clear_data[MAX_AV1_HEADER_SIZE]) {
+    const uint8_t *data_end) {
   rb->bit_offset = 0;
   rb->error_handler = error_handler;
   rb->error_handler_data = &pbi->common;
-  if (pbi->decrypt_cb) {
-    const int n = (int)AOMMIN(MAX_AV1_HEADER_SIZE, data_end - data);
-    pbi->decrypt_cb(pbi->decrypt_state, data, clear_data, n);
-    rb->bit_buffer = clear_data;
-    rb->bit_buffer_end = clear_data + n;
-  } else {
-    rb->bit_buffer = data;
-    rb->bit_buffer_end = data_end;
-  }
+  rb->bit_buffer = data;
+  rb->bit_buffer_end = data_end;
   return rb;
 }
 
-//------------------------------------------------------------------------------
-
-void av1_read_frame_size(struct aom_read_bit_buffer *rb, int *width,
-                         int *height) {
-  *width = aom_rb_read_literal(rb, 16) + 1;
-  *height = aom_rb_read_literal(rb, 16) + 1;
+void av1_read_frame_size(struct aom_read_bit_buffer *rb, int num_bits_width,
+                         int num_bits_height, int *width, int *height) {
+  *width = aom_rb_read_literal(rb, num_bits_width) + 1;
+  *height = aom_rb_read_literal(rb, num_bits_height) + 1;
 }
 
 BITSTREAM_PROFILE av1_read_profile(struct aom_read_bit_buffer *rb) {
-  int profile = aom_rb_read_bit(rb);
-  profile |= aom_rb_read_bit(rb) << 1;
-  if (profile > 2) profile += aom_rb_read_bit(rb);
+  int profile = aom_rb_read_literal(rb, PROFILE_BITS);
   return (BITSTREAM_PROFILE)profile;
 }
 
-static void make_update_tile_list_dec(AV1Decoder *pbi, int tile_rows,
-                                      int tile_cols, FRAME_CONTEXT *ec_ctxs[]) {
-  int i;
-  for (i = 0; i < tile_rows * tile_cols; ++i)
-    ec_ctxs[i] = &pbi->tile_data[i].tctx;
-}
-
-#if CONFIG_FRAME_SUPERRES
 void superres_post_decode(AV1Decoder *pbi) {
   AV1_COMMON *const cm = &pbi->common;
   BufferPool *const pool = cm->buffer_pool;
 
-  if (av1_superres_unscaled(cm)) return;
+  if (!av1_superres_scaled(cm)) return;
+  assert(!cm->all_lossless);
 
   lock_buffer_pool(pool);
   av1_superres_upscale(cm, pool);
   unlock_buffer_pool(pool);
 }
-#endif  // CONFIG_FRAME_SUPERRES
-
-static void dec_setup_frame_boundary_info(AV1_COMMON *const cm) {
-// Note: When LOOPFILTERING_ACROSS_TILES is enabled, we need to clear the
-// boundary information every frame, since the tile boundaries may
-// change every frame (particularly when dependent-horztiles is also
-// enabled); when it is disabled, the only information stored is the frame
-// boundaries, which only depend on the frame size.
-#if !CONFIG_LOOPFILTERING_ACROSS_TILES
-  if (cm->width != cm->last_width || cm->height != cm->last_height)
-#endif  // CONFIG_LOOPFILTERING_ACROSS_TILES
-  {
-    int row, col;
-    for (row = 0; row < cm->mi_rows; ++row) {
-      MODE_INFO *mi = cm->mi + row * cm->mi_stride;
-      for (col = 0; col < cm->mi_cols; ++col) {
-        mi->mbmi.boundary_info = 0;
-        mi++;
-      }
-    }
-    av1_setup_frame_boundary_info(cm);
-  }
-}
 
-size_t av1_decode_frame_headers_and_setup(AV1Decoder *pbi, const uint8_t *data,
-                                          const uint8_t *data_end,
-                                          const uint8_t **p_data_end) {
+int av1_decode_frame_headers_and_setup(AV1Decoder *pbi,
+                                       struct aom_read_bit_buffer *rb,
+                                       const uint8_t *data,
+                                       const uint8_t **p_data_end,
+                                       int trailing_bits_present) {
   AV1_COMMON *const cm = &pbi->common;
+  const int num_planes = av1_num_planes(cm);
   MACROBLOCKD *const xd = &pbi->mb;
-  struct aom_read_bit_buffer rb;
-  uint8_t clear_data[MAX_AV1_HEADER_SIZE];
-  size_t first_partition_size;
-  YV12_BUFFER_CONFIG *new_fb;
-#if CONFIG_EXT_REFS || CONFIG_TEMPMV_SIGNALING
-  RefBuffer *last_fb_ref_buf = &cm->frame_refs[LAST_FRAME - LAST_FRAME];
-#endif  // CONFIG_EXT_REFS || CONFIG_TEMPMV_SIGNALING
-
-#if CONFIG_ADAPT_SCAN
-  av1_deliver_eob_threshold(cm, xd);
-#endif
+
 #if CONFIG_BITSTREAM_DEBUG
   bitstream_queue_set_frame_read(cm->current_video_frame * 2 + cm->show_frame);
 #endif
+#if CONFIG_MISMATCH_DEBUG
+  mismatch_move_frame_idx_r();
+#endif
 
-#if CONFIG_GLOBAL_MOTION
-  int i;
-  for (i = LAST_FRAME; i <= ALTREF_FRAME; ++i) {
+  for (int i = LAST_FRAME; i <= ALTREF_FRAME; ++i) {
     cm->global_motion[i] = default_warp_params;
     cm->cur_frame->global_motion[i] = default_warp_params;
   }
   xd->global_motion = cm->global_motion;
-#endif  // CONFIG_GLOBAL_MOTION
 
-  first_partition_size = read_uncompressed_header(
-      pbi, init_read_bit_buffer(pbi, &rb, data, data_end, clear_data));
+  read_uncompressed_header(pbi, rb);
+
+  if (trailing_bits_present) av1_check_trailing_bits(pbi, rb);
 
-#if CONFIG_EXT_TILE
   // If cm->single_tile_decoding = 0, the independent decoding of a single tile
   // or a section of a frame is not allowed.
   if (!cm->single_tile_decoding &&
@@ -5534,268 +4315,160 @@ size_t av1_decode_frame_headers_and_setup(AV1Decoder *pbi, const uint8_t *data,
     pbi->dec_tile_row = -1;
     pbi->dec_tile_col = -1;
   }
-#endif  // CONFIG_EXT_TILE
 
-  pbi->first_partition_size = first_partition_size;
-  pbi->uncomp_hdr_size = aom_rb_bytes_read(&rb);
-  new_fb = get_frame_new_buffer(cm);
+  pbi->uncomp_hdr_size = aom_rb_bytes_read(rb);
+  YV12_BUFFER_CONFIG *new_fb = get_frame_new_buffer(cm);
   xd->cur_buf = new_fb;
-#if CONFIG_INTRABC
-#if CONFIG_HIGHBITDEPTH
-  av1_setup_scale_factors_for_frame(
-      &xd->sf_identity, xd->cur_buf->y_crop_width, xd->cur_buf->y_crop_height,
-      xd->cur_buf->y_crop_width, xd->cur_buf->y_crop_height,
-      cm->use_highbitdepth);
-#else
-  av1_setup_scale_factors_for_frame(
-      &xd->sf_identity, xd->cur_buf->y_crop_width, xd->cur_buf->y_crop_height,
-      xd->cur_buf->y_crop_width, xd->cur_buf->y_crop_height);
-#endif  // CONFIG_HIGHBITDEPTH
-#endif  // CONFIG_INTRABC
+  if (av1_allow_intrabc(cm)) {
+    av1_setup_scale_factors_for_frame(
+        &cm->sf_identity, xd->cur_buf->y_crop_width, xd->cur_buf->y_crop_height,
+        xd->cur_buf->y_crop_width, xd->cur_buf->y_crop_height);
+  }
 
   if (cm->show_existing_frame) {
     // showing a frame directly
-    *p_data_end = data + aom_rb_bytes_read(&rb);
+    *p_data_end = data + aom_rb_bytes_read(rb);
+    if (cm->reset_decoder_state) {
+      // Use the default frame context values.
+      *cm->fc = cm->frame_contexts[FRAME_CONTEXT_DEFAULTS];
+      if (!cm->fc->initialized)
+        aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
+                           "Uninitialized entropy context.");
+    }
     return 0;
   }
 
-  data += aom_rb_bytes_read(&rb);
-  if (first_partition_size)
-    if (!read_is_valid(data, first_partition_size, data_end))
-      aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
-                         "Truncated packet or corrupt header length");
-
   cm->setup_mi(cm);
 
-#if CONFIG_EXT_REFS || CONFIG_TEMPMV_SIGNALING
-  // NOTE(zoeliu): As cm->prev_frame can take neither a frame of
-  //               show_exisiting_frame=1, nor can it take a frame not used as
-  //               a reference, it is probable that by the time it is being
-  //               referred to, the frame buffer it originally points to may
-  //               already get expired and have been reassigned to the current
-  //               newly coded frame. Hence, we need to check whether this is
-  //               the case, and if yes, we have 2 choices:
-  //               (1) Simply disable the use of previous frame mvs; or
-  //               (2) Have cm->prev_frame point to one reference frame buffer,
-  //                   e.g. LAST_FRAME.
-  if (!dec_is_ref_frame_buf(pbi, cm->prev_frame)) {
-    // Reassign the LAST_FRAME buffer to cm->prev_frame.
-    cm->prev_frame = last_fb_ref_buf->idx != INVALID_IDX
-                         ? &cm->buffer_pool->frame_bufs[last_fb_ref_buf->idx]
-                         : NULL;
-  }
-#endif  // CONFIG_EXT_REFS || CONFIG_TEMPMV_SIGNALING
-
-#if CONFIG_TEMPMV_SIGNALING
-  if (cm->use_prev_frame_mvs && !frame_can_use_prev_frame_mvs(cm)) {
-    aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
-                       "Frame wrongly requests previous frame MVs");
-  }
-#else
-  cm->use_prev_frame_mvs = !cm->error_resilient_mode && cm->prev_frame &&
-#if CONFIG_FRAME_SUPERRES
-                           cm->width == cm->last_width &&
-                           cm->height == cm->last_height &&
-#else
-                           cm->width == cm->prev_frame->buf.y_crop_width &&
-                           cm->height == cm->prev_frame->buf.y_crop_height &&
-#endif  // CONFIG_FRAME_SUPERRES
-                           !cm->last_intra_only && cm->last_show_frame &&
-                           (cm->last_frame_type != KEY_FRAME);
-#endif  // CONFIG_TEMPMV_SIGNALING
-
-#if CONFIG_MFMV
+  cm->current_frame_seg_map = cm->cur_frame->seg_map;
+
   av1_setup_motion_field(cm);
-#endif  // CONFIG_MFMV
 
-  av1_setup_block_planes(xd, cm->subsampling_x, cm->subsampling_y);
-#if CONFIG_NO_FRAME_CONTEXT_SIGNALING
-  if (cm->error_resilient_mode || frame_is_intra_only(cm)) {
+  av1_setup_block_planes(xd, cm->subsampling_x, cm->subsampling_y, num_planes);
+  if (cm->primary_ref_frame == PRIMARY_REF_NONE) {
     // use the default frame context values
     *cm->fc = cm->frame_contexts[FRAME_CONTEXT_DEFAULTS];
-    cm->pre_fc = &cm->frame_contexts[FRAME_CONTEXT_DEFAULTS];
   } else {
-    *cm->fc = cm->frame_contexts[cm->frame_refs[0].idx];
-    cm->pre_fc = &cm->frame_contexts[cm->frame_refs[0].idx];
+    *cm->fc = cm->frame_contexts[cm->frame_refs[cm->primary_ref_frame].idx];
   }
-#else
-  *cm->fc = cm->frame_contexts[cm->frame_context_idx];
-  cm->pre_fc = &cm->frame_contexts[cm->frame_context_idx];
-#endif  // CONFIG_NO_FRAME_CONTEXT_SIGNALING
   if (!cm->fc->initialized)
     aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
                        "Uninitialized entropy context.");
 
-  av1_zero(cm->counts);
-
   xd->corrupted = 0;
-  if (first_partition_size) {
-    new_fb->corrupted = read_compressed_header(pbi, data, first_partition_size);
-    if (new_fb->corrupted)
-      aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
-                         "Decode failed. Frame data header is corrupted.");
-  }
-  return first_partition_size;
+  return 0;
 }
 
-void av1_decode_tg_tiles_and_wrapup(AV1Decoder *pbi, const uint8_t *data,
-                                    const uint8_t *data_end,
-                                    const uint8_t **p_data_end, int startTile,
-                                    int endTile, int initialize_flag) {
+// Once-per-frame initialization
+static void setup_frame_info(AV1Decoder *pbi) {
   AV1_COMMON *const cm = &pbi->common;
-  MACROBLOCKD *const xd = &pbi->mb;
-  int context_updated = 0;
 
-#if CONFIG_LOOP_RESTORATION
   if (cm->rst_info[0].frame_restoration_type != RESTORE_NONE ||
       cm->rst_info[1].frame_restoration_type != RESTORE_NONE ||
       cm->rst_info[2].frame_restoration_type != RESTORE_NONE) {
     av1_alloc_restoration_buffers(cm);
   }
-#endif
-
-#if !CONFIG_LOOPFILTER_LEVEL
-  if (cm->lf.filter_level && !cm->skip_loop_filter) {
-    av1_loop_filter_frame_init(cm, cm->lf.filter_level, cm->lf.filter_level);
+  const int use_highbd = cm->use_highbitdepth ? 1 : 0;
+  const int buf_size = MC_TEMP_BUF_PELS << use_highbd;
+  if (pbi->td.mc_buf_size != buf_size) {
+    av1_free_mc_tmp_buf(&pbi->td, use_highbd);
+    allocate_mc_tmp_buf(cm, &pbi->td, buf_size, use_highbd);
   }
-#endif
+}
 
-  // If encoded in frame parallel mode, frame context is ready after decoding
-  // the frame header.
-  if (cm->frame_parallel_decode && initialize_flag &&
-      cm->refresh_frame_context != REFRESH_FRAME_CONTEXT_BACKWARD) {
-    AVxWorker *const worker = pbi->frame_worker_owner;
-    FrameWorkerData *const frame_worker_data = worker->data1;
-    if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_FORWARD) {
-      context_updated = 1;
-#if CONFIG_NO_FRAME_CONTEXT_SIGNALING
-      cm->frame_contexts[cm->new_fb_idx] = *cm->fc;
-#else
-      cm->frame_contexts[cm->frame_context_idx] = *cm->fc;
-#endif  // CONFIG_NO_FRAME_CONTEXT_SIGNALING
-    }
-    av1_frameworker_lock_stats(worker);
-    pbi->cur_buf->row = -1;
-    pbi->cur_buf->col = -1;
-    frame_worker_data->frame_context_ready = 1;
-    // Signal the main thread that context is ready.
-    av1_frameworker_signal_stats(worker);
-    av1_frameworker_unlock_stats(worker);
-  }
-
-  dec_setup_frame_boundary_info(cm);
-
-  if (pbi->max_threads > 1 && !CONFIG_CB4X4 &&
-#if CONFIG_EXT_TILE
-      pbi->dec_tile_col < 0 &&  // Decoding all columns
-#endif                          // CONFIG_EXT_TILE
-      cm->tile_cols > 1) {
-    // Multi-threaded tile decoder
-    *p_data_end =
-        decode_tiles_mt(pbi, data + pbi->first_partition_size, data_end);
-    if (!xd->corrupted) {
-      if (!cm->skip_loop_filter) {
-// If multiple threads are used to decode tiles, then we use those
-// threads to do parallel loopfiltering.
-#if CONFIG_LOOPFILTER_LEVEL
-        av1_loop_filter_frame_mt(
-            (YV12_BUFFER_CONFIG *)xd->cur_buf, cm, pbi->mb.plane,
-            cm->lf.filter_level[0], cm->lf.filter_level[1], 0, 0,
-            pbi->tile_workers, pbi->num_tile_workers, &pbi->lf_row_sync);
-#else
-        av1_loop_filter_frame_mt((YV12_BUFFER_CONFIG *)xd->cur_buf, cm,
-                                 pbi->mb.plane, cm->lf.filter_level, 0, 0,
-                                 pbi->tile_workers, pbi->num_tile_workers,
-                                 &pbi->lf_row_sync);
-#endif  // CONFIG_LOOPFILTER_LEVEL
-      }
-    } else {
-      aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
-                         "Decode failed. Frame data is corrupted.");
-    }
-  } else {
-#if CONFIG_OBU
-    *p_data_end = decode_tiles(pbi, data, data_end, startTile, endTile);
-#else
-    *p_data_end = decode_tiles(
-        pbi, data + pbi->uncomp_hdr_size + pbi->first_partition_size, data_end,
-        startTile, endTile);
-#endif
-  }
+void av1_decode_tg_tiles_and_wrapup(AV1Decoder *pbi, const uint8_t *data,
+                                    const uint8_t *data_end,
+                                    const uint8_t **p_data_end, int start_tile,
+                                    int end_tile, int initialize_flag) {
+  AV1_COMMON *const cm = &pbi->common;
+  MACROBLOCKD *const xd = &pbi->mb;
+  const int tile_count_tg = end_tile - start_tile + 1;
 
-  if (endTile != cm->tile_rows * cm->tile_cols - 1) {
-    return;
-  }
+  if (initialize_flag) setup_frame_info(pbi);
 
-#if CONFIG_STRIPED_LOOP_RESTORATION
-  if (cm->rst_info[0].frame_restoration_type != RESTORE_NONE ||
-      cm->rst_info[1].frame_restoration_type != RESTORE_NONE ||
-      cm->rst_info[2].frame_restoration_type != RESTORE_NONE) {
-    av1_loop_restoration_save_boundary_lines(&pbi->cur_buf->buf, cm);
-  }
-#endif
+  if (pbi->max_threads > 1 && tile_count_tg > 1 && !cm->large_scale_tile)
+    *p_data_end = decode_tiles_mt(pbi, data, data_end, start_tile, end_tile);
+  else
+    *p_data_end = decode_tiles(pbi, data, data_end, start_tile, end_tile);
 
-#if CONFIG_CDEF
-  if (!cm->skip_loop_filter && !cm->all_lossless) {
-    av1_cdef_frame(&pbi->cur_buf->buf, cm, &pbi->mb);
+  const int num_planes = av1_num_planes(cm);
+  // If the bit stream is monochrome, set the U and V buffers to a constant.
+  if (num_planes < 3) set_planes_to_neutral_grey(cm, xd->cur_buf, 1);
+
+  if (end_tile != cm->tile_rows * cm->tile_cols - 1) {
+    return;
   }
-#endif  // CONFIG_CDEF
 
-#if CONFIG_FRAME_SUPERRES
-  superres_post_decode(pbi);
-#endif  // CONFIG_FRAME_SUPERRES
+  if (!cm->allow_intrabc && !cm->single_tile_decoding) {
+    if (cm->lf.filter_level[0] || cm->lf.filter_level[1]) {
+#if LOOP_FILTER_BITMASK
+      av1_loop_filter_frame(get_frame_new_buffer(cm), cm, &pbi->mb, 0,
+                            num_planes, 0);
+#else
+      if (pbi->num_workers > 1) {
+        av1_loop_filter_frame_mt(get_frame_new_buffer(cm), cm, &pbi->mb, 0,
+                                 num_planes, 0, pbi->tile_workers,
+                                 pbi->num_workers, &pbi->lf_row_sync);
+      } else {
+        av1_loop_filter_frame(get_frame_new_buffer(cm), cm, &pbi->mb, 0,
+                              num_planes, 0);
+      }
+#endif
+    }
 
-#if CONFIG_LOOP_RESTORATION
-  if (cm->rst_info[0].frame_restoration_type != RESTORE_NONE ||
-      cm->rst_info[1].frame_restoration_type != RESTORE_NONE ||
-      cm->rst_info[2].frame_restoration_type != RESTORE_NONE) {
-    aom_extend_frame_borders((YV12_BUFFER_CONFIG *)xd->cur_buf);
-    av1_loop_restoration_frame((YV12_BUFFER_CONFIG *)xd->cur_buf, cm,
-                               cm->rst_info, 7, 0, NULL);
+    const int do_loop_restoration =
+        cm->rst_info[0].frame_restoration_type != RESTORE_NONE ||
+        cm->rst_info[1].frame_restoration_type != RESTORE_NONE ||
+        cm->rst_info[2].frame_restoration_type != RESTORE_NONE;
+    const int do_cdef =
+        !cm->skip_loop_filter && !cm->coded_lossless &&
+        (cm->cdef_bits || cm->cdef_strengths[0] || cm->cdef_uv_strengths[0]);
+    const int do_superres = av1_superres_scaled(cm);
+    const int optimized_loop_restoration = !do_cdef && !do_superres;
+
+    if (!optimized_loop_restoration) {
+      if (do_loop_restoration)
+        av1_loop_restoration_save_boundary_lines(&pbi->cur_buf->buf, cm, 0);
+
+      if (do_cdef) av1_cdef_frame(&pbi->cur_buf->buf, cm, &pbi->mb);
+
+      superres_post_decode(pbi);
+
+      if (do_loop_restoration) {
+        av1_loop_restoration_save_boundary_lines(&pbi->cur_buf->buf, cm, 1);
+        if (pbi->num_workers > 1) {
+          av1_loop_restoration_filter_frame_mt(
+              (YV12_BUFFER_CONFIG *)xd->cur_buf, cm, optimized_loop_restoration,
+              pbi->tile_workers, pbi->num_workers, &pbi->lr_row_sync,
+              &pbi->lr_ctxt);
+        } else {
+          av1_loop_restoration_filter_frame((YV12_BUFFER_CONFIG *)xd->cur_buf,
+                                            cm, optimized_loop_restoration,
+                                            &pbi->lr_ctxt);
+        }
+      }
+    } else {
+      // In no cdef and no superres case. Provide an optimized version of
+      // loop_restoration_filter.
+      if (do_loop_restoration) {
+        if (pbi->num_workers > 1) {
+          av1_loop_restoration_filter_frame_mt(
+              (YV12_BUFFER_CONFIG *)xd->cur_buf, cm, optimized_loop_restoration,
+              pbi->tile_workers, pbi->num_workers, &pbi->lr_row_sync,
+              &pbi->lr_ctxt);
+        } else {
+          av1_loop_restoration_filter_frame((YV12_BUFFER_CONFIG *)xd->cur_buf,
+                                            cm, optimized_loop_restoration,
+                                            &pbi->lr_ctxt);
+        }
+      }
+    }
   }
-#endif  // CONFIG_LOOP_RESTORATION
 
   if (!xd->corrupted) {
     if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) {
-      FRAME_CONTEXT **tile_ctxs = aom_malloc(cm->tile_rows * cm->tile_cols *
-                                             sizeof(&pbi->tile_data[0].tctx));
-      aom_cdf_prob **cdf_ptrs =
-          aom_malloc(cm->tile_rows * cm->tile_cols *
-                     sizeof(&pbi->tile_data[0].tctx.partition_cdf[0][0]));
-      make_update_tile_list_dec(pbi, cm->tile_rows, cm->tile_cols, tile_ctxs);
-#if CONFIG_LV_MAP
-      av1_adapt_coef_probs(cm);
-#endif  // CONFIG_LV_MAP
-#if CONFIG_SYMBOLRATE
-      av1_dump_symbol_rate(cm);
-#endif
-      av1_adapt_intra_frame_probs(cm);
-      av1_average_tile_coef_cdfs(pbi->common.fc, tile_ctxs, cdf_ptrs,
-                                 cm->tile_rows * cm->tile_cols);
-      av1_average_tile_intra_cdfs(pbi->common.fc, tile_ctxs, cdf_ptrs,
-                                  cm->tile_rows * cm->tile_cols);
-#if CONFIG_PVQ
-      av1_average_tile_pvq_cdfs(pbi->common.fc, tile_ctxs,
-                                cm->tile_rows * cm->tile_cols);
-#endif  // CONFIG_PVQ
-#if CONFIG_ADAPT_SCAN
-      av1_adapt_scan_order(cm);
-#endif  // CONFIG_ADAPT_SCAN
-
-      if (!frame_is_intra_only(cm)) {
-        av1_adapt_inter_frame_probs(cm);
-#if !CONFIG_NEW_MULTISYMBOL
-        av1_adapt_mv_probs(cm, cm->allow_high_precision_mv);
-#endif
-        av1_average_tile_inter_cdfs(&pbi->common, pbi->common.fc, tile_ctxs,
-                                    cdf_ptrs, cm->tile_rows * cm->tile_cols);
-        av1_average_tile_mv_cdfs(pbi->common.fc, tile_ctxs, cdf_ptrs,
-                                 cm->tile_rows * cm->tile_cols);
-      }
-      aom_free(tile_ctxs);
-      aom_free(cdf_ptrs);
-    } else {
-      debug_check_frame_counts(cm);
+      *cm->fc = pbi->tile_data[cm->context_update_tile_id].tctx;
+      av1_reset_cdf_symbol_counters(cm->fc);
     }
   } else {
     aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
@@ -5808,153 +4481,8 @@ void av1_decode_tg_tiles_and_wrapup(AV1Decoder *pbi, const uint8_t *data,
   }
 #endif
 
-// Non frame parallel update frame context here.
-#if CONFIG_NO_FRAME_CONTEXT_SIGNALING
-  if (!context_updated) cm->frame_contexts[cm->new_fb_idx] = *cm->fc;
-#else
-  if (!cm->error_resilient_mode && !context_updated)
-    cm->frame_contexts[cm->frame_context_idx] = *cm->fc;
-#endif
-}
-
-#if CONFIG_OBU
-
-static OBU_TYPE read_obu_header(struct aom_read_bit_buffer *rb,
-                                uint32_t *header_size) {
-  OBU_TYPE obu_type;
-  int obu_extension_flag;
-
-  *header_size = 1;
-
-  obu_type = (OBU_TYPE)aom_rb_read_literal(rb, 5);
-  aom_rb_read_literal(rb, 2);  // reserved
-  obu_extension_flag = aom_rb_read_bit(rb);
-  if (obu_extension_flag) {
-    *header_size += 1;
-    aom_rb_read_literal(rb, 3);  // temporal_id
-    aom_rb_read_literal(rb, 2);
-    aom_rb_read_literal(rb, 2);
-    aom_rb_read_literal(rb, 1);  // reserved
-  }
-
-  return obu_type;
-}
-
-static uint32_t read_temporal_delimiter_obu() { return 0; }
-
-static uint32_t read_sequence_header_obu(AV1Decoder *pbi,
-                                         struct aom_read_bit_buffer *rb) {
-  AV1_COMMON *const cm = &pbi->common;
-  SequenceHeader *const seq_params = &cm->seq_params;
-  uint32_t saved_bit_offset = rb->bit_offset;
-
-  cm->profile = av1_read_profile(rb);
-  aom_rb_read_literal(rb, 4);  // level
-
-  seq_params->frame_id_numbers_present_flag = aom_rb_read_bit(rb);
-  if (seq_params->frame_id_numbers_present_flag) {
-    seq_params->frame_id_length_minus7 = aom_rb_read_literal(rb, 4);
-    seq_params->delta_frame_id_length_minus2 = aom_rb_read_literal(rb, 4);
-  }
-
-  read_bitdepth_colorspace_sampling(cm, rb, pbi->allow_lowbitdepth);
-
-  return ((rb->bit_offset - saved_bit_offset + 7) >> 3);
-}
-
-static uint32_t read_frame_header_obu(AV1Decoder *pbi, const uint8_t *data,
-                                      const uint8_t *data_end,
-                                      const uint8_t **p_data_end) {
-  size_t header_size;
-
-  header_size =
-      av1_decode_frame_headers_and_setup(pbi, data, data_end, p_data_end);
-  return (uint32_t)(pbi->uncomp_hdr_size + header_size);
-}
-
-static uint32_t read_tile_group_header(AV1Decoder *pbi,
-                                       struct aom_read_bit_buffer *rb,
-                                       int *startTile, int *endTile) {
-  AV1_COMMON *const cm = &pbi->common;
-  uint32_t saved_bit_offset = rb->bit_offset;
-
-  *startTile = aom_rb_read_literal(rb, cm->log2_tile_rows + cm->log2_tile_cols);
-  *endTile = aom_rb_read_literal(rb, cm->log2_tile_rows + cm->log2_tile_cols);
-
-  return ((rb->bit_offset - saved_bit_offset + 7) >> 3);
-}
-
-static uint32_t read_one_tile_group_obu(AV1Decoder *pbi,
-                                        struct aom_read_bit_buffer *rb,
-                                        int is_first_tg, const uint8_t *data,
-                                        const uint8_t *data_end,
-                                        const uint8_t **p_data_end,
-                                        int *is_last_tg) {
-  AV1_COMMON *const cm = &pbi->common;
-  int startTile, endTile;
-  uint32_t header_size, tg_payload_size;
-
-  header_size = read_tile_group_header(pbi, rb, &startTile, &endTile);
-  data += header_size;
-  av1_decode_tg_tiles_and_wrapup(pbi, data, data_end, p_data_end, startTile,
-                                 endTile, is_first_tg);
-  tg_payload_size = (uint32_t)(*p_data_end - data);
-
-  // TODO(shan):  For now, assume all tile groups received in order
-  *is_last_tg = endTile == cm->tile_rows * cm->tile_cols - 1;
-
-  return header_size + tg_payload_size;
-}
-
-void av1_decode_frame_from_obus(struct AV1Decoder *pbi, const uint8_t *data,
-                                const uint8_t *data_end,
-                                const uint8_t **p_data_end) {
-  AV1_COMMON *const cm = &pbi->common;
-  int frame_decoding_finished = 0;
-  int is_first_tg_obu_received = 1;
-  int frame_header_received = 0;
-  int frame_header_size = 0;
-
-  // decode frame as a series of OBUs
-  while (!frame_decoding_finished && !cm->error.error_code) {
-    struct aom_read_bit_buffer rb;
-    uint8_t clear_data[80];
-    uint32_t obu_size, obu_header_size, obu_payload_size = 0;
-    OBU_TYPE obu_type;
-
-    init_read_bit_buffer(pbi, &rb, data + 4, data_end, clear_data);
-
-    // every obu is preceded by 4-byte size of obu (obu header + payload size)
-    // The obu size is only needed for tile group OBUs
-    obu_size = mem_get_le32(data);
-    obu_type = read_obu_header(&rb, &obu_header_size);
-    data += (4 + obu_header_size);
-
-    switch (obu_type) {
-      case OBU_TD: obu_payload_size = read_temporal_delimiter_obu(); break;
-      case OBU_SEQUENCE_HEADER:
-        obu_payload_size = read_sequence_header_obu(pbi, &rb);
-        break;
-      case OBU_FRAME_HEADER:
-        // Only decode first frame header received
-        if (!frame_header_received) {
-          frame_header_size = obu_payload_size =
-              read_frame_header_obu(pbi, data, data_end, p_data_end);
-          frame_header_received = 1;
-        } else {
-          obu_payload_size = frame_header_size;
-        }
-        if (cm->show_existing_frame) frame_decoding_finished = 1;
-        break;
-      case OBU_TILE_GROUP:
-        obu_payload_size = read_one_tile_group_obu(
-            pbi, &rb, is_first_tg_obu_received, data, data + obu_size - 1,
-            p_data_end, &frame_decoding_finished);
-        is_first_tg_obu_received = 0;
-        break;
-      default: break;
-    }
-    data += obu_payload_size;
+  // Non frame parallel update frame context here.
+  if (!cm->large_scale_tile) {
+    cm->frame_contexts[cm->new_fb_idx] = *cm->fc;
   }
 }
-#endif
diff --git a/third_party/aom/av1/decoder/decodeframe.h b/third_party/aom/av1/decoder/decodeframe.h
index 0e7eb6a1d..330cedcdc 100644
--- a/third_party/aom/av1/decoder/decodeframe.h
+++ b/third_party/aom/av1/decoder/decodeframe.h
@@ -19,35 +19,59 @@ extern "C" {
 struct AV1Decoder;
 struct aom_read_bit_buffer;
 
-#if CONFIG_REFERENCE_BUFFER
-/* Placeholder for now */
-void read_sequence_header(SequenceHeader *seq_params,
-                          struct aom_read_bit_buffer *rb);
-#endif
+// Reads the middle part of the sequence header OBU (from
+// frame_width_bits_minus_1 to enable_restoration) into cm->seq_params (a
+// SequenceHeader). Reports errors by calling rb->error_handler() or
+// aom_internal_error().
+void read_sequence_header(AV1_COMMON *cm, struct aom_read_bit_buffer *rb);
 
-void av1_read_frame_size(struct aom_read_bit_buffer *rb, int *width,
-                         int *height);
+void av1_read_frame_size(struct aom_read_bit_buffer *rb, int num_bits_width,
+                         int num_bits_height, int *width, int *height);
 BITSTREAM_PROFILE av1_read_profile(struct aom_read_bit_buffer *rb);
 
-// This function is now obsolete
-void av1_decode_frame(struct AV1Decoder *pbi, const uint8_t *data,
-                      const uint8_t *data_end, const uint8_t **p_data_end);
-size_t av1_decode_frame_headers_and_setup(struct AV1Decoder *pbi,
-                                          const uint8_t *data,
-                                          const uint8_t *data_end,
-                                          const uint8_t **p_data_end);
+// Returns 0 on success. Sets pbi->common.error.error_code and returns -1 on
+// failure.
+int av1_check_trailing_bits(struct AV1Decoder *pbi,
+                            struct aom_read_bit_buffer *rb);
+
+int av1_decode_frame_headers_and_setup(struct AV1Decoder *pbi,
+                                       struct aom_read_bit_buffer *rb,
+                                       const uint8_t *data,
+                                       const uint8_t **p_data_end,
+                                       int trailing_bits_present);
 
 void av1_decode_tg_tiles_and_wrapup(struct AV1Decoder *pbi, const uint8_t *data,
                                     const uint8_t *data_end,
                                     const uint8_t **p_data_end, int startTile,
                                     int endTile, int initialize_flag);
 
-#if CONFIG_OBU
-// replaces av1_decode_frame
-void av1_decode_frame_from_obus(struct AV1Decoder *pbi, const uint8_t *data,
-                                const uint8_t *data_end,
-                                const uint8_t **p_data_end);
-#endif
+// Implements the color_config() function in the spec. Reports errors by
+// calling rb->error_handler() or aom_internal_error().
+void av1_read_color_config(AV1_COMMON *cm, struct aom_read_bit_buffer *rb,
+                           int allow_lowbitdepth);
+
+// Implements the timing_info() function in the spec. Reports errors by calling
+// rb->error_handler().
+void av1_read_timing_info_header(AV1_COMMON *cm,
+                                 struct aom_read_bit_buffer *rb);
+
+// Implements the decoder_model_info() function in the spec. Reports errors by
+// calling rb->error_handler().
+void av1_read_decoder_model_info(AV1_COMMON *cm,
+                                 struct aom_read_bit_buffer *rb);
+
+// Implements the operating_parameters_info() function in the spec. Reports
+// errors by calling rb->error_handler() or aom_internal_error().
+void av1_read_op_parameters_info(AV1_COMMON *const cm,
+                                 struct aom_read_bit_buffer *rb, int op_num);
+
+struct aom_read_bit_buffer *av1_init_read_bit_buffer(
+    struct AV1Decoder *pbi, struct aom_read_bit_buffer *rb, const uint8_t *data,
+    const uint8_t *data_end);
+
+void av1_free_mc_tmp_buf(void *td, int use_highbd);
+
+void av1_set_single_tile_decoding_mode(AV1_COMMON *const cm);
 
 #ifdef __cplusplus
 }  // extern "C"
diff --git a/third_party/aom/av1/decoder/decodemv.c b/third_party/aom/av1/decoder/decodemv.c
index cac27e9a6..cc8f4d29e 100644
--- a/third_party/aom/av1/decoder/decodemv.c
+++ b/third_party/aom/av1/decoder/decodemv.c
@@ -11,6 +11,7 @@
 
 #include <assert.h>
 
+#include "av1/common/cfl.h"
 #include "av1/common/common.h"
 #include "av1/common/entropy.h"
 #include "av1/common/entropymode.h"
@@ -18,13 +19,9 @@
 #include "av1/common/mvref_common.h"
 #include "av1/common/pred_common.h"
 #include "av1/common/reconinter.h"
-#if CONFIG_EXT_INTRA
 #include "av1/common/reconintra.h"
-#endif  // CONFIG_EXT_INTRA
 #include "av1/common/seg_common.h"
-#if CONFIG_WARPED_MOTION
 #include "av1/common/warped_motion.h"
-#endif  // CONFIG_WARPED_MOTION
 
 #include "av1/decoder/decodeframe.h"
 #include "av1/decoder/decodemv.h"
@@ -39,30 +36,51 @@ static PREDICTION_MODE read_intra_mode(aom_reader *r, aom_cdf_prob *cdf) {
   return (PREDICTION_MODE)aom_read_symbol(r, cdf, INTRA_MODES, ACCT_STR);
 }
 
-static int read_delta_qindex(AV1_COMMON *cm, MACROBLOCKD *xd, aom_reader *r,
-                             MB_MODE_INFO *const mbmi, int mi_col, int mi_row) {
-  FRAME_COUNTS *counts = xd->counts;
+static void read_cdef(AV1_COMMON *cm, aom_reader *r, MACROBLOCKD *const xd,
+                      int mi_col, int mi_row) {
+  MB_MODE_INFO *const mbmi = xd->mi[0];
+  if (cm->coded_lossless) return;
+  if (cm->allow_intrabc) {
+    assert(cm->cdef_bits == 0);
+    return;
+  }
+
+  if (!(mi_col & (cm->seq_params.mib_size - 1)) &&
+      !(mi_row & (cm->seq_params.mib_size - 1))) {  // Top left?
+    xd->cdef_preset[0] = xd->cdef_preset[1] = xd->cdef_preset[2] =
+        xd->cdef_preset[3] = -1;
+  }
+  // Read CDEF param at the first non-skip coding block
+  const int mask = (1 << (6 - MI_SIZE_LOG2));
+  const int m = ~(mask - 1);
+  const int index = cm->seq_params.sb_size == BLOCK_128X128
+                        ? !!(mi_col & mask) + 2 * !!(mi_row & mask)
+                        : 0;
+  cm->mi_grid_visible[(mi_row & m) * cm->mi_stride + (mi_col & m)]
+      ->cdef_strength = xd->cdef_preset[index] =
+      xd->cdef_preset[index] == -1 && !mbmi->skip
+          ? aom_read_literal(r, cm->cdef_bits, ACCT_STR)
+          : xd->cdef_preset[index];
+}
+
+static int read_delta_qindex(AV1_COMMON *cm, const MACROBLOCKD *xd,
+                             aom_reader *r, MB_MODE_INFO *const mbmi,
+                             int mi_col, int mi_row) {
   int sign, abs, reduced_delta_qindex = 0;
   BLOCK_SIZE bsize = mbmi->sb_type;
-  const int b_col = mi_col & MAX_MIB_MASK;
-  const int b_row = mi_row & MAX_MIB_MASK;
+  const int b_col = mi_col & (cm->seq_params.mib_size - 1);
+  const int b_row = mi_row & (cm->seq_params.mib_size - 1);
   const int read_delta_q_flag = (b_col == 0 && b_row == 0);
-  int rem_bits, thr;
-  int i, smallval;
   FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
-  (void)cm;
 
-  if ((bsize != BLOCK_LARGEST || mbmi->skip == 0) && read_delta_q_flag) {
+  if ((bsize != cm->seq_params.sb_size || mbmi->skip == 0) &&
+      read_delta_q_flag) {
     abs = aom_read_symbol(r, ec_ctx->delta_q_cdf, DELTA_Q_PROBS + 1, ACCT_STR);
-    smallval = (abs < DELTA_Q_SMALL);
-    if (counts) {
-      for (i = 0; i < abs; ++i) counts->delta_q[i][1]++;
-      if (smallval) counts->delta_q[abs][0]++;
-    }
+    const int smallval = (abs < DELTA_Q_SMALL);
 
     if (!smallval) {
-      rem_bits = aom_read_literal(r, 3, ACCT_STR) + 1;
-      thr = (1 << rem_bits) + 1;
+      const int rem_bits = aom_read_literal(r, 3, ACCT_STR) + 1;
+      const int thr = (1 << rem_bits) + 1;
       abs = aom_read_literal(r, rem_bits, ACCT_STR) + thr;
     }
 
@@ -76,56 +94,33 @@ static int read_delta_qindex(AV1_COMMON *cm, MACROBLOCKD *xd, aom_reader *r,
   }
   return reduced_delta_qindex;
 }
-#if CONFIG_EXT_DELTA_Q
-static int read_delta_lflevel(AV1_COMMON *cm, MACROBLOCKD *xd, aom_reader *r,
-#if CONFIG_LOOPFILTER_LEVEL
-                              int lf_id,
-#endif
+static int read_delta_lflevel(AV1_COMMON *cm, const MACROBLOCKD *xd,
+                              aom_reader *r, int lf_id,
                               MB_MODE_INFO *const mbmi, int mi_col,
                               int mi_row) {
-  FRAME_COUNTS *counts = xd->counts;
   int sign, abs, reduced_delta_lflevel = 0;
   BLOCK_SIZE bsize = mbmi->sb_type;
-  const int b_col = mi_col & MAX_MIB_MASK;
-  const int b_row = mi_row & MAX_MIB_MASK;
+  const int b_col = mi_col & (cm->seq_params.mib_size - 1);
+  const int b_row = mi_row & (cm->seq_params.mib_size - 1);
   const int read_delta_lf_flag = (b_col == 0 && b_row == 0);
-  int rem_bits, thr;
-  int i, smallval;
   FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
-  (void)cm;
 
-  if ((bsize != cm->sb_size || mbmi->skip == 0) && read_delta_lf_flag) {
-#if CONFIG_LOOPFILTER_LEVEL
+  if ((bsize != cm->seq_params.sb_size || mbmi->skip == 0) &&
+      read_delta_lf_flag) {
     if (cm->delta_lf_multi) {
-      assert(lf_id >= 0 && lf_id < FRAME_LF_COUNT);
+      assert(lf_id >= 0 &&
+             lf_id < (av1_num_planes(cm) > 1 ? FRAME_LF_COUNT
+                                             : FRAME_LF_COUNT - 2));
       abs = aom_read_symbol(r, ec_ctx->delta_lf_multi_cdf[lf_id],
                             DELTA_LF_PROBS + 1, ACCT_STR);
     } else {
       abs = aom_read_symbol(r, ec_ctx->delta_lf_cdf, DELTA_LF_PROBS + 1,
                             ACCT_STR);
     }
-#else
-    abs =
-        aom_read_symbol(r, ec_ctx->delta_lf_cdf, DELTA_LF_PROBS + 1, ACCT_STR);
-#endif  // CONFIG_LOOPFILTER_LEVEL
-    smallval = (abs < DELTA_LF_SMALL);
-    if (counts) {
-#if CONFIG_LOOPFILTER_LEVEL
-      if (cm->delta_lf_multi) {
-        for (i = 0; i < abs; ++i) counts->delta_lf_multi[lf_id][i][1]++;
-        if (smallval) counts->delta_lf_multi[lf_id][abs][0]++;
-      } else {
-        for (i = 0; i < abs; ++i) counts->delta_lf[i][1]++;
-        if (smallval) counts->delta_lf[abs][0]++;
-      }
-#else
-      for (i = 0; i < abs; ++i) counts->delta_lf[i][1]++;
-      if (smallval) counts->delta_lf[abs][0]++;
-#endif  // CONFIG_LOOPFILTER_LEVEL
-    }
+    const int smallval = (abs < DELTA_LF_SMALL);
     if (!smallval) {
-      rem_bits = aom_read_literal(r, 3, ACCT_STR) + 1;
-      thr = (1 << rem_bits) + 1;
+      const int rem_bits = aom_read_literal(r, 3, ACCT_STR) + 1;
+      const int thr = (1 << rem_bits) + 1;
       abs = aom_read_literal(r, rem_bits, ACCT_STR) + thr;
     }
 
@@ -139,21 +134,17 @@ static int read_delta_lflevel(AV1_COMMON *cm, MACROBLOCKD *xd, aom_reader *r,
   }
   return reduced_delta_lflevel;
 }
-#endif
 
 static UV_PREDICTION_MODE read_intra_mode_uv(FRAME_CONTEXT *ec_ctx,
                                              aom_reader *r,
+                                             CFL_ALLOWED_TYPE cfl_allowed,
                                              PREDICTION_MODE y_mode) {
   const UV_PREDICTION_MODE uv_mode =
-#if CONFIG_CFL
-      aom_read_symbol(r, ec_ctx->uv_mode_cdf[y_mode], UV_INTRA_MODES, ACCT_STR);
-#else
-      read_intra_mode(r, ec_ctx->uv_mode_cdf[y_mode]);
-#endif  // CONFIG_CFL
+      aom_read_symbol(r, ec_ctx->uv_mode_cdf[cfl_allowed][y_mode],
+                      UV_INTRA_MODES - !cfl_allowed, ACCT_STR);
   return uv_mode;
 }
 
-#if CONFIG_CFL
 static int read_cfl_alphas(FRAME_CONTEXT *const ec_ctx, aom_reader *r,
                            int *signs_out) {
   const int joint_sign =
@@ -172,400 +163,145 @@ static int read_cfl_alphas(FRAME_CONTEXT *const ec_ctx, aom_reader *r,
   *signs_out = joint_sign;
   return idx;
 }
-#endif
 
-#if CONFIG_INTERINTRA
-static INTERINTRA_MODE read_interintra_mode(AV1_COMMON *cm, MACROBLOCKD *xd,
-                                            aom_reader *r, int size_group) {
-  (void)cm;
+static INTERINTRA_MODE read_interintra_mode(MACROBLOCKD *xd, aom_reader *r,
+                                            int size_group) {
   const INTERINTRA_MODE ii_mode = (INTERINTRA_MODE)aom_read_symbol(
       r, xd->tile_ctx->interintra_mode_cdf[size_group], INTERINTRA_MODES,
       ACCT_STR);
-  FRAME_COUNTS *counts = xd->counts;
-  if (counts) ++counts->interintra_mode[size_group][ii_mode];
   return ii_mode;
 }
-#endif  // CONFIG_INTERINTRA
 
-static PREDICTION_MODE read_inter_mode(FRAME_CONTEXT *ec_ctx, MACROBLOCKD *xd,
-                                       aom_reader *r, int16_t ctx) {
-  FRAME_COUNTS *counts = xd->counts;
+static PREDICTION_MODE read_inter_mode(FRAME_CONTEXT *ec_ctx, aom_reader *r,
+                                       int16_t ctx) {
   int16_t mode_ctx = ctx & NEWMV_CTX_MASK;
   int is_newmv, is_zeromv, is_refmv;
-#if CONFIG_NEW_MULTISYMBOL
   is_newmv = aom_read_symbol(r, ec_ctx->newmv_cdf[mode_ctx], 2, ACCT_STR) == 0;
-#else
-  is_newmv = aom_read(r, ec_ctx->newmv_prob[mode_ctx], ACCT_STR) == 0;
-#endif
-
-  if (is_newmv) {
-    if (counts) ++counts->newmv_mode[mode_ctx][0];
-    return NEWMV;
-  }
-  if (counts) ++counts->newmv_mode[mode_ctx][1];
-
-  if (ctx & (1 << ALL_ZERO_FLAG_OFFSET)) return ZEROMV;
+  if (is_newmv) return NEWMV;
 
-  mode_ctx = (ctx >> ZEROMV_OFFSET) & ZEROMV_CTX_MASK;
-
-#if CONFIG_NEW_MULTISYMBOL
+  mode_ctx = (ctx >> GLOBALMV_OFFSET) & GLOBALMV_CTX_MASK;
   is_zeromv =
       aom_read_symbol(r, ec_ctx->zeromv_cdf[mode_ctx], 2, ACCT_STR) == 0;
-#else
-  is_zeromv = aom_read(r, ec_ctx->zeromv_prob[mode_ctx], ACCT_STR) == 0;
-#endif
-  if (is_zeromv) {
-    if (counts) ++counts->zeromv_mode[mode_ctx][0];
-    return ZEROMV;
-  }
-  if (counts) ++counts->zeromv_mode[mode_ctx][1];
+  if (is_zeromv) return GLOBALMV;
 
   mode_ctx = (ctx >> REFMV_OFFSET) & REFMV_CTX_MASK;
-
-  if (ctx & (1 << SKIP_NEARESTMV_OFFSET)) mode_ctx = 6;
-  if (ctx & (1 << SKIP_NEARMV_OFFSET)) mode_ctx = 7;
-  if (ctx & (1 << SKIP_NEARESTMV_SUB8X8_OFFSET)) mode_ctx = 8;
-
-#if CONFIG_NEW_MULTISYMBOL
   is_refmv = aom_read_symbol(r, ec_ctx->refmv_cdf[mode_ctx], 2, ACCT_STR) == 0;
-#else
-  is_refmv = aom_read(r, ec_ctx->refmv_prob[mode_ctx], ACCT_STR) == 0;
-#endif
-
-  if (is_refmv) {
-    if (counts) ++counts->refmv_mode[mode_ctx][0];
-
+  if (is_refmv)
     return NEARESTMV;
-  } else {
-    if (counts) ++counts->refmv_mode[mode_ctx][1];
+  else
     return NEARMV;
-  }
-
-  // Invalid prediction mode.
-  assert(0);
 }
 
 static void read_drl_idx(FRAME_CONTEXT *ec_ctx, MACROBLOCKD *xd,
                          MB_MODE_INFO *mbmi, aom_reader *r) {
   uint8_t ref_frame_type = av1_ref_frame_type(mbmi->ref_frame);
   mbmi->ref_mv_idx = 0;
-
-  if (mbmi->mode == NEWMV || mbmi->mode == NEW_NEWMV
-#if CONFIG_COMPOUND_SINGLEREF
-      || mbmi->mode == SR_NEW_NEWMV
-#endif  // CONFIG_COMPOUND_SINGLEREF
-      ) {
-    int idx;
-    for (idx = 0; idx < 2; ++idx) {
+  if (mbmi->mode == NEWMV || mbmi->mode == NEW_NEWMV) {
+    for (int idx = 0; idx < 2; ++idx) {
       if (xd->ref_mv_count[ref_frame_type] > idx + 1) {
         uint8_t drl_ctx = av1_drl_ctx(xd->ref_mv_stack[ref_frame_type], idx);
-#if CONFIG_NEW_MULTISYMBOL
         int drl_idx = aom_read_symbol(r, ec_ctx->drl_cdf[drl_ctx], 2, ACCT_STR);
-#else
-        int drl_idx = aom_read(r, ec_ctx->drl_prob[drl_ctx], ACCT_STR);
-#endif
         mbmi->ref_mv_idx = idx + drl_idx;
-        if (xd->counts) ++xd->counts->drl_mode[drl_ctx][drl_idx];
         if (!drl_idx) return;
       }
     }
   }
-
   if (have_nearmv_in_inter_mode(mbmi->mode)) {
-    int idx;
     // Offset the NEARESTMV mode.
     // TODO(jingning): Unify the two syntax decoding loops after the NEARESTMV
     // mode is factored in.
-    for (idx = 1; idx < 3; ++idx) {
+    for (int idx = 1; idx < 3; ++idx) {
       if (xd->ref_mv_count[ref_frame_type] > idx + 1) {
         uint8_t drl_ctx = av1_drl_ctx(xd->ref_mv_stack[ref_frame_type], idx);
-#if CONFIG_NEW_MULTISYMBOL
         int drl_idx = aom_read_symbol(r, ec_ctx->drl_cdf[drl_ctx], 2, ACCT_STR);
-#else
-        int drl_idx = aom_read(r, ec_ctx->drl_prob[drl_ctx], ACCT_STR);
-#endif
         mbmi->ref_mv_idx = idx + drl_idx - 1;
-        if (xd->counts) ++xd->counts->drl_mode[drl_ctx][drl_idx];
         if (!drl_idx) return;
       }
     }
   }
 }
 
-#if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
 static MOTION_MODE read_motion_mode(AV1_COMMON *cm, MACROBLOCKD *xd,
-                                    MODE_INFO *mi, aom_reader *r) {
-  MB_MODE_INFO *mbmi = &mi->mbmi;
-#if !CONFIG_MOTION_VAR || !CONFIG_WARPED_MOTION || CONFIG_NEW_MULTISYMBOL || \
-    CONFIG_NCOBMC_ADAPT_WEIGHT
-  (void)cm;
-#endif
+                                    MB_MODE_INFO *mbmi, aom_reader *r) {
+  if (cm->switchable_motion_mode == 0) return SIMPLE_TRANSLATION;
+  if (mbmi->skip_mode) return SIMPLE_TRANSLATION;
 
-  const MOTION_MODE last_motion_mode_allowed = motion_mode_allowed(
-#if CONFIG_GLOBAL_MOTION
-      0, xd->global_motion,
-#endif  // CONFIG_GLOBAL_MOTION
-#if CONFIG_WARPED_MOTION
-      xd,
-#endif
-      mi);
+  const MOTION_MODE last_motion_mode_allowed =
+      motion_mode_allowed(xd->global_motion, xd, mbmi, cm->allow_warped_motion);
   int motion_mode;
-  FRAME_COUNTS *counts = xd->counts;
 
   if (last_motion_mode_allowed == SIMPLE_TRANSLATION) return SIMPLE_TRANSLATION;
-#if CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
-#if CONFIG_NCOBMC_ADAPT_WEIGHT
-  if (last_motion_mode_allowed == NCOBMC_ADAPT_WEIGHT) {
-    motion_mode = aom_read_symbol(r, xd->tile_ctx->ncobmc_cdf[mbmi->sb_type],
-                                  OBMC_FAMILY_MODES, ACCT_STR);
-    if (counts) ++counts->ncobmc[mbmi->sb_type][motion_mode];
-    return (MOTION_MODE)(SIMPLE_TRANSLATION + motion_mode);
-  } else if (last_motion_mode_allowed == OBMC_CAUSAL) {
-    motion_mode =
-        aom_read_symbol(r, xd->tile_ctx->obmc_cdf[mbmi->sb_type], 2, ACCT_STR);
-    if (counts) ++counts->obmc[mbmi->sb_type][motion_mode];
-    return (MOTION_MODE)(SIMPLE_TRANSLATION + motion_mode);
-  } else {
-#else
+
   if (last_motion_mode_allowed == OBMC_CAUSAL) {
-#if CONFIG_NEW_MULTISYMBOL
     motion_mode =
         aom_read_symbol(r, xd->tile_ctx->obmc_cdf[mbmi->sb_type], 2, ACCT_STR);
-#else
-    motion_mode = aom_read(r, cm->fc->obmc_prob[mbmi->sb_type], ACCT_STR);
-#endif
-    if (counts) ++counts->obmc[mbmi->sb_type][motion_mode];
     return (MOTION_MODE)(SIMPLE_TRANSLATION + motion_mode);
   } else {
-#endif  // CONFIG_NCOBMC_ADAPT_WEIGHT
-#endif  // CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
     motion_mode =
         aom_read_symbol(r, xd->tile_ctx->motion_mode_cdf[mbmi->sb_type],
                         MOTION_MODES, ACCT_STR);
-    if (counts) ++counts->motion_mode[mbmi->sb_type][motion_mode];
     return (MOTION_MODE)(SIMPLE_TRANSLATION + motion_mode);
-#if CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
   }
-#endif  // CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
 }
 
-#if CONFIG_NCOBMC_ADAPT_WEIGHT
-static void read_ncobmc_mode(MACROBLOCKD *xd, MODE_INFO *mi,
-                             NCOBMC_MODE ncobmc_mode[2], aom_reader *r) {
-  MB_MODE_INFO *mbmi = &mi->mbmi;
-  FRAME_COUNTS *counts = xd->counts;
-  ADAPT_OVERLAP_BLOCK ao_block = adapt_overlap_block_lookup[mbmi->sb_type];
-  if (mbmi->motion_mode != NCOBMC_ADAPT_WEIGHT) return;
-
-  ncobmc_mode[0] = aom_read_symbol(r, xd->tile_ctx->ncobmc_mode_cdf[ao_block],
-                                   MAX_NCOBMC_MODES, ACCT_STR);
-  if (counts) ++counts->ncobmc_mode[ao_block][ncobmc_mode[0]];
-
-  if (mi_size_wide[mbmi->sb_type] != mi_size_high[mbmi->sb_type]) {
-    ncobmc_mode[1] = aom_read_symbol(r, xd->tile_ctx->ncobmc_mode_cdf[ao_block],
-                                     MAX_NCOBMC_MODES, ACCT_STR);
-    if (counts) ++counts->ncobmc_mode[ao_block][ncobmc_mode[1]];
-  }
-}
-#endif  // CONFIG_NCOBMC_ADAPT_WEIGHT
-#endif  // CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
-
-static PREDICTION_MODE read_inter_compound_mode(AV1_COMMON *cm, MACROBLOCKD *xd,
-                                                aom_reader *r, int16_t ctx) {
-  (void)cm;
+static PREDICTION_MODE read_inter_compound_mode(MACROBLOCKD *xd, aom_reader *r,
+                                                int16_t ctx) {
   const int mode =
       aom_read_symbol(r, xd->tile_ctx->inter_compound_mode_cdf[ctx],
                       INTER_COMPOUND_MODES, ACCT_STR);
-  FRAME_COUNTS *counts = xd->counts;
-
-  if (counts) ++counts->inter_compound_mode[ctx][mode];
-
   assert(is_inter_compound_mode(NEAREST_NEARESTMV + mode));
   return NEAREST_NEARESTMV + mode;
 }
 
-#if CONFIG_COMPOUND_SINGLEREF
-static PREDICTION_MODE read_inter_singleref_comp_mode(MACROBLOCKD *xd,
-                                                      aom_reader *r,
-                                                      int16_t ctx) {
-  const int mode =
-      aom_read_symbol(r, xd->tile_ctx->inter_singleref_comp_mode_cdf[ctx],
-                      INTER_SINGLEREF_COMP_MODES, ACCT_STR);
-  FRAME_COUNTS *counts = xd->counts;
-
-  if (counts) ++counts->inter_singleref_comp_mode[ctx][mode];
-
-  assert(is_inter_singleref_comp_mode(SR_NEAREST_NEARMV + mode));
-  return SR_NEAREST_NEARMV + mode;
-}
-#endif  // CONFIG_COMPOUND_SINGLEREF
-
-static int read_segment_id(aom_reader *r, struct segmentation_probs *segp) {
-  return aom_read_symbol(r, segp->tree_cdf, MAX_SEGMENTS, ACCT_STR);
-}
-
-#if CONFIG_VAR_TX
-static void read_tx_size_vartx(AV1_COMMON *cm, MACROBLOCKD *xd,
-                               MB_MODE_INFO *mbmi, FRAME_COUNTS *counts,
-                               TX_SIZE tx_size, int depth, int blk_row,
-                               int blk_col, aom_reader *r) {
-#if CONFIG_NEW_MULTISYMBOL
-  FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
-  (void)cm;
-#endif
-  int is_split = 0;
-  const int tx_row = blk_row >> 1;
-  const int tx_col = blk_col >> 1;
-  const int max_blocks_high = max_block_high(xd, mbmi->sb_type, 0);
-  const int max_blocks_wide = max_block_wide(xd, mbmi->sb_type, 0);
-  int ctx = txfm_partition_context(xd->above_txfm_context + blk_col,
-                                   xd->left_txfm_context + blk_row,
-                                   mbmi->sb_type, tx_size);
-  TX_SIZE(*const inter_tx_size)
-  [MAX_MIB_SIZE] =
-      (TX_SIZE(*)[MAX_MIB_SIZE]) & mbmi->inter_tx_size[tx_row][tx_col];
-  if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
-  assert(tx_size > TX_4X4);
-
-  if (depth == MAX_VARTX_DEPTH) {
-    int idx, idy;
-    inter_tx_size[0][0] = tx_size;
-    for (idy = 0; idy < tx_size_high_unit[tx_size] / 2; ++idy)
-      for (idx = 0; idx < tx_size_wide_unit[tx_size] / 2; ++idx)
-        inter_tx_size[idy][idx] = tx_size;
-    mbmi->tx_size = tx_size;
-    mbmi->min_tx_size = AOMMIN(mbmi->min_tx_size, get_min_tx_size(tx_size));
-    txfm_partition_update(xd->above_txfm_context + blk_col,
-                          xd->left_txfm_context + blk_row, tx_size, tx_size);
-    return;
-  }
-
-#if CONFIG_NEW_MULTISYMBOL
-  is_split = aom_read_symbol(r, ec_ctx->txfm_partition_cdf[ctx], 2, ACCT_STR);
-#else
-  is_split = aom_read(r, cm->fc->txfm_partition_prob[ctx], ACCT_STR);
-#endif
-
-  if (is_split) {
-    const TX_SIZE sub_txs = sub_tx_size_map[tx_size];
-    const int bsl = tx_size_wide_unit[sub_txs];
-    int i;
-
-    if (counts) ++counts->txfm_partition[ctx][1];
-
-    if (sub_txs == TX_4X4) {
-      int idx, idy;
-      inter_tx_size[0][0] = sub_txs;
-      for (idy = 0; idy < tx_size_high_unit[tx_size] / 2; ++idy)
-        for (idx = 0; idx < tx_size_wide_unit[tx_size] / 2; ++idx)
-          inter_tx_size[idy][idx] = inter_tx_size[0][0];
-      mbmi->tx_size = sub_txs;
-      mbmi->min_tx_size = get_min_tx_size(mbmi->tx_size);
-      txfm_partition_update(xd->above_txfm_context + blk_col,
-                            xd->left_txfm_context + blk_row, sub_txs, tx_size);
-      return;
-    }
-
-    assert(bsl > 0);
-    for (i = 0; i < 4; ++i) {
-      int offsetr = blk_row + (i >> 1) * bsl;
-      int offsetc = blk_col + (i & 0x01) * bsl;
-      read_tx_size_vartx(cm, xd, mbmi, counts, sub_txs, depth + 1, offsetr,
-                         offsetc, r);
+int av1_neg_deinterleave(int diff, int ref, int max) {
+  if (!ref) return diff;
+  if (ref >= (max - 1)) return max - diff - 1;
+  if (2 * ref < max) {
+    if (diff <= 2 * ref) {
+      if (diff & 1)
+        return ref + ((diff + 1) >> 1);
+      else
+        return ref - (diff >> 1);
     }
+    return diff;
   } else {
-    int idx, idy;
-    inter_tx_size[0][0] = tx_size;
-    for (idy = 0; idy < tx_size_high_unit[tx_size] / 2; ++idy)
-      for (idx = 0; idx < tx_size_wide_unit[tx_size] / 2; ++idx)
-        inter_tx_size[idy][idx] = tx_size;
-    mbmi->tx_size = tx_size;
-    mbmi->min_tx_size = AOMMIN(mbmi->min_tx_size, get_min_tx_size(tx_size));
-    if (counts) ++counts->txfm_partition[ctx][0];
-    txfm_partition_update(xd->above_txfm_context + blk_col,
-                          xd->left_txfm_context + blk_row, tx_size, tx_size);
+    if (diff <= 2 * (max - ref - 1)) {
+      if (diff & 1)
+        return ref + ((diff + 1) >> 1);
+      else
+        return ref - (diff >> 1);
+    }
+    return max - (diff + 1);
   }
 }
-#endif
 
-static TX_SIZE read_selected_tx_size(AV1_COMMON *cm, MACROBLOCKD *xd,
-                                     int32_t tx_size_cat, aom_reader *r) {
-  FRAME_COUNTS *counts = xd->counts;
-  const int ctx = get_tx_size_context(xd);
-  FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
-  (void)cm;
-
-  const int depth = aom_read_symbol(r, ec_ctx->tx_size_cdf[tx_size_cat][ctx],
-                                    tx_size_cat + 2, ACCT_STR);
-  const TX_SIZE tx_size = depth_to_tx_size(depth);
-#if CONFIG_RECT_TX
-  assert(!is_rect_tx(tx_size));
-#endif  // CONFIG_RECT_TX
-  if (counts) ++counts->tx_size[tx_size_cat][ctx][depth];
-  return tx_size;
-}
+static int read_segment_id(AV1_COMMON *const cm, const MACROBLOCKD *const xd,
+                           int mi_row, int mi_col, aom_reader *r, int skip) {
+  int cdf_num;
+  const int pred = av1_get_spatial_seg_pred(cm, xd, mi_row, mi_col, &cdf_num);
+  if (skip) return pred;
 
-static TX_SIZE read_tx_size(AV1_COMMON *cm, MACROBLOCKD *xd, int is_inter,
-                            int allow_select_inter, aom_reader *r) {
-  const TX_MODE tx_mode = cm->tx_mode;
-  const BLOCK_SIZE bsize = xd->mi[0]->mbmi.sb_type;
-  if (xd->lossless[xd->mi[0]->mbmi.segment_id]) return TX_4X4;
-
-  if (block_signals_txsize(bsize)) {
-    if ((!is_inter || allow_select_inter) && tx_mode == TX_MODE_SELECT) {
-      const int32_t tx_size_cat = is_inter ? inter_tx_size_cat_lookup[bsize]
-                                           : intra_tx_size_cat_lookup[bsize];
-      const TX_SIZE coded_tx_size =
-          read_selected_tx_size(cm, xd, tx_size_cat, r);
-#if CONFIG_RECT_TX && (CONFIG_EXT_TX || CONFIG_VAR_TX)
-      if (coded_tx_size > max_txsize_lookup[bsize]) {
-        assert(coded_tx_size == max_txsize_lookup[bsize] + 1);
-#if CONFIG_RECT_TX_EXT
-        if (is_quarter_tx_allowed(xd, &xd->mi[0]->mbmi, is_inter)) {
-          int quarter_tx;
-
-          if (quarter_txsize_lookup[bsize] != max_txsize_lookup[bsize]) {
-#if CONFIG_NEW_MULTISYMBOL
-            quarter_tx =
-                aom_read_symbol(r, cm->fc->quarter_tx_size_cdf, 2, ACCT_STR);
-#else
-            quarter_tx = aom_read(r, cm->fc->quarter_tx_size_prob, ACCT_STR);
-            FRAME_COUNTS *counts = xd->counts;
-            if (counts) ++counts->quarter_tx_size[quarter_tx];
-#endif
-          } else {
-            quarter_tx = 1;
-          }
-          return quarter_tx ? quarter_txsize_lookup[bsize]
-                            : max_txsize_rect_lookup[bsize];
-        }
-#endif  // CONFIG_RECT_TX_EXT
-
-        return max_txsize_rect_lookup[bsize];
-      }
-#else
-      assert(coded_tx_size <= max_txsize_lookup[bsize]);
-#endif  // CONFIG_RECT_TX && (CONFIG_EXT_TX || CONFIG_VAR_TX)
-      return coded_tx_size;
-    } else {
-      return tx_size_from_tx_mode(bsize, tx_mode, is_inter);
-    }
-  } else {
-#if CONFIG_EXT_TX && CONFIG_RECT_TX
-    assert(IMPLIES(tx_mode == ONLY_4X4, bsize == BLOCK_4X4));
-    return max_txsize_rect_lookup[bsize];
-#else
-    return TX_4X4;
-#endif  // CONFIG_EXT_TX && CONFIG_RECT_TX
+  FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
+  struct segmentation *const seg = &cm->seg;
+  struct segmentation_probs *const segp = &ec_ctx->seg;
+  aom_cdf_prob *pred_cdf = segp->spatial_pred_seg_cdf[cdf_num];
+  const int coded_id = aom_read_symbol(r, pred_cdf, MAX_SEGMENTS, ACCT_STR);
+  const int segment_id =
+      av1_neg_deinterleave(coded_id, pred, seg->last_active_segid + 1);
+
+  if (segment_id < 0 || segment_id > seg->last_active_segid) {
+    aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
+                       "Corrupted segment_ids");
   }
+  return segment_id;
 }
 
 static int dec_get_segment_id(const AV1_COMMON *cm, const uint8_t *segment_ids,
                               int mi_offset, int x_mis, int y_mis) {
-  int x, y, segment_id = INT_MAX;
+  int segment_id = INT_MAX;
 
-  for (y = 0; y < y_mis; y++)
-    for (x = 0; x < x_mis; x++)
+  for (int y = 0; y < y_mis; y++)
+    for (int x = 0; x < x_mis; x++)
       segment_id =
           AOMMIN(segment_id, segment_ids[mi_offset + y * cm->mi_cols + x]);
 
@@ -575,30 +311,28 @@ static int dec_get_segment_id(const AV1_COMMON *cm, const uint8_t *segment_ids,
 
 static void set_segment_id(AV1_COMMON *cm, int mi_offset, int x_mis, int y_mis,
                            int segment_id) {
-  int x, y;
-
   assert(segment_id >= 0 && segment_id < MAX_SEGMENTS);
 
-  for (y = 0; y < y_mis; y++)
-    for (x = 0; x < x_mis; x++)
+  for (int y = 0; y < y_mis; y++)
+    for (int x = 0; x < x_mis; x++)
       cm->current_frame_seg_map[mi_offset + y * cm->mi_cols + x] = segment_id;
 }
 
-static int read_intra_segment_id(AV1_COMMON *const cm, MACROBLOCKD *const xd,
-                                 int mi_offset, int x_mis, int y_mis,
-                                 aom_reader *r) {
+static int read_intra_segment_id(AV1_COMMON *const cm,
+                                 const MACROBLOCKD *const xd, int mi_row,
+                                 int mi_col, int bsize, aom_reader *r,
+                                 int skip) {
   struct segmentation *const seg = &cm->seg;
-  FRAME_COUNTS *counts = xd->counts;
-  FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
-  struct segmentation_probs *const segp = &ec_ctx->seg;
-  int segment_id;
-
   if (!seg->enabled) return 0;  // Default for disabled segmentation
 
   assert(seg->update_map && !seg->temporal_update);
 
-  segment_id = read_segment_id(r, segp);
-  if (counts) ++counts->seg.tree_total[segment_id];
+  const int mi_offset = mi_row * cm->mi_cols + mi_col;
+  const int bw = mi_size_wide[bsize];
+  const int bh = mi_size_high[bsize];
+  const int x_mis = AOMMIN(cm->mi_cols - mi_col, bw);
+  const int y_mis = AOMMIN(cm->mi_rows - mi_row, bh);
+  const int segment_id = read_segment_id(cm, xd, mi_row, mi_col, r, skip);
   set_segment_id(cm, mi_offset, x_mis, y_mis, segment_id);
   return segment_id;
 }
@@ -607,24 +341,25 @@ static void copy_segment_id(const AV1_COMMON *cm,
                             const uint8_t *last_segment_ids,
                             uint8_t *current_segment_ids, int mi_offset,
                             int x_mis, int y_mis) {
-  int x, y;
-
-  for (y = 0; y < y_mis; y++)
-    for (x = 0; x < x_mis; x++)
+  for (int y = 0; y < y_mis; y++)
+    for (int x = 0; x < x_mis; x++)
       current_segment_ids[mi_offset + y * cm->mi_cols + x] =
           last_segment_ids ? last_segment_ids[mi_offset + y * cm->mi_cols + x]
                            : 0;
 }
 
+static int get_predicted_segment_id(AV1_COMMON *const cm, int mi_offset,
+                                    int x_mis, int y_mis) {
+  return cm->last_frame_seg_map ? dec_get_segment_id(cm, cm->last_frame_seg_map,
+                                                     mi_offset, x_mis, y_mis)
+                                : 0;
+}
+
 static int read_inter_segment_id(AV1_COMMON *const cm, MACROBLOCKD *const xd,
-                                 int mi_row, int mi_col, aom_reader *r) {
+                                 int mi_row, int mi_col, int preskip,
+                                 aom_reader *r) {
   struct segmentation *const seg = &cm->seg;
-  FRAME_COUNTS *counts = xd->counts;
-  FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
-  struct segmentation_probs *const segp = &ec_ctx->seg;
-
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
-  int predicted_segment_id, segment_id;
+  MB_MODE_INFO *const mbmi = xd->mi[0];
   const int mi_offset = mi_row * cm->mi_cols + mi_col;
   const int bw = mi_size_wide[mbmi->sb_type];
   const int bh = mi_size_high[mbmi->sb_type];
@@ -635,60 +370,82 @@ static int read_inter_segment_id(AV1_COMMON *const cm, MACROBLOCKD *const xd,
 
   if (!seg->enabled) return 0;  // Default for disabled segmentation
 
-  predicted_segment_id = cm->last_frame_seg_map
-                             ? dec_get_segment_id(cm, cm->last_frame_seg_map,
-                                                  mi_offset, x_mis, y_mis)
-                             : 0;
-
   if (!seg->update_map) {
     copy_segment_id(cm, cm->last_frame_seg_map, cm->current_frame_seg_map,
                     mi_offset, x_mis, y_mis);
-    return predicted_segment_id;
+    return get_predicted_segment_id(cm, mi_offset, x_mis, y_mis);
+  }
+
+  int segment_id;
+  if (preskip) {
+    if (!seg->segid_preskip) return 0;
+  } else {
+    if (seg->segid_preskip) return mbmi->segment_id;
+    if (mbmi->skip) {
+      if (seg->temporal_update) {
+        mbmi->seg_id_predicted = 0;
+      }
+      segment_id = read_segment_id(cm, xd, mi_row, mi_col, r, 1);
+      set_segment_id(cm, mi_offset, x_mis, y_mis, segment_id);
+      return segment_id;
+    }
   }
 
   if (seg->temporal_update) {
     const int ctx = av1_get_pred_context_seg_id(xd);
-#if CONFIG_NEW_MULTISYMBOL
+    FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
+    struct segmentation_probs *const segp = &ec_ctx->seg;
     aom_cdf_prob *pred_cdf = segp->pred_cdf[ctx];
     mbmi->seg_id_predicted = aom_read_symbol(r, pred_cdf, 2, ACCT_STR);
-#else
-    const aom_prob pred_prob = segp->pred_probs[ctx];
-    mbmi->seg_id_predicted = aom_read(r, pred_prob, ACCT_STR);
-#endif
-    if (counts) ++counts->seg.pred[ctx][mbmi->seg_id_predicted];
     if (mbmi->seg_id_predicted) {
-      segment_id = predicted_segment_id;
+      segment_id = get_predicted_segment_id(cm, mi_offset, x_mis, y_mis);
     } else {
-      segment_id = read_segment_id(r, segp);
-      if (counts) ++counts->seg.tree_mispred[segment_id];
+      segment_id = read_segment_id(cm, xd, mi_row, mi_col, r, 0);
     }
   } else {
-    segment_id = read_segment_id(r, segp);
-    if (counts) ++counts->seg.tree_total[segment_id];
+    segment_id = read_segment_id(cm, xd, mi_row, mi_col, r, 0);
   }
   set_segment_id(cm, mi_offset, x_mis, y_mis, segment_id);
   return segment_id;
 }
 
+static int read_skip_mode(AV1_COMMON *cm, const MACROBLOCKD *xd, int segment_id,
+                          aom_reader *r) {
+  if (!cm->skip_mode_flag) return 0;
+
+  if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP)) {
+    return 0;
+  }
+
+  if (!is_comp_ref_allowed(xd->mi[0]->sb_type)) return 0;
+
+  if (segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME) ||
+      segfeature_active(&cm->seg, segment_id, SEG_LVL_GLOBALMV)) {
+    // These features imply single-reference mode, while skip mode implies
+    // compound reference. Hence, the two are mutually exclusive.
+    // In other words, skip_mode is implicitly 0 here.
+    return 0;
+  }
+
+  const int ctx = av1_get_skip_mode_context(xd);
+  FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
+  const int skip_mode =
+      aom_read_symbol(r, ec_ctx->skip_mode_cdfs[ctx], 2, ACCT_STR);
+  return skip_mode;
+}
+
 static int read_skip(AV1_COMMON *cm, const MACROBLOCKD *xd, int segment_id,
                      aom_reader *r) {
   if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP)) {
     return 1;
   } else {
     const int ctx = av1_get_skip_context(xd);
-#if CONFIG_NEW_MULTISYMBOL
     FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
     const int skip = aom_read_symbol(r, ec_ctx->skip_cdfs[ctx], 2, ACCT_STR);
-#else
-    const int skip = aom_read(r, cm->fc->skip_probs[ctx], ACCT_STR);
-#endif
-    FRAME_COUNTS *counts = xd->counts;
-    if (counts) ++counts->skip[ctx][skip];
     return skip;
   }
 }
 
-#if CONFIG_PALETTE_DELTA_ENCODING
 // Merge the sorted list of cached colors(cached_colors[0...n_cached_colors-1])
 // and the sorted list of transmitted colors(colors[n_cached_colors...n-1]) into
 // one single sorted list(colors[...]).
@@ -796,346 +553,114 @@ static void read_palette_colors_uv(MACROBLOCKD *const xd, int bit_depth,
     }
   }
 }
-#endif  // CONFIG_PALETTE_DELTA_ENCODING
 
 static void read_palette_mode_info(AV1_COMMON *const cm, MACROBLOCKD *const xd,
-                                   aom_reader *r) {
-  MODE_INFO *const mi = xd->mi[0];
-  MB_MODE_INFO *const mbmi = &mi->mbmi;
-  const MODE_INFO *const above_mi = xd->above_mi;
-  const MODE_INFO *const left_mi = xd->left_mi;
+                                   int mi_row, int mi_col, aom_reader *r) {
+  const int num_planes = av1_num_planes(cm);
+  MB_MODE_INFO *const mbmi = xd->mi[0];
   const BLOCK_SIZE bsize = mbmi->sb_type;
+  assert(av1_allow_palette(cm->allow_screen_content_tools, bsize));
   PALETTE_MODE_INFO *const pmi = &mbmi->palette_mode_info;
-
-  assert(bsize >= BLOCK_8X8 && bsize <= BLOCK_LARGEST);
-  const int block_palette_idx = bsize - BLOCK_8X8;
-  int modev;
+  const int bsize_ctx = av1_get_palette_bsize_ctx(bsize);
 
   if (mbmi->mode == DC_PRED) {
-    int palette_y_mode_ctx = 0;
-    if (above_mi) {
-      palette_y_mode_ctx +=
-          (above_mi->mbmi.palette_mode_info.palette_size[0] > 0);
-    }
-    if (left_mi) {
-      palette_y_mode_ctx +=
-          (left_mi->mbmi.palette_mode_info.palette_size[0] > 0);
-    }
-#if CONFIG_NEW_MULTISYMBOL
-    modev = aom_read_symbol(
-        r,
-        xd->tile_ctx->palette_y_mode_cdf[block_palette_idx][palette_y_mode_ctx],
-        2, ACCT_STR);
-#else
-    modev = aom_read(
-        r,
-        av1_default_palette_y_mode_prob[block_palette_idx][palette_y_mode_ctx],
+    const int palette_mode_ctx = av1_get_palette_mode_ctx(xd);
+    const int modev = aom_read_symbol(
+        r, xd->tile_ctx->palette_y_mode_cdf[bsize_ctx][palette_mode_ctx], 2,
         ACCT_STR);
-#endif
     if (modev) {
       pmi->palette_size[0] =
-          aom_read_symbol(r,
-                          xd->tile_ctx->palette_y_size_cdf[block_palette_idx],
+          aom_read_symbol(r, xd->tile_ctx->palette_y_size_cdf[bsize_ctx],
                           PALETTE_SIZES, ACCT_STR) +
           2;
-#if CONFIG_PALETTE_DELTA_ENCODING
       read_palette_colors_y(xd, cm->bit_depth, pmi, r);
-#else
-      for (int i = 0; i < pmi->palette_size[0]; ++i)
-        pmi->palette_colors[i] = aom_read_literal(r, cm->bit_depth, ACCT_STR);
-#endif  // CONFIG_PALETTE_DELTA_ENCODING
     }
   }
-  if (mbmi->uv_mode == UV_DC_PRED) {
+  if (num_planes > 1 && mbmi->uv_mode == UV_DC_PRED &&
+      is_chroma_reference(mi_row, mi_col, bsize, xd->plane[1].subsampling_x,
+                          xd->plane[1].subsampling_y)) {
     const int palette_uv_mode_ctx = (pmi->palette_size[0] > 0);
-#if CONFIG_NEW_MULTISYMBOL
-    modev = aom_read_symbol(
+    const int modev = aom_read_symbol(
         r, xd->tile_ctx->palette_uv_mode_cdf[palette_uv_mode_ctx], 2, ACCT_STR);
-#else
-    modev = aom_read(r, av1_default_palette_uv_mode_prob[palette_uv_mode_ctx],
-                     ACCT_STR);
-#endif
     if (modev) {
       pmi->palette_size[1] =
-          aom_read_symbol(r,
-                          xd->tile_ctx->palette_uv_size_cdf[block_palette_idx],
+          aom_read_symbol(r, xd->tile_ctx->palette_uv_size_cdf[bsize_ctx],
                           PALETTE_SIZES, ACCT_STR) +
           2;
-#if CONFIG_PALETTE_DELTA_ENCODING
       read_palette_colors_uv(xd, cm->bit_depth, pmi, r);
-#else
-      for (int i = 0; i < pmi->palette_size[1]; ++i) {
-        pmi->palette_colors[PALETTE_MAX_SIZE + i] =
-            aom_read_literal(r, cm->bit_depth, ACCT_STR);
-        pmi->palette_colors[2 * PALETTE_MAX_SIZE + i] =
-            aom_read_literal(r, cm->bit_depth, ACCT_STR);
-      }
-#endif  // CONFIG_PALETTE_DELTA_ENCODING
     }
   }
 }
 
-#if CONFIG_FILTER_INTRA
-static void read_filter_intra_mode_info(AV1_COMMON *const cm,
-                                        MACROBLOCKD *const xd, int mi_row,
-                                        int mi_col, aom_reader *r) {
-  MODE_INFO *const mi = xd->mi[0];
-  MB_MODE_INFO *const mbmi = &mi->mbmi;
-  FRAME_COUNTS *counts = xd->counts;
+static int read_angle_delta(aom_reader *r, aom_cdf_prob *cdf) {
+  const int sym = aom_read_symbol(r, cdf, 2 * MAX_ANGLE_DELTA + 1, ACCT_STR);
+  return sym - MAX_ANGLE_DELTA;
+}
+
+static void read_filter_intra_mode_info(const AV1_COMMON *const cm,
+                                        MACROBLOCKD *const xd, aom_reader *r) {
+  MB_MODE_INFO *const mbmi = xd->mi[0];
   FILTER_INTRA_MODE_INFO *filter_intra_mode_info =
       &mbmi->filter_intra_mode_info;
 
-  if (mbmi->mode == DC_PRED && mbmi->palette_mode_info.palette_size[0] == 0) {
-    filter_intra_mode_info->use_filter_intra_mode[0] =
-        aom_read(r, cm->fc->filter_intra_probs[0], ACCT_STR);
-    if (filter_intra_mode_info->use_filter_intra_mode[0]) {
-      filter_intra_mode_info->filter_intra_mode[0] =
-          av1_read_uniform(r, FILTER_INTRA_MODES);
+  if (av1_filter_intra_allowed(cm, mbmi)) {
+    filter_intra_mode_info->use_filter_intra = aom_read_symbol(
+        r, xd->tile_ctx->filter_intra_cdfs[mbmi->sb_type], 2, ACCT_STR);
+    if (filter_intra_mode_info->use_filter_intra) {
+      filter_intra_mode_info->filter_intra_mode = aom_read_symbol(
+          r, xd->tile_ctx->filter_intra_mode_cdf, FILTER_INTRA_MODES, ACCT_STR);
     }
-    if (counts) {
-      ++counts
-            ->filter_intra[0][filter_intra_mode_info->use_filter_intra_mode[0]];
-    }
-  }
-
-#if CONFIG_CB4X4
-  if (!is_chroma_reference(mi_row, mi_col, mbmi->sb_type,
-                           xd->plane[1].subsampling_x,
-                           xd->plane[1].subsampling_y))
-    return;
-#else
-  (void)mi_row;
-  (void)mi_col;
-#endif  // CONFIG_CB4X4
-
-  if (mbmi->uv_mode == UV_DC_PRED &&
-      mbmi->palette_mode_info.palette_size[1] == 0) {
-    filter_intra_mode_info->use_filter_intra_mode[1] =
-        aom_read(r, cm->fc->filter_intra_probs[1], ACCT_STR);
-    if (filter_intra_mode_info->use_filter_intra_mode[1]) {
-      filter_intra_mode_info->filter_intra_mode[1] =
-          av1_read_uniform(r, FILTER_INTRA_MODES);
-    }
-    if (counts) {
-      ++counts
-            ->filter_intra[1][filter_intra_mode_info->use_filter_intra_mode[1]];
-    }
-  }
-}
-#endif  // CONFIG_FILTER_INTRA
-
-#if CONFIG_EXT_INTRA
-static void read_intra_angle_info(AV1_COMMON *const cm, MACROBLOCKD *const xd,
-                                  aom_reader *r) {
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
-  const BLOCK_SIZE bsize = mbmi->sb_type;
-#if CONFIG_INTRA_INTERP
-  FRAME_CONTEXT *const ec_ctx = xd->tile_ctx;
-  const int ctx = av1_get_pred_context_intra_interp(xd);
-  int p_angle;
-#endif  // CONFIG_INTRA_INTERP
-
-  (void)cm;
-
-  mbmi->angle_delta[0] = 0;
-  mbmi->angle_delta[1] = 0;
-#if CONFIG_INTRA_INTERP
-  mbmi->intra_filter = INTRA_FILTER_LINEAR;
-#endif  // CONFIG_INTRA_INTERP
-
-  if (!av1_use_angle_delta(bsize)) return;
-
-  if (av1_is_directional_mode(mbmi->mode, bsize)) {
-    mbmi->angle_delta[0] =
-        av1_read_uniform(r, 2 * MAX_ANGLE_DELTA + 1) - MAX_ANGLE_DELTA;
-#if CONFIG_INTRA_INTERP
-    p_angle = mode_to_angle_map[mbmi->mode] + mbmi->angle_delta[0] * ANGLE_STEP;
-    if (av1_is_intra_filter_switchable(p_angle)) {
-      FRAME_COUNTS *counts = xd->counts;
-      mbmi->intra_filter = aom_read_symbol(r, ec_ctx->intra_filter_cdf[ctx],
-                                           INTRA_FILTERS, ACCT_STR);
-      if (counts) ++counts->intra_filter[ctx][mbmi->intra_filter];
-    }
-#endif  // CONFIG_INTRA_INTERP
-  }
-
-  if (av1_is_directional_mode(get_uv_mode(mbmi->uv_mode), bsize)) {
-    mbmi->angle_delta[1] =
-        av1_read_uniform(r, 2 * MAX_ANGLE_DELTA + 1) - MAX_ANGLE_DELTA;
+  } else {
+    filter_intra_mode_info->use_filter_intra = 0;
   }
 }
-#endif  // CONFIG_EXT_INTRA
 
-void av1_read_tx_type(const AV1_COMMON *const cm, MACROBLOCKD *xd,
-#if CONFIG_SUPERTX
-                      int supertx_enabled,
-#endif
-#if CONFIG_TXK_SEL
-                      int blk_row, int blk_col, int block, int plane,
-                      TX_SIZE tx_size,
-#endif
-                      aom_reader *r) {
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+void av1_read_tx_type(const AV1_COMMON *const cm, MACROBLOCKD *xd, int blk_row,
+                      int blk_col, TX_SIZE tx_size, aom_reader *r) {
+  MB_MODE_INFO *mbmi = xd->mi[0];
   const int inter_block = is_inter_block(mbmi);
-#if !CONFIG_TXK_SEL
-#if CONFIG_VAR_TX
-  const TX_SIZE tx_size = inter_block ? mbmi->min_tx_size : mbmi->tx_size;
-#else
-  const TX_SIZE tx_size = mbmi->tx_size;
-#endif
-#endif  // !CONFIG_TXK_SEL
   FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
 
-#if !CONFIG_TXK_SEL
-  TX_TYPE *tx_type = &mbmi->tx_type;
-#else
-  // only y plane's tx_type is transmitted
-  if (plane > 0) return;
-  (void)block;
-  TX_TYPE *tx_type = &mbmi->txk_type[(blk_row << 4) + blk_col];
-#endif
-#if CONFIG_LGT_FROM_PRED
-  mbmi->use_lgt = 0;
-#endif
-
-  if (!FIXED_TX_TYPE) {
-#if CONFIG_EXT_TX
-    const TX_SIZE square_tx_size = txsize_sqr_map[tx_size];
-    if (get_ext_tx_types(tx_size, mbmi->sb_type, inter_block,
-                         cm->reduced_tx_set_used) > 1 &&
-        ((!cm->seg.enabled && cm->base_qindex > 0) ||
-         (cm->seg.enabled && xd->qindex[mbmi->segment_id] > 0)) &&
-        !mbmi->skip &&
-#if CONFIG_SUPERTX
-        !supertx_enabled &&
-#endif  // CONFIG_SUPERTX
-        !segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP)) {
-      const TxSetType tx_set_type = get_ext_tx_set_type(
-          tx_size, mbmi->sb_type, inter_block, cm->reduced_tx_set_used);
-      const int eset = get_ext_tx_set(tx_size, mbmi->sb_type, inter_block,
-                                      cm->reduced_tx_set_used);
-      // eset == 0 should correspond to a set with only DCT_DCT and
-      // there is no need to read the tx_type
-      assert(eset != 0);
-
-#if !CONFIG_LGT_FROM_PRED
-      if (inter_block) {
-        *tx_type = av1_ext_tx_inv[tx_set_type][aom_read_symbol(
-            r, ec_ctx->inter_ext_tx_cdf[eset][square_tx_size],
-            av1_num_ext_tx_set[tx_set_type], ACCT_STR)];
-      } else if (ALLOW_INTRA_EXT_TX) {
-        *tx_type = av1_ext_tx_inv[tx_set_type][aom_read_symbol(
-            r, ec_ctx->intra_ext_tx_cdf[eset][square_tx_size][mbmi->mode],
-            av1_num_ext_tx_set[tx_set_type], ACCT_STR)];
-      }
-#else
-      // only signal tx_type when lgt is not allowed or not selected
-      if (inter_block) {
-        if (LGT_FROM_PRED_INTER) {
-          if (is_lgt_allowed(mbmi->mode, tx_size) && !cm->reduced_tx_set_used) {
-            mbmi->use_lgt =
-                aom_read(r, ec_ctx->inter_lgt_prob[square_tx_size], ACCT_STR);
-#if CONFIG_ENTROPY_STATS
-            if (counts) ++counts->inter_lgt[square_tx_size][mbmi->use_lgt];
-#endif  // CONFIG_ENTROPY_STATS
-          }
-          if (!mbmi->use_lgt) {
-            *tx_type = av1_ext_tx_inv[tx_set_type][aom_read_symbol(
-                r, ec_ctx->inter_ext_tx_cdf[eset][square_tx_size],
-                av1_num_ext_tx_set[tx_set_type], ACCT_STR)];
-#if CONFIG_ENTROPY_STATS
-            if (counts) ++counts->inter_ext_tx[eset][square_tx_size][*tx_type];
-#endif  // CONFIG_ENTROPY_STATS
-          } else {
-            *tx_type = DCT_DCT;  // assign a dummy tx_type
-          }
-        } else {
-          *tx_type = av1_ext_tx_inv[tx_set_type][aom_read_symbol(
-              r, ec_ctx->inter_ext_tx_cdf[eset][square_tx_size],
-              av1_num_ext_tx_set[tx_set_type], ACCT_STR)];
-#if CONFIG_ENTROPY_STATS
-          if (counts) ++counts->inter_ext_tx[eset][square_tx_size][*tx_type];
-#endif  // CONFIG_ENTROPY_STATS
-        }
-      } else if (ALLOW_INTRA_EXT_TX) {
-        if (LGT_FROM_PRED_INTRA) {
-          if (is_lgt_allowed(mbmi->mode, tx_size) && !cm->reduced_tx_set_used) {
-            mbmi->use_lgt =
-                aom_read(r, ec_ctx->intra_lgt_prob[square_tx_size][mbmi->mode],
-                         ACCT_STR);
-#if CONFIG_ENTROPY_STATS
-            if (counts)
-              ++counts->intra_lgt[square_tx_size][mbmi->mode][mbmi->use_lgt];
-#endif  // CONFIG_ENTROPY_STATS
-          }
-          if (!mbmi->use_lgt) {
-            *tx_type = av1_ext_tx_inv[tx_set_type][aom_read_symbol(
-                r, ec_ctx->intra_ext_tx_cdf[eset][square_tx_size][mbmi->mode],
-                av1_num_ext_tx_set[tx_set_type], ACCT_STR)];
-#if CONFIG_ENTROPY_STATS
-            if (counts)
-              ++counts
-                    ->intra_ext_tx[eset][square_tx_size][mbmi->mode][*tx_type];
-#endif  // CONFIG_ENTROPY_STATS
-          } else {
-            *tx_type = DCT_DCT;  // assign a dummy tx_type
-          }
-        } else {
-          *tx_type = av1_ext_tx_inv[tx_set_type][aom_read_symbol(
-              r, ec_ctx->intra_ext_tx_cdf[eset][square_tx_size][mbmi->mode],
-              av1_num_ext_tx_set[tx_set_type], ACCT_STR)];
-#if CONFIG_ENTROPY_STATS
-          if (counts)
-            ++counts->intra_ext_tx[eset][square_tx_size][mbmi->mode][*tx_type];
-#endif  // CONFIG_ENTROPY_STATS
-        }
-      }
-#endif  // CONFIG_LGT_FROM_PRED
+  const int txk_type_idx =
+      av1_get_txk_type_index(mbmi->sb_type, blk_row, blk_col);
+  TX_TYPE *tx_type = &mbmi->txk_type[txk_type_idx];
+
+  const TX_SIZE square_tx_size = txsize_sqr_map[tx_size];
+  if (get_ext_tx_types(tx_size, inter_block, cm->reduced_tx_set_used) > 1 &&
+      ((!cm->seg.enabled && cm->base_qindex > 0) ||
+       (cm->seg.enabled && xd->qindex[mbmi->segment_id] > 0)) &&
+      !mbmi->skip &&
+      !segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP)) {
+    const TxSetType tx_set_type =
+        av1_get_ext_tx_set_type(tx_size, inter_block, cm->reduced_tx_set_used);
+    const int eset =
+        get_ext_tx_set(tx_size, inter_block, cm->reduced_tx_set_used);
+    // eset == 0 should correspond to a set with only DCT_DCT and
+    // there is no need to read the tx_type
+    assert(eset != 0);
+
+    if (inter_block) {
+      *tx_type = av1_ext_tx_inv[tx_set_type][aom_read_symbol(
+          r, ec_ctx->inter_ext_tx_cdf[eset][square_tx_size],
+          av1_num_ext_tx_set[tx_set_type], ACCT_STR)];
     } else {
-      *tx_type = DCT_DCT;
-    }
-#else  // CONFIG_EXT_TX
-
-    if (tx_size < TX_32X32 &&
-        ((!cm->seg.enabled && cm->base_qindex > 0) ||
-         (cm->seg.enabled && xd->qindex[mbmi->segment_id] > 0)) &&
-        !mbmi->skip &&
-#if CONFIG_SUPERTX
-        !supertx_enabled &&
-#endif  // CONFIG_SUPERTX
-        !segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP)) {
-#if CONFIG_ENTROPY_STATS
-      FRAME_COUNTS *counts = xd->counts;
-#endif  // CONFIG_ENTROPY_STATS
-      if (inter_block) {
-        *tx_type = av1_ext_tx_inv[aom_read_symbol(
-            r, ec_ctx->inter_ext_tx_cdf[tx_size], TX_TYPES, ACCT_STR)];
-#if CONFIG_ENTROPY_STATS
-        if (counts) ++counts->inter_ext_tx[tx_size][*tx_type];
-#endif  // CONFIG_ENTROPY_STATS
-      } else {
-        const TX_TYPE tx_type_nom = intra_mode_to_tx_type_context[mbmi->mode];
-        *tx_type = av1_ext_tx_inv[aom_read_symbol(
-            r, ec_ctx->intra_ext_tx_cdf[tx_size][tx_type_nom], TX_TYPES,
-            ACCT_STR)];
-#if CONFIG_ENTROPY_STATS
-        if (counts) ++counts->intra_ext_tx[tx_size][tx_type_nom][*tx_type];
-#endif  // CONFIG_ENTROPY_STATS
-      }
-    } else {
-      *tx_type = DCT_DCT;
+      PREDICTION_MODE intra_dir;
+      if (mbmi->filter_intra_mode_info.use_filter_intra)
+        intra_dir =
+            fimode_to_intradir[mbmi->filter_intra_mode_info.filter_intra_mode];
+      else
+        intra_dir = mbmi->mode;
+      *tx_type = av1_ext_tx_inv[tx_set_type][aom_read_symbol(
+          r, ec_ctx->intra_ext_tx_cdf[eset][square_tx_size][intra_dir],
+          av1_num_ext_tx_set[tx_set_type], ACCT_STR)];
     }
-#endif  // CONFIG_EXT_TX
+  } else {
+    *tx_type = DCT_DCT;
   }
-#if FIXED_TX_TYPE
-  assert(mbmi->tx_type == DCT_DCT);
-#endif
 }
 
-#if CONFIG_INTRABC
 static INLINE void read_mv(aom_reader *r, MV *mv, const MV *ref,
-                           nmv_context *ctx, nmv_context_counts *counts,
-                           MvSubpelPrecision precision);
+                           nmv_context *ctx, MvSubpelPrecision precision);
 
 static INLINE int is_mv_valid(const MV *mv);
 
@@ -1143,267 +668,195 @@ static INLINE int assign_dv(AV1_COMMON *cm, MACROBLOCKD *xd, int_mv *mv,
                             const int_mv *ref_mv, int mi_row, int mi_col,
                             BLOCK_SIZE bsize, aom_reader *r) {
   FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
-  (void)cm;
-  FRAME_COUNTS *counts = xd->counts;
-  nmv_context_counts *const dv_counts = counts ? &counts->dv : NULL;
-  read_mv(r, &mv->as_mv, &ref_mv->as_mv, &ec_ctx->ndvc, dv_counts,
-          MV_SUBPEL_NONE);
+  read_mv(r, &mv->as_mv, &ref_mv->as_mv, &ec_ctx->ndvc, MV_SUBPEL_NONE);
+  // DV should not have sub-pel.
+  assert((mv->as_mv.col & 7) == 0);
+  assert((mv->as_mv.row & 7) == 0);
+  mv->as_mv.col = (mv->as_mv.col >> 3) * 8;
+  mv->as_mv.row = (mv->as_mv.row >> 3) * 8;
   int valid = is_mv_valid(&mv->as_mv) &&
-              is_dv_valid(mv->as_mv, &xd->tile, mi_row, mi_col, bsize);
+              av1_is_dv_valid(mv->as_mv, cm, xd, mi_row, mi_col, bsize,
+                              cm->seq_params.mib_size_log2);
   return valid;
 }
-#endif  // CONFIG_INTRABC
+
+static void read_intrabc_info(AV1_COMMON *const cm, MACROBLOCKD *const xd,
+                              int mi_row, int mi_col, aom_reader *r) {
+  MB_MODE_INFO *const mbmi = xd->mi[0];
+  FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
+  mbmi->use_intrabc = aom_read_symbol(r, ec_ctx->intrabc_cdf, 2, ACCT_STR);
+  if (mbmi->use_intrabc) {
+    BLOCK_SIZE bsize = mbmi->sb_type;
+    mbmi->mode = DC_PRED;
+    mbmi->uv_mode = UV_DC_PRED;
+    mbmi->interp_filters = av1_broadcast_interp_filter(BILINEAR);
+    mbmi->motion_mode = SIMPLE_TRANSLATION;
+
+    int16_t inter_mode_ctx[MODE_CTX_REF_FRAMES];
+    int_mv ref_mvs[INTRA_FRAME + 1][MAX_MV_REF_CANDIDATES];
+    int_mv global_mvs[REF_FRAMES];
+
+    av1_find_mv_refs(cm, xd, mbmi, INTRA_FRAME, xd->ref_mv_count,
+                     xd->ref_mv_stack, ref_mvs, global_mvs, mi_row, mi_col,
+                     inter_mode_ctx);
+
+    int_mv nearestmv, nearmv;
+
+    av1_find_best_ref_mvs(0, ref_mvs[INTRA_FRAME], &nearestmv, &nearmv, 0);
+    int_mv dv_ref = nearestmv.as_int == 0 ? nearmv : nearestmv;
+    if (dv_ref.as_int == 0)
+      av1_find_ref_dv(&dv_ref, &xd->tile, cm->seq_params.mib_size, mi_row,
+                      mi_col);
+    // Ref DV should not have sub-pel.
+    int valid_dv = (dv_ref.as_mv.col & 7) == 0 && (dv_ref.as_mv.row & 7) == 0;
+    dv_ref.as_mv.col = (dv_ref.as_mv.col >> 3) * 8;
+    dv_ref.as_mv.row = (dv_ref.as_mv.row >> 3) * 8;
+    valid_dv = valid_dv && assign_dv(cm, xd, &mbmi->mv[0], &dv_ref, mi_row,
+                                     mi_col, bsize, r);
+    if (!valid_dv) {
+      // Intra bc motion vectors are not valid - signal corrupt frame
+      aom_merge_corrupted_flag(&xd->corrupted, 1);
+    }
+  }
+}
 
 static void read_intra_frame_mode_info(AV1_COMMON *const cm,
                                        MACROBLOCKD *const xd, int mi_row,
                                        int mi_col, aom_reader *r) {
-  MODE_INFO *const mi = xd->mi[0];
-  MB_MODE_INFO *const mbmi = &mi->mbmi;
-  const MODE_INFO *above_mi = xd->above_mi;
-  const MODE_INFO *left_mi = xd->left_mi;
+  MB_MODE_INFO *const mbmi = xd->mi[0];
+  const MB_MODE_INFO *above_mi = xd->above_mbmi;
+  const MB_MODE_INFO *left_mi = xd->left_mbmi;
   const BLOCK_SIZE bsize = mbmi->sb_type;
-  int i;
-  const int mi_offset = mi_row * cm->mi_cols + mi_col;
-  const int bw = mi_size_wide[bsize];
-  const int bh = mi_size_high[bsize];
+  struct segmentation *const seg = &cm->seg;
 
-  // TODO(slavarnway): move x_mis, y_mis into xd ?????
-  const int x_mis = AOMMIN(cm->mi_cols - mi_col, bw);
-  const int y_mis = AOMMIN(cm->mi_rows - mi_row, bh);
   FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
 
-  mbmi->segment_id = read_intra_segment_id(cm, xd, mi_offset, x_mis, y_mis, r);
+  if (seg->segid_preskip)
+    mbmi->segment_id =
+        read_intra_segment_id(cm, xd, mi_row, mi_col, bsize, r, 0);
+
   mbmi->skip = read_skip(cm, xd, mbmi->segment_id, r);
 
+  if (!seg->segid_preskip)
+    mbmi->segment_id =
+        read_intra_segment_id(cm, xd, mi_row, mi_col, bsize, r, mbmi->skip);
+
+  read_cdef(cm, r, xd, mi_col, mi_row);
+
   if (cm->delta_q_present_flag) {
-    xd->current_qindex =
-        xd->prev_qindex +
+    xd->current_qindex +=
         read_delta_qindex(cm, xd, r, mbmi, mi_col, mi_row) * cm->delta_q_res;
     /* Normative: Clamp to [1,MAXQ] to not interfere with lossless mode */
     xd->current_qindex = clamp(xd->current_qindex, 1, MAXQ);
-    xd->prev_qindex = xd->current_qindex;
-#if CONFIG_EXT_DELTA_Q
     if (cm->delta_lf_present_flag) {
-#if CONFIG_LOOPFILTER_LEVEL
       if (cm->delta_lf_multi) {
-        for (int lf_id = 0; lf_id < FRAME_LF_COUNT; ++lf_id) {
-          mbmi->curr_delta_lf[lf_id] = xd->curr_delta_lf[lf_id] =
-              xd->prev_delta_lf[lf_id] +
+        const int frame_lf_count =
+            av1_num_planes(cm) > 1 ? FRAME_LF_COUNT : FRAME_LF_COUNT - 2;
+        for (int lf_id = 0; lf_id < frame_lf_count; ++lf_id) {
+          const int tmp_lvl =
+              xd->delta_lf[lf_id] +
               read_delta_lflevel(cm, xd, r, lf_id, mbmi, mi_col, mi_row) *
                   cm->delta_lf_res;
-          xd->prev_delta_lf[lf_id] = xd->curr_delta_lf[lf_id];
+          mbmi->delta_lf[lf_id] = xd->delta_lf[lf_id] =
+              clamp(tmp_lvl, -MAX_LOOP_FILTER, MAX_LOOP_FILTER);
         }
       } else {
-        mbmi->current_delta_lf_from_base = xd->current_delta_lf_from_base =
-            xd->prev_delta_lf_from_base +
+        const int tmp_lvl =
+            xd->delta_lf_from_base +
             read_delta_lflevel(cm, xd, r, -1, mbmi, mi_col, mi_row) *
                 cm->delta_lf_res;
-        xd->prev_delta_lf_from_base = xd->current_delta_lf_from_base;
+        mbmi->delta_lf_from_base = xd->delta_lf_from_base =
+            clamp(tmp_lvl, -MAX_LOOP_FILTER, MAX_LOOP_FILTER);
       }
-#else
-      const int current_delta_lf_from_base =
-          xd->prev_delta_lf_from_base +
-          read_delta_lflevel(cm, xd, r, mbmi, mi_col, mi_row) *
-              cm->delta_lf_res;
-      mbmi->current_delta_lf_from_base = xd->current_delta_lf_from_base =
-          clamp(current_delta_lf_from_base, 0, MAX_LOOP_FILTER);
-      xd->prev_delta_lf_from_base = xd->current_delta_lf_from_base;
-#endif  // CONFIG_LOOPFILTER_LEVEL
     }
-#endif
   }
 
+  mbmi->current_qindex = xd->current_qindex;
+
   mbmi->ref_frame[0] = INTRA_FRAME;
   mbmi->ref_frame[1] = NONE_FRAME;
+  mbmi->palette_mode_info.palette_size[0] = 0;
+  mbmi->palette_mode_info.palette_size[1] = 0;
+  mbmi->filter_intra_mode_info.use_filter_intra = 0;
 
-#if CONFIG_INTRABC
-  if (av1_allow_intrabc(bsize, cm)) {
-    mbmi->use_intrabc = aom_read_symbol(r, ec_ctx->intrabc_cdf, 2, ACCT_STR);
-    if (mbmi->use_intrabc) {
-      mbmi->tx_size = read_tx_size(cm, xd, 1, !mbmi->skip, r);
-      mbmi->mode = mbmi->uv_mode = UV_DC_PRED;
-      mbmi->interp_filters = av1_broadcast_interp_filter(BILINEAR);
-
-      int16_t inter_mode_ctx[MODE_CTX_REF_FRAMES];
-      int_mv ref_mvs[MAX_MV_REF_CANDIDATES];
-
-      av1_find_mv_refs(cm, xd, mi, INTRA_FRAME, &xd->ref_mv_count[INTRA_FRAME],
-                       xd->ref_mv_stack[INTRA_FRAME], NULL, ref_mvs, mi_row,
-                       mi_col, NULL, NULL, inter_mode_ctx);
-
-      int_mv nearestmv, nearmv;
-      av1_find_best_ref_mvs(0, ref_mvs, &nearestmv, &nearmv);
-
-      int_mv dv_ref = nearestmv.as_int == 0 ? nearmv : nearestmv;
-      if (dv_ref.as_int == 0) av1_find_ref_dv(&dv_ref, mi_row, mi_col);
-
-      xd->corrupted |=
-          !assign_dv(cm, xd, &mbmi->mv[0], &dv_ref, mi_row, mi_col, bsize, r);
-#if CONFIG_VAR_TX
-      // TODO(aconverse@google.com): Evaluate allowing VAR TX on intrabc blocks
-      const int width = block_size_wide[bsize] >> tx_size_wide_log2[0];
-      const int height = block_size_high[bsize] >> tx_size_high_log2[0];
-      int idx, idy;
-      for (idy = 0; idy < height; ++idy)
-        for (idx = 0; idx < width; ++idx)
-          mbmi->inter_tx_size[idy >> 1][idx >> 1] = mbmi->tx_size;
-      mbmi->min_tx_size = get_min_tx_size(mbmi->tx_size);
-#endif  // CONFIG_VAR_TX
-#if CONFIG_EXT_TX && !CONFIG_TXK_SEL
-      av1_read_tx_type(cm, xd,
-#if CONFIG_SUPERTX
-                       0,
-#endif
-                       r);
-#endif  // CONFIG_EXT_TX && !CONFIG_TXK_SEL
-      return;
-    }
-  }
-#endif  // CONFIG_INTRABC
-
-  mbmi->tx_size = read_tx_size(cm, xd, 0, 1, r);
-
-#if CONFIG_CB4X4
-  (void)i;
-  mbmi->mode =
-      read_intra_mode(r, get_y_mode_cdf(ec_ctx, mi, above_mi, left_mi, 0));
-#else
-  switch (bsize) {
-    case BLOCK_4X4:
-      for (i = 0; i < 4; ++i)
-        mi->bmi[i].as_mode = read_intra_mode(
-            r, get_y_mode_cdf(ec_ctx, mi, above_mi, left_mi, i));
-      mbmi->mode = mi->bmi[3].as_mode;
-      break;
-    case BLOCK_4X8:
-      mi->bmi[0].as_mode = mi->bmi[2].as_mode =
-          read_intra_mode(r, get_y_mode_cdf(ec_ctx, mi, above_mi, left_mi, 0));
-      mi->bmi[1].as_mode = mi->bmi[3].as_mode = mbmi->mode =
-          read_intra_mode(r, get_y_mode_cdf(ec_ctx, mi, above_mi, left_mi, 1));
-      break;
-    case BLOCK_8X4:
-      mi->bmi[0].as_mode = mi->bmi[1].as_mode =
-          read_intra_mode(r, get_y_mode_cdf(ec_ctx, mi, above_mi, left_mi, 0));
-      mi->bmi[2].as_mode = mi->bmi[3].as_mode = mbmi->mode =
-          read_intra_mode(r, get_y_mode_cdf(ec_ctx, mi, above_mi, left_mi, 2));
-      break;
-    default:
-      mbmi->mode =
-          read_intra_mode(r, get_y_mode_cdf(ec_ctx, mi, above_mi, left_mi, 0));
+  xd->above_txfm_context = cm->above_txfm_context[xd->tile.tile_row] + mi_col;
+  xd->left_txfm_context =
+      xd->left_txfm_context_buffer + (mi_row & MAX_MIB_MASK);
+
+  if (av1_allow_intrabc(cm)) {
+    read_intrabc_info(cm, xd, mi_row, mi_col, r);
+    if (is_intrabc_block(mbmi)) return;
   }
-#endif
 
-#if CONFIG_CB4X4
-  if (is_chroma_reference(mi_row, mi_col, bsize, xd->plane[1].subsampling_x,
-                          xd->plane[1].subsampling_y)) {
-#if CONFIG_CFL
-    xd->cfl->is_chroma_reference = 1;
-#endif  // CONFIG_CFL
-#endif  // CONFIG_CB4X4
-    mbmi->uv_mode = read_intra_mode_uv(ec_ctx, r, mbmi->mode);
+  mbmi->mode = read_intra_mode(r, get_y_mode_cdf(ec_ctx, above_mi, left_mi));
 
-#if CONFIG_CFL
+  const int use_angle_delta = av1_use_angle_delta(bsize);
+  mbmi->angle_delta[PLANE_TYPE_Y] =
+      (use_angle_delta && av1_is_directional_mode(mbmi->mode))
+          ? read_angle_delta(r, ec_ctx->angle_delta_cdf[mbmi->mode - V_PRED])
+          : 0;
+
+  if (!cm->seq_params.monochrome &&
+      is_chroma_reference(mi_row, mi_col, bsize, xd->plane[1].subsampling_x,
+                          xd->plane[1].subsampling_y)) {
+    xd->cfl.is_chroma_reference = 1;
+    mbmi->uv_mode =
+        read_intra_mode_uv(ec_ctx, r, is_cfl_allowed(xd), mbmi->mode);
     if (mbmi->uv_mode == UV_CFL_PRED) {
       mbmi->cfl_alpha_idx = read_cfl_alphas(ec_ctx, r, &mbmi->cfl_alpha_signs);
-      xd->cfl->store_y = 1;
-    } else {
-      xd->cfl->store_y = 0;
     }
-#endif  // CONFIG_CFL
-
-#if CONFIG_CB4X4
+    mbmi->angle_delta[PLANE_TYPE_UV] =
+        (use_angle_delta && av1_is_directional_mode(get_uv_mode(mbmi->uv_mode)))
+            ? read_angle_delta(r,
+                               ec_ctx->angle_delta_cdf[mbmi->uv_mode - V_PRED])
+            : 0;
   } else {
     // Avoid decoding angle_info if there is is no chroma prediction
     mbmi->uv_mode = UV_DC_PRED;
-#if CONFIG_CFL
-    xd->cfl->is_chroma_reference = 0;
-    xd->cfl->store_y = 1;
-#endif
+    xd->cfl.is_chroma_reference = 0;
   }
-#endif
+  xd->cfl.store_y = store_cfl_required(cm, xd);
 
-#if CONFIG_EXT_INTRA
-  read_intra_angle_info(cm, xd, r);
-#endif  // CONFIG_EXT_INTRA
-  mbmi->palette_mode_info.palette_size[0] = 0;
-  mbmi->palette_mode_info.palette_size[1] = 0;
   if (av1_allow_palette(cm->allow_screen_content_tools, bsize))
-    read_palette_mode_info(cm, xd, r);
-#if CONFIG_FILTER_INTRA
-  mbmi->filter_intra_mode_info.use_filter_intra_mode[0] = 0;
-  mbmi->filter_intra_mode_info.use_filter_intra_mode[1] = 0;
-  if (bsize >= BLOCK_8X8 || CONFIG_CB4X4)
-    read_filter_intra_mode_info(cm, xd, mi_row, mi_col, r);
-#endif  // CONFIG_FILTER_INTRA
-
-#if !CONFIG_TXK_SEL
-  av1_read_tx_type(cm, xd,
-#if CONFIG_SUPERTX
-                   0,
-#endif
-                   r);
-#endif  // !CONFIG_TXK_SEL
+    read_palette_mode_info(cm, xd, mi_row, mi_col, r);
+
+  read_filter_intra_mode_info(cm, xd, r);
 }
 
 static int read_mv_component(aom_reader *r, nmv_component *mvcomp,
-#if CONFIG_INTRABC || CONFIG_AMVR
-                             int use_subpel,
-#endif  // CONFIG_INTRABC || CONFIG_AMVR
-                             int usehp) {
+                             int use_subpel, int usehp) {
   int mag, d, fr, hp;
-#if CONFIG_NEW_MULTISYMBOL
-  const int sign = aom_read_bit(r, ACCT_STR);
-#else
-  const int sign = aom_read(r, mvcomp->sign, ACCT_STR);
-#endif
+  const int sign = aom_read_symbol(r, mvcomp->sign_cdf, 2, ACCT_STR);
   const int mv_class =
-      aom_read_symbol(r, mvcomp->class_cdf, MV_CLASSES, ACCT_STR);
+      aom_read_symbol(r, mvcomp->classes_cdf, MV_CLASSES, ACCT_STR);
   const int class0 = mv_class == MV_CLASS_0;
 
   // Integer part
   if (class0) {
-#if CONFIG_NEW_MULTISYMBOL
     d = aom_read_symbol(r, mvcomp->class0_cdf, CLASS0_SIZE, ACCT_STR);
-#else
-    d = aom_read(r, mvcomp->class0[0], ACCT_STR);
-#endif
     mag = 0;
   } else {
-    int i;
     const int n = mv_class + CLASS0_BITS - 1;  // number of bits
     d = 0;
-#if CONFIG_NEW_MULTISYMBOL
-    for (i = 0; i < n; ++i)
-      d |= aom_read_symbol(r, mvcomp->bits_cdf[(i + 1) / 2], 2, ACCT_STR) << i;
-#else
-    for (i = 0; i < n; ++i) d |= aom_read(r, mvcomp->bits[i], ACCT_STR) << i;
-#endif
+    for (int i = 0; i < n; ++i)
+      d |= aom_read_symbol(r, mvcomp->bits_cdf[i], 2, ACCT_STR) << i;
     mag = CLASS0_SIZE << (mv_class + 2);
   }
 
-#if CONFIG_INTRABC || CONFIG_AMVR
   if (use_subpel) {
-#endif  // CONFIG_INTRABC || CONFIG_AMVR
-        // Fractional part
+    // Fractional part
     fr = aom_read_symbol(r, class0 ? mvcomp->class0_fp_cdf[d] : mvcomp->fp_cdf,
                          MV_FP_SIZE, ACCT_STR);
 
-// High precision part (if hp is not used, the default value of the hp is 1)
-#if CONFIG_NEW_MULTISYMBOL
+    // High precision part (if hp is not used, the default value of the hp is 1)
     hp = usehp ? aom_read_symbol(
                      r, class0 ? mvcomp->class0_hp_cdf : mvcomp->hp_cdf, 2,
                      ACCT_STR)
                : 1;
-#else
-  hp = usehp ? aom_read(r, class0 ? mvcomp->class0_hp : mvcomp->hp, ACCT_STR)
-             : 1;
-#endif
-#if CONFIG_INTRABC || CONFIG_AMVR
   } else {
     fr = 3;
     hp = 1;
   }
-#endif  // CONFIG_INTRABC || CONFIG_AMVR
 
   // Result
   mag += ((d << 3) | (fr << 1) | hp) + 1;
@@ -1411,29 +864,19 @@ static int read_mv_component(aom_reader *r, nmv_component *mvcomp,
 }
 
 static INLINE void read_mv(aom_reader *r, MV *mv, const MV *ref,
-                           nmv_context *ctx, nmv_context_counts *counts,
-                           MvSubpelPrecision precision) {
-  MV_JOINT_TYPE joint_type;
-  MV diff = { 0, 0 };
-  joint_type =
-      (MV_JOINT_TYPE)aom_read_symbol(r, ctx->joint_cdf, MV_JOINTS, ACCT_STR);
+                           nmv_context *ctx, MvSubpelPrecision precision) {
+  MV diff = kZeroMv;
+  const MV_JOINT_TYPE joint_type =
+      (MV_JOINT_TYPE)aom_read_symbol(r, ctx->joints_cdf, MV_JOINTS, ACCT_STR);
 
   if (mv_joint_vertical(joint_type))
-    diff.row = read_mv_component(r, &ctx->comps[0],
-#if CONFIG_INTRABC || CONFIG_AMVR
-                                 precision > MV_SUBPEL_NONE,
-#endif  // CONFIG_INTRABC || CONFIG_AMVR
+    diff.row = read_mv_component(r, &ctx->comps[0], precision > MV_SUBPEL_NONE,
                                  precision > MV_SUBPEL_LOW_PRECISION);
 
   if (mv_joint_horizontal(joint_type))
-    diff.col = read_mv_component(r, &ctx->comps[1],
-#if CONFIG_INTRABC || CONFIG_AMVR
-                                 precision > MV_SUBPEL_NONE,
-#endif  // CONFIG_INTRABC || CONFIG_AMVR
+    diff.col = read_mv_component(r, &ctx->comps[1], precision > MV_SUBPEL_NONE,
                                  precision > MV_SUBPEL_LOW_PRECISION);
 
-  av1_inc_mv(&diff, counts, precision);
-
   mv->row = ref->row + diff.row;
   mv->col = ref->col + diff.col;
 }
@@ -1441,138 +884,68 @@ static INLINE void read_mv(aom_reader *r, MV *mv, const MV *ref,
 static REFERENCE_MODE read_block_reference_mode(AV1_COMMON *cm,
                                                 const MACROBLOCKD *xd,
                                                 aom_reader *r) {
-  if (!is_comp_ref_allowed(xd->mi[0]->mbmi.sb_type)) return SINGLE_REFERENCE;
+  if (!is_comp_ref_allowed(xd->mi[0]->sb_type)) return SINGLE_REFERENCE;
   if (cm->reference_mode == REFERENCE_MODE_SELECT) {
-    const int ctx = av1_get_reference_mode_context(cm, xd);
-#if CONFIG_NEW_MULTISYMBOL
+    const int ctx = av1_get_reference_mode_context(xd);
     const REFERENCE_MODE mode = (REFERENCE_MODE)aom_read_symbol(
         r, xd->tile_ctx->comp_inter_cdf[ctx], 2, ACCT_STR);
-#else
-    const REFERENCE_MODE mode =
-        (REFERENCE_MODE)aom_read(r, cm->fc->comp_inter_prob[ctx], ACCT_STR);
-#endif
-    FRAME_COUNTS *counts = xd->counts;
-    if (counts) ++counts->comp_inter[ctx][mode];
     return mode;  // SINGLE_REFERENCE or COMPOUND_REFERENCE
   } else {
+    assert(cm->reference_mode == SINGLE_REFERENCE);
     return cm->reference_mode;
   }
 }
 
-#if CONFIG_NEW_MULTISYMBOL
 #define READ_REF_BIT(pname) \
-  aom_read_symbol(r, av1_get_pred_cdf_##pname(cm, xd), 2, ACCT_STR)
-#define READ_REF_BIT2(pname) \
   aom_read_symbol(r, av1_get_pred_cdf_##pname(xd), 2, ACCT_STR)
-#else
-#define READ_REF_BIT(pname) \
-  aom_read(r, av1_get_pred_prob_##pname(cm, xd), ACCT_STR)
-#define READ_REF_BIT2(pname) \
-  aom_read(r, av1_get_pred_prob_##pname(cm, xd), ACCT_STR)
-#endif
 
-#if CONFIG_EXT_COMP_REFS
-static COMP_REFERENCE_TYPE read_comp_reference_type(AV1_COMMON *cm,
-                                                    const MACROBLOCKD *xd,
+static COMP_REFERENCE_TYPE read_comp_reference_type(const MACROBLOCKD *xd,
                                                     aom_reader *r) {
   const int ctx = av1_get_comp_reference_type_context(xd);
-#if USE_UNI_COMP_REFS
-  COMP_REFERENCE_TYPE comp_ref_type;
-#if CONFIG_VAR_REFS
-  if ((L_OR_L2(cm) || L3_OR_G(cm)) && BWD_OR_ALT(cm)) {
-    if (L_AND_L2(cm) || L_AND_L3(cm) || L_AND_G(cm) || BWD_AND_ALT(cm)) {
-#endif  // CONFIG_VAR_REFS
-#if CONFIG_NEW_MULTISYMBOL
-      (void)cm;
-      comp_ref_type = (COMP_REFERENCE_TYPE)aom_read_symbol(
+  const COMP_REFERENCE_TYPE comp_ref_type =
+      (COMP_REFERENCE_TYPE)aom_read_symbol(
           r, xd->tile_ctx->comp_ref_type_cdf[ctx], 2, ACCT_STR);
-#else
-  comp_ref_type = (COMP_REFERENCE_TYPE)aom_read(
-      r, cm->fc->comp_ref_type_prob[ctx], ACCT_STR);
-#endif
-#if CONFIG_VAR_REFS
-    } else {
-      comp_ref_type = BIDIR_COMP_REFERENCE;
-    }
-  } else {
-    comp_ref_type = UNIDIR_COMP_REFERENCE;
-  }
-#endif  // CONFIG_VAR_REFS
-#else   // !USE_UNI_COMP_REFS
-  // TODO(zoeliu): Temporarily turn off uni-directional comp refs
-  const COMP_REFERENCE_TYPE comp_ref_type = BIDIR_COMP_REFERENCE;
-#endif  // USE_UNI_COMP_REFS
-  FRAME_COUNTS *counts = xd->counts;
-  if (counts) ++counts->comp_ref_type[ctx][comp_ref_type];
   return comp_ref_type;  // UNIDIR_COMP_REFERENCE or BIDIR_COMP_REFERENCE
 }
-#endif  // CONFIG_EXT_COMP_REFS
+
+static void set_ref_frames_for_skip_mode(AV1_COMMON *const cm,
+                                         MV_REFERENCE_FRAME ref_frame[2]) {
+  ref_frame[0] = LAST_FRAME + cm->ref_frame_idx_0;
+  ref_frame[1] = LAST_FRAME + cm->ref_frame_idx_1;
+}
 
 // Read the referncence frame
 static void read_ref_frames(AV1_COMMON *const cm, MACROBLOCKD *const xd,
                             aom_reader *r, int segment_id,
                             MV_REFERENCE_FRAME ref_frame[2]) {
-  FRAME_COUNTS *counts = xd->counts;
+  if (xd->mi[0]->skip_mode) {
+    set_ref_frames_for_skip_mode(cm, ref_frame);
+    return;
+  }
 
   if (segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME)) {
     ref_frame[0] = (MV_REFERENCE_FRAME)get_segdata(&cm->seg, segment_id,
                                                    SEG_LVL_REF_FRAME);
     ref_frame[1] = NONE_FRAME;
+  } else if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP) ||
+             segfeature_active(&cm->seg, segment_id, SEG_LVL_GLOBALMV)) {
+    ref_frame[0] = LAST_FRAME;
+    ref_frame[1] = NONE_FRAME;
   } else {
     const REFERENCE_MODE mode = read_block_reference_mode(cm, xd, r);
-    // FIXME(rbultje) I'm pretty sure this breaks segmentation ref frame coding
-    if (mode == COMPOUND_REFERENCE) {
-#if CONFIG_EXT_COMP_REFS
-      const COMP_REFERENCE_TYPE comp_ref_type =
-          read_comp_reference_type(cm, xd, r);
 
-#if !USE_UNI_COMP_REFS
-      // TODO(zoeliu): Temporarily turn off uni-directional comp refs
-      assert(comp_ref_type == BIDIR_COMP_REFERENCE);
-#endif  // !USE_UNI_COMP_REFS
+    if (mode == COMPOUND_REFERENCE) {
+      const COMP_REFERENCE_TYPE comp_ref_type = read_comp_reference_type(xd, r);
 
       if (comp_ref_type == UNIDIR_COMP_REFERENCE) {
-        const int ctx = av1_get_pred_context_uni_comp_ref_p(xd);
-        int bit;
-#if CONFIG_VAR_REFS
-        if ((L_AND_L2(cm) || L_AND_L3(cm) || L_AND_G(cm)) && BWD_AND_ALT(cm))
-#endif  // CONFIG_VAR_REFS
-          bit = READ_REF_BIT2(uni_comp_ref_p);
-#if CONFIG_VAR_REFS
-        else
-          bit = BWD_AND_ALT(cm);
-#endif  // CONFIG_VAR_REFS
-        if (counts) ++counts->uni_comp_ref[ctx][0][bit];
-
+        const int bit = READ_REF_BIT(uni_comp_ref_p);
         if (bit) {
           ref_frame[0] = BWDREF_FRAME;
           ref_frame[1] = ALTREF_FRAME;
         } else {
-          const int ctx1 = av1_get_pred_context_uni_comp_ref_p1(xd);
-          int bit1;
-#if CONFIG_VAR_REFS
-          if (L_AND_L2(cm) && (L_AND_L3(cm) || L_AND_G(cm)))
-#endif  // CONFIG_VAR_REFS
-            bit1 = READ_REF_BIT2(uni_comp_ref_p1);
-#if CONFIG_VAR_REFS
-          else
-            bit1 = L_AND_L3(cm) || L_AND_G(cm);
-#endif  // CONFIG_VAR_REFS
-          if (counts) ++counts->uni_comp_ref[ctx1][1][bit1];
-
+          const int bit1 = READ_REF_BIT(uni_comp_ref_p1);
           if (bit1) {
-            const int ctx2 = av1_get_pred_context_uni_comp_ref_p2(xd);
-            int bit2;
-#if CONFIG_VAR_REFS
-            if (L_AND_L3(cm) && L_AND_G(cm))
-#endif  // CONFIG_VAR_REFS
-              bit2 = READ_REF_BIT2(uni_comp_ref_p2);
-#if CONFIG_VAR_REFS
-            else
-              bit2 = L_AND_G(cm);
-#endif  // CONFIG_VAR_REFS
-            if (counts) ++counts->uni_comp_ref[ctx2][2][bit2];
-
+            const int bit2 = READ_REF_BIT(uni_comp_ref_p2);
             if (bit2) {
               ref_frame[0] = LAST_FRAME;
               ref_frame[1] = GOLDEN_FRAME;
@@ -1590,202 +963,46 @@ static void read_ref_frames(AV1_COMMON *const cm, MACROBLOCKD *const xd,
       }
 
       assert(comp_ref_type == BIDIR_COMP_REFERENCE);
-#endif  // CONFIG_EXT_COMP_REFS
 
-// Normative in decoder (for low delay)
-#if CONFIG_ONE_SIDED_COMPOUND || CONFIG_FRAME_SIGN_BIAS
       const int idx = 1;
-#else  // !(CONFIG_ONE_SIDED_COMPOUND || CONFIG_FRAME_SIGN_BIAS)
-#if CONFIG_EXT_REFS
-      const int idx = cm->ref_frame_sign_bias[cm->comp_bwd_ref[0]];
-#else   // !CONFIG_EXT_REFS
-      const int idx = cm->ref_frame_sign_bias[cm->comp_fixed_ref];
-#endif  // CONFIG_EXT_REFS
-#endif  // CONFIG_ONE_SIDED_COMPOUND || CONFIG_FRAME_SIGN_BIAS)
-
-      const int ctx = av1_get_pred_context_comp_ref_p(cm, xd);
-#if CONFIG_VAR_REFS
-      int bit;
-      // Test need to explicitly code (L,L2) vs (L3,G) branch node in tree
-      if (L_OR_L2(cm) && L3_OR_G(cm))
-        bit = READ_REF_BIT(comp_ref_p);
-      else
-        bit = L3_OR_G(cm);
-#else   // !CONFIG_VAR_REFS
       const int bit = READ_REF_BIT(comp_ref_p);
-#endif  // CONFIG_VAR_REFS
-      if (counts) ++counts->comp_ref[ctx][0][bit];
-
-#if CONFIG_EXT_REFS
       // Decode forward references.
       if (!bit) {
-        const int ctx1 = av1_get_pred_context_comp_ref_p1(cm, xd);
-#if CONFIG_VAR_REFS
-        int bit1;
-        // Test need to explicitly code (L) vs (L2) branch node in tree
-        if (L_AND_L2(cm))
-          bit1 = READ_REF_BIT(comp_ref_p1);
-        else
-          bit1 = LAST_IS_VALID(cm);
-#else   // !CONFIG_VAR_REFS
         const int bit1 = READ_REF_BIT(comp_ref_p1);
-#endif  // CONFIG_VAR_REFS
-        if (counts) ++counts->comp_ref[ctx1][1][bit1];
-        ref_frame[!idx] = cm->comp_fwd_ref[bit1 ? 0 : 1];
+        ref_frame[!idx] = cm->comp_fwd_ref[bit1 ? 1 : 0];
       } else {
-        const int ctx2 = av1_get_pred_context_comp_ref_p2(cm, xd);
-#if CONFIG_VAR_REFS
-        int bit2;
-        // Test need to explicitly code (L3) vs (G) branch node in tree
-        if (L3_AND_G(cm))
-          bit2 = READ_REF_BIT(comp_ref_p2);
-        else
-          bit2 = GOLDEN_IS_VALID(cm);
-#else   // !CONFIG_VAR_REFS
         const int bit2 = READ_REF_BIT(comp_ref_p2);
-#endif  // CONFIG_VAR_REFS
-        if (counts) ++counts->comp_ref[ctx2][2][bit2];
         ref_frame[!idx] = cm->comp_fwd_ref[bit2 ? 3 : 2];
       }
 
       // Decode backward references.
-      const int ctx_bwd = av1_get_pred_context_comp_bwdref_p(cm, xd);
-#if CONFIG_VAR_REFS
-      int bit_bwd;
-      // Test need to explicitly code (BWD/ALT2) vs (ALT) branch node in tree
-      const int bit_bwd_uncertain = BWD_OR_ALT2(cm) && ALTREF_IS_VALID(cm);
-      if (bit_bwd_uncertain)
-        bit_bwd = READ_REF_BIT(comp_bwdref_p);
-      else
-        bit_bwd = ALTREF_IS_VALID(cm);
-#else  // !CONFIG_VAR_REFS
       const int bit_bwd = READ_REF_BIT(comp_bwdref_p);
-#endif  // CONFIG_VAR_REFS
-      if (counts) ++counts->comp_bwdref[ctx_bwd][0][bit_bwd];
       if (!bit_bwd) {
-        const int ctx1_bwd = av1_get_pred_context_comp_bwdref_p1(cm, xd);
-#if CONFIG_VAR_REFS
-        int bit1_bwd;
-        if (BWD_AND_ALT2(cm))
-          bit1_bwd = READ_REF_BIT(comp_bwdref_p1);
-        else
-          bit1_bwd = ALTREF2_IS_VALID(cm);
-#else  // !CONFIG_VAR_REFS
         const int bit1_bwd = READ_REF_BIT(comp_bwdref_p1);
-#endif  // CONFIG_VAR_REFS
-        if (counts) ++counts->comp_bwdref[ctx1_bwd][1][bit1_bwd];
         ref_frame[idx] = cm->comp_bwd_ref[bit1_bwd];
       } else {
         ref_frame[idx] = cm->comp_bwd_ref[2];
       }
-#else   // !CONFIG_EXT_REFS
-      ref_frame[!idx] = cm->comp_var_ref[bit];
-      ref_frame[idx] = cm->comp_fixed_ref;
-#endif  // CONFIG_EXT_REFS
     } else if (mode == SINGLE_REFERENCE) {
-#if CONFIG_EXT_REFS
-      const int ctx0 = av1_get_pred_context_single_ref_p1(xd);
-#if CONFIG_VAR_REFS
-      int bit0;
-      // Test need to explicitly code (L,L2,L3,G) vs (BWD,ALT2,ALT) branch node
-      // in tree
-      if ((L_OR_L2(cm) || L3_OR_G(cm)) &&
-          (BWD_OR_ALT2(cm) || ALTREF_IS_VALID(cm)))
-        bit0 = READ_REF_BIT(single_ref_p1);
-      else
-        bit0 = (BWD_OR_ALT2(cm) || ALTREF_IS_VALID(cm));
-#else   // !CONFIG_VAR_REFS
       const int bit0 = READ_REF_BIT(single_ref_p1);
-#endif  // CONFIG_VAR_REFS
-      if (counts) ++counts->single_ref[ctx0][0][bit0];
-
       if (bit0) {
-        const int ctx1 = av1_get_pred_context_single_ref_p2(xd);
-#if CONFIG_VAR_REFS
-        int bit1;
-        // Test need to explicitly code (BWD/ALT2) vs (ALT) branch node in tree
-        const int bit1_uncertain = BWD_OR_ALT2(cm) && ALTREF_IS_VALID(cm);
-        if (bit1_uncertain)
-          bit1 = READ_REF_BIT(single_ref_p2);
-        else
-          bit1 = ALTREF_IS_VALID(cm);
-#else  // !CONFIG_VAR_REFS
         const int bit1 = READ_REF_BIT(single_ref_p2);
-#endif  // CONFIG_VAR_REFS
-        if (counts) ++counts->single_ref[ctx1][1][bit1];
         if (!bit1) {
-          const int ctx5 = av1_get_pred_context_single_ref_p6(xd);
-#if CONFIG_VAR_REFS
-          int bit5;
-          if (BWD_AND_ALT2(cm))
-            bit5 = READ_REF_BIT(single_ref_p6);
-          else
-            bit5 = ALTREF2_IS_VALID(cm);
-#else  // !CONFIG_VAR_REFS
           const int bit5 = READ_REF_BIT(single_ref_p6);
-#endif  // CONFIG_VAR_REFS
-          if (counts) ++counts->single_ref[ctx5][5][bit5];
           ref_frame[0] = bit5 ? ALTREF2_FRAME : BWDREF_FRAME;
         } else {
           ref_frame[0] = ALTREF_FRAME;
         }
       } else {
-        const int ctx2 = av1_get_pred_context_single_ref_p3(xd);
-#if CONFIG_VAR_REFS
-        int bit2;
-        // Test need to explicitly code (L,L2) vs (L3,G) branch node in tree
-        if (L_OR_L2(cm) && L3_OR_G(cm))
-          bit2 = READ_REF_BIT(single_ref_p3);
-        else
-          bit2 = L3_OR_G(cm);
-#else  // !CONFIG_VAR_REFS
         const int bit2 = READ_REF_BIT(single_ref_p3);
-#endif  // CONFIG_VAR_REFS
-        if (counts) ++counts->single_ref[ctx2][2][bit2];
         if (bit2) {
-          const int ctx4 = av1_get_pred_context_single_ref_p5(xd);
-#if CONFIG_VAR_REFS
-          int bit4;
-          // Test need to explicitly code (L3) vs (G) branch node in tree
-          if (L3_AND_G(cm))
-            bit4 = READ_REF_BIT(single_ref_p5);
-          else
-            bit4 = GOLDEN_IS_VALID(cm);
-#else  // !CONFIG_VAR_REFS
           const int bit4 = READ_REF_BIT(single_ref_p5);
-#endif  // CONFIG_VAR_REFS
-          if (counts) ++counts->single_ref[ctx4][4][bit4];
           ref_frame[0] = bit4 ? GOLDEN_FRAME : LAST3_FRAME;
         } else {
-          const int ctx3 = av1_get_pred_context_single_ref_p4(xd);
-#if CONFIG_VAR_REFS
-          int bit3;
-          // Test need to explicitly code (L) vs (L2) branch node in tree
-          if (L_AND_L2(cm))
-            bit3 = READ_REF_BIT(single_ref_p4);
-          else
-            bit3 = LAST2_IS_VALID(cm);
-#else  // !CONFIG_VAR_REFS
           const int bit3 = READ_REF_BIT(single_ref_p4);
-#endif  // CONFIG_VAR_REFS
-          if (counts) ++counts->single_ref[ctx3][3][bit3];
           ref_frame[0] = bit3 ? LAST2_FRAME : LAST_FRAME;
         }
       }
-#else   // !CONFIG_EXT_REFS
-      const int ctx0 = av1_get_pred_context_single_ref_p1(xd);
-      const int bit0 = READ_REF_BIT(single_ref_p1);
-      if (counts) ++counts->single_ref[ctx0][0][bit0];
-
-      if (bit0) {
-        const int ctx1 = av1_get_pred_context_single_ref_p2(xd);
-        const int bit1 = READ_REF_BIT(single_ref_p2);
-        if (counts) ++counts->single_ref[ctx1][1][bit1];
-        ref_frame[0] = bit1 ? ALTREF_FRAME : GOLDEN_FRAME;
-      } else {
-        ref_frame[0] = LAST_FRAME;
-      }
-#endif  // CONFIG_EXT_REFS
 
       ref_frame[1] = NONE_FRAME;
     } else {
@@ -1798,7 +1015,6 @@ static INLINE void read_mb_interp_filter(AV1_COMMON *const cm,
                                          MACROBLOCKD *const xd,
                                          MB_MODE_INFO *const mbmi,
                                          aom_reader *r) {
-  FRAME_COUNTS *counts = xd->counts;
   FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
 
   if (!av1_is_interp_needed(xd)) {
@@ -1809,120 +1025,68 @@ static INLINE void read_mb_interp_filter(AV1_COMMON *const cm,
   if (cm->interp_filter != SWITCHABLE) {
     mbmi->interp_filters = av1_broadcast_interp_filter(cm->interp_filter);
   } else {
-#if CONFIG_DUAL_FILTER
     InterpFilter ref0_filter[2] = { EIGHTTAP_REGULAR, EIGHTTAP_REGULAR };
     for (int dir = 0; dir < 2; ++dir) {
-      if (has_subpel_mv_component(xd->mi[0], xd, dir) ||
-          (mbmi->ref_frame[1] > INTRA_FRAME &&
-           has_subpel_mv_component(xd->mi[0], xd, dir + 2))) {
-        const int ctx = av1_get_pred_context_switchable_interp(xd, dir);
-        ref0_filter[dir] =
-            (InterpFilter)aom_read_symbol(r, ec_ctx->switchable_interp_cdf[ctx],
-                                          SWITCHABLE_FILTERS, ACCT_STR);
-        if (counts) ++counts->switchable_interp[ctx][ref0_filter[dir]];
+      const int ctx = av1_get_pred_context_switchable_interp(xd, dir);
+      ref0_filter[dir] = (InterpFilter)aom_read_symbol(
+          r, ec_ctx->switchable_interp_cdf[ctx], SWITCHABLE_FILTERS, ACCT_STR);
+      if (cm->seq_params.enable_dual_filter == 0) {
+        ref0_filter[1] = ref0_filter[0];
+        break;
       }
     }
     // The index system works as: (0, 1) -> (vertical, horizontal) filter types
     mbmi->interp_filters =
         av1_make_interp_filters(ref0_filter[0], ref0_filter[1]);
-#else   // CONFIG_DUAL_FILTER
-    const int ctx = av1_get_pred_context_switchable_interp(xd);
-    InterpFilter filter = (InterpFilter)aom_read_symbol(
-        r, ec_ctx->switchable_interp_cdf[ctx], SWITCHABLE_FILTERS, ACCT_STR);
-    mbmi->interp_filters = av1_broadcast_interp_filter(filter);
-    if (counts) ++counts->switchable_interp[ctx][filter];
-#endif  // CONFIG_DUAL_FILTER
   }
 }
 
 static void read_intra_block_mode_info(AV1_COMMON *const cm, const int mi_row,
                                        const int mi_col, MACROBLOCKD *const xd,
-                                       MODE_INFO *mi, aom_reader *r) {
-  MB_MODE_INFO *const mbmi = &mi->mbmi;
-  const BLOCK_SIZE bsize = mi->mbmi.sb_type;
-  int i;
+                                       MB_MODE_INFO *const mbmi,
+                                       aom_reader *r) {
+  const BLOCK_SIZE bsize = mbmi->sb_type;
+  const int use_angle_delta = av1_use_angle_delta(bsize);
 
   mbmi->ref_frame[0] = INTRA_FRAME;
   mbmi->ref_frame[1] = NONE_FRAME;
 
   FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
 
-#if CONFIG_CB4X4
-  (void)i;
   mbmi->mode = read_intra_mode(r, ec_ctx->y_mode_cdf[size_group_lookup[bsize]]);
-#else
-  switch (bsize) {
-    case BLOCK_4X4:
-      for (i = 0; i < 4; ++i)
-        mi->bmi[i].as_mode = read_intra_mode(r, ec_ctx->y_mode_cdf[0]);
-      mbmi->mode = mi->bmi[3].as_mode;
-      break;
-    case BLOCK_4X8:
-      mi->bmi[0].as_mode = mi->bmi[2].as_mode =
-          read_intra_mode(r, ec_ctx->y_mode_cdf[0]);
-      mi->bmi[1].as_mode = mi->bmi[3].as_mode = mbmi->mode =
-          read_intra_mode(r, ec_ctx->y_mode_cdf[0]);
-      break;
-    case BLOCK_8X4:
-      mi->bmi[0].as_mode = mi->bmi[1].as_mode =
-          read_intra_mode(r, ec_ctx->y_mode_cdf[0]);
-      mi->bmi[2].as_mode = mi->bmi[3].as_mode = mbmi->mode =
-          read_intra_mode(r, ec_ctx->y_mode_cdf[0]);
-      break;
-    default:
-      mbmi->mode =
-          read_intra_mode(r, ec_ctx->y_mode_cdf[size_group_lookup[bsize]]);
-  }
-#endif
 
-#if CONFIG_CB4X4
-  if (is_chroma_reference(mi_row, mi_col, bsize, xd->plane[1].subsampling_x,
-                          xd->plane[1].subsampling_y)) {
-    mbmi->uv_mode = read_intra_mode_uv(ec_ctx, r, mbmi->mode);
-#else
-  mbmi->uv_mode = read_intra_mode_uv(ec_ctx, r, mbmi->mode);
-  (void)mi_row;
-  (void)mi_col;
-#endif
-
-#if CONFIG_CFL
+  mbmi->angle_delta[PLANE_TYPE_Y] =
+      use_angle_delta && av1_is_directional_mode(mbmi->mode)
+          ? read_angle_delta(r, ec_ctx->angle_delta_cdf[mbmi->mode - V_PRED])
+          : 0;
+  const int has_chroma =
+      is_chroma_reference(mi_row, mi_col, bsize, xd->plane[1].subsampling_x,
+                          xd->plane[1].subsampling_y);
+  xd->cfl.is_chroma_reference = has_chroma;
+  if (!cm->seq_params.monochrome && has_chroma) {
+    mbmi->uv_mode =
+        read_intra_mode_uv(ec_ctx, r, is_cfl_allowed(xd), mbmi->mode);
     if (mbmi->uv_mode == UV_CFL_PRED) {
       mbmi->cfl_alpha_idx =
           read_cfl_alphas(xd->tile_ctx, r, &mbmi->cfl_alpha_signs);
-      xd->cfl->store_y = 1;
-    } else {
-      xd->cfl->store_y = 0;
     }
-#endif  // CONFIG_CFL
-
-#if CONFIG_CB4X4
+    mbmi->angle_delta[PLANE_TYPE_UV] =
+        use_angle_delta && av1_is_directional_mode(get_uv_mode(mbmi->uv_mode))
+            ? read_angle_delta(r,
+                               ec_ctx->angle_delta_cdf[mbmi->uv_mode - V_PRED])
+            : 0;
   } else {
     // Avoid decoding angle_info if there is is no chroma prediction
     mbmi->uv_mode = UV_DC_PRED;
-#if CONFIG_CFL
-    xd->cfl->is_chroma_reference = 0;
-    xd->cfl->store_y = 1;
-#endif
   }
-#endif
-
-  // Explicitly ignore cm here to avoid a compile warning if none of
-  // ext-intra, palette and filter-intra are enabled.
-  (void)cm;
+  xd->cfl.store_y = store_cfl_required(cm, xd);
 
-#if CONFIG_EXT_INTRA
-  read_intra_angle_info(cm, xd, r);
-#endif  // CONFIG_EXT_INTRA
   mbmi->palette_mode_info.palette_size[0] = 0;
   mbmi->palette_mode_info.palette_size[1] = 0;
   if (av1_allow_palette(cm->allow_screen_content_tools, bsize))
-    read_palette_mode_info(cm, xd, r);
-#if CONFIG_FILTER_INTRA
-  mbmi->filter_intra_mode_info.use_filter_intra_mode[0] = 0;
-  mbmi->filter_intra_mode_info.use_filter_intra_mode[1] = 0;
-  if (bsize >= BLOCK_8X8 || CONFIG_CB4X4)
-    read_filter_intra_mode_info(cm, xd, mi_row, mi_col, r);
-#endif  // CONFIG_FILTER_INTRA
+    read_palette_mode_info(cm, xd, mi_row, mi_col, r);
+
+  read_filter_intra_mode_info(cm, xd, r);
 }
 
 static INLINE int is_mv_valid(const MV *mv) {
@@ -1932,188 +1096,43 @@ static INLINE int is_mv_valid(const MV *mv) {
 
 static INLINE int assign_mv(AV1_COMMON *cm, MACROBLOCKD *xd,
                             PREDICTION_MODE mode,
-                            MV_REFERENCE_FRAME ref_frame[2], int block,
-                            int_mv mv[2], int_mv ref_mv[2],
-                            int_mv nearest_mv[2], int_mv near_mv[2], int mi_row,
-                            int mi_col, int is_compound, int allow_hp,
-                            aom_reader *r) {
-  int i;
-  int ret = 1;
+                            MV_REFERENCE_FRAME ref_frame[2], int_mv mv[2],
+                            int_mv ref_mv[2], int_mv nearest_mv[2],
+                            int_mv near_mv[2], int mi_row, int mi_col,
+                            int is_compound, int allow_hp, aom_reader *r) {
   FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
-  BLOCK_SIZE bsize = xd->mi[0]->mbmi.sb_type;
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
-#if CONFIG_CB4X4
-  int_mv *pred_mv = mbmi->pred_mv;
-  (void)block;
-#else
-  int_mv *pred_mv =
-      (bsize >= BLOCK_8X8) ? mbmi->pred_mv : xd->mi[0]->bmi[block].pred_mv;
-#endif  // CONFIG_CB4X4
-  (void)ref_frame;
-  (void)cm;
-  (void)mi_row;
-  (void)mi_col;
-  (void)bsize;
-#if CONFIG_AMVR
-  if (cm->cur_frame_mv_precision_level) {
+  MB_MODE_INFO *mbmi = xd->mi[0];
+  BLOCK_SIZE bsize = mbmi->sb_type;
+  if (cm->cur_frame_force_integer_mv) {
     allow_hp = MV_SUBPEL_NONE;
   }
-#endif
   switch (mode) {
     case NEWMV: {
-      FRAME_COUNTS *counts = xd->counts;
-      for (i = 0; i < 1 + is_compound; ++i) {
-        int8_t rf_type = av1_ref_frame_type(mbmi->ref_frame);
-        int nmv_ctx =
-            av1_nmv_ctx(xd->ref_mv_count[rf_type], xd->ref_mv_stack[rf_type], i,
-                        mbmi->ref_mv_idx);
-        nmv_context *const nmvc = &ec_ctx->nmvc[nmv_ctx];
-        nmv_context_counts *const mv_counts =
-            counts ? &counts->mv[nmv_ctx] : NULL;
-        read_mv(r, &mv[i].as_mv, &ref_mv[i].as_mv, nmvc, mv_counts, allow_hp);
-        ret = ret && is_mv_valid(&mv[i].as_mv);
-
-        pred_mv[i].as_int = ref_mv[i].as_int;
-      }
+      nmv_context *const nmvc = &ec_ctx->nmvc;
+      read_mv(r, &mv[0].as_mv, &ref_mv[0].as_mv, nmvc, allow_hp);
       break;
     }
     case NEARESTMV: {
       mv[0].as_int = nearest_mv[0].as_int;
-      if (is_compound) mv[1].as_int = nearest_mv[1].as_int;
-
-      pred_mv[0].as_int = nearest_mv[0].as_int;
-      if (is_compound) pred_mv[1].as_int = nearest_mv[1].as_int;
       break;
     }
     case NEARMV: {
       mv[0].as_int = near_mv[0].as_int;
-      if (is_compound) mv[1].as_int = near_mv[1].as_int;
-
-      pred_mv[0].as_int = near_mv[0].as_int;
-      if (is_compound) pred_mv[1].as_int = near_mv[1].as_int;
-      break;
-    }
-    case ZEROMV: {
-#if CONFIG_GLOBAL_MOTION
-      mv[0].as_int = gm_get_motion_vector(&cm->global_motion[ref_frame[0]],
-                                          cm->allow_high_precision_mv, bsize,
-                                          mi_col, mi_row, block
-#if CONFIG_AMVR
-                                          ,
-                                          cm->cur_frame_mv_precision_level
-#endif
-                                          )
-                         .as_int;
-      if (is_compound)
-        mv[1].as_int = gm_get_motion_vector(&cm->global_motion[ref_frame[1]],
-                                            cm->allow_high_precision_mv, bsize,
-                                            mi_col, mi_row, block
-#if CONFIG_AMVR
-                                            ,
-                                            cm->cur_frame_mv_precision_level
-#endif
-                                            )
-                           .as_int;
-#else
-      mv[0].as_int = 0;
-      if (is_compound) mv[1].as_int = 0;
-#endif  // CONFIG_GLOBAL_MOTION
-
-      pred_mv[0].as_int = mv[0].as_int;
-      if (is_compound) pred_mv[1].as_int = mv[1].as_int;
-      break;
-    }
-#if CONFIG_COMPOUND_SINGLEREF
-    case SR_NEAREST_NEARMV: {
-      assert(!is_compound);
-      mv[0].as_int = nearest_mv[0].as_int;
-      mv[1].as_int = near_mv[0].as_int;
-      break;
-    }
-    /*
-    case SR_NEAREST_NEWMV: {
-      assert(!is_compound);
-      mv[0].as_int = nearest_mv[0].as_int;
-
-      FRAME_COUNTS *counts = xd->counts;
-      int8_t rf_type = av1_ref_frame_type(mbmi->ref_frame);
-      int nmv_ctx = av1_nmv_ctx(xd->ref_mv_count[rf_type],
-                                xd->ref_mv_stack[rf_type], 0, mbmi->ref_mv_idx);
-      nmv_context *const nmvc = &ec_ctx->nmvc[nmv_ctx];
-      nmv_context_counts *const mv_counts =
-          counts ? &counts->mv[nmv_ctx] : NULL;
-      read_mv(r, &mv[1].as_mv, &ref_mv[0].as_mv, nmvc, mv_counts, allow_hp);
-      ret = ret && is_mv_valid(&mv[1].as_mv);
-      break;
-    }*/
-    case SR_NEAR_NEWMV: {
-      assert(!is_compound);
-      mv[0].as_int = near_mv[0].as_int;
-
-      FRAME_COUNTS *counts = xd->counts;
-      int8_t rf_type = av1_ref_frame_type(mbmi->ref_frame);
-      int nmv_ctx = av1_nmv_ctx(xd->ref_mv_count[rf_type],
-                                xd->ref_mv_stack[rf_type], 0, mbmi->ref_mv_idx);
-      nmv_context *const nmvc = &ec_ctx->nmvc[nmv_ctx];
-      nmv_context_counts *const mv_counts =
-          counts ? &counts->mv[nmv_ctx] : NULL;
-      read_mv(r, &mv[1].as_mv, &ref_mv[0].as_mv, nmvc, mv_counts, allow_hp);
-      ret = ret && is_mv_valid(&mv[1].as_mv);
-      break;
-    }
-    case SR_ZERO_NEWMV: {
-      assert(!is_compound);
-#if CONFIG_GLOBAL_MOTION
-      mv[0].as_int = gm_get_motion_vector(&cm->global_motion[ref_frame[0]],
-                                          cm->allow_high_precision_mv, bsize,
-                                          mi_col, mi_row, block)
-                         .as_int;
-#else
-      mv[0].as_int = 0;
-#endif  // CONFIG_GLOBAL_MOTION
-
-      FRAME_COUNTS *counts = xd->counts;
-      int8_t rf_type = av1_ref_frame_type(mbmi->ref_frame);
-      int nmv_ctx = av1_nmv_ctx(xd->ref_mv_count[rf_type],
-                                xd->ref_mv_stack[rf_type], 0, mbmi->ref_mv_idx);
-      nmv_context *const nmvc = &ec_ctx->nmvc[nmv_ctx];
-      nmv_context_counts *const mv_counts =
-          counts ? &counts->mv[nmv_ctx] : NULL;
-      read_mv(r, &mv[1].as_mv, &ref_mv[0].as_mv, nmvc, mv_counts, allow_hp);
-      ret = ret && is_mv_valid(&mv[1].as_mv);
       break;
     }
-    case SR_NEW_NEWMV: {
-      assert(!is_compound);
-
-      FRAME_COUNTS *counts = xd->counts;
-      for (i = 0; i < 2; ++i) {
-        int8_t rf_type = av1_ref_frame_type(mbmi->ref_frame);
-        int nmv_ctx =
-            av1_nmv_ctx(xd->ref_mv_count[rf_type], xd->ref_mv_stack[rf_type], 0,
-                        mbmi->ref_mv_idx);
-        nmv_context *const nmvc = &ec_ctx->nmvc[nmv_ctx];
-        nmv_context_counts *const mv_counts =
-            counts ? &counts->mv[nmv_ctx] : NULL;
-        read_mv(r, &mv[i].as_mv, &ref_mv[0].as_mv, nmvc, mv_counts, allow_hp);
-        ret = ret && is_mv_valid(&mv[i].as_mv);
-      }
+    case GLOBALMV: {
+      mv[0].as_int =
+          gm_get_motion_vector(&cm->global_motion[ref_frame[0]],
+                               cm->allow_high_precision_mv, bsize, mi_col,
+                               mi_row, cm->cur_frame_force_integer_mv)
+              .as_int;
       break;
     }
-#endif  // CONFIG_COMPOUND_SINGLEREF
     case NEW_NEWMV: {
-      FRAME_COUNTS *counts = xd->counts;
       assert(is_compound);
-      for (i = 0; i < 2; ++i) {
-        int8_t rf_type = av1_ref_frame_type(mbmi->ref_frame);
-        int nmv_ctx =
-            av1_nmv_ctx(xd->ref_mv_count[rf_type], xd->ref_mv_stack[rf_type], i,
-                        mbmi->ref_mv_idx);
-        nmv_context *const nmvc = &ec_ctx->nmvc[nmv_ctx];
-        nmv_context_counts *const mv_counts =
-            counts ? &counts->mv[nmv_ctx] : NULL;
-        read_mv(r, &mv[i].as_mv, &ref_mv[i].as_mv, nmvc, mv_counts, allow_hp);
-        ret = ret && is_mv_valid(&mv[i].as_mv);
+      for (int i = 0; i < 2; ++i) {
+        nmv_context *const nmvc = &ec_ctx->nmvc;
+        read_mv(r, &mv[i].as_mv, &ref_mv[i].as_mv, nmvc, allow_hp);
       }
       break;
     }
@@ -2130,984 +1149,440 @@ static INLINE int assign_mv(AV1_COMMON *cm, MACROBLOCKD *xd,
       break;
     }
     case NEW_NEARESTMV: {
-      FRAME_COUNTS *counts = xd->counts;
-      int8_t rf_type = av1_ref_frame_type(mbmi->ref_frame);
-      int nmv_ctx = av1_nmv_ctx(xd->ref_mv_count[rf_type],
-                                xd->ref_mv_stack[rf_type], 0, mbmi->ref_mv_idx);
-      nmv_context *const nmvc = &ec_ctx->nmvc[nmv_ctx];
-      nmv_context_counts *const mv_counts =
-          counts ? &counts->mv[nmv_ctx] : NULL;
-      read_mv(r, &mv[0].as_mv, &ref_mv[0].as_mv, nmvc, mv_counts, allow_hp);
+      nmv_context *const nmvc = &ec_ctx->nmvc;
+      read_mv(r, &mv[0].as_mv, &ref_mv[0].as_mv, nmvc, allow_hp);
       assert(is_compound);
-      ret = ret && is_mv_valid(&mv[0].as_mv);
       mv[1].as_int = nearest_mv[1].as_int;
       break;
     }
     case NEAREST_NEWMV: {
-      FRAME_COUNTS *counts = xd->counts;
-      int8_t rf_type = av1_ref_frame_type(mbmi->ref_frame);
-      int nmv_ctx = av1_nmv_ctx(xd->ref_mv_count[rf_type],
-                                xd->ref_mv_stack[rf_type], 1, mbmi->ref_mv_idx);
-      nmv_context_counts *const mv_counts =
-          counts ? &counts->mv[nmv_ctx] : NULL;
-      nmv_context *const nmvc = &ec_ctx->nmvc[nmv_ctx];
+      nmv_context *const nmvc = &ec_ctx->nmvc;
       mv[0].as_int = nearest_mv[0].as_int;
-      read_mv(r, &mv[1].as_mv, &ref_mv[1].as_mv, nmvc, mv_counts, allow_hp);
+      read_mv(r, &mv[1].as_mv, &ref_mv[1].as_mv, nmvc, allow_hp);
       assert(is_compound);
-      ret = ret && is_mv_valid(&mv[1].as_mv);
       break;
     }
     case NEAR_NEWMV: {
-      FRAME_COUNTS *counts = xd->counts;
-      int8_t rf_type = av1_ref_frame_type(mbmi->ref_frame);
-      int nmv_ctx = av1_nmv_ctx(xd->ref_mv_count[rf_type],
-                                xd->ref_mv_stack[rf_type], 1, mbmi->ref_mv_idx);
-      nmv_context *const nmvc = &ec_ctx->nmvc[nmv_ctx];
-      nmv_context_counts *const mv_counts =
-          counts ? &counts->mv[nmv_ctx] : NULL;
+      nmv_context *const nmvc = &ec_ctx->nmvc;
       mv[0].as_int = near_mv[0].as_int;
-      read_mv(r, &mv[1].as_mv, &ref_mv[1].as_mv, nmvc, mv_counts, allow_hp);
+      read_mv(r, &mv[1].as_mv, &ref_mv[1].as_mv, nmvc, allow_hp);
       assert(is_compound);
-
-      ret = ret && is_mv_valid(&mv[1].as_mv);
       break;
     }
     case NEW_NEARMV: {
-      FRAME_COUNTS *counts = xd->counts;
-      int8_t rf_type = av1_ref_frame_type(mbmi->ref_frame);
-      int nmv_ctx = av1_nmv_ctx(xd->ref_mv_count[rf_type],
-                                xd->ref_mv_stack[rf_type], 0, mbmi->ref_mv_idx);
-      nmv_context *const nmvc = &ec_ctx->nmvc[nmv_ctx];
-      nmv_context_counts *const mv_counts =
-          counts ? &counts->mv[nmv_ctx] : NULL;
-      read_mv(r, &mv[0].as_mv, &ref_mv[0].as_mv, nmvc, mv_counts, allow_hp);
+      nmv_context *const nmvc = &ec_ctx->nmvc;
+      read_mv(r, &mv[0].as_mv, &ref_mv[0].as_mv, nmvc, allow_hp);
       assert(is_compound);
-      ret = ret && is_mv_valid(&mv[0].as_mv);
       mv[1].as_int = near_mv[1].as_int;
       break;
     }
-    case ZERO_ZEROMV: {
+    case GLOBAL_GLOBALMV: {
       assert(is_compound);
-#if CONFIG_GLOBAL_MOTION
-      mv[0].as_int = gm_get_motion_vector(&cm->global_motion[ref_frame[0]],
-                                          cm->allow_high_precision_mv, bsize,
-                                          mi_col, mi_row, block
-#if CONFIG_AMVR
-                                          ,
-                                          cm->cur_frame_mv_precision_level
-#endif
-                                          )
-                         .as_int;
-      mv[1].as_int = gm_get_motion_vector(&cm->global_motion[ref_frame[1]],
-                                          cm->allow_high_precision_mv, bsize,
-                                          mi_col, mi_row, block
-#if CONFIG_AMVR
-                                          ,
-                                          cm->cur_frame_mv_precision_level
-#endif
-                                          )
-                         .as_int;
-#else
-      mv[0].as_int = 0;
-      mv[1].as_int = 0;
-#endif  // CONFIG_GLOBAL_MOTION
+      mv[0].as_int =
+          gm_get_motion_vector(&cm->global_motion[ref_frame[0]],
+                               cm->allow_high_precision_mv, bsize, mi_col,
+                               mi_row, cm->cur_frame_force_integer_mv)
+              .as_int;
+      mv[1].as_int =
+          gm_get_motion_vector(&cm->global_motion[ref_frame[1]],
+                               cm->allow_high_precision_mv, bsize, mi_col,
+                               mi_row, cm->cur_frame_force_integer_mv)
+              .as_int;
       break;
     }
     default: { return 0; }
   }
+
+  int ret = is_mv_valid(&mv[0].as_mv);
+  if (is_compound) {
+    ret = ret && is_mv_valid(&mv[1].as_mv);
+  }
   return ret;
 }
 
 static int read_is_inter_block(AV1_COMMON *const cm, MACROBLOCKD *const xd,
                                int segment_id, aom_reader *r) {
   if (segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME)) {
-    return get_segdata(&cm->seg, segment_id, SEG_LVL_REF_FRAME) != INTRA_FRAME;
-  } else {
-    const int ctx = av1_get_intra_inter_context(xd);
-#if CONFIG_NEW_MULTISYMBOL
-    FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
-    const int is_inter =
-        aom_read_symbol(r, ec_ctx->intra_inter_cdf[ctx], 2, ACCT_STR);
-#else
-    const int is_inter = aom_read(r, cm->fc->intra_inter_prob[ctx], ACCT_STR);
-#endif
-    FRAME_COUNTS *counts = xd->counts;
-    if (counts) ++counts->intra_inter[ctx][is_inter];
-    return is_inter;
+    const int frame = get_segdata(&cm->seg, segment_id, SEG_LVL_REF_FRAME);
+    if (frame < LAST_FRAME) return 0;
+    return frame != INTRA_FRAME;
   }
-}
-
-#if CONFIG_COMPOUND_SINGLEREF
-static int read_is_inter_singleref_comp_mode(AV1_COMMON *const cm,
-                                             MACROBLOCKD *const xd,
-                                             int segment_id, aom_reader *r) {
-  if (segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME)) return 0;
-
-  const int ctx = av1_get_inter_mode_context(xd);
-  const int is_singleref_comp_mode =
-      aom_read(r, cm->fc->comp_inter_mode_prob[ctx], ACCT_STR);
-  FRAME_COUNTS *counts = xd->counts;
-
-  if (counts) ++counts->comp_inter_mode[ctx][is_singleref_comp_mode];
-  return is_singleref_comp_mode;
-}
-#endif  // CONFIG_COMPOUND_SINGLEREF
-
-static void fpm_sync(void *const data, int mi_row) {
-  AV1Decoder *const pbi = (AV1Decoder *)data;
-  av1_frameworker_wait(pbi->frame_worker_owner, pbi->common.prev_frame,
-                       mi_row << pbi->common.mib_size_log2);
+  if (segfeature_active(&cm->seg, segment_id, SEG_LVL_GLOBALMV)) {
+    return 1;
+  }
+  const int ctx = av1_get_intra_inter_context(xd);
+  FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
+  const int is_inter =
+      aom_read_symbol(r, ec_ctx->intra_inter_cdf[ctx], 2, ACCT_STR);
+  return is_inter;
 }
 
 #if DEC_MISMATCH_DEBUG
-static void dec_dump_logs(AV1_COMMON *cm, MODE_INFO *const mi, int mi_row,
-                          int mi_col,
-                          int16_t inter_mode_ctx[MODE_CTX_REF_FRAMES],
-                          int16_t mode_ctx) {
+static void dec_dump_logs(AV1_COMMON *cm, MB_MODE_INFO *const mbmi, int mi_row,
+                          int mi_col, int16_t mode_ctx) {
   int_mv mv[2] = { { 0 } };
-  int ref;
-  MB_MODE_INFO *const mbmi = &mi->mbmi;
-  for (ref = 0; ref < 1 + has_second_ref(mbmi); ++ref)
+  for (int ref = 0; ref < 1 + has_second_ref(mbmi); ++ref)
     mv[ref].as_mv = mbmi->mv[ref].as_mv;
 
   const int16_t newmv_ctx = mode_ctx & NEWMV_CTX_MASK;
   int16_t zeromv_ctx = -1;
   int16_t refmv_ctx = -1;
   if (mbmi->mode != NEWMV) {
-    if (mode_ctx & (1 << ALL_ZERO_FLAG_OFFSET)) assert(mbmi->mode == ZEROMV);
-    zeromv_ctx = (mode_ctx >> ZEROMV_OFFSET) & ZEROMV_CTX_MASK;
-    if (mbmi->mode != ZEROMV) {
+    zeromv_ctx = (mode_ctx >> GLOBALMV_OFFSET) & GLOBALMV_CTX_MASK;
+    if (mbmi->mode != GLOBALMV)
       refmv_ctx = (mode_ctx >> REFMV_OFFSET) & REFMV_CTX_MASK;
-      if (mode_ctx & (1 << SKIP_NEARESTMV_OFFSET)) refmv_ctx = 6;
-      if (mode_ctx & (1 << SKIP_NEARMV_OFFSET)) refmv_ctx = 7;
-      if (mode_ctx & (1 << SKIP_NEARESTMV_SUB8X8_OFFSET)) refmv_ctx = 8;
-    }
   }
 
-  int8_t ref_frame_type = av1_ref_frame_type(mbmi->ref_frame);
-#define FRAME_TO_CHECK 1
+#define FRAME_TO_CHECK 11
   if (cm->current_video_frame == FRAME_TO_CHECK && cm->show_frame == 1) {
     printf(
         "=== DECODER ===: "
-        "Frame=%d, (mi_row,mi_col)=(%d,%d), mode=%d, bsize=%d, "
+        "Frame=%d, (mi_row,mi_col)=(%d,%d), skip_mode=%d, mode=%d, bsize=%d, "
         "show_frame=%d, mv[0]=(%d,%d), mv[1]=(%d,%d), ref[0]=%d, "
-        "ref[1]=%d, motion_mode=%d, inter_mode_ctx=%d, mode_ctx=%d, "
-        "newmv_ctx=%d, zeromv_ctx=%d, refmv_ctx=%d\n",
-        cm->current_video_frame, mi_row, mi_col, mbmi->mode, mbmi->sb_type,
-        cm->show_frame, mv[0].as_mv.row, mv[0].as_mv.col, mv[1].as_mv.row,
-        mv[1].as_mv.col, mbmi->ref_frame[0], mbmi->ref_frame[1],
-        mbmi->motion_mode, inter_mode_ctx[ref_frame_type], mode_ctx, newmv_ctx,
-        zeromv_ctx, refmv_ctx);
+        "ref[1]=%d, motion_mode=%d, mode_ctx=%d, "
+        "newmv_ctx=%d, zeromv_ctx=%d, refmv_ctx=%d, tx_size=%d\n",
+        cm->current_video_frame, mi_row, mi_col, mbmi->skip_mode, mbmi->mode,
+        mbmi->sb_type, cm->show_frame, mv[0].as_mv.row, mv[0].as_mv.col,
+        mv[1].as_mv.row, mv[1].as_mv.col, mbmi->ref_frame[0],
+        mbmi->ref_frame[1], mbmi->motion_mode, mode_ctx, newmv_ctx, zeromv_ctx,
+        refmv_ctx, mbmi->tx_size);
   }
 }
 #endif  // DEC_MISMATCH_DEBUG
 
 static void read_inter_block_mode_info(AV1Decoder *const pbi,
                                        MACROBLOCKD *const xd,
-                                       MODE_INFO *const mi,
-#if CONFIG_SUPERTX
-                                       int mi_row, int mi_col, aom_reader *r,
-                                       int supertx_enabled) {
-#else
-                                       int mi_row, int mi_col, aom_reader *r) {
-#endif  // CONFIG_MOTION_VAR && CONFIG_SUPERTX
+                                       MB_MODE_INFO *const mbmi, int mi_row,
+                                       int mi_col, aom_reader *r) {
   AV1_COMMON *const cm = &pbi->common;
-  MB_MODE_INFO *const mbmi = &mi->mbmi;
   const BLOCK_SIZE bsize = mbmi->sb_type;
   const int allow_hp = cm->allow_high_precision_mv;
-  const int unify_bsize = CONFIG_CB4X4;
   int_mv nearestmv[2], nearmv[2];
-  int_mv ref_mvs[MODE_CTX_REF_FRAMES][MAX_MV_REF_CANDIDATES];
-  int ref, is_compound;
-#if CONFIG_COMPOUND_SINGLEREF
-  int is_singleref_comp_mode = 0;
-#endif  // CONFIG_COMPOUND_SINGLEREF
+  int_mv ref_mvs[MODE_CTX_REF_FRAMES][MAX_MV_REF_CANDIDATES] = { { { 0 } } };
   int16_t inter_mode_ctx[MODE_CTX_REF_FRAMES];
-  int16_t compound_inter_mode_ctx[MODE_CTX_REF_FRAMES];
-  int16_t mode_ctx = 0;
-#if CONFIG_WARPED_MOTION
   int pts[SAMPLES_ARRAY_SIZE], pts_inref[SAMPLES_ARRAY_SIZE];
-#if WARPED_MOTION_SORT_SAMPLES
-  int pts_mv[SAMPLES_ARRAY_SIZE];
-#endif  // WARPED_MOTION_SORT_SAMPLES
-#endif  // CONFIG_WARPED_MOTION
   FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
 
-  assert(NELEMENTS(mode_2_counter) == MB_MODE_COUNT);
-
   mbmi->uv_mode = UV_DC_PRED;
   mbmi->palette_mode_info.palette_size[0] = 0;
   mbmi->palette_mode_info.palette_size[1] = 0;
 
-  memset(ref_mvs, 0, sizeof(ref_mvs));
+  av1_collect_neighbors_ref_counts(xd);
 
   read_ref_frames(cm, xd, r, mbmi->segment_id, mbmi->ref_frame);
-  is_compound = has_second_ref(mbmi);
-
-#if CONFIG_EXT_COMP_REFS
-#if !USE_UNI_COMP_REFS
-  // NOTE: uni-directional comp refs disabled
-  if (is_compound)
-    assert(mbmi->ref_frame[0] < BWDREF_FRAME &&
-           mbmi->ref_frame[1] >= BWDREF_FRAME);
-#endif  // !USE_UNI_COMP_REFS
-#endif  // CONFIG_EXT_COMP_REFS
-
-#if CONFIG_COMPOUND_SINGLEREF
-  if (!is_compound)
-    is_singleref_comp_mode =
-        read_is_inter_singleref_comp_mode(cm, xd, mbmi->segment_id, r);
-#endif  // CONFIG_COMPOUND_SINGLEREF
-
-  for (ref = 0; ref < 1 + is_compound; ++ref) {
-    MV_REFERENCE_FRAME frame = mbmi->ref_frame[ref];
-
-    av1_find_mv_refs(cm, xd, mi, frame, &xd->ref_mv_count[frame],
-                     xd->ref_mv_stack[frame], compound_inter_mode_ctx,
-                     ref_mvs[frame], mi_row, mi_col, fpm_sync, (void *)pbi,
-                     inter_mode_ctx);
-  }
+  const int is_compound = has_second_ref(mbmi);
 
-  if (is_compound) {
-    MV_REFERENCE_FRAME ref_frame = av1_ref_frame_type(mbmi->ref_frame);
-    av1_find_mv_refs(cm, xd, mi, ref_frame, &xd->ref_mv_count[ref_frame],
-                     xd->ref_mv_stack[ref_frame], compound_inter_mode_ctx,
-                     ref_mvs[ref_frame], mi_row, mi_col, fpm_sync, (void *)pbi,
-                     inter_mode_ctx);
+  MV_REFERENCE_FRAME ref_frame = av1_ref_frame_type(mbmi->ref_frame);
+  int_mv global_mvs[REF_FRAMES];
+  av1_find_mv_refs(cm, xd, mbmi, ref_frame, xd->ref_mv_count, xd->ref_mv_stack,
+                   ref_mvs, global_mvs, mi_row, mi_col, inter_mode_ctx);
 
-    if (xd->ref_mv_count[ref_frame] < 2) {
-      MV_REFERENCE_FRAME rf[2];
-      int_mv zeromv[2];
-      av1_set_ref_frame(rf, ref_frame);
-#if CONFIG_GLOBAL_MOTION
-      zeromv[0].as_int = gm_get_motion_vector(&cm->global_motion[rf[0]],
-                                              cm->allow_high_precision_mv,
-                                              bsize, mi_col, mi_row, 0
-#if CONFIG_AMVR
-                                              ,
-                                              cm->cur_frame_mv_precision_level
-#endif
-                                              )
-                             .as_int;
-      zeromv[1].as_int =
-          (rf[1] != NONE_FRAME)
-              ? gm_get_motion_vector(&cm->global_motion[rf[1]],
-                                     cm->allow_high_precision_mv, bsize, mi_col,
-                                     mi_row, 0
-#if CONFIG_AMVR
-                                     ,
-                                     cm->cur_frame_mv_precision_level
-#endif
-                                     )
-                    .as_int
-              : 0;
-#else
-      zeromv[0].as_int = zeromv[1].as_int = 0;
-#endif
-      for (ref = 0; ref < 2; ++ref) {
-        if (rf[ref] == NONE_FRAME) continue;
-#if CONFIG_AMVR
-        lower_mv_precision(&ref_mvs[rf[ref]][0].as_mv, allow_hp,
-                           cm->cur_frame_mv_precision_level);
-        lower_mv_precision(&ref_mvs[rf[ref]][1].as_mv, allow_hp,
-                           cm->cur_frame_mv_precision_level);
-#else
-        lower_mv_precision(&ref_mvs[rf[ref]][0].as_mv, allow_hp);
-        lower_mv_precision(&ref_mvs[rf[ref]][1].as_mv, allow_hp);
-#endif
-        if (ref_mvs[rf[ref]][0].as_int != zeromv[ref].as_int ||
-            ref_mvs[rf[ref]][1].as_int != zeromv[ref].as_int)
-          inter_mode_ctx[ref_frame] &= ~(1 << ALL_ZERO_FLAG_OFFSET);
-      }
-    }
-  }
-
-#if CONFIG_COMPOUND_SINGLEREF
-  if (is_compound || is_singleref_comp_mode)
-#else   // !CONFIG_COMPOUND_SINGLEREF
-  if (is_compound)
-#endif  // CONFIG_COMPOUND_SINGLEREF
-    mode_ctx = compound_inter_mode_ctx[mbmi->ref_frame[0]];
-  else
-    mode_ctx =
-        av1_mode_context_analyzer(inter_mode_ctx, mbmi->ref_frame, bsize, -1);
+  int mode_ctx = av1_mode_context_analyzer(inter_mode_ctx, mbmi->ref_frame);
   mbmi->ref_mv_idx = 0;
 
-#if CONFIG_SEGMENT_ZEROMV
-  if (segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP) ||
-      segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_ZEROMV)) {
-#else
-  if (segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP)) {
-#endif
-    mbmi->mode = ZEROMV;
-    if (bsize < BLOCK_8X8 && !unify_bsize) {
-      aom_internal_error(xd->error_info, AOM_CODEC_UNSUP_BITSTREAM,
-                         "Invalid usage of segment feature on small blocks");
-      return;
-    }
+  if (mbmi->skip_mode) {
+    assert(is_compound);
+    mbmi->mode = NEAREST_NEARESTMV;
   } else {
-    if (bsize >= BLOCK_8X8 || unify_bsize) {
+    if (segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP) ||
+        segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_GLOBALMV)) {
+      mbmi->mode = GLOBALMV;
+    } else {
       if (is_compound)
-        mbmi->mode = read_inter_compound_mode(cm, xd, r, mode_ctx);
-#if CONFIG_COMPOUND_SINGLEREF
-      else if (is_singleref_comp_mode)
-        mbmi->mode = read_inter_singleref_comp_mode(xd, r, mode_ctx);
-#endif  // CONFIG_COMPOUND_SINGLEREF
+        mbmi->mode = read_inter_compound_mode(xd, r, mode_ctx);
       else
-        mbmi->mode = read_inter_mode(ec_ctx, xd, r, mode_ctx);
+        mbmi->mode = read_inter_mode(ec_ctx, r, mode_ctx);
       if (mbmi->mode == NEWMV || mbmi->mode == NEW_NEWMV ||
-#if CONFIG_COMPOUND_SINGLEREF
-          mbmi->mode == SR_NEW_NEWMV ||
-#endif  // CONFIG_COMPOUND_SINGLEREF
           have_nearmv_in_inter_mode(mbmi->mode))
         read_drl_idx(ec_ctx, xd, mbmi, r);
     }
   }
 
-  if ((bsize < BLOCK_8X8 && !unify_bsize) ||
-      (mbmi->mode != ZEROMV && mbmi->mode != ZERO_ZEROMV)) {
-    for (ref = 0; ref < 1 + is_compound; ++ref) {
-#if CONFIG_AMVR
-      av1_find_best_ref_mvs(allow_hp, ref_mvs[mbmi->ref_frame[ref]],
-                            &nearestmv[ref], &nearmv[ref],
-                            cm->cur_frame_mv_precision_level);
-#else
-      av1_find_best_ref_mvs(allow_hp, ref_mvs[mbmi->ref_frame[ref]],
-                            &nearestmv[ref], &nearmv[ref]);
-#endif
-    }
+  if (is_compound != is_inter_compound_mode(mbmi->mode)) {
+    aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
+                       "Prediction mode %d invalid with ref frame %d %d",
+                       mbmi->mode, mbmi->ref_frame[0], mbmi->ref_frame[1]);
   }
 
-#if CONFIG_COMPOUND_SINGLEREF
-  if ((is_compound || is_singleref_comp_mode) &&
-      (bsize >= BLOCK_8X8 || unify_bsize) && mbmi->mode != ZERO_ZEROMV)
-#else   // !CONFIG_COMPOUND_SINGLEREF
-  if (is_compound && (bsize >= BLOCK_8X8 || unify_bsize) &&
-      mbmi->mode != ZERO_ZEROMV)
-#endif  // CONFIG_COMPOUND_SINGLEREF
-  {
-    uint8_t ref_frame_type = av1_ref_frame_type(mbmi->ref_frame);
-
-    if (xd->ref_mv_count[ref_frame_type] > 0) {
-      if (mbmi->mode == NEAREST_NEARESTMV) {
-        nearestmv[0] = xd->ref_mv_stack[ref_frame_type][0].this_mv;
-        nearestmv[1] = xd->ref_mv_stack[ref_frame_type][0].comp_mv;
-#if CONFIG_AMVR
-        lower_mv_precision(&nearestmv[0].as_mv, allow_hp,
-                           cm->cur_frame_mv_precision_level);
-        lower_mv_precision(&nearestmv[1].as_mv, allow_hp,
-                           cm->cur_frame_mv_precision_level);
-#else
-        lower_mv_precision(&nearestmv[0].as_mv, allow_hp);
-        lower_mv_precision(&nearestmv[1].as_mv, allow_hp);
-#endif
-      } else if (mbmi->mode == NEAREST_NEWMV
-#if CONFIG_COMPOUND_SINGLEREF
-                 || mbmi->mode == SR_NEAREST_NEARMV
-// || mbmi->mode == SR_NEAREST_NEWMV
-#endif  // CONFIG_COMPOUND_SINGLEREF
-                 ) {
-        nearestmv[0] = xd->ref_mv_stack[ref_frame_type][0].this_mv;
-
-#if CONFIG_AMVR
-        lower_mv_precision(&nearestmv[0].as_mv, allow_hp,
-                           cm->cur_frame_mv_precision_level);
-#else
-        lower_mv_precision(&nearestmv[0].as_mv, allow_hp);
-#endif
-      } else if (mbmi->mode == NEW_NEARESTMV) {
-        nearestmv[1] = xd->ref_mv_stack[ref_frame_type][0].comp_mv;
-#if CONFIG_AMVR
-        lower_mv_precision(&nearestmv[1].as_mv, allow_hp,
-                           cm->cur_frame_mv_precision_level);
-#else
-        lower_mv_precision(&nearestmv[1].as_mv, allow_hp);
-#endif
-      }
-    }
-
-    if (xd->ref_mv_count[ref_frame_type] > 1) {
-      int ref_mv_idx = 1 + mbmi->ref_mv_idx;
-#if CONFIG_COMPOUND_SINGLEREF
-      if (is_compound) {
-#endif  // CONFIG_COMPOUND_SINGLEREF
-        if (compound_ref0_mode(mbmi->mode) == NEARMV) {
-          nearmv[0] = xd->ref_mv_stack[ref_frame_type][ref_mv_idx].this_mv;
-#if CONFIG_AMVR
-          lower_mv_precision(&nearmv[0].as_mv, allow_hp,
-                             cm->cur_frame_mv_precision_level);
-#else
-        lower_mv_precision(&nearmv[0].as_mv, allow_hp);
-#endif
-        }
+  if (!is_compound && mbmi->mode != GLOBALMV) {
+    av1_find_best_ref_mvs(allow_hp, ref_mvs[mbmi->ref_frame[0]], &nearestmv[0],
+                          &nearmv[0], cm->cur_frame_force_integer_mv);
+  }
 
-        if (compound_ref1_mode(mbmi->mode) == NEARMV) {
-          nearmv[1] = xd->ref_mv_stack[ref_frame_type][ref_mv_idx].comp_mv;
-#if CONFIG_AMVR
-          lower_mv_precision(&nearmv[1].as_mv, allow_hp,
-                             cm->cur_frame_mv_precision_level);
-#else
-        lower_mv_precision(&nearmv[1].as_mv, allow_hp);
-#endif
-        }
-#if CONFIG_COMPOUND_SINGLEREF
-      } else {
-        assert(is_singleref_comp_mode);
-        if (compound_ref0_mode(mbmi->mode) == NEARMV ||
-            compound_ref1_mode(mbmi->mode) == NEARMV) {
-          nearmv[0] = xd->ref_mv_stack[ref_frame_type][ref_mv_idx].this_mv;
-          lower_mv_precision(&nearmv[0].as_mv, allow_hp);
-        }
-      }
-#endif  // CONFIG_COMPOUND_SINGLEREF
-    }
+  if (is_compound && mbmi->mode != GLOBAL_GLOBALMV) {
+    int ref_mv_idx = mbmi->ref_mv_idx + 1;
+    nearestmv[0] = xd->ref_mv_stack[ref_frame][0].this_mv;
+    nearestmv[1] = xd->ref_mv_stack[ref_frame][0].comp_mv;
+    nearmv[0] = xd->ref_mv_stack[ref_frame][ref_mv_idx].this_mv;
+    nearmv[1] = xd->ref_mv_stack[ref_frame][ref_mv_idx].comp_mv;
+    lower_mv_precision(&nearestmv[0].as_mv, allow_hp,
+                       cm->cur_frame_force_integer_mv);
+    lower_mv_precision(&nearestmv[1].as_mv, allow_hp,
+                       cm->cur_frame_force_integer_mv);
+    lower_mv_precision(&nearmv[0].as_mv, allow_hp,
+                       cm->cur_frame_force_integer_mv);
+    lower_mv_precision(&nearmv[1].as_mv, allow_hp,
+                       cm->cur_frame_force_integer_mv);
   } else if (mbmi->ref_mv_idx > 0 && mbmi->mode == NEARMV) {
     int_mv cur_mv =
         xd->ref_mv_stack[mbmi->ref_frame[0]][1 + mbmi->ref_mv_idx].this_mv;
     nearmv[0] = cur_mv;
   }
 
-#if !CONFIG_DUAL_FILTER && !CONFIG_WARPED_MOTION && !CONFIG_GLOBAL_MOTION
-  read_mb_interp_filter(cm, xd, mbmi, r);
-#endif  // !CONFIG_DUAL_FILTER && !CONFIG_WARPED_MOTION
-
-  if (bsize < BLOCK_8X8 && !unify_bsize) {
-    const int num_4x4_w = 1 << xd->bmode_blocks_wl;
-    const int num_4x4_h = 1 << xd->bmode_blocks_hl;
-    int idx, idy;
-    PREDICTION_MODE b_mode;
-    int_mv nearest_sub8x8[2], near_sub8x8[2];
-    int_mv ref_mv[2][2];
-    for (idy = 0; idy < 2; idy += num_4x4_h) {
-      for (idx = 0; idx < 2; idx += num_4x4_w) {
-        int_mv block[2];
-        const int j = idy * 2 + idx;
-        int_mv ref_mv_s8[2];
-        if (!is_compound)
-          mode_ctx = av1_mode_context_analyzer(inter_mode_ctx, mbmi->ref_frame,
-                                               bsize, j);
-        if (is_compound)
-          b_mode = read_inter_compound_mode(cm, xd, r, mode_ctx);
-        else
-          b_mode = read_inter_mode(ec_ctx, xd, r, mode_ctx);
-
-        if (b_mode != ZEROMV && b_mode != ZERO_ZEROMV) {
-          CANDIDATE_MV ref_mv_stack[2][MAX_REF_MV_STACK_SIZE];
-          uint8_t ref_mv_count[2];
-          for (ref = 0; ref < 1 + is_compound; ++ref) {
-            int_mv mv_ref_list[MAX_MV_REF_CANDIDATES];
-            av1_update_mv_context(cm, xd, mi, mbmi->ref_frame[ref], mv_ref_list,
-                                  j, mi_row, mi_col, NULL);
-            av1_append_sub8x8_mvs_for_idx(cm, xd, j, ref, mi_row, mi_col,
-                                          ref_mv_stack[ref], &ref_mv_count[ref],
-                                          mv_ref_list, &nearest_sub8x8[ref],
-                                          &near_sub8x8[ref]);
-            if (have_newmv_in_inter_mode(b_mode)) {
-              mv_ref_list[0].as_int = nearest_sub8x8[ref].as_int;
-              mv_ref_list[1].as_int = near_sub8x8[ref].as_int;
-#if CONFIG_AMVR
-              av1_find_best_ref_mvs(allow_hp, mv_ref_list, &ref_mv[0][ref],
-                                    &ref_mv[1][ref],
-                                    cm->cur_frame_mv_precision_level);
-#else
-              av1_find_best_ref_mvs(allow_hp, mv_ref_list, &ref_mv[0][ref],
-                                    &ref_mv[1][ref]);
-#endif
-            }
-          }
-        }
-
-        for (ref = 0; ref < 1 + is_compound && b_mode != ZEROMV; ++ref) {
-          ref_mv_s8[ref] = nearest_sub8x8[ref];
-#if CONFIG_AMVR
-          lower_mv_precision(&ref_mv_s8[ref].as_mv, allow_hp,
-                             cm->cur_frame_mv_precision_level);
-#else
-          lower_mv_precision(&ref_mv_s8[ref].as_mv, allow_hp);
-#endif
-        }
-        (void)ref_mv_s8;
-
-        if (!assign_mv(cm, xd, b_mode, mbmi->ref_frame, j, block, ref_mv[0],
-                       nearest_sub8x8, near_sub8x8, mi_row, mi_col, is_compound,
-                       allow_hp, r)) {
-          aom_merge_corrupted_flag(&xd->corrupted, 1);
-          break;
-        };
-
-        mi->bmi[j].as_mv[0].as_int = block[0].as_int;
-        mi->bmi[j].as_mode = b_mode;
-        if (is_compound) mi->bmi[j].as_mv[1].as_int = block[1].as_int;
-
-        if (num_4x4_h == 2) mi->bmi[j + 2] = mi->bmi[j];
-        if (num_4x4_w == 2) mi->bmi[j + 1] = mi->bmi[j];
-      }
-    }
+  int_mv ref_mv[2];
+  ref_mv[0] = nearestmv[0];
+  ref_mv[1] = nearestmv[1];
 
-    mbmi->pred_mv[0].as_int = mi->bmi[3].pred_mv[0].as_int;
-    mbmi->pred_mv[1].as_int = mi->bmi[3].pred_mv[1].as_int;
-    mi->mbmi.mode = b_mode;
-
-    mbmi->mv[0].as_int = mi->bmi[3].as_mv[0].as_int;
-    mbmi->mv[1].as_int = mi->bmi[3].as_mv[1].as_int;
+  if (is_compound) {
+    int ref_mv_idx = mbmi->ref_mv_idx;
+    // Special case: NEAR_NEWMV and NEW_NEARMV modes use
+    // 1 + mbmi->ref_mv_idx (like NEARMV) instead of
+    // mbmi->ref_mv_idx (like NEWMV)
+    if (mbmi->mode == NEAR_NEWMV || mbmi->mode == NEW_NEARMV)
+      ref_mv_idx = 1 + mbmi->ref_mv_idx;
+
+    // TODO(jingning, yunqing): Do we need a lower_mv_precision() call here?
+    if (compound_ref0_mode(mbmi->mode) == NEWMV)
+      ref_mv[0] = xd->ref_mv_stack[ref_frame][ref_mv_idx].this_mv;
+
+    if (compound_ref1_mode(mbmi->mode) == NEWMV)
+      ref_mv[1] = xd->ref_mv_stack[ref_frame][ref_mv_idx].comp_mv;
   } else {
-    int_mv ref_mv[2];
-    ref_mv[0] = nearestmv[0];
-    ref_mv[1] = nearestmv[1];
-
-    if (is_compound) {
-      int ref_mv_idx = mbmi->ref_mv_idx;
-      // Special case: NEAR_NEWMV and NEW_NEARMV modes use
-      // 1 + mbmi->ref_mv_idx (like NEARMV) instead of
-      // mbmi->ref_mv_idx (like NEWMV)
-      if (mbmi->mode == NEAR_NEWMV || mbmi->mode == NEW_NEARMV)
-        ref_mv_idx = 1 + mbmi->ref_mv_idx;
-
-      if (compound_ref0_mode(mbmi->mode) == NEWMV) {
-        uint8_t ref_frame_type = av1_ref_frame_type(mbmi->ref_frame);
-        if (xd->ref_mv_count[ref_frame_type] > 1) {
-          ref_mv[0] = xd->ref_mv_stack[ref_frame_type][ref_mv_idx].this_mv;
-          clamp_mv_ref(&ref_mv[0].as_mv, xd->n8_w << MI_SIZE_LOG2,
-                       xd->n8_h << MI_SIZE_LOG2, xd);
-        }
-        nearestmv[0] = ref_mv[0];
-      }
-      if (compound_ref1_mode(mbmi->mode) == NEWMV) {
-        uint8_t ref_frame_type = av1_ref_frame_type(mbmi->ref_frame);
-        if (xd->ref_mv_count[ref_frame_type] > 1) {
-          ref_mv[1] = xd->ref_mv_stack[ref_frame_type][ref_mv_idx].comp_mv;
-          clamp_mv_ref(&ref_mv[1].as_mv, xd->n8_w << MI_SIZE_LOG2,
-                       xd->n8_h << MI_SIZE_LOG2, xd);
-        }
-        nearestmv[1] = ref_mv[1];
-      }
-#if CONFIG_COMPOUND_SINGLEREF
-    } else if (is_singleref_comp_mode) {
-      int ref_mv_idx = mbmi->ref_mv_idx;
-      // Special case: SR_NEAR_NEWMV use 1 + mbmi->ref_mv_idx (like NEARMV)
-      //               instead of mbmi->ref_mv_idx (like NEWMV)
-      if (mbmi->mode == SR_NEAR_NEWMV) ref_mv_idx = 1 + mbmi->ref_mv_idx;
-
-      if (compound_ref0_mode(mbmi->mode) == NEWMV ||
-          compound_ref1_mode(mbmi->mode) == NEWMV) {
-        uint8_t ref_frame_type = av1_ref_frame_type(mbmi->ref_frame);
-        if (xd->ref_mv_count[ref_frame_type] > 1) {
-          ref_mv[0] = xd->ref_mv_stack[ref_frame_type][ref_mv_idx].this_mv;
-          clamp_mv_ref(&ref_mv[0].as_mv, xd->n8_w << MI_SIZE_LOG2,
-                       xd->n8_h << MI_SIZE_LOG2, xd);
-        }
-        // TODO(zoeliu): To further investigate why this would not cause a
-        //               mismatch for the mode of SR_NEAREST_NEWMV.
-        nearestmv[0] = ref_mv[0];
-      }
-#endif  // CONFIG_COMPOUND_SINGLEREF
-    } else {
-      if (mbmi->mode == NEWMV) {
-        for (ref = 0; ref < 1 + is_compound; ++ref) {
-          uint8_t ref_frame_type = av1_ref_frame_type(mbmi->ref_frame);
-          if (xd->ref_mv_count[ref_frame_type] > 1) {
-            ref_mv[ref] =
-                (ref == 0)
-                    ? xd->ref_mv_stack[ref_frame_type][mbmi->ref_mv_idx].this_mv
-                    : xd->ref_mv_stack[ref_frame_type][mbmi->ref_mv_idx]
-                          .comp_mv;
-            clamp_mv_ref(&ref_mv[ref].as_mv, xd->n8_w << MI_SIZE_LOG2,
-                         xd->n8_h << MI_SIZE_LOG2, xd);
-          }
-          nearestmv[ref] = ref_mv[ref];
-        }
-      }
+    if (mbmi->mode == NEWMV) {
+      if (xd->ref_mv_count[ref_frame] > 1)
+        ref_mv[0] = xd->ref_mv_stack[ref_frame][mbmi->ref_mv_idx].this_mv;
     }
+  }
 
+  if (mbmi->skip_mode) {
+    assert(mbmi->mode == NEAREST_NEARESTMV);
+    mbmi->mv[0].as_int = nearestmv[0].as_int;
+    mbmi->mv[1].as_int = nearestmv[1].as_int;
+  } else {
     int mv_corrupted_flag =
-        !assign_mv(cm, xd, mbmi->mode, mbmi->ref_frame, 0, mbmi->mv, ref_mv,
+        !assign_mv(cm, xd, mbmi->mode, mbmi->ref_frame, mbmi->mv, ref_mv,
                    nearestmv, nearmv, mi_row, mi_col, is_compound, allow_hp, r);
     aom_merge_corrupted_flag(&xd->corrupted, mv_corrupted_flag);
   }
 
-#if CONFIG_INTERINTRA
   mbmi->use_wedge_interintra = 0;
-  if (cm->reference_mode != COMPOUND_REFERENCE &&
-#if CONFIG_SUPERTX
-      !supertx_enabled &&
-#endif
-      cm->allow_interintra_compound && is_interintra_allowed(mbmi)) {
+  if (cm->seq_params.enable_interintra_compound && !mbmi->skip_mode &&
+      is_interintra_allowed(mbmi)) {
     const int bsize_group = size_group_lookup[bsize];
-#if CONFIG_NEW_MULTISYMBOL
     const int interintra =
         aom_read_symbol(r, ec_ctx->interintra_cdf[bsize_group], 2, ACCT_STR);
-#else
-    const int interintra =
-        aom_read(r, cm->fc->interintra_prob[bsize_group], ACCT_STR);
-#endif
-    if (xd->counts) xd->counts->interintra[bsize_group][interintra]++;
     assert(mbmi->ref_frame[1] == NONE_FRAME);
     if (interintra) {
       const INTERINTRA_MODE interintra_mode =
-          read_interintra_mode(cm, xd, r, bsize_group);
+          read_interintra_mode(xd, r, bsize_group);
       mbmi->ref_frame[1] = INTRA_FRAME;
       mbmi->interintra_mode = interintra_mode;
-#if CONFIG_EXT_INTRA
-      mbmi->angle_delta[0] = 0;
-      mbmi->angle_delta[1] = 0;
-#if CONFIG_INTRA_INTERP
-      mbmi->intra_filter = INTRA_FILTER_LINEAR;
-#endif  // CONFIG_INTRA_INTERP
-#endif  // CONFIG_EXT_INTRA
-#if CONFIG_FILTER_INTRA
-      mbmi->filter_intra_mode_info.use_filter_intra_mode[0] = 0;
-      mbmi->filter_intra_mode_info.use_filter_intra_mode[1] = 0;
-#endif  // CONFIG_FILTER_INTRA
+      mbmi->angle_delta[PLANE_TYPE_Y] = 0;
+      mbmi->angle_delta[PLANE_TYPE_UV] = 0;
+      mbmi->filter_intra_mode_info.use_filter_intra = 0;
       if (is_interintra_wedge_used(bsize)) {
-#if CONFIG_NEW_MULTISYMBOL
         mbmi->use_wedge_interintra = aom_read_symbol(
             r, ec_ctx->wedge_interintra_cdf[bsize], 2, ACCT_STR);
-#else
-        mbmi->use_wedge_interintra =
-            aom_read(r, cm->fc->wedge_interintra_prob[bsize], ACCT_STR);
-#endif
-        if (xd->counts)
-          xd->counts->wedge_interintra[bsize][mbmi->use_wedge_interintra]++;
         if (mbmi->use_wedge_interintra) {
           mbmi->interintra_wedge_index =
-              aom_read_literal(r, get_wedge_bits_lookup(bsize), ACCT_STR);
+              aom_read_symbol(r, ec_ctx->wedge_idx_cdf[bsize], 16, ACCT_STR);
           mbmi->interintra_wedge_sign = 0;
         }
       }
     }
   }
-#endif  // CONFIG_INTERINTRA
 
-#if CONFIG_WARPED_MOTION
-  for (ref = 0; ref < 1 + has_second_ref(mbmi); ++ref) {
+  for (int ref = 0; ref < 1 + has_second_ref(mbmi); ++ref) {
     const MV_REFERENCE_FRAME frame = mbmi->ref_frame[ref];
     RefBuffer *ref_buf = &cm->frame_refs[frame - LAST_FRAME];
 
     xd->block_refs[ref] = ref_buf;
   }
-#endif
 
-#if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
   mbmi->motion_mode = SIMPLE_TRANSLATION;
-#if CONFIG_WARPED_MOTION
-  if (mbmi->sb_type >= BLOCK_8X8 && !has_second_ref(mbmi))
-#if WARPED_MOTION_SORT_SAMPLES
-    mbmi->num_proj_ref[0] =
-        findSamples(cm, xd, mi_row, mi_col, pts, pts_inref, pts_mv);
-#else
+  if (is_motion_variation_allowed_bsize(mbmi->sb_type) && !mbmi->skip_mode &&
+      !has_second_ref(mbmi))
     mbmi->num_proj_ref[0] = findSamples(cm, xd, mi_row, mi_col, pts, pts_inref);
-#endif  // WARPED_MOTION_SORT_SAMPLES
-#endif  // CONFIG_WARPED_MOTION
-#if CONFIG_MOTION_VAR
   av1_count_overlappable_neighbors(cm, xd, mi_row, mi_col);
-#endif
 
-#if CONFIG_SUPERTX
-  if (!supertx_enabled) {
-#endif  // CONFIG_SUPERTX
-    if (mbmi->ref_frame[1] != INTRA_FRAME)
-      mbmi->motion_mode = read_motion_mode(cm, xd, mi, r);
+  if (mbmi->ref_frame[1] != INTRA_FRAME)
+    mbmi->motion_mode = read_motion_mode(cm, xd, mbmi, r);
 
-#if CONFIG_NCOBMC_ADAPT_WEIGHT
-    read_ncobmc_mode(xd, mi, mbmi->ncobmc_mode, r);
-#endif
+  // init
+  mbmi->comp_group_idx = 0;
+  mbmi->compound_idx = 1;
+  mbmi->interinter_comp.type = COMPOUND_AVERAGE;
 
-#if CONFIG_COMPOUND_SINGLEREF
-    if (is_singleref_comp_mode) assert(mbmi->motion_mode == SIMPLE_TRANSLATION);
-#endif  // CONFIG_COMPOUND_SINGLEREF
-#if CONFIG_WARPED_MOTION
-    if (mbmi->motion_mode == WARPED_CAUSAL) {
-      mbmi->wm_params[0].wmtype = DEFAULT_WMTYPE;
-
-#if WARPED_MOTION_SORT_SAMPLES
-      if (mbmi->num_proj_ref[0] > 1)
-        mbmi->num_proj_ref[0] = sortSamples(pts_mv, &mbmi->mv[0].as_mv, pts,
-                                            pts_inref, mbmi->num_proj_ref[0]);
-#endif  // WARPED_MOTION_SORT_SAMPLES
-
-      if (find_projection(mbmi->num_proj_ref[0], pts, pts_inref, bsize,
-                          mbmi->mv[0].as_mv.row, mbmi->mv[0].as_mv.col,
-                          &mbmi->wm_params[0], mi_row, mi_col)) {
-        aom_internal_error(&cm->error, AOM_CODEC_ERROR, "Invalid Warped Model");
-      }
+  if (has_second_ref(mbmi) && !mbmi->skip_mode) {
+    // Read idx to indicate current compound inter prediction mode group
+    const int masked_compound_used = is_any_masked_compound_used(bsize) &&
+                                     cm->seq_params.enable_masked_compound;
+
+    if (masked_compound_used) {
+      const int ctx_comp_group_idx = get_comp_group_idx_context(xd);
+      mbmi->comp_group_idx = aom_read_symbol(
+          r, ec_ctx->comp_group_idx_cdf[ctx_comp_group_idx], 2, ACCT_STR);
     }
-#endif  // CONFIG_WARPED_MOTION
-#if CONFIG_SUPERTX
-  }
-#endif  // CONFIG_SUPERTX
-#endif  // CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
-
-  mbmi->interinter_compound_type = COMPOUND_AVERAGE;
-  if (
-#if CONFIG_COMPOUND_SINGLEREF
-      is_inter_anyref_comp_mode(mbmi->mode)
-#else   // !CONFIG_COMPOUND_SINGLEREF
-      cm->reference_mode != SINGLE_REFERENCE &&
-      is_inter_compound_mode(mbmi->mode)
-#endif  // CONFIG_COMPOUND_SINGLEREF
-#if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
-      && mbmi->motion_mode == SIMPLE_TRANSLATION
-#endif  // CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
-      ) {
-    if (is_any_masked_compound_used(bsize)) {
-#if CONFIG_COMPOUND_SEGMENT || CONFIG_WEDGE
-      if (cm->allow_masked_compound) {
-#if CONFIG_WEDGE && CONFIG_COMPOUND_SEGMENT
-        if (!is_interinter_compound_used(COMPOUND_WEDGE, bsize))
-          mbmi->interinter_compound_type =
-              aom_read_bit(r, ACCT_STR) ? COMPOUND_AVERAGE : COMPOUND_SEG;
-        else
-#endif  // CONFIG_WEDGE && CONFIG_COMPOUND_SEGMENT
-          mbmi->interinter_compound_type = aom_read_symbol(
-              r, ec_ctx->compound_type_cdf[bsize], COMPOUND_TYPES, ACCT_STR);
-#if CONFIG_WEDGE
-        if (mbmi->interinter_compound_type == COMPOUND_WEDGE) {
-          assert(is_interinter_compound_used(COMPOUND_WEDGE, bsize));
-          mbmi->wedge_index =
-              aom_read_literal(r, get_wedge_bits_lookup(bsize), ACCT_STR);
-          mbmi->wedge_sign = aom_read_bit(r, ACCT_STR);
-        }
-#endif  // CONFIG_WEDGE
-#if CONFIG_COMPOUND_SEGMENT
-        if (mbmi->interinter_compound_type == COMPOUND_SEG) {
-          mbmi->mask_type = aom_read_literal(r, MAX_SEG_MASK_BITS, ACCT_STR);
-        }
-#endif  // CONFIG_COMPOUND_SEGMENT
+
+    if (mbmi->comp_group_idx == 0) {
+      if (cm->seq_params.enable_jnt_comp) {
+        const int comp_index_ctx = get_comp_index_context(cm, xd);
+        mbmi->compound_idx = aom_read_symbol(
+            r, ec_ctx->compound_index_cdf[comp_index_ctx], 2, ACCT_STR);
+      } else {
+        // Distance-weighted compound is disabled, so always use average
+        mbmi->compound_idx = 1;
       }
-#endif  // CONFIG_COMPOUND_SEGMENT || CONFIG_WEDGE
     } else {
-      mbmi->interinter_compound_type = COMPOUND_AVERAGE;
+      assert(cm->reference_mode != SINGLE_REFERENCE &&
+             is_inter_compound_mode(mbmi->mode) &&
+             mbmi->motion_mode == SIMPLE_TRANSLATION);
+      assert(masked_compound_used);
+
+      // compound_diffwtd, wedge
+      if (is_interinter_compound_used(COMPOUND_WEDGE, bsize))
+        mbmi->interinter_comp.type =
+            1 + aom_read_symbol(r, ec_ctx->compound_type_cdf[bsize],
+                                COMPOUND_TYPES - 1, ACCT_STR);
+      else
+        mbmi->interinter_comp.type = COMPOUND_DIFFWTD;
+
+      if (mbmi->interinter_comp.type == COMPOUND_WEDGE) {
+        assert(is_interinter_compound_used(COMPOUND_WEDGE, bsize));
+        mbmi->interinter_comp.wedge_index =
+            aom_read_symbol(r, ec_ctx->wedge_idx_cdf[bsize], 16, ACCT_STR);
+        mbmi->interinter_comp.wedge_sign = aom_read_bit(r, ACCT_STR);
+      } else {
+        assert(mbmi->interinter_comp.type == COMPOUND_DIFFWTD);
+        mbmi->interinter_comp.mask_type =
+            aom_read_literal(r, MAX_DIFFWTD_MASK_BITS, ACCT_STR);
+      }
     }
-    if (xd->counts)
-      xd->counts->compound_interinter[bsize][mbmi->interinter_compound_type]++;
   }
 
-#if CONFIG_DUAL_FILTER || CONFIG_WARPED_MOTION || CONFIG_GLOBAL_MOTION
   read_mb_interp_filter(cm, xd, mbmi, r);
-#endif  // CONFIG_DUAL_FILTER || CONFIG_WARPED_MOTION
+
+  if (mbmi->motion_mode == WARPED_CAUSAL) {
+    mbmi->wm_params[0].wmtype = DEFAULT_WMTYPE;
+    mbmi->wm_params[0].invalid = 0;
+
+    if (mbmi->num_proj_ref[0] > 1)
+      mbmi->num_proj_ref[0] = selectSamples(&mbmi->mv[0].as_mv, pts, pts_inref,
+                                            mbmi->num_proj_ref[0], bsize);
+
+    if (find_projection(mbmi->num_proj_ref[0], pts, pts_inref, bsize,
+                        mbmi->mv[0].as_mv.row, mbmi->mv[0].as_mv.col,
+                        &mbmi->wm_params[0], mi_row, mi_col)) {
+#if WARPED_MOTION_DEBUG
+      printf("Warning: unexpected warped model from aomenc\n");
+#endif
+      mbmi->wm_params[0].invalid = 1;
+    }
+  }
+
+  xd->cfl.is_chroma_reference = is_chroma_reference(
+      mi_row, mi_col, bsize, cm->subsampling_x, cm->subsampling_y);
+  xd->cfl.store_y = store_cfl_required(cm, xd);
 
 #if DEC_MISMATCH_DEBUG
-  dec_dump_logs(cm, mi, mi_row, mi_col, inter_mode_ctx, mode_ctx);
+  dec_dump_logs(cm, mi, mi_row, mi_col, mode_ctx);
 #endif  // DEC_MISMATCH_DEBUG
 }
 
 static void read_inter_frame_mode_info(AV1Decoder *const pbi,
-                                       MACROBLOCKD *const xd,
-#if CONFIG_SUPERTX
-                                       int supertx_enabled,
-#endif  // CONFIG_SUPERTX
-                                       int mi_row, int mi_col, aom_reader *r) {
+                                       MACROBLOCKD *const xd, int mi_row,
+                                       int mi_col, aom_reader *r) {
   AV1_COMMON *const cm = &pbi->common;
-  MODE_INFO *const mi = xd->mi[0];
-  MB_MODE_INFO *const mbmi = &mi->mbmi;
+  MB_MODE_INFO *const mbmi = xd->mi[0];
   int inter_block = 1;
-#if CONFIG_VAR_TX
-  BLOCK_SIZE bsize = mbmi->sb_type;
-#endif  // CONFIG_VAR_TX
 
   mbmi->mv[0].as_int = 0;
   mbmi->mv[1].as_int = 0;
-  mbmi->segment_id = read_inter_segment_id(cm, xd, mi_row, mi_col, r);
-#if CONFIG_SUPERTX
-  if (!supertx_enabled)
-#endif  // CONFIG_SUPERTX
+  mbmi->segment_id = read_inter_segment_id(cm, xd, mi_row, mi_col, 1, r);
+
+  mbmi->skip_mode = read_skip_mode(cm, xd, mbmi->segment_id, r);
+
+  if (mbmi->skip_mode)
+    mbmi->skip = 1;
+  else
     mbmi->skip = read_skip(cm, xd, mbmi->segment_id, r);
 
+  mbmi->segment_id = read_inter_segment_id(cm, xd, mi_row, mi_col, 0, r);
+
+  read_cdef(cm, r, xd, mi_col, mi_row);
+
   if (cm->delta_q_present_flag) {
-    xd->current_qindex =
-        xd->prev_qindex +
+    xd->current_qindex +=
         read_delta_qindex(cm, xd, r, mbmi, mi_col, mi_row) * cm->delta_q_res;
     /* Normative: Clamp to [1,MAXQ] to not interfere with lossless mode */
     xd->current_qindex = clamp(xd->current_qindex, 1, MAXQ);
-    xd->prev_qindex = xd->current_qindex;
-#if CONFIG_EXT_DELTA_Q
     if (cm->delta_lf_present_flag) {
-#if CONFIG_LOOPFILTER_LEVEL
       if (cm->delta_lf_multi) {
-        for (int lf_id = 0; lf_id < FRAME_LF_COUNT; ++lf_id) {
-          mbmi->curr_delta_lf[lf_id] = xd->curr_delta_lf[lf_id] =
-              xd->prev_delta_lf[lf_id] +
+        const int frame_lf_count =
+            av1_num_planes(cm) > 1 ? FRAME_LF_COUNT : FRAME_LF_COUNT - 2;
+        for (int lf_id = 0; lf_id < frame_lf_count; ++lf_id) {
+          const int tmp_lvl =
+              xd->delta_lf[lf_id] +
               read_delta_lflevel(cm, xd, r, lf_id, mbmi, mi_col, mi_row) *
                   cm->delta_lf_res;
-          xd->prev_delta_lf[lf_id] = xd->curr_delta_lf[lf_id];
+          mbmi->delta_lf[lf_id] = xd->delta_lf[lf_id] =
+              clamp(tmp_lvl, -MAX_LOOP_FILTER, MAX_LOOP_FILTER);
         }
       } else {
-        mbmi->current_delta_lf_from_base = xd->current_delta_lf_from_base =
-            xd->prev_delta_lf_from_base +
+        const int tmp_lvl =
+            xd->delta_lf_from_base +
             read_delta_lflevel(cm, xd, r, -1, mbmi, mi_col, mi_row) *
                 cm->delta_lf_res;
-        xd->prev_delta_lf_from_base = xd->current_delta_lf_from_base;
+        mbmi->delta_lf_from_base = xd->delta_lf_from_base =
+            clamp(tmp_lvl, -MAX_LOOP_FILTER, MAX_LOOP_FILTER);
       }
-#else
-      const int current_delta_lf_from_base =
-          xd->prev_delta_lf_from_base +
-          read_delta_lflevel(cm, xd, r, mbmi, mi_col, mi_row) *
-              cm->delta_lf_res;
-      mbmi->current_delta_lf_from_base = xd->current_delta_lf_from_base =
-          clamp(current_delta_lf_from_base, 0, MAX_LOOP_FILTER);
-      xd->prev_delta_lf_from_base = xd->current_delta_lf_from_base;
-#endif  // CONFIG_LOOPFILTER_LEVEL
     }
-#endif
   }
 
-#if CONFIG_SUPERTX
-  if (!supertx_enabled) {
-#endif  // CONFIG_SUPERTX
+  if (!mbmi->skip_mode)
     inter_block = read_is_inter_block(cm, xd, mbmi->segment_id, r);
 
-#if CONFIG_VAR_TX
-    xd->above_txfm_context =
-        cm->above_txfm_context + (mi_col << TX_UNIT_WIDE_LOG2);
-    xd->left_txfm_context = xd->left_txfm_context_buffer +
-                            ((mi_row & MAX_MIB_MASK) << TX_UNIT_HIGH_LOG2);
-
-    if (cm->tx_mode == TX_MODE_SELECT &&
-#if CONFIG_CB4X4
-        bsize > BLOCK_4X4 &&
-#else
-        bsize >= BLOCK_8X8 &&
-#endif
-        !mbmi->skip && inter_block && !xd->lossless[mbmi->segment_id]) {
-      const TX_SIZE max_tx_size = max_txsize_rect_lookup[bsize];
-      const int bh = tx_size_high_unit[max_tx_size];
-      const int bw = tx_size_wide_unit[max_tx_size];
-      const int width = block_size_wide[bsize] >> tx_size_wide_log2[0];
-      const int height = block_size_high[bsize] >> tx_size_wide_log2[0];
-      int idx, idy;
-      int init_depth =
-          (height != width) ? RECT_VARTX_DEPTH_INIT : SQR_VARTX_DEPTH_INIT;
-
-      mbmi->min_tx_size = TX_SIZES_ALL;
-      for (idy = 0; idy < height; idy += bh)
-        for (idx = 0; idx < width; idx += bw)
-          read_tx_size_vartx(cm, xd, mbmi, xd->counts, max_tx_size, init_depth,
-                             idy, idx, r);
-#if CONFIG_RECT_TX_EXT
-      if (is_quarter_tx_allowed(xd, mbmi, inter_block) &&
-          mbmi->tx_size == max_tx_size) {
-        int quarter_tx;
-
-        if (quarter_txsize_lookup[bsize] != max_tx_size) {
-#if CONFIG_NEW_MULTISYMBOL
-          quarter_tx =
-              aom_read_symbol(r, cm->fc->quarter_tx_size_cdf, 2, ACCT_STR);
-#else
-          quarter_tx = aom_read(r, cm->fc->quarter_tx_size_prob, ACCT_STR);
-          if (xd->counts) ++xd->counts->quarter_tx_size[quarter_tx];
-#endif
-        } else {
-          quarter_tx = 1;
-        }
-        if (quarter_tx) {
-          mbmi->tx_size = quarter_txsize_lookup[bsize];
-          for (idy = 0; idy < tx_size_high_unit[max_tx_size] / 2; ++idy)
-            for (idx = 0; idx < tx_size_wide_unit[max_tx_size] / 2; ++idx)
-              mbmi->inter_tx_size[idy][idx] = mbmi->tx_size;
-          mbmi->min_tx_size = get_min_tx_size(mbmi->tx_size);
-        }
-      }
-#endif
-    } else {
-      mbmi->tx_size = read_tx_size(cm, xd, inter_block, !mbmi->skip, r);
-
-      if (inter_block) {
-        const int width = block_size_wide[bsize] >> tx_size_wide_log2[0];
-        const int height = block_size_high[bsize] >> tx_size_high_log2[0];
-        int idx, idy;
-        for (idy = 0; idy < height; ++idy)
-          for (idx = 0; idx < width; ++idx)
-            mbmi->inter_tx_size[idy >> 1][idx >> 1] = mbmi->tx_size;
-      }
-      mbmi->min_tx_size = get_min_tx_size(mbmi->tx_size);
-      set_txfm_ctxs(mbmi->tx_size, xd->n8_w, xd->n8_h, mbmi->skip, xd);
-    }
-#else
-  mbmi->tx_size = read_tx_size(cm, xd, inter_block, !mbmi->skip, r);
-#endif  // CONFIG_VAR_TX
-#if CONFIG_SUPERTX
-  }
-#if CONFIG_VAR_TX
-  else if (inter_block) {
-    const int width = num_4x4_blocks_wide_lookup[bsize];
-    const int height = num_4x4_blocks_high_lookup[bsize];
-    int idx, idy;
-    xd->mi[0]->mbmi.tx_size = xd->supertx_size;
-    for (idy = 0; idy < height; ++idy)
-      for (idx = 0; idx < width; ++idx)
-        xd->mi[0]->mbmi.inter_tx_size[idy >> 1][idx >> 1] = xd->supertx_size;
-  }
-#endif  // CONFIG_VAR_TX
-#endif  // CONFIG_SUPERTX
+  mbmi->current_qindex = xd->current_qindex;
+
+  xd->above_txfm_context = cm->above_txfm_context[xd->tile.tile_row] + mi_col;
+  xd->left_txfm_context =
+      xd->left_txfm_context_buffer + (mi_row & MAX_MIB_MASK);
 
   if (inter_block)
-    read_inter_block_mode_info(pbi, xd,
-#if CONFIG_SUPERTX
-                               mi, mi_row, mi_col, r, supertx_enabled);
-#else
-                               mi, mi_row, mi_col, r);
-#endif  // CONFIG_MOTION_VAR && CONFIG_SUPERTX
+    read_inter_block_mode_info(pbi, xd, mbmi, mi_row, mi_col, r);
   else
-    read_intra_block_mode_info(cm, mi_row, mi_col, xd, mi, r);
-
-#if !CONFIG_TXK_SEL
-  av1_read_tx_type(cm, xd,
-#if CONFIG_SUPERTX
-                   supertx_enabled,
-#endif
-                   r);
-#endif  // !CONFIG_TXK_SEL
+    read_intra_block_mode_info(cm, mi_row, mi_col, xd, mbmi, r);
 }
 
-static void av1_intra_copy_frame_mvs(AV1_COMMON *const cm, int mi_row,
-                                     int mi_col, int x_mis, int y_mis) {
-#if CONFIG_TMV
+static void intra_copy_frame_mvs(AV1_COMMON *const cm, int mi_row, int mi_col,
+                                 int x_mis, int y_mis) {
   const int frame_mvs_stride = ROUND_POWER_OF_TWO(cm->mi_cols, 1);
-  MV_REF *frame_mvs = cm->cur_frame->mvs +
-                      ((mi_row & 0xfffe) >> 1) * frame_mvs_stride +
-                      ((mi_col & 0xfffe) >> 1);
+  MV_REF *frame_mvs =
+      cm->cur_frame->mvs + (mi_row >> 1) * frame_mvs_stride + (mi_col >> 1);
   x_mis = ROUND_POWER_OF_TWO(x_mis, 1);
   y_mis = ROUND_POWER_OF_TWO(y_mis, 1);
-#else
-  const int frame_mvs_stride = cm->mi_cols;
-  MV_REF *frame_mvs = cm->cur_frame->mvs +
-                      (mi_row & 0xfffe) * frame_mvs_stride + (mi_col & 0xfffe);
-  x_mis = AOMMAX(x_mis, 2);
-  y_mis = AOMMAX(y_mis, 2);
-#endif  // CONFIG_TMV
-  int w, h;
-
-  for (h = 0; h < y_mis; h++) {
-    MV_REF *const frame_mv = frame_mvs + h * frame_mvs_stride;
-    for (w = 0; w < x_mis; w++) {
-      MV_REF *const mv = frame_mv + w;
-      mv->ref_frame[0] = NONE_FRAME;
-      mv->ref_frame[1] = NONE_FRAME;
+
+  for (int h = 0; h < y_mis; h++) {
+    MV_REF *mv = frame_mvs;
+    for (int w = 0; w < x_mis; w++) {
+      mv->ref_frame = NONE_FRAME;
+      mv++;
     }
+    frame_mvs += frame_mvs_stride;
   }
 }
 
-void av1_read_mode_info(AV1Decoder *const pbi, MACROBLOCKD *xd,
-#if CONFIG_SUPERTX
-                        int supertx_enabled,
-#endif  // CONFIG_SUPERTX
-                        int mi_row, int mi_col, aom_reader *r, int x_mis,
-                        int y_mis) {
+void av1_read_mode_info(AV1Decoder *const pbi, MACROBLOCKD *xd, int mi_row,
+                        int mi_col, aom_reader *r, int x_mis, int y_mis) {
   AV1_COMMON *const cm = &pbi->common;
-  MODE_INFO *const mi = xd->mi[0];
-#if CONFIG_INTRABC
-  mi->mbmi.use_intrabc = 0;
-#endif  // CONFIG_INTRABC
+  MB_MODE_INFO *const mi = xd->mi[0];
+  mi->use_intrabc = 0;
 
   if (frame_is_intra_only(cm)) {
     read_intra_frame_mode_info(cm, xd, mi_row, mi_col, r);
-    av1_intra_copy_frame_mvs(cm, mi_row, mi_col, x_mis, y_mis);
+    intra_copy_frame_mvs(cm, mi_row, mi_col, x_mis, y_mis);
   } else {
-    read_inter_frame_mode_info(pbi, xd,
-#if CONFIG_SUPERTX
-                               supertx_enabled,
-#endif  // CONFIG_SUPERTX
-                               mi_row, mi_col, r);
+    read_inter_frame_mode_info(pbi, xd, mi_row, mi_col, r);
     av1_copy_frame_mvs(cm, mi, mi_row, mi_col, x_mis, y_mis);
   }
 }
diff --git a/third_party/aom/av1/decoder/decodemv.h b/third_party/aom/av1/decoder/decodemv.h
index 162cf3254..6243bb168 100644
--- a/third_party/aom/av1/decoder/decodemv.h
+++ b/third_party/aom/av1/decoder/decodemv.h
@@ -21,9 +21,6 @@ extern "C" {
 #endif
 
 void av1_read_mode_info(AV1Decoder *const pbi, MACROBLOCKD *xd,
-#if CONFIG_SUPERTX
-                        int supertx_enabled,
-#endif
 
                         int mi_row, int mi_col, aom_reader *r, int x_mis,
                         int y_mis);
@@ -32,14 +29,7 @@ void av1_read_mode_info(AV1Decoder *const pbi, MACROBLOCKD *xd,
 }  // extern "C"
 #endif
 
-void av1_read_tx_type(const AV1_COMMON *const cm, MACROBLOCKD *xd,
-#if CONFIG_SUPERTX
-                      int supertx_enabled,
-#endif
-#if CONFIG_TXK_SEL
-                      int blk_row, int blk_col, int block, int plane,
-                      TX_SIZE tx_size,
-#endif
-                      aom_reader *r);
+void av1_read_tx_type(const AV1_COMMON *const cm, MACROBLOCKD *xd, int blk_row,
+                      int blk_col, TX_SIZE tx_size, aom_reader *r);
 
 #endif  // AV1_DECODER_DECODEMV_H_
diff --git a/third_party/aom/av1/decoder/decoder.c b/third_party/aom/av1/decoder/decoder.c
index cd82d5b53..2e91d27d3 100644
--- a/third_party/aom/av1/decoder/decoder.c
+++ b/third_party/aom/av1/decoder/decoder.c
@@ -13,9 +13,9 @@
 #include <limits.h>
 #include <stdio.h>
 
-#include "./av1_rtcd.h"
-#include "./aom_dsp_rtcd.h"
-#include "./aom_scale_rtcd.h"
+#include "config/av1_rtcd.h"
+#include "config/aom_dsp_rtcd.h"
+#include "config/aom_scale_rtcd.h"
 
 #include "aom_mem/aom_mem.h"
 #include "aom_ports/system_state.h"
@@ -33,12 +33,8 @@
 
 #include "av1/decoder/decodeframe.h"
 #include "av1/decoder/decoder.h"
-#if CONFIG_NCOBMC_ADAPT_WEIGHT
-#include "av1/common/ncobmc_kernels.h"
-#endif  // CONFIG_NCOBMC_ADAPT_WEIGHT
-#if !CONFIG_PVQ
 #include "av1/decoder/detokenize.h"
-#endif
+#include "av1/decoder/obu.h"
 
 static void initialize_dec(void) {
   static volatile int init_done = 0;
@@ -53,23 +49,24 @@ static void initialize_dec(void) {
   }
 }
 
-static void av1_dec_setup_mi(AV1_COMMON *cm) {
-  cm->mi = cm->mip + cm->mi_stride + 1;
-  cm->mi_grid_visible = cm->mi_grid_base + cm->mi_stride + 1;
+static void dec_setup_mi(AV1_COMMON *cm) {
+  cm->mi = cm->mip;
+  cm->mi_grid_visible = cm->mi_grid_base;
   memset(cm->mi_grid_base, 0,
-         cm->mi_stride * (cm->mi_rows + 1) * sizeof(*cm->mi_grid_base));
+         cm->mi_stride * cm->mi_rows * sizeof(*cm->mi_grid_base));
 }
 
 static int av1_dec_alloc_mi(AV1_COMMON *cm, int mi_size) {
   cm->mip = aom_calloc(mi_size, sizeof(*cm->mip));
   if (!cm->mip) return 1;
   cm->mi_alloc_size = mi_size;
-  cm->mi_grid_base = (MODE_INFO **)aom_calloc(mi_size, sizeof(MODE_INFO *));
+  cm->mi_grid_base =
+      (MB_MODE_INFO **)aom_calloc(mi_size, sizeof(MB_MODE_INFO *));
   if (!cm->mi_grid_base) return 1;
   return 0;
 }
 
-static void av1_dec_free_mi(AV1_COMMON *cm) {
+static void dec_free_mi(AV1_COMMON *cm) {
   aom_free(cm->mip);
   cm->mip = NULL;
   aom_free(cm->mi_grid_base);
@@ -108,28 +105,20 @@ AV1Decoder *av1_decoder_create(BufferPool *const pool) {
   memset(&cm->next_ref_frame_map, -1, sizeof(cm->next_ref_frame_map));
 
   cm->current_video_frame = 0;
-  pbi->ready_for_new_data = 1;
+  pbi->decoding_first_frame = 1;
   pbi->common.buffer_pool = pool;
 
   cm->bit_depth = AOM_BITS_8;
   cm->dequant_bit_depth = AOM_BITS_8;
 
   cm->alloc_mi = av1_dec_alloc_mi;
-  cm->free_mi = av1_dec_free_mi;
-  cm->setup_mi = av1_dec_setup_mi;
+  cm->free_mi = dec_free_mi;
+  cm->setup_mi = dec_setup_mi;
 
   av1_loop_filter_init(cm);
 
-#if CONFIG_NCOBMC_ADAPT_WEIGHT
-  get_default_ncobmc_kernels(cm);
-#endif  // CONFIG_NCOBMC_ADAPT_WEIGHT
-
-#if CONFIG_AOM_QM
-  aom_qm_init(cm);
-#endif
-#if CONFIG_LOOP_RESTORATION
+  av1_qm_init(cm);
   av1_loop_restoration_precal();
-#endif  // CONFIG_LOOP_RESTORATION
 #if CONFIG_ACCOUNTING
   pbi->acct_enabled = 1;
   aom_accounting_init(&pbi->accounting);
@@ -142,33 +131,83 @@ AV1Decoder *av1_decoder_create(BufferPool *const pool) {
   return pbi;
 }
 
+void av1_dealloc_dec_jobs(struct AV1DecTileMTData *tile_mt_info) {
+  if (tile_mt_info != NULL) {
+#if CONFIG_MULTITHREAD
+    if (tile_mt_info->job_mutex != NULL) {
+      pthread_mutex_destroy(tile_mt_info->job_mutex);
+      aom_free(tile_mt_info->job_mutex);
+    }
+#endif
+    aom_free(tile_mt_info->job_queue);
+    // clear the structure as the source of this call may be a resize in which
+    // case this call will be followed by an _alloc() which may fail.
+    av1_zero(*tile_mt_info);
+  }
+}
+
 void av1_decoder_remove(AV1Decoder *pbi) {
   int i;
 
   if (!pbi) return;
 
+  // Free the tile list output buffer.
+  if (pbi->tile_list_output != NULL) aom_free(pbi->tile_list_output);
+  pbi->tile_list_output = NULL;
+
   aom_get_worker_interface()->end(&pbi->lf_worker);
   aom_free(pbi->lf_worker.data1);
-  aom_free(pbi->tile_data);
-  for (i = 0; i < pbi->num_tile_workers; ++i) {
+
+  if (pbi->thread_data) {
+    for (int worker_idx = 0; worker_idx < pbi->max_threads - 1; worker_idx++) {
+      DecWorkerData *const thread_data = pbi->thread_data + worker_idx;
+      const int use_highbd = pbi->common.use_highbitdepth ? 1 : 0;
+      av1_free_mc_tmp_buf(thread_data->td, use_highbd);
+      aom_free(thread_data->td);
+    }
+    aom_free(pbi->thread_data);
+  }
+
+  for (i = 0; i < pbi->num_workers; ++i) {
     AVxWorker *const worker = &pbi->tile_workers[i];
     aom_get_worker_interface()->end(worker);
   }
-  aom_free(pbi->tile_worker_data);
-  aom_free(pbi->tile_worker_info);
+  aom_free(pbi->tile_data);
   aom_free(pbi->tile_workers);
 
-  if (pbi->num_tile_workers > 0) {
+  if (pbi->num_workers > 0) {
     av1_loop_filter_dealloc(&pbi->lf_row_sync);
+    av1_loop_restoration_dealloc(&pbi->lr_row_sync, pbi->num_workers);
+    av1_dealloc_dec_jobs(&pbi->tile_mt_info);
   }
 
 #if CONFIG_ACCOUNTING
   aom_accounting_clear(&pbi->accounting);
 #endif
+  const int use_highbd = pbi->common.use_highbitdepth ? 1 : 0;
+  av1_free_mc_tmp_buf(&pbi->td, use_highbd);
 
   aom_free(pbi);
 }
 
+void av1_visit_palette(AV1Decoder *const pbi, MACROBLOCKD *const xd, int mi_row,
+                       int mi_col, aom_reader *r, BLOCK_SIZE bsize,
+                       palette_visitor_fn_t visit) {
+  if (!is_inter_block(xd->mi[0])) {
+    for (int plane = 0; plane < AOMMIN(2, av1_num_planes(&pbi->common));
+         ++plane) {
+      const struct macroblockd_plane *const pd = &xd->plane[plane];
+      if (is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
+                              pd->subsampling_y)) {
+        if (xd->mi[0]->palette_mode_info.palette_size[plane])
+          visit(xd, plane, r);
+      } else {
+        assert(xd->mi[0]->palette_mode_info.palette_size[plane] == 0);
+      }
+    }
+  }
+}
+
 static int equal_dimensions(const YV12_BUFFER_CONFIG *a,
                             const YV12_BUFFER_CONFIG *b) {
   return a->y_height == b->y_height && a->y_width == b->y_width &&
@@ -178,6 +217,7 @@ static int equal_dimensions(const YV12_BUFFER_CONFIG *a,
 aom_codec_err_t av1_copy_reference_dec(AV1Decoder *pbi, int idx,
                                        YV12_BUFFER_CONFIG *sd) {
   AV1_COMMON *cm = &pbi->common;
+  const int num_planes = av1_num_planes(cm);
 
   const YV12_BUFFER_CONFIG *const cfg = get_ref_frame(cm, idx);
   if (cfg == NULL) {
@@ -188,13 +228,25 @@ aom_codec_err_t av1_copy_reference_dec(AV1Decoder *pbi, int idx,
     aom_internal_error(&cm->error, AOM_CODEC_ERROR,
                        "Incorrect buffer dimensions");
   else
-    aom_yv12_copy_frame(cfg, sd);
+    aom_yv12_copy_frame(cfg, sd, num_planes);
 
   return cm->error.error_code;
 }
 
+static int equal_dimensions_and_border(const YV12_BUFFER_CONFIG *a,
+                                       const YV12_BUFFER_CONFIG *b) {
+  return a->y_height == b->y_height && a->y_width == b->y_width &&
+         a->uv_height == b->uv_height && a->uv_width == b->uv_width &&
+         a->y_stride == b->y_stride && a->uv_stride == b->uv_stride &&
+         a->border == b->border &&
+         (a->flags & YV12_FLAG_HIGHBITDEPTH) ==
+             (b->flags & YV12_FLAG_HIGHBITDEPTH);
+}
+
 aom_codec_err_t av1_set_reference_dec(AV1_COMMON *cm, int idx,
+                                      int use_external_ref,
                                       YV12_BUFFER_CONFIG *sd) {
+  const int num_planes = av1_num_planes(cm);
   YV12_BUFFER_CONFIG *ref_buf = NULL;
 
   // Get the destination reference buffer.
@@ -205,60 +257,132 @@ aom_codec_err_t av1_set_reference_dec(AV1_COMMON *cm, int idx,
     return AOM_CODEC_ERROR;
   }
 
-  if (!equal_dimensions(ref_buf, sd)) {
-    aom_internal_error(&cm->error, AOM_CODEC_ERROR,
-                       "Incorrect buffer dimensions");
+  if (!use_external_ref) {
+    if (!equal_dimensions(ref_buf, sd)) {
+      aom_internal_error(&cm->error, AOM_CODEC_ERROR,
+                         "Incorrect buffer dimensions");
+    } else {
+      // Overwrite the reference frame buffer.
+      aom_yv12_copy_frame(sd, ref_buf, num_planes);
+    }
   } else {
-    // Overwrite the reference frame buffer.
-    aom_yv12_copy_frame(sd, ref_buf);
+    if (!equal_dimensions_and_border(ref_buf, sd)) {
+      aom_internal_error(&cm->error, AOM_CODEC_ERROR,
+                         "Incorrect buffer dimensions");
+    } else {
+      // Overwrite the reference frame buffer pointers.
+      // Once we no longer need the external reference buffer, these pointers
+      // are restored.
+      ref_buf->store_buf_adr[0] = ref_buf->y_buffer;
+      ref_buf->store_buf_adr[1] = ref_buf->u_buffer;
+      ref_buf->store_buf_adr[2] = ref_buf->v_buffer;
+      ref_buf->y_buffer = sd->y_buffer;
+      ref_buf->u_buffer = sd->u_buffer;
+      ref_buf->v_buffer = sd->v_buffer;
+      ref_buf->use_external_refernce_buffers = 1;
+    }
   }
 
   return cm->error.error_code;
 }
 
-/* If any buffer updating is signaled it should be done here. */
-static void swap_frame_buffers(AV1Decoder *pbi) {
+aom_codec_err_t av1_copy_new_frame_dec(AV1_COMMON *cm,
+                                       YV12_BUFFER_CONFIG *new_frame,
+                                       YV12_BUFFER_CONFIG *sd) {
+  const int num_planes = av1_num_planes(cm);
+
+  if (!equal_dimensions_and_border(new_frame, sd))
+    aom_internal_error(&cm->error, AOM_CODEC_ERROR,
+                       "Incorrect buffer dimensions");
+  else
+    aom_yv12_copy_frame(new_frame, sd, num_planes);
+
+  return cm->error.error_code;
+}
+
+/* If any buffer updating is signaled it should be done here.
+   Consumes a reference to cm->new_fb_idx.
+*/
+static void swap_frame_buffers(AV1Decoder *pbi, int frame_decoded) {
   int ref_index = 0, mask;
   AV1_COMMON *const cm = &pbi->common;
   BufferPool *const pool = cm->buffer_pool;
   RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs;
 
-  lock_buffer_pool(pool);
-  for (mask = pbi->refresh_frame_flags; mask; mask >>= 1) {
-    const int old_idx = cm->ref_frame_map[ref_index];
-    // Current thread releases the holding of reference frame.
-    decrease_ref_count(old_idx, frame_bufs, pool);
-
-    // Release the reference frame holding in the reference map for the decoding
-    // of the next frame.
-    if (mask & 1) decrease_ref_count(old_idx, frame_bufs, pool);
-    cm->ref_frame_map[ref_index] = cm->next_ref_frame_map[ref_index];
-    ++ref_index;
-  }
+  if (frame_decoded) {
+    lock_buffer_pool(pool);
 
-  // Current thread releases the holding of reference frame.
-  for (; ref_index < REF_FRAMES && !cm->show_existing_frame; ++ref_index) {
-    const int old_idx = cm->ref_frame_map[ref_index];
-    decrease_ref_count(old_idx, frame_bufs, pool);
-    cm->ref_frame_map[ref_index] = cm->next_ref_frame_map[ref_index];
-  }
+    // In ext-tile decoding, the camera frame header is only decoded once. So,
+    // we don't release the references here.
+    if (!pbi->camera_frame_header_ready) {
+      for (mask = pbi->refresh_frame_flags; mask; mask >>= 1) {
+        const int old_idx = cm->ref_frame_map[ref_index];
+        // Current thread releases the holding of reference frame.
+        decrease_ref_count(old_idx, frame_bufs, pool);
 
-  unlock_buffer_pool(pool);
-  pbi->hold_ref_buf = 0;
-  cm->frame_to_show = get_frame_new_buffer(cm);
+        // Release the reference frame holding in the reference map for the
+        // decoding of the next frame.
+        if (mask & 1) decrease_ref_count(old_idx, frame_bufs, pool);
+        cm->ref_frame_map[ref_index] = cm->next_ref_frame_map[ref_index];
+        ++ref_index;
+      }
 
-  // TODO(zoeliu): To fix the ref frame buffer update for the scenario of
-  //               cm->frame_parellel_decode == 1
-  if (!cm->frame_parallel_decode || !cm->show_frame) {
+      // Current thread releases the holding of reference frame.
+      const int check_on_show_existing_frame =
+          !cm->show_existing_frame || cm->reset_decoder_state;
+      for (; ref_index < REF_FRAMES && check_on_show_existing_frame;
+           ++ref_index) {
+        const int old_idx = cm->ref_frame_map[ref_index];
+        decrease_ref_count(old_idx, frame_bufs, pool);
+        cm->ref_frame_map[ref_index] = cm->next_ref_frame_map[ref_index];
+      }
+    }
+
+    YV12_BUFFER_CONFIG *cur_frame = get_frame_new_buffer(cm);
+
+    if (cm->show_existing_frame || cm->show_frame) {
+      if (pbi->output_all_layers) {
+        // Append this frame to the output queue
+        if (pbi->num_output_frames >= MAX_NUM_SPATIAL_LAYERS) {
+          // We can't store the new frame anywhere, so drop it and return an
+          // error
+          decrease_ref_count(cm->new_fb_idx, frame_bufs, pool);
+          cm->error.error_code = AOM_CODEC_UNSUP_BITSTREAM;
+        } else {
+          pbi->output_frames[pbi->num_output_frames] = cur_frame;
+          pbi->output_frame_index[pbi->num_output_frames] = cm->new_fb_idx;
+          pbi->num_output_frames++;
+        }
+      } else {
+        // Replace any existing output frame
+        assert(pbi->num_output_frames == 0 || pbi->num_output_frames == 1);
+        if (pbi->num_output_frames > 0) {
+          decrease_ref_count((int)pbi->output_frame_index[0], frame_bufs, pool);
+        }
+        pbi->output_frames[0] = cur_frame;
+        pbi->output_frame_index[0] = cm->new_fb_idx;
+        pbi->num_output_frames = 1;
+      }
+    } else {
+      decrease_ref_count(cm->new_fb_idx, frame_bufs, pool);
+    }
+
+    unlock_buffer_pool(pool);
+  } else {
+    // Nothing was decoded, so just drop this frame buffer
     lock_buffer_pool(pool);
-    --frame_bufs[cm->new_fb_idx].ref_count;
+    decrease_ref_count(cm->new_fb_idx, frame_bufs, pool);
     unlock_buffer_pool(pool);
   }
 
-  // Invalidate these references until the next frame starts.
-  for (ref_index = 0; ref_index < INTER_REFS_PER_FRAME; ref_index++) {
-    cm->frame_refs[ref_index].idx = INVALID_IDX;
-    cm->frame_refs[ref_index].buf = NULL;
+  if (!pbi->camera_frame_header_ready) {
+    pbi->hold_ref_buf = 0;
+
+    // Invalidate these references until the next frame starts.
+    for (ref_index = 0; ref_index < INTER_REFS_PER_FRAME; ref_index++) {
+      cm->frame_refs[ref_index].idx = INVALID_IDX;
+      cm->frame_refs[ref_index].buf = NULL;
+    }
   }
 }
 
@@ -268,7 +392,6 @@ int av1_receive_compressed_data(AV1Decoder *pbi, size_t size,
   BufferPool *volatile const pool = cm->buffer_pool;
   RefCntBuffer *volatile const frame_bufs = cm->buffer_pool->frame_bufs;
   const uint8_t *source = *psource;
-  int retcode = 0;
   cm->error.error_code = AOM_CODEC_OK;
 
   if (size == 0) {
@@ -286,18 +409,9 @@ int av1_receive_compressed_data(AV1Decoder *pbi, size_t size,
     }
   }
 
-  pbi->ready_for_new_data = 0;
-
   // Find a free buffer for the new frame, releasing the reference previously
   // held.
 
-  // Check if the previous frame was a frame without any references to it.
-  // Release frame buffer if not decoding in frame parallel mode.
-  if (!cm->frame_parallel_decode && cm->new_fb_idx >= 0 &&
-      frame_bufs[cm->new_fb_idx].ref_count == 0)
-    pool->release_fb_cb(pool->cb_priv,
-                        &frame_bufs[cm->new_fb_idx].raw_frame_buffer);
-
   // Find a free frame buffer. Return error if can not find any.
   cm->new_fb_idx = get_free_fb(cm);
   if (cm->new_fb_idx == INVALID_IDX) return AOM_CODEC_MEM_ERROR;
@@ -305,31 +419,20 @@ int av1_receive_compressed_data(AV1Decoder *pbi, size_t size,
   // Assign a MV array to the frame buffer.
   cm->cur_frame = &pool->frame_bufs[cm->new_fb_idx];
 
-  pbi->hold_ref_buf = 0;
-  if (cm->frame_parallel_decode) {
-    AVxWorker *const worker = pbi->frame_worker_owner;
-    av1_frameworker_lock_stats(worker);
-    frame_bufs[cm->new_fb_idx].frame_worker_owner = worker;
-    // Reset decoding progress.
-    pbi->cur_buf = &frame_bufs[cm->new_fb_idx];
-    pbi->cur_buf->row = -1;
-    pbi->cur_buf->col = -1;
-    av1_frameworker_unlock_stats(worker);
-  } else {
-    pbi->cur_buf = &frame_bufs[cm->new_fb_idx];
-  }
+  if (!pbi->camera_frame_header_ready) pbi->hold_ref_buf = 0;
+
+  pbi->cur_buf = &frame_bufs[cm->new_fb_idx];
 
   if (setjmp(cm->error.jmp)) {
     const AVxWorkerInterface *const winterface = aom_get_worker_interface();
     int i;
 
     cm->error.setjmp = 0;
-    pbi->ready_for_new_data = 1;
 
     // Synchronize all threads immediately as a subsequent decode call may
     // cause a resize invalidating some allocations.
     winterface->sync(&pbi->lf_worker);
-    for (i = 0; i < pbi->num_tile_workers; ++i) {
+    for (i = 0; i < pbi->num_workers; ++i) {
       winterface->sync(&pbi->tile_workers[i]);
     }
 
@@ -349,7 +452,10 @@ int av1_receive_compressed_data(AV1Decoder *pbi, size_t size,
       }
 
       // Current thread releases the holding of reference frame.
-      for (; ref_index < REF_FRAMES && !cm->show_existing_frame; ++ref_index) {
+      const int check_on_show_existing_frame =
+          !cm->show_existing_frame || cm->reset_decoder_state;
+      for (; ref_index < REF_FRAMES && check_on_show_existing_frame;
+           ++ref_index) {
         const int old_idx = cm->ref_frame_map[ref_index];
         decrease_ref_count(old_idx, frame_bufs, pool);
       }
@@ -365,160 +471,72 @@ int av1_receive_compressed_data(AV1Decoder *pbi, size_t size,
 
   cm->error.setjmp = 1;
 
-#if !CONFIG_OBU
-  av1_decode_frame_headers_and_setup(pbi, source, source + size, psource);
-  if (!cm->show_existing_frame) {
-    av1_decode_tg_tiles_and_wrapup(pbi, source, source + size, psource, 0,
-                                   cm->tile_rows * cm->tile_cols - 1, 1);
-  }
-#else
-  av1_decode_frame_from_obus(pbi, source, source + size, psource);
+  int frame_decoded =
+      aom_decode_frame_from_obus(pbi, source, source + size, psource);
+
+  if (cm->error.error_code != AOM_CODEC_OK) return 1;
+
+#if TXCOEFF_TIMER
+  cm->cum_txcoeff_timer += cm->txcoeff_timer;
+  fprintf(stderr,
+          "txb coeff block number: %d, frame time: %ld, cum time %ld in us\n",
+          cm->txb_count, cm->txcoeff_timer, cm->cum_txcoeff_timer);
+  cm->txcoeff_timer = 0;
+  cm->txb_count = 0;
 #endif
 
-  swap_frame_buffers(pbi);
+  // Note: At this point, this function holds a reference to cm->new_fb_idx
+  // in the buffer pool. This reference is consumed by swap_frame_buffers().
+  swap_frame_buffers(pbi, frame_decoded);
+
+  if (frame_decoded) {
+    pbi->decoding_first_frame = 0;
+  }
 
-#if CONFIG_EXT_TILE
-  // For now, we only extend the frame borders when the whole frame is decoded.
-  // Later, if needed, extend the border for the decoded tile on the frame
-  // border.
-  if (pbi->dec_tile_row == -1 && pbi->dec_tile_col == -1)
-#endif  // CONFIG_EXT_TILE
-    // TODO(debargha): Fix encoder side mv range, so that we can use the
-    // inner border extension. As of now use the larger extension.
-    // aom_extend_frame_inner_borders(cm->frame_to_show);
-    aom_extend_frame_borders(cm->frame_to_show);
+  if (cm->error.error_code != AOM_CODEC_OK) return 1;
 
   aom_clear_system_state();
 
   if (!cm->show_existing_frame) {
     cm->last_show_frame = cm->show_frame;
 
-#if CONFIG_EXT_REFS
-    // NOTE: It is not supposed to ref to any frame not used as reference
-    if (cm->is_reference_frame)
-#endif  // CONFIG_EXT_REFS
-      cm->prev_frame = cm->cur_frame;
-
-    if (cm->seg.enabled && !cm->frame_parallel_decode)
-      av1_swap_current_and_last_seg_map(cm);
-  }
-
-  // Update progress in frame parallel decode.
-  if (cm->frame_parallel_decode) {
-    // Need to lock the mutex here as another thread may
-    // be accessing this buffer.
-    AVxWorker *const worker = pbi->frame_worker_owner;
-    FrameWorkerData *const frame_worker_data = worker->data1;
-    av1_frameworker_lock_stats(worker);
-
-    if (cm->show_frame) {
-      cm->current_video_frame++;
-    }
-    frame_worker_data->frame_decoded = 1;
-    frame_worker_data->frame_context_ready = 1;
-    av1_frameworker_signal_stats(worker);
-    av1_frameworker_unlock_stats(worker);
-  } else {
-    cm->last_width = cm->width;
-    cm->last_height = cm->height;
-    cm->last_tile_cols = cm->tile_cols;
-    cm->last_tile_rows = cm->tile_rows;
-    if (cm->show_frame) {
-      cm->current_video_frame++;
+    if (cm->seg.enabled) {
+      if (cm->prev_frame && (cm->mi_rows == cm->prev_frame->mi_rows) &&
+          (cm->mi_cols == cm->prev_frame->mi_cols)) {
+        cm->last_frame_seg_map = cm->prev_frame->seg_map;
+      } else {
+        cm->last_frame_seg_map = NULL;
+      }
     }
   }
 
+  // Update progress in frame parallel decode.
+  cm->last_width = cm->width;
+  cm->last_height = cm->height;
+  cm->last_tile_cols = cm->tile_cols;
+  cm->last_tile_rows = cm->tile_rows;
   cm->error.setjmp = 0;
-  return retcode;
-}
-
-int av1_get_raw_frame(AV1Decoder *pbi, YV12_BUFFER_CONFIG *sd) {
-  AV1_COMMON *const cm = &pbi->common;
-  int ret = -1;
-  if (pbi->ready_for_new_data == 1) return ret;
 
-  pbi->ready_for_new_data = 1;
+  return 0;
+}
 
-  /* no raw frame to show!!! */
-  if (!cm->show_frame) return ret;
+// Get the frame at a particular index in the output queue
+int av1_get_raw_frame(AV1Decoder *pbi, size_t index, YV12_BUFFER_CONFIG **sd,
+                      aom_film_grain_t **grain_params) {
+  RefCntBuffer *const frame_bufs = pbi->common.buffer_pool->frame_bufs;
 
-  *sd = *cm->frame_to_show;
-  ret = 0;
+  if (index >= pbi->num_output_frames) return -1;
+  *sd = pbi->output_frames[index];
+  *grain_params = &frame_bufs[pbi->output_frame_index[index]].film_grain_params;
   aom_clear_system_state();
-  return ret;
+  return 0;
 }
 
+// Get the highest-spatial-layer output
+// TODO(david.barker): What should this do?
 int av1_get_frame_to_show(AV1Decoder *pbi, YV12_BUFFER_CONFIG *frame) {
-  AV1_COMMON *const cm = &pbi->common;
-
-  if (!cm->show_frame || !cm->frame_to_show) return -1;
+  if (pbi->num_output_frames == 0) return -1;
 
-  *frame = *cm->frame_to_show;
+  *frame = *pbi->output_frames[pbi->num_output_frames - 1];
   return 0;
 }
-
-aom_codec_err_t av1_parse_superframe_index(const uint8_t *data, size_t data_sz,
-                                           uint32_t sizes[8], int *count,
-                                           int *index_size,
-                                           aom_decrypt_cb decrypt_cb,
-                                           void *decrypt_state) {
-  // A chunk ending with a byte matching 0xc0 is an invalid chunk unless
-  // it is a super frame index. If the last byte of real video compression
-  // data is 0xc0 the encoder must add a 0 byte. If we have the marker but
-  // not the associated matching marker byte at the front of the index we have
-  // an invalid bitstream and need to return an error.
-
-  uint8_t marker;
-  size_t frame_sz_sum = 0;
-
-  assert(data_sz);
-  marker = read_marker(decrypt_cb, decrypt_state, data);
-  *count = 0;
-
-  if ((marker & 0xe0) == 0xc0) {
-    const uint32_t frames = (marker & 0x7) + 1;
-    const uint32_t mag = ((marker >> 3) & 0x3) + 1;
-    const size_t index_sz = 2 + mag * (frames - 1);
-    *index_size = (int)index_sz;
-
-    // This chunk is marked as having a superframe index but doesn't have
-    // enough data for it, thus it's an invalid superframe index.
-    if (data_sz < index_sz) return AOM_CODEC_CORRUPT_FRAME;
-
-    {
-      const uint8_t marker2 =
-          read_marker(decrypt_cb, decrypt_state, data + index_sz - 1);
-
-      // This chunk is marked as having a superframe index but doesn't have
-      // the matching marker byte at the front of the index therefore it's an
-      // invalid chunk.
-      if (marker != marker2) return AOM_CODEC_CORRUPT_FRAME;
-    }
-
-    {
-      // Found a valid superframe index.
-      uint32_t i, j;
-      const uint8_t *x = &data[1];
-
-      // Frames has a maximum of 8 and mag has a maximum of 4.
-      uint8_t clear_buffer[28];
-      assert(sizeof(clear_buffer) >= (frames - 1) * mag);
-      if (decrypt_cb) {
-        decrypt_cb(decrypt_state, x, clear_buffer, (frames - 1) * mag);
-        x = clear_buffer;
-      }
-
-      for (i = 0; i < frames - 1; ++i) {
-        uint32_t this_sz = 0;
-
-        for (j = 0; j < mag; ++j) this_sz |= (*x++) << (j * 8);
-        this_sz += 1;
-        sizes[i] = this_sz;
-        frame_sz_sum += this_sz;
-      }
-      sizes[i] = (uint32_t)(data_sz - index_sz - frame_sz_sum);
-      *count = frames;
-    }
-  }
-  return AOM_CODEC_OK;
-}
diff --git a/third_party/aom/av1/decoder/decoder.h b/third_party/aom/av1/decoder/decoder.h
index 20129b669..42fcc1256 100644
--- a/third_party/aom/av1/decoder/decoder.h
+++ b/third_party/aom/av1/decoder/decoder.h
@@ -12,7 +12,7 @@
 #ifndef AV1_DECODER_DECODER_H_
 #define AV1_DECODER_DECODER_H_
 
-#include "./aom_config.h"
+#include "config/aom_config.h"
 
 #include "aom/aom_codec.h"
 #include "aom_dsp/bitreader.h"
@@ -29,73 +29,61 @@
 #include "av1/decoder/inspection.h"
 #endif
 
-#if CONFIG_PVQ
-#include "aom_dsp/entdec.h"
-#include "av1/decoder/decint.h"
-#include "av1/encoder/encodemb.h"
-#endif
-
 #ifdef __cplusplus
 extern "C" {
 #endif
 
-// TODO(hkuang): combine this with TileWorkerData.
-typedef struct TileData {
-  AV1_COMMON *cm;
-  aom_reader bit_reader;
-  DECLARE_ALIGNED(16, MACROBLOCKD, xd);
+typedef struct ThreadData {
+  aom_reader *bit_reader;
+  DECLARE_ALIGNED(32, MACROBLOCKD, xd);
   /* dqcoeff are shared by all the planes. So planes must be decoded serially */
-  DECLARE_ALIGNED(16, tran_low_t, dqcoeff[MAX_TX_SQUARE]);
-#if CONFIG_PVQ
-  /* forward transformed predicted image, a reference for PVQ */
-  DECLARE_ALIGNED(16, tran_low_t, pvq_ref_coeff[OD_TXSIZE_MAX * OD_TXSIZE_MAX]);
-#endif
-#if CONFIG_CFL
-  CFL_CTX cfl;
-#endif
-  DECLARE_ALIGNED(16, FRAME_CONTEXT, tctx);
-  DECLARE_ALIGNED(16, uint8_t, color_index_map[2][MAX_SB_SQUARE]);
-#if CONFIG_MRC_TX
-  DECLARE_ALIGNED(16, uint8_t, mrc_mask[MAX_SB_SQUARE]);
-#endif  // CONFIG_MRC_TX
-} TileData;
-
-typedef struct TileWorkerData {
-  struct AV1Decoder *pbi;
+  DECLARE_ALIGNED(32, tran_low_t, dqcoeff[MAX_TX_SQUARE]);
+  CB_BUFFER cb_buffer_base;
+  uint8_t *mc_buf[2];
+  int32_t mc_buf_size;
+} ThreadData;
+
+typedef struct TileDataDec {
+  TileInfo tile_info;
   aom_reader bit_reader;
-  FRAME_COUNTS counts;
-  DECLARE_ALIGNED(16, MACROBLOCKD, xd);
-  /* dqcoeff are shared by all the planes. So planes must be decoded serially */
-  DECLARE_ALIGNED(16, tran_low_t, dqcoeff[MAX_TX_SQUARE]);
-#if CONFIG_PVQ
-  /* forward transformed predicted image, a reference for PVQ */
-  DECLARE_ALIGNED(16, tran_low_t, pvq_ref_coeff[OD_TXSIZE_MAX * OD_TXSIZE_MAX]);
-#endif
-#if CONFIG_CFL
-  CFL_CTX cfl;
-#endif
-  FRAME_CONTEXT tctx;
-  DECLARE_ALIGNED(16, uint8_t, color_index_map[2][MAX_SB_SQUARE]);
-#if CONFIG_MRC_TX
-  DECLARE_ALIGNED(16, uint8_t, mrc_mask[MAX_SB_SQUARE]);
-#endif  // CONFIG_MRC_TX
-  struct aom_internal_error_info error_info;
-} TileWorkerData;
+  DECLARE_ALIGNED(16, FRAME_CONTEXT, tctx);
+} TileDataDec;
 
 typedef struct TileBufferDec {
   const uint8_t *data;
   size_t size;
-  const uint8_t *raw_data_end;  // The end of the raw tile buffer in the
-                                // bit stream.
-  int col;                      // only used with multi-threaded decoding
 } TileBufferDec;
 
-typedef struct AV1Decoder {
-  DECLARE_ALIGNED(16, MACROBLOCKD, mb);
+typedef struct DataBuffer {
+  const uint8_t *data;
+  size_t size;
+} DataBuffer;
+
+typedef struct EXTERNAL_REFERENCES {
+  YV12_BUFFER_CONFIG refs[MAX_EXTERNAL_REFERENCES];
+  int num;
+} EXTERNAL_REFERENCES;
+
+typedef struct TileJobsDec {
+  TileBufferDec *tile_buffer;
+  TileDataDec *tile_data;
+} TileJobsDec;
 
-  DECLARE_ALIGNED(16, AV1_COMMON, common);
+typedef struct AV1DecTileMTData {
+#if CONFIG_MULTITHREAD
+  pthread_mutex_t *job_mutex;
+#endif
+  TileJobsDec *job_queue;
+  int jobs_enqueued;
+  int jobs_dequeued;
+  int alloc_tile_rows;
+  int alloc_tile_cols;
+} AV1DecTileMT;
+
+typedef struct AV1Decoder {
+  DECLARE_ALIGNED(32, MACROBLOCKD, mb);
 
-  int ready_for_new_data;
+  DECLARE_ALIGNED(32, AV1_COMMON, common);
 
   int refresh_frame_flags;
 
@@ -105,20 +93,38 @@ typedef struct AV1Decoder {
 
   AVxWorker *frame_worker_owner;  // frame_worker that owns this pbi.
   AVxWorker lf_worker;
+  AV1LfSync lf_row_sync;
+  AV1LrSync lr_row_sync;
+  AV1LrStruct lr_ctxt;
   AVxWorker *tile_workers;
-  TileWorkerData *tile_worker_data;
-  TileInfo *tile_worker_info;
-  int num_tile_workers;
-
-  TileData *tile_data;
+  int num_workers;
+  DecWorkerData *thread_data;
+  ThreadData td;
+  TileDataDec *tile_data;
   int allocated_tiles;
 
   TileBufferDec tile_buffers[MAX_TILE_ROWS][MAX_TILE_COLS];
-
-  AV1LfSync lf_row_sync;
-
-  aom_decrypt_cb decrypt_cb;
-  void *decrypt_state;
+  AV1DecTileMT tile_mt_info;
+
+  // Each time the decoder is called, we expect to receive a full temporal unit.
+  // This can contain up to one shown frame per spatial layer in the current
+  // operating point (note that some layers may be entirely omitted).
+  // If the 'output_all_layers' option is true, we save all of these shown
+  // frames so that they can be returned to the application. If the
+  // 'output_all_layers' option is false, then we only output one image per
+  // temporal unit.
+  //
+  // Note: The saved buffers are released at the start of the next time the
+  // application calls aom_codec_decode().
+  int output_all_layers;
+  YV12_BUFFER_CONFIG *output_frames[MAX_NUM_SPATIAL_LAYERS];
+  size_t output_frame_index[MAX_NUM_SPATIAL_LAYERS];  // Buffer pool indices
+  size_t num_output_frames;  // How many frames are queued up so far?
+
+  // In order to properly support random-access decoding, we need
+  // to behave slightly differently for the very first frame we decode.
+  // So we track whether this is the first frame or not.
+  int decoding_first_frame;
 
   int allow_lowbitdepth;
   int max_threads;
@@ -127,29 +133,47 @@ typedef struct AV1Decoder {
   int hold_ref_buf;  // hold the reference buffer.
 
   int tile_size_bytes;
-#if CONFIG_EXT_TILE
   int tile_col_size_bytes;
   int dec_tile_row, dec_tile_col;  // always -1 for non-VR tile encoding
-#endif                             // CONFIG_EXT_TILE
 #if CONFIG_ACCOUNTING
   int acct_enabled;
   Accounting accounting;
 #endif
-  size_t uncomp_hdr_size;       // Size of the uncompressed header
-  size_t first_partition_size;  // Size of the compressed header
-  int tg_size;                  // Number of tiles in the current tilegroup
-  int tg_start;                 // First tile in the current tilegroup
+  size_t uncomp_hdr_size;  // Size of the uncompressed header
+  int tg_size;             // Number of tiles in the current tilegroup
+  int tg_start;            // First tile in the current tilegroup
   int tg_size_bit_offset;
+  int sequence_header_ready;
 #if CONFIG_INSPECTION
   aom_inspect_cb inspect_cb;
   void *inspect_ctx;
 #endif
+  int operating_point;
+  int current_operating_point;
+  int seen_frame_header;
+
+  // State if the camera frame header is already decoded while
+  // large_scale_tile = 1.
+  int camera_frame_header_ready;
+  size_t frame_header_size;
+  DataBuffer obu_size_hdr;
+  int output_frame_width_in_tiles_minus_1;
+  int output_frame_height_in_tiles_minus_1;
+  int tile_count_minus_1;
+  uint32_t coded_tile_data_size;
+  unsigned int ext_tile_debug;  // for ext-tile software debug & testing
+  EXTERNAL_REFERENCES ext_refs;
+  size_t tile_list_size;
+  uint8_t *tile_list_output;
+  size_t buffer_sz;
 } AV1Decoder;
 
 int av1_receive_compressed_data(struct AV1Decoder *pbi, size_t size,
                                 const uint8_t **dest);
 
-int av1_get_raw_frame(struct AV1Decoder *pbi, YV12_BUFFER_CONFIG *sd);
+// Get the frame at a particular index in the output queue
+int av1_get_raw_frame(AV1Decoder *pbi, size_t index, YV12_BUFFER_CONFIG **sd,
+                      aom_film_grain_t **grain_params);
 
 int av1_get_frame_to_show(struct AV1Decoder *pbi, YV12_BUFFER_CONFIG *frame);
 
@@ -157,29 +181,16 @@ aom_codec_err_t av1_copy_reference_dec(struct AV1Decoder *pbi, int idx,
                                        YV12_BUFFER_CONFIG *sd);
 
 aom_codec_err_t av1_set_reference_dec(AV1_COMMON *cm, int idx,
+                                      int use_external_ref,
                                       YV12_BUFFER_CONFIG *sd);
-
-static INLINE uint8_t read_marker(aom_decrypt_cb decrypt_cb,
-                                  void *decrypt_state, const uint8_t *data) {
-  if (decrypt_cb) {
-    uint8_t marker;
-    decrypt_cb(decrypt_state, data, &marker, 1);
-    return marker;
-  }
-  return *data;
-}
-
-// This function is exposed for use in tests, as well as the inlined function
-// "read_marker".
-aom_codec_err_t av1_parse_superframe_index(const uint8_t *data, size_t data_sz,
-                                           uint32_t sizes[8], int *count,
-                                           int *index_size,
-                                           aom_decrypt_cb decrypt_cb,
-                                           void *decrypt_state);
+aom_codec_err_t av1_copy_new_frame_dec(AV1_COMMON *cm,
+                                       YV12_BUFFER_CONFIG *new_frame,
+                                       YV12_BUFFER_CONFIG *sd);
 
 struct AV1Decoder *av1_decoder_create(BufferPool *const pool);
 
 void av1_decoder_remove(struct AV1Decoder *pbi);
+void av1_dealloc_dec_jobs(struct AV1DecTileMTData *tile_jobs_sync);
 
 static INLINE void decrease_ref_count(int idx, RefCntBuffer *const frame_bufs,
                                       BufferPool *const pool) {
@@ -196,7 +207,6 @@ static INLINE void decrease_ref_count(int idx, RefCntBuffer *const frame_bufs,
   }
 }
 
-#if CONFIG_EXT_REFS || CONFIG_TEMPMV_SIGNALING
 static INLINE int dec_is_ref_frame_buf(AV1Decoder *const pbi,
                                        RefCntBuffer *frame_buf) {
   AV1_COMMON *const cm = &pbi->common;
@@ -208,7 +218,6 @@ static INLINE int dec_is_ref_frame_buf(AV1Decoder *const pbi,
   }
   return (i < INTER_REFS_PER_FRAME);
 }
-#endif  // CONFIG_EXT_REFS
 
 #define ACCT_STR __func__
 static INLINE int av1_read_uniform(aom_reader *r, int n) {
@@ -222,6 +231,13 @@ static INLINE int av1_read_uniform(aom_reader *r, int n) {
     return (v << 1) - m + aom_read_literal(r, 1, ACCT_STR);
 }
 
+typedef void (*palette_visitor_fn_t)(MACROBLOCKD *const xd, int plane,
+                                     aom_reader *r);
+
+void av1_visit_palette(AV1Decoder *const pbi, MACROBLOCKD *const xd, int mi_row,
+                       int mi_col, aom_reader *r, BLOCK_SIZE bsize,
+                       palette_visitor_fn_t visit);
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/third_party/aom/av1/decoder/decodetxb.c b/third_party/aom/av1/decoder/decodetxb.c
index 13f944b35..f9a3e8578 100644
--- a/third_party/aom/av1/decoder/decodetxb.c
+++ b/third_party/aom/av1/decoder/decodetxb.c
@@ -9,28 +9,25 @@
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */
 
-#include "av1/common/scan.h"
+#include "av1/decoder/decodetxb.h"
+
+#include "aom_ports/mem.h"
 #include "av1/common/idct.h"
+#include "av1/common/scan.h"
 #include "av1/common/txb_common.h"
 #include "av1/decoder/decodemv.h"
-#include "av1/decoder/decodetxb.h"
-#include "av1/decoder/dsubexp.h"
-#include "av1/decoder/symbolrate.h"
 
 #define ACCT_STR __func__
 
-static int read_golomb(MACROBLOCKD *xd, aom_reader *r, FRAME_COUNTS *counts) {
-#if !CONFIG_SYMBOLRATE
-  (void)counts;
-#endif
+static int read_golomb(MACROBLOCKD *xd, aom_reader *r) {
   int x = 1;
   int length = 0;
   int i = 0;
 
   while (!i) {
-    i = av1_read_record_bit(counts, r, ACCT_STR);
+    i = aom_read_bit(r, ACCT_STR);
     ++length;
-    if (length >= 32) {
+    if (length > 20) {
       aom_internal_error(xd->error_info, AOM_CODEC_CORRUPT_FRAME,
                          "Invalid length in read_golomb");
       break;
@@ -39,570 +36,306 @@ static int read_golomb(MACROBLOCKD *xd, aom_reader *r, FRAME_COUNTS *counts) {
 
   for (i = 0; i < length - 1; ++i) {
     x <<= 1;
-    x += av1_read_record_bit(counts, r, ACCT_STR);
+    x += aom_read_bit(r, ACCT_STR);
   }
 
   return x - 1;
 }
 
-static INLINE int read_nz_map(aom_reader *r, tran_low_t *tcoeffs, int plane,
-                              const int16_t *scan, TX_SIZE tx_size,
-                              TX_TYPE tx_type, FRAME_CONTEXT *fc,
-                              FRAME_COUNTS *counts) {
-  TX_SIZE txs_ctx = get_txsize_context(tx_size);
-  const int bwl = b_width_log2_lookup[txsize_to_bsize[tx_size]] + 2;
-  const int height = tx_size_high[tx_size];
-#if CONFIG_CTX1D
-  const int width = tx_size_wide[tx_size];
-  const int eob_offset = width + height;
-  const TX_CLASS tx_class = get_tx_class(tx_type);
-  const int seg_eob =
-      (tx_class == TX_CLASS_2D) ? tx_size_2d[tx_size] : eob_offset;
-#else
-  const int seg_eob = tx_size_2d[tx_size];
-#endif
-  const PLANE_TYPE plane_type = get_plane_type(plane);
-  unsigned int(*nz_map_count)[SIG_COEF_CONTEXTS][2] =
-      (counts) ? &counts->nz_map[txs_ctx][plane_type] : NULL;
-#if !LV_MAP_PROB
-  aom_prob *nz_map = fc->nz_map[txs_ctx][plane_type];
-  aom_prob *eob_flag = fc->eob_flag[txs_ctx][plane_type];
-#endif
-  int c;
-  for (c = 0; c < seg_eob; ++c) {
-    int is_nz;
-    int coeff_ctx = get_nz_map_ctx(tcoeffs, c, scan, bwl, height, tx_type);
-    int eob_ctx = get_eob_ctx(tcoeffs, scan[c], txs_ctx, tx_type);
-
-    if (c < seg_eob - 1) {
-#if LV_MAP_PROB
-      is_nz = av1_read_record_bin(
-          counts, r, fc->nz_map_cdf[txs_ctx][plane_type][coeff_ctx], 2,
-          ACCT_STR);
-#else
-      is_nz = aom_read(r, nz_map[coeff_ctx], ACCT_STR);
-#endif
-    } else {
-      is_nz = 1;
-    }
-
-    // set non-zero coefficient map.
-    tcoeffs[scan[c]] = is_nz;
-
-    if (c == seg_eob - 1) {
-      ++c;
-      break;
-    }
-
-    if (counts) ++(*nz_map_count)[coeff_ctx][is_nz];
-
-    if (is_nz) {
-#if LV_MAP_PROB
-      int is_eob = av1_read_record_bin(
-          counts, r, fc->eob_flag_cdf[txs_ctx][plane_type][eob_ctx], 2,
-          ACCT_STR);
-#else
-      int is_eob = aom_read(r, eob_flag[eob_ctx], ACCT_STR);
-#endif
-      if (counts) ++counts->eob_flag[txs_ctx][plane_type][eob_ctx][is_eob];
-      if (is_eob) break;
-    }
+static INLINE int rec_eob_pos(const int eob_token, const int extra) {
+  int eob = k_eob_group_start[eob_token];
+  if (eob > 2) {
+    eob += extra;
   }
-  return AOMMIN(seg_eob, c + 1);
+  return eob;
 }
 
-#if CONFIG_CTX1D
-static INLINE int read_nz_map_vert(aom_reader *r, tran_low_t *tcoeffs,
-                                   int plane, const int16_t *scan,
-                                   const int16_t *iscan, TX_SIZE tx_size,
-                                   TX_TYPE tx_type, FRAME_CONTEXT *fc,
-                                   FRAME_COUNTS *counts) {
-  const TX_SIZE txs_ctx = get_txsize_context(tx_size);
-  const PLANE_TYPE plane_type = get_plane_type(plane);
-  const TX_CLASS tx_class = get_tx_class(tx_type);
-  const int bwl = b_width_log2_lookup[txsize_to_bsize[tx_size]] + 2;
-  const int width = tx_size_wide[tx_size];
-  const int height = tx_size_high[tx_size];
-  int16_t eob_ls[MAX_HVTX_SIZE];
-  int eob = 0;
-#if !LV_MAP_PROB
-  aom_prob *nz_map = fc->nz_map[txs_ctx][plane_type];
-#endif
-  for (int col = 0; col < width; ++col) {
-    int el_ctx = get_empty_line_ctx(col, eob_ls);
-#if LV_MAP_PROB
-    int empty_line = av1_read_record_bin(
-        counts, r, fc->empty_line_cdf[txs_ctx][plane_type][tx_class][el_ctx], 2,
-        ACCT_STR);
-#else
-    int empty_line = aom_read(
-        r, fc->empty_line[txs_ctx][plane_type][tx_class][el_ctx], ACCT_STR);
-#endif
-    if (counts)
-      ++counts->empty_line[txs_ctx][plane_type][tx_class][el_ctx][empty_line];
-    if (!empty_line) {
-      int row;
-      for (row = 0; row < height; ++row) {
-        if (row + 1 != height) {
-          int coeff_idx = row * width + col;
-          int scan_idx = iscan[coeff_idx];
-          int coeff_ctx =
-              get_nz_map_ctx(tcoeffs, scan_idx, scan, bwl, height, tx_type);
-#if LV_MAP_PROB
-          int is_nz = av1_read_record_bin(
-              counts, r, fc->nz_map_cdf[txs_ctx][plane_type][coeff_ctx], 2,
-              ACCT_STR);
-#else
-          int is_nz = aom_read(r, nz_map[coeff_ctx], ACCT_STR);
-#endif
-          if (counts) ++counts->nz_map[txs_ctx][plane_type][coeff_ctx][is_nz];
-          tcoeffs[coeff_idx] = is_nz;
-          if (is_nz) {
-            eob = AOMMAX(eob, iscan[coeff_idx] + 1);
-            if (row + 1 != height) {
-              int eob_ctx = get_hv_eob_ctx(col, row, eob_ls);
-#if LV_MAP_PROB
-              int is_eob = av1_read_record_bin(
-                  counts, r,
-                  fc->hv_eob_cdf[txs_ctx][plane_type][tx_class][eob_ctx], 2,
-                  ACCT_STR);
-#else
-              int is_eob = aom_read(
-                  r, fc->hv_eob[txs_ctx][plane_type][tx_class][eob_ctx],
-                  ACCT_STR);
-#endif
-              if (counts)
-                ++counts
-                      ->hv_eob[txs_ctx][plane_type][tx_class][eob_ctx][is_eob];
-              if (is_eob) break;
-            }
-          }
-        } else {
-          int coeff_idx = row * width + col;
-          tcoeffs[coeff_idx] = 1;
-          eob = AOMMAX(eob, iscan[coeff_idx] + 1);
-        }
+static INLINE int get_dqv(const int16_t *dequant, int coeff_idx,
+                          const qm_val_t *iqmatrix) {
+  int dqv = dequant[!!coeff_idx];
+  if (iqmatrix != NULL)
+    dqv =
+        ((iqmatrix[coeff_idx] * dqv) + (1 << (AOM_QM_BITS - 1))) >> AOM_QM_BITS;
+  return dqv;
+}
+
+static INLINE void read_coeffs_reverse_2d(aom_reader *r, TX_SIZE tx_size,
+                                          int start_si, int end_si,
+                                          const int16_t *scan, int bwl,
+                                          uint8_t *levels,
+                                          base_cdf_arr base_cdf,
+                                          br_cdf_arr br_cdf) {
+  for (int c = end_si; c >= start_si; --c) {
+    const int pos = scan[c];
+    const int coeff_ctx = get_lower_levels_ctx_2d(levels, pos, bwl, tx_size);
+    const int nsymbs = 4;
+    int level = aom_read_symbol(r, base_cdf[coeff_ctx], nsymbs, ACCT_STR);
+    if (level > NUM_BASE_LEVELS) {
+      const int br_ctx = get_br_ctx_2d(levels, pos, bwl);
+      aom_cdf_prob *cdf = br_cdf[br_ctx];
+      for (int idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) {
+        const int k = aom_read_symbol(r, cdf, BR_CDF_SIZE, ACCT_STR);
+        level += k;
+        if (k < BR_CDF_SIZE - 1) break;
       }
-      eob_ls[col] = AOMMIN(height, row + 1);
-    } else {
-      eob_ls[col] = 0;
     }
+    levels[get_padded_idx(pos, bwl)] = level;
   }
-  return eob;
 }
 
-static INLINE int read_nz_map_horiz(aom_reader *r, tran_low_t *tcoeffs,
-                                    int plane, const int16_t *scan,
-                                    const int16_t *iscan, TX_SIZE tx_size,
-                                    TX_TYPE tx_type, FRAME_CONTEXT *fc,
-                                    FRAME_COUNTS *counts) {
-  const TX_SIZE txs_ctx = get_txsize_context(tx_size);
-  const PLANE_TYPE plane_type = get_plane_type(plane);
-  const TX_CLASS tx_class = get_tx_class(tx_type);
-  const int bwl = b_width_log2_lookup[txsize_to_bsize[tx_size]] + 2;
-  const int width = tx_size_wide[tx_size];
-  const int height = tx_size_high[tx_size];
-  int16_t eob_ls[MAX_HVTX_SIZE];
-  int eob = 0;
-#if !LV_MAP_PROB
-  aom_prob *nz_map = fc->nz_map[txs_ctx][plane_type];
-#endif
-  for (int row = 0; row < height; ++row) {
-    int el_ctx = get_empty_line_ctx(row, eob_ls);
-#if LV_MAP_PROB
-    int empty_line = av1_read_record_bin(
-        counts, r, fc->empty_line_cdf[txs_ctx][plane_type][tx_class][el_ctx], 2,
-        ACCT_STR);
-#else
-    int empty_line = aom_read(
-        r, fc->empty_line[txs_ctx][plane_type][tx_class][el_ctx], ACCT_STR);
-#endif
-    if (counts)
-      ++counts->empty_line[txs_ctx][plane_type][tx_class][el_ctx][empty_line];
-    if (!empty_line) {
-      int col;
-      for (col = 0; col < width; ++col) {
-        if (col + 1 != width) {
-          int coeff_idx = row * width + col;
-          int scan_idx = iscan[coeff_idx];
-          int coeff_ctx =
-              get_nz_map_ctx(tcoeffs, scan_idx, scan, bwl, height, tx_type);
-#if LV_MAP_PROB
-          int is_nz = av1_read_record_bin(
-              counts, r, fc->nz_map_cdf[txs_ctx][plane_type][coeff_ctx], 2,
-              ACCT_STR);
-#else
-          int is_nz = aom_read(r, nz_map[coeff_ctx], ACCT_STR);
-#endif
-          if (counts) ++counts->nz_map[txs_ctx][plane_type][coeff_ctx][is_nz];
-          tcoeffs[coeff_idx] = is_nz;
-          if (is_nz) {
-            eob = AOMMAX(eob, iscan[coeff_idx] + 1);
-            int eob_ctx = get_hv_eob_ctx(row, col, eob_ls);
-#if LV_MAP_PROB
-            int is_eob = av1_read_record_bin(
-                counts, r,
-                fc->hv_eob_cdf[txs_ctx][plane_type][tx_class][eob_ctx], 2,
-                ACCT_STR);
-#else
-            int is_eob =
-                aom_read(r, fc->hv_eob[txs_ctx][plane_type][tx_class][eob_ctx],
-                         ACCT_STR);
-#endif
-            if (counts)
-              ++counts->hv_eob[txs_ctx][plane_type][tx_class][eob_ctx][is_eob];
-            if (is_eob) break;
-          }
-        } else {
-          int coeff_idx = row * width + col;
-          tcoeffs[coeff_idx] = 1;
-          eob = AOMMAX(eob, iscan[coeff_idx] + 1);
-        }
+static INLINE void read_coeffs_reverse(aom_reader *r, TX_SIZE tx_size,
+                                       TX_CLASS tx_class, int start_si,
+                                       int end_si, const int16_t *scan, int bwl,
+                                       uint8_t *levels, base_cdf_arr base_cdf,
+                                       br_cdf_arr br_cdf) {
+  for (int c = end_si; c >= start_si; --c) {
+    const int pos = scan[c];
+    const int coeff_ctx =
+        get_lower_levels_ctx(levels, pos, bwl, tx_size, tx_class);
+    const int nsymbs = 4;
+    int level = aom_read_symbol(r, base_cdf[coeff_ctx], nsymbs, ACCT_STR);
+    if (level > NUM_BASE_LEVELS) {
+      const int br_ctx = get_br_ctx(levels, pos, bwl, tx_class);
+      aom_cdf_prob *cdf = br_cdf[br_ctx];
+      for (int idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) {
+        const int k = aom_read_symbol(r, cdf, BR_CDF_SIZE, ACCT_STR);
+        level += k;
+        if (k < BR_CDF_SIZE - 1) break;
       }
-      eob_ls[row] = AOMMIN(width, col + 1);
-    } else {
-      eob_ls[row] = 0;
     }
+    levels[get_padded_idx(pos, bwl)] = level;
   }
-  return eob;
 }
-#endif
 
-uint8_t av1_read_coeffs_txb(const AV1_COMMON *const cm, MACROBLOCKD *xd,
-                            aom_reader *r, int blk_row, int blk_col, int block,
-                            int plane, tran_low_t *tcoeffs, TXB_CTX *txb_ctx,
-                            TX_SIZE tx_size, int16_t *max_scan_line, int *eob) {
-  FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
-  FRAME_COUNTS *counts = xd->counts;
-  TX_SIZE txs_ctx = get_txsize_context(tx_size);
-  PLANE_TYPE plane_type = get_plane_type(plane);
-#if !LV_MAP_PROB
-  aom_prob *nz_map = ec_ctx->nz_map[txs_ctx][plane_type];
-  aom_prob *eob_flag = ec_ctx->eob_flag[txs_ctx][plane_type];
-#endif
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
-  const int seg_eob = tx_size_2d[tx_size];
-  int c = 0;
-  int update_eob = -1;
-  const int16_t *const dequant = xd->plane[plane].seg_dequant[mbmi->segment_id];
+uint8_t av1_read_coeffs_txb(const AV1_COMMON *const cm, MACROBLOCKD *const xd,
+                            aom_reader *const r, const int blk_row,
+                            const int blk_col, const int plane,
+                            const TXB_CTX *const txb_ctx,
+                            const TX_SIZE tx_size) {
+  FRAME_CONTEXT *const ec_ctx = xd->tile_ctx;
+  const int32_t max_value = (1 << (7 + xd->bd)) - 1;
+  const int32_t min_value = -(1 << (7 + xd->bd));
+  const TX_SIZE txs_ctx = get_txsize_entropy_ctx(tx_size);
+  const PLANE_TYPE plane_type = get_plane_type(plane);
+  MB_MODE_INFO *const mbmi = xd->mi[0];
+  struct macroblockd_plane *const pd = &xd->plane[plane];
+  const int16_t *const dequant = pd->seg_dequant_QTX[mbmi->segment_id];
+  tran_low_t *const tcoeffs = pd->dqcoeff_block + xd->cb_offset[plane];
   const int shift = av1_get_tx_scale(tx_size);
-  const int bwl = b_width_log2_lookup[txsize_to_bsize[tx_size]] + 2;
-  const int height = tx_size_high[tx_size];
+  const int bwl = get_txb_bwl(tx_size);
+  const int width = get_txb_wide(tx_size);
+  const int height = get_txb_high(tx_size);
   int cul_level = 0;
-  memset(tcoeffs, 0, sizeof(*tcoeffs) * seg_eob);
-
-#if LV_MAP_PROB
-  int all_zero = av1_read_record_bin(
-      counts, r, ec_ctx->txb_skip_cdf[txs_ctx][txb_ctx->txb_skip_ctx], 2,
-      ACCT_STR);
-#else
-  int all_zero =
-      aom_read(r, ec_ctx->txb_skip[txs_ctx][txb_ctx->txb_skip_ctx], ACCT_STR);
-#endif
-  if (xd->counts)
-    ++xd->counts->txb_skip[txs_ctx][txb_ctx->txb_skip_ctx][all_zero];
-
+  int dc_val = 0;
+  uint8_t levels_buf[TX_PAD_2D];
+  uint8_t *const levels = set_levels(levels_buf, width);
+  const int all_zero = aom_read_symbol(
+      r, ec_ctx->txb_skip_cdf[txs_ctx][txb_ctx->txb_skip_ctx], 2, ACCT_STR);
+  eob_info *eob_data = pd->eob_data + xd->txb_offset[plane];
+  uint16_t *const eob = &(eob_data->eob);
+  uint16_t *const max_scan_line = &(eob_data->max_scan_line);
+  *max_scan_line = 0;
   *eob = 0;
   if (all_zero) {
     *max_scan_line = 0;
-#if CONFIG_TXK_SEL
-    if (plane == 0) mbmi->txk_type[(blk_row << 4) + blk_col] = DCT_DCT;
-#endif
+    if (plane == 0) {
+      const int txk_type_idx =
+          av1_get_txk_type_index(mbmi->sb_type, blk_row, blk_col);
+      mbmi->txk_type[txk_type_idx] = DCT_DCT;
+    }
     return 0;
   }
 
-  (void)blk_row;
-  (void)blk_col;
-#if CONFIG_TXK_SEL
-  av1_read_tx_type(cm, xd, blk_row, blk_col, block, plane,
-                   get_min_tx_size(tx_size), r);
-#endif
-  const TX_TYPE tx_type =
-      av1_get_tx_type(plane_type, xd, blk_row, blk_col, block, tx_size);
-  const SCAN_ORDER *const scan_order = get_scan(cm, tx_size, tx_type, mbmi);
-  const int16_t *scan = scan_order->scan;
-
-#if CONFIG_CTX1D
-  const int16_t *iscan = scan_order->iscan;
-  TX_CLASS tx_class = get_tx_class(tx_type);
-  if (tx_class == TX_CLASS_2D) {
-    *eob =
-        read_nz_map(r, tcoeffs, plane, scan, tx_size, tx_type, ec_ctx, counts);
-  } else {
-#if LV_MAP_PROB
-    const int eob_mode = av1_read_record_bin(
-        counts, r, ec_ctx->eob_mode_cdf[txs_ctx][plane_type][tx_class], 2,
-        ACCT_STR);
-#else
-    const int eob_mode =
-        aom_read(r, ec_ctx->eob_mode[txs_ctx][plane_type][tx_class], ACCT_STR);
-#endif
-    if (counts) ++counts->eob_mode[txs_ctx][plane_type][tx_class][eob_mode];
-    if (eob_mode == 0) {
-      *eob = read_nz_map(r, tcoeffs, plane, scan, tx_size, tx_type, ec_ctx,
-                         counts);
-    } else {
-      assert(tx_class == TX_CLASS_VERT || tx_class == TX_CLASS_HORIZ);
-      if (tx_class == TX_CLASS_VERT)
-        *eob = read_nz_map_vert(r, tcoeffs, plane, scan, iscan, tx_size,
-                                tx_type, ec_ctx, counts);
-      else
-        *eob = read_nz_map_horiz(r, tcoeffs, plane, scan, iscan, tx_size,
-                                 tx_type, ec_ctx, counts);
-    }
+  memset(levels_buf, 0,
+         sizeof(*levels_buf) *
+             ((width + TX_PAD_HOR) * (height + TX_PAD_VER) + TX_PAD_END));
+  if (plane == AOM_PLANE_Y) {
+    // only y plane's tx_type is transmitted
+    av1_read_tx_type(cm, xd, blk_row, blk_col, tx_size, r);
+  }
+  const TX_TYPE tx_type = av1_get_tx_type(plane_type, xd, blk_row, blk_col,
+                                          tx_size, cm->reduced_tx_set_used);
+  const TX_CLASS tx_class = tx_type_to_class[tx_type];
+  const TX_SIZE qm_tx_size = av1_get_adjusted_tx_size(tx_size);
+  const qm_val_t *iqmatrix =
+      IS_2D_TRANSFORM(tx_type)
+          ? pd->seg_iqmatrix[mbmi->segment_id][qm_tx_size]
+          : cm->giqmatrix[NUM_QM_LEVELS - 1][0][qm_tx_size];
+  const SCAN_ORDER *const scan_order = get_scan(tx_size, tx_type);
+  const int16_t *const scan = scan_order->scan;
+  int eob_extra = 0;
+  int eob_pt = 1;
+
+  const int eob_multi_size = txsize_log2_minus4[tx_size];
+  const int eob_multi_ctx = (tx_class == TX_CLASS_2D) ? 0 : 1;
+  switch (eob_multi_size) {
+    case 0:
+      eob_pt =
+          aom_read_symbol(r, ec_ctx->eob_flag_cdf16[plane_type][eob_multi_ctx],
+                          5, ACCT_STR) +
+          1;
+      break;
+    case 1:
+      eob_pt =
+          aom_read_symbol(r, ec_ctx->eob_flag_cdf32[plane_type][eob_multi_ctx],
+                          6, ACCT_STR) +
+          1;
+      break;
+    case 2:
+      eob_pt =
+          aom_read_symbol(r, ec_ctx->eob_flag_cdf64[plane_type][eob_multi_ctx],
+                          7, ACCT_STR) +
+          1;
+      break;
+    case 3:
+      eob_pt =
+          aom_read_symbol(r, ec_ctx->eob_flag_cdf128[plane_type][eob_multi_ctx],
+                          8, ACCT_STR) +
+          1;
+      break;
+    case 4:
+      eob_pt =
+          aom_read_symbol(r, ec_ctx->eob_flag_cdf256[plane_type][eob_multi_ctx],
+                          9, ACCT_STR) +
+          1;
+      break;
+    case 5:
+      eob_pt =
+          aom_read_symbol(r, ec_ctx->eob_flag_cdf512[plane_type][eob_multi_ctx],
+                          10, ACCT_STR) +
+          1;
+      break;
+    case 6:
+    default:
+      eob_pt = aom_read_symbol(
+                   r, ec_ctx->eob_flag_cdf1024[plane_type][eob_multi_ctx], 11,
+                   ACCT_STR) +
+               1;
+      break;
   }
-#else
-  *eob = read_nz_map(r, tcoeffs, plane, scan, tx_size, tx_type, ec_ctx, counts);
-#endif
-  *max_scan_line = *eob;
-
-  int i;
-  for (i = 0; i < NUM_BASE_LEVELS; ++i) {
-#if !LV_MAP_PROB
-    aom_prob *coeff_base = ec_ctx->coeff_base[txs_ctx][plane_type][i];
-#endif
-    update_eob = 0;
-    for (c = *eob - 1; c >= 0; --c) {
-      tran_low_t *v = &tcoeffs[scan[c]];
-      int sign;
-      int ctx;
-
-      if (*v <= i) continue;
-
-      ctx = get_base_ctx(tcoeffs, scan[c], bwl, height, i + 1);
-
-#if LV_MAP_PROB
-      if (av1_read_record_bin(
-              counts, r, ec_ctx->coeff_base_cdf[txs_ctx][plane_type][i][ctx], 2,
-              ACCT_STR))
-#else
-      if (aom_read(r, coeff_base[ctx], ACCT_STR))
-#endif
-      {
-        *v = i + 1;
-        cul_level += i + 1;
 
-        if (counts) ++counts->coeff_base[txs_ctx][plane_type][i][ctx][1];
+  if (k_eob_offset_bits[eob_pt] > 0) {
+    const int eob_ctx = eob_pt - 3;
+    int bit = aom_read_symbol(
+        r, ec_ctx->eob_extra_cdf[txs_ctx][plane_type][eob_ctx], 2, ACCT_STR);
+    if (bit) {
+      eob_extra += (1 << (k_eob_offset_bits[eob_pt] - 1));
+    }
 
-        if (c == 0) {
-          int dc_sign_ctx = txb_ctx->dc_sign_ctx;
-#if LV_MAP_PROB
-          sign = av1_read_record_bin(
-              counts, r, ec_ctx->dc_sign_cdf[plane_type][dc_sign_ctx], 2,
-              ACCT_STR);
-#else
-          sign =
-              aom_read(r, ec_ctx->dc_sign[plane_type][dc_sign_ctx], ACCT_STR);
-#endif
-          if (counts) ++counts->dc_sign[plane_type][dc_sign_ctx][sign];
-        } else {
-          sign = av1_read_record_bit(counts, r, ACCT_STR);
-        }
-        if (sign) *v = -(*v);
-        continue;
+    for (int i = 1; i < k_eob_offset_bits[eob_pt]; i++) {
+      bit = aom_read_bit(r, ACCT_STR);
+      if (bit) {
+        eob_extra += (1 << (k_eob_offset_bits[eob_pt] - 1 - i));
       }
-      *v = i + 2;
-      if (counts) ++counts->coeff_base[txs_ctx][plane_type][i][ctx][0];
-
-      // update the eob flag for coefficients with magnitude above 1.
-      update_eob = AOMMAX(update_eob, c);
     }
   }
-
-  for (c = update_eob; c >= 0; --c) {
-    tran_low_t *v = &tcoeffs[scan[c]];
-    int sign;
-    int idx;
-    int ctx;
-
-    if (*v <= NUM_BASE_LEVELS) continue;
-
-    if (c == 0) {
-      int dc_sign_ctx = txb_ctx->dc_sign_ctx;
-#if LV_MAP_PROB
-      sign = av1_read_record_bin(
-          counts, r, ec_ctx->dc_sign_cdf[plane_type][dc_sign_ctx], 2, ACCT_STR);
-#else
-      sign = aom_read(r, ec_ctx->dc_sign[plane_type][dc_sign_ctx], ACCT_STR);
-#endif
-      if (counts) ++counts->dc_sign[plane_type][dc_sign_ctx][sign];
+  *eob = rec_eob_pos(eob_pt, eob_extra);
+
+  {
+    // Read the non-zero coefficient with scan index eob-1
+    // TODO(angiebird): Put this into a function
+    const int c = *eob - 1;
+    const int pos = scan[c];
+    const int coeff_ctx = get_lower_levels_ctx_eob(bwl, height, c);
+    const int nsymbs = 3;
+    aom_cdf_prob *cdf =
+        ec_ctx->coeff_base_eob_cdf[txs_ctx][plane_type][coeff_ctx];
+    int level = aom_read_symbol(r, cdf, nsymbs, ACCT_STR) + 1;
+    if (level > NUM_BASE_LEVELS) {
+      const int br_ctx = get_br_ctx(levels, pos, bwl, tx_class);
+      for (int idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) {
+        const int k = aom_read_symbol(
+            r,
+            ec_ctx->coeff_br_cdf[AOMMIN(txs_ctx, TX_32X32)][plane_type][br_ctx],
+            BR_CDF_SIZE, ACCT_STR);
+        level += k;
+        if (k < BR_CDF_SIZE - 1) break;
+      }
+    }
+    levels[get_padded_idx(pos, bwl)] = level;
+  }
+  if (*eob > 1) {
+    base_cdf_arr base_cdf = ec_ctx->coeff_base_cdf[txs_ctx][plane_type];
+    br_cdf_arr br_cdf =
+        ec_ctx->coeff_br_cdf[AOMMIN(txs_ctx, TX_32X32)][plane_type];
+    if (tx_class == TX_CLASS_2D) {
+      read_coeffs_reverse_2d(r, tx_size, 1, *eob - 1 - 1, scan, bwl, levels,
+                             base_cdf, br_cdf);
+      read_coeffs_reverse(r, tx_size, tx_class, 0, 0, scan, bwl, levels,
+                          base_cdf, br_cdf);
     } else {
-      sign = av1_read_record_bit(counts, r, ACCT_STR);
+      read_coeffs_reverse(r, tx_size, tx_class, 0, *eob - 1 - 1, scan, bwl,
+                          levels, base_cdf, br_cdf);
     }
+  }
 
-    ctx = get_br_ctx(tcoeffs, scan[c], bwl, height);
-
-#if BR_NODE
-    for (idx = 0; idx < BASE_RANGE_SETS; ++idx) {
-#if LV_MAP_PROB
-      if (av1_read_record_bin(
-              counts, r, ec_ctx->coeff_br_cdf[txs_ctx][plane_type][idx][ctx], 2,
-              ACCT_STR))
-#else   // LV_MAP_PROB
-      if (aom_read(r, ec_ctx->coeff_br[txs_ctx][plane_type][idx][ctx],
-                   ACCT_STR))
-#endif  // LV_MAP_PROB
-      {
-        int extra_bits = (1 << br_extra_bits[idx]) - 1;
-        //        int br_offset = aom_read_literal(r, extra_bits, ACCT_STR);
-        int br_offset = 0;
-        int tok;
-        if (counts) ++counts->coeff_br[txs_ctx][plane_type][idx][ctx][1];
-        for (tok = 0; tok < extra_bits; ++tok) {
-#if LV_MAP_PROB
-          if (av1_read_record_bin(
-                  counts, r, ec_ctx->coeff_lps_cdf[txs_ctx][plane_type][ctx], 2,
-                  ACCT_STR))
-#else
-          if (aom_read(r, ec_ctx->coeff_lps[txs_ctx][plane_type][ctx],
-                       ACCT_STR))
-#endif
-          {
-            br_offset = tok;
-            if (counts) ++counts->coeff_lps[txs_ctx][plane_type][ctx][1];
-            break;
-          }
-          if (counts) ++counts->coeff_lps[txs_ctx][plane_type][ctx][0];
-        }
-        if (tok == extra_bits) br_offset = extra_bits;
-
-        int br_base = br_index_to_coeff[idx];
-
-        *v = NUM_BASE_LEVELS + 1 + br_base + br_offset;
-        cul_level += *v;
-        if (sign) *v = -(*v);
-        break;
+  int16_t num_zero_coeffs = 0;
+  for (int c = 0; c < *eob; ++c) {
+    const int pos = scan[c];
+    num_zero_coeffs = AOMMAX(num_zero_coeffs, pos);
+  }
+  memset(tcoeffs, 0, (num_zero_coeffs + 1) * sizeof(tcoeffs[0]));
+
+  for (int c = 0; c < *eob; ++c) {
+    const int pos = scan[c];
+    uint8_t sign;
+    tran_low_t level = levels[get_padded_idx(pos, bwl)];
+    if (level) {
+      *max_scan_line = AOMMAX(*max_scan_line, pos);
+      if (c == 0) {
+        const int dc_sign_ctx = txb_ctx->dc_sign_ctx;
+        sign = aom_read_symbol(r, ec_ctx->dc_sign_cdf[plane_type][dc_sign_ctx],
+                               2, ACCT_STR);
+      } else {
+        sign = aom_read_bit(r, ACCT_STR);
+      }
+      if (level >= MAX_BASE_BR_RANGE) {
+        level += read_golomb(xd, r);
       }
-      if (counts) ++counts->coeff_br[txs_ctx][plane_type][idx][ctx][0];
-    }
-
-    if (idx < BASE_RANGE_SETS) continue;
-#else
-    for (idx = 0; idx < COEFF_BASE_RANGE; ++idx) {
-#if LV_MAP_PROB
-      if (av1_read_record_bin(counts, r,
-                              ec_ctx->coeff_lps_cdf[txs_ctx][plane_type][ctx],
-                              2, ACCT_STR))
-#else
-      if (aom_read(r, ec_ctx->coeff_lps[txs_ctx][plane_type][ctx], ACCT_STR))
-#endif
-      {
-        *v = (idx + 1 + NUM_BASE_LEVELS);
-        if (sign) *v = -(*v);
-        cul_level += abs(*v);
 
-        if (counts) ++counts->coeff_lps[txs_ctx][plane_type][ctx][1];
-        break;
+      if (c == 0) dc_val = sign ? -level : level;
+
+      // Bitmasking to clamp level to valid range:
+      //   The valid range for 8/10/12 bit vdieo is at most 14/16/18 bit
+      level &= 0xfffff;
+      cul_level += level;
+      tran_low_t dq_coeff;
+      // Bitmasking to clamp dq_coeff to valid range:
+      //   The valid range for 8/10/12 bit video is at most 17/19/21 bit
+      dq_coeff = (tran_low_t)(
+          (int64_t)level * get_dqv(dequant, scan[c], iqmatrix) & 0xffffff);
+      dq_coeff = dq_coeff >> shift;
+      if (sign) {
+        dq_coeff = -dq_coeff;
       }
-      if (counts) ++counts->coeff_lps[txs_ctx][plane_type][ctx][0];
+      tcoeffs[pos] = clamp(dq_coeff, min_value, max_value);
     }
-    if (idx < COEFF_BASE_RANGE) continue;
-#endif
-
-    // decode 0-th order Golomb code
-    *v = read_golomb(xd, r, counts) + COEFF_BASE_RANGE + 1 + NUM_BASE_LEVELS;
-    if (sign) *v = -(*v);
-    cul_level += abs(*v);
-  }
-
-  for (c = 0; c < *eob; ++c) {
-    int16_t dqv = (c == 0) ? dequant[0] : dequant[1];
-    tran_low_t *v = &tcoeffs[scan[c]];
-#if CONFIG_SYMBOLRATE
-    av1_record_coeff(counts, abs(*v));
-#endif
-    int sign = (*v) < 0;
-    *v = (abs(*v) * dqv) >> shift;
-    if (sign) *v = -(*v);
   }
 
-  cul_level = AOMMIN(63, cul_level);
+  cul_level = AOMMIN(COEFF_CONTEXT_MASK, cul_level);
 
   // DC value
-  set_dc_sign(&cul_level, tcoeffs[0]);
+  set_dc_sign(&cul_level, dc_val);
 
   return cul_level;
 }
 
-uint8_t av1_read_coeffs_txb_facade(AV1_COMMON *cm, MACROBLOCKD *xd,
-                                   aom_reader *r, int row, int col, int block,
-                                   int plane, tran_low_t *tcoeffs,
-                                   TX_SIZE tx_size, int16_t *max_scan_line,
-                                   int *eob) {
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
-  struct macroblockd_plane *pd = &xd->plane[plane];
+uint8_t av1_read_coeffs_txb_facade(const AV1_COMMON *const cm,
+                                   MACROBLOCKD *const xd, aom_reader *const r,
+                                   const int row, const int col,
+                                   const int plane, const TX_SIZE tx_size) {
+  MB_MODE_INFO *const mbmi = xd->mi[0];
+  struct macroblockd_plane *const pd = &xd->plane[plane];
 
   const BLOCK_SIZE bsize = mbmi->sb_type;
-#if CONFIG_CHROMA_SUB8X8
-  const BLOCK_SIZE plane_bsize =
-      AOMMAX(BLOCK_4X4, get_plane_block_size(bsize, pd));
-#elif CONFIG_CB4X4
-  const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
-#else   // CONFIG_CB4X4
   const BLOCK_SIZE plane_bsize =
-      get_plane_block_size(AOMMAX(BLOCK_8X8, bsize), pd);
-#endif  // CONFIG_CB4X4
+      get_plane_block_size(bsize, pd->subsampling_x, pd->subsampling_y);
 
   TXB_CTX txb_ctx;
   get_txb_ctx(plane_bsize, tx_size, plane, pd->above_context + col,
               pd->left_context + row, &txb_ctx);
-  uint8_t cul_level =
-      av1_read_coeffs_txb(cm, xd, r, row, col, block, plane, tcoeffs, &txb_ctx,
-                          tx_size, max_scan_line, eob);
-#if CONFIG_ADAPT_SCAN
-  PLANE_TYPE plane_type = get_plane_type(plane);
-  TX_TYPE tx_type = av1_get_tx_type(plane_type, xd, row, col, block, tx_size);
-  if (xd->counts && *eob > 0)
-    av1_update_scan_count_facade(cm, xd->counts, tx_size, tx_type, pd->dqcoeff,
-                                 *eob);
-#endif
-  av1_set_contexts(xd, pd, plane, tx_size, cul_level, col, row);
+  const uint8_t cul_level =
+      av1_read_coeffs_txb(cm, xd, r, row, col, plane, &txb_ctx, tx_size);
+  av1_set_contexts(xd, pd, plane, plane_bsize, tx_size, cul_level, col, row);
   return cul_level;
 }
-
-#if !LV_MAP_PROB
-static void read_txb_probs(FRAME_CONTEXT *fc, const TX_SIZE tx_size,
-                           aom_reader *r, FRAME_COUNTS *counts) {
-#if !CONFIG_SYMBOLRATE
-  (void)counts;
-#endif
-  int plane, ctx, level;
-
-  if (av1_read_record_bit(counts, r, ACCT_STR) == 0) return;
-
-  for (ctx = 0; ctx < TXB_SKIP_CONTEXTS; ++ctx)
-    av1_diff_update_prob(r, &fc->txb_skip[tx_size][ctx], ACCT_STR);
-
-  for (plane = 0; plane < PLANE_TYPES; ++plane)
-    for (ctx = 0; ctx < SIG_COEF_CONTEXTS; ++ctx)
-      av1_diff_update_prob(r, &fc->nz_map[tx_size][plane][ctx], ACCT_STR);
-
-  for (plane = 0; plane < PLANE_TYPES; ++plane)
-    for (ctx = 0; ctx < EOB_COEF_CONTEXTS; ++ctx)
-      av1_diff_update_prob(r, &fc->eob_flag[tx_size][plane][ctx], ACCT_STR);
-
-  for (level = 0; level < NUM_BASE_LEVELS; ++level)
-    for (plane = 0; plane < PLANE_TYPES; ++plane)
-      for (ctx = 0; ctx < COEFF_BASE_CONTEXTS; ++ctx)
-        av1_diff_update_prob(r, &fc->coeff_base[tx_size][plane][level][ctx],
-                             ACCT_STR);
-
-  for (plane = 0; plane < PLANE_TYPES; ++plane)
-    for (ctx = 0; ctx < LEVEL_CONTEXTS; ++ctx)
-      av1_diff_update_prob(r, &fc->coeff_lps[tx_size][plane][ctx], ACCT_STR);
-}
-
-void av1_read_txb_probs(FRAME_CONTEXT *fc, TX_MODE tx_mode, aom_reader *r,
-                        FRAME_COUNTS *counts) {
-  const TX_SIZE max_tx_size = tx_mode_to_biggest_tx_size[tx_mode];
-  TX_SIZE tx_size;
-  int ctx, plane;
-
-  for (plane = 0; plane < PLANE_TYPES; ++plane)
-    for (ctx = 0; ctx < DC_SIGN_CONTEXTS; ++ctx)
-      av1_diff_update_prob(r, &fc->dc_sign[plane][ctx], ACCT_STR);
-
-  for (tx_size = TX_4X4; tx_size <= max_tx_size; ++tx_size)
-    read_txb_probs(fc, tx_size, r, counts);
-}
-#endif  // !LV_MAP_PROB
diff --git a/third_party/aom/av1/decoder/decodetxb.h b/third_party/aom/av1/decoder/decodetxb.h
index 1c6512e97..d0b3d8c7a 100644
--- a/third_party/aom/av1/decoder/decodetxb.h
+++ b/third_party/aom/av1/decoder/decodetxb.h
@@ -12,24 +12,21 @@
 #ifndef DECODETXB_H_
 #define DECODETXB_H_
 
-#include "./aom_config.h"
+#include "config/aom_config.h"
+
 #include "av1/common/blockd.h"
 #include "av1/common/onyxc_int.h"
 #include "av1/common/txb_common.h"
 #include "aom_dsp/bitreader.h"
 
-uint8_t av1_read_coeffs_txb(const AV1_COMMON *const cm, MACROBLOCKD *xd,
-                            aom_reader *r, int blk_row, int blk_col, int block,
-                            int plane, tran_low_t *tcoeffs, TXB_CTX *txb_ctx,
-                            TX_SIZE tx_size, int16_t *max_scan_line, int *eob);
+uint8_t av1_read_coeffs_txb(const AV1_COMMON *const cm, MACROBLOCKD *const xd,
+                            aom_reader *const r, const int blk_row,
+                            const int blk_col, const int plane,
+                            const TXB_CTX *const txb_ctx,
+                            const TX_SIZE tx_size);
 
-uint8_t av1_read_coeffs_txb_facade(AV1_COMMON *cm, MACROBLOCKD *xd,
-                                   aom_reader *r, int row, int col, int block,
-                                   int plane, tran_low_t *tcoeffs,
-                                   TX_SIZE tx_size, int16_t *max_scan_line,
-                                   int *eob);
-#if !LV_MAP_PROB
-void av1_read_txb_probs(FRAME_CONTEXT *fc, TX_MODE tx_mode, aom_reader *r,
-                        FRAME_COUNTS *counts);
-#endif  // !LV_MAP_PROB
+uint8_t av1_read_coeffs_txb_facade(const AV1_COMMON *const cm,
+                                   MACROBLOCKD *const xd, aom_reader *const r,
+                                   const int row, const int col,
+                                   const int plane, const TX_SIZE tx_size);
 #endif  //  DECODETXB_H_
diff --git a/third_party/aom/av1/decoder/detokenize.c b/third_party/aom/av1/decoder/detokenize.c
index a59a7bac1..9d54bd13d 100644
--- a/third_party/aom/av1/decoder/detokenize.c
+++ b/third_party/aom/av1/decoder/detokenize.c
@@ -9,245 +9,18 @@
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */
 
-#include "./aom_config.h"
-#if !CONFIG_PVQ
+#include "config/aom_config.h"
+
 #include "aom_mem/aom_mem.h"
 #include "aom_ports/mem.h"
-#endif  // !CONFIG_PVQ
-
 #include "av1/common/blockd.h"
 #include "av1/decoder/detokenize.h"
 
 #define ACCT_STR __func__
 
-#if !CONFIG_PVQ || CONFIG_VAR_TX
 #include "av1/common/common.h"
 #include "av1/common/entropy.h"
 #include "av1/common/idct.h"
-#endif
-
-#include "av1/decoder/symbolrate.h"
-
-#if !CONFIG_PVQ || CONFIG_VAR_TX
-#define EOB_CONTEXT_NODE 0
-#define ZERO_CONTEXT_NODE 1
-#define ONE_CONTEXT_NODE 2
-#define LOW_VAL_CONTEXT_NODE 0
-#define TWO_CONTEXT_NODE 1
-#define THREE_CONTEXT_NODE 2
-#define HIGH_LOW_CONTEXT_NODE 3
-#define CAT_ONE_CONTEXT_NODE 4
-#define CAT_THREEFOUR_CONTEXT_NODE 5
-#define CAT_THREE_CONTEXT_NODE 6
-#define CAT_FIVE_CONTEXT_NODE 7
-
-#define INCREMENT_COUNT(token)                   \
-  do {                                           \
-    if (counts) ++coef_counts[band][ctx][token]; \
-  } while (0)
-
-#if CONFIG_NEW_MULTISYMBOL
-#define READ_COEFF(counts, prob_name, cdf_name, num, r) \
-  read_coeff(counts, cdf_name, num, r);
-static INLINE int read_coeff(FRAME_COUNTS *counts,
-                             const aom_cdf_prob *const *cdf, int n,
-                             aom_reader *r) {
-#if !CONFIG_SYMBOLRATE
-  (void)counts;
-#endif
-  int val = 0;
-  int i = 0;
-  int count = 0;
-  while (count < n) {
-    const int size = AOMMIN(n - count, 4);
-    val |= av1_read_record_cdf(counts, r, cdf[i++], 1 << size, ACCT_STR)
-           << count;
-    count += size;
-  }
-  return val;
-}
-#else
-#define READ_COEFF(counts, prob_name, cdf_name, num, r) \
-  read_coeff(counts, prob_name, num, r);
-static INLINE int read_coeff(FRAME_COUNTS *counts, const aom_prob *probs, int n,
-                             aom_reader *r) {
-#if !CONFIG_SYMBOLRATE
-  (void)counts;
-#endif
-  int i, val = 0;
-  for (i = 0; i < n; ++i)
-    val = (val << 1) | av1_read_record(counts, r, probs[i], ACCT_STR);
-  return val;
-}
-
-#endif
-
-static int token_to_value(FRAME_COUNTS *counts, aom_reader *const r, int token,
-                          TX_SIZE tx_size, int bit_depth) {
-#if !CONFIG_HIGHBITDEPTH
-  assert(bit_depth == 8);
-#endif  // !CONFIG_HIGHBITDEPTH
-
-  switch (token) {
-    case ZERO_TOKEN:
-    case ONE_TOKEN:
-    case TWO_TOKEN:
-    case THREE_TOKEN:
-    case FOUR_TOKEN: return token;
-    case CATEGORY1_TOKEN:
-      return CAT1_MIN_VAL +
-             READ_COEFF(counts, av1_cat1_prob, av1_cat1_cdf, 1, r);
-    case CATEGORY2_TOKEN:
-      return CAT2_MIN_VAL +
-             READ_COEFF(counts, av1_cat2_prob, av1_cat2_cdf, 2, r);
-    case CATEGORY3_TOKEN:
-      return CAT3_MIN_VAL +
-             READ_COEFF(counts, av1_cat3_prob, av1_cat3_cdf, 3, r);
-    case CATEGORY4_TOKEN:
-      return CAT4_MIN_VAL +
-             READ_COEFF(counts, av1_cat4_prob, av1_cat4_cdf, 4, r);
-    case CATEGORY5_TOKEN:
-      return CAT5_MIN_VAL +
-             READ_COEFF(counts, av1_cat5_prob, av1_cat5_cdf, 5, r);
-    case CATEGORY6_TOKEN: {
-      const int skip_bits = (int)sizeof(av1_cat6_prob) -
-                            av1_get_cat6_extrabits_size(tx_size, bit_depth);
-      return CAT6_MIN_VAL + READ_COEFF(counts, av1_cat6_prob + skip_bits,
-                                       av1_cat6_cdf, 18 - skip_bits, r);
-    }
-    default:
-      assert(0);  // Invalid token.
-      return -1;
-  }
-}
-
-static int decode_coefs(MACROBLOCKD *xd, PLANE_TYPE type, tran_low_t *dqcoeff,
-                        TX_SIZE tx_size, TX_TYPE tx_type, const int16_t *dq,
-#if CONFIG_NEW_QUANT
-                        dequant_val_type_nuq *dq_val,
-#else
-#if CONFIG_AOM_QM
-                        qm_val_t *iqm[2][TX_SIZES_ALL],
-#endif  // CONFIG_AOM_QM
-#endif  // CONFIG_NEW_QUANT
-                        int ctx, const int16_t *scan, const int16_t *nb,
-                        int16_t *max_scan_line, aom_reader *r) {
-  FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
-  const int max_eob = tx_size_2d[tx_size];
-  const int ref = is_inter_block(&xd->mi[0]->mbmi);
-#if CONFIG_AOM_QM && !CONFIG_NEW_QUANT
-  const qm_val_t *iqmatrix = iqm[!ref][tx_size];
-#endif  // CONFIG_AOM_QM
-  (void)tx_type;
-  int band, c = 0;
-  const TX_SIZE tx_size_ctx = txsize_sqr_map[tx_size];
-  aom_cdf_prob(*coef_head_cdfs)[COEFF_CONTEXTS][CDF_SIZE(ENTROPY_TOKENS)] =
-      ec_ctx->coef_head_cdfs[tx_size_ctx][type][ref];
-  aom_cdf_prob(*coef_tail_cdfs)[COEFF_CONTEXTS][CDF_SIZE(ENTROPY_TOKENS)] =
-      ec_ctx->coef_tail_cdfs[tx_size_ctx][type][ref];
-  int val = 0;
-
-  uint8_t token_cache[MAX_TX_SQUARE];
-  const uint8_t *band_translate = get_band_translate(tx_size);
-  int dq_shift;
-  int v, token;
-  int32_t dqv = dq[0];
-#if CONFIG_NEW_QUANT
-  const tran_low_t *dqv_val = &dq_val[0][0];
-#endif  // CONFIG_NEW_QUANT
-
-  dq_shift = av1_get_tx_scale(tx_size);
-
-  band = *band_translate++;
-
-  int more_data = 1;
-  while (more_data) {
-    int comb_token;
-    int last_pos = (c + 1 == max_eob);
-    int first_pos = (c == 0);
-
-#if CONFIG_NEW_QUANT
-    dqv_val = &dq_val[band][0];
-#endif  // CONFIG_NEW_QUANT
-
-    comb_token = last_pos ? 2 * av1_read_record_bit(xd->counts, r, ACCT_STR) + 2
-                          : av1_read_record_symbol(
-                                xd->counts, r, coef_head_cdfs[band][ctx],
-                                HEAD_TOKENS + first_pos, ACCT_STR) +
-                                !first_pos;
-    if (first_pos) {
-      if (comb_token == 0) return 0;
-    }
-    token = comb_token >> 1;
-
-    while (!token) {
-      *max_scan_line = AOMMAX(*max_scan_line, scan[c]);
-      token_cache[scan[c]] = 0;
-#if CONFIG_SYMBOLRATE
-      av1_record_coeff(xd->counts, 0);
-#endif
-      ++c;
-      dqv = dq[1];
-      ctx = get_coef_context(nb, token_cache, c);
-      band = *band_translate++;
-
-      last_pos = (c + 1 == max_eob);
-
-      comb_token =
-          last_pos
-              ? 2 * av1_read_record_bit(xd->counts, r, ACCT_STR) + 2
-              : av1_read_record_symbol(xd->counts, r, coef_head_cdfs[band][ctx],
-                                       HEAD_TOKENS, ACCT_STR) +
-                    1;
-      token = comb_token >> 1;
-    }
-
-    more_data = comb_token & 1;
-
-    if (token > ONE_TOKEN)
-      token += av1_read_record_symbol(xd->counts, r, coef_tail_cdfs[band][ctx],
-                                      TAIL_TOKENS, ACCT_STR);
-#if CONFIG_NEW_QUANT
-    dqv_val = &dq_val[band][0];
-#endif  // CONFIG_NEW_QUANT
-
-    *max_scan_line = AOMMAX(*max_scan_line, scan[c]);
-    token_cache[scan[c]] = av1_pt_energy_class[token];
-
-    val = token_to_value(xd->counts, r, token, tx_size, xd->bd);
-#if CONFIG_SYMBOLRATE
-    av1_record_coeff(xd->counts, val);
-#endif
-
-#if CONFIG_NEW_QUANT
-    v = av1_dequant_abscoeff_nuq(val, dqv, dqv_val);
-    v = dq_shift ? ROUND_POWER_OF_TWO(v, dq_shift) : v;
-#else
-#if CONFIG_AOM_QM
-    // Apply quant matrix only for 2D transforms
-    if (IS_2D_TRANSFORM(tx_type) && iqmatrix != NULL)
-      dqv = ((iqmatrix[scan[c]] * (int)dqv) + (1 << (AOM_QM_BITS - 1))) >>
-            AOM_QM_BITS;
-#endif
-    v = (val * dqv) >> dq_shift;
-#endif
-
-    v = (int)check_range(av1_read_record_bit(xd->counts, r, ACCT_STR) ? -v : v,
-                         xd->bd);
-
-    dqcoeff[scan[c]] = v;
-
-    ++c;
-    more_data &= (c < max_eob);
-    if (!more_data) break;
-    dqv = dq[1];
-    ctx = get_coef_context(nb, token_cache, c);
-    band = *band_translate++;
-  }
-
-  return c;
-}
-#endif  // !CONFIG_PVQ
 
 static void decode_color_map_tokens(Av1ColorMapParam *param, aom_reader *r) {
   uint8_t color_order[PALETTE_MAX_SIZE];
@@ -263,7 +36,6 @@ static void decode_color_map_tokens(Av1ColorMapParam *param, aom_reader *r) {
   color_map[0] = av1_read_uniform(r, n);
   assert(color_map[0] < n);
 
-#if CONFIG_PALETTE_THROUGHPUT
   // Run wavefront on the palette map index decoding.
   for (int i = 1; i < rows + cols - 1; ++i) {
     for (int j = AOMMIN(i, cols - 1); j >= AOMMAX(0, i - rows + 1); --j) {
@@ -283,21 +55,6 @@ static void decode_color_map_tokens(Av1ColorMapParam *param, aom_reader *r) {
              (plane_block_width - cols));
     }
   }
-#else
-  for (int i = 0; i < rows; ++i) {
-    for (int j = (i == 0 ? 1 : 0); j < cols; ++j) {
-      const int color_ctx = av1_get_palette_color_index_context(
-          color_map, plane_block_width, i, j, n, color_order, NULL);
-      const int color_idx = aom_read_symbol(
-          r, color_map_cdf[n - PALETTE_MIN_SIZE][color_ctx], n, ACCT_STR);
-      assert(color_idx >= 0 && color_idx < n);
-      color_map[i * plane_block_width + j] = color_order[color_idx];
-    }
-    memset(color_map + i * plane_block_width + cols,
-           color_map[i * plane_block_width + cols - 1],
-           (plane_block_width - cols));  // Copy last column to extra columns.
-  }
-#endif  // CONFIG_PALETTE_THROUGHPUT
   // Copy last row to extra rows.
   for (int i = rows; i < plane_block_height; ++i) {
     memcpy(color_map + i * plane_block_width,
@@ -305,97 +62,17 @@ static void decode_color_map_tokens(Av1ColorMapParam *param, aom_reader *r) {
   }
 }
 
-static void get_palette_params(const MACROBLOCKD *const xd, int plane,
-                               BLOCK_SIZE bsize, Av1ColorMapParam *params) {
-  assert(plane == 0 || plane == 1);
-  const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
-  const PALETTE_MODE_INFO *const pmi = &mbmi->palette_mode_info;
-  params->color_map = xd->plane[plane].color_index_map;
-  params->map_cdf = plane ? xd->tile_ctx->palette_uv_color_index_cdf
-                          : xd->tile_ctx->palette_y_color_index_cdf;
-  params->n_colors = pmi->palette_size[plane];
-  av1_get_block_dimensions(bsize, plane, xd, &params->plane_width,
-                           &params->plane_height, &params->rows, &params->cols);
-}
-
-#if CONFIG_MRC_TX && SIGNAL_ANY_MRC_MASK
-static void get_mrc_params(const MACROBLOCKD *const xd, TX_SIZE tx_size,
-                           Av1ColorMapParam *params) {
-  memset(params, 0, sizeof(*params));
-  const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
-  const int is_inter = is_inter_block(mbmi);
-  params->color_map = xd->mrc_mask;
-  params->map_cdf = is_inter ? xd->tile_ctx->mrc_mask_inter_cdf
-                             : xd->tile_ctx->mrc_mask_intra_cdf;
-  params->n_colors = 2;
-  params->plane_width = tx_size_wide[tx_size];
-  params->rows = tx_size_high[tx_size];
-  params->cols = tx_size_wide[tx_size];
-}
-#endif  // CONFIG_MRC_TX && SIGNAL_ANY_MRC_MASK
-
 void av1_decode_palette_tokens(MACROBLOCKD *const xd, int plane,
                                aom_reader *r) {
-  const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
   assert(plane == 0 || plane == 1);
-  assert(mbmi->sb_type >= BLOCK_8X8);
-  Av1ColorMapParam color_map_params;
-  memset(&color_map_params, 0, sizeof(color_map_params));
-  get_palette_params(xd, plane, mbmi->sb_type, &color_map_params);
-  decode_color_map_tokens(&color_map_params, r);
-}
-
-#if CONFIG_MRC_TX && SIGNAL_ANY_MRC_MASK
-static void decode_mrc_tokens(MACROBLOCKD *const xd, TX_TYPE tx_size,
-                              aom_reader *r) {
-  const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
-  const int is_inter = is_inter_block(mbmi);
-  if ((is_inter && !SIGNAL_MRC_MASK_INTER) ||
-      (!is_inter && !SIGNAL_MRC_MASK_INTRA))
-    return;
-  Av1ColorMapParam color_map_params;
-  get_mrc_params(xd, tx_size, &color_map_params);
-  decode_color_map_tokens(&color_map_params, r);
-}
-#endif  // CONFIG_MRC_TX && SIGNAL_ANY_MRC_MASK
-
-#if !CONFIG_PVQ || CONFIG_VAR_TX
-int av1_decode_block_tokens(AV1_COMMON *cm, MACROBLOCKD *const xd, int plane,
-                            const SCAN_ORDER *sc, int x, int y, TX_SIZE tx_size,
-                            TX_TYPE tx_type, int16_t *max_scan_line,
-                            aom_reader *r, int seg_id) {
-  struct macroblockd_plane *const pd = &xd->plane[plane];
-  const int16_t *const dequant = pd->seg_dequant[seg_id];
-  const int ctx =
-      get_entropy_context(tx_size, pd->above_context + x, pd->left_context + y);
-#if CONFIG_NEW_QUANT
-  const int ref = is_inter_block(&xd->mi[0]->mbmi);
-  int dq =
-      get_dq_profile_from_ctx(xd->qindex[seg_id], ctx, ref, pd->plane_type);
-#endif  //  CONFIG_NEW_QUANT
-
-#if CONFIG_MRC_TX && SIGNAL_ANY_MRC_MASK
-  if (tx_type == MRC_DCT) decode_mrc_tokens(xd, tx_size, r);
-#endif  // CONFIG_MRC_TX && SIGNAL_ANY_MRC_MASK
-
-  const int eob =
-      decode_coefs(xd, pd->plane_type, pd->dqcoeff, tx_size, tx_type, dequant,
-#if CONFIG_NEW_QUANT
-                   pd->seg_dequant_nuq[seg_id][dq],
-#else
-#if CONFIG_AOM_QM
-                   pd->seg_iqmatrix[seg_id],
-#endif  // CONFIG_AOM_QM
-#endif  // CONFIG_NEW_QUANT
-                   ctx, sc->scan, sc->neighbors, max_scan_line, r);
-  av1_set_contexts(xd, pd, plane, tx_size, eob > 0, x, y);
-#if CONFIG_ADAPT_SCAN
-  if (xd->counts)
-    av1_update_scan_count_facade(cm, xd->counts, tx_size, tx_type, pd->dqcoeff,
-                                 eob);
-#else
-  (void)cm;
-#endif
-  return eob;
+  Av1ColorMapParam params;
+  params.color_map =
+      xd->plane[plane].color_index_map + xd->color_index_map_offset[plane];
+  params.map_cdf = plane ? xd->tile_ctx->palette_uv_color_index_cdf
+                         : xd->tile_ctx->palette_y_color_index_cdf;
+  const MB_MODE_INFO *const mbmi = xd->mi[0];
+  params.n_colors = mbmi->palette_mode_info.palette_size[plane];
+  av1_get_block_dimensions(mbmi->sb_type, plane, xd, &params.plane_width,
+                           &params.plane_height, &params.rows, &params.cols);
+  decode_color_map_tokens(&params, r);
 }
-#endif  // !CONFIG_PVQ
diff --git a/third_party/aom/av1/decoder/detokenize.h b/third_party/aom/av1/decoder/detokenize.h
index eb31d58c6..ec85bf7ea 100644
--- a/third_party/aom/av1/decoder/detokenize.h
+++ b/third_party/aom/av1/decoder/detokenize.h
@@ -12,10 +12,9 @@
 #ifndef AV1_DECODER_DETOKENIZE_H_
 #define AV1_DECODER_DETOKENIZE_H_
 
-#include "./aom_config.h"
-#if !CONFIG_PVQ || CONFIG_VAR_TX
+#include "config/aom_config.h"
+
 #include "av1/common/scan.h"
-#endif  // !CONFIG_PVQ || CONFIG_VAR_TX
 #include "av1/decoder/decoder.h"
 
 #ifdef __cplusplus
@@ -24,12 +23,6 @@ extern "C" {
 
 void av1_decode_palette_tokens(MACROBLOCKD *const xd, int plane, aom_reader *r);
 
-#if !CONFIG_PVQ || CONFIG_VAR_TX
-int av1_decode_block_tokens(AV1_COMMON *cm, MACROBLOCKD *const xd, int plane,
-                            const SCAN_ORDER *sc, int x, int y, TX_SIZE tx_size,
-                            TX_TYPE tx_type, int16_t *max_scan_line,
-                            aom_reader *r, int seg_id);
-#endif  // !CONFIG_PVQ
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/third_party/aom/av1/decoder/dsubexp.c b/third_party/aom/av1/decoder/dsubexp.c
deleted file mode 100644
index 5171f1144..000000000
--- a/third_party/aom/av1/decoder/dsubexp.c
+++ /dev/null
@@ -1,82 +0,0 @@
-/*
- * Copyright (c) 2016, Alliance for Open Media. All rights reserved
- *
- * This source code is subject to the terms of the BSD 2 Clause License and
- * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
- * was not distributed with this source code in the LICENSE file, you can
- * obtain it at www.aomedia.org/license/software. If the Alliance for Open
- * Media Patent License 1.0 was not distributed with this source code in the
- * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
- */
-
-#include <assert.h>
-
-#include "av1/common/entropy.h"
-
-#include "av1/decoder/dsubexp.h"
-
-static int inv_recenter_nonneg(int v, int m) {
-  if (v > 2 * m) return v;
-
-  return (v & 1) ? m - ((v + 1) >> 1) : m + (v >> 1);
-}
-
-#define decode_uniform(r, ACCT_STR_NAME) \
-  decode_uniform_(r ACCT_STR_ARG(ACCT_STR_NAME))
-#define decode_term_subexp(r, ACCT_STR_NAME) \
-  decode_term_subexp_(r ACCT_STR_ARG(ACCT_STR_NAME))
-
-static int decode_uniform_(aom_reader *r ACCT_STR_PARAM) {
-  const int l = 8;
-  const int m = (1 << l) - 190;
-  const int v = aom_read_literal(r, l - 1, ACCT_STR_NAME);
-  return v < m ? v : (v << 1) - m + aom_read_bit(r, ACCT_STR_NAME);
-}
-
-static int inv_remap_prob(int v, int m) {
-  /* clang-format off */
-  static uint8_t inv_map_table[MAX_PROB - 1] = {
-      7,  20,  33,  46,  59,  72,  85,  98, 111, 124, 137, 150, 163, 176, 189,
-    202, 215, 228, 241, 254,   1,   2,   3,   4,   5,   6,   8,   9,  10,  11,
-     12,  13,  14,  15,  16,  17,  18,  19,  21,  22,  23,  24,  25,  26,  27,
-     28,  29,  30,  31,  32,  34,  35,  36,  37,  38,  39,  40,  41,  42,  43,
-     44,  45,  47,  48,  49,  50,  51,  52,  53,  54,  55,  56,  57,  58,  60,
-     61,  62,  63,  64,  65,  66,  67,  68,  69,  70,  71,  73,  74,  75,  76,
-     77,  78,  79,  80,  81,  82,  83,  84,  86,  87,  88,  89,  90,  91,  92,
-     93,  94,  95,  96,  97,  99, 100, 101, 102, 103, 104, 105, 106, 107, 108,
-    109, 110, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 125,
-    126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 138, 139, 140, 141,
-    142, 143, 144, 145, 146, 147, 148, 149, 151, 152, 153, 154, 155, 156, 157,
-    158, 159, 160, 161, 162, 164, 165, 166, 167, 168, 169, 170, 171, 172, 173,
-    174, 175, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 188, 190,
-    191, 192, 193, 194, 195, 196, 197, 198, 199, 200, 201, 203, 204, 205, 206,
-    207, 208, 209, 210, 211, 212, 213, 214, 216, 217, 218, 219, 220, 221, 222,
-    223, 224, 225, 226, 227, 229, 230, 231, 232, 233, 234, 235, 236, 237, 238,
-    239, 240, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252, 253
-  }; /* clang-format on */
-  assert(v < (int)(sizeof(inv_map_table) / sizeof(inv_map_table[0])));
-  v = inv_map_table[v];
-  m--;
-  if ((m << 1) <= MAX_PROB) {
-    return 1 + inv_recenter_nonneg(v, m);
-  } else {
-    return MAX_PROB - inv_recenter_nonneg(v, MAX_PROB - 1 - m);
-  }
-}
-
-static int decode_term_subexp_(aom_reader *r ACCT_STR_PARAM) {
-  if (!aom_read_bit(r, ACCT_STR_NAME))
-    return aom_read_literal(r, 4, ACCT_STR_NAME);
-  if (!aom_read_bit(r, ACCT_STR_NAME))
-    return aom_read_literal(r, 4, ACCT_STR_NAME) + 16;
-  if (!aom_read_bit(r, ACCT_STR_NAME))
-    return aom_read_literal(r, 5, ACCT_STR_NAME) + 32;
-  return decode_uniform(r, ACCT_STR_NAME) + 64;
-}
-
-void av1_diff_update_prob_(aom_reader *r, aom_prob *p ACCT_STR_PARAM) {
-  if (aom_read(r, DIFF_UPDATE_PROB, ACCT_STR_NAME)) {
-    const int delp = decode_term_subexp(r, ACCT_STR_NAME);
-    *p = (aom_prob)inv_remap_prob(delp, *p);
-  }
-}
diff --git a/third_party/aom/av1/decoder/dsubexp.h b/third_party/aom/av1/decoder/dsubexp.h
deleted file mode 100644
index 4bc38578c..000000000
--- a/third_party/aom/av1/decoder/dsubexp.h
+++ /dev/null
@@ -1,32 +0,0 @@
-/*
- * Copyright (c) 2016, Alliance for Open Media. All rights reserved
- *
- * This source code is subject to the terms of the BSD 2 Clause License and
- * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
- * was not distributed with this source code in the LICENSE file, you can
- * obtain it at www.aomedia.org/license/software. If the Alliance for Open
- * Media Patent License 1.0 was not distributed with this source code in the
- * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
- */
-
-#ifndef AV1_DECODER_DSUBEXP_H_
-#define AV1_DECODER_DSUBEXP_H_
-
-#include "aom_dsp/bitreader.h"
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-#if CONFIG_ACCOUNTING
-#define av1_diff_update_prob(r, p, str) av1_diff_update_prob_(r, p, str)
-#else
-#define av1_diff_update_prob(r, p, str) av1_diff_update_prob_(r, p)
-#endif
-
-void av1_diff_update_prob_(aom_reader *r, aom_prob *p ACCT_STR_PARAM);
-
-#ifdef __cplusplus
-}  // extern "C"
-#endif
-#endif  // AV1_DECODER_DSUBEXP_H_
diff --git a/third_party/aom/av1/decoder/dthread.c b/third_party/aom/av1/decoder/dthread.c
index 7f16b233c..ff03502e6 100644
--- a/third_party/aom/av1/decoder/dthread.c
+++ b/third_party/aom/av1/decoder/dthread.c
@@ -9,7 +9,8 @@
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */
 
-#include "./aom_config.h"
+#include "config/aom_config.h"
+
 #include "aom_mem/aom_mem.h"
 #include "av1/common/reconinter.h"
 #include "av1/decoder/dthread.h"
@@ -157,12 +158,8 @@ void av1_frameworker_copy_context(AVxWorker *const dst_worker,
   av1_frameworker_unlock_stats(src_worker);
 
   dst_cm->bit_depth = src_cm->bit_depth;
-#if CONFIG_HIGHBITDEPTH
   dst_cm->use_highbitdepth = src_cm->use_highbitdepth;
-#endif
-#if CONFIG_EXT_REFS
-// TODO(zoeliu): To handle parallel decoding
-#endif  // CONFIG_EXT_REFS
+  // TODO(zoeliu): To handle parallel decoding
   dst_cm->prev_frame =
       src_cm->show_existing_frame ? src_cm->prev_frame : src_cm->cur_frame;
   dst_cm->last_width =
@@ -180,14 +177,10 @@ void av1_frameworker_copy_context(AVxWorker *const dst_worker,
 
   memcpy(dst_cm->lf_info.lfthr, src_cm->lf_info.lfthr,
          (MAX_LOOP_FILTER + 1) * sizeof(loop_filter_thresh));
-  dst_cm->lf.last_sharpness_level = src_cm->lf.sharpness_level;
-#if CONFIG_LOOPFILTER_LEVEL
+  dst_cm->lf.sharpness_level = src_cm->lf.sharpness_level;
   dst_cm->lf.filter_level[0] = src_cm->lf.filter_level[0];
   dst_cm->lf.filter_level[1] = src_cm->lf.filter_level[1];
-#else
-  dst_cm->lf.filter_level = src_cm->lf.filter_level;
-#endif
-  memcpy(dst_cm->lf.ref_deltas, src_cm->lf.ref_deltas, TOTAL_REFS_PER_FRAME);
+  memcpy(dst_cm->lf.ref_deltas, src_cm->lf.ref_deltas, REF_FRAMES);
   memcpy(dst_cm->lf.mode_deltas, src_cm->lf.mode_deltas, MAX_MODE_LF_DELTAS);
   dst_cm->seg = src_cm->seg;
   memcpy(dst_cm->frame_contexts, src_cm->frame_contexts,
diff --git a/third_party/aom/av1/decoder/dthread.h b/third_party/aom/av1/decoder/dthread.h
index c17053d9c..33d89006e 100644
--- a/third_party/aom/av1/decoder/dthread.h
+++ b/third_party/aom/av1/decoder/dthread.h
@@ -12,7 +12,8 @@
 #ifndef AV1_DECODER_DTHREAD_H_
 #define AV1_DECODER_DTHREAD_H_
 
-#include "./aom_config.h"
+#include "config/aom_config.h"
+
 #include "aom_util/aom_thread.h"
 #include "aom/internal/aom_codec_internal.h"
 
@@ -22,6 +23,13 @@ extern "C" {
 
 struct AV1Common;
 struct AV1Decoder;
+struct ThreadData;
+
+typedef struct DecWorkerData {
+  struct ThreadData *td;
+  const uint8_t *data_end;
+  struct aom_internal_error_info error_info;
+} DecWorkerData;
 
 // WorkerData for the FrameWorker thread. It contains all the information of
 // the worker and decode structures for decoding a frame.
diff --git a/third_party/aom/av1/decoder/generic_decoder.c b/third_party/aom/av1/decoder/generic_decoder.c
deleted file mode 100644
index 0c7d71b9f..000000000
--- a/third_party/aom/av1/decoder/generic_decoder.c
+++ /dev/null
@@ -1,110 +0,0 @@
-/*
- * Copyright (c) 2001-2016, Alliance for Open Media. All rights reserved
- *
- * This source code is subject to the terms of the BSD 2 Clause License and
- * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
- * was not distributed with this source code in the LICENSE file, you can
- * obtain it at www.aomedia.org/license/software. If the Alliance for Open
- * Media Patent License 1.0 was not distributed with this source code in the
- * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
- */
-
-/* clang-format off */
-
-#ifdef HAVE_CONFIG_H
-# include "config.h"
-#endif
-
-#include <stdio.h>
-
-#include "aom_dsp/bitreader.h"
-#include "av1/common/generic_code.h"
-#include "av1/common/odintrin.h"
-#include "pvq_decoder.h"
-
-/** Decodes a value from 0 to N-1 (with N up to 16) based on a cdf and adapts
- * the cdf accordingly.
- *
- * @param [in,out] r     multi-symbol entropy decoder
- * @param [in,out] cdf   CDF of the variable (Q15)
- * @param [in]     n     number of values possible
- * @param [in,out] count number of symbols encoded with that cdf so far
- * @param [in]     rate  adaptation rate shift (smaller is faster)
- * @return decoded variable
- */
-int aom_decode_cdf_adapt_q15_(aom_reader *r, uint16_t *cdf, int n,
- int *count, int rate ACCT_STR_PARAM) {
-  int val;
-  int i;
-  if (*count == 0) {
-    int ft;
-    ft = cdf[n - 1];
-    for (i = 0; i < n; i++) {
-      cdf[i] = AOM_ICDF(cdf[i]*32768/ft);
-    }
-  }
-  val = aom_read_cdf(r, cdf, n, ACCT_STR_NAME);
-  aom_cdf_adapt_q15(val, cdf, n, count, rate);
-  return val;
-}
-
-/** Encodes a random variable using a "generic" model, assuming that the
- * distribution is one-sided (zero and up), has a single mode, and decays
- * exponentially past the model.
- *
- * @param [in,out] r     multi-symbol entropy decoder
- * @param [in,out] model generic probability model
- * @param [in]     x     variable being encoded
- * @param [in,out] ExQ16 expectation of x (adapted)
- * @param [in]     integration integration period of ExQ16 (leaky average over
- * 1<<integration samples)
- *
- * @retval decoded variable x
- */
-int generic_decode_(aom_reader *r, generic_encoder *model,
- int *ex_q16, int integration ACCT_STR_PARAM) {
-  int lg_q1;
-  int shift;
-  int id;
-  uint16_t *cdf;
-  int xs;
-  int lsb;
-  int x;
-  lsb = 0;
-  lg_q1 = log_ex(*ex_q16);
-  /* If expectation is too large, shift x to ensure that
-     all we have past xs=15 is the exponentially decaying tail
-     of the distribution. */
-  shift = OD_MAXI(0, (lg_q1 - 5) >> 1);
-  /* Choose the cdf to use: we have two per "octave" of ExQ16. */
-  id = OD_MINI(GENERIC_TABLES - 1, lg_q1);
-  cdf = model->cdf[id];
-  xs = aom_read_symbol_pvq(r, cdf, 16, ACCT_STR_NAME);
-  if (xs == 15) {
-    int e;
-    unsigned decay;
-    /* Estimate decay based on the assumption that the distribution is close
-       to Laplacian for large values. We should probably have an adaptive
-       estimate instead. Note: The 2* is a kludge that's not fully understood
-       yet. */
-    OD_ASSERT(*ex_q16 < INT_MAX >> 1);
-    e = ((2**ex_q16 >> 8) + (1 << shift >> 1)) >> shift;
-    decay = OD_MAXI(2, OD_MINI(254, 256*e/(e + 256)));
-    xs += aom_laplace_decode_special(r, decay, ACCT_STR_NAME);
-  }
-  if (shift != 0) {
-    int special;
-    /* Because of the rounding, there's only half the number of possibilities
-       for xs=0 */
-    special = xs == 0;
-    if (shift - special > 0) {
-      lsb = aom_read_literal(r, shift - special, ACCT_STR_NAME);
-    }
-    lsb -= !special << (shift - 1);
-  }
-  x = (xs << shift) + lsb;
-  generic_model_update(ex_q16, x, integration);
-  OD_LOG((OD_LOG_ENTROPY_CODER, OD_LOG_DEBUG,
-   "dec: %d %d %d %d %d %x", *ex_q16, x, shift, id, xs, dec->rng));
-  return x;
-}
diff --git a/third_party/aom/av1/decoder/inspection.c b/third_party/aom/av1/decoder/inspection.c
index 98c51d4ba..e6c89298a 100644
--- a/third_party/aom/av1/decoder/inspection.c
+++ b/third_party/aom/av1/decoder/inspection.c
@@ -11,12 +11,7 @@
 #include "av1/decoder/decoder.h"
 #include "av1/decoder/inspection.h"
 #include "av1/common/enums.h"
-#if CONFIG_CDEF
 #include "av1/common/cdef.h"
-#endif
-#if CONFIG_CFL
-#include "av1/common/cfl.h"
-#endif
 
 static void ifd_init_mi_rc(insp_frame_data *fd, int mi_cols, int mi_rows) {
   fd->mi_cols = mi_cols;
@@ -48,25 +43,29 @@ int ifd_inspect(insp_frame_data *fd, void *decoder) {
   fd->show_frame = cm->show_frame;
   fd->frame_type = cm->frame_type;
   fd->base_qindex = cm->base_qindex;
-  fd->tile_mi_cols = cm->tile_width;
-  fd->tile_mi_rows = cm->tile_height;
+  // Set width and height of the first tile until generic support can be added
+  TileInfo tile_info;
+  av1_tile_set_row(&tile_info, cm, 0);
+  av1_tile_set_col(&tile_info, cm, 0);
+  fd->tile_mi_cols = tile_info.mi_col_end - tile_info.mi_col_start;
+  fd->tile_mi_rows = tile_info.mi_row_end - tile_info.mi_row_start;
+  fd->delta_q_present_flag = cm->delta_q_present_flag;
+  fd->delta_q_res = cm->delta_q_res;
 #if CONFIG_ACCOUNTING
   fd->accounting = &pbi->accounting;
 #endif
-#if CONFIG_CDEF
-// TODO(negge): copy per frame CDEF data
-#endif
+  // TODO(negge): copy per frame CDEF data
   int i, j;
   for (i = 0; i < MAX_SEGMENTS; i++) {
     for (j = 0; j < 2; j++) {
-      fd->y_dequant[i][j] = cm->y_dequant[i][j];
-      fd->uv_dequant[i][j] = cm->uv_dequant[i][j];
+      fd->y_dequant[i][j] = cm->y_dequant_QTX[i][j];
+      fd->u_dequant[i][j] = cm->u_dequant_QTX[i][j];
+      fd->v_dequant[i][j] = cm->v_dequant_QTX[i][j];
     }
   }
   for (j = 0; j < cm->mi_rows; j++) {
     for (i = 0; i < cm->mi_cols; i++) {
-      const MB_MODE_INFO *mbmi =
-          &cm->mi_grid_visible[j * cm->mi_stride + i]->mbmi;
+      const MB_MODE_INFO *mbmi = cm->mi_grid_visible[j * cm->mi_stride + i];
       insp_mi_data *mi = &fd->mi_grid[j * cm->mi_cols + i];
       // Segment
       mi->segment_id = mbmi->segment_id;
@@ -90,24 +89,19 @@ int ifd_inspect(insp_frame_data *fd, void *decoder) {
       mi->sb_type = mbmi->sb_type;
       // Skip Flag
       mi->skip = mbmi->skip;
-#if CONFIG_DUAL_FILTER
       mi->filter[0] = av1_extract_interp_filter(mbmi->interp_filters, 0);
       mi->filter[1] = av1_extract_interp_filter(mbmi->interp_filters, 1);
-#else
-      mi->filter = av1_extract_interp_filter(mbmi->interp_filters, 0);
-#endif
+      mi->dual_filter_type = mi->filter[0] * 3 + mi->filter[1];
       // Transform
-      mi->tx_type = mbmi->tx_type;
+      // TODO(anyone): extract tx type info from mbmi->txk_type[].
+      mi->tx_type = DCT_DCT;
       mi->tx_size = mbmi->tx_size;
 
-#if CONFIG_CDEF
       mi->cdef_level =
           cm->cdef_strengths[mbmi->cdef_strength] / CDEF_SEC_STRENGTHS;
       mi->cdef_strength =
           cm->cdef_strengths[mbmi->cdef_strength] % CDEF_SEC_STRENGTHS;
       mi->cdef_strength += mi->cdef_strength == 3;
-#endif
-#if CONFIG_CFL
       if (mbmi->uv_mode == UV_CFL_PRED) {
         mi->cfl_alpha_idx = mbmi->cfl_alpha_idx;
         mi->cfl_alpha_sign = mbmi->cfl_alpha_signs;
@@ -115,7 +109,8 @@ int ifd_inspect(insp_frame_data *fd, void *decoder) {
         mi->cfl_alpha_idx = 0;
         mi->cfl_alpha_sign = 0;
       }
-#endif
+      // delta_q
+      mi->current_qindex = mbmi->current_qindex;
     }
   }
   return 1;
diff --git a/third_party/aom/av1/decoder/inspection.h b/third_party/aom/av1/decoder/inspection.h
index 06a94b737..bb604f684 100644
--- a/third_party/aom/av1/decoder/inspection.h
+++ b/third_party/aom/av1/decoder/inspection.h
@@ -20,7 +20,9 @@ extern "C" {
 #include "av1/decoder/accounting.h"
 #endif
 
+#ifndef AOM_AOMDX_H_
 typedef void (*aom_inspect_cb)(void *decoder, void *data);
+#endif
 
 typedef struct insp_mv insp_mv;
 
@@ -33,27 +35,21 @@ typedef struct insp_mi_data insp_mi_data;
 
 struct insp_mi_data {
   insp_mv mv[2];
-  int8_t ref_frame[2];
-  int8_t mode;
-  int8_t uv_mode;
-  int8_t sb_type;
-  int8_t skip;
-  int8_t segment_id;
-#if CONFIG_DUAL_FILTER
-  int8_t filter[2];
-#else
-  int8_t filter;
-#endif
-  int8_t tx_type;
-  int8_t tx_size;
-#if CONFIG_CDEF
-  int8_t cdef_level;
-  int8_t cdef_strength;
-#endif
-#if CONFIG_CFL
-  int8_t cfl_alpha_idx;
-  int8_t cfl_alpha_sign;
-#endif
+  int16_t ref_frame[2];
+  int16_t mode;
+  int16_t uv_mode;
+  int16_t sb_type;
+  int16_t skip;
+  int16_t segment_id;
+  int16_t dual_filter_type;
+  int16_t filter[2];
+  int16_t tx_type;
+  int16_t tx_size;
+  int16_t cdef_level;
+  int16_t cdef_strength;
+  int16_t cfl_alpha_idx;
+  int16_t cfl_alpha_sign;
+  int16_t current_qindex;
 };
 
 typedef struct insp_frame_data insp_frame_data;
@@ -71,10 +67,11 @@ struct insp_frame_data {
   int tile_mi_rows;
   int tile_mi_cols;
   int16_t y_dequant[MAX_SEGMENTS][2];
-  int16_t uv_dequant[MAX_SEGMENTS][2];
-#if CONFIG_CDEF
-// TODO(negge): add per frame CDEF data
-#endif
+  int16_t u_dequant[MAX_SEGMENTS][2];
+  int16_t v_dequant[MAX_SEGMENTS][2];
+  // TODO(negge): add per frame CDEF data
+  int delta_q_present_flag;
+  int delta_q_res;
 };
 
 void ifd_init(insp_frame_data *fd, int frame_width, int frame_height);
diff --git a/third_party/aom/av1/decoder/laplace_decoder.c b/third_party/aom/av1/decoder/laplace_decoder.c
deleted file mode 100644
index 5cc080ea7..000000000
--- a/third_party/aom/av1/decoder/laplace_decoder.c
+++ /dev/null
@@ -1,121 +0,0 @@
-/*
- * Copyright (c) 2001-2016, Alliance for Open Media. All rights reserved
- *
- * This source code is subject to the terms of the BSD 2 Clause License and
- * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
- * was not distributed with this source code in the LICENSE file, you can
- * obtain it at www.aomedia.org/license/software. If the Alliance for Open
- * Media Patent License 1.0 was not distributed with this source code in the
- * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
- */
-/* clang-format off */
-
-#ifdef HAVE_CONFIG_H
-# include "config.h"
-#endif
-
-#include <stdio.h>
-
-#include "aom_dsp/bitreader.h"
-#include "av1/common/pvq.h"
-#include "pvq_decoder.h"
-
-#define aom_decode_pvq_split(r, adapt, sum, ctx, ACCT_STR_NAME) \
-  aom_decode_pvq_split_(r, adapt, sum, ctx ACCT_STR_ARG(ACCT_STR_NAME))
-
-static int aom_decode_pvq_split_(aom_reader *r, od_pvq_codeword_ctx *adapt,
- int sum, int ctx ACCT_STR_PARAM) {
-  int shift;
-  int count;
-  int msbs;
-  int fctx;
-  count = 0;
-  if (sum == 0) return 0;
-  shift = OD_MAXI(0, OD_ILOG(sum) - 3);
-  fctx = 7*ctx + (sum >> shift) - 1;
-  msbs = aom_read_symbol_pvq(r, adapt->pvq_split_cdf[fctx], (sum >> shift) + 1,
-      ACCT_STR_NAME);
-  if (shift) count = aom_read_literal(r, shift, ACCT_STR_NAME);
-  count += msbs << shift;
-  if (count > sum) {
-    count = sum;
-#if !CONFIG_ANS
-    r->ec.error = 1;
-#else
-# error "CONFIG_PVQ currently requires !CONFIG_ANS."
-#endif
-  }
-  return count;
-}
-
-void aom_decode_band_pvq_splits(aom_reader *r, od_pvq_codeword_ctx *adapt,
- od_coeff *y, int n, int k, int level) {
-  int mid;
-  int count_right;
-  if (n == 1) {
-    y[0] = k;
-  }
-  else if (k == 0) {
-    OD_CLEAR(y, n);
-  }
-  else if (k == 1 && n <= 16) {
-    int cdf_id;
-    int pos;
-    cdf_id = od_pvq_k1_ctx(n, level == 0);
-    OD_CLEAR(y, n);
-    pos = aom_read_symbol_pvq(r, adapt->pvq_k1_cdf[cdf_id], n, "pvq:k1");
-    y[pos] = 1;
-  }
-  else {
-    mid = n >> 1;
-    count_right = aom_decode_pvq_split(r, adapt, k, od_pvq_size_ctx(n),
-     "pvq:split");
-    aom_decode_band_pvq_splits(r, adapt, y, mid, k - count_right, level + 1);
-    aom_decode_band_pvq_splits(r, adapt, y + mid, n - mid, count_right,
-     level + 1);
-  }
-}
-
-/** Decodes the tail of a Laplace-distributed variable, i.e. it doesn't
- * do anything special for the zero case.
- *
- * @param [dec] range decoder
- * @param [decay] decay factor of the distribution, i.e. pdf ~= decay^x
- *
- * @retval decoded variable x
- */
-int aom_laplace_decode_special_(aom_reader *r, unsigned decay ACCT_STR_PARAM) {
-  int pos;
-  int shift;
-  int xs;
-  int sym;
-  const uint16_t *cdf;
-  shift = 0;
-  /* We don't want a large decay value because that would require too many
-     symbols. */
-  while (decay > 235) {
-    decay = (decay*decay + 128) >> 8;
-    shift++;
-  }
-  decay = OD_MINI(decay, 254);
-  decay = OD_MAXI(decay, 2);
-  cdf = EXP_CDF_TABLE[(decay + 1) >> 1];
-  OD_LOG((OD_LOG_PVQ, OD_LOG_DEBUG, "decay = %d\n", decay));
-  xs = 0;
-  do {
-    sym = OD_MINI(xs, 15);
-    {
-      int i;
-      OD_LOG((OD_LOG_PVQ, OD_LOG_DEBUG, "%d %d %d", xs, shift, sym));
-      for (i = 0; i < 16; i++) {
-        OD_LOG_PARTIAL((OD_LOG_PVQ, OD_LOG_DEBUG, "%d ", cdf[i]));
-      }
-      OD_LOG_PARTIAL((OD_LOG_PVQ, OD_LOG_DEBUG, "\n"));
-    }
-    sym = aom_read_cdf(r, cdf, 16, ACCT_STR_NAME);
-    xs += sym;
-  } while (sym >= 15);
-  if (shift) pos = (xs << shift) + aom_read_literal(r, shift, ACCT_STR_NAME);
-  else pos = xs;
-  return pos;
-}
diff --git a/third_party/aom/av1/decoder/obu.c b/third_party/aom/av1/decoder/obu.c
new file mode 100644
index 000000000..482b6415e
--- /dev/null
+++ b/third_party/aom/av1/decoder/obu.c
@@ -0,0 +1,907 @@
+/*
+ * Copyright (c) 2017, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <assert.h>
+
+#include "config/aom_config.h"
+
+#include "aom/aom_codec.h"
+#include "aom_dsp/bitreader_buffer.h"
+#include "aom_ports/mem_ops.h"
+
+#include "av1/common/common.h"
+#include "av1/common/timing.h"
+#include "av1/decoder/decoder.h"
+#include "av1/decoder/decodeframe.h"
+#include "av1/decoder/obu.h"
+
+// Picture prediction structures (0-12 are predefined) in scalability metadata.
+typedef enum {
+  SCALABILITY_L1T2 = 0,
+  SCALABILITY_L1T3 = 1,
+  SCALABILITY_L2T1 = 2,
+  SCALABILITY_L2T2 = 3,
+  SCALABILITY_L2T3 = 4,
+  SCALABILITY_S2T1 = 5,
+  SCALABILITY_S2T2 = 6,
+  SCALABILITY_S2T3 = 7,
+  SCALABILITY_L2T1h = 8,
+  SCALABILITY_L2T2h = 9,
+  SCALABILITY_L2T3h = 10,
+  SCALABILITY_S2T1h = 11,
+  SCALABILITY_S2T2h = 12,
+  SCALABILITY_S2T3h = 13,
+  SCALABILITY_SS = 14
+} SCALABILITY_STRUCTURES;
+
+// Returns 1 when OBU type is valid, and 0 otherwise.
+static int valid_obu_type(int obu_type) {
+  int valid_type = 0;
+  switch (obu_type) {
+    case OBU_SEQUENCE_HEADER:
+    case OBU_TEMPORAL_DELIMITER:
+    case OBU_FRAME_HEADER:
+    case OBU_TILE_GROUP:
+    case OBU_METADATA:
+    case OBU_FRAME:
+    case OBU_REDUNDANT_FRAME_HEADER:
+    case OBU_TILE_LIST:
+    case OBU_PADDING: valid_type = 1; break;
+    default: break;
+  }
+  return valid_type;
+}
+
+// Parses OBU header and stores values in 'header'.
+static aom_codec_err_t read_obu_header(struct aom_read_bit_buffer *rb,
+                                       int is_annexb, ObuHeader *header) {
+  if (!rb || !header) return AOM_CODEC_INVALID_PARAM;
+
+  const ptrdiff_t bit_buffer_byte_length = rb->bit_buffer_end - rb->bit_buffer;
+  if (bit_buffer_byte_length < 1) return AOM_CODEC_CORRUPT_FRAME;
+
+  header->size = 1;
+
+  if (aom_rb_read_bit(rb) != 0) {
+    // Forbidden bit. Must not be set.
+    return AOM_CODEC_CORRUPT_FRAME;
+  }
+
+  header->type = (OBU_TYPE)aom_rb_read_literal(rb, 4);
+
+  if (!valid_obu_type(header->type)) return AOM_CODEC_CORRUPT_FRAME;
+
+  header->has_extension = aom_rb_read_bit(rb);
+  header->has_size_field = aom_rb_read_bit(rb);
+
+  if (!header->has_size_field && !is_annexb) {
+    // section 5 obu streams must have obu_size field set.
+    return AOM_CODEC_UNSUP_BITSTREAM;
+  }
+
+  if (aom_rb_read_bit(rb) != 0) {
+    // obu_reserved_1bit must be set to 0.
+    return AOM_CODEC_CORRUPT_FRAME;
+  }
+
+  if (header->has_extension) {
+    if (bit_buffer_byte_length == 1) return AOM_CODEC_CORRUPT_FRAME;
+
+    header->size += 1;
+    header->temporal_layer_id = aom_rb_read_literal(rb, 3);
+    header->spatial_layer_id = aom_rb_read_literal(rb, 2);
+    if (aom_rb_read_literal(rb, 3) != 0) {
+      // extension_header_reserved_3bits must be set to 0.
+      return AOM_CODEC_CORRUPT_FRAME;
+    }
+  }
+
+  return AOM_CODEC_OK;
+}
+
+aom_codec_err_t aom_read_obu_header(uint8_t *buffer, size_t buffer_length,
+                                    size_t *consumed, ObuHeader *header,
+                                    int is_annexb) {
+  if (buffer_length < 1 || !consumed || !header) return AOM_CODEC_INVALID_PARAM;
+
+  // TODO(tomfinegan): Set the error handler here and throughout this file, and
+  // confirm parsing work done via aom_read_bit_buffer is successful.
+  struct aom_read_bit_buffer rb = { buffer, buffer + buffer_length, 0, NULL,
+                                    NULL };
+  aom_codec_err_t parse_result = read_obu_header(&rb, is_annexb, header);
+  if (parse_result == AOM_CODEC_OK) *consumed = header->size;
+  return parse_result;
+}
+
+aom_codec_err_t aom_get_num_layers_from_operating_point_idc(
+    int operating_point_idc, unsigned int *number_spatial_layers,
+    unsigned int *number_temporal_layers) {
+  // derive number of spatial/temporal layers from operating_point_idc
+
+  if (!number_spatial_layers || !number_temporal_layers)
+    return AOM_CODEC_INVALID_PARAM;
+
+  if (operating_point_idc == 0) {
+    *number_temporal_layers = 1;
+    *number_spatial_layers = 1;
+  } else {
+    *number_spatial_layers = 0;
+    *number_temporal_layers = 0;
+    for (int j = 0; j < MAX_NUM_SPATIAL_LAYERS; j++) {
+      *number_spatial_layers +=
+          (operating_point_idc >> (j + MAX_NUM_TEMPORAL_LAYERS)) & 0x1;
+    }
+    for (int j = 0; j < MAX_NUM_TEMPORAL_LAYERS; j++) {
+      *number_temporal_layers += (operating_point_idc >> j) & 0x1;
+    }
+  }
+
+  return AOM_CODEC_OK;
+}
+
+static int is_obu_in_current_operating_point(AV1Decoder *pbi,
+                                             ObuHeader obu_header) {
+  if (!pbi->current_operating_point) {
+    return 1;
+  }
+
+  if ((pbi->current_operating_point >> obu_header.temporal_layer_id) & 0x1 &&
+      (pbi->current_operating_point >> (obu_header.spatial_layer_id + 8)) &
+          0x1) {
+    return 1;
+  }
+  return 0;
+}
+
+static uint32_t read_temporal_delimiter_obu() { return 0; }
+
+// Returns a boolean that indicates success.
+static int read_bitstream_level(BitstreamLevel *bl,
+                                struct aom_read_bit_buffer *rb) {
+  const uint8_t seq_level_idx = aom_rb_read_literal(rb, LEVEL_BITS);
+  if (!is_valid_seq_level_idx(seq_level_idx)) return 0;
+  bl->major = (seq_level_idx >> LEVEL_MINOR_BITS) + LEVEL_MAJOR_MIN;
+  bl->minor = seq_level_idx & ((1 << LEVEL_MINOR_BITS) - 1);
+  return 1;
+}
+
+// On success, sets pbi->sequence_header_ready to 1 and returns the number of
+// bytes read from 'rb'.
+// On failure, sets pbi->common.error.error_code and returns 0.
+static uint32_t read_sequence_header_obu(AV1Decoder *pbi,
+                                         struct aom_read_bit_buffer *rb) {
+  AV1_COMMON *const cm = &pbi->common;
+  const uint32_t saved_bit_offset = rb->bit_offset;
+
+  // Verify rb has been configured to report errors.
+  assert(rb->error_handler);
+
+  cm->profile = av1_read_profile(rb);
+  if (cm->profile > PROFILE_2) {
+    cm->error.error_code = AOM_CODEC_UNSUP_BITSTREAM;
+    return 0;
+  }
+
+  SequenceHeader *const seq_params = &cm->seq_params;
+
+  // Still picture or not
+  seq_params->still_picture = aom_rb_read_bit(rb);
+  seq_params->reduced_still_picture_hdr = aom_rb_read_bit(rb);
+  // Video must have reduced_still_picture_hdr = 0
+  if (!seq_params->still_picture && seq_params->reduced_still_picture_hdr) {
+    cm->error.error_code = AOM_CODEC_UNSUP_BITSTREAM;
+    return 0;
+  }
+
+  if (seq_params->reduced_still_picture_hdr) {
+    cm->timing_info_present = 0;
+    seq_params->decoder_model_info_present_flag = 0;
+    seq_params->display_model_info_present_flag = 0;
+    seq_params->operating_points_cnt_minus_1 = 0;
+    seq_params->operating_point_idc[0] = 0;
+    if (!read_bitstream_level(&seq_params->level[0], rb)) {
+      cm->error.error_code = AOM_CODEC_UNSUP_BITSTREAM;
+      return 0;
+    }
+    seq_params->tier[0] = 0;
+    cm->op_params[0].decoder_model_param_present_flag = 0;
+    cm->op_params[0].display_model_param_present_flag = 0;
+  } else {
+    cm->timing_info_present = aom_rb_read_bit(rb);  // timing_info_present_flag
+    if (cm->timing_info_present) {
+      av1_read_timing_info_header(cm, rb);
+
+      seq_params->decoder_model_info_present_flag = aom_rb_read_bit(rb);
+      if (seq_params->decoder_model_info_present_flag)
+        av1_read_decoder_model_info(cm, rb);
+    } else {
+      seq_params->decoder_model_info_present_flag = 0;
+    }
+    seq_params->display_model_info_present_flag = aom_rb_read_bit(rb);
+    seq_params->operating_points_cnt_minus_1 =
+        aom_rb_read_literal(rb, OP_POINTS_CNT_MINUS_1_BITS);
+    for (int i = 0; i < seq_params->operating_points_cnt_minus_1 + 1; i++) {
+      seq_params->operating_point_idc[i] =
+          aom_rb_read_literal(rb, OP_POINTS_IDC_BITS);
+      if (!read_bitstream_level(&seq_params->level[i], rb)) {
+        cm->error.error_code = AOM_CODEC_UNSUP_BITSTREAM;
+        return 0;
+      }
+      // This is the seq_level_idx[i] > 7 check in the spec. seq_level_idx 7
+      // is equivalent to level 3.3.
+      if (seq_params->level[i].major > 3)
+        seq_params->tier[i] = aom_rb_read_bit(rb);
+      else
+        seq_params->tier[i] = 0;
+      if (seq_params->decoder_model_info_present_flag) {
+        cm->op_params[i].decoder_model_param_present_flag = aom_rb_read_bit(rb);
+        if (cm->op_params[i].decoder_model_param_present_flag)
+          av1_read_op_parameters_info(cm, rb, i);
+      } else {
+        cm->op_params[i].decoder_model_param_present_flag = 0;
+      }
+      if (cm->timing_info_present &&
+          (cm->timing_info.equal_picture_interval ||
+           cm->op_params[i].decoder_model_param_present_flag)) {
+        cm->op_params[i].bitrate = max_level_bitrate(
+            cm->profile, major_minor_to_seq_level_idx(seq_params->level[i]),
+            seq_params->tier[i]);
+        // Level with seq_level_idx = 31 returns a high "dummy" bitrate to pass
+        // the check
+        if (cm->op_params[i].bitrate == 0)
+          aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+                             "AV1 does not support this combination of "
+                             "profile, level, and tier.");
+        // Buffer size in bits/s is bitrate in bits/s * 1 s
+        cm->op_params[i].buffer_size = cm->op_params[i].bitrate;
+      }
+      if (cm->timing_info_present && cm->timing_info.equal_picture_interval &&
+          !cm->op_params[i].decoder_model_param_present_flag) {
+        // When the decoder_model_parameters are not sent for this op, set
+        // the default ones that can be used with the resource availability mode
+        cm->op_params[i].decoder_buffer_delay = 70000;
+        cm->op_params[i].encoder_buffer_delay = 20000;
+        cm->op_params[i].low_delay_mode_flag = 0;
+      }
+
+      if (seq_params->display_model_info_present_flag) {
+        cm->op_params[i].display_model_param_present_flag = aom_rb_read_bit(rb);
+        if (cm->op_params[i].display_model_param_present_flag) {
+          cm->op_params[i].initial_display_delay =
+              aom_rb_read_literal(rb, 4) + 1;
+          if (cm->op_params[i].initial_display_delay > 10)
+            aom_internal_error(
+                &cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+                "AV1 does not support more than 10 decoded frames delay");
+        } else {
+          cm->op_params[i].initial_display_delay = 10;
+        }
+      } else {
+        cm->op_params[i].display_model_param_present_flag = 0;
+        cm->op_params[i].initial_display_delay = 10;
+      }
+    }
+  }
+  // This decoder supports all levels.  Choose operating point provided by
+  // external means
+  int operating_point = pbi->operating_point;
+  if (operating_point < 0 ||
+      operating_point > seq_params->operating_points_cnt_minus_1)
+    operating_point = 0;
+  pbi->current_operating_point =
+      seq_params->operating_point_idc[operating_point];
+  if (aom_get_num_layers_from_operating_point_idc(
+          pbi->current_operating_point, &cm->number_spatial_layers,
+          &cm->number_temporal_layers) != AOM_CODEC_OK) {
+    cm->error.error_code = AOM_CODEC_ERROR;
+    return 0;
+  }
+
+  read_sequence_header(cm, rb);
+
+  av1_read_color_config(cm, rb, pbi->allow_lowbitdepth);
+
+  cm->film_grain_params_present = aom_rb_read_bit(rb);
+
+  if (av1_check_trailing_bits(pbi, rb) != 0) {
+    // cm->error.error_code is already set.
+    return 0;
+  }
+
+  pbi->sequence_header_ready = 1;
+
+  return ((rb->bit_offset - saved_bit_offset + 7) >> 3);
+}
+
+static uint32_t read_frame_header_obu(AV1Decoder *pbi,
+                                      struct aom_read_bit_buffer *rb,
+                                      const uint8_t *data,
+                                      const uint8_t **p_data_end,
+                                      int trailing_bits_present) {
+  av1_decode_frame_headers_and_setup(pbi, rb, data, p_data_end,
+                                     trailing_bits_present);
+  return (uint32_t)(pbi->uncomp_hdr_size);
+}
+
+static int32_t read_tile_group_header(AV1Decoder *pbi,
+                                      struct aom_read_bit_buffer *rb,
+                                      int *start_tile, int *end_tile,
+                                      int tile_start_implicit) {
+  AV1_COMMON *const cm = &pbi->common;
+  uint32_t saved_bit_offset = rb->bit_offset;
+  int tile_start_and_end_present_flag = 0;
+  const int num_tiles = pbi->common.tile_rows * pbi->common.tile_cols;
+
+  if (!pbi->common.large_scale_tile && num_tiles > 1) {
+    tile_start_and_end_present_flag = aom_rb_read_bit(rb);
+  }
+  if (pbi->common.large_scale_tile || num_tiles == 1 ||
+      !tile_start_and_end_present_flag) {
+    *start_tile = 0;
+    *end_tile = num_tiles - 1;
+    return ((rb->bit_offset - saved_bit_offset + 7) >> 3);
+  }
+  if (tile_start_implicit && tile_start_and_end_present_flag) {
+    aom_internal_error(
+        &cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+        "For OBU_FRAME type obu tile_start_and_end_present_flag must be 0");
+    cm->error.error_code = AOM_CODEC_CORRUPT_FRAME;
+    return -1;
+  }
+  *start_tile =
+      aom_rb_read_literal(rb, cm->log2_tile_rows + cm->log2_tile_cols);
+  *end_tile = aom_rb_read_literal(rb, cm->log2_tile_rows + cm->log2_tile_cols);
+
+  return ((rb->bit_offset - saved_bit_offset + 7) >> 3);
+}
+
+static uint32_t read_one_tile_group_obu(
+    AV1Decoder *pbi, struct aom_read_bit_buffer *rb, int is_first_tg,
+    const uint8_t *data, const uint8_t *data_end, const uint8_t **p_data_end,
+    int *is_last_tg, int tile_start_implicit) {
+  AV1_COMMON *const cm = &pbi->common;
+  int start_tile, end_tile;
+  int32_t header_size, tg_payload_size;
+
+  header_size = read_tile_group_header(pbi, rb, &start_tile, &end_tile,
+                                       tile_start_implicit);
+  if (header_size == -1) return 0;
+  if (start_tile > end_tile) return header_size;
+  data += header_size;
+  av1_decode_tg_tiles_and_wrapup(pbi, data, data_end, p_data_end, start_tile,
+                                 end_tile, is_first_tg);
+
+  tg_payload_size = (uint32_t)(*p_data_end - data);
+
+  // TODO(shan):  For now, assume all tile groups received in order
+  *is_last_tg = end_tile == cm->tile_rows * cm->tile_cols - 1;
+  return header_size + tg_payload_size;
+}
+
+// Only called while large_scale_tile = 1.
+static uint32_t read_and_decode_one_tile_list(AV1Decoder *pbi,
+                                              struct aom_read_bit_buffer *rb,
+                                              const uint8_t *data,
+                                              const uint8_t *data_end,
+                                              const uint8_t **p_data_end,
+                                              int *frame_decoding_finished) {
+  AV1_COMMON *const cm = &pbi->common;
+  uint32_t tile_list_payload_size = 0;
+  const int num_tiles = cm->tile_cols * cm->tile_rows;
+  const int start_tile = 0;
+  const int end_tile = num_tiles - 1;
+  int i = 0;
+
+  // Process the tile list info.
+  pbi->output_frame_width_in_tiles_minus_1 = aom_rb_read_literal(rb, 8);
+  pbi->output_frame_height_in_tiles_minus_1 = aom_rb_read_literal(rb, 8);
+  pbi->tile_count_minus_1 = aom_rb_read_literal(rb, 16);
+  if (pbi->tile_count_minus_1 > 511) {
+    cm->error.error_code = AOM_CODEC_CORRUPT_FRAME;
+    return 0;
+  }
+
+  // Allocate output frame buffer for the tile list.
+  // TODO(yunqing): for now, copy each tile's decoded YUV data directly to the
+  // output buffer. This needs to be modified according to the application
+  // requirement.
+  const int tile_width_in_pixels = cm->tile_width * MI_SIZE;
+  const int tile_height_in_pixels = cm->tile_height * MI_SIZE;
+  const int ssy = cm->subsampling_y;
+  const int ssx = cm->subsampling_x;
+  const int num_planes = av1_num_planes(cm);
+  const size_t yplane_tile_size = tile_height_in_pixels * tile_width_in_pixels;
+  const size_t uvplane_tile_size =
+      (num_planes > 1)
+          ? (tile_height_in_pixels >> ssy) * (tile_width_in_pixels >> ssx)
+          : 0;
+  const size_t tile_size = (cm->use_highbitdepth ? 2 : 1) *
+                           (yplane_tile_size + 2 * uvplane_tile_size);
+  pbi->tile_list_size = tile_size * (pbi->tile_count_minus_1 + 1);
+
+  if (pbi->tile_list_size > pbi->buffer_sz) {
+    if (pbi->tile_list_output != NULL) aom_free(pbi->tile_list_output);
+    pbi->tile_list_output = NULL;
+
+    pbi->tile_list_output = (uint8_t *)aom_memalign(32, pbi->tile_list_size);
+    if (pbi->tile_list_output == NULL)
+      aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
+                         "Failed to allocate the tile list output buffer");
+    pbi->buffer_sz = pbi->tile_list_size;
+  }
+
+  uint32_t tile_list_info_bytes = 4;
+  tile_list_payload_size += tile_list_info_bytes;
+  data += tile_list_info_bytes;
+  uint8_t *output = pbi->tile_list_output;
+
+  for (i = 0; i <= pbi->tile_count_minus_1; i++) {
+    // Process 1 tile.
+    // Reset the bit reader.
+    rb->bit_offset = 0;
+    rb->bit_buffer = data;
+
+    // Read out the tile info.
+    uint32_t tile_info_bytes = 5;
+    // Set reference for each tile.
+    int ref_idx = aom_rb_read_literal(rb, 8);
+    if (ref_idx >= MAX_EXTERNAL_REFERENCES) {
+      cm->error.error_code = AOM_CODEC_CORRUPT_FRAME;
+      return 0;
+    }
+    av1_set_reference_dec(cm, 0, 1, &pbi->ext_refs.refs[ref_idx]);
+
+    pbi->dec_tile_row = aom_rb_read_literal(rb, 8);
+    pbi->dec_tile_col = aom_rb_read_literal(rb, 8);
+    if (pbi->dec_tile_row < 0 || pbi->dec_tile_col < 0 ||
+        pbi->dec_tile_row >= cm->tile_rows ||
+        pbi->dec_tile_col >= cm->tile_cols) {
+      cm->error.error_code = AOM_CODEC_CORRUPT_FRAME;
+      return 0;
+    }
+
+    pbi->coded_tile_data_size = aom_rb_read_literal(rb, 16) + 1;
+    data += tile_info_bytes;
+    if ((size_t)(data_end - data) < pbi->coded_tile_data_size) {
+      cm->error.error_code = AOM_CODEC_CORRUPT_FRAME;
+      return 0;
+    }
+
+    av1_decode_tg_tiles_and_wrapup(pbi, data, data + pbi->coded_tile_data_size,
+                                   p_data_end, start_tile, end_tile, 0);
+    uint32_t tile_payload_size = (uint32_t)(*p_data_end - data);
+
+    tile_list_payload_size += tile_info_bytes + tile_payload_size;
+
+    // Update data ptr for next tile decoding.
+    data = *p_data_end;
+    assert(data <= data_end);
+
+    // Copy decoded tile to the tile list output buffer.
+    YV12_BUFFER_CONFIG *cur_frame = get_frame_new_buffer(cm);
+    const int mi_row = pbi->dec_tile_row * cm->tile_height;
+    const int mi_col = pbi->dec_tile_col * cm->tile_width;
+    const int is_hbd = (cur_frame->flags & YV12_FLAG_HIGHBITDEPTH) ? 1 : 0;
+    uint8_t *bufs[MAX_MB_PLANE] = { NULL, NULL, NULL };
+    int strides[MAX_MB_PLANE] = { 0, 0, 0 };
+    int plane;
+
+    for (plane = 0; plane < num_planes; ++plane) {
+      int shift_x = plane > 0 ? ssx : 0;
+      int shift_y = plane > 0 ? ssy : 0;
+
+      bufs[plane] = cur_frame->buffers[plane];
+      strides[plane] =
+          (plane > 0) ? cur_frame->strides[1] : cur_frame->strides[0];
+      if (is_hbd) {
+        bufs[plane] = (uint8_t *)CONVERT_TO_SHORTPTR(cur_frame->buffers[plane]);
+        strides[plane] =
+            (plane > 0) ? 2 * cur_frame->strides[1] : 2 * cur_frame->strides[0];
+      }
+
+      bufs[plane] += mi_row * (MI_SIZE >> shift_y) * strides[plane] +
+                     mi_col * (MI_SIZE >> shift_x);
+
+      int w, h;
+      w = (plane > 0 && shift_x > 0) ? ((tile_width_in_pixels + 1) >> shift_x)
+                                     : tile_width_in_pixels;
+      w *= (1 + is_hbd);
+      h = (plane > 0 && shift_y > 0) ? ((tile_height_in_pixels + 1) >> shift_y)
+                                     : tile_height_in_pixels;
+      int j;
+
+      for (j = 0; j < h; ++j) {
+        memcpy(output, bufs[plane], w);
+        bufs[plane] += strides[plane];
+        output += w;
+      }
+    }
+  }
+
+  *frame_decoding_finished = 1;
+  return tile_list_payload_size;
+}
+
+static void read_metadata_itut_t35(const uint8_t *data, size_t sz) {
+  struct aom_read_bit_buffer rb = { data, data + sz, 0, NULL, NULL };
+  for (size_t i = 0; i < sz; i++) {
+    aom_rb_read_literal(&rb, 8);
+  }
+}
+
+static void read_metadata_hdr_cll(const uint8_t *data, size_t sz) {
+  struct aom_read_bit_buffer rb = { data, data + sz, 0, NULL, NULL };
+  aom_rb_read_literal(&rb, 16);  // max_cll
+  aom_rb_read_literal(&rb, 16);  // max_fall
+}
+
+static void read_metadata_hdr_mdcv(const uint8_t *data, size_t sz) {
+  struct aom_read_bit_buffer rb = { data, data + sz, 0, NULL, NULL };
+  for (int i = 0; i < 3; i++) {
+    aom_rb_read_literal(&rb, 16);  // primary_i_chromaticity_x
+    aom_rb_read_literal(&rb, 16);  // primary_i_chromaticity_y
+  }
+
+  aom_rb_read_literal(&rb, 16);  // white_point_chromaticity_x
+  aom_rb_read_literal(&rb, 16);  // white_point_chromaticity_y
+
+  aom_rb_read_unsigned_literal(&rb, 32);  // luminance_max
+  aom_rb_read_unsigned_literal(&rb, 32);  // luminance_min
+}
+
+static void scalability_structure(struct aom_read_bit_buffer *rb) {
+  int spatial_layers_cnt = aom_rb_read_literal(rb, 2);
+  int spatial_layer_dimensions_present_flag = aom_rb_read_literal(rb, 1);
+  int spatial_layer_description_present_flag = aom_rb_read_literal(rb, 1);
+  int temporal_group_description_present_flag = aom_rb_read_literal(rb, 1);
+  aom_rb_read_literal(rb, 3);  // reserved
+
+  if (spatial_layer_dimensions_present_flag) {
+    int i;
+    for (i = 0; i < spatial_layers_cnt + 1; i++) {
+      aom_rb_read_literal(rb, 16);
+      aom_rb_read_literal(rb, 16);
+    }
+  }
+  if (spatial_layer_description_present_flag) {
+    int i;
+    for (i = 0; i < spatial_layers_cnt + 1; i++) {
+      aom_rb_read_literal(rb, 8);
+    }
+  }
+  if (temporal_group_description_present_flag) {
+    int i, j, temporal_group_size;
+    temporal_group_size = aom_rb_read_literal(rb, 8);
+    for (i = 0; i < temporal_group_size; i++) {
+      aom_rb_read_literal(rb, 3);
+      aom_rb_read_literal(rb, 1);
+      aom_rb_read_literal(rb, 1);
+      int temporal_group_ref_cnt = aom_rb_read_literal(rb, 3);
+      for (j = 0; j < temporal_group_ref_cnt; j++) {
+        aom_rb_read_literal(rb, 8);
+      }
+    }
+  }
+}
+
+static void read_metadata_scalability(const uint8_t *data, size_t sz) {
+  struct aom_read_bit_buffer rb = { data, data + sz, 0, NULL, NULL };
+  int scalability_mode_idc = aom_rb_read_literal(&rb, 8);
+  if (scalability_mode_idc == SCALABILITY_SS) {
+    scalability_structure(&rb);
+  }
+}
+
+static void read_metadata_timecode(const uint8_t *data, size_t sz) {
+  struct aom_read_bit_buffer rb = { data, data + sz, 0, NULL, NULL };
+  aom_rb_read_literal(&rb, 5);                     // counting_type f(5)
+  int full_timestamp_flag = aom_rb_read_bit(&rb);  // full_timestamp_flag f(1)
+  aom_rb_read_bit(&rb);                            // discontinuity_flag (f1)
+  aom_rb_read_bit(&rb);                            // cnt_dropped_flag f(1)
+  aom_rb_read_literal(&rb, 9);                     // n_frames f(9)
+  if (full_timestamp_flag) {
+    aom_rb_read_literal(&rb, 6);  // seconds_value f(6)
+    aom_rb_read_literal(&rb, 6);  // minutes_value f(6)
+    aom_rb_read_literal(&rb, 5);  // hours_value f(5)
+  } else {
+    int seconds_flag = aom_rb_read_bit(&rb);  // seconds_flag f(1)
+    if (seconds_flag) {
+      aom_rb_read_literal(&rb, 6);              // seconds_value f(6)
+      int minutes_flag = aom_rb_read_bit(&rb);  // minutes_flag f(1)
+      if (minutes_flag) {
+        aom_rb_read_literal(&rb, 6);            // minutes_value f(6)
+        int hours_flag = aom_rb_read_bit(&rb);  // hours_flag f(1)
+        if (hours_flag) {
+          aom_rb_read_literal(&rb, 5);  // hours_value f(5)
+        }
+      }
+    }
+  }
+  // time_offset_length f(5)
+  int time_offset_length = aom_rb_read_literal(&rb, 5);
+  if (time_offset_length) {
+    aom_rb_read_literal(&rb, time_offset_length);  // f(time_offset_length)
+  }
+}
+
+static size_t read_metadata(const uint8_t *data, size_t sz) {
+  size_t type_length;
+  uint64_t type_value;
+  OBU_METADATA_TYPE metadata_type;
+  if (aom_uleb_decode(data, sz, &type_value, &type_length) < 0) {
+    return sz;
+  }
+  metadata_type = (OBU_METADATA_TYPE)type_value;
+  if (metadata_type == OBU_METADATA_TYPE_ITUT_T35) {
+    read_metadata_itut_t35(data + type_length, sz - type_length);
+  } else if (metadata_type == OBU_METADATA_TYPE_HDR_CLL) {
+    read_metadata_hdr_cll(data + type_length, sz - type_length);
+  } else if (metadata_type == OBU_METADATA_TYPE_HDR_MDCV) {
+    read_metadata_hdr_mdcv(data + type_length, sz - type_length);
+  } else if (metadata_type == OBU_METADATA_TYPE_SCALABILITY) {
+    read_metadata_scalability(data + type_length, sz - type_length);
+  } else if (metadata_type == OBU_METADATA_TYPE_TIMECODE) {
+    read_metadata_timecode(data + type_length, sz - type_length);
+  }
+
+  return sz;
+}
+
+static aom_codec_err_t read_obu_size(const uint8_t *data,
+                                     size_t bytes_available,
+                                     size_t *const obu_size,
+                                     size_t *const length_field_size) {
+  uint64_t u_obu_size = 0;
+  if (aom_uleb_decode(data, bytes_available, &u_obu_size, length_field_size) !=
+      0) {
+    return AOM_CODEC_CORRUPT_FRAME;
+  }
+
+  if (u_obu_size > UINT32_MAX) return AOM_CODEC_CORRUPT_FRAME;
+  *obu_size = (size_t)u_obu_size;
+  return AOM_CODEC_OK;
+}
+
+aom_codec_err_t aom_read_obu_header_and_size(const uint8_t *data,
+                                             size_t bytes_available,
+                                             int is_annexb,
+                                             ObuHeader *obu_header,
+                                             size_t *const payload_size,
+                                             size_t *const bytes_read) {
+  size_t length_field_size = 0, obu_size = 0;
+  aom_codec_err_t status;
+
+  if (is_annexb) {
+    // Size field comes before the OBU header, and includes the OBU header
+    status =
+        read_obu_size(data, bytes_available, &obu_size, &length_field_size);
+
+    if (status != AOM_CODEC_OK) return status;
+  }
+
+  struct aom_read_bit_buffer rb = { data + length_field_size,
+                                    data + bytes_available, 0, NULL, NULL };
+
+  status = read_obu_header(&rb, is_annexb, obu_header);
+  if (status != AOM_CODEC_OK) return status;
+
+  if (is_annexb) {
+    // Derive the payload size from the data we've already read
+    if (obu_size < obu_header->size) return AOM_CODEC_CORRUPT_FRAME;
+
+    *payload_size = obu_size - obu_header->size;
+  } else {
+    // Size field comes after the OBU header, and is just the payload size
+    status = read_obu_size(data + obu_header->size,
+                           bytes_available - obu_header->size, payload_size,
+                           &length_field_size);
+    if (status != AOM_CODEC_OK) return status;
+  }
+
+  *bytes_read = length_field_size + obu_header->size;
+  return AOM_CODEC_OK;
+}
+
+#define EXT_TILE_DEBUG 0
+// On success, returns a boolean that indicates whether the decoding of the
+// current frame is finished. On failure, sets cm->error.error_code and
+// returns -1.
+int aom_decode_frame_from_obus(struct AV1Decoder *pbi, const uint8_t *data,
+                               const uint8_t *data_end,
+                               const uint8_t **p_data_end) {
+  AV1_COMMON *const cm = &pbi->common;
+  int frame_decoding_finished = 0;
+  int is_first_tg_obu_received = 1;
+  int frame_header_size = 0;
+  int seq_header_received = 0;
+  size_t seq_header_size = 0;
+  ObuHeader obu_header;
+  memset(&obu_header, 0, sizeof(obu_header));
+  pbi->seen_frame_header = 0;
+
+  if (data_end < data) {
+    cm->error.error_code = AOM_CODEC_CORRUPT_FRAME;
+    return -1;
+  }
+
+  // Reset pbi->camera_frame_header_ready to 0 if cm->large_scale_tile = 0.
+  if (!cm->large_scale_tile) pbi->camera_frame_header_ready = 0;
+
+  // decode frame as a series of OBUs
+  while (!frame_decoding_finished && !cm->error.error_code) {
+    struct aom_read_bit_buffer rb;
+    size_t payload_size = 0;
+    size_t decoded_payload_size = 0;
+    size_t obu_payload_offset = 0;
+    size_t bytes_read = 0;
+    const size_t bytes_available = data_end - data;
+
+    if (bytes_available == 0 && !pbi->seen_frame_header) {
+      *p_data_end = data;
+      cm->error.error_code = AOM_CODEC_OK;
+      break;
+    }
+
+    aom_codec_err_t status =
+        aom_read_obu_header_and_size(data, bytes_available, cm->is_annexb,
+                                     &obu_header, &payload_size, &bytes_read);
+
+    if (status != AOM_CODEC_OK) {
+      cm->error.error_code = status;
+      return -1;
+    }
+
+    // Record obu size header information.
+    pbi->obu_size_hdr.data = data + obu_header.size;
+    pbi->obu_size_hdr.size = bytes_read - obu_header.size;
+
+    // Note: aom_read_obu_header_and_size() takes care of checking that this
+    // doesn't cause 'data' to advance past 'data_end'.
+    data += bytes_read;
+
+    if ((size_t)(data_end - data) < payload_size) {
+      cm->error.error_code = AOM_CODEC_CORRUPT_FRAME;
+      return -1;
+    }
+
+    cm->temporal_layer_id = obu_header.temporal_layer_id;
+    cm->spatial_layer_id = obu_header.spatial_layer_id;
+
+    if (obu_header.type != OBU_TEMPORAL_DELIMITER &&
+        obu_header.type != OBU_SEQUENCE_HEADER &&
+        obu_header.type != OBU_PADDING) {
+      // don't decode obu if it's not in current operating mode
+      if (!is_obu_in_current_operating_point(pbi, obu_header)) {
+        data += payload_size;
+        continue;
+      }
+    }
+
+    av1_init_read_bit_buffer(pbi, &rb, data, data_end);
+
+    switch (obu_header.type) {
+      case OBU_TEMPORAL_DELIMITER:
+        decoded_payload_size = read_temporal_delimiter_obu();
+        pbi->seen_frame_header = 0;
+        break;
+      case OBU_SEQUENCE_HEADER:
+        if (!seq_header_received) {
+          decoded_payload_size = read_sequence_header_obu(pbi, &rb);
+          if (cm->error.error_code != AOM_CODEC_OK) return -1;
+
+          seq_header_size = decoded_payload_size;
+          seq_header_received = 1;
+        } else {
+          // Seeing another sequence header, skip as all sequence headers are
+          // required to be identical except for the contents of
+          // operating_parameters_info and the amount of trailing bits.
+          // TODO(yaowu): verifying redundant sequence headers are identical.
+          decoded_payload_size = seq_header_size;
+        }
+        break;
+      case OBU_FRAME_HEADER:
+      case OBU_REDUNDANT_FRAME_HEADER:
+      case OBU_FRAME:
+        // Only decode first frame header received
+        if (!pbi->seen_frame_header ||
+            (cm->large_scale_tile && !pbi->camera_frame_header_ready)) {
+          pbi->seen_frame_header = 1;
+          frame_header_size = read_frame_header_obu(
+              pbi, &rb, data, p_data_end, obu_header.type != OBU_FRAME);
+          if (cm->large_scale_tile) pbi->camera_frame_header_ready = 1;
+        }
+        decoded_payload_size = frame_header_size;
+        pbi->frame_header_size = (size_t)frame_header_size;
+
+        if (cm->show_existing_frame) {
+          frame_decoding_finished = 1;
+          pbi->seen_frame_header = 0;
+          break;
+        }
+
+#if !EXT_TILE_DEBUG
+        // In large scale tile coding, decode the common camera frame header
+        // before any tile list OBU.
+        if (!pbi->ext_tile_debug && pbi->camera_frame_header_ready) {
+          frame_decoding_finished = 1;
+          // Skip the rest of the frame data.
+          decoded_payload_size = payload_size;
+          // Update data_end.
+          *p_data_end = data_end;
+          break;
+        }
+#endif  // EXT_TILE_DEBUG
+
+        if (obu_header.type != OBU_FRAME) break;
+        obu_payload_offset = frame_header_size;
+        AOM_FALLTHROUGH_INTENDED;  // fall through to read tile group.
+      case OBU_TILE_GROUP:
+        if (!pbi->seen_frame_header) {
+          cm->error.error_code = AOM_CODEC_CORRUPT_FRAME;
+          return -1;
+        }
+        if ((size_t)(data_end - data) < obu_payload_offset) {
+          cm->error.error_code = AOM_CODEC_CORRUPT_FRAME;
+          return -1;
+        }
+        decoded_payload_size += read_one_tile_group_obu(
+            pbi, &rb, is_first_tg_obu_received, data + obu_payload_offset,
+            data + payload_size, p_data_end, &frame_decoding_finished,
+            obu_header.type == OBU_FRAME);
+        is_first_tg_obu_received = 0;
+        if (frame_decoding_finished) pbi->seen_frame_header = 0;
+        break;
+      case OBU_METADATA:
+        decoded_payload_size = read_metadata(data, payload_size);
+        break;
+      case OBU_TILE_LIST:
+        // This OBU type is purely for the large scale tile coding mode.
+        // The common camera frame header has to be already decoded.
+        if (!pbi->camera_frame_header_ready) {
+          cm->error.error_code = AOM_CODEC_CORRUPT_FRAME;
+          return -1;
+        }
+
+        cm->large_scale_tile = 1;
+        av1_set_single_tile_decoding_mode(cm);
+        decoded_payload_size =
+            read_and_decode_one_tile_list(pbi, &rb, data, data + payload_size,
+                                          p_data_end, &frame_decoding_finished);
+        if (cm->error.error_code != AOM_CODEC_OK) return -1;
+        break;
+      case OBU_PADDING:
+      default:
+        // Skip unrecognized OBUs
+        decoded_payload_size = payload_size;
+        break;
+    }
+
+    // Check that the signalled OBU size matches the actual amount of data read
+    if (decoded_payload_size > payload_size) {
+      cm->error.error_code = AOM_CODEC_CORRUPT_FRAME;
+      return -1;
+    }
+
+    // If there are extra padding bytes, they should all be zero
+    while (decoded_payload_size < payload_size) {
+      uint8_t padding_byte = data[decoded_payload_size++];
+      if (padding_byte != 0) {
+        cm->error.error_code = AOM_CODEC_CORRUPT_FRAME;
+        return -1;
+      }
+    }
+
+    data += payload_size;
+  }
+
+  return frame_decoding_finished;
+}
+#undef EXT_TILE_DEBUG
diff --git a/third_party/aom/av1/decoder/obu.h b/third_party/aom/av1/decoder/obu.h
new file mode 100644
index 000000000..5f2197058
--- /dev/null
+++ b/third_party/aom/av1/decoder/obu.h
@@ -0,0 +1,54 @@
+/*
+ * Copyright (c) 2017, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AV1_DECODER_OBU_H
+#define AV1_DECODER_OBU_H
+
+#include "aom/aom_codec.h"
+#include "av1/decoder/decoder.h"
+
+typedef struct {
+  size_t size;  // Size (1 or 2 bytes) of the OBU header (including the
+                // optional OBU extension header) in the bitstream.
+  OBU_TYPE type;
+  int has_size_field;
+  int has_extension;
+  // The following fields come from the OBU extension header and therefore are
+  // only used if has_extension is true.
+  int temporal_layer_id;
+  int spatial_layer_id;
+} ObuHeader;
+
+aom_codec_err_t aom_read_obu_header(uint8_t *buffer, size_t buffer_length,
+                                    size_t *consumed, ObuHeader *header,
+                                    int is_annexb);
+
+aom_codec_err_t aom_read_obu_header_and_size(const uint8_t *data,
+                                             size_t bytes_available,
+                                             int is_annexb,
+                                             ObuHeader *obu_header,
+                                             size_t *const payload_size,
+                                             size_t *const bytes_read);
+
+// Try to decode one frame from a buffer.
+// Returns 1 if we decoded a frame,
+//         0 if we didn't decode a frame but that's okay
+//           (eg, if there was a frame but we skipped it),
+//     or -1 on error
+int aom_decode_frame_from_obus(struct AV1Decoder *pbi, const uint8_t *data,
+                               const uint8_t *data_end,
+                               const uint8_t **p_data_end);
+
+aom_codec_err_t aom_get_num_layers_from_operating_point_idc(
+    int operating_point_idc, unsigned int *num_spatial_layers,
+    unsigned int *num_temporal_layers);
+
+#endif
diff --git a/third_party/aom/av1/decoder/pvq_decoder.c b/third_party/aom/av1/decoder/pvq_decoder.c
deleted file mode 100644
index d9a8e8056..000000000
--- a/third_party/aom/av1/decoder/pvq_decoder.c
+++ /dev/null
@@ -1,378 +0,0 @@
-/*
- * Copyright (c) 2001-2016, Alliance for Open Media. All rights reserved
- *
- * This source code is subject to the terms of the BSD 2 Clause License and
- * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
- * was not distributed with this source code in the LICENSE file, you can
- * obtain it at www.aomedia.org/license/software. If the Alliance for Open
- * Media Patent License 1.0 was not distributed with this source code in the
- * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
- */
-
-/* clang-format off */
-
-#ifdef HAVE_CONFIG_H
-# include "config.h"
-#endif
-
-#include <stdio.h>
-#include <stdlib.h>
-#include "./aom_config.h"
-#include "aom_dsp/bitreader.h"
-#include "aom_dsp/entcode.h"
-#include "aom_dsp/entdec.h"
-#include "av1/common/odintrin.h"
-#include "av1/common/partition.h"
-#include "av1/common/pvq_state.h"
-#include "av1/decoder/decint.h"
-#include "av1/decoder/pvq_decoder.h"
-#include "aom_ports/system_state.h"
-
-int aom_read_symbol_pvq_(aom_reader *r, aom_cdf_prob *cdf, int nsymbs
- ACCT_STR_PARAM) {
-  if (cdf[0] == 0)
-    aom_cdf_init_q15_1D(cdf, nsymbs, CDF_SIZE(nsymbs));
-  return aom_read_symbol(r, cdf, nsymbs, ACCT_STR_NAME);
-}
-
-static void aom_decode_pvq_codeword(aom_reader *r, od_pvq_codeword_ctx *ctx,
- od_coeff *y, int n, int k) {
-  int i;
-  aom_decode_band_pvq_splits(r, ctx, y, n, k, 0);
-  for (i = 0; i < n; i++) {
-    if (y[i] && aom_read_bit(r, "pvq:sign")) y[i] = -y[i];
-  }
-}
-
-/** Inverse of neg_interleave; decodes the interleaved gain.
- *
- * @param [in]      x      quantized/interleaved gain to decode
- * @param [in]      ref    quantized gain of the reference
- * @return                 original quantized gain value
- */
-static int neg_deinterleave(int x, int ref) {
-  if (x < 2*ref-1) {
-    if (x & 1) return ref - 1 - (x >> 1);
-    else return ref + (x >> 1);
-  }
-  else return x+1;
-}
-
-/** Synthesizes one parition of coefficient values from a PVQ-encoded
- * vector.
- *
- * @param [out]     xcoeff  output coefficient partition (x in math doc)
- * @param [in]      ypulse  PVQ-encoded values (y in math doc); in the noref
- *                          case, this vector has n entries, in the
- *                          reference case it contains n-1 entries
- *                          (the m-th entry is not included)
- * @param [in]      ref     reference vector (prediction)
- * @param [in]      n       number of elements in this partition
- * @param [in]      gr      gain of the reference vector (prediction)
- * @param [in]      noref   indicates presence or lack of prediction
- * @param [in]      g       decoded quantized vector gain
- * @param [in]      theta   decoded theta (prediction error)
- * @param [in]      qm      QM with magnitude compensation
- * @param [in]      qm_inv  Inverse of QM with magnitude compensation
- */
-static void pvq_synthesis(od_coeff *xcoeff, od_coeff *ypulse, od_val16 *r16,
- int n, od_val32 gr, int noref, od_val32 g, od_val32 theta, const int16_t *qm_inv,
- int shift) {
-  int s;
-  int m;
-  /* Sign of the Householder reflection vector */
-  s = 0;
-  /* Direction of the Householder reflection vector */
-  m = noref ? 0 : od_compute_householder(r16, n, gr, &s, shift);
-  od_pvq_synthesis_partial(xcoeff, ypulse, r16, n, noref, g, theta, m, s,
-   qm_inv);
-}
-
-typedef struct {
-  od_coeff *ref;
-  int nb_coeffs;
-  int allow_flip;
-} cfl_ctx;
-
-/** Decodes a single vector of integers (eg, a partition within a
- *  coefficient block) encoded using PVQ
- *
- * @param [in,out] ec          range encoder
- * @param [in]     q0          scale/quantizer
- * @param [in]     n           number of coefficients in partition
- * @param [in,out] model       entropy decoder state
- * @param [in,out] adapt       adaptation context
- * @param [in,out] exg         ExQ16 expectation of decoded gain value
- * @param [in,out] ext         ExQ16 expectation of decoded theta value
- * @param [in]     ref         'reference' (prediction) vector
- * @param [out]    out         decoded partition
- * @param [out]    noref       boolean indicating absence of reference
- * @param [in]     beta        per-band activity masking beta param
- * @param [in]     is_keyframe whether we're encoding a keyframe
- * @param [in]     pli         plane index
- * @param [in]     cdf_ctx     selects which cdf context to use
- * @param [in,out] skip_rest   whether to skip further bands in each direction
- * @param [in]     band        index of the band being decoded
- * @param [in]     band        index of the band being decoded
- * @param [out]    skip        skip flag with range [0,1]
- * @param [in]     qm          QM with magnitude compensation
- * @param [in]     qm_inv      Inverse of QM with magnitude compensation
- */
-static void pvq_decode_partition(aom_reader *r,
-                                 int q0,
-                                 int n,
-                                 generic_encoder model[3],
-                                 od_adapt_ctx *adapt,
-                                 int *exg,
-                                 int *ext,
-                                 od_coeff *ref,
-                                 od_coeff *out,
-                                 int *noref,
-                                 od_val16 beta,
-                                 int is_keyframe,
-                                 int pli,
-                                 int cdf_ctx,
-                                 cfl_ctx *cfl,
-                                 int has_skip,
-                                 int *skip_rest,
-                                 int band,
-                                 int *skip,
-                                 const int16_t *qm,
-                                 const int16_t *qm_inv) {
-  int k;
-  od_val32 qcg;
-  int itheta;
-  od_val32 theta;
-  od_val32 gr;
-  od_val32 gain_offset;
-  od_coeff y[MAXN];
-  int qg;
-  int id;
-  int i;
-  od_val16 ref16[MAXN];
-  int rshift;
-  theta = 0;
-  gr = 0;
-  gain_offset = 0;
-  /* Skip is per-direction. For band=0, we can use any of the flags. */
-  if (skip_rest[(band + 2) % 3]) {
-    qg = 0;
-    if (is_keyframe) {
-      itheta = -1;
-      *noref = 1;
-    }
-    else {
-      itheta = 0;
-      *noref = 0;
-    }
-  }
-  else {
-    /* Jointly decode gain, itheta and noref for small values. Then we handle
-       larger gain. */
-    id = aom_read_symbol_pvq(r, &adapt->pvq.pvq_gaintheta_cdf[cdf_ctx][0],
-     8 + 7*has_skip, "pvq:gaintheta");
-    if (!is_keyframe && id >= 10) id++;
-    if (is_keyframe && id >= 8) id++;
-    if (id >= 8) {
-      id -= 8;
-      skip_rest[0] = skip_rest[1] = skip_rest[2] = 1;
-    }
-    qg = id & 1;
-    itheta = (id >> 1) - 1;
-    *noref = (itheta == -1);
-  }
-  /* The CfL flip bit is only decoded on the first band that has noref=0. */
-  if (cfl->allow_flip && !*noref) {
-    int flip;
-    flip = aom_read_bit(r, "cfl:flip");
-    if (flip) {
-      for (i = 0; i < cfl->nb_coeffs; i++) cfl->ref[i] = -cfl->ref[i];
-    }
-    cfl->allow_flip = 0;
-  }
-  if (qg > 0) {
-    int tmp;
-    tmp = *exg;
-    qg = 1 + generic_decode(r, &model[!*noref], &tmp, 2, "pvq:gain");
-    OD_IIR_DIADIC(*exg, qg << 16, 2);
-  }
-  *skip = 0;
-#if defined(OD_FLOAT_PVQ)
-  rshift = 0;
-#else
-  /* Shift needed to make the reference fit in 15 bits, so that the Householder
-     vector can fit in 16 bits. */
-  rshift = OD_MAXI(0, od_vector_log_mag(ref, n) - 14);
-#endif
-  for (i = 0; i < n; i++) {
-#if defined(OD_FLOAT_PVQ)
-    ref16[i] = ref[i]*(double)qm[i]*OD_QM_SCALE_1;
-#else
-    ref16[i] = OD_SHR_ROUND(ref[i]*qm[i], OD_QM_SHIFT + rshift);
-#endif
-  }
-  if(!*noref){
-    /* we have a reference; compute its gain */
-    od_val32 cgr;
-    int icgr;
-    int cfl_enabled;
-    cfl_enabled = pli != 0 && is_keyframe && !OD_DISABLE_CFL;
-    cgr = od_pvq_compute_gain(ref16, n, q0, &gr, beta, rshift);
-    if (cfl_enabled) cgr = OD_CGAIN_SCALE;
-#if defined(OD_FLOAT_PVQ)
-    icgr = (int)floor(.5 + cgr);
-#else
-    icgr = OD_SHR_ROUND(cgr, OD_CGAIN_SHIFT);
-#endif
-    /* quantized gain is interleave encoded when there's a reference;
-       deinterleave it now */
-    if (is_keyframe) qg = neg_deinterleave(qg, icgr);
-    else {
-      qg = neg_deinterleave(qg, icgr + 1) - 1;
-      if (qg == 0) *skip = (icgr ? OD_PVQ_SKIP_ZERO : OD_PVQ_SKIP_COPY);
-    }
-    if (qg == icgr && itheta == 0 && !cfl_enabled) *skip = OD_PVQ_SKIP_COPY;
-    gain_offset = cgr - OD_SHL(icgr, OD_CGAIN_SHIFT);
-    qcg = OD_SHL(qg, OD_CGAIN_SHIFT) + gain_offset;
-    /* read and decode first-stage PVQ error theta */
-    if (itheta > 1) {
-      int tmp;
-      tmp = *ext;
-      itheta = 2 + generic_decode(r, &model[2], &tmp, 2, "pvq:theta");
-      OD_IIR_DIADIC(*ext, itheta << 16, 2);
-    }
-    theta = od_pvq_compute_theta(itheta, od_pvq_compute_max_theta(qcg, beta));
-  }
-  else{
-    itheta = 0;
-    if (!is_keyframe) qg++;
-    qcg = OD_SHL(qg, OD_CGAIN_SHIFT);
-    if (qg == 0) *skip = OD_PVQ_SKIP_ZERO;
-  }
-
-  k = od_pvq_compute_k(qcg, itheta, *noref, n, beta);
-  if (k != 0) {
-    /* when noref==0, y is actually size n-1 */
-    aom_decode_pvq_codeword(r, &adapt->pvq.pvq_codeword_ctx, y,
-     n - !*noref, k);
-  }
-  else {
-    OD_CLEAR(y, n);
-  }
-  if (*skip) {
-    if (*skip == OD_PVQ_SKIP_COPY) OD_COPY(out, ref, n);
-    else OD_CLEAR(out, n);
-  }
-  else {
-    od_val32 g;
-    g = od_gain_expand(qcg, q0, beta);
-    pvq_synthesis(out, y, ref16, n, gr, *noref, g, theta, qm_inv, rshift);
-  }
-  /* If OD_PVQ_SKIP_ZERO or OD_PVQ_SKIP_COPY, set skip to 1 for visualization */
-  if (*skip) *skip = 1;
-}
-
-/** Decodes a coefficient block (except for DC) encoded using PVQ
- *
- * @param [in,out] dec         daala decoder context
- * @param [in]     ref         'reference' (prediction) vector
- * @param [out]    out         decoded partition
- * @param [in]     q0          quantizer
- * @param [in]     pli         plane index
- * @param [in]     bs          log of the block size minus two
- * @param [in]     beta        per-band activity masking beta param
- * @param [in]     is_keyframe whether we're encoding a keyframe
- * @param [out]    flags       bitmask of the per band skip and noref flags
- * @param [in]     ac_dc_coded skip flag for the block (range 0-3)
- * @param [in]     qm          QM with magnitude compensation
- * @param [in]     qm_inv      Inverse of QM with magnitude compensation
- */
-void od_pvq_decode(daala_dec_ctx *dec,
-                   od_coeff *ref,
-                   od_coeff *out,
-                   int q0,
-                   int pli,
-                   int bs,
-                   const od_val16 *beta,
-                   int is_keyframe,
-                   unsigned int *flags,
-                   PVQ_SKIP_TYPE ac_dc_coded,
-                   const int16_t *qm,
-                   const int16_t *qm_inv){
-
-  int noref[PVQ_MAX_PARTITIONS];
-  int skip[PVQ_MAX_PARTITIONS];
-  int *exg;
-  int *ext;
-  int nb_bands;
-  int i;
-  const int *off;
-  int size[PVQ_MAX_PARTITIONS];
-  generic_encoder *model;
-  int skip_rest[3] = {0};
-  cfl_ctx cfl;
-  const unsigned char *pvq_qm;
-  int use_masking;
-
-  aom_clear_system_state();
-
-  /*Default to skip=1 and noref=0 for all bands.*/
-  for (i = 0; i < PVQ_MAX_PARTITIONS; i++) {
-    noref[i] = 0;
-    skip[i] = 1;
-  }
-
-  use_masking = dec->use_activity_masking;
-
-  if (use_masking)
-    pvq_qm = &dec->state.pvq_qm_q4[pli][0];
-  else
-    pvq_qm = 0;
-
-  exg = &dec->state.adapt->pvq.pvq_exg[pli][bs][0];
-  ext = dec->state.adapt->pvq.pvq_ext + bs*PVQ_MAX_PARTITIONS;
-  model = dec->state.adapt->pvq.pvq_param_model;
-  nb_bands = OD_BAND_OFFSETS[bs][0];
-  off = &OD_BAND_OFFSETS[bs][1];
-  out[0] = ac_dc_coded & DC_CODED;
-  if (ac_dc_coded < AC_CODED) {
-    if (is_keyframe) for (i = 1; i < 1 << (2*bs + 4); i++) out[i] = 0;
-    else for (i = 1; i < 1 << (2*bs + 4); i++) out[i] = ref[i];
-  }
-  else {
-    for (i = 0; i < nb_bands; i++) size[i] = off[i+1] - off[i];
-    cfl.ref = ref;
-    cfl.nb_coeffs = off[nb_bands];
-    cfl.allow_flip = pli != 0 && is_keyframe;
-    for (i = 0; i < nb_bands; i++) {
-      int q;
-
-      if (use_masking)
-        q = OD_MAXI(1, q0 * pvq_qm[od_qm_get_index(bs, i + 1)] >> 4);
-      else
-        q = OD_MAXI(1, q0);
-
-      pvq_decode_partition(dec->r, q, size[i],
-       model, dec->state.adapt, exg + i, ext + i, ref + off[i], out + off[i],
-       &noref[i], beta[i], is_keyframe, pli,
-       (pli != 0)*OD_TXSIZES*PVQ_MAX_PARTITIONS + bs*PVQ_MAX_PARTITIONS + i,
-       &cfl, i == 0 && (i < nb_bands - 1), skip_rest, i, &skip[i],
-       qm + off[i], qm_inv + off[i]);
-      if (i == 0 && !skip_rest[0] && bs > 0) {
-        int skip_dir;
-        int j;
-        skip_dir = aom_read_symbol(dec->r,
-         &dec->state.adapt->pvq.pvq_skip_dir_cdf[(pli != 0) + 2*(bs - 1)][0], 7,
-         "pvq:skiprest");
-        for (j = 0; j < 3; j++) skip_rest[j] = !!(skip_dir & (1 << j));
-      }
-    }
-  }
-  *flags = 0;
-  for (i = nb_bands - 1; i >= 0; i--) {
-    *flags <<= 1;
-    *flags |= noref[i]&1;
-    *flags <<= 1;
-    *flags |= skip[i]&1;
-  }
-}
diff --git a/third_party/aom/av1/decoder/pvq_decoder.h b/third_party/aom/av1/decoder/pvq_decoder.h
deleted file mode 100644
index 98970663b..000000000
--- a/third_party/aom/av1/decoder/pvq_decoder.h
+++ /dev/null
@@ -1,40 +0,0 @@
-/*
- * Copyright (c) 2001-2016, Alliance for Open Media. All rights reserved
- *
- * This source code is subject to the terms of the BSD 2 Clause License and
- * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
- * was not distributed with this source code in the LICENSE file, you can
- * obtain it at www.aomedia.org/license/software. If the Alliance for Open
- * Media Patent License 1.0 was not distributed with this source code in the
- * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
- */
-
-/* clang-format off */
-
-#if !defined(_pvq_decoder_H)
-# define _pvq_decoder_H (1)
-# include "aom_dsp/bitreader.h"
-# include "aom_dsp/entdec.h"
-# include "av1/common/pvq.h"
-# include "av1/decoder/decint.h"
-
-#define aom_read_symbol_pvq(r, cdf, nsymbs, ACCT_STR_NAME) \
-  aom_read_symbol_pvq_(r, cdf, nsymbs ACCT_STR_ARG(ACCT_STR_NAME))
-
-int aom_read_symbol_pvq_(aom_reader *r, aom_cdf_prob *cdf, int nsymbs
-  ACCT_STR_PARAM);
-
-void aom_decode_band_pvq_splits(aom_reader *r, od_pvq_codeword_ctx *adapt,
- od_coeff *y, int n, int k, int level);
-
-#define aom_laplace_decode_special(r, decay, ACCT_STR_NAME) \
-  aom_laplace_decode_special_(r, decay ACCT_STR_ARG(ACCT_STR_NAME))
-
-int aom_laplace_decode_special_(aom_reader *r, unsigned decay ACCT_STR_PARAM);
-
-void od_pvq_decode(daala_dec_ctx *dec, od_coeff *ref, od_coeff *out, int q0,
-    int pli, int bs, const od_val16 *beta, int is_keyframe,
-    unsigned int *flags, PVQ_SKIP_TYPE ac_dc_coded, const int16_t *qm,
-    const int16_t *qm_inv);
-
-#endif
diff --git a/third_party/aom/av1/decoder/symbolrate.h b/third_party/aom/av1/decoder/symbolrate.h
deleted file mode 100644
index 023287732..000000000
--- a/third_party/aom/av1/decoder/symbolrate.h
+++ /dev/null
@@ -1,88 +0,0 @@
-/*
- * Copyright (c) 2017, Alliance for Open Media. All rights reserved
- *
- * This source code is subject to the terms of the BSD 2 Clause License and
- * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
- * was not distributed with this source code in the LICENSE file, you can
- * obtain it at www.aomedia.org/license/software. If the Alliance for Open
- * Media Patent License 1.0 was not distributed with this source code in the
- * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
- */
-
-#include "aom_dsp/bitreader.h"
-
-#ifndef AV1_DECODER_SYMBOLRATE_H_
-#define AV1_DECODER_SYMBOLRATE_H_
-
-#if CONFIG_SYMBOLRATE
-static INLINE void av1_dump_symbol_rate(struct AV1Common *cm) {
-  const FRAME_COUNTS *counts = &cm->counts;
-  printf("%d %d %d %d\n", counts->coeff_num[0], counts->coeff_num[1],
-         counts->symbol_num[0], counts->symbol_num[1]);
-}
-static INLINE int av1_read_record_symbol(FRAME_COUNTS *counts, aom_reader *r,
-                                         aom_cdf_prob *cdf, int nsymbs,
-                                         const char *str) {
-  (void)str;
-  if (counts) ++counts->symbol_num[0];
-  return aom_read_symbol(r, cdf, nsymbs, str);
-}
-
-#if CONFIG_LV_MAP
-static INLINE int av1_read_record_bin(FRAME_COUNTS *counts, aom_reader *r,
-                                      aom_cdf_prob *cdf, int nsymbs,
-                                      const char *str) {
-  (void)str;
-  if (counts) ++counts->symbol_num[0];
-  return aom_read_bin(r, cdf, nsymbs, str);
-}
-#endif
-
-static INLINE int av1_read_record(FRAME_COUNTS *counts, aom_reader *r, int prob,
-                                  const char *str) {
-  (void)str;
-  if (counts) ++counts->symbol_num[0];
-  return aom_read(r, prob, str);
-}
-
-static INLINE int av1_read_record_cdf(FRAME_COUNTS *counts, aom_reader *r,
-                                      const aom_cdf_prob *cdf, int nsymbs,
-                                      const char *str) {
-  (void)str;
-  if (counts) ++counts->symbol_num[0];
-  return aom_read_cdf(r, cdf, nsymbs, str);
-}
-
-static INLINE int av1_read_record_bit(FRAME_COUNTS *counts, aom_reader *r,
-                                      const char *str) {
-  (void)str;
-  if (counts) ++counts->symbol_num[1];
-  return aom_read_bit(r, str);
-}
-
-static INLINE void av1_record_coeff(FRAME_COUNTS *counts, tran_low_t qcoeff) {
-  assert(qcoeff >= 0);
-  if (counts) ++counts->coeff_num[qcoeff != 0];
-}
-#else  // CONFIG_SYMBOLRATE
-
-#define av1_read_record_symbol(counts, r, cdf, nsymbs, ACCT_STR_NAME) \
-  aom_read_symbol(r, cdf, nsymbs, ACCT_STR_NAME)
-
-#if CONFIG_LV_MAP
-#define av1_read_record_bin(counts, r, cdf, nsymbs, ACCT_STR_NAME) \
-  aom_read_bin(r, cdf, nsymbs, ACCT_STR_NAME)
-#endif
-
-#define av1_read_record(counts, r, prob, ACCT_STR_NAME) \
-  aom_read(r, prob, ACCT_STR_NAME)
-
-#define av1_read_record_cdf(counts, r, cdf, nsymbs, ACCT_STR_NAME) \
-  aom_read_cdf(r, cdf, nsymbs, ACCT_STR_NAME)
-
-#define av1_read_record_bit(counts, r, ACCT_STR_NAME) \
-  aom_read_bit(r, ACCT_STR_NAME)
-
-#endif  // CONFIG_SYMBOLRATE
-
-#endif  // AV1_DECODER_SYMBOLRATE_H_