Update aom to commit id f5bdeac22930ff4c6b219be49c843db35970b918

1 files changed, 224 insertions, 754 deletions

diff --git a/third_party/aom/av1/encoder/encodemb.c b/third_party/aom/av1/encoder/encodemb.c
index 7c9781533..e7f4d313d 100644
--- a/third_party/aom/av1/encoder/encodemb.c
+++ b/third_party/aom/av1/encoder/encodemb.c
@@ -112,19 +112,9 @@ void av1_subtract_plane(MACROBLOCK *x, BLOCK_SIZE bsize, int plane) {
 
 // These numbers are empirically obtained.
 static const int plane_rd_mult[REF_TYPES][PLANE_TYPES] = {
-#if CONFIG_EC_ADAPT
   { 10, 7 }, { 8, 5 },
-#else
-  { 10, 6 }, { 8, 6 },
-#endif
 };
 
-#define UPDATE_RD_COST()                             \
-  {                                                  \
-    rd_cost0 = RDCOST(rdmult, rddiv, rate0, error0); \
-    rd_cost1 = RDCOST(rdmult, rddiv, rate1, error1); \
-  }
-
 static INLINE unsigned int get_token_bit_costs(
     unsigned int token_costs[2][COEFF_CONTEXTS][ENTROPY_TOKENS], int skip_eob,
     int ctx, int token) {
@@ -133,23 +123,14 @@ static INLINE unsigned int get_token_bit_costs(
 }
 
 #if !CONFIG_LV_MAP
-#define USE_GREEDY_OPTIMIZE_B 0
-
-#if USE_GREEDY_OPTIMIZE_B
-
-typedef struct av1_token_state_greedy {
-  int16_t token;
-  tran_low_t qc;
-  tran_low_t dqc;
-} av1_token_state_greedy;
 
 static int optimize_b_greedy(const AV1_COMMON *cm, MACROBLOCK *mb, int plane,
-                             int block, TX_SIZE tx_size, int ctx) {
+                             int blk_row, int blk_col, int block,
+                             TX_SIZE tx_size, int ctx) {
   MACROBLOCKD *const xd = &mb->e_mbd;
   struct macroblock_plane *const p = &mb->plane[plane];
   struct macroblockd_plane *const pd = &xd->plane[plane];
   const int ref = is_inter_block(&xd->mi[0]->mbmi);
-  av1_token_state_greedy tokens[MAX_TX_SQUARE + 1][2];
   uint8_t token_cache[MAX_TX_SQUARE];
   const tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
   tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
@@ -158,23 +139,27 @@ static int optimize_b_greedy(const AV1_COMMON *cm, MACROBLOCK *mb, int plane,
   const PLANE_TYPE plane_type = pd->plane_type;
   const int16_t *const dequant_ptr = pd->dequant;
   const uint8_t *const band_translate = get_band_translate(tx_size);
-  TX_TYPE tx_type = get_tx_type(plane_type, xd, block, tx_size);
+  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, is_inter_block(&xd->mi[0]->mbmi));
+      get_scan(cm, tx_size, tx_type, &xd->mi[0]->mbmi);
   const int16_t *const scan = scan_order->scan;
   const int16_t *const nb = scan_order->neighbors;
   int dqv;
   const int shift = av1_get_tx_scale(tx_size);
 #if CONFIG_AOM_QM
   int seg_id = xd->mi[0]->mbmi.segment_id;
-  const qm_val_t *iqmatrix = pd->seg_iqmatrix[seg_id][!ref][tx_size];
+  // Use a flat matrix (i.e. no weighting) for 1D and Identity transforms
+  const qm_val_t *iqmatrix =
+      IS_2D_TRANSFORM(tx_type)
+          ? pd->seg_iqmatrix[seg_id][!ref][tx_size]
+          : cm->giqmatrix[NUM_QM_LEVELS - 1][0][0][tx_size];
 #endif
 #if CONFIG_NEW_QUANT
   int dq = get_dq_profile_from_ctx(mb->qindex, ctx, ref, plane_type);
   const dequant_val_type_nuq *dequant_val = pd->dequant_val_nuq[dq];
 #endif  // CONFIG_NEW_QUANT
   int sz = 0;
-  const int64_t rddiv = mb->rddiv;
   int64_t rd_cost0, rd_cost1;
   int16_t t0, t1;
   int i, final_eob;
@@ -193,19 +178,8 @@ static int optimize_b_greedy(const AV1_COMMON *cm, MACROBLOCK *mb, int plane,
   int64_t rate0, rate1;
   for (i = 0; i < eob; i++) {
     const int rc = scan[i];
-    int x = qcoeff[rc];
-    t0 = av1_get_token(x);
-
-    tokens[i][0].qc = x;
-    tokens[i][0].token = t0;
-    tokens[i][0].dqc = dqcoeff[rc];
-
-    token_cache[rc] = av1_pt_energy_class[t0];
+    token_cache[rc] = av1_pt_energy_class[av1_get_token(qcoeff[rc])];
   }
-  tokens[eob][0].token = EOB_TOKEN;
-  tokens[eob][0].qc = 0;
-  tokens[eob][0].dqc = 0;
-  tokens[eob][1] = tokens[eob][0];
 
   unsigned int(*token_costs_ptr)[2][COEFF_CONTEXTS][ENTROPY_TOKENS] =
       token_costs;
@@ -213,20 +187,22 @@ static int optimize_b_greedy(const AV1_COMMON *cm, MACROBLOCK *mb, int plane,
   final_eob = 0;
 
   int64_t eob_cost0, eob_cost1;
+  tran_low_t before_best_eob_qc = 0;
+  tran_low_t before_best_eob_dqc = 0;
 
   const int ctx0 = ctx;
   /* Record the r-d cost */
   int64_t accu_rate = 0;
-  int64_t accu_error = 0;
+  // Initialized to the worst possible error for the largest transform size.
+  // This ensures that it never goes negative.
+  int64_t accu_error = ((int64_t)1) << 50;
 
   rate0 = get_token_bit_costs(*(token_costs_ptr + band_translate[0]), 0, ctx0,
                               EOB_TOKEN);
-  int64_t best_block_rd_cost = RDCOST(rdmult, rddiv, rate0, accu_error);
+  int64_t best_block_rd_cost = RDCOST(rdmult, rate0, accu_error);
 
   // int64_t best_block_rd_cost_all0 = best_block_rd_cost;
-
   int x_prev = 1;
-
   for (i = 0; i < eob; i++) {
     const int rc = scan[i];
     int x = qcoeff[rc];
@@ -238,9 +214,9 @@ static int optimize_b_greedy(const AV1_COMMON *cm, MACROBLOCK *mb, int plane,
 
     if (x == 0) {
       // no need to search when x == 0
-      rate0 =
-          get_token_bit_costs(*(token_costs_ptr + band_cur), token_tree_sel_cur,
-                              ctx_cur, tokens[i][0].token);
+      int token = av1_get_token(x);
+      rate0 = get_token_bit_costs(*(token_costs_ptr + band_cur),
+                                  token_tree_sel_cur, ctx_cur, token);
       accu_rate += rate0;
       x_prev = 0;
       // accu_error does not change when x==0
@@ -249,7 +225,7 @@ static int optimize_b_greedy(const AV1_COMMON *cm, MACROBLOCK *mb, int plane,
        */
       // compute the distortion for the first candidate
       // and the distortion for quantizing to 0.
-      int dx0 = (-coeff[rc]) * (1 << shift);
+      int dx0 = abs(coeff[rc]) * (1 << shift);
 #if CONFIG_HIGHBITDEPTH
       if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
         dx0 >>= xd->bd - 8;
@@ -273,7 +249,9 @@ static int optimize_b_greedy(const AV1_COMMON *cm, MACROBLOCK *mb, int plane,
       dx = (dqcoeff[rc] - coeff[rc]) * (1 << shift);
 #if CONFIG_HIGHBITDEPTH
       if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-        dx >>= xd->bd - 8;
+        int dx_sign = dx < 0 ? 1 : 0;
+        dx = abs(dx) >> (xd->bd - 8);
+        if (dx_sign) dx = -dx;
       }
 #endif  // CONFIG_HIGHBITDEPTH
       d2 = (int64_t)dx * dx;
@@ -329,14 +307,16 @@ static int optimize_b_greedy(const AV1_COMMON *cm, MACROBLOCK *mb, int plane,
       if (i < default_eob - 1) {
         int ctx_next, token_tree_sel_next;
         int band_next = band_translate[i + 1];
+        int token_next =
+            i + 1 != eob ? av1_get_token(qcoeff[scan[i + 1]]) : EOB_TOKEN;
 
         token_cache[rc] = av1_pt_energy_class[t0];
         ctx_next = get_coef_context(nb, token_cache, i + 1);
         token_tree_sel_next = (x == 0);
 
-        next_bits0 = get_token_bit_costs(*(token_costs_ptr + band_next),
-                                         token_tree_sel_next, ctx_next,
-                                         tokens[i + 1][0].token);
+        next_bits0 =
+            get_token_bit_costs(*(token_costs_ptr + band_next),
+                                token_tree_sel_next, ctx_next, token_next);
         next_eob_bits0 =
             get_token_bit_costs(*(token_costs_ptr + band_next),
                                 token_tree_sel_next, ctx_next, EOB_TOKEN);
@@ -345,9 +325,9 @@ static int optimize_b_greedy(const AV1_COMMON *cm, MACROBLOCK *mb, int plane,
         ctx_next = get_coef_context(nb, token_cache, i + 1);
         token_tree_sel_next = (x_a == 0);
 
-        next_bits1 = get_token_bit_costs(*(token_costs_ptr + band_next),
-                                         token_tree_sel_next, ctx_next,
-                                         tokens[i + 1][0].token);
+        next_bits1 =
+            get_token_bit_costs(*(token_costs_ptr + band_next),
+                                token_tree_sel_next, ctx_next, token_next);
 
         if (x_a != 0) {
           next_eob_bits1 =
@@ -356,16 +336,16 @@ static int optimize_b_greedy(const AV1_COMMON *cm, MACROBLOCK *mb, int plane,
         }
       }
 
-      rd_cost0 = RDCOST(rdmult, rddiv, (rate0 + next_bits0), d2);
-      rd_cost1 = RDCOST(rdmult, rddiv, (rate1 + next_bits1), d2_a);
+      rd_cost0 = RDCOST(rdmult, (rate0 + next_bits0), d2);
+      rd_cost1 = RDCOST(rdmult, (rate1 + next_bits1), d2_a);
 
       best_x = (rd_cost1 < rd_cost0);
 
-      eob_cost0 = RDCOST(rdmult, rddiv, (accu_rate + rate0 + next_eob_bits0),
+      eob_cost0 = RDCOST(rdmult, (accu_rate + rate0 + next_eob_bits0),
                          (accu_error + d2 - d0));
       eob_cost1 = eob_cost0;
       if (x_a != 0) {
-        eob_cost1 = RDCOST(rdmult, rddiv, (accu_rate + rate1 + next_eob_bits1),
+        eob_cost1 = RDCOST(rdmult, (accu_rate + rate1 + next_eob_bits1),
                            (accu_error + d2_a - d0));
         best_eob_x = (eob_cost1 < eob_cost0);
       } else {
@@ -410,38 +390,35 @@ static int optimize_b_greedy(const AV1_COMMON *cm, MACROBLOCK *mb, int plane,
 
         token_cache[rc] = av1_pt_energy_class[t0];
       }
+      assert(accu_error >= 0);
 
       x_prev = qcoeff[rc];
 
       // determine whether to move the eob position to i+1
-      int64_t best_eob_cost_i = eob_cost0;
-
-      tokens[i][1].token = t0;
-      tokens[i][1].qc = x;
-      tokens[i][1].dqc = dqc;
-
-      if ((x_a != 0) && (best_eob_x)) {
-        best_eob_cost_i = eob_cost1;
-
-        tokens[i][1].token = t1;
-        tokens[i][1].qc = x_a;
-        tokens[i][1].dqc = dqc_a;
-      }
+      int use_a = (x_a != 0) && (best_eob_x);
+      int64_t best_eob_cost_i = use_a ? eob_cost1 : eob_cost0;
 
       if (best_eob_cost_i < best_block_rd_cost) {
         best_block_rd_cost = best_eob_cost_i;
         final_eob = i + 1;
+        if (use_a) {
+          before_best_eob_qc = x_a;
+          before_best_eob_dqc = dqc_a;
+        } else {
+          before_best_eob_qc = x;
+          before_best_eob_dqc = dqc;
+        }
       }
     }  // if (x==0)
   }    // for (i)
 
   assert(final_eob <= eob);
   if (final_eob > 0) {
-    assert(tokens[final_eob - 1][1].qc != 0);
+    assert(before_best_eob_qc != 0);
     i = final_eob - 1;
     int rc = scan[i];
-    qcoeff[rc] = tokens[i][1].qc;
-    dqcoeff[rc] = tokens[i][1].dqc;
+    qcoeff[rc] = before_best_eob_qc;
+    dqcoeff[rc] = before_best_eob_dqc;
   }
 
   for (i = final_eob; i < eob; i++) {
@@ -453,366 +430,19 @@ static int optimize_b_greedy(const AV1_COMMON *cm, MACROBLOCK *mb, int plane,
   mb->plane[plane].eobs[block] = final_eob;
   return final_eob;
 }
-
-#else  // USE_GREEDY_OPTIMIZE_B
-
-typedef struct av1_token_state_org {
-  int64_t error;
-  int rate;
-  int16_t next;
-  int16_t token;
-  tran_low_t qc;
-  tran_low_t dqc;
-  uint8_t best_index;
-} av1_token_state_org;
-
-static int optimize_b_org(const AV1_COMMON *cm, MACROBLOCK *mb, int plane,
-                          int block, TX_SIZE tx_size, int ctx) {
-  MACROBLOCKD *const xd = &mb->e_mbd;
-  struct macroblock_plane *const p = &mb->plane[plane];
-  struct macroblockd_plane *const pd = &xd->plane[plane];
-  const int ref = is_inter_block(&xd->mi[0]->mbmi);
-  av1_token_state_org tokens[MAX_TX_SQUARE + 1][2];
-  uint8_t token_cache[MAX_TX_SQUARE];
-  const tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
-  tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
-  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
-  const int eob = p->eobs[block];
-  const PLANE_TYPE plane_type = pd->plane_type;
-  const int default_eob = tx_size_2d[tx_size];
-  const int16_t *const dequant_ptr = pd->dequant;
-  const uint8_t *const band_translate = get_band_translate(tx_size);
-  TX_TYPE tx_type = get_tx_type(plane_type, xd, block, tx_size);
-  const SCAN_ORDER *const scan_order =
-      get_scan(cm, tx_size, tx_type, is_inter_block(&xd->mi[0]->mbmi));
-  const int16_t *const scan = scan_order->scan;
-  const int16_t *const nb = scan_order->neighbors;
-  int dqv;
-  const int shift = av1_get_tx_scale(tx_size);
-#if CONFIG_AOM_QM
-  int seg_id = xd->mi[0]->mbmi.segment_id;
-  const qm_val_t *iqmatrix = pd->seg_iqmatrix[seg_id][!ref][tx_size];
-#endif
-#if CONFIG_NEW_QUANT
-  int dq = get_dq_profile_from_ctx(mb->qindex, ctx, ref, plane_type);
-  const dequant_val_type_nuq *dequant_val = pd->dequant_val_nuq[dq];
-#endif  // CONFIG_NEW_QUANT
-  int next = eob, sz = 0;
-  const int64_t rdmult = (mb->rdmult * plane_rd_mult[ref][plane_type]) >> 1;
-  const int64_t rddiv = mb->rddiv;
-  int64_t rd_cost0, rd_cost1;
-  int rate0, rate1;
-  int64_t error0, error1;
-  int16_t t0, t1;
-  int best, band = (eob < default_eob) ? band_translate[eob]
-                                       : band_translate[eob - 1];
-  int pt, i, final_eob;
-  const int cat6_bits = av1_get_cat6_extrabits_size(tx_size, xd->bd);
-  unsigned int(*token_costs)[2][COEFF_CONTEXTS][ENTROPY_TOKENS] =
-      mb->token_costs[txsize_sqr_map[tx_size]][plane_type][ref];
-  const uint16_t *band_counts = &band_count_table[tx_size][band];
-  uint16_t band_left = eob - band_cum_count_table[tx_size][band] + 1;
-  int shortcut = 0;
-  int next_shortcut = 0;
-
-#if CONFIG_EXT_DELTA_Q
-  const int qindex = cm->seg.enabled
-                         ? av1_get_qindex(&cm->seg, xd->mi[0]->mbmi.segment_id,
-                                          cm->base_qindex)
-                         : cm->base_qindex;
-  assert(qindex > 0);
-  (void)qindex;
-#else
-  assert(mb->qindex > 0);
-#endif
-
-  token_costs += band;
-
-  assert((!plane_type && !plane) || (plane_type && plane));
-  assert(eob <= default_eob);
-
-  /* Now set up a Viterbi trellis to evaluate alternative roundings. */
-  /* Initialize the sentinel node of the trellis. */
-  tokens[eob][0].rate = 0;
-  tokens[eob][0].error = 0;
-  tokens[eob][0].next = default_eob;
-  tokens[eob][0].token = EOB_TOKEN;
-  tokens[eob][0].qc = 0;
-  tokens[eob][1] = tokens[eob][0];
-
-  for (i = 0; i < eob; i++) {
-    const int rc = scan[i];
-    tokens[i][0].rate = av1_get_token_cost(qcoeff[rc], &t0, cat6_bits);
-    tokens[i][0].token = t0;
-    token_cache[rc] = av1_pt_energy_class[t0];
-  }
-
-  for (i = eob; i-- > 0;) {
-    int base_bits, dx;
-    int64_t d2;
-    const int rc = scan[i];
-    int x = qcoeff[rc];
-#if CONFIG_AOM_QM
-    int iwt = iqmatrix[rc];
-    dqv = dequant_ptr[rc != 0];
-    dqv = ((iwt * (int)dqv) + (1 << (AOM_QM_BITS - 1))) >> AOM_QM_BITS;
-#else
-    dqv = dequant_ptr[rc != 0];
-#endif
-    next_shortcut = shortcut;
-
-    /* Only add a trellis state for non-zero coefficients. */
-    if (UNLIKELY(x)) {
-      error0 = tokens[next][0].error;
-      error1 = tokens[next][1].error;
-      /* Evaluate the first possibility for this state. */
-      rate0 = tokens[next][0].rate;
-      rate1 = tokens[next][1].rate;
-
-      if (next_shortcut) {
-        /* Consider both possible successor states. */
-        if (next < default_eob) {
-          pt = get_coef_context(nb, token_cache, i + 1);
-          rate0 +=
-              get_token_bit_costs(*token_costs, 0, pt, tokens[next][0].token);
-          rate1 +=
-              get_token_bit_costs(*token_costs, 0, pt, tokens[next][1].token);
-        }
-        UPDATE_RD_COST();
-        /* And pick the best. */
-        best = rd_cost1 < rd_cost0;
-      } else {
-        if (next < default_eob) {
-          pt = get_coef_context(nb, token_cache, i + 1);
-          rate0 +=
-              get_token_bit_costs(*token_costs, 0, pt, tokens[next][0].token);
-        }
-        best = 0;
-      }
-
-      dx = (dqcoeff[rc] - coeff[rc]) * (1 << shift);
-#if CONFIG_HIGHBITDEPTH
-      if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-        dx >>= xd->bd - 8;
-      }
-#endif  // CONFIG_HIGHBITDEPTH
-      d2 = (int64_t)dx * dx;
-      tokens[i][0].rate += (best ? rate1 : rate0);
-      tokens[i][0].error = d2 + (best ? error1 : error0);
-      tokens[i][0].next = next;
-      tokens[i][0].qc = x;
-      tokens[i][0].dqc = dqcoeff[rc];
-      tokens[i][0].best_index = best;
-
-      /* Evaluate the second possibility for this state. */
-      rate0 = tokens[next][0].rate;
-      rate1 = tokens[next][1].rate;
-
-      // The threshold of 3 is empirically obtained.
-      if (UNLIKELY(abs(x) > 3)) {
-        shortcut = 0;
-      } else {
-#if CONFIG_NEW_QUANT
-        shortcut = ((av1_dequant_abscoeff_nuq(abs(x), dqv,
-                                              dequant_val[band_translate[i]]) >
-                     (abs(coeff[rc]) << shift)) &&
-                    (av1_dequant_abscoeff_nuq(abs(x) - 1, dqv,
-                                              dequant_val[band_translate[i]]) <
-                     (abs(coeff[rc]) << shift)));
-#else  // CONFIG_NEW_QUANT
-#if CONFIG_AOM_QM
-        if ((abs(x) * dequant_ptr[rc != 0] * iwt >
-             ((abs(coeff[rc]) << shift) << AOM_QM_BITS)) &&
-            (abs(x) * dequant_ptr[rc != 0] * iwt <
-             (((abs(coeff[rc]) << shift) + dequant_ptr[rc != 0])
-              << AOM_QM_BITS)))
-#else
-        if ((abs(x) * dequant_ptr[rc != 0] > (abs(coeff[rc]) << shift)) &&
-            (abs(x) * dequant_ptr[rc != 0] <
-             (abs(coeff[rc]) << shift) + dequant_ptr[rc != 0]))
-#endif  // CONFIG_AOM_QM
-          shortcut = 1;
-        else
-          shortcut = 0;
-#endif  // CONFIG_NEW_QUANT
-      }
-
-      if (shortcut) {
-        sz = -(x < 0);
-        x -= 2 * sz + 1;
-      } else {
-        tokens[i][1] = tokens[i][0];
-        next = i;
-
-        if (UNLIKELY(!(--band_left))) {
-          --band_counts;
-          band_left = *band_counts;
-          --token_costs;
-        }
-        continue;
-      }
-
-      /* Consider both possible successor states. */
-      if (!x) {
-        /* If we reduced this coefficient to zero, check to see if
-         *  we need to move the EOB back here.
-         */
-        t0 = tokens[next][0].token == EOB_TOKEN ? EOB_TOKEN : ZERO_TOKEN;
-        t1 = tokens[next][1].token == EOB_TOKEN ? EOB_TOKEN : ZERO_TOKEN;
-        base_bits = 0;
-      } else {
-        base_bits = av1_get_token_cost(x, &t0, cat6_bits);
-        t1 = t0;
-      }
-
-      if (next_shortcut) {
-        if (LIKELY(next < default_eob)) {
-          if (t0 != EOB_TOKEN) {
-            token_cache[rc] = av1_pt_energy_class[t0];
-            pt = get_coef_context(nb, token_cache, i + 1);
-            rate0 += get_token_bit_costs(*token_costs, !x, pt,
-                                         tokens[next][0].token);
-          }
-          if (t1 != EOB_TOKEN) {
-            token_cache[rc] = av1_pt_energy_class[t1];
-            pt = get_coef_context(nb, token_cache, i + 1);
-            rate1 += get_token_bit_costs(*token_costs, !x, pt,
-                                         tokens[next][1].token);
-          }
-        }
-
-        UPDATE_RD_COST();
-        /* And pick the best. */
-        best = rd_cost1 < rd_cost0;
-      } else {
-        // The two states in next stage are identical.
-        if (next < default_eob && t0 != EOB_TOKEN) {
-          token_cache[rc] = av1_pt_energy_class[t0];
-          pt = get_coef_context(nb, token_cache, i + 1);
-          rate0 +=
-              get_token_bit_costs(*token_costs, !x, pt, tokens[next][0].token);
-        }
-        best = 0;
-      }
-
-#if CONFIG_NEW_QUANT
-      dx = av1_dequant_coeff_nuq(x, dqv, dequant_val[band_translate[i]]) -
-           (coeff[rc] << shift);
-#if CONFIG_HIGHBITDEPTH
-      if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-        dx >>= xd->bd - 8;
-      }
-#endif  // CONFIG_HIGHBITDEPTH
-#else   // CONFIG_NEW_QUANT
-#if CONFIG_HIGHBITDEPTH
-      if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-        dx -= ((dqv >> (xd->bd - 8)) + sz) ^ sz;
-      } else {
-        dx -= (dqv + sz) ^ sz;
-      }
-#else
-      dx -= (dqv + sz) ^ sz;
-#endif  // CONFIG_HIGHBITDEPTH
-#endif  // CONFIG_NEW_QUANT
-      d2 = (int64_t)dx * dx;
-
-      tokens[i][1].rate = base_bits + (best ? rate1 : rate0);
-      tokens[i][1].error = d2 + (best ? error1 : error0);
-      tokens[i][1].next = next;
-      tokens[i][1].token = best ? t1 : t0;
-      tokens[i][1].qc = x;
-
-      if (x) {
-#if CONFIG_NEW_QUANT
-        tokens[i][1].dqc = av1_dequant_abscoeff_nuq(
-            abs(x), dqv, dequant_val[band_translate[i]]);
-        tokens[i][1].dqc = shift ? ROUND_POWER_OF_TWO(tokens[i][1].dqc, shift)
-                                 : tokens[i][1].dqc;
-        if (sz) tokens[i][1].dqc = -tokens[i][1].dqc;
-#else
-        if (x < 0)
-          tokens[i][1].dqc = -((-x * dqv) >> shift);
-        else
-          tokens[i][1].dqc = (x * dqv) >> shift;
-#endif  // CONFIG_NEW_QUANT
-      } else {
-        tokens[i][1].dqc = 0;
-      }
-
-      tokens[i][1].best_index = best;
-      /* Finally, make this the new head of the trellis. */
-      next = i;
-    } else {
-      /* There's no choice to make for a zero coefficient, so we don't
-       *  add a new trellis node, but we do need to update the costs.
-       */
-      t0 = tokens[next][0].token;
-      t1 = tokens[next][1].token;
-      pt = get_coef_context(nb, token_cache, i + 1);
-      /* Update the cost of each path if we're past the EOB token. */
-      if (t0 != EOB_TOKEN) {
-        tokens[next][0].rate += get_token_bit_costs(*token_costs, 1, pt, t0);
-        tokens[next][0].token = ZERO_TOKEN;
-      }
-      if (t1 != EOB_TOKEN) {
-        tokens[next][1].rate += get_token_bit_costs(*token_costs, 1, pt, t1);
-        tokens[next][1].token = ZERO_TOKEN;
-      }
-      tokens[i][0].best_index = tokens[i][1].best_index = 0;
-      shortcut = (tokens[next][0].rate != tokens[next][1].rate);
-      /* Don't update next, because we didn't add a new node. */
-    }
-
-    if (UNLIKELY(!(--band_left))) {
-      --band_counts;
-      band_left = *band_counts;
-      --token_costs;
-    }
-  }
-
-  /* Now pick the best path through the whole trellis. */
-  rate0 = tokens[next][0].rate;
-  rate1 = tokens[next][1].rate;
-  error0 = tokens[next][0].error;
-  error1 = tokens[next][1].error;
-  t0 = tokens[next][0].token;
-  t1 = tokens[next][1].token;
-  rate0 += get_token_bit_costs(*token_costs, 0, ctx, t0);
-  rate1 += get_token_bit_costs(*token_costs, 0, ctx, t1);
-  UPDATE_RD_COST();
-  best = rd_cost1 < rd_cost0;
-
-  final_eob = -1;
-
-  for (i = next; i < eob; i = next) {
-    const int x = tokens[i][best].qc;
-    const int rc = scan[i];
-    if (x) final_eob = i;
-    qcoeff[rc] = x;
-    dqcoeff[rc] = tokens[i][best].dqc;
-
-    next = tokens[i][best].next;
-    best = tokens[i][best].best_index;
-  }
-  final_eob++;
-
-  mb->plane[plane].eobs[block] = final_eob;
-  assert(final_eob <= default_eob);
-  return final_eob;
-}
-
-#endif  // USE_GREEDY_OPTIMIZE_B
 #endif  // !CONFIG_LV_MAP
 
-int av1_optimize_b(const AV1_COMMON *cm, MACROBLOCK *mb, int plane, int block,
-                   BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
-                   const ENTROPY_CONTEXT *a, const ENTROPY_CONTEXT *l) {
+int av1_optimize_b(const AV1_COMMON *cm, MACROBLOCK *mb, int plane, int blk_row,
+                   int blk_col, int block, BLOCK_SIZE plane_bsize,
+                   TX_SIZE tx_size, const ENTROPY_CONTEXT *a,
+                   const ENTROPY_CONTEXT *l) {
   MACROBLOCKD *const xd = &mb->e_mbd;
   struct macroblock_plane *const p = &mb->plane[plane];
   const int eob = p->eobs[block];
   assert((mb->qindex == 0) ^ (xd->lossless[xd->mi[0]->mbmi.segment_id] == 0));
   if (eob == 0) return eob;
   if (xd->lossless[xd->mi[0]->mbmi.segment_id]) return eob;
+
 #if CONFIG_PVQ
   (void)cm;
   (void)tx_size;
@@ -823,26 +453,24 @@ int av1_optimize_b(const AV1_COMMON *cm, MACROBLOCK *mb, int plane, int block,
 
 #if !CONFIG_LV_MAP
   (void)plane_bsize;
+  (void)blk_row;
+  (void)blk_col;
 #if CONFIG_VAR_TX
   int ctx = get_entropy_context(tx_size, a, l);
 #else
   int ctx = combine_entropy_contexts(*a, *l);
-#endif
-
-#if USE_GREEDY_OPTIMIZE_B
-  return optimize_b_greedy(cm, mb, plane, block, tx_size, ctx);
-#else   // USE_GREEDY_OPTIMIZE_B
-  return optimize_b_org(cm, mb, plane, block, tx_size, ctx);
-#endif  // USE_GREEDY_OPTIMIZE_B
+#endif  // CONFIG_VAR_TX
+  return optimize_b_greedy(cm, mb, plane, blk_row, blk_col, block, tx_size,
+                           ctx);
 #else   // !CONFIG_LV_MAP
   TXB_CTX txb_ctx;
   get_txb_ctx(plane_bsize, tx_size, plane, a, l, &txb_ctx);
-  return av1_optimize_txb(cm, mb, plane, block, tx_size, &txb_ctx);
+  return av1_optimize_txb(cm, mb, plane, blk_row, blk_col, block, tx_size,
+                          &txb_ctx);
 #endif  // !CONFIG_LV_MAP
 }
 
 #if !CONFIG_PVQ
-#if CONFIG_HIGHBITDEPTH
 typedef enum QUANT_FUNC {
   QUANT_FUNC_LOWBD = 0,
   QUANT_FUNC_HIGHBD = 1,
@@ -862,29 +490,12 @@ static AV1_QUANT_FACADE
 #endif  // !CONFIG_NEW_QUANT
       { NULL, NULL }
     };
+#endif  // !CONFIG_PVQ
 
-#else
-
-typedef enum QUANT_FUNC {
-  QUANT_FUNC_LOWBD = 0,
-  QUANT_FUNC_TYPES = 1
-} QUANT_FUNC;
-
-static AV1_QUANT_FACADE quant_func_list[AV1_XFORM_QUANT_TYPES]
-                                       [QUANT_FUNC_TYPES] = {
-#if !CONFIG_NEW_QUANT
-                                         { av1_quantize_fp_facade },
-                                         { av1_quantize_b_facade },
-                                         { av1_quantize_dc_facade },
-#else   // !CONFIG_NEW_QUANT
-                                         { av1_quantize_fp_nuq_facade },
-                                         { av1_quantize_b_nuq_facade },
-                                         { av1_quantize_dc_nuq_facade },
-#endif  // !CONFIG_NEW_QUANT
-                                         { NULL }
-                                       };
-#endif  // CONFIG_HIGHBITDEPTH
-#endif  // CONFIG_PVQ
+typedef void (*fwdTxfmFunc)(const int16_t *diff, tran_low_t *coeff, int stride,
+                            TxfmParam *txfm_param);
+static const fwdTxfmFunc fwd_txfm_func[2] = { av1_fwd_txfm,
+                                              av1_highbd_fwd_txfm };
 
 void av1_xform_quant(const AV1_COMMON *cm, MACROBLOCK *x, int plane, int block,
                      int blk_row, int blk_col, BLOCK_SIZE plane_bsize,
@@ -892,7 +503,7 @@ void av1_xform_quant(const AV1_COMMON *cm, MACROBLOCK *x, int plane, int block,
                      AV1_XFORM_QUANT xform_quant_idx) {
   MACROBLOCKD *const xd = &x->e_mbd;
   MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
-#if !(CONFIG_PVQ || CONFIG_DAALA_DIST)
+#if !(CONFIG_PVQ || CONFIG_DIST_8X8)
   const struct macroblock_plane *const p = &x->plane[plane];
   const struct macroblockd_plane *const pd = &xd->plane[plane];
 #else
@@ -900,9 +511,14 @@ void av1_xform_quant(const AV1_COMMON *cm, MACROBLOCK *x, int plane, int block,
   struct macroblockd_plane *const pd = &xd->plane[plane];
 #endif
   PLANE_TYPE plane_type = get_plane_type(plane);
-  TX_TYPE tx_type = get_tx_type(plane_type, xd, block, tx_size);
+  TX_TYPE tx_type =
+      av1_get_tx_type(plane_type, xd, blk_row, blk_col, block, tx_size);
+
+#if CONFIG_AOM_QM || CONFIG_NEW_QUANT
   const int is_inter = is_inter_block(mbmi);
-  const SCAN_ORDER *const scan_order = get_scan(cm, tx_size, tx_type, is_inter);
+#endif
+
+  const SCAN_ORDER *const scan_order = get_scan(cm, tx_size, tx_type, mbmi);
   tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
   tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
   tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
@@ -910,19 +526,28 @@ void av1_xform_quant(const AV1_COMMON *cm, MACROBLOCK *x, int plane, int block,
   const int diff_stride = block_size_wide[plane_bsize];
 #if CONFIG_AOM_QM
   int seg_id = mbmi->segment_id;
-  const qm_val_t *qmatrix = pd->seg_qmatrix[seg_id][!is_inter][tx_size];
-  const qm_val_t *iqmatrix = pd->seg_iqmatrix[seg_id][!is_inter][tx_size];
+  // Use a flat matrix (i.e. no weighting) for 1D and Identity transforms
+  const qm_val_t *qmatrix =
+      IS_2D_TRANSFORM(tx_type) ? pd->seg_qmatrix[seg_id][!is_inter][tx_size]
+                               : cm->gqmatrix[NUM_QM_LEVELS - 1][0][0][tx_size];
+  const qm_val_t *iqmatrix =
+      IS_2D_TRANSFORM(tx_type)
+          ? pd->seg_iqmatrix[seg_id][!is_inter][tx_size]
+          : cm->giqmatrix[NUM_QM_LEVELS - 1][0][0][tx_size];
 #endif
 
-  FWD_TXFM_PARAM fwd_txfm_param;
+  TxfmParam txfm_param;
 
-#if CONFIG_PVQ || CONFIG_DAALA_DIST
+#if CONFIG_PVQ || CONFIG_DIST_8X8 || CONFIG_LGT || CONFIG_MRC_TX
   uint8_t *dst;
-  int16_t *pred;
   const int dst_stride = pd->dst.stride;
-  int tx_blk_size;
+#if CONFIG_PVQ || CONFIG_DIST_8X8
+  int16_t *pred;
+  const int txw = tx_size_wide[tx_size];
+  const int txh = tx_size_high[tx_size];
   int i, j;
 #endif
+#endif
 
 #if !CONFIG_PVQ
   const int tx2d_size = tx_size_2d[tx_size];
@@ -960,79 +585,68 @@ void av1_xform_quant(const AV1_COMMON *cm, MACROBLOCK *x, int plane, int block,
   src_int16 =
       &p->src_int16[(blk_row * diff_stride + blk_col) << tx_size_wide_log2[0]];
 
-  // transform block size in pixels
-  tx_blk_size = tx_size_wide[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++)
+    for (j = 0; j < txh; j++)
+      for (i = 0; i < txw; i++)
         src_int16[diff_stride * j + i] =
             CONVERT_TO_SHORTPTR(src)[src_stride * j + i];
   } else {
 #endif  // CONFIG_HIGHBITDEPTH
-    for (j = 0; j < tx_blk_size; j++)
-      for (i = 0; i < tx_blk_size; i++)
+    for (j = 0; j < txh; j++)
+      for (i = 0; i < txw; i++)
         src_int16[diff_stride * j + i] = src[src_stride * j + i];
 #if CONFIG_HIGHBITDEPTH
   }
 #endif  // CONFIG_HIGHBITDEPTH
 #endif
 
-#if CONFIG_PVQ || CONFIG_DAALA_DIST
+#if CONFIG_PVQ || CONFIG_DIST_8X8 || CONFIG_LGT || CONFIG_MRC_TX
   dst = &pd->dst.buf[(blk_row * dst_stride + blk_col) << tx_size_wide_log2[0]];
+#if CONFIG_PVQ || CONFIG_DIST_8X8
   pred = &pd->pred[(blk_row * diff_stride + blk_col) << tx_size_wide_log2[0]];
 
-  // transform block size in pixels
-  tx_blk_size = tx_size_wide[tx_size];
-
 // copy uint8 orig and predicted block to int16 buffer
 // in order to use existing VP10 transform functions
 #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++)
+    for (j = 0; j < txh; j++)
+      for (i = 0; i < txw; i++)
         pred[diff_stride * j + i] =
             CONVERT_TO_SHORTPTR(dst)[dst_stride * j + i];
   } else {
 #endif  // CONFIG_HIGHBITDEPTH
-    for (j = 0; j < tx_blk_size; j++)
-      for (i = 0; i < tx_blk_size; i++)
+    for (j = 0; j < txh; j++)
+      for (i = 0; i < txw; i++)
         pred[diff_stride * j + i] = dst[dst_stride * j + i];
 #if CONFIG_HIGHBITDEPTH
   }
 #endif  // CONFIG_HIGHBITDEPTH
-#endif
+#endif  // CONFIG_PVQ || CONFIG_DIST_8X8
+#endif  // CONFIG_PVQ || CONFIG_DIST_8X8 || CONFIG_LGT || CONFIG_MRC_TX
 
   (void)ctx;
 
-  fwd_txfm_param.tx_type = tx_type;
-  fwd_txfm_param.tx_size = tx_size;
-  fwd_txfm_param.lossless = xd->lossless[mbmi->segment_id];
+  txfm_param.tx_type = tx_type;
+  txfm_param.tx_size = tx_size;
+  txfm_param.lossless = xd->lossless[mbmi->segment_id];
+#if CONFIG_MRC_TX || CONFIG_LGT
+  txfm_param.dst = dst;
+  txfm_param.stride = dst_stride;
+#endif  // CONFIG_MRC_TX || CONFIG_LGT
+#if CONFIG_LGT
+  txfm_param.is_inter = is_inter_block(mbmi);
+  txfm_param.mode = get_prediction_mode(xd->mi[0], plane, tx_size, block);
+#endif
 
 #if !CONFIG_PVQ
-#if CONFIG_HIGHBITDEPTH
-  fwd_txfm_param.bd = xd->bd;
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-    av1_highbd_fwd_txfm(src_diff, coeff, diff_stride, &fwd_txfm_param);
-    if (xform_quant_idx != AV1_XFORM_QUANT_SKIP_QUANT) {
-      if (LIKELY(!x->skip_block)) {
-        quant_func_list[xform_quant_idx][QUANT_FUNC_HIGHBD](
-            coeff, tx2d_size, p, qcoeff, pd, dqcoeff, eob, scan_order, &qparam);
-      } else {
-        av1_quantize_skip(tx2d_size, qcoeff, dqcoeff, eob);
-      }
-    }
-#if CONFIG_LV_MAP
-    p->txb_entropy_ctx[block] =
-        (uint8_t)av1_get_txb_entropy_context(qcoeff, scan_order, *eob);
-#endif  // CONFIG_LV_MAP
-    return;
-  }
-#endif  // CONFIG_HIGHBITDEPTH
-  av1_fwd_txfm(src_diff, coeff, diff_stride, &fwd_txfm_param);
+  txfm_param.bd = xd->bd;
+  const int is_hbd = get_bitdepth_data_path_index(xd);
+  fwd_txfm_func[is_hbd](src_diff, coeff, diff_stride, &txfm_param);
+
   if (xform_quant_idx != AV1_XFORM_QUANT_SKIP_QUANT) {
     if (LIKELY(!x->skip_block)) {
-      quant_func_list[xform_quant_idx][QUANT_FUNC_LOWBD](
+      quant_func_list[xform_quant_idx][is_hbd](
           coeff, tx2d_size, p, qcoeff, pd, dqcoeff, eob, scan_order, &qparam);
     } else {
       av1_quantize_skip(tx2d_size, qcoeff, dqcoeff, eob);
@@ -1042,17 +656,18 @@ void av1_xform_quant(const AV1_COMMON *cm, MACROBLOCK *x, int plane, int block,
   p->txb_entropy_ctx[block] =
       (uint8_t)av1_get_txb_entropy_context(qcoeff, scan_order, *eob);
 #endif  // CONFIG_LV_MAP
-#else   // #if !CONFIG_PVQ
+  return;
+#else  // CONFIG_PVQ
   (void)xform_quant_idx;
 #if CONFIG_HIGHBITDEPTH
-  fwd_txfm_param.bd = xd->bd;
+  txfm_param.bd = xd->bd;
   if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-    av1_highbd_fwd_txfm(src_int16, coeff, diff_stride, &fwd_txfm_param);
-    av1_highbd_fwd_txfm(pred, ref_coeff, diff_stride, &fwd_txfm_param);
+    av1_highbd_fwd_txfm(src_int16, coeff, diff_stride, &txfm_param);
+    av1_highbd_fwd_txfm(pred, ref_coeff, diff_stride, &txfm_param);
   } else {
 #endif
-    av1_fwd_txfm(src_int16, coeff, diff_stride, &fwd_txfm_param);
-    av1_fwd_txfm(pred, ref_coeff, diff_stride, &fwd_txfm_param);
+    av1_fwd_txfm(src_int16, coeff, diff_stride, &txfm_param);
+    av1_fwd_txfm(pred, ref_coeff, diff_stride, &txfm_param);
 #if CONFIG_HIGHBITDEPTH
   }
 #endif
@@ -1130,7 +745,8 @@ static void encode_block(int plane, int block, int blk_row, int blk_col,
 #endif
 
 #if !CONFIG_PVQ
-  av1_optimize_b(cm, x, plane, block, plane_bsize, tx_size, a, l);
+  av1_optimize_b(cm, x, plane, blk_row, blk_col, block, plane_bsize, tx_size, a,
+                 l);
 
   av1_set_txb_context(x, plane, block, tx_size, a, l);
 
@@ -1143,9 +759,16 @@ static void encode_block(int plane, int block, int blk_row, int blk_col,
 
   if (x->pvq_skip[plane]) return;
 #endif
-  TX_TYPE tx_type = get_tx_type(pd->plane_type, xd, block, tx_size);
+  TX_TYPE tx_type =
+      av1_get_tx_type(pd->plane_type, xd, blk_row, blk_col, block, tx_size);
+#if CONFIG_LGT
+  PREDICTION_MODE mode = get_prediction_mode(xd->mi[0], plane, tx_size, block);
+  av1_inverse_transform_block(xd, dqcoeff, mode, tx_type, tx_size, dst,
+                              pd->dst.stride, p->eobs[block]);
+#else
   av1_inverse_transform_block(xd, dqcoeff, tx_type, tx_size, dst,
                               pd->dst.stride, p->eobs[block]);
+#endif
 }
 
 #if CONFIG_VAR_TX
@@ -1174,16 +797,32 @@ static void encode_block_inter(int plane, int block, int blk_row, int blk_col,
     encode_block(plane, block, blk_row, blk_col, plane_bsize, tx_size, arg);
   } else {
     assert(tx_size < TX_SIZES_ALL);
+#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(sub_txs < tx_size);
+#endif
     // This is the square transform block partition entry point.
     int bsl = tx_size_wide_unit[sub_txs];
     int i;
     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
       int step = tx_size_wide_unit[sub_txs] * tx_size_high_unit[sub_txs];
 
       if (offsetr >= max_blocks_high || offsetc >= max_blocks_wide) continue;
@@ -1211,6 +850,7 @@ static void encode_block_pass1(int plane, int block, int blk_row, int blk_col,
   struct macroblock_plane *const p = &x->plane[plane];
   struct macroblockd_plane *const pd = &xd->plane[plane];
   tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
+  TxfmParam txfm_param;
   uint8_t *dst;
   int ctx = 0;
   dst = &pd->dst
@@ -1246,22 +886,20 @@ static void encode_block_pass1(int plane, int block, int blk_row, int blk_col,
 #endif  // CONFIG_HIGHBITDEPTH
     }
 #endif  // !CONFIG_PVQ
+    txfm_param.bd = xd->bd;
+    txfm_param.tx_type = DCT_DCT;
+    txfm_param.eob = p->eobs[block];
+    txfm_param.lossless = xd->lossless[xd->mi[0]->mbmi.segment_id];
 #if CONFIG_HIGHBITDEPTH
     if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-      if (xd->lossless[xd->mi[0]->mbmi.segment_id]) {
-        av1_highbd_iwht4x4_add(dqcoeff, dst, pd->dst.stride, p->eobs[block],
-                               xd->bd);
-      } else {
-        av1_highbd_idct4x4_add(dqcoeff, dst, pd->dst.stride, p->eobs[block],
-                               xd->bd);
-      }
+      av1_highbd_inv_txfm_add_4x4(dqcoeff, dst, pd->dst.stride, &txfm_param);
       return;
     }
 #endif  //  CONFIG_HIGHBITDEPTH
     if (xd->lossless[xd->mi[0]->mbmi.segment_id]) {
-      av1_iwht4x4_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
+      av1_iwht4x4_add(dqcoeff, dst, pd->dst.stride, &txfm_param);
     } else {
-      av1_idct4x4_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
+      av1_idct4x4_add(dqcoeff, dst, pd->dst.stride, &txfm_param);
     }
   }
 }
@@ -1316,7 +954,7 @@ void av1_encode_sb(AV1_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE bsize, int mi_row,
     av1_get_entropy_contexts(bsize, 0, pd, ctx.ta[plane], ctx.tl[plane]);
 #else
     const struct macroblockd_plane *const pd = &xd->plane[plane];
-    const TX_SIZE tx_size = get_tx_size(plane, xd);
+    const TX_SIZE tx_size = av1_get_tx_size(plane, xd);
     av1_get_entropy_contexts(bsize, tx_size, pd, ctx.ta[plane], ctx.tl[plane]);
 #endif
 
@@ -1327,11 +965,27 @@ void av1_encode_sb(AV1_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE bsize, int mi_row,
     arg.tl = ctx.tl[plane];
 
 #if CONFIG_VAR_TX
-    for (idy = 0; idy < mi_height; idy += bh) {
-      for (idx = 0; idx < mi_width; idx += bw) {
-        encode_block_inter(plane, block, idy, idx, plane_bsize, max_tx_size,
-                           &arg);
-        block += step;
+    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(mi_width, mu_blocks_wide);
+    mu_blocks_high = AOMMIN(mi_height, mu_blocks_high);
+
+    for (idy = 0; idy < mi_height; idy += mu_blocks_high) {
+      for (idx = 0; idx < mi_width; idx += mu_blocks_wide) {
+        int blk_row, blk_col;
+        const int unit_height = AOMMIN(mu_blocks_high + idy, mi_height);
+        const int unit_width = AOMMIN(mu_blocks_wide + idx, mi_width);
+        for (blk_row = idy; blk_row < unit_height; blk_row += bh) {
+          for (blk_col = idx; blk_col < unit_width; blk_col += bw) {
+            encode_block_inter(plane, block, blk_row, blk_col, plane_bsize,
+                               max_tx_size, &arg);
+            block += step;
+          }
+        }
       }
     }
 #else
@@ -1357,7 +1011,7 @@ void av1_encode_sb_supertx(AV1_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE bsize) {
 #if CONFIG_VAR_TX
     const TX_SIZE tx_size = TX_4X4;
 #else
-    const TX_SIZE tx_size = get_tx_size(plane, xd);
+    const TX_SIZE tx_size = av1_get_tx_size(plane, xd);
 #endif
     av1_subtract_plane(x, bsize, plane);
     av1_get_entropy_contexts(bsize, tx_size, pd, ctx.ta[plane], ctx.tl[plane]);
@@ -1435,6 +1089,24 @@ static void quantize_scaler(int coeff, int16_t zbin, int16_t round_value,
   }
 }
 
+#if CONFIG_HIGHBITDEPTH
+typedef void (*hbd_dpcm_fwd_tx_func)(const int16_t *input, int stride,
+                                     TX_TYPE_1D tx_type, tran_low_t *output,
+                                     int dir);
+
+static hbd_dpcm_fwd_tx_func get_hbd_dpcm_fwd_tx_func(int tx_length) {
+  switch (tx_length) {
+    case 4: return av1_hbd_dpcm_ft4_c;
+    case 8: return av1_hbd_dpcm_ft8_c;
+    case 16: return av1_hbd_dpcm_ft16_c;
+    case 32:
+      return av1_hbd_dpcm_ft32_c;
+    // TODO(huisu): add support for TX_64X64.
+    default: assert(0); return NULL;
+  }
+}
+#endif  // CONFIG_HIGHBITDEPTH
+
 typedef void (*dpcm_fwd_tx_func)(const int16_t *input, int stride,
                                  TX_TYPE_1D tx_type, tran_low_t *output);
 
@@ -1539,7 +1211,7 @@ static void hbd_process_block_dpcm_vert(
     int16_t *src_diff, int diff_stride, tran_low_t *coeff, tran_low_t *qcoeff,
     tran_low_t *dqcoeff) {
   const int tx1d_width = tx_size_wide[tx_size];
-  dpcm_fwd_tx_func forward_tx = get_dpcm_fwd_tx_func(tx1d_width);
+  hbd_dpcm_fwd_tx_func forward_tx = get_hbd_dpcm_fwd_tx_func(tx1d_width);
   hbd_dpcm_inv_txfm_add_func inverse_tx =
       av1_get_hbd_dpcm_inv_txfm_add_func(tx1d_width);
   uint16_t *src = CONVERT_TO_SHORTPTR(src8);
@@ -1553,7 +1225,7 @@ static void hbd_process_block_dpcm_vert(
     // Subtraction.
     for (int c = 0; c < tx1d_width; ++c) src_diff[c] = src[c] - dst[c];
     // Forward transform.
-    forward_tx(src_diff, 1, tx_type_1d, coeff);
+    forward_tx(src_diff, 1, tx_type_1d, coeff, 1);
     // Quantization.
     for (int c = 0; c < tx1d_width; ++c) {
       quantize_scaler(coeff[c], p->zbin[q_idx], p->round[q_idx],
@@ -1562,7 +1234,7 @@ static void hbd_process_block_dpcm_vert(
       q_idx = 1;
     }
     // Inverse transform.
-    inverse_tx(dqcoeff, 1, tx_type_1d, bd, dst);
+    inverse_tx(dqcoeff, 1, tx_type_1d, bd, dst, 1);
     // Move to the next row.
     coeff += tx1d_width;
     qcoeff += tx1d_width;
@@ -1580,7 +1252,7 @@ static void hbd_process_block_dpcm_horz(
     int16_t *src_diff, int diff_stride, tran_low_t *coeff, tran_low_t *qcoeff,
     tran_low_t *dqcoeff) {
   const int tx1d_height = tx_size_high[tx_size];
-  dpcm_fwd_tx_func forward_tx = get_dpcm_fwd_tx_func(tx1d_height);
+  hbd_dpcm_fwd_tx_func forward_tx = get_hbd_dpcm_fwd_tx_func(tx1d_height);
   hbd_dpcm_inv_txfm_add_func inverse_tx =
       av1_get_hbd_dpcm_inv_txfm_add_func(tx1d_height);
   uint16_t *src = CONVERT_TO_SHORTPTR(src8);
@@ -1597,7 +1269,7 @@ static void hbd_process_block_dpcm_horz(
     }
     // Forward transform.
     tran_low_t tx_buff[64];
-    forward_tx(src_diff, diff_stride, tx_type_1d, tx_buff);
+    forward_tx(src_diff, diff_stride, tx_type_1d, tx_buff, 0);
     for (int r = 0; r < tx1d_height; ++r) coeff[r * tx1d_width] = tx_buff[r];
     // Quantization.
     for (int r = 0; r < tx1d_height; ++r) {
@@ -1609,7 +1281,7 @@ static void hbd_process_block_dpcm_horz(
     }
     // Inverse transform.
     for (int r = 0; r < tx1d_height; ++r) tx_buff[r] = dqcoeff[r * tx1d_width];
-    inverse_tx(tx_buff, dst_stride, tx_type_1d, bd, dst);
+    inverse_tx(tx_buff, dst_stride, tx_type_1d, bd, dst, 0);
     // Move to the next column.
     ++coeff, ++qcoeff, ++dqcoeff, ++src_diff, ++dst, ++src;
   }
@@ -1631,7 +1303,8 @@ void av1_encode_block_intra_dpcm(const AV1_COMMON *cm, MACROBLOCK *x,
   const int dst_stride = pd->dst.stride;
   const int tx1d_width = tx_size_wide[tx_size];
   const int tx1d_height = tx_size_high[tx_size];
-  const SCAN_ORDER *const scan_order = get_scan(cm, tx_size, tx_type, 0);
+  const SCAN_ORDER *const scan_order =
+      get_scan(cm, tx_size, tx_type, &xd->mi[0]->mbmi);
   tran_low_t *coeff = BLOCK_OFFSET(p->coeff, block);
   tran_low_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
   uint8_t *dst =
@@ -1711,30 +1384,20 @@ void av1_encode_block_intra(int plane, int block, int blk_row, int blk_col,
   struct macroblockd_plane *const pd = &xd->plane[plane];
   tran_low_t *dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
   PLANE_TYPE plane_type = get_plane_type(plane);
-  const TX_TYPE tx_type = get_tx_type(plane_type, xd, block, tx_size);
+  const TX_TYPE tx_type =
+      av1_get_tx_type(plane_type, xd, blk_row, blk_col, block, tx_size);
   uint16_t *eob = &p->eobs[block];
   const int dst_stride = pd->dst.stride;
   uint8_t *dst =
       &pd->dst.buf[(blk_row * dst_stride + blk_col) << tx_size_wide_log2[0]];
-#if CONFIG_CFL
-
-#if CONFIG_EC_ADAPT
-  FRAME_CONTEXT *const ec_ctx = xd->tile_ctx;
-#else
-  FRAME_CONTEXT *const ec_ctx = cm->fc;
-#endif  // CONFIG_EC_ADAPT
 
-  av1_predict_intra_block_encoder_facade(x, ec_ctx, plane, block, blk_col,
-                                         blk_row, tx_size, plane_bsize);
-#else
   av1_predict_intra_block_facade(xd, plane, block, blk_col, blk_row, tx_size);
-#endif
 
+#if CONFIG_DPCM_INTRA || CONFIG_LGT
+  const PREDICTION_MODE mode =
+      get_prediction_mode(xd->mi[0], plane, tx_size, block);
 #if CONFIG_DPCM_INTRA
-  const int block_raster_idx = av1_block_index_to_raster_order(tx_size, block);
   const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
-  const PREDICTION_MODE mode =
-      (plane == 0) ? get_y_mode(xd->mi[0], block_raster_idx) : mbmi->uv_mode;
   if (av1_use_dpcm_intra(plane, mode, tx_type, mbmi)) {
     av1_encode_block_intra_dpcm(cm, x, mode, plane, block, blk_row, blk_col,
                                 plane_bsize, tx_size, tx_type, args->ta,
@@ -1742,6 +1405,7 @@ void av1_encode_block_intra(int plane, int block, int blk_row, int blk_col,
     return;
   }
 #endif  // CONFIG_DPCM_INTRA
+#endif  // CONFIG_DPCM_INTRA || CONFIG_LGT
 
   av1_subtract_txb(x, plane, plane_bsize, blk_col, blk_row, tx_size);
 
@@ -1751,7 +1415,8 @@ void av1_encode_block_intra(int plane, int block, int blk_row, int blk_col,
   if (args->enable_optimize_b) {
     av1_xform_quant(cm, x, plane, block, blk_row, blk_col, plane_bsize, tx_size,
                     ctx, AV1_XFORM_QUANT_FP);
-    av1_optimize_b(cm, x, plane, block, plane_bsize, tx_size, a, l);
+    av1_optimize_b(cm, x, plane, blk_row, blk_col, block, plane_bsize, tx_size,
+                   a, l);
   } else {
     av1_xform_quant(cm, x, plane, block, blk_row, blk_col, plane_bsize, tx_size,
                     ctx, AV1_XFORM_QUANT_B);
@@ -1763,220 +1428,25 @@ void av1_encode_block_intra(int plane, int block, int blk_row, int blk_col,
 
   if (x->pvq_skip[plane]) return;
 #endif  // CONFIG_PVQ
-  av1_inverse_transform_block(xd, dqcoeff, tx_type, tx_size, dst, dst_stride,
-                              *eob);
+  av1_inverse_transform_block(xd, dqcoeff,
+#if CONFIG_LGT
+                              mode,
+#endif
+                              tx_type, tx_size, dst, dst_stride, *eob);
 #if !CONFIG_PVQ
   if (*eob) *(args->skip) = 0;
 #else
 // Note : *(args->skip) == mbmi->skip
 #endif
 #if CONFIG_CFL
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
   if (plane == AOM_PLANE_Y && x->cfl_store_y) {
-    cfl_store(xd->cfl, dst, dst_stride, blk_row, blk_col, tx_size);
-  }
-
-  if (mbmi->uv_mode == DC_PRED) {
-    // TODO(ltrudeau) find a cleaner way to detect last transform block
-    if (plane == AOM_PLANE_U) {
-      xd->cfl->num_tx_blk[CFL_PRED_U] =
-          (blk_row == 0 && blk_col == 0) ? 1
-                                         : xd->cfl->num_tx_blk[CFL_PRED_U] + 1;
-    }
-
-    if (plane == AOM_PLANE_V) {
-      xd->cfl->num_tx_blk[CFL_PRED_V] =
-          (blk_row == 0 && blk_col == 0) ? 1
-                                         : xd->cfl->num_tx_blk[CFL_PRED_V] + 1;
-
-      if (mbmi->skip &&
-          xd->cfl->num_tx_blk[CFL_PRED_U] == xd->cfl->num_tx_blk[CFL_PRED_V]) {
-        assert(plane_bsize != BLOCK_INVALID);
-        const int block_width = block_size_wide[plane_bsize];
-        const int block_height = block_size_high[plane_bsize];
-
-        // if SKIP is chosen at the block level, and ind != 0, we must change
-        // the prediction
-        if (mbmi->cfl_alpha_idx != 0) {
-          const struct macroblockd_plane *const pd_cb = &xd->plane[AOM_PLANE_U];
-          uint8_t *const dst_cb = pd_cb->dst.buf;
-          const int dst_stride_cb = pd_cb->dst.stride;
-          uint8_t *const dst_cr = pd->dst.buf;
-          const int dst_stride_cr = pd->dst.stride;
-          for (int j = 0; j < block_height; j++) {
-            for (int i = 0; i < block_width; i++) {
-              dst_cb[dst_stride_cb * j + i] =
-                  (uint8_t)(xd->cfl->dc_pred[CFL_PRED_U] + 0.5);
-              dst_cr[dst_stride_cr * j + i] =
-                  (uint8_t)(xd->cfl->dc_pred[CFL_PRED_V] + 0.5);
-            }
-          }
-          mbmi->cfl_alpha_idx = 0;
-          mbmi->cfl_alpha_signs[CFL_PRED_U] = CFL_SIGN_POS;
-          mbmi->cfl_alpha_signs[CFL_PRED_V] = CFL_SIGN_POS;
-        }
-      }
-    }
+    // TODO (ltrudeau) Store sub-8x8 inter blocks when bottom right block is
+    // intra predicted.
+    cfl_store(xd->cfl, dst, dst_stride, blk_row, blk_col, tx_size, plane_bsize);
   }
 #endif
 }
 
-#if CONFIG_CFL
-static int cfl_alpha_dist(const uint8_t *y_pix, int y_stride, double y_avg,
-                          const uint8_t *src, int src_stride, int blk_width,
-                          int blk_height, double dc_pred, double alpha,
-                          int *dist_neg_out) {
-  const double dc_pred_bias = dc_pred + 0.5;
-  int dist = 0;
-  int diff;
-
-  if (alpha == 0.0) {
-    const int dc_pred_i = (int)dc_pred_bias;
-    for (int j = 0; j < blk_height; j++) {
-      for (int i = 0; i < blk_width; i++) {
-        diff = src[i] - dc_pred_i;
-        dist += diff * diff;
-      }
-      src += src_stride;
-    }
-
-    if (dist_neg_out) *dist_neg_out = dist;
-
-    return dist;
-  }
-
-  int dist_neg = 0;
-  for (int j = 0; j < blk_height; j++) {
-    for (int i = 0; i < blk_width; i++) {
-      const double scaled_luma = alpha * (y_pix[i] - y_avg);
-      const int uv = src[i];
-      diff = uv - (int)(scaled_luma + dc_pred_bias);
-      dist += diff * diff;
-      diff = uv + (int)(scaled_luma - dc_pred_bias);
-      dist_neg += diff * diff;
-    }
-    y_pix += y_stride;
-    src += src_stride;
-  }
-
-  if (dist_neg_out) *dist_neg_out = dist_neg;
-
-  return dist;
-}
-
-static int cfl_compute_alpha_ind(MACROBLOCK *const x, const CFL_CTX *const cfl,
-                                 BLOCK_SIZE bsize,
-                                 CFL_SIGN_TYPE signs_out[CFL_SIGNS]) {
-  const struct macroblock_plane *const p_u = &x->plane[AOM_PLANE_U];
-  const struct macroblock_plane *const p_v = &x->plane[AOM_PLANE_V];
-  const uint8_t *const src_u = p_u->src.buf;
-  const uint8_t *const src_v = p_v->src.buf;
-  const int src_stride_u = p_u->src.stride;
-  const int src_stride_v = p_v->src.stride;
-  const int block_width = block_size_wide[bsize];
-  const int block_height = block_size_high[bsize];
-  const double dc_pred_u = cfl->dc_pred[CFL_PRED_U];
-  const double dc_pred_v = cfl->dc_pred[CFL_PRED_V];
-
-  // Temporary pixel buffer used to store the CfL prediction when we compute the
-  // alpha index.
-  uint8_t tmp_pix[MAX_SB_SQUARE];
-  // Load CfL Prediction over the entire block
-  const double y_avg =
-      cfl_load(cfl, tmp_pix, MAX_SB_SIZE, 0, 0, block_width, block_height);
-
-  int sse[CFL_PRED_PLANES][CFL_MAGS_SIZE];
-  sse[CFL_PRED_U][0] =
-      cfl_alpha_dist(tmp_pix, MAX_SB_SIZE, y_avg, src_u, src_stride_u,
-                     block_width, block_height, dc_pred_u, 0, NULL);
-  sse[CFL_PRED_V][0] =
-      cfl_alpha_dist(tmp_pix, MAX_SB_SIZE, y_avg, src_v, src_stride_v,
-                     block_width, block_height, dc_pred_v, 0, NULL);
-  for (int m = 1; m < CFL_MAGS_SIZE; m += 2) {
-    assert(cfl_alpha_mags[m + 1] == -cfl_alpha_mags[m]);
-    sse[CFL_PRED_U][m] = cfl_alpha_dist(
-        tmp_pix, MAX_SB_SIZE, y_avg, src_u, src_stride_u, block_width,
-        block_height, dc_pred_u, cfl_alpha_mags[m], &sse[CFL_PRED_U][m + 1]);
-    sse[CFL_PRED_V][m] = cfl_alpha_dist(
-        tmp_pix, MAX_SB_SIZE, y_avg, src_v, src_stride_v, block_width,
-        block_height, dc_pred_v, cfl_alpha_mags[m], &sse[CFL_PRED_V][m + 1]);
-  }
-
-  int dist;
-  int64_t cost;
-  int64_t best_cost;
-
-  // Compute least squares parameter of the entire block
-  // IMPORTANT: We assume that the first code is 0,0
-  int ind = 0;
-  signs_out[CFL_PRED_U] = CFL_SIGN_POS;
-  signs_out[CFL_PRED_V] = CFL_SIGN_POS;
-
-  dist = sse[CFL_PRED_U][0] + sse[CFL_PRED_V][0];
-  dist *= 16;
-  best_cost = RDCOST(x->rdmult, x->rddiv, cfl->costs[0], dist);
-
-  for (int c = 1; c < CFL_ALPHABET_SIZE; c++) {
-    const int idx_u = cfl_alpha_codes[c][CFL_PRED_U];
-    const int idx_v = cfl_alpha_codes[c][CFL_PRED_V];
-    for (CFL_SIGN_TYPE sign_u = idx_u == 0; sign_u < CFL_SIGNS; sign_u++) {
-      for (CFL_SIGN_TYPE sign_v = idx_v == 0; sign_v < CFL_SIGNS; sign_v++) {
-        dist = sse[CFL_PRED_U][idx_u + (sign_u == CFL_SIGN_NEG)] +
-               sse[CFL_PRED_V][idx_v + (sign_v == CFL_SIGN_NEG)];
-        dist *= 16;
-        cost = RDCOST(x->rdmult, x->rddiv, cfl->costs[c], dist);
-        if (cost < best_cost) {
-          best_cost = cost;
-          ind = c;
-          signs_out[CFL_PRED_U] = sign_u;
-          signs_out[CFL_PRED_V] = sign_v;
-        }
-      }
-    }
-  }
-
-  return ind;
-}
-
-static inline void cfl_update_costs(CFL_CTX *cfl, FRAME_CONTEXT *ec_ctx) {
-  assert(ec_ctx->cfl_alpha_cdf[CFL_ALPHABET_SIZE - 1] ==
-         AOM_ICDF(CDF_PROB_TOP));
-  const int prob_den = CDF_PROB_TOP;
-
-  int prob_num = AOM_ICDF(ec_ctx->cfl_alpha_cdf[0]);
-  cfl->costs[0] = av1_cost_zero(get_prob(prob_num, prob_den));
-
-  for (int c = 1; c < CFL_ALPHABET_SIZE; c++) {
-    int sign_bit_cost = (cfl_alpha_codes[c][CFL_PRED_U] != 0) +
-                        (cfl_alpha_codes[c][CFL_PRED_V] != 0);
-    prob_num = AOM_ICDF(ec_ctx->cfl_alpha_cdf[c]) -
-               AOM_ICDF(ec_ctx->cfl_alpha_cdf[c - 1]);
-    cfl->costs[c] = av1_cost_zero(get_prob(prob_num, prob_den)) +
-                    av1_cost_literal(sign_bit_cost);
-  }
-}
-
-void av1_predict_intra_block_encoder_facade(MACROBLOCK *x,
-                                            FRAME_CONTEXT *ec_ctx, int plane,
-                                            int block_idx, int blk_col,
-                                            int blk_row, TX_SIZE tx_size,
-                                            BLOCK_SIZE plane_bsize) {
-  MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
-  if (plane != AOM_PLANE_Y && mbmi->uv_mode == DC_PRED) {
-    if (blk_col == 0 && blk_row == 0 && plane == AOM_PLANE_U) {
-      CFL_CTX *const cfl = xd->cfl;
-      cfl_update_costs(cfl, ec_ctx);
-      cfl_dc_pred(xd, plane_bsize, tx_size);
-      mbmi->cfl_alpha_idx =
-          cfl_compute_alpha_ind(x, cfl, plane_bsize, mbmi->cfl_alpha_signs);
-    }
-  }
-  av1_predict_intra_block_facade(xd, plane, block_idx, blk_col, blk_row,
-                                 tx_size);
-}
-#endif
-
 void av1_encode_intra_block_plane(AV1_COMMON *cm, MACROBLOCK *x,
                                   BLOCK_SIZE bsize, int plane,
                                   int enable_optimize_b, int mi_row,
@@ -2001,7 +1471,7 @@ void av1_encode_intra_block_plane(AV1_COMMON *cm, MACROBLOCK *x,
 
   if (enable_optimize_b) {
     const struct macroblockd_plane *const pd = &xd->plane[plane];
-    const TX_SIZE tx_size = get_tx_size(plane, xd);
+    const TX_SIZE tx_size = av1_get_tx_size(plane, xd);
     av1_get_entropy_contexts(bsize, tx_size, pd, ta, tl);
   }
   av1_foreach_transformed_block_in_plane(