diff --git a/av1/common/av1_inv_txfm1d_cfg.h b/av1/common/av1_inv_txfm1d_cfg.h
index ba037c2..97beacf 100644
--- a/av1/common/av1_inv_txfm1d_cfg.h
+++ b/av1/common/av1_inv_txfm1d_cfg.h
@@ -16,8 +16,13 @@
 // sum of fwd_shift_##
 #if CONFIG_TX64X64
 static const int8_t fwd_shift_sum[TX_SIZES] = { 2, 1, 0, -2, -4 };
-#else  // CONFIG_TX64X64
+static const int8_t inv_start_range[TX_SIZES_ALL] = {
+  5, 6, 7, 7, 7, 6, 6, 7, 7, 7, 7, 7, 7, 6, 6, 7, 7, 7, 7
+};
+#else   // CONFIG_TX64X64
 static const int8_t fwd_shift_sum[TX_SIZES] = { 2, 1, 0, -2 };
+static const int8_t inv_start_range[TX_SIZES_ALL] = { 5, 6, 7, 7, 6, 6, 7,
+                                                      7, 7, 7, 6, 6, 7, 7 };
 #endif  // CONFIG_TX64X64
 
 //  ---------------- 4x4 1D config -----------------------
@@ -25,7 +30,7 @@
 static const int8_t inv_shift_4[2] = { 0, -4 };
 
 // stage range
-static const int8_t inv_stage_range_col_dct_4[4] = { 3, 3, 2, 2 };
+static const int8_t inv_stage_range_col_dct_4[4] = { 3, 3, 3, 3 };
 static const int8_t inv_stage_range_row_dct_4[4] = { 3, 3, 3, 3 };
 static const int8_t inv_stage_range_col_adst_4[6] = { 3, 3, 3, 3, 2, 2 };
 static const int8_t inv_stage_range_row_adst_4[6] = { 3, 3, 3, 3, 3, 3 };
@@ -111,11 +116,18 @@
 //  ---------------- 8x16 1D constants -----------------------
 #define inv_shift_8x16 inv_shift_16
 // stage range
+static const int8_t inv_stage_range_row_dct_8x16[6] = { 5, 5, 5, 5, 5, 5 };
+static const int8_t inv_stage_range_row_adst_8x16[8] = {
+  5, 5, 5, 5, 5, 5, 5, 5
+};
 static const int8_t inv_stage_range_col_dct_8x16[8] =
     ARRAYOFFSET8(-2, 7, 7, 7, 7, 7, 7, 7, 7);
 static const int8_t inv_stage_range_col_adst_8x16[10] =
     ARRAYOFFSET10(-2, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7);
 // cos bit
+static const int8_t inv_cos_bit_row_dct_8x16[6] = { 12, 12, 12, 12, 12, 12 };
+static const int8_t inv_cos_bit_row_adst_8x16[8] = { 12, 12, 12, 12,
+                                                     12, 12, 12, 12 };
 static const int8_t inv_cos_bit_col_dct_8x16[8] = { 13, 13, 13, 13,
                                                     13, 13, 13, 13 };
 static const int8_t inv_cos_bit_col_adst_8x16[10] = { 13, 13, 13, 13, 13,
@@ -249,11 +261,18 @@
 //  ---------------- 8x32 1D constants -----------------------
 #define inv_shift_8x32 inv_shift_32
 // stage range
+static const int8_t inv_stage_range_row_dct_8x32[6] = { 5, 5, 5, 5, 5, 5 };
+static const int8_t inv_stage_range_row_adst_8x32[8] = {
+  5, 5, 5, 5, 5, 5, 5, 5
+};
 static const int8_t inv_stage_range_col_dct_8x32[10] =
     ARRAYOFFSET10(-4, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9);
 static const int8_t inv_stage_range_col_adst_8x32[12] =
     ARRAYOFFSET12(-4, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9);
 // cos bit
+static const int8_t inv_cos_bit_row_dct_8x32[6] = { 12, 12, 12, 12, 12, 12 };
+static const int8_t inv_cos_bit_row_adst_8x32[8] = { 12, 12, 12, 12,
+                                                     12, 12, 12, 12 };
 static const int8_t inv_cos_bit_col_dct_8x32[10] = { 13, 13, 13, 13, 13,
                                                      13, 13, 13, 13, 13 };
 static const int8_t inv_cos_bit_col_adst_8x32[12] = { 13, 13, 13, 13, 13, 13,
@@ -692,6 +711,16 @@
   TXFM_TYPE_ADST16,               // .txfm_type
 };
 
+//  ---------------- row config inv_dct_8x16 ----------------
+static const TXFM_1D_CFG inv_txfm_1d_row_cfg_dct_8x16 = {
+  8,                             // .txfm_size
+  6,                             // .stage_num
+  inv_shift_8x16,                // .shift
+  inv_stage_range_row_dct_8x16,  // .stage_range
+  inv_cos_bit_row_dct_8x16,      // .cos_bit
+  TXFM_TYPE_DCT8                 // .txfm_type
+};
+
 //  ---------------- col config inv_dct_8x16 ----------------
 static const TXFM_1D_CFG inv_txfm_1d_col_cfg_dct_8x16 = {
   16,                            // .txfm_size
@@ -702,6 +731,16 @@
   TXFM_TYPE_DCT16                // .txfm_type
 };
 
+//  ---------------- row config inv_adst_8x16 ----------------
+static const TXFM_1D_CFG inv_txfm_1d_row_cfg_adst_8x16 = {
+  8,                              // .txfm_size
+  8,                              // .stage_num
+  inv_shift_8x16,                 // .shift
+  inv_stage_range_row_adst_8x16,  // .stage_range
+  inv_cos_bit_row_adst_8x16,      // .cos_bit
+  TXFM_TYPE_ADST8,                // .txfm_type
+};
+
 //  ---------------- col config inv_adst_8x16 ----------------
 static const TXFM_1D_CFG inv_txfm_1d_col_cfg_adst_8x16 = {
   16,                             // .txfm_size
@@ -744,6 +783,16 @@
 };
 #endif  // CONFIG_TX64X64
 
+//  ---------------- row config inv_dct_8x32 ----------------
+static const TXFM_1D_CFG inv_txfm_1d_row_cfg_dct_8x32 = {
+  8,                             // .txfm_size
+  6,                             // .stage_num
+  inv_shift_8x32,                // .shift
+  inv_stage_range_row_dct_8x32,  // .stage_range
+  inv_cos_bit_row_dct_8x32,      // .cos_bit_col
+  TXFM_TYPE_DCT8                 // .txfm_type
+};
+
 //  ---------------- col config inv_dct_8x32 ----------------
 static const TXFM_1D_CFG inv_txfm_1d_col_cfg_dct_8x32 = {
   32,                            // .txfm_size
@@ -754,6 +803,16 @@
   TXFM_TYPE_DCT32                // .txfm_type
 };
 
+//  ---------------- row config inv_adst_8x32 ----------------
+static const TXFM_1D_CFG inv_txfm_1d_row_cfg_adst_8x32 = {
+  8,                              // .txfm_size
+  8,                              // .stage_num
+  inv_shift_8x32,                 // .shift
+  inv_stage_range_row_adst_8x32,  // .stage_range
+  inv_cos_bit_row_adst_8x32,      // .cos_bit
+  TXFM_TYPE_ADST8,                // .txfm_type
+};
+
 //  ---------------- col config inv_adst_8x32 ----------------
 static const TXFM_1D_CFG inv_txfm_1d_col_cfg_adst_8x32 = {
   32,                             // .txfm_size
diff --git a/av1/common/av1_inv_txfm2d.c b/av1/common/av1_inv_txfm2d.c
index 1874819..3c83769 100644
--- a/av1/common/av1_inv_txfm2d.c
+++ b/av1/common/av1_inv_txfm2d.c
@@ -17,6 +17,8 @@
 #include "av1/common/av1_inv_txfm1d.h"
 #include "av1/common/av1_inv_txfm1d_cfg.h"
 
+#define NO_INV_TRANSPOSE 1
+
 static INLINE TxfmFunc inv_txfm_type_to_func(TXFM_TYPE txfm_type) {
   switch (txfm_type) {
     case TXFM_TYPE_DCT4: return av1_idct4_new;
@@ -141,21 +143,21 @@
 static const TXFM_1D_CFG *inv_txfm_row_cfg_ls[TX_TYPES_1D][TX_SIZES_ALL] = {
   // DCT
   {
-      &inv_txfm_1d_row_cfg_dct_4,  &inv_txfm_1d_row_cfg_dct_8,
-      &inv_txfm_1d_row_cfg_dct_16, &inv_txfm_1d_row_cfg_dct_32,
+      &inv_txfm_1d_row_cfg_dct_4,    &inv_txfm_1d_row_cfg_dct_8,
+      &inv_txfm_1d_row_cfg_dct_16,   &inv_txfm_1d_row_cfg_dct_32,
 #if CONFIG_TX64X64
       &inv_txfm_1d_row_cfg_dct_64,
 #endif  // CONFIG_TX64X64
-      &inv_txfm_1d_row_cfg_dct_4,  &inv_txfm_1d_row_cfg_dct_8,
-      &inv_txfm_1d_row_cfg_dct_8,  &inv_txfm_1d_row_cfg_dct_16,
-      &inv_txfm_1d_row_cfg_dct_16, &inv_txfm_1d_row_cfg_dct_32,
+      &inv_txfm_1d_row_cfg_dct_4,    &inv_txfm_1d_row_cfg_dct_8,
+      &inv_txfm_1d_row_cfg_dct_8x16, &inv_txfm_1d_row_cfg_dct_16,
+      &inv_txfm_1d_row_cfg_dct_16,   &inv_txfm_1d_row_cfg_dct_32,
 #if CONFIG_TX64X64
-      &inv_txfm_1d_row_cfg_dct_32, &inv_txfm_1d_row_cfg_dct_64,
+      &inv_txfm_1d_row_cfg_dct_32,   &inv_txfm_1d_row_cfg_dct_64,
 #endif  // CONFIG_TX64X64
-      &inv_txfm_1d_row_cfg_dct_4,  &inv_txfm_1d_row_cfg_dct_16,
-      &inv_txfm_1d_row_cfg_dct_8,  &inv_txfm_1d_row_cfg_dct_32,
+      &inv_txfm_1d_row_cfg_dct_4,    &inv_txfm_1d_row_cfg_dct_16,
+      &inv_txfm_1d_row_cfg_dct_8x32, &inv_txfm_1d_row_cfg_dct_32,
 #if CONFIG_TX64X64
-      &inv_txfm_1d_row_cfg_dct_16, &inv_txfm_1d_row_cfg_dct_64,
+      &inv_txfm_1d_row_cfg_dct_16,   &inv_txfm_1d_row_cfg_dct_64,
 #endif  // CONFIG_TX64X64
   },
   // ADST
@@ -169,7 +171,7 @@
 #endif  // CONFIG_TX64X64
       &inv_txfm_1d_row_cfg_adst_4,
       &inv_txfm_1d_row_cfg_adst_8,
-      &inv_txfm_1d_row_cfg_adst_8,
+      &inv_txfm_1d_row_cfg_adst_8x16,
       &inv_txfm_1d_row_cfg_adst_16,
       &inv_txfm_1d_row_cfg_adst_16,
       &inv_txfm_1d_row_cfg_adst_32,
@@ -179,7 +181,7 @@
 #endif  // CONFIG_TX64X64
       &inv_txfm_1d_row_cfg_adst_4,
       &inv_txfm_1d_row_cfg_adst_16,
-      &inv_txfm_1d_row_cfg_adst_8,
+      &inv_txfm_1d_row_cfg_adst_8x32,
       &inv_txfm_1d_row_cfg_adst_32,
 #if CONFIG_TX64X64
       &inv_txfm_1d_row_cfg_adst_16,
@@ -197,7 +199,7 @@
 #endif  // CONFIG_TX64X64
       &inv_txfm_1d_row_cfg_adst_4,
       &inv_txfm_1d_row_cfg_adst_8,
-      &inv_txfm_1d_row_cfg_adst_8,
+      &inv_txfm_1d_row_cfg_adst_8x16,
       &inv_txfm_1d_row_cfg_adst_16,
       &inv_txfm_1d_row_cfg_adst_16,
       &inv_txfm_1d_row_cfg_adst_32,
@@ -207,7 +209,7 @@
 #endif  // CONFIG_TX64X64
       &inv_txfm_1d_row_cfg_adst_4,
       &inv_txfm_1d_row_cfg_adst_16,
-      &inv_txfm_1d_row_cfg_adst_8,
+      &inv_txfm_1d_row_cfg_adst_8x32,
       &inv_txfm_1d_row_cfg_adst_32,
 #if CONFIG_TX64X64
       &inv_txfm_1d_row_cfg_adst_16,
@@ -246,7 +248,7 @@
 }
 
 void av1_gen_inv_stage_range(int8_t *stage_range_col, int8_t *stage_range_row,
-                             const TXFM_2D_FLIP_CFG *cfg, int8_t fwd_shift,
+                             const TXFM_2D_FLIP_CFG *cfg, TX_SIZE tx_size,
                              int bd) {
   // Note when assigning txfm_size_col, we use the txfm_size from the
   // row configuration and vice versa. This is intentionally done to
@@ -254,35 +256,12 @@
   // rectangular, the number of columns will be the same as the
   // txfm_size stored in the row cfg struct. It will make no difference
   // for square transforms.
-  const int txfm_size_col = cfg->row_cfg->txfm_size;
-  const int txfm_size_row = cfg->col_cfg->txfm_size;
-  // Take the shift from the larger dimension in the rectangular case.
-  const int8_t *shift = (txfm_size_col > txfm_size_row) ? cfg->row_cfg->shift
-                                                        : cfg->col_cfg->shift;
-  // i < MAX_TXFM_STAGE_NUM will mute above array bounds warning
-  for (int i = 0; i < cfg->row_cfg->stage_num && i < MAX_TXFM_STAGE_NUM; ++i) {
-    stage_range_row[i] = cfg->row_cfg->stage_range[i] + fwd_shift + bd + 1;
-  }
-  // i < MAX_TXFM_STAGE_NUM will mute above array bounds warning
-  for (int i = 0; i < cfg->col_cfg->stage_num && i < MAX_TXFM_STAGE_NUM; ++i) {
-    stage_range_col[i] =
-        cfg->col_cfg->stage_range[i] + fwd_shift + shift[0] + bd + 1;
-  }
-}
-
-static INLINE void inv_txfm2d_add_c(const int32_t *input, uint16_t *output,
-                                    int stride, TXFM_2D_FLIP_CFG *cfg,
-                                    int32_t *txfm_buf, int8_t fwd_shift,
-                                    int bd) {
-  // Note when assigning txfm_size_col, we use the txfm_size from the
-  // row configuration and vice versa. This is intentionally done to
-  // accurately perform rectangular transforms. When the transform is
-  // rectangular, the number of columns will be the same as the
-  // txfm_size stored in the row cfg struct. It will make no difference
-  // for square transforms.
+  // const int fwd_shift = fwd_shift_sum[txsize_sqr_up_map[tx_size]];
+  const int fwd_shift = inv_start_range[tx_size];
   const int txfm_size_col = cfg->row_cfg->txfm_size;
   const int txfm_size_row = cfg->col_cfg->txfm_size;
   const int rect_type = get_rect_tx_log_ratio(txfm_size_col, txfm_size_row);
+  if (txfm_size_col == txfm_size_row) assert(rect_type == 0);
   int rect_type2_shift = 0;
   if (rect_type == 2 || rect_type == -2) {
     const int txfm_size_max = AOMMAX(txfm_size_col, txfm_size_row);
@@ -293,11 +272,57 @@
   // Take the shift from the larger dimension in the rectangular case.
   const int8_t *shift = (txfm_size_col > txfm_size_row) ? cfg->row_cfg->shift
                                                         : cfg->col_cfg->shift;
+  int shift1 = shift[1];
+  while (rect_type2_shift > 0 && shift1 < 0) {
+    shift1++;
+    rect_type2_shift--;
+  }
+  // i < MAX_TXFM_STAGE_NUM will mute above array bounds warning
+  for (int i = 0; i < cfg->row_cfg->stage_num && i < MAX_TXFM_STAGE_NUM; ++i) {
+    stage_range_row[i] = cfg->row_cfg->stage_range[i] + fwd_shift + bd + 1 -
+                         cfg->row_cfg->stage_range[0];
+  }
+  // i < MAX_TXFM_STAGE_NUM will mute above array bounds warning
+  for (int i = 0; i < cfg->col_cfg->stage_num && i < MAX_TXFM_STAGE_NUM; ++i) {
+    stage_range_col[i] = cfg->col_cfg->stage_range[i] + fwd_shift + shift[0] +
+                         bd + 1 - cfg->col_cfg->stage_range[0] +
+                         rect_type2_shift;
+  }
+}
+
+static INLINE void inv_txfm2d_add_c(const int32_t *input, uint16_t *output,
+                                    int stride, TXFM_2D_FLIP_CFG *cfg,
+                                    int32_t *txfm_buf, TX_SIZE tx_size,
+                                    int bd) {
+  // Note when assigning txfm_size_col, we use the txfm_size from the
+  // row configuration and vice versa. This is intentionally done to
+  // accurately perform rectangular transforms. When the transform is
+  // rectangular, the number of columns will be the same as the
+  // txfm_size stored in the row cfg struct. It will make no difference
+  // for square transforms.
+  const int txfm_size_col = cfg->row_cfg->txfm_size;
+  const int txfm_size_row = cfg->col_cfg->txfm_size;
+  // Take the shift from the larger dimension in the rectangular case.
+  const int8_t *shift = (txfm_size_col > txfm_size_row) ? cfg->row_cfg->shift
+                                                        : cfg->col_cfg->shift;
+  const int rect_type = get_rect_tx_log_ratio(txfm_size_col, txfm_size_row);
+  int rect_type2_shift = 0;
+  int shift1 = shift[1];
+  if (rect_type == 2 || rect_type == -2) {
+    const int txfm_size_max = AOMMAX(txfm_size_col, txfm_size_row);
+    // For 16x4 / 4x16 shift 1 bit, for 32x8 / 8x32 / 64x16 / 16x64 no need
+    // for any additional shift.
+    rect_type2_shift = (txfm_size_max == 16 ? 1 : 0);
+    while (rect_type2_shift > 0 && shift1 < 0) {
+      shift1++;
+      rect_type2_shift--;
+    }
+  }
   int8_t stage_range_row[MAX_TXFM_STAGE_NUM];
   int8_t stage_range_col[MAX_TXFM_STAGE_NUM];
   assert(cfg->row_cfg->stage_num <= MAX_TXFM_STAGE_NUM);
   assert(cfg->col_cfg->stage_num <= MAX_TXFM_STAGE_NUM);
-  av1_gen_inv_stage_range(stage_range_col, stage_range_row, cfg, fwd_shift, bd);
+  av1_gen_inv_stage_range(stage_range_col, stage_range_row, cfg, tx_size, bd);
 
   const int8_t *cos_bit_col = cfg->col_cfg->cos_bit;
   const int8_t *cos_bit_row = cfg->row_cfg->cos_bit;
@@ -340,7 +365,7 @@
         temp_in[r] = buf[r * txfm_size_col + (txfm_size_col - c - 1)];
     }
     txfm_func_col(temp_in, temp_out, cos_bit_col, stage_range_col);
-    av1_round_shift_array(temp_out, txfm_size_row, -shift[1]);
+    av1_round_shift_array(temp_out, txfm_size_row, -shift1);
     if (cfg->ud_flip == 0) {
       for (r = 0; r < txfm_size_row; ++r) {
         output[r * stride + c] =
@@ -364,9 +389,7 @@
   av1_get_inv_txfm_cfg(tx_type, tx_size, &cfg);
   // Forward shift sum uses larger square size, to be consistent with what
   // av1_gen_inv_stage_range() does for inverse shifts.
-  const TX_SIZE tx_size_sqr_up = txsize_sqr_up_map[tx_size];
-  inv_txfm2d_add_c(input, output, stride, &cfg, txfm_buf,
-                   fwd_shift_sum[tx_size_sqr_up], bd);
+  inv_txfm2d_add_c(input, output, stride, &cfg, txfm_buf, tx_size, bd);
 }
 
 void av1_inv_txfm2d_add_4x8_c(const int32_t *input, uint16_t *output,
@@ -378,6 +401,9 @@
 void av1_inv_txfm2d_add_8x4_c(const int32_t *input, uint16_t *output,
                               int stride, TX_TYPE tx_type, int bd) {
   DECLARE_ALIGNED(32, int, txfm_buf[8 * 4 + 8 + 8]);
+#if NO_INV_TRANSPOSE
+  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X4, bd);
+#else
   int32_t rinput[8 * 4];
   uint16_t routput[8 * 4];
   TX_SIZE tx_size = TX_8X4;
@@ -391,6 +417,7 @@
   transpose_uint16(routput, rw, output, stride, w, h);
   inv_txfm2d_add_facade(rinput, routput, rw, txfm_buf, rtx_type, rtx_size, bd);
   transpose_uint16(output, stride, routput, rw, rw, rh);
+#endif  // NO_INV_TRANSPOSE
 }
 
 void av1_inv_txfm2d_add_8x16_c(const int32_t *input, uint16_t *output,
@@ -402,6 +429,9 @@
 void av1_inv_txfm2d_add_16x8_c(const int32_t *input, uint16_t *output,
                                int stride, TX_TYPE tx_type, int bd) {
   DECLARE_ALIGNED(32, int, txfm_buf[16 * 8 + 16 + 16]);
+#if NO_INV_TRANSPOSE
+  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X8, bd);
+#else
   int32_t rinput[16 * 8];
   uint16_t routput[16 * 8];
   TX_SIZE tx_size = TX_16X8;
@@ -415,6 +445,7 @@
   transpose_uint16(routput, rw, output, stride, w, h);
   inv_txfm2d_add_facade(rinput, routput, rw, txfm_buf, rtx_type, rtx_size, bd);
   transpose_uint16(output, stride, routput, rw, rw, rh);
+#endif  // NO_INV_TRANSPOSE
 }
 
 void av1_inv_txfm2d_add_16x32_c(const int32_t *input, uint16_t *output,
@@ -426,6 +457,9 @@
 void av1_inv_txfm2d_add_32x16_c(const int32_t *input, uint16_t *output,
                                 int stride, TX_TYPE tx_type, int bd) {
   DECLARE_ALIGNED(32, int, txfm_buf[32 * 16 + 32 + 32]);
+#if NO_INV_TRANSPOSE
+  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_32X16, bd);
+#else
   int32_t rinput[32 * 16];
   uint16_t routput[32 * 16];
   TX_SIZE tx_size = TX_32X16;
@@ -439,6 +473,7 @@
   transpose_uint16(routput, rw, output, stride, w, h);
   inv_txfm2d_add_facade(rinput, routput, rw, txfm_buf, rtx_type, rtx_size, bd);
   transpose_uint16(output, stride, routput, rw, rw, rh);
+#endif  // NO_INV_TRANSPOSE
 }
 
 void av1_inv_txfm2d_add_4x4_c(const int32_t *input, uint16_t *output,
@@ -494,6 +529,10 @@
     memset(mod_input + row * 64 + 32, 0, 32 * sizeof(*mod_input));
   }
   DECLARE_ALIGNED(32, int, txfm_buf[64 * 32 + 64 + 64]);
+#if NO_INV_TRANSPOSE
+  inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_64X32,
+                        bd);
+#else
   int32_t rinput[64 * 32];
   uint16_t routput[64 * 32];
   TX_SIZE tx_size = TX_64X32;
@@ -507,6 +546,7 @@
   transpose_uint16(routput, rw, output, stride, w, h);
   inv_txfm2d_add_facade(rinput, routput, rw, txfm_buf, rtx_type, rtx_size, bd);
   transpose_uint16(output, stride, routput, rw, rw, rh);
+#endif  // NO_INV_TRANSPOSE
 }
 
 void av1_inv_txfm2d_add_32x64_c(const int32_t *input, uint16_t *output,
@@ -546,6 +586,10 @@
     memset(mod_input + row * 64 + 32, 0, 32 * sizeof(*mod_input));
   }
   DECLARE_ALIGNED(32, int, txfm_buf[16 * 64 + 64 + 64]);
+#if NO_INV_TRANSPOSE
+  inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_64X16,
+                        bd);
+#else
   int32_t rinput[16 * 64];
   uint16_t routput[16 * 64];
   TX_SIZE tx_size = TX_64X16;
@@ -559,6 +603,7 @@
   transpose_uint16(routput, rw, output, stride, w, h);
   inv_txfm2d_add_facade(rinput, routput, rw, txfm_buf, rtx_type, rtx_size, bd);
   transpose_uint16(output, stride, routput, rw, rw, rh);
+#endif  // NO_INV_TRANSPOSE
 }
 #endif  // CONFIG_TX64X64
 
@@ -571,6 +616,9 @@
 void av1_inv_txfm2d_add_16x4_c(const int32_t *input, uint16_t *output,
                                int stride, TX_TYPE tx_type, int bd) {
   DECLARE_ALIGNED(32, int, txfm_buf[4 * 16 + 16 + 16]);
+#if NO_INV_TRANSPOSE
+  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X4, bd);
+#else
   int32_t rinput[4 * 16];
   uint16_t routput[4 * 16];
   TX_SIZE tx_size = TX_16X4;
@@ -584,6 +632,7 @@
   transpose_uint16(routput, rw, output, stride, w, h);
   inv_txfm2d_add_facade(rinput, routput, rw, txfm_buf, rtx_type, rtx_size, bd);
   transpose_uint16(output, stride, routput, rw, rw, rh);
+#endif  // NO_INV_TRANSPOSE
 }
 
 void av1_inv_txfm2d_add_8x32_c(const int32_t *input, uint16_t *output,
@@ -595,6 +644,9 @@
 void av1_inv_txfm2d_add_32x8_c(const int32_t *input, uint16_t *output,
                                int stride, TX_TYPE tx_type, int bd) {
   DECLARE_ALIGNED(32, int, txfm_buf[8 * 32 + 32 + 32]);
+#if NO_INV_TRANSPOSE
+  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_32X8, bd);
+#else
   int32_t rinput[8 * 32];
   uint16_t routput[8 * 32];
   TX_SIZE tx_size = TX_32X8;
@@ -608,4 +660,5 @@
   transpose_uint16(routput, rw, output, stride, w, h);
   inv_txfm2d_add_facade(rinput, routput, rw, txfm_buf, rtx_type, rtx_size, bd);
   transpose_uint16(output, stride, routput, rw, rw, rh);
+#endif  // NO_INV_TRANSPOSE
 }
diff --git a/av1/common/av1_txfm.h b/av1/common/av1_txfm.h
index 061e0d8..df9fc3c 100644
--- a/av1/common/av1_txfm.h
+++ b/av1/common/av1_txfm.h
@@ -263,7 +263,7 @@
                              const TXFM_2D_FLIP_CFG *cfg, int bd);
 
 void av1_gen_inv_stage_range(int8_t *stage_range_col, int8_t *stage_range_row,
-                             const TXFM_2D_FLIP_CFG *cfg, int8_t fwd_shift,
+                             const TXFM_2D_FLIP_CFG *cfg, TX_SIZE tx_size,
                              int bd);
 
 void av1_get_fwd_txfm_cfg(TX_TYPE tx_type, TX_SIZE tx_size,
diff --git a/av1/encoder/av1_fwd_txfm1d_cfg.h b/av1/encoder/av1_fwd_txfm1d_cfg.h
index 1d925ed..187f62d 100644
--- a/av1/encoder/av1_fwd_txfm1d_cfg.h
+++ b/av1/encoder/av1_fwd_txfm1d_cfg.h
@@ -209,8 +209,8 @@
     ARRAYOFFSET4(4, 0, 1, 2, 2);
 static const int8_t fwd_stage_range_row_adst_4x16[6] =
     ARRAYOFFSET6(4, 0, 0, 1, 2, 2, 2);
-static const int8_t fwd_cos_bit_row_dct_4x16[6] = { 13, 13, 13, 13 };
-static const int8_t fwd_cos_bit_row_adst_4x16[6] = { 13, 13, 13, 13, 13, 13 };
+static const int8_t fwd_cos_bit_row_dct_4x16[6] = { 12, 12, 12, 12 };
+static const int8_t fwd_cos_bit_row_adst_4x16[6] = { 12, 12, 12, 12, 12, 12 };
 
 //  ---------------- 16x4 1D constants -----------------------
 #define fwd_shift_16x4 fwd_shift_16
@@ -218,10 +218,10 @@
     ARRAYOFFSET8(2, 0, 1, 2, 3, 4, 4, 4, 4);
 static const int8_t fwd_stage_range_row_adst_16x4[10] =
     ARRAYOFFSET10(2, 0, 0, 1, 2, 2, 3, 3, 4, 4, 4);
-static const int8_t fwd_cos_bit_row_dct_16x4[8] = { 13, 13, 13, 13,
-                                                    13, 13, 13, 13 };
-static const int8_t fwd_cos_bit_row_adst_16x4[10] = { 13, 13, 13, 13, 13,
-                                                      13, 13, 13, 13, 13 };
+static const int8_t fwd_cos_bit_row_dct_16x4[8] = { 12, 12, 12, 12,
+                                                    12, 12, 12, 12 };
+static const int8_t fwd_cos_bit_row_adst_16x4[10] = { 12, 12, 12, 12, 12,
+                                                      12, 12, 12, 12, 12 };
 
 //  ---------------- 8x32 1D constants -----------------------
 #define fwd_shift_8x32 fwd_shift_32
@@ -229,9 +229,9 @@
     ARRAYOFFSET6(5, 0, 1, 2, 3, 3, 3);
 static const int8_t fwd_stage_range_row_adst_8x32[8] =
     ARRAYOFFSET8(5, 0, 0, 1, 2, 2, 3, 3, 3);
-static const int8_t fwd_cos_bit_row_dct_8x32[6] = { 13, 13, 13, 13, 13, 13 };
-static const int8_t fwd_cos_bit_row_adst_8x32[8] = { 13, 13, 13, 13,
-                                                     13, 13, 13, 13 };
+static const int8_t fwd_cos_bit_row_dct_8x32[6] = { 12, 12, 11, 11, 11, 11 };
+static const int8_t fwd_cos_bit_row_adst_8x32[8] = { 12, 12, 12, 12,
+                                                     11, 11, 11, 11 };
 
 //  ---------------- 32x8 1D constants -----------------------
 #define fwd_shift_32x8 fwd_shift_32
@@ -239,17 +239,17 @@
     ARRAYOFFSET10(3, 0, 1, 2, 3, 4, 5, 5, 5, 5, 5);
 static const int8_t fwd_stage_range_row_adst_32x8[12] =
     ARRAYOFFSET12(3, 0, 0, 1, 2, 2, 3, 3, 4, 4, 5, 5, 5);
-static const int8_t fwd_cos_bit_row_dct_32x8[10] = { 12, 12, 12, 12, 12,
-                                                     12, 12, 12, 12, 12 };
+static const int8_t fwd_cos_bit_row_dct_32x8[10] = { 12, 12, 12, 12, 11,
+                                                     11, 11, 11, 11, 11 };
 static const int8_t fwd_cos_bit_row_adst_32x8[12] = { 12, 12, 12, 12, 12, 12,
-                                                      12, 12, 12, 12, 12, 12 };
+                                                      12, 11, 11, 11, 11, 11 };
 
 //  ---------------- 16x64 1D constants -----------------------
 #define fwd_shift_16x64 fwd_shift_64
 static const int8_t fwd_stage_range_row_dct_16x64[8] =
     ARRAYOFFSET8(6, 0, 1, 2, 3, 4, 4, 4, 4);
-static const int8_t fwd_cos_bit_row_dct_16x64[8] = { 12, 12, 12, 11,
-                                                     11, 11, 11, 11 };
+static const int8_t fwd_cos_bit_row_dct_16x64[8] = { 12, 11, 10, 10,
+                                                     10, 10, 10, 10 };
 
 //  ---------------- 64x16 1D constants -----------------------
 #define fwd_shift_64x16 fwd_shift_64
diff --git a/av1/encoder/av1_fwd_txfm2d.c b/av1/encoder/av1_fwd_txfm2d.c
index 99102f2..250521e 100644
--- a/av1/encoder/av1_fwd_txfm2d.c
+++ b/av1/encoder/av1_fwd_txfm2d.c
@@ -19,6 +19,8 @@
 #include "av1/encoder/av1_fwd_txfm1d.h"
 #include "av1/encoder/av1_fwd_txfm1d_cfg.h"
 
+#define NO_FWD_TRANSPOSE 1
+
 static INLINE TxfmFunc fwd_txfm_type_to_func(TXFM_TYPE txfm_type) {
   switch (txfm_type) {
     case TXFM_TYPE_DCT4: return av1_fdct4_new;
@@ -61,10 +63,23 @@
     stage_range_col[i] = cfg->col_cfg->stage_range[i] + shift[0] + bd + 1;
   }
 
+  const int rect_type = get_rect_tx_log_ratio(txfm_size_col, txfm_size_row);
+  int rect_shift = 0;
+  int shift2 = shift[2];
+  if (rect_type == 2 || rect_type == -2) {
+    const int txfm_size_max = AOMMAX(txfm_size_col, txfm_size_row);
+    // For 64x16 / 16x64 / 32x8 / 8x32 shift 2 bits, and
+    // For 16x4 / 4x16 shift by 1 bit.
+    rect_shift = (txfm_size_max >= 32) ? 2 : 1;
+  }
+  while (rect_shift > 0 && shift2 < 0) {
+    shift2++;
+    rect_shift--;
+  }
   // i < MAX_TXFM_STAGE_NUM will mute above array bounds warning
   for (int i = 0; i < cfg->row_cfg->stage_num && i < MAX_TXFM_STAGE_NUM; ++i) {
-    stage_range_row[i] =
-        cfg->row_cfg->stage_range[i] + shift[0] + shift[1] + bd + 1;
+    stage_range_row[i] = cfg->row_cfg->stage_range[i] + shift[0] + shift[1] +
+                         bd + 1 + rect_shift;
   }
 }
 
@@ -80,6 +95,10 @@
   // for square transforms.
   const int txfm_size_col = cfg->row_cfg->txfm_size;
   const int txfm_size_row = cfg->col_cfg->txfm_size;
+  // Take the shift from the larger dimension in the rectangular case.
+  const int8_t *shift = (txfm_size_col > txfm_size_row) ? cfg->row_cfg->shift
+                                                        : cfg->col_cfg->shift;
+  int shift2 = shift[2];
   const int rect_type = get_rect_tx_log_ratio(txfm_size_col, txfm_size_row);
   int rect_type2_shift = 0;
   if (rect_type == 2 || rect_type == -2) {
@@ -87,10 +106,11 @@
     // For 64x16 / 16x64 / 32x8 / 8x32 shift 2 bits, and
     // For 16x4 / 4x16 shift by 1 bit.
     rect_type2_shift = (txfm_size_max >= 32) ? 2 : 1;
+    while (rect_type2_shift > 0 && shift2 < 0) {
+      shift2++;
+      rect_type2_shift--;
+    }
   }
-  // Take the shift from the larger dimension in the rectangular case.
-  const int8_t *shift = (txfm_size_col > txfm_size_row) ? cfg->row_cfg->shift
-                                                        : cfg->col_cfg->shift;
   int8_t stage_range_col[MAX_TXFM_STAGE_NUM];
   int8_t stage_range_row[MAX_TXFM_STAGE_NUM];
   assert(cfg->col_cfg->stage_num <= MAX_TXFM_STAGE_NUM);
@@ -116,17 +136,17 @@
         temp_in[r] = input[(txfm_size_row - r - 1) * stride + c];
     }
     av1_round_shift_array(temp_in, txfm_size_row, -shift[0]);
+    txfm_func_col(temp_in, temp_out, cos_bit_col, stage_range_col);
     // Multiply everything by Sqrt2 on the larger dimension if the
     // transform is rectangular and the size difference is a factor of 2.
     // If the size difference is a factor of 4, multiply by
     // 2^rect_type_2_extra_shift.
     if (rect_type == 1) {
       for (r = 0; r < txfm_size_row; ++r)
-        temp_in[r] = (int32_t)fdct_round_shift(temp_in[r] * Sqrt2);
+        temp_out[r] = (int32_t)fdct_round_shift(temp_out[r] * Sqrt2);
     } else if (rect_type == 2) {
-      av1_round_shift_array(temp_in, txfm_size_row, -rect_type2_shift);
+      av1_round_shift_array(temp_out, txfm_size_row, -rect_type2_shift);
     }
-    txfm_func_col(temp_in, temp_out, cos_bit_col, stage_range_col);
     av1_round_shift_array(temp_out, txfm_size_row, -shift[1]);
     if (cfg->lr_flip == 0) {
       for (r = 0; r < txfm_size_row; ++r)
@@ -154,13 +174,18 @@
     }
     txfm_func_row(buf + r * txfm_size_col, output + r * txfm_size_col,
                   cos_bit_row, stage_range_row);
-    av1_round_shift_array(output + r * txfm_size_col, txfm_size_col, -shift[2]);
+    av1_round_shift_array(output + r * txfm_size_col, txfm_size_col, -shift2);
   }
 }
 
 void av1_fwd_txfm2d_4x8_c(const int16_t *input, int32_t *output, int stride,
                           TX_TYPE tx_type, int bd) {
   DECLARE_ALIGNED(32, int32_t, txfm_buf[4 * 8]);
+  TXFM_2D_FLIP_CFG cfg;
+#if NO_FWD_TRANSPOSE
+  av1_get_fwd_txfm_cfg(tx_type, TX_4X8, &cfg);
+  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
+#else
   int16_t rinput[4 * 8];
   TX_SIZE tx_size = TX_4X8;
   TX_SIZE rtx_size = av1_rotate_tx_size(tx_size);
@@ -170,10 +195,10 @@
   int rw = h;
   int rh = w;
   transpose_int16(rinput, rw, input, stride, w, h);
-  TXFM_2D_FLIP_CFG cfg;
   av1_get_fwd_txfm_cfg(rtx_type, rtx_size, &cfg);
   fwd_txfm2d_c(rinput, txfm_buf, rw, &cfg, output, bd);
   transpose_int32(output, w, txfm_buf, rw, rw, rh);
+#endif  // NO_FWD_TRANSPOSE
 }
 
 void av1_fwd_txfm2d_8x4_c(const int16_t *input, int32_t *output, int stride,
@@ -187,6 +212,11 @@
 void av1_fwd_txfm2d_8x16_c(const int16_t *input, int32_t *output, int stride,
                            TX_TYPE tx_type, int bd) {
   DECLARE_ALIGNED(32, int32_t, txfm_buf[8 * 16]);
+  TXFM_2D_FLIP_CFG cfg;
+#if NO_FWD_TRANSPOSE
+  av1_get_fwd_txfm_cfg(tx_type, TX_8X16, &cfg);
+  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
+#else
   int16_t rinput[8 * 16];
   TX_SIZE tx_size = TX_8X16;
   TX_SIZE rtx_size = av1_rotate_tx_size(tx_size);
@@ -196,10 +226,10 @@
   int rw = h;
   int rh = w;
   transpose_int16(rinput, rw, input, stride, w, h);
-  TXFM_2D_FLIP_CFG cfg;
   av1_get_fwd_txfm_cfg(rtx_type, rtx_size, &cfg);
   fwd_txfm2d_c(rinput, txfm_buf, rw, &cfg, output, bd);
   transpose_int32(output, w, txfm_buf, rw, rw, rh);
+#endif  // NO_FWD_TRANSPOSE
 }
 
 void av1_fwd_txfm2d_16x8_c(const int16_t *input, int32_t *output, int stride,
@@ -213,6 +243,11 @@
 void av1_fwd_txfm2d_16x32_c(const int16_t *input, int32_t *output, int stride,
                             TX_TYPE tx_type, int bd) {
   DECLARE_ALIGNED(32, int32_t, txfm_buf[16 * 32]);
+  TXFM_2D_FLIP_CFG cfg;
+#if NO_FWD_TRANSPOSE
+  av1_get_fwd_txfm_cfg(tx_type, TX_16X32, &cfg);
+  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
+#else
   int16_t rinput[16 * 32];
   TX_SIZE tx_size = TX_16X32;
   TX_SIZE rtx_size = av1_rotate_tx_size(tx_size);
@@ -222,10 +257,10 @@
   int rw = h;
   int rh = w;
   transpose_int16(rinput, rw, input, stride, w, h);
-  TXFM_2D_FLIP_CFG cfg;
   av1_get_fwd_txfm_cfg(rtx_type, rtx_size, &cfg);
   fwd_txfm2d_c(rinput, txfm_buf, rw, &cfg, output, bd);
   transpose_int32(output, w, txfm_buf, rw, rw, rh);
+#endif  // NO_FWD_TRANSPOSE
 }
 
 void av1_fwd_txfm2d_32x16_c(const int16_t *input, int32_t *output, int stride,
@@ -239,6 +274,11 @@
 void av1_fwd_txfm2d_4x16_c(const int16_t *input, int32_t *output, int stride,
                            TX_TYPE tx_type, int bd) {
   DECLARE_ALIGNED(32, int32_t, txfm_buf[4 * 16]);
+  TXFM_2D_FLIP_CFG cfg;
+#if NO_FWD_TRANSPOSE
+  av1_get_fwd_txfm_cfg(tx_type, TX_4X16, &cfg);
+  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
+#else
   int16_t rinput[4 * 16];
   TX_SIZE tx_size = TX_4X16;
   TX_SIZE rtx_size = av1_rotate_tx_size(tx_size);
@@ -248,10 +288,10 @@
   int rw = h;
   int rh = w;
   transpose_int16(rinput, rw, input, stride, w, h);
-  TXFM_2D_FLIP_CFG cfg;
   av1_get_fwd_txfm_cfg(rtx_type, rtx_size, &cfg);
   fwd_txfm2d_c(rinput, txfm_buf, rw, &cfg, output, bd);
   transpose_int32(output, w, txfm_buf, rw, rw, rh);
+#endif  // NO_FWD_TRANSPOSE
 }
 
 void av1_fwd_txfm2d_16x4_c(const int16_t *input, int32_t *output, int stride,
@@ -265,6 +305,11 @@
 void av1_fwd_txfm2d_8x32_c(const int16_t *input, int32_t *output, int stride,
                            TX_TYPE tx_type, int bd) {
   DECLARE_ALIGNED(32, int32_t, txfm_buf[32 * 8]);
+  TXFM_2D_FLIP_CFG cfg;
+#if NO_FWD_TRANSPOSE
+  av1_get_fwd_txfm_cfg(tx_type, TX_8X32, &cfg);
+  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
+#else
   int16_t rinput[32 * 8];
   TX_SIZE tx_size = TX_8X32;
   TX_SIZE rtx_size = av1_rotate_tx_size(tx_size);
@@ -274,10 +319,10 @@
   int rw = h;
   int rh = w;
   transpose_int16(rinput, rw, input, stride, w, h);
-  TXFM_2D_FLIP_CFG cfg;
   av1_get_fwd_txfm_cfg(rtx_type, rtx_size, &cfg);
   fwd_txfm2d_c(rinput, txfm_buf, rw, &cfg, output, bd);
   transpose_int32(output, w, txfm_buf, rw, rw, rh);
+#endif  // NO_FWD_TRANSPOSE
 }
 
 void av1_fwd_txfm2d_32x8_c(const int16_t *input, int32_t *output, int stride,
@@ -343,6 +388,11 @@
 void av1_fwd_txfm2d_32x64_c(const int16_t *input, int32_t *output, int stride,
                             TX_TYPE tx_type, int bd) {
   DECLARE_ALIGNED(32, int32_t, txfm_buf[32 * 64]);
+  TXFM_2D_FLIP_CFG cfg;
+#if NO_FWD_TRANSPOSE
+  av1_get_fwd_txfm_cfg(tx_type, TX_32X64, &cfg);
+  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
+#else
   int16_t rinput[64 * 32];
   TX_SIZE tx_size = TX_32X64;
   TX_SIZE rtx_size = av1_rotate_tx_size(tx_size);
@@ -352,10 +402,10 @@
   int rw = h;
   int rh = w;
   transpose_int16(rinput, rw, input, stride, w, h);
-  TXFM_2D_FLIP_CFG cfg;
   av1_get_fwd_txfm_cfg(rtx_type, rtx_size, &cfg);
   fwd_txfm2d_c(rinput, txfm_buf, rw, &cfg, output, bd);
   transpose_int32(output, w, txfm_buf, rw, rw, rh);
+#endif  // NO_FWD_TRANSPOSE
 
   // Zero out the bottom 32x32 area.
   memset(output + 32 * 32, 0, 32 * 32 * sizeof(*output));
@@ -382,6 +432,11 @@
 void av1_fwd_txfm2d_16x64_c(const int16_t *input, int32_t *output, int stride,
                             TX_TYPE tx_type, int bd) {
   DECLARE_ALIGNED(32, int32_t, txfm_buf[64 * 16]);
+  TXFM_2D_FLIP_CFG cfg;
+#if NO_FWD_TRANSPOSE
+  av1_get_fwd_txfm_cfg(tx_type, TX_16X64, &cfg);
+  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
+#else
   int16_t rinput[64 * 16];
   TX_SIZE tx_size = TX_16X64;
   TX_SIZE rtx_size = av1_rotate_tx_size(tx_size);
@@ -391,10 +446,10 @@
   int rw = h;
   int rh = w;
   transpose_int16(rinput, rw, input, stride, w, h);
-  TXFM_2D_FLIP_CFG cfg;
   av1_get_fwd_txfm_cfg(rtx_type, rtx_size, &cfg);
   fwd_txfm2d_c(rinput, txfm_buf, rw, &cfg, output, bd);
   transpose_int32(output, w, txfm_buf, rw, rw, rh);
+#endif  // NO_FWD_TRANSPOSE
   // Zero out the bottom 16x32 area.
   memset(output + 16 * 32, 0, 16 * 32 * sizeof(*output));
   // Note: no repacking needed here.
diff --git a/test/av1_fwd_txfm2d_test.cc b/test/av1_fwd_txfm2d_test.cc
index ed388db..2a6a308 100644
--- a/test/av1_fwd_txfm2d_test.cc
+++ b/test/av1_fwd_txfm2d_test.cc
@@ -148,24 +148,24 @@
     }
 #endif  // CONFIG_TX64X64
 
-    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_4X8, 2.9, 0.55));
-    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_8X4, 3.2, 0.56));
+    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_4X8, 3.2, 0.58));
+    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_8X4, 3.2, 0.58));
     param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_8X16, 6.5, 1));
     param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_16X8, 6.5, 1));
-    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_16X32, 46, 7));
+    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_16X32, 50, 7));
     param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_32X16, 30, 7));
 
     param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_4X16, 5, 0.7));
     param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_16X4, 5.5, 0.9));
-    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_8X32, 14, 2.1));
-    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_32X8, 11, 1.6));
+    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_8X32, 20, 2.2));
+    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_32X8, 15, 1.8));
 
 #if CONFIG_TX64X64
     if (tx_type == DCT_DCT) {  // Other types not supported by these tx sizes.
       param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_32X64, 136, 7));
       param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_64X32, 136, 7));
-      param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_16X64, 16, 1.6));
-      param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_64X16, 20, 2.0));
+      param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_16X64, 30, 3.6));
+      param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_64X16, 50, 6.0));
     }
 #endif  // CONFIG_TX64X64
   }
diff --git a/test/av1_inv_txfm2d_test.cc b/test/av1_inv_txfm2d_test.cc
index bf15d41..5f85148 100644
--- a/test/av1_inv_txfm2d_test.cc
+++ b/test/av1_inv_txfm2d_test.cc
@@ -158,20 +158,20 @@
     param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_8X4, 2, 0.016));
     param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_8X16, 2, 0.033));
     param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_16X8, 2, 0.033));
-    param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_16X32, 2, 0.4));
-    param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_32X16, 2, 0.4));
+    param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_16X32, 3, 0.4));
+    param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_32X16, 3, 0.4));
 
-    param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_4X16, 2, 0.1));
-    param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_16X4, 2, 0.1));
+    param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_4X16, 2, 0.2));
+    param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_16X4, 2, 0.2));
     param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_8X32, 2, 0.2));
-    param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_32X8, 2, 0.1));
+    param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_32X8, 2, 0.2));
 
 #if CONFIG_TX64X64
     if (tx_type == DCT_DCT) {  // Other types not supported by these tx sizes.
-      param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_32X64, 3, 0.3));
-      param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_64X32, 3, 0.31));
-      param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_16X64, 2, 0.16));
-      param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_64X16, 2, 0.16));
+      param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_32X64, 4, 0.38));
+      param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_64X32, 4, 0.38));
+      param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_16X64, 3, 0.38));
+      param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_64X16, 3, 0.38));
     }
 #endif  // CONFIG_TX64X64
   }
@@ -201,9 +201,8 @@
                              static_cast<TX_SIZE>(tx_size), &cfg);
         int8_t stage_range_col[MAX_TXFM_STAGE_NUM];
         int8_t stage_range_row[MAX_TXFM_STAGE_NUM];
-        const TX_SIZE tx_size_sqr_up = txsize_sqr_up_map[tx_size];
         av1_gen_inv_stage_range(stage_range_col, stage_range_row, &cfg,
-                                fwd_shift_sum[tx_size_sqr_up], bd);
+                                (TX_SIZE)tx_size, bd);
         const TXFM_1D_CFG *col_cfg = cfg.col_cfg;
         const TXFM_1D_CFG *row_cfg = cfg.row_cfg;
         libaom_test::txfm_stage_range_check(stage_range_col, col_cfg->stage_num,