Fix 2x2 high bit-depth transform setups

This commit fixes the 2x2 transform system setups for high bit-
depth setting. It enables the cb4x4 mode to support high bit-depth
process. The coding performance is improved over high bit-depth +
ref-mv:

lowres  2.5%
midres  1.2%

Change-Id: I351f9d72bdc7e15b2bd00e94286b98966a295e6d
diff --git a/av1/common/av1_fwd_txfm2d.c b/av1/common/av1_fwd_txfm2d.c
index 2bba193..41f3e71 100644
--- a/av1/common/av1_fwd_txfm2d.c
+++ b/av1/common/av1_fwd_txfm2d.c
@@ -138,16 +138,32 @@
   { &fwd_txfm_2d_cfg_adst_adst_4, &fwd_txfm_2d_cfg_adst_adst_8,
     &fwd_txfm_2d_cfg_adst_adst_16, &fwd_txfm_2d_cfg_adst_adst_32 },
 };
-#else   // CONFIG_EXT_TX
+#else  // CONFIG_EXT_TX
 static const TXFM_2D_CFG *fwd_txfm_cfg_ls[TX_TYPES][TX_SIZES] = {
-  { &fwd_txfm_2d_cfg_dct_dct_4, &fwd_txfm_2d_cfg_dct_dct_8,
-    &fwd_txfm_2d_cfg_dct_dct_16, &fwd_txfm_2d_cfg_dct_dct_32 },
-  { &fwd_txfm_2d_cfg_adst_dct_4, &fwd_txfm_2d_cfg_adst_dct_8,
-    &fwd_txfm_2d_cfg_adst_dct_16, &fwd_txfm_2d_cfg_adst_dct_32 },
-  { &fwd_txfm_2d_cfg_dct_adst_4, &fwd_txfm_2d_cfg_dct_adst_8,
-    &fwd_txfm_2d_cfg_dct_adst_16, &fwd_txfm_2d_cfg_dct_adst_32 },
-  { &fwd_txfm_2d_cfg_adst_adst_4, &fwd_txfm_2d_cfg_adst_adst_8,
-    &fwd_txfm_2d_cfg_adst_adst_16, &fwd_txfm_2d_cfg_adst_adst_32 },
+  {
+#if CONFIG_CB4X4
+      NULL,
+#endif
+      &fwd_txfm_2d_cfg_dct_dct_4, &fwd_txfm_2d_cfg_dct_dct_8,
+      &fwd_txfm_2d_cfg_dct_dct_16, &fwd_txfm_2d_cfg_dct_dct_32 },
+  {
+#if CONFIG_CB4X4
+      NULL,
+#endif
+      &fwd_txfm_2d_cfg_adst_dct_4, &fwd_txfm_2d_cfg_adst_dct_8,
+      &fwd_txfm_2d_cfg_adst_dct_16, &fwd_txfm_2d_cfg_adst_dct_32 },
+  {
+#if CONFIG_CB4X4
+      NULL,
+#endif
+      &fwd_txfm_2d_cfg_dct_adst_4, &fwd_txfm_2d_cfg_dct_adst_8,
+      &fwd_txfm_2d_cfg_dct_adst_16, &fwd_txfm_2d_cfg_dct_adst_32 },
+  {
+#if CONFIG_CB4X4
+      NULL,
+#endif
+      &fwd_txfm_2d_cfg_adst_adst_4, &fwd_txfm_2d_cfg_adst_adst_8,
+      &fwd_txfm_2d_cfg_adst_adst_16, &fwd_txfm_2d_cfg_adst_adst_32 },
 };
 #endif  // CONFIG_EXT_TX
 
diff --git a/av1/common/av1_inv_txfm2d.c b/av1/common/av1_inv_txfm2d.c
index d8dc83f..df772ef 100644
--- a/av1/common/av1_inv_txfm2d.c
+++ b/av1/common/av1_inv_txfm2d.c
@@ -52,14 +52,30 @@
 };
 #else
 static const TXFM_2D_CFG *inv_txfm_cfg_ls[TX_TYPES][TX_SIZES] = {
-  { &inv_txfm_2d_cfg_dct_dct_4, &inv_txfm_2d_cfg_dct_dct_8,
-    &inv_txfm_2d_cfg_dct_dct_16, &inv_txfm_2d_cfg_dct_dct_32 },
-  { &inv_txfm_2d_cfg_adst_dct_4, &inv_txfm_2d_cfg_adst_dct_8,
-    &inv_txfm_2d_cfg_adst_dct_16, &inv_txfm_2d_cfg_adst_dct_32 },
-  { &inv_txfm_2d_cfg_dct_adst_4, &inv_txfm_2d_cfg_dct_adst_8,
-    &inv_txfm_2d_cfg_dct_adst_16, &inv_txfm_2d_cfg_dct_adst_32 },
-  { &inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
-    &inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
+  {
+#if CONFIG_CB4X4
+      NULL,
+#endif
+      &inv_txfm_2d_cfg_dct_dct_4, &inv_txfm_2d_cfg_dct_dct_8,
+      &inv_txfm_2d_cfg_dct_dct_16, &inv_txfm_2d_cfg_dct_dct_32 },
+  {
+#if CONFIG_CB4X4
+      NULL,
+#endif
+      &inv_txfm_2d_cfg_adst_dct_4, &inv_txfm_2d_cfg_adst_dct_8,
+      &inv_txfm_2d_cfg_adst_dct_16, &inv_txfm_2d_cfg_adst_dct_32 },
+  {
+#if CONFIG_CB4X4
+      NULL,
+#endif
+      &inv_txfm_2d_cfg_dct_adst_4, &inv_txfm_2d_cfg_dct_adst_8,
+      &inv_txfm_2d_cfg_dct_adst_16, &inv_txfm_2d_cfg_dct_adst_32 },
+  {
+#if CONFIG_CB4X4
+      NULL,
+#endif
+      &inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
+      &inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
 };
 #endif
 
diff --git a/av1/common/idct.c b/av1/common/idct.c
index 948ac14..1858a0a 100644
--- a/av1/common/idct.c
+++ b/av1/common/idct.c
@@ -2511,8 +2511,8 @@
 
 #if CONFIG_CB4X4
 void av1_highbd_inv_txfm_add_2x2(const tran_low_t *input, uint8_t *dest,
-                                 int stride, int eob, TX_TYPE tx_type,
-                                 int lossless, int bd) {
+                                 int stride, int eob, int bd, TX_TYPE tx_type,
+                                 int lossless) {
   tran_high_t a1 = input[0] >> UNIT_QUANT_SHIFT;
   tran_high_t b1 = input[1] >> UNIT_QUANT_SHIFT;
   tran_high_t c1 = input[2] >> UNIT_QUANT_SHIFT;