commit 1a2f245278aec2e945e55d5f1b9471e32da255b0
author Angie Chiang <angiebird@google.com> Wed Feb 07 19:42:13 2018 -0800
committer Angie Chiang <angiebird@google.com> Fri Feb 09 18:07:47 2018 +0000
tree 70974f936b5680a695b1bfe15f93ae181347f7f5
parent dbc677db5d64c683900ea38e83202cb9480bc60d

Hook in av1_lowbd_inv_txfm2d_add_8x8_sse2

Change-Id: Ib385a62aa269fd4eae883080bf1172ef58f9bf2d

av1/common/av1_rtcd_defs.pl

diff --git a/av1/common/av1_rtcd_defs.pl b/av1/common/av1_rtcd_defs.pl
index f3fd840..64c052e 100755
--- a/av1/common/av1_rtcd_defs.pl
+++ b/av1/common/av1_rtcd_defs.pl
@@ -227,6 +227,9 @@
 add_proto qw/void av1_highbd_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, const struct txfm_param *param";
 
 #inv txfm
+add_proto qw/void av1_inv_txfm_add/, "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param";
+specialize qw/av1_inv_txfm_add sse2/;
+
 add_proto qw/void av1_inv_txfm2d_add_4x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 add_proto qw/void av1_inv_txfm2d_add_8x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 add_proto qw/void av1_inv_txfm2d_add_8x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";

av1/common/idct.c

diff --git a/av1/common/idct.c b/av1/common/idct.c
index 07fef41..e939795 100644
--- a/av1/common/idct.c
+++ b/av1/common/idct.c
@@ -1625,7 +1625,7 @@
 }
 
 static void av1_highbd_inv_txfm_add(const tran_low_t *input, uint8_t *dest,
-                                    int stride, TxfmParam *txfm_param) {
+                                    int stride, const TxfmParam *txfm_param) {
   assert(av1_ext_tx_used[txfm_param->tx_set_type][txfm_param->tx_type]);
   const TX_SIZE tx_size = txfm_param->tx_size;
   switch (tx_size) {
@@ -1695,8 +1695,8 @@
   }
 }
 
-static void av1_inv_txfm_add(const tran_low_t *dqcoeff, uint8_t *dst,
-                             int stride, TxfmParam *txfm_param) {
+void av1_inv_txfm_add_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride,
+                        const TxfmParam *txfm_param) {
   const TX_SIZE tx_size = txfm_param->tx_size;
   DECLARE_ALIGNED(16, uint16_t, tmp[MAX_TX_SQUARE]);
   int tmp_stride = MAX_TX_SIZE;
@@ -1719,7 +1719,7 @@
 }
 
 typedef void (*InvTxfmFunc)(const tran_low_t *dqcoeff, uint8_t *dst, int stride,
-                            TxfmParam *txfm_param);
+                            const TxfmParam *txfm_param);
 void av1_inverse_transform_block(const MACROBLOCKD *xd,
                                  const tran_low_t *dqcoeff, int plane,
                                  TX_TYPE tx_type, TX_SIZE tx_size, uint8_t *dst,
@@ -1733,9 +1733,8 @@
                   &txfm_param);
   assert(av1_ext_tx_used[txfm_param.tx_set_type][txfm_param.tx_type]);
 
-  static const InvTxfmFunc inv_txfm_func[2] = {
-    av1_inv_txfm_add, av1_highbd_inv_txfm_add,
-  };
-
-  inv_txfm_func[txfm_param.is_hbd](dqcoeff, dst, stride, &txfm_param);
+  if (txfm_param.is_hbd)
+    av1_highbd_inv_txfm_add(dqcoeff, dst, stride, &txfm_param);
+  else
+    av1_inv_txfm_add(dqcoeff, dst, stride, &txfm_param);
 }

av1/common/x86/av1_inv_txfm_sse2.c

diff --git a/av1/common/x86/av1_inv_txfm_sse2.c b/av1/common/x86/av1_inv_txfm_sse2.c
index 1122316..269e2fc 100644
--- a/av1/common/x86/av1_inv_txfm_sse2.c
+++ b/av1/common/x86/av1_inv_txfm_sse2.c
@@ -1849,3 +1849,45 @@
   round_shift_16bit(buf0, buf_size_h, shift[1]);
   lowbd_write_buffer_8xn_sse2(buf0, output, stride, ud_flip, buf_size_h);
 }
+
+typedef void (*inv_txfm_func)(const int32_t *input, uint8_t *output, int stride,
+                              TX_TYPE tx_type, int bd);
+
+static inv_txfm_func inv_txfm_func_ls[TX_SIZES_ALL] = {
+  NULL,                               // 4x4 transform
+  av1_lowbd_inv_txfm2d_add_8x8_sse2,  // 8x8 transform
+  NULL,                               // 16x16 transform
+  NULL,                               // 32x32 transform
+#if CONFIG_TX64X64
+  NULL,  // 64x64 transform
+#endif   // CONFIG_TX64X64
+  NULL,  // 4x8 transform
+  NULL,  // 8x4 transform
+  NULL,  // 8x16 transform
+  NULL,  // 16x8 transform
+  NULL,  // 16x32 transform
+  NULL,  // 32x16 transform
+#if CONFIG_TX64X64
+  NULL,  // 32x64 transform
+  NULL,  // 64x32 transform
+#endif   // CONFIG_TX64X64
+  NULL,  // 4x16 transform
+  NULL,  // 16x4 transform
+  NULL,  // 8x32 transform
+  NULL,  // 32x8 transform
+#if CONFIG_TX64X64
+  NULL,  // 16x64 transform
+  NULL,  // 64x16 transform
+#endif   // CONFIG_TX64X64
+};
+
+void av1_inv_txfm_add_sse2(const tran_low_t *dqcoeff, uint8_t *dst, int stride,
+                           const TxfmParam *txfm_param) {
+  const TX_TYPE tx_type = txfm_param->tx_type;
+  const inv_txfm_func inv_func = inv_txfm_func_ls[txfm_param->tx_size];
+  if (inv_func != NULL) {
+    inv_func(dqcoeff, dst, stride, tx_type, txfm_param->bd);
+  } else {
+    av1_inv_txfm_add_c(dqcoeff, dst, stride, txfm_param);
+  }
+}