tx64x64: Use C version for DCT/IDCT transform.

The SSE4 function does not support 64x64 size, and was triggering an
assertion failure lowbitdepth is disabled.

BUG=aomedia:672

Change-Id: Id14e76b5c180a211a84c2e933b07e8acf72dddbc

av1/common/av1_fwd_txfm2d.c

diff --git a/av1/common/av1_fwd_txfm2d.c b/av1/common/av1_fwd_txfm2d.c
index bbd08e2..90c571e 100644
--- a/av1/common/av1_fwd_txfm2d.c
+++ b/av1/common/av1_fwd_txfm2d.c
@@ -24,6 +24,9 @@
     case TXFM_TYPE_DCT8: return av1_fdct8_new;
     case TXFM_TYPE_DCT16: return av1_fdct16_new;
     case TXFM_TYPE_DCT32: return av1_fdct32_new;
+#if CONFIG_TX64X64
+    case TXFM_TYPE_DCT64: return av1_fdct64_new;
+#endif  // CONFIG_TX64X64
     case TXFM_TYPE_ADST4: return av1_fadst4_new;
     case TXFM_TYPE_ADST8: return av1_fadst8_new;
     case TXFM_TYPE_ADST16: return av1_fadst16_new;

av1/common/av1_inv_txfm2d.c

diff --git a/av1/common/av1_inv_txfm2d.c b/av1/common/av1_inv_txfm2d.c
index ddcb00f..ba5ae4d 100644
--- a/av1/common/av1_inv_txfm2d.c
+++ b/av1/common/av1_inv_txfm2d.c
@@ -22,6 +22,9 @@
     case TXFM_TYPE_DCT8: return av1_idct8_new;
     case TXFM_TYPE_DCT16: return av1_idct16_new;
     case TXFM_TYPE_DCT32: return av1_idct32_new;
+#if CONFIG_TX64X64
+    case TXFM_TYPE_DCT64: return av1_idct64_new;
+#endif  // CONFIG_TX64X64
     case TXFM_TYPE_ADST4: return av1_iadst4_new;
     case TXFM_TYPE_ADST8: return av1_iadst8_new;
     case TXFM_TYPE_ADST16: return av1_iadst16_new;

av1/common/av1_rtcd_defs.pl

diff --git a/av1/common/av1_rtcd_defs.pl b/av1/common/av1_rtcd_defs.pl
index dc6a844..999d975 100755
--- a/av1/common/av1_rtcd_defs.pl
+++ b/av1/common/av1_rtcd_defs.pl
@@ -412,7 +412,6 @@
 add_proto qw/void av1_fwd_txfm2d_32x32/, "const int16_t *input, int32_t *output, int stride, int tx_type, int bd";
 specialize qw/av1_fwd_txfm2d_32x32 sse4_1/;
 add_proto qw/void av1_fwd_txfm2d_64x64/, "const int16_t *input, int32_t *output, int stride, int tx_type, int bd";
-specialize qw/av1_fwd_txfm2d_64x64 sse4_1/;
 
 #
 # Motion search
@@ -457,11 +456,6 @@
   add_proto qw/int64_t av1_highbd_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz, int bd";
   specialize qw/av1_highbd_block_error sse2/;
 
-  # fdct functions
-  if (aom_config("CONFIG_TX64X64") eq "yes") {
-    add_proto qw/void av1_highbd_fht64x64/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
-  }
-
   add_proto qw/void av1_highbd_temporal_filter_apply/, "uint8_t *frame1, unsigned int stride, uint8_t *frame2, unsigned int block_width, unsigned int block_height, int strength, int filter_weight, unsigned int *accumulator, uint16_t *count";
 
 }