Refactor highbd forward transform use case

Separate the hybrid transform case from 2D-DCT case. This will
allow us to clear up cross dependency between c and SIMD
implementations later.

Change-Id: Iaa499e8b096850a1c5a0c50a3b6e63e15d0184bf

vp9/encoder/vp9_encodemb.c

@@ -823,7 +823,10 @@ void vp9_encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
         if (!x->skip_recode) {
           vpx_highbd_subtract_block(16, 16, src_diff, diff_stride,
                                     src, src_stride, dst, dst_stride, xd->bd);
-          vp9_highbd_fht16x16(src_diff, coeff, diff_stride, tx_type);
+          if (tx_type == DCT_DCT)
+            vp9_highbd_fdct16x16(src_diff, coeff, diff_stride);
+          else
+            vp9_highbd_fht16x16(src_diff, coeff, diff_stride, tx_type);
           vp9_highbd_quantize_b(coeff, 256, x->skip_block, p->zbin, p->round,
                                 p->quant, p->quant_shift, qcoeff, dqcoeff,
                                 pd->dequant, eob,
@@ -845,7 +848,10 @@ void vp9_encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
         if (!x->skip_recode) {
           vpx_highbd_subtract_block(8, 8, src_diff, diff_stride,
                                     src, src_stride, dst, dst_stride, xd->bd);
-          vp9_highbd_fht8x8(src_diff, coeff, diff_stride, tx_type);
+          if (tx_type == DCT_DCT)
+            vp9_highbd_fdct8x8(src_diff, coeff, diff_stride);
+          else
+            vp9_highbd_fht8x8(src_diff, coeff, diff_stride, tx_type);
           vp9_highbd_quantize_b(coeff, 64, x->skip_block, p->zbin, p->round,
                                 p->quant, p->quant_shift, qcoeff, dqcoeff,
                                 pd->dequant, eob,

vp9/encoder/vp9_rdopt.c

@@ -845,7 +845,10 @@ static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x,
             int64_t unused;
             const TX_TYPE tx_type = get_tx_type_4x4(PLANE_TYPE_Y, xd, block);
             const scan_order *so = &vp9_scan_orders[TX_4X4][tx_type];
-            vp9_highbd_fht4x4(src_diff, coeff, 8, tx_type);
+            if (tx_type == DCT_DCT)
+              vp9_highbd_fdct4x4(src_diff, coeff, 8);
+            else
+              vp9_highbd_fht4x4(src_diff, coeff, 8, tx_type);
             vp9_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
             ratey += cost_coeffs(x, 0, block, tempa + idx, templ + idy, TX_4X4,
                                  so->scan, so->neighbors,