Use low precision 32x32fdct for encodemb in speed1

The low precision 32x32 fdct has all the intermediate steps within
16-bit depth, hence allowing faster SSE2 implementation, at the
expense of larger round-trip error. It was used in the rate-distortion
optimization search loop only.

Using the low precision version, in replace of the high precision one,
affects the compression performance by about 0.7% (derf, stdhd) at
speed 0. For speed 1, it makes derf set down by only 0.017%.

Change-Id: I4e7d18fac5bea5317b91c8e7dabae143bc6b5c8b

1 files changed, 2 insertions, 2 deletions

diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 82859c5d7..39ca5efc4 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -565,7 +565,7 @@ static void pick_sb_modes(VP9_COMP *cpi, int mi_row, int mi_col,
   MACROBLOCK *const x = &cpi->mb;
   MACROBLOCKD *const xd = &x->e_mbd;
 
-  x->rd_search = 1;
+  x->use_lp32x32fdct = 1;
 
   if (bsize < BLOCK_8X8) {
     // When ab_index = 0 all sub-blocks are handled, so for ab_index != 0
@@ -2546,7 +2546,7 @@ static void encode_superblock(VP9_COMP *cpi, TOKENEXTRA **t, int output_enabled,
   const int mis = cm->mode_info_stride;
   const int mi_width = num_8x8_blocks_wide_lookup[bsize];
   const int mi_height = num_8x8_blocks_high_lookup[bsize];
-  x->rd_search = 0;
+  x->use_lp32x32fdct = cpi->sf.use_lp32x32fdct;
   x->skip_encode = (!output_enabled && cpi->sf.skip_encode_frame &&
                     xd->q_index < QIDX_SKIP_THRESH);
   if (x->skip_encode)