Improve motion search in real-time mode

Applied better MV prediction in real-time mode, which improves
the encoding quality.

Used quarter-pixel search instead of iterative sub-pixel search
for speed >=5 to improve encoding performance.

Tests on the test set showed:
1. For speed=-5, quality improvement: 1.7% on AvgPSNR and 2.1%
on SSIM, performance improvement: 3.6% (This counts in the
performance lose caused by MV prediction calculation in "Improve
MV prediction in vp8_pick_inter_mode() for speed>3").
2. For speed=-8, quality improvement: 2.1% on AvgPSNR and 2.5%
on SSIM. but, 6.9% performance decrease because of MV prediction
calculation. This should be improved later.

Change-Id: I349a96c452bd691081d8c8e3e54419e7f477bebd

6 files changed, 13 insertions, 11 deletions

diff --git a/vp8/encoder/mcomp.c b/vp8/encoder/mcomp.c
index d9923fbe9..33aaa2ca9 100644
--- a/vp8/encoder/mcomp.c
+++ b/vp8/encoder/mcomp.c
@@ -779,15 +779,17 @@ int vp8_hex_search
     int *num00,
     const vp8_variance_fn_ptr_t *vfp,
     int *mvsadcost[2],
-    int *mvcost[2]
+    int *mvcost[2],
+    MV *center_mv
 )
 {
     MV hex[6] = { { -1, -2}, {1, -2}, {2, 0}, {1, 2}, { -1, 2}, { -2, 0} } ;
-    MV neighbors[8] = { { -1, -1}, { -1, 0}, { -1, 1}, {0, -1}, {0, 1}, {1, -1}, {1, 0}, {1, 1} } ;
+    MV neighbors[8] = { { -1, -1}, {0, -1}, {1, -1}, { -1, 0}, {1, 0}, { -1, 1}, {0, 1}, {1, 1} } ;
     int i, j;
     unsigned char *src = (*(b->base_src) + b->src);
     int src_stride = b->src_stride;
-    int rr = ref_mv->row, rc = ref_mv->col, br = rr >> 3, bc = rc >> 3, tr, tc;
+    int rr = center_mv->row, rc = center_mv->col;
+    int br = ref_mv->row >> 3, bc = ref_mv->col >> 3, tr, tc;
     unsigned int besterr, thiserr = 0x7fffffff;
     int k = -1, tk;
 
@@ -892,7 +894,7 @@ cal_neighbors:
     best_mv->row = br;
     best_mv->col = bc;
 
-    return vfp->vf(src, src_stride, PRE(br, bc), d->pre_stride, &thiserr) + MVC(br, bc) ;
+    return vfp->vf(src, src_stride, PRE(br, bc), d->pre_stride, &thiserr) + vp8_mv_err_cost(best_mv, center_mv, mvcost, error_per_bit) ;
 }
 #undef MVC
 #undef PRE
diff --git a/vp8/encoder/mcomp.h b/vp8/encoder/mcomp.h
index 7600f87fc..83f95c6e0 100644
--- a/vp8/encoder/mcomp.h
+++ b/vp8/encoder/mcomp.h
@@ -43,8 +43,8 @@ extern int vp8_hex_search
     int *num00,
     const vp8_variance_fn_ptr_t *vf,
     int *mvsadcost[2],
-    int *mvcost[2]
-
+    int *mvcost[2],
+    MV *center_mv
 );
 
 typedef int (fractional_mv_step_fp)
diff --git a/vp8/encoder/onyx_if.c b/vp8/encoder/onyx_if.c
index 912d64179..595c31cec 100644
--- a/vp8/encoder/onyx_if.c
+++ b/vp8/encoder/onyx_if.c
@@ -1118,6 +1118,7 @@ void vp8_set_speed_features(VP8_COMP *cpi)
 #else
             sf->search_method = DIAMOND;
 #endif
+            sf->iterative_sub_pixel = 0;
 
             cpi->mode_check_freq[THR_V_PRED] = 4;
             cpi->mode_check_freq[THR_H_PRED] = 4;
@@ -1169,7 +1170,6 @@ void vp8_set_speed_features(VP8_COMP *cpi)
             int total_skip;
 
             int min = 2000;
-            sf->iterative_sub_pixel = 0;
 
             if (cpi->oxcf.encode_breakout > 2000)
                 min = cpi->oxcf.encode_breakout;
diff --git a/vp8/encoder/pickinter.c b/vp8/encoder/pickinter.c
index 287ca618e..0bfcd38a6 100644
--- a/vp8/encoder/pickinter.c
+++ b/vp8/encoder/pickinter.c
@@ -718,7 +718,7 @@ int vp8_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int rec
 
             if (cpi->sf.search_method == HEX)
             {
-                bestsme = vp8_hex_search(x, b, d, &best_ref_mv, &d->bmi.mv.as_mv, step_param, sadpb/*x->errorperbit*/, &num00, &cpi->fn_ptr[BLOCK_16X16], x->mvsadcost, x->mvcost);
+                bestsme = vp8_hex_search(x, b, d, &mvp, &d->bmi.mv.as_mv, step_param, sadpb/*x->errorperbit*/, &num00, &cpi->fn_ptr[BLOCK_16X16], x->mvsadcost, x->mvcost, &best_ref_mv);
                 mode_mv[NEWMV].row = d->bmi.mv.as_mv.row;
                 mode_mv[NEWMV].col = d->bmi.mv.as_mv.col;
             }
diff --git a/vp8/encoder/rdopt.c b/vp8/encoder/rdopt.c
index ea21cf452..d8b4f5ca7 100644
--- a/vp8/encoder/rdopt.c
+++ b/vp8/encoder/rdopt.c
@@ -1139,7 +1139,7 @@ void vp8_rd_check_segment(VP8_COMP *cpi, MACROBLOCK *x, BEST_SEG_INFO *bsi,
 
                     if (cpi->sf.search_method == HEX)
                         bestsme = vp8_hex_search(x, c, e, bsi->ref_mv,
-                                                 &mode_mv[NEW4X4], step_param, sadpb, &num00, v_fn_ptr, x->mvsadcost, x->mvcost);
+                                                 &mode_mv[NEW4X4], step_param, sadpb, &num00, v_fn_ptr, x->mvsadcost, x->mvcost, bsi->ref_mv);
 
                     else
                     {
@@ -2068,7 +2068,7 @@ int vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int
 
                     if (cpi->sf.search_method == HEX)
                     {
-                        bestsme = vp8_hex_search(x, b, d, &best_ref_mv, &d->bmi.mv.as_mv, step_param, sadpb/*x->errorperbit*/, &num00, &cpi->fn_ptr[BLOCK_16X16], x->mvsadcost, x->mvcost);
+                        bestsme = vp8_hex_search(x, b, d, &best_ref_mv, &d->bmi.mv.as_mv, step_param, sadpb/*x->errorperbit*/, &num00, &cpi->fn_ptr[BLOCK_16X16], x->mvsadcost, x->mvcost, &best_ref_mv);
                         mode_mv[NEWMV].row = d->bmi.mv.as_mv.row;
                         mode_mv[NEWMV].col = d->bmi.mv.as_mv.col;
                     }
diff --git a/vp8/encoder/temporal_filter.c b/vp8/encoder/temporal_filter.c
index f28daaff9..dbe1d0259 100644
--- a/vp8/encoder/temporal_filter.c
+++ b/vp8/encoder/temporal_filter.c
@@ -204,7 +204,7 @@ static int vp8_temporal_filter_find_matching_mb_c
             step_param,
             sadpb/*x->errorperbit*/,
             &num00, &cpi->fn_ptr[BLOCK_16X16],
-            mvsadcost, mvcost);
+            mvsadcost, mvcost, &best_ref_mv1);
     }
     else
     {