diff --git a/vp8/encoder/mcomp.c b/vp8/encoder/mcomp.c
index d9923fbe99040838068cc542c1e7a63f542ae72a..33aaa2ca9350757035610a96c660335ec599df7f 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 7600f87fc2fea1a3e2f8fbb35f85f0abe42c6f5e..83f95c6e012665eeb9dd684d3d757986f896136b 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 fc908df064d387c8f4454b8987071e56d3990a95..511631ae4675317d4272db11b7808bf755197031 100644
--- a/vp8/encoder/onyx_if.c
+++ b/vp8/encoder/onyx_if.c
@@ -1121,6 +1121,7 @@ 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;
@@ -1172,7 +1173,6 @@ 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 287ca618e5894728ef145c54f129f600c317dbed..0bfcd38a6d525818899a5543b38b48283d3363e9 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 632bca4190a51ef5a2fd4c72ab6999c209915865..7080425c1aa1c30ecdfa14c9cc5110c12581e01e 100644
--- a/vp8/encoder/rdopt.c
+++ b/vp8/encoder/rdopt.c
@@ -1149,7 +1149,7 @@ 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
                     {
@@ -2091,7 +2091,7 @@ 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 3c6d1a4d963dc26e1553ea138e8221fdf389ebc4..e4c3db1b86ba1d787bfc3ac67cf384745868f207 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
     {