Fixing a bug in obmc prediction in the rd loop

This bug made the rd loop use one-side obmc (compound of the current
predictor and the predictors of the left mi's, while the above ones
are ignored by mistake) to determine whether to use obmc. This fix
improved the compression performance by ~0.6% on different test sets.

Coding gain (%) of obmc experiment on derflr/derfhd/hevcmr/hevchd:
1.568/TBD/1.628/TBD

Change-Id: I43b239bedf9a8eebfd02315b1b036e140a998140
diff --git a/vp10/common/reconinter.c b/vp10/common/reconinter.c
index efc1ee1..c1cb69d 100644
--- a/vp10/common/reconinter.c
+++ b/vp10/common/reconinter.c
@@ -654,8 +654,6 @@
       uint8_t *dst = use_tmp_dst_buf ?
           &final_buf[plane][(i * 8) >> pd->subsampling_x] :
           &pd->dst.buf[(i * 8) >> pd->subsampling_x];
-      int bmc_stride = pd->dst.stride;
-      uint8_t *bmc = &pd->dst.buf[(i * 8) >> pd->subsampling_x];
       int tmp_stride = tmp_stride1[plane];
       uint8_t *tmp = &tmp_buf1[plane][(i * 8) >> pd->subsampling_x];
       const uint8_t *mask[2];
@@ -665,27 +663,22 @@
 #if CONFIG_VP9_HIGHBITDEPTH
       if (is_hbd) {
         uint16_t *dst16 = CONVERT_TO_SHORTPTR(dst);
-        uint16_t *bmc16 = CONVERT_TO_SHORTPTR(bmc);
         uint16_t *tmp16 = CONVERT_TO_SHORTPTR(tmp);
 
         for (row = 0; row < bh; ++row) {
-          for (col = 0; col < bw; ++col) {
-            dst16[col] = (mask[0][row] * bmc16[col] + mask[1][row] * tmp16[col]
+          for (col = 0; col < bw; ++col)
+            dst16[col] = (mask[0][row] * dst16[col] + mask[1][row] * tmp16[col]
                           + 32) >> 6;
-          }
           dst16 += dst_stride;
-          bmc16 += bmc_stride;
           tmp16 += tmp_stride;
         }
       } else {
 #endif  // CONFIG_VP9_HIGHBITDEPTH
       for (row = 0; row < bh; ++row) {
-        for (col = 0; col < bw; ++col) {
-          dst[col] = (mask[0][row] * bmc[col] + mask[1][row] * tmp[col] + 32)
+        for (col = 0; col < bw; ++col)
+          dst[col] = (mask[0][row] * dst[col] + mask[1][row] * tmp[col] + 32)
                      >> 6;
-        }
         dst += dst_stride;
-        bmc += bmc_stride;
         tmp += tmp_stride;
       }
 #if CONFIG_VP9_HIGHBITDEPTH
@@ -727,8 +720,6 @@
       uint8_t *dst = use_tmp_dst_buf ?
           &final_buf[plane][(i * 8 * dst_stride) >> pd->subsampling_y] :
           &pd->dst.buf[(i * 8 * dst_stride) >> pd->subsampling_y];
-      int bmc_stride = pd->dst.stride;
-      uint8_t *bmc = &pd->dst.buf[(i * 8 * bmc_stride) >> pd->subsampling_y];
       int tmp_stride = tmp_stride2[plane];
       uint8_t *tmp = &tmp_buf2[plane]
                               [(i * 8 * tmp_stride) >> pd->subsampling_y];
@@ -739,27 +730,22 @@
 #if CONFIG_VP9_HIGHBITDEPTH
       if (is_hbd) {
         uint16_t *dst16 = CONVERT_TO_SHORTPTR(dst);
-        uint16_t *bmc16 = CONVERT_TO_SHORTPTR(bmc);
         uint16_t *tmp16 = CONVERT_TO_SHORTPTR(tmp);
 
         for (row = 0; row < bh; ++row) {
-          for (col = 0; col < bw; ++col) {
-            dst16[col] = (mask[0][row] * bmc16[col] + mask[1][row] * tmp16[col]
+          for (col = 0; col < bw; ++col)
+            dst16[col] = (mask[0][row] * dst16[col] + mask[1][row] * tmp16[col]
                           + 32) >> 6;
-          }
           dst16 += dst_stride;
-          bmc16 += bmc_stride;
           tmp16 += tmp_stride;
         }
       } else {
 #endif  // CONFIG_VP9_HIGHBITDEPTH
       for (row = 0; row < bh; ++row) {
-        for (col = 0; col < bw; ++col) {
-          dst[col] = (mask[0][col] * bmc[col] + mask[1][col] * tmp[col] + 32)
+        for (col = 0; col < bw; ++col)
+          dst[col] = (mask[0][col] * dst[col] + mask[1][col] * tmp[col] + 32)
                      >> 6;
-        }
         dst += dst_stride;
-        bmc += bmc_stride;
         tmp += tmp_stride;
       }
 #if CONFIG_VP9_HIGHBITDEPTH
diff --git a/vp10/encoder/rdopt.c b/vp10/encoder/rdopt.c
index 30b8406..736adbb 100644
--- a/vp10/encoder/rdopt.c
+++ b/vp10/encoder/rdopt.c
@@ -5089,6 +5089,8 @@
   DECLARE_ALIGNED(16, uint8_t, tmp_buf[MAX_MB_PLANE * 64 * 64]);
 #endif  // CONFIG_VP9_HIGHBITDEPTH
 #if CONFIG_OBMC
+  int allow_obmc = is_obmc_allowed(mbmi);
+  int best_obmc_flag = 0;
 #if CONFIG_VP9_HIGHBITDEPTH
   DECLARE_ALIGNED(16, uint16_t, tmp_buf1_16[MAX_MB_PLANE * 64 * 64]);
   uint8_t *tmp_buf1;
@@ -5098,13 +5100,11 @@
   uint8_t *obmc_tmp_buf[3] = {tmp_buf1, tmp_buf1 + 4096, tmp_buf1 + 8192};
 #endif  // CONFIG_VP9_HIGHBITDEPTH
   int obmc_tmp_stride[3] = {64, 64, 64};
-  int best_obmc_flag = 0;
   uint8_t tmp_skip_txfm[MAX_MB_PLANE << 2] = {0};
   int64_t tmp_bsse[MAX_MB_PLANE << 2] = {0};
   int64_t rdobmc;
   int skip_txfm_sb_obmc = 0;
   int64_t skip_sse_sb_obmc = INT64_MAX;
-  int allow_obmc = is_obmc_allowed(mbmi);
 #endif  // CONFIG_OBMC
   int pred_exists = 0;
   int intpel_mv;
@@ -5587,6 +5587,7 @@
 #if CONFIG_OBMC
     int tmp_rate_obmc;
     int64_t tmp_dist_obmc;
+    restore_dst_buf(xd, orig_dst, orig_dst_stride);
 #endif  // CONFIG_OBMC
     // Handles the special case when a filter that is not in the
     // switchable list (ex. bilinear) is indicated at the frame level, or
@@ -5594,19 +5595,14 @@
     vp10_build_inter_predictors_sb(xd, mi_row, mi_col, bsize);
 #if CONFIG_OBMC
     if (mbmi->obmc) {
-      vp10_build_obmc_inter_prediction(cm, xd, mi_row, mi_col, 1,
-                                       obmc_tmp_buf, obmc_tmp_stride,
+      vp10_build_obmc_inter_prediction(cm, xd, mi_row, mi_col, 0,
+                                       NULL, NULL,
                                        dst_buf1, dst_stride1,
                                        dst_buf2, dst_stride2);
-      for (i = 0; i < MAX_MB_PLANE; ++i) {
-        xd->plane[i].dst.buf = obmc_tmp_buf[i];
-        xd->plane[i].dst.stride = obmc_tmp_stride[i];
-      }
       model_rd_for_sb(cpi, bsize, x, xd, &tmp_rate, &tmp_dist,
                       &skip_txfm_sb, &skip_sse_sb);
       rd = RDCOST(x->rdmult, x->rddiv,
-                  rs + tmp_rate + cpi->obmc_cost[bsize][1],
-                  tmp_dist);
+                  rs + tmp_rate + cpi->obmc_cost[bsize][1], tmp_dist);
     } else {
 #endif  // CONFIG_OBMC
     model_rd_for_sb(cpi, bsize, x, xd, &tmp_rate, &tmp_dist,