Optimization for EXT_INTER + OBMC combination

In the rd loop, check the perf of obmc, whose mv is copied from regular
inter predictor, when wedge interinter is better than regular inter
(previously it will force allow_obmc = 0). The condition of the early
termination before this step is relaxed to avoid skipping too many obmc
predictions. The rates of the overhead are properly calculated for these tools.

The logic of the bitstream syntax:
(a single ref) the interintra flag is sent first, only if it is 0, we
send the obmc flag;
(compound refs) the obmc flag is sent first, only if it is 0, we send
the wedge interinter flag

Coding gain
lowres: 0.428% (2.287%->2.715%)

Change-Id: I5f3a34640b398e313cbf84235c9fe2073eb2173f

diff --git a/vp10/common/blockd.h b/vp10/common/blockd.h
index 5cecf79d32800a367cd13f632415bc4361fd5b75..65a8298a2e97ed8ea5d6df6522fd1d5e0b62a445 100644 (file)
--- a/vp10/common/blockd.h
+++ b/vp10/common/blockd.h
@@ -700,7 +700,11 @@ static INLINE int is_interintra_pred(const MB_MODE_INFO *mbmi) {
 
 #if CONFIG_OBMC
 static INLINE int is_obmc_allowed(const MB_MODE_INFO *mbmi) {
+#if CONFIG_EXT_INTER
+  return (mbmi->sb_type >= BLOCK_8X8 && mbmi->ref_frame[1] != INTRA_FRAME);
+#else
   return (mbmi->sb_type >= BLOCK_8X8);
+#endif  // CONFIG_EXT_INTER
 }
 
 static INLINE int is_neighbor_overlappable(const MB_MODE_INFO *mbmi) {

diff --git a/vp10/decoder/decodemv.c b/vp10/decoder/decodemv.c
index b52696d1348b6c76253ec4d22a7a2c1a36e31bfe..32ed20037e52ad656efec1bda103a856ed28e5f3 100644 (file)
--- a/vp10/decoder/decodemv.c
+++ b/vp10/decoder/decodemv.c
@@ -1230,14 +1230,6 @@ static void read_inter_block_mode_info(VP10Decoder *const pbi,
                       mi_row, mi_col, fpm_sync, (void *)pbi, inter_mode_ctx);
   }
 
-#if CONFIG_OBMC
-  mbmi->obmc = 0;
-#if CONFIG_SUPERTX
-  if (!supertx_enabled)
-#endif  // CONFIG_SUPERTX
-    mbmi->obmc = read_is_obmc_block(cm, xd, r);
-#endif  // CONFIG_OBMC
-
 #if CONFIG_REF_MV
 #if CONFIG_EXT_INTER
   if (is_compound)
@@ -1516,12 +1508,8 @@ static void read_inter_block_mode_info(VP10Decoder *const pbi,
   }
 
 #if CONFIG_EXT_INTER
-    mbmi->use_wedge_interintra = 0;
-    mbmi->use_wedge_interinter = 0;
+  mbmi->use_wedge_interintra = 0;
   if (cm->reference_mode != COMPOUND_REFERENCE &&
-#if CONFIG_OBMC
-      !(is_obmc_allowed(mbmi) && mbmi->obmc) &&
-#endif  // CONFIG_OBMC
 #if CONFIG_SUPERTX
       !supertx_enabled &&
 #endif
@@ -1557,6 +1545,21 @@ static void read_inter_block_mode_info(VP10Decoder *const pbi,
       }
     }
   }
+#endif  // CONFIG_EXT_INTER
+
+#if CONFIG_OBMC
+  mbmi->obmc = 0;
+#if CONFIG_SUPERTX
+  if (!supertx_enabled)
+#endif  // CONFIG_SUPERTX
+#if CONFIG_EXT_INTER
+    if (mbmi->ref_frame[1] != INTRA_FRAME)
+#endif  // CONFIG_EXT_INTER
+    mbmi->obmc = read_is_obmc_block(cm, xd, r);
+#endif  // CONFIG_OBMC
+
+#if CONFIG_EXT_INTER
+  mbmi->use_wedge_interinter = 0;
   if (cm->reference_mode != SINGLE_REFERENCE &&
       is_inter_compound_mode(mbmi->mode) &&
 #if CONFIG_OBMC

diff --git a/vp10/encoder/bitstream.c b/vp10/encoder/bitstream.c
index da1885d3ef6a00c889e57954bc2c42c33b3e7b86..dccdf3b75bf823898c0da834ee296d56d57e8f8b 100644 (file)
--- a/vp10/encoder/bitstream.c
+++ b/vp10/encoder/bitstream.c
@@ -1115,14 +1115,6 @@ static void pack_inter_mode_mvs(VP10_COMP *cpi, const MODE_INFO *mi,
     int16_t mode_ctx = mbmi_ext->mode_context[mbmi->ref_frame[0]];
     write_ref_frames(cm, xd, w);
 
-#if CONFIG_OBMC
-#if CONFIG_SUPERTX
-    if (!supertx_enabled)
-#endif  // CONFIG_SUPERTX
-      if (is_obmc_allowed(mbmi))
-        vp10_write(w, mbmi->obmc, cm->fc->obmc_prob[bsize]);
-#endif  // CONFIG_OBMC
-
 #if CONFIG_REF_MV
 #if CONFIG_EXT_INTER
     if (is_compound)
@@ -1286,9 +1278,6 @@ static void pack_inter_mode_mvs(VP10_COMP *cpi, const MODE_INFO *mi,
 
 #if CONFIG_EXT_INTER
     if (cpi->common.reference_mode != COMPOUND_REFERENCE &&
-#if CONFIG_OBMC
-        !(is_obmc_allowed(mbmi) && mbmi->obmc) &&
-#endif  // CONFIG_OBMC
 #if CONFIG_SUPERTX
         !supertx_enabled &&
 #endif  // CONFIG_SUPERTX
@@ -1311,6 +1300,20 @@ static void pack_inter_mode_mvs(VP10_COMP *cpi, const MODE_INFO *mi,
         }
       }
     }
+#endif  // CONFIG_EXT_INTER
+
+#if CONFIG_OBMC
+#if CONFIG_SUPERTX
+    if (!supertx_enabled)
+#endif  // CONFIG_SUPERTX
+#if CONFIG_EXT_INTER
+      if (mbmi->ref_frame[1] != INTRA_FRAME)
+#endif  // CONFIG_EXT_INTER
+      if (is_obmc_allowed(mbmi))
+        vp10_write(w, mbmi->obmc, cm->fc->obmc_prob[bsize]);
+#endif  // CONFIG_OBMC
+
+#if CONFIG_EXT_INTER
     if (cpi->common.reference_mode != SINGLE_REFERENCE &&
         is_inter_compound_mode(mbmi->mode) &&
 #if CONFIG_OBMC

diff --git a/vp10/encoder/encodeframe.c b/vp10/encoder/encodeframe.c
index 6aba4752dcfb770c3eed1606ece3a489d084f3f1..d93d7f35f224a6e8af51fbdd77a0cdede7a3c441 100644 (file)
--- a/vp10/encoder/encodeframe.c
+++ b/vp10/encoder/encodeframe.c
@@ -2008,21 +2008,8 @@ static void update_stats(VP10_COMMON *cm, ThreadData *td
 #endif  // CONFIG_EXT_REFS
         }
 
-#if CONFIG_OBMC
-#if CONFIG_SUPERTX
-        if (!supertx_enabled)
-#endif  // CONFIG_SUPERTX
-          if (is_obmc_allowed(mbmi))
-            counts->obmc[mbmi->sb_type][mbmi->obmc]++;
-#endif  // CONFIG_OBMC
-      }
-    }
-
 #if CONFIG_EXT_INTER
     if (cm->reference_mode != COMPOUND_REFERENCE &&
-#if CONFIG_OBMC
-        !(is_obmc_allowed(mbmi) && mbmi->obmc) &&
-#endif
 #if CONFIG_SUPERTX
         !supertx_enabled &&
 #endif
@@ -2037,15 +2024,31 @@ static void update_stats(VP10_COMMON *cm, ThreadData *td
         counts->interintra[bsize_group][0]++;
       }
     }
-    if (cm->reference_mode != SINGLE_REFERENCE &&
-        is_inter_compound_mode(mbmi->mode) &&
+#endif  // CONFIG_EXT_INTER
+
 #if CONFIG_OBMC
-        !(is_obmc_allowed(mbmi) && mbmi->obmc) &&
+#if CONFIG_SUPERTX
+        if (!supertx_enabled)
+#endif  // CONFIG_SUPERTX
+#if CONFIG_EXT_INTER
+        if (mbmi->ref_frame[1] != INTRA_FRAME)
+#endif  // CONFIG_EXT_INTER
+          if (is_obmc_allowed(mbmi))
+            counts->obmc[mbmi->sb_type][mbmi->obmc]++;
 #endif  // CONFIG_OBMC
-        is_interinter_wedge_used(bsize)) {
-      counts->wedge_interinter[bsize][mbmi->use_wedge_interinter]++;
-    }
+
+#if CONFIG_EXT_INTER
+        if (cm->reference_mode != SINGLE_REFERENCE &&
+            is_inter_compound_mode(mbmi->mode) &&
+#if CONFIG_OBMC
+            !(is_obmc_allowed(mbmi) && mbmi->obmc) &&
+#endif  // CONFIG_OBMC
+            is_interinter_wedge_used(bsize)) {
+          counts->wedge_interinter[bsize][mbmi->use_wedge_interinter]++;
+        }
 #endif  // CONFIG_EXT_INTER
+      }
+    }
 
     if (inter_block &&
         !segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP)) {

diff --git a/vp10/encoder/rdopt.c b/vp10/encoder/rdopt.c
index 918ad3e983fc5277f101d5fcbc12e4b13e9e6bc9..2b08c46c87128dfd917dd026d7e6ea9d258c39ea 100644 (file)
--- a/vp10/encoder/rdopt.c
+++ b/vp10/encoder/rdopt.c
@@ -6248,6 +6248,10 @@ static int64_t handle_inter_mode(VP10_COMP *cpi, MACROBLOCK *x,
   int64_t best_distortion = INT64_MAX;
   unsigned int best_pred_var = UINT_MAX;
   MB_MODE_INFO best_mbmi;
+#if CONFIG_EXT_INTER
+  int rate2_bmc_nocoeff;
+  MB_MODE_INFO best_bmc_mbmi;
+#endif  // CONFIG_EXT_INTER
 #endif  // CONFIG_OBMC
 
   int pred_exists = 0;
@@ -6278,9 +6282,6 @@ static int64_t handle_inter_mode(VP10_COMP *cpi, MACROBLOCK *x,
   assert(!is_comp_interintra_pred || is_interintra_allowed(mbmi));
 #endif  // CONFIG_EXT_INTER
 
-#if CONFIG_OBMC
-  tmp_rd = 0;
-#endif  // CONFIG_OBMC
 #if CONFIG_REF_MV
 #if CONFIG_EXT_INTER
   if (is_comp_pred)
@@ -6643,6 +6644,12 @@ static int64_t handle_inter_mode(VP10_COMP *cpi, MACROBLOCK *x,
   rs = cm->interp_filter == SWITCHABLE ? vp10_get_switchable_rate(cpi, xd) : 0;
 
 #if CONFIG_EXT_INTER
+#if CONFIG_OBMC
+  best_bmc_mbmi = *mbmi;
+  rate2_bmc_nocoeff = *rate2;
+  if (cm->interp_filter == SWITCHABLE)
+    rate2_bmc_nocoeff += rs;
+#endif  // CONFIG_OBMC
   if (is_comp_pred && is_interinter_wedge_used(bsize)) {
     int wedge_index, best_wedge_index = WEDGE_NONE, rs;
     int rate_sum;
@@ -6755,7 +6762,6 @@ static int64_t handle_inter_mode(VP10_COMP *cpi, MACROBLOCK *x,
           xd, bsize, mi_row, mi_col, 1, preds1, strides);
       for (wedge_index = 0; wedge_index < wedge_types; ++wedge_index) {
         mbmi->interinter_wedge_index = wedge_index;
-        // vp10_build_inter_predictors_sb(xd, mi_row, mi_col, bsize);
         vp10_build_wedge_inter_predictor_from_buf(xd, bsize, mi_row, mi_col,
                                                   preds0, strides,
                                                   preds1, strides);
@@ -6774,12 +6780,8 @@ static int64_t handle_inter_mode(VP10_COMP *cpi, MACROBLOCK *x,
         mbmi->use_wedge_interinter = 0;
       }
     }
-#if CONFIG_OBMC
-    if (mbmi->use_wedge_interinter)
-      allow_obmc = 0;
-#endif  // CONFIG_OBMC
     if (ref_best_rd < INT64_MAX &&
-        VPXMIN(best_rd_wedge, best_rd_nowedge) / 2 > ref_best_rd)
+        VPXMIN(best_rd_wedge, best_rd_nowedge) / 3 > ref_best_rd)
       return INT64_MAX;
 
     pred_exists = 0;
@@ -7079,7 +7081,6 @@ static int64_t handle_inter_mode(VP10_COMP *cpi, MACROBLOCK *x,
   memcpy(x->skip_txfm, skip_txfm, sizeof(skip_txfm));
   memcpy(x->bsse, bsse, sizeof(bsse));
 
-
 #if CONFIG_OBMC
   best_rd = INT64_MAX;
   for (mbmi->obmc = 0; mbmi->obmc <= allow_obmc; mbmi->obmc++) {
@@ -7087,6 +7088,12 @@ static int64_t handle_inter_mode(VP10_COMP *cpi, MACROBLOCK *x,
     int tmp_rate;
 
     if (mbmi->obmc) {
+#if CONFIG_EXT_INTER
+      *mbmi = best_bmc_mbmi;
+      assert(!mbmi->use_wedge_interinter);
+      vp10_build_inter_predictors_sb(xd, mi_row, mi_col, bsize);
+      mbmi->obmc = 1;
+#endif  // CONFIG_EXT_INTER
       vp10_build_obmc_inter_prediction(cm, xd, mi_row, mi_col, 0,
                                        NULL, NULL,
                                        dst_buf1, dst_stride1,
@@ -7106,13 +7113,16 @@ static int64_t handle_inter_mode(VP10_COMP *cpi, MACROBLOCK *x,
     x->pred_variance =
         vp10_get_sby_perpixel_variance(cpi, &xd->plane[0].dst, bsize);
 #endif  // CONFIG_VP9_HIGHBITDEPTH
-
     x->skip = 0;
 
+#if CONFIG_EXT_INTER
+    *rate2 = mbmi->obmc ? rate2_bmc_nocoeff : rate2_nocoeff;
+#else
     *rate2 = rate2_nocoeff;
-    *distortion = 0;
+#endif  // CONFIG_EXT_INTER
     if (allow_obmc)
       *rate2 += cpi->obmc_cost[bsize][mbmi->obmc];
+    *distortion = 0;
 #endif  // CONFIG_OBMC
   if (!skip_txfm_sb) {
     int skippable_y, skippable_uv;
@@ -8472,6 +8482,9 @@ void vp10_rd_pick_inter_mode_sb(VP10_COMP *cpi,
 #if CONFIG_EXT_INTER
     rate2 += compmode_interintra_cost;
     if (cm->reference_mode != SINGLE_REFERENCE && comp_pred)
+#if CONFIG_OBMC
+      if (mbmi->obmc == 0)
+#endif  // CONFIG_OBMC
       rate2 += compmode_wedge_cost;
 #endif  // CONFIG_EXT_INTER
 
@@ -8578,7 +8591,7 @@ void vp10_rd_pick_inter_mode_sb(VP10_COMP *cpi,
         *returnrate_nocoef -= vp10_cost_bit(vp10_get_intra_inter_prob(cm, xd),
                                             mbmi->ref_frame[0] != INTRA_FRAME);
 #if CONFIG_OBMC
-        if (is_neighbor_overlappable(mbmi) && is_obmc_allowed(mbmi))
+        if (is_inter_block(mbmi) && is_obmc_allowed(mbmi))
           *returnrate_nocoef -= cpi->obmc_cost[bsize][mbmi->obmc];
 #endif  // CONFIG_OBMC
 #endif  // CONFIG_SUPERTX