Enable mode search threshold update in non-RD coding mode

Adaptively adjust the mode thresholds after each mode search round
to skip checking less likely selected modes. Local tests indicate
5% - 10% speed-up in speed -5 and -6. Average coding performance
loss is -1.055%.

speed -5
vidyo1 720p 1000 kbps
16533 b/f, 40.851 dB, 12607 ms -> 16556 b/f, 40.796 dB, 11831 ms

nik 720p 1000 kbps
33229 b/f, 39.127 dB, 11468 ms -> 33235 b/f, 39.131 dB, 10919 ms

speed -6
vidyo1 720p 1000 kbps
16549 b/f, 40.268 dB, 10138 ms -> 16538 b/f, 40.212 dB, 8456 ms

nik 720p 1000 kbps
33271 b/f, 38.433 dB,  7886 ms -> 33279 b/f, 38.416 dB, 7843 ms

Change-Id: I2c2963f1ce4ed9c1cf233b5b2c880b682e1c1e8b

diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index 1e4c9b1..6928338 100644 (file)
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -450,7 +450,8 @@ static void estimate_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
   args->dist += dist;
 }
 
-static const THR_MODES mode_idx[MAX_REF_FRAMES - 1][INTER_MODES] = {
+static const THR_MODES mode_idx[MAX_REF_FRAMES][4] = {
+  {THR_DC, THR_H_PRED, THR_V_PRED},
   {THR_NEARESTMV, THR_NEARMV, THR_ZEROMV, THR_NEWMV},
   {THR_NEARESTG, THR_NEARG, THR_ZEROG, THR_NEWG},
   {THR_NEARESTA, THR_NEARA, THR_ZEROA, THR_NEWA},
@@ -558,7 +559,7 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
                         EIGHTTAP : cm->interp_filter;
   mbmi->segment_id = segment_id;
 
-  for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
+  for (ref_frame = LAST_FRAME; ref_frame <= GOLDEN_FRAME; ++ref_frame) {
     PREDICTION_MODE this_mode;
     x->pred_mv_sad[ref_frame] = INT_MAX;
     frame_mv[NEWMV][ref_frame].as_int = INVALID_MV;
@@ -610,8 +611,7 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
         continue;
 
       mode_rd_thresh =
-          rd_threshes[mode_idx[ref_frame -
-                               LAST_FRAME][INTER_OFFSET(this_mode)]];
+          rd_threshes[mode_idx[ref_frame][INTER_OFFSET(this_mode)]];
       if (rd_less_than_thresh(best_rdc.rdcost, mode_rd_thresh,
                               rd_thresh_freq_fact[this_mode]))
         continue;
@@ -757,10 +757,9 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
       if (x->skip)
         break;
     }
-    // If the current reference frame is valid and we found a usable mode,
-    // we are done.
-    if (best_rdc.rdcost < INT64_MAX && ref_frame == GOLDEN_FRAME)
-      break;
+
+    // Check that a prediction mode has been selected.
+    assert(best_rdc.rdcost < INT64_MAX);
   }
 
   // If best prediction is not in dst buf, then copy the prediction block from
@@ -836,5 +835,12 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
       pd->dst = orig_dst;
   }
 
+  if (is_inter_block(mbmi))
+    vp9_update_rd_thresh_fact(cpi, tile_data, bsize,
+                              mode_idx[ref_frame][INTER_OFFSET(mbmi->mode)]);
+  else
+    vp9_update_rd_thresh_fact(cpi, tile_data, bsize,
+                              mode_idx[ref_frame][mbmi->mode]);
+
   *rd_cost = best_rdc;
 }

diff --git a/vp9/encoder/vp9_rd.c b/vp9/encoder/vp9_rd.c
index d758430..587ffad 100644 (file)
--- a/vp9/encoder/vp9_rd.c
+++ b/vp9/encoder/vp9_rd.c
@@ -611,6 +611,29 @@ void vp9_set_rd_speed_thresholds_sub8x8(VP9_COMP *cpi) {
       rd->thresh_mult_sub8x8[i] = INT_MAX;
 }
 
+// TODO(jingning) Refactor this function. Use targeted smaller struct as inputs.
+void vp9_update_rd_thresh_fact(VP9_COMP *cpi, TileDataEnc *tile_data,
+                               int bsize, int best_mode_index) {
+  if (cpi->sf.adaptive_rd_thresh > 0) {
+    const int top_mode = bsize < BLOCK_8X8 ? MAX_REFS : MAX_MODES;
+    int mode;
+    for (mode = 0; mode < top_mode; ++mode) {
+      const BLOCK_SIZE min_size = MAX(bsize - 1, BLOCK_4X4);
+      const BLOCK_SIZE max_size = MIN(bsize + 2, BLOCK_64X64);
+      BLOCK_SIZE bs;
+      for (bs = min_size; bs <= max_size; ++bs) {
+        int *const fact = &tile_data->thresh_freq_fact[bs][mode];
+        if (mode == best_mode_index) {
+          *fact -= (*fact >> 4);
+        } else {
+          *fact = MIN(*fact + RD_THRESH_INC,
+                      cpi->sf.adaptive_rd_thresh * RD_THRESH_MAX_FACT);
+        }
+      }
+    }
+  }
+}
+
 int vp9_get_intra_cost_penalty(int qindex, int qdelta,
                                vpx_bit_depth_t bit_depth) {
   const int q = vp9_dc_quant(qindex, qdelta, bit_depth);

diff --git a/vp9/encoder/vp9_rd.h b/vp9/encoder/vp9_rd.h
index 1aa5266..aecca0b 100644 (file)
--- a/vp9/encoder/vp9_rd.h
+++ b/vp9/encoder/vp9_rd.h
@@ -36,6 +36,9 @@ extern "C" {
 #define MAX_MODES 30
 #define MAX_REFS  6
 
+#define RD_THRESH_MAX_FACT 64
+#define RD_THRESH_INC      1
+
 // This enumerator type needs to be kept aligned with the mode order in
 // const MODE_DEFINITION vp9_mode_order[MAX_MODES] used in the rd code.
 typedef enum {
@@ -129,6 +132,7 @@ void vp9_rd_cost_reset(RD_COST *rd_cost);
 void vp9_rd_cost_init(RD_COST *rd_cost);
 
 struct TileInfo;
+struct TileDataEnc;
 struct VP9_COMP;
 struct macroblock;
 
@@ -158,6 +162,10 @@ void vp9_set_rd_speed_thresholds(struct VP9_COMP *cpi);
 
 void vp9_set_rd_speed_thresholds_sub8x8(struct VP9_COMP *cpi);
 
+void vp9_update_rd_thresh_fact(struct VP9_COMP *cpi,
+                               struct TileDataEnc *tile_data,
+                               int bsize, int best_mode_index);
+
 static INLINE int rd_less_than_thresh(int64_t best_rd, int thresh,
                                       int thresh_fact) {
     return best_rd < ((int64_t)thresh * thresh_fact >> 5) || thresh == INT_MAX;

diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 646cc89..795820b 100644 (file)
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -38,9 +38,6 @@
 #include "vp9/encoder/vp9_rdopt.h"
 #include "vp9/encoder/vp9_variance.h"
 
-#define RD_THRESH_MAX_FACT 64
-#define RD_THRESH_INC      1
-
 #define LAST_FRAME_MODE_MASK    ((1 << GOLDEN_FRAME) | (1 << ALTREF_FRAME) | \
                                  (1 << INTRA_FRAME))
 #define GOLDEN_FRAME_MODE_MASK  ((1 << LAST_FRAME) | (1 << ALTREF_FRAME) | \
@@ -2765,29 +2762,6 @@ void vp9_rd_pick_intra_mode_sb(VP9_COMP *cpi, MACROBLOCK *x,
   rd_cost->rdcost = RDCOST(x->rdmult, x->rddiv, rd_cost->rate, rd_cost->dist);
 }
 
-// TODO(jingning) Refactor this function. Use targeted smaller struct as inputs.
-static void update_rd_thresh_fact(VP9_COMP *cpi, TileDataEnc *tile_data,
-                                  int bsize, int best_mode_index) {
-  if (cpi->sf.adaptive_rd_thresh > 0) {
-    const int top_mode = bsize < BLOCK_8X8 ? MAX_REFS : MAX_MODES;
-    int mode;
-    for (mode = 0; mode < top_mode; ++mode) {
-      const BLOCK_SIZE min_size = MAX(bsize - 1, BLOCK_4X4);
-      const BLOCK_SIZE max_size = MIN(bsize + 2, BLOCK_64X64);
-      BLOCK_SIZE bs;
-      for (bs = min_size; bs <= max_size; ++bs) {
-        int *const fact = &tile_data->thresh_freq_fact[bs][mode];
-        if (mode == best_mode_index) {
-          *fact -= (*fact >> 4);
-        } else {
-          *fact = MIN(*fact + RD_THRESH_INC,
-                      cpi->sf.adaptive_rd_thresh * RD_THRESH_MAX_FACT);
-        }
-      }
-    }
-  }
-}
-
 void vp9_rd_pick_inter_mode_sb(VP9_COMP *cpi,
                                TileDataEnc *tile_data,
                                MACROBLOCK *x,
@@ -3423,7 +3397,7 @@ void vp9_rd_pick_inter_mode_sb(VP9_COMP *cpi,
          !is_inter_block(&best_mbmode));
 
   if (!cpi->rc.is_src_frame_alt_ref)
-    update_rd_thresh_fact(cpi, tile_data, bsize, best_mode_index);
+    vp9_update_rd_thresh_fact(cpi, tile_data, bsize, best_mode_index);
 
   // macroblock modes
   *mbmi = best_mbmode;
@@ -3578,7 +3552,7 @@ void vp9_rd_pick_inter_mode_sb_seg_skip(VP9_COMP *cpi,
   assert((cm->interp_filter == SWITCHABLE) ||
          (cm->interp_filter == mbmi->interp_filter));
 
-  update_rd_thresh_fact(cpi, tile_data, bsize, THR_ZEROMV);
+  vp9_update_rd_thresh_fact(cpi, tile_data, bsize, THR_ZEROMV);
 
   vp9_zero(best_pred_diff);
   vp9_zero(best_filter_diff);
@@ -4153,7 +4127,7 @@ void vp9_rd_pick_inter_mode_sub8x8(VP9_COMP *cpi,
          (cm->interp_filter == best_mbmode.interp_filter) ||
          !is_inter_block(&best_mbmode));
 
-  update_rd_thresh_fact(cpi, tile_data, bsize, best_ref_index);
+  vp9_update_rd_thresh_fact(cpi, tile_data, bsize, best_ref_index);
 
   // macroblock modes
   *mbmi = best_mbmode;

diff --git a/vp9/encoder/vp9_speed_features.c b/vp9/encoder/vp9_speed_features.c
index e148bf9..294d5cc 100644 (file)
--- a/vp9/encoder/vp9_speed_features.c
+++ b/vp9/encoder/vp9_speed_features.c
@@ -269,7 +269,7 @@ static void set_rt_speed_feature(VP9_COMP *cpi, SPEED_FEATURES *sf,
     sf->inter_mode_mask[BLOCK_32X64] = INTER_NEAREST_NEW_ZERO;
     sf->inter_mode_mask[BLOCK_64X32] = INTER_NEAREST_NEW_ZERO;
     sf->inter_mode_mask[BLOCK_64X64] = INTER_NEAREST_NEW_ZERO;
-
+    sf->adaptive_rd_thresh = 2;
     // This feature is only enabled when partition search is disabled.
     sf->reuse_inter_pred_sby = 1;
 
@@ -292,12 +292,7 @@ static void set_rt_speed_feature(VP9_COMP *cpi, SPEED_FEATURES *sf,
     sf->partition_search_type = VAR_BASED_PARTITION;
     sf->search_type_check_frequency = 50;
     sf->mv.search_method = NSTEP;
-
     sf->tx_size_search_method = is_keyframe ? USE_LARGESTALL : USE_TX_8X8;
-
-    // Increase mode checking threshold for NEWMV.
-    sf->elevate_newmv_thresh = 1000;
-
     sf->mv.reduce_first_step_size = 1;
   }