vp9: Use sb content measure to bias against golden.

For each superblock, keep track of how far from current frame
was the last significant content change, and use that (along
with GF distance), to turnoff GF search in non-rd pickmode.

Only enabled for speed >= 8.

avgPNSR on RTC/RTC_derf down by ~0.9/1.2.
Speedup on mac: ~3-5%.
Speedup on arm: 3.6% for VGA and 4.4% for HD.

Change-Id: Ic3f3d6a2af650aca6ba0064d2b1db8d48c035ac7

diff --git a/vp9/encoder/vp9_block.h b/vp9/encoder/vp9_block.h
index c86c818aaadba485fc770fa3dc595e9c9ec3885e..071a9b3e353396960a56cea1330027e8215f4772 100644 (file)
--- a/vp9/encoder/vp9_block.h
+++ b/vp9/encoder/vp9_block.h
@@ -170,6 +170,8 @@ struct macroblock {
 
   uint8_t skip_low_source_sad;
 
+  uint8_t last_sb_high_content;
+
   // Used to save the status of whether a block has a low variance in
   // choose_partitioning. 0 for 64x64, 1~2 for 64x32, 3~4 for 32x64, 5~8 for
   // 32x32, 9~24 for 16x16.

diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index df72667d4244bc0f9d97220e9371f20ad44e039a..e6f4901fdb7dbb0a98762485ef5b71993e1a5fe6 100644 (file)
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -1018,6 +1018,7 @@ static int choose_partitioning(VP9_COMP *cpi, const TileInfo *const tile,
                               content_state == kLowSadHighSumdiff)
                                  ? 1
                                  : 0;
+    x->last_sb_high_content = cpi->content_state_sb_fd[sb_offset2];
     // If source_sad is low copy the partition without computing the y_sad.
     if (x->skip_low_source_sad && cpi->sf.copy_partition_flag &&
         copy_partitioning(cpi, x, mi_row, mi_col, segment_id, sb_offset)) {

diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index 64d6985b55fba3042c305ab8d0017ba027102fb2..aba2c115a370f7b7ae1628f61e2fc53b60eb6620 100644 (file)
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -466,6 +466,9 @@ static void dealloc_compressor_data(VP9_COMP *cpi) {
   vpx_free(cpi->content_state_sb);
   cpi->content_state_sb = NULL;
 
+  vpx_free(cpi->content_state_sb_fd);
+  cpi->content_state_sb_fd = NULL;
+
   vp9_cyclic_refresh_free(cpi->cyclic_refresh);
   cpi->cyclic_refresh = NULL;
 
@@ -3141,10 +3144,14 @@ static void encode_without_recode_loop(VP9_COMP *cpi, size_t *size,
         cpi->svc.current_superframe < 1)) ||
       cpi->resize_pending || cpi->resize_state || cpi->external_resize) {
     compute_source_sad = 0;
-    if (cpi->content_state_sb != NULL)
+    if (cpi->content_state_sb != NULL) {
       memset(cpi->content_state_sb, 0, (cm->mi_stride >> 3) *
                                            ((cm->mi_rows >> 3) + 1) *
                                            sizeof(*cpi->content_state_sb));
+      memset(cpi->content_state_sb_fd, 0,
+             (cm->mi_stride >> 3) * ((cm->mi_rows >> 3) + 1) *
+                 sizeof(*cpi->content_state_sb_fd));
+    }
   }
 
   // Avoid scaling last_source unless its needed.

diff --git a/vp9/encoder/vp9_encoder.h b/vp9/encoder/vp9_encoder.h
index ba29cabb1ffd2ca6764d5f842f1d263e96f4c1d6..72c0ba78706ac1604cb6a1a642bf58e5f4a3f40d 100644 (file)
--- a/vp9/encoder/vp9_encoder.h
+++ b/vp9/encoder/vp9_encoder.h
@@ -705,7 +705,12 @@ typedef struct VP9_COMP {
   uint8_t *copied_frame_cnt;
   uint8_t max_copied_frame;
 
+  // For each superblock: saves the content value (e.g., low/high sad/sumdiff)
+  // based on source sad, prior to encoding the frame.
   uint8_t *content_state_sb;
+  // For each superblock: keeps track of the last time (in frame distance) the
+  // the superblock did not have low source sad.
+  uint8_t *content_state_sb_fd;
 
   LevelConstraint level_constraint;
 } VP9_COMP;

diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index 46fb67a93818f24e03488d89a2bc08aaa92a0bd2..ecc9afbad656232aaf03de0e915b06a14da31f74 100644 (file)
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -1563,6 +1563,11 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data,
     usable_ref_frame = LAST_FRAME;
 #endif
 
+  if (cpi->oxcf.speed >= 8 && !cpi->use_svc &&
+      ((cpi->rc.frames_since_golden + 1) < x->last_sb_high_content ||
+       x->last_sb_high_content > 40))
+    usable_ref_frame = LAST_FRAME;
+
   for (ref_frame = LAST_FRAME; ref_frame <= usable_ref_frame; ++ref_frame) {
     if (!skip_ref_find_pred[ref_frame]) {
       find_predictors(cpi, x, ref_frame, frame_mv, const_motion,

diff --git a/vp9/encoder/vp9_ratectrl.c b/vp9/encoder/vp9_ratectrl.c
index b95cd77736b687c71137818810c6bbc6f1b239e0..fb41427eab62b2544e8f1de869a31bcf01517ef1 100644 (file)
--- a/vp9/encoder/vp9_ratectrl.c
+++ b/vp9/encoder/vp9_ratectrl.c
@@ -2285,6 +2285,7 @@ void vp9_avg_source_sad(VP9_COMP *cpi) {
         int sb_cols = (cm->mi_cols + MI_BLOCK_SIZE - 1) / MI_BLOCK_SIZE;
         int sb_rows = (cm->mi_rows + MI_BLOCK_SIZE - 1) / MI_BLOCK_SIZE;
         uint64_t avg_source_sad_threshold = 10000;
+        uint64_t avg_source_sad_threshold2 = 12000;
         if (cpi->oxcf.lag_in_frames > 0) {
           src_y = frames[frame]->y_buffer;
           src_ystride = frames[frame]->y_stride;
@@ -2315,6 +2316,13 @@ void vp9_avg_source_sad(VP9_COMP *cpi) {
                   cpi->content_state_sb[num_samples] =
                       ((tmp_sse - tmp_variance) < 25) ? kHighSadLowSumdiff
                                                       : kHighSadHighSumdiff;
+                if (tmp_sad < avg_source_sad_threshold2) {
+                  // Cap the increment to 255.
+                  if (cpi->content_state_sb_fd[num_samples] < 255)
+                    cpi->content_state_sb_fd[num_samples]++;
+                } else {
+                  cpi->content_state_sb_fd[num_samples] = 0;
+                }
               }
               avg_sad += tmp_sad;
               num_samples++;

diff --git a/vp9/encoder/vp9_speed_features.c b/vp9/encoder/vp9_speed_features.c
index 2ce836f2eba35e6f4ced5fa4701760f95d4c4944..fd589f9909a3a6253b929206861923a4957535a6 100644 (file)
--- a/vp9/encoder/vp9_speed_features.c
+++ b/vp9/encoder/vp9_speed_features.c
@@ -518,6 +518,8 @@ static void set_rt_speed_feature_framesize_independent(
            cpi->svc.spatial_layer_id == cpi->svc.number_spatial_layers - 1)) {
         cpi->content_state_sb = (uint8_t *)vpx_calloc(
             (cm->mi_stride >> 3) * ((cm->mi_rows >> 3) + 1), sizeof(uint8_t));
+        cpi->content_state_sb_fd = (uint8_t *)vpx_calloc(
+            (cm->mi_stride >> 3) * ((cm->mi_rows >> 3) + 1), sizeof(uint8_t));
       }
     }
   }