Optimize coef update

1. move the check of search method of USE_TX_8X8 up one level to
avoid operations of build_tree_distributions()
2. count tx used and avoid computaton for coef udpate when one size
is not used at all.

Change-Id: Ia3e54a2588aa531c41377a1bfaa64385d04a592c

3 files changed, 15 insertions, 9 deletions

diff --git a/vp9/common/vp9_entropymode.h b/vp9/common/vp9_entropymode.h
index 6db10806d..972509d7d 100644
--- a/vp9/common/vp9_entropymode.h
+++ b/vp9/common/vp9_entropymode.h
@@ -33,6 +33,7 @@ struct tx_counts {
   unsigned int p32x32[TX_SIZE_CONTEXTS][TX_SIZES];
   unsigned int p16x16[TX_SIZE_CONTEXTS][TX_SIZES - 1];
   unsigned int p8x8[TX_SIZE_CONTEXTS][TX_SIZES - 2];
+  unsigned int tx_totals[TX_SIZES];
 };
 
 typedef struct frame_contexts {
diff --git a/vp9/encoder/vp9_bitstream.c b/vp9/encoder/vp9_bitstream.c
index a72856db4..c7d489db6 100644
--- a/vp9/encoder/vp9_bitstream.c
+++ b/vp9/encoder/vp9_bitstream.c
@@ -611,10 +611,6 @@ static void update_coef_probs_common(vp9_writer* const bc, VP9_COMP *cpi,
     case ONE_LOOP_REDUCED: {
       int updates = 0;
       int noupdates_before_first = 0;
-      if (tx_size >= TX_16X16 && cpi->sf.tx_size_search_method == USE_TX_8X8) {
-        vp9_write_bit(bc, 0);
-        return;
-      }
       for (i = 0; i < PLANE_TYPES; ++i) {
         for (j = 0; j < REF_TYPES; ++j) {
           for (k = 0; k < COEF_BANDS; ++k) {
@@ -678,10 +674,15 @@ static void update_coef_probs(VP9_COMP *cpi, vp9_writer* w) {
   for (tx_size = TX_4X4; tx_size <= max_tx_size; ++tx_size) {
     vp9_coeff_stats frame_branch_ct[PLANE_TYPES];
     vp9_coeff_probs_model frame_coef_probs[PLANE_TYPES];
-    build_tree_distribution(cpi, tx_size, frame_branch_ct,
-                            frame_coef_probs);
-    update_coef_probs_common(w, cpi, tx_size, frame_branch_ct,
-                             frame_coef_probs);
+    if (cpi->td.counts->tx.tx_totals[tx_size] == 0 ||
+        (tx_size >= TX_16X16 && cpi->sf.tx_size_search_method == USE_TX_8X8)) {
+      vp9_write_bit(w, 0);
+    } else {
+      build_tree_distribution(cpi, tx_size, frame_branch_ct,
+                              frame_coef_probs);
+      update_coef_probs_common(w, cpi, tx_size, frame_branch_ct,
+                               frame_coef_probs);
+    }
   }
 }
 
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 730a229ca..18929c74a 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -3805,6 +3805,9 @@ void vp9_encode_frame(VP9_COMP *cpi) {
     }
   }
 
+  vpx_memset(cpi->td.counts->tx.tx_totals, 0,
+             sizeof(cpi->td.counts->tx.tx_totals));
+
   if (cpi->sf.frame_parameter_update) {
     int i;
 
@@ -3891,7 +3894,6 @@ void vp9_encode_frame(VP9_COMP *cpi) {
         count16x16_lp += counts->tx.p32x32[i][TX_16X16];
         count32x32 += counts->tx.p32x32[i][TX_32X32];
       }
-
       if (count4x4 == 0 && count16x16_lp == 0 && count16x16_16x16p == 0 &&
           count32x32 == 0) {
         cm->tx_mode = ALLOW_8X8;
@@ -4015,5 +4017,7 @@ static void encode_superblock(VP9_COMP *cpi, ThreadData *td,
           if (mi_col + x < cm->mi_cols && mi_row + y < cm->mi_rows)
             mi_8x8[mis * y + x].src_mi->mbmi.tx_size = tx_size;
     }
+    ++td->counts->tx.tx_totals[mbmi->tx_size];
+    ++td->counts->tx.tx_totals[get_uv_tx_size(mbmi, &xd->plane[1])];
   }
 }