shithub: libvpx

Info  •  Files  •  Log  •  Branches

Calculate transform size cost once per frame

Instead of doing it in every transform search loop.

Change-Id: I12dc402a6633d1a27d32cb6b58710b8c0ebf0fd4

--- a/vp9/encoder/vp9_encoder.h

+++ b/vp9/encoder/vp9_encoder.h

@@ -645,6 +645,8 @@

   int y_mode_costs[INTRA_MODES][INTRA_MODES][INTRA_MODES];

   int switchable_interp_costs[SWITCHABLE_FILTER_CONTEXTS][SWITCHABLE_FILTERS];

   int partition_cost[PARTITION_CONTEXTS][PARTITION_TYPES];

+  // Indices are:  max_tx_size-1,  tx_size_ctx,    tx_size

+  int tx_size_cost[TX_SIZES - 1][TX_SIZE_CONTEXTS][TX_SIZES];

   int multi_arf_allowed;

   int multi_arf_enabled;

--- a/vp9/encoder/vp9_rd.c

+++ b/vp9/encoder/vp9_rd.c

@@ -69,10 +69,12 @@

   const FRAME_CONTEXT *const fc = cpi->common.fc;

   int i, j;

-  for (i = 0; i < INTRA_MODES; ++i)

-    for (j = 0; j < INTRA_MODES; ++j)

+  for (i = 0; i < INTRA_MODES; ++i) {

+    for (j = 0; j < INTRA_MODES; ++j) {

       vp9_cost_tokens(cpi->y_mode_costs[i][j], vp9_kf_y_mode_prob[i][j],

                       vp9_intra_mode_tree);

+    }

+  }

   vp9_cost_tokens(cpi->mbmode_cost, fc->y_mode_prob[1], vp9_intra_mode_tree);

   for (i = 0; i < INTRA_MODES; ++i) {

@@ -82,9 +84,28 @@

                     fc->uv_mode_prob[i], vp9_intra_mode_tree);

   }

-  for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; ++i)

+  for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; ++i) {

     vp9_cost_tokens(cpi->switchable_interp_costs[i],

                     fc->switchable_interp_prob[i], vp9_switchable_interp_tree);

+  }

+

+  for (i = TX_8X8; i < TX_SIZES; ++i) {

+    for (j = 0; j < TX_SIZE_CONTEXTS; ++j) {

+      const vpx_prob *tx_probs = get_tx_probs(i, j, &fc->tx_probs);

+      int k;

+      for (k = 0; k <= i; ++k) {

+        int cost = 0;

+        int m;

+        for (m = 0; m <= k - (k == i); ++m) {

+          if (m == k)

+            cost += vp9_cost_zero(tx_probs[m]);

+          else

+            cost += vp9_cost_one(tx_probs[m]);

+        }

+        cpi->tx_size_cost[i - 1][j][k] = cost;

+      }

+    }

+  }

 }

 static void fill_token_costs(vp9_coeff_cost *c,

--- a/vp9/encoder/vp9_rdopt.c

+++ b/vp9/encoder/vp9_rdopt.c

@@ -845,14 +845,12 @@

                               { INT64_MAX, INT64_MAX },

                               { INT64_MAX, INT64_MAX },

                               { INT64_MAX, INT64_MAX } };

-  int n, m;

+  int n;

   int s0, s1;

   int64_t best_rd = INT64_MAX;

   TX_SIZE best_tx = max_tx_size;

   int start_tx, end_tx;

-

-  const vpx_prob *tx_probs =

-      get_tx_probs(max_tx_size, get_tx_size_context(xd), &cm->fc->tx_probs);

+  const int tx_size_ctx = get_tx_size_context(xd);

   assert(skip_prob > 0);

   s0 = vp9_cost_bit(skip_prob, 0);

   s1 = vp9_cost_bit(skip_prob, 1);

@@ -868,13 +866,7 @@

   }

   for (n = start_tx; n >= end_tx; n--) {

-    int r_tx_size = 0;

-    for (m = 0; m <= n - (n == (int)max_tx_size); m++) {

-      if (m == n)

-        r_tx_size += vp9_cost_zero(tx_probs[m]);

-      else

-        r_tx_size += vp9_cost_one(tx_probs[m]);

-    }

+    const int r_tx_size = cpi->tx_size_cost[max_tx_size - 1][tx_size_ctx][n];

     txfm_rd_in_plane(cpi, x, &r[n][0], &d[n], &s[n], &sse[n], ref_best_rd, 0,

                      bs, n, cpi->sf.use_fast_coef_costing);

     r[n][1] = r[n][0];