Move token_cache from cost_coeffs to MACROBLOCK

This commit moves token_cache buffer into macroblock struct, instead
of defining as a local variable in cost_coeffs. This avoids repeatedly
re-allocating memory space in the rate-distortion optimization loop.

The runtime at speed 0 reduces:
bus 2000kbps, 161692ms to 159951ms
football 600kbps, 229505ms to 225821ms

Change-Id: If7da6b0b6d8c5138a16271a33c4548fba33d8840

diff --git a/vp9/encoder/vp9_block.h b/vp9/encoder/vp9_block.h
index a63bd1b8af521ec57c7d86745c678a84c2d00773..9b57bc364ddc530434cfcdb3294fe5ec78bfabcc 100644 (file)
--- a/vp9/encoder/vp9_block.h
+++ b/vp9/encoder/vp9_block.h
@@ -135,6 +135,7 @@ struct macroblock {
 
   // note that token_costs is the cost when eob node is skipped
   vp9_coeff_cost token_costs[TX_SIZES];
+  uint8_t token_cache[1024];
 
   int optimize;
 

diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index a1be626233665516ca3f651e335d90427643fb9f..5eb0ed0a5187e55fdaf79b931a432b2befd5046f 100644 (file)
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -467,12 +467,12 @@ static const int16_t band_counts[TX_SIZES][8] = {
   { 1, 2, 3, 4, 11, 1024 - 21, 0 },
 };
 
-static INLINE int cost_coeffs(MACROBLOCK *mb,
+static INLINE int cost_coeffs(MACROBLOCK *x,
                               int plane, int block,
                               ENTROPY_CONTEXT *A, ENTROPY_CONTEXT *L,
                               TX_SIZE tx_size,
                               const int16_t *scan, const int16_t *nb) {
-  MACROBLOCKD *const xd = &mb->e_mbd;
+  MACROBLOCKD *const xd = &x->e_mbd;
   MB_MODE_INFO *mbmi = &xd->this_mi->mbmi;
   struct macroblockd_plane *pd = &xd->plane[plane];
   const PLANE_TYPE type = pd->plane_type;
@@ -481,9 +481,9 @@ static INLINE int cost_coeffs(MACROBLOCK *mb,
   const int16_t *const qcoeff_ptr = BLOCK_OFFSET(pd->qcoeff, block);
   const int ref = mbmi->ref_frame[0] != INTRA_FRAME;
   unsigned int (*token_costs)[2][PREV_COEF_CONTEXTS][MAX_ENTROPY_TOKENS] =
-                   mb->token_costs[tx_size][type][ref];
+                   x->token_costs[tx_size][type][ref];
   const ENTROPY_CONTEXT above_ec = !!*A, left_ec = !!*L;
-  uint8_t token_cache[1024];
+  uint8_t *p_tok = x->token_cache;
   int pt = combine_entropy_contexts(above_ec, left_ec);
   int c, cost;
 
@@ -502,7 +502,7 @@ static INLINE int cost_coeffs(MACROBLOCK *mb,
     int v = qcoeff_ptr[0];
     int prev_t = vp9_dct_value_tokens_ptr[v].token;
     cost = (*token_costs)[0][pt][prev_t] + vp9_dct_value_cost_ptr[v];
-    token_cache[0] = vp9_pt_energy_class[prev_t];
+    p_tok[0] = vp9_pt_energy_class[prev_t];
     ++token_costs;
 
     // ac tokens
@@ -512,9 +512,9 @@ static INLINE int cost_coeffs(MACROBLOCK *mb,
 
       v = qcoeff_ptr[rc];
       t = vp9_dct_value_tokens_ptr[v].token;
-      pt = get_coef_context(nb, token_cache, c);
+      pt = get_coef_context(nb, p_tok, c);
       cost += (*token_costs)[!prev_t][pt][t] + vp9_dct_value_cost_ptr[v];
-      token_cache[rc] = vp9_pt_energy_class[t];
+      p_tok[rc] = vp9_pt_energy_class[t];
       prev_t = t;
       if (!--band_left) {
         band_left = *band_count++;
@@ -524,7 +524,7 @@ static INLINE int cost_coeffs(MACROBLOCK *mb,
 
     // eob token
     if (band_left) {
-      pt = get_coef_context(nb, token_cache, c);
+      pt = get_coef_context(nb, p_tok, c);
       cost += (*token_costs)[0][pt][DCT_EOB_TOKEN];
     }
   }