Residual coding to cache energy class of tokens.

Proposal for tuning the residual coding by changing how the context
from previous tokens is calculated. Storing the energy class of previous
tokens instead of the token itself eases the critical path of
HW implementations.

Change-Id: I6d71d856b84518f6c88de771ddd818436f794bab

diff --git a/vp9/common/vp9_entropy.c b/vp9/common/vp9_entropy.c
index 847fee6f34b4d92935f235148275d203d0bfa907..1ae35864c736425e8b3f305c98270deac152bd48 100644 (file)
--- a/vp9/common/vp9_entropy.c
+++ b/vp9/common/vp9_entropy.c
@@ -468,7 +468,7 @@ int vp9_get_coef_context(const int *scan, const int *neighbors,
     } else {
       ctx = token_cache[scan[neighbors[MAX_NEIGHBORS * c + 0]]];
     }
-    return vp9_pt_energy_class[ctx];
+    return ctx;
   }
 };
 

diff --git a/vp9/decoder/vp9_detokenize.c b/vp9/decoder/vp9_detokenize.c
index 02900c08c66f7fb312f6fac2aa67bc25bc300c04..890d5d0d6f27698c308dfe2ddfc7569035ceb422 100644 (file)
--- a/vp9/decoder/vp9_detokenize.c
+++ b/vp9/decoder/vp9_detokenize.c
@@ -58,13 +58,15 @@ static const vp9_prob cat6_prob[15] = {
   254, 254, 254, 252, 249, 243, 230, 196, 177, 153, 140, 133, 130, 129, 0
 };
 
+DECLARE_ALIGNED(16, extern const uint8_t,
+                vp9_pt_energy_class[MAX_ENTROPY_TOKENS]);
 #define INCREMENT_COUNT(token)               \
   do {                                       \
     coef_counts[type][ref][band][pt]         \
                [token >= TWO_TOKEN ?     \
                 (token == DCT_EOB_TOKEN ? DCT_EOB_MODEL_TOKEN : TWO_TOKEN) : \
                 token]++;     \
-    token_cache[scan[c]] = token; \
+    token_cache[scan[c]] = vp9_pt_energy_class[token]; \
   } while (0)
 
 #define WRITE_COEF_CONTINUE(val, token)                  \

diff --git a/vp9/encoder/vp9_encodemb.c b/vp9/encoder/vp9_encodemb.c
index b7f60b1272992b87b0a09386bd94f606d7e94c0a..62ef9a04f4aad10138db3e1c39b872aebf387e1a 100644 (file)
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -20,6 +20,9 @@
 #include "vp9/common/vp9_systemdependent.h"
 #include "vp9_rtcd.h"
 
+DECLARE_ALIGNED(16, extern const uint8_t,
+                vp9_pt_energy_class[MAX_ENTROPY_TOKENS]);
+
 void vp9_subtract_block(int rows, int cols,
                         int16_t *diff_ptr, int diff_stride,
                         const uint8_t *src_ptr, int src_stride,
@@ -105,7 +108,7 @@ static int trellis_get_coeff_context(const int *scan,
                                      uint8_t *token_cache,
                                      int pad, int l) {
   int bak = token_cache[scan[idx]], pt;
-  token_cache[scan[idx]] = token;
+  token_cache[scan[idx]] = vp9_pt_energy_class[token];
   pt = vp9_get_coef_context(scan, nb, pad, token_cache, idx + 1, l);
   token_cache[scan[idx]] = bak;
   return pt;
@@ -189,7 +192,8 @@ static void optimize_b(VP9_COMMON *const cm, MACROBLOCK *mb,
   *(tokens[eob] + 1) = *(tokens[eob] + 0);
   next = eob;
   for (i = 0; i < eob; i++)
-    token_cache[scan[i]] = vp9_dct_value_tokens_ptr[qcoeff_ptr[scan[i]]].token;
+    token_cache[scan[i]] = vp9_pt_energy_class[vp9_dct_value_tokens_ptr[
+        qcoeff_ptr[scan[i]]].token];
   nb = vp9_get_coef_neighbors_handle(scan, &pad);
 
   for (i = eob; i-- > i0;) {

diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 2eb3f9b29070669f50a0ccb4b9e51059f9daf2b1..aaba2bb11f2bfaabc85550d9f7686109ce998596 100644 (file)
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -46,6 +46,9 @@
 /* Factor to weigh the rate for switchable interp filters */
 #define SWITCHABLE_INTERP_RATE_FACTOR 1
 
+DECLARE_ALIGNED(16, extern const uint8_t,
+                vp9_pt_energy_class[MAX_ENTROPY_TOKENS]);
+
 const MODE_DEFINITION vp9_mode_order[MAX_MODES] = {
   {ZEROMV,    LAST_FRAME,   NONE},
   {DC_PRED,   INTRA_FRAME,  NONE},
@@ -366,7 +369,7 @@ static INLINE int cost_coeffs(VP9_COMMON *const cm, MACROBLOCK *mb,
 
       if (!c || token_cache[scan[c - 1]])
         cost += vp9_cost_bit(coef_probs[band][pt][0], 1);
-      token_cache[scan[c]] = t;
+      token_cache[scan[c]] = vp9_pt_energy_class[t];
     }
     if (c < seg_eob) {
       if (c)

diff --git a/vp9/encoder/vp9_tokenize.c b/vp9/encoder/vp9_tokenize.c
index 08efc84d43b0c2a9b3397a07000bebfae5bd4c1d..eb79de1d96eef3b07862c240a54b666c2bcf52d9 100644 (file)
--- a/vp9/encoder/vp9_tokenize.c
+++ b/vp9/encoder/vp9_tokenize.c
@@ -36,6 +36,9 @@ extern vp9_coeff_stats tree_update_hist_16x16[BLOCK_TYPES];
 extern vp9_coeff_stats tree_update_hist_32x32[BLOCK_TYPES];
 #endif  /* ENTROPY_STATS */
 
+DECLARE_ALIGNED(16, extern const uint8_t,
+                vp9_pt_energy_class[MAX_ENTROPY_TOKENS]);
+
 static TOKENVALUE dct_value_tokens[DCT_MAX_VALUE * 2];
 const TOKENVALUE *vp9_dct_value_tokens_ptr;
 static int dct_value_cost[DCT_MAX_VALUE * 2];
@@ -228,7 +231,7 @@ static void tokenize_b(int plane, int block, BLOCK_SIZE_TYPE bsize,
       if (!t->skip_eob_node)
         ++cpi->common.fc.eob_branch_counts[tx_size][type][ref][band][pt];
     }
-    token_cache[scan[c]] = token;
+    token_cache[scan[c]] = vp9_pt_energy_class[token];
     ++t;
   } while (c < eob && ++c < seg_eob);