fixed an issue with 8x8 token cost in trellisquant

changed the token cost for 8x8 transformed macroblock used in trellisquant
from those derived from 4x4 transform coefficient distribution to those
derived from 8x8 transform coefficient distribution. Test results show
this fix help 8x8 transform based compression consistently on cif and hd
sets:

http://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_cost8x8only.html
(avg psnr:.14% glb psnr: .17% ssim: .20%)
http://www.corp.google.com/~yaowu/no_crawl/t8x8/hd_cost8x8only.html
(avg psnr:.17% glb psnr: .18% ssim: .58%)

Note: To test the effect of this change, 8x8 transform was forced to be used
only on 16x16 predicted macroblocks on inter frames, the effect would be
bigger had all macroblocks been forcd to use 8x8 transform.

Change-Id: If9b7868b75357c66541f511e5ee78e4d2d4929a4

3 files changed, 23 insertions, 10 deletions

diff --git a/vp8/encoder/block.h b/vp8/encoder/block.h
index f992630a3..68de571c5 100644
--- a/vp8/encoder/block.h
+++ b/vp8/encoder/block.h
@@ -118,7 +118,14 @@ typedef struct
     unsigned char *active_ptr;
     MV_CONTEXT *mvc;
 
-    unsigned int token_costs[BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [MAX_ENTROPY_TOKENS];
+    unsigned int token_costs[BLOCK_TYPES] [COEF_BANDS]
+                            [PREV_COEF_CONTEXTS][MAX_ENTROPY_TOKENS];
+
+#if CONFIG_T8X8
+    unsigned int token_costs_8x8[BLOCK_TYPES] [COEF_BANDS]
+                            [PREV_COEF_CONTEXTS] [MAX_ENTROPY_TOKENS];
+#endif
+
     int optimize;
     int q_index;
 
diff --git a/vp8/encoder/encodemb.c b/vp8/encoder/encodemb.c
index 82ac021fd..7916066eb 100644
--- a/vp8/encoder/encodemb.c
+++ b/vp8/encoder/encodemb.c
@@ -907,9 +907,9 @@ void optimize_b_8x8(MACROBLOCK *mb, int i, int type,
                 band = vp8_coef_bands_8x8[i + 1];
                 pt = vp8_prev_token_class[t0];
                 rate0 +=
-                    mb->token_costs[type][band][pt][tokens[next][0].token];
+                    mb->token_costs_8x8[type][band][pt][tokens[next][0].token];
                 rate1 +=
-                    mb->token_costs[type][band][pt][tokens[next][1].token];
+                    mb->token_costs_8x8[type][band][pt][tokens[next][1].token];
             }
             rd_cost0 = RDCOST_8x8(rdmult, rddiv, rate0, error0);
             rd_cost1 = RDCOST_8x8(rdmult, rddiv, rate1, error1);
@@ -966,13 +966,13 @@ void optimize_b_8x8(MACROBLOCK *mb, int i, int type,
                 if(t0!=DCT_EOB_TOKEN)
                 {
                     pt = vp8_prev_token_class[t0];
-                    rate0 += mb->token_costs[type][band][pt][
+                    rate0 += mb->token_costs_8x8[type][band][pt][
                         tokens[next][0].token];
                 }
                 if(t1!=DCT_EOB_TOKEN)
                 {
                     pt = vp8_prev_token_class[t1];
-                    rate1 += mb->token_costs[type][band][pt][
+                    rate1 += mb->token_costs_8x8[type][band][pt][
                         tokens[next][1].token];
                 }
             }
@@ -1013,12 +1013,12 @@ void optimize_b_8x8(MACROBLOCK *mb, int i, int type,
             /* Update the cost of each path if we're past the EOB token. */
             if (t0 != DCT_EOB_TOKEN)
             {
-                tokens[next][0].rate += mb->token_costs[type][band][0][t0];
+                tokens[next][0].rate += mb->token_costs_8x8[type][band][0][t0];
                 tokens[next][0].token = ZERO_TOKEN;
             }
             if (t1 != DCT_EOB_TOKEN)
             {
-                tokens[next][1].rate += mb->token_costs[type][band][0][t1];
+                tokens[next][1].rate += mb->token_costs_8x8[type][band][0][t1];
                 tokens[next][1].token = ZERO_TOKEN;
             }
             /* Don't update next, because we didn't add a new node. */
@@ -1034,8 +1034,8 @@ void optimize_b_8x8(MACROBLOCK *mb, int i, int type,
     error1 = tokens[next][1].error;
     t0 = tokens[next][0].token;
     t1 = tokens[next][1].token;
-    rate0 += mb->token_costs[type][band][pt][t0];
-    rate1 += mb->token_costs[type][band][pt][t1];
+    rate0 += mb->token_costs_8x8[type][band][pt][t0];
+    rate1 += mb->token_costs_8x8[type][band][pt][t1];
     rd_cost0 = RDCOST_8x8(rdmult, rddiv, rate0, error0);
     rd_cost1 = RDCOST_8x8(rdmult, rddiv, rate1, error1);
     if (rd_cost0 == rd_cost1)
diff --git a/vp8/encoder/rdopt.c b/vp8/encoder/rdopt.c
index 99bad881d..45db403f6 100644
--- a/vp8/encoder/rdopt.c
+++ b/vp8/encoder/rdopt.c
@@ -215,7 +215,6 @@ static void fill_token_costs(
     for (i = 0; i < BLOCK_TYPES; i++)
         for (j = 0; j < COEF_BANDS; j++)
             for (k = 0; k < PREV_COEF_CONTEXTS; k++)
-
                 vp8_cost_tokens((int *)(c [i][j][k]), p [i][j][k], vp8_coef_tree);
 
 }
@@ -363,6 +362,12 @@ void vp8_initialize_rd_consts(VP8_COMP *cpi, int QIndex)
         (const vp8_prob( *)[8][3][11]) cpi->common.fc.coef_probs
     );
 
+#if CONFIG_T8X8
+    fill_token_costs(
+        cpi->mb.token_costs_8x8,
+        (const vp8_prob( *)[8][3][11]) cpi->common.fc.coef_probs_8x8
+    );
+#endif
 #if CONFIG_QIMODE
     //rough estimate for costing
     cpi->common.kf_ymode_probs_index = cpi->common.base_qindex>>4;
@@ -666,6 +671,7 @@ static void macro_block_yrd( MACROBLOCK *mb,
     *Rate = vp8_rdcost_mby(mb);
 }
 
+
 static void copy_predictor(unsigned char *dst, const unsigned char *predictor)
 {
     const unsigned int *p = (const unsigned int *)predictor;