further clean-ups on intra4x4 coding

Removed one 4x4 prediction step that was unnessary in the rd loop.
Removed a unused modecosts estimate from encoder side.

Change-Id: I65221a52719d6876492996955ef04142d2752d86

vp9/common/vp9_reconintra.c

diff --git a/vp9/common/vp9_reconintra.c b/vp9/common/vp9_reconintra.c
index dd60a76..b1f327b 100644
--- a/vp9/common/vp9_reconintra.c
+++ b/vp9/common/vp9_reconintra.c
@@ -402,7 +402,7 @@
       (block_idx >> bwl) || xd->up_available;
   const int have_left =
       (block_idx & wmask) || xd->left_available;
-  const int have_right = ((block_idx & wmask) != wmask);
+  int have_right = ((block_idx & wmask) != wmask);
   const int txfm_block_size = 4 << tx_size;
 
   assert(bwl >= 0);

vp9/encoder/vp9_block.h

diff --git a/vp9/encoder/vp9_block.h b/vp9/encoder/vp9_block.h
index 211eca4..0e9b680 100644
--- a/vp9/encoder/vp9_block.h
+++ b/vp9/encoder/vp9_block.h
@@ -117,7 +117,6 @@
   int mbmode_cost[2][MB_MODE_COUNT];
   int intra_uv_mode_cost[2][MB_MODE_COUNT];
   int bmode_costs[VP9_BINTRAMODES][VP9_BINTRAMODES][VP9_BINTRAMODES];
-  int inter_bmode_costs[INTRA_MODE_COUNT];
   int switchable_interp_costs[VP9_SWITCHABLE_FILTERS + 1]
                              [VP9_SWITCHABLE_FILTERS];
 

vp9/encoder/vp9_encodemb.c

diff --git a/vp9/encoder/vp9_encodemb.c b/vp9/encoder/vp9_encodemb.c
index b7f60b1..87323f9 100644
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -610,6 +610,7 @@
   struct encode_b_args* const args = arg;
   MACROBLOCK* const x = args->x;
   MACROBLOCKD* const xd = &x->e_mbd;
+  MB_MODE_INFO* const mbmi = &xd->mode_info_context->mbmi;
   const TX_SIZE tx_size = (TX_SIZE)(ss_txfrm_size / 2);
   const int bw = 4 << (b_width_log2(bsize) - xd->plane[plane].subsampling_x);
   const int raster_block = txfrm_block_to_raster_block(xd, bsize, plane,
@@ -634,9 +635,9 @@
   TX_TYPE tx_type;
   int mode, b_mode;
 
-  mode = plane == 0? xd->mode_info_context->mbmi.mode:
-                     xd->mode_info_context->mbmi.uv_mode;
-  if (bsize <= BLOCK_SIZE_SB8X8 && mode == I4X4_PRED && plane == 0)
+  mode = plane == 0? mbmi->mode: mbmi->uv_mode;
+  if (mbmi->sb_type < BLOCK_SIZE_SB8X8 && plane == 0 &&
+      mbmi->ref_frame == INTRA_FRAME)
     b_mode = xd->mode_info_context->bmi[ib].as_mode.first;
   else
     b_mode = mode;

vp9/encoder/vp9_modecosts.c

diff --git a/vp9/encoder/vp9_modecosts.c b/vp9/encoder/vp9_modecosts.c
index 171b44b..67658f5 100644
--- a/vp9/encoder/vp9_modecosts.c
+++ b/vp9/encoder/vp9_modecosts.c
@@ -17,7 +17,6 @@
 
 void vp9_init_mode_costs(VP9_COMP *c) {
   VP9_COMMON *x = &c->common;
-  const vp9_tree_p T = vp9_bmode_tree;
   const vp9_tree_p KT = vp9_bmode_tree;
   int i, j;
 
@@ -28,8 +27,6 @@
     }
   }
 
-  vp9_cost_tokens((int *)c->mb.inter_bmode_costs, x->fc.bmode_prob, T);
-
   // TODO(rbultje) separate tables for superblock costing?
   vp9_cost_tokens(c->mb.mbmode_cost[1], x->fc.sb_ymode_prob,
                   vp9_sb_ymode_tree);

vp9/encoder/vp9_rdopt.c

diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 2eb3f9b..f64bd97 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -611,7 +611,6 @@
     int64_t this_rd;
     int ratey = 0;
 
-    xd->mode_info_context->bmi[ib].as_mode.first = mode;
     if (cm->frame_type == KEY_FRAME)
       rate = bmode_costs[mode];
     else
@@ -653,9 +652,6 @@
         distortion += vp9_block_error(coeff, BLOCK_OFFSET(xd->plane[0].dqcoeff,
                                                          block, 16), 16) >> 2;
 
-        vp9_intra4x4_predict(xd, block, BLOCK_SIZE_SB8X8, mode,
-                             dst, xd->plane[0].dst.stride);
-
         if (best_tx_type != DCT_DCT)
           vp9_short_iht4x4_add(BLOCK_OFFSET(xd->plane[0].dqcoeff, block, 16),
                                dst, xd->plane[0].dst.stride, best_tx_type);
@@ -731,7 +727,7 @@
   vpx_memcpy(t_left, xd->plane[0].left_context, sizeof(t_left));
 
   xd->mode_info_context->mbmi.mode = I4X4_PRED;
-  bmode_costs = mb->inter_bmode_costs;
+  bmode_costs = mb->mbmode_cost[cpi->common.frame_type];
 
   for (idy = 0; idy < 2; idy += bh) {
     for (idx = 0; idx < 2; idx += bw) {
@@ -2684,8 +2680,6 @@
     if (this_mode == I4X4_PRED) {
       int rate;
 
-      // Note the rate value returned here includes the cost of coding
-      // the I4X4_PRED mode : x->mbmode_cost[xd->frame_type][I4X4_PRED];
       mbmi->txfm_size = TX_4X4;
       rd_pick_intra4x4mby_modes(cpi, x, &rate, &rate_y,
                                 &distortion_y, INT64_MAX);