Improve the performance by caching the left_mi and right_mi in macroblockd.

This improve the deocde performance by ~2% on Nexus 7 2013.

Change-Id: Ie9c4ba0371a149eb7fddc687a6a291c17298d6c3

6 files changed, 58 insertions, 52 deletions

diff --git a/vp9/common/vp9_blockd.h b/vp9/common/vp9_blockd.h
index 893a2bb63..7d7209c56 100644
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -192,6 +192,10 @@ typedef struct macroblockd {
   int mi_stride;
 
   MODE_INFO *mi;
+  MODE_INFO *left_mi;
+  MODE_INFO *above_mi;
+  MB_MODE_INFO *left_mbmi;
+  MB_MODE_INFO *above_mbmi;
 
   int up_available;
   int left_available;
diff --git a/vp9/common/vp9_onyxc_int.h b/vp9/common/vp9_onyxc_int.h
index ae69c0c00..c5af6eaf9 100644
--- a/vp9/common/vp9_onyxc_int.h
+++ b/vp9/common/vp9_onyxc_int.h
@@ -309,6 +309,21 @@ static INLINE void set_mi_row_col(MACROBLOCKD *xd, const TileInfo *const tile,
   // Are edges available for intra prediction?
   xd->up_available    = (mi_row != 0);
   xd->left_available  = (mi_col > tile->mi_col_start);
+  if (xd->up_available) {
+    xd->above_mi = xd->mi[-xd->mi_stride].src_mi;
+    xd->above_mbmi = &xd->above_mi->mbmi;;
+  } else {
+    xd->above_mi = NULL;
+    xd->above_mbmi = NULL;
+  }
+
+  if (xd->left_available) {
+    xd->left_mi = xd->mi[-1].src_mi;
+    xd->left_mbmi = &xd->left_mi->mbmi;;
+  } else {
+    xd->left_mi = NULL;
+    xd->left_mbmi = NULL;
+  }
 }
 
 static INLINE void set_prev_mi(VP9_COMMON *cm) {
diff --git a/vp9/common/vp9_pred_common.c b/vp9/common/vp9_pred_common.c
index 901a043f6..fd735f483 100644
--- a/vp9/common/vp9_pred_common.c
+++ b/vp9/common/vp9_pred_common.c
@@ -15,21 +15,17 @@
 #include "vp9/common/vp9_pred_common.h"
 #include "vp9/common/vp9_seg_common.h"
 
-static INLINE const MB_MODE_INFO *get_mbmi(const MODE_INFO *const mi) {
-  return (mi != NULL) ? &mi->mbmi : NULL;
-}
-
 // Returns a context number for the given MB prediction signal
 int vp9_get_pred_context_switchable_interp(const MACROBLOCKD *xd) {
   // Note:
   // The mode info data structure has a one element border above and to the
   // left of the entries correpsonding to real macroblocks.
   // The prediction flags in these dummy entries are initialised to 0.
-  const MB_MODE_INFO *const left_mbmi = get_mbmi(get_left_mi(xd));
-  const int left_type = left_mbmi != NULL && is_inter_block(left_mbmi) ?
-                           left_mbmi->interp_filter : SWITCHABLE_FILTERS;
-  const MB_MODE_INFO *const above_mbmi = get_mbmi(get_above_mi(xd));
-  const int above_type = above_mbmi != NULL && is_inter_block(above_mbmi) ?
+  const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
+  const int left_type = xd->left_available && is_inter_block(left_mbmi) ?
+                            left_mbmi->interp_filter : SWITCHABLE_FILTERS;
+  const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
+  const int above_type = xd->up_available && is_inter_block(above_mbmi) ?
                              above_mbmi->interp_filter : SWITCHABLE_FILTERS;
 
   if (left_type == above_type)
@@ -50,10 +46,10 @@ int vp9_get_pred_context_switchable_interp(const MACROBLOCKD *xd) {
 // 2 - intra/--, --/intra
 // 3 - intra/intra
 int vp9_get_intra_inter_context(const MACROBLOCKD *xd) {
-  const MB_MODE_INFO *const above_mbmi = get_mbmi(get_above_mi(xd));
-  const MB_MODE_INFO *const left_mbmi = get_mbmi(get_left_mi(xd));
-  const int has_above = above_mbmi != NULL;
-  const int has_left = left_mbmi != NULL;
+  const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
+  const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
+  const int has_above = xd->up_available;
+  const int has_left = xd->left_available;
 
   if (has_above && has_left) {  // both edges available
     const int above_intra = !is_inter_block(above_mbmi);
@@ -70,10 +66,10 @@ int vp9_get_intra_inter_context(const MACROBLOCKD *xd) {
 int vp9_get_reference_mode_context(const VP9_COMMON *cm,
                                    const MACROBLOCKD *xd) {
   int ctx;
-  const MB_MODE_INFO *const above_mbmi = get_mbmi(get_above_mi(xd));
-  const MB_MODE_INFO *const left_mbmi = get_mbmi(get_left_mi(xd));
-  const int has_above = above_mbmi != NULL;
-  const int has_left = left_mbmi != NULL;
+  const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
+  const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
+  const int has_above = xd->up_available;
+  const int has_left = xd->left_available;
   // Note:
   // The mode info data structure has a one element border above and to the
   // left of the entries correpsonding to real macroblocks.
@@ -113,10 +109,10 @@ int vp9_get_reference_mode_context(const VP9_COMMON *cm,
 int vp9_get_pred_context_comp_ref_p(const VP9_COMMON *cm,
                                     const MACROBLOCKD *xd) {
   int pred_context;
-  const MB_MODE_INFO *const above_mbmi = get_mbmi(get_above_mi(xd));
-  const MB_MODE_INFO *const left_mbmi = get_mbmi(get_left_mi(xd));
-  const int above_in_image = above_mbmi != NULL;
-  const int left_in_image = left_mbmi != NULL;
+  const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
+  const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
+  const int above_in_image = xd->up_available;
+  const int left_in_image = xd->left_available;
 
   // Note:
   // The mode info data structure has a one element border above and to the
@@ -194,10 +190,10 @@ int vp9_get_pred_context_comp_ref_p(const VP9_COMMON *cm,
 
 int vp9_get_pred_context_single_ref_p1(const MACROBLOCKD *xd) {
   int pred_context;
-  const MB_MODE_INFO *const above_mbmi = get_mbmi(get_above_mi(xd));
-  const MB_MODE_INFO *const left_mbmi = get_mbmi(get_left_mi(xd));
-  const int has_above = above_mbmi != NULL;
-  const int has_left = left_mbmi != NULL;
+  const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
+  const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
+  const int has_above = xd->up_available;
+  const int has_left = xd->left_available;
   // Note:
   // The mode info data structure has a one element border above and to the
   // left of the entries correpsonding to real macroblocks.
@@ -260,10 +256,10 @@ int vp9_get_pred_context_single_ref_p1(const MACROBLOCKD *xd) {
 
 int vp9_get_pred_context_single_ref_p2(const MACROBLOCKD *xd) {
   int pred_context;
-  const MB_MODE_INFO *const above_mbmi = get_mbmi(get_above_mi(xd));
-  const MB_MODE_INFO *const left_mbmi = get_mbmi(get_left_mi(xd));
-  const int has_above = above_mbmi != NULL;
-  const int has_left = left_mbmi != NULL;
+  const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
+  const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
+  const int has_above = xd->up_available;
+  const int has_left = xd->left_available;
 
   // Note:
   // The mode info data structure has a one element border above and to the
@@ -349,10 +345,10 @@ int vp9_get_pred_context_single_ref_p2(const MACROBLOCKD *xd) {
 // The prediction flags in these dummy entries are initialized to 0.
 int vp9_get_tx_size_context(const MACROBLOCKD *xd) {
   const int max_tx_size = max_txsize_lookup[xd->mi[0].src_mi->mbmi.sb_type];
-  const MB_MODE_INFO *const above_mbmi = get_mbmi(get_above_mi(xd));
-  const MB_MODE_INFO *const left_mbmi = get_mbmi(get_left_mi(xd));
-  const int has_above = above_mbmi != NULL;
-  const int has_left = left_mbmi != NULL;
+  const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
+  const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
+  const int has_above = xd->up_available;
+  const int has_left = xd->left_available;
   int above_ctx = (has_above && !above_mbmi->skip) ? (int)above_mbmi->tx_size
                                                    : max_tx_size;
   int left_ctx = (has_left && !left_mbmi->skip) ? (int)left_mbmi->tx_size
diff --git a/vp9/common/vp9_pred_common.h b/vp9/common/vp9_pred_common.h
index cf13e4a91..bc19d28b9 100644
--- a/vp9/common/vp9_pred_common.h
+++ b/vp9/common/vp9_pred_common.h
@@ -18,20 +18,12 @@
 extern "C" {
 #endif
 
-static INLINE const MODE_INFO *get_above_mi(const MACROBLOCKD *const xd) {
-  return xd->up_available ? xd->mi[-xd->mi_stride].src_mi : NULL;
-}
-
-static INLINE const MODE_INFO *get_left_mi(const MACROBLOCKD *const xd) {
-  return xd->left_available ? xd->mi[-1].src_mi : NULL;
-}
-
 int vp9_get_segment_id(const VP9_COMMON *cm, const uint8_t *segment_ids,
                        BLOCK_SIZE bsize, int mi_row, int mi_col);
 
 static INLINE int vp9_get_pred_context_seg_id(const MACROBLOCKD *xd) {
-  const MODE_INFO *const above_mi = get_above_mi(xd);
-  const MODE_INFO *const left_mi = get_left_mi(xd);
+  const MODE_INFO *const above_mi = xd->above_mi;
+  const MODE_INFO *const left_mi = xd->left_mi;
   const int above_sip = (above_mi != NULL) ?
                         above_mi->mbmi.seg_id_predicted : 0;
   const int left_sip = (left_mi != NULL) ? left_mi->mbmi.seg_id_predicted : 0;
@@ -45,8 +37,8 @@ static INLINE vp9_prob vp9_get_pred_prob_seg_id(const struct segmentation *seg,
 }
 
 static INLINE int vp9_get_skip_context(const MACROBLOCKD *xd) {
-  const MODE_INFO *const above_mi = get_above_mi(xd);
-  const MODE_INFO *const left_mi = get_left_mi(xd);
+  const MODE_INFO *const above_mi = xd->above_mi;
+  const MODE_INFO *const left_mi = xd->left_mi;
   const int above_skip = (above_mi != NULL) ? above_mi->mbmi.skip : 0;
   const int left_skip = (left_mi != NULL) ? left_mi->mbmi.skip : 0;
   return above_skip + left_skip;
diff --git a/vp9/encoder/vp9_bitstream.c b/vp9/encoder/vp9_bitstream.c
index 4d88fb5a5..20368f096 100644
--- a/vp9/encoder/vp9_bitstream.c
+++ b/vp9/encoder/vp9_bitstream.c
@@ -346,9 +346,8 @@ static void write_mb_modes_kf(const VP9_COMMON *cm, const MACROBLOCKD *xd,
                               MODE_INFO *mi_8x8, vp9_writer *w) {
   const struct segmentation *const seg = &cm->seg;
   const MODE_INFO *const mi = mi_8x8;
-  const MODE_INFO *const above_mi = mi_8x8[-xd->mi_stride].src_mi;
-  const MODE_INFO *const left_mi =
-      xd->left_available ? mi_8x8[-1].src_mi : NULL;
+  const MODE_INFO *const above_mi = xd->above_mi;
+  const MODE_INFO *const left_mi = xd->left_mi;
   const MB_MODE_INFO *const mbmi = &mi->mbmi;
   const BLOCK_SIZE bsize = mbmi->sb_type;
 
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index c1bdff77a..600a3eb1a 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -982,8 +982,8 @@ static int64_t rd_pick_intra_sub_8x8_y_mode(VP9_COMP *cpi, MACROBLOCK *mb,
   int i, j;
   const MACROBLOCKD *const xd = &mb->e_mbd;
   MODE_INFO *const mic = xd->mi[0].src_mi;
-  const MODE_INFO *above_mi = xd->mi[-xd->mi_stride].src_mi;
-  const MODE_INFO *left_mi = xd->left_available ? xd->mi[-1].src_mi : NULL;
+  const MODE_INFO *above_mi = xd->above_mi;
+  const MODE_INFO *left_mi = xd->left_mi;
   const BLOCK_SIZE bsize = xd->mi[0].src_mi->mbmi.sb_type;
   const int num_4x4_blocks_wide = num_4x4_blocks_wide_lookup[bsize];
   const int num_4x4_blocks_high = num_4x4_blocks_high_lookup[bsize];
@@ -1058,8 +1058,8 @@ static int64_t rd_pick_intra_sby_mode(VP9_COMP *cpi, MACROBLOCK *x,
   TX_SIZE best_tx = TX_4X4;
   int i;
   int *bmode_costs;
-  const MODE_INFO *above_mi = xd->mi[-xd->mi_stride].src_mi;
-  const MODE_INFO *left_mi = xd->left_available ? xd->mi[-1].src_mi : NULL;
+  const MODE_INFO *above_mi = xd->above_mi;
+  const MODE_INFO *left_mi = xd->left_mi;
   const PREDICTION_MODE A = vp9_above_block_mode(mic, above_mi, 0);
   const PREDICTION_MODE L = vp9_left_block_mode(mic, left_mi, 0);
   bmode_costs = cpi->y_mode_costs[A][L];