vp9: Encoding cycle reduction for speed 8.

1. Skip golden non-zeromv and newmv-last for bsize >= 16x16 if the
temporal variance obtained from choose_partitioning is very low.
2. Skip horz and vert INTRA mode for speed 8.

This change works best on the clips with little noise and with some
motion (e.g. gips_motion which has > 5% speed up). PSNR drop is 1.78%
on rtc test set, no obvious visual quality regression found.

Change-Id: Ib43b5b20e67809d03c5a6890818ddff59e1fc94a

1 files changed, 32 insertions, 2 deletions

diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index 554409b74..7e3bc1b66 100644
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -40,6 +40,14 @@ typedef struct {
   int in_use;
 } PRED_BUFFER;
 
+
+static const int pos_shift_16x16[4][4] = {
+  {9, 10, 13, 14},
+  {11, 12, 15, 16},
+  {17, 18, 21, 22},
+  {19, 20, 23, 24}
+};
+
 static int mv_refs_rt(VP9_COMP *cpi, const VP9_COMMON *cm,
                       const MACROBLOCK *x,
                       const MACROBLOCKD *xd,
@@ -1274,6 +1282,8 @@ static INLINE int set_force_skip_low_temp_var(uint8_t *variance_low,
                                               int mi_row, int mi_col,
                                               BLOCK_SIZE bsize) {
   int force_skip_low_temp_var = 0;
+  int i = (mi_row & 0x7) >> 1;
+  int j = (mi_col & 0x7) >> 1;
   // Set force_skip_low_temp_var based on the block size and block offset.
   if (bsize == BLOCK_64X64) {
     force_skip_low_temp_var = variance_low[0];
@@ -1299,6 +1309,19 @@ static INLINE int set_force_skip_low_temp_var(uint8_t *variance_low,
     } else if ((mi_col & 0x7) && (mi_row & 0x7)) {
       force_skip_low_temp_var = variance_low[8];
     }
+  } else if (bsize == BLOCK_16X16) {
+    force_skip_low_temp_var = variance_low[pos_shift_16x16[i][j]];
+  } else if (bsize == BLOCK_32X16) {
+    // The col shift index for the second 16x16 block.
+    int j2 = ((mi_col + 2) & 0x7) >> 1;
+    // Only if each 16x16 block inside has low temporal variance.
+    force_skip_low_temp_var = variance_low[pos_shift_16x16[i][j]] &&
+      variance_low[pos_shift_16x16[i][j2]];
+  } else if (bsize == BLOCK_16X32) {
+    // The row shift index for the second 16x16 block.
+    int i2 = ((mi_row + 2) & 0x7) >> 1;
+    force_skip_low_temp_var = variance_low[pos_shift_16x16[i][j]] &&
+      variance_low[pos_shift_16x16[i2][j]];
   }
   return force_skip_low_temp_var;
 }
@@ -1503,6 +1526,12 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
       continue;
     }
 
+    if (cpi->sf.short_circuit_low_temp_var == 2 &&
+        force_skip_low_temp_var && ref_frame == LAST_FRAME &&
+        this_mode == NEWMV) {
+      continue;
+    }
+
     if (cpi->use_svc) {
       if (svc_force_zero_mode[ref_frame - 1] &&
           frame_mv[this_mode][ref_frame].as_int != 0)
@@ -1842,8 +1871,9 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
     inter_mode_thresh = (inter_mode_thresh << 1) + inter_mode_thresh;
   }
   // Perform intra prediction search, if the best SAD is above a certain
-  // threshold. Skip intra prediction if force_skip_low_temp_var is set.
-  if (!force_skip_low_temp_var && perform_intra_pred &&
+  // threshold.
+  if ((!force_skip_low_temp_var || bsize < BLOCK_32X32) &&
+      perform_intra_pred &&
       (best_rdc.rdcost == INT64_MAX ||
        (!x->skip && best_rdc.rdcost > inter_mode_thresh &&
         bsize <= cpi->sf.max_intra_bsize))) {