Speed up non-rd mode decision search

This commit makes the encoder to explicitly calculate the SAD
associated with the LAST_FRAME motion vector and compare it to
that of the GOLDEN_FRAME given by integral projection motion
estimation. It skips the expensive sub-pixel motion search over
GOLDEN_FRAME when the LAST_FRAME can provide fairly good motion
compensated prediction quality.

For dark720p speed -6 single thread goes from
33304 b/f, 40.070 dB, 18156 ms ->
33319 b/f, 40.061 dB, 17611 ms

Change-Id: I01bc94b9b598075567a392111046b97a9bc30efe

diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index a96a97453e0249b6669c6ecbbbe33c095462b547..6dfb9eedde14e15b3bff64b5a32625fdfcb67156 100644 (file)
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -681,6 +681,7 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
   int reuse_inter_pred = cpi->sf.reuse_inter_pred_sby && ctx->pred_pixel_ready;
   int ref_frame_skip_mask = 0;
   int idx;
+  int best_pred_sad = INT_MAX;
 
   if (reuse_inter_pred) {
     int i;
@@ -811,8 +812,11 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
           continue;
 
         tmp_sad = vp9_int_pro_motion_estimation(cpi, x, bsize);
+
         if (tmp_sad > x->pred_mv_sad[LAST_FRAME])
           continue;
+        if (tmp_sad + (num_pels_log2_lookup[bsize] << 4) > best_pred_sad)
+          continue;
 
         frame_mv[NEWMV][ref_frame].as_int = mbmi->mv[0].as_int;
         rate_mv = vp9_mv_bit_cost(&frame_mv[NEWMV][ref_frame].as_mv,
@@ -837,6 +841,17 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
       }
     }
 
+    if (this_mode == NEWMV && ref_frame == LAST_FRAME &&
+        frame_mv[NEWMV][LAST_FRAME].as_int != INVALID_MV) {
+      const int pre_stride = xd->plane[0].pre[0].stride;
+      const uint8_t * const pre_buf = xd->plane[0].pre[0].buf +
+          (frame_mv[NEWMV][LAST_FRAME].as_mv.row >> 3) * pre_stride +
+          (frame_mv[NEWMV][LAST_FRAME].as_mv.col >> 3);
+      best_pred_sad = cpi->fn_ptr[bsize].sdf(x->plane[0].src.buf,
+                                   x->plane[0].src.stride,
+                                   pre_buf, pre_stride);
+    }
+
     if (this_mode != NEARESTMV &&
         frame_mv[this_mode][ref_frame].as_int ==
             frame_mv[NEARESTMV][ref_frame].as_int)