vp9: Change the scheme for modeling rd for bsize 32x32.

For real-time CBR mode, use model_rd_for_sb_y_large instead of
model_rd_for_sb_y for 32x32 block. In the former model, transform
might be skipped more aggressively in some condtions, which speeds
up encoding time with only a little PSNR/SSIM drop on rtc test set.
No obvious visual quality regression.

PSNR effect on different speed settings:
speed 8 rtc:  0.129% overall PSNR drop, 0.137% SSIM drop
speed 7 rtc:  0.135% overall PSNR drop, 0.062% SSIM drop
speed 5 rtc_derf: 0.105% overall PSNR drop, 0.095% SSIM drop

Speed up:
gips_motion_WHD, 1mbps: 3.29% faster on speed 7, 2.56% faster on speed8
gips_stat_WHD, 1mbps: 2.17% faster on speed 7, 1.62% faster on speed8

BUG=webm:1250

Change-Id: I818babce5b8549b4b1a7c3978df8591bffde7173

diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index 6fcc2a7..fa0fea0 100644 (file)
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -1727,9 +1727,9 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
       vp9_build_inter_predictors_sby(xd, mi_row, mi_col, bsize);
 
       // For large partition blocks, extra testing is done.
-      if (cpi->oxcf.rc_mode == VPX_CBR && bsize > BLOCK_32X32 &&
-        !cyclic_refresh_segment_id_boosted(xd->mi[0]->segment_id) &&
-        cm->base_qindex) {
+      if (cpi->oxcf.rc_mode == VPX_CBR && bsize >= BLOCK_32X32 &&
+          !cyclic_refresh_segment_id_boosted(xd->mi[0]->segment_id) &&
+          cm->base_qindex) {
         model_rd_for_sb_y_large(cpi, bsize, x, xd, &this_rdc.rate,
                                 &this_rdc.dist, &var_y, &sse_y, mi_row, mi_col,
                                 &this_early_term);