tree 83501666e1ede6a108c0f6103468495a69456ac7
parent d58711f4fcdea41793bc64761e452d6ee8c42b6b
author Ranjit Kumar Tulabandu <ranjit.tulabandu@ittiam.com> 1774355254 +0530
committer Yunqing Wang <yunqingwang@google.com> 1774388482 -0700

Prune the evaluation of inter transform split

In the CL, top N estimated RD Costs of transform no-split are tracked
for inter modes and transform split RD evaluation of a inter mode is
skipped if the relevant no-split RD Cost is worser than top Nth
no-split RD Cost.

The encoder performance results averaged over all resolutions are
as below:

    Instruction Count                 BD-Rate Loss(%)
cpu   Reduction(%)     avg.psnr   ovr.psnr   ssim     vmaf   vmaf_neg
 1       1.647          0.0253     0.0305   0.0261   0.0594   0.0524
 2       1.144          0.0121     0.0081  -0.0006   0.0096   0.0342

The CL is bit-exact for speed>=3

STATS_CHANGED

Change-Id: I7f477093a84b27d19ac2d6be41e4aeeab26a8fb7