[FFmpeg-devel] [PATCH] vp9: add avx2 iadst16 implementations.

[James Almer]

On 11/8/2016 1:22 PM, Ronald S. Bultje wrote:
> Also a small cosmetic change to the avx2 idct16 version to make it
> explicit that one of the arguments to the write-out macros is unused
> for >=avx2 (it uses pmovzxbw instead of punpcklbw).

A braindead test (ffmpeg -i 4kHDRsample.webm -benchmark -f null -) on an i5
Haswell went from

frame= 2000 fps= 73 q=-0.0 Lsize=N/A time=00:00:33.36 bitrate=N/A speed=1.21x
bench: utime=92.250s

To

frame= 2000 fps= 77 q=-0.0 Lsize=N/A time=00:00:33.36 bitrate=N/A speed=1.28x
bench: utime=86.891s

In comparison, a 1080p version of the same video now reaches ~360fps.

FATE passes, so LGTM (After the x86_32 fix).