Skip to content

NEON port of svt_av1_highbd_convolve_2d_sr_c

Rodrigo Causarano requested to merge rjcausarano/SVT-AV1:rjcausarano/svt_av1_highbd_convolve_2d_sr_c into master

Description

Issue

NEON port of svt_av1_highbd_convolve_2d_sr_c from the SSSE3 implementation.

The system under test is c7g.4xlarge AWS graviton instance.

The measurements below were obtained as follows:

wget wget http://ultravideo.fi/video/Bosphorus_3840x2160_120fps_420_8bit_YUV_Y4M.7z
7z x Bosphorus_3840x2160_120fps_420_8bit_YUV_Y4M.7z

SvtAv1EncApp -i Bosphorus_3840x2160.y4m --crf [C] --preset [P]

Improvement:

filename preset crf e2e 3x encoding time
After Before Improv.
Bosphorus_3840x2160.y4m' 4 30 147.18 146.46 -0.49%
Bosphorus_3840x2160.y4m' 6 30 51.38 51.15 -0.45%
Bosphorus_3840x2160.y4m' 8 30 24.41 24.20 -0.86%
Bosphorus_3840x2160.y4m' 10 30 11.34 11.28 -0.53%
Bosphorus_3840x2160.y4m' 12 30 7.60 7.57 -0.39%
Bosphorus_3840x2160.y4m' 13 30 7.61 7.59 -0.26%
Bosphorus_1920x1080.y4m' 4 30 70.08 70.23 0.21%
Bosphorus_1920x1080.y4m' 6 30 25.89 26.03 0.54%
Bosphorus_1920x1080.y4m' 8 30 13.12 13.08 -0.30%
Bosphorus_1920x1080.y4m' 10 30 5.75 5.71 -0.70%
Bosphorus_1920x1080.y4m' 12 30 2.90 2.89 -0.34%
Bosphorus_1920x1080.y4m' 13 30 2.42 2.39 -1.24%
jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' 4 30 97.89 99.62 1.77%
jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' 6 30 96.43 98.57 2.22%
jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' 8 30 97.49 98.59 1.13%
jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' 10 30 26.12 26.23 0.42%
jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' 12 30 14.07 14.28 1.49%
jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' 13 30 14.96 15.08 0.80%
jellyfish-1080p-hevc-10bit.yuv' 4 30 75.99 78.36 3.12%
jellyfish-1080p-hevc-10bit.yuv' 6 30 76.30 77.98 2.20%
jellyfish-1080p-hevc-10bit.yuv' 8 30 76.69 78.59 2.48%
jellyfish-1080p-hevc-10bit.yuv' 10 30 21.71 22.21 2.30%
jellyfish-1080p-hevc-10bit.yuv' 12 30 14.21 14.40 1.34%
jellyfish-1080p-hevc-10bit.yuv' 13 30 12.63 12.74 0.87%

Author(s)

Rodrigo Causarano ( @rjcausarano) Gerardo Puga ( @glpuga)

Performance impact

  • quality
  • memory
  • speed
  • 8 bit
  • 10 bit
  • N/A

Test set

  • obj-1-fast can be found here
  • other
  • N/A

Merge method

  • Allow the maintainer to squash and merge when PR is ready to create a 1-commit to the master branch. The maintainer will be able to fix typos / combine commit messages to create a more readable 1-commit message or use whatever is stated in the 'Description' section
  • I will clean up my commits and the maintainer shall use 'rebase and merge' to the master branch

Merge request reports