]> git.sesse.net Git - ffmpeg/commit
avfilter/vf_gblur: add x86 SIMD optimizations
authorRuiling Song <ruiling.song@intel.com>
Wed, 15 May 2019 09:54:10 +0000 (17:54 +0800)
committerRuiling Song <ruiling.song@intel.com>
Wed, 12 Jun 2019 00:53:11 +0000 (08:53 +0800)
commit83f9da77684e7ea0d8e9f9712ec716424140043a
tree36264e4571080e5cdd0d9c9eb649d6c31a9c1cb7
parent5fc8d87ba6954d3917a9095fb3b8c1d8caf0b0f4
avfilter/vf_gblur: add x86 SIMD optimizations

The horizontal pass get ~2x performance with the patch
under single thread.

Tested overall performance using the command(avx2 enabled):
./ffmpeg -i 1080p.mp4 -vf gblur -f null /dev/null
./ffmpeg -i 1080p.mp4 -vf gblur=threads=1 -f null /dev/null
For single thread, the fps improves from 43 to 60, about 40%.
For multi-thread, the fps improves from 110 to 130, about 20%.

Signed-off-by: Ruiling Song <ruiling.song@intel.com>
libavfilter/gblur.h [new file with mode: 0644]
libavfilter/vf_gblur.c
libavfilter/x86/Makefile
libavfilter/x86/vf_gblur.asm [new file with mode: 0644]
libavfilter/x86/vf_gblur_init.c [new file with mode: 0644]