Based on testing from the discord, opt=3 seems to give reliable speed boosts compared over opt=0 if mclip is used regardless of CPU. Furthermore, avx512 seems to ALWAYS perform worse than avx. The latter is not part of this PR, but worth keeping in mind for future PRs.
Zander (7950x):
opt=0 - encoded 200 frames in 0:00:35.65 (5.61 fps) (auto)
opt=1 - encoded 200 frames in 0:00:40.24 (4.97 fps) (c)
opt=2 -encoded 200 frames in 0:00:27.95 (7.15 fps) (sse2)
opt=3 - encoded 200 frames in 0:00:26.53 (7.54 fps) (sse4.1)
opt=4 - encoded 200 frames in 0:00:31.01 (6.45 fps) (avx)
opt=5 - encoded 200 frames in 0:00:35.67 (5.61 fps) (avx512)
Based on testing from the discord, opt=3 seems to give reliable speed boosts compared over opt=0 if
mclip
is used regardless of CPU. Furthermore, avx512 seems to ALWAYS perform worse than avx. The latter is not part of this PR, but worth keeping in mind for future PRs.Zander (7950x):
sinc:
noiy (7700x - 32GB - Linux 6.10.7):
smol (i9-13900k (-100mv undervolt, ICCMax=307A PL1=165 PL2=253, new bios)):
basic: (first is 3, second is 0)