Closed mikk9 closed 10 months ago
The parameters in the log is acquired by calling MFXVideoENCODE_GetVideoParam and somehow it is always returning tskip:on, ctu:64, sao:all, even if I set "--sao none" or "--ctu 16"...
I'm not sure the cause of this (whether my implementation is wrong or the driver error), but as the other parameters are returned correctly, I think there are some problem specific to these parameters (ctu, tskip and sao).
As feature check shows it is not supported, I think that it is not enabled in actual encoding.
I will remove printing these parameters from the next version, as it shows confusing results as you have mentioned.
On Kabylake Gen9 it didn't show tskip and sao in the encoding log. CTU seems correct also, ctu 32 for Gen9 hardware which is correct and ctu 64 for Gen12, so I would guess tskip+sao is enabled on Gen12 hardware but for whatever reason it can't be enabled/disabled and properly detected from the feature check.
I can confirm that I cannot deactivate SAO on the 12th generation.
Gen12 graphics seems to have problem with ctu, tskip and sao. These will be not shown from QSVEnc 6.00 as it is rather confusing.
I've tried QSVEnc 7.05 with newest driver and it works now, I can disable/enable SAO. I can also enable/disable tskip. I can say SAO and tskip were enabled from the beginning, the parameter in the log were correct. I think you can re-enable the tskip, SAO, ctu log parameter.
There is a confusion about SampleAdaptiveOffset. According to feature check it's not supported, but in the log (from Tigerlake) it says "sao: all" which suggests it's enabled. We can't disable it though, same for tskip by the way.
Ext. Features WeightP WeightB QPOffset tskip ctu:64 sao:all
There is also a hevcehw_g12_sao.h on github: https://github.com/Intel-Media-SDK/MediaSDK/blob/510d19dcace1d8c57567fdd40b557155ab11ab8e/_studio/mfx_lib/encode_hw/hevc/agnostic/g12/hevcehw_g12_sao.h
For grainy video content it can be beneficial to disable SAO because it tends to smooth things out, sometimes it's not needed.