Frame scaling using VQE or VEBOX in Skylake P580 graphics card

[rajesh_k_]

Hi,

I have a question related to Scaling operation. I am building a transcode pipeline application using P580 graphics card. 

1. I want to know how many Mfx engines and VQE engines are present in P580 Iris pro Graphics card -GT4e  version?

2.I have read that VEBOX engine can do the scaling operation. but when i used the "RunFrameVPPAsync"  API  , from the metrics monitor logs i see that VEBOX is idle . VIDEO_E is zero all the time from the below logs.I want to offload the scaling task to VEBOX . 

RENDER usage: 13.00,    VIDEO usage: 26.00,     VIDEO_E usage: 0.00     GT Freq: 950.00
RENDER usage: 31.00,    VIDEO usage: 22.00,     VIDEO_E usage: 0.00     GT Freq: 950.00
RENDER usage: 31.00,    VIDEO usage: 27.00,     VIDEO_E usage: 0.00     GT Freq: 950.00
RENDER usage: 30.00,    VIDEO usage: 23.00,     VIDEO_E usage: 0.00     GT Freq: 950.00
RENDER usage: 28.00,    VIDEO usage: 22.00,     VIDEO_E usage: 0.00     GT Freq: 950.00
RENDER usage: 28.00,    VIDEO usage: 25.00,     VIDEO_E usage: 0.00     GT Freq: 950.00
RENDER usage: 28.00,    VIDEO usage: 27.00,     VIDEO_E usage: 0.00     GT Freq: 950.00
RENDER usage: 37.00,    VIDEO usage: 29.00,     VIDEO_E usage: 0.00     GT Freq: 950.00

could you help me how to do that?

thanks

Rajesh

[Artem_S_Intel]

Hi Rajesh, what is your operation system? Linux? On linux VEBox scaling is not supported.

For transcoding on P580 Iris pro Graphics card -GT4e  VEBox scaling will show lower performance and lower quality than render based.

[rajesh_k_]

Hi Artem,

Thanks for your quick reply.

we are working on Linux . we are looking for performance. since VeBox is  Fixed function we are thinking that it would give higher performance than doing it on EU's.

could you also please answer my first question?

1.  I want to know how many Mfx engines and VQE engines are present in P580 Iris pro Graphics card -GT4e  version?

Thanks 

Rajesh

[Artem_S_Intel]

You can find out specification at 01.org site, particular one for SKL:

https://01.org/sites/default/files/documentation/intel-gfx-prm-osrc-skl-vol04-configurations.pdf

It has 2 VEBox units. Only one of them can do scaling as it is associated with SFC(only one unit) - that is why performance won't be better than using render for scaling. So far you can try it on windows  via mfxExtVPPScaling(set mode to lowpower) to verify.

[rajesh_k_]

Thanks Artem for the confirmation.

-Rajesh