[Bug]: vaGetConfigAttributes returns wrong value of cu_qp_delta in VaConfigAttribValEncHevcFeatures with low-power mode

[mengker33]

Which component impacted?

Encode

Is it regression? Good in old configuration?

None

What happened?

In hevc encoding with low-power mode, vaGetConfigAtttibutes return an incorrect value (i.e. 2)of cu_qp_delta in VaConfigAttribValEncHevcFeatures, which causes artifacts in the encoded clip. See the wrong one:

If returned value is 3 in low-power, the result is correct: See the correct one:

What's the usage scenario when you are seeing the problem?

Others

What impacted?

It impacts the whole hevc encoding in low-power mode, simply means it can possibly impact every listed selection above.

Debug Information

No response

Do you want to contribute a patch to fix the issue?

None

[Sherry-Lin]

Is it common issue or specific to which platform?

[mengker33]

Is it common issue or specific to which platform?

I think it is a common one as I can reproduce on tgl as well as on mtl.

[intel-mediadev]

Auto Created VSMGWL-66648 for further analysis.

[mengker33]

Hey, any updates?

[bai-isaac]

hi mengker33, which va version do you use? and could you please share how to reproduce this issue? Thanks!

[mengker33]

hi mengker33, which va version do you use? and could you please share how to reproduce this issue? Thanks!

I use libva 2.19.0. The issue can be reproduced using gstreamer-va low-power hevc encoder, i.e. vah265lpenc

gst-launch-1.0 -vf videotestsrc num-buffers=30 ! video/x-raw,format=NV12 ! vah265lpenc ! h265parse ! filesink location=output.h265 

You can see the output.h265 has the artifacts as above one.

[OttoYang]

Hi @mengker33 We return this param cu_qp_delta as an enable flag not as depth.
The depth only supports 0 or 3 due to HW limitation in low-power mode. So you may need to change the code in gstreamer as shown below:

pps->cu_qp_delta_enabled_flag = !!features.bits.cu_qp_delta;
pps->diff_cu_qp_delta_depth = sps->log2_diff_max_min_luma_coding_block_size == 3 ？sps->log2_diff_max_min_luma_coding_block_size: 0;

[mengker33]

Understood！ Appreciate your quick response :)

[OttoYang]

Hi @mengker33 May I know if the fix works for you?

[mengker33]

Yes, I will modify codes on GStreamer side BTW, one more question, why need to set depth to maximum when this flag is enabled?

[OttoYang]

Yes, I will modify codes on GStreamer side BTW, one more question, why need to set depth to maximum when this flag is enabled?

Hardware only supports 0 or 3(LCU-64) in low-power mode now. So other depths may lead unexpected behavior.

[mengker33]

Got it. Thank you, let me close the ticket.