latency/slowness in video of simultaneous decode and display with composition on ATOM E3940

[SaurabhShah]

We have done some performance related test cases in below environment:

Board: AAEON UP2, Intel Atom E3940, 4GB RAM, 32 GB eMMC

OS: Ubuntu 20.04

Sample: SVET Linux Sample 2020.3.0 (video_e2e_sample)

Please find attached .par files for reference, we have used. We are focused on pipelines with Decode + Compose + Display with DRM backend.

Observations with Rendering on a HDMI display with 1080p resolution

1. 16 channel, H264 codec, D1 resolution, 30 fps, 1024 Kbps bitrate (working fine)
2. 25 channel, H264 codec, D1 resolution, 30 fps, 1024 Kbps bitrate (working fine)
3. 36 channel, H264 codec, D1 resolution, 30 fps, 1024 Kbps bitrate (not working: Latency & slowness is observed on display)
4. 64 channel, H264 codec, D1 resolution, 30 fps, 1024 Kbps bitrate (working fine with higher CPU usage)

Observations with Rendering on a HDMI display with 4k resolution 

1. 16 channel, H264 codec, D1 resolution, 30 fps, 1024 Kbps bitrate (working fine)
2. 25 channel, H264 codec, D1 resolution, 30 fps, 1024 Kbps bitrate (not working: Latency & slowness is observed on display)
3. 36 channel, H264 codec, D1 resolution, 30 fps, 1024 Kbps bitrate (not working: Latency & slowness is observed on display)
4. 64 channel, H264 codec, D1 resolution, 30 fps, 1024 Kbps bitrate (not working: Latency & slowness is observed on display)

Overall, we are observing latency/slowness on Display while doing the composition of various combinations.

What is the root cause of latency/slowness in video though there is free room available in CPU/GPU?

[AthiraM_Intel]

Hi,

Thanks for reaching out to us.

Could you please let us know whether you are using sample(video_e2e_sample) from the below link:

https://github.com/intel-iot-devkit/concurrent-video-analytic-pipeline-optimization-sample-l

If your issue is related to sample given in the link then you can post your query in github because we don't support those samples.

Thanks.

[SaurabhShah]

Yes.

But we also took same trials with sample_multi_transcode that is part of Media SDK and got same observations.

https://github.com/Intel-Media-SDK/MediaSDK/tree/master/samples

[AthiraM_Intel]

Hi,

Sorry for the delay. The issue may be due to your hardware.

Could you please try the same test in a different hardware.

If you observe the same result, please let us know.

Thanks

[SaurabhShah]

Hi,

Update from my end.

We came across a given link for a similar kind of issue reported earlier.
https://github.com/Intel-Media-SDK/MediaSDK/issues/803
After modifying our par file with tile_id option, 25 channel and 36 channels video wall combinations are working fine on 4k resolution. But the 64 channels wall is still facing slowness/latency. Please advice.

Addition to this, we are also supporting custom display resolution(TV adjust) features in our product. Please find the attached par file for the same. 16 channels with standard resolution was working fine but 16 channels with custom display resolution(TV adjust) is facing the same issue of slowness/latency. After using the tile_id option, it also starts working fine.

16_d1_4k.par  (working fine)
16_d1_4k_tvadjust.par (not working: Latency & slowness is observed on display)
16_d1_4k_tvadjust_tile.par (working fine)

Hence with no known side effect but with better performance, we concluded that tile option should always be used irrespective of no. of channels or display resolution. Please confirm our understanding and also let us know whether any reference document available on how to use tile_id.

[AthiraM_Intel]

Hi,

We are looking into this internally and will get back to you soon.

Thanks.

[Mark_L_Intel1]

Hi SaurabhShah,

Sorry for the late update, this issues was reported to dev team in March and we are waiting for their investigation. I didn't get their update yet.

Mark Liu