[BRX/APL] Very Slow frame fetch from the GPU to CPU #1191

[Kevin_T_Tang]

J4125(Gen9) has low performance for sample_decode:

sample_decode h264 -i 1080p.264 -o /dev/null -rgb4

On the Windows side, with 2021R1, it is 30fps

On the Linux side, only 15fps, any ideas to improve the performance?

Related links:

https://github.com/intel/media-driver/issues/1191 

https://github.com/intel/media-driver/pull/646  

[AlekhyaV_Intel]

Hi,

Thank you for posting in Intel Communities. We are working on this internally. Could you please let us know your OS details?

Regards,

Alekhya

[AlekhyaV_Intel]

Hi Kevin,

We tried reproducing your issue on a Coffee Lake machine but couldn't reproduce your issue. IoT processors, e.g., Gemini Lake Processor, have relatively simple/slow CPU cache hierarchies, so I/O will be slow. That's why the problem isn’t apparent on Coffee Lake Processor.

To see the true performance of the HW, please use the following command line:

sample_decode h264 -i 1080p.264

(Remove the pixel format conversion options and output options)

You may need to add flags to keep the frames in video memory like '-vaapi' for Linux.

Performance should increase with multiple simultaneous decodes. 

If this resolves your issue, please accept this as a solution. This might help others with similar issue. Thanks.

Regards,

Alekhya

[AlekhyaV_Intel]

Hi,

Is your issue resolved? Could you please give us an update regarding this issue?

Regards,

Alekhya

[AlekhyaV_Intel]

Hi,

We assume that your issue is resolved. If you need any further assistance, please post a new question as this query will no longer be monitored by Intel.

Regards,

Alekhya