Closed VunnySodhi closed 1 year ago
Attached vainfo and libvatrace files. libva_trace.log.083318.thd-0x00000055.txt vainfo.log libva_trace.log.083317.thd-0x00000053.txt libva_trace.log.083318.thd-0x0000005b.txt
Found out issue is coming from https://github.com/intel/media-driver/blob/intel-media-23.1.0/media_driver/agnostic/gen12/hw/vdbox/mhw_vdbox_vdenc_g12_X.h#L1768 When I run pipeline: on #line 1768 avcPicParams->NumROI is coming as 5 whereas on #line 1772 we are allocating array of 3(ENCODE_VDENC_AVC_MAX_ROI_NUMBER_G9), that why stack is corrupted and pipeline terminates after that.
Not sure we can we increase range of array to ENCODE_VDENC_AVC_MAX_ROI_NUMBER_ADV as its G12 and I am also running on Tigerlake machine.
Hi @XinfengZhang and @Jexu any update on this issue?
Hi @Jexu / @XinfengZhang any update on raised issue?
Hi @VunnySodhi , We will take a look at this issue, most likely on the next week (based in engineering Team bandwidth).
Thanks for the update @DenWolf
Auto Created VSMGWL-65418 for further analysis.
Hi @VunnySodhi ,
I prepared patch for the problem.
You could check if it is work for you
You should add the codtion to https://github.com/intel/media-driver/blob/intel-media-23.1.0/media_driver/agnostic/gen12/hw/vdbox/mhw_vdbox_vdenc_g12_X.h#L1768
like this:
if (!params->bVdencBRCEnabled && avcPicParams->NumROI && avcPicParams->bNativeROI)
Hi @VunnySodhi, Did you have a chance to check if the solution works for you?
Hi @dmenshov I have tried changing https://github.com/intel/media-driver/blob/intel-media-23.1.0/media_driver/agnostic/gen12/hw/vdbox/mhw_vdbox_vdenc_g12_X.h#L1768 to (!params->bVdencBRCEnabled && avcPicParams->NumROI && avcPicParams->bNativeROI) It worked, pipeline works fine with changes and got proper output.
@dmenshov could you help provide a patch for it?
THe patch is on review now. It will back merged to opensource after finishing testing.
Thanks @dmenshov. I'm going to close this issue which has been fixed by https://github.com/intel/media-driver/commit/6f483dd9fb430c439dc324fc105a1e0312d1c9c0. Please feel free to re-open it if any concerns.
Thanks all for fix.
Which component impacted?
Encode
Is it regression? Good in old configuration?
No, this issue exist a long time
What happened?
1) In Linux pull DL Streamer docker image from https://dlstreamer.github.io/get_started/install/install_guide_ubuntu.html#option-2-pull-and-run-intel-dl-streamer-pipeline-framework-docker-image 2) Run following pipeline gst-launch-1.0 filesrc location=test.mp4 ! qtdemux ! h264parse ! vaapih264dec ! video/x-raw(memory:VASurface) ! gvadetect device=CPU pre-process-backend=vaapi model=FP32/face-detection-retail-0004.xml model-proc=face-detection-retail-0004.json ! vaapih264enc tune=low-power ! h264parse ! mp4mux ! fakesink 3) Issue is not coming if we modify above command with vaapih264enc tune=none
Resources: Video file: https://cdn.videvo.net/videvo_files/video/free/video0488/large_watermarked/_import_629da40d4fc106.16643143_preview.mp4 Face-detection-retail-004 model files can be downloaded using: https://www.intel.com/content/www/us/en/developer/articles/technical/video-analytics-service.html#inpage-nav-3-1
What's the usage scenario when you are seeing the problem?
Video Analytics
What impacted?
Not impacting any milestone as of now, only performance impact.
Debug Information
1) Libva: 2.17.0.2-44 / gmmlib: 22.3.3+i550 / media-driver: 23.1.0+i553 2) ls /dev/dri card0 renderD128 3) lspci -nn |grep -Ei 'VGA|DISPLAY' 00:02.0 VGA compatible controller [0300]: Intel Corporation TigerLake-LP GT2 [Iris Xe Graphics] [8086:9a49] (rev 01) 4) Attached vinfo.log 5) Attached libva_trace.log files 6) No dmesg seen (not attached)
Do you want to contribute a patch to fix the issue?
None