sample_decode jpeg fail with gpu hang

[Daydaysee]

Platform: Centos 7.4 Kernel 5.2.4 with E3-1515M
decode gpu-hang.jpg using sample_decode, the program will get stuck for a while and then return normally, but the decoded image is invalid, while dmesg show "i915 0000:00:02.0: Resetting vcs0 for hang on vcs0"

The attachment is gpu-hang.jpg，gpu hang can be reproduced with decoding this image.

#dmesg -T
[Tue Aug 25 16:17:48 2020] i915 0000:00:02.0: Resetting vcs0 for hang on vcs0
[Tue Aug 25 16:25:29 2020] i915 0000:00:02.0: Resetting vcs0 for hang on vcs0

#./sample_decode jpeg -i gpu-hang.jpg -o out.yuv
pretending that aspect ratio is 1:1
libva info: VA-API version 1.4.1
libva info: va_getDriverName() returns 0
libva info: User requested driver 'iHD'
libva info: Trying to open /usr/lib/dri//iHD_drv_video.so
libva info: Found init function __vaDriverInit_1_4
libva info: va_openDriver() returns 0
Decoding Sample Version 8.3.26.

Input video JPEG
Output format NV12
Input:
Resolution 352x288
Crop X,Y,W,H 0,0,352,288
Output:
Resolution 352x288
Frame rate 30.00
Memory type system
MediaSDK impl hw
MediaSDK version 1.29

Decoding started
Frame number: 1, fps: 0.128, fread_fps: 0.000, fwrite_fps: 2597.403
Decoding finished

[IntelSupport]

Hi Daydaysee,

What version of Media SDK are you using? I believe API 1.29 was included in release 2019 Q1 which is quite old. Do you see the same behavior on the latest Media SDK 20.2 release?

Regards,

Jesus

[Daydaysee]

We have saw the same behavior  with MediaSDK 20.2.0.

[Tue Sep 8 11:18:57 2020] i915 0000:00:02.0: Resetting vcs0 for preemption time out
[root@localhost samples]# ./sample_decode jpeg -i gpu-hang.jpg
libva info: VA-API version 1.8.0
libva info: Trying to open /usr/lib/dri/iHD_drv_video.so
libva info: Found init function __vaDriverInit_1_8
libva info: va_openDriver() returns 0
pretending that aspect ratio is 1:1
Decoding Sample Version 20.2.0

Input video JPEG
Output format NV12
Input:
Resolution 352x288
Crop X,Y,W,H 0,0,352,288
Output:
Resolution 352x288
Frame rate 30.00
Memory type system
MediaSDK impl hw
MediaSDK version 1.33

Decoding started
Frame number: 1, fps: 0.148, fread_fps: 0.000, fwrite_fps: 0.000
Decoding finished
[root@localhost samples]#

[Mark_L_Intel1]

Hi Guoqing,

Sorry for the late response.

Yes, I can reproduce this issue. I also did some investigation on the input file, its meta data looks normal, but I got the same error when I open it with my Preview app in Mac.

I have reported this to dev team and I will update you with the result.

Mark

[Mark_L_Intel1]

Hi Guoqing,

Just an update.

Dev team has completed the investigation and identify the root cause. They will put the fix next.

Mark

[Mark_L_Intel1]

Hi Guoqing,

I have confirmed this issue is fixed by the commit:

https://github.com/intel/media-driver/commit/d0571ead99efae0c36fb16133a4c8cd987575a8d

Please try it and let me know if this issue is still exists in the latest code.

Mark