-
Notifications
You must be signed in to change notification settings - Fork 5.8k
New issue
Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? # to your account
Accuracy cudacodec test fails always with -DWITH_NVCUVID=ON #3362
Comments
Is it possible that you have copied the stub ( Does
give you something similar to
instead of
If so, can you remove the stub library (
or
|
@cudawarped, Yes, I had similar to
and after this removed |
@cudawarped I use the latest nvidia driver 515.65.01 version and 11.7 CUDA. |
@asenyaev If
is pointing to the correct place
and the error code is still garbage (not 1 to 999) and coming from the same location
I'm out of idea's. |
Should it pointing certainly to
Or it can be like this?
|
No it shouldn't be pointing to that, that's just the stub, OpenCV can use that to link against but, it needs to call into the actual library provided by the driver. If you don't pass the stub
does cmake pick up the driver lib here
or in a similar location? Did you remove both stubs libnvcuvid.so and libnvcuvid.so.1 from your cuda install? |
Cmake picks up from here Sure, removed both of them. |
The stubs should allow you to link against them then use the driver lib, that's there purpose, maybe the driver lib can't be found or doesn't exist for some reason. I would try just passing in the location of the driver lib e.g.
|
No, it also is not a solution :( Even if a linkage like this
all tests fail |
I'm not sure I can be any help, my linux skills are limited at best. @asmorkalov has it working on his local machine so maybe he can shed some light on why its not working. To summarize I can recreate the issue by copying the stubs into the
and then fails because its a stub, so the first call into fails. Then if I remove the lib ldd gives me the correct path (this is on wsl so hence the different path)
and everything works. I am not sure why you are seeing
if this isn't on your load library path, unless Additionaly if I don't pass the stub library then cmake picks up the driver version automatically and everything works
which it doesn't on your system. |
@cudawarped thanks for assistance. We are working with @asenyaev together trying to debug the issue on CI side. |
@cudawarped, the only one difference can be on my side, I run it inside |
So, I tried to do it on a host (w/o container usage), but the same results happened. My steps:
Output:
Unfortunately, all tests are failed. Failure messages are the same as they are in the body of this issue. |
Don't do this. The shared lib should already exist in your driver folder. The stub is there so you can compile on a system which doen't have an nvidia gpu and therefore won't have any cuda shared libs in What is your location for |
@cudawarped, thank you for your support! Finally, I made it to work. I re-installed all nvidia components and just did not install NVIDIA Video codec SDK over CUDA, but copied headers from Firstly, Tests started to run successfully also, but some of them failed anyway:
It happened in case of hardware decoding, example:
|
I have test failure for av1 codec too. Most probably it depends on FFmpeg version. |
AV1 is only supported on 5th gen NVDEC, P2000 is 3rd gen and not supported but this should get picked up here. My guess is that check needs to be made earlier using FFmpeg's provided codec (if its supported by FFmpeg) as the parser is failing to create the decoder. Looks like a bug in so much as it should throw a more informative exception, I'll test on older hardware and fix when I have chance. Strangely VP8 is also not supported on your P2000. |
Some updates for this issue how to build OpenCV with
I'm closing this issue, because the error has disappeared and new failures do not relate to the main issue. @cudawarped thank you! |
Accuracy:cudacodec
fails always if-DWITH_NVCUVID=ON
flag is set for a build, all tests are failed there.List of failed tests:
An example of a failed test:
System information (version)
The text was updated successfully, but these errors were encountered: