Stream waits in corners.cu before textures destruction

[gagicb]

Using a texture handle after calling cudaDestroyTextureObject results in undefined behavior.
In GFTT/corners, we had the following sequence:

Enter function F
Create texture T
Launch kernel K on stream S
Return from function F, which destroys texture T

At step 4, we must guarantee that kernel K launched on stream S has completed and that T is no longer used before it is destroyed. Thats why I added stream synchronization before function return.

Relevant description from the CUDA SDK documentation:

https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TEXTURE__OBJECT.html#group__CUDART__TEXTURE__OBJECT__1_gae41ef138ba69d54b91068247f8f6c4e8:~:text=Use%20of%20the%20handle%20after%20this%20call%20is%20undefined%20behavior.

Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

• I agree to contribute to the project under Apache 2 License.
• To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
• The PR is proposed to the proper branch
• There is a reference to the original bug report and related work
• There is accuracy test, performance test and test data in opencv_extra repository, if applicable
  Patch to opencv_extra has the same branch name.
• The feature is well documented and sample code can be built with the project CMake

[asmorkalov]

cc @cudawarped

[asmorkalov]

@gagicb Could you add some issue description and why you add the synchronization for non-default stream. Some reference to documentation is useful too.

[gagicb]

@gagicb Could you add some issue description and why you add the synchronization for non-default stream. Some reference to documentation is useful too.

Hi @asmorkalov,

Using a texture handle after calling cudaDestroyTextureObject results in undefined behavior.
In GFTT/corners, we had the following sequence:

1. Enter function F
2. Create texture T
3. Launch kernel K on stream S
4. Return from function F, which destroys texture T

At step 4, we must guarantee that kernel K launched on stream S has completed and that T is no longer used before it is destroyed. Thats why I added stream synchronization before function return.

Relevant description from the CUDA SDK documentation:

https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TEXTURE__OBJECT.html#group__CUDART__TEXTURE__OBJECT__1_gae41ef138ba69d54b91068247f8f6c4e8:~:text=Use%20of%20the%20handle%20after%20this%20call%20is%20undefined%20behavior.

Best regards,
gagik