Hi I've been trying to update some code to run on a 1080 that ran fine on a 970 under Cuda 7.5. Initially I had some problems with the build, but after replacing the thrust that comes with Cuda 8.0rc it seems to build fine.
Some really basic debugging -not super familiar with OpenCVs inner workings- has narrowed it down to when I call cvtColor or setTo on a gpuMat it breaks on the transformSimple with OpenCV error reporting an (invalid device function). The weird thing is that this happens only the second time I execute the function as I wrote a simple Cuda setTo and that ran fine but instead of breaking on setTo it broke on the cvtColor after running the first loop.
Thought it might be an issue with the PTX version in the build config but all the other OpenCV Cuda functions -like Canny and MinEigenValCorner detectors- are executing in between and not causing a problem.
Not sure whats going on if anyone has any insight that would be great.