I have been trying to compile OpenCV 3.0 with CUDA support on a NVidia Jetson TK1 (which supports CUDA). According to the OpenCV docs:
"To enable CUDA support, configure OpenCV using CMake with WITH_CUDA=ON . When the flag is set and if CUDA is installed, the full-featured OpenCV CUDA module is built. Otherwise, the module is still built but at runtime all functions from the module throw Exception <http://docs.opencv.org/3.0-beta/modules/core/doc/utility_and_system_functions_and_macros.html#Exception%20:%20public%20std::exception> " So CUDA is a build time dependency for OpenCV. Cuda libraries are provided by Nvidia in binary form via .deb files. So what I did was to make recipes that installs the binaries from the .deb files. I wrote them in such a way that all other binaries are dependencies (DEPENDS) of cuda-toolkit-6-5 so that when I install cuda-toolkit-6-5, every library associated with CUDA gets installed. I then made cuda-toolkit-6-5 a build dependency (DEPENDS) of opencv, and added the -WITH_CUDA=ON cmake flag. But still when I inspect the do_configure logs of OpenCV, I find that CUDA support has not been turned on. I am guessing that FindCUDA.cmake is not finding the binaries installed by cuda-toolkit-6-5. How do I fix this? Or were do I check for the problem? Here are the recipes I wrote for installing CUDA binaries: https://github.com/daltonfury42/meta-yocto-jetson-mylayer/tree/master/recipes-cuda/cuda Here is the recipe that I used for OpenCV which is a modification of the one from meta-oe: https://github.com/daltonfury42/meta-yocto-jetson-mylayer/blob/master/recipes-opencv/opencv/opencv_3.0.bb Here is the do_configure log: https://paste.ubuntu.com/15042198/
-- _______________________________________________ yocto mailing list [email protected] https://lists.yoctoproject.org/listinfo/yocto
