Problem of building OpenCV 2.4.5 with Cuda 5.5 RC

asked 2013-05-30 22:47:47 -0600

148 ●1 ●4 ●14

When I use cmake to configure, it will it cannot found CUDA_npp_LIBRARY

CMake Error: The following variables are used in this project, but they are set to NOTFOUND.
Please set them or make sure they are set and tested correctly in the CMake files:
CUDA_npp_LIBRARY (ADVANCED)
    linked by target "opencv_core" in directory /home/gianluigi/Lib/opencv-2.4.5/modules/core
    linked by target "opencv_gpu" in directory /home/gianluigi/Lib/opencv-2.4.5/modules/gpu
    linked by target "opencv_nonfree" in directory /home/gianluigi/Lib/opencv-2.4.5/modules/nonfree

I think this is caused by NVIDIA split the libnpp to libnppc, libnppi and libnpps in CUDA 5.5. Is there any way to fix this problem?

edit retag flag offensive close merge delete

add a comment

5

answered 2013-05-31 15:41:51 -0600

cuda.geek
121 ●2 ●3

NVIDIA divide npp into 3 libraries.

libnppc.so is general libnppi.so is for image processing libnpps.so is for signal processing

So, just a hack for now open your CMakeCache.txt and manually set it to your path. Like following //"npp" library CUDA_npp_LIBRARY:FILEPATH=/opt/cuda/lib64/libnppc.so;/opt/cuda/lib64/libnppi.so

I guess it'll be fixed in next version of cmake. Anyway we should workaround it in OpenCV.

edit flag offensive delete link

Comments

Thanks, it works!

Gianluigi ( 2013-06-02 21:25:37 -0600 )edit

Sorry, it can configure now, but it appear many error when I build it. Like, error: calling a __device__ function("AddMat") from a __host__ function("addMat") is not allowed detected during instantiation of "void arithm::addMat<T,D>(cv::gpu::PtrStepSzb, cv::gpu::PtrStepSzb, cv::gpu::PtrStepSzb, cv::gpu::PtrStepb, cudaStream_t) [with T=ushort, D=double]"

Gianluigi ( 2013-06-02 21:51:35 -0600 )edit

add a comment

1

answered 2013-06-05 02:47:37 -0600