I have to paralelize a svm aplication, basically I have a vector and I must call the svm kernell in each element in a paralell way. So in my cuda kernell y have a call to the svm routine.
But cuda don’t support calling external functions from a kernel function.
Any idea or help to make it possible? I already try using http://mklab.iti.gr/project/GPU-LIBSVM but when I tried to compile it following the instructions it says the same error “calling a host function from a device function is not allowed”