Performance study of using the Direct Compute API for implementing Support vector machines on GPUs