Java bindings for the CUDA runtime and driver API. The JCuda Runtime API is mainly intended for the interaction with the Java bindings of the the CUDA Runtime libraries, like JCublas and JCufft instead of creating own kernels. Own CUDA kernels can be launched in JCuda using the Driver API.
With JCuda it is possible to interact with the CUDA runtime and driver API from Java programs. JCuda is the common platform for libraries such as JCublas, JCufft, JCudpp, JCurand and JCusparse. The JCuda Runtime API is mainly intended for the interaction with the Java bindings of the the CUDA Runtime libraries as mentioned above instead of creating own kernels. Java program cannot be compiled using NVCC(NVIDIA CUDA Compiler), hence own CUDA kernels can be launched in JCuda using the Driver API.
For Jcuda to work the CUDA driver and toolkit has to be first installed from http://developer.nvidia.com/cuda-toolkit-archive based on the desired version that suits your GPU. Then Jcuda can be downloaded from http://www.jcuda.org/downloads/downloads.html. Set the jar files location to the CLASSPATH (either a path that is java.library.path for the JVM, or the root directory of project).
Useful links
- JCuda site
- Sample Examples - http://www.jcuda.org/samples/samples.html
