A Framework for Transparent Execution of Massively-Parallel Applications on CUDA and OpenCL