GSParLib is a C++ object-oriented multi-level programming interface for GPU programming that allows code portability between different GPU platforms and targets stream and data parallelism.

It provides a unified interface and driver-agnostic runtime, which enables applying the GPU parallelism to a given application once and using it with different GPU vendors or GPU backends (currently CUDA and OpenCL) without introducing any modifications in the source code.

GSParLib's programming interface is composed of two layers, and each layer offers different abstraction levels: