https://hgpu.org/?p=3371
Programming Massively Parallel Architectures using MARTE: a Case Study