https://hgpu.org/?p=6503
Automatic Fusions of CUDA-GPU Kernels for Parallel Map