https://hgpu.org/?p=11746
Implementation of Just In Time Value Specialization for the Optimization of Data Parallel Kernels