https://hgpu.org/?p=5588
A Light-Weight Approach to Dynamical Runtime Linking Supporting Heterogenous, Parallel, and Reconfigurable Architectures