https://hgpu.org/?p=18844
On the Representation of Partially Specified Implementations and its Application to the Optimization of Linear Algebra Kernels on GPU