CUDA C++ template library and Python DSLs for high-performance GEMM and linear algebra with reusable tiling and Tensor Core abstractions.