Efficient Interleaved Batch Matrix Solvers for CUDA