State-of-the-art neural networks depend on linear operations, such as matrix-vector multiplications and convolutions. While dedicated processors like GPUs and TPUs exist for these operations, they ...