Query Processing on Tensor Computation Runtimes