https://hgpu.org/?p=26330
Query Processing on Tensor Computation Runtimes