A Thread-Register Decoupled GPU Execution Model for Efficient Tensor Computation

17 points | by matt_d 7 hours ago