#pragma once #include namespace torch::nativert { // lay out all tensors contiguously in memory // this doesn't take into account lifetimes, // it literally just puts them all next to each other LayoutPlan BumpAllocationPlanner( const std::vector& allocation_specs); } // namespace torch::nativert