TensorAccessor APIs
If one uses TensorAccessor<T, N>
, is there a way to leverage high level operations with the performance benefits of bypassing dynamic typing?
In other words, can we do things like
auto X_ = X.accessor<float, 1>();
auto Y_ = Y.accessor<float, 1>();
at::Tensor Z = torch::dot(X_, Y_);
Similar to how we work with Tensor
types?