Roadmap

[lgarithm]

Overview

The goal of stdtensor is to provide a C++ library for tensors in STL style. It will allow user to write program in a type safe and generic way.

Components

generic container data types

template <typename R, rank_t r>
tensor_t<R, r>;  // the data owner type

template <typename R, rank_t r>
tensor_ref_t<R, r>; // the reference type

template <typename R, rank_t r>
tensor_view_t<R, r>; // the read only reference type, like https://en.cppreference.com/w/cpp/header/string_view

All the 3 kinds of data types should have the following methods: They should be as consist as https://en.cppreference.com/w/cpp/container/vector

R* data();
const R* data();

R &at(...);
const R at();

rank_t rank();
std::array_t<dim_t, r> dims();

size_t size(); // the product of all dims

generic algorithms for tensors

advanced features

• allow memory to be allocated on GPU via thrust::device_vector
• small tensors with constexpr shape #3

Potential usages:

• type safe memory buffer
• the tensor counter part of std::vector
• the C++ port of np.array
• build deep neural network framework (like http://github.com/lgarithm/crystalnet)
• as a wrapper interface to other linear algebra libraries, including the Blas family, Eigen, OpenCV, thrust, etc.

[lgarithm]

Lattice of types

	typed	flat	raw
own	tensor<R, r>	flat_tensor<R>	raw_tensor
ref	tensor_ref<R, r>	flat_tensor_ref<R>	raw_tensor_ref
view	tensor_view<R, r>	flat_tensor_view<R>	raw_tensor_view