tdegeus / cppmat Goto Github PK

The initial design started out was decoupled as much as possible to allow a lot of flexibility. It is now starting to become a bit bulky.

In 11a31a9 the default pybind11 interface for vectors was introduced for all vector types. This does however not yet include bounds-checks. This needs to be done.

(The upper-bound is checked by push_back, the lower bound is however not checked)

For example use shields.io

See Eigen

Rename classes:

• matrix -> array (rename header to regular_array.h, modify pybind11)
• matrix2 -> matrix (rename header to regular_matrix.h, modify pybind11)
• vector -> vector (rename header to regular_vector.h, modify pybind11)
• periodic::matrix -> periodic::array (rename header to periodic_array.h, modify pybind11)
• periodic::matrix2 -> periodic::matrix (rename header to periodic_matrix.h, modify pybind11)
• periodic::vector -> periodic::vector (rename header to periodic_vector.h, modify pybind11)
• tiny::matrix2 -> tiny::matrix (rename header to tiny_matrix.h, modify pybind11)
• tiny::vector -> tiny::vector (rename header to tiny_vector.h, modify pybind11)
• view::matrix2 -> view::matrix (rename header to view_matrix.h, modify pybind11)
• view::vector -> view::vector (rename header to view_vector.h, modify pybind11)
• Introduce a private.h and corresponding namespace for specific support functions

• Document release of this clean version.
• Updated documentation

Introduce bound checks using assert.

See regular_array.h

Equip every class with:

• copy(...) to copy from (many) overloaded templates.
• cast<...>() to copy to many overloaded templates. This will incorporate the assymmetric, ... functions for the tensors.

This will replace the more implicit conversion operators, which have the risk making casts (including approximations) that stay unnoticed by the user.

N.B. The conversions from tensor2d -> tensor2 will stay as this cast is very common, and allowing this conversion actually leads to more readable code.

Implement cppmat::abs for

• cppmat::array
• cppmat::matrix
• cppmat::vector
• cppmat::periodic::array
• cppmat::periodic::matrix
• cppmat::periodic::vector
• cppmat::tiny::matrix
• cppmat::tiny::vector
• cppmat::view::matrix
• cppmat::view::vector
• cppmat::cartesian::tensor4
• cppmat::cartesian::tensor2
• cppmat::cartesian::tensor2s
• cppmat::cartesian::tensor2d
• cppmat::cartesian::vector
• cppmat::cartesian2d::tensor4
• cppmat::cartesian2d::tensor2
• cppmat::cartesian2d::tensor2s
• cppmat::cartesian2d::tensor2d
• cppmat::cartesian2d::vector
• cppmat::cartesian3d::tensor4
• cppmat::cartesian3d::tensor2
• cppmat::cartesian3d::tensor2s
• cppmat::cartesian3d::tensor2d
• cppmat::cartesian3d::vector
• cppmat::view::cartesian2d::tensor4
• cppmat::view::cartesian2d::tensor2
• cppmat::view::cartesian2d::tensor2s
• cppmat::view::cartesian2d::tensor2d
• cppmat::view::cartesian2d::vector
• cppmat::view::cartesian3d::tensor4
• cppmat::view::cartesian3d::tensor2
• cppmat::view::cartesian3d::tensor2s
• cppmat::view::cartesian3d::tensor2d
• cppmat::view::cartesian3d::vector
• Document

• Disconnect 'header' from implementation everywhere. The finished cppmat::matrix can be used as guideline.
• Make sure that all commuting implementations have the same order, to quickly checks the differences and similarities.

Include copy to eigen everywhere.

• regular_array.h
• regular_matrix.h
• regular_vector.h
• periodic_array.h
• periodic_matrix.h
• periodic_vector.h
• tiny_matrix.h
• tiny_vector.h
• tensor.h
• tensor2.h
• tensor3.h
• view_matrix2.h
• view_vector.h
• view_tensor2.h
• view_tensor3.h

Complete switching to ['Catch'])(https://github.com/philsquared/Catch) for testing. This will immediately allow to compile and run all tests at once.

Completed:

• verify_matrix.cpp

Todo:

• verify_tensor.cpp
• verify_tensor2.cpp
• verify_tensor3.cpp

Ultimately it would be good to switch to some automatic build and test system like travis CI.

Complete mixed size_t and int operators for 4-d, 5-d, and 6-d matrices

Increase readability of CMakelists.txt, see https://github.com/tdegeus/ElastoPlasticQPot/blob/master/src/CMakeLists.txt

• tensor2.h: <-> view_tensor2.h: operator? (X, X) (2e79399)
• tensor2.h: <-> view_tensor2.h: operator? (X), class member
• tensor2.h: <-> view_tensor2.h: tensor products
• tensor2.h: <-> view_tensor2.h: tensor products, class member
• tensor3.h: <-> view_tensor3.h: operator? (X, X) (2e79399)
• tensor3.h: <-> view_tensor3.h: operator? (X), class member
• tensor3.h: <-> view_tensor3.h: tensor products
• tensor3.h: <-> view_tensor3.h: tensor products, class member
• matrix2.h: <-> view_matrix2.h: operator? (X, X)
• matrix2.h: <-> view_matrix2.h: operator? (X), class member
• vector.h: <-> view_vector.h: operator? (X, X)
• vector.h: <-> view_vector.h: operator? (X), class member

That is the real issue here really. Awesome work, and awesome documentation. You deserve alot of cred for your work 👍

Implementation option 1

#include <iostream>

template<class X, int nd=2>
class Foo
{
  static_assert( nd == 2 );

  private:
    X data[nd];

  public:
    Foo() = default;
};

Implementation option 2

#include <iostream>

template<class X>
class Foo
{
  private:
    const expr int nd = 2;
    X data[nd];

  public:
    Foo() = default;
};

Warning: this has some limitations, such as binding to a reference

Create .cpp functions with the implementation, that are normally included and inlined, but can be excluded using for example a define like CPPMAT_LIBRARY. See doctopt.cpp.

Benefits:

• Leave all compilation strategies open to the user.

Introduce argmax and argmin in all modules:

• regular_array.h
• regular_matrix.h
• regular_vector.h
• periodic_array.h
• periodic_matrix.h
• periodic_vector.h
• tiny_matrix.h
• tiny_vector.h
• tensor.h
• tensor2.h
• tensor3.h
• view_matrix2.h
• view_vector.h
• view_tensor2.h
• view_tensor3.h