Minimal header only linear algebra library with expression templates and low footprint designed to run on embedded devices.
A matrix is generic over the scalar type, the number of rows and the number
of columns, dynamically sized matrices are not supported.
The internal buffer is always allocated on the stack as a contiguous
row-major array and by default is set to 0.
There are three classes of vectors, two concrete and one as a view.
A column_vector<Type, Size> is just a type alias for matrix<Type, Size, 1>
but in addition to expression access it implements direct access through the
[] operator.
A row_vector<Type, Size> is just a type alias for matrix<Type, 1, Size> but
in addition to expression access it implements direct access through the []
operator.
A vector is a column or row vector view into an expression (or matrix if
mutable access is required).
ela::matrix<float, 3, 3> a{{1, 0, 3}, {4, 0, 6}, {7, 0, 9}};
// Assign a column of a matrix.
a.column(1) = {2, 5, 8};
// Scale a row of the matrix and save it as a column vector.
ela::column_vector<float, 3> b = ~(a.row(1) * 2);Bound checks are used at any runtime indexing operation through a call to
ELA_ASSUME.
Unless ELA_ASSUME is defined before ela/ela.hpp is included its definition
will become an assert call when NDEBUG is not defined and a compiler
unreachable hint when NDEBUG is defined.
In your custom expression implementations you're advised to call ELA_ASSUME
yourself to make sure the compiler knows what's going on.
All matrix operations are implemented as expression templates so no
intermediary objects are created unless explictly assigned to a matrix.
If all your matrix values are known at compile time compilers are able to completely constant unfold the results most of the time.
An expression is any type that implements the ela::expression::traits and
provides the appropriate indexing operator.
We'll implement a generic RGB type as if it were a column vector.
/* Inheriting from `ela::expression::base` is not required but it automatically
* implements all generic expression operators for free, it doesn't add any
* data.
*/
template <typename Type>
struct RGB: public ela::expression::base<RGB<Type>>
{
public:
using ela::expression::base<RGB<Type>>::operator =;
public:
Type r = 0;
Type g = 0;
Type b = 0;
/* Create an empty color.
*/
RGB () noexcept
{ }
RGB (Type r, Type g, Type b) noexcept
: r(r), g(g), b(b)
{ }
template <typename Input, typename T = Type>
RGB (Input const& expr, typename std::enable_if<
3 == ela::expression::traits<Input>::rows &&
1 == ela::expression::traits<Input>::columns &&
std::is_same<T, typename ela::expression::traits<Input>::type>::value>::type* = 0) noexcept
{
ela::expression::base<RGB<Type>>::operator=(expr);
}
RGB (std::initializer_list<std::initializer_list<Type>> elements) noexcept
{
ela::expression::base<RGB<Type>>::operator=(elements);
}
RGB (std::initializer_list<Type> elements) noexcept
{
ela::expression::base<RGB<Type>>::operator=(elements);
}
/* This is the expression access operator.
*/
inline
Type const&
operator () (size_t row, size_t column) const noexcept
{
ELA_ASSUME(row < 3 && column == 0);
if (row == 0) {
return r;
}
else if (row == 1) {
return g;
}
else {
return b;
}
}
/* This is the expression access operator.
*/
inline
Type&
operator () (size_t row, size_t column) noexcept
{
ELA_ASSUME(row < 3 && column == 0);
if (row == 0) {
return r;
}
else if (row == 1) {
return g;
}
else {
return b;
}
}
};
namespace ela { namespace expression {
/* This defines the experssion traits for `RGB<T>`.
*/
template <typename Type>
struct traits<RGB<Type>>
{
/* This is always the scalar type.
*/
typedef Type type;
/* This is the number of rows the expression will produce.
*/
static constexpr size_t rows = 3;
/* This is the number of columns the expression will produce.
*/
static constexpr size_t columns = 1;
/* This says the expression can return references.
*/
static constexpr bool concrete = true;
};
} }
This code will make RGB behave as an expression, you can look into tests/
for more examples of expressions and other stuff.