xfunction¶

Defined in xtensor/xfunction.hpp

template<class F, class ...CT> class xfunction : private xt::xconst_iterable<xfunction<F, CT...>>, public xt::xsharable_expression<xfunction<F, CT...>>, private xt::xconst_accessible<xfunction<F, CT...>>, public extension::xfunction_base_t<F, CT...>¶

Multidimensional function operating on xtensor expressions.

The xfunction class implements a multidimensional function operating on xtensor expressions.

Template Parameters

F: the function type
CT: the closure types for arguments of the function

Constructor

template<class Func, class ...CTA, class U = std::enable_if_t<!std::is_base_of<std::decay_t<Func>, self_type>::value>> xfunction(Func &&f, CTA&&... e)¶

Constructs an xfunction applying the specified function to the given arguments.

Parameters

f: the function to apply
e: the xexpression arguments

Size and shape

auto dimension() const¶: Returns the number of dimensions of the function.

auto shape() const¶: Returns the shape of the xfunction.

layout_type layout() const¶: Returns the layout_type of the xfunction.

Data

template<class ...Args> auto operator()(Args... args) const¶

Returns a constant reference to the element at the specified position in the function.

Parameters

args: a list of indices specifying the position in the function. Indices must be unsigned integers, the number of indices should be equal or greater than the number of dimensions of the function.

template<class ...Args> auto unchecked(Args... args) const¶

Returns a constant reference to the element at the specified position in the expression.

Warning

This method is meant for performance, for expressions with a dynamic number of dimensions (i.e. not known at compile time). Since it may have undefined behavior (see parameters), operator() should be preferred whenever it is possible.

Warning

This method is NOT compatible with broadcasting, meaning the following code has undefined behavior:

xt::xarray<double> a = {{0, 1}, {2, 3}};
xt::xarray<double> b = {0, 1};
auto fd = a + b;
double res = fd.unchecked(0, 1);

Parameters

args: a list of indices specifying the position in the expression. Indices must be unsigned integers, the number of indices must be equal to the number of dimensions of the expression, else the behavior is undefined.

template<class It> auto element(It first, It last) const¶

Returns a constant reference to the element at the specified position in the function.

Parameters

first: iterator starting the sequence of indices
last: iterator ending the sequence of indices The number of indices in the sequence should be equal to or greater than the number of dimensions of the container.

Broadcasting

template<class S> bool broadcast_shape(S &shape, bool reuse_cache = false) const¶

Broadcast the shape of the function to the specified parameter.

Return

a boolean indicating whether the broadcasting is trivial

Parameters

shape: the result shape
reuse_cache: boolean for reusing a previously computed shape

template<class S> bool has_linear_assign(const S &strides) const¶

Checks whether the xfunction can be linearly assigned to an expression with the specified strides.

Return: a boolean indicating whether a linear assign is possible

Defined in xtensor/xmath.hpp

template<class F, class ...E> auto xt::make_lambda_xfunction(F &&lambda, E&&... args)¶

Create a xfunction from a lambda.

This function can be used to easily create performant xfunctions from lambdas:

template <class E1>
inline auto square(E1&& e1) noexcept
{
    auto fnct = [](auto x) -> decltype(x * x) {
        return x * x;
    };
    return make_lambda_xfunction(std::move(fnct), std::forward<E1>(e1));
}

Lambda function allow the reusal of a single arguments in multiple places (otherwise only correctly possible when using xshared_expressions). auto lambda functions are automatically vectorized with xsimd if possible (note that the trailing -> decltype(...) is mandatory for the feature detection to work).

Return

lazy xfunction

Parameters

lambda: the lambda to be vectorized
args: forwarded arguments