Wrap any callable void(const Vector&, Vector&) as a LinearOp. More...

#include <subspace.hpp>

Inheritance diagram for num::kernel::subspace::CallableOp< F >:

Public Member Functions
	CallableOp (F f, idx n)

void	apply (const Vector &x, Vector &y) const override
	y = A*x (y must be pre-allocated to the correct size)

idx	rows () const noexcept override

idx	cols () const noexcept override

Public Member Functions inherited from num::kernel::subspace::LinearOp
virtual	~LinearOp ()=default

Detailed Description

template<typename F>
struct num::kernel::subspace::CallableOp< F >

Wrap any callable void(const Vector&, Vector&) as a LinearOp.

Useful for structured operators – stencils, FFT-based, implicit matrices – without forming an explicit matrix. The functor is stored by value.

apply() pre-allocates y to the correct size (n_) before calling f_ so that callers (e.g. arnoldi_step passing a pre-allocated scratch buffer) never trigger a heap allocation inside the hot Arnoldi loop. Lambdas passed to make_op / make_op_ptr can safely assume y.size() == n on entry.

auto op = num::kernel::subspace::make_op(

[&](const Vector& x, Vector& y) { laplacian(x, y, N); }, N * N);

num::BasicVector< real >

num::kernel::subspace::make_op

CallableOp< F > make_op(F f, idx n)

Factory: wrap a callable as a stack-allocated CallableOp<F>.

Definition subspace.hpp:128

Definition at line 108 of file subspace.hpp.