apply_per_channel()

template<ComputeData InputType = std::vector<Kakshya::DataVariant>, ComputeData OutputType = InputType>

template<typename Func >

output_type MayaFlux::Yantra::ConvolutionTransformer< InputType, OutputType >::apply_per_channel ( input_type &  input, Func &&  func  )

inlineprivate

Extracts per-channel spans, applies func to each, and reconstructs.

Template Parameters

Func	Callable matching either void(std::span<double>) or std::vector<double>(std::span<const double>)

In-place: results are copied back into the original channel spans of input. Out-of-place: input is not mutated.

Definition at line 172 of file ConvolutionTransformer.hpp.

    {
        auto [channels, structure_info] = OperationHelper::extract_structured_double(input);
        m_working_buffer.resize(channels.size());
        for (size_t i = 0; i < channels.size(); ++i) {
            if constexpr (std::is_void_v<std::invoke_result_t<Func, std::span<double>>>) {
                m_working_buffer[i].assign(channels[i].begin(), channels[i].end());
                func(std::span<double> { m_working_buffer[i] });
            } else {
                m_working_buffer[i] = func(std::span<const double> { channels[i] });
            }
        }
        if (this->is_in_place())
            for (size_t i = 0; i < channels.size(); ++i)
                std::ranges::copy(m_working_buffer[i], channels[i].begin());
        return create_output(
            OperationHelper::reconstruct_from_double<InputType>(m_working_buffer, structure_info));
    }