Method configure

src/gpu/cl/kernels/ClCopyKernel.cpp:86–142

}

void ClCopyKernel::configure(const CLCompileContext &compile_context,
                             const ITensorInfo      *src,
                             ITensorInfo            *dst,
                             Window                 *dst_window)
{
    ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst, dst_window));

    auto padding_info = get_padding_info({src, dst});

    // Create kernel
    CLBuildOptions build_opts;
    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(src->data_type()));

    // Output auto inizialitation if not yet initialized
    auto_init_if_empty(*dst, *src);

    // Configure window
    const unsigned int vec_size_x = adjust_vec_size(16 / src->element_size(), src->dimension(0));

    const Window win_config = calculate_max_window(*src, Steps(vec_size_x));

    if (dst_window != nullptr)
    {
        _has_dst_window                = true;
        _dst_window                    = Window(*dst_window);
        const int  width_x             = dst_window->num_iterations(0);
        const int  vec_size_x_leftover = width_x % vec_size_x;
        const bool multi_access_x      = width_x >= static_cast<int32_t>(vec_size_x);

        if (multi_access_x)
        {
            _dst_window.set(Window::DimX,
                            Window::Dimension(dst_window->x().start(),
                                              ceil_to_multiple(dst_window->x().end(), vec_size_x), vec_size_x));
        }

        build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(vec_size_x_leftover));
    }
    else
    {
        const int width_x             = src->tensor_shape().x();
        const int vec_size_x_leftover = width_x % vec_size_x;

        build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(vec_size_x_leftover));
    }

    build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(vec_size_x));

    // Build kernel
    _kernel = create_kernel(compile_context, "copy_tensor", build_opts.options());

    // Validate and set the window
    ICLKernel::configure_internal(win_config);

    ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}