Method run_op

src/gpu/cl/kernels/ClWinogradOutputTransformKernel.cpp:317–362

}

void ClWinogradOutputTransformKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
{
    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(IClKernel::window(), window);

    auto src  = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_0));
    auto bias = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_1));
    auto dst  = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));

    // Collapse window
    Window window_collapsed = window.collapse_if_possible(IClKernel::window(), Window::DimZ);

    // Get initial windows
    Window slice = window_collapsed.first_slice_window_4D();
    slice.set(Window::DimZ, Window::Dimension(0, 1, 1));

    // Setup output slice
    Window slice_out(slice);
    slice_out.set(Window::DimX, Window::Dimension(0, 0, 0));
    slice_out.set(Window::DimY, Window::Dimension(0, 0, 0));

    if (bias != nullptr)
    {
        unsigned int idx1 = 2 * num_arguments_per_4D_tensor();
        Window       slice_biases;
        slice_biases.use_tensor_dimensions(bias->info()->tensor_shape());
        add_1D_tensor_argument(idx1, bias, slice_biases);
    }

    if (_is_nhwc)
    {
        unsigned int idx2 = 2 * num_arguments_per_4D_tensor() + ((bias != nullptr) ? num_arguments_per_1D_tensor() : 0);
        _kernel.setArg(idx2++, static_cast<int>(dst->info()->total_size() - dst->info()->strides_in_bytes().y()));
        _kernel.setArg<cl_int>(idx2++, _src_height);
        _kernel.setArg<cl_int>(idx2++, _dst_width);
        _kernel.setArg<cl_int>(idx2++, _dst_height);
    }

    do
    {
        unsigned int idx = 0;
        add_4D_tensor_argument(idx, src, slice);
        add_4D_tensor_argument(idx, dst, slice_out);
        enqueue(queue, *this, slice, lws_hint());
    } while (window.slide_window_slice_3D(slice) && window.slide_window_slice_3D(slice_out));
}
} // namespace kernels
} // namespace opencl
} // namespace arm_compute