Method run_op

src/gpu/cl/kernels/ClWinogradInputTransformKernel.cpp:274–327

}

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

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

    const size_t idx_w         = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::WIDTH);
    const size_t idx_h         = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::HEIGHT);
    const size_t idx_c         = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::CHANNEL);
    const size_t total_batches = window.shape().total_size_upper(3);

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

    if (_data_layout == DataLayout::NHWC)
    {
        Window slice = window_collapsed.first_slice_window_3D();
        slice.set(1, Window::Dimension(0, _num_tiles_x * _num_tiles_y, 1));
        slice.set(2, Window::Dimension(0, total_batches, 1));

        unsigned int idx = 0;
        add_4D_tensor_argument(idx, src, slice);
        add_4D_tensor_argument(idx, dst, slice);
        _kernel.setArg<cl_uint>(idx++, _src_width);
        _kernel.setArg<cl_uint>(idx++, _src_height);
        _kernel.setArg<cl_uint>(idx++, _num_tiles_x);
        _kernel.setArg<cl_uint>(idx++, _num_tiles_y);
        enqueue(queue, *this, slice, lws_hint());
    }
    else
    {
        Window slice = window_collapsed.first_slice_window_3D();
        slice.set(idx_w, Window::Dimension(0, _num_tiles_x, 1));
        slice.set(idx_h, Window::Dimension(0, _num_tiles_y, 1));

        ARM_COMPUTE_ERROR_ON(((slice[idx_c].end() - slice[idx_c].start()) % _step_z) != 0);
        slice.set(idx_c, Window::Dimension(slice[idx_c].start(), slice[idx_c].end(), _step_z));

        unsigned int idx = 2 * num_arguments_per_3D_tensor();
        _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(src->info()->strides_in_bytes()[3]));
        _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(dst->info()->strides_in_bytes()[3]));

        do
        {
            unsigned int idx = 0;
            add_3D_tensor_argument(idx, src, slice);
            add_3D_tensor_argument(idx, dst, slice);

            enqueue(queue, *this, slice, lws_hint());
        } while (window_collapsed.slide_window_slice_3D(slice));
    }
}
} // namespace kernels
} // namespace opencl
} // namespace arm_compute