Function copy_tensor

dlib/cuda/cpu_dlib.cpp:3071–3108

     // ------------------------------------------------------------------------------------

        void copy_tensor(
            bool add_to,
            tensor& dest,
            size_t dest_k_offset,
            const tensor& src,
            size_t src_k_offset,
            size_t count_k
        )
        {
            const size_t dest_sample_size = static_cast<size_t>(dest.nc() * dest.nr() * dest.k());
            const size_t src_sample_size = static_cast<size_t>(src.nc() * src.nr() * src.k());

            const size_t block_size = count_k * dest.nc() * dest.nr();

            DLIB_CASSERT(dest.num_samples() == src.num_samples() &&
                dest.nc() == src.nc() && dest.nr() == src.nr(), "All sources should fit into dest tensor size");
            DLIB_CASSERT(dest.k() - dest_k_offset >= count_k, "Not enough space in dest tensor");
            DLIB_CASSERT(src.k() - src_k_offset >= count_k, "Not enough space in src tensor");

            float* dest_p = dest.host() + dest_k_offset * dest.nc() * dest.nr();
            const float* src_p = src.host() + src_k_offset * src.nc() * src.nr();

            for (long i = 0; i < src.num_samples(); ++i)
            {
                if (add_to)
                {
                    for (size_t j = 0; j < block_size; ++j)
                        dest_p[j] += src_p[j];
                }
                else
                {
                    ::memcpy(dest_p, src_p, block_size * sizeof(float));
                }

                dest_p += dest_sample_size;
                src_p  += src_sample_size;
            }
        }

    // ------------------------------------------------------------------------------------