Method run_op

src/cpu/kernels/CpuConcatenateHeightKernel.cpp:91–189

}

void CpuConcatenateHeightKernel::run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info)
{
    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuConcatenateHeightKernel::run_op");
    ARM_COMPUTE_UNUSED(info);
    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICpuKernel::window(), window);

    const auto src = tensors.get_const_tensor(TensorType::ACL_SRC);
    auto       dst = tensors.get_tensor(TensorType::ACL_DST);

    // Offset destination pointer to the correct position
    uint8_t *dst_ptr = dst->buffer() + dst->info()->offset_first_element_in_bytes() +
                       _height_offset * dst->info()->strides_in_bytes()[Window::DimY];

    const auto window_start_x = static_cast<int>(window.x().start());
    const auto window_end_x   = static_cast<int>(window.x().end()) * static_cast<int>(dst->info()->element_size());
    const int  window_step_x  = 16;

    Window win{window};
    win.set(Window::DimX, Window::Dimension(0, 1, 1));
    win.set(Window::DimY, Window::Dimension(0, src->info()->tensor_shape().y(), 1));

    // Create iterators
    Iterator src_it(src, win);
    Iterator dst_it(dst, win);

    const DataType                 dt        = src->info()->data_type();
    const UniformQuantizationInfo &src_qinfo = src->info()->quantization_info().uniform();
    const UniformQuantizationInfo &dst_qinfo = dst->info()->quantization_info().uniform();
    if (dt == DataType::QASYMM8 && src_qinfo != dst_qinfo)
    {
        execute_window_loop(
            win,
            [&](const Coordinates &)
            {
                int x = window_start_x;
                for (; x <= (window_end_x - window_step_x); x += window_step_x)
                {
                    vst1q_u8(dst_ptr + dst_it.offset() + x,
                             vquantize(vdequantize(vld1q_u8(src_it.ptr() + x), src_qinfo), dst_qinfo));
                }

                // Compute left-over elements
                for (; x < window_end_x; ++x)
                {
                    *(dst_ptr + dst_it.offset() + x) =
                        quantize_qasymm8(dequantize_qasymm8(*(src_it.ptr() + x), src_qinfo), dst_qinfo);
                }
            },
            src_it, dst_it);
    }
    else if (dt == DataType::QASYMM8_SIGNED && src_qinfo != dst_qinfo)
    {
        execute_window_loop(
            win,
            [&](const Coordinates &)
            {
                int x = window_start_x;