Method put_impl

imperative/src/impl/interpreter/interpreter_impl.cpp:149–180

}

TensorInfo* ChannelImpl::put_impl(const HostTensorND& value, bool no_cache) {
    if (value.empty()) {
        auto layout = value.layout();
        layout.init_contiguous_stride();
        const_cast<HostTensorND&>(value).reset(value.storage(), layout);
    }
    auto info = alloc();
    constexpr int size_threshold = TensorShape::MAX_NDIM;
    init(info, {value.layout(), value.comp_node()});
    if (value.layout().total_nr_elems() <= size_threshold) {
        info->h_value = value;
        info->desc.value = value.proxy_to_default_cpu();
    }
    if (Profiler::is_profiling()) {
        m_worker.add_task(
                {Profiler::next_id(), Put{info, value, no_cache},
                 get_channel_state().stack_manager.dump()});
    } else {
        m_worker.add_task({
                Profiler::next_id(),
                Put{info, value, no_cache},
        });
    }

    if (get_channel_state().options.async_level == 0) {
        sync_impl();
        info->desc.comp_node.sync();
        auto err = info->desc.comp_node.check_async_error();
        mgb_assert(!err, "%s", err->what());
    }
    return info;
}

Handle ChannelImpl::put(const DeviceTensorND& data, const HostTensorND& hvalue) {
    MGB_LOCK_GUARD(m_spin);