RFID-based item localization for large-scale warehousing and inventorying applications has been gradually developing over the years. In these settings, fast and accurate item localization with low deployment cost is vital for inventory visibility and warehouse efficiency. However, existing tag positioning methods based on a single antenna offer either fast inference with low accuracy or high accuracy with slow inference. In this paper, we propose a phase-based SAR tag positioning method that offers both merits. Firstly, the traditional holograms formed in existing SAR methods are completely replaced by a one-step version of the recently proposed hologram mask (Angle-of-Arrival hologram), to accelerate the tag positioning task. Secondly, a pair of hashtables is implemented to output directly hologram grids from phase measurements, reducing further the hologram construction time. Thirdly, a pre-processing cross-validation stage tests the measurements at each operating frequency to determine the one leading to the lowest positioning error. After this, the model is trained and tested solely on this frequency, limiting both the inference time and on-site collection time. Finally, we efficiently infer the height information of tags from a single antenna moving along a rectilinear horizontal trajectory. Our comparative experiments show that our model achieves similar positioning accuracy to the state-of-the-art accuracy methods while offering a much lower inference time, in both short-range and long-range settings.