The throughput and latency measured using Vitis Analyzer are shown to be 144 ns and 443.2 ns, in the following figures. This throughput level is equivalent to 32/144e-9 = 222 Msps. Note that a higher QoR is achieved by the Vitis DSP library IP.
The same “batch processing” approach to improving throughput is supported by the Vitis DSP library. Set the TP_WINDOW_SIZE parameter to hold multiple transforms and reduce the overhead. The previous code block is set up to do this using the REPEAT parameter. When set to REPEAT=128, a throughput and latency of 11.14 μs and 22.29 μs are achieved, respectively. This throughput level is equivalent to 128*32/11.14e-6 = 367 Msps. These results are summarized and compared to the earlier design in the table above.