PYNQ HLS Moving Average

This repository contains an HLS design of a moving average IP for PYNQ-based systems.

SETUP

We relied on Vivado HLS 2019.2 to perform the High-Level Synthesis step, Vivado 2019.2 to generate the bitstream, and Pynq 2.5 to evaluate the IP. In particular, we tested our IP on three Zynq-based platforms, namely Ultra96-V2, Pynq Z2, and ZCU104 .

HLS

The HLS folder contains the IP HLS files moving_average.cpp, moving_average.hpp, and a testbench testbench.cpp. The current version of the IP receives an input stream of integer values and produces an output stream of float averages. The first two elements of the input stream represent the number of input elements and the window of elements to average. Currently, the IP supports moving averages from 1 to 16 elements. You can modify this parameter within the moving_average.hpp file.

Vivado Design

After exporting the IP from Vivado HLS, run Vivado and create an RTL project for your target board. Then, create a block design and add the following IPs:

• Zynq IP (Zynq 7000 or Zynq Ultrascale+ MPSoC according to your board)
• AXI DMA
• Moving Average IP

Enable the High Performance (HP) port on the Zynq IP, disable the Scatter Gather Engine on the AXI DMA. We also suggest to set the AXI DMA Buffer Length Register to the maximum value. Run the available automations and, finally, connect the Moving Average IP to the AXI DMA and to the clock and reset wires. You can also create a hierarchy containing both the AXI DMA and the Moving Average IP. At this point, validate the design, create the HDL wrapper and run the bitstream generation.
Once the bitstream is ready, deploy the following files to your target board:

• bitstream file: projectFolder/projectName.runs/impl_1/design_1_wrapper.bit
• Block design file: projectFolder/projectName.srcs/sources_1/bd/design_1/hw_handoff/design_1_bd.tcl
• hwh file: projectFolder/projectName.srcs/sources_1/bd/design_1/hw_handoff/design_1.hwh

Remember that these files should have the same name and that you cannot change the name of the bitstream file.

PYNQ evaluation

Python folder contains two Jupyter notebooks to evaluate the Moving Average IP: moving_average.ipynb and moving_average_with_class.ipynb. These files evaluate the IP similar to the HLS testbench using PYNQ APIs. moving_average_with_class.ipynb wraps the IP usage within a Python class.