Skip to main content

Framework for functional hardware design approach

Project description

Welcome to PyGears

HW Design: A Functional Approach

PyGears is an ambitious attempt to create a Python framework that facilitates describing digital hardware. It aims to augment current RTL methodology to drastically increase composability of hardware modules. Ease of composition leads to better reusability, since modules that compose better can be used in a wider variety of contexts. Set of reusable components can then form a well-tested and documented library that significantly speeds up the development process.

For a guide through PyGears methodology, checkout blog series on implementing RISC-V in PyGears.

For an introductory PyGears example, checkout echo. A snippet is given below:

@gear
def echo(samples: Int, *, fifo_depth, feedback_gain, precision):
    dout = Intf(din.dtype)

    feedback = dout \
        | fifo(depth=fifo_depth, threshold=fifo_depth - 1) \
        | fill_void(fill=Int[16](0)) \
        | decoupler

    feedback_attenuated = (feedback * feedback_gain) >> precision

    dout |= (din + feedback_attenuated) | dout.dtype

    return dout

PyGears proposes a single generic interface for all modules (read about the hardware implementation of the interface here) and provides a way to use powerful features of Python language to compose modules written in an existing HDL (currently only supports SystemVerilog). Based on the Python description, PyGears generates functionally equivalent, synthetizable RTL code.

Furthermore, PyGears offers a way to write verification environment in high-level Python language and co-simulate the generated RTL with an external HDL simulator. PyGears features a completely free solution using Verilator simulator and standard SystemVerilog simulators via the DPI (tested on proprietary Questa and NCSim simulators).

PyGears also features a library of standard modules and the cookbook library that are ready to be used in a PyGears design.

In PyGears, each HDL module is considered a Python function, called the gear, hence the design is described in form of a functional (gear) composition. In order for HDL modules to be composable in this way, they need to be designed in accordance with the Gears methodology. You should probably read a short intro to Gears in order to understand this project from the hardware perspective.

PyGears supports also the hierarchical gears which do not have a HDL implementation, but are defined in terms of other gears. Each gear accepts and returns interface objects as arguments, which represents module connections. This allows for a module composition to be described in terms of powerful functional concepts, such as: partial application, higher-order functions, function polymorphism.

PyGears features a powerful system of generic types, which allows for generic modules to be described, as well as to perform type checking of the gear composition.

Installation Instructions

For the instruction checkout Installation page.

Read the documentation

PyGears documentation

Checkout the examples

Echo: Hardware module that applies echo audio effect to a continuous audio stream.

RISC-V processor: PyGears implementation. Checkout also the RISC-V implementation blog series.

Tests: Contain many examples on how individual PyGears components operate

Library of standard modules

Cookbook library

References

  • Kortiq’s AIScale Deep Learning Processor was completely developed using PyGears

Contributions

Special thanks to the people that helped develop this framework:

  • Andrea Erdeljan

  • Damjan Rakanović

  • Nemanja Kajtez

  • Risto Pejašinović

  • Stefan Tambur

  • Vladimir Nikić

  • Vladimir Vrbaški

In order to contribute, pull your copy from github repository and create a pull request.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

pygears-0.1.2.tar.gz (109.0 kB view hashes)

Uploaded Source

Built Distribution

pygears-0.1.2-py3-none-any.whl (209.4 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page