Python-based Portable IP-core Synthesis Framework for FPGA-based Computing
Copyright (C) 2013, Shinya Takamaeda-Yamazaki
E-mail: takamaeda_at_ist.hokudai.ac.jp
Apache License 2.0 (http://www.apache.org/licenses/LICENSE-2.0)
If you use PyCoRAM in your research, please cite our paper.
- Shinya Takamaeda-Yamazaki, Kenji Kise and James C. Hoe: PyCoRAM: Yet Another Implementation of CoRAM Memory Architecture for Modern FPGA-based Computing, The Third Workshop on the Intersections of Computer Architecture and Reconfigurable Logic (CARL 2013) (Co-located with MICRO-46), December 2013. Paper Slide
@inproceedings{Takamaeda:2013:CARL:PyCoRAM,
author = {Takamaeda-Yamazaki, Shinya and Kise, Kenji and Hoe, James C.},
title = {{PyCoRAM: Yet Another Implementation of CoRAM Memory Architecture for Modern FPGA-based Computing}},
booktitle={Intersections of Computer Architecture and Reconfigurable Logic (CARL 2013)},
month={Dec},
year = {2013},
location = {Davis, CA},
}
-
Zynq + PyCoRAM (+ Debian) (slideshare, in Japanese) Slide
-
PyCoRAM for HLS meet up (slideshare, in Japanese) Slide
PyCoRAM is a Python-based portable IP-core synthesis framework with CoRAM (Connected RAM) memory architecture.
PyCoRAM framework generates a portable IP-core package from computing logic descriptions in Verilog HDL and memory access pattern descriptions in Python. Designers can easily build an FPGA-based custom accelerator using a generated IP-core with any common IP-cores on vendor-provided EDA tools. PyCoRAM framework includes (1) the Verilog-to-Verilog design translation compiler and (2) the Python-to-Verilog high-level synthesis (HLS) compiler for generating control circuits of memory operations.
There are some major differences between PyCoRAM and the original soft-logic implementation of CoRAM.
- Memory access pattern representation in Python
- The original CoRAM uses C language to represent a memory access pattern (called 'control thread').
- In PyCoRAM, you can easily describe them by using popular lightweight scripting language.
- A Python script of memory access patterns is translated into an RT-level hardware design in Verilog HDL by the Python-to-Verilog high-level synthesis compiler.
- Commercial interconnect supports (AMBA AXI4 and Altera Avalon)
- The original CoRAM uses CONNECT to generate an on-chip interconnect.
- PyCoRAM compiler generates a IP-core design with AMBA AXI4 or Altera Avalon. Both are commonly used on vendor-provided EDA tools.
- Parameterized RTL Design Support
- The original CoRAM has some limitations in Verilog HDL description of computing logic, such as no supports of generate statement.
- PyCoRAM has a sophisticated RTL analyzer/translator to convert RTL descriptions into synthesizable IP-core package under memory abstractions of CoRAM.
- Python: 2.7, 3.4 or later
Python3 is recommended.
- Icarus Verilog: 0.9.7 or later
Install on your platform. For exmple, on Ubuntu:
sudo apt-get install iverilog
- Jinja2: 2.8 or later
Install on your python environment by using pip.
pip install jinja2
- Pyverilog: 1.0.0 or later
Install from pip:
pip install pyverilog
Install PyCoRAM.
python setup.py install
Dockerfile is available, so that you can try PyCoRAM on Docker without any installation on your host platform.
cd docker
sudo docker build -t user/pycoram .
sudo docker run --name pycoram -i -t user/pycoram /bin/bash
cd PyCoRAM/tests/single_memory/
make build
make sim
You can use the pycoram command from your console.
pycoram
You can find some examples in 'PyCoRAM/examples/' and 'PyCoRAM/tests'.
Let's begin PyCoRAM by an example in 'tests/single_memory'. You will find two source files.
- ctrl_thread.py : Control-thread definition in Python
- userlogic.v : User-defined Verilog code using CoRAM memory blocks
Type 'make' to build a PyCoRAM IP-core from the source files. Then type 'make run' to simulate sample system.
make build
make sim
Instead, you can type commands as below directly at 'PyCoRAM' directory.
pycoram default.config -t userlogic -I include tests/single_memory/ctrl_thread.py tests/single_memory/userlogic.v
iverilog -I pycoram_userlogic_v1_00_a/hdl/verilog/ pycoram_userlogic_v1_00_a/test/test_pycoram_userlogic.v
./a.out
PyCoRAM compiler generates a directory for IP-core (pycoram_userlogic_v1_00_a, in this example).
'pycoram_userlogic_v1_00_a.v' includes
- IP-core RTL design (hdl/verilog/pycoram_userlogic.v)
- Test bench (test/test_pycoram_userlogic.v)
- XPS setting files (pycoram_userlogic_v2_1_0.{mpd,pao,tcl})
- IP-XACT file (component.xml)
A bit-stream can be synthesized by using Xilinx Platform Studio, Xilinx Vivado, and Altera Qsys. In case of XPS, please copy the generated IP-core into 'pcores' directory of XPS project.
This project has some examples in 'PyCoRAM/examples/' and 'PyCoRAM/tests'. To build them, please modify 'Makefile', so that the corresponding files and parameters are selected (especially INPUT, MEMIMG and USERTEST).
pycoram [config] [-t topmodule] [-I includepath]+ [--memimg=filename] [--usertest=filename] [file]+
- file
- User-logic Verilog file (.v) and control-thread definition file (.py). Automatically, .v file is recognized as a user-logic Verilog file, and .py file recongnized as a control-thread definition, respectively.
- config
- Configuration file which includes memory and device specification
- -t
- Name of user-defined top module, default is "userlogic".
- -I
- Include path for input Verilog HDL files.
- --memimg
- DRAM image file in HEX DRAM (option, if you need). The file is copied into test directory. If no file is assigned, the array is initialized with incremental values.
- --usertest
- User-defined test code file (option, if you need). The code is copied into testbench script.
- Python-based Hardware Design Processing Toolkit for Verilog HDL
- Used as basic code analyser and generator in PyCoRAM
- A General Purpose Memory Architecture for FPGAs
- The original CoRAM developed at CMU