GithubHelp home page GithubHelp logo

tempname993 / flowtune Goto Github PK

View Code? Open in Web Editor NEW

This project forked from yu-maryland/flowtune

0.0 0.0 0.0 190.92 MB

Shell 0.11% C++ 1.43% Perl 0.31% C 96.06% Tcl 0.01% Verilog 1.54% Makefile 0.15% HTML 0.38% CMake 0.01%

flowtune's Introduction

FlowTune

FlowTune Overview

What FlowTune does?

  • Efficiently explore ABC synthesis flows (sequential decision making)

  • Push-button tool/script for various QoR metrics are included

    • #AIG minimization (Boolean network minimization)
    • #LUT minimization (ABC LUT mapper "if -K")
    • STD Technology mapping Delay and Area optimization
    • STA-Aware STD Technology mapping Delay optimization
    • SAT-CNF minimization (#clauses)

What is included in this repo?

  • FlowTune framework (implemented in ABC; Thanks @Alan Mishchenko)

  • A set of seqeuntial FPU benchmarks in BLIF (original benchmarks from VTR 8.0 Thanks!@VTR Team)

  • Push-button FlowTune for Boolean Networks, LUT (FPGA mapping), STD (ASIC) mapping optimizations.

  • FlowTune full integration with VTR 8.0 (see bash script install.sh). End-to-end evaluation of FlowTune up to post-PnR stage is enabled! (Thanks @VTR team).

  • FlowTune full integration with Yosys (Thanks! @Yosys Team)

  • FlowTune -> Vivado demo and FlowTune -> Cadence Genus demo are included.

Reference

@inproceedings{yu-mab-iccad2020,
title={Practical Multi-armed Bandits in Boolean Optimization},
author={Cunxi Yu},
booktitle={2020 International Conference On Computer Aided Design (ICCAD'20)},
year={2020},
}

Required Packages:

- readline: sudo apt-get install libreadline6 libreadline6-dev (Ubuntu)
- readline: sudo yum install readline-devel (CentOS 7)
- -stdc++11 (tested on gcc4.9 and gcc7.0) 
- OpenMP (https://www.openmp.org/resources/openmp-compilers-tools/ version tested 4.0.1 -- https://download.open-mpi.org/release/open-mpi/v4.0/openmpi-4.0.1.tar.gz)
	- wget https://download.open-mpi.org/release/open-mpi/v4.0/openmpi-4.0.1.tar.gz; tar -xvf openmpi-4.0.1.tar.gz
	- cd openmpi-4.0.1; ./configure; make -j12 all install
- Yosys -- https://github.com/YosysHQ/yosys
- export "abc" and "yosys" PATH TO .bashrc to have a global access
	- Instruction:  echo "export PATH=$(pwd):\${PATH}" >> ~/.bashrc; source ~/.bashrc
	- Testing : If PATH added succesfully, you should be able to type "abc" and "yosys" at any location of your LINUX system 
- VTR 8.0 -- https://github.com/verilog-to-routing/vtr-verilog-to-routing

Installation Demo on Ubuntu 20.x.x

Installation bash script

This is an installation script (tested) for Ubuntu system to setup FlowTune+VTR8.0

# NOTE: This is an installation script (tested) for Ubuntu system
# git clone https://github.com/Yu-Utah/FlowTune;

# NOTE: Let's first install the required packages
# 	You can skip these if you have the packages in your machine, or use different methods to install these packages (e.g., if you use RedHat/CentOS, you will need to use "yum")

# sudo apt-get install libreadline-dev g++ make flex bison libgtk-3-dev;
# wget https://download.open-mpi.org/release/open-mpi/v4.0/openmpi-4.0.1.tar.gz; tar -xvf openmpi-4.0.1.tar.gz
# cd openmpi-4.0.1; ./configure; make -j12 all install;
# you can skip this OpenMP installation if you have OpenMP library
# you can check whether you have OpenMP or the version using "echo |cpp -fopenmp -dM |grep -i open"
# Ubuntu 20.x.x includes OpenMP automatically


if ! [ -x "$(command -v cmake3)" ] && ! [ -x "$(command -v cmake)" ];
then
    echo "cmake and cmake3 could not be found"
    exit
fi

# NOTE: Let's compile FlowTune with ABC
cd src/; mkdir build; cd build; cmake3 ..;
abc_dir_path="my \$abc_dir_path = \"$(pwd)\"" 
make -j12; export PATH=$PATH:$(pwd)

# NOTE: A quick test see if FlowTune has been installed correctly
cd ../../FlowTune-AIG-Optimization/
./single_design.sh adder2 adder2.blif 1 1 0 1 1

# NOTE: Now let's setup FlowTune + VTR 8.0
cd ../; pwd
git clone https://github.com/verilog-to-routing/vtr-verilog-to-routing;
cd vtr-verilog-to-routing; make -j12;
cd vtr_flow;
vtr_flow_pth="my \$vtr_flow_path = \"$(pwd)\""
cd scripts/perl_libs/XML-TreePP-0.41/lib;
vtr_flow_xml_lib_path="$(pwd)"
cd ../../../../;
# follow VTR installation instruction to install other packages if you cannot compile VTR
echo $vtr_flow_pth # check your vtr flow path
echo $abc_dir_path # check your flowtune-abc path

cd ../../FlowTune-VTR-Flow;ls ftune_vtr_flow.pl
echo "IMPORTANT !! - You need to change ftune_vtr_flow.pl from line 56 - 60 accordingly"
echo "1) change the use lib path using <$vtr_flow_xml_lib_path>"
echo "2) change the vtr_flow_path using <$vtr_flow_pth>"
echo "3) change abc_dir_path using <$abc_dir_path>"

# NOTE: Setup the ftune_vtr_flow.pl
# IMPORTANT !! - You need to change ftune_vtr_flow.pl from line 56 - 60 according to your VTR installation
# 1) change the use lib path using "$vtr_flow_xml_lib_path"
# 2) change the vtr_flow_path using "$vtr_flow_pth"
# 3) change abc_dir_path using "$abc_dir_path"

# To run FlowTune with VTR, just run "./ftune_vtr_flow.pl <circuit_file> <architecture_file>"

FlowTune implementation in ABC

abc 01> ftune -h
usage: ftune [drtfpihSL]
	flowtune(ftune): Multi-Armed Bandit (MAB) synthesis flow tuning for Logic Minimization
-t    : Targeted metric (default = 0, i.e., AIG Minization targeting number of AIG nodes)
      : t=0 AIG Minization - Minimizing AIG nodes                       :  ftune -d i10.aig -r 4 -t 0 -p 1 -i 10 -s 5 -L [other options]
      : t=1 AIG Minization - minimizing AIG levels                      :  ftune -d i10.aig -r 4 -t 1 -p 1 -i 10 -s 5 -L [other options]
      : t=2 Technology mapping (w Gate Sizing + STA) Min STA-Delay      :  ftune -d i10.aig -r 4 -t 2 -p 1 -i 10 -s 5 -L your.lib(Liberty) [other options]
      : t=3 Technology mapping (w Gate Sizing + STA) Min Area           :  ftune -d i10.aig -r 4 -t 3 -p 1 -i 10 -s 5 -L your.lib(Liberty) [other options]
      : t=4 FPGA Mapping - Miniziming Number of 6-input LUTs            :  ftune -d i10.aig -r 4 -t 4 -p 1 -i 10 -s 5 [other options]
      : t=5 FPGA Mapping - Miniziming Levels of 6-input LUT network     :  ftune -d i10.aig -r 4 -t 5 -p 1 -i 10 -s 5 [other options]
      : t=6 SAT (CNF) Minization - Miniziming Number of Clauses         :  ftune -d cnf.aig -r 4 -t 6 -p 1 -i 10 -s 5 [other options]
      : t=7 SAT (CNF) Minization - Miniziming Number of literals        :  ftune -d cnf.aig -r 4 -t 7 -p 1 -i 10 -s 5 [other options]
      : t=8 Regular Technology mapping using GENLIB (map) Min Delay     :  ftune -d i10.aig -r 4 -t 8 -p 1 -i 10 -s 5 -L your.genlib(GENLIB) [other options]
      : t=9 Regular Technology mapping using GENLIB (map) Min Area      :  ftune -d i10.aig -r 4 -t 9 -p 1 -i 10 -s 5 -L your.genlib(GENLIB) [other options]
-h    : Print the command usage
-r    : Number of appearances of each synthesis command (default = 3)
-p    : Factor of number of Arms (default = 1)
-i    : Number of MAB iterations (default = 10)
-s    : Number of MAB sampling for each iteration per Arm (default = 5)
-f    : Toggle using long-short-term memory for probability matching (default=FALSE)
-C    : Customize the logic transformations (commands) for tuning instead of default commands. 
	 Example 1:  -C b;rw;rwz;rf (your flow will use these 4 opts).
		
	 Example 2: -C b;rw;rwz;rf;your_cmd (You can define your new command in abc.rc namely your_cmd and tunes for these 4 opts + your_cmd).
-S    : Toggle using Softmax() or LogSoftmax() for finalizing samples at each iteration (default=FALSE)
	    -S 0 : Using winning rate Pr
	    -S 1 : Using Softmax()
	    -S 2 : Using LogSoftmax()
-L    : Liberty or GENLIB file. Required for Technology mapping tuning (t=2,3,8,9) 

FlowTune Demos

(a) -- AIG Node Minimization

Design: bfly.v
Baseline: resyn (default abc flow) **applied 25 times consecutively**
Result: FlowTune provides 20%+ more AIG reduction compared to resyn*25 
(see video below)

Baseline

bfly-baseline

FlowTune (20% more AIG reductions compared to resyn applied 25 times consecutively)

bfly-ftune

(b) -- VTR (vtr-verilog-to-routing) Integration

See ./FlowTune-Integration-VTR/ftune_vtr_flow.pl for details

Comparisons between default VTR vs FTune VTR

Design: bfly.v
FPGA Architecture = k6_frac_N10_frac_chain_mem32K_40nm
Result: 15% area/block reductions at Post-PnR stage, with 13% #Nets reductions.

bfly-ftune

(c) -- More can be found in the listed folders in this repo.

flowtune's People

Contributors

ycunxi avatar

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    ๐Ÿ–– Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. ๐Ÿ“Š๐Ÿ“ˆ๐ŸŽ‰

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google โค๏ธ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.