Welcome to CirKit’s documentation!

Installation

Requirements and dependencies

The following software is required in order to build CirKit

  • git (at least version 1.6.5)
  • cmake (at least version 3.0.0)
  • g++ (at least version 4.9.0) or clang++ (at least version 3.5.0)
  • boost (at least version 1.56.0)
  • GNU MP, and its C++ interface GMP++
  • GNU readline

Installing dependencies in Ubuntu Linux

In Ubuntu the packages can be installed with:

sudo apt-get install build-essential git g++ cmake libboost-all-dev libgmp3-dev libxml2-dev zlib1g-dev lapack openblas

Installing dependencies in Arch Linux

In arch the packages can be installed with:

sudo pacman -S base-devel git g++ cmake boost boost-libs gmp libxml2 zlib lapack

Installing dependencies in Mac OS

In Mac it’s recommended to use Homebrew to install the required packages:

brew install boost cmake gmp readline git lapack openblas

Build and run CirKit

After extracting or cloning CirKit perform the following steps:

git clone --recursive https://github.com/msoeken/cirkit.git
cd cirkit
mkdir build
cd build
cmake ..
make external
make cirkit

CirKit can be executed with:

build/programs/cirkit

Build and run RevKit

After extracting or cloning CirKit perform the following steps:

git clone --recursive https://github.com/msoeken/cirkit.git
cd cirkit
mkdir build
cd build
cmake -Denable_cirkit-addon-reversible=ON -Denable_cirkit-addon-formal=ON ..
make external
make revkit

RevKit can be executed with:

build/programs/revkit

Python interface

The CLI from CirKit and RevKit can be invoked using a Python interface. The implementation is based on pybind11 and can be enabled and compiled as follows:

cd build
cmake -Dcirkit_ENABLE_PYTHON_API=ON -DPYBIND11_PYTHON_VERSION=3.6 -DPYTHON_EXECUTABLE=/path/to/python ..
make cirkit_python

Replace /path/to/python to the path of a Python3 executable on the system and replace cirkit_python by revkit_python to build the RevKit CLI Python API.

Start Jupyter notebook

The Python CLI can conveniently be used through a Jupyter notebooks and some notebooks are provided in the jupyter directory. From the main directory in CirKit, one can do the following:

cd jupyter
PYTHONPATH=`pwd`/../build/programs jupyter notebook

Troubleshooting

No recent cmake version

If for some reason only older versions of cmake are available on the system, you can install cmake using utils/tools.py install cmake directly from CirKit. Afterwards update the $PATH variable by typing export PATH=<path-to-cirkit>/ext/bin:$PATH.

Libraries not found

It’s best to set the environment variable CIRKIT_HOME to the directory of CirKit using:

export CIRKIT_HOME=<full-path-to-cirkit>

or some other command depending on the user’s shell. Also, sometimes depending libraries are not found, then run:

export LD_LIBRARY_PATH=$CIRKIT_HOME/ext/lib:$LD_LIBRARY_PATH

in Linux or:

export DYLD_LIBRARY_PATH=$CIRKIT_HOME/ext/lib:$DYLD_LIBRARY_PATH

in Mac OS.

No recent Boost version

There is a manual way to install Boost and CirKit’s package manager can help. Before installing CirKit run:

mkdir build
./utils/tools.py install boost
cd build
cmake -DBoost_NO_SYSTEM_PATHS=TRUE -DBOOST_ROOT:PATHNAME=`pwd`/tools/boost_1_63_0/ ..
make external
make cirkit

Of course, one can add further options to the cmake command in the fourth line, e.g., to build RevKit.

Getting Started

Running CirKit

CirKit has a command-line interface. After calling build/programs/cirkit a shell prompt is printed to the screen. In order to see all available commands type help . This prints a list with all commands together with a short description for each of them. To see more details about a command and its usage call the command together with the -h flag. For example read_aiger -h prints options to read an AIGER file.

CirKit can be called in three different modes:

  1. Interactive mode: This is the default interactive mode that is described above.

  2. Bash mode: -c In this mode, commands are given to CirKit as command line arguments, e.g.,:

    build/programs/cirkit -c "read_aiger file.aig; cone -o y; simulate -at; quit"
    

    This example reads an AIGER from file.aig, reduces the network to the cone of output y, simulates the resulting network as truth table and quits. By adding the flag -e additionally each command is printed to the screen before execution, e.g.,:

    build/programs/cirkit -ec "read_aiger file.aig; cone -o y; simulate -at; quit"
    

    Note that single character command options (that start eith with a single -) can be concatenated, e.g., -e -c can be written as -ec.

  3. Batch mode: -f In this mode, commands are read line-by-line from a file, e.g.,:

    build/programs/cirkit -f command_file
    

    This mode also accepts the -e flag to print each command before execution. It is possible to comment some commands in the command file by starting a line with a # character.

Some of the main commands are:

Command Description
alias Creates an alias
help Shows all available commands
quit Quits CirKit
set Sets (internal) environment variables

Stores

Shared data in CirKit such as circuits or truth tables are stored in stores and commands can access the data from them. Each data structure has its own store and each store can hold more than one element. For example, there are seperate stores for truth table, AIGs, and BDDs. Call store -h to see all available stores. Each store comes with its own command flag to access it, e.g., -a for AIGs and -t for truth tables. Although a store can hold more than one element, it is not necessary and possible to specify which store element to access. Instead each store indivually has a pointer to the current store element and commands always access this one. In order to access a different store element, one can change the pointer using the command current. For example, current -a 1 will set the pointer in the AIG store to the element with index 1 (which is the second element in the store).

Example

read_aiger file1.aig
read_aiger file2.aig
store --show -a
read_aiger -n file3.aig
store --show -a
current -a 0
store --clear -a

First file1.aig is read into the AIG store. The second command reads file2.aig and overrides the current entry. Overriding the current store element is the default behavior of most commands. The current content of a store can be displayed with store --show -a or store -a as a short-hand. In the third line of the example, one AIG is in the store. When passing the flag -n to read_aiger as in the fourth command a new entry is added to the store and the current index is updated to the new entry, i.e., at this time the AIG store contains two elements with the current element being the second one (index 1). With current -a 0 the current index is reset to 0 and store --clear -a clears the store from all AIGs.

For many commands it is clear which store they access and it’s not necessary to specify the store. There are some generic commands which work on all data structures and require to pass the store access flag, e.g., the command store. The generic commands are:

Command Description
convert Converts store elements into other types, e.g., AIGs to BDDs
current Changes the current store pointer
print Prints a textual ASCII representation of the current store element
ps Prints statistical information about the current store element
show Visualizes the current store element (writes to a dot file)
store Shows and clears elements from the store

Logging

Passing -l file.log to cirkit creates a log file of the session. This option is particularly useful in batch mode. The log file contains a JSON array with an entry for each command. Each entry contains at least the full command that was run and the time at which the command was started to execute. Some commands write additional data into the log file. For example, ps -a writes number of inputs, outputs, and AND gates of an AIG, and quit writes several information about the computer on which CirKit has been executed. Being a JSON array, the log file can be easily parsed as many programming languages have a JSON library.

Some helper functions to parse the log file and, e.g., create ASCII tables from them can be found in utils/experiments.py. Further, the Python program utils/extract_script.py extracts a CirKit script file from the the log that can be run in batch mode. This can be helpful when logging an interactive session and then rerunning the commands:

$ cirkit -l session.log
cirkit> read_aiger file.aig
cirkit> ps -a
...
cirkit> quit
$ utils/extract_script.py session.log > session.cs
$ cirkit -f session.cs

For performing experimental evaluations, the following workflow is suggested. Create two Python programs (or any other programming language) called make_script.py and make_table.py. The program make_script.py writes a CirKit script. The program make_table.py reads the log file created for the script and prints out a table:

$ ./make_script.cs experiments.cs
$ cirkit -f experiments.cs -l experiments.log
$ ./make_table.cs experiments.log

Aliases

The command alias allows to create aliases, which are shortcuts to commands or sequences of commands. The best place for aliass is the init file alias located in the directory that is specified in the $CIRKIT_HOME environment variable. It is recommended to set $CIRKIT_HOME to the root directory of CirKit. Examples for entries in an alias file are::

alias e2t "convert --expr_to_tt"

The alias command gets two arguments, the key and the value that is used for substituion. If the key or the value contain a space they need to be put into quotes, and internal quotes need to be escaped.

Note that they key can be any regular expression with capture groups and that the value is a formatted string that can contain placeholders for each capture string: %1% for the first capture group, %2% for the second one and so on. Note that the % sign needs to be escaped. A more complex example is an alias to read a Verilog file into an AIG using ABC::

alias "abc_verilog (.*)" "abc -c \"%%read %1%; %%blast\""

This will translate, e.g., the command abc_verilog file.v into:

abc -c "%read file.v; %blast"

Since the key is any regular expression, we can create aliases which are very expressive. The alias:

alias "(\\w+) > (\\w+)" "convert --%1%_to_%2%"

allows, e.g., to convert a truth table into an AIG using tt > aig. Putting everything together we can write scripts in CirKit such as:

abc_verilog file.v
aig > bdd
bdd -c

which reads a Verilog file into a CirKit AIG using ABC’s API, then converts the AIG into a BDD and finally computes the characteristic function of the BDD.

Aliases are also useful inside scripts when they are only required locally. Consider, e.g., one wants to convert several truth tables into AIGs, optimize them, and then write them into a file. A script for this task could look as follows:

alias "tt_aig_prog ([01]+)" "tt %1%; tt > aig; abc -c &dc2; ps -a; write_aiger %1%.aag"
tt_aig_prog 11101000
tt_aig_prog 01011101
tt_aig_prog 0110
tt_aig_prog 1001100111010111
tt_aig_prog 1101110011000000

Data Structures

CirKit (and RevKit) provide the analysis and manipulation of several data structures. These data structures are explained in this section. As described above, instances of these data structures are stored in individual stores. Not all data structures are available in both CirKit and RevKit. The following table gives an overview over the existing data structures, their access option for the store, and their availability in CirKit and RevKit.

Data structure Access option CirKit RevKit
Truth table -t --tt
Expression -e --expr
And-inverter graph (AIG) -a --aig
Majority-inverter graph (MIG) -m --mig  
XOR majority graph (XMG) -x --xmg
Binary decision diagram (BDD) -b --bdd
Reversible circuit -c --circuit  
Reversible specification -s --spec  
BDD of a characteristic reversible function (RCBDD) -r --rcbdd  

Truth tables

Truth tables are bitstrings of length $2^k$ and represent Boolean functions over $k$ variables. The most significant bit is the first bit in the bitstring. For example, to load a truth table that represents the AND function $a land b$, type tt 1000. We assume that the least significant variable is $a$, then $b$, then $c$, and so on. The truth tables for $a$, $b$, and $c$ are therefore 10, 1100, and 11110000. In order to meet size requirements, truth tables can be extended. If, e.g., 1011 is the current truth table in store, the command tt -e 3 extends the truth table to be defined over 3 variables, yielding 10111011.

On can convert truth tables into AIGs using convert --tt_to_aig. This will construct an AIG in a very naïve way by constructing each minterm explicitly and then ORing them all. Conversely, one can obtain truth tables from AIGs using simulation. For this purpose use the command simulate with the flags -a to simulate from AIGs, -t to simulate to truth tables, and -n to store the simulatuion results. The following example illustrates the usage for the c17 benchmark from the ISCAS benchmark suite. It also employs NPN canonization on the resulting truth tables using the command npn.

Example

cirkit> read_aiger c17.aig
cirkit> simulate -atn
[i] G16 : 1011100011111000101110001111100010111000111110001011100011111000 (B8F8B8F8B8F8B8F8)
[i] G17 : 0011001111111111001100001111000000110011111111110011000011110000 (33FF30F033FF30F0)
[i] runtime: 0.00 secs
cirkit> store -t
[i] truth tables in store:
     0: 1011100011111000101110001111100010111000111110001011100011111000
  *  1: 0011001111111111001100001111000000110011111111110011000011110000
cirkit> current -t 0
cirkit> npn -t --approach 0
[i] run-time: 0.89 secs
[i] NPN class for 1011100011111000101110001111100010111000111110001011100011111000 is 0000000000000000000000000000111111110000111100001111111111111111
[i] - phase: 1001010 perm: 5 4 1 3 0 2
cirkit> current -t 1
cirkit> npn -t --approach 0
[i] run-time: 0.89 secs
[i] NPN class for 0011001111111111001100001111000000110011111111110011000011110000 is 0000000000000000000000001111111100001111000011110000111111111111
[i] - phase: 1001010 perm: 5 0 1 2 4 3

The current truth table in the store corresponds to the last output of the AIG. Notice that truth table simulation only scales for AIGs with a small number of inputs. One can obtain a truth table from an expression using convert --expr_to_tt or its alias expr > tt.

Some truth table related commands are:

Command Description
tt Load and modify truth tables
npn NPN canonization (exact and heuristic)
convert --tt_to_aig, Alias: tt > aig Convert truth table to AIG
convert --expr_to_tt, Alias: expr > tt Convert expression to truth table
simulate -atn Simulates AIGs as truth table and stores them
simulate -mtn Simulates MIGs as truth table and stores them

Expressions

Expressions provide an easy way to enter Boolean functions into CirKit. The expressions are multi-level expressions that can contain constants (0, 1), Boolean variables (a, b, c, …), inversion (!), binary AND (()), binary OR ({}), binary XOR ([]), and ternary MAJ (<>). The whole syntax is given as follows:

expr ::= 0 | 1 | var | ! expr | ( expr expr ) | { expr expr } | [ expr expr ] | < expr expr expr >
var ::= a | b | c | ...

Note that a always corresponds to the least significant bit, b to the second least significant bit, and so on. Expressions can be loaded into its store (access flag -e) using the command expr. Expressions can be used as starting point to create truth tables (expr > tt) or binary decision diagrams (expr > bdd) for simple functions and avoid to create a file. The following example illustrates its usage.

Example

cirkit> expr (ab)
cirkit> expr > tt
cirkit> print -t
1000
cirkit> expr !{ac}
cirkit> expr > tt
cirkit> print -t
00000101
cirkit> expr {{(ab)(ac)}(bc)}
cirkit> expr > tt
cirkit> print -t
11101000
cirkit> expr
cirkit> expr > tt
cirkit> print -t
11101000

Note that when loading !{ac} the resulting truth table represents a 3-variable Boolean function which does not functially depend on the value for b. The last two examples are both Boolean expressions for MAJ, the majority-of-three function.

Some commands related to expressions are:

Command Description
expr Load expressions
convert --expr_to_tt, Alias: expr > tt Convert expression to truth table
convert --expr_to_bdd, Alias: expr > bdd Convert expression to binary decision diagram

And-inverter Graphs (AIG)

Loading an AIG into CirKit

There are several ways to load an AIG into CirKit. If the AIG is represented as AIGER file with extension *.aig if in binary format and *.aag if in ASCII format, one can use the command read_aiger to parse the file and create an AIG in the store. If already an AIG is in the store, it will be overriden, unless one calls read_aiger -n. If the AIG is represented in Verilog such that ABC’s command %read is able to parse it, one can use read_verilog -a to read the Verilog file, convert it into an AIG and put it into the store. Also BENCH files can be read into AIGs with the command read_bench. The command tt > aig allows to translate the current truth table into an AIG. Internally, ABC’s API will be used for that purpose and the AIG is optimized using dc2.

This summary lists commands to load AIGs into CirKit:

Command Description
read_aiger Read AIG from binary or ASCII AIGER file
read_verilog -a Read AIG from Verilog file (using ABC’s %read command)
read_bench Read AIG from BENCH file
convert -tt_to_aig, Alias: tt > aig Convert truth table to AIG

Manipulating the AIG

ABC is a powerful tool for AIG optimization and manipulation and using the tight integration of CirKit with ABC using the command abc, it is very easy to use ABC to optimize AIGs in CirKit directly. Hence, few commands in CirKit exist to perform AIG optimization, but mainly utility commands.

This list some commands in CirKit to manipulate an AIG:

Command Description
cone Extracts AIG based on output cones
cuts -a Performs cut enumeration
propagate Propagates constants through inputs
rename Renames inputs and outputs
shuffle -a Shuffles I/O of an AIG
strash Strashes an AIG (removes dangling nodes)
unate Computes unateness properties and functional dependencies of the AIG

Writing an AIG

AIGs can be written into AIGER files using write_aiger or into Verilog files using write_verilog -a.

This summary lists commands to write AIGs:

Command Description
write_aiger Write AIG to ASCII file (ASCII if suffix is .aag)
write_verilog -a Write AIG to Verilog file

Integration

CirKit offers functions to interact with certain tools. This section shows with which tools CirKit interacts well and gives some illustrative examples.

ABC

CirKit is tightly integrated with ABC. ABC can be accessed as a subshell inside CirKit with the command abc. If an AIG is present in CirKit’s AIG store, it will be copied to the ABC subshell and available in the &-space (ABC9 commands). Furthermore, when leaving the ABC subshell using quit ABC’s AIG will be copied back to CirKit and replace the current AIG in the store (unless abc is called with the option -n). The following example illustrates this interaction in which an AIG is copied to ABC in order to optimize it and then copied back:

cirkit> read_aiger c432.aig
cirkit> ps -a
[i]                 c432: i/o =      36 /       7  and =     136  lev =   25
cirkit> abc
UC Berkeley, ABC 1.01 (compiled Apr 22 2016 19:45:32)
abc 01> &ps
c432     : i/o =     36/      7  and =     136  lev =   25 (19.14)  mem = 0.00 MB
abc 01> &dc2
abc 01> &ps
c432     : i/o =     36/      7  and =     123  lev =   25 (19.14)  mem = 0.00 MB
abc 01> quit
cirkit> ps -a
[i]                 c432: i/o =      36 /       7  and =     123  lev =   25
cirkit> quit

RevKit

This page lists some RevKit specific documentation. Many parts of this documentation were contributed from Aldo Sayeg.

Input and output

In general, in order to create an instance of a data structure in RevKit we need to load the data structure from a file in a format that describes the data structure. Expressions and truth tables are the only structure that can be created inside of Revkit by using the expr and the tt command, respectively.

The data structures can often also be written to the same file format. The following table lists all of the data structures’ read and write commands used in RevKit along with a short decription.

Data structure Format Read command Write command Description
Reversible circuits REAL read_real write_real Reversible circuit representation using different gates as basis. Part of RevLib supported formats.
Reversible specification SPEC read_spec write_spec Truth table of a reversible circuit. Part of RevLib supported formats.
Binary decision diagram PLA read_pla write_pla Sum of products representation of a Boolean function.
AND-inverter graph AIGER read_aiger write_aiger Format developed for the AIGER utilities.
AND-inverter graph Verilog read_verilog -a write_verilog -a Parses whatever can be read with ABC’s %read command.
AND-inverter graph Bench read_bench   Format developed for traditional circuits as part of ABC.
XOR-majority graph Verilog read_verilog -x write_verilog -x Simple single-statement assignment Verilog file.

Adding a command to RevKit

This tutorial explains how to integrate a new simple command into RevKit. As an example, a command called unopt is implemented, that copies gate in a reversible circuit without modifying the functionality.

CirKit provides utility scripts in order to create new files easily. We create a file for the new command using the following script:

./utils/make_src_file.py cli/commands/unopt reversible

Note, that the first argument is the path to the filename without the src/ in the beginning and without an extension in the end. Two files, a header and a source file, are created. The second parameter ensures that the files are created for the RevKit add-on. The third parameter is optional and can have a name of the file author. If not specified, the name is fetched from the user’s git configuration.

The header file in addons/cirkit-addon-reversible/src/cli/commands/unopt.hpp contains already some skeleton code and we extend it as follows (all comments are omitted in the code):

// unopt.hpp
#ifndef CLI_UNOPT_COMMAND_HPP
#define CLI_UNOPT_COMMAND_HPP

#include <cli/cirkit_command.hpp>

namespace cirkit
{

class unopt_command : public cirkit_command
{
public:
  unopt_command( const environment::ptr& env );
  rules_t validity_rules() const;

protected:
  bool execute();

private:
  unsigned copies = 1u;
};

}

#endif

We define a command with the base class cirkit_command. It is important that the class name is unopt_command to be used in later code that makes use of macros and relies on some naming conventions. Two methods need to be implemented, the constructor that will set up the arguments which can be passed to the command, and execute which executes the code and calls our algorithm. We also implement the method validity_rules to ensure that the store contains at least one reversible circuit when calling the command. More details on how to write commands can be found in the abc_cli example program.

// unopt.cpp
#include "unopt.hpp"

#include <alice/rules.hpp>
#include <cli/reversible_stores.hpp>
#include <core/utils/program_options.hpp>
#include <reversible/target_tags.hpp>
#include <reversible/functions/copy_metadata.hpp>

namespace cirkit
{

unopt_command::unopt_command( const environment::ptr& env )
  : cirkit_command( env, "unoptimize circuits" )
{
  opts.add_options()
    ( "copies,c", value_with_default( &copies ), "number of gate copies" )
    ;

  add_new_option(); /* adds a flag --new, or -n that can be used to add a new
                       store entry instead of overwriting it */
}

command::rules_t unopt_command::validity_rules() const
{
  return {has_store_element<circuit>( env )};
}

bool unopt_command::execute()
{
  auto& circuits = env->store<circuit>(); /* access store with reversible circuits */

  /* reference to current circuit, and new circuit with same properties */
  const auto& circ = circuits.current();
  circuit circ_new;
  copy_metadata( circ, circ_new );

  for ( const auto& g : circ )     /* iterate through the gates */
  {
    circ_new.append_gate() = g;    /* copy existing gate */
    if ( is_toffoli( g ) )         /* some more copies, if gate is Toffoli */
    {
      for ( auto i = 0u; i < 2u * copies; ++i )
      {
        circ_new.append_gate() = g;
      }
    }
  }

  extend_if_new( circuits ); /* extend store by empty element if --new option is set */
  circuits.current() = circ_new;

  return true; /* always return true */
}

}

The function should always return true.

We are almost done. Next, we add the command to the RevKit executable. For this purpose, open the file addons/cirkit-addon-reversible/programs/reversible/revkit.cpp and add the following header, where other headers are included:

#include <cli/commands/unopt.hpp>

And then add the command in the same style as other commands are added using:

ADD_COMMAND( unopt );

That’s it. We rebuild RevKit with:

make -C build revkit

and then call it to try out the new command:

revkit> read_spec -p "0 4 2 1 0 3 7 5"
revkit> tbs
[i] run-time: 0.00 secs
revkit> ps -c
Lines:        3
Gates:        7
T-count:      21
Logic qubits: 4
revkit> unopt
revkit> ps -c
Lines:        3
Gates:        21
T-count:      63
Logic qubits: 4

Exercises

Here are some suggestions for exercises (with a difficulty estimation from 0–50) to extend the add-on.

  1. [25] Copy all gates which are self-inverse in this manner based on a syntactic comparison.

Indices and tables