Script generated finite field arithmetic for elliptic curve cryptography
This repository contains Python 3 scripts for the automatic generation of efficent code for multi-precision modular arithmetic, for 16, 32 and 64-bit architectures, in C and in Rust.
The code uses a multi-limb unsaturated-radix representation for big numbers.
For both languages there are two scripts, one specialised for pseudo-Mersenne moduli of the form
The code generated includes functions for modular addition, subtraction, multiplication, inversion, quadratic residuosity and square roots. In other words all of the requirements to implement field arithmetic in the context of elliptic curve cryptography.
As dependencies it is required that an appropriate C compiler (gcc/clang/icx) and/or the Rust compiler (rustc) is included in the path. For the C code also ensure that the clang-format and cppcheck utilities are installed.
Also in the path must be the very useful utility addchain
https://github.com/mmcloughlin/addchain
For accurate timings across a range of architectures for the C code, install Dan Bernstein's libcpucycles utility from https://cpucycles.cr.yp.to/ . It may be necessary to run ldconfig.
For a quick start copy the files from here into a working directory, and try
python3 pseudo.py 64 2**255-19
./time
Then 64-bit code for the suggested modulus is generated and tested. An executable that times important functions will be created if the platform allows it. The standalone C timing code is output to time.c, and code for production use is output to code.c and header.h
For Rust
python3 pseudo_rust.py 64 2**255-19
./time
In this case the output is directed to files time.rs and code.rs
For more details read the comments in the provided scripts.
As a Proof of Concept, elliptic curve code for RFC7748 is provided in the files rfc7748.c and rfc7748.rs
Read comments in these files for simple build instructions.
(RFC7748 describes an implementation of Diffie-Hellman key exchange on the Montgomery elliptic curves C25519 and C448. You can create your own Montgomery curve using the sagemath script provided in the file bowe.sage)
Assume the modulus is
nres() -- Convert a big number to internal format
redc() -- Convert back from internal format, result
modfsb() -- Perform final subtraction to reduce from
modadd() -- Modular addition, result
modsub() -- Modular subtraction, result
modneg() -- Modular negation, result
modmul() -- Modular multiplication, result
modsqr() -- Modular squaring, result
modmli() -- Modular multiplication by a small integer, result
modcpy() -- Copy a big number
modpro() -- Calculate progenitor, for subsequent use for modular inverses and square roots
modinv() -- Modular inversion
modsqrt() -- Modular square root
modis1() -- Test for equal to unity
modis0() -- Test for equal to zero
modone() -- Set equal to unity
modzer() -- Set equal to zero
modint() -- Convert an integer to internal format
modqr() -- Test for quadratic residue
modcmv() -- Conditional constant time move
modcsw() -- Conditional constant time swap
modshl() -- shift left by bits
modshr() -- shift right by bits
modexp() -- export from internal format to byte array
modimp() -- import to internal format from byte array
modsign() -- Extract sign (parity bit)
modcmp() -- Test for equality \
The scripts can be used out-of-the-box using simple command-line arguments, as in the examples above.
The scripts can also be tailored in various ways at the top of the script, and in the ``user editable area''.
There are default settings for the choice of compiler, choice of using a clock cycle counter, encouragement for inlining certain functions, code formatting, the use of Karatsuba for (maybe) faster multiplication, and, for the C code, an option to ``decorate'' function names to avoid name clashes.
New named moduli can also be provided in the user editable area, and some settings (like radix choice) adapted individually.
Function name decoration may be required to avoid name clashes in C. If using C++ namespaces can be used to avoid this necessity. It is not an issue for Rust.
All generated functions are written with the expectation that they will execute in constant time. But high level code is nevertheless at the mercy of both the compiler and the architecture.
It is strongly recommended that the generated assembly language be closely studied to ensure that there are no compiler introduced timing leaks. In particular code generated for the functions modcmv and modcsw should be checked, bearing in mind that they may be inlined by the compiler. If necessary compiler-specific measures should be taken to prevent inlining, and/or place these functions into a separately compiled module.
- Decide on wordlength (32 or 64 bit)
- Choose the field prime.
- Automatically generate field code to files code.c and header.h. If (pseudo)-Mersenne prime, use script pseudo.py, else use monty.py
- Decide on Edwards or Weierstrass curve. The number of points on the curve is assumed to be a prime for Weierstrass, or 4 or 8 times a prime for Edwards. This prime is the order of the group.
- Drop code.c into edwards.c or weierstrass.c and provide some curve constants (a constant B or d and a group generator point x, y) where indicated (some are provided already). If larger constants are required, a tool make.py is provided to generate them
- Insert the prime group order into testcurve.c where indicated (several examples are there already)
- Make sure fixed API header file curve.h is in the path. Note that this API is independent of the curve, its associated field and its parameters.
- Compile and link testcurve.c with edwards.c or weierstrass.c
- Run testcurve to test the arithmetic and perform some timings.
The API interface is as indicated in curve.h. The API is completely implemented in edwards.c or weierstrass.c
python pseudo.py 64 Ed25519
Drop code.c into edwards.c where indicated
gcc -O2 testcurve.c edwards.c -lcpucycles -o testcurve
./testcurve
Note that this intermediate API only provides the elliptic curve functionality. A higher level algorithm API (like that provided for Ed448 signature) would use this API while itself providing additional algorithm specific random number and hashing functionality. It may also use the monty.py script to generate code to perform arithmetic modulo the prime group order, if so required by the algorithm.
python monty.py 64 0x3fffffffffffffffffffffffffffffffffffffffffffffffffffffff7cca23e9c44edb49aed63690216cc2728dc58f552378c292ab5844f3
Drop code.c into Ed448.c where indicated
python monty.py 64 Ed448
Drop code.c into edwards.c where indicated
gcc -O2 Ed448.c edwards.c -o Ed448
./Ed448