AVXECC is a throughput-optimized Elliptic Curve Cryptography (ECC) library that exploits the parallel processing capabilities of Intel's Advanced Vector eXtensions (AVX2). It can execute four instances of scalar multiplication in parallel, whereby each instance uses a 64-bit element of a 256-bit AVX2 vector. AVXECC supports variable-base scalar multiplication on Curve25519 and fixed-base scalar multiplication on the twisted Edwards curve that is birationally-equivalent to Curve25519. Each of the four instances can use a different scalar and, in the case of variable-base scalar multiplication, also a different base point. The implementation is resistant against timing attacks since it does not execute any secret-dependent branches or loads. Further details can be found in the paper linked below.
Supported cryptosystems and platforms:
- X25519 (key-pair generation and shared-secret computation) using AVX2
- planned: Ed25519 (signature generation and verification) using AVX2
- planned: X25519 and Ed25519 using AVX512 and AVX512-IFMA
A paper describing the "limb-slicing" technique upon which the implementation of AVXECC is based was presented at the 27th International Workshop on Selected Areas in Cryptography (SAC 2020). A preliminary version of the paper is available here.
Clang/GCC
The source code contains a simple makefile for Clang. This makefile can be
easily modified for other compilers (e.g. GCC), but the performance may be
affected since AVXECC was specifically "tuned" for Clang. The library can be
compiled and tested with the following commands.
$ make
$ ./test_benchMicrosoft Visual Studio 2017
The subdirectory vs2017 contains the project/solution file AVXECC.sln for
Visual Studio 2017. After loading the project it will most likely be necessary
to adjust the Windows SDK version in the project settings (currently Windows SDK
version 10.0.17763.0 is used). To modify the SDK version go to Project - AVXECC
Properties (Alt F7) - Configuration Properties - General - Windows SDK Version.
Hao Cheng, Johann Groszschaedl, and Jiaqi Tian (SnT and DCS, University of Luxembourg). Copyright © 2020-2024 University of Luxembourg.
GNU General Public License (GPL) version 3 (see LICENSE file for details). Other licenses may be available on request, please contact Johann Groszschaedl (firstname(dot)lastname(at)uni(dot)lu).