formosa-mldsa

Jasmin implementations of the pure, hedged version of ML-DSA for x86-64 and ARMv7M.

• Overview
• Quickstart

Overview

This repository contains, for all parameter sets, the following implementations of the pure, hedged version of ML-DSA:

x86-64

1. A reference implementation (found under x86-64/ref).
2. An implementation optimized using the AVX2 instruction set (found under x86-64/avx2).

ARMv7M

1. A reference implementation (found under arm-m4/ref).
2. An implementation that minimizes use of stack space (found under arm-m4/lowram).

Quickstart

0. Ensure you have the Jasmin compiler installed, using the latest commit on the main branch of the project. Also ensure that jasminc is in the PATH, since this is what the Makefile invokes.

1. To generate assembly code for, say, the AVX2 implementation of ML-DSA-65 on the x86-64 platform, run:

env ARCHITECTURE=x86-64 PARAMETER_SET=65 IMPLEMENTATION_TYPE=avx2 make

The resulting assembly implementation will be found in ml_dsa_65_avx2_x86-64.s.

1. Similarly, to generate assembly code for, say, the minimal-stack-usage implementation of ML-DSA-44 on the ARMv7M platform, run:

env ARCHITECTURE=arm-m4 PARAMETER_SET=44 IMPLEMENTATION_TYPE=lowram make

The resulting assembly implementation will be found in ml_dsa_44_lowram_arm-m4.s.