LP-012: Post-Quantum Cryptography GPU Acceleration

Abstract

All NIST FIPS post-quantum algorithms run natively on GPU. The Number Theoretic Transform (NTT) is the shared primitive — 6 GPU NTT variants accelerate ML-DSA, ML-KEM, SLH-DSA, and Pulsar simultaneously.

Algorithms

Shared NTT

6 NTT variants optimized for different use cases:

• ntt.metal: General-purpose butterfly
• four_step_ntt.metal: Large-degree polynomials
• ntt_unified_memory.metal: Apple Silicon zero-copy
• ntt_kernels.metal: Fused NTT + pointwise multiply
• poly_mul.metal: Polynomial multiplication via NTT
• twiddle_cache.metal: Pre-computed twiddle factors