International Association
for Cryptologic Research

Dot-product is a widely used operation in many machine learning and scientific computing algorithms. Recently, NVIDIA has introduced dot-product instructions (DP2A and DP4A) in modern GPU architectures, with the aim of accelerating machine learning and scientific computing applications. These dot-product instructions allow the computation of multiply-and-add instructions in a clock cycle, effectively achieving higher throughput compared to conventional 32-bit integer units. In this paper, we show that the dot-product instruction can also be used to accelerate matrix-multiplication and polynomial convolution operations, which are commonly found in post-quantum lattice-based cryptographic schemes. In particular, we propose a highly optimized implementation of FrodoKEM, wherein the matrix-multiplication is accelerated by the dot-product instruction. We also present specially designed data structures that allow an efficient implementation of Saber key encapsulation mechanism, utilizing the dot-product instruction to speed-up the polynomial convolution. The proposed FrodoKEM implementation achieves 4.37x higher throughput in terms of key exchange operations per second than the state-of-the-art implementation on V100 GPU. This paper also presents the first implementation of Saber on GPU platforms, achieving 124,418, 120,463, and 31,658 key exchange operations per second on RTX3080, V100, and T4 GPUs, respectively. Since matrix-multiplication and polynomial convolution operations are the most time-consuming operations in lattice-based cryptographic schemes, our proposed techniques are likely to benefit other similar algorithms. The proposed high throughput implementation of KEMs on various GPU platforms allows the heavy computations (KEMs) to be offloaded from the server. This is very useful for many emerging applications like Internet of Things and cloud computing.