Workshops Archive

A Highly Optimized Double-Word Arithmetic Implementation using Versal DSPs


Workshop: Twelfth International Workshop on Heterogeneous High-Performance Reconfigurable Computing (H2RC 2026)

Authors: David Volz, Torben Kalkhof, Tim Noack, and Andreas Koch (Technical University of Darmstadt)

Abstract: Scientific applications often require high-precision floating-point operations. Modern FPGAs provide dedicated hardware for single-precision (FP32) multiply-accumulate operations, but double-precision (FP64) still requires costly implementations using lookup tables (LUTs). As an alternative to FP64, double-word arithmetic can nearly double the precision of FP32 using only single-precision instructions. In double-word arithmetic, a number is represented as a pair of FP32 values, which is carefully normalized so that its two components never overlap.

This work presents, to the best of our knowledge, the first implementation of double-word arithmetic on FPGAs. We develop both an HLS-based and a hand-optimized implementation. Our hand-optimized implementation fully exploits the functionality available in the DSP blocks, requiring zero LUTs or flip-flops for multiplication and almost no LUTs or flip-flops for addition.

We evaluate our double-word implementation by comparing to HLS and vendor provided IP, then demonstrate it to be a viable alternative to FP64 on modern FPGA.


Back to Twelfth International Workshop on Heterogeneous High-Performance Reconfigurable Computing (H2RC 2026) Archive Listing

Back to Full Workshop Archive Listing