Synthesis of Sorting Kernels
Year: 2025
Type: conference-paper
Abstract: Recently, AlphaDev has shown significant advances in the synthesis of branchless sorting kernels for arrays of lengths 3 to 5. In this paper, we propose an enumerative search technique based on A* search and present novel optimality-preserving heuristics and non-optimality-preserving cuts for sorting kernel synthesis. Our algorithm outperforms AlphaDev in synthesis time by two orders of magnitude ran on a standard notebook instead of a TPU cluster. Because our algorithm can explore the solution space, we are able to enumerate all correct sorting kernels for length 3 and simply select the best-performing one. For larger array lengths, we intelligently sample the solution space and find a sorting kernel that outperforms the state-of-the-art. Furthermore, we establish a new tight lower bound for the shortest sorting kernel for length 4. Finally, we provide a comprehensive comparison against several other existing synthesis techniques and show that none of them is able to synthesize sorting kernels for arrays longer than 3.
Authors: Marcel Ullrich, Sebastian Hack
Institution: Saarland University
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4407857908