Algorithm-Aware Hardware Optimization using E-Graph Rewriting: how should we marry software and hardware?
Software-defined hardware design, also known as high-level synthesis (HLS), automatically translates a software program in a high-level language into a low-level hardware description. The software representation of a hardware design allows existing software optimization techniques to be applied at the algorithm level for efficient hardware synthesis. However, the hardware designs produced by these tools still suffer from a significant performance gap compared to manual implementations. This is because the input programs must still be written using hardware design principles.
Existing techniques either leave the program source unchanged or perform a fixed sequence of source transformation passes, potentially missing opportunities to find the optimal design. Our prior work proposes an efficient approach named SEER (Super-optimization Exploration using E-graph Rewriting) for software-defined hardware design that automatically rewrites a software algorithm into a representation that can be used to generate an efficient hardware design. SEER provides an extensible framework, orchestrating existing software compiler passes and hardware optimizers. In this paper, we would like to share the learned lessons and future challenges about algorithm-aware hardware optimizations using e-graph rewriting.
Mon 24 JunDisplayed time zone: Windhoek change
13:40 - 15:20 | |||
13:40 25mTalk | Powered by Less: Low Power Circuit Synthesis EGRAPHS Samuel Coward Imperial College London, UK / Intel Corporation, Theo Drane Intel Corporation, USA, Emiliano Morini Intel Corporation, George A. Constantinides Imperial College London, UK | ||
14:05 25mTalk | Algorithm-Aware Hardware Optimization using E-Graph Rewriting: how should we marry software and hardware? EGRAPHS Jianyi Cheng University of Cambridge, Samuel Coward Imperial College London, UK / Intel Corporation, Rafael Barbalho Intel Corporation, Theo Drane Intel Corporation, USA Link to publication DOI | ||
14:30 25mTalk | Loop Saturation for Scalable High-Level Synthesis EGRAPHS Camille Bossut Georgia Institute of Technology, Qirun Zhang Georgia Institute of Technology, Cong "Callie" Hao Georgia Institute of Technology | ||
14:55 25mTalk | SpEQ: Translation of Sparse Codes using Equivalences EGRAPHS Avery Laird University of Toronto, Bangtian Liu University of Toronto, Nikolaj Bjørner Microsoft Research, Maryam Mehri Dehnavi University of Toronto |