DiffHLS: Differential Learning for High-Level Synthesis QoR Prediction with GNNs and LLM Code Embeddings

arXiv cs.LG / 4/13/2026

💬 OpinionIdeas & Deep AnalysisModels & Research

共有:

Key Points

DiffHLS is proposed as a differential-learning framework to predict High-Level Synthesis (HLS) Quality-of-Result (QoR) by learning from pairs of a kernel baseline and a pragma-modified design variant.
The method encodes both kernel and design intermediate-representation graphs using dedicated GNN branches, then predicts the baseline and the design-induced delta jointly for composition into the final design QoR.
It incorporates pretrained LLM code embeddings into the delta pathway, and the paper reports consistent improvements over a GNN-only ablation.
Experiments on PolyBench show lower average MAPE than GNN baselines across four GNN backbones, and additional scalability validation is performed on the ForgeHLS dataset.

Abstract

High-Level Synthesis (HLS) compiles C/C++ into RTL, but exploring pragma-driven optimization choices remains expensive because each design point requires time-consuming synthesis. We propose \textbf{\DiffHLS}, a differential learning framework for HLS Quality-of-Result (QoR) prediction that learns from kernel--design pairs: a kernel baseline and a pragma-inserted design variant. \DiffHLS~encodes kernel and design intermediate-representation graphs with dedicated graph neural network (GNN) branches, and augments the delta pathway with code embeddings from a pretrained code large language model (LLM). Instead of regressing absolute targets directly, we jointly predict the kernel baseline and the design-induced delta, and compose them to obtain the design prediction. On PolyBench, \DiffHLS~attains lower average MAPE than GNN baselines under four GNN backbones, and LLM code embeddings consistently improve over a GNN-only ablation. We further validate scalability on the ForgeHLS dataset.