Cross-Route Dataset for QoR Surrogate Modeling in High-Level Synthesis: Synthetic, HLSynCSV, and Bambu Tiny-n

This dataset accompanies the manuscript \u201cCross-Route Evaluation of Surrogate Models for QoR Prediction in High-Level Synthesis.\u201d It provides all data and scripts necessary to reproduce the experiments reported in the paper. The dataset integrates three complementary routes for quality-of-result (QoR) prediction in high-level synthesis (HLS):Synthetic generator: 480 toy samples with clean ground-truth values, used to validate surrogate correctness under idealized conditions.Benchmark-derived corpus: 13,493 design points extracted from the HLSyn benchmark, capturing large-scale but noisy measurements.Bambu HLS trials: six valid design points obtained from direct invocations of the open-source Bambu HLS tool, representing the tiny-$n$ regime with realistic tool constraints.The dataset includes raw CSV files, feature extraction utilities, and training scripts for surrogate modeling using multilayer perceptrons (MLPs). It also provides code for evaluation metrics, including error, correlation, ranking, and cost-aware constrained next-point (CNP@budget) analysis. All experiments can be executed in a Google Colab environment with modest CPU resources, ensuring accessibility and reproducibility.By making these resources publicly available, this dataset establishes a baseline for future research on surrogate-assisted design space exploration in HLS. It enables researchers to replicate our results, compare alternative models, and extend the analysis to new surrogate learning strategies under both abundant and scarce data regimes.