Hybrid Graph Representation and Learning Framework for High-Level Synthesis Design Space Exploration
Bibliographic record
Abstract
Optimizing hardware accelerators in high-level synthesis (HLS) relies on design space exploration (DSE), which involves experimenting with different pragma options and trading off hardware cost and performance metrics (HCPMs) to identify Pareto-optimal solutions. The exponential growth of the design space, poor quality-of-results (QoR) estimation by HLS tools, and lengthy post-implementation runtime have made the HLS DSE process highly challenging and time-consuming. Automating this process could reduce time-to-market and associated development costs. Learning-based methods, particularly graph neural networks (GNNs), have shown considerable potential in addressing HLS QoR/DSE problems by modeling the mapping function from control data flow graphs (CDFGs) of HLS designs to their logic, enabling early estimation of QoR during the compilation phase of the hardware design flow. However, there is still a gap in terms of their prediction accuracy. Indeed, modeling HLS-related problems using GNNs that efficiently capture the complex patterns arising from applied pragmas and low-level characteristics of HLS specifications is challenging. This paper introduces a novel hybrid graph representation and learning framework under a multi-task setting, featuring two distinct types of CDFGs derived from two different sources. Furthermore, various models are proposed to fuse features and knowledge in joint, sequential, and parallel learning architectures, aiming to improve the overall accuracy and generalization in predicting QoR and approximating the Pareto frontier (PF). Experimental results show that our framework can attain a higher level of performance than the state-of-the-art baseline models over unseen designs, with an average relative improvement of 47.4 % and 16.0 % for resource utilization and performance metrics, respectively. Additionally, considering various HLS designs with different design space sizes, a 26.8 % enhancement in DSE PF approximation is observed.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".