pith. sign in

arxiv: 2604.14248 · v1 · submitted 2026-04-15 · 💻 cs.SE

Graph-Based ECO and Patch Generation for High-Level Synthesis

Alireza Azadi , Paul Rigge , Ethan Mahintorabi , Kenneth B. Kent This is my paper

Pith reviewed 2026-05-10 13:26 UTC · model grok-4.3

classification 💻 cs.SE
keywords high-level synthesisengineering change ordersgraph edit distancepatch generationintermediate representationschedule preservationstructural reuse
0
0 comments X

The pith

Graph edit distance on intermediate representations generates patches for late-stage changes in high-level synthesis that preserve original schedules and semantics.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper develops a graph-based method to support Engineering Change Orders in high-level synthesis using the XLS tool. It computes graph edit distance between the original and revised versions of a design's intermediate representation to locate structural differences. Those differences are mapped to a sequence of patch operations that can be applied while obeying XLS IR rules. The patches also incorporate a schedule constraining scheme to keep the original pipeline registers in place. This matters because current HLS flows make late modifications expensive, and a working approach would let designers revise designs incrementally without full resynthesis.

Core claim

Differences between an original and revised XLS intermediate representation are detected using a Graph Edit Distance algorithm. These differences are transformed into patch operations that enforce XLS IR constraints and preserve semantic correctness. A schedule constraining scheme maintains the original pipeline registers. Experiments across several XLS designs demonstrate high structural reuse ratios, effective schedule preservation, and full functional correctness.

What carries the argument

Graph Edit Distance algorithm applied to the IR graphs, which identifies structural differences that are then mapped to patch operations under a mechanism that enforces XLS IR constraints while a schedule constraining scheme preserves pipeline registers.

If this is right

  • Designers achieve high structural reuse when applying changes to existing HLS designs.
  • The original pipeline schedule and registers remain unchanged after patch application.
  • Full functional correctness is retained for the modified design.
  • The method supports arbitrary late-stage changes in production XLS flows.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar graph comparison and patching could be adapted to other compiler or synthesis IRs outside XLS.
  • The technique might shorten hardware design cycles by allowing more frequent incremental updates.
  • Combining the patches with timing analysis tools could help catch new violations early.
  • Automated ECO systems in HLS might incorporate this as a core step to reduce manual rework.

Load-bearing premise

That differences found by graph edit distance can always be turned into patches that satisfy XLS IR constraints and keep semantic correctness for any late-stage design change.

What would settle it

A concrete design modification where the generated patch either breaks an XLS IR constraint, changes the original schedule, or produces a functionally incorrect result on simulation.

Figures

Figures reproduced from arXiv: 2604.14248 by Alireza Azadi, Ethan Mahintorabi, Kenneth B. Kent, Paul Rigge.

Figure 1
Figure 1. Figure 1: Illustration of IR patching: (a) original IR implementing [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the XLS ECO Flow. Components with [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Hierarchical structure of the patch protobuf. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Fraction of reuse (substitutions and unchanged struc [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Resource usage during ECO patch generation across [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Schedule preservation across designs: (a) Percentage [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
read the original abstract

High-level synthesis (HLS) tools offer limited support for Engineering Change Orders (ECOs), making late-stage design modifications challenging and costly. This paper introduces a graph-based ECO methodology tailored for Google XLS. A Graph Edit Distance (GED) algorithm is used to detect structural differences between original and revised intermediate representations (IRs), which are then transformed into patch operations. A patch application mechanism is developed to enforce XLS IR constraints while preserving semantic correctness, together with a schedule constraining scheme that maintains the original pipeline registers. Experiments across several XLS designs demonstrate high structural reuse ratios, effective schedule preservation, and full functional correctness, highlighting the practicality of the approach for production HLS flows.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper introduces a graph-based ECO methodology for Google XLS in high-level synthesis. It uses Graph Edit Distance (GED) to detect differences between original and revised IRs, transforms them into patches, applies the patches enforcing IR constraints and preserving semantics, and uses a schedule constraining scheme to maintain pipeline registers. Experiments on several XLS designs are claimed to show high structural reuse, schedule preservation, and full functional correctness, demonstrating practicality for production HLS flows.

Significance. If the experimental claims hold, this could significantly impact HLS design flows by enabling efficient handling of late-stage changes, reducing costs associated with ECOs. The graph-based approach leveraging GED and custom patching with schedule preservation offers a structured solution to a practical problem. The reported full functional correctness and high reuse ratios, if quantitatively validated, would provide strong evidence for its utility in production settings.

major comments (2)
  1. The abstract reports positive results on reuse, schedule preservation, and correctness but supplies no quantitative metrics, baselines, design details, or discussion of failure cases. This makes it difficult to evaluate the soundness of the central claim that the approach is practical for production HLS flows.
  2. The description of transforming GED differences into patches that enforce XLS IR constraints and preserve semantic correctness for arbitrary changes lacks specifics on the mechanism's completeness, edge cases (e.g., pipeline stages or new dependencies), or guarantees against introducing bugs or timing violations. This is critical as the practicality rests on this transformation always succeeding.
minor comments (1)
  1. The abstract could be strengthened by including at least one key quantitative result to illustrate the 'high structural reuse ratios'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful and constructive review. The comments highlight important areas for improving clarity and completeness. We address each major comment below and have revised the manuscript to incorporate additional details and quantitative information where feasible.

read point-by-point responses

  1. Referee: The abstract reports positive results on reuse, schedule preservation, and correctness but supplies no quantitative metrics, baselines, design details, or discussion of failure cases. This makes it difficult to evaluate the soundness of the central claim that the approach is practical for production HLS flows.

    Authors: We agree that the abstract would be strengthened by including key quantitative results. In the revised manuscript, we have updated the abstract to report specific metrics from our experiments, including average structural reuse ratios exceeding 85% across designs, full schedule preservation in all tested cases, and 100% functional correctness verified via equivalence checking. We have also briefly noted the benchmark designs used (several representative XLS modules). Detailed baselines (comparison to full resynthesis), design characteristics, and discussion of any edge cases or failure modes are expanded in Section 5 (Experiments), as abstract length limits preclude full coverage there. revision: yes

  2. Referee: The description of transforming GED differences into patches that enforce XLS IR constraints and preserve semantic correctness for arbitrary changes lacks specifics on the mechanism's completeness, edge cases (e.g., pipeline stages or new dependencies), or guarantees against introducing bugs or timing violations. This is critical as the practicality rests on this transformation always succeeding.

    Authors: We acknowledge the need for greater specificity on the patch transformation process. In the revised manuscript, we have substantially expanded Section 4 to include: (1) a step-by-step description of mapping GED operations to XLS IR patches while enforcing constraints such as type consistency and dataflow validity; (2) explicit handling of edge cases including modifications to pipeline stages (via schedule constraining to avoid register insertion/deletion) and introduction of new dependencies (resolved by minimal rescheduling or fallback to targeted resynthesis); and (3) discussion of guarantees, noting that semantic correctness is preserved through post-application equivalence verification and that timing violations are prevented by the original schedule preservation scheme. We clarify that the method does not claim to succeed for arbitrary changes in all possible scenarios; instead, it reports full success on the evaluated benchmarks, with a new subsection on potential failure modes and mitigation strategies. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected in the graph-based ECO derivation

full rationale

The paper's chain proceeds from standard Graph Edit Distance computation on XLS IRs to patch transformation and application, with added schedule constraints. These are algorithmic steps presented as novel engineering, backed by experimental results on multiple designs rather than any self-referential definition, fitted parameter renamed as prediction, or load-bearing self-citation. No equations or uniqueness theorems are shown that collapse the output to the input by construction. The methodology remains self-contained against external benchmarks and does not invoke prior author work to force its own premises.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on domain assumptions about graph representations of HLS IRs and the feasibility of constraint-preserving edits; no free parameters or new entities are introduced.

axioms (2)
  • domain assumption Graph Edit Distance accurately captures structural differences between original and revised HLS IRs
    Invoked as the detection mechanism in the methodology.
  • domain assumption Detected edits can be mapped to patches that enforce XLS IR constraints while preserving semantics
    Core premise of the patch application mechanism.

pith-pipeline@v0.9.0 · 5411 in / 1234 out tokens · 40948 ms · 2026-05-10T13:26:25.044700+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages

  1. [1]

    Xls eco results repository,

    A. Azadi, “Xls eco results repository,” https://github.com/alirezazd/ xls-eco/tree/eco-paper-2025, 2025, accessed: 2025-09-08

    work page 2025

  2. [2]

    Cramming more components onto integrated circuits,

    G. Moore, “Cramming more components onto integrated circuits,” Proceedings of the IEEE, vol. 86, no. 1, pp. 82–85, 1998

    work page 1998

  3. [3]

    High-level synthesis: Past, present, and future,

    G. Martin and G. Smith, “High-level synthesis: Past, present, and future,” IEEE Design & Test of Computers, vol. 26, no. 4, pp. 18–25, 2009

    work page 2009

  4. [4]

    An introduction to high-level synthesis,

    P. Coussy, D. D. Gajski, M. Meredith, and A. Takach, “An introduction to high-level synthesis,”IEEE Design & Test of Computers, vol. 26, no. 4, pp. 8–17, 2009

    work page 2009

  5. [5]

    Parallel Programming for FPGAs,

    R. Kastner, J. Matai, and S. Neuendorffer, “Parallel Programming for FPGAs,”ArXiv e-prints, May 2018

    work page 2018

  6. [6]

    Google xls: High-level synthesis framework,

    Google, “Google xls: High-level synthesis framework,” 2024, accessed: 2025-09-08. [Online]. Available: https://google.github.io/xls

    work page 2024

  7. [7]

    Pytfhe: An end-to-end compilation and execution framework for fully homomorphic encryption applications,

    J. Ma, C. Xu, and L. W. Wills, “Pytfhe: An end-to-end compilation and execution framework for fully homomorphic encryption applications,” in2023 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS), 2023, pp. 24–34

    work page 2023

  8. [8]

    Accelerating digital block design with google’s xls,

    Antmicro, “Accelerating digital block design with google’s xls,” 2023, accessed: 2025-09-08. [Online]. Available: https://antmicro.com/blog/ 2023/09/accelerating-digital-block-design-with-googles-xls/

    work page 2023

  9. [9]

    Subgraph extraction- based feedback-guided iterative scheduling for hls,

    H. Ye, D. Z. Pan, C. Leary, D. Chen, and X. Xu, “Subgraph extraction- based feedback-guided iterative scheduling for hls,” in2024 Design, Automation & Test in Europe Conference & Exhibition (DATE), 2024, pp. 1–6

    work page 2024

  10. [10]

    An exact graph edit distance algorithm for solving pattern recognition problems,

    Z. Abu-Aisheh, R. Raveaux, J.-Y . Ramel, and P. Martineau, “An exact graph edit distance algorithm for solving pattern recognition problems,” inProceedings of the 4th International Conference on Pattern Recogni- tion Applications and Methods, Lisbon, Portugal, January 2015

    work page 2015

  11. [11]

    Naval Research Logistics Quarterly2(1–2), 83–97 (1955) https://doi.org/10.1002/nav

    H. W. Kuhn, “The hungarian method for the assignment problem,” Naval Research Logistics Quarterly, vol. 2, no. 1-2, pp. 83–97, 1955. [Online]. Available: https://onlinelibrary.wiley.com/doi/abs/10.1002/nav. 3800020109

    work page doi:10.1002/nav 1955

  12. [12]

    [Online]

    NetworkX Developers,NetworkX: optimize edit paths, NetworkX Project, 2025, online; accessed 8 September 2025. [Online]. Avail- able: https://networkx.org/documentation/latest/reference/algorithms/ generated/networkx.algorithms.similarity.optimize edit paths.html

    work page 2025

  13. [13]

    Incremental high-level synthesis,

    L. Lavagno, A. Kondratyev, Y . Watanabe, Q. Zhu, M. Fujii, M. Tatesawa, and N. Nakayama, “Incremental high-level synthesis,” in2010 15th Asia and South Pacific Design Automation Conference (ASP-DAC), 2010, pp. 701–706

    work page 2010

  14. [14]

    Data-path aware high-level eco synthesis,

    M. Shiroei, B. Alizadeh, and M. Fujita, “Data-path aware high-level eco synthesis,”Integration, vol. 65, pp. 88–96, 2019. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0167926018303766

    work page 2019

  15. [15]

    High-level eco via programmable datapath and smt,

    R.-Y . Wang, C.-C. Pai, J.-J. Wang, H.-T. Wen, Y .-C. Pai, Y .-W. Chang, J. C.-M. Li, and J.-H. R. Jiang, “High-level eco via programmable datapath and smt,” inProceedings of the IEEE International Symposium on Circuits and Systems (ISCAS), 2019, pp. 1–5

    work page 2019

  16. [16]

    Automatic correction of rtl designs using a lightweight partial high level synthesis,

    B. Alizadeh and M. Shiroei, “Automatic correction of rtl designs using a lightweight partial high level synthesis,”Integration, vol. 91, pp. 173–181, 2023. [Online]. Available: https://www.sciencedirect.com/ science/article/pii/S0167926023000573

    work page 2023

  17. [17]

    Cost-aware patch generation for multi- target function rectification of engineering change orders,

    H.-T. Zhang and J.-H. R. Jiang, “Cost-aware patch generation for multi- target function rectification of engineering change orders,” in2018 55th ACM/ESDA/IEEE Design Automation Conference (DAC), 2018, pp. 1–6

    work page 2018

  18. [18]

    Efficient computation of eco patch functions,

    A. Q. Dao, N.-Z. Lee, L.-C. Chen, M. P.-H. Lin, J.-H. R. Jiang, A. Mishchenko, and R. Brayton, “Efficient computation of eco patch functions,” inProceedings of the 55th Annual Design Automation Conference, ser. DAC ’18. New York, NY , USA: Association for Computing Machinery, 2018. [Online]. Available: https://doi.org/10.1145/3195970.3196039

    work page doi:10.1145/3195970.3196039 2018

  19. [19]

    Resource-aware functional eco patch generation,

    A.-C. Cheng, H.-R. Jiang, and J.-Y . Jou, “Resource-aware functional eco patch generation,” inProceedings of the Design, Automation & Test in Europe Conference & Exhibition (DATE), 2016, pp. 1037–1042. IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS 10

    work page 2016

  20. [20]

    Comprehensive search for eco rectification using symbolic sampling,

    V . N. Kravets, N.-Z. Lee, and J.-H. R. Jiang, “Comprehensive search for eco rectification using symbolic sampling,” inProceedings of the 56th Annual Design Automation Conference (DAC), 2019, pp. 71:1–71:6

    work page 2019

  21. [21]

    Deltasyn: An efficient logic difference optimizer for eco synthesis,

    S. Krishnaswamy, H. Ren, N. Modi, and R. Puri, “Deltasyn: An efficient logic difference optimizer for eco synthesis,” in2009 IEEE/ACM Inter- national Conference on Computer-Aided Design - Digest of Technical Papers, 2009, pp. 789–796

    work page 2009

  22. [22]

    Z3: An efficient SMT solver,

    L. de Moura and N. Bjørner, “Z3: An efficient SMT solver,” inTools and Algorithms for the Construction and Analysis of Systems (TACAS), ser. Lecture Notes in Computer Science, vol. 4963. Springer, 2008, pp. 337–340

    work page 2008

  23. [23]

    Eco with stratus hls and the digital implementation flow,

    D. Pursley, “Eco with stratus hls and the digital implementation flow,”Cadence Community Blog, 2018, accessed: 2025-09-08. [Online]. Available: https://community.cadence.com/cadence blogs 8/b/di/posts/ eco-with-stratus-hls

    work page 2018

  24. [24]

    Cadence announces stratus high-level synthesis platform,

    Cadence Design Systems, “Cadence announces stratus high-level synthesis platform,” https://www.prnewswire.com/news-releases/ cadence-announces-stratus-high-level-synthesis-platform-300038873. html, 2015, press release, accessed 2025-09-08

    work page 2015

  25. [25]

    Catapult 8 a major hls upgrade,

    M. Dunn, “Catapult 8 a major hls upgrade,”EDN Network, 2014, accessed: 2025-09-08. [Online]. Available: https://www.edn.com/ catapult-8-a-major-hls-upgrade/

    work page 2014

  26. [26]

    Rapid recompile,

    Intel Corporation, “Rapid recompile,” https://www.intel.com/content/ www/us/en/software/programmable/quartus-prime/rapid-recompile. html, 2023, accessed: 2025-09-08

    work page 2023

  27. [27]

    Enable faster design iterations with incremental compile and abstract shell in vivado,

    AMD Xilinx, “Enable faster design iterations with incremental compile and abstract shell in vivado,” https://www.amd.com/en/products/ software/adaptive-socs-and-fpgas/vivado/faster-design-iterations.html, 2023, accessed: 2025-09-08

    work page 2023

  28. [28]

    Xls: Z3-based ir equivalence checker,

    Google, “Xls: Z3-based ir equivalence checker,” https://github.com/ google/xls/blob/main/xls/dev tools/check ir equivalence main.cc, 2024, accessed: 2025-09-08

    work page 2024

  29. [29]

    A parallel graph edit distance algorithm,

    Z. Abu-Aisheh, R. Raveaux, J.-Y . Ramel, and P. Martineau, “A parallel graph edit distance algorithm,”Expert Systems with Applications, vol. 94, pp. 41–57, 2018. [Online]. Available: https: //www.sciencedirect.com/science/article/pii/S095741741730725X

    work page 2018

  30. [30]

    Gedlib: A c++ library for graph edit distance computation,

    D. B. Blumenthal, S. Bougleux, J. Gamper, and L. Brun, “Gedlib: A c++ library for graph edit distance computation,” inGraph-Based Representations in Pattern Recognition, D. Conte, J.-Y . Ramel, and P. Foggia, Eds. Cham: Springer International Publishing, 2019, pp. 14–24

    work page 2019

  31. [31]

    Computing graph edit distance via neural graph matching,

    C. Piao, T. Xu, X. Sun, Y . Rong, K. Zhao, and H. Cheng, “Computing graph edit distance via neural graph matching,”Proc. VLDB Endow., vol. 16, no. 8, p. 1817–1829, Apr. 2023. [Online]. Available: https://doi.org/10.14778/3594512.3594514

    work page doi:10.14778/3594512.3594514 2023

  32. [32]

    Graph Edit Distance with General Costs Using Neural Set Divergence,

    E. Jain, I. Roy, S. Meher, S. Chakrabarti, and A. De, “Graph Edit Distance with General Costs Using Neural Set Divergence,” inThe Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024. [Online]. Available: https://openreview.net/forum?id= u7JRmrGutT

    work page 2024

  33. [33]

    Combinatorial learning of graph edit distance via dynamic embedding,

    R. Wang, T. Zhang, T. Yu, J. Yan, and X. Yang, “Combinatorial learning of graph edit distance via dynamic embedding,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2021, pp. 5241–5250

    work page 2021