QLLVM: A Scalable Quantum-Classical Co-Compilation Framework based on LLVM
Yu Zhu, Qiming Du, Yuqiong Jin, Woji He, Hang Lian, Xin Zhou, Jinchen Xu, Zheng Shan
Abstract
To address the urgent need in the NISQ era for high-performance, scalable quantum compilers and to advance the integration of classical and quantum computing, we present QLLVM, an advanced Quantum-Classical co-compilation framework built on LLVM. To our knowledge, QLLVM delivers an end-to-end, LLVM-based compilation workflow that unifies the build of classical high-performance programs, including CUDA, MPI, and C++, together with quantum programs into a single executable. For quantum program compilation, QLLVM adopts a three-stage design: high-level optimizations are implemented in the MLIR Quantum dialect and then lowered to QIR, an LLVM IR-based representation, for low-level optimization and hardware mapping. Its extensible architecture and seamless interoperability with classical high-performance computing provide an efficient, flexible, industrial-grade compilation infrastructure for future quantum software development. Experimental results show that, on the MQTBench benchmark suite, QLLVM reduces circuit depth and gate counts compared with state-of-the-art compilers and demonstrates clear advantages in compiling hybrid classical-quantum programs.
AI Impact Assessments
(3 models)Scientific Impact Assessment: QLLVM — A Scalable Quantum-Classical Co-Compilation Framework based on LLVM
1. Core Contribution
QLLVM presents a unified compilation framework that integrates quantum program compilation into the LLVM/MLIR ecosystem alongside classical high-performance computing toolchains (CUDA, MPI, C++). The framework adopts a three-stage design: (1) frontend parsing of heterogeneous source files, (2) MLIR-level high-level quantum optimizations followed by lowering to QIR (Quantum Intermediate Representation) for low-level optimization and hardware-aware qubit mapping/routing, and (3) backend code generation that links all components into a single executable. The main claimed novelty is performing hardware-aware qubit mapping and routing at the LLVM IR (QIR) level—something the authors state is absent from prior LLVM/MLIR-based quantum compilers like ScaffCC and PennyLane.
2. Methodological Rigor
The methodological rigor of this paper has several notable shortcomings:
Experimental comparison is limited. The benchmarking compares QLLVM against Qiskit, Cirq, and PennyLane on gate count and circuit depth using MQTBench. However, the reported improvements over Qiskit (3.98% gate count, 3.56% depth) and Cirq (1.19% gate count, 1.61% depth) are modest and within ranges that could be sensitive to configuration details. The massive improvement over PennyLane (74-77%) likely reflects PennyLane's different design philosophy (focused on differentiable programming) rather than a meaningful compiler comparison. Notably, TKET is mentioned in the introduction but not included in benchmarks, which is a significant omission given it is one of the most competitive quantum compilers.
No statistical analysis. The paper provides aggregate percentage improvements without error bars, variance analysis, or statistical significance tests. Given the modest improvements over Qiskit and Cirq, this is a critical gap.
All-to-all topology. The experiments use "all-to-all connected logical topology," which means the SABRE-based mapping and routing—one of the key claimed contributions—is never actually tested in the benchmarks. This fundamentally undermines the claim that QLLVM performs better hardware-aware compilation.
Hybrid compilation evaluation is superficial. The hybrid classical-quantum compilation capability is demonstrated with a trivial toy example (a 2-qubit Bell circuit + simple CUDA kernel). There are no performance benchmarks of real hybrid workloads, no compilation time comparisons, and no demonstration of non-trivial quantum-classical interaction patterns.
The quantum optimization strategy (single-qubit gate fusion followed by multi-decomposition selection among ZYZ/XYX/ZXZ) is a well-known technique, not a novel contribution.
3. Potential Impact
The vision of a unified LLVM-based compilation framework for heterogeneous quantum-classical workloads addresses a genuine need. If matured, such infrastructure could:
However, the current implementation falls short of realizing this vision. The CUDA and MPI integration amounts to delegating compilation to nvcc and mpicc respectively—QLLVM acts essentially as a build orchestrator rather than performing cross-domain optimization. There is no evidence of joint classical-quantum optimization, which would be the truly impactful contribution. The quantum compilation itself uses standard, well-known techniques (SABRE routing, Euler decomposition selection).
4. Timeliness & Relevance
The paper addresses a timely topic. As quantum hardware scales and hybrid algorithms (VQE, QAOA, quantum machine learning) become more complex, the need for robust compiler infrastructure grows. The choice to build on LLVM/MLIR aligns with industry trends (Microsoft's QIR, NVIDIA's CUDA-Q). However, the paper's positioning relative to CUDA-Q is somewhat misleading—CUDA-Q already provides much of the claimed functionality with deeper integration and active industrial support.
5. Strengths & Limitations
Strengths:
Limitations:
Overall Assessment
QLLVM presents reasonable engineering work toward a unified quantum-classical compilation framework, but the paper overpromises relative to what it delivers. The benchmarking is insufficient to support the claimed advantages, the hybrid compilation is demonstrated only at the most superficial level, and the algorithmic contributions (gate fusion, SABRE routing) are implementations of existing techniques rather than novel methods. The paper would benefit substantially from: (1) benchmarks on realistic hardware topologies, (2) comparison with TKET and CUDA-Q, (3) meaningful hybrid workload evaluation, and (4) compilation time and scalability analysis.
Generated Apr 17, 2026
Comparison History (33)
Paper 2 is more likely to have higher scientific impact: it proposes and experimentally demonstrates a fundamentally different, measurement-defined description of single-particle interference with near-unity visibility and a nontrivial “three-scan equivalence” unavailable in standard interferometers. The results appear broadly relevant across quantum optics and quantum foundations, and it links multiple phenomena (CPT, EIT, diffraction) under one framework, suggesting cross-field influence and new quantum-photonic resources. Paper 1 is timely and useful for quantum software engineering, but its impact is more incremental and primarily within compiler/HPC toolchain communities.
While Paper 1 offers strong theoretical contributions and solves an open question in quantum Hamiltonians, Paper 2 presents a highly practical, scalable co-compilation framework based on LLVM. By seamlessly integrating quantum programs with classical high-performance computing architectures (CUDA, MPI), QLLVM provides foundational infrastructure that can accelerate a wide array of experimental and applied quantum research, giving it broader applicability and higher potential for immediate real-world impact in the NISQ era.
Paper 1 likely has higher scientific impact due to strong real-world applicability and timeliness: an LLVM-based end-to-end quantum–classical co-compilation stack can be adopted by HPC and quantum-software ecosystems, affecting many users and enabling hybrid workloads now in the NISQ era. Its integration with CUDA/MPI/C++ and benchmarked improvements suggest practical, scalable tooling. Paper 2 is methodologically rigorous and novel in CV resource theory with operational activation results, but its impact is more specialized to quantum information theory and may diffuse more slowly into broadly used systems.
Paper 2 likely has higher impact: it tackles a timely, widely shared bottleneck in the NISQ era (scalable quantum-classical compilation) with an engineering contribution that can be adopted broadly via the LLVM/MLIR ecosystem. Its end-to-end workflow integrating CUDA/MPI/C++ with quantum code enables real-world applications in HPC+quantum, and reported benchmark improvements support methodological rigor and practical value. Paper 1 is novel within nonequilibrium/Floquet many-body physics but is narrower in applicability and likely impacts a smaller subcommunity compared to a compiler framework that can influence many quantum software and hardware efforts.
Paper 2 presents a practical, scalable co-compilation framework that directly addresses the immediate needs of the NISQ era. By integrating quantum compilation with classical high-performance computing (CUDA, MPI) via LLVM, it provides a foundational tool for researchers and engineers. Such infrastructure tools typically have a broader and more immediate impact across multiple fields compared to the highly specialized, theoretical quantum information framework presented in Paper 1.
Paper 1 presents a foundational tool for quantum-classical co-compilation integrating with widely used frameworks (LLVM, CUDA, MPI). Its practical utility and broad applicability across quantum software development give it a significantly higher potential for widespread adoption and real-world impact compared to Paper 2, which focuses on a narrow theoretical aspect of a specific quantum error correction code family.
Paper 2 presents fundamental experimental discoveries about quantum emitters in hBN—a material of intense current interest for quantum technologies. The identification of spin signatures, spectral diffusion mechanisms, and donor-acceptor-pair recombination pathways provides new physical insights that advance understanding of solid-state quantum emitters. These findings have broad implications for quantum networking, sensing, and photonics. Paper 1, while practically useful, is primarily an engineering contribution (a compiler framework) with incremental improvements over existing tools, limiting its broader scientific impact beyond the quantum computing software community.
Paper 2 likely has higher scientific impact due to broader and more immediate real-world applicability: an end-to-end LLVM-based quantum-classical co-compilation stack affects many users, platforms, and workloads, and can be adopted by industry and HPC communities. Its benchmarking claims (depth/gate reductions on MQTBench) indicate practical performance gains and methodological evaluation. Paper 1 is theoretically novel and rigorous for photonic quantum advantage verification, but its impact is narrower (specialized to boson-sampling/LXEB theory) and less directly enabling across the wider quantum software ecosystem.
Paper 1 has higher potential scientific impact: it addresses a central open problem in quantum circuit complexity (PARITY vs QAC0) and reframes it via a Fourier-analytic characterization, adds an average-case separation between AC0 and QAC0, and connects complexity to quantum state-synthesis with a new state measure (“felinity”). These contributions are conceptually novel, broadly relevant across complexity theory and quantum information, and likely to influence future lower-bound techniques. Paper 2 is timely and useful engineering (LLVM-based co-compilation) with real-world applicability, but its impact is more incremental and domain-specific.
QLLVM addresses a critical infrastructure need for quantum-classical computing integration by building on the widely-adopted LLVM framework, providing an end-to-end compilation workflow that unifies classical HPC (CUDA, MPI, C++) with quantum programs. This has broader impact across the entire quantum computing ecosystem as a foundational tool. Paper 2, while technically sound, addresses a more specialized problem (variational state preparation for quantum metrology) with incremental improvements in a narrower domain. QLLVM's industrial-grade infrastructure approach and scalability advantages position it for wider adoption and longer-term impact.
Paper 1 introduces a novel theoretical framework connecting stabilizer codes, quantum Fisher information, and string order parameters—bridging quantum information, metrology, and condensed matter physics. This conceptual advance has broad interdisciplinary impact and addresses fundamental questions about detecting nonlocal quantum order metrologically. Paper 2, while practically useful, is primarily an engineering contribution (a compiler framework) that represents incremental improvement over existing quantum compilation tools. Its impact is more narrowly confined to the quantum software toolchain, whereas Paper 1 opens new research directions across multiple subfields.
Paper 2 introduces an industrial-grade, LLVM-based compilation framework for quantum-classical hybrid computing. While Paper 1 provides a significant fundamental theoretical result in quantum physics, Paper 2 addresses a critical and highly timely bottleneck in the NISQ era: scalable software infrastructure. By enabling seamless integration of high-performance classical and quantum code, QLLVM has the potential to become a foundational tool in the rapidly growing quantum computing industry, likely leading to widespread adoption, broad cross-disciplinary applications, and a higher overall scientific impact.
QLLVM provides a foundational, scalable software infrastructure that bridges classical high-performance computing and quantum programming. Its practical utility for developing and optimizing hybrid algorithms in the NISQ era gives it broader applicability and higher immediate real-world impact compared to the specialized, albeit rigorous, experimental physics results presented in Paper 1.
Paper 1 likely has higher scientific impact due to its broad, timely applicability to NISQ-era quantum software: an end-to-end LLVM-based quantum-classical co-compilation pipeline integrating CUDA/MPI/C++ with quantum code is a significant systems contribution with clear real-world adoption potential. It also reports benchmarked improvements versus state-of-the-art compilers, indicating methodological rigor and practical value. Paper 2 is innovative in optomechanics and could enable quantum devices, but it is purely theoretical and more specialized, with impact dependent on experimental feasibility and narrower immediate reach across fields.
While Paper 1 presents a highly rigorous and significant fundamental physics measurement, Paper 2 introduces a foundational compilation framework that bridges classical high-performance computing and quantum programming. By optimizing quantum circuits and enabling seamless hybrid execution, Paper 2 provides a scalable, industrial-grade tool. This will likely accelerate software development and algorithmic research across the entire quantum computing ecosystem in the NISQ era, offering a broader and more immediate scientific and practical impact.
Paper 2 likely has higher scientific impact: it targets an urgent, widely shared bottleneck (scalable quantum-classical compilation) with immediate applicability across NISQ hardware and HPC workflows. Building on LLVM/MLIR/QIR enables broad adoption, extensibility, and community integration, and benchmarking shows practical performance gains. Its impact spans quantum computing, compiler research, and HPC. Paper 1 is scientifically intriguing and potentially transformative for quantum metrology, but claims of surpassing the Heisenberg limit may face interpretational/assumption scrutiny and its applicability is narrower (specialized waveguide-QED platform).
QLLVM presents a novel end-to-end quantum-classical co-compilation framework built on LLVM, addressing a fundamental infrastructure need for the quantum computing community. Its integration of CUDA, MPI, C++ with quantum programs into a unified compilation pipeline is highly innovative and has broad impact potential as an industrial-grade tool. Paper 2 addresses important scheduling concerns but is more incremental, comparing existing scheduling strategies. QLLVM's contribution as foundational compiler infrastructure is likely to have deeper, longer-lasting impact across quantum software development, similar to how LLVM transformed classical compilation.
Paper 2 likely has higher scientific impact due to broad, timely relevance and strong real-world applicability: an LLVM/MLIR-based end-to-end quantum-classical co-compilation stack can be adopted across many quantum platforms and HPC workflows, influencing both academia and industry. Its reported benchmark improvements (depth/gate reductions) indicate methodological rigor and measurable benefits. Paper 1 is novel within photonic entanglement generation, but its impact is narrower (specialized linear-optical postselection schemes) and may face experimental scalability constraints, limiting cross-field uptake compared to a compiler infrastructure.
Paper 1 introduces a fundamentally novel concept—an Ising 'selector' machine that can target specific excited states rather than just ground states—which addresses an open challenge with broad implications for combinatorial optimization, Boltzmann sampling, and computational complexity characterization. This represents a conceptual breakthrough in analog computing. Paper 2, while practically useful, is primarily an engineering contribution (a compilation framework) that incrementally improves existing quantum compilation tools. Paper 1's novelty and potential to reshape how Ising machines are used gives it higher scientific impact potential.
Paper 1 likely has higher impact due to timeliness and breadth: scalable quantum-classical co-compilation on LLVM targets an urgent, widely shared bottleneck in NISQ software stacks and HPC integration. Its end-to-end workflow, interoperability with CUDA/MPI/C++, and benchmarked improvements suggest immediate real-world applicability and adoption potential across quantum computing, compilers, and HPC. Paper 2 is methodologically solid and relevant to photonic quantum state engineering, but as a theoretical study of a specific platform, its near-term impact and cross-field reach are narrower and more contingent on experimental realization.