发表机构
Delft University of Technology（代尔夫特理工大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
QuWARP提出工作负载级复用规划器，通过识别共享前缀和物化边界工件，在量子电路重复运行仿真中实现2.95-32.84倍加速，并保证决策可审计。
AI 中文摘要
量子电路仿真通常表现为重复运行的工作负载，而非孤立的电路：变分量子本征求解器（VQE）扫描、噪声多次采样研究和量子纠错（QEC）周期在多次运行中反复访问紧密相关的结构。现有仿真器能很好地优化单次执行，但很大程度上忽略了跨任务的共享状态，因此重复了本可安全复用的工作。我们提出QuWARP，一种基于规划器的工作负载优化器，适用于最先进的受限仿真器执行表面：它对相关任务执行工作负载级规划，识别共享前缀，并选择何时物化精确类型的边界工件，以便在评估的态矢量模式和稳定子混合模式中后续复用。其规划器将延续合法性视为狭窄的正确性护栏，在复用无利可图时弃权（不执行），并通过EXPLAIN风格的轨迹（即带有来源和实际成本摘要的可检查规划器报告）使每次复用、弃权或拒绝决策可审计。在真实世界的量子应用工作负载上，QuWARP在复用正向工作负载上比本工作的主要直接逐任务Qrack基线实现了2.95倍至32.84倍的加速。这些结果表明，工作负载级复用规划改善了重复运行的仿真，同时将不支持的交接保持可审计且不进入执行路径。
英文摘要
Quantum circuit simulation often appears as repeated-run workloads rather than isolated circuits: variational quantum eigensolver (VQE) sweeps, noisy multishot studies, and quantum error correction (QEC) cycles revisit closely related structure across many runs. Existing simulators optimise individual executions well, but they largely ignore cross-task shared state and therefore repeat work that could be reused safely. We propose QuWARP, a planner-based workload optimiser for a bounded state of the art simulator execution surface: it performs workload-level planning over related tasks, identifies shared prefixes, and chooses when to materialize exact typed boundary artifacts for later reuse across the evaluated statevector mode, and stabilizer-hybrid mode. Its planner treats continuation legality as a narrow correctness guardrail, abstains when reuse is unprofitable, and keeps each reuse, abstention, or refusal decision auditable through EXPLAIN-style traces, meaning inspectable planner reports with provenance and realized-cost summaries. Across real world quantum application workloads QuWARP delivers 2.95x-32.84x speedups over this work's main direct per-task Qrack denominator on reuse-positive workloads. These results show workload-level reuse planning improves repeated-run simulation while keeping unsupported handoffs auditable and out of the execution path.
CommentsAccepted for presentation at IEEE QCE 2026 and publication in the Proceedings of the IEEE International Conference on Quantum Computing and Engineering (QCE) on the Quantum System Software Paper Track
