摘要
Quantum error correction (QEC) via error decoding is essential towards fault-tolerant quantum computing, but extremely costly to simulate. However, existing simulators are often CPU-based and suffer from severe performance bottlenecks, especially when scaling to extensive simulations with large code distances or low physical error rate. In this work, we introduce Syndrilla, a PyTorch-based across-platform QEC simulation framework. The framework is highly modular, allowing flexible integration and customization of the error model, syndrome model, decoding configuration (algorithm and data format), and logical check type. Experimental results show that Syndrilla achieves 10 × ∼ 20 × speedup over CPU, when running on both AMD and NVIDIA GPUs, and decoding data format does not degrade accuracy, demonstrating the practicality and efficiency of Syndrilla to instigate future QEC research.