ArchAgent v2: A Case Study with the Data Prefetching Championship
ArchAgent v2 automatically found a three-level data prefetcher that beat the prior DPC4 champion under the same rules.
The paper says the system adds cascaded evolutionary search across cache levels and a hardware-size feedback loop during evolution. Its discovered policy posts a 3.8% geometric mean IPC speedup over baseline, and 0.3% over BertiGO. The gains are stronger in low-bandwidth single-core settings, where it reports 4.6% versus BertiGO’s 2.6%. Multi-core evolution remains limited by simulation latency. ArXiv · AI/CL/LG's note
The paper says the system adds cascaded evolutionary search across cache levels and a hardware-size feedback loop during evolution. Its discovered policy posts a 3.8% geometric mean IPC speedup over baseline, and 0.3% over BertiGO. The gains are stronger in low-bandwidth single-core settings, where it reports 4.6% versus BertiGO’s 2.6%. Multi-core evolution remains limited by simulation latency. ArXiv · AI/CL/LG's note
score 5