KernelArc: A Multi-Agent Framework for GPU Kernel Optimization

KernelArc is introduced as a multi-agent framework for autonomous GPU kernel optimization across heterogeneous workloads, utilizing strategy-specialized agents and shared memory. It was evaluated on NVIDIA H100 and B200 GPUs using SOL-ExecBench. Simultaneously, PTXBench provides a benchmark to evaluate LLMs using architecture-specific PTX for GPU kernel optimization. Testing on H100 and B200 GPUs reveals that LLM success rates drop on complex attention backward workloads, with no model consistently matching frontier libraries across the suite.