High-performance kernels underpin efficient accelerator execution but require expert tuning and lengthy manual optimization cycles. LLM coding agents promise…
机构:复旦
来源:arXiv 2609.35568 | AI4Papers 论文推荐平台