Research map / Learning theory and algorithms / Combinatorial and search optimization
Compiler Auto-Vectorization with Imitation Learning
Charith Mendis, Cambridge Yang, Yewen Pu, Dr.Saman Amarasinghe, Michael Carbin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| increase rollouts from 1 to 10 (best-of-n) (ablation) | Vemal p100 forward · Test set cost | average cost reduction: 13.3633 -> 14.9421 | Improved |
| increase rollouts from 1 to 10 (ablation) | Vemal p100 · NAS runtime | speedup over LLVM SLP: 0.979 -> 1.015 | Improved |
| use forward instead of backward traversal (ablation) | Vemal p100, 10 rollouts · Test set cost | average cost reduction: 11.3227 -> 14.9421 | Improved |
| partition size 200 instead of 100 (ablation) | Vemal forward 10 rollouts · Test set cost | average cost reduction: 14.9421 -> 14.7340 | No clear effect |
| replace LLVM SLP with learnt Vemal p100 10 rollouts | vectorization · Test set cost | average cost reduction: 12.1872 -> 14.9421 | Improved |
| replace goSLP-p100 ILP with learnt imitation policy | Vemal p100 10 rollouts · Test set cost | average cost reduction: 19.4815 -> 14.9421 | Hurt |
| replace LLVM SLP with Vemal p100 10 rollouts | vectorization · NAS runtime | speedup over LLVM SLP: 1.015x | Improved |
| replace random packing with learnt policy | Vemal p100 forward 1 rollout · Test set cost | average cost reduction: 1.0320 -> 13.3633 | Improved |
Closest papers by meaning
- Efficiently Vectorized MCMC on Modern Accelerators (ICML 2025 · similarity 0.81)
- Compiler-R1: Towards Agentic Compiler Auto-tuning with Reinforcement Learning (NeurIPS 2025 · similarity 0.81)
- UniAP: Unifying Inter- and Intra-Layer Automatic Parallelism by Mixed Integer Quadratic Programming (CVPR 2025 · similarity 0.80)
- REASONING COMPILER: LLM-Guided Optimizations for Efficient Model Serving (NeurIPS 2025 · similarity 0.80)
- Dynamic Operator Optimization for Efficient Multi-Tenant LoRA Model Serving (AAAI 2025 · similarity 0.80)
- QiMeng-NeuComBack: Self-Evolving Translation from IR to Assembly Code (NeurIPS 2025 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
