REVIEW 13 cited by
Large Language Models for Compiler Optimization
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
We explore the novel application of Large Language Models to code optimization. We present a 7B-parameter transformer model trained from scratch to optimize LLVM assembly for code size. The model takes as input unoptimized assembly and outputs a list of compiler options to best optimize the program. Crucially, during training, we ask the model to predict the instruction counts before and after optimization, and the optimized code itself. These auxiliary learning tasks significantly improve the optimization performance of the model and improve the model's depth of understanding. We evaluate on a large suite of test programs. Our approach achieves a 3.0% improvement in reducing instruction counts over the compiler, outperforming two state-of-the-art baselines that require thousands of compilations. Furthermore, the model shows surprisingly strong code reasoning abilities, generating compilable code 91% of the time and perfectly emulating the output of the compiler 70% of the time.
Forward citations
Cited by 13 Pith papers
-
JETO-Bench: A Reproducible Benchmark for Execution Time Improvement Patches in Java
JETO-Mine is a reusable three-phase pipeline that mines 1.8 million Java commits to produce JETO-Bench containing 91 verified executable ETIPs, on which OpenHands succeeds at 14.3%.
-
Can Coding Agents Implement Missed Compiler Optimizations? Evaluating LLM Agents on LLVM Peephole Optimizations
No evaluated coding agent simultaneously matches human experts on behavioral validity and profitability when implementing real missed InstCombine optimizations from LLVM issues.
-
AtumAI: A Principled Framework for Agentic Generation of Datacenter Control-Plane Policies
AtumAI compiles plain-language goals into a formal search problem and uses an LLM-plus-evolution loop to generate datacenter control-plane policies that beat hand-tuned baselines in simulation.
-
WarmTuner: Program-Specific Warm Starts for Compiler Autotuning via Offline-to-Online Reinforcement Learning
WarmTuner learns a source-code-conditioned flag policy offline and refines it online with GRPO, reporting 1.732x average speedup over GCC -O3 but not a significant gain over GroupTuner.
-
AgentCompile: An LLM-Guided Compiler for Direct CUDA Inference
A hybrid LLM-guided CUDA compiler for transformer decode reports 3.2-6.6x over PyTorch eager and 1.02-1.12x over vLLM, but only for single-request, fixed-shape generation.
-
Bridging the Gap in Ophthalmic AI: MM-Retinal-Reason Dataset and OphthaReason Model toward Dynamic Multimodal Reasoning
A new retinal-imaging multimodal dataset and an ophthalmology-specific reasoning model claim state-of-the-art gains of 15 to 25 percent over existing medical and general multimodal LLMs.
-
Re-thinking Memory-Bound Limitations in CGRAs
Adding a cache hierarchy, CGRA-specific runahead prefetching, and per-PE cache reconfiguration allows CGRAs to process irregular memory-access kernels at near-ideal speed with 1.27% of the storage.
-
CompilerGPT: Leveraging Large Language Models for Analyzing and Acting on Compiler Optimization Reports
An open-source framework in which GPT-4o or Claude reads Clang/GCC optimization reports and iteratively rewrites C++ code, achieving up to 6.5x speedups on some benchmarks but not consistently.
-
Compiler-R1: Towards Agentic Compiler Auto-tuning with Reinforcement Learning
A two-stage SFT+RL pipeline trains LLM agents to call compiler-analysis tools and select pass sequences, achieving 8.46% mean IR instruction reduction, but the tool's own contribution is not controlled.
-
LEGO-Compiler: Enhancing Neural Compilation Through Translation Composability
LEGO-Compiler decomposes C functions into control blocks, translates each block to assembly with a shared symbol table, and reassembles them, reporting over 99% pass rates on ExeBench and near order-of-magnitude scala...
-
Agentic CPU-GPU Scheduling for Heterogeneous AI Workloads
An LLM plus a runtime measurer picked the brute-force-optimal CPU/GPU/queue mapping in all 13 evaluated four-tool workflows, though probe costs were excluded and the prompt was tuned on these scenarios.
-
Position: Quantum Program Generation Must Prioritize Validity Over Probabilistic Scaling
The paper argues that probabilistic scaling alone cannot fix the validity gap in quantum circuit generation, so quantum code assistants must build verification into generation rather than filter outputs after the fact.
-
Let's Revise Step-by-Step: A Unified Local Search Framework for Code Generation with LLMs
The abstract claims a new local search framework for code generation, but the manuscript body is a different mathematics paper.
Discussion (0). Continue with ORCID to comment.