Generating Energy-Efficient Code via Large-Language Models – Where are we now?

This repository contains the replication package and dataset of the paper published at ICSE 2026 (research track) with the title Generating Energy-Efficient Code via Large-Language Models – Where are we now?. The preprint of the study is available here

How to cite the dataset

If this replication package is helping your research, consider to cite it is as follows, thanks!

@inproceedings{ICSE_2026,
  url = { http://www.ivanomalavolta.com/files/papers/ICSE_2026.pdf },
  numpages = { 13 },
  pages = { To appear },
  year = { 2026 },
  booktitle = { Proceedings of the ACM/IEEE 48th International Conference on Software Engineering },
  author = { Radu Apsan and Vincenzo Stoico and Michel Albonico and Rudra Dhar and Karthik Vaidhyanathan and Ivano Malavolta },
  title = {{Generating Energy-Efficient Code via Large-Language Models - Where are we now?}},
}

Overview of the replication package

The replication package contains the data folder and one folder corresponding to each subsection of section 3 of the paper. The structure is as follows:

• ./data: Contains the datasets used in the study.
• ./3.1 LLMs and coding problems selection: Contains the code corresponding to the selection of coding problems and LLMs, along with the functional solutions from the EvoEval study.
• ./3.2 Green guidelines elicitations: Contains the process of guideline selection.
• ./3.3 Code generation and implementation: Contains the code related to prompting the LLMs and testing the generated solutions.
• ./3.4 Experiment execution: Contains the configurations for replicating the experiment execution along with Experiment-Runner.
• ./3.5 Data analysis and synthesis: Contains the data analysis using the data in the folder ./data.

Replicating the data analysis

To replicate the data analysis, please follow the instructions in the folder ./3.5 Data analysis and synthesis.

Running the entire replication package

To run the replication package, please follow the instructions in each folder. The replication package is designed to be run in the order of the subsections of section 3 of the paper.