Published November 14, 2026 | Version artifact-v1

ANEForge: Python for direct computation on the Apple Neural Engine

Authors/Creators

  • 1. The University of Texas at Austin

Description

Archival snapshot of the code, benchmark drivers, and committed result files behind A predictive, per-engine energy roofline of the Apple Neural Engine (ACM Transactions on Architecture and Code Optimization, accepted 2026).

The measurement drivers are in bench/ (device_saturation_sweep.py, device_bandwidth_roofline.py, roofline_suite.py, roofline_analysis.py, decode_measurement.py, and related scripts), and their outputs are in bench/results/.

This is not a package release, and nothing is published to PyPI. For the library, use the latest v* release. Cite the software as Bryngelson, Software: Practice and Experience (2026), doi:10.1002/spe.70124.

Notes

If you use ANEForge, please cite the paper below.

Files

sbryngelson/ANEForge-artifact-v1.zip

Files (6.7 MB)

Name Size
md5:417318797960cec41a896d3c84e2bcd4
6.7 MB Preview Download

Additional details

Related works