cid f675e8e8416ba59eacb08d668d99d2f6c1803ac5c1f2885aa16e8771e58f3bd0 · original · open access · via openalex
Title: tchanah/chipyard: IEEE SBAC-PAD 2026 — Artifact Snapshot Authors: Chanaka Hettige; Martin Swany A Venue: Zenodo (CERN European Organization for Nuclear Research) Year: 2026 Canonical: doi:10.5281/zenodo.22049730 URL: https://doi.org/10.5281/zenodo.22049730 Collector: openalex OA: true ABSTRACT Snapshot of the collective-communication network accelerator and its PyTorch/Gloo integration used to produce the results reported in Offloading Floating Point based Collective Operations to Programmable Logic for Distributed Learning, IEEE SBAC-PAD 2026. This artifact spans three repositories, all tagged IEEE-SBAC-PAD-2026. All three are needed together for a working build: tchanah/chipyard @ IEEE-SBAC-PAD-2026 tchanah/icenet @ IEEE-SBAC-PAD-2026 tchanah/gloo @ IEEE-SBAC-PAD-2026 Environment (global, host octo250) Ubuntu 20.04.6 LTS (Focal Fossa), kernel 5.4.0-233-generic, x86_64 System compiler: gcc 9.4.0 Vivado 2021.1 (required for the Xilinx Alveo U250) PyTorch (not a fork — informational, for reproducing the build environment) Version: 2.10.0a0 (source checkout, untagged trunk snapshot between the 2.9 and 2.10 releases) Branch main, tracking pytorch/pytorch, commit a273475b01e912f402378a522bb9c4ed37e8413a (2025-09-19) Working tree carries exactly two intentional local edits: .gitmodules and the third_party/gloo pin (both point at the tchanah/gloo fork above — expected, not drift) Build & run instructions Chipyard/FireSim build & simulation workflow: Complete Workflow for Building and Simulating a New Module PyTorch/Gloo integration guide: Testing with PyTorch