FastTrackGPU: Static-Analysis–Guided Analytical Modeling for Softcore GPUs
IEEE Computer Architecture Letters · 2026-01 · 25 (1) · 162-165
DOI: 10.1109/lca.2026.3677349 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Computer Architecture Letters, 2026-01. Verfasst von Chihyo Ahn und 8 weiteren Autor:innen. Die Arbeit nennt 7 Literaturangaben. Gefördert von: NSF, Booz Allen Hamilton.
Abstract (Originalsprache)
SoftGPUs make FPGA platforms practical data-parallel accelerators by providing a GPU-like programming model on top of an open RTL-to-software stack. However, their high configurability creates a large configuration space, making it difficult to identify workload-optimal designs under FPGA resource constraints. We address this challenge by proposing the first analytical model for softGPUs and four decision policies that aggressively prune the design space. Implemented in our FastTrackGPU framework, our lightweight analytical model driven by static analysis enables orders-of-magnitude faster configuration selection than Bayesian optimization while boosting kernel throughput by 1.55× (geomean) over area-greedy baselines.
Förderung
- NSF
- Booz Allen Hamilton
Zitieren
APA
Ahn, C., Jeong, S., Cooper, L. P., Han, R., Pu, H., Parnenzini, N., Zhao, J., Tine, B., & Kim, H. (2026). FastTrackGPU: Static-Analysis–Guided Analytical Modeling for Softcore GPUs. IEEE Computer Architecture Letters, 25(1), 162-165. https://doi.org/10.1109/lca.2026.3677349
Vancouver
Ahn C, Jeong S, Cooper LP, Han R, Pu H, Parnenzini N, et al. FastTrackGPU: Static-Analysis–Guided Analytical Modeling for Softcore GPUs. IEEE Computer Architecture Letters. 2026;25(1):162-165. doi:10.1109/lca.2026.3677349
BibTeX
@article{Ahn2026,
author = {Ahn, Chihyo and Jeong, Shinnung and Cooper, Liam Paul and Han, Ruobing and Pu, Huanzhi and Parnenzini, Nicholas and Zhao, Jisheng and Tine, Blaise and Kim, Hyesoon},
title = {FastTrackGPU: Static-Analysis–Guided Analytical Modeling for Softcore GPUs},
journal = {IEEE Computer Architecture Letters},
year = {2026},
volume = {25},
number = {1},
pages = {162--165},
doi = {10.1109/lca.2026.3677349}
}
EUVIMED