Metraipolis › archive › dc_6cceeb23d501
CREDIT: Cost-guided Reduction-reuse with Efficient DSMEM Inter-CTA Tiling — arXiv:2609.01864
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.01864
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-17T04:19:19.606Z
- SHA-256 of the archived text
e081c79c5b75408978d63d6303b9782fb792a5189e9cf7c8e1e3ebe2c94ffe02- Stable URL of this page
https://metraipolis.com/documento/dc_6cceeb23d501
How to cite
CREDIT: Cost-guided Reduction-reuse with Efficient DSMEM Inter-CTA Tiling — arXiv:2609.01864 — arXiv (en), https://arxiv.org/abs/2609.01864 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_6cceeb23d501 (Metraipolis archive dc_6cceeb23d501, sha256 e081c79c5b75408978d63d6303b9782fb792a5189e9cf7c8e1e3ebe2c94ffe02)
BibTeX
@misc{dc6cceeb23d501,
title = {CREDIT: Cost-guided Reduction-reuse with Efficient DSMEM Inter-CTA Tiling — arXiv:2609.01864},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.01864},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_6cceeb23d501; sha256 e081c79c5b75408978d63d6303b9782fb792a5189e9cf7c8e1e3ebe2c94ffe02},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.01864. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
NVIDIA distributed shared memory (DSMEM) enables direct shared-memory access within a thread block cluster. However, cluster synchronization, remote access, and resource costs make it difficult to determine when DSMEM improves performance. To fill this gap, we propose CREDIT, a cost-guided framework that identifies DSMEM-profitable workload patterns, predicts their profitability range, and delivers consistent speedups across diverse workloads. CREDIT combines three innovations: (1) a profiling-driven characterization that identifies workload patterns likely to benefit from DSMEM; (2) a transformation that applies DSMEM to reduction-reuse workloads; (3) a cost model based on profiling data, to determine its profitability range. Evaluations on diverse workloads show CREDIT achieves 91.7% prediction accuracy on profitability. CREDIT beats this http URL , Triton, and optimized non-DSMEM CUDA baselines on all six workloads, with geometric-mean speedups of 1.466x on RTX 5090 and 1.318x on H100. CREDIT's source code is publicly available at this https URL . Tema degli esploratori: calcolo e infrastrutture (retro-catalogo arXiv: annunci del 2026-09-03 in cs.DC). Categorie arXiv: Distributed, Parallel, and Cluster Computing (cs.DC). Pagina del lavoro: https://arxiv.org/abs/2609.01864 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-17T03:27:00.761Z dalla pagina https://arxiv.org/catchup/cs.DC/2026-09-03?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).