Metraipolis › archive › dc_2fb2b16a7692

Decoupling Logical Masks from GPU Execution for Dynamic Block-Sparse Attention — arXiv:2609.25869

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.25869
License (as archived)
CC0-1.0
Archived
2026-10-08T14:37:39.375Z
SHA-256 of the archived text
38215ca53b168cf5094505421ffc37c100bd962c9132094ec2ba750a31983934
Stable URL of this page
https://metraipolis.com/documento/dc_2fb2b16a7692

How to cite

Decoupling Logical Masks from GPU Execution for Dynamic Block-Sparse Attention — arXiv:2609.25869 — arXiv (en), https://arxiv.org/abs/2609.25869 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_2fb2b16a7692 (Metraipolis archive dc_2fb2b16a7692, sha256 38215ca53b168cf5094505421ffc37c100bd962c9132094ec2ba750a31983934)
BibTeX
@misc{dc2fb2b16a7692,
  title = {Decoupling Logical Masks from GPU Execution for Dynamic Block-Sparse Attention — arXiv:2609.25869},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.25869},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_2fb2b16a7692; sha256 38215ca53b168cf5094505421ffc37c100bd962c9132094ec2ba750a31983934},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.25869. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Attention computation makes inference expensive in video diffusion transformers (vDiTs), which generate videos through iterative denoising. Block-sparse attention (BSA) reduces this cost by computing only blocks selected by a logical mask, which specifies attention interactions to compute. However, coupling logical block geometry to execution choices limits adaptation to varying masks and graphics processing units (GPUs), while runtime kernel specialization can incur preparation overhead that outweighs execution time savings. We present Tessera, a specialized runtime for dynamic BSA that decouples logical masks from GPU execution while preserving specified attention interactions. Its physical mapping layer retains, combines, or subdivides logical attention blocks into physical tiles suited to different attention mask shapes and GPU architectures. Its task organization layer groups and schedules tiles within GPU tasks to reuse data, expose parallelism, and overlap data movement with computation. Finally, profile-guided regime selection enables low- overhead execution plan selection through a lookup table constructed from offline profiling. We implement Tessera with specialized CUDA kernels supporting four NVIDIA GPU generations. Evaluated on 2,315 real attention masks and industrial video diffusion models, Tessera achieves up to 6.79x BSA request speedup over baseline systems in the evaluated video diffusion models.

Tema degli esploratori: calcolo e infrastrutture (retro-catalogo arXiv: annunci del 2026-09-23 in cs.AR).
Categorie arXiv: Hardware Architecture (cs.AR).
Pagina del lavoro: https://arxiv.org/abs/2609.25869 — 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 2026-10-08T13:47:52.525Z dalla pagina https://arxiv.org/catchup/cs.AR/2026-09-23?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).