Metraipolis › archive › dc_dd6f1d4e3ca5

AI as a Compiler: Compiling Triton kernels without the Triton compiler — arXiv:2609.36800

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.36800
License (as archived)
CC0-1.0
Archived
2026-10-07T16:23:02.241Z
SHA-256 of the archived text
9604afe4509b11021342a4f56d538dd6bfe1256d0bdcd306e323b40a8637e4ca
Stable URL of this page
https://metraipolis.com/documento/dc_dd6f1d4e3ca5

How to cite

AI as a Compiler: Compiling Triton kernels without the Triton compiler — arXiv:2609.36800 — arXiv (en), https://arxiv.org/abs/2609.36800 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_dd6f1d4e3ca5 (Metraipolis archive dc_dd6f1d4e3ca5, sha256 9604afe4509b11021342a4f56d538dd6bfe1256d0bdcd306e323b40a8637e4ca)
BibTeX
@misc{dcdd6f1d4e3ca5,
  title = {AI as a Compiler: Compiling Triton kernels without the Triton compiler — arXiv:2609.36800},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.36800},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_dd6f1d4e3ca5; sha256 9604afe4509b11021342a4f56d538dd6bfe1256d0bdcd306e323b40a8637e4ca},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.36800. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Compiler backends are expensive to build and maintain as programming models, workloads, and accelerators evolve. We investigate whether large language models can replace the conventional optimizing and lowering pipeline, a process that we call AI lowering. We study AI lowering from Triton to NVIDIA PTX: an LLM agent translates Triton kernels directly into PTX. We build an environment that evaluates candidate PTX, and an agentic harness in which an LLM translates Triton kernels into PTX. Across twelve common kernels on Ada, Hopper, and Blackwell GPUs and ten kernels from recent ML papers, AI lowering achieves 0.83x-3.34x the performance of autotuned Triton. The largest gains come from transformations that Triton's lowering pipeline does not perform, such as decoding packed binary weights directly into Tensor Core operands (3.34x on BitDelta), assigning each thread a complete softmax row in tensor memory (1.37x on FlashAttention), and reusing overlapping convolution windows (up to 2.23x). These results rely on a robust evaluation harness with comprehensive verification support. We build on Volta, an existing PTX verifier, and substantially extend it to support modern GPU architectures by introducing support for Blackwell's tcgen05 Tensor Core interface. This requires modeling three architectural features: managed tensor memory, descriptor-based operand layouts, and asynchronous execution coordinated through commits, waits, memory barriers, and proxy fences. We discuss the challenges involved in formalizing them, as well as the current limitations. Our results suggest an emerging future in which AI compilers replace custom-written intermediate representations and checkers, reducing the time and engineering effort required to bring up software for new general-purpose and custom chips.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-30 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.36800 — 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 2026-10-07T15:54:53.542Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-30?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).