Paper
Hexagon-MLIR: An AI Compilation Stack For Qualcomm's Neural Processing Units (NPUs)
Authors
Mohammed Javed Absar, Muthu Baskaran, Abhikrant Sharma, Abhilash Bhandari, Ankit Aggarwal, Arun Rangasamy, Dibyendu Das, Fateme Hosseini, Franck Slama, Iulian Brumar, Jyotsna Verma, Krishnaprasad Bindumadhavan, Mitesh Kothari, Mohit Gupta, Ravishankar Kolachana, Richard Lethin, Samarth Narang, Sanjay Motilal Ladwa, Shalini Jain, Snigdha Suresh Dalvi, Tasmia Rahman, Venkat Rasagna Reddy Komatireddy, Vivek Vasudevbhai Pandya, Xiyue Shi, Zachary Zipper
Abstract
In this paper, we present Hexagon-MLIR,an open-source compilation stack that targets Qualcomm Hexagon Neural Processing Unit (NPU) and provides unified support for lowering Triton kernels and PyTorch models . Built using the MLIR framework, our compiler applies a structured sequence of passes to exploit NPU architectural features to accelerate AI workloads. It enables faster deployment of new Triton kernels (hand-written or subgraphs from PyTorch 2.0), for our target by providing automated compilation from kernel to binary. By ingesting Triton kernels, we generate mega-kernels that maximize data locality in the NPU's Tightly Coupled Memory (TCM), reducing the bandwidth bottlenecks inherent in library-based approaches. This initiative complements our commercial toolchains by providing developers with an open-source MLIR-based compilation stack that gives them a path to advance AI compilation capabilities through a more flexible approach. Hexagon-MLIR is a work-in-progress, and we are continuing to add many more optimizations and capabilities in this effort.
Metadata
Related papers
Vibe Coding XR: Accelerating AI + XR Prototyping with XR Blocks and Gemini
Ruofei Du, Benjamin Hersh, David Li, Nels Numan, Xun Qian, Yanhe Chen, Zhongy... • 2026-03-25
Comparing Developer and LLM Biases in Code Evaluation
Aditya Mittal, Ryan Shar, Zichu Wu, Shyam Agarwal, Tongshuang Wu, Chris Donah... • 2026-03-25
The Stochastic Gap: A Markovian Framework for Pre-Deployment Reliability and Oversight-Cost Auditing in Agentic Artificial Intelligence
Biplab Pal, Santanu Bhattacharya • 2026-03-25
Retrieval Improvements Do Not Guarantee Better Answers: A Study of RAG for AI Policy QA
Saahil Mathur, Ryan David Rittner, Vedant Ajit Thakur, Daniel Stuart Schiff, ... • 2026-03-25
MARCH: Multi-Agent Reinforced Self-Check for LLM Hallucination
Zhuo Li, Yupeng Zhang, Pengyu Cheng, Jiajun Song, Mengyu Zhou, Hao Li, Shujie... • 2026-03-25
Raw Data (Debug)
{
"raw_xml": "<entry>\n <id>http://arxiv.org/abs/2602.19762v1</id>\n <title>Hexagon-MLIR: An AI Compilation Stack For Qualcomm's Neural Processing Units (NPUs)</title>\n <updated>2026-02-23T12:12:39Z</updated>\n <link href='https://arxiv.org/abs/2602.19762v1' rel='alternate' type='text/html'/>\n <link href='https://arxiv.org/pdf/2602.19762v1' rel='related' title='pdf' type='application/pdf'/>\n <summary>In this paper, we present Hexagon-MLIR,an open-source compilation stack that targets Qualcomm Hexagon Neural Processing Unit (NPU) and provides unified support for lowering Triton kernels and PyTorch models . Built using the MLIR framework, our compiler applies a structured sequence of passes to exploit NPU architectural features to accelerate AI workloads. It enables faster deployment of new Triton kernels (hand-written or subgraphs from PyTorch 2.0), for our target by providing automated compilation from kernel to binary. By ingesting Triton kernels, we generate mega-kernels that maximize data locality in the NPU's Tightly Coupled Memory (TCM), reducing the bandwidth bottlenecks inherent in library-based approaches. This initiative complements our commercial toolchains by providing developers with an open-source MLIR-based compilation stack that gives them a path to advance AI compilation capabilities through a more flexible approach. Hexagon-MLIR is a work-in-progress, and we are continuing to add many more optimizations and capabilities in this effort.</summary>\n <category scheme='http://arxiv.org/schemas/atom' term='cs.PL'/>\n <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n <published>2026-02-23T12:12:39Z</published>\n <arxiv:primary_category term='cs.PL'/>\n <author>\n <name>Mohammed Javed Absar</name>\n </author>\n <author>\n <name>Muthu Baskaran</name>\n </author>\n <author>\n <name>Abhikrant Sharma</name>\n </author>\n <author>\n <name>Abhilash Bhandari</name>\n </author>\n <author>\n <name>Ankit Aggarwal</name>\n </author>\n <author>\n <name>Arun Rangasamy</name>\n </author>\n <author>\n <name>Dibyendu Das</name>\n </author>\n <author>\n <name>Fateme Hosseini</name>\n </author>\n <author>\n <name>Franck Slama</name>\n </author>\n <author>\n <name>Iulian Brumar</name>\n </author>\n <author>\n <name>Jyotsna Verma</name>\n </author>\n <author>\n <name>Krishnaprasad Bindumadhavan</name>\n </author>\n <author>\n <name>Mitesh Kothari</name>\n </author>\n <author>\n <name>Mohit Gupta</name>\n </author>\n <author>\n <name>Ravishankar Kolachana</name>\n </author>\n <author>\n <name>Richard Lethin</name>\n </author>\n <author>\n <name>Samarth Narang</name>\n </author>\n <author>\n <name>Sanjay Motilal Ladwa</name>\n </author>\n <author>\n <name>Shalini Jain</name>\n </author>\n <author>\n <name>Snigdha Suresh Dalvi</name>\n </author>\n <author>\n <name>Tasmia Rahman</name>\n </author>\n <author>\n <name>Venkat Rasagna Reddy Komatireddy</name>\n </author>\n <author>\n <name>Vivek Vasudevbhai Pandya</name>\n </author>\n <author>\n <name>Xiyue Shi</name>\n </author>\n <author>\n <name>Zachary Zipper</name>\n </author>\n </entry>"
}