[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-651087-105":59,"doc-detail-651087-en":130},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":123,"head_meta":125,"extra_data":127,"updated_unix":129},105,"en","axon-a-synthesizing-superoptimizer-for-tensor-programs","Axon - A Synthesizing Superoptimizer for Tensor Programs","","Writing efficient kernels for AI accelerators is hard because high performance depends on tiling, instruction selection, data layout, and operator fusion. The Axon framework synthesizes kernels for tile-based accelerator tensor programs from semantic specifications, automatically generating target instructions and exploring semantically equivalent variants to select the best-performing option empirically. It discovers algebraic transformations via computation-graph propagation and verifies semantic preservation with SMT over unbounded tensors, then lowers operations to ISA instructions, searches tiling configurations, and fuses to reduce memory traffic.",{"@graph":69,"@context":122},[70,84,105],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/axon-a-synthesizing-superoptimizer-for-tensor-programs/651087/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/axon-a-synthesizing-superoptimizer-for-tensor-programs/651087.png","ImageObject",300,407,{"name":92,"@type":93},"The weather in my hometown","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-29","2026-10-24",true,{"@type":102,"interactionType":103,"userInteractionCount":81},"InteractionCounter",{"@type":104},"ViewAction",{"@type":106,"mainEntity":107},"FAQPage",[108,114,118],{"name":109,"@type":110,"acceptedAnswer":111},"What problem does Axon address for AI accelerator kernel programming?","Question",{"text":112,"@type":113},"Axon targets the difficulty of writing high-performance kernels for AI accelerators, where achieving good utilization requires expertise in tiling, instruction selection, data layout, and operator fusion.","Answer",{"name":115,"@type":110,"acceptedAnswer":116},"How does Axon generate and verify candidate kernels?",{"text":117,"@type":113},"Axon uses program synthesis to generate target instructions from semantics specifications and explores semantically equivalent program variants. It guarantees semantic preservation by checking equivalence with an SMT solver over unbounded tensors.",{"name":119,"@type":110,"acceptedAnswer":120},"What benefits does Axon show in evaluation on Amazon Trainium?",{"text":121,"@type":113},"Compared to Amazon’s Neuron compiler, Axon reports speedups up to 3.7x on individual operators and up to 19x on multi-operator LLM kernels, and it matches or outperforms hand-optimized NKI kernels by up to 1.35x.","https://schema.org",{"og:url":83,"og:type":124,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":126,"canonical":83},"index,follow",{"doc_id":128,"site_id":62},651087,1792885022,{"code":4,"msg":5,"data":131},{"doc_id":128,"user_id":132,"nickname":92,"user_avatar":133,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":81,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":139,"language":140,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":67,"update_tm":144,"read_time":145},1374408744247,"https://ap-avatar.wpscdn.com/avatar/14001251d37d565ec6f?x-image-process=image/resize,m_fixed,w_180,h_180&k=1790672120290115972","arXiv :2606 .26344v1 [ cs .PL] 24 Jun 2026  \nAxon: A Synthesizing Superoptimizer for Tensor Programs  \nAKASH KOTHARI, University of Illinois at Urbana-Champaign, USA  \nSHAOWEI ZHU, Amazon Web Services, USA DANIEL KROENING, Amazon Web Services, USA CHUNGHA SUNG, Amazon Web Services, USA  \nWriting high-performance kernels for AI accelerators requires deep expertise in tiling, instruction selection, data layout, and operator fusion—placing a significant burden on programmers.  \nIn this paper, we focus on tile-based AI accelerator programs and present Axon, a synthesizing superoptimizer for tensor programs: it uses program synthesis to automatically generate target instructions from semantics specifications, and explores semantically equivalent program variants to select the best-performing kernel empirically. Axon discovers algebraic transformations by propagating operators through computation graphs and uses SMT over unbounded tensors to guarantee that all transformations preserve semantics—without requiring hand-crafted rewrite rules. It then lowers tensor operations to target ISA instructions, explores tiling configurations constrained by hardware descriptions, and fuses operators and instructions to minimize memory traffic.  \nWe evaluate Axon on Amazon’s Trainium across 20 benchmarks spanning individual tensor operators and multi-operator LLM kernels, including Group Query Attention and GatedMLP. Compared to Amazon’s Neuron compiler, Axon achieves speedups of up to 3.7x on individual operators and up to 19x on multi-operator kernels. Axon matches or outperforms hand-optimized NKI kernels by up to 1.35x, and achieves geomean speedups of 2–10% over Mirage, a state-of-the-art search-based compiler, on supported benchmarks—while synthesizing kernels for 16 additional benchmarks that Mirage cannot handle.  \nACM Reference Format:  \nAkash Kothari, Shaowei Zhu, Daniel Kroening, and Chungha Sung. 2026. Axon: A Synthesizing Superoptimizer  \nfor Tensor Programs. 1, 1 (June 2026), 29 pages. [https://doi.org/10.1145/nnnnnnn.nnnnnnn](https://doi.org/10.1145/nnnnnnn.nnnnnnn)  \n1 Introduction  \nPerformance of modern AI accelerators depends critically on the quality of kernels—low-level programs that describe how hardware resources should be orchestrated to execute a workload. Even though tile-based kernel languages (e.g., NKI for Amazon’s Trainium [6], Pallas for Google’s TPUs [25], and Triton [44] for GPUs) provide easier pathways to good performance by abstracting away many low-level details, it remains challenging to write kernels that achieve high utilization of hardware. Kernel programmers still need to understand hardware details such as memory hierarchies and specialized compute and data movement engines to reason about tiling, data layout, operator fusion, and engine selection—tasks that require deep hardware expertise and are out of reach for most ML engineers.  \nWe present Axon, a synthesizing superoptimizer for tile-based AI accelerator programs: given tensor programs written in a Numpy-like interface [22], it uses program synthesis to automatically  \nAuthors’ Contact Information: Akash KothariUniversity of Illinois at Urbana-Champaign, USA; Shaowei ZhuAmazon Web Services, USA; Daniel KroeningAmazon Web Services, USA; Chungha SungAmazon Web Services, USA.  \nPermission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than the author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires [prior specific permission and/or a fee. Request permissions from permissions@acm.org](prior specific permission and/or a fee. Request permissions from permissions@acm.org).  \n© 2026 Cop","cbCaik16nLsr3uVD","https://ap.wps.com/l/cbCaik16nLsr3uVD","pdf",1429386,29,"English","# Introduction\n## Axon: Synthesizing superoptimization for tensor programs\n## Program synthesis, semantic verification, and lowering to ISA\n## Evaluation on Trainium and benchmark results","[{\"question\":\"What problem does Axon address for AI accelerator kernel programming?\",\"answer\":\"Axon targets the difficulty of writing high-performance kernels for AI accelerators, where achieving good utilization requires expertise in tiling, instruction selection, data layout, and operator fusion.\"},{\"question\":\"How does Axon generate and verify candidate kernels?\",\"answer\":\"Axon uses program synthesis to generate target instructions from semantics specifications and explores semantically equivalent program variants. It guarantees semantic preservation by checking equivalence with an SMT solver over unbounded tensors.\"},{\"question\":\"What benefits does Axon show in evaluation on Amazon Trainium?\",\"answer\":\"Compared to Amazon’s Neuron compiler, Axon reports speedups up to 3.7x on individual operators and up to 19x on multi-operator LLM kernels, and it matches or outperforms hand-optimized NKI kernels by up to 1.35x.\"}]","Axon - A Synthesizing Superoptimizer for Tensor Programs | PDF",1792860296,73]