[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"site-home-live-entry-setting":3,"site-header-contests":8,"site-header-camps":49,"study-material-detail-study-68":104},{"enabled":4,"title":5,"statusText":5,"buttonText":5,"campSlug":5,"stage":6,"classroomId":7,"hasClassroom":4},false,"",0,null,{"total":9,"list":10,"pageNum":44,"pageSize":45,"size":9,"startRow":44,"endRow":9,"pages":44,"prePage":6,"nextPage":6,"isFirstPage":46,"isLastPage":46,"hasPreviousPage":4,"hasNextPage":4,"navigatePages":47,"navigatepageNums":48,"navigateFirstPage":44,"navigateLastPage":44},3,[11,25,35],{"slug":12,"title":13,"logo":7,"liveUrl":14,"prize":15,"sponsor":16,"description":17,"organizer":18,"startTime":19,"endTime":20,"registrationDeadline":21,"status":22,"dashboardVisibility":23,"createdAt":24},"spring2026","2026春季启元人工智能大赛","https:\u002F\u002Fwww.bilibili.com\u002Fblackboard\u002Flive\u002Flive-activity-player.html?cid=1822703816&quality=0","200,000","麒麟软件有限公司","“2026春季启元人工智能大赛”是由启元实验室主办的面向全国人工智能领域开发者、科研人员及高校学生的权威赛事。大赛旨在促进与培养具备工程实践能力和创新能力的优秀人工智能领域开发者、推广国产软硬件技术与生态、助力国家人工智能产业高质量发展。\n各参赛选手需在规定的比赛周期内，依据赛事规则选定赛题并完成相应挑战，优胜者将共同角逐总额20万元的赛事奖金池，并且还将有机会获得实习和知名企业优质岗位内推资格。","启元实验室","2026-04-19 16:00:00","2026-07-13 16:00:00","2026-06-16 16:00:00","ongoing","private","2026-04-16 08:00:00",{"slug":26,"title":27,"logo":7,"liveUrl":14,"prize":28,"sponsor":7,"description":29,"organizer":18,"startTime":30,"endTime":31,"registrationDeadline":32,"status":33,"dashboardVisibility":23,"createdAt":34},"autumn2025","2025秋季启元人工智能大赛","150,000","“2025秋季启元人工智能大赛”是由启元实验室主办的面向全国人工智能领域开发者、科研人员及高校学生的权威赛事。大赛旨在促进与培养具备工程实践能力和创新能力的优秀人工智能领域开发者、推广国产软硬件技术与生态、助力国家人工智能产业高质量发展。","2025-11-08 08:00:00","2026-01-13 08:00:00","2025-12-30 08:00:00","ended","2025-11-03 08:00:00",{"slug":36,"title":37,"logo":7,"liveUrl":14,"prize":38,"sponsor":7,"description":39,"organizer":18,"startTime":40,"endTime":41,"registrationDeadline":42,"status":33,"dashboardVisibility":23,"createdAt":43},"summer2025","启元人工智能大赛(系统方向)","80,000","启元人工智能大赛由启元实验室主办，其中系统赛道基于启元实验室九源软件栈的相关开源项目，通过拉动开源社区力量，一方面促进国产智能计算生态的发展，另一方面也为人工智能系统方向的学生及从业者提供一个实践平台。","2025-07-08 08:00:00","2025-08-25 08:00:00","2025-07-22 08:00:00","2025-06-22 08:00:00",1,5,true,8,[44],{"total":45,"list":50,"pageNum":44,"pageSize":45,"size":45,"startRow":44,"endRow":45,"pages":44,"prePage":6,"nextPage":6,"isFirstPage":46,"isLastPage":46,"hasPreviousPage":4,"hasNextPage":4,"navigatePages":47,"navigatepageNums":103,"navigateFirstPage":44,"navigateLastPage":44},[51,62,73,84,94],{"id":52,"slug":53,"title":54,"logo":55,"publicLiveEmbedUrl":14,"description":56,"introduction":57,"organizer":18,"coOrganizer":5,"sponsor":16,"startTime":58,"endTime":59,"status":22,"dashboardVisibility":60,"createdAt":58,"updatedAt":61},"019eeeff-6b00-7d55-a629-efafded3e232","summer2026","2026年夏季InfiniTensor大模型与人工智能系统训练营","https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fuploads\u002F2026-06\u002F019eef7f-7816-7828-9309-79c2c61dd568.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=fLKIdTmv5ztTF5lNyW4vFO%2FwiU4%3D","由启元实验室发起的InfiniTensor大模型与人工智能系统训练营，面向开源社区、高校及产业界伙伴开放。活动以「普及AI全栈知识技术、培养工程实践与创新型人才、推广国产软硬件技术生态」为核心目标，助力国家人工智能产业高质量发展。","...","2026-06-22 08:00:00","2026-10-19 08:00:00","public","2026-10-12 08:00:00",{"id":63,"slug":64,"title":65,"logo":66,"publicLiveEmbedUrl":7,"description":67,"introduction":57,"organizer":18,"coOrganizer":5,"sponsor":68,"startTime":69,"endTime":70,"status":33,"dashboardVisibility":23,"createdAt":71,"updatedAt":72},"019af3f3-e06a-7245-aa8d-37dc99a69c73","winter2025","2025年冬季InfiniTensor大模型与人工智能系统训练营","https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fuploads\u002F2025-12\u002F019afc69-b7ca-7805-91fb-605429671c01.png?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=IfcKyBkOe6Ug2cb8ODO9I1IUNkA%3D","由启元实验室发起的InfiniTensor大模型与人工智能系统训练营，面向开源社区、高校及产业界伙伴开放。活动以「普及AI全栈知识技术、培养工程实践与创新型人才、推广国产软硬件技术生态」为核心目标，助力国家人工智能产业高质量发展。 \n","新华三技术有限公司","2025-12-25 08:00:00","2026-03-27 08:00:00","2025-12-06 08:00:00","2026-10-22 20:24:30",{"id":74,"slug":36,"title":75,"logo":76,"publicLiveEmbedUrl":7,"description":77,"introduction":57,"organizer":18,"coOrganizer":5,"sponsor":78,"startTime":79,"endTime":80,"status":81,"dashboardVisibility":23,"createdAt":82,"updatedAt":83},"01980342-208c-7803-b3f3-19e1d31960ba","2025年夏季InfiniTensor大模型与人工智能系统训练营","https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fuploads\u002F2025-07\u002F0198045e-28fe-78f0-95fe-23835ce58395.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Ev4DWdZSXJ0t1T4u2JL5qybAazs%3D","InfiniTensor大模型与人工智能系统训练营由启元实验室及清华大学 AI 系统方向团队发起，面向开源社区、高校和产业界，普及人工智能各方面知识和技术，培养相关人才。训练营联合开源项目开发者、学术界大牛、工业界资深工程师参与授课与项目指导。课程采用「线上直播授课 + 在线答疑 + 项目实践辅导 」的方式进行，确保每位同学都能获得充分成长。训练营全程免费，通过视频、直播课、习题和项目实践，在人工智能飞速发展的背景下，为学员提供一窥深层技术的机会。","北京积算科技有限公司","2025-07-14 08:00:00","2025-09-30 08:00:00","upcoming","2025-07-13 08:00:00","2026-10-22 20:24:58",{"id":85,"slug":86,"title":87,"logo":88,"publicLiveEmbedUrl":7,"description":89,"introduction":57,"organizer":18,"coOrganizer":5,"sponsor":5,"startTime":90,"endTime":91,"status":81,"dashboardVisibility":23,"createdAt":92,"updatedAt":93},"01980cf9-a7d8-7796-ae86-0bbe05334add","winter2024","2024年冬季InfiniTensor大模型与人工智能系统训练营","https:\u002F\u002Ftesting-cdn.infinitensor.com\u002Fuploads\u002F2025-07\u002F01981269-ce23-7624-a287-9f07dd0d85f6.jpg","InfiniTensor大模型与人工智能系统训练营由清华大学发起，面向开源社区、高校和产业界，普及人工智能各方面知识和技术，培养相关人才。训练营全程免费，通过视频、直播课、习题和项目实践，在人工智能飞速发展的背景下，为学员提供一窥深层技术的机会。报名学员请阅读“导学”课程页的训练营介绍。","2025-01-05 08:00:00","2025-03-22 08:00:00","2025-07-15 08:00:00","2026-10-22 20:24:39",{"id":95,"slug":96,"title":97,"logo":98,"publicLiveEmbedUrl":7,"description":99,"introduction":57,"organizer":18,"coOrganizer":5,"sponsor":5,"startTime":100,"endTime":101,"status":81,"dashboardVisibility":23,"createdAt":92,"updatedAt":102},"01980c20-363a-7dc3-8f5a-58a68e08dcc4","summer2024","2024年夏季InfiniTensor大模型与人工智能系统训练营","https:\u002F\u002Ftesting-cdn.infinitensor.com\u002Fuploads\u002F2025-07\u002F01981269-fbfa-788a-afcd-17fa9be74ff9.jpg","InfiniTensor大模型与人工智能系统训练营由清华大学发起，面向开源社区和高校，普及人工智能各方面知识和技术，培养相关人才。训练营全程免费，通过课程、习题和项目实践，在人工智能飞速发展的背景下，为学员提供一窥深层技术的机会。","2024-07-20 08:00:00","2024-09-29 08:00:00","2026-10-22 20:24:49",[44],{"currentItem":105,"processedData":125},{"img":106,"keywords":107,"treeType":44,"description":108,"showStatus":44,"updateTime":109,"type":110,"title":111,"content":112,"nextArticle":113,"createTime":109,"sortOrder":117,"canonicalPath":118,"id":119,"contentType":120,"prevArticle":121},"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_fm.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=pje%2BVzV90DDIupXnQ45j9zs9qtw%3D","Triton 编译器,寒武纪 MLU,Linalg,国产 DSA 芯片,算子开发","讲解 Triton 适配国产芯片价值，对比 CUDA 开发差异，拆解寒武纪 Triton 编译器、Triton 转 Linalg 流程、指针连续性分析，配套开源 triton-linalg 项目介绍。","2026-10-22 20:38:11",2,"2026 夏季 InfiniTensor 训练营 Triton & 九齿 2：《Triton 编译器探索与寒武纪 MLU 实践》","![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw01.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=vl%2BeMhiU53Zv2c7IZpozWBMyass%3D)\r\n\r\n\r\n\u003Cspan style=\"font-size:12px;color:#555;\">7 月 20 日，2026 夏季 InfiniTensor 训练营 Triton & 九齿方向第二课《Triton 编译器探索与寒武纪 MLU 实践》开讲。\r\n前置课程《Triton 练气术》讲解了 Triton 并行编程语言核心知识，涵盖 DSL 特性、与 CUDA 对比优势及 program_id、arange、load、store 四大基础 API。\u003C\u002Fspan>\r\n\r\n本节课将从硬件厂商视角出发，系统讲解 Triton 编程语言的核心价值，并深入解析寒武纪 Cambricon Triton 编译器的架构设计，重点拆解 Triton 到 Linalg 的转换方案及指针连续性分析等关键技术。\r\n\r\n\r\n## 为什么使用 Triton？\r\n### 多元芯片生态的痛点\r\n- CUDA 生态垄断与国内厂商“各自为战”的困境\r\n- 用户侧：多厂商语言切换成本高、学习曲线陡峭\r\n- 产业侧：缺乏统一语言层，难以形成合力对抗 CUDA 生态\r\n\r\n### 为何选择 Triton 作为“最大公约数”\r\n1.**生态拉通层面**：芯片层与编译器后端无法统一，唯有厂商语言层可实现共建\r\n2.**技术特性层面**：硬件无关的中间层表示（IR），屏蔽架构差异\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw02.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=4GFpohK7ir0lmeDsdrhjUAyR4PQ%3D)\r\n\r\n3.**开发体验对比**：CUDA vs Triton\r\n- CUDA：类单反相机，上限高下限低，需手动管理 Shared Memory\u002F流水\u002FBarriers\r\n- Triton：类智能相机，下限高上限适中，编译器托管片上内存与指令融合\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw03.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=UqveHZbBAevyztjiBe15Wkw0rIc%3D)\r\n\r\n\r\n### Triton 的四大核心优势\r\n- **快速高效**：高效支持上层 AI 框架（PyTorch 等）\r\n- **易于上手**：类 Python 语法，低学习成本\r\n- **生态兼容性强**：减少对单一生态依赖\r\n- **适配度高**：原生支持 Torch Compile \u002F Inductor 路径\r\n\r\n\r\n## Cambricon Triton 编译器\r\n### Cambricon Triton 定位与特点\r\n1. 基于 Triton 语言适配寒武纪 MLU（DSA 架构）\r\n2. 最大程度复用 Triton 开发生态，性能接近手写算子\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw04.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=xHGEq%2BShpvYGgqzEcXHM5XcxETY%3D)\r\n\r\n### 版本现状（v2.1 \u002F 基于原生 Triton 3.4）\r\n1. 支持 JIT \u002F AOT 编译模式\r\n2. 全量支持 Triton 官方原语及后端无关特性（Auto-tune, Interpreter 等）\r\n3. `num_warps` 参数重定义：映射 IPU 任务类型（Block 任务 \u002F U1-U4 Cluster 任务）\r\n4. 解除 Block Size 必须为 2 的幂次限制\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw05.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=%2BZm1UXmNoTphnB7RUQvCc1BMgV0%3D)\r\n\r\n\r\n### 编译器架构与 Triton-to-Linalg 核心技术\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw06.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=CO0juJ3iuvcEe0Zfda8SCJgZ7lo%3D)\r\n\r\n\r\n#### 1. 整体技术路线\r\n- 国际主流 GPU 路线：Triton → Triton GPU → LLVM → PTX\r\n- **寒武纪 DSA 路线**：Triton → **Linalg** → MLU → MLVM → LLVM IR → MLISA\r\n\r\n#### 2. 为何选择 Linalg Dialect？\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw07.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=IosfLtanliniwyWGNc9LsTDoIO0%3D)\r\n\r\n- 避免重复造轮子：GPU 优化 Pass 无法复用于 DSA\r\n- MLIR 社区广泛采用，基础设施丰富（Tile, Reduction, Fields Promotion）\r\n- 易于扩展 OP 与定制化 Pass（Linalg Extension）\r\n- 基于 Tensor\u002FMemRef 定义，与 Triton IR 对应\r\n- 业界主流方案（Triton-shared 项目）\r\n\r\n#### 3. Triton → Linalg 转换原则\r\n- 尽可能使用 structured 算子，不使用 `linalg.generic` 算子\r\n- 尽早识别算子语义（如 `tt.reduce` → `linalg_ext.argmax`）\r\n\r\n#### 4. Triton-Linalg 构成部分\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw08.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=%2BB99HfVwvxQTgWeL4XCros32m4k%3D)\r\n\r\n\r\n#### 5. 核心难点：指针分析\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw09.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=yn25QavXpqqW2fntzlO8aFDNtUg%3D)\r\n\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw10.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=o%2B3FWOzj%2BADbmzVQ%2FzZ%2Bsa%2Bo4t8%3D)\r\n\r\n- **问题本质**：`tt.load` 可能对应 `copy`（连续访存）或 `gather`（离散访存），错误下降导致 DMA 效率骤降\r\n\r\n- **解决方案**：自动化多维连续性识别算法\r\n\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw11.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=eFrYuPI8ozZf8sXedkfB4BCS5gs%3D)\r\n\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw12.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=LiBn%2FWB%2FofFwzgeqShCLY2lY62o%3D)\r\n\r\n#### 6. Linalg → MLU 下降流程\r\n- Tile Fusion \u002F Buffer Rise 等 Pass\r\n- 复用 MLIR 官方基础设施 + 寒武纪定制优化\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw13.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Z9fPwiihdtjefj4tVQTBcyrEi8I%3D)\r\n\r\n\r\n\r\n#### 7. Triton-Linalg 开源项目\r\n寒武纪开源跨平台 AI 编译器前端 Triton-Linalg，帮助开发者降低硬件适配成本，提高集成效率。\r\n- GitHub 开源地址：http:\u002F\u002Fgithub.com\u002FCambricon\u002Ftriton-linalg\r\n\r\n\r\n\r\n\r\n\r\n## 加入训练营\r\n#### 📺完整课程内容，请观看直播或查看课程回放：\r\n**InfiniTensor 官网**： https:\u002F\u002Fwww.infinitensor.com\r\n\r\n**B站直播**： InfiniTensor 官方直播间\r\n\r\n**视频号直播**： InfiniTensor 视频号预约\r\n\r\n**答疑交流**： 训练营官方社群⬇️\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002Fportal_news\u002F2026-xjq.png?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=RCGzL2mm8Hipb2jgTTOzg7DOtUs%3D)\r\n\r\n## 关注与交流\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002Fportal_news\u002Fnews_ewm02.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=ne2IbUxGKCs8rwPiAjr5B%2BpodTI%3D)",{"publishTime":7,"createTime":109,"canonicalPath":114,"id":115,"title":116},"\u002Fstudy-material\u002Fstudy-67","study-67","2026 夏季 InfiniTensor 训练营前沿模型 1：《大模型学习方法》",41,"\u002Fstudy-material\u002Fstudy-68","study-68","markdown",{"publishTime":7,"createTime":109,"canonicalPath":122,"id":123,"title":124},"\u002Fstudy-material\u002Fstudy-69","study-69","2026 夏季 InfiniTensor 训练营前沿模型 2：《检索增强的语言模型》",{"htmlContent":126,"headings":127},"\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw01.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=vl%2BeMhiU53Zv2c7IZpozWBMyass%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cspan style=\"font-size:12px;color:#555;\">7 月 20 日，2026 夏季 InfiniTensor 训练营 Triton & 九齿方向第二课《Triton 编译器探索与寒武纪 MLU 实践》开讲。\u003Cbr>前置课程《Triton 练气术》讲解了 Triton 并行编程语言核心知识，涵盖 DSL 特性、与 CUDA 对比优势及 program_id、arange、load、store 四大基础 API。\u003C\u002Fspan>\u003C\u002Fp>\n\u003Cp>本节课将从硬件厂商视角出发，系统讲解 Triton 编程语言的核心价值，并深入解析寒武纪 Cambricon Triton 编译器的架构设计，重点拆解 Triton 到 Linalg 的转换方案及指针连续性分析等关键技术。\u003C\u002Fp>\n\u003Ch2 id=\"为什么使用-triton\">\u003Ca class=\"anchor\" href=\"#为什么使用-triton\" aria-label=\"跳转到标题：为什么使用 Triton？\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>为什么使用 Triton？\u003C\u002Fh2>\u003Ch3 id=\"多元芯片生态的痛点\">\u003Ca class=\"anchor\" href=\"#多元芯片生态的痛点\" aria-label=\"跳转到标题：多元芯片生态的痛点\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>多元芯片生态的痛点\u003C\u002Fh3>\u003Cul>\n\u003Cli>CUDA 生态垄断与国内厂商“各自为战”的困境\u003C\u002Fli>\n\u003Cli>用户侧：多厂商语言切换成本高、学习曲线陡峭\u003C\u002Fli>\n\u003Cli>产业侧：缺乏统一语言层，难以形成合力对抗 CUDA 生态\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3 id=\"为何选择-triton-作为-最大公约数\">\u003Ca class=\"anchor\" href=\"#为何选择-triton-作为-最大公约数\" aria-label=\"跳转到标题：为何选择 Triton 作为“最大公约数”\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>为何选择 Triton 作为“最大公约数”\u003C\u002Fh3>\u003Cp>1.\u003Cstrong>生态拉通层面\u003C\u002Fstrong>：芯片层与编译器后端无法统一，唯有厂商语言层可实现共建\u003Cbr>2.\u003Cstrong>技术特性层面\u003C\u002Fstrong>：硬件无关的中间层表示（IR），屏蔽架构差异\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw02.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=4GFpohK7ir0lmeDsdrhjUAyR4PQ%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>3.\u003Cstrong>开发体验对比\u003C\u002Fstrong>：CUDA vs Triton\u003C\u002Fp>\n\u003Cul>\n\u003Cli>CUDA：类单反相机，上限高下限低，需手动管理 Shared Memory\u002F流水\u002FBarriers\u003C\u002Fli>\n\u003Cli>Triton：类智能相机，下限高上限适中，编译器托管片上内存与指令融合\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw03.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=UqveHZbBAevyztjiBe15Wkw0rIc%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"triton-的四大核心优势\">\u003Ca class=\"anchor\" href=\"#triton-的四大核心优势\" aria-label=\"跳转到标题：Triton 的四大核心优势\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Triton 的四大核心优势\u003C\u002Fh3>\u003Cul>\n\u003Cli>\u003Cstrong>快速高效\u003C\u002Fstrong>：高效支持上层 AI 框架（PyTorch 等）\u003C\u002Fli>\n\u003Cli>\u003Cstrong>易于上手\u003C\u002Fstrong>：类 Python 语法，低学习成本\u003C\u002Fli>\n\u003Cli>\u003Cstrong>生态兼容性强\u003C\u002Fstrong>：减少对单一生态依赖\u003C\u002Fli>\n\u003Cli>\u003Cstrong>适配度高\u003C\u002Fstrong>：原生支持 Torch Compile \u002F Inductor 路径\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2 id=\"cambricon-triton-编译器\">\u003Ca class=\"anchor\" href=\"#cambricon-triton-编译器\" aria-label=\"跳转到标题：Cambricon Triton 编译器\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Cambricon Triton 编译器\u003C\u002Fh2>\u003Ch3 id=\"cambricon-triton-定位与特点\">\u003Ca class=\"anchor\" href=\"#cambricon-triton-定位与特点\" aria-label=\"跳转到标题：Cambricon Triton 定位与特点\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Cambricon Triton 定位与特点\u003C\u002Fh3>\u003Col>\n\u003Cli>基于 Triton 语言适配寒武纪 MLU（DSA 架构）\u003C\u002Fli>\n\u003Cli>最大程度复用 Triton 开发生态，性能接近手写算子\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw04.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=xHGEq%2BShpvYGgqzEcXHM5XcxETY%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"版本现状-v2-1-基于原生-triton-3-4\">\u003Ca class=\"anchor\" href=\"#版本现状-v2-1-基于原生-triton-3-4\" aria-label=\"跳转到标题：版本现状（v2.1 \u002F 基于原生 Triton 3.4）\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>版本现状（v2.1 \u002F 基于原生 Triton 3.4）\u003C\u002Fh3>\u003Col>\n\u003Cli>支持 JIT \u002F AOT 编译模式\u003C\u002Fli>\n\u003Cli>全量支持 Triton 官方原语及后端无关特性（Auto-tune, Interpreter 等）\u003C\u002Fli>\n\u003Cli>\u003Ccode>num_warps\u003C\u002Fcode> 参数重定义：映射 IPU 任务类型（Block 任务 \u002F U1-U4 Cluster 任务）\u003C\u002Fli>\n\u003Cli>解除 Block Size 必须为 2 的幂次限制\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw05.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=%2BZm1UXmNoTphnB7RUQvCc1BMgV0%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"编译器架构与-triton-to-linalg-核心技术\">\u003Ca class=\"anchor\" href=\"#编译器架构与-triton-to-linalg-核心技术\" aria-label=\"跳转到标题：编译器架构与 Triton-to-Linalg 核心技术\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>编译器架构与 Triton-to-Linalg 核心技术\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw06.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=CO0juJ3iuvcEe0Zfda8SCJgZ7lo%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"1-整体技术路线\">\u003Ca class=\"anchor\" href=\"#1-整体技术路线\" aria-label=\"跳转到标题：1. 整体技术路线\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1. 整体技术路线\u003C\u002Fh4>\u003Cul>\n\u003Cli>国际主流 GPU 路线：Triton → Triton GPU → LLVM → PTX\u003C\u002Fli>\n\u003Cli>\u003Cstrong>寒武纪 DSA 路线\u003C\u002Fstrong>：Triton → \u003Cstrong>Linalg\u003C\u002Fstrong> → MLU → MLVM → LLVM IR → MLISA\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"2-为何选择-linalg-dialect\">\u003Ca class=\"anchor\" href=\"#2-为何选择-linalg-dialect\" aria-label=\"跳转到标题：2. 为何选择 Linalg Dialect？\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2. 为何选择 Linalg Dialect？\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw07.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=IosfLtanliniwyWGNc9LsTDoIO0%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>避免重复造轮子：GPU 优化 Pass 无法复用于 DSA\u003C\u002Fli>\n\u003Cli>MLIR 社区广泛采用，基础设施丰富（Tile, Reduction, Fields Promotion）\u003C\u002Fli>\n\u003Cli>易于扩展 OP 与定制化 Pass（Linalg Extension）\u003C\u002Fli>\n\u003Cli>基于 Tensor\u002FMemRef 定义，与 Triton IR 对应\u003C\u002Fli>\n\u003Cli>业界主流方案（Triton-shared 项目）\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"3-triton-linalg-转换原则\">\u003Ca class=\"anchor\" href=\"#3-triton-linalg-转换原则\" aria-label=\"跳转到标题：3. Triton → Linalg 转换原则\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>3. Triton → Linalg 转换原则\u003C\u002Fh4>\u003Cul>\n\u003Cli>尽可能使用 structured 算子，不使用 \u003Ccode>linalg.generic\u003C\u002Fcode> 算子\u003C\u002Fli>\n\u003Cli>尽早识别算子语义（如 \u003Ccode>tt.reduce\u003C\u002Fcode> → \u003Ccode>linalg_ext.argmax\u003C\u002Fcode>）\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"4-triton-linalg-构成部分\">\u003Ca class=\"anchor\" href=\"#4-triton-linalg-构成部分\" aria-label=\"跳转到标题：4. Triton-Linalg 构成部分\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>4. Triton-Linalg 构成部分\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw08.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=%2BB99HfVwvxQTgWeL4XCros32m4k%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"5-核心难点-指针分析\">\u003Ca class=\"anchor\" href=\"#5-核心难点-指针分析\" aria-label=\"跳转到标题：5. 核心难点：指针分析\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>5. 核心难点：指针分析\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw09.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=yn25QavXpqqW2fntzlO8aFDNtUg%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw10.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=o%2B3FWOzj%2BADbmzVQ%2FzZ%2Bsa%2Bo4t8%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cp>\u003Cstrong>问题本质\u003C\u002Fstrong>：\u003Ccode>tt.load\u003C\u002Fcode> 可能对应 \u003Ccode>copy\u003C\u002Fcode>（连续访存）或 \u003Ccode>gather\u003C\u002Fcode>（离散访存），错误下降导致 DMA 效率骤降\u003C\u002Fp>\n\u003C\u002Fli>\n\u003Cli>\u003Cp>\u003Cstrong>解决方案\u003C\u002Fstrong>：自动化多维连续性识别算法\u003C\u002Fp>\n\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw11.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=eFrYuPI8ozZf8sXedkfB4BCS5gs%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw12.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=LiBn%2FWB%2FofFwzgeqShCLY2lY62o%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"6-linalg-mlu-下降流程\">\u003Ca class=\"anchor\" href=\"#6-linalg-mlu-下降流程\" aria-label=\"跳转到标题：6. Linalg → MLU 下降流程\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>6. Linalg → MLU 下降流程\u003C\u002Fh4>\u003Cul>\n\u003Cli>Tile Fusion \u002F Buffer Rise 等 Pass\u003C\u002Fli>\n\u003Cli>复用 MLIR 官方基础设施 + 寒武纪定制优化\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_68_zw13.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Z9fPwiihdtjefj4tVQTBcyrEi8I%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"7-triton-linalg-开源项目\">\u003Ca class=\"anchor\" href=\"#7-triton-linalg-开源项目\" aria-label=\"跳转到标题：7. Triton-Linalg 开源项目\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>7. Triton-Linalg 开源项目\u003C\u002Fh4>\u003Cp>寒武纪开源跨平台 AI 编译器前端 Triton-Linalg，帮助开发者降低硬件适配成本，提高集成效率。\u003C\u002Fp>\n\u003Cul>\n\u003Cli>GitHub 开源地址：\u003Ca href=\"http:\u002F\u002Fgithub.com\u002FCambricon\u002Ftriton-linalg\">http:\u002F\u002Fgithub.com\u002FCambricon\u002Ftriton-linalg\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2 id=\"加入训练营\">\u003Ca class=\"anchor\" href=\"#加入训练营\" aria-label=\"跳转到标题：加入训练营\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>加入训练营\u003C\u002Fh2>\u003Ch4 id=\"完整课程内容-请观看直播或查看课程回放\">\u003Ca class=\"anchor\" href=\"#完整课程内容-请观看直播或查看课程回放\" aria-label=\"跳转到标题：📺完整课程内容，请观看直播或查看课程回放：\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>📺完整课程内容，请观看直播或查看课程回放：\u003C\u002Fh4>\u003Cp>\u003Cstrong>InfiniTensor 官网\u003C\u002Fstrong>： \u003Ca href=\"https:\u002F\u002Fwww.infinitensor.com\">https:\u002F\u002Fwww.infinitensor.com\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>\u003Cstrong>B站直播\u003C\u002Fstrong>： InfiniTensor 官方直播间\u003C\u002Fp>\n\u003Cp>\u003Cstrong>视频号直播\u003C\u002Fstrong>： InfiniTensor 视频号预约\u003C\u002Fp>\n\u003Cp>\u003Cstrong>答疑交流\u003C\u002Fstrong>： 训练营官方社群⬇️\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002Fportal_news\u002F2026-xjq.png?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=RCGzL2mm8Hipb2jgTTOzg7DOtUs%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"关注与交流\">\u003Ca class=\"anchor\" href=\"#关注与交流\" aria-label=\"跳转到标题：关注与交流\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>关注与交流\u003C\u002Fh2>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002Fportal_news\u002Fnews_ewm02.jpg?Expires=1792752590&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=ne2IbUxGKCs8rwPiAjr5B%2BpodTI%3D\" alt=\"\">\u003C\u002Fp>\n",[128,131,133,136,139,142,145,148,151,155,158,161,164,167,170,173,175,178],{"id":129,"text":130,"level":110},"为什么使用-triton","为什么使用 Triton？",{"id":132,"text":132,"level":9},"多元芯片生态的痛点",{"id":134,"text":135,"level":9},"为何选择-triton-作为-最大公约数","为何选择 Triton 作为“最大公约数”",{"id":137,"text":138,"level":9},"triton-的四大核心优势","Triton 的四大核心优势",{"id":140,"text":141,"level":110},"cambricon-triton-编译器","Cambricon Triton 编译器",{"id":143,"text":144,"level":9},"cambricon-triton-定位与特点","Cambricon Triton 定位与特点",{"id":146,"text":147,"level":9},"版本现状-v2-1-基于原生-triton-3-4","版本现状（v2.1 \u002F 基于原生 Triton 3.4）",{"id":149,"text":150,"level":9},"编译器架构与-triton-to-linalg-核心技术","编译器架构与 Triton-to-Linalg 核心技术",{"id":152,"text":153,"level":154},"1-整体技术路线","1. 整体技术路线",4,{"id":156,"text":157,"level":154},"2-为何选择-linalg-dialect","2. 为何选择 Linalg Dialect？",{"id":159,"text":160,"level":154},"3-triton-linalg-转换原则","3. Triton → Linalg 转换原则",{"id":162,"text":163,"level":154},"4-triton-linalg-构成部分","4. Triton-Linalg 构成部分",{"id":165,"text":166,"level":154},"5-核心难点-指针分析","5. 核心难点：指针分析",{"id":168,"text":169,"level":154},"6-linalg-mlu-下降流程","6. Linalg → MLU 下降流程",{"id":171,"text":172,"level":154},"7-triton-linalg-开源项目","7. Triton-Linalg 开源项目",{"id":174,"text":174,"level":110},"加入训练营",{"id":176,"text":177,"level":154},"完整课程内容-请观看直播或查看课程回放","📺完整课程内容，请观看直播或查看课程回放：",{"id":179,"text":179,"level":110},"关注与交流"]