[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-EnzymeAD--Enzyme":3,"tool-EnzymeAD--Enzyme":64},[4,17,27,35,43,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",140436,2,"2026-04-05T23:32:43",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",107662,"2026-04-03T11:11:01",[13,14,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},3704,"NextChat","ChatGPTNextWeb\u002FNextChat","NextChat 是一款轻量且极速的 AI 助手，旨在为用户提供流畅、跨平台的大模型交互体验。它完美解决了用户在多设备间切换时难以保持对话连续性，以及面对众多 AI 模型不知如何统一管理的痛点。无论是日常办公、学习辅助还是创意激发，NextChat 都能让用户随时随地通过网页、iOS、Android、Windows、MacOS 或 Linux 端无缝接入智能服务。\n\n这款工具非常适合普通用户、学生、职场人士以及需要私有化部署的企业团队使用。对于开发者而言，它也提供了便捷的自托管方案，支持一键部署到 Vercel 或 Zeabur 等平台。\n\nNextChat 的核心亮点在于其广泛的模型兼容性，原生支持 Claude、DeepSeek、GPT-4 及 Gemini Pro 等主流大模型，让用户在一个界面即可自由切换不同 AI 能力。此外，它还率先支持 MCP（Model Context Protocol）协议，增强了上下文处理能力。针对企业用户，NextChat 提供专业版解决方案，具备品牌定制、细粒度权限控制、内部知识库整合及安全审计等功能，满足公司对数据隐私和个性化管理的高标准要求。",87618,"2026-04-05T07:20:52",[13,26],{"id":44,"name":45,"github_repo":46,"description_zh":47,"stars":48,"difficulty_score":23,"last_commit_at":49,"category_tags":50,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,51,52,53,15,54,26,13,55],"数据工具","视频","插件","其他","音频",{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,54],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":76,"owner_avatar_url":77,"owner_bio":78,"owner_company":79,"owner_location":79,"owner_email":80,"owner_twitter":79,"owner_website":81,"owner_url":82,"languages":83,"stars":122,"forks":123,"last_commit_at":124,"license":125,"difficulty_score":126,"env_os":127,"env_gpu":128,"env_ram":129,"env_deps":130,"category_tags":135,"github_topics":136,"view_count":23,"oss_zip_url":79,"oss_zip_packed_at":79,"status":16,"created_at":157,"updated_at":158,"faqs":159,"releases":188},2922,"EnzymeAD\u002FEnzyme","Enzyme","High-performance automatic differentiation of LLVM and MLIR.","Enzyme 是一款专为 LLVM 和 MLIR 编译器基础设施设计的高性能自动微分插件。在机器学习、科学计算及物理模拟等领域，开发者常需手动推导复杂的梯度公式或依赖效率有限的传统工具，这不仅耗时且容易出错。Enzyme 通过直接分析编译后的中间代码，自动合成高效的梯度计算逻辑，完美解决了这一痛点。\n\n它最大的技术亮点在于“基于优化代码的微分”能力。不同于仅在源代码层面操作的工具，Enzyme 能在编译器完成循环展开、向量化等高级优化之后再进行微分，从而生成极致的执行效率，性能往往媲美甚至超越当前最先进的专用微分库。此外，它对 GPU 内核（如 CUDA、ROCm）以及多种并行范式（OpenMP、MPI 等）提供了卓越的支持。\n\nEnzyme 非常适合编译器开发者、高性能计算研究人员以及需要构建自定义机器学习框架的工程师使用。无论是希望通过 Julia、Rust 还是 C++ 进行底层算法加速，Enzyme 都能让你无需重写外部代码，即可轻松获得快速且准确的梯度结果，是连接高级算法与底层硬件性能的强力桥梁。","# \u003Cimg src=\"https:\u002F\u002Fenzyme.mit.edu\u002Flogo.svg\" width=\"75\" align=left> The Enzyme High-Performance Automatic Differentiator of LLVM and MLIR\n\n\nEnzyme is a plugin that performs automatic differentiation (AD) of statically analyzable LLVM and MLIR.\n\nEnzyme can be used by calling `__enzyme_autodiff` on a function to be differentiated as shown below. \nRunning the Enzyme transformation pass then replaces the call to `__enzyme_autodiff` with the gradient of its first argument.\n```c\ndouble foo(double);\n\ndouble grad_foo(double x) {\n    return __enzyme_autodiff(foo, x);\n}\n```\n\nEnzyme is highly-efficient and its ability to perform AD on optimized code allows Enzyme to meet or exceed the performance of state-of-the-art AD tools.\n\n\u003Cdiv style=\"padding:2em\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FEnzymeAD_Enzyme_readme_ec6d931ce8ba.png\" width=\"500\" align=center>\n\u003C\u002Fdiv>\n\nDetailed information on installing and using Enzyme can be found on our website: [https:\u002F\u002Fenzyme.mit.edu](https:\u002F\u002Fenzyme.mit.edu).\n\nA short example of how to install Enzyme is below:\n```\ncd \u002Fpath\u002Fto\u002FEnzyme\u002Fenzyme\nmkdir build && cd build\ncmake -G Ninja .. -DLLVM_DIR=\u002Fpath\u002Fto\u002Fllvm\u002Flib\u002Fcmake\u002Fllvm -DLLVM_EXTERNAL_LIT=\u002Fpath\u002Fto\u002Flit\u002Flit.py\nninja\n```\n\nOr, install Enzyme using a package manager:\n\n[Homebrew](https:\u002F\u002Fbrew.sh)\n```\nbrew install enzyme\n```\n[Spack](https:\u002F\u002Fspack.io)\n```\nspack install enzyme\n```\n[Nix](https:\u002F\u002Fnixos.org\u002F)\n```\nnix-shell -p enzyme\n```\n\nTo get involved or if you have questions, please join our [mailing list](https:\u002F\u002Fgroups.google.com\u002Fd\u002Fforum\u002Fenzyme-dev).\n\nIf using this code in an academic setting, please cite the following three papers (first for Enzyme as a whole, second for GPU+optimizations, and third for AD of all other parallel programs (OpenMP, MPI, Julia Tasks, etc.)):\n```\n@inproceedings{NEURIPS2020_9332c513,\n author = {Moses, William and Churavy, Valentin},\n booktitle = {Advances in Neural Information Processing Systems},\n editor = {H. Larochelle and M. Ranzato and R. Hadsell and M. F. Balcan and H. Lin},\n pages = {12472--12485},\n publisher = {Curran Associates, Inc.},\n title = {Instead of Rewriting Foreign Code for Machine Learning, Automatically Synthesize Fast Gradients},\n url = {https:\u002F\u002Fproceedings.neurips.cc\u002Fpaper\u002F2020\u002Ffile\u002F9332c513ef44b682e9347822c2e457ac-Paper.pdf},\n volume = {33},\n year = {2020}\n}\n@inproceedings{10.1145\u002F3458817.3476165,\nauthor = {Moses, William S. and Churavy, Valentin and Paehler, Ludger and H\\\"{u}ckelheim, Jan and Narayanan, Sri Hari Krishna and Schanen, Michel and Doerfert, Johannes},\ntitle = {Reverse-Mode Automatic Differentiation and Optimization of GPU Kernels via Enzyme},\nyear = {2021},\nisbn = {9781450384421},\npublisher = {Association for Computing Machinery},\naddress = {New York, NY, USA},\nurl = {https:\u002F\u002Fdoi.org\u002F10.1145\u002F3458817.3476165},\ndoi = {10.1145\u002F3458817.3476165},\nbooktitle = {Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis},\narticleno = {61},\nnumpages = {16},\nkeywords = {CUDA, LLVM, ROCm, HPC, AD, GPU, automatic differentiation},\nlocation = {St. Louis, Missouri},\nseries = {SC '21}\n}\n@inproceedings{10.5555\u002F3571885.3571964,\nauthor = {Moses, William S. and Narayanan, Sri Hari Krishna and Paehler, Ludger and Churavy, Valentin and Schanen, Michel and H\\\"{u}ckelheim, Jan and Doerfert, Johannes and Hovland, Paul},\ntitle = {Scalable Automatic Differentiation of Multiple Parallel Paradigms through Compiler Augmentation},\nyear = {2022},\nisbn = {9784665454445},\npublisher = {IEEE Press},\nbooktitle = {Proceedings of the International Conference on High Performance Computing, Networking, Storage and Analysis},\narticleno = {60},\nnumpages = {18},\nkeywords = {automatic differentiation, tasks, OpenMP, compiler, Julia, parallel, Enzyme, C++, RAJA, hybrid parallelization, MPI, distributed, LLVM},\nlocation = {Dallas, Texas},\nseries = {SC '22}\n}\n```\n\nBoth [Julia bindings](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl#readme) and [Rust bindings](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002Frust#readme) are available for Enzyme.\n","# \u003Cimg src=\"https:\u002F\u002Fenzyme.mit.edu\u002Flogo.svg\" width=\"75\" align=left> LLVM 和 MLIR 的高性能自动微分工具 Enzyme\n\n\nEnzyme 是一个插件，用于对可静态分析的 LLVM 和 MLIR 代码执行自动微分（AD）。\n\n使用 Enzyme 时，只需在需要求导的函数上调用 `__enzyme_autodiff`，如下所示。运行 Enzyme 转换 pass 后，对 `__enzyme_autodiff` 的调用将被其第一个参数的梯度所替换。\n```c\ndouble foo(double);\n\ndouble grad_foo(double x) {\n    return __enzyme_autodiff(foo, x);\n}\n```\n\nEnzyme 效率极高，并且能够在优化后的代码上执行自动微分，因此其性能可以达到甚至超越当前最先进的自动微分工具。\n\n\u003Cdiv style=\"padding:2em\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FEnzymeAD_Enzyme_readme_ec6d931ce8ba.png\" width=\"500\" align=center>\n\u003C\u002Fdiv>\n\n有关安装和使用 Enzyme 的详细信息，请访问我们的官网：[https:\u002F\u002Fenzyme.mit.edu](https:\u002F\u002Fenzyme.mit.edu)。\n\n以下是一个简短的安装示例：\n```\ncd \u002Fpath\u002Fto\u002FEnzyme\u002Fenzyme\nmkdir build && cd build\ncmake -G Ninja .. -DLLVM_DIR=\u002Fpath\u002Fto\u002Fllvm\u002Flib\u002Fcmake\u002Fllvm -DLLVM_EXTERNAL_LIT=\u002Fpath\u002Fto\u002Flit\u002Flit.py\nninja\n```\n\n或者，您也可以通过包管理器安装 Enzyme：\n\n[Homebrew](https:\u002F\u002Fbrew.sh)\n```\nbrew install enzyme\n```\n[Spack](https:\u002F\u002Fspack.io)\n```\nspack install enzyme\n```\n[Nix](https:\u002F\u002Fnixos.org\u002F)\n```\nnix-shell -p enzyme\n```\n\n如需参与或有任何问题，请加入我们的 [邮件列表](https:\u002F\u002Fgroups.google.com\u002Fd\u002Fforum\u002Fenzyme-dev)。\n\n如果您在学术环境中使用此代码，请引用以下三篇论文（第一篇为 Enzyme 整体介绍，第二篇为 GPU 加速及优化相关工作，第三篇则涉及对其他并行程序（OpenMP、MPI、Julia Tasks 等）的自动微分）：\n```\n@inproceedings{NEURIPS2020_9332c513,\n author = {Moses, William and Churavy, Valentin},\n booktitle = {Advances in Neural Information Processing Systems},\n editor = {H. Larochelle and M. Ranzato and R. Hadsell and M. F. Balcan and H. Lin},\n pages = {12472--12485},\n publisher = {Curran Associates, Inc.},\n title = {Instead of Rewriting Foreign Code for Machine Learning, Automatically Synthesize Fast Gradients},\n url = {https:\u002F\u002Fproceedings.neurips.cc\u002Fpaper\u002F2020\u002Ffile\u002F9332c513ef44b682e9347822c2e457ac-Paper.pdf},\n volume = {33},\n year = {2020}\n}\n@inproceedings{10.1145\u002F3458817.3476165,\nauthor = {Moses, William S. and Churavy, Valentin and Paehler, Ludger and H\\\"{u}ckelheim, Jan and Narayanan, Sri Hari Krishna and Schanen, Michel and Doerfert, Johannes},\ntitle = {Reverse-Mode Automatic Differentiation and Optimization of GPU Kernels via Enzyme},\nyear = {2021},\nisbn = {9781450384421},\npublisher = {Association for Computing Machinery},\naddress = {New York, NY, USA},\nurl = {https:\u002F\u002Fdoi.org\u002F10.1145\u002F3458817.3476165},\ndoi = {10.1145\u002F3458817.3476165},\nbooktitle = {Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis},\narticleno = {61},\nnumpages = {16},\nkeywords = {CUDA, LLVM, ROCm, HPC, AD, GPU, automatic differentiation},\nlocation = {St. Louis, Missouri},\nseries = {SC '21}\n}\n@inproceedings{10.5555\u002F3571885.3571964,\nauthor = {Moses, William S. and Narayanan, Sri Hari Krishna and Paehler, Ludger and Churavy, Valentin and Schanen, Michel and H\\\"{u}ckelheim, Jan and Doerfert, Johannes and Hovland, Paul},\ntitle = {Scalable Automatic Differentiation of Multiple Parallel Paradigms through Compiler Augmentation},\nyear = {2022},\nisbn = {9784665454445},\npublisher = {IEEE Press},\nbooktitle = {Proceedings of the International Conference on High Performance Computing, Networking, Storage and Analysis},\narticleno = {60},\nnumpages = {18},\nkeywords = {automatic differentiation, tasks, OpenMP, compiler, Julia, parallel, Enzyme, C++, RAJA, hybrid parallelization, MPI, distributed, LLVM},\nlocation = {Dallas, Texas},\nseries = {SC '22}\n}\n```\n\nEnzyme 提供了 [Julia 绑定](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl#readme) 和 [Rust 绑定](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002Frust#readme)，方便用户在这些语言中使用。","# Enzyme 快速上手指南\n\nEnzyme 是一个基于 LLVM 和 MLIR 的高性能自动微分（Automatic Differentiation, AD）插件。它能够对静态可分析的代码进行高效的梯度计算，性能媲美甚至超越现有的主流 AD 工具，特别适用于科学计算和高性能计算场景。\n\n## 环境准备\n\n在开始之前，请确保您的开发环境满足以下要求：\n\n*   **操作系统**：Linux 或 macOS（Windows 需通过 WSL 或类似环境）。\n*   **核心依赖**：\n    *   **LLVM**：需要安装与 Enzyme 版本兼容的 LLVM 库及开发头文件。\n    *   **构建工具**：`CMake` (推荐 3.10+)，`Ninja` 或 `Make`。\n    *   **编译器**：支持 LLVM 后端的主流编译器（如 Clang\u002FGCC）。\n*   **可选依赖**：若需运行测试套件，需安装 `lit` (LLVM Integrated Tester)。\n\n> **提示**：对于国内开发者，若直接从源码编译遇到网络问题，建议优先使用包管理器安装，或配置国内镜像源加速 LLVM 相关依赖的下载。\n\n## 安装步骤\n\n您可以选择从源码编译或使用包管理器安装。\n\n### 方式一：使用包管理器（推荐）\n\n这是最快捷的安装方式，适合大多数用户。\n\n**macOS (Homebrew):**\n```bash\nbrew install enzyme\n```\n\n**Linux (Spack):**\n```bash\nspack install enzyme\n```\n\n**Nix 环境:**\n```bash\nnix-shell -p enzyme\n```\n\n### 方式二：从源码编译\n\n如果您需要特定版本的 LLVM 支持或进行开发调试，请按以下步骤操作：\n\n1.  克隆仓库并进入目录：\n    ```bash\n    cd \u002Fpath\u002Fto\u002FEnzyme\u002Fenzyme\n    ```\n2.  创建构建目录并配置 CMake（请替换 `\u002Fpath\u002Fto\u002Fllvm...` 为您实际的 LLVM 路径）：\n    ```bash\n    mkdir build && cd build\n    cmake -G Ninja .. -DLLVM_DIR=\u002Fpath\u002Fto\u002Fllvm\u002Flib\u002Fcmake\u002Fllvm -DLLVM_EXTERNAL_LIT=\u002Fpath\u002Fto\u002Flit\u002Flit.py\n    ```\n3.  执行编译：\n    ```bash\n    ninja\n    ```\n\n## 基本使用\n\nEnzyme 的核心用法是在 C\u002FC++ 代码中调用 `__enzyme_autodiff` 函数。编译时加载 Enzyme 插件后，该调用会被自动替换为对应函数的梯度计算逻辑。\n\n### 最简单示例\n\n假设您有一个函数 `foo`，想要计算其在点 `x` 处的导数：\n\n```c\ndouble foo(double);\n\ndouble grad_foo(double x) {\n    \u002F\u002F __enzyme_autodiff(目标函数，输入变量)\n    \u002F\u002F 运行时将被替换为 foo 在 x 处的梯度值\n    return __enzyme_autodiff(foo, x);\n}\n```\n\n**使用说明：**\n1.  编写包含上述代码的 C\u002FC++ 文件。\n2.  使用集成了 Enzyme _pass_ 的编译器命令进行编译（具体编译命令取决于您的构建系统如何链接 Enzyme 插件）。\n3.  程序运行时，`grad_foo` 将直接返回 `foo` 的导数值，无需手动推导公式。\n\n> **多语言支持**：除了 C\u002FC++，Enzyme 官方还提供了 [Julia bindings](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl#readme) 和 [Rust bindings](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002Frust#readme)，可在相应语言的生态中直接使用。","某高性能计算团队正在开发基于 CUDA 的自定义物理仿真器，需要为复杂的核函数实现反向传播以进行参数优化。\n\n### 没有 Enzyme 时\n- 开发者必须手动推导并编写梯度代码，面对数千行的 GPU 核函数，极易出现数学推导错误或索引越界。\n- 若借助传统自动微分工具，往往需要将核心算法从优化的 LLVM IR 或 C++ 重写为特定框架格式，导致原本针对硬件做的极致优化失效。\n- 每次修改物理模型公式后，都需要重新手工调整对应的梯度逻辑，迭代周期长达数天，严重拖慢研发进度。\n- 手动实现的梯度代码难以利用编译器的高级优化特性，运行效率通常低于理论峰值的 50%。\n\n### 使用 Enzyme 后\n- 只需在原有 C++\u002FCUDA 函数调用处添加 `__enzyme_autodiff` 标记，Enzyme 直接在 LLVM IR 层面自动生成精确的梯度核函数，彻底消除人工推导错误。\n- 直接对经过 clang\u002Fnvcc 高度优化后的二进制代码进行微分，无需重写任何业务逻辑，完美保留原有的内存布局与指令级优化。\n- 修改物理公式后，重新编译即可瞬间获得新的梯度实现，将算法迭代时间从“天”级缩短至“分钟”级。\n- 生成的梯度代码性能卓越，基准测试显示其执行效率常优于手写代码，甚至能超越主流深度学习框架的原生算子性能。\n\nEnzyme 让科学家能专注于物理模型创新，而将繁琐且易错的梯度计算交给编译器自动高效完成。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FEnzymeAD_Enzyme_57ee8448.png","EnzymeAD","Enzyme Automatic Differentiation Compiler","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002FEnzymeAD_519ef65a.png","",null,"enzyme-dev@googlegroups.com","https:\u002F\u002Fenzyme.mit.edu","https:\u002F\u002Fgithub.com\u002FEnzymeAD",[84,88,92,96,100,104,108,112,116,119],{"name":85,"color":86,"percentage":87},"LLVM","#185619",57.8,{"name":89,"color":90,"percentage":91},"C++","#f34b7d",36,{"name":93,"color":94,"percentage":95},"MLIR","#5EC8DB",3.6,{"name":97,"color":98,"percentage":99},"C","#555555",1.7,{"name":101,"color":102,"percentage":103},"CMake","#DA3434",0.5,{"name":105,"color":106,"percentage":107},"Starlark","#76d275",0.3,{"name":109,"color":110,"percentage":111},"Julia","#a270ba",0.1,{"name":113,"color":114,"percentage":115},"Fortran","#4d41b1",0,{"name":117,"color":118,"percentage":115},"Python","#3572A5",{"name":120,"color":121,"percentage":115},"Shell","#89e051",1572,157,"2026-04-03T13:09:43","NOASSERTION",4,"Linux, macOS","非必需。支持通过 LLVM\u002FMLIR 对 GPU 内核（如 CUDA, ROCm）进行自动微分，具体取决于底层 LLVM 配置及目标硬件。","未说明",{"notes":131,"python":132,"dependencies":133},"Enzyme 是 LLVM 和 MLIR 的自动微分插件，主要通过 C++ 使用。安装需要预先编译好的 LLVM 环境（需指定 LLVM_DIR）。虽然 README 未明确列出 Windows 支持，但提供了 Homebrew (macOS)、Spack 和 Nix 包管理器安装方式。学术引用需参考三篇特定论文。","未说明 (核心为 LLVM 插件，提供 Julia 和 Rust 绑定)",[85,93,134,101],"Ninja (构建工具)",[54,14,51,53,15,55,13,26],[137,138,139,140,141,142,143,144,145,146,147,148,149,150,151,152,153,154,155,156],"enzyme","llvm","ad","automatic-differentiation","c","cpp","tensorflow","pytorch","machine-learning","gradient","derivative","high-performance","rust","clang","deep-learning","scientific-computing","simulation","differentiable-programming","compiler","llvm-enzyme","2026-03-27T02:49:30.150509","2026-04-06T08:09:02.454550",[160,165,170,175,180,184],{"id":161,"question_zh":162,"answer_zh":163,"source_url":164},13509,"Enzyme 在计算 sqrt(0) 的导数时为什么会引入分支，导致性能下降？","这是 Enzyme 的故意设计决策。虽然数学上 sqrt 在 0 处的导数是未定义或无穷大，但 Enzyme 将其特殊处理为 0（即生成 `(x = 0.0) ? 0.0 : 1\u002F(sqrt(2.0*x))` 的逻辑），以确保自动微分几乎处处（almost everywhere）计算的是标准导数（基于 PAP 函数理论）。这种处理引入了比较和跳转指令（如 ucomisd -> je），在某些基准测试中可能导致约 40% 的性能下降。如果追求极致性能且能接受数学上的不严谨，可以尝试使用 `-fast-math` 编译选项，这可能会优化掉相关分支或使等价变换生效。","https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fissues\u002F1295",{"id":166,"question_zh":167,"answer_zh":168,"source_url":169},13510,"Enzyme 是否支持多源（multisource）自动微分？文档似乎过时了。","是的，Enzyme 支持多源功能。之前教程中的示例可能因版本更新而失效，但该功能本身并未移除。维护者已确认相关问题已解决，并正在更新官方网站文档以提供最新的用法说明。如果遇到阻碍，建议查看最新的 PR 或等待官网文档更新，也可以直接联系维护者（如 @ZuseZ4 或 @wsmoses）获取最新指引。","https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fissues\u002F1462",{"id":171,"question_zh":172,"answer_zh":173,"source_url":174},13511,"Enzyme 能否处理 C++ 中的虚函数（virtual functions）？","Enzyme 可以处理虚函数，但由于无法在编译期静态推导具体调用的函数类型，需要特殊处理。对于涉及虚函数调用的代码，Enzyme 需要在正向传递中缓存必要的信息（tape），以便在反向传递中使用。通常需要使用特定的命令行参数（如 `-Xclang -new-struct-path-tbaa`）来辅助类型分析。对于复杂的虚函数表场景，可能需要手动创建修改后的虚函数表或使用特定的包装器来明确正向和反向传递的返回值及缓存数据。","https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fissues\u002F316",{"id":176,"question_zh":177,"answer_zh":178,"source_url":179},13512,"遇到 'cannot find deal with ptr that isnt arg' 断言错误怎么办？","该错误通常发生在 Enzyme 分析包含原子操作（如 `atomicrmw`）的代码时，特别是当原子操作作用于纯整数内存时。维护者指出，这是因为活动性分析（ActivityAnalysis）未能正确识别此类原子操作为“非活跃”（inactive）。解决方案是修复内部逻辑，告知分析器如果 `atomicrmw` 操作的是纯整数内存，则应视为非活跃。该问题在较新的提交中已被修复，建议升级到最新版本或应用相关的补丁。","https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fissues\u002F677",{"id":181,"question_zh":182,"answer_zh":183,"source_url":164},13513,"为什么在使用 `-fast-math` 时 sqrt 相关的微分行为会发生变化？","在不使用 `-fast-math` 时，浮点数学严格遵循 IEEE 标准，此时 `sqrt(x^2)` 并不等价于 `|x|`，导致微分规则需要处理边界情况（如 0 值分支）。而启用 `-fast-math` 后，编译器会进行更激进的优化和代数简化，使得 `sqrt(x^2)` 被视为等价于 `|x|`，从而消除了特殊的分支逻辑，使整个微分过程运行得更顺畅且效率更高。如果性能至关重要且对极端边界值的数学严格性要求不高，建议启用此选项。",{"id":185,"question_zh":186,"answer_zh":187,"source_url":174},13514,"如何在编译时正确加载 Enzyme 插件以支持 C++ 项目？","对于 C++ 项目，通常需要使用 clang++ 并通过 `-Xclang -load` 参数加载 Enzyme 的共享库文件（如 `ClangEnzyme-\u003Cversion>.so`）。一个典型的编译命令示例如下：`clang++ file.cpp -Xclang -load -Xclang \u002Fpath\u002Fto\u002FClangEnzyme-12.so -O2 -Xclang -new-struct-path-tbaa`。确保路径指向正确的构建产物，并根据 LLVM\u002FClang 版本选择对应的插件版本。如果使用 LTO（链接时优化），还需配合 `-flto` 和相应的 linker 参数（如 `-Wl,-mllvm=-load=...`）。",[189,194,199,204,209,214,219,224,229,234,239,244,249,254,259,264,269,274,278,283],{"id":190,"version":191,"summary_zh":192,"released_at":193},72325,"v0.0.256","## 变更内容\n* 由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2760 中修复了 syrk 的导数问题\n\n\n**完整变更日志**: https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.255...v0.0.256","2026-03-23T05:34:14",{"id":195,"version":196,"summary_zh":197,"released_at":198},72326,"v0.0.255","## 变更内容\n* 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2758 中，@wsmoses 提升了 LLVM 的提交版本。\n* 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2759 中，@wsmoses 改进了 Julia 调用卷积修复，并将其拆分为独立的优化传递。\n\n\n**完整变更日志**: https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.254...v0.0.255","2026-03-23T01:50:55",{"id":200,"version":201,"summary_zh":202,"released_at":203},72327,"v0.0.254","## 变更内容\n* 由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2751 中升级了 LLVM\n* 由 @vimarsh6739 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2752 中修复了 Bazel 构建问题\n* 由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2754 中修复了未知内建函数的编译时错误\n* 由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2755 中优化了在需要时对 LI 的使用，并替换了 OrigLI\n\n\n**完整变更日志**: https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.253...v0.0.254","2026-03-20T19:41:56",{"id":205,"version":206,"summary_zh":207,"released_at":208},72328,"v0.0.253","## 变更内容\n* 更多 libdevice 标记为 noinline，由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2745 中完成\n* 修复 cmpxchg 上的虚构 phi 节点问题，由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2746 中完成\n* 处理导数的打包返回值，由 @vimarsh6739 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2748 中完成\n* 移除未使用的原函数参数，由 @vimarsh6739 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2741 中完成\n\n\n**完整变更日志**: https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.252...v0.0.253","2026-03-17T17:54:38",{"id":210,"version":211,"summary_zh":212,"released_at":213},72329,"v0.0.252","## 变更内容\n* MLIR：由 @etphipp 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2722 中为 scf.parallel、scf.reduce 和 scf.reduce.return 添加了前向模式支持。\n* 修复 MLIR 构建依赖和警告，由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2736 中完成。\n* 通过松散类型分析处理 lshr 操作，由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2738 中实现。\n* MLIR：修复结果属性的克隆问题，由 @Pangoraw 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2739 中完成。\n* 将某些 h 转换函数标记为保持链接性，由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2740 中完成。\n* PreserveNVVM：添加更多 bfloat 类型支持，由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2744 中实现。\n* 传播 Dereferenceable 属性，但将其从 DUP_NONEED 参数中移除，由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2742 中完成。\n\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.251...v0.0.252","2026-03-16T21:25:19",{"id":215,"version":216,"summary_zh":217,"released_at":218},72330,"v0.0.251","## 变更内容\n* ProbProg：@sbrantq 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2727 中增加了对 MCMC 状态的支持\n* MLIR：@etphipp 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2723 中修复了反向模式应用于 `scf.if` 和 `scf.parallel` 时的几个问题\n* 性能：@wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2730 中添加了 O1 级别的内联后优化\n* C++ NaN 检查：@wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2731 中实现\n\n## 新贡献者\n* @etphipp 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2723 中完成了首次贡献\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.250...v0.0.251","2026-03-11T14:35:55",{"id":220,"version":221,"summary_zh":222,"released_at":223},72331,"v0.0.250","## 变更内容\n* 由 @sbrantq 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2687 中为 ProbProg 添加 C API 属性\n* 由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2688 中将 Julia 1.12 版本添加到工作流矩阵\n* 修复：更新 API 以适配新的 LLVM，由 @avik-pal 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2690 中完成\n* 修复：returnShadow 大小问题，由 @sbrantq 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2693 中解决\n* ProbProg：纯净 StableHLO 迁移，由 @sbrantq 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2692 中完成\n* mlir：实现更多复数导数，由 @Pangoraw 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2691 中完成\n* ProbProg：原始 MCMC 实现，由 @sbrantq 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2695 中完成\n* ProbProg：支持额外的 logpdf 参数，由 @sbrantq 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2696 中实现\n* [mlir] 修复某些 scf\u002Faffine.for 结果不活跃时的 bug，由 @pengmai 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2697 中完成\n* 将 raja 注册表标记为不可微，由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2702 中完成\n* 向 CI 工作流中添加夜间发布步骤，由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2704 中实现\n* 删除 _codeql_detected_source_root，由 @Pangoraw 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2707 中完成\n* bazel：导出 PreserveNVVM，由 @vimarsh6739 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2705 中完成\n* 进一步拆分 bazel 配置，由 @vimarsh6739 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2709 中完成\n* 从 CMake 和 Bazel 中导出 PreserveNVVM，由 @vimarsh6739 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2710 中完成\n* 改进前向模式下的错误信息反混淆，由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2703 中完成\n* ProbProg：对角质量矩阵修复，由 @sbrantq 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2698 中完成\n* 修复 LLVM 版本升级问题，由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2717 中完成\n* 修复 Eigen::array 初始化语法，由 @cantonios 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2720 中完成\n* LLVM isZero 实现，由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2721 中完成\n* 暂时将 ActivityAnalysis 摘要标记为 xfail，由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2725 中完成\n* WIP：实现回溯功能，由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2718 中完成\n\n## 新贡献者\n* @cantonios 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2720 中完成了首次贡献\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.249...v0.0.250","2026-03-07T12:08:12",{"id":225,"version":226,"summary_zh":227,"released_at":228},72332,"nightly","Enzyme 的最新夜间构建版本。","2026-02-16T23:56:34",{"id":230,"version":231,"summary_zh":232,"released_at":233},72333,"v0.0.249","## 变更内容\n* @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2686 中进行了更多本地分配相关的活动\n\n\n**完整变更日志**: https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.248...v0.0.249","2026-01-31T22:46:03",{"id":235,"version":236,"summary_zh":237,"released_at":238},72334,"v0.0.248","## 变更内容\n* 由 @wsmoses 在 https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2685 中添加了对 cublas 的 int 类型调用支持\n\n\n**完整变更日志**: https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.247...v0.0.248","2026-01-31T11:41:47",{"id":240,"version":241,"summary_zh":242,"released_at":243},72335,"v0.0.247","## What's Changed\r\n* Skip debug fn by @wsmoses in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2684\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.246...v0.0.247","2026-01-31T04:18:34",{"id":245,"version":246,"summary_zh":247,"released_at":248},72336,"v0.0.246","## What's Changed\r\n* [mlir] Support mincut with nested regions by @pengmai in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2676\r\n* ProbProg: MCMC by @sbrantq in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2646\r\n* Keep rooted attribute for v by @wsmoses in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2683\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.245...v0.0.246","2026-01-31T00:11:47",{"id":250,"version":251,"summary_zh":252,"released_at":253},72337,"v0.0.245","## What's Changed\r\n* Ignore icmp usage of sret by @wsmoses in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2682\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.244...v0.0.245","2026-01-29T05:31:31",{"id":255,"version":256,"summary_zh":257,"released_at":258},72338,"v0.0.244","## What's Changed\r\n* Correct rooted ordering by @wsmoses in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2681\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.243...v0.0.244","2026-01-29T01:43:23",{"id":260,"version":261,"summary_zh":262,"released_at":263},72339,"v0.0.243","## What's Changed\r\n* Canonicalization: Remove unused primal arguments for region op by @vimarsh6739 in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2640\r\n* Revert \"Canonicalization: Remove unused primal arguments for region op\" by @vimarsh6739 in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2677\r\n* Fix batched attribute usage by @wsmoses in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2678\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.242...v0.0.243","2026-01-28T02:21:59",{"id":265,"version":266,"summary_zh":267,"released_at":268},72340,"v0.0.242","## What's Changed\r\n* cmake: make Enzyme Apple dynamic_lookup opt-in by @sgasho in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2670\r\n* LLVM integrate by @wsmoses in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2673\r\n* Fix syrk derivatives to use optimized symm+axpy logic by @wsmoses in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2675\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.241...v0.0.242","2026-01-27T09:18:47",{"id":270,"version":271,"summary_zh":272,"released_at":273},72341,"v0.0.241","## What's Changed\r\n* Fix custom zero segfault by @wsmoses in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2672\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.240...v0.0.241","2026-01-26T22:51:14",{"id":275,"version":276,"summary_zh":79,"released_at":277},72342,"v0.0.240","2026-01-25T23:53:01",{"id":279,"version":280,"summary_zh":281,"released_at":282},72343,"v0.0.239","## What's Changed\r\n* primal code motion by @vimarsh6739 in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2562\r\n* [NFC] Migrate enzyme-tblgen from builder.create\u003COp>() to Op::create() by @rupprecht in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2660\r\n* More llvm integration by @wsmoses in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2665\r\n* fix: memcpy -> memset uses destination alignment by @martinjrobins in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2664\r\n* Preserve enzymejl_gc_alloc_rt by @wsmoses in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2667\r\n* Fix needs cached check by @wsmoses in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2668\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.238...v0.0.239","2026-01-20T00:01:48",{"id":284,"version":285,"summary_zh":286,"released_at":287},72344,"v0.0.238","## What's Changed\r\n* Fix pointer address space issues by @wsmoses in https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fpull\u002F2662\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme\u002Fcompare\u002Fv0.0.237...v0.0.238","2026-01-16T15:13:07"]