LLVM Pass 开发指南<\/h1>

一、LLVM 基础概念<\/h2>

LLVM 最初是 "Low Level Virtual Machine" 的缩写，但现在已发展为一个编译器基础架构(infrastructure)。它不是一个完整的编译器，而是将编译器所需功能以模块化形式实现并包装成库。<\/p>

现代编译器通常分为三个部分：<\/p>

前端(Frontend)<\/strong>：将高级语言代码转换为中间表示(IR)<\/li>
优化层(Optimization)<\/strong>：对IR进行分析和优化<\/li>

后端(Backend)<\/strong>：将优化后的IR转换为目标机器代码<\/li> <\/ol>
LLVM 采用这种分层设计，使其能够灵活适应不同编程语言和目标平台。<\/p>
二、环境配置<\/h2>
1. 源码获取<\/h3>
从 GitHub 下载 LLVM 源码(示例使用 LLVM 18.1.8)<\/p>
2. 依赖安装<\/h3>
sudo apt-get install g++ <\/span><\/span>sudo apt-get install make <\/span><\/span>sudo apt-get install cmake <\/span><\/span>sudo apt install ninja-build <\/span><\/span><\/code><\/pre>3. 编译LLVM<\/h3> cd llvm-project <\/span><\/span>sudo cmake -S llvm -B build -G Ninja -DCMAKE_BUILD_TYPE=<\/span>Release -DLLVM_ENABLE_PROJECTS=<\/span>"clang"<\/span> <\/span><\/span>cd build <\/span><\/span>ninja -j8 <\/span><\/span><\/code><\/pre>4. CLion配置<\/h3> 导入LLVM源码到CLion<\/li> 找到llvm\/CMakeLists.txt<\/code>并打开<\/li> 在Settings → CMake中添加Release配置<\/li> 设置CMake选项：<\/li> <\/ol> -G Ninja -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_PROJECTS="clang" -DBUILD_SHARED_LIBS=ON <\/code><\/pre> 三、LLVM Pass开发<\/h2> 1. 新版Pass管理器特点<\/h3> 使用基于概念的多态性而非显式接口<\/li> 所有Pass继承自PassInfoMixin<PassT><\/code>模板类<\/li> Pass必须实现run()<\/code>方法，返回PreservedAnalyses<\/code><\/li> <\/ul> 2. 创建Pass头文件<\/h3> 在llvm\/include\/llvm\/Transforms\/Utils\/<\/code>下创建Encode.h<\/code>:<\/p> #ifndef LLVM_TRANSFORMS_UTILS_ENCODE_H <\/span><\/span><\/span>#define LLVM_TRANSFORMS_UTILS_ENCODE_H <\/span><\/span><\/span><\/span> <\/span><\/span>#include<\/span> "llvm\/IR\/PassManager.h"<\/span> <\/span><\/span><\/span><\/span> <\/span><\/span>namespace<\/span> llvm { <\/span><\/span>class<\/span> EncodePass<\/span> :<\/span> public<\/span> PassInfoMixin<<\/span>EncodePass><\/span> { <\/span><\/span>public<\/span>:<\/span> <\/span><\/span> PreservedAnalyses run(Function &<\/span>F, FunctionAnalysisManager &<\/span>AM); <\/span><\/span>}; <\/span><\/span>} \/\/ namespace llvm <\/span><\/span><\/span><\/span> <\/span><\/span>#endif <\/span>\/\/ LLVM_TRANSFORMS_UTILS_ENCODE_H <\/span><\/span><\/span><\/code><\/pre>3. 实现Pass<\/h3> 在llvm\/lib\/Transforms\/Utils\/<\/code>下创建Encode.cpp<\/code>:<\/p> #include<\/span> "llvm\/Transforms\/Utils\/Encode.h"<\/span> <\/span><\/span><\/span><\/span> <\/span><\/span>using<\/span> namespace<\/span> llvm; <\/span><\/span> <\/span><\/span>PreservedAnalyses EncodePass::<\/span>run(Function &<\/span>F, FunctionAnalysisManager &<\/span>AM) { <\/span><\/span> errs() <<<\/span> F.getName() <<<\/span> "<\/span>\n<\/span>"<\/span>; <\/span><\/span> return<\/span> PreservedAnalyses::<\/span>all(); <\/span><\/span>} <\/span><\/span><\/code><\/pre>4. 修改构建配置<\/h3> 在llvm\/lib\/Transforms\/Utils\/CMakeLists.txt<\/code>中添加Encode.cpp<\/code><\/li> 在llvm\/lib\/Passes\/PassRegistry.def<\/code>中添加: FUNCTION_PASS("encode"<\/span>, EncodePass()) <\/span><\/span><\/code><\/pre><\/li> 在llvm\/lib\/Passes\/PassBuilder.cpp<\/code>中添加头文件引用: #include<\/span> "llvm\/Transforms\/Utils\/Encode.h"<\/span> <\/span><\/span><\/span><\/code><\/pre><\/li> <\/ol> 5. 编译Pass<\/h3> 在CLion中重新加载CMake项目，然后在cmake-build-release<\/code>目录下执行:<\/p> ninja -j2 opt <\/span><\/span><\/code><\/pre>注意<\/strong>：如果遇到循环依赖问题，检查lib\/support\/BLAKE3\/CMakeLists.txt<\/code>中是否意外添加了Encode.cpp相关内容。<\/p> 四、使用Pass<\/h2> 1. 准备测试代码<\/h3> 创建hello_clang.c<\/code>:<\/p> #include<\/span> <stdio.h><\/span> <\/span><\/span><\/span><\/span> <\/span><\/span>void<\/span> test_hello1<\/span>() { <\/span><\/span> printf("test_hello1<\/span>\n<\/span>"<\/span>); <\/span><\/span> return<\/span>; <\/span><\/span>} <\/span><\/span> <\/span><\/span>void<\/span> test_hello2<\/span>() { <\/span><\/span> printf("test_hello2<\/span>\n<\/span>"<\/span>); <\/span><\/span> return<\/span>; <\/span><\/span>} <\/span><\/span> <\/span><\/span>int<\/span> main<\/span>(int<\/span> argc, char<\/span> const<\/span> *<\/span>argv[]) { <\/span><\/span> printf("hello clang<\/span>\n<\/span>"<\/span>); <\/span><\/span> return<\/span> 0<\/span>; <\/span><\/span>} <\/span><\/span><\/code><\/pre>2. 生成LLVM IR<\/h3> clang -O3 -emit-llvm hello_clang.c -S -o hello_clang.ll <\/span><\/span><\/code><\/pre>重要<\/strong>：必须使用-O3<\/code>优化选项，否则生成的IR可能无法被Pass正确处理。<\/p> 3. 运行Pass<\/h3> 有两种方式运行Pass：<\/p> 直接使用opt工具<\/strong>:<\/p> opt -passes=<\/span>encode hello_clang.ll <\/span><\/span><\/code><\/pre><\/li> 构建.so文件<\/strong>(旧版方式)<\/p> <\/li> <\/ol> 五、应用场景<\/h2> LLVM Pass在以下领域有广泛应用：<\/p> 代码混淆(如OLLVM的控制流平坦化、虚假控制流)<\/li> IR信息提取<\/li> 静态分析<\/li> 代码优化<\/li> <\/ul> 相比自行开发脚本或插件，基于LLVM框架的项目更容易实现，因为它提供了丰富的API和基础设施。<\/p>

LLVM Pass 开发指南<\/h1>

二、环境配置<\/h2>

1. 源码获取<\/h3> 从 GitHub 下载 LLVM 源码(示例使用 LLVM 18.1.8)<\/p>

三、LLVM Pass开发<\/h2>

1. 源码获取<\/h3>
从 GitHub 下载 LLVM 源码(示例使用 LLVM 18.1.8)<\/p>