agentleFS
mohitmishra786mohitmishra786/low-level-dev-skills/skills/compiler-internals/llvm-passes/SKILL.md

llvm-passes

Skill290 starsChanged 7 months agoMITView on GitHub

LLVM passes skill for writing compiler optimizations. Use when writing FunctionPass or ModulePass, registering PassPlugins, running with opt, using analysis utilities, or testing with llvm-lit. Activates on queries about LLVM pass, PassPlugin, opt -passes, DominatorTree, llvm-lit, or New Pass Manager.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Always update SSA and invalidate analyses after structural changes.

    line 182

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

LLVM Passes

Purpose

Guide agents through writing LLVM optimization passes with the New Pass Manager: FunctionPass and ModulePass structure, PassPluginLibraryInfo registration, running via opt -load-pass-plugin, common analysis utilities (DominatorTree, LoopInfo, AliasAnalysis), IR modification patterns, llvm-lit testing, and debugging with opt -print-after-all.

When to Use

  • Adding a custom optimization to an LLVM-based compiler
  • Writing an IR transformation pass (inlining, DCE, custom lowering)
  • Analyzing control flow with dominator trees or loop info
  • Testing passes with FileCheck and llvm-lit
  • Debugging pass ordering and IR corruption
  • Integrating passes into Clang via plugin

Workflow

1. New Pass Manager architecture

opt / clang
├── ModulePassManager
│   └── FunctionPassManager (per function)
│       └── FunctionPass instances
└── AnalysisManager (cached analyses)

Passes declare analysis usage; analyses are invalidated on IR mutation.

2. Minimal FunctionPass (C++ plugin)

// MyPass.cpp
#include "llvm/IR/PassManager.h"
#include "llvm/Passes/PassBuilder.h"
#include "llvm/Passes/PassPlugin.h"
#include "llvm/Support/raw_ostream.h"

using namespace llvm;

namespace {

struct MyPass : PassInfoMixin<MyPass> {
    PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM) {
        bool changed = false;
        for (BasicBlock &BB : F) {
            for (Instruction &I : BB) {
                if (auto *Call = dyn_cast<CallInst>(&I)) {
                    if (Call->getCalledFunction() &&
                        Call->getCalledFunction()->getName() == "dead_func") {
                        Call->eraseFromParent();
                        changed = true;
                    }
                }
            }
        }
        return changed ? PreservedAnalyses::none() : PreservedAnalyses::all();
    }
};

} // namespace

extern "C" LLVM_ATTRIBUTE_WEAK PassPluginLibraryInfo llvmGetPassPluginInfo() {
    return {
        LLVM_PLUGIN_API_VERSION, "MyPass", "v0.1",
        [](PassBuilder &PB) {
            PB.registerPipelineParsingCallback(
                [](StringRef Name, FunctionPassManager &FPM,
                   ArrayRef<PassBuilder::PipelineElement>) {
                    if (Name == "my-pass") {
                        FPM.addPass(MyPass());
                        return true;
                    }
                    return false;
                });
        }
    };
}
# CMakeLists.txt
find_package(LLVM REQUIRED CONFIG)
add_library(MyPass MODULE MyPass.cpp)
target_include_directories(MyPass SYSTEM PRIVATE ${LLVM_INCLUDE_DIRS})
target_compile_definitions(MyPass PRIVATE ${LLVM_DEFINITIONS})
llvm_map_components_to_libnames(llvm_libs core passes support)
target_link_libraries(MyPass PRIVATE ${llvm_libs})
cmake -B build -DLLVM_DIR=$(llvm-config --cmakedir)
cmake --build build

3. Running with opt

# Run pass on IR file
opt -load-pass-plugin ./build/MyPass.so -passes=my-pass -S input.ll -o output.ll

# Print IR after each pass
opt -load-pass-plugin ./build/MyPass.so -passes=my-pass -print-after-all input.ll -o /dev/null

# Pass pipeline string
opt -passes="function(instcombine),my-pass,function(dce)" input.ll -S -o out.ll

4. ModulePass example

struct MyModulePass : PassInfoMixin<MyModulePass> {
    PreservedAnalyses run(Module &M, ModuleAnalysisManager &AM) {
        for (Function &F : M) {
            if (F.isDeclaration()) continue;
            // module-level transformation
        }
        return PreservedAnalyses::none();
    }
};

Register on ModulePassManager in plugin callback.

5. Analysis utilities

#include "llvm/Analysis/DominatorTree.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/AliasAnalysis.h"

PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM) {
    auto &DT = AM.getResult<DominatorTreeAnalysis>(F);
    auto &LI = AM.getResult<LoopAnalysis>(F);
    auto &AA = AM.getResult<AAManager>(F);

    for (Loop *L : LI) {
        BasicBlock *Header = L->getHeader();
        // Loop invariant code motion, etc.
    }

    DominatorTreeNode *IDom = DT.getNode(&F.getEntryBlock());
    (void)IDom;
    return PreservedAnalyses::all();
}

Declare analyses used:

AnalysisUsage MyLegacyPass::getAnalysisUsage(AnalysisUsage &AU) const {
    AU.addRequired<DominatorTreeWrapperPass>();
    AU.addRequired<LoopInfoWrapperPass>();
    return AU;
}

(New PM: analyses requested via AM.getResult<> — dependency auto-tracked.)

6. IR modification patterns

// Insert instruction before iterator
IRBuilder<> Builder(&*I.getIterator());
Value *NewVal = Builder.CreateAdd(I.getOperand(0), ConstantInt::get(I.getType(), 1));
I.replaceAllUsesWith(NewVal);
I.eraseFromParent();

// Clone basic block
BasicBlock *Clone = CloneBasicBlock(OrigBB, VMap, ".clone", &F);

// Create function
FunctionCallee Fn = M.getOrInsertFunction("my_fn",
    FunctionType::get(Builder.getVoidTy(), false));

Always update SSA and invalidate analyses after structural changes.

7. llvm-lit testing

test/
├── lit.cfg.py
└── my-pass.test
# my-pass.test
# RUN: opt -load-pass-plugin %shlibdir/MyPass.so -passes=my-pass -S %s | FileCheck %s

define void @test() {
  call void @dead_func()
  ret void
}

; CHECK-NOT: dead_func
llvm-lit test/my-pass.test -v

8. Debugging passes

# Verify IR after pass
opt -load-pass-plugin ./MyPass.so -passes=my-pass input.ll -o out.ll
opt -verify-each out.ll

# Time passes
opt -passes=my-pass -time-passes input.ll -o /dev/null

# Debug pass manager
OPT_DEBUG=1 opt -passes=my-pass input.ll -o /dev/null

Common Problems

SymptomCauseFix
Pass not found: my-passPlugin not loaded or name mismatchCheck registerPipelineParsingCallback name
Verify module failedBroken SSA after transformRun -verify-each; check use-def chains
Analysis staleModified IR without invalidationReturn PreservedAnalyses::none()
Plugin load errorLLVM version mismatchBuild against same LLVM as opt
Empty opt outputForgot -SAdd -S for textual IR
lit test FAILCHECK line mismatchopt -S and compare manually

Related Skills

  • skills/compilers/llvm — LLVM IR inspection and opt pipeline as user
  • skills/compiler-internals/compiler-frontend — IR generation feeding passes
  • skills/compiler-internals/jit-compilation — ORC JIT running optimized IR
  • skills/compiler-internals/mlir — MLIR passes (similar concepts)
  • skills/compilers/clang — Clang optimization flags
  • skills/build-systems/cmake — building pass plugins

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.