llvm-ir-and-passes
Skill290 starsChanged 7 months agoMITView on GitHub
LLVM IR and passes skill for reading compiler IR. Use when analyzing LLVM IR, understanding SSA, pass pipeline order, or running opt on bitcode. Activates on queries about LLVM IR, SSA form, opt passes, llvm-dis, pass pipeline, or reading .ll files.
In this repository142 other files it gives its agents.
- AGENTS.mdAGENTS.md
- custom-allocatorsSkill
- numa-programmingSkill
- af-xdpSkill
- dpdkSkill
- io-uringSkill
- adc-dac-baremetalSkill
- baremetal-startupSkill
- bootloaders-embeddedSkill
- datasheet-and-refmanual-readingSkill
- dma-baremetalSkill
- gpio-baremetalSkill
- interrupts-and-exceptions-baremetalSkill
- low-power-embeddedSkill
- mmio-and-bit-manipulationSkill
- peripherals-from-datasheetSkill
- spi-i2c-baremetalSkill
- stm32-baremetalSkill
- timers-pwm-baremetalSkill
- uart-serial-baremetalSkill
- binutilsSkill
- dynamic-linkingSkill
- elf-inspectionSkill
- linkers-ltoSkill
- bazelSkill
- build-accelerationSkill
- cmakeSkill
- conan-vcpkgSkill
- include-what-you-useSkill
- makeSkill
- mesonSkill
- ninjaSkill
- static-analysisSkill
- code-generation-and-backendsSkill
- compiler-frontendSkill
- compiler-optimizations-deepSkill
- jit-compilationSkill
- llvm-passesSkill
- mlirSkill
- clangSkill
- cpp-modulesSkill
- cpp-templatesSkill
- cross-gccSkill
- gccSkill
- llvmSkill
- msvc-clSkill
- pgoSkill
- abi-and-calling-conventionsSkill
- branch-prediction-and-speculationSkill
- cpu-pipelines-and-hazardsSkill
- memory-hierarchy-and-cachesSkill
- virtual-memory-paging-and-tlbSkill
- concurrency-debuggingSkill
- core-dumpsSkill
- debug-optimized-buildsSkill
- dwarf-debug-formatSkill
- gdbSkill
- lldbSkill
- embedded-rustSkill
- freertosSkill
- The first 60 shown.
The file
Files
LLVM IR and Passes
Purpose
Teach agents to read and reason about LLVM IR (SSA, types, terminators), navigate the standard pass pipeline, and use opt/llvm-dis for inspection. Complements skills/compilers/llvm (toolchain) and skills/compiler-internals/llvm-passes (writing plugins) — not merged.
When to Use
- Understanding what
-O2changed in generated IR - Triaging miscompiles between Clang versions
- Preparing to write a custom pass
- Teaching SSA and dominance without writing C++ passes yet
Workflow
1. LLVM IR structure
; Function in SSA form
define i32 @add(i32 %a, i32 %b) {
entry:
%sum = add i32 %a, %b
ret i32 %sum
}Key concepts: basic blocks, PHI nodes at merges, typed values (i32, ptr, vectors).
2. Emit IR from Clang
clang -S -emit-llvm -O0 -o foo.ll foo.c
clang -c -emit-llvm -O2 -o foo.bc foo.c
llvm-dis foo.bc -o foo.ll3. Standard optimization pipeline (New PM)
Clang -O2 roughly runs:
SimplifyCFG → InstCombine → GVN → LICM → LoopVectorize → ...Inspect with:
opt -passes='default<O2>' -S foo.ll -o foo-opt.ll
opt -passes='default<O2>' -print-passes foo.ll 2>&1 | head4. Useful analysis passes
| Pass | Shows |
-passes=print<domtree> | Dominator tree |
-passes=print<loops> | Loop nests |
-passes=print-alias-sets | Alias sets |
opt -passes='print<domtree>' -disable-output foo.ll5. Reading PHI and UB
merge:
%v = phi i32 [ %a, %then ], [ %b, %else ]undef and poison in IR model LLVM poison semantics — distinct from C UB but related.
6. Compare before/after
opt -passes='instcombine,simplifycfg' -S foo.ll -o - | diff -u foo.ll -7. Agent usage
/llvm-ir-and-passes Explain this PHI node and which pass likely created itCommon Problems
| Symptom | Cause | Fix |
| Pass not found | LLVM version rename | opt --print-passes for your build |
| IR mismatch | Different LLVM major | Match clang/opt versions |
| "optnone" blocks opts | -O0 attribute | Compile with -O1+ |
| Huge IR | Inlined headers | -fno-discard-value-names off; filter function |
| Wrong pipeline | Legacy vs NPM | Use -passes= syntax |
Related Skills
skills/compilers/llvm— toolchain overviewskills/compiler-internals/llvm-passes— writing PassPluginsskills/compiler-internals/compiler-optimizations-deep— RA and ISelskills/compiler-internals/compiler-frontend— IR generationskills/rust/rustc-basics—rustc --emit=llvm-ir
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
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.

