Glossary

LLVM has its own dialect of compiler vocabulary — some terms are widely used across compiler textbooks, and some are specific to this codebase. The following entries are the ones most likely to confuse a new contributor.

Term	Meaning
LLVM IR	The intermediate representation that flows from front end to back end. Strongly typed, SSA, three-address. Lives as in-memory llvm::Module, bitcode, or textual .ll. See llvm/include/llvm/IR/.
SSA	Static Single Assignment. Every value is defined exactly once. The LLVM optimizer relies on SSA invariants.
Module	A llvm::Module (llvm/include/llvm/IR/Module.h) is roughly the IR equivalent of one translation unit: a collection of functions, globals, and metadata.
Pass	A transformation or analysis over IR (or MachineFunction). Run by the pass manager in llvm/lib/Passes/.
New PM	The "new pass manager", the current pass-manager API based on PassManager<> templates and PassBuilder. The "legacy PM" still exists for the codegen pipeline.
MIR	Machine IR — MachineFunction, MachineBasicBlock, MachineInstr. The IR used inside the back end after instruction selection. Serializable as .mir YAML. See llvm/include/llvm/CodeGen/.
MC	The "machine code" layer — target-independent assembler/disassembler/object-file machinery in llvm/lib/MC/.
MCA	LLVM Machine Code Analyzer (llvm/lib/MCA/) — performance analysis tool that simulates a CPU pipeline.
TableGen / .td	LLVM's data-description DSL. Used for instruction definitions, register classes, intrinsics, attributes, schedulers, ISel patterns, etc. Source in *.td files; build-time tool is llvm-tblgen (llvm/utils/TableGen/).
TargetMachine / Subtarget	Per-target classes describing CPU features and code-generation options (llvm/lib/Target/<Name>/).
GlobalISel	A modern instruction-selection framework in llvm/lib/CodeGen/GlobalISel/. Built around MachineIRBuilder and matchers.
SelectionDAG	The original instruction selector framework (llvm/lib/CodeGen/SelectionDAG/), still used by most targets.
MLIR	Multi-Level Intermediate Representation. A separate IR framework in mlir/ that supports user-defined "dialects".
Dialect	An MLIR concept — a namespaced collection of operations, types, and attributes. Examples: llvm, linalg, tensor, func.
Operation / Op	The MLIR equivalent of an instruction. Generic mlir::Operation plus dialect-specific C++ wrappers.
lit	The LLVM Integrated Tester (llvm/utils/lit/). Drives test files by extracting RUN: directives and shelling out commands.
FileCheck	A pattern-matching tool (llvm/utils/FileCheck/, library at llvm/lib/FileCheck/) used by lit tests to assert on compiler output.
opt	The LLVM IR optimizer driver (llvm/tools/opt/).
llc	The LLVM static codegen driver (llvm/tools/llc/) — runs the back end and emits assembly or object code.
clang driver	The user-facing executable that picks subcommands (preprocessor, compiler, assembler, linker) based on the input. Source in clang/lib/Driver/ and clang/tools/driver/.
clangd	The LSP language server in clang-tools-extra/clangd/ that powers IDE features.
Sema / AST / CodeGen (in Clang)	Three of Clang's main libraries: clang/lib/Sema/ (semantic analysis), clang/lib/AST/ (the C/C++/Objective-C AST), clang/lib/CodeGen/ (lower AST to LLVM IR).
CFE	"C front end" — historical alias for Clang. Still appears in commit messages and bug reports.
lld	The LLVM linker (lld/). Has separate ports for ELF, COFF, Mach-O, and WebAssembly.
lldb	The LLVM debugger (lldb/).
compiler-rt	The compiler runtime (compiler-rt/) — provides intrinsics like __udivsi3 plus the sanitizer runtimes (ASan, TSan, MSan, UBSan, etc.).
libc++ / libcxx	LLVM's C++ standard library implementation (libcxx/).
libc++abi / libcxxabi	LLVM's C++ ABI runtime (libcxxabi/) that pairs with libc++.
libunwind	LLVM's Itanium-ABI stack-unwinding library (libunwind/).
libc	LLVM's C standard library (libc/) — distinct from glibc/musl.
BOLT	Binary Optimization and Layout Tool (bolt/) — a post-link optimizer that rewrites code layout based on profile data.
ORC / ORCv2	"On-Request Compilation". The current LLVM JIT framework (llvm/lib/ExecutionEngine/Orc/ and the orc-rt/ runtime).
ThinLTO	A scalable form of link-time optimization (llvm/lib/LTO/).
DTLTO	Distributed ThinLTO (llvm/lib/DTLTO/) — runs ThinLTO across a build farm.
PGO	Profile-Guided Optimization. Source in llvm/lib/Transforms/Instrumentation/ and compiler-rt/lib/profile.
PDB / DWARF	Debug-info formats. DWARF code: llvm/lib/DebugInfo/. PDB code lives there too.
CIR	Clang IR — an MLIR dialect Clang can emit between AST and LLVM IR. See clang/lib/CIR/.
Polly	A polyhedral loop optimizer (polly/) that hooks into LLVM as a transform-pass plugin.
OpenMP	Both the language extension and a runtime. Runtime lives in openmp/; device-side execution lives in offload/.
SYCL	A Khronos C++ programming model for accelerators. The libsycl/ subproject is LLVM's runtime implementation.
OpenCL C	A C dialect for accelerator kernels. The libclc/ library implements its standard library.
flang-rt	The Fortran runtime (flang-rt/) used by code that Flang compiles.
target triple	A <arch>-<vendor>-<os>-<env> string that fully identifies a compilation target (e.g., x86_64-unknown-linux-gnu, aarch64-apple-darwin).
ABI	Application Binary Interface — the rules for how compiled code interoperates (calling conventions, struct layout, name mangling, exception unwinding).
maintainer	The person or persons responsible for an area, listed in the per-subproject Maintainers.md. They merge and review code in their area. See the LLVM developer policy.