Examples

Enabling ClangIR

By default, Clang goes through its traditional CodeGen pipeline. After you've built a ClangIR-enabled Clang, you can pass -fclangir to enable the ClangIR pipeline. Godbolt also hosts a ClangIR-enabled compiler for quick experimentation.

Emitting CIR

Pass -S -Xclang -emit-cir to Clang to emit CIR instead of assembly. Godbolt shows an example with various language features and how they're translated to CIR.

Viewing the pass pipeline

ClangIR runs its own pass pipeline. Some useful flags to introspect this pipeline are:

• -mmlir --mlir-print-ir-before-all prints the CIR before each pass.
• -mmlir --mlir-print-ir-after-all prints the CIR after each pass.
• -mmlir --mlir-print-ir-before=<pass> and -mmlir --mlir-print-ir-after=<pass> print the CIR before and after a particular pass, respectively.

One particularly useful pass to introspect is cir-lowering-prepare (LoweringPrepare), which goes from higher-level CIR constructs to lower-level ones. Godbolt's pipeline viewer is convenient for this.

Emitting flat CIR

ClangIR tries to preserve high-level scope information till the end of its pipeline, where it runs a flattening pass to convert to pure basic block control flow. You can pass -Xclang -emit-cir-flat to see this flattened representation. Godbolt shows the difference in the representations.

Generating LLVM IR

Pass -fclangir -S -emit-llvm to emit LLVM through the ClangIR pipeline. Godbolt shows an example. All the standard Clang flags can be used as well, e.g. to build with optimizations.

Using the C++ lifetime-checker

via clang driver

TBD

via clang-tidy / clangd

TBD