A compiler driver coordinates source transformation and ordinary compilation. llmc++ can stop at generated C++, or continue to a native program:
# Generate function bodies and write main.llm.cpp. $ llmc++ --llm main.cpp # Reuse those cached bodies and build an executable. $ llmc++ -fllm-offline main.cpp -o main
The first command requests source output; the second requires usable cache entries and performs normal compilation. Generation needs a configured backend when the cache cannot satisfy a target.
The driver separates -fllm-* settings from arguments passed to Clang, including include paths and language options. The finished program contains ordinary compiled implementations, so calling a generated function does not contact a model.