ecmascript/compiler/codegen

Translates the optimized Circuit IR into machine code through two pluggable backends: LLVM and Maple/LiteCG. The backend is selected at build time via COMPILE_MAPLE flag. Primary Language: C++

Design

Both backends implement the same abstract interface (CodeGeneratorImpl) and follow the same flow:

  Circuit IR (after optimization passes)
       │
       ▼
  CodeGenerator (dispatches to active backend)
       │
       ├──▶ LLVMIRGeneratorImpl path:
       │      LLVMIRGeneratorImpl.GenerateCode()
       │        └── LLVMIRBuilder → builds LLVM IR
       │      LLVMAssembler.Run()
       │        └── LLVM JIT engine → machine code
       │
       └──▶ Maple/LiteCG path:
              LiteCGIRGeneratorImpl.GenerateCode()
                └── LiteCGIRBuilder → builds Maple LMIR
              LiteCGAssembler.Run()
                └── LiteCG backend → machine code

Backend abstraction:

• CodeGeneratorImpl — GenerateCodeForStub() and GenerateCode() translate Circuit gates to backend IR.
• Assembler (base class in assembler/) — Run() produces machine code from backend-specific IR.
• Target-specific builders (aarch64_builder, x64_builder) handle architecture differences within each backend.

Key Files

File	Role
llvm/llvm_codegen.h	LLVM backend: LLVMAssembler (IR → machine code)
llvm/llvm_ir_builder.cpp/.h	Circuit IR → LLVM IR translation
llvm/lib/llvm_interface.cpp/.h	LLVM shared library wrapper (libark_llvmcodegen)
llvm/aarch64/aarch64_builder.cpp	ARM64-specific LLVM IR patterns
llvm/x64/x64_builder.cpp	x64-specific LLVM IR patterns
maple/litecg_codegen.cpp/.h	Maple backend: LiteCGAssembler
maple/litecg_ir_builder.cpp/.h	Circuit IR → Maple LMIR translation
maple/BUILD.gn	Maple backend build configuration

How to Modify

Add support for a new Circuit gate in codegen:

1. Add the lowering case in llvm/llvm_ir_builder.cpp (LLVM path)
2. Add the lowering case in maple/litecg_ir_builder.cpp (Maple path)
3. Add target-specific handling in aarch64_builder / x64_builder if needed
4. Add test coverage in compiler/tests/ or codegen/maple/test/

Tune backend compilation options:

• LLVM: Modify LOptions in llvm/llvm_codegen.h (opt level, FP elimination, reloc mode)
• Maple: Modify litecgOptions vector passed to LiteCGAssembler

Building

Standalone repo (from repo root ../../):

python ark.py x64.debug

Full OHOS tree:

./build.sh --product-name <product-name> --build-target ark_js_host_linux_tools_packages

Tests

Standalone repo (from repo root ../../):

python ark.py x64.debug unittest   # all unit tests (includes codegen tests)

Full OHOS tree:

# Maple backend tests
./build.sh --product-name <product-name> --build-target ecmascript/compiler/codegen/maple/test:host_unittest

# Full compiler tests (includes backend integration)
./build.sh --product-name <product-name> --build-target ecmascript/compiler/tests:host_unittest

Boundaries

• Keep backend logic strictly inside maple/ or llvm/ — no cross-backend coupling.
• Both backends must handle the same set of Circuit gates; when adding gates, update both.
• Maintain target portability (aarch64 and x64) in both backends.
• Codegen ABI/layout changes must be validated by compiler host unittests.