gcc backend (initial pass) via libgccjit

[SamuelMarks]

NOTE: this is not intended to be merged; just trying to start a conversation…

---

PR: Native GCC (libgccjit) Backend (gcc_gen_be) with Modular Subsystems

Overview

This pull request introduces a native, direct code-generation backend in the EDG C++ front end (cpfe) powered by libgccjit (gcc_gen_be).

Historically, EDG lowers its Intermediate Language (IL) primarily to C or C++ source code (c_gen_be, cp_gen_be) or passes IL to proprietary code generators. This PR establishes a direct lowering pipeline from EDG's lowered Intermediate Language directly to libgccjit IR. This enables cpfe to directly emit machine code artifacts (object files .o, assembly .s, and shared libraries .so/.dylib/.dll) without requiring a secondary textual compiler invocation.

Across the commits comprising this branch, the backend is architected into 10 decoupled, testable subsystems with comprehensive CLI options, full CMake/CTest/CPack integration, dynamic loader support across POSIX and Windows, and robust coverage verification tooling.

---

Commit Summary

1. cc0d088356 GCC [libgccjit] backend (initial pass):
   • Established the baseline gcc_gen_be architecture, hooking into src/cfe.c, src/cmd_line.[ch], and src/host_envir.h.
   • Introduced CMake detection via cmake/FindLibGCCJIT.cmake and cross-platform CPack packaging.
   • Implemented initial end-to-end IL lowering across scalar/aggregate types, declarations, control flow (conditionals, loops, switch tables, inline assembly), and expression trees.
2. 533a5b95ac Modularize GCC backend into dedicated subsystems and expand CLI options:
   • Refactored the monolithic implementation into 10 dedicated compilation units (gcc_gen_be_*.c and matching headers).
   • Introduced centralized error handling (gcc_gen_be_error.h) with typed error codes (gcc_gen_be_error_t), error stringification, and GCC_GEN_BE_NODISCARD / GCC_GEN_BE_CHECK error propagation macros.
   • Expanded driver CLI options in src/cmd_line.[ch] for backend optimization levels (-O0 through -O3, -Os, -Ofast), debug generation (-g), position-independent code (-fPIC, -fPIE), and JIT intermediate representation dumping (--dump-initial-tree, --dump-gimple).
   • Expanded the test suite under tests/gcc_be/ with 11 targeted test suites and modernized check_coverage.py to evaluate line and branch coverage across all backend compilation units.

---

Architectural Highlights

1. Modular Subsystem Architecture

The backend is decomposed into specialized subsystems located under src/:

Subsystem	Header / Source	Description
Error Handling	gcc_gen_be_error.[ch]	Defines gcc_gen_be_error_t status codes, GCC_GEN_BE_NODISCARD attributes, GCC_GEN_BE_CHECK macro, and gcc_gen_be_error_string().
Context & State	gcc_gen_be_context.[ch]	Manages JIT context lifecycle (gcc_gen_be_init, gcc_gen_be_cleanup), active blocks, break/continue/switch exit stacks, cleanup stacks, and driver options.
Memoization & Cache	gcc_gen_be_cache.[ch]	Fast hash-map memoization structures mapping EDG AST pointers (a_type_ptr, a_variable_ptr, a_routine_ptr, a_label_ptr, a_field_ptr) to JIT equivalents.
Dynamic Library Loader	gcc_gen_be_lib_loader.[ch]	Cross-platform runtime loader handling dynamic symbol resolution (dlopen on POSIX, LoadLibraryA on Windows with function wrappers) to avoid hard build-time link dependencies where appropriate.
Location Tracking	gcc_gen_be_location.[ch]	Translates EDG sequence numbers (conv_seq_to_file_and_line) from expressions and statements into gcc_jit_location* records for debug info and diagnostics.
Type Lowering	gcc_gen_be_type.[ch]	Translates scalar types, pointers, arrays, aggregate structs/classes/unions (with cycle detection via opaque forward declarations), bitfields, and routine signatures.
Declaration Lowering	gcc_gen_be_decl.[ch]	Lowers global variables, local variables, parameters, linkage classification (INTERNAL, EXPORTED, IMPORTED), and dynamic initializers.
Expression Lowering	gcc_gen_be_expr.[ch]	Lowers expressions to gcc_jit_rvalue* and gcc_jit_lvalue*: constants, arithmetic/bitwise/logical operations, comparisons, casts, assignments, calls, and builtins.
Statement & CFG Lowering	gcc_gen_be_stmt.[ch]	Builds basic blocks and control-flow graphs for blocks, expressions, conditionals (if, constexpr if, if consteval), multi-way switch tables, loops (while, do-while, for), gotos, and extended inline asm.
Main Driver	gcc_gen_be_main.[ch]	Orchestrates the top-level backend entry point (gcc_gen_be(), back_end()), visits the primary scope, compiles to target files, and handles backend error reporting.

2. Frontend Integration & CLI Switches

• Minimal Core Intrusion: EDG core files (src/cfe.c, src/cmd_line.c, src/cmd_line.h, src/host_envir.h) guard backend hooks behind #if BACK_END_IS_GCC_GEN_BE.
• Output Target Redirection: Command-line flag -o sets gcc_be_output_file_name to determine the emitted artifact kind based on extension (.o for object files, .s for assembly, .so/.dylib/.dll for shared libraries).
• Optimization & Debug Flags: Added CLI option descriptors in src/cmd_line.c and flags in src/cmd_line.h:
  • Optimization levels: --O0, --O1, --O2, --O3, --Os, --Ofast
  • Debugging: -g
  • Code generation flags: -fPIC, -fPIE
  • Tree/GIMPLE dumps: --dump-initial-tree, --dump-gimple

3. Build, Packaging & CTest Integration

• cmake/FindLibGCCJIT.cmake: Probes standard system directories (Linux multiarch, Homebrew, MacPorts, MinGW/MSYS2), creates LibGCCJIT::LibGCCJIT, and inspects header feature availability macros (LIBGCCJIT_HAVE_gcc_jit_context_add_command_line_option, LIBGCCJIT_HAVE_gcc_jit_version, etc.).
• Cross-Platform RPATH & Linking: Automatically sets $ORIGIN/../lib on Linux/FreeBSD and @executable_path/../lib;@rpath on macOS. Configures /DELAYLOAD:libgccjit.dll for MSVC.
• Coverage Instrumentation: Automatic --coverage and -fprofile-arcs -ftest-coverage configuration for all gcc_gen_be_*.c compilation units when building with GCC or Clang.
• CPack Packaging: Preconfigured Debian (.deb), Red Hat (.rpm), macOS (.dmg, .tar.gz), Windows (.zip, .msi), and FreeBSD (.txz) packagers with runtime library dependencies.

---

File-by-File Breakdown

File	Change Summary
CMakeLists.txt	Adds CMake module path; declares EDG_ENABLE_GCC_BACKEND option; enables testing (CTest); adds tests/gcc_be subdirectory; defines multi-platform CPack packager metadata.
cmake/FindLibGCCJIT.cmake	Finds libgccjit.h and libraries; creates imported target LibGCCJIT::LibGCCJIT; checks feature availability macros.
src/CMakeLists.txt	Conditionally compiles all gcc_gen_be_*.c sources; links LibGCCJIT::LibGCCJIT; sets platform RPATHs and delay-load settings; adds coverage flags for GCC backend sources.
src/cfe.c	Adds conditional includes for gcc_gen_be_main.h and gcc_gen_be_context.h; registers backend invocation.
src/cmd_line.c	Registers backend options (-O0..-O3, -Os, -Ofast, -g, -fPIC, -fPIE, dumps, -o); parses and assigns corresponding variables.
src/cmd_line.h	Declares option kind enums and thread-local configuration variables (gcc_be_output_file_name, gcc_be_opt_level, gcc_be_debug_info, gcc_be_fPIC, etc.).
src/host_envir.h	Configures BACK_END_IS_GCC_GEN_BE; deactivates BACK_END_IS_C_GEN_BE; asserts DO_IL_LOWERING requirement.
src/gcc_gen_be_error.[ch]	New module: centralized status codes (gcc_gen_be_error_t), error strings, and propagation macros.
src/gcc_gen_be_context.[ch]	New module: global context state, option application, control stacks, and lifecycle management.
src/gcc_gen_be_cache.[ch]	New module: hash-map caches for AST-to-JIT memoization across types, vars, functions, labels, and fields.
src/gcc_gen_be_lib_loader.[ch]	New module: runtime dynamic loader and function wrappers for Windows and POSIX.
src/gcc_gen_be_location.[ch]	New module: source position and AST node location translation to JIT locations.
src/gcc_gen_be_type.[ch]	New module: lowering scalar types, pointers, arrays, structs/unions/classes, and routine prototypes.
src/gcc_gen_be_decl.[ch]	New module: lowering variable declarations, parameters, routines, and initializers.
src/gcc_gen_be_expr.[ch]	New module: lowering lvalues and rvalues (operators, casts, calls, intrinsics).
src/gcc_gen_be_stmt.[ch]	New module: lowering basic blocks, loops, switch statements, jumps, and inline asm.
src/gcc_gen_be_main.[ch]	New module: driver entry points gcc_gen_be() and back_end(), and compilation output.
tests/gcc_be/CMakeLists.txt	Defines CTest test cases for C and C++ test files and registers coverage targets (coverage_gcc_be).
tests/gcc_be/check_coverage.py	Automated gcov test runner calculating line and branch coverage across all gcc_gen_be_*.c files.
tests/gcc_be/test_*.c[pp]	Test source fixtures for scalars, types, expressions, short-circuiting, control flow, functions, aggregates, C++ classes, virtual methods, exception handling, inline asm, and edge cases.

---

Verification & Testing

• Configuration & Build: Verified CMake configuration and compilation under EDG_ENABLE_GCC_BACKEND=ON.
• Target Platforms: Validated build configuration paths on Linux, macOS (Apple Silicon & x86_64), FreeBSD, and Windows (MSVC delay-load & MinGW).
• Test Suite: Verified test discovery and fixture setups via ctest and target run_gcc_be_tests.
• Coverage Runner: Modernized check_coverage.py to evaluate execution and branch metrics across all modular backend compilation units.

[wchilders-nvidia]

Thanks, I've opened #245 for a central discussion of the topic of some kind of "native IR" back end for the EDG compiler project.