add clang frontend support for more reliable ABI generation

.clang-format

@@ -0,0 +1,2 @@
+BasedOnStyle: LLVM
+LineEnding: LF

.gitignore

@@ -26,3 +26,4 @@ __pycache__
 
 # Build folder
 build/
+llvm-project/

CLAUDE.md

@@ -0,0 +1,33 @@
+# LLVM-CBE Build Instructions
+
+## Install CMake (if needed)
+
+```bash
+curl -L -O https://github.com/Kitware/CMake/releases/download/v3.30.0/cmake-3.30.0-linux-x86_64.tar.gz
+tar -xzf cmake-3.30.0-linux-x86_64.tar.gz
+export PATH=$PWD/cmake-3.30.0-linux-x86_64/bin:$PATH
+```
+
+## Build LLVM (if needed)
+
+```bash
+git clone --single-branch --branch llvmorg-20.1.8 --depth 1 https://github.com/llvm/llvm-project.git
+cd llvm-project
+cmake -S llvm -B build -G "Ninja" -DCMAKE_BUILD_TYPE=Debug -DLLVM_ENABLE_PROJECTS="clang" -DLLVM_ENABLE_ASSERTIONS=ON -DBUILD_SHARED_LIBS=ON
+ninja
+```
+
+## Run Tests
+
+### pytest
+```bash
+cd /home/vtjnash/llvm-cbe
+pip install pytest pytest-xdist
+pytest -n 4
+```
+
+### Unit tests
+```bash
+cd /home/vtjnash/llvm-project/llvm/build
+make CBEUnitTests && projects/llvm-cbe/unittests/CWriterTest
+```

CMakeLists.txt

@@ -13,10 +13,13 @@ if (NOT DEFINED LLVM_VERSION_MAJOR)
   include(AddLLVM)
 
   set (CMAKE_CXX_STANDARD 17)
+else()
+  set (USE_SYSTEM_LLVM 0)
 endif()
 
 add_subdirectory(lib)
 add_subdirectory(tools)
+add_subdirectory(clang-plugin)
 
 if( LLVM_INCLUDE_TESTS )
   add_subdirectory(unittests)

README.md

@@ -3,6 +3,10 @@ llvm-cbe
 
 This LLVM C backend has been resurrected by Julia Computing with various improvements.
 
+See clang-plugin/README for an enhanced version which provides much more reliable support for linking the output against existing C libraries.
+Linking against C++ libraries (or other non-C front-ends ABIs) is not likely to be reliable,
+unless it has been defined in terms of a C ABI (such as GCC's Itanium ABI that it uses for most C++ code).
+
 Installation instructions
 =========================
 

clang-plugin/CBackend.cpp

@@ -0,0 +1,91 @@
+//===-- CBackend.cpp - C Backend ABI Plugin ------------------------------===//
+//
+// Loadable plugin that provides a type-preserving ABI for LLVM C backend.
+// Load with: clang -Xclang -load -Xclang CBackendPlugin.so
+//
+// Triple format: llvm32-<os>[-<env>] or llvm64-<os>[-<env>]
+//   e.g. llvm32-linux-gnu, llvm64-linux-gnu, llvm64-darwin
+//
+//===----------------------------------------------------------------------===//
+
+#include "ABIInfo.h"
+#include "ABIInfoImpl.h"
+#include "TargetInfo.h"
+#include "clang/CodeGen/CGFunctionInfo.h"
+#include "llvm/Support/Compiler.h"
+#include "llvm/TargetParser/Triple.h"
+
+using namespace clang;
+using namespace clang::CodeGen;
+
+namespace {
+
+class CBackendABIInfo : public ABIInfo {
+public:
+  CBackendABIInfo(CodeGenTypes &CGT) : ABIInfo(CGT) {}
+
+  ABIArgInfo classifyReturnType(QualType RetTy) const {
+    if (RetTy->isVoidType())
+      return ABIArgInfo::getIgnore();
+    if (RetTy->isSignedIntegerOrEnumerationType())
+      return ABIArgInfo::getSignExtend(RetTy);
+    else if (RetTy->isUnsignedIntegerOrEnumerationType())
+      return ABIArgInfo::getZeroExtend(RetTy);
+    return ABIArgInfo::getDirect(nullptr, 0, nullptr, /*CanBeFlattened*/ false);
+  }
+
+  ABIArgInfo classifyArgumentType(QualType Ty) const {
+    Ty = useFirstFieldIfTransparentUnion(Ty);
+    if (Ty->isVoidType())
+      return ABIArgInfo::getIgnore();
+    if (Ty->isSignedIntegerOrEnumerationType())
+      return ABIArgInfo::getSignExtend(Ty);
+    else if (Ty->isUnsignedIntegerOrEnumerationType())
+      return ABIArgInfo::getZeroExtend(Ty);
+    if (isAggregateTypeForABI(Ty)) {
+      // Structures with either a non-trivial destructor or a non-trivial
+      // copy constructor are always indirect.
+      if (CGCXXABI::RecordArgABI RAA = getRecordArgABI(Ty, getCXXABI())) {
+        return getNaturalAlignIndirect(Ty, /*ByVal=*/RAA ==
+                                               CGCXXABI::RAA_DirectInMemory);
+      }
+    }
+    return ABIArgInfo::getDirect(nullptr, 0, nullptr, /*CanBeFlattened*/ false);
+  }
+
+  void computeInfo(CGFunctionInfo &FI) const override {
+    if (!::classifyReturnType(getCXXABI(), FI, *this))
+      FI.getReturnInfo() = classifyReturnType(FI.getReturnType());
+    for (auto &Arg : FI.arguments())
+      Arg.info = classifyArgumentType(Arg.type);
+  }
+
+  RValue EmitVAArg(CodeGenFunction &CGF, Address VAListAddr, QualType Ty,
+                   AggValueSlot Slot) const override {
+    return emitVoidPtrVAArg(CGF, VAListAddr, Ty, /*IsIndirect=*/false,
+                            CGF.getContext().getTypeInfoInChars(Ty),
+                            CharUnits::fromQuantity(8),
+                            /*AllowHigherAlign=*/true, Slot);
+  }
+};
+
+class CBackendTargetCodeGenInfo : public TargetCodeGenInfo {
+public:
+  CBackendTargetCodeGenInfo(CodeGenTypes &CGT)
+      : TargetCodeGenInfo(std::make_unique<CBackendABIInfo>(CGT)) {}
+};
+} // namespace
+
+namespace clang {
+namespace CodeGen {
+inline std::unique_ptr<TargetCodeGenInfo>
+createCBackendTargetCodeGenInfo(CodeGenModule &CGM) {
+  return std::make_unique<CBackendTargetCodeGenInfo>(CGM.getTypes());
+}
+} // namespace CodeGen
+} // namespace clang
+
+extern "C" LLVM_ABI_EXPORT void *getCBackendABIForTarget(const llvm::Triple *T,
+                                                         CodeGenModule *CGM) {
+  return createCBackendTargetCodeGenInfo(*CGM).release();
+}

clang-plugin/CMakeLists.txt

@@ -0,0 +1,35 @@
+IF (NOT DEFINED LLVM_VERSION_MAJOR)
+  # Standalone build: requires both LLVM and Clang CMake packages
+  project(CBackendPlugin)
+  cmake_minimum_required(VERSION 3.20.0)
+  find_package(LLVM 20.1 REQUIRED CONFIG)
+  find_package(Clang REQUIRED CONFIG)
+  list(APPEND CMAKE_MODULE_PATH "${LLVM_CMAKE_DIR}")
+  include(AddLLVM)
+  set(CMAKE_CXX_STANDARD 17)
+  include_directories(${LLVM_INCLUDE_DIRS} ${CLANG_INCLUDE_DIRS})
+  add_definitions(${LLVM_DEFINITIONS})
+else()
+  find_package(Clang REQUIRED CONFIG)
+  include_directories(${CLANG_INCLUDE_DIRS})
+endif()
+
+# Internal Clang CodeGen headers (ABIInfo.h, TargetInfo.h, etc.) are not
+# exported; derive path from the LLVM source include directory.
+get_filename_component(_llvm_src "${LLVM_MAIN_INCLUDE_DIR}" DIRECTORY)
+include_directories(${_llvm_src}/../clang/lib/CodeGen)
+
+add_llvm_library(CBackendPlugin MODULE
+  CBackend.cpp
+
+  LINK_LIBS
+  clangAST
+  clangBasic
+  clangCodeGen
+  LLVMSupport
+)
+
+# Plugins must not have undefined symbols
+set_target_properties(CBackendPlugin PROPERTIES
+  CXX_VISIBILITY_PRESET hidden
+)