clang/lib/CIR/CodeGen/CIRGenModule.cpp

//===- CIRGenModule.cpp - Per-Module state for CIR generation -------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This is the internal per-translation-unit state used for CIR translation.
//
//===----------------------------------------------------------------------===//

#include "CIRGenModule.h"
#include "CIRGenConstantEmitter.h"
#include "CIRGenFunction.h"

#include "clang/AST/ASTContext.h"
#include "clang/AST/DeclBase.h"
#include "clang/AST/DeclOpenACC.h"
#include "clang/AST/GlobalDecl.h"
#include "clang/Basic/SourceManager.h"
#include "clang/CIR/Dialect/IR/CIRDialect.h"
#include "clang/CIR/MissingFeatures.h"

#include "mlir/IR/BuiltinOps.h"
#include "mlir/IR/Location.h"
#include "mlir/IR/MLIRContext.h"
#include "mlir/IR/Verifier.h"

using namespace clang;
using namespace clang::CIRGen;

CIRGenModule::CIRGenModule(mlir::MLIRContext &mlirContext,
                           clang::ASTContext &astContext,
                           const clang::CodeGenOptions &cgo,
                           DiagnosticsEngine &diags)
    : builder(mlirContext, *this), astContext(astContext),
      langOpts(astContext.getLangOpts()), codeGenOpts(cgo),
      theModule{mlir::ModuleOp::create(mlir::UnknownLoc::get(&mlirContext))},
      diags(diags), target(astContext.getTargetInfo()), genTypes(*this) {

  // Initialize cached types
  VoidTy = cir::VoidType::get(&getMLIRContext());
  SInt8Ty = cir::IntType::get(&getMLIRContext(), 8, /*isSigned=*/true);
  SInt16Ty = cir::IntType::get(&getMLIRContext(), 16, /*isSigned=*/true);
  SInt32Ty = cir::IntType::get(&getMLIRContext(), 32, /*isSigned=*/true);
  SInt64Ty = cir::IntType::get(&getMLIRContext(), 64, /*isSigned=*/true);
  SInt128Ty = cir::IntType::get(&getMLIRContext(), 128, /*isSigned=*/true);
  UInt8Ty = cir::IntType::get(&getMLIRContext(), 8, /*isSigned=*/false);
  UInt16Ty = cir::IntType::get(&getMLIRContext(), 16, /*isSigned=*/false);
  UInt32Ty = cir::IntType::get(&getMLIRContext(), 32, /*isSigned=*/false);
  UInt64Ty = cir::IntType::get(&getMLIRContext(), 64, /*isSigned=*/false);
  UInt128Ty = cir::IntType::get(&getMLIRContext(), 128, /*isSigned=*/false);
  FP16Ty = cir::FP16Type::get(&getMLIRContext());
  BFloat16Ty = cir::BF16Type::get(&getMLIRContext());
  FloatTy = cir::SingleType::get(&getMLIRContext());
  DoubleTy = cir::DoubleType::get(&getMLIRContext());
  FP80Ty = cir::FP80Type::get(&getMLIRContext());
  FP128Ty = cir::FP128Type::get(&getMLIRContext());

  PointerAlignInBytes =
      astContext
          .toCharUnitsFromBits(
              astContext.getTargetInfo().getPointerAlign(LangAS::Default))
          .getQuantity();

  // TODO(CIR): Should be updated once TypeSizeInfoAttr is upstreamed
  const unsigned sizeTypeSize =
      astContext.getTypeSize(astContext.getSignedSizeType());
  PtrDiffTy =
      cir::IntType::get(&getMLIRContext(), sizeTypeSize, /*isSigned=*/true);

  theModule->setAttr(cir::CIRDialect::getTripleAttrName(),
                     builder.getStringAttr(getTriple().str()));
}

CharUnits CIRGenModule::getNaturalTypeAlignment(QualType t,
                                                LValueBaseInfo *baseInfo) {
  assert(!cir::MissingFeatures::opTBAA());

  // FIXME: This duplicates logic in ASTContext::getTypeAlignIfKnown, but
  // that doesn't return the information we need to compute baseInfo.

  // Honor alignment typedef attributes even on incomplete types.
  // We also honor them straight for C++ class types, even as pointees;
  // there's an expressivity gap here.
  if (const auto *tt = t->getAs<TypedefType>()) {
    if (unsigned align = tt->getDecl()->getMaxAlignment()) {
      if (baseInfo)
        *baseInfo = LValueBaseInfo(AlignmentSource::AttributedType);
      return astContext.toCharUnitsFromBits(align);
    }
  }

  // Analyze the base element type, so we don't get confused by incomplete
  // array types.
  t = astContext.getBaseElementType(t);

  if (t->isIncompleteType()) {
    // We could try to replicate the logic from
    // ASTContext::getTypeAlignIfKnown, but nothing uses the alignment if the
    // type is incomplete, so it's impossible to test. We could try to reuse
    // getTypeAlignIfKnown, but that doesn't return the information we need
    // to set baseInfo.  So just ignore the possibility that the alignment is
    // greater than one.
    if (baseInfo)
      *baseInfo = LValueBaseInfo(AlignmentSource::Type);
    return CharUnits::One();
  }

  if (baseInfo)
    *baseInfo = LValueBaseInfo(AlignmentSource::Type);

  CharUnits alignment;
  if (t.getQualifiers().hasUnaligned()) {
    alignment = CharUnits::One();
  } else {
    assert(!cir::MissingFeatures::alignCXXRecordDecl());
    alignment = astContext.getTypeAlignInChars(t);
  }

  // Cap to the global maximum type alignment unless the alignment
  // was somehow explicit on the type.
  if (unsigned maxAlign = astContext.getLangOpts().MaxTypeAlign) {
    if (alignment.getQuantity() > maxAlign &&
        !astContext.isAlignmentRequired(t))
      alignment = CharUnits::fromQuantity(maxAlign);
  }
  return alignment;
}

mlir::Location CIRGenModule::getLoc(SourceLocation cLoc) {
  assert(cLoc.isValid() && "expected valid source location");
  const SourceManager &sm = astContext.getSourceManager();
  PresumedLoc pLoc = sm.getPresumedLoc(cLoc);
  StringRef filename = pLoc.getFilename();
  return mlir::FileLineColLoc::get(builder.getStringAttr(filename),
                                   pLoc.getLine(), pLoc.getColumn());
}

mlir::Location CIRGenModule::getLoc(SourceRange cRange) {
  assert(cRange.isValid() && "expected a valid source range");
  mlir::Location begin = getLoc(cRange.getBegin());
  mlir::Location end = getLoc(cRange.getEnd());
  mlir::Attribute metadata;
  return mlir::FusedLoc::get({begin, end}, metadata, builder.getContext());
}

void CIRGenModule::emitGlobal(clang::GlobalDecl gd) {
  if (const auto *cd = dyn_cast<clang::OpenACCConstructDecl>(gd.getDecl())) {
    emitGlobalOpenACCDecl(cd);
    return;
  }

  const auto *global = cast<ValueDecl>(gd.getDecl());

  if (const auto *fd = dyn_cast<FunctionDecl>(global)) {
    // Update deferred annotations with the latest declaration if the function
    // was already used or defined.
    if (fd->hasAttr<AnnotateAttr>())
      errorNYI(fd->getSourceRange(), "deferredAnnotations");
    if (!fd->doesThisDeclarationHaveABody()) {
      if (!fd->doesDeclarationForceExternallyVisibleDefinition())
        return;

      errorNYI(fd->getSourceRange(),
               "function declaration that forces code gen");
      return;
    }
  } else {
    assert(cast<VarDecl>(global)->isFileVarDecl() &&
           "Cannot emit local var decl as global");
  }

  // TODO(CIR): Defer emitting some global definitions until later
  emitGlobalDefinition(gd);
}

void CIRGenModule::emitGlobalFunctionDefinition(clang::GlobalDecl gd,
                                                mlir::Operation *op) {
  auto const *funcDecl = cast<FunctionDecl>(gd.getDecl());
  if (funcDecl->getIdentifier() == nullptr) {
    errorNYI(funcDecl->getSourceRange().getBegin(),
             "function definition with a non-identifier for a name");
    return;
  }
  cir::FuncType funcType =
      cast<cir::FuncType>(convertType(funcDecl->getType()));

  cir::FuncOp funcOp = dyn_cast_if_present<cir::FuncOp>(op);
  if (!funcOp || funcOp.getFunctionType() != funcType) {
    funcOp = getAddrOfFunction(gd, funcType, /*ForVTable=*/false,
                               /*DontDefer=*/true, ForDefinition);
  }

  CIRGenFunction cgf(*this, builder);
  curCGF = &cgf;
  {
    mlir::OpBuilder::InsertionGuard guard(builder);
    cgf.generateCode(gd, funcOp, funcType);
  }
  curCGF = nullptr;
}

void CIRGenModule::emitGlobalVarDefinition(const clang::VarDecl *vd,
                                           bool isTentative) {
  const QualType astTy = vd->getType();
  const mlir::Type type = convertType(vd->getType());
  if (clang::IdentifierInfo *identifier = vd->getIdentifier()) {
    auto varOp = builder.create<cir::GlobalOp>(getLoc(vd->getSourceRange()),
                                               identifier->getName(), type);
    // TODO(CIR): This code for processing initial values is a placeholder
    // until class ConstantEmitter is upstreamed and the code for processing
    // constant expressions is filled out.  Only the most basic handling of
    // certain constant expressions is implemented for now.
    const VarDecl *initDecl;
    const Expr *initExpr = vd->getAnyInitializer(initDecl);
    mlir::Attribute initializer;
    if (initExpr) {
      if (APValue *value = initDecl->evaluateValue()) {
        ConstantEmitter emitter(*this);
        initializer = emitter.tryEmitPrivateForMemory(*value, astTy);
      } else {
        errorNYI(initExpr->getSourceRange(), "non-constant initializer");
      }
    } else {
      initializer = builder.getZeroInitAttr(convertType(astTy));
    }

    varOp.setInitialValueAttr(initializer);

    // Set CIR's linkage type as appropriate.
    cir::GlobalLinkageKind linkage =
        getCIRLinkageVarDefinition(vd, /*IsConstant=*/false);

    // Set CIR linkage and DLL storage class.
    varOp.setLinkage(linkage);

    if (linkage == cir::GlobalLinkageKind::CommonLinkage)
      errorNYI(initExpr->getSourceRange(), "common linkage");

    theModule.push_back(varOp);
  } else {
    errorNYI(vd->getSourceRange().getBegin(),
             "variable definition with a non-identifier for a name");
  }
}

void CIRGenModule::emitGlobalDefinition(clang::GlobalDecl gd,
                                        mlir::Operation *op) {
  const auto *decl = cast<ValueDecl>(gd.getDecl());
  if (const auto *fd = dyn_cast<FunctionDecl>(decl)) {
    // TODO(CIR): Skip generation of CIR for functions with available_externally
    // linkage at -O0.

    if (const auto *method = dyn_cast<CXXMethodDecl>(decl)) {
      // Make sure to emit the definition(s) before we emit the thunks. This is
      // necessary for the generation of certain thunks.
      (void)method;
      errorNYI(method->getSourceRange(), "member function");
      return;
    }

    if (fd->isMultiVersion())
      errorNYI(fd->getSourceRange(), "multiversion functions");
    emitGlobalFunctionDefinition(gd, op);
    return;
  }

  if (const auto *vd = dyn_cast<VarDecl>(decl))
    return emitGlobalVarDefinition(vd, !vd->hasDefinition());

  llvm_unreachable("Invalid argument to CIRGenModule::emitGlobalDefinition");
}

static bool shouldBeInCOMDAT(CIRGenModule &cgm, const Decl &d) {
  assert(!cir::MissingFeatures::supportComdat());

  if (d.hasAttr<SelectAnyAttr>())
    return true;

  GVALinkage linkage;
  if (auto *vd = dyn_cast<VarDecl>(&d))
    linkage = cgm.getASTContext().GetGVALinkageForVariable(vd);
  else
    linkage =
        cgm.getASTContext().GetGVALinkageForFunction(cast<FunctionDecl>(&d));

  switch (linkage) {
  case clang::GVA_Internal:
  case clang::GVA_AvailableExternally:
  case clang::GVA_StrongExternal:
    return false;
  case clang::GVA_DiscardableODR:
  case clang::GVA_StrongODR:
    return true;
  }
  llvm_unreachable("No such linkage");
}

// TODO(CIR): this could be a common method between LLVM codegen.
static bool isVarDeclStrongDefinition(const ASTContext &astContext,
                                      CIRGenModule &cgm, const VarDecl *vd,
                                      bool noCommon) {
  // Don't give variables common linkage if -fno-common was specified unless it
  // was overridden by a NoCommon attribute.
  if ((noCommon || vd->hasAttr<NoCommonAttr>()) && !vd->hasAttr<CommonAttr>())
    return true;

  // C11 6.9.2/2:
  //   A declaration of an identifier for an object that has file scope without
  //   an initializer, and without a storage-class specifier or with the
  //   storage-class specifier static, constitutes a tentative definition.
  if (vd->getInit() || vd->hasExternalStorage())
    return true;

  // A variable cannot be both common and exist in a section.
  if (vd->hasAttr<SectionAttr>())
    return true;

  // A variable cannot be both common and exist in a section.
  // We don't try to determine which is the right section in the front-end.
  // If no specialized section name is applicable, it will resort to default.
  if (vd->hasAttr<PragmaClangBSSSectionAttr>() ||
      vd->hasAttr<PragmaClangDataSectionAttr>() ||
      vd->hasAttr<PragmaClangRelroSectionAttr>() ||
      vd->hasAttr<PragmaClangRodataSectionAttr>())
    return true;

  // Thread local vars aren't considered common linkage.
  if (vd->getTLSKind())
    return true;

  // Tentative definitions marked with WeakImportAttr are true definitions.
  if (vd->hasAttr<WeakImportAttr>())
    return true;

  // A variable cannot be both common and exist in a comdat.
  if (shouldBeInCOMDAT(cgm, *vd))
    return true;

  // Declarations with a required alignment do not have common linkage in MSVC
  // mode.
  if (astContext.getTargetInfo().getCXXABI().isMicrosoft()) {
    if (vd->hasAttr<AlignedAttr>())
      return true;
    QualType varType = vd->getType();
    if (astContext.isAlignmentRequired(varType))
      return true;

    if (const auto *rt = varType->getAs<RecordType>()) {
      const RecordDecl *rd = rt->getDecl();
      for (const FieldDecl *fd : rd->fields()) {
        if (fd->isBitField())
          continue;
        if (fd->hasAttr<AlignedAttr>())
          return true;
        if (astContext.isAlignmentRequired(fd->getType()))
          return true;
      }
    }
  }

  // Microsoft's link.exe doesn't support alignments greater than 32 bytes for
  // common symbols, so symbols with greater alignment requirements cannot be
  // common.
  // Other COFF linkers (ld.bfd and LLD) support arbitrary power-of-two
  // alignments for common symbols via the aligncomm directive, so this
  // restriction only applies to MSVC environments.
  if (astContext.getTargetInfo().getTriple().isKnownWindowsMSVCEnvironment() &&
      astContext.getTypeAlignIfKnown(vd->getType()) >
          astContext.toBits(CharUnits::fromQuantity(32)))
    return true;

  return false;
}

cir::GlobalLinkageKind CIRGenModule::getCIRLinkageForDeclarator(
    const DeclaratorDecl *dd, GVALinkage linkage, bool isConstantVariable) {
  if (linkage == GVA_Internal)
    return cir::GlobalLinkageKind::InternalLinkage;

  if (dd->hasAttr<WeakAttr>()) {
    if (isConstantVariable)
      return cir::GlobalLinkageKind::WeakODRLinkage;
    return cir::GlobalLinkageKind::WeakAnyLinkage;
  }

  if (const auto *fd = dd->getAsFunction())
    if (fd->isMultiVersion() && linkage == GVA_AvailableExternally)
      return cir::GlobalLinkageKind::LinkOnceAnyLinkage;

  // We are guaranteed to have a strong definition somewhere else,
  // so we can use available_externally linkage.
  if (linkage == GVA_AvailableExternally)
    return cir::GlobalLinkageKind::AvailableExternallyLinkage;

  // Note that Apple's kernel linker doesn't support symbol
  // coalescing, so we need to avoid linkonce and weak linkages there.
  // Normally, this means we just map to internal, but for explicit
  // instantiations we'll map to external.

  // In C++, the compiler has to emit a definition in every translation unit
  // that references the function.  We should use linkonce_odr because
  // a) if all references in this translation unit are optimized away, we
  // don't need to codegen it.  b) if the function persists, it needs to be
  // merged with other definitions. c) C++ has the ODR, so we know the
  // definition is dependable.
  if (linkage == GVA_DiscardableODR)
    return !astContext.getLangOpts().AppleKext
               ? cir::GlobalLinkageKind::LinkOnceODRLinkage
               : cir::GlobalLinkageKind::InternalLinkage;

  // An explicit instantiation of a template has weak linkage, since
  // explicit instantiations can occur in multiple translation units
  // and must all be equivalent. However, we are not allowed to
  // throw away these explicit instantiations.
  //
  // CUDA/HIP: For -fno-gpu-rdc case, device code is limited to one TU,
  // so say that CUDA templates are either external (for kernels) or internal.
  // This lets llvm perform aggressive inter-procedural optimizations. For
  // -fgpu-rdc case, device function calls across multiple TU's are allowed,
  // therefore we need to follow the normal linkage paradigm.
  if (linkage == GVA_StrongODR) {
    if (getLangOpts().AppleKext)
      return cir::GlobalLinkageKind::ExternalLinkage;
    if (getLangOpts().CUDA && getLangOpts().CUDAIsDevice &&
        !getLangOpts().GPURelocatableDeviceCode)
      return dd->hasAttr<CUDAGlobalAttr>()
                 ? cir::GlobalLinkageKind::ExternalLinkage
                 : cir::GlobalLinkageKind::InternalLinkage;
    return cir::GlobalLinkageKind::WeakODRLinkage;
  }

  // C++ doesn't have tentative definitions and thus cannot have common
  // linkage.
  if (!getLangOpts().CPlusPlus && isa<VarDecl>(dd) &&
      !isVarDeclStrongDefinition(astContext, *this, cast<VarDecl>(dd),
                                 getCodeGenOpts().NoCommon)) {
    errorNYI(dd->getBeginLoc(), "common linkage", dd->getDeclKindName());
    return cir::GlobalLinkageKind::CommonLinkage;
  }

  // selectany symbols are externally visible, so use weak instead of
  // linkonce.  MSVC optimizes away references to const selectany globals, so
  // all definitions should be the same and ODR linkage should be used.
  // http://msdn.microsoft.com/en-us/library/5tkz6s71.aspx
  if (dd->hasAttr<SelectAnyAttr>())
    return cir::GlobalLinkageKind::WeakODRLinkage;

  // Otherwise, we have strong external linkage.
  assert(linkage == GVA_StrongExternal);
  return cir::GlobalLinkageKind::ExternalLinkage;
}

cir::GlobalLinkageKind
CIRGenModule::getCIRLinkageVarDefinition(const VarDecl *vd, bool isConstant) {
  assert(!isConstant && "constant variables NYI");
  GVALinkage linkage = astContext.GetGVALinkageForVariable(vd);
  return getCIRLinkageForDeclarator(vd, linkage, isConstant);
}

// Emit code for a single top level declaration.
void CIRGenModule::emitTopLevelDecl(Decl *decl) {

  // Ignore dependent declarations.
  if (decl->isTemplated())
    return;

  switch (decl->getKind()) {
  default:
    errorNYI(decl->getBeginLoc(), "declaration of kind",
             decl->getDeclKindName());
    break;

  case Decl::Function: {
    auto *fd = cast<FunctionDecl>(decl);
    // Consteval functions shouldn't be emitted.
    if (!fd->isConsteval())
      emitGlobal(fd);
    break;
  }

  case Decl::Var: {
    auto *vd = cast<VarDecl>(decl);
    emitGlobal(vd);
    break;
  }
  case Decl::OpenACCRoutine:
    emitGlobalOpenACCDecl(cast<OpenACCRoutineDecl>(decl));
    break;
  case Decl::OpenACCDeclare:
    emitGlobalOpenACCDecl(cast<OpenACCDeclareDecl>(decl));
    break;
  }
}

cir::FuncOp CIRGenModule::getAddrOfFunction(clang::GlobalDecl gd,
                                            mlir::Type funcType, bool forVTable,
                                            bool dontDefer,
                                            ForDefinition_t isForDefinition) {
  assert(!cast<FunctionDecl>(gd.getDecl())->isConsteval() &&
         "consteval function should never be emitted");

  if (!funcType) {
    const auto *fd = cast<FunctionDecl>(gd.getDecl());
    funcType = convertType(fd->getType());
  }

  cir::FuncOp func = getOrCreateCIRFunction(
      cast<NamedDecl>(gd.getDecl())->getIdentifier()->getName(), funcType, gd,
      forVTable, dontDefer, /*isThunk=*/false, isForDefinition);
  return func;
}

cir::FuncOp CIRGenModule::getOrCreateCIRFunction(
    StringRef mangledName, mlir::Type funcType, GlobalDecl gd, bool forVTable,
    bool dontDefer, bool isThunk, ForDefinition_t isForDefinition,
    mlir::ArrayAttr extraAttrs) {
  auto *funcDecl = llvm::cast_or_null<FunctionDecl>(gd.getDecl());
  bool invalidLoc = !funcDecl ||
                    funcDecl->getSourceRange().getBegin().isInvalid() ||
                    funcDecl->getSourceRange().getEnd().isInvalid();
  cir::FuncOp funcOp = createCIRFunction(
      invalidLoc ? theModule->getLoc() : getLoc(funcDecl->getSourceRange()),
      mangledName, mlir::cast<cir::FuncType>(funcType), funcDecl);
  return funcOp;
}

cir::FuncOp
CIRGenModule::createCIRFunction(mlir::Location loc, StringRef name,
                                cir::FuncType funcType,
                                const clang::FunctionDecl *funcDecl) {
  cir::FuncOp func;
  {
    mlir::OpBuilder::InsertionGuard guard(builder);

    // Some global emissions are triggered while emitting a function, e.g.
    // void s() { x.method() }
    //
    // Be sure to insert a new function before a current one.
    CIRGenFunction *cgf = this->curCGF;
    if (cgf)
      builder.setInsertionPoint(cgf->curFn);

    func = builder.create<cir::FuncOp>(loc, name, funcType);

    if (!cgf)
      theModule.push_back(func);
  }
  return func;
}

mlir::Type CIRGenModule::convertType(QualType type) {
  return genTypes.convertType(type);
}

bool CIRGenModule::verifyModule() const {
  // Verify the module after we have finished constructing it, this will
  // check the structural properties of the IR and invoke any specific
  // verifiers we have on the CIR operations.
  return mlir::verify(theModule).succeeded();
}

DiagnosticBuilder CIRGenModule::errorNYI(SourceLocation loc,
                                         llvm::StringRef feature) {
  unsigned diagID = diags.getCustomDiagID(
      DiagnosticsEngine::Error, "ClangIR code gen Not Yet Implemented: %0");
  return diags.Report(loc, diagID) << feature;
}

DiagnosticBuilder CIRGenModule::errorNYI(SourceRange loc,
                                         llvm::StringRef feature) {
  return errorNYI(loc.getBegin(), feature) << loc;
}