Erich Keane 0a3fdd30e5
[CIR] Handle vtable-lowering-with-incomplete types (#190216)
The NYI diagnostic in getFunctionTypeForVTable showed up a few times in
testing, so this patch is attempting to fix that up.

The reproducer here is a function type for a vtable that has an
incomplete type in it(return or parameter). Classic codegen chooses to
represent this as an opaque type.

This patch instead removes the special v-table handling here, so that we
can instead just represent the types as incomplete record types.

At the moment, this patch ends up lowering incomplete types as 'empty'
types in LLVM-IR, which we may find we need to modify in the future,
however at the moment, it seems to work.

This patch ALSO changes the definition of RecordType::isSized to only be
true for complete types, which prevents a number of other things from
attempting to add attributes/check the size of the type/etc, but those
are irrelevant for the purposes of vtable emission.
2026-04-03 05:59:46 -07:00

1430 lines
59 KiB
C++

//===--- CIRGenCall.cpp - Encapsulate calling convention details ----------===//
//
// 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
//
//===----------------------------------------------------------------------===//
//
// These classes wrap the information about a call or function definition used
// to handle ABI compliancy.
//
//===----------------------------------------------------------------------===//
#include "CIRGenCall.h"
#include "CIRGenCXXABI.h"
#include "CIRGenFunction.h"
#include "CIRGenFunctionInfo.h"
#include "mlir/Dialect/LLVMIR/LLVMDialect.h"
#include "clang/CIR/ABIArgInfo.h"
#include "clang/CIR/MissingFeatures.h"
#include "llvm/Support/TypeSize.h"
using namespace clang;
using namespace clang::CIRGen;
CIRGenFunctionInfo *CIRGenFunctionInfo::create(
FunctionType::ExtInfo info, bool isInstanceMethod, CanQualType resultType,
llvm::ArrayRef<CanQualType> argTypes, RequiredArgs required) {
// The first slot allocated for arg type slot is for the return value.
void *buffer = operator new(
totalSizeToAlloc<CanQualType>(argTypes.size() + 1));
assert(!cir::MissingFeatures::opCallCIRGenFuncInfoParamInfo());
CIRGenFunctionInfo *fi = new (buffer) CIRGenFunctionInfo();
fi->noReturn = info.getNoReturn();
fi->instanceMethod = isInstanceMethod;
fi->required = required;
fi->numArgs = argTypes.size();
fi->getArgTypes()[0] = resultType;
std::copy(argTypes.begin(), argTypes.end(), fi->argTypesBegin());
assert(!cir::MissingFeatures::opCallCIRGenFuncInfoExtParamInfo());
return fi;
}
cir::FuncType CIRGenTypes::getFunctionType(GlobalDecl gd) {
const CIRGenFunctionInfo &fi = arrangeGlobalDeclaration(gd);
return getFunctionType(fi);
}
cir::FuncType CIRGenTypes::getFunctionType(const CIRGenFunctionInfo &info) {
mlir::Type resultType = convertType(info.getReturnType());
SmallVector<mlir::Type, 8> argTypes;
argTypes.reserve(info.getNumRequiredArgs());
for (const CanQualType &argType : info.requiredArguments())
argTypes.push_back(convertType(argType));
return cir::FuncType::get(argTypes,
(resultType ? resultType : builder.getVoidTy()),
info.isVariadic());
}
CIRGenCallee CIRGenCallee::prepareConcreteCallee(CIRGenFunction &cgf) const {
if (isVirtual()) {
const CallExpr *ce = getVirtualCallExpr();
return cgf.cgm.getCXXABI().getVirtualFunctionPointer(
cgf, getVirtualMethodDecl(), getThisAddress(), getVirtualFunctionType(),
ce ? ce->getBeginLoc() : SourceLocation());
}
return *this;
}
void CIRGenFunction::emitAggregateStore(mlir::Value value, Address dest) {
// In classic codegen:
// Function to store a first-class aggregate into memory. We prefer to
// store the elements rather than the aggregate to be more friendly to
// fast-isel.
// In CIR codegen:
// Emit the most simple cir.store possible (e.g. a store for a whole
// record), which can later be broken down in other CIR levels (or prior
// to dialect codegen).
// Stored result for the callers of this function expected to be in the same
// scope as the value, don't make assumptions about current insertion point.
mlir::OpBuilder::InsertionGuard guard(builder);
builder.setInsertionPointAfter(value.getDefiningOp());
builder.createStore(*currSrcLoc, value, dest);
}
static void addAttributesFromFunctionProtoType(CIRGenBuilderTy &builder,
mlir::NamedAttrList &attrs,
const FunctionProtoType *fpt) {
if (!fpt)
return;
if (!isUnresolvedExceptionSpec(fpt->getExceptionSpecType()) &&
fpt->isNothrow())
attrs.set(cir::CIRDialect::getNoThrowAttrName(),
mlir::UnitAttr::get(builder.getContext()));
}
static void addNoBuiltinAttributes(mlir::MLIRContext &ctx,
mlir::NamedAttrList &attrs,
const LangOptions &langOpts,
const NoBuiltinAttr *nba = nullptr) {
// First, handle the language options passed through -fno-builtin.
// or, if there is a wildcard in the builtin names specified through the
// attribute, disable them all.
if (langOpts.NoBuiltin ||
(nba && llvm::is_contained(nba->builtinNames(), "*"))) {
// -fno-builtin disables them all.
// Empty attribute means 'all'.
attrs.set(cir::CIRDialect::getNoBuiltinsAttrName(),
mlir::ArrayAttr::get(&ctx, {}));
return;
}
llvm::SetVector<mlir::Attribute> nbFuncs;
auto addNoBuiltinAttr = [&ctx, &nbFuncs](StringRef builtinName) {
nbFuncs.insert(mlir::StringAttr::get(&ctx, builtinName));
};
// Then, add attributes for builtins specified through -fno-builtin-<name>.
llvm::for_each(langOpts.NoBuiltinFuncs, addNoBuiltinAttr);
// Now, let's check the __attribute__((no_builtin("...")) attribute added to
// the source.
if (nba)
llvm::for_each(nba->builtinNames(), addNoBuiltinAttr);
if (!nbFuncs.empty())
attrs.set(cir::CIRDialect::getNoBuiltinsAttrName(),
mlir::ArrayAttr::get(&ctx, nbFuncs.getArrayRef()));
}
/// Add denormal-fp-math and denormal-fp-math-f32 as appropriate for the
/// requested denormal behavior, accounting for the overriding behavior of the
/// -f32 case.
static void addDenormalModeAttrs(llvm::DenormalMode fpDenormalMode,
llvm::DenormalMode fp32DenormalMode,
mlir::NamedAttrList &attrs) {
// TODO(cir): Classic-codegen sets the denormal modes here. There are two
// values, both with a string, but it seems that perhaps we could combine
// these into a single attribute? It seems a little silly to have two so
// similar named attributes that do the same thing.
}
/// Add default attributes to a function, which have merge semantics under
/// -mlink-builtin-bitcode and should not simply overwrite any existing
/// attributes in the linked library.
static void
addMergeableDefaultFunctionAttributes(const CodeGenOptions &codeGenOpts,
mlir::NamedAttrList &attrs) {
addDenormalModeAttrs(codeGenOpts.FPDenormalMode, codeGenOpts.FP32DenormalMode,
attrs);
}
static llvm::StringLiteral
getZeroCallUsedRegsKindStr(llvm::ZeroCallUsedRegs::ZeroCallUsedRegsKind k) {
switch (k) {
case llvm::ZeroCallUsedRegs::ZeroCallUsedRegsKind::Skip:
llvm_unreachable("No string value, shouldn't be able to get here");
case llvm::ZeroCallUsedRegs::ZeroCallUsedRegsKind::UsedGPRArg:
return "used-gpr-arg";
case llvm::ZeroCallUsedRegs::ZeroCallUsedRegsKind::UsedGPR:
return "used-gpr";
case llvm::ZeroCallUsedRegs::ZeroCallUsedRegsKind::UsedArg:
return "used-arg";
case llvm::ZeroCallUsedRegs::ZeroCallUsedRegsKind::Used:
return "used";
case llvm::ZeroCallUsedRegs::ZeroCallUsedRegsKind::AllGPRArg:
return "all-gpr-arg";
case llvm::ZeroCallUsedRegs::ZeroCallUsedRegsKind::AllGPR:
return "all-gpr";
case llvm::ZeroCallUsedRegs::ZeroCallUsedRegsKind::AllArg:
return "all-arg";
case llvm::ZeroCallUsedRegs::ZeroCallUsedRegsKind::All:
return "all";
}
llvm_unreachable("Unknown kind?");
}
/// Add default attributes to a function, which have merge semantics under
/// -mlink-builtin-bitcode and should not simply overwrite any existing
/// attributes in the linked library.
static void addTrivialDefaultFunctionAttributes(
mlir::MLIRContext *mlirCtx, StringRef name, bool hasOptNoneAttr,
const CodeGenOptions &codeGenOpts, const LangOptions &langOpts,
bool attrOnCallSite, mlir::NamedAttrList &attrs) {
// TODO(cir): Handle optimize attribute flag here.
// OptimizeNoneAttr takes precedence over -Os or -Oz. No warning needed.
if (!hasOptNoneAttr) {
if (codeGenOpts.OptimizeSize)
attrs.set(cir::CIRDialect::getOptimizeForSizeAttrName(),
mlir::UnitAttr::get(mlirCtx));
if (codeGenOpts.OptimizeSize == 2)
attrs.set(cir::CIRDialect::getMinSizeAttrName(),
mlir::UnitAttr::get(mlirCtx));
}
// TODO(cir): Classic codegen adds 'DisableRedZone', 'indirect-tls-seg-refs'
// and 'NoImplicitFloat' here.
if (attrOnCallSite) {
// Add the 'nobuiltin' tag, which is different from 'no-builtins'.
if (!codeGenOpts.SimplifyLibCalls || langOpts.isNoBuiltinFunc(name))
attrs.set(cir::CIRDialect::getNoBuiltinAttrName(),
mlir::UnitAttr::get(mlirCtx));
if (!codeGenOpts.TrapFuncName.empty())
attrs.set(cir::CIRDialect::getTrapFuncNameAttrName(),
mlir::StringAttr::get(mlirCtx, codeGenOpts.TrapFuncName));
} else {
// TODO(cir): Set frame pointer attribute here.
// TODO(cir): a number of other attribute 1-offs based on codegen/lang opts
// should be done here: less-recise-fpmad null-pointer-is-valid
// no-trapping-math
// various inf/nan/nsz/etc work here.
//
// TODO(cir): set stack-protector buffer size attribute (sorted oddly in
// classic compiler inside of the above region, but should be done on its
// own).
// TODO(cir): other attributes here:
// reciprocal estimates, prefer-vector-width, stackrealign, backchain,
// split-stack, speculative-load-hardening.
if (codeGenOpts.getZeroCallUsedRegs() ==
llvm::ZeroCallUsedRegs::ZeroCallUsedRegsKind::Skip)
attrs.erase(cir::CIRDialect::getZeroCallUsedRegsAttrName());
else
attrs.set(cir::CIRDialect::getZeroCallUsedRegsAttrName(),
mlir::StringAttr::get(mlirCtx,
getZeroCallUsedRegsKindStr(
codeGenOpts.getZeroCallUsedRegs())));
}
if (langOpts.assumeFunctionsAreConvergent()) {
// Conservatively, mark all functions and calls in CUDA and OpenCL as
// convergent (meaning, they may call an intrinsically convergent op, such
// as __syncthreads() / barrier(), and so can't have certain optimizations
// applied around them). LLVM will remove this attribute where it safely
// can.
attrs.set(cir::CIRDialect::getConvergentAttrName(),
mlir::UnitAttr::get(mlirCtx));
}
// TODO(cir): Classic codegen adds 'nounwind' here in a bunch of offload
// targets.
if (codeGenOpts.SaveRegParams && !attrOnCallSite)
attrs.set(cir::CIRDialect::getSaveRegParamsAttrName(),
mlir::UnitAttr::get(mlirCtx));
// These come in the form of an optional equality sign, so make sure we pass
// these on correctly. These will eventually just be passed through to
// LLVM-IR, but we want to put them all in 1 array to simplify the
// LLVM-MLIR dialect.
SmallVector<mlir::NamedAttribute> defaultFuncAttrs;
llvm::transform(
codeGenOpts.DefaultFunctionAttrs, std::back_inserter(defaultFuncAttrs),
[mlirCtx](llvm::StringRef arg) {
auto [var, value] = arg.split('=');
auto valueAttr =
value.empty()
? cast<mlir::Attribute>(mlir::UnitAttr::get(mlirCtx))
: cast<mlir::Attribute>(mlir::StringAttr::get(mlirCtx, value));
return mlir::NamedAttribute(var, valueAttr);
});
if (!defaultFuncAttrs.empty())
attrs.set(cir::CIRDialect::getDefaultFuncAttrsAttrName(),
mlir::DictionaryAttr::get(mlirCtx, defaultFuncAttrs));
// TODO(cir): Do branch protection attributes here.
}
/// This function matches the behavior of 'getDefaultFunctionAttributes' from
/// classic codegen, despite the similarity of its name to
/// 'addDefaultFunctionDefinitionAttributes', which is a caller of this
/// function.
void CIRGenModule::addDefaultFunctionAttributes(StringRef name,
bool hasOptNoneAttr,
bool attrOnCallSite,
mlir::NamedAttrList &attrs) {
addTrivialDefaultFunctionAttributes(&getMLIRContext(), name, hasOptNoneAttr,
codeGenOpts, langOpts, attrOnCallSite,
attrs);
if (!attrOnCallSite) {
// TODO(cir): Classic codegen adds pointer-auth attributes here, by calling
// into TargetCodeGenInfo. At the moment, we've not looked into this as it
// is somewhat less used.
addMergeableDefaultFunctionAttributes(codeGenOpts, attrs);
}
}
/// Construct the CIR attribute list of a function or call.
void CIRGenModule::constructAttributeList(
llvm::StringRef name, const CIRGenFunctionInfo &info,
CIRGenCalleeInfo calleeInfo, mlir::NamedAttrList &attrs,
llvm::MutableArrayRef<mlir::NamedAttrList> argAttrs,
mlir::NamedAttrList &retAttrs, cir::CallingConv &callingConv,
cir::SideEffect &sideEffect, bool attrOnCallSite, bool isThunk) {
assert(!cir::MissingFeatures::opCallCallConv());
sideEffect = cir::SideEffect::All;
auto addUnitAttr = [&](llvm::StringRef name) {
attrs.set(name, mlir::UnitAttr::get(&getMLIRContext()));
};
if (info.isNoReturn())
addUnitAttr(cir::CIRDialect::getNoReturnAttrName());
// TODO(cir): Implement/check the CSME Nonsecure call attribute here. This
// requires being in CSME mode.
addAttributesFromFunctionProtoType(getBuilder(), attrs,
calleeInfo.getCalleeFunctionProtoType());
const Decl *targetDecl = calleeInfo.getCalleeDecl().getDecl();
// TODO(cir): OMP Assume Attributes should be here.
const NoBuiltinAttr *nba = nullptr;
// TODO(cir): Some work for arg memory effects can be done here, as it is in
// classic codegen.
if (targetDecl) {
if (targetDecl->hasAttr<NoThrowAttr>())
addUnitAttr(cir::CIRDialect::getNoThrowAttrName());
// TODO(cir): This is actually only possible if targetDecl isn't a
// declarator, which ObjCMethodDecl seems to be the only way to get this to
// happen. We're including it here for completeness, but we should add a
// test for this when we start generating ObjectiveC.
if (targetDecl->hasAttr<NoReturnAttr>())
addUnitAttr(cir::CIRDialect::getNoReturnAttrName());
if (targetDecl->hasAttr<ReturnsTwiceAttr>())
addUnitAttr(cir::CIRDialect::getReturnsTwiceAttrName());
if (targetDecl->hasAttr<ColdAttr>())
addUnitAttr(cir::CIRDialect::getColdAttrName());
if (targetDecl->hasAttr<HotAttr>())
addUnitAttr(cir::CIRDialect::getHotAttrName());
if (targetDecl->hasAttr<NoDuplicateAttr>())
addUnitAttr(cir::CIRDialect::getNoDuplicatesAttrName());
if (targetDecl->hasAttr<ConvergentAttr>())
addUnitAttr(cir::CIRDialect::getConvergentAttrName());
if (const FunctionDecl *func = dyn_cast<FunctionDecl>(targetDecl)) {
addAttributesFromFunctionProtoType(
getBuilder(), attrs, func->getType()->getAs<FunctionProtoType>());
// TODO(cir): When doing 'return attrs' we need to cover the 'NoAlias' for
// global allocation functions here.
assert(!cir::MissingFeatures::opCallAttrs());
const CXXMethodDecl *md = dyn_cast<CXXMethodDecl>(func);
bool isVirtualCall = md && md->isVirtual();
// Don't use [[noreturn]], _Noreturn or [[no_builtin]] for a call to a
// virtual function. These attributes are not inherited by overloads.
if (!(attrOnCallSite && isVirtualCall)) {
if (func->isNoReturn())
addUnitAttr(cir::CIRDialect::getNoReturnAttrName());
nba = func->getAttr<NoBuiltinAttr>();
}
}
assert(!cir::MissingFeatures::opCallAttrs());
// 'const', 'pure' and 'noalias' attributed functions are also nounwind.
if (targetDecl->hasAttr<ConstAttr>()) {
// gcc specifies that 'const' functions have greater restrictions than
// 'pure' functions, so they also cannot have infinite loops.
sideEffect = cir::SideEffect::Const;
} else if (targetDecl->hasAttr<PureAttr>()) {
// gcc specifies that 'pure' functions cannot have infinite loops.
sideEffect = cir::SideEffect::Pure;
}
attrs.set(cir::CIRDialect::getSideEffectAttrName(),
cir::SideEffectAttr::get(&getMLIRContext(), sideEffect));
// TODO(cir): When doing 'return attrs' we need to cover the Restrict and
// ReturnsNonNull attributes here.
if (targetDecl->hasAttr<AnyX86NoCallerSavedRegistersAttr>())
addUnitAttr(cir::CIRDialect::getNoCallerSavedRegsAttrName());
// TODO(cir): Implement 'NoCFCheck' attribute here. This requires
// fcf-protection mode.
if (targetDecl->hasAttr<LeafAttr>())
addUnitAttr(cir::CIRDialect::getNoCallbackAttrName());
// TODO(cir): Implement 'BPFFastCall' attribute here. This requires C, and
// the BPF target.
if (auto *allocSizeAttr = targetDecl->getAttr<AllocSizeAttr>()) {
unsigned size = allocSizeAttr->getElemSizeParam().getLLVMIndex();
if (allocSizeAttr->getNumElemsParam().isValid()) {
unsigned numElts = allocSizeAttr->getNumElemsParam().getLLVMIndex();
attrs.set(cir::CIRDialect::getAllocSizeAttrName(),
builder.getDenseI32ArrayAttr(
{static_cast<int>(size), static_cast<int>(numElts)}));
} else {
attrs.set(cir::CIRDialect::getAllocSizeAttrName(),
builder.getDenseI32ArrayAttr({static_cast<int>(size)}));
}
}
// TODO(cir): Quite a few CUDA and OpenCL attributes are added here, like
// uniform-work-group-size.
if (langOpts.CUDA && !langOpts.CUDAIsDevice &&
targetDecl->hasAttr<CUDAGlobalAttr>()) {
GlobalDecl kernel(calleeInfo.getCalleeDecl());
llvm::StringRef kernelName = getMangledName(
kernel.getWithKernelReferenceKind(KernelReferenceKind::Kernel));
auto attr =
cir::CUDAKernelNameAttr::get(&getMLIRContext(), kernelName.str());
attrs.set(attr.getMnemonic(), attr);
}
// TODO(cir): we should also do 'aarch64_pstate_sm_body' here.
if (auto *modularFormat = targetDecl->getAttr<ModularFormatAttr>()) {
FormatAttr *format = targetDecl->getAttr<FormatAttr>();
StringRef type = format->getType()->getName();
std::string formatIdx = std::to_string(format->getFormatIdx());
std::string firstArg = std::to_string(format->getFirstArg());
SmallVector<StringRef> args = {
type, formatIdx, firstArg,
modularFormat->getModularImplFn()->getName(),
modularFormat->getImplName()};
llvm::append_range(args, modularFormat->aspects());
attrs.set(cir::CIRDialect::getModularFormatAttrName(),
builder.getStringAttr(llvm::join(args, ",")));
}
}
addNoBuiltinAttributes(getMLIRContext(), attrs, getLangOpts(), nba);
bool hasOptNoneAttr = targetDecl && targetDecl->hasAttr<OptimizeNoneAttr>();
addDefaultFunctionAttributes(name, hasOptNoneAttr, attrOnCallSite, attrs);
if (targetDecl) {
// TODO(cir): There is another region of `if (targetDecl)` that handles
// removing some attributes that are necessary modifications of the
// default-function attrs. Including:
// NoSpeculativeLoadHardening
// SpeculativeLoadHardening
// NoSplitStack
// Non-lazy-bind
// 'sample-profile-suffix-elision-policy'.
if (targetDecl->hasAttr<ZeroCallUsedRegsAttr>()) {
// A function "__attribute__((...))" overrides the command-line flag.
auto kind =
targetDecl->getAttr<ZeroCallUsedRegsAttr>()->getZeroCallUsedRegs();
attrs.set(
cir::CIRDialect::getZeroCallUsedRegsAttrName(),
mlir::StringAttr::get(
&getMLIRContext(),
ZeroCallUsedRegsAttr::ConvertZeroCallUsedRegsKindToStr(kind)));
}
if (targetDecl->hasAttr<NoConvergentAttr>())
attrs.erase(cir::CIRDialect::getConvergentAttrName());
}
// TODO(cir): A bunch of non-call-site function IR attributes from
// declaration-specific information, including tail calls,
// cmse_nonsecure_entry, additional/automatic 'returns-twice' functions,
// CPU-features/overrides, and hotpatch support.
// TODO(cir): Add loader-replaceable attribute here.
constructFunctionReturnAttributes(info, targetDecl, isThunk, retAttrs);
constructFunctionArgumentAttributes(info, isThunk, argAttrs);
}
bool CIRGenModule::hasStrictReturn(QualType retTy, const Decl *targetDecl) {
// As-is msan can not tolerate noundef mismatch between caller and
// implementation. Mismatch is possible for e.g. indirect calls from C-caller
// into C++. Such mismatches lead to confusing false reports. To avoid
// expensive workaround on msan we enforce initialization event in uncommon
// cases where it's allowed.
if (getLangOpts().Sanitize.has(SanitizerKind::Memory))
return true;
// C++ explicitly makes returning undefined values UB. C's rule only applies
// to used values, so we never mark them noundef for now.
if (!getLangOpts().CPlusPlus)
return false;
if (targetDecl) {
if (const FunctionDecl *func = dyn_cast<FunctionDecl>(targetDecl)) {
if (func->isExternC())
return false;
} else if (const VarDecl *var = dyn_cast<VarDecl>(targetDecl)) {
// Function pointer.
if (var->isExternC())
return false;
}
}
// We don't want to be too aggressive with the return checking, unless
// it's explicit in the code opts or we're using an appropriate sanitizer.
// Try to respect what the programmer intended.
return getCodeGenOpts().StrictReturn ||
!mayDropFunctionReturn(getASTContext(), retTy) ||
getLangOpts().Sanitize.has(SanitizerKind::Return);
}
bool CIRGenModule::mayDropFunctionReturn(const ASTContext &context,
QualType retTy) {
// We can't just discard the return value for a record type with a
// complex destructor or a non-trivially copyable type.
if (const RecordType *recTy =
retTy.getCanonicalType()->getAsCanonical<RecordType>()) {
if (const auto *record = dyn_cast<CXXRecordDecl>(recTy->getDecl()))
return record->hasTrivialDestructor();
}
return retTy.isTriviallyCopyableType(context);
}
static bool determineNoUndef(QualType clangTy, CIRGenTypes &types,
const cir::CIRDataLayout &layout,
const cir::ABIArgInfo &argInfo) {
mlir::Type ty = types.convertTypeForMem(clangTy);
assert(!cir::MissingFeatures::abiArgInfo());
if (argInfo.isIndirect() || argInfo.isIndirectAliased())
return true;
if (argInfo.isExtend() && !argInfo.isNoExt())
return true;
if (cir::isSized(ty) && !layout.typeSizeEqualsStoreSize(ty))
// TODO: This will result in a modest amount of values not marked noundef
// when they could be. We care about values that *invisibly* contain undef
// bits from the perspective of LLVM IR.
return false;
assert(!cir::MissingFeatures::opCallCallConv());
// TODO(cir): The calling convention code needs to figure if the
// coerced-to-type is larger than the actual type, and remove the noundef
// attribute. Classic compiler did it here.
if (clangTy->isBitIntType())
return true;
if (clangTy->isReferenceType())
return true;
if (clangTy->isNullPtrType())
return false;
if (clangTy->isMemberPointerType())
// TODO: Some member pointers are `noundef`, but it depends on the ABI. For
// now, never mark them.
return false;
if (clangTy->isScalarType()) {
if (const ComplexType *Complex = dyn_cast<ComplexType>(clangTy))
return determineNoUndef(Complex->getElementType(), types, layout,
argInfo);
return true;
}
if (const VectorType *Vector = dyn_cast<VectorType>(clangTy))
return determineNoUndef(Vector->getElementType(), types, layout, argInfo);
if (const MatrixType *Matrix = dyn_cast<MatrixType>(clangTy))
return determineNoUndef(Matrix->getElementType(), types, layout, argInfo);
if (const ArrayType *Array = dyn_cast<ArrayType>(clangTy))
return determineNoUndef(Array->getElementType(), types, layout, argInfo);
// TODO: Some structs may be `noundef`, in specific situations.
return false;
}
void CIRGenModule::constructFunctionReturnAttributes(
const CIRGenFunctionInfo &info, const Decl *targetDecl, bool isThunk,
mlir::NamedAttrList &retAttrs) {
// Collect attributes from arguments and return values.
QualType retTy = info.getReturnType();
const cir::ABIArgInfo retInfo = info.getReturnInfo();
const cir::CIRDataLayout &layout = getDataLayout();
if (codeGenOpts.EnableNoundefAttrs && hasStrictReturn(retTy, targetDecl) &&
!retTy->isVoidType() &&
determineNoUndef(retTy, getTypes(), layout, retInfo))
retAttrs.set(mlir::LLVM::LLVMDialect::getNoUndefAttrName(),
mlir::UnitAttr::get(&getMLIRContext()));
// TODO(cir): classic codegen adds a bunch of attributes based on
// calling-convention lowering results. However, since calling conventions
// haven't happened yet, this work likely has to happen there.
if (!isThunk) {
// TODO(cir): following comment taken from classic codegen, so if anything
// happens there, we should reflect it here.
// FIXME: fix this properly, https://reviews.llvm.org/D100388
if (const auto *refTy = retTy->getAs<ReferenceType>()) {
QualType pointeeTy = refTy->getPointeeType();
if (!pointeeTy->isIncompleteType() && pointeeTy->isConstantSizeType())
retAttrs.set(mlir::LLVM::LLVMDialect::getDereferenceableAttrName(),
builder.getI64IntegerAttr(
getMinimumObjectSize(pointeeTy).getQuantity()));
if (getTypes().getTargetAddressSpace(pointeeTy) == 0 &&
!codeGenOpts.NullPointerIsValid)
retAttrs.set(mlir::LLVM::LLVMDialect::getNonNullAttrName(),
mlir::UnitAttr::get(&getMLIRContext()));
if (pointeeTy->isObjectType())
retAttrs.set(mlir::LLVM::LLVMDialect::getAlignAttrName(),
builder.getI64IntegerAttr(
getNaturalPointeeTypeAlignment(retTy).getQuantity()));
}
}
}
void CIRGenModule::constructFunctionArgumentAttributes(
const CIRGenFunctionInfo &info, bool isThunk,
llvm::MutableArrayRef<mlir::NamedAttrList> argAttrs) {
assert(!cir::MissingFeatures::abiArgInfo());
// TODO(cir): classic codegen does a lot of work here based on the ABIArgInfo
// to set things based on calling convention.
// At the moment, only nonnull, dereferenceable, align, and noundef are being
// implemented here, using similar logic to how we do so for return types.
if (info.isInstanceMethod() && !isThunk) {
QualType thisPtrTy = info.arguments()[0];
// Member allocation functions are instance methods, but setting attributes
// on them is nonsensical and not correct. Make sure we skip that here.
if (!thisPtrTy->isVoidPointerType()) {
QualType thisTy = thisPtrTy->getPointeeType();
if (!codeGenOpts.NullPointerIsValid &&
getTypes().getTargetAddressSpace(thisPtrTy) == 0) {
argAttrs[0].set(mlir::LLVM::LLVMDialect::getDereferenceableAttrName(),
builder.getI64IntegerAttr(
getMinimumObjectSize(thisTy).getQuantity()));
argAttrs[0].set(mlir::LLVM::LLVMDialect::getNonNullAttrName(),
mlir::UnitAttr::get(&getMLIRContext()));
} else {
uint64_t bytes = getMinimumObjectSize(thisTy).getQuantity();
if (bytes != 0)
argAttrs[0].set(
mlir::LLVM::LLVMDialect::getDereferenceableOrNullAttrName(),
builder.getI64IntegerAttr(bytes));
}
argAttrs[0].set(
mlir::LLVM::LLVMDialect::getAlignAttrName(),
builder.getI64IntegerAttr(
getNaturalPointeeTypeAlignment(thisPtrTy).getQuantity()));
// TODO(cir): the classic codegen has a recently-added bunch of logic for
// 'dead_on_return' as an attribute. This both doesn't exist in the LLVM
// dialect, and is 'too new' at the time of writing this to be considered
// stable enough here. For now, we'll leave this as a TODO so that when
// we come back, it is hopefully a more stabilized implementation.
}
}
// TODO(cir): the logic between 'this', return, and normal arguments hsould
// probably be merged at one point, however the logic is unfortunately mildly
// different between each in classic codegen, so trying to do anything like
// that seems risky at the moment. At one point we should evaluate if at least
// dereferenceable, nonnull, and align can be combined.
const cir::CIRDataLayout &layout = getDataLayout();
for (const auto &[argAttrList, argCanType] :
llvm::zip_equal(argAttrs, info.arguments())) {
assert(!cir::MissingFeatures::abiArgInfo());
QualType argType = argCanType;
const cir::ABIArgInfo argInfo = cir::ABIArgInfo::getDirect();
if (codeGenOpts.EnableNoundefAttrs &&
determineNoUndef(argType, getTypes(), layout, argInfo))
argAttrList.set(mlir::LLVM::LLVMDialect::getNoUndefAttrName(),
mlir::UnitAttr::get(&getMLIRContext()));
assert(!cir::MissingFeatures::abiArgInfo());
// TODO(cir): there is plenty of other attributes here added due to ABI
// decisions. While these probably won't end up here, we note that the
// classic codegen does it here and perhaps we should pay attention to that.
if (const auto *refTy = argType->getAs<ReferenceType>()) {
QualType pointeeTy = refTy->getPointeeType();
if (!pointeeTy->isIncompleteType() && pointeeTy->isConstantSizeType())
argAttrList.set(mlir::LLVM::LLVMDialect::getDereferenceableAttrName(),
builder.getI64IntegerAttr(
getMinimumObjectSize(pointeeTy).getQuantity()));
if (getTypes().getTargetAddressSpace(pointeeTy) == 0 &&
!codeGenOpts.NullPointerIsValid)
argAttrList.set(mlir::LLVM::LLVMDialect::getNonNullAttrName(),
mlir::UnitAttr::get(&getMLIRContext()));
if (pointeeTy->isObjectType())
argAttrList.set(
mlir::LLVM::LLVMDialect::getAlignAttrName(),
builder.getI64IntegerAttr(
getNaturalPointeeTypeAlignment(argType).getQuantity()));
}
}
}
/// Returns the canonical formal type of the given C++ method.
static CanQual<FunctionProtoType> getFormalType(const CXXMethodDecl *md) {
return md->getType()
->getCanonicalTypeUnqualified()
.getAs<FunctionProtoType>();
}
/// Adds the formal parameters in FPT to the given prefix. If any parameter in
/// FPT has pass_object_size_attrs, then we'll add parameters for those, too.
/// TODO(cir): this should be shared with LLVM codegen
static void appendParameterTypes(const CIRGenTypes &cgt,
SmallVectorImpl<CanQualType> &prefix,
CanQual<FunctionProtoType> fpt) {
assert(!cir::MissingFeatures::opCallExtParameterInfo());
// Fast path: don't touch param info if we don't need to.
if (!fpt->hasExtParameterInfos()) {
prefix.append(fpt->param_type_begin(), fpt->param_type_end());
return;
}
cgt.getCGModule().errorNYI("appendParameterTypes: hasExtParameterInfos");
}
const CIRGenFunctionInfo &
CIRGenTypes::arrangeCXXStructorDeclaration(GlobalDecl gd) {
auto *md = cast<CXXMethodDecl>(gd.getDecl());
llvm::SmallVector<CanQualType, 16> argTypes;
argTypes.push_back(deriveThisType(md->getParent(), md));
bool passParams = true;
if (auto *cd = dyn_cast<CXXConstructorDecl>(md)) {
// A base class inheriting constructor doesn't get forwarded arguments
// needed to construct a virtual base (or base class thereof)
if (cd->getInheritedConstructor())
cgm.errorNYI(cd->getSourceRange(),
"arrangeCXXStructorDeclaration: inheriting constructor");
}
CanQual<FunctionProtoType> fpt = getFormalType(md);
if (passParams)
appendParameterTypes(*this, argTypes, fpt);
// The structor signature may include implicit parameters.
[[maybe_unused]] CIRGenCXXABI::AddedStructorArgCounts addedArgs =
theCXXABI.buildStructorSignature(gd, argTypes);
assert(!cir::MissingFeatures::opCallCIRGenFuncInfoExtParamInfo());
RequiredArgs required =
(passParams && md->isVariadic() ? RequiredArgs(argTypes.size())
: RequiredArgs::All);
CanQualType resultType = theCXXABI.hasThisReturn(gd) ? argTypes.front()
: theCXXABI.hasMostDerivedReturn(gd)
? astContext.VoidPtrTy
: astContext.VoidTy;
assert(!theCXXABI.hasThisReturn(gd) &&
"Please send PR with a test and remove this");
assert(!cir::MissingFeatures::opCallCIRGenFuncInfoExtParamInfo());
assert(!cir::MissingFeatures::opCallFnInfoOpts());
return arrangeCIRFunctionInfo(resultType, /*isInstanceMethod=*/true, argTypes,
fpt->getExtInfo(), required);
}
/// Derives the 'this' type for CIRGen purposes, i.e. ignoring method CVR
/// qualification. Either or both of `rd` and `md` may be null. A null `rd`
/// indicates that there is no meaningful 'this' type, and a null `md` can occur
/// when calling a method pointer.
CanQualType CIRGenTypes::deriveThisType(const CXXRecordDecl *rd,
const CXXMethodDecl *md) {
CanQualType recTy;
if (rd) {
recTy = getASTContext().getCanonicalTagType(rd);
} else {
// This can happen with the MS ABI. It shouldn't need anything more than
// setting recTy to VoidTy here, but we're flagging it for now because we
// don't have the full handling implemented.
cgm.errorNYI("deriveThisType: no record decl");
recTy = getASTContext().VoidTy;
}
if (md)
recTy = CanQualType::CreateUnsafe(getASTContext().getAddrSpaceQualType(
recTy, md->getMethodQualifiers().getAddressSpace()));
return getASTContext().getPointerType(recTy);
}
/// Arrange the CIR function layout for a value of the given function type, on
/// top of any implicit parameters already stored.
static const CIRGenFunctionInfo &
arrangeCIRFunctionInfo(CIRGenTypes &cgt, bool instanceMethod,
SmallVectorImpl<CanQualType> &prefix,
CanQual<FunctionProtoType> fpt) {
assert(!cir::MissingFeatures::opCallFnInfoOpts());
RequiredArgs required =
RequiredArgs::getFromProtoWithExtraSlots(fpt, prefix.size());
assert(!cir::MissingFeatures::opCallExtParameterInfo());
appendParameterTypes(cgt, prefix, fpt);
CanQualType resultType = fpt->getReturnType().getUnqualifiedType();
return cgt.arrangeCIRFunctionInfo(resultType, instanceMethod, prefix,
fpt->getExtInfo(), required);
}
void CIRGenFunction::emitDelegateCallArg(CallArgList &args,
const VarDecl *param,
SourceLocation loc) {
// StartFunction converted the ABI-lowered parameter(s) into a local alloca.
// We need to turn that into an r-value suitable for emitCall
Address local = getAddrOfLocalVar(param);
QualType type = param->getType();
if (type->getAsCXXRecordDecl()) {
cgm.errorNYI(param->getSourceRange(),
"emitDelegateCallArg: record argument");
return;
}
// GetAddrOfLocalVar returns a pointer-to-pointer for references, but the
// argument needs to be the original pointer.
if (type->isReferenceType()) {
args.add(
RValue::get(builder.createLoad(getLoc(param->getSourceRange()), local)),
type);
} else if (getLangOpts().ObjCAutoRefCount) {
cgm.errorNYI(param->getSourceRange(),
"emitDelegateCallArg: ObjCAutoRefCount");
// For the most part, we just need to load the alloca, except that aggregate
// r-values are actually pointers to temporaries.
} else {
args.add(convertTempToRValue(local, type, loc), type);
}
// Deactivate the cleanup for the callee-destructed param that was pushed.
assert(!cir::MissingFeatures::thunks());
if (type->isRecordType() &&
type->castAsRecordDecl()->isParamDestroyedInCallee() &&
param->needsDestruction(getContext())) {
cgm.errorNYI(param->getSourceRange(),
"emitDelegateCallArg: callee-destructed param");
}
}
static const CIRGenFunctionInfo &
arrangeFreeFunctionLikeCall(CIRGenTypes &cgt, CIRGenModule &cgm,
const CallArgList &args,
const FunctionType *fnType) {
RequiredArgs required = RequiredArgs::All;
if (const auto *proto = dyn_cast<FunctionProtoType>(fnType)) {
if (proto->isVariadic())
required = RequiredArgs::getFromProtoWithExtraSlots(proto, 0);
if (proto->hasExtParameterInfos())
cgm.errorNYI("call to functions with extra parameter info");
} else if (cgm.getTargetCIRGenInfo().isNoProtoCallVariadic(
cast<FunctionNoProtoType>(fnType)))
cgm.errorNYI("call to function without a prototype");
SmallVector<CanQualType, 16> argTypes;
for (const CallArg &arg : args)
argTypes.push_back(cgt.getASTContext().getCanonicalParamType(arg.ty));
CanQualType retType = fnType->getReturnType()->getCanonicalTypeUnqualified();
assert(!cir::MissingFeatures::opCallFnInfoOpts());
return cgt.arrangeCIRFunctionInfo(retType, /*isInstanceMethod=*/false,
argTypes, fnType->getExtInfo(), required);
}
/// Arrange a call to a C++ method, passing the given arguments.
///
/// extraPrefixArgs is the number of ABI-specific args passed after the `this`
/// parameter.
/// passProtoArgs indicates whether `args` has args for the parameters in the
/// given CXXConstructorDecl.
const CIRGenFunctionInfo &CIRGenTypes::arrangeCXXConstructorCall(
const CallArgList &args, const CXXConstructorDecl *d, CXXCtorType ctorKind,
unsigned extraPrefixArgs, unsigned extraSuffixArgs, bool passProtoArgs) {
// FIXME: Kill copy.
llvm::SmallVector<CanQualType, 16> argTypes;
for (const auto &arg : args)
argTypes.push_back(astContext.getCanonicalParamType(arg.ty));
// +1 for implicit this, which should always be args[0]
unsigned totalPrefixArgs = 1 + extraPrefixArgs;
CanQual<FunctionProtoType> fpt = getFormalType(d);
RequiredArgs required = passProtoArgs
? RequiredArgs::getFromProtoWithExtraSlots(
fpt, totalPrefixArgs + extraSuffixArgs)
: RequiredArgs::All;
GlobalDecl gd(d, ctorKind);
if (theCXXABI.hasThisReturn(gd))
cgm.errorNYI(d->getSourceRange(),
"arrangeCXXConstructorCall: hasThisReturn");
if (theCXXABI.hasMostDerivedReturn(gd))
cgm.errorNYI(d->getSourceRange(),
"arrangeCXXConstructorCall: hasMostDerivedReturn");
CanQualType resultType = astContext.VoidTy;
assert(!cir::MissingFeatures::opCallFnInfoOpts());
assert(!cir::MissingFeatures::opCallCIRGenFuncInfoExtParamInfo());
return arrangeCIRFunctionInfo(resultType, /*isInstanceMethod=*/true, argTypes,
fpt->getExtInfo(), required);
}
/// Arrange a call to a C++ method, passing the given arguments.
///
/// numPrefixArgs is the number of the ABI-specific prefix arguments we have. It
/// does not count `this`.
const CIRGenFunctionInfo &CIRGenTypes::arrangeCXXMethodCall(
const CallArgList &args, const FunctionProtoType *proto,
RequiredArgs required, unsigned numPrefixArgs) {
assert(!cir::MissingFeatures::opCallExtParameterInfo());
assert(numPrefixArgs + 1 <= args.size() &&
"Emitting a call with less args than the required prefix?");
// FIXME: Kill copy.
llvm::SmallVector<CanQualType, 16> argTypes;
for (const CallArg &arg : args)
argTypes.push_back(astContext.getCanonicalParamType(arg.ty));
assert(!cir::MissingFeatures::opCallFnInfoOpts());
return arrangeCIRFunctionInfo(
proto->getReturnType()->getCanonicalTypeUnqualified(),
/*isInstanceMethod=*/true, argTypes, proto->getExtInfo(), required);
}
const CIRGenFunctionInfo &
CIRGenTypes::arrangeFreeFunctionCall(const CallArgList &args,
const FunctionType *fnType) {
return arrangeFreeFunctionLikeCall(*this, cgm, args, fnType);
}
/// Arrange the argument and result information for a declaration or definition
/// of the given C++ non-static member function. The member function must be an
/// ordinary function, i.e. not a constructor or destructor.
const CIRGenFunctionInfo &
CIRGenTypes::arrangeCXXMethodDeclaration(const CXXMethodDecl *md) {
assert(!isa<CXXConstructorDecl>(md) && "wrong method for constructors!");
assert(!isa<CXXDestructorDecl>(md) && "wrong method for destructors!");
auto prototype =
md->getType()->getCanonicalTypeUnqualified().getAs<FunctionProtoType>();
assert(!cir::MissingFeatures::cudaSupport());
if (md->isInstance()) {
// The abstract case is perfectly fine.
auto *thisType = theCXXABI.getThisArgumentTypeForMethod(md);
return arrangeCXXMethodType(thisType, prototype.getTypePtr(), md);
}
return arrangeFreeFunctionType(prototype);
}
/// Arrange the argument and result information for a call to an unknown C++
/// non-static member function of the given abstract type. (A null RD means we
/// don't have any meaningful "this" argument type, so fall back to a generic
/// pointer type). The member fucntion must be an ordinary function, i.e. not a
/// constructor or destructor.
const CIRGenFunctionInfo &
CIRGenTypes::arrangeCXXMethodType(const CXXRecordDecl *rd,
const FunctionProtoType *fpt,
const CXXMethodDecl *md) {
llvm::SmallVector<CanQualType, 16> argTypes;
// Add the 'this' pointer.
argTypes.push_back(deriveThisType(rd, md));
assert(!cir::MissingFeatures::opCallFnInfoOpts());
return ::arrangeCIRFunctionInfo(
*this, /*isInstanceMethod=*/true, argTypes,
fpt->getCanonicalTypeUnqualified().getAs<FunctionProtoType>());
}
/// Arrange the argument and result information for the declaration or
/// definition of the given function.
const CIRGenFunctionInfo &
CIRGenTypes::arrangeFunctionDeclaration(const FunctionDecl *fd) {
if (const auto *md = dyn_cast<CXXMethodDecl>(fd))
if (md->isInstance())
return arrangeCXXMethodDeclaration(md);
CanQualType funcTy = fd->getType()->getCanonicalTypeUnqualified();
assert(isa<FunctionType>(funcTy));
// TODO: setCUDAKernelCallingConvention
assert(!cir::MissingFeatures::cudaSupport());
// When declaring a function without a prototype, always use a non-variadic
// type.
if (CanQual<FunctionNoProtoType> noProto =
funcTy.getAs<FunctionNoProtoType>()) {
assert(!cir::MissingFeatures::opCallCIRGenFuncInfoExtParamInfo());
assert(!cir::MissingFeatures::opCallFnInfoOpts());
return arrangeCIRFunctionInfo(noProto->getReturnType(),
/*isInstanceMethod=*/false, {},
noProto->getExtInfo(), RequiredArgs::All);
}
return arrangeFreeFunctionType(funcTy.castAs<FunctionProtoType>());
}
RValue CallArg::getRValue(CIRGenFunction &cgf, mlir::Location loc) const {
if (!hasLV)
return rv;
LValue copy = cgf.makeAddrLValue(cgf.createMemTemp(ty, loc), ty);
cgf.emitAggregateCopy(copy, lv, ty, AggValueSlot::DoesNotOverlap,
lv.isVolatile());
isUsed = true;
return RValue::getAggregate(copy.getAddress());
}
void CIRGenFunction::emitNonNullArgCheck(RValue rv, QualType argType,
SourceLocation argLoc,
AbstractCallee ac, unsigned paramNum) {
if (!ac.getDecl() || !(sanOpts.has(SanitizerKind::NonnullAttribute) ||
sanOpts.has(SanitizerKind::NullabilityArg)))
return;
cgm.errorNYI("non-null arg check is NYI");
}
static cir::CIRCallOpInterface
emitCallLikeOp(CIRGenFunction &cgf, mlir::Location callLoc,
cir::FuncType indirectFuncTy, mlir::Value indirectFuncVal,
cir::FuncOp directFuncOp,
const SmallVectorImpl<mlir::Value> &cirCallArgs, bool isInvoke,
const mlir::NamedAttrList &attrs,
llvm::ArrayRef<mlir::NamedAttrList> argAttrs,
const mlir::NamedAttrList &retAttrs) {
CIRGenBuilderTy &builder = cgf.getBuilder();
assert(!cir::MissingFeatures::opCallSurroundingTry());
assert(builder.getInsertionBlock() && "expected valid basic block");
cir::CallOp op;
if (indirectFuncTy) {
// TODO(cir): Set calling convention for indirect calls.
assert(!cir::MissingFeatures::opCallCallConv());
op = builder.createIndirectCallOp(callLoc, indirectFuncVal, indirectFuncTy,
cirCallArgs, attrs, argAttrs, retAttrs);
} else {
op = builder.createCallOp(callLoc, directFuncOp, cirCallArgs, attrs,
argAttrs, retAttrs);
}
return op;
}
const CIRGenFunctionInfo &
CIRGenTypes::arrangeFreeFunctionType(CanQual<FunctionProtoType> fpt) {
SmallVector<CanQualType, 16> argTypes;
assert(!cir::MissingFeatures::opCallFnInfoOpts());
return ::arrangeCIRFunctionInfo(*this, /*isInstanceMethod=*/false, argTypes,
fpt);
}
const CIRGenFunctionInfo &
CIRGenTypes::arrangeFreeFunctionType(CanQual<FunctionNoProtoType> fnpt) {
CanQualType resultType = fnpt->getReturnType().getUnqualifiedType();
assert(!cir::MissingFeatures::opCallFnInfoOpts());
return arrangeCIRFunctionInfo(resultType, /*isInstanceMethod=*/false, {},
fnpt->getExtInfo(), RequiredArgs(0));
}
RValue CIRGenFunction::emitCall(const CIRGenFunctionInfo &funcInfo,
const CIRGenCallee &callee,
ReturnValueSlot returnValue,
const CallArgList &args,
cir::CIRCallOpInterface *callOp,
mlir::Location loc) {
QualType retTy = funcInfo.getReturnType();
cir::FuncType cirFuncTy = getTypes().getFunctionType(funcInfo);
SmallVector<mlir::Value, 16> cirCallArgs(args.size());
assert(!cir::MissingFeatures::emitLifetimeMarkers());
// Translate all of the arguments as necessary to match the CIR lowering.
for (auto [argNo, arg, canQualArgType] :
llvm::enumerate(args, funcInfo.argTypes())) {
// Insert a padding argument to ensure proper alignment.
assert(!cir::MissingFeatures::opCallPaddingArgs());
mlir::Type argType = convertType(canQualArgType);
if (!mlir::isa<cir::RecordType>(argType) &&
!mlir::isa<cir::ComplexType>(argType)) {
mlir::Value v;
if (arg.isAggregate())
cgm.errorNYI(loc, "emitCall: aggregate call argument");
v = arg.getKnownRValue().getValue();
// We might have to widen integers, but we should never truncate.
if (argType != v.getType() && mlir::isa<cir::IntType>(v.getType()))
cgm.errorNYI(loc, "emitCall: widening integer call argument");
// If the argument doesn't match, perform a bitcast to coerce it. This
// can happen due to trivial type mismatches.
// TODO(cir): When getFunctionType is added, assert that this isn't
// needed.
assert(!cir::MissingFeatures::opCallBitcastArg());
cirCallArgs[argNo] = v;
} else {
Address src = Address::invalid();
if (!arg.isAggregate()) {
src = createMemTemp(arg.ty, loc, "coerce");
arg.copyInto(*this, src, loc);
} else {
src = arg.hasLValue() ? arg.getKnownLValue().getAddress()
: arg.getKnownRValue().getAggregateAddress();
}
// Fast-isel and the optimizer generally like scalar values better than
// FCAs, so we flatten them if this is safe to do for this argument.
mlir::Type srcTy = src.getElementType();
// FIXME(cir): get proper location for each argument.
mlir::Location argLoc = loc;
// If the source type is smaller than the destination type of the
// coerce-to logic, copy the source value into a temp alloca the size
// of the destination type to allow loading all of it. The bits past
// the source value are left undef.
// FIXME(cir): add data layout info and compare sizes instead of
// matching the types.
//
// uint64_t SrcSize = CGM.getDataLayout().getTypeAllocSize(SrcTy);
// uint64_t DstSize = CGM.getDataLayout().getTypeAllocSize(STy);
// if (SrcSize < DstSize) {
assert(!cir::MissingFeatures::dataLayoutTypeAllocSize());
if (srcTy != argType) {
cgm.errorNYI(loc, "emitCall: source type does not match argument type");
} else {
// FIXME(cir): this currently only runs when the types are exactly the
// same, but should be when alloc sizes are the same, fix this as soon
// as datalayout gets introduced.
assert(!cir::MissingFeatures::dataLayoutTypeAllocSize());
}
// assert(NumCIRArgs == STy.getMembers().size());
// In LLVMGen: Still only pass the struct without any gaps but mark it
// as such somehow.
//
// In CIRGen: Emit a load from the "whole" struct,
// which shall be broken later by some lowering step into multiple
// loads.
assert(!cir::MissingFeatures::lowerAggregateLoadStore());
cirCallArgs[argNo] = builder.createLoad(argLoc, src);
}
}
const CIRGenCallee &concreteCallee = callee.prepareConcreteCallee(*this);
mlir::Operation *calleePtr = concreteCallee.getFunctionPointer();
assert(!cir::MissingFeatures::opCallInAlloca());
mlir::NamedAttrList attrs;
std::vector<mlir::NamedAttrList> argAttrs(funcInfo.arguments().size());
mlir::NamedAttrList retAttrs;
StringRef funcName;
if (auto calleeFuncOp = dyn_cast<cir::FuncOp>(calleePtr))
funcName = calleeFuncOp.getName();
assert(!cir::MissingFeatures::opCallCallConv());
assert(!cir::MissingFeatures::opCallAttrs());
cir::CallingConv callingConv;
cir::SideEffect sideEffect;
cgm.constructAttributeList(funcName, funcInfo, callee.getAbstractInfo(),
attrs, argAttrs, retAttrs, callingConv, sideEffect,
/*attrOnCallSite=*/true, /*isThunk=*/false);
cir::FuncType indirectFuncTy;
mlir::Value indirectFuncVal;
cir::FuncOp directFuncOp;
if (auto fnOp = dyn_cast<cir::FuncOp>(calleePtr)) {
directFuncOp = fnOp;
} else if (auto getGlobalOp = mlir::dyn_cast<cir::GetGlobalOp>(calleePtr)) {
// FIXME(cir): This peephole optimization avoids indirect calls for
// builtins. This should be fixed in the builtin declaration instead by
// not emitting an unecessary get_global in the first place.
// However, this is also used for no-prototype functions.
mlir::Operation *globalOp = cgm.getGlobalValue(getGlobalOp.getName());
assert(globalOp && "undefined global function");
directFuncOp = mlir::cast<cir::FuncOp>(globalOp);
} else {
[[maybe_unused]] mlir::ValueTypeRange<mlir::ResultRange> resultTypes =
calleePtr->getResultTypes();
[[maybe_unused]] auto funcPtrTy =
mlir::dyn_cast<cir::PointerType>(resultTypes.front());
assert(funcPtrTy && mlir::isa<cir::FuncType>(funcPtrTy.getPointee()) &&
"expected pointer to function");
indirectFuncTy = cirFuncTy;
indirectFuncVal = calleePtr->getResult(0);
}
assert(!cir::MissingFeatures::msvcCXXPersonality());
assert(!cir::MissingFeatures::functionUsesSEHTry());
assert(!cir::MissingFeatures::nothrowAttr());
bool cannotThrow = attrs.getNamed("nothrow").has_value();
bool isInvoke = !cannotThrow && isCatchOrCleanupRequired();
mlir::Location callLoc = loc;
cir::CIRCallOpInterface theCall =
emitCallLikeOp(*this, loc, indirectFuncTy, indirectFuncVal, directFuncOp,
cirCallArgs, isInvoke, attrs, argAttrs, retAttrs);
if (callOp)
*callOp = theCall;
assert(!cir::MissingFeatures::opCallMustTail());
assert(!cir::MissingFeatures::opCallReturn());
mlir::Type retCIRTy = convertType(retTy);
if (isa<cir::VoidType>(retCIRTy))
return getUndefRValue(retTy);
switch (getEvaluationKind(retTy)) {
case cir::TEK_Aggregate: {
Address destPtr = returnValue.getValue();
if (!destPtr.isValid())
destPtr = createMemTemp(retTy, callLoc, getCounterAggTmpAsString());
mlir::ResultRange results = theCall->getOpResults();
assert(results.size() <= 1 && "multiple returns from a call");
SourceLocRAIIObject loc{*this, callLoc};
emitAggregateStore(results[0], destPtr);
return RValue::getAggregate(destPtr);
}
case cir::TEK_Scalar: {
mlir::ResultRange results = theCall->getOpResults();
assert(results.size() == 1 && "unexpected number of returns");
// If the argument doesn't match, perform a bitcast to coerce it. This
// can happen due to trivial type mismatches.
if (results[0].getType() != retCIRTy)
cgm.errorNYI(loc, "bitcast on function return value");
mlir::Region *region = builder.getBlock()->getParent();
if (region != theCall->getParentRegion())
cgm.errorNYI(loc, "function calls with cleanup");
return RValue::get(results[0]);
}
case cir::TEK_Complex: {
mlir::ResultRange results = theCall->getOpResults();
assert(!results.empty() &&
"Expected at least one result for complex rvalue");
return RValue::getComplex(results[0]);
}
}
llvm_unreachable("Invalid evaluation kind");
}
void CallArg::copyInto(CIRGenFunction &cgf, Address addr,
mlir::Location loc) const {
LValue dst = cgf.makeAddrLValue(addr, ty);
if (!hasLV && rv.isScalar())
cgf.cgm.errorNYI(loc, "copyInto scalar value");
else if (!hasLV && rv.isComplex())
cgf.emitStoreOfComplex(loc, rv.getComplexValue(), dst, /*isInit=*/true);
else
cgf.cgm.errorNYI(loc, "copyInto hasLV");
isUsed = true;
}
mlir::Value CIRGenFunction::emitRuntimeCall(mlir::Location loc,
cir::FuncOp callee,
ArrayRef<mlir::Value> args,
mlir::NamedAttrList attrs) {
// TODO(cir): set the calling convention to this runtime call.
assert(!cir::MissingFeatures::opFuncCallingConv());
cir::CallOp call = builder.createCallOp(loc, callee, args);
assert(call->getNumResults() <= 1 &&
"runtime functions have at most 1 result");
if (!attrs.empty())
call->setAttrs(attrs);
if (call->getNumResults() == 0)
return nullptr;
return call->getResult(0);
}
void CIRGenFunction::emitCallArg(CallArgList &args, const clang::Expr *e,
clang::QualType argType) {
assert(argType->isReferenceType() == e->isGLValue() &&
"reference binding to unmaterialized r-value!");
if (e->isGLValue()) {
assert(e->getObjectKind() == OK_Ordinary);
return args.add(emitReferenceBindingToExpr(e), argType);
}
bool hasAggregateEvalKind = hasAggregateEvaluationKind(argType);
// In the Microsoft C++ ABI, aggregate arguments are destructed by the callee.
// However, we still have to push an EH-only cleanup in case we unwind before
// we make it to the call.
if (argType->isRecordType() &&
argType->castAsRecordDecl()->isParamDestroyedInCallee()) {
assert(!cir::MissingFeatures::msabi());
cgm.errorNYI(e->getSourceRange(), "emitCallArg: msabi is NYI");
}
if (hasAggregateEvalKind && isa<ImplicitCastExpr>(e) &&
cast<CastExpr>(e)->getCastKind() == CK_LValueToRValue) {
LValue lv = emitLValue(cast<CastExpr>(e)->getSubExpr());
assert(lv.isSimple());
args.addUncopiedAggregate(lv, argType);
return;
}
args.add(emitAnyExprToTemp(e), argType);
}
QualType CIRGenFunction::getVarArgType(const Expr *arg) {
// System headers on Windows define NULL to 0 instead of 0LL on Win64. MSVC
// implicitly widens null pointer constants that are arguments to varargs
// functions to pointer-sized ints.
if (!getTarget().getTriple().isOSWindows())
return arg->getType();
assert(!cir::MissingFeatures::msabi());
cgm.errorNYI(arg->getSourceRange(), "getVarArgType: NYI for Windows target");
return arg->getType();
}
/// Similar to emitAnyExpr(), however, the result will always be accessible
/// even if no aggregate location is provided.
RValue CIRGenFunction::emitAnyExprToTemp(const Expr *e) {
AggValueSlot aggSlot = AggValueSlot::ignored();
if (hasAggregateEvaluationKind(e->getType()))
aggSlot = createAggTemp(e->getType(), getLoc(e->getSourceRange()),
getCounterAggTmpAsString());
return emitAnyExpr(e, aggSlot);
}
void CIRGenFunction::emitCallArgs(
CallArgList &args, PrototypeWrapper prototype,
llvm::iterator_range<clang::CallExpr::const_arg_iterator> argRange,
AbstractCallee callee, unsigned paramsToSkip) {
llvm::SmallVector<QualType, 16> argTypes;
assert(!cir::MissingFeatures::opCallCallConv());
// First, if a prototype was provided, use those argument types.
bool isVariadic = false;
if (prototype.p) {
assert(!cir::MissingFeatures::opCallObjCMethod());
const auto *fpt = cast<const FunctionProtoType *>(prototype.p);
isVariadic = fpt->isVariadic();
assert(!cir::MissingFeatures::opCallCallConv());
argTypes.assign(fpt->param_type_begin() + paramsToSkip,
fpt->param_type_end());
}
// If we still have any arguments, emit them using the type of the argument.
for (const clang::Expr *a : llvm::drop_begin(argRange, argTypes.size()))
argTypes.push_back(isVariadic ? getVarArgType(a) : a->getType());
assert(argTypes.size() == (size_t)(argRange.end() - argRange.begin()));
// We must evaluate arguments from right to left in the MS C++ ABI, because
// arguments are destroyed left to right in the callee. As a special case,
// there are certain language constructs taht require left-to-right
// evaluation, and in those cases we consider the evaluation order requirement
// to trump the "destruction order is reverse construction order" guarantee.
auto leftToRight = true;
assert(!cir::MissingFeatures::msabi());
auto maybeEmitImplicitObjectSize = [&](size_t i, const Expr *arg,
RValue emittedArg) {
if (!callee.hasFunctionDecl() || i >= callee.getNumParams())
return;
auto *ps = callee.getParamDecl(i)->getAttr<PassObjectSizeAttr>();
if (!ps)
return;
assert(!cir::MissingFeatures::opCallImplicitObjectSizeArgs());
cgm.errorNYI("emit implicit object size for call arg");
};
// Evaluate each argument in the appropriate order.
size_t callArgsStart = args.size();
for (size_t i = 0; i != argTypes.size(); ++i) {
size_t idx = leftToRight ? i : argTypes.size() - i - 1;
CallExpr::const_arg_iterator currentArg = argRange.begin() + idx;
size_t initialArgSize = args.size();
emitCallArg(args, *currentArg, argTypes[idx]);
// In particular, we depend on it being the last arg in Args, and the
// objectsize bits depend on there only being one arg if !LeftToRight.
assert(initialArgSize + 1 == args.size() &&
"The code below depends on only adding one arg per emitCallArg");
(void)initialArgSize;
// Since pointer argument are never emitted as LValue, it is safe to emit
// non-null argument check for r-value only.
if (!args.back().hasLValue()) {
RValue rvArg = args.back().getKnownRValue();
assert(!cir::MissingFeatures::sanitizers());
maybeEmitImplicitObjectSize(idx, *currentArg, rvArg);
}
if (!leftToRight)
std::reverse(args.begin() + callArgsStart, args.end());
}
}