bool Inliner::runOnSCC(CallGraphSCC &SCC) {
CallGraph &CG = getAnalysis<CallGraph>();
const TargetData *TD = getAnalysisIfAvailable<TargetData>();
SmallPtrSet<Function*, 8> SCCFunctions;
DEBUG(dbgs() << "Inliner visiting SCC:");
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
Function *F = (*I)->getFunction();
if (F) SCCFunctions.insert(F);
DEBUG(dbgs() << " " << (F ? F->getName() : "INDIRECTNODE"));
}
// Scan through and identify all call sites ahead of time so that we only
// inline call sites in the original functions, not call sites that result
// from inlining other functions.
SmallVector<std::pair<CallSite, int>, 16> CallSites;
// When inlining a callee produces new call sites, we want to keep track of
// the fact that they were inlined from the callee. This allows us to avoid
// infinite inlining in some obscure cases. To represent this, we use an
// index into the InlineHistory vector.
SmallVector<std::pair<Function*, int>, 8> InlineHistory;
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
Function *F = (*I)->getFunction();
if (!F) continue;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
CallSite CS(cast<Value>(I));
// If this isn't a call, or it is a call to an intrinsic, it can
// never be inlined.
if (!CS || isa<IntrinsicInst>(I))
continue;
// If this is a direct call to an external function, we can never inline
// it. If it is an indirect call, inlining may resolve it to be a
// direct call, so we keep it.
if (CS.getCalledFunction() && CS.getCalledFunction()->isDeclaration())
continue;
CallSites.push_back(std::make_pair(CS, -1));
}
}
DEBUG(dbgs() << ": " << CallSites.size() << " call sites.\n");
// If there are no calls in this function, exit early.
if (CallSites.empty())
return false;
// Now that we have all of the call sites, move the ones to functions in the
// current SCC to the end of the list.
unsigned FirstCallInSCC = CallSites.size();
for (unsigned i = 0; i < FirstCallInSCC; ++i)
if (Function *F = CallSites[i].first.getCalledFunction())
if (SCCFunctions.count(F))
std::swap(CallSites[i--], CallSites[--FirstCallInSCC]);
InlinedArrayAllocasTy InlinedArrayAllocas;
InlineFunctionInfo InlineInfo(&CG, TD);
// Now that we have all of the call sites, loop over them and inline them if
// it looks profitable to do so.
bool Changed = false;
bool LocalChange;
do {
LocalChange = false;
// Iterate over the outer loop because inlining functions can cause indirect
// calls to become direct calls.
for (unsigned CSi = 0; CSi != CallSites.size(); ++CSi) {
CallSite CS = CallSites[CSi].first;
Function *Caller = CS.getCaller();
Function *Callee = CS.getCalledFunction();
// If this call site is dead and it is to a readonly function, we should
// just delete the call instead of trying to inline it, regardless of
// size. This happens because IPSCCP propagates the result out of the
// call and then we're left with the dead call.
if (isInstructionTriviallyDead(CS.getInstruction())) {
DEBUG(dbgs() << " -> Deleting dead call: "
<< *CS.getInstruction() << "\n");
// Update the call graph by deleting the edge from Callee to Caller.
CG[Caller]->removeCallEdgeFor(CS);
CS.getInstruction()->eraseFromParent();
++NumCallsDeleted;
// Update the cached cost info with the missing call
growCachedCostInfo(Caller, NULL);
} else {
// We can only inline direct calls to non-declarations.
if (Callee == 0 || Callee->isDeclaration()) continue;
// If this call site was obtained by inlining another function, verify
// that the include path for the function did not include the callee
// itself. If so, we'd be recursively inlining the same function,
// which would provide the same callsites, which would cause us to
// infinitely inline.
int InlineHistoryID = CallSites[CSi].second;
if (InlineHistoryID != -1 &&
InlineHistoryIncludes(Callee, InlineHistoryID, InlineHistory))
continue;
// If the policy determines that we should inline this function,
// try to do so.
if (!shouldInline(CS))
continue;
// Attempt to inline the function.
if (!InlineCallIfPossible(CS, InlineInfo, InlinedArrayAllocas,
InlineHistoryID, InsertLifetime))
continue;
++NumInlined;
// If inlining this function gave us any new call sites, throw them
// onto our worklist to process. They are useful inline candidates.
if (!InlineInfo.InlinedCalls.empty()) {
// Create a new inline history entry for this, so that we remember
// that these new callsites came about due to inlining Callee.
int NewHistoryID = InlineHistory.size();
InlineHistory.push_back(std::make_pair(Callee, InlineHistoryID));
for (unsigned i = 0, e = InlineInfo.InlinedCalls.size();
i != e; ++i) {
Value *Ptr = InlineInfo.InlinedCalls[i];
CallSite NewCS = Ptr;
simplifyCallSiteArguments(TD, NewCS);
CallSites.push_back(std::make_pair(NewCS, NewHistoryID));
}
}
// Update the cached cost info with the inlined call.
growCachedCostInfo(Caller, Callee);
}
// If we inlined or deleted the last possible call site to the function,
// delete the function body now.
if (Callee && Callee->use_empty() && Callee->hasLocalLinkage() &&
// TODO: Can remove if in SCC now.
!SCCFunctions.count(Callee) &&
// The function may be apparently dead, but if there are indirect
// callgraph references to the node, we cannot delete it yet, this
// could invalidate the CGSCC iterator.
CG[Callee]->getNumReferences() == 0) {
DEBUG(dbgs() << " -> Deleting dead function: "
<< Callee->getName() << "\n");
CallGraphNode *CalleeNode = CG[Callee];
// Remove any call graph edges from the callee to its callees.
CalleeNode->removeAllCalledFunctions();
resetCachedCostInfo(Callee);
// Removing the node for callee from the call graph and delete it.
delete CG.removeFunctionFromModule(CalleeNode);
++NumDeleted;
}
// Remove this call site from the list. If possible, use
// swap/pop_back for efficiency, but do not use it if doing so would
// move a call site to a function in this SCC before the
// 'FirstCallInSCC' barrier.
if (SCC.isSingular()) {
CallSites[CSi] = CallSites.back();
CallSites.pop_back();
} else {
CallSites.erase(CallSites.begin()+CSi);
}
--CSi;
Changed = true;
LocalChange = true;
}
} while (LocalChange);
return Changed;
}
void UseOverrideCheck::check(const MatchFinder::MatchResult &Result) {
const FunctionDecl *Method = Result.Nodes.getStmtAs<FunctionDecl>("method");
const SourceManager &Sources = *Result.SourceManager;
assert(Method != nullptr);
if (Method->getInstantiatedFromMemberFunction() != nullptr)
Method = Method->getInstantiatedFromMemberFunction();
if (Method->isImplicit() || Method->getLocation().isMacroID() ||
Method->isOutOfLine())
return;
bool HasVirtual = Method->isVirtualAsWritten();
bool HasOverride = Method->getAttr<OverrideAttr>();
bool HasFinal = Method->getAttr<FinalAttr>();
bool OnlyVirtualSpecified = HasVirtual && !HasOverride && !HasFinal;
unsigned KeywordCount = HasVirtual + HasOverride + HasFinal;
if (!OnlyVirtualSpecified && KeywordCount == 1)
return; // Nothing to do.
std::string Message;
if (OnlyVirtualSpecified) {
Message =
"prefer using 'override' or (rarely) 'final' instead of 'virtual'";
} else if (KeywordCount == 0) {
Message = "annotate this function with 'override' or (rarely) 'final'";
} else {
StringRef Redundant =
HasVirtual ? (HasOverride && HasFinal ? "'virtual' and 'override' are"
: "'virtual' is")
: "'override' is";
StringRef Correct = HasFinal ? "'final'" : "'override'";
Message =
(llvm::Twine(Redundant) +
" redundant since the function is already declared " + Correct).str();
}
DiagnosticBuilder Diag = diag(Method->getLocation(), Message);
CharSourceRange FileRange = Lexer::makeFileCharRange(
CharSourceRange::getTokenRange(Method->getSourceRange()), Sources,
Result.Context->getLangOpts());
if (!FileRange.isValid())
return;
// FIXME: Instead of re-lexing and looking for specific macros such as
// 'ABSTRACT', properly store the location of 'virtual' and '= 0' in each
// FunctionDecl.
SmallVector<Token, 16> Tokens = ParseTokens(FileRange, Result);
// Add 'override' on inline declarations that don't already have it.
if (!HasFinal && !HasOverride) {
SourceLocation InsertLoc;
StringRef ReplacementText = "override ";
for (Token T : Tokens) {
if (T.is(tok::kw___attribute)) {
InsertLoc = T.getLocation();
break;
}
}
if (Method->hasAttrs()) {
for (const clang::Attr *A : Method->getAttrs()) {
if (!A->isImplicit()) {
SourceLocation Loc =
Sources.getExpansionLoc(A->getRange().getBegin());
if (!InsertLoc.isValid() ||
Sources.isBeforeInTranslationUnit(Loc, InsertLoc))
InsertLoc = Loc;
}
}
}
if (InsertLoc.isInvalid() && Method->doesThisDeclarationHaveABody() &&
Method->getBody() && !Method->isDefaulted()) {
// For methods with inline definition, add the override keyword at the
// end of the declaration of the function, but prefer to put it on the
// same line as the declaration if the beginning brace for the start of
// the body falls on the next line.
Token LastNonCommentToken;
for (Token T : Tokens) {
if (!T.is(tok::comment)) {
LastNonCommentToken = T;
}
}
InsertLoc = LastNonCommentToken.getEndLoc();
ReplacementText = " override";
}
if (!InsertLoc.isValid()) {
// For declarations marked with "= 0" or "= [default|delete]", the end
// location will point until after those markings. Therefore, the override
// keyword shouldn't be inserted at the end, but before the '='.
if (Tokens.size() > 2 && (GetText(Tokens.back(), Sources) == "0" ||
Tokens.back().is(tok::kw_default) ||
Tokens.back().is(tok::kw_delete)) &&
GetText(Tokens[Tokens.size() - 2], Sources) == "=") {
InsertLoc = Tokens[Tokens.size() - 2].getLocation();
} else if (GetText(Tokens.back(), Sources) == "ABSTRACT") {
InsertLoc = Tokens.back().getLocation();
}
}
if (!InsertLoc.isValid()) {
InsertLoc = FileRange.getEnd();
ReplacementText = " override";
}
Diag << FixItHint::CreateInsertion(InsertLoc, ReplacementText);
}
if (HasFinal && HasOverride) {
SourceLocation OverrideLoc = Method->getAttr<OverrideAttr>()->getLocation();
Diag << FixItHint::CreateRemoval(
CharSourceRange::getTokenRange(OverrideLoc, OverrideLoc));
}
if (HasVirtual) {
for (Token Tok : Tokens) {
if (Tok.is(tok::kw_virtual)) {
Diag << FixItHint::CreateRemoval(CharSourceRange::getTokenRange(
Tok.getLocation(), Tok.getLocation()));
break;
}
}
}
}
// Propagate existing explicit probabilities from either profile data or
// 'expect' intrinsic processing. Examine metadata against unreachable
// heuristic. The probability of the edge coming to unreachable block is
// set to min of metadata and unreachable heuristic.
bool BranchProbabilityInfo::calcMetadataWeights(const BasicBlock *BB) {
const TerminatorInst *TI = BB->getTerminator();
assert(TI->getNumSuccessors() > 1 && "expected more than one successor!");
if (!isa<BranchInst>(TI) && !isa<SwitchInst>(TI))
return false;
MDNode *WeightsNode = TI->getMetadata(LLVMContext::MD_prof);
if (!WeightsNode)
return false;
// Check that the number of successors is manageable.
assert(TI->getNumSuccessors() < UINT32_MAX && "Too many successors");
// Ensure there are weights for all of the successors. Note that the first
// operand to the metadata node is a name, not a weight.
if (WeightsNode->getNumOperands() != TI->getNumSuccessors() + 1)
return false;
// Build up the final weights that will be used in a temporary buffer.
// Compute the sum of all weights to later decide whether they need to
// be scaled to fit in 32 bits.
uint64_t WeightSum = 0;
SmallVector<uint32_t, 2> Weights;
SmallVector<unsigned, 2> UnreachableIdxs;
SmallVector<unsigned, 2> ReachableIdxs;
Weights.reserve(TI->getNumSuccessors());
for (unsigned i = 1, e = WeightsNode->getNumOperands(); i != e; ++i) {
ConstantInt *Weight =
mdconst::dyn_extract<ConstantInt>(WeightsNode->getOperand(i));
if (!Weight)
return false;
assert(Weight->getValue().getActiveBits() <= 32 &&
"Too many bits for uint32_t");
Weights.push_back(Weight->getZExtValue());
WeightSum += Weights.back();
if (PostDominatedByUnreachable.count(TI->getSuccessor(i - 1)))
UnreachableIdxs.push_back(i - 1);
else
ReachableIdxs.push_back(i - 1);
}
assert(Weights.size() == TI->getNumSuccessors() && "Checked above");
// If the sum of weights does not fit in 32 bits, scale every weight down
// accordingly.
uint64_t ScalingFactor =
(WeightSum > UINT32_MAX) ? WeightSum / UINT32_MAX + 1 : 1;
if (ScalingFactor > 1) {
WeightSum = 0;
for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i) {
Weights[i] /= ScalingFactor;
WeightSum += Weights[i];
}
}
assert(WeightSum <= UINT32_MAX &&
"Expected weights to scale down to 32 bits");
if (WeightSum == 0 || ReachableIdxs.size() == 0) {
for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
Weights[i] = 1;
WeightSum = TI->getNumSuccessors();
}
// Set the probability.
SmallVector<BranchProbability, 2> BP;
for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
BP.push_back({ Weights[i], static_cast<uint32_t>(WeightSum) });
// Examine the metadata against unreachable heuristic.
// If the unreachable heuristic is more strong then we use it for this edge.
if (UnreachableIdxs.size() > 0 && ReachableIdxs.size() > 0) {
auto ToDistribute = BranchProbability::getZero();
auto UnreachableProb = UR_TAKEN_PROB;
for (auto i : UnreachableIdxs)
if (UnreachableProb < BP[i]) {
ToDistribute += BP[i] - UnreachableProb;
BP[i] = UnreachableProb;
}
// If we modified the probability of some edges then we must distribute
// the difference between reachable blocks.
if (ToDistribute > BranchProbability::getZero()) {
BranchProbability PerEdge = ToDistribute / ReachableIdxs.size();
for (auto i : ReachableIdxs)
BP[i] += PerEdge;
}
}
for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
setEdgeProbability(BB, i, BP[i]);
return true;
}
void FinalOverriderCollector::Collect(const CXXRecordDecl *RD,
bool VirtualBase,
const CXXRecordDecl *InVirtualSubobject,
CXXFinalOverriderMap &Overriders) {
unsigned SubobjectNumber = 0;
if (!VirtualBase)
SubobjectNumber
= ++SubobjectCount[cast<CXXRecordDecl>(RD->getCanonicalDecl())];
for (CXXRecordDecl::base_class_const_iterator Base = RD->bases_begin(),
BaseEnd = RD->bases_end(); Base != BaseEnd; ++Base) {
if (const RecordType *RT = Base->getType()->getAs<RecordType>()) {
const CXXRecordDecl *BaseDecl = cast<CXXRecordDecl>(RT->getDecl());
if (!BaseDecl->isPolymorphic())
continue;
if (Overriders.empty() && !Base->isVirtual()) {
// There are no other overriders of virtual member functions,
// so let the base class fill in our overriders for us.
Collect(BaseDecl, false, InVirtualSubobject, Overriders);
continue;
}
// Collect all of the overridders from the base class subobject
// and merge them into the set of overridders for this class.
// For virtual base classes, populate or use the cached virtual
// overrides so that we do not walk the virtual base class (and
// its base classes) more than once.
CXXFinalOverriderMap ComputedBaseOverriders;
CXXFinalOverriderMap *BaseOverriders = &ComputedBaseOverriders;
if (Base->isVirtual()) {
CXXFinalOverriderMap *&MyVirtualOverriders = VirtualOverriders[BaseDecl];
BaseOverriders = MyVirtualOverriders;
if (!MyVirtualOverriders) {
MyVirtualOverriders = new CXXFinalOverriderMap;
// Collect may cause VirtualOverriders to reallocate, invalidating the
// MyVirtualOverriders reference. Set BaseOverriders to the right
// value now.
BaseOverriders = MyVirtualOverriders;
Collect(BaseDecl, true, BaseDecl, *MyVirtualOverriders);
}
} else
Collect(BaseDecl, false, InVirtualSubobject, ComputedBaseOverriders);
// Merge the overriders from this base class into our own set of
// overriders.
for (CXXFinalOverriderMap::iterator OM = BaseOverriders->begin(),
OMEnd = BaseOverriders->end();
OM != OMEnd;
++OM) {
const CXXMethodDecl *CanonOM
= cast<CXXMethodDecl>(OM->first->getCanonicalDecl());
Overriders[CanonOM].add(OM->second);
}
}
}
for (CXXRecordDecl::method_iterator M = RD->method_begin(),
MEnd = RD->method_end();
M != MEnd;
++M) {
// We only care about virtual methods.
if (!M->isVirtual())
continue;
CXXMethodDecl *CanonM = cast<CXXMethodDecl>(M->getCanonicalDecl());
if (CanonM->begin_overridden_methods()
== CanonM->end_overridden_methods()) {
// This is a new virtual function that does not override any
// other virtual function. Add it to the map of virtual
// functions for which we are tracking overridders.
// C++ [class.virtual]p2:
// For convenience we say that any virtual function overrides itself.
Overriders[CanonM].add(SubobjectNumber,
UniqueVirtualMethod(CanonM, SubobjectNumber,
InVirtualSubobject));
continue;
}
// This virtual method overrides other virtual methods, so it does
// not add any new slots into the set of overriders. Instead, we
// replace entries in the set of overriders with the new
// overrider. To do so, we dig down to the original virtual
// functions using data recursion and update all of the methods it
// overrides.
typedef std::pair<CXXMethodDecl::method_iterator,
CXXMethodDecl::method_iterator> OverriddenMethods;
SmallVector<OverriddenMethods, 4> Stack;
Stack.push_back(std::make_pair(CanonM->begin_overridden_methods(),
CanonM->end_overridden_methods()));
while (!Stack.empty()) {
OverriddenMethods OverMethods = Stack.back();
Stack.pop_back();
for (; OverMethods.first != OverMethods.second; ++OverMethods.first) {
const CXXMethodDecl *CanonOM
= cast<CXXMethodDecl>((*OverMethods.first)->getCanonicalDecl());
// C++ [class.virtual]p2:
// A virtual member function C::vf of a class object S is
// a final overrider unless the most derived class (1.8)
// of which S is a base class subobject (if any) declares
// or inherits another member function that overrides vf.
//
// Treating this object like the most derived class, we
// replace any overrides from base classes with this
// overriding virtual function.
Overriders[CanonOM].replaceAll(
UniqueVirtualMethod(CanonM, SubobjectNumber,
InVirtualSubobject));
if (CanonOM->begin_overridden_methods()
== CanonOM->end_overridden_methods())
continue;
// Continue recursion to the methods that this virtual method
// overrides.
Stack.push_back(std::make_pair(CanonOM->begin_overridden_methods(),
CanonOM->end_overridden_methods()));
}
}
// C++ [class.virtual]p2:
// For convenience we say that any virtual function overrides itself.
Overriders[CanonM].add(SubobjectNumber,
UniqueVirtualMethod(CanonM, SubobjectNumber,
InVirtualSubobject));
}
}
void UnwrappedLineParser::calculateBraceTypes() {
// We'll parse forward through the tokens until we hit
// a closing brace or eof - note that getNextToken() will
// parse macros, so this will magically work inside macro
// definitions, too.
unsigned StoredPosition = Tokens->getPosition();
unsigned Position = StoredPosition;
FormatToken *Tok = FormatTok;
// Keep a stack of positions of lbrace tokens. We will
// update information about whether an lbrace starts a
// braced init list or a different block during the loop.
SmallVector<unsigned, 8> LBraceStack;
assert(Tok->Tok.is(tok::l_brace));
do {
FormatToken *NextTok = Tokens->getNextToken();
switch (Tok->Tok.getKind()) {
case tok::l_brace:
LBraceStack.push_back(Position);
break;
case tok::r_brace:
if (!LBraceStack.empty()) {
if (LBraces[LBraceStack.back()] == BS_Unknown) {
// If there is a comma, semicolon or right paren after the closing
// brace, we assume this is a braced initializer list.
// FIXME: Note that this currently works only because we do not
// use the brace information while inside a braced init list.
// Thus, if the parent is a braced init list, we consider all
// brace blocks inside it braced init list. That works good enough
// for now, but we will need to fix it to correctly handle lambdas.
if (NextTok->Tok.is(tok::comma) || NextTok->Tok.is(tok::semi) ||
NextTok->Tok.is(tok::r_paren) || NextTok->Tok.is(tok::l_brace))
LBraces[LBraceStack.back()] = BS_BracedInit;
else
LBraces[LBraceStack.back()] = BS_Block;
}
LBraceStack.pop_back();
}
break;
case tok::semi:
case tok::kw_if:
case tok::kw_while:
case tok::kw_for:
case tok::kw_switch:
case tok::kw_try:
if (!LBraceStack.empty())
LBraces[LBraceStack.back()] = BS_Block;
break;
default:
break;
}
Tok = NextTok;
++Position;
} while (Tok->Tok.isNot(tok::eof));
// Assume other blocks for all unclosed opening braces.
for (unsigned i = 0, e = LBraceStack.size(); i != e; ++i) {
if (LBraces[LBraceStack[i]] == BS_Unknown)
LBraces[LBraceStack[i]] = BS_Block;
}
FormatTok = Tokens->setPosition(StoredPosition);
}
bool GuardWideningImpl::combineRangeChecks(
SmallVectorImpl<GuardWideningImpl::RangeCheck> &Checks,
SmallVectorImpl<GuardWideningImpl::RangeCheck> &RangeChecksOut) {
unsigned OldCount = Checks.size();
while (!Checks.empty()) {
// Pick all of the range checks with a specific base and length, and try to
// merge them.
Value *CurrentBase = Checks.front().getBase();
Value *CurrentLength = Checks.front().getLength();
SmallVector<GuardWideningImpl::RangeCheck, 3> CurrentChecks;
auto IsCurrentCheck = [&](GuardWideningImpl::RangeCheck &RC) {
return RC.getBase() == CurrentBase && RC.getLength() == CurrentLength;
};
std::copy_if(Checks.begin(), Checks.end(),
std::back_inserter(CurrentChecks), IsCurrentCheck);
Checks.erase(remove_if(Checks, IsCurrentCheck), Checks.end());
assert(CurrentChecks.size() != 0 && "We know we have at least one!");
if (CurrentChecks.size() < 3) {
RangeChecksOut.insert(RangeChecksOut.end(), CurrentChecks.begin(),
CurrentChecks.end());
continue;
}
// CurrentChecks.size() will typically be 3 here, but so far there has been
// no need to hard-code that fact.
std::sort(CurrentChecks.begin(), CurrentChecks.end(),
[&](const GuardWideningImpl::RangeCheck &LHS,
const GuardWideningImpl::RangeCheck &RHS) {
return LHS.getOffsetValue().slt(RHS.getOffsetValue());
});
// Note: std::sort should not invalidate the ChecksStart iterator.
ConstantInt *MinOffset = CurrentChecks.front().getOffset(),
*MaxOffset = CurrentChecks.back().getOffset();
unsigned BitWidth = MaxOffset->getValue().getBitWidth();
if ((MaxOffset->getValue() - MinOffset->getValue())
.ugt(APInt::getSignedMinValue(BitWidth)))
return false;
APInt MaxDiff = MaxOffset->getValue() - MinOffset->getValue();
const APInt &HighOffset = MaxOffset->getValue();
auto OffsetOK = [&](const GuardWideningImpl::RangeCheck &RC) {
return (HighOffset - RC.getOffsetValue()).ult(MaxDiff);
};
if (MaxDiff.isMinValue() ||
!std::all_of(std::next(CurrentChecks.begin()), CurrentChecks.end(),
OffsetOK))
return false;
// We have a series of f+1 checks as:
//
// I+k_0 u< L ... Chk_0
// I_k_1 u< L ... Chk_1
// ...
// I_k_f u< L ... Chk_(f+1)
//
// with forall i in [0,f): k_f-k_i u< k_f-k_0 ... Precond_0
// k_f-k_0 u< INT_MIN+k_f ... Precond_1
// k_f != k_0 ... Precond_2
//
// Claim:
// Chk_0 AND Chk_(f+1) implies all the other checks
//
// Informal proof sketch:
//
// We will show that the integer range [I+k_0,I+k_f] does not unsigned-wrap
// (i.e. going from I+k_0 to I+k_f does not cross the -1,0 boundary) and
// thus I+k_f is the greatest unsigned value in that range.
//
// This combined with Ckh_(f+1) shows that everything in that range is u< L.
// Via Precond_0 we know that all of the indices in Chk_0 through Chk_(f+1)
// lie in [I+k_0,I+k_f], this proving our claim.
//
// To see that [I+k_0,I+k_f] is not a wrapping range, note that there are
// two possibilities: I+k_0 u< I+k_f or I+k_0 >u I+k_f (they can't be equal
// since k_0 != k_f). In the former case, [I+k_0,I+k_f] is not a wrapping
// range by definition, and the latter case is impossible:
//
// 0-----I+k_f---I+k_0----L---INT_MAX,INT_MIN------------------(-1)
// xxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
//
// For Chk_0 to succeed, we'd have to have k_f-k_0 (the range highlighted
// with 'x' above) to be at least >u INT_MIN.
RangeChecksOut.emplace_back(CurrentChecks.front());
RangeChecksOut.emplace_back(CurrentChecks.back());
}
assert(RangeChecksOut.size() <= OldCount && "We pessimized!");
return RangeChecksOut.size() != OldCount;
}
/// \brief Recursively emit notes for each macro expansion and caret
/// diagnostics where appropriate.
///
/// Walks up the macro expansion stack printing expansion notes, the code
/// snippet, caret, underlines and FixItHint display as appropriate at each
/// level.
///
/// \param Loc The location for this caret.
/// \param Level The diagnostic level currently being emitted.
/// \param Ranges The underlined ranges for this code snippet.
/// \param Hints The FixIt hints active for this diagnostic.
void DiagnosticRenderer::emitMacroExpansions(SourceLocation Loc,
DiagnosticsEngine::Level Level,
ArrayRef<CharSourceRange> Ranges,
ArrayRef<FixItHint> Hints,
const SourceManager &SM) {
assert(Loc.isValid() && "must have a valid source location here");
// Produce a stack of macro backtraces.
SmallVector<SourceLocation, 8> LocationStack;
unsigned IgnoredEnd = 0;
while (Loc.isMacroID()) {
// If this is the expansion of a macro argument, point the caret at the
// use of the argument in the definition of the macro, not the expansion.
if (SM.isMacroArgExpansion(Loc))
LocationStack.push_back(SM.getImmediateExpansionRange(Loc).first);
else
LocationStack.push_back(Loc);
if (checkRangesForMacroArgExpansion(Loc, Ranges, SM))
IgnoredEnd = LocationStack.size();
Loc = SM.getImmediateMacroCallerLoc(Loc);
// Once the location no longer points into a macro, try stepping through
// the last found location. This sometimes produces additional useful
// backtraces.
if (Loc.isFileID())
Loc = SM.getImmediateMacroCallerLoc(LocationStack.back());
assert(Loc.isValid() && "must have a valid source location here");
}
LocationStack.erase(LocationStack.begin(),
LocationStack.begin() + IgnoredEnd);
unsigned MacroDepth = LocationStack.size();
unsigned MacroLimit = DiagOpts->MacroBacktraceLimit;
if (MacroDepth <= MacroLimit || MacroLimit == 0) {
for (auto I = LocationStack.rbegin(), E = LocationStack.rend();
I != E; ++I)
emitSingleMacroExpansion(*I, Level, Ranges, SM);
return;
}
unsigned MacroStartMessages = MacroLimit / 2;
unsigned MacroEndMessages = MacroLimit / 2 + MacroLimit % 2;
for (auto I = LocationStack.rbegin(),
E = LocationStack.rbegin() + MacroStartMessages;
I != E; ++I)
emitSingleMacroExpansion(*I, Level, Ranges, SM);
SmallString<200> MessageStorage;
llvm::raw_svector_ostream Message(MessageStorage);
Message << "(skipping " << (MacroDepth - MacroLimit)
<< " expansions in backtrace; use -fmacro-backtrace-limit=0 to "
"see all)";
emitBasicNote(Message.str());
for (auto I = LocationStack.rend() - MacroEndMessages,
E = LocationStack.rend();
I != E; ++I)
emitSingleMacroExpansion(*I, Level, Ranges, SM);
}
static bool
inlineCallsImpl(CallGraphSCC &SCC, CallGraph &CG,
std::function<AssumptionCache &(Function &)> GetAssumptionCache,
ProfileSummaryInfo *PSI, TargetLibraryInfo &TLI,
bool InsertLifetime,
function_ref<InlineCost(CallSite CS)> GetInlineCost,
function_ref<AAResults &(Function &)> AARGetter,
ImportedFunctionsInliningStatistics &ImportedFunctionsStats) {
SmallPtrSet<Function *, 8> SCCFunctions;
DEBUG(dbgs() << "Inliner visiting SCC:");
for (CallGraphNode *Node : SCC) {
Function *F = Node->getFunction();
if (F)
SCCFunctions.insert(F);
DEBUG(dbgs() << " " << (F ? F->getName() : "INDIRECTNODE"));
}
// Scan through and identify all call sites ahead of time so that we only
// inline call sites in the original functions, not call sites that result
// from inlining other functions.
SmallVector<std::pair<CallSite, int>, 16> CallSites;
// When inlining a callee produces new call sites, we want to keep track of
// the fact that they were inlined from the callee. This allows us to avoid
// infinite inlining in some obscure cases. To represent this, we use an
// index into the InlineHistory vector.
SmallVector<std::pair<Function *, int>, 8> InlineHistory;
for (CallGraphNode *Node : SCC) {
Function *F = Node->getFunction();
if (!F || F->isDeclaration())
continue;
OptimizationRemarkEmitter ORE(F);
for (BasicBlock &BB : *F)
for (Instruction &I : BB) {
CallSite CS(cast<Value>(&I));
// If this isn't a call, or it is a call to an intrinsic, it can
// never be inlined.
if (!CS || isa<IntrinsicInst>(I))
continue;
// If this is a direct call to an external function, we can never inline
// it. If it is an indirect call, inlining may resolve it to be a
// direct call, so we keep it.
if (Function *Callee = CS.getCalledFunction())
if (Callee->isDeclaration()) {
using namespace ore;
ORE.emit(OptimizationRemarkMissed(DEBUG_TYPE, "NoDefinition", &I)
<< NV("Callee", Callee) << " will not be inlined into "
<< NV("Caller", CS.getCaller())
<< " because its definition is unavailable"
<< setIsVerbose());
continue;
}
CallSites.push_back(std::make_pair(CS, -1));
}
}
DEBUG(dbgs() << ": " << CallSites.size() << " call sites.\n");
// If there are no calls in this function, exit early.
if (CallSites.empty())
return false;
// Now that we have all of the call sites, move the ones to functions in the
// current SCC to the end of the list.
unsigned FirstCallInSCC = CallSites.size();
for (unsigned i = 0; i < FirstCallInSCC; ++i)
if (Function *F = CallSites[i].first.getCalledFunction())
if (SCCFunctions.count(F))
std::swap(CallSites[i--], CallSites[--FirstCallInSCC]);
InlinedArrayAllocasTy InlinedArrayAllocas;
InlineFunctionInfo InlineInfo(&CG, &GetAssumptionCache);
// Now that we have all of the call sites, loop over them and inline them if
// it looks profitable to do so.
bool Changed = false;
bool LocalChange;
do {
LocalChange = false;
// Iterate over the outer loop because inlining functions can cause indirect
// calls to become direct calls.
// CallSites may be modified inside so ranged for loop can not be used.
for (unsigned CSi = 0; CSi != CallSites.size(); ++CSi) {
CallSite CS = CallSites[CSi].first;
Function *Caller = CS.getCaller();
Function *Callee = CS.getCalledFunction();
// If this call site is dead and it is to a readonly function, we should
// just delete the call instead of trying to inline it, regardless of
// size. This happens because IPSCCP propagates the result out of the
// call and then we're left with the dead call.
if (isInstructionTriviallyDead(CS.getInstruction(), &TLI)) {
DEBUG(dbgs() << " -> Deleting dead call: " << *CS.getInstruction()
<< "\n");
// Update the call graph by deleting the edge from Callee to Caller.
CG[Caller]->removeCallEdgeFor(CS);
CS.getInstruction()->eraseFromParent();
++NumCallsDeleted;
} else {
// We can only inline direct calls to non-declarations.
if (!Callee || Callee->isDeclaration())
continue;
// If this call site was obtained by inlining another function, verify
// that the include path for the function did not include the callee
// itself. If so, we'd be recursively inlining the same function,
// which would provide the same callsites, which would cause us to
// infinitely inline.
int InlineHistoryID = CallSites[CSi].second;
if (InlineHistoryID != -1 &&
InlineHistoryIncludes(Callee, InlineHistoryID, InlineHistory))
continue;
// Get DebugLoc to report. CS will be invalid after Inliner.
DebugLoc DLoc = CS.getInstruction()->getDebugLoc();
BasicBlock *Block = CS.getParent();
// FIXME for new PM: because of the old PM we currently generate ORE and
// in turn BFI on demand. With the new PM, the ORE dependency should
// just become a regular analysis dependency.
OptimizationRemarkEmitter ORE(Caller);
// If the policy determines that we should inline this function,
// try to do so.
using namespace ore;
if (!shouldInline(CS, GetInlineCost, ORE)) {
ORE.emit(
OptimizationRemarkMissed(DEBUG_TYPE, "NotInlined", DLoc, Block)
<< NV("Callee", Callee) << " will not be inlined into "
<< NV("Caller", Caller));
continue;
}
// Attempt to inline the function.
if (!InlineCallIfPossible(CS, InlineInfo, InlinedArrayAllocas,
InlineHistoryID, InsertLifetime, AARGetter,
ImportedFunctionsStats)) {
ORE.emit(
OptimizationRemarkMissed(DEBUG_TYPE, "NotInlined", DLoc, Block)
<< NV("Callee", Callee) << " will not be inlined into "
<< NV("Caller", Caller));
continue;
}
++NumInlined;
// Report the inline decision.
ORE.emit(OptimizationRemark(DEBUG_TYPE, "Inlined", DLoc, Block)
<< NV("Callee", Callee) << " inlined into "
<< NV("Caller", Caller));
// If inlining this function gave us any new call sites, throw them
// onto our worklist to process. They are useful inline candidates.
if (!InlineInfo.InlinedCalls.empty()) {
// Create a new inline history entry for this, so that we remember
// that these new callsites came about due to inlining Callee.
int NewHistoryID = InlineHistory.size();
InlineHistory.push_back(std::make_pair(Callee, InlineHistoryID));
for (Value *Ptr : InlineInfo.InlinedCalls)
CallSites.push_back(std::make_pair(CallSite(Ptr), NewHistoryID));
}
}
// If we inlined or deleted the last possible call site to the function,
// delete the function body now.
if (Callee && Callee->use_empty() && Callee->hasLocalLinkage() &&
// TODO: Can remove if in SCC now.
!SCCFunctions.count(Callee) &&
// The function may be apparently dead, but if there are indirect
// callgraph references to the node, we cannot delete it yet, this
// could invalidate the CGSCC iterator.
CG[Callee]->getNumReferences() == 0) {
DEBUG(dbgs() << " -> Deleting dead function: " << Callee->getName()
<< "\n");
CallGraphNode *CalleeNode = CG[Callee];
// Remove any call graph edges from the callee to its callees.
CalleeNode->removeAllCalledFunctions();
// Removing the node for callee from the call graph and delete it.
delete CG.removeFunctionFromModule(CalleeNode);
++NumDeleted;
}
// Remove this call site from the list. If possible, use
// swap/pop_back for efficiency, but do not use it if doing so would
// move a call site to a function in this SCC before the
// 'FirstCallInSCC' barrier.
if (SCC.isSingular()) {
CallSites[CSi] = CallSites.back();
CallSites.pop_back();
} else {
CallSites.erase(CallSites.begin() + CSi);
}
--CSi;
Changed = true;
LocalChange = true;
}
} while (LocalChange);
return Changed;
}
// Propagate existing explicit probabilities from either profile data or
// 'expect' intrinsic processing.
bool BranchProbabilityInfo::calcMetadataWeights(const BasicBlock *BB) {
const TerminatorInst *TI = BB->getTerminator();
if (TI->getNumSuccessors() == 1)
return false;
if (!isa<BranchInst>(TI) && !isa<SwitchInst>(TI))
return false;
MDNode *WeightsNode = TI->getMetadata(LLVMContext::MD_prof);
if (!WeightsNode)
return false;
// Check that the number of successors is manageable.
assert(TI->getNumSuccessors() < UINT32_MAX && "Too many successors");
// Ensure there are weights for all of the successors. Note that the first
// operand to the metadata node is a name, not a weight.
if (WeightsNode->getNumOperands() != TI->getNumSuccessors() + 1)
return false;
// Build up the final weights that will be used in a temporary buffer.
// Compute the sum of all weights to later decide whether they need to
// be scaled to fit in 32 bits.
uint64_t WeightSum = 0;
SmallVector<uint32_t, 2> Weights;
Weights.reserve(TI->getNumSuccessors());
for (unsigned i = 1, e = WeightsNode->getNumOperands(); i != e; ++i) {
ConstantInt *Weight =
mdconst::dyn_extract<ConstantInt>(WeightsNode->getOperand(i));
if (!Weight)
return false;
assert(Weight->getValue().getActiveBits() <= 32 &&
"Too many bits for uint32_t");
Weights.push_back(Weight->getZExtValue());
WeightSum += Weights.back();
}
assert(Weights.size() == TI->getNumSuccessors() && "Checked above");
// If the sum of weights does not fit in 32 bits, scale every weight down
// accordingly.
uint64_t ScalingFactor =
(WeightSum > UINT32_MAX) ? WeightSum / UINT32_MAX + 1 : 1;
WeightSum = 0;
for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i) {
Weights[i] /= ScalingFactor;
WeightSum += Weights[i];
}
if (WeightSum == 0) {
for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
setEdgeProbability(BB, i, {1, e});
} else {
for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
setEdgeProbability(BB, i, {Weights[i], static_cast<uint32_t>(WeightSum)});
}
assert(WeightSum <= UINT32_MAX &&
"Expected weights to scale down to 32 bits");
return true;
}
void CGObjCRuntime::EmitTryCatchStmt(CodeGenFunction &CGF,
const ObjCAtTryStmt &S,
llvm::Constant *beginCatchFn,
llvm::Constant *endCatchFn,
llvm::Constant *exceptionRethrowFn) {
// Jump destination for falling out of catch bodies.
CodeGenFunction::JumpDest Cont;
if (S.getNumCatchStmts())
Cont = CGF.getJumpDestInCurrentScope("eh.cont");
CodeGenFunction::FinallyInfo FinallyInfo;
if (const ObjCAtFinallyStmt *Finally = S.getFinallyStmt())
FinallyInfo.enter(CGF, Finally->getFinallyBody(),
beginCatchFn, endCatchFn, exceptionRethrowFn);
SmallVector<CatchHandler, 8> Handlers;
// Enter the catch, if there is one.
if (S.getNumCatchStmts()) {
for (unsigned I = 0, N = S.getNumCatchStmts(); I != N; ++I) {
const ObjCAtCatchStmt *CatchStmt = S.getCatchStmt(I);
const VarDecl *CatchDecl = CatchStmt->getCatchParamDecl();
Handlers.push_back(CatchHandler());
CatchHandler &Handler = Handlers.back();
Handler.Variable = CatchDecl;
Handler.Body = CatchStmt->getCatchBody();
Handler.Block = CGF.createBasicBlock("catch");
// @catch(...) always matches.
if (!CatchDecl) {
Handler.TypeInfo = 0; // catch-all
// Don't consider any other catches.
break;
}
Handler.TypeInfo = GetEHType(CatchDecl->getType());
}
EHCatchScope *Catch = CGF.EHStack.pushCatch(Handlers.size());
for (unsigned I = 0, E = Handlers.size(); I != E; ++I)
Catch->setHandler(I, Handlers[I].TypeInfo, Handlers[I].Block);
}
// Emit the try body.
CGF.EmitStmt(S.getTryBody());
// Leave the try.
if (S.getNumCatchStmts())
CGF.popCatchScope();
// Remember where we were.
CGBuilderTy::InsertPoint SavedIP = CGF.Builder.saveAndClearIP();
// Emit the handlers.
for (unsigned I = 0, E = Handlers.size(); I != E; ++I) {
CatchHandler &Handler = Handlers[I];
CGF.EmitBlock(Handler.Block);
llvm::Value *RawExn = CGF.getExceptionFromSlot();
// Enter the catch.
llvm::Value *Exn = RawExn;
if (beginCatchFn) {
Exn = CGF.Builder.CreateCall(beginCatchFn, RawExn, "exn.adjusted");
cast<llvm::CallInst>(Exn)->setDoesNotThrow();
}
CodeGenFunction::LexicalScope cleanups(CGF, Handler.Body->getSourceRange());
if (endCatchFn) {
// Add a cleanup to leave the catch.
bool EndCatchMightThrow = (Handler.Variable == 0);
CGF.EHStack.pushCleanup<CallObjCEndCatch>(NormalAndEHCleanup,
EndCatchMightThrow,
endCatchFn);
}
// Bind the catch parameter if it exists.
if (const VarDecl *CatchParam = Handler.Variable) {
llvm::Type *CatchType = CGF.ConvertType(CatchParam->getType());
llvm::Value *CastExn = CGF.Builder.CreateBitCast(Exn, CatchType);
CGF.EmitAutoVarDecl(*CatchParam);
llvm::Value *CatchParamAddr = CGF.GetAddrOfLocalVar(CatchParam);
switch (CatchParam->getType().getQualifiers().getObjCLifetime()) {
case Qualifiers::OCL_Strong:
CastExn = CGF.EmitARCRetainNonBlock(CastExn);
// fallthrough
case Qualifiers::OCL_None:
case Qualifiers::OCL_ExplicitNone:
case Qualifiers::OCL_Autoreleasing:
CGF.Builder.CreateStore(CastExn, CatchParamAddr);
break;
case Qualifiers::OCL_Weak:
CGF.EmitARCInitWeak(CatchParamAddr, CastExn);
break;
}
}
CGF.ObjCEHValueStack.push_back(Exn);
CGF.EmitStmt(Handler.Body);
CGF.ObjCEHValueStack.pop_back();
// Leave any cleanups associated with the catch.
cleanups.ForceCleanup();
CGF.EmitBranchThroughCleanup(Cont);
}
// Go back to the try-statement fallthrough.
CGF.Builder.restoreIP(SavedIP);
// Pop out of the finally.
if (S.getFinallyStmt())
FinallyInfo.exit(CGF);
if (Cont.isValid())
CGF.EmitBlock(Cont.getBlock());
}
/// shrinkToUses - After removing some uses of a register, shrink its live
/// range to just the remaining uses. This method does not compute reaching
/// defs for new uses, and it doesn't remove dead defs.
bool LiveIntervals::shrinkToUses(LiveInterval *li,
SmallVectorImpl<MachineInstr*> *dead) {
DEBUG(dbgs() << "Shrink: " << *li << '\n');
assert(TargetRegisterInfo::isVirtualRegister(li->reg)
&& "Can only shrink virtual registers");
// Find all the values used, including PHI kills.
SmallVector<std::pair<SlotIndex, VNInfo*>, 16> WorkList;
// Blocks that have already been added to WorkList as live-out.
SmallPtrSet<MachineBasicBlock*, 16> LiveOut;
// Visit all instructions reading li->reg.
for (MachineRegisterInfo::reg_iterator I = MRI->reg_begin(li->reg);
MachineInstr *UseMI = I.skipInstruction();) {
if (UseMI->isDebugValue() || !UseMI->readsVirtualRegister(li->reg))
continue;
SlotIndex Idx = getInstructionIndex(UseMI).getRegSlot();
LiveQueryResult LRQ = li->Query(Idx);
VNInfo *VNI = LRQ.valueIn();
if (!VNI) {
// This shouldn't happen: readsVirtualRegister returns true, but there is
// no live value. It is likely caused by a target getting <undef> flags
// wrong.
DEBUG(dbgs() << Idx << '\t' << *UseMI
<< "Warning: Instr claims to read non-existent value in "
<< *li << '\n');
continue;
}
// Special case: An early-clobber tied operand reads and writes the
// register one slot early.
if (VNInfo *DefVNI = LRQ.valueDefined())
Idx = DefVNI->def;
WorkList.push_back(std::make_pair(Idx, VNI));
}
// Create new live ranges with only minimal live segments per def.
LiveRange NewLR;
for (LiveInterval::vni_iterator I = li->vni_begin(), E = li->vni_end();
I != E; ++I) {
VNInfo *VNI = *I;
if (VNI->isUnused())
continue;
NewLR.addSegment(LiveRange::Segment(VNI->def, VNI->def.getDeadSlot(), VNI));
}
// Keep track of the PHIs that are in use.
SmallPtrSet<VNInfo*, 8> UsedPHIs;
// Extend intervals to reach all uses in WorkList.
while (!WorkList.empty()) {
SlotIndex Idx = WorkList.back().first;
VNInfo *VNI = WorkList.back().second;
WorkList.pop_back();
const MachineBasicBlock *MBB = getMBBFromIndex(Idx.getPrevSlot());
SlotIndex BlockStart = getMBBStartIdx(MBB);
// Extend the live range for VNI to be live at Idx.
if (VNInfo *ExtVNI = NewLR.extendInBlock(BlockStart, Idx)) {
(void)ExtVNI;
assert(ExtVNI == VNI && "Unexpected existing value number");
// Is this a PHIDef we haven't seen before?
if (!VNI->isPHIDef() || VNI->def != BlockStart || !UsedPHIs.insert(VNI))
continue;
// The PHI is live, make sure the predecessors are live-out.
for (MachineBasicBlock::const_pred_iterator PI = MBB->pred_begin(),
PE = MBB->pred_end(); PI != PE; ++PI) {
if (!LiveOut.insert(*PI))
continue;
SlotIndex Stop = getMBBEndIdx(*PI);
// A predecessor is not required to have a live-out value for a PHI.
if (VNInfo *PVNI = li->getVNInfoBefore(Stop))
WorkList.push_back(std::make_pair(Stop, PVNI));
}
continue;
}
// VNI is live-in to MBB.
DEBUG(dbgs() << " live-in at " << BlockStart << '\n');
NewLR.addSegment(LiveRange::Segment(BlockStart, Idx, VNI));
// Make sure VNI is live-out from the predecessors.
for (MachineBasicBlock::const_pred_iterator PI = MBB->pred_begin(),
PE = MBB->pred_end(); PI != PE; ++PI) {
if (!LiveOut.insert(*PI))
continue;
SlotIndex Stop = getMBBEndIdx(*PI);
assert(li->getVNInfoBefore(Stop) == VNI &&
"Wrong value out of predecessor");
WorkList.push_back(std::make_pair(Stop, VNI));
}
}
// Handle dead values.
bool CanSeparate = false;
for (LiveInterval::vni_iterator I = li->vni_begin(), E = li->vni_end();
I != E; ++I) {
VNInfo *VNI = *I;
if (VNI->isUnused())
continue;
LiveRange::iterator LRI = NewLR.FindSegmentContaining(VNI->def);
assert(LRI != NewLR.end() && "Missing segment for PHI");
if (LRI->end != VNI->def.getDeadSlot())
continue;
if (VNI->isPHIDef()) {
// This is a dead PHI. Remove it.
VNI->markUnused();
NewLR.removeSegment(LRI->start, LRI->end);
DEBUG(dbgs() << "Dead PHI at " << VNI->def << " may separate interval\n");
CanSeparate = true;
} else {
// This is a dead def. Make sure the instruction knows.
MachineInstr *MI = getInstructionFromIndex(VNI->def);
assert(MI && "No instruction defining live value");
MI->addRegisterDead(li->reg, TRI);
if (dead && MI->allDefsAreDead()) {
DEBUG(dbgs() << "All defs dead: " << VNI->def << '\t' << *MI);
dead->push_back(MI);
}
}
}
// Move the trimmed segments back.
li->segments.swap(NewLR.segments);
DEBUG(dbgs() << "Shrunk: " << *li << '\n');
return CanSeparate;
}
void AArch64FrameLowering::processFunctionBeforeCalleeSavedScan(
MachineFunction &MF, RegScavenger *RS) const {
const AArch64RegisterInfo *RegInfo = static_cast<const AArch64RegisterInfo *>(
MF.getSubtarget().getRegisterInfo());
AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>();
MachineRegisterInfo *MRI = &MF.getRegInfo();
SmallVector<unsigned, 4> UnspilledCSGPRs;
SmallVector<unsigned, 4> UnspilledCSFPRs;
// The frame record needs to be created by saving the appropriate registers
if (hasFP(MF)) {
MRI->setPhysRegUsed(AArch64::FP);
MRI->setPhysRegUsed(AArch64::LR);
}
// Spill the BasePtr if it's used. Do this first thing so that the
// getCalleeSavedRegs() below will get the right answer.
if (RegInfo->hasBasePointer(MF))
MRI->setPhysRegUsed(RegInfo->getBaseRegister());
// If any callee-saved registers are used, the frame cannot be eliminated.
unsigned NumGPRSpilled = 0;
unsigned NumFPRSpilled = 0;
bool ExtraCSSpill = false;
bool CanEliminateFrame = true;
DEBUG(dbgs() << "*** processFunctionBeforeCalleeSavedScan\nUsed CSRs:");
const MCPhysReg *CSRegs = RegInfo->getCalleeSavedRegs(&MF);
// Check pairs of consecutive callee-saved registers.
for (unsigned i = 0; CSRegs[i]; i += 2) {
assert(CSRegs[i + 1] && "Odd number of callee-saved registers!");
const unsigned OddReg = CSRegs[i];
const unsigned EvenReg = CSRegs[i + 1];
assert((AArch64::GPR64RegClass.contains(OddReg) &&
AArch64::GPR64RegClass.contains(EvenReg)) ^
(AArch64::FPR64RegClass.contains(OddReg) &&
AArch64::FPR64RegClass.contains(EvenReg)) &&
"Register class mismatch!");
const bool OddRegUsed = MRI->isPhysRegUsed(OddReg);
const bool EvenRegUsed = MRI->isPhysRegUsed(EvenReg);
// Early exit if none of the registers in the register pair is actually
// used.
if (!OddRegUsed && !EvenRegUsed) {
if (AArch64::GPR64RegClass.contains(OddReg)) {
UnspilledCSGPRs.push_back(OddReg);
UnspilledCSGPRs.push_back(EvenReg);
} else {
UnspilledCSFPRs.push_back(OddReg);
UnspilledCSFPRs.push_back(EvenReg);
}
continue;
}
unsigned Reg = AArch64::NoRegister;
// If only one of the registers of the register pair is used, make sure to
// mark the other one as used as well.
if (OddRegUsed ^ EvenRegUsed) {
// Find out which register is the additional spill.
Reg = OddRegUsed ? EvenReg : OddReg;
MRI->setPhysRegUsed(Reg);
}
DEBUG(dbgs() << ' ' << PrintReg(OddReg, RegInfo));
DEBUG(dbgs() << ' ' << PrintReg(EvenReg, RegInfo));
assert(((OddReg == AArch64::LR && EvenReg == AArch64::FP) ||
(RegInfo->getEncodingValue(OddReg) + 1 ==
RegInfo->getEncodingValue(EvenReg))) &&
"Register pair of non-adjacent registers!");
if (AArch64::GPR64RegClass.contains(OddReg)) {
NumGPRSpilled += 2;
// If it's not a reserved register, we can use it in lieu of an
// emergency spill slot for the register scavenger.
// FIXME: It would be better to instead keep looking and choose another
// unspilled register that isn't reserved, if there is one.
if (Reg != AArch64::NoRegister && !RegInfo->isReservedReg(MF, Reg))
ExtraCSSpill = true;
} else
NumFPRSpilled += 2;
CanEliminateFrame = false;
}
// FIXME: Set BigStack if any stack slot references may be out of range.
// For now, just conservatively guestimate based on unscaled indexing
// range. We'll end up allocating an unnecessary spill slot a lot, but
// realistically that's not a big deal at this stage of the game.
// The CSR spill slots have not been allocated yet, so estimateStackSize
// won't include them.
MachineFrameInfo *MFI = MF.getFrameInfo();
unsigned CFSize = estimateStackSize(MF) + 8 * (NumGPRSpilled + NumFPRSpilled);
DEBUG(dbgs() << "Estimated stack frame size: " << CFSize << " bytes.\n");
bool BigStack = (CFSize >= 256);
if (BigStack || !CanEliminateFrame || RegInfo->cannotEliminateFrame(MF))
AFI->setHasStackFrame(true);
// Estimate if we might need to scavenge a register at some point in order
// to materialize a stack offset. If so, either spill one additional
// callee-saved register or reserve a special spill slot to facilitate
// register scavenging. If we already spilled an extra callee-saved register
// above to keep the number of spills even, we don't need to do anything else
// here.
if (BigStack && !ExtraCSSpill) {
// If we're adding a register to spill here, we have to add two of them
// to keep the number of regs to spill even.
assert(((UnspilledCSGPRs.size() & 1) == 0) && "Odd number of registers!");
unsigned Count = 0;
while (!UnspilledCSGPRs.empty() && Count < 2) {
unsigned Reg = UnspilledCSGPRs.back();
UnspilledCSGPRs.pop_back();
DEBUG(dbgs() << "Spilling " << PrintReg(Reg, RegInfo)
<< " to get a scratch register.\n");
MRI->setPhysRegUsed(Reg);
ExtraCSSpill = true;
++Count;
}
// If we didn't find an extra callee-saved register to spill, create
// an emergency spill slot.
if (!ExtraCSSpill) {
const TargetRegisterClass *RC = &AArch64::GPR64RegClass;
int FI = MFI->CreateStackObject(RC->getSize(), RC->getAlignment(), false);
RS->addScavengingFrameIndex(FI);
DEBUG(dbgs() << "No available CS registers, allocated fi#" << FI
<< " as the emergency spill slot.\n");
}
}
}
/// SplitCriticalEdge - If this edge is a critical edge, insert a new node to
/// split the critical edge. This will update DominatorTree and
/// DominatorFrontier information if it is available, thus calling this pass
/// will not invalidate either of them. This returns the new block if the edge
/// was split, null otherwise.
///
/// If MergeIdenticalEdges is true (not the default), *all* edges from TI to the
/// specified successor will be merged into the same critical edge block.
/// This is most commonly interesting with switch instructions, which may
/// have many edges to any one destination. This ensures that all edges to that
/// dest go to one block instead of each going to a different block, but isn't
/// the standard definition of a "critical edge".
///
/// It is invalid to call this function on a critical edge that starts at an
/// IndirectBrInst. Splitting these edges will almost always create an invalid
/// program because the address of the new block won't be the one that is jumped
/// to.
///
BasicBlock *llvm::SplitCriticalEdge(TerminatorInst *TI, unsigned SuccNum,
Pass *P, bool MergeIdenticalEdges) {
if (!isCriticalEdge(TI, SuccNum, MergeIdenticalEdges)) return 0;
assert(!isa<IndirectBrInst>(TI) &&
"Cannot split critical edge from IndirectBrInst");
BasicBlock *TIBB = TI->getParent();
BasicBlock *DestBB = TI->getSuccessor(SuccNum);
// Create a new basic block, linking it into the CFG.
BasicBlock *NewBB = BasicBlock::Create(TI->getContext(),
TIBB->getName() + "." + DestBB->getName() + "_crit_edge");
// Create our unconditional branch.
BranchInst::Create(DestBB, NewBB);
// Branch to the new block, breaking the edge.
TI->setSuccessor(SuccNum, NewBB);
// Insert the block into the function... right after the block TI lives in.
Function &F = *TIBB->getParent();
Function::iterator FBBI = TIBB;
F.getBasicBlockList().insert(++FBBI, NewBB);
// If there are any PHI nodes in DestBB, we need to update them so that they
// merge incoming values from NewBB instead of from TIBB.
if (PHINode *APHI = dyn_cast<PHINode>(DestBB->begin())) {
// This conceptually does:
// foreach (PHINode *PN in DestBB)
// PN->setIncomingBlock(PN->getIncomingBlock(TIBB), NewBB);
// but is optimized for two cases.
if (APHI->getNumIncomingValues() <= 8) { // Small # preds case.
unsigned BBIdx = 0;
for (BasicBlock::iterator I = DestBB->begin(); isa<PHINode>(I); ++I) {
// We no longer enter through TIBB, now we come in through NewBB.
// Revector exactly one entry in the PHI node that used to come from
// TIBB to come from NewBB.
PHINode *PN = cast<PHINode>(I);
// Reuse the previous value of BBIdx if it lines up. In cases where we
// have multiple phi nodes with *lots* of predecessors, this is a speed
// win because we don't have to scan the PHI looking for TIBB. This
// happens because the BB list of PHI nodes are usually in the same
// order.
if (PN->getIncomingBlock(BBIdx) != TIBB)
BBIdx = PN->getBasicBlockIndex(TIBB);
PN->setIncomingBlock(BBIdx, NewBB);
}
} else {
// However, the foreach loop is slow for blocks with lots of predecessors
// because PHINode::getIncomingBlock is O(n) in # preds. Instead, walk
// the user list of TIBB to find the PHI nodes.
SmallPtrSet<PHINode*, 16> UpdatedPHIs;
for (Value::use_iterator UI = TIBB->use_begin(), E = TIBB->use_end();
UI != E; ) {
Value::use_iterator Use = UI++;
if (PHINode *PN = dyn_cast<PHINode>(*Use)) {
// Remove one entry from each PHI.
if (PN->getParent() == DestBB && UpdatedPHIs.insert(PN))
PN->setOperand(Use.getOperandNo(), NewBB);
}
}
}
}
// If there are any other edges from TIBB to DestBB, update those to go
// through the split block, making those edges non-critical as well (and
// reducing the number of phi entries in the DestBB if relevant).
if (MergeIdenticalEdges) {
for (unsigned i = SuccNum+1, e = TI->getNumSuccessors(); i != e; ++i) {
if (TI->getSuccessor(i) != DestBB) continue;
// Remove an entry for TIBB from DestBB phi nodes.
DestBB->removePredecessor(TIBB);
// We found another edge to DestBB, go to NewBB instead.
TI->setSuccessor(i, NewBB);
}
}
// If we don't have a pass object, we can't update anything...
if (P == 0) return NewBB;
DominatorTree *DT = P->getAnalysisIfAvailable<DominatorTree>();
DominanceFrontier *DF = P->getAnalysisIfAvailable<DominanceFrontier>();
LoopInfo *LI = P->getAnalysisIfAvailable<LoopInfo>();
ProfileInfo *PI = P->getAnalysisIfAvailable<ProfileInfo>();
// If we have nothing to update, just return.
if (DT == 0 && DF == 0 && LI == 0 && PI == 0)
return NewBB;
// Now update analysis information. Since the only predecessor of NewBB is
// the TIBB, TIBB clearly dominates NewBB. TIBB usually doesn't dominate
// anything, as there are other successors of DestBB. However, if all other
// predecessors of DestBB are already dominated by DestBB (e.g. DestBB is a
// loop header) then NewBB dominates DestBB.
SmallVector<BasicBlock*, 8> OtherPreds;
// If there is a PHI in the block, loop over predecessors with it, which is
// faster than iterating pred_begin/end.
if (PHINode *PN = dyn_cast<PHINode>(DestBB->begin())) {
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
if (PN->getIncomingBlock(i) != NewBB)
OtherPreds.push_back(PN->getIncomingBlock(i));
} else {
for (pred_iterator I = pred_begin(DestBB), E = pred_end(DestBB);
I != E; ++I) {
BasicBlock *P = *I;
if (P != NewBB)
OtherPreds.push_back(P);
}
}
bool NewBBDominatesDestBB = true;
// Should we update DominatorTree information?
if (DT) {
DomTreeNode *TINode = DT->getNode(TIBB);
// The new block is not the immediate dominator for any other nodes, but
// TINode is the immediate dominator for the new node.
//
if (TINode) { // Don't break unreachable code!
DomTreeNode *NewBBNode = DT->addNewBlock(NewBB, TIBB);
DomTreeNode *DestBBNode = 0;
// If NewBBDominatesDestBB hasn't been computed yet, do so with DT.
if (!OtherPreds.empty()) {
DestBBNode = DT->getNode(DestBB);
while (!OtherPreds.empty() && NewBBDominatesDestBB) {
if (DomTreeNode *OPNode = DT->getNode(OtherPreds.back()))
NewBBDominatesDestBB = DT->dominates(DestBBNode, OPNode);
OtherPreds.pop_back();
}
OtherPreds.clear();
}
// If NewBBDominatesDestBB, then NewBB dominates DestBB, otherwise it
// doesn't dominate anything.
if (NewBBDominatesDestBB) {
if (!DestBBNode) DestBBNode = DT->getNode(DestBB);
DT->changeImmediateDominator(DestBBNode, NewBBNode);
}
}
}
// Should we update DominanceFrontier information?
if (DF) {
// If NewBBDominatesDestBB hasn't been computed yet, do so with DF.
if (!OtherPreds.empty()) {
// FIXME: IMPLEMENT THIS!
llvm_unreachable("Requiring domfrontiers but not idom/domtree/domset."
" not implemented yet!");
}
// Since the new block is dominated by its only predecessor TIBB,
// it cannot be in any block's dominance frontier. If NewBB dominates
// DestBB, its dominance frontier is the same as DestBB's, otherwise it is
// just {DestBB}.
DominanceFrontier::DomSetType NewDFSet;
if (NewBBDominatesDestBB) {
DominanceFrontier::iterator I = DF->find(DestBB);
if (I != DF->end()) {
DF->addBasicBlock(NewBB, I->second);
if (I->second.count(DestBB)) {
// However NewBB's frontier does not include DestBB.
DominanceFrontier::iterator NF = DF->find(NewBB);
DF->removeFromFrontier(NF, DestBB);
}
}
else
DF->addBasicBlock(NewBB, DominanceFrontier::DomSetType());
} else {
DominanceFrontier::DomSetType NewDFSet;
NewDFSet.insert(DestBB);
DF->addBasicBlock(NewBB, NewDFSet);
}
}
// Update LoopInfo if it is around.
if (LI) {
if (Loop *TIL = LI->getLoopFor(TIBB)) {
// If one or the other blocks were not in a loop, the new block is not
// either, and thus LI doesn't need to be updated.
if (Loop *DestLoop = LI->getLoopFor(DestBB)) {
if (TIL == DestLoop) {
// Both in the same loop, the NewBB joins loop.
DestLoop->addBasicBlockToLoop(NewBB, LI->getBase());
} else if (TIL->contains(DestLoop)) {
// Edge from an outer loop to an inner loop. Add to the outer loop.
TIL->addBasicBlockToLoop(NewBB, LI->getBase());
} else if (DestLoop->contains(TIL)) {
// Edge from an inner loop to an outer loop. Add to the outer loop.
DestLoop->addBasicBlockToLoop(NewBB, LI->getBase());
} else {
// Edge from two loops with no containment relation. Because these
// are natural loops, we know that the destination block must be the
// header of its loop (adding a branch into a loop elsewhere would
// create an irreducible loop).
assert(DestLoop->getHeader() == DestBB &&
"Should not create irreducible loops!");
if (Loop *P = DestLoop->getParentLoop())
P->addBasicBlockToLoop(NewBB, LI->getBase());
}
}
// If TIBB is in a loop and DestBB is outside of that loop, split the
// other exit blocks of the loop that also have predecessors outside
// the loop, to maintain a LoopSimplify guarantee.
if (!TIL->contains(DestBB) &&
P->mustPreserveAnalysisID(LoopSimplifyID)) {
assert(!TIL->contains(NewBB) &&
"Split point for loop exit is contained in loop!");
// Update LCSSA form in the newly created exit block.
if (P->mustPreserveAnalysisID(LCSSAID)) {
SmallVector<BasicBlock *, 1> OrigPred;
OrigPred.push_back(TIBB);
CreatePHIsForSplitLoopExit(OrigPred, NewBB, DestBB);
}
// For each unique exit block...
SmallVector<BasicBlock *, 4> ExitBlocks;
TIL->getExitBlocks(ExitBlocks);
for (unsigned i = 0, e = ExitBlocks.size(); i != e; ++i) {
// Collect all the preds that are inside the loop, and note
// whether there are any preds outside the loop.
SmallVector<BasicBlock *, 4> Preds;
bool HasPredOutsideOfLoop = false;
BasicBlock *Exit = ExitBlocks[i];
for (pred_iterator I = pred_begin(Exit), E = pred_end(Exit);
I != E; ++I) {
BasicBlock *P = *I;
if (TIL->contains(P))
Preds.push_back(P);
else
HasPredOutsideOfLoop = true;
}
// If there are any preds not in the loop, we'll need to split
// the edges. The Preds.empty() check is needed because a block
// may appear multiple times in the list. We can't use
// getUniqueExitBlocks above because that depends on LoopSimplify
// form, which we're in the process of restoring!
if (!Preds.empty() && HasPredOutsideOfLoop) {
BasicBlock *NewExitBB =
SplitBlockPredecessors(Exit, Preds.data(), Preds.size(),
"split", P);
if (P->mustPreserveAnalysisID(LCSSAID))
CreatePHIsForSplitLoopExit(Preds, NewExitBB, Exit);
}
}
}
// LCSSA form was updated above for the case where LoopSimplify is
// available, which means that all predecessors of loop exit blocks
// are within the loop. Without LoopSimplify form, it would be
// necessary to insert a new phi.
assert((!P->mustPreserveAnalysisID(LCSSAID) ||
P->mustPreserveAnalysisID(LoopSimplifyID)) &&
"SplitCriticalEdge doesn't know how to update LCCSA form "
"without LoopSimplify!");
}
}
// Update ProfileInfo if it is around.
if (PI)
PI->splitEdge(TIBB, DestBB, NewBB, MergeIdenticalEdges);
return NewBB;
}
/// EmitSpecialNode - Generate machine code for a target-independent node and
/// needed dependencies.
void InstrEmitter::
EmitSpecialNode(SDNode *Node, bool IsClone, bool IsCloned,
DenseMap<SDValue, unsigned> &VRBaseMap) {
switch (Node->getOpcode()) {
default:
#ifndef NDEBUG
Node->dump();
#endif
llvm_unreachable("This target-independent node should have been selected!");
case ISD::EntryToken:
llvm_unreachable("EntryToken should have been excluded from the schedule!");
case ISD::MERGE_VALUES:
case ISD::TokenFactor: // fall thru
break;
case ISD::CopyToReg: {
unsigned SrcReg;
SDValue SrcVal = Node->getOperand(2);
if (RegisterSDNode *R = dyn_cast<RegisterSDNode>(SrcVal))
SrcReg = R->getReg();
else
SrcReg = getVR(SrcVal, VRBaseMap);
unsigned DestReg = cast<RegisterSDNode>(Node->getOperand(1))->getReg();
if (SrcReg == DestReg) // Coalesced away the copy? Ignore.
break;
BuildMI(*MBB, InsertPos, Node->getDebugLoc(), TII->get(TargetOpcode::COPY),
DestReg).addReg(SrcReg);
break;
}
case ISD::CopyFromReg: {
unsigned SrcReg = cast<RegisterSDNode>(Node->getOperand(1))->getReg();
EmitCopyFromReg(Node, 0, IsClone, IsCloned, SrcReg, VRBaseMap);
break;
}
case ISD::EH_LABEL:
case ISD::ANNOTATION_LABEL: {
unsigned Opc = (Node->getOpcode() == ISD::EH_LABEL)
? TargetOpcode::EH_LABEL
: TargetOpcode::ANNOTATION_LABEL;
MCSymbol *S = cast<LabelSDNode>(Node)->getLabel();
BuildMI(*MBB, InsertPos, Node->getDebugLoc(),
TII->get(Opc)).addSym(S);
break;
}
case ISD::LIFETIME_START:
case ISD::LIFETIME_END: {
unsigned TarOp = (Node->getOpcode() == ISD::LIFETIME_START) ?
TargetOpcode::LIFETIME_START : TargetOpcode::LIFETIME_END;
FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(Node->getOperand(1));
BuildMI(*MBB, InsertPos, Node->getDebugLoc(), TII->get(TarOp))
.addFrameIndex(FI->getIndex());
break;
}
case ISD::INLINEASM: {
unsigned NumOps = Node->getNumOperands();
if (Node->getOperand(NumOps-1).getValueType() == MVT::Glue)
--NumOps; // Ignore the glue operand.
// Create the inline asm machine instruction.
MachineInstrBuilder MIB = BuildMI(*MF, Node->getDebugLoc(),
TII->get(TargetOpcode::INLINEASM));
// Add the asm string as an external symbol operand.
SDValue AsmStrV = Node->getOperand(InlineAsm::Op_AsmString);
const char *AsmStr = cast<ExternalSymbolSDNode>(AsmStrV)->getSymbol();
MIB.addExternalSymbol(AsmStr);
// Add the HasSideEffect, isAlignStack, AsmDialect, MayLoad and MayStore
// bits.
int64_t ExtraInfo =
cast<ConstantSDNode>(Node->getOperand(InlineAsm::Op_ExtraInfo))->
getZExtValue();
MIB.addImm(ExtraInfo);
// Remember to operand index of the group flags.
SmallVector<unsigned, 8> GroupIdx;
// Remember registers that are part of early-clobber defs.
SmallVector<unsigned, 8> ECRegs;
// Add all of the operand registers to the instruction.
for (unsigned i = InlineAsm::Op_FirstOperand; i != NumOps;) {
unsigned Flags =
cast<ConstantSDNode>(Node->getOperand(i))->getZExtValue();
const unsigned NumVals = InlineAsm::getNumOperandRegisters(Flags);
GroupIdx.push_back(MIB->getNumOperands());
MIB.addImm(Flags);
++i; // Skip the ID value.
switch (InlineAsm::getKind(Flags)) {
default: llvm_unreachable("Bad flags!");
case InlineAsm::Kind_RegDef:
for (unsigned j = 0; j != NumVals; ++j, ++i) {
unsigned Reg = cast<RegisterSDNode>(Node->getOperand(i))->getReg();
// FIXME: Add dead flags for physical and virtual registers defined.
// For now, mark physical register defs as implicit to help fast
// regalloc. This makes inline asm look a lot like calls.
MIB.addReg(Reg, RegState::Define |
getImplRegState(TargetRegisterInfo::isPhysicalRegister(Reg)));
}
break;
case InlineAsm::Kind_RegDefEarlyClobber:
case InlineAsm::Kind_Clobber:
for (unsigned j = 0; j != NumVals; ++j, ++i) {
unsigned Reg = cast<RegisterSDNode>(Node->getOperand(i))->getReg();
MIB.addReg(Reg, RegState::Define | RegState::EarlyClobber |
getImplRegState(TargetRegisterInfo::isPhysicalRegister(Reg)));
ECRegs.push_back(Reg);
}
break;
case InlineAsm::Kind_RegUse: // Use of register.
case InlineAsm::Kind_Imm: // Immediate.
case InlineAsm::Kind_Mem: // Addressing mode.
// The addressing mode has been selected, just add all of the
// operands to the machine instruction.
for (unsigned j = 0; j != NumVals; ++j, ++i)
AddOperand(MIB, Node->getOperand(i), 0, nullptr, VRBaseMap,
/*IsDebug=*/false, IsClone, IsCloned);
// Manually set isTied bits.
if (InlineAsm::getKind(Flags) == InlineAsm::Kind_RegUse) {
unsigned DefGroup = 0;
if (InlineAsm::isUseOperandTiedToDef(Flags, DefGroup)) {
unsigned DefIdx = GroupIdx[DefGroup] + 1;
unsigned UseIdx = GroupIdx.back() + 1;
for (unsigned j = 0; j != NumVals; ++j)
MIB->tieOperands(DefIdx + j, UseIdx + j);
}
}
break;
}
}
// GCC inline assembly allows input operands to also be early-clobber
// output operands (so long as the operand is written only after it's
// used), but this does not match the semantics of our early-clobber flag.
// If an early-clobber operand register is also an input operand register,
// then remove the early-clobber flag.
for (unsigned Reg : ECRegs) {
if (MIB->readsRegister(Reg, TRI)) {
MachineOperand *MO = MIB->findRegisterDefOperand(Reg, false, TRI);
assert(MO && "No def operand for clobbered register?");
MO->setIsEarlyClobber(false);
}
}
// Get the mdnode from the asm if it exists and add it to the instruction.
SDValue MDV = Node->getOperand(InlineAsm::Op_MDNode);
const MDNode *MD = cast<MDNodeSDNode>(MDV)->getMD();
if (MD)
MIB.addMetadata(MD);
MBB->insert(InsertPos, MIB);
break;
}
}
}
/// lowerAcrossUnwindEdges - Find all variables which are alive across an unwind
/// edge and spill them.
void SjLjEHPrepare::lowerAcrossUnwindEdges(Function &F,
ArrayRef<InvokeInst *> Invokes) {
// Finally, scan the code looking for instructions with bad live ranges.
for (Function::iterator BB = F.begin(), BBE = F.end(); BB != BBE; ++BB) {
for (BasicBlock::iterator II = BB->begin(), IIE = BB->end(); II != IIE;
++II) {
// Ignore obvious cases we don't have to handle. In particular, most
// instructions either have no uses or only have a single use inside the
// current block. Ignore them quickly.
Instruction *Inst = II;
if (Inst->use_empty())
continue;
if (Inst->hasOneUse() &&
cast<Instruction>(Inst->use_back())->getParent() == BB &&
!isa<PHINode>(Inst->use_back()))
continue;
// If this is an alloca in the entry block, it's not a real register
// value.
if (AllocaInst *AI = dyn_cast<AllocaInst>(Inst))
if (isa<ConstantInt>(AI->getArraySize()) && BB == F.begin())
continue;
// Avoid iterator invalidation by copying users to a temporary vector.
SmallVector<Instruction *, 16> Users;
for (Value::use_iterator UI = Inst->use_begin(), E = Inst->use_end();
UI != E; ++UI) {
Instruction *User = cast<Instruction>(*UI);
if (User->getParent() != BB || isa<PHINode>(User))
Users.push_back(User);
}
// Find all of the blocks that this value is live in.
SmallPtrSet<BasicBlock *, 64> LiveBBs;
LiveBBs.insert(Inst->getParent());
while (!Users.empty()) {
Instruction *U = Users.back();
Users.pop_back();
if (!isa<PHINode>(U)) {
MarkBlocksLiveIn(U->getParent(), LiveBBs);
} else {
// Uses for a PHI node occur in their predecessor block.
PHINode *PN = cast<PHINode>(U);
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
if (PN->getIncomingValue(i) == Inst)
MarkBlocksLiveIn(PN->getIncomingBlock(i), LiveBBs);
}
}
// Now that we know all of the blocks that this thing is live in, see if
// it includes any of the unwind locations.
bool NeedsSpill = false;
for (unsigned i = 0, e = Invokes.size(); i != e; ++i) {
BasicBlock *UnwindBlock = Invokes[i]->getUnwindDest();
if (UnwindBlock != BB && LiveBBs.count(UnwindBlock)) {
DEBUG(dbgs() << "SJLJ Spill: " << *Inst << " around "
<< UnwindBlock->getName() << "\n");
NeedsSpill = true;
break;
}
}
// If we decided we need a spill, do it.
// FIXME: Spilling this way is overkill, as it forces all uses of
// the value to be reloaded from the stack slot, even those that aren't
// in the unwind blocks. We should be more selective.
if (NeedsSpill) {
DemoteRegToStack(*Inst, true);
++NumSpilled;
}
}
}
// Go through the landing pads and remove any PHIs there.
for (unsigned i = 0, e = Invokes.size(); i != e; ++i) {
BasicBlock *UnwindBlock = Invokes[i]->getUnwindDest();
LandingPadInst *LPI = UnwindBlock->getLandingPadInst();
// Place PHIs into a set to avoid invalidating the iterator.
SmallPtrSet<PHINode *, 8> PHIsToDemote;
for (BasicBlock::iterator PN = UnwindBlock->begin(); isa<PHINode>(PN); ++PN)
PHIsToDemote.insert(cast<PHINode>(PN));
if (PHIsToDemote.empty())
continue;
// Demote the PHIs to the stack.
for (SmallPtrSet<PHINode *, 8>::iterator I = PHIsToDemote.begin(),
E = PHIsToDemote.end();
I != E; ++I)
DemotePHIToStack(*I);
// Move the landingpad instruction back to the top of the landing pad block.
LPI->moveBefore(UnwindBlock->begin());
}
}
