/// \brief Checks if the padding bytes of an argument could be accessed.
bool ArgPromotion::canPaddingBeAccessed(Argument *arg) {
assert(arg->hasByValAttr());
// Track all the pointers to the argument to make sure they are not captured.
SmallPtrSet<Value *, 16> PtrValues;
PtrValues.insert(arg);
// Track all of the stores.
SmallVector<StoreInst *, 16> Stores;
// Scan through the uses recursively to make sure the pointer is always used
// sanely.
SmallVector<Value *, 16> WorkList;
WorkList.insert(WorkList.end(), arg->user_begin(), arg->user_end());
while (!WorkList.empty()) {
Value *V = WorkList.back();
WorkList.pop_back();
if (isa<GetElementPtrInst>(V) || isa<PHINode>(V)) {
if (PtrValues.insert(V))
WorkList.insert(WorkList.end(), V->user_begin(), V->user_end());
} else if (StoreInst *Store = dyn_cast<StoreInst>(V)) {
Stores.push_back(Store);
} else if (!isa<LoadInst>(V)) {
return true;
}
}
// Check to make sure the pointers aren't captured
for (StoreInst *Store : Stores)
if (PtrValues.count(Store->getValueOperand()))
return true;
return false;
}
// Get the value we should change this callsite to call instead.
Value *CSDataRando::getCloneCalledValue(CallSite CS, FuncInfo &CalleeInfo) {
if (CalleeInfo.ArgNodes.size() == 0) {
return nullptr;
}
// Find the function type we want based on how many args need to be added. We
// do this in case the original function has been cast to a different type.
FunctionType *FT = CS.getFunctionType();
SmallVector<Type*, 8> Params;
Params.insert(Params.end(), FT->param_begin(), FT->param_end());
Params.insert(Params.end(), CalleeInfo.ArgNodes.size(), MaskTy);
FunctionType *TargetType = FunctionType::get(FT->getReturnType(), Params, FT->isVarArg());
IRBuilder<> Builder(CS.getInstruction());
// Direct call, find the clone and cast it to what we want.
if (Function *F = dyn_cast<Function>(CS.getCalledValue()->stripPointerCasts())) {
Value *Clone = OldToNewFuncMap[F];
if (Clone) {
Clone = Builder.CreateBitCast(Clone, PointerType::getUnqual(TargetType));
}
return Clone;
}
// Indirect calls, cast the called value to the type we want.
Value *CalledValue = CS.getCalledValue();
return Builder.CreateBitCast(CalledValue, PointerType::getUnqual(TargetType));
}
Value *Reassociate::ReassociateExpression(BinaryOperator *I) {
// First, walk the expression tree, linearizing the tree, collecting the
// operand information.
SmallVector<ValueEntry, 8> Ops;
LinearizeExprTree(I, Ops);
DEBUG(dbgs() << "RAIn:\t"; PrintOps(I, Ops); dbgs() << '\n');
// Now that we have linearized the tree to a list and have gathered all of
// the operands and their ranks, sort the operands by their rank. Use a
// stable_sort so that values with equal ranks will have their relative
// positions maintained (and so the compiler is deterministic). Note that
// this sorts so that the highest ranking values end up at the beginning of
// the vector.
std::stable_sort(Ops.begin(), Ops.end());
// OptimizeExpression - Now that we have the expression tree in a convenient
// sorted form, optimize it globally if possible.
if (Value *V = OptimizeExpression(I, Ops)) {
// This expression tree simplified to something that isn't a tree,
// eliminate it.
DEBUG(dbgs() << "Reassoc to scalar: " << *V << '\n');
I->replaceAllUsesWith(V);
if (Instruction *VI = dyn_cast<Instruction>(V))
VI->setDebugLoc(I->getDebugLoc());
RemoveDeadBinaryOp(I);
++NumAnnihil;
return V;
}
// We want to sink immediates as deeply as possible except in the case where
// this is a multiply tree used only by an add, and the immediate is a -1.
// In this case we reassociate to put the negation on the outside so that we
// can fold the negation into the add: (-X)*Y + Z -> Z-X*Y
if (I->getOpcode() == Instruction::Mul && I->hasOneUse() &&
cast<Instruction>(I->use_back())->getOpcode() == Instruction::Add &&
isa<ConstantInt>(Ops.back().Op) &&
cast<ConstantInt>(Ops.back().Op)->isAllOnesValue()) {
ValueEntry Tmp = Ops.pop_back_val();
Ops.insert(Ops.begin(), Tmp);
}
DEBUG(dbgs() << "RAOut:\t"; PrintOps(I, Ops); dbgs() << '\n');
if (Ops.size() == 1) {
// This expression tree simplified to something that isn't a tree,
// eliminate it.
I->replaceAllUsesWith(Ops[0].Op);
if (Instruction *OI = dyn_cast<Instruction>(Ops[0].Op))
OI->setDebugLoc(I->getDebugLoc());
RemoveDeadBinaryOp(I);
return Ops[0].Op;
}
// Now that we ordered and optimized the expressions, splat them back into
// the expression tree, removing any unneeded nodes.
RewriteExprTree(I, Ops);
return I;
}
AttrListPtr AttrListPtr::removeAttr(unsigned Idx, Attributes Attrs) const {
#ifndef NDEBUG
// FIXME it is not obvious how this should work for alignment.
// For now, say we can't pass in alignment, which no current use does.
assert(!(Attrs & Attribute::Alignment) && "Attempt to exclude alignment!");
#endif
if (AttrList == 0) return AttrListPtr();
Attributes OldAttrs = getAttributes(Idx);
Attributes NewAttrs = OldAttrs & ~Attrs;
if (NewAttrs == OldAttrs)
return *this;
SmallVector<AttributeWithIndex, 8> NewAttrList;
const SmallVector<AttributeWithIndex, 4> &OldAttrList = AttrList->Attrs;
unsigned i = 0, e = OldAttrList.size();
// Copy attributes for arguments before this one.
for (; i != e && OldAttrList[i].Index < Idx; ++i)
NewAttrList.push_back(OldAttrList[i]);
// If there are attributes already at this index, merge them in.
assert(OldAttrList[i].Index == Idx && "Attribute isn't set?");
Attrs = OldAttrList[i].Attrs & ~Attrs;
++i;
if (Attrs) // If any attributes left for this parameter, add them.
NewAttrList.push_back(AttributeWithIndex::get(Idx, Attrs));
// Copy attributes for arguments after this one.
NewAttrList.insert(NewAttrList.end(),
OldAttrList.begin()+i, OldAttrList.end());
return get(NewAttrList.data(), NewAttrList.size());
}
TEST(SmallVectorTest, MidInsert) {
SmallVector<MovedFrom, 3> v;
v.push_back(MovedFrom());
v.insert(v.begin(), MovedFrom());
for (MovedFrom &m : v)
EXPECT_TRUE(m.hasValue);
}
/// createInvocationFromCommandLine - Construct a compiler invocation object for
/// a command line argument vector.
///
/// \return A CompilerInvocation, or 0 if none was built for the given
/// argument vector.
CompilerInvocation *
clang::createInvocationFromCommandLine(ArrayRef<const char *> ArgList,
IntrusiveRefCntPtr<DiagnosticsEngine> Diags) {
if (!Diags.getPtr()) {
// No diagnostics engine was provided, so create our own diagnostics object
// with the default options.
Diags = CompilerInstance::createDiagnostics(new DiagnosticOptions,
ArgList.size(),
ArgList.begin());
}
SmallVector<const char *, 16> Args;
Args.push_back("<clang>"); // FIXME: Remove dummy argument.
Args.insert(Args.end(), ArgList.begin(), ArgList.end());
// FIXME: Find a cleaner way to force the driver into restricted modes.
Args.push_back("-fsyntax-only");
// FIXME: We shouldn't have to pass in the path info.
driver::Driver TheDriver("clang", llvm::sys::getDefaultTargetTriple(),
"a.out", *Diags);
// Don't check that inputs exist, they may have been remapped.
TheDriver.setCheckInputsExist(false);
OwningPtr<driver::Compilation> C(TheDriver.BuildCompilation(Args));
// Just print the cc1 options if -### was present.
if (C->getArgs().hasArg(driver::options::OPT__HASH_HASH_HASH)) {
C->PrintJob(llvm::errs(), C->getJobs(), "\n", true);
return 0;
}
// We expect to get back exactly one command job, if we didn't something
// failed.
const driver::JobList &Jobs = C->getJobs();
if (Jobs.size() != 1 || !isa<driver::Command>(*Jobs.begin())) {
SmallString<256> Msg;
llvm::raw_svector_ostream OS(Msg);
C->PrintJob(OS, C->getJobs(), "; ", true);
Diags->Report(diag::err_fe_expected_compiler_job) << OS.str();
return 0;
}
const driver::Command *Cmd = cast<driver::Command>(*Jobs.begin());
if (StringRef(Cmd->getCreator().getName()) != "clang") {
Diags->Report(diag::err_fe_expected_clang_command);
return 0;
}
const driver::ArgStringList &CCArgs = Cmd->getArguments();
OwningPtr<CompilerInvocation> CI(new CompilerInvocation());
if (!CompilerInvocation::CreateFromArgs(*CI,
const_cast<const char **>(CCArgs.data()),
const_cast<const char **>(CCArgs.data()) +
CCArgs.size(),
*Diags))
return 0;
return CI.take();
}
bool DevirtModule::tryEvaluateFunctionsWithArgs(
MutableArrayRef<VirtualCallTarget> TargetsForSlot,
ArrayRef<ConstantInt *> Args) {
// Evaluate each function and store the result in each target's RetVal
// field.
for (VirtualCallTarget &Target : TargetsForSlot) {
if (Target.Fn->arg_size() != Args.size() + 1)
return false;
for (unsigned I = 0; I != Args.size(); ++I)
if (Target.Fn->getFunctionType()->getParamType(I + 1) !=
Args[I]->getType())
return false;
Evaluator Eval(M.getDataLayout(), nullptr);
SmallVector<Constant *, 2> EvalArgs;
EvalArgs.push_back(
Constant::getNullValue(Target.Fn->getFunctionType()->getParamType(0)));
EvalArgs.insert(EvalArgs.end(), Args.begin(), Args.end());
Constant *RetVal;
if (!Eval.EvaluateFunction(Target.Fn, RetVal, EvalArgs) ||
!isa<ConstantInt>(RetVal))
return false;
Target.RetVal = cast<ConstantInt>(RetVal)->getZExtValue();
}
return true;
}
std::unique_ptr<CompilerInvocation>
swift::driver::createCompilerInvocation(ArrayRef<const char *> ArgList,
DiagnosticEngine &Diags) {
SmallVector<const char *, 16> Args;
Args.push_back("<swiftc>"); // FIXME: Remove dummy argument.
Args.insert(Args.end(), ArgList.begin(), ArgList.end());
// When creating a CompilerInvocation, ensure that the driver creates a single
// frontend command.
Args.push_back("-force-single-frontend-invocation");
// Force the driver into batch mode by specifying "swiftc" as the name.
Driver TheDriver("swiftc", "swiftc", Args, Diags);
// Don't check for the existence of input files, since the user of the
// CompilerInvocation may wish to remap inputs to source buffers.
TheDriver.setCheckInputFilesExist(false);
std::unique_ptr<Compilation> C = TheDriver.buildCompilation(Args);
if (!C || C->getJobs().empty())
return nullptr; // Don't emit an error; one should already have been emitted
SmallPtrSet<const Job *, 4> CompileCommands;
for (const Job *Cmd : C->getJobs())
if (isa<CompileJobAction>(Cmd->getSource()))
CompileCommands.insert(Cmd);
if (CompileCommands.size() != 1) {
// TODO: include Jobs in the diagnostic.
Diags.diagnose(SourceLoc(), diag::error_expected_one_frontend_job);
return nullptr;
}
const Job *Cmd = *CompileCommands.begin();
if (StringRef("-frontend") != Cmd->getArguments().front()) {
Diags.diagnose(SourceLoc(), diag::error_expected_frontend_command);
return nullptr;
}
std::unique_ptr<CompilerInvocation> Invocation(new CompilerInvocation());
const llvm::opt::ArgStringList &BaseFrontendArgs = Cmd->getArguments();
ArrayRef<const char *> FrontendArgs =
llvm::makeArrayRef(BaseFrontendArgs.data() + 1,
BaseFrontendArgs.data() + BaseFrontendArgs.size());
if (Invocation->parseArgs(FrontendArgs, Diags))
return nullptr; // Don't emit an error; one should already have been emitted
return Invocation;
}
/// CreateComplexVariable - Create a new descriptor for the specified variable
/// which has a complex address expression for its address.
DIVariable DIFactory::CreateComplexVariable(unsigned Tag, DIDescriptor Context,
const std::string &Name,
DICompileUnit CompileUnit,
unsigned LineNo,
DIType Type, SmallVector<Value *, 9> &addr) {
SmallVector<Value *, 9> Elts;
Elts.push_back(GetTagConstant(Tag));
Elts.push_back(Context.getNode());
Elts.push_back(MDString::get(VMContext, Name));
Elts.push_back(CompileUnit.getNode());
Elts.push_back(ConstantInt::get(Type::getInt32Ty(VMContext), LineNo));
Elts.push_back(Type.getNode());
Elts.insert(Elts.end(), addr.begin(), addr.end());
return DIVariable(MDNode::get(VMContext, &Elts[0], 6+addr.size()));
}
int main(int argc, char **argv) {
cl::ParseCommandLineOptions(argc, argv, "Module concatenation");
ExitOnError ExitOnErr("llvm-cat: ");
LLVMContext Context;
SmallVector<char, 0> Buffer;
BitcodeWriter Writer(Buffer);
if (BinaryCat) {
for (const auto &InputFilename : InputFilenames) {
std::unique_ptr<MemoryBuffer> MB = ExitOnErr(
errorOrToExpected(MemoryBuffer::getFileOrSTDIN(InputFilename)));
std::vector<BitcodeModule> Mods = ExitOnErr(getBitcodeModuleList(*MB));
for (auto &BitcodeMod : Mods) {
Buffer.insert(Buffer.end(), BitcodeMod.getBuffer().begin(),
BitcodeMod.getBuffer().end());
Writer.copyStrtab(BitcodeMod.getStrtab());
}
}
} else {
// The string table does not own strings added to it, some of which are
// owned by the modules; keep them alive until we write the string table.
std::vector<std::unique_ptr<Module>> OwnedMods;
for (const auto &InputFilename : InputFilenames) {
SMDiagnostic Err;
std::unique_ptr<Module> M = parseIRFile(InputFilename, Err, Context);
if (!M) {
Err.print(argv[0], errs());
return 1;
}
Writer.writeModule(M.get());
OwnedMods.push_back(std::move(M));
}
Writer.writeStrtab();
}
std::error_code EC;
raw_fd_ostream OS(OutputFilename, EC, sys::fs::OpenFlags::F_None);
if (EC) {
errs() << argv[0] << ": cannot open " << OutputFilename << " for writing: "
<< EC.message();
return 1;
}
OS.write(Buffer.data(), Buffer.size());
return 0;
}
int csabase::run(int argc_, const char **argv_)
{
sys::PrintStackTraceOnErrorSignal();
PrettyStackTraceProgram X(argc_, argv_);
SmallVector<const char *, 1024> argv;
SpecificBumpPtrAllocator<char> ArgAllocator;
StringSetSaver Saver;
sys::Process::GetArgumentVector(argv,
ArrayRef<const char *>(argv_, argc_),
ArgAllocator);
cl::ExpandResponseFiles(Saver, cl::TokenizeGNUCommandLine, argv);
argv.insert(argv.begin() == argv.end() ? argv.begin() : argv.begin() + 1,
"-xc++");
llvm::InitializeNativeTarget();
llvm::InitializeNativeTargetAsmParser();
CompilerInstance Clang;
TextDiagnosticBuffer DiagsBuffer;
IntrusiveRefCntPtr<DiagnosticIDs> DiagID(new DiagnosticIDs());
IntrusiveRefCntPtr<DiagnosticOptions> DiagOpts(new DiagnosticOptions());
DiagnosticsEngine Diags(DiagID, &*DiagOpts, &DiagsBuffer, false);
bool Success = CompilerInvocation::CreateFromArgs(
Clang.getInvocation(),
argv.data() + 1,
argv.data() + argv.size(),
Diags);
Clang.createDiagnostics();
install_fatal_error_handler(LLVMErrorHandler, &Clang.getDiagnostics());
DiagsBuffer.FlushDiagnostics(Clang.getDiagnostics());
if (Success) {
Success = ExecuteCompilerInvocation(&Clang);
}
remove_fatal_error_handler();
llvm::llvm_shutdown();
return !Success;
}
AttrListPtr AttrListPtr::addAttr(unsigned Idx, Attributes Attrs) const {
Attributes OldAttrs = getAttributes(Idx);
#ifndef NDEBUG
// FIXME it is not obvious how this should work for alignment.
// For now, say we can't change a known alignment.
Attributes OldAlign = OldAttrs & Attribute::Alignment;
Attributes NewAlign = Attrs & Attribute::Alignment;
assert((!OldAlign || !NewAlign || OldAlign == NewAlign) &&
"Attempt to change alignment!");
#endif
Attributes NewAttrs = OldAttrs | Attrs;
if (NewAttrs == OldAttrs)
return *this;
SmallVector<AttributeWithIndex, 8> NewAttrList;
if (AttrList == 0)
NewAttrList.push_back(AttributeWithIndex::get(Idx, Attrs));
else {
const SmallVector<AttributeWithIndex, 4> &OldAttrList = AttrList->Attrs;
unsigned i = 0, e = OldAttrList.size();
// Copy attributes for arguments before this one.
for (; i != e && OldAttrList[i].Index < Idx; ++i)
NewAttrList.push_back(OldAttrList[i]);
// If there are attributes already at this index, merge them in.
if (i != e && OldAttrList[i].Index == Idx) {
Attrs |= OldAttrList[i].Attrs;
++i;
}
NewAttrList.push_back(AttributeWithIndex::get(Idx, Attrs));
// Copy attributes for arguments after this one.
NewAttrList.insert(NewAttrList.end(),
OldAttrList.begin()+i, OldAttrList.end());
}
return get(NewAttrList.data(), NewAttrList.size());
}
AstBackEnd::RegionType AstBackEnd::isRegion(BasicBlock &entry, BasicBlock *exit)
{
// LLVM's algorithm for finding regions (as of this early LLVM 3.7 fork) seems over-eager. For instance, with the
// following graph:
//
// 0
// |\
// | 1
// | |
// | 2=<| (where =<| denotes an edge to itself)
// |/
// 3
//
// LLVM thinks that BBs 2 and 3 form a region. After asking for help on the mailing list, it appears that LLVM
// tags it as an "extended region"; that is, a set of nodes that would be a region if we only added one basic block.
// This is not helpful for our purposes.
//
// Sine the classical definition of regions apply to edges and edges are second-class citizens in the LLVM graph
// world, we're going to roll with this inefficient-but-working, home-baked definition instead:
//
// A region is an ordered pair (A, B) of nodes, where A dominates, and B postdominates, every node
// traversed in any given iteration order from A to B. Additionally, no path starts after B such that a node of the
// region can be reached again without traversing A.
// This definition means that B is *excluded* from the region, because B could have predecessors that are not
// dominated by A. And I'm okay with it, I like [) ranges. To compensate, nullptr represents the end of a function.
bool cyclic = false;
unordered_set<BasicBlock*> toVisit { &entry };
unordered_set<BasicBlock*> visited { exit };
SmallVector<BasicBlock*, 2> nodeSuccessors;
// Step one: check domination
while (toVisit.size() > 0)
{
auto iter = toVisit.begin();
BasicBlock* bb = *iter;
// We use `exit = nullptr` to denote that the exit is the end of the function, which post-dominates
// every basic block. This is a deviation from the normal LLVM dominator tree behavior, where
// nullptr is considered unreachable (and thus does not dominate or post-dominate anything).
if (!domTree->dominates(&entry, bb) || (exit != nullptr && !postDomTree->dominates(exit, bb)))
{
return NotARegion;
}
toVisit.erase(iter);
visited.insert(bb);
// Only visit region successors. This saves times, and saves us from spuriously declaring that regions are
// cyclic by skipping cycles that have already been identified.
nodeSuccessors.clear();
AstGraphNode* graphNode = grapher->getGraphNodeFromEntry(bb);
if (graphNode->hasExit())
{
nodeSuccessors.push_back(graphNode->getExit());
}
else
{
nodeSuccessors.insert(nodeSuccessors.end(), succ_begin(bb), succ_end(bb));
}
for (BasicBlock* succ : nodeSuccessors)
{
if (visited.count(succ) == 0)
{
toVisit.insert(succ);
}
else if (succ == &entry)
{
cyclic = true;
}
}
}
// Step two: check that no path starting after the exit goes back into the region without first going through the
// entry.
unordered_set<BasicBlock*> regionMembers;
regionMembers.swap(visited);
regionMembers.erase(exit);
if (exit != nullptr)
{
toVisit.insert(succ_begin(exit), succ_end(exit));
}
visited.insert(&entry);
while (toVisit.size() > 0)
{
auto iter = toVisit.begin();
BasicBlock* bb = *iter;
if (regionMembers.count(bb) != 0)
{
return NotARegion;
}
toVisit.erase(iter);
visited.insert(bb);
for (BasicBlock* succ : successors(bb))
{
if (visited.count(succ) == 0)
{
toVisit.insert(succ);
}
}
}
return cyclic ? Cyclic : Acyclic;
}
void print(raw_ostream &OS, FuncIdConversionHelper &FN,
RootVector RootValues) {
// Go through each of the roots, and traverse the call stack, producing the
// aggregates as you go along. Remember these aggregates and stacks, and
// show summary statistics about:
//
// - Total number of unique stacks
// - Top 10 stacks by count
// - Top 10 stacks by aggregate duration
SmallVector<std::pair<const StackTrieNode *, uint64_t>, 11>
TopStacksByCount;
SmallVector<std::pair<const StackTrieNode *, uint64_t>, 11> TopStacksBySum;
auto greater_second =
[](const std::pair<const StackTrieNode *, uint64_t> &A,
const std::pair<const StackTrieNode *, uint64_t> &B) {
return A.second > B.second;
};
uint64_t UniqueStacks = 0;
for (const auto *N : RootValues) {
SmallVector<const StackTrieNode *, 16> S;
S.emplace_back(N);
while (!S.empty()) {
auto *Top = S.pop_back_val();
// We only start printing the stack (by walking up the parent pointers)
// when we get to a leaf function.
if (!Top->ExtraData.TerminalDurations.empty()) {
++UniqueStacks;
auto TopSum =
std::accumulate(Top->ExtraData.TerminalDurations.begin(),
Top->ExtraData.TerminalDurations.end(), 0uLL);
{
auto E = std::make_pair(Top, TopSum);
TopStacksBySum.insert(std::lower_bound(TopStacksBySum.begin(),
TopStacksBySum.end(), E,
greater_second),
E);
if (TopStacksBySum.size() == 11)
TopStacksBySum.pop_back();
}
{
auto E =
std::make_pair(Top, Top->ExtraData.TerminalDurations.size());
TopStacksByCount.insert(std::lower_bound(TopStacksByCount.begin(),
TopStacksByCount.end(), E,
greater_second),
E);
if (TopStacksByCount.size() == 11)
TopStacksByCount.pop_back();
}
}
for (const auto *C : Top->Callees)
S.push_back(C);
}
}
// Now print the statistics in the end.
OS << "\n";
OS << "Unique Stacks: " << UniqueStacks << "\n";
OS << "Top 10 Stacks by leaf sum:\n\n";
for (const auto &P : TopStacksBySum) {
OS << "Sum: " << P.second << "\n";
printStack(OS, P.first, FN);
}
OS << "\n";
OS << "Top 10 Stacks by leaf count:\n\n";
for (const auto &P : TopStacksByCount) {
OS << "Count: " << P.second << "\n";
printStack(OS, P.first, FN);
}
OS << "\n";
}
SDValue SelectionDAGBuilder::LowerAsSTATEPOINT(
SelectionDAGBuilder::StatepointLoweringInfo &SI) {
// The basic scheme here is that information about both the original call and
// the safepoint is encoded in the CallInst. We create a temporary call and
// lower it, then reverse engineer the calling sequence.
NumOfStatepoints++;
// Clear state
StatepointLowering.startNewStatepoint(*this);
#ifndef NDEBUG
// We schedule gc relocates before removeDuplicateGCPtrs since we _will_
// encounter the duplicate gc relocates we elide in removeDuplicateGCPtrs.
for (auto *Reloc : SI.GCRelocates)
if (Reloc->getParent() == SI.StatepointInstr->getParent())
StatepointLowering.scheduleRelocCall(*Reloc);
#endif
// Remove any redundant llvm::Values which map to the same SDValue as another
// input. Also has the effect of removing duplicates in the original
// llvm::Value input list as well. This is a useful optimization for
// reducing the size of the StackMap section. It has no other impact.
removeDuplicateGCPtrs(SI.Bases, SI.Ptrs, SI.GCRelocates, *this,
FuncInfo.StatepointSpillMaps[SI.StatepointInstr]);
assert(SI.Bases.size() == SI.Ptrs.size() &&
SI.Ptrs.size() == SI.GCRelocates.size());
// Lower statepoint vmstate and gcstate arguments
SmallVector<SDValue, 10> LoweredMetaArgs;
lowerStatepointMetaArgs(LoweredMetaArgs, SI, *this);
// Now that we've emitted the spills, we need to update the root so that the
// call sequence is ordered correctly.
SI.CLI.setChain(getRoot());
// Get call node, we will replace it later with statepoint
SDValue ReturnVal;
SDNode *CallNode;
std::tie(ReturnVal, CallNode) =
lowerCallFromStatepointLoweringInfo(SI, *this, PendingExports);
// Construct the actual GC_TRANSITION_START, STATEPOINT, and GC_TRANSITION_END
// nodes with all the appropriate arguments and return values.
// Call Node: Chain, Target, {Args}, RegMask, [Glue]
SDValue Chain = CallNode->getOperand(0);
SDValue Glue;
bool CallHasIncomingGlue = CallNode->getGluedNode();
if (CallHasIncomingGlue) {
// Glue is always last operand
Glue = CallNode->getOperand(CallNode->getNumOperands() - 1);
}
// Build the GC_TRANSITION_START node if necessary.
//
// The operands to the GC_TRANSITION_{START,END} nodes are laid out in the
// order in which they appear in the call to the statepoint intrinsic. If
// any of the operands is a pointer-typed, that operand is immediately
// followed by a SRCVALUE for the pointer that may be used during lowering
// (e.g. to form MachinePointerInfo values for loads/stores).
const bool IsGCTransition =
(SI.StatepointFlags & (uint64_t)StatepointFlags::GCTransition) ==
(uint64_t)StatepointFlags::GCTransition;
if (IsGCTransition) {
SmallVector<SDValue, 8> TSOps;
// Add chain
TSOps.push_back(Chain);
// Add GC transition arguments
for (const Value *V : SI.GCTransitionArgs) {
TSOps.push_back(getValue(V));
if (V->getType()->isPointerTy())
TSOps.push_back(DAG.getSrcValue(V));
}
// Add glue if necessary
if (CallHasIncomingGlue)
TSOps.push_back(Glue);
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue GCTransitionStart =
DAG.getNode(ISD::GC_TRANSITION_START, getCurSDLoc(), NodeTys, TSOps);
Chain = GCTransitionStart.getValue(0);
Glue = GCTransitionStart.getValue(1);
}
// TODO: Currently, all of these operands are being marked as read/write in
// PrologEpilougeInserter.cpp, we should special case the VMState arguments
// and flags to be read-only.
SmallVector<SDValue, 40> Ops;
// Add the <id> and <numBytes> constants.
Ops.push_back(DAG.getTargetConstant(SI.ID, getCurSDLoc(), MVT::i64));
Ops.push_back(
DAG.getTargetConstant(SI.NumPatchBytes, getCurSDLoc(), MVT::i32));
// Calculate and push starting position of vmstate arguments
// Get number of arguments incoming directly into call node
unsigned NumCallRegArgs =
CallNode->getNumOperands() - (CallHasIncomingGlue ? 4 : 3);
Ops.push_back(DAG.getTargetConstant(NumCallRegArgs, getCurSDLoc(), MVT::i32));
// Add call target
SDValue CallTarget = SDValue(CallNode->getOperand(1).getNode(), 0);
Ops.push_back(CallTarget);
// Add call arguments
// Get position of register mask in the call
SDNode::op_iterator RegMaskIt;
if (CallHasIncomingGlue)
RegMaskIt = CallNode->op_end() - 2;
else
RegMaskIt = CallNode->op_end() - 1;
Ops.insert(Ops.end(), CallNode->op_begin() + 2, RegMaskIt);
// Add a constant argument for the calling convention
pushStackMapConstant(Ops, *this, SI.CLI.CallConv);
// Add a constant argument for the flags
uint64_t Flags = SI.StatepointFlags;
assert(((Flags & ~(uint64_t)StatepointFlags::MaskAll) == 0) &&
"Unknown flag used");
pushStackMapConstant(Ops, *this, Flags);
// Insert all vmstate and gcstate arguments
Ops.insert(Ops.end(), LoweredMetaArgs.begin(), LoweredMetaArgs.end());
// Add register mask from call node
Ops.push_back(*RegMaskIt);
// Add chain
Ops.push_back(Chain);
// Same for the glue, but we add it only if original call had it
if (Glue.getNode())
Ops.push_back(Glue);
// Compute return values. Provide a glue output since we consume one as
// input. This allows someone else to chain off us as needed.
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
SDNode *StatepointMCNode =
DAG.getMachineNode(TargetOpcode::STATEPOINT, getCurSDLoc(), NodeTys, Ops);
SDNode *SinkNode = StatepointMCNode;
// Build the GC_TRANSITION_END node if necessary.
//
// See the comment above regarding GC_TRANSITION_START for the layout of
// the operands to the GC_TRANSITION_END node.
if (IsGCTransition) {
SmallVector<SDValue, 8> TEOps;
// Add chain
TEOps.push_back(SDValue(StatepointMCNode, 0));
// Add GC transition arguments
for (const Value *V : SI.GCTransitionArgs) {
TEOps.push_back(getValue(V));
if (V->getType()->isPointerTy())
TEOps.push_back(DAG.getSrcValue(V));
}
// Add glue
TEOps.push_back(SDValue(StatepointMCNode, 1));
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue GCTransitionStart =
DAG.getNode(ISD::GC_TRANSITION_END, getCurSDLoc(), NodeTys, TEOps);
SinkNode = GCTransitionStart.getNode();
}
// Replace original call
DAG.ReplaceAllUsesWith(CallNode, SinkNode); // This may update Root
// Remove original call node
DAG.DeleteNode(CallNode);
// DON'T set the root - under the assumption that it's already set past the
// inserted node we created.
// TODO: A better future implementation would be to emit a single variable
// argument, variable return value STATEPOINT node here and then hookup the
// return value of each gc.relocate to the respective output of the
// previously emitted STATEPOINT value. Unfortunately, this doesn't appear
// to actually be possible today.
return ReturnVal;
}
bool AllocaArraysMerging::runOnFunction(Function& F)
{
class ArraysToMerge
{
private:
std::map<AllocaInst*, uint32_t> arraysToMerge;
uint32_t currentOffset;
public:
ArraysToMerge():currentOffset(0)
{
}
bool empty() const
{
return arraysToMerge.empty();
}
std::map<AllocaInst*, uint32_t>::iterator begin()
{
return arraysToMerge.begin();
}
std::map<AllocaInst*, uint32_t>::iterator end()
{
return arraysToMerge.end();
}
void add(AllocaInst* a)
{
arraysToMerge.insert(std::make_pair(a, currentOffset));
currentOffset+=cast<ArrayType>(a->getAllocatedType())->getNumElements();
}
uint32_t getNewSize() const
{
return currentOffset;
}
};
cheerp::PointerAnalyzer & PA = getAnalysis<cheerp::PointerAnalyzer>();
cheerp::Registerize & registerize = getAnalysis<cheerp::Registerize>();
cheerp::GlobalDepsAnalyzer & GDA = getAnalysis<cheerp::GlobalDepsAnalyzer>();
std::list<std::pair<AllocaInst*, cheerp::Registerize::LiveRange>> allocaInfos;
// Gather all the allocas
for(BasicBlock& BB: F)
analyzeBlock(registerize, BB, allocaInfos);
if (allocaInfos.size() < 2)
return false;
bool Changed = false;
// We can also try to merge arrays of the same type, if only pointers to values are passed around
while(!allocaInfos.empty())
{
// Build a map of array to be merged and their offseet into the new array
ArraysToMerge arraysToMerge;
auto targetCandidate = allocaInfos.begin();
AllocaInst* targetAlloca = targetCandidate->first;
if(!targetAlloca->getAllocatedType()->isArrayTy() ||
// Check target uses
!checkUsesForArrayMerging(targetAlloca))
{
allocaInfos.erase(targetCandidate);
continue;
}
Type* targetElementType = targetAlloca->getAllocatedType()->getSequentialElementType();
auto sourceCandidate=targetCandidate;
++sourceCandidate;
// Now that we have computed the sourceCandidate we can invalidate the targetCandidate
allocaInfos.erase(targetCandidate);
while(sourceCandidate!=allocaInfos.end())
{
AllocaInst* sourceAlloca = sourceCandidate->first;
// Check that allocas are arrays of the same type
if(!sourceAlloca->getAllocatedType()->isArrayTy())
{
++sourceCandidate;
continue;
}
// Both are arrays, check the types
if(targetElementType != sourceAlloca->getAllocatedType()->getSequentialElementType())
{
++sourceCandidate;
continue;
}
// Verify that the source candidate has supported uses
if(!checkUsesForArrayMerging(sourceAlloca))
{
++sourceCandidate;
continue;
}
// We can merge the source and the target
// If the set is empty add the target as well
if(arraysToMerge.empty())
arraysToMerge.add(targetAlloca);
arraysToMerge.add(sourceAlloca);
auto oldCandidate = sourceCandidate;
++sourceCandidate;
// Now that we have moved to the next candidate, we can invalidate the old one
allocaInfos.erase(oldCandidate);
}
// If we have a non-empty set of alloca merge them
if (arraysToMerge.empty())
continue;
if(!Changed)
registerize.invalidateLiveRangeForAllocas(F);
// Build new alloca
Type* newAllocaType = ArrayType::get(targetElementType, arraysToMerge.getNewSize());
// Add the new struct type to the GlobalDepsAnalyzer, it may need the createArray helper
GDA.visitType(newAllocaType, /*forceTypedArray*/ false);
AllocaInst* newAlloca = new AllocaInst(newAllocaType, "mergedArray", &(*F.getEntryBlock().begin()));
Type* indexType = IntegerType::get(newAllocaType->getContext(), 32);
// Change every use of every merged array with an appropiate GEP
for(auto it: arraysToMerge)
{
AllocaInst* allocaToMerge = it.first;
uint32_t baseOffset = it.second;
SmallVector<User*, 4> users(allocaToMerge->users());
for(User* u: users)
{
if(GetElementPtrInst* oldGep = dyn_cast<GetElementPtrInst>(u))
{
// Build 2 GEPs, one to reach the first element in the merged array
// and the other for the rest of the offsets
SmallVector<Value*, 4> indices;
// Dereference array
indices.push_back(ConstantInt::get(indexType, 0));
// Reach offset
indices.push_back(ConstantInt::get(indexType, baseOffset));
Value* gep1 = GetElementPtrInst::Create(newAlloca, indices, "", oldGep);
// Apply all the old offsets but the first one using a new GEP
indices.clear();
indices.insert(indices.begin(), oldGep->idx_begin()+1, oldGep->idx_end());
Value* gep2 = GetElementPtrInst::Create(gep1, indices, "", oldGep);
// Replace all uses with gep2
oldGep->replaceAllUsesWith(gep2);
PA.invalidate(oldGep);
oldGep->eraseFromParent();
}
else if(BitCastInst* BI=dyn_cast<BitCastInst>(u))
{
//Only used for lifetime intrinsics
Value* newBitCast=new BitCastInst(newAlloca, BI->getType(), "", BI);
BI->replaceAllUsesWith(newBitCast);
PA.invalidate(BI);
BI->eraseFromParent();
}
else
assert(false && "Unexpected use while merging arrays");
}
// Kill the alloca itself now
PA.invalidate(allocaToMerge);
allocaToMerge->eraseFromParent();
Changed = true;
}
}
if(Changed)
registerize.computeLiveRangeForAllocas(F);
return Changed;
}
/// DoPromotion - This method actually performs the promotion of the specified
/// arguments, and returns the new function. At this point, we know that it's
/// safe to do so.
static Function *
doPromotion(Function *F, SmallPtrSetImpl<Argument *> &ArgsToPromote,
SmallPtrSetImpl<Argument *> &ByValArgsToTransform,
Optional<function_ref<void(CallSite OldCS, CallSite NewCS)>>
ReplaceCallSite) {
// Start by computing a new prototype for the function, which is the same as
// the old function, but has modified arguments.
FunctionType *FTy = F->getFunctionType();
std::vector<Type *> Params;
using ScalarizeTable = std::set<std::pair<Type *, IndicesVector>>;
// ScalarizedElements - If we are promoting a pointer that has elements
// accessed out of it, keep track of which elements are accessed so that we
// can add one argument for each.
//
// Arguments that are directly loaded will have a zero element value here, to
// handle cases where there are both a direct load and GEP accesses.
std::map<Argument *, ScalarizeTable> ScalarizedElements;
// OriginalLoads - Keep track of a representative load instruction from the
// original function so that we can tell the alias analysis implementation
// what the new GEP/Load instructions we are inserting look like.
// We need to keep the original loads for each argument and the elements
// of the argument that are accessed.
std::map<std::pair<Argument *, IndicesVector>, LoadInst *> OriginalLoads;
// Attribute - Keep track of the parameter attributes for the arguments
// that we are *not* promoting. For the ones that we do promote, the parameter
// attributes are lost
SmallVector<AttributeSet, 8> ArgAttrVec;
AttributeList PAL = F->getAttributes();
// First, determine the new argument list
unsigned ArgNo = 0;
for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(); I != E;
++I, ++ArgNo) {
if (ByValArgsToTransform.count(&*I)) {
// Simple byval argument? Just add all the struct element types.
Type *AgTy = cast<PointerType>(I->getType())->getElementType();
StructType *STy = cast<StructType>(AgTy);
Params.insert(Params.end(), STy->element_begin(), STy->element_end());
ArgAttrVec.insert(ArgAttrVec.end(), STy->getNumElements(),
AttributeSet());
++NumByValArgsPromoted;
} else if (!ArgsToPromote.count(&*I)) {
// Unchanged argument
Params.push_back(I->getType());
ArgAttrVec.push_back(PAL.getParamAttributes(ArgNo));
} else if (I->use_empty()) {
// Dead argument (which are always marked as promotable)
++NumArgumentsDead;
// There may be remaining metadata uses of the argument for things like
// llvm.dbg.value. Replace them with undef.
I->replaceAllUsesWith(UndefValue::get(I->getType()));
} else {
// Okay, this is being promoted. This means that the only uses are loads
// or GEPs which are only used by loads
// In this table, we will track which indices are loaded from the argument
// (where direct loads are tracked as no indices).
ScalarizeTable &ArgIndices = ScalarizedElements[&*I];
for (User *U : I->users()) {
Instruction *UI = cast<Instruction>(U);
Type *SrcTy;
if (LoadInst *L = dyn_cast<LoadInst>(UI))
SrcTy = L->getType();
else
SrcTy = cast<GetElementPtrInst>(UI)->getSourceElementType();
IndicesVector Indices;
Indices.reserve(UI->getNumOperands() - 1);
// Since loads will only have a single operand, and GEPs only a single
// non-index operand, this will record direct loads without any indices,
// and gep+loads with the GEP indices.
for (User::op_iterator II = UI->op_begin() + 1, IE = UI->op_end();
II != IE; ++II)
Indices.push_back(cast<ConstantInt>(*II)->getSExtValue());
// GEPs with a single 0 index can be merged with direct loads
if (Indices.size() == 1 && Indices.front() == 0)
Indices.clear();
ArgIndices.insert(std::make_pair(SrcTy, Indices));
LoadInst *OrigLoad;
if (LoadInst *L = dyn_cast<LoadInst>(UI))
OrigLoad = L;
else
// Take any load, we will use it only to update Alias Analysis
OrigLoad = cast<LoadInst>(UI->user_back());
OriginalLoads[std::make_pair(&*I, Indices)] = OrigLoad;
}
// Add a parameter to the function for each element passed in.
for (const auto &ArgIndex : ArgIndices) {
// not allowed to dereference ->begin() if size() is 0
Params.push_back(GetElementPtrInst::getIndexedType(
cast<PointerType>(I->getType()->getScalarType())->getElementType(),
ArgIndex.second));
ArgAttrVec.push_back(AttributeSet());
assert(Params.back());
}
if (ArgIndices.size() == 1 && ArgIndices.begin()->second.empty())
++NumArgumentsPromoted;
else
++NumAggregatesPromoted;
}
}
Type *RetTy = FTy->getReturnType();
// Construct the new function type using the new arguments.
FunctionType *NFTy = FunctionType::get(RetTy, Params, FTy->isVarArg());
// Create the new function body and insert it into the module.
Function *NF = Function::Create(NFTy, F->getLinkage(), F->getName());
NF->copyAttributesFrom(F);
// Patch the pointer to LLVM function in debug info descriptor.
NF->setSubprogram(F->getSubprogram());
F->setSubprogram(nullptr);
DEBUG(dbgs() << "ARG PROMOTION: Promoting to:" << *NF << "\n"
<< "From: " << *F);
// Recompute the parameter attributes list based on the new arguments for
// the function.
NF->setAttributes(AttributeList::get(F->getContext(), PAL.getFnAttributes(),
PAL.getRetAttributes(), ArgAttrVec));
ArgAttrVec.clear();
F->getParent()->getFunctionList().insert(F->getIterator(), NF);
NF->takeName(F);
// Loop over all of the callers of the function, transforming the call sites
// to pass in the loaded pointers.
//
SmallVector<Value *, 16> Args;
while (!F->use_empty()) {
CallSite CS(F->user_back());
assert(CS.getCalledFunction() == F);
Instruction *Call = CS.getInstruction();
const AttributeList &CallPAL = CS.getAttributes();
// Loop over the operands, inserting GEP and loads in the caller as
// appropriate.
CallSite::arg_iterator AI = CS.arg_begin();
ArgNo = 0;
for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(); I != E;
++I, ++AI, ++ArgNo)
if (!ArgsToPromote.count(&*I) && !ByValArgsToTransform.count(&*I)) {
Args.push_back(*AI); // Unmodified argument
ArgAttrVec.push_back(CallPAL.getParamAttributes(ArgNo));
} else if (ByValArgsToTransform.count(&*I)) {
// Emit a GEP and load for each element of the struct.
Type *AgTy = cast<PointerType>(I->getType())->getElementType();
StructType *STy = cast<StructType>(AgTy);
Value *Idxs[2] = {
ConstantInt::get(Type::getInt32Ty(F->getContext()), 0), nullptr};
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
Idxs[1] = ConstantInt::get(Type::getInt32Ty(F->getContext()), i);
Value *Idx = GetElementPtrInst::Create(
STy, *AI, Idxs, (*AI)->getName() + "." + Twine(i), Call);
// TODO: Tell AA about the new values?
Args.push_back(new LoadInst(Idx, Idx->getName() + ".val", Call));
ArgAttrVec.push_back(AttributeSet());
}
} else if (!I->use_empty()) {
// Non-dead argument: insert GEPs and loads as appropriate.
ScalarizeTable &ArgIndices = ScalarizedElements[&*I];
// Store the Value* version of the indices in here, but declare it now
// for reuse.
std::vector<Value *> Ops;
for (const auto &ArgIndex : ArgIndices) {
Value *V = *AI;
LoadInst *OrigLoad =
OriginalLoads[std::make_pair(&*I, ArgIndex.second)];
if (!ArgIndex.second.empty()) {
Ops.reserve(ArgIndex.second.size());
Type *ElTy = V->getType();
for (auto II : ArgIndex.second) {
// Use i32 to index structs, and i64 for others (pointers/arrays).
// This satisfies GEP constraints.
Type *IdxTy =
(ElTy->isStructTy() ? Type::getInt32Ty(F->getContext())
: Type::getInt64Ty(F->getContext()));
Ops.push_back(ConstantInt::get(IdxTy, II));
// Keep track of the type we're currently indexing.
if (auto *ElPTy = dyn_cast<PointerType>(ElTy))
ElTy = ElPTy->getElementType();
else
ElTy = cast<CompositeType>(ElTy)->getTypeAtIndex(II);
}
// And create a GEP to extract those indices.
V = GetElementPtrInst::Create(ArgIndex.first, V, Ops,
V->getName() + ".idx", Call);
Ops.clear();
}
// Since we're replacing a load make sure we take the alignment
// of the previous load.
LoadInst *newLoad = new LoadInst(V, V->getName() + ".val", Call);
newLoad->setAlignment(OrigLoad->getAlignment());
// Transfer the AA info too.
AAMDNodes AAInfo;
OrigLoad->getAAMetadata(AAInfo);
newLoad->setAAMetadata(AAInfo);
Args.push_back(newLoad);
ArgAttrVec.push_back(AttributeSet());
}
}
// Push any varargs arguments on the list.
for (; AI != CS.arg_end(); ++AI, ++ArgNo) {
Args.push_back(*AI);
ArgAttrVec.push_back(CallPAL.getParamAttributes(ArgNo));
}
SmallVector<OperandBundleDef, 1> OpBundles;
CS.getOperandBundlesAsDefs(OpBundles);
CallSite NewCS;
if (InvokeInst *II = dyn_cast<InvokeInst>(Call)) {
NewCS = InvokeInst::Create(NF, II->getNormalDest(), II->getUnwindDest(),
Args, OpBundles, "", Call);
} else {
auto *NewCall = CallInst::Create(NF, Args, OpBundles, "", Call);
NewCall->setTailCallKind(cast<CallInst>(Call)->getTailCallKind());
NewCS = NewCall;
}
NewCS.setCallingConv(CS.getCallingConv());
NewCS.setAttributes(
AttributeList::get(F->getContext(), CallPAL.getFnAttributes(),
CallPAL.getRetAttributes(), ArgAttrVec));
NewCS->setDebugLoc(Call->getDebugLoc());
uint64_t W;
if (Call->extractProfTotalWeight(W))
NewCS->setProfWeight(W);
Args.clear();
ArgAttrVec.clear();
// Update the callgraph to know that the callsite has been transformed.
if (ReplaceCallSite)
(*ReplaceCallSite)(CS, NewCS);
if (!Call->use_empty()) {
Call->replaceAllUsesWith(NewCS.getInstruction());
NewCS->takeName(Call);
}
// Finally, remove the old call from the program, reducing the use-count of
// F.
Call->eraseFromParent();
}
const DataLayout &DL = F->getParent()->getDataLayout();
// Since we have now created the new function, splice the body of the old
// function right into the new function, leaving the old rotting hulk of the
// function empty.
NF->getBasicBlockList().splice(NF->begin(), F->getBasicBlockList());
// Loop over the argument list, transferring uses of the old arguments over to
// the new arguments, also transferring over the names as well.
for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(),
I2 = NF->arg_begin();
I != E; ++I) {
if (!ArgsToPromote.count(&*I) && !ByValArgsToTransform.count(&*I)) {
// If this is an unmodified argument, move the name and users over to the
// new version.
I->replaceAllUsesWith(&*I2);
I2->takeName(&*I);
++I2;
continue;
}
if (ByValArgsToTransform.count(&*I)) {
// In the callee, we create an alloca, and store each of the new incoming
// arguments into the alloca.
Instruction *InsertPt = &NF->begin()->front();
// Just add all the struct element types.
Type *AgTy = cast<PointerType>(I->getType())->getElementType();
Value *TheAlloca = new AllocaInst(AgTy, DL.getAllocaAddrSpace(), nullptr,
I->getParamAlignment(), "", InsertPt);
StructType *STy = cast<StructType>(AgTy);
Value *Idxs[2] = {ConstantInt::get(Type::getInt32Ty(F->getContext()), 0),
nullptr};
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
Idxs[1] = ConstantInt::get(Type::getInt32Ty(F->getContext()), i);
Value *Idx = GetElementPtrInst::Create(
AgTy, TheAlloca, Idxs, TheAlloca->getName() + "." + Twine(i),
InsertPt);
I2->setName(I->getName() + "." + Twine(i));
new StoreInst(&*I2++, Idx, InsertPt);
}
// Anything that used the arg should now use the alloca.
I->replaceAllUsesWith(TheAlloca);
TheAlloca->takeName(&*I);
// If the alloca is used in a call, we must clear the tail flag since
// the callee now uses an alloca from the caller.
for (User *U : TheAlloca->users()) {
CallInst *Call = dyn_cast<CallInst>(U);
if (!Call)
continue;
Call->setTailCall(false);
}
continue;
}
if (I->use_empty())
continue;
// Otherwise, if we promoted this argument, then all users are load
// instructions (or GEPs with only load users), and all loads should be
// using the new argument that we added.
ScalarizeTable &ArgIndices = ScalarizedElements[&*I];
while (!I->use_empty()) {
if (LoadInst *LI = dyn_cast<LoadInst>(I->user_back())) {
assert(ArgIndices.begin()->second.empty() &&
"Load element should sort to front!");
I2->setName(I->getName() + ".val");
LI->replaceAllUsesWith(&*I2);
LI->eraseFromParent();
DEBUG(dbgs() << "*** Promoted load of argument '" << I->getName()
<< "' in function '" << F->getName() << "'\n");
} else {
GetElementPtrInst *GEP = cast<GetElementPtrInst>(I->user_back());
IndicesVector Operands;
Operands.reserve(GEP->getNumIndices());
for (User::op_iterator II = GEP->idx_begin(), IE = GEP->idx_end();
II != IE; ++II)
Operands.push_back(cast<ConstantInt>(*II)->getSExtValue());
// GEPs with a single 0 index can be merged with direct loads
if (Operands.size() == 1 && Operands.front() == 0)
Operands.clear();
Function::arg_iterator TheArg = I2;
for (ScalarizeTable::iterator It = ArgIndices.begin();
It->second != Operands; ++It, ++TheArg) {
assert(It != ArgIndices.end() && "GEP not handled??");
}
std::string NewName = I->getName();
for (unsigned i = 0, e = Operands.size(); i != e; ++i) {
NewName += "." + utostr(Operands[i]);
}
NewName += ".val";
TheArg->setName(NewName);
DEBUG(dbgs() << "*** Promoted agg argument '" << TheArg->getName()
<< "' of function '" << NF->getName() << "'\n");
// All of the uses must be load instructions. Replace them all with
// the argument specified by ArgNo.
while (!GEP->use_empty()) {
LoadInst *L = cast<LoadInst>(GEP->user_back());
L->replaceAllUsesWith(&*TheArg);
L->eraseFromParent();
}
GEP->eraseFromParent();
}
}
// Increment I2 past all of the arguments added for this promoted pointer.
std::advance(I2, ArgIndices.size());
}
return NF;
}
// A soft instruction can be changed to work in other domains given by mask.
void ExeDepsFix::visitSoftInstr(MachineInstr *mi, unsigned mask) {
// Bitmask of available domains for this instruction after taking collapsed
// operands into account.
unsigned available = mask;
// Scan the explicit use operands for incoming domains.
SmallVector<int, 4> used;
if (LiveRegs)
for (unsigned i = mi->getDesc().getNumDefs(),
e = mi->getDesc().getNumOperands(); i != e; ++i) {
MachineOperand &mo = mi->getOperand(i);
if (!mo.isReg()) continue;
int rx = regIndex(mo.getReg());
if (rx < 0) continue;
if (DomainValue *dv = LiveRegs[rx].Value) {
// Bitmask of domains that dv and available have in common.
unsigned common = dv->getCommonDomains(available);
// Is it possible to use this collapsed register for free?
if (dv->isCollapsed()) {
// Restrict available domains to the ones in common with the operand.
// If there are no common domains, we must pay the cross-domain
// penalty for this operand.
if (common) available = common;
} else if (common)
// Open DomainValue is compatible, save it for merging.
used.push_back(rx);
else
// Open DomainValue is not compatible with instruction. It is useless
// now.
kill(rx);
}
}
// If the collapsed operands force a single domain, propagate the collapse.
if (isPowerOf2_32(available)) {
unsigned domain = countTrailingZeros(available);
TII->setExecutionDomain(mi, domain);
visitHardInstr(mi, domain);
return;
}
// Kill off any remaining uses that don't match available, and build a list of
// incoming DomainValues that we want to merge.
SmallVector<LiveReg, 4> Regs;
for (SmallVector<int, 4>::iterator i=used.begin(), e=used.end(); i!=e; ++i) {
int rx = *i;
const LiveReg &LR = LiveRegs[rx];
// This useless DomainValue could have been missed above.
if (!LR.Value->getCommonDomains(available)) {
kill(rx);
continue;
}
// Sorted insertion.
bool Inserted = false;
for (SmallVector<LiveReg, 4>::iterator i = Regs.begin(), e = Regs.end();
i != e && !Inserted; ++i) {
if (LR.Def < i->Def) {
Inserted = true;
Regs.insert(i, LR);
}
}
if (!Inserted)
Regs.push_back(LR);
}
// doms are now sorted in order of appearance. Try to merge them all, giving
// priority to the latest ones.
DomainValue *dv = 0;
while (!Regs.empty()) {
if (!dv) {
dv = Regs.pop_back_val().Value;
// Force the first dv to match the current instruction.
dv->AvailableDomains = dv->getCommonDomains(available);
assert(dv->AvailableDomains && "Domain should have been filtered");
continue;
}
DomainValue *Latest = Regs.pop_back_val().Value;
// Skip already merged values.
if (Latest == dv || Latest->Next)
continue;
if (merge(dv, Latest))
continue;
// If latest didn't merge, it is useless now. Kill all registers using it.
for (SmallVector<int,4>::iterator i=used.begin(), e=used.end(); i != e; ++i)
if (LiveRegs[*i].Value == Latest)
kill(*i);
}
// dv is the DomainValue we are going to use for this instruction.
if (!dv) {
dv = alloc();
dv->AvailableDomains = available;
}
dv->Instrs.push_back(mi);
// Finally set all defs and non-collapsed uses to dv. We must iterate through
// all the operators, including imp-def ones.
for (MachineInstr::mop_iterator ii = mi->operands_begin(),
ee = mi->operands_end();
ii != ee; ++ii) {
MachineOperand &mo = *ii;
if (!mo.isReg()) continue;
int rx = regIndex(mo.getReg());
if (rx < 0) continue;
if (!LiveRegs[rx].Value || (mo.isDef() && LiveRegs[rx].Value != dv)) {
kill(rx);
setLiveReg(rx, dv);
}
}
}
/// runOnLoop - Remove dead loops, by which we mean loops that do not impact the
/// observable behavior of the program other than finite running time. Note
/// we do ensure that this never remove a loop that might be infinite, as doing
/// so could change the halting/non-halting nature of a program.
/// NOTE: This entire process relies pretty heavily on LoopSimplify and LCSSA
/// in order to make various safety checks work.
bool LoopDeletion::runOnLoop(Loop *L, LPPassManager &LPM) {
// We can only remove the loop if there is a preheader that we can
// branch from after removing it.
BasicBlock *preheader = L->getLoopPreheader();
if (!preheader)
return false;
// If LoopSimplify form is not available, stay out of trouble.
if (!L->hasDedicatedExits())
return false;
// We can't remove loops that contain subloops. If the subloops were dead,
// they would already have been removed in earlier executions of this pass.
if (L->begin() != L->end())
return false;
SmallVector<BasicBlock*, 4> exitingBlocks;
L->getExitingBlocks(exitingBlocks);
SmallVector<BasicBlock*, 4> exitBlocks;
L->getUniqueExitBlocks(exitBlocks);
// We require that the loop only have a single exit block. Otherwise, we'd
// be in the situation of needing to be able to solve statically which exit
// block will be branched to, or trying to preserve the branching logic in
// a loop invariant manner.
if (exitBlocks.size() != 1)
return false;
// Finally, we have to check that the loop really is dead.
bool Changed = false;
if (!isLoopDead(L, exitingBlocks, exitBlocks, Changed, preheader))
return Changed;
// Don't remove loops for which we can't solve the trip count.
// They could be infinite, in which case we'd be changing program behavior.
ScalarEvolution &SE = getAnalysis<ScalarEvolution>();
const SCEV *S = SE.getMaxBackedgeTakenCount(L);
if (isa<SCEVCouldNotCompute>(S))
return Changed;
// Now that we know the removal is safe, remove the loop by changing the
// branch from the preheader to go to the single exit block.
BasicBlock *exitBlock = exitBlocks[0];
// Because we're deleting a large chunk of code at once, the sequence in which
// we remove things is very important to avoid invalidation issues. Don't
// mess with this unless you have good reason and know what you're doing.
// Tell ScalarEvolution that the loop is deleted. Do this before
// deleting the loop so that ScalarEvolution can look at the loop
// to determine what it needs to clean up.
SE.forgetLoop(L);
// Connect the preheader directly to the exit block.
TerminatorInst *TI = preheader->getTerminator();
TI->replaceUsesOfWith(L->getHeader(), exitBlock);
// Rewrite phis in the exit block to get their inputs from
// the preheader instead of the exiting block.
BasicBlock *exitingBlock = exitingBlocks[0];
BasicBlock::iterator BI = exitBlock->begin();
while (PHINode *P = dyn_cast<PHINode>(BI)) {
int j = P->getBasicBlockIndex(exitingBlock);
assert(j >= 0 && "Can't find exiting block in exit block's phi node!");
P->setIncomingBlock(j, preheader);
for (unsigned i = 1; i < exitingBlocks.size(); ++i)
P->removeIncomingValue(exitingBlocks[i]);
++BI;
}
// Update the dominator tree and remove the instructions and blocks that will
// be deleted from the reference counting scheme.
DominatorTree &DT = getAnalysis<DominatorTree>();
SmallVector<DomTreeNode*, 8> ChildNodes;
for (Loop::block_iterator LI = L->block_begin(), LE = L->block_end();
LI != LE; ++LI) {
// Move all of the block's children to be children of the preheader, which
// allows us to remove the domtree entry for the block.
ChildNodes.insert(ChildNodes.begin(), DT[*LI]->begin(), DT[*LI]->end());
for (SmallVectorImpl<DomTreeNode *>::iterator DI = ChildNodes.begin(),
DE = ChildNodes.end(); DI != DE; ++DI) {
DT.changeImmediateDominator(*DI, DT[preheader]);
}
ChildNodes.clear();
DT.eraseNode(*LI);
// Remove the block from the reference counting scheme, so that we can
// delete it freely later.
(*LI)->dropAllReferences();
}
// Erase the instructions and the blocks without having to worry
// about ordering because we already dropped the references.
// NOTE: This iteration is safe because erasing the block does not remove its
// entry from the loop's block list. We do that in the next section.
for (Loop::block_iterator LI = L->block_begin(), LE = L->block_end();
LI != LE; ++LI)
(*LI)->eraseFromParent();
// Finally, the blocks from loopinfo. This has to happen late because
// otherwise our loop iterators won't work.
LoopInfo &loopInfo = getAnalysis<LoopInfo>();
SmallPtrSet<BasicBlock*, 8> blocks;
blocks.insert(L->block_begin(), L->block_end());
for (SmallPtrSet<BasicBlock*,8>::iterator I = blocks.begin(),
E = blocks.end(); I != E; ++I)
loopInfo.removeBlock(*I);
// The last step is to inform the loop pass manager that we've
// eliminated this loop.
LPM.deleteLoopFromQueue(L);
Changed = true;
++NumDeleted;
return Changed;
}
// Sinks \p I from the loop \p L's preheader to its uses. Returns true if
// sinking is successful.
// \p LoopBlockNumber is used to sort the insertion blocks to ensure
// determinism.
static bool sinkInstruction(Loop &L, Instruction &I,
const SmallVectorImpl<BasicBlock *> &ColdLoopBBs,
const SmallDenseMap<BasicBlock *, int, 16> &LoopBlockNumber,
LoopInfo &LI, DominatorTree &DT,
BlockFrequencyInfo &BFI) {
// Compute the set of blocks in loop L which contain a use of I.
SmallPtrSet<BasicBlock *, 2> BBs;
for (auto &U : I.uses()) {
Instruction *UI = cast<Instruction>(U.getUser());
// We cannot sink I to PHI-uses.
if (dyn_cast<PHINode>(UI))
return false;
// We cannot sink I if it has uses outside of the loop.
if (!L.contains(LI.getLoopFor(UI->getParent())))
return false;
BBs.insert(UI->getParent());
}
// findBBsToSinkInto is O(BBs.size() * ColdLoopBBs.size()). We cap the max
// BBs.size() to avoid expensive computation.
// FIXME: Handle code size growth for min_size and opt_size.
if (BBs.size() > MaxNumberOfUseBBsForSinking)
return false;
// Find the set of BBs that we should insert a copy of I.
SmallPtrSet<BasicBlock *, 2> BBsToSinkInto =
findBBsToSinkInto(L, BBs, ColdLoopBBs, DT, BFI);
if (BBsToSinkInto.empty())
return false;
// Copy the final BBs into a vector and sort them using the total ordering
// of the loop block numbers as iterating the set doesn't give a useful
// order. No need to stable sort as the block numbers are a total ordering.
SmallVector<BasicBlock *, 2> SortedBBsToSinkInto;
SortedBBsToSinkInto.insert(SortedBBsToSinkInto.begin(), BBsToSinkInto.begin(),
BBsToSinkInto.end());
std::sort(SortedBBsToSinkInto.begin(), SortedBBsToSinkInto.end(),
[&](BasicBlock *A, BasicBlock *B) {
return *LoopBlockNumber.find(A) < *LoopBlockNumber.find(B);
});
BasicBlock *MoveBB = *SortedBBsToSinkInto.begin();
// FIXME: Optimize the efficiency for cloned value replacement. The current
// implementation is O(SortedBBsToSinkInto.size() * I.num_uses()).
for (BasicBlock *N : SortedBBsToSinkInto) {
if (N == MoveBB)
continue;
// Clone I and replace its uses.
Instruction *IC = I.clone();
IC->setName(I.getName());
IC->insertBefore(&*N->getFirstInsertionPt());
// Replaces uses of I with IC in N
for (Value::use_iterator UI = I.use_begin(), UE = I.use_end(); UI != UE;) {
Use &U = *UI++;
auto *I = cast<Instruction>(U.getUser());
if (I->getParent() == N)
U.set(IC);
}
// Replaces uses of I with IC in blocks dominated by N
replaceDominatedUsesWith(&I, IC, DT, N);
DEBUG(dbgs() << "Sinking a clone of " << I << " To: " << N->getName()
<< '\n');
NumLoopSunkCloned++;
}
DEBUG(dbgs() << "Sinking " << I << " To: " << MoveBB->getName() << '\n');
NumLoopSunk++;
I.moveBefore(&*MoveBB->getFirstInsertionPt());
return true;
}
/// ParseMicrosoftAsmStatement. When -fms-extensions/-fasm-blocks is enabled,
/// this routine is called to collect the tokens for an MS asm statement.
///
/// [MS] ms-asm-statement:
/// ms-asm-block
/// ms-asm-block ms-asm-statement
///
/// [MS] ms-asm-block:
/// '__asm' ms-asm-line '\n'
/// '__asm' '{' ms-asm-instruction-block[opt] '}' ';'[opt]
///
/// [MS] ms-asm-instruction-block
/// ms-asm-line
/// ms-asm-line '\n' ms-asm-instruction-block
///
StmtResult Parser::ParseMicrosoftAsmStatement(SourceLocation AsmLoc) {
SourceManager &SrcMgr = PP.getSourceManager();
SourceLocation EndLoc = AsmLoc;
SmallVector<Token, 4> AsmToks;
bool SingleLineMode = true;
unsigned BraceNesting = 0;
unsigned short savedBraceCount = BraceCount;
bool InAsmComment = false;
FileID FID;
unsigned LineNo = 0;
unsigned NumTokensRead = 0;
SmallVector<SourceLocation, 4> LBraceLocs;
bool SkippedStartOfLine = false;
if (Tok.is(tok::l_brace)) {
// Braced inline asm: consume the opening brace.
SingleLineMode = false;
BraceNesting = 1;
EndLoc = ConsumeBrace();
LBraceLocs.push_back(EndLoc);
++NumTokensRead;
} else {
// Single-line inline asm; compute which line it is on.
std::pair<FileID, unsigned> ExpAsmLoc =
SrcMgr.getDecomposedExpansionLoc(EndLoc);
FID = ExpAsmLoc.first;
LineNo = SrcMgr.getLineNumber(FID, ExpAsmLoc.second);
LBraceLocs.push_back(SourceLocation());
}
SourceLocation TokLoc = Tok.getLocation();
do {
// If we hit EOF, we're done, period.
if (isEofOrEom())
break;
if (!InAsmComment && Tok.is(tok::l_brace)) {
// Consume the opening brace.
SkippedStartOfLine = Tok.isAtStartOfLine();
EndLoc = ConsumeBrace();
BraceNesting++;
LBraceLocs.push_back(EndLoc);
TokLoc = Tok.getLocation();
++NumTokensRead;
continue;
} else if (!InAsmComment && Tok.is(tok::semi)) {
// A semicolon in an asm is the start of a comment.
InAsmComment = true;
if (!SingleLineMode) {
// Compute which line the comment is on.
std::pair<FileID, unsigned> ExpSemiLoc =
SrcMgr.getDecomposedExpansionLoc(TokLoc);
FID = ExpSemiLoc.first;
LineNo = SrcMgr.getLineNumber(FID, ExpSemiLoc.second);
}
} else if (SingleLineMode || InAsmComment) {
// If end-of-line is significant, check whether this token is on a
// new line.
std::pair<FileID, unsigned> ExpLoc =
SrcMgr.getDecomposedExpansionLoc(TokLoc);
if (ExpLoc.first != FID ||
SrcMgr.getLineNumber(ExpLoc.first, ExpLoc.second) != LineNo) {
// If this is a single-line __asm, we're done, except if the next
// line begins with an __asm too, in which case we finish a comment
// if needed and then keep processing the next line as a single
// line __asm.
bool isAsm = Tok.is(tok::kw_asm);
if (SingleLineMode && !isAsm)
break;
// We're no longer in a comment.
InAsmComment = false;
if (isAsm) {
LineNo = SrcMgr.getLineNumber(ExpLoc.first, ExpLoc.second);
SkippedStartOfLine = Tok.isAtStartOfLine();
}
} else if (!InAsmComment && Tok.is(tok::r_brace)) {
// In MSVC mode, braces only participate in brace matching and
// separating the asm statements. This is an intentional
// departure from the Apple gcc behavior.
if (!BraceNesting)
break;
}
}
if (!InAsmComment && BraceNesting && Tok.is(tok::r_brace) &&
BraceCount == (savedBraceCount + BraceNesting)) {
// Consume the closing brace.
SkippedStartOfLine = Tok.isAtStartOfLine();
EndLoc = ConsumeBrace();
BraceNesting--;
// Finish if all of the opened braces in the inline asm section were
// consumed.
if (BraceNesting == 0 && !SingleLineMode)
break;
else {
LBraceLocs.pop_back();
TokLoc = Tok.getLocation();
++NumTokensRead;
continue;
}
}
// Consume the next token; make sure we don't modify the brace count etc.
// if we are in a comment.
EndLoc = TokLoc;
if (InAsmComment)
PP.Lex(Tok);
else {
// Set the token as the start of line if we skipped the original start
// of line token in case it was a nested brace.
if (SkippedStartOfLine)
Tok.setFlag(Token::StartOfLine);
AsmToks.push_back(Tok);
ConsumeAnyToken();
}
TokLoc = Tok.getLocation();
++NumTokensRead;
SkippedStartOfLine = false;
} while (1);
if (BraceNesting && BraceCount != savedBraceCount) {
// __asm without closing brace (this can happen at EOF).
for (unsigned i = 0; i < BraceNesting; ++i) {
Diag(Tok, diag::err_expected) << tok::r_brace;
Diag(LBraceLocs.back(), diag::note_matching) << tok::l_brace;
LBraceLocs.pop_back();
}
return StmtError();
} else if (NumTokensRead == 0) {
// Empty __asm.
Diag(Tok, diag::err_expected) << tok::l_brace;
return StmtError();
}
// Okay, prepare to use MC to parse the assembly.
SmallVector<StringRef, 4> ConstraintRefs;
SmallVector<Expr *, 4> Exprs;
SmallVector<StringRef, 4> ClobberRefs;
// We need an actual supported target.
const llvm::Triple &TheTriple = Actions.Context.getTargetInfo().getTriple();
llvm::Triple::ArchType ArchTy = TheTriple.getArch();
const std::string &TT = TheTriple.getTriple();
const llvm::Target *TheTarget = nullptr;
bool UnsupportedArch =
(ArchTy != llvm::Triple::x86 && ArchTy != llvm::Triple::x86_64);
if (UnsupportedArch) {
Diag(AsmLoc, diag::err_msasm_unsupported_arch) << TheTriple.getArchName();
} else {
std::string Error;
TheTarget = llvm::TargetRegistry::lookupTarget(TT, Error);
if (!TheTarget)
Diag(AsmLoc, diag::err_msasm_unable_to_create_target) << Error;
}
assert(!LBraceLocs.empty() && "Should have at least one location here");
// If we don't support assembly, or the assembly is empty, we don't
// need to instantiate the AsmParser, etc.
if (!TheTarget || AsmToks.empty()) {
return Actions.ActOnMSAsmStmt(AsmLoc, LBraceLocs[0], AsmToks, StringRef(),
/*NumOutputs*/ 0, /*NumInputs*/ 0,
ConstraintRefs, ClobberRefs, Exprs, EndLoc);
}
// Expand the tokens into a string buffer.
SmallString<512> AsmString;
SmallVector<unsigned, 8> TokOffsets;
if (buildMSAsmString(PP, AsmLoc, AsmToks, TokOffsets, AsmString))
return StmtError();
std::unique_ptr<llvm::MCRegisterInfo> MRI(TheTarget->createMCRegInfo(TT));
std::unique_ptr<llvm::MCAsmInfo> MAI(TheTarget->createMCAsmInfo(*MRI, TT));
// Get the instruction descriptor.
std::unique_ptr<llvm::MCInstrInfo> MII(TheTarget->createMCInstrInfo());
std::unique_ptr<llvm::MCObjectFileInfo> MOFI(new llvm::MCObjectFileInfo());
std::unique_ptr<llvm::MCSubtargetInfo> STI(
TheTarget->createMCSubtargetInfo(TT, "", ""));
llvm::SourceMgr TempSrcMgr;
llvm::MCContext Ctx(MAI.get(), MRI.get(), MOFI.get(), &TempSrcMgr);
MOFI->InitMCObjectFileInfo(TheTriple, llvm::Reloc::Default,
llvm::CodeModel::Default, Ctx);
std::unique_ptr<llvm::MemoryBuffer> Buffer =
llvm::MemoryBuffer::getMemBuffer(AsmString, "<MS inline asm>");
// Tell SrcMgr about this buffer, which is what the parser will pick up.
TempSrcMgr.AddNewSourceBuffer(std::move(Buffer), llvm::SMLoc());
std::unique_ptr<llvm::MCStreamer> Str(createNullStreamer(Ctx));
std::unique_ptr<llvm::MCAsmParser> Parser(
createMCAsmParser(TempSrcMgr, Ctx, *Str.get(), *MAI));
// FIXME: init MCOptions from sanitizer flags here.
llvm::MCTargetOptions MCOptions;
std::unique_ptr<llvm::MCTargetAsmParser> TargetParser(
TheTarget->createMCAsmParser(*STI, *Parser, *MII, MCOptions));
std::unique_ptr<llvm::MCInstPrinter> IP(
TheTarget->createMCInstPrinter(llvm::Triple(TT), 1, *MAI, *MII, *MRI));
// Change to the Intel dialect.
Parser->setAssemblerDialect(1);
Parser->setTargetParser(*TargetParser.get());
Parser->setParsingInlineAsm(true);
TargetParser->setParsingInlineAsm(true);
ClangAsmParserCallback Callback(*this, AsmLoc, AsmString, AsmToks,
TokOffsets);
TargetParser->setSemaCallback(&Callback);
TempSrcMgr.setDiagHandler(ClangAsmParserCallback::DiagHandlerCallback,
&Callback);
unsigned NumOutputs;
unsigned NumInputs;
std::string AsmStringIR;
SmallVector<std::pair<void *, bool>, 4> OpExprs;
SmallVector<std::string, 4> Constraints;
SmallVector<std::string, 4> Clobbers;
if (Parser->parseMSInlineAsm(AsmLoc.getPtrEncoding(), AsmStringIR, NumOutputs,
NumInputs, OpExprs, Constraints, Clobbers,
MII.get(), IP.get(), Callback))
return StmtError();
// Filter out "fpsw". Clang doesn't accept it, and it always lists flags and
// fpsr as clobbers.
auto End = std::remove(Clobbers.begin(), Clobbers.end(), "fpsw");
Clobbers.erase(End, Clobbers.end());
// Build the vector of clobber StringRefs.
ClobberRefs.insert(ClobberRefs.end(), Clobbers.begin(), Clobbers.end());
// Recast the void pointers and build the vector of constraint StringRefs.
unsigned NumExprs = NumOutputs + NumInputs;
ConstraintRefs.resize(NumExprs);
Exprs.resize(NumExprs);
for (unsigned i = 0, e = NumExprs; i != e; ++i) {
Expr *OpExpr = static_cast<Expr *>(OpExprs[i].first);
if (!OpExpr)
return StmtError();
// Need address of variable.
if (OpExprs[i].second)
OpExpr =
Actions.BuildUnaryOp(getCurScope(), AsmLoc, UO_AddrOf, OpExpr).get();
ConstraintRefs[i] = StringRef(Constraints[i]);
Exprs[i] = OpExpr;
}
// FIXME: We should be passing source locations for better diagnostics.
return Actions.ActOnMSAsmStmt(AsmLoc, LBraceLocs[0], AsmToks, AsmStringIR,
NumOutputs, NumInputs, ConstraintRefs,
ClobberRefs, Exprs, EndLoc);
}
/// ComputeLiveInBlocks - Determine which blocks the value is live in. These
/// are blocks which lead to uses. Knowing this allows us to avoid inserting
/// PHI nodes into blocks which don't lead to uses (thus, the inserted phi nodes
/// would be dead).
void PromoteMem2Reg::
ComputeLiveInBlocks(AllocaInst *AI, AllocaInfo &Info,
const SmallPtrSet<BasicBlock*, 32> &DefBlocks,
SmallPtrSet<BasicBlock*, 32> &LiveInBlocks) {
// To determine liveness, we must iterate through the predecessors of blocks
// where the def is live. Blocks are added to the worklist if we need to
// check their predecessors. Start with all the using blocks.
SmallVector<BasicBlock*, 64> LiveInBlockWorklist;
LiveInBlockWorklist.insert(LiveInBlockWorklist.end(),
Info.UsingBlocks.begin(), Info.UsingBlocks.end());
// If any of the using blocks is also a definition block, check to see if the
// definition occurs before or after the use. If it happens before the use,
// the value isn't really live-in.
for (unsigned i = 0, e = LiveInBlockWorklist.size(); i != e; ++i) {
BasicBlock *BB = LiveInBlockWorklist[i];
if (!DefBlocks.count(BB)) continue;
// Okay, this is a block that both uses and defines the value. If the first
// reference to the alloca is a def (store), then we know it isn't live-in.
for (BasicBlock::iterator I = BB->begin(); ; ++I) {
if (StoreInst *SI = dyn_cast<StoreInst>(I)) {
if (SI->getOperand(1) != AI) continue;
// We found a store to the alloca before a load. The alloca is not
// actually live-in here.
LiveInBlockWorklist[i] = LiveInBlockWorklist.back();
LiveInBlockWorklist.pop_back();
--i, --e;
break;
}
if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
if (LI->getOperand(0) != AI) continue;
// Okay, we found a load before a store to the alloca. It is actually
// live into this block.
break;
}
}
}
// Now that we have a set of blocks where the phi is live-in, recursively add
// their predecessors until we find the full region the value is live.
while (!LiveInBlockWorklist.empty()) {
BasicBlock *BB = LiveInBlockWorklist.pop_back_val();
// The block really is live in here, insert it into the set. If already in
// the set, then it has already been processed.
if (!LiveInBlocks.insert(BB))
continue;
// Since the value is live into BB, it is either defined in a predecessor or
// live into it to. Add the preds to the worklist unless they are a
// defining block.
for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
BasicBlock *P = *PI;
// The value is not live into a predecessor if it defines the value.
if (DefBlocks.count(P))
continue;
// Otherwise it is, add to the worklist.
LiveInBlockWorklist.push_back(P);
}
}
}
void swift::ide::printSubmoduleInterface(
Module *M,
ArrayRef<StringRef> FullModuleName,
Optional<StringRef> GroupName,
ModuleTraversalOptions TraversalOptions,
ASTPrinter &Printer,
const PrintOptions &Options,
const bool PrintSynthesizedExtensions) {
auto AdjustedOptions = Options;
adjustPrintOptions(AdjustedOptions);
SmallVector<Decl *, 1> Decls;
M->getDisplayDecls(Decls);
auto &SwiftContext = M->getASTContext();
auto &Importer =
static_cast<ClangImporter &>(*SwiftContext.getClangModuleLoader());
const clang::Module *InterestingClangModule = nullptr;
SmallVector<ImportDecl *, 1> ImportDecls;
llvm::DenseSet<const clang::Module *> ClangModulesForImports;
SmallVector<Decl *, 1> SwiftDecls;
llvm::DenseMap<const clang::Module *,
SmallVector<std::pair<Decl *, clang::SourceLocation>, 1>>
ClangDecls;
// Drop top-level module name.
FullModuleName = FullModuleName.slice(1);
InterestingClangModule = M->findUnderlyingClangModule();
if (InterestingClangModule) {
for (StringRef Name : FullModuleName) {
InterestingClangModule = InterestingClangModule->findSubmodule(Name);
if (!InterestingClangModule)
return;
}
} else {
assert(FullModuleName.empty());
}
// If we're printing recursively, find all of the submodules to print.
if (InterestingClangModule) {
if (TraversalOptions) {
SmallVector<const clang::Module *, 8> Worklist;
SmallPtrSet<const clang::Module *, 8> Visited;
Worklist.push_back(InterestingClangModule);
Visited.insert(InterestingClangModule);
while (!Worklist.empty()) {
const clang::Module *CM = Worklist.pop_back_val();
if (!(TraversalOptions & ModuleTraversal::VisitHidden) &&
CM->IsExplicit)
continue;
ClangDecls.insert({ CM, {} });
// If we're supposed to visit submodules, add them now.
if (TraversalOptions & ModuleTraversal::VisitSubmodules) {
for (auto Sub = CM->submodule_begin(), SubEnd = CM->submodule_end();
Sub != SubEnd; ++Sub) {
if (Visited.insert(*Sub).second)
Worklist.push_back(*Sub);
}
}
}
} else {
ClangDecls.insert({ InterestingClangModule, {} });
}
}
// Collect those submodules that are actually imported but have no import decls
// in the module.
llvm::SmallPtrSet<const clang::Module *, 16> NoImportSubModules;
if (InterestingClangModule) {
// Assume all submodules are missing.
for (auto It =InterestingClangModule->submodule_begin();
It != InterestingClangModule->submodule_end(); It ++) {
NoImportSubModules.insert(*It);
}
}
// Separate the declarations that we are going to print into different
// buckets.
for (Decl *D : Decls) {
// Skip declarations that are not accessible.
if (auto *VD = dyn_cast<ValueDecl>(D)) {
if (Options.AccessibilityFilter > Accessibility::Private &&
VD->hasAccessibility() &&
VD->getFormalAccess() < Options.AccessibilityFilter)
continue;
}
auto ShouldPrintImport = [&](ImportDecl *ImportD) -> bool {
if (!InterestingClangModule)
return true;
auto ClangMod = ImportD->getClangModule();
if (!ClangMod)
return true;
if (!ClangMod->isSubModule())
return true;
if (ClangMod == InterestingClangModule)
return false;
// FIXME: const-ness on the clang API.
return ClangMod->isSubModuleOf(
const_cast<clang::Module*>(InterestingClangModule));
};
if (auto ID = dyn_cast<ImportDecl>(D)) {
if (ShouldPrintImport(ID)) {
if (ID->getClangModule())
// Erase those submodules that are not missing.
NoImportSubModules.erase(ID->getClangModule());
if (ID->getImportKind() == ImportKind::Module) {
// Make sure we don't print duplicate imports, due to getting imports
// for both a clang module and its overlay.
if (auto *ClangMod = getUnderlyingClangModuleForImport(ID)) {
auto P = ClangModulesForImports.insert(ClangMod);
bool IsNew = P.second;
if (!IsNew)
continue;
}
}
ImportDecls.push_back(ID);
}
continue;
}
auto addToClangDecls = [&](Decl *D) {
assert(D->hasClangNode());
auto CN = D->getClangNode();
clang::SourceLocation Loc = CN.getLocation();
auto *OwningModule = Importer.getClangOwningModule(CN);
auto I = ClangDecls.find(OwningModule);
if (I != ClangDecls.end()) {
I->second.push_back({ D, Loc });
}
};
if (D->hasClangNode()) {
addToClangDecls(D);
continue;
}
if (FullModuleName.empty()) {
// If group name is given and the decl does not belong to the group, skip it.
if (GroupName && (!D->getGroupName() ||
D->getGroupName().getValue() != GroupName.getValue()))
continue;
// Add Swift decls if we are printing the top-level module.
SwiftDecls.push_back(D);
}
}
// Create the missing import decls and add to the collector.
for (auto *SM : NoImportSubModules) {
ImportDecls.push_back(createImportDecl(M->getASTContext(), M, SM, {}));
}
auto &ClangSourceManager = Importer.getClangASTContext().getSourceManager();
// Sort imported declarations in source order *within a submodule*.
for (auto &P : ClangDecls) {
std::sort(P.second.begin(), P.second.end(),
[&](std::pair<Decl *, clang::SourceLocation> LHS,
std::pair<Decl *, clang::SourceLocation> RHS) -> bool {
return ClangSourceManager.isBeforeInTranslationUnit(LHS.second,
RHS.second);
});
}
// Sort Swift declarations so that we print them in a consistent order.
std::sort(ImportDecls.begin(), ImportDecls.end(),
[](ImportDecl *LHS, ImportDecl *RHS) -> bool {
auto LHSPath = LHS->getFullAccessPath();
auto RHSPath = RHS->getFullAccessPath();
for (unsigned i = 0, e = std::min(LHSPath.size(), RHSPath.size()); i != e;
i++) {
if (int Ret = LHSPath[i].first.str().compare(RHSPath[i].first.str()))
return Ret < 0;
}
return false;
});
// If the group name is specified, we sort them according to their source order,
// which is the order preserved by getTopLeveDecls.
if (!GroupName) {
std::sort(SwiftDecls.begin(), SwiftDecls.end(),
[&](Decl *LHS, Decl *RHS) -> bool {
auto *LHSValue = dyn_cast<ValueDecl>(LHS);
auto *RHSValue = dyn_cast<ValueDecl>(RHS);
if (LHSValue && RHSValue) {
StringRef LHSName = LHSValue->getName().str();
StringRef RHSName = RHSValue->getName().str();
if (int Ret = LHSName.compare(RHSName))
return Ret < 0;
// FIXME: this is not sufficient to establish a total order for overloaded
// decls.
return LHS->getKind() < RHS->getKind();
}
return LHS->getKind() < RHS->getKind();
});
}
ASTPrinter *PrinterToUse = &Printer;
ClangCommentPrinter RegularCommentPrinter(Printer, Importer);
if (Options.PrintRegularClangComments)
PrinterToUse = &RegularCommentPrinter;
auto PrintDecl = [&](Decl *D) -> bool {
ASTPrinter &Printer = *PrinterToUse;
if (!shouldPrint(D, AdjustedOptions)) {
Printer.avoidPrintDeclPost(D);
return false;
}
if (auto Ext = dyn_cast<ExtensionDecl>(D)) {
// Clang extensions (categories) are always printed in source order.
// Swift extensions are printed with their associated type unless it's
// a cross-module extension.
if (!Ext->hasClangNode()) {
auto ExtendedNominal = Ext->getExtendedType()->getAnyNominal();
if (Ext->getModuleContext() == ExtendedNominal->getModuleContext())
return false;
}
}
if (D->print(Printer, AdjustedOptions)) {
Printer << "\n";
if (auto NTD = dyn_cast<NominalTypeDecl>(D)) {
std::queue<NominalTypeDecl *> SubDecls{{NTD}};
while (!SubDecls.empty()) {
auto NTD = SubDecls.front();
SubDecls.pop();
// Add sub-types of NTD.
for (auto Sub : NTD->getMembers())
if (auto N = dyn_cast<NominalTypeDecl>(Sub))
SubDecls.push(N);
// Print Ext and add sub-types of Ext.
for (auto Ext : NTD->getExtensions()) {
if (!shouldPrint(Ext, AdjustedOptions)) {
Printer.avoidPrintDeclPost(Ext);
continue;
}
if (Ext->hasClangNode())
continue; // will be printed in its source location, see above.
Printer << "\n";
Ext->print(Printer, AdjustedOptions);
Printer << "\n";
for (auto Sub : Ext->getMembers())
if (auto N = dyn_cast<NominalTypeDecl>(Sub))
SubDecls.push(N);
}
if (!PrintSynthesizedExtensions)
continue;
// Print synthesized extensions.
llvm::SmallPtrSet<ExtensionDecl *, 10> ExtensionsFromConformances;
findExtensionsFromConformingProtocols(D, ExtensionsFromConformances);
AdjustedOptions.initArchetypeTransformerForSynthesizedExtensions(NTD);
for (auto ET : ExtensionsFromConformances) {
if (!shouldPrint(ET, AdjustedOptions))
continue;
Printer << "\n";
Printer << "/// Synthesized extension from ";
ET->getExtendedTypeLoc().getType().print(Printer, AdjustedOptions);
Printer << "\n";
ET->print(Printer, AdjustedOptions);
Printer << "\n";
}
AdjustedOptions.clearArchetypeTransformerForSynthesizedExtensions();
}
}
return true;
}
return false;
};
// Imports from the stdlib are internal details that don't need to be exposed.
if (!M->isStdlibModule()) {
for (auto *D : ImportDecls)
PrintDecl(D);
Printer << "\n";
}
{
using ModuleAndName = std::pair<const clang::Module *, std::string>;
SmallVector<ModuleAndName, 8> ClangModules;
for (auto P : ClangDecls) {
ClangModules.push_back({ P.first, P.first->getFullModuleName() });
}
// Sort modules by name.
std::sort(ClangModules.begin(), ClangModules.end(),
[](const ModuleAndName &LHS, const ModuleAndName &RHS)
-> bool {
return LHS.second < RHS.second;
});
for (auto CM : ClangModules) {
for (auto DeclAndLoc : ClangDecls[CM.first])
PrintDecl(DeclAndLoc.first);
}
}
if (!(TraversalOptions & ModuleTraversal::SkipOverlay) ||
!InterestingClangModule) {
for (auto *D : SwiftDecls) {
if (PrintDecl(D))
Printer << "\n";
}
}
}
int main(int argc_, const char **argv_) {
llvm::sys::PrintStackTraceOnErrorSignal(argv_[0]);
llvm::PrettyStackTraceProgram X(argc_, argv_);
llvm::llvm_shutdown_obj Y; // Call llvm_shutdown() on exit.
if (llvm::sys::Process::FixupStandardFileDescriptors())
return 1;
SmallVector<const char *, 256> argv;
llvm::SpecificBumpPtrAllocator<char> ArgAllocator;
std::error_code EC = llvm::sys::Process::GetArgumentVector(
argv, llvm::makeArrayRef(argv_, argc_), ArgAllocator);
if (EC) {
llvm::errs() << "error: couldn't get arguments: " << EC.message() << '\n';
return 1;
}
llvm::InitializeAllTargets();
std::string ProgName = argv[0];
std::pair<std::string, std::string> TargetAndMode =
ToolChain::getTargetAndModeFromProgramName(ProgName);
llvm::BumpPtrAllocator A;
llvm::StringSaver Saver(A);
// Parse response files using the GNU syntax, unless we're in CL mode. There
// are two ways to put clang in CL compatibility mode: argv[0] is either
// clang-cl or cl, or --driver-mode=cl is on the command line. The normal
// command line parsing can't happen until after response file parsing, so we
// have to manually search for a --driver-mode=cl argument the hard way.
// Finally, our -cc1 tools don't care which tokenization mode we use because
// response files written by clang will tokenize the same way in either mode.
bool ClangCLMode = false;
if (TargetAndMode.second == "--driver-mode=cl" ||
std::find_if(argv.begin(), argv.end(), [](const char *F) {
return F && strcmp(F, "--driver-mode=cl") == 0;
}) != argv.end()) {
ClangCLMode = true;
}
enum { Default, POSIX, Windows } RSPQuoting = Default;
for (const char *F : argv) {
if (strcmp(F, "--rsp-quoting=posix") == 0)
RSPQuoting = POSIX;
else if (strcmp(F, "--rsp-quoting=windows") == 0)
RSPQuoting = Windows;
}
// Determines whether we want nullptr markers in argv to indicate response
// files end-of-lines. We only use this for the /LINK driver argument with
// clang-cl.exe on Windows.
bool MarkEOLs = ClangCLMode;
llvm::cl::TokenizerCallback Tokenizer;
if (RSPQuoting == Windows || (RSPQuoting == Default && ClangCLMode))
Tokenizer = &llvm::cl::TokenizeWindowsCommandLine;
else
Tokenizer = &llvm::cl::TokenizeGNUCommandLine;
if (MarkEOLs && argv.size() > 1 && StringRef(argv[1]).startswith("-cc1"))
MarkEOLs = false;
llvm::cl::ExpandResponseFiles(Saver, Tokenizer, argv, MarkEOLs);
// Handle -cc1 integrated tools, even if -cc1 was expanded from a response
// file.
auto FirstArg = std::find_if(argv.begin() + 1, argv.end(),
[](const char *A) { return A != nullptr; });
if (FirstArg != argv.end() && StringRef(*FirstArg).startswith("-cc1")) {
// If -cc1 came from a response file, remove the EOL sentinels.
if (MarkEOLs) {
auto newEnd = std::remove(argv.begin(), argv.end(), nullptr);
argv.resize(newEnd - argv.begin());
}
return ExecuteCC1Tool(argv, argv[1] + 4);
}
bool CanonicalPrefixes = true;
for (int i = 1, size = argv.size(); i < size; ++i) {
// Skip end-of-line response file markers
if (argv[i] == nullptr)
continue;
if (StringRef(argv[i]) == "-no-canonical-prefixes") {
CanonicalPrefixes = false;
break;
}
}
// Handle CL and _CL_ which permits additional command line options to be
// prepended or appended.
if (ClangCLMode) {
// Arguments in "CL" are prepended.
llvm::Optional<std::string> OptCL = llvm::sys::Process::GetEnv("CL");
if (OptCL.hasValue()) {
SmallVector<const char *, 8> PrependedOpts;
getCLEnvVarOptions(OptCL.getValue(), Saver, PrependedOpts);
// Insert right after the program name to prepend to the argument list.
argv.insert(argv.begin() + 1, PrependedOpts.begin(), PrependedOpts.end());
}
// Arguments in "_CL_" are appended.
llvm::Optional<std::string> Opt_CL_ = llvm::sys::Process::GetEnv("_CL_");
if (Opt_CL_.hasValue()) {
SmallVector<const char *, 8> AppendedOpts;
getCLEnvVarOptions(Opt_CL_.getValue(), Saver, AppendedOpts);
// Insert at the end of the argument list to append.
argv.append(AppendedOpts.begin(), AppendedOpts.end());
}
}
std::set<std::string> SavedStrings;
// Handle CCC_OVERRIDE_OPTIONS, used for editing a command line behind the
// scenes.
if (const char *OverrideStr = ::getenv("CCC_OVERRIDE_OPTIONS")) {
// FIXME: Driver shouldn't take extra initial argument.
ApplyQAOverride(argv, OverrideStr, SavedStrings);
}
std::string Path = GetExecutablePath(argv[0], CanonicalPrefixes);
IntrusiveRefCntPtr<DiagnosticOptions> DiagOpts =
CreateAndPopulateDiagOpts(argv);
TextDiagnosticPrinter *DiagClient
= new TextDiagnosticPrinter(llvm::errs(), &*DiagOpts);
FixupDiagPrefixExeName(DiagClient, Path);
IntrusiveRefCntPtr<DiagnosticIDs> DiagID(new DiagnosticIDs());
DiagnosticsEngine Diags(DiagID, &*DiagOpts, DiagClient);
if (!DiagOpts->DiagnosticSerializationFile.empty()) {
auto SerializedConsumer =
clang::serialized_diags::create(DiagOpts->DiagnosticSerializationFile,
&*DiagOpts, /*MergeChildRecords=*/true);
Diags.setClient(new ChainedDiagnosticConsumer(
Diags.takeClient(), std::move(SerializedConsumer)));
}
ProcessWarningOptions(Diags, *DiagOpts, /*ReportDiags=*/false);
Driver TheDriver(Path, llvm::sys::getDefaultTargetTriple(), Diags);
SetInstallDir(argv, TheDriver, CanonicalPrefixes);
insertTargetAndModeArgs(TargetAndMode.first, TargetAndMode.second, argv,
SavedStrings);
SetBackdoorDriverOutputsFromEnvVars(TheDriver);
std::unique_ptr<Compilation> C(TheDriver.BuildCompilation(argv));
int Res = 0;
SmallVector<std::pair<int, const Command *>, 4> FailingCommands;
if (C.get())
Res = TheDriver.ExecuteCompilation(*C, FailingCommands);
// Force a crash to test the diagnostics.
if (TheDriver.GenReproducer) {
Diags.Report(diag::err_drv_force_crash)
<< !::getenv("FORCE_CLANG_DIAGNOSTICS_CRASH");
// Pretend that every command failed.
FailingCommands.clear();
for (const auto &J : C->getJobs())
if (const Command *C = dyn_cast<Command>(&J))
FailingCommands.push_back(std::make_pair(-1, C));
}
for (const auto &P : FailingCommands) {
int CommandRes = P.first;
const Command *FailingCommand = P.second;
if (!Res)
Res = CommandRes;
// If result status is < 0, then the driver command signalled an error.
// If result status is 70, then the driver command reported a fatal error.
// On Windows, abort will return an exit code of 3. In these cases,
// generate additional diagnostic information if possible.
bool DiagnoseCrash = CommandRes < 0 || CommandRes == 70;
#ifdef LLVM_ON_WIN32
DiagnoseCrash |= CommandRes == 3;
#endif
if (DiagnoseCrash) {
TheDriver.generateCompilationDiagnostics(*C, *FailingCommand);
break;
}
}
Diags.getClient()->finish();
// If any timers were active but haven't been destroyed yet, print their
// results now. This happens in -disable-free mode.
llvm::TimerGroup::printAll(llvm::errs());
#ifdef LLVM_ON_WIN32
// Exit status should not be negative on Win32, unless abnormal termination.
// Once abnormal termiation was caught, negative status should not be
// propagated.
if (Res < 0)
Res = 1;
#endif
// If we have multiple failing commands, we return the result of the first
// failing command.
return Res;
}
void SelectionDAGBuilder::LowerStatepoint(
ImmutableStatepoint ISP, MachineBasicBlock *LandingPad /*=nullptr*/) {
// The basic scheme here is that information about both the original call and
// the safepoint is encoded in the CallInst. We create a temporary call and
// lower it, then reverse engineer the calling sequence.
NumOfStatepoints++;
// Clear state
StatepointLowering.startNewStatepoint(*this);
ImmutableCallSite CS(ISP.getCallSite());
#ifndef NDEBUG
// Consistency check. Don't do this for invokes. It would be too
// expensive to preserve this information across different basic blocks
if (!CS.isInvoke()) {
for (const User *U : CS->users()) {
const CallInst *Call = cast<CallInst>(U);
if (isGCRelocate(Call))
StatepointLowering.scheduleRelocCall(*Call);
}
}
#endif
#ifndef NDEBUG
// If this is a malformed statepoint, report it early to simplify debugging.
// This should catch any IR level mistake that's made when constructing or
// transforming statepoints.
ISP.verify();
// Check that the associated GCStrategy expects to encounter statepoints.
assert(GFI->getStrategy().useStatepoints() &&
"GCStrategy does not expect to encounter statepoints");
#endif
// Lower statepoint vmstate and gcstate arguments
SmallVector<SDValue, 10> LoweredMetaArgs;
lowerStatepointMetaArgs(LoweredMetaArgs, ISP, *this);
// Get call node, we will replace it later with statepoint
SDNode *CallNode =
lowerCallFromStatepoint(ISP, LandingPad, *this, PendingExports);
// Construct the actual GC_TRANSITION_START, STATEPOINT, and GC_TRANSITION_END
// nodes with all the appropriate arguments and return values.
// Call Node: Chain, Target, {Args}, RegMask, [Glue]
SDValue Chain = CallNode->getOperand(0);
SDValue Glue;
bool CallHasIncomingGlue = CallNode->getGluedNode();
if (CallHasIncomingGlue) {
// Glue is always last operand
Glue = CallNode->getOperand(CallNode->getNumOperands() - 1);
}
// Build the GC_TRANSITION_START node if necessary.
//
// The operands to the GC_TRANSITION_{START,END} nodes are laid out in the
// order in which they appear in the call to the statepoint intrinsic. If
// any of the operands is a pointer-typed, that operand is immediately
// followed by a SRCVALUE for the pointer that may be used during lowering
// (e.g. to form MachinePointerInfo values for loads/stores).
const bool IsGCTransition =
(ISP.getFlags() & (uint64_t)StatepointFlags::GCTransition) ==
(uint64_t)StatepointFlags::GCTransition;
if (IsGCTransition) {
SmallVector<SDValue, 8> TSOps;
// Add chain
TSOps.push_back(Chain);
// Add GC transition arguments
for (const Value *V : ISP.gc_transition_args()) {
TSOps.push_back(getValue(V));
if (V->getType()->isPointerTy())
TSOps.push_back(DAG.getSrcValue(V));
}
// Add glue if necessary
if (CallHasIncomingGlue)
TSOps.push_back(Glue);
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue GCTransitionStart =
DAG.getNode(ISD::GC_TRANSITION_START, getCurSDLoc(), NodeTys, TSOps);
Chain = GCTransitionStart.getValue(0);
Glue = GCTransitionStart.getValue(1);
}
// TODO: Currently, all of these operands are being marked as read/write in
// PrologEpilougeInserter.cpp, we should special case the VMState arguments
// and flags to be read-only.
SmallVector<SDValue, 40> Ops;
// Add the <id> and <numBytes> constants.
Ops.push_back(DAG.getTargetConstant(ISP.getID(), getCurSDLoc(), MVT::i64));
Ops.push_back(
DAG.getTargetConstant(ISP.getNumPatchBytes(), getCurSDLoc(), MVT::i32));
// Calculate and push starting position of vmstate arguments
// Get number of arguments incoming directly into call node
unsigned NumCallRegArgs =
CallNode->getNumOperands() - (CallHasIncomingGlue ? 4 : 3);
Ops.push_back(DAG.getTargetConstant(NumCallRegArgs, getCurSDLoc(), MVT::i32));
// Add call target
SDValue CallTarget = SDValue(CallNode->getOperand(1).getNode(), 0);
Ops.push_back(CallTarget);
// Add call arguments
// Get position of register mask in the call
SDNode::op_iterator RegMaskIt;
if (CallHasIncomingGlue)
RegMaskIt = CallNode->op_end() - 2;
else
RegMaskIt = CallNode->op_end() - 1;
Ops.insert(Ops.end(), CallNode->op_begin() + 2, RegMaskIt);
// Add a constant argument for the calling convention
pushStackMapConstant(Ops, *this, CS.getCallingConv());
// Add a constant argument for the flags
uint64_t Flags = ISP.getFlags();
assert(
((Flags & ~(uint64_t)StatepointFlags::MaskAll) == 0)
&& "unknown flag used");
pushStackMapConstant(Ops, *this, Flags);
// Insert all vmstate and gcstate arguments
Ops.insert(Ops.end(), LoweredMetaArgs.begin(), LoweredMetaArgs.end());
// Add register mask from call node
Ops.push_back(*RegMaskIt);
// Add chain
Ops.push_back(Chain);
// Same for the glue, but we add it only if original call had it
if (Glue.getNode())
Ops.push_back(Glue);
// Compute return values. Provide a glue output since we consume one as
// input. This allows someone else to chain off us as needed.
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
SDNode *StatepointMCNode =
DAG.getMachineNode(TargetOpcode::STATEPOINT, getCurSDLoc(), NodeTys, Ops);
SDNode *SinkNode = StatepointMCNode;
// Build the GC_TRANSITION_END node if necessary.
//
// See the comment above regarding GC_TRANSITION_START for the layout of
// the operands to the GC_TRANSITION_END node.
if (IsGCTransition) {
SmallVector<SDValue, 8> TEOps;
// Add chain
TEOps.push_back(SDValue(StatepointMCNode, 0));
// Add GC transition arguments
for (const Value *V : ISP.gc_transition_args()) {
TEOps.push_back(getValue(V));
if (V->getType()->isPointerTy())
TEOps.push_back(DAG.getSrcValue(V));
}
// Add glue
TEOps.push_back(SDValue(StatepointMCNode, 1));
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue GCTransitionStart =
DAG.getNode(ISD::GC_TRANSITION_END, getCurSDLoc(), NodeTys, TEOps);
SinkNode = GCTransitionStart.getNode();
}
// Replace original call
DAG.ReplaceAllUsesWith(CallNode, SinkNode); // This may update Root
// Remove originall call node
DAG.DeleteNode(CallNode);
// DON'T set the root - under the assumption that it's already set past the
// inserted node we created.
// TODO: A better future implementation would be to emit a single variable
// argument, variable return value STATEPOINT node here and then hookup the
// return value of each gc.relocate to the respective output of the
// previously emitted STATEPOINT value. Unfortunately, this doesn't appear
// to actually be possible today.
}
int main(int argc_, const char **argv_) {
llvm::sys::PrintStackTraceOnErrorSignal();
llvm::PrettyStackTraceProgram X(argc_, argv_);
std::set<std::string> SavedStrings;
SmallVector<const char*, 256> argv;
ExpandArgv(argc_, argv_, argv, SavedStrings);
// Handle -cc1 integrated tools.
if (argv.size() > 1 && StringRef(argv[1]).startswith("-cc1")) {
StringRef Tool = argv[1] + 4;
if (Tool == "")
return cc1_main(argv.data()+2, argv.data()+argv.size(), argv[0],
(void*) (intptr_t) GetExecutablePath);
if (Tool == "as")
return cc1as_main(argv.data()+2, argv.data()+argv.size(), argv[0],
(void*) (intptr_t) GetExecutablePath);
// Reject unknown tools.
llvm::errs() << "error: unknown integrated tool '" << Tool << "'\n";
return 1;
}
bool CanonicalPrefixes = true;
for (int i = 1, size = argv.size(); i < size; ++i) {
if (StringRef(argv[i]) == "-no-canonical-prefixes") {
CanonicalPrefixes = false;
break;
}
}
llvm::sys::Path Path = GetExecutablePath(argv[0], CanonicalPrefixes);
IntrusiveRefCntPtr<DiagnosticOptions> DiagOpts = new DiagnosticOptions;
{
// Note that ParseDiagnosticArgs() uses the cc1 option table.
OwningPtr<OptTable> CC1Opts(createDriverOptTable());
unsigned MissingArgIndex, MissingArgCount;
OwningPtr<InputArgList> Args(CC1Opts->ParseArgs(argv.begin()+1, argv.end(),
MissingArgIndex, MissingArgCount));
// We ignore MissingArgCount and the return value of ParseDiagnosticArgs.
// Any errors that would be diagnosed here will also be diagnosed later,
// when the DiagnosticsEngine actually exists.
(void) ParseDiagnosticArgs(*DiagOpts, *Args);
}
// Now we can create the DiagnosticsEngine with a properly-filled-out
// DiagnosticOptions instance.
TextDiagnosticPrinter *DiagClient
= new TextDiagnosticPrinter(llvm::errs(), &*DiagOpts);
DiagClient->setPrefix(llvm::sys::path::stem(Path.str()));
IntrusiveRefCntPtr<DiagnosticIDs> DiagID(new DiagnosticIDs());
DiagnosticsEngine Diags(DiagID, &*DiagOpts, DiagClient);
ProcessWarningOptions(Diags, *DiagOpts);
Driver TheDriver(Path.str(), llvm::sys::getDefaultTargetTriple(),
"a.out", Diags);
// Attempt to find the original path used to invoke the driver, to determine
// the installed path. We do this manually, because we want to support that
// path being a symlink.
{
SmallString<128> InstalledPath(argv[0]);
// Do a PATH lookup, if there are no directory components.
if (llvm::sys::path::filename(InstalledPath) == InstalledPath) {
llvm::sys::Path Tmp = llvm::sys::Program::FindProgramByName(
llvm::sys::path::filename(InstalledPath.str()));
if (!Tmp.empty())
InstalledPath = Tmp.str();
}
llvm::sys::fs::make_absolute(InstalledPath);
InstalledPath = llvm::sys::path::parent_path(InstalledPath);
bool exists;
if (!llvm::sys::fs::exists(InstalledPath.str(), exists) && exists)
TheDriver.setInstalledDir(InstalledPath);
}
llvm::InitializeAllTargets();
ParseProgName(argv, SavedStrings, TheDriver);
// Handle CC_PRINT_OPTIONS and CC_PRINT_OPTIONS_FILE.
TheDriver.CCPrintOptions = !!::getenv("CC_PRINT_OPTIONS");
if (TheDriver.CCPrintOptions)
TheDriver.CCPrintOptionsFilename = ::getenv("CC_PRINT_OPTIONS_FILE");
// Handle CC_PRINT_HEADERS and CC_PRINT_HEADERS_FILE.
TheDriver.CCPrintHeaders = !!::getenv("CC_PRINT_HEADERS");
if (TheDriver.CCPrintHeaders)
TheDriver.CCPrintHeadersFilename = ::getenv("CC_PRINT_HEADERS_FILE");
// Handle CC_LOG_DIAGNOSTICS and CC_LOG_DIAGNOSTICS_FILE.
TheDriver.CCLogDiagnostics = !!::getenv("CC_LOG_DIAGNOSTICS");
if (TheDriver.CCLogDiagnostics)
TheDriver.CCLogDiagnosticsFilename = ::getenv("CC_LOG_DIAGNOSTICS_FILE");
// Handle QA_OVERRIDE_GCC3_OPTIONS and CCC_ADD_ARGS, used for editing a
// command line behind the scenes.
if (const char *OverrideStr = ::getenv("QA_OVERRIDE_GCC3_OPTIONS")) {
// FIXME: Driver shouldn't take extra initial argument.
ApplyQAOverride(argv, OverrideStr, SavedStrings);
} else if (const char *Cur = ::getenv("CCC_ADD_ARGS")) {
// FIXME: Driver shouldn't take extra initial argument.
std::vector<const char*> ExtraArgs;
for (;;) {
const char *Next = strchr(Cur, ',');
if (Next) {
ExtraArgs.push_back(SaveStringInSet(SavedStrings,
std::string(Cur, Next)));
Cur = Next + 1;
} else {
if (*Cur != '\0')
ExtraArgs.push_back(SaveStringInSet(SavedStrings, Cur));
break;
}
}
argv.insert(&argv[1], ExtraArgs.begin(), ExtraArgs.end());
}
OwningPtr<Compilation> C(TheDriver.BuildCompilation(argv));
int Res = 0;
const Command *FailingCommand = 0;
if (C.get())
Res = TheDriver.ExecuteCompilation(*C, FailingCommand);
// Force a crash to test the diagnostics.
if(::getenv("FORCE_CLANG_DIAGNOSTICS_CRASH"))
Res = -1;
// If result status is < 0, then the driver command signalled an error.
// If result status is 70, then the driver command reported a fatal error.
// In these cases, generate additional diagnostic information if possible.
if (Res < 0 || Res == 70)
TheDriver.generateCompilationDiagnostics(*C, FailingCommand);
// If any timers were active but haven't been destroyed yet, print their
// results now. This happens in -disable-free mode.
llvm::TimerGroup::printAll(llvm::errs());
llvm::llvm_shutdown();
#ifdef _WIN32
// Exit status should not be negative on Win32, unless abnormal termination.
// Once abnormal termiation was caught, negative status should not be
// propagated.
if (Res < 0)
Res = 1;
#endif
return Res;
}
int main(int argc_, const char **argv_) {
llvm::sys::PrintStackTraceOnErrorSignal();
llvm::PrettyStackTraceProgram X(argc_, argv_);
std::set<std::string> SavedStrings;
SmallVector<const char*, 256> argv;
ExpandArgv(argc_, argv_, argv, SavedStrings);
// Handle -cc1 integrated tools.
if (argv.size() > 1 && StringRef(argv[1]).startswith("-cc1")) {
StringRef Tool = argv[1] + 4;
if (Tool == "")
return cc1_main(argv.data()+2, argv.data()+argv.size(), argv[0],
(void*) (intptr_t) GetExecutablePath);
if (Tool == "as")
return cc1as_main(argv.data()+2, argv.data()+argv.size(), argv[0],
(void*) (intptr_t) GetExecutablePath);
// Reject unknown tools.
llvm::errs() << "error: unknown integrated tool '" << Tool << "'\n";
return 1;
}
bool CanonicalPrefixes = true;
for (int i = 1, size = argv.size(); i < size; ++i) {
if (StringRef(argv[i]) == "-no-canonical-prefixes") {
CanonicalPrefixes = false;
break;
}
}
llvm::sys::Path Path = GetExecutablePath(argv[0], CanonicalPrefixes);
TextDiagnosticPrinter *DiagClient
= new TextDiagnosticPrinter(llvm::errs(), DiagnosticOptions());
DiagClient->setPrefix(llvm::sys::path::stem(Path.str()));
llvm::IntrusiveRefCntPtr<DiagnosticIDs> DiagID(new DiagnosticIDs());
DiagnosticsEngine Diags(DiagID, DiagClient);
#ifdef CLANG_IS_PRODUCTION
const bool IsProduction = true;
#else
const bool IsProduction = false;
#endif
Driver TheDriver(Path.str(), llvm::sys::getHostTriple(),
"a.out", IsProduction, Diags);
// Attempt to find the original path used to invoke the driver, to determine
// the installed path. We do this manually, because we want to support that
// path being a symlink.
{
llvm::SmallString<128> InstalledPath(argv[0]);
// Do a PATH lookup, if there are no directory components.
if (llvm::sys::path::filename(InstalledPath) == InstalledPath) {
llvm::sys::Path Tmp = llvm::sys::Program::FindProgramByName(
llvm::sys::path::filename(InstalledPath.str()));
if (!Tmp.empty())
InstalledPath = Tmp.str();
}
llvm::sys::fs::make_absolute(InstalledPath);
InstalledPath = llvm::sys::path::parent_path(InstalledPath);
bool exists;
if (!llvm::sys::fs::exists(InstalledPath.str(), exists) && exists)
TheDriver.setInstalledDir(InstalledPath);
}
llvm::InitializeAllTargets();
ParseProgName(argv, SavedStrings, TheDriver);
// Handle CC_PRINT_OPTIONS and CC_PRINT_OPTIONS_FILE.
TheDriver.CCPrintOptions = !!::getenv("CC_PRINT_OPTIONS");
if (TheDriver.CCPrintOptions)
TheDriver.CCPrintOptionsFilename = ::getenv("CC_PRINT_OPTIONS_FILE");
// Handle CC_PRINT_HEADERS and CC_PRINT_HEADERS_FILE.
TheDriver.CCPrintHeaders = !!::getenv("CC_PRINT_HEADERS");
if (TheDriver.CCPrintHeaders)
TheDriver.CCPrintHeadersFilename = ::getenv("CC_PRINT_HEADERS_FILE");
// Handle CC_LOG_DIAGNOSTICS and CC_LOG_DIAGNOSTICS_FILE.
TheDriver.CCLogDiagnostics = !!::getenv("CC_LOG_DIAGNOSTICS");
if (TheDriver.CCLogDiagnostics)
TheDriver.CCLogDiagnosticsFilename = ::getenv("CC_LOG_DIAGNOSTICS_FILE");
// Handle QA_OVERRIDE_GCC3_OPTIONS and CCC_ADD_ARGS, used for editing a
// command line behind the scenes.
if (const char *OverrideStr = ::getenv("QA_OVERRIDE_GCC3_OPTIONS")) {
// FIXME: Driver shouldn't take extra initial argument.
ApplyQAOverride(argv, OverrideStr, SavedStrings);
} else if (const char *Cur = ::getenv("CCC_ADD_ARGS")) {
// FIXME: Driver shouldn't take extra initial argument.
std::vector<const char*> ExtraArgs;
for (;;) {
const char *Next = strchr(Cur, ',');
if (Next) {
ExtraArgs.push_back(SaveStringInSet(SavedStrings,
std::string(Cur, Next)));
Cur = Next + 1;
} else {
if (*Cur != '\0')
ExtraArgs.push_back(SaveStringInSet(SavedStrings, Cur));
break;
}
}
argv.insert(&argv[1], ExtraArgs.begin(), ExtraArgs.end());
}
llvm::OwningPtr<Compilation> C(TheDriver.BuildCompilation(argv));
int Res = 0;
const Command *FailingCommand = 0;
if (C.get())
Res = TheDriver.ExecuteCompilation(*C, FailingCommand);
// If result status is < 0, then the driver command signalled an error.
// In this case, generate additional diagnostic information if possible.
if (Res < 0)
TheDriver.generateCompilationDiagnostics(*C, FailingCommand);
// If any timers were active but haven't been destroyed yet, print their
// results now. This happens in -disable-free mode.
llvm::TimerGroup::printAll(llvm::errs());
llvm::llvm_shutdown();
return Res;
}
