// Check to see if this function returns one or more constants. If so, replace
// all callers that use those return values with the constant value. This will
// leave in the actual return values and instructions, but deadargelim will
// clean that up.
//
// Additionally if a function always returns one of its arguments directly,
// callers will be updated to use the value they pass in directly instead of
// using the return value.
bool IPCP::PropagateConstantReturn(Function &F) {
if (F.getReturnType()->isVoidTy())
return false; // No return value.
// If this function could be overridden later in the link stage, we can't
// propagate information about its results into callers.
if (F.mayBeOverridden())
return false;
// Check to see if this function returns a constant.
SmallVector<Value *,4> RetVals;
StructType *STy = dyn_cast<StructType>(F.getReturnType());
if (STy)
for (unsigned i = 0, e = STy->getNumElements(); i < e; ++i)
RetVals.push_back(UndefValue::get(STy->getElementType(i)));
else
RetVals.push_back(UndefValue::get(F.getReturnType()));
unsigned NumNonConstant = 0;
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
if (ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator())) {
for (unsigned i = 0, e = RetVals.size(); i != e; ++i) {
// Already found conflicting return values?
Value *RV = RetVals[i];
if (!RV)
continue;
// Find the returned value
Value *V;
if (!STy)
V = RI->getOperand(0);
else
V = FindInsertedValue(RI->getOperand(0), i);
if (V) {
// Ignore undefs, we can change them into anything
if (isa<UndefValue>(V))
continue;
// Try to see if all the rets return the same constant or argument.
if (isa<Constant>(V) || isa<Argument>(V)) {
if (isa<UndefValue>(RV)) {
// No value found yet? Try the current one.
RetVals[i] = V;
continue;
}
// Returning the same value? Good.
if (RV == V)
continue;
}
}
// Different or no known return value? Don't propagate this return
// value.
RetVals[i] = 0;
// All values non constant? Stop looking.
if (++NumNonConstant == RetVals.size())
return false;
}
}
// If we got here, the function returns at least one constant value. Loop
// over all users, replacing any uses of the return value with the returned
// constant.
bool MadeChange = false;
for (Value::use_iterator UI = F.use_begin(), E = F.use_end(); UI != E; ++UI) {
CallSite CS(*UI);
Instruction* Call = CS.getInstruction();
// Not a call instruction or a call instruction that's not calling F
// directly?
if (!Call || !CS.isCallee(UI))
continue;
// Call result not used?
if (Call->use_empty())
continue;
MadeChange = true;
if (STy == 0) {
Value* New = RetVals[0];
if (Argument *A = dyn_cast<Argument>(New))
// Was an argument returned? Then find the corresponding argument in
// the call instruction and use that.
New = CS.getArgument(A->getArgNo());
Call->replaceAllUsesWith(New);
continue;
}
for (Value::use_iterator I = Call->use_begin(), E = Call->use_end();
I != E;) {
Instruction *Ins = cast<Instruction>(*I);
// Increment now, so we can remove the use
++I;
// Find the index of the retval to replace with
int index = -1;
if (ExtractValueInst *EV = dyn_cast<ExtractValueInst>(Ins))
if (EV->hasIndices())
index = *EV->idx_begin();
// If this use uses a specific return value, and we have a replacement,
// replace it.
if (index != -1) {
Value *New = RetVals[index];
if (New) {
if (Argument *A = dyn_cast<Argument>(New))
// Was an argument returned? Then find the corresponding argument in
// the call instruction and use that.
New = CS.getArgument(A->getArgNo());
Ins->replaceAllUsesWith(New);
Ins->eraseFromParent();
}
}
}
}
if (MadeChange) ++NumReturnValProped;
return MadeChange;
}
void Diagnostic::
FormatDiagnostic(const char *DiagStr, const char *DiagEnd,
SmallVectorImpl<char> &OutStr) const {
/// FormattedArgs - Keep track of all of the arguments formatted by
/// ConvertArgToString and pass them into subsequent calls to
/// ConvertArgToString, allowing the implementation to avoid redundancies in
/// obvious cases.
SmallVector<DiagnosticsEngine::ArgumentValue, 8> FormattedArgs;
/// QualTypeVals - Pass a vector of arrays so that QualType names can be
/// compared to see if more information is needed to be printed.
SmallVector<intptr_t, 2> QualTypeVals;
SmallVector<char, 64> Tree;
for (unsigned i = 0, e = getNumArgs(); i < e; ++i)
if (getArgKind(i) == DiagnosticsEngine::ak_qualtype)
QualTypeVals.push_back(getRawArg(i));
while (DiagStr != DiagEnd) {
if (DiagStr[0] != '%') {
// Append non-%0 substrings to Str if we have one.
const char *StrEnd = std::find(DiagStr, DiagEnd, '%');
OutStr.append(DiagStr, StrEnd);
DiagStr = StrEnd;
continue;
} else if (ispunct(DiagStr[1])) {
OutStr.push_back(DiagStr[1]); // %% -> %.
DiagStr += 2;
continue;
}
// Skip the %.
++DiagStr;
// This must be a placeholder for a diagnostic argument. The format for a
// placeholder is one of "%0", "%modifier0", or "%modifier{arguments}0".
// The digit is a number from 0-9 indicating which argument this comes from.
// The modifier is a string of digits from the set [-a-z]+, arguments is a
// brace enclosed string.
const char *Modifier = 0, *Argument = 0;
unsigned ModifierLen = 0, ArgumentLen = 0;
// Check to see if we have a modifier. If so eat it.
if (!isdigit(DiagStr[0])) {
Modifier = DiagStr;
while (DiagStr[0] == '-' ||
(DiagStr[0] >= 'a' && DiagStr[0] <= 'z'))
++DiagStr;
ModifierLen = DiagStr-Modifier;
// If we have an argument, get it next.
if (DiagStr[0] == '{') {
++DiagStr; // Skip {.
Argument = DiagStr;
DiagStr = ScanFormat(DiagStr, DiagEnd, '}');
assert(DiagStr != DiagEnd && "Mismatched {}'s in diagnostic string!");
ArgumentLen = DiagStr-Argument;
++DiagStr; // Skip }.
}
}
assert(isdigit(*DiagStr) && "Invalid format for argument in diagnostic");
unsigned ArgNo = *DiagStr++ - '0';
// Only used for type diffing.
unsigned ArgNo2 = ArgNo;
DiagnosticsEngine::ArgumentKind Kind = getArgKind(ArgNo);
if (Kind == DiagnosticsEngine::ak_qualtype &&
ModifierIs(Modifier, ModifierLen, "diff")) {
Kind = DiagnosticsEngine::ak_qualtype_pair;
assert(*DiagStr == ',' && isdigit(*(DiagStr + 1)) &&
"Invalid format for diff modifier");
++DiagStr; // Comma.
ArgNo2 = *DiagStr++ - '0';
assert(getArgKind(ArgNo2) == DiagnosticsEngine::ak_qualtype &&
"Second value of type diff must be a qualtype");
}
switch (Kind) {
// ---- STRINGS ----
case DiagnosticsEngine::ak_std_string: {
const std::string &S = getArgStdStr(ArgNo);
assert(ModifierLen == 0 && "No modifiers for strings yet");
OutStr.append(S.begin(), S.end());
break;
}
case DiagnosticsEngine::ak_c_string: {
const char *S = getArgCStr(ArgNo);
assert(ModifierLen == 0 && "No modifiers for strings yet");
// Don't crash if get passed a null pointer by accident.
if (!S)
S = "(null)";
OutStr.append(S, S + strlen(S));
break;
}
// ---- INTEGERS ----
case DiagnosticsEngine::ak_sint: {
int Val = getArgSInt(ArgNo);
if (ModifierIs(Modifier, ModifierLen, "select")) {
HandleSelectModifier(*this, (unsigned)Val, Argument, ArgumentLen,
OutStr);
} else if (ModifierIs(Modifier, ModifierLen, "s")) {
HandleIntegerSModifier(Val, OutStr);
} else if (ModifierIs(Modifier, ModifierLen, "plural")) {
HandlePluralModifier(*this, (unsigned)Val, Argument, ArgumentLen,
OutStr);
} else if (ModifierIs(Modifier, ModifierLen, "ordinal")) {
HandleOrdinalModifier((unsigned)Val, OutStr);
} else {
assert(ModifierLen == 0 && "Unknown integer modifier");
llvm::raw_svector_ostream(OutStr) << Val;
}
break;
}
case DiagnosticsEngine::ak_uint: {
unsigned Val = getArgUInt(ArgNo);
if (ModifierIs(Modifier, ModifierLen, "select")) {
HandleSelectModifier(*this, Val, Argument, ArgumentLen, OutStr);
} else if (ModifierIs(Modifier, ModifierLen, "s")) {
HandleIntegerSModifier(Val, OutStr);
} else if (ModifierIs(Modifier, ModifierLen, "plural")) {
HandlePluralModifier(*this, (unsigned)Val, Argument, ArgumentLen,
OutStr);
} else if (ModifierIs(Modifier, ModifierLen, "ordinal")) {
HandleOrdinalModifier(Val, OutStr);
} else {
assert(ModifierLen == 0 && "Unknown integer modifier");
llvm::raw_svector_ostream(OutStr) << Val;
}
break;
}
// ---- NAMES and TYPES ----
case DiagnosticsEngine::ak_identifierinfo: {
const IdentifierInfo *II = getArgIdentifier(ArgNo);
assert(ModifierLen == 0 && "No modifiers for strings yet");
// Don't crash if get passed a null pointer by accident.
if (!II) {
const char *S = "(null)";
OutStr.append(S, S + strlen(S));
continue;
}
llvm::raw_svector_ostream(OutStr) << '\'' << II->getName() << '\'';
break;
}
case DiagnosticsEngine::ak_qualtype:
case DiagnosticsEngine::ak_declarationname:
case DiagnosticsEngine::ak_nameddecl:
case DiagnosticsEngine::ak_nestednamespec:
case DiagnosticsEngine::ak_declcontext:
getDiags()->ConvertArgToString(Kind, getRawArg(ArgNo),
Modifier, ModifierLen,
Argument, ArgumentLen,
FormattedArgs.data(), FormattedArgs.size(),
OutStr, QualTypeVals);
break;
case DiagnosticsEngine::ak_qualtype_pair:
// Create a struct with all the info needed for printing.
TemplateDiffTypes TDT;
TDT.FromType = getRawArg(ArgNo);
TDT.ToType = getRawArg(ArgNo2);
TDT.ElideType = getDiags()->ElideType;
TDT.ShowColors = getDiags()->ShowColors;
TDT.TemplateDiffUsed = false;
intptr_t val = reinterpret_cast<intptr_t>(&TDT);
const char *ArgumentEnd = Argument + ArgumentLen;
const char *Pipe = ScanFormat(Argument, ArgumentEnd, '|');
// Print the tree. If this diagnostic already has a tree, skip the
// second tree.
if (getDiags()->PrintTemplateTree && Tree.empty()) {
TDT.PrintFromType = true;
TDT.PrintTree = true;
getDiags()->ConvertArgToString(Kind, val,
Modifier, ModifierLen,
Argument, ArgumentLen,
FormattedArgs.data(),
FormattedArgs.size(),
Tree, QualTypeVals);
// If there is no tree information, fall back to regular printing.
if (!Tree.empty()) {
FormatDiagnostic(Pipe + 1, ArgumentEnd, OutStr);
break;
}
}
// Non-tree printing, also the fall-back when tree printing fails.
// The fall-back is triggered when the types compared are not templates.
const char *FirstDollar = ScanFormat(Argument, ArgumentEnd, '$');
const char *SecondDollar = ScanFormat(FirstDollar + 1, ArgumentEnd, '$');
// Append before text
FormatDiagnostic(Argument, FirstDollar, OutStr);
// Append first type
TDT.PrintTree = false;
TDT.PrintFromType = true;
getDiags()->ConvertArgToString(Kind, val,
Modifier, ModifierLen,
Argument, ArgumentLen,
FormattedArgs.data(), FormattedArgs.size(),
OutStr, QualTypeVals);
if (!TDT.TemplateDiffUsed)
FormattedArgs.push_back(std::make_pair(DiagnosticsEngine::ak_qualtype,
TDT.FromType));
// Append middle text
FormatDiagnostic(FirstDollar + 1, SecondDollar, OutStr);
// Append second type
TDT.PrintFromType = false;
getDiags()->ConvertArgToString(Kind, val,
Modifier, ModifierLen,
Argument, ArgumentLen,
FormattedArgs.data(), FormattedArgs.size(),
OutStr, QualTypeVals);
if (!TDT.TemplateDiffUsed)
FormattedArgs.push_back(std::make_pair(DiagnosticsEngine::ak_qualtype,
TDT.ToType));
// Append end text
FormatDiagnostic(SecondDollar + 1, Pipe, OutStr);
break;
}
// Remember this argument info for subsequent formatting operations. Turn
// std::strings into a null terminated string to make it be the same case as
// all the other ones.
if (Kind == DiagnosticsEngine::ak_qualtype_pair)
continue;
else if (Kind != DiagnosticsEngine::ak_std_string)
FormattedArgs.push_back(std::make_pair(Kind, getRawArg(ArgNo)));
else
FormattedArgs.push_back(std::make_pair(DiagnosticsEngine::ak_c_string,
(intptr_t)getArgStdStr(ArgNo).c_str()));
}
// Append the type tree to the end of the diagnostics.
OutStr.append(Tree.begin(), Tree.end());
}
void
UserValue::addDefsFromCopies(LiveInterval *LI, unsigned LocNo,
const SmallVectorImpl<SlotIndex> &Kills,
SmallVectorImpl<std::pair<SlotIndex, unsigned> > &NewDefs,
MachineRegisterInfo &MRI, LiveIntervals &LIS) {
if (Kills.empty())
return;
// Don't track copies from physregs, there are too many uses.
if (!TargetRegisterInfo::isVirtualRegister(LI->reg))
return;
// Collect all the (vreg, valno) pairs that are copies of LI.
SmallVector<std::pair<LiveInterval*, const VNInfo*>, 8> CopyValues;
for (MachineOperand &MO : MRI.use_nodbg_operands(LI->reg)) {
MachineInstr *MI = MO.getParent();
// Copies of the full value.
if (MO.getSubReg() || !MI->isCopy())
continue;
unsigned DstReg = MI->getOperand(0).getReg();
// Don't follow copies to physregs. These are usually setting up call
// arguments, and the argument registers are always call clobbered. We are
// better off in the source register which could be a callee-saved register,
// or it could be spilled.
if (!TargetRegisterInfo::isVirtualRegister(DstReg))
continue;
// Is LocNo extended to reach this copy? If not, another def may be blocking
// it, or we are looking at a wrong value of LI.
SlotIndex Idx = LIS.getInstructionIndex(MI);
LocMap::iterator I = locInts.find(Idx.getRegSlot(true));
if (!I.valid() || I.value() != LocNo)
continue;
if (!LIS.hasInterval(DstReg))
continue;
LiveInterval *DstLI = &LIS.getInterval(DstReg);
const VNInfo *DstVNI = DstLI->getVNInfoAt(Idx.getRegSlot());
assert(DstVNI && DstVNI->def == Idx.getRegSlot() && "Bad copy value");
CopyValues.push_back(std::make_pair(DstLI, DstVNI));
}
if (CopyValues.empty())
return;
DEBUG(dbgs() << "Got " << CopyValues.size() << " copies of " << *LI << '\n');
// Try to add defs of the copied values for each kill point.
for (unsigned i = 0, e = Kills.size(); i != e; ++i) {
SlotIndex Idx = Kills[i];
for (unsigned j = 0, e = CopyValues.size(); j != e; ++j) {
LiveInterval *DstLI = CopyValues[j].first;
const VNInfo *DstVNI = CopyValues[j].second;
if (DstLI->getVNInfoAt(Idx) != DstVNI)
continue;
// Check that there isn't already a def at Idx
LocMap::iterator I = locInts.find(Idx);
if (I.valid() && I.start() <= Idx)
continue;
DEBUG(dbgs() << "Kill at " << Idx << " covered by valno #"
<< DstVNI->id << " in " << *DstLI << '\n');
MachineInstr *CopyMI = LIS.getInstructionFromIndex(DstVNI->def);
assert(CopyMI && CopyMI->isCopy() && "Bad copy value");
unsigned LocNo = getLocationNo(CopyMI->getOperand(0));
I.insert(Idx, Idx.getNextSlot(), LocNo);
NewDefs.push_back(std::make_pair(Idx, LocNo));
break;
}
}
}
/// Compute the actions table and gather the first action index for each landing
/// pad site.
unsigned EHStreamer::
computeActionsTable(const SmallVectorImpl<const LandingPadInfo*> &LandingPads,
SmallVectorImpl<ActionEntry> &Actions,
SmallVectorImpl<unsigned> &FirstActions) {
// The action table follows the call-site table in the LSDA. The individual
// records are of two types:
//
// * Catch clause
// * Exception specification
//
// The two record kinds have the same format, with only small differences.
// They are distinguished by the "switch value" field: Catch clauses
// (TypeInfos) have strictly positive switch values, and exception
// specifications (FilterIds) have strictly negative switch values. Value 0
// indicates a catch-all clause.
//
// Negative type IDs index into FilterIds. Positive type IDs index into
// TypeInfos. The value written for a positive type ID is just the type ID
// itself. For a negative type ID, however, the value written is the
// (negative) byte offset of the corresponding FilterIds entry. The byte
// offset is usually equal to the type ID (because the FilterIds entries are
// written using a variable width encoding, which outputs one byte per entry
// as long as the value written is not too large) but can differ. This kind
// of complication does not occur for positive type IDs because type infos are
// output using a fixed width encoding. FilterOffsets[i] holds the byte
// offset corresponding to FilterIds[i].
const std::vector<unsigned> &FilterIds = MMI->getFilterIds();
SmallVector<int, 16> FilterOffsets;
FilterOffsets.reserve(FilterIds.size());
int Offset = -1;
for (std::vector<unsigned>::const_iterator
I = FilterIds.begin(), E = FilterIds.end(); I != E; ++I) {
FilterOffsets.push_back(Offset);
Offset -= getULEB128Size(*I);
}
FirstActions.reserve(LandingPads.size());
int FirstAction = 0;
unsigned SizeActions = 0;
const LandingPadInfo *PrevLPI = nullptr;
for (SmallVectorImpl<const LandingPadInfo *>::const_iterator
I = LandingPads.begin(), E = LandingPads.end(); I != E; ++I) {
const LandingPadInfo *LPI = *I;
const std::vector<int> &TypeIds = LPI->TypeIds;
unsigned NumShared = PrevLPI ? sharedTypeIDs(LPI, PrevLPI) : 0;
unsigned SizeSiteActions = 0;
if (NumShared < TypeIds.size()) {
unsigned SizeAction = 0;
unsigned PrevAction = (unsigned)-1;
if (NumShared) {
unsigned SizePrevIds = PrevLPI->TypeIds.size();
assert(Actions.size());
PrevAction = Actions.size() - 1;
SizeAction = getSLEB128Size(Actions[PrevAction].NextAction) +
getSLEB128Size(Actions[PrevAction].ValueForTypeID);
for (unsigned j = NumShared; j != SizePrevIds; ++j) {
assert(PrevAction != (unsigned)-1 && "PrevAction is invalid!");
SizeAction -= getSLEB128Size(Actions[PrevAction].ValueForTypeID);
SizeAction += -Actions[PrevAction].NextAction;
PrevAction = Actions[PrevAction].Previous;
}
}
// Compute the actions.
for (unsigned J = NumShared, M = TypeIds.size(); J != M; ++J) {
int TypeID = TypeIds[J];
assert(-1 - TypeID < (int)FilterOffsets.size() && "Unknown filter id!");
int ValueForTypeID = TypeID < 0 ? FilterOffsets[-1 - TypeID] : TypeID;
unsigned SizeTypeID = getSLEB128Size(ValueForTypeID);
int NextAction = SizeAction ? -(SizeAction + SizeTypeID) : 0;
SizeAction = SizeTypeID + getSLEB128Size(NextAction);
SizeSiteActions += SizeAction;
ActionEntry Action = { ValueForTypeID, NextAction, PrevAction };
Actions.push_back(Action);
PrevAction = Actions.size() - 1;
}
// Record the first action of the landing pad site.
FirstAction = SizeActions + SizeSiteActions - SizeAction + 1;
} // else identical - re-use previous FirstAction
// Information used when created the call-site table. The action record
// field of the call site record is the offset of the first associated
// action record, relative to the start of the actions table. This value is
// biased by 1 (1 indicating the start of the actions table), and 0
// indicates that there are no actions.
FirstActions.push_back(FirstAction);
// Compute this sites contribution to size.
SizeActions += SizeSiteActions;
PrevLPI = LPI;
}
return SizeActions;
}
/// ClusterNeighboringLoads - Force nearby loads together by "gluing" them.
/// This function finds loads of the same base and different offsets. If the
/// offsets are not far apart (target specific), it add MVT::Glue inputs and
/// outputs to ensure they are scheduled together and in order. This
/// optimization may benefit some targets by improving cache locality.
void ScheduleDAGSDNodes::ClusterNeighboringLoads(SDNode *Node) {
SDNode *Chain = nullptr;
unsigned NumOps = Node->getNumOperands();
if (Node->getOperand(NumOps-1).getValueType() == MVT::Other)
Chain = Node->getOperand(NumOps-1).getNode();
if (!Chain)
return;
// Look for other loads of the same chain. Find loads that are loading from
// the same base pointer and different offsets.
SmallPtrSet<SDNode*, 16> Visited;
SmallVector<int64_t, 4> Offsets;
DenseMap<long long, SDNode*> O2SMap; // Map from offset to SDNode.
bool Cluster = false;
SDNode *Base = Node;
// This algorithm requires a reasonably low use count before finding a match
// to avoid uselessly blowing up compile time in large blocks.
unsigned UseCount = 0;
for (SDNode::use_iterator I = Chain->use_begin(), E = Chain->use_end();
I != E && UseCount < 100; ++I, ++UseCount) {
SDNode *User = *I;
if (User == Node || !Visited.insert(User).second)
continue;
int64_t Offset1, Offset2;
if (!TII->areLoadsFromSameBasePtr(Base, User, Offset1, Offset2) ||
Offset1 == Offset2)
// FIXME: Should be ok if they addresses are identical. But earlier
// optimizations really should have eliminated one of the loads.
continue;
if (O2SMap.insert(std::make_pair(Offset1, Base)).second)
Offsets.push_back(Offset1);
O2SMap.insert(std::make_pair(Offset2, User));
Offsets.push_back(Offset2);
if (Offset2 < Offset1)
Base = User;
Cluster = true;
// Reset UseCount to allow more matches.
UseCount = 0;
}
if (!Cluster)
return;
// Sort them in increasing order.
std::sort(Offsets.begin(), Offsets.end());
// Check if the loads are close enough.
SmallVector<SDNode*, 4> Loads;
unsigned NumLoads = 0;
int64_t BaseOff = Offsets[0];
SDNode *BaseLoad = O2SMap[BaseOff];
Loads.push_back(BaseLoad);
for (unsigned i = 1, e = Offsets.size(); i != e; ++i) {
int64_t Offset = Offsets[i];
SDNode *Load = O2SMap[Offset];
if (!TII->shouldScheduleLoadsNear(BaseLoad, Load, BaseOff, Offset,NumLoads))
break; // Stop right here. Ignore loads that are further away.
Loads.push_back(Load);
++NumLoads;
}
if (NumLoads == 0)
return;
// Cluster loads by adding MVT::Glue outputs and inputs. This also
// ensure they are scheduled in order of increasing addresses.
SDNode *Lead = Loads[0];
SDValue InGlue = SDValue(nullptr, 0);
if (AddGlue(Lead, InGlue, true, DAG))
InGlue = SDValue(Lead, Lead->getNumValues() - 1);
for (unsigned I = 1, E = Loads.size(); I != E; ++I) {
bool OutGlue = I < E - 1;
SDNode *Load = Loads[I];
// If AddGlue fails, we could leave an unsused glue value. This should not
// cause any
if (AddGlue(Load, InGlue, OutGlue, DAG)) {
if (OutGlue)
InGlue = SDValue(Load, Load->getNumValues() - 1);
++LoadsClustered;
}
else if (!OutGlue && InGlue.getNode())
RemoveUnusedGlue(InGlue.getNode(), DAG);
}
}
/// Emit a checked cast to a protocol or protocol composition.
void irgen::emitScalarExistentialDowncast(IRGenFunction &IGF,
llvm::Value *value,
SILType srcType,
SILType destType,
CheckedCastMode mode,
Optional<MetatypeRepresentation> metatypeKind,
Explosion &ex) {
SmallVector<ProtocolDecl*, 4> allProtos;
destType.getSwiftRValueType().getAnyExistentialTypeProtocols(allProtos);
// Look up witness tables for the protocols that need them and get
// references to the ObjC Protocol* values for the objc protocols.
SmallVector<llvm::Value*, 4> objcProtos;
SmallVector<llvm::Value*, 4> witnessTableProtos;
bool hasClassConstraint = false;
bool hasClassConstraintByProtocol = false;
for (auto proto : allProtos) {
// If the protocol introduces a class constraint, track whether we need
// to check for it independent of protocol witnesses.
if (proto->requiresClass()) {
hasClassConstraint = true;
if (proto->getKnownProtocolKind()
&& *proto->getKnownProtocolKind() == KnownProtocolKind::AnyObject) {
// AnyObject only requires that the type be a class.
continue;
}
// If this protocol is class-constrained but not AnyObject, checking its
// conformance will check the class constraint too.
hasClassConstraintByProtocol = true;
}
if (Lowering::TypeConverter::protocolRequiresWitnessTable(proto)) {
auto descriptor = emitProtocolDescriptorRef(IGF, proto);
witnessTableProtos.push_back(descriptor);
}
if (!proto->isObjC())
continue;
objcProtos.push_back(emitReferenceToObjCProtocol(IGF, proto));
}
llvm::Type *resultType;
if (metatypeKind) {
switch (*metatypeKind) {
case MetatypeRepresentation::Thin:
llvm_unreachable("can't cast to thin metatype");
case MetatypeRepresentation::Thick:
resultType = IGF.IGM.TypeMetadataPtrTy;
break;
case MetatypeRepresentation::ObjC:
resultType = IGF.IGM.ObjCClassPtrTy;
break;
}
} else {
auto schema = IGF.getTypeInfo(destType).getSchema();
resultType = schema[0].getScalarType();
}
// We only need to check the class constraint for metatype casts where
// no protocol conformance indirectly requires the constraint for us.
bool checkClassConstraint =
(bool)metatypeKind && hasClassConstraint && !hasClassConstraintByProtocol;
llvm::Value *resultValue = value;
// If we don't have anything we really need to check, then trivially succeed.
if (objcProtos.empty() && witnessTableProtos.empty() &&
!checkClassConstraint) {
resultValue = IGF.Builder.CreateBitCast(value, resultType);
ex.add(resultValue);
return;
}
// Check the ObjC protocol conformances if there were any.
llvm::Value *objcCast = nullptr;
if (!objcProtos.empty()) {
// Get the ObjC instance or class object to check for these conformances.
llvm::Value *objcObject;
if (metatypeKind) {
switch (*metatypeKind) {
case MetatypeRepresentation::Thin:
llvm_unreachable("can't cast to thin metatype");
case MetatypeRepresentation::Thick: {
// The metadata might be for a non-class type, which wouldn't have
// an ObjC class object.
objcObject = nullptr;
break;
}
case MetatypeRepresentation::ObjC:
// Metatype is already an ObjC object.
objcObject = value;
break;
}
} else {
// Class instance is already an ObjC object.
objcObject = value;
}
if (objcObject)
objcObject = IGF.Builder.CreateBitCast(objcObject,
IGF.IGM.UnknownRefCountedPtrTy);
// Pick the cast function based on the cast mode and on whether we're
// casting a Swift metatype or ObjC object.
llvm::Value *castFn;
switch (mode) {
case CheckedCastMode::Unconditional:
castFn = objcObject
? IGF.IGM.getDynamicCastObjCProtocolUnconditionalFn()
: IGF.IGM.getDynamicCastTypeToObjCProtocolUnconditionalFn();
break;
case CheckedCastMode::Conditional:
castFn = objcObject
? IGF.IGM.getDynamicCastObjCProtocolConditionalFn()
: IGF.IGM.getDynamicCastTypeToObjCProtocolConditionalFn();
break;
}
llvm::Value *objcCastObject = objcObject ? objcObject : value;
Address protoRefsBuf = IGF.createAlloca(
llvm::ArrayType::get(IGF.IGM.Int8PtrTy,
objcProtos.size()),
IGF.IGM.getPointerAlignment(),
"objc_protocols");
protoRefsBuf = IGF.Builder.CreateBitCast(protoRefsBuf,
IGF.IGM.Int8PtrPtrTy);
for (unsigned index : indices(objcProtos)) {
Address protoRefSlot = IGF.Builder.CreateConstArrayGEP(
protoRefsBuf, index,
IGF.IGM.getPointerSize());
IGF.Builder.CreateStore(objcProtos[index], protoRefSlot);
++index;
}
auto cc = IGF.IGM.DefaultCC;
if (auto fun = dyn_cast<llvm::Function>(castFn))
cc = fun->getCallingConv();
auto call = IGF.Builder.CreateCall(
castFn,
{objcCastObject, IGF.IGM.getSize(Size(objcProtos.size())),
protoRefsBuf.getAddress()});
call->setCallingConv(cc);
objcCast = call;
resultValue = IGF.Builder.CreateBitCast(objcCast, resultType);
}
// If we don't need to look up any witness tables, we're done.
if (witnessTableProtos.empty() && !checkClassConstraint) {
ex.add(resultValue);
return;
}
// If we're doing a conditional cast, and the ObjC protocol checks failed,
// then the cast is done.
Optional<ConditionalDominanceScope> condition;
llvm::BasicBlock *origBB = nullptr, *successBB = nullptr, *contBB = nullptr;
if (!objcProtos.empty()) {
switch (mode) {
case CheckedCastMode::Unconditional:
break;
case CheckedCastMode::Conditional: {
origBB = IGF.Builder.GetInsertBlock();
successBB = IGF.createBasicBlock("success");
contBB = IGF.createBasicBlock("cont");
auto isNull = IGF.Builder.CreateICmpEQ(objcCast,
llvm::ConstantPointerNull::get(
cast<llvm::PointerType>(objcCast->getType())));
IGF.Builder.CreateCondBr(isNull, contBB, successBB);
IGF.Builder.emitBlock(successBB);
condition.emplace(IGF);
}
}
}
// Get the Swift type metadata for the type.
llvm::Value *metadataValue;
if (metatypeKind) {
switch (*metatypeKind) {
case MetatypeRepresentation::Thin:
llvm_unreachable("can't cast to thin metatype");
case MetatypeRepresentation::Thick:
// The value is already a native metatype.
metadataValue = value;
break;
case MetatypeRepresentation::ObjC:
// Get the type metadata from the ObjC class, which may be a wrapper.
metadataValue = emitObjCMetadataRefForMetadata(IGF, value);
}
} else {
// Get the type metadata for the instance.
metadataValue = emitDynamicTypeOfHeapObject(IGF, value, srcType);
}
// Look up witness tables for the protocols that need them.
auto fn = emitExistentialScalarCastFn(IGF.IGM, witnessTableProtos.size(),
mode, checkClassConstraint);
llvm::SmallVector<llvm::Value *, 4> args;
if (resultValue->getType() != IGF.IGM.Int8PtrTy)
resultValue = IGF.Builder.CreateBitCast(resultValue, IGF.IGM.Int8PtrTy);
args.push_back(resultValue);
args.push_back(metadataValue);
for (auto proto : witnessTableProtos)
args.push_back(proto);
auto valueAndWitnessTables = IGF.Builder.CreateCall(fn, args);
resultValue = IGF.Builder.CreateExtractValue(valueAndWitnessTables, 0);
if (resultValue->getType() != resultType)
resultValue = IGF.Builder.CreateBitCast(resultValue, resultType);
ex.add(resultValue);
for (unsigned i = 0, e = witnessTableProtos.size(); i < e; ++i) {
auto wt = IGF.Builder.CreateExtractValue(valueAndWitnessTables, i + 1);
ex.add(wt);
}
// If we had conditional ObjC checks, join the failure paths.
if (contBB) {
condition.reset();
IGF.Builder.CreateBr(contBB);
IGF.Builder.emitBlock(contBB);
// Return null on the failure path.
Explosion successEx = std::move(ex);
ex.reset();
while (!successEx.empty()) {
auto successVal = successEx.claimNext();
auto failureVal = llvm::Constant::getNullValue(successVal->getType());
auto phi = IGF.Builder.CreatePHI(successVal->getType(), 2);
phi->addIncoming(successVal, successBB);
phi->addIncoming(failureVal, origBB);
ex.add(phi);
}
}
}
/// ParseBlock - Read a block, updating statistics, etc.
static bool ParseBlock(BitstreamCursor &Stream, unsigned IndentLevel) {
std::string Indent(IndentLevel*2, ' ');
uint64_t BlockBitStart = Stream.GetCurrentBitNo();
unsigned BlockID = Stream.ReadSubBlockID();
// Get the statistics for this BlockID.
PerBlockIDStats &BlockStats = BlockIDStats[BlockID];
BlockStats.NumInstances++;
// BLOCKINFO is a special part of the stream.
if (BlockID == bitc::BLOCKINFO_BLOCK_ID) {
if (Dump) errs() << Indent << "<BLOCKINFO_BLOCK/>\n";
if (Stream.ReadBlockInfoBlock())
return Error("Malformed BlockInfoBlock");
uint64_t BlockBitEnd = Stream.GetCurrentBitNo();
BlockStats.NumBits += BlockBitEnd-BlockBitStart;
return false;
}
unsigned NumWords = 0;
if (Stream.EnterSubBlock(BlockID, &NumWords))
return Error("Malformed block record");
const char *BlockName = 0;
if (Dump) {
errs() << Indent << "<";
if ((BlockName = GetBlockName(BlockID, *Stream.getBitStreamReader())))
errs() << BlockName;
else
errs() << "UnknownBlock" << BlockID;
if (NonSymbolic && BlockName)
errs() << " BlockID=" << BlockID;
errs() << " NumWords=" << NumWords
<< " BlockCodeSize=" << Stream.GetAbbrevIDWidth() << ">\n";
}
SmallVector<uint64_t, 64> Record;
// Read all the records for this block.
while (1) {
if (Stream.AtEndOfStream())
return Error("Premature end of bitstream");
uint64_t RecordStartBit = Stream.GetCurrentBitNo();
// Read the code for this record.
unsigned AbbrevID = Stream.ReadCode();
switch (AbbrevID) {
case bitc::END_BLOCK: {
if (Stream.ReadBlockEnd())
return Error("Error at end of block");
uint64_t BlockBitEnd = Stream.GetCurrentBitNo();
BlockStats.NumBits += BlockBitEnd-BlockBitStart;
if (Dump) {
errs() << Indent << "</";
if (BlockName)
errs() << BlockName << ">\n";
else
errs() << "UnknownBlock" << BlockID << ">\n";
}
return false;
}
case bitc::ENTER_SUBBLOCK: {
uint64_t SubBlockBitStart = Stream.GetCurrentBitNo();
if (ParseBlock(Stream, IndentLevel+1))
return true;
++BlockStats.NumSubBlocks;
uint64_t SubBlockBitEnd = Stream.GetCurrentBitNo();
// Don't include subblock sizes in the size of this block.
BlockBitStart += SubBlockBitEnd-SubBlockBitStart;
break;
}
case bitc::DEFINE_ABBREV:
Stream.ReadAbbrevRecord();
++BlockStats.NumAbbrevs;
break;
default:
Record.clear();
++BlockStats.NumRecords;
if (AbbrevID != bitc::UNABBREV_RECORD)
++BlockStats.NumAbbreviatedRecords;
const char *BlobStart = 0;
unsigned BlobLen = 0;
unsigned Code = Stream.ReadRecord(AbbrevID, Record, BlobStart, BlobLen);
// Increment the # occurrences of this code.
if (BlockStats.CodeFreq.size() <= Code)
BlockStats.CodeFreq.resize(Code+1);
BlockStats.CodeFreq[Code].NumInstances++;
BlockStats.CodeFreq[Code].TotalBits +=
Stream.GetCurrentBitNo()-RecordStartBit;
if (AbbrevID != bitc::UNABBREV_RECORD)
BlockStats.CodeFreq[Code].NumAbbrev++;
if (Dump) {
errs() << Indent << " <";
if (const char *CodeName =
GetCodeName(Code, BlockID, *Stream.getBitStreamReader()))
errs() << CodeName;
else
errs() << "UnknownCode" << Code;
if (NonSymbolic &&
GetCodeName(Code, BlockID, *Stream.getBitStreamReader()))
errs() << " codeid=" << Code;
if (AbbrevID != bitc::UNABBREV_RECORD)
errs() << " abbrevid=" << AbbrevID;
for (unsigned i = 0, e = Record.size(); i != e; ++i)
errs() << " op" << i << "=" << (int64_t)Record[i];
errs() << "/>";
if (BlobStart) {
errs() << " blob data = ";
bool BlobIsPrintable = true;
for (unsigned i = 0; i != BlobLen; ++i)
if (!isprint(BlobStart[i])) {
BlobIsPrintable = false;
break;
}
if (BlobIsPrintable)
errs() << "'" << std::string(BlobStart, BlobStart+BlobLen) <<"'";
else
errs() << "unprintable, " << BlobLen << " bytes.";
}
errs() << "\n";
}
break;
}
}
}
void ValueEnumerator::incorporateFunction(const Function &F) {
InstructionCount = 0;
NumModuleValues = Values.size();
NumModuleMDValues = MDValues.size();
// Adding function arguments to the value table.
for (Function::const_arg_iterator I = F.arg_begin(), E = F.arg_end();
I != E; ++I)
EnumerateValue(I);
FirstFuncConstantID = Values.size();
// Add all function-level constants to the value table.
for (Function::const_iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I!=E; ++I)
for (User::const_op_iterator OI = I->op_begin(), E = I->op_end();
OI != E; ++OI) {
if ((isa<Constant>(*OI) && !isa<GlobalValue>(*OI)) ||
isa<InlineAsm>(*OI))
EnumerateValue(*OI);
}
BasicBlocks.push_back(BB);
ValueMap[BB] = BasicBlocks.size();
}
// Optimize the constant layout.
OptimizeConstants(FirstFuncConstantID, Values.size());
// Add the function's parameter attributes so they are available for use in
// the function's instruction.
EnumerateAttributes(F.getAttributes());
FirstInstID = Values.size();
SmallVector<MDNode *, 8> FnLocalMDVector;
// Add all of the instructions.
for (Function::const_iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I!=E; ++I) {
for (User::const_op_iterator OI = I->op_begin(), E = I->op_end();
OI != E; ++OI) {
if (MDNode *MD = dyn_cast<MDNode>(*OI))
if (MD->isFunctionLocal() && MD->getFunction())
// Enumerate metadata after the instructions they might refer to.
FnLocalMDVector.push_back(MD);
}
SmallVector<std::pair<unsigned, MDNode*>, 8> MDs;
I->getAllMetadataOtherThanDebugLoc(MDs);
for (unsigned i = 0, e = MDs.size(); i != e; ++i) {
MDNode *N = MDs[i].second;
if (N->isFunctionLocal() && N->getFunction())
FnLocalMDVector.push_back(N);
}
if (!I->getType()->isVoidTy())
EnumerateValue(I);
}
}
// Add all of the function-local metadata.
for (unsigned i = 0, e = FnLocalMDVector.size(); i != e; ++i)
EnumerateFunctionLocalMetadata(FnLocalMDVector[i]);
}
// recurseBasicBlock() - This calculates the ProfileInfo estimation for a
// single block and then recurses into the successors.
// The algorithm preserves the flow condition, meaning that the sum of the
// weight of the incoming edges must be equal the block weight which must in
// turn be equal to the sume of the weights of the outgoing edges.
// Since the flow of an block is deterimined from the current state of the
// flow, once an edge has a flow assigned this flow is never changed again,
// otherwise it would be possible to violate the flow condition in another
// block.
void ProfileEstimatorPass::recurseBasicBlock(BasicBlock *BB) {
// Break the recursion if this BasicBlock was already visited.
if (BBToVisit.find(BB) == BBToVisit.end()) return;
// Read the LoopInfo for this block.
bool BBisHeader = LI->isLoopHeader(BB);
Loop* BBLoop = LI->getLoopFor(BB);
// To get the block weight, read all incoming edges.
double BBWeight = 0;
std::set<BasicBlock*> ProcessedPreds;
for ( pred_iterator bbi = pred_begin(BB), bbe = pred_end(BB);
bbi != bbe; ++bbi ) {
// If this block was not considered already, add weight.
Edge edge = getEdge(*bbi,BB);
double w = getEdgeWeight(edge);
if (ProcessedPreds.insert(*bbi).second) {
BBWeight += ignoreMissing(w);
}
// If this block is a loop header and the predecessor is contained in this
// loop, thus the edge is a backedge, continue and do not check if the
// value is valid.
if (BBisHeader && BBLoop->contains(*bbi)) {
printEdgeError(edge, "but is backedge, continuing");
continue;
}
// If the edges value is missing (and this is no loop header, and this is
// no backedge) return, this block is currently non estimatable.
if (w == MissingValue) {
printEdgeError(edge, "returning");
return;
}
}
if (getExecutionCount(BB) != MissingValue) {
BBWeight = getExecutionCount(BB);
}
// Fetch all necessary information for current block.
SmallVector<Edge, 8> ExitEdges;
SmallVector<Edge, 8> Edges;
if (BBLoop) {
BBLoop->getExitEdges(ExitEdges);
}
// If this is a loop header, consider the following:
// Exactly the flow that is entering this block, must exit this block too. So
// do the following:
// *) get all the exit edges, read the flow that is already leaving this
// loop, remember the edges that do not have any flow on them right now.
// (The edges that have already flow on them are most likely exiting edges of
// other loops, do not touch those flows because the previously caclulated
// loopheaders would not be exact anymore.)
// *) In case there is not a single exiting edge left, create one at the loop
// latch to prevent the flow from building up in the loop.
// *) Take the flow that is not leaving the loop already and distribute it on
// the remaining exiting edges.
// (This ensures that all flow that enters the loop also leaves it.)
// *) Increase the flow into the loop by increasing the weight of this block.
// There is at least one incoming backedge that will bring us this flow later
// on. (So that the flow condition in this node is valid again.)
if (BBisHeader) {
double incoming = BBWeight;
// Subtract the flow leaving the loop.
std::set<Edge> ProcessedExits;
for (SmallVector<Edge, 8>::iterator ei = ExitEdges.begin(),
ee = ExitEdges.end(); ei != ee; ++ei) {
if (ProcessedExits.insert(*ei).second) {
double w = getEdgeWeight(*ei);
if (w == MissingValue) {
Edges.push_back(*ei);
// Check if there is a necessary minimal weight, if yes, subtract it
// from weight.
if (MinimalWeight.find(*ei) != MinimalWeight.end()) {
incoming -= MinimalWeight[*ei];
DEBUG(dbgs() << "Reserving " << format("%.20g",MinimalWeight[*ei]) << " at " << (*ei) << "\n");
}
} else {
incoming -= w;
}
}
}
// If no exit edges, create one:
if (Edges.size() == 0) {
BasicBlock *Latch = BBLoop->getLoopLatch();
if (Latch) {
Edge edge = getEdge(Latch,0);
EdgeInformation[BB->getParent()][edge] = BBWeight;
printEdgeWeight(edge);
edge = getEdge(Latch, BB);
EdgeInformation[BB->getParent()][edge] = BBWeight * ExecCount;
printEdgeWeight(edge);
}
}
// Distribute remaining weight to the exting edges. To prevent fractions
// from building up and provoking precision problems the weight which is to
// be distributed is split and the rounded, the last edge gets a somewhat
// bigger value, but we are close enough for an estimation.
double fraction = floor(incoming/Edges.size());
for (SmallVector<Edge, 8>::iterator ei = Edges.begin(), ee = Edges.end();
ei != ee; ++ei) {
double w = 0;
if (ei != (ee-1)) {
w = fraction;
incoming -= fraction;
} else {
w = incoming;
}
EdgeInformation[BB->getParent()][*ei] += w;
// Read necessary minimal weight.
if (MinimalWeight.find(*ei) != MinimalWeight.end()) {
EdgeInformation[BB->getParent()][*ei] += MinimalWeight[*ei];
DEBUG(dbgs() << "Additionally " << format("%.20g",MinimalWeight[*ei]) << " at " << (*ei) << "\n");
}
printEdgeWeight(*ei);
// Add minimal weight to paths to all exit edges, this is used to ensure
// that enough flow is reaching this edges.
Path p;
const BasicBlock *Dest = GetPath(BB, (*ei).first, p, GetPathToDest);
while (Dest != BB) {
const BasicBlock *Parent = p.find(Dest)->second;
Edge e = getEdge(Parent, Dest);
if (MinimalWeight.find(e) == MinimalWeight.end()) {
MinimalWeight[e] = 0;
}
MinimalWeight[e] += w;
DEBUG(dbgs() << "Minimal Weight for " << e << ": " << format("%.20g",MinimalWeight[e]) << "\n");
Dest = Parent;
}
}
// Increase flow into the loop.
BBWeight *= (ExecCount+1);
}
BlockInformation[BB->getParent()][BB] = BBWeight;
// Up until now we considered only the loop exiting edges, now we have a
// definite block weight and must distribute this onto the outgoing edges.
// Since there may be already flow attached to some of the edges, read this
// flow first and remember the edges that have still now flow attached.
Edges.clear();
std::set<BasicBlock*> ProcessedSuccs;
succ_iterator bbi = succ_begin(BB), bbe = succ_end(BB);
// Also check for (BB,0) edges that may already contain some flow. (But only
// in case there are no successors.)
if (bbi == bbe) {
Edge edge = getEdge(BB,0);
EdgeInformation[BB->getParent()][edge] = BBWeight;
printEdgeWeight(edge);
}
for ( ; bbi != bbe; ++bbi ) {
if (ProcessedSuccs.insert(*bbi).second) {
Edge edge = getEdge(BB,*bbi);
double w = getEdgeWeight(edge);
if (w != MissingValue) {
BBWeight -= getEdgeWeight(edge);
} else {
Edges.push_back(edge);
// If minimal weight is necessary, reserve weight by subtracting weight
// from block weight, this is readded later on.
if (MinimalWeight.find(edge) != MinimalWeight.end()) {
BBWeight -= MinimalWeight[edge];
DEBUG(dbgs() << "Reserving " << format("%.20g",MinimalWeight[edge]) << " at " << edge << "\n");
}
}
}
}
double fraction = floor(BBWeight/Edges.size());
// Finally we know what flow is still not leaving the block, distribute this
// flow onto the empty edges.
for (SmallVector<Edge, 8>::iterator ei = Edges.begin(), ee = Edges.end();
ei != ee; ++ei) {
if (ei != (ee-1)) {
EdgeInformation[BB->getParent()][*ei] += fraction;
BBWeight -= fraction;
} else {
EdgeInformation[BB->getParent()][*ei] += BBWeight;
}
// Readd minial necessary weight.
if (MinimalWeight.find(*ei) != MinimalWeight.end()) {
EdgeInformation[BB->getParent()][*ei] += MinimalWeight[*ei];
DEBUG(dbgs() << "Additionally " << format("%.20g",MinimalWeight[*ei]) << " at " << (*ei) << "\n");
}
printEdgeWeight(*ei);
}
// This block is visited, mark this before the recursion.
BBToVisit.erase(BB);
// Recurse into successors.
for (succ_iterator bbi = succ_begin(BB), bbe = succ_end(BB);
bbi != bbe; ++bbi) {
recurseBasicBlock(*bbi);
}
}
/// LowerFormalArguments - transform physical registers into virtual registers
/// and generate load operations for arguments places on the stack.
SDValue
Cpu0TargetLowering::LowerFormalArguments(SDValue Chain,
CallingConv::ID CallConv,
bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals)
const {
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
Cpu0FunctionInfo *Cpu0FI = MF.getInfo<Cpu0FunctionInfo>();
Cpu0FI->setVarArgsFrameIndex(0);
// Used with vargs to acumulate store chains.
std::vector<SDValue> OutChains;
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
getTargetMachine(), ArgLocs, *DAG.getContext());
CCInfo.AnalyzeFormalArguments(Ins, CC_Cpu0);
Function::const_arg_iterator FuncArg =
DAG.getMachineFunction().getFunction()->arg_begin();
int LastFI = 0;// Cpu0FI->LastInArgFI is 0 at the entry of this function.
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i, ++FuncArg) {
CCValAssign &VA = ArgLocs[i];
EVT ValVT = VA.getValVT();
ISD::ArgFlagsTy Flags = Ins[i].Flags;
bool IsRegLoc = VA.isRegLoc();
if (Flags.isByVal()) {
assert(Flags.getByValSize() &&
"ByVal args of size 0 should have been ignored by front-end.");
continue;
}
// sanity check
assert(VA.isMemLoc());
// The stack pointer offset is relative to the caller stack frame.
LastFI = MFI->CreateFixedObject(ValVT.getSizeInBits()/8,
VA.getLocMemOffset(), true);
// Create load nodes to retrieve arguments from the stack
SDValue FIN = DAG.getFrameIndex(LastFI, getPointerTy());
InVals.push_back(DAG.getLoad(ValVT, dl, Chain, FIN,
MachinePointerInfo::getFixedStack(LastFI),
false, false, false, 0));
}
Cpu0FI->setLastInArgFI(LastFI);
// All stores are grouped in one node to allow the matching between
// the size of Ins and InVals. This only happens when on varg functions
if (!OutChains.empty()) {
OutChains.push_back(Chain);
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
&OutChains[0], OutChains.size());
}
return Chain;
}
/// ValueEnumerator - Enumerate module-level information.
ValueEnumerator::ValueEnumerator(const Module *M) {
// Enumerate the global variables.
for (Module::const_global_iterator I = M->global_begin(),
E = M->global_end(); I != E; ++I)
EnumerateValue(I);
// Enumerate the functions.
for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I) {
EnumerateValue(I);
EnumerateAttributes(cast<Function>(I)->getAttributes());
}
// Enumerate the aliases.
for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I)
EnumerateValue(I);
// Remember what is the cutoff between globalvalue's and other constants.
unsigned FirstConstant = Values.size();
// Enumerate the global variable initializers.
for (Module::const_global_iterator I = M->global_begin(),
E = M->global_end(); I != E; ++I)
if (I->hasInitializer())
EnumerateValue(I->getInitializer());
// Enumerate the aliasees.
for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I)
EnumerateValue(I->getAliasee());
// Insert constants and metadata that are named at module level into the slot
// pool so that the module symbol table can refer to them...
EnumerateValueSymbolTable(M->getValueSymbolTable());
EnumerateNamedMetadata(M);
SmallVector<std::pair<unsigned, MDNode*>, 8> MDs;
// Enumerate types used by function bodies and argument lists.
for (Module::const_iterator F = M->begin(), E = M->end(); F != E; ++F) {
for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
I != E; ++I)
EnumerateType(I->getType());
for (Function::const_iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I!=E;++I){
for (User::const_op_iterator OI = I->op_begin(), E = I->op_end();
OI != E; ++OI) {
if (MDNode *MD = dyn_cast<MDNode>(*OI))
if (MD->isFunctionLocal() && MD->getFunction())
// These will get enumerated during function-incorporation.
continue;
EnumerateOperandType(*OI);
}
EnumerateType(I->getType());
if (const CallInst *CI = dyn_cast<CallInst>(I))
EnumerateAttributes(CI->getAttributes());
else if (const InvokeInst *II = dyn_cast<InvokeInst>(I))
EnumerateAttributes(II->getAttributes());
// Enumerate metadata attached with this instruction.
MDs.clear();
I->getAllMetadataOtherThanDebugLoc(MDs);
for (unsigned i = 0, e = MDs.size(); i != e; ++i)
EnumerateMetadata(MDs[i].second);
if (!I->getDebugLoc().isUnknown()) {
MDNode *Scope, *IA;
I->getDebugLoc().getScopeAndInlinedAt(Scope, IA, I->getContext());
if (Scope) EnumerateMetadata(Scope);
if (IA) EnumerateMetadata(IA);
}
}
}
// Optimize constant ordering.
OptimizeConstants(FirstConstant, Values.size());
}
SDValue
Cpu0TargetLowering::LowerCall(SDValue InChain, SDValue Callee,
CallingConv::ID CallConv, bool isVarArg,
bool doesNotRet, bool &isTailCall,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
// Cpu0 target does not yet support tail call optimization.
isTailCall = false;
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
const TargetFrameLowering *TFL = MF.getTarget().getFrameLowering();
bool IsPIC = getTargetMachine().getRelocationModel() == Reloc::PIC_;
Cpu0FunctionInfo *Cpu0FI = MF.getInfo<Cpu0FunctionInfo>();
// Analyze operands of the call, assigning locations to each operand.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
getTargetMachine(), ArgLocs, *DAG.getContext());
CCInfo.AnalyzeCallOperands(Outs, CC_Cpu0);
// Get a count of how many bytes are to be pushed on the stack.
unsigned NextStackOffset = CCInfo.getNextStackOffset();
// If this is the first call, create a stack frame object that points to
// a location to which .cprestore saves $gp.
if (IsPIC && Cpu0FI->globalBaseRegFixed() && !Cpu0FI->getGPFI())
Cpu0FI->setGPFI(MFI->CreateFixedObject(4, 0, true));
// Get the frame index of the stack frame object that points to the location
// of dynamically allocated area on the stack.
int DynAllocFI = Cpu0FI->getDynAllocFI();
unsigned MaxCallFrameSize = Cpu0FI->getMaxCallFrameSize();
if (MaxCallFrameSize < NextStackOffset) {
Cpu0FI->setMaxCallFrameSize(NextStackOffset);
// Set the offsets relative to $sp of the $gp restore slot and dynamically
// allocated stack space. These offsets must be aligned to a boundary
// determined by the stack alignment of the ABI.
unsigned StackAlignment = TFL->getStackAlignment();
NextStackOffset = (NextStackOffset + StackAlignment - 1) /
StackAlignment * StackAlignment;
MFI->setObjectOffset(DynAllocFI, NextStackOffset);
}
// Chain is the output chain of the last Load/Store or CopyToReg node.
// ByValChain is the output chain of the last Memcpy node created for copying
// byval arguments to the stack.
SDValue Chain, CallSeqStart, ByValChain;
SDValue NextStackOffsetVal = DAG.getIntPtrConstant(NextStackOffset, true);
Chain = CallSeqStart = DAG.getCALLSEQ_START(InChain, NextStackOffsetVal);
ByValChain = InChain;
// With EABI is it possible to have 16 args on registers.
SmallVector<std::pair<unsigned, SDValue>, 16> RegsToPass;
SmallVector<SDValue, 8> MemOpChains;
int FirstFI = -MFI->getNumFixedObjects() - 1, LastFI = 0;
// Walk the register/memloc assignments, inserting copies/loads.
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
SDValue Arg = OutVals[i];
CCValAssign &VA = ArgLocs[i];
MVT ValVT = VA.getValVT(), LocVT = VA.getLocVT();
ISD::ArgFlagsTy Flags = Outs[i].Flags;
// ByVal Arg.
if (Flags.isByVal()) {
assert("!!!Error!!!, Flags.isByVal()==true");
assert(Flags.getByValSize() &&
"ByVal args of size 0 should have been ignored by front-end.");
continue;
}
// Register can't get to this point...
assert(VA.isMemLoc());
// Create the frame index object for this incoming parameter
LastFI = MFI->CreateFixedObject(ValVT.getSizeInBits()/8,
VA.getLocMemOffset(), true);
SDValue PtrOff = DAG.getFrameIndex(LastFI, getPointerTy());
// emit ISD::STORE whichs stores the
// parameter value to a stack Location
MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
MachinePointerInfo(), false, false, 0));
}
// Extend range of indices of frame objects for outgoing arguments that were
// created during this function call. Skip this step if no such objects were
// created.
if (LastFI)
Cpu0FI->extendOutArgFIRange(FirstFI, LastFI);
// If a memcpy has been created to copy a byval arg to a stack, replace the
// chain input of CallSeqStart with ByValChain.
if (InChain != ByValChain)
DAG.UpdateNodeOperands(CallSeqStart.getNode(), ByValChain,
NextStackOffsetVal);
// Transform all store nodes into one single node because all store
// nodes are independent of each other.
if (!MemOpChains.empty())
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
&MemOpChains[0], MemOpChains.size());
// If the callee is a GlobalAddress/ExternalSymbol node (quite common, every
// direct call is) turn it into a TargetGlobalAddress/TargetExternalSymbol
// node so that legalize doesn't hack it.
unsigned char OpFlag;
bool IsPICCall = IsPIC; // true if calls are translated to jalr $25
bool GlobalOrExternal = false;
SDValue CalleeLo;
if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
OpFlag = IsPICCall ? Cpu0II::MO_GOT_CALL : Cpu0II::MO_NO_FLAG;
Callee = DAG.getTargetGlobalAddress(G->getGlobal(), dl,
getPointerTy(), 0, OpFlag);
GlobalOrExternal = true;
}
else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
if (!IsPIC) // static
OpFlag = Cpu0II::MO_NO_FLAG;
else // O32 & PIC
OpFlag = Cpu0II::MO_GOT_CALL;
Callee = DAG.getTargetExternalSymbol(S->getSymbol(), getPointerTy(),
OpFlag);
GlobalOrExternal = true;
}
SDValue InFlag;
// Create nodes that load address of callee and copy it to T9
if (IsPICCall) {
if (GlobalOrExternal) {
// Load callee address
Callee = DAG.getNode(Cpu0ISD::Wrapper, dl, getPointerTy(),
GetGlobalReg(DAG, getPointerTy()), Callee);
SDValue LoadValue = DAG.getLoad(getPointerTy(), dl, DAG.getEntryNode(),
Callee, MachinePointerInfo::getGOT(),
false, false, false, 0);
// Use GOT+LO if callee has internal linkage.
if (CalleeLo.getNode()) {
SDValue Lo = DAG.getNode(Cpu0ISD::Lo, dl, getPointerTy(), CalleeLo);
Callee = DAG.getNode(ISD::ADD, dl, getPointerTy(), LoadValue, Lo);
} else
Callee = LoadValue;
}
}
// T9 should contain the address of the callee function if
// -reloction-model=pic or it is an indirect call.
if (IsPICCall || !GlobalOrExternal) {
// copy to T9
unsigned T9Reg = Cpu0::T9;
Chain = DAG.getCopyToReg(Chain, dl, T9Reg, Callee, SDValue(0, 0));
InFlag = Chain.getValue(1);
Callee = DAG.getRegister(T9Reg, getPointerTy());
}
// Cpu0JmpLink = #chain, #target_address, #opt_in_flags...
// = Chain, Callee, Reg#1, Reg#2, ...
//
// Returns a chain & a flag for retval copy to use.
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
SmallVector<SDValue, 8> Ops;
Ops.push_back(Chain);
Ops.push_back(Callee);
// Add argument registers to the end of the list so that they are
// known live into the call.
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i)
Ops.push_back(DAG.getRegister(RegsToPass[i].first,
RegsToPass[i].second.getValueType()));
// Add a register mask operand representing the call-preserved registers.
const TargetRegisterInfo *TRI = getTargetMachine().getRegisterInfo();
const uint32_t *Mask = TRI->getCallPreservedMask(CallConv);
assert(Mask && "Missing call preserved mask for calling convention");
Ops.push_back(DAG.getRegisterMask(Mask));
if (InFlag.getNode())
Ops.push_back(InFlag);
Chain = DAG.getNode(Cpu0ISD::JmpLink, dl, NodeTys, &Ops[0], Ops.size());
InFlag = Chain.getValue(1);
// Create the CALLSEQ_END node.
Chain = DAG.getCALLSEQ_END(Chain,
DAG.getIntPtrConstant(NextStackOffset, true),
DAG.getIntPtrConstant(0, true), InFlag);
InFlag = Chain.getValue(1);
// Handle result values, copying them out of physregs into vregs that we
// return.
return LowerCallResult(Chain, InFlag, CallConv, isVarArg,
Ins, dl, DAG, InVals);
}
/// ParseAsmStatement - Parse a GNU extended asm statement.
/// asm-statement:
/// gnu-asm-statement
/// ms-asm-statement
///
/// [GNU] gnu-asm-statement:
/// 'asm' type-qualifier[opt] '(' asm-argument ')' ';'
///
/// [GNU] asm-argument:
/// asm-string-literal
/// asm-string-literal ':' asm-operands[opt]
/// asm-string-literal ':' asm-operands[opt] ':' asm-operands[opt]
/// asm-string-literal ':' asm-operands[opt] ':' asm-operands[opt]
/// ':' asm-clobbers
///
/// [GNU] asm-clobbers:
/// asm-string-literal
/// asm-clobbers ',' asm-string-literal
///
StmtResult Parser::ParseAsmStatement(bool &msAsm) {
assert(Tok.is(tok::kw_asm) && "Not an asm stmt");
SourceLocation AsmLoc = ConsumeToken();
if (getLangOpts().AsmBlocks && !isGCCAsmStatement(Tok)) {
msAsm = true;
return ParseMicrosoftAsmStatement(AsmLoc);
}
DeclSpec DS(AttrFactory);
SourceLocation Loc = Tok.getLocation();
ParseTypeQualifierListOpt(DS, AR_VendorAttributesParsed);
// GNU asms accept, but warn, about type-qualifiers other than volatile.
if (DS.getTypeQualifiers() & DeclSpec::TQ_const)
Diag(Loc, diag::warn_asm_qualifier_ignored) << "const";
if (DS.getTypeQualifiers() & DeclSpec::TQ_restrict)
Diag(Loc, diag::warn_asm_qualifier_ignored) << "restrict";
// FIXME: Once GCC supports _Atomic, check whether it permits it here.
if (DS.getTypeQualifiers() & DeclSpec::TQ_atomic)
Diag(Loc, diag::warn_asm_qualifier_ignored) << "_Atomic";
// Remember if this was a volatile asm.
bool isVolatile = DS.getTypeQualifiers() & DeclSpec::TQ_volatile;
// TODO: support "asm goto" constructs (PR#9295).
if (Tok.is(tok::kw_goto)) {
Diag(Tok, diag::err_asm_goto_not_supported_yet);
SkipUntil(tok::r_paren, StopAtSemi);
return StmtError();
}
if (Tok.isNot(tok::l_paren)) {
Diag(Tok, diag::err_expected_lparen_after) << "asm";
SkipUntil(tok::r_paren, StopAtSemi);
return StmtError();
}
BalancedDelimiterTracker T(*this, tok::l_paren);
T.consumeOpen();
ExprResult AsmString(ParseAsmStringLiteral());
// Check if GNU-style InlineAsm is disabled.
// Error on anything other than empty string.
if (!(getLangOpts().GNUAsm || AsmString.isInvalid())) {
const auto *SL = cast<StringLiteral>(AsmString.get());
if (!SL->getString().trim().empty())
Diag(Loc, diag::err_gnu_inline_asm_disabled);
}
if (AsmString.isInvalid()) {
// Consume up to and including the closing paren.
T.skipToEnd();
return StmtError();
}
SmallVector<IdentifierInfo *, 4> Names;
ExprVector Constraints;
ExprVector Exprs;
ExprVector Clobbers;
if (Tok.is(tok::r_paren)) {
// We have a simple asm expression like 'asm("foo")'.
T.consumeClose();
return Actions.ActOnGCCAsmStmt(AsmLoc, /*isSimple*/ true, isVolatile,
/*NumOutputs*/ 0, /*NumInputs*/ 0, nullptr,
Constraints, Exprs, AsmString.get(),
Clobbers, T.getCloseLocation());
}
// Parse Outputs, if present.
bool AteExtraColon = false;
if (Tok.is(tok::colon) || Tok.is(tok::coloncolon)) {
// In C++ mode, parse "::" like ": :".
AteExtraColon = Tok.is(tok::coloncolon);
ConsumeToken();
if (!AteExtraColon && ParseAsmOperandsOpt(Names, Constraints, Exprs))
return StmtError();
}
unsigned NumOutputs = Names.size();
// Parse Inputs, if present.
if (AteExtraColon || Tok.is(tok::colon) || Tok.is(tok::coloncolon)) {
// In C++ mode, parse "::" like ": :".
if (AteExtraColon)
AteExtraColon = false;
else {
AteExtraColon = Tok.is(tok::coloncolon);
ConsumeToken();
}
if (!AteExtraColon && ParseAsmOperandsOpt(Names, Constraints, Exprs))
return StmtError();
}
assert(Names.size() == Constraints.size() &&
Constraints.size() == Exprs.size() && "Input operand size mismatch!");
unsigned NumInputs = Names.size() - NumOutputs;
// Parse the clobbers, if present.
if (AteExtraColon || Tok.is(tok::colon)) {
if (!AteExtraColon)
ConsumeToken();
// Parse the asm-string list for clobbers if present.
if (Tok.isNot(tok::r_paren)) {
while (1) {
ExprResult Clobber(ParseAsmStringLiteral());
if (Clobber.isInvalid())
break;
Clobbers.push_back(Clobber.get());
if (!TryConsumeToken(tok::comma))
break;
}
}
}
T.consumeClose();
return Actions.ActOnGCCAsmStmt(
AsmLoc, false, isVolatile, NumOutputs, NumInputs, Names.data(),
Constraints, Exprs, AsmString.get(), Clobbers, T.getCloseLocation());
}
/// PropagateConstantsIntoArguments - Look at all uses of the specified
/// function. If all uses are direct call sites, and all pass a particular
/// constant in for an argument, propagate that constant in as the argument.
///
bool IPCP::PropagateConstantsIntoArguments(Function &F) {
if (F.arg_empty() || F.use_empty()) return false; // No arguments? Early exit.
// For each argument, keep track of its constant value and whether it is a
// constant or not. The bool is driven to true when found to be non-constant.
SmallVector<std::pair<Constant*, bool>, 16> ArgumentConstants;
ArgumentConstants.resize(F.arg_size());
unsigned NumNonconstant = 0;
for (Value::use_iterator UI = F.use_begin(), E = F.use_end(); UI != E; ++UI) {
User *U = *UI;
// Ignore blockaddress uses.
if (isa<BlockAddress>(U)) continue;
// Used by a non-instruction, or not the callee of a function, do not
// transform.
if (!isa<CallInst>(U) && !isa<InvokeInst>(U))
return false;
CallSite CS(cast<Instruction>(U));
if (!CS.isCallee(UI))
return false;
// Check out all of the potentially constant arguments. Note that we don't
// inspect varargs here.
CallSite::arg_iterator AI = CS.arg_begin();
Function::arg_iterator Arg = F.arg_begin();
for (unsigned i = 0, e = ArgumentConstants.size(); i != e;
++i, ++AI, ++Arg) {
// If this argument is known non-constant, ignore it.
if (ArgumentConstants[i].second)
continue;
Constant *C = dyn_cast<Constant>(*AI);
if (C && ArgumentConstants[i].first == 0) {
ArgumentConstants[i].first = C; // First constant seen.
} else if (C && ArgumentConstants[i].first == C) {
// Still the constant value we think it is.
} else if (*AI == &*Arg) {
// Ignore recursive calls passing argument down.
} else {
// Argument became non-constant. If all arguments are non-constant now,
// give up on this function.
if (++NumNonconstant == ArgumentConstants.size())
return false;
ArgumentConstants[i].second = true;
}
}
}
// If we got to this point, there is a constant argument!
assert(NumNonconstant != ArgumentConstants.size());
bool MadeChange = false;
Function::arg_iterator AI = F.arg_begin();
for (unsigned i = 0, e = ArgumentConstants.size(); i != e; ++i, ++AI) {
// Do we have a constant argument?
if (ArgumentConstants[i].second || AI->use_empty() ||
(AI->hasByValAttr() && !F.onlyReadsMemory()))
continue;
Value *V = ArgumentConstants[i].first;
if (V == 0) V = UndefValue::get(AI->getType());
AI->replaceAllUsesWith(V);
++NumArgumentsProped;
MadeChange = true;
}
return MadeChange;
}
/// spillAroundUses - insert spill code around each use of Reg.
void InlineSpiller::spillAroundUses(unsigned Reg) {
DEBUG(dbgs() << "spillAroundUses " << PrintReg(Reg) << '\n');
LiveInterval &OldLI = LIS.getInterval(Reg);
// Iterate over instructions using Reg.
for (MachineRegisterInfo::reg_iterator RegI = MRI.reg_begin(Reg);
MachineInstr *MI = RegI.skipBundle();) {
// Debug values are not allowed to affect codegen.
if (MI->isDebugValue()) {
// Modify DBG_VALUE now that the value is in a spill slot.
uint64_t Offset = MI->getOperand(1).getImm();
const MDNode *MDPtr = MI->getOperand(2).getMetadata();
DebugLoc DL = MI->getDebugLoc();
DEBUG(dbgs() << "Modifying debug info due to spill:" << "\t" << *MI);
MachineBasicBlock *MBB = MI->getParent();
BuildMI(*MBB, MBB->erase(MI), DL, TII.get(TargetOpcode::DBG_VALUE))
.addFrameIndex(StackSlot).addImm(Offset).addMetadata(MDPtr);
continue;
}
// Ignore copies to/from snippets. We'll delete them.
if (SnippetCopies.count(MI))
continue;
// Stack slot accesses may coalesce away.
if (coalesceStackAccess(MI, Reg))
continue;
// Analyze instruction.
SmallVector<std::pair<MachineInstr*, unsigned>, 8> Ops;
MIBundleOperands::VirtRegInfo RI =
MIBundleOperands(MI).analyzeVirtReg(Reg, &Ops);
// Find the slot index where this instruction reads and writes OldLI.
// This is usually the def slot, except for tied early clobbers.
SlotIndex Idx = LIS.getInstructionIndex(MI).getRegSlot();
if (VNInfo *VNI = OldLI.getVNInfoAt(Idx.getRegSlot(true)))
if (SlotIndex::isSameInstr(Idx, VNI->def))
Idx = VNI->def;
// Check for a sibling copy.
unsigned SibReg = isFullCopyOf(MI, Reg);
if (SibReg && isSibling(SibReg)) {
// This may actually be a copy between snippets.
if (isRegToSpill(SibReg)) {
DEBUG(dbgs() << "Found new snippet copy: " << *MI);
SnippetCopies.insert(MI);
continue;
}
if (RI.Writes) {
// Hoist the spill of a sib-reg copy.
if (hoistSpill(OldLI, MI)) {
// This COPY is now dead, the value is already in the stack slot.
MI->getOperand(0).setIsDead();
DeadDefs.push_back(MI);
continue;
}
} else {
// This is a reload for a sib-reg copy. Drop spills downstream.
LiveInterval &SibLI = LIS.getInterval(SibReg);
eliminateRedundantSpills(SibLI, SibLI.getVNInfoAt(Idx));
// The COPY will fold to a reload below.
}
}
// Attempt to fold memory ops.
if (foldMemoryOperand(Ops))
continue;
// Allocate interval around instruction.
// FIXME: Infer regclass from instruction alone.
LiveInterval &NewLI = Edit->createFrom(Reg);
NewLI.markNotSpillable();
if (RI.Reads)
insertReload(NewLI, Idx, MI);
// Rewrite instruction operands.
bool hasLiveDef = false;
for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
MachineOperand &MO = Ops[i].first->getOperand(Ops[i].second);
MO.setReg(NewLI.reg);
if (MO.isUse()) {
if (!Ops[i].first->isRegTiedToDefOperand(Ops[i].second))
MO.setIsKill();
} else {
if (!MO.isDead())
hasLiveDef = true;
}
}
DEBUG(dbgs() << "\trewrite: " << Idx << '\t' << *MI);
// FIXME: Use a second vreg if instruction has no tied ops.
if (RI.Writes) {
if (hasLiveDef)
insertSpill(NewLI, OldLI, Idx, MI);
else {
// This instruction defines a dead value. We don't need to spill it,
// but do create a live range for the dead value.
VNInfo *VNI = NewLI.getNextValue(Idx, LIS.getVNInfoAllocator());
NewLI.addRange(LiveRange(Idx, Idx.getDeadSlot(), VNI));
}
}
DEBUG(dbgs() << "\tinterval: " << NewLI << '\n');
}
}
Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
// Ensure that the alloca array size argument has type intptr_t, so that
// any casting is exposed early.
if (DL) {
Type *IntPtrTy = DL->getIntPtrType(AI.getType());
if (AI.getArraySize()->getType() != IntPtrTy) {
Value *V = Builder->CreateIntCast(AI.getArraySize(),
IntPtrTy, false);
AI.setOperand(0, V);
return &AI;
}
}
// Convert: alloca Ty, C - where C is a constant != 1 into: alloca [C x Ty], 1
if (AI.isArrayAllocation()) { // Check C != 1
if (const ConstantInt *C = dyn_cast<ConstantInt>(AI.getArraySize())) {
Type *NewTy =
ArrayType::get(AI.getAllocatedType(), C->getZExtValue());
AllocaInst *New = Builder->CreateAlloca(NewTy, nullptr, AI.getName());
New->setAlignment(AI.getAlignment());
// Scan to the end of the allocation instructions, to skip over a block of
// allocas if possible...also skip interleaved debug info
//
BasicBlock::iterator It = New;
while (isa<AllocaInst>(*It) || isa<DbgInfoIntrinsic>(*It)) ++It;
// Now that I is pointing to the first non-allocation-inst in the block,
// insert our getelementptr instruction...
//
Type *IdxTy = DL
? DL->getIntPtrType(AI.getType())
: Type::getInt64Ty(AI.getContext());
Value *NullIdx = Constant::getNullValue(IdxTy);
Value *Idx[2] = { NullIdx, NullIdx };
Instruction *GEP =
GetElementPtrInst::CreateInBounds(New, Idx, New->getName() + ".sub");
InsertNewInstBefore(GEP, *It);
// Now make everything use the getelementptr instead of the original
// allocation.
return ReplaceInstUsesWith(AI, GEP);
} else if (isa<UndefValue>(AI.getArraySize())) {
return ReplaceInstUsesWith(AI, Constant::getNullValue(AI.getType()));
}
}
if (DL && AI.getAllocatedType()->isSized()) {
// If the alignment is 0 (unspecified), assign it the preferred alignment.
if (AI.getAlignment() == 0)
AI.setAlignment(DL->getPrefTypeAlignment(AI.getAllocatedType()));
// Move all alloca's of zero byte objects to the entry block and merge them
// together. Note that we only do this for alloca's, because malloc should
// allocate and return a unique pointer, even for a zero byte allocation.
if (DL->getTypeAllocSize(AI.getAllocatedType()) == 0) {
// For a zero sized alloca there is no point in doing an array allocation.
// This is helpful if the array size is a complicated expression not used
// elsewhere.
if (AI.isArrayAllocation()) {
AI.setOperand(0, ConstantInt::get(AI.getArraySize()->getType(), 1));
return &AI;
}
// Get the first instruction in the entry block.
BasicBlock &EntryBlock = AI.getParent()->getParent()->getEntryBlock();
Instruction *FirstInst = EntryBlock.getFirstNonPHIOrDbg();
if (FirstInst != &AI) {
// If the entry block doesn't start with a zero-size alloca then move
// this one to the start of the entry block. There is no problem with
// dominance as the array size was forced to a constant earlier already.
AllocaInst *EntryAI = dyn_cast<AllocaInst>(FirstInst);
if (!EntryAI || !EntryAI->getAllocatedType()->isSized() ||
DL->getTypeAllocSize(EntryAI->getAllocatedType()) != 0) {
AI.moveBefore(FirstInst);
return &AI;
}
// If the alignment of the entry block alloca is 0 (unspecified),
// assign it the preferred alignment.
if (EntryAI->getAlignment() == 0)
EntryAI->setAlignment(
DL->getPrefTypeAlignment(EntryAI->getAllocatedType()));
// Replace this zero-sized alloca with the one at the start of the entry
// block after ensuring that the address will be aligned enough for both
// types.
unsigned MaxAlign = std::max(EntryAI->getAlignment(),
AI.getAlignment());
EntryAI->setAlignment(MaxAlign);
if (AI.getType() != EntryAI->getType())
return new BitCastInst(EntryAI, AI.getType());
return ReplaceInstUsesWith(AI, EntryAI);
}
}
}
if (AI.getAlignment()) {
// Check to see if this allocation is only modified by a memcpy/memmove from
// a constant global whose alignment is equal to or exceeds that of the
// allocation. If this is the case, we can change all users to use
// the constant global instead. This is commonly produced by the CFE by
// constructs like "void foo() { int A[] = {1,2,3,4,5,6,7,8,9...}; }" if 'A'
// is only subsequently read.
SmallVector<Instruction *, 4> ToDelete;
if (MemTransferInst *Copy = isOnlyCopiedFromConstantGlobal(&AI, ToDelete)) {
unsigned SourceAlign = getOrEnforceKnownAlignment(Copy->getSource(),
AI.getAlignment(),
DL, AT, &AI, DT);
if (AI.getAlignment() <= SourceAlign) {
DEBUG(dbgs() << "Found alloca equal to global: " << AI << '\n');
DEBUG(dbgs() << " memcpy = " << *Copy << '\n');
for (unsigned i = 0, e = ToDelete.size(); i != e; ++i)
EraseInstFromFunction(*ToDelete[i]);
Constant *TheSrc = cast<Constant>(Copy->getSource());
Constant *Cast
= ConstantExpr::getPointerBitCastOrAddrSpaceCast(TheSrc, AI.getType());
Instruction *NewI = ReplaceInstUsesWith(AI, Cast);
EraseInstFromFunction(*Copy);
++NumGlobalCopies;
return NewI;
}
}
}
// At last, use the generic allocation site handler to aggressively remove
// unused allocas.
return visitAllocSite(AI);
}
/// reMaterializeFor - Attempt to rematerialize before MI instead of reloading.
bool InlineSpiller::reMaterializeFor(LiveInterval &VirtReg,
MachineBasicBlock::iterator MI) {
SlotIndex UseIdx = LIS.getInstructionIndex(MI).getRegSlot(true);
VNInfo *ParentVNI = VirtReg.getVNInfoAt(UseIdx.getBaseIndex());
if (!ParentVNI) {
DEBUG(dbgs() << "\tadding <undef> flags: ");
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
MachineOperand &MO = MI->getOperand(i);
if (MO.isReg() && MO.isUse() && MO.getReg() == VirtReg.reg)
MO.setIsUndef();
}
DEBUG(dbgs() << UseIdx << '\t' << *MI);
return true;
}
if (SnippetCopies.count(MI))
return false;
// Use an OrigVNI from traceSiblingValue when ParentVNI is a sibling copy.
LiveRangeEdit::Remat RM(ParentVNI);
SibValueMap::const_iterator SibI = SibValues.find(ParentVNI);
if (SibI != SibValues.end())
RM.OrigMI = SibI->second.DefMI;
if (!Edit->canRematerializeAt(RM, UseIdx, false)) {
markValueUsed(&VirtReg, ParentVNI);
DEBUG(dbgs() << "\tcannot remat for " << UseIdx << '\t' << *MI);
return false;
}
// If the instruction also writes VirtReg.reg, it had better not require the
// same register for uses and defs.
SmallVector<std::pair<MachineInstr*, unsigned>, 8> Ops;
MIBundleOperands::VirtRegInfo RI =
MIBundleOperands(MI).analyzeVirtReg(VirtReg.reg, &Ops);
if (RI.Tied) {
markValueUsed(&VirtReg, ParentVNI);
DEBUG(dbgs() << "\tcannot remat tied reg: " << UseIdx << '\t' << *MI);
return false;
}
// Before rematerializing into a register for a single instruction, try to
// fold a load into the instruction. That avoids allocating a new register.
if (RM.OrigMI->canFoldAsLoad() &&
foldMemoryOperand(Ops, RM.OrigMI)) {
Edit->markRematerialized(RM.ParentVNI);
++NumFoldedLoads;
return true;
}
// Alocate a new register for the remat.
LiveInterval &NewLI = Edit->createFrom(Original);
NewLI.markNotSpillable();
// Finally we can rematerialize OrigMI before MI.
SlotIndex DefIdx = Edit->rematerializeAt(*MI->getParent(), MI, NewLI.reg, RM,
TRI);
DEBUG(dbgs() << "\tremat: " << DefIdx << '\t'
<< *LIS.getInstructionFromIndex(DefIdx));
// Replace operands
for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
MachineOperand &MO = MI->getOperand(Ops[i].second);
if (MO.isReg() && MO.isUse() && MO.getReg() == VirtReg.reg) {
MO.setReg(NewLI.reg);
MO.setIsKill();
}
}
DEBUG(dbgs() << "\t " << UseIdx << '\t' << *MI);
VNInfo *DefVNI = NewLI.getNextValue(DefIdx, LIS.getVNInfoAllocator());
NewLI.addRange(LiveRange(DefIdx, UseIdx.getRegSlot(), DefVNI));
DEBUG(dbgs() << "\tinterval: " << NewLI << '\n');
++NumRemats;
return true;
}
StmtResult Sema::ActOnGCCAsmStmt(SourceLocation AsmLoc, bool IsSimple,
bool IsVolatile, unsigned NumOutputs,
unsigned NumInputs, IdentifierInfo **Names,
MultiExprArg constraints, MultiExprArg Exprs,
Expr *asmString, MultiExprArg clobbers,
SourceLocation RParenLoc) {
unsigned NumClobbers = clobbers.size();
StringLiteral **Constraints =
reinterpret_cast<StringLiteral**>(constraints.data());
StringLiteral *AsmString = cast<StringLiteral>(asmString);
StringLiteral **Clobbers = reinterpret_cast<StringLiteral**>(clobbers.data());
SmallVector<TargetInfo::ConstraintInfo, 4> OutputConstraintInfos;
// The parser verifies that there is a string literal here.
assert(AsmString->isAscii());
for (unsigned i = 0; i != NumOutputs; i++) {
StringLiteral *Literal = Constraints[i];
assert(Literal->isAscii());
StringRef OutputName;
if (Names[i])
OutputName = Names[i]->getName();
TargetInfo::ConstraintInfo Info(Literal->getString(), OutputName);
if (!Context.getTargetInfo().validateOutputConstraint(Info))
return StmtError(Diag(Literal->getLocStart(),
diag::err_asm_invalid_output_constraint)
<< Info.getConstraintStr());
ExprResult ER = CheckPlaceholderExpr(Exprs[i]);
if (ER.isInvalid())
return StmtError();
Exprs[i] = ER.get();
// Check that the output exprs are valid lvalues.
Expr *OutputExpr = Exprs[i];
// Referring to parameters is not allowed in naked functions.
if (CheckNakedParmReference(OutputExpr, *this))
return StmtError();
OutputConstraintInfos.push_back(Info);
// If this is dependent, just continue.
if (OutputExpr->isTypeDependent())
continue;
Expr::isModifiableLvalueResult IsLV =
OutputExpr->isModifiableLvalue(Context, /*Loc=*/nullptr);
switch (IsLV) {
case Expr::MLV_Valid:
// Cool, this is an lvalue.
break;
case Expr::MLV_ArrayType:
// This is OK too.
break;
case Expr::MLV_LValueCast: {
const Expr *LVal = OutputExpr->IgnoreParenNoopCasts(Context);
if (!getLangOpts().HeinousExtensions) {
Diag(LVal->getLocStart(), diag::err_invalid_asm_cast_lvalue)
<< OutputExpr->getSourceRange();
} else {
Diag(LVal->getLocStart(), diag::warn_invalid_asm_cast_lvalue)
<< OutputExpr->getSourceRange();
}
// Accept, even if we emitted an error diagnostic.
break;
}
case Expr::MLV_IncompleteType:
case Expr::MLV_IncompleteVoidType:
if (RequireCompleteType(OutputExpr->getLocStart(), Exprs[i]->getType(),
diag::err_dereference_incomplete_type))
return StmtError();
default:
return StmtError(Diag(OutputExpr->getLocStart(),
diag::err_asm_invalid_lvalue_in_output)
<< OutputExpr->getSourceRange());
}
unsigned Size = Context.getTypeSize(OutputExpr->getType());
if (!Context.getTargetInfo().validateOutputSize(Literal->getString(),
Size))
return StmtError(Diag(OutputExpr->getLocStart(),
diag::err_asm_invalid_output_size)
<< Info.getConstraintStr());
}
SmallVector<TargetInfo::ConstraintInfo, 4> InputConstraintInfos;
for (unsigned i = NumOutputs, e = NumOutputs + NumInputs; i != e; i++) {
StringLiteral *Literal = Constraints[i];
assert(Literal->isAscii());
StringRef InputName;
if (Names[i])
InputName = Names[i]->getName();
TargetInfo::ConstraintInfo Info(Literal->getString(), InputName);
if (!Context.getTargetInfo().validateInputConstraint(OutputConstraintInfos.data(),
NumOutputs, Info)) {
return StmtError(Diag(Literal->getLocStart(),
diag::err_asm_invalid_input_constraint)
<< Info.getConstraintStr());
}
ExprResult ER = CheckPlaceholderExpr(Exprs[i]);
if (ER.isInvalid())
return StmtError();
Exprs[i] = ER.get();
Expr *InputExpr = Exprs[i];
// Referring to parameters is not allowed in naked functions.
if (CheckNakedParmReference(InputExpr, *this))
return StmtError();
// Only allow void types for memory constraints.
if (Info.allowsMemory() && !Info.allowsRegister()) {
if (CheckAsmLValue(InputExpr, *this))
return StmtError(Diag(InputExpr->getLocStart(),
diag::err_asm_invalid_lvalue_in_input)
<< Info.getConstraintStr()
<< InputExpr->getSourceRange());
} else if (Info.requiresImmediateConstant() && !Info.allowsRegister()) {
llvm::APSInt Result;
if (!InputExpr->EvaluateAsInt(Result, Context))
return StmtError(
Diag(InputExpr->getLocStart(), diag::err_asm_invalid_type_in_input)
<< InputExpr->getType() << Info.getConstraintStr()
<< InputExpr->getSourceRange());
if (Result.slt(Info.getImmConstantMin()) ||
Result.sgt(Info.getImmConstantMax()))
return StmtError(Diag(InputExpr->getLocStart(),
diag::err_invalid_asm_value_for_constraint)
<< Result.toString(10) << Info.getConstraintStr()
<< InputExpr->getSourceRange());
} else {
ExprResult Result = DefaultFunctionArrayLvalueConversion(Exprs[i]);
if (Result.isInvalid())
return StmtError();
Exprs[i] = Result.get();
}
if (Info.allowsRegister()) {
if (InputExpr->getType()->isVoidType()) {
return StmtError(Diag(InputExpr->getLocStart(),
diag::err_asm_invalid_type_in_input)
<< InputExpr->getType() << Info.getConstraintStr()
<< InputExpr->getSourceRange());
}
}
InputConstraintInfos.push_back(Info);
const Type *Ty = Exprs[i]->getType().getTypePtr();
if (Ty->isDependentType())
continue;
if (!Ty->isVoidType() || !Info.allowsMemory())
if (RequireCompleteType(InputExpr->getLocStart(), Exprs[i]->getType(),
diag::err_dereference_incomplete_type))
return StmtError();
unsigned Size = Context.getTypeSize(Ty);
if (!Context.getTargetInfo().validateInputSize(Literal->getString(),
Size))
return StmtError(Diag(InputExpr->getLocStart(),
diag::err_asm_invalid_input_size)
<< Info.getConstraintStr());
}
// Check that the clobbers are valid.
for (unsigned i = 0; i != NumClobbers; i++) {
StringLiteral *Literal = Clobbers[i];
assert(Literal->isAscii());
StringRef Clobber = Literal->getString();
if (!Context.getTargetInfo().isValidClobber(Clobber))
return StmtError(Diag(Literal->getLocStart(),
diag::err_asm_unknown_register_name) << Clobber);
}
GCCAsmStmt *NS =
new (Context) GCCAsmStmt(Context, AsmLoc, IsSimple, IsVolatile, NumOutputs,
NumInputs, Names, Constraints, Exprs.data(),
AsmString, NumClobbers, Clobbers, RParenLoc);
// Validate the asm string, ensuring it makes sense given the operands we
// have.
SmallVector<GCCAsmStmt::AsmStringPiece, 8> Pieces;
unsigned DiagOffs;
if (unsigned DiagID = NS->AnalyzeAsmString(Pieces, Context, DiagOffs)) {
Diag(getLocationOfStringLiteralByte(AsmString, DiagOffs), DiagID)
<< AsmString->getSourceRange();
return StmtError();
}
// Validate constraints and modifiers.
for (unsigned i = 0, e = Pieces.size(); i != e; ++i) {
GCCAsmStmt::AsmStringPiece &Piece = Pieces[i];
if (!Piece.isOperand()) continue;
// Look for the correct constraint index.
unsigned Idx = 0;
unsigned ConstraintIdx = 0;
for (unsigned i = 0, e = NS->getNumOutputs(); i != e; ++i, ++ConstraintIdx) {
TargetInfo::ConstraintInfo &Info = OutputConstraintInfos[i];
if (Idx == Piece.getOperandNo())
break;
++Idx;
if (Info.isReadWrite()) {
if (Idx == Piece.getOperandNo())
break;
++Idx;
}
}
for (unsigned i = 0, e = NS->getNumInputs(); i != e; ++i, ++ConstraintIdx) {
TargetInfo::ConstraintInfo &Info = InputConstraintInfos[i];
if (Idx == Piece.getOperandNo())
break;
++Idx;
if (Info.isReadWrite()) {
if (Idx == Piece.getOperandNo())
break;
++Idx;
}
}
// Now that we have the right indexes go ahead and check.
StringLiteral *Literal = Constraints[ConstraintIdx];
const Type *Ty = Exprs[ConstraintIdx]->getType().getTypePtr();
if (Ty->isDependentType() || Ty->isIncompleteType())
continue;
unsigned Size = Context.getTypeSize(Ty);
std::string SuggestedModifier;
if (!Context.getTargetInfo().validateConstraintModifier(
Literal->getString(), Piece.getModifier(), Size,
SuggestedModifier)) {
Diag(Exprs[ConstraintIdx]->getLocStart(),
diag::warn_asm_mismatched_size_modifier);
if (!SuggestedModifier.empty()) {
auto B = Diag(Piece.getRange().getBegin(),
diag::note_asm_missing_constraint_modifier)
<< SuggestedModifier;
SuggestedModifier = "%" + SuggestedModifier + Piece.getString();
B.AddFixItHint(FixItHint::CreateReplacement(Piece.getRange(),
SuggestedModifier));
}
}
}
// Validate tied input operands for type mismatches.
unsigned NumAlternatives = ~0U;
for (unsigned i = 0, e = OutputConstraintInfos.size(); i != e; ++i) {
TargetInfo::ConstraintInfo &Info = OutputConstraintInfos[i];
StringRef ConstraintStr = Info.getConstraintStr();
unsigned AltCount = ConstraintStr.count(',') + 1;
if (NumAlternatives == ~0U)
NumAlternatives = AltCount;
else if (NumAlternatives != AltCount)
return StmtError(Diag(NS->getOutputExpr(i)->getLocStart(),
diag::err_asm_unexpected_constraint_alternatives)
<< NumAlternatives << AltCount);
}
for (unsigned i = 0, e = InputConstraintInfos.size(); i != e; ++i) {
TargetInfo::ConstraintInfo &Info = InputConstraintInfos[i];
StringRef ConstraintStr = Info.getConstraintStr();
unsigned AltCount = ConstraintStr.count(',') + 1;
if (NumAlternatives == ~0U)
NumAlternatives = AltCount;
else if (NumAlternatives != AltCount)
return StmtError(Diag(NS->getInputExpr(i)->getLocStart(),
diag::err_asm_unexpected_constraint_alternatives)
<< NumAlternatives << AltCount);
// If this is a tied constraint, verify that the output and input have
// either exactly the same type, or that they are int/ptr operands with the
// same size (int/long, int*/long, are ok etc).
if (!Info.hasTiedOperand()) continue;
unsigned TiedTo = Info.getTiedOperand();
unsigned InputOpNo = i+NumOutputs;
Expr *OutputExpr = Exprs[TiedTo];
Expr *InputExpr = Exprs[InputOpNo];
if (OutputExpr->isTypeDependent() || InputExpr->isTypeDependent())
continue;
QualType InTy = InputExpr->getType();
QualType OutTy = OutputExpr->getType();
if (Context.hasSameType(InTy, OutTy))
continue; // All types can be tied to themselves.
// Decide if the input and output are in the same domain (integer/ptr or
// floating point.
enum AsmDomain {
AD_Int, AD_FP, AD_Other
} InputDomain, OutputDomain;
if (InTy->isIntegerType() || InTy->isPointerType())
InputDomain = AD_Int;
else if (InTy->isRealFloatingType())
InputDomain = AD_FP;
else
InputDomain = AD_Other;
if (OutTy->isIntegerType() || OutTy->isPointerType())
OutputDomain = AD_Int;
else if (OutTy->isRealFloatingType())
OutputDomain = AD_FP;
else
OutputDomain = AD_Other;
// They are ok if they are the same size and in the same domain. This
// allows tying things like:
// void* to int*
// void* to int if they are the same size.
// double to long double if they are the same size.
//
uint64_t OutSize = Context.getTypeSize(OutTy);
uint64_t InSize = Context.getTypeSize(InTy);
if (OutSize == InSize && InputDomain == OutputDomain &&
InputDomain != AD_Other)
continue;
// If the smaller input/output operand is not mentioned in the asm string,
// then we can promote the smaller one to a larger input and the asm string
// won't notice.
bool SmallerValueMentioned = false;
// If this is a reference to the input and if the input was the smaller
// one, then we have to reject this asm.
if (isOperandMentioned(InputOpNo, Pieces)) {
// This is a use in the asm string of the smaller operand. Since we
// codegen this by promoting to a wider value, the asm will get printed
// "wrong".
SmallerValueMentioned |= InSize < OutSize;
}
if (isOperandMentioned(TiedTo, Pieces)) {
// If this is a reference to the output, and if the output is the larger
// value, then it's ok because we'll promote the input to the larger type.
SmallerValueMentioned |= OutSize < InSize;
}
// If the smaller value wasn't mentioned in the asm string, and if the
// output was a register, just extend the shorter one to the size of the
// larger one.
if (!SmallerValueMentioned && InputDomain != AD_Other &&
OutputConstraintInfos[TiedTo].allowsRegister())
continue;
// Either both of the operands were mentioned or the smaller one was
// mentioned. One more special case that we'll allow: if the tied input is
// integer, unmentioned, and is a constant, then we'll allow truncating it
// down to the size of the destination.
if (InputDomain == AD_Int && OutputDomain == AD_Int &&
!isOperandMentioned(InputOpNo, Pieces) &&
InputExpr->isEvaluatable(Context)) {
CastKind castKind =
(OutTy->isBooleanType() ? CK_IntegralToBoolean : CK_IntegralCast);
InputExpr = ImpCastExprToType(InputExpr, OutTy, castKind).get();
Exprs[InputOpNo] = InputExpr;
NS->setInputExpr(i, InputExpr);
continue;
}
Diag(InputExpr->getLocStart(),
diag::err_asm_tying_incompatible_types)
<< InTy << OutTy << OutputExpr->getSourceRange()
<< InputExpr->getSourceRange();
return StmtError();
}
return NS;
}
ChainedIncludesSource *ChainedIncludesSource::create(CompilerInstance &CI) {
std::vector<std::string> &includes = CI.getPreprocessorOpts().ChainedIncludes;
assert(!includes.empty() && "No '-chain-include' in options!");
llvm::OwningPtr<ChainedIncludesSource> source(new ChainedIncludesSource());
InputKind IK = CI.getFrontendOpts().Inputs[0].first;
SmallVector<llvm::MemoryBuffer *, 4> serialBufs;
SmallVector<std::string, 4> serialBufNames;
for (unsigned i = 0, e = includes.size(); i != e; ++i) {
bool firstInclude = (i == 0);
llvm::OwningPtr<CompilerInvocation> CInvok;
CInvok.reset(new CompilerInvocation(CI.getInvocation()));
CInvok->getPreprocessorOpts().ChainedIncludes.clear();
CInvok->getPreprocessorOpts().ImplicitPCHInclude.clear();
CInvok->getPreprocessorOpts().ImplicitPTHInclude.clear();
CInvok->getPreprocessorOpts().DisablePCHValidation = true;
CInvok->getPreprocessorOpts().Includes.clear();
CInvok->getPreprocessorOpts().MacroIncludes.clear();
CInvok->getPreprocessorOpts().Macros.clear();
CInvok->getFrontendOpts().Inputs.clear();
CInvok->getFrontendOpts().Inputs.push_back(std::make_pair(IK, includes[i]));
TextDiagnosticPrinter *DiagClient =
new TextDiagnosticPrinter(llvm::errs(), DiagnosticOptions());
llvm::IntrusiveRefCntPtr<DiagnosticIDs> DiagID(new DiagnosticIDs());
llvm::IntrusiveRefCntPtr<DiagnosticsEngine> Diags(
new DiagnosticsEngine(DiagID, DiagClient));
llvm::OwningPtr<CompilerInstance> Clang(new CompilerInstance());
Clang->setInvocation(CInvok.take());
Clang->setDiagnostics(Diags.getPtr());
Clang->setTarget(TargetInfo::CreateTargetInfo(Clang->getDiagnostics(),
Clang->getTargetOpts()));
Clang->createFileManager();
Clang->createSourceManager(Clang->getFileManager());
Clang->createPreprocessor();
Clang->getDiagnosticClient().BeginSourceFile(Clang->getLangOpts(),
&Clang->getPreprocessor());
Clang->createASTContext();
SmallVector<char, 256> serialAST;
llvm::raw_svector_ostream OS(serialAST);
llvm::OwningPtr<ASTConsumer> consumer;
consumer.reset(new PCHGenerator(Clang->getPreprocessor(), "-",
/*IsModule=*/false, /*isysroot=*/"", &OS));
Clang->getASTContext().setASTMutationListener(
consumer->GetASTMutationListener());
Clang->setASTConsumer(consumer.take());
Clang->createSema(TU_Prefix, 0);
if (firstInclude) {
Preprocessor &PP = Clang->getPreprocessor();
PP.getBuiltinInfo().InitializeBuiltins(PP.getIdentifierTable(),
PP.getLangOptions());
} else {
assert(!serialBufs.empty());
SmallVector<llvm::MemoryBuffer *, 4> bufs;
for (unsigned si = 0, se = serialBufs.size(); si != se; ++si) {
bufs.push_back(llvm::MemoryBuffer::getMemBufferCopy(
StringRef(serialBufs[si]->getBufferStart(),
serialBufs[si]->getBufferSize())));
}
std::string pchName = includes[i-1];
llvm::raw_string_ostream os(pchName);
os << ".pch" << i-1;
os.flush();
serialBufNames.push_back(pchName);
llvm::OwningPtr<ExternalASTSource> Reader;
Reader.reset(createASTReader(*Clang, pchName, bufs, serialBufNames,
Clang->getASTConsumer().GetASTDeserializationListener()));
if (!Reader)
return 0;
Clang->getASTContext().setExternalSource(Reader);
}
if (!Clang->InitializeSourceManager(includes[i]))
return 0;
ParseAST(Clang->getSema());
OS.flush();
Clang->getDiagnosticClient().EndSourceFile();
serialBufs.push_back(
llvm::MemoryBuffer::getMemBufferCopy(StringRef(serialAST.data(),
serialAST.size())));
source->CIs.push_back(Clang.take());
}
assert(!serialBufs.empty());
std::string pchName = includes.back() + ".pch-final";
serialBufNames.push_back(pchName);
llvm::OwningPtr<ASTReader> Reader;
Reader.reset(createASTReader(CI, pchName, serialBufs, serialBufNames));
if (!Reader)
return 0;
source->FinalReader.reset(Reader.take());
return source.take();
}
// With -fembed-bitcode, save a copy of the llvm IR as data in the
// __LLVM,__bitcode section.
void clang::EmbedBitcode(llvm::Module *M, const CodeGenOptions &CGOpts,
llvm::MemoryBufferRef Buf) {
if (CGOpts.getEmbedBitcode() == CodeGenOptions::Embed_Off)
return;
// Save llvm.compiler.used and remote it.
SmallVector<Constant*, 2> UsedArray;
SmallSet<GlobalValue*, 4> UsedGlobals;
Type *UsedElementType = Type::getInt8Ty(M->getContext())->getPointerTo(0);
GlobalVariable *Used = collectUsedGlobalVariables(*M, UsedGlobals, true);
for (auto *GV : UsedGlobals) {
if (GV->getName() != "llvm.embedded.module" &&
GV->getName() != "llvm.cmdline")
UsedArray.push_back(
ConstantExpr::getPointerBitCastOrAddrSpaceCast(GV, UsedElementType));
}
if (Used)
Used->eraseFromParent();
// Embed the bitcode for the llvm module.
std::string Data;
ArrayRef<uint8_t> ModuleData;
Triple T(M->getTargetTriple());
// Create a constant that contains the bitcode.
// In case of embedding a marker, ignore the input Buf and use the empty
// ArrayRef. It is also legal to create a bitcode marker even Buf is empty.
if (CGOpts.getEmbedBitcode() != CodeGenOptions::Embed_Marker) {
if (!isBitcode((const unsigned char *)Buf.getBufferStart(),
(const unsigned char *)Buf.getBufferEnd())) {
// If the input is LLVM Assembly, bitcode is produced by serializing
// the module. Use-lists order need to be perserved in this case.
llvm::raw_string_ostream OS(Data);
llvm::WriteBitcodeToFile(M, OS, /* ShouldPreserveUseListOrder */ true);
ModuleData =
ArrayRef<uint8_t>((const uint8_t *)OS.str().data(), OS.str().size());
} else
// If the input is LLVM bitcode, write the input byte stream directly.
ModuleData = ArrayRef<uint8_t>((const uint8_t *)Buf.getBufferStart(),
Buf.getBufferSize());
}
llvm::Constant *ModuleConstant =
llvm::ConstantDataArray::get(M->getContext(), ModuleData);
llvm::GlobalVariable *GV = new llvm::GlobalVariable(
*M, ModuleConstant->getType(), true, llvm::GlobalValue::PrivateLinkage,
ModuleConstant);
GV->setSection(getSectionNameForBitcode(T));
UsedArray.push_back(
ConstantExpr::getPointerBitCastOrAddrSpaceCast(GV, UsedElementType));
if (llvm::GlobalVariable *Old =
M->getGlobalVariable("llvm.embedded.module", true)) {
assert(Old->hasOneUse() &&
"llvm.embedded.module can only be used once in llvm.compiler.used");
GV->takeName(Old);
Old->eraseFromParent();
} else {
GV->setName("llvm.embedded.module");
}
// Skip if only bitcode needs to be embedded.
if (CGOpts.getEmbedBitcode() != CodeGenOptions::Embed_Bitcode) {
// Embed command-line options.
ArrayRef<uint8_t> CmdData(const_cast<uint8_t *>(CGOpts.CmdArgs.data()),
CGOpts.CmdArgs.size());
llvm::Constant *CmdConstant =
llvm::ConstantDataArray::get(M->getContext(), CmdData);
GV = new llvm::GlobalVariable(*M, CmdConstant->getType(), true,
llvm::GlobalValue::PrivateLinkage,
CmdConstant);
GV->setSection(getSectionNameForCommandline(T));
UsedArray.push_back(
ConstantExpr::getPointerBitCastOrAddrSpaceCast(GV, UsedElementType));
if (llvm::GlobalVariable *Old =
M->getGlobalVariable("llvm.cmdline", true)) {
assert(Old->hasOneUse() &&
"llvm.cmdline can only be used once in llvm.compiler.used");
GV->takeName(Old);
Old->eraseFromParent();
} else {
GV->setName("llvm.cmdline");
}
}
if (UsedArray.empty())
return;
// Recreate llvm.compiler.used.
ArrayType *ATy = ArrayType::get(UsedElementType, UsedArray.size());
auto *NewUsed = new GlobalVariable(
*M, ATy, false, llvm::GlobalValue::AppendingLinkage,
llvm::ConstantArray::get(ATy, UsedArray), "llvm.compiler.used");
NewUsed->setSection("llvm.metadata");
}
/// Emit landing pads and actions.
///
/// The general organization of the table is complex, but the basic concepts are
/// easy. First there is a header which describes the location and organization
/// of the three components that follow.
///
/// 1. The landing pad site information describes the range of code covered by
/// the try. In our case it's an accumulation of the ranges covered by the
/// invokes in the try. There is also a reference to the landing pad that
/// handles the exception once processed. Finally an index into the actions
/// table.
/// 2. The action table, in our case, is composed of pairs of type IDs and next
/// action offset. Starting with the action index from the landing pad
/// site, each type ID is checked for a match to the current exception. If
/// it matches then the exception and type id are passed on to the landing
/// pad. Otherwise the next action is looked up. This chain is terminated
/// with a next action of zero. If no type id is found then the frame is
/// unwound and handling continues.
/// 3. Type ID table contains references to all the C++ typeinfo for all
/// catches in the function. This tables is reverse indexed base 1.
void EHStreamer::emitExceptionTable() {
const std::vector<const GlobalValue *> &TypeInfos = MMI->getTypeInfos();
const std::vector<unsigned> &FilterIds = MMI->getFilterIds();
const std::vector<LandingPadInfo> &PadInfos = MMI->getLandingPads();
// Sort the landing pads in order of their type ids. This is used to fold
// duplicate actions.
SmallVector<const LandingPadInfo *, 64> LandingPads;
LandingPads.reserve(PadInfos.size());
for (unsigned i = 0, N = PadInfos.size(); i != N; ++i)
LandingPads.push_back(&PadInfos[i]);
// Order landing pads lexicographically by type id.
std::sort(LandingPads.begin(), LandingPads.end(),
[](const LandingPadInfo *L,
const LandingPadInfo *R) { return L->TypeIds < R->TypeIds; });
// Compute the actions table and gather the first action index for each
// landing pad site.
SmallVector<ActionEntry, 32> Actions;
SmallVector<unsigned, 64> FirstActions;
unsigned SizeActions =
computeActionsTable(LandingPads, Actions, FirstActions);
// Invokes and nounwind calls have entries in PadMap (due to being bracketed
// by try-range labels when lowered). Ordinary calls do not, so appropriate
// try-ranges for them need be deduced when using DWARF exception handling.
RangeMapType PadMap;
for (unsigned i = 0, N = LandingPads.size(); i != N; ++i) {
const LandingPadInfo *LandingPad = LandingPads[i];
for (unsigned j = 0, E = LandingPad->BeginLabels.size(); j != E; ++j) {
MCSymbol *BeginLabel = LandingPad->BeginLabels[j];
assert(!PadMap.count(BeginLabel) && "Duplicate landing pad labels!");
PadRange P = { i, j };
PadMap[BeginLabel] = P;
}
}
// Compute the call-site table.
SmallVector<CallSiteEntry, 64> CallSites;
computeCallSiteTable(CallSites, PadMap, LandingPads, FirstActions);
// Final tallies.
// Call sites.
bool IsSJLJ = Asm->MAI->getExceptionHandlingType() == ExceptionHandling::SjLj;
bool HaveTTData = IsSJLJ ? (!TypeInfos.empty() || !FilterIds.empty()) : true;
unsigned CallSiteTableLength;
if (IsSJLJ)
CallSiteTableLength = 0;
else {
unsigned SiteStartSize = 4; // dwarf::DW_EH_PE_udata4
unsigned SiteLengthSize = 4; // dwarf::DW_EH_PE_udata4
unsigned LandingPadSize = 4; // dwarf::DW_EH_PE_udata4
CallSiteTableLength =
CallSites.size() * (SiteStartSize + SiteLengthSize + LandingPadSize);
}
for (unsigned i = 0, e = CallSites.size(); i < e; ++i) {
CallSiteTableLength += getULEB128Size(CallSites[i].Action);
if (IsSJLJ)
CallSiteTableLength += getULEB128Size(i);
}
// Type infos.
const MCSection *LSDASection = Asm->getObjFileLowering().getLSDASection();
unsigned TTypeEncoding;
unsigned TypeFormatSize;
if (!HaveTTData) {
// For SjLj exceptions, if there is no TypeInfo, then we just explicitly say
// that we're omitting that bit.
TTypeEncoding = dwarf::DW_EH_PE_omit;
// dwarf::DW_EH_PE_absptr
TypeFormatSize = Asm->getDataLayout().getPointerSize();
} else {
// Okay, we have actual filters or typeinfos to emit. As such, we need to
// pick a type encoding for them. We're about to emit a list of pointers to
// typeinfo objects at the end of the LSDA. However, unless we're in static
// mode, this reference will require a relocation by the dynamic linker.
//
// Because of this, we have a couple of options:
//
// 1) If we are in -static mode, we can always use an absolute reference
// from the LSDA, because the static linker will resolve it.
//
// 2) Otherwise, if the LSDA section is writable, we can output the direct
// reference to the typeinfo and allow the dynamic linker to relocate
// it. Since it is in a writable section, the dynamic linker won't
// have a problem.
//
// 3) Finally, if we're in PIC mode and the LDSA section isn't writable,
// we need to use some form of indirection. For example, on Darwin,
// we can output a statically-relocatable reference to a dyld stub. The
// offset to the stub is constant, but the contents are in a section
// that is updated by the dynamic linker. This is easy enough, but we
// need to tell the personality function of the unwinder to indirect
// through the dyld stub.
//
// FIXME: When (3) is actually implemented, we'll have to emit the stubs
// somewhere. This predicate should be moved to a shared location that is
// in target-independent code.
//
TTypeEncoding = Asm->getObjFileLowering().getTTypeEncoding();
TypeFormatSize = Asm->GetSizeOfEncodedValue(TTypeEncoding);
}
// Begin the exception table.
// Sometimes we want not to emit the data into separate section (e.g. ARM
// EHABI). In this case LSDASection will be NULL.
if (LSDASection)
Asm->OutStreamer.SwitchSection(LSDASection);
Asm->EmitAlignment(2);
// Emit the LSDA.
MCSymbol *GCCETSym =
Asm->OutContext.GetOrCreateSymbol(Twine("GCC_except_table")+
Twine(Asm->getFunctionNumber()));
Asm->OutStreamer.EmitLabel(GCCETSym);
Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("exception",
Asm->getFunctionNumber()));
if (IsSJLJ)
Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("_LSDA_",
Asm->getFunctionNumber()));
// Emit the LSDA header.
Asm->EmitEncodingByte(dwarf::DW_EH_PE_omit, "@LPStart");
Asm->EmitEncodingByte(TTypeEncoding, "@TType");
// The type infos need to be aligned. GCC does this by inserting padding just
// before the type infos. However, this changes the size of the exception
// table, so you need to take this into account when you output the exception
// table size. However, the size is output using a variable length encoding.
// So by increasing the size by inserting padding, you may increase the number
// of bytes used for writing the size. If it increases, say by one byte, then
// you now need to output one less byte of padding to get the type infos
// aligned. However this decreases the size of the exception table. This
// changes the value you have to output for the exception table size. Due to
// the variable length encoding, the number of bytes used for writing the
// length may decrease. If so, you then have to increase the amount of
// padding. And so on. If you look carefully at the GCC code you will see that
// it indeed does this in a loop, going on and on until the values stabilize.
// We chose another solution: don't output padding inside the table like GCC
// does, instead output it before the table.
unsigned SizeTypes = TypeInfos.size() * TypeFormatSize;
unsigned CallSiteTableLengthSize = getULEB128Size(CallSiteTableLength);
unsigned TTypeBaseOffset =
sizeof(int8_t) + // Call site format
CallSiteTableLengthSize + // Call site table length size
CallSiteTableLength + // Call site table length
SizeActions + // Actions size
SizeTypes;
unsigned TTypeBaseOffsetSize = getULEB128Size(TTypeBaseOffset);
unsigned TotalSize =
sizeof(int8_t) + // LPStart format
sizeof(int8_t) + // TType format
(HaveTTData ? TTypeBaseOffsetSize : 0) + // TType base offset size
TTypeBaseOffset; // TType base offset
unsigned SizeAlign = (4 - TotalSize) & 3;
if (HaveTTData) {
// Account for any extra padding that will be added to the call site table
// length.
Asm->EmitULEB128(TTypeBaseOffset, "@TType base offset", SizeAlign);
SizeAlign = 0;
}
bool VerboseAsm = Asm->OutStreamer.isVerboseAsm();
// SjLj Exception handling
if (IsSJLJ) {
Asm->EmitEncodingByte(dwarf::DW_EH_PE_udata4, "Call site");
// Add extra padding if it wasn't added to the TType base offset.
Asm->EmitULEB128(CallSiteTableLength, "Call site table length", SizeAlign);
// Emit the landing pad site information.
unsigned idx = 0;
for (SmallVectorImpl<CallSiteEntry>::const_iterator
I = CallSites.begin(), E = CallSites.end(); I != E; ++I, ++idx) {
const CallSiteEntry &S = *I;
// Offset of the landing pad, counted in 16-byte bundles relative to the
// @LPStart address.
if (VerboseAsm) {
Asm->OutStreamer.AddComment(">> Call Site " + Twine(idx) + " <<");
Asm->OutStreamer.AddComment(" On exception at call site "+Twine(idx));
}
Asm->EmitULEB128(idx);
// Offset of the first associated action record, relative to the start of
// the action table. This value is biased by 1 (1 indicates the start of
// the action table), and 0 indicates that there are no actions.
if (VerboseAsm) {
if (S.Action == 0)
Asm->OutStreamer.AddComment(" Action: cleanup");
else
Asm->OutStreamer.AddComment(" Action: " +
Twine((S.Action - 1) / 2 + 1));
}
Asm->EmitULEB128(S.Action);
}
} else {
// DWARF Exception handling
assert(Asm->MAI->usesItaniumLSDAForExceptions());
// The call-site table is a list of all call sites that may throw an
// exception (including C++ 'throw' statements) in the procedure
// fragment. It immediately follows the LSDA header. Each entry indicates,
// for a given call, the first corresponding action record and corresponding
// landing pad.
//
// The table begins with the number of bytes, stored as an LEB128
// compressed, unsigned integer. The records immediately follow the record
// count. They are sorted in increasing call-site address. Each record
// indicates:
//
// * The position of the call-site.
// * The position of the landing pad.
// * The first action record for that call site.
//
// A missing entry in the call-site table indicates that a call is not
// supposed to throw.
// Emit the landing pad call site table.
Asm->EmitEncodingByte(dwarf::DW_EH_PE_udata4, "Call site");
// Add extra padding if it wasn't added to the TType base offset.
Asm->EmitULEB128(CallSiteTableLength, "Call site table length", SizeAlign);
unsigned Entry = 0;
for (SmallVectorImpl<CallSiteEntry>::const_iterator
I = CallSites.begin(), E = CallSites.end(); I != E; ++I) {
const CallSiteEntry &S = *I;
MCSymbol *EHFuncBeginSym =
Asm->GetTempSymbol("eh_func_begin", Asm->getFunctionNumber());
MCSymbol *BeginLabel = S.BeginLabel;
if (!BeginLabel)
BeginLabel = EHFuncBeginSym;
MCSymbol *EndLabel = S.EndLabel;
if (!EndLabel)
EndLabel = Asm->GetTempSymbol("eh_func_end", Asm->getFunctionNumber());
// Offset of the call site relative to the previous call site, counted in
// number of 16-byte bundles. The first call site is counted relative to
// the start of the procedure fragment.
if (VerboseAsm)
Asm->OutStreamer.AddComment(">> Call Site " + Twine(++Entry) + " <<");
Asm->EmitLabelDifference(BeginLabel, EHFuncBeginSym, 4);
if (VerboseAsm)
Asm->OutStreamer.AddComment(Twine(" Call between ") +
BeginLabel->getName() + " and " +
EndLabel->getName());
Asm->EmitLabelDifference(EndLabel, BeginLabel, 4);
// Offset of the landing pad, counted in 16-byte bundles relative to the
// @LPStart address.
if (!S.PadLabel) {
if (VerboseAsm)
Asm->OutStreamer.AddComment(" has no landing pad");
Asm->OutStreamer.EmitIntValue(0, 4/*size*/);
} else {
if (VerboseAsm)
Asm->OutStreamer.AddComment(Twine(" jumps to ") +
S.PadLabel->getName());
Asm->EmitLabelDifference(S.PadLabel, EHFuncBeginSym, 4);
}
// Offset of the first associated action record, relative to the start of
// the action table. This value is biased by 1 (1 indicates the start of
// the action table), and 0 indicates that there are no actions.
if (VerboseAsm) {
if (S.Action == 0)
Asm->OutStreamer.AddComment(" On action: cleanup");
else
Asm->OutStreamer.AddComment(" On action: " +
Twine((S.Action - 1) / 2 + 1));
}
Asm->EmitULEB128(S.Action);
}
}
// Emit the Action Table.
int Entry = 0;
for (SmallVectorImpl<ActionEntry>::const_iterator
I = Actions.begin(), E = Actions.end(); I != E; ++I) {
const ActionEntry &Action = *I;
if (VerboseAsm) {
// Emit comments that decode the action table.
Asm->OutStreamer.AddComment(">> Action Record " + Twine(++Entry) + " <<");
}
// Type Filter
//
// Used by the runtime to match the type of the thrown exception to the
// type of the catch clauses or the types in the exception specification.
if (VerboseAsm) {
if (Action.ValueForTypeID > 0)
Asm->OutStreamer.AddComment(" Catch TypeInfo " +
Twine(Action.ValueForTypeID));
else if (Action.ValueForTypeID < 0)
Asm->OutStreamer.AddComment(" Filter TypeInfo " +
Twine(Action.ValueForTypeID));
else
Asm->OutStreamer.AddComment(" Cleanup");
}
Asm->EmitSLEB128(Action.ValueForTypeID);
// Action Record
//
// Self-relative signed displacement in bytes of the next action record,
// or 0 if there is no next action record.
if (VerboseAsm) {
if (Action.NextAction == 0) {
Asm->OutStreamer.AddComment(" No further actions");
} else {
unsigned NextAction = Entry + (Action.NextAction + 1) / 2;
Asm->OutStreamer.AddComment(" Continue to action "+Twine(NextAction));
}
}
Asm->EmitSLEB128(Action.NextAction);
}
emitTypeInfos(TTypeEncoding);
Asm->EmitAlignment(2);
}
/// ParseBlock - Read a block, updating statistics, etc.
static bool ParseBlock(BitstreamCursor &Stream, unsigned BlockID,
unsigned IndentLevel) {
std::string Indent(IndentLevel*2, ' ');
uint64_t BlockBitStart = Stream.GetCurrentBitNo();
// Get the statistics for this BlockID.
PerBlockIDStats &BlockStats = BlockIDStats[BlockID];
BlockStats.NumInstances++;
// BLOCKINFO is a special part of the stream.
if (BlockID == bitc::BLOCKINFO_BLOCK_ID) {
if (Dump) outs() << Indent << "<BLOCKINFO_BLOCK/>\n";
if (Stream.ReadBlockInfoBlock())
return Error("Malformed BlockInfoBlock");
uint64_t BlockBitEnd = Stream.GetCurrentBitNo();
BlockStats.NumBits += BlockBitEnd-BlockBitStart;
return false;
}
unsigned NumWords = 0;
if (Stream.EnterSubBlock(BlockID, &NumWords))
return Error("Malformed block record");
const char *BlockName = 0;
if (Dump) {
outs() << Indent << "<";
if ((BlockName = GetBlockName(BlockID, *Stream.getBitStreamReader())))
outs() << BlockName;
else
outs() << "UnknownBlock" << BlockID;
if (NonSymbolic && BlockName)
outs() << " BlockID=" << BlockID;
outs() << " NumWords=" << NumWords
<< " BlockCodeSize=" << Stream.getAbbrevIDWidth() << ">\n";
}
SmallVector<uint64_t, 64> Record;
// Read all the records for this block.
while (1) {
if (Stream.AtEndOfStream())
return Error("Premature end of bitstream");
uint64_t RecordStartBit = Stream.GetCurrentBitNo();
BitstreamEntry Entry =
Stream.advance(BitstreamCursor::AF_DontAutoprocessAbbrevs);
switch (Entry.Kind) {
case BitstreamEntry::Error:
return Error("malformed bitcode file");
case BitstreamEntry::EndBlock: {
uint64_t BlockBitEnd = Stream.GetCurrentBitNo();
BlockStats.NumBits += BlockBitEnd-BlockBitStart;
if (Dump) {
outs() << Indent << "</";
if (BlockName)
outs() << BlockName << ">\n";
else
outs() << "UnknownBlock" << BlockID << ">\n";
}
return false;
}
case BitstreamEntry::SubBlock: {
uint64_t SubBlockBitStart = Stream.GetCurrentBitNo();
if (ParseBlock(Stream, Entry.ID, IndentLevel+1))
return true;
++BlockStats.NumSubBlocks;
uint64_t SubBlockBitEnd = Stream.GetCurrentBitNo();
// Don't include subblock sizes in the size of this block.
BlockBitStart += SubBlockBitEnd-SubBlockBitStart;
continue;
}
case BitstreamEntry::Record:
// The interesting case.
break;
}
if (Entry.ID == bitc::DEFINE_ABBREV) {
Stream.ReadAbbrevRecord();
++BlockStats.NumAbbrevs;
continue;
}
Record.clear();
++BlockStats.NumRecords;
StringRef Blob;
unsigned Code = Stream.readRecord(Entry.ID, Record, &Blob);
// Increment the # occurrences of this code.
if (BlockStats.CodeFreq.size() <= Code)
BlockStats.CodeFreq.resize(Code+1);
BlockStats.CodeFreq[Code].NumInstances++;
BlockStats.CodeFreq[Code].TotalBits +=
Stream.GetCurrentBitNo()-RecordStartBit;
if (Entry.ID != bitc::UNABBREV_RECORD) {
BlockStats.CodeFreq[Code].NumAbbrev++;
++BlockStats.NumAbbreviatedRecords;
}
if (Dump) {
outs() << Indent << " <";
if (const char *CodeName =
GetCodeName(Code, BlockID, *Stream.getBitStreamReader()))
outs() << CodeName;
else
outs() << "UnknownCode" << Code;
if (NonSymbolic &&
GetCodeName(Code, BlockID, *Stream.getBitStreamReader()))
outs() << " codeid=" << Code;
if (Entry.ID != bitc::UNABBREV_RECORD)
outs() << " abbrevid=" << Entry.ID;
for (unsigned i = 0, e = Record.size(); i != e; ++i)
outs() << " op" << i << "=" << (int64_t)Record[i];
outs() << "/>";
if (Blob.data()) {
outs() << " blob data = ";
bool BlobIsPrintable = true;
for (unsigned i = 0, e = Blob.size(); i != e; ++i)
if (!isprint(static_cast<unsigned char>(Blob[i]))) {
BlobIsPrintable = false;
break;
}
if (BlobIsPrintable)
outs() << "'" << Blob << "'";
else
outs() << "unprintable, " << Blob.size() << " bytes.";
}
outs() << "\n";
}
}
}
SDValue
BlackfinTargetLowering::LowerCall(SDValue Chain, SDValue Callee,
CallingConv::ID CallConv, bool isVarArg,
bool &isTailCall,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
// Blackfin target does not yet support tail call optimization.
isTailCall = false;
// Analyze operands of the call, assigning locations to each operand.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
DAG.getTarget(), ArgLocs, *DAG.getContext());
CCInfo.AllocateStack(12, 4); // ABI requires 12 bytes stack space
CCInfo.AnalyzeCallOperands(Outs, CC_Blackfin);
// Get the size of the outgoing arguments stack space requirement.
unsigned ArgsSize = CCInfo.getNextStackOffset();
Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(ArgsSize, true));
SmallVector<std::pair<unsigned, SDValue>, 8> RegsToPass;
SmallVector<SDValue, 8> MemOpChains;
// Walk the register/memloc assignments, inserting copies/loads.
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
SDValue Arg = OutVals[i];
// Promote the value if needed.
switch (VA.getLocInfo()) {
default: llvm_unreachable("Unknown loc info!");
case CCValAssign::Full: break;
case CCValAssign::SExt:
Arg = DAG.getNode(ISD::SIGN_EXTEND, dl, VA.getLocVT(), Arg);
break;
case CCValAssign::ZExt:
Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, VA.getLocVT(), Arg);
break;
case CCValAssign::AExt:
Arg = DAG.getNode(ISD::ANY_EXTEND, dl, VA.getLocVT(), Arg);
break;
}
// Arguments that can be passed on register must be kept at
// RegsToPass vector
if (VA.isRegLoc()) {
RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
} else {
assert(VA.isMemLoc() && "CCValAssign must be RegLoc or MemLoc");
int Offset = VA.getLocMemOffset();
assert(Offset%4 == 0 && "Unaligned LocMemOffset");
assert(VA.getLocVT()==MVT::i32 && "Illegal CCValAssign type");
SDValue SPN = DAG.getCopyFromReg(Chain, dl, BF::SP, MVT::i32);
SDValue OffsetN = DAG.getIntPtrConstant(Offset);
OffsetN = DAG.getNode(ISD::ADD, dl, MVT::i32, SPN, OffsetN);
MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, OffsetN,
MachinePointerInfo(),false, false, 0));
}
}
// Transform all store nodes into one single node because
// all store nodes are independent of each other.
if (!MemOpChains.empty())
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
&MemOpChains[0], MemOpChains.size());
// Build a sequence of copy-to-reg nodes chained together with token
// chain and flag operands which copy the outgoing args into registers.
// The InFlag in necessary since all emitted instructions must be
// stuck together.
SDValue InFlag;
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
RegsToPass[i].second, InFlag);
InFlag = Chain.getValue(1);
}
// If the callee is a GlobalAddress node (quite common, every direct call is)
// turn it into a TargetGlobalAddress node so that legalize doesn't hack it.
// Likewise ExternalSymbol -> TargetExternalSymbol.
if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee))
Callee = DAG.getTargetGlobalAddress(G->getGlobal(), dl, MVT::i32);
else if (ExternalSymbolSDNode *E = dyn_cast<ExternalSymbolSDNode>(Callee))
Callee = DAG.getTargetExternalSymbol(E->getSymbol(), MVT::i32);
std::vector<EVT> NodeTys;
NodeTys.push_back(MVT::Other); // Returns a chain
NodeTys.push_back(MVT::Glue); // Returns a flag for retval copy to use.
SDValue Ops[] = { Chain, Callee, InFlag };
Chain = DAG.getNode(BFISD::CALL, dl, NodeTys, Ops,
InFlag.getNode() ? 3 : 2);
InFlag = Chain.getValue(1);
Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(ArgsSize, true),
DAG.getIntPtrConstant(0, true), InFlag);
InFlag = Chain.getValue(1);
// Assign locations to each value returned by this call.
SmallVector<CCValAssign, 16> RVLocs;
CCState RVInfo(CallConv, isVarArg, DAG.getMachineFunction(),
DAG.getTarget(), RVLocs, *DAG.getContext());
RVInfo.AnalyzeCallResult(Ins, RetCC_Blackfin);
// Copy all of the result registers out of their specified physreg.
for (unsigned i = 0; i != RVLocs.size(); ++i) {
CCValAssign &RV = RVLocs[i];
unsigned Reg = RV.getLocReg();
Chain = DAG.getCopyFromReg(Chain, dl, Reg,
RVLocs[i].getLocVT(), InFlag);
SDValue Val = Chain.getValue(0);
InFlag = Chain.getValue(2);
Chain = Chain.getValue(1);
// Callee is responsible for extending any i16 return values.
switch (RV.getLocInfo()) {
case CCValAssign::SExt:
Val = DAG.getNode(ISD::AssertSext, dl, RV.getLocVT(), Val,
DAG.getValueType(RV.getValVT()));
break;
case CCValAssign::ZExt:
Val = DAG.getNode(ISD::AssertZext, dl, RV.getLocVT(), Val,
DAG.getValueType(RV.getValVT()));
break;
default:
break;
}
// Truncate to valtype
if (RV.getLocInfo() != CCValAssign::Full)
Val = DAG.getNode(ISD::TRUNCATE, dl, RV.getValVT(), Val);
InVals.push_back(Val);
}
return Chain;
}
TargetMachine *EmitAssemblyHelper::CreateTargetMachine(bool MustCreateTM) {
// Create the TargetMachine for generating code.
std::string Error;
std::string Triple = TheModule->getTargetTriple();
const llvm::Target *TheTarget = TargetRegistry::lookupTarget(Triple, Error);
if (!TheTarget) {
if (MustCreateTM)
Diags.Report(diag::err_fe_unable_to_create_target) << Error;
return 0;
}
// FIXME: Expose these capabilities via actual APIs!!!! Aside from just
// being gross, this is also totally broken if we ever care about
// concurrency.
//TargetMachine::setAsmVerbosityDefault(CodeGenOpts.AsmVerbose);
//TargetMachine::setFunctionSections(CodeGenOpts.FunctionSections);
//TargetMachine::setDataSections (CodeGenOpts.DataSections);
// FIXME: Parse this earlier.
llvm::CodeModel::Model CM;
if (CodeGenOpts.CodeModel == "small") {
CM = llvm::CodeModel::Small;
} else if (CodeGenOpts.CodeModel == "kernel") {
CM = llvm::CodeModel::Kernel;
} else if (CodeGenOpts.CodeModel == "medium") {
CM = llvm::CodeModel::Medium;
} else if (CodeGenOpts.CodeModel == "large") {
CM = llvm::CodeModel::Large;
} else {
assert(CodeGenOpts.CodeModel.empty() && "Invalid code model!");
CM = llvm::CodeModel::Default;
}
SmallVector<const char *, 16> BackendArgs;
BackendArgs.push_back("clang"); // Fake program name.
if (!CodeGenOpts.DebugPass.empty()) {
BackendArgs.push_back("-debug-pass");
BackendArgs.push_back(CodeGenOpts.DebugPass.c_str());
}
if (!CodeGenOpts.LimitFloatPrecision.empty()) {
BackendArgs.push_back("-limit-float-precision");
BackendArgs.push_back(CodeGenOpts.LimitFloatPrecision.c_str());
}
if (llvm::TimePassesIsEnabled)
BackendArgs.push_back("-time-passes");
for (unsigned i = 0, e = CodeGenOpts.BackendOptions.size(); i != e; ++i)
BackendArgs.push_back(CodeGenOpts.BackendOptions[i].c_str());
if (CodeGenOpts.NoGlobalMerge)
BackendArgs.push_back("-global-merge=false");
BackendArgs.push_back(0);
llvm::cl::ParseCommandLineOptions(BackendArgs.size() - 1,
BackendArgs.data());
std::string FeaturesStr;
if (TargetOpts.Features.size()) {
SubtargetFeatures Features;
for (std::vector<std::string>::const_iterator
it = TargetOpts.Features.begin(),
ie = TargetOpts.Features.end(); it != ie; ++it)
Features.AddFeature(*it);
FeaturesStr = Features.getString();
}
llvm::Reloc::Model RM = llvm::Reloc::Default;
if (CodeGenOpts.RelocationModel == "static") {
RM = llvm::Reloc::Static;
} else if (CodeGenOpts.RelocationModel == "pic") {
RM = llvm::Reloc::PIC_;
} else {
assert(CodeGenOpts.RelocationModel == "dynamic-no-pic" &&
"Invalid PIC model!");
RM = llvm::Reloc::DynamicNoPIC;
}
CodeGenOpt::Level OptLevel = CodeGenOpt::Default;
switch (CodeGenOpts.OptimizationLevel) {
default: break;
case 0: OptLevel = CodeGenOpt::None; break;
case 3: OptLevel = CodeGenOpt::Aggressive; break;
}
llvm::TargetOptions Options;
// Set frame pointer elimination mode.
// WARNING: This code is now dead for LLVM trunk checkin r238244
// The checkin in question removes the global variable NoFramePointerElim
// from TargetOptions in favor of using resetTargetOptions. This function
// resides in TargetMachine and called for individual function instances
// via the DAG selection.
#if 0
if (!CodeGenOpts.DisableFPElim) {
Options.NoFramePointerElim = false;
//Options.NoFramePointerElimNonLeaf = false;
} else if (CodeGenOpts.OmitLeafFramePointer) {
Options.NoFramePointerElim = false;
//Options.NoFramePointerElimNonLeaf = true;
} else {
Options.NoFramePointerElim = true;
//Options.NoFramePointerElimNonLeaf = true;
}
#endif
if (CodeGenOpts.UseInitArray)
Options.UseInitArray = true;
// Set float ABI type.
if (CodeGenOpts.FloatABI == "soft" || CodeGenOpts.FloatABI == "softfp")
Options.FloatABIType = llvm::FloatABI::Soft;
else if (CodeGenOpts.FloatABI == "hard")
Options.FloatABIType = llvm::FloatABI::Hard;
else {
assert(CodeGenOpts.FloatABI.empty() && "Invalid float abi!");
Options.FloatABIType = llvm::FloatABI::Default;
}
// Set FP fusion mode.
switch (CodeGenOpts.getFPContractMode()) {
case CodeGenOptions::FPC_Off:
Options.AllowFPOpFusion = llvm::FPOpFusion::Strict;
break;
case CodeGenOptions::FPC_On:
Options.AllowFPOpFusion = llvm::FPOpFusion::Standard;
break;
case CodeGenOptions::FPC_Fast:
Options.AllowFPOpFusion = llvm::FPOpFusion::Fast;
break;
}
Options.LessPreciseFPMADOption = CodeGenOpts.LessPreciseFPMAD;
Options.NoInfsFPMath = CodeGenOpts.NoInfsFPMath;
Options.NoNaNsFPMath = CodeGenOpts.NoNaNsFPMath;
Options.NoZerosInBSS = CodeGenOpts.NoZeroInitializedInBSS;
Options.UnsafeFPMath = CodeGenOpts.UnsafeFPMath;
//TODO : This definition has moved to Module code rather thn TargetOptions
// This is true from LLVM trunk r237079 and beyond
//Options.UseSoftFloat = CodeGenOpts.SoftFloat;
Options.StackAlignmentOverride = CodeGenOpts.StackAlignment;
//Options.RealignStack = CodeGenOpts.StackRealignment;
//Options.DisableTailCalls = CodeGenOpts.DisableTailCalls;
//Options.TrapFuncName = CodeGenOpts.TrapFuncName;
//Options.PositionIndependentExecutable = LangOpts.PIELevel != 0;
//Options.EnableSegmentedStacks = CodeGenOpts.EnableSegmentedStacks;
TargetMachine *TM = TheTarget->createTargetMachine(Triple, TargetOpts.CPU,
FeaturesStr, Options,
RM, CM, OptLevel);
//if (CodeGenOpts.RelaxAll)
// TM->setMCRelaxAll(true);
//if (CodeGenOpts.SaveTempLabels)
// TM->setMCSaveTempLabels(true);
//if (CodeGenOpts.NoDwarf2CFIAsm)
// TM->setMCUseCFI(false);
//if (!CodeGenOpts.NoDwarfDirectoryAsm)
// TM->setMCUseDwarfDirectory(true);
//if (CodeGenOpts.NoExecStack)
// TM->setMCNoExecStack(true);
return TM;
}
/// EmitSchedule - Emit the machine code in scheduled order. Return the new
/// InsertPos and MachineBasicBlock that contains this insertion
/// point. ScheduleDAGSDNodes holds a BB pointer for convenience, but this does
/// not necessarily refer to returned BB. The emitter may split blocks.
MachineBasicBlock *ScheduleDAGSDNodes::
EmitSchedule(MachineBasicBlock::iterator &InsertPos) {
InstrEmitter Emitter(BB, InsertPos);
DenseMap<SDValue, unsigned> VRBaseMap;
DenseMap<SUnit*, unsigned> CopyVRBaseMap;
SmallVector<std::pair<unsigned, MachineInstr*>, 32> Orders;
SmallSet<unsigned, 8> Seen;
bool HasDbg = DAG->hasDebugValues();
// If this is the first BB, emit byval parameter dbg_value's.
if (HasDbg && BB->getParent()->begin() == MachineFunction::iterator(BB)) {
SDDbgInfo::DbgIterator PDI = DAG->ByvalParmDbgBegin();
SDDbgInfo::DbgIterator PDE = DAG->ByvalParmDbgEnd();
for (; PDI != PDE; ++PDI) {
MachineInstr *DbgMI= Emitter.EmitDbgValue(*PDI, VRBaseMap);
if (DbgMI)
BB->insert(InsertPos, DbgMI);
}
}
for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
SUnit *SU = Sequence[i];
if (!SU) {
// Null SUnit* is a noop.
TII->insertNoop(*Emitter.getBlock(), InsertPos);
continue;
}
// For pre-regalloc scheduling, create instructions corresponding to the
// SDNode and any glued SDNodes and append them to the block.
if (!SU->getNode()) {
// Emit a copy.
EmitPhysRegCopy(SU, CopyVRBaseMap, InsertPos);
continue;
}
SmallVector<SDNode *, 4> GluedNodes;
for (SDNode *N = SU->getNode()->getGluedNode(); N; N = N->getGluedNode())
GluedNodes.push_back(N);
while (!GluedNodes.empty()) {
SDNode *N = GluedNodes.back();
Emitter.EmitNode(GluedNodes.back(), SU->OrigNode != SU, SU->isCloned,
VRBaseMap);
// Remember the source order of the inserted instruction.
if (HasDbg)
ProcessSourceNode(N, DAG, Emitter, VRBaseMap, Orders, Seen);
GluedNodes.pop_back();
}
Emitter.EmitNode(SU->getNode(), SU->OrigNode != SU, SU->isCloned,
VRBaseMap);
// Remember the source order of the inserted instruction.
if (HasDbg)
ProcessSourceNode(SU->getNode(), DAG, Emitter, VRBaseMap, Orders,
Seen);
}
// Insert all the dbg_values which have not already been inserted in source
// order sequence.
if (HasDbg) {
MachineBasicBlock::iterator BBBegin = BB->getFirstNonPHI();
// Sort the source order instructions and use the order to insert debug
// values.
std::sort(Orders.begin(), Orders.end(), less_first());
SDDbgInfo::DbgIterator DI = DAG->DbgBegin();
SDDbgInfo::DbgIterator DE = DAG->DbgEnd();
// Now emit the rest according to source order.
unsigned LastOrder = 0;
for (unsigned i = 0, e = Orders.size(); i != e && DI != DE; ++i) {
unsigned Order = Orders[i].first;
MachineInstr *MI = Orders[i].second;
// Insert all SDDbgValue's whose order(s) are before "Order".
if (!MI)
continue;
for (; DI != DE &&
(*DI)->getOrder() >= LastOrder && (*DI)->getOrder() < Order; ++DI) {
if ((*DI)->isInvalidated())
continue;
MachineInstr *DbgMI = Emitter.EmitDbgValue(*DI, VRBaseMap);
if (DbgMI) {
if (!LastOrder)
// Insert to start of the BB (after PHIs).
BB->insert(BBBegin, DbgMI);
else {
// Insert at the instruction, which may be in a different
// block, if the block was split by a custom inserter.
MachineBasicBlock::iterator Pos = MI;
MI->getParent()->insert(Pos, DbgMI);
}
}
}
LastOrder = Order;
}
// Add trailing DbgValue's before the terminator. FIXME: May want to add
// some of them before one or more conditional branches?
SmallVector<MachineInstr*, 8> DbgMIs;
while (DI != DE) {
if (!(*DI)->isInvalidated())
if (MachineInstr *DbgMI = Emitter.EmitDbgValue(*DI, VRBaseMap))
DbgMIs.push_back(DbgMI);
++DI;
}
MachineBasicBlock *InsertBB = Emitter.getBlock();
MachineBasicBlock::iterator Pos = InsertBB->getFirstTerminator();
InsertBB->insert(Pos, DbgMIs.begin(), DbgMIs.end());
}
InsertPos = Emitter.getInsertPos();
return Emitter.getBlock();
}
/// ComputePHILiveOutRegInfo - Compute LiveOutInfo for a PHI's destination
/// register based on the LiveOutInfo of its operands.
void FunctionLoweringInfo::ComputePHILiveOutRegInfo(const PHINode *PN) {
Type *Ty = PN->getType();
if (!Ty->isIntegerTy() || Ty->isVectorTy())
return;
const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering();
SmallVector<EVT, 1> ValueVTs;
ComputeValueVTs(*TLI, Ty, ValueVTs);
assert(ValueVTs.size() == 1 &&
"PHIs with non-vector integer types should have a single VT.");
EVT IntVT = ValueVTs[0];
if (TLI->getNumRegisters(PN->getContext(), IntVT) != 1)
return;
IntVT = TLI->getTypeToTransformTo(PN->getContext(), IntVT);
unsigned BitWidth = IntVT.getSizeInBits();
unsigned DestReg = ValueMap[PN];
if (!TargetRegisterInfo::isVirtualRegister(DestReg))
return;
LiveOutRegInfo.grow(DestReg);
LiveOutInfo &DestLOI = LiveOutRegInfo[DestReg];
Value *V = PN->getIncomingValue(0);
if (isa<UndefValue>(V) || isa<ConstantExpr>(V)) {
DestLOI.NumSignBits = 1;
APInt Zero(BitWidth, 0);
DestLOI.KnownZero = Zero;
DestLOI.KnownOne = Zero;
return;
}
if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
APInt Val = CI->getValue().zextOrTrunc(BitWidth);
DestLOI.NumSignBits = Val.getNumSignBits();
DestLOI.KnownZero = ~Val;
DestLOI.KnownOne = Val;
} else {
assert(ValueMap.count(V) && "V should have been placed in ValueMap when its"
"CopyToReg node was created.");
unsigned SrcReg = ValueMap[V];
if (!TargetRegisterInfo::isVirtualRegister(SrcReg)) {
DestLOI.IsValid = false;
return;
}
const LiveOutInfo *SrcLOI = GetLiveOutRegInfo(SrcReg, BitWidth);
if (!SrcLOI) {
DestLOI.IsValid = false;
return;
}
DestLOI = *SrcLOI;
}
assert(DestLOI.KnownZero.getBitWidth() == BitWidth &&
DestLOI.KnownOne.getBitWidth() == BitWidth &&
"Masks should have the same bit width as the type.");
for (unsigned i = 1, e = PN->getNumIncomingValues(); i != e; ++i) {
Value *V = PN->getIncomingValue(i);
if (isa<UndefValue>(V) || isa<ConstantExpr>(V)) {
DestLOI.NumSignBits = 1;
APInt Zero(BitWidth, 0);
DestLOI.KnownZero = Zero;
DestLOI.KnownOne = Zero;
return;
}
if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
APInt Val = CI->getValue().zextOrTrunc(BitWidth);
DestLOI.NumSignBits = std::min(DestLOI.NumSignBits, Val.getNumSignBits());
DestLOI.KnownZero &= ~Val;
DestLOI.KnownOne &= Val;
continue;
}
assert(ValueMap.count(V) && "V should have been placed in ValueMap when "
"its CopyToReg node was created.");
unsigned SrcReg = ValueMap[V];
if (!TargetRegisterInfo::isVirtualRegister(SrcReg)) {
DestLOI.IsValid = false;
return;
}
const LiveOutInfo *SrcLOI = GetLiveOutRegInfo(SrcReg, BitWidth);
if (!SrcLOI) {
DestLOI.IsValid = false;
return;
}
DestLOI.NumSignBits = std::min(DestLOI.NumSignBits, SrcLOI->NumSignBits);
DestLOI.KnownZero &= SrcLOI->KnownZero;
DestLOI.KnownOne &= SrcLOI->KnownOne;
}
}
/// sink - When an instruction is found to only be used outside of the loop,
/// this function moves it to the exit blocks and patches up SSA form as needed.
/// This method is guaranteed to remove the original instruction from its
/// position, and may either delete it or move it to outside of the loop.
///
void LICM::sink(Instruction &I) {
DEBUG(dbgs() << "LICM sinking instruction: " << I << "\n");
SmallVector<BasicBlock*, 8> ExitBlocks;
CurLoop->getUniqueExitBlocks(ExitBlocks);
if (isa<LoadInst>(I)) ++NumMovedLoads;
else if (isa<CallInst>(I)) ++NumMovedCalls;
++NumSunk;
Changed = true;
// The case where there is only a single exit node of this loop is common
// enough that we handle it as a special (more efficient) case. It is more
// efficient to handle because there are no PHI nodes that need to be placed.
if (ExitBlocks.size() == 1) {
if (!DT->dominates(I.getParent(), ExitBlocks[0])) {
// Instruction is not used, just delete it.
CurAST->deleteValue(&I);
// If I has users in unreachable blocks, eliminate.
// If I is not void type then replaceAllUsesWith undef.
// This allows ValueHandlers and custom metadata to adjust itself.
if (!I.use_empty())
I.replaceAllUsesWith(UndefValue::get(I.getType()));
I.eraseFromParent();
} else {
// Move the instruction to the start of the exit block, after any PHI
// nodes in it.
I.moveBefore(ExitBlocks[0]->getFirstInsertionPt());
// This instruction is no longer in the AST for the current loop, because
// we just sunk it out of the loop. If we just sunk it into an outer
// loop, we will rediscover the operation when we process it.
CurAST->deleteValue(&I);
}
return;
}
if (ExitBlocks.empty()) {
// The instruction is actually dead if there ARE NO exit blocks.
CurAST->deleteValue(&I);
// If I has users in unreachable blocks, eliminate.
// If I is not void type then replaceAllUsesWith undef.
// This allows ValueHandlers and custom metadata to adjust itself.
if (!I.use_empty())
I.replaceAllUsesWith(UndefValue::get(I.getType()));
I.eraseFromParent();
return;
}
// Otherwise, if we have multiple exits, use the SSAUpdater to do all of the
// hard work of inserting PHI nodes as necessary.
SmallVector<PHINode*, 8> NewPHIs;
SSAUpdater SSA(&NewPHIs);
if (!I.use_empty())
SSA.Initialize(I.getType(), I.getName());
// Insert a copy of the instruction in each exit block of the loop that is
// dominated by the instruction. Each exit block is known to only be in the
// ExitBlocks list once.
BasicBlock *InstOrigBB = I.getParent();
unsigned NumInserted = 0;
for (unsigned i = 0, e = ExitBlocks.size(); i != e; ++i) {
BasicBlock *ExitBlock = ExitBlocks[i];
if (!DT->dominates(InstOrigBB, ExitBlock))
continue;
// Insert the code after the last PHI node.
BasicBlock::iterator InsertPt = ExitBlock->getFirstInsertionPt();
// If this is the first exit block processed, just move the original
// instruction, otherwise clone the original instruction and insert
// the copy.
Instruction *New;
if (NumInserted++ == 0) {
I.moveBefore(InsertPt);
New = &I;
} else {
New = I.clone();
if (!I.getName().empty())
New->setName(I.getName()+".le");
ExitBlock->getInstList().insert(InsertPt, New);
}
// Now that we have inserted the instruction, inform SSAUpdater.
if (!I.use_empty())
SSA.AddAvailableValue(ExitBlock, New);
}
// If the instruction doesn't dominate any exit blocks, it must be dead.
if (NumInserted == 0) {
CurAST->deleteValue(&I);
if (!I.use_empty())
I.replaceAllUsesWith(UndefValue::get(I.getType()));
I.eraseFromParent();
return;
}
// Next, rewrite uses of the instruction, inserting PHI nodes as needed.
for (Value::use_iterator UI = I.use_begin(), UE = I.use_end(); UI != UE; ) {
// Grab the use before incrementing the iterator.
Use &U = UI.getUse();
// Increment the iterator before removing the use from the list.
++UI;
SSA.RewriteUseAfterInsertions(U);
}
// Update CurAST for NewPHIs if I had pointer type.
if (I.getType()->isPointerTy())
for (unsigned i = 0, e = NewPHIs.size(); i != e; ++i)
CurAST->copyValue(&I, NewPHIs[i]);
// Finally, remove the instruction from CurAST. It is no longer in the loop.
CurAST->deleteValue(&I);
}
void FunctionLoweringInfo::set(const Function &fn, MachineFunction &mf,
SelectionDAG *DAG) {
const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering();
Fn = &fn;
MF = &mf;
RegInfo = &MF->getRegInfo();
// Check whether the function can return without sret-demotion.
SmallVector<ISD::OutputArg, 4> Outs;
GetReturnInfo(Fn->getReturnType(), Fn->getAttributes(), Outs, *TLI);
CanLowerReturn = TLI->CanLowerReturn(Fn->getCallingConv(), *MF,
Fn->isVarArg(),
Outs, Fn->getContext());
// Initialize the mapping of values to registers. This is only set up for
// instruction values that are used outside of the block that defines
// them.
Function::const_iterator BB = Fn->begin(), EB = Fn->end();
for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I != E; ++I)
if (const AllocaInst *AI = dyn_cast<AllocaInst>(I)) {
// Don't fold inalloca allocas or other dynamic allocas into the initial
// stack frame allocation, even if they are in the entry block.
if (!AI->isStaticAlloca())
continue;
if (const ConstantInt *CUI = dyn_cast<ConstantInt>(AI->getArraySize())) {
Type *Ty = AI->getAllocatedType();
uint64_t TySize = TLI->getDataLayout()->getTypeAllocSize(Ty);
unsigned Align =
std::max((unsigned)TLI->getDataLayout()->getPrefTypeAlignment(Ty),
AI->getAlignment());
TySize *= CUI->getZExtValue(); // Get total allocated size.
if (TySize == 0) TySize = 1; // Don't create zero-sized stack objects.
StaticAllocaMap[AI] =
MF->getFrameInfo()->CreateStackObject(TySize, Align, false, AI);
}
}
for (; BB != EB; ++BB)
for (BasicBlock::const_iterator I = BB->begin(), E = BB->end();
I != E; ++I) {
// Look for dynamic allocas.
if (const AllocaInst *AI = dyn_cast<AllocaInst>(I)) {
if (!AI->isStaticAlloca()) {
unsigned Align = std::max(
(unsigned)TLI->getDataLayout()->getPrefTypeAlignment(
AI->getAllocatedType()),
AI->getAlignment());
unsigned StackAlign =
TM.getSubtargetImpl()->getFrameLowering()->getStackAlignment();
if (Align <= StackAlign)
Align = 0;
// Inform the Frame Information that we have variable-sized objects.
MF->getFrameInfo()->CreateVariableSizedObject(Align ? Align : 1, AI);
}
}
// Look for inline asm that clobbers the SP register.
if (isa<CallInst>(I) || isa<InvokeInst>(I)) {
ImmutableCallSite CS(I);
if (isa<InlineAsm>(CS.getCalledValue())) {
unsigned SP = TLI->getStackPointerRegisterToSaveRestore();
std::vector<TargetLowering::AsmOperandInfo> Ops =
TLI->ParseConstraints(CS);
for (size_t I = 0, E = Ops.size(); I != E; ++I) {
TargetLowering::AsmOperandInfo &Op = Ops[I];
if (Op.Type == InlineAsm::isClobber) {
// Clobbers don't have SDValue operands, hence SDValue().
TLI->ComputeConstraintToUse(Op, SDValue(), DAG);
std::pair<unsigned, const TargetRegisterClass*> PhysReg =
TLI->getRegForInlineAsmConstraint(Op.ConstraintCode,
Op.ConstraintVT);
if (PhysReg.first == SP)
MF->getFrameInfo()->setHasInlineAsmWithSPAdjust(true);
}
}
}
}
// Mark values used outside their block as exported, by allocating
// a virtual register for them.
if (isUsedOutsideOfDefiningBlock(I))
if (!isa<AllocaInst>(I) ||
!StaticAllocaMap.count(cast<AllocaInst>(I)))
InitializeRegForValue(I);
// Collect llvm.dbg.declare information. This is done now instead of
// during the initial isel pass through the IR so that it is done
// in a predictable order.
if (const DbgDeclareInst *DI = dyn_cast<DbgDeclareInst>(I)) {
MachineModuleInfo &MMI = MF->getMMI();
DIVariable DIVar(DI->getVariable());
assert((!DIVar || DIVar.isVariable()) &&
"Variable in DbgDeclareInst should be either null or a DIVariable.");
if (MMI.hasDebugInfo() &&
DIVar &&
!DI->getDebugLoc().isUnknown()) {
// Don't handle byval struct arguments or VLAs, for example.
// Non-byval arguments are handled here (they refer to the stack
// temporary alloca at this point).
const Value *Address = DI->getAddress();
if (Address) {
if (const BitCastInst *BCI = dyn_cast<BitCastInst>(Address))
Address = BCI->getOperand(0);
if (const AllocaInst *AI = dyn_cast<AllocaInst>(Address)) {
DenseMap<const AllocaInst *, int>::iterator SI =
StaticAllocaMap.find(AI);
if (SI != StaticAllocaMap.end()) { // Check for VLAs.
int FI = SI->second;
MMI.setVariableDbgInfo(DI->getVariable(),
FI, DI->getDebugLoc());
}
}
}
}
}
}
// Create an initial MachineBasicBlock for each LLVM BasicBlock in F. This
// also creates the initial PHI MachineInstrs, though none of the input
// operands are populated.
for (BB = Fn->begin(); BB != EB; ++BB) {
MachineBasicBlock *MBB = mf.CreateMachineBasicBlock(BB);
MBBMap[BB] = MBB;
MF->push_back(MBB);
// Transfer the address-taken flag. This is necessary because there could
// be multiple MachineBasicBlocks corresponding to one BasicBlock, and only
// the first one should be marked.
if (BB->hasAddressTaken())
MBB->setHasAddressTaken();
// Create Machine PHI nodes for LLVM PHI nodes, lowering them as
// appropriate.
for (BasicBlock::const_iterator I = BB->begin();
const PHINode *PN = dyn_cast<PHINode>(I); ++I) {
if (PN->use_empty()) continue;
// Skip empty types
if (PN->getType()->isEmptyTy())
continue;
DebugLoc DL = PN->getDebugLoc();
unsigned PHIReg = ValueMap[PN];
assert(PHIReg && "PHI node does not have an assigned virtual register!");
SmallVector<EVT, 4> ValueVTs;
ComputeValueVTs(*TLI, PN->getType(), ValueVTs);
for (unsigned vti = 0, vte = ValueVTs.size(); vti != vte; ++vti) {
EVT VT = ValueVTs[vti];
unsigned NumRegisters = TLI->getNumRegisters(Fn->getContext(), VT);
const TargetInstrInfo *TII = MF->getSubtarget().getInstrInfo();
for (unsigned i = 0; i != NumRegisters; ++i)
BuildMI(MBB, DL, TII->get(TargetOpcode::PHI), PHIReg + i);
PHIReg += NumRegisters;
}
}
}
// Mark landing pad blocks.
for (BB = Fn->begin(); BB != EB; ++BB)
if (const InvokeInst *Invoke = dyn_cast<InvokeInst>(BB->getTerminator()))
MBBMap[Invoke->getSuccessor(1)]->setIsLandingPad();
}
void LiveIntervals::addKillFlags(const VirtRegMap *VRM) {
// Keep track of regunit ranges.
SmallVector<std::pair<LiveInterval*, LiveInterval::iterator>, 8> RU;
for (unsigned i = 0, e = MRI->getNumVirtRegs(); i != e; ++i) {
unsigned Reg = TargetRegisterInfo::index2VirtReg(i);
if (MRI->reg_nodbg_empty(Reg))
continue;
LiveInterval *LI = &getInterval(Reg);
if (LI->empty())
continue;
// Find the regunit intervals for the assigned register. They may overlap
// the virtual register live range, cancelling any kills.
RU.clear();
for (MCRegUnitIterator Units(VRM->getPhys(Reg), TRI); Units.isValid();
++Units) {
LiveInterval *RUInt = &getRegUnit(*Units);
if (RUInt->empty())
continue;
RU.push_back(std::make_pair(RUInt, RUInt->find(LI->begin()->end)));
}
// Every instruction that kills Reg corresponds to a live range end point.
for (LiveInterval::iterator RI = LI->begin(), RE = LI->end(); RI != RE;
++RI) {
// A block index indicates an MBB edge.
if (RI->end.isBlock())
continue;
MachineInstr *MI = getInstructionFromIndex(RI->end);
if (!MI)
continue;
// Check if any of the reguints are live beyond the end of RI. That could
// happen when a physreg is defined as a copy of a virtreg:
//
// %EAX = COPY %vreg5
// FOO %vreg5 <--- MI, cancel kill because %EAX is live.
// BAR %EAX<kill>
//
// There should be no kill flag on FOO when %vreg5 is rewritten as %EAX.
bool CancelKill = false;
for (unsigned u = 0, e = RU.size(); u != e; ++u) {
LiveInterval *RInt = RU[u].first;
LiveInterval::iterator &I = RU[u].second;
if (I == RInt->end())
continue;
I = RInt->advanceTo(I, RI->end);
if (I == RInt->end() || I->start >= RI->end)
continue;
// I is overlapping RI.
CancelKill = true;
break;
}
if (CancelKill)
MI->clearRegisterKills(Reg, NULL);
else
MI->addRegisterKilled(Reg, NULL);
}
}
}
unsigned
JITDwarfEmitter::GetExceptionTableSizeInBytes(MachineFunction* MF) const {
unsigned FinalSize = 0;
// Map all labels and get rid of any dead landing pads.
MMI->TidyLandingPads();
const std::vector<GlobalVariable *> &TypeInfos = MMI->getTypeInfos();
const std::vector<unsigned> &FilterIds = MMI->getFilterIds();
const std::vector<LandingPadInfo> &PadInfos = MMI->getLandingPads();
if (PadInfos.empty()) return 0;
// Sort the landing pads in order of their type ids. This is used to fold
// duplicate actions.
SmallVector<const LandingPadInfo *, 64> LandingPads;
LandingPads.reserve(PadInfos.size());
for (unsigned i = 0, N = PadInfos.size(); i != N; ++i)
LandingPads.push_back(&PadInfos[i]);
std::sort(LandingPads.begin(), LandingPads.end(), PadLT);
// Negative type ids index into FilterIds, positive type ids index into
// TypeInfos. The value written for a positive type id is just the type
// id itself. For a negative type id, however, the value written is the
// (negative) byte offset of the corresponding FilterIds entry. The byte
// offset is usually equal to the type id, because the FilterIds entries
// are written using a variable width encoding which outputs one byte per
// entry as long as the value written is not too large, but can differ.
// This kind of complication does not occur for positive type ids because
// type infos are output using a fixed width encoding.
// FilterOffsets[i] holds the byte offset corresponding to FilterIds[i].
SmallVector<int, 16> FilterOffsets;
FilterOffsets.reserve(FilterIds.size());
int Offset = -1;
for(std::vector<unsigned>::const_iterator I = FilterIds.begin(),
E = FilterIds.end(); I != E; ++I) {
FilterOffsets.push_back(Offset);
Offset -= MCAsmInfo::getULEB128Size(*I);
}
// Compute the actions table and gather the first action index for each
// landing pad site.
SmallVector<ActionEntry, 32> Actions;
SmallVector<unsigned, 64> FirstActions;
FirstActions.reserve(LandingPads.size());
int FirstAction = 0;
unsigned SizeActions = 0;
for (unsigned i = 0, N = LandingPads.size(); i != N; ++i) {
const LandingPadInfo *LP = LandingPads[i];
const std::vector<int> &TypeIds = LP->TypeIds;
const unsigned NumShared = i ? SharedTypeIds(LP, LandingPads[i-1]) : 0;
unsigned SizeSiteActions = 0;
if (NumShared < TypeIds.size()) {
unsigned SizeAction = 0;
ActionEntry *PrevAction = 0;
if (NumShared) {
const unsigned SizePrevIds = LandingPads[i-1]->TypeIds.size();
assert(Actions.size());
PrevAction = &Actions.back();
SizeAction = MCAsmInfo::getSLEB128Size(PrevAction->NextAction) +
MCAsmInfo::getSLEB128Size(PrevAction->ValueForTypeID);
for (unsigned j = NumShared; j != SizePrevIds; ++j) {
SizeAction -= MCAsmInfo::getSLEB128Size(PrevAction->ValueForTypeID);
SizeAction += -PrevAction->NextAction;
PrevAction = PrevAction->Previous;
}
}
// Compute the actions.
for (unsigned I = NumShared, M = TypeIds.size(); I != M; ++I) {
int TypeID = TypeIds[I];
assert(-1-TypeID < (int)FilterOffsets.size() && "Unknown filter id!");
int ValueForTypeID = TypeID < 0 ? FilterOffsets[-1 - TypeID] : TypeID;
unsigned SizeTypeID = MCAsmInfo::getSLEB128Size(ValueForTypeID);
int NextAction = SizeAction ? -(SizeAction + SizeTypeID) : 0;
SizeAction = SizeTypeID + MCAsmInfo::getSLEB128Size(NextAction);
SizeSiteActions += SizeAction;
ActionEntry Action = {ValueForTypeID, NextAction, PrevAction};
Actions.push_back(Action);
PrevAction = &Actions.back();
}
// Record the first action of the landing pad site.
FirstAction = SizeActions + SizeSiteActions - SizeAction + 1;
} // else identical - re-use previous FirstAction
FirstActions.push_back(FirstAction);
// Compute this sites contribution to size.
SizeActions += SizeSiteActions;
}
// Compute the call-site table. Entries must be ordered by address.
SmallVector<CallSiteEntry, 64> CallSites;
RangeMapType PadMap;
for (unsigned i = 0, N = LandingPads.size(); i != N; ++i) {
const LandingPadInfo *LandingPad = LandingPads[i];
for (unsigned j=0, E = LandingPad->BeginLabels.size(); j != E; ++j) {
unsigned BeginLabel = LandingPad->BeginLabels[j];
assert(!PadMap.count(BeginLabel) && "Duplicate landing pad labels!");
PadRange P = { i, j };
PadMap[BeginLabel] = P;
}
}
bool MayThrow = false;
unsigned LastLabel = 0;
for (MachineFunction::const_iterator I = MF->begin(), E = MF->end();
I != E; ++I) {
for (MachineBasicBlock::const_iterator MI = I->begin(), E = I->end();
MI != E; ++MI) {
if (!MI->isLabel()) {
MayThrow |= MI->getDesc().isCall();
continue;
}
unsigned BeginLabel = MI->getOperand(0).getImm();
assert(BeginLabel && "Invalid label!");
if (BeginLabel == LastLabel)
MayThrow = false;
RangeMapType::iterator L = PadMap.find(BeginLabel);
if (L == PadMap.end())
continue;
PadRange P = L->second;
const LandingPadInfo *LandingPad = LandingPads[P.PadIndex];
assert(BeginLabel == LandingPad->BeginLabels[P.RangeIndex] &&
"Inconsistent landing pad map!");
// If some instruction between the previous try-range and this one may
// throw, create a call-site entry with no landing pad for the region
// between the try-ranges.
if (MayThrow) {
CallSiteEntry Site = {LastLabel, BeginLabel, 0, 0};
CallSites.push_back(Site);
}
LastLabel = LandingPad->EndLabels[P.RangeIndex];
CallSiteEntry Site = {BeginLabel, LastLabel,
LandingPad->LandingPadLabel, FirstActions[P.PadIndex]};
assert(Site.BeginLabel && Site.EndLabel && Site.PadLabel &&
"Invalid landing pad!");
// Try to merge with the previous call-site.
if (CallSites.size()) {
CallSiteEntry &Prev = CallSites.back();
if (Site.PadLabel == Prev.PadLabel && Site.Action == Prev.Action) {
// Extend the range of the previous entry.
Prev.EndLabel = Site.EndLabel;
continue;
}
}
// Otherwise, create a new call-site.
CallSites.push_back(Site);
}
}
// If some instruction between the previous try-range and the end of the
// function may throw, create a call-site entry with no landing pad for the
// region following the try-range.
if (MayThrow) {
CallSiteEntry Site = {LastLabel, 0, 0, 0};
CallSites.push_back(Site);
}
// Final tallies.
unsigned SizeSites = CallSites.size() * (sizeof(int32_t) + // Site start.
sizeof(int32_t) + // Site length.
sizeof(int32_t)); // Landing pad.
for (unsigned i = 0, e = CallSites.size(); i < e; ++i)
SizeSites += MCAsmInfo::getULEB128Size(CallSites[i].Action);
unsigned SizeTypes = TypeInfos.size() * TD->getPointerSize();
unsigned TypeOffset = sizeof(int8_t) + // Call site format
// Call-site table length
MCAsmInfo::getULEB128Size(SizeSites) +
SizeSites + SizeActions + SizeTypes;
unsigned TotalSize = sizeof(int8_t) + // LPStart format
sizeof(int8_t) + // TType format
MCAsmInfo::getULEB128Size(TypeOffset) + // TType base offset
TypeOffset;
unsigned SizeAlign = (4 - TotalSize) & 3;
// Begin the exception table.
FinalSize = RoundUpToAlign(FinalSize, 4);
for (unsigned i = 0; i != SizeAlign; ++i) {
++FinalSize;
}
unsigned PointerSize = TD->getPointerSize();
// Emit the header.
++FinalSize;
// Asm->EOL("LPStart format (DW_EH_PE_omit)");
++FinalSize;
// Asm->EOL("TType format (DW_EH_PE_absptr)");
++FinalSize;
// Asm->EOL("TType base offset");
++FinalSize;
// Asm->EOL("Call site format (DW_EH_PE_udata4)");
++FinalSize;
// Asm->EOL("Call-site table length");
// Emit the landing pad site information.
for (unsigned i = 0; i < CallSites.size(); ++i) {
CallSiteEntry &S = CallSites[i];
// Asm->EOL("Region start");
FinalSize += PointerSize;
//Asm->EOL("Region length");
FinalSize += PointerSize;
// Asm->EOL("Landing pad");
FinalSize += PointerSize;
FinalSize += MCAsmInfo::getULEB128Size(S.Action);
// Asm->EOL("Action");
}
// Emit the actions.
for (unsigned I = 0, N = Actions.size(); I != N; ++I) {
ActionEntry &Action = Actions[I];
//Asm->EOL("TypeInfo index");
FinalSize += MCAsmInfo::getSLEB128Size(Action.ValueForTypeID);
//Asm->EOL("Next action");
FinalSize += MCAsmInfo::getSLEB128Size(Action.NextAction);
}
// Emit the type ids.
for (unsigned M = TypeInfos.size(); M; --M) {
// Asm->EOL("TypeInfo");
FinalSize += PointerSize;
}
// Emit the filter typeids.
for (unsigned j = 0, M = FilterIds.size(); j < M; ++j) {
unsigned TypeID = FilterIds[j];
FinalSize += MCAsmInfo::getULEB128Size(TypeID);
//Asm->EOL("Filter TypeInfo index");
}
FinalSize = RoundUpToAlign(FinalSize, 4);
return FinalSize;
}
