#include "llvm/CodeGen/MachineJumpTableInfo.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/CodeGen/PseudoSourceValue.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/Analysis/DebugInfo.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Target/TargetFrameLowering.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetIntrinsicInfo.h"
+#include "llvm/Target/TargetLibraryInfo.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/IntegersSubsetMapping.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
// FP_ROUND's are always exact here.
if (ValueVT.bitsLT(Val.getValueType()))
return DAG.getNode(ISD::FP_ROUND, DL, ValueVT, Val,
- DAG.getIntPtrConstant(1));
+ DAG.getTargetConstant(1, TLI.getPointerTy()));
return DAG.getNode(ISD::FP_EXTEND, DL, ValueVT, Val);
}
return DAG.getNode(ISD::BITCAST, DL, ValueVT, Val);
llvm_unreachable("Unknown mismatch!");
- return SDValue();
}
/// getCopyFromParts - Create a value that contains the specified legal parts
assert(NumParts == 1 && "Do not know what to promote to!");
Val = DAG.getNode(ISD::FP_EXTEND, DL, PartVT, Val);
} else {
- assert(PartVT.isInteger() && ValueVT.isInteger() &&
+ assert((PartVT.isInteger() || PartVT == MVT::x86mmx) &&
+ ValueVT.isInteger() &&
"Unknown mismatch!");
ValueVT = EVT::getIntegerVT(*DAG.getContext(), NumParts * PartBits);
Val = DAG.getNode(ExtendKind, DL, ValueVT, Val);
+ if (PartVT == MVT::x86mmx)
+ Val = DAG.getNode(ISD::BITCAST, DL, PartVT, Val);
}
} else if (PartBits == ValueVT.getSizeInBits()) {
// Different types of the same size.
Val = DAG.getNode(ISD::BITCAST, DL, PartVT, Val);
} else if (NumParts * PartBits < ValueVT.getSizeInBits()) {
// If the parts cover less bits than value has, truncate the value.
- assert(PartVT.isInteger() && ValueVT.isInteger() &&
+ assert((PartVT.isInteger() || PartVT == MVT::x86mmx) &&
+ ValueVT.isInteger() &&
"Unknown mismatch!");
ValueVT = EVT::getIntegerVT(*DAG.getContext(), NumParts * PartBits);
Val = DAG.getNode(ISD::TRUNCATE, DL, ValueVT, Val);
+ if (PartVT == MVT::x86mmx)
+ Val = DAG.getNode(ISD::BITCAST, DL, PartVT, Val);
}
// The value may have changed - recompute ValueVT.
unsigned Flag = InlineAsm::getFlagWord(Code, Regs.size());
if (HasMatching)
Flag = InlineAsm::getFlagWordForMatchingOp(Flag, MatchingIdx);
+ else if (!Regs.empty() &&
+ TargetRegisterInfo::isVirtualRegister(Regs.front())) {
+ // Put the register class of the virtual registers in the flag word. That
+ // way, later passes can recompute register class constraints for inline
+ // assembly as well as normal instructions.
+ // Don't do this for tied operands that can use the regclass information
+ // from the def.
+ const MachineRegisterInfo &MRI = DAG.getMachineFunction().getRegInfo();
+ const TargetRegisterClass *RC = MRI.getRegClass(Regs.front());
+ Flag = InlineAsm::getFlagWordForRegClass(Flag, RC->getID());
+ }
+
SDValue Res = DAG.getTargetConstant(Flag, MVT::i32);
Ops.push_back(Res);
}
}
-void SelectionDAGBuilder::init(GCFunctionInfo *gfi, AliasAnalysis &aa) {
+void SelectionDAGBuilder::init(GCFunctionInfo *gfi, AliasAnalysis &aa,
+ const TargetLibraryInfo *li) {
AA = &aa;
GFI = gfi;
+ LibInfo = li;
TD = DAG.getTarget().getTargetData();
+ LPadToCallSiteMap.clear();
}
/// clear - Clear out the current SelectionDAG and the associated
}
/// clearDanglingDebugInfo - Clear the dangling debug information
-/// map. This function is seperated from the clear so that debug
+/// map. This function is separated from the clear so that debug
/// information that is dangling in a basic block can be properly
/// resolved in a different basic block. This allows the
/// SelectionDAG to resolve dangling debug information attached
DAG.AddDbgValue(SDV, Val.getNode(), false);
}
} else
- DEBUG(dbgs() << "Dropping debug info for " << DI);
+ DEBUG(dbgs() << "Dropping debug info for " << DI << "\n");
DanglingDebugInfoMap[V] = DanglingDebugInfo();
}
}
-// getValue - Return an SDValue for the given Value.
+/// getValue - Return an SDValue for the given Value.
SDValue SelectionDAGBuilder::getValue(const Value *V) {
// If we already have an SDValue for this value, use it. It's important
// to do this first, so that we don't create a CopyFromReg if we already
unsigned InReg = It->second;
RegsForValue RFV(*DAG.getContext(), TLI, InReg, V->getType());
SDValue Chain = DAG.getEntryNode();
- N = RFV.getCopyFromRegs(DAG, FuncInfo, getCurDebugLoc(), Chain,NULL);
+ N = RFV.getCopyFromRegs(DAG, FuncInfo, getCurDebugLoc(), Chain, NULL);
resolveDanglingDebugInfo(V, N);
return N;
}
return DAG.getMergeValues(&Constants[0], Constants.size(),
getCurDebugLoc());
}
+
+ if (const ConstantDataSequential *CDS =
+ dyn_cast<ConstantDataSequential>(C)) {
+ SmallVector<SDValue, 4> Ops;
+ for (unsigned i = 0, e = CDS->getNumElements(); i != e; ++i) {
+ SDNode *Val = getValue(CDS->getElementAsConstant(i)).getNode();
+ // Add each leaf value from the operand to the Constants list
+ // to form a flattened list of all the values.
+ for (unsigned i = 0, e = Val->getNumValues(); i != e; ++i)
+ Ops.push_back(SDValue(Val, i));
+ }
+
+ if (isa<ArrayType>(CDS->getType()))
+ return DAG.getMergeValues(&Ops[0], Ops.size(), getCurDebugLoc());
+ return NodeMap[V] = DAG.getNode(ISD::BUILD_VECTOR, getCurDebugLoc(),
+ VT, &Ops[0], Ops.size());
+ }
if (C->getType()->isStructTy() || C->getType()->isArrayTy()) {
assert((isa<ConstantAggregateZero>(C) || isa<UndefValue>(C)) &&
// Now that we know the number and type of the elements, get that number of
// elements into the Ops array based on what kind of constant it is.
SmallVector<SDValue, 16> Ops;
- if (const ConstantVector *CP = dyn_cast<ConstantVector>(C)) {
+ if (const ConstantVector *CV = dyn_cast<ConstantVector>(C)) {
for (unsigned i = 0; i != NumElements; ++i)
- Ops.push_back(getValue(CP->getOperand(i)));
+ Ops.push_back(getValue(CV->getOperand(i)));
} else {
assert(isa<ConstantAggregateZero>(C) && "Unknown vector constant!");
EVT EltVT = TLI.getValueType(VecTy->getElementType());
}
llvm_unreachable("Can't get register for value!");
- return SDValue();
}
void SelectionDAGBuilder::visitRet(const ReturnInst &I) {
}
/// Return branch probability calculated by BranchProbabilityInfo for IR blocks.
-uint32_t SelectionDAGBuilder::getEdgeWeight(MachineBasicBlock *Src,
- MachineBasicBlock *Dst) {
+uint32_t SelectionDAGBuilder::getEdgeWeight(const MachineBasicBlock *Src,
+ const MachineBasicBlock *Dst) const {
BranchProbabilityInfo *BPI = FuncInfo.BPI;
if (!BPI)
return 0;
Condition = getICmpCondCode(IC->getPredicate());
} else if (const FCmpInst *FC = dyn_cast<FCmpInst>(Cond)) {
Condition = getFCmpCondCode(FC->getPredicate());
+ if (TM.Options.NoNaNsFPMath)
+ Condition = getFCmpCodeWithoutNaN(Condition);
} else {
Condition = ISD::SETEQ; // silence warning.
llvm_unreachable("Unknown compare instruction");
} else
Cond = DAG.getSetCC(dl, MVT::i1, CondLHS, getValue(CB.CmpRHS), CB.CC);
} else {
- assert(CB.CC == ISD::SETLE && "Can handle only LE ranges now");
+ assert(CB.CC == ISD::SETCC_INVALID &&
+ "Condition is undefined for to-the-range belonging check.");
const APInt& Low = cast<ConstantInt>(CB.CmpLHS)->getValue();
const APInt& High = cast<ConstantInt>(CB.CmpRHS)->getValue();
SDValue CmpOp = getValue(CB.CmpMHS);
EVT VT = CmpOp.getValueType();
-
- if (cast<ConstantInt>(CB.CmpLHS)->isMinValue(true)) {
+
+ if (cast<ConstantInt>(CB.CmpLHS)->isMinValue(false)) {
Cond = DAG.getSetCC(dl, MVT::i1, CmpOp, DAG.getConstant(High, VT),
- ISD::SETLE);
+ ISD::SETULE);
} else {
SDValue SUB = DAG.getNode(ISD::SUB, dl,
VT, CmpOp, DAG.getConstant(Low, VT));
UsePtrType = true;
else {
for (unsigned i = 0, e = B.Cases.size(); i != e; ++i)
- if ((uint64_t)((int64_t)B.Cases[i].Mask >> VT.getSizeInBits()) + 1 >= 2) {
+ if (!isUIntN(VT.getSizeInBits(), B.Cases[i].Mask)) {
// Switch table case range are encoded into series of masks.
// Just use pointer type, it's guaranteed to fit.
UsePtrType = true;
CopyToExportRegsIfNeeded(&I);
// Update successor info
- InvokeMBB->addSuccessor(Return);
- InvokeMBB->addSuccessor(LandingPad);
+ addSuccessorWithWeight(InvokeMBB, Return);
+ addSuccessorWithWeight(InvokeMBB, LandingPad);
// Drop into normal successor.
DAG.setRoot(DAG.getNode(ISD::BR, getCurDebugLoc(),
DAG.getBasicBlock(Return)));
}
-void SelectionDAGBuilder::visitUnwind(const UnwindInst &I) {
-}
-
void SelectionDAGBuilder::visitResume(const ResumeInst &RI) {
llvm_unreachable("SelectionDAGBuilder shouldn't visit resume instructions!");
}
MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI();
AddLandingPadInfo(LP, MMI, MBB);
+ // If there aren't registers to copy the values into (e.g., during SjLj
+ // exceptions), then don't bother to create these DAG nodes.
+ if (TLI.getExceptionPointerRegister() == 0 &&
+ TLI.getExceptionSelectorRegister() == 0)
+ return;
+
SmallVector<EVT, 2> ValueVTs;
ComputeValueVTs(TLI, LP.getType(), ValueVTs);
const Value* SV,
MachineBasicBlock *Default,
MachineBasicBlock *SwitchBB) {
- Case& BackCase = *(CR.Range.second-1);
-
// Size is the number of Cases represented by this range.
size_t Size = CR.Range.second - CR.Range.first;
if (Size > 3)
}
}
+ // Order cases by weight so the most likely case will be checked first.
+ BranchProbabilityInfo *BPI = FuncInfo.BPI;
+ if (BPI) {
+ for (CaseItr I = CR.Range.first, IE = CR.Range.second; I != IE; ++I) {
+ uint32_t IWeight = BPI->getEdgeWeight(SwitchBB->getBasicBlock(),
+ I->BB->getBasicBlock());
+ for (CaseItr J = CR.Range.first; J < I; ++J) {
+ uint32_t JWeight = BPI->getEdgeWeight(SwitchBB->getBasicBlock(),
+ J->BB->getBasicBlock());
+ if (IWeight > JWeight)
+ std::swap(*I, *J);
+ }
+ }
+ }
// Rearrange the case blocks so that the last one falls through if possible.
- if (NextBlock && Default != NextBlock && BackCase.BB != NextBlock) {
+ Case &BackCase = *(CR.Range.second-1);
+ if (Size > 1 &&
+ NextBlock && Default != NextBlock && BackCase.BB != NextBlock) {
// The last case block won't fall through into 'NextBlock' if we emit the
// branches in this order. See if rearranging a case value would help.
- for (CaseItr I = CR.Range.first, E = CR.Range.second-1; I != E; ++I) {
+ // We start at the bottom as it's the case with the least weight.
+ for (Case *I = &*(CR.Range.second-2), *E = &*CR.Range.first-1; I != E; --I){
if (I->BB == NextBlock) {
std::swap(*I, BackCase);
break;
CC = ISD::SETEQ;
LHS = SV; RHS = I->High; MHS = NULL;
} else {
- CC = ISD::SETLE;
+ CC = ISD::SETCC_INVALID;
LHS = I->Low; MHS = SV; RHS = I->High;
}
}
static inline bool areJTsAllowed(const TargetLowering &TLI) {
- return !DisableJumpTables &&
+ return !TLI.getTargetMachine().Options.DisableJumpTables &&
(TLI.isOperationLegalOrCustom(ISD::BR_JT, MVT::Other) ||
TLI.isOperationLegalOrCustom(ISD::BRIND, MVT::Other));
}
static APInt ComputeRange(const APInt &First, const APInt &Last) {
uint32_t BitWidth = std::max(Last.getBitWidth(), First.getBitWidth()) + 1;
- APInt LastExt = Last.sext(BitWidth), FirstExt = First.sext(BitWidth);
+ APInt LastExt = Last.zext(BitWidth), FirstExt = First.zext(BitWidth);
return (LastExt - FirstExt + 1ULL);
}
/// handleJTSwitchCase - Emit jumptable for current switch case range
-bool SelectionDAGBuilder::handleJTSwitchCase(CaseRec& CR,
- CaseRecVector& WorkList,
- const Value* SV,
- MachineBasicBlock* Default,
+bool SelectionDAGBuilder::handleJTSwitchCase(CaseRec &CR,
+ CaseRecVector &WorkList,
+ const Value *SV,
+ MachineBasicBlock *Default,
MachineBasicBlock *SwitchBB) {
Case& FrontCase = *CR.Range.first;
Case& BackCase = *(CR.Range.second-1);
const APInt &Last = cast<ConstantInt>(BackCase.High)->getValue();
APInt TSize(First.getBitWidth(), 0);
- for (CaseItr I = CR.Range.first, E = CR.Range.second;
- I!=E; ++I)
+ for (CaseItr I = CR.Range.first, E = CR.Range.second; I != E; ++I)
TSize += I->size();
if (!areJTsAllowed(TLI) || TSize.ult(4))
return false;
APInt Range = ComputeRange(First, Last);
- double Density = TSize.roundToDouble() / Range.roundToDouble();
- if (Density < 0.4)
+ // The density is TSize / Range. Require at least 40%.
+ // It should not be possible for IntTSize to saturate for sane code, but make
+ // sure we handle Range saturation correctly.
+ uint64_t IntRange = Range.getLimitedValue(UINT64_MAX/10);
+ uint64_t IntTSize = TSize.getLimitedValue(UINT64_MAX/10);
+ if (IntTSize * 10 < IntRange * 4)
return false;
DEBUG(dbgs() << "Lowering jump table\n"
<< "First entry: " << First << ". Last entry: " << Last << '\n'
- << "Range: " << Range
- << ". Size: " << TSize << ". Density: " << Density << "\n\n");
+ << "Range: " << Range << ". Size: " << TSize << ".\n\n");
// Get the MachineFunction which holds the current MBB. This is used when
// inserting any additional MBBs necessary to represent the switch.
const APInt &Low = cast<ConstantInt>(I->Low)->getValue();
const APInt &High = cast<ConstantInt>(I->High)->getValue();
- if (Low.sle(TEI) && TEI.sle(High)) {
+ if (Low.ule(TEI) && TEI.ule(High)) {
DestBBs.push_back(I->BB);
if (TEI==High)
++I;
visitJumpTableHeader(JT, JTH, SwitchBB);
JTCases.push_back(JumpTableBlock(JTH, JT));
-
return true;
}
CaseRange LHSR(CR.Range.first, Pivot);
CaseRange RHSR(Pivot, CR.Range.second);
- Constant *C = Pivot->Low;
+ const Constant *C = Pivot->Low;
MachineBasicBlock *FalseBB = 0, *TrueBB = 0;
// We know that we branch to the LHS if the Value being switched on is
// Create a CaseBlock record representing a conditional branch to
// the LHS node if the value being switched on SV is less than C.
// Otherwise, branch to LHS.
- CaseBlock CB(ISD::SETLT, SV, C, NULL, TrueBB, FalseBB, CR.CaseBB);
+ CaseBlock CB(ISD::SETULT, SV, C, NULL, TrueBB, FalseBB, CR.CaseBB);
if (CR.CaseBB == SwitchBB)
visitSwitchCase(CB, SwitchBB);
// Optimize the case where all the case values fit in a
// word without having to subtract minValue. In this case,
// we can optimize away the subtraction.
- if (minValue.isNonNegative() && maxValue.slt(IntPtrBits)) {
+ if (maxValue.ult(IntPtrBits)) {
cmpRange = maxValue;
} else {
lowBound = minValue;
/// Clusterify - Transform simple list of Cases into list of CaseRange's
size_t SelectionDAGBuilder::Clusterify(CaseVector& Cases,
const SwitchInst& SI) {
- size_t numCmps = 0;
+
+ /// Use a shorter form of declaration, and also
+ /// show the we want to use CRSBuilder as Clusterifier.
+ typedef IntegersSubsetMapping<MachineBasicBlock> Clusterifier;
+
+ Clusterifier TheClusterifier;
- BranchProbabilityInfo *BPI = FuncInfo.BPI;
// Start with "simple" cases
- for (size_t i = 1; i < SI.getNumSuccessors(); ++i) {
- BasicBlock *SuccBB = SI.getSuccessor(i);
+ for (SwitchInst::ConstCaseIt i = SI.case_begin(), e = SI.case_end();
+ i != e; ++i) {
+ const BasicBlock *SuccBB = i.getCaseSuccessor();
MachineBasicBlock *SMBB = FuncInfo.MBBMap[SuccBB];
- uint32_t ExtraWeight = BPI ? BPI->getEdgeWeight(SI.getParent(), SuccBB) : 0;
-
- Cases.push_back(Case(SI.getSuccessorValue(i),
- SI.getSuccessorValue(i),
- SMBB, ExtraWeight));
- }
- std::sort(Cases.begin(), Cases.end(), CaseCmp());
-
- // Merge case into clusters
- if (Cases.size() >= 2)
- // Must recompute end() each iteration because it may be
- // invalidated by erase if we hold on to it
- for (CaseItr I = Cases.begin(), J = llvm::next(Cases.begin());
- J != Cases.end(); ) {
- const APInt& nextValue = cast<ConstantInt>(J->Low)->getValue();
- const APInt& currentValue = cast<ConstantInt>(I->High)->getValue();
- MachineBasicBlock* nextBB = J->BB;
- MachineBasicBlock* currentBB = I->BB;
-
- // If the two neighboring cases go to the same destination, merge them
- // into a single case.
- if ((nextValue - currentValue == 1) && (currentBB == nextBB)) {
- I->High = J->High;
- J = Cases.erase(J);
-
- if (BranchProbabilityInfo *BPI = FuncInfo.BPI) {
- uint32_t CurWeight = currentBB->getBasicBlock() ?
- BPI->getEdgeWeight(SI.getParent(), currentBB->getBasicBlock()) : 16;
- uint32_t NextWeight = nextBB->getBasicBlock() ?
- BPI->getEdgeWeight(SI.getParent(), nextBB->getBasicBlock()) : 16;
-
- BPI->setEdgeWeight(SI.getParent(), currentBB->getBasicBlock(),
- CurWeight + NextWeight);
- }
- } else {
- I = J++;
- }
+ TheClusterifier.add(i.getCaseValueEx(), SMBB);
+ }
+
+ TheClusterifier.optimize();
+
+ BranchProbabilityInfo *BPI = FuncInfo.BPI;
+ size_t numCmps = 0;
+ for (Clusterifier::RangeIterator i = TheClusterifier.begin(),
+ e = TheClusterifier.end(); i != e; ++i, ++numCmps) {
+ Clusterifier::Cluster &C = *i;
+ unsigned W = 0;
+ if (BPI) {
+ W = BPI->getEdgeWeight(SI.getParent(), C.second->getBasicBlock());
+ if (!W)
+ W = 16;
+ W *= C.first.Weight;
+ BPI->setEdgeWeight(SI.getParent(), C.second->getBasicBlock(), W);
}
- for (CaseItr I=Cases.begin(), E=Cases.end(); I!=E; ++I, ++numCmps) {
- if (I->Low != I->High)
- // A range counts double, since it requires two compares.
- ++numCmps;
+ // FIXME: Currently work with ConstantInt based numbers.
+ // Changing it to APInt based is a pretty heavy for this commit.
+ Cases.push_back(Case(C.first.getLow().toConstantInt(),
+ C.first.getHigh().toConstantInt(), C.second, W));
+
+ if (C.first.getLow() != C.first.getHigh())
+ // A range counts double, since it requires two compares.
+ ++numCmps;
}
return numCmps;
// If there is only the default destination, branch to it if it is not the
// next basic block. Otherwise, just fall through.
- if (SI.getNumOperands() == 2) {
+ if (!SI.getNumCases()) {
// Update machine-CFG edges.
// If this is not a fall-through branch, emit the branch.
size_t numCmps = Clusterify(Cases, SI);
DEBUG(dbgs() << "Clusterify finished. Total clusters: " << Cases.size()
<< ". Total compares: " << numCmps << '\n');
- numCmps = 0;
+ (void)numCmps;
// Get the Value to be switched on and default basic blocks, which will be
// inserted into CaseBlock records, representing basic blocks in the binary
// search tree.
- const Value *SV = SI.getOperand(0);
+ const Value *SV = SI.getCondition();
// Push the initial CaseRec onto the worklist
CaseRecVector WorkList;
SDValue Op1 = getValue(I.getOperand(0));
SDValue Op2 = getValue(I.getOperand(1));
ISD::CondCode Condition = getFCmpCondCode(predicate);
+ if (TM.Options.NoNaNsFPMath)
+ Condition = getFCmpCodeWithoutNaN(Condition);
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getSetCC(getCurDebugLoc(), DestVT, Op1, Op2, Condition));
}
SDValue Cond = getValue(I.getOperand(0));
SDValue TrueVal = getValue(I.getOperand(1));
SDValue FalseVal = getValue(I.getOperand(2));
+ ISD::NodeType OpCode = Cond.getValueType().isVector() ?
+ ISD::VSELECT : ISD::SELECT;
for (unsigned i = 0; i != NumValues; ++i)
- Values[i] = DAG.getNode(ISD::SELECT, getCurDebugLoc(),
- TrueVal.getNode()->getValueType(TrueVal.getResNo()+i),
+ Values[i] = DAG.getNode(OpCode, getCurDebugLoc(),
+ TrueVal.getNode()->getValueType(TrueVal.getResNo()+i),
Cond,
SDValue(TrueVal.getNode(),
TrueVal.getResNo() + i),
SDValue N = getValue(I.getOperand(0));
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getNode(ISD::FP_ROUND, getCurDebugLoc(),
- DestVT, N, DAG.getIntPtrConstant(0)));
+ DestVT, N,
+ DAG.getTargetConstant(0, TLI.getPointerTy())));
}
void SelectionDAGBuilder::visitFPExt(const User &I){
- // FPTrunc is never a no-op cast, no need to check
+ // FPExt is never a no-op cast, no need to check
SDValue N = getValue(I.getOperand(0));
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getNode(ISD::FP_EXTEND, getCurDebugLoc(), DestVT, N));
TLI.getValueType(I.getType()), InVec, InIdx));
}
-// Utility for visitShuffleVector - Returns true if the mask is mask starting
-// from SIndx and increasing to the element length (undefs are allowed).
-static bool SequentialMask(SmallVectorImpl<int> &Mask, unsigned SIndx) {
- unsigned MaskNumElts = Mask.size();
- for (unsigned i = 0; i != MaskNumElts; ++i)
- if ((Mask[i] >= 0) && (Mask[i] != (int)(i + SIndx)))
+// Utility for visitShuffleVector - Return true if every element in Mask,
+// beginning from position Pos and ending in Pos+Size, falls within the
+// specified sequential range [L, L+Pos). or is undef.
+static bool isSequentialInRange(const SmallVectorImpl<int> &Mask,
+ unsigned Pos, unsigned Size, int Low) {
+ for (unsigned i = Pos, e = Pos+Size; i != e; ++i, ++Low)
+ if (Mask[i] >= 0 && Mask[i] != Low)
return false;
return true;
}
void SelectionDAGBuilder::visitShuffleVector(const User &I) {
- SmallVector<int, 8> Mask;
SDValue Src1 = getValue(I.getOperand(0));
SDValue Src2 = getValue(I.getOperand(1));
- // Convert the ConstantVector mask operand into an array of ints, with -1
- // representing undef values.
- SmallVector<Constant*, 8> MaskElts;
- cast<Constant>(I.getOperand(2))->getVectorElements(MaskElts);
- unsigned MaskNumElts = MaskElts.size();
- for (unsigned i = 0; i != MaskNumElts; ++i) {
- if (isa<UndefValue>(MaskElts[i]))
- Mask.push_back(-1);
- else
- Mask.push_back(cast<ConstantInt>(MaskElts[i])->getSExtValue());
- }
-
+ SmallVector<int, 8> Mask;
+ ShuffleVectorInst::getShuffleMask(cast<Constant>(I.getOperand(2)), Mask);
+ unsigned MaskNumElts = Mask.size();
+
EVT VT = TLI.getValueType(I.getType());
EVT SrcVT = Src1.getValueType();
unsigned SrcNumElts = SrcVT.getVectorNumElements();
// Mask is longer than the source vectors and is a multiple of the source
// vectors. We can use concatenate vector to make the mask and vectors
// lengths match.
- if (SrcNumElts*2 == MaskNumElts && SequentialMask(Mask, 0)) {
- // The shuffle is concatenating two vectors together.
- setValue(&I, DAG.getNode(ISD::CONCAT_VECTORS, getCurDebugLoc(),
- VT, Src1, Src2));
- return;
+ if (SrcNumElts*2 == MaskNumElts) {
+ // First check for Src1 in low and Src2 in high
+ if (isSequentialInRange(Mask, 0, SrcNumElts, 0) &&
+ isSequentialInRange(Mask, SrcNumElts, SrcNumElts, SrcNumElts)) {
+ // The shuffle is concatenating two vectors together.
+ setValue(&I, DAG.getNode(ISD::CONCAT_VECTORS, getCurDebugLoc(),
+ VT, Src1, Src2));
+ return;
+ }
+ // Then check for Src2 in low and Src1 in high
+ if (isSequentialInRange(Mask, 0, SrcNumElts, SrcNumElts) &&
+ isSequentialInRange(Mask, SrcNumElts, SrcNumElts, 0)) {
+ // The shuffle is concatenating two vectors together.
+ setValue(&I, DAG.getNode(ISD::CONCAT_VECTORS, getCurDebugLoc(),
+ VT, Src2, Src1));
+ return;
+ }
}
// Pad both vectors with undefs to make them the same length as the mask.
SmallVector<int, 8> MappedOps;
for (unsigned i = 0; i != MaskNumElts; ++i) {
int Idx = Mask[i];
- if (Idx < (int)SrcNumElts)
- MappedOps.push_back(Idx);
- else
- MappedOps.push_back(Idx + MaskNumElts - SrcNumElts);
+ if (Idx >= (int)SrcNumElts)
+ Idx -= SrcNumElts - MaskNumElts;
+ MappedOps.push_back(Idx);
}
setValue(&I, DAG.getVectorShuffle(VT, getCurDebugLoc(), Src1, Src2,
// Analyze the access pattern of the vector to see if we can extract
// two subvectors and do the shuffle. The analysis is done by calculating
// the range of elements the mask access on both vectors.
- int MinRange[2] = { static_cast<int>(SrcNumElts+1),
- static_cast<int>(SrcNumElts+1)};
+ int MinRange[2] = { static_cast<int>(SrcNumElts),
+ static_cast<int>(SrcNumElts)};
int MaxRange[2] = {-1, -1};
for (unsigned i = 0; i != MaskNumElts; ++i) {
int Idx = Mask[i];
- int Input = 0;
+ unsigned Input = 0;
if (Idx < 0)
continue;
// Check if the access is smaller than the vector size and can we find
// a reasonable extract index.
- int RangeUse[2] = { 2, 2 }; // 0 = Unused, 1 = Extract, 2 = Can not
- // Extract.
+ int RangeUse[2] = { -1, -1 }; // 0 = Unused, 1 = Extract, -1 = Can not
+ // Extract.
int StartIdx[2]; // StartIdx to extract from
- for (int Input=0; Input < 2; ++Input) {
- if (MinRange[Input] == (int)(SrcNumElts+1) && MaxRange[Input] == -1) {
+ for (unsigned Input = 0; Input < 2; ++Input) {
+ if (MinRange[Input] >= (int)SrcNumElts && MaxRange[Input] < 0) {
RangeUse[Input] = 0; // Unused
StartIdx[Input] = 0;
- } else if (MaxRange[Input] - MinRange[Input] < (int)MaskNumElts) {
- // Fits within range but we should see if we can find a good
- // start index that is a multiple of the mask length.
- if (MaxRange[Input] < (int)MaskNumElts) {
- RangeUse[Input] = 1; // Extract from beginning of the vector
- StartIdx[Input] = 0;
- } else {
- StartIdx[Input] = (MinRange[Input]/MaskNumElts)*MaskNumElts;
- if (MaxRange[Input] - StartIdx[Input] < (int)MaskNumElts &&
- StartIdx[Input] + MaskNumElts <= SrcNumElts)
- RangeUse[Input] = 1; // Extract from a multiple of the mask length.
- }
+ continue;
}
+
+ // Find a good start index that is a multiple of the mask length. Then
+ // see if the rest of the elements are in range.
+ StartIdx[Input] = (MinRange[Input]/MaskNumElts)*MaskNumElts;
+ if (MaxRange[Input] - StartIdx[Input] < (int)MaskNumElts &&
+ StartIdx[Input] + MaskNumElts <= SrcNumElts)
+ RangeUse[Input] = 1; // Extract from a multiple of the mask length.
}
if (RangeUse[0] == 0 && RangeUse[1] == 0) {
setValue(&I, DAG.getUNDEF(VT)); // Vectors are not used.
return;
}
- else if (RangeUse[0] < 2 && RangeUse[1] < 2) {
+ if (RangeUse[0] >= 0 && RangeUse[1] >= 0) {
// Extract appropriate subvector and generate a vector shuffle
- for (int Input=0; Input < 2; ++Input) {
+ for (unsigned Input = 0; Input < 2; ++Input) {
SDValue &Src = Input == 0 ? Src1 : Src2;
if (RangeUse[Input] == 0)
Src = DAG.getUNDEF(VT);
SmallVector<int, 8> MappedOps;
for (unsigned i = 0; i != MaskNumElts; ++i) {
int Idx = Mask[i];
- if (Idx < 0)
- MappedOps.push_back(Idx);
- else if (Idx < (int)SrcNumElts)
- MappedOps.push_back(Idx - StartIdx[0]);
- else
- MappedOps.push_back(Idx - SrcNumElts - StartIdx[1] + MaskNumElts);
+ if (Idx >= 0) {
+ if (Idx < (int)SrcNumElts)
+ Idx -= StartIdx[0];
+ else
+ Idx -= SrcNumElts + StartIdx[1] - MaskNumElts;
+ }
+ MappedOps.push_back(Idx);
}
setValue(&I, DAG.getVectorShuffle(VT, getCurDebugLoc(), Src1, Src2,
EVT PtrVT = TLI.getPointerTy();
SmallVector<SDValue,8> Ops;
for (unsigned i = 0; i != MaskNumElts; ++i) {
- if (Mask[i] < 0) {
- Ops.push_back(DAG.getUNDEF(EltVT));
- } else {
- int Idx = Mask[i];
- SDValue Res;
+ int Idx = Mask[i];
+ SDValue Res;
- if (Idx < (int)SrcNumElts)
- Res = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, getCurDebugLoc(),
- EltVT, Src1, DAG.getConstant(Idx, PtrVT));
- else
- Res = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, getCurDebugLoc(),
- EltVT, Src2,
- DAG.getConstant(Idx - SrcNumElts, PtrVT));
+ if (Idx < 0) {
+ Res = DAG.getUNDEF(EltVT);
+ } else {
+ SDValue &Src = Idx < (int)SrcNumElts ? Src1 : Src2;
+ if (Idx >= (int)SrcNumElts) Idx -= SrcNumElts;
- Ops.push_back(Res);
+ Res = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, getCurDebugLoc(),
+ EltVT, Src, DAG.getConstant(Idx, PtrVT));
}
+
+ Ops.push_back(Res);
}
setValue(&I, DAG.getNode(ISD::BUILD_VECTOR, getCurDebugLoc(),
void SelectionDAGBuilder::visitGetElementPtr(const User &I) {
SDValue N = getValue(I.getOperand(0));
- Type *Ty = I.getOperand(0)->getType();
+ // Note that the pointer operand may be a vector of pointers. Take the scalar
+ // element which holds a pointer.
+ Type *Ty = I.getOperand(0)->getType()->getScalarType();
for (GetElementPtrInst::const_op_iterator OI = I.op_begin()+1, E = I.op_end();
OI != E; ++OI) {
unsigned Amt = ElementSize.logBase2();
IdxN = DAG.getNode(ISD::SHL, getCurDebugLoc(),
N.getValueType(), IdxN,
- DAG.getConstant(Amt, TLI.getPointerTy()));
+ DAG.getConstant(Amt, IdxN.getValueType()));
} else {
SDValue Scale = DAG.getConstant(ElementSize, TLI.getPointerTy());
IdxN = DAG.getNode(ISD::MUL, getCurDebugLoc(),
}
void SelectionDAGBuilder::visitLoad(const LoadInst &I) {
+ if (I.isAtomic())
+ return visitAtomicLoad(I);
+
const Value *SV = I.getOperand(0);
SDValue Ptr = getValue(SV);
bool isVolatile = I.isVolatile();
bool isNonTemporal = I.getMetadata("nontemporal") != 0;
+ bool isInvariant = I.getMetadata("invariant.load") != 0;
unsigned Alignment = I.getAlignment();
const MDNode *TBAAInfo = I.getMetadata(LLVMContext::MD_tbaa);
+ const MDNode *Ranges = I.getMetadata(LLVMContext::MD_range);
SmallVector<EVT, 4> ValueVTs;
SmallVector<uint64_t, 4> Offsets;
DAG.getConstant(Offsets[i], PtrVT));
SDValue L = DAG.getLoad(ValueVTs[i], getCurDebugLoc(), Root,
A, MachinePointerInfo(SV, Offsets[i]), isVolatile,
- isNonTemporal, Alignment, TBAAInfo);
+ isNonTemporal, isInvariant, Alignment, TBAAInfo,
+ Ranges);
Values[i] = L;
Chains[ChainI] = L.getValue(1);
}
void SelectionDAGBuilder::visitStore(const StoreInst &I) {
+ if (I.isAtomic())
+ return visitAtomicStore(I);
+
const Value *SrcV = I.getOperand(0);
const Value *PtrV = I.getOperand(1);
}
static SDValue InsertFenceForAtomic(SDValue Chain, AtomicOrdering Order,
+ SynchronizationScope Scope,
bool Before, DebugLoc dl,
SelectionDAG &DAG,
const TargetLowering &TLI) {
// Fence, if necessary
if (Before) {
- if (Order == AcquireRelease)
+ if (Order == AcquireRelease || Order == SequentiallyConsistent)
Order = Release;
else if (Order == Acquire || Order == Monotonic)
return Chain;
}
SDValue Ops[3];
Ops[0] = Chain;
- Ops[1] = DAG.getConstant(SequentiallyConsistent, TLI.getPointerTy());
- Ops[2] = DAG.getConstant(Order, TLI.getPointerTy());
+ Ops[1] = DAG.getConstant(Order, TLI.getPointerTy());
+ Ops[2] = DAG.getConstant(Scope, TLI.getPointerTy());
return DAG.getNode(ISD::ATOMIC_FENCE, dl, MVT::Other, Ops, 3);
}
void SelectionDAGBuilder::visitAtomicCmpXchg(const AtomicCmpXchgInst &I) {
DebugLoc dl = getCurDebugLoc();
AtomicOrdering Order = I.getOrdering();
+ SynchronizationScope Scope = I.getSynchScope();
SDValue InChain = getRoot();
if (TLI.getInsertFencesForAtomic())
- InChain = InsertFenceForAtomic(InChain, Order, true, dl, DAG, TLI);
+ InChain = InsertFenceForAtomic(InChain, Order, Scope, true, dl,
+ DAG, TLI);
SDValue L =
DAG.getAtomic(ISD::ATOMIC_CMP_SWAP, dl,
getValue(I.getCompareOperand()),
getValue(I.getNewValOperand()),
MachinePointerInfo(I.getPointerOperand()), 0 /* Alignment */,
- I.getOrdering(), I.getSynchScope());
+ TLI.getInsertFencesForAtomic() ? Monotonic : Order,
+ Scope);
SDValue OutChain = L.getValue(1);
if (TLI.getInsertFencesForAtomic())
- OutChain = InsertFenceForAtomic(OutChain, Order, false, dl, DAG, TLI);
+ OutChain = InsertFenceForAtomic(OutChain, Order, Scope, false, dl,
+ DAG, TLI);
setValue(&I, L);
DAG.setRoot(OutChain);
DebugLoc dl = getCurDebugLoc();
ISD::NodeType NT;
switch (I.getOperation()) {
- default: llvm_unreachable("Unknown atomicrmw operation"); return;
+ default: llvm_unreachable("Unknown atomicrmw operation");
case AtomicRMWInst::Xchg: NT = ISD::ATOMIC_SWAP; break;
case AtomicRMWInst::Add: NT = ISD::ATOMIC_LOAD_ADD; break;
case AtomicRMWInst::Sub: NT = ISD::ATOMIC_LOAD_SUB; break;
case AtomicRMWInst::UMin: NT = ISD::ATOMIC_LOAD_UMIN; break;
}
AtomicOrdering Order = I.getOrdering();
+ SynchronizationScope Scope = I.getSynchScope();
SDValue InChain = getRoot();
if (TLI.getInsertFencesForAtomic())
- InChain = InsertFenceForAtomic(InChain, Order, true, dl, DAG, TLI);
+ InChain = InsertFenceForAtomic(InChain, Order, Scope, true, dl,
+ DAG, TLI);
SDValue L =
DAG.getAtomic(NT, dl,
getValue(I.getValOperand()),
I.getPointerOperand(), 0 /* Alignment */,
TLI.getInsertFencesForAtomic() ? Monotonic : Order,
- I.getSynchScope());
+ Scope);
SDValue OutChain = L.getValue(1);
if (TLI.getInsertFencesForAtomic())
- OutChain = InsertFenceForAtomic(OutChain, Order, false, dl, DAG, TLI);
+ OutChain = InsertFenceForAtomic(OutChain, Order, Scope, false, dl,
+ DAG, TLI);
setValue(&I, L);
DAG.setRoot(OutChain);
DAG.setRoot(DAG.getNode(ISD::ATOMIC_FENCE, dl, MVT::Other, Ops, 3));
}
+void SelectionDAGBuilder::visitAtomicLoad(const LoadInst &I) {
+ DebugLoc dl = getCurDebugLoc();
+ AtomicOrdering Order = I.getOrdering();
+ SynchronizationScope Scope = I.getSynchScope();
+
+ SDValue InChain = getRoot();
+
+ EVT VT = EVT::getEVT(I.getType());
+
+ if (I.getAlignment() * 8 < VT.getSizeInBits())
+ report_fatal_error("Cannot generate unaligned atomic load");
+
+ SDValue L =
+ DAG.getAtomic(ISD::ATOMIC_LOAD, dl, VT, VT, InChain,
+ getValue(I.getPointerOperand()),
+ I.getPointerOperand(), I.getAlignment(),
+ TLI.getInsertFencesForAtomic() ? Monotonic : Order,
+ Scope);
+
+ SDValue OutChain = L.getValue(1);
+
+ if (TLI.getInsertFencesForAtomic())
+ OutChain = InsertFenceForAtomic(OutChain, Order, Scope, false, dl,
+ DAG, TLI);
+
+ setValue(&I, L);
+ DAG.setRoot(OutChain);
+}
+
+void SelectionDAGBuilder::visitAtomicStore(const StoreInst &I) {
+ DebugLoc dl = getCurDebugLoc();
+
+ AtomicOrdering Order = I.getOrdering();
+ SynchronizationScope Scope = I.getSynchScope();
+
+ SDValue InChain = getRoot();
+
+ EVT VT = EVT::getEVT(I.getValueOperand()->getType());
+
+ if (I.getAlignment() * 8 < VT.getSizeInBits())
+ report_fatal_error("Cannot generate unaligned atomic store");
+
+ if (TLI.getInsertFencesForAtomic())
+ InChain = InsertFenceForAtomic(InChain, Order, Scope, true, dl,
+ DAG, TLI);
+
+ SDValue OutChain =
+ DAG.getAtomic(ISD::ATOMIC_STORE, dl, VT,
+ InChain,
+ getValue(I.getPointerOperand()),
+ getValue(I.getValueOperand()),
+ I.getPointerOperand(), I.getAlignment(),
+ TLI.getInsertFencesForAtomic() ? Monotonic : Order,
+ Scope);
+
+ if (TLI.getInsertFencesForAtomic())
+ OutChain = InsertFenceForAtomic(OutChain, Order, Scope, false, dl,
+ DAG, TLI);
+
+ DAG.setRoot(OutChain);
+}
+
/// visitTargetIntrinsic - Lower a call of a target intrinsic to an INTRINSIC
/// node.
void SelectionDAGBuilder::visitTargetIntrinsic(const CallInst &I,
// Add the intrinsic ID as an integer operand if it's not a target intrinsic.
if (!IsTgtIntrinsic || Info.opc == ISD::INTRINSIC_VOID ||
Info.opc == ISD::INTRINSIC_W_CHAIN)
- Ops.push_back(DAG.getConstant(Intrinsic, TLI.getPointerTy()));
+ Ops.push_back(DAG.getTargetConstant(Intrinsic, TLI.getPointerTy()));
// Add all operands of the call to the operand list.
for (unsigned i = 0, e = I.getNumArgOperands(); i != e; ++i) {
SDValue Op = getValue(I.getArgOperand(i));
- assert(TLI.isTypeLegal(Op.getValueType()) &&
- "Intrinsic uses a non-legal type?");
Ops.push_back(Op);
}
SmallVector<EVT, 4> ValueVTs;
ComputeValueVTs(TLI, I.getType(), ValueVTs);
-#ifndef NDEBUG
- for (unsigned Val = 0, E = ValueVTs.size(); Val != E; ++Val) {
- assert(TLI.isTypeLegal(ValueVTs[Val]) &&
- "Intrinsic uses a non-legal type?");
- }
-#endif // NDEBUG
if (HasChain)
ValueVTs.push_back(MVT::Other);
}
setValue(&I, Result);
+ } else {
+ // Assign order to result here. If the intrinsic does not produce a result,
+ // it won't be mapped to a SDNode and visit() will not assign it an order
+ // number.
+ ++SDNodeOrder;
+ AssignOrderingToNode(Result.getNode());
}
}
return DAG.getConstantFP(APFloat(APInt(32, Flt)), MVT::f32);
}
-/// Inlined utility function to implement binary input atomic intrinsics for
-/// visitIntrinsicCall: I is a call instruction
-/// Op is the associated NodeType for I
-const char *
-SelectionDAGBuilder::implVisitBinaryAtomic(const CallInst& I,
- ISD::NodeType Op) {
- SDValue Root = getRoot();
- SDValue L =
- DAG.getAtomic(Op, getCurDebugLoc(),
- getValue(I.getArgOperand(1)).getValueType().getSimpleVT(),
- Root,
- getValue(I.getArgOperand(0)),
- getValue(I.getArgOperand(1)),
- I.getArgOperand(0), 0 /* Alignment */,
- Monotonic, CrossThread);
- setValue(&I, L);
- DAG.setRoot(L.getValue(1));
- return 0;
-}
-
-// implVisitAluOverflow - Lower arithmetic overflow instrinsics.
-const char *
-SelectionDAGBuilder::implVisitAluOverflow(const CallInst &I, ISD::NodeType Op) {
- SDValue Op1 = getValue(I.getArgOperand(0));
- SDValue Op2 = getValue(I.getArgOperand(1));
-
- SDVTList VTs = DAG.getVTList(Op1.getValueType(), MVT::i1);
- setValue(&I, DAG.getNode(Op, getCurDebugLoc(), VTs, Op1, Op2));
- return 0;
-}
-
/// visitExp - Lower an exp intrinsic. Handles the special sequences for
/// limited-precision mode.
void
const SDValue &CFR = Ext.getOperand(0);
if (CFR.getOpcode() == ISD::CopyFromReg)
return cast<RegisterSDNode>(CFR.getOperand(1))->getReg();
- else
- if (CFR.getOpcode() == ISD::TRUNCATE)
- return getTruncatedArgReg(CFR);
+ if (CFR.getOpcode() == ISD::TRUNCATE)
+ return getTruncatedArgReg(CFR);
}
return 0;
}
return false;
unsigned Reg = 0;
- if (Arg->hasByValAttr()) {
- // Byval arguments' frame index is recorded during argument lowering.
- // Use this info directly.
+ // Some arguments' frame index is recorded during argument lowering.
+ Offset = FuncInfo.getArgumentFrameIndex(Arg);
+ if (Offset)
Reg = TRI->getFrameRegister(MF);
- Offset = FuncInfo.getByValArgumentFrameIndex(Arg);
- // If byval argument ofset is not recorded then ignore this.
- if (!Offset)
- Reg = 0;
- }
- if (N.getNode()) {
+ if (!Reg && N.getNode()) {
if (N.getOpcode() == ISD::CopyFromReg)
Reg = cast<RegisterSDNode>(N.getOperand(1))->getReg();
else
getValue(I.getArgOperand(0))));
return 0;
case Intrinsic::setjmp:
- return "_setjmp"+!TLI.usesUnderscoreSetJmp();
+ return &"_setjmp"[!TLI.usesUnderscoreSetJmp()];
case Intrinsic::longjmp:
- return "_longjmp"+!TLI.usesUnderscoreLongJmp();
+ return &"_longjmp"[!TLI.usesUnderscoreLongJmp()];
case Intrinsic::memcpy: {
// Assert for address < 256 since we support only user defined address
// spaces.
const DbgDeclareInst &DI = cast<DbgDeclareInst>(I);
MDNode *Variable = DI.getVariable();
const Value *Address = DI.getAddress();
- if (!Address || !DIVariable(DI.getVariable()).Verify())
+ if (!Address || !DIVariable(Variable).Verify()) {
+ DEBUG(dbgs() << "Dropping debug info for " << DI << "\n");
return 0;
+ }
// Build an entry in DbgOrdering. Debug info input nodes get an SDNodeOrder
// but do not always have a corresponding SDNode built. The SDNodeOrder
// Check if address has undef value.
if (isa<UndefValue>(Address) ||
(Address->use_empty() && !isa<Argument>(Address))) {
- DEBUG(dbgs() << "Dropping debug info for " << DI);
+ DEBUG(dbgs() << "Dropping debug info for " << DI << "\n");
return 0;
}
N = UnusedArgNodeMap[Address];
SDDbgValue *SDV;
if (N.getNode()) {
- // Parameters are handled specially.
- bool isParameter =
- DIVariable(Variable).getTag() == dwarf::DW_TAG_arg_variable;
if (const BitCastInst *BCI = dyn_cast<BitCastInst>(Address))
Address = BCI->getOperand(0);
+ // Parameters are handled specially.
+ bool isParameter =
+ (DIVariable(Variable).getTag() == dwarf::DW_TAG_arg_variable ||
+ isa<Argument>(Address));
+
const AllocaInst *AI = dyn_cast<AllocaInst>(Address);
if (isParameter && !AI) {
0, dl, SDNodeOrder);
else {
// Can't do anything with other non-AI cases yet.
- DEBUG(dbgs() << "Dropping debug info for " << DI);
+ DEBUG(dbgs() << "Dropping debug info for " << DI << "\n");
+ DEBUG(dbgs() << "non-AllocaInst issue for Address: \n\t");
+ DEBUG(Address->dump());
return 0;
}
DAG.AddDbgValue(SDV, N.getNode(), isParameter);
}
}
}
- DEBUG(dbgs() << "Dropping debug info for " << DI);
+ DEBUG(dbgs() << "Dropping debug info for " << DI << "\n");
}
}
return 0;
} else {
// We may expand this to cover more cases. One case where we have no
// data available is an unreferenced parameter.
- DEBUG(dbgs() << "Dropping debug info for " << DI);
+ DEBUG(dbgs() << "Dropping debug info for " << DI << "\n");
}
}
V = BCI->getOperand(0);
const AllocaInst *AI = dyn_cast<AllocaInst>(V);
// Don't handle byval struct arguments or VLAs, for example.
- if (!AI)
+ if (!AI) {
+ DEBUG(dbgs() << "Dropping debug location info for:\n " << DI << "\n");
+ DEBUG(dbgs() << " Last seen at:\n " << *V << "\n");
return 0;
+ }
DenseMap<const AllocaInst*, int>::iterator SI =
FuncInfo.StaticAllocaMap.find(AI);
if (SI == FuncInfo.StaticAllocaMap.end())
MMI.setVariableDbgInfo(Variable, FI, DI.getDebugLoc());
return 0;
}
- case Intrinsic::eh_exception: {
- // Insert the EXCEPTIONADDR instruction.
- assert(FuncInfo.MBB->isLandingPad() &&
- "Call to eh.exception not in landing pad!");
- SDVTList VTs = DAG.getVTList(TLI.getPointerTy(), MVT::Other);
- SDValue Ops[1];
- Ops[0] = DAG.getRoot();
- SDValue Op = DAG.getNode(ISD::EXCEPTIONADDR, dl, VTs, Ops, 1);
- setValue(&I, Op);
- DAG.setRoot(Op.getValue(1));
- return 0;
- }
-
- case Intrinsic::eh_selector: {
- MachineBasicBlock *CallMBB = FuncInfo.MBB;
- MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI();
- if (CallMBB->isLandingPad())
- AddCatchInfo(I, &MMI, CallMBB);
- else {
-#ifndef NDEBUG
- FuncInfo.CatchInfoLost.insert(&I);
-#endif
- // FIXME: Mark exception selector register as live in. Hack for PR1508.
- unsigned Reg = TLI.getExceptionSelectorRegister();
- if (Reg) FuncInfo.MBB->addLiveIn(Reg);
- }
-
- // Insert the EHSELECTION instruction.
- SDVTList VTs = DAG.getVTList(TLI.getPointerTy(), MVT::Other);
- SDValue Ops[2];
- Ops[0] = getValue(I.getArgOperand(0));
- Ops[1] = getRoot();
- SDValue Op = DAG.getNode(ISD::EHSELECTION, dl, VTs, Ops, 2);
- DAG.setRoot(Op.getValue(1));
- setValue(&I, DAG.getSExtOrTrunc(Op, dl, MVT::i32));
- return 0;
- }
case Intrinsic::eh_typeid_for: {
// Find the type id for the given typeinfo.
MMI.setCurrentCallSite(CI->getZExtValue());
return 0;
}
+ case Intrinsic::eh_sjlj_functioncontext: {
+ // Get and store the index of the function context.
+ MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
+ AllocaInst *FnCtx =
+ cast<AllocaInst>(I.getArgOperand(0)->stripPointerCasts());
+ int FI = FuncInfo.StaticAllocaMap[FnCtx];
+ MFI->setFunctionContextIndex(FI);
+ return 0;
+ }
case Intrinsic::eh_sjlj_setjmp: {
- setValue(&I, DAG.getNode(ISD::EH_SJLJ_SETJMP, dl, MVT::i32, getRoot(),
- getValue(I.getArgOperand(0))));
+ SDValue Ops[2];
+ Ops[0] = getRoot();
+ Ops[1] = getValue(I.getArgOperand(0));
+ SDValue Op = DAG.getNode(ISD::EH_SJLJ_SETJMP, dl,
+ DAG.getVTList(MVT::i32, MVT::Other),
+ Ops, 2);
+ setValue(&I, Op.getValue(0));
+ DAG.setRoot(Op.getValue(1));
return 0;
}
case Intrinsic::eh_sjlj_longjmp: {
getRoot(), getValue(I.getArgOperand(0))));
return 0;
}
- case Intrinsic::eh_sjlj_dispatch_setup: {
- DAG.setRoot(DAG.getNode(ISD::EH_SJLJ_DISPATCHSETUP, dl, MVT::Other,
- getRoot(), getValue(I.getArgOperand(0))));
- return 0;
- }
case Intrinsic::x86_mmx_pslli_w:
case Intrinsic::x86_mmx_pslli_d:
setValue(&I, Res);
return 0;
}
+ case Intrinsic::x86_avx_vinsertf128_pd_256:
+ case Intrinsic::x86_avx_vinsertf128_ps_256:
+ case Intrinsic::x86_avx_vinsertf128_si_256:
+ case Intrinsic::x86_avx2_vinserti128: {
+ DebugLoc dl = getCurDebugLoc();
+ EVT DestVT = TLI.getValueType(I.getType());
+ EVT ElVT = TLI.getValueType(I.getArgOperand(1)->getType());
+ uint64_t Idx = (cast<ConstantInt>(I.getArgOperand(2))->getZExtValue() & 1) *
+ ElVT.getVectorNumElements();
+ Res = DAG.getNode(ISD::INSERT_SUBVECTOR, dl, DestVT,
+ getValue(I.getArgOperand(0)),
+ getValue(I.getArgOperand(1)),
+ DAG.getConstant(Idx, MVT::i32));
+ setValue(&I, Res);
+ return 0;
+ }
case Intrinsic::convertff:
case Intrinsic::convertfsi:
case Intrinsic::convertfui:
case Intrinsic::convertuu: {
ISD::CvtCode Code = ISD::CVT_INVALID;
switch (Intrinsic) {
+ default: llvm_unreachable("Impossible intrinsic"); // Can't reach here.
case Intrinsic::convertff: Code = ISD::CVT_FF; break;
case Intrinsic::convertfsi: Code = ISD::CVT_FS; break;
case Intrinsic::convertfui: Code = ISD::CVT_FU; break;
case Intrinsic::pow:
visitPow(I);
return 0;
+ case Intrinsic::fabs:
+ setValue(&I, DAG.getNode(ISD::FABS, dl,
+ getValue(I.getArgOperand(0)).getValueType(),
+ getValue(I.getArgOperand(0))));
+ return 0;
case Intrinsic::fma:
setValue(&I, DAG.getNode(ISD::FMA, dl,
getValue(I.getArgOperand(0)).getValueType(),
return 0;
case Intrinsic::cttz: {
SDValue Arg = getValue(I.getArgOperand(0));
+ ConstantInt *CI = cast<ConstantInt>(I.getArgOperand(1));
EVT Ty = Arg.getValueType();
- setValue(&I, DAG.getNode(ISD::CTTZ, dl, Ty, Arg));
+ setValue(&I, DAG.getNode(CI->isZero() ? ISD::CTTZ : ISD::CTTZ_ZERO_UNDEF,
+ dl, Ty, Arg));
return 0;
}
case Intrinsic::ctlz: {
SDValue Arg = getValue(I.getArgOperand(0));
+ ConstantInt *CI = cast<ConstantInt>(I.getArgOperand(1));
EVT Ty = Arg.getValueType();
- setValue(&I, DAG.getNode(ISD::CTLZ, dl, Ty, Arg));
+ setValue(&I, DAG.getNode(CI->isZero() ? ISD::CTLZ : ISD::CTLZ_ZERO_UNDEF,
+ dl, Ty, Arg));
return 0;
}
case Intrinsic::ctpop: {
Ops[4] = DAG.getSrcValue(I.getArgOperand(0));
Ops[5] = DAG.getSrcValue(F);
- Res = DAG.getNode(ISD::TRAMPOLINE, dl,
- DAG.getVTList(TLI.getPointerTy(), MVT::Other),
- Ops, 6);
+ Res = DAG.getNode(ISD::INIT_TRAMPOLINE, dl, MVT::Other, Ops, 6);
- setValue(&I, Res);
- DAG.setRoot(Res.getValue(1));
+ DAG.setRoot(Res);
+ return 0;
+ }
+ case Intrinsic::adjust_trampoline: {
+ setValue(&I, DAG.getNode(ISD::ADJUST_TRAMPOLINE, dl,
+ TLI.getPointerTy(),
+ getValue(I.getArgOperand(0))));
return 0;
}
case Intrinsic::gcroot:
if (GFI) {
- const Value *Alloca = I.getArgOperand(0);
+ const Value *Alloca = I.getArgOperand(0)->stripPointerCasts();
const Constant *TypeMap = cast<Constant>(I.getArgOperand(1));
FrameIndexSDNode *FI = cast<FrameIndexSDNode>(getValue(Alloca).getNode());
case Intrinsic::gcread:
case Intrinsic::gcwrite:
llvm_unreachable("GC failed to lower gcread/gcwrite intrinsics!");
- return 0;
case Intrinsic::flt_rounds:
setValue(&I, DAG.getNode(ISD::FLT_ROUNDS_, dl, MVT::i32));
return 0;
}
case Intrinsic::trap: {
- StringRef TrapFuncName = getTrapFunctionName();
+ StringRef TrapFuncName = TM.Options.getTrapFunctionName();
if (TrapFuncName.empty()) {
DAG.setRoot(DAG.getNode(ISD::TRAP, dl,MVT::Other, getRoot()));
return 0;
}
TargetLowering::ArgListTy Args;
- std::pair<SDValue, SDValue> Result =
- TLI.LowerCallTo(getRoot(), I.getType(),
+ TargetLowering::
+ CallLoweringInfo CLI(getRoot(), I.getType(),
false, false, false, false, 0, CallingConv::C,
- /*isTailCall=*/false, /*isReturnValueUsed=*/true,
+ /*isTailCall=*/false,
+ /*doesNotRet=*/false, /*isReturnValueUsed=*/true,
DAG.getExternalSymbol(TrapFuncName.data(), TLI.getPointerTy()),
Args, DAG, getCurDebugLoc());
+ std::pair<SDValue, SDValue> Result = TLI.LowerCallTo(CLI);
DAG.setRoot(Result.second);
return 0;
}
+ case Intrinsic::debugtrap: {
+ DAG.setRoot(DAG.getNode(ISD::DEBUGTRAP, dl,MVT::Other, getRoot()));
+ return 0;
+ }
case Intrinsic::uadd_with_overflow:
- return implVisitAluOverflow(I, ISD::UADDO);
case Intrinsic::sadd_with_overflow:
- return implVisitAluOverflow(I, ISD::SADDO);
case Intrinsic::usub_with_overflow:
- return implVisitAluOverflow(I, ISD::USUBO);
case Intrinsic::ssub_with_overflow:
- return implVisitAluOverflow(I, ISD::SSUBO);
case Intrinsic::umul_with_overflow:
- return implVisitAluOverflow(I, ISD::UMULO);
- case Intrinsic::smul_with_overflow:
- return implVisitAluOverflow(I, ISD::SMULO);
+ case Intrinsic::smul_with_overflow: {
+ ISD::NodeType Op;
+ switch (Intrinsic) {
+ default: llvm_unreachable("Impossible intrinsic"); // Can't reach here.
+ case Intrinsic::uadd_with_overflow: Op = ISD::UADDO; break;
+ case Intrinsic::sadd_with_overflow: Op = ISD::SADDO; break;
+ case Intrinsic::usub_with_overflow: Op = ISD::USUBO; break;
+ case Intrinsic::ssub_with_overflow: Op = ISD::SSUBO; break;
+ case Intrinsic::umul_with_overflow: Op = ISD::UMULO; break;
+ case Intrinsic::smul_with_overflow: Op = ISD::SMULO; break;
+ }
+ SDValue Op1 = getValue(I.getArgOperand(0));
+ SDValue Op2 = getValue(I.getArgOperand(1));
+ SDVTList VTs = DAG.getVTList(Op1.getValueType(), MVT::i1);
+ setValue(&I, DAG.getNode(Op, getCurDebugLoc(), VTs, Op1, Op2));
+ return 0;
+ }
case Intrinsic::prefetch: {
SDValue Ops[5];
unsigned rw = cast<ConstantInt>(I.getArgOperand(1))->getZExtValue();
rw==1)); /* write */
return 0;
}
- case Intrinsic::memory_barrier: {
- SDValue Ops[6];
- Ops[0] = getRoot();
- for (int x = 1; x < 6; ++x)
- Ops[x] = getValue(I.getArgOperand(x - 1));
-
- DAG.setRoot(DAG.getNode(ISD::MEMBARRIER, dl, MVT::Other, &Ops[0], 6));
- return 0;
- }
- case Intrinsic::atomic_cmp_swap: {
- SDValue Root = getRoot();
- SDValue L =
- DAG.getAtomic(ISD::ATOMIC_CMP_SWAP, getCurDebugLoc(),
- getValue(I.getArgOperand(1)).getValueType().getSimpleVT(),
- Root,
- getValue(I.getArgOperand(0)),
- getValue(I.getArgOperand(1)),
- getValue(I.getArgOperand(2)),
- MachinePointerInfo(I.getArgOperand(0)), 0 /* Alignment */,
- Monotonic, CrossThread);
- setValue(&I, L);
- DAG.setRoot(L.getValue(1));
- return 0;
- }
- case Intrinsic::atomic_load_add:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_ADD);
- case Intrinsic::atomic_load_sub:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_SUB);
- case Intrinsic::atomic_load_or:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_OR);
- case Intrinsic::atomic_load_xor:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_XOR);
- case Intrinsic::atomic_load_and:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_AND);
- case Intrinsic::atomic_load_nand:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_NAND);
- case Intrinsic::atomic_load_max:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_MAX);
- case Intrinsic::atomic_load_min:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_MIN);
- case Intrinsic::atomic_load_umin:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_UMIN);
- case Intrinsic::atomic_load_umax:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_LOAD_UMAX);
- case Intrinsic::atomic_swap:
- return implVisitBinaryAtomic(I, ISD::ATOMIC_SWAP);
case Intrinsic::invariant_start:
case Intrinsic::lifetime_start:
// Check whether the function can return without sret-demotion.
SmallVector<ISD::OutputArg, 4> Outs;
- SmallVector<uint64_t, 4> Offsets;
GetReturnInfo(RetTy, CS.getAttributes().getRetAttributes(),
- Outs, TLI, &Offsets);
+ Outs, TLI);
bool CanLowerReturn = TLI.CanLowerReturn(CS.getCallingConv(),
DAG.getMachineFunction(),
unsigned CallSiteIndex = MMI.getCurrentCallSite();
if (CallSiteIndex) {
MMI.setCallSiteBeginLabel(BeginLabel, CallSiteIndex);
+ LPadToCallSiteMap[LandingPad].push_back(CallSiteIndex);
+
// Now that the call site is handled, stop tracking it.
MMI.setCurrentCallSite(0);
}
// If there's a possibility that fast-isel has already selected some amount
// of the current basic block, don't emit a tail call.
- if (isTailCall && EnableFastISel)
+ if (isTailCall && TM.Options.EnableFastISel)
isTailCall = false;
- std::pair<SDValue,SDValue> Result =
- TLI.LowerCallTo(getRoot(), RetTy,
- CS.paramHasAttr(0, Attribute::SExt),
- CS.paramHasAttr(0, Attribute::ZExt), FTy->isVarArg(),
- CS.paramHasAttr(0, Attribute::InReg), FTy->getNumParams(),
- CS.getCallingConv(),
- isTailCall,
- !CS.getInstruction()->use_empty(),
- Callee, Args, DAG, getCurDebugLoc());
+ TargetLowering::
+ CallLoweringInfo CLI(getRoot(), RetTy, FTy, isTailCall, Callee, Args, DAG,
+ getCurDebugLoc(), CS);
+ std::pair<SDValue,SDValue> Result = TLI.LowerCallTo(CLI);
assert((isTailCall || Result.second.getNode()) &&
"Non-null chain expected with non-tail call!");
assert((Result.second.getNode() || !Result.first.getNode()) &&
ComputeValueVTs(TLI, PtrRetTy, PVTs);
assert(PVTs.size() == 1 && "Pointers should fit in one register");
EVT PtrVT = PVTs[0];
- unsigned NumValues = Outs.size();
+
+ SmallVector<EVT, 4> RetTys;
+ SmallVector<uint64_t, 4> Offsets;
+ RetTy = FTy->getReturnType();
+ ComputeValueVTs(TLI, RetTy, RetTys, &Offsets);
+
+ unsigned NumValues = RetTys.size();
SmallVector<SDValue, 4> Values(NumValues);
SmallVector<SDValue, 4> Chains(NumValues);
SDValue Add = DAG.getNode(ISD::ADD, getCurDebugLoc(), PtrVT,
DemoteStackSlot,
DAG.getConstant(Offsets[i], PtrVT));
- SDValue L = DAG.getLoad(Outs[i].VT, getCurDebugLoc(), Result.second,
- Add,
+ SDValue L = DAG.getLoad(RetTys[i], getCurDebugLoc(), Result.second, Add,
MachinePointerInfo::getFixedStack(DemoteStackIdx, Offsets[i]),
- false, false, 1);
+ false, false, false, 1);
Values[i] = L;
Chains[i] = L.getValue(1);
}
MVT::Other, &Chains[0], NumValues);
PendingLoads.push_back(Chain);
- // Collect the legal value parts into potentially illegal values
- // that correspond to the original function's return values.
- SmallVector<EVT, 4> RetTys;
- RetTy = FTy->getReturnType();
- ComputeValueVTs(TLI, RetTy, RetTys);
- ISD::NodeType AssertOp = ISD::DELETED_NODE;
- SmallVector<SDValue, 4> ReturnValues;
- unsigned CurReg = 0;
- for (unsigned I = 0, E = RetTys.size(); I != E; ++I) {
- EVT VT = RetTys[I];
- EVT RegisterVT = TLI.getRegisterType(RetTy->getContext(), VT);
- unsigned NumRegs = TLI.getNumRegisters(RetTy->getContext(), VT);
-
- SDValue ReturnValue =
- getCopyFromParts(DAG, getCurDebugLoc(), &Values[CurReg], NumRegs,
- RegisterVT, VT, AssertOp);
- ReturnValues.push_back(ReturnValue);
- CurReg += NumRegs;
- }
-
setValue(CS.getInstruction(),
DAG.getNode(ISD::MERGE_VALUES, getCurDebugLoc(),
DAG.getVTList(&RetTys[0], RetTys.size()),
- &ReturnValues[0], ReturnValues.size()));
+ &Values[0], Values.size()));
}
// Assign order to nodes here. If the call does not produce a result, it won't
SDValue LoadVal = Builder.DAG.getLoad(LoadVT, Builder.getCurDebugLoc(), Root,
Ptr, MachinePointerInfo(PtrVal),
false /*volatile*/,
- false /*nontemporal*/, 1 /* align=1 */);
+ false /*nontemporal*/,
+ false /*isinvariant*/, 1 /* align=1 */);
if (!ConstantMemory)
Builder.PendingLoads.push_back(LoadVal.getValue(1));
return;
}
- // See if any floating point values are being passed to this function. This is
- // used to emit an undefined reference to fltused on Windows.
- FunctionType *FT =
- cast<FunctionType>(I.getCalledValue()->getType()->getContainedType(0));
MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI();
- if (FT->isVarArg() &&
- !MMI.callsExternalVAFunctionWithFloatingPointArguments()) {
- for (unsigned i = 0, e = I.getNumArgOperands(); i != e; ++i) {
- Type* T = I.getArgOperand(i)->getType();
- for (po_iterator<Type*> i = po_begin(T), e = po_end(T);
- i != e; ++i) {
- if (!i->isFloatingPointTy()) continue;
- MMI.setCallsExternalVAFunctionWithFloatingPointArguments(true);
- break;
- }
- }
- }
+ ComputeUsesVAFloatArgument(I, &MMI);
const char *RenameFn = 0;
if (Function *F = I.getCalledFunction()) {
// can't be a library call.
if (!F->hasLocalLinkage() && F->hasName()) {
StringRef Name = F->getName();
- if (Name == "copysign" || Name == "copysignf" || Name == "copysignl") {
+ if ((LibInfo->has(LibFunc::copysign) && Name == "copysign") ||
+ (LibInfo->has(LibFunc::copysignf) && Name == "copysignf") ||
+ (LibInfo->has(LibFunc::copysignl) && Name == "copysignl")) {
if (I.getNumArgOperands() == 2 && // Basic sanity checks.
I.getArgOperand(0)->getType()->isFloatingPointTy() &&
I.getType() == I.getArgOperand(0)->getType() &&
LHS.getValueType(), LHS, RHS));
return;
}
- } else if (Name == "fabs" || Name == "fabsf" || Name == "fabsl") {
+ } else if ((LibInfo->has(LibFunc::fabs) && Name == "fabs") ||
+ (LibInfo->has(LibFunc::fabsf) && Name == "fabsf") ||
+ (LibInfo->has(LibFunc::fabsl) && Name == "fabsl")) {
if (I.getNumArgOperands() == 1 && // Basic sanity checks.
I.getArgOperand(0)->getType()->isFloatingPointTy() &&
I.getType() == I.getArgOperand(0)->getType()) {
Tmp.getValueType(), Tmp));
return;
}
- } else if (Name == "sin" || Name == "sinf" || Name == "sinl") {
+ } else if ((LibInfo->has(LibFunc::sin) && Name == "sin") ||
+ (LibInfo->has(LibFunc::sinf) && Name == "sinf") ||
+ (LibInfo->has(LibFunc::sinl) && Name == "sinl")) {
if (I.getNumArgOperands() == 1 && // Basic sanity checks.
I.getArgOperand(0)->getType()->isFloatingPointTy() &&
I.getType() == I.getArgOperand(0)->getType() &&
Tmp.getValueType(), Tmp));
return;
}
- } else if (Name == "cos" || Name == "cosf" || Name == "cosl") {
+ } else if ((LibInfo->has(LibFunc::cos) && Name == "cos") ||
+ (LibInfo->has(LibFunc::cosf) && Name == "cosf") ||
+ (LibInfo->has(LibFunc::cosl) && Name == "cosl")) {
if (I.getNumArgOperands() == 1 && // Basic sanity checks.
I.getArgOperand(0)->getType()->isFloatingPointTy() &&
I.getType() == I.getArgOperand(0)->getType() &&
Tmp.getValueType(), Tmp));
return;
}
- } else if (Name == "sqrt" || Name == "sqrtf" || Name == "sqrtl") {
+ } else if ((LibInfo->has(LibFunc::sqrt) && Name == "sqrt") ||
+ (LibInfo->has(LibFunc::sqrtf) && Name == "sqrtf") ||
+ (LibInfo->has(LibFunc::sqrtl) && Name == "sqrtl")) {
if (I.getNumArgOperands() == 1 && // Basic sanity checks.
I.getArgOperand(0)->getType()->isFloatingPointTy() &&
I.getType() == I.getArgOperand(0)->getType() &&
Tmp.getValueType(), Tmp));
return;
}
+ } else if ((LibInfo->has(LibFunc::floor) && Name == "floor") ||
+ (LibInfo->has(LibFunc::floorf) && Name == "floorf") ||
+ (LibInfo->has(LibFunc::floorl) && Name == "floorl")) {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FFLOOR, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
+ } else if ((LibInfo->has(LibFunc::nearbyint) && Name == "nearbyint") ||
+ (LibInfo->has(LibFunc::nearbyintf) && Name == "nearbyintf") ||
+ (LibInfo->has(LibFunc::nearbyintl) && Name == "nearbyintl")) {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FNEARBYINT, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
+ } else if ((LibInfo->has(LibFunc::ceil) && Name == "ceil") ||
+ (LibInfo->has(LibFunc::ceilf) && Name == "ceilf") ||
+ (LibInfo->has(LibFunc::ceill) && Name == "ceill")) {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FCEIL, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
+ } else if ((LibInfo->has(LibFunc::rint) && Name == "rint") ||
+ (LibInfo->has(LibFunc::rintf) && Name == "rintf") ||
+ (LibInfo->has(LibFunc::rintl) && Name == "rintl")) {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FRINT, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
+ } else if ((LibInfo->has(LibFunc::trunc) && Name == "trunc") ||
+ (LibInfo->has(LibFunc::truncf) && Name == "truncf") ||
+ (LibInfo->has(LibFunc::truncl) && Name == "truncl")) {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FTRUNC, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
+ } else if ((LibInfo->has(LibFunc::log2) && Name == "log2") ||
+ (LibInfo->has(LibFunc::log2f) && Name == "log2f") ||
+ (LibInfo->has(LibFunc::log2l) && Name == "log2l")) {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType() &&
+ I.onlyReadsMemory()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FLOG2, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
+ } else if ((LibInfo->has(LibFunc::exp2) && Name == "exp2") ||
+ (LibInfo->has(LibFunc::exp2f) && Name == "exp2f") ||
+ (LibInfo->has(LibFunc::exp2l) && Name == "exp2l")) {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType() &&
+ I.onlyReadsMemory()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FEXP2, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
} else if (Name == "memcmp") {
if (visitMemCmpCall(I))
return;
: TargetLowering::AsmOperandInfo(info), CallOperand(0,0) {
}
- /// MarkAllocatedRegs - Once AssignedRegs is set, mark the assigned registers
- /// busy in OutputRegs/InputRegs.
- void MarkAllocatedRegs(bool isOutReg, bool isInReg,
- std::set<unsigned> &OutputRegs,
- std::set<unsigned> &InputRegs,
- const TargetRegisterInfo &TRI) const {
- if (isOutReg) {
- for (unsigned i = 0, e = AssignedRegs.Regs.size(); i != e; ++i)
- MarkRegAndAliases(AssignedRegs.Regs[i], OutputRegs, TRI);
- }
- if (isInReg) {
- for (unsigned i = 0, e = AssignedRegs.Regs.size(); i != e; ++i)
- MarkRegAndAliases(AssignedRegs.Regs[i], InputRegs, TRI);
- }
- }
-
/// getCallOperandValEVT - Return the EVT of the Value* that this operand
/// corresponds to. If there is no Value* for this operand, it returns
/// MVT::Other.
return TLI.getValueType(OpTy, true);
}
-
-private:
- /// MarkRegAndAliases - Mark the specified register and all aliases in the
- /// specified set.
- static void MarkRegAndAliases(unsigned Reg, std::set<unsigned> &Regs,
- const TargetRegisterInfo &TRI) {
- assert(TargetRegisterInfo::isPhysicalRegister(Reg) && "Isn't a physreg");
- Regs.insert(Reg);
- if (const unsigned *Aliases = TRI.getAliasSet(Reg))
- for (; *Aliases; ++Aliases)
- Regs.insert(*Aliases);
- }
};
typedef SmallVector<SDISelAsmOperandInfo,16> SDISelAsmOperandInfoVector;
/// allocation. This produces generally horrible, but correct, code.
///
/// OpInfo describes the operand.
-/// Input and OutputRegs are the set of already allocated physical registers.
///
static void GetRegistersForValue(SelectionDAG &DAG,
const TargetLowering &TLI,
DebugLoc DL,
- SDISelAsmOperandInfo &OpInfo,
- std::set<unsigned> &OutputRegs,
- std::set<unsigned> &InputRegs) {
+ SDISelAsmOperandInfo &OpInfo) {
LLVMContext &Context = *DAG.getContext();
- // Compute whether this value requires an input register, an output register,
- // or both.
- bool isOutReg = false;
- bool isInReg = false;
- switch (OpInfo.Type) {
- case InlineAsm::isOutput:
- isOutReg = true;
-
- // If there is an input constraint that matches this, we need to reserve
- // the input register so no other inputs allocate to it.
- isInReg = OpInfo.hasMatchingInput();
- break;
- case InlineAsm::isInput:
- isInReg = true;
- isOutReg = false;
- break;
- case InlineAsm::isClobber:
- isOutReg = true;
- isInReg = true;
- break;
- }
-
-
MachineFunction &MF = DAG.getMachineFunction();
SmallVector<unsigned, 4> Regs;
}
OpInfo.AssignedRegs = RegsForValue(Regs, RegVT, ValueVT);
- const TargetRegisterInfo *TRI = DAG.getTarget().getRegisterInfo();
- OpInfo.MarkAllocatedRegs(isOutReg, isInReg, OutputRegs, InputRegs, *TRI);
return;
}
/// ConstraintOperands - Information about all of the constraints.
SDISelAsmOperandInfoVector ConstraintOperands;
- std::set<unsigned> OutputRegs, InputRegs;
-
TargetLowering::AsmOperandInfoVector
TargetConstraints = TLI.ParseConstraints(CS);
// The return value of the call is this value. As such, there is no
// corresponding argument.
- assert(!CS.getType()->isVoidTy() &&
- "Bad inline asm!");
+ assert(!CS.getType()->isVoidTy() && "Bad inline asm!");
if (StructType *STy = dyn_cast<StructType>(CS.getType())) {
OpVT = TLI.getValueType(STy->getElementType(ResNo));
} else {
if (OpInfo.ConstraintVT != Input.ConstraintVT) {
std::pair<unsigned, const TargetRegisterClass*> MatchRC =
- TLI.getRegForInlineAsmConstraint(OpInfo.ConstraintCode, OpInfo.ConstraintVT);
+ TLI.getRegForInlineAsmConstraint(OpInfo.ConstraintCode,
+ OpInfo.ConstraintVT);
std::pair<unsigned, const TargetRegisterClass*> InputRC =
- TLI.getRegForInlineAsmConstraint(Input.ConstraintCode, Input.ConstraintVT);
+ TLI.getRegForInlineAsmConstraint(Input.ConstraintCode,
+ Input.ConstraintVT);
if ((OpInfo.ConstraintVT.isInteger() !=
Input.ConstraintVT.isInteger()) ||
(MatchRC.second != InputRC.second)) {
// constant pool entry to get its address.
const Value *OpVal = OpInfo.CallOperandVal;
if (isa<ConstantFP>(OpVal) || isa<ConstantInt>(OpVal) ||
- isa<ConstantVector>(OpVal)) {
+ isa<ConstantVector>(OpVal) || isa<ConstantDataVector>(OpVal)) {
OpInfo.CallOperand = DAG.getConstantPool(cast<Constant>(OpVal),
TLI.getPointerTy());
} else {
// If this constraint is for a specific register, allocate it before
// anything else.
if (OpInfo.ConstraintType == TargetLowering::C_Register)
- GetRegistersForValue(DAG, TLI, getCurDebugLoc(), OpInfo, OutputRegs,
- InputRegs);
+ GetRegistersForValue(DAG, TLI, getCurDebugLoc(), OpInfo);
}
// Second pass - Loop over all of the operands, assigning virtual or physregs
// C_Register operands have already been allocated, Other/Memory don't need
// to be.
if (OpInfo.ConstraintType == TargetLowering::C_RegisterClass)
- GetRegistersForValue(DAG, TLI, getCurDebugLoc(), OpInfo, OutputRegs,
- InputRegs);
+ GetRegistersForValue(DAG, TLI, getCurDebugLoc(), OpInfo);
}
// AsmNodeOperands - The operands for the ISD::INLINEASM node.
// Copy the output from the appropriate register. Find a register that
// we can use.
- if (OpInfo.AssignedRegs.Regs.empty())
- report_fatal_error("Couldn't allocate output reg for constraint '" +
- Twine(OpInfo.ConstraintCode) + "'!");
+ if (OpInfo.AssignedRegs.Regs.empty()) {
+ LLVMContext &Ctx = *DAG.getContext();
+ Ctx.emitError(CS.getInstruction(),
+ "couldn't allocate output register for constraint '" +
+ Twine(OpInfo.ConstraintCode) + "'");
+ break;
+ }
// If this is an indirect operand, store through the pointer after the
// asm.
std::vector<SDValue> Ops;
TLI.LowerAsmOperandForConstraint(InOperandVal, OpInfo.ConstraintCode,
Ops, DAG);
- if (Ops.empty())
- report_fatal_error("Invalid operand for inline asm constraint '" +
- Twine(OpInfo.ConstraintCode) + "'!");
+ if (Ops.empty()) {
+ LLVMContext &Ctx = *DAG.getContext();
+ Ctx.emitError(CS.getInstruction(),
+ "invalid operand for inline asm constraint '" +
+ Twine(OpInfo.ConstraintCode) + "'");
+ break;
+ }
// Add information to the INLINEASM node to know about this input.
unsigned ResOpType =
"Don't know how to handle indirect register inputs yet!");
// Copy the input into the appropriate registers.
- if (OpInfo.AssignedRegs.Regs.empty())
- report_fatal_error("Couldn't allocate input reg for constraint '" +
- Twine(OpInfo.ConstraintCode) + "'!");
+ if (OpInfo.AssignedRegs.Regs.empty()) {
+ LLVMContext &Ctx = *DAG.getContext();
+ Ctx.emitError(CS.getInstruction(),
+ "couldn't allocate input reg for constraint '" +
+ Twine(OpInfo.ConstraintCode) + "'");
+ break;
+ }
OpInfo.AssignedRegs.getCopyToRegs(InOperandVal, DAG, getCurDebugLoc(),
Chain, &Flag);
/// FIXME: When all targets are
/// migrated to using LowerCall, this hook should be integrated into SDISel.
std::pair<SDValue, SDValue>
-TargetLowering::LowerCallTo(SDValue Chain, Type *RetTy,
- bool RetSExt, bool RetZExt, bool isVarArg,
- bool isInreg, unsigned NumFixedArgs,
- CallingConv::ID CallConv, bool isTailCall,
- bool isReturnValueUsed,
- SDValue Callee,
- ArgListTy &Args, SelectionDAG &DAG,
- DebugLoc dl) const {
+TargetLowering::LowerCallTo(TargetLowering::CallLoweringInfo &CLI) const {
// Handle all of the outgoing arguments.
- SmallVector<ISD::OutputArg, 32> Outs;
- SmallVector<SDValue, 32> OutVals;
+ CLI.Outs.clear();
+ CLI.OutVals.clear();
+ ArgListTy &Args = CLI.Args;
for (unsigned i = 0, e = Args.size(); i != e; ++i) {
SmallVector<EVT, 4> ValueVTs;
ComputeValueVTs(*this, Args[i].Ty, ValueVTs);
for (unsigned Value = 0, NumValues = ValueVTs.size();
Value != NumValues; ++Value) {
EVT VT = ValueVTs[Value];
- Type *ArgTy = VT.getTypeForEVT(RetTy->getContext());
+ Type *ArgTy = VT.getTypeForEVT(CLI.RetTy->getContext());
SDValue Op = SDValue(Args[i].Node.getNode(),
Args[i].Node.getResNo() + Value);
ISD::ArgFlagsTy Flags;
Flags.setNest();
Flags.setOrigAlign(OriginalAlignment);
- EVT PartVT = getRegisterType(RetTy->getContext(), VT);
- unsigned NumParts = getNumRegisters(RetTy->getContext(), VT);
+ EVT PartVT = getRegisterType(CLI.RetTy->getContext(), VT);
+ unsigned NumParts = getNumRegisters(CLI.RetTy->getContext(), VT);
SmallVector<SDValue, 4> Parts(NumParts);
ISD::NodeType ExtendKind = ISD::ANY_EXTEND;
else if (Args[i].isZExt)
ExtendKind = ISD::ZERO_EXTEND;
- getCopyToParts(DAG, dl, Op, &Parts[0], NumParts,
+ getCopyToParts(CLI.DAG, CLI.DL, Op, &Parts[0], NumParts,
PartVT, ExtendKind);
for (unsigned j = 0; j != NumParts; ++j) {
// if it isn't first piece, alignment must be 1
ISD::OutputArg MyFlags(Flags, Parts[j].getValueType(),
- i < NumFixedArgs);
+ i < CLI.NumFixedArgs);
if (NumParts > 1 && j == 0)
MyFlags.Flags.setSplit();
else if (j != 0)
MyFlags.Flags.setOrigAlign(1);
- Outs.push_back(MyFlags);
- OutVals.push_back(Parts[j]);
+ CLI.Outs.push_back(MyFlags);
+ CLI.OutVals.push_back(Parts[j]);
}
}
}
// Handle the incoming return values from the call.
- SmallVector<ISD::InputArg, 32> Ins;
+ CLI.Ins.clear();
SmallVector<EVT, 4> RetTys;
- ComputeValueVTs(*this, RetTy, RetTys);
+ ComputeValueVTs(*this, CLI.RetTy, RetTys);
for (unsigned I = 0, E = RetTys.size(); I != E; ++I) {
EVT VT = RetTys[I];
- EVT RegisterVT = getRegisterType(RetTy->getContext(), VT);
- unsigned NumRegs = getNumRegisters(RetTy->getContext(), VT);
+ EVT RegisterVT = getRegisterType(CLI.RetTy->getContext(), VT);
+ unsigned NumRegs = getNumRegisters(CLI.RetTy->getContext(), VT);
for (unsigned i = 0; i != NumRegs; ++i) {
ISD::InputArg MyFlags;
MyFlags.VT = RegisterVT.getSimpleVT();
- MyFlags.Used = isReturnValueUsed;
- if (RetSExt)
+ MyFlags.Used = CLI.IsReturnValueUsed;
+ if (CLI.RetSExt)
MyFlags.Flags.setSExt();
- if (RetZExt)
+ if (CLI.RetZExt)
MyFlags.Flags.setZExt();
- if (isInreg)
+ if (CLI.IsInReg)
MyFlags.Flags.setInReg();
- Ins.push_back(MyFlags);
+ CLI.Ins.push_back(MyFlags);
}
}
SmallVector<SDValue, 4> InVals;
- Chain = LowerCall(Chain, Callee, CallConv, isVarArg, isTailCall,
- Outs, OutVals, Ins, dl, DAG, InVals);
+ CLI.Chain = LowerCall(CLI, InVals);
// Verify that the target's LowerCall behaved as expected.
- assert(Chain.getNode() && Chain.getValueType() == MVT::Other &&
+ assert(CLI.Chain.getNode() && CLI.Chain.getValueType() == MVT::Other &&
"LowerCall didn't return a valid chain!");
- assert((!isTailCall || InVals.empty()) &&
+ assert((!CLI.IsTailCall || InVals.empty()) &&
"LowerCall emitted a return value for a tail call!");
- assert((isTailCall || InVals.size() == Ins.size()) &&
+ assert((CLI.IsTailCall || InVals.size() == CLI.Ins.size()) &&
"LowerCall didn't emit the correct number of values!");
// For a tail call, the return value is merely live-out and there aren't
// any nodes in the DAG representing it. Return a special value to
// indicate that a tail call has been emitted and no more Instructions
// should be processed in the current block.
- if (isTailCall) {
- DAG.setRoot(Chain);
+ if (CLI.IsTailCall) {
+ CLI.DAG.setRoot(CLI.Chain);
return std::make_pair(SDValue(), SDValue());
}
- DEBUG(for (unsigned i = 0, e = Ins.size(); i != e; ++i) {
+ DEBUG(for (unsigned i = 0, e = CLI.Ins.size(); i != e; ++i) {
assert(InVals[i].getNode() &&
"LowerCall emitted a null value!");
- assert(EVT(Ins[i].VT) == InVals[i].getValueType() &&
+ assert(EVT(CLI.Ins[i].VT) == InVals[i].getValueType() &&
"LowerCall emitted a value with the wrong type!");
});
// Collect the legal value parts into potentially illegal values
// that correspond to the original function's return values.
ISD::NodeType AssertOp = ISD::DELETED_NODE;
- if (RetSExt)
+ if (CLI.RetSExt)
AssertOp = ISD::AssertSext;
- else if (RetZExt)
+ else if (CLI.RetZExt)
AssertOp = ISD::AssertZext;
SmallVector<SDValue, 4> ReturnValues;
unsigned CurReg = 0;
for (unsigned I = 0, E = RetTys.size(); I != E; ++I) {
EVT VT = RetTys[I];
- EVT RegisterVT = getRegisterType(RetTy->getContext(), VT);
- unsigned NumRegs = getNumRegisters(RetTy->getContext(), VT);
+ EVT RegisterVT = getRegisterType(CLI.RetTy->getContext(), VT);
+ unsigned NumRegs = getNumRegisters(CLI.RetTy->getContext(), VT);
- ReturnValues.push_back(getCopyFromParts(DAG, dl, &InVals[CurReg],
+ ReturnValues.push_back(getCopyFromParts(CLI.DAG, CLI.DL, &InVals[CurReg],
NumRegs, RegisterVT, VT,
AssertOp));
CurReg += NumRegs;
// such a node, so we just return a null return value in that case. In
// that case, nothing will actually look at the value.
if (ReturnValues.empty())
- return std::make_pair(SDValue(), Chain);
+ return std::make_pair(SDValue(), CLI.Chain);
- SDValue Res = DAG.getNode(ISD::MERGE_VALUES, dl,
- DAG.getVTList(&RetTys[0], RetTys.size()),
+ SDValue Res = CLI.DAG.getNode(ISD::MERGE_VALUES, CLI.DL,
+ CLI.DAG.getVTList(&RetTys[0], RetTys.size()),
&ReturnValues[0], ReturnValues.size());
- return std::make_pair(Res, Chain);
+ return std::make_pair(Res, CLI.Chain);
}
void TargetLowering::LowerOperationWrapper(SDNode *N,
SDValue TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
llvm_unreachable("LowerOperation not implemented for this target!");
- return SDValue();
}
void
/// isOnlyUsedInEntryBlock - If the specified argument is only used in the
/// entry block, return true. This includes arguments used by switches, since
/// the switch may expand into multiple basic blocks.
-static bool isOnlyUsedInEntryBlock(const Argument *A) {
+static bool isOnlyUsedInEntryBlock(const Argument *A, bool FastISel) {
// With FastISel active, we may be splitting blocks, so force creation
// of virtual registers for all non-dead arguments.
- if (EnableFastISel)
+ if (FastISel)
return A->use_empty();
const BasicBlock *Entry = A->getParent()->begin();
if (ArgValues.empty())
continue;
- // Note down frame index for byval arguments.
- if (I->hasByValAttr())
- if (FrameIndexSDNode *FI =
- dyn_cast<FrameIndexSDNode>(ArgValues[0].getNode()))
- FuncInfo->setByValArgumentFrameIndex(I, FI->getIndex());
+ // Note down frame index.
+ if (FrameIndexSDNode *FI =
+ dyn_cast<FrameIndexSDNode>(ArgValues[0].getNode()))
+ FuncInfo->setArgumentFrameIndex(I, FI->getIndex());
SDValue Res = DAG.getMergeValues(&ArgValues[0], NumValues,
SDB->getCurDebugLoc());
+
SDB->setValue(I, Res);
+ if (!TM.Options.EnableFastISel && Res.getOpcode() == ISD::BUILD_PAIR) {
+ if (LoadSDNode *LNode =
+ dyn_cast<LoadSDNode>(Res.getOperand(0).getNode()))
+ if (FrameIndexSDNode *FI =
+ dyn_cast<FrameIndexSDNode>(LNode->getBasePtr().getNode()))
+ FuncInfo->setArgumentFrameIndex(I, FI->getIndex());
+ }
// If this argument is live outside of the entry block, insert a copy from
// wherever we got it to the vreg that other BB's will reference it as.
- if (!EnableFastISel && Res.getOpcode() == ISD::CopyFromReg) {
+ if (!TM.Options.EnableFastISel && Res.getOpcode() == ISD::CopyFromReg) {
// If we can, though, try to skip creating an unnecessary vreg.
// FIXME: This isn't very clean... it would be nice to make this more
// general. It's also subtly incompatible with the hacks FastISel
continue;
}
}
- if (!isOnlyUsedInEntryBlock(I)) {
+ if (!isOnlyUsedInEntryBlock(I, TM.Options.EnableFastISel)) {
FuncInfo->InitializeRegForValue(I);
SDB->CopyToExportRegsIfNeeded(I);
}
projects / oota-llvm.git / blobdiff