namespace {
/// RegsForValue - This struct represents the registers (physical or virtual)
- /// that a particular set of values is assigned, and the type information about
- /// the value. The most common situation is to represent one value at a time,
- /// but struct or array values are handled element-wise as multiple values.
- /// The splitting of aggregates is performed recursively, so that we never
- /// have aggregate-typed registers. The values at this point do not necessarily
- /// have legal types, so each value may require one or more registers of some
- /// legal type.
+ /// that a particular set of values is assigned, and the type information
+ /// about the value. The most common situation is to represent one value at a
+ /// time, but struct or array values are handled element-wise as multiple
+ /// values. The splitting of aggregates is performed recursively, so that we
+ /// never have aggregate-typed registers. The values at this point do not
+ /// necessarily have legal types, so each value may require one or more
+ /// registers of some legal type.
///
struct RegsForValue {
/// TLI - The TargetLowering object.
EVT IntermediateVT, RegisterVT;
unsigned NumIntermediates;
unsigned NumRegs =
- TLI.getVectorTypeBreakdown(*DAG.getContext(), ValueVT, IntermediateVT,
+ TLI.getVectorTypeBreakdown(*DAG.getContext(), ValueVT, IntermediateVT,
NumIntermediates, RegisterVT);
- assert(NumRegs == NumParts && "Part count doesn't match vector breakdown!");
+ assert(NumRegs == NumParts
+ && "Part count doesn't match vector breakdown!");
NumParts = NumRegs; // Silence a compiler warning.
- assert(RegisterVT == PartVT && "Part type doesn't match vector breakdown!");
+ assert(RegisterVT == PartVT
+ && "Part type doesn't match vector breakdown!");
assert(RegisterVT == Parts[0].getValueType() &&
"Part type doesn't match part!");
Ops[i] = getCopyFromParts(DAG, dl, Order, &Parts[i], 1,
PartVT, IntermediateVT);
} else if (NumParts > 0) {
- // If the intermediate type was expanded, build the intermediate operands
- // from the parts.
+ // If the intermediate type was expanded, build the intermediate
+ // operands from the parts.
assert(NumParts % NumIntermediates == 0 &&
"Must expand into a divisible number of parts!");
unsigned Factor = NumParts / NumIntermediates;
PartVT, IntermediateVT);
}
- // Build a vector with BUILD_VECTOR or CONCAT_VECTORS from the intermediate
- // operands.
+ // Build a vector with BUILD_VECTOR or CONCAT_VECTORS from the
+ // intermediate operands.
Val = DAG.getNode(IntermediateVT.isVector() ?
ISD::CONCAT_VECTORS : ISD::BUILD_VECTOR, dl,
ValueVT, &Ops[0], NumIntermediates);
// The number of parts is a power of 2. Repeatedly bisect the value using
// EXTRACT_ELEMENT.
Parts[0] = DAG.getNode(ISD::BIT_CONVERT, dl,
- EVT::getIntegerVT(*DAG.getContext(), ValueVT.getSizeInBits()),
+ EVT::getIntegerVT(*DAG.getContext(),
+ ValueVT.getSizeInBits()),
Val);
if (DisableScheduling)
SDNodeOrder, Chain, NULL);
}
-/// Get the EVTs and ArgFlags collections that represent the return type
-/// of the given function. This does not require a DAG or a return value, and
-/// is suitable for use before any DAGs for the function are constructed.
+/// Get the EVTs and ArgFlags collections that represent the legalized return
+/// type of the given function. This does not require a DAG or a return value,
+/// and is suitable for use before any DAGs for the function are constructed.
static void getReturnInfo(const Type* ReturnType,
Attributes attr, SmallVectorImpl<EVT> &OutVTs,
SmallVectorImpl<ISD::ArgFlagsTy> &OutFlags,
TargetLowering &TLI,
SmallVectorImpl<uint64_t> *Offsets = 0) {
SmallVector<EVT, 4> ValueVTs;
- ComputeValueVTs(TLI, ReturnType, ValueVTs, Offsets);
+ ComputeValueVTs(TLI, ReturnType, ValueVTs);
unsigned NumValues = ValueVTs.size();
- if ( NumValues == 0 ) return;
+ if (NumValues == 0) return;
+ unsigned Offset = 0;
for (unsigned j = 0, f = NumValues; j != f; ++j) {
EVT VT = ValueVTs[j];
unsigned NumParts = TLI.getNumRegisters(ReturnType->getContext(), VT);
EVT PartVT = TLI.getRegisterType(ReturnType->getContext(), VT);
+ unsigned PartSize = TLI.getTargetData()->getTypeAllocSize(
+ PartVT.getTypeForEVT(ReturnType->getContext()));
+
// 'inreg' on function refers to return value
ISD::ArgFlagsTy Flags = ISD::ArgFlagsTy();
if (attr & Attribute::InReg)
for (unsigned i = 0; i < NumParts; ++i) {
OutVTs.push_back(PartVT);
OutFlags.push_back(Flags);
+ if (Offsets)
+ {
+ Offsets->push_back(Offset);
+ Offset += PartSize;
+ }
}
}
}
SDValue Chain = getControlRoot();
SmallVector<ISD::OutputArg, 8> Outs;
FunctionLoweringInfo &FLI = DAG.getFunctionLoweringInfo();
-
+
if (!FLI.CanLowerReturn) {
unsigned DemoteReg = FLI.DemoteRegister;
const Function *F = I.getParent()->getParent();
// Leave Outs empty so that LowerReturn won't try to load return
// registers the usual way.
SmallVector<EVT, 1> PtrValueVTs;
- ComputeValueVTs(TLI, PointerType::getUnqual(F->getReturnType()),
+ ComputeValueVTs(TLI, PointerType::getUnqual(F->getReturnType()),
PtrValueVTs);
SDValue RetPtr = DAG.getRegister(DemoteReg, PtrValueVTs[0]);
SDValue RetOp = getValue(I.getOperand(0));
-
+
SmallVector<EVT, 4> ValueVTs;
SmallVector<uint64_t, 4> Offsets;
ComputeValueVTs(TLI, I.getOperand(0)->getType(), ValueVTs, &Offsets);
ComputeValueVTs(TLI, I.getOperand(i)->getType(), ValueVTs);
unsigned NumValues = ValueVTs.size();
if (NumValues == 0) continue;
-
+
SDValue RetOp = getValue(I.getOperand(i));
for (unsigned j = 0, f = NumValues; j != f; ++j) {
EVT VT = ValueVTs[j];
else if (F->paramHasAttr(0, Attribute::ZExt))
ExtendKind = ISD::ZERO_EXTEND;
- // FIXME: C calling convention requires the return type to be promoted to
- // at least 32-bit. But this is not necessary for non-C calling
+ // FIXME: C calling convention requires the return type to be promoted
+ // to at least 32-bit. But this is not necessary for non-C calling
// conventions. The frontend should mark functions whose return values
// require promoting with signext or zeroext attributes.
if (ExtendKind != ISD::ANY_EXTEND && VT.isInteger()) {
return false;
}
+ // Handle: (X != null) | (Y != null) --> (X|Y) != 0
+ // Handle: (X == null) & (Y == null) --> (X|Y) == 0
+ if (Cases[0].CmpRHS == Cases[1].CmpRHS &&
+ Cases[0].CC == Cases[1].CC &&
+ isa<Constant>(Cases[0].CmpRHS) &&
+ cast<Constant>(Cases[0].CmpRHS)->isNullValue()) {
+ if (Cases[0].CC == ISD::SETEQ && Cases[0].TrueBB == Cases[1].ThisBB)
+ return false;
+ if (Cases[0].CC == ISD::SETNE && Cases[0].FalseBB == Cases[1].ThisBB)
+ return false;
+ }
+
return true;
}
if (Density < 0.4)
return false;
- DEBUG(errs() << "Lowering jump table\n"
+ DEBUG(dbgs() << "Lowering jump table\n"
<< "First entry: " << First << ". Last entry: " << Last << '\n'
<< "Range: " << Range
<< "Size: " << TSize << ". Density: " << Density << "\n\n");
APInt LSize = FrontCase.size();
APInt RSize = TSize-LSize;
- DEBUG(errs() << "Selecting best pivot: \n"
+ DEBUG(dbgs() << "Selecting best pivot: \n"
<< "First: " << First << ", Last: " << Last <<'\n'
<< "LSize: " << LSize << ", RSize: " << RSize << '\n');
for (CaseItr I = CR.Range.first, J=I+1, E = CR.Range.second;
APInt Range = ComputeRange(LEnd, RBegin);
assert((Range - 2ULL).isNonNegative() &&
"Invalid case distance");
- double LDensity = (double)LSize.roundToDouble() /
+ double LDensity = (double)LSize.roundToDouble() /
(LEnd - First + 1ULL).roundToDouble();
double RDensity = (double)RSize.roundToDouble() /
(Last - RBegin + 1ULL).roundToDouble();
double Metric = Range.logBase2()*(LDensity+RDensity);
// Should always split in some non-trivial place
- DEBUG(errs() <<"=>Step\n"
+ DEBUG(dbgs() <<"=>Step\n"
<< "LEnd: " << LEnd << ", RBegin: " << RBegin << '\n'
<< "LDensity: " << LDensity
<< ", RDensity: " << RDensity << '\n'
if (FMetric < Metric) {
Pivot = J;
FMetric = Metric;
- DEBUG(errs() << "Current metric set to: " << FMetric << '\n');
+ DEBUG(dbgs() << "Current metric set to: " << FMetric << '\n');
}
LSize += J->size();
// Don't bother the code below, if there are too much unique destinations
return false;
}
- DEBUG(errs() << "Total number of unique destinations: " << Dests.size() << '\n'
- << "Total number of comparisons: " << numCmps << '\n');
+ DEBUG(dbgs() << "Total number of unique destinations: "
+ << Dests.size() << '\n'
+ << "Total number of comparisons: " << numCmps << '\n');
// Compute span of values.
const APInt& minValue = cast<ConstantInt>(FrontCase.Low)->getValue();
const APInt& maxValue = cast<ConstantInt>(BackCase.High)->getValue();
APInt cmpRange = maxValue - minValue;
- DEBUG(errs() << "Compare range: " << cmpRange << '\n'
+ DEBUG(dbgs() << "Compare range: " << cmpRange << '\n'
<< "Low bound: " << minValue << '\n'
<< "High bound: " << maxValue << '\n');
!(Dests.size() >= 3 && numCmps >= 6)))
return false;
- DEBUG(errs() << "Emitting bit tests\n");
+ DEBUG(dbgs() << "Emitting bit tests\n");
APInt lowBound = APInt::getNullValue(cmpRange.getBitWidth());
// Optimize the case where all the case values fit in a
const BasicBlock *LLVMBB = CR.CaseBB->getBasicBlock();
- DEBUG(errs() << "Cases:\n");
+ DEBUG(dbgs() << "Cases:\n");
for (unsigned i = 0, e = CasesBits.size(); i!=e; ++i) {
- DEBUG(errs() << "Mask: " << CasesBits[i].Mask
+ DEBUG(dbgs() << "Mask: " << CasesBits[i].Mask
<< ", Bits: " << CasesBits[i].Bits
<< ", BB: " << CasesBits[i].BB << '\n');
// create a binary search tree from them.
CaseVector Cases;
size_t numCmps = Clusterify(Cases, SI);
- DEBUG(errs() << "Clusterify finished. Total clusters: " << Cases.size()
+ DEBUG(dbgs() << "Clusterify finished. Total clusters: " << Cases.size()
<< ". Total compares: " << numCmps << '\n');
numCmps = 0;
if (CV == CNZ) {
SDValue Op2 = getValue(I.getOperand(1));
SDValue Res = DAG.getNode(ISD::FNEG, getCurDebugLoc(),
- Op2.getValueType(), Op2);
+ Op2.getValueType(), Op2);
setValue(&I, Res);
if (DisableScheduling)
SDValue Op1 = getValue(I.getOperand(0));
SDValue Op2 = getValue(I.getOperand(1));
ISD::CondCode Opcode = getICmpCondCode(predicate);
-
+
EVT DestVT = TLI.getValueType(I.getType());
SDValue Res = DAG.getSetCC(getCurDebugLoc(), DestVT, Op1, Op2, Opcode);
setValue(&I, Res);
// 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(*DAG.getContext(),
+ cast<Constant>(I.getOperand(2))->getVectorElements(*DAG.getContext(),
MaskElts);
unsigned MaskNumElts = MaskElts.size();
for (unsigned i = 0; i != MaskNumElts; ++i) {
else
Mask.push_back(cast<ConstantInt>(MaskElts[i])->getSExtValue());
}
-
+
EVT VT = TLI.getValueType(I.getType());
EVT SrcVT = Src1.getValueType();
unsigned SrcNumElts = SrcVT.getVectorNumElements();
SmallVector<SDValue, 8> MOps2(NumConcat, UndefVal);
MOps1[0] = Src1;
MOps2[0] = Src2;
-
- Src1 = Src1U ? DAG.getUNDEF(VT) : DAG.getNode(ISD::CONCAT_VECTORS,
- getCurDebugLoc(), VT,
+
+ Src1 = Src1U ? DAG.getUNDEF(VT) : DAG.getNode(ISD::CONCAT_VECTORS,
+ getCurDebugLoc(), VT,
&MOps1[0], NumConcat);
Src2 = Src2U ? DAG.getUNDEF(VT) : DAG.getNode(ISD::CONCAT_VECTORS,
- getCurDebugLoc(), VT,
+ getCurDebugLoc(), VT,
&MOps2[0], NumConcat);
// Readjust mask for new input vector length.
MappedOps.push_back(Idx + MaskNumElts - SrcNumElts);
}
- SDValue Res = DAG.getVectorShuffle(VT, getCurDebugLoc(), Src1, Src2,
+ SDValue Res = DAG.getVectorShuffle(VT, getCurDebugLoc(), Src1, Src2,
&MappedOps[0]);
setValue(&I, Res);
int Input = 0;
if (Idx < 0)
continue;
-
+
if (Idx >= (int)SrcNumElts) {
Input = 1;
Idx -= SrcNumElts;
// 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] = { 2, 2 }; // 0 = Unused, 1 = Extract, 2 = 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) {
I.getAlignment());
SDValue AllocSize = getValue(I.getArraySize());
-
+
AllocSize = DAG.getNode(ISD::MUL, getCurDebugLoc(), AllocSize.getValueType(),
AllocSize,
DAG.getConstant(TySize, AllocSize.getValueType()));
-
+
if (DisableScheduling)
DAG.AssignOrdering(AllocSize.getNode(), SDNodeOrder);
-
+
EVT IntPtr = TLI.getPointerTy();
AllocSize = DAG.getZExtOrTrunc(AllocSize, getCurDebugLoc(), IntPtr);
if (DisableScheduling)
DAG.AssignOrdering(AllocSize.getNode(), SDNodeOrder);
-
+
// Handle alignment. If the requested alignment is less than or equal to
// the stack alignment, ignore it. If the size is greater than or equal to
// the stack alignment, we note this in the DYNAMIC_STACKALLOC node.
} else if (!HasChain) {
Result = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, getCurDebugLoc(),
VTs, &Ops[0], Ops.size());
- } else if (I.getType() != Type::getVoidTy(*DAG.getContext())) {
+ } else if (!I.getType()->isVoidTy()) {
Result = DAG.getNode(ISD::INTRINSIC_W_CHAIN, getCurDebugLoc(),
VTs, &Ops[0], Ops.size());
} else {
DAG.setRoot(Chain);
}
- if (I.getType() != Type::getVoidTy(*DAG.getContext())) {
+ if (!I.getType()->isVoidTy()) {
if (const VectorType *PTy = dyn_cast<VectorType>(I.getType())) {
EVT VT = TLI.getValueType(PTy);
Result = DAG.getNode(ISD::BIT_CONVERT, getCurDebugLoc(), VT, Result);
setValue(&I, result);
}
+
+/// ExpandPowI - Expand a llvm.powi intrinsic.
+static SDValue ExpandPowI(DebugLoc DL, SDValue LHS, SDValue RHS,
+ SelectionDAG &DAG) {
+ // If RHS is a constant, we can expand this out to a multiplication tree,
+ // otherwise we end up lowering to a call to __powidf2 (for example). When
+ // optimizing for size, we only want to do this if the expansion would produce
+ // a small number of multiplies, otherwise we do the full expansion.
+ if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(RHS)) {
+ // Get the exponent as a positive value.
+ unsigned Val = RHSC->getSExtValue();
+ if ((int)Val < 0) Val = -Val;
+
+ // powi(x, 0) -> 1.0
+ if (Val == 0)
+ return DAG.getConstantFP(1.0, LHS.getValueType());
+
+ Function *F = DAG.getMachineFunction().getFunction();
+ if (!F->hasFnAttr(Attribute::OptimizeForSize) ||
+ // If optimizing for size, don't insert too many multiplies. This
+ // inserts up to 5 multiplies.
+ CountPopulation_32(Val)+Log2_32(Val) < 7) {
+ // We use the simple binary decomposition method to generate the multiply
+ // sequence. There are more optimal ways to do this (for example,
+ // powi(x,15) generates one more multiply than it should), but this has
+ // the benefit of being both really simple and much better than a libcall.
+ SDValue Res; // Logically starts equal to 1.0
+ SDValue CurSquare = LHS;
+ while (Val) {
+ if (Val & 1) {
+ if (Res.getNode())
+ Res = DAG.getNode(ISD::FMUL, DL,Res.getValueType(), Res, CurSquare);
+ else
+ Res = CurSquare; // 1.0*CurSquare.
+ }
+
+ CurSquare = DAG.getNode(ISD::FMUL, DL, CurSquare.getValueType(),
+ CurSquare, CurSquare);
+ Val >>= 1;
+ }
+
+ // If the original was negative, invert the result, producing 1/(x*x*x).
+ if (RHSC->getSExtValue() < 0)
+ Res = DAG.getNode(ISD::FDIV, DL, LHS.getValueType(),
+ DAG.getConstantFP(1.0, LHS.getValueType()), Res);
+ return Res;
+ }
+ }
+
+ // Otherwise, expand to a libcall.
+ return DAG.getNode(ISD::FPOWI, DL, LHS.getValueType(), LHS, RHS);
+}
+
+
/// visitIntrinsicCall - Lower the call to the specified intrinsic function. If
/// we want to emit this as a call to a named external function, return the name
/// otherwise lower it and return null.
DAG.AssignOrdering(Res.getNode(), SDNodeOrder);
return 0;
}
- case Intrinsic::dbg_stoppoint:
- case Intrinsic::dbg_region_start:
- case Intrinsic::dbg_region_end:
- case Intrinsic::dbg_func_start:
- // FIXME - Remove this instructions once the dust settles.
- return 0;
case Intrinsic::dbg_declare: {
- if (OptLevel != CodeGenOpt::None)
+ if (OptLevel != CodeGenOpt::None)
// FIXME: Variable debug info is not supported here.
return 0;
DwarfWriter *DW = DAG.getDwarfWriter();
if (!DW)
return 0;
DbgDeclareInst &DI = cast<DbgDeclareInst>(I);
- if (!isValidDebugInfoIntrinsic(DI, CodeGenOpt::None))
+ if (!DIDescriptor::ValidDebugInfo(DI.getVariable(), CodeGenOpt::None))
return 0;
MDNode *Variable = DI.getVariable();
return 0;
DenseMap<const AllocaInst*, int>::iterator SI =
FuncInfo.StaticAllocaMap.find(AI);
- if (SI == FuncInfo.StaticAllocaMap.end())
+ if (SI == FuncInfo.StaticAllocaMap.end())
return 0; // VLAs.
int FI = SI->second;
DAG.AssignOrdering(Res.getNode(), SDNodeOrder);
return 0;
case Intrinsic::powi:
- Res = DAG.getNode(ISD::FPOWI, dl,
- getValue(I.getOperand(1)).getValueType(),
- getValue(I.getOperand(1)),
- getValue(I.getOperand(2)));
+ Res = ExpandPowI(dl, getValue(I.getOperand(1)), getValue(I.getOperand(2)),
+ DAG);
setValue(&I, Res);
if (DisableScheduling)
DAG.AssignOrdering(Res.getNode(), SDNodeOrder);
SmallVector<EVT, 4> OutVTs;
SmallVector<ISD::ArgFlagsTy, 4> OutsFlags;
SmallVector<uint64_t, 4> Offsets;
- getReturnInfo(RetTy, CS.getAttributes().getRetAttributes(),
+ getReturnInfo(RetTy, CS.getAttributes().getRetAttributes(),
OutVTs, OutsFlags, TLI, &Offsets);
- bool CanLowerReturn = TLI.CanLowerReturn(CS.getCallingConv(),
+ bool CanLowerReturn = TLI.CanLowerReturn(CS.getCallingConv(),
FTy->isVarArg(), OutVTs, OutsFlags, DAG);
SDValue DemoteStackSlot;
SDValue Chain = DAG.getNode(ISD::TokenFactor, getCurDebugLoc(),
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(), SDNodeOrder, &Values[CurReg], NumRegs,
+ RegisterVT, VT, AssertOp);
+ ReturnValues.push_back(ReturnValue);
+ if (DisableScheduling)
+ DAG.AssignOrdering(ReturnValue.getNode(), SDNodeOrder);
+ CurReg += NumRegs;
+ }
+ SDValue Res = DAG.getNode(ISD::MERGE_VALUES, getCurDebugLoc(),
+ DAG.getVTList(&RetTys[0], RetTys.size()),
+ &ReturnValues[0], ReturnValues.size());
- SDValue MV = DAG.getNode(ISD::MERGE_VALUES,
- getCurDebugLoc(),
- DAG.getVTList(&OutVTs[0], NumValues),
- &Values[0], NumValues);
- setValue(CS.getInstruction(), MV);
+ setValue(CS.getInstruction(), Res);
if (DisableScheduling) {
DAG.AssignOrdering(Chain.getNode(), SDNodeOrder);
- DAG.AssignOrdering(MV.getNode(), SDNodeOrder);
+ DAG.AssignOrdering(Res.getNode(), SDNodeOrder);
}
}
static SDValue getMemCmpLoad(Value *PtrVal, MVT LoadVT, const Type *LoadTy,
SelectionDAGBuilder &Builder) {
-
+
// Check to see if this load can be trivially constant folded, e.g. if the
// input is from a string literal.
if (Constant *LoadInput = dyn_cast<Constant>(PtrVal)) {
// Cast pointer to the type we really want to load.
LoadInput = ConstantExpr::getBitCast(LoadInput,
PointerType::getUnqual(LoadTy));
-
+
if (Constant *LoadCst = ConstantFoldLoadFromConstPtr(LoadInput, Builder.TD))
return Builder.getValue(LoadCst);
}
-
+
// Otherwise, we have to emit the load. If the pointer is to unfoldable but
// still constant memory, the input chain can be the entry node.
SDValue Root;
bool ConstantMemory = false;
-
+
// Do not serialize (non-volatile) loads of constant memory with anything.
if (Builder.AA->pointsToConstantMemory(PtrVal)) {
Root = Builder.DAG.getEntryNode();
// Do not serialize non-volatile loads against each other.
Root = Builder.DAG.getRoot();
}
-
+
SDValue Ptr = Builder.getValue(PtrVal);
SDValue LoadVal = Builder.DAG.getLoad(LoadVT, Builder.getCurDebugLoc(), Root,
Ptr, PtrVal /*SrcValue*/, 0/*SVOffset*/,
false /*volatile*/, 1 /* align=1 */);
-
+
if (!ConstantMemory)
Builder.PendingLoads.push_back(LoadVal.getValue(1));
return LoadVal;
// Verify that the prototype makes sense. int memcmp(void*,void*,size_t)
if (I.getNumOperands() != 4)
return false;
-
+
Value *LHS = I.getOperand(1), *RHS = I.getOperand(2);
if (!isa<PointerType>(LHS->getType()) || !isa<PointerType>(RHS->getType()) ||
!isa<IntegerType>(I.getOperand(3)->getType()) ||
!isa<IntegerType>(I.getType()))
- return false;
-
+ return false;
+
ConstantInt *Size = dyn_cast<ConstantInt>(I.getOperand(3));
-
+
// memcmp(S1,S2,2) != 0 -> (*(short*)LHS != *(short*)RHS) != 0
// memcmp(S1,S2,4) != 0 -> (*(int*)LHS != *(int*)RHS) != 0
if (Size && IsOnlyUsedInZeroEqualityComparison(&I)) {
break;
case 4:
LoadVT = MVT::i32;
- LoadTy = Type::getInt32Ty(Size->getContext());
+ LoadTy = Type::getInt32Ty(Size->getContext());
break;
case 8:
LoadVT = MVT::i64;
- LoadTy = Type::getInt64Ty(Size->getContext());
+ LoadTy = Type::getInt64Ty(Size->getContext());
break;
/*
case 16:
LoadVT = MVT::v4i32;
- LoadTy = Type::getInt32Ty(Size->getContext());
+ LoadTy = Type::getInt32Ty(Size->getContext());
LoadTy = VectorType::get(LoadTy, 4);
break;
*/
}
-
+
// This turns into unaligned loads. We only do this if the target natively
// supports the MVT we'll be loading or if it is small enough (<= 4) that
// we'll only produce a small number of byte loads.
-
+
// Require that we can find a legal MVT, and only do this if the target
// supports unaligned loads of that type. Expanding into byte loads would
// bloat the code.
if (!TLI.isTypeLegal(LoadVT) ||!TLI.allowsUnalignedMemoryAccesses(LoadVT))
ActuallyDoIt = false;
}
-
+
if (ActuallyDoIt) {
SDValue LHSVal = getMemCmpLoad(LHS, LoadVT, LoadTy, *this);
SDValue RHSVal = getMemCmpLoad(RHS, LoadVT, LoadTy, *this);
-
+
SDValue Res = DAG.getSetCC(getCurDebugLoc(), MVT::i1, LHSVal, RHSVal,
ISD::SETNE);
EVT CallVT = TLI.getValueType(I.getType(), true);
return true;
}
}
-
-
+
+
return false;
}
/// getCallOperandValEVT - Return the EVT of the Value* that this operand
/// corresponds to. If there is no Value* for this operand, it returns
/// MVT::Other.
- EVT getCallOperandValEVT(LLVMContext &Context,
+ EVT getCallOperandValEVT(LLVMContext &Context,
const TargetLowering &TLI,
const TargetData *TD) const {
if (CallOperandVal == 0) return MVT::Other;
// bitcast to the corresponding integer type. This turns an f64 value
// into i64, which can be passed with two i32 values on a 32-bit
// machine.
- RegVT = EVT::getIntegerVT(Context,
+ RegVT = EVT::getIntegerVT(Context,
OpInfo.ConstraintVT.getSizeInBits());
OpInfo.CallOperand = DAG.getNode(ISD::BIT_CONVERT, getCurDebugLoc(),
RegVT, OpInfo.CallOperand);
OpInfo.AssignedRegs = RegsForValue(TLI, Regs, RegVT, ValueVT);
return;
}
-
+
// This is a reference to a register class that doesn't directly correspond
// to an LLVM register class. Allocate NumRegs consecutive, available,
// registers from the class.
if (CType == TargetLowering::C_Memory)
return true;
}
-
+
// Indirect operand accesses access memory.
if (CI.isIndirect)
return true;
ConstraintInfos = IA->ParseConstraints();
bool hasMemory = hasInlineAsmMemConstraint(ConstraintInfos, TLI);
-
+
SDValue Chain, Flag;
-
+
// We won't need to flush pending loads if this asm doesn't touch
// memory and is nonvolatile.
if (hasMemory || IA->hasSideEffects())
// The return value of the call is this value. As such, there is no
// corresponding argument.
- assert(CS.getType() != Type::getVoidTy(*DAG.getContext()) &&
+ assert(!CS.getType()->isVoidTy() &&
"Bad inline asm!");
if (const StructType *STy = dyn_cast<StructType>(CS.getType())) {
OpVT = TLI.getValueType(STy->getElementType(ResNo));
std::vector<SDValue> AsmNodeOperands;
AsmNodeOperands.push_back(SDValue()); // reserve space for input chain
AsmNodeOperands.push_back(
- DAG.getTargetExternalSymbol(IA->getAsmString().c_str(), MVT::Other));
+ DAG.getTargetExternalSymbol(IA->getAsmString().c_str(),
+ TLI.getPointerTy()));
// Loop over all of the inputs, copying the operand values into the
OpInfo.CallOperandVal));
} else {
// This is the result value of the call.
- assert(CS.getType() != Type::getVoidTy(*DAG.getContext()) &&
- "Bad inline asm!");
+ assert(!CS.getType()->isVoidTy() && "Bad inline asm!");
// Concatenate this output onto the outputs list.
RetValRegs.append(OpInfo.AssignedRegs);
}
MachineRegisterInfo &RegInfo = DAG.getMachineFunction().getRegInfo();
for (unsigned i = 0, e = InlineAsm::getNumOperandRegisters(OpFlag);
i != e; ++i)
- MatchedRegs.Regs.
- push_back(RegInfo.createVirtualRegister(TLI.getRegClassFor(RegVT)));
+ MatchedRegs.Regs.push_back
+ (RegInfo.createVirtualRegister(TLI.getRegClassFor(RegVT)));
// Use the produced MatchedRegs object to
MatchedRegs.getCopyToRegs(InOperandVal, DAG, getCurDebugLoc(),
// Check whether the function can return without sret-demotion.
SmallVector<EVT, 4> OutVTs;
SmallVector<ISD::ArgFlagsTy, 4> OutsFlags;
- getReturnInfo(F.getReturnType(), F.getAttributes().getRetAttributes(),
+ getReturnInfo(F.getReturnType(), F.getAttributes().getRetAttributes(),
OutVTs, OutsFlags, TLI);
FunctionLoweringInfo &FLI = DAG.getFunctionLoweringInfo();
- FLI.CanLowerReturn = TLI.CanLowerReturn(F.getCallingConv(), F.isVarArg(),
+ FLI.CanLowerReturn = TLI.CanLowerReturn(F.getCallingConv(), F.isVarArg(),
OutVTs, OutsFlags, DAG);
if (!FLI.CanLowerReturn) {
// Put in an sret pointer parameter before all the other parameters.
MachineRegisterInfo& RegInfo = MF.getRegInfo();
unsigned SRetReg = RegInfo.createVirtualRegister(TLI.getRegClassFor(RegVT));
FLI.DemoteRegister = SRetReg;
- NewRoot = SDB->DAG.getCopyToReg(NewRoot, SDB->getCurDebugLoc(), SRetReg, ArgValue);
+ NewRoot = SDB->DAG.getCopyToReg(NewRoot, SDB->getCurDebugLoc(),
+ SRetReg, ArgValue);
DAG.setRoot(NewRoot);
// i indexes lowered arguments. Bump it past the hidden sret argument.
projects / oota-llvm.git / blobdiff