#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"
}
/// 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
} 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));
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 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);
}
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;
// 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 = 0; i < SI.getNumCases(); ++i) {
- BasicBlock *SuccBB = SI.getCaseSuccessor(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.getCaseValue(i),
- SI.getCaseValue(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;
}
// Utility for visitShuffleVector - Return true if every element in Mask,
-// begining // from position Pos and ending in Pos+Size, falls within the
+// 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,
- int Pos, int Size, int Low) {
- for (int i = Pos, e = Pos+Size; i != e; ++i, ++Low)
+ 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;
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) {
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, isInvariant, Alignment, TBAAInfo);
+ isNonTemporal, isInvariant, Alignment, TBAAInfo,
+ Ranges);
Values[i] = L;
Chains[ChainI] = L.getValue(1);
}
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);
}
-// 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;
}
// Some arguments' frame index is recorded during argument lowering.
Offset = FuncInfo.getArgumentFrameIndex(Arg);
if (Offset)
- Reg = TRI->getFrameRegister(MF);
+ Reg = TRI->getFrameRegister(MF);
if (!Reg && N.getNode()) {
if (N.getOpcode() == ISD::CopyFromReg)
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(Variable).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
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())
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(),
}
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());
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();
// 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(),
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, false, 1);
Values[i] = L;
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
(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.getType() == I.getArgOperand(0)->getType() &&
+ I.onlyReadsMemory()) {
SDValue Tmp = getValue(I.getArgOperand(0));
setValue(&I, DAG.getNode(ISD::FLOG2, getCurDebugLoc(),
Tmp.getValueType(), Tmp));
(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.getType() == I.getArgOperand(0)->getType() &&
+ I.onlyReadsMemory()) {
SDValue Tmp = getValue(I.getArgOperand(0));
setValue(&I, DAG.getNode(ISD::FEXP2, getCurDebugLoc(),
Tmp.getValueType(), Tmp));
: 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);
// 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.
/// 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,
projects / oota-llvm.git / blobdiff